projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'pm+acpi-3.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
[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 "compat.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
61 struct btrfs_iget_args {
62 u64 ino;
63 struct btrfs_root *root;
64 };
65
66 static const struct inode_operations btrfs_dir_inode_operations;
67 static const struct inode_operations btrfs_symlink_inode_operations;
68 static const struct inode_operations btrfs_dir_ro_inode_operations;
69 static const struct inode_operations btrfs_special_inode_operations;
70 static const struct inode_operations btrfs_file_inode_operations;
71 static const struct address_space_operations btrfs_aops;
72 static const struct address_space_operations btrfs_symlink_aops;
73 static const struct file_operations btrfs_dir_file_operations;
74 static struct extent_io_ops btrfs_extent_io_ops;
75
76 static struct kmem_cache *btrfs_inode_cachep;
77 static struct kmem_cache *btrfs_delalloc_work_cachep;
78 struct kmem_cache *btrfs_trans_handle_cachep;
79 struct kmem_cache *btrfs_transaction_cachep;
80 struct kmem_cache *btrfs_path_cachep;
81 struct kmem_cache *btrfs_free_space_cachep;
82
83 #define S_SHIFT 12
84 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
85 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
86 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
87 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
88 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
89 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
90 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
91 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
92 };
93
94 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
95 static int btrfs_truncate(struct inode *inode);
96 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
97 static noinline int cow_file_range(struct inode *inode,
98 struct page *locked_page,
99 u64 start, u64 end, int *page_started,
100 unsigned long *nr_written, int unlock);
101 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
102 u64 len, u64 orig_start,
103 u64 block_start, u64 block_len,
104 u64 orig_block_len, u64 ram_bytes,
105 int type);
106
107 static int btrfs_dirty_inode(struct inode *inode);
108
109 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
110 struct inode *inode, struct inode *dir,
111 const struct qstr *qstr)
112 {
113 int err;
114
115 err = btrfs_init_acl(trans, inode, dir);
116 if (!err)
117 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
118 return err;
119 }
120
121 /*
122 * this does all the hard work for inserting an inline extent into
123 * the btree. The caller should have done a btrfs_drop_extents so that
124 * no overlapping inline items exist in the btree
125 */
126 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
127 struct btrfs_root *root, struct inode *inode,
128 u64 start, size_t size, size_t compressed_size,
129 int compress_type,
130 struct page **compressed_pages)
131 {
132 struct btrfs_key key;
133 struct btrfs_path *path;
134 struct extent_buffer *leaf;
135 struct page *page = NULL;
136 char *kaddr;
137 unsigned long ptr;
138 struct btrfs_file_extent_item *ei;
139 int err = 0;
140 int ret;
141 size_t cur_size = size;
142 size_t datasize;
143 unsigned long offset;
144
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
147
148 path = btrfs_alloc_path();
149 if (!path)
150 return -ENOMEM;
151
152 path->leave_spinning = 1;
153
154 key.objectid = btrfs_ino(inode);
155 key.offset = start;
156 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
157 datasize = btrfs_file_extent_calc_inline_size(cur_size);
158
159 inode_add_bytes(inode, size);
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
161 datasize);
162 if (ret) {
163 err = ret;
164 goto fail;
165 }
166 leaf = path->nodes[0];
167 ei = btrfs_item_ptr(leaf, path->slots[0],
168 struct btrfs_file_extent_item);
169 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
170 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
171 btrfs_set_file_extent_encryption(leaf, ei, 0);
172 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
173 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
174 ptr = btrfs_file_extent_inline_start(ei);
175
176 if (compress_type != BTRFS_COMPRESS_NONE) {
177 struct page *cpage;
178 int i = 0;
179 while (compressed_size > 0) {
180 cpage = compressed_pages[i];
181 cur_size = min_t(unsigned long, compressed_size,
182 PAGE_CACHE_SIZE);
183
184 kaddr = kmap_atomic(cpage);
185 write_extent_buffer(leaf, kaddr, ptr, cur_size);
186 kunmap_atomic(kaddr);
187
188 i++;
189 ptr += cur_size;
190 compressed_size -= cur_size;
191 }
192 btrfs_set_file_extent_compression(leaf, ei,
193 compress_type);
194 } else {
195 page = find_get_page(inode->i_mapping,
196 start >> PAGE_CACHE_SHIFT);
197 btrfs_set_file_extent_compression(leaf, ei, 0);
198 kaddr = kmap_atomic(page);
199 offset = start & (PAGE_CACHE_SIZE - 1);
200 write_extent_buffer(leaf, kaddr + offset, ptr, size);
201 kunmap_atomic(kaddr);
202 page_cache_release(page);
203 }
204 btrfs_mark_buffer_dirty(leaf);
205 btrfs_free_path(path);
206
207 /*
208 * we're an inline extent, so nobody can
209 * extend the file past i_size without locking
210 * a page we already have locked.
211 *
212 * We must do any isize and inode updates
213 * before we unlock the pages. Otherwise we
214 * could end up racing with unlink.
215 */
216 BTRFS_I(inode)->disk_i_size = inode->i_size;
217 ret = btrfs_update_inode(trans, root, inode);
218
219 return ret;
220 fail:
221 btrfs_free_path(path);
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_trans_handle *trans,
232 struct btrfs_root *root,
233 struct inode *inode, u64 start, u64 end,
234 size_t compressed_size, int compress_type,
235 struct page **compressed_pages)
236 {
237 u64 isize = i_size_read(inode);
238 u64 actual_end = min(end + 1, isize);
239 u64 inline_len = actual_end - start;
240 u64 aligned_end = ALIGN(end, root->sectorsize);
241 u64 data_len = inline_len;
242 int ret;
243
244 if (compressed_size)
245 data_len = compressed_size;
246
247 if (start > 0 ||
248 actual_end >= PAGE_CACHE_SIZE ||
249 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
250 (!compressed_size &&
251 (actual_end & (root->sectorsize - 1)) == 0) ||
252 end + 1 < isize ||
253 data_len > root->fs_info->max_inline) {
254 return 1;
255 }
256
257 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
258 if (ret)
259 return ret;
260
261 if (isize > actual_end)
262 inline_len = min_t(u64, isize, actual_end);
263 ret = insert_inline_extent(trans, root, inode, start,
264 inline_len, compressed_size,
265 compress_type, compressed_pages);
266 if (ret && ret != -ENOSPC) {
267 btrfs_abort_transaction(trans, root, ret);
268 return ret;
269 } else if (ret == -ENOSPC) {
270 return 1;
271 }
272
273 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
274 btrfs_delalloc_release_metadata(inode, end + 1 - start);
275 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
276 return 0;
277 }
278
279 struct async_extent {
280 u64 start;
281 u64 ram_size;
282 u64 compressed_size;
283 struct page **pages;
284 unsigned long nr_pages;
285 int compress_type;
286 struct list_head list;
287 };
288
289 struct async_cow {
290 struct inode *inode;
291 struct btrfs_root *root;
292 struct page *locked_page;
293 u64 start;
294 u64 end;
295 struct list_head extents;
296 struct btrfs_work work;
297 };
298
299 static noinline int add_async_extent(struct async_cow *cow,
300 u64 start, u64 ram_size,
301 u64 compressed_size,
302 struct page **pages,
303 unsigned long nr_pages,
304 int compress_type)
305 {
306 struct async_extent *async_extent;
307
308 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
309 BUG_ON(!async_extent); /* -ENOMEM */
310 async_extent->start = start;
311 async_extent->ram_size = ram_size;
312 async_extent->compressed_size = compressed_size;
313 async_extent->pages = pages;
314 async_extent->nr_pages = nr_pages;
315 async_extent->compress_type = compress_type;
316 list_add_tail(&async_extent->list, &cow->extents);
317 return 0;
318 }
319
320 /*
321 * we create compressed extents in two phases. The first
322 * phase compresses a range of pages that have already been
323 * locked (both pages and state bits are locked).
324 *
325 * This is done inside an ordered work queue, and the compression
326 * is spread across many cpus. The actual IO submission is step
327 * two, and the ordered work queue takes care of making sure that
328 * happens in the same order things were put onto the queue by
329 * writepages and friends.
330 *
331 * If this code finds it can't get good compression, it puts an
332 * entry onto the work queue to write the uncompressed bytes. This
333 * makes sure that both compressed inodes and uncompressed inodes
334 * are written in the same order that the flusher thread sent them
335 * down.
336 */
337 static noinline int compress_file_range(struct inode *inode,
338 struct page *locked_page,
339 u64 start, u64 end,
340 struct async_cow *async_cow,
341 int *num_added)
342 {
343 struct btrfs_root *root = BTRFS_I(inode)->root;
344 struct btrfs_trans_handle *trans;
345 u64 num_bytes;
346 u64 blocksize = root->sectorsize;
347 u64 actual_end;
348 u64 isize = i_size_read(inode);
349 int ret = 0;
350 struct page **pages = NULL;
351 unsigned long nr_pages;
352 unsigned long nr_pages_ret = 0;
353 unsigned long total_compressed = 0;
354 unsigned long total_in = 0;
355 unsigned long max_compressed = 128 * 1024;
356 unsigned long max_uncompressed = 128 * 1024;
357 int i;
358 int will_compress;
359 int compress_type = root->fs_info->compress_type;
360 int redirty = 0;
361
362 /* if this is a small write inside eof, kick off a defrag */
363 if ((end - start + 1) < 16 * 1024 &&
364 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
365 btrfs_add_inode_defrag(NULL, inode);
366
367 actual_end = min_t(u64, isize, end + 1);
368 again:
369 will_compress = 0;
370 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
371 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
372
373 /*
374 * we don't want to send crud past the end of i_size through
375 * compression, that's just a waste of CPU time. So, if the
376 * end of the file is before the start of our current
377 * requested range of bytes, we bail out to the uncompressed
378 * cleanup code that can deal with all of this.
379 *
380 * It isn't really the fastest way to fix things, but this is a
381 * very uncommon corner.
382 */
383 if (actual_end <= start)
384 goto cleanup_and_bail_uncompressed;
385
386 total_compressed = actual_end - start;
387
388 /* we want to make sure that amount of ram required to uncompress
389 * an extent is reasonable, so we limit the total size in ram
390 * of a compressed extent to 128k. This is a crucial number
391 * because it also controls how easily we can spread reads across
392 * cpus for decompression.
393 *
394 * We also want to make sure the amount of IO required to do
395 * a random read is reasonably small, so we limit the size of
396 * a compressed extent to 128k.
397 */
398 total_compressed = min(total_compressed, max_uncompressed);
399 num_bytes = ALIGN(end - start + 1, blocksize);
400 num_bytes = max(blocksize, num_bytes);
401 total_in = 0;
402 ret = 0;
403
404 /*
405 * we do compression for mount -o compress and when the
406 * inode has not been flagged as nocompress. This flag can
407 * change at any time if we discover bad compression ratios.
408 */
409 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
410 (btrfs_test_opt(root, COMPRESS) ||
411 (BTRFS_I(inode)->force_compress) ||
412 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
413 WARN_ON(pages);
414 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
415 if (!pages) {
416 /* just bail out to the uncompressed code */
417 goto cont;
418 }
419
420 if (BTRFS_I(inode)->force_compress)
421 compress_type = BTRFS_I(inode)->force_compress;
422
423 /*
424 * we need to call clear_page_dirty_for_io on each
425 * page in the range. Otherwise applications with the file
426 * mmap'd can wander in and change the page contents while
427 * we are compressing them.
428 *
429 * If the compression fails for any reason, we set the pages
430 * dirty again later on.
431 */
432 extent_range_clear_dirty_for_io(inode, start, end);
433 redirty = 1;
434 ret = btrfs_compress_pages(compress_type,
435 inode->i_mapping, start,
436 total_compressed, pages,
437 nr_pages, &nr_pages_ret,
438 &total_in,
439 &total_compressed,
440 max_compressed);
441
442 if (!ret) {
443 unsigned long offset = total_compressed &
444 (PAGE_CACHE_SIZE - 1);
445 struct page *page = pages[nr_pages_ret - 1];
446 char *kaddr;
447
448 /* zero the tail end of the last page, we might be
449 * sending it down to disk
450 */
451 if (offset) {
452 kaddr = kmap_atomic(page);
453 memset(kaddr + offset, 0,
454 PAGE_CACHE_SIZE - offset);
455 kunmap_atomic(kaddr);
456 }
457 will_compress = 1;
458 }
459 }
460 cont:
461 if (start == 0) {
462 trans = btrfs_join_transaction(root);
463 if (IS_ERR(trans)) {
464 ret = PTR_ERR(trans);
465 trans = NULL;
466 goto cleanup_and_out;
467 }
468 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
469
470 /* lets try to make an inline extent */
471 if (ret || total_in < (actual_end - start)) {
472 /* we didn't compress the entire range, try
473 * to make an uncompressed inline extent.
474 */
475 ret = cow_file_range_inline(trans, root, inode,
476 start, end, 0, 0, NULL);
477 } else {
478 /* try making a compressed inline extent */
479 ret = cow_file_range_inline(trans, root, inode,
480 start, end,
481 total_compressed,
482 compress_type, pages);
483 }
484 if (ret <= 0) {
485 /*
486 * inline extent creation worked or returned error,
487 * we don't need to create any more async work items.
488 * Unlock and free up our temp pages.
489 */
490 extent_clear_unlock_delalloc(inode,
491 &BTRFS_I(inode)->io_tree,
492 start, end, NULL,
493 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
494 EXTENT_CLEAR_DELALLOC |
495 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
496
497 btrfs_end_transaction(trans, root);
498 goto free_pages_out;
499 }
500 btrfs_end_transaction(trans, root);
501 }
502
503 if (will_compress) {
504 /*
505 * we aren't doing an inline extent round the compressed size
506 * up to a block size boundary so the allocator does sane
507 * things
508 */
509 total_compressed = ALIGN(total_compressed, blocksize);
510
511 /*
512 * one last check to make sure the compression is really a
513 * win, compare the page count read with the blocks on disk
514 */
515 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
516 if (total_compressed >= total_in) {
517 will_compress = 0;
518 } else {
519 num_bytes = total_in;
520 }
521 }
522 if (!will_compress && pages) {
523 /*
524 * the compression code ran but failed to make things smaller,
525 * free any pages it allocated and our page pointer array
526 */
527 for (i = 0; i < nr_pages_ret; i++) {
528 WARN_ON(pages[i]->mapping);
529 page_cache_release(pages[i]);
530 }
531 kfree(pages);
532 pages = NULL;
533 total_compressed = 0;
534 nr_pages_ret = 0;
535
536 /* flag the file so we don't compress in the future */
537 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
538 !(BTRFS_I(inode)->force_compress)) {
539 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
540 }
541 }
542 if (will_compress) {
543 *num_added += 1;
544
545 /* the async work queues will take care of doing actual
546 * allocation on disk for these compressed pages,
547 * and will submit them to the elevator.
548 */
549 add_async_extent(async_cow, start, num_bytes,
550 total_compressed, pages, nr_pages_ret,
551 compress_type);
552
553 if (start + num_bytes < end) {
554 start += num_bytes;
555 pages = NULL;
556 cond_resched();
557 goto again;
558 }
559 } else {
560 cleanup_and_bail_uncompressed:
561 /*
562 * No compression, but we still need to write the pages in
563 * the file we've been given so far. redirty the locked
564 * page if it corresponds to our extent and set things up
565 * for the async work queue to run cow_file_range to do
566 * the normal delalloc dance
567 */
568 if (page_offset(locked_page) >= start &&
569 page_offset(locked_page) <= end) {
570 __set_page_dirty_nobuffers(locked_page);
571 /* unlocked later on in the async handlers */
572 }
573 if (redirty)
574 extent_range_redirty_for_io(inode, start, end);
575 add_async_extent(async_cow, start, end - start + 1,
576 0, NULL, 0, BTRFS_COMPRESS_NONE);
577 *num_added += 1;
578 }
579
580 out:
581 return ret;
582
583 free_pages_out:
584 for (i = 0; i < nr_pages_ret; i++) {
585 WARN_ON(pages[i]->mapping);
586 page_cache_release(pages[i]);
587 }
588 kfree(pages);
589
590 goto out;
591
592 cleanup_and_out:
593 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
594 start, end, NULL,
595 EXTENT_CLEAR_UNLOCK_PAGE |
596 EXTENT_CLEAR_DIRTY |
597 EXTENT_CLEAR_DELALLOC |
598 EXTENT_SET_WRITEBACK |
599 EXTENT_END_WRITEBACK);
600 if (!trans || IS_ERR(trans))
601 btrfs_error(root->fs_info, ret, "Failed to join transaction");
602 else
603 btrfs_abort_transaction(trans, root, ret);
604 goto free_pages_out;
605 }
606
607 /*
608 * phase two of compressed writeback. This is the ordered portion
609 * of the code, which only gets called in the order the work was
610 * queued. We walk all the async extents created by compress_file_range
611 * and send them down to the disk.
612 */
613 static noinline int submit_compressed_extents(struct inode *inode,
614 struct async_cow *async_cow)
615 {
616 struct async_extent *async_extent;
617 u64 alloc_hint = 0;
618 struct btrfs_trans_handle *trans;
619 struct btrfs_key ins;
620 struct extent_map *em;
621 struct btrfs_root *root = BTRFS_I(inode)->root;
622 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
623 struct extent_io_tree *io_tree;
624 int ret = 0;
625
626 if (list_empty(&async_cow->extents))
627 return 0;
628
629 again:
630 while (!list_empty(&async_cow->extents)) {
631 async_extent = list_entry(async_cow->extents.next,
632 struct async_extent, list);
633 list_del(&async_extent->list);
634
635 io_tree = &BTRFS_I(inode)->io_tree;
636
637 retry:
638 /* did the compression code fall back to uncompressed IO? */
639 if (!async_extent->pages) {
640 int page_started = 0;
641 unsigned long nr_written = 0;
642
643 lock_extent(io_tree, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1);
646
647 /* allocate blocks */
648 ret = cow_file_range(inode, async_cow->locked_page,
649 async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1,
652 &page_started, &nr_written, 0);
653
654 /* JDM XXX */
655
656 /*
657 * if page_started, cow_file_range inserted an
658 * inline extent and took care of all the unlocking
659 * and IO for us. Otherwise, we need to submit
660 * all those pages down to the drive.
661 */
662 if (!page_started && !ret)
663 extent_write_locked_range(io_tree,
664 inode, async_extent->start,
665 async_extent->start +
666 async_extent->ram_size - 1,
667 btrfs_get_extent,
668 WB_SYNC_ALL);
669 else if (ret)
670 unlock_page(async_cow->locked_page);
671 kfree(async_extent);
672 cond_resched();
673 continue;
674 }
675
676 lock_extent(io_tree, async_extent->start,
677 async_extent->start + async_extent->ram_size - 1);
678
679 trans = btrfs_join_transaction(root);
680 if (IS_ERR(trans)) {
681 ret = PTR_ERR(trans);
682 } else {
683 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
684 ret = btrfs_reserve_extent(trans, root,
685 async_extent->compressed_size,
686 async_extent->compressed_size,
687 0, alloc_hint, &ins, 1);
688 if (ret && ret != -ENOSPC)
689 btrfs_abort_transaction(trans, root, ret);
690 btrfs_end_transaction(trans, root);
691 }
692
693 if (ret) {
694 int i;
695
696 for (i = 0; i < async_extent->nr_pages; i++) {
697 WARN_ON(async_extent->pages[i]->mapping);
698 page_cache_release(async_extent->pages[i]);
699 }
700 kfree(async_extent->pages);
701 async_extent->nr_pages = 0;
702 async_extent->pages = NULL;
703
704 if (ret == -ENOSPC)
705 goto retry;
706 goto out_free;
707 }
708
709 /*
710 * here we're doing allocation and writeback of the
711 * compressed pages
712 */
713 btrfs_drop_extent_cache(inode, async_extent->start,
714 async_extent->start +
715 async_extent->ram_size - 1, 0);
716
717 em = alloc_extent_map();
718 if (!em) {
719 ret = -ENOMEM;
720 goto out_free_reserve;
721 }
722 em->start = async_extent->start;
723 em->len = async_extent->ram_size;
724 em->orig_start = em->start;
725 em->mod_start = em->start;
726 em->mod_len = em->len;
727
728 em->block_start = ins.objectid;
729 em->block_len = ins.offset;
730 em->orig_block_len = ins.offset;
731 em->ram_bytes = async_extent->ram_size;
732 em->bdev = root->fs_info->fs_devices->latest_bdev;
733 em->compress_type = async_extent->compress_type;
734 set_bit(EXTENT_FLAG_PINNED, &em->flags);
735 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
736 em->generation = -1;
737
738 while (1) {
739 write_lock(&em_tree->lock);
740 ret = add_extent_mapping(em_tree, em, 1);
741 write_unlock(&em_tree->lock);
742 if (ret != -EEXIST) {
743 free_extent_map(em);
744 break;
745 }
746 btrfs_drop_extent_cache(inode, async_extent->start,
747 async_extent->start +
748 async_extent->ram_size - 1, 0);
749 }
750
751 if (ret)
752 goto out_free_reserve;
753
754 ret = btrfs_add_ordered_extent_compress(inode,
755 async_extent->start,
756 ins.objectid,
757 async_extent->ram_size,
758 ins.offset,
759 BTRFS_ORDERED_COMPRESSED,
760 async_extent->compress_type);
761 if (ret)
762 goto out_free_reserve;
763
764 /*
765 * clear dirty, set writeback and unlock the pages.
766 */
767 extent_clear_unlock_delalloc(inode,
768 &BTRFS_I(inode)->io_tree,
769 async_extent->start,
770 async_extent->start +
771 async_extent->ram_size - 1,
772 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
773 EXTENT_CLEAR_UNLOCK |
774 EXTENT_CLEAR_DELALLOC |
775 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
776
777 ret = btrfs_submit_compressed_write(inode,
778 async_extent->start,
779 async_extent->ram_size,
780 ins.objectid,
781 ins.offset, async_extent->pages,
782 async_extent->nr_pages);
783 alloc_hint = ins.objectid + ins.offset;
784 kfree(async_extent);
785 if (ret)
786 goto out;
787 cond_resched();
788 }
789 ret = 0;
790 out:
791 return ret;
792 out_free_reserve:
793 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
794 out_free:
795 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
796 async_extent->start,
797 async_extent->start +
798 async_extent->ram_size - 1,
799 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
800 EXTENT_CLEAR_UNLOCK |
801 EXTENT_CLEAR_DELALLOC |
802 EXTENT_CLEAR_DIRTY |
803 EXTENT_SET_WRITEBACK |
804 EXTENT_END_WRITEBACK);
805 kfree(async_extent);
806 goto again;
807 }
808
809 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
810 u64 num_bytes)
811 {
812 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
813 struct extent_map *em;
814 u64 alloc_hint = 0;
815
816 read_lock(&em_tree->lock);
817 em = search_extent_mapping(em_tree, start, num_bytes);
818 if (em) {
819 /*
820 * if block start isn't an actual block number then find the
821 * first block in this inode and use that as a hint. If that
822 * block is also bogus then just don't worry about it.
823 */
824 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
825 free_extent_map(em);
826 em = search_extent_mapping(em_tree, 0, 0);
827 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
828 alloc_hint = em->block_start;
829 if (em)
830 free_extent_map(em);
831 } else {
832 alloc_hint = em->block_start;
833 free_extent_map(em);
834 }
835 }
836 read_unlock(&em_tree->lock);
837
838 return alloc_hint;
839 }
840
841 /*
842 * when extent_io.c finds a delayed allocation range in the file,
843 * the call backs end up in this code. The basic idea is to
844 * allocate extents on disk for the range, and create ordered data structs
845 * in ram to track those extents.
846 *
847 * locked_page is the page that writepage had locked already. We use
848 * it to make sure we don't do extra locks or unlocks.
849 *
850 * *page_started is set to one if we unlock locked_page and do everything
851 * required to start IO on it. It may be clean and already done with
852 * IO when we return.
853 */
854 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
855 struct inode *inode,
856 struct btrfs_root *root,
857 struct page *locked_page,
858 u64 start, u64 end, int *page_started,
859 unsigned long *nr_written,
860 int unlock)
861 {
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 BUG_ON(btrfs_is_free_space_inode(inode));
874
875 num_bytes = ALIGN(end - start + 1, blocksize);
876 num_bytes = max(blocksize, num_bytes);
877 disk_num_bytes = num_bytes;
878
879 /* if this is a small write inside eof, kick off defrag */
880 if (num_bytes < 64 * 1024 &&
881 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
882 btrfs_add_inode_defrag(trans, inode);
883
884 if (start == 0) {
885 /* lets try to make an inline extent */
886 ret = cow_file_range_inline(trans, root, inode,
887 start, end, 0, 0, NULL);
888 if (ret == 0) {
889 extent_clear_unlock_delalloc(inode,
890 &BTRFS_I(inode)->io_tree,
891 start, end, NULL,
892 EXTENT_CLEAR_UNLOCK_PAGE |
893 EXTENT_CLEAR_UNLOCK |
894 EXTENT_CLEAR_DELALLOC |
895 EXTENT_CLEAR_DIRTY |
896 EXTENT_SET_WRITEBACK |
897 EXTENT_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 btrfs_abort_transaction(trans, root, ret);
905 goto out_unlock;
906 }
907 }
908
909 BUG_ON(disk_num_bytes >
910 btrfs_super_total_bytes(root->fs_info->super_copy));
911
912 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
913 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
914
915 while (disk_num_bytes > 0) {
916 unsigned long op;
917
918 cur_alloc_size = disk_num_bytes;
919 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
920 root->sectorsize, 0, alloc_hint,
921 &ins, 1);
922 if (ret < 0) {
923 btrfs_abort_transaction(trans, root, ret);
924 goto out_unlock;
925 }
926
927 em = alloc_extent_map();
928 if (!em) {
929 ret = -ENOMEM;
930 goto out_reserve;
931 }
932 em->start = start;
933 em->orig_start = em->start;
934 ram_size = ins.offset;
935 em->len = ins.offset;
936 em->mod_start = em->start;
937 em->mod_len = em->len;
938
939 em->block_start = ins.objectid;
940 em->block_len = ins.offset;
941 em->orig_block_len = ins.offset;
942 em->ram_bytes = ram_size;
943 em->bdev = root->fs_info->fs_devices->latest_bdev;
944 set_bit(EXTENT_FLAG_PINNED, &em->flags);
945 em->generation = -1;
946
947 while (1) {
948 write_lock(&em_tree->lock);
949 ret = add_extent_mapping(em_tree, em, 1);
950 write_unlock(&em_tree->lock);
951 if (ret != -EEXIST) {
952 free_extent_map(em);
953 break;
954 }
955 btrfs_drop_extent_cache(inode, start,
956 start + ram_size - 1, 0);
957 }
958 if (ret)
959 goto out_reserve;
960
961 cur_alloc_size = ins.offset;
962 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
963 ram_size, cur_alloc_size, 0);
964 if (ret)
965 goto out_reserve;
966
967 if (root->root_key.objectid ==
968 BTRFS_DATA_RELOC_TREE_OBJECTID) {
969 ret = btrfs_reloc_clone_csums(inode, start,
970 cur_alloc_size);
971 if (ret) {
972 btrfs_abort_transaction(trans, root, ret);
973 goto out_reserve;
974 }
975 }
976
977 if (disk_num_bytes < cur_alloc_size)
978 break;
979
980 /* we're not doing compressed IO, don't unlock the first
981 * page (which the caller expects to stay locked), don't
982 * clear any dirty bits and don't set any writeback bits
983 *
984 * Do set the Private2 bit so we know this page was properly
985 * setup for writepage
986 */
987 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
988 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
989 EXTENT_SET_PRIVATE2;
990
991 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
992 start, start + ram_size - 1,
993 locked_page, op);
994 disk_num_bytes -= cur_alloc_size;
995 num_bytes -= cur_alloc_size;
996 alloc_hint = ins.objectid + ins.offset;
997 start += cur_alloc_size;
998 }
999 out:
1000 return ret;
1001
1002 out_reserve:
1003 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1004 out_unlock:
1005 extent_clear_unlock_delalloc(inode,
1006 &BTRFS_I(inode)->io_tree,
1007 start, end, locked_page,
1008 EXTENT_CLEAR_UNLOCK_PAGE |
1009 EXTENT_CLEAR_UNLOCK |
1010 EXTENT_CLEAR_DELALLOC |
1011 EXTENT_CLEAR_DIRTY |
1012 EXTENT_SET_WRITEBACK |
1013 EXTENT_END_WRITEBACK);
1014
1015 goto out;
1016 }
1017
1018 static noinline int cow_file_range(struct inode *inode,
1019 struct page *locked_page,
1020 u64 start, u64 end, int *page_started,
1021 unsigned long *nr_written,
1022 int unlock)
1023 {
1024 struct btrfs_trans_handle *trans;
1025 struct btrfs_root *root = BTRFS_I(inode)->root;
1026 int ret;
1027
1028 trans = btrfs_join_transaction(root);
1029 if (IS_ERR(trans)) {
1030 extent_clear_unlock_delalloc(inode,
1031 &BTRFS_I(inode)->io_tree,
1032 start, end, locked_page,
1033 EXTENT_CLEAR_UNLOCK_PAGE |
1034 EXTENT_CLEAR_UNLOCK |
1035 EXTENT_CLEAR_DELALLOC |
1036 EXTENT_CLEAR_DIRTY |
1037 EXTENT_SET_WRITEBACK |
1038 EXTENT_END_WRITEBACK);
1039 return PTR_ERR(trans);
1040 }
1041 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1042
1043 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1044 page_started, nr_written, unlock);
1045
1046 btrfs_end_transaction(trans, root);
1047
1048 return ret;
1049 }
1050
1051 /*
1052 * work queue call back to started compression on a file and pages
1053 */
1054 static noinline void async_cow_start(struct btrfs_work *work)
1055 {
1056 struct async_cow *async_cow;
1057 int num_added = 0;
1058 async_cow = container_of(work, struct async_cow, work);
1059
1060 compress_file_range(async_cow->inode, async_cow->locked_page,
1061 async_cow->start, async_cow->end, async_cow,
1062 &num_added);
1063 if (num_added == 0) {
1064 btrfs_add_delayed_iput(async_cow->inode);
1065 async_cow->inode = NULL;
1066 }
1067 }
1068
1069 /*
1070 * work queue call back to submit previously compressed pages
1071 */
1072 static noinline void async_cow_submit(struct btrfs_work *work)
1073 {
1074 struct async_cow *async_cow;
1075 struct btrfs_root *root;
1076 unsigned long nr_pages;
1077
1078 async_cow = container_of(work, struct async_cow, work);
1079
1080 root = async_cow->root;
1081 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1082 PAGE_CACHE_SHIFT;
1083
1084 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1085 5 * 1024 * 1024 &&
1086 waitqueue_active(&root->fs_info->async_submit_wait))
1087 wake_up(&root->fs_info->async_submit_wait);
1088
1089 if (async_cow->inode)
1090 submit_compressed_extents(async_cow->inode, async_cow);
1091 }
1092
1093 static noinline void async_cow_free(struct btrfs_work *work)
1094 {
1095 struct async_cow *async_cow;
1096 async_cow = container_of(work, struct async_cow, work);
1097 if (async_cow->inode)
1098 btrfs_add_delayed_iput(async_cow->inode);
1099 kfree(async_cow);
1100 }
1101
1102 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1103 u64 start, u64 end, int *page_started,
1104 unsigned long *nr_written)
1105 {
1106 struct async_cow *async_cow;
1107 struct btrfs_root *root = BTRFS_I(inode)->root;
1108 unsigned long nr_pages;
1109 u64 cur_end;
1110 int limit = 10 * 1024 * 1024;
1111
1112 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1113 1, 0, NULL, GFP_NOFS);
1114 while (start < end) {
1115 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1116 BUG_ON(!async_cow); /* -ENOMEM */
1117 async_cow->inode = igrab(inode);
1118 async_cow->root = root;
1119 async_cow->locked_page = locked_page;
1120 async_cow->start = start;
1121
1122 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1123 cur_end = end;
1124 else
1125 cur_end = min(end, start + 512 * 1024 - 1);
1126
1127 async_cow->end = cur_end;
1128 INIT_LIST_HEAD(&async_cow->extents);
1129
1130 async_cow->work.func = async_cow_start;
1131 async_cow->work.ordered_func = async_cow_submit;
1132 async_cow->work.ordered_free = async_cow_free;
1133 async_cow->work.flags = 0;
1134
1135 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1136 PAGE_CACHE_SHIFT;
1137 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1138
1139 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1140 &async_cow->work);
1141
1142 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1143 wait_event(root->fs_info->async_submit_wait,
1144 (atomic_read(&root->fs_info->async_delalloc_pages) <
1145 limit));
1146 }
1147
1148 while (atomic_read(&root->fs_info->async_submit_draining) &&
1149 atomic_read(&root->fs_info->async_delalloc_pages)) {
1150 wait_event(root->fs_info->async_submit_wait,
1151 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1152 0));
1153 }
1154
1155 *nr_written += nr_pages;
1156 start = cur_end + 1;
1157 }
1158 *page_started = 1;
1159 return 0;
1160 }
1161
1162 static noinline int csum_exist_in_range(struct btrfs_root *root,
1163 u64 bytenr, u64 num_bytes)
1164 {
1165 int ret;
1166 struct btrfs_ordered_sum *sums;
1167 LIST_HEAD(list);
1168
1169 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1170 bytenr + num_bytes - 1, &list, 0);
1171 if (ret == 0 && list_empty(&list))
1172 return 0;
1173
1174 while (!list_empty(&list)) {
1175 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1176 list_del(&sums->list);
1177 kfree(sums);
1178 }
1179 return 1;
1180 }
1181
1182 /*
1183 * when nowcow writeback call back. This checks for snapshots or COW copies
1184 * of the extents that exist in the file, and COWs the file as required.
1185 *
1186 * If no cow copies or snapshots exist, we write directly to the existing
1187 * blocks on disk
1188 */
1189 static noinline int run_delalloc_nocow(struct inode *inode,
1190 struct page *locked_page,
1191 u64 start, u64 end, int *page_started, int force,
1192 unsigned long *nr_written)
1193 {
1194 struct btrfs_root *root = BTRFS_I(inode)->root;
1195 struct btrfs_trans_handle *trans;
1196 struct extent_buffer *leaf;
1197 struct btrfs_path *path;
1198 struct btrfs_file_extent_item *fi;
1199 struct btrfs_key found_key;
1200 u64 cow_start;
1201 u64 cur_offset;
1202 u64 extent_end;
1203 u64 extent_offset;
1204 u64 disk_bytenr;
1205 u64 num_bytes;
1206 u64 disk_num_bytes;
1207 u64 ram_bytes;
1208 int extent_type;
1209 int ret, err;
1210 int type;
1211 int nocow;
1212 int check_prev = 1;
1213 bool nolock;
1214 u64 ino = btrfs_ino(inode);
1215
1216 path = btrfs_alloc_path();
1217 if (!path) {
1218 extent_clear_unlock_delalloc(inode,
1219 &BTRFS_I(inode)->io_tree,
1220 start, end, locked_page,
1221 EXTENT_CLEAR_UNLOCK_PAGE |
1222 EXTENT_CLEAR_UNLOCK |
1223 EXTENT_CLEAR_DELALLOC |
1224 EXTENT_CLEAR_DIRTY |
1225 EXTENT_SET_WRITEBACK |
1226 EXTENT_END_WRITEBACK);
1227 return -ENOMEM;
1228 }
1229
1230 nolock = btrfs_is_free_space_inode(inode);
1231
1232 if (nolock)
1233 trans = btrfs_join_transaction_nolock(root);
1234 else
1235 trans = btrfs_join_transaction(root);
1236
1237 if (IS_ERR(trans)) {
1238 extent_clear_unlock_delalloc(inode,
1239 &BTRFS_I(inode)->io_tree,
1240 start, end, locked_page,
1241 EXTENT_CLEAR_UNLOCK_PAGE |
1242 EXTENT_CLEAR_UNLOCK |
1243 EXTENT_CLEAR_DELALLOC |
1244 EXTENT_CLEAR_DIRTY |
1245 EXTENT_SET_WRITEBACK |
1246 EXTENT_END_WRITEBACK);
1247 btrfs_free_path(path);
1248 return PTR_ERR(trans);
1249 }
1250
1251 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1252
1253 cow_start = (u64)-1;
1254 cur_offset = start;
1255 while (1) {
1256 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1257 cur_offset, 0);
1258 if (ret < 0) {
1259 btrfs_abort_transaction(trans, root, ret);
1260 goto error;
1261 }
1262 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1263 leaf = path->nodes[0];
1264 btrfs_item_key_to_cpu(leaf, &found_key,
1265 path->slots[0] - 1);
1266 if (found_key.objectid == ino &&
1267 found_key.type == BTRFS_EXTENT_DATA_KEY)
1268 path->slots[0]--;
1269 }
1270 check_prev = 0;
1271 next_slot:
1272 leaf = path->nodes[0];
1273 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1274 ret = btrfs_next_leaf(root, path);
1275 if (ret < 0) {
1276 btrfs_abort_transaction(trans, root, ret);
1277 goto error;
1278 }
1279 if (ret > 0)
1280 break;
1281 leaf = path->nodes[0];
1282 }
1283
1284 nocow = 0;
1285 disk_bytenr = 0;
1286 num_bytes = 0;
1287 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1288
1289 if (found_key.objectid > ino ||
1290 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1291 found_key.offset > end)
1292 break;
1293
1294 if (found_key.offset > cur_offset) {
1295 extent_end = found_key.offset;
1296 extent_type = 0;
1297 goto out_check;
1298 }
1299
1300 fi = btrfs_item_ptr(leaf, path->slots[0],
1301 struct btrfs_file_extent_item);
1302 extent_type = btrfs_file_extent_type(leaf, fi);
1303
1304 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1305 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1306 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1307 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1308 extent_offset = btrfs_file_extent_offset(leaf, fi);
1309 extent_end = found_key.offset +
1310 btrfs_file_extent_num_bytes(leaf, fi);
1311 disk_num_bytes =
1312 btrfs_file_extent_disk_num_bytes(leaf, fi);
1313 if (extent_end <= start) {
1314 path->slots[0]++;
1315 goto next_slot;
1316 }
1317 if (disk_bytenr == 0)
1318 goto out_check;
1319 if (btrfs_file_extent_compression(leaf, fi) ||
1320 btrfs_file_extent_encryption(leaf, fi) ||
1321 btrfs_file_extent_other_encoding(leaf, fi))
1322 goto out_check;
1323 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1324 goto out_check;
1325 if (btrfs_extent_readonly(root, disk_bytenr))
1326 goto out_check;
1327 if (btrfs_cross_ref_exist(trans, root, ino,
1328 found_key.offset -
1329 extent_offset, disk_bytenr))
1330 goto out_check;
1331 disk_bytenr += extent_offset;
1332 disk_bytenr += cur_offset - found_key.offset;
1333 num_bytes = min(end + 1, extent_end) - cur_offset;
1334 /*
1335 * force cow if csum exists in the range.
1336 * this ensure that csum for a given extent are
1337 * either valid or do not exist.
1338 */
1339 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1340 goto out_check;
1341 nocow = 1;
1342 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1343 extent_end = found_key.offset +
1344 btrfs_file_extent_inline_len(leaf, fi);
1345 extent_end = ALIGN(extent_end, root->sectorsize);
1346 } else {
1347 BUG_ON(1);
1348 }
1349 out_check:
1350 if (extent_end <= start) {
1351 path->slots[0]++;
1352 goto next_slot;
1353 }
1354 if (!nocow) {
1355 if (cow_start == (u64)-1)
1356 cow_start = cur_offset;
1357 cur_offset = extent_end;
1358 if (cur_offset > end)
1359 break;
1360 path->slots[0]++;
1361 goto next_slot;
1362 }
1363
1364 btrfs_release_path(path);
1365 if (cow_start != (u64)-1) {
1366 ret = __cow_file_range(trans, inode, root, locked_page,
1367 cow_start, found_key.offset - 1,
1368 page_started, nr_written, 1);
1369 if (ret) {
1370 btrfs_abort_transaction(trans, root, ret);
1371 goto error;
1372 }
1373 cow_start = (u64)-1;
1374 }
1375
1376 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1377 struct extent_map *em;
1378 struct extent_map_tree *em_tree;
1379 em_tree = &BTRFS_I(inode)->extent_tree;
1380 em = alloc_extent_map();
1381 BUG_ON(!em); /* -ENOMEM */
1382 em->start = cur_offset;
1383 em->orig_start = found_key.offset - extent_offset;
1384 em->len = num_bytes;
1385 em->block_len = num_bytes;
1386 em->block_start = disk_bytenr;
1387 em->orig_block_len = disk_num_bytes;
1388 em->ram_bytes = ram_bytes;
1389 em->bdev = root->fs_info->fs_devices->latest_bdev;
1390 em->mod_start = em->start;
1391 em->mod_len = em->len;
1392 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1393 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1394 em->generation = -1;
1395 while (1) {
1396 write_lock(&em_tree->lock);
1397 ret = add_extent_mapping(em_tree, em, 1);
1398 write_unlock(&em_tree->lock);
1399 if (ret != -EEXIST) {
1400 free_extent_map(em);
1401 break;
1402 }
1403 btrfs_drop_extent_cache(inode, em->start,
1404 em->start + em->len - 1, 0);
1405 }
1406 type = BTRFS_ORDERED_PREALLOC;
1407 } else {
1408 type = BTRFS_ORDERED_NOCOW;
1409 }
1410
1411 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1412 num_bytes, num_bytes, type);
1413 BUG_ON(ret); /* -ENOMEM */
1414
1415 if (root->root_key.objectid ==
1416 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1417 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1418 num_bytes);
1419 if (ret) {
1420 btrfs_abort_transaction(trans, root, ret);
1421 goto error;
1422 }
1423 }
1424
1425 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1426 cur_offset, cur_offset + num_bytes - 1,
1427 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1428 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1429 EXTENT_SET_PRIVATE2);
1430 cur_offset = extent_end;
1431 if (cur_offset > end)
1432 break;
1433 }
1434 btrfs_release_path(path);
1435
1436 if (cur_offset <= end && cow_start == (u64)-1) {
1437 cow_start = cur_offset;
1438 cur_offset = end;
1439 }
1440
1441 if (cow_start != (u64)-1) {
1442 ret = __cow_file_range(trans, inode, root, locked_page,
1443 cow_start, end,
1444 page_started, nr_written, 1);
1445 if (ret) {
1446 btrfs_abort_transaction(trans, root, ret);
1447 goto error;
1448 }
1449 }
1450
1451 error:
1452 err = btrfs_end_transaction(trans, root);
1453 if (!ret)
1454 ret = err;
1455
1456 if (ret && cur_offset < end)
1457 extent_clear_unlock_delalloc(inode,
1458 &BTRFS_I(inode)->io_tree,
1459 cur_offset, end, locked_page,
1460 EXTENT_CLEAR_UNLOCK_PAGE |
1461 EXTENT_CLEAR_UNLOCK |
1462 EXTENT_CLEAR_DELALLOC |
1463 EXTENT_CLEAR_DIRTY |
1464 EXTENT_SET_WRITEBACK |
1465 EXTENT_END_WRITEBACK);
1466
1467 btrfs_free_path(path);
1468 return ret;
1469 }
1470
1471 /*
1472 * extent_io.c call back to do delayed allocation processing
1473 */
1474 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1475 u64 start, u64 end, int *page_started,
1476 unsigned long *nr_written)
1477 {
1478 int ret;
1479 struct btrfs_root *root = BTRFS_I(inode)->root;
1480
1481 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1482 ret = run_delalloc_nocow(inode, locked_page, start, end,
1483 page_started, 1, nr_written);
1484 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1485 ret = run_delalloc_nocow(inode, locked_page, start, end,
1486 page_started, 0, nr_written);
1487 } else if (!btrfs_test_opt(root, COMPRESS) &&
1488 !(BTRFS_I(inode)->force_compress) &&
1489 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1490 ret = cow_file_range(inode, locked_page, start, end,
1491 page_started, nr_written, 1);
1492 } else {
1493 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1494 &BTRFS_I(inode)->runtime_flags);
1495 ret = cow_file_range_async(inode, locked_page, start, end,
1496 page_started, nr_written);
1497 }
1498 return ret;
1499 }
1500
1501 static void btrfs_split_extent_hook(struct inode *inode,
1502 struct extent_state *orig, u64 split)
1503 {
1504 /* not delalloc, ignore it */
1505 if (!(orig->state & EXTENT_DELALLOC))
1506 return;
1507
1508 spin_lock(&BTRFS_I(inode)->lock);
1509 BTRFS_I(inode)->outstanding_extents++;
1510 spin_unlock(&BTRFS_I(inode)->lock);
1511 }
1512
1513 /*
1514 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1515 * extents so we can keep track of new extents that are just merged onto old
1516 * extents, such as when we are doing sequential writes, so we can properly
1517 * account for the metadata space we'll need.
1518 */
1519 static void btrfs_merge_extent_hook(struct inode *inode,
1520 struct extent_state *new,
1521 struct extent_state *other)
1522 {
1523 /* not delalloc, ignore it */
1524 if (!(other->state & EXTENT_DELALLOC))
1525 return;
1526
1527 spin_lock(&BTRFS_I(inode)->lock);
1528 BTRFS_I(inode)->outstanding_extents--;
1529 spin_unlock(&BTRFS_I(inode)->lock);
1530 }
1531
1532 /*
1533 * extent_io.c set_bit_hook, used to track delayed allocation
1534 * bytes in this file, and to maintain the list of inodes that
1535 * have pending delalloc work to be done.
1536 */
1537 static void btrfs_set_bit_hook(struct inode *inode,
1538 struct extent_state *state, unsigned long *bits)
1539 {
1540
1541 /*
1542 * set_bit and clear bit hooks normally require _irqsave/restore
1543 * but in this case, we are only testing for the DELALLOC
1544 * bit, which is only set or cleared with irqs on
1545 */
1546 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1547 struct btrfs_root *root = BTRFS_I(inode)->root;
1548 u64 len = state->end + 1 - state->start;
1549 bool do_list = !btrfs_is_free_space_inode(inode);
1550
1551 if (*bits & EXTENT_FIRST_DELALLOC) {
1552 *bits &= ~EXTENT_FIRST_DELALLOC;
1553 } else {
1554 spin_lock(&BTRFS_I(inode)->lock);
1555 BTRFS_I(inode)->outstanding_extents++;
1556 spin_unlock(&BTRFS_I(inode)->lock);
1557 }
1558
1559 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1560 root->fs_info->delalloc_batch);
1561 spin_lock(&BTRFS_I(inode)->lock);
1562 BTRFS_I(inode)->delalloc_bytes += len;
1563 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1564 &BTRFS_I(inode)->runtime_flags)) {
1565 spin_lock(&root->fs_info->delalloc_lock);
1566 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1567 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1568 &root->fs_info->delalloc_inodes);
1569 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1570 &BTRFS_I(inode)->runtime_flags);
1571 }
1572 spin_unlock(&root->fs_info->delalloc_lock);
1573 }
1574 spin_unlock(&BTRFS_I(inode)->lock);
1575 }
1576 }
1577
1578 /*
1579 * extent_io.c clear_bit_hook, see set_bit_hook for why
1580 */
1581 static void btrfs_clear_bit_hook(struct inode *inode,
1582 struct extent_state *state,
1583 unsigned long *bits)
1584 {
1585 /*
1586 * set_bit and clear bit hooks normally require _irqsave/restore
1587 * but in this case, we are only testing for the DELALLOC
1588 * bit, which is only set or cleared with irqs on
1589 */
1590 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1591 struct btrfs_root *root = BTRFS_I(inode)->root;
1592 u64 len = state->end + 1 - state->start;
1593 bool do_list = !btrfs_is_free_space_inode(inode);
1594
1595 if (*bits & EXTENT_FIRST_DELALLOC) {
1596 *bits &= ~EXTENT_FIRST_DELALLOC;
1597 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1598 spin_lock(&BTRFS_I(inode)->lock);
1599 BTRFS_I(inode)->outstanding_extents--;
1600 spin_unlock(&BTRFS_I(inode)->lock);
1601 }
1602
1603 if (*bits & EXTENT_DO_ACCOUNTING)
1604 btrfs_delalloc_release_metadata(inode, len);
1605
1606 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1607 && do_list)
1608 btrfs_free_reserved_data_space(inode, len);
1609
1610 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1611 root->fs_info->delalloc_batch);
1612 spin_lock(&BTRFS_I(inode)->lock);
1613 BTRFS_I(inode)->delalloc_bytes -= len;
1614 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1615 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1616 &BTRFS_I(inode)->runtime_flags)) {
1617 spin_lock(&root->fs_info->delalloc_lock);
1618 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1619 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1620 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1621 &BTRFS_I(inode)->runtime_flags);
1622 }
1623 spin_unlock(&root->fs_info->delalloc_lock);
1624 }
1625 spin_unlock(&BTRFS_I(inode)->lock);
1626 }
1627 }
1628
1629 /*
1630 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1631 * we don't create bios that span stripes or chunks
1632 */
1633 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1634 size_t size, struct bio *bio,
1635 unsigned long bio_flags)
1636 {
1637 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1638 u64 logical = (u64)bio->bi_sector << 9;
1639 u64 length = 0;
1640 u64 map_length;
1641 int ret;
1642
1643 if (bio_flags & EXTENT_BIO_COMPRESSED)
1644 return 0;
1645
1646 length = bio->bi_size;
1647 map_length = length;
1648 ret = btrfs_map_block(root->fs_info, rw, logical,
1649 &map_length, NULL, 0);
1650 /* Will always return 0 with map_multi == NULL */
1651 BUG_ON(ret < 0);
1652 if (map_length < length + size)
1653 return 1;
1654 return 0;
1655 }
1656
1657 /*
1658 * in order to insert checksums into the metadata in large chunks,
1659 * we wait until bio submission time. All the pages in the bio are
1660 * checksummed and sums are attached onto the ordered extent record.
1661 *
1662 * At IO completion time the cums attached on the ordered extent record
1663 * are inserted into the btree
1664 */
1665 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1666 struct bio *bio, int mirror_num,
1667 unsigned long bio_flags,
1668 u64 bio_offset)
1669 {
1670 struct btrfs_root *root = BTRFS_I(inode)->root;
1671 int ret = 0;
1672
1673 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1674 BUG_ON(ret); /* -ENOMEM */
1675 return 0;
1676 }
1677
1678 /*
1679 * in order to insert checksums into the metadata in large chunks,
1680 * we wait until bio submission time. All the pages in the bio are
1681 * checksummed and sums are attached onto the ordered extent record.
1682 *
1683 * At IO completion time the cums attached on the ordered extent record
1684 * are inserted into the btree
1685 */
1686 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1687 int mirror_num, unsigned long bio_flags,
1688 u64 bio_offset)
1689 {
1690 struct btrfs_root *root = BTRFS_I(inode)->root;
1691 int ret;
1692
1693 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1694 if (ret)
1695 bio_endio(bio, ret);
1696 return ret;
1697 }
1698
1699 /*
1700 * extent_io.c submission hook. This does the right thing for csum calculation
1701 * on write, or reading the csums from the tree before a read
1702 */
1703 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1704 int mirror_num, unsigned long bio_flags,
1705 u64 bio_offset)
1706 {
1707 struct btrfs_root *root = BTRFS_I(inode)->root;
1708 int ret = 0;
1709 int skip_sum;
1710 int metadata = 0;
1711 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1712
1713 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1714
1715 if (btrfs_is_free_space_inode(inode))
1716 metadata = 2;
1717
1718 if (!(rw & REQ_WRITE)) {
1719 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1720 if (ret)
1721 goto out;
1722
1723 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1724 ret = btrfs_submit_compressed_read(inode, bio,
1725 mirror_num,
1726 bio_flags);
1727 goto out;
1728 } else if (!skip_sum) {
1729 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1730 if (ret)
1731 goto out;
1732 }
1733 goto mapit;
1734 } else if (async && !skip_sum) {
1735 /* csum items have already been cloned */
1736 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1737 goto mapit;
1738 /* we're doing a write, do the async checksumming */
1739 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1740 inode, rw, bio, mirror_num,
1741 bio_flags, bio_offset,
1742 __btrfs_submit_bio_start,
1743 __btrfs_submit_bio_done);
1744 goto out;
1745 } else if (!skip_sum) {
1746 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1747 if (ret)
1748 goto out;
1749 }
1750
1751 mapit:
1752 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1753
1754 out:
1755 if (ret < 0)
1756 bio_endio(bio, ret);
1757 return ret;
1758 }
1759
1760 /*
1761 * given a list of ordered sums record them in the inode. This happens
1762 * at IO completion time based on sums calculated at bio submission time.
1763 */
1764 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1765 struct inode *inode, u64 file_offset,
1766 struct list_head *list)
1767 {
1768 struct btrfs_ordered_sum *sum;
1769
1770 list_for_each_entry(sum, list, list) {
1771 trans->adding_csums = 1;
1772 btrfs_csum_file_blocks(trans,
1773 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1774 trans->adding_csums = 0;
1775 }
1776 return 0;
1777 }
1778
1779 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1780 struct extent_state **cached_state)
1781 {
1782 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1783 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1784 cached_state, GFP_NOFS);
1785 }
1786
1787 /* see btrfs_writepage_start_hook for details on why this is required */
1788 struct btrfs_writepage_fixup {
1789 struct page *page;
1790 struct btrfs_work work;
1791 };
1792
1793 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1794 {
1795 struct btrfs_writepage_fixup *fixup;
1796 struct btrfs_ordered_extent *ordered;
1797 struct extent_state *cached_state = NULL;
1798 struct page *page;
1799 struct inode *inode;
1800 u64 page_start;
1801 u64 page_end;
1802 int ret;
1803
1804 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1805 page = fixup->page;
1806 again:
1807 lock_page(page);
1808 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1809 ClearPageChecked(page);
1810 goto out_page;
1811 }
1812
1813 inode = page->mapping->host;
1814 page_start = page_offset(page);
1815 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1816
1817 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1818 &cached_state);
1819
1820 /* already ordered? We're done */
1821 if (PagePrivate2(page))
1822 goto out;
1823
1824 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1825 if (ordered) {
1826 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1827 page_end, &cached_state, GFP_NOFS);
1828 unlock_page(page);
1829 btrfs_start_ordered_extent(inode, ordered, 1);
1830 btrfs_put_ordered_extent(ordered);
1831 goto again;
1832 }
1833
1834 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1835 if (ret) {
1836 mapping_set_error(page->mapping, ret);
1837 end_extent_writepage(page, ret, page_start, page_end);
1838 ClearPageChecked(page);
1839 goto out;
1840 }
1841
1842 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1843 ClearPageChecked(page);
1844 set_page_dirty(page);
1845 out:
1846 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1847 &cached_state, GFP_NOFS);
1848 out_page:
1849 unlock_page(page);
1850 page_cache_release(page);
1851 kfree(fixup);
1852 }
1853
1854 /*
1855 * There are a few paths in the higher layers of the kernel that directly
1856 * set the page dirty bit without asking the filesystem if it is a
1857 * good idea. This causes problems because we want to make sure COW
1858 * properly happens and the data=ordered rules are followed.
1859 *
1860 * In our case any range that doesn't have the ORDERED bit set
1861 * hasn't been properly setup for IO. We kick off an async process
1862 * to fix it up. The async helper will wait for ordered extents, set
1863 * the delalloc bit and make it safe to write the page.
1864 */
1865 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1866 {
1867 struct inode *inode = page->mapping->host;
1868 struct btrfs_writepage_fixup *fixup;
1869 struct btrfs_root *root = BTRFS_I(inode)->root;
1870
1871 /* this page is properly in the ordered list */
1872 if (TestClearPagePrivate2(page))
1873 return 0;
1874
1875 if (PageChecked(page))
1876 return -EAGAIN;
1877
1878 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1879 if (!fixup)
1880 return -EAGAIN;
1881
1882 SetPageChecked(page);
1883 page_cache_get(page);
1884 fixup->work.func = btrfs_writepage_fixup_worker;
1885 fixup->page = page;
1886 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1887 return -EBUSY;
1888 }
1889
1890 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1891 struct inode *inode, u64 file_pos,
1892 u64 disk_bytenr, u64 disk_num_bytes,
1893 u64 num_bytes, u64 ram_bytes,
1894 u8 compression, u8 encryption,
1895 u16 other_encoding, int extent_type)
1896 {
1897 struct btrfs_root *root = BTRFS_I(inode)->root;
1898 struct btrfs_file_extent_item *fi;
1899 struct btrfs_path *path;
1900 struct extent_buffer *leaf;
1901 struct btrfs_key ins;
1902 int ret;
1903
1904 path = btrfs_alloc_path();
1905 if (!path)
1906 return -ENOMEM;
1907
1908 path->leave_spinning = 1;
1909
1910 /*
1911 * we may be replacing one extent in the tree with another.
1912 * The new extent is pinned in the extent map, and we don't want
1913 * to drop it from the cache until it is completely in the btree.
1914 *
1915 * So, tell btrfs_drop_extents to leave this extent in the cache.
1916 * the caller is expected to unpin it and allow it to be merged
1917 * with the others.
1918 */
1919 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1920 file_pos + num_bytes, 0);
1921 if (ret)
1922 goto out;
1923
1924 ins.objectid = btrfs_ino(inode);
1925 ins.offset = file_pos;
1926 ins.type = BTRFS_EXTENT_DATA_KEY;
1927 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1928 if (ret)
1929 goto out;
1930 leaf = path->nodes[0];
1931 fi = btrfs_item_ptr(leaf, path->slots[0],
1932 struct btrfs_file_extent_item);
1933 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1934 btrfs_set_file_extent_type(leaf, fi, extent_type);
1935 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1936 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1937 btrfs_set_file_extent_offset(leaf, fi, 0);
1938 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1939 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1940 btrfs_set_file_extent_compression(leaf, fi, compression);
1941 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1942 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1943
1944 btrfs_mark_buffer_dirty(leaf);
1945 btrfs_release_path(path);
1946
1947 inode_add_bytes(inode, num_bytes);
1948
1949 ins.objectid = disk_bytenr;
1950 ins.offset = disk_num_bytes;
1951 ins.type = BTRFS_EXTENT_ITEM_KEY;
1952 ret = btrfs_alloc_reserved_file_extent(trans, root,
1953 root->root_key.objectid,
1954 btrfs_ino(inode), file_pos, &ins);
1955 out:
1956 btrfs_free_path(path);
1957
1958 return ret;
1959 }
1960
1961 /* snapshot-aware defrag */
1962 struct sa_defrag_extent_backref {
1963 struct rb_node node;
1964 struct old_sa_defrag_extent *old;
1965 u64 root_id;
1966 u64 inum;
1967 u64 file_pos;
1968 u64 extent_offset;
1969 u64 num_bytes;
1970 u64 generation;
1971 };
1972
1973 struct old_sa_defrag_extent {
1974 struct list_head list;
1975 struct new_sa_defrag_extent *new;
1976
1977 u64 extent_offset;
1978 u64 bytenr;
1979 u64 offset;
1980 u64 len;
1981 int count;
1982 };
1983
1984 struct new_sa_defrag_extent {
1985 struct rb_root root;
1986 struct list_head head;
1987 struct btrfs_path *path;
1988 struct inode *inode;
1989 u64 file_pos;
1990 u64 len;
1991 u64 bytenr;
1992 u64 disk_len;
1993 u8 compress_type;
1994 };
1995
1996 static int backref_comp(struct sa_defrag_extent_backref *b1,
1997 struct sa_defrag_extent_backref *b2)
1998 {
1999 if (b1->root_id < b2->root_id)
2000 return -1;
2001 else if (b1->root_id > b2->root_id)
2002 return 1;
2003
2004 if (b1->inum < b2->inum)
2005 return -1;
2006 else if (b1->inum > b2->inum)
2007 return 1;
2008
2009 if (b1->file_pos < b2->file_pos)
2010 return -1;
2011 else if (b1->file_pos > b2->file_pos)
2012 return 1;
2013
2014 /*
2015 * [------------------------------] ===> (a range of space)
2016 * |<--->| |<---->| =============> (fs/file tree A)
2017 * |<---------------------------->| ===> (fs/file tree B)
2018 *
2019 * A range of space can refer to two file extents in one tree while
2020 * refer to only one file extent in another tree.
2021 *
2022 * So we may process a disk offset more than one time(two extents in A)
2023 * and locate at the same extent(one extent in B), then insert two same
2024 * backrefs(both refer to the extent in B).
2025 */
2026 return 0;
2027 }
2028
2029 static void backref_insert(struct rb_root *root,
2030 struct sa_defrag_extent_backref *backref)
2031 {
2032 struct rb_node **p = &root->rb_node;
2033 struct rb_node *parent = NULL;
2034 struct sa_defrag_extent_backref *entry;
2035 int ret;
2036
2037 while (*p) {
2038 parent = *p;
2039 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2040
2041 ret = backref_comp(backref, entry);
2042 if (ret < 0)
2043 p = &(*p)->rb_left;
2044 else
2045 p = &(*p)->rb_right;
2046 }
2047
2048 rb_link_node(&backref->node, parent, p);
2049 rb_insert_color(&backref->node, root);
2050 }
2051
2052 /*
2053 * Note the backref might has changed, and in this case we just return 0.
2054 */
2055 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2056 void *ctx)
2057 {
2058 struct btrfs_file_extent_item *extent;
2059 struct btrfs_fs_info *fs_info;
2060 struct old_sa_defrag_extent *old = ctx;
2061 struct new_sa_defrag_extent *new = old->new;
2062 struct btrfs_path *path = new->path;
2063 struct btrfs_key key;
2064 struct btrfs_root *root;
2065 struct sa_defrag_extent_backref *backref;
2066 struct extent_buffer *leaf;
2067 struct inode *inode = new->inode;
2068 int slot;
2069 int ret;
2070 u64 extent_offset;
2071 u64 num_bytes;
2072
2073 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2074 inum == btrfs_ino(inode))
2075 return 0;
2076
2077 key.objectid = root_id;
2078 key.type = BTRFS_ROOT_ITEM_KEY;
2079 key.offset = (u64)-1;
2080
2081 fs_info = BTRFS_I(inode)->root->fs_info;
2082 root = btrfs_read_fs_root_no_name(fs_info, &key);
2083 if (IS_ERR(root)) {
2084 if (PTR_ERR(root) == -ENOENT)
2085 return 0;
2086 WARN_ON(1);
2087 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2088 inum, offset, root_id);
2089 return PTR_ERR(root);
2090 }
2091
2092 key.objectid = inum;
2093 key.type = BTRFS_EXTENT_DATA_KEY;
2094 if (offset > (u64)-1 << 32)
2095 key.offset = 0;
2096 else
2097 key.offset = offset;
2098
2099 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2100 if (ret < 0) {
2101 WARN_ON(1);
2102 return ret;
2103 }
2104
2105 while (1) {
2106 cond_resched();
2107
2108 leaf = path->nodes[0];
2109 slot = path->slots[0];
2110
2111 if (slot >= btrfs_header_nritems(leaf)) {
2112 ret = btrfs_next_leaf(root, path);
2113 if (ret < 0) {
2114 goto out;
2115 } else if (ret > 0) {
2116 ret = 0;
2117 goto out;
2118 }
2119 continue;
2120 }
2121
2122 path->slots[0]++;
2123
2124 btrfs_item_key_to_cpu(leaf, &key, slot);
2125
2126 if (key.objectid > inum)
2127 goto out;
2128
2129 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2130 continue;
2131
2132 extent = btrfs_item_ptr(leaf, slot,
2133 struct btrfs_file_extent_item);
2134
2135 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2136 continue;
2137
2138 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139 if (key.offset - extent_offset != offset)
2140 continue;
2141
2142 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2143 if (extent_offset >= old->extent_offset + old->offset +
2144 old->len || extent_offset + num_bytes <=
2145 old->extent_offset + old->offset)
2146 continue;
2147
2148 break;
2149 }
2150
2151 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2152 if (!backref) {
2153 ret = -ENOENT;
2154 goto out;
2155 }
2156
2157 backref->root_id = root_id;
2158 backref->inum = inum;
2159 backref->file_pos = offset + extent_offset;
2160 backref->num_bytes = num_bytes;
2161 backref->extent_offset = extent_offset;
2162 backref->generation = btrfs_file_extent_generation(leaf, extent);
2163 backref->old = old;
2164 backref_insert(&new->root, backref);
2165 old->count++;
2166 out:
2167 btrfs_release_path(path);
2168 WARN_ON(ret);
2169 return ret;
2170 }
2171
2172 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2173 struct new_sa_defrag_extent *new)
2174 {
2175 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2176 struct old_sa_defrag_extent *old, *tmp;
2177 int ret;
2178
2179 new->path = path;
2180
2181 list_for_each_entry_safe(old, tmp, &new->head, list) {
2182 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2183 path, record_one_backref,
2184 old);
2185 BUG_ON(ret < 0 && ret != -ENOENT);
2186
2187 /* no backref to be processed for this extent */
2188 if (!old->count) {
2189 list_del(&old->list);
2190 kfree(old);
2191 }
2192 }
2193
2194 if (list_empty(&new->head))
2195 return false;
2196
2197 return true;
2198 }
2199
2200 static int relink_is_mergable(struct extent_buffer *leaf,
2201 struct btrfs_file_extent_item *fi,
2202 u64 disk_bytenr)
2203 {
2204 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2205 return 0;
2206
2207 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2208 return 0;
2209
2210 if (btrfs_file_extent_compression(leaf, fi) ||
2211 btrfs_file_extent_encryption(leaf, fi) ||
2212 btrfs_file_extent_other_encoding(leaf, fi))
2213 return 0;
2214
2215 return 1;
2216 }
2217
2218 /*
2219 * Note the backref might has changed, and in this case we just return 0.
2220 */
2221 static noinline int relink_extent_backref(struct btrfs_path *path,
2222 struct sa_defrag_extent_backref *prev,
2223 struct sa_defrag_extent_backref *backref)
2224 {
2225 struct btrfs_file_extent_item *extent;
2226 struct btrfs_file_extent_item *item;
2227 struct btrfs_ordered_extent *ordered;
2228 struct btrfs_trans_handle *trans;
2229 struct btrfs_fs_info *fs_info;
2230 struct btrfs_root *root;
2231 struct btrfs_key key;
2232 struct extent_buffer *leaf;
2233 struct old_sa_defrag_extent *old = backref->old;
2234 struct new_sa_defrag_extent *new = old->new;
2235 struct inode *src_inode = new->inode;
2236 struct inode *inode;
2237 struct extent_state *cached = NULL;
2238 int ret = 0;
2239 u64 start;
2240 u64 len;
2241 u64 lock_start;
2242 u64 lock_end;
2243 bool merge = false;
2244 int index;
2245
2246 if (prev && prev->root_id == backref->root_id &&
2247 prev->inum == backref->inum &&
2248 prev->file_pos + prev->num_bytes == backref->file_pos)
2249 merge = true;
2250
2251 /* step 1: get root */
2252 key.objectid = backref->root_id;
2253 key.type = BTRFS_ROOT_ITEM_KEY;
2254 key.offset = (u64)-1;
2255
2256 fs_info = BTRFS_I(src_inode)->root->fs_info;
2257 index = srcu_read_lock(&fs_info->subvol_srcu);
2258
2259 root = btrfs_read_fs_root_no_name(fs_info, &key);
2260 if (IS_ERR(root)) {
2261 srcu_read_unlock(&fs_info->subvol_srcu, index);
2262 if (PTR_ERR(root) == -ENOENT)
2263 return 0;
2264 return PTR_ERR(root);
2265 }
2266 if (btrfs_root_refs(&root->root_item) == 0) {
2267 srcu_read_unlock(&fs_info->subvol_srcu, index);
2268 /* parse ENOENT to 0 */
2269 return 0;
2270 }
2271
2272 /* step 2: get inode */
2273 key.objectid = backref->inum;
2274 key.type = BTRFS_INODE_ITEM_KEY;
2275 key.offset = 0;
2276
2277 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2278 if (IS_ERR(inode)) {
2279 srcu_read_unlock(&fs_info->subvol_srcu, index);
2280 return 0;
2281 }
2282
2283 srcu_read_unlock(&fs_info->subvol_srcu, index);
2284
2285 /* step 3: relink backref */
2286 lock_start = backref->file_pos;
2287 lock_end = backref->file_pos + backref->num_bytes - 1;
2288 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2289 0, &cached);
2290
2291 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2292 if (ordered) {
2293 btrfs_put_ordered_extent(ordered);
2294 goto out_unlock;
2295 }
2296
2297 trans = btrfs_join_transaction(root);
2298 if (IS_ERR(trans)) {
2299 ret = PTR_ERR(trans);
2300 goto out_unlock;
2301 }
2302
2303 key.objectid = backref->inum;
2304 key.type = BTRFS_EXTENT_DATA_KEY;
2305 key.offset = backref->file_pos;
2306
2307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2308 if (ret < 0) {
2309 goto out_free_path;
2310 } else if (ret > 0) {
2311 ret = 0;
2312 goto out_free_path;
2313 }
2314
2315 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2316 struct btrfs_file_extent_item);
2317
2318 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2319 backref->generation)
2320 goto out_free_path;
2321
2322 btrfs_release_path(path);
2323
2324 start = backref->file_pos;
2325 if (backref->extent_offset < old->extent_offset + old->offset)
2326 start += old->extent_offset + old->offset -
2327 backref->extent_offset;
2328
2329 len = min(backref->extent_offset + backref->num_bytes,
2330 old->extent_offset + old->offset + old->len);
2331 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2332
2333 ret = btrfs_drop_extents(trans, root, inode, start,
2334 start + len, 1);
2335 if (ret)
2336 goto out_free_path;
2337 again:
2338 key.objectid = btrfs_ino(inode);
2339 key.type = BTRFS_EXTENT_DATA_KEY;
2340 key.offset = start;
2341
2342 path->leave_spinning = 1;
2343 if (merge) {
2344 struct btrfs_file_extent_item *fi;
2345 u64 extent_len;
2346 struct btrfs_key found_key;
2347
2348 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2349 if (ret < 0)
2350 goto out_free_path;
2351
2352 path->slots[0]--;
2353 leaf = path->nodes[0];
2354 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356 fi = btrfs_item_ptr(leaf, path->slots[0],
2357 struct btrfs_file_extent_item);
2358 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2359
2360 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2361 extent_len + found_key.offset == start) {
2362 btrfs_set_file_extent_num_bytes(leaf, fi,
2363 extent_len + len);
2364 btrfs_mark_buffer_dirty(leaf);
2365 inode_add_bytes(inode, len);
2366
2367 ret = 1;
2368 goto out_free_path;
2369 } else {
2370 merge = false;
2371 btrfs_release_path(path);
2372 goto again;
2373 }
2374 }
2375
2376 ret = btrfs_insert_empty_item(trans, root, path, &key,
2377 sizeof(*extent));
2378 if (ret) {
2379 btrfs_abort_transaction(trans, root, ret);
2380 goto out_free_path;
2381 }
2382
2383 leaf = path->nodes[0];
2384 item = btrfs_item_ptr(leaf, path->slots[0],
2385 struct btrfs_file_extent_item);
2386 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2387 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2388 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2389 btrfs_set_file_extent_num_bytes(leaf, item, len);
2390 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2391 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2392 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2393 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2394 btrfs_set_file_extent_encryption(leaf, item, 0);
2395 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2396
2397 btrfs_mark_buffer_dirty(leaf);
2398 inode_add_bytes(inode, len);
2399 btrfs_release_path(path);
2400
2401 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2402 new->disk_len, 0,
2403 backref->root_id, backref->inum,
2404 new->file_pos, 0); /* start - extent_offset */
2405 if (ret) {
2406 btrfs_abort_transaction(trans, root, ret);
2407 goto out_free_path;
2408 }
2409
2410 ret = 1;
2411 out_free_path:
2412 btrfs_release_path(path);
2413 path->leave_spinning = 0;
2414 btrfs_end_transaction(trans, root);
2415 out_unlock:
2416 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2417 &cached, GFP_NOFS);
2418 iput(inode);
2419 return ret;
2420 }
2421
2422 static void relink_file_extents(struct new_sa_defrag_extent *new)
2423 {
2424 struct btrfs_path *path;
2425 struct old_sa_defrag_extent *old, *tmp;
2426 struct sa_defrag_extent_backref *backref;
2427 struct sa_defrag_extent_backref *prev = NULL;
2428 struct inode *inode;
2429 struct btrfs_root *root;
2430 struct rb_node *node;
2431 int ret;
2432
2433 inode = new->inode;
2434 root = BTRFS_I(inode)->root;
2435
2436 path = btrfs_alloc_path();
2437 if (!path)
2438 return;
2439
2440 if (!record_extent_backrefs(path, new)) {
2441 btrfs_free_path(path);
2442 goto out;
2443 }
2444 btrfs_release_path(path);
2445
2446 while (1) {
2447 node = rb_first(&new->root);
2448 if (!node)
2449 break;
2450 rb_erase(node, &new->root);
2451
2452 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2453
2454 ret = relink_extent_backref(path, prev, backref);
2455 WARN_ON(ret < 0);
2456
2457 kfree(prev);
2458
2459 if (ret == 1)
2460 prev = backref;
2461 else
2462 prev = NULL;
2463 cond_resched();
2464 }
2465 kfree(prev);
2466
2467 btrfs_free_path(path);
2468
2469 list_for_each_entry_safe(old, tmp, &new->head, list) {
2470 list_del(&old->list);
2471 kfree(old);
2472 }
2473 out:
2474 atomic_dec(&root->fs_info->defrag_running);
2475 wake_up(&root->fs_info->transaction_wait);
2476
2477 kfree(new);
2478 }
2479
2480 static struct new_sa_defrag_extent *
2481 record_old_file_extents(struct inode *inode,
2482 struct btrfs_ordered_extent *ordered)
2483 {
2484 struct btrfs_root *root = BTRFS_I(inode)->root;
2485 struct btrfs_path *path;
2486 struct btrfs_key key;
2487 struct old_sa_defrag_extent *old, *tmp;
2488 struct new_sa_defrag_extent *new;
2489 int ret;
2490
2491 new = kmalloc(sizeof(*new), GFP_NOFS);
2492 if (!new)
2493 return NULL;
2494
2495 new->inode = inode;
2496 new->file_pos = ordered->file_offset;
2497 new->len = ordered->len;
2498 new->bytenr = ordered->start;
2499 new->disk_len = ordered->disk_len;
2500 new->compress_type = ordered->compress_type;
2501 new->root = RB_ROOT;
2502 INIT_LIST_HEAD(&new->head);
2503
2504 path = btrfs_alloc_path();
2505 if (!path)
2506 goto out_kfree;
2507
2508 key.objectid = btrfs_ino(inode);
2509 key.type = BTRFS_EXTENT_DATA_KEY;
2510 key.offset = new->file_pos;
2511
2512 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2513 if (ret < 0)
2514 goto out_free_path;
2515 if (ret > 0 && path->slots[0] > 0)
2516 path->slots[0]--;
2517
2518 /* find out all the old extents for the file range */
2519 while (1) {
2520 struct btrfs_file_extent_item *extent;
2521 struct extent_buffer *l;
2522 int slot;
2523 u64 num_bytes;
2524 u64 offset;
2525 u64 end;
2526 u64 disk_bytenr;
2527 u64 extent_offset;
2528
2529 l = path->nodes[0];
2530 slot = path->slots[0];
2531
2532 if (slot >= btrfs_header_nritems(l)) {
2533 ret = btrfs_next_leaf(root, path);
2534 if (ret < 0)
2535 goto out_free_list;
2536 else if (ret > 0)
2537 break;
2538 continue;
2539 }
2540
2541 btrfs_item_key_to_cpu(l, &key, slot);
2542
2543 if (key.objectid != btrfs_ino(inode))
2544 break;
2545 if (key.type != BTRFS_EXTENT_DATA_KEY)
2546 break;
2547 if (key.offset >= new->file_pos + new->len)
2548 break;
2549
2550 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2551
2552 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2553 if (key.offset + num_bytes < new->file_pos)
2554 goto next;
2555
2556 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2557 if (!disk_bytenr)
2558 goto next;
2559
2560 extent_offset = btrfs_file_extent_offset(l, extent);
2561
2562 old = kmalloc(sizeof(*old), GFP_NOFS);
2563 if (!old)
2564 goto out_free_list;
2565
2566 offset = max(new->file_pos, key.offset);
2567 end = min(new->file_pos + new->len, key.offset + num_bytes);
2568
2569 old->bytenr = disk_bytenr;
2570 old->extent_offset = extent_offset;
2571 old->offset = offset - key.offset;
2572 old->len = end - offset;
2573 old->new = new;
2574 old->count = 0;
2575 list_add_tail(&old->list, &new->head);
2576 next:
2577 path->slots[0]++;
2578 cond_resched();
2579 }
2580
2581 btrfs_free_path(path);
2582 atomic_inc(&root->fs_info->defrag_running);
2583
2584 return new;
2585
2586 out_free_list:
2587 list_for_each_entry_safe(old, tmp, &new->head, list) {
2588 list_del(&old->list);
2589 kfree(old);
2590 }
2591 out_free_path:
2592 btrfs_free_path(path);
2593 out_kfree:
2594 kfree(new);
2595 return NULL;
2596 }
2597
2598 /*
2599 * helper function for btrfs_finish_ordered_io, this
2600 * just reads in some of the csum leaves to prime them into ram
2601 * before we start the transaction. It limits the amount of btree
2602 * reads required while inside the transaction.
2603 */
2604 /* as ordered data IO finishes, this gets called so we can finish
2605 * an ordered extent if the range of bytes in the file it covers are
2606 * fully written.
2607 */
2608 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2609 {
2610 struct inode *inode = ordered_extent->inode;
2611 struct btrfs_root *root = BTRFS_I(inode)->root;
2612 struct btrfs_trans_handle *trans = NULL;
2613 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2614 struct extent_state *cached_state = NULL;
2615 struct new_sa_defrag_extent *new = NULL;
2616 int compress_type = 0;
2617 int ret;
2618 bool nolock;
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_NOCOW, &ordered_extent->flags)) {
2628 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2629 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2630 if (nolock)
2631 trans = btrfs_join_transaction_nolock(root);
2632 else
2633 trans = btrfs_join_transaction(root);
2634 if (IS_ERR(trans)) {
2635 ret = PTR_ERR(trans);
2636 trans = NULL;
2637 goto out;
2638 }
2639 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2640 ret = btrfs_update_inode_fallback(trans, root, inode);
2641 if (ret) /* -ENOMEM or corruption */
2642 btrfs_abort_transaction(trans, root, ret);
2643 goto out;
2644 }
2645
2646 lock_extent_bits(io_tree, ordered_extent->file_offset,
2647 ordered_extent->file_offset + ordered_extent->len - 1,
2648 0, &cached_state);
2649
2650 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2651 ordered_extent->file_offset + ordered_extent->len - 1,
2652 EXTENT_DEFRAG, 1, cached_state);
2653 if (ret) {
2654 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2655 if (last_snapshot >= BTRFS_I(inode)->generation)
2656 /* the inode is shared */
2657 new = record_old_file_extents(inode, ordered_extent);
2658
2659 clear_extent_bit(io_tree, ordered_extent->file_offset,
2660 ordered_extent->file_offset + ordered_extent->len - 1,
2661 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2662 }
2663
2664 if (nolock)
2665 trans = btrfs_join_transaction_nolock(root);
2666 else
2667 trans = btrfs_join_transaction(root);
2668 if (IS_ERR(trans)) {
2669 ret = PTR_ERR(trans);
2670 trans = NULL;
2671 goto out_unlock;
2672 }
2673 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2674
2675 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2676 compress_type = ordered_extent->compress_type;
2677 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2678 BUG_ON(compress_type);
2679 ret = btrfs_mark_extent_written(trans, inode,
2680 ordered_extent->file_offset,
2681 ordered_extent->file_offset +
2682 ordered_extent->len);
2683 } else {
2684 BUG_ON(root == root->fs_info->tree_root);
2685 ret = insert_reserved_file_extent(trans, inode,
2686 ordered_extent->file_offset,
2687 ordered_extent->start,
2688 ordered_extent->disk_len,
2689 ordered_extent->len,
2690 ordered_extent->len,
2691 compress_type, 0, 0,
2692 BTRFS_FILE_EXTENT_REG);
2693 }
2694 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2695 ordered_extent->file_offset, ordered_extent->len,
2696 trans->transid);
2697 if (ret < 0) {
2698 btrfs_abort_transaction(trans, root, ret);
2699 goto out_unlock;
2700 }
2701
2702 add_pending_csums(trans, inode, ordered_extent->file_offset,
2703 &ordered_extent->list);
2704
2705 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2706 ret = btrfs_update_inode_fallback(trans, root, inode);
2707 if (ret) { /* -ENOMEM or corruption */
2708 btrfs_abort_transaction(trans, root, ret);
2709 goto out_unlock;
2710 }
2711 ret = 0;
2712 out_unlock:
2713 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2714 ordered_extent->file_offset +
2715 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2716 out:
2717 if (root != root->fs_info->tree_root)
2718 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2719 if (trans)
2720 btrfs_end_transaction(trans, root);
2721
2722 if (ret) {
2723 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2724 ordered_extent->file_offset +
2725 ordered_extent->len - 1, NULL, GFP_NOFS);
2726
2727 /*
2728 * If the ordered extent had an IOERR or something else went
2729 * wrong we need to return the space for this ordered extent
2730 * back to the allocator.
2731 */
2732 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2733 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2734 btrfs_free_reserved_extent(root, ordered_extent->start,
2735 ordered_extent->disk_len);
2736 }
2737
2738
2739 /*
2740 * This needs to be done to make sure anybody waiting knows we are done
2741 * updating everything for this ordered extent.
2742 */
2743 btrfs_remove_ordered_extent(inode, ordered_extent);
2744
2745 /* for snapshot-aware defrag */
2746 if (new)
2747 relink_file_extents(new);
2748
2749 /* once for us */
2750 btrfs_put_ordered_extent(ordered_extent);
2751 /* once for the tree */
2752 btrfs_put_ordered_extent(ordered_extent);
2753
2754 return ret;
2755 }
2756
2757 static void finish_ordered_fn(struct btrfs_work *work)
2758 {
2759 struct btrfs_ordered_extent *ordered_extent;
2760 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2761 btrfs_finish_ordered_io(ordered_extent);
2762 }
2763
2764 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2765 struct extent_state *state, int uptodate)
2766 {
2767 struct inode *inode = page->mapping->host;
2768 struct btrfs_root *root = BTRFS_I(inode)->root;
2769 struct btrfs_ordered_extent *ordered_extent = NULL;
2770 struct btrfs_workers *workers;
2771
2772 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2773
2774 ClearPagePrivate2(page);
2775 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2776 end - start + 1, uptodate))
2777 return 0;
2778
2779 ordered_extent->work.func = finish_ordered_fn;
2780 ordered_extent->work.flags = 0;
2781
2782 if (btrfs_is_free_space_inode(inode))
2783 workers = &root->fs_info->endio_freespace_worker;
2784 else
2785 workers = &root->fs_info->endio_write_workers;
2786 btrfs_queue_worker(workers, &ordered_extent->work);
2787
2788 return 0;
2789 }
2790
2791 /*
2792 * when reads are done, we need to check csums to verify the data is correct
2793 * if there's a match, we allow the bio to finish. If not, the code in
2794 * extent_io.c will try to find good copies for us.
2795 */
2796 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2797 struct extent_state *state, int mirror)
2798 {
2799 size_t offset = start - page_offset(page);
2800 struct inode *inode = page->mapping->host;
2801 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2802 char *kaddr;
2803 u64 private = ~(u32)0;
2804 int ret;
2805 struct btrfs_root *root = BTRFS_I(inode)->root;
2806 u32 csum = ~(u32)0;
2807 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2808 DEFAULT_RATELIMIT_BURST);
2809
2810 if (PageChecked(page)) {
2811 ClearPageChecked(page);
2812 goto good;
2813 }
2814
2815 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2816 goto good;
2817
2818 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2819 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2820 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2821 GFP_NOFS);
2822 return 0;
2823 }
2824
2825 if (state && state->start == start) {
2826 private = state->private;
2827 ret = 0;
2828 } else {
2829 ret = get_state_private(io_tree, start, &private);
2830 }
2831 kaddr = kmap_atomic(page);
2832 if (ret)
2833 goto zeroit;
2834
2835 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2836 btrfs_csum_final(csum, (char *)&csum);
2837 if (csum != private)
2838 goto zeroit;
2839
2840 kunmap_atomic(kaddr);
2841 good:
2842 return 0;
2843
2844 zeroit:
2845 if (__ratelimit(&_rs))
2846 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2847 (unsigned long long)btrfs_ino(page->mapping->host),
2848 (unsigned long long)start, csum,
2849 (unsigned long long)private);
2850 memset(kaddr + offset, 1, end - start + 1);
2851 flush_dcache_page(page);
2852 kunmap_atomic(kaddr);
2853 if (private == 0)
2854 return 0;
2855 return -EIO;
2856 }
2857
2858 struct delayed_iput {
2859 struct list_head list;
2860 struct inode *inode;
2861 };
2862
2863 /* JDM: If this is fs-wide, why can't we add a pointer to
2864 * btrfs_inode instead and avoid the allocation? */
2865 void btrfs_add_delayed_iput(struct inode *inode)
2866 {
2867 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2868 struct delayed_iput *delayed;
2869
2870 if (atomic_add_unless(&inode->i_count, -1, 1))
2871 return;
2872
2873 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2874 delayed->inode = inode;
2875
2876 spin_lock(&fs_info->delayed_iput_lock);
2877 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2878 spin_unlock(&fs_info->delayed_iput_lock);
2879 }
2880
2881 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2882 {
2883 LIST_HEAD(list);
2884 struct btrfs_fs_info *fs_info = root->fs_info;
2885 struct delayed_iput *delayed;
2886 int empty;
2887
2888 spin_lock(&fs_info->delayed_iput_lock);
2889 empty = list_empty(&fs_info->delayed_iputs);
2890 spin_unlock(&fs_info->delayed_iput_lock);
2891 if (empty)
2892 return;
2893
2894 spin_lock(&fs_info->delayed_iput_lock);
2895 list_splice_init(&fs_info->delayed_iputs, &list);
2896 spin_unlock(&fs_info->delayed_iput_lock);
2897
2898 while (!list_empty(&list)) {
2899 delayed = list_entry(list.next, struct delayed_iput, list);
2900 list_del(&delayed->list);
2901 iput(delayed->inode);
2902 kfree(delayed);
2903 }
2904 }
2905
2906 /*
2907 * This is called in transaction commit time. If there are no orphan
2908 * files in the subvolume, it removes orphan item and frees block_rsv
2909 * structure.
2910 */
2911 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2912 struct btrfs_root *root)
2913 {
2914 struct btrfs_block_rsv *block_rsv;
2915 int ret;
2916
2917 if (atomic_read(&root->orphan_inodes) ||
2918 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2919 return;
2920
2921 spin_lock(&root->orphan_lock);
2922 if (atomic_read(&root->orphan_inodes)) {
2923 spin_unlock(&root->orphan_lock);
2924 return;
2925 }
2926
2927 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2928 spin_unlock(&root->orphan_lock);
2929 return;
2930 }
2931
2932 block_rsv = root->orphan_block_rsv;
2933 root->orphan_block_rsv = NULL;
2934 spin_unlock(&root->orphan_lock);
2935
2936 if (root->orphan_item_inserted &&
2937 btrfs_root_refs(&root->root_item) > 0) {
2938 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2939 root->root_key.objectid);
2940 BUG_ON(ret);
2941 root->orphan_item_inserted = 0;
2942 }
2943
2944 if (block_rsv) {
2945 WARN_ON(block_rsv->size > 0);
2946 btrfs_free_block_rsv(root, block_rsv);
2947 }
2948 }
2949
2950 /*
2951 * This creates an orphan entry for the given inode in case something goes
2952 * wrong in the middle of an unlink/truncate.
2953 *
2954 * NOTE: caller of this function should reserve 5 units of metadata for
2955 * this function.
2956 */
2957 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2958 {
2959 struct btrfs_root *root = BTRFS_I(inode)->root;
2960 struct btrfs_block_rsv *block_rsv = NULL;
2961 int reserve = 0;
2962 int insert = 0;
2963 int ret;
2964
2965 if (!root->orphan_block_rsv) {
2966 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2967 if (!block_rsv)
2968 return -ENOMEM;
2969 }
2970
2971 spin_lock(&root->orphan_lock);
2972 if (!root->orphan_block_rsv) {
2973 root->orphan_block_rsv = block_rsv;
2974 } else if (block_rsv) {
2975 btrfs_free_block_rsv(root, block_rsv);
2976 block_rsv = NULL;
2977 }
2978
2979 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980 &BTRFS_I(inode)->runtime_flags)) {
2981 #if 0
2982 /*
2983 * For proper ENOSPC handling, we should do orphan
2984 * cleanup when mounting. But this introduces backward
2985 * compatibility issue.
2986 */
2987 if (!xchg(&root->orphan_item_inserted, 1))
2988 insert = 2;
2989 else
2990 insert = 1;
2991 #endif
2992 insert = 1;
2993 atomic_inc(&root->orphan_inodes);
2994 }
2995
2996 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2997 &BTRFS_I(inode)->runtime_flags))
2998 reserve = 1;
2999 spin_unlock(&root->orphan_lock);
3000
3001 /* grab metadata reservation from transaction handle */
3002 if (reserve) {
3003 ret = btrfs_orphan_reserve_metadata(trans, inode);
3004 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3005 }
3006
3007 /* insert an orphan item to track this unlinked/truncated file */
3008 if (insert >= 1) {
3009 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3010 if (ret && ret != -EEXIST) {
3011 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3012 &BTRFS_I(inode)->runtime_flags);
3013 btrfs_abort_transaction(trans, root, ret);
3014 return ret;
3015 }
3016 ret = 0;
3017 }
3018
3019 /* insert an orphan item to track subvolume contains orphan files */
3020 if (insert >= 2) {
3021 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3022 root->root_key.objectid);
3023 if (ret && ret != -EEXIST) {
3024 btrfs_abort_transaction(trans, root, ret);
3025 return ret;
3026 }
3027 }
3028 return 0;
3029 }
3030
3031 /*
3032 * We have done the truncate/delete so we can go ahead and remove the orphan
3033 * item for this particular inode.
3034 */
3035 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3036 struct inode *inode)
3037 {
3038 struct btrfs_root *root = BTRFS_I(inode)->root;
3039 int delete_item = 0;
3040 int release_rsv = 0;
3041 int ret = 0;
3042
3043 spin_lock(&root->orphan_lock);
3044 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3045 &BTRFS_I(inode)->runtime_flags))
3046 delete_item = 1;
3047
3048 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3049 &BTRFS_I(inode)->runtime_flags))
3050 release_rsv = 1;
3051 spin_unlock(&root->orphan_lock);
3052
3053 if (trans && delete_item) {
3054 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3055 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3056 }
3057
3058 if (release_rsv) {
3059 btrfs_orphan_release_metadata(inode);
3060 atomic_dec(&root->orphan_inodes);
3061 }
3062
3063 return 0;
3064 }
3065
3066 /*
3067 * this cleans up any orphans that may be left on the list from the last use
3068 * of this root.
3069 */
3070 int btrfs_orphan_cleanup(struct btrfs_root *root)
3071 {
3072 struct btrfs_path *path;
3073 struct extent_buffer *leaf;
3074 struct btrfs_key key, found_key;
3075 struct btrfs_trans_handle *trans;
3076 struct inode *inode;
3077 u64 last_objectid = 0;
3078 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3079
3080 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3081 return 0;
3082
3083 path = btrfs_alloc_path();
3084 if (!path) {
3085 ret = -ENOMEM;
3086 goto out;
3087 }
3088 path->reada = -1;
3089
3090 key.objectid = BTRFS_ORPHAN_OBJECTID;
3091 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3092 key.offset = (u64)-1;
3093
3094 while (1) {
3095 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3096 if (ret < 0)
3097 goto out;
3098
3099 /*
3100 * if ret == 0 means we found what we were searching for, which
3101 * is weird, but possible, so only screw with path if we didn't
3102 * find the key and see if we have stuff that matches
3103 */
3104 if (ret > 0) {
3105 ret = 0;
3106 if (path->slots[0] == 0)
3107 break;
3108 path->slots[0]--;
3109 }
3110
3111 /* pull out the item */
3112 leaf = path->nodes[0];
3113 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3114
3115 /* make sure the item matches what we want */
3116 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3117 break;
3118 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3119 break;
3120
3121 /* release the path since we're done with it */
3122 btrfs_release_path(path);
3123
3124 /*
3125 * this is where we are basically btrfs_lookup, without the
3126 * crossing root thing. we store the inode number in the
3127 * offset of the orphan item.
3128 */
3129
3130 if (found_key.offset == last_objectid) {
3131 btrfs_err(root->fs_info,
3132 "Error removing orphan entry, stopping orphan cleanup");
3133 ret = -EINVAL;
3134 goto out;
3135 }
3136
3137 last_objectid = found_key.offset;
3138
3139 found_key.objectid = found_key.offset;
3140 found_key.type = BTRFS_INODE_ITEM_KEY;
3141 found_key.offset = 0;
3142 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3143 ret = PTR_RET(inode);
3144 if (ret && ret != -ESTALE)
3145 goto out;
3146
3147 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3148 struct btrfs_root *dead_root;
3149 struct btrfs_fs_info *fs_info = root->fs_info;
3150 int is_dead_root = 0;
3151
3152 /*
3153 * this is an orphan in the tree root. Currently these
3154 * could come from 2 sources:
3155 * a) a snapshot deletion in progress
3156 * b) a free space cache inode
3157 * We need to distinguish those two, as the snapshot
3158 * orphan must not get deleted.
3159 * find_dead_roots already ran before us, so if this
3160 * is a snapshot deletion, we should find the root
3161 * in the dead_roots list
3162 */
3163 spin_lock(&fs_info->trans_lock);
3164 list_for_each_entry(dead_root, &fs_info->dead_roots,
3165 root_list) {
3166 if (dead_root->root_key.objectid ==
3167 found_key.objectid) {
3168 is_dead_root = 1;
3169 break;
3170 }
3171 }
3172 spin_unlock(&fs_info->trans_lock);
3173 if (is_dead_root) {
3174 /* prevent this orphan from being found again */
3175 key.offset = found_key.objectid - 1;
3176 continue;
3177 }
3178 }
3179 /*
3180 * Inode is already gone but the orphan item is still there,
3181 * kill the orphan item.
3182 */
3183 if (ret == -ESTALE) {
3184 trans = btrfs_start_transaction(root, 1);
3185 if (IS_ERR(trans)) {
3186 ret = PTR_ERR(trans);
3187 goto out;
3188 }
3189 btrfs_debug(root->fs_info, "auto deleting %Lu",
3190 found_key.objectid);
3191 ret = btrfs_del_orphan_item(trans, root,
3192 found_key.objectid);
3193 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3194 btrfs_end_transaction(trans, root);
3195 continue;
3196 }
3197
3198 /*
3199 * add this inode to the orphan list so btrfs_orphan_del does
3200 * the proper thing when we hit it
3201 */
3202 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3203 &BTRFS_I(inode)->runtime_flags);
3204 atomic_inc(&root->orphan_inodes);
3205
3206 /* if we have links, this was a truncate, lets do that */
3207 if (inode->i_nlink) {
3208 if (!S_ISREG(inode->i_mode)) {
3209 WARN_ON(1);
3210 iput(inode);
3211 continue;
3212 }
3213 nr_truncate++;
3214
3215 /* 1 for the orphan item deletion. */
3216 trans = btrfs_start_transaction(root, 1);
3217 if (IS_ERR(trans)) {
3218 ret = PTR_ERR(trans);
3219 goto out;
3220 }
3221 ret = btrfs_orphan_add(trans, inode);
3222 btrfs_end_transaction(trans, root);
3223 if (ret)
3224 goto out;
3225
3226 ret = btrfs_truncate(inode);
3227 if (ret)
3228 btrfs_orphan_del(NULL, inode);
3229 } else {
3230 nr_unlink++;
3231 }
3232
3233 /* this will do delete_inode and everything for us */
3234 iput(inode);
3235 if (ret)
3236 goto out;
3237 }
3238 /* release the path since we're done with it */
3239 btrfs_release_path(path);
3240
3241 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3242
3243 if (root->orphan_block_rsv)
3244 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3245 (u64)-1);
3246
3247 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3248 trans = btrfs_join_transaction(root);
3249 if (!IS_ERR(trans))
3250 btrfs_end_transaction(trans, root);
3251 }
3252
3253 if (nr_unlink)
3254 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3255 if (nr_truncate)
3256 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3257
3258 out:
3259 if (ret)
3260 btrfs_crit(root->fs_info,
3261 "could not do orphan cleanup %d", ret);
3262 btrfs_free_path(path);
3263 return ret;
3264 }
3265
3266 /*
3267 * very simple check to peek ahead in the leaf looking for xattrs. If we
3268 * don't find any xattrs, we know there can't be any acls.
3269 *
3270 * slot is the slot the inode is in, objectid is the objectid of the inode
3271 */
3272 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3273 int slot, u64 objectid)
3274 {
3275 u32 nritems = btrfs_header_nritems(leaf);
3276 struct btrfs_key found_key;
3277 int scanned = 0;
3278
3279 slot++;
3280 while (slot < nritems) {
3281 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3282
3283 /* we found a different objectid, there must not be acls */
3284 if (found_key.objectid != objectid)
3285 return 0;
3286
3287 /* we found an xattr, assume we've got an acl */
3288 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3289 return 1;
3290
3291 /*
3292 * we found a key greater than an xattr key, there can't
3293 * be any acls later on
3294 */
3295 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3296 return 0;
3297
3298 slot++;
3299 scanned++;
3300
3301 /*
3302 * it goes inode, inode backrefs, xattrs, extents,
3303 * so if there are a ton of hard links to an inode there can
3304 * be a lot of backrefs. Don't waste time searching too hard,
3305 * this is just an optimization
3306 */
3307 if (scanned >= 8)
3308 break;
3309 }
3310 /* we hit the end of the leaf before we found an xattr or
3311 * something larger than an xattr. We have to assume the inode
3312 * has acls
3313 */
3314 return 1;
3315 }
3316
3317 /*
3318 * read an inode from the btree into the in-memory inode
3319 */
3320 static void btrfs_read_locked_inode(struct inode *inode)
3321 {
3322 struct btrfs_path *path;
3323 struct extent_buffer *leaf;
3324 struct btrfs_inode_item *inode_item;
3325 struct btrfs_timespec *tspec;
3326 struct btrfs_root *root = BTRFS_I(inode)->root;
3327 struct btrfs_key location;
3328 int maybe_acls;
3329 u32 rdev;
3330 int ret;
3331 bool filled = false;
3332
3333 ret = btrfs_fill_inode(inode, &rdev);
3334 if (!ret)
3335 filled = true;
3336
3337 path = btrfs_alloc_path();
3338 if (!path)
3339 goto make_bad;
3340
3341 path->leave_spinning = 1;
3342 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3343
3344 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3345 if (ret)
3346 goto make_bad;
3347
3348 leaf = path->nodes[0];
3349
3350 if (filled)
3351 goto cache_acl;
3352
3353 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3354 struct btrfs_inode_item);
3355 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3356 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3357 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3358 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3359 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3360
3361 tspec = btrfs_inode_atime(inode_item);
3362 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3363 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3364
3365 tspec = btrfs_inode_mtime(inode_item);
3366 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3367 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3368
3369 tspec = btrfs_inode_ctime(inode_item);
3370 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3371 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3372
3373 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3374 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3375 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3376
3377 /*
3378 * If we were modified in the current generation and evicted from memory
3379 * and then re-read we need to do a full sync since we don't have any
3380 * idea about which extents were modified before we were evicted from
3381 * cache.
3382 */
3383 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3384 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3385 &BTRFS_I(inode)->runtime_flags);
3386
3387 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3388 inode->i_generation = BTRFS_I(inode)->generation;
3389 inode->i_rdev = 0;
3390 rdev = btrfs_inode_rdev(leaf, inode_item);
3391
3392 BTRFS_I(inode)->index_cnt = (u64)-1;
3393 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3394 cache_acl:
3395 /*
3396 * try to precache a NULL acl entry for files that don't have
3397 * any xattrs or acls
3398 */
3399 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3400 btrfs_ino(inode));
3401 if (!maybe_acls)
3402 cache_no_acl(inode);
3403
3404 btrfs_free_path(path);
3405
3406 switch (inode->i_mode & S_IFMT) {
3407 case S_IFREG:
3408 inode->i_mapping->a_ops = &btrfs_aops;
3409 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3410 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3411 inode->i_fop = &btrfs_file_operations;
3412 inode->i_op = &btrfs_file_inode_operations;
3413 break;
3414 case S_IFDIR:
3415 inode->i_fop = &btrfs_dir_file_operations;
3416 if (root == root->fs_info->tree_root)
3417 inode->i_op = &btrfs_dir_ro_inode_operations;
3418 else
3419 inode->i_op = &btrfs_dir_inode_operations;
3420 break;
3421 case S_IFLNK:
3422 inode->i_op = &btrfs_symlink_inode_operations;
3423 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3424 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3425 break;
3426 default:
3427 inode->i_op = &btrfs_special_inode_operations;
3428 init_special_inode(inode, inode->i_mode, rdev);
3429 break;
3430 }
3431
3432 btrfs_update_iflags(inode);
3433 return;
3434
3435 make_bad:
3436 btrfs_free_path(path);
3437 make_bad_inode(inode);
3438 }
3439
3440 /*
3441 * given a leaf and an inode, copy the inode fields into the leaf
3442 */
3443 static void fill_inode_item(struct btrfs_trans_handle *trans,
3444 struct extent_buffer *leaf,
3445 struct btrfs_inode_item *item,
3446 struct inode *inode)
3447 {
3448 struct btrfs_map_token token;
3449
3450 btrfs_init_map_token(&token);
3451
3452 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3453 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3454 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3455 &token);
3456 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3457 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3458
3459 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3460 inode->i_atime.tv_sec, &token);
3461 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3462 inode->i_atime.tv_nsec, &token);
3463
3464 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3465 inode->i_mtime.tv_sec, &token);
3466 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3467 inode->i_mtime.tv_nsec, &token);
3468
3469 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3470 inode->i_ctime.tv_sec, &token);
3471 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3472 inode->i_ctime.tv_nsec, &token);
3473
3474 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3475 &token);
3476 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3477 &token);
3478 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3479 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3480 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3481 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3482 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3483 }
3484
3485 /*
3486 * copy everything in the in-memory inode into the btree.
3487 */
3488 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *root, struct inode *inode)
3490 {
3491 struct btrfs_inode_item *inode_item;
3492 struct btrfs_path *path;
3493 struct extent_buffer *leaf;
3494 int ret;
3495
3496 path = btrfs_alloc_path();
3497 if (!path)
3498 return -ENOMEM;
3499
3500 path->leave_spinning = 1;
3501 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3502 1);
3503 if (ret) {
3504 if (ret > 0)
3505 ret = -ENOENT;
3506 goto failed;
3507 }
3508
3509 btrfs_unlock_up_safe(path, 1);
3510 leaf = path->nodes[0];
3511 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3512 struct btrfs_inode_item);
3513
3514 fill_inode_item(trans, leaf, inode_item, inode);
3515 btrfs_mark_buffer_dirty(leaf);
3516 btrfs_set_inode_last_trans(trans, inode);
3517 ret = 0;
3518 failed:
3519 btrfs_free_path(path);
3520 return ret;
3521 }
3522
3523 /*
3524 * copy everything in the in-memory inode into the btree.
3525 */
3526 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3527 struct btrfs_root *root, struct inode *inode)
3528 {
3529 int ret;
3530
3531 /*
3532 * If the inode is a free space inode, we can deadlock during commit
3533 * if we put it into the delayed code.
3534 *
3535 * The data relocation inode should also be directly updated
3536 * without delay
3537 */
3538 if (!btrfs_is_free_space_inode(inode)
3539 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3540 btrfs_update_root_times(trans, root);
3541
3542 ret = btrfs_delayed_update_inode(trans, root, inode);
3543 if (!ret)
3544 btrfs_set_inode_last_trans(trans, inode);
3545 return ret;
3546 }
3547
3548 return btrfs_update_inode_item(trans, root, inode);
3549 }
3550
3551 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3552 struct btrfs_root *root,
3553 struct inode *inode)
3554 {
3555 int ret;
3556
3557 ret = btrfs_update_inode(trans, root, inode);
3558 if (ret == -ENOSPC)
3559 return btrfs_update_inode_item(trans, root, inode);
3560 return ret;
3561 }
3562
3563 /*
3564 * unlink helper that gets used here in inode.c and in the tree logging
3565 * recovery code. It remove a link in a directory with a given name, and
3566 * also drops the back refs in the inode to the directory
3567 */
3568 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3569 struct btrfs_root *root,
3570 struct inode *dir, struct inode *inode,
3571 const char *name, int name_len)
3572 {
3573 struct btrfs_path *path;
3574 int ret = 0;
3575 struct extent_buffer *leaf;
3576 struct btrfs_dir_item *di;
3577 struct btrfs_key key;
3578 u64 index;
3579 u64 ino = btrfs_ino(inode);
3580 u64 dir_ino = btrfs_ino(dir);
3581
3582 path = btrfs_alloc_path();
3583 if (!path) {
3584 ret = -ENOMEM;
3585 goto out;
3586 }
3587
3588 path->leave_spinning = 1;
3589 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3590 name, name_len, -1);
3591 if (IS_ERR(di)) {
3592 ret = PTR_ERR(di);
3593 goto err;
3594 }
3595 if (!di) {
3596 ret = -ENOENT;
3597 goto err;
3598 }
3599 leaf = path->nodes[0];
3600 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3601 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3602 if (ret)
3603 goto err;
3604 btrfs_release_path(path);
3605
3606 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3607 dir_ino, &index);
3608 if (ret) {
3609 btrfs_info(root->fs_info,
3610 "failed to delete reference to %.*s, inode %llu parent %llu",
3611 name_len, name,
3612 (unsigned long long)ino, (unsigned long long)dir_ino);
3613 btrfs_abort_transaction(trans, root, ret);
3614 goto err;
3615 }
3616
3617 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3618 if (ret) {
3619 btrfs_abort_transaction(trans, root, ret);
3620 goto err;
3621 }
3622
3623 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3624 inode, dir_ino);
3625 if (ret != 0 && ret != -ENOENT) {
3626 btrfs_abort_transaction(trans, root, ret);
3627 goto err;
3628 }
3629
3630 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3631 dir, index);
3632 if (ret == -ENOENT)
3633 ret = 0;
3634 else if (ret)
3635 btrfs_abort_transaction(trans, root, ret);
3636 err:
3637 btrfs_free_path(path);
3638 if (ret)
3639 goto out;
3640
3641 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3642 inode_inc_iversion(inode);
3643 inode_inc_iversion(dir);
3644 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3645 ret = btrfs_update_inode(trans, root, dir);
3646 out:
3647 return ret;
3648 }
3649
3650 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3651 struct btrfs_root *root,
3652 struct inode *dir, struct inode *inode,
3653 const char *name, int name_len)
3654 {
3655 int ret;
3656 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3657 if (!ret) {
3658 btrfs_drop_nlink(inode);
3659 ret = btrfs_update_inode(trans, root, inode);
3660 }
3661 return ret;
3662 }
3663
3664
3665 /* helper to check if there is any shared block in the path */
3666 static int check_path_shared(struct btrfs_root *root,
3667 struct btrfs_path *path)
3668 {
3669 struct extent_buffer *eb;
3670 int level;
3671 u64 refs = 1;
3672
3673 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3674 int ret;
3675
3676 if (!path->nodes[level])
3677 break;
3678 eb = path->nodes[level];
3679 if (!btrfs_block_can_be_shared(root, eb))
3680 continue;
3681 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3682 &refs, NULL);
3683 if (refs > 1)
3684 return 1;
3685 }
3686 return 0;
3687 }
3688
3689 /*
3690 * helper to start transaction for unlink and rmdir.
3691 *
3692 * unlink and rmdir are special in btrfs, they do not always free space.
3693 * so in enospc case, we should make sure they will free space before
3694 * allowing them to use the global metadata reservation.
3695 */
3696 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3697 struct dentry *dentry)
3698 {
3699 struct btrfs_trans_handle *trans;
3700 struct btrfs_root *root = BTRFS_I(dir)->root;
3701 struct btrfs_path *path;
3702 struct btrfs_dir_item *di;
3703 struct inode *inode = dentry->d_inode;
3704 u64 index;
3705 int check_link = 1;
3706 int err = -ENOSPC;
3707 int ret;
3708 u64 ino = btrfs_ino(inode);
3709 u64 dir_ino = btrfs_ino(dir);
3710
3711 /*
3712 * 1 for the possible orphan item
3713 * 1 for the dir item
3714 * 1 for the dir index
3715 * 1 for the inode ref
3716 * 1 for the inode
3717 */
3718 trans = btrfs_start_transaction(root, 5);
3719 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3720 return trans;
3721
3722 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3723 return ERR_PTR(-ENOSPC);
3724
3725 /* check if there is someone else holds reference */
3726 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3727 return ERR_PTR(-ENOSPC);
3728
3729 if (atomic_read(&inode->i_count) > 2)
3730 return ERR_PTR(-ENOSPC);
3731
3732 if (xchg(&root->fs_info->enospc_unlink, 1))
3733 return ERR_PTR(-ENOSPC);
3734
3735 path = btrfs_alloc_path();
3736 if (!path) {
3737 root->fs_info->enospc_unlink = 0;
3738 return ERR_PTR(-ENOMEM);
3739 }
3740
3741 /* 1 for the orphan item */
3742 trans = btrfs_start_transaction(root, 1);
3743 if (IS_ERR(trans)) {
3744 btrfs_free_path(path);
3745 root->fs_info->enospc_unlink = 0;
3746 return trans;
3747 }
3748
3749 path->skip_locking = 1;
3750 path->search_commit_root = 1;
3751
3752 ret = btrfs_lookup_inode(trans, root, path,
3753 &BTRFS_I(dir)->location, 0);
3754 if (ret < 0) {
3755 err = ret;
3756 goto out;
3757 }
3758 if (ret == 0) {
3759 if (check_path_shared(root, path))
3760 goto out;
3761 } else {
3762 check_link = 0;
3763 }
3764 btrfs_release_path(path);
3765
3766 ret = btrfs_lookup_inode(trans, root, path,
3767 &BTRFS_I(inode)->location, 0);
3768 if (ret < 0) {
3769 err = ret;
3770 goto out;
3771 }
3772 if (ret == 0) {
3773 if (check_path_shared(root, path))
3774 goto out;
3775 } else {
3776 check_link = 0;
3777 }
3778 btrfs_release_path(path);
3779
3780 if (ret == 0 && S_ISREG(inode->i_mode)) {
3781 ret = btrfs_lookup_file_extent(trans, root, path,
3782 ino, (u64)-1, 0);
3783 if (ret < 0) {
3784 err = ret;
3785 goto out;
3786 }
3787 BUG_ON(ret == 0); /* Corruption */
3788 if (check_path_shared(root, path))
3789 goto out;
3790 btrfs_release_path(path);
3791 }
3792
3793 if (!check_link) {
3794 err = 0;
3795 goto out;
3796 }
3797
3798 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3799 dentry->d_name.name, dentry->d_name.len, 0);
3800 if (IS_ERR(di)) {
3801 err = PTR_ERR(di);
3802 goto out;
3803 }
3804 if (di) {
3805 if (check_path_shared(root, path))
3806 goto out;
3807 } else {
3808 err = 0;
3809 goto out;
3810 }
3811 btrfs_release_path(path);
3812
3813 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3814 dentry->d_name.len, ino, dir_ino, 0,
3815 &index);
3816 if (ret) {
3817 err = ret;
3818 goto out;
3819 }
3820
3821 if (check_path_shared(root, path))
3822 goto out;
3823
3824 btrfs_release_path(path);
3825
3826 /*
3827 * This is a commit root search, if we can lookup inode item and other
3828 * relative items in the commit root, it means the transaction of
3829 * dir/file creation has been committed, and the dir index item that we
3830 * delay to insert has also been inserted into the commit root. So
3831 * we needn't worry about the delayed insertion of the dir index item
3832 * here.
3833 */
3834 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3835 dentry->d_name.name, dentry->d_name.len, 0);
3836 if (IS_ERR(di)) {
3837 err = PTR_ERR(di);
3838 goto out;
3839 }
3840 BUG_ON(ret == -ENOENT);
3841 if (check_path_shared(root, path))
3842 goto out;
3843
3844 err = 0;
3845 out:
3846 btrfs_free_path(path);
3847 /* Migrate the orphan reservation over */
3848 if (!err)
3849 err = btrfs_block_rsv_migrate(trans->block_rsv,
3850 &root->fs_info->global_block_rsv,
3851 trans->bytes_reserved);
3852
3853 if (err) {
3854 btrfs_end_transaction(trans, root);
3855 root->fs_info->enospc_unlink = 0;
3856 return ERR_PTR(err);
3857 }
3858
3859 trans->block_rsv = &root->fs_info->global_block_rsv;
3860 return trans;
3861 }
3862
3863 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3864 struct btrfs_root *root)
3865 {
3866 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3867 btrfs_block_rsv_release(root, trans->block_rsv,
3868 trans->bytes_reserved);
3869 trans->block_rsv = &root->fs_info->trans_block_rsv;
3870 BUG_ON(!root->fs_info->enospc_unlink);
3871 root->fs_info->enospc_unlink = 0;
3872 }
3873 btrfs_end_transaction(trans, root);
3874 }
3875
3876 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3877 {
3878 struct btrfs_root *root = BTRFS_I(dir)->root;
3879 struct btrfs_trans_handle *trans;
3880 struct inode *inode = dentry->d_inode;
3881 int ret;
3882
3883 trans = __unlink_start_trans(dir, dentry);
3884 if (IS_ERR(trans))
3885 return PTR_ERR(trans);
3886
3887 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3888
3889 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3890 dentry->d_name.name, dentry->d_name.len);
3891 if (ret)
3892 goto out;
3893
3894 if (inode->i_nlink == 0) {
3895 ret = btrfs_orphan_add(trans, inode);
3896 if (ret)
3897 goto out;
3898 }
3899
3900 out:
3901 __unlink_end_trans(trans, root);
3902 btrfs_btree_balance_dirty(root);
3903 return ret;
3904 }
3905
3906 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3907 struct btrfs_root *root,
3908 struct inode *dir, u64 objectid,
3909 const char *name, int name_len)
3910 {
3911 struct btrfs_path *path;
3912 struct extent_buffer *leaf;
3913 struct btrfs_dir_item *di;
3914 struct btrfs_key key;
3915 u64 index;
3916 int ret;
3917 u64 dir_ino = btrfs_ino(dir);
3918
3919 path = btrfs_alloc_path();
3920 if (!path)
3921 return -ENOMEM;
3922
3923 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3924 name, name_len, -1);
3925 if (IS_ERR_OR_NULL(di)) {
3926 if (!di)
3927 ret = -ENOENT;
3928 else
3929 ret = PTR_ERR(di);
3930 goto out;
3931 }
3932
3933 leaf = path->nodes[0];
3934 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3935 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3936 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3937 if (ret) {
3938 btrfs_abort_transaction(trans, root, ret);
3939 goto out;
3940 }
3941 btrfs_release_path(path);
3942
3943 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3944 objectid, root->root_key.objectid,
3945 dir_ino, &index, name, name_len);
3946 if (ret < 0) {
3947 if (ret != -ENOENT) {
3948 btrfs_abort_transaction(trans, root, ret);
3949 goto out;
3950 }
3951 di = btrfs_search_dir_index_item(root, path, dir_ino,
3952 name, name_len);
3953 if (IS_ERR_OR_NULL(di)) {
3954 if (!di)
3955 ret = -ENOENT;
3956 else
3957 ret = PTR_ERR(di);
3958 btrfs_abort_transaction(trans, root, ret);
3959 goto out;
3960 }
3961
3962 leaf = path->nodes[0];
3963 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3964 btrfs_release_path(path);
3965 index = key.offset;
3966 }
3967 btrfs_release_path(path);
3968
3969 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3970 if (ret) {
3971 btrfs_abort_transaction(trans, root, ret);
3972 goto out;
3973 }
3974
3975 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3976 inode_inc_iversion(dir);
3977 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3978 ret = btrfs_update_inode_fallback(trans, root, dir);
3979 if (ret)
3980 btrfs_abort_transaction(trans, root, ret);
3981 out:
3982 btrfs_free_path(path);
3983 return ret;
3984 }
3985
3986 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3987 {
3988 struct inode *inode = dentry->d_inode;
3989 int err = 0;
3990 struct btrfs_root *root = BTRFS_I(dir)->root;
3991 struct btrfs_trans_handle *trans;
3992
3993 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3994 return -ENOTEMPTY;
3995 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3996 return -EPERM;
3997
3998 trans = __unlink_start_trans(dir, dentry);
3999 if (IS_ERR(trans))
4000 return PTR_ERR(trans);
4001
4002 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4003 err = btrfs_unlink_subvol(trans, root, dir,
4004 BTRFS_I(inode)->location.objectid,
4005 dentry->d_name.name,
4006 dentry->d_name.len);
4007 goto out;
4008 }
4009
4010 err = btrfs_orphan_add(trans, inode);
4011 if (err)
4012 goto out;
4013
4014 /* now the directory is empty */
4015 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4016 dentry->d_name.name, dentry->d_name.len);
4017 if (!err)
4018 btrfs_i_size_write(inode, 0);
4019 out:
4020 __unlink_end_trans(trans, root);
4021 btrfs_btree_balance_dirty(root);
4022
4023 return err;
4024 }
4025
4026 /*
4027 * this can truncate away extent items, csum items and directory items.
4028 * It starts at a high offset and removes keys until it can't find
4029 * any higher than new_size
4030 *
4031 * csum items that cross the new i_size are truncated to the new size
4032 * as well.
4033 *
4034 * min_type is the minimum key type to truncate down to. If set to 0, this
4035 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4036 */
4037 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4038 struct btrfs_root *root,
4039 struct inode *inode,
4040 u64 new_size, u32 min_type)
4041 {
4042 struct btrfs_path *path;
4043 struct extent_buffer *leaf;
4044 struct btrfs_file_extent_item *fi;
4045 struct btrfs_key key;
4046 struct btrfs_key found_key;
4047 u64 extent_start = 0;
4048 u64 extent_num_bytes = 0;
4049 u64 extent_offset = 0;
4050 u64 item_end = 0;
4051 u32 found_type = (u8)-1;
4052 int found_extent;
4053 int del_item;
4054 int pending_del_nr = 0;
4055 int pending_del_slot = 0;
4056 int extent_type = -1;
4057 int ret;
4058 int err = 0;
4059 u64 ino = btrfs_ino(inode);
4060
4061 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4062
4063 path = btrfs_alloc_path();
4064 if (!path)
4065 return -ENOMEM;
4066 path->reada = -1;
4067
4068 /*
4069 * We want to drop from the next block forward in case this new size is
4070 * not block aligned since we will be keeping the last block of the
4071 * extent just the way it is.
4072 */
4073 if (root->ref_cows || root == root->fs_info->tree_root)
4074 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4075 root->sectorsize), (u64)-1, 0);
4076
4077 /*
4078 * This function is also used to drop the items in the log tree before
4079 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4080 * it is used to drop the loged items. So we shouldn't kill the delayed
4081 * items.
4082 */
4083 if (min_type == 0 && root == BTRFS_I(inode)->root)
4084 btrfs_kill_delayed_inode_items(inode);
4085
4086 key.objectid = ino;
4087 key.offset = (u64)-1;
4088 key.type = (u8)-1;
4089
4090 search_again:
4091 path->leave_spinning = 1;
4092 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4093 if (ret < 0) {
4094 err = ret;
4095 goto out;
4096 }
4097
4098 if (ret > 0) {
4099 /* there are no items in the tree for us to truncate, we're
4100 * done
4101 */
4102 if (path->slots[0] == 0)
4103 goto out;
4104 path->slots[0]--;
4105 }
4106
4107 while (1) {
4108 fi = NULL;
4109 leaf = path->nodes[0];
4110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4111 found_type = btrfs_key_type(&found_key);
4112
4113 if (found_key.objectid != ino)
4114 break;
4115
4116 if (found_type < min_type)
4117 break;
4118
4119 item_end = found_key.offset;
4120 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4121 fi = btrfs_item_ptr(leaf, path->slots[0],
4122 struct btrfs_file_extent_item);
4123 extent_type = btrfs_file_extent_type(leaf, fi);
4124 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4125 item_end +=
4126 btrfs_file_extent_num_bytes(leaf, fi);
4127 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4128 item_end += btrfs_file_extent_inline_len(leaf,
4129 fi);
4130 }
4131 item_end--;
4132 }
4133 if (found_type > min_type) {
4134 del_item = 1;
4135 } else {
4136 if (item_end < new_size)
4137 break;
4138 if (found_key.offset >= new_size)
4139 del_item = 1;
4140 else
4141 del_item = 0;
4142 }
4143 found_extent = 0;
4144 /* FIXME, shrink the extent if the ref count is only 1 */
4145 if (found_type != BTRFS_EXTENT_DATA_KEY)
4146 goto delete;
4147
4148 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4149 u64 num_dec;
4150 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4151 if (!del_item) {
4152 u64 orig_num_bytes =
4153 btrfs_file_extent_num_bytes(leaf, fi);
4154 extent_num_bytes = ALIGN(new_size -
4155 found_key.offset,
4156 root->sectorsize);
4157 btrfs_set_file_extent_num_bytes(leaf, fi,
4158 extent_num_bytes);
4159 num_dec = (orig_num_bytes -
4160 extent_num_bytes);
4161 if (root->ref_cows && extent_start != 0)
4162 inode_sub_bytes(inode, num_dec);
4163 btrfs_mark_buffer_dirty(leaf);
4164 } else {
4165 extent_num_bytes =
4166 btrfs_file_extent_disk_num_bytes(leaf,
4167 fi);
4168 extent_offset = found_key.offset -
4169 btrfs_file_extent_offset(leaf, fi);
4170
4171 /* FIXME blocksize != 4096 */
4172 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4173 if (extent_start != 0) {
4174 found_extent = 1;
4175 if (root->ref_cows)
4176 inode_sub_bytes(inode, num_dec);
4177 }
4178 }
4179 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4180 /*
4181 * we can't truncate inline items that have had
4182 * special encodings
4183 */
4184 if (!del_item &&
4185 btrfs_file_extent_compression(leaf, fi) == 0 &&
4186 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4187 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4188 u32 size = new_size - found_key.offset;
4189
4190 if (root->ref_cows) {
4191 inode_sub_bytes(inode, item_end + 1 -
4192 new_size);
4193 }
4194 size =
4195 btrfs_file_extent_calc_inline_size(size);
4196 btrfs_truncate_item(root, path, size, 1);
4197 } else if (root->ref_cows) {
4198 inode_sub_bytes(inode, item_end + 1 -
4199 found_key.offset);
4200 }
4201 }
4202 delete:
4203 if (del_item) {
4204 if (!pending_del_nr) {
4205 /* no pending yet, add ourselves */
4206 pending_del_slot = path->slots[0];
4207 pending_del_nr = 1;
4208 } else if (pending_del_nr &&
4209 path->slots[0] + 1 == pending_del_slot) {
4210 /* hop on the pending chunk */
4211 pending_del_nr++;
4212 pending_del_slot = path->slots[0];
4213 } else {
4214 BUG();
4215 }
4216 } else {
4217 break;
4218 }
4219 if (found_extent && (root->ref_cows ||
4220 root == root->fs_info->tree_root)) {
4221 btrfs_set_path_blocking(path);
4222 ret = btrfs_free_extent(trans, root, extent_start,
4223 extent_num_bytes, 0,
4224 btrfs_header_owner(leaf),
4225 ino, extent_offset, 0);
4226 BUG_ON(ret);
4227 }
4228
4229 if (found_type == BTRFS_INODE_ITEM_KEY)
4230 break;
4231
4232 if (path->slots[0] == 0 ||
4233 path->slots[0] != pending_del_slot) {
4234 if (pending_del_nr) {
4235 ret = btrfs_del_items(trans, root, path,
4236 pending_del_slot,
4237 pending_del_nr);
4238 if (ret) {
4239 btrfs_abort_transaction(trans,
4240 root, ret);
4241 goto error;
4242 }
4243 pending_del_nr = 0;
4244 }
4245 btrfs_release_path(path);
4246 goto search_again;
4247 } else {
4248 path->slots[0]--;
4249 }
4250 }
4251 out:
4252 if (pending_del_nr) {
4253 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4254 pending_del_nr);
4255 if (ret)
4256 btrfs_abort_transaction(trans, root, ret);
4257 }
4258 error:
4259 btrfs_free_path(path);
4260 return err;
4261 }
4262
4263 /*
4264 * btrfs_truncate_page - read, zero a chunk and write a page
4265 * @inode - inode that we're zeroing
4266 * @from - the offset to start zeroing
4267 * @len - the length to zero, 0 to zero the entire range respective to the
4268 * offset
4269 * @front - zero up to the offset instead of from the offset on
4270 *
4271 * This will find the page for the "from" offset and cow the page and zero the
4272 * part we want to zero. This is used with truncate and hole punching.
4273 */
4274 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4275 int front)
4276 {
4277 struct address_space *mapping = inode->i_mapping;
4278 struct btrfs_root *root = BTRFS_I(inode)->root;
4279 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4280 struct btrfs_ordered_extent *ordered;
4281 struct extent_state *cached_state = NULL;
4282 char *kaddr;
4283 u32 blocksize = root->sectorsize;
4284 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4285 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4286 struct page *page;
4287 gfp_t mask = btrfs_alloc_write_mask(mapping);
4288 int ret = 0;
4289 u64 page_start;
4290 u64 page_end;
4291
4292 if ((offset & (blocksize - 1)) == 0 &&
4293 (!len || ((len & (blocksize - 1)) == 0)))
4294 goto out;
4295 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4296 if (ret)
4297 goto out;
4298
4299 again:
4300 page = find_or_create_page(mapping, index, mask);
4301 if (!page) {
4302 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4303 ret = -ENOMEM;
4304 goto out;
4305 }
4306
4307 page_start = page_offset(page);
4308 page_end = page_start + PAGE_CACHE_SIZE - 1;
4309
4310 if (!PageUptodate(page)) {
4311 ret = btrfs_readpage(NULL, page);
4312 lock_page(page);
4313 if (page->mapping != mapping) {
4314 unlock_page(page);
4315 page_cache_release(page);
4316 goto again;
4317 }
4318 if (!PageUptodate(page)) {
4319 ret = -EIO;
4320 goto out_unlock;
4321 }
4322 }
4323 wait_on_page_writeback(page);
4324
4325 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4326 set_page_extent_mapped(page);
4327
4328 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4329 if (ordered) {
4330 unlock_extent_cached(io_tree, page_start, page_end,
4331 &cached_state, GFP_NOFS);
4332 unlock_page(page);
4333 page_cache_release(page);
4334 btrfs_start_ordered_extent(inode, ordered, 1);
4335 btrfs_put_ordered_extent(ordered);
4336 goto again;
4337 }
4338
4339 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4340 EXTENT_DIRTY | EXTENT_DELALLOC |
4341 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4342 0, 0, &cached_state, GFP_NOFS);
4343
4344 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4345 &cached_state);
4346 if (ret) {
4347 unlock_extent_cached(io_tree, page_start, page_end,
4348 &cached_state, GFP_NOFS);
4349 goto out_unlock;
4350 }
4351
4352 if (offset != PAGE_CACHE_SIZE) {
4353 if (!len)
4354 len = PAGE_CACHE_SIZE - offset;
4355 kaddr = kmap(page);
4356 if (front)
4357 memset(kaddr, 0, offset);
4358 else
4359 memset(kaddr + offset, 0, len);
4360 flush_dcache_page(page);
4361 kunmap(page);
4362 }
4363 ClearPageChecked(page);
4364 set_page_dirty(page);
4365 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4366 GFP_NOFS);
4367
4368 out_unlock:
4369 if (ret)
4370 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4371 unlock_page(page);
4372 page_cache_release(page);
4373 out:
4374 return ret;
4375 }
4376
4377 /*
4378 * This function puts in dummy file extents for the area we're creating a hole
4379 * for. So if we are truncating this file to a larger size we need to insert
4380 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4381 * the range between oldsize and size
4382 */
4383 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4384 {
4385 struct btrfs_trans_handle *trans;
4386 struct btrfs_root *root = BTRFS_I(inode)->root;
4387 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4388 struct extent_map *em = NULL;
4389 struct extent_state *cached_state = NULL;
4390 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4391 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4392 u64 block_end = ALIGN(size, root->sectorsize);
4393 u64 last_byte;
4394 u64 cur_offset;
4395 u64 hole_size;
4396 int err = 0;
4397
4398 if (size <= hole_start)
4399 return 0;
4400
4401 while (1) {
4402 struct btrfs_ordered_extent *ordered;
4403 btrfs_wait_ordered_range(inode, hole_start,
4404 block_end - hole_start);
4405 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4406 &cached_state);
4407 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4408 if (!ordered)
4409 break;
4410 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4411 &cached_state, GFP_NOFS);
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 trans = btrfs_start_transaction(root, 3);
4431 if (IS_ERR(trans)) {
4432 err = PTR_ERR(trans);
4433 break;
4434 }
4435
4436 err = btrfs_drop_extents(trans, root, inode,
4437 cur_offset,
4438 cur_offset + hole_size, 1);
4439 if (err) {
4440 btrfs_abort_transaction(trans, root, err);
4441 btrfs_end_transaction(trans, root);
4442 break;
4443 }
4444
4445 err = btrfs_insert_file_extent(trans, root,
4446 btrfs_ino(inode), cur_offset, 0,
4447 0, hole_size, 0, hole_size,
4448 0, 0, 0);
4449 if (err) {
4450 btrfs_abort_transaction(trans, root, err);
4451 btrfs_end_transaction(trans, root);
4452 break;
4453 }
4454
4455 btrfs_drop_extent_cache(inode, cur_offset,
4456 cur_offset + hole_size - 1, 0);
4457 hole_em = alloc_extent_map();
4458 if (!hole_em) {
4459 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4460 &BTRFS_I(inode)->runtime_flags);
4461 goto next;
4462 }
4463 hole_em->start = cur_offset;
4464 hole_em->len = hole_size;
4465 hole_em->orig_start = cur_offset;
4466
4467 hole_em->block_start = EXTENT_MAP_HOLE;
4468 hole_em->block_len = 0;
4469 hole_em->orig_block_len = 0;
4470 hole_em->ram_bytes = hole_size;
4471 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4472 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4473 hole_em->generation = trans->transid;
4474
4475 while (1) {
4476 write_lock(&em_tree->lock);
4477 err = add_extent_mapping(em_tree, hole_em, 1);
4478 write_unlock(&em_tree->lock);
4479 if (err != -EEXIST)
4480 break;
4481 btrfs_drop_extent_cache(inode, cur_offset,
4482 cur_offset +
4483 hole_size - 1, 0);
4484 }
4485 free_extent_map(hole_em);
4486 next:
4487 btrfs_update_inode(trans, root, inode);
4488 btrfs_end_transaction(trans, root);
4489 }
4490 free_extent_map(em);
4491 em = NULL;
4492 cur_offset = last_byte;
4493 if (cur_offset >= block_end)
4494 break;
4495 }
4496
4497 free_extent_map(em);
4498 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4499 GFP_NOFS);
4500 return err;
4501 }
4502
4503 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4504 {
4505 struct btrfs_root *root = BTRFS_I(inode)->root;
4506 struct btrfs_trans_handle *trans;
4507 loff_t oldsize = i_size_read(inode);
4508 loff_t newsize = attr->ia_size;
4509 int mask = attr->ia_valid;
4510 int ret;
4511
4512 if (newsize == oldsize)
4513 return 0;
4514
4515 /*
4516 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4517 * special case where we need to update the times despite not having
4518 * these flags set. For all other operations the VFS set these flags
4519 * explicitly if it wants a timestamp update.
4520 */
4521 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4522 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4523
4524 if (newsize > oldsize) {
4525 truncate_pagecache(inode, oldsize, newsize);
4526 ret = btrfs_cont_expand(inode, oldsize, newsize);
4527 if (ret)
4528 return ret;
4529
4530 trans = btrfs_start_transaction(root, 1);
4531 if (IS_ERR(trans))
4532 return PTR_ERR(trans);
4533
4534 i_size_write(inode, newsize);
4535 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4536 ret = btrfs_update_inode(trans, root, inode);
4537 btrfs_end_transaction(trans, root);
4538 } else {
4539
4540 /*
4541 * We're truncating a file that used to have good data down to
4542 * zero. Make sure it gets into the ordered flush list so that
4543 * any new writes get down to disk quickly.
4544 */
4545 if (newsize == 0)
4546 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4547 &BTRFS_I(inode)->runtime_flags);
4548
4549 /*
4550 * 1 for the orphan item we're going to add
4551 * 1 for the orphan item deletion.
4552 */
4553 trans = btrfs_start_transaction(root, 2);
4554 if (IS_ERR(trans))
4555 return PTR_ERR(trans);
4556
4557 /*
4558 * We need to do this in case we fail at _any_ point during the
4559 * actual truncate. Once we do the truncate_setsize we could
4560 * invalidate pages which forces any outstanding ordered io to
4561 * be instantly completed which will give us extents that need
4562 * to be truncated. If we fail to get an orphan inode down we
4563 * could have left over extents that were never meant to live,
4564 * so we need to garuntee from this point on that everything
4565 * will be consistent.
4566 */
4567 ret = btrfs_orphan_add(trans, inode);
4568 btrfs_end_transaction(trans, root);
4569 if (ret)
4570 return ret;
4571
4572 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4573 truncate_setsize(inode, newsize);
4574
4575 /* Disable nonlocked read DIO to avoid the end less truncate */
4576 btrfs_inode_block_unlocked_dio(inode);
4577 inode_dio_wait(inode);
4578 btrfs_inode_resume_unlocked_dio(inode);
4579
4580 ret = btrfs_truncate(inode);
4581 if (ret && inode->i_nlink)
4582 btrfs_orphan_del(NULL, inode);
4583 }
4584
4585 return ret;
4586 }
4587
4588 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4589 {
4590 struct inode *inode = dentry->d_inode;
4591 struct btrfs_root *root = BTRFS_I(inode)->root;
4592 int err;
4593
4594 if (btrfs_root_readonly(root))
4595 return -EROFS;
4596
4597 err = inode_change_ok(inode, attr);
4598 if (err)
4599 return err;
4600
4601 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4602 err = btrfs_setsize(inode, attr);
4603 if (err)
4604 return err;
4605 }
4606
4607 if (attr->ia_valid) {
4608 setattr_copy(inode, attr);
4609 inode_inc_iversion(inode);
4610 err = btrfs_dirty_inode(inode);
4611
4612 if (!err && attr->ia_valid & ATTR_MODE)
4613 err = btrfs_acl_chmod(inode);
4614 }
4615
4616 return err;
4617 }
4618
4619 void btrfs_evict_inode(struct inode *inode)
4620 {
4621 struct btrfs_trans_handle *trans;
4622 struct btrfs_root *root = BTRFS_I(inode)->root;
4623 struct btrfs_block_rsv *rsv, *global_rsv;
4624 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4625 int ret;
4626
4627 trace_btrfs_inode_evict(inode);
4628
4629 truncate_inode_pages(&inode->i_data, 0);
4630 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4631 btrfs_is_free_space_inode(inode)))
4632 goto no_delete;
4633
4634 if (is_bad_inode(inode)) {
4635 btrfs_orphan_del(NULL, inode);
4636 goto no_delete;
4637 }
4638 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4639 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4640
4641 if (root->fs_info->log_root_recovering) {
4642 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4643 &BTRFS_I(inode)->runtime_flags));
4644 goto no_delete;
4645 }
4646
4647 if (inode->i_nlink > 0) {
4648 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4649 goto no_delete;
4650 }
4651
4652 ret = btrfs_commit_inode_delayed_inode(inode);
4653 if (ret) {
4654 btrfs_orphan_del(NULL, inode);
4655 goto no_delete;
4656 }
4657
4658 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4659 if (!rsv) {
4660 btrfs_orphan_del(NULL, inode);
4661 goto no_delete;
4662 }
4663 rsv->size = min_size;
4664 rsv->failfast = 1;
4665 global_rsv = &root->fs_info->global_block_rsv;
4666
4667 btrfs_i_size_write(inode, 0);
4668
4669 /*
4670 * This is a bit simpler than btrfs_truncate since we've already
4671 * reserved our space for our orphan item in the unlink, so we just
4672 * need to reserve some slack space in case we add bytes and update
4673 * inode item when doing the truncate.
4674 */
4675 while (1) {
4676 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4677 BTRFS_RESERVE_FLUSH_LIMIT);
4678
4679 /*
4680 * Try and steal from the global reserve since we will
4681 * likely not use this space anyway, we want to try as
4682 * hard as possible to get this to work.
4683 */
4684 if (ret)
4685 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4686
4687 if (ret) {
4688 btrfs_warn(root->fs_info,
4689 "Could not get space for a delete, will truncate on mount %d",
4690 ret);
4691 btrfs_orphan_del(NULL, inode);
4692 btrfs_free_block_rsv(root, rsv);
4693 goto no_delete;
4694 }
4695
4696 trans = btrfs_join_transaction(root);
4697 if (IS_ERR(trans)) {
4698 btrfs_orphan_del(NULL, inode);
4699 btrfs_free_block_rsv(root, rsv);
4700 goto no_delete;
4701 }
4702
4703 trans->block_rsv = rsv;
4704
4705 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4706 if (ret != -ENOSPC)
4707 break;
4708
4709 trans->block_rsv = &root->fs_info->trans_block_rsv;
4710 btrfs_end_transaction(trans, root);
4711 trans = NULL;
4712 btrfs_btree_balance_dirty(root);
4713 }
4714
4715 btrfs_free_block_rsv(root, rsv);
4716
4717 if (ret == 0) {
4718 trans->block_rsv = root->orphan_block_rsv;
4719 ret = btrfs_orphan_del(trans, inode);
4720 BUG_ON(ret);
4721 }
4722
4723 trans->block_rsv = &root->fs_info->trans_block_rsv;
4724 if (!(root == root->fs_info->tree_root ||
4725 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4726 btrfs_return_ino(root, btrfs_ino(inode));
4727
4728 btrfs_end_transaction(trans, root);
4729 btrfs_btree_balance_dirty(root);
4730 no_delete:
4731 btrfs_remove_delayed_node(inode);
4732 clear_inode(inode);
4733 return;
4734 }
4735
4736 /*
4737 * this returns the key found in the dir entry in the location pointer.
4738 * If no dir entries were found, location->objectid is 0.
4739 */
4740 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4741 struct btrfs_key *location)
4742 {
4743 const char *name = dentry->d_name.name;
4744 int namelen = dentry->d_name.len;
4745 struct btrfs_dir_item *di;
4746 struct btrfs_path *path;
4747 struct btrfs_root *root = BTRFS_I(dir)->root;
4748 int ret = 0;
4749
4750 path = btrfs_alloc_path();
4751 if (!path)
4752 return -ENOMEM;
4753
4754 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4755 namelen, 0);
4756 if (IS_ERR(di))
4757 ret = PTR_ERR(di);
4758
4759 if (IS_ERR_OR_NULL(di))
4760 goto out_err;
4761
4762 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4763 out:
4764 btrfs_free_path(path);
4765 return ret;
4766 out_err:
4767 location->objectid = 0;
4768 goto out;
4769 }
4770
4771 /*
4772 * when we hit a tree root in a directory, the btrfs part of the inode
4773 * needs to be changed to reflect the root directory of the tree root. This
4774 * is kind of like crossing a mount point.
4775 */
4776 static int fixup_tree_root_location(struct btrfs_root *root,
4777 struct inode *dir,
4778 struct dentry *dentry,
4779 struct btrfs_key *location,
4780 struct btrfs_root **sub_root)
4781 {
4782 struct btrfs_path *path;
4783 struct btrfs_root *new_root;
4784 struct btrfs_root_ref *ref;
4785 struct extent_buffer *leaf;
4786 int ret;
4787 int err = 0;
4788
4789 path = btrfs_alloc_path();
4790 if (!path) {
4791 err = -ENOMEM;
4792 goto out;
4793 }
4794
4795 err = -ENOENT;
4796 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4797 BTRFS_I(dir)->root->root_key.objectid,
4798 location->objectid);
4799 if (ret) {
4800 if (ret < 0)
4801 err = ret;
4802 goto out;
4803 }
4804
4805 leaf = path->nodes[0];
4806 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4807 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4808 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4809 goto out;
4810
4811 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4812 (unsigned long)(ref + 1),
4813 dentry->d_name.len);
4814 if (ret)
4815 goto out;
4816
4817 btrfs_release_path(path);
4818
4819 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4820 if (IS_ERR(new_root)) {
4821 err = PTR_ERR(new_root);
4822 goto out;
4823 }
4824
4825 if (btrfs_root_refs(&new_root->root_item) == 0) {
4826 err = -ENOENT;
4827 goto out;
4828 }
4829
4830 *sub_root = new_root;
4831 location->objectid = btrfs_root_dirid(&new_root->root_item);
4832 location->type = BTRFS_INODE_ITEM_KEY;
4833 location->offset = 0;
4834 err = 0;
4835 out:
4836 btrfs_free_path(path);
4837 return err;
4838 }
4839
4840 static void inode_tree_add(struct inode *inode)
4841 {
4842 struct btrfs_root *root = BTRFS_I(inode)->root;
4843 struct btrfs_inode *entry;
4844 struct rb_node **p;
4845 struct rb_node *parent;
4846 u64 ino = btrfs_ino(inode);
4847
4848 if (inode_unhashed(inode))
4849 return;
4850 again:
4851 parent = NULL;
4852 spin_lock(&root->inode_lock);
4853 p = &root->inode_tree.rb_node;
4854 while (*p) {
4855 parent = *p;
4856 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4857
4858 if (ino < btrfs_ino(&entry->vfs_inode))
4859 p = &parent->rb_left;
4860 else if (ino > btrfs_ino(&entry->vfs_inode))
4861 p = &parent->rb_right;
4862 else {
4863 WARN_ON(!(entry->vfs_inode.i_state &
4864 (I_WILL_FREE | I_FREEING)));
4865 rb_erase(parent, &root->inode_tree);
4866 RB_CLEAR_NODE(parent);
4867 spin_unlock(&root->inode_lock);
4868 goto again;
4869 }
4870 }
4871 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4872 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4873 spin_unlock(&root->inode_lock);
4874 }
4875
4876 static void inode_tree_del(struct inode *inode)
4877 {
4878 struct btrfs_root *root = BTRFS_I(inode)->root;
4879 int empty = 0;
4880
4881 spin_lock(&root->inode_lock);
4882 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4883 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4884 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4885 empty = RB_EMPTY_ROOT(&root->inode_tree);
4886 }
4887 spin_unlock(&root->inode_lock);
4888
4889 /*
4890 * Free space cache has inodes in the tree root, but the tree root has a
4891 * root_refs of 0, so this could end up dropping the tree root as a
4892 * snapshot, so we need the extra !root->fs_info->tree_root check to
4893 * make sure we don't drop it.
4894 */
4895 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4896 root != root->fs_info->tree_root) {
4897 synchronize_srcu(&root->fs_info->subvol_srcu);
4898 spin_lock(&root->inode_lock);
4899 empty = RB_EMPTY_ROOT(&root->inode_tree);
4900 spin_unlock(&root->inode_lock);
4901 if (empty)
4902 btrfs_add_dead_root(root);
4903 }
4904 }
4905
4906 void btrfs_invalidate_inodes(struct btrfs_root *root)
4907 {
4908 struct rb_node *node;
4909 struct rb_node *prev;
4910 struct btrfs_inode *entry;
4911 struct inode *inode;
4912 u64 objectid = 0;
4913
4914 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4915
4916 spin_lock(&root->inode_lock);
4917 again:
4918 node = root->inode_tree.rb_node;
4919 prev = NULL;
4920 while (node) {
4921 prev = node;
4922 entry = rb_entry(node, struct btrfs_inode, rb_node);
4923
4924 if (objectid < btrfs_ino(&entry->vfs_inode))
4925 node = node->rb_left;
4926 else if (objectid > btrfs_ino(&entry->vfs_inode))
4927 node = node->rb_right;
4928 else
4929 break;
4930 }
4931 if (!node) {
4932 while (prev) {
4933 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4934 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4935 node = prev;
4936 break;
4937 }
4938 prev = rb_next(prev);
4939 }
4940 }
4941 while (node) {
4942 entry = rb_entry(node, struct btrfs_inode, rb_node);
4943 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4944 inode = igrab(&entry->vfs_inode);
4945 if (inode) {
4946 spin_unlock(&root->inode_lock);
4947 if (atomic_read(&inode->i_count) > 1)
4948 d_prune_aliases(inode);
4949 /*
4950 * btrfs_drop_inode will have it removed from
4951 * the inode cache when its usage count
4952 * hits zero.
4953 */
4954 iput(inode);
4955 cond_resched();
4956 spin_lock(&root->inode_lock);
4957 goto again;
4958 }
4959
4960 if (cond_resched_lock(&root->inode_lock))
4961 goto again;
4962
4963 node = rb_next(node);
4964 }
4965 spin_unlock(&root->inode_lock);
4966 }
4967
4968 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4969 {
4970 struct btrfs_iget_args *args = p;
4971 inode->i_ino = args->ino;
4972 BTRFS_I(inode)->root = args->root;
4973 return 0;
4974 }
4975
4976 static int btrfs_find_actor(struct inode *inode, void *opaque)
4977 {
4978 struct btrfs_iget_args *args = opaque;
4979 return args->ino == btrfs_ino(inode) &&
4980 args->root == BTRFS_I(inode)->root;
4981 }
4982
4983 static struct inode *btrfs_iget_locked(struct super_block *s,
4984 u64 objectid,
4985 struct btrfs_root *root)
4986 {
4987 struct inode *inode;
4988 struct btrfs_iget_args args;
4989 args.ino = objectid;
4990 args.root = root;
4991
4992 inode = iget5_locked(s, objectid, btrfs_find_actor,
4993 btrfs_init_locked_inode,
4994 (void *)&args);
4995 return inode;
4996 }
4997
4998 /* Get an inode object given its location and corresponding root.
4999 * Returns in *is_new if the inode was read from disk
5000 */
5001 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5002 struct btrfs_root *root, int *new)
5003 {
5004 struct inode *inode;
5005
5006 inode = btrfs_iget_locked(s, location->objectid, root);
5007 if (!inode)
5008 return ERR_PTR(-ENOMEM);
5009
5010 if (inode->i_state & I_NEW) {
5011 BTRFS_I(inode)->root = root;
5012 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5013 btrfs_read_locked_inode(inode);
5014 if (!is_bad_inode(inode)) {
5015 inode_tree_add(inode);
5016 unlock_new_inode(inode);
5017 if (new)
5018 *new = 1;
5019 } else {
5020 unlock_new_inode(inode);
5021 iput(inode);
5022 inode = ERR_PTR(-ESTALE);
5023 }
5024 }
5025
5026 return inode;
5027 }
5028
5029 static struct inode *new_simple_dir(struct super_block *s,
5030 struct btrfs_key *key,
5031 struct btrfs_root *root)
5032 {
5033 struct inode *inode = new_inode(s);
5034
5035 if (!inode)
5036 return ERR_PTR(-ENOMEM);
5037
5038 BTRFS_I(inode)->root = root;
5039 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5040 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5041
5042 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5043 inode->i_op = &btrfs_dir_ro_inode_operations;
5044 inode->i_fop = &simple_dir_operations;
5045 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5046 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5047
5048 return inode;
5049 }
5050
5051 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5052 {
5053 struct inode *inode;
5054 struct btrfs_root *root = BTRFS_I(dir)->root;
5055 struct btrfs_root *sub_root = root;
5056 struct btrfs_key location;
5057 int index;
5058 int ret = 0;
5059
5060 if (dentry->d_name.len > BTRFS_NAME_LEN)
5061 return ERR_PTR(-ENAMETOOLONG);
5062
5063 ret = btrfs_inode_by_name(dir, dentry, &location);
5064 if (ret < 0)
5065 return ERR_PTR(ret);
5066
5067 if (location.objectid == 0)
5068 return NULL;
5069
5070 if (location.type == BTRFS_INODE_ITEM_KEY) {
5071 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5072 return inode;
5073 }
5074
5075 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5076
5077 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5078 ret = fixup_tree_root_location(root, dir, dentry,
5079 &location, &sub_root);
5080 if (ret < 0) {
5081 if (ret != -ENOENT)
5082 inode = ERR_PTR(ret);
5083 else
5084 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5085 } else {
5086 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5087 }
5088 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5089
5090 if (!IS_ERR(inode) && root != sub_root) {
5091 down_read(&root->fs_info->cleanup_work_sem);
5092 if (!(inode->i_sb->s_flags & MS_RDONLY))
5093 ret = btrfs_orphan_cleanup(sub_root);
5094 up_read(&root->fs_info->cleanup_work_sem);
5095 if (ret)
5096 inode = ERR_PTR(ret);
5097 }
5098
5099 return inode;
5100 }
5101
5102 static int btrfs_dentry_delete(const struct dentry *dentry)
5103 {
5104 struct btrfs_root *root;
5105 struct inode *inode = dentry->d_inode;
5106
5107 if (!inode && !IS_ROOT(dentry))
5108 inode = dentry->d_parent->d_inode;
5109
5110 if (inode) {
5111 root = BTRFS_I(inode)->root;
5112 if (btrfs_root_refs(&root->root_item) == 0)
5113 return 1;
5114
5115 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5116 return 1;
5117 }
5118 return 0;
5119 }
5120
5121 static void btrfs_dentry_release(struct dentry *dentry)
5122 {
5123 if (dentry->d_fsdata)
5124 kfree(dentry->d_fsdata);
5125 }
5126
5127 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5128 unsigned int flags)
5129 {
5130 struct dentry *ret;
5131
5132 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5133 return ret;
5134 }
5135
5136 unsigned char btrfs_filetype_table[] = {
5137 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5138 };
5139
5140 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5141 {
5142 struct inode *inode = file_inode(file);
5143 struct btrfs_root *root = BTRFS_I(inode)->root;
5144 struct btrfs_item *item;
5145 struct btrfs_dir_item *di;
5146 struct btrfs_key key;
5147 struct btrfs_key found_key;
5148 struct btrfs_path *path;
5149 struct list_head ins_list;
5150 struct list_head del_list;
5151 int ret;
5152 struct extent_buffer *leaf;
5153 int slot;
5154 unsigned char d_type;
5155 int over = 0;
5156 u32 di_cur;
5157 u32 di_total;
5158 u32 di_len;
5159 int key_type = BTRFS_DIR_INDEX_KEY;
5160 char tmp_name[32];
5161 char *name_ptr;
5162 int name_len;
5163 int is_curr = 0; /* ctx->pos points to the current index? */
5164
5165 /* FIXME, use a real flag for deciding about the key type */
5166 if (root->fs_info->tree_root == root)
5167 key_type = BTRFS_DIR_ITEM_KEY;
5168
5169 if (!dir_emit_dots(file, ctx))
5170 return 0;
5171
5172 path = btrfs_alloc_path();
5173 if (!path)
5174 return -ENOMEM;
5175
5176 path->reada = 1;
5177
5178 if (key_type == BTRFS_DIR_INDEX_KEY) {
5179 INIT_LIST_HEAD(&ins_list);
5180 INIT_LIST_HEAD(&del_list);
5181 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5182 }
5183
5184 btrfs_set_key_type(&key, key_type);
5185 key.offset = ctx->pos;
5186 key.objectid = btrfs_ino(inode);
5187
5188 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5189 if (ret < 0)
5190 goto err;
5191
5192 while (1) {
5193 leaf = path->nodes[0];
5194 slot = path->slots[0];
5195 if (slot >= btrfs_header_nritems(leaf)) {
5196 ret = btrfs_next_leaf(root, path);
5197 if (ret < 0)
5198 goto err;
5199 else if (ret > 0)
5200 break;
5201 continue;
5202 }
5203
5204 item = btrfs_item_nr(leaf, slot);
5205 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5206
5207 if (found_key.objectid != key.objectid)
5208 break;
5209 if (btrfs_key_type(&found_key) != key_type)
5210 break;
5211 if (found_key.offset < ctx->pos)
5212 goto next;
5213 if (key_type == BTRFS_DIR_INDEX_KEY &&
5214 btrfs_should_delete_dir_index(&del_list,
5215 found_key.offset))
5216 goto next;
5217
5218 ctx->pos = found_key.offset;
5219 is_curr = 1;
5220
5221 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5222 di_cur = 0;
5223 di_total = btrfs_item_size(leaf, item);
5224
5225 while (di_cur < di_total) {
5226 struct btrfs_key location;
5227
5228 if (verify_dir_item(root, leaf, di))
5229 break;
5230
5231 name_len = btrfs_dir_name_len(leaf, di);
5232 if (name_len <= sizeof(tmp_name)) {
5233 name_ptr = tmp_name;
5234 } else {
5235 name_ptr = kmalloc(name_len, GFP_NOFS);
5236 if (!name_ptr) {
5237 ret = -ENOMEM;
5238 goto err;
5239 }
5240 }
5241 read_extent_buffer(leaf, name_ptr,
5242 (unsigned long)(di + 1), name_len);
5243
5244 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5245 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5246
5247
5248 /* is this a reference to our own snapshot? If so
5249 * skip it.
5250 *
5251 * In contrast to old kernels, we insert the snapshot's
5252 * dir item and dir index after it has been created, so
5253 * we won't find a reference to our own snapshot. We
5254 * still keep the following code for backward
5255 * compatibility.
5256 */
5257 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5258 location.objectid == root->root_key.objectid) {
5259 over = 0;
5260 goto skip;
5261 }
5262 over = !dir_emit(ctx, name_ptr, name_len,
5263 location.objectid, d_type);
5264
5265 skip:
5266 if (name_ptr != tmp_name)
5267 kfree(name_ptr);
5268
5269 if (over)
5270 goto nopos;
5271 di_len = btrfs_dir_name_len(leaf, di) +
5272 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5273 di_cur += di_len;
5274 di = (struct btrfs_dir_item *)((char *)di + di_len);
5275 }
5276 next:
5277 path->slots[0]++;
5278 }
5279
5280 if (key_type == BTRFS_DIR_INDEX_KEY) {
5281 if (is_curr)
5282 ctx->pos++;
5283 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5284 if (ret)
5285 goto nopos;
5286 }
5287
5288 /* Reached end of directory/root. Bump pos past the last item. */
5289 if (key_type == BTRFS_DIR_INDEX_KEY)
5290 /*
5291 * 32-bit glibc will use getdents64, but then strtol -
5292 * so the last number we can serve is this.
5293 */
5294 ctx->pos = 0x7fffffff;
5295 else
5296 ctx->pos++;
5297 nopos:
5298 ret = 0;
5299 err:
5300 if (key_type == BTRFS_DIR_INDEX_KEY)
5301 btrfs_put_delayed_items(&ins_list, &del_list);
5302 btrfs_free_path(path);
5303 return ret;
5304 }
5305
5306 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5307 {
5308 struct btrfs_root *root = BTRFS_I(inode)->root;
5309 struct btrfs_trans_handle *trans;
5310 int ret = 0;
5311 bool nolock = false;
5312
5313 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5314 return 0;
5315
5316 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5317 nolock = true;
5318
5319 if (wbc->sync_mode == WB_SYNC_ALL) {
5320 if (nolock)
5321 trans = btrfs_join_transaction_nolock(root);
5322 else
5323 trans = btrfs_join_transaction(root);
5324 if (IS_ERR(trans))
5325 return PTR_ERR(trans);
5326 ret = btrfs_commit_transaction(trans, root);
5327 }
5328 return ret;
5329 }
5330
5331 /*
5332 * This is somewhat expensive, updating the tree every time the
5333 * inode changes. But, it is most likely to find the inode in cache.
5334 * FIXME, needs more benchmarking...there are no reasons other than performance
5335 * to keep or drop this code.
5336 */
5337 static int btrfs_dirty_inode(struct inode *inode)
5338 {
5339 struct btrfs_root *root = BTRFS_I(inode)->root;
5340 struct btrfs_trans_handle *trans;
5341 int ret;
5342
5343 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5344 return 0;
5345
5346 trans = btrfs_join_transaction(root);
5347 if (IS_ERR(trans))
5348 return PTR_ERR(trans);
5349
5350 ret = btrfs_update_inode(trans, root, inode);
5351 if (ret && ret == -ENOSPC) {
5352 /* whoops, lets try again with the full transaction */
5353 btrfs_end_transaction(trans, root);
5354 trans = btrfs_start_transaction(root, 1);
5355 if (IS_ERR(trans))
5356 return PTR_ERR(trans);
5357
5358 ret = btrfs_update_inode(trans, root, inode);
5359 }
5360 btrfs_end_transaction(trans, root);
5361 if (BTRFS_I(inode)->delayed_node)
5362 btrfs_balance_delayed_items(root);
5363
5364 return ret;
5365 }
5366
5367 /*
5368 * This is a copy of file_update_time. We need this so we can return error on
5369 * ENOSPC for updating the inode in the case of file write and mmap writes.
5370 */
5371 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5372 int flags)
5373 {
5374 struct btrfs_root *root = BTRFS_I(inode)->root;
5375
5376 if (btrfs_root_readonly(root))
5377 return -EROFS;
5378
5379 if (flags & S_VERSION)
5380 inode_inc_iversion(inode);
5381 if (flags & S_CTIME)
5382 inode->i_ctime = *now;
5383 if (flags & S_MTIME)
5384 inode->i_mtime = *now;
5385 if (flags & S_ATIME)
5386 inode->i_atime = *now;
5387 return btrfs_dirty_inode(inode);
5388 }
5389
5390 /*
5391 * find the highest existing sequence number in a directory
5392 * and then set the in-memory index_cnt variable to reflect
5393 * free sequence numbers
5394 */
5395 static int btrfs_set_inode_index_count(struct inode *inode)
5396 {
5397 struct btrfs_root *root = BTRFS_I(inode)->root;
5398 struct btrfs_key key, found_key;
5399 struct btrfs_path *path;
5400 struct extent_buffer *leaf;
5401 int ret;
5402
5403 key.objectid = btrfs_ino(inode);
5404 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5405 key.offset = (u64)-1;
5406
5407 path = btrfs_alloc_path();
5408 if (!path)
5409 return -ENOMEM;
5410
5411 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5412 if (ret < 0)
5413 goto out;
5414 /* FIXME: we should be able to handle this */
5415 if (ret == 0)
5416 goto out;
5417 ret = 0;
5418
5419 /*
5420 * MAGIC NUMBER EXPLANATION:
5421 * since we search a directory based on f_pos we have to start at 2
5422 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5423 * else has to start at 2
5424 */
5425 if (path->slots[0] == 0) {
5426 BTRFS_I(inode)->index_cnt = 2;
5427 goto out;
5428 }
5429
5430 path->slots[0]--;
5431
5432 leaf = path->nodes[0];
5433 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5434
5435 if (found_key.objectid != btrfs_ino(inode) ||
5436 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5437 BTRFS_I(inode)->index_cnt = 2;
5438 goto out;
5439 }
5440
5441 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5442 out:
5443 btrfs_free_path(path);
5444 return ret;
5445 }
5446
5447 /*
5448 * helper to find a free sequence number in a given directory. This current
5449 * code is very simple, later versions will do smarter things in the btree
5450 */
5451 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5452 {
5453 int ret = 0;
5454
5455 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5456 ret = btrfs_inode_delayed_dir_index_count(dir);
5457 if (ret) {
5458 ret = btrfs_set_inode_index_count(dir);
5459 if (ret)
5460 return ret;
5461 }
5462 }
5463
5464 *index = BTRFS_I(dir)->index_cnt;
5465 BTRFS_I(dir)->index_cnt++;
5466
5467 return ret;
5468 }
5469
5470 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5471 struct btrfs_root *root,
5472 struct inode *dir,
5473 const char *name, int name_len,
5474 u64 ref_objectid, u64 objectid,
5475 umode_t mode, u64 *index)
5476 {
5477 struct inode *inode;
5478 struct btrfs_inode_item *inode_item;
5479 struct btrfs_key *location;
5480 struct btrfs_path *path;
5481 struct btrfs_inode_ref *ref;
5482 struct btrfs_key key[2];
5483 u32 sizes[2];
5484 unsigned long ptr;
5485 int ret;
5486 int owner;
5487
5488 path = btrfs_alloc_path();
5489 if (!path)
5490 return ERR_PTR(-ENOMEM);
5491
5492 inode = new_inode(root->fs_info->sb);
5493 if (!inode) {
5494 btrfs_free_path(path);
5495 return ERR_PTR(-ENOMEM);
5496 }
5497
5498 /*
5499 * we have to initialize this early, so we can reclaim the inode
5500 * number if we fail afterwards in this function.
5501 */
5502 inode->i_ino = objectid;
5503
5504 if (dir) {
5505 trace_btrfs_inode_request(dir);
5506
5507 ret = btrfs_set_inode_index(dir, index);
5508 if (ret) {
5509 btrfs_free_path(path);
5510 iput(inode);
5511 return ERR_PTR(ret);
5512 }
5513 }
5514 /*
5515 * index_cnt is ignored for everything but a dir,
5516 * btrfs_get_inode_index_count has an explanation for the magic
5517 * number
5518 */
5519 BTRFS_I(inode)->index_cnt = 2;
5520 BTRFS_I(inode)->root = root;
5521 BTRFS_I(inode)->generation = trans->transid;
5522 inode->i_generation = BTRFS_I(inode)->generation;
5523
5524 /*
5525 * We could have gotten an inode number from somebody who was fsynced
5526 * and then removed in this same transaction, so let's just set full
5527 * sync since it will be a full sync anyway and this will blow away the
5528 * old info in the log.
5529 */
5530 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5531
5532 if (S_ISDIR(mode))
5533 owner = 0;
5534 else
5535 owner = 1;
5536
5537 key[0].objectid = objectid;
5538 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5539 key[0].offset = 0;
5540
5541 /*
5542 * Start new inodes with an inode_ref. This is slightly more
5543 * efficient for small numbers of hard links since they will
5544 * be packed into one item. Extended refs will kick in if we
5545 * add more hard links than can fit in the ref item.
5546 */
5547 key[1].objectid = objectid;
5548 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5549 key[1].offset = ref_objectid;
5550
5551 sizes[0] = sizeof(struct btrfs_inode_item);
5552 sizes[1] = name_len + sizeof(*ref);
5553
5554 path->leave_spinning = 1;
5555 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5556 if (ret != 0)
5557 goto fail;
5558
5559 inode_init_owner(inode, dir, mode);
5560 inode_set_bytes(inode, 0);
5561 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5562 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5563 struct btrfs_inode_item);
5564 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5565 sizeof(*inode_item));
5566 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5567
5568 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5569 struct btrfs_inode_ref);
5570 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5571 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5572 ptr = (unsigned long)(ref + 1);
5573 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5574
5575 btrfs_mark_buffer_dirty(path->nodes[0]);
5576 btrfs_free_path(path);
5577
5578 location = &BTRFS_I(inode)->location;
5579 location->objectid = objectid;
5580 location->offset = 0;
5581 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5582
5583 btrfs_inherit_iflags(inode, dir);
5584
5585 if (S_ISREG(mode)) {
5586 if (btrfs_test_opt(root, NODATASUM))
5587 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5588 if (btrfs_test_opt(root, NODATACOW))
5589 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5590 BTRFS_INODE_NODATASUM;
5591 }
5592
5593 insert_inode_hash(inode);
5594 inode_tree_add(inode);
5595
5596 trace_btrfs_inode_new(inode);
5597 btrfs_set_inode_last_trans(trans, inode);
5598
5599 btrfs_update_root_times(trans, root);
5600
5601 return inode;
5602 fail:
5603 if (dir)
5604 BTRFS_I(dir)->index_cnt--;
5605 btrfs_free_path(path);
5606 iput(inode);
5607 return ERR_PTR(ret);
5608 }
5609
5610 static inline u8 btrfs_inode_type(struct inode *inode)
5611 {
5612 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5613 }
5614
5615 /*
5616 * utility function to add 'inode' into 'parent_inode' with
5617 * a give name and a given sequence number.
5618 * if 'add_backref' is true, also insert a backref from the
5619 * inode to the parent directory.
5620 */
5621 int btrfs_add_link(struct btrfs_trans_handle *trans,
5622 struct inode *parent_inode, struct inode *inode,
5623 const char *name, int name_len, int add_backref, u64 index)
5624 {
5625 int ret = 0;
5626 struct btrfs_key key;
5627 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5628 u64 ino = btrfs_ino(inode);
5629 u64 parent_ino = btrfs_ino(parent_inode);
5630
5631 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5632 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5633 } else {
5634 key.objectid = ino;
5635 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5636 key.offset = 0;
5637 }
5638
5639 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5640 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5641 key.objectid, root->root_key.objectid,
5642 parent_ino, index, name, name_len);
5643 } else if (add_backref) {
5644 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5645 parent_ino, index);
5646 }
5647
5648 /* Nothing to clean up yet */
5649 if (ret)
5650 return ret;
5651
5652 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5653 parent_inode, &key,
5654 btrfs_inode_type(inode), index);
5655 if (ret == -EEXIST || ret == -EOVERFLOW)
5656 goto fail_dir_item;
5657 else if (ret) {
5658 btrfs_abort_transaction(trans, root, ret);
5659 return ret;
5660 }
5661
5662 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5663 name_len * 2);
5664 inode_inc_iversion(parent_inode);
5665 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5666 ret = btrfs_update_inode(trans, root, parent_inode);
5667 if (ret)
5668 btrfs_abort_transaction(trans, root, ret);
5669 return ret;
5670
5671 fail_dir_item:
5672 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5673 u64 local_index;
5674 int err;
5675 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5676 key.objectid, root->root_key.objectid,
5677 parent_ino, &local_index, name, name_len);
5678
5679 } else if (add_backref) {
5680 u64 local_index;
5681 int err;
5682
5683 err = btrfs_del_inode_ref(trans, root, name, name_len,
5684 ino, parent_ino, &local_index);
5685 }
5686 return ret;
5687 }
5688
5689 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5690 struct inode *dir, struct dentry *dentry,
5691 struct inode *inode, int backref, u64 index)
5692 {
5693 int err = btrfs_add_link(trans, dir, inode,
5694 dentry->d_name.name, dentry->d_name.len,
5695 backref, index);
5696 if (err > 0)
5697 err = -EEXIST;
5698 return err;
5699 }
5700
5701 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5702 umode_t mode, dev_t rdev)
5703 {
5704 struct btrfs_trans_handle *trans;
5705 struct btrfs_root *root = BTRFS_I(dir)->root;
5706 struct inode *inode = NULL;
5707 int err;
5708 int drop_inode = 0;
5709 u64 objectid;
5710 u64 index = 0;
5711
5712 if (!new_valid_dev(rdev))
5713 return -EINVAL;
5714
5715 /*
5716 * 2 for inode item and ref
5717 * 2 for dir items
5718 * 1 for xattr if selinux is on
5719 */
5720 trans = btrfs_start_transaction(root, 5);
5721 if (IS_ERR(trans))
5722 return PTR_ERR(trans);
5723
5724 err = btrfs_find_free_ino(root, &objectid);
5725 if (err)
5726 goto out_unlock;
5727
5728 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5729 dentry->d_name.len, btrfs_ino(dir), objectid,
5730 mode, &index);
5731 if (IS_ERR(inode)) {
5732 err = PTR_ERR(inode);
5733 goto out_unlock;
5734 }
5735
5736 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5737 if (err) {
5738 drop_inode = 1;
5739 goto out_unlock;
5740 }
5741
5742 /*
5743 * If the active LSM wants to access the inode during
5744 * d_instantiate it needs these. Smack checks to see
5745 * if the filesystem supports xattrs by looking at the
5746 * ops vector.
5747 */
5748
5749 inode->i_op = &btrfs_special_inode_operations;
5750 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5751 if (err)
5752 drop_inode = 1;
5753 else {
5754 init_special_inode(inode, inode->i_mode, rdev);
5755 btrfs_update_inode(trans, root, inode);
5756 d_instantiate(dentry, inode);
5757 }
5758 out_unlock:
5759 btrfs_end_transaction(trans, root);
5760 btrfs_btree_balance_dirty(root);
5761 if (drop_inode) {
5762 inode_dec_link_count(inode);
5763 iput(inode);
5764 }
5765 return err;
5766 }
5767
5768 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5769 umode_t mode, bool excl)
5770 {
5771 struct btrfs_trans_handle *trans;
5772 struct btrfs_root *root = BTRFS_I(dir)->root;
5773 struct inode *inode = NULL;
5774 int drop_inode_on_err = 0;
5775 int err;
5776 u64 objectid;
5777 u64 index = 0;
5778
5779 /*
5780 * 2 for inode item and ref
5781 * 2 for dir items
5782 * 1 for xattr if selinux is on
5783 */
5784 trans = btrfs_start_transaction(root, 5);
5785 if (IS_ERR(trans))
5786 return PTR_ERR(trans);
5787
5788 err = btrfs_find_free_ino(root, &objectid);
5789 if (err)
5790 goto out_unlock;
5791
5792 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5793 dentry->d_name.len, btrfs_ino(dir), objectid,
5794 mode, &index);
5795 if (IS_ERR(inode)) {
5796 err = PTR_ERR(inode);
5797 goto out_unlock;
5798 }
5799 drop_inode_on_err = 1;
5800
5801 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5802 if (err)
5803 goto out_unlock;
5804
5805 err = btrfs_update_inode(trans, root, inode);
5806 if (err)
5807 goto out_unlock;
5808
5809 /*
5810 * If the active LSM wants to access the inode during
5811 * d_instantiate it needs these. Smack checks to see
5812 * if the filesystem supports xattrs by looking at the
5813 * ops vector.
5814 */
5815 inode->i_fop = &btrfs_file_operations;
5816 inode->i_op = &btrfs_file_inode_operations;
5817
5818 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5819 if (err)
5820 goto out_unlock;
5821
5822 inode->i_mapping->a_ops = &btrfs_aops;
5823 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5824 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5825 d_instantiate(dentry, inode);
5826
5827 out_unlock:
5828 btrfs_end_transaction(trans, root);
5829 if (err && drop_inode_on_err) {
5830 inode_dec_link_count(inode);
5831 iput(inode);
5832 }
5833 btrfs_btree_balance_dirty(root);
5834 return err;
5835 }
5836
5837 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5838 struct dentry *dentry)
5839 {
5840 struct btrfs_trans_handle *trans;
5841 struct btrfs_root *root = BTRFS_I(dir)->root;
5842 struct inode *inode = old_dentry->d_inode;
5843 u64 index;
5844 int err;
5845 int drop_inode = 0;
5846
5847 /* do not allow sys_link's with other subvols of the same device */
5848 if (root->objectid != BTRFS_I(inode)->root->objectid)
5849 return -EXDEV;
5850
5851 if (inode->i_nlink >= BTRFS_LINK_MAX)
5852 return -EMLINK;
5853
5854 err = btrfs_set_inode_index(dir, &index);
5855 if (err)
5856 goto fail;
5857
5858 /*
5859 * 2 items for inode and inode ref
5860 * 2 items for dir items
5861 * 1 item for parent inode
5862 */
5863 trans = btrfs_start_transaction(root, 5);
5864 if (IS_ERR(trans)) {
5865 err = PTR_ERR(trans);
5866 goto fail;
5867 }
5868
5869 btrfs_inc_nlink(inode);
5870 inode_inc_iversion(inode);
5871 inode->i_ctime = CURRENT_TIME;
5872 ihold(inode);
5873 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5874
5875 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5876
5877 if (err) {
5878 drop_inode = 1;
5879 } else {
5880 struct dentry *parent = dentry->d_parent;
5881 err = btrfs_update_inode(trans, root, inode);
5882 if (err)
5883 goto fail;
5884 d_instantiate(dentry, inode);
5885 btrfs_log_new_name(trans, inode, NULL, parent);
5886 }
5887
5888 btrfs_end_transaction(trans, root);
5889 fail:
5890 if (drop_inode) {
5891 inode_dec_link_count(inode);
5892 iput(inode);
5893 }
5894 btrfs_btree_balance_dirty(root);
5895 return err;
5896 }
5897
5898 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5899 {
5900 struct inode *inode = NULL;
5901 struct btrfs_trans_handle *trans;
5902 struct btrfs_root *root = BTRFS_I(dir)->root;
5903 int err = 0;
5904 int drop_on_err = 0;
5905 u64 objectid = 0;
5906 u64 index = 0;
5907
5908 /*
5909 * 2 items for inode and ref
5910 * 2 items for dir items
5911 * 1 for xattr if selinux is on
5912 */
5913 trans = btrfs_start_transaction(root, 5);
5914 if (IS_ERR(trans))
5915 return PTR_ERR(trans);
5916
5917 err = btrfs_find_free_ino(root, &objectid);
5918 if (err)
5919 goto out_fail;
5920
5921 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5922 dentry->d_name.len, btrfs_ino(dir), objectid,
5923 S_IFDIR | mode, &index);
5924 if (IS_ERR(inode)) {
5925 err = PTR_ERR(inode);
5926 goto out_fail;
5927 }
5928
5929 drop_on_err = 1;
5930
5931 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5932 if (err)
5933 goto out_fail;
5934
5935 inode->i_op = &btrfs_dir_inode_operations;
5936 inode->i_fop = &btrfs_dir_file_operations;
5937
5938 btrfs_i_size_write(inode, 0);
5939 err = btrfs_update_inode(trans, root, inode);
5940 if (err)
5941 goto out_fail;
5942
5943 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5944 dentry->d_name.len, 0, index);
5945 if (err)
5946 goto out_fail;
5947
5948 d_instantiate(dentry, inode);
5949 drop_on_err = 0;
5950
5951 out_fail:
5952 btrfs_end_transaction(trans, root);
5953 if (drop_on_err)
5954 iput(inode);
5955 btrfs_btree_balance_dirty(root);
5956 return err;
5957 }
5958
5959 /* helper for btfs_get_extent. Given an existing extent in the tree,
5960 * and an extent that you want to insert, deal with overlap and insert
5961 * the new extent into the tree.
5962 */
5963 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5964 struct extent_map *existing,
5965 struct extent_map *em,
5966 u64 map_start, u64 map_len)
5967 {
5968 u64 start_diff;
5969
5970 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5971 start_diff = map_start - em->start;
5972 em->start = map_start;
5973 em->len = map_len;
5974 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5975 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5976 em->block_start += start_diff;
5977 em->block_len -= start_diff;
5978 }
5979 return add_extent_mapping(em_tree, em, 0);
5980 }
5981
5982 static noinline int uncompress_inline(struct btrfs_path *path,
5983 struct inode *inode, struct page *page,
5984 size_t pg_offset, u64 extent_offset,
5985 struct btrfs_file_extent_item *item)
5986 {
5987 int ret;
5988 struct extent_buffer *leaf = path->nodes[0];
5989 char *tmp;
5990 size_t max_size;
5991 unsigned long inline_size;
5992 unsigned long ptr;
5993 int compress_type;
5994
5995 WARN_ON(pg_offset != 0);
5996 compress_type = btrfs_file_extent_compression(leaf, item);
5997 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5998 inline_size = btrfs_file_extent_inline_item_len(leaf,
5999 btrfs_item_nr(leaf, path->slots[0]));
6000 tmp = kmalloc(inline_size, GFP_NOFS);
6001 if (!tmp)
6002 return -ENOMEM;
6003 ptr = btrfs_file_extent_inline_start(item);
6004
6005 read_extent_buffer(leaf, tmp, ptr, inline_size);
6006
6007 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6008 ret = btrfs_decompress(compress_type, tmp, page,
6009 extent_offset, inline_size, max_size);
6010 if (ret) {
6011 char *kaddr = kmap_atomic(page);
6012 unsigned long copy_size = min_t(u64,
6013 PAGE_CACHE_SIZE - pg_offset,
6014 max_size - extent_offset);
6015 memset(kaddr + pg_offset, 0, copy_size);
6016 kunmap_atomic(kaddr);
6017 }
6018 kfree(tmp);
6019 return 0;
6020 }
6021
6022 /*
6023 * a bit scary, this does extent mapping from logical file offset to the disk.
6024 * the ugly parts come from merging extents from the disk with the in-ram
6025 * representation. This gets more complex because of the data=ordered code,
6026 * where the in-ram extents might be locked pending data=ordered completion.
6027 *
6028 * This also copies inline extents directly into the page.
6029 */
6030
6031 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6032 size_t pg_offset, u64 start, u64 len,
6033 int create)
6034 {
6035 int ret;
6036 int err = 0;
6037 u64 bytenr;
6038 u64 extent_start = 0;
6039 u64 extent_end = 0;
6040 u64 objectid = btrfs_ino(inode);
6041 u32 found_type;
6042 struct btrfs_path *path = NULL;
6043 struct btrfs_root *root = BTRFS_I(inode)->root;
6044 struct btrfs_file_extent_item *item;
6045 struct extent_buffer *leaf;
6046 struct btrfs_key found_key;
6047 struct extent_map *em = NULL;
6048 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6049 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6050 struct btrfs_trans_handle *trans = NULL;
6051 int compress_type;
6052
6053 again:
6054 read_lock(&em_tree->lock);
6055 em = lookup_extent_mapping(em_tree, start, len);
6056 if (em)
6057 em->bdev = root->fs_info->fs_devices->latest_bdev;
6058 read_unlock(&em_tree->lock);
6059
6060 if (em) {
6061 if (em->start > start || em->start + em->len <= start)
6062 free_extent_map(em);
6063 else if (em->block_start == EXTENT_MAP_INLINE && page)
6064 free_extent_map(em);
6065 else
6066 goto out;
6067 }
6068 em = alloc_extent_map();
6069 if (!em) {
6070 err = -ENOMEM;
6071 goto out;
6072 }
6073 em->bdev = root->fs_info->fs_devices->latest_bdev;
6074 em->start = EXTENT_MAP_HOLE;
6075 em->orig_start = EXTENT_MAP_HOLE;
6076 em->len = (u64)-1;
6077 em->block_len = (u64)-1;
6078
6079 if (!path) {
6080 path = btrfs_alloc_path();
6081 if (!path) {
6082 err = -ENOMEM;
6083 goto out;
6084 }
6085 /*
6086 * Chances are we'll be called again, so go ahead and do
6087 * readahead
6088 */
6089 path->reada = 1;
6090 }
6091
6092 ret = btrfs_lookup_file_extent(trans, root, path,
6093 objectid, start, trans != NULL);
6094 if (ret < 0) {
6095 err = ret;
6096 goto out;
6097 }
6098
6099 if (ret != 0) {
6100 if (path->slots[0] == 0)
6101 goto not_found;
6102 path->slots[0]--;
6103 }
6104
6105 leaf = path->nodes[0];
6106 item = btrfs_item_ptr(leaf, path->slots[0],
6107 struct btrfs_file_extent_item);
6108 /* are we inside the extent that was found? */
6109 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6110 found_type = btrfs_key_type(&found_key);
6111 if (found_key.objectid != objectid ||
6112 found_type != BTRFS_EXTENT_DATA_KEY) {
6113 goto not_found;
6114 }
6115
6116 found_type = btrfs_file_extent_type(leaf, item);
6117 extent_start = found_key.offset;
6118 compress_type = btrfs_file_extent_compression(leaf, item);
6119 if (found_type == BTRFS_FILE_EXTENT_REG ||
6120 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6121 extent_end = extent_start +
6122 btrfs_file_extent_num_bytes(leaf, item);
6123 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6124 size_t size;
6125 size = btrfs_file_extent_inline_len(leaf, item);
6126 extent_end = ALIGN(extent_start + size, root->sectorsize);
6127 }
6128
6129 if (start >= extent_end) {
6130 path->slots[0]++;
6131 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6132 ret = btrfs_next_leaf(root, path);
6133 if (ret < 0) {
6134 err = ret;
6135 goto out;
6136 }
6137 if (ret > 0)
6138 goto not_found;
6139 leaf = path->nodes[0];
6140 }
6141 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6142 if (found_key.objectid != objectid ||
6143 found_key.type != BTRFS_EXTENT_DATA_KEY)
6144 goto not_found;
6145 if (start + len <= found_key.offset)
6146 goto not_found;
6147 em->start = start;
6148 em->orig_start = start;
6149 em->len = found_key.offset - start;
6150 goto not_found_em;
6151 }
6152
6153 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6154 if (found_type == BTRFS_FILE_EXTENT_REG ||
6155 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6156 em->start = extent_start;
6157 em->len = extent_end - extent_start;
6158 em->orig_start = extent_start -
6159 btrfs_file_extent_offset(leaf, item);
6160 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6161 item);
6162 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6163 if (bytenr == 0) {
6164 em->block_start = EXTENT_MAP_HOLE;
6165 goto insert;
6166 }
6167 if (compress_type != BTRFS_COMPRESS_NONE) {
6168 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6169 em->compress_type = compress_type;
6170 em->block_start = bytenr;
6171 em->block_len = em->orig_block_len;
6172 } else {
6173 bytenr += btrfs_file_extent_offset(leaf, item);
6174 em->block_start = bytenr;
6175 em->block_len = em->len;
6176 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6177 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6178 }
6179 goto insert;
6180 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6181 unsigned long ptr;
6182 char *map;
6183 size_t size;
6184 size_t extent_offset;
6185 size_t copy_size;
6186
6187 em->block_start = EXTENT_MAP_INLINE;
6188 if (!page || create) {
6189 em->start = extent_start;
6190 em->len = extent_end - extent_start;
6191 goto out;
6192 }
6193
6194 size = btrfs_file_extent_inline_len(leaf, item);
6195 extent_offset = page_offset(page) + pg_offset - extent_start;
6196 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6197 size - extent_offset);
6198 em->start = extent_start + extent_offset;
6199 em->len = ALIGN(copy_size, root->sectorsize);
6200 em->orig_block_len = em->len;
6201 em->orig_start = em->start;
6202 if (compress_type) {
6203 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6204 em->compress_type = compress_type;
6205 }
6206 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6207 if (create == 0 && !PageUptodate(page)) {
6208 if (btrfs_file_extent_compression(leaf, item) !=
6209 BTRFS_COMPRESS_NONE) {
6210 ret = uncompress_inline(path, inode, page,
6211 pg_offset,
6212 extent_offset, item);
6213 BUG_ON(ret); /* -ENOMEM */
6214 } else {
6215 map = kmap(page);
6216 read_extent_buffer(leaf, map + pg_offset, ptr,
6217 copy_size);
6218 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6219 memset(map + pg_offset + copy_size, 0,
6220 PAGE_CACHE_SIZE - pg_offset -
6221 copy_size);
6222 }
6223 kunmap(page);
6224 }
6225 flush_dcache_page(page);
6226 } else if (create && PageUptodate(page)) {
6227 BUG();
6228 if (!trans) {
6229 kunmap(page);
6230 free_extent_map(em);
6231 em = NULL;
6232
6233 btrfs_release_path(path);
6234 trans = btrfs_join_transaction(root);
6235
6236 if (IS_ERR(trans))
6237 return ERR_CAST(trans);
6238 goto again;
6239 }
6240 map = kmap(page);
6241 write_extent_buffer(leaf, map + pg_offset, ptr,
6242 copy_size);
6243 kunmap(page);
6244 btrfs_mark_buffer_dirty(leaf);
6245 }
6246 set_extent_uptodate(io_tree, em->start,
6247 extent_map_end(em) - 1, NULL, GFP_NOFS);
6248 goto insert;
6249 } else {
6250 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6251 }
6252 not_found:
6253 em->start = start;
6254 em->orig_start = start;
6255 em->len = len;
6256 not_found_em:
6257 em->block_start = EXTENT_MAP_HOLE;
6258 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6259 insert:
6260 btrfs_release_path(path);
6261 if (em->start > start || extent_map_end(em) <= start) {
6262 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6263 (unsigned long long)em->start,
6264 (unsigned long long)em->len,
6265 (unsigned long long)start,
6266 (unsigned long long)len);
6267 err = -EIO;
6268 goto out;
6269 }
6270
6271 err = 0;
6272 write_lock(&em_tree->lock);
6273 ret = add_extent_mapping(em_tree, em, 0);
6274 /* it is possible that someone inserted the extent into the tree
6275 * while we had the lock dropped. It is also possible that
6276 * an overlapping map exists in the tree
6277 */
6278 if (ret == -EEXIST) {
6279 struct extent_map *existing;
6280
6281 ret = 0;
6282
6283 existing = lookup_extent_mapping(em_tree, start, len);
6284 if (existing && (existing->start > start ||
6285 existing->start + existing->len <= start)) {
6286 free_extent_map(existing);
6287 existing = NULL;
6288 }
6289 if (!existing) {
6290 existing = lookup_extent_mapping(em_tree, em->start,
6291 em->len);
6292 if (existing) {
6293 err = merge_extent_mapping(em_tree, existing,
6294 em, start,
6295 root->sectorsize);
6296 free_extent_map(existing);
6297 if (err) {
6298 free_extent_map(em);
6299 em = NULL;
6300 }
6301 } else {
6302 err = -EIO;
6303 free_extent_map(em);
6304 em = NULL;
6305 }
6306 } else {
6307 free_extent_map(em);
6308 em = existing;
6309 err = 0;
6310 }
6311 }
6312 write_unlock(&em_tree->lock);
6313 out:
6314
6315 if (em)
6316 trace_btrfs_get_extent(root, em);
6317
6318 if (path)
6319 btrfs_free_path(path);
6320 if (trans) {
6321 ret = btrfs_end_transaction(trans, root);
6322 if (!err)
6323 err = ret;
6324 }
6325 if (err) {
6326 free_extent_map(em);
6327 return ERR_PTR(err);
6328 }
6329 BUG_ON(!em); /* Error is always set */
6330 return em;
6331 }
6332
6333 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6334 size_t pg_offset, u64 start, u64 len,
6335 int create)
6336 {
6337 struct extent_map *em;
6338 struct extent_map *hole_em = NULL;
6339 u64 range_start = start;
6340 u64 end;
6341 u64 found;
6342 u64 found_end;
6343 int err = 0;
6344
6345 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6346 if (IS_ERR(em))
6347 return em;
6348 if (em) {
6349 /*
6350 * if our em maps to
6351 * - a hole or
6352 * - a pre-alloc extent,
6353 * there might actually be delalloc bytes behind it.
6354 */
6355 if (em->block_start != EXTENT_MAP_HOLE &&
6356 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6357 return em;
6358 else
6359 hole_em = em;
6360 }
6361
6362 /* check to see if we've wrapped (len == -1 or similar) */
6363 end = start + len;
6364 if (end < start)
6365 end = (u64)-1;
6366 else
6367 end -= 1;
6368
6369 em = NULL;
6370
6371 /* ok, we didn't find anything, lets look for delalloc */
6372 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6373 end, len, EXTENT_DELALLOC, 1);
6374 found_end = range_start + found;
6375 if (found_end < range_start)
6376 found_end = (u64)-1;
6377
6378 /*
6379 * we didn't find anything useful, return
6380 * the original results from get_extent()
6381 */
6382 if (range_start > end || found_end <= start) {
6383 em = hole_em;
6384 hole_em = NULL;
6385 goto out;
6386 }
6387
6388 /* adjust the range_start to make sure it doesn't
6389 * go backwards from the start they passed in
6390 */
6391 range_start = max(start,range_start);
6392 found = found_end - range_start;
6393
6394 if (found > 0) {
6395 u64 hole_start = start;
6396 u64 hole_len = len;
6397
6398 em = alloc_extent_map();
6399 if (!em) {
6400 err = -ENOMEM;
6401 goto out;
6402 }
6403 /*
6404 * when btrfs_get_extent can't find anything it
6405 * returns one huge hole
6406 *
6407 * make sure what it found really fits our range, and
6408 * adjust to make sure it is based on the start from
6409 * the caller
6410 */
6411 if (hole_em) {
6412 u64 calc_end = extent_map_end(hole_em);
6413
6414 if (calc_end <= start || (hole_em->start > end)) {
6415 free_extent_map(hole_em);
6416 hole_em = NULL;
6417 } else {
6418 hole_start = max(hole_em->start, start);
6419 hole_len = calc_end - hole_start;
6420 }
6421 }
6422 em->bdev = NULL;
6423 if (hole_em && range_start > hole_start) {
6424 /* our hole starts before our delalloc, so we
6425 * have to return just the parts of the hole
6426 * that go until the delalloc starts
6427 */
6428 em->len = min(hole_len,
6429 range_start - hole_start);
6430 em->start = hole_start;
6431 em->orig_start = hole_start;
6432 /*
6433 * don't adjust block start at all,
6434 * it is fixed at EXTENT_MAP_HOLE
6435 */
6436 em->block_start = hole_em->block_start;
6437 em->block_len = hole_len;
6438 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6439 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6440 } else {
6441 em->start = range_start;
6442 em->len = found;
6443 em->orig_start = range_start;
6444 em->block_start = EXTENT_MAP_DELALLOC;
6445 em->block_len = found;
6446 }
6447 } else if (hole_em) {
6448 return hole_em;
6449 }
6450 out:
6451
6452 free_extent_map(hole_em);
6453 if (err) {
6454 free_extent_map(em);
6455 return ERR_PTR(err);
6456 }
6457 return em;
6458 }
6459
6460 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6461 u64 start, u64 len)
6462 {
6463 struct btrfs_root *root = BTRFS_I(inode)->root;
6464 struct btrfs_trans_handle *trans;
6465 struct extent_map *em;
6466 struct btrfs_key ins;
6467 u64 alloc_hint;
6468 int ret;
6469
6470 trans = btrfs_join_transaction(root);
6471 if (IS_ERR(trans))
6472 return ERR_CAST(trans);
6473
6474 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6475
6476 alloc_hint = get_extent_allocation_hint(inode, start, len);
6477 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6478 alloc_hint, &ins, 1);
6479 if (ret) {
6480 em = ERR_PTR(ret);
6481 goto out;
6482 }
6483
6484 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6485 ins.offset, ins.offset, ins.offset, 0);
6486 if (IS_ERR(em))
6487 goto out;
6488
6489 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6490 ins.offset, ins.offset, 0);
6491 if (ret) {
6492 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6493 em = ERR_PTR(ret);
6494 }
6495 out:
6496 btrfs_end_transaction(trans, root);
6497 return em;
6498 }
6499
6500 /*
6501 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6502 * block must be cow'd
6503 */
6504 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6505 struct inode *inode, u64 offset, u64 *len,
6506 u64 *orig_start, u64 *orig_block_len,
6507 u64 *ram_bytes)
6508 {
6509 struct btrfs_path *path;
6510 int ret;
6511 struct extent_buffer *leaf;
6512 struct btrfs_root *root = BTRFS_I(inode)->root;
6513 struct btrfs_file_extent_item *fi;
6514 struct btrfs_key key;
6515 u64 disk_bytenr;
6516 u64 backref_offset;
6517 u64 extent_end;
6518 u64 num_bytes;
6519 int slot;
6520 int found_type;
6521
6522 path = btrfs_alloc_path();
6523 if (!path)
6524 return -ENOMEM;
6525
6526 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6527 offset, 0);
6528 if (ret < 0)
6529 goto out;
6530
6531 slot = path->slots[0];
6532 if (ret == 1) {
6533 if (slot == 0) {
6534 /* can't find the item, must cow */
6535 ret = 0;
6536 goto out;
6537 }
6538 slot--;
6539 }
6540 ret = 0;
6541 leaf = path->nodes[0];
6542 btrfs_item_key_to_cpu(leaf, &key, slot);
6543 if (key.objectid != btrfs_ino(inode) ||
6544 key.type != BTRFS_EXTENT_DATA_KEY) {
6545 /* not our file or wrong item type, must cow */
6546 goto out;
6547 }
6548
6549 if (key.offset > offset) {
6550 /* Wrong offset, must cow */
6551 goto out;
6552 }
6553
6554 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6555 found_type = btrfs_file_extent_type(leaf, fi);
6556 if (found_type != BTRFS_FILE_EXTENT_REG &&
6557 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6558 /* not a regular extent, must cow */
6559 goto out;
6560 }
6561 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6562 backref_offset = btrfs_file_extent_offset(leaf, fi);
6563
6564 *orig_start = key.offset - backref_offset;
6565 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6566 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6567
6568 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6569 if (extent_end < offset + *len) {
6570 /* extent doesn't include our full range, must cow */
6571 goto out;
6572 }
6573
6574 if (btrfs_extent_readonly(root, disk_bytenr))
6575 goto out;
6576
6577 /*
6578 * look for other files referencing this extent, if we
6579 * find any we must cow
6580 */
6581 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6582 key.offset - backref_offset, disk_bytenr))
6583 goto out;
6584
6585 /*
6586 * adjust disk_bytenr and num_bytes to cover just the bytes
6587 * in this extent we are about to write. If there
6588 * are any csums in that range we have to cow in order
6589 * to keep the csums correct
6590 */
6591 disk_bytenr += backref_offset;
6592 disk_bytenr += offset - key.offset;
6593 num_bytes = min(offset + *len, extent_end) - offset;
6594 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6595 goto out;
6596 /*
6597 * all of the above have passed, it is safe to overwrite this extent
6598 * without cow
6599 */
6600 *len = num_bytes;
6601 ret = 1;
6602 out:
6603 btrfs_free_path(path);
6604 return ret;
6605 }
6606
6607 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6608 struct extent_state **cached_state, int writing)
6609 {
6610 struct btrfs_ordered_extent *ordered;
6611 int ret = 0;
6612
6613 while (1) {
6614 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6615 0, cached_state);
6616 /*
6617 * We're concerned with the entire range that we're going to be
6618 * doing DIO to, so we need to make sure theres no ordered
6619 * extents in this range.
6620 */
6621 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6622 lockend - lockstart + 1);
6623
6624 /*
6625 * We need to make sure there are no buffered pages in this
6626 * range either, we could have raced between the invalidate in
6627 * generic_file_direct_write and locking the extent. The
6628 * invalidate needs to happen so that reads after a write do not
6629 * get stale data.
6630 */
6631 if (!ordered && (!writing ||
6632 !test_range_bit(&BTRFS_I(inode)->io_tree,
6633 lockstart, lockend, EXTENT_UPTODATE, 0,
6634 *cached_state)))
6635 break;
6636
6637 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6638 cached_state, GFP_NOFS);
6639
6640 if (ordered) {
6641 btrfs_start_ordered_extent(inode, ordered, 1);
6642 btrfs_put_ordered_extent(ordered);
6643 } else {
6644 /* Screw you mmap */
6645 ret = filemap_write_and_wait_range(inode->i_mapping,
6646 lockstart,
6647 lockend);
6648 if (ret)
6649 break;
6650
6651 /*
6652 * If we found a page that couldn't be invalidated just
6653 * fall back to buffered.
6654 */
6655 ret = invalidate_inode_pages2_range(inode->i_mapping,
6656 lockstart >> PAGE_CACHE_SHIFT,
6657 lockend >> PAGE_CACHE_SHIFT);
6658 if (ret)
6659 break;
6660 }
6661
6662 cond_resched();
6663 }
6664
6665 return ret;
6666 }
6667
6668 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6669 u64 len, u64 orig_start,
6670 u64 block_start, u64 block_len,
6671 u64 orig_block_len, u64 ram_bytes,
6672 int type)
6673 {
6674 struct extent_map_tree *em_tree;
6675 struct extent_map *em;
6676 struct btrfs_root *root = BTRFS_I(inode)->root;
6677 int ret;
6678
6679 em_tree = &BTRFS_I(inode)->extent_tree;
6680 em = alloc_extent_map();
6681 if (!em)
6682 return ERR_PTR(-ENOMEM);
6683
6684 em->start = start;
6685 em->orig_start = orig_start;
6686 em->mod_start = start;
6687 em->mod_len = len;
6688 em->len = len;
6689 em->block_len = block_len;
6690 em->block_start = block_start;
6691 em->bdev = root->fs_info->fs_devices->latest_bdev;
6692 em->orig_block_len = orig_block_len;
6693 em->ram_bytes = ram_bytes;
6694 em->generation = -1;
6695 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6696 if (type == BTRFS_ORDERED_PREALLOC)
6697 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6698
6699 do {
6700 btrfs_drop_extent_cache(inode, em->start,
6701 em->start + em->len - 1, 0);
6702 write_lock(&em_tree->lock);
6703 ret = add_extent_mapping(em_tree, em, 1);
6704 write_unlock(&em_tree->lock);
6705 } while (ret == -EEXIST);
6706
6707 if (ret) {
6708 free_extent_map(em);
6709 return ERR_PTR(ret);
6710 }
6711
6712 return em;
6713 }
6714
6715
6716 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6717 struct buffer_head *bh_result, int create)
6718 {
6719 struct extent_map *em;
6720 struct btrfs_root *root = BTRFS_I(inode)->root;
6721 struct extent_state *cached_state = NULL;
6722 u64 start = iblock << inode->i_blkbits;
6723 u64 lockstart, lockend;
6724 u64 len = bh_result->b_size;
6725 struct btrfs_trans_handle *trans;
6726 int unlock_bits = EXTENT_LOCKED;
6727 int ret = 0;
6728
6729 if (create)
6730 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6731 else
6732 len = min_t(u64, len, root->sectorsize);
6733
6734 lockstart = start;
6735 lockend = start + len - 1;
6736
6737 /*
6738 * If this errors out it's because we couldn't invalidate pagecache for
6739 * this range and we need to fallback to buffered.
6740 */
6741 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6742 return -ENOTBLK;
6743
6744 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6745 if (IS_ERR(em)) {
6746 ret = PTR_ERR(em);
6747 goto unlock_err;
6748 }
6749
6750 /*
6751 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6752 * io. INLINE is special, and we could probably kludge it in here, but
6753 * it's still buffered so for safety lets just fall back to the generic
6754 * buffered path.
6755 *
6756 * For COMPRESSED we _have_ to read the entire extent in so we can
6757 * decompress it, so there will be buffering required no matter what we
6758 * do, so go ahead and fallback to buffered.
6759 *
6760 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6761 * to buffered IO. Don't blame me, this is the price we pay for using
6762 * the generic code.
6763 */
6764 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6765 em->block_start == EXTENT_MAP_INLINE) {
6766 free_extent_map(em);
6767 ret = -ENOTBLK;
6768 goto unlock_err;
6769 }
6770
6771 /* Just a good old fashioned hole, return */
6772 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6773 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6774 free_extent_map(em);
6775 goto unlock_err;
6776 }
6777
6778 /*
6779 * We don't allocate a new extent in the following cases
6780 *
6781 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6782 * existing extent.
6783 * 2) The extent is marked as PREALLOC. We're good to go here and can
6784 * just use the extent.
6785 *
6786 */
6787 if (!create) {
6788 len = min(len, em->len - (start - em->start));
6789 lockstart = start + len;
6790 goto unlock;
6791 }
6792
6793 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6794 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6795 em->block_start != EXTENT_MAP_HOLE)) {
6796 int type;
6797 int ret;
6798 u64 block_start, orig_start, orig_block_len, ram_bytes;
6799
6800 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6801 type = BTRFS_ORDERED_PREALLOC;
6802 else
6803 type = BTRFS_ORDERED_NOCOW;
6804 len = min(len, em->len - (start - em->start));
6805 block_start = em->block_start + (start - em->start);
6806
6807 /*
6808 * we're not going to log anything, but we do need
6809 * to make sure the current transaction stays open
6810 * while we look for nocow cross refs
6811 */
6812 trans = btrfs_join_transaction(root);
6813 if (IS_ERR(trans))
6814 goto must_cow;
6815
6816 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6817 &orig_block_len, &ram_bytes) == 1) {
6818 if (type == BTRFS_ORDERED_PREALLOC) {
6819 free_extent_map(em);
6820 em = create_pinned_em(inode, start, len,
6821 orig_start,
6822 block_start, len,
6823 orig_block_len,
6824 ram_bytes, type);
6825 if (IS_ERR(em)) {
6826 btrfs_end_transaction(trans, root);
6827 goto unlock_err;
6828 }
6829 }
6830
6831 ret = btrfs_add_ordered_extent_dio(inode, start,
6832 block_start, len, len, type);
6833 btrfs_end_transaction(trans, root);
6834 if (ret) {
6835 free_extent_map(em);
6836 goto unlock_err;
6837 }
6838 goto unlock;
6839 }
6840 btrfs_end_transaction(trans, root);
6841 }
6842 must_cow:
6843 /*
6844 * this will cow the extent, reset the len in case we changed
6845 * it above
6846 */
6847 len = bh_result->b_size;
6848 free_extent_map(em);
6849 em = btrfs_new_extent_direct(inode, start, len);
6850 if (IS_ERR(em)) {
6851 ret = PTR_ERR(em);
6852 goto unlock_err;
6853 }
6854 len = min(len, em->len - (start - em->start));
6855 unlock:
6856 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6857 inode->i_blkbits;
6858 bh_result->b_size = len;
6859 bh_result->b_bdev = em->bdev;
6860 set_buffer_mapped(bh_result);
6861 if (create) {
6862 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6863 set_buffer_new(bh_result);
6864
6865 /*
6866 * Need to update the i_size under the extent lock so buffered
6867 * readers will get the updated i_size when we unlock.
6868 */
6869 if (start + len > i_size_read(inode))
6870 i_size_write(inode, start + len);
6871
6872 spin_lock(&BTRFS_I(inode)->lock);
6873 BTRFS_I(inode)->outstanding_extents++;
6874 spin_unlock(&BTRFS_I(inode)->lock);
6875
6876 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6877 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6878 &cached_state, GFP_NOFS);
6879 BUG_ON(ret);
6880 }
6881
6882 /*
6883 * In the case of write we need to clear and unlock the entire range,
6884 * in the case of read we need to unlock only the end area that we
6885 * aren't using if there is any left over space.
6886 */
6887 if (lockstart < lockend) {
6888 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6889 lockend, unlock_bits, 1, 0,
6890 &cached_state, GFP_NOFS);
6891 } else {
6892 free_extent_state(cached_state);
6893 }
6894
6895 free_extent_map(em);
6896
6897 return 0;
6898
6899 unlock_err:
6900 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6901 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6902 return ret;
6903 }
6904
6905 struct btrfs_dio_private {
6906 struct inode *inode;
6907 u64 logical_offset;
6908 u64 disk_bytenr;
6909 u64 bytes;
6910 void *private;
6911
6912 /* number of bios pending for this dio */
6913 atomic_t pending_bios;
6914
6915 /* IO errors */
6916 int errors;
6917
6918 /* orig_bio is our btrfs_io_bio */
6919 struct bio *orig_bio;
6920
6921 /* dio_bio came from fs/direct-io.c */
6922 struct bio *dio_bio;
6923 };
6924
6925 static void btrfs_endio_direct_read(struct bio *bio, int err)
6926 {
6927 struct btrfs_dio_private *dip = bio->bi_private;
6928 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6929 struct bio_vec *bvec = bio->bi_io_vec;
6930 struct inode *inode = dip->inode;
6931 struct btrfs_root *root = BTRFS_I(inode)->root;
6932 struct bio *dio_bio;
6933 u64 start;
6934
6935 start = dip->logical_offset;
6936 do {
6937 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6938 struct page *page = bvec->bv_page;
6939 char *kaddr;
6940 u32 csum = ~(u32)0;
6941 u64 private = ~(u32)0;
6942 unsigned long flags;
6943
6944 if (get_state_private(&BTRFS_I(inode)->io_tree,
6945 start, &private))
6946 goto failed;
6947 local_irq_save(flags);
6948 kaddr = kmap_atomic(page);
6949 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6950 csum, bvec->bv_len);
6951 btrfs_csum_final(csum, (char *)&csum);
6952 kunmap_atomic(kaddr);
6953 local_irq_restore(flags);
6954
6955 flush_dcache_page(bvec->bv_page);
6956 if (csum != private) {
6957 failed:
6958 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6959 (unsigned long long)btrfs_ino(inode),
6960 (unsigned long long)start,
6961 csum, (unsigned)private);
6962 err = -EIO;
6963 }
6964 }
6965
6966 start += bvec->bv_len;
6967 bvec++;
6968 } while (bvec <= bvec_end);
6969
6970 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6971 dip->logical_offset + dip->bytes - 1);
6972 dio_bio = dip->dio_bio;
6973
6974 kfree(dip);
6975
6976 /* If we had a csum failure make sure to clear the uptodate flag */
6977 if (err)
6978 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6979 dio_end_io(dio_bio, err);
6980 bio_put(bio);
6981 }
6982
6983 static void btrfs_endio_direct_write(struct bio *bio, int err)
6984 {
6985 struct btrfs_dio_private *dip = bio->bi_private;
6986 struct inode *inode = dip->inode;
6987 struct btrfs_root *root = BTRFS_I(inode)->root;
6988 struct btrfs_ordered_extent *ordered = NULL;
6989 u64 ordered_offset = dip->logical_offset;
6990 u64 ordered_bytes = dip->bytes;
6991 struct bio *dio_bio;
6992 int ret;
6993
6994 if (err)
6995 goto out_done;
6996 again:
6997 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6998 &ordered_offset,
6999 ordered_bytes, !err);
7000 if (!ret)
7001 goto out_test;
7002
7003 ordered->work.func = finish_ordered_fn;
7004 ordered->work.flags = 0;
7005 btrfs_queue_worker(&root->fs_info->endio_write_workers,
7006 &ordered->work);
7007 out_test:
7008 /*
7009 * our bio might span multiple ordered extents. If we haven't
7010 * completed the accounting for the whole dio, go back and try again
7011 */
7012 if (ordered_offset < dip->logical_offset + dip->bytes) {
7013 ordered_bytes = dip->logical_offset + dip->bytes -
7014 ordered_offset;
7015 ordered = NULL;
7016 goto again;
7017 }
7018 out_done:
7019 dio_bio = dip->dio_bio;
7020
7021 kfree(dip);
7022
7023 /* If we had an error make sure to clear the uptodate flag */
7024 if (err)
7025 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7026 dio_end_io(dio_bio, err);
7027 bio_put(bio);
7028 }
7029
7030 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7031 struct bio *bio, int mirror_num,
7032 unsigned long bio_flags, u64 offset)
7033 {
7034 int ret;
7035 struct btrfs_root *root = BTRFS_I(inode)->root;
7036 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7037 BUG_ON(ret); /* -ENOMEM */
7038 return 0;
7039 }
7040
7041 static void btrfs_end_dio_bio(struct bio *bio, int err)
7042 {
7043 struct btrfs_dio_private *dip = bio->bi_private;
7044
7045 if (err) {
7046 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7047 "sector %#Lx len %u err no %d\n",
7048 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7049 (unsigned long long)bio->bi_sector, bio->bi_size, err);
7050 dip->errors = 1;
7051
7052 /*
7053 * before atomic variable goto zero, we must make sure
7054 * dip->errors is perceived to be set.
7055 */
7056 smp_mb__before_atomic_dec();
7057 }
7058
7059 /* if there are more bios still pending for this dio, just exit */
7060 if (!atomic_dec_and_test(&dip->pending_bios))
7061 goto out;
7062
7063 if (dip->errors) {
7064 bio_io_error(dip->orig_bio);
7065 } else {
7066 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7067 bio_endio(dip->orig_bio, 0);
7068 }
7069 out:
7070 bio_put(bio);
7071 }
7072
7073 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7074 u64 first_sector, gfp_t gfp_flags)
7075 {
7076 int nr_vecs = bio_get_nr_vecs(bdev);
7077 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7078 }
7079
7080 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7081 int rw, u64 file_offset, int skip_sum,
7082 int async_submit)
7083 {
7084 int write = rw & REQ_WRITE;
7085 struct btrfs_root *root = BTRFS_I(inode)->root;
7086 int ret;
7087
7088 if (async_submit)
7089 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7090
7091 bio_get(bio);
7092
7093 if (!write) {
7094 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7095 if (ret)
7096 goto err;
7097 }
7098
7099 if (skip_sum)
7100 goto map;
7101
7102 if (write && async_submit) {
7103 ret = btrfs_wq_submit_bio(root->fs_info,
7104 inode, rw, bio, 0, 0,
7105 file_offset,
7106 __btrfs_submit_bio_start_direct_io,
7107 __btrfs_submit_bio_done);
7108 goto err;
7109 } else if (write) {
7110 /*
7111 * If we aren't doing async submit, calculate the csum of the
7112 * bio now.
7113 */
7114 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7115 if (ret)
7116 goto err;
7117 } else if (!skip_sum) {
7118 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7119 if (ret)
7120 goto err;
7121 }
7122
7123 map:
7124 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7125 err:
7126 bio_put(bio);
7127 return ret;
7128 }
7129
7130 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7131 int skip_sum)
7132 {
7133 struct inode *inode = dip->inode;
7134 struct btrfs_root *root = BTRFS_I(inode)->root;
7135 struct bio *bio;
7136 struct bio *orig_bio = dip->orig_bio;
7137 struct bio_vec *bvec = orig_bio->bi_io_vec;
7138 u64 start_sector = orig_bio->bi_sector;
7139 u64 file_offset = dip->logical_offset;
7140 u64 submit_len = 0;
7141 u64 map_length;
7142 int nr_pages = 0;
7143 int ret = 0;
7144 int async_submit = 0;
7145
7146 map_length = orig_bio->bi_size;
7147 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7148 &map_length, NULL, 0);
7149 if (ret) {
7150 bio_put(orig_bio);
7151 return -EIO;
7152 }
7153 if (map_length >= orig_bio->bi_size) {
7154 bio = orig_bio;
7155 goto submit;
7156 }
7157
7158 /* async crcs make it difficult to collect full stripe writes. */
7159 if (btrfs_get_alloc_profile(root, 1) &
7160 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7161 async_submit = 0;
7162 else
7163 async_submit = 1;
7164
7165 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7166 if (!bio)
7167 return -ENOMEM;
7168 bio->bi_private = dip;
7169 bio->bi_end_io = btrfs_end_dio_bio;
7170 atomic_inc(&dip->pending_bios);
7171
7172 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7173 if (unlikely(map_length < submit_len + bvec->bv_len ||
7174 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7175 bvec->bv_offset) < bvec->bv_len)) {
7176 /*
7177 * inc the count before we submit the bio so
7178 * we know the end IO handler won't happen before
7179 * we inc the count. Otherwise, the dip might get freed
7180 * before we're done setting it up
7181 */
7182 atomic_inc(&dip->pending_bios);
7183 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7184 file_offset, skip_sum,
7185 async_submit);
7186 if (ret) {
7187 bio_put(bio);
7188 atomic_dec(&dip->pending_bios);
7189 goto out_err;
7190 }
7191
7192 start_sector += submit_len >> 9;
7193 file_offset += submit_len;
7194
7195 submit_len = 0;
7196 nr_pages = 0;
7197
7198 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7199 start_sector, GFP_NOFS);
7200 if (!bio)
7201 goto out_err;
7202 bio->bi_private = dip;
7203 bio->bi_end_io = btrfs_end_dio_bio;
7204
7205 map_length = orig_bio->bi_size;
7206 ret = btrfs_map_block(root->fs_info, rw,
7207 start_sector << 9,
7208 &map_length, NULL, 0);
7209 if (ret) {
7210 bio_put(bio);
7211 goto out_err;
7212 }
7213 } else {
7214 submit_len += bvec->bv_len;
7215 nr_pages ++;
7216 bvec++;
7217 }
7218 }
7219
7220 submit:
7221 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7222 async_submit);
7223 if (!ret)
7224 return 0;
7225
7226 bio_put(bio);
7227 out_err:
7228 dip->errors = 1;
7229 /*
7230 * before atomic variable goto zero, we must
7231 * make sure dip->errors is perceived to be set.
7232 */
7233 smp_mb__before_atomic_dec();
7234 if (atomic_dec_and_test(&dip->pending_bios))
7235 bio_io_error(dip->orig_bio);
7236
7237 /* bio_end_io() will handle error, so we needn't return it */
7238 return 0;
7239 }
7240
7241 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7242 struct inode *inode, loff_t file_offset)
7243 {
7244 struct btrfs_root *root = BTRFS_I(inode)->root;
7245 struct btrfs_dio_private *dip;
7246 struct bio_vec *bvec = dio_bio->bi_io_vec;
7247 struct bio *io_bio;
7248 int skip_sum;
7249 int write = rw & REQ_WRITE;
7250 int ret = 0;
7251
7252 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7253
7254 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7255
7256 if (!io_bio) {
7257 ret = -ENOMEM;
7258 goto free_ordered;
7259 }
7260
7261 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7262 if (!dip) {
7263 ret = -ENOMEM;
7264 goto free_io_bio;
7265 }
7266
7267 dip->private = dio_bio->bi_private;
7268 io_bio->bi_private = dio_bio->bi_private;
7269 dip->inode = inode;
7270 dip->logical_offset = file_offset;
7271
7272 dip->bytes = 0;
7273 do {
7274 dip->bytes += bvec->bv_len;
7275 bvec++;
7276 } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7277
7278 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7279 io_bio->bi_private = dip;
7280 dip->errors = 0;
7281 dip->orig_bio = io_bio;
7282 dip->dio_bio = dio_bio;
7283 atomic_set(&dip->pending_bios, 0);
7284
7285 if (write)
7286 io_bio->bi_end_io = btrfs_endio_direct_write;
7287 else
7288 io_bio->bi_end_io = btrfs_endio_direct_read;
7289
7290 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7291 if (!ret)
7292 return;
7293
7294 free_io_bio:
7295 bio_put(io_bio);
7296
7297 free_ordered:
7298 /*
7299 * If this is a write, we need to clean up the reserved space and kill
7300 * the ordered extent.
7301 */
7302 if (write) {
7303 struct btrfs_ordered_extent *ordered;
7304 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7305 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7306 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7307 btrfs_free_reserved_extent(root, ordered->start,
7308 ordered->disk_len);
7309 btrfs_put_ordered_extent(ordered);
7310 btrfs_put_ordered_extent(ordered);
7311 }
7312 bio_endio(dio_bio, ret);
7313 }
7314
7315 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7316 const struct iovec *iov, loff_t offset,
7317 unsigned long nr_segs)
7318 {
7319 int seg;
7320 int i;
7321 size_t size;
7322 unsigned long addr;
7323 unsigned blocksize_mask = root->sectorsize - 1;
7324 ssize_t retval = -EINVAL;
7325 loff_t end = offset;
7326
7327 if (offset & blocksize_mask)
7328 goto out;
7329
7330 /* Check the memory alignment. Blocks cannot straddle pages */
7331 for (seg = 0; seg < nr_segs; seg++) {
7332 addr = (unsigned long)iov[seg].iov_base;
7333 size = iov[seg].iov_len;
7334 end += size;
7335 if ((addr & blocksize_mask) || (size & blocksize_mask))
7336 goto out;
7337
7338 /* If this is a write we don't need to check anymore */
7339 if (rw & WRITE)
7340 continue;
7341
7342 /*
7343 * Check to make sure we don't have duplicate iov_base's in this
7344 * iovec, if so return EINVAL, otherwise we'll get csum errors
7345 * when reading back.
7346 */
7347 for (i = seg + 1; i < nr_segs; i++) {
7348 if (iov[seg].iov_base == iov[i].iov_base)
7349 goto out;
7350 }
7351 }
7352 retval = 0;
7353 out:
7354 return retval;
7355 }
7356
7357 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7358 const struct iovec *iov, loff_t offset,
7359 unsigned long nr_segs)
7360 {
7361 struct file *file = iocb->ki_filp;
7362 struct inode *inode = file->f_mapping->host;
7363 size_t count = 0;
7364 int flags = 0;
7365 bool wakeup = true;
7366 bool relock = false;
7367 ssize_t ret;
7368
7369 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7370 offset, nr_segs))
7371 return 0;
7372
7373 atomic_inc(&inode->i_dio_count);
7374 smp_mb__after_atomic_inc();
7375
7376 if (rw & WRITE) {
7377 count = iov_length(iov, nr_segs);
7378 /*
7379 * If the write DIO is beyond the EOF, we need update
7380 * the isize, but it is protected by i_mutex. So we can
7381 * not unlock the i_mutex at this case.
7382 */
7383 if (offset + count <= inode->i_size) {
7384 mutex_unlock(&inode->i_mutex);
7385 relock = true;
7386 }
7387 ret = btrfs_delalloc_reserve_space(inode, count);
7388 if (ret)
7389 goto out;
7390 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7391 &BTRFS_I(inode)->runtime_flags))) {
7392 inode_dio_done(inode);
7393 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7394 wakeup = false;
7395 }
7396
7397 ret = __blockdev_direct_IO(rw, iocb, inode,
7398 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7399 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7400 btrfs_submit_direct, flags);
7401 if (rw & WRITE) {
7402 if (ret < 0 && ret != -EIOCBQUEUED)
7403 btrfs_delalloc_release_space(inode, count);
7404 else if (ret >= 0 && (size_t)ret < count)
7405 btrfs_delalloc_release_space(inode,
7406 count - (size_t)ret);
7407 else
7408 btrfs_delalloc_release_metadata(inode, 0);
7409 }
7410 out:
7411 if (wakeup)
7412 inode_dio_done(inode);
7413 if (relock)
7414 mutex_lock(&inode->i_mutex);
7415
7416 return ret;
7417 }
7418
7419 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7420
7421 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7422 __u64 start, __u64 len)
7423 {
7424 int ret;
7425
7426 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7427 if (ret)
7428 return ret;
7429
7430 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7431 }
7432
7433 int btrfs_readpage(struct file *file, struct page *page)
7434 {
7435 struct extent_io_tree *tree;
7436 tree = &BTRFS_I(page->mapping->host)->io_tree;
7437 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7438 }
7439
7440 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7441 {
7442 struct extent_io_tree *tree;
7443
7444
7445 if (current->flags & PF_MEMALLOC) {
7446 redirty_page_for_writepage(wbc, page);
7447 unlock_page(page);
7448 return 0;
7449 }
7450 tree = &BTRFS_I(page->mapping->host)->io_tree;
7451 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7452 }
7453
7454 static int btrfs_writepages(struct address_space *mapping,
7455 struct writeback_control *wbc)
7456 {
7457 struct extent_io_tree *tree;
7458
7459 tree = &BTRFS_I(mapping->host)->io_tree;
7460 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7461 }
7462
7463 static int
7464 btrfs_readpages(struct file *file, struct address_space *mapping,
7465 struct list_head *pages, unsigned nr_pages)
7466 {
7467 struct extent_io_tree *tree;
7468 tree = &BTRFS_I(mapping->host)->io_tree;
7469 return extent_readpages(tree, mapping, pages, nr_pages,
7470 btrfs_get_extent);
7471 }
7472 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7473 {
7474 struct extent_io_tree *tree;
7475 struct extent_map_tree *map;
7476 int ret;
7477
7478 tree = &BTRFS_I(page->mapping->host)->io_tree;
7479 map = &BTRFS_I(page->mapping->host)->extent_tree;
7480 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7481 if (ret == 1) {
7482 ClearPagePrivate(page);
7483 set_page_private(page, 0);
7484 page_cache_release(page);
7485 }
7486 return ret;
7487 }
7488
7489 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7490 {
7491 if (PageWriteback(page) || PageDirty(page))
7492 return 0;
7493 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7494 }
7495
7496 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7497 unsigned int length)
7498 {
7499 struct inode *inode = page->mapping->host;
7500 struct extent_io_tree *tree;
7501 struct btrfs_ordered_extent *ordered;
7502 struct extent_state *cached_state = NULL;
7503 u64 page_start = page_offset(page);
7504 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7505
7506 /*
7507 * we have the page locked, so new writeback can't start,
7508 * and the dirty bit won't be cleared while we are here.
7509 *
7510 * Wait for IO on this page so that we can safely clear
7511 * the PagePrivate2 bit and do ordered accounting
7512 */
7513 wait_on_page_writeback(page);
7514
7515 tree = &BTRFS_I(inode)->io_tree;
7516 if (offset) {
7517 btrfs_releasepage(page, GFP_NOFS);
7518 return;
7519 }
7520 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7521 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7522 if (ordered) {
7523 /*
7524 * IO on this page will never be started, so we need
7525 * to account for any ordered extents now
7526 */
7527 clear_extent_bit(tree, page_start, page_end,
7528 EXTENT_DIRTY | EXTENT_DELALLOC |
7529 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7530 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7531 /*
7532 * whoever cleared the private bit is responsible
7533 * for the finish_ordered_io
7534 */
7535 if (TestClearPagePrivate2(page) &&
7536 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7537 PAGE_CACHE_SIZE, 1)) {
7538 btrfs_finish_ordered_io(ordered);
7539 }
7540 btrfs_put_ordered_extent(ordered);
7541 cached_state = NULL;
7542 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7543 }
7544 clear_extent_bit(tree, page_start, page_end,
7545 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7546 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7547 &cached_state, GFP_NOFS);
7548 __btrfs_releasepage(page, GFP_NOFS);
7549
7550 ClearPageChecked(page);
7551 if (PagePrivate(page)) {
7552 ClearPagePrivate(page);
7553 set_page_private(page, 0);
7554 page_cache_release(page);
7555 }
7556 }
7557
7558 /*
7559 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7560 * called from a page fault handler when a page is first dirtied. Hence we must
7561 * be careful to check for EOF conditions here. We set the page up correctly
7562 * for a written page which means we get ENOSPC checking when writing into
7563 * holes and correct delalloc and unwritten extent mapping on filesystems that
7564 * support these features.
7565 *
7566 * We are not allowed to take the i_mutex here so we have to play games to
7567 * protect against truncate races as the page could now be beyond EOF. Because
7568 * vmtruncate() writes the inode size before removing pages, once we have the
7569 * page lock we can determine safely if the page is beyond EOF. If it is not
7570 * beyond EOF, then the page is guaranteed safe against truncation until we
7571 * unlock the page.
7572 */
7573 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7574 {
7575 struct page *page = vmf->page;
7576 struct inode *inode = file_inode(vma->vm_file);
7577 struct btrfs_root *root = BTRFS_I(inode)->root;
7578 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7579 struct btrfs_ordered_extent *ordered;
7580 struct extent_state *cached_state = NULL;
7581 char *kaddr;
7582 unsigned long zero_start;
7583 loff_t size;
7584 int ret;
7585 int reserved = 0;
7586 u64 page_start;
7587 u64 page_end;
7588
7589 sb_start_pagefault(inode->i_sb);
7590 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7591 if (!ret) {
7592 ret = file_update_time(vma->vm_file);
7593 reserved = 1;
7594 }
7595 if (ret) {
7596 if (ret == -ENOMEM)
7597 ret = VM_FAULT_OOM;
7598 else /* -ENOSPC, -EIO, etc */
7599 ret = VM_FAULT_SIGBUS;
7600 if (reserved)
7601 goto out;
7602 goto out_noreserve;
7603 }
7604
7605 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7606 again:
7607 lock_page(page);
7608 size = i_size_read(inode);
7609 page_start = page_offset(page);
7610 page_end = page_start + PAGE_CACHE_SIZE - 1;
7611
7612 if ((page->mapping != inode->i_mapping) ||
7613 (page_start >= size)) {
7614 /* page got truncated out from underneath us */
7615 goto out_unlock;
7616 }
7617 wait_on_page_writeback(page);
7618
7619 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7620 set_page_extent_mapped(page);
7621
7622 /*
7623 * we can't set the delalloc bits if there are pending ordered
7624 * extents. Drop our locks and wait for them to finish
7625 */
7626 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7627 if (ordered) {
7628 unlock_extent_cached(io_tree, page_start, page_end,
7629 &cached_state, GFP_NOFS);
7630 unlock_page(page);
7631 btrfs_start_ordered_extent(inode, ordered, 1);
7632 btrfs_put_ordered_extent(ordered);
7633 goto again;
7634 }
7635
7636 /*
7637 * XXX - page_mkwrite gets called every time the page is dirtied, even
7638 * if it was already dirty, so for space accounting reasons we need to
7639 * clear any delalloc bits for the range we are fixing to save. There
7640 * is probably a better way to do this, but for now keep consistent with
7641 * prepare_pages in the normal write path.
7642 */
7643 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7644 EXTENT_DIRTY | EXTENT_DELALLOC |
7645 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7646 0, 0, &cached_state, GFP_NOFS);
7647
7648 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7649 &cached_state);
7650 if (ret) {
7651 unlock_extent_cached(io_tree, page_start, page_end,
7652 &cached_state, GFP_NOFS);
7653 ret = VM_FAULT_SIGBUS;
7654 goto out_unlock;
7655 }
7656 ret = 0;
7657
7658 /* page is wholly or partially inside EOF */
7659 if (page_start + PAGE_CACHE_SIZE > size)
7660 zero_start = size & ~PAGE_CACHE_MASK;
7661 else
7662 zero_start = PAGE_CACHE_SIZE;
7663
7664 if (zero_start != PAGE_CACHE_SIZE) {
7665 kaddr = kmap(page);
7666 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7667 flush_dcache_page(page);
7668 kunmap(page);
7669 }
7670 ClearPageChecked(page);
7671 set_page_dirty(page);
7672 SetPageUptodate(page);
7673
7674 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7675 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7676 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7677
7678 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7679
7680 out_unlock:
7681 if (!ret) {
7682 sb_end_pagefault(inode->i_sb);
7683 return VM_FAULT_LOCKED;
7684 }
7685 unlock_page(page);
7686 out:
7687 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7688 out_noreserve:
7689 sb_end_pagefault(inode->i_sb);
7690 return ret;
7691 }
7692
7693 static int btrfs_truncate(struct inode *inode)
7694 {
7695 struct btrfs_root *root = BTRFS_I(inode)->root;
7696 struct btrfs_block_rsv *rsv;
7697 int ret;
7698 int err = 0;
7699 struct btrfs_trans_handle *trans;
7700 u64 mask = root->sectorsize - 1;
7701 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7702
7703 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7704 if (ret)
7705 return ret;
7706
7707 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7708 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7709
7710 /*
7711 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7712 * 3 things going on here
7713 *
7714 * 1) We need to reserve space for our orphan item and the space to
7715 * delete our orphan item. Lord knows we don't want to have a dangling
7716 * orphan item because we didn't reserve space to remove it.
7717 *
7718 * 2) We need to reserve space to update our inode.
7719 *
7720 * 3) We need to have something to cache all the space that is going to
7721 * be free'd up by the truncate operation, but also have some slack
7722 * space reserved in case it uses space during the truncate (thank you
7723 * very much snapshotting).
7724 *
7725 * And we need these to all be seperate. The fact is we can use alot of
7726 * space doing the truncate, and we have no earthly idea how much space
7727 * we will use, so we need the truncate reservation to be seperate so it
7728 * doesn't end up using space reserved for updating the inode or
7729 * removing the orphan item. We also need to be able to stop the
7730 * transaction and start a new one, which means we need to be able to
7731 * update the inode several times, and we have no idea of knowing how
7732 * many times that will be, so we can't just reserve 1 item for the
7733 * entirety of the opration, so that has to be done seperately as well.
7734 * Then there is the orphan item, which does indeed need to be held on
7735 * to for the whole operation, and we need nobody to touch this reserved
7736 * space except the orphan code.
7737 *
7738 * So that leaves us with
7739 *
7740 * 1) root->orphan_block_rsv - for the orphan deletion.
7741 * 2) rsv - for the truncate reservation, which we will steal from the
7742 * transaction reservation.
7743 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7744 * updating the inode.
7745 */
7746 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7747 if (!rsv)
7748 return -ENOMEM;
7749 rsv->size = min_size;
7750 rsv->failfast = 1;
7751
7752 /*
7753 * 1 for the truncate slack space
7754 * 1 for updating the inode.
7755 */
7756 trans = btrfs_start_transaction(root, 2);
7757 if (IS_ERR(trans)) {
7758 err = PTR_ERR(trans);
7759 goto out;
7760 }
7761
7762 /* Migrate the slack space for the truncate to our reserve */
7763 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7764 min_size);
7765 BUG_ON(ret);
7766
7767 /*
7768 * setattr is responsible for setting the ordered_data_close flag,
7769 * but that is only tested during the last file release. That
7770 * could happen well after the next commit, leaving a great big
7771 * window where new writes may get lost if someone chooses to write
7772 * to this file after truncating to zero
7773 *
7774 * The inode doesn't have any dirty data here, and so if we commit
7775 * this is a noop. If someone immediately starts writing to the inode
7776 * it is very likely we'll catch some of their writes in this
7777 * transaction, and the commit will find this file on the ordered
7778 * data list with good things to send down.
7779 *
7780 * This is a best effort solution, there is still a window where
7781 * using truncate to replace the contents of the file will
7782 * end up with a zero length file after a crash.
7783 */
7784 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7785 &BTRFS_I(inode)->runtime_flags))
7786 btrfs_add_ordered_operation(trans, root, inode);
7787
7788 /*
7789 * So if we truncate and then write and fsync we normally would just
7790 * write the extents that changed, which is a problem if we need to
7791 * first truncate that entire inode. So set this flag so we write out
7792 * all of the extents in the inode to the sync log so we're completely
7793 * safe.
7794 */
7795 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7796 trans->block_rsv = rsv;
7797
7798 while (1) {
7799 ret = btrfs_truncate_inode_items(trans, root, inode,
7800 inode->i_size,
7801 BTRFS_EXTENT_DATA_KEY);
7802 if (ret != -ENOSPC) {
7803 err = ret;
7804 break;
7805 }
7806
7807 trans->block_rsv = &root->fs_info->trans_block_rsv;
7808 ret = btrfs_update_inode(trans, root, inode);
7809 if (ret) {
7810 err = ret;
7811 break;
7812 }
7813
7814 btrfs_end_transaction(trans, root);
7815 btrfs_btree_balance_dirty(root);
7816
7817 trans = btrfs_start_transaction(root, 2);
7818 if (IS_ERR(trans)) {
7819 ret = err = PTR_ERR(trans);
7820 trans = NULL;
7821 break;
7822 }
7823
7824 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7825 rsv, min_size);
7826 BUG_ON(ret); /* shouldn't happen */
7827 trans->block_rsv = rsv;
7828 }
7829
7830 if (ret == 0 && inode->i_nlink > 0) {
7831 trans->block_rsv = root->orphan_block_rsv;
7832 ret = btrfs_orphan_del(trans, inode);
7833 if (ret)
7834 err = ret;
7835 }
7836
7837 if (trans) {
7838 trans->block_rsv = &root->fs_info->trans_block_rsv;
7839 ret = btrfs_update_inode(trans, root, inode);
7840 if (ret && !err)
7841 err = ret;
7842
7843 ret = btrfs_end_transaction(trans, root);
7844 btrfs_btree_balance_dirty(root);
7845 }
7846
7847 out:
7848 btrfs_free_block_rsv(root, rsv);
7849
7850 if (ret && !err)
7851 err = ret;
7852
7853 return err;
7854 }
7855
7856 /*
7857 * create a new subvolume directory/inode (helper for the ioctl).
7858 */
7859 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7860 struct btrfs_root *new_root, u64 new_dirid)
7861 {
7862 struct inode *inode;
7863 int err;
7864 u64 index = 0;
7865
7866 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7867 new_dirid, new_dirid,
7868 S_IFDIR | (~current_umask() & S_IRWXUGO),
7869 &index);
7870 if (IS_ERR(inode))
7871 return PTR_ERR(inode);
7872 inode->i_op = &btrfs_dir_inode_operations;
7873 inode->i_fop = &btrfs_dir_file_operations;
7874
7875 set_nlink(inode, 1);
7876 btrfs_i_size_write(inode, 0);
7877
7878 err = btrfs_update_inode(trans, new_root, inode);
7879
7880 iput(inode);
7881 return err;
7882 }
7883
7884 struct inode *btrfs_alloc_inode(struct super_block *sb)
7885 {
7886 struct btrfs_inode *ei;
7887 struct inode *inode;
7888
7889 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7890 if (!ei)
7891 return NULL;
7892
7893 ei->root = NULL;
7894 ei->generation = 0;
7895 ei->last_trans = 0;
7896 ei->last_sub_trans = 0;
7897 ei->logged_trans = 0;
7898 ei->delalloc_bytes = 0;
7899 ei->disk_i_size = 0;
7900 ei->flags = 0;
7901 ei->csum_bytes = 0;
7902 ei->index_cnt = (u64)-1;
7903 ei->last_unlink_trans = 0;
7904 ei->last_log_commit = 0;
7905
7906 spin_lock_init(&ei->lock);
7907 ei->outstanding_extents = 0;
7908 ei->reserved_extents = 0;
7909
7910 ei->runtime_flags = 0;
7911 ei->force_compress = BTRFS_COMPRESS_NONE;
7912
7913 ei->delayed_node = NULL;
7914
7915 inode = &ei->vfs_inode;
7916 extent_map_tree_init(&ei->extent_tree);
7917 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7918 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7919 ei->io_tree.track_uptodate = 1;
7920 ei->io_failure_tree.track_uptodate = 1;
7921 atomic_set(&ei->sync_writers, 0);
7922 mutex_init(&ei->log_mutex);
7923 mutex_init(&ei->delalloc_mutex);
7924 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7925 INIT_LIST_HEAD(&ei->delalloc_inodes);
7926 INIT_LIST_HEAD(&ei->ordered_operations);
7927 RB_CLEAR_NODE(&ei->rb_node);
7928
7929 return inode;
7930 }
7931
7932 static void btrfs_i_callback(struct rcu_head *head)
7933 {
7934 struct inode *inode = container_of(head, struct inode, i_rcu);
7935 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7936 }
7937
7938 void btrfs_destroy_inode(struct inode *inode)
7939 {
7940 struct btrfs_ordered_extent *ordered;
7941 struct btrfs_root *root = BTRFS_I(inode)->root;
7942
7943 WARN_ON(!hlist_empty(&inode->i_dentry));
7944 WARN_ON(inode->i_data.nrpages);
7945 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7946 WARN_ON(BTRFS_I(inode)->reserved_extents);
7947 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7948 WARN_ON(BTRFS_I(inode)->csum_bytes);
7949
7950 /*
7951 * This can happen where we create an inode, but somebody else also
7952 * created the same inode and we need to destroy the one we already
7953 * created.
7954 */
7955 if (!root)
7956 goto free;
7957
7958 /*
7959 * Make sure we're properly removed from the ordered operation
7960 * lists.
7961 */
7962 smp_mb();
7963 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7964 spin_lock(&root->fs_info->ordered_extent_lock);
7965 list_del_init(&BTRFS_I(inode)->ordered_operations);
7966 spin_unlock(&root->fs_info->ordered_extent_lock);
7967 }
7968
7969 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7970 &BTRFS_I(inode)->runtime_flags)) {
7971 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7972 (unsigned long long)btrfs_ino(inode));
7973 atomic_dec(&root->orphan_inodes);
7974 }
7975
7976 while (1) {
7977 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7978 if (!ordered)
7979 break;
7980 else {
7981 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7982 (unsigned long long)ordered->file_offset,
7983 (unsigned long long)ordered->len);
7984 btrfs_remove_ordered_extent(inode, ordered);
7985 btrfs_put_ordered_extent(ordered);
7986 btrfs_put_ordered_extent(ordered);
7987 }
7988 }
7989 inode_tree_del(inode);
7990 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7991 free:
7992 call_rcu(&inode->i_rcu, btrfs_i_callback);
7993 }
7994
7995 int btrfs_drop_inode(struct inode *inode)
7996 {
7997 struct btrfs_root *root = BTRFS_I(inode)->root;
7998
7999 if (root == NULL)
8000 return 1;
8001
8002 /* the snap/subvol tree is on deleting */
8003 if (btrfs_root_refs(&root->root_item) == 0 &&
8004 root != root->fs_info->tree_root)
8005 return 1;
8006 else
8007 return generic_drop_inode(inode);
8008 }
8009
8010 static void init_once(void *foo)
8011 {
8012 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8013
8014 inode_init_once(&ei->vfs_inode);
8015 }
8016
8017 void btrfs_destroy_cachep(void)
8018 {
8019 /*
8020 * Make sure all delayed rcu free inodes are flushed before we
8021 * destroy cache.
8022 */
8023 rcu_barrier();
8024 if (btrfs_inode_cachep)
8025 kmem_cache_destroy(btrfs_inode_cachep);
8026 if (btrfs_trans_handle_cachep)
8027 kmem_cache_destroy(btrfs_trans_handle_cachep);
8028 if (btrfs_transaction_cachep)
8029 kmem_cache_destroy(btrfs_transaction_cachep);
8030 if (btrfs_path_cachep)
8031 kmem_cache_destroy(btrfs_path_cachep);
8032 if (btrfs_free_space_cachep)
8033 kmem_cache_destroy(btrfs_free_space_cachep);
8034 if (btrfs_delalloc_work_cachep)
8035 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8036 }
8037
8038 int btrfs_init_cachep(void)
8039 {
8040 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8041 sizeof(struct btrfs_inode), 0,
8042 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8043 if (!btrfs_inode_cachep)
8044 goto fail;
8045
8046 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8047 sizeof(struct btrfs_trans_handle), 0,
8048 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8049 if (!btrfs_trans_handle_cachep)
8050 goto fail;
8051
8052 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8053 sizeof(struct btrfs_transaction), 0,
8054 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8055 if (!btrfs_transaction_cachep)
8056 goto fail;
8057
8058 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8059 sizeof(struct btrfs_path), 0,
8060 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8061 if (!btrfs_path_cachep)
8062 goto fail;
8063
8064 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8065 sizeof(struct btrfs_free_space), 0,
8066 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8067 if (!btrfs_free_space_cachep)
8068 goto fail;
8069
8070 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8071 sizeof(struct btrfs_delalloc_work), 0,
8072 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8073 NULL);
8074 if (!btrfs_delalloc_work_cachep)
8075 goto fail;
8076
8077 return 0;
8078 fail:
8079 btrfs_destroy_cachep();
8080 return -ENOMEM;
8081 }
8082
8083 static int btrfs_getattr(struct vfsmount *mnt,
8084 struct dentry *dentry, struct kstat *stat)
8085 {
8086 u64 delalloc_bytes;
8087 struct inode *inode = dentry->d_inode;
8088 u32 blocksize = inode->i_sb->s_blocksize;
8089
8090 generic_fillattr(inode, stat);
8091 stat->dev = BTRFS_I(inode)->root->anon_dev;
8092 stat->blksize = PAGE_CACHE_SIZE;
8093
8094 spin_lock(&BTRFS_I(inode)->lock);
8095 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8096 spin_unlock(&BTRFS_I(inode)->lock);
8097 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8098 ALIGN(delalloc_bytes, blocksize)) >> 9;
8099 return 0;
8100 }
8101
8102 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8103 struct inode *new_dir, struct dentry *new_dentry)
8104 {
8105 struct btrfs_trans_handle *trans;
8106 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8107 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8108 struct inode *new_inode = new_dentry->d_inode;
8109 struct inode *old_inode = old_dentry->d_inode;
8110 struct timespec ctime = CURRENT_TIME;
8111 u64 index = 0;
8112 u64 root_objectid;
8113 int ret;
8114 u64 old_ino = btrfs_ino(old_inode);
8115
8116 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8117 return -EPERM;
8118
8119 /* we only allow rename subvolume link between subvolumes */
8120 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8121 return -EXDEV;
8122
8123 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8124 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8125 return -ENOTEMPTY;
8126
8127 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8128 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8129 return -ENOTEMPTY;
8130
8131
8132 /* check for collisions, even if the name isn't there */
8133 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8134 new_dentry->d_name.name,
8135 new_dentry->d_name.len);
8136
8137 if (ret) {
8138 if (ret == -EEXIST) {
8139 /* we shouldn't get
8140 * eexist without a new_inode */
8141 if (!new_inode) {
8142 WARN_ON(1);
8143 return ret;
8144 }
8145 } else {
8146 /* maybe -EOVERFLOW */
8147 return ret;
8148 }
8149 }
8150 ret = 0;
8151
8152 /*
8153 * we're using rename to replace one file with another.
8154 * and the replacement file is large. Start IO on it now so
8155 * we don't add too much work to the end of the transaction
8156 */
8157 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8158 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8159 filemap_flush(old_inode->i_mapping);
8160
8161 /* close the racy window with snapshot create/destroy ioctl */
8162 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8163 down_read(&root->fs_info->subvol_sem);
8164 /*
8165 * We want to reserve the absolute worst case amount of items. So if
8166 * both inodes are subvols and we need to unlink them then that would
8167 * require 4 item modifications, but if they are both normal inodes it
8168 * would require 5 item modifications, so we'll assume their normal
8169 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8170 * should cover the worst case number of items we'll modify.
8171 */
8172 trans = btrfs_start_transaction(root, 11);
8173 if (IS_ERR(trans)) {
8174 ret = PTR_ERR(trans);
8175 goto out_notrans;
8176 }
8177
8178 if (dest != root)
8179 btrfs_record_root_in_trans(trans, dest);
8180
8181 ret = btrfs_set_inode_index(new_dir, &index);
8182 if (ret)
8183 goto out_fail;
8184
8185 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8186 /* force full log commit if subvolume involved. */
8187 root->fs_info->last_trans_log_full_commit = trans->transid;
8188 } else {
8189 ret = btrfs_insert_inode_ref(trans, dest,
8190 new_dentry->d_name.name,
8191 new_dentry->d_name.len,
8192 old_ino,
8193 btrfs_ino(new_dir), index);
8194 if (ret)
8195 goto out_fail;
8196 /*
8197 * this is an ugly little race, but the rename is required
8198 * to make sure that if we crash, the inode is either at the
8199 * old name or the new one. pinning the log transaction lets
8200 * us make sure we don't allow a log commit to come in after
8201 * we unlink the name but before we add the new name back in.
8202 */
8203 btrfs_pin_log_trans(root);
8204 }
8205 /*
8206 * make sure the inode gets flushed if it is replacing
8207 * something.
8208 */
8209 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8210 btrfs_add_ordered_operation(trans, root, old_inode);
8211
8212 inode_inc_iversion(old_dir);
8213 inode_inc_iversion(new_dir);
8214 inode_inc_iversion(old_inode);
8215 old_dir->i_ctime = old_dir->i_mtime = ctime;
8216 new_dir->i_ctime = new_dir->i_mtime = ctime;
8217 old_inode->i_ctime = ctime;
8218
8219 if (old_dentry->d_parent != new_dentry->d_parent)
8220 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8221
8222 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8223 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8224 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8225 old_dentry->d_name.name,
8226 old_dentry->d_name.len);
8227 } else {
8228 ret = __btrfs_unlink_inode(trans, root, old_dir,
8229 old_dentry->d_inode,
8230 old_dentry->d_name.name,
8231 old_dentry->d_name.len);
8232 if (!ret)
8233 ret = btrfs_update_inode(trans, root, old_inode);
8234 }
8235 if (ret) {
8236 btrfs_abort_transaction(trans, root, ret);
8237 goto out_fail;
8238 }
8239
8240 if (new_inode) {
8241 inode_inc_iversion(new_inode);
8242 new_inode->i_ctime = CURRENT_TIME;
8243 if (unlikely(btrfs_ino(new_inode) ==
8244 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8245 root_objectid = BTRFS_I(new_inode)->location.objectid;
8246 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8247 root_objectid,
8248 new_dentry->d_name.name,
8249 new_dentry->d_name.len);
8250 BUG_ON(new_inode->i_nlink == 0);
8251 } else {
8252 ret = btrfs_unlink_inode(trans, dest, new_dir,
8253 new_dentry->d_inode,
8254 new_dentry->d_name.name,
8255 new_dentry->d_name.len);
8256 }
8257 if (!ret && new_inode->i_nlink == 0) {
8258 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8259 BUG_ON(ret);
8260 }
8261 if (ret) {
8262 btrfs_abort_transaction(trans, root, ret);
8263 goto out_fail;
8264 }
8265 }
8266
8267 ret = btrfs_add_link(trans, new_dir, old_inode,
8268 new_dentry->d_name.name,
8269 new_dentry->d_name.len, 0, index);
8270 if (ret) {
8271 btrfs_abort_transaction(trans, root, ret);
8272 goto out_fail;
8273 }
8274
8275 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8276 struct dentry *parent = new_dentry->d_parent;
8277 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8278 btrfs_end_log_trans(root);
8279 }
8280 out_fail:
8281 btrfs_end_transaction(trans, root);
8282 out_notrans:
8283 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8284 up_read(&root->fs_info->subvol_sem);
8285
8286 return ret;
8287 }
8288
8289 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8290 {
8291 struct btrfs_delalloc_work *delalloc_work;
8292
8293 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8294 work);
8295 if (delalloc_work->wait)
8296 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8297 else
8298 filemap_flush(delalloc_work->inode->i_mapping);
8299
8300 if (delalloc_work->delay_iput)
8301 btrfs_add_delayed_iput(delalloc_work->inode);
8302 else
8303 iput(delalloc_work->inode);
8304 complete(&delalloc_work->completion);
8305 }
8306
8307 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8308 int wait, int delay_iput)
8309 {
8310 struct btrfs_delalloc_work *work;
8311
8312 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8313 if (!work)
8314 return NULL;
8315
8316 init_completion(&work->completion);
8317 INIT_LIST_HEAD(&work->list);
8318 work->inode = inode;
8319 work->wait = wait;
8320 work->delay_iput = delay_iput;
8321 work->work.func = btrfs_run_delalloc_work;
8322
8323 return work;
8324 }
8325
8326 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8327 {
8328 wait_for_completion(&work->completion);
8329 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8330 }
8331
8332 /*
8333 * some fairly slow code that needs optimization. This walks the list
8334 * of all the inodes with pending delalloc and forces them to disk.
8335 */
8336 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8337 {
8338 struct btrfs_inode *binode;
8339 struct inode *inode;
8340 struct btrfs_delalloc_work *work, *next;
8341 struct list_head works;
8342 struct list_head splice;
8343 int ret = 0;
8344
8345 if (root->fs_info->sb->s_flags & MS_RDONLY)
8346 return -EROFS;
8347
8348 INIT_LIST_HEAD(&works);
8349 INIT_LIST_HEAD(&splice);
8350
8351 spin_lock(&root->fs_info->delalloc_lock);
8352 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8353 while (!list_empty(&splice)) {
8354 binode = list_entry(splice.next, struct btrfs_inode,
8355 delalloc_inodes);
8356
8357 list_del_init(&binode->delalloc_inodes);
8358
8359 inode = igrab(&binode->vfs_inode);
8360 if (!inode) {
8361 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8362 &binode->runtime_flags);
8363 continue;
8364 }
8365
8366 list_add_tail(&binode->delalloc_inodes,
8367 &root->fs_info->delalloc_inodes);
8368 spin_unlock(&root->fs_info->delalloc_lock);
8369
8370 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8371 if (unlikely(!work)) {
8372 ret = -ENOMEM;
8373 goto out;
8374 }
8375 list_add_tail(&work->list, &works);
8376 btrfs_queue_worker(&root->fs_info->flush_workers,
8377 &work->work);
8378
8379 cond_resched();
8380 spin_lock(&root->fs_info->delalloc_lock);
8381 }
8382 spin_unlock(&root->fs_info->delalloc_lock);
8383
8384 list_for_each_entry_safe(work, next, &works, list) {
8385 list_del_init(&work->list);
8386 btrfs_wait_and_free_delalloc_work(work);
8387 }
8388
8389 /* the filemap_flush will queue IO into the worker threads, but
8390 * we have to make sure the IO is actually started and that
8391 * ordered extents get created before we return
8392 */
8393 atomic_inc(&root->fs_info->async_submit_draining);
8394 while (atomic_read(&root->fs_info->nr_async_submits) ||
8395 atomic_read(&root->fs_info->async_delalloc_pages)) {
8396 wait_event(root->fs_info->async_submit_wait,
8397 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8398 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8399 }
8400 atomic_dec(&root->fs_info->async_submit_draining);
8401 return 0;
8402 out:
8403 list_for_each_entry_safe(work, next, &works, list) {
8404 list_del_init(&work->list);
8405 btrfs_wait_and_free_delalloc_work(work);
8406 }
8407
8408 if (!list_empty_careful(&splice)) {
8409 spin_lock(&root->fs_info->delalloc_lock);
8410 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8411 spin_unlock(&root->fs_info->delalloc_lock);
8412 }
8413 return ret;
8414 }
8415
8416 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8417 const char *symname)
8418 {
8419 struct btrfs_trans_handle *trans;
8420 struct btrfs_root *root = BTRFS_I(dir)->root;
8421 struct btrfs_path *path;
8422 struct btrfs_key key;
8423 struct inode *inode = NULL;
8424 int err;
8425 int drop_inode = 0;
8426 u64 objectid;
8427 u64 index = 0 ;
8428 int name_len;
8429 int datasize;
8430 unsigned long ptr;
8431 struct btrfs_file_extent_item *ei;
8432 struct extent_buffer *leaf;
8433
8434 name_len = strlen(symname) + 1;
8435 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8436 return -ENAMETOOLONG;
8437
8438 /*
8439 * 2 items for inode item and ref
8440 * 2 items for dir items
8441 * 1 item for xattr if selinux is on
8442 */
8443 trans = btrfs_start_transaction(root, 5);
8444 if (IS_ERR(trans))
8445 return PTR_ERR(trans);
8446
8447 err = btrfs_find_free_ino(root, &objectid);
8448 if (err)
8449 goto out_unlock;
8450
8451 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8452 dentry->d_name.len, btrfs_ino(dir), objectid,
8453 S_IFLNK|S_IRWXUGO, &index);
8454 if (IS_ERR(inode)) {
8455 err = PTR_ERR(inode);
8456 goto out_unlock;
8457 }
8458
8459 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8460 if (err) {
8461 drop_inode = 1;
8462 goto out_unlock;
8463 }
8464
8465 /*
8466 * If the active LSM wants to access the inode during
8467 * d_instantiate it needs these. Smack checks to see
8468 * if the filesystem supports xattrs by looking at the
8469 * ops vector.
8470 */
8471 inode->i_fop = &btrfs_file_operations;
8472 inode->i_op = &btrfs_file_inode_operations;
8473
8474 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8475 if (err)
8476 drop_inode = 1;
8477 else {
8478 inode->i_mapping->a_ops = &btrfs_aops;
8479 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8480 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8481 }
8482 if (drop_inode)
8483 goto out_unlock;
8484
8485 path = btrfs_alloc_path();
8486 if (!path) {
8487 err = -ENOMEM;
8488 drop_inode = 1;
8489 goto out_unlock;
8490 }
8491 key.objectid = btrfs_ino(inode);
8492 key.offset = 0;
8493 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8494 datasize = btrfs_file_extent_calc_inline_size(name_len);
8495 err = btrfs_insert_empty_item(trans, root, path, &key,
8496 datasize);
8497 if (err) {
8498 drop_inode = 1;
8499 btrfs_free_path(path);
8500 goto out_unlock;
8501 }
8502 leaf = path->nodes[0];
8503 ei = btrfs_item_ptr(leaf, path->slots[0],
8504 struct btrfs_file_extent_item);
8505 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8506 btrfs_set_file_extent_type(leaf, ei,
8507 BTRFS_FILE_EXTENT_INLINE);
8508 btrfs_set_file_extent_encryption(leaf, ei, 0);
8509 btrfs_set_file_extent_compression(leaf, ei, 0);
8510 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8511 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8512
8513 ptr = btrfs_file_extent_inline_start(ei);
8514 write_extent_buffer(leaf, symname, ptr, name_len);
8515 btrfs_mark_buffer_dirty(leaf);
8516 btrfs_free_path(path);
8517
8518 inode->i_op = &btrfs_symlink_inode_operations;
8519 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8520 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8521 inode_set_bytes(inode, name_len);
8522 btrfs_i_size_write(inode, name_len - 1);
8523 err = btrfs_update_inode(trans, root, inode);
8524 if (err)
8525 drop_inode = 1;
8526
8527 out_unlock:
8528 if (!err)
8529 d_instantiate(dentry, inode);
8530 btrfs_end_transaction(trans, root);
8531 if (drop_inode) {
8532 inode_dec_link_count(inode);
8533 iput(inode);
8534 }
8535 btrfs_btree_balance_dirty(root);
8536 return err;
8537 }
8538
8539 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8540 u64 start, u64 num_bytes, u64 min_size,
8541 loff_t actual_len, u64 *alloc_hint,
8542 struct btrfs_trans_handle *trans)
8543 {
8544 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8545 struct extent_map *em;
8546 struct btrfs_root *root = BTRFS_I(inode)->root;
8547 struct btrfs_key ins;
8548 u64 cur_offset = start;
8549 u64 i_size;
8550 u64 cur_bytes;
8551 int ret = 0;
8552 bool own_trans = true;
8553
8554 if (trans)
8555 own_trans = false;
8556 while (num_bytes > 0) {
8557 if (own_trans) {
8558 trans = btrfs_start_transaction(root, 3);
8559 if (IS_ERR(trans)) {
8560 ret = PTR_ERR(trans);
8561 break;
8562 }
8563 }
8564
8565 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8566 cur_bytes = max(cur_bytes, min_size);
8567 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8568 min_size, 0, *alloc_hint, &ins, 1);
8569 if (ret) {
8570 if (own_trans)
8571 btrfs_end_transaction(trans, root);
8572 break;
8573 }
8574
8575 ret = insert_reserved_file_extent(trans, inode,
8576 cur_offset, ins.objectid,
8577 ins.offset, ins.offset,
8578 ins.offset, 0, 0, 0,
8579 BTRFS_FILE_EXTENT_PREALLOC);
8580 if (ret) {
8581 btrfs_abort_transaction(trans, root, ret);
8582 if (own_trans)
8583 btrfs_end_transaction(trans, root);
8584 break;
8585 }
8586 btrfs_drop_extent_cache(inode, cur_offset,
8587 cur_offset + ins.offset -1, 0);
8588
8589 em = alloc_extent_map();
8590 if (!em) {
8591 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8592 &BTRFS_I(inode)->runtime_flags);
8593 goto next;
8594 }
8595
8596 em->start = cur_offset;
8597 em->orig_start = cur_offset;
8598 em->len = ins.offset;
8599 em->block_start = ins.objectid;
8600 em->block_len = ins.offset;
8601 em->orig_block_len = ins.offset;
8602 em->ram_bytes = ins.offset;
8603 em->bdev = root->fs_info->fs_devices->latest_bdev;
8604 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8605 em->generation = trans->transid;
8606
8607 while (1) {
8608 write_lock(&em_tree->lock);
8609 ret = add_extent_mapping(em_tree, em, 1);
8610 write_unlock(&em_tree->lock);
8611 if (ret != -EEXIST)
8612 break;
8613 btrfs_drop_extent_cache(inode, cur_offset,
8614 cur_offset + ins.offset - 1,
8615 0);
8616 }
8617 free_extent_map(em);
8618 next:
8619 num_bytes -= ins.offset;
8620 cur_offset += ins.offset;
8621 *alloc_hint = ins.objectid + ins.offset;
8622
8623 inode_inc_iversion(inode);
8624 inode->i_ctime = CURRENT_TIME;
8625 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8626 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8627 (actual_len > inode->i_size) &&
8628 (cur_offset > inode->i_size)) {
8629 if (cur_offset > actual_len)
8630 i_size = actual_len;
8631 else
8632 i_size = cur_offset;
8633 i_size_write(inode, i_size);
8634 btrfs_ordered_update_i_size(inode, i_size, NULL);
8635 }
8636
8637 ret = btrfs_update_inode(trans, root, inode);
8638
8639 if (ret) {
8640 btrfs_abort_transaction(trans, root, ret);
8641 if (own_trans)
8642 btrfs_end_transaction(trans, root);
8643 break;
8644 }
8645
8646 if (own_trans)
8647 btrfs_end_transaction(trans, root);
8648 }
8649 return ret;
8650 }
8651
8652 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8653 u64 start, u64 num_bytes, u64 min_size,
8654 loff_t actual_len, u64 *alloc_hint)
8655 {
8656 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8657 min_size, actual_len, alloc_hint,
8658 NULL);
8659 }
8660
8661 int btrfs_prealloc_file_range_trans(struct inode *inode,
8662 struct btrfs_trans_handle *trans, int mode,
8663 u64 start, u64 num_bytes, u64 min_size,
8664 loff_t actual_len, u64 *alloc_hint)
8665 {
8666 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8667 min_size, actual_len, alloc_hint, trans);
8668 }
8669
8670 static int btrfs_set_page_dirty(struct page *page)
8671 {
8672 return __set_page_dirty_nobuffers(page);
8673 }
8674
8675 static int btrfs_permission(struct inode *inode, int mask)
8676 {
8677 struct btrfs_root *root = BTRFS_I(inode)->root;
8678 umode_t mode = inode->i_mode;
8679
8680 if (mask & MAY_WRITE &&
8681 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8682 if (btrfs_root_readonly(root))
8683 return -EROFS;
8684 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8685 return -EACCES;
8686 }
8687 return generic_permission(inode, mask);
8688 }
8689
8690 static const struct inode_operations btrfs_dir_inode_operations = {
8691 .getattr = btrfs_getattr,
8692 .lookup = btrfs_lookup,
8693 .create = btrfs_create,
8694 .unlink = btrfs_unlink,
8695 .link = btrfs_link,
8696 .mkdir = btrfs_mkdir,
8697 .rmdir = btrfs_rmdir,
8698 .rename = btrfs_rename,
8699 .symlink = btrfs_symlink,
8700 .setattr = btrfs_setattr,
8701 .mknod = btrfs_mknod,
8702 .setxattr = btrfs_setxattr,
8703 .getxattr = btrfs_getxattr,
8704 .listxattr = btrfs_listxattr,
8705 .removexattr = btrfs_removexattr,
8706 .permission = btrfs_permission,
8707 .get_acl = btrfs_get_acl,
8708 };
8709 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8710 .lookup = btrfs_lookup,
8711 .permission = btrfs_permission,
8712 .get_acl = btrfs_get_acl,
8713 };
8714
8715 static const struct file_operations btrfs_dir_file_operations = {
8716 .llseek = generic_file_llseek,
8717 .read = generic_read_dir,
8718 .iterate = btrfs_real_readdir,
8719 .unlocked_ioctl = btrfs_ioctl,
8720 #ifdef CONFIG_COMPAT
8721 .compat_ioctl = btrfs_ioctl,
8722 #endif
8723 .release = btrfs_release_file,
8724 .fsync = btrfs_sync_file,
8725 };
8726
8727 static struct extent_io_ops btrfs_extent_io_ops = {
8728 .fill_delalloc = run_delalloc_range,
8729 .submit_bio_hook = btrfs_submit_bio_hook,
8730 .merge_bio_hook = btrfs_merge_bio_hook,
8731 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8732 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8733 .writepage_start_hook = btrfs_writepage_start_hook,
8734 .set_bit_hook = btrfs_set_bit_hook,
8735 .clear_bit_hook = btrfs_clear_bit_hook,
8736 .merge_extent_hook = btrfs_merge_extent_hook,
8737 .split_extent_hook = btrfs_split_extent_hook,
8738 };
8739
8740 /*
8741 * btrfs doesn't support the bmap operation because swapfiles
8742 * use bmap to make a mapping of extents in the file. They assume
8743 * these extents won't change over the life of the file and they
8744 * use the bmap result to do IO directly to the drive.
8745 *
8746 * the btrfs bmap call would return logical addresses that aren't
8747 * suitable for IO and they also will change frequently as COW
8748 * operations happen. So, swapfile + btrfs == corruption.
8749 *
8750 * For now we're avoiding this by dropping bmap.
8751 */
8752 static const struct address_space_operations btrfs_aops = {
8753 .readpage = btrfs_readpage,
8754 .writepage = btrfs_writepage,
8755 .writepages = btrfs_writepages,
8756 .readpages = btrfs_readpages,
8757 .direct_IO = btrfs_direct_IO,
8758 .invalidatepage = btrfs_invalidatepage,
8759 .releasepage = btrfs_releasepage,
8760 .set_page_dirty = btrfs_set_page_dirty,
8761 .error_remove_page = generic_error_remove_page,
8762 };
8763
8764 static const struct address_space_operations btrfs_symlink_aops = {
8765 .readpage = btrfs_readpage,
8766 .writepage = btrfs_writepage,
8767 .invalidatepage = btrfs_invalidatepage,
8768 .releasepage = btrfs_releasepage,
8769 };
8770
8771 static const struct inode_operations btrfs_file_inode_operations = {
8772 .getattr = btrfs_getattr,
8773 .setattr = btrfs_setattr,
8774 .setxattr = btrfs_setxattr,
8775 .getxattr = btrfs_getxattr,
8776 .listxattr = btrfs_listxattr,
8777 .removexattr = btrfs_removexattr,
8778 .permission = btrfs_permission,
8779 .fiemap = btrfs_fiemap,
8780 .get_acl = btrfs_get_acl,
8781 .update_time = btrfs_update_time,
8782 };
8783 static const struct inode_operations btrfs_special_inode_operations = {
8784 .getattr = btrfs_getattr,
8785 .setattr = btrfs_setattr,
8786 .permission = btrfs_permission,
8787 .setxattr = btrfs_setxattr,
8788 .getxattr = btrfs_getxattr,
8789 .listxattr = btrfs_listxattr,
8790 .removexattr = btrfs_removexattr,
8791 .get_acl = btrfs_get_acl,
8792 .update_time = btrfs_update_time,
8793 };
8794 static const struct inode_operations btrfs_symlink_inode_operations = {
8795 .readlink = generic_readlink,
8796 .follow_link = page_follow_link_light,
8797 .put_link = page_put_link,
8798 .getattr = btrfs_getattr,
8799 .setattr = btrfs_setattr,
8800 .permission = btrfs_permission,
8801 .setxattr = btrfs_setxattr,
8802 .getxattr = btrfs_getxattr,
8803 .listxattr = btrfs_listxattr,
8804 .removexattr = btrfs_removexattr,
8805 .get_acl = btrfs_get_acl,
8806 .update_time = btrfs_update_time,
8807 };
8808
8809 const struct dentry_operations btrfs_dentry_operations = {
8810 .d_delete = btrfs_dentry_delete,
8811 .d_release = btrfs_dentry_release,
8812 };
Linux kernel - Deliverables
This page took 0.306538 seconds and 6 git commands to generate.