projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: remove stale newlines from log messages
[deliverable/linux.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/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 /*
398 * skip compression for a small file range(<=blocksize) that
399 * isn't an inline extent, since it dosen't save disk space at all.
400 */
401 if ((end - start + 1) <= blocksize &&
402 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
403 goto cleanup_and_bail_uncompressed;
404
405 actual_end = min_t(u64, isize, end + 1);
406 again:
407 will_compress = 0;
408 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
409 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
410
411 /*
412 * we don't want to send crud past the end of i_size through
413 * compression, that's just a waste of CPU time. So, if the
414 * end of the file is before the start of our current
415 * requested range of bytes, we bail out to the uncompressed
416 * cleanup code that can deal with all of this.
417 *
418 * It isn't really the fastest way to fix things, but this is a
419 * very uncommon corner.
420 */
421 if (actual_end <= start)
422 goto cleanup_and_bail_uncompressed;
423
424 total_compressed = actual_end - start;
425
426 /* we want to make sure that amount of ram required to uncompress
427 * an extent is reasonable, so we limit the total size in ram
428 * of a compressed extent to 128k. This is a crucial number
429 * because it also controls how easily we can spread reads across
430 * cpus for decompression.
431 *
432 * We also want to make sure the amount of IO required to do
433 * a random read is reasonably small, so we limit the size of
434 * a compressed extent to 128k.
435 */
436 total_compressed = min(total_compressed, max_uncompressed);
437 num_bytes = ALIGN(end - start + 1, blocksize);
438 num_bytes = max(blocksize, num_bytes);
439 total_in = 0;
440 ret = 0;
441
442 /*
443 * we do compression for mount -o compress and when the
444 * inode has not been flagged as nocompress. This flag can
445 * change at any time if we discover bad compression ratios.
446 */
447 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
448 (btrfs_test_opt(root, COMPRESS) ||
449 (BTRFS_I(inode)->force_compress) ||
450 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
451 WARN_ON(pages);
452 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
453 if (!pages) {
454 /* just bail out to the uncompressed code */
455 goto cont;
456 }
457
458 if (BTRFS_I(inode)->force_compress)
459 compress_type = BTRFS_I(inode)->force_compress;
460
461 /*
462 * we need to call clear_page_dirty_for_io on each
463 * page in the range. Otherwise applications with the file
464 * mmap'd can wander in and change the page contents while
465 * we are compressing them.
466 *
467 * If the compression fails for any reason, we set the pages
468 * dirty again later on.
469 */
470 extent_range_clear_dirty_for_io(inode, start, end);
471 redirty = 1;
472 ret = btrfs_compress_pages(compress_type,
473 inode->i_mapping, start,
474 total_compressed, pages,
475 nr_pages, &nr_pages_ret,
476 &total_in,
477 &total_compressed,
478 max_compressed);
479
480 if (!ret) {
481 unsigned long offset = total_compressed &
482 (PAGE_CACHE_SIZE - 1);
483 struct page *page = pages[nr_pages_ret - 1];
484 char *kaddr;
485
486 /* zero the tail end of the last page, we might be
487 * sending it down to disk
488 */
489 if (offset) {
490 kaddr = kmap_atomic(page);
491 memset(kaddr + offset, 0,
492 PAGE_CACHE_SIZE - offset);
493 kunmap_atomic(kaddr);
494 }
495 will_compress = 1;
496 }
497 }
498 cont:
499 if (start == 0) {
500 /* lets try to make an inline extent */
501 if (ret || total_in < (actual_end - start)) {
502 /* we didn't compress the entire range, try
503 * to make an uncompressed inline extent.
504 */
505 ret = cow_file_range_inline(root, inode, start, end,
506 0, 0, NULL);
507 } else {
508 /* try making a compressed inline extent */
509 ret = cow_file_range_inline(root, inode, start, end,
510 total_compressed,
511 compress_type, pages);
512 }
513 if (ret <= 0) {
514 unsigned long clear_flags = EXTENT_DELALLOC |
515 EXTENT_DEFRAG;
516 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
517
518 /*
519 * inline extent creation worked or returned error,
520 * we don't need to create any more async work items.
521 * Unlock and free up our temp pages.
522 */
523 extent_clear_unlock_delalloc(inode, start, end, NULL,
524 clear_flags, PAGE_UNLOCK |
525 PAGE_CLEAR_DIRTY |
526 PAGE_SET_WRITEBACK |
527 PAGE_END_WRITEBACK);
528 goto free_pages_out;
529 }
530 }
531
532 if (will_compress) {
533 /*
534 * we aren't doing an inline extent round the compressed size
535 * up to a block size boundary so the allocator does sane
536 * things
537 */
538 total_compressed = ALIGN(total_compressed, blocksize);
539
540 /*
541 * one last check to make sure the compression is really a
542 * win, compare the page count read with the blocks on disk
543 */
544 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
545 if (total_compressed >= total_in) {
546 will_compress = 0;
547 } else {
548 num_bytes = total_in;
549 }
550 }
551 if (!will_compress && pages) {
552 /*
553 * the compression code ran but failed to make things smaller,
554 * free any pages it allocated and our page pointer array
555 */
556 for (i = 0; i < nr_pages_ret; i++) {
557 WARN_ON(pages[i]->mapping);
558 page_cache_release(pages[i]);
559 }
560 kfree(pages);
561 pages = NULL;
562 total_compressed = 0;
563 nr_pages_ret = 0;
564
565 /* flag the file so we don't compress in the future */
566 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
567 !(BTRFS_I(inode)->force_compress)) {
568 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
569 }
570 }
571 if (will_compress) {
572 *num_added += 1;
573
574 /* the async work queues will take care of doing actual
575 * allocation on disk for these compressed pages,
576 * and will submit them to the elevator.
577 */
578 add_async_extent(async_cow, start, num_bytes,
579 total_compressed, pages, nr_pages_ret,
580 compress_type);
581
582 if (start + num_bytes < end) {
583 start += num_bytes;
584 pages = NULL;
585 cond_resched();
586 goto again;
587 }
588 } else {
589 cleanup_and_bail_uncompressed:
590 /*
591 * No compression, but we still need to write the pages in
592 * the file we've been given so far. redirty the locked
593 * page if it corresponds to our extent and set things up
594 * for the async work queue to run cow_file_range to do
595 * the normal delalloc dance
596 */
597 if (page_offset(locked_page) >= start &&
598 page_offset(locked_page) <= end) {
599 __set_page_dirty_nobuffers(locked_page);
600 /* unlocked later on in the async handlers */
601 }
602 if (redirty)
603 extent_range_redirty_for_io(inode, start, end);
604 add_async_extent(async_cow, start, end - start + 1,
605 0, NULL, 0, BTRFS_COMPRESS_NONE);
606 *num_added += 1;
607 }
608
609 out:
610 return ret;
611
612 free_pages_out:
613 for (i = 0; i < nr_pages_ret; i++) {
614 WARN_ON(pages[i]->mapping);
615 page_cache_release(pages[i]);
616 }
617 kfree(pages);
618
619 goto out;
620 }
621
622 /*
623 * phase two of compressed writeback. This is the ordered portion
624 * of the code, which only gets called in the order the work was
625 * queued. We walk all the async extents created by compress_file_range
626 * and send them down to the disk.
627 */
628 static noinline int submit_compressed_extents(struct inode *inode,
629 struct async_cow *async_cow)
630 {
631 struct async_extent *async_extent;
632 u64 alloc_hint = 0;
633 struct btrfs_key ins;
634 struct extent_map *em;
635 struct btrfs_root *root = BTRFS_I(inode)->root;
636 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
637 struct extent_io_tree *io_tree;
638 int ret = 0;
639
640 if (list_empty(&async_cow->extents))
641 return 0;
642
643 again:
644 while (!list_empty(&async_cow->extents)) {
645 async_extent = list_entry(async_cow->extents.next,
646 struct async_extent, list);
647 list_del(&async_extent->list);
648
649 io_tree = &BTRFS_I(inode)->io_tree;
650
651 retry:
652 /* did the compression code fall back to uncompressed IO? */
653 if (!async_extent->pages) {
654 int page_started = 0;
655 unsigned long nr_written = 0;
656
657 lock_extent(io_tree, async_extent->start,
658 async_extent->start +
659 async_extent->ram_size - 1);
660
661 /* allocate blocks */
662 ret = cow_file_range(inode, async_cow->locked_page,
663 async_extent->start,
664 async_extent->start +
665 async_extent->ram_size - 1,
666 &page_started, &nr_written, 0);
667
668 /* JDM XXX */
669
670 /*
671 * if page_started, cow_file_range inserted an
672 * inline extent and took care of all the unlocking
673 * and IO for us. Otherwise, we need to submit
674 * all those pages down to the drive.
675 */
676 if (!page_started && !ret)
677 extent_write_locked_range(io_tree,
678 inode, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1,
681 btrfs_get_extent,
682 WB_SYNC_ALL);
683 else if (ret)
684 unlock_page(async_cow->locked_page);
685 kfree(async_extent);
686 cond_resched();
687 continue;
688 }
689
690 lock_extent(io_tree, async_extent->start,
691 async_extent->start + async_extent->ram_size - 1);
692
693 ret = btrfs_reserve_extent(root,
694 async_extent->compressed_size,
695 async_extent->compressed_size,
696 0, alloc_hint, &ins, 1);
697 if (ret) {
698 int i;
699
700 for (i = 0; i < async_extent->nr_pages; i++) {
701 WARN_ON(async_extent->pages[i]->mapping);
702 page_cache_release(async_extent->pages[i]);
703 }
704 kfree(async_extent->pages);
705 async_extent->nr_pages = 0;
706 async_extent->pages = NULL;
707
708 if (ret == -ENOSPC) {
709 unlock_extent(io_tree, async_extent->start,
710 async_extent->start +
711 async_extent->ram_size - 1);
712 goto retry;
713 }
714 goto out_free;
715 }
716
717 /*
718 * here we're doing allocation and writeback of the
719 * compressed pages
720 */
721 btrfs_drop_extent_cache(inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1, 0);
724
725 em = alloc_extent_map();
726 if (!em) {
727 ret = -ENOMEM;
728 goto out_free_reserve;
729 }
730 em->start = async_extent->start;
731 em->len = async_extent->ram_size;
732 em->orig_start = em->start;
733 em->mod_start = em->start;
734 em->mod_len = em->len;
735
736 em->block_start = ins.objectid;
737 em->block_len = ins.offset;
738 em->orig_block_len = ins.offset;
739 em->ram_bytes = async_extent->ram_size;
740 em->bdev = root->fs_info->fs_devices->latest_bdev;
741 em->compress_type = async_extent->compress_type;
742 set_bit(EXTENT_FLAG_PINNED, &em->flags);
743 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
744 em->generation = -1;
745
746 while (1) {
747 write_lock(&em_tree->lock);
748 ret = add_extent_mapping(em_tree, em, 1);
749 write_unlock(&em_tree->lock);
750 if (ret != -EEXIST) {
751 free_extent_map(em);
752 break;
753 }
754 btrfs_drop_extent_cache(inode, async_extent->start,
755 async_extent->start +
756 async_extent->ram_size - 1, 0);
757 }
758
759 if (ret)
760 goto out_free_reserve;
761
762 ret = btrfs_add_ordered_extent_compress(inode,
763 async_extent->start,
764 ins.objectid,
765 async_extent->ram_size,
766 ins.offset,
767 BTRFS_ORDERED_COMPRESSED,
768 async_extent->compress_type);
769 if (ret)
770 goto out_free_reserve;
771
772 /*
773 * clear dirty, set writeback and unlock the pages.
774 */
775 extent_clear_unlock_delalloc(inode, async_extent->start,
776 async_extent->start +
777 async_extent->ram_size - 1,
778 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
779 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
780 PAGE_SET_WRITEBACK);
781 ret = btrfs_submit_compressed_write(inode,
782 async_extent->start,
783 async_extent->ram_size,
784 ins.objectid,
785 ins.offset, async_extent->pages,
786 async_extent->nr_pages);
787 alloc_hint = ins.objectid + ins.offset;
788 kfree(async_extent);
789 if (ret)
790 goto out;
791 cond_resched();
792 }
793 ret = 0;
794 out:
795 return ret;
796 out_free_reserve:
797 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
798 out_free:
799 extent_clear_unlock_delalloc(inode, async_extent->start,
800 async_extent->start +
801 async_extent->ram_size - 1,
802 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
803 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
804 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
805 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
806 kfree(async_extent);
807 goto again;
808 }
809
810 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
811 u64 num_bytes)
812 {
813 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
814 struct extent_map *em;
815 u64 alloc_hint = 0;
816
817 read_lock(&em_tree->lock);
818 em = search_extent_mapping(em_tree, start, num_bytes);
819 if (em) {
820 /*
821 * if block start isn't an actual block number then find the
822 * first block in this inode and use that as a hint. If that
823 * block is also bogus then just don't worry about it.
824 */
825 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
826 free_extent_map(em);
827 em = search_extent_mapping(em_tree, 0, 0);
828 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
829 alloc_hint = em->block_start;
830 if (em)
831 free_extent_map(em);
832 } else {
833 alloc_hint = em->block_start;
834 free_extent_map(em);
835 }
836 }
837 read_unlock(&em_tree->lock);
838
839 return alloc_hint;
840 }
841
842 /*
843 * when extent_io.c finds a delayed allocation range in the file,
844 * the call backs end up in this code. The basic idea is to
845 * allocate extents on disk for the range, and create ordered data structs
846 * in ram to track those extents.
847 *
848 * locked_page is the page that writepage had locked already. We use
849 * it to make sure we don't do extra locks or unlocks.
850 *
851 * *page_started is set to one if we unlock locked_page and do everything
852 * required to start IO on it. It may be clean and already done with
853 * IO when we return.
854 */
855 static noinline int cow_file_range(struct inode *inode,
856 struct page *locked_page,
857 u64 start, u64 end, int *page_started,
858 unsigned long *nr_written,
859 int unlock)
860 {
861 struct btrfs_root *root = BTRFS_I(inode)->root;
862 u64 alloc_hint = 0;
863 u64 num_bytes;
864 unsigned long ram_size;
865 u64 disk_num_bytes;
866 u64 cur_alloc_size;
867 u64 blocksize = root->sectorsize;
868 struct btrfs_key ins;
869 struct extent_map *em;
870 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
871 int ret = 0;
872
873 if (btrfs_is_free_space_inode(inode)) {
874 WARN_ON_ONCE(1);
875 ret = -EINVAL;
876 goto out_unlock;
877 }
878
879 num_bytes = ALIGN(end - start + 1, blocksize);
880 num_bytes = max(blocksize, num_bytes);
881 disk_num_bytes = num_bytes;
882
883 /* if this is a small write inside eof, kick off defrag */
884 if (num_bytes < 64 * 1024 &&
885 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
886 btrfs_add_inode_defrag(NULL, inode);
887
888 if (start == 0) {
889 /* lets try to make an inline extent */
890 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
891 NULL);
892 if (ret == 0) {
893 extent_clear_unlock_delalloc(inode, start, end, NULL,
894 EXTENT_LOCKED | EXTENT_DELALLOC |
895 EXTENT_DEFRAG, PAGE_UNLOCK |
896 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
897 PAGE_END_WRITEBACK);
898
899 *nr_written = *nr_written +
900 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
901 *page_started = 1;
902 goto out;
903 } else if (ret < 0) {
904 goto out_unlock;
905 }
906 }
907
908 BUG_ON(disk_num_bytes >
909 btrfs_super_total_bytes(root->fs_info->super_copy));
910
911 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
912 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
913
914 while (disk_num_bytes > 0) {
915 unsigned long op;
916
917 cur_alloc_size = disk_num_bytes;
918 ret = btrfs_reserve_extent(root, cur_alloc_size,
919 root->sectorsize, 0, alloc_hint,
920 &ins, 1);
921 if (ret < 0)
922 goto out_unlock;
923
924 em = alloc_extent_map();
925 if (!em) {
926 ret = -ENOMEM;
927 goto out_reserve;
928 }
929 em->start = start;
930 em->orig_start = em->start;
931 ram_size = ins.offset;
932 em->len = ins.offset;
933 em->mod_start = em->start;
934 em->mod_len = em->len;
935
936 em->block_start = ins.objectid;
937 em->block_len = ins.offset;
938 em->orig_block_len = ins.offset;
939 em->ram_bytes = ram_size;
940 em->bdev = root->fs_info->fs_devices->latest_bdev;
941 set_bit(EXTENT_FLAG_PINNED, &em->flags);
942 em->generation = -1;
943
944 while (1) {
945 write_lock(&em_tree->lock);
946 ret = add_extent_mapping(em_tree, em, 1);
947 write_unlock(&em_tree->lock);
948 if (ret != -EEXIST) {
949 free_extent_map(em);
950 break;
951 }
952 btrfs_drop_extent_cache(inode, start,
953 start + ram_size - 1, 0);
954 }
955 if (ret)
956 goto out_reserve;
957
958 cur_alloc_size = ins.offset;
959 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
960 ram_size, cur_alloc_size, 0);
961 if (ret)
962 goto out_reserve;
963
964 if (root->root_key.objectid ==
965 BTRFS_DATA_RELOC_TREE_OBJECTID) {
966 ret = btrfs_reloc_clone_csums(inode, start,
967 cur_alloc_size);
968 if (ret)
969 goto out_reserve;
970 }
971
972 if (disk_num_bytes < cur_alloc_size)
973 break;
974
975 /* we're not doing compressed IO, don't unlock the first
976 * page (which the caller expects to stay locked), don't
977 * clear any dirty bits and don't set any writeback bits
978 *
979 * Do set the Private2 bit so we know this page was properly
980 * setup for writepage
981 */
982 op = unlock ? PAGE_UNLOCK : 0;
983 op |= PAGE_SET_PRIVATE2;
984
985 extent_clear_unlock_delalloc(inode, start,
986 start + ram_size - 1, locked_page,
987 EXTENT_LOCKED | EXTENT_DELALLOC,
988 op);
989 disk_num_bytes -= cur_alloc_size;
990 num_bytes -= cur_alloc_size;
991 alloc_hint = ins.objectid + ins.offset;
992 start += cur_alloc_size;
993 }
994 out:
995 return ret;
996
997 out_reserve:
998 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
999 out_unlock:
1000 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1001 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1002 EXTENT_DELALLOC | EXTENT_DEFRAG,
1003 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1004 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1005 goto out;
1006 }
1007
1008 /*
1009 * work queue call back to started compression on a file and pages
1010 */
1011 static noinline void async_cow_start(struct btrfs_work *work)
1012 {
1013 struct async_cow *async_cow;
1014 int num_added = 0;
1015 async_cow = container_of(work, struct async_cow, work);
1016
1017 compress_file_range(async_cow->inode, async_cow->locked_page,
1018 async_cow->start, async_cow->end, async_cow,
1019 &num_added);
1020 if (num_added == 0) {
1021 btrfs_add_delayed_iput(async_cow->inode);
1022 async_cow->inode = NULL;
1023 }
1024 }
1025
1026 /*
1027 * work queue call back to submit previously compressed pages
1028 */
1029 static noinline void async_cow_submit(struct btrfs_work *work)
1030 {
1031 struct async_cow *async_cow;
1032 struct btrfs_root *root;
1033 unsigned long nr_pages;
1034
1035 async_cow = container_of(work, struct async_cow, work);
1036
1037 root = async_cow->root;
1038 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1039 PAGE_CACHE_SHIFT;
1040
1041 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1042 5 * 1024 * 1024 &&
1043 waitqueue_active(&root->fs_info->async_submit_wait))
1044 wake_up(&root->fs_info->async_submit_wait);
1045
1046 if (async_cow->inode)
1047 submit_compressed_extents(async_cow->inode, async_cow);
1048 }
1049
1050 static noinline void async_cow_free(struct btrfs_work *work)
1051 {
1052 struct async_cow *async_cow;
1053 async_cow = container_of(work, struct async_cow, work);
1054 if (async_cow->inode)
1055 btrfs_add_delayed_iput(async_cow->inode);
1056 kfree(async_cow);
1057 }
1058
1059 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1060 u64 start, u64 end, int *page_started,
1061 unsigned long *nr_written)
1062 {
1063 struct async_cow *async_cow;
1064 struct btrfs_root *root = BTRFS_I(inode)->root;
1065 unsigned long nr_pages;
1066 u64 cur_end;
1067 int limit = 10 * 1024 * 1024;
1068
1069 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1070 1, 0, NULL, GFP_NOFS);
1071 while (start < end) {
1072 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1073 BUG_ON(!async_cow); /* -ENOMEM */
1074 async_cow->inode = igrab(inode);
1075 async_cow->root = root;
1076 async_cow->locked_page = locked_page;
1077 async_cow->start = start;
1078
1079 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1080 cur_end = end;
1081 else
1082 cur_end = min(end, start + 512 * 1024 - 1);
1083
1084 async_cow->end = cur_end;
1085 INIT_LIST_HEAD(&async_cow->extents);
1086
1087 btrfs_init_work(&async_cow->work, async_cow_start,
1088 async_cow_submit, async_cow_free);
1089
1090 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1091 PAGE_CACHE_SHIFT;
1092 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1093
1094 btrfs_queue_work(root->fs_info->delalloc_workers,
1095 &async_cow->work);
1096
1097 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1098 wait_event(root->fs_info->async_submit_wait,
1099 (atomic_read(&root->fs_info->async_delalloc_pages) <
1100 limit));
1101 }
1102
1103 while (atomic_read(&root->fs_info->async_submit_draining) &&
1104 atomic_read(&root->fs_info->async_delalloc_pages)) {
1105 wait_event(root->fs_info->async_submit_wait,
1106 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1107 0));
1108 }
1109
1110 *nr_written += nr_pages;
1111 start = cur_end + 1;
1112 }
1113 *page_started = 1;
1114 return 0;
1115 }
1116
1117 static noinline int csum_exist_in_range(struct btrfs_root *root,
1118 u64 bytenr, u64 num_bytes)
1119 {
1120 int ret;
1121 struct btrfs_ordered_sum *sums;
1122 LIST_HEAD(list);
1123
1124 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1125 bytenr + num_bytes - 1, &list, 0);
1126 if (ret == 0 && list_empty(&list))
1127 return 0;
1128
1129 while (!list_empty(&list)) {
1130 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1131 list_del(&sums->list);
1132 kfree(sums);
1133 }
1134 return 1;
1135 }
1136
1137 /*
1138 * when nowcow writeback call back. This checks for snapshots or COW copies
1139 * of the extents that exist in the file, and COWs the file as required.
1140 *
1141 * If no cow copies or snapshots exist, we write directly to the existing
1142 * blocks on disk
1143 */
1144 static noinline int run_delalloc_nocow(struct inode *inode,
1145 struct page *locked_page,
1146 u64 start, u64 end, int *page_started, int force,
1147 unsigned long *nr_written)
1148 {
1149 struct btrfs_root *root = BTRFS_I(inode)->root;
1150 struct btrfs_trans_handle *trans;
1151 struct extent_buffer *leaf;
1152 struct btrfs_path *path;
1153 struct btrfs_file_extent_item *fi;
1154 struct btrfs_key found_key;
1155 u64 cow_start;
1156 u64 cur_offset;
1157 u64 extent_end;
1158 u64 extent_offset;
1159 u64 disk_bytenr;
1160 u64 num_bytes;
1161 u64 disk_num_bytes;
1162 u64 ram_bytes;
1163 int extent_type;
1164 int ret, err;
1165 int type;
1166 int nocow;
1167 int check_prev = 1;
1168 bool nolock;
1169 u64 ino = btrfs_ino(inode);
1170
1171 path = btrfs_alloc_path();
1172 if (!path) {
1173 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1174 EXTENT_LOCKED | EXTENT_DELALLOC |
1175 EXTENT_DO_ACCOUNTING |
1176 EXTENT_DEFRAG, PAGE_UNLOCK |
1177 PAGE_CLEAR_DIRTY |
1178 PAGE_SET_WRITEBACK |
1179 PAGE_END_WRITEBACK);
1180 return -ENOMEM;
1181 }
1182
1183 nolock = btrfs_is_free_space_inode(inode);
1184
1185 if (nolock)
1186 trans = btrfs_join_transaction_nolock(root);
1187 else
1188 trans = btrfs_join_transaction(root);
1189
1190 if (IS_ERR(trans)) {
1191 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1192 EXTENT_LOCKED | EXTENT_DELALLOC |
1193 EXTENT_DO_ACCOUNTING |
1194 EXTENT_DEFRAG, PAGE_UNLOCK |
1195 PAGE_CLEAR_DIRTY |
1196 PAGE_SET_WRITEBACK |
1197 PAGE_END_WRITEBACK);
1198 btrfs_free_path(path);
1199 return PTR_ERR(trans);
1200 }
1201
1202 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1203
1204 cow_start = (u64)-1;
1205 cur_offset = start;
1206 while (1) {
1207 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1208 cur_offset, 0);
1209 if (ret < 0)
1210 goto error;
1211 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1212 leaf = path->nodes[0];
1213 btrfs_item_key_to_cpu(leaf, &found_key,
1214 path->slots[0] - 1);
1215 if (found_key.objectid == ino &&
1216 found_key.type == BTRFS_EXTENT_DATA_KEY)
1217 path->slots[0]--;
1218 }
1219 check_prev = 0;
1220 next_slot:
1221 leaf = path->nodes[0];
1222 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1223 ret = btrfs_next_leaf(root, path);
1224 if (ret < 0)
1225 goto error;
1226 if (ret > 0)
1227 break;
1228 leaf = path->nodes[0];
1229 }
1230
1231 nocow = 0;
1232 disk_bytenr = 0;
1233 num_bytes = 0;
1234 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1235
1236 if (found_key.objectid > ino ||
1237 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1238 found_key.offset > end)
1239 break;
1240
1241 if (found_key.offset > cur_offset) {
1242 extent_end = found_key.offset;
1243 extent_type = 0;
1244 goto out_check;
1245 }
1246
1247 fi = btrfs_item_ptr(leaf, path->slots[0],
1248 struct btrfs_file_extent_item);
1249 extent_type = btrfs_file_extent_type(leaf, fi);
1250
1251 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1252 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1253 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1254 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1255 extent_offset = btrfs_file_extent_offset(leaf, fi);
1256 extent_end = found_key.offset +
1257 btrfs_file_extent_num_bytes(leaf, fi);
1258 disk_num_bytes =
1259 btrfs_file_extent_disk_num_bytes(leaf, fi);
1260 if (extent_end <= start) {
1261 path->slots[0]++;
1262 goto next_slot;
1263 }
1264 if (disk_bytenr == 0)
1265 goto out_check;
1266 if (btrfs_file_extent_compression(leaf, fi) ||
1267 btrfs_file_extent_encryption(leaf, fi) ||
1268 btrfs_file_extent_other_encoding(leaf, fi))
1269 goto out_check;
1270 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1271 goto out_check;
1272 if (btrfs_extent_readonly(root, disk_bytenr))
1273 goto out_check;
1274 if (btrfs_cross_ref_exist(trans, root, ino,
1275 found_key.offset -
1276 extent_offset, disk_bytenr))
1277 goto out_check;
1278 disk_bytenr += extent_offset;
1279 disk_bytenr += cur_offset - found_key.offset;
1280 num_bytes = min(end + 1, extent_end) - cur_offset;
1281 /*
1282 * if there are pending snapshots for this root,
1283 * we fall into common COW way.
1284 */
1285 if (!nolock) {
1286 err = btrfs_start_nocow_write(root);
1287 if (!err)
1288 goto out_check;
1289 }
1290 /*
1291 * force cow if csum exists in the range.
1292 * this ensure that csum for a given extent are
1293 * either valid or do not exist.
1294 */
1295 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1296 goto out_check;
1297 nocow = 1;
1298 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299 extent_end = found_key.offset +
1300 btrfs_file_extent_inline_len(leaf,
1301 path->slots[0], fi);
1302 extent_end = ALIGN(extent_end, root->sectorsize);
1303 } else {
1304 BUG_ON(1);
1305 }
1306 out_check:
1307 if (extent_end <= start) {
1308 path->slots[0]++;
1309 if (!nolock && nocow)
1310 btrfs_end_nocow_write(root);
1311 goto next_slot;
1312 }
1313 if (!nocow) {
1314 if (cow_start == (u64)-1)
1315 cow_start = cur_offset;
1316 cur_offset = extent_end;
1317 if (cur_offset > end)
1318 break;
1319 path->slots[0]++;
1320 goto next_slot;
1321 }
1322
1323 btrfs_release_path(path);
1324 if (cow_start != (u64)-1) {
1325 ret = cow_file_range(inode, locked_page,
1326 cow_start, found_key.offset - 1,
1327 page_started, nr_written, 1);
1328 if (ret) {
1329 if (!nolock && nocow)
1330 btrfs_end_nocow_write(root);
1331 goto error;
1332 }
1333 cow_start = (u64)-1;
1334 }
1335
1336 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337 struct extent_map *em;
1338 struct extent_map_tree *em_tree;
1339 em_tree = &BTRFS_I(inode)->extent_tree;
1340 em = alloc_extent_map();
1341 BUG_ON(!em); /* -ENOMEM */
1342 em->start = cur_offset;
1343 em->orig_start = found_key.offset - extent_offset;
1344 em->len = num_bytes;
1345 em->block_len = num_bytes;
1346 em->block_start = disk_bytenr;
1347 em->orig_block_len = disk_num_bytes;
1348 em->ram_bytes = ram_bytes;
1349 em->bdev = root->fs_info->fs_devices->latest_bdev;
1350 em->mod_start = em->start;
1351 em->mod_len = em->len;
1352 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1353 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1354 em->generation = -1;
1355 while (1) {
1356 write_lock(&em_tree->lock);
1357 ret = add_extent_mapping(em_tree, em, 1);
1358 write_unlock(&em_tree->lock);
1359 if (ret != -EEXIST) {
1360 free_extent_map(em);
1361 break;
1362 }
1363 btrfs_drop_extent_cache(inode, em->start,
1364 em->start + em->len - 1, 0);
1365 }
1366 type = BTRFS_ORDERED_PREALLOC;
1367 } else {
1368 type = BTRFS_ORDERED_NOCOW;
1369 }
1370
1371 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1372 num_bytes, num_bytes, type);
1373 BUG_ON(ret); /* -ENOMEM */
1374
1375 if (root->root_key.objectid ==
1376 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1377 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1378 num_bytes);
1379 if (ret) {
1380 if (!nolock && nocow)
1381 btrfs_end_nocow_write(root);
1382 goto error;
1383 }
1384 }
1385
1386 extent_clear_unlock_delalloc(inode, cur_offset,
1387 cur_offset + num_bytes - 1,
1388 locked_page, EXTENT_LOCKED |
1389 EXTENT_DELALLOC, PAGE_UNLOCK |
1390 PAGE_SET_PRIVATE2);
1391 if (!nolock && nocow)
1392 btrfs_end_nocow_write(root);
1393 cur_offset = extent_end;
1394 if (cur_offset > end)
1395 break;
1396 }
1397 btrfs_release_path(path);
1398
1399 if (cur_offset <= end && cow_start == (u64)-1) {
1400 cow_start = cur_offset;
1401 cur_offset = end;
1402 }
1403
1404 if (cow_start != (u64)-1) {
1405 ret = cow_file_range(inode, locked_page, cow_start, end,
1406 page_started, nr_written, 1);
1407 if (ret)
1408 goto error;
1409 }
1410
1411 error:
1412 err = btrfs_end_transaction(trans, root);
1413 if (!ret)
1414 ret = err;
1415
1416 if (ret && cur_offset < end)
1417 extent_clear_unlock_delalloc(inode, cur_offset, end,
1418 locked_page, EXTENT_LOCKED |
1419 EXTENT_DELALLOC | EXTENT_DEFRAG |
1420 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1421 PAGE_CLEAR_DIRTY |
1422 PAGE_SET_WRITEBACK |
1423 PAGE_END_WRITEBACK);
1424 btrfs_free_path(path);
1425 return ret;
1426 }
1427
1428 /*
1429 * extent_io.c call back to do delayed allocation processing
1430 */
1431 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1432 u64 start, u64 end, int *page_started,
1433 unsigned long *nr_written)
1434 {
1435 int ret;
1436 struct btrfs_root *root = BTRFS_I(inode)->root;
1437
1438 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1439 ret = run_delalloc_nocow(inode, locked_page, start, end,
1440 page_started, 1, nr_written);
1441 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1442 ret = run_delalloc_nocow(inode, locked_page, start, end,
1443 page_started, 0, nr_written);
1444 } else if (!btrfs_test_opt(root, COMPRESS) &&
1445 !(BTRFS_I(inode)->force_compress) &&
1446 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1447 ret = cow_file_range(inode, locked_page, start, end,
1448 page_started, nr_written, 1);
1449 } else {
1450 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1451 &BTRFS_I(inode)->runtime_flags);
1452 ret = cow_file_range_async(inode, locked_page, start, end,
1453 page_started, nr_written);
1454 }
1455 return ret;
1456 }
1457
1458 static void btrfs_split_extent_hook(struct inode *inode,
1459 struct extent_state *orig, u64 split)
1460 {
1461 /* not delalloc, ignore it */
1462 if (!(orig->state & EXTENT_DELALLOC))
1463 return;
1464
1465 spin_lock(&BTRFS_I(inode)->lock);
1466 BTRFS_I(inode)->outstanding_extents++;
1467 spin_unlock(&BTRFS_I(inode)->lock);
1468 }
1469
1470 /*
1471 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1472 * extents so we can keep track of new extents that are just merged onto old
1473 * extents, such as when we are doing sequential writes, so we can properly
1474 * account for the metadata space we'll need.
1475 */
1476 static void btrfs_merge_extent_hook(struct inode *inode,
1477 struct extent_state *new,
1478 struct extent_state *other)
1479 {
1480 /* not delalloc, ignore it */
1481 if (!(other->state & EXTENT_DELALLOC))
1482 return;
1483
1484 spin_lock(&BTRFS_I(inode)->lock);
1485 BTRFS_I(inode)->outstanding_extents--;
1486 spin_unlock(&BTRFS_I(inode)->lock);
1487 }
1488
1489 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1490 struct inode *inode)
1491 {
1492 spin_lock(&root->delalloc_lock);
1493 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1494 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1495 &root->delalloc_inodes);
1496 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1497 &BTRFS_I(inode)->runtime_flags);
1498 root->nr_delalloc_inodes++;
1499 if (root->nr_delalloc_inodes == 1) {
1500 spin_lock(&root->fs_info->delalloc_root_lock);
1501 BUG_ON(!list_empty(&root->delalloc_root));
1502 list_add_tail(&root->delalloc_root,
1503 &root->fs_info->delalloc_roots);
1504 spin_unlock(&root->fs_info->delalloc_root_lock);
1505 }
1506 }
1507 spin_unlock(&root->delalloc_lock);
1508 }
1509
1510 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1511 struct inode *inode)
1512 {
1513 spin_lock(&root->delalloc_lock);
1514 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1516 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1517 &BTRFS_I(inode)->runtime_flags);
1518 root->nr_delalloc_inodes--;
1519 if (!root->nr_delalloc_inodes) {
1520 spin_lock(&root->fs_info->delalloc_root_lock);
1521 BUG_ON(list_empty(&root->delalloc_root));
1522 list_del_init(&root->delalloc_root);
1523 spin_unlock(&root->fs_info->delalloc_root_lock);
1524 }
1525 }
1526 spin_unlock(&root->delalloc_lock);
1527 }
1528
1529 /*
1530 * extent_io.c set_bit_hook, used to track delayed allocation
1531 * bytes in this file, and to maintain the list of inodes that
1532 * have pending delalloc work to be done.
1533 */
1534 static void btrfs_set_bit_hook(struct inode *inode,
1535 struct extent_state *state, unsigned long *bits)
1536 {
1537
1538 /*
1539 * set_bit and clear bit hooks normally require _irqsave/restore
1540 * but in this case, we are only testing for the DELALLOC
1541 * bit, which is only set or cleared with irqs on
1542 */
1543 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1544 struct btrfs_root *root = BTRFS_I(inode)->root;
1545 u64 len = state->end + 1 - state->start;
1546 bool do_list = !btrfs_is_free_space_inode(inode);
1547
1548 if (*bits & EXTENT_FIRST_DELALLOC) {
1549 *bits &= ~EXTENT_FIRST_DELALLOC;
1550 } else {
1551 spin_lock(&BTRFS_I(inode)->lock);
1552 BTRFS_I(inode)->outstanding_extents++;
1553 spin_unlock(&BTRFS_I(inode)->lock);
1554 }
1555
1556 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1557 root->fs_info->delalloc_batch);
1558 spin_lock(&BTRFS_I(inode)->lock);
1559 BTRFS_I(inode)->delalloc_bytes += len;
1560 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1561 &BTRFS_I(inode)->runtime_flags))
1562 btrfs_add_delalloc_inodes(root, inode);
1563 spin_unlock(&BTRFS_I(inode)->lock);
1564 }
1565 }
1566
1567 /*
1568 * extent_io.c clear_bit_hook, see set_bit_hook for why
1569 */
1570 static void btrfs_clear_bit_hook(struct inode *inode,
1571 struct extent_state *state,
1572 unsigned long *bits)
1573 {
1574 /*
1575 * set_bit and clear bit hooks normally require _irqsave/restore
1576 * but in this case, we are only testing for the DELALLOC
1577 * bit, which is only set or cleared with irqs on
1578 */
1579 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1580 struct btrfs_root *root = BTRFS_I(inode)->root;
1581 u64 len = state->end + 1 - state->start;
1582 bool do_list = !btrfs_is_free_space_inode(inode);
1583
1584 if (*bits & EXTENT_FIRST_DELALLOC) {
1585 *bits &= ~EXTENT_FIRST_DELALLOC;
1586 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1587 spin_lock(&BTRFS_I(inode)->lock);
1588 BTRFS_I(inode)->outstanding_extents--;
1589 spin_unlock(&BTRFS_I(inode)->lock);
1590 }
1591
1592 /*
1593 * We don't reserve metadata space for space cache inodes so we
1594 * don't need to call dellalloc_release_metadata if there is an
1595 * error.
1596 */
1597 if (*bits & EXTENT_DO_ACCOUNTING &&
1598 root != root->fs_info->tree_root)
1599 btrfs_delalloc_release_metadata(inode, len);
1600
1601 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1602 && do_list && !(state->state & EXTENT_NORESERVE))
1603 btrfs_free_reserved_data_space(inode, len);
1604
1605 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1606 root->fs_info->delalloc_batch);
1607 spin_lock(&BTRFS_I(inode)->lock);
1608 BTRFS_I(inode)->delalloc_bytes -= len;
1609 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1610 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1611 &BTRFS_I(inode)->runtime_flags))
1612 btrfs_del_delalloc_inode(root, inode);
1613 spin_unlock(&BTRFS_I(inode)->lock);
1614 }
1615 }
1616
1617 /*
1618 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1619 * we don't create bios that span stripes or chunks
1620 */
1621 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1622 size_t size, struct bio *bio,
1623 unsigned long bio_flags)
1624 {
1625 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1626 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1627 u64 length = 0;
1628 u64 map_length;
1629 int ret;
1630
1631 if (bio_flags & EXTENT_BIO_COMPRESSED)
1632 return 0;
1633
1634 length = bio->bi_iter.bi_size;
1635 map_length = length;
1636 ret = btrfs_map_block(root->fs_info, rw, logical,
1637 &map_length, NULL, 0);
1638 /* Will always return 0 with map_multi == NULL */
1639 BUG_ON(ret < 0);
1640 if (map_length < length + size)
1641 return 1;
1642 return 0;
1643 }
1644
1645 /*
1646 * in order to insert checksums into the metadata in large chunks,
1647 * we wait until bio submission time. All the pages in the bio are
1648 * checksummed and sums are attached onto the ordered extent record.
1649 *
1650 * At IO completion time the cums attached on the ordered extent record
1651 * are inserted into the btree
1652 */
1653 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1654 struct bio *bio, int mirror_num,
1655 unsigned long bio_flags,
1656 u64 bio_offset)
1657 {
1658 struct btrfs_root *root = BTRFS_I(inode)->root;
1659 int ret = 0;
1660
1661 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1662 BUG_ON(ret); /* -ENOMEM */
1663 return 0;
1664 }
1665
1666 /*
1667 * in order to insert checksums into the metadata in large chunks,
1668 * we wait until bio submission time. All the pages in the bio are
1669 * checksummed and sums are attached onto the ordered extent record.
1670 *
1671 * At IO completion time the cums attached on the ordered extent record
1672 * are inserted into the btree
1673 */
1674 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1675 int mirror_num, unsigned long bio_flags,
1676 u64 bio_offset)
1677 {
1678 struct btrfs_root *root = BTRFS_I(inode)->root;
1679 int ret;
1680
1681 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1682 if (ret)
1683 bio_endio(bio, ret);
1684 return ret;
1685 }
1686
1687 /*
1688 * extent_io.c submission hook. This does the right thing for csum calculation
1689 * on write, or reading the csums from the tree before a read
1690 */
1691 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1692 int mirror_num, unsigned long bio_flags,
1693 u64 bio_offset)
1694 {
1695 struct btrfs_root *root = BTRFS_I(inode)->root;
1696 int ret = 0;
1697 int skip_sum;
1698 int metadata = 0;
1699 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1700
1701 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1702
1703 if (btrfs_is_free_space_inode(inode))
1704 metadata = 2;
1705
1706 if (!(rw & REQ_WRITE)) {
1707 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1708 if (ret)
1709 goto out;
1710
1711 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1712 ret = btrfs_submit_compressed_read(inode, bio,
1713 mirror_num,
1714 bio_flags);
1715 goto out;
1716 } else if (!skip_sum) {
1717 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1718 if (ret)
1719 goto out;
1720 }
1721 goto mapit;
1722 } else if (async && !skip_sum) {
1723 /* csum items have already been cloned */
1724 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1725 goto mapit;
1726 /* we're doing a write, do the async checksumming */
1727 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1728 inode, rw, bio, mirror_num,
1729 bio_flags, bio_offset,
1730 __btrfs_submit_bio_start,
1731 __btrfs_submit_bio_done);
1732 goto out;
1733 } else if (!skip_sum) {
1734 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1735 if (ret)
1736 goto out;
1737 }
1738
1739 mapit:
1740 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1741
1742 out:
1743 if (ret < 0)
1744 bio_endio(bio, ret);
1745 return ret;
1746 }
1747
1748 /*
1749 * given a list of ordered sums record them in the inode. This happens
1750 * at IO completion time based on sums calculated at bio submission time.
1751 */
1752 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1753 struct inode *inode, u64 file_offset,
1754 struct list_head *list)
1755 {
1756 struct btrfs_ordered_sum *sum;
1757
1758 list_for_each_entry(sum, list, list) {
1759 trans->adding_csums = 1;
1760 btrfs_csum_file_blocks(trans,
1761 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1762 trans->adding_csums = 0;
1763 }
1764 return 0;
1765 }
1766
1767 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1768 struct extent_state **cached_state)
1769 {
1770 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1771 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1772 cached_state, GFP_NOFS);
1773 }
1774
1775 /* see btrfs_writepage_start_hook for details on why this is required */
1776 struct btrfs_writepage_fixup {
1777 struct page *page;
1778 struct btrfs_work work;
1779 };
1780
1781 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1782 {
1783 struct btrfs_writepage_fixup *fixup;
1784 struct btrfs_ordered_extent *ordered;
1785 struct extent_state *cached_state = NULL;
1786 struct page *page;
1787 struct inode *inode;
1788 u64 page_start;
1789 u64 page_end;
1790 int ret;
1791
1792 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1793 page = fixup->page;
1794 again:
1795 lock_page(page);
1796 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1797 ClearPageChecked(page);
1798 goto out_page;
1799 }
1800
1801 inode = page->mapping->host;
1802 page_start = page_offset(page);
1803 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1804
1805 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1806 &cached_state);
1807
1808 /* already ordered? We're done */
1809 if (PagePrivate2(page))
1810 goto out;
1811
1812 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1813 if (ordered) {
1814 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1815 page_end, &cached_state, GFP_NOFS);
1816 unlock_page(page);
1817 btrfs_start_ordered_extent(inode, ordered, 1);
1818 btrfs_put_ordered_extent(ordered);
1819 goto again;
1820 }
1821
1822 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1823 if (ret) {
1824 mapping_set_error(page->mapping, ret);
1825 end_extent_writepage(page, ret, page_start, page_end);
1826 ClearPageChecked(page);
1827 goto out;
1828 }
1829
1830 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1831 ClearPageChecked(page);
1832 set_page_dirty(page);
1833 out:
1834 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1835 &cached_state, GFP_NOFS);
1836 out_page:
1837 unlock_page(page);
1838 page_cache_release(page);
1839 kfree(fixup);
1840 }
1841
1842 /*
1843 * There are a few paths in the higher layers of the kernel that directly
1844 * set the page dirty bit without asking the filesystem if it is a
1845 * good idea. This causes problems because we want to make sure COW
1846 * properly happens and the data=ordered rules are followed.
1847 *
1848 * In our case any range that doesn't have the ORDERED bit set
1849 * hasn't been properly setup for IO. We kick off an async process
1850 * to fix it up. The async helper will wait for ordered extents, set
1851 * the delalloc bit and make it safe to write the page.
1852 */
1853 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1854 {
1855 struct inode *inode = page->mapping->host;
1856 struct btrfs_writepage_fixup *fixup;
1857 struct btrfs_root *root = BTRFS_I(inode)->root;
1858
1859 /* this page is properly in the ordered list */
1860 if (TestClearPagePrivate2(page))
1861 return 0;
1862
1863 if (PageChecked(page))
1864 return -EAGAIN;
1865
1866 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1867 if (!fixup)
1868 return -EAGAIN;
1869
1870 SetPageChecked(page);
1871 page_cache_get(page);
1872 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1873 fixup->page = page;
1874 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1875 return -EBUSY;
1876 }
1877
1878 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1879 struct inode *inode, u64 file_pos,
1880 u64 disk_bytenr, u64 disk_num_bytes,
1881 u64 num_bytes, u64 ram_bytes,
1882 u8 compression, u8 encryption,
1883 u16 other_encoding, int extent_type)
1884 {
1885 struct btrfs_root *root = BTRFS_I(inode)->root;
1886 struct btrfs_file_extent_item *fi;
1887 struct btrfs_path *path;
1888 struct extent_buffer *leaf;
1889 struct btrfs_key ins;
1890 int extent_inserted = 0;
1891 int ret;
1892
1893 path = btrfs_alloc_path();
1894 if (!path)
1895 return -ENOMEM;
1896
1897 /*
1898 * we may be replacing one extent in the tree with another.
1899 * The new extent is pinned in the extent map, and we don't want
1900 * to drop it from the cache until it is completely in the btree.
1901 *
1902 * So, tell btrfs_drop_extents to leave this extent in the cache.
1903 * the caller is expected to unpin it and allow it to be merged
1904 * with the others.
1905 */
1906 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1907 file_pos + num_bytes, NULL, 0,
1908 1, sizeof(*fi), &extent_inserted);
1909 if (ret)
1910 goto out;
1911
1912 if (!extent_inserted) {
1913 ins.objectid = btrfs_ino(inode);
1914 ins.offset = file_pos;
1915 ins.type = BTRFS_EXTENT_DATA_KEY;
1916
1917 path->leave_spinning = 1;
1918 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1919 sizeof(*fi));
1920 if (ret)
1921 goto out;
1922 }
1923 leaf = path->nodes[0];
1924 fi = btrfs_item_ptr(leaf, path->slots[0],
1925 struct btrfs_file_extent_item);
1926 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1927 btrfs_set_file_extent_type(leaf, fi, extent_type);
1928 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1929 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1930 btrfs_set_file_extent_offset(leaf, fi, 0);
1931 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1932 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1933 btrfs_set_file_extent_compression(leaf, fi, compression);
1934 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1935 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1936
1937 btrfs_mark_buffer_dirty(leaf);
1938 btrfs_release_path(path);
1939
1940 inode_add_bytes(inode, num_bytes);
1941
1942 ins.objectid = disk_bytenr;
1943 ins.offset = disk_num_bytes;
1944 ins.type = BTRFS_EXTENT_ITEM_KEY;
1945 ret = btrfs_alloc_reserved_file_extent(trans, root,
1946 root->root_key.objectid,
1947 btrfs_ino(inode), file_pos, &ins);
1948 out:
1949 btrfs_free_path(path);
1950
1951 return ret;
1952 }
1953
1954 /* snapshot-aware defrag */
1955 struct sa_defrag_extent_backref {
1956 struct rb_node node;
1957 struct old_sa_defrag_extent *old;
1958 u64 root_id;
1959 u64 inum;
1960 u64 file_pos;
1961 u64 extent_offset;
1962 u64 num_bytes;
1963 u64 generation;
1964 };
1965
1966 struct old_sa_defrag_extent {
1967 struct list_head list;
1968 struct new_sa_defrag_extent *new;
1969
1970 u64 extent_offset;
1971 u64 bytenr;
1972 u64 offset;
1973 u64 len;
1974 int count;
1975 };
1976
1977 struct new_sa_defrag_extent {
1978 struct rb_root root;
1979 struct list_head head;
1980 struct btrfs_path *path;
1981 struct inode *inode;
1982 u64 file_pos;
1983 u64 len;
1984 u64 bytenr;
1985 u64 disk_len;
1986 u8 compress_type;
1987 };
1988
1989 static int backref_comp(struct sa_defrag_extent_backref *b1,
1990 struct sa_defrag_extent_backref *b2)
1991 {
1992 if (b1->root_id < b2->root_id)
1993 return -1;
1994 else if (b1->root_id > b2->root_id)
1995 return 1;
1996
1997 if (b1->inum < b2->inum)
1998 return -1;
1999 else if (b1->inum > b2->inum)
2000 return 1;
2001
2002 if (b1->file_pos < b2->file_pos)
2003 return -1;
2004 else if (b1->file_pos > b2->file_pos)
2005 return 1;
2006
2007 /*
2008 * [------------------------------] ===> (a range of space)
2009 * |<--->| |<---->| =============> (fs/file tree A)
2010 * |<---------------------------->| ===> (fs/file tree B)
2011 *
2012 * A range of space can refer to two file extents in one tree while
2013 * refer to only one file extent in another tree.
2014 *
2015 * So we may process a disk offset more than one time(two extents in A)
2016 * and locate at the same extent(one extent in B), then insert two same
2017 * backrefs(both refer to the extent in B).
2018 */
2019 return 0;
2020 }
2021
2022 static void backref_insert(struct rb_root *root,
2023 struct sa_defrag_extent_backref *backref)
2024 {
2025 struct rb_node **p = &root->rb_node;
2026 struct rb_node *parent = NULL;
2027 struct sa_defrag_extent_backref *entry;
2028 int ret;
2029
2030 while (*p) {
2031 parent = *p;
2032 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2033
2034 ret = backref_comp(backref, entry);
2035 if (ret < 0)
2036 p = &(*p)->rb_left;
2037 else
2038 p = &(*p)->rb_right;
2039 }
2040
2041 rb_link_node(&backref->node, parent, p);
2042 rb_insert_color(&backref->node, root);
2043 }
2044
2045 /*
2046 * Note the backref might has changed, and in this case we just return 0.
2047 */
2048 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2049 void *ctx)
2050 {
2051 struct btrfs_file_extent_item *extent;
2052 struct btrfs_fs_info *fs_info;
2053 struct old_sa_defrag_extent *old = ctx;
2054 struct new_sa_defrag_extent *new = old->new;
2055 struct btrfs_path *path = new->path;
2056 struct btrfs_key key;
2057 struct btrfs_root *root;
2058 struct sa_defrag_extent_backref *backref;
2059 struct extent_buffer *leaf;
2060 struct inode *inode = new->inode;
2061 int slot;
2062 int ret;
2063 u64 extent_offset;
2064 u64 num_bytes;
2065
2066 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2067 inum == btrfs_ino(inode))
2068 return 0;
2069
2070 key.objectid = root_id;
2071 key.type = BTRFS_ROOT_ITEM_KEY;
2072 key.offset = (u64)-1;
2073
2074 fs_info = BTRFS_I(inode)->root->fs_info;
2075 root = btrfs_read_fs_root_no_name(fs_info, &key);
2076 if (IS_ERR(root)) {
2077 if (PTR_ERR(root) == -ENOENT)
2078 return 0;
2079 WARN_ON(1);
2080 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2081 inum, offset, root_id);
2082 return PTR_ERR(root);
2083 }
2084
2085 key.objectid = inum;
2086 key.type = BTRFS_EXTENT_DATA_KEY;
2087 if (offset > (u64)-1 << 32)
2088 key.offset = 0;
2089 else
2090 key.offset = offset;
2091
2092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2093 if (WARN_ON(ret < 0))
2094 return ret;
2095 ret = 0;
2096
2097 while (1) {
2098 cond_resched();
2099
2100 leaf = path->nodes[0];
2101 slot = path->slots[0];
2102
2103 if (slot >= btrfs_header_nritems(leaf)) {
2104 ret = btrfs_next_leaf(root, path);
2105 if (ret < 0) {
2106 goto out;
2107 } else if (ret > 0) {
2108 ret = 0;
2109 goto out;
2110 }
2111 continue;
2112 }
2113
2114 path->slots[0]++;
2115
2116 btrfs_item_key_to_cpu(leaf, &key, slot);
2117
2118 if (key.objectid > inum)
2119 goto out;
2120
2121 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2122 continue;
2123
2124 extent = btrfs_item_ptr(leaf, slot,
2125 struct btrfs_file_extent_item);
2126
2127 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2128 continue;
2129
2130 /*
2131 * 'offset' refers to the exact key.offset,
2132 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2133 * (key.offset - extent_offset).
2134 */
2135 if (key.offset != offset)
2136 continue;
2137
2138 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2140
2141 if (extent_offset >= old->extent_offset + old->offset +
2142 old->len || extent_offset + num_bytes <=
2143 old->extent_offset + old->offset)
2144 continue;
2145 break;
2146 }
2147
2148 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2149 if (!backref) {
2150 ret = -ENOENT;
2151 goto out;
2152 }
2153
2154 backref->root_id = root_id;
2155 backref->inum = inum;
2156 backref->file_pos = offset;
2157 backref->num_bytes = num_bytes;
2158 backref->extent_offset = extent_offset;
2159 backref->generation = btrfs_file_extent_generation(leaf, extent);
2160 backref->old = old;
2161 backref_insert(&new->root, backref);
2162 old->count++;
2163 out:
2164 btrfs_release_path(path);
2165 WARN_ON(ret);
2166 return ret;
2167 }
2168
2169 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2170 struct new_sa_defrag_extent *new)
2171 {
2172 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2173 struct old_sa_defrag_extent *old, *tmp;
2174 int ret;
2175
2176 new->path = path;
2177
2178 list_for_each_entry_safe(old, tmp, &new->head, list) {
2179 ret = iterate_inodes_from_logical(old->bytenr +
2180 old->extent_offset, fs_info,
2181 path, record_one_backref,
2182 old);
2183 if (ret < 0 && ret != -ENOENT)
2184 return false;
2185
2186 /* no backref to be processed for this extent */
2187 if (!old->count) {
2188 list_del(&old->list);
2189 kfree(old);
2190 }
2191 }
2192
2193 if (list_empty(&new->head))
2194 return false;
2195
2196 return true;
2197 }
2198
2199 static int relink_is_mergable(struct extent_buffer *leaf,
2200 struct btrfs_file_extent_item *fi,
2201 struct new_sa_defrag_extent *new)
2202 {
2203 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2204 return 0;
2205
2206 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2207 return 0;
2208
2209 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2210 return 0;
2211
2212 if (btrfs_file_extent_encryption(leaf, fi) ||
2213 btrfs_file_extent_other_encoding(leaf, fi))
2214 return 0;
2215
2216 return 1;
2217 }
2218
2219 /*
2220 * Note the backref might has changed, and in this case we just return 0.
2221 */
2222 static noinline int relink_extent_backref(struct btrfs_path *path,
2223 struct sa_defrag_extent_backref *prev,
2224 struct sa_defrag_extent_backref *backref)
2225 {
2226 struct btrfs_file_extent_item *extent;
2227 struct btrfs_file_extent_item *item;
2228 struct btrfs_ordered_extent *ordered;
2229 struct btrfs_trans_handle *trans;
2230 struct btrfs_fs_info *fs_info;
2231 struct btrfs_root *root;
2232 struct btrfs_key key;
2233 struct extent_buffer *leaf;
2234 struct old_sa_defrag_extent *old = backref->old;
2235 struct new_sa_defrag_extent *new = old->new;
2236 struct inode *src_inode = new->inode;
2237 struct inode *inode;
2238 struct extent_state *cached = NULL;
2239 int ret = 0;
2240 u64 start;
2241 u64 len;
2242 u64 lock_start;
2243 u64 lock_end;
2244 bool merge = false;
2245 int index;
2246
2247 if (prev && prev->root_id == backref->root_id &&
2248 prev->inum == backref->inum &&
2249 prev->file_pos + prev->num_bytes == backref->file_pos)
2250 merge = true;
2251
2252 /* step 1: get root */
2253 key.objectid = backref->root_id;
2254 key.type = BTRFS_ROOT_ITEM_KEY;
2255 key.offset = (u64)-1;
2256
2257 fs_info = BTRFS_I(src_inode)->root->fs_info;
2258 index = srcu_read_lock(&fs_info->subvol_srcu);
2259
2260 root = btrfs_read_fs_root_no_name(fs_info, &key);
2261 if (IS_ERR(root)) {
2262 srcu_read_unlock(&fs_info->subvol_srcu, index);
2263 if (PTR_ERR(root) == -ENOENT)
2264 return 0;
2265 return PTR_ERR(root);
2266 }
2267
2268 if (btrfs_root_readonly(root)) {
2269 srcu_read_unlock(&fs_info->subvol_srcu, index);
2270 return 0;
2271 }
2272
2273 /* step 2: get inode */
2274 key.objectid = backref->inum;
2275 key.type = BTRFS_INODE_ITEM_KEY;
2276 key.offset = 0;
2277
2278 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2279 if (IS_ERR(inode)) {
2280 srcu_read_unlock(&fs_info->subvol_srcu, index);
2281 return 0;
2282 }
2283
2284 srcu_read_unlock(&fs_info->subvol_srcu, index);
2285
2286 /* step 3: relink backref */
2287 lock_start = backref->file_pos;
2288 lock_end = backref->file_pos + backref->num_bytes - 1;
2289 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2290 0, &cached);
2291
2292 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2293 if (ordered) {
2294 btrfs_put_ordered_extent(ordered);
2295 goto out_unlock;
2296 }
2297
2298 trans = btrfs_join_transaction(root);
2299 if (IS_ERR(trans)) {
2300 ret = PTR_ERR(trans);
2301 goto out_unlock;
2302 }
2303
2304 key.objectid = backref->inum;
2305 key.type = BTRFS_EXTENT_DATA_KEY;
2306 key.offset = backref->file_pos;
2307
2308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2309 if (ret < 0) {
2310 goto out_free_path;
2311 } else if (ret > 0) {
2312 ret = 0;
2313 goto out_free_path;
2314 }
2315
2316 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2317 struct btrfs_file_extent_item);
2318
2319 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2320 backref->generation)
2321 goto out_free_path;
2322
2323 btrfs_release_path(path);
2324
2325 start = backref->file_pos;
2326 if (backref->extent_offset < old->extent_offset + old->offset)
2327 start += old->extent_offset + old->offset -
2328 backref->extent_offset;
2329
2330 len = min(backref->extent_offset + backref->num_bytes,
2331 old->extent_offset + old->offset + old->len);
2332 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2333
2334 ret = btrfs_drop_extents(trans, root, inode, start,
2335 start + len, 1);
2336 if (ret)
2337 goto out_free_path;
2338 again:
2339 key.objectid = btrfs_ino(inode);
2340 key.type = BTRFS_EXTENT_DATA_KEY;
2341 key.offset = start;
2342
2343 path->leave_spinning = 1;
2344 if (merge) {
2345 struct btrfs_file_extent_item *fi;
2346 u64 extent_len;
2347 struct btrfs_key found_key;
2348
2349 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2350 if (ret < 0)
2351 goto out_free_path;
2352
2353 path->slots[0]--;
2354 leaf = path->nodes[0];
2355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2356
2357 fi = btrfs_item_ptr(leaf, path->slots[0],
2358 struct btrfs_file_extent_item);
2359 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2360
2361 if (extent_len + found_key.offset == start &&
2362 relink_is_mergable(leaf, fi, new)) {
2363 btrfs_set_file_extent_num_bytes(leaf, fi,
2364 extent_len + len);
2365 btrfs_mark_buffer_dirty(leaf);
2366 inode_add_bytes(inode, len);
2367
2368 ret = 1;
2369 goto out_free_path;
2370 } else {
2371 merge = false;
2372 btrfs_release_path(path);
2373 goto again;
2374 }
2375 }
2376
2377 ret = btrfs_insert_empty_item(trans, root, path, &key,
2378 sizeof(*extent));
2379 if (ret) {
2380 btrfs_abort_transaction(trans, root, ret);
2381 goto out_free_path;
2382 }
2383
2384 leaf = path->nodes[0];
2385 item = btrfs_item_ptr(leaf, path->slots[0],
2386 struct btrfs_file_extent_item);
2387 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2388 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2389 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2390 btrfs_set_file_extent_num_bytes(leaf, item, len);
2391 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2392 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2393 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2394 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2395 btrfs_set_file_extent_encryption(leaf, item, 0);
2396 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2397
2398 btrfs_mark_buffer_dirty(leaf);
2399 inode_add_bytes(inode, len);
2400 btrfs_release_path(path);
2401
2402 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2403 new->disk_len, 0,
2404 backref->root_id, backref->inum,
2405 new->file_pos, 0); /* start - extent_offset */
2406 if (ret) {
2407 btrfs_abort_transaction(trans, root, ret);
2408 goto out_free_path;
2409 }
2410
2411 ret = 1;
2412 out_free_path:
2413 btrfs_release_path(path);
2414 path->leave_spinning = 0;
2415 btrfs_end_transaction(trans, root);
2416 out_unlock:
2417 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2418 &cached, GFP_NOFS);
2419 iput(inode);
2420 return ret;
2421 }
2422
2423 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2424 {
2425 struct old_sa_defrag_extent *old, *tmp;
2426
2427 if (!new)
2428 return;
2429
2430 list_for_each_entry_safe(old, tmp, &new->head, list) {
2431 list_del(&old->list);
2432 kfree(old);
2433 }
2434 kfree(new);
2435 }
2436
2437 static void relink_file_extents(struct new_sa_defrag_extent *new)
2438 {
2439 struct btrfs_path *path;
2440 struct sa_defrag_extent_backref *backref;
2441 struct sa_defrag_extent_backref *prev = NULL;
2442 struct inode *inode;
2443 struct btrfs_root *root;
2444 struct rb_node *node;
2445 int ret;
2446
2447 inode = new->inode;
2448 root = BTRFS_I(inode)->root;
2449
2450 path = btrfs_alloc_path();
2451 if (!path)
2452 return;
2453
2454 if (!record_extent_backrefs(path, new)) {
2455 btrfs_free_path(path);
2456 goto out;
2457 }
2458 btrfs_release_path(path);
2459
2460 while (1) {
2461 node = rb_first(&new->root);
2462 if (!node)
2463 break;
2464 rb_erase(node, &new->root);
2465
2466 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2467
2468 ret = relink_extent_backref(path, prev, backref);
2469 WARN_ON(ret < 0);
2470
2471 kfree(prev);
2472
2473 if (ret == 1)
2474 prev = backref;
2475 else
2476 prev = NULL;
2477 cond_resched();
2478 }
2479 kfree(prev);
2480
2481 btrfs_free_path(path);
2482 out:
2483 free_sa_defrag_extent(new);
2484
2485 atomic_dec(&root->fs_info->defrag_running);
2486 wake_up(&root->fs_info->transaction_wait);
2487 }
2488
2489 static struct new_sa_defrag_extent *
2490 record_old_file_extents(struct inode *inode,
2491 struct btrfs_ordered_extent *ordered)
2492 {
2493 struct btrfs_root *root = BTRFS_I(inode)->root;
2494 struct btrfs_path *path;
2495 struct btrfs_key key;
2496 struct old_sa_defrag_extent *old;
2497 struct new_sa_defrag_extent *new;
2498 int ret;
2499
2500 new = kmalloc(sizeof(*new), GFP_NOFS);
2501 if (!new)
2502 return NULL;
2503
2504 new->inode = inode;
2505 new->file_pos = ordered->file_offset;
2506 new->len = ordered->len;
2507 new->bytenr = ordered->start;
2508 new->disk_len = ordered->disk_len;
2509 new->compress_type = ordered->compress_type;
2510 new->root = RB_ROOT;
2511 INIT_LIST_HEAD(&new->head);
2512
2513 path = btrfs_alloc_path();
2514 if (!path)
2515 goto out_kfree;
2516
2517 key.objectid = btrfs_ino(inode);
2518 key.type = BTRFS_EXTENT_DATA_KEY;
2519 key.offset = new->file_pos;
2520
2521 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2522 if (ret < 0)
2523 goto out_free_path;
2524 if (ret > 0 && path->slots[0] > 0)
2525 path->slots[0]--;
2526
2527 /* find out all the old extents for the file range */
2528 while (1) {
2529 struct btrfs_file_extent_item *extent;
2530 struct extent_buffer *l;
2531 int slot;
2532 u64 num_bytes;
2533 u64 offset;
2534 u64 end;
2535 u64 disk_bytenr;
2536 u64 extent_offset;
2537
2538 l = path->nodes[0];
2539 slot = path->slots[0];
2540
2541 if (slot >= btrfs_header_nritems(l)) {
2542 ret = btrfs_next_leaf(root, path);
2543 if (ret < 0)
2544 goto out_free_path;
2545 else if (ret > 0)
2546 break;
2547 continue;
2548 }
2549
2550 btrfs_item_key_to_cpu(l, &key, slot);
2551
2552 if (key.objectid != btrfs_ino(inode))
2553 break;
2554 if (key.type != BTRFS_EXTENT_DATA_KEY)
2555 break;
2556 if (key.offset >= new->file_pos + new->len)
2557 break;
2558
2559 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2560
2561 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2562 if (key.offset + num_bytes < new->file_pos)
2563 goto next;
2564
2565 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2566 if (!disk_bytenr)
2567 goto next;
2568
2569 extent_offset = btrfs_file_extent_offset(l, extent);
2570
2571 old = kmalloc(sizeof(*old), GFP_NOFS);
2572 if (!old)
2573 goto out_free_path;
2574
2575 offset = max(new->file_pos, key.offset);
2576 end = min(new->file_pos + new->len, key.offset + num_bytes);
2577
2578 old->bytenr = disk_bytenr;
2579 old->extent_offset = extent_offset;
2580 old->offset = offset - key.offset;
2581 old->len = end - offset;
2582 old->new = new;
2583 old->count = 0;
2584 list_add_tail(&old->list, &new->head);
2585 next:
2586 path->slots[0]++;
2587 cond_resched();
2588 }
2589
2590 btrfs_free_path(path);
2591 atomic_inc(&root->fs_info->defrag_running);
2592
2593 return new;
2594
2595 out_free_path:
2596 btrfs_free_path(path);
2597 out_kfree:
2598 free_sa_defrag_extent(new);
2599 return NULL;
2600 }
2601
2602 /* as ordered data IO finishes, this gets called so we can finish
2603 * an ordered extent if the range of bytes in the file it covers are
2604 * fully written.
2605 */
2606 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2607 {
2608 struct inode *inode = ordered_extent->inode;
2609 struct btrfs_root *root = BTRFS_I(inode)->root;
2610 struct btrfs_trans_handle *trans = NULL;
2611 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2612 struct extent_state *cached_state = NULL;
2613 struct new_sa_defrag_extent *new = NULL;
2614 int compress_type = 0;
2615 int ret = 0;
2616 u64 logical_len = ordered_extent->len;
2617 bool nolock;
2618 bool truncated = false;
2619
2620 nolock = btrfs_is_free_space_inode(inode);
2621
2622 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2623 ret = -EIO;
2624 goto out;
2625 }
2626
2627 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2628 truncated = true;
2629 logical_len = ordered_extent->truncated_len;
2630 /* Truncated the entire extent, don't bother adding */
2631 if (!logical_len)
2632 goto out;
2633 }
2634
2635 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2636 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2637 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2638 if (nolock)
2639 trans = btrfs_join_transaction_nolock(root);
2640 else
2641 trans = btrfs_join_transaction(root);
2642 if (IS_ERR(trans)) {
2643 ret = PTR_ERR(trans);
2644 trans = NULL;
2645 goto out;
2646 }
2647 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2648 ret = btrfs_update_inode_fallback(trans, root, inode);
2649 if (ret) /* -ENOMEM or corruption */
2650 btrfs_abort_transaction(trans, root, ret);
2651 goto out;
2652 }
2653
2654 lock_extent_bits(io_tree, ordered_extent->file_offset,
2655 ordered_extent->file_offset + ordered_extent->len - 1,
2656 0, &cached_state);
2657
2658 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2659 ordered_extent->file_offset + ordered_extent->len - 1,
2660 EXTENT_DEFRAG, 1, cached_state);
2661 if (ret) {
2662 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2663 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2664 /* the inode is shared */
2665 new = record_old_file_extents(inode, ordered_extent);
2666
2667 clear_extent_bit(io_tree, ordered_extent->file_offset,
2668 ordered_extent->file_offset + ordered_extent->len - 1,
2669 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2670 }
2671
2672 if (nolock)
2673 trans = btrfs_join_transaction_nolock(root);
2674 else
2675 trans = btrfs_join_transaction(root);
2676 if (IS_ERR(trans)) {
2677 ret = PTR_ERR(trans);
2678 trans = NULL;
2679 goto out_unlock;
2680 }
2681 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2682
2683 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2684 compress_type = ordered_extent->compress_type;
2685 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2686 BUG_ON(compress_type);
2687 ret = btrfs_mark_extent_written(trans, inode,
2688 ordered_extent->file_offset,
2689 ordered_extent->file_offset +
2690 logical_len);
2691 } else {
2692 BUG_ON(root == root->fs_info->tree_root);
2693 ret = insert_reserved_file_extent(trans, inode,
2694 ordered_extent->file_offset,
2695 ordered_extent->start,
2696 ordered_extent->disk_len,
2697 logical_len, logical_len,
2698 compress_type, 0, 0,
2699 BTRFS_FILE_EXTENT_REG);
2700 }
2701 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2702 ordered_extent->file_offset, ordered_extent->len,
2703 trans->transid);
2704 if (ret < 0) {
2705 btrfs_abort_transaction(trans, root, ret);
2706 goto out_unlock;
2707 }
2708
2709 add_pending_csums(trans, inode, ordered_extent->file_offset,
2710 &ordered_extent->list);
2711
2712 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2713 ret = btrfs_update_inode_fallback(trans, root, inode);
2714 if (ret) { /* -ENOMEM or corruption */
2715 btrfs_abort_transaction(trans, root, ret);
2716 goto out_unlock;
2717 }
2718 ret = 0;
2719 out_unlock:
2720 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2721 ordered_extent->file_offset +
2722 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2723 out:
2724 if (root != root->fs_info->tree_root)
2725 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2726 if (trans)
2727 btrfs_end_transaction(trans, root);
2728
2729 if (ret || truncated) {
2730 u64 start, end;
2731
2732 if (truncated)
2733 start = ordered_extent->file_offset + logical_len;
2734 else
2735 start = ordered_extent->file_offset;
2736 end = ordered_extent->file_offset + ordered_extent->len - 1;
2737 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2738
2739 /* Drop the cache for the part of the extent we didn't write. */
2740 btrfs_drop_extent_cache(inode, start, end, 0);
2741
2742 /*
2743 * If the ordered extent had an IOERR or something else went
2744 * wrong we need to return the space for this ordered extent
2745 * back to the allocator. We only free the extent in the
2746 * truncated case if we didn't write out the extent at all.
2747 */
2748 if ((ret || !logical_len) &&
2749 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2750 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2751 btrfs_free_reserved_extent(root, ordered_extent->start,
2752 ordered_extent->disk_len);
2753 }
2754
2755
2756 /*
2757 * This needs to be done to make sure anybody waiting knows we are done
2758 * updating everything for this ordered extent.
2759 */
2760 btrfs_remove_ordered_extent(inode, ordered_extent);
2761
2762 /* for snapshot-aware defrag */
2763 if (new) {
2764 if (ret) {
2765 free_sa_defrag_extent(new);
2766 atomic_dec(&root->fs_info->defrag_running);
2767 } else {
2768 relink_file_extents(new);
2769 }
2770 }
2771
2772 /* once for us */
2773 btrfs_put_ordered_extent(ordered_extent);
2774 /* once for the tree */
2775 btrfs_put_ordered_extent(ordered_extent);
2776
2777 return ret;
2778 }
2779
2780 static void finish_ordered_fn(struct btrfs_work *work)
2781 {
2782 struct btrfs_ordered_extent *ordered_extent;
2783 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2784 btrfs_finish_ordered_io(ordered_extent);
2785 }
2786
2787 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2788 struct extent_state *state, int uptodate)
2789 {
2790 struct inode *inode = page->mapping->host;
2791 struct btrfs_root *root = BTRFS_I(inode)->root;
2792 struct btrfs_ordered_extent *ordered_extent = NULL;
2793 struct btrfs_workqueue *workers;
2794
2795 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2796
2797 ClearPagePrivate2(page);
2798 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2799 end - start + 1, uptodate))
2800 return 0;
2801
2802 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2803
2804 if (btrfs_is_free_space_inode(inode))
2805 workers = root->fs_info->endio_freespace_worker;
2806 else
2807 workers = root->fs_info->endio_write_workers;
2808 btrfs_queue_work(workers, &ordered_extent->work);
2809
2810 return 0;
2811 }
2812
2813 /*
2814 * when reads are done, we need to check csums to verify the data is correct
2815 * if there's a match, we allow the bio to finish. If not, the code in
2816 * extent_io.c will try to find good copies for us.
2817 */
2818 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2819 u64 phy_offset, struct page *page,
2820 u64 start, u64 end, int mirror)
2821 {
2822 size_t offset = start - page_offset(page);
2823 struct inode *inode = page->mapping->host;
2824 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2825 char *kaddr;
2826 struct btrfs_root *root = BTRFS_I(inode)->root;
2827 u32 csum_expected;
2828 u32 csum = ~(u32)0;
2829 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2830 DEFAULT_RATELIMIT_BURST);
2831
2832 if (PageChecked(page)) {
2833 ClearPageChecked(page);
2834 goto good;
2835 }
2836
2837 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2838 goto good;
2839
2840 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2841 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2842 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2843 GFP_NOFS);
2844 return 0;
2845 }
2846
2847 phy_offset >>= inode->i_sb->s_blocksize_bits;
2848 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2849
2850 kaddr = kmap_atomic(page);
2851 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2852 btrfs_csum_final(csum, (char *)&csum);
2853 if (csum != csum_expected)
2854 goto zeroit;
2855
2856 kunmap_atomic(kaddr);
2857 good:
2858 return 0;
2859
2860 zeroit:
2861 if (__ratelimit(&_rs))
2862 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2863 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2864 memset(kaddr + offset, 1, end - start + 1);
2865 flush_dcache_page(page);
2866 kunmap_atomic(kaddr);
2867 if (csum_expected == 0)
2868 return 0;
2869 return -EIO;
2870 }
2871
2872 struct delayed_iput {
2873 struct list_head list;
2874 struct inode *inode;
2875 };
2876
2877 /* JDM: If this is fs-wide, why can't we add a pointer to
2878 * btrfs_inode instead and avoid the allocation? */
2879 void btrfs_add_delayed_iput(struct inode *inode)
2880 {
2881 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2882 struct delayed_iput *delayed;
2883
2884 if (atomic_add_unless(&inode->i_count, -1, 1))
2885 return;
2886
2887 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2888 delayed->inode = inode;
2889
2890 spin_lock(&fs_info->delayed_iput_lock);
2891 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2892 spin_unlock(&fs_info->delayed_iput_lock);
2893 }
2894
2895 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2896 {
2897 LIST_HEAD(list);
2898 struct btrfs_fs_info *fs_info = root->fs_info;
2899 struct delayed_iput *delayed;
2900 int empty;
2901
2902 spin_lock(&fs_info->delayed_iput_lock);
2903 empty = list_empty(&fs_info->delayed_iputs);
2904 spin_unlock(&fs_info->delayed_iput_lock);
2905 if (empty)
2906 return;
2907
2908 spin_lock(&fs_info->delayed_iput_lock);
2909 list_splice_init(&fs_info->delayed_iputs, &list);
2910 spin_unlock(&fs_info->delayed_iput_lock);
2911
2912 while (!list_empty(&list)) {
2913 delayed = list_entry(list.next, struct delayed_iput, list);
2914 list_del(&delayed->list);
2915 iput(delayed->inode);
2916 kfree(delayed);
2917 }
2918 }
2919
2920 /*
2921 * This is called in transaction commit time. If there are no orphan
2922 * files in the subvolume, it removes orphan item and frees block_rsv
2923 * structure.
2924 */
2925 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2926 struct btrfs_root *root)
2927 {
2928 struct btrfs_block_rsv *block_rsv;
2929 int ret;
2930
2931 if (atomic_read(&root->orphan_inodes) ||
2932 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2933 return;
2934
2935 spin_lock(&root->orphan_lock);
2936 if (atomic_read(&root->orphan_inodes)) {
2937 spin_unlock(&root->orphan_lock);
2938 return;
2939 }
2940
2941 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2942 spin_unlock(&root->orphan_lock);
2943 return;
2944 }
2945
2946 block_rsv = root->orphan_block_rsv;
2947 root->orphan_block_rsv = NULL;
2948 spin_unlock(&root->orphan_lock);
2949
2950 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
2951 btrfs_root_refs(&root->root_item) > 0) {
2952 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2953 root->root_key.objectid);
2954 if (ret)
2955 btrfs_abort_transaction(trans, root, ret);
2956 else
2957 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
2958 &root->state);
2959 }
2960
2961 if (block_rsv) {
2962 WARN_ON(block_rsv->size > 0);
2963 btrfs_free_block_rsv(root, block_rsv);
2964 }
2965 }
2966
2967 /*
2968 * This creates an orphan entry for the given inode in case something goes
2969 * wrong in the middle of an unlink/truncate.
2970 *
2971 * NOTE: caller of this function should reserve 5 units of metadata for
2972 * this function.
2973 */
2974 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2975 {
2976 struct btrfs_root *root = BTRFS_I(inode)->root;
2977 struct btrfs_block_rsv *block_rsv = NULL;
2978 int reserve = 0;
2979 int insert = 0;
2980 int ret;
2981
2982 if (!root->orphan_block_rsv) {
2983 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2984 if (!block_rsv)
2985 return -ENOMEM;
2986 }
2987
2988 spin_lock(&root->orphan_lock);
2989 if (!root->orphan_block_rsv) {
2990 root->orphan_block_rsv = block_rsv;
2991 } else if (block_rsv) {
2992 btrfs_free_block_rsv(root, block_rsv);
2993 block_rsv = NULL;
2994 }
2995
2996 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2997 &BTRFS_I(inode)->runtime_flags)) {
2998 #if 0
2999 /*
3000 * For proper ENOSPC handling, we should do orphan
3001 * cleanup when mounting. But this introduces backward
3002 * compatibility issue.
3003 */
3004 if (!xchg(&root->orphan_item_inserted, 1))
3005 insert = 2;
3006 else
3007 insert = 1;
3008 #endif
3009 insert = 1;
3010 atomic_inc(&root->orphan_inodes);
3011 }
3012
3013 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3014 &BTRFS_I(inode)->runtime_flags))
3015 reserve = 1;
3016 spin_unlock(&root->orphan_lock);
3017
3018 /* grab metadata reservation from transaction handle */
3019 if (reserve) {
3020 ret = btrfs_orphan_reserve_metadata(trans, inode);
3021 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3022 }
3023
3024 /* insert an orphan item to track this unlinked/truncated file */
3025 if (insert >= 1) {
3026 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3027 if (ret) {
3028 atomic_dec(&root->orphan_inodes);
3029 if (reserve) {
3030 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3031 &BTRFS_I(inode)->runtime_flags);
3032 btrfs_orphan_release_metadata(inode);
3033 }
3034 if (ret != -EEXIST) {
3035 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3036 &BTRFS_I(inode)->runtime_flags);
3037 btrfs_abort_transaction(trans, root, ret);
3038 return ret;
3039 }
3040 }
3041 ret = 0;
3042 }
3043
3044 /* insert an orphan item to track subvolume contains orphan files */
3045 if (insert >= 2) {
3046 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3047 root->root_key.objectid);
3048 if (ret && ret != -EEXIST) {
3049 btrfs_abort_transaction(trans, root, ret);
3050 return ret;
3051 }
3052 }
3053 return 0;
3054 }
3055
3056 /*
3057 * We have done the truncate/delete so we can go ahead and remove the orphan
3058 * item for this particular inode.
3059 */
3060 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3061 struct inode *inode)
3062 {
3063 struct btrfs_root *root = BTRFS_I(inode)->root;
3064 int delete_item = 0;
3065 int release_rsv = 0;
3066 int ret = 0;
3067
3068 spin_lock(&root->orphan_lock);
3069 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3070 &BTRFS_I(inode)->runtime_flags))
3071 delete_item = 1;
3072
3073 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3074 &BTRFS_I(inode)->runtime_flags))
3075 release_rsv = 1;
3076 spin_unlock(&root->orphan_lock);
3077
3078 if (delete_item) {
3079 atomic_dec(&root->orphan_inodes);
3080 if (trans)
3081 ret = btrfs_del_orphan_item(trans, root,
3082 btrfs_ino(inode));
3083 }
3084
3085 if (release_rsv)
3086 btrfs_orphan_release_metadata(inode);
3087
3088 return ret;
3089 }
3090
3091 /*
3092 * this cleans up any orphans that may be left on the list from the last use
3093 * of this root.
3094 */
3095 int btrfs_orphan_cleanup(struct btrfs_root *root)
3096 {
3097 struct btrfs_path *path;
3098 struct extent_buffer *leaf;
3099 struct btrfs_key key, found_key;
3100 struct btrfs_trans_handle *trans;
3101 struct inode *inode;
3102 u64 last_objectid = 0;
3103 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3104
3105 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3106 return 0;
3107
3108 path = btrfs_alloc_path();
3109 if (!path) {
3110 ret = -ENOMEM;
3111 goto out;
3112 }
3113 path->reada = -1;
3114
3115 key.objectid = BTRFS_ORPHAN_OBJECTID;
3116 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3117 key.offset = (u64)-1;
3118
3119 while (1) {
3120 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3121 if (ret < 0)
3122 goto out;
3123
3124 /*
3125 * if ret == 0 means we found what we were searching for, which
3126 * is weird, but possible, so only screw with path if we didn't
3127 * find the key and see if we have stuff that matches
3128 */
3129 if (ret > 0) {
3130 ret = 0;
3131 if (path->slots[0] == 0)
3132 break;
3133 path->slots[0]--;
3134 }
3135
3136 /* pull out the item */
3137 leaf = path->nodes[0];
3138 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3139
3140 /* make sure the item matches what we want */
3141 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3142 break;
3143 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3144 break;
3145
3146 /* release the path since we're done with it */
3147 btrfs_release_path(path);
3148
3149 /*
3150 * this is where we are basically btrfs_lookup, without the
3151 * crossing root thing. we store the inode number in the
3152 * offset of the orphan item.
3153 */
3154
3155 if (found_key.offset == last_objectid) {
3156 btrfs_err(root->fs_info,
3157 "Error removing orphan entry, stopping orphan cleanup");
3158 ret = -EINVAL;
3159 goto out;
3160 }
3161
3162 last_objectid = found_key.offset;
3163
3164 found_key.objectid = found_key.offset;
3165 found_key.type = BTRFS_INODE_ITEM_KEY;
3166 found_key.offset = 0;
3167 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3168 ret = PTR_ERR_OR_ZERO(inode);
3169 if (ret && ret != -ESTALE)
3170 goto out;
3171
3172 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3173 struct btrfs_root *dead_root;
3174 struct btrfs_fs_info *fs_info = root->fs_info;
3175 int is_dead_root = 0;
3176
3177 /*
3178 * this is an orphan in the tree root. Currently these
3179 * could come from 2 sources:
3180 * a) a snapshot deletion in progress
3181 * b) a free space cache inode
3182 * We need to distinguish those two, as the snapshot
3183 * orphan must not get deleted.
3184 * find_dead_roots already ran before us, so if this
3185 * is a snapshot deletion, we should find the root
3186 * in the dead_roots list
3187 */
3188 spin_lock(&fs_info->trans_lock);
3189 list_for_each_entry(dead_root, &fs_info->dead_roots,
3190 root_list) {
3191 if (dead_root->root_key.objectid ==
3192 found_key.objectid) {
3193 is_dead_root = 1;
3194 break;
3195 }
3196 }
3197 spin_unlock(&fs_info->trans_lock);
3198 if (is_dead_root) {
3199 /* prevent this orphan from being found again */
3200 key.offset = found_key.objectid - 1;
3201 continue;
3202 }
3203 }
3204 /*
3205 * Inode is already gone but the orphan item is still there,
3206 * kill the orphan item.
3207 */
3208 if (ret == -ESTALE) {
3209 trans = btrfs_start_transaction(root, 1);
3210 if (IS_ERR(trans)) {
3211 ret = PTR_ERR(trans);
3212 goto out;
3213 }
3214 btrfs_debug(root->fs_info, "auto deleting %Lu",
3215 found_key.objectid);
3216 ret = btrfs_del_orphan_item(trans, root,
3217 found_key.objectid);
3218 btrfs_end_transaction(trans, root);
3219 if (ret)
3220 goto out;
3221 continue;
3222 }
3223
3224 /*
3225 * add this inode to the orphan list so btrfs_orphan_del does
3226 * the proper thing when we hit it
3227 */
3228 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3229 &BTRFS_I(inode)->runtime_flags);
3230 atomic_inc(&root->orphan_inodes);
3231
3232 /* if we have links, this was a truncate, lets do that */
3233 if (inode->i_nlink) {
3234 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3235 iput(inode);
3236 continue;
3237 }
3238 nr_truncate++;
3239
3240 /* 1 for the orphan item deletion. */
3241 trans = btrfs_start_transaction(root, 1);
3242 if (IS_ERR(trans)) {
3243 iput(inode);
3244 ret = PTR_ERR(trans);
3245 goto out;
3246 }
3247 ret = btrfs_orphan_add(trans, inode);
3248 btrfs_end_transaction(trans, root);
3249 if (ret) {
3250 iput(inode);
3251 goto out;
3252 }
3253
3254 ret = btrfs_truncate(inode);
3255 if (ret)
3256 btrfs_orphan_del(NULL, inode);
3257 } else {
3258 nr_unlink++;
3259 }
3260
3261 /* this will do delete_inode and everything for us */
3262 iput(inode);
3263 if (ret)
3264 goto out;
3265 }
3266 /* release the path since we're done with it */
3267 btrfs_release_path(path);
3268
3269 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3270
3271 if (root->orphan_block_rsv)
3272 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3273 (u64)-1);
3274
3275 if (root->orphan_block_rsv ||
3276 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3277 trans = btrfs_join_transaction(root);
3278 if (!IS_ERR(trans))
3279 btrfs_end_transaction(trans, root);
3280 }
3281
3282 if (nr_unlink)
3283 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3284 if (nr_truncate)
3285 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3286
3287 out:
3288 if (ret)
3289 btrfs_crit(root->fs_info,
3290 "could not do orphan cleanup %d", ret);
3291 btrfs_free_path(path);
3292 return ret;
3293 }
3294
3295 /*
3296 * very simple check to peek ahead in the leaf looking for xattrs. If we
3297 * don't find any xattrs, we know there can't be any acls.
3298 *
3299 * slot is the slot the inode is in, objectid is the objectid of the inode
3300 */
3301 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3302 int slot, u64 objectid,
3303 int *first_xattr_slot)
3304 {
3305 u32 nritems = btrfs_header_nritems(leaf);
3306 struct btrfs_key found_key;
3307 static u64 xattr_access = 0;
3308 static u64 xattr_default = 0;
3309 int scanned = 0;
3310
3311 if (!xattr_access) {
3312 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3313 strlen(POSIX_ACL_XATTR_ACCESS));
3314 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3315 strlen(POSIX_ACL_XATTR_DEFAULT));
3316 }
3317
3318 slot++;
3319 *first_xattr_slot = -1;
3320 while (slot < nritems) {
3321 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3322
3323 /* we found a different objectid, there must not be acls */
3324 if (found_key.objectid != objectid)
3325 return 0;
3326
3327 /* we found an xattr, assume we've got an acl */
3328 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3329 if (*first_xattr_slot == -1)
3330 *first_xattr_slot = slot;
3331 if (found_key.offset == xattr_access ||
3332 found_key.offset == xattr_default)
3333 return 1;
3334 }
3335
3336 /*
3337 * we found a key greater than an xattr key, there can't
3338 * be any acls later on
3339 */
3340 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3341 return 0;
3342
3343 slot++;
3344 scanned++;
3345
3346 /*
3347 * it goes inode, inode backrefs, xattrs, extents,
3348 * so if there are a ton of hard links to an inode there can
3349 * be a lot of backrefs. Don't waste time searching too hard,
3350 * this is just an optimization
3351 */
3352 if (scanned >= 8)
3353 break;
3354 }
3355 /* we hit the end of the leaf before we found an xattr or
3356 * something larger than an xattr. We have to assume the inode
3357 * has acls
3358 */
3359 if (*first_xattr_slot == -1)
3360 *first_xattr_slot = slot;
3361 return 1;
3362 }
3363
3364 /*
3365 * read an inode from the btree into the in-memory inode
3366 */
3367 static void btrfs_read_locked_inode(struct inode *inode)
3368 {
3369 struct btrfs_path *path;
3370 struct extent_buffer *leaf;
3371 struct btrfs_inode_item *inode_item;
3372 struct btrfs_timespec *tspec;
3373 struct btrfs_root *root = BTRFS_I(inode)->root;
3374 struct btrfs_key location;
3375 unsigned long ptr;
3376 int maybe_acls;
3377 u32 rdev;
3378 int ret;
3379 bool filled = false;
3380 int first_xattr_slot;
3381
3382 ret = btrfs_fill_inode(inode, &rdev);
3383 if (!ret)
3384 filled = true;
3385
3386 path = btrfs_alloc_path();
3387 if (!path)
3388 goto make_bad;
3389
3390 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3391
3392 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3393 if (ret)
3394 goto make_bad;
3395
3396 leaf = path->nodes[0];
3397
3398 if (filled)
3399 goto cache_index;
3400
3401 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3402 struct btrfs_inode_item);
3403 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3404 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3405 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3406 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3407 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3408
3409 tspec = btrfs_inode_atime(inode_item);
3410 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3411 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3412
3413 tspec = btrfs_inode_mtime(inode_item);
3414 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3415 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3416
3417 tspec = btrfs_inode_ctime(inode_item);
3418 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3419 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3420
3421 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3422 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3423 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3424
3425 /*
3426 * If we were modified in the current generation and evicted from memory
3427 * and then re-read we need to do a full sync since we don't have any
3428 * idea about which extents were modified before we were evicted from
3429 * cache.
3430 */
3431 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3432 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3433 &BTRFS_I(inode)->runtime_flags);
3434
3435 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3436 inode->i_generation = BTRFS_I(inode)->generation;
3437 inode->i_rdev = 0;
3438 rdev = btrfs_inode_rdev(leaf, inode_item);
3439
3440 BTRFS_I(inode)->index_cnt = (u64)-1;
3441 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3442
3443 cache_index:
3444 path->slots[0]++;
3445 if (inode->i_nlink != 1 ||
3446 path->slots[0] >= btrfs_header_nritems(leaf))
3447 goto cache_acl;
3448
3449 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3450 if (location.objectid != btrfs_ino(inode))
3451 goto cache_acl;
3452
3453 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3454 if (location.type == BTRFS_INODE_REF_KEY) {
3455 struct btrfs_inode_ref *ref;
3456
3457 ref = (struct btrfs_inode_ref *)ptr;
3458 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3459 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3460 struct btrfs_inode_extref *extref;
3461
3462 extref = (struct btrfs_inode_extref *)ptr;
3463 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3464 extref);
3465 }
3466 cache_acl:
3467 /*
3468 * try to precache a NULL acl entry for files that don't have
3469 * any xattrs or acls
3470 */
3471 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3472 btrfs_ino(inode), &first_xattr_slot);
3473 if (first_xattr_slot != -1) {
3474 path->slots[0] = first_xattr_slot;
3475 ret = btrfs_load_inode_props(inode, path);
3476 if (ret)
3477 btrfs_err(root->fs_info,
3478 "error loading props for ino %llu (root %llu): %d",
3479 btrfs_ino(inode),
3480 root->root_key.objectid, ret);
3481 }
3482 btrfs_free_path(path);
3483
3484 if (!maybe_acls)
3485 cache_no_acl(inode);
3486
3487 switch (inode->i_mode & S_IFMT) {
3488 case S_IFREG:
3489 inode->i_mapping->a_ops = &btrfs_aops;
3490 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3491 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3492 inode->i_fop = &btrfs_file_operations;
3493 inode->i_op = &btrfs_file_inode_operations;
3494 break;
3495 case S_IFDIR:
3496 inode->i_fop = &btrfs_dir_file_operations;
3497 if (root == root->fs_info->tree_root)
3498 inode->i_op = &btrfs_dir_ro_inode_operations;
3499 else
3500 inode->i_op = &btrfs_dir_inode_operations;
3501 break;
3502 case S_IFLNK:
3503 inode->i_op = &btrfs_symlink_inode_operations;
3504 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3505 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3506 break;
3507 default:
3508 inode->i_op = &btrfs_special_inode_operations;
3509 init_special_inode(inode, inode->i_mode, rdev);
3510 break;
3511 }
3512
3513 btrfs_update_iflags(inode);
3514 return;
3515
3516 make_bad:
3517 btrfs_free_path(path);
3518 make_bad_inode(inode);
3519 }
3520
3521 /*
3522 * given a leaf and an inode, copy the inode fields into the leaf
3523 */
3524 static void fill_inode_item(struct btrfs_trans_handle *trans,
3525 struct extent_buffer *leaf,
3526 struct btrfs_inode_item *item,
3527 struct inode *inode)
3528 {
3529 struct btrfs_map_token token;
3530
3531 btrfs_init_map_token(&token);
3532
3533 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3534 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3535 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3536 &token);
3537 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3538 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3539
3540 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3541 inode->i_atime.tv_sec, &token);
3542 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3543 inode->i_atime.tv_nsec, &token);
3544
3545 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3546 inode->i_mtime.tv_sec, &token);
3547 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3548 inode->i_mtime.tv_nsec, &token);
3549
3550 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3551 inode->i_ctime.tv_sec, &token);
3552 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3553 inode->i_ctime.tv_nsec, &token);
3554
3555 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3556 &token);
3557 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3558 &token);
3559 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3560 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3561 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3562 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3563 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3564 }
3565
3566 /*
3567 * copy everything in the in-memory inode into the btree.
3568 */
3569 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3570 struct btrfs_root *root, struct inode *inode)
3571 {
3572 struct btrfs_inode_item *inode_item;
3573 struct btrfs_path *path;
3574 struct extent_buffer *leaf;
3575 int ret;
3576
3577 path = btrfs_alloc_path();
3578 if (!path)
3579 return -ENOMEM;
3580
3581 path->leave_spinning = 1;
3582 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3583 1);
3584 if (ret) {
3585 if (ret > 0)
3586 ret = -ENOENT;
3587 goto failed;
3588 }
3589
3590 leaf = path->nodes[0];
3591 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3592 struct btrfs_inode_item);
3593
3594 fill_inode_item(trans, leaf, inode_item, inode);
3595 btrfs_mark_buffer_dirty(leaf);
3596 btrfs_set_inode_last_trans(trans, inode);
3597 ret = 0;
3598 failed:
3599 btrfs_free_path(path);
3600 return ret;
3601 }
3602
3603 /*
3604 * copy everything in the in-memory inode into the btree.
3605 */
3606 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3607 struct btrfs_root *root, struct inode *inode)
3608 {
3609 int ret;
3610
3611 /*
3612 * If the inode is a free space inode, we can deadlock during commit
3613 * if we put it into the delayed code.
3614 *
3615 * The data relocation inode should also be directly updated
3616 * without delay
3617 */
3618 if (!btrfs_is_free_space_inode(inode)
3619 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3620 btrfs_update_root_times(trans, root);
3621
3622 ret = btrfs_delayed_update_inode(trans, root, inode);
3623 if (!ret)
3624 btrfs_set_inode_last_trans(trans, inode);
3625 return ret;
3626 }
3627
3628 return btrfs_update_inode_item(trans, root, inode);
3629 }
3630
3631 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3632 struct btrfs_root *root,
3633 struct inode *inode)
3634 {
3635 int ret;
3636
3637 ret = btrfs_update_inode(trans, root, inode);
3638 if (ret == -ENOSPC)
3639 return btrfs_update_inode_item(trans, root, inode);
3640 return ret;
3641 }
3642
3643 /*
3644 * unlink helper that gets used here in inode.c and in the tree logging
3645 * recovery code. It remove a link in a directory with a given name, and
3646 * also drops the back refs in the inode to the directory
3647 */
3648 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3649 struct btrfs_root *root,
3650 struct inode *dir, struct inode *inode,
3651 const char *name, int name_len)
3652 {
3653 struct btrfs_path *path;
3654 int ret = 0;
3655 struct extent_buffer *leaf;
3656 struct btrfs_dir_item *di;
3657 struct btrfs_key key;
3658 u64 index;
3659 u64 ino = btrfs_ino(inode);
3660 u64 dir_ino = btrfs_ino(dir);
3661
3662 path = btrfs_alloc_path();
3663 if (!path) {
3664 ret = -ENOMEM;
3665 goto out;
3666 }
3667
3668 path->leave_spinning = 1;
3669 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3670 name, name_len, -1);
3671 if (IS_ERR(di)) {
3672 ret = PTR_ERR(di);
3673 goto err;
3674 }
3675 if (!di) {
3676 ret = -ENOENT;
3677 goto err;
3678 }
3679 leaf = path->nodes[0];
3680 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3681 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3682 if (ret)
3683 goto err;
3684 btrfs_release_path(path);
3685
3686 /*
3687 * If we don't have dir index, we have to get it by looking up
3688 * the inode ref, since we get the inode ref, remove it directly,
3689 * it is unnecessary to do delayed deletion.
3690 *
3691 * But if we have dir index, needn't search inode ref to get it.
3692 * Since the inode ref is close to the inode item, it is better
3693 * that we delay to delete it, and just do this deletion when
3694 * we update the inode item.
3695 */
3696 if (BTRFS_I(inode)->dir_index) {
3697 ret = btrfs_delayed_delete_inode_ref(inode);
3698 if (!ret) {
3699 index = BTRFS_I(inode)->dir_index;
3700 goto skip_backref;
3701 }
3702 }
3703
3704 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3705 dir_ino, &index);
3706 if (ret) {
3707 btrfs_info(root->fs_info,
3708 "failed to delete reference to %.*s, inode %llu parent %llu",
3709 name_len, name, ino, dir_ino);
3710 btrfs_abort_transaction(trans, root, ret);
3711 goto err;
3712 }
3713 skip_backref:
3714 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3715 if (ret) {
3716 btrfs_abort_transaction(trans, root, ret);
3717 goto err;
3718 }
3719
3720 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3721 inode, dir_ino);
3722 if (ret != 0 && ret != -ENOENT) {
3723 btrfs_abort_transaction(trans, root, ret);
3724 goto err;
3725 }
3726
3727 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3728 dir, index);
3729 if (ret == -ENOENT)
3730 ret = 0;
3731 else if (ret)
3732 btrfs_abort_transaction(trans, root, ret);
3733 err:
3734 btrfs_free_path(path);
3735 if (ret)
3736 goto out;
3737
3738 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3739 inode_inc_iversion(inode);
3740 inode_inc_iversion(dir);
3741 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3742 ret = btrfs_update_inode(trans, root, dir);
3743 out:
3744 return ret;
3745 }
3746
3747 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3748 struct btrfs_root *root,
3749 struct inode *dir, struct inode *inode,
3750 const char *name, int name_len)
3751 {
3752 int ret;
3753 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3754 if (!ret) {
3755 drop_nlink(inode);
3756 ret = btrfs_update_inode(trans, root, inode);
3757 }
3758 return ret;
3759 }
3760
3761 /*
3762 * helper to start transaction for unlink and rmdir.
3763 *
3764 * unlink and rmdir are special in btrfs, they do not always free space, so
3765 * if we cannot make our reservations the normal way try and see if there is
3766 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3767 * allow the unlink to occur.
3768 */
3769 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3770 {
3771 struct btrfs_trans_handle *trans;
3772 struct btrfs_root *root = BTRFS_I(dir)->root;
3773 int ret;
3774
3775 /*
3776 * 1 for the possible orphan item
3777 * 1 for the dir item
3778 * 1 for the dir index
3779 * 1 for the inode ref
3780 * 1 for the inode
3781 */
3782 trans = btrfs_start_transaction(root, 5);
3783 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3784 return trans;
3785
3786 if (PTR_ERR(trans) == -ENOSPC) {
3787 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3788
3789 trans = btrfs_start_transaction(root, 0);
3790 if (IS_ERR(trans))
3791 return trans;
3792 ret = btrfs_cond_migrate_bytes(root->fs_info,
3793 &root->fs_info->trans_block_rsv,
3794 num_bytes, 5);
3795 if (ret) {
3796 btrfs_end_transaction(trans, root);
3797 return ERR_PTR(ret);
3798 }
3799 trans->block_rsv = &root->fs_info->trans_block_rsv;
3800 trans->bytes_reserved = num_bytes;
3801 }
3802 return trans;
3803 }
3804
3805 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3806 {
3807 struct btrfs_root *root = BTRFS_I(dir)->root;
3808 struct btrfs_trans_handle *trans;
3809 struct inode *inode = dentry->d_inode;
3810 int ret;
3811
3812 trans = __unlink_start_trans(dir);
3813 if (IS_ERR(trans))
3814 return PTR_ERR(trans);
3815
3816 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3817
3818 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3819 dentry->d_name.name, dentry->d_name.len);
3820 if (ret)
3821 goto out;
3822
3823 if (inode->i_nlink == 0) {
3824 ret = btrfs_orphan_add(trans, inode);
3825 if (ret)
3826 goto out;
3827 }
3828
3829 out:
3830 btrfs_end_transaction(trans, root);
3831 btrfs_btree_balance_dirty(root);
3832 return ret;
3833 }
3834
3835 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3836 struct btrfs_root *root,
3837 struct inode *dir, u64 objectid,
3838 const char *name, int name_len)
3839 {
3840 struct btrfs_path *path;
3841 struct extent_buffer *leaf;
3842 struct btrfs_dir_item *di;
3843 struct btrfs_key key;
3844 u64 index;
3845 int ret;
3846 u64 dir_ino = btrfs_ino(dir);
3847
3848 path = btrfs_alloc_path();
3849 if (!path)
3850 return -ENOMEM;
3851
3852 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3853 name, name_len, -1);
3854 if (IS_ERR_OR_NULL(di)) {
3855 if (!di)
3856 ret = -ENOENT;
3857 else
3858 ret = PTR_ERR(di);
3859 goto out;
3860 }
3861
3862 leaf = path->nodes[0];
3863 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3864 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3865 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3866 if (ret) {
3867 btrfs_abort_transaction(trans, root, ret);
3868 goto out;
3869 }
3870 btrfs_release_path(path);
3871
3872 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3873 objectid, root->root_key.objectid,
3874 dir_ino, &index, name, name_len);
3875 if (ret < 0) {
3876 if (ret != -ENOENT) {
3877 btrfs_abort_transaction(trans, root, ret);
3878 goto out;
3879 }
3880 di = btrfs_search_dir_index_item(root, path, dir_ino,
3881 name, name_len);
3882 if (IS_ERR_OR_NULL(di)) {
3883 if (!di)
3884 ret = -ENOENT;
3885 else
3886 ret = PTR_ERR(di);
3887 btrfs_abort_transaction(trans, root, ret);
3888 goto out;
3889 }
3890
3891 leaf = path->nodes[0];
3892 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3893 btrfs_release_path(path);
3894 index = key.offset;
3895 }
3896 btrfs_release_path(path);
3897
3898 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3899 if (ret) {
3900 btrfs_abort_transaction(trans, root, ret);
3901 goto out;
3902 }
3903
3904 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3905 inode_inc_iversion(dir);
3906 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3907 ret = btrfs_update_inode_fallback(trans, root, dir);
3908 if (ret)
3909 btrfs_abort_transaction(trans, root, ret);
3910 out:
3911 btrfs_free_path(path);
3912 return ret;
3913 }
3914
3915 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3916 {
3917 struct inode *inode = dentry->d_inode;
3918 int err = 0;
3919 struct btrfs_root *root = BTRFS_I(dir)->root;
3920 struct btrfs_trans_handle *trans;
3921
3922 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3923 return -ENOTEMPTY;
3924 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3925 return -EPERM;
3926
3927 trans = __unlink_start_trans(dir);
3928 if (IS_ERR(trans))
3929 return PTR_ERR(trans);
3930
3931 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3932 err = btrfs_unlink_subvol(trans, root, dir,
3933 BTRFS_I(inode)->location.objectid,
3934 dentry->d_name.name,
3935 dentry->d_name.len);
3936 goto out;
3937 }
3938
3939 err = btrfs_orphan_add(trans, inode);
3940 if (err)
3941 goto out;
3942
3943 /* now the directory is empty */
3944 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3945 dentry->d_name.name, dentry->d_name.len);
3946 if (!err)
3947 btrfs_i_size_write(inode, 0);
3948 out:
3949 btrfs_end_transaction(trans, root);
3950 btrfs_btree_balance_dirty(root);
3951
3952 return err;
3953 }
3954
3955 /*
3956 * this can truncate away extent items, csum items and directory items.
3957 * It starts at a high offset and removes keys until it can't find
3958 * any higher than new_size
3959 *
3960 * csum items that cross the new i_size are truncated to the new size
3961 * as well.
3962 *
3963 * min_type is the minimum key type to truncate down to. If set to 0, this
3964 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3965 */
3966 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3967 struct btrfs_root *root,
3968 struct inode *inode,
3969 u64 new_size, u32 min_type)
3970 {
3971 struct btrfs_path *path;
3972 struct extent_buffer *leaf;
3973 struct btrfs_file_extent_item *fi;
3974 struct btrfs_key key;
3975 struct btrfs_key found_key;
3976 u64 extent_start = 0;
3977 u64 extent_num_bytes = 0;
3978 u64 extent_offset = 0;
3979 u64 item_end = 0;
3980 u64 last_size = (u64)-1;
3981 u32 found_type = (u8)-1;
3982 int found_extent;
3983 int del_item;
3984 int pending_del_nr = 0;
3985 int pending_del_slot = 0;
3986 int extent_type = -1;
3987 int ret;
3988 int err = 0;
3989 u64 ino = btrfs_ino(inode);
3990
3991 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3992
3993 path = btrfs_alloc_path();
3994 if (!path)
3995 return -ENOMEM;
3996 path->reada = -1;
3997
3998 /*
3999 * We want to drop from the next block forward in case this new size is
4000 * not block aligned since we will be keeping the last block of the
4001 * extent just the way it is.
4002 */
4003 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4004 root == root->fs_info->tree_root)
4005 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4006 root->sectorsize), (u64)-1, 0);
4007
4008 /*
4009 * This function is also used to drop the items in the log tree before
4010 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4011 * it is used to drop the loged items. So we shouldn't kill the delayed
4012 * items.
4013 */
4014 if (min_type == 0 && root == BTRFS_I(inode)->root)
4015 btrfs_kill_delayed_inode_items(inode);
4016
4017 key.objectid = ino;
4018 key.offset = (u64)-1;
4019 key.type = (u8)-1;
4020
4021 search_again:
4022 path->leave_spinning = 1;
4023 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4024 if (ret < 0) {
4025 err = ret;
4026 goto out;
4027 }
4028
4029 if (ret > 0) {
4030 /* there are no items in the tree for us to truncate, we're
4031 * done
4032 */
4033 if (path->slots[0] == 0)
4034 goto out;
4035 path->slots[0]--;
4036 }
4037
4038 while (1) {
4039 fi = NULL;
4040 leaf = path->nodes[0];
4041 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4042 found_type = btrfs_key_type(&found_key);
4043
4044 if (found_key.objectid != ino)
4045 break;
4046
4047 if (found_type < min_type)
4048 break;
4049
4050 item_end = found_key.offset;
4051 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4052 fi = btrfs_item_ptr(leaf, path->slots[0],
4053 struct btrfs_file_extent_item);
4054 extent_type = btrfs_file_extent_type(leaf, fi);
4055 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4056 item_end +=
4057 btrfs_file_extent_num_bytes(leaf, fi);
4058 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4059 item_end += btrfs_file_extent_inline_len(leaf,
4060 path->slots[0], fi);
4061 }
4062 item_end--;
4063 }
4064 if (found_type > min_type) {
4065 del_item = 1;
4066 } else {
4067 if (item_end < new_size)
4068 break;
4069 if (found_key.offset >= new_size)
4070 del_item = 1;
4071 else
4072 del_item = 0;
4073 }
4074 found_extent = 0;
4075 /* FIXME, shrink the extent if the ref count is only 1 */
4076 if (found_type != BTRFS_EXTENT_DATA_KEY)
4077 goto delete;
4078
4079 if (del_item)
4080 last_size = found_key.offset;
4081 else
4082 last_size = new_size;
4083
4084 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4085 u64 num_dec;
4086 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4087 if (!del_item) {
4088 u64 orig_num_bytes =
4089 btrfs_file_extent_num_bytes(leaf, fi);
4090 extent_num_bytes = ALIGN(new_size -
4091 found_key.offset,
4092 root->sectorsize);
4093 btrfs_set_file_extent_num_bytes(leaf, fi,
4094 extent_num_bytes);
4095 num_dec = (orig_num_bytes -
4096 extent_num_bytes);
4097 if (test_bit(BTRFS_ROOT_REF_COWS,
4098 &root->state) &&
4099 extent_start != 0)
4100 inode_sub_bytes(inode, num_dec);
4101 btrfs_mark_buffer_dirty(leaf);
4102 } else {
4103 extent_num_bytes =
4104 btrfs_file_extent_disk_num_bytes(leaf,
4105 fi);
4106 extent_offset = found_key.offset -
4107 btrfs_file_extent_offset(leaf, fi);
4108
4109 /* FIXME blocksize != 4096 */
4110 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4111 if (extent_start != 0) {
4112 found_extent = 1;
4113 if (test_bit(BTRFS_ROOT_REF_COWS,
4114 &root->state))
4115 inode_sub_bytes(inode, num_dec);
4116 }
4117 }
4118 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4119 /*
4120 * we can't truncate inline items that have had
4121 * special encodings
4122 */
4123 if (!del_item &&
4124 btrfs_file_extent_compression(leaf, fi) == 0 &&
4125 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4126 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4127 u32 size = new_size - found_key.offset;
4128
4129 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4130 inode_sub_bytes(inode, item_end + 1 -
4131 new_size);
4132
4133 /*
4134 * update the ram bytes to properly reflect
4135 * the new size of our item
4136 */
4137 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4138 size =
4139 btrfs_file_extent_calc_inline_size(size);
4140 btrfs_truncate_item(root, path, size, 1);
4141 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4142 &root->state)) {
4143 inode_sub_bytes(inode, item_end + 1 -
4144 found_key.offset);
4145 }
4146 }
4147 delete:
4148 if (del_item) {
4149 if (!pending_del_nr) {
4150 /* no pending yet, add ourselves */
4151 pending_del_slot = path->slots[0];
4152 pending_del_nr = 1;
4153 } else if (pending_del_nr &&
4154 path->slots[0] + 1 == pending_del_slot) {
4155 /* hop on the pending chunk */
4156 pending_del_nr++;
4157 pending_del_slot = path->slots[0];
4158 } else {
4159 BUG();
4160 }
4161 } else {
4162 break;
4163 }
4164 if (found_extent &&
4165 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4166 root == root->fs_info->tree_root)) {
4167 btrfs_set_path_blocking(path);
4168 ret = btrfs_free_extent(trans, root, extent_start,
4169 extent_num_bytes, 0,
4170 btrfs_header_owner(leaf),
4171 ino, extent_offset, 0);
4172 BUG_ON(ret);
4173 }
4174
4175 if (found_type == BTRFS_INODE_ITEM_KEY)
4176 break;
4177
4178 if (path->slots[0] == 0 ||
4179 path->slots[0] != pending_del_slot) {
4180 if (pending_del_nr) {
4181 ret = btrfs_del_items(trans, root, path,
4182 pending_del_slot,
4183 pending_del_nr);
4184 if (ret) {
4185 btrfs_abort_transaction(trans,
4186 root, ret);
4187 goto error;
4188 }
4189 pending_del_nr = 0;
4190 }
4191 btrfs_release_path(path);
4192 goto search_again;
4193 } else {
4194 path->slots[0]--;
4195 }
4196 }
4197 out:
4198 if (pending_del_nr) {
4199 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4200 pending_del_nr);
4201 if (ret)
4202 btrfs_abort_transaction(trans, root, ret);
4203 }
4204 error:
4205 if (last_size != (u64)-1)
4206 btrfs_ordered_update_i_size(inode, last_size, NULL);
4207 btrfs_free_path(path);
4208 return err;
4209 }
4210
4211 /*
4212 * btrfs_truncate_page - read, zero a chunk and write a page
4213 * @inode - inode that we're zeroing
4214 * @from - the offset to start zeroing
4215 * @len - the length to zero, 0 to zero the entire range respective to the
4216 * offset
4217 * @front - zero up to the offset instead of from the offset on
4218 *
4219 * This will find the page for the "from" offset and cow the page and zero the
4220 * part we want to zero. This is used with truncate and hole punching.
4221 */
4222 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4223 int front)
4224 {
4225 struct address_space *mapping = inode->i_mapping;
4226 struct btrfs_root *root = BTRFS_I(inode)->root;
4227 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4228 struct btrfs_ordered_extent *ordered;
4229 struct extent_state *cached_state = NULL;
4230 char *kaddr;
4231 u32 blocksize = root->sectorsize;
4232 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4233 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4234 struct page *page;
4235 gfp_t mask = btrfs_alloc_write_mask(mapping);
4236 int ret = 0;
4237 u64 page_start;
4238 u64 page_end;
4239
4240 if ((offset & (blocksize - 1)) == 0 &&
4241 (!len || ((len & (blocksize - 1)) == 0)))
4242 goto out;
4243 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4244 if (ret)
4245 goto out;
4246
4247 again:
4248 page = find_or_create_page(mapping, index, mask);
4249 if (!page) {
4250 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4251 ret = -ENOMEM;
4252 goto out;
4253 }
4254
4255 page_start = page_offset(page);
4256 page_end = page_start + PAGE_CACHE_SIZE - 1;
4257
4258 if (!PageUptodate(page)) {
4259 ret = btrfs_readpage(NULL, page);
4260 lock_page(page);
4261 if (page->mapping != mapping) {
4262 unlock_page(page);
4263 page_cache_release(page);
4264 goto again;
4265 }
4266 if (!PageUptodate(page)) {
4267 ret = -EIO;
4268 goto out_unlock;
4269 }
4270 }
4271 wait_on_page_writeback(page);
4272
4273 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4274 set_page_extent_mapped(page);
4275
4276 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4277 if (ordered) {
4278 unlock_extent_cached(io_tree, page_start, page_end,
4279 &cached_state, GFP_NOFS);
4280 unlock_page(page);
4281 page_cache_release(page);
4282 btrfs_start_ordered_extent(inode, ordered, 1);
4283 btrfs_put_ordered_extent(ordered);
4284 goto again;
4285 }
4286
4287 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4288 EXTENT_DIRTY | EXTENT_DELALLOC |
4289 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4290 0, 0, &cached_state, GFP_NOFS);
4291
4292 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4293 &cached_state);
4294 if (ret) {
4295 unlock_extent_cached(io_tree, page_start, page_end,
4296 &cached_state, GFP_NOFS);
4297 goto out_unlock;
4298 }
4299
4300 if (offset != PAGE_CACHE_SIZE) {
4301 if (!len)
4302 len = PAGE_CACHE_SIZE - offset;
4303 kaddr = kmap(page);
4304 if (front)
4305 memset(kaddr, 0, offset);
4306 else
4307 memset(kaddr + offset, 0, len);
4308 flush_dcache_page(page);
4309 kunmap(page);
4310 }
4311 ClearPageChecked(page);
4312 set_page_dirty(page);
4313 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4314 GFP_NOFS);
4315
4316 out_unlock:
4317 if (ret)
4318 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4319 unlock_page(page);
4320 page_cache_release(page);
4321 out:
4322 return ret;
4323 }
4324
4325 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4326 u64 offset, u64 len)
4327 {
4328 struct btrfs_trans_handle *trans;
4329 int ret;
4330
4331 /*
4332 * Still need to make sure the inode looks like it's been updated so
4333 * that any holes get logged if we fsync.
4334 */
4335 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4336 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4337 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4338 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4339 return 0;
4340 }
4341
4342 /*
4343 * 1 - for the one we're dropping
4344 * 1 - for the one we're adding
4345 * 1 - for updating the inode.
4346 */
4347 trans = btrfs_start_transaction(root, 3);
4348 if (IS_ERR(trans))
4349 return PTR_ERR(trans);
4350
4351 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4352 if (ret) {
4353 btrfs_abort_transaction(trans, root, ret);
4354 btrfs_end_transaction(trans, root);
4355 return ret;
4356 }
4357
4358 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4359 0, 0, len, 0, len, 0, 0, 0);
4360 if (ret)
4361 btrfs_abort_transaction(trans, root, ret);
4362 else
4363 btrfs_update_inode(trans, root, inode);
4364 btrfs_end_transaction(trans, root);
4365 return ret;
4366 }
4367
4368 /*
4369 * This function puts in dummy file extents for the area we're creating a hole
4370 * for. So if we are truncating this file to a larger size we need to insert
4371 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4372 * the range between oldsize and size
4373 */
4374 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4375 {
4376 struct btrfs_root *root = BTRFS_I(inode)->root;
4377 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4378 struct extent_map *em = NULL;
4379 struct extent_state *cached_state = NULL;
4380 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4381 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4382 u64 block_end = ALIGN(size, root->sectorsize);
4383 u64 last_byte;
4384 u64 cur_offset;
4385 u64 hole_size;
4386 int err = 0;
4387
4388 /*
4389 * If our size started in the middle of a page we need to zero out the
4390 * rest of the page before we expand the i_size, otherwise we could
4391 * expose stale data.
4392 */
4393 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4394 if (err)
4395 return err;
4396
4397 if (size <= hole_start)
4398 return 0;
4399
4400 while (1) {
4401 struct btrfs_ordered_extent *ordered;
4402
4403 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4404 &cached_state);
4405 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4406 block_end - hole_start);
4407 if (!ordered)
4408 break;
4409 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4410 &cached_state, GFP_NOFS);
4411 btrfs_start_ordered_extent(inode, ordered, 1);
4412 btrfs_put_ordered_extent(ordered);
4413 }
4414
4415 cur_offset = hole_start;
4416 while (1) {
4417 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4418 block_end - cur_offset, 0);
4419 if (IS_ERR(em)) {
4420 err = PTR_ERR(em);
4421 em = NULL;
4422 break;
4423 }
4424 last_byte = min(extent_map_end(em), block_end);
4425 last_byte = ALIGN(last_byte , root->sectorsize);
4426 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4427 struct extent_map *hole_em;
4428 hole_size = last_byte - cur_offset;
4429
4430 err = maybe_insert_hole(root, inode, cur_offset,
4431 hole_size);
4432 if (err)
4433 break;
4434 btrfs_drop_extent_cache(inode, cur_offset,
4435 cur_offset + hole_size - 1, 0);
4436 hole_em = alloc_extent_map();
4437 if (!hole_em) {
4438 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4439 &BTRFS_I(inode)->runtime_flags);
4440 goto next;
4441 }
4442 hole_em->start = cur_offset;
4443 hole_em->len = hole_size;
4444 hole_em->orig_start = cur_offset;
4445
4446 hole_em->block_start = EXTENT_MAP_HOLE;
4447 hole_em->block_len = 0;
4448 hole_em->orig_block_len = 0;
4449 hole_em->ram_bytes = hole_size;
4450 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4451 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4452 hole_em->generation = root->fs_info->generation;
4453
4454 while (1) {
4455 write_lock(&em_tree->lock);
4456 err = add_extent_mapping(em_tree, hole_em, 1);
4457 write_unlock(&em_tree->lock);
4458 if (err != -EEXIST)
4459 break;
4460 btrfs_drop_extent_cache(inode, cur_offset,
4461 cur_offset +
4462 hole_size - 1, 0);
4463 }
4464 free_extent_map(hole_em);
4465 }
4466 next:
4467 free_extent_map(em);
4468 em = NULL;
4469 cur_offset = last_byte;
4470 if (cur_offset >= block_end)
4471 break;
4472 }
4473 free_extent_map(em);
4474 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4475 GFP_NOFS);
4476 return err;
4477 }
4478
4479 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4480 {
4481 struct btrfs_root *root = BTRFS_I(inode)->root;
4482 struct btrfs_trans_handle *trans;
4483 loff_t oldsize = i_size_read(inode);
4484 loff_t newsize = attr->ia_size;
4485 int mask = attr->ia_valid;
4486 int ret;
4487
4488 /*
4489 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4490 * special case where we need to update the times despite not having
4491 * these flags set. For all other operations the VFS set these flags
4492 * explicitly if it wants a timestamp update.
4493 */
4494 if (newsize != oldsize) {
4495 inode_inc_iversion(inode);
4496 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4497 inode->i_ctime = inode->i_mtime =
4498 current_fs_time(inode->i_sb);
4499 }
4500
4501 if (newsize > oldsize) {
4502 truncate_pagecache(inode, newsize);
4503 ret = btrfs_cont_expand(inode, oldsize, newsize);
4504 if (ret)
4505 return ret;
4506
4507 trans = btrfs_start_transaction(root, 1);
4508 if (IS_ERR(trans))
4509 return PTR_ERR(trans);
4510
4511 i_size_write(inode, newsize);
4512 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4513 ret = btrfs_update_inode(trans, root, inode);
4514 btrfs_end_transaction(trans, root);
4515 } else {
4516
4517 /*
4518 * We're truncating a file that used to have good data down to
4519 * zero. Make sure it gets into the ordered flush list so that
4520 * any new writes get down to disk quickly.
4521 */
4522 if (newsize == 0)
4523 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4524 &BTRFS_I(inode)->runtime_flags);
4525
4526 /*
4527 * 1 for the orphan item we're going to add
4528 * 1 for the orphan item deletion.
4529 */
4530 trans = btrfs_start_transaction(root, 2);
4531 if (IS_ERR(trans))
4532 return PTR_ERR(trans);
4533
4534 /*
4535 * We need to do this in case we fail at _any_ point during the
4536 * actual truncate. Once we do the truncate_setsize we could
4537 * invalidate pages which forces any outstanding ordered io to
4538 * be instantly completed which will give us extents that need
4539 * to be truncated. If we fail to get an orphan inode down we
4540 * could have left over extents that were never meant to live,
4541 * so we need to garuntee from this point on that everything
4542 * will be consistent.
4543 */
4544 ret = btrfs_orphan_add(trans, inode);
4545 btrfs_end_transaction(trans, root);
4546 if (ret)
4547 return ret;
4548
4549 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4550 truncate_setsize(inode, newsize);
4551
4552 /* Disable nonlocked read DIO to avoid the end less truncate */
4553 btrfs_inode_block_unlocked_dio(inode);
4554 inode_dio_wait(inode);
4555 btrfs_inode_resume_unlocked_dio(inode);
4556
4557 ret = btrfs_truncate(inode);
4558 if (ret && inode->i_nlink) {
4559 int err;
4560
4561 /*
4562 * failed to truncate, disk_i_size is only adjusted down
4563 * as we remove extents, so it should represent the true
4564 * size of the inode, so reset the in memory size and
4565 * delete our orphan entry.
4566 */
4567 trans = btrfs_join_transaction(root);
4568 if (IS_ERR(trans)) {
4569 btrfs_orphan_del(NULL, inode);
4570 return ret;
4571 }
4572 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4573 err = btrfs_orphan_del(trans, inode);
4574 if (err)
4575 btrfs_abort_transaction(trans, root, err);
4576 btrfs_end_transaction(trans, root);
4577 }
4578 }
4579
4580 return ret;
4581 }
4582
4583 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4584 {
4585 struct inode *inode = dentry->d_inode;
4586 struct btrfs_root *root = BTRFS_I(inode)->root;
4587 int err;
4588
4589 if (btrfs_root_readonly(root))
4590 return -EROFS;
4591
4592 err = inode_change_ok(inode, attr);
4593 if (err)
4594 return err;
4595
4596 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4597 err = btrfs_setsize(inode, attr);
4598 if (err)
4599 return err;
4600 }
4601
4602 if (attr->ia_valid) {
4603 setattr_copy(inode, attr);
4604 inode_inc_iversion(inode);
4605 err = btrfs_dirty_inode(inode);
4606
4607 if (!err && attr->ia_valid & ATTR_MODE)
4608 err = posix_acl_chmod(inode, inode->i_mode);
4609 }
4610
4611 return err;
4612 }
4613
4614 /*
4615 * While truncating the inode pages during eviction, we get the VFS calling
4616 * btrfs_invalidatepage() against each page of the inode. This is slow because
4617 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4618 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4619 * extent_state structures over and over, wasting lots of time.
4620 *
4621 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4622 * those expensive operations on a per page basis and do only the ordered io
4623 * finishing, while we release here the extent_map and extent_state structures,
4624 * without the excessive merging and splitting.
4625 */
4626 static void evict_inode_truncate_pages(struct inode *inode)
4627 {
4628 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4629 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4630 struct rb_node *node;
4631
4632 ASSERT(inode->i_state & I_FREEING);
4633 truncate_inode_pages_final(&inode->i_data);
4634
4635 write_lock(&map_tree->lock);
4636 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4637 struct extent_map *em;
4638
4639 node = rb_first(&map_tree->map);
4640 em = rb_entry(node, struct extent_map, rb_node);
4641 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4642 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4643 remove_extent_mapping(map_tree, em);
4644 free_extent_map(em);
4645 }
4646 write_unlock(&map_tree->lock);
4647
4648 spin_lock(&io_tree->lock);
4649 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4650 struct extent_state *state;
4651 struct extent_state *cached_state = NULL;
4652
4653 node = rb_first(&io_tree->state);
4654 state = rb_entry(node, struct extent_state, rb_node);
4655 atomic_inc(&state->refs);
4656 spin_unlock(&io_tree->lock);
4657
4658 lock_extent_bits(io_tree, state->start, state->end,
4659 0, &cached_state);
4660 clear_extent_bit(io_tree, state->start, state->end,
4661 EXTENT_LOCKED | EXTENT_DIRTY |
4662 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4663 EXTENT_DEFRAG, 1, 1,
4664 &cached_state, GFP_NOFS);
4665 free_extent_state(state);
4666
4667 spin_lock(&io_tree->lock);
4668 }
4669 spin_unlock(&io_tree->lock);
4670 }
4671
4672 void btrfs_evict_inode(struct inode *inode)
4673 {
4674 struct btrfs_trans_handle *trans;
4675 struct btrfs_root *root = BTRFS_I(inode)->root;
4676 struct btrfs_block_rsv *rsv, *global_rsv;
4677 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4678 int ret;
4679
4680 trace_btrfs_inode_evict(inode);
4681
4682 evict_inode_truncate_pages(inode);
4683
4684 if (inode->i_nlink &&
4685 ((btrfs_root_refs(&root->root_item) != 0 &&
4686 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4687 btrfs_is_free_space_inode(inode)))
4688 goto no_delete;
4689
4690 if (is_bad_inode(inode)) {
4691 btrfs_orphan_del(NULL, inode);
4692 goto no_delete;
4693 }
4694 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4695 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4696
4697 if (root->fs_info->log_root_recovering) {
4698 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4699 &BTRFS_I(inode)->runtime_flags));
4700 goto no_delete;
4701 }
4702
4703 if (inode->i_nlink > 0) {
4704 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4705 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4706 goto no_delete;
4707 }
4708
4709 ret = btrfs_commit_inode_delayed_inode(inode);
4710 if (ret) {
4711 btrfs_orphan_del(NULL, inode);
4712 goto no_delete;
4713 }
4714
4715 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4716 if (!rsv) {
4717 btrfs_orphan_del(NULL, inode);
4718 goto no_delete;
4719 }
4720 rsv->size = min_size;
4721 rsv->failfast = 1;
4722 global_rsv = &root->fs_info->global_block_rsv;
4723
4724 btrfs_i_size_write(inode, 0);
4725
4726 /*
4727 * This is a bit simpler than btrfs_truncate since we've already
4728 * reserved our space for our orphan item in the unlink, so we just
4729 * need to reserve some slack space in case we add bytes and update
4730 * inode item when doing the truncate.
4731 */
4732 while (1) {
4733 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4734 BTRFS_RESERVE_FLUSH_LIMIT);
4735
4736 /*
4737 * Try and steal from the global reserve since we will
4738 * likely not use this space anyway, we want to try as
4739 * hard as possible to get this to work.
4740 */
4741 if (ret)
4742 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4743
4744 if (ret) {
4745 btrfs_warn(root->fs_info,
4746 "Could not get space for a delete, will truncate on mount %d",
4747 ret);
4748 btrfs_orphan_del(NULL, inode);
4749 btrfs_free_block_rsv(root, rsv);
4750 goto no_delete;
4751 }
4752
4753 trans = btrfs_join_transaction(root);
4754 if (IS_ERR(trans)) {
4755 btrfs_orphan_del(NULL, inode);
4756 btrfs_free_block_rsv(root, rsv);
4757 goto no_delete;
4758 }
4759
4760 trans->block_rsv = rsv;
4761
4762 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4763 if (ret != -ENOSPC)
4764 break;
4765
4766 trans->block_rsv = &root->fs_info->trans_block_rsv;
4767 btrfs_end_transaction(trans, root);
4768 trans = NULL;
4769 btrfs_btree_balance_dirty(root);
4770 }
4771
4772 btrfs_free_block_rsv(root, rsv);
4773
4774 /*
4775 * Errors here aren't a big deal, it just means we leave orphan items
4776 * in the tree. They will be cleaned up on the next mount.
4777 */
4778 if (ret == 0) {
4779 trans->block_rsv = root->orphan_block_rsv;
4780 btrfs_orphan_del(trans, inode);
4781 } else {
4782 btrfs_orphan_del(NULL, inode);
4783 }
4784
4785 trans->block_rsv = &root->fs_info->trans_block_rsv;
4786 if (!(root == root->fs_info->tree_root ||
4787 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4788 btrfs_return_ino(root, btrfs_ino(inode));
4789
4790 btrfs_end_transaction(trans, root);
4791 btrfs_btree_balance_dirty(root);
4792 no_delete:
4793 btrfs_remove_delayed_node(inode);
4794 clear_inode(inode);
4795 return;
4796 }
4797
4798 /*
4799 * this returns the key found in the dir entry in the location pointer.
4800 * If no dir entries were found, location->objectid is 0.
4801 */
4802 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4803 struct btrfs_key *location)
4804 {
4805 const char *name = dentry->d_name.name;
4806 int namelen = dentry->d_name.len;
4807 struct btrfs_dir_item *di;
4808 struct btrfs_path *path;
4809 struct btrfs_root *root = BTRFS_I(dir)->root;
4810 int ret = 0;
4811
4812 path = btrfs_alloc_path();
4813 if (!path)
4814 return -ENOMEM;
4815
4816 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4817 namelen, 0);
4818 if (IS_ERR(di))
4819 ret = PTR_ERR(di);
4820
4821 if (IS_ERR_OR_NULL(di))
4822 goto out_err;
4823
4824 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4825 out:
4826 btrfs_free_path(path);
4827 return ret;
4828 out_err:
4829 location->objectid = 0;
4830 goto out;
4831 }
4832
4833 /*
4834 * when we hit a tree root in a directory, the btrfs part of the inode
4835 * needs to be changed to reflect the root directory of the tree root. This
4836 * is kind of like crossing a mount point.
4837 */
4838 static int fixup_tree_root_location(struct btrfs_root *root,
4839 struct inode *dir,
4840 struct dentry *dentry,
4841 struct btrfs_key *location,
4842 struct btrfs_root **sub_root)
4843 {
4844 struct btrfs_path *path;
4845 struct btrfs_root *new_root;
4846 struct btrfs_root_ref *ref;
4847 struct extent_buffer *leaf;
4848 int ret;
4849 int err = 0;
4850
4851 path = btrfs_alloc_path();
4852 if (!path) {
4853 err = -ENOMEM;
4854 goto out;
4855 }
4856
4857 err = -ENOENT;
4858 ret = btrfs_find_item(root->fs_info->tree_root, path,
4859 BTRFS_I(dir)->root->root_key.objectid,
4860 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4861 if (ret) {
4862 if (ret < 0)
4863 err = ret;
4864 goto out;
4865 }
4866
4867 leaf = path->nodes[0];
4868 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4869 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4870 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4871 goto out;
4872
4873 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4874 (unsigned long)(ref + 1),
4875 dentry->d_name.len);
4876 if (ret)
4877 goto out;
4878
4879 btrfs_release_path(path);
4880
4881 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4882 if (IS_ERR(new_root)) {
4883 err = PTR_ERR(new_root);
4884 goto out;
4885 }
4886
4887 *sub_root = new_root;
4888 location->objectid = btrfs_root_dirid(&new_root->root_item);
4889 location->type = BTRFS_INODE_ITEM_KEY;
4890 location->offset = 0;
4891 err = 0;
4892 out:
4893 btrfs_free_path(path);
4894 return err;
4895 }
4896
4897 static void inode_tree_add(struct inode *inode)
4898 {
4899 struct btrfs_root *root = BTRFS_I(inode)->root;
4900 struct btrfs_inode *entry;
4901 struct rb_node **p;
4902 struct rb_node *parent;
4903 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4904 u64 ino = btrfs_ino(inode);
4905
4906 if (inode_unhashed(inode))
4907 return;
4908 parent = NULL;
4909 spin_lock(&root->inode_lock);
4910 p = &root->inode_tree.rb_node;
4911 while (*p) {
4912 parent = *p;
4913 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4914
4915 if (ino < btrfs_ino(&entry->vfs_inode))
4916 p = &parent->rb_left;
4917 else if (ino > btrfs_ino(&entry->vfs_inode))
4918 p = &parent->rb_right;
4919 else {
4920 WARN_ON(!(entry->vfs_inode.i_state &
4921 (I_WILL_FREE | I_FREEING)));
4922 rb_replace_node(parent, new, &root->inode_tree);
4923 RB_CLEAR_NODE(parent);
4924 spin_unlock(&root->inode_lock);
4925 return;
4926 }
4927 }
4928 rb_link_node(new, parent, p);
4929 rb_insert_color(new, &root->inode_tree);
4930 spin_unlock(&root->inode_lock);
4931 }
4932
4933 static void inode_tree_del(struct inode *inode)
4934 {
4935 struct btrfs_root *root = BTRFS_I(inode)->root;
4936 int empty = 0;
4937
4938 spin_lock(&root->inode_lock);
4939 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4940 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4941 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4942 empty = RB_EMPTY_ROOT(&root->inode_tree);
4943 }
4944 spin_unlock(&root->inode_lock);
4945
4946 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4947 synchronize_srcu(&root->fs_info->subvol_srcu);
4948 spin_lock(&root->inode_lock);
4949 empty = RB_EMPTY_ROOT(&root->inode_tree);
4950 spin_unlock(&root->inode_lock);
4951 if (empty)
4952 btrfs_add_dead_root(root);
4953 }
4954 }
4955
4956 void btrfs_invalidate_inodes(struct btrfs_root *root)
4957 {
4958 struct rb_node *node;
4959 struct rb_node *prev;
4960 struct btrfs_inode *entry;
4961 struct inode *inode;
4962 u64 objectid = 0;
4963
4964 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
4965 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4966
4967 spin_lock(&root->inode_lock);
4968 again:
4969 node = root->inode_tree.rb_node;
4970 prev = NULL;
4971 while (node) {
4972 prev = node;
4973 entry = rb_entry(node, struct btrfs_inode, rb_node);
4974
4975 if (objectid < btrfs_ino(&entry->vfs_inode))
4976 node = node->rb_left;
4977 else if (objectid > btrfs_ino(&entry->vfs_inode))
4978 node = node->rb_right;
4979 else
4980 break;
4981 }
4982 if (!node) {
4983 while (prev) {
4984 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4985 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4986 node = prev;
4987 break;
4988 }
4989 prev = rb_next(prev);
4990 }
4991 }
4992 while (node) {
4993 entry = rb_entry(node, struct btrfs_inode, rb_node);
4994 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4995 inode = igrab(&entry->vfs_inode);
4996 if (inode) {
4997 spin_unlock(&root->inode_lock);
4998 if (atomic_read(&inode->i_count) > 1)
4999 d_prune_aliases(inode);
5000 /*
5001 * btrfs_drop_inode will have it removed from
5002 * the inode cache when its usage count
5003 * hits zero.
5004 */
5005 iput(inode);
5006 cond_resched();
5007 spin_lock(&root->inode_lock);
5008 goto again;
5009 }
5010
5011 if (cond_resched_lock(&root->inode_lock))
5012 goto again;
5013
5014 node = rb_next(node);
5015 }
5016 spin_unlock(&root->inode_lock);
5017 }
5018
5019 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5020 {
5021 struct btrfs_iget_args *args = p;
5022 inode->i_ino = args->location->objectid;
5023 memcpy(&BTRFS_I(inode)->location, args->location,
5024 sizeof(*args->location));
5025 BTRFS_I(inode)->root = args->root;
5026 return 0;
5027 }
5028
5029 static int btrfs_find_actor(struct inode *inode, void *opaque)
5030 {
5031 struct btrfs_iget_args *args = opaque;
5032 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5033 args->root == BTRFS_I(inode)->root;
5034 }
5035
5036 static struct inode *btrfs_iget_locked(struct super_block *s,
5037 struct btrfs_key *location,
5038 struct btrfs_root *root)
5039 {
5040 struct inode *inode;
5041 struct btrfs_iget_args args;
5042 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5043
5044 args.location = location;
5045 args.root = root;
5046
5047 inode = iget5_locked(s, hashval, btrfs_find_actor,
5048 btrfs_init_locked_inode,
5049 (void *)&args);
5050 return inode;
5051 }
5052
5053 /* Get an inode object given its location and corresponding root.
5054 * Returns in *is_new if the inode was read from disk
5055 */
5056 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5057 struct btrfs_root *root, int *new)
5058 {
5059 struct inode *inode;
5060
5061 inode = btrfs_iget_locked(s, location, root);
5062 if (!inode)
5063 return ERR_PTR(-ENOMEM);
5064
5065 if (inode->i_state & I_NEW) {
5066 btrfs_read_locked_inode(inode);
5067 if (!is_bad_inode(inode)) {
5068 inode_tree_add(inode);
5069 unlock_new_inode(inode);
5070 if (new)
5071 *new = 1;
5072 } else {
5073 unlock_new_inode(inode);
5074 iput(inode);
5075 inode = ERR_PTR(-ESTALE);
5076 }
5077 }
5078
5079 return inode;
5080 }
5081
5082 static struct inode *new_simple_dir(struct super_block *s,
5083 struct btrfs_key *key,
5084 struct btrfs_root *root)
5085 {
5086 struct inode *inode = new_inode(s);
5087
5088 if (!inode)
5089 return ERR_PTR(-ENOMEM);
5090
5091 BTRFS_I(inode)->root = root;
5092 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5093 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5094
5095 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5096 inode->i_op = &btrfs_dir_ro_inode_operations;
5097 inode->i_fop = &simple_dir_operations;
5098 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5099 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5100
5101 return inode;
5102 }
5103
5104 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5105 {
5106 struct inode *inode;
5107 struct btrfs_root *root = BTRFS_I(dir)->root;
5108 struct btrfs_root *sub_root = root;
5109 struct btrfs_key location;
5110 int index;
5111 int ret = 0;
5112
5113 if (dentry->d_name.len > BTRFS_NAME_LEN)
5114 return ERR_PTR(-ENAMETOOLONG);
5115
5116 ret = btrfs_inode_by_name(dir, dentry, &location);
5117 if (ret < 0)
5118 return ERR_PTR(ret);
5119
5120 if (location.objectid == 0)
5121 return ERR_PTR(-ENOENT);
5122
5123 if (location.type == BTRFS_INODE_ITEM_KEY) {
5124 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5125 return inode;
5126 }
5127
5128 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5129
5130 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5131 ret = fixup_tree_root_location(root, dir, dentry,
5132 &location, &sub_root);
5133 if (ret < 0) {
5134 if (ret != -ENOENT)
5135 inode = ERR_PTR(ret);
5136 else
5137 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5138 } else {
5139 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5140 }
5141 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5142
5143 if (!IS_ERR(inode) && root != sub_root) {
5144 down_read(&root->fs_info->cleanup_work_sem);
5145 if (!(inode->i_sb->s_flags & MS_RDONLY))
5146 ret = btrfs_orphan_cleanup(sub_root);
5147 up_read(&root->fs_info->cleanup_work_sem);
5148 if (ret) {
5149 iput(inode);
5150 inode = ERR_PTR(ret);
5151 }
5152 }
5153
5154 return inode;
5155 }
5156
5157 static int btrfs_dentry_delete(const struct dentry *dentry)
5158 {
5159 struct btrfs_root *root;
5160 struct inode *inode = dentry->d_inode;
5161
5162 if (!inode && !IS_ROOT(dentry))
5163 inode = dentry->d_parent->d_inode;
5164
5165 if (inode) {
5166 root = BTRFS_I(inode)->root;
5167 if (btrfs_root_refs(&root->root_item) == 0)
5168 return 1;
5169
5170 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5171 return 1;
5172 }
5173 return 0;
5174 }
5175
5176 static void btrfs_dentry_release(struct dentry *dentry)
5177 {
5178 kfree(dentry->d_fsdata);
5179 }
5180
5181 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5182 unsigned int flags)
5183 {
5184 struct inode *inode;
5185
5186 inode = btrfs_lookup_dentry(dir, dentry);
5187 if (IS_ERR(inode)) {
5188 if (PTR_ERR(inode) == -ENOENT)
5189 inode = NULL;
5190 else
5191 return ERR_CAST(inode);
5192 }
5193
5194 return d_materialise_unique(dentry, inode);
5195 }
5196
5197 unsigned char btrfs_filetype_table[] = {
5198 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5199 };
5200
5201 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5202 {
5203 struct inode *inode = file_inode(file);
5204 struct btrfs_root *root = BTRFS_I(inode)->root;
5205 struct btrfs_item *item;
5206 struct btrfs_dir_item *di;
5207 struct btrfs_key key;
5208 struct btrfs_key found_key;
5209 struct btrfs_path *path;
5210 struct list_head ins_list;
5211 struct list_head del_list;
5212 int ret;
5213 struct extent_buffer *leaf;
5214 int slot;
5215 unsigned char d_type;
5216 int over = 0;
5217 u32 di_cur;
5218 u32 di_total;
5219 u32 di_len;
5220 int key_type = BTRFS_DIR_INDEX_KEY;
5221 char tmp_name[32];
5222 char *name_ptr;
5223 int name_len;
5224 int is_curr = 0; /* ctx->pos points to the current index? */
5225
5226 /* FIXME, use a real flag for deciding about the key type */
5227 if (root->fs_info->tree_root == root)
5228 key_type = BTRFS_DIR_ITEM_KEY;
5229
5230 if (!dir_emit_dots(file, ctx))
5231 return 0;
5232
5233 path = btrfs_alloc_path();
5234 if (!path)
5235 return -ENOMEM;
5236
5237 path->reada = 1;
5238
5239 if (key_type == BTRFS_DIR_INDEX_KEY) {
5240 INIT_LIST_HEAD(&ins_list);
5241 INIT_LIST_HEAD(&del_list);
5242 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5243 }
5244
5245 btrfs_set_key_type(&key, key_type);
5246 key.offset = ctx->pos;
5247 key.objectid = btrfs_ino(inode);
5248
5249 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5250 if (ret < 0)
5251 goto err;
5252
5253 while (1) {
5254 leaf = path->nodes[0];
5255 slot = path->slots[0];
5256 if (slot >= btrfs_header_nritems(leaf)) {
5257 ret = btrfs_next_leaf(root, path);
5258 if (ret < 0)
5259 goto err;
5260 else if (ret > 0)
5261 break;
5262 continue;
5263 }
5264
5265 item = btrfs_item_nr(slot);
5266 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5267
5268 if (found_key.objectid != key.objectid)
5269 break;
5270 if (btrfs_key_type(&found_key) != key_type)
5271 break;
5272 if (found_key.offset < ctx->pos)
5273 goto next;
5274 if (key_type == BTRFS_DIR_INDEX_KEY &&
5275 btrfs_should_delete_dir_index(&del_list,
5276 found_key.offset))
5277 goto next;
5278
5279 ctx->pos = found_key.offset;
5280 is_curr = 1;
5281
5282 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5283 di_cur = 0;
5284 di_total = btrfs_item_size(leaf, item);
5285
5286 while (di_cur < di_total) {
5287 struct btrfs_key location;
5288
5289 if (verify_dir_item(root, leaf, di))
5290 break;
5291
5292 name_len = btrfs_dir_name_len(leaf, di);
5293 if (name_len <= sizeof(tmp_name)) {
5294 name_ptr = tmp_name;
5295 } else {
5296 name_ptr = kmalloc(name_len, GFP_NOFS);
5297 if (!name_ptr) {
5298 ret = -ENOMEM;
5299 goto err;
5300 }
5301 }
5302 read_extent_buffer(leaf, name_ptr,
5303 (unsigned long)(di + 1), name_len);
5304
5305 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5306 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5307
5308
5309 /* is this a reference to our own snapshot? If so
5310 * skip it.
5311 *
5312 * In contrast to old kernels, we insert the snapshot's
5313 * dir item and dir index after it has been created, so
5314 * we won't find a reference to our own snapshot. We
5315 * still keep the following code for backward
5316 * compatibility.
5317 */
5318 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5319 location.objectid == root->root_key.objectid) {
5320 over = 0;
5321 goto skip;
5322 }
5323 over = !dir_emit(ctx, name_ptr, name_len,
5324 location.objectid, d_type);
5325
5326 skip:
5327 if (name_ptr != tmp_name)
5328 kfree(name_ptr);
5329
5330 if (over)
5331 goto nopos;
5332 di_len = btrfs_dir_name_len(leaf, di) +
5333 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5334 di_cur += di_len;
5335 di = (struct btrfs_dir_item *)((char *)di + di_len);
5336 }
5337 next:
5338 path->slots[0]++;
5339 }
5340
5341 if (key_type == BTRFS_DIR_INDEX_KEY) {
5342 if (is_curr)
5343 ctx->pos++;
5344 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5345 if (ret)
5346 goto nopos;
5347 }
5348
5349 /* Reached end of directory/root. Bump pos past the last item. */
5350 ctx->pos++;
5351
5352 /*
5353 * Stop new entries from being returned after we return the last
5354 * entry.
5355 *
5356 * New directory entries are assigned a strictly increasing
5357 * offset. This means that new entries created during readdir
5358 * are *guaranteed* to be seen in the future by that readdir.
5359 * This has broken buggy programs which operate on names as
5360 * they're returned by readdir. Until we re-use freed offsets
5361 * we have this hack to stop new entries from being returned
5362 * under the assumption that they'll never reach this huge
5363 * offset.
5364 *
5365 * This is being careful not to overflow 32bit loff_t unless the
5366 * last entry requires it because doing so has broken 32bit apps
5367 * in the past.
5368 */
5369 if (key_type == BTRFS_DIR_INDEX_KEY) {
5370 if (ctx->pos >= INT_MAX)
5371 ctx->pos = LLONG_MAX;
5372 else
5373 ctx->pos = INT_MAX;
5374 }
5375 nopos:
5376 ret = 0;
5377 err:
5378 if (key_type == BTRFS_DIR_INDEX_KEY)
5379 btrfs_put_delayed_items(&ins_list, &del_list);
5380 btrfs_free_path(path);
5381 return ret;
5382 }
5383
5384 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5385 {
5386 struct btrfs_root *root = BTRFS_I(inode)->root;
5387 struct btrfs_trans_handle *trans;
5388 int ret = 0;
5389 bool nolock = false;
5390
5391 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5392 return 0;
5393
5394 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5395 nolock = true;
5396
5397 if (wbc->sync_mode == WB_SYNC_ALL) {
5398 if (nolock)
5399 trans = btrfs_join_transaction_nolock(root);
5400 else
5401 trans = btrfs_join_transaction(root);
5402 if (IS_ERR(trans))
5403 return PTR_ERR(trans);
5404 ret = btrfs_commit_transaction(trans, root);
5405 }
5406 return ret;
5407 }
5408
5409 /*
5410 * This is somewhat expensive, updating the tree every time the
5411 * inode changes. But, it is most likely to find the inode in cache.
5412 * FIXME, needs more benchmarking...there are no reasons other than performance
5413 * to keep or drop this code.
5414 */
5415 static int btrfs_dirty_inode(struct inode *inode)
5416 {
5417 struct btrfs_root *root = BTRFS_I(inode)->root;
5418 struct btrfs_trans_handle *trans;
5419 int ret;
5420
5421 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5422 return 0;
5423
5424 trans = btrfs_join_transaction(root);
5425 if (IS_ERR(trans))
5426 return PTR_ERR(trans);
5427
5428 ret = btrfs_update_inode(trans, root, inode);
5429 if (ret && ret == -ENOSPC) {
5430 /* whoops, lets try again with the full transaction */
5431 btrfs_end_transaction(trans, root);
5432 trans = btrfs_start_transaction(root, 1);
5433 if (IS_ERR(trans))
5434 return PTR_ERR(trans);
5435
5436 ret = btrfs_update_inode(trans, root, inode);
5437 }
5438 btrfs_end_transaction(trans, root);
5439 if (BTRFS_I(inode)->delayed_node)
5440 btrfs_balance_delayed_items(root);
5441
5442 return ret;
5443 }
5444
5445 /*
5446 * This is a copy of file_update_time. We need this so we can return error on
5447 * ENOSPC for updating the inode in the case of file write and mmap writes.
5448 */
5449 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5450 int flags)
5451 {
5452 struct btrfs_root *root = BTRFS_I(inode)->root;
5453
5454 if (btrfs_root_readonly(root))
5455 return -EROFS;
5456
5457 if (flags & S_VERSION)
5458 inode_inc_iversion(inode);
5459 if (flags & S_CTIME)
5460 inode->i_ctime = *now;
5461 if (flags & S_MTIME)
5462 inode->i_mtime = *now;
5463 if (flags & S_ATIME)
5464 inode->i_atime = *now;
5465 return btrfs_dirty_inode(inode);
5466 }
5467
5468 /*
5469 * find the highest existing sequence number in a directory
5470 * and then set the in-memory index_cnt variable to reflect
5471 * free sequence numbers
5472 */
5473 static int btrfs_set_inode_index_count(struct inode *inode)
5474 {
5475 struct btrfs_root *root = BTRFS_I(inode)->root;
5476 struct btrfs_key key, found_key;
5477 struct btrfs_path *path;
5478 struct extent_buffer *leaf;
5479 int ret;
5480
5481 key.objectid = btrfs_ino(inode);
5482 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5483 key.offset = (u64)-1;
5484
5485 path = btrfs_alloc_path();
5486 if (!path)
5487 return -ENOMEM;
5488
5489 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5490 if (ret < 0)
5491 goto out;
5492 /* FIXME: we should be able to handle this */
5493 if (ret == 0)
5494 goto out;
5495 ret = 0;
5496
5497 /*
5498 * MAGIC NUMBER EXPLANATION:
5499 * since we search a directory based on f_pos we have to start at 2
5500 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5501 * else has to start at 2
5502 */
5503 if (path->slots[0] == 0) {
5504 BTRFS_I(inode)->index_cnt = 2;
5505 goto out;
5506 }
5507
5508 path->slots[0]--;
5509
5510 leaf = path->nodes[0];
5511 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5512
5513 if (found_key.objectid != btrfs_ino(inode) ||
5514 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5515 BTRFS_I(inode)->index_cnt = 2;
5516 goto out;
5517 }
5518
5519 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5520 out:
5521 btrfs_free_path(path);
5522 return ret;
5523 }
5524
5525 /*
5526 * helper to find a free sequence number in a given directory. This current
5527 * code is very simple, later versions will do smarter things in the btree
5528 */
5529 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5530 {
5531 int ret = 0;
5532
5533 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5534 ret = btrfs_inode_delayed_dir_index_count(dir);
5535 if (ret) {
5536 ret = btrfs_set_inode_index_count(dir);
5537 if (ret)
5538 return ret;
5539 }
5540 }
5541
5542 *index = BTRFS_I(dir)->index_cnt;
5543 BTRFS_I(dir)->index_cnt++;
5544
5545 return ret;
5546 }
5547
5548 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5549 struct btrfs_root *root,
5550 struct inode *dir,
5551 const char *name, int name_len,
5552 u64 ref_objectid, u64 objectid,
5553 umode_t mode, u64 *index)
5554 {
5555 struct inode *inode;
5556 struct btrfs_inode_item *inode_item;
5557 struct btrfs_key *location;
5558 struct btrfs_path *path;
5559 struct btrfs_inode_ref *ref;
5560 struct btrfs_key key[2];
5561 u32 sizes[2];
5562 int nitems = name ? 2 : 1;
5563 unsigned long ptr;
5564 int ret;
5565
5566 path = btrfs_alloc_path();
5567 if (!path)
5568 return ERR_PTR(-ENOMEM);
5569
5570 inode = new_inode(root->fs_info->sb);
5571 if (!inode) {
5572 btrfs_free_path(path);
5573 return ERR_PTR(-ENOMEM);
5574 }
5575
5576 /*
5577 * we have to initialize this early, so we can reclaim the inode
5578 * number if we fail afterwards in this function.
5579 */
5580 inode->i_ino = objectid;
5581
5582 if (dir && name) {
5583 trace_btrfs_inode_request(dir);
5584
5585 ret = btrfs_set_inode_index(dir, index);
5586 if (ret) {
5587 btrfs_free_path(path);
5588 iput(inode);
5589 return ERR_PTR(ret);
5590 }
5591 } else if (dir) {
5592 *index = 0;
5593 }
5594 /*
5595 * index_cnt is ignored for everything but a dir,
5596 * btrfs_get_inode_index_count has an explanation for the magic
5597 * number
5598 */
5599 BTRFS_I(inode)->index_cnt = 2;
5600 BTRFS_I(inode)->dir_index = *index;
5601 BTRFS_I(inode)->root = root;
5602 BTRFS_I(inode)->generation = trans->transid;
5603 inode->i_generation = BTRFS_I(inode)->generation;
5604
5605 /*
5606 * We could have gotten an inode number from somebody who was fsynced
5607 * and then removed in this same transaction, so let's just set full
5608 * sync since it will be a full sync anyway and this will blow away the
5609 * old info in the log.
5610 */
5611 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5612
5613 key[0].objectid = objectid;
5614 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5615 key[0].offset = 0;
5616
5617 sizes[0] = sizeof(struct btrfs_inode_item);
5618
5619 if (name) {
5620 /*
5621 * Start new inodes with an inode_ref. This is slightly more
5622 * efficient for small numbers of hard links since they will
5623 * be packed into one item. Extended refs will kick in if we
5624 * add more hard links than can fit in the ref item.
5625 */
5626 key[1].objectid = objectid;
5627 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5628 key[1].offset = ref_objectid;
5629
5630 sizes[1] = name_len + sizeof(*ref);
5631 }
5632
5633 path->leave_spinning = 1;
5634 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5635 if (ret != 0)
5636 goto fail;
5637
5638 inode_init_owner(inode, dir, mode);
5639 inode_set_bytes(inode, 0);
5640 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5641 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5642 struct btrfs_inode_item);
5643 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5644 sizeof(*inode_item));
5645 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5646
5647 if (name) {
5648 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5649 struct btrfs_inode_ref);
5650 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5651 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5652 ptr = (unsigned long)(ref + 1);
5653 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5654 }
5655
5656 btrfs_mark_buffer_dirty(path->nodes[0]);
5657 btrfs_free_path(path);
5658
5659 location = &BTRFS_I(inode)->location;
5660 location->objectid = objectid;
5661 location->offset = 0;
5662 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5663
5664 btrfs_inherit_iflags(inode, dir);
5665
5666 if (S_ISREG(mode)) {
5667 if (btrfs_test_opt(root, NODATASUM))
5668 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5669 if (btrfs_test_opt(root, NODATACOW))
5670 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5671 BTRFS_INODE_NODATASUM;
5672 }
5673
5674 btrfs_insert_inode_hash(inode);
5675 inode_tree_add(inode);
5676
5677 trace_btrfs_inode_new(inode);
5678 btrfs_set_inode_last_trans(trans, inode);
5679
5680 btrfs_update_root_times(trans, root);
5681
5682 ret = btrfs_inode_inherit_props(trans, inode, dir);
5683 if (ret)
5684 btrfs_err(root->fs_info,
5685 "error inheriting props for ino %llu (root %llu): %d",
5686 btrfs_ino(inode), root->root_key.objectid, ret);
5687
5688 return inode;
5689 fail:
5690 if (dir && name)
5691 BTRFS_I(dir)->index_cnt--;
5692 btrfs_free_path(path);
5693 iput(inode);
5694 return ERR_PTR(ret);
5695 }
5696
5697 static inline u8 btrfs_inode_type(struct inode *inode)
5698 {
5699 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5700 }
5701
5702 /*
5703 * utility function to add 'inode' into 'parent_inode' with
5704 * a give name and a given sequence number.
5705 * if 'add_backref' is true, also insert a backref from the
5706 * inode to the parent directory.
5707 */
5708 int btrfs_add_link(struct btrfs_trans_handle *trans,
5709 struct inode *parent_inode, struct inode *inode,
5710 const char *name, int name_len, int add_backref, u64 index)
5711 {
5712 int ret = 0;
5713 struct btrfs_key key;
5714 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5715 u64 ino = btrfs_ino(inode);
5716 u64 parent_ino = btrfs_ino(parent_inode);
5717
5718 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5719 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5720 } else {
5721 key.objectid = ino;
5722 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5723 key.offset = 0;
5724 }
5725
5726 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5727 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5728 key.objectid, root->root_key.objectid,
5729 parent_ino, index, name, name_len);
5730 } else if (add_backref) {
5731 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5732 parent_ino, index);
5733 }
5734
5735 /* Nothing to clean up yet */
5736 if (ret)
5737 return ret;
5738
5739 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5740 parent_inode, &key,
5741 btrfs_inode_type(inode), index);
5742 if (ret == -EEXIST || ret == -EOVERFLOW)
5743 goto fail_dir_item;
5744 else if (ret) {
5745 btrfs_abort_transaction(trans, root, ret);
5746 return ret;
5747 }
5748
5749 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5750 name_len * 2);
5751 inode_inc_iversion(parent_inode);
5752 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5753 ret = btrfs_update_inode(trans, root, parent_inode);
5754 if (ret)
5755 btrfs_abort_transaction(trans, root, ret);
5756 return ret;
5757
5758 fail_dir_item:
5759 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5760 u64 local_index;
5761 int err;
5762 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5763 key.objectid, root->root_key.objectid,
5764 parent_ino, &local_index, name, name_len);
5765
5766 } else if (add_backref) {
5767 u64 local_index;
5768 int err;
5769
5770 err = btrfs_del_inode_ref(trans, root, name, name_len,
5771 ino, parent_ino, &local_index);
5772 }
5773 return ret;
5774 }
5775
5776 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5777 struct inode *dir, struct dentry *dentry,
5778 struct inode *inode, int backref, u64 index)
5779 {
5780 int err = btrfs_add_link(trans, dir, inode,
5781 dentry->d_name.name, dentry->d_name.len,
5782 backref, index);
5783 if (err > 0)
5784 err = -EEXIST;
5785 return err;
5786 }
5787
5788 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5789 umode_t mode, dev_t rdev)
5790 {
5791 struct btrfs_trans_handle *trans;
5792 struct btrfs_root *root = BTRFS_I(dir)->root;
5793 struct inode *inode = NULL;
5794 int err;
5795 int drop_inode = 0;
5796 u64 objectid;
5797 u64 index = 0;
5798
5799 if (!new_valid_dev(rdev))
5800 return -EINVAL;
5801
5802 /*
5803 * 2 for inode item and ref
5804 * 2 for dir items
5805 * 1 for xattr if selinux is on
5806 */
5807 trans = btrfs_start_transaction(root, 5);
5808 if (IS_ERR(trans))
5809 return PTR_ERR(trans);
5810
5811 err = btrfs_find_free_ino(root, &objectid);
5812 if (err)
5813 goto out_unlock;
5814
5815 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5816 dentry->d_name.len, btrfs_ino(dir), objectid,
5817 mode, &index);
5818 if (IS_ERR(inode)) {
5819 err = PTR_ERR(inode);
5820 goto out_unlock;
5821 }
5822
5823 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5824 if (err) {
5825 drop_inode = 1;
5826 goto out_unlock;
5827 }
5828
5829 /*
5830 * If the active LSM wants to access the inode during
5831 * d_instantiate it needs these. Smack checks to see
5832 * if the filesystem supports xattrs by looking at the
5833 * ops vector.
5834 */
5835
5836 inode->i_op = &btrfs_special_inode_operations;
5837 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5838 if (err)
5839 drop_inode = 1;
5840 else {
5841 init_special_inode(inode, inode->i_mode, rdev);
5842 btrfs_update_inode(trans, root, inode);
5843 d_instantiate(dentry, inode);
5844 }
5845 out_unlock:
5846 btrfs_end_transaction(trans, root);
5847 btrfs_balance_delayed_items(root);
5848 btrfs_btree_balance_dirty(root);
5849 if (drop_inode) {
5850 inode_dec_link_count(inode);
5851 iput(inode);
5852 }
5853 return err;
5854 }
5855
5856 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5857 umode_t mode, bool excl)
5858 {
5859 struct btrfs_trans_handle *trans;
5860 struct btrfs_root *root = BTRFS_I(dir)->root;
5861 struct inode *inode = NULL;
5862 int drop_inode_on_err = 0;
5863 int err;
5864 u64 objectid;
5865 u64 index = 0;
5866
5867 /*
5868 * 2 for inode item and ref
5869 * 2 for dir items
5870 * 1 for xattr if selinux is on
5871 */
5872 trans = btrfs_start_transaction(root, 5);
5873 if (IS_ERR(trans))
5874 return PTR_ERR(trans);
5875
5876 err = btrfs_find_free_ino(root, &objectid);
5877 if (err)
5878 goto out_unlock;
5879
5880 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5881 dentry->d_name.len, btrfs_ino(dir), objectid,
5882 mode, &index);
5883 if (IS_ERR(inode)) {
5884 err = PTR_ERR(inode);
5885 goto out_unlock;
5886 }
5887 drop_inode_on_err = 1;
5888
5889 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5890 if (err)
5891 goto out_unlock;
5892
5893 err = btrfs_update_inode(trans, root, inode);
5894 if (err)
5895 goto out_unlock;
5896
5897 /*
5898 * If the active LSM wants to access the inode during
5899 * d_instantiate it needs these. Smack checks to see
5900 * if the filesystem supports xattrs by looking at the
5901 * ops vector.
5902 */
5903 inode->i_fop = &btrfs_file_operations;
5904 inode->i_op = &btrfs_file_inode_operations;
5905
5906 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5907 if (err)
5908 goto out_unlock;
5909
5910 inode->i_mapping->a_ops = &btrfs_aops;
5911 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5912 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5913 d_instantiate(dentry, inode);
5914
5915 out_unlock:
5916 btrfs_end_transaction(trans, root);
5917 if (err && drop_inode_on_err) {
5918 inode_dec_link_count(inode);
5919 iput(inode);
5920 }
5921 btrfs_balance_delayed_items(root);
5922 btrfs_btree_balance_dirty(root);
5923 return err;
5924 }
5925
5926 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5927 struct dentry *dentry)
5928 {
5929 struct btrfs_trans_handle *trans;
5930 struct btrfs_root *root = BTRFS_I(dir)->root;
5931 struct inode *inode = old_dentry->d_inode;
5932 u64 index;
5933 int err;
5934 int drop_inode = 0;
5935
5936 /* do not allow sys_link's with other subvols of the same device */
5937 if (root->objectid != BTRFS_I(inode)->root->objectid)
5938 return -EXDEV;
5939
5940 if (inode->i_nlink >= BTRFS_LINK_MAX)
5941 return -EMLINK;
5942
5943 err = btrfs_set_inode_index(dir, &index);
5944 if (err)
5945 goto fail;
5946
5947 /*
5948 * 2 items for inode and inode ref
5949 * 2 items for dir items
5950 * 1 item for parent inode
5951 */
5952 trans = btrfs_start_transaction(root, 5);
5953 if (IS_ERR(trans)) {
5954 err = PTR_ERR(trans);
5955 goto fail;
5956 }
5957
5958 /* There are several dir indexes for this inode, clear the cache. */
5959 BTRFS_I(inode)->dir_index = 0ULL;
5960 inc_nlink(inode);
5961 inode_inc_iversion(inode);
5962 inode->i_ctime = CURRENT_TIME;
5963 ihold(inode);
5964 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5965
5966 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5967
5968 if (err) {
5969 drop_inode = 1;
5970 } else {
5971 struct dentry *parent = dentry->d_parent;
5972 err = btrfs_update_inode(trans, root, inode);
5973 if (err)
5974 goto fail;
5975 if (inode->i_nlink == 1) {
5976 /*
5977 * If new hard link count is 1, it's a file created
5978 * with open(2) O_TMPFILE flag.
5979 */
5980 err = btrfs_orphan_del(trans, inode);
5981 if (err)
5982 goto fail;
5983 }
5984 d_instantiate(dentry, inode);
5985 btrfs_log_new_name(trans, inode, NULL, parent);
5986 }
5987
5988 btrfs_end_transaction(trans, root);
5989 btrfs_balance_delayed_items(root);
5990 fail:
5991 if (drop_inode) {
5992 inode_dec_link_count(inode);
5993 iput(inode);
5994 }
5995 btrfs_btree_balance_dirty(root);
5996 return err;
5997 }
5998
5999 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6000 {
6001 struct inode *inode = NULL;
6002 struct btrfs_trans_handle *trans;
6003 struct btrfs_root *root = BTRFS_I(dir)->root;
6004 int err = 0;
6005 int drop_on_err = 0;
6006 u64 objectid = 0;
6007 u64 index = 0;
6008
6009 /*
6010 * 2 items for inode and ref
6011 * 2 items for dir items
6012 * 1 for xattr if selinux is on
6013 */
6014 trans = btrfs_start_transaction(root, 5);
6015 if (IS_ERR(trans))
6016 return PTR_ERR(trans);
6017
6018 err = btrfs_find_free_ino(root, &objectid);
6019 if (err)
6020 goto out_fail;
6021
6022 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6023 dentry->d_name.len, btrfs_ino(dir), objectid,
6024 S_IFDIR | mode, &index);
6025 if (IS_ERR(inode)) {
6026 err = PTR_ERR(inode);
6027 goto out_fail;
6028 }
6029
6030 drop_on_err = 1;
6031
6032 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6033 if (err)
6034 goto out_fail;
6035
6036 inode->i_op = &btrfs_dir_inode_operations;
6037 inode->i_fop = &btrfs_dir_file_operations;
6038
6039 btrfs_i_size_write(inode, 0);
6040 err = btrfs_update_inode(trans, root, inode);
6041 if (err)
6042 goto out_fail;
6043
6044 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6045 dentry->d_name.len, 0, index);
6046 if (err)
6047 goto out_fail;
6048
6049 d_instantiate(dentry, inode);
6050 drop_on_err = 0;
6051
6052 out_fail:
6053 btrfs_end_transaction(trans, root);
6054 if (drop_on_err)
6055 iput(inode);
6056 btrfs_balance_delayed_items(root);
6057 btrfs_btree_balance_dirty(root);
6058 return err;
6059 }
6060
6061 /* helper for btfs_get_extent. Given an existing extent in the tree,
6062 * and an extent that you want to insert, deal with overlap and insert
6063 * the new extent into the tree.
6064 */
6065 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6066 struct extent_map *existing,
6067 struct extent_map *em,
6068 u64 map_start, u64 map_len)
6069 {
6070 u64 start_diff;
6071
6072 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6073 start_diff = map_start - em->start;
6074 em->start = map_start;
6075 em->len = map_len;
6076 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6077 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6078 em->block_start += start_diff;
6079 em->block_len -= start_diff;
6080 }
6081 return add_extent_mapping(em_tree, em, 0);
6082 }
6083
6084 static noinline int uncompress_inline(struct btrfs_path *path,
6085 struct inode *inode, struct page *page,
6086 size_t pg_offset, u64 extent_offset,
6087 struct btrfs_file_extent_item *item)
6088 {
6089 int ret;
6090 struct extent_buffer *leaf = path->nodes[0];
6091 char *tmp;
6092 size_t max_size;
6093 unsigned long inline_size;
6094 unsigned long ptr;
6095 int compress_type;
6096
6097 WARN_ON(pg_offset != 0);
6098 compress_type = btrfs_file_extent_compression(leaf, item);
6099 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6100 inline_size = btrfs_file_extent_inline_item_len(leaf,
6101 btrfs_item_nr(path->slots[0]));
6102 tmp = kmalloc(inline_size, GFP_NOFS);
6103 if (!tmp)
6104 return -ENOMEM;
6105 ptr = btrfs_file_extent_inline_start(item);
6106
6107 read_extent_buffer(leaf, tmp, ptr, inline_size);
6108
6109 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6110 ret = btrfs_decompress(compress_type, tmp, page,
6111 extent_offset, inline_size, max_size);
6112 kfree(tmp);
6113 return ret;
6114 }
6115
6116 /*
6117 * a bit scary, this does extent mapping from logical file offset to the disk.
6118 * the ugly parts come from merging extents from the disk with the in-ram
6119 * representation. This gets more complex because of the data=ordered code,
6120 * where the in-ram extents might be locked pending data=ordered completion.
6121 *
6122 * This also copies inline extents directly into the page.
6123 */
6124
6125 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6126 size_t pg_offset, u64 start, u64 len,
6127 int create)
6128 {
6129 int ret;
6130 int err = 0;
6131 u64 bytenr;
6132 u64 extent_start = 0;
6133 u64 extent_end = 0;
6134 u64 objectid = btrfs_ino(inode);
6135 u32 found_type;
6136 struct btrfs_path *path = NULL;
6137 struct btrfs_root *root = BTRFS_I(inode)->root;
6138 struct btrfs_file_extent_item *item;
6139 struct extent_buffer *leaf;
6140 struct btrfs_key found_key;
6141 struct extent_map *em = NULL;
6142 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6143 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6144 struct btrfs_trans_handle *trans = NULL;
6145 int compress_type;
6146
6147 again:
6148 read_lock(&em_tree->lock);
6149 em = lookup_extent_mapping(em_tree, start, len);
6150 if (em)
6151 em->bdev = root->fs_info->fs_devices->latest_bdev;
6152 read_unlock(&em_tree->lock);
6153
6154 if (em) {
6155 if (em->start > start || em->start + em->len <= start)
6156 free_extent_map(em);
6157 else if (em->block_start == EXTENT_MAP_INLINE && page)
6158 free_extent_map(em);
6159 else
6160 goto out;
6161 }
6162 em = alloc_extent_map();
6163 if (!em) {
6164 err = -ENOMEM;
6165 goto out;
6166 }
6167 em->bdev = root->fs_info->fs_devices->latest_bdev;
6168 em->start = EXTENT_MAP_HOLE;
6169 em->orig_start = EXTENT_MAP_HOLE;
6170 em->len = (u64)-1;
6171 em->block_len = (u64)-1;
6172
6173 if (!path) {
6174 path = btrfs_alloc_path();
6175 if (!path) {
6176 err = -ENOMEM;
6177 goto out;
6178 }
6179 /*
6180 * Chances are we'll be called again, so go ahead and do
6181 * readahead
6182 */
6183 path->reada = 1;
6184 }
6185
6186 ret = btrfs_lookup_file_extent(trans, root, path,
6187 objectid, start, trans != NULL);
6188 if (ret < 0) {
6189 err = ret;
6190 goto out;
6191 }
6192
6193 if (ret != 0) {
6194 if (path->slots[0] == 0)
6195 goto not_found;
6196 path->slots[0]--;
6197 }
6198
6199 leaf = path->nodes[0];
6200 item = btrfs_item_ptr(leaf, path->slots[0],
6201 struct btrfs_file_extent_item);
6202 /* are we inside the extent that was found? */
6203 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6204 found_type = btrfs_key_type(&found_key);
6205 if (found_key.objectid != objectid ||
6206 found_type != BTRFS_EXTENT_DATA_KEY) {
6207 /*
6208 * If we backup past the first extent we want to move forward
6209 * and see if there is an extent in front of us, otherwise we'll
6210 * say there is a hole for our whole search range which can
6211 * cause problems.
6212 */
6213 extent_end = start;
6214 goto next;
6215 }
6216
6217 found_type = btrfs_file_extent_type(leaf, item);
6218 extent_start = found_key.offset;
6219 compress_type = btrfs_file_extent_compression(leaf, item);
6220 if (found_type == BTRFS_FILE_EXTENT_REG ||
6221 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6222 extent_end = extent_start +
6223 btrfs_file_extent_num_bytes(leaf, item);
6224 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6225 size_t size;
6226 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6227 extent_end = ALIGN(extent_start + size, root->sectorsize);
6228 }
6229 next:
6230 if (start >= extent_end) {
6231 path->slots[0]++;
6232 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6233 ret = btrfs_next_leaf(root, path);
6234 if (ret < 0) {
6235 err = ret;
6236 goto out;
6237 }
6238 if (ret > 0)
6239 goto not_found;
6240 leaf = path->nodes[0];
6241 }
6242 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6243 if (found_key.objectid != objectid ||
6244 found_key.type != BTRFS_EXTENT_DATA_KEY)
6245 goto not_found;
6246 if (start + len <= found_key.offset)
6247 goto not_found;
6248 em->start = start;
6249 em->orig_start = start;
6250 em->len = found_key.offset - start;
6251 goto not_found_em;
6252 }
6253
6254 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6255 if (found_type == BTRFS_FILE_EXTENT_REG ||
6256 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6257 em->start = extent_start;
6258 em->len = extent_end - extent_start;
6259 em->orig_start = extent_start -
6260 btrfs_file_extent_offset(leaf, item);
6261 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6262 item);
6263 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6264 if (bytenr == 0) {
6265 em->block_start = EXTENT_MAP_HOLE;
6266 goto insert;
6267 }
6268 if (compress_type != BTRFS_COMPRESS_NONE) {
6269 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6270 em->compress_type = compress_type;
6271 em->block_start = bytenr;
6272 em->block_len = em->orig_block_len;
6273 } else {
6274 bytenr += btrfs_file_extent_offset(leaf, item);
6275 em->block_start = bytenr;
6276 em->block_len = em->len;
6277 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6278 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6279 }
6280 goto insert;
6281 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6282 unsigned long ptr;
6283 char *map;
6284 size_t size;
6285 size_t extent_offset;
6286 size_t copy_size;
6287
6288 em->block_start = EXTENT_MAP_INLINE;
6289 if (!page || create) {
6290 em->start = extent_start;
6291 em->len = extent_end - extent_start;
6292 goto out;
6293 }
6294
6295 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6296 extent_offset = page_offset(page) + pg_offset - extent_start;
6297 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6298 size - extent_offset);
6299 em->start = extent_start + extent_offset;
6300 em->len = ALIGN(copy_size, root->sectorsize);
6301 em->orig_block_len = em->len;
6302 em->orig_start = em->start;
6303 if (compress_type) {
6304 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6305 em->compress_type = compress_type;
6306 }
6307 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6308 if (create == 0 && !PageUptodate(page)) {
6309 if (btrfs_file_extent_compression(leaf, item) !=
6310 BTRFS_COMPRESS_NONE) {
6311 ret = uncompress_inline(path, inode, page,
6312 pg_offset,
6313 extent_offset, item);
6314 if (ret) {
6315 err = ret;
6316 goto out;
6317 }
6318 } else {
6319 map = kmap(page);
6320 read_extent_buffer(leaf, map + pg_offset, ptr,
6321 copy_size);
6322 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6323 memset(map + pg_offset + copy_size, 0,
6324 PAGE_CACHE_SIZE - pg_offset -
6325 copy_size);
6326 }
6327 kunmap(page);
6328 }
6329 flush_dcache_page(page);
6330 } else if (create && PageUptodate(page)) {
6331 BUG();
6332 if (!trans) {
6333 kunmap(page);
6334 free_extent_map(em);
6335 em = NULL;
6336
6337 btrfs_release_path(path);
6338 trans = btrfs_join_transaction(root);
6339
6340 if (IS_ERR(trans))
6341 return ERR_CAST(trans);
6342 goto again;
6343 }
6344 map = kmap(page);
6345 write_extent_buffer(leaf, map + pg_offset, ptr,
6346 copy_size);
6347 kunmap(page);
6348 btrfs_mark_buffer_dirty(leaf);
6349 }
6350 set_extent_uptodate(io_tree, em->start,
6351 extent_map_end(em) - 1, NULL, GFP_NOFS);
6352 goto insert;
6353 } else {
6354 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6355 }
6356 not_found:
6357 em->start = start;
6358 em->orig_start = start;
6359 em->len = len;
6360 not_found_em:
6361 em->block_start = EXTENT_MAP_HOLE;
6362 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6363 insert:
6364 btrfs_release_path(path);
6365 if (em->start > start || extent_map_end(em) <= start) {
6366 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6367 em->start, em->len, start, len);
6368 err = -EIO;
6369 goto out;
6370 }
6371
6372 err = 0;
6373 write_lock(&em_tree->lock);
6374 ret = add_extent_mapping(em_tree, em, 0);
6375 /* it is possible that someone inserted the extent into the tree
6376 * while we had the lock dropped. It is also possible that
6377 * an overlapping map exists in the tree
6378 */
6379 if (ret == -EEXIST) {
6380 struct extent_map *existing;
6381
6382 ret = 0;
6383
6384 existing = lookup_extent_mapping(em_tree, start, len);
6385 if (existing && (existing->start > start ||
6386 existing->start + existing->len <= start)) {
6387 free_extent_map(existing);
6388 existing = NULL;
6389 }
6390 if (!existing) {
6391 existing = lookup_extent_mapping(em_tree, em->start,
6392 em->len);
6393 if (existing) {
6394 err = merge_extent_mapping(em_tree, existing,
6395 em, start,
6396 root->sectorsize);
6397 free_extent_map(existing);
6398 if (err) {
6399 free_extent_map(em);
6400 em = NULL;
6401 }
6402 } else {
6403 err = -EIO;
6404 free_extent_map(em);
6405 em = NULL;
6406 }
6407 } else {
6408 free_extent_map(em);
6409 em = existing;
6410 err = 0;
6411 }
6412 }
6413 write_unlock(&em_tree->lock);
6414 out:
6415
6416 trace_btrfs_get_extent(root, em);
6417
6418 if (path)
6419 btrfs_free_path(path);
6420 if (trans) {
6421 ret = btrfs_end_transaction(trans, root);
6422 if (!err)
6423 err = ret;
6424 }
6425 if (err) {
6426 free_extent_map(em);
6427 return ERR_PTR(err);
6428 }
6429 BUG_ON(!em); /* Error is always set */
6430 return em;
6431 }
6432
6433 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6434 size_t pg_offset, u64 start, u64 len,
6435 int create)
6436 {
6437 struct extent_map *em;
6438 struct extent_map *hole_em = NULL;
6439 u64 range_start = start;
6440 u64 end;
6441 u64 found;
6442 u64 found_end;
6443 int err = 0;
6444
6445 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6446 if (IS_ERR(em))
6447 return em;
6448 if (em) {
6449 /*
6450 * if our em maps to
6451 * - a hole or
6452 * - a pre-alloc extent,
6453 * there might actually be delalloc bytes behind it.
6454 */
6455 if (em->block_start != EXTENT_MAP_HOLE &&
6456 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6457 return em;
6458 else
6459 hole_em = em;
6460 }
6461
6462 /* check to see if we've wrapped (len == -1 or similar) */
6463 end = start + len;
6464 if (end < start)
6465 end = (u64)-1;
6466 else
6467 end -= 1;
6468
6469 em = NULL;
6470
6471 /* ok, we didn't find anything, lets look for delalloc */
6472 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6473 end, len, EXTENT_DELALLOC, 1);
6474 found_end = range_start + found;
6475 if (found_end < range_start)
6476 found_end = (u64)-1;
6477
6478 /*
6479 * we didn't find anything useful, return
6480 * the original results from get_extent()
6481 */
6482 if (range_start > end || found_end <= start) {
6483 em = hole_em;
6484 hole_em = NULL;
6485 goto out;
6486 }
6487
6488 /* adjust the range_start to make sure it doesn't
6489 * go backwards from the start they passed in
6490 */
6491 range_start = max(start, range_start);
6492 found = found_end - range_start;
6493
6494 if (found > 0) {
6495 u64 hole_start = start;
6496 u64 hole_len = len;
6497
6498 em = alloc_extent_map();
6499 if (!em) {
6500 err = -ENOMEM;
6501 goto out;
6502 }
6503 /*
6504 * when btrfs_get_extent can't find anything it
6505 * returns one huge hole
6506 *
6507 * make sure what it found really fits our range, and
6508 * adjust to make sure it is based on the start from
6509 * the caller
6510 */
6511 if (hole_em) {
6512 u64 calc_end = extent_map_end(hole_em);
6513
6514 if (calc_end <= start || (hole_em->start > end)) {
6515 free_extent_map(hole_em);
6516 hole_em = NULL;
6517 } else {
6518 hole_start = max(hole_em->start, start);
6519 hole_len = calc_end - hole_start;
6520 }
6521 }
6522 em->bdev = NULL;
6523 if (hole_em && range_start > hole_start) {
6524 /* our hole starts before our delalloc, so we
6525 * have to return just the parts of the hole
6526 * that go until the delalloc starts
6527 */
6528 em->len = min(hole_len,
6529 range_start - hole_start);
6530 em->start = hole_start;
6531 em->orig_start = hole_start;
6532 /*
6533 * don't adjust block start at all,
6534 * it is fixed at EXTENT_MAP_HOLE
6535 */
6536 em->block_start = hole_em->block_start;
6537 em->block_len = hole_len;
6538 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6539 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6540 } else {
6541 em->start = range_start;
6542 em->len = found;
6543 em->orig_start = range_start;
6544 em->block_start = EXTENT_MAP_DELALLOC;
6545 em->block_len = found;
6546 }
6547 } else if (hole_em) {
6548 return hole_em;
6549 }
6550 out:
6551
6552 free_extent_map(hole_em);
6553 if (err) {
6554 free_extent_map(em);
6555 return ERR_PTR(err);
6556 }
6557 return em;
6558 }
6559
6560 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6561 u64 start, u64 len)
6562 {
6563 struct btrfs_root *root = BTRFS_I(inode)->root;
6564 struct extent_map *em;
6565 struct btrfs_key ins;
6566 u64 alloc_hint;
6567 int ret;
6568
6569 alloc_hint = get_extent_allocation_hint(inode, start, len);
6570 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6571 alloc_hint, &ins, 1);
6572 if (ret)
6573 return ERR_PTR(ret);
6574
6575 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6576 ins.offset, ins.offset, ins.offset, 0);
6577 if (IS_ERR(em)) {
6578 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6579 return em;
6580 }
6581
6582 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6583 ins.offset, ins.offset, 0);
6584 if (ret) {
6585 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6586 free_extent_map(em);
6587 return ERR_PTR(ret);
6588 }
6589
6590 return em;
6591 }
6592
6593 /*
6594 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6595 * block must be cow'd
6596 */
6597 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6598 u64 *orig_start, u64 *orig_block_len,
6599 u64 *ram_bytes)
6600 {
6601 struct btrfs_trans_handle *trans;
6602 struct btrfs_path *path;
6603 int ret;
6604 struct extent_buffer *leaf;
6605 struct btrfs_root *root = BTRFS_I(inode)->root;
6606 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6607 struct btrfs_file_extent_item *fi;
6608 struct btrfs_key key;
6609 u64 disk_bytenr;
6610 u64 backref_offset;
6611 u64 extent_end;
6612 u64 num_bytes;
6613 int slot;
6614 int found_type;
6615 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6616
6617 path = btrfs_alloc_path();
6618 if (!path)
6619 return -ENOMEM;
6620
6621 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6622 offset, 0);
6623 if (ret < 0)
6624 goto out;
6625
6626 slot = path->slots[0];
6627 if (ret == 1) {
6628 if (slot == 0) {
6629 /* can't find the item, must cow */
6630 ret = 0;
6631 goto out;
6632 }
6633 slot--;
6634 }
6635 ret = 0;
6636 leaf = path->nodes[0];
6637 btrfs_item_key_to_cpu(leaf, &key, slot);
6638 if (key.objectid != btrfs_ino(inode) ||
6639 key.type != BTRFS_EXTENT_DATA_KEY) {
6640 /* not our file or wrong item type, must cow */
6641 goto out;
6642 }
6643
6644 if (key.offset > offset) {
6645 /* Wrong offset, must cow */
6646 goto out;
6647 }
6648
6649 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6650 found_type = btrfs_file_extent_type(leaf, fi);
6651 if (found_type != BTRFS_FILE_EXTENT_REG &&
6652 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6653 /* not a regular extent, must cow */
6654 goto out;
6655 }
6656
6657 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6658 goto out;
6659
6660 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6661 if (extent_end <= offset)
6662 goto out;
6663
6664 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6665 if (disk_bytenr == 0)
6666 goto out;
6667
6668 if (btrfs_file_extent_compression(leaf, fi) ||
6669 btrfs_file_extent_encryption(leaf, fi) ||
6670 btrfs_file_extent_other_encoding(leaf, fi))
6671 goto out;
6672
6673 backref_offset = btrfs_file_extent_offset(leaf, fi);
6674
6675 if (orig_start) {
6676 *orig_start = key.offset - backref_offset;
6677 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6678 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6679 }
6680
6681 if (btrfs_extent_readonly(root, disk_bytenr))
6682 goto out;
6683
6684 num_bytes = min(offset + *len, extent_end) - offset;
6685 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6686 u64 range_end;
6687
6688 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6689 ret = test_range_bit(io_tree, offset, range_end,
6690 EXTENT_DELALLOC, 0, NULL);
6691 if (ret) {
6692 ret = -EAGAIN;
6693 goto out;
6694 }
6695 }
6696
6697 btrfs_release_path(path);
6698
6699 /*
6700 * look for other files referencing this extent, if we
6701 * find any we must cow
6702 */
6703 trans = btrfs_join_transaction(root);
6704 if (IS_ERR(trans)) {
6705 ret = 0;
6706 goto out;
6707 }
6708
6709 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6710 key.offset - backref_offset, disk_bytenr);
6711 btrfs_end_transaction(trans, root);
6712 if (ret) {
6713 ret = 0;
6714 goto out;
6715 }
6716
6717 /*
6718 * adjust disk_bytenr and num_bytes to cover just the bytes
6719 * in this extent we are about to write. If there
6720 * are any csums in that range we have to cow in order
6721 * to keep the csums correct
6722 */
6723 disk_bytenr += backref_offset;
6724 disk_bytenr += offset - key.offset;
6725 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6726 goto out;
6727 /*
6728 * all of the above have passed, it is safe to overwrite this extent
6729 * without cow
6730 */
6731 *len = num_bytes;
6732 ret = 1;
6733 out:
6734 btrfs_free_path(path);
6735 return ret;
6736 }
6737
6738 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6739 struct extent_state **cached_state, int writing)
6740 {
6741 struct btrfs_ordered_extent *ordered;
6742 int ret = 0;
6743
6744 while (1) {
6745 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6746 0, cached_state);
6747 /*
6748 * We're concerned with the entire range that we're going to be
6749 * doing DIO to, so we need to make sure theres no ordered
6750 * extents in this range.
6751 */
6752 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6753 lockend - lockstart + 1);
6754
6755 /*
6756 * We need to make sure there are no buffered pages in this
6757 * range either, we could have raced between the invalidate in
6758 * generic_file_direct_write and locking the extent. The
6759 * invalidate needs to happen so that reads after a write do not
6760 * get stale data.
6761 */
6762 if (!ordered && (!writing ||
6763 !test_range_bit(&BTRFS_I(inode)->io_tree,
6764 lockstart, lockend, EXTENT_UPTODATE, 0,
6765 *cached_state)))
6766 break;
6767
6768 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6769 cached_state, GFP_NOFS);
6770
6771 if (ordered) {
6772 btrfs_start_ordered_extent(inode, ordered, 1);
6773 btrfs_put_ordered_extent(ordered);
6774 } else {
6775 /* Screw you mmap */
6776 ret = filemap_write_and_wait_range(inode->i_mapping,
6777 lockstart,
6778 lockend);
6779 if (ret)
6780 break;
6781
6782 /*
6783 * If we found a page that couldn't be invalidated just
6784 * fall back to buffered.
6785 */
6786 ret = invalidate_inode_pages2_range(inode->i_mapping,
6787 lockstart >> PAGE_CACHE_SHIFT,
6788 lockend >> PAGE_CACHE_SHIFT);
6789 if (ret)
6790 break;
6791 }
6792
6793 cond_resched();
6794 }
6795
6796 return ret;
6797 }
6798
6799 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6800 u64 len, u64 orig_start,
6801 u64 block_start, u64 block_len,
6802 u64 orig_block_len, u64 ram_bytes,
6803 int type)
6804 {
6805 struct extent_map_tree *em_tree;
6806 struct extent_map *em;
6807 struct btrfs_root *root = BTRFS_I(inode)->root;
6808 int ret;
6809
6810 em_tree = &BTRFS_I(inode)->extent_tree;
6811 em = alloc_extent_map();
6812 if (!em)
6813 return ERR_PTR(-ENOMEM);
6814
6815 em->start = start;
6816 em->orig_start = orig_start;
6817 em->mod_start = start;
6818 em->mod_len = len;
6819 em->len = len;
6820 em->block_len = block_len;
6821 em->block_start = block_start;
6822 em->bdev = root->fs_info->fs_devices->latest_bdev;
6823 em->orig_block_len = orig_block_len;
6824 em->ram_bytes = ram_bytes;
6825 em->generation = -1;
6826 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6827 if (type == BTRFS_ORDERED_PREALLOC)
6828 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6829
6830 do {
6831 btrfs_drop_extent_cache(inode, em->start,
6832 em->start + em->len - 1, 0);
6833 write_lock(&em_tree->lock);
6834 ret = add_extent_mapping(em_tree, em, 1);
6835 write_unlock(&em_tree->lock);
6836 } while (ret == -EEXIST);
6837
6838 if (ret) {
6839 free_extent_map(em);
6840 return ERR_PTR(ret);
6841 }
6842
6843 return em;
6844 }
6845
6846
6847 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6848 struct buffer_head *bh_result, int create)
6849 {
6850 struct extent_map *em;
6851 struct btrfs_root *root = BTRFS_I(inode)->root;
6852 struct extent_state *cached_state = NULL;
6853 u64 start = iblock << inode->i_blkbits;
6854 u64 lockstart, lockend;
6855 u64 len = bh_result->b_size;
6856 int unlock_bits = EXTENT_LOCKED;
6857 int ret = 0;
6858
6859 if (create)
6860 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6861 else
6862 len = min_t(u64, len, root->sectorsize);
6863
6864 lockstart = start;
6865 lockend = start + len - 1;
6866
6867 /*
6868 * If this errors out it's because we couldn't invalidate pagecache for
6869 * this range and we need to fallback to buffered.
6870 */
6871 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6872 return -ENOTBLK;
6873
6874 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6875 if (IS_ERR(em)) {
6876 ret = PTR_ERR(em);
6877 goto unlock_err;
6878 }
6879
6880 /*
6881 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6882 * io. INLINE is special, and we could probably kludge it in here, but
6883 * it's still buffered so for safety lets just fall back to the generic
6884 * buffered path.
6885 *
6886 * For COMPRESSED we _have_ to read the entire extent in so we can
6887 * decompress it, so there will be buffering required no matter what we
6888 * do, so go ahead and fallback to buffered.
6889 *
6890 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6891 * to buffered IO. Don't blame me, this is the price we pay for using
6892 * the generic code.
6893 */
6894 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6895 em->block_start == EXTENT_MAP_INLINE) {
6896 free_extent_map(em);
6897 ret = -ENOTBLK;
6898 goto unlock_err;
6899 }
6900
6901 /* Just a good old fashioned hole, return */
6902 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6903 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6904 free_extent_map(em);
6905 goto unlock_err;
6906 }
6907
6908 /*
6909 * We don't allocate a new extent in the following cases
6910 *
6911 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6912 * existing extent.
6913 * 2) The extent is marked as PREALLOC. We're good to go here and can
6914 * just use the extent.
6915 *
6916 */
6917 if (!create) {
6918 len = min(len, em->len - (start - em->start));
6919 lockstart = start + len;
6920 goto unlock;
6921 }
6922
6923 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6924 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6925 em->block_start != EXTENT_MAP_HOLE)) {
6926 int type;
6927 int ret;
6928 u64 block_start, orig_start, orig_block_len, ram_bytes;
6929
6930 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6931 type = BTRFS_ORDERED_PREALLOC;
6932 else
6933 type = BTRFS_ORDERED_NOCOW;
6934 len = min(len, em->len - (start - em->start));
6935 block_start = em->block_start + (start - em->start);
6936
6937 if (can_nocow_extent(inode, start, &len, &orig_start,
6938 &orig_block_len, &ram_bytes) == 1) {
6939 if (type == BTRFS_ORDERED_PREALLOC) {
6940 free_extent_map(em);
6941 em = create_pinned_em(inode, start, len,
6942 orig_start,
6943 block_start, len,
6944 orig_block_len,
6945 ram_bytes, type);
6946 if (IS_ERR(em))
6947 goto unlock_err;
6948 }
6949
6950 ret = btrfs_add_ordered_extent_dio(inode, start,
6951 block_start, len, len, type);
6952 if (ret) {
6953 free_extent_map(em);
6954 goto unlock_err;
6955 }
6956 goto unlock;
6957 }
6958 }
6959
6960 /*
6961 * this will cow the extent, reset the len in case we changed
6962 * it above
6963 */
6964 len = bh_result->b_size;
6965 free_extent_map(em);
6966 em = btrfs_new_extent_direct(inode, start, len);
6967 if (IS_ERR(em)) {
6968 ret = PTR_ERR(em);
6969 goto unlock_err;
6970 }
6971 len = min(len, em->len - (start - em->start));
6972 unlock:
6973 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6974 inode->i_blkbits;
6975 bh_result->b_size = len;
6976 bh_result->b_bdev = em->bdev;
6977 set_buffer_mapped(bh_result);
6978 if (create) {
6979 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6980 set_buffer_new(bh_result);
6981
6982 /*
6983 * Need to update the i_size under the extent lock so buffered
6984 * readers will get the updated i_size when we unlock.
6985 */
6986 if (start + len > i_size_read(inode))
6987 i_size_write(inode, start + len);
6988
6989 spin_lock(&BTRFS_I(inode)->lock);
6990 BTRFS_I(inode)->outstanding_extents++;
6991 spin_unlock(&BTRFS_I(inode)->lock);
6992
6993 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6994 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6995 &cached_state, GFP_NOFS);
6996 BUG_ON(ret);
6997 }
6998
6999 /*
7000 * In the case of write we need to clear and unlock the entire range,
7001 * in the case of read we need to unlock only the end area that we
7002 * aren't using if there is any left over space.
7003 */
7004 if (lockstart < lockend) {
7005 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7006 lockend, unlock_bits, 1, 0,
7007 &cached_state, GFP_NOFS);
7008 } else {
7009 free_extent_state(cached_state);
7010 }
7011
7012 free_extent_map(em);
7013
7014 return 0;
7015
7016 unlock_err:
7017 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7018 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7019 return ret;
7020 }
7021
7022 static void btrfs_endio_direct_read(struct bio *bio, int err)
7023 {
7024 struct btrfs_dio_private *dip = bio->bi_private;
7025 struct bio_vec *bvec;
7026 struct inode *inode = dip->inode;
7027 struct btrfs_root *root = BTRFS_I(inode)->root;
7028 struct bio *dio_bio;
7029 u32 *csums = (u32 *)dip->csum;
7030 u64 start;
7031 int i;
7032
7033 start = dip->logical_offset;
7034 bio_for_each_segment_all(bvec, bio, i) {
7035 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7036 struct page *page = bvec->bv_page;
7037 char *kaddr;
7038 u32 csum = ~(u32)0;
7039 unsigned long flags;
7040
7041 local_irq_save(flags);
7042 kaddr = kmap_atomic(page);
7043 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7044 csum, bvec->bv_len);
7045 btrfs_csum_final(csum, (char *)&csum);
7046 kunmap_atomic(kaddr);
7047 local_irq_restore(flags);
7048
7049 flush_dcache_page(bvec->bv_page);
7050 if (csum != csums[i]) {
7051 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7052 btrfs_ino(inode), start, csum,
7053 csums[i]);
7054 err = -EIO;
7055 }
7056 }
7057
7058 start += bvec->bv_len;
7059 }
7060
7061 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7062 dip->logical_offset + dip->bytes - 1);
7063 dio_bio = dip->dio_bio;
7064
7065 kfree(dip);
7066
7067 /* If we had a csum failure make sure to clear the uptodate flag */
7068 if (err)
7069 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7070 dio_end_io(dio_bio, err);
7071 bio_put(bio);
7072 }
7073
7074 static void btrfs_endio_direct_write(struct bio *bio, int err)
7075 {
7076 struct btrfs_dio_private *dip = bio->bi_private;
7077 struct inode *inode = dip->inode;
7078 struct btrfs_root *root = BTRFS_I(inode)->root;
7079 struct btrfs_ordered_extent *ordered = NULL;
7080 u64 ordered_offset = dip->logical_offset;
7081 u64 ordered_bytes = dip->bytes;
7082 struct bio *dio_bio;
7083 int ret;
7084
7085 if (err)
7086 goto out_done;
7087 again:
7088 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7089 &ordered_offset,
7090 ordered_bytes, !err);
7091 if (!ret)
7092 goto out_test;
7093
7094 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7095 btrfs_queue_work(root->fs_info->endio_write_workers,
7096 &ordered->work);
7097 out_test:
7098 /*
7099 * our bio might span multiple ordered extents. If we haven't
7100 * completed the accounting for the whole dio, go back and try again
7101 */
7102 if (ordered_offset < dip->logical_offset + dip->bytes) {
7103 ordered_bytes = dip->logical_offset + dip->bytes -
7104 ordered_offset;
7105 ordered = NULL;
7106 goto again;
7107 }
7108 out_done:
7109 dio_bio = dip->dio_bio;
7110
7111 kfree(dip);
7112
7113 /* If we had an error make sure to clear the uptodate flag */
7114 if (err)
7115 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7116 dio_end_io(dio_bio, err);
7117 bio_put(bio);
7118 }
7119
7120 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7121 struct bio *bio, int mirror_num,
7122 unsigned long bio_flags, u64 offset)
7123 {
7124 int ret;
7125 struct btrfs_root *root = BTRFS_I(inode)->root;
7126 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7127 BUG_ON(ret); /* -ENOMEM */
7128 return 0;
7129 }
7130
7131 static void btrfs_end_dio_bio(struct bio *bio, int err)
7132 {
7133 struct btrfs_dio_private *dip = bio->bi_private;
7134
7135 if (err) {
7136 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7137 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7138 btrfs_ino(dip->inode), bio->bi_rw,
7139 (unsigned long long)bio->bi_iter.bi_sector,
7140 bio->bi_iter.bi_size, err);
7141 dip->errors = 1;
7142
7143 /*
7144 * before atomic variable goto zero, we must make sure
7145 * dip->errors is perceived to be set.
7146 */
7147 smp_mb__before_atomic_dec();
7148 }
7149
7150 /* if there are more bios still pending for this dio, just exit */
7151 if (!atomic_dec_and_test(&dip->pending_bios))
7152 goto out;
7153
7154 if (dip->errors) {
7155 bio_io_error(dip->orig_bio);
7156 } else {
7157 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7158 bio_endio(dip->orig_bio, 0);
7159 }
7160 out:
7161 bio_put(bio);
7162 }
7163
7164 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7165 u64 first_sector, gfp_t gfp_flags)
7166 {
7167 int nr_vecs = bio_get_nr_vecs(bdev);
7168 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7169 }
7170
7171 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7172 int rw, u64 file_offset, int skip_sum,
7173 int async_submit)
7174 {
7175 struct btrfs_dio_private *dip = bio->bi_private;
7176 int write = rw & REQ_WRITE;
7177 struct btrfs_root *root = BTRFS_I(inode)->root;
7178 int ret;
7179
7180 if (async_submit)
7181 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7182
7183 bio_get(bio);
7184
7185 if (!write) {
7186 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7187 if (ret)
7188 goto err;
7189 }
7190
7191 if (skip_sum)
7192 goto map;
7193
7194 if (write && async_submit) {
7195 ret = btrfs_wq_submit_bio(root->fs_info,
7196 inode, rw, bio, 0, 0,
7197 file_offset,
7198 __btrfs_submit_bio_start_direct_io,
7199 __btrfs_submit_bio_done);
7200 goto err;
7201 } else if (write) {
7202 /*
7203 * If we aren't doing async submit, calculate the csum of the
7204 * bio now.
7205 */
7206 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7207 if (ret)
7208 goto err;
7209 } else if (!skip_sum) {
7210 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7211 file_offset);
7212 if (ret)
7213 goto err;
7214 }
7215
7216 map:
7217 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7218 err:
7219 bio_put(bio);
7220 return ret;
7221 }
7222
7223 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7224 int skip_sum)
7225 {
7226 struct inode *inode = dip->inode;
7227 struct btrfs_root *root = BTRFS_I(inode)->root;
7228 struct bio *bio;
7229 struct bio *orig_bio = dip->orig_bio;
7230 struct bio_vec *bvec = orig_bio->bi_io_vec;
7231 u64 start_sector = orig_bio->bi_iter.bi_sector;
7232 u64 file_offset = dip->logical_offset;
7233 u64 submit_len = 0;
7234 u64 map_length;
7235 int nr_pages = 0;
7236 int ret = 0;
7237 int async_submit = 0;
7238
7239 map_length = orig_bio->bi_iter.bi_size;
7240 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7241 &map_length, NULL, 0);
7242 if (ret) {
7243 bio_put(orig_bio);
7244 return -EIO;
7245 }
7246
7247 if (map_length >= orig_bio->bi_iter.bi_size) {
7248 bio = orig_bio;
7249 goto submit;
7250 }
7251
7252 /* async crcs make it difficult to collect full stripe writes. */
7253 if (btrfs_get_alloc_profile(root, 1) &
7254 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7255 async_submit = 0;
7256 else
7257 async_submit = 1;
7258
7259 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7260 if (!bio)
7261 return -ENOMEM;
7262 bio->bi_private = dip;
7263 bio->bi_end_io = btrfs_end_dio_bio;
7264 atomic_inc(&dip->pending_bios);
7265
7266 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7267 if (unlikely(map_length < submit_len + bvec->bv_len ||
7268 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7269 bvec->bv_offset) < bvec->bv_len)) {
7270 /*
7271 * inc the count before we submit the bio so
7272 * we know the end IO handler won't happen before
7273 * we inc the count. Otherwise, the dip might get freed
7274 * before we're done setting it up
7275 */
7276 atomic_inc(&dip->pending_bios);
7277 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7278 file_offset, skip_sum,
7279 async_submit);
7280 if (ret) {
7281 bio_put(bio);
7282 atomic_dec(&dip->pending_bios);
7283 goto out_err;
7284 }
7285
7286 start_sector += submit_len >> 9;
7287 file_offset += submit_len;
7288
7289 submit_len = 0;
7290 nr_pages = 0;
7291
7292 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7293 start_sector, GFP_NOFS);
7294 if (!bio)
7295 goto out_err;
7296 bio->bi_private = dip;
7297 bio->bi_end_io = btrfs_end_dio_bio;
7298
7299 map_length = orig_bio->bi_iter.bi_size;
7300 ret = btrfs_map_block(root->fs_info, rw,
7301 start_sector << 9,
7302 &map_length, NULL, 0);
7303 if (ret) {
7304 bio_put(bio);
7305 goto out_err;
7306 }
7307 } else {
7308 submit_len += bvec->bv_len;
7309 nr_pages++;
7310 bvec++;
7311 }
7312 }
7313
7314 submit:
7315 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7316 async_submit);
7317 if (!ret)
7318 return 0;
7319
7320 bio_put(bio);
7321 out_err:
7322 dip->errors = 1;
7323 /*
7324 * before atomic variable goto zero, we must
7325 * make sure dip->errors is perceived to be set.
7326 */
7327 smp_mb__before_atomic_dec();
7328 if (atomic_dec_and_test(&dip->pending_bios))
7329 bio_io_error(dip->orig_bio);
7330
7331 /* bio_end_io() will handle error, so we needn't return it */
7332 return 0;
7333 }
7334
7335 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7336 struct inode *inode, loff_t file_offset)
7337 {
7338 struct btrfs_root *root = BTRFS_I(inode)->root;
7339 struct btrfs_dio_private *dip;
7340 struct bio *io_bio;
7341 int skip_sum;
7342 int sum_len;
7343 int write = rw & REQ_WRITE;
7344 int ret = 0;
7345 u16 csum_size;
7346
7347 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7348
7349 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7350 if (!io_bio) {
7351 ret = -ENOMEM;
7352 goto free_ordered;
7353 }
7354
7355 if (!skip_sum && !write) {
7356 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7357 sum_len = dio_bio->bi_iter.bi_size >>
7358 inode->i_sb->s_blocksize_bits;
7359 sum_len *= csum_size;
7360 } else {
7361 sum_len = 0;
7362 }
7363
7364 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7365 if (!dip) {
7366 ret = -ENOMEM;
7367 goto free_io_bio;
7368 }
7369
7370 dip->private = dio_bio->bi_private;
7371 dip->inode = inode;
7372 dip->logical_offset = file_offset;
7373 dip->bytes = dio_bio->bi_iter.bi_size;
7374 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7375 io_bio->bi_private = dip;
7376 dip->errors = 0;
7377 dip->orig_bio = io_bio;
7378 dip->dio_bio = dio_bio;
7379 atomic_set(&dip->pending_bios, 0);
7380
7381 if (write)
7382 io_bio->bi_end_io = btrfs_endio_direct_write;
7383 else
7384 io_bio->bi_end_io = btrfs_endio_direct_read;
7385
7386 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7387 if (!ret)
7388 return;
7389
7390 free_io_bio:
7391 bio_put(io_bio);
7392
7393 free_ordered:
7394 /*
7395 * If this is a write, we need to clean up the reserved space and kill
7396 * the ordered extent.
7397 */
7398 if (write) {
7399 struct btrfs_ordered_extent *ordered;
7400 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7401 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7402 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7403 btrfs_free_reserved_extent(root, ordered->start,
7404 ordered->disk_len);
7405 btrfs_put_ordered_extent(ordered);
7406 btrfs_put_ordered_extent(ordered);
7407 }
7408 bio_endio(dio_bio, ret);
7409 }
7410
7411 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7412 const struct iovec *iov, loff_t offset,
7413 unsigned long nr_segs)
7414 {
7415 int seg;
7416 int i;
7417 size_t size;
7418 unsigned long addr;
7419 unsigned blocksize_mask = root->sectorsize - 1;
7420 ssize_t retval = -EINVAL;
7421 loff_t end = offset;
7422
7423 if (offset & blocksize_mask)
7424 goto out;
7425
7426 /* Check the memory alignment. Blocks cannot straddle pages */
7427 for (seg = 0; seg < nr_segs; seg++) {
7428 addr = (unsigned long)iov[seg].iov_base;
7429 size = iov[seg].iov_len;
7430 end += size;
7431 if ((addr & blocksize_mask) || (size & blocksize_mask))
7432 goto out;
7433
7434 /* If this is a write we don't need to check anymore */
7435 if (rw & WRITE)
7436 continue;
7437
7438 /*
7439 * Check to make sure we don't have duplicate iov_base's in this
7440 * iovec, if so return EINVAL, otherwise we'll get csum errors
7441 * when reading back.
7442 */
7443 for (i = seg + 1; i < nr_segs; i++) {
7444 if (iov[seg].iov_base == iov[i].iov_base)
7445 goto out;
7446 }
7447 }
7448 retval = 0;
7449 out:
7450 return retval;
7451 }
7452
7453 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7454 const struct iovec *iov, loff_t offset,
7455 unsigned long nr_segs)
7456 {
7457 struct file *file = iocb->ki_filp;
7458 struct inode *inode = file->f_mapping->host;
7459 size_t count = 0;
7460 int flags = 0;
7461 bool wakeup = true;
7462 bool relock = false;
7463 ssize_t ret;
7464
7465 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7466 offset, nr_segs))
7467 return 0;
7468
7469 atomic_inc(&inode->i_dio_count);
7470 smp_mb__after_atomic_inc();
7471
7472 /*
7473 * The generic stuff only does filemap_write_and_wait_range, which
7474 * isn't enough if we've written compressed pages to this area, so
7475 * we need to flush the dirty pages again to make absolutely sure
7476 * that any outstanding dirty pages are on disk.
7477 */
7478 count = iov_length(iov, nr_segs);
7479 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7480 &BTRFS_I(inode)->runtime_flags))
7481 filemap_fdatawrite_range(inode->i_mapping, offset, count);
7482
7483 if (rw & WRITE) {
7484 /*
7485 * If the write DIO is beyond the EOF, we need update
7486 * the isize, but it is protected by i_mutex. So we can
7487 * not unlock the i_mutex at this case.
7488 */
7489 if (offset + count <= inode->i_size) {
7490 mutex_unlock(&inode->i_mutex);
7491 relock = true;
7492 }
7493 ret = btrfs_delalloc_reserve_space(inode, count);
7494 if (ret)
7495 goto out;
7496 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7497 &BTRFS_I(inode)->runtime_flags))) {
7498 inode_dio_done(inode);
7499 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7500 wakeup = false;
7501 }
7502
7503 ret = __blockdev_direct_IO(rw, iocb, inode,
7504 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7505 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7506 btrfs_submit_direct, flags);
7507 if (rw & WRITE) {
7508 if (ret < 0 && ret != -EIOCBQUEUED)
7509 btrfs_delalloc_release_space(inode, count);
7510 else if (ret >= 0 && (size_t)ret < count)
7511 btrfs_delalloc_release_space(inode,
7512 count - (size_t)ret);
7513 else
7514 btrfs_delalloc_release_metadata(inode, 0);
7515 }
7516 out:
7517 if (wakeup)
7518 inode_dio_done(inode);
7519 if (relock)
7520 mutex_lock(&inode->i_mutex);
7521
7522 return ret;
7523 }
7524
7525 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7526
7527 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7528 __u64 start, __u64 len)
7529 {
7530 int ret;
7531
7532 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7533 if (ret)
7534 return ret;
7535
7536 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7537 }
7538
7539 int btrfs_readpage(struct file *file, struct page *page)
7540 {
7541 struct extent_io_tree *tree;
7542 tree = &BTRFS_I(page->mapping->host)->io_tree;
7543 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7544 }
7545
7546 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7547 {
7548 struct extent_io_tree *tree;
7549
7550
7551 if (current->flags & PF_MEMALLOC) {
7552 redirty_page_for_writepage(wbc, page);
7553 unlock_page(page);
7554 return 0;
7555 }
7556 tree = &BTRFS_I(page->mapping->host)->io_tree;
7557 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7558 }
7559
7560 static int btrfs_writepages(struct address_space *mapping,
7561 struct writeback_control *wbc)
7562 {
7563 struct extent_io_tree *tree;
7564
7565 tree = &BTRFS_I(mapping->host)->io_tree;
7566 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7567 }
7568
7569 static int
7570 btrfs_readpages(struct file *file, struct address_space *mapping,
7571 struct list_head *pages, unsigned nr_pages)
7572 {
7573 struct extent_io_tree *tree;
7574 tree = &BTRFS_I(mapping->host)->io_tree;
7575 return extent_readpages(tree, mapping, pages, nr_pages,
7576 btrfs_get_extent);
7577 }
7578 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7579 {
7580 struct extent_io_tree *tree;
7581 struct extent_map_tree *map;
7582 int ret;
7583
7584 tree = &BTRFS_I(page->mapping->host)->io_tree;
7585 map = &BTRFS_I(page->mapping->host)->extent_tree;
7586 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7587 if (ret == 1) {
7588 ClearPagePrivate(page);
7589 set_page_private(page, 0);
7590 page_cache_release(page);
7591 }
7592 return ret;
7593 }
7594
7595 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7596 {
7597 if (PageWriteback(page) || PageDirty(page))
7598 return 0;
7599 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7600 }
7601
7602 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7603 unsigned int length)
7604 {
7605 struct inode *inode = page->mapping->host;
7606 struct extent_io_tree *tree;
7607 struct btrfs_ordered_extent *ordered;
7608 struct extent_state *cached_state = NULL;
7609 u64 page_start = page_offset(page);
7610 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7611 int inode_evicting = inode->i_state & I_FREEING;
7612
7613 /*
7614 * we have the page locked, so new writeback can't start,
7615 * and the dirty bit won't be cleared while we are here.
7616 *
7617 * Wait for IO on this page so that we can safely clear
7618 * the PagePrivate2 bit and do ordered accounting
7619 */
7620 wait_on_page_writeback(page);
7621
7622 tree = &BTRFS_I(inode)->io_tree;
7623 if (offset) {
7624 btrfs_releasepage(page, GFP_NOFS);
7625 return;
7626 }
7627
7628 if (!inode_evicting)
7629 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7630 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7631 if (ordered) {
7632 /*
7633 * IO on this page will never be started, so we need
7634 * to account for any ordered extents now
7635 */
7636 if (!inode_evicting)
7637 clear_extent_bit(tree, page_start, page_end,
7638 EXTENT_DIRTY | EXTENT_DELALLOC |
7639 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7640 EXTENT_DEFRAG, 1, 0, &cached_state,
7641 GFP_NOFS);
7642 /*
7643 * whoever cleared the private bit is responsible
7644 * for the finish_ordered_io
7645 */
7646 if (TestClearPagePrivate2(page)) {
7647 struct btrfs_ordered_inode_tree *tree;
7648 u64 new_len;
7649
7650 tree = &BTRFS_I(inode)->ordered_tree;
7651
7652 spin_lock_irq(&tree->lock);
7653 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7654 new_len = page_start - ordered->file_offset;
7655 if (new_len < ordered->truncated_len)
7656 ordered->truncated_len = new_len;
7657 spin_unlock_irq(&tree->lock);
7658
7659 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7660 page_start,
7661 PAGE_CACHE_SIZE, 1))
7662 btrfs_finish_ordered_io(ordered);
7663 }
7664 btrfs_put_ordered_extent(ordered);
7665 if (!inode_evicting) {
7666 cached_state = NULL;
7667 lock_extent_bits(tree, page_start, page_end, 0,
7668 &cached_state);
7669 }
7670 }
7671
7672 if (!inode_evicting) {
7673 clear_extent_bit(tree, page_start, page_end,
7674 EXTENT_LOCKED | EXTENT_DIRTY |
7675 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7676 EXTENT_DEFRAG, 1, 1,
7677 &cached_state, GFP_NOFS);
7678
7679 __btrfs_releasepage(page, GFP_NOFS);
7680 }
7681
7682 ClearPageChecked(page);
7683 if (PagePrivate(page)) {
7684 ClearPagePrivate(page);
7685 set_page_private(page, 0);
7686 page_cache_release(page);
7687 }
7688 }
7689
7690 /*
7691 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7692 * called from a page fault handler when a page is first dirtied. Hence we must
7693 * be careful to check for EOF conditions here. We set the page up correctly
7694 * for a written page which means we get ENOSPC checking when writing into
7695 * holes and correct delalloc and unwritten extent mapping on filesystems that
7696 * support these features.
7697 *
7698 * We are not allowed to take the i_mutex here so we have to play games to
7699 * protect against truncate races as the page could now be beyond EOF. Because
7700 * vmtruncate() writes the inode size before removing pages, once we have the
7701 * page lock we can determine safely if the page is beyond EOF. If it is not
7702 * beyond EOF, then the page is guaranteed safe against truncation until we
7703 * unlock the page.
7704 */
7705 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7706 {
7707 struct page *page = vmf->page;
7708 struct inode *inode = file_inode(vma->vm_file);
7709 struct btrfs_root *root = BTRFS_I(inode)->root;
7710 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7711 struct btrfs_ordered_extent *ordered;
7712 struct extent_state *cached_state = NULL;
7713 char *kaddr;
7714 unsigned long zero_start;
7715 loff_t size;
7716 int ret;
7717 int reserved = 0;
7718 u64 page_start;
7719 u64 page_end;
7720
7721 sb_start_pagefault(inode->i_sb);
7722 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7723 if (!ret) {
7724 ret = file_update_time(vma->vm_file);
7725 reserved = 1;
7726 }
7727 if (ret) {
7728 if (ret == -ENOMEM)
7729 ret = VM_FAULT_OOM;
7730 else /* -ENOSPC, -EIO, etc */
7731 ret = VM_FAULT_SIGBUS;
7732 if (reserved)
7733 goto out;
7734 goto out_noreserve;
7735 }
7736
7737 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7738 again:
7739 lock_page(page);
7740 size = i_size_read(inode);
7741 page_start = page_offset(page);
7742 page_end = page_start + PAGE_CACHE_SIZE - 1;
7743
7744 if ((page->mapping != inode->i_mapping) ||
7745 (page_start >= size)) {
7746 /* page got truncated out from underneath us */
7747 goto out_unlock;
7748 }
7749 wait_on_page_writeback(page);
7750
7751 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7752 set_page_extent_mapped(page);
7753
7754 /*
7755 * we can't set the delalloc bits if there are pending ordered
7756 * extents. Drop our locks and wait for them to finish
7757 */
7758 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7759 if (ordered) {
7760 unlock_extent_cached(io_tree, page_start, page_end,
7761 &cached_state, GFP_NOFS);
7762 unlock_page(page);
7763 btrfs_start_ordered_extent(inode, ordered, 1);
7764 btrfs_put_ordered_extent(ordered);
7765 goto again;
7766 }
7767
7768 /*
7769 * XXX - page_mkwrite gets called every time the page is dirtied, even
7770 * if it was already dirty, so for space accounting reasons we need to
7771 * clear any delalloc bits for the range we are fixing to save. There
7772 * is probably a better way to do this, but for now keep consistent with
7773 * prepare_pages in the normal write path.
7774 */
7775 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7776 EXTENT_DIRTY | EXTENT_DELALLOC |
7777 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7778 0, 0, &cached_state, GFP_NOFS);
7779
7780 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7781 &cached_state);
7782 if (ret) {
7783 unlock_extent_cached(io_tree, page_start, page_end,
7784 &cached_state, GFP_NOFS);
7785 ret = VM_FAULT_SIGBUS;
7786 goto out_unlock;
7787 }
7788 ret = 0;
7789
7790 /* page is wholly or partially inside EOF */
7791 if (page_start + PAGE_CACHE_SIZE > size)
7792 zero_start = size & ~PAGE_CACHE_MASK;
7793 else
7794 zero_start = PAGE_CACHE_SIZE;
7795
7796 if (zero_start != PAGE_CACHE_SIZE) {
7797 kaddr = kmap(page);
7798 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7799 flush_dcache_page(page);
7800 kunmap(page);
7801 }
7802 ClearPageChecked(page);
7803 set_page_dirty(page);
7804 SetPageUptodate(page);
7805
7806 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7807 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7808 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7809
7810 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7811
7812 out_unlock:
7813 if (!ret) {
7814 sb_end_pagefault(inode->i_sb);
7815 return VM_FAULT_LOCKED;
7816 }
7817 unlock_page(page);
7818 out:
7819 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7820 out_noreserve:
7821 sb_end_pagefault(inode->i_sb);
7822 return ret;
7823 }
7824
7825 static int btrfs_truncate(struct inode *inode)
7826 {
7827 struct btrfs_root *root = BTRFS_I(inode)->root;
7828 struct btrfs_block_rsv *rsv;
7829 int ret = 0;
7830 int err = 0;
7831 struct btrfs_trans_handle *trans;
7832 u64 mask = root->sectorsize - 1;
7833 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7834
7835 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7836 (u64)-1);
7837 if (ret)
7838 return ret;
7839
7840 /*
7841 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7842 * 3 things going on here
7843 *
7844 * 1) We need to reserve space for our orphan item and the space to
7845 * delete our orphan item. Lord knows we don't want to have a dangling
7846 * orphan item because we didn't reserve space to remove it.
7847 *
7848 * 2) We need to reserve space to update our inode.
7849 *
7850 * 3) We need to have something to cache all the space that is going to
7851 * be free'd up by the truncate operation, but also have some slack
7852 * space reserved in case it uses space during the truncate (thank you
7853 * very much snapshotting).
7854 *
7855 * And we need these to all be seperate. The fact is we can use alot of
7856 * space doing the truncate, and we have no earthly idea how much space
7857 * we will use, so we need the truncate reservation to be seperate so it
7858 * doesn't end up using space reserved for updating the inode or
7859 * removing the orphan item. We also need to be able to stop the
7860 * transaction and start a new one, which means we need to be able to
7861 * update the inode several times, and we have no idea of knowing how
7862 * many times that will be, so we can't just reserve 1 item for the
7863 * entirety of the opration, so that has to be done seperately as well.
7864 * Then there is the orphan item, which does indeed need to be held on
7865 * to for the whole operation, and we need nobody to touch this reserved
7866 * space except the orphan code.
7867 *
7868 * So that leaves us with
7869 *
7870 * 1) root->orphan_block_rsv - for the orphan deletion.
7871 * 2) rsv - for the truncate reservation, which we will steal from the
7872 * transaction reservation.
7873 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7874 * updating the inode.
7875 */
7876 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7877 if (!rsv)
7878 return -ENOMEM;
7879 rsv->size = min_size;
7880 rsv->failfast = 1;
7881
7882 /*
7883 * 1 for the truncate slack space
7884 * 1 for updating the inode.
7885 */
7886 trans = btrfs_start_transaction(root, 2);
7887 if (IS_ERR(trans)) {
7888 err = PTR_ERR(trans);
7889 goto out;
7890 }
7891
7892 /* Migrate the slack space for the truncate to our reserve */
7893 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7894 min_size);
7895 BUG_ON(ret);
7896
7897 /*
7898 * setattr is responsible for setting the ordered_data_close flag,
7899 * but that is only tested during the last file release. That
7900 * could happen well after the next commit, leaving a great big
7901 * window where new writes may get lost if someone chooses to write
7902 * to this file after truncating to zero
7903 *
7904 * The inode doesn't have any dirty data here, and so if we commit
7905 * this is a noop. If someone immediately starts writing to the inode
7906 * it is very likely we'll catch some of their writes in this
7907 * transaction, and the commit will find this file on the ordered
7908 * data list with good things to send down.
7909 *
7910 * This is a best effort solution, there is still a window where
7911 * using truncate to replace the contents of the file will
7912 * end up with a zero length file after a crash.
7913 */
7914 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7915 &BTRFS_I(inode)->runtime_flags))
7916 btrfs_add_ordered_operation(trans, root, inode);
7917
7918 /*
7919 * So if we truncate and then write and fsync we normally would just
7920 * write the extents that changed, which is a problem if we need to
7921 * first truncate that entire inode. So set this flag so we write out
7922 * all of the extents in the inode to the sync log so we're completely
7923 * safe.
7924 */
7925 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7926 trans->block_rsv = rsv;
7927
7928 while (1) {
7929 ret = btrfs_truncate_inode_items(trans, root, inode,
7930 inode->i_size,
7931 BTRFS_EXTENT_DATA_KEY);
7932 if (ret != -ENOSPC) {
7933 err = ret;
7934 break;
7935 }
7936
7937 trans->block_rsv = &root->fs_info->trans_block_rsv;
7938 ret = btrfs_update_inode(trans, root, inode);
7939 if (ret) {
7940 err = ret;
7941 break;
7942 }
7943
7944 btrfs_end_transaction(trans, root);
7945 btrfs_btree_balance_dirty(root);
7946
7947 trans = btrfs_start_transaction(root, 2);
7948 if (IS_ERR(trans)) {
7949 ret = err = PTR_ERR(trans);
7950 trans = NULL;
7951 break;
7952 }
7953
7954 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7955 rsv, min_size);
7956 BUG_ON(ret); /* shouldn't happen */
7957 trans->block_rsv = rsv;
7958 }
7959
7960 if (ret == 0 && inode->i_nlink > 0) {
7961 trans->block_rsv = root->orphan_block_rsv;
7962 ret = btrfs_orphan_del(trans, inode);
7963 if (ret)
7964 err = ret;
7965 }
7966
7967 if (trans) {
7968 trans->block_rsv = &root->fs_info->trans_block_rsv;
7969 ret = btrfs_update_inode(trans, root, inode);
7970 if (ret && !err)
7971 err = ret;
7972
7973 ret = btrfs_end_transaction(trans, root);
7974 btrfs_btree_balance_dirty(root);
7975 }
7976
7977 out:
7978 btrfs_free_block_rsv(root, rsv);
7979
7980 if (ret && !err)
7981 err = ret;
7982
7983 return err;
7984 }
7985
7986 /*
7987 * create a new subvolume directory/inode (helper for the ioctl).
7988 */
7989 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7990 struct btrfs_root *new_root,
7991 struct btrfs_root *parent_root,
7992 u64 new_dirid)
7993 {
7994 struct inode *inode;
7995 int err;
7996 u64 index = 0;
7997
7998 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7999 new_dirid, new_dirid,
8000 S_IFDIR | (~current_umask() & S_IRWXUGO),
8001 &index);
8002 if (IS_ERR(inode))
8003 return PTR_ERR(inode);
8004 inode->i_op = &btrfs_dir_inode_operations;
8005 inode->i_fop = &btrfs_dir_file_operations;
8006
8007 set_nlink(inode, 1);
8008 btrfs_i_size_write(inode, 0);
8009
8010 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8011 if (err)
8012 btrfs_err(new_root->fs_info,
8013 "error inheriting subvolume %llu properties: %d",
8014 new_root->root_key.objectid, err);
8015
8016 err = btrfs_update_inode(trans, new_root, inode);
8017
8018 iput(inode);
8019 return err;
8020 }
8021
8022 struct inode *btrfs_alloc_inode(struct super_block *sb)
8023 {
8024 struct btrfs_inode *ei;
8025 struct inode *inode;
8026
8027 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8028 if (!ei)
8029 return NULL;
8030
8031 ei->root = NULL;
8032 ei->generation = 0;
8033 ei->last_trans = 0;
8034 ei->last_sub_trans = 0;
8035 ei->logged_trans = 0;
8036 ei->delalloc_bytes = 0;
8037 ei->disk_i_size = 0;
8038 ei->flags = 0;
8039 ei->csum_bytes = 0;
8040 ei->index_cnt = (u64)-1;
8041 ei->dir_index = 0;
8042 ei->last_unlink_trans = 0;
8043 ei->last_log_commit = 0;
8044
8045 spin_lock_init(&ei->lock);
8046 ei->outstanding_extents = 0;
8047 ei->reserved_extents = 0;
8048
8049 ei->runtime_flags = 0;
8050 ei->force_compress = BTRFS_COMPRESS_NONE;
8051
8052 ei->delayed_node = NULL;
8053
8054 inode = &ei->vfs_inode;
8055 extent_map_tree_init(&ei->extent_tree);
8056 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8057 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8058 ei->io_tree.track_uptodate = 1;
8059 ei->io_failure_tree.track_uptodate = 1;
8060 atomic_set(&ei->sync_writers, 0);
8061 mutex_init(&ei->log_mutex);
8062 mutex_init(&ei->delalloc_mutex);
8063 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8064 INIT_LIST_HEAD(&ei->delalloc_inodes);
8065 INIT_LIST_HEAD(&ei->ordered_operations);
8066 RB_CLEAR_NODE(&ei->rb_node);
8067
8068 return inode;
8069 }
8070
8071 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8072 void btrfs_test_destroy_inode(struct inode *inode)
8073 {
8074 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8075 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8076 }
8077 #endif
8078
8079 static void btrfs_i_callback(struct rcu_head *head)
8080 {
8081 struct inode *inode = container_of(head, struct inode, i_rcu);
8082 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8083 }
8084
8085 void btrfs_destroy_inode(struct inode *inode)
8086 {
8087 struct btrfs_ordered_extent *ordered;
8088 struct btrfs_root *root = BTRFS_I(inode)->root;
8089
8090 WARN_ON(!hlist_empty(&inode->i_dentry));
8091 WARN_ON(inode->i_data.nrpages);
8092 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8093 WARN_ON(BTRFS_I(inode)->reserved_extents);
8094 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8095 WARN_ON(BTRFS_I(inode)->csum_bytes);
8096
8097 /*
8098 * This can happen where we create an inode, but somebody else also
8099 * created the same inode and we need to destroy the one we already
8100 * created.
8101 */
8102 if (!root)
8103 goto free;
8104
8105 /*
8106 * Make sure we're properly removed from the ordered operation
8107 * lists.
8108 */
8109 smp_mb();
8110 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8111 spin_lock(&root->fs_info->ordered_root_lock);
8112 list_del_init(&BTRFS_I(inode)->ordered_operations);
8113 spin_unlock(&root->fs_info->ordered_root_lock);
8114 }
8115
8116 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8117 &BTRFS_I(inode)->runtime_flags)) {
8118 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8119 btrfs_ino(inode));
8120 atomic_dec(&root->orphan_inodes);
8121 }
8122
8123 while (1) {
8124 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8125 if (!ordered)
8126 break;
8127 else {
8128 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8129 ordered->file_offset, ordered->len);
8130 btrfs_remove_ordered_extent(inode, ordered);
8131 btrfs_put_ordered_extent(ordered);
8132 btrfs_put_ordered_extent(ordered);
8133 }
8134 }
8135 inode_tree_del(inode);
8136 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8137 free:
8138 call_rcu(&inode->i_rcu, btrfs_i_callback);
8139 }
8140
8141 int btrfs_drop_inode(struct inode *inode)
8142 {
8143 struct btrfs_root *root = BTRFS_I(inode)->root;
8144
8145 if (root == NULL)
8146 return 1;
8147
8148 /* the snap/subvol tree is on deleting */
8149 if (btrfs_root_refs(&root->root_item) == 0)
8150 return 1;
8151 else
8152 return generic_drop_inode(inode);
8153 }
8154
8155 static void init_once(void *foo)
8156 {
8157 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8158
8159 inode_init_once(&ei->vfs_inode);
8160 }
8161
8162 void btrfs_destroy_cachep(void)
8163 {
8164 /*
8165 * Make sure all delayed rcu free inodes are flushed before we
8166 * destroy cache.
8167 */
8168 rcu_barrier();
8169 if (btrfs_inode_cachep)
8170 kmem_cache_destroy(btrfs_inode_cachep);
8171 if (btrfs_trans_handle_cachep)
8172 kmem_cache_destroy(btrfs_trans_handle_cachep);
8173 if (btrfs_transaction_cachep)
8174 kmem_cache_destroy(btrfs_transaction_cachep);
8175 if (btrfs_path_cachep)
8176 kmem_cache_destroy(btrfs_path_cachep);
8177 if (btrfs_free_space_cachep)
8178 kmem_cache_destroy(btrfs_free_space_cachep);
8179 if (btrfs_delalloc_work_cachep)
8180 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8181 }
8182
8183 int btrfs_init_cachep(void)
8184 {
8185 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8186 sizeof(struct btrfs_inode), 0,
8187 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8188 if (!btrfs_inode_cachep)
8189 goto fail;
8190
8191 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8192 sizeof(struct btrfs_trans_handle), 0,
8193 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8194 if (!btrfs_trans_handle_cachep)
8195 goto fail;
8196
8197 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8198 sizeof(struct btrfs_transaction), 0,
8199 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8200 if (!btrfs_transaction_cachep)
8201 goto fail;
8202
8203 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8204 sizeof(struct btrfs_path), 0,
8205 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8206 if (!btrfs_path_cachep)
8207 goto fail;
8208
8209 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8210 sizeof(struct btrfs_free_space), 0,
8211 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8212 if (!btrfs_free_space_cachep)
8213 goto fail;
8214
8215 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8216 sizeof(struct btrfs_delalloc_work), 0,
8217 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8218 NULL);
8219 if (!btrfs_delalloc_work_cachep)
8220 goto fail;
8221
8222 return 0;
8223 fail:
8224 btrfs_destroy_cachep();
8225 return -ENOMEM;
8226 }
8227
8228 static int btrfs_getattr(struct vfsmount *mnt,
8229 struct dentry *dentry, struct kstat *stat)
8230 {
8231 u64 delalloc_bytes;
8232 struct inode *inode = dentry->d_inode;
8233 u32 blocksize = inode->i_sb->s_blocksize;
8234
8235 generic_fillattr(inode, stat);
8236 stat->dev = BTRFS_I(inode)->root->anon_dev;
8237 stat->blksize = PAGE_CACHE_SIZE;
8238
8239 spin_lock(&BTRFS_I(inode)->lock);
8240 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8241 spin_unlock(&BTRFS_I(inode)->lock);
8242 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8243 ALIGN(delalloc_bytes, blocksize)) >> 9;
8244 return 0;
8245 }
8246
8247 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8248 struct inode *new_dir, struct dentry *new_dentry)
8249 {
8250 struct btrfs_trans_handle *trans;
8251 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8252 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8253 struct inode *new_inode = new_dentry->d_inode;
8254 struct inode *old_inode = old_dentry->d_inode;
8255 struct timespec ctime = CURRENT_TIME;
8256 u64 index = 0;
8257 u64 root_objectid;
8258 int ret;
8259 u64 old_ino = btrfs_ino(old_inode);
8260
8261 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8262 return -EPERM;
8263
8264 /* we only allow rename subvolume link between subvolumes */
8265 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8266 return -EXDEV;
8267
8268 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8269 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8270 return -ENOTEMPTY;
8271
8272 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8273 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8274 return -ENOTEMPTY;
8275
8276
8277 /* check for collisions, even if the name isn't there */
8278 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8279 new_dentry->d_name.name,
8280 new_dentry->d_name.len);
8281
8282 if (ret) {
8283 if (ret == -EEXIST) {
8284 /* we shouldn't get
8285 * eexist without a new_inode */
8286 if (WARN_ON(!new_inode)) {
8287 return ret;
8288 }
8289 } else {
8290 /* maybe -EOVERFLOW */
8291 return ret;
8292 }
8293 }
8294 ret = 0;
8295
8296 /*
8297 * we're using rename to replace one file with another.
8298 * and the replacement file is large. Start IO on it now so
8299 * we don't add too much work to the end of the transaction
8300 */
8301 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8302 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8303 filemap_flush(old_inode->i_mapping);
8304
8305 /* close the racy window with snapshot create/destroy ioctl */
8306 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8307 down_read(&root->fs_info->subvol_sem);
8308 /*
8309 * We want to reserve the absolute worst case amount of items. So if
8310 * both inodes are subvols and we need to unlink them then that would
8311 * require 4 item modifications, but if they are both normal inodes it
8312 * would require 5 item modifications, so we'll assume their normal
8313 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8314 * should cover the worst case number of items we'll modify.
8315 */
8316 trans = btrfs_start_transaction(root, 11);
8317 if (IS_ERR(trans)) {
8318 ret = PTR_ERR(trans);
8319 goto out_notrans;
8320 }
8321
8322 if (dest != root)
8323 btrfs_record_root_in_trans(trans, dest);
8324
8325 ret = btrfs_set_inode_index(new_dir, &index);
8326 if (ret)
8327 goto out_fail;
8328
8329 BTRFS_I(old_inode)->dir_index = 0ULL;
8330 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8331 /* force full log commit if subvolume involved. */
8332 btrfs_set_log_full_commit(root->fs_info, trans);
8333 } else {
8334 ret = btrfs_insert_inode_ref(trans, dest,
8335 new_dentry->d_name.name,
8336 new_dentry->d_name.len,
8337 old_ino,
8338 btrfs_ino(new_dir), index);
8339 if (ret)
8340 goto out_fail;
8341 /*
8342 * this is an ugly little race, but the rename is required
8343 * to make sure that if we crash, the inode is either at the
8344 * old name or the new one. pinning the log transaction lets
8345 * us make sure we don't allow a log commit to come in after
8346 * we unlink the name but before we add the new name back in.
8347 */
8348 btrfs_pin_log_trans(root);
8349 }
8350 /*
8351 * make sure the inode gets flushed if it is replacing
8352 * something.
8353 */
8354 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8355 btrfs_add_ordered_operation(trans, root, old_inode);
8356
8357 inode_inc_iversion(old_dir);
8358 inode_inc_iversion(new_dir);
8359 inode_inc_iversion(old_inode);
8360 old_dir->i_ctime = old_dir->i_mtime = ctime;
8361 new_dir->i_ctime = new_dir->i_mtime = ctime;
8362 old_inode->i_ctime = ctime;
8363
8364 if (old_dentry->d_parent != new_dentry->d_parent)
8365 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8366
8367 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8368 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8369 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8370 old_dentry->d_name.name,
8371 old_dentry->d_name.len);
8372 } else {
8373 ret = __btrfs_unlink_inode(trans, root, old_dir,
8374 old_dentry->d_inode,
8375 old_dentry->d_name.name,
8376 old_dentry->d_name.len);
8377 if (!ret)
8378 ret = btrfs_update_inode(trans, root, old_inode);
8379 }
8380 if (ret) {
8381 btrfs_abort_transaction(trans, root, ret);
8382 goto out_fail;
8383 }
8384
8385 if (new_inode) {
8386 inode_inc_iversion(new_inode);
8387 new_inode->i_ctime = CURRENT_TIME;
8388 if (unlikely(btrfs_ino(new_inode) ==
8389 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8390 root_objectid = BTRFS_I(new_inode)->location.objectid;
8391 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8392 root_objectid,
8393 new_dentry->d_name.name,
8394 new_dentry->d_name.len);
8395 BUG_ON(new_inode->i_nlink == 0);
8396 } else {
8397 ret = btrfs_unlink_inode(trans, dest, new_dir,
8398 new_dentry->d_inode,
8399 new_dentry->d_name.name,
8400 new_dentry->d_name.len);
8401 }
8402 if (!ret && new_inode->i_nlink == 0)
8403 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8404 if (ret) {
8405 btrfs_abort_transaction(trans, root, ret);
8406 goto out_fail;
8407 }
8408 }
8409
8410 ret = btrfs_add_link(trans, new_dir, old_inode,
8411 new_dentry->d_name.name,
8412 new_dentry->d_name.len, 0, index);
8413 if (ret) {
8414 btrfs_abort_transaction(trans, root, ret);
8415 goto out_fail;
8416 }
8417
8418 if (old_inode->i_nlink == 1)
8419 BTRFS_I(old_inode)->dir_index = index;
8420
8421 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8422 struct dentry *parent = new_dentry->d_parent;
8423 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8424 btrfs_end_log_trans(root);
8425 }
8426 out_fail:
8427 btrfs_end_transaction(trans, root);
8428 out_notrans:
8429 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8430 up_read(&root->fs_info->subvol_sem);
8431
8432 return ret;
8433 }
8434
8435 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8436 {
8437 struct btrfs_delalloc_work *delalloc_work;
8438 struct inode *inode;
8439
8440 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8441 work);
8442 inode = delalloc_work->inode;
8443 if (delalloc_work->wait) {
8444 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8445 } else {
8446 filemap_flush(inode->i_mapping);
8447 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8448 &BTRFS_I(inode)->runtime_flags))
8449 filemap_flush(inode->i_mapping);
8450 }
8451
8452 if (delalloc_work->delay_iput)
8453 btrfs_add_delayed_iput(inode);
8454 else
8455 iput(inode);
8456 complete(&delalloc_work->completion);
8457 }
8458
8459 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8460 int wait, int delay_iput)
8461 {
8462 struct btrfs_delalloc_work *work;
8463
8464 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8465 if (!work)
8466 return NULL;
8467
8468 init_completion(&work->completion);
8469 INIT_LIST_HEAD(&work->list);
8470 work->inode = inode;
8471 work->wait = wait;
8472 work->delay_iput = delay_iput;
8473 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8474
8475 return work;
8476 }
8477
8478 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8479 {
8480 wait_for_completion(&work->completion);
8481 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8482 }
8483
8484 /*
8485 * some fairly slow code that needs optimization. This walks the list
8486 * of all the inodes with pending delalloc and forces them to disk.
8487 */
8488 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8489 int nr)
8490 {
8491 struct btrfs_inode *binode;
8492 struct inode *inode;
8493 struct btrfs_delalloc_work *work, *next;
8494 struct list_head works;
8495 struct list_head splice;
8496 int ret = 0;
8497
8498 INIT_LIST_HEAD(&works);
8499 INIT_LIST_HEAD(&splice);
8500
8501 mutex_lock(&root->delalloc_mutex);
8502 spin_lock(&root->delalloc_lock);
8503 list_splice_init(&root->delalloc_inodes, &splice);
8504 while (!list_empty(&splice)) {
8505 binode = list_entry(splice.next, struct btrfs_inode,
8506 delalloc_inodes);
8507
8508 list_move_tail(&binode->delalloc_inodes,
8509 &root->delalloc_inodes);
8510 inode = igrab(&binode->vfs_inode);
8511 if (!inode) {
8512 cond_resched_lock(&root->delalloc_lock);
8513 continue;
8514 }
8515 spin_unlock(&root->delalloc_lock);
8516
8517 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8518 if (unlikely(!work)) {
8519 if (delay_iput)
8520 btrfs_add_delayed_iput(inode);
8521 else
8522 iput(inode);
8523 ret = -ENOMEM;
8524 goto out;
8525 }
8526 list_add_tail(&work->list, &works);
8527 btrfs_queue_work(root->fs_info->flush_workers,
8528 &work->work);
8529 ret++;
8530 if (nr != -1 && ret >= nr)
8531 goto out;
8532 cond_resched();
8533 spin_lock(&root->delalloc_lock);
8534 }
8535 spin_unlock(&root->delalloc_lock);
8536
8537 out:
8538 list_for_each_entry_safe(work, next, &works, list) {
8539 list_del_init(&work->list);
8540 btrfs_wait_and_free_delalloc_work(work);
8541 }
8542
8543 if (!list_empty_careful(&splice)) {
8544 spin_lock(&root->delalloc_lock);
8545 list_splice_tail(&splice, &root->delalloc_inodes);
8546 spin_unlock(&root->delalloc_lock);
8547 }
8548 mutex_unlock(&root->delalloc_mutex);
8549 return ret;
8550 }
8551
8552 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8553 {
8554 int ret;
8555
8556 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8557 return -EROFS;
8558
8559 ret = __start_delalloc_inodes(root, delay_iput, -1);
8560 if (ret > 0)
8561 ret = 0;
8562 /*
8563 * the filemap_flush will queue IO into the worker threads, but
8564 * we have to make sure the IO is actually started and that
8565 * ordered extents get created before we return
8566 */
8567 atomic_inc(&root->fs_info->async_submit_draining);
8568 while (atomic_read(&root->fs_info->nr_async_submits) ||
8569 atomic_read(&root->fs_info->async_delalloc_pages)) {
8570 wait_event(root->fs_info->async_submit_wait,
8571 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8572 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8573 }
8574 atomic_dec(&root->fs_info->async_submit_draining);
8575 return ret;
8576 }
8577
8578 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8579 int nr)
8580 {
8581 struct btrfs_root *root;
8582 struct list_head splice;
8583 int ret;
8584
8585 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8586 return -EROFS;
8587
8588 INIT_LIST_HEAD(&splice);
8589
8590 mutex_lock(&fs_info->delalloc_root_mutex);
8591 spin_lock(&fs_info->delalloc_root_lock);
8592 list_splice_init(&fs_info->delalloc_roots, &splice);
8593 while (!list_empty(&splice) && nr) {
8594 root = list_first_entry(&splice, struct btrfs_root,
8595 delalloc_root);
8596 root = btrfs_grab_fs_root(root);
8597 BUG_ON(!root);
8598 list_move_tail(&root->delalloc_root,
8599 &fs_info->delalloc_roots);
8600 spin_unlock(&fs_info->delalloc_root_lock);
8601
8602 ret = __start_delalloc_inodes(root, delay_iput, nr);
8603 btrfs_put_fs_root(root);
8604 if (ret < 0)
8605 goto out;
8606
8607 if (nr != -1) {
8608 nr -= ret;
8609 WARN_ON(nr < 0);
8610 }
8611 spin_lock(&fs_info->delalloc_root_lock);
8612 }
8613 spin_unlock(&fs_info->delalloc_root_lock);
8614
8615 ret = 0;
8616 atomic_inc(&fs_info->async_submit_draining);
8617 while (atomic_read(&fs_info->nr_async_submits) ||
8618 atomic_read(&fs_info->async_delalloc_pages)) {
8619 wait_event(fs_info->async_submit_wait,
8620 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8621 atomic_read(&fs_info->async_delalloc_pages) == 0));
8622 }
8623 atomic_dec(&fs_info->async_submit_draining);
8624 out:
8625 if (!list_empty_careful(&splice)) {
8626 spin_lock(&fs_info->delalloc_root_lock);
8627 list_splice_tail(&splice, &fs_info->delalloc_roots);
8628 spin_unlock(&fs_info->delalloc_root_lock);
8629 }
8630 mutex_unlock(&fs_info->delalloc_root_mutex);
8631 return ret;
8632 }
8633
8634 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8635 const char *symname)
8636 {
8637 struct btrfs_trans_handle *trans;
8638 struct btrfs_root *root = BTRFS_I(dir)->root;
8639 struct btrfs_path *path;
8640 struct btrfs_key key;
8641 struct inode *inode = NULL;
8642 int err;
8643 int drop_inode = 0;
8644 u64 objectid;
8645 u64 index = 0;
8646 int name_len;
8647 int datasize;
8648 unsigned long ptr;
8649 struct btrfs_file_extent_item *ei;
8650 struct extent_buffer *leaf;
8651
8652 name_len = strlen(symname);
8653 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8654 return -ENAMETOOLONG;
8655
8656 /*
8657 * 2 items for inode item and ref
8658 * 2 items for dir items
8659 * 1 item for xattr if selinux is on
8660 */
8661 trans = btrfs_start_transaction(root, 5);
8662 if (IS_ERR(trans))
8663 return PTR_ERR(trans);
8664
8665 err = btrfs_find_free_ino(root, &objectid);
8666 if (err)
8667 goto out_unlock;
8668
8669 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8670 dentry->d_name.len, btrfs_ino(dir), objectid,
8671 S_IFLNK|S_IRWXUGO, &index);
8672 if (IS_ERR(inode)) {
8673 err = PTR_ERR(inode);
8674 goto out_unlock;
8675 }
8676
8677 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8678 if (err) {
8679 drop_inode = 1;
8680 goto out_unlock;
8681 }
8682
8683 /*
8684 * If the active LSM wants to access the inode during
8685 * d_instantiate it needs these. Smack checks to see
8686 * if the filesystem supports xattrs by looking at the
8687 * ops vector.
8688 */
8689 inode->i_fop = &btrfs_file_operations;
8690 inode->i_op = &btrfs_file_inode_operations;
8691
8692 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8693 if (err)
8694 drop_inode = 1;
8695 else {
8696 inode->i_mapping->a_ops = &btrfs_aops;
8697 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8698 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8699 }
8700 if (drop_inode)
8701 goto out_unlock;
8702
8703 path = btrfs_alloc_path();
8704 if (!path) {
8705 err = -ENOMEM;
8706 drop_inode = 1;
8707 goto out_unlock;
8708 }
8709 key.objectid = btrfs_ino(inode);
8710 key.offset = 0;
8711 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8712 datasize = btrfs_file_extent_calc_inline_size(name_len);
8713 err = btrfs_insert_empty_item(trans, root, path, &key,
8714 datasize);
8715 if (err) {
8716 drop_inode = 1;
8717 btrfs_free_path(path);
8718 goto out_unlock;
8719 }
8720 leaf = path->nodes[0];
8721 ei = btrfs_item_ptr(leaf, path->slots[0],
8722 struct btrfs_file_extent_item);
8723 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8724 btrfs_set_file_extent_type(leaf, ei,
8725 BTRFS_FILE_EXTENT_INLINE);
8726 btrfs_set_file_extent_encryption(leaf, ei, 0);
8727 btrfs_set_file_extent_compression(leaf, ei, 0);
8728 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8729 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8730
8731 ptr = btrfs_file_extent_inline_start(ei);
8732 write_extent_buffer(leaf, symname, ptr, name_len);
8733 btrfs_mark_buffer_dirty(leaf);
8734 btrfs_free_path(path);
8735
8736 inode->i_op = &btrfs_symlink_inode_operations;
8737 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8738 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8739 inode_set_bytes(inode, name_len);
8740 btrfs_i_size_write(inode, name_len);
8741 err = btrfs_update_inode(trans, root, inode);
8742 if (err)
8743 drop_inode = 1;
8744
8745 out_unlock:
8746 if (!err)
8747 d_instantiate(dentry, inode);
8748 btrfs_end_transaction(trans, root);
8749 if (drop_inode) {
8750 inode_dec_link_count(inode);
8751 iput(inode);
8752 }
8753 btrfs_btree_balance_dirty(root);
8754 return err;
8755 }
8756
8757 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8758 u64 start, u64 num_bytes, u64 min_size,
8759 loff_t actual_len, u64 *alloc_hint,
8760 struct btrfs_trans_handle *trans)
8761 {
8762 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8763 struct extent_map *em;
8764 struct btrfs_root *root = BTRFS_I(inode)->root;
8765 struct btrfs_key ins;
8766 u64 cur_offset = start;
8767 u64 i_size;
8768 u64 cur_bytes;
8769 int ret = 0;
8770 bool own_trans = true;
8771
8772 if (trans)
8773 own_trans = false;
8774 while (num_bytes > 0) {
8775 if (own_trans) {
8776 trans = btrfs_start_transaction(root, 3);
8777 if (IS_ERR(trans)) {
8778 ret = PTR_ERR(trans);
8779 break;
8780 }
8781 }
8782
8783 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8784 cur_bytes = max(cur_bytes, min_size);
8785 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8786 *alloc_hint, &ins, 1);
8787 if (ret) {
8788 if (own_trans)
8789 btrfs_end_transaction(trans, root);
8790 break;
8791 }
8792
8793 ret = insert_reserved_file_extent(trans, inode,
8794 cur_offset, ins.objectid,
8795 ins.offset, ins.offset,
8796 ins.offset, 0, 0, 0,
8797 BTRFS_FILE_EXTENT_PREALLOC);
8798 if (ret) {
8799 btrfs_free_reserved_extent(root, ins.objectid,
8800 ins.offset);
8801 btrfs_abort_transaction(trans, root, ret);
8802 if (own_trans)
8803 btrfs_end_transaction(trans, root);
8804 break;
8805 }
8806 btrfs_drop_extent_cache(inode, cur_offset,
8807 cur_offset + ins.offset -1, 0);
8808
8809 em = alloc_extent_map();
8810 if (!em) {
8811 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8812 &BTRFS_I(inode)->runtime_flags);
8813 goto next;
8814 }
8815
8816 em->start = cur_offset;
8817 em->orig_start = cur_offset;
8818 em->len = ins.offset;
8819 em->block_start = ins.objectid;
8820 em->block_len = ins.offset;
8821 em->orig_block_len = ins.offset;
8822 em->ram_bytes = ins.offset;
8823 em->bdev = root->fs_info->fs_devices->latest_bdev;
8824 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8825 em->generation = trans->transid;
8826
8827 while (1) {
8828 write_lock(&em_tree->lock);
8829 ret = add_extent_mapping(em_tree, em, 1);
8830 write_unlock(&em_tree->lock);
8831 if (ret != -EEXIST)
8832 break;
8833 btrfs_drop_extent_cache(inode, cur_offset,
8834 cur_offset + ins.offset - 1,
8835 0);
8836 }
8837 free_extent_map(em);
8838 next:
8839 num_bytes -= ins.offset;
8840 cur_offset += ins.offset;
8841 *alloc_hint = ins.objectid + ins.offset;
8842
8843 inode_inc_iversion(inode);
8844 inode->i_ctime = CURRENT_TIME;
8845 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8846 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8847 (actual_len > inode->i_size) &&
8848 (cur_offset > inode->i_size)) {
8849 if (cur_offset > actual_len)
8850 i_size = actual_len;
8851 else
8852 i_size = cur_offset;
8853 i_size_write(inode, i_size);
8854 btrfs_ordered_update_i_size(inode, i_size, NULL);
8855 }
8856
8857 ret = btrfs_update_inode(trans, root, inode);
8858
8859 if (ret) {
8860 btrfs_abort_transaction(trans, root, ret);
8861 if (own_trans)
8862 btrfs_end_transaction(trans, root);
8863 break;
8864 }
8865
8866 if (own_trans)
8867 btrfs_end_transaction(trans, root);
8868 }
8869 return ret;
8870 }
8871
8872 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8873 u64 start, u64 num_bytes, u64 min_size,
8874 loff_t actual_len, u64 *alloc_hint)
8875 {
8876 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8877 min_size, actual_len, alloc_hint,
8878 NULL);
8879 }
8880
8881 int btrfs_prealloc_file_range_trans(struct inode *inode,
8882 struct btrfs_trans_handle *trans, int mode,
8883 u64 start, u64 num_bytes, u64 min_size,
8884 loff_t actual_len, u64 *alloc_hint)
8885 {
8886 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8887 min_size, actual_len, alloc_hint, trans);
8888 }
8889
8890 static int btrfs_set_page_dirty(struct page *page)
8891 {
8892 return __set_page_dirty_nobuffers(page);
8893 }
8894
8895 static int btrfs_permission(struct inode *inode, int mask)
8896 {
8897 struct btrfs_root *root = BTRFS_I(inode)->root;
8898 umode_t mode = inode->i_mode;
8899
8900 if (mask & MAY_WRITE &&
8901 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8902 if (btrfs_root_readonly(root))
8903 return -EROFS;
8904 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8905 return -EACCES;
8906 }
8907 return generic_permission(inode, mask);
8908 }
8909
8910 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
8911 {
8912 struct btrfs_trans_handle *trans;
8913 struct btrfs_root *root = BTRFS_I(dir)->root;
8914 struct inode *inode = NULL;
8915 u64 objectid;
8916 u64 index;
8917 int ret = 0;
8918
8919 /*
8920 * 5 units required for adding orphan entry
8921 */
8922 trans = btrfs_start_transaction(root, 5);
8923 if (IS_ERR(trans))
8924 return PTR_ERR(trans);
8925
8926 ret = btrfs_find_free_ino(root, &objectid);
8927 if (ret)
8928 goto out;
8929
8930 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
8931 btrfs_ino(dir), objectid, mode, &index);
8932 if (IS_ERR(inode)) {
8933 ret = PTR_ERR(inode);
8934 inode = NULL;
8935 goto out;
8936 }
8937
8938 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
8939 if (ret)
8940 goto out;
8941
8942 ret = btrfs_update_inode(trans, root, inode);
8943 if (ret)
8944 goto out;
8945
8946 inode->i_fop = &btrfs_file_operations;
8947 inode->i_op = &btrfs_file_inode_operations;
8948
8949 inode->i_mapping->a_ops = &btrfs_aops;
8950 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8951 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8952
8953 ret = btrfs_orphan_add(trans, inode);
8954 if (ret)
8955 goto out;
8956
8957 d_tmpfile(dentry, inode);
8958 mark_inode_dirty(inode);
8959
8960 out:
8961 btrfs_end_transaction(trans, root);
8962 if (ret)
8963 iput(inode);
8964 btrfs_balance_delayed_items(root);
8965 btrfs_btree_balance_dirty(root);
8966
8967 return ret;
8968 }
8969
8970 static const struct inode_operations btrfs_dir_inode_operations = {
8971 .getattr = btrfs_getattr,
8972 .lookup = btrfs_lookup,
8973 .create = btrfs_create,
8974 .unlink = btrfs_unlink,
8975 .link = btrfs_link,
8976 .mkdir = btrfs_mkdir,
8977 .rmdir = btrfs_rmdir,
8978 .rename = btrfs_rename,
8979 .symlink = btrfs_symlink,
8980 .setattr = btrfs_setattr,
8981 .mknod = btrfs_mknod,
8982 .setxattr = btrfs_setxattr,
8983 .getxattr = btrfs_getxattr,
8984 .listxattr = btrfs_listxattr,
8985 .removexattr = btrfs_removexattr,
8986 .permission = btrfs_permission,
8987 .get_acl = btrfs_get_acl,
8988 .set_acl = btrfs_set_acl,
8989 .update_time = btrfs_update_time,
8990 .tmpfile = btrfs_tmpfile,
8991 };
8992 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8993 .lookup = btrfs_lookup,
8994 .permission = btrfs_permission,
8995 .get_acl = btrfs_get_acl,
8996 .set_acl = btrfs_set_acl,
8997 .update_time = btrfs_update_time,
8998 };
8999
9000 static const struct file_operations btrfs_dir_file_operations = {
9001 .llseek = generic_file_llseek,
9002 .read = generic_read_dir,
9003 .iterate = btrfs_real_readdir,
9004 .unlocked_ioctl = btrfs_ioctl,
9005 #ifdef CONFIG_COMPAT
9006 .compat_ioctl = btrfs_ioctl,
9007 #endif
9008 .release = btrfs_release_file,
9009 .fsync = btrfs_sync_file,
9010 };
9011
9012 static struct extent_io_ops btrfs_extent_io_ops = {
9013 .fill_delalloc = run_delalloc_range,
9014 .submit_bio_hook = btrfs_submit_bio_hook,
9015 .merge_bio_hook = btrfs_merge_bio_hook,
9016 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9017 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9018 .writepage_start_hook = btrfs_writepage_start_hook,
9019 .set_bit_hook = btrfs_set_bit_hook,
9020 .clear_bit_hook = btrfs_clear_bit_hook,
9021 .merge_extent_hook = btrfs_merge_extent_hook,
9022 .split_extent_hook = btrfs_split_extent_hook,
9023 };
9024
9025 /*
9026 * btrfs doesn't support the bmap operation because swapfiles
9027 * use bmap to make a mapping of extents in the file. They assume
9028 * these extents won't change over the life of the file and they
9029 * use the bmap result to do IO directly to the drive.
9030 *
9031 * the btrfs bmap call would return logical addresses that aren't
9032 * suitable for IO and they also will change frequently as COW
9033 * operations happen. So, swapfile + btrfs == corruption.
9034 *
9035 * For now we're avoiding this by dropping bmap.
9036 */
9037 static const struct address_space_operations btrfs_aops = {
9038 .readpage = btrfs_readpage,
9039 .writepage = btrfs_writepage,
9040 .writepages = btrfs_writepages,
9041 .readpages = btrfs_readpages,
9042 .direct_IO = btrfs_direct_IO,
9043 .invalidatepage = btrfs_invalidatepage,
9044 .releasepage = btrfs_releasepage,
9045 .set_page_dirty = btrfs_set_page_dirty,
9046 .error_remove_page = generic_error_remove_page,
9047 };
9048
9049 static const struct address_space_operations btrfs_symlink_aops = {
9050 .readpage = btrfs_readpage,
9051 .writepage = btrfs_writepage,
9052 .invalidatepage = btrfs_invalidatepage,
9053 .releasepage = btrfs_releasepage,
9054 };
9055
9056 static const struct inode_operations btrfs_file_inode_operations = {
9057 .getattr = btrfs_getattr,
9058 .setattr = btrfs_setattr,
9059 .setxattr = btrfs_setxattr,
9060 .getxattr = btrfs_getxattr,
9061 .listxattr = btrfs_listxattr,
9062 .removexattr = btrfs_removexattr,
9063 .permission = btrfs_permission,
9064 .fiemap = btrfs_fiemap,
9065 .get_acl = btrfs_get_acl,
9066 .set_acl = btrfs_set_acl,
9067 .update_time = btrfs_update_time,
9068 };
9069 static const struct inode_operations btrfs_special_inode_operations = {
9070 .getattr = btrfs_getattr,
9071 .setattr = btrfs_setattr,
9072 .permission = btrfs_permission,
9073 .setxattr = btrfs_setxattr,
9074 .getxattr = btrfs_getxattr,
9075 .listxattr = btrfs_listxattr,
9076 .removexattr = btrfs_removexattr,
9077 .get_acl = btrfs_get_acl,
9078 .set_acl = btrfs_set_acl,
9079 .update_time = btrfs_update_time,
9080 };
9081 static const struct inode_operations btrfs_symlink_inode_operations = {
9082 .readlink = generic_readlink,
9083 .follow_link = page_follow_link_light,
9084 .put_link = page_put_link,
9085 .getattr = btrfs_getattr,
9086 .setattr = btrfs_setattr,
9087 .permission = btrfs_permission,
9088 .setxattr = btrfs_setxattr,
9089 .getxattr = btrfs_getxattr,
9090 .listxattr = btrfs_listxattr,
9091 .removexattr = btrfs_removexattr,
9092 .update_time = btrfs_update_time,
9093 };
9094
9095 const struct dentry_operations btrfs_dentry_operations = {
9096 .d_delete = btrfs_dentry_delete,
9097 .d_release = btrfs_dentry_release,
9098 };
Linux kernel - Deliverables
This page took 0.227223 seconds and 5 git commands to generate.