?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: convert printk to btrfs_ and fix BTRFS prefix
[linux-2.6/btrfs-unstable.git] / fs / btrfs / inode.c
blob06bcf5b53cb00dac7cc15a50538c2a93564f3982
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61
62 struct btrfs_iget_args {
63 u64 ino;
64 struct btrfs_root *root;
65 };
66
67 static const struct inode_operations btrfs_dir_inode_operations;
68 static const struct inode_operations btrfs_symlink_inode_operations;
69 static const struct inode_operations btrfs_dir_ro_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct address_space_operations btrfs_symlink_aops;
74 static const struct file_operations btrfs_dir_file_operations;
75 static struct extent_io_ops btrfs_extent_io_ops;
76
77 static struct kmem_cache *btrfs_inode_cachep;
78 static struct kmem_cache *btrfs_delalloc_work_cachep;
79 struct kmem_cache *btrfs_trans_handle_cachep;
80 struct kmem_cache *btrfs_transaction_cachep;
81 struct kmem_cache *btrfs_path_cachep;
82 struct kmem_cache *btrfs_free_space_cachep;
83
84 #define S_SHIFT 12
85 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
86 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
87 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
88 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
89 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
90 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
91 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
92 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
93 };
94
95 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
96 static int btrfs_truncate(struct inode *inode);
97 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
98 static noinline int cow_file_range(struct inode *inode,
99 struct page *locked_page,
100 u64 start, u64 end, int *page_started,
101 unsigned long *nr_written, int unlock);
102 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
103 u64 len, u64 orig_start,
104 u64 block_start, u64 block_len,
105 u64 orig_block_len, u64 ram_bytes,
106 int type);
107
108 static int btrfs_dirty_inode(struct inode *inode);
109
110 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
111 struct inode *inode, struct inode *dir,
112 const struct qstr *qstr)
113 {
114 int err;
115
116 err = btrfs_init_acl(trans, inode, dir);
117 if (!err)
118 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
119 return err;
120 }
121
122 /*
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
126 */
127 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
130 int compress_type,
131 struct page **compressed_pages)
132 {
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 size_t datasize;
144 unsigned long offset;
145
146 if (compressed_size && compressed_pages)
147 cur_size = compressed_size;
148
149 path = btrfs_alloc_path();
150 if (!path)
151 return -ENOMEM;
152
153 path->leave_spinning = 1;
154
155 key.objectid = btrfs_ino(inode);
156 key.offset = start;
157 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159
160 inode_add_bytes(inode, size);
161 ret = btrfs_insert_empty_item(trans, root, path, &key,
162 datasize);
163 if (ret) {
164 err = ret;
165 goto fail;
166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
177 if (compress_type != BTRFS_COMPRESS_NONE) {
178 struct page *cpage;
179 int i = 0;
180 while (compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min_t(unsigned long, compressed_size,
183 PAGE_CACHE_SIZE);
184
185 kaddr = kmap_atomic(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap_atomic(kaddr);
188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
194 compress_type);
195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr);
203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_free_path(path);
207
208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
217 BTRFS_I(inode)->disk_i_size = inode->i_size;
218 ret = btrfs_update_inode(trans, root, inode);
219
220 return ret;
221 fail:
222 btrfs_free_path(path);
223 return err;
224 }
225
226
227 /*
228 * conditionally insert an inline extent into the file. This
229 * does the checks required to make sure the data is small enough
230 * to fit as an inline extent.
231 */
232 static noinline int cow_file_range_inline(struct btrfs_root *root,
233 struct inode *inode, u64 start,
234 u64 end, size_t compressed_size,
235 int compress_type,
236 struct page **compressed_pages)
237 {
238 struct btrfs_trans_handle *trans;
239 u64 isize = i_size_read(inode);
240 u64 actual_end = min(end + 1, isize);
241 u64 inline_len = actual_end - start;
242 u64 aligned_end = ALIGN(end, root->sectorsize);
243 u64 data_len = inline_len;
244 int ret;
245
246 if (compressed_size)
247 data_len = compressed_size;
248
249 if (start > 0 ||
250 actual_end >= PAGE_CACHE_SIZE ||
251 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
252 (!compressed_size &&
253 (actual_end & (root->sectorsize - 1)) == 0) ||
254 end + 1 < isize ||
255 data_len > root->fs_info->max_inline) {
256 return 1;
257 }
258
259 trans = btrfs_join_transaction(root);
260 if (IS_ERR(trans))
261 return PTR_ERR(trans);
262 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
263
264 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
265 if (ret) {
266 btrfs_abort_transaction(trans, root, ret);
267 goto out;
268 }
269
270 if (isize > actual_end)
271 inline_len = min_t(u64, isize, actual_end);
272 ret = insert_inline_extent(trans, root, inode, start,
273 inline_len, compressed_size,
274 compress_type, compressed_pages);
275 if (ret && ret != -ENOSPC) {
276 btrfs_abort_transaction(trans, root, ret);
277 goto out;
278 } else if (ret == -ENOSPC) {
279 ret = 1;
280 goto out;
281 }
282
283 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
284 btrfs_delalloc_release_metadata(inode, end + 1 - start);
285 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
286 out:
287 btrfs_end_transaction(trans, root);
288 return ret;
289 }
290
291 struct async_extent {
292 u64 start;
293 u64 ram_size;
294 u64 compressed_size;
295 struct page **pages;
296 unsigned long nr_pages;
297 int compress_type;
298 struct list_head list;
299 };
300
301 struct async_cow {
302 struct inode *inode;
303 struct btrfs_root *root;
304 struct page *locked_page;
305 u64 start;
306 u64 end;
307 struct list_head extents;
308 struct btrfs_work work;
309 };
310
311 static noinline int add_async_extent(struct async_cow *cow,
312 u64 start, u64 ram_size,
313 u64 compressed_size,
314 struct page **pages,
315 unsigned long nr_pages,
316 int compress_type)
317 {
318 struct async_extent *async_extent;
319
320 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
321 BUG_ON(!async_extent); /* -ENOMEM */
322 async_extent->start = start;
323 async_extent->ram_size = ram_size;
324 async_extent->compressed_size = compressed_size;
325 async_extent->pages = pages;
326 async_extent->nr_pages = nr_pages;
327 async_extent->compress_type = compress_type;
328 list_add_tail(&async_extent->list, &cow->extents);
329 return 0;
330 }
331
332 /*
333 * we create compressed extents in two phases. The first
334 * phase compresses a range of pages that have already been
335 * locked (both pages and state bits are locked).
336 *
337 * This is done inside an ordered work queue, and the compression
338 * is spread across many cpus. The actual IO submission is step
339 * two, and the ordered work queue takes care of making sure that
340 * happens in the same order things were put onto the queue by
341 * writepages and friends.
342 *
343 * If this code finds it can't get good compression, it puts an
344 * entry onto the work queue to write the uncompressed bytes. This
345 * makes sure that both compressed inodes and uncompressed inodes
346 * are written in the same order that the flusher thread sent them
347 * down.
348 */
349 static noinline int compress_file_range(struct inode *inode,
350 struct page *locked_page,
351 u64 start, u64 end,
352 struct async_cow *async_cow,
353 int *num_added)
354 {
355 struct btrfs_root *root = BTRFS_I(inode)->root;
356 u64 num_bytes;
357 u64 blocksize = root->sectorsize;
358 u64 actual_end;
359 u64 isize = i_size_read(inode);
360 int ret = 0;
361 struct page **pages = NULL;
362 unsigned long nr_pages;
363 unsigned long nr_pages_ret = 0;
364 unsigned long total_compressed = 0;
365 unsigned long total_in = 0;
366 unsigned long max_compressed = 128 * 1024;
367 unsigned long max_uncompressed = 128 * 1024;
368 int i;
369 int will_compress;
370 int compress_type = root->fs_info->compress_type;
371 int redirty = 0;
372
373 /* if this is a small write inside eof, kick off a defrag */
374 if ((end - start + 1) < 16 * 1024 &&
375 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
376 btrfs_add_inode_defrag(NULL, inode);
377
378 actual_end = min_t(u64, isize, end + 1);
379 again:
380 will_compress = 0;
381 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
382 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
383
384 /*
385 * we don't want to send crud past the end of i_size through
386 * compression, that's just a waste of CPU time. So, if the
387 * end of the file is before the start of our current
388 * requested range of bytes, we bail out to the uncompressed
389 * cleanup code that can deal with all of this.
390 *
391 * It isn't really the fastest way to fix things, but this is a
392 * very uncommon corner.
393 */
394 if (actual_end <= start)
395 goto cleanup_and_bail_uncompressed;
396
397 total_compressed = actual_end - start;
398
399 /* we want to make sure that amount of ram required to uncompress
400 * an extent is reasonable, so we limit the total size in ram
401 * of a compressed extent to 128k. This is a crucial number
402 * because it also controls how easily we can spread reads across
403 * cpus for decompression.
404 *
405 * We also want to make sure the amount of IO required to do
406 * a random read is reasonably small, so we limit the size of
407 * a compressed extent to 128k.
408 */
409 total_compressed = min(total_compressed, max_uncompressed);
410 num_bytes = ALIGN(end - start + 1, blocksize);
411 num_bytes = max(blocksize, num_bytes);
412 total_in = 0;
413 ret = 0;
414
415 /*
416 * we do compression for mount -o compress and when the
417 * inode has not been flagged as nocompress. This flag can
418 * change at any time if we discover bad compression ratios.
419 */
420 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
421 (btrfs_test_opt(root, COMPRESS) ||
422 (BTRFS_I(inode)->force_compress) ||
423 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
424 WARN_ON(pages);
425 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
426 if (!pages) {
427 /* just bail out to the uncompressed code */
428 goto cont;
429 }
430
431 if (BTRFS_I(inode)->force_compress)
432 compress_type = BTRFS_I(inode)->force_compress;
433
434 /*
435 * we need to call clear_page_dirty_for_io on each
436 * page in the range. Otherwise applications with the file
437 * mmap'd can wander in and change the page contents while
438 * we are compressing them.
439 *
440 * If the compression fails for any reason, we set the pages
441 * dirty again later on.
442 */
443 extent_range_clear_dirty_for_io(inode, start, end);
444 redirty = 1;
445 ret = btrfs_compress_pages(compress_type,
446 inode->i_mapping, start,
447 total_compressed, pages,
448 nr_pages, &nr_pages_ret,
449 &total_in,
450 &total_compressed,
451 max_compressed);
452
453 if (!ret) {
454 unsigned long offset = total_compressed &
455 (PAGE_CACHE_SIZE - 1);
456 struct page *page = pages[nr_pages_ret - 1];
457 char *kaddr;
458
459 /* zero the tail end of the last page, we might be
460 * sending it down to disk
461 */
462 if (offset) {
463 kaddr = kmap_atomic(page);
464 memset(kaddr + offset, 0,
465 PAGE_CACHE_SIZE - offset);
466 kunmap_atomic(kaddr);
467 }
468 will_compress = 1;
469 }
470 }
471 cont:
472 if (start == 0) {
473 /* lets try to make an inline extent */
474 if (ret || total_in < (actual_end - start)) {
475 /* we didn't compress the entire range, try
476 * to make an uncompressed inline extent.
477 */
478 ret = cow_file_range_inline(root, inode, start, end,
479 0, 0, NULL);
480 } else {
481 /* try making a compressed inline extent */
482 ret = cow_file_range_inline(root, inode, start, end,
483 total_compressed,
484 compress_type, pages);
485 }
486 if (ret <= 0) {
487 unsigned long clear_flags = EXTENT_DELALLOC |
488 EXTENT_DEFRAG;
489 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
490
491 /*
492 * inline extent creation worked or returned error,
493 * we don't need to create any more async work items.
494 * Unlock and free up our temp pages.
495 */
496 extent_clear_unlock_delalloc(inode, start, end, NULL,
497 clear_flags, PAGE_UNLOCK |
498 PAGE_CLEAR_DIRTY |
499 PAGE_SET_WRITEBACK |
500 PAGE_END_WRITEBACK);
501 goto free_pages_out;
502 }
503 }
504
505 if (will_compress) {
506 /*
507 * we aren't doing an inline extent round the compressed size
508 * up to a block size boundary so the allocator does sane
509 * things
510 */
511 total_compressed = ALIGN(total_compressed, blocksize);
512
513 /*
514 * one last check to make sure the compression is really a
515 * win, compare the page count read with the blocks on disk
516 */
517 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
518 if (total_compressed >= total_in) {
519 will_compress = 0;
520 } else {
521 num_bytes = total_in;
522 }
523 }
524 if (!will_compress && pages) {
525 /*
526 * the compression code ran but failed to make things smaller,
527 * free any pages it allocated and our page pointer array
528 */
529 for (i = 0; i < nr_pages_ret; i++) {
530 WARN_ON(pages[i]->mapping);
531 page_cache_release(pages[i]);
532 }
533 kfree(pages);
534 pages = NULL;
535 total_compressed = 0;
536 nr_pages_ret = 0;
537
538 /* flag the file so we don't compress in the future */
539 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
540 !(BTRFS_I(inode)->force_compress)) {
541 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
542 }
543 }
544 if (will_compress) {
545 *num_added += 1;
546
547 /* the async work queues will take care of doing actual
548 * allocation on disk for these compressed pages,
549 * and will submit them to the elevator.
550 */
551 add_async_extent(async_cow, start, num_bytes,
552 total_compressed, pages, nr_pages_ret,
553 compress_type);
554
555 if (start + num_bytes < end) {
556 start += num_bytes;
557 pages = NULL;
558 cond_resched();
559 goto again;
560 }
561 } else {
562 cleanup_and_bail_uncompressed:
563 /*
564 * No compression, but we still need to write the pages in
565 * the file we've been given so far. redirty the locked
566 * page if it corresponds to our extent and set things up
567 * for the async work queue to run cow_file_range to do
568 * the normal delalloc dance
569 */
570 if (page_offset(locked_page) >= start &&
571 page_offset(locked_page) <= end) {
572 __set_page_dirty_nobuffers(locked_page);
573 /* unlocked later on in the async handlers */
574 }
575 if (redirty)
576 extent_range_redirty_for_io(inode, start, end);
577 add_async_extent(async_cow, start, end - start + 1,
578 0, NULL, 0, BTRFS_COMPRESS_NONE);
579 *num_added += 1;
580 }
581
582 out:
583 return ret;
584
585 free_pages_out:
586 for (i = 0; i < nr_pages_ret; i++) {
587 WARN_ON(pages[i]->mapping);
588 page_cache_release(pages[i]);
589 }
590 kfree(pages);
591
592 goto out;
593 }
594
595 /*
596 * phase two of compressed writeback. This is the ordered portion
597 * of the code, which only gets called in the order the work was
598 * queued. We walk all the async extents created by compress_file_range
599 * and send them down to the disk.
600 */
601 static noinline int submit_compressed_extents(struct inode *inode,
602 struct async_cow *async_cow)
603 {
604 struct async_extent *async_extent;
605 u64 alloc_hint = 0;
606 struct btrfs_key ins;
607 struct extent_map *em;
608 struct btrfs_root *root = BTRFS_I(inode)->root;
609 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
610 struct extent_io_tree *io_tree;
611 int ret = 0;
612
613 if (list_empty(&async_cow->extents))
614 return 0;
615
616 again:
617 while (!list_empty(&async_cow->extents)) {
618 async_extent = list_entry(async_cow->extents.next,
619 struct async_extent, list);
620 list_del(&async_extent->list);
621
622 io_tree = &BTRFS_I(inode)->io_tree;
623
624 retry:
625 /* did the compression code fall back to uncompressed IO? */
626 if (!async_extent->pages) {
627 int page_started = 0;
628 unsigned long nr_written = 0;
629
630 lock_extent(io_tree, async_extent->start,
631 async_extent->start +
632 async_extent->ram_size - 1);
633
634 /* allocate blocks */
635 ret = cow_file_range(inode, async_cow->locked_page,
636 async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1,
639 &page_started, &nr_written, 0);
640
641 /* JDM XXX */
642
643 /*
644 * if page_started, cow_file_range inserted an
645 * inline extent and took care of all the unlocking
646 * and IO for us. Otherwise, we need to submit
647 * all those pages down to the drive.
648 */
649 if (!page_started && !ret)
650 extent_write_locked_range(io_tree,
651 inode, async_extent->start,
652 async_extent->start +
653 async_extent->ram_size - 1,
654 btrfs_get_extent,
655 WB_SYNC_ALL);
656 else if (ret)
657 unlock_page(async_cow->locked_page);
658 kfree(async_extent);
659 cond_resched();
660 continue;
661 }
662
663 lock_extent(io_tree, async_extent->start,
664 async_extent->start + async_extent->ram_size - 1);
665
666 ret = btrfs_reserve_extent(root,
667 async_extent->compressed_size,
668 async_extent->compressed_size,
669 0, alloc_hint, &ins, 1);
670 if (ret) {
671 int i;
672
673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
676 }
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
680
681 if (ret == -ENOSPC) {
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
684 async_extent->ram_size - 1);
685 goto retry;
686 }
687 goto out_free;
688 }
689
690 /*
691 * here we're doing allocation and writeback of the
692 * compressed pages
693 */
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
697
698 em = alloc_extent_map();
699 if (!em) {
700 ret = -ENOMEM;
701 goto out_free_reserve;
702 }
703 em->start = async_extent->start;
704 em->len = async_extent->ram_size;
705 em->orig_start = em->start;
706 em->mod_start = em->start;
707 em->mod_len = em->len;
708
709 em->block_start = ins.objectid;
710 em->block_len = ins.offset;
711 em->orig_block_len = ins.offset;
712 em->ram_bytes = async_extent->ram_size;
713 em->bdev = root->fs_info->fs_devices->latest_bdev;
714 em->compress_type = async_extent->compress_type;
715 set_bit(EXTENT_FLAG_PINNED, &em->flags);
716 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
717 em->generation = -1;
718
719 while (1) {
720 write_lock(&em_tree->lock);
721 ret = add_extent_mapping(em_tree, em, 1);
722 write_unlock(&em_tree->lock);
723 if (ret != -EEXIST) {
724 free_extent_map(em);
725 break;
726 }
727 btrfs_drop_extent_cache(inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1, 0);
730 }
731
732 if (ret)
733 goto out_free_reserve;
734
735 ret = btrfs_add_ordered_extent_compress(inode,
736 async_extent->start,
737 ins.objectid,
738 async_extent->ram_size,
739 ins.offset,
740 BTRFS_ORDERED_COMPRESSED,
741 async_extent->compress_type);
742 if (ret)
743 goto out_free_reserve;
744
745 /*
746 * clear dirty, set writeback and unlock the pages.
747 */
748 extent_clear_unlock_delalloc(inode, async_extent->start,
749 async_extent->start +
750 async_extent->ram_size - 1,
751 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
752 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
753 PAGE_SET_WRITEBACK);
754 ret = btrfs_submit_compressed_write(inode,
755 async_extent->start,
756 async_extent->ram_size,
757 ins.objectid,
758 ins.offset, async_extent->pages,
759 async_extent->nr_pages);
760 alloc_hint = ins.objectid + ins.offset;
761 kfree(async_extent);
762 if (ret)
763 goto out;
764 cond_resched();
765 }
766 ret = 0;
767 out:
768 return ret;
769 out_free_reserve:
770 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
771 out_free:
772 extent_clear_unlock_delalloc(inode, async_extent->start,
773 async_extent->start +
774 async_extent->ram_size - 1,
775 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
776 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
777 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
778 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
779 kfree(async_extent);
780 goto again;
781 }
782
783 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
784 u64 num_bytes)
785 {
786 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
787 struct extent_map *em;
788 u64 alloc_hint = 0;
789
790 read_lock(&em_tree->lock);
791 em = search_extent_mapping(em_tree, start, num_bytes);
792 if (em) {
793 /*
794 * if block start isn't an actual block number then find the
795 * first block in this inode and use that as a hint. If that
796 * block is also bogus then just don't worry about it.
797 */
798 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
799 free_extent_map(em);
800 em = search_extent_mapping(em_tree, 0, 0);
801 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
802 alloc_hint = em->block_start;
803 if (em)
804 free_extent_map(em);
805 } else {
806 alloc_hint = em->block_start;
807 free_extent_map(em);
808 }
809 }
810 read_unlock(&em_tree->lock);
811
812 return alloc_hint;
813 }
814
815 /*
816 * when extent_io.c finds a delayed allocation range in the file,
817 * the call backs end up in this code. The basic idea is to
818 * allocate extents on disk for the range, and create ordered data structs
819 * in ram to track those extents.
820 *
821 * locked_page is the page that writepage had locked already. We use
822 * it to make sure we don't do extra locks or unlocks.
823 *
824 * *page_started is set to one if we unlock locked_page and do everything
825 * required to start IO on it. It may be clean and already done with
826 * IO when we return.
827 */
828 static noinline int cow_file_range(struct inode *inode,
829 struct page *locked_page,
830 u64 start, u64 end, int *page_started,
831 unsigned long *nr_written,
832 int unlock)
833 {
834 struct btrfs_root *root = BTRFS_I(inode)->root;
835 u64 alloc_hint = 0;
836 u64 num_bytes;
837 unsigned long ram_size;
838 u64 disk_num_bytes;
839 u64 cur_alloc_size;
840 u64 blocksize = root->sectorsize;
841 struct btrfs_key ins;
842 struct extent_map *em;
843 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
844 int ret = 0;
845
846 if (btrfs_is_free_space_inode(inode)) {
847 WARN_ON_ONCE(1);
848 return -EINVAL;
849 }
850
851 num_bytes = ALIGN(end - start + 1, blocksize);
852 num_bytes = max(blocksize, num_bytes);
853 disk_num_bytes = num_bytes;
854
855 /* if this is a small write inside eof, kick off defrag */
856 if (num_bytes < 64 * 1024 &&
857 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
858 btrfs_add_inode_defrag(NULL, inode);
859
860 if (start == 0) {
861 /* lets try to make an inline extent */
862 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
863 NULL);
864 if (ret == 0) {
865 extent_clear_unlock_delalloc(inode, start, end, NULL,
866 EXTENT_LOCKED | EXTENT_DELALLOC |
867 EXTENT_DEFRAG, PAGE_UNLOCK |
868 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
869 PAGE_END_WRITEBACK);
870
871 *nr_written = *nr_written +
872 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
873 *page_started = 1;
874 goto out;
875 } else if (ret < 0) {
876 goto out_unlock;
877 }
878 }
879
880 BUG_ON(disk_num_bytes >
881 btrfs_super_total_bytes(root->fs_info->super_copy));
882
883 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
884 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
885
886 while (disk_num_bytes > 0) {
887 unsigned long op;
888
889 cur_alloc_size = disk_num_bytes;
890 ret = btrfs_reserve_extent(root, cur_alloc_size,
891 root->sectorsize, 0, alloc_hint,
892 &ins, 1);
893 if (ret < 0)
894 goto out_unlock;
895
896 em = alloc_extent_map();
897 if (!em) {
898 ret = -ENOMEM;
899 goto out_reserve;
900 }
901 em->start = start;
902 em->orig_start = em->start;
903 ram_size = ins.offset;
904 em->len = ins.offset;
905 em->mod_start = em->start;
906 em->mod_len = em->len;
907
908 em->block_start = ins.objectid;
909 em->block_len = ins.offset;
910 em->orig_block_len = ins.offset;
911 em->ram_bytes = ram_size;
912 em->bdev = root->fs_info->fs_devices->latest_bdev;
913 set_bit(EXTENT_FLAG_PINNED, &em->flags);
914 em->generation = -1;
915
916 while (1) {
917 write_lock(&em_tree->lock);
918 ret = add_extent_mapping(em_tree, em, 1);
919 write_unlock(&em_tree->lock);
920 if (ret != -EEXIST) {
921 free_extent_map(em);
922 break;
923 }
924 btrfs_drop_extent_cache(inode, start,
925 start + ram_size - 1, 0);
926 }
927 if (ret)
928 goto out_reserve;
929
930 cur_alloc_size = ins.offset;
931 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
932 ram_size, cur_alloc_size, 0);
933 if (ret)
934 goto out_reserve;
935
936 if (root->root_key.objectid ==
937 BTRFS_DATA_RELOC_TREE_OBJECTID) {
938 ret = btrfs_reloc_clone_csums(inode, start,
939 cur_alloc_size);
940 if (ret)
941 goto out_reserve;
942 }
943
944 if (disk_num_bytes < cur_alloc_size)
945 break;
946
947 /* we're not doing compressed IO, don't unlock the first
948 * page (which the caller expects to stay locked), don't
949 * clear any dirty bits and don't set any writeback bits
950 *
951 * Do set the Private2 bit so we know this page was properly
952 * setup for writepage
953 */
954 op = unlock ? PAGE_UNLOCK : 0;
955 op |= PAGE_SET_PRIVATE2;
956
957 extent_clear_unlock_delalloc(inode, start,
958 start + ram_size - 1, locked_page,
959 EXTENT_LOCKED | EXTENT_DELALLOC,
960 op);
961 disk_num_bytes -= cur_alloc_size;
962 num_bytes -= cur_alloc_size;
963 alloc_hint = ins.objectid + ins.offset;
964 start += cur_alloc_size;
965 }
966 out:
967 return ret;
968
969 out_reserve:
970 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
971 out_unlock:
972 extent_clear_unlock_delalloc(inode, start, end, locked_page,
973 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
974 EXTENT_DELALLOC | EXTENT_DEFRAG,
975 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
976 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
977 goto out;
978 }
979
980 /*
981 * work queue call back to started compression on a file and pages
982 */
983 static noinline void async_cow_start(struct btrfs_work *work)
984 {
985 struct async_cow *async_cow;
986 int num_added = 0;
987 async_cow = container_of(work, struct async_cow, work);
988
989 compress_file_range(async_cow->inode, async_cow->locked_page,
990 async_cow->start, async_cow->end, async_cow,
991 &num_added);
992 if (num_added == 0) {
993 btrfs_add_delayed_iput(async_cow->inode);
994 async_cow->inode = NULL;
995 }
996 }
997
998 /*
999 * work queue call back to submit previously compressed pages
1000 */
1001 static noinline void async_cow_submit(struct btrfs_work *work)
1002 {
1003 struct async_cow *async_cow;
1004 struct btrfs_root *root;
1005 unsigned long nr_pages;
1006
1007 async_cow = container_of(work, struct async_cow, work);
1008
1009 root = async_cow->root;
1010 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1011 PAGE_CACHE_SHIFT;
1012
1013 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1014 5 * 1024 * 1024 &&
1015 waitqueue_active(&root->fs_info->async_submit_wait))
1016 wake_up(&root->fs_info->async_submit_wait);
1017
1018 if (async_cow->inode)
1019 submit_compressed_extents(async_cow->inode, async_cow);
1020 }
1021
1022 static noinline void async_cow_free(struct btrfs_work *work)
1023 {
1024 struct async_cow *async_cow;
1025 async_cow = container_of(work, struct async_cow, work);
1026 if (async_cow->inode)
1027 btrfs_add_delayed_iput(async_cow->inode);
1028 kfree(async_cow);
1029 }
1030
1031 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032 u64 start, u64 end, int *page_started,
1033 unsigned long *nr_written)
1034 {
1035 struct async_cow *async_cow;
1036 struct btrfs_root *root = BTRFS_I(inode)->root;
1037 unsigned long nr_pages;
1038 u64 cur_end;
1039 int limit = 10 * 1024 * 1024;
1040
1041 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042 1, 0, NULL, GFP_NOFS);
1043 while (start < end) {
1044 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1045 BUG_ON(!async_cow); /* -ENOMEM */
1046 async_cow->inode = igrab(inode);
1047 async_cow->root = root;
1048 async_cow->locked_page = locked_page;
1049 async_cow->start = start;
1050
1051 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1052 cur_end = end;
1053 else
1054 cur_end = min(end, start + 512 * 1024 - 1);
1055
1056 async_cow->end = cur_end;
1057 INIT_LIST_HEAD(&async_cow->extents);
1058
1059 async_cow->work.func = async_cow_start;
1060 async_cow->work.ordered_func = async_cow_submit;
1061 async_cow->work.ordered_free = async_cow_free;
1062 async_cow->work.flags = 0;
1063
1064 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1065 PAGE_CACHE_SHIFT;
1066 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1067
1068 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1069 &async_cow->work);
1070
1071 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072 wait_event(root->fs_info->async_submit_wait,
1073 (atomic_read(&root->fs_info->async_delalloc_pages) <
1074 limit));
1075 }
1076
1077 while (atomic_read(&root->fs_info->async_submit_draining) &&
1078 atomic_read(&root->fs_info->async_delalloc_pages)) {
1079 wait_event(root->fs_info->async_submit_wait,
1080 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1081 0));
1082 }
1083
1084 *nr_written += nr_pages;
1085 start = cur_end + 1;
1086 }
1087 *page_started = 1;
1088 return 0;
1089 }
1090
1091 static noinline int csum_exist_in_range(struct btrfs_root *root,
1092 u64 bytenr, u64 num_bytes)
1093 {
1094 int ret;
1095 struct btrfs_ordered_sum *sums;
1096 LIST_HEAD(list);
1097
1098 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1099 bytenr + num_bytes - 1, &list, 0);
1100 if (ret == 0 && list_empty(&list))
1101 return 0;
1102
1103 while (!list_empty(&list)) {
1104 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105 list_del(&sums->list);
1106 kfree(sums);
1107 }
1108 return 1;
1109 }
1110
1111 /*
1112 * when nowcow writeback call back. This checks for snapshots or COW copies
1113 * of the extents that exist in the file, and COWs the file as required.
1114 *
1115 * If no cow copies or snapshots exist, we write directly to the existing
1116 * blocks on disk
1117 */
1118 static noinline int run_delalloc_nocow(struct inode *inode,
1119 struct page *locked_page,
1120 u64 start, u64 end, int *page_started, int force,
1121 unsigned long *nr_written)
1122 {
1123 struct btrfs_root *root = BTRFS_I(inode)->root;
1124 struct btrfs_trans_handle *trans;
1125 struct extent_buffer *leaf;
1126 struct btrfs_path *path;
1127 struct btrfs_file_extent_item *fi;
1128 struct btrfs_key found_key;
1129 u64 cow_start;
1130 u64 cur_offset;
1131 u64 extent_end;
1132 u64 extent_offset;
1133 u64 disk_bytenr;
1134 u64 num_bytes;
1135 u64 disk_num_bytes;
1136 u64 ram_bytes;
1137 int extent_type;
1138 int ret, err;
1139 int type;
1140 int nocow;
1141 int check_prev = 1;
1142 bool nolock;
1143 u64 ino = btrfs_ino(inode);
1144
1145 path = btrfs_alloc_path();
1146 if (!path) {
1147 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1148 EXTENT_LOCKED | EXTENT_DELALLOC |
1149 EXTENT_DO_ACCOUNTING |
1150 EXTENT_DEFRAG, PAGE_UNLOCK |
1151 PAGE_CLEAR_DIRTY |
1152 PAGE_SET_WRITEBACK |
1153 PAGE_END_WRITEBACK);
1154 return -ENOMEM;
1155 }
1156
1157 nolock = btrfs_is_free_space_inode(inode);
1158
1159 if (nolock)
1160 trans = btrfs_join_transaction_nolock(root);
1161 else
1162 trans = btrfs_join_transaction(root);
1163
1164 if (IS_ERR(trans)) {
1165 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166 EXTENT_LOCKED | EXTENT_DELALLOC |
1167 EXTENT_DO_ACCOUNTING |
1168 EXTENT_DEFRAG, PAGE_UNLOCK |
1169 PAGE_CLEAR_DIRTY |
1170 PAGE_SET_WRITEBACK |
1171 PAGE_END_WRITEBACK);
1172 btrfs_free_path(path);
1173 return PTR_ERR(trans);
1174 }
1175
1176 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1177
1178 cow_start = (u64)-1;
1179 cur_offset = start;
1180 while (1) {
1181 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1182 cur_offset, 0);
1183 if (ret < 0)
1184 goto error;
1185 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186 leaf = path->nodes[0];
1187 btrfs_item_key_to_cpu(leaf, &found_key,
1188 path->slots[0] - 1);
1189 if (found_key.objectid == ino &&
1190 found_key.type == BTRFS_EXTENT_DATA_KEY)
1191 path->slots[0]--;
1192 }
1193 check_prev = 0;
1194 next_slot:
1195 leaf = path->nodes[0];
1196 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197 ret = btrfs_next_leaf(root, path);
1198 if (ret < 0)
1199 goto error;
1200 if (ret > 0)
1201 break;
1202 leaf = path->nodes[0];
1203 }
1204
1205 nocow = 0;
1206 disk_bytenr = 0;
1207 num_bytes = 0;
1208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1209
1210 if (found_key.objectid > ino ||
1211 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1212 found_key.offset > end)
1213 break;
1214
1215 if (found_key.offset > cur_offset) {
1216 extent_end = found_key.offset;
1217 extent_type = 0;
1218 goto out_check;
1219 }
1220
1221 fi = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_file_extent_item);
1223 extent_type = btrfs_file_extent_type(leaf, fi);
1224
1225 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1226 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1227 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1228 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1229 extent_offset = btrfs_file_extent_offset(leaf, fi);
1230 extent_end = found_key.offset +
1231 btrfs_file_extent_num_bytes(leaf, fi);
1232 disk_num_bytes =
1233 btrfs_file_extent_disk_num_bytes(leaf, fi);
1234 if (extent_end <= start) {
1235 path->slots[0]++;
1236 goto next_slot;
1237 }
1238 if (disk_bytenr == 0)
1239 goto out_check;
1240 if (btrfs_file_extent_compression(leaf, fi) ||
1241 btrfs_file_extent_encryption(leaf, fi) ||
1242 btrfs_file_extent_other_encoding(leaf, fi))
1243 goto out_check;
1244 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1245 goto out_check;
1246 if (btrfs_extent_readonly(root, disk_bytenr))
1247 goto out_check;
1248 if (btrfs_cross_ref_exist(trans, root, ino,
1249 found_key.offset -
1250 extent_offset, disk_bytenr))
1251 goto out_check;
1252 disk_bytenr += extent_offset;
1253 disk_bytenr += cur_offset - found_key.offset;
1254 num_bytes = min(end + 1, extent_end) - cur_offset;
1255 /*
1256 * force cow if csum exists in the range.
1257 * this ensure that csum for a given extent are
1258 * either valid or do not exist.
1259 */
1260 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1261 goto out_check;
1262 nocow = 1;
1263 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1264 extent_end = found_key.offset +
1265 btrfs_file_extent_inline_len(leaf, fi);
1266 extent_end = ALIGN(extent_end, root->sectorsize);
1267 } else {
1268 BUG_ON(1);
1269 }
1270 out_check:
1271 if (extent_end <= start) {
1272 path->slots[0]++;
1273 goto next_slot;
1274 }
1275 if (!nocow) {
1276 if (cow_start == (u64)-1)
1277 cow_start = cur_offset;
1278 cur_offset = extent_end;
1279 if (cur_offset > end)
1280 break;
1281 path->slots[0]++;
1282 goto next_slot;
1283 }
1284
1285 btrfs_release_path(path);
1286 if (cow_start != (u64)-1) {
1287 ret = cow_file_range(inode, locked_page,
1288 cow_start, found_key.offset - 1,
1289 page_started, nr_written, 1);
1290 if (ret)
1291 goto error;
1292 cow_start = (u64)-1;
1293 }
1294
1295 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1296 struct extent_map *em;
1297 struct extent_map_tree *em_tree;
1298 em_tree = &BTRFS_I(inode)->extent_tree;
1299 em = alloc_extent_map();
1300 BUG_ON(!em); /* -ENOMEM */
1301 em->start = cur_offset;
1302 em->orig_start = found_key.offset - extent_offset;
1303 em->len = num_bytes;
1304 em->block_len = num_bytes;
1305 em->block_start = disk_bytenr;
1306 em->orig_block_len = disk_num_bytes;
1307 em->ram_bytes = ram_bytes;
1308 em->bdev = root->fs_info->fs_devices->latest_bdev;
1309 em->mod_start = em->start;
1310 em->mod_len = em->len;
1311 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1313 em->generation = -1;
1314 while (1) {
1315 write_lock(&em_tree->lock);
1316 ret = add_extent_mapping(em_tree, em, 1);
1317 write_unlock(&em_tree->lock);
1318 if (ret != -EEXIST) {
1319 free_extent_map(em);
1320 break;
1321 }
1322 btrfs_drop_extent_cache(inode, em->start,
1323 em->start + em->len - 1, 0);
1324 }
1325 type = BTRFS_ORDERED_PREALLOC;
1326 } else {
1327 type = BTRFS_ORDERED_NOCOW;
1328 }
1329
1330 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1331 num_bytes, num_bytes, type);
1332 BUG_ON(ret); /* -ENOMEM */
1333
1334 if (root->root_key.objectid ==
1335 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1336 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1337 num_bytes);
1338 if (ret)
1339 goto error;
1340 }
1341
1342 extent_clear_unlock_delalloc(inode, cur_offset,
1343 cur_offset + num_bytes - 1,
1344 locked_page, EXTENT_LOCKED |
1345 EXTENT_DELALLOC, PAGE_UNLOCK |
1346 PAGE_SET_PRIVATE2);
1347 cur_offset = extent_end;
1348 if (cur_offset > end)
1349 break;
1350 }
1351 btrfs_release_path(path);
1352
1353 if (cur_offset <= end && cow_start == (u64)-1) {
1354 cow_start = cur_offset;
1355 cur_offset = end;
1356 }
1357
1358 if (cow_start != (u64)-1) {
1359 ret = cow_file_range(inode, locked_page, cow_start, end,
1360 page_started, nr_written, 1);
1361 if (ret)
1362 goto error;
1363 }
1364
1365 error:
1366 err = btrfs_end_transaction(trans, root);
1367 if (!ret)
1368 ret = err;
1369
1370 if (ret && cur_offset < end)
1371 extent_clear_unlock_delalloc(inode, cur_offset, end,
1372 locked_page, EXTENT_LOCKED |
1373 EXTENT_DELALLOC | EXTENT_DEFRAG |
1374 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1375 PAGE_CLEAR_DIRTY |
1376 PAGE_SET_WRITEBACK |
1377 PAGE_END_WRITEBACK);
1378 btrfs_free_path(path);
1379 return ret;
1380 }
1381
1382 /*
1383 * extent_io.c call back to do delayed allocation processing
1384 */
1385 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1386 u64 start, u64 end, int *page_started,
1387 unsigned long *nr_written)
1388 {
1389 int ret;
1390 struct btrfs_root *root = BTRFS_I(inode)->root;
1391
1392 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1393 ret = run_delalloc_nocow(inode, locked_page, start, end,
1394 page_started, 1, nr_written);
1395 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1396 ret = run_delalloc_nocow(inode, locked_page, start, end,
1397 page_started, 0, nr_written);
1398 } else if (!btrfs_test_opt(root, COMPRESS) &&
1399 !(BTRFS_I(inode)->force_compress) &&
1400 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1401 ret = cow_file_range(inode, locked_page, start, end,
1402 page_started, nr_written, 1);
1403 } else {
1404 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1405 &BTRFS_I(inode)->runtime_flags);
1406 ret = cow_file_range_async(inode, locked_page, start, end,
1407 page_started, nr_written);
1408 }
1409 return ret;
1410 }
1411
1412 static void btrfs_split_extent_hook(struct inode *inode,
1413 struct extent_state *orig, u64 split)
1414 {
1415 /* not delalloc, ignore it */
1416 if (!(orig->state & EXTENT_DELALLOC))
1417 return;
1418
1419 spin_lock(&BTRFS_I(inode)->lock);
1420 BTRFS_I(inode)->outstanding_extents++;
1421 spin_unlock(&BTRFS_I(inode)->lock);
1422 }
1423
1424 /*
1425 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1426 * extents so we can keep track of new extents that are just merged onto old
1427 * extents, such as when we are doing sequential writes, so we can properly
1428 * account for the metadata space we'll need.
1429 */
1430 static void btrfs_merge_extent_hook(struct inode *inode,
1431 struct extent_state *new,
1432 struct extent_state *other)
1433 {
1434 /* not delalloc, ignore it */
1435 if (!(other->state & EXTENT_DELALLOC))
1436 return;
1437
1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents--;
1440 spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1444 struct inode *inode)
1445 {
1446 spin_lock(&root->delalloc_lock);
1447 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1448 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1449 &root->delalloc_inodes);
1450 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1451 &BTRFS_I(inode)->runtime_flags);
1452 root->nr_delalloc_inodes++;
1453 if (root->nr_delalloc_inodes == 1) {
1454 spin_lock(&root->fs_info->delalloc_root_lock);
1455 BUG_ON(!list_empty(&root->delalloc_root));
1456 list_add_tail(&root->delalloc_root,
1457 &root->fs_info->delalloc_roots);
1458 spin_unlock(&root->fs_info->delalloc_root_lock);
1459 }
1460 }
1461 spin_unlock(&root->delalloc_lock);
1462 }
1463
1464 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1465 struct inode *inode)
1466 {
1467 spin_lock(&root->delalloc_lock);
1468 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1470 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1471 &BTRFS_I(inode)->runtime_flags);
1472 root->nr_delalloc_inodes--;
1473 if (!root->nr_delalloc_inodes) {
1474 spin_lock(&root->fs_info->delalloc_root_lock);
1475 BUG_ON(list_empty(&root->delalloc_root));
1476 list_del_init(&root->delalloc_root);
1477 spin_unlock(&root->fs_info->delalloc_root_lock);
1478 }
1479 }
1480 spin_unlock(&root->delalloc_lock);
1481 }
1482
1483 /*
1484 * extent_io.c set_bit_hook, used to track delayed allocation
1485 * bytes in this file, and to maintain the list of inodes that
1486 * have pending delalloc work to be done.
1487 */
1488 static void btrfs_set_bit_hook(struct inode *inode,
1489 struct extent_state *state, unsigned long *bits)
1490 {
1491
1492 /*
1493 * set_bit and clear bit hooks normally require _irqsave/restore
1494 * but in this case, we are only testing for the DELALLOC
1495 * bit, which is only set or cleared with irqs on
1496 */
1497 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1498 struct btrfs_root *root = BTRFS_I(inode)->root;
1499 u64 len = state->end + 1 - state->start;
1500 bool do_list = !btrfs_is_free_space_inode(inode);
1501
1502 if (*bits & EXTENT_FIRST_DELALLOC) {
1503 *bits &= ~EXTENT_FIRST_DELALLOC;
1504 } else {
1505 spin_lock(&BTRFS_I(inode)->lock);
1506 BTRFS_I(inode)->outstanding_extents++;
1507 spin_unlock(&BTRFS_I(inode)->lock);
1508 }
1509
1510 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1511 root->fs_info->delalloc_batch);
1512 spin_lock(&BTRFS_I(inode)->lock);
1513 BTRFS_I(inode)->delalloc_bytes += len;
1514 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1515 &BTRFS_I(inode)->runtime_flags))
1516 btrfs_add_delalloc_inodes(root, inode);
1517 spin_unlock(&BTRFS_I(inode)->lock);
1518 }
1519 }
1520
1521 /*
1522 * extent_io.c clear_bit_hook, see set_bit_hook for why
1523 */
1524 static void btrfs_clear_bit_hook(struct inode *inode,
1525 struct extent_state *state,
1526 unsigned long *bits)
1527 {
1528 /*
1529 * set_bit and clear bit hooks normally require _irqsave/restore
1530 * but in this case, we are only testing for the DELALLOC
1531 * bit, which is only set or cleared with irqs on
1532 */
1533 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1534 struct btrfs_root *root = BTRFS_I(inode)->root;
1535 u64 len = state->end + 1 - state->start;
1536 bool do_list = !btrfs_is_free_space_inode(inode);
1537
1538 if (*bits & EXTENT_FIRST_DELALLOC) {
1539 *bits &= ~EXTENT_FIRST_DELALLOC;
1540 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541 spin_lock(&BTRFS_I(inode)->lock);
1542 BTRFS_I(inode)->outstanding_extents--;
1543 spin_unlock(&BTRFS_I(inode)->lock);
1544 }
1545
1546 /*
1547 * We don't reserve metadata space for space cache inodes so we
1548 * don't need to call dellalloc_release_metadata if there is an
1549 * error.
1550 */
1551 if (*bits & EXTENT_DO_ACCOUNTING &&
1552 root != root->fs_info->tree_root)
1553 btrfs_delalloc_release_metadata(inode, len);
1554
1555 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1556 && do_list && !(state->state & EXTENT_NORESERVE))
1557 btrfs_free_reserved_data_space(inode, len);
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 && BTRFS_I(inode)->delalloc_bytes == 0 &&
1564 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1565 &BTRFS_I(inode)->runtime_flags))
1566 btrfs_del_delalloc_inode(root, inode);
1567 spin_unlock(&BTRFS_I(inode)->lock);
1568 }
1569 }
1570
1571 /*
1572 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1573 * we don't create bios that span stripes or chunks
1574 */
1575 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1576 size_t size, struct bio *bio,
1577 unsigned long bio_flags)
1578 {
1579 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1580 u64 logical = (u64)bio->bi_sector << 9;
1581 u64 length = 0;
1582 u64 map_length;
1583 int ret;
1584
1585 if (bio_flags & EXTENT_BIO_COMPRESSED)
1586 return 0;
1587
1588 length = bio->bi_size;
1589 map_length = length;
1590 ret = btrfs_map_block(root->fs_info, rw, logical,
1591 &map_length, NULL, 0);
1592 /* Will always return 0 with map_multi == NULL */
1593 BUG_ON(ret < 0);
1594 if (map_length < length + size)
1595 return 1;
1596 return 0;
1597 }
1598
1599 /*
1600 * in order to insert checksums into the metadata in large chunks,
1601 * we wait until bio submission time. All the pages in the bio are
1602 * checksummed and sums are attached onto the ordered extent record.
1603 *
1604 * At IO completion time the cums attached on the ordered extent record
1605 * are inserted into the btree
1606 */
1607 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1608 struct bio *bio, int mirror_num,
1609 unsigned long bio_flags,
1610 u64 bio_offset)
1611 {
1612 struct btrfs_root *root = BTRFS_I(inode)->root;
1613 int ret = 0;
1614
1615 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1616 BUG_ON(ret); /* -ENOMEM */
1617 return 0;
1618 }
1619
1620 /*
1621 * in order to insert checksums into the metadata in large chunks,
1622 * we wait until bio submission time. All the pages in the bio are
1623 * checksummed and sums are attached onto the ordered extent record.
1624 *
1625 * At IO completion time the cums attached on the ordered extent record
1626 * are inserted into the btree
1627 */
1628 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1629 int mirror_num, unsigned long bio_flags,
1630 u64 bio_offset)
1631 {
1632 struct btrfs_root *root = BTRFS_I(inode)->root;
1633 int ret;
1634
1635 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1636 if (ret)
1637 bio_endio(bio, ret);
1638 return ret;
1639 }
1640
1641 /*
1642 * extent_io.c submission hook. This does the right thing for csum calculation
1643 * on write, or reading the csums from the tree before a read
1644 */
1645 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1646 int mirror_num, unsigned long bio_flags,
1647 u64 bio_offset)
1648 {
1649 struct btrfs_root *root = BTRFS_I(inode)->root;
1650 int ret = 0;
1651 int skip_sum;
1652 int metadata = 0;
1653 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1654
1655 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1656
1657 if (btrfs_is_free_space_inode(inode))
1658 metadata = 2;
1659
1660 if (!(rw & REQ_WRITE)) {
1661 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1662 if (ret)
1663 goto out;
1664
1665 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1666 ret = btrfs_submit_compressed_read(inode, bio,
1667 mirror_num,
1668 bio_flags);
1669 goto out;
1670 } else if (!skip_sum) {
1671 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1672 if (ret)
1673 goto out;
1674 }
1675 goto mapit;
1676 } else if (async && !skip_sum) {
1677 /* csum items have already been cloned */
1678 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1679 goto mapit;
1680 /* we're doing a write, do the async checksumming */
1681 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1682 inode, rw, bio, mirror_num,
1683 bio_flags, bio_offset,
1684 __btrfs_submit_bio_start,
1685 __btrfs_submit_bio_done);
1686 goto out;
1687 } else if (!skip_sum) {
1688 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1689 if (ret)
1690 goto out;
1691 }
1692
1693 mapit:
1694 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1695
1696 out:
1697 if (ret < 0)
1698 bio_endio(bio, ret);
1699 return ret;
1700 }
1701
1702 /*
1703 * given a list of ordered sums record them in the inode. This happens
1704 * at IO completion time based on sums calculated at bio submission time.
1705 */
1706 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1707 struct inode *inode, u64 file_offset,
1708 struct list_head *list)
1709 {
1710 struct btrfs_ordered_sum *sum;
1711
1712 list_for_each_entry(sum, list, list) {
1713 trans->adding_csums = 1;
1714 btrfs_csum_file_blocks(trans,
1715 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1716 trans->adding_csums = 0;
1717 }
1718 return 0;
1719 }
1720
1721 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1722 struct extent_state **cached_state)
1723 {
1724 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1725 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1726 cached_state, GFP_NOFS);
1727 }
1728
1729 /* see btrfs_writepage_start_hook for details on why this is required */
1730 struct btrfs_writepage_fixup {
1731 struct page *page;
1732 struct btrfs_work work;
1733 };
1734
1735 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1736 {
1737 struct btrfs_writepage_fixup *fixup;
1738 struct btrfs_ordered_extent *ordered;
1739 struct extent_state *cached_state = NULL;
1740 struct page *page;
1741 struct inode *inode;
1742 u64 page_start;
1743 u64 page_end;
1744 int ret;
1745
1746 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1747 page = fixup->page;
1748 again:
1749 lock_page(page);
1750 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1751 ClearPageChecked(page);
1752 goto out_page;
1753 }
1754
1755 inode = page->mapping->host;
1756 page_start = page_offset(page);
1757 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1758
1759 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1760 &cached_state);
1761
1762 /* already ordered? We're done */
1763 if (PagePrivate2(page))
1764 goto out;
1765
1766 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1767 if (ordered) {
1768 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1769 page_end, &cached_state, GFP_NOFS);
1770 unlock_page(page);
1771 btrfs_start_ordered_extent(inode, ordered, 1);
1772 btrfs_put_ordered_extent(ordered);
1773 goto again;
1774 }
1775
1776 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1777 if (ret) {
1778 mapping_set_error(page->mapping, ret);
1779 end_extent_writepage(page, ret, page_start, page_end);
1780 ClearPageChecked(page);
1781 goto out;
1782 }
1783
1784 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1785 ClearPageChecked(page);
1786 set_page_dirty(page);
1787 out:
1788 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1789 &cached_state, GFP_NOFS);
1790 out_page:
1791 unlock_page(page);
1792 page_cache_release(page);
1793 kfree(fixup);
1794 }
1795
1796 /*
1797 * There are a few paths in the higher layers of the kernel that directly
1798 * set the page dirty bit without asking the filesystem if it is a
1799 * good idea. This causes problems because we want to make sure COW
1800 * properly happens and the data=ordered rules are followed.
1801 *
1802 * In our case any range that doesn't have the ORDERED bit set
1803 * hasn't been properly setup for IO. We kick off an async process
1804 * to fix it up. The async helper will wait for ordered extents, set
1805 * the delalloc bit and make it safe to write the page.
1806 */
1807 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1808 {
1809 struct inode *inode = page->mapping->host;
1810 struct btrfs_writepage_fixup *fixup;
1811 struct btrfs_root *root = BTRFS_I(inode)->root;
1812
1813 /* this page is properly in the ordered list */
1814 if (TestClearPagePrivate2(page))
1815 return 0;
1816
1817 if (PageChecked(page))
1818 return -EAGAIN;
1819
1820 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1821 if (!fixup)
1822 return -EAGAIN;
1823
1824 SetPageChecked(page);
1825 page_cache_get(page);
1826 fixup->work.func = btrfs_writepage_fixup_worker;
1827 fixup->page = page;
1828 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1829 return -EBUSY;
1830 }
1831
1832 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1833 struct inode *inode, u64 file_pos,
1834 u64 disk_bytenr, u64 disk_num_bytes,
1835 u64 num_bytes, u64 ram_bytes,
1836 u8 compression, u8 encryption,
1837 u16 other_encoding, int extent_type)
1838 {
1839 struct btrfs_root *root = BTRFS_I(inode)->root;
1840 struct btrfs_file_extent_item *fi;
1841 struct btrfs_path *path;
1842 struct extent_buffer *leaf;
1843 struct btrfs_key ins;
1844 int ret;
1845
1846 path = btrfs_alloc_path();
1847 if (!path)
1848 return -ENOMEM;
1849
1850 path->leave_spinning = 1;
1851
1852 /*
1853 * we may be replacing one extent in the tree with another.
1854 * The new extent is pinned in the extent map, and we don't want
1855 * to drop it from the cache until it is completely in the btree.
1856 *
1857 * So, tell btrfs_drop_extents to leave this extent in the cache.
1858 * the caller is expected to unpin it and allow it to be merged
1859 * with the others.
1860 */
1861 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1862 file_pos + num_bytes, 0);
1863 if (ret)
1864 goto out;
1865
1866 ins.objectid = btrfs_ino(inode);
1867 ins.offset = file_pos;
1868 ins.type = BTRFS_EXTENT_DATA_KEY;
1869 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1870 if (ret)
1871 goto out;
1872 leaf = path->nodes[0];
1873 fi = btrfs_item_ptr(leaf, path->slots[0],
1874 struct btrfs_file_extent_item);
1875 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1876 btrfs_set_file_extent_type(leaf, fi, extent_type);
1877 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1878 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1879 btrfs_set_file_extent_offset(leaf, fi, 0);
1880 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1881 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1882 btrfs_set_file_extent_compression(leaf, fi, compression);
1883 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1884 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1885
1886 btrfs_mark_buffer_dirty(leaf);
1887 btrfs_release_path(path);
1888
1889 inode_add_bytes(inode, num_bytes);
1890
1891 ins.objectid = disk_bytenr;
1892 ins.offset = disk_num_bytes;
1893 ins.type = BTRFS_EXTENT_ITEM_KEY;
1894 ret = btrfs_alloc_reserved_file_extent(trans, root,
1895 root->root_key.objectid,
1896 btrfs_ino(inode), file_pos, &ins);
1897 out:
1898 btrfs_free_path(path);
1899
1900 return ret;
1901 }
1902
1903 /* snapshot-aware defrag */
1904 struct sa_defrag_extent_backref {
1905 struct rb_node node;
1906 struct old_sa_defrag_extent *old;
1907 u64 root_id;
1908 u64 inum;
1909 u64 file_pos;
1910 u64 extent_offset;
1911 u64 num_bytes;
1912 u64 generation;
1913 };
1914
1915 struct old_sa_defrag_extent {
1916 struct list_head list;
1917 struct new_sa_defrag_extent *new;
1918
1919 u64 extent_offset;
1920 u64 bytenr;
1921 u64 offset;
1922 u64 len;
1923 int count;
1924 };
1925
1926 struct new_sa_defrag_extent {
1927 struct rb_root root;
1928 struct list_head head;
1929 struct btrfs_path *path;
1930 struct inode *inode;
1931 u64 file_pos;
1932 u64 len;
1933 u64 bytenr;
1934 u64 disk_len;
1935 u8 compress_type;
1936 };
1937
1938 static int backref_comp(struct sa_defrag_extent_backref *b1,
1939 struct sa_defrag_extent_backref *b2)
1940 {
1941 if (b1->root_id < b2->root_id)
1942 return -1;
1943 else if (b1->root_id > b2->root_id)
1944 return 1;
1945
1946 if (b1->inum < b2->inum)
1947 return -1;
1948 else if (b1->inum > b2->inum)
1949 return 1;
1950
1951 if (b1->file_pos < b2->file_pos)
1952 return -1;
1953 else if (b1->file_pos > b2->file_pos)
1954 return 1;
1955
1956 /*
1957 * [------------------------------] ===> (a range of space)
1958 * |<--->| |<---->| =============> (fs/file tree A)
1959 * |<---------------------------->| ===> (fs/file tree B)
1960 *
1961 * A range of space can refer to two file extents in one tree while
1962 * refer to only one file extent in another tree.
1963 *
1964 * So we may process a disk offset more than one time(two extents in A)
1965 * and locate at the same extent(one extent in B), then insert two same
1966 * backrefs(both refer to the extent in B).
1967 */
1968 return 0;
1969 }
1970
1971 static void backref_insert(struct rb_root *root,
1972 struct sa_defrag_extent_backref *backref)
1973 {
1974 struct rb_node **p = &root->rb_node;
1975 struct rb_node *parent = NULL;
1976 struct sa_defrag_extent_backref *entry;
1977 int ret;
1978
1979 while (*p) {
1980 parent = *p;
1981 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1982
1983 ret = backref_comp(backref, entry);
1984 if (ret < 0)
1985 p = &(*p)->rb_left;
1986 else
1987 p = &(*p)->rb_right;
1988 }
1989
1990 rb_link_node(&backref->node, parent, p);
1991 rb_insert_color(&backref->node, root);
1992 }
1993
1994 /*
1995 * Note the backref might has changed, and in this case we just return 0.
1996 */
1997 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
1998 void *ctx)
1999 {
2000 struct btrfs_file_extent_item *extent;
2001 struct btrfs_fs_info *fs_info;
2002 struct old_sa_defrag_extent *old = ctx;
2003 struct new_sa_defrag_extent *new = old->new;
2004 struct btrfs_path *path = new->path;
2005 struct btrfs_key key;
2006 struct btrfs_root *root;
2007 struct sa_defrag_extent_backref *backref;
2008 struct extent_buffer *leaf;
2009 struct inode *inode = new->inode;
2010 int slot;
2011 int ret;
2012 u64 extent_offset;
2013 u64 num_bytes;
2014
2015 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2016 inum == btrfs_ino(inode))
2017 return 0;
2018
2019 key.objectid = root_id;
2020 key.type = BTRFS_ROOT_ITEM_KEY;
2021 key.offset = (u64)-1;
2022
2023 fs_info = BTRFS_I(inode)->root->fs_info;
2024 root = btrfs_read_fs_root_no_name(fs_info, &key);
2025 if (IS_ERR(root)) {
2026 if (PTR_ERR(root) == -ENOENT)
2027 return 0;
2028 WARN_ON(1);
2029 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2030 inum, offset, root_id);
2031 return PTR_ERR(root);
2032 }
2033
2034 key.objectid = inum;
2035 key.type = BTRFS_EXTENT_DATA_KEY;
2036 if (offset > (u64)-1 << 32)
2037 key.offset = 0;
2038 else
2039 key.offset = offset;
2040
2041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2042 if (WARN_ON(ret < 0))
2043 return ret;
2044 ret = 0;
2045
2046 while (1) {
2047 cond_resched();
2048
2049 leaf = path->nodes[0];
2050 slot = path->slots[0];
2051
2052 if (slot >= btrfs_header_nritems(leaf)) {
2053 ret = btrfs_next_leaf(root, path);
2054 if (ret < 0) {
2055 goto out;
2056 } else if (ret > 0) {
2057 ret = 0;
2058 goto out;
2059 }
2060 continue;
2061 }
2062
2063 path->slots[0]++;
2064
2065 btrfs_item_key_to_cpu(leaf, &key, slot);
2066
2067 if (key.objectid > inum)
2068 goto out;
2069
2070 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2071 continue;
2072
2073 extent = btrfs_item_ptr(leaf, slot,
2074 struct btrfs_file_extent_item);
2075
2076 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2077 continue;
2078
2079 /*
2080 * 'offset' refers to the exact key.offset,
2081 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2082 * (key.offset - extent_offset).
2083 */
2084 if (key.offset != offset)
2085 continue;
2086
2087 extent_offset = btrfs_file_extent_offset(leaf, extent);
2088 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2089
2090 if (extent_offset >= old->extent_offset + old->offset +
2091 old->len || extent_offset + num_bytes <=
2092 old->extent_offset + old->offset)
2093 continue;
2094 break;
2095 }
2096
2097 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2098 if (!backref) {
2099 ret = -ENOENT;
2100 goto out;
2101 }
2102
2103 backref->root_id = root_id;
2104 backref->inum = inum;
2105 backref->file_pos = offset;
2106 backref->num_bytes = num_bytes;
2107 backref->extent_offset = extent_offset;
2108 backref->generation = btrfs_file_extent_generation(leaf, extent);
2109 backref->old = old;
2110 backref_insert(&new->root, backref);
2111 old->count++;
2112 out:
2113 btrfs_release_path(path);
2114 WARN_ON(ret);
2115 return ret;
2116 }
2117
2118 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2119 struct new_sa_defrag_extent *new)
2120 {
2121 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2122 struct old_sa_defrag_extent *old, *tmp;
2123 int ret;
2124
2125 new->path = path;
2126
2127 list_for_each_entry_safe(old, tmp, &new->head, list) {
2128 ret = iterate_inodes_from_logical(old->bytenr +
2129 old->extent_offset, fs_info,
2130 path, record_one_backref,
2131 old);
2132 if (ret < 0 && ret != -ENOENT)
2133 return false;
2134
2135 /* no backref to be processed for this extent */
2136 if (!old->count) {
2137 list_del(&old->list);
2138 kfree(old);
2139 }
2140 }
2141
2142 if (list_empty(&new->head))
2143 return false;
2144
2145 return true;
2146 }
2147
2148 static int relink_is_mergable(struct extent_buffer *leaf,
2149 struct btrfs_file_extent_item *fi,
2150 struct new_sa_defrag_extent *new)
2151 {
2152 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2153 return 0;
2154
2155 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2156 return 0;
2157
2158 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2159 return 0;
2160
2161 if (btrfs_file_extent_encryption(leaf, fi) ||
2162 btrfs_file_extent_other_encoding(leaf, fi))
2163 return 0;
2164
2165 return 1;
2166 }
2167
2168 /*
2169 * Note the backref might has changed, and in this case we just return 0.
2170 */
2171 static noinline int relink_extent_backref(struct btrfs_path *path,
2172 struct sa_defrag_extent_backref *prev,
2173 struct sa_defrag_extent_backref *backref)
2174 {
2175 struct btrfs_file_extent_item *extent;
2176 struct btrfs_file_extent_item *item;
2177 struct btrfs_ordered_extent *ordered;
2178 struct btrfs_trans_handle *trans;
2179 struct btrfs_fs_info *fs_info;
2180 struct btrfs_root *root;
2181 struct btrfs_key key;
2182 struct extent_buffer *leaf;
2183 struct old_sa_defrag_extent *old = backref->old;
2184 struct new_sa_defrag_extent *new = old->new;
2185 struct inode *src_inode = new->inode;
2186 struct inode *inode;
2187 struct extent_state *cached = NULL;
2188 int ret = 0;
2189 u64 start;
2190 u64 len;
2191 u64 lock_start;
2192 u64 lock_end;
2193 bool merge = false;
2194 int index;
2195
2196 if (prev && prev->root_id == backref->root_id &&
2197 prev->inum == backref->inum &&
2198 prev->file_pos + prev->num_bytes == backref->file_pos)
2199 merge = true;
2200
2201 /* step 1: get root */
2202 key.objectid = backref->root_id;
2203 key.type = BTRFS_ROOT_ITEM_KEY;
2204 key.offset = (u64)-1;
2205
2206 fs_info = BTRFS_I(src_inode)->root->fs_info;
2207 index = srcu_read_lock(&fs_info->subvol_srcu);
2208
2209 root = btrfs_read_fs_root_no_name(fs_info, &key);
2210 if (IS_ERR(root)) {
2211 srcu_read_unlock(&fs_info->subvol_srcu, index);
2212 if (PTR_ERR(root) == -ENOENT)
2213 return 0;
2214 return PTR_ERR(root);
2215 }
2216
2217 /* step 2: get inode */
2218 key.objectid = backref->inum;
2219 key.type = BTRFS_INODE_ITEM_KEY;
2220 key.offset = 0;
2221
2222 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2223 if (IS_ERR(inode)) {
2224 srcu_read_unlock(&fs_info->subvol_srcu, index);
2225 return 0;
2226 }
2227
2228 srcu_read_unlock(&fs_info->subvol_srcu, index);
2229
2230 /* step 3: relink backref */
2231 lock_start = backref->file_pos;
2232 lock_end = backref->file_pos + backref->num_bytes - 1;
2233 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2234 0, &cached);
2235
2236 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2237 if (ordered) {
2238 btrfs_put_ordered_extent(ordered);
2239 goto out_unlock;
2240 }
2241
2242 trans = btrfs_join_transaction(root);
2243 if (IS_ERR(trans)) {
2244 ret = PTR_ERR(trans);
2245 goto out_unlock;
2246 }
2247
2248 key.objectid = backref->inum;
2249 key.type = BTRFS_EXTENT_DATA_KEY;
2250 key.offset = backref->file_pos;
2251
2252 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2253 if (ret < 0) {
2254 goto out_free_path;
2255 } else if (ret > 0) {
2256 ret = 0;
2257 goto out_free_path;
2258 }
2259
2260 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2261 struct btrfs_file_extent_item);
2262
2263 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2264 backref->generation)
2265 goto out_free_path;
2266
2267 btrfs_release_path(path);
2268
2269 start = backref->file_pos;
2270 if (backref->extent_offset < old->extent_offset + old->offset)
2271 start += old->extent_offset + old->offset -
2272 backref->extent_offset;
2273
2274 len = min(backref->extent_offset + backref->num_bytes,
2275 old->extent_offset + old->offset + old->len);
2276 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2277
2278 ret = btrfs_drop_extents(trans, root, inode, start,
2279 start + len, 1);
2280 if (ret)
2281 goto out_free_path;
2282 again:
2283 key.objectid = btrfs_ino(inode);
2284 key.type = BTRFS_EXTENT_DATA_KEY;
2285 key.offset = start;
2286
2287 path->leave_spinning = 1;
2288 if (merge) {
2289 struct btrfs_file_extent_item *fi;
2290 u64 extent_len;
2291 struct btrfs_key found_key;
2292
2293 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2294 if (ret < 0)
2295 goto out_free_path;
2296
2297 path->slots[0]--;
2298 leaf = path->nodes[0];
2299 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2300
2301 fi = btrfs_item_ptr(leaf, path->slots[0],
2302 struct btrfs_file_extent_item);
2303 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2304
2305 if (extent_len + found_key.offset == start &&
2306 relink_is_mergable(leaf, fi, new)) {
2307 btrfs_set_file_extent_num_bytes(leaf, fi,
2308 extent_len + len);
2309 btrfs_mark_buffer_dirty(leaf);
2310 inode_add_bytes(inode, len);
2311
2312 ret = 1;
2313 goto out_free_path;
2314 } else {
2315 merge = false;
2316 btrfs_release_path(path);
2317 goto again;
2318 }
2319 }
2320
2321 ret = btrfs_insert_empty_item(trans, root, path, &key,
2322 sizeof(*extent));
2323 if (ret) {
2324 btrfs_abort_transaction(trans, root, ret);
2325 goto out_free_path;
2326 }
2327
2328 leaf = path->nodes[0];
2329 item = btrfs_item_ptr(leaf, path->slots[0],
2330 struct btrfs_file_extent_item);
2331 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2332 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2333 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2334 btrfs_set_file_extent_num_bytes(leaf, item, len);
2335 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2336 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2337 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2338 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2339 btrfs_set_file_extent_encryption(leaf, item, 0);
2340 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2341
2342 btrfs_mark_buffer_dirty(leaf);
2343 inode_add_bytes(inode, len);
2344 btrfs_release_path(path);
2345
2346 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2347 new->disk_len, 0,
2348 backref->root_id, backref->inum,
2349 new->file_pos, 0); /* start - extent_offset */
2350 if (ret) {
2351 btrfs_abort_transaction(trans, root, ret);
2352 goto out_free_path;
2353 }
2354
2355 ret = 1;
2356 out_free_path:
2357 btrfs_release_path(path);
2358 path->leave_spinning = 0;
2359 btrfs_end_transaction(trans, root);
2360 out_unlock:
2361 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2362 &cached, GFP_NOFS);
2363 iput(inode);
2364 return ret;
2365 }
2366
2367 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2368 {
2369 struct old_sa_defrag_extent *old, *tmp;
2370
2371 if (!new)
2372 return;
2373
2374 list_for_each_entry_safe(old, tmp, &new->head, list) {
2375 list_del(&old->list);
2376 kfree(old);
2377 }
2378 kfree(new);
2379 }
2380
2381 static void relink_file_extents(struct new_sa_defrag_extent *new)
2382 {
2383 struct btrfs_path *path;
2384 struct sa_defrag_extent_backref *backref;
2385 struct sa_defrag_extent_backref *prev = NULL;
2386 struct inode *inode;
2387 struct btrfs_root *root;
2388 struct rb_node *node;
2389 int ret;
2390
2391 inode = new->inode;
2392 root = BTRFS_I(inode)->root;
2393
2394 path = btrfs_alloc_path();
2395 if (!path)
2396 return;
2397
2398 if (!record_extent_backrefs(path, new)) {
2399 btrfs_free_path(path);
2400 goto out;
2401 }
2402 btrfs_release_path(path);
2403
2404 while (1) {
2405 node = rb_first(&new->root);
2406 if (!node)
2407 break;
2408 rb_erase(node, &new->root);
2409
2410 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2411
2412 ret = relink_extent_backref(path, prev, backref);
2413 WARN_ON(ret < 0);
2414
2415 kfree(prev);
2416
2417 if (ret == 1)
2418 prev = backref;
2419 else
2420 prev = NULL;
2421 cond_resched();
2422 }
2423 kfree(prev);
2424
2425 btrfs_free_path(path);
2426 out:
2427 free_sa_defrag_extent(new);
2428
2429 atomic_dec(&root->fs_info->defrag_running);
2430 wake_up(&root->fs_info->transaction_wait);
2431 }
2432
2433 static struct new_sa_defrag_extent *
2434 record_old_file_extents(struct inode *inode,
2435 struct btrfs_ordered_extent *ordered)
2436 {
2437 struct btrfs_root *root = BTRFS_I(inode)->root;
2438 struct btrfs_path *path;
2439 struct btrfs_key key;
2440 struct old_sa_defrag_extent *old;
2441 struct new_sa_defrag_extent *new;
2442 int ret;
2443
2444 new = kmalloc(sizeof(*new), GFP_NOFS);
2445 if (!new)
2446 return NULL;
2447
2448 new->inode = inode;
2449 new->file_pos = ordered->file_offset;
2450 new->len = ordered->len;
2451 new->bytenr = ordered->start;
2452 new->disk_len = ordered->disk_len;
2453 new->compress_type = ordered->compress_type;
2454 new->root = RB_ROOT;
2455 INIT_LIST_HEAD(&new->head);
2456
2457 path = btrfs_alloc_path();
2458 if (!path)
2459 goto out_kfree;
2460
2461 key.objectid = btrfs_ino(inode);
2462 key.type = BTRFS_EXTENT_DATA_KEY;
2463 key.offset = new->file_pos;
2464
2465 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2466 if (ret < 0)
2467 goto out_free_path;
2468 if (ret > 0 && path->slots[0] > 0)
2469 path->slots[0]--;
2470
2471 /* find out all the old extents for the file range */
2472 while (1) {
2473 struct btrfs_file_extent_item *extent;
2474 struct extent_buffer *l;
2475 int slot;
2476 u64 num_bytes;
2477 u64 offset;
2478 u64 end;
2479 u64 disk_bytenr;
2480 u64 extent_offset;
2481
2482 l = path->nodes[0];
2483 slot = path->slots[0];
2484
2485 if (slot >= btrfs_header_nritems(l)) {
2486 ret = btrfs_next_leaf(root, path);
2487 if (ret < 0)
2488 goto out_free_path;
2489 else if (ret > 0)
2490 break;
2491 continue;
2492 }
2493
2494 btrfs_item_key_to_cpu(l, &key, slot);
2495
2496 if (key.objectid != btrfs_ino(inode))
2497 break;
2498 if (key.type != BTRFS_EXTENT_DATA_KEY)
2499 break;
2500 if (key.offset >= new->file_pos + new->len)
2501 break;
2502
2503 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2504
2505 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2506 if (key.offset + num_bytes < new->file_pos)
2507 goto next;
2508
2509 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2510 if (!disk_bytenr)
2511 goto next;
2512
2513 extent_offset = btrfs_file_extent_offset(l, extent);
2514
2515 old = kmalloc(sizeof(*old), GFP_NOFS);
2516 if (!old)
2517 goto out_free_path;
2518
2519 offset = max(new->file_pos, key.offset);
2520 end = min(new->file_pos + new->len, key.offset + num_bytes);
2521
2522 old->bytenr = disk_bytenr;
2523 old->extent_offset = extent_offset;
2524 old->offset = offset - key.offset;
2525 old->len = end - offset;
2526 old->new = new;
2527 old->count = 0;
2528 list_add_tail(&old->list, &new->head);
2529 next:
2530 path->slots[0]++;
2531 cond_resched();
2532 }
2533
2534 btrfs_free_path(path);
2535 atomic_inc(&root->fs_info->defrag_running);
2536
2537 return new;
2538
2539 out_free_path:
2540 btrfs_free_path(path);
2541 out_kfree:
2542 free_sa_defrag_extent(new);
2543 return NULL;
2544 }
2545
2546 /* as ordered data IO finishes, this gets called so we can finish
2547 * an ordered extent if the range of bytes in the file it covers are
2548 * fully written.
2549 */
2550 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2551 {
2552 struct inode *inode = ordered_extent->inode;
2553 struct btrfs_root *root = BTRFS_I(inode)->root;
2554 struct btrfs_trans_handle *trans = NULL;
2555 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2556 struct extent_state *cached_state = NULL;
2557 struct new_sa_defrag_extent *new = NULL;
2558 int compress_type = 0;
2559 int ret = 0;
2560 u64 logical_len = ordered_extent->len;
2561 bool nolock;
2562 bool truncated = false;
2563
2564 nolock = btrfs_is_free_space_inode(inode);
2565
2566 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2567 ret = -EIO;
2568 goto out;
2569 }
2570
2571 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2572 truncated = true;
2573 logical_len = ordered_extent->truncated_len;
2574 /* Truncated the entire extent, don't bother adding */
2575 if (!logical_len)
2576 goto out;
2577 }
2578
2579 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2580 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2581 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2582 if (nolock)
2583 trans = btrfs_join_transaction_nolock(root);
2584 else
2585 trans = btrfs_join_transaction(root);
2586 if (IS_ERR(trans)) {
2587 ret = PTR_ERR(trans);
2588 trans = NULL;
2589 goto out;
2590 }
2591 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2592 ret = btrfs_update_inode_fallback(trans, root, inode);
2593 if (ret) /* -ENOMEM or corruption */
2594 btrfs_abort_transaction(trans, root, ret);
2595 goto out;
2596 }
2597
2598 lock_extent_bits(io_tree, ordered_extent->file_offset,
2599 ordered_extent->file_offset + ordered_extent->len - 1,
2600 0, &cached_state);
2601
2602 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2603 ordered_extent->file_offset + ordered_extent->len - 1,
2604 EXTENT_DEFRAG, 1, cached_state);
2605 if (ret) {
2606 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2607 if (last_snapshot >= BTRFS_I(inode)->generation)
2608 /* the inode is shared */
2609 new = record_old_file_extents(inode, ordered_extent);
2610
2611 clear_extent_bit(io_tree, ordered_extent->file_offset,
2612 ordered_extent->file_offset + ordered_extent->len - 1,
2613 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2614 }
2615
2616 if (nolock)
2617 trans = btrfs_join_transaction_nolock(root);
2618 else
2619 trans = btrfs_join_transaction(root);
2620 if (IS_ERR(trans)) {
2621 ret = PTR_ERR(trans);
2622 trans = NULL;
2623 goto out_unlock;
2624 }
2625 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2626
2627 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2628 compress_type = ordered_extent->compress_type;
2629 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2630 BUG_ON(compress_type);
2631 ret = btrfs_mark_extent_written(trans, inode,
2632 ordered_extent->file_offset,
2633 ordered_extent->file_offset +
2634 logical_len);
2635 } else {
2636 BUG_ON(root == root->fs_info->tree_root);
2637 ret = insert_reserved_file_extent(trans, inode,
2638 ordered_extent->file_offset,
2639 ordered_extent->start,
2640 ordered_extent->disk_len,
2641 logical_len, logical_len,
2642 compress_type, 0, 0,
2643 BTRFS_FILE_EXTENT_REG);
2644 }
2645 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2646 ordered_extent->file_offset, ordered_extent->len,
2647 trans->transid);
2648 if (ret < 0) {
2649 btrfs_abort_transaction(trans, root, ret);
2650 goto out_unlock;
2651 }
2652
2653 add_pending_csums(trans, inode, ordered_extent->file_offset,
2654 &ordered_extent->list);
2655
2656 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2657 ret = btrfs_update_inode_fallback(trans, root, inode);
2658 if (ret) { /* -ENOMEM or corruption */
2659 btrfs_abort_transaction(trans, root, ret);
2660 goto out_unlock;
2661 }
2662 ret = 0;
2663 out_unlock:
2664 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2665 ordered_extent->file_offset +
2666 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2667 out:
2668 if (root != root->fs_info->tree_root)
2669 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2670 if (trans)
2671 btrfs_end_transaction(trans, root);
2672
2673 if (ret || truncated) {
2674 u64 start, end;
2675
2676 if (truncated)
2677 start = ordered_extent->file_offset + logical_len;
2678 else
2679 start = ordered_extent->file_offset;
2680 end = ordered_extent->file_offset + ordered_extent->len - 1;
2681 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2682
2683 /* Drop the cache for the part of the extent we didn't write. */
2684 btrfs_drop_extent_cache(inode, start, end, 0);
2685
2686 /*
2687 * If the ordered extent had an IOERR or something else went
2688 * wrong we need to return the space for this ordered extent
2689 * back to the allocator. We only free the extent in the
2690 * truncated case if we didn't write out the extent at all.
2691 */
2692 if ((ret || !logical_len) &&
2693 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2694 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2695 btrfs_free_reserved_extent(root, ordered_extent->start,
2696 ordered_extent->disk_len);
2697 }
2698
2699
2700 /*
2701 * This needs to be done to make sure anybody waiting knows we are done
2702 * updating everything for this ordered extent.
2703 */
2704 btrfs_remove_ordered_extent(inode, ordered_extent);
2705
2706 /* for snapshot-aware defrag */
2707 if (new) {
2708 if (ret) {
2709 free_sa_defrag_extent(new);
2710 atomic_dec(&root->fs_info->defrag_running);
2711 } else {
2712 relink_file_extents(new);
2713 }
2714 }
2715
2716 /* once for us */
2717 btrfs_put_ordered_extent(ordered_extent);
2718 /* once for the tree */
2719 btrfs_put_ordered_extent(ordered_extent);
2720
2721 return ret;
2722 }
2723
2724 static void finish_ordered_fn(struct btrfs_work *work)
2725 {
2726 struct btrfs_ordered_extent *ordered_extent;
2727 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2728 btrfs_finish_ordered_io(ordered_extent);
2729 }
2730
2731 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2732 struct extent_state *state, int uptodate)
2733 {
2734 struct inode *inode = page->mapping->host;
2735 struct btrfs_root *root = BTRFS_I(inode)->root;
2736 struct btrfs_ordered_extent *ordered_extent = NULL;
2737 struct btrfs_workers *workers;
2738
2739 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2740
2741 ClearPagePrivate2(page);
2742 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2743 end - start + 1, uptodate))
2744 return 0;
2745
2746 ordered_extent->work.func = finish_ordered_fn;
2747 ordered_extent->work.flags = 0;
2748
2749 if (btrfs_is_free_space_inode(inode))
2750 workers = &root->fs_info->endio_freespace_worker;
2751 else
2752 workers = &root->fs_info->endio_write_workers;
2753 btrfs_queue_worker(workers, &ordered_extent->work);
2754
2755 return 0;
2756 }
2757
2758 /*
2759 * when reads are done, we need to check csums to verify the data is correct
2760 * if there's a match, we allow the bio to finish. If not, the code in
2761 * extent_io.c will try to find good copies for us.
2762 */
2763 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2764 u64 phy_offset, struct page *page,
2765 u64 start, u64 end, int mirror)
2766 {
2767 size_t offset = start - page_offset(page);
2768 struct inode *inode = page->mapping->host;
2769 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2770 char *kaddr;
2771 struct btrfs_root *root = BTRFS_I(inode)->root;
2772 u32 csum_expected;
2773 u32 csum = ~(u32)0;
2774 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2775 DEFAULT_RATELIMIT_BURST);
2776
2777 if (PageChecked(page)) {
2778 ClearPageChecked(page);
2779 goto good;
2780 }
2781
2782 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2783 goto good;
2784
2785 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2786 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2787 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2788 GFP_NOFS);
2789 return 0;
2790 }
2791
2792 phy_offset >>= inode->i_sb->s_blocksize_bits;
2793 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2794
2795 kaddr = kmap_atomic(page);
2796 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2797 btrfs_csum_final(csum, (char *)&csum);
2798 if (csum != csum_expected)
2799 goto zeroit;
2800
2801 kunmap_atomic(kaddr);
2802 good:
2803 return 0;
2804
2805 zeroit:
2806 if (__ratelimit(&_rs))
2807 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2808 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2809 memset(kaddr + offset, 1, end - start + 1);
2810 flush_dcache_page(page);
2811 kunmap_atomic(kaddr);
2812 if (csum_expected == 0)
2813 return 0;
2814 return -EIO;
2815 }
2816
2817 struct delayed_iput {
2818 struct list_head list;
2819 struct inode *inode;
2820 };
2821
2822 /* JDM: If this is fs-wide, why can't we add a pointer to
2823 * btrfs_inode instead and avoid the allocation? */
2824 void btrfs_add_delayed_iput(struct inode *inode)
2825 {
2826 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2827 struct delayed_iput *delayed;
2828
2829 if (atomic_add_unless(&inode->i_count, -1, 1))
2830 return;
2831
2832 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2833 delayed->inode = inode;
2834
2835 spin_lock(&fs_info->delayed_iput_lock);
2836 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2837 spin_unlock(&fs_info->delayed_iput_lock);
2838 }
2839
2840 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2841 {
2842 LIST_HEAD(list);
2843 struct btrfs_fs_info *fs_info = root->fs_info;
2844 struct delayed_iput *delayed;
2845 int empty;
2846
2847 spin_lock(&fs_info->delayed_iput_lock);
2848 empty = list_empty(&fs_info->delayed_iputs);
2849 spin_unlock(&fs_info->delayed_iput_lock);
2850 if (empty)
2851 return;
2852
2853 spin_lock(&fs_info->delayed_iput_lock);
2854 list_splice_init(&fs_info->delayed_iputs, &list);
2855 spin_unlock(&fs_info->delayed_iput_lock);
2856
2857 while (!list_empty(&list)) {
2858 delayed = list_entry(list.next, struct delayed_iput, list);
2859 list_del(&delayed->list);
2860 iput(delayed->inode);
2861 kfree(delayed);
2862 }
2863 }
2864
2865 /*
2866 * This is called in transaction commit time. If there are no orphan
2867 * files in the subvolume, it removes orphan item and frees block_rsv
2868 * structure.
2869 */
2870 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root)
2872 {
2873 struct btrfs_block_rsv *block_rsv;
2874 int ret;
2875
2876 if (atomic_read(&root->orphan_inodes) ||
2877 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2878 return;
2879
2880 spin_lock(&root->orphan_lock);
2881 if (atomic_read(&root->orphan_inodes)) {
2882 spin_unlock(&root->orphan_lock);
2883 return;
2884 }
2885
2886 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2887 spin_unlock(&root->orphan_lock);
2888 return;
2889 }
2890
2891 block_rsv = root->orphan_block_rsv;
2892 root->orphan_block_rsv = NULL;
2893 spin_unlock(&root->orphan_lock);
2894
2895 if (root->orphan_item_inserted &&
2896 btrfs_root_refs(&root->root_item) > 0) {
2897 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2898 root->root_key.objectid);
2899 if (ret)
2900 btrfs_abort_transaction(trans, root, ret);
2901 else
2902 root->orphan_item_inserted = 0;
2903 }
2904
2905 if (block_rsv) {
2906 WARN_ON(block_rsv->size > 0);
2907 btrfs_free_block_rsv(root, block_rsv);
2908 }
2909 }
2910
2911 /*
2912 * This creates an orphan entry for the given inode in case something goes
2913 * wrong in the middle of an unlink/truncate.
2914 *
2915 * NOTE: caller of this function should reserve 5 units of metadata for
2916 * this function.
2917 */
2918 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2919 {
2920 struct btrfs_root *root = BTRFS_I(inode)->root;
2921 struct btrfs_block_rsv *block_rsv = NULL;
2922 int reserve = 0;
2923 int insert = 0;
2924 int ret;
2925
2926 if (!root->orphan_block_rsv) {
2927 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2928 if (!block_rsv)
2929 return -ENOMEM;
2930 }
2931
2932 spin_lock(&root->orphan_lock);
2933 if (!root->orphan_block_rsv) {
2934 root->orphan_block_rsv = block_rsv;
2935 } else if (block_rsv) {
2936 btrfs_free_block_rsv(root, block_rsv);
2937 block_rsv = NULL;
2938 }
2939
2940 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2941 &BTRFS_I(inode)->runtime_flags)) {
2942 #if 0
2943 /*
2944 * For proper ENOSPC handling, we should do orphan
2945 * cleanup when mounting. But this introduces backward
2946 * compatibility issue.
2947 */
2948 if (!xchg(&root->orphan_item_inserted, 1))
2949 insert = 2;
2950 else
2951 insert = 1;
2952 #endif
2953 insert = 1;
2954 atomic_inc(&root->orphan_inodes);
2955 }
2956
2957 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2958 &BTRFS_I(inode)->runtime_flags))
2959 reserve = 1;
2960 spin_unlock(&root->orphan_lock);
2961
2962 /* grab metadata reservation from transaction handle */
2963 if (reserve) {
2964 ret = btrfs_orphan_reserve_metadata(trans, inode);
2965 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2966 }
2967
2968 /* insert an orphan item to track this unlinked/truncated file */
2969 if (insert >= 1) {
2970 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2971 if (ret) {
2972 atomic_dec(&root->orphan_inodes);
2973 if (reserve) {
2974 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2975 &BTRFS_I(inode)->runtime_flags);
2976 btrfs_orphan_release_metadata(inode);
2977 }
2978 if (ret != -EEXIST) {
2979 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980 &BTRFS_I(inode)->runtime_flags);
2981 btrfs_abort_transaction(trans, root, ret);
2982 return ret;
2983 }
2984 }
2985 ret = 0;
2986 }
2987
2988 /* insert an orphan item to track subvolume contains orphan files */
2989 if (insert >= 2) {
2990 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2991 root->root_key.objectid);
2992 if (ret && ret != -EEXIST) {
2993 btrfs_abort_transaction(trans, root, ret);
2994 return ret;
2995 }
2996 }
2997 return 0;
2998 }
2999
3000 /*
3001 * We have done the truncate/delete so we can go ahead and remove the orphan
3002 * item for this particular inode.
3003 */
3004 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3005 struct inode *inode)
3006 {
3007 struct btrfs_root *root = BTRFS_I(inode)->root;
3008 int delete_item = 0;
3009 int release_rsv = 0;
3010 int ret = 0;
3011
3012 spin_lock(&root->orphan_lock);
3013 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3014 &BTRFS_I(inode)->runtime_flags))
3015 delete_item = 1;
3016
3017 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3018 &BTRFS_I(inode)->runtime_flags))
3019 release_rsv = 1;
3020 spin_unlock(&root->orphan_lock);
3021
3022 if (delete_item) {
3023 atomic_dec(&root->orphan_inodes);
3024 if (trans)
3025 ret = btrfs_del_orphan_item(trans, root,
3026 btrfs_ino(inode));
3027 }
3028
3029 if (release_rsv)
3030 btrfs_orphan_release_metadata(inode);
3031
3032 return ret;
3033 }
3034
3035 /*
3036 * this cleans up any orphans that may be left on the list from the last use
3037 * of this root.
3038 */
3039 int btrfs_orphan_cleanup(struct btrfs_root *root)
3040 {
3041 struct btrfs_path *path;
3042 struct extent_buffer *leaf;
3043 struct btrfs_key key, found_key;
3044 struct btrfs_trans_handle *trans;
3045 struct inode *inode;
3046 u64 last_objectid = 0;
3047 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3048
3049 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3050 return 0;
3051
3052 path = btrfs_alloc_path();
3053 if (!path) {
3054 ret = -ENOMEM;
3055 goto out;
3056 }
3057 path->reada = -1;
3058
3059 key.objectid = BTRFS_ORPHAN_OBJECTID;
3060 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3061 key.offset = (u64)-1;
3062
3063 while (1) {
3064 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3065 if (ret < 0)
3066 goto out;
3067
3068 /*
3069 * if ret == 0 means we found what we were searching for, which
3070 * is weird, but possible, so only screw with path if we didn't
3071 * find the key and see if we have stuff that matches
3072 */
3073 if (ret > 0) {
3074 ret = 0;
3075 if (path->slots[0] == 0)
3076 break;
3077 path->slots[0]--;
3078 }
3079
3080 /* pull out the item */
3081 leaf = path->nodes[0];
3082 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3083
3084 /* make sure the item matches what we want */
3085 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3086 break;
3087 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3088 break;
3089
3090 /* release the path since we're done with it */
3091 btrfs_release_path(path);
3092
3093 /*
3094 * this is where we are basically btrfs_lookup, without the
3095 * crossing root thing. we store the inode number in the
3096 * offset of the orphan item.
3097 */
3098
3099 if (found_key.offset == last_objectid) {
3100 btrfs_err(root->fs_info,
3101 "Error removing orphan entry, stopping orphan cleanup");
3102 ret = -EINVAL;
3103 goto out;
3104 }
3105
3106 last_objectid = found_key.offset;
3107
3108 found_key.objectid = found_key.offset;
3109 found_key.type = BTRFS_INODE_ITEM_KEY;
3110 found_key.offset = 0;
3111 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3112 ret = PTR_ERR_OR_ZERO(inode);
3113 if (ret && ret != -ESTALE)
3114 goto out;
3115
3116 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3117 struct btrfs_root *dead_root;
3118 struct btrfs_fs_info *fs_info = root->fs_info;
3119 int is_dead_root = 0;
3120
3121 /*
3122 * this is an orphan in the tree root. Currently these
3123 * could come from 2 sources:
3124 * a) a snapshot deletion in progress
3125 * b) a free space cache inode
3126 * We need to distinguish those two, as the snapshot
3127 * orphan must not get deleted.
3128 * find_dead_roots already ran before us, so if this
3129 * is a snapshot deletion, we should find the root
3130 * in the dead_roots list
3131 */
3132 spin_lock(&fs_info->trans_lock);
3133 list_for_each_entry(dead_root, &fs_info->dead_roots,
3134 root_list) {
3135 if (dead_root->root_key.objectid ==
3136 found_key.objectid) {
3137 is_dead_root = 1;
3138 break;
3139 }
3140 }
3141 spin_unlock(&fs_info->trans_lock);
3142 if (is_dead_root) {
3143 /* prevent this orphan from being found again */
3144 key.offset = found_key.objectid - 1;
3145 continue;
3146 }
3147 }
3148 /*
3149 * Inode is already gone but the orphan item is still there,
3150 * kill the orphan item.
3151 */
3152 if (ret == -ESTALE) {
3153 trans = btrfs_start_transaction(root, 1);
3154 if (IS_ERR(trans)) {
3155 ret = PTR_ERR(trans);
3156 goto out;
3157 }
3158 btrfs_debug(root->fs_info, "auto deleting %Lu",
3159 found_key.objectid);
3160 ret = btrfs_del_orphan_item(trans, root,
3161 found_key.objectid);
3162 btrfs_end_transaction(trans, root);
3163 if (ret)
3164 goto out;
3165 continue;
3166 }
3167
3168 /*
3169 * add this inode to the orphan list so btrfs_orphan_del does
3170 * the proper thing when we hit it
3171 */
3172 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3173 &BTRFS_I(inode)->runtime_flags);
3174 atomic_inc(&root->orphan_inodes);
3175
3176 /* if we have links, this was a truncate, lets do that */
3177 if (inode->i_nlink) {
3178 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3179 iput(inode);
3180 continue;
3181 }
3182 nr_truncate++;
3183
3184 /* 1 for the orphan item deletion. */
3185 trans = btrfs_start_transaction(root, 1);
3186 if (IS_ERR(trans)) {
3187 iput(inode);
3188 ret = PTR_ERR(trans);
3189 goto out;
3190 }
3191 ret = btrfs_orphan_add(trans, inode);
3192 btrfs_end_transaction(trans, root);
3193 if (ret) {
3194 iput(inode);
3195 goto out;
3196 }
3197
3198 ret = btrfs_truncate(inode);
3199 if (ret)
3200 btrfs_orphan_del(NULL, inode);
3201 } else {
3202 nr_unlink++;
3203 }
3204
3205 /* this will do delete_inode and everything for us */
3206 iput(inode);
3207 if (ret)
3208 goto out;
3209 }
3210 /* release the path since we're done with it */
3211 btrfs_release_path(path);
3212
3213 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3214
3215 if (root->orphan_block_rsv)
3216 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3217 (u64)-1);
3218
3219 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3220 trans = btrfs_join_transaction(root);
3221 if (!IS_ERR(trans))
3222 btrfs_end_transaction(trans, root);
3223 }
3224
3225 if (nr_unlink)
3226 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3227 if (nr_truncate)
3228 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3229
3230 out:
3231 if (ret)
3232 btrfs_crit(root->fs_info,
3233 "could not do orphan cleanup %d", ret);
3234 btrfs_free_path(path);
3235 return ret;
3236 }
3237
3238 /*
3239 * very simple check to peek ahead in the leaf looking for xattrs. If we
3240 * don't find any xattrs, we know there can't be any acls.
3241 *
3242 * slot is the slot the inode is in, objectid is the objectid of the inode
3243 */
3244 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3245 int slot, u64 objectid)
3246 {
3247 u32 nritems = btrfs_header_nritems(leaf);
3248 struct btrfs_key found_key;
3249 static u64 xattr_access = 0;
3250 static u64 xattr_default = 0;
3251 int scanned = 0;
3252
3253 if (!xattr_access) {
3254 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3255 strlen(POSIX_ACL_XATTR_ACCESS));
3256 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3257 strlen(POSIX_ACL_XATTR_DEFAULT));
3258 }
3259
3260 slot++;
3261 while (slot < nritems) {
3262 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3263
3264 /* we found a different objectid, there must not be acls */
3265 if (found_key.objectid != objectid)
3266 return 0;
3267
3268 /* we found an xattr, assume we've got an acl */
3269 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3270 if (found_key.offset == xattr_access ||
3271 found_key.offset == xattr_default)
3272 return 1;
3273 }
3274
3275 /*
3276 * we found a key greater than an xattr key, there can't
3277 * be any acls later on
3278 */
3279 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3280 return 0;
3281
3282 slot++;
3283 scanned++;
3284
3285 /*
3286 * it goes inode, inode backrefs, xattrs, extents,
3287 * so if there are a ton of hard links to an inode there can
3288 * be a lot of backrefs. Don't waste time searching too hard,
3289 * this is just an optimization
3290 */
3291 if (scanned >= 8)
3292 break;
3293 }
3294 /* we hit the end of the leaf before we found an xattr or
3295 * something larger than an xattr. We have to assume the inode
3296 * has acls
3297 */
3298 return 1;
3299 }
3300
3301 /*
3302 * read an inode from the btree into the in-memory inode
3303 */
3304 static void btrfs_read_locked_inode(struct inode *inode)
3305 {
3306 struct btrfs_path *path;
3307 struct extent_buffer *leaf;
3308 struct btrfs_inode_item *inode_item;
3309 struct btrfs_timespec *tspec;
3310 struct btrfs_root *root = BTRFS_I(inode)->root;
3311 struct btrfs_key location;
3312 int maybe_acls;
3313 u32 rdev;
3314 int ret;
3315 bool filled = false;
3316
3317 ret = btrfs_fill_inode(inode, &rdev);
3318 if (!ret)
3319 filled = true;
3320
3321 path = btrfs_alloc_path();
3322 if (!path)
3323 goto make_bad;
3324
3325 path->leave_spinning = 1;
3326 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3327
3328 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3329 if (ret)
3330 goto make_bad;
3331
3332 leaf = path->nodes[0];
3333
3334 if (filled)
3335 goto cache_acl;
3336
3337 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3338 struct btrfs_inode_item);
3339 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3340 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3341 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3342 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3343 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3344
3345 tspec = btrfs_inode_atime(inode_item);
3346 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3347 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3348
3349 tspec = btrfs_inode_mtime(inode_item);
3350 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3351 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3352
3353 tspec = btrfs_inode_ctime(inode_item);
3354 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3355 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3356
3357 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3358 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3359 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3360
3361 /*
3362 * If we were modified in the current generation and evicted from memory
3363 * and then re-read we need to do a full sync since we don't have any
3364 * idea about which extents were modified before we were evicted from
3365 * cache.
3366 */
3367 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3368 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3369 &BTRFS_I(inode)->runtime_flags);
3370
3371 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3372 inode->i_generation = BTRFS_I(inode)->generation;
3373 inode->i_rdev = 0;
3374 rdev = btrfs_inode_rdev(leaf, inode_item);
3375
3376 BTRFS_I(inode)->index_cnt = (u64)-1;
3377 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3378 cache_acl:
3379 /*
3380 * try to precache a NULL acl entry for files that don't have
3381 * any xattrs or acls
3382 */
3383 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3384 btrfs_ino(inode));
3385 if (!maybe_acls)
3386 cache_no_acl(inode);
3387
3388 btrfs_free_path(path);
3389
3390 switch (inode->i_mode & S_IFMT) {
3391 case S_IFREG:
3392 inode->i_mapping->a_ops = &btrfs_aops;
3393 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3394 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3395 inode->i_fop = &btrfs_file_operations;
3396 inode->i_op = &btrfs_file_inode_operations;
3397 break;
3398 case S_IFDIR:
3399 inode->i_fop = &btrfs_dir_file_operations;
3400 if (root == root->fs_info->tree_root)
3401 inode->i_op = &btrfs_dir_ro_inode_operations;
3402 else
3403 inode->i_op = &btrfs_dir_inode_operations;
3404 break;
3405 case S_IFLNK:
3406 inode->i_op = &btrfs_symlink_inode_operations;
3407 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3408 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3409 break;
3410 default:
3411 inode->i_op = &btrfs_special_inode_operations;
3412 init_special_inode(inode, inode->i_mode, rdev);
3413 break;
3414 }
3415
3416 btrfs_update_iflags(inode);
3417 return;
3418
3419 make_bad:
3420 btrfs_free_path(path);
3421 make_bad_inode(inode);
3422 }
3423
3424 /*
3425 * given a leaf and an inode, copy the inode fields into the leaf
3426 */
3427 static void fill_inode_item(struct btrfs_trans_handle *trans,
3428 struct extent_buffer *leaf,
3429 struct btrfs_inode_item *item,
3430 struct inode *inode)
3431 {
3432 struct btrfs_map_token token;
3433
3434 btrfs_init_map_token(&token);
3435
3436 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3437 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3438 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3439 &token);
3440 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3441 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3442
3443 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3444 inode->i_atime.tv_sec, &token);
3445 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3446 inode->i_atime.tv_nsec, &token);
3447
3448 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3449 inode->i_mtime.tv_sec, &token);
3450 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3451 inode->i_mtime.tv_nsec, &token);
3452
3453 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3454 inode->i_ctime.tv_sec, &token);
3455 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3456 inode->i_ctime.tv_nsec, &token);
3457
3458 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3459 &token);
3460 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3461 &token);
3462 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3463 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3464 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3465 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3466 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3467 }
3468
3469 /*
3470 * copy everything in the in-memory inode into the btree.
3471 */
3472 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3473 struct btrfs_root *root, struct inode *inode)
3474 {
3475 struct btrfs_inode_item *inode_item;
3476 struct btrfs_path *path;
3477 struct extent_buffer *leaf;
3478 int ret;
3479
3480 path = btrfs_alloc_path();
3481 if (!path)
3482 return -ENOMEM;
3483
3484 path->leave_spinning = 1;
3485 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3486 1);
3487 if (ret) {
3488 if (ret > 0)
3489 ret = -ENOENT;
3490 goto failed;
3491 }
3492
3493 btrfs_unlock_up_safe(path, 1);
3494 leaf = path->nodes[0];
3495 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3496 struct btrfs_inode_item);
3497
3498 fill_inode_item(trans, leaf, inode_item, inode);
3499 btrfs_mark_buffer_dirty(leaf);
3500 btrfs_set_inode_last_trans(trans, inode);
3501 ret = 0;
3502 failed:
3503 btrfs_free_path(path);
3504 return ret;
3505 }
3506
3507 /*
3508 * copy everything in the in-memory inode into the btree.
3509 */
3510 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3511 struct btrfs_root *root, struct inode *inode)
3512 {
3513 int ret;
3514
3515 /*
3516 * If the inode is a free space inode, we can deadlock during commit
3517 * if we put it into the delayed code.
3518 *
3519 * The data relocation inode should also be directly updated
3520 * without delay
3521 */
3522 if (!btrfs_is_free_space_inode(inode)
3523 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3524 btrfs_update_root_times(trans, root);
3525
3526 ret = btrfs_delayed_update_inode(trans, root, inode);
3527 if (!ret)
3528 btrfs_set_inode_last_trans(trans, inode);
3529 return ret;
3530 }
3531
3532 return btrfs_update_inode_item(trans, root, inode);
3533 }
3534
3535 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3536 struct btrfs_root *root,
3537 struct inode *inode)
3538 {
3539 int ret;
3540
3541 ret = btrfs_update_inode(trans, root, inode);
3542 if (ret == -ENOSPC)
3543 return btrfs_update_inode_item(trans, root, inode);
3544 return ret;
3545 }
3546
3547 /*
3548 * unlink helper that gets used here in inode.c and in the tree logging
3549 * recovery code. It remove a link in a directory with a given name, and
3550 * also drops the back refs in the inode to the directory
3551 */
3552 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3553 struct btrfs_root *root,
3554 struct inode *dir, struct inode *inode,
3555 const char *name, int name_len)
3556 {
3557 struct btrfs_path *path;
3558 int ret = 0;
3559 struct extent_buffer *leaf;
3560 struct btrfs_dir_item *di;
3561 struct btrfs_key key;
3562 u64 index;
3563 u64 ino = btrfs_ino(inode);
3564 u64 dir_ino = btrfs_ino(dir);
3565
3566 path = btrfs_alloc_path();
3567 if (!path) {
3568 ret = -ENOMEM;
3569 goto out;
3570 }
3571
3572 path->leave_spinning = 1;
3573 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3574 name, name_len, -1);
3575 if (IS_ERR(di)) {
3576 ret = PTR_ERR(di);
3577 goto err;
3578 }
3579 if (!di) {
3580 ret = -ENOENT;
3581 goto err;
3582 }
3583 leaf = path->nodes[0];
3584 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3585 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3586 if (ret)
3587 goto err;
3588 btrfs_release_path(path);
3589
3590 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3591 dir_ino, &index);
3592 if (ret) {
3593 btrfs_info(root->fs_info,
3594 "failed to delete reference to %.*s, inode %llu parent %llu",
3595 name_len, name, ino, dir_ino);
3596 btrfs_abort_transaction(trans, root, ret);
3597 goto err;
3598 }
3599
3600 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3601 if (ret) {
3602 btrfs_abort_transaction(trans, root, ret);
3603 goto err;
3604 }
3605
3606 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3607 inode, dir_ino);
3608 if (ret != 0 && ret != -ENOENT) {
3609 btrfs_abort_transaction(trans, root, ret);
3610 goto err;
3611 }
3612
3613 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3614 dir, index);
3615 if (ret == -ENOENT)
3616 ret = 0;
3617 else if (ret)
3618 btrfs_abort_transaction(trans, root, ret);
3619 err:
3620 btrfs_free_path(path);
3621 if (ret)
3622 goto out;
3623
3624 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3625 inode_inc_iversion(inode);
3626 inode_inc_iversion(dir);
3627 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3628 ret = btrfs_update_inode(trans, root, dir);
3629 out:
3630 return ret;
3631 }
3632
3633 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3634 struct btrfs_root *root,
3635 struct inode *dir, struct inode *inode,
3636 const char *name, int name_len)
3637 {
3638 int ret;
3639 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3640 if (!ret) {
3641 drop_nlink(inode);
3642 ret = btrfs_update_inode(trans, root, inode);
3643 }
3644 return ret;
3645 }
3646
3647 /*
3648 * helper to start transaction for unlink and rmdir.
3649 *
3650 * unlink and rmdir are special in btrfs, they do not always free space, so
3651 * if we cannot make our reservations the normal way try and see if there is
3652 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3653 * allow the unlink to occur.
3654 */
3655 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3656 {
3657 struct btrfs_trans_handle *trans;
3658 struct btrfs_root *root = BTRFS_I(dir)->root;
3659 int ret;
3660
3661 /*
3662 * 1 for the possible orphan item
3663 * 1 for the dir item
3664 * 1 for the dir index
3665 * 1 for the inode ref
3666 * 1 for the inode
3667 */
3668 trans = btrfs_start_transaction(root, 5);
3669 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3670 return trans;
3671
3672 if (PTR_ERR(trans) == -ENOSPC) {
3673 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3674
3675 trans = btrfs_start_transaction(root, 0);
3676 if (IS_ERR(trans))
3677 return trans;
3678 ret = btrfs_cond_migrate_bytes(root->fs_info,
3679 &root->fs_info->trans_block_rsv,
3680 num_bytes, 5);
3681 if (ret) {
3682 btrfs_end_transaction(trans, root);
3683 return ERR_PTR(ret);
3684 }
3685 trans->block_rsv = &root->fs_info->trans_block_rsv;
3686 trans->bytes_reserved = num_bytes;
3687 }
3688 return trans;
3689 }
3690
3691 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3692 {
3693 struct btrfs_root *root = BTRFS_I(dir)->root;
3694 struct btrfs_trans_handle *trans;
3695 struct inode *inode = dentry->d_inode;
3696 int ret;
3697
3698 trans = __unlink_start_trans(dir);
3699 if (IS_ERR(trans))
3700 return PTR_ERR(trans);
3701
3702 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3703
3704 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3705 dentry->d_name.name, dentry->d_name.len);
3706 if (ret)
3707 goto out;
3708
3709 if (inode->i_nlink == 0) {
3710 ret = btrfs_orphan_add(trans, inode);
3711 if (ret)
3712 goto out;
3713 }
3714
3715 out:
3716 btrfs_end_transaction(trans, root);
3717 btrfs_btree_balance_dirty(root);
3718 return ret;
3719 }
3720
3721 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3722 struct btrfs_root *root,
3723 struct inode *dir, u64 objectid,
3724 const char *name, int name_len)
3725 {
3726 struct btrfs_path *path;
3727 struct extent_buffer *leaf;
3728 struct btrfs_dir_item *di;
3729 struct btrfs_key key;
3730 u64 index;
3731 int ret;
3732 u64 dir_ino = btrfs_ino(dir);
3733
3734 path = btrfs_alloc_path();
3735 if (!path)
3736 return -ENOMEM;
3737
3738 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3739 name, name_len, -1);
3740 if (IS_ERR_OR_NULL(di)) {
3741 if (!di)
3742 ret = -ENOENT;
3743 else
3744 ret = PTR_ERR(di);
3745 goto out;
3746 }
3747
3748 leaf = path->nodes[0];
3749 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3750 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3751 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3752 if (ret) {
3753 btrfs_abort_transaction(trans, root, ret);
3754 goto out;
3755 }
3756 btrfs_release_path(path);
3757
3758 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3759 objectid, root->root_key.objectid,
3760 dir_ino, &index, name, name_len);
3761 if (ret < 0) {
3762 if (ret != -ENOENT) {
3763 btrfs_abort_transaction(trans, root, ret);
3764 goto out;
3765 }
3766 di = btrfs_search_dir_index_item(root, path, dir_ino,
3767 name, name_len);
3768 if (IS_ERR_OR_NULL(di)) {
3769 if (!di)
3770 ret = -ENOENT;
3771 else
3772 ret = PTR_ERR(di);
3773 btrfs_abort_transaction(trans, root, ret);
3774 goto out;
3775 }
3776
3777 leaf = path->nodes[0];
3778 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3779 btrfs_release_path(path);
3780 index = key.offset;
3781 }
3782 btrfs_release_path(path);
3783
3784 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3785 if (ret) {
3786 btrfs_abort_transaction(trans, root, ret);
3787 goto out;
3788 }
3789
3790 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3791 inode_inc_iversion(dir);
3792 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3793 ret = btrfs_update_inode_fallback(trans, root, dir);
3794 if (ret)
3795 btrfs_abort_transaction(trans, root, ret);
3796 out:
3797 btrfs_free_path(path);
3798 return ret;
3799 }
3800
3801 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3802 {
3803 struct inode *inode = dentry->d_inode;
3804 int err = 0;
3805 struct btrfs_root *root = BTRFS_I(dir)->root;
3806 struct btrfs_trans_handle *trans;
3807
3808 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3809 return -ENOTEMPTY;
3810 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3811 return -EPERM;
3812
3813 trans = __unlink_start_trans(dir);
3814 if (IS_ERR(trans))
3815 return PTR_ERR(trans);
3816
3817 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3818 err = btrfs_unlink_subvol(trans, root, dir,
3819 BTRFS_I(inode)->location.objectid,
3820 dentry->d_name.name,
3821 dentry->d_name.len);
3822 goto out;
3823 }
3824
3825 err = btrfs_orphan_add(trans, inode);
3826 if (err)
3827 goto out;
3828
3829 /* now the directory is empty */
3830 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3831 dentry->d_name.name, dentry->d_name.len);
3832 if (!err)
3833 btrfs_i_size_write(inode, 0);
3834 out:
3835 btrfs_end_transaction(trans, root);
3836 btrfs_btree_balance_dirty(root);
3837
3838 return err;
3839 }
3840
3841 /*
3842 * this can truncate away extent items, csum items and directory items.
3843 * It starts at a high offset and removes keys until it can't find
3844 * any higher than new_size
3845 *
3846 * csum items that cross the new i_size are truncated to the new size
3847 * as well.
3848 *
3849 * min_type is the minimum key type to truncate down to. If set to 0, this
3850 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3851 */
3852 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3853 struct btrfs_root *root,
3854 struct inode *inode,
3855 u64 new_size, u32 min_type)
3856 {
3857 struct btrfs_path *path;
3858 struct extent_buffer *leaf;
3859 struct btrfs_file_extent_item *fi;
3860 struct btrfs_key key;
3861 struct btrfs_key found_key;
3862 u64 extent_start = 0;
3863 u64 extent_num_bytes = 0;
3864 u64 extent_offset = 0;
3865 u64 item_end = 0;
3866 u64 last_size = (u64)-1;
3867 u32 found_type = (u8)-1;
3868 int found_extent;
3869 int del_item;
3870 int pending_del_nr = 0;
3871 int pending_del_slot = 0;
3872 int extent_type = -1;
3873 int ret;
3874 int err = 0;
3875 u64 ino = btrfs_ino(inode);
3876
3877 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3878
3879 path = btrfs_alloc_path();
3880 if (!path)
3881 return -ENOMEM;
3882 path->reada = -1;
3883
3884 /*
3885 * We want to drop from the next block forward in case this new size is
3886 * not block aligned since we will be keeping the last block of the
3887 * extent just the way it is.
3888 */
3889 if (root->ref_cows || root == root->fs_info->tree_root)
3890 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3891 root->sectorsize), (u64)-1, 0);
3892
3893 /*
3894 * This function is also used to drop the items in the log tree before
3895 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3896 * it is used to drop the loged items. So we shouldn't kill the delayed
3897 * items.
3898 */
3899 if (min_type == 0 && root == BTRFS_I(inode)->root)
3900 btrfs_kill_delayed_inode_items(inode);
3901
3902 key.objectid = ino;
3903 key.offset = (u64)-1;
3904 key.type = (u8)-1;
3905
3906 search_again:
3907 path->leave_spinning = 1;
3908 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3909 if (ret < 0) {
3910 err = ret;
3911 goto out;
3912 }
3913
3914 if (ret > 0) {
3915 /* there are no items in the tree for us to truncate, we're
3916 * done
3917 */
3918 if (path->slots[0] == 0)
3919 goto out;
3920 path->slots[0]--;
3921 }
3922
3923 while (1) {
3924 fi = NULL;
3925 leaf = path->nodes[0];
3926 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3927 found_type = btrfs_key_type(&found_key);
3928
3929 if (found_key.objectid != ino)
3930 break;
3931
3932 if (found_type < min_type)
3933 break;
3934
3935 item_end = found_key.offset;
3936 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3937 fi = btrfs_item_ptr(leaf, path->slots[0],
3938 struct btrfs_file_extent_item);
3939 extent_type = btrfs_file_extent_type(leaf, fi);
3940 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3941 item_end +=
3942 btrfs_file_extent_num_bytes(leaf, fi);
3943 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3944 item_end += btrfs_file_extent_inline_len(leaf,
3945 fi);
3946 }
3947 item_end--;
3948 }
3949 if (found_type > min_type) {
3950 del_item = 1;
3951 } else {
3952 if (item_end < new_size)
3953 break;
3954 if (found_key.offset >= new_size)
3955 del_item = 1;
3956 else
3957 del_item = 0;
3958 }
3959 found_extent = 0;
3960 /* FIXME, shrink the extent if the ref count is only 1 */
3961 if (found_type != BTRFS_EXTENT_DATA_KEY)
3962 goto delete;
3963
3964 if (del_item)
3965 last_size = found_key.offset;
3966 else
3967 last_size = new_size;
3968
3969 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3970 u64 num_dec;
3971 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3972 if (!del_item) {
3973 u64 orig_num_bytes =
3974 btrfs_file_extent_num_bytes(leaf, fi);
3975 extent_num_bytes = ALIGN(new_size -
3976 found_key.offset,
3977 root->sectorsize);
3978 btrfs_set_file_extent_num_bytes(leaf, fi,
3979 extent_num_bytes);
3980 num_dec = (orig_num_bytes -
3981 extent_num_bytes);
3982 if (root->ref_cows && extent_start != 0)
3983 inode_sub_bytes(inode, num_dec);
3984 btrfs_mark_buffer_dirty(leaf);
3985 } else {
3986 extent_num_bytes =
3987 btrfs_file_extent_disk_num_bytes(leaf,
3988 fi);
3989 extent_offset = found_key.offset -
3990 btrfs_file_extent_offset(leaf, fi);
3991
3992 /* FIXME blocksize != 4096 */
3993 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3994 if (extent_start != 0) {
3995 found_extent = 1;
3996 if (root->ref_cows)
3997 inode_sub_bytes(inode, num_dec);
3998 }
3999 }
4000 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4001 /*
4002 * we can't truncate inline items that have had
4003 * special encodings
4004 */
4005 if (!del_item &&
4006 btrfs_file_extent_compression(leaf, fi) == 0 &&
4007 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4008 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4009 u32 size = new_size - found_key.offset;
4010
4011 if (root->ref_cows) {
4012 inode_sub_bytes(inode, item_end + 1 -
4013 new_size);
4014 }
4015 size =
4016 btrfs_file_extent_calc_inline_size(size);
4017 btrfs_truncate_item(root, path, size, 1);
4018 } else if (root->ref_cows) {
4019 inode_sub_bytes(inode, item_end + 1 -
4020 found_key.offset);
4021 }
4022 }
4023 delete:
4024 if (del_item) {
4025 if (!pending_del_nr) {
4026 /* no pending yet, add ourselves */
4027 pending_del_slot = path->slots[0];
4028 pending_del_nr = 1;
4029 } else if (pending_del_nr &&
4030 path->slots[0] + 1 == pending_del_slot) {
4031 /* hop on the pending chunk */
4032 pending_del_nr++;
4033 pending_del_slot = path->slots[0];
4034 } else {
4035 BUG();
4036 }
4037 } else {
4038 break;
4039 }
4040 if (found_extent && (root->ref_cows ||
4041 root == root->fs_info->tree_root)) {
4042 btrfs_set_path_blocking(path);
4043 ret = btrfs_free_extent(trans, root, extent_start,
4044 extent_num_bytes, 0,
4045 btrfs_header_owner(leaf),
4046 ino, extent_offset, 0);
4047 BUG_ON(ret);
4048 }
4049
4050 if (found_type == BTRFS_INODE_ITEM_KEY)
4051 break;
4052
4053 if (path->slots[0] == 0 ||
4054 path->slots[0] != pending_del_slot) {
4055 if (pending_del_nr) {
4056 ret = btrfs_del_items(trans, root, path,
4057 pending_del_slot,
4058 pending_del_nr);
4059 if (ret) {
4060 btrfs_abort_transaction(trans,
4061 root, ret);
4062 goto error;
4063 }
4064 pending_del_nr = 0;
4065 }
4066 btrfs_release_path(path);
4067 goto search_again;
4068 } else {
4069 path->slots[0]--;
4070 }
4071 }
4072 out:
4073 if (pending_del_nr) {
4074 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4075 pending_del_nr);
4076 if (ret)
4077 btrfs_abort_transaction(trans, root, ret);
4078 }
4079 error:
4080 if (last_size != (u64)-1)
4081 btrfs_ordered_update_i_size(inode, last_size, NULL);
4082 btrfs_free_path(path);
4083 return err;
4084 }
4085
4086 /*
4087 * btrfs_truncate_page - read, zero a chunk and write a page
4088 * @inode - inode that we're zeroing
4089 * @from - the offset to start zeroing
4090 * @len - the length to zero, 0 to zero the entire range respective to the
4091 * offset
4092 * @front - zero up to the offset instead of from the offset on
4093 *
4094 * This will find the page for the "from" offset and cow the page and zero the
4095 * part we want to zero. This is used with truncate and hole punching.
4096 */
4097 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4098 int front)
4099 {
4100 struct address_space *mapping = inode->i_mapping;
4101 struct btrfs_root *root = BTRFS_I(inode)->root;
4102 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4103 struct btrfs_ordered_extent *ordered;
4104 struct extent_state *cached_state = NULL;
4105 char *kaddr;
4106 u32 blocksize = root->sectorsize;
4107 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4108 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4109 struct page *page;
4110 gfp_t mask = btrfs_alloc_write_mask(mapping);
4111 int ret = 0;
4112 u64 page_start;
4113 u64 page_end;
4114
4115 if ((offset & (blocksize - 1)) == 0 &&
4116 (!len || ((len & (blocksize - 1)) == 0)))
4117 goto out;
4118 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4119 if (ret)
4120 goto out;
4121
4122 again:
4123 page = find_or_create_page(mapping, index, mask);
4124 if (!page) {
4125 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4126 ret = -ENOMEM;
4127 goto out;
4128 }
4129
4130 page_start = page_offset(page);
4131 page_end = page_start + PAGE_CACHE_SIZE - 1;
4132
4133 if (!PageUptodate(page)) {
4134 ret = btrfs_readpage(NULL, page);
4135 lock_page(page);
4136 if (page->mapping != mapping) {
4137 unlock_page(page);
4138 page_cache_release(page);
4139 goto again;
4140 }
4141 if (!PageUptodate(page)) {
4142 ret = -EIO;
4143 goto out_unlock;
4144 }
4145 }
4146 wait_on_page_writeback(page);
4147
4148 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4149 set_page_extent_mapped(page);
4150
4151 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4152 if (ordered) {
4153 unlock_extent_cached(io_tree, page_start, page_end,
4154 &cached_state, GFP_NOFS);
4155 unlock_page(page);
4156 page_cache_release(page);
4157 btrfs_start_ordered_extent(inode, ordered, 1);
4158 btrfs_put_ordered_extent(ordered);
4159 goto again;
4160 }
4161
4162 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4163 EXTENT_DIRTY | EXTENT_DELALLOC |
4164 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4165 0, 0, &cached_state, GFP_NOFS);
4166
4167 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4168 &cached_state);
4169 if (ret) {
4170 unlock_extent_cached(io_tree, page_start, page_end,
4171 &cached_state, GFP_NOFS);
4172 goto out_unlock;
4173 }
4174
4175 if (offset != PAGE_CACHE_SIZE) {
4176 if (!len)
4177 len = PAGE_CACHE_SIZE - offset;
4178 kaddr = kmap(page);
4179 if (front)
4180 memset(kaddr, 0, offset);
4181 else
4182 memset(kaddr + offset, 0, len);
4183 flush_dcache_page(page);
4184 kunmap(page);
4185 }
4186 ClearPageChecked(page);
4187 set_page_dirty(page);
4188 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4189 GFP_NOFS);
4190
4191 out_unlock:
4192 if (ret)
4193 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4194 unlock_page(page);
4195 page_cache_release(page);
4196 out:
4197 return ret;
4198 }
4199
4200 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4201 u64 offset, u64 len)
4202 {
4203 struct btrfs_trans_handle *trans;
4204 int ret;
4205
4206 /*
4207 * Still need to make sure the inode looks like it's been updated so
4208 * that any holes get logged if we fsync.
4209 */
4210 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4211 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4212 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4213 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4214 return 0;
4215 }
4216
4217 /*
4218 * 1 - for the one we're dropping
4219 * 1 - for the one we're adding
4220 * 1 - for updating the inode.
4221 */
4222 trans = btrfs_start_transaction(root, 3);
4223 if (IS_ERR(trans))
4224 return PTR_ERR(trans);
4225
4226 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4227 if (ret) {
4228 btrfs_abort_transaction(trans, root, ret);
4229 btrfs_end_transaction(trans, root);
4230 return ret;
4231 }
4232
4233 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4234 0, 0, len, 0, len, 0, 0, 0);
4235 if (ret)
4236 btrfs_abort_transaction(trans, root, ret);
4237 else
4238 btrfs_update_inode(trans, root, inode);
4239 btrfs_end_transaction(trans, root);
4240 return ret;
4241 }
4242
4243 /*
4244 * This function puts in dummy file extents for the area we're creating a hole
4245 * for. So if we are truncating this file to a larger size we need to insert
4246 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4247 * the range between oldsize and size
4248 */
4249 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4250 {
4251 struct btrfs_root *root = BTRFS_I(inode)->root;
4252 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4253 struct extent_map *em = NULL;
4254 struct extent_state *cached_state = NULL;
4255 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4256 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4257 u64 block_end = ALIGN(size, root->sectorsize);
4258 u64 last_byte;
4259 u64 cur_offset;
4260 u64 hole_size;
4261 int err = 0;
4262
4263 /*
4264 * If our size started in the middle of a page we need to zero out the
4265 * rest of the page before we expand the i_size, otherwise we could
4266 * expose stale data.
4267 */
4268 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4269 if (err)
4270 return err;
4271
4272 if (size <= hole_start)
4273 return 0;
4274
4275 while (1) {
4276 struct btrfs_ordered_extent *ordered;
4277
4278 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4279 &cached_state);
4280 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4281 block_end - hole_start);
4282 if (!ordered)
4283 break;
4284 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4285 &cached_state, GFP_NOFS);
4286 btrfs_start_ordered_extent(inode, ordered, 1);
4287 btrfs_put_ordered_extent(ordered);
4288 }
4289
4290 cur_offset = hole_start;
4291 while (1) {
4292 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4293 block_end - cur_offset, 0);
4294 if (IS_ERR(em)) {
4295 err = PTR_ERR(em);
4296 em = NULL;
4297 break;
4298 }
4299 last_byte = min(extent_map_end(em), block_end);
4300 last_byte = ALIGN(last_byte , root->sectorsize);
4301 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4302 struct extent_map *hole_em;
4303 hole_size = last_byte - cur_offset;
4304
4305 err = maybe_insert_hole(root, inode, cur_offset,
4306 hole_size);
4307 if (err)
4308 break;
4309 btrfs_drop_extent_cache(inode, cur_offset,
4310 cur_offset + hole_size - 1, 0);
4311 hole_em = alloc_extent_map();
4312 if (!hole_em) {
4313 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4314 &BTRFS_I(inode)->runtime_flags);
4315 goto next;
4316 }
4317 hole_em->start = cur_offset;
4318 hole_em->len = hole_size;
4319 hole_em->orig_start = cur_offset;
4320
4321 hole_em->block_start = EXTENT_MAP_HOLE;
4322 hole_em->block_len = 0;
4323 hole_em->orig_block_len = 0;
4324 hole_em->ram_bytes = hole_size;
4325 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4326 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4327 hole_em->generation = root->fs_info->generation;
4328
4329 while (1) {
4330 write_lock(&em_tree->lock);
4331 err = add_extent_mapping(em_tree, hole_em, 1);
4332 write_unlock(&em_tree->lock);
4333 if (err != -EEXIST)
4334 break;
4335 btrfs_drop_extent_cache(inode, cur_offset,
4336 cur_offset +
4337 hole_size - 1, 0);
4338 }
4339 free_extent_map(hole_em);
4340 }
4341 next:
4342 free_extent_map(em);
4343 em = NULL;
4344 cur_offset = last_byte;
4345 if (cur_offset >= block_end)
4346 break;
4347 }
4348 free_extent_map(em);
4349 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4350 GFP_NOFS);
4351 return err;
4352 }
4353
4354 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4355 {
4356 struct btrfs_root *root = BTRFS_I(inode)->root;
4357 struct btrfs_trans_handle *trans;
4358 loff_t oldsize = i_size_read(inode);
4359 loff_t newsize = attr->ia_size;
4360 int mask = attr->ia_valid;
4361 int ret;
4362
4363 /*
4364 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4365 * special case where we need to update the times despite not having
4366 * these flags set. For all other operations the VFS set these flags
4367 * explicitly if it wants a timestamp update.
4368 */
4369 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4370 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4371
4372 if (newsize > oldsize) {
4373 truncate_pagecache(inode, newsize);
4374 ret = btrfs_cont_expand(inode, oldsize, newsize);
4375 if (ret)
4376 return ret;
4377
4378 trans = btrfs_start_transaction(root, 1);
4379 if (IS_ERR(trans))
4380 return PTR_ERR(trans);
4381
4382 i_size_write(inode, newsize);
4383 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4384 ret = btrfs_update_inode(trans, root, inode);
4385 btrfs_end_transaction(trans, root);
4386 } else {
4387
4388 /*
4389 * We're truncating a file that used to have good data down to
4390 * zero. Make sure it gets into the ordered flush list so that
4391 * any new writes get down to disk quickly.
4392 */
4393 if (newsize == 0)
4394 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4395 &BTRFS_I(inode)->runtime_flags);
4396
4397 /*
4398 * 1 for the orphan item we're going to add
4399 * 1 for the orphan item deletion.
4400 */
4401 trans = btrfs_start_transaction(root, 2);
4402 if (IS_ERR(trans))
4403 return PTR_ERR(trans);
4404
4405 /*
4406 * We need to do this in case we fail at _any_ point during the
4407 * actual truncate. Once we do the truncate_setsize we could
4408 * invalidate pages which forces any outstanding ordered io to
4409 * be instantly completed which will give us extents that need
4410 * to be truncated. If we fail to get an orphan inode down we
4411 * could have left over extents that were never meant to live,
4412 * so we need to garuntee from this point on that everything
4413 * will be consistent.
4414 */
4415 ret = btrfs_orphan_add(trans, inode);
4416 btrfs_end_transaction(trans, root);
4417 if (ret)
4418 return ret;
4419
4420 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4421 truncate_setsize(inode, newsize);
4422
4423 /* Disable nonlocked read DIO to avoid the end less truncate */
4424 btrfs_inode_block_unlocked_dio(inode);
4425 inode_dio_wait(inode);
4426 btrfs_inode_resume_unlocked_dio(inode);
4427
4428 ret = btrfs_truncate(inode);
4429 if (ret && inode->i_nlink) {
4430 int err;
4431
4432 /*
4433 * failed to truncate, disk_i_size is only adjusted down
4434 * as we remove extents, so it should represent the true
4435 * size of the inode, so reset the in memory size and
4436 * delete our orphan entry.
4437 */
4438 trans = btrfs_join_transaction(root);
4439 if (IS_ERR(trans)) {
4440 btrfs_orphan_del(NULL, inode);
4441 return ret;
4442 }
4443 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4444 err = btrfs_orphan_del(trans, inode);
4445 if (err)
4446 btrfs_abort_transaction(trans, root, err);
4447 btrfs_end_transaction(trans, root);
4448 }
4449 }
4450
4451 return ret;
4452 }
4453
4454 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4455 {
4456 struct inode *inode = dentry->d_inode;
4457 struct btrfs_root *root = BTRFS_I(inode)->root;
4458 int err;
4459
4460 if (btrfs_root_readonly(root))
4461 return -EROFS;
4462
4463 err = inode_change_ok(inode, attr);
4464 if (err)
4465 return err;
4466
4467 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4468 err = btrfs_setsize(inode, attr);
4469 if (err)
4470 return err;
4471 }
4472
4473 if (attr->ia_valid) {
4474 setattr_copy(inode, attr);
4475 inode_inc_iversion(inode);
4476 err = btrfs_dirty_inode(inode);
4477
4478 if (!err && attr->ia_valid & ATTR_MODE)
4479 err = btrfs_acl_chmod(inode);
4480 }
4481
4482 return err;
4483 }
4484
4485 /*
4486 * While truncating the inode pages during eviction, we get the VFS calling
4487 * btrfs_invalidatepage() against each page of the inode. This is slow because
4488 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4489 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4490 * extent_state structures over and over, wasting lots of time.
4491 *
4492 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4493 * those expensive operations on a per page basis and do only the ordered io
4494 * finishing, while we release here the extent_map and extent_state structures,
4495 * without the excessive merging and splitting.
4496 */
4497 static void evict_inode_truncate_pages(struct inode *inode)
4498 {
4499 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4500 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4501 struct rb_node *node;
4502
4503 ASSERT(inode->i_state & I_FREEING);
4504 truncate_inode_pages(&inode->i_data, 0);
4505
4506 write_lock(&map_tree->lock);
4507 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4508 struct extent_map *em;
4509
4510 node = rb_first(&map_tree->map);
4511 em = rb_entry(node, struct extent_map, rb_node);
4512 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4513 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4514 remove_extent_mapping(map_tree, em);
4515 free_extent_map(em);
4516 }
4517 write_unlock(&map_tree->lock);
4518
4519 spin_lock(&io_tree->lock);
4520 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4521 struct extent_state *state;
4522 struct extent_state *cached_state = NULL;
4523
4524 node = rb_first(&io_tree->state);
4525 state = rb_entry(node, struct extent_state, rb_node);
4526 atomic_inc(&state->refs);
4527 spin_unlock(&io_tree->lock);
4528
4529 lock_extent_bits(io_tree, state->start, state->end,
4530 0, &cached_state);
4531 clear_extent_bit(io_tree, state->start, state->end,
4532 EXTENT_LOCKED | EXTENT_DIRTY |
4533 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4534 EXTENT_DEFRAG, 1, 1,
4535 &cached_state, GFP_NOFS);
4536 free_extent_state(state);
4537
4538 spin_lock(&io_tree->lock);
4539 }
4540 spin_unlock(&io_tree->lock);
4541 }
4542
4543 void btrfs_evict_inode(struct inode *inode)
4544 {
4545 struct btrfs_trans_handle *trans;
4546 struct btrfs_root *root = BTRFS_I(inode)->root;
4547 struct btrfs_block_rsv *rsv, *global_rsv;
4548 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4549 int ret;
4550
4551 trace_btrfs_inode_evict(inode);
4552
4553 evict_inode_truncate_pages(inode);
4554
4555 if (inode->i_nlink &&
4556 ((btrfs_root_refs(&root->root_item) != 0 &&
4557 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4558 btrfs_is_free_space_inode(inode)))
4559 goto no_delete;
4560
4561 if (is_bad_inode(inode)) {
4562 btrfs_orphan_del(NULL, inode);
4563 goto no_delete;
4564 }
4565 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4566 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4567
4568 if (root->fs_info->log_root_recovering) {
4569 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4570 &BTRFS_I(inode)->runtime_flags));
4571 goto no_delete;
4572 }
4573
4574 if (inode->i_nlink > 0) {
4575 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4576 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4577 goto no_delete;
4578 }
4579
4580 ret = btrfs_commit_inode_delayed_inode(inode);
4581 if (ret) {
4582 btrfs_orphan_del(NULL, inode);
4583 goto no_delete;
4584 }
4585
4586 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4587 if (!rsv) {
4588 btrfs_orphan_del(NULL, inode);
4589 goto no_delete;
4590 }
4591 rsv->size = min_size;
4592 rsv->failfast = 1;
4593 global_rsv = &root->fs_info->global_block_rsv;
4594
4595 btrfs_i_size_write(inode, 0);
4596
4597 /*
4598 * This is a bit simpler than btrfs_truncate since we've already
4599 * reserved our space for our orphan item in the unlink, so we just
4600 * need to reserve some slack space in case we add bytes and update
4601 * inode item when doing the truncate.
4602 */
4603 while (1) {
4604 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4605 BTRFS_RESERVE_FLUSH_LIMIT);
4606
4607 /*
4608 * Try and steal from the global reserve since we will
4609 * likely not use this space anyway, we want to try as
4610 * hard as possible to get this to work.
4611 */
4612 if (ret)
4613 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4614
4615 if (ret) {
4616 btrfs_warn(root->fs_info,
4617 "Could not get space for a delete, will truncate on mount %d",
4618 ret);
4619 btrfs_orphan_del(NULL, inode);
4620 btrfs_free_block_rsv(root, rsv);
4621 goto no_delete;
4622 }
4623
4624 trans = btrfs_join_transaction(root);
4625 if (IS_ERR(trans)) {
4626 btrfs_orphan_del(NULL, inode);
4627 btrfs_free_block_rsv(root, rsv);
4628 goto no_delete;
4629 }
4630
4631 trans->block_rsv = rsv;
4632
4633 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4634 if (ret != -ENOSPC)
4635 break;
4636
4637 trans->block_rsv = &root->fs_info->trans_block_rsv;
4638 btrfs_end_transaction(trans, root);
4639 trans = NULL;
4640 btrfs_btree_balance_dirty(root);
4641 }
4642
4643 btrfs_free_block_rsv(root, rsv);
4644
4645 /*
4646 * Errors here aren't a big deal, it just means we leave orphan items
4647 * in the tree. They will be cleaned up on the next mount.
4648 */
4649 if (ret == 0) {
4650 trans->block_rsv = root->orphan_block_rsv;
4651 btrfs_orphan_del(trans, inode);
4652 } else {
4653 btrfs_orphan_del(NULL, inode);
4654 }
4655
4656 trans->block_rsv = &root->fs_info->trans_block_rsv;
4657 if (!(root == root->fs_info->tree_root ||
4658 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4659 btrfs_return_ino(root, btrfs_ino(inode));
4660
4661 btrfs_end_transaction(trans, root);
4662 btrfs_btree_balance_dirty(root);
4663 no_delete:
4664 btrfs_remove_delayed_node(inode);
4665 clear_inode(inode);
4666 return;
4667 }
4668
4669 /*
4670 * this returns the key found in the dir entry in the location pointer.
4671 * If no dir entries were found, location->objectid is 0.
4672 */
4673 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4674 struct btrfs_key *location)
4675 {
4676 const char *name = dentry->d_name.name;
4677 int namelen = dentry->d_name.len;
4678 struct btrfs_dir_item *di;
4679 struct btrfs_path *path;
4680 struct btrfs_root *root = BTRFS_I(dir)->root;
4681 int ret = 0;
4682
4683 path = btrfs_alloc_path();
4684 if (!path)
4685 return -ENOMEM;
4686
4687 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4688 namelen, 0);
4689 if (IS_ERR(di))
4690 ret = PTR_ERR(di);
4691
4692 if (IS_ERR_OR_NULL(di))
4693 goto out_err;
4694
4695 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4696 out:
4697 btrfs_free_path(path);
4698 return ret;
4699 out_err:
4700 location->objectid = 0;
4701 goto out;
4702 }
4703
4704 /*
4705 * when we hit a tree root in a directory, the btrfs part of the inode
4706 * needs to be changed to reflect the root directory of the tree root. This
4707 * is kind of like crossing a mount point.
4708 */
4709 static int fixup_tree_root_location(struct btrfs_root *root,
4710 struct inode *dir,
4711 struct dentry *dentry,
4712 struct btrfs_key *location,
4713 struct btrfs_root **sub_root)
4714 {
4715 struct btrfs_path *path;
4716 struct btrfs_root *new_root;
4717 struct btrfs_root_ref *ref;
4718 struct extent_buffer *leaf;
4719 int ret;
4720 int err = 0;
4721
4722 path = btrfs_alloc_path();
4723 if (!path) {
4724 err = -ENOMEM;
4725 goto out;
4726 }
4727
4728 err = -ENOENT;
4729 ret = btrfs_find_item(root->fs_info->tree_root, path,
4730 BTRFS_I(dir)->root->root_key.objectid,
4731 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4732 if (ret) {
4733 if (ret < 0)
4734 err = ret;
4735 goto out;
4736 }
4737
4738 leaf = path->nodes[0];
4739 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4740 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4741 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4742 goto out;
4743
4744 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4745 (unsigned long)(ref + 1),
4746 dentry->d_name.len);
4747 if (ret)
4748 goto out;
4749
4750 btrfs_release_path(path);
4751
4752 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4753 if (IS_ERR(new_root)) {
4754 err = PTR_ERR(new_root);
4755 goto out;
4756 }
4757
4758 *sub_root = new_root;
4759 location->objectid = btrfs_root_dirid(&new_root->root_item);
4760 location->type = BTRFS_INODE_ITEM_KEY;
4761 location->offset = 0;
4762 err = 0;
4763 out:
4764 btrfs_free_path(path);
4765 return err;
4766 }
4767
4768 static void inode_tree_add(struct inode *inode)
4769 {
4770 struct btrfs_root *root = BTRFS_I(inode)->root;
4771 struct btrfs_inode *entry;
4772 struct rb_node **p;
4773 struct rb_node *parent;
4774 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4775 u64 ino = btrfs_ino(inode);
4776
4777 if (inode_unhashed(inode))
4778 return;
4779 parent = NULL;
4780 spin_lock(&root->inode_lock);
4781 p = &root->inode_tree.rb_node;
4782 while (*p) {
4783 parent = *p;
4784 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4785
4786 if (ino < btrfs_ino(&entry->vfs_inode))
4787 p = &parent->rb_left;
4788 else if (ino > btrfs_ino(&entry->vfs_inode))
4789 p = &parent->rb_right;
4790 else {
4791 WARN_ON(!(entry->vfs_inode.i_state &
4792 (I_WILL_FREE | I_FREEING)));
4793 rb_replace_node(parent, new, &root->inode_tree);
4794 RB_CLEAR_NODE(parent);
4795 spin_unlock(&root->inode_lock);
4796 return;
4797 }
4798 }
4799 rb_link_node(new, parent, p);
4800 rb_insert_color(new, &root->inode_tree);
4801 spin_unlock(&root->inode_lock);
4802 }
4803
4804 static void inode_tree_del(struct inode *inode)
4805 {
4806 struct btrfs_root *root = BTRFS_I(inode)->root;
4807 int empty = 0;
4808
4809 spin_lock(&root->inode_lock);
4810 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4811 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4812 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4813 empty = RB_EMPTY_ROOT(&root->inode_tree);
4814 }
4815 spin_unlock(&root->inode_lock);
4816
4817 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4818 synchronize_srcu(&root->fs_info->subvol_srcu);
4819 spin_lock(&root->inode_lock);
4820 empty = RB_EMPTY_ROOT(&root->inode_tree);
4821 spin_unlock(&root->inode_lock);
4822 if (empty)
4823 btrfs_add_dead_root(root);
4824 }
4825 }
4826
4827 void btrfs_invalidate_inodes(struct btrfs_root *root)
4828 {
4829 struct rb_node *node;
4830 struct rb_node *prev;
4831 struct btrfs_inode *entry;
4832 struct inode *inode;
4833 u64 objectid = 0;
4834
4835 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4836
4837 spin_lock(&root->inode_lock);
4838 again:
4839 node = root->inode_tree.rb_node;
4840 prev = NULL;
4841 while (node) {
4842 prev = node;
4843 entry = rb_entry(node, struct btrfs_inode, rb_node);
4844
4845 if (objectid < btrfs_ino(&entry->vfs_inode))
4846 node = node->rb_left;
4847 else if (objectid > btrfs_ino(&entry->vfs_inode))
4848 node = node->rb_right;
4849 else
4850 break;
4851 }
4852 if (!node) {
4853 while (prev) {
4854 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4855 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4856 node = prev;
4857 break;
4858 }
4859 prev = rb_next(prev);
4860 }
4861 }
4862 while (node) {
4863 entry = rb_entry(node, struct btrfs_inode, rb_node);
4864 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4865 inode = igrab(&entry->vfs_inode);
4866 if (inode) {
4867 spin_unlock(&root->inode_lock);
4868 if (atomic_read(&inode->i_count) > 1)
4869 d_prune_aliases(inode);
4870 /*
4871 * btrfs_drop_inode will have it removed from
4872 * the inode cache when its usage count
4873 * hits zero.
4874 */
4875 iput(inode);
4876 cond_resched();
4877 spin_lock(&root->inode_lock);
4878 goto again;
4879 }
4880
4881 if (cond_resched_lock(&root->inode_lock))
4882 goto again;
4883
4884 node = rb_next(node);
4885 }
4886 spin_unlock(&root->inode_lock);
4887 }
4888
4889 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4890 {
4891 struct btrfs_iget_args *args = p;
4892 inode->i_ino = args->ino;
4893 BTRFS_I(inode)->root = args->root;
4894 return 0;
4895 }
4896
4897 static int btrfs_find_actor(struct inode *inode, void *opaque)
4898 {
4899 struct btrfs_iget_args *args = opaque;
4900 return args->ino == btrfs_ino(inode) &&
4901 args->root == BTRFS_I(inode)->root;
4902 }
4903
4904 static struct inode *btrfs_iget_locked(struct super_block *s,
4905 u64 objectid,
4906 struct btrfs_root *root)
4907 {
4908 struct inode *inode;
4909 struct btrfs_iget_args args;
4910 unsigned long hashval = btrfs_inode_hash(objectid, root);
4911
4912 args.ino = objectid;
4913 args.root = root;
4914
4915 inode = iget5_locked(s, hashval, btrfs_find_actor,
4916 btrfs_init_locked_inode,
4917 (void *)&args);
4918 return inode;
4919 }
4920
4921 /* Get an inode object given its location and corresponding root.
4922 * Returns in *is_new if the inode was read from disk
4923 */
4924 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4925 struct btrfs_root *root, int *new)
4926 {
4927 struct inode *inode;
4928
4929 inode = btrfs_iget_locked(s, location->objectid, root);
4930 if (!inode)
4931 return ERR_PTR(-ENOMEM);
4932
4933 if (inode->i_state & I_NEW) {
4934 BTRFS_I(inode)->root = root;
4935 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4936 btrfs_read_locked_inode(inode);
4937 if (!is_bad_inode(inode)) {
4938 inode_tree_add(inode);
4939 unlock_new_inode(inode);
4940 if (new)
4941 *new = 1;
4942 } else {
4943 unlock_new_inode(inode);
4944 iput(inode);
4945 inode = ERR_PTR(-ESTALE);
4946 }
4947 }
4948
4949 return inode;
4950 }
4951
4952 static struct inode *new_simple_dir(struct super_block *s,
4953 struct btrfs_key *key,
4954 struct btrfs_root *root)
4955 {
4956 struct inode *inode = new_inode(s);
4957
4958 if (!inode)
4959 return ERR_PTR(-ENOMEM);
4960
4961 BTRFS_I(inode)->root = root;
4962 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4963 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4964
4965 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4966 inode->i_op = &btrfs_dir_ro_inode_operations;
4967 inode->i_fop = &simple_dir_operations;
4968 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4969 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4970
4971 return inode;
4972 }
4973
4974 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4975 {
4976 struct inode *inode;
4977 struct btrfs_root *root = BTRFS_I(dir)->root;
4978 struct btrfs_root *sub_root = root;
4979 struct btrfs_key location;
4980 int index;
4981 int ret = 0;
4982
4983 if (dentry->d_name.len > BTRFS_NAME_LEN)
4984 return ERR_PTR(-ENAMETOOLONG);
4985
4986 ret = btrfs_inode_by_name(dir, dentry, &location);
4987 if (ret < 0)
4988 return ERR_PTR(ret);
4989
4990 if (location.objectid == 0)
4991 return ERR_PTR(-ENOENT);
4992
4993 if (location.type == BTRFS_INODE_ITEM_KEY) {
4994 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4995 return inode;
4996 }
4997
4998 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4999
5000 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5001 ret = fixup_tree_root_location(root, dir, dentry,
5002 &location, &sub_root);
5003 if (ret < 0) {
5004 if (ret != -ENOENT)
5005 inode = ERR_PTR(ret);
5006 else
5007 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5008 } else {
5009 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5010 }
5011 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5012
5013 if (!IS_ERR(inode) && root != sub_root) {
5014 down_read(&root->fs_info->cleanup_work_sem);
5015 if (!(inode->i_sb->s_flags & MS_RDONLY))
5016 ret = btrfs_orphan_cleanup(sub_root);
5017 up_read(&root->fs_info->cleanup_work_sem);
5018 if (ret) {
5019 iput(inode);
5020 inode = ERR_PTR(ret);
5021 }
5022 }
5023
5024 return inode;
5025 }
5026
5027 static int btrfs_dentry_delete(const struct dentry *dentry)
5028 {
5029 struct btrfs_root *root;
5030 struct inode *inode = dentry->d_inode;
5031
5032 if (!inode && !IS_ROOT(dentry))
5033 inode = dentry->d_parent->d_inode;
5034
5035 if (inode) {
5036 root = BTRFS_I(inode)->root;
5037 if (btrfs_root_refs(&root->root_item) == 0)
5038 return 1;
5039
5040 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5041 return 1;
5042 }
5043 return 0;
5044 }
5045
5046 static void btrfs_dentry_release(struct dentry *dentry)
5047 {
5048 if (dentry->d_fsdata)
5049 kfree(dentry->d_fsdata);
5050 }
5051
5052 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5053 unsigned int flags)
5054 {
5055 struct inode *inode;
5056
5057 inode = btrfs_lookup_dentry(dir, dentry);
5058 if (IS_ERR(inode)) {
5059 if (PTR_ERR(inode) == -ENOENT)
5060 inode = NULL;
5061 else
5062 return ERR_CAST(inode);
5063 }
5064
5065 return d_splice_alias(inode, dentry);
5066 }
5067
5068 unsigned char btrfs_filetype_table[] = {
5069 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5070 };
5071
5072 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5073 {
5074 struct inode *inode = file_inode(file);
5075 struct btrfs_root *root = BTRFS_I(inode)->root;
5076 struct btrfs_item *item;
5077 struct btrfs_dir_item *di;
5078 struct btrfs_key key;
5079 struct btrfs_key found_key;
5080 struct btrfs_path *path;
5081 struct list_head ins_list;
5082 struct list_head del_list;
5083 int ret;
5084 struct extent_buffer *leaf;
5085 int slot;
5086 unsigned char d_type;
5087 int over = 0;
5088 u32 di_cur;
5089 u32 di_total;
5090 u32 di_len;
5091 int key_type = BTRFS_DIR_INDEX_KEY;
5092 char tmp_name[32];
5093 char *name_ptr;
5094 int name_len;
5095 int is_curr = 0; /* ctx->pos points to the current index? */
5096
5097 /* FIXME, use a real flag for deciding about the key type */
5098 if (root->fs_info->tree_root == root)
5099 key_type = BTRFS_DIR_ITEM_KEY;
5100
5101 if (!dir_emit_dots(file, ctx))
5102 return 0;
5103
5104 path = btrfs_alloc_path();
5105 if (!path)
5106 return -ENOMEM;
5107
5108 path->reada = 1;
5109
5110 if (key_type == BTRFS_DIR_INDEX_KEY) {
5111 INIT_LIST_HEAD(&ins_list);
5112 INIT_LIST_HEAD(&del_list);
5113 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5114 }
5115
5116 btrfs_set_key_type(&key, key_type);
5117 key.offset = ctx->pos;
5118 key.objectid = btrfs_ino(inode);
5119
5120 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5121 if (ret < 0)
5122 goto err;
5123
5124 while (1) {
5125 leaf = path->nodes[0];
5126 slot = path->slots[0];
5127 if (slot >= btrfs_header_nritems(leaf)) {
5128 ret = btrfs_next_leaf(root, path);
5129 if (ret < 0)
5130 goto err;
5131 else if (ret > 0)
5132 break;
5133 continue;
5134 }
5135
5136 item = btrfs_item_nr(slot);
5137 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5138
5139 if (found_key.objectid != key.objectid)
5140 break;
5141 if (btrfs_key_type(&found_key) != key_type)
5142 break;
5143 if (found_key.offset < ctx->pos)
5144 goto next;
5145 if (key_type == BTRFS_DIR_INDEX_KEY &&
5146 btrfs_should_delete_dir_index(&del_list,
5147 found_key.offset))
5148 goto next;
5149
5150 ctx->pos = found_key.offset;
5151 is_curr = 1;
5152
5153 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5154 di_cur = 0;
5155 di_total = btrfs_item_size(leaf, item);
5156
5157 while (di_cur < di_total) {
5158 struct btrfs_key location;
5159
5160 if (verify_dir_item(root, leaf, di))
5161 break;
5162
5163 name_len = btrfs_dir_name_len(leaf, di);
5164 if (name_len <= sizeof(tmp_name)) {
5165 name_ptr = tmp_name;
5166 } else {
5167 name_ptr = kmalloc(name_len, GFP_NOFS);
5168 if (!name_ptr) {
5169 ret = -ENOMEM;
5170 goto err;
5171 }
5172 }
5173 read_extent_buffer(leaf, name_ptr,
5174 (unsigned long)(di + 1), name_len);
5175
5176 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5177 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5178
5179
5180 /* is this a reference to our own snapshot? If so
5181 * skip it.
5182 *
5183 * In contrast to old kernels, we insert the snapshot's
5184 * dir item and dir index after it has been created, so
5185 * we won't find a reference to our own snapshot. We
5186 * still keep the following code for backward
5187 * compatibility.
5188 */
5189 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5190 location.objectid == root->root_key.objectid) {
5191 over = 0;
5192 goto skip;
5193 }
5194 over = !dir_emit(ctx, name_ptr, name_len,
5195 location.objectid, d_type);
5196
5197 skip:
5198 if (name_ptr != tmp_name)
5199 kfree(name_ptr);
5200
5201 if (over)
5202 goto nopos;
5203 di_len = btrfs_dir_name_len(leaf, di) +
5204 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5205 di_cur += di_len;
5206 di = (struct btrfs_dir_item *)((char *)di + di_len);
5207 }
5208 next:
5209 path->slots[0]++;
5210 }
5211
5212 if (key_type == BTRFS_DIR_INDEX_KEY) {
5213 if (is_curr)
5214 ctx->pos++;
5215 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5216 if (ret)
5217 goto nopos;
5218 }
5219
5220 /* Reached end of directory/root. Bump pos past the last item. */
5221 ctx->pos++;
5222
5223 /*
5224 * Stop new entries from being returned after we return the last
5225 * entry.
5226 *
5227 * New directory entries are assigned a strictly increasing
5228 * offset. This means that new entries created during readdir
5229 * are *guaranteed* to be seen in the future by that readdir.
5230 * This has broken buggy programs which operate on names as
5231 * they're returned by readdir. Until we re-use freed offsets
5232 * we have this hack to stop new entries from being returned
5233 * under the assumption that they'll never reach this huge
5234 * offset.
5235 *
5236 * This is being careful not to overflow 32bit loff_t unless the
5237 * last entry requires it because doing so has broken 32bit apps
5238 * in the past.
5239 */
5240 if (key_type == BTRFS_DIR_INDEX_KEY) {
5241 if (ctx->pos >= INT_MAX)
5242 ctx->pos = LLONG_MAX;
5243 else
5244 ctx->pos = INT_MAX;
5245 }
5246 nopos:
5247 ret = 0;
5248 err:
5249 if (key_type == BTRFS_DIR_INDEX_KEY)
5250 btrfs_put_delayed_items(&ins_list, &del_list);
5251 btrfs_free_path(path);
5252 return ret;
5253 }
5254
5255 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5256 {
5257 struct btrfs_root *root = BTRFS_I(inode)->root;
5258 struct btrfs_trans_handle *trans;
5259 int ret = 0;
5260 bool nolock = false;
5261
5262 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5263 return 0;
5264
5265 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5266 nolock = true;
5267
5268 if (wbc->sync_mode == WB_SYNC_ALL) {
5269 if (nolock)
5270 trans = btrfs_join_transaction_nolock(root);
5271 else
5272 trans = btrfs_join_transaction(root);
5273 if (IS_ERR(trans))
5274 return PTR_ERR(trans);
5275 ret = btrfs_commit_transaction(trans, root);
5276 }
5277 return ret;
5278 }
5279
5280 /*
5281 * This is somewhat expensive, updating the tree every time the
5282 * inode changes. But, it is most likely to find the inode in cache.
5283 * FIXME, needs more benchmarking...there are no reasons other than performance
5284 * to keep or drop this code.
5285 */
5286 static int btrfs_dirty_inode(struct inode *inode)
5287 {
5288 struct btrfs_root *root = BTRFS_I(inode)->root;
5289 struct btrfs_trans_handle *trans;
5290 int ret;
5291
5292 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5293 return 0;
5294
5295 trans = btrfs_join_transaction(root);
5296 if (IS_ERR(trans))
5297 return PTR_ERR(trans);
5298
5299 ret = btrfs_update_inode(trans, root, inode);
5300 if (ret && ret == -ENOSPC) {
5301 /* whoops, lets try again with the full transaction */
5302 btrfs_end_transaction(trans, root);
5303 trans = btrfs_start_transaction(root, 1);
5304 if (IS_ERR(trans))
5305 return PTR_ERR(trans);
5306
5307 ret = btrfs_update_inode(trans, root, inode);
5308 }
5309 btrfs_end_transaction(trans, root);
5310 if (BTRFS_I(inode)->delayed_node)
5311 btrfs_balance_delayed_items(root);
5312
5313 return ret;
5314 }
5315
5316 /*
5317 * This is a copy of file_update_time. We need this so we can return error on
5318 * ENOSPC for updating the inode in the case of file write and mmap writes.
5319 */
5320 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5321 int flags)
5322 {
5323 struct btrfs_root *root = BTRFS_I(inode)->root;
5324
5325 if (btrfs_root_readonly(root))
5326 return -EROFS;
5327
5328 if (flags & S_VERSION)
5329 inode_inc_iversion(inode);
5330 if (flags & S_CTIME)
5331 inode->i_ctime = *now;
5332 if (flags & S_MTIME)
5333 inode->i_mtime = *now;
5334 if (flags & S_ATIME)
5335 inode->i_atime = *now;
5336 return btrfs_dirty_inode(inode);
5337 }
5338
5339 /*
5340 * find the highest existing sequence number in a directory
5341 * and then set the in-memory index_cnt variable to reflect
5342 * free sequence numbers
5343 */
5344 static int btrfs_set_inode_index_count(struct inode *inode)
5345 {
5346 struct btrfs_root *root = BTRFS_I(inode)->root;
5347 struct btrfs_key key, found_key;
5348 struct btrfs_path *path;
5349 struct extent_buffer *leaf;
5350 int ret;
5351
5352 key.objectid = btrfs_ino(inode);
5353 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5354 key.offset = (u64)-1;
5355
5356 path = btrfs_alloc_path();
5357 if (!path)
5358 return -ENOMEM;
5359
5360 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5361 if (ret < 0)
5362 goto out;
5363 /* FIXME: we should be able to handle this */
5364 if (ret == 0)
5365 goto out;
5366 ret = 0;
5367
5368 /*
5369 * MAGIC NUMBER EXPLANATION:
5370 * since we search a directory based on f_pos we have to start at 2
5371 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5372 * else has to start at 2
5373 */
5374 if (path->slots[0] == 0) {
5375 BTRFS_I(inode)->index_cnt = 2;
5376 goto out;
5377 }
5378
5379 path->slots[0]--;
5380
5381 leaf = path->nodes[0];
5382 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5383
5384 if (found_key.objectid != btrfs_ino(inode) ||
5385 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5386 BTRFS_I(inode)->index_cnt = 2;
5387 goto out;
5388 }
5389
5390 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5391 out:
5392 btrfs_free_path(path);
5393 return ret;
5394 }
5395
5396 /*
5397 * helper to find a free sequence number in a given directory. This current
5398 * code is very simple, later versions will do smarter things in the btree
5399 */
5400 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5401 {
5402 int ret = 0;
5403
5404 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5405 ret = btrfs_inode_delayed_dir_index_count(dir);
5406 if (ret) {
5407 ret = btrfs_set_inode_index_count(dir);
5408 if (ret)
5409 return ret;
5410 }
5411 }
5412
5413 *index = BTRFS_I(dir)->index_cnt;
5414 BTRFS_I(dir)->index_cnt++;
5415
5416 return ret;
5417 }
5418
5419 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5420 struct btrfs_root *root,
5421 struct inode *dir,
5422 const char *name, int name_len,
5423 u64 ref_objectid, u64 objectid,
5424 umode_t mode, u64 *index)
5425 {
5426 struct inode *inode;
5427 struct btrfs_inode_item *inode_item;
5428 struct btrfs_key *location;
5429 struct btrfs_path *path;
5430 struct btrfs_inode_ref *ref;
5431 struct btrfs_key key[2];
5432 u32 sizes[2];
5433 unsigned long ptr;
5434 int ret;
5435
5436 path = btrfs_alloc_path();
5437 if (!path)
5438 return ERR_PTR(-ENOMEM);
5439
5440 inode = new_inode(root->fs_info->sb);
5441 if (!inode) {
5442 btrfs_free_path(path);
5443 return ERR_PTR(-ENOMEM);
5444 }
5445
5446 /*
5447 * we have to initialize this early, so we can reclaim the inode
5448 * number if we fail afterwards in this function.
5449 */
5450 inode->i_ino = objectid;
5451
5452 if (dir) {
5453 trace_btrfs_inode_request(dir);
5454
5455 ret = btrfs_set_inode_index(dir, index);
5456 if (ret) {
5457 btrfs_free_path(path);
5458 iput(inode);
5459 return ERR_PTR(ret);
5460 }
5461 }
5462 /*
5463 * index_cnt is ignored for everything but a dir,
5464 * btrfs_get_inode_index_count has an explanation for the magic
5465 * number
5466 */
5467 BTRFS_I(inode)->index_cnt = 2;
5468 BTRFS_I(inode)->root = root;
5469 BTRFS_I(inode)->generation = trans->transid;
5470 inode->i_generation = BTRFS_I(inode)->generation;
5471
5472 /*
5473 * We could have gotten an inode number from somebody who was fsynced
5474 * and then removed in this same transaction, so let's just set full
5475 * sync since it will be a full sync anyway and this will blow away the
5476 * old info in the log.
5477 */
5478 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5479
5480 key[0].objectid = objectid;
5481 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5482 key[0].offset = 0;
5483
5484 /*
5485 * Start new inodes with an inode_ref. This is slightly more
5486 * efficient for small numbers of hard links since they will
5487 * be packed into one item. Extended refs will kick in if we
5488 * add more hard links than can fit in the ref item.
5489 */
5490 key[1].objectid = objectid;
5491 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5492 key[1].offset = ref_objectid;
5493
5494 sizes[0] = sizeof(struct btrfs_inode_item);
5495 sizes[1] = name_len + sizeof(*ref);
5496
5497 path->leave_spinning = 1;
5498 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5499 if (ret != 0)
5500 goto fail;
5501
5502 inode_init_owner(inode, dir, mode);
5503 inode_set_bytes(inode, 0);
5504 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5505 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5506 struct btrfs_inode_item);
5507 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5508 sizeof(*inode_item));
5509 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5510
5511 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5512 struct btrfs_inode_ref);
5513 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5514 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5515 ptr = (unsigned long)(ref + 1);
5516 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5517
5518 btrfs_mark_buffer_dirty(path->nodes[0]);
5519 btrfs_free_path(path);
5520
5521 location = &BTRFS_I(inode)->location;
5522 location->objectid = objectid;
5523 location->offset = 0;
5524 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5525
5526 btrfs_inherit_iflags(inode, dir);
5527
5528 if (S_ISREG(mode)) {
5529 if (btrfs_test_opt(root, NODATASUM))
5530 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5531 if (btrfs_test_opt(root, NODATACOW))
5532 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5533 BTRFS_INODE_NODATASUM;
5534 }
5535
5536 btrfs_insert_inode_hash(inode);
5537 inode_tree_add(inode);
5538
5539 trace_btrfs_inode_new(inode);
5540 btrfs_set_inode_last_trans(trans, inode);
5541
5542 btrfs_update_root_times(trans, root);
5543
5544 return inode;
5545 fail:
5546 if (dir)
5547 BTRFS_I(dir)->index_cnt--;
5548 btrfs_free_path(path);
5549 iput(inode);
5550 return ERR_PTR(ret);
5551 }
5552
5553 static inline u8 btrfs_inode_type(struct inode *inode)
5554 {
5555 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5556 }
5557
5558 /*
5559 * utility function to add 'inode' into 'parent_inode' with
5560 * a give name and a given sequence number.
5561 * if 'add_backref' is true, also insert a backref from the
5562 * inode to the parent directory.
5563 */
5564 int btrfs_add_link(struct btrfs_trans_handle *trans,
5565 struct inode *parent_inode, struct inode *inode,
5566 const char *name, int name_len, int add_backref, u64 index)
5567 {
5568 int ret = 0;
5569 struct btrfs_key key;
5570 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5571 u64 ino = btrfs_ino(inode);
5572 u64 parent_ino = btrfs_ino(parent_inode);
5573
5574 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5575 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5576 } else {
5577 key.objectid = ino;
5578 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5579 key.offset = 0;
5580 }
5581
5582 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5583 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5584 key.objectid, root->root_key.objectid,
5585 parent_ino, index, name, name_len);
5586 } else if (add_backref) {
5587 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5588 parent_ino, index);
5589 }
5590
5591 /* Nothing to clean up yet */
5592 if (ret)
5593 return ret;
5594
5595 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5596 parent_inode, &key,
5597 btrfs_inode_type(inode), index);
5598 if (ret == -EEXIST || ret == -EOVERFLOW)
5599 goto fail_dir_item;
5600 else if (ret) {
5601 btrfs_abort_transaction(trans, root, ret);
5602 return ret;
5603 }
5604
5605 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5606 name_len * 2);
5607 inode_inc_iversion(parent_inode);
5608 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5609 ret = btrfs_update_inode(trans, root, parent_inode);
5610 if (ret)
5611 btrfs_abort_transaction(trans, root, ret);
5612 return ret;
5613
5614 fail_dir_item:
5615 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5616 u64 local_index;
5617 int err;
5618 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5619 key.objectid, root->root_key.objectid,
5620 parent_ino, &local_index, name, name_len);
5621
5622 } else if (add_backref) {
5623 u64 local_index;
5624 int err;
5625
5626 err = btrfs_del_inode_ref(trans, root, name, name_len,
5627 ino, parent_ino, &local_index);
5628 }
5629 return ret;
5630 }
5631
5632 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5633 struct inode *dir, struct dentry *dentry,
5634 struct inode *inode, int backref, u64 index)
5635 {
5636 int err = btrfs_add_link(trans, dir, inode,
5637 dentry->d_name.name, dentry->d_name.len,
5638 backref, index);
5639 if (err > 0)
5640 err = -EEXIST;
5641 return err;
5642 }
5643
5644 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5645 umode_t mode, dev_t rdev)
5646 {
5647 struct btrfs_trans_handle *trans;
5648 struct btrfs_root *root = BTRFS_I(dir)->root;
5649 struct inode *inode = NULL;
5650 int err;
5651 int drop_inode = 0;
5652 u64 objectid;
5653 u64 index = 0;
5654
5655 if (!new_valid_dev(rdev))
5656 return -EINVAL;
5657
5658 /*
5659 * 2 for inode item and ref
5660 * 2 for dir items
5661 * 1 for xattr if selinux is on
5662 */
5663 trans = btrfs_start_transaction(root, 5);
5664 if (IS_ERR(trans))
5665 return PTR_ERR(trans);
5666
5667 err = btrfs_find_free_ino(root, &objectid);
5668 if (err)
5669 goto out_unlock;
5670
5671 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5672 dentry->d_name.len, btrfs_ino(dir), objectid,
5673 mode, &index);
5674 if (IS_ERR(inode)) {
5675 err = PTR_ERR(inode);
5676 goto out_unlock;
5677 }
5678
5679 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5680 if (err) {
5681 drop_inode = 1;
5682 goto out_unlock;
5683 }
5684
5685 /*
5686 * If the active LSM wants to access the inode during
5687 * d_instantiate it needs these. Smack checks to see
5688 * if the filesystem supports xattrs by looking at the
5689 * ops vector.
5690 */
5691
5692 inode->i_op = &btrfs_special_inode_operations;
5693 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5694 if (err)
5695 drop_inode = 1;
5696 else {
5697 init_special_inode(inode, inode->i_mode, rdev);
5698 btrfs_update_inode(trans, root, inode);
5699 d_instantiate(dentry, inode);
5700 }
5701 out_unlock:
5702 btrfs_end_transaction(trans, root);
5703 btrfs_btree_balance_dirty(root);
5704 if (drop_inode) {
5705 inode_dec_link_count(inode);
5706 iput(inode);
5707 }
5708 return err;
5709 }
5710
5711 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5712 umode_t mode, bool excl)
5713 {
5714 struct btrfs_trans_handle *trans;
5715 struct btrfs_root *root = BTRFS_I(dir)->root;
5716 struct inode *inode = NULL;
5717 int drop_inode_on_err = 0;
5718 int err;
5719 u64 objectid;
5720 u64 index = 0;
5721
5722 /*
5723 * 2 for inode item and ref
5724 * 2 for dir items
5725 * 1 for xattr if selinux is on
5726 */
5727 trans = btrfs_start_transaction(root, 5);
5728 if (IS_ERR(trans))
5729 return PTR_ERR(trans);
5730
5731 err = btrfs_find_free_ino(root, &objectid);
5732 if (err)
5733 goto out_unlock;
5734
5735 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5736 dentry->d_name.len, btrfs_ino(dir), objectid,
5737 mode, &index);
5738 if (IS_ERR(inode)) {
5739 err = PTR_ERR(inode);
5740 goto out_unlock;
5741 }
5742 drop_inode_on_err = 1;
5743
5744 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5745 if (err)
5746 goto out_unlock;
5747
5748 err = btrfs_update_inode(trans, root, inode);
5749 if (err)
5750 goto out_unlock;
5751
5752 /*
5753 * If the active LSM wants to access the inode during
5754 * d_instantiate it needs these. Smack checks to see
5755 * if the filesystem supports xattrs by looking at the
5756 * ops vector.
5757 */
5758 inode->i_fop = &btrfs_file_operations;
5759 inode->i_op = &btrfs_file_inode_operations;
5760
5761 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5762 if (err)
5763 goto out_unlock;
5764
5765 inode->i_mapping->a_ops = &btrfs_aops;
5766 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5767 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5768 d_instantiate(dentry, inode);
5769
5770 out_unlock:
5771 btrfs_end_transaction(trans, root);
5772 if (err && drop_inode_on_err) {
5773 inode_dec_link_count(inode);
5774 iput(inode);
5775 }
5776 btrfs_btree_balance_dirty(root);
5777 return err;
5778 }
5779
5780 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5781 struct dentry *dentry)
5782 {
5783 struct btrfs_trans_handle *trans;
5784 struct btrfs_root *root = BTRFS_I(dir)->root;
5785 struct inode *inode = old_dentry->d_inode;
5786 u64 index;
5787 int err;
5788 int drop_inode = 0;
5789
5790 /* do not allow sys_link's with other subvols of the same device */
5791 if (root->objectid != BTRFS_I(inode)->root->objectid)
5792 return -EXDEV;
5793
5794 if (inode->i_nlink >= BTRFS_LINK_MAX)
5795 return -EMLINK;
5796
5797 err = btrfs_set_inode_index(dir, &index);
5798 if (err)
5799 goto fail;
5800
5801 /*
5802 * 2 items for inode and inode ref
5803 * 2 items for dir items
5804 * 1 item for parent inode
5805 */
5806 trans = btrfs_start_transaction(root, 5);
5807 if (IS_ERR(trans)) {
5808 err = PTR_ERR(trans);
5809 goto fail;
5810 }
5811
5812 inc_nlink(inode);
5813 inode_inc_iversion(inode);
5814 inode->i_ctime = CURRENT_TIME;
5815 ihold(inode);
5816 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5817
5818 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5819
5820 if (err) {
5821 drop_inode = 1;
5822 } else {
5823 struct dentry *parent = dentry->d_parent;
5824 err = btrfs_update_inode(trans, root, inode);
5825 if (err)
5826 goto fail;
5827 d_instantiate(dentry, inode);
5828 btrfs_log_new_name(trans, inode, NULL, parent);
5829 }
5830
5831 btrfs_end_transaction(trans, root);
5832 fail:
5833 if (drop_inode) {
5834 inode_dec_link_count(inode);
5835 iput(inode);
5836 }
5837 btrfs_btree_balance_dirty(root);
5838 return err;
5839 }
5840
5841 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5842 {
5843 struct inode *inode = NULL;
5844 struct btrfs_trans_handle *trans;
5845 struct btrfs_root *root = BTRFS_I(dir)->root;
5846 int err = 0;
5847 int drop_on_err = 0;
5848 u64 objectid = 0;
5849 u64 index = 0;
5850
5851 /*
5852 * 2 items for inode and ref
5853 * 2 items for dir items
5854 * 1 for xattr if selinux is on
5855 */
5856 trans = btrfs_start_transaction(root, 5);
5857 if (IS_ERR(trans))
5858 return PTR_ERR(trans);
5859
5860 err = btrfs_find_free_ino(root, &objectid);
5861 if (err)
5862 goto out_fail;
5863
5864 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5865 dentry->d_name.len, btrfs_ino(dir), objectid,
5866 S_IFDIR | mode, &index);
5867 if (IS_ERR(inode)) {
5868 err = PTR_ERR(inode);
5869 goto out_fail;
5870 }
5871
5872 drop_on_err = 1;
5873
5874 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5875 if (err)
5876 goto out_fail;
5877
5878 inode->i_op = &btrfs_dir_inode_operations;
5879 inode->i_fop = &btrfs_dir_file_operations;
5880
5881 btrfs_i_size_write(inode, 0);
5882 err = btrfs_update_inode(trans, root, inode);
5883 if (err)
5884 goto out_fail;
5885
5886 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5887 dentry->d_name.len, 0, index);
5888 if (err)
5889 goto out_fail;
5890
5891 d_instantiate(dentry, inode);
5892 drop_on_err = 0;
5893
5894 out_fail:
5895 btrfs_end_transaction(trans, root);
5896 if (drop_on_err)
5897 iput(inode);
5898 btrfs_btree_balance_dirty(root);
5899 return err;
5900 }
5901
5902 /* helper for btfs_get_extent. Given an existing extent in the tree,
5903 * and an extent that you want to insert, deal with overlap and insert
5904 * the new extent into the tree.
5905 */
5906 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5907 struct extent_map *existing,
5908 struct extent_map *em,
5909 u64 map_start, u64 map_len)
5910 {
5911 u64 start_diff;
5912
5913 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5914 start_diff = map_start - em->start;
5915 em->start = map_start;
5916 em->len = map_len;
5917 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5918 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5919 em->block_start += start_diff;
5920 em->block_len -= start_diff;
5921 }
5922 return add_extent_mapping(em_tree, em, 0);
5923 }
5924
5925 static noinline int uncompress_inline(struct btrfs_path *path,
5926 struct inode *inode, struct page *page,
5927 size_t pg_offset, u64 extent_offset,
5928 struct btrfs_file_extent_item *item)
5929 {
5930 int ret;
5931 struct extent_buffer *leaf = path->nodes[0];
5932 char *tmp;
5933 size_t max_size;
5934 unsigned long inline_size;
5935 unsigned long ptr;
5936 int compress_type;
5937
5938 WARN_ON(pg_offset != 0);
5939 compress_type = btrfs_file_extent_compression(leaf, item);
5940 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5941 inline_size = btrfs_file_extent_inline_item_len(leaf,
5942 btrfs_item_nr(path->slots[0]));
5943 tmp = kmalloc(inline_size, GFP_NOFS);
5944 if (!tmp)
5945 return -ENOMEM;
5946 ptr = btrfs_file_extent_inline_start(item);
5947
5948 read_extent_buffer(leaf, tmp, ptr, inline_size);
5949
5950 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5951 ret = btrfs_decompress(compress_type, tmp, page,
5952 extent_offset, inline_size, max_size);
5953 if (ret) {
5954 char *kaddr = kmap_atomic(page);
5955 unsigned long copy_size = min_t(u64,
5956 PAGE_CACHE_SIZE - pg_offset,
5957 max_size - extent_offset);
5958 memset(kaddr + pg_offset, 0, copy_size);
5959 kunmap_atomic(kaddr);
5960 }
5961 kfree(tmp);
5962 return 0;
5963 }
5964
5965 /*
5966 * a bit scary, this does extent mapping from logical file offset to the disk.
5967 * the ugly parts come from merging extents from the disk with the in-ram
5968 * representation. This gets more complex because of the data=ordered code,
5969 * where the in-ram extents might be locked pending data=ordered completion.
5970 *
5971 * This also copies inline extents directly into the page.
5972 */
5973
5974 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5975 size_t pg_offset, u64 start, u64 len,
5976 int create)
5977 {
5978 int ret;
5979 int err = 0;
5980 u64 bytenr;
5981 u64 extent_start = 0;
5982 u64 extent_end = 0;
5983 u64 objectid = btrfs_ino(inode);
5984 u32 found_type;
5985 struct btrfs_path *path = NULL;
5986 struct btrfs_root *root = BTRFS_I(inode)->root;
5987 struct btrfs_file_extent_item *item;
5988 struct extent_buffer *leaf;
5989 struct btrfs_key found_key;
5990 struct extent_map *em = NULL;
5991 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5992 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5993 struct btrfs_trans_handle *trans = NULL;
5994 int compress_type;
5995
5996 again:
5997 read_lock(&em_tree->lock);
5998 em = lookup_extent_mapping(em_tree, start, len);
5999 if (em)
6000 em->bdev = root->fs_info->fs_devices->latest_bdev;
6001 read_unlock(&em_tree->lock);
6002
6003 if (em) {
6004 if (em->start > start || em->start + em->len <= start)
6005 free_extent_map(em);
6006 else if (em->block_start == EXTENT_MAP_INLINE && page)
6007 free_extent_map(em);
6008 else
6009 goto out;
6010 }
6011 em = alloc_extent_map();
6012 if (!em) {
6013 err = -ENOMEM;
6014 goto out;
6015 }
6016 em->bdev = root->fs_info->fs_devices->latest_bdev;
6017 em->start = EXTENT_MAP_HOLE;
6018 em->orig_start = EXTENT_MAP_HOLE;
6019 em->len = (u64)-1;
6020 em->block_len = (u64)-1;
6021
6022 if (!path) {
6023 path = btrfs_alloc_path();
6024 if (!path) {
6025 err = -ENOMEM;
6026 goto out;
6027 }
6028 /*
6029 * Chances are we'll be called again, so go ahead and do
6030 * readahead
6031 */
6032 path->reada = 1;
6033 }
6034
6035 ret = btrfs_lookup_file_extent(trans, root, path,
6036 objectid, start, trans != NULL);
6037 if (ret < 0) {
6038 err = ret;
6039 goto out;
6040 }
6041
6042 if (ret != 0) {
6043 if (path->slots[0] == 0)
6044 goto not_found;
6045 path->slots[0]--;
6046 }
6047
6048 leaf = path->nodes[0];
6049 item = btrfs_item_ptr(leaf, path->slots[0],
6050 struct btrfs_file_extent_item);
6051 /* are we inside the extent that was found? */
6052 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6053 found_type = btrfs_key_type(&found_key);
6054 if (found_key.objectid != objectid ||
6055 found_type != BTRFS_EXTENT_DATA_KEY) {
6056 /*
6057 * If we backup past the first extent we want to move forward
6058 * and see if there is an extent in front of us, otherwise we'll
6059 * say there is a hole for our whole search range which can
6060 * cause problems.
6061 */
6062 extent_end = start;
6063 goto next;
6064 }
6065
6066 found_type = btrfs_file_extent_type(leaf, item);
6067 extent_start = found_key.offset;
6068 compress_type = btrfs_file_extent_compression(leaf, item);
6069 if (found_type == BTRFS_FILE_EXTENT_REG ||
6070 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6071 extent_end = extent_start +
6072 btrfs_file_extent_num_bytes(leaf, item);
6073 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6074 size_t size;
6075 size = btrfs_file_extent_inline_len(leaf, item);
6076 extent_end = ALIGN(extent_start + size, root->sectorsize);
6077 }
6078 next:
6079 if (start >= extent_end) {
6080 path->slots[0]++;
6081 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6082 ret = btrfs_next_leaf(root, path);
6083 if (ret < 0) {
6084 err = ret;
6085 goto out;
6086 }
6087 if (ret > 0)
6088 goto not_found;
6089 leaf = path->nodes[0];
6090 }
6091 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6092 if (found_key.objectid != objectid ||
6093 found_key.type != BTRFS_EXTENT_DATA_KEY)
6094 goto not_found;
6095 if (start + len <= found_key.offset)
6096 goto not_found;
6097 em->start = start;
6098 em->orig_start = start;
6099 em->len = found_key.offset - start;
6100 goto not_found_em;
6101 }
6102
6103 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6104 if (found_type == BTRFS_FILE_EXTENT_REG ||
6105 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6106 em->start = extent_start;
6107 em->len = extent_end - extent_start;
6108 em->orig_start = extent_start -
6109 btrfs_file_extent_offset(leaf, item);
6110 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6111 item);
6112 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6113 if (bytenr == 0) {
6114 em->block_start = EXTENT_MAP_HOLE;
6115 goto insert;
6116 }
6117 if (compress_type != BTRFS_COMPRESS_NONE) {
6118 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6119 em->compress_type = compress_type;
6120 em->block_start = bytenr;
6121 em->block_len = em->orig_block_len;
6122 } else {
6123 bytenr += btrfs_file_extent_offset(leaf, item);
6124 em->block_start = bytenr;
6125 em->block_len = em->len;
6126 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6127 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6128 }
6129 goto insert;
6130 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6131 unsigned long ptr;
6132 char *map;
6133 size_t size;
6134 size_t extent_offset;
6135 size_t copy_size;
6136
6137 em->block_start = EXTENT_MAP_INLINE;
6138 if (!page || create) {
6139 em->start = extent_start;
6140 em->len = extent_end - extent_start;
6141 goto out;
6142 }
6143
6144 size = btrfs_file_extent_inline_len(leaf, item);
6145 extent_offset = page_offset(page) + pg_offset - extent_start;
6146 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6147 size - extent_offset);
6148 em->start = extent_start + extent_offset;
6149 em->len = ALIGN(copy_size, root->sectorsize);
6150 em->orig_block_len = em->len;
6151 em->orig_start = em->start;
6152 if (compress_type) {
6153 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6154 em->compress_type = compress_type;
6155 }
6156 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6157 if (create == 0 && !PageUptodate(page)) {
6158 if (btrfs_file_extent_compression(leaf, item) !=
6159 BTRFS_COMPRESS_NONE) {
6160 ret = uncompress_inline(path, inode, page,
6161 pg_offset,
6162 extent_offset, item);
6163 BUG_ON(ret); /* -ENOMEM */
6164 } else {
6165 map = kmap(page);
6166 read_extent_buffer(leaf, map + pg_offset, ptr,
6167 copy_size);
6168 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6169 memset(map + pg_offset + copy_size, 0,
6170 PAGE_CACHE_SIZE - pg_offset -
6171 copy_size);
6172 }
6173 kunmap(page);
6174 }
6175 flush_dcache_page(page);
6176 } else if (create && PageUptodate(page)) {
6177 BUG();
6178 if (!trans) {
6179 kunmap(page);
6180 free_extent_map(em);
6181 em = NULL;
6182
6183 btrfs_release_path(path);
6184 trans = btrfs_join_transaction(root);
6185
6186 if (IS_ERR(trans))
6187 return ERR_CAST(trans);
6188 goto again;
6189 }
6190 map = kmap(page);
6191 write_extent_buffer(leaf, map + pg_offset, ptr,
6192 copy_size);
6193 kunmap(page);
6194 btrfs_mark_buffer_dirty(leaf);
6195 }
6196 set_extent_uptodate(io_tree, em->start,
6197 extent_map_end(em) - 1, NULL, GFP_NOFS);
6198 goto insert;
6199 } else {
6200 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6201 }
6202 not_found:
6203 em->start = start;
6204 em->orig_start = start;
6205 em->len = len;
6206 not_found_em:
6207 em->block_start = EXTENT_MAP_HOLE;
6208 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6209 insert:
6210 btrfs_release_path(path);
6211 if (em->start > start || extent_map_end(em) <= start) {
6212 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6213 em->start, em->len, start, len);
6214 err = -EIO;
6215 goto out;
6216 }
6217
6218 err = 0;
6219 write_lock(&em_tree->lock);
6220 ret = add_extent_mapping(em_tree, em, 0);
6221 /* it is possible that someone inserted the extent into the tree
6222 * while we had the lock dropped. It is also possible that
6223 * an overlapping map exists in the tree
6224 */
6225 if (ret == -EEXIST) {
6226 struct extent_map *existing;
6227
6228 ret = 0;
6229
6230 existing = lookup_extent_mapping(em_tree, start, len);
6231 if (existing && (existing->start > start ||
6232 existing->start + existing->len <= start)) {
6233 free_extent_map(existing);
6234 existing = NULL;
6235 }
6236 if (!existing) {
6237 existing = lookup_extent_mapping(em_tree, em->start,
6238 em->len);
6239 if (existing) {
6240 err = merge_extent_mapping(em_tree, existing,
6241 em, start,
6242 root->sectorsize);
6243 free_extent_map(existing);
6244 if (err) {
6245 free_extent_map(em);
6246 em = NULL;
6247 }
6248 } else {
6249 err = -EIO;
6250 free_extent_map(em);
6251 em = NULL;
6252 }
6253 } else {
6254 free_extent_map(em);
6255 em = existing;
6256 err = 0;
6257 }
6258 }
6259 write_unlock(&em_tree->lock);
6260 out:
6261
6262 trace_btrfs_get_extent(root, em);
6263
6264 if (path)
6265 btrfs_free_path(path);
6266 if (trans) {
6267 ret = btrfs_end_transaction(trans, root);
6268 if (!err)
6269 err = ret;
6270 }
6271 if (err) {
6272 free_extent_map(em);
6273 return ERR_PTR(err);
6274 }
6275 BUG_ON(!em); /* Error is always set */
6276 return em;
6277 }
6278
6279 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6280 size_t pg_offset, u64 start, u64 len,
6281 int create)
6282 {
6283 struct extent_map *em;
6284 struct extent_map *hole_em = NULL;
6285 u64 range_start = start;
6286 u64 end;
6287 u64 found;
6288 u64 found_end;
6289 int err = 0;
6290
6291 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6292 if (IS_ERR(em))
6293 return em;
6294 if (em) {
6295 /*
6296 * if our em maps to
6297 * - a hole or
6298 * - a pre-alloc extent,
6299 * there might actually be delalloc bytes behind it.
6300 */
6301 if (em->block_start != EXTENT_MAP_HOLE &&
6302 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6303 return em;
6304 else
6305 hole_em = em;
6306 }
6307
6308 /* check to see if we've wrapped (len == -1 or similar) */
6309 end = start + len;
6310 if (end < start)
6311 end = (u64)-1;
6312 else
6313 end -= 1;
6314
6315 em = NULL;
6316
6317 /* ok, we didn't find anything, lets look for delalloc */
6318 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6319 end, len, EXTENT_DELALLOC, 1);
6320 found_end = range_start + found;
6321 if (found_end < range_start)
6322 found_end = (u64)-1;
6323
6324 /*
6325 * we didn't find anything useful, return
6326 * the original results from get_extent()
6327 */
6328 if (range_start > end || found_end <= start) {
6329 em = hole_em;
6330 hole_em = NULL;
6331 goto out;
6332 }
6333
6334 /* adjust the range_start to make sure it doesn't
6335 * go backwards from the start they passed in
6336 */
6337 range_start = max(start, range_start);
6338 found = found_end - range_start;
6339
6340 if (found > 0) {
6341 u64 hole_start = start;
6342 u64 hole_len = len;
6343
6344 em = alloc_extent_map();
6345 if (!em) {
6346 err = -ENOMEM;
6347 goto out;
6348 }
6349 /*
6350 * when btrfs_get_extent can't find anything it
6351 * returns one huge hole
6352 *
6353 * make sure what it found really fits our range, and
6354 * adjust to make sure it is based on the start from
6355 * the caller
6356 */
6357 if (hole_em) {
6358 u64 calc_end = extent_map_end(hole_em);
6359
6360 if (calc_end <= start || (hole_em->start > end)) {
6361 free_extent_map(hole_em);
6362 hole_em = NULL;
6363 } else {
6364 hole_start = max(hole_em->start, start);
6365 hole_len = calc_end - hole_start;
6366 }
6367 }
6368 em->bdev = NULL;
6369 if (hole_em && range_start > hole_start) {
6370 /* our hole starts before our delalloc, so we
6371 * have to return just the parts of the hole
6372 * that go until the delalloc starts
6373 */
6374 em->len = min(hole_len,
6375 range_start - hole_start);
6376 em->start = hole_start;
6377 em->orig_start = hole_start;
6378 /*
6379 * don't adjust block start at all,
6380 * it is fixed at EXTENT_MAP_HOLE
6381 */
6382 em->block_start = hole_em->block_start;
6383 em->block_len = hole_len;
6384 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6385 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6386 } else {
6387 em->start = range_start;
6388 em->len = found;
6389 em->orig_start = range_start;
6390 em->block_start = EXTENT_MAP_DELALLOC;
6391 em->block_len = found;
6392 }
6393 } else if (hole_em) {
6394 return hole_em;
6395 }
6396 out:
6397
6398 free_extent_map(hole_em);
6399 if (err) {
6400 free_extent_map(em);
6401 return ERR_PTR(err);
6402 }
6403 return em;
6404 }
6405
6406 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6407 u64 start, u64 len)
6408 {
6409 struct btrfs_root *root = BTRFS_I(inode)->root;
6410 struct extent_map *em;
6411 struct btrfs_key ins;
6412 u64 alloc_hint;
6413 int ret;
6414
6415 alloc_hint = get_extent_allocation_hint(inode, start, len);
6416 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6417 alloc_hint, &ins, 1);
6418 if (ret)
6419 return ERR_PTR(ret);
6420
6421 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6422 ins.offset, ins.offset, ins.offset, 0);
6423 if (IS_ERR(em)) {
6424 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6425 return em;
6426 }
6427
6428 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6429 ins.offset, ins.offset, 0);
6430 if (ret) {
6431 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6432 free_extent_map(em);
6433 return ERR_PTR(ret);
6434 }
6435
6436 return em;
6437 }
6438
6439 /*
6440 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6441 * block must be cow'd
6442 */
6443 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6444 u64 *orig_start, u64 *orig_block_len,
6445 u64 *ram_bytes)
6446 {
6447 struct btrfs_trans_handle *trans;
6448 struct btrfs_path *path;
6449 int ret;
6450 struct extent_buffer *leaf;
6451 struct btrfs_root *root = BTRFS_I(inode)->root;
6452 struct btrfs_file_extent_item *fi;
6453 struct btrfs_key key;
6454 u64 disk_bytenr;
6455 u64 backref_offset;
6456 u64 extent_end;
6457 u64 num_bytes;
6458 int slot;
6459 int found_type;
6460 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6461 path = btrfs_alloc_path();
6462 if (!path)
6463 return -ENOMEM;
6464
6465 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6466 offset, 0);
6467 if (ret < 0)
6468 goto out;
6469
6470 slot = path->slots[0];
6471 if (ret == 1) {
6472 if (slot == 0) {
6473 /* can't find the item, must cow */
6474 ret = 0;
6475 goto out;
6476 }
6477 slot--;
6478 }
6479 ret = 0;
6480 leaf = path->nodes[0];
6481 btrfs_item_key_to_cpu(leaf, &key, slot);
6482 if (key.objectid != btrfs_ino(inode) ||
6483 key.type != BTRFS_EXTENT_DATA_KEY) {
6484 /* not our file or wrong item type, must cow */
6485 goto out;
6486 }
6487
6488 if (key.offset > offset) {
6489 /* Wrong offset, must cow */
6490 goto out;
6491 }
6492
6493 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6494 found_type = btrfs_file_extent_type(leaf, fi);
6495 if (found_type != BTRFS_FILE_EXTENT_REG &&
6496 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6497 /* not a regular extent, must cow */
6498 goto out;
6499 }
6500
6501 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6502 goto out;
6503
6504 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6505 if (disk_bytenr == 0)
6506 goto out;
6507
6508 if (btrfs_file_extent_compression(leaf, fi) ||
6509 btrfs_file_extent_encryption(leaf, fi) ||
6510 btrfs_file_extent_other_encoding(leaf, fi))
6511 goto out;
6512
6513 backref_offset = btrfs_file_extent_offset(leaf, fi);
6514
6515 if (orig_start) {
6516 *orig_start = key.offset - backref_offset;
6517 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6518 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6519 }
6520
6521 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6522
6523 if (btrfs_extent_readonly(root, disk_bytenr))
6524 goto out;
6525 btrfs_release_path(path);
6526
6527 /*
6528 * look for other files referencing this extent, if we
6529 * find any we must cow
6530 */
6531 trans = btrfs_join_transaction(root);
6532 if (IS_ERR(trans)) {
6533 ret = 0;
6534 goto out;
6535 }
6536
6537 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6538 key.offset - backref_offset, disk_bytenr);
6539 btrfs_end_transaction(trans, root);
6540 if (ret) {
6541 ret = 0;
6542 goto out;
6543 }
6544
6545 /*
6546 * adjust disk_bytenr and num_bytes to cover just the bytes
6547 * in this extent we are about to write. If there
6548 * are any csums in that range we have to cow in order
6549 * to keep the csums correct
6550 */
6551 disk_bytenr += backref_offset;
6552 disk_bytenr += offset - key.offset;
6553 num_bytes = min(offset + *len, extent_end) - offset;
6554 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6555 goto out;
6556 /*
6557 * all of the above have passed, it is safe to overwrite this extent
6558 * without cow
6559 */
6560 *len = num_bytes;
6561 ret = 1;
6562 out:
6563 btrfs_free_path(path);
6564 return ret;
6565 }
6566
6567 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6568 struct extent_state **cached_state, int writing)
6569 {
6570 struct btrfs_ordered_extent *ordered;
6571 int ret = 0;
6572
6573 while (1) {
6574 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6575 0, cached_state);
6576 /*
6577 * We're concerned with the entire range that we're going to be
6578 * doing DIO to, so we need to make sure theres no ordered
6579 * extents in this range.
6580 */
6581 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6582 lockend - lockstart + 1);
6583
6584 /*
6585 * We need to make sure there are no buffered pages in this
6586 * range either, we could have raced between the invalidate in
6587 * generic_file_direct_write and locking the extent. The
6588 * invalidate needs to happen so that reads after a write do not
6589 * get stale data.
6590 */
6591 if (!ordered && (!writing ||
6592 !test_range_bit(&BTRFS_I(inode)->io_tree,
6593 lockstart, lockend, EXTENT_UPTODATE, 0,
6594 *cached_state)))
6595 break;
6596
6597 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6598 cached_state, GFP_NOFS);
6599
6600 if (ordered) {
6601 btrfs_start_ordered_extent(inode, ordered, 1);
6602 btrfs_put_ordered_extent(ordered);
6603 } else {
6604 /* Screw you mmap */
6605 ret = filemap_write_and_wait_range(inode->i_mapping,
6606 lockstart,
6607 lockend);
6608 if (ret)
6609 break;
6610
6611 /*
6612 * If we found a page that couldn't be invalidated just
6613 * fall back to buffered.
6614 */
6615 ret = invalidate_inode_pages2_range(inode->i_mapping,
6616 lockstart >> PAGE_CACHE_SHIFT,
6617 lockend >> PAGE_CACHE_SHIFT);
6618 if (ret)
6619 break;
6620 }
6621
6622 cond_resched();
6623 }
6624
6625 return ret;
6626 }
6627
6628 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6629 u64 len, u64 orig_start,
6630 u64 block_start, u64 block_len,
6631 u64 orig_block_len, u64 ram_bytes,
6632 int type)
6633 {
6634 struct extent_map_tree *em_tree;
6635 struct extent_map *em;
6636 struct btrfs_root *root = BTRFS_I(inode)->root;
6637 int ret;
6638
6639 em_tree = &BTRFS_I(inode)->extent_tree;
6640 em = alloc_extent_map();
6641 if (!em)
6642 return ERR_PTR(-ENOMEM);
6643
6644 em->start = start;
6645 em->orig_start = orig_start;
6646 em->mod_start = start;
6647 em->mod_len = len;
6648 em->len = len;
6649 em->block_len = block_len;
6650 em->block_start = block_start;
6651 em->bdev = root->fs_info->fs_devices->latest_bdev;
6652 em->orig_block_len = orig_block_len;
6653 em->ram_bytes = ram_bytes;
6654 em->generation = -1;
6655 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6656 if (type == BTRFS_ORDERED_PREALLOC)
6657 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6658
6659 do {
6660 btrfs_drop_extent_cache(inode, em->start,
6661 em->start + em->len - 1, 0);
6662 write_lock(&em_tree->lock);
6663 ret = add_extent_mapping(em_tree, em, 1);
6664 write_unlock(&em_tree->lock);
6665 } while (ret == -EEXIST);
6666
6667 if (ret) {
6668 free_extent_map(em);
6669 return ERR_PTR(ret);
6670 }
6671
6672 return em;
6673 }
6674
6675
6676 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6677 struct buffer_head *bh_result, int create)
6678 {
6679 struct extent_map *em;
6680 struct btrfs_root *root = BTRFS_I(inode)->root;
6681 struct extent_state *cached_state = NULL;
6682 u64 start = iblock << inode->i_blkbits;
6683 u64 lockstart, lockend;
6684 u64 len = bh_result->b_size;
6685 int unlock_bits = EXTENT_LOCKED;
6686 int ret = 0;
6687
6688 if (create)
6689 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6690 else
6691 len = min_t(u64, len, root->sectorsize);
6692
6693 lockstart = start;
6694 lockend = start + len - 1;
6695
6696 /*
6697 * If this errors out it's because we couldn't invalidate pagecache for
6698 * this range and we need to fallback to buffered.
6699 */
6700 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6701 return -ENOTBLK;
6702
6703 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6704 if (IS_ERR(em)) {
6705 ret = PTR_ERR(em);
6706 goto unlock_err;
6707 }
6708
6709 /*
6710 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6711 * io. INLINE is special, and we could probably kludge it in here, but
6712 * it's still buffered so for safety lets just fall back to the generic
6713 * buffered path.
6714 *
6715 * For COMPRESSED we _have_ to read the entire extent in so we can
6716 * decompress it, so there will be buffering required no matter what we
6717 * do, so go ahead and fallback to buffered.
6718 *
6719 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6720 * to buffered IO. Don't blame me, this is the price we pay for using
6721 * the generic code.
6722 */
6723 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6724 em->block_start == EXTENT_MAP_INLINE) {
6725 free_extent_map(em);
6726 ret = -ENOTBLK;
6727 goto unlock_err;
6728 }
6729
6730 /* Just a good old fashioned hole, return */
6731 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6732 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6733 free_extent_map(em);
6734 goto unlock_err;
6735 }
6736
6737 /*
6738 * We don't allocate a new extent in the following cases
6739 *
6740 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6741 * existing extent.
6742 * 2) The extent is marked as PREALLOC. We're good to go here and can
6743 * just use the extent.
6744 *
6745 */
6746 if (!create) {
6747 len = min(len, em->len - (start - em->start));
6748 lockstart = start + len;
6749 goto unlock;
6750 }
6751
6752 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6753 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6754 em->block_start != EXTENT_MAP_HOLE)) {
6755 int type;
6756 int ret;
6757 u64 block_start, orig_start, orig_block_len, ram_bytes;
6758
6759 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6760 type = BTRFS_ORDERED_PREALLOC;
6761 else
6762 type = BTRFS_ORDERED_NOCOW;
6763 len = min(len, em->len - (start - em->start));
6764 block_start = em->block_start + (start - em->start);
6765
6766 if (can_nocow_extent(inode, start, &len, &orig_start,
6767 &orig_block_len, &ram_bytes) == 1) {
6768 if (type == BTRFS_ORDERED_PREALLOC) {
6769 free_extent_map(em);
6770 em = create_pinned_em(inode, start, len,
6771 orig_start,
6772 block_start, len,
6773 orig_block_len,
6774 ram_bytes, type);
6775 if (IS_ERR(em))
6776 goto unlock_err;
6777 }
6778
6779 ret = btrfs_add_ordered_extent_dio(inode, start,
6780 block_start, len, len, type);
6781 if (ret) {
6782 free_extent_map(em);
6783 goto unlock_err;
6784 }
6785 goto unlock;
6786 }
6787 }
6788
6789 /*
6790 * this will cow the extent, reset the len in case we changed
6791 * it above
6792 */
6793 len = bh_result->b_size;
6794 free_extent_map(em);
6795 em = btrfs_new_extent_direct(inode, start, len);
6796 if (IS_ERR(em)) {
6797 ret = PTR_ERR(em);
6798 goto unlock_err;
6799 }
6800 len = min(len, em->len - (start - em->start));
6801 unlock:
6802 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6803 inode->i_blkbits;
6804 bh_result->b_size = len;
6805 bh_result->b_bdev = em->bdev;
6806 set_buffer_mapped(bh_result);
6807 if (create) {
6808 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6809 set_buffer_new(bh_result);
6810
6811 /*
6812 * Need to update the i_size under the extent lock so buffered
6813 * readers will get the updated i_size when we unlock.
6814 */
6815 if (start + len > i_size_read(inode))
6816 i_size_write(inode, start + len);
6817
6818 spin_lock(&BTRFS_I(inode)->lock);
6819 BTRFS_I(inode)->outstanding_extents++;
6820 spin_unlock(&BTRFS_I(inode)->lock);
6821
6822 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6823 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6824 &cached_state, GFP_NOFS);
6825 BUG_ON(ret);
6826 }
6827
6828 /*
6829 * In the case of write we need to clear and unlock the entire range,
6830 * in the case of read we need to unlock only the end area that we
6831 * aren't using if there is any left over space.
6832 */
6833 if (lockstart < lockend) {
6834 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6835 lockend, unlock_bits, 1, 0,
6836 &cached_state, GFP_NOFS);
6837 } else {
6838 free_extent_state(cached_state);
6839 }
6840
6841 free_extent_map(em);
6842
6843 return 0;
6844
6845 unlock_err:
6846 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6847 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6848 return ret;
6849 }
6850
6851 static void btrfs_endio_direct_read(struct bio *bio, int err)
6852 {
6853 struct btrfs_dio_private *dip = bio->bi_private;
6854 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6855 struct bio_vec *bvec = bio->bi_io_vec;
6856 struct inode *inode = dip->inode;
6857 struct btrfs_root *root = BTRFS_I(inode)->root;
6858 struct bio *dio_bio;
6859 u32 *csums = (u32 *)dip->csum;
6860 int index = 0;
6861 u64 start;
6862
6863 start = dip->logical_offset;
6864 do {
6865 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6866 struct page *page = bvec->bv_page;
6867 char *kaddr;
6868 u32 csum = ~(u32)0;
6869 unsigned long flags;
6870
6871 local_irq_save(flags);
6872 kaddr = kmap_atomic(page);
6873 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6874 csum, bvec->bv_len);
6875 btrfs_csum_final(csum, (char *)&csum);
6876 kunmap_atomic(kaddr);
6877 local_irq_restore(flags);
6878
6879 flush_dcache_page(bvec->bv_page);
6880 if (csum != csums[index]) {
6881 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6882 btrfs_ino(inode), start, csum,
6883 csums[index]);
6884 err = -EIO;
6885 }
6886 }
6887
6888 start += bvec->bv_len;
6889 bvec++;
6890 index++;
6891 } while (bvec <= bvec_end);
6892
6893 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6894 dip->logical_offset + dip->bytes - 1);
6895 dio_bio = dip->dio_bio;
6896
6897 kfree(dip);
6898
6899 /* If we had a csum failure make sure to clear the uptodate flag */
6900 if (err)
6901 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6902 dio_end_io(dio_bio, err);
6903 bio_put(bio);
6904 }
6905
6906 static void btrfs_endio_direct_write(struct bio *bio, int err)
6907 {
6908 struct btrfs_dio_private *dip = bio->bi_private;
6909 struct inode *inode = dip->inode;
6910 struct btrfs_root *root = BTRFS_I(inode)->root;
6911 struct btrfs_ordered_extent *ordered = NULL;
6912 u64 ordered_offset = dip->logical_offset;
6913 u64 ordered_bytes = dip->bytes;
6914 struct bio *dio_bio;
6915 int ret;
6916
6917 if (err)
6918 goto out_done;
6919 again:
6920 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6921 &ordered_offset,
6922 ordered_bytes, !err);
6923 if (!ret)
6924 goto out_test;
6925
6926 ordered->work.func = finish_ordered_fn;
6927 ordered->work.flags = 0;
6928 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6929 &ordered->work);
6930 out_test:
6931 /*
6932 * our bio might span multiple ordered extents. If we haven't
6933 * completed the accounting for the whole dio, go back and try again
6934 */
6935 if (ordered_offset < dip->logical_offset + dip->bytes) {
6936 ordered_bytes = dip->logical_offset + dip->bytes -
6937 ordered_offset;
6938 ordered = NULL;
6939 goto again;
6940 }
6941 out_done:
6942 dio_bio = dip->dio_bio;
6943
6944 kfree(dip);
6945
6946 /* If we had an error make sure to clear the uptodate flag */
6947 if (err)
6948 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6949 dio_end_io(dio_bio, err);
6950 bio_put(bio);
6951 }
6952
6953 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6954 struct bio *bio, int mirror_num,
6955 unsigned long bio_flags, u64 offset)
6956 {
6957 int ret;
6958 struct btrfs_root *root = BTRFS_I(inode)->root;
6959 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6960 BUG_ON(ret); /* -ENOMEM */
6961 return 0;
6962 }
6963
6964 static void btrfs_end_dio_bio(struct bio *bio, int err)
6965 {
6966 struct btrfs_dio_private *dip = bio->bi_private;
6967
6968 if (err) {
6969 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
6970 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
6971 btrfs_ino(dip->inode), bio->bi_rw,
6972 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6973 dip->errors = 1;
6974
6975 /*
6976 * before atomic variable goto zero, we must make sure
6977 * dip->errors is perceived to be set.
6978 */
6979 smp_mb__before_atomic_dec();
6980 }
6981
6982 /* if there are more bios still pending for this dio, just exit */
6983 if (!atomic_dec_and_test(&dip->pending_bios))
6984 goto out;
6985
6986 if (dip->errors) {
6987 bio_io_error(dip->orig_bio);
6988 } else {
6989 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6990 bio_endio(dip->orig_bio, 0);
6991 }
6992 out:
6993 bio_put(bio);
6994 }
6995
6996 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6997 u64 first_sector, gfp_t gfp_flags)
6998 {
6999 int nr_vecs = bio_get_nr_vecs(bdev);
7000 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7001 }
7002
7003 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7004 int rw, u64 file_offset, int skip_sum,
7005 int async_submit)
7006 {
7007 struct btrfs_dio_private *dip = bio->bi_private;
7008 int write = rw & REQ_WRITE;
7009 struct btrfs_root *root = BTRFS_I(inode)->root;
7010 int ret;
7011
7012 if (async_submit)
7013 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7014
7015 bio_get(bio);
7016
7017 if (!write) {
7018 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7019 if (ret)
7020 goto err;
7021 }
7022
7023 if (skip_sum)
7024 goto map;
7025
7026 if (write && async_submit) {
7027 ret = btrfs_wq_submit_bio(root->fs_info,
7028 inode, rw, bio, 0, 0,
7029 file_offset,
7030 __btrfs_submit_bio_start_direct_io,
7031 __btrfs_submit_bio_done);
7032 goto err;
7033 } else if (write) {
7034 /*
7035 * If we aren't doing async submit, calculate the csum of the
7036 * bio now.
7037 */
7038 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7039 if (ret)
7040 goto err;
7041 } else if (!skip_sum) {
7042 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7043 file_offset);
7044 if (ret)
7045 goto err;
7046 }
7047
7048 map:
7049 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7050 err:
7051 bio_put(bio);
7052 return ret;
7053 }
7054
7055 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7056 int skip_sum)
7057 {
7058 struct inode *inode = dip->inode;
7059 struct btrfs_root *root = BTRFS_I(inode)->root;
7060 struct bio *bio;
7061 struct bio *orig_bio = dip->orig_bio;
7062 struct bio_vec *bvec = orig_bio->bi_io_vec;
7063 u64 start_sector = orig_bio->bi_sector;
7064 u64 file_offset = dip->logical_offset;
7065 u64 submit_len = 0;
7066 u64 map_length;
7067 int nr_pages = 0;
7068 int ret = 0;
7069 int async_submit = 0;
7070
7071 map_length = orig_bio->bi_size;
7072 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7073 &map_length, NULL, 0);
7074 if (ret) {
7075 bio_put(orig_bio);
7076 return -EIO;
7077 }
7078
7079 if (map_length >= orig_bio->bi_size) {
7080 bio = orig_bio;
7081 goto submit;
7082 }
7083
7084 /* async crcs make it difficult to collect full stripe writes. */
7085 if (btrfs_get_alloc_profile(root, 1) &
7086 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7087 async_submit = 0;
7088 else
7089 async_submit = 1;
7090
7091 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7092 if (!bio)
7093 return -ENOMEM;
7094 bio->bi_private = dip;
7095 bio->bi_end_io = btrfs_end_dio_bio;
7096 atomic_inc(&dip->pending_bios);
7097
7098 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7099 if (unlikely(map_length < submit_len + bvec->bv_len ||
7100 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7101 bvec->bv_offset) < bvec->bv_len)) {
7102 /*
7103 * inc the count before we submit the bio so
7104 * we know the end IO handler won't happen before
7105 * we inc the count. Otherwise, the dip might get freed
7106 * before we're done setting it up
7107 */
7108 atomic_inc(&dip->pending_bios);
7109 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7110 file_offset, skip_sum,
7111 async_submit);
7112 if (ret) {
7113 bio_put(bio);
7114 atomic_dec(&dip->pending_bios);
7115 goto out_err;
7116 }
7117
7118 start_sector += submit_len >> 9;
7119 file_offset += submit_len;
7120
7121 submit_len = 0;
7122 nr_pages = 0;
7123
7124 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7125 start_sector, GFP_NOFS);
7126 if (!bio)
7127 goto out_err;
7128 bio->bi_private = dip;
7129 bio->bi_end_io = btrfs_end_dio_bio;
7130
7131 map_length = orig_bio->bi_size;
7132 ret = btrfs_map_block(root->fs_info, rw,
7133 start_sector << 9,
7134 &map_length, NULL, 0);
7135 if (ret) {
7136 bio_put(bio);
7137 goto out_err;
7138 }
7139 } else {
7140 submit_len += bvec->bv_len;
7141 nr_pages++;
7142 bvec++;
7143 }
7144 }
7145
7146 submit:
7147 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7148 async_submit);
7149 if (!ret)
7150 return 0;
7151
7152 bio_put(bio);
7153 out_err:
7154 dip->errors = 1;
7155 /*
7156 * before atomic variable goto zero, we must
7157 * make sure dip->errors is perceived to be set.
7158 */
7159 smp_mb__before_atomic_dec();
7160 if (atomic_dec_and_test(&dip->pending_bios))
7161 bio_io_error(dip->orig_bio);
7162
7163 /* bio_end_io() will handle error, so we needn't return it */
7164 return 0;
7165 }
7166
7167 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7168 struct inode *inode, loff_t file_offset)
7169 {
7170 struct btrfs_root *root = BTRFS_I(inode)->root;
7171 struct btrfs_dio_private *dip;
7172 struct bio *io_bio;
7173 int skip_sum;
7174 int sum_len;
7175 int write = rw & REQ_WRITE;
7176 int ret = 0;
7177 u16 csum_size;
7178
7179 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7180
7181 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7182 if (!io_bio) {
7183 ret = -ENOMEM;
7184 goto free_ordered;
7185 }
7186
7187 if (!skip_sum && !write) {
7188 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7189 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7190 sum_len *= csum_size;
7191 } else {
7192 sum_len = 0;
7193 }
7194
7195 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7196 if (!dip) {
7197 ret = -ENOMEM;
7198 goto free_io_bio;
7199 }
7200
7201 dip->private = dio_bio->bi_private;
7202 dip->inode = inode;
7203 dip->logical_offset = file_offset;
7204 dip->bytes = dio_bio->bi_size;
7205 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7206 io_bio->bi_private = dip;
7207 dip->errors = 0;
7208 dip->orig_bio = io_bio;
7209 dip->dio_bio = dio_bio;
7210 atomic_set(&dip->pending_bios, 0);
7211
7212 if (write)
7213 io_bio->bi_end_io = btrfs_endio_direct_write;
7214 else
7215 io_bio->bi_end_io = btrfs_endio_direct_read;
7216
7217 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7218 if (!ret)
7219 return;
7220
7221 free_io_bio:
7222 bio_put(io_bio);
7223
7224 free_ordered:
7225 /*
7226 * If this is a write, we need to clean up the reserved space and kill
7227 * the ordered extent.
7228 */
7229 if (write) {
7230 struct btrfs_ordered_extent *ordered;
7231 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7232 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7233 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7234 btrfs_free_reserved_extent(root, ordered->start,
7235 ordered->disk_len);
7236 btrfs_put_ordered_extent(ordered);
7237 btrfs_put_ordered_extent(ordered);
7238 }
7239 bio_endio(dio_bio, ret);
7240 }
7241
7242 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7243 const struct iovec *iov, loff_t offset,
7244 unsigned long nr_segs)
7245 {
7246 int seg;
7247 int i;
7248 size_t size;
7249 unsigned long addr;
7250 unsigned blocksize_mask = root->sectorsize - 1;
7251 ssize_t retval = -EINVAL;
7252 loff_t end = offset;
7253
7254 if (offset & blocksize_mask)
7255 goto out;
7256
7257 /* Check the memory alignment. Blocks cannot straddle pages */
7258 for (seg = 0; seg < nr_segs; seg++) {
7259 addr = (unsigned long)iov[seg].iov_base;
7260 size = iov[seg].iov_len;
7261 end += size;
7262 if ((addr & blocksize_mask) || (size & blocksize_mask))
7263 goto out;
7264
7265 /* If this is a write we don't need to check anymore */
7266 if (rw & WRITE)
7267 continue;
7268
7269 /*
7270 * Check to make sure we don't have duplicate iov_base's in this
7271 * iovec, if so return EINVAL, otherwise we'll get csum errors
7272 * when reading back.
7273 */
7274 for (i = seg + 1; i < nr_segs; i++) {
7275 if (iov[seg].iov_base == iov[i].iov_base)
7276 goto out;
7277 }
7278 }
7279 retval = 0;
7280 out:
7281 return retval;
7282 }
7283
7284 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7285 const struct iovec *iov, loff_t offset,
7286 unsigned long nr_segs)
7287 {
7288 struct file *file = iocb->ki_filp;
7289 struct inode *inode = file->f_mapping->host;
7290 size_t count = 0;
7291 int flags = 0;
7292 bool wakeup = true;
7293 bool relock = false;
7294 ssize_t ret;
7295
7296 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7297 offset, nr_segs))
7298 return 0;
7299
7300 atomic_inc(&inode->i_dio_count);
7301 smp_mb__after_atomic_inc();
7302
7303 /*
7304 * The generic stuff only does filemap_write_and_wait_range, which isn't
7305 * enough if we've written compressed pages to this area, so we need to
7306 * call btrfs_wait_ordered_range to make absolutely sure that any
7307 * outstanding dirty pages are on disk.
7308 */
7309 count = iov_length(iov, nr_segs);
7310 ret = btrfs_wait_ordered_range(inode, offset, count);
7311 if (ret)
7312 return ret;
7313
7314 if (rw & WRITE) {
7315 /*
7316 * If the write DIO is beyond the EOF, we need update
7317 * the isize, but it is protected by i_mutex. So we can
7318 * not unlock the i_mutex at this case.
7319 */
7320 if (offset + count <= inode->i_size) {
7321 mutex_unlock(&inode->i_mutex);
7322 relock = true;
7323 }
7324 ret = btrfs_delalloc_reserve_space(inode, count);
7325 if (ret)
7326 goto out;
7327 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7328 &BTRFS_I(inode)->runtime_flags))) {
7329 inode_dio_done(inode);
7330 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7331 wakeup = false;
7332 }
7333
7334 ret = __blockdev_direct_IO(rw, iocb, inode,
7335 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7336 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7337 btrfs_submit_direct, flags);
7338 if (rw & WRITE) {
7339 if (ret < 0 && ret != -EIOCBQUEUED)
7340 btrfs_delalloc_release_space(inode, count);
7341 else if (ret >= 0 && (size_t)ret < count)
7342 btrfs_delalloc_release_space(inode,
7343 count - (size_t)ret);
7344 else
7345 btrfs_delalloc_release_metadata(inode, 0);
7346 }
7347 out:
7348 if (wakeup)
7349 inode_dio_done(inode);
7350 if (relock)
7351 mutex_lock(&inode->i_mutex);
7352
7353 return ret;
7354 }
7355
7356 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7357
7358 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7359 __u64 start, __u64 len)
7360 {
7361 int ret;
7362
7363 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7364 if (ret)
7365 return ret;
7366
7367 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7368 }
7369
7370 int btrfs_readpage(struct file *file, struct page *page)
7371 {
7372 struct extent_io_tree *tree;
7373 tree = &BTRFS_I(page->mapping->host)->io_tree;
7374 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7375 }
7376
7377 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7378 {
7379 struct extent_io_tree *tree;
7380
7381
7382 if (current->flags & PF_MEMALLOC) {
7383 redirty_page_for_writepage(wbc, page);
7384 unlock_page(page);
7385 return 0;
7386 }
7387 tree = &BTRFS_I(page->mapping->host)->io_tree;
7388 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7389 }
7390
7391 static int btrfs_writepages(struct address_space *mapping,
7392 struct writeback_control *wbc)
7393 {
7394 struct extent_io_tree *tree;
7395
7396 tree = &BTRFS_I(mapping->host)->io_tree;
7397 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7398 }
7399
7400 static int
7401 btrfs_readpages(struct file *file, struct address_space *mapping,
7402 struct list_head *pages, unsigned nr_pages)
7403 {
7404 struct extent_io_tree *tree;
7405 tree = &BTRFS_I(mapping->host)->io_tree;
7406 return extent_readpages(tree, mapping, pages, nr_pages,
7407 btrfs_get_extent);
7408 }
7409 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7410 {
7411 struct extent_io_tree *tree;
7412 struct extent_map_tree *map;
7413 int ret;
7414
7415 tree = &BTRFS_I(page->mapping->host)->io_tree;
7416 map = &BTRFS_I(page->mapping->host)->extent_tree;
7417 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7418 if (ret == 1) {
7419 ClearPagePrivate(page);
7420 set_page_private(page, 0);
7421 page_cache_release(page);
7422 }
7423 return ret;
7424 }
7425
7426 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7427 {
7428 if (PageWriteback(page) || PageDirty(page))
7429 return 0;
7430 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7431 }
7432
7433 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7434 unsigned int length)
7435 {
7436 struct inode *inode = page->mapping->host;
7437 struct extent_io_tree *tree;
7438 struct btrfs_ordered_extent *ordered;
7439 struct extent_state *cached_state = NULL;
7440 u64 page_start = page_offset(page);
7441 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7442 int inode_evicting = inode->i_state & I_FREEING;
7443
7444 /*
7445 * we have the page locked, so new writeback can't start,
7446 * and the dirty bit won't be cleared while we are here.
7447 *
7448 * Wait for IO on this page so that we can safely clear
7449 * the PagePrivate2 bit and do ordered accounting
7450 */
7451 wait_on_page_writeback(page);
7452
7453 tree = &BTRFS_I(inode)->io_tree;
7454 if (offset) {
7455 btrfs_releasepage(page, GFP_NOFS);
7456 return;
7457 }
7458
7459 if (!inode_evicting)
7460 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7461 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7462 if (ordered) {
7463 /*
7464 * IO on this page will never be started, so we need
7465 * to account for any ordered extents now
7466 */
7467 if (!inode_evicting)
7468 clear_extent_bit(tree, page_start, page_end,
7469 EXTENT_DIRTY | EXTENT_DELALLOC |
7470 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7471 EXTENT_DEFRAG, 1, 0, &cached_state,
7472 GFP_NOFS);
7473 /*
7474 * whoever cleared the private bit is responsible
7475 * for the finish_ordered_io
7476 */
7477 if (TestClearPagePrivate2(page)) {
7478 struct btrfs_ordered_inode_tree *tree;
7479 u64 new_len;
7480
7481 tree = &BTRFS_I(inode)->ordered_tree;
7482
7483 spin_lock_irq(&tree->lock);
7484 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7485 new_len = page_start - ordered->file_offset;
7486 if (new_len < ordered->truncated_len)
7487 ordered->truncated_len = new_len;
7488 spin_unlock_irq(&tree->lock);
7489
7490 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7491 page_start,
7492 PAGE_CACHE_SIZE, 1))
7493 btrfs_finish_ordered_io(ordered);
7494 }
7495 btrfs_put_ordered_extent(ordered);
7496 if (!inode_evicting) {
7497 cached_state = NULL;
7498 lock_extent_bits(tree, page_start, page_end, 0,
7499 &cached_state);
7500 }
7501 }
7502
7503 if (!inode_evicting) {
7504 clear_extent_bit(tree, page_start, page_end,
7505 EXTENT_LOCKED | EXTENT_DIRTY |
7506 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7507 EXTENT_DEFRAG, 1, 1,
7508 &cached_state, GFP_NOFS);
7509
7510 __btrfs_releasepage(page, GFP_NOFS);
7511 }
7512
7513 ClearPageChecked(page);
7514 if (PagePrivate(page)) {
7515 ClearPagePrivate(page);
7516 set_page_private(page, 0);
7517 page_cache_release(page);
7518 }
7519 }
7520
7521 /*
7522 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7523 * called from a page fault handler when a page is first dirtied. Hence we must
7524 * be careful to check for EOF conditions here. We set the page up correctly
7525 * for a written page which means we get ENOSPC checking when writing into
7526 * holes and correct delalloc and unwritten extent mapping on filesystems that
7527 * support these features.
7528 *
7529 * We are not allowed to take the i_mutex here so we have to play games to
7530 * protect against truncate races as the page could now be beyond EOF. Because
7531 * vmtruncate() writes the inode size before removing pages, once we have the
7532 * page lock we can determine safely if the page is beyond EOF. If it is not
7533 * beyond EOF, then the page is guaranteed safe against truncation until we
7534 * unlock the page.
7535 */
7536 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7537 {
7538 struct page *page = vmf->page;
7539 struct inode *inode = file_inode(vma->vm_file);
7540 struct btrfs_root *root = BTRFS_I(inode)->root;
7541 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7542 struct btrfs_ordered_extent *ordered;
7543 struct extent_state *cached_state = NULL;
7544 char *kaddr;
7545 unsigned long zero_start;
7546 loff_t size;
7547 int ret;
7548 int reserved = 0;
7549 u64 page_start;
7550 u64 page_end;
7551
7552 sb_start_pagefault(inode->i_sb);
7553 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7554 if (!ret) {
7555 ret = file_update_time(vma->vm_file);
7556 reserved = 1;
7557 }
7558 if (ret) {
7559 if (ret == -ENOMEM)
7560 ret = VM_FAULT_OOM;
7561 else /* -ENOSPC, -EIO, etc */
7562 ret = VM_FAULT_SIGBUS;
7563 if (reserved)
7564 goto out;
7565 goto out_noreserve;
7566 }
7567
7568 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7569 again:
7570 lock_page(page);
7571 size = i_size_read(inode);
7572 page_start = page_offset(page);
7573 page_end = page_start + PAGE_CACHE_SIZE - 1;
7574
7575 if ((page->mapping != inode->i_mapping) ||
7576 (page_start >= size)) {
7577 /* page got truncated out from underneath us */
7578 goto out_unlock;
7579 }
7580 wait_on_page_writeback(page);
7581
7582 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7583 set_page_extent_mapped(page);
7584
7585 /*
7586 * we can't set the delalloc bits if there are pending ordered
7587 * extents. Drop our locks and wait for them to finish
7588 */
7589 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7590 if (ordered) {
7591 unlock_extent_cached(io_tree, page_start, page_end,
7592 &cached_state, GFP_NOFS);
7593 unlock_page(page);
7594 btrfs_start_ordered_extent(inode, ordered, 1);
7595 btrfs_put_ordered_extent(ordered);
7596 goto again;
7597 }
7598
7599 /*
7600 * XXX - page_mkwrite gets called every time the page is dirtied, even
7601 * if it was already dirty, so for space accounting reasons we need to
7602 * clear any delalloc bits for the range we are fixing to save. There
7603 * is probably a better way to do this, but for now keep consistent with
7604 * prepare_pages in the normal write path.
7605 */
7606 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7607 EXTENT_DIRTY | EXTENT_DELALLOC |
7608 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7609 0, 0, &cached_state, GFP_NOFS);
7610
7611 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7612 &cached_state);
7613 if (ret) {
7614 unlock_extent_cached(io_tree, page_start, page_end,
7615 &cached_state, GFP_NOFS);
7616 ret = VM_FAULT_SIGBUS;
7617 goto out_unlock;
7618 }
7619 ret = 0;
7620
7621 /* page is wholly or partially inside EOF */
7622 if (page_start + PAGE_CACHE_SIZE > size)
7623 zero_start = size & ~PAGE_CACHE_MASK;
7624 else
7625 zero_start = PAGE_CACHE_SIZE;
7626
7627 if (zero_start != PAGE_CACHE_SIZE) {
7628 kaddr = kmap(page);
7629 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7630 flush_dcache_page(page);
7631 kunmap(page);
7632 }
7633 ClearPageChecked(page);
7634 set_page_dirty(page);
7635 SetPageUptodate(page);
7636
7637 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7638 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7639 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7640
7641 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7642
7643 out_unlock:
7644 if (!ret) {
7645 sb_end_pagefault(inode->i_sb);
7646 return VM_FAULT_LOCKED;
7647 }
7648 unlock_page(page);
7649 out:
7650 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7651 out_noreserve:
7652 sb_end_pagefault(inode->i_sb);
7653 return ret;
7654 }
7655
7656 static int btrfs_truncate(struct inode *inode)
7657 {
7658 struct btrfs_root *root = BTRFS_I(inode)->root;
7659 struct btrfs_block_rsv *rsv;
7660 int ret = 0;
7661 int err = 0;
7662 struct btrfs_trans_handle *trans;
7663 u64 mask = root->sectorsize - 1;
7664 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7665
7666 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7667 (u64)-1);
7668 if (ret)
7669 return ret;
7670
7671 /*
7672 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7673 * 3 things going on here
7674 *
7675 * 1) We need to reserve space for our orphan item and the space to
7676 * delete our orphan item. Lord knows we don't want to have a dangling
7677 * orphan item because we didn't reserve space to remove it.
7678 *
7679 * 2) We need to reserve space to update our inode.
7680 *
7681 * 3) We need to have something to cache all the space that is going to
7682 * be free'd up by the truncate operation, but also have some slack
7683 * space reserved in case it uses space during the truncate (thank you
7684 * very much snapshotting).
7685 *
7686 * And we need these to all be seperate. The fact is we can use alot of
7687 * space doing the truncate, and we have no earthly idea how much space
7688 * we will use, so we need the truncate reservation to be seperate so it
7689 * doesn't end up using space reserved for updating the inode or
7690 * removing the orphan item. We also need to be able to stop the
7691 * transaction and start a new one, which means we need to be able to
7692 * update the inode several times, and we have no idea of knowing how
7693 * many times that will be, so we can't just reserve 1 item for the
7694 * entirety of the opration, so that has to be done seperately as well.
7695 * Then there is the orphan item, which does indeed need to be held on
7696 * to for the whole operation, and we need nobody to touch this reserved
7697 * space except the orphan code.
7698 *
7699 * So that leaves us with
7700 *
7701 * 1) root->orphan_block_rsv - for the orphan deletion.
7702 * 2) rsv - for the truncate reservation, which we will steal from the
7703 * transaction reservation.
7704 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7705 * updating the inode.
7706 */
7707 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7708 if (!rsv)
7709 return -ENOMEM;
7710 rsv->size = min_size;
7711 rsv->failfast = 1;
7712
7713 /*
7714 * 1 for the truncate slack space
7715 * 1 for updating the inode.
7716 */
7717 trans = btrfs_start_transaction(root, 2);
7718 if (IS_ERR(trans)) {
7719 err = PTR_ERR(trans);
7720 goto out;
7721 }
7722
7723 /* Migrate the slack space for the truncate to our reserve */
7724 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7725 min_size);
7726 BUG_ON(ret);
7727
7728 /*
7729 * setattr is responsible for setting the ordered_data_close flag,
7730 * but that is only tested during the last file release. That
7731 * could happen well after the next commit, leaving a great big
7732 * window where new writes may get lost if someone chooses to write
7733 * to this file after truncating to zero
7734 *
7735 * The inode doesn't have any dirty data here, and so if we commit
7736 * this is a noop. If someone immediately starts writing to the inode
7737 * it is very likely we'll catch some of their writes in this
7738 * transaction, and the commit will find this file on the ordered
7739 * data list with good things to send down.
7740 *
7741 * This is a best effort solution, there is still a window where
7742 * using truncate to replace the contents of the file will
7743 * end up with a zero length file after a crash.
7744 */
7745 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7746 &BTRFS_I(inode)->runtime_flags))
7747 btrfs_add_ordered_operation(trans, root, inode);
7748
7749 /*
7750 * So if we truncate and then write and fsync we normally would just
7751 * write the extents that changed, which is a problem if we need to
7752 * first truncate that entire inode. So set this flag so we write out
7753 * all of the extents in the inode to the sync log so we're completely
7754 * safe.
7755 */
7756 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7757 trans->block_rsv = rsv;
7758
7759 while (1) {
7760 ret = btrfs_truncate_inode_items(trans, root, inode,
7761 inode->i_size,
7762 BTRFS_EXTENT_DATA_KEY);
7763 if (ret != -ENOSPC) {
7764 err = ret;
7765 break;
7766 }
7767
7768 trans->block_rsv = &root->fs_info->trans_block_rsv;
7769 ret = btrfs_update_inode(trans, root, inode);
7770 if (ret) {
7771 err = ret;
7772 break;
7773 }
7774
7775 btrfs_end_transaction(trans, root);
7776 btrfs_btree_balance_dirty(root);
7777
7778 trans = btrfs_start_transaction(root, 2);
7779 if (IS_ERR(trans)) {
7780 ret = err = PTR_ERR(trans);
7781 trans = NULL;
7782 break;
7783 }
7784
7785 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7786 rsv, min_size);
7787 BUG_ON(ret); /* shouldn't happen */
7788 trans->block_rsv = rsv;
7789 }
7790
7791 if (ret == 0 && inode->i_nlink > 0) {
7792 trans->block_rsv = root->orphan_block_rsv;
7793 ret = btrfs_orphan_del(trans, inode);
7794 if (ret)
7795 err = ret;
7796 }
7797
7798 if (trans) {
7799 trans->block_rsv = &root->fs_info->trans_block_rsv;
7800 ret = btrfs_update_inode(trans, root, inode);
7801 if (ret && !err)
7802 err = ret;
7803
7804 ret = btrfs_end_transaction(trans, root);
7805 btrfs_btree_balance_dirty(root);
7806 }
7807
7808 out:
7809 btrfs_free_block_rsv(root, rsv);
7810
7811 if (ret && !err)
7812 err = ret;
7813
7814 return err;
7815 }
7816
7817 /*
7818 * create a new subvolume directory/inode (helper for the ioctl).
7819 */
7820 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7821 struct btrfs_root *new_root, u64 new_dirid)
7822 {
7823 struct inode *inode;
7824 int err;
7825 u64 index = 0;
7826
7827 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7828 new_dirid, new_dirid,
7829 S_IFDIR | (~current_umask() & S_IRWXUGO),
7830 &index);
7831 if (IS_ERR(inode))
7832 return PTR_ERR(inode);
7833 inode->i_op = &btrfs_dir_inode_operations;
7834 inode->i_fop = &btrfs_dir_file_operations;
7835
7836 set_nlink(inode, 1);
7837 btrfs_i_size_write(inode, 0);
7838
7839 err = btrfs_update_inode(trans, new_root, inode);
7840
7841 iput(inode);
7842 return err;
7843 }
7844
7845 struct inode *btrfs_alloc_inode(struct super_block *sb)
7846 {
7847 struct btrfs_inode *ei;
7848 struct inode *inode;
7849
7850 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7851 if (!ei)
7852 return NULL;
7853
7854 ei->root = NULL;
7855 ei->generation = 0;
7856 ei->last_trans = 0;
7857 ei->last_sub_trans = 0;
7858 ei->logged_trans = 0;
7859 ei->delalloc_bytes = 0;
7860 ei->disk_i_size = 0;
7861 ei->flags = 0;
7862 ei->csum_bytes = 0;
7863 ei->index_cnt = (u64)-1;
7864 ei->last_unlink_trans = 0;
7865 ei->last_log_commit = 0;
7866
7867 spin_lock_init(&ei->lock);
7868 ei->outstanding_extents = 0;
7869 ei->reserved_extents = 0;
7870
7871 ei->runtime_flags = 0;
7872 ei->force_compress = BTRFS_COMPRESS_NONE;
7873
7874 ei->delayed_node = NULL;
7875
7876 inode = &ei->vfs_inode;
7877 extent_map_tree_init(&ei->extent_tree);
7878 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7879 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7880 ei->io_tree.track_uptodate = 1;
7881 ei->io_failure_tree.track_uptodate = 1;
7882 atomic_set(&ei->sync_writers, 0);
7883 mutex_init(&ei->log_mutex);
7884 mutex_init(&ei->delalloc_mutex);
7885 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7886 INIT_LIST_HEAD(&ei->delalloc_inodes);
7887 INIT_LIST_HEAD(&ei->ordered_operations);
7888 RB_CLEAR_NODE(&ei->rb_node);
7889
7890 return inode;
7891 }
7892
7893 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7894 void btrfs_test_destroy_inode(struct inode *inode)
7895 {
7896 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7897 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7898 }
7899 #endif
7900
7901 static void btrfs_i_callback(struct rcu_head *head)
7902 {
7903 struct inode *inode = container_of(head, struct inode, i_rcu);
7904 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7905 }
7906
7907 void btrfs_destroy_inode(struct inode *inode)
7908 {
7909 struct btrfs_ordered_extent *ordered;
7910 struct btrfs_root *root = BTRFS_I(inode)->root;
7911
7912 WARN_ON(!hlist_empty(&inode->i_dentry));
7913 WARN_ON(inode->i_data.nrpages);
7914 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7915 WARN_ON(BTRFS_I(inode)->reserved_extents);
7916 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7917 WARN_ON(BTRFS_I(inode)->csum_bytes);
7918
7919 /*
7920 * This can happen where we create an inode, but somebody else also
7921 * created the same inode and we need to destroy the one we already
7922 * created.
7923 */
7924 if (!root)
7925 goto free;
7926
7927 /*
7928 * Make sure we're properly removed from the ordered operation
7929 * lists.
7930 */
7931 smp_mb();
7932 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7933 spin_lock(&root->fs_info->ordered_root_lock);
7934 list_del_init(&BTRFS_I(inode)->ordered_operations);
7935 spin_unlock(&root->fs_info->ordered_root_lock);
7936 }
7937
7938 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7939 &BTRFS_I(inode)->runtime_flags)) {
7940 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7941 btrfs_ino(inode));
7942 atomic_dec(&root->orphan_inodes);
7943 }
7944
7945 while (1) {
7946 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7947 if (!ordered)
7948 break;
7949 else {
7950 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7951 ordered->file_offset, ordered->len);
7952 btrfs_remove_ordered_extent(inode, ordered);
7953 btrfs_put_ordered_extent(ordered);
7954 btrfs_put_ordered_extent(ordered);
7955 }
7956 }
7957 inode_tree_del(inode);
7958 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7959 free:
7960 call_rcu(&inode->i_rcu, btrfs_i_callback);
7961 }
7962
7963 int btrfs_drop_inode(struct inode *inode)
7964 {
7965 struct btrfs_root *root = BTRFS_I(inode)->root;
7966
7967 if (root == NULL)
7968 return 1;
7969
7970 /* the snap/subvol tree is on deleting */
7971 if (btrfs_root_refs(&root->root_item) == 0)
7972 return 1;
7973 else
7974 return generic_drop_inode(inode);
7975 }
7976
7977 static void init_once(void *foo)
7978 {
7979 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7980
7981 inode_init_once(&ei->vfs_inode);
7982 }
7983
7984 void btrfs_destroy_cachep(void)
7985 {
7986 /*
7987 * Make sure all delayed rcu free inodes are flushed before we
7988 * destroy cache.
7989 */
7990 rcu_barrier();
7991 if (btrfs_inode_cachep)
7992 kmem_cache_destroy(btrfs_inode_cachep);
7993 if (btrfs_trans_handle_cachep)
7994 kmem_cache_destroy(btrfs_trans_handle_cachep);
7995 if (btrfs_transaction_cachep)
7996 kmem_cache_destroy(btrfs_transaction_cachep);
7997 if (btrfs_path_cachep)
7998 kmem_cache_destroy(btrfs_path_cachep);
7999 if (btrfs_free_space_cachep)
8000 kmem_cache_destroy(btrfs_free_space_cachep);
8001 if (btrfs_delalloc_work_cachep)
8002 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8003 }
8004
8005 int btrfs_init_cachep(void)
8006 {
8007 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8008 sizeof(struct btrfs_inode), 0,
8009 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8010 if (!btrfs_inode_cachep)
8011 goto fail;
8012
8013 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8014 sizeof(struct btrfs_trans_handle), 0,
8015 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8016 if (!btrfs_trans_handle_cachep)
8017 goto fail;
8018
8019 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8020 sizeof(struct btrfs_transaction), 0,
8021 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8022 if (!btrfs_transaction_cachep)
8023 goto fail;
8024
8025 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8026 sizeof(struct btrfs_path), 0,
8027 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8028 if (!btrfs_path_cachep)
8029 goto fail;
8030
8031 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8032 sizeof(struct btrfs_free_space), 0,
8033 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8034 if (!btrfs_free_space_cachep)
8035 goto fail;
8036
8037 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8038 sizeof(struct btrfs_delalloc_work), 0,
8039 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8040 NULL);
8041 if (!btrfs_delalloc_work_cachep)
8042 goto fail;
8043
8044 return 0;
8045 fail:
8046 btrfs_destroy_cachep();
8047 return -ENOMEM;
8048 }
8049
8050 static int btrfs_getattr(struct vfsmount *mnt,
8051 struct dentry *dentry, struct kstat *stat)
8052 {
8053 u64 delalloc_bytes;
8054 struct inode *inode = dentry->d_inode;
8055 u32 blocksize = inode->i_sb->s_blocksize;
8056
8057 generic_fillattr(inode, stat);
8058 stat->dev = BTRFS_I(inode)->root->anon_dev;
8059 stat->blksize = PAGE_CACHE_SIZE;
8060
8061 spin_lock(&BTRFS_I(inode)->lock);
8062 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8063 spin_unlock(&BTRFS_I(inode)->lock);
8064 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8065 ALIGN(delalloc_bytes, blocksize)) >> 9;
8066 return 0;
8067 }
8068
8069 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8070 struct inode *new_dir, struct dentry *new_dentry)
8071 {
8072 struct btrfs_trans_handle *trans;
8073 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8074 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8075 struct inode *new_inode = new_dentry->d_inode;
8076 struct inode *old_inode = old_dentry->d_inode;
8077 struct timespec ctime = CURRENT_TIME;
8078 u64 index = 0;
8079 u64 root_objectid;
8080 int ret;
8081 u64 old_ino = btrfs_ino(old_inode);
8082
8083 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8084 return -EPERM;
8085
8086 /* we only allow rename subvolume link between subvolumes */
8087 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8088 return -EXDEV;
8089
8090 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8091 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8092 return -ENOTEMPTY;
8093
8094 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8095 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8096 return -ENOTEMPTY;
8097
8098
8099 /* check for collisions, even if the name isn't there */
8100 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8101 new_dentry->d_name.name,
8102 new_dentry->d_name.len);
8103
8104 if (ret) {
8105 if (ret == -EEXIST) {
8106 /* we shouldn't get
8107 * eexist without a new_inode */
8108 if (WARN_ON(!new_inode)) {
8109 return ret;
8110 }
8111 } else {
8112 /* maybe -EOVERFLOW */
8113 return ret;
8114 }
8115 }
8116 ret = 0;
8117
8118 /*
8119 * we're using rename to replace one file with another.
8120 * and the replacement file is large. Start IO on it now so
8121 * we don't add too much work to the end of the transaction
8122 */
8123 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8124 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8125 filemap_flush(old_inode->i_mapping);
8126
8127 /* close the racy window with snapshot create/destroy ioctl */
8128 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8129 down_read(&root->fs_info->subvol_sem);
8130 /*
8131 * We want to reserve the absolute worst case amount of items. So if
8132 * both inodes are subvols and we need to unlink them then that would
8133 * require 4 item modifications, but if they are both normal inodes it
8134 * would require 5 item modifications, so we'll assume their normal
8135 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8136 * should cover the worst case number of items we'll modify.
8137 */
8138 trans = btrfs_start_transaction(root, 11);
8139 if (IS_ERR(trans)) {
8140 ret = PTR_ERR(trans);
8141 goto out_notrans;
8142 }
8143
8144 if (dest != root)
8145 btrfs_record_root_in_trans(trans, dest);
8146
8147 ret = btrfs_set_inode_index(new_dir, &index);
8148 if (ret)
8149 goto out_fail;
8150
8151 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8152 /* force full log commit if subvolume involved. */
8153 root->fs_info->last_trans_log_full_commit = trans->transid;
8154 } else {
8155 ret = btrfs_insert_inode_ref(trans, dest,
8156 new_dentry->d_name.name,
8157 new_dentry->d_name.len,
8158 old_ino,
8159 btrfs_ino(new_dir), index);
8160 if (ret)
8161 goto out_fail;
8162 /*
8163 * this is an ugly little race, but the rename is required
8164 * to make sure that if we crash, the inode is either at the
8165 * old name or the new one. pinning the log transaction lets
8166 * us make sure we don't allow a log commit to come in after
8167 * we unlink the name but before we add the new name back in.
8168 */
8169 btrfs_pin_log_trans(root);
8170 }
8171 /*
8172 * make sure the inode gets flushed if it is replacing
8173 * something.
8174 */
8175 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8176 btrfs_add_ordered_operation(trans, root, old_inode);
8177
8178 inode_inc_iversion(old_dir);
8179 inode_inc_iversion(new_dir);
8180 inode_inc_iversion(old_inode);
8181 old_dir->i_ctime = old_dir->i_mtime = ctime;
8182 new_dir->i_ctime = new_dir->i_mtime = ctime;
8183 old_inode->i_ctime = ctime;
8184
8185 if (old_dentry->d_parent != new_dentry->d_parent)
8186 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8187
8188 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8189 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8190 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8191 old_dentry->d_name.name,
8192 old_dentry->d_name.len);
8193 } else {
8194 ret = __btrfs_unlink_inode(trans, root, old_dir,
8195 old_dentry->d_inode,
8196 old_dentry->d_name.name,
8197 old_dentry->d_name.len);
8198 if (!ret)
8199 ret = btrfs_update_inode(trans, root, old_inode);
8200 }
8201 if (ret) {
8202 btrfs_abort_transaction(trans, root, ret);
8203 goto out_fail;
8204 }
8205
8206 if (new_inode) {
8207 inode_inc_iversion(new_inode);
8208 new_inode->i_ctime = CURRENT_TIME;
8209 if (unlikely(btrfs_ino(new_inode) ==
8210 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8211 root_objectid = BTRFS_I(new_inode)->location.objectid;
8212 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8213 root_objectid,
8214 new_dentry->d_name.name,
8215 new_dentry->d_name.len);
8216 BUG_ON(new_inode->i_nlink == 0);
8217 } else {
8218 ret = btrfs_unlink_inode(trans, dest, new_dir,
8219 new_dentry->d_inode,
8220 new_dentry->d_name.name,
8221 new_dentry->d_name.len);
8222 }
8223 if (!ret && new_inode->i_nlink == 0)
8224 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8225 if (ret) {
8226 btrfs_abort_transaction(trans, root, ret);
8227 goto out_fail;
8228 }
8229 }
8230
8231 ret = btrfs_add_link(trans, new_dir, old_inode,
8232 new_dentry->d_name.name,
8233 new_dentry->d_name.len, 0, index);
8234 if (ret) {
8235 btrfs_abort_transaction(trans, root, ret);
8236 goto out_fail;
8237 }
8238
8239 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8240 struct dentry *parent = new_dentry->d_parent;
8241 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8242 btrfs_end_log_trans(root);
8243 }
8244 out_fail:
8245 btrfs_end_transaction(trans, root);
8246 out_notrans:
8247 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8248 up_read(&root->fs_info->subvol_sem);
8249
8250 return ret;
8251 }
8252
8253 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8254 {
8255 struct btrfs_delalloc_work *delalloc_work;
8256 struct inode *inode;
8257
8258 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8259 work);
8260 inode = delalloc_work->inode;
8261 if (delalloc_work->wait) {
8262 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8263 } else {
8264 filemap_flush(inode->i_mapping);
8265 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8266 &BTRFS_I(inode)->runtime_flags))
8267 filemap_flush(inode->i_mapping);
8268 }
8269
8270 if (delalloc_work->delay_iput)
8271 btrfs_add_delayed_iput(inode);
8272 else
8273 iput(inode);
8274 complete(&delalloc_work->completion);
8275 }
8276
8277 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8278 int wait, int delay_iput)
8279 {
8280 struct btrfs_delalloc_work *work;
8281
8282 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8283 if (!work)
8284 return NULL;
8285
8286 init_completion(&work->completion);
8287 INIT_LIST_HEAD(&work->list);
8288 work->inode = inode;
8289 work->wait = wait;
8290 work->delay_iput = delay_iput;
8291 work->work.func = btrfs_run_delalloc_work;
8292
8293 return work;
8294 }
8295
8296 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8297 {
8298 wait_for_completion(&work->completion);
8299 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8300 }
8301
8302 /*
8303 * some fairly slow code that needs optimization. This walks the list
8304 * of all the inodes with pending delalloc and forces them to disk.
8305 */
8306 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8307 {
8308 struct btrfs_inode *binode;
8309 struct inode *inode;
8310 struct btrfs_delalloc_work *work, *next;
8311 struct list_head works;
8312 struct list_head splice;
8313 int ret = 0;
8314
8315 INIT_LIST_HEAD(&works);
8316 INIT_LIST_HEAD(&splice);
8317
8318 spin_lock(&root->delalloc_lock);
8319 list_splice_init(&root->delalloc_inodes, &splice);
8320 while (!list_empty(&splice)) {
8321 binode = list_entry(splice.next, struct btrfs_inode,
8322 delalloc_inodes);
8323
8324 list_move_tail(&binode->delalloc_inodes,
8325 &root->delalloc_inodes);
8326 inode = igrab(&binode->vfs_inode);
8327 if (!inode) {
8328 cond_resched_lock(&root->delalloc_lock);
8329 continue;
8330 }
8331 spin_unlock(&root->delalloc_lock);
8332
8333 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8334 if (unlikely(!work)) {
8335 if (delay_iput)
8336 btrfs_add_delayed_iput(inode);
8337 else
8338 iput(inode);
8339 ret = -ENOMEM;
8340 goto out;
8341 }
8342 list_add_tail(&work->list, &works);
8343 btrfs_queue_worker(&root->fs_info->flush_workers,
8344 &work->work);
8345
8346 cond_resched();
8347 spin_lock(&root->delalloc_lock);
8348 }
8349 spin_unlock(&root->delalloc_lock);
8350
8351 list_for_each_entry_safe(work, next, &works, list) {
8352 list_del_init(&work->list);
8353 btrfs_wait_and_free_delalloc_work(work);
8354 }
8355 return 0;
8356 out:
8357 list_for_each_entry_safe(work, next, &works, list) {
8358 list_del_init(&work->list);
8359 btrfs_wait_and_free_delalloc_work(work);
8360 }
8361
8362 if (!list_empty_careful(&splice)) {
8363 spin_lock(&root->delalloc_lock);
8364 list_splice_tail(&splice, &root->delalloc_inodes);
8365 spin_unlock(&root->delalloc_lock);
8366 }
8367 return ret;
8368 }
8369
8370 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8371 {
8372 int ret;
8373
8374 if (root->fs_info->sb->s_flags & MS_RDONLY)
8375 return -EROFS;
8376
8377 ret = __start_delalloc_inodes(root, delay_iput);
8378 /*
8379 * the filemap_flush will queue IO into the worker threads, but
8380 * we have to make sure the IO is actually started and that
8381 * ordered extents get created before we return
8382 */
8383 atomic_inc(&root->fs_info->async_submit_draining);
8384 while (atomic_read(&root->fs_info->nr_async_submits) ||
8385 atomic_read(&root->fs_info->async_delalloc_pages)) {
8386 wait_event(root->fs_info->async_submit_wait,
8387 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8388 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8389 }
8390 atomic_dec(&root->fs_info->async_submit_draining);
8391 return ret;
8392 }
8393
8394 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8395 {
8396 struct btrfs_root *root;
8397 struct list_head splice;
8398 int ret;
8399
8400 if (fs_info->sb->s_flags & MS_RDONLY)
8401 return -EROFS;
8402
8403 INIT_LIST_HEAD(&splice);
8404
8405 spin_lock(&fs_info->delalloc_root_lock);
8406 list_splice_init(&fs_info->delalloc_roots, &splice);
8407 while (!list_empty(&splice)) {
8408 root = list_first_entry(&splice, struct btrfs_root,
8409 delalloc_root);
8410 root = btrfs_grab_fs_root(root);
8411 BUG_ON(!root);
8412 list_move_tail(&root->delalloc_root,
8413 &fs_info->delalloc_roots);
8414 spin_unlock(&fs_info->delalloc_root_lock);
8415
8416 ret = __start_delalloc_inodes(root, delay_iput);
8417 btrfs_put_fs_root(root);
8418 if (ret)
8419 goto out;
8420
8421 spin_lock(&fs_info->delalloc_root_lock);
8422 }
8423 spin_unlock(&fs_info->delalloc_root_lock);
8424
8425 atomic_inc(&fs_info->async_submit_draining);
8426 while (atomic_read(&fs_info->nr_async_submits) ||
8427 atomic_read(&fs_info->async_delalloc_pages)) {
8428 wait_event(fs_info->async_submit_wait,
8429 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8430 atomic_read(&fs_info->async_delalloc_pages) == 0));
8431 }
8432 atomic_dec(&fs_info->async_submit_draining);
8433 return 0;
8434 out:
8435 if (!list_empty_careful(&splice)) {
8436 spin_lock(&fs_info->delalloc_root_lock);
8437 list_splice_tail(&splice, &fs_info->delalloc_roots);
8438 spin_unlock(&fs_info->delalloc_root_lock);
8439 }
8440 return ret;
8441 }
8442
8443 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8444 const char *symname)
8445 {
8446 struct btrfs_trans_handle *trans;
8447 struct btrfs_root *root = BTRFS_I(dir)->root;
8448 struct btrfs_path *path;
8449 struct btrfs_key key;
8450 struct inode *inode = NULL;
8451 int err;
8452 int drop_inode = 0;
8453 u64 objectid;
8454 u64 index = 0;
8455 int name_len;
8456 int datasize;
8457 unsigned long ptr;
8458 struct btrfs_file_extent_item *ei;
8459 struct extent_buffer *leaf;
8460
8461 name_len = strlen(symname);
8462 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8463 return -ENAMETOOLONG;
8464
8465 /*
8466 * 2 items for inode item and ref
8467 * 2 items for dir items
8468 * 1 item for xattr if selinux is on
8469 */
8470 trans = btrfs_start_transaction(root, 5);
8471 if (IS_ERR(trans))
8472 return PTR_ERR(trans);
8473
8474 err = btrfs_find_free_ino(root, &objectid);
8475 if (err)
8476 goto out_unlock;
8477
8478 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8479 dentry->d_name.len, btrfs_ino(dir), objectid,
8480 S_IFLNK|S_IRWXUGO, &index);
8481 if (IS_ERR(inode)) {
8482 err = PTR_ERR(inode);
8483 goto out_unlock;
8484 }
8485
8486 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8487 if (err) {
8488 drop_inode = 1;
8489 goto out_unlock;
8490 }
8491
8492 /*
8493 * If the active LSM wants to access the inode during
8494 * d_instantiate it needs these. Smack checks to see
8495 * if the filesystem supports xattrs by looking at the
8496 * ops vector.
8497 */
8498 inode->i_fop = &btrfs_file_operations;
8499 inode->i_op = &btrfs_file_inode_operations;
8500
8501 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8502 if (err)
8503 drop_inode = 1;
8504 else {
8505 inode->i_mapping->a_ops = &btrfs_aops;
8506 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8507 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8508 }
8509 if (drop_inode)
8510 goto out_unlock;
8511
8512 path = btrfs_alloc_path();
8513 if (!path) {
8514 err = -ENOMEM;
8515 drop_inode = 1;
8516 goto out_unlock;
8517 }
8518 key.objectid = btrfs_ino(inode);
8519 key.offset = 0;
8520 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8521 datasize = btrfs_file_extent_calc_inline_size(name_len);
8522 err = btrfs_insert_empty_item(trans, root, path, &key,
8523 datasize);
8524 if (err) {
8525 drop_inode = 1;
8526 btrfs_free_path(path);
8527 goto out_unlock;
8528 }
8529 leaf = path->nodes[0];
8530 ei = btrfs_item_ptr(leaf, path->slots[0],
8531 struct btrfs_file_extent_item);
8532 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8533 btrfs_set_file_extent_type(leaf, ei,
8534 BTRFS_FILE_EXTENT_INLINE);
8535 btrfs_set_file_extent_encryption(leaf, ei, 0);
8536 btrfs_set_file_extent_compression(leaf, ei, 0);
8537 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8538 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8539
8540 ptr = btrfs_file_extent_inline_start(ei);
8541 write_extent_buffer(leaf, symname, ptr, name_len);
8542 btrfs_mark_buffer_dirty(leaf);
8543 btrfs_free_path(path);
8544
8545 inode->i_op = &btrfs_symlink_inode_operations;
8546 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8547 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8548 inode_set_bytes(inode, name_len);
8549 btrfs_i_size_write(inode, name_len);
8550 err = btrfs_update_inode(trans, root, inode);
8551 if (err)
8552 drop_inode = 1;
8553
8554 out_unlock:
8555 if (!err)
8556 d_instantiate(dentry, inode);
8557 btrfs_end_transaction(trans, root);
8558 if (drop_inode) {
8559 inode_dec_link_count(inode);
8560 iput(inode);
8561 }
8562 btrfs_btree_balance_dirty(root);
8563 return err;
8564 }
8565
8566 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8567 u64 start, u64 num_bytes, u64 min_size,
8568 loff_t actual_len, u64 *alloc_hint,
8569 struct btrfs_trans_handle *trans)
8570 {
8571 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8572 struct extent_map *em;
8573 struct btrfs_root *root = BTRFS_I(inode)->root;
8574 struct btrfs_key ins;
8575 u64 cur_offset = start;
8576 u64 i_size;
8577 u64 cur_bytes;
8578 int ret = 0;
8579 bool own_trans = true;
8580
8581 if (trans)
8582 own_trans = false;
8583 while (num_bytes > 0) {
8584 if (own_trans) {
8585 trans = btrfs_start_transaction(root, 3);
8586 if (IS_ERR(trans)) {
8587 ret = PTR_ERR(trans);
8588 break;
8589 }
8590 }
8591
8592 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8593 cur_bytes = max(cur_bytes, min_size);
8594 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8595 *alloc_hint, &ins, 1);
8596 if (ret) {
8597 if (own_trans)
8598 btrfs_end_transaction(trans, root);
8599 break;
8600 }
8601
8602 ret = insert_reserved_file_extent(trans, inode,
8603 cur_offset, ins.objectid,
8604 ins.offset, ins.offset,
8605 ins.offset, 0, 0, 0,
8606 BTRFS_FILE_EXTENT_PREALLOC);
8607 if (ret) {
8608 btrfs_free_reserved_extent(root, ins.objectid,
8609 ins.offset);
8610 btrfs_abort_transaction(trans, root, ret);
8611 if (own_trans)
8612 btrfs_end_transaction(trans, root);
8613 break;
8614 }
8615 btrfs_drop_extent_cache(inode, cur_offset,
8616 cur_offset + ins.offset -1, 0);
8617
8618 em = alloc_extent_map();
8619 if (!em) {
8620 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8621 &BTRFS_I(inode)->runtime_flags);
8622 goto next;
8623 }
8624
8625 em->start = cur_offset;
8626 em->orig_start = cur_offset;
8627 em->len = ins.offset;
8628 em->block_start = ins.objectid;
8629 em->block_len = ins.offset;
8630 em->orig_block_len = ins.offset;
8631 em->ram_bytes = ins.offset;
8632 em->bdev = root->fs_info->fs_devices->latest_bdev;
8633 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8634 em->generation = trans->transid;
8635
8636 while (1) {
8637 write_lock(&em_tree->lock);
8638 ret = add_extent_mapping(em_tree, em, 1);
8639 write_unlock(&em_tree->lock);
8640 if (ret != -EEXIST)
8641 break;
8642 btrfs_drop_extent_cache(inode, cur_offset,
8643 cur_offset + ins.offset - 1,
8644 0);
8645 }
8646 free_extent_map(em);
8647 next:
8648 num_bytes -= ins.offset;
8649 cur_offset += ins.offset;
8650 *alloc_hint = ins.objectid + ins.offset;
8651
8652 inode_inc_iversion(inode);
8653 inode->i_ctime = CURRENT_TIME;
8654 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8655 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8656 (actual_len > inode->i_size) &&
8657 (cur_offset > inode->i_size)) {
8658 if (cur_offset > actual_len)
8659 i_size = actual_len;
8660 else
8661 i_size = cur_offset;
8662 i_size_write(inode, i_size);
8663 btrfs_ordered_update_i_size(inode, i_size, NULL);
8664 }
8665
8666 ret = btrfs_update_inode(trans, root, inode);
8667
8668 if (ret) {
8669 btrfs_abort_transaction(trans, root, ret);
8670 if (own_trans)
8671 btrfs_end_transaction(trans, root);
8672 break;
8673 }
8674
8675 if (own_trans)
8676 btrfs_end_transaction(trans, root);
8677 }
8678 return ret;
8679 }
8680
8681 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8682 u64 start, u64 num_bytes, u64 min_size,
8683 loff_t actual_len, u64 *alloc_hint)
8684 {
8685 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8686 min_size, actual_len, alloc_hint,
8687 NULL);
8688 }
8689
8690 int btrfs_prealloc_file_range_trans(struct inode *inode,
8691 struct btrfs_trans_handle *trans, int mode,
8692 u64 start, u64 num_bytes, u64 min_size,
8693 loff_t actual_len, u64 *alloc_hint)
8694 {
8695 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8696 min_size, actual_len, alloc_hint, trans);
8697 }
8698
8699 static int btrfs_set_page_dirty(struct page *page)
8700 {
8701 return __set_page_dirty_nobuffers(page);
8702 }
8703
8704 static int btrfs_permission(struct inode *inode, int mask)
8705 {
8706 struct btrfs_root *root = BTRFS_I(inode)->root;
8707 umode_t mode = inode->i_mode;
8708
8709 if (mask & MAY_WRITE &&
8710 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8711 if (btrfs_root_readonly(root))
8712 return -EROFS;
8713 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8714 return -EACCES;
8715 }
8716 return generic_permission(inode, mask);
8717 }
8718
8719 static const struct inode_operations btrfs_dir_inode_operations = {
8720 .getattr = btrfs_getattr,
8721 .lookup = btrfs_lookup,
8722 .create = btrfs_create,
8723 .unlink = btrfs_unlink,
8724 .link = btrfs_link,
8725 .mkdir = btrfs_mkdir,
8726 .rmdir = btrfs_rmdir,
8727 .rename = btrfs_rename,
8728 .symlink = btrfs_symlink,
8729 .setattr = btrfs_setattr,
8730 .mknod = btrfs_mknod,
8731 .setxattr = btrfs_setxattr,
8732 .getxattr = btrfs_getxattr,
8733 .listxattr = btrfs_listxattr,
8734 .removexattr = btrfs_removexattr,
8735 .permission = btrfs_permission,
8736 .get_acl = btrfs_get_acl,
8737 .update_time = btrfs_update_time,
8738 };
8739 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8740 .lookup = btrfs_lookup,
8741 .permission = btrfs_permission,
8742 .get_acl = btrfs_get_acl,
8743 .update_time = btrfs_update_time,
8744 };
8745
8746 static const struct file_operations btrfs_dir_file_operations = {
8747 .llseek = generic_file_llseek,
8748 .read = generic_read_dir,
8749 .iterate = btrfs_real_readdir,
8750 .unlocked_ioctl = btrfs_ioctl,
8751 #ifdef CONFIG_COMPAT
8752 .compat_ioctl = btrfs_ioctl,
8753 #endif
8754 .release = btrfs_release_file,
8755 .fsync = btrfs_sync_file,
8756 };
8757
8758 static struct extent_io_ops btrfs_extent_io_ops = {
8759 .fill_delalloc = run_delalloc_range,
8760 .submit_bio_hook = btrfs_submit_bio_hook,
8761 .merge_bio_hook = btrfs_merge_bio_hook,
8762 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8763 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8764 .writepage_start_hook = btrfs_writepage_start_hook,
8765 .set_bit_hook = btrfs_set_bit_hook,
8766 .clear_bit_hook = btrfs_clear_bit_hook,
8767 .merge_extent_hook = btrfs_merge_extent_hook,
8768 .split_extent_hook = btrfs_split_extent_hook,
8769 };
8770
8771 /*
8772 * btrfs doesn't support the bmap operation because swapfiles
8773 * use bmap to make a mapping of extents in the file. They assume
8774 * these extents won't change over the life of the file and they
8775 * use the bmap result to do IO directly to the drive.
8776 *
8777 * the btrfs bmap call would return logical addresses that aren't
8778 * suitable for IO and they also will change frequently as COW
8779 * operations happen. So, swapfile + btrfs == corruption.
8780 *
8781 * For now we're avoiding this by dropping bmap.
8782 */
8783 static const struct address_space_operations btrfs_aops = {
8784 .readpage = btrfs_readpage,
8785 .writepage = btrfs_writepage,
8786 .writepages = btrfs_writepages,
8787 .readpages = btrfs_readpages,
8788 .direct_IO = btrfs_direct_IO,
8789 .invalidatepage = btrfs_invalidatepage,
8790 .releasepage = btrfs_releasepage,
8791 .set_page_dirty = btrfs_set_page_dirty,
8792 .error_remove_page = generic_error_remove_page,
8793 };
8794
8795 static const struct address_space_operations btrfs_symlink_aops = {
8796 .readpage = btrfs_readpage,
8797 .writepage = btrfs_writepage,
8798 .invalidatepage = btrfs_invalidatepage,
8799 .releasepage = btrfs_releasepage,
8800 };
8801
8802 static const struct inode_operations btrfs_file_inode_operations = {
8803 .getattr = btrfs_getattr,
8804 .setattr = btrfs_setattr,
8805 .setxattr = btrfs_setxattr,
8806 .getxattr = btrfs_getxattr,
8807 .listxattr = btrfs_listxattr,
8808 .removexattr = btrfs_removexattr,
8809 .permission = btrfs_permission,
8810 .fiemap = btrfs_fiemap,
8811 .get_acl = btrfs_get_acl,
8812 .update_time = btrfs_update_time,
8813 };
8814 static const struct inode_operations btrfs_special_inode_operations = {
8815 .getattr = btrfs_getattr,
8816 .setattr = btrfs_setattr,
8817 .permission = btrfs_permission,
8818 .setxattr = btrfs_setxattr,
8819 .getxattr = btrfs_getxattr,
8820 .listxattr = btrfs_listxattr,
8821 .removexattr = btrfs_removexattr,
8822 .get_acl = btrfs_get_acl,
8823 .update_time = btrfs_update_time,
8824 };
8825 static const struct inode_operations btrfs_symlink_inode_operations = {
8826 .readlink = generic_readlink,
8827 .follow_link = page_follow_link_light,
8828 .put_link = page_put_link,
8829 .getattr = btrfs_getattr,
8830 .setattr = btrfs_setattr,
8831 .permission = btrfs_permission,
8832 .setxattr = btrfs_setxattr,
8833 .getxattr = btrfs_getxattr,
8834 .listxattr = btrfs_listxattr,
8835 .removexattr = btrfs_removexattr,
8836 .get_acl = btrfs_get_acl,
8837 .update_time = btrfs_update_time,
8838 };
8839
8840 const struct dentry_operations btrfs_dentry_operations = {
8841 .d_delete = btrfs_dentry_delete,
8842 .d_release = btrfs_dentry_release,
8843 };
(deprecated) Btrfs devel tree