]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/inode.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'samsung-fixes-4.6' of git://git.kernel.org/pub/scm/linux/kernel/git/krzk...
[mirror_ubuntu-artful-kernel.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61 #include "props.h"
62 #include "qgroup.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74 };
75
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
85
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
91
92 #define S_SHIFT 12
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101 };
102
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
113 u64 orig_block_len, u64 ram_bytes,
114 int type);
115
116 static int btrfs_dirty_inode(struct inode *inode);
117
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
120 {
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
122 }
123 #endif
124
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
128 {
129 int err;
130
131 err = btrfs_init_acl(trans, inode, dir);
132 if (!err)
133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
134 return err;
135 }
136
137 /*
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
141 */
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143 struct btrfs_path *path, int extent_inserted,
144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
146 int compress_type,
147 struct page **compressed_pages)
148 {
149 struct extent_buffer *leaf;
150 struct page *page = NULL;
151 char *kaddr;
152 unsigned long ptr;
153 struct btrfs_file_extent_item *ei;
154 int err = 0;
155 int ret;
156 size_t cur_size = size;
157 unsigned long offset;
158
159 if (compressed_size && compressed_pages)
160 cur_size = compressed_size;
161
162 inode_add_bytes(inode, size);
163
164 if (!extent_inserted) {
165 struct btrfs_key key;
166 size_t datasize;
167
168 key.objectid = btrfs_ino(inode);
169 key.offset = start;
170 key.type = BTRFS_EXTENT_DATA_KEY;
171
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 datasize);
176 if (ret) {
177 err = ret;
178 goto fail;
179 }
180 }
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
190
191 if (compress_type != BTRFS_COMPRESS_NONE) {
192 struct page *cpage;
193 int i = 0;
194 while (compressed_size > 0) {
195 cpage = compressed_pages[i];
196 cur_size = min_t(unsigned long, compressed_size,
197 PAGE_SIZE);
198
199 kaddr = kmap_atomic(cpage);
200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
201 kunmap_atomic(kaddr);
202
203 i++;
204 ptr += cur_size;
205 compressed_size -= cur_size;
206 }
207 btrfs_set_file_extent_compression(leaf, ei,
208 compress_type);
209 } else {
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
217 put_page(page);
218 }
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
221
222 /*
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
226 *
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
230 */
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
233
234 return ret;
235 fail:
236 return err;
237 }
238
239
240 /*
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
244 */
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
248 int compress_type,
249 struct page **compressed_pages)
250 {
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
257 int ret;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
261
262 if (compressed_size)
263 data_len = compressed_size;
264
265 if (start > 0 ||
266 actual_end > root->sectorsize ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
268 (!compressed_size &&
269 (actual_end & (root->sectorsize - 1)) == 0) ||
270 end + 1 < isize ||
271 data_len > root->fs_info->max_inline) {
272 return 1;
273 }
274
275 path = btrfs_alloc_path();
276 if (!path)
277 return -ENOMEM;
278
279 trans = btrfs_join_transaction(root);
280 if (IS_ERR(trans)) {
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
283 }
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
285
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
288 compressed_size);
289 else
290 extent_item_size = btrfs_file_extent_calc_inline_size(
291 inline_len);
292
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
296 if (ret) {
297 btrfs_abort_transaction(trans, root, ret);
298 goto out;
299 }
300
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
304 root, inode, start,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
309 goto out;
310 } else if (ret == -ENOSPC) {
311 ret = 1;
312 goto out;
313 }
314
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
318 out:
319 /*
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
324 */
325 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
326 btrfs_free_path(path);
327 btrfs_end_transaction(trans, root);
328 return ret;
329 }
330
331 struct async_extent {
332 u64 start;
333 u64 ram_size;
334 u64 compressed_size;
335 struct page **pages;
336 unsigned long nr_pages;
337 int compress_type;
338 struct list_head list;
339 };
340
341 struct async_cow {
342 struct inode *inode;
343 struct btrfs_root *root;
344 struct page *locked_page;
345 u64 start;
346 u64 end;
347 struct list_head extents;
348 struct btrfs_work work;
349 };
350
351 static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
353 u64 compressed_size,
354 struct page **pages,
355 unsigned long nr_pages,
356 int compress_type)
357 {
358 struct async_extent *async_extent;
359
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361 BUG_ON(!async_extent); /* -ENOMEM */
362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
367 async_extent->compress_type = compress_type;
368 list_add_tail(&async_extent->list, &cow->extents);
369 return 0;
370 }
371
372 static inline int inode_need_compress(struct inode *inode)
373 {
374 struct btrfs_root *root = BTRFS_I(inode)->root;
375
376 /* force compress */
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
378 return 1;
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
381 return 0;
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
385 return 1;
386 return 0;
387 }
388
389 /*
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
393 *
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
399 *
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
404 * down.
405 */
406 static noinline void compress_file_range(struct inode *inode,
407 struct page *locked_page,
408 u64 start, u64 end,
409 struct async_cow *async_cow,
410 int *num_added)
411 {
412 struct btrfs_root *root = BTRFS_I(inode)->root;
413 u64 num_bytes;
414 u64 blocksize = root->sectorsize;
415 u64 actual_end;
416 u64 isize = i_size_read(inode);
417 int ret = 0;
418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
425 int i;
426 int will_compress;
427 int compress_type = root->fs_info->compress_type;
428 int redirty = 0;
429
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end - start + 1) < SZ_16K &&
432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
434
435 actual_end = min_t(u64, isize, end + 1);
436 again:
437 will_compress = 0;
438 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
440
441 /*
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
447 *
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
450 */
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
453
454 total_compressed = actual_end - start;
455
456 /*
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
459 */
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
463
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
469 *
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
473 */
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
477 total_in = 0;
478 ret = 0;
479
480 /*
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
484 */
485 if (inode_need_compress(inode)) {
486 WARN_ON(pages);
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
488 if (!pages) {
489 /* just bail out to the uncompressed code */
490 goto cont;
491 }
492
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
495
496 /*
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
501 *
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
504 */
505 extent_range_clear_dirty_for_io(inode, start, end);
506 redirty = 1;
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
511 &total_in,
512 &total_compressed,
513 max_compressed);
514
515 if (!ret) {
516 unsigned long offset = total_compressed &
517 (PAGE_SIZE - 1);
518 struct page *page = pages[nr_pages_ret - 1];
519 char *kaddr;
520
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
523 */
524 if (offset) {
525 kaddr = kmap_atomic(page);
526 memset(kaddr + offset, 0,
527 PAGE_SIZE - offset);
528 kunmap_atomic(kaddr);
529 }
530 will_compress = 1;
531 }
532 }
533 cont:
534 if (start == 0) {
535 /* lets try to make an inline extent */
536 if (ret || total_in < (actual_end - start)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
539 */
540 ret = cow_file_range_inline(root, inode, start, end,
541 0, 0, NULL);
542 } else {
543 /* try making a compressed inline extent */
544 ret = cow_file_range_inline(root, inode, start, end,
545 total_compressed,
546 compress_type, pages);
547 }
548 if (ret <= 0) {
549 unsigned long clear_flags = EXTENT_DELALLOC |
550 EXTENT_DEFRAG;
551 unsigned long page_error_op;
552
553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
555
556 /*
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
560 */
561 extent_clear_unlock_delalloc(inode, start, end, NULL,
562 clear_flags, PAGE_UNLOCK |
563 PAGE_CLEAR_DIRTY |
564 PAGE_SET_WRITEBACK |
565 page_error_op |
566 PAGE_END_WRITEBACK);
567 goto free_pages_out;
568 }
569 }
570
571 if (will_compress) {
572 /*
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
575 * things
576 */
577 total_compressed = ALIGN(total_compressed, blocksize);
578
579 /*
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
582 */
583 total_in = ALIGN(total_in, PAGE_SIZE);
584 if (total_compressed >= total_in) {
585 will_compress = 0;
586 } else {
587 num_bytes = total_in;
588 }
589 }
590 if (!will_compress && pages) {
591 /*
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
594 */
595 for (i = 0; i < nr_pages_ret; i++) {
596 WARN_ON(pages[i]->mapping);
597 put_page(pages[i]);
598 }
599 kfree(pages);
600 pages = NULL;
601 total_compressed = 0;
602 nr_pages_ret = 0;
603
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
608 }
609 }
610 if (will_compress) {
611 *num_added += 1;
612
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
616 */
617 add_async_extent(async_cow, start, num_bytes,
618 total_compressed, pages, nr_pages_ret,
619 compress_type);
620
621 if (start + num_bytes < end) {
622 start += num_bytes;
623 pages = NULL;
624 cond_resched();
625 goto again;
626 }
627 } else {
628 cleanup_and_bail_uncompressed:
629 /*
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
635 */
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
640 }
641 if (redirty)
642 extent_range_redirty_for_io(inode, start, end);
643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
645 *num_added += 1;
646 }
647
648 return;
649
650 free_pages_out:
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
653 put_page(pages[i]);
654 }
655 kfree(pages);
656 }
657
658 static void free_async_extent_pages(struct async_extent *async_extent)
659 {
660 int i;
661
662 if (!async_extent->pages)
663 return;
664
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 put_page(async_extent->pages[i]);
668 }
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
672 }
673
674 /*
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
679 */
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
682 {
683 struct async_extent *async_extent;
684 u64 alloc_hint = 0;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
690 int ret = 0;
691
692 again:
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
697
698 io_tree = &BTRFS_I(inode)->io_tree;
699
700 retry:
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
705
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
709
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
712 async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
716
717 /* JDM XXX */
718
719 /*
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
724 */
725 if (!page_started && !ret)
726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1,
730 btrfs_get_extent,
731 WB_SYNC_ALL);
732 else if (ret)
733 unlock_page(async_cow->locked_page);
734 kfree(async_extent);
735 cond_resched();
736 continue;
737 }
738
739 lock_extent(io_tree, async_extent->start,
740 async_extent->start + async_extent->ram_size - 1);
741
742 ret = btrfs_reserve_extent(root,
743 async_extent->compressed_size,
744 async_extent->compressed_size,
745 0, alloc_hint, &ins, 1, 1);
746 if (ret) {
747 free_async_extent_pages(async_extent);
748
749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
753
754 /*
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
758 * layers.
759 */
760 extent_range_redirty_for_io(inode,
761 async_extent->start,
762 async_extent->start +
763 async_extent->ram_size - 1);
764
765 goto retry;
766 }
767 goto out_free;
768 }
769 /*
770 * here we're doing allocation and writeback of the
771 * compressed pages
772 */
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
776
777 em = alloc_extent_map();
778 if (!em) {
779 ret = -ENOMEM;
780 goto out_free_reserve;
781 }
782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
784 em->orig_start = em->start;
785 em->mod_start = em->start;
786 em->mod_len = em->len;
787
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
790 em->orig_block_len = ins.offset;
791 em->ram_bytes = async_extent->ram_size;
792 em->bdev = root->fs_info->fs_devices->latest_bdev;
793 em->compress_type = async_extent->compress_type;
794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
796 em->generation = -1;
797
798 while (1) {
799 write_lock(&em_tree->lock);
800 ret = add_extent_mapping(em_tree, em, 1);
801 write_unlock(&em_tree->lock);
802 if (ret != -EEXIST) {
803 free_extent_map(em);
804 break;
805 }
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
809 }
810
811 if (ret)
812 goto out_free_reserve;
813
814 ret = btrfs_add_ordered_extent_compress(inode,
815 async_extent->start,
816 ins.objectid,
817 async_extent->ram_size,
818 ins.offset,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
821 if (ret) {
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
825 goto out_free_reserve;
826 }
827
828 /*
829 * clear dirty, set writeback and unlock the pages.
830 */
831 extent_clear_unlock_delalloc(inode, async_extent->start,
832 async_extent->start +
833 async_extent->ram_size - 1,
834 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
836 PAGE_SET_WRITEBACK);
837 ret = btrfs_submit_compressed_write(inode,
838 async_extent->start,
839 async_extent->ram_size,
840 ins.objectid,
841 ins.offset, async_extent->pages,
842 async_extent->nr_pages);
843 if (ret) {
844 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845 struct page *p = async_extent->pages[0];
846 const u64 start = async_extent->start;
847 const u64 end = start + async_extent->ram_size - 1;
848
849 p->mapping = inode->i_mapping;
850 tree->ops->writepage_end_io_hook(p, start, end,
851 NULL, 0);
852 p->mapping = NULL;
853 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
854 PAGE_END_WRITEBACK |
855 PAGE_SET_ERROR);
856 free_async_extent_pages(async_extent);
857 }
858 alloc_hint = ins.objectid + ins.offset;
859 kfree(async_extent);
860 cond_resched();
861 }
862 return;
863 out_free_reserve:
864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
865 out_free:
866 extent_clear_unlock_delalloc(inode, async_extent->start,
867 async_extent->start +
868 async_extent->ram_size - 1,
869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
873 PAGE_SET_ERROR);
874 free_async_extent_pages(async_extent);
875 kfree(async_extent);
876 goto again;
877 }
878
879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
880 u64 num_bytes)
881 {
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
884 u64 alloc_hint = 0;
885
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
888 if (em) {
889 /*
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
893 */
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
895 free_extent_map(em);
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
899 if (em)
900 free_extent_map(em);
901 } else {
902 alloc_hint = em->block_start;
903 free_extent_map(em);
904 }
905 }
906 read_unlock(&em_tree->lock);
907
908 return alloc_hint;
909 }
910
911 /*
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
916 *
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
919 *
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
922 * IO when we return.
923 */
924 static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
928 int unlock)
929 {
930 struct btrfs_root *root = BTRFS_I(inode)->root;
931 u64 alloc_hint = 0;
932 u64 num_bytes;
933 unsigned long ram_size;
934 u64 disk_num_bytes;
935 u64 cur_alloc_size;
936 u64 blocksize = root->sectorsize;
937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
940 int ret = 0;
941
942 if (btrfs_is_free_space_inode(inode)) {
943 WARN_ON_ONCE(1);
944 ret = -EINVAL;
945 goto out_unlock;
946 }
947
948 num_bytes = ALIGN(end - start + 1, blocksize);
949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
951
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes < SZ_64K &&
954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
955 btrfs_add_inode_defrag(NULL, inode);
956
957 if (start == 0) {
958 /* lets try to make an inline extent */
959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
960 NULL);
961 if (ret == 0) {
962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
964 EXTENT_DEFRAG, PAGE_UNLOCK |
965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
966 PAGE_END_WRITEBACK);
967
968 *nr_written = *nr_written +
969 (end - start + PAGE_SIZE) / PAGE_SIZE;
970 *page_started = 1;
971 goto out;
972 } else if (ret < 0) {
973 goto out_unlock;
974 }
975 }
976
977 BUG_ON(disk_num_bytes >
978 btrfs_super_total_bytes(root->fs_info->super_copy));
979
980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
982
983 while (disk_num_bytes > 0) {
984 unsigned long op;
985
986 cur_alloc_size = disk_num_bytes;
987 ret = btrfs_reserve_extent(root, cur_alloc_size,
988 root->sectorsize, 0, alloc_hint,
989 &ins, 1, 1);
990 if (ret < 0)
991 goto out_unlock;
992
993 em = alloc_extent_map();
994 if (!em) {
995 ret = -ENOMEM;
996 goto out_reserve;
997 }
998 em->start = start;
999 em->orig_start = em->start;
1000 ram_size = ins.offset;
1001 em->len = ins.offset;
1002 em->mod_start = em->start;
1003 em->mod_len = em->len;
1004
1005 em->block_start = ins.objectid;
1006 em->block_len = ins.offset;
1007 em->orig_block_len = ins.offset;
1008 em->ram_bytes = ram_size;
1009 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011 em->generation = -1;
1012
1013 while (1) {
1014 write_lock(&em_tree->lock);
1015 ret = add_extent_mapping(em_tree, em, 1);
1016 write_unlock(&em_tree->lock);
1017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1019 break;
1020 }
1021 btrfs_drop_extent_cache(inode, start,
1022 start + ram_size - 1, 0);
1023 }
1024 if (ret)
1025 goto out_reserve;
1026
1027 cur_alloc_size = ins.offset;
1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029 ram_size, cur_alloc_size, 0);
1030 if (ret)
1031 goto out_drop_extent_cache;
1032
1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1036 cur_alloc_size);
1037 if (ret)
1038 goto out_drop_extent_cache;
1039 }
1040
1041 if (disk_num_bytes < cur_alloc_size)
1042 break;
1043
1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
1047 *
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
1050 */
1051 op = unlock ? PAGE_UNLOCK : 0;
1052 op |= PAGE_SET_PRIVATE2;
1053
1054 extent_clear_unlock_delalloc(inode, start,
1055 start + ram_size - 1, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1057 op);
1058 disk_num_bytes -= cur_alloc_size;
1059 num_bytes -= cur_alloc_size;
1060 alloc_hint = ins.objectid + ins.offset;
1061 start += cur_alloc_size;
1062 }
1063 out:
1064 return ret;
1065
1066 out_drop_extent_cache:
1067 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1068 out_reserve:
1069 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1070 out_unlock:
1071 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073 EXTENT_DELALLOC | EXTENT_DEFRAG,
1074 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1076 goto out;
1077 }
1078
1079 /*
1080 * work queue call back to started compression on a file and pages
1081 */
1082 static noinline void async_cow_start(struct btrfs_work *work)
1083 {
1084 struct async_cow *async_cow;
1085 int num_added = 0;
1086 async_cow = container_of(work, struct async_cow, work);
1087
1088 compress_file_range(async_cow->inode, async_cow->locked_page,
1089 async_cow->start, async_cow->end, async_cow,
1090 &num_added);
1091 if (num_added == 0) {
1092 btrfs_add_delayed_iput(async_cow->inode);
1093 async_cow->inode = NULL;
1094 }
1095 }
1096
1097 /*
1098 * work queue call back to submit previously compressed pages
1099 */
1100 static noinline void async_cow_submit(struct btrfs_work *work)
1101 {
1102 struct async_cow *async_cow;
1103 struct btrfs_root *root;
1104 unsigned long nr_pages;
1105
1106 async_cow = container_of(work, struct async_cow, work);
1107
1108 root = async_cow->root;
1109 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1110 PAGE_SHIFT;
1111
1112 /*
1113 * atomic_sub_return implies a barrier for waitqueue_active
1114 */
1115 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1116 5 * SZ_1M &&
1117 waitqueue_active(&root->fs_info->async_submit_wait))
1118 wake_up(&root->fs_info->async_submit_wait);
1119
1120 if (async_cow->inode)
1121 submit_compressed_extents(async_cow->inode, async_cow);
1122 }
1123
1124 static noinline void async_cow_free(struct btrfs_work *work)
1125 {
1126 struct async_cow *async_cow;
1127 async_cow = container_of(work, struct async_cow, work);
1128 if (async_cow->inode)
1129 btrfs_add_delayed_iput(async_cow->inode);
1130 kfree(async_cow);
1131 }
1132
1133 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1136 {
1137 struct async_cow *async_cow;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 unsigned long nr_pages;
1140 u64 cur_end;
1141 int limit = 10 * SZ_1M;
1142
1143 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144 1, 0, NULL, GFP_NOFS);
1145 while (start < end) {
1146 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1147 BUG_ON(!async_cow); /* -ENOMEM */
1148 async_cow->inode = igrab(inode);
1149 async_cow->root = root;
1150 async_cow->locked_page = locked_page;
1151 async_cow->start = start;
1152
1153 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154 !btrfs_test_opt(root, FORCE_COMPRESS))
1155 cur_end = end;
1156 else
1157 cur_end = min(end, start + SZ_512K - 1);
1158
1159 async_cow->end = cur_end;
1160 INIT_LIST_HEAD(&async_cow->extents);
1161
1162 btrfs_init_work(&async_cow->work,
1163 btrfs_delalloc_helper,
1164 async_cow_start, async_cow_submit,
1165 async_cow_free);
1166
1167 nr_pages = (cur_end - start + PAGE_SIZE) >>
1168 PAGE_SHIFT;
1169 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1170
1171 btrfs_queue_work(root->fs_info->delalloc_workers,
1172 &async_cow->work);
1173
1174 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175 wait_event(root->fs_info->async_submit_wait,
1176 (atomic_read(&root->fs_info->async_delalloc_pages) <
1177 limit));
1178 }
1179
1180 while (atomic_read(&root->fs_info->async_submit_draining) &&
1181 atomic_read(&root->fs_info->async_delalloc_pages)) {
1182 wait_event(root->fs_info->async_submit_wait,
1183 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1184 0));
1185 }
1186
1187 *nr_written += nr_pages;
1188 start = cur_end + 1;
1189 }
1190 *page_started = 1;
1191 return 0;
1192 }
1193
1194 static noinline int csum_exist_in_range(struct btrfs_root *root,
1195 u64 bytenr, u64 num_bytes)
1196 {
1197 int ret;
1198 struct btrfs_ordered_sum *sums;
1199 LIST_HEAD(list);
1200
1201 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1202 bytenr + num_bytes - 1, &list, 0);
1203 if (ret == 0 && list_empty(&list))
1204 return 0;
1205
1206 while (!list_empty(&list)) {
1207 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208 list_del(&sums->list);
1209 kfree(sums);
1210 }
1211 return 1;
1212 }
1213
1214 /*
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1217 *
1218 * If no cow copies or snapshots exist, we write directly to the existing
1219 * blocks on disk
1220 */
1221 static noinline int run_delalloc_nocow(struct inode *inode,
1222 struct page *locked_page,
1223 u64 start, u64 end, int *page_started, int force,
1224 unsigned long *nr_written)
1225 {
1226 struct btrfs_root *root = BTRFS_I(inode)->root;
1227 struct btrfs_trans_handle *trans;
1228 struct extent_buffer *leaf;
1229 struct btrfs_path *path;
1230 struct btrfs_file_extent_item *fi;
1231 struct btrfs_key found_key;
1232 u64 cow_start;
1233 u64 cur_offset;
1234 u64 extent_end;
1235 u64 extent_offset;
1236 u64 disk_bytenr;
1237 u64 num_bytes;
1238 u64 disk_num_bytes;
1239 u64 ram_bytes;
1240 int extent_type;
1241 int ret, err;
1242 int type;
1243 int nocow;
1244 int check_prev = 1;
1245 bool nolock;
1246 u64 ino = btrfs_ino(inode);
1247
1248 path = btrfs_alloc_path();
1249 if (!path) {
1250 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251 EXTENT_LOCKED | EXTENT_DELALLOC |
1252 EXTENT_DO_ACCOUNTING |
1253 EXTENT_DEFRAG, PAGE_UNLOCK |
1254 PAGE_CLEAR_DIRTY |
1255 PAGE_SET_WRITEBACK |
1256 PAGE_END_WRITEBACK);
1257 return -ENOMEM;
1258 }
1259
1260 nolock = btrfs_is_free_space_inode(inode);
1261
1262 if (nolock)
1263 trans = btrfs_join_transaction_nolock(root);
1264 else
1265 trans = btrfs_join_transaction(root);
1266
1267 if (IS_ERR(trans)) {
1268 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269 EXTENT_LOCKED | EXTENT_DELALLOC |
1270 EXTENT_DO_ACCOUNTING |
1271 EXTENT_DEFRAG, PAGE_UNLOCK |
1272 PAGE_CLEAR_DIRTY |
1273 PAGE_SET_WRITEBACK |
1274 PAGE_END_WRITEBACK);
1275 btrfs_free_path(path);
1276 return PTR_ERR(trans);
1277 }
1278
1279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1280
1281 cow_start = (u64)-1;
1282 cur_offset = start;
1283 while (1) {
1284 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1285 cur_offset, 0);
1286 if (ret < 0)
1287 goto error;
1288 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key,
1291 path->slots[0] - 1);
1292 if (found_key.objectid == ino &&
1293 found_key.type == BTRFS_EXTENT_DATA_KEY)
1294 path->slots[0]--;
1295 }
1296 check_prev = 0;
1297 next_slot:
1298 leaf = path->nodes[0];
1299 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300 ret = btrfs_next_leaf(root, path);
1301 if (ret < 0)
1302 goto error;
1303 if (ret > 0)
1304 break;
1305 leaf = path->nodes[0];
1306 }
1307
1308 nocow = 0;
1309 disk_bytenr = 0;
1310 num_bytes = 0;
1311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1312
1313 if (found_key.objectid > ino)
1314 break;
1315 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1317 path->slots[0]++;
1318 goto next_slot;
1319 }
1320 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1321 found_key.offset > end)
1322 break;
1323
1324 if (found_key.offset > cur_offset) {
1325 extent_end = found_key.offset;
1326 extent_type = 0;
1327 goto out_check;
1328 }
1329
1330 fi = btrfs_item_ptr(leaf, path->slots[0],
1331 struct btrfs_file_extent_item);
1332 extent_type = btrfs_file_extent_type(leaf, fi);
1333
1334 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1335 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1338 extent_offset = btrfs_file_extent_offset(leaf, fi);
1339 extent_end = found_key.offset +
1340 btrfs_file_extent_num_bytes(leaf, fi);
1341 disk_num_bytes =
1342 btrfs_file_extent_disk_num_bytes(leaf, fi);
1343 if (extent_end <= start) {
1344 path->slots[0]++;
1345 goto next_slot;
1346 }
1347 if (disk_bytenr == 0)
1348 goto out_check;
1349 if (btrfs_file_extent_compression(leaf, fi) ||
1350 btrfs_file_extent_encryption(leaf, fi) ||
1351 btrfs_file_extent_other_encoding(leaf, fi))
1352 goto out_check;
1353 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1354 goto out_check;
1355 if (btrfs_extent_readonly(root, disk_bytenr))
1356 goto out_check;
1357 if (btrfs_cross_ref_exist(trans, root, ino,
1358 found_key.offset -
1359 extent_offset, disk_bytenr))
1360 goto out_check;
1361 disk_bytenr += extent_offset;
1362 disk_bytenr += cur_offset - found_key.offset;
1363 num_bytes = min(end + 1, extent_end) - cur_offset;
1364 /*
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1367 */
1368 if (!nolock) {
1369 err = btrfs_start_write_no_snapshoting(root);
1370 if (!err)
1371 goto out_check;
1372 }
1373 /*
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1377 */
1378 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1379 goto out_check;
1380 nocow = 1;
1381 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382 extent_end = found_key.offset +
1383 btrfs_file_extent_inline_len(leaf,
1384 path->slots[0], fi);
1385 extent_end = ALIGN(extent_end, root->sectorsize);
1386 } else {
1387 BUG_ON(1);
1388 }
1389 out_check:
1390 if (extent_end <= start) {
1391 path->slots[0]++;
1392 if (!nolock && nocow)
1393 btrfs_end_write_no_snapshoting(root);
1394 goto next_slot;
1395 }
1396 if (!nocow) {
1397 if (cow_start == (u64)-1)
1398 cow_start = cur_offset;
1399 cur_offset = extent_end;
1400 if (cur_offset > end)
1401 break;
1402 path->slots[0]++;
1403 goto next_slot;
1404 }
1405
1406 btrfs_release_path(path);
1407 if (cow_start != (u64)-1) {
1408 ret = cow_file_range(inode, locked_page,
1409 cow_start, found_key.offset - 1,
1410 page_started, nr_written, 1);
1411 if (ret) {
1412 if (!nolock && nocow)
1413 btrfs_end_write_no_snapshoting(root);
1414 goto error;
1415 }
1416 cow_start = (u64)-1;
1417 }
1418
1419 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420 struct extent_map *em;
1421 struct extent_map_tree *em_tree;
1422 em_tree = &BTRFS_I(inode)->extent_tree;
1423 em = alloc_extent_map();
1424 BUG_ON(!em); /* -ENOMEM */
1425 em->start = cur_offset;
1426 em->orig_start = found_key.offset - extent_offset;
1427 em->len = num_bytes;
1428 em->block_len = num_bytes;
1429 em->block_start = disk_bytenr;
1430 em->orig_block_len = disk_num_bytes;
1431 em->ram_bytes = ram_bytes;
1432 em->bdev = root->fs_info->fs_devices->latest_bdev;
1433 em->mod_start = em->start;
1434 em->mod_len = em->len;
1435 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1436 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1437 em->generation = -1;
1438 while (1) {
1439 write_lock(&em_tree->lock);
1440 ret = add_extent_mapping(em_tree, em, 1);
1441 write_unlock(&em_tree->lock);
1442 if (ret != -EEXIST) {
1443 free_extent_map(em);
1444 break;
1445 }
1446 btrfs_drop_extent_cache(inode, em->start,
1447 em->start + em->len - 1, 0);
1448 }
1449 type = BTRFS_ORDERED_PREALLOC;
1450 } else {
1451 type = BTRFS_ORDERED_NOCOW;
1452 }
1453
1454 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1455 num_bytes, num_bytes, type);
1456 BUG_ON(ret); /* -ENOMEM */
1457
1458 if (root->root_key.objectid ==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1461 num_bytes);
1462 if (ret) {
1463 if (!nolock && nocow)
1464 btrfs_end_write_no_snapshoting(root);
1465 goto error;
1466 }
1467 }
1468
1469 extent_clear_unlock_delalloc(inode, cur_offset,
1470 cur_offset + num_bytes - 1,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC, PAGE_UNLOCK |
1473 PAGE_SET_PRIVATE2);
1474 if (!nolock && nocow)
1475 btrfs_end_write_no_snapshoting(root);
1476 cur_offset = extent_end;
1477 if (cur_offset > end)
1478 break;
1479 }
1480 btrfs_release_path(path);
1481
1482 if (cur_offset <= end && cow_start == (u64)-1) {
1483 cow_start = cur_offset;
1484 cur_offset = end;
1485 }
1486
1487 if (cow_start != (u64)-1) {
1488 ret = cow_file_range(inode, locked_page, cow_start, end,
1489 page_started, nr_written, 1);
1490 if (ret)
1491 goto error;
1492 }
1493
1494 error:
1495 err = btrfs_end_transaction(trans, root);
1496 if (!ret)
1497 ret = err;
1498
1499 if (ret && cur_offset < end)
1500 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501 locked_page, EXTENT_LOCKED |
1502 EXTENT_DELALLOC | EXTENT_DEFRAG |
1503 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1504 PAGE_CLEAR_DIRTY |
1505 PAGE_SET_WRITEBACK |
1506 PAGE_END_WRITEBACK);
1507 btrfs_free_path(path);
1508 return ret;
1509 }
1510
1511 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1512 {
1513
1514 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1516 return 0;
1517
1518 /*
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1522 */
1523 if (BTRFS_I(inode)->defrag_bytes &&
1524 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525 EXTENT_DEFRAG, 0, NULL))
1526 return 1;
1527
1528 return 0;
1529 }
1530
1531 /*
1532 * extent_io.c call back to do delayed allocation processing
1533 */
1534 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1535 u64 start, u64 end, int *page_started,
1536 unsigned long *nr_written)
1537 {
1538 int ret;
1539 int force_cow = need_force_cow(inode, start, end);
1540
1541 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1542 ret = run_delalloc_nocow(inode, locked_page, start, end,
1543 page_started, 1, nr_written);
1544 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1545 ret = run_delalloc_nocow(inode, locked_page, start, end,
1546 page_started, 0, nr_written);
1547 } else if (!inode_need_compress(inode)) {
1548 ret = cow_file_range(inode, locked_page, start, end,
1549 page_started, nr_written, 1);
1550 } else {
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 &BTRFS_I(inode)->runtime_flags);
1553 ret = cow_file_range_async(inode, locked_page, start, end,
1554 page_started, nr_written);
1555 }
1556 return ret;
1557 }
1558
1559 static void btrfs_split_extent_hook(struct inode *inode,
1560 struct extent_state *orig, u64 split)
1561 {
1562 u64 size;
1563
1564 /* not delalloc, ignore it */
1565 if (!(orig->state & EXTENT_DELALLOC))
1566 return;
1567
1568 size = orig->end - orig->start + 1;
1569 if (size > BTRFS_MAX_EXTENT_SIZE) {
1570 u64 num_extents;
1571 u64 new_size;
1572
1573 /*
1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
1576 */
1577 new_size = orig->end - split + 1;
1578 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1579 BTRFS_MAX_EXTENT_SIZE);
1580 new_size = split - orig->start;
1581 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582 BTRFS_MAX_EXTENT_SIZE);
1583 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1585 return;
1586 }
1587
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
1591 }
1592
1593 /*
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1598 */
1599 static void btrfs_merge_extent_hook(struct inode *inode,
1600 struct extent_state *new,
1601 struct extent_state *other)
1602 {
1603 u64 new_size, old_size;
1604 u64 num_extents;
1605
1606 /* not delalloc, ignore it */
1607 if (!(other->state & EXTENT_DELALLOC))
1608 return;
1609
1610 if (new->start > other->start)
1611 new_size = new->end - other->start + 1;
1612 else
1613 new_size = other->end - new->start + 1;
1614
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617 spin_lock(&BTRFS_I(inode)->lock);
1618 BTRFS_I(inode)->outstanding_extents--;
1619 spin_unlock(&BTRFS_I(inode)->lock);
1620 return;
1621 }
1622
1623 /*
1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1628 *
1629 * [ 4k][MAX_SIZE]
1630 *
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1634 *
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1636 *
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1639 * this case.
1640 */
1641 old_size = other->end - other->start + 1;
1642 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE);
1644 old_size = new->end - new->start + 1;
1645 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646 BTRFS_MAX_EXTENT_SIZE);
1647
1648 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1649 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1650 return;
1651
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
1655 }
1656
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658 struct inode *inode)
1659 {
1660 spin_lock(&root->delalloc_lock);
1661 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663 &root->delalloc_inodes);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665 &BTRFS_I(inode)->runtime_flags);
1666 root->nr_delalloc_inodes++;
1667 if (root->nr_delalloc_inodes == 1) {
1668 spin_lock(&root->fs_info->delalloc_root_lock);
1669 BUG_ON(!list_empty(&root->delalloc_root));
1670 list_add_tail(&root->delalloc_root,
1671 &root->fs_info->delalloc_roots);
1672 spin_unlock(&root->fs_info->delalloc_root_lock);
1673 }
1674 }
1675 spin_unlock(&root->delalloc_lock);
1676 }
1677
1678 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679 struct inode *inode)
1680 {
1681 spin_lock(&root->delalloc_lock);
1682 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes--;
1687 if (!root->nr_delalloc_inodes) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(list_empty(&root->delalloc_root));
1690 list_del_init(&root->delalloc_root);
1691 spin_unlock(&root->fs_info->delalloc_root_lock);
1692 }
1693 }
1694 spin_unlock(&root->delalloc_lock);
1695 }
1696
1697 /*
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1701 */
1702 static void btrfs_set_bit_hook(struct inode *inode,
1703 struct extent_state *state, unsigned *bits)
1704 {
1705
1706 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1707 WARN_ON(1);
1708 /*
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1712 */
1713 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714 struct btrfs_root *root = BTRFS_I(inode)->root;
1715 u64 len = state->end + 1 - state->start;
1716 bool do_list = !btrfs_is_free_space_inode(inode);
1717
1718 if (*bits & EXTENT_FIRST_DELALLOC) {
1719 *bits &= ~EXTENT_FIRST_DELALLOC;
1720 } else {
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 BTRFS_I(inode)->outstanding_extents++;
1723 spin_unlock(&BTRFS_I(inode)->lock);
1724 }
1725
1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root))
1728 return;
1729
1730 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731 root->fs_info->delalloc_batch);
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 BTRFS_I(inode)->delalloc_bytes += len;
1734 if (*bits & EXTENT_DEFRAG)
1735 BTRFS_I(inode)->defrag_bytes += len;
1736 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &BTRFS_I(inode)->runtime_flags))
1738 btrfs_add_delalloc_inodes(root, inode);
1739 spin_unlock(&BTRFS_I(inode)->lock);
1740 }
1741 }
1742
1743 /*
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1745 */
1746 static void btrfs_clear_bit_hook(struct inode *inode,
1747 struct extent_state *state,
1748 unsigned *bits)
1749 {
1750 u64 len = state->end + 1 - state->start;
1751 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752 BTRFS_MAX_EXTENT_SIZE);
1753
1754 spin_lock(&BTRFS_I(inode)->lock);
1755 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756 BTRFS_I(inode)->defrag_bytes -= len;
1757 spin_unlock(&BTRFS_I(inode)->lock);
1758
1759 /*
1760 * set_bit and clear bit hooks normally require _irqsave/restore
1761 * but in this case, we are only testing for the DELALLOC
1762 * bit, which is only set or cleared with irqs on
1763 */
1764 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1765 struct btrfs_root *root = BTRFS_I(inode)->root;
1766 bool do_list = !btrfs_is_free_space_inode(inode);
1767
1768 if (*bits & EXTENT_FIRST_DELALLOC) {
1769 *bits &= ~EXTENT_FIRST_DELALLOC;
1770 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771 spin_lock(&BTRFS_I(inode)->lock);
1772 BTRFS_I(inode)->outstanding_extents -= num_extents;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1774 }
1775
1776 /*
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1779 * error.
1780 */
1781 if (*bits & EXTENT_DO_ACCOUNTING &&
1782 root != root->fs_info->tree_root)
1783 btrfs_delalloc_release_metadata(inode, len);
1784
1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root))
1787 return;
1788
1789 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1790 && do_list && !(state->state & EXTENT_NORESERVE))
1791 btrfs_free_reserved_data_space_noquota(inode,
1792 state->start, len);
1793
1794 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795 root->fs_info->delalloc_batch);
1796 spin_lock(&BTRFS_I(inode)->lock);
1797 BTRFS_I(inode)->delalloc_bytes -= len;
1798 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1800 &BTRFS_I(inode)->runtime_flags))
1801 btrfs_del_delalloc_inode(root, inode);
1802 spin_unlock(&BTRFS_I(inode)->lock);
1803 }
1804 }
1805
1806 /*
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1809 */
1810 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1811 size_t size, struct bio *bio,
1812 unsigned long bio_flags)
1813 {
1814 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1815 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1816 u64 length = 0;
1817 u64 map_length;
1818 int ret;
1819
1820 if (bio_flags & EXTENT_BIO_COMPRESSED)
1821 return 0;
1822
1823 length = bio->bi_iter.bi_size;
1824 map_length = length;
1825 ret = btrfs_map_block(root->fs_info, rw, logical,
1826 &map_length, NULL, 0);
1827 /* Will always return 0 with map_multi == NULL */
1828 BUG_ON(ret < 0);
1829 if (map_length < length + size)
1830 return 1;
1831 return 0;
1832 }
1833
1834 /*
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1838 *
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1841 */
1842 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843 struct bio *bio, int mirror_num,
1844 unsigned long bio_flags,
1845 u64 bio_offset)
1846 {
1847 struct btrfs_root *root = BTRFS_I(inode)->root;
1848 int ret = 0;
1849
1850 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1851 BUG_ON(ret); /* -ENOMEM */
1852 return 0;
1853 }
1854
1855 /*
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1859 *
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1862 */
1863 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1864 int mirror_num, unsigned long bio_flags,
1865 u64 bio_offset)
1866 {
1867 struct btrfs_root *root = BTRFS_I(inode)->root;
1868 int ret;
1869
1870 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1871 if (ret) {
1872 bio->bi_error = ret;
1873 bio_endio(bio);
1874 }
1875 return ret;
1876 }
1877
1878 /*
1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
1881 */
1882 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1883 int mirror_num, unsigned long bio_flags,
1884 u64 bio_offset)
1885 {
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
1887 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1888 int ret = 0;
1889 int skip_sum;
1890 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1891
1892 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1893
1894 if (btrfs_is_free_space_inode(inode))
1895 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1896
1897 if (!(rw & REQ_WRITE)) {
1898 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1899 if (ret)
1900 goto out;
1901
1902 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1903 ret = btrfs_submit_compressed_read(inode, bio,
1904 mirror_num,
1905 bio_flags);
1906 goto out;
1907 } else if (!skip_sum) {
1908 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1909 if (ret)
1910 goto out;
1911 }
1912 goto mapit;
1913 } else if (async && !skip_sum) {
1914 /* csum items have already been cloned */
1915 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1916 goto mapit;
1917 /* we're doing a write, do the async checksumming */
1918 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1919 inode, rw, bio, mirror_num,
1920 bio_flags, bio_offset,
1921 __btrfs_submit_bio_start,
1922 __btrfs_submit_bio_done);
1923 goto out;
1924 } else if (!skip_sum) {
1925 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1926 if (ret)
1927 goto out;
1928 }
1929
1930 mapit:
1931 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1932
1933 out:
1934 if (ret < 0) {
1935 bio->bi_error = ret;
1936 bio_endio(bio);
1937 }
1938 return ret;
1939 }
1940
1941 /*
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1944 */
1945 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1946 struct inode *inode, u64 file_offset,
1947 struct list_head *list)
1948 {
1949 struct btrfs_ordered_sum *sum;
1950
1951 list_for_each_entry(sum, list, list) {
1952 trans->adding_csums = 1;
1953 btrfs_csum_file_blocks(trans,
1954 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1955 trans->adding_csums = 0;
1956 }
1957 return 0;
1958 }
1959
1960 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961 struct extent_state **cached_state)
1962 {
1963 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1965 cached_state, GFP_NOFS);
1966 }
1967
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup {
1970 struct page *page;
1971 struct btrfs_work work;
1972 };
1973
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1975 {
1976 struct btrfs_writepage_fixup *fixup;
1977 struct btrfs_ordered_extent *ordered;
1978 struct extent_state *cached_state = NULL;
1979 struct page *page;
1980 struct inode *inode;
1981 u64 page_start;
1982 u64 page_end;
1983 int ret;
1984
1985 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1986 page = fixup->page;
1987 again:
1988 lock_page(page);
1989 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990 ClearPageChecked(page);
1991 goto out_page;
1992 }
1993
1994 inode = page->mapping->host;
1995 page_start = page_offset(page);
1996 page_end = page_offset(page) + PAGE_SIZE - 1;
1997
1998 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
1999 &cached_state);
2000
2001 /* already ordered? We're done */
2002 if (PagePrivate2(page))
2003 goto out;
2004
2005 ordered = btrfs_lookup_ordered_range(inode, page_start,
2006 PAGE_SIZE);
2007 if (ordered) {
2008 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2009 page_end, &cached_state, GFP_NOFS);
2010 unlock_page(page);
2011 btrfs_start_ordered_extent(inode, ordered, 1);
2012 btrfs_put_ordered_extent(ordered);
2013 goto again;
2014 }
2015
2016 ret = btrfs_delalloc_reserve_space(inode, page_start,
2017 PAGE_SIZE);
2018 if (ret) {
2019 mapping_set_error(page->mapping, ret);
2020 end_extent_writepage(page, ret, page_start, page_end);
2021 ClearPageChecked(page);
2022 goto out;
2023 }
2024
2025 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2026 ClearPageChecked(page);
2027 set_page_dirty(page);
2028 out:
2029 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2030 &cached_state, GFP_NOFS);
2031 out_page:
2032 unlock_page(page);
2033 put_page(page);
2034 kfree(fixup);
2035 }
2036
2037 /*
2038 * There are a few paths in the higher layers of the kernel that directly
2039 * set the page dirty bit without asking the filesystem if it is a
2040 * good idea. This causes problems because we want to make sure COW
2041 * properly happens and the data=ordered rules are followed.
2042 *
2043 * In our case any range that doesn't have the ORDERED bit set
2044 * hasn't been properly setup for IO. We kick off an async process
2045 * to fix it up. The async helper will wait for ordered extents, set
2046 * the delalloc bit and make it safe to write the page.
2047 */
2048 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2049 {
2050 struct inode *inode = page->mapping->host;
2051 struct btrfs_writepage_fixup *fixup;
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
2053
2054 /* this page is properly in the ordered list */
2055 if (TestClearPagePrivate2(page))
2056 return 0;
2057
2058 if (PageChecked(page))
2059 return -EAGAIN;
2060
2061 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2062 if (!fixup)
2063 return -EAGAIN;
2064
2065 SetPageChecked(page);
2066 get_page(page);
2067 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2068 btrfs_writepage_fixup_worker, NULL, NULL);
2069 fixup->page = page;
2070 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2071 return -EBUSY;
2072 }
2073
2074 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2075 struct inode *inode, u64 file_pos,
2076 u64 disk_bytenr, u64 disk_num_bytes,
2077 u64 num_bytes, u64 ram_bytes,
2078 u8 compression, u8 encryption,
2079 u16 other_encoding, int extent_type)
2080 {
2081 struct btrfs_root *root = BTRFS_I(inode)->root;
2082 struct btrfs_file_extent_item *fi;
2083 struct btrfs_path *path;
2084 struct extent_buffer *leaf;
2085 struct btrfs_key ins;
2086 int extent_inserted = 0;
2087 int ret;
2088
2089 path = btrfs_alloc_path();
2090 if (!path)
2091 return -ENOMEM;
2092
2093 /*
2094 * we may be replacing one extent in the tree with another.
2095 * The new extent is pinned in the extent map, and we don't want
2096 * to drop it from the cache until it is completely in the btree.
2097 *
2098 * So, tell btrfs_drop_extents to leave this extent in the cache.
2099 * the caller is expected to unpin it and allow it to be merged
2100 * with the others.
2101 */
2102 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2103 file_pos + num_bytes, NULL, 0,
2104 1, sizeof(*fi), &extent_inserted);
2105 if (ret)
2106 goto out;
2107
2108 if (!extent_inserted) {
2109 ins.objectid = btrfs_ino(inode);
2110 ins.offset = file_pos;
2111 ins.type = BTRFS_EXTENT_DATA_KEY;
2112
2113 path->leave_spinning = 1;
2114 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2115 sizeof(*fi));
2116 if (ret)
2117 goto out;
2118 }
2119 leaf = path->nodes[0];
2120 fi = btrfs_item_ptr(leaf, path->slots[0],
2121 struct btrfs_file_extent_item);
2122 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123 btrfs_set_file_extent_type(leaf, fi, extent_type);
2124 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2125 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2126 btrfs_set_file_extent_offset(leaf, fi, 0);
2127 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2128 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2129 btrfs_set_file_extent_compression(leaf, fi, compression);
2130 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2131 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2132
2133 btrfs_mark_buffer_dirty(leaf);
2134 btrfs_release_path(path);
2135
2136 inode_add_bytes(inode, num_bytes);
2137
2138 ins.objectid = disk_bytenr;
2139 ins.offset = disk_num_bytes;
2140 ins.type = BTRFS_EXTENT_ITEM_KEY;
2141 ret = btrfs_alloc_reserved_file_extent(trans, root,
2142 root->root_key.objectid,
2143 btrfs_ino(inode), file_pos,
2144 ram_bytes, &ins);
2145 /*
2146 * Release the reserved range from inode dirty range map, as it is
2147 * already moved into delayed_ref_head
2148 */
2149 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2150 out:
2151 btrfs_free_path(path);
2152
2153 return ret;
2154 }
2155
2156 /* snapshot-aware defrag */
2157 struct sa_defrag_extent_backref {
2158 struct rb_node node;
2159 struct old_sa_defrag_extent *old;
2160 u64 root_id;
2161 u64 inum;
2162 u64 file_pos;
2163 u64 extent_offset;
2164 u64 num_bytes;
2165 u64 generation;
2166 };
2167
2168 struct old_sa_defrag_extent {
2169 struct list_head list;
2170 struct new_sa_defrag_extent *new;
2171
2172 u64 extent_offset;
2173 u64 bytenr;
2174 u64 offset;
2175 u64 len;
2176 int count;
2177 };
2178
2179 struct new_sa_defrag_extent {
2180 struct rb_root root;
2181 struct list_head head;
2182 struct btrfs_path *path;
2183 struct inode *inode;
2184 u64 file_pos;
2185 u64 len;
2186 u64 bytenr;
2187 u64 disk_len;
2188 u8 compress_type;
2189 };
2190
2191 static int backref_comp(struct sa_defrag_extent_backref *b1,
2192 struct sa_defrag_extent_backref *b2)
2193 {
2194 if (b1->root_id < b2->root_id)
2195 return -1;
2196 else if (b1->root_id > b2->root_id)
2197 return 1;
2198
2199 if (b1->inum < b2->inum)
2200 return -1;
2201 else if (b1->inum > b2->inum)
2202 return 1;
2203
2204 if (b1->file_pos < b2->file_pos)
2205 return -1;
2206 else if (b1->file_pos > b2->file_pos)
2207 return 1;
2208
2209 /*
2210 * [------------------------------] ===> (a range of space)
2211 * |<--->| |<---->| =============> (fs/file tree A)
2212 * |<---------------------------->| ===> (fs/file tree B)
2213 *
2214 * A range of space can refer to two file extents in one tree while
2215 * refer to only one file extent in another tree.
2216 *
2217 * So we may process a disk offset more than one time(two extents in A)
2218 * and locate at the same extent(one extent in B), then insert two same
2219 * backrefs(both refer to the extent in B).
2220 */
2221 return 0;
2222 }
2223
2224 static void backref_insert(struct rb_root *root,
2225 struct sa_defrag_extent_backref *backref)
2226 {
2227 struct rb_node **p = &root->rb_node;
2228 struct rb_node *parent = NULL;
2229 struct sa_defrag_extent_backref *entry;
2230 int ret;
2231
2232 while (*p) {
2233 parent = *p;
2234 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2235
2236 ret = backref_comp(backref, entry);
2237 if (ret < 0)
2238 p = &(*p)->rb_left;
2239 else
2240 p = &(*p)->rb_right;
2241 }
2242
2243 rb_link_node(&backref->node, parent, p);
2244 rb_insert_color(&backref->node, root);
2245 }
2246
2247 /*
2248 * Note the backref might has changed, and in this case we just return 0.
2249 */
2250 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2251 void *ctx)
2252 {
2253 struct btrfs_file_extent_item *extent;
2254 struct btrfs_fs_info *fs_info;
2255 struct old_sa_defrag_extent *old = ctx;
2256 struct new_sa_defrag_extent *new = old->new;
2257 struct btrfs_path *path = new->path;
2258 struct btrfs_key key;
2259 struct btrfs_root *root;
2260 struct sa_defrag_extent_backref *backref;
2261 struct extent_buffer *leaf;
2262 struct inode *inode = new->inode;
2263 int slot;
2264 int ret;
2265 u64 extent_offset;
2266 u64 num_bytes;
2267
2268 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2269 inum == btrfs_ino(inode))
2270 return 0;
2271
2272 key.objectid = root_id;
2273 key.type = BTRFS_ROOT_ITEM_KEY;
2274 key.offset = (u64)-1;
2275
2276 fs_info = BTRFS_I(inode)->root->fs_info;
2277 root = btrfs_read_fs_root_no_name(fs_info, &key);
2278 if (IS_ERR(root)) {
2279 if (PTR_ERR(root) == -ENOENT)
2280 return 0;
2281 WARN_ON(1);
2282 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2283 inum, offset, root_id);
2284 return PTR_ERR(root);
2285 }
2286
2287 key.objectid = inum;
2288 key.type = BTRFS_EXTENT_DATA_KEY;
2289 if (offset > (u64)-1 << 32)
2290 key.offset = 0;
2291 else
2292 key.offset = offset;
2293
2294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2295 if (WARN_ON(ret < 0))
2296 return ret;
2297 ret = 0;
2298
2299 while (1) {
2300 cond_resched();
2301
2302 leaf = path->nodes[0];
2303 slot = path->slots[0];
2304
2305 if (slot >= btrfs_header_nritems(leaf)) {
2306 ret = btrfs_next_leaf(root, path);
2307 if (ret < 0) {
2308 goto out;
2309 } else if (ret > 0) {
2310 ret = 0;
2311 goto out;
2312 }
2313 continue;
2314 }
2315
2316 path->slots[0]++;
2317
2318 btrfs_item_key_to_cpu(leaf, &key, slot);
2319
2320 if (key.objectid > inum)
2321 goto out;
2322
2323 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2324 continue;
2325
2326 extent = btrfs_item_ptr(leaf, slot,
2327 struct btrfs_file_extent_item);
2328
2329 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2330 continue;
2331
2332 /*
2333 * 'offset' refers to the exact key.offset,
2334 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2335 * (key.offset - extent_offset).
2336 */
2337 if (key.offset != offset)
2338 continue;
2339
2340 extent_offset = btrfs_file_extent_offset(leaf, extent);
2341 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2342
2343 if (extent_offset >= old->extent_offset + old->offset +
2344 old->len || extent_offset + num_bytes <=
2345 old->extent_offset + old->offset)
2346 continue;
2347 break;
2348 }
2349
2350 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2351 if (!backref) {
2352 ret = -ENOENT;
2353 goto out;
2354 }
2355
2356 backref->root_id = root_id;
2357 backref->inum = inum;
2358 backref->file_pos = offset;
2359 backref->num_bytes = num_bytes;
2360 backref->extent_offset = extent_offset;
2361 backref->generation = btrfs_file_extent_generation(leaf, extent);
2362 backref->old = old;
2363 backref_insert(&new->root, backref);
2364 old->count++;
2365 out:
2366 btrfs_release_path(path);
2367 WARN_ON(ret);
2368 return ret;
2369 }
2370
2371 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2372 struct new_sa_defrag_extent *new)
2373 {
2374 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2375 struct old_sa_defrag_extent *old, *tmp;
2376 int ret;
2377
2378 new->path = path;
2379
2380 list_for_each_entry_safe(old, tmp, &new->head, list) {
2381 ret = iterate_inodes_from_logical(old->bytenr +
2382 old->extent_offset, fs_info,
2383 path, record_one_backref,
2384 old);
2385 if (ret < 0 && ret != -ENOENT)
2386 return false;
2387
2388 /* no backref to be processed for this extent */
2389 if (!old->count) {
2390 list_del(&old->list);
2391 kfree(old);
2392 }
2393 }
2394
2395 if (list_empty(&new->head))
2396 return false;
2397
2398 return true;
2399 }
2400
2401 static int relink_is_mergable(struct extent_buffer *leaf,
2402 struct btrfs_file_extent_item *fi,
2403 struct new_sa_defrag_extent *new)
2404 {
2405 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2406 return 0;
2407
2408 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2409 return 0;
2410
2411 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2412 return 0;
2413
2414 if (btrfs_file_extent_encryption(leaf, fi) ||
2415 btrfs_file_extent_other_encoding(leaf, fi))
2416 return 0;
2417
2418 return 1;
2419 }
2420
2421 /*
2422 * Note the backref might has changed, and in this case we just return 0.
2423 */
2424 static noinline int relink_extent_backref(struct btrfs_path *path,
2425 struct sa_defrag_extent_backref *prev,
2426 struct sa_defrag_extent_backref *backref)
2427 {
2428 struct btrfs_file_extent_item *extent;
2429 struct btrfs_file_extent_item *item;
2430 struct btrfs_ordered_extent *ordered;
2431 struct btrfs_trans_handle *trans;
2432 struct btrfs_fs_info *fs_info;
2433 struct btrfs_root *root;
2434 struct btrfs_key key;
2435 struct extent_buffer *leaf;
2436 struct old_sa_defrag_extent *old = backref->old;
2437 struct new_sa_defrag_extent *new = old->new;
2438 struct inode *src_inode = new->inode;
2439 struct inode *inode;
2440 struct extent_state *cached = NULL;
2441 int ret = 0;
2442 u64 start;
2443 u64 len;
2444 u64 lock_start;
2445 u64 lock_end;
2446 bool merge = false;
2447 int index;
2448
2449 if (prev && prev->root_id == backref->root_id &&
2450 prev->inum == backref->inum &&
2451 prev->file_pos + prev->num_bytes == backref->file_pos)
2452 merge = true;
2453
2454 /* step 1: get root */
2455 key.objectid = backref->root_id;
2456 key.type = BTRFS_ROOT_ITEM_KEY;
2457 key.offset = (u64)-1;
2458
2459 fs_info = BTRFS_I(src_inode)->root->fs_info;
2460 index = srcu_read_lock(&fs_info->subvol_srcu);
2461
2462 root = btrfs_read_fs_root_no_name(fs_info, &key);
2463 if (IS_ERR(root)) {
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 if (PTR_ERR(root) == -ENOENT)
2466 return 0;
2467 return PTR_ERR(root);
2468 }
2469
2470 if (btrfs_root_readonly(root)) {
2471 srcu_read_unlock(&fs_info->subvol_srcu, index);
2472 return 0;
2473 }
2474
2475 /* step 2: get inode */
2476 key.objectid = backref->inum;
2477 key.type = BTRFS_INODE_ITEM_KEY;
2478 key.offset = 0;
2479
2480 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2481 if (IS_ERR(inode)) {
2482 srcu_read_unlock(&fs_info->subvol_srcu, index);
2483 return 0;
2484 }
2485
2486 srcu_read_unlock(&fs_info->subvol_srcu, index);
2487
2488 /* step 3: relink backref */
2489 lock_start = backref->file_pos;
2490 lock_end = backref->file_pos + backref->num_bytes - 1;
2491 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2492 &cached);
2493
2494 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2495 if (ordered) {
2496 btrfs_put_ordered_extent(ordered);
2497 goto out_unlock;
2498 }
2499
2500 trans = btrfs_join_transaction(root);
2501 if (IS_ERR(trans)) {
2502 ret = PTR_ERR(trans);
2503 goto out_unlock;
2504 }
2505
2506 key.objectid = backref->inum;
2507 key.type = BTRFS_EXTENT_DATA_KEY;
2508 key.offset = backref->file_pos;
2509
2510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2511 if (ret < 0) {
2512 goto out_free_path;
2513 } else if (ret > 0) {
2514 ret = 0;
2515 goto out_free_path;
2516 }
2517
2518 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2519 struct btrfs_file_extent_item);
2520
2521 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2522 backref->generation)
2523 goto out_free_path;
2524
2525 btrfs_release_path(path);
2526
2527 start = backref->file_pos;
2528 if (backref->extent_offset < old->extent_offset + old->offset)
2529 start += old->extent_offset + old->offset -
2530 backref->extent_offset;
2531
2532 len = min(backref->extent_offset + backref->num_bytes,
2533 old->extent_offset + old->offset + old->len);
2534 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2535
2536 ret = btrfs_drop_extents(trans, root, inode, start,
2537 start + len, 1);
2538 if (ret)
2539 goto out_free_path;
2540 again:
2541 key.objectid = btrfs_ino(inode);
2542 key.type = BTRFS_EXTENT_DATA_KEY;
2543 key.offset = start;
2544
2545 path->leave_spinning = 1;
2546 if (merge) {
2547 struct btrfs_file_extent_item *fi;
2548 u64 extent_len;
2549 struct btrfs_key found_key;
2550
2551 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2552 if (ret < 0)
2553 goto out_free_path;
2554
2555 path->slots[0]--;
2556 leaf = path->nodes[0];
2557 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2558
2559 fi = btrfs_item_ptr(leaf, path->slots[0],
2560 struct btrfs_file_extent_item);
2561 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2562
2563 if (extent_len + found_key.offset == start &&
2564 relink_is_mergable(leaf, fi, new)) {
2565 btrfs_set_file_extent_num_bytes(leaf, fi,
2566 extent_len + len);
2567 btrfs_mark_buffer_dirty(leaf);
2568 inode_add_bytes(inode, len);
2569
2570 ret = 1;
2571 goto out_free_path;
2572 } else {
2573 merge = false;
2574 btrfs_release_path(path);
2575 goto again;
2576 }
2577 }
2578
2579 ret = btrfs_insert_empty_item(trans, root, path, &key,
2580 sizeof(*extent));
2581 if (ret) {
2582 btrfs_abort_transaction(trans, root, ret);
2583 goto out_free_path;
2584 }
2585
2586 leaf = path->nodes[0];
2587 item = btrfs_item_ptr(leaf, path->slots[0],
2588 struct btrfs_file_extent_item);
2589 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2590 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2591 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2592 btrfs_set_file_extent_num_bytes(leaf, item, len);
2593 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2594 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2595 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2596 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2597 btrfs_set_file_extent_encryption(leaf, item, 0);
2598 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2599
2600 btrfs_mark_buffer_dirty(leaf);
2601 inode_add_bytes(inode, len);
2602 btrfs_release_path(path);
2603
2604 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2605 new->disk_len, 0,
2606 backref->root_id, backref->inum,
2607 new->file_pos); /* start - extent_offset */
2608 if (ret) {
2609 btrfs_abort_transaction(trans, root, ret);
2610 goto out_free_path;
2611 }
2612
2613 ret = 1;
2614 out_free_path:
2615 btrfs_release_path(path);
2616 path->leave_spinning = 0;
2617 btrfs_end_transaction(trans, root);
2618 out_unlock:
2619 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2620 &cached, GFP_NOFS);
2621 iput(inode);
2622 return ret;
2623 }
2624
2625 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2626 {
2627 struct old_sa_defrag_extent *old, *tmp;
2628
2629 if (!new)
2630 return;
2631
2632 list_for_each_entry_safe(old, tmp, &new->head, list) {
2633 kfree(old);
2634 }
2635 kfree(new);
2636 }
2637
2638 static void relink_file_extents(struct new_sa_defrag_extent *new)
2639 {
2640 struct btrfs_path *path;
2641 struct sa_defrag_extent_backref *backref;
2642 struct sa_defrag_extent_backref *prev = NULL;
2643 struct inode *inode;
2644 struct btrfs_root *root;
2645 struct rb_node *node;
2646 int ret;
2647
2648 inode = new->inode;
2649 root = BTRFS_I(inode)->root;
2650
2651 path = btrfs_alloc_path();
2652 if (!path)
2653 return;
2654
2655 if (!record_extent_backrefs(path, new)) {
2656 btrfs_free_path(path);
2657 goto out;
2658 }
2659 btrfs_release_path(path);
2660
2661 while (1) {
2662 node = rb_first(&new->root);
2663 if (!node)
2664 break;
2665 rb_erase(node, &new->root);
2666
2667 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2668
2669 ret = relink_extent_backref(path, prev, backref);
2670 WARN_ON(ret < 0);
2671
2672 kfree(prev);
2673
2674 if (ret == 1)
2675 prev = backref;
2676 else
2677 prev = NULL;
2678 cond_resched();
2679 }
2680 kfree(prev);
2681
2682 btrfs_free_path(path);
2683 out:
2684 free_sa_defrag_extent(new);
2685
2686 atomic_dec(&root->fs_info->defrag_running);
2687 wake_up(&root->fs_info->transaction_wait);
2688 }
2689
2690 static struct new_sa_defrag_extent *
2691 record_old_file_extents(struct inode *inode,
2692 struct btrfs_ordered_extent *ordered)
2693 {
2694 struct btrfs_root *root = BTRFS_I(inode)->root;
2695 struct btrfs_path *path;
2696 struct btrfs_key key;
2697 struct old_sa_defrag_extent *old;
2698 struct new_sa_defrag_extent *new;
2699 int ret;
2700
2701 new = kmalloc(sizeof(*new), GFP_NOFS);
2702 if (!new)
2703 return NULL;
2704
2705 new->inode = inode;
2706 new->file_pos = ordered->file_offset;
2707 new->len = ordered->len;
2708 new->bytenr = ordered->start;
2709 new->disk_len = ordered->disk_len;
2710 new->compress_type = ordered->compress_type;
2711 new->root = RB_ROOT;
2712 INIT_LIST_HEAD(&new->head);
2713
2714 path = btrfs_alloc_path();
2715 if (!path)
2716 goto out_kfree;
2717
2718 key.objectid = btrfs_ino(inode);
2719 key.type = BTRFS_EXTENT_DATA_KEY;
2720 key.offset = new->file_pos;
2721
2722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2723 if (ret < 0)
2724 goto out_free_path;
2725 if (ret > 0 && path->slots[0] > 0)
2726 path->slots[0]--;
2727
2728 /* find out all the old extents for the file range */
2729 while (1) {
2730 struct btrfs_file_extent_item *extent;
2731 struct extent_buffer *l;
2732 int slot;
2733 u64 num_bytes;
2734 u64 offset;
2735 u64 end;
2736 u64 disk_bytenr;
2737 u64 extent_offset;
2738
2739 l = path->nodes[0];
2740 slot = path->slots[0];
2741
2742 if (slot >= btrfs_header_nritems(l)) {
2743 ret = btrfs_next_leaf(root, path);
2744 if (ret < 0)
2745 goto out_free_path;
2746 else if (ret > 0)
2747 break;
2748 continue;
2749 }
2750
2751 btrfs_item_key_to_cpu(l, &key, slot);
2752
2753 if (key.objectid != btrfs_ino(inode))
2754 break;
2755 if (key.type != BTRFS_EXTENT_DATA_KEY)
2756 break;
2757 if (key.offset >= new->file_pos + new->len)
2758 break;
2759
2760 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2761
2762 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2763 if (key.offset + num_bytes < new->file_pos)
2764 goto next;
2765
2766 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2767 if (!disk_bytenr)
2768 goto next;
2769
2770 extent_offset = btrfs_file_extent_offset(l, extent);
2771
2772 old = kmalloc(sizeof(*old), GFP_NOFS);
2773 if (!old)
2774 goto out_free_path;
2775
2776 offset = max(new->file_pos, key.offset);
2777 end = min(new->file_pos + new->len, key.offset + num_bytes);
2778
2779 old->bytenr = disk_bytenr;
2780 old->extent_offset = extent_offset;
2781 old->offset = offset - key.offset;
2782 old->len = end - offset;
2783 old->new = new;
2784 old->count = 0;
2785 list_add_tail(&old->list, &new->head);
2786 next:
2787 path->slots[0]++;
2788 cond_resched();
2789 }
2790
2791 btrfs_free_path(path);
2792 atomic_inc(&root->fs_info->defrag_running);
2793
2794 return new;
2795
2796 out_free_path:
2797 btrfs_free_path(path);
2798 out_kfree:
2799 free_sa_defrag_extent(new);
2800 return NULL;
2801 }
2802
2803 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2804 u64 start, u64 len)
2805 {
2806 struct btrfs_block_group_cache *cache;
2807
2808 cache = btrfs_lookup_block_group(root->fs_info, start);
2809 ASSERT(cache);
2810
2811 spin_lock(&cache->lock);
2812 cache->delalloc_bytes -= len;
2813 spin_unlock(&cache->lock);
2814
2815 btrfs_put_block_group(cache);
2816 }
2817
2818 /* as ordered data IO finishes, this gets called so we can finish
2819 * an ordered extent if the range of bytes in the file it covers are
2820 * fully written.
2821 */
2822 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2823 {
2824 struct inode *inode = ordered_extent->inode;
2825 struct btrfs_root *root = BTRFS_I(inode)->root;
2826 struct btrfs_trans_handle *trans = NULL;
2827 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2828 struct extent_state *cached_state = NULL;
2829 struct new_sa_defrag_extent *new = NULL;
2830 int compress_type = 0;
2831 int ret = 0;
2832 u64 logical_len = ordered_extent->len;
2833 bool nolock;
2834 bool truncated = false;
2835
2836 nolock = btrfs_is_free_space_inode(inode);
2837
2838 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2839 ret = -EIO;
2840 goto out;
2841 }
2842
2843 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2844 ordered_extent->file_offset +
2845 ordered_extent->len - 1);
2846
2847 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2848 truncated = true;
2849 logical_len = ordered_extent->truncated_len;
2850 /* Truncated the entire extent, don't bother adding */
2851 if (!logical_len)
2852 goto out;
2853 }
2854
2855 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2856 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2857
2858 /*
2859 * For mwrite(mmap + memset to write) case, we still reserve
2860 * space for NOCOW range.
2861 * As NOCOW won't cause a new delayed ref, just free the space
2862 */
2863 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2864 ordered_extent->len);
2865 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2866 if (nolock)
2867 trans = btrfs_join_transaction_nolock(root);
2868 else
2869 trans = btrfs_join_transaction(root);
2870 if (IS_ERR(trans)) {
2871 ret = PTR_ERR(trans);
2872 trans = NULL;
2873 goto out;
2874 }
2875 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2876 ret = btrfs_update_inode_fallback(trans, root, inode);
2877 if (ret) /* -ENOMEM or corruption */
2878 btrfs_abort_transaction(trans, root, ret);
2879 goto out;
2880 }
2881
2882 lock_extent_bits(io_tree, ordered_extent->file_offset,
2883 ordered_extent->file_offset + ordered_extent->len - 1,
2884 &cached_state);
2885
2886 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2887 ordered_extent->file_offset + ordered_extent->len - 1,
2888 EXTENT_DEFRAG, 1, cached_state);
2889 if (ret) {
2890 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2891 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2892 /* the inode is shared */
2893 new = record_old_file_extents(inode, ordered_extent);
2894
2895 clear_extent_bit(io_tree, ordered_extent->file_offset,
2896 ordered_extent->file_offset + ordered_extent->len - 1,
2897 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2898 }
2899
2900 if (nolock)
2901 trans = btrfs_join_transaction_nolock(root);
2902 else
2903 trans = btrfs_join_transaction(root);
2904 if (IS_ERR(trans)) {
2905 ret = PTR_ERR(trans);
2906 trans = NULL;
2907 goto out_unlock;
2908 }
2909
2910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2911
2912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2913 compress_type = ordered_extent->compress_type;
2914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2915 BUG_ON(compress_type);
2916 ret = btrfs_mark_extent_written(trans, inode,
2917 ordered_extent->file_offset,
2918 ordered_extent->file_offset +
2919 logical_len);
2920 } else {
2921 BUG_ON(root == root->fs_info->tree_root);
2922 ret = insert_reserved_file_extent(trans, inode,
2923 ordered_extent->file_offset,
2924 ordered_extent->start,
2925 ordered_extent->disk_len,
2926 logical_len, logical_len,
2927 compress_type, 0, 0,
2928 BTRFS_FILE_EXTENT_REG);
2929 if (!ret)
2930 btrfs_release_delalloc_bytes(root,
2931 ordered_extent->start,
2932 ordered_extent->disk_len);
2933 }
2934 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2935 ordered_extent->file_offset, ordered_extent->len,
2936 trans->transid);
2937 if (ret < 0) {
2938 btrfs_abort_transaction(trans, root, ret);
2939 goto out_unlock;
2940 }
2941
2942 add_pending_csums(trans, inode, ordered_extent->file_offset,
2943 &ordered_extent->list);
2944
2945 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2946 ret = btrfs_update_inode_fallback(trans, root, inode);
2947 if (ret) { /* -ENOMEM or corruption */
2948 btrfs_abort_transaction(trans, root, ret);
2949 goto out_unlock;
2950 }
2951 ret = 0;
2952 out_unlock:
2953 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2954 ordered_extent->file_offset +
2955 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2956 out:
2957 if (root != root->fs_info->tree_root)
2958 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2959 if (trans)
2960 btrfs_end_transaction(trans, root);
2961
2962 if (ret || truncated) {
2963 u64 start, end;
2964
2965 if (truncated)
2966 start = ordered_extent->file_offset + logical_len;
2967 else
2968 start = ordered_extent->file_offset;
2969 end = ordered_extent->file_offset + ordered_extent->len - 1;
2970 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2971
2972 /* Drop the cache for the part of the extent we didn't write. */
2973 btrfs_drop_extent_cache(inode, start, end, 0);
2974
2975 /*
2976 * If the ordered extent had an IOERR or something else went
2977 * wrong we need to return the space for this ordered extent
2978 * back to the allocator. We only free the extent in the
2979 * truncated case if we didn't write out the extent at all.
2980 */
2981 if ((ret || !logical_len) &&
2982 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2983 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2984 btrfs_free_reserved_extent(root, ordered_extent->start,
2985 ordered_extent->disk_len, 1);
2986 }
2987
2988
2989 /*
2990 * This needs to be done to make sure anybody waiting knows we are done
2991 * updating everything for this ordered extent.
2992 */
2993 btrfs_remove_ordered_extent(inode, ordered_extent);
2994
2995 /* for snapshot-aware defrag */
2996 if (new) {
2997 if (ret) {
2998 free_sa_defrag_extent(new);
2999 atomic_dec(&root->fs_info->defrag_running);
3000 } else {
3001 relink_file_extents(new);
3002 }
3003 }
3004
3005 /* once for us */
3006 btrfs_put_ordered_extent(ordered_extent);
3007 /* once for the tree */
3008 btrfs_put_ordered_extent(ordered_extent);
3009
3010 return ret;
3011 }
3012
3013 static void finish_ordered_fn(struct btrfs_work *work)
3014 {
3015 struct btrfs_ordered_extent *ordered_extent;
3016 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3017 btrfs_finish_ordered_io(ordered_extent);
3018 }
3019
3020 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3021 struct extent_state *state, int uptodate)
3022 {
3023 struct inode *inode = page->mapping->host;
3024 struct btrfs_root *root = BTRFS_I(inode)->root;
3025 struct btrfs_ordered_extent *ordered_extent = NULL;
3026 struct btrfs_workqueue *wq;
3027 btrfs_work_func_t func;
3028
3029 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3030
3031 ClearPagePrivate2(page);
3032 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3033 end - start + 1, uptodate))
3034 return 0;
3035
3036 if (btrfs_is_free_space_inode(inode)) {
3037 wq = root->fs_info->endio_freespace_worker;
3038 func = btrfs_freespace_write_helper;
3039 } else {
3040 wq = root->fs_info->endio_write_workers;
3041 func = btrfs_endio_write_helper;
3042 }
3043
3044 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3045 NULL);
3046 btrfs_queue_work(wq, &ordered_extent->work);
3047
3048 return 0;
3049 }
3050
3051 static int __readpage_endio_check(struct inode *inode,
3052 struct btrfs_io_bio *io_bio,
3053 int icsum, struct page *page,
3054 int pgoff, u64 start, size_t len)
3055 {
3056 char *kaddr;
3057 u32 csum_expected;
3058 u32 csum = ~(u32)0;
3059
3060 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3061
3062 kaddr = kmap_atomic(page);
3063 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3064 btrfs_csum_final(csum, (char *)&csum);
3065 if (csum != csum_expected)
3066 goto zeroit;
3067
3068 kunmap_atomic(kaddr);
3069 return 0;
3070 zeroit:
3071 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3072 "csum failed ino %llu off %llu csum %u expected csum %u",
3073 btrfs_ino(inode), start, csum, csum_expected);
3074 memset(kaddr + pgoff, 1, len);
3075 flush_dcache_page(page);
3076 kunmap_atomic(kaddr);
3077 if (csum_expected == 0)
3078 return 0;
3079 return -EIO;
3080 }
3081
3082 /*
3083 * when reads are done, we need to check csums to verify the data is correct
3084 * if there's a match, we allow the bio to finish. If not, the code in
3085 * extent_io.c will try to find good copies for us.
3086 */
3087 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3088 u64 phy_offset, struct page *page,
3089 u64 start, u64 end, int mirror)
3090 {
3091 size_t offset = start - page_offset(page);
3092 struct inode *inode = page->mapping->host;
3093 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3094 struct btrfs_root *root = BTRFS_I(inode)->root;
3095
3096 if (PageChecked(page)) {
3097 ClearPageChecked(page);
3098 return 0;
3099 }
3100
3101 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3102 return 0;
3103
3104 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3105 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3106 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3107 GFP_NOFS);
3108 return 0;
3109 }
3110
3111 phy_offset >>= inode->i_sb->s_blocksize_bits;
3112 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3113 start, (size_t)(end - start + 1));
3114 }
3115
3116 void btrfs_add_delayed_iput(struct inode *inode)
3117 {
3118 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3119 struct btrfs_inode *binode = BTRFS_I(inode);
3120
3121 if (atomic_add_unless(&inode->i_count, -1, 1))
3122 return;
3123
3124 spin_lock(&fs_info->delayed_iput_lock);
3125 if (binode->delayed_iput_count == 0) {
3126 ASSERT(list_empty(&binode->delayed_iput));
3127 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3128 } else {
3129 binode->delayed_iput_count++;
3130 }
3131 spin_unlock(&fs_info->delayed_iput_lock);
3132 }
3133
3134 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3135 {
3136 struct btrfs_fs_info *fs_info = root->fs_info;
3137
3138 spin_lock(&fs_info->delayed_iput_lock);
3139 while (!list_empty(&fs_info->delayed_iputs)) {
3140 struct btrfs_inode *inode;
3141
3142 inode = list_first_entry(&fs_info->delayed_iputs,
3143 struct btrfs_inode, delayed_iput);
3144 if (inode->delayed_iput_count) {
3145 inode->delayed_iput_count--;
3146 list_move_tail(&inode->delayed_iput,
3147 &fs_info->delayed_iputs);
3148 } else {
3149 list_del_init(&inode->delayed_iput);
3150 }
3151 spin_unlock(&fs_info->delayed_iput_lock);
3152 iput(&inode->vfs_inode);
3153 spin_lock(&fs_info->delayed_iput_lock);
3154 }
3155 spin_unlock(&fs_info->delayed_iput_lock);
3156 }
3157
3158 /*
3159 * This is called in transaction commit time. If there are no orphan
3160 * files in the subvolume, it removes orphan item and frees block_rsv
3161 * structure.
3162 */
3163 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3164 struct btrfs_root *root)
3165 {
3166 struct btrfs_block_rsv *block_rsv;
3167 int ret;
3168
3169 if (atomic_read(&root->orphan_inodes) ||
3170 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3171 return;
3172
3173 spin_lock(&root->orphan_lock);
3174 if (atomic_read(&root->orphan_inodes)) {
3175 spin_unlock(&root->orphan_lock);
3176 return;
3177 }
3178
3179 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3180 spin_unlock(&root->orphan_lock);
3181 return;
3182 }
3183
3184 block_rsv = root->orphan_block_rsv;
3185 root->orphan_block_rsv = NULL;
3186 spin_unlock(&root->orphan_lock);
3187
3188 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3189 btrfs_root_refs(&root->root_item) > 0) {
3190 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3191 root->root_key.objectid);
3192 if (ret)
3193 btrfs_abort_transaction(trans, root, ret);
3194 else
3195 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3196 &root->state);
3197 }
3198
3199 if (block_rsv) {
3200 WARN_ON(block_rsv->size > 0);
3201 btrfs_free_block_rsv(root, block_rsv);
3202 }
3203 }
3204
3205 /*
3206 * This creates an orphan entry for the given inode in case something goes
3207 * wrong in the middle of an unlink/truncate.
3208 *
3209 * NOTE: caller of this function should reserve 5 units of metadata for
3210 * this function.
3211 */
3212 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3213 {
3214 struct btrfs_root *root = BTRFS_I(inode)->root;
3215 struct btrfs_block_rsv *block_rsv = NULL;
3216 int reserve = 0;
3217 int insert = 0;
3218 int ret;
3219
3220 if (!root->orphan_block_rsv) {
3221 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3222 if (!block_rsv)
3223 return -ENOMEM;
3224 }
3225
3226 spin_lock(&root->orphan_lock);
3227 if (!root->orphan_block_rsv) {
3228 root->orphan_block_rsv = block_rsv;
3229 } else if (block_rsv) {
3230 btrfs_free_block_rsv(root, block_rsv);
3231 block_rsv = NULL;
3232 }
3233
3234 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3235 &BTRFS_I(inode)->runtime_flags)) {
3236 #if 0
3237 /*
3238 * For proper ENOSPC handling, we should do orphan
3239 * cleanup when mounting. But this introduces backward
3240 * compatibility issue.
3241 */
3242 if (!xchg(&root->orphan_item_inserted, 1))
3243 insert = 2;
3244 else
3245 insert = 1;
3246 #endif
3247 insert = 1;
3248 atomic_inc(&root->orphan_inodes);
3249 }
3250
3251 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3252 &BTRFS_I(inode)->runtime_flags))
3253 reserve = 1;
3254 spin_unlock(&root->orphan_lock);
3255
3256 /* grab metadata reservation from transaction handle */
3257 if (reserve) {
3258 ret = btrfs_orphan_reserve_metadata(trans, inode);
3259 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3260 }
3261
3262 /* insert an orphan item to track this unlinked/truncated file */
3263 if (insert >= 1) {
3264 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3265 if (ret) {
3266 atomic_dec(&root->orphan_inodes);
3267 if (reserve) {
3268 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3269 &BTRFS_I(inode)->runtime_flags);
3270 btrfs_orphan_release_metadata(inode);
3271 }
3272 if (ret != -EEXIST) {
3273 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3274 &BTRFS_I(inode)->runtime_flags);
3275 btrfs_abort_transaction(trans, root, ret);
3276 return ret;
3277 }
3278 }
3279 ret = 0;
3280 }
3281
3282 /* insert an orphan item to track subvolume contains orphan files */
3283 if (insert >= 2) {
3284 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3285 root->root_key.objectid);
3286 if (ret && ret != -EEXIST) {
3287 btrfs_abort_transaction(trans, root, ret);
3288 return ret;
3289 }
3290 }
3291 return 0;
3292 }
3293
3294 /*
3295 * We have done the truncate/delete so we can go ahead and remove the orphan
3296 * item for this particular inode.
3297 */
3298 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3299 struct inode *inode)
3300 {
3301 struct btrfs_root *root = BTRFS_I(inode)->root;
3302 int delete_item = 0;
3303 int release_rsv = 0;
3304 int ret = 0;
3305
3306 spin_lock(&root->orphan_lock);
3307 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3308 &BTRFS_I(inode)->runtime_flags))
3309 delete_item = 1;
3310
3311 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3312 &BTRFS_I(inode)->runtime_flags))
3313 release_rsv = 1;
3314 spin_unlock(&root->orphan_lock);
3315
3316 if (delete_item) {
3317 atomic_dec(&root->orphan_inodes);
3318 if (trans)
3319 ret = btrfs_del_orphan_item(trans, root,
3320 btrfs_ino(inode));
3321 }
3322
3323 if (release_rsv)
3324 btrfs_orphan_release_metadata(inode);
3325
3326 return ret;
3327 }
3328
3329 /*
3330 * this cleans up any orphans that may be left on the list from the last use
3331 * of this root.
3332 */
3333 int btrfs_orphan_cleanup(struct btrfs_root *root)
3334 {
3335 struct btrfs_path *path;
3336 struct extent_buffer *leaf;
3337 struct btrfs_key key, found_key;
3338 struct btrfs_trans_handle *trans;
3339 struct inode *inode;
3340 u64 last_objectid = 0;
3341 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3342
3343 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3344 return 0;
3345
3346 path = btrfs_alloc_path();
3347 if (!path) {
3348 ret = -ENOMEM;
3349 goto out;
3350 }
3351 path->reada = READA_BACK;
3352
3353 key.objectid = BTRFS_ORPHAN_OBJECTID;
3354 key.type = BTRFS_ORPHAN_ITEM_KEY;
3355 key.offset = (u64)-1;
3356
3357 while (1) {
3358 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3359 if (ret < 0)
3360 goto out;
3361
3362 /*
3363 * if ret == 0 means we found what we were searching for, which
3364 * is weird, but possible, so only screw with path if we didn't
3365 * find the key and see if we have stuff that matches
3366 */
3367 if (ret > 0) {
3368 ret = 0;
3369 if (path->slots[0] == 0)
3370 break;
3371 path->slots[0]--;
3372 }
3373
3374 /* pull out the item */
3375 leaf = path->nodes[0];
3376 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3377
3378 /* make sure the item matches what we want */
3379 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3380 break;
3381 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3382 break;
3383
3384 /* release the path since we're done with it */
3385 btrfs_release_path(path);
3386
3387 /*
3388 * this is where we are basically btrfs_lookup, without the
3389 * crossing root thing. we store the inode number in the
3390 * offset of the orphan item.
3391 */
3392
3393 if (found_key.offset == last_objectid) {
3394 btrfs_err(root->fs_info,
3395 "Error removing orphan entry, stopping orphan cleanup");
3396 ret = -EINVAL;
3397 goto out;
3398 }
3399
3400 last_objectid = found_key.offset;
3401
3402 found_key.objectid = found_key.offset;
3403 found_key.type = BTRFS_INODE_ITEM_KEY;
3404 found_key.offset = 0;
3405 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3406 ret = PTR_ERR_OR_ZERO(inode);
3407 if (ret && ret != -ESTALE)
3408 goto out;
3409
3410 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3411 struct btrfs_root *dead_root;
3412 struct btrfs_fs_info *fs_info = root->fs_info;
3413 int is_dead_root = 0;
3414
3415 /*
3416 * this is an orphan in the tree root. Currently these
3417 * could come from 2 sources:
3418 * a) a snapshot deletion in progress
3419 * b) a free space cache inode
3420 * We need to distinguish those two, as the snapshot
3421 * orphan must not get deleted.
3422 * find_dead_roots already ran before us, so if this
3423 * is a snapshot deletion, we should find the root
3424 * in the dead_roots list
3425 */
3426 spin_lock(&fs_info->trans_lock);
3427 list_for_each_entry(dead_root, &fs_info->dead_roots,
3428 root_list) {
3429 if (dead_root->root_key.objectid ==
3430 found_key.objectid) {
3431 is_dead_root = 1;
3432 break;
3433 }
3434 }
3435 spin_unlock(&fs_info->trans_lock);
3436 if (is_dead_root) {
3437 /* prevent this orphan from being found again */
3438 key.offset = found_key.objectid - 1;
3439 continue;
3440 }
3441 }
3442 /*
3443 * Inode is already gone but the orphan item is still there,
3444 * kill the orphan item.
3445 */
3446 if (ret == -ESTALE) {
3447 trans = btrfs_start_transaction(root, 1);
3448 if (IS_ERR(trans)) {
3449 ret = PTR_ERR(trans);
3450 goto out;
3451 }
3452 btrfs_debug(root->fs_info, "auto deleting %Lu",
3453 found_key.objectid);
3454 ret = btrfs_del_orphan_item(trans, root,
3455 found_key.objectid);
3456 btrfs_end_transaction(trans, root);
3457 if (ret)
3458 goto out;
3459 continue;
3460 }
3461
3462 /*
3463 * add this inode to the orphan list so btrfs_orphan_del does
3464 * the proper thing when we hit it
3465 */
3466 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3467 &BTRFS_I(inode)->runtime_flags);
3468 atomic_inc(&root->orphan_inodes);
3469
3470 /* if we have links, this was a truncate, lets do that */
3471 if (inode->i_nlink) {
3472 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3473 iput(inode);
3474 continue;
3475 }
3476 nr_truncate++;
3477
3478 /* 1 for the orphan item deletion. */
3479 trans = btrfs_start_transaction(root, 1);
3480 if (IS_ERR(trans)) {
3481 iput(inode);
3482 ret = PTR_ERR(trans);
3483 goto out;
3484 }
3485 ret = btrfs_orphan_add(trans, inode);
3486 btrfs_end_transaction(trans, root);
3487 if (ret) {
3488 iput(inode);
3489 goto out;
3490 }
3491
3492 ret = btrfs_truncate(inode);
3493 if (ret)
3494 btrfs_orphan_del(NULL, inode);
3495 } else {
3496 nr_unlink++;
3497 }
3498
3499 /* this will do delete_inode and everything for us */
3500 iput(inode);
3501 if (ret)
3502 goto out;
3503 }
3504 /* release the path since we're done with it */
3505 btrfs_release_path(path);
3506
3507 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3508
3509 if (root->orphan_block_rsv)
3510 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3511 (u64)-1);
3512
3513 if (root->orphan_block_rsv ||
3514 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3515 trans = btrfs_join_transaction(root);
3516 if (!IS_ERR(trans))
3517 btrfs_end_transaction(trans, root);
3518 }
3519
3520 if (nr_unlink)
3521 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3522 if (nr_truncate)
3523 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3524
3525 out:
3526 if (ret)
3527 btrfs_err(root->fs_info,
3528 "could not do orphan cleanup %d", ret);
3529 btrfs_free_path(path);
3530 return ret;
3531 }
3532
3533 /*
3534 * very simple check to peek ahead in the leaf looking for xattrs. If we
3535 * don't find any xattrs, we know there can't be any acls.
3536 *
3537 * slot is the slot the inode is in, objectid is the objectid of the inode
3538 */
3539 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3540 int slot, u64 objectid,
3541 int *first_xattr_slot)
3542 {
3543 u32 nritems = btrfs_header_nritems(leaf);
3544 struct btrfs_key found_key;
3545 static u64 xattr_access = 0;
3546 static u64 xattr_default = 0;
3547 int scanned = 0;
3548
3549 if (!xattr_access) {
3550 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3551 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3552 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3553 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3554 }
3555
3556 slot++;
3557 *first_xattr_slot = -1;
3558 while (slot < nritems) {
3559 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3560
3561 /* we found a different objectid, there must not be acls */
3562 if (found_key.objectid != objectid)
3563 return 0;
3564
3565 /* we found an xattr, assume we've got an acl */
3566 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3567 if (*first_xattr_slot == -1)
3568 *first_xattr_slot = slot;
3569 if (found_key.offset == xattr_access ||
3570 found_key.offset == xattr_default)
3571 return 1;
3572 }
3573
3574 /*
3575 * we found a key greater than an xattr key, there can't
3576 * be any acls later on
3577 */
3578 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3579 return 0;
3580
3581 slot++;
3582 scanned++;
3583
3584 /*
3585 * it goes inode, inode backrefs, xattrs, extents,
3586 * so if there are a ton of hard links to an inode there can
3587 * be a lot of backrefs. Don't waste time searching too hard,
3588 * this is just an optimization
3589 */
3590 if (scanned >= 8)
3591 break;
3592 }
3593 /* we hit the end of the leaf before we found an xattr or
3594 * something larger than an xattr. We have to assume the inode
3595 * has acls
3596 */
3597 if (*first_xattr_slot == -1)
3598 *first_xattr_slot = slot;
3599 return 1;
3600 }
3601
3602 /*
3603 * read an inode from the btree into the in-memory inode
3604 */
3605 static void btrfs_read_locked_inode(struct inode *inode)
3606 {
3607 struct btrfs_path *path;
3608 struct extent_buffer *leaf;
3609 struct btrfs_inode_item *inode_item;
3610 struct btrfs_root *root = BTRFS_I(inode)->root;
3611 struct btrfs_key location;
3612 unsigned long ptr;
3613 int maybe_acls;
3614 u32 rdev;
3615 int ret;
3616 bool filled = false;
3617 int first_xattr_slot;
3618
3619 ret = btrfs_fill_inode(inode, &rdev);
3620 if (!ret)
3621 filled = true;
3622
3623 path = btrfs_alloc_path();
3624 if (!path)
3625 goto make_bad;
3626
3627 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3628
3629 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3630 if (ret)
3631 goto make_bad;
3632
3633 leaf = path->nodes[0];
3634
3635 if (filled)
3636 goto cache_index;
3637
3638 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3639 struct btrfs_inode_item);
3640 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3641 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3642 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3643 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3644 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3645
3646 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3647 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3648
3649 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3650 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3651
3652 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3653 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3654
3655 BTRFS_I(inode)->i_otime.tv_sec =
3656 btrfs_timespec_sec(leaf, &inode_item->otime);
3657 BTRFS_I(inode)->i_otime.tv_nsec =
3658 btrfs_timespec_nsec(leaf, &inode_item->otime);
3659
3660 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3661 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3662 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3663
3664 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3665 inode->i_generation = BTRFS_I(inode)->generation;
3666 inode->i_rdev = 0;
3667 rdev = btrfs_inode_rdev(leaf, inode_item);
3668
3669 BTRFS_I(inode)->index_cnt = (u64)-1;
3670 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3671
3672 cache_index:
3673 /*
3674 * If we were modified in the current generation and evicted from memory
3675 * and then re-read we need to do a full sync since we don't have any
3676 * idea about which extents were modified before we were evicted from
3677 * cache.
3678 *
3679 * This is required for both inode re-read from disk and delayed inode
3680 * in delayed_nodes_tree.
3681 */
3682 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3683 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3684 &BTRFS_I(inode)->runtime_flags);
3685
3686 /*
3687 * We don't persist the id of the transaction where an unlink operation
3688 * against the inode was last made. So here we assume the inode might
3689 * have been evicted, and therefore the exact value of last_unlink_trans
3690 * lost, and set it to last_trans to avoid metadata inconsistencies
3691 * between the inode and its parent if the inode is fsync'ed and the log
3692 * replayed. For example, in the scenario:
3693 *
3694 * touch mydir/foo
3695 * ln mydir/foo mydir/bar
3696 * sync
3697 * unlink mydir/bar
3698 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3699 * xfs_io -c fsync mydir/foo
3700 * <power failure>
3701 * mount fs, triggers fsync log replay
3702 *
3703 * We must make sure that when we fsync our inode foo we also log its
3704 * parent inode, otherwise after log replay the parent still has the
3705 * dentry with the "bar" name but our inode foo has a link count of 1
3706 * and doesn't have an inode ref with the name "bar" anymore.
3707 *
3708 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3709 * but it guarantees correctness at the expense of ocassional full
3710 * transaction commits on fsync if our inode is a directory, or if our
3711 * inode is not a directory, logging its parent unnecessarily.
3712 */
3713 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3714
3715 path->slots[0]++;
3716 if (inode->i_nlink != 1 ||
3717 path->slots[0] >= btrfs_header_nritems(leaf))
3718 goto cache_acl;
3719
3720 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3721 if (location.objectid != btrfs_ino(inode))
3722 goto cache_acl;
3723
3724 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3725 if (location.type == BTRFS_INODE_REF_KEY) {
3726 struct btrfs_inode_ref *ref;
3727
3728 ref = (struct btrfs_inode_ref *)ptr;
3729 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3730 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3731 struct btrfs_inode_extref *extref;
3732
3733 extref = (struct btrfs_inode_extref *)ptr;
3734 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3735 extref);
3736 }
3737 cache_acl:
3738 /*
3739 * try to precache a NULL acl entry for files that don't have
3740 * any xattrs or acls
3741 */
3742 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3743 btrfs_ino(inode), &first_xattr_slot);
3744 if (first_xattr_slot != -1) {
3745 path->slots[0] = first_xattr_slot;
3746 ret = btrfs_load_inode_props(inode, path);
3747 if (ret)
3748 btrfs_err(root->fs_info,
3749 "error loading props for ino %llu (root %llu): %d",
3750 btrfs_ino(inode),
3751 root->root_key.objectid, ret);
3752 }
3753 btrfs_free_path(path);
3754
3755 if (!maybe_acls)
3756 cache_no_acl(inode);
3757
3758 switch (inode->i_mode & S_IFMT) {
3759 case S_IFREG:
3760 inode->i_mapping->a_ops = &btrfs_aops;
3761 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3762 inode->i_fop = &btrfs_file_operations;
3763 inode->i_op = &btrfs_file_inode_operations;
3764 break;
3765 case S_IFDIR:
3766 inode->i_fop = &btrfs_dir_file_operations;
3767 if (root == root->fs_info->tree_root)
3768 inode->i_op = &btrfs_dir_ro_inode_operations;
3769 else
3770 inode->i_op = &btrfs_dir_inode_operations;
3771 break;
3772 case S_IFLNK:
3773 inode->i_op = &btrfs_symlink_inode_operations;
3774 inode_nohighmem(inode);
3775 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3776 break;
3777 default:
3778 inode->i_op = &btrfs_special_inode_operations;
3779 init_special_inode(inode, inode->i_mode, rdev);
3780 break;
3781 }
3782
3783 btrfs_update_iflags(inode);
3784 return;
3785
3786 make_bad:
3787 btrfs_free_path(path);
3788 make_bad_inode(inode);
3789 }
3790
3791 /*
3792 * given a leaf and an inode, copy the inode fields into the leaf
3793 */
3794 static void fill_inode_item(struct btrfs_trans_handle *trans,
3795 struct extent_buffer *leaf,
3796 struct btrfs_inode_item *item,
3797 struct inode *inode)
3798 {
3799 struct btrfs_map_token token;
3800
3801 btrfs_init_map_token(&token);
3802
3803 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3804 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3805 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3806 &token);
3807 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3808 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3809
3810 btrfs_set_token_timespec_sec(leaf, &item->atime,
3811 inode->i_atime.tv_sec, &token);
3812 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3813 inode->i_atime.tv_nsec, &token);
3814
3815 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3816 inode->i_mtime.tv_sec, &token);
3817 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3818 inode->i_mtime.tv_nsec, &token);
3819
3820 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3821 inode->i_ctime.tv_sec, &token);
3822 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3823 inode->i_ctime.tv_nsec, &token);
3824
3825 btrfs_set_token_timespec_sec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_sec, &token);
3827 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3828 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3829
3830 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3831 &token);
3832 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3833 &token);
3834 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3835 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3836 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3837 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3838 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3839 }
3840
3841 /*
3842 * copy everything in the in-memory inode into the btree.
3843 */
3844 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3845 struct btrfs_root *root, struct inode *inode)
3846 {
3847 struct btrfs_inode_item *inode_item;
3848 struct btrfs_path *path;
3849 struct extent_buffer *leaf;
3850 int ret;
3851
3852 path = btrfs_alloc_path();
3853 if (!path)
3854 return -ENOMEM;
3855
3856 path->leave_spinning = 1;
3857 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3858 1);
3859 if (ret) {
3860 if (ret > 0)
3861 ret = -ENOENT;
3862 goto failed;
3863 }
3864
3865 leaf = path->nodes[0];
3866 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3867 struct btrfs_inode_item);
3868
3869 fill_inode_item(trans, leaf, inode_item, inode);
3870 btrfs_mark_buffer_dirty(leaf);
3871 btrfs_set_inode_last_trans(trans, inode);
3872 ret = 0;
3873 failed:
3874 btrfs_free_path(path);
3875 return ret;
3876 }
3877
3878 /*
3879 * copy everything in the in-memory inode into the btree.
3880 */
3881 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3882 struct btrfs_root *root, struct inode *inode)
3883 {
3884 int ret;
3885
3886 /*
3887 * If the inode is a free space inode, we can deadlock during commit
3888 * if we put it into the delayed code.
3889 *
3890 * The data relocation inode should also be directly updated
3891 * without delay
3892 */
3893 if (!btrfs_is_free_space_inode(inode)
3894 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3895 && !root->fs_info->log_root_recovering) {
3896 btrfs_update_root_times(trans, root);
3897
3898 ret = btrfs_delayed_update_inode(trans, root, inode);
3899 if (!ret)
3900 btrfs_set_inode_last_trans(trans, inode);
3901 return ret;
3902 }
3903
3904 return btrfs_update_inode_item(trans, root, inode);
3905 }
3906
3907 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3908 struct btrfs_root *root,
3909 struct inode *inode)
3910 {
3911 int ret;
3912
3913 ret = btrfs_update_inode(trans, root, inode);
3914 if (ret == -ENOSPC)
3915 return btrfs_update_inode_item(trans, root, inode);
3916 return ret;
3917 }
3918
3919 /*
3920 * unlink helper that gets used here in inode.c and in the tree logging
3921 * recovery code. It remove a link in a directory with a given name, and
3922 * also drops the back refs in the inode to the directory
3923 */
3924 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3925 struct btrfs_root *root,
3926 struct inode *dir, struct inode *inode,
3927 const char *name, int name_len)
3928 {
3929 struct btrfs_path *path;
3930 int ret = 0;
3931 struct extent_buffer *leaf;
3932 struct btrfs_dir_item *di;
3933 struct btrfs_key key;
3934 u64 index;
3935 u64 ino = btrfs_ino(inode);
3936 u64 dir_ino = btrfs_ino(dir);
3937
3938 path = btrfs_alloc_path();
3939 if (!path) {
3940 ret = -ENOMEM;
3941 goto out;
3942 }
3943
3944 path->leave_spinning = 1;
3945 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3946 name, name_len, -1);
3947 if (IS_ERR(di)) {
3948 ret = PTR_ERR(di);
3949 goto err;
3950 }
3951 if (!di) {
3952 ret = -ENOENT;
3953 goto err;
3954 }
3955 leaf = path->nodes[0];
3956 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3957 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3958 if (ret)
3959 goto err;
3960 btrfs_release_path(path);
3961
3962 /*
3963 * If we don't have dir index, we have to get it by looking up
3964 * the inode ref, since we get the inode ref, remove it directly,
3965 * it is unnecessary to do delayed deletion.
3966 *
3967 * But if we have dir index, needn't search inode ref to get it.
3968 * Since the inode ref is close to the inode item, it is better
3969 * that we delay to delete it, and just do this deletion when
3970 * we update the inode item.
3971 */
3972 if (BTRFS_I(inode)->dir_index) {
3973 ret = btrfs_delayed_delete_inode_ref(inode);
3974 if (!ret) {
3975 index = BTRFS_I(inode)->dir_index;
3976 goto skip_backref;
3977 }
3978 }
3979
3980 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3981 dir_ino, &index);
3982 if (ret) {
3983 btrfs_info(root->fs_info,
3984 "failed to delete reference to %.*s, inode %llu parent %llu",
3985 name_len, name, ino, dir_ino);
3986 btrfs_abort_transaction(trans, root, ret);
3987 goto err;
3988 }
3989 skip_backref:
3990 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3991 if (ret) {
3992 btrfs_abort_transaction(trans, root, ret);
3993 goto err;
3994 }
3995
3996 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3997 inode, dir_ino);
3998 if (ret != 0 && ret != -ENOENT) {
3999 btrfs_abort_transaction(trans, root, ret);
4000 goto err;
4001 }
4002
4003 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4004 dir, index);
4005 if (ret == -ENOENT)
4006 ret = 0;
4007 else if (ret)
4008 btrfs_abort_transaction(trans, root, ret);
4009 err:
4010 btrfs_free_path(path);
4011 if (ret)
4012 goto out;
4013
4014 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4015 inode_inc_iversion(inode);
4016 inode_inc_iversion(dir);
4017 inode->i_ctime = dir->i_mtime =
4018 dir->i_ctime = current_fs_time(inode->i_sb);
4019 ret = btrfs_update_inode(trans, root, dir);
4020 out:
4021 return ret;
4022 }
4023
4024 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4025 struct btrfs_root *root,
4026 struct inode *dir, struct inode *inode,
4027 const char *name, int name_len)
4028 {
4029 int ret;
4030 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4031 if (!ret) {
4032 drop_nlink(inode);
4033 ret = btrfs_update_inode(trans, root, inode);
4034 }
4035 return ret;
4036 }
4037
4038 /*
4039 * helper to start transaction for unlink and rmdir.
4040 *
4041 * unlink and rmdir are special in btrfs, they do not always free space, so
4042 * if we cannot make our reservations the normal way try and see if there is
4043 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4044 * allow the unlink to occur.
4045 */
4046 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4047 {
4048 struct btrfs_root *root = BTRFS_I(dir)->root;
4049
4050 /*
4051 * 1 for the possible orphan item
4052 * 1 for the dir item
4053 * 1 for the dir index
4054 * 1 for the inode ref
4055 * 1 for the inode
4056 */
4057 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4058 }
4059
4060 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4061 {
4062 struct btrfs_root *root = BTRFS_I(dir)->root;
4063 struct btrfs_trans_handle *trans;
4064 struct inode *inode = d_inode(dentry);
4065 int ret;
4066
4067 trans = __unlink_start_trans(dir);
4068 if (IS_ERR(trans))
4069 return PTR_ERR(trans);
4070
4071 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4072
4073 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4074 dentry->d_name.name, dentry->d_name.len);
4075 if (ret)
4076 goto out;
4077
4078 if (inode->i_nlink == 0) {
4079 ret = btrfs_orphan_add(trans, inode);
4080 if (ret)
4081 goto out;
4082 }
4083
4084 out:
4085 btrfs_end_transaction(trans, root);
4086 btrfs_btree_balance_dirty(root);
4087 return ret;
4088 }
4089
4090 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4091 struct btrfs_root *root,
4092 struct inode *dir, u64 objectid,
4093 const char *name, int name_len)
4094 {
4095 struct btrfs_path *path;
4096 struct extent_buffer *leaf;
4097 struct btrfs_dir_item *di;
4098 struct btrfs_key key;
4099 u64 index;
4100 int ret;
4101 u64 dir_ino = btrfs_ino(dir);
4102
4103 path = btrfs_alloc_path();
4104 if (!path)
4105 return -ENOMEM;
4106
4107 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4108 name, name_len, -1);
4109 if (IS_ERR_OR_NULL(di)) {
4110 if (!di)
4111 ret = -ENOENT;
4112 else
4113 ret = PTR_ERR(di);
4114 goto out;
4115 }
4116
4117 leaf = path->nodes[0];
4118 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4119 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4120 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4121 if (ret) {
4122 btrfs_abort_transaction(trans, root, ret);
4123 goto out;
4124 }
4125 btrfs_release_path(path);
4126
4127 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4128 objectid, root->root_key.objectid,
4129 dir_ino, &index, name, name_len);
4130 if (ret < 0) {
4131 if (ret != -ENOENT) {
4132 btrfs_abort_transaction(trans, root, ret);
4133 goto out;
4134 }
4135 di = btrfs_search_dir_index_item(root, path, dir_ino,
4136 name, name_len);
4137 if (IS_ERR_OR_NULL(di)) {
4138 if (!di)
4139 ret = -ENOENT;
4140 else
4141 ret = PTR_ERR(di);
4142 btrfs_abort_transaction(trans, root, ret);
4143 goto out;
4144 }
4145
4146 leaf = path->nodes[0];
4147 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4148 btrfs_release_path(path);
4149 index = key.offset;
4150 }
4151 btrfs_release_path(path);
4152
4153 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4154 if (ret) {
4155 btrfs_abort_transaction(trans, root, ret);
4156 goto out;
4157 }
4158
4159 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4160 inode_inc_iversion(dir);
4161 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4162 ret = btrfs_update_inode_fallback(trans, root, dir);
4163 if (ret)
4164 btrfs_abort_transaction(trans, root, ret);
4165 out:
4166 btrfs_free_path(path);
4167 return ret;
4168 }
4169
4170 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4171 {
4172 struct inode *inode = d_inode(dentry);
4173 int err = 0;
4174 struct btrfs_root *root = BTRFS_I(dir)->root;
4175 struct btrfs_trans_handle *trans;
4176
4177 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4178 return -ENOTEMPTY;
4179 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4180 return -EPERM;
4181
4182 trans = __unlink_start_trans(dir);
4183 if (IS_ERR(trans))
4184 return PTR_ERR(trans);
4185
4186 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4187 err = btrfs_unlink_subvol(trans, root, dir,
4188 BTRFS_I(inode)->location.objectid,
4189 dentry->d_name.name,
4190 dentry->d_name.len);
4191 goto out;
4192 }
4193
4194 err = btrfs_orphan_add(trans, inode);
4195 if (err)
4196 goto out;
4197
4198 /* now the directory is empty */
4199 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4200 dentry->d_name.name, dentry->d_name.len);
4201 if (!err)
4202 btrfs_i_size_write(inode, 0);
4203 out:
4204 btrfs_end_transaction(trans, root);
4205 btrfs_btree_balance_dirty(root);
4206
4207 return err;
4208 }
4209
4210 static int truncate_space_check(struct btrfs_trans_handle *trans,
4211 struct btrfs_root *root,
4212 u64 bytes_deleted)
4213 {
4214 int ret;
4215
4216 /*
4217 * This is only used to apply pressure to the enospc system, we don't
4218 * intend to use this reservation at all.
4219 */
4220 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4221 bytes_deleted *= root->nodesize;
4222 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4223 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4224 if (!ret) {
4225 trace_btrfs_space_reservation(root->fs_info, "transaction",
4226 trans->transid,
4227 bytes_deleted, 1);
4228 trans->bytes_reserved += bytes_deleted;
4229 }
4230 return ret;
4231
4232 }
4233
4234 static int truncate_inline_extent(struct inode *inode,
4235 struct btrfs_path *path,
4236 struct btrfs_key *found_key,
4237 const u64 item_end,
4238 const u64 new_size)
4239 {
4240 struct extent_buffer *leaf = path->nodes[0];
4241 int slot = path->slots[0];
4242 struct btrfs_file_extent_item *fi;
4243 u32 size = (u32)(new_size - found_key->offset);
4244 struct btrfs_root *root = BTRFS_I(inode)->root;
4245
4246 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4247
4248 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4249 loff_t offset = new_size;
4250 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4251
4252 /*
4253 * Zero out the remaining of the last page of our inline extent,
4254 * instead of directly truncating our inline extent here - that
4255 * would be much more complex (decompressing all the data, then
4256 * compressing the truncated data, which might be bigger than
4257 * the size of the inline extent, resize the extent, etc).
4258 * We release the path because to get the page we might need to
4259 * read the extent item from disk (data not in the page cache).
4260 */
4261 btrfs_release_path(path);
4262 return btrfs_truncate_block(inode, offset, page_end - offset,
4263 0);
4264 }
4265
4266 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4267 size = btrfs_file_extent_calc_inline_size(size);
4268 btrfs_truncate_item(root, path, size, 1);
4269
4270 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4271 inode_sub_bytes(inode, item_end + 1 - new_size);
4272
4273 return 0;
4274 }
4275
4276 /*
4277 * this can truncate away extent items, csum items and directory items.
4278 * It starts at a high offset and removes keys until it can't find
4279 * any higher than new_size
4280 *
4281 * csum items that cross the new i_size are truncated to the new size
4282 * as well.
4283 *
4284 * min_type is the minimum key type to truncate down to. If set to 0, this
4285 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4286 */
4287 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4288 struct btrfs_root *root,
4289 struct inode *inode,
4290 u64 new_size, u32 min_type)
4291 {
4292 struct btrfs_path *path;
4293 struct extent_buffer *leaf;
4294 struct btrfs_file_extent_item *fi;
4295 struct btrfs_key key;
4296 struct btrfs_key found_key;
4297 u64 extent_start = 0;
4298 u64 extent_num_bytes = 0;
4299 u64 extent_offset = 0;
4300 u64 item_end = 0;
4301 u64 last_size = new_size;
4302 u32 found_type = (u8)-1;
4303 int found_extent;
4304 int del_item;
4305 int pending_del_nr = 0;
4306 int pending_del_slot = 0;
4307 int extent_type = -1;
4308 int ret;
4309 int err = 0;
4310 u64 ino = btrfs_ino(inode);
4311 u64 bytes_deleted = 0;
4312 bool be_nice = 0;
4313 bool should_throttle = 0;
4314 bool should_end = 0;
4315
4316 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4317
4318 /*
4319 * for non-free space inodes and ref cows, we want to back off from
4320 * time to time
4321 */
4322 if (!btrfs_is_free_space_inode(inode) &&
4323 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4324 be_nice = 1;
4325
4326 path = btrfs_alloc_path();
4327 if (!path)
4328 return -ENOMEM;
4329 path->reada = READA_BACK;
4330
4331 /*
4332 * We want to drop from the next block forward in case this new size is
4333 * not block aligned since we will be keeping the last block of the
4334 * extent just the way it is.
4335 */
4336 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4337 root == root->fs_info->tree_root)
4338 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4339 root->sectorsize), (u64)-1, 0);
4340
4341 /*
4342 * This function is also used to drop the items in the log tree before
4343 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4344 * it is used to drop the loged items. So we shouldn't kill the delayed
4345 * items.
4346 */
4347 if (min_type == 0 && root == BTRFS_I(inode)->root)
4348 btrfs_kill_delayed_inode_items(inode);
4349
4350 key.objectid = ino;
4351 key.offset = (u64)-1;
4352 key.type = (u8)-1;
4353
4354 search_again:
4355 /*
4356 * with a 16K leaf size and 128MB extents, you can actually queue
4357 * up a huge file in a single leaf. Most of the time that
4358 * bytes_deleted is > 0, it will be huge by the time we get here
4359 */
4360 if (be_nice && bytes_deleted > SZ_32M) {
4361 if (btrfs_should_end_transaction(trans, root)) {
4362 err = -EAGAIN;
4363 goto error;
4364 }
4365 }
4366
4367
4368 path->leave_spinning = 1;
4369 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4370 if (ret < 0) {
4371 err = ret;
4372 goto out;
4373 }
4374
4375 if (ret > 0) {
4376 /* there are no items in the tree for us to truncate, we're
4377 * done
4378 */
4379 if (path->slots[0] == 0)
4380 goto out;
4381 path->slots[0]--;
4382 }
4383
4384 while (1) {
4385 fi = NULL;
4386 leaf = path->nodes[0];
4387 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4388 found_type = found_key.type;
4389
4390 if (found_key.objectid != ino)
4391 break;
4392
4393 if (found_type < min_type)
4394 break;
4395
4396 item_end = found_key.offset;
4397 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4398 fi = btrfs_item_ptr(leaf, path->slots[0],
4399 struct btrfs_file_extent_item);
4400 extent_type = btrfs_file_extent_type(leaf, fi);
4401 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4402 item_end +=
4403 btrfs_file_extent_num_bytes(leaf, fi);
4404 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4405 item_end += btrfs_file_extent_inline_len(leaf,
4406 path->slots[0], fi);
4407 }
4408 item_end--;
4409 }
4410 if (found_type > min_type) {
4411 del_item = 1;
4412 } else {
4413 if (item_end < new_size)
4414 break;
4415 if (found_key.offset >= new_size)
4416 del_item = 1;
4417 else
4418 del_item = 0;
4419 }
4420 found_extent = 0;
4421 /* FIXME, shrink the extent if the ref count is only 1 */
4422 if (found_type != BTRFS_EXTENT_DATA_KEY)
4423 goto delete;
4424
4425 if (del_item)
4426 last_size = found_key.offset;
4427 else
4428 last_size = new_size;
4429
4430 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4431 u64 num_dec;
4432 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4433 if (!del_item) {
4434 u64 orig_num_bytes =
4435 btrfs_file_extent_num_bytes(leaf, fi);
4436 extent_num_bytes = ALIGN(new_size -
4437 found_key.offset,
4438 root->sectorsize);
4439 btrfs_set_file_extent_num_bytes(leaf, fi,
4440 extent_num_bytes);
4441 num_dec = (orig_num_bytes -
4442 extent_num_bytes);
4443 if (test_bit(BTRFS_ROOT_REF_COWS,
4444 &root->state) &&
4445 extent_start != 0)
4446 inode_sub_bytes(inode, num_dec);
4447 btrfs_mark_buffer_dirty(leaf);
4448 } else {
4449 extent_num_bytes =
4450 btrfs_file_extent_disk_num_bytes(leaf,
4451 fi);
4452 extent_offset = found_key.offset -
4453 btrfs_file_extent_offset(leaf, fi);
4454
4455 /* FIXME blocksize != 4096 */
4456 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4457 if (extent_start != 0) {
4458 found_extent = 1;
4459 if (test_bit(BTRFS_ROOT_REF_COWS,
4460 &root->state))
4461 inode_sub_bytes(inode, num_dec);
4462 }
4463 }
4464 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4465 /*
4466 * we can't truncate inline items that have had
4467 * special encodings
4468 */
4469 if (!del_item &&
4470 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4471 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4472
4473 /*
4474 * Need to release path in order to truncate a
4475 * compressed extent. So delete any accumulated
4476 * extent items so far.
4477 */
4478 if (btrfs_file_extent_compression(leaf, fi) !=
4479 BTRFS_COMPRESS_NONE && pending_del_nr) {
4480 err = btrfs_del_items(trans, root, path,
4481 pending_del_slot,
4482 pending_del_nr);
4483 if (err) {
4484 btrfs_abort_transaction(trans,
4485 root,
4486 err);
4487 goto error;
4488 }
4489 pending_del_nr = 0;
4490 }
4491
4492 err = truncate_inline_extent(inode, path,
4493 &found_key,
4494 item_end,
4495 new_size);
4496 if (err) {
4497 btrfs_abort_transaction(trans,
4498 root, err);
4499 goto error;
4500 }
4501 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4502 &root->state)) {
4503 inode_sub_bytes(inode, item_end + 1 - new_size);
4504 }
4505 }
4506 delete:
4507 if (del_item) {
4508 if (!pending_del_nr) {
4509 /* no pending yet, add ourselves */
4510 pending_del_slot = path->slots[0];
4511 pending_del_nr = 1;
4512 } else if (pending_del_nr &&
4513 path->slots[0] + 1 == pending_del_slot) {
4514 /* hop on the pending chunk */
4515 pending_del_nr++;
4516 pending_del_slot = path->slots[0];
4517 } else {
4518 BUG();
4519 }
4520 } else {
4521 break;
4522 }
4523 should_throttle = 0;
4524
4525 if (found_extent &&
4526 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4527 root == root->fs_info->tree_root)) {
4528 btrfs_set_path_blocking(path);
4529 bytes_deleted += extent_num_bytes;
4530 ret = btrfs_free_extent(trans, root, extent_start,
4531 extent_num_bytes, 0,
4532 btrfs_header_owner(leaf),
4533 ino, extent_offset);
4534 BUG_ON(ret);
4535 if (btrfs_should_throttle_delayed_refs(trans, root))
4536 btrfs_async_run_delayed_refs(root,
4537 trans->delayed_ref_updates * 2, 0);
4538 if (be_nice) {
4539 if (truncate_space_check(trans, root,
4540 extent_num_bytes)) {
4541 should_end = 1;
4542 }
4543 if (btrfs_should_throttle_delayed_refs(trans,
4544 root)) {
4545 should_throttle = 1;
4546 }
4547 }
4548 }
4549
4550 if (found_type == BTRFS_INODE_ITEM_KEY)
4551 break;
4552
4553 if (path->slots[0] == 0 ||
4554 path->slots[0] != pending_del_slot ||
4555 should_throttle || should_end) {
4556 if (pending_del_nr) {
4557 ret = btrfs_del_items(trans, root, path,
4558 pending_del_slot,
4559 pending_del_nr);
4560 if (ret) {
4561 btrfs_abort_transaction(trans,
4562 root, ret);
4563 goto error;
4564 }
4565 pending_del_nr = 0;
4566 }
4567 btrfs_release_path(path);
4568 if (should_throttle) {
4569 unsigned long updates = trans->delayed_ref_updates;
4570 if (updates) {
4571 trans->delayed_ref_updates = 0;
4572 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4573 if (ret && !err)
4574 err = ret;
4575 }
4576 }
4577 /*
4578 * if we failed to refill our space rsv, bail out
4579 * and let the transaction restart
4580 */
4581 if (should_end) {
4582 err = -EAGAIN;
4583 goto error;
4584 }
4585 goto search_again;
4586 } else {
4587 path->slots[0]--;
4588 }
4589 }
4590 out:
4591 if (pending_del_nr) {
4592 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4593 pending_del_nr);
4594 if (ret)
4595 btrfs_abort_transaction(trans, root, ret);
4596 }
4597 error:
4598 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4599 btrfs_ordered_update_i_size(inode, last_size, NULL);
4600
4601 btrfs_free_path(path);
4602
4603 if (be_nice && bytes_deleted > SZ_32M) {
4604 unsigned long updates = trans->delayed_ref_updates;
4605 if (updates) {
4606 trans->delayed_ref_updates = 0;
4607 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4608 if (ret && !err)
4609 err = ret;
4610 }
4611 }
4612 return err;
4613 }
4614
4615 /*
4616 * btrfs_truncate_block - read, zero a chunk and write a block
4617 * @inode - inode that we're zeroing
4618 * @from - the offset to start zeroing
4619 * @len - the length to zero, 0 to zero the entire range respective to the
4620 * offset
4621 * @front - zero up to the offset instead of from the offset on
4622 *
4623 * This will find the block for the "from" offset and cow the block and zero the
4624 * part we want to zero. This is used with truncate and hole punching.
4625 */
4626 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4627 int front)
4628 {
4629 struct address_space *mapping = inode->i_mapping;
4630 struct btrfs_root *root = BTRFS_I(inode)->root;
4631 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4632 struct btrfs_ordered_extent *ordered;
4633 struct extent_state *cached_state = NULL;
4634 char *kaddr;
4635 u32 blocksize = root->sectorsize;
4636 pgoff_t index = from >> PAGE_SHIFT;
4637 unsigned offset = from & (blocksize - 1);
4638 struct page *page;
4639 gfp_t mask = btrfs_alloc_write_mask(mapping);
4640 int ret = 0;
4641 u64 block_start;
4642 u64 block_end;
4643
4644 if ((offset & (blocksize - 1)) == 0 &&
4645 (!len || ((len & (blocksize - 1)) == 0)))
4646 goto out;
4647
4648 ret = btrfs_delalloc_reserve_space(inode,
4649 round_down(from, blocksize), blocksize);
4650 if (ret)
4651 goto out;
4652
4653 again:
4654 page = find_or_create_page(mapping, index, mask);
4655 if (!page) {
4656 btrfs_delalloc_release_space(inode,
4657 round_down(from, blocksize),
4658 blocksize);
4659 ret = -ENOMEM;
4660 goto out;
4661 }
4662
4663 block_start = round_down(from, blocksize);
4664 block_end = block_start + blocksize - 1;
4665
4666 if (!PageUptodate(page)) {
4667 ret = btrfs_readpage(NULL, page);
4668 lock_page(page);
4669 if (page->mapping != mapping) {
4670 unlock_page(page);
4671 put_page(page);
4672 goto again;
4673 }
4674 if (!PageUptodate(page)) {
4675 ret = -EIO;
4676 goto out_unlock;
4677 }
4678 }
4679 wait_on_page_writeback(page);
4680
4681 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4682 set_page_extent_mapped(page);
4683
4684 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4685 if (ordered) {
4686 unlock_extent_cached(io_tree, block_start, block_end,
4687 &cached_state, GFP_NOFS);
4688 unlock_page(page);
4689 put_page(page);
4690 btrfs_start_ordered_extent(inode, ordered, 1);
4691 btrfs_put_ordered_extent(ordered);
4692 goto again;
4693 }
4694
4695 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4696 EXTENT_DIRTY | EXTENT_DELALLOC |
4697 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4698 0, 0, &cached_state, GFP_NOFS);
4699
4700 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4701 &cached_state);
4702 if (ret) {
4703 unlock_extent_cached(io_tree, block_start, block_end,
4704 &cached_state, GFP_NOFS);
4705 goto out_unlock;
4706 }
4707
4708 if (offset != blocksize) {
4709 if (!len)
4710 len = blocksize - offset;
4711 kaddr = kmap(page);
4712 if (front)
4713 memset(kaddr + (block_start - page_offset(page)),
4714 0, offset);
4715 else
4716 memset(kaddr + (block_start - page_offset(page)) + offset,
4717 0, len);
4718 flush_dcache_page(page);
4719 kunmap(page);
4720 }
4721 ClearPageChecked(page);
4722 set_page_dirty(page);
4723 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4724 GFP_NOFS);
4725
4726 out_unlock:
4727 if (ret)
4728 btrfs_delalloc_release_space(inode, block_start,
4729 blocksize);
4730 unlock_page(page);
4731 put_page(page);
4732 out:
4733 return ret;
4734 }
4735
4736 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4737 u64 offset, u64 len)
4738 {
4739 struct btrfs_trans_handle *trans;
4740 int ret;
4741
4742 /*
4743 * Still need to make sure the inode looks like it's been updated so
4744 * that any holes get logged if we fsync.
4745 */
4746 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4747 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4748 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4749 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4750 return 0;
4751 }
4752
4753 /*
4754 * 1 - for the one we're dropping
4755 * 1 - for the one we're adding
4756 * 1 - for updating the inode.
4757 */
4758 trans = btrfs_start_transaction(root, 3);
4759 if (IS_ERR(trans))
4760 return PTR_ERR(trans);
4761
4762 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4763 if (ret) {
4764 btrfs_abort_transaction(trans, root, ret);
4765 btrfs_end_transaction(trans, root);
4766 return ret;
4767 }
4768
4769 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4770 0, 0, len, 0, len, 0, 0, 0);
4771 if (ret)
4772 btrfs_abort_transaction(trans, root, ret);
4773 else
4774 btrfs_update_inode(trans, root, inode);
4775 btrfs_end_transaction(trans, root);
4776 return ret;
4777 }
4778
4779 /*
4780 * This function puts in dummy file extents for the area we're creating a hole
4781 * for. So if we are truncating this file to a larger size we need to insert
4782 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4783 * the range between oldsize and size
4784 */
4785 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4786 {
4787 struct btrfs_root *root = BTRFS_I(inode)->root;
4788 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4789 struct extent_map *em = NULL;
4790 struct extent_state *cached_state = NULL;
4791 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4792 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4793 u64 block_end = ALIGN(size, root->sectorsize);
4794 u64 last_byte;
4795 u64 cur_offset;
4796 u64 hole_size;
4797 int err = 0;
4798
4799 /*
4800 * If our size started in the middle of a block we need to zero out the
4801 * rest of the block before we expand the i_size, otherwise we could
4802 * expose stale data.
4803 */
4804 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4805 if (err)
4806 return err;
4807
4808 if (size <= hole_start)
4809 return 0;
4810
4811 while (1) {
4812 struct btrfs_ordered_extent *ordered;
4813
4814 lock_extent_bits(io_tree, hole_start, block_end - 1,
4815 &cached_state);
4816 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4817 block_end - hole_start);
4818 if (!ordered)
4819 break;
4820 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4821 &cached_state, GFP_NOFS);
4822 btrfs_start_ordered_extent(inode, ordered, 1);
4823 btrfs_put_ordered_extent(ordered);
4824 }
4825
4826 cur_offset = hole_start;
4827 while (1) {
4828 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4829 block_end - cur_offset, 0);
4830 if (IS_ERR(em)) {
4831 err = PTR_ERR(em);
4832 em = NULL;
4833 break;
4834 }
4835 last_byte = min(extent_map_end(em), block_end);
4836 last_byte = ALIGN(last_byte , root->sectorsize);
4837 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4838 struct extent_map *hole_em;
4839 hole_size = last_byte - cur_offset;
4840
4841 err = maybe_insert_hole(root, inode, cur_offset,
4842 hole_size);
4843 if (err)
4844 break;
4845 btrfs_drop_extent_cache(inode, cur_offset,
4846 cur_offset + hole_size - 1, 0);
4847 hole_em = alloc_extent_map();
4848 if (!hole_em) {
4849 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4850 &BTRFS_I(inode)->runtime_flags);
4851 goto next;
4852 }
4853 hole_em->start = cur_offset;
4854 hole_em->len = hole_size;
4855 hole_em->orig_start = cur_offset;
4856
4857 hole_em->block_start = EXTENT_MAP_HOLE;
4858 hole_em->block_len = 0;
4859 hole_em->orig_block_len = 0;
4860 hole_em->ram_bytes = hole_size;
4861 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4862 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4863 hole_em->generation = root->fs_info->generation;
4864
4865 while (1) {
4866 write_lock(&em_tree->lock);
4867 err = add_extent_mapping(em_tree, hole_em, 1);
4868 write_unlock(&em_tree->lock);
4869 if (err != -EEXIST)
4870 break;
4871 btrfs_drop_extent_cache(inode, cur_offset,
4872 cur_offset +
4873 hole_size - 1, 0);
4874 }
4875 free_extent_map(hole_em);
4876 }
4877 next:
4878 free_extent_map(em);
4879 em = NULL;
4880 cur_offset = last_byte;
4881 if (cur_offset >= block_end)
4882 break;
4883 }
4884 free_extent_map(em);
4885 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4886 GFP_NOFS);
4887 return err;
4888 }
4889
4890 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4891 {
4892 struct btrfs_root *root = BTRFS_I(inode)->root;
4893 struct btrfs_trans_handle *trans;
4894 loff_t oldsize = i_size_read(inode);
4895 loff_t newsize = attr->ia_size;
4896 int mask = attr->ia_valid;
4897 int ret;
4898
4899 /*
4900 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4901 * special case where we need to update the times despite not having
4902 * these flags set. For all other operations the VFS set these flags
4903 * explicitly if it wants a timestamp update.
4904 */
4905 if (newsize != oldsize) {
4906 inode_inc_iversion(inode);
4907 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4908 inode->i_ctime = inode->i_mtime =
4909 current_fs_time(inode->i_sb);
4910 }
4911
4912 if (newsize > oldsize) {
4913 /*
4914 * Don't do an expanding truncate while snapshoting is ongoing.
4915 * This is to ensure the snapshot captures a fully consistent
4916 * state of this file - if the snapshot captures this expanding
4917 * truncation, it must capture all writes that happened before
4918 * this truncation.
4919 */
4920 btrfs_wait_for_snapshot_creation(root);
4921 ret = btrfs_cont_expand(inode, oldsize, newsize);
4922 if (ret) {
4923 btrfs_end_write_no_snapshoting(root);
4924 return ret;
4925 }
4926
4927 trans = btrfs_start_transaction(root, 1);
4928 if (IS_ERR(trans)) {
4929 btrfs_end_write_no_snapshoting(root);
4930 return PTR_ERR(trans);
4931 }
4932
4933 i_size_write(inode, newsize);
4934 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4935 pagecache_isize_extended(inode, oldsize, newsize);
4936 ret = btrfs_update_inode(trans, root, inode);
4937 btrfs_end_write_no_snapshoting(root);
4938 btrfs_end_transaction(trans, root);
4939 } else {
4940
4941 /*
4942 * We're truncating a file that used to have good data down to
4943 * zero. Make sure it gets into the ordered flush list so that
4944 * any new writes get down to disk quickly.
4945 */
4946 if (newsize == 0)
4947 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4948 &BTRFS_I(inode)->runtime_flags);
4949
4950 /*
4951 * 1 for the orphan item we're going to add
4952 * 1 for the orphan item deletion.
4953 */
4954 trans = btrfs_start_transaction(root, 2);
4955 if (IS_ERR(trans))
4956 return PTR_ERR(trans);
4957
4958 /*
4959 * We need to do this in case we fail at _any_ point during the
4960 * actual truncate. Once we do the truncate_setsize we could
4961 * invalidate pages which forces any outstanding ordered io to
4962 * be instantly completed which will give us extents that need
4963 * to be truncated. If we fail to get an orphan inode down we
4964 * could have left over extents that were never meant to live,
4965 * so we need to garuntee from this point on that everything
4966 * will be consistent.
4967 */
4968 ret = btrfs_orphan_add(trans, inode);
4969 btrfs_end_transaction(trans, root);
4970 if (ret)
4971 return ret;
4972
4973 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4974 truncate_setsize(inode, newsize);
4975
4976 /* Disable nonlocked read DIO to avoid the end less truncate */
4977 btrfs_inode_block_unlocked_dio(inode);
4978 inode_dio_wait(inode);
4979 btrfs_inode_resume_unlocked_dio(inode);
4980
4981 ret = btrfs_truncate(inode);
4982 if (ret && inode->i_nlink) {
4983 int err;
4984
4985 /*
4986 * failed to truncate, disk_i_size is only adjusted down
4987 * as we remove extents, so it should represent the true
4988 * size of the inode, so reset the in memory size and
4989 * delete our orphan entry.
4990 */
4991 trans = btrfs_join_transaction(root);
4992 if (IS_ERR(trans)) {
4993 btrfs_orphan_del(NULL, inode);
4994 return ret;
4995 }
4996 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4997 err = btrfs_orphan_del(trans, inode);
4998 if (err)
4999 btrfs_abort_transaction(trans, root, err);
5000 btrfs_end_transaction(trans, root);
5001 }
5002 }
5003
5004 return ret;
5005 }
5006
5007 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5008 {
5009 struct inode *inode = d_inode(dentry);
5010 struct btrfs_root *root = BTRFS_I(inode)->root;
5011 int err;
5012
5013 if (btrfs_root_readonly(root))
5014 return -EROFS;
5015
5016 err = inode_change_ok(inode, attr);
5017 if (err)
5018 return err;
5019
5020 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5021 err = btrfs_setsize(inode, attr);
5022 if (err)
5023 return err;
5024 }
5025
5026 if (attr->ia_valid) {
5027 setattr_copy(inode, attr);
5028 inode_inc_iversion(inode);
5029 err = btrfs_dirty_inode(inode);
5030
5031 if (!err && attr->ia_valid & ATTR_MODE)
5032 err = posix_acl_chmod(inode, inode->i_mode);
5033 }
5034
5035 return err;
5036 }
5037
5038 /*
5039 * While truncating the inode pages during eviction, we get the VFS calling
5040 * btrfs_invalidatepage() against each page of the inode. This is slow because
5041 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5042 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5043 * extent_state structures over and over, wasting lots of time.
5044 *
5045 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5046 * those expensive operations on a per page basis and do only the ordered io
5047 * finishing, while we release here the extent_map and extent_state structures,
5048 * without the excessive merging and splitting.
5049 */
5050 static void evict_inode_truncate_pages(struct inode *inode)
5051 {
5052 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5053 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5054 struct rb_node *node;
5055
5056 ASSERT(inode->i_state & I_FREEING);
5057 truncate_inode_pages_final(&inode->i_data);
5058
5059 write_lock(&map_tree->lock);
5060 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5061 struct extent_map *em;
5062
5063 node = rb_first(&map_tree->map);
5064 em = rb_entry(node, struct extent_map, rb_node);
5065 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5066 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5067 remove_extent_mapping(map_tree, em);
5068 free_extent_map(em);
5069 if (need_resched()) {
5070 write_unlock(&map_tree->lock);
5071 cond_resched();
5072 write_lock(&map_tree->lock);
5073 }
5074 }
5075 write_unlock(&map_tree->lock);
5076
5077 /*
5078 * Keep looping until we have no more ranges in the io tree.
5079 * We can have ongoing bios started by readpages (called from readahead)
5080 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5081 * still in progress (unlocked the pages in the bio but did not yet
5082 * unlocked the ranges in the io tree). Therefore this means some
5083 * ranges can still be locked and eviction started because before
5084 * submitting those bios, which are executed by a separate task (work
5085 * queue kthread), inode references (inode->i_count) were not taken
5086 * (which would be dropped in the end io callback of each bio).
5087 * Therefore here we effectively end up waiting for those bios and
5088 * anyone else holding locked ranges without having bumped the inode's
5089 * reference count - if we don't do it, when they access the inode's
5090 * io_tree to unlock a range it may be too late, leading to an
5091 * use-after-free issue.
5092 */
5093 spin_lock(&io_tree->lock);
5094 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5095 struct extent_state *state;
5096 struct extent_state *cached_state = NULL;
5097 u64 start;
5098 u64 end;
5099
5100 node = rb_first(&io_tree->state);
5101 state = rb_entry(node, struct extent_state, rb_node);
5102 start = state->start;
5103 end = state->end;
5104 spin_unlock(&io_tree->lock);
5105
5106 lock_extent_bits(io_tree, start, end, &cached_state);
5107
5108 /*
5109 * If still has DELALLOC flag, the extent didn't reach disk,
5110 * and its reserved space won't be freed by delayed_ref.
5111 * So we need to free its reserved space here.
5112 * (Refer to comment in btrfs_invalidatepage, case 2)
5113 *
5114 * Note, end is the bytenr of last byte, so we need + 1 here.
5115 */
5116 if (state->state & EXTENT_DELALLOC)
5117 btrfs_qgroup_free_data(inode, start, end - start + 1);
5118
5119 clear_extent_bit(io_tree, start, end,
5120 EXTENT_LOCKED | EXTENT_DIRTY |
5121 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5122 EXTENT_DEFRAG, 1, 1,
5123 &cached_state, GFP_NOFS);
5124
5125 cond_resched();
5126 spin_lock(&io_tree->lock);
5127 }
5128 spin_unlock(&io_tree->lock);
5129 }
5130
5131 void btrfs_evict_inode(struct inode *inode)
5132 {
5133 struct btrfs_trans_handle *trans;
5134 struct btrfs_root *root = BTRFS_I(inode)->root;
5135 struct btrfs_block_rsv *rsv, *global_rsv;
5136 int steal_from_global = 0;
5137 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5138 int ret;
5139
5140 trace_btrfs_inode_evict(inode);
5141
5142 evict_inode_truncate_pages(inode);
5143
5144 if (inode->i_nlink &&
5145 ((btrfs_root_refs(&root->root_item) != 0 &&
5146 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5147 btrfs_is_free_space_inode(inode)))
5148 goto no_delete;
5149
5150 if (is_bad_inode(inode)) {
5151 btrfs_orphan_del(NULL, inode);
5152 goto no_delete;
5153 }
5154 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5155 if (!special_file(inode->i_mode))
5156 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5157
5158 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5159
5160 if (root->fs_info->log_root_recovering) {
5161 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5162 &BTRFS_I(inode)->runtime_flags));
5163 goto no_delete;
5164 }
5165
5166 if (inode->i_nlink > 0) {
5167 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5168 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5169 goto no_delete;
5170 }
5171
5172 ret = btrfs_commit_inode_delayed_inode(inode);
5173 if (ret) {
5174 btrfs_orphan_del(NULL, inode);
5175 goto no_delete;
5176 }
5177
5178 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5179 if (!rsv) {
5180 btrfs_orphan_del(NULL, inode);
5181 goto no_delete;
5182 }
5183 rsv->size = min_size;
5184 rsv->failfast = 1;
5185 global_rsv = &root->fs_info->global_block_rsv;
5186
5187 btrfs_i_size_write(inode, 0);
5188
5189 /*
5190 * This is a bit simpler than btrfs_truncate since we've already
5191 * reserved our space for our orphan item in the unlink, so we just
5192 * need to reserve some slack space in case we add bytes and update
5193 * inode item when doing the truncate.
5194 */
5195 while (1) {
5196 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5197 BTRFS_RESERVE_FLUSH_LIMIT);
5198
5199 /*
5200 * Try and steal from the global reserve since we will
5201 * likely not use this space anyway, we want to try as
5202 * hard as possible to get this to work.
5203 */
5204 if (ret)
5205 steal_from_global++;
5206 else
5207 steal_from_global = 0;
5208 ret = 0;
5209
5210 /*
5211 * steal_from_global == 0: we reserved stuff, hooray!
5212 * steal_from_global == 1: we didn't reserve stuff, boo!
5213 * steal_from_global == 2: we've committed, still not a lot of
5214 * room but maybe we'll have room in the global reserve this
5215 * time.
5216 * steal_from_global == 3: abandon all hope!
5217 */
5218 if (steal_from_global > 2) {
5219 btrfs_warn(root->fs_info,
5220 "Could not get space for a delete, will truncate on mount %d",
5221 ret);
5222 btrfs_orphan_del(NULL, inode);
5223 btrfs_free_block_rsv(root, rsv);
5224 goto no_delete;
5225 }
5226
5227 trans = btrfs_join_transaction(root);
5228 if (IS_ERR(trans)) {
5229 btrfs_orphan_del(NULL, inode);
5230 btrfs_free_block_rsv(root, rsv);
5231 goto no_delete;
5232 }
5233
5234 /*
5235 * We can't just steal from the global reserve, we need tomake
5236 * sure there is room to do it, if not we need to commit and try
5237 * again.
5238 */
5239 if (steal_from_global) {
5240 if (!btrfs_check_space_for_delayed_refs(trans, root))
5241 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5242 min_size);
5243 else
5244 ret = -ENOSPC;
5245 }
5246
5247 /*
5248 * Couldn't steal from the global reserve, we have too much
5249 * pending stuff built up, commit the transaction and try it
5250 * again.
5251 */
5252 if (ret) {
5253 ret = btrfs_commit_transaction(trans, root);
5254 if (ret) {
5255 btrfs_orphan_del(NULL, inode);
5256 btrfs_free_block_rsv(root, rsv);
5257 goto no_delete;
5258 }
5259 continue;
5260 } else {
5261 steal_from_global = 0;
5262 }
5263
5264 trans->block_rsv = rsv;
5265
5266 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5267 if (ret != -ENOSPC && ret != -EAGAIN)
5268 break;
5269
5270 trans->block_rsv = &root->fs_info->trans_block_rsv;
5271 btrfs_end_transaction(trans, root);
5272 trans = NULL;
5273 btrfs_btree_balance_dirty(root);
5274 }
5275
5276 btrfs_free_block_rsv(root, rsv);
5277
5278 /*
5279 * Errors here aren't a big deal, it just means we leave orphan items
5280 * in the tree. They will be cleaned up on the next mount.
5281 */
5282 if (ret == 0) {
5283 trans->block_rsv = root->orphan_block_rsv;
5284 btrfs_orphan_del(trans, inode);
5285 } else {
5286 btrfs_orphan_del(NULL, inode);
5287 }
5288
5289 trans->block_rsv = &root->fs_info->trans_block_rsv;
5290 if (!(root == root->fs_info->tree_root ||
5291 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5292 btrfs_return_ino(root, btrfs_ino(inode));
5293
5294 btrfs_end_transaction(trans, root);
5295 btrfs_btree_balance_dirty(root);
5296 no_delete:
5297 btrfs_remove_delayed_node(inode);
5298 clear_inode(inode);
5299 }
5300
5301 /*
5302 * this returns the key found in the dir entry in the location pointer.
5303 * If no dir entries were found, location->objectid is 0.
5304 */
5305 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5306 struct btrfs_key *location)
5307 {
5308 const char *name = dentry->d_name.name;
5309 int namelen = dentry->d_name.len;
5310 struct btrfs_dir_item *di;
5311 struct btrfs_path *path;
5312 struct btrfs_root *root = BTRFS_I(dir)->root;
5313 int ret = 0;
5314
5315 path = btrfs_alloc_path();
5316 if (!path)
5317 return -ENOMEM;
5318
5319 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5320 namelen, 0);
5321 if (IS_ERR(di))
5322 ret = PTR_ERR(di);
5323
5324 if (IS_ERR_OR_NULL(di))
5325 goto out_err;
5326
5327 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5328 out:
5329 btrfs_free_path(path);
5330 return ret;
5331 out_err:
5332 location->objectid = 0;
5333 goto out;
5334 }
5335
5336 /*
5337 * when we hit a tree root in a directory, the btrfs part of the inode
5338 * needs to be changed to reflect the root directory of the tree root. This
5339 * is kind of like crossing a mount point.
5340 */
5341 static int fixup_tree_root_location(struct btrfs_root *root,
5342 struct inode *dir,
5343 struct dentry *dentry,
5344 struct btrfs_key *location,
5345 struct btrfs_root **sub_root)
5346 {
5347 struct btrfs_path *path;
5348 struct btrfs_root *new_root;
5349 struct btrfs_root_ref *ref;
5350 struct extent_buffer *leaf;
5351 struct btrfs_key key;
5352 int ret;
5353 int err = 0;
5354
5355 path = btrfs_alloc_path();
5356 if (!path) {
5357 err = -ENOMEM;
5358 goto out;
5359 }
5360
5361 err = -ENOENT;
5362 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5363 key.type = BTRFS_ROOT_REF_KEY;
5364 key.offset = location->objectid;
5365
5366 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5367 0, 0);
5368 if (ret) {
5369 if (ret < 0)
5370 err = ret;
5371 goto out;
5372 }
5373
5374 leaf = path->nodes[0];
5375 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5376 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5377 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5378 goto out;
5379
5380 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5381 (unsigned long)(ref + 1),
5382 dentry->d_name.len);
5383 if (ret)
5384 goto out;
5385
5386 btrfs_release_path(path);
5387
5388 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5389 if (IS_ERR(new_root)) {
5390 err = PTR_ERR(new_root);
5391 goto out;
5392 }
5393
5394 *sub_root = new_root;
5395 location->objectid = btrfs_root_dirid(&new_root->root_item);
5396 location->type = BTRFS_INODE_ITEM_KEY;
5397 location->offset = 0;
5398 err = 0;
5399 out:
5400 btrfs_free_path(path);
5401 return err;
5402 }
5403
5404 static void inode_tree_add(struct inode *inode)
5405 {
5406 struct btrfs_root *root = BTRFS_I(inode)->root;
5407 struct btrfs_inode *entry;
5408 struct rb_node **p;
5409 struct rb_node *parent;
5410 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5411 u64 ino = btrfs_ino(inode);
5412
5413 if (inode_unhashed(inode))
5414 return;
5415 parent = NULL;
5416 spin_lock(&root->inode_lock);
5417 p = &root->inode_tree.rb_node;
5418 while (*p) {
5419 parent = *p;
5420 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5421
5422 if (ino < btrfs_ino(&entry->vfs_inode))
5423 p = &parent->rb_left;
5424 else if (ino > btrfs_ino(&entry->vfs_inode))
5425 p = &parent->rb_right;
5426 else {
5427 WARN_ON(!(entry->vfs_inode.i_state &
5428 (I_WILL_FREE | I_FREEING)));
5429 rb_replace_node(parent, new, &root->inode_tree);
5430 RB_CLEAR_NODE(parent);
5431 spin_unlock(&root->inode_lock);
5432 return;
5433 }
5434 }
5435 rb_link_node(new, parent, p);
5436 rb_insert_color(new, &root->inode_tree);
5437 spin_unlock(&root->inode_lock);
5438 }
5439
5440 static void inode_tree_del(struct inode *inode)
5441 {
5442 struct btrfs_root *root = BTRFS_I(inode)->root;
5443 int empty = 0;
5444
5445 spin_lock(&root->inode_lock);
5446 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5447 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5448 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5449 empty = RB_EMPTY_ROOT(&root->inode_tree);
5450 }
5451 spin_unlock(&root->inode_lock);
5452
5453 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5454 synchronize_srcu(&root->fs_info->subvol_srcu);
5455 spin_lock(&root->inode_lock);
5456 empty = RB_EMPTY_ROOT(&root->inode_tree);
5457 spin_unlock(&root->inode_lock);
5458 if (empty)
5459 btrfs_add_dead_root(root);
5460 }
5461 }
5462
5463 void btrfs_invalidate_inodes(struct btrfs_root *root)
5464 {
5465 struct rb_node *node;
5466 struct rb_node *prev;
5467 struct btrfs_inode *entry;
5468 struct inode *inode;
5469 u64 objectid = 0;
5470
5471 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5472 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5473
5474 spin_lock(&root->inode_lock);
5475 again:
5476 node = root->inode_tree.rb_node;
5477 prev = NULL;
5478 while (node) {
5479 prev = node;
5480 entry = rb_entry(node, struct btrfs_inode, rb_node);
5481
5482 if (objectid < btrfs_ino(&entry->vfs_inode))
5483 node = node->rb_left;
5484 else if (objectid > btrfs_ino(&entry->vfs_inode))
5485 node = node->rb_right;
5486 else
5487 break;
5488 }
5489 if (!node) {
5490 while (prev) {
5491 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5492 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5493 node = prev;
5494 break;
5495 }
5496 prev = rb_next(prev);
5497 }
5498 }
5499 while (node) {
5500 entry = rb_entry(node, struct btrfs_inode, rb_node);
5501 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5502 inode = igrab(&entry->vfs_inode);
5503 if (inode) {
5504 spin_unlock(&root->inode_lock);
5505 if (atomic_read(&inode->i_count) > 1)
5506 d_prune_aliases(inode);
5507 /*
5508 * btrfs_drop_inode will have it removed from
5509 * the inode cache when its usage count
5510 * hits zero.
5511 */
5512 iput(inode);
5513 cond_resched();
5514 spin_lock(&root->inode_lock);
5515 goto again;
5516 }
5517
5518 if (cond_resched_lock(&root->inode_lock))
5519 goto again;
5520
5521 node = rb_next(node);
5522 }
5523 spin_unlock(&root->inode_lock);
5524 }
5525
5526 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5527 {
5528 struct btrfs_iget_args *args = p;
5529 inode->i_ino = args->location->objectid;
5530 memcpy(&BTRFS_I(inode)->location, args->location,
5531 sizeof(*args->location));
5532 BTRFS_I(inode)->root = args->root;
5533 return 0;
5534 }
5535
5536 static int btrfs_find_actor(struct inode *inode, void *opaque)
5537 {
5538 struct btrfs_iget_args *args = opaque;
5539 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5540 args->root == BTRFS_I(inode)->root;
5541 }
5542
5543 static struct inode *btrfs_iget_locked(struct super_block *s,
5544 struct btrfs_key *location,
5545 struct btrfs_root *root)
5546 {
5547 struct inode *inode;
5548 struct btrfs_iget_args args;
5549 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5550
5551 args.location = location;
5552 args.root = root;
5553
5554 inode = iget5_locked(s, hashval, btrfs_find_actor,
5555 btrfs_init_locked_inode,
5556 (void *)&args);
5557 return inode;
5558 }
5559
5560 /* Get an inode object given its location and corresponding root.
5561 * Returns in *is_new if the inode was read from disk
5562 */
5563 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5564 struct btrfs_root *root, int *new)
5565 {
5566 struct inode *inode;
5567
5568 inode = btrfs_iget_locked(s, location, root);
5569 if (!inode)
5570 return ERR_PTR(-ENOMEM);
5571
5572 if (inode->i_state & I_NEW) {
5573 btrfs_read_locked_inode(inode);
5574 if (!is_bad_inode(inode)) {
5575 inode_tree_add(inode);
5576 unlock_new_inode(inode);
5577 if (new)
5578 *new = 1;
5579 } else {
5580 unlock_new_inode(inode);
5581 iput(inode);
5582 inode = ERR_PTR(-ESTALE);
5583 }
5584 }
5585
5586 return inode;
5587 }
5588
5589 static struct inode *new_simple_dir(struct super_block *s,
5590 struct btrfs_key *key,
5591 struct btrfs_root *root)
5592 {
5593 struct inode *inode = new_inode(s);
5594
5595 if (!inode)
5596 return ERR_PTR(-ENOMEM);
5597
5598 BTRFS_I(inode)->root = root;
5599 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5600 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5601
5602 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5603 inode->i_op = &btrfs_dir_ro_inode_operations;
5604 inode->i_fop = &simple_dir_operations;
5605 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5606 inode->i_mtime = current_fs_time(inode->i_sb);
5607 inode->i_atime = inode->i_mtime;
5608 inode->i_ctime = inode->i_mtime;
5609 BTRFS_I(inode)->i_otime = inode->i_mtime;
5610
5611 return inode;
5612 }
5613
5614 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5615 {
5616 struct inode *inode;
5617 struct btrfs_root *root = BTRFS_I(dir)->root;
5618 struct btrfs_root *sub_root = root;
5619 struct btrfs_key location;
5620 int index;
5621 int ret = 0;
5622
5623 if (dentry->d_name.len > BTRFS_NAME_LEN)
5624 return ERR_PTR(-ENAMETOOLONG);
5625
5626 ret = btrfs_inode_by_name(dir, dentry, &location);
5627 if (ret < 0)
5628 return ERR_PTR(ret);
5629
5630 if (location.objectid == 0)
5631 return ERR_PTR(-ENOENT);
5632
5633 if (location.type == BTRFS_INODE_ITEM_KEY) {
5634 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5635 return inode;
5636 }
5637
5638 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5639
5640 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5641 ret = fixup_tree_root_location(root, dir, dentry,
5642 &location, &sub_root);
5643 if (ret < 0) {
5644 if (ret != -ENOENT)
5645 inode = ERR_PTR(ret);
5646 else
5647 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5648 } else {
5649 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5650 }
5651 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5652
5653 if (!IS_ERR(inode) && root != sub_root) {
5654 down_read(&root->fs_info->cleanup_work_sem);
5655 if (!(inode->i_sb->s_flags & MS_RDONLY))
5656 ret = btrfs_orphan_cleanup(sub_root);
5657 up_read(&root->fs_info->cleanup_work_sem);
5658 if (ret) {
5659 iput(inode);
5660 inode = ERR_PTR(ret);
5661 }
5662 }
5663
5664 return inode;
5665 }
5666
5667 static int btrfs_dentry_delete(const struct dentry *dentry)
5668 {
5669 struct btrfs_root *root;
5670 struct inode *inode = d_inode(dentry);
5671
5672 if (!inode && !IS_ROOT(dentry))
5673 inode = d_inode(dentry->d_parent);
5674
5675 if (inode) {
5676 root = BTRFS_I(inode)->root;
5677 if (btrfs_root_refs(&root->root_item) == 0)
5678 return 1;
5679
5680 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5681 return 1;
5682 }
5683 return 0;
5684 }
5685
5686 static void btrfs_dentry_release(struct dentry *dentry)
5687 {
5688 kfree(dentry->d_fsdata);
5689 }
5690
5691 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5692 unsigned int flags)
5693 {
5694 struct inode *inode;
5695
5696 inode = btrfs_lookup_dentry(dir, dentry);
5697 if (IS_ERR(inode)) {
5698 if (PTR_ERR(inode) == -ENOENT)
5699 inode = NULL;
5700 else
5701 return ERR_CAST(inode);
5702 }
5703
5704 return d_splice_alias(inode, dentry);
5705 }
5706
5707 unsigned char btrfs_filetype_table[] = {
5708 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5709 };
5710
5711 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5712 {
5713 struct inode *inode = file_inode(file);
5714 struct btrfs_root *root = BTRFS_I(inode)->root;
5715 struct btrfs_item *item;
5716 struct btrfs_dir_item *di;
5717 struct btrfs_key key;
5718 struct btrfs_key found_key;
5719 struct btrfs_path *path;
5720 struct list_head ins_list;
5721 struct list_head del_list;
5722 int ret;
5723 struct extent_buffer *leaf;
5724 int slot;
5725 unsigned char d_type;
5726 int over = 0;
5727 u32 di_cur;
5728 u32 di_total;
5729 u32 di_len;
5730 int key_type = BTRFS_DIR_INDEX_KEY;
5731 char tmp_name[32];
5732 char *name_ptr;
5733 int name_len;
5734 int is_curr = 0; /* ctx->pos points to the current index? */
5735 bool emitted;
5736
5737 /* FIXME, use a real flag for deciding about the key type */
5738 if (root->fs_info->tree_root == root)
5739 key_type = BTRFS_DIR_ITEM_KEY;
5740
5741 if (!dir_emit_dots(file, ctx))
5742 return 0;
5743
5744 path = btrfs_alloc_path();
5745 if (!path)
5746 return -ENOMEM;
5747
5748 path->reada = READA_FORWARD;
5749
5750 if (key_type == BTRFS_DIR_INDEX_KEY) {
5751 INIT_LIST_HEAD(&ins_list);
5752 INIT_LIST_HEAD(&del_list);
5753 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5754 }
5755
5756 key.type = key_type;
5757 key.offset = ctx->pos;
5758 key.objectid = btrfs_ino(inode);
5759
5760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5761 if (ret < 0)
5762 goto err;
5763
5764 emitted = false;
5765 while (1) {
5766 leaf = path->nodes[0];
5767 slot = path->slots[0];
5768 if (slot >= btrfs_header_nritems(leaf)) {
5769 ret = btrfs_next_leaf(root, path);
5770 if (ret < 0)
5771 goto err;
5772 else if (ret > 0)
5773 break;
5774 continue;
5775 }
5776
5777 item = btrfs_item_nr(slot);
5778 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5779
5780 if (found_key.objectid != key.objectid)
5781 break;
5782 if (found_key.type != key_type)
5783 break;
5784 if (found_key.offset < ctx->pos)
5785 goto next;
5786 if (key_type == BTRFS_DIR_INDEX_KEY &&
5787 btrfs_should_delete_dir_index(&del_list,
5788 found_key.offset))
5789 goto next;
5790
5791 ctx->pos = found_key.offset;
5792 is_curr = 1;
5793
5794 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5795 di_cur = 0;
5796 di_total = btrfs_item_size(leaf, item);
5797
5798 while (di_cur < di_total) {
5799 struct btrfs_key location;
5800
5801 if (verify_dir_item(root, leaf, di))
5802 break;
5803
5804 name_len = btrfs_dir_name_len(leaf, di);
5805 if (name_len <= sizeof(tmp_name)) {
5806 name_ptr = tmp_name;
5807 } else {
5808 name_ptr = kmalloc(name_len, GFP_KERNEL);
5809 if (!name_ptr) {
5810 ret = -ENOMEM;
5811 goto err;
5812 }
5813 }
5814 read_extent_buffer(leaf, name_ptr,
5815 (unsigned long)(di + 1), name_len);
5816
5817 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5818 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5819
5820
5821 /* is this a reference to our own snapshot? If so
5822 * skip it.
5823 *
5824 * In contrast to old kernels, we insert the snapshot's
5825 * dir item and dir index after it has been created, so
5826 * we won't find a reference to our own snapshot. We
5827 * still keep the following code for backward
5828 * compatibility.
5829 */
5830 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5831 location.objectid == root->root_key.objectid) {
5832 over = 0;
5833 goto skip;
5834 }
5835 over = !dir_emit(ctx, name_ptr, name_len,
5836 location.objectid, d_type);
5837
5838 skip:
5839 if (name_ptr != tmp_name)
5840 kfree(name_ptr);
5841
5842 if (over)
5843 goto nopos;
5844 emitted = true;
5845 di_len = btrfs_dir_name_len(leaf, di) +
5846 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5847 di_cur += di_len;
5848 di = (struct btrfs_dir_item *)((char *)di + di_len);
5849 }
5850 next:
5851 path->slots[0]++;
5852 }
5853
5854 if (key_type == BTRFS_DIR_INDEX_KEY) {
5855 if (is_curr)
5856 ctx->pos++;
5857 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5858 if (ret)
5859 goto nopos;
5860 }
5861
5862 /*
5863 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5864 * it was was set to the termination value in previous call. We assume
5865 * that "." and ".." were emitted if we reach this point and set the
5866 * termination value as well for an empty directory.
5867 */
5868 if (ctx->pos > 2 && !emitted)
5869 goto nopos;
5870
5871 /* Reached end of directory/root. Bump pos past the last item. */
5872 ctx->pos++;
5873
5874 /*
5875 * Stop new entries from being returned after we return the last
5876 * entry.
5877 *
5878 * New directory entries are assigned a strictly increasing
5879 * offset. This means that new entries created during readdir
5880 * are *guaranteed* to be seen in the future by that readdir.
5881 * This has broken buggy programs which operate on names as
5882 * they're returned by readdir. Until we re-use freed offsets
5883 * we have this hack to stop new entries from being returned
5884 * under the assumption that they'll never reach this huge
5885 * offset.
5886 *
5887 * This is being careful not to overflow 32bit loff_t unless the
5888 * last entry requires it because doing so has broken 32bit apps
5889 * in the past.
5890 */
5891 if (key_type == BTRFS_DIR_INDEX_KEY) {
5892 if (ctx->pos >= INT_MAX)
5893 ctx->pos = LLONG_MAX;
5894 else
5895 ctx->pos = INT_MAX;
5896 }
5897 nopos:
5898 ret = 0;
5899 err:
5900 if (key_type == BTRFS_DIR_INDEX_KEY)
5901 btrfs_put_delayed_items(&ins_list, &del_list);
5902 btrfs_free_path(path);
5903 return ret;
5904 }
5905
5906 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5907 {
5908 struct btrfs_root *root = BTRFS_I(inode)->root;
5909 struct btrfs_trans_handle *trans;
5910 int ret = 0;
5911 bool nolock = false;
5912
5913 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5914 return 0;
5915
5916 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5917 nolock = true;
5918
5919 if (wbc->sync_mode == WB_SYNC_ALL) {
5920 if (nolock)
5921 trans = btrfs_join_transaction_nolock(root);
5922 else
5923 trans = btrfs_join_transaction(root);
5924 if (IS_ERR(trans))
5925 return PTR_ERR(trans);
5926 ret = btrfs_commit_transaction(trans, root);
5927 }
5928 return ret;
5929 }
5930
5931 /*
5932 * This is somewhat expensive, updating the tree every time the
5933 * inode changes. But, it is most likely to find the inode in cache.
5934 * FIXME, needs more benchmarking...there are no reasons other than performance
5935 * to keep or drop this code.
5936 */
5937 static int btrfs_dirty_inode(struct inode *inode)
5938 {
5939 struct btrfs_root *root = BTRFS_I(inode)->root;
5940 struct btrfs_trans_handle *trans;
5941 int ret;
5942
5943 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5944 return 0;
5945
5946 trans = btrfs_join_transaction(root);
5947 if (IS_ERR(trans))
5948 return PTR_ERR(trans);
5949
5950 ret = btrfs_update_inode(trans, root, inode);
5951 if (ret && ret == -ENOSPC) {
5952 /* whoops, lets try again with the full transaction */
5953 btrfs_end_transaction(trans, root);
5954 trans = btrfs_start_transaction(root, 1);
5955 if (IS_ERR(trans))
5956 return PTR_ERR(trans);
5957
5958 ret = btrfs_update_inode(trans, root, inode);
5959 }
5960 btrfs_end_transaction(trans, root);
5961 if (BTRFS_I(inode)->delayed_node)
5962 btrfs_balance_delayed_items(root);
5963
5964 return ret;
5965 }
5966
5967 /*
5968 * This is a copy of file_update_time. We need this so we can return error on
5969 * ENOSPC for updating the inode in the case of file write and mmap writes.
5970 */
5971 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5972 int flags)
5973 {
5974 struct btrfs_root *root = BTRFS_I(inode)->root;
5975
5976 if (btrfs_root_readonly(root))
5977 return -EROFS;
5978
5979 if (flags & S_VERSION)
5980 inode_inc_iversion(inode);
5981 if (flags & S_CTIME)
5982 inode->i_ctime = *now;
5983 if (flags & S_MTIME)
5984 inode->i_mtime = *now;
5985 if (flags & S_ATIME)
5986 inode->i_atime = *now;
5987 return btrfs_dirty_inode(inode);
5988 }
5989
5990 /*
5991 * find the highest existing sequence number in a directory
5992 * and then set the in-memory index_cnt variable to reflect
5993 * free sequence numbers
5994 */
5995 static int btrfs_set_inode_index_count(struct inode *inode)
5996 {
5997 struct btrfs_root *root = BTRFS_I(inode)->root;
5998 struct btrfs_key key, found_key;
5999 struct btrfs_path *path;
6000 struct extent_buffer *leaf;
6001 int ret;
6002
6003 key.objectid = btrfs_ino(inode);
6004 key.type = BTRFS_DIR_INDEX_KEY;
6005 key.offset = (u64)-1;
6006
6007 path = btrfs_alloc_path();
6008 if (!path)
6009 return -ENOMEM;
6010
6011 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6012 if (ret < 0)
6013 goto out;
6014 /* FIXME: we should be able to handle this */
6015 if (ret == 0)
6016 goto out;
6017 ret = 0;
6018
6019 /*
6020 * MAGIC NUMBER EXPLANATION:
6021 * since we search a directory based on f_pos we have to start at 2
6022 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6023 * else has to start at 2
6024 */
6025 if (path->slots[0] == 0) {
6026 BTRFS_I(inode)->index_cnt = 2;
6027 goto out;
6028 }
6029
6030 path->slots[0]--;
6031
6032 leaf = path->nodes[0];
6033 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6034
6035 if (found_key.objectid != btrfs_ino(inode) ||
6036 found_key.type != BTRFS_DIR_INDEX_KEY) {
6037 BTRFS_I(inode)->index_cnt = 2;
6038 goto out;
6039 }
6040
6041 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6042 out:
6043 btrfs_free_path(path);
6044 return ret;
6045 }
6046
6047 /*
6048 * helper to find a free sequence number in a given directory. This current
6049 * code is very simple, later versions will do smarter things in the btree
6050 */
6051 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6052 {
6053 int ret = 0;
6054
6055 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6056 ret = btrfs_inode_delayed_dir_index_count(dir);
6057 if (ret) {
6058 ret = btrfs_set_inode_index_count(dir);
6059 if (ret)
6060 return ret;
6061 }
6062 }
6063
6064 *index = BTRFS_I(dir)->index_cnt;
6065 BTRFS_I(dir)->index_cnt++;
6066
6067 return ret;
6068 }
6069
6070 static int btrfs_insert_inode_locked(struct inode *inode)
6071 {
6072 struct btrfs_iget_args args;
6073 args.location = &BTRFS_I(inode)->location;
6074 args.root = BTRFS_I(inode)->root;
6075
6076 return insert_inode_locked4(inode,
6077 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6078 btrfs_find_actor, &args);
6079 }
6080
6081 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6082 struct btrfs_root *root,
6083 struct inode *dir,
6084 const char *name, int name_len,
6085 u64 ref_objectid, u64 objectid,
6086 umode_t mode, u64 *index)
6087 {
6088 struct inode *inode;
6089 struct btrfs_inode_item *inode_item;
6090 struct btrfs_key *location;
6091 struct btrfs_path *path;
6092 struct btrfs_inode_ref *ref;
6093 struct btrfs_key key[2];
6094 u32 sizes[2];
6095 int nitems = name ? 2 : 1;
6096 unsigned long ptr;
6097 int ret;
6098
6099 path = btrfs_alloc_path();
6100 if (!path)
6101 return ERR_PTR(-ENOMEM);
6102
6103 inode = new_inode(root->fs_info->sb);
6104 if (!inode) {
6105 btrfs_free_path(path);
6106 return ERR_PTR(-ENOMEM);
6107 }
6108
6109 /*
6110 * O_TMPFILE, set link count to 0, so that after this point,
6111 * we fill in an inode item with the correct link count.
6112 */
6113 if (!name)
6114 set_nlink(inode, 0);
6115
6116 /*
6117 * we have to initialize this early, so we can reclaim the inode
6118 * number if we fail afterwards in this function.
6119 */
6120 inode->i_ino = objectid;
6121
6122 if (dir && name) {
6123 trace_btrfs_inode_request(dir);
6124
6125 ret = btrfs_set_inode_index(dir, index);
6126 if (ret) {
6127 btrfs_free_path(path);
6128 iput(inode);
6129 return ERR_PTR(ret);
6130 }
6131 } else if (dir) {
6132 *index = 0;
6133 }
6134 /*
6135 * index_cnt is ignored for everything but a dir,
6136 * btrfs_get_inode_index_count has an explanation for the magic
6137 * number
6138 */
6139 BTRFS_I(inode)->index_cnt = 2;
6140 BTRFS_I(inode)->dir_index = *index;
6141 BTRFS_I(inode)->root = root;
6142 BTRFS_I(inode)->generation = trans->transid;
6143 inode->i_generation = BTRFS_I(inode)->generation;
6144
6145 /*
6146 * We could have gotten an inode number from somebody who was fsynced
6147 * and then removed in this same transaction, so let's just set full
6148 * sync since it will be a full sync anyway and this will blow away the
6149 * old info in the log.
6150 */
6151 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6152
6153 key[0].objectid = objectid;
6154 key[0].type = BTRFS_INODE_ITEM_KEY;
6155 key[0].offset = 0;
6156
6157 sizes[0] = sizeof(struct btrfs_inode_item);
6158
6159 if (name) {
6160 /*
6161 * Start new inodes with an inode_ref. This is slightly more
6162 * efficient for small numbers of hard links since they will
6163 * be packed into one item. Extended refs will kick in if we
6164 * add more hard links than can fit in the ref item.
6165 */
6166 key[1].objectid = objectid;
6167 key[1].type = BTRFS_INODE_REF_KEY;
6168 key[1].offset = ref_objectid;
6169
6170 sizes[1] = name_len + sizeof(*ref);
6171 }
6172
6173 location = &BTRFS_I(inode)->location;
6174 location->objectid = objectid;
6175 location->offset = 0;
6176 location->type = BTRFS_INODE_ITEM_KEY;
6177
6178 ret = btrfs_insert_inode_locked(inode);
6179 if (ret < 0)
6180 goto fail;
6181
6182 path->leave_spinning = 1;
6183 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6184 if (ret != 0)
6185 goto fail_unlock;
6186
6187 inode_init_owner(inode, dir, mode);
6188 inode_set_bytes(inode, 0);
6189
6190 inode->i_mtime = current_fs_time(inode->i_sb);
6191 inode->i_atime = inode->i_mtime;
6192 inode->i_ctime = inode->i_mtime;
6193 BTRFS_I(inode)->i_otime = inode->i_mtime;
6194
6195 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6196 struct btrfs_inode_item);
6197 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6198 sizeof(*inode_item));
6199 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6200
6201 if (name) {
6202 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6203 struct btrfs_inode_ref);
6204 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6205 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6206 ptr = (unsigned long)(ref + 1);
6207 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6208 }
6209
6210 btrfs_mark_buffer_dirty(path->nodes[0]);
6211 btrfs_free_path(path);
6212
6213 btrfs_inherit_iflags(inode, dir);
6214
6215 if (S_ISREG(mode)) {
6216 if (btrfs_test_opt(root, NODATASUM))
6217 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6218 if (btrfs_test_opt(root, NODATACOW))
6219 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6220 BTRFS_INODE_NODATASUM;
6221 }
6222
6223 inode_tree_add(inode);
6224
6225 trace_btrfs_inode_new(inode);
6226 btrfs_set_inode_last_trans(trans, inode);
6227
6228 btrfs_update_root_times(trans, root);
6229
6230 ret = btrfs_inode_inherit_props(trans, inode, dir);
6231 if (ret)
6232 btrfs_err(root->fs_info,
6233 "error inheriting props for ino %llu (root %llu): %d",
6234 btrfs_ino(inode), root->root_key.objectid, ret);
6235
6236 return inode;
6237
6238 fail_unlock:
6239 unlock_new_inode(inode);
6240 fail:
6241 if (dir && name)
6242 BTRFS_I(dir)->index_cnt--;
6243 btrfs_free_path(path);
6244 iput(inode);
6245 return ERR_PTR(ret);
6246 }
6247
6248 static inline u8 btrfs_inode_type(struct inode *inode)
6249 {
6250 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6251 }
6252
6253 /*
6254 * utility function to add 'inode' into 'parent_inode' with
6255 * a give name and a given sequence number.
6256 * if 'add_backref' is true, also insert a backref from the
6257 * inode to the parent directory.
6258 */
6259 int btrfs_add_link(struct btrfs_trans_handle *trans,
6260 struct inode *parent_inode, struct inode *inode,
6261 const char *name, int name_len, int add_backref, u64 index)
6262 {
6263 int ret = 0;
6264 struct btrfs_key key;
6265 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6266 u64 ino = btrfs_ino(inode);
6267 u64 parent_ino = btrfs_ino(parent_inode);
6268
6269 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6270 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6271 } else {
6272 key.objectid = ino;
6273 key.type = BTRFS_INODE_ITEM_KEY;
6274 key.offset = 0;
6275 }
6276
6277 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6278 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6279 key.objectid, root->root_key.objectid,
6280 parent_ino, index, name, name_len);
6281 } else if (add_backref) {
6282 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6283 parent_ino, index);
6284 }
6285
6286 /* Nothing to clean up yet */
6287 if (ret)
6288 return ret;
6289
6290 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6291 parent_inode, &key,
6292 btrfs_inode_type(inode), index);
6293 if (ret == -EEXIST || ret == -EOVERFLOW)
6294 goto fail_dir_item;
6295 else if (ret) {
6296 btrfs_abort_transaction(trans, root, ret);
6297 return ret;
6298 }
6299
6300 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6301 name_len * 2);
6302 inode_inc_iversion(parent_inode);
6303 parent_inode->i_mtime = parent_inode->i_ctime =
6304 current_fs_time(parent_inode->i_sb);
6305 ret = btrfs_update_inode(trans, root, parent_inode);
6306 if (ret)
6307 btrfs_abort_transaction(trans, root, ret);
6308 return ret;
6309
6310 fail_dir_item:
6311 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6312 u64 local_index;
6313 int err;
6314 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6315 key.objectid, root->root_key.objectid,
6316 parent_ino, &local_index, name, name_len);
6317
6318 } else if (add_backref) {
6319 u64 local_index;
6320 int err;
6321
6322 err = btrfs_del_inode_ref(trans, root, name, name_len,
6323 ino, parent_ino, &local_index);
6324 }
6325 return ret;
6326 }
6327
6328 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6329 struct inode *dir, struct dentry *dentry,
6330 struct inode *inode, int backref, u64 index)
6331 {
6332 int err = btrfs_add_link(trans, dir, inode,
6333 dentry->d_name.name, dentry->d_name.len,
6334 backref, index);
6335 if (err > 0)
6336 err = -EEXIST;
6337 return err;
6338 }
6339
6340 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6341 umode_t mode, dev_t rdev)
6342 {
6343 struct btrfs_trans_handle *trans;
6344 struct btrfs_root *root = BTRFS_I(dir)->root;
6345 struct inode *inode = NULL;
6346 int err;
6347 int drop_inode = 0;
6348 u64 objectid;
6349 u64 index = 0;
6350
6351 /*
6352 * 2 for inode item and ref
6353 * 2 for dir items
6354 * 1 for xattr if selinux is on
6355 */
6356 trans = btrfs_start_transaction(root, 5);
6357 if (IS_ERR(trans))
6358 return PTR_ERR(trans);
6359
6360 err = btrfs_find_free_ino(root, &objectid);
6361 if (err)
6362 goto out_unlock;
6363
6364 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6365 dentry->d_name.len, btrfs_ino(dir), objectid,
6366 mode, &index);
6367 if (IS_ERR(inode)) {
6368 err = PTR_ERR(inode);
6369 goto out_unlock;
6370 }
6371
6372 /*
6373 * If the active LSM wants to access the inode during
6374 * d_instantiate it needs these. Smack checks to see
6375 * if the filesystem supports xattrs by looking at the
6376 * ops vector.
6377 */
6378 inode->i_op = &btrfs_special_inode_operations;
6379 init_special_inode(inode, inode->i_mode, rdev);
6380
6381 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6382 if (err)
6383 goto out_unlock_inode;
6384
6385 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6386 if (err) {
6387 goto out_unlock_inode;
6388 } else {
6389 btrfs_update_inode(trans, root, inode);
6390 unlock_new_inode(inode);
6391 d_instantiate(dentry, inode);
6392 }
6393
6394 out_unlock:
6395 btrfs_end_transaction(trans, root);
6396 btrfs_balance_delayed_items(root);
6397 btrfs_btree_balance_dirty(root);
6398 if (drop_inode) {
6399 inode_dec_link_count(inode);
6400 iput(inode);
6401 }
6402 return err;
6403
6404 out_unlock_inode:
6405 drop_inode = 1;
6406 unlock_new_inode(inode);
6407 goto out_unlock;
6408
6409 }
6410
6411 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6412 umode_t mode, bool excl)
6413 {
6414 struct btrfs_trans_handle *trans;
6415 struct btrfs_root *root = BTRFS_I(dir)->root;
6416 struct inode *inode = NULL;
6417 int drop_inode_on_err = 0;
6418 int err;
6419 u64 objectid;
6420 u64 index = 0;
6421
6422 /*
6423 * 2 for inode item and ref
6424 * 2 for dir items
6425 * 1 for xattr if selinux is on
6426 */
6427 trans = btrfs_start_transaction(root, 5);
6428 if (IS_ERR(trans))
6429 return PTR_ERR(trans);
6430
6431 err = btrfs_find_free_ino(root, &objectid);
6432 if (err)
6433 goto out_unlock;
6434
6435 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6436 dentry->d_name.len, btrfs_ino(dir), objectid,
6437 mode, &index);
6438 if (IS_ERR(inode)) {
6439 err = PTR_ERR(inode);
6440 goto out_unlock;
6441 }
6442 drop_inode_on_err = 1;
6443 /*
6444 * If the active LSM wants to access the inode during
6445 * d_instantiate it needs these. Smack checks to see
6446 * if the filesystem supports xattrs by looking at the
6447 * ops vector.
6448 */
6449 inode->i_fop = &btrfs_file_operations;
6450 inode->i_op = &btrfs_file_inode_operations;
6451 inode->i_mapping->a_ops = &btrfs_aops;
6452
6453 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6454 if (err)
6455 goto out_unlock_inode;
6456
6457 err = btrfs_update_inode(trans, root, inode);
6458 if (err)
6459 goto out_unlock_inode;
6460
6461 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6462 if (err)
6463 goto out_unlock_inode;
6464
6465 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6466 unlock_new_inode(inode);
6467 d_instantiate(dentry, inode);
6468
6469 out_unlock:
6470 btrfs_end_transaction(trans, root);
6471 if (err && drop_inode_on_err) {
6472 inode_dec_link_count(inode);
6473 iput(inode);
6474 }
6475 btrfs_balance_delayed_items(root);
6476 btrfs_btree_balance_dirty(root);
6477 return err;
6478
6479 out_unlock_inode:
6480 unlock_new_inode(inode);
6481 goto out_unlock;
6482
6483 }
6484
6485 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6486 struct dentry *dentry)
6487 {
6488 struct btrfs_trans_handle *trans = NULL;
6489 struct btrfs_root *root = BTRFS_I(dir)->root;
6490 struct inode *inode = d_inode(old_dentry);
6491 u64 index;
6492 int err;
6493 int drop_inode = 0;
6494
6495 /* do not allow sys_link's with other subvols of the same device */
6496 if (root->objectid != BTRFS_I(inode)->root->objectid)
6497 return -EXDEV;
6498
6499 if (inode->i_nlink >= BTRFS_LINK_MAX)
6500 return -EMLINK;
6501
6502 err = btrfs_set_inode_index(dir, &index);
6503 if (err)
6504 goto fail;
6505
6506 /*
6507 * 2 items for inode and inode ref
6508 * 2 items for dir items
6509 * 1 item for parent inode
6510 */
6511 trans = btrfs_start_transaction(root, 5);
6512 if (IS_ERR(trans)) {
6513 err = PTR_ERR(trans);
6514 trans = NULL;
6515 goto fail;
6516 }
6517
6518 /* There are several dir indexes for this inode, clear the cache. */
6519 BTRFS_I(inode)->dir_index = 0ULL;
6520 inc_nlink(inode);
6521 inode_inc_iversion(inode);
6522 inode->i_ctime = current_fs_time(inode->i_sb);
6523 ihold(inode);
6524 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6525
6526 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6527
6528 if (err) {
6529 drop_inode = 1;
6530 } else {
6531 struct dentry *parent = dentry->d_parent;
6532 err = btrfs_update_inode(trans, root, inode);
6533 if (err)
6534 goto fail;
6535 if (inode->i_nlink == 1) {
6536 /*
6537 * If new hard link count is 1, it's a file created
6538 * with open(2) O_TMPFILE flag.
6539 */
6540 err = btrfs_orphan_del(trans, inode);
6541 if (err)
6542 goto fail;
6543 }
6544 d_instantiate(dentry, inode);
6545 btrfs_log_new_name(trans, inode, NULL, parent);
6546 }
6547
6548 btrfs_balance_delayed_items(root);
6549 fail:
6550 if (trans)
6551 btrfs_end_transaction(trans, root);
6552 if (drop_inode) {
6553 inode_dec_link_count(inode);
6554 iput(inode);
6555 }
6556 btrfs_btree_balance_dirty(root);
6557 return err;
6558 }
6559
6560 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6561 {
6562 struct inode *inode = NULL;
6563 struct btrfs_trans_handle *trans;
6564 struct btrfs_root *root = BTRFS_I(dir)->root;
6565 int err = 0;
6566 int drop_on_err = 0;
6567 u64 objectid = 0;
6568 u64 index = 0;
6569
6570 /*
6571 * 2 items for inode and ref
6572 * 2 items for dir items
6573 * 1 for xattr if selinux is on
6574 */
6575 trans = btrfs_start_transaction(root, 5);
6576 if (IS_ERR(trans))
6577 return PTR_ERR(trans);
6578
6579 err = btrfs_find_free_ino(root, &objectid);
6580 if (err)
6581 goto out_fail;
6582
6583 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6584 dentry->d_name.len, btrfs_ino(dir), objectid,
6585 S_IFDIR | mode, &index);
6586 if (IS_ERR(inode)) {
6587 err = PTR_ERR(inode);
6588 goto out_fail;
6589 }
6590
6591 drop_on_err = 1;
6592 /* these must be set before we unlock the inode */
6593 inode->i_op = &btrfs_dir_inode_operations;
6594 inode->i_fop = &btrfs_dir_file_operations;
6595
6596 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6597 if (err)
6598 goto out_fail_inode;
6599
6600 btrfs_i_size_write(inode, 0);
6601 err = btrfs_update_inode(trans, root, inode);
6602 if (err)
6603 goto out_fail_inode;
6604
6605 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6606 dentry->d_name.len, 0, index);
6607 if (err)
6608 goto out_fail_inode;
6609
6610 d_instantiate(dentry, inode);
6611 /*
6612 * mkdir is special. We're unlocking after we call d_instantiate
6613 * to avoid a race with nfsd calling d_instantiate.
6614 */
6615 unlock_new_inode(inode);
6616 drop_on_err = 0;
6617
6618 out_fail:
6619 btrfs_end_transaction(trans, root);
6620 if (drop_on_err) {
6621 inode_dec_link_count(inode);
6622 iput(inode);
6623 }
6624 btrfs_balance_delayed_items(root);
6625 btrfs_btree_balance_dirty(root);
6626 return err;
6627
6628 out_fail_inode:
6629 unlock_new_inode(inode);
6630 goto out_fail;
6631 }
6632
6633 /* Find next extent map of a given extent map, caller needs to ensure locks */
6634 static struct extent_map *next_extent_map(struct extent_map *em)
6635 {
6636 struct rb_node *next;
6637
6638 next = rb_next(&em->rb_node);
6639 if (!next)
6640 return NULL;
6641 return container_of(next, struct extent_map, rb_node);
6642 }
6643
6644 static struct extent_map *prev_extent_map(struct extent_map *em)
6645 {
6646 struct rb_node *prev;
6647
6648 prev = rb_prev(&em->rb_node);
6649 if (!prev)
6650 return NULL;
6651 return container_of(prev, struct extent_map, rb_node);
6652 }
6653
6654 /* helper for btfs_get_extent. Given an existing extent in the tree,
6655 * the existing extent is the nearest extent to map_start,
6656 * and an extent that you want to insert, deal with overlap and insert
6657 * the best fitted new extent into the tree.
6658 */
6659 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6660 struct extent_map *existing,
6661 struct extent_map *em,
6662 u64 map_start)
6663 {
6664 struct extent_map *prev;
6665 struct extent_map *next;
6666 u64 start;
6667 u64 end;
6668 u64 start_diff;
6669
6670 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6671
6672 if (existing->start > map_start) {
6673 next = existing;
6674 prev = prev_extent_map(next);
6675 } else {
6676 prev = existing;
6677 next = next_extent_map(prev);
6678 }
6679
6680 start = prev ? extent_map_end(prev) : em->start;
6681 start = max_t(u64, start, em->start);
6682 end = next ? next->start : extent_map_end(em);
6683 end = min_t(u64, end, extent_map_end(em));
6684 start_diff = start - em->start;
6685 em->start = start;
6686 em->len = end - start;
6687 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6688 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6689 em->block_start += start_diff;
6690 em->block_len -= start_diff;
6691 }
6692 return add_extent_mapping(em_tree, em, 0);
6693 }
6694
6695 static noinline int uncompress_inline(struct btrfs_path *path,
6696 struct page *page,
6697 size_t pg_offset, u64 extent_offset,
6698 struct btrfs_file_extent_item *item)
6699 {
6700 int ret;
6701 struct extent_buffer *leaf = path->nodes[0];
6702 char *tmp;
6703 size_t max_size;
6704 unsigned long inline_size;
6705 unsigned long ptr;
6706 int compress_type;
6707
6708 WARN_ON(pg_offset != 0);
6709 compress_type = btrfs_file_extent_compression(leaf, item);
6710 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6711 inline_size = btrfs_file_extent_inline_item_len(leaf,
6712 btrfs_item_nr(path->slots[0]));
6713 tmp = kmalloc(inline_size, GFP_NOFS);
6714 if (!tmp)
6715 return -ENOMEM;
6716 ptr = btrfs_file_extent_inline_start(item);
6717
6718 read_extent_buffer(leaf, tmp, ptr, inline_size);
6719
6720 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6721 ret = btrfs_decompress(compress_type, tmp, page,
6722 extent_offset, inline_size, max_size);
6723 kfree(tmp);
6724 return ret;
6725 }
6726
6727 /*
6728 * a bit scary, this does extent mapping from logical file offset to the disk.
6729 * the ugly parts come from merging extents from the disk with the in-ram
6730 * representation. This gets more complex because of the data=ordered code,
6731 * where the in-ram extents might be locked pending data=ordered completion.
6732 *
6733 * This also copies inline extents directly into the page.
6734 */
6735
6736 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6737 size_t pg_offset, u64 start, u64 len,
6738 int create)
6739 {
6740 int ret;
6741 int err = 0;
6742 u64 extent_start = 0;
6743 u64 extent_end = 0;
6744 u64 objectid = btrfs_ino(inode);
6745 u32 found_type;
6746 struct btrfs_path *path = NULL;
6747 struct btrfs_root *root = BTRFS_I(inode)->root;
6748 struct btrfs_file_extent_item *item;
6749 struct extent_buffer *leaf;
6750 struct btrfs_key found_key;
6751 struct extent_map *em = NULL;
6752 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6753 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6754 struct btrfs_trans_handle *trans = NULL;
6755 const bool new_inline = !page || create;
6756
6757 again:
6758 read_lock(&em_tree->lock);
6759 em = lookup_extent_mapping(em_tree, start, len);
6760 if (em)
6761 em->bdev = root->fs_info->fs_devices->latest_bdev;
6762 read_unlock(&em_tree->lock);
6763
6764 if (em) {
6765 if (em->start > start || em->start + em->len <= start)
6766 free_extent_map(em);
6767 else if (em->block_start == EXTENT_MAP_INLINE && page)
6768 free_extent_map(em);
6769 else
6770 goto out;
6771 }
6772 em = alloc_extent_map();
6773 if (!em) {
6774 err = -ENOMEM;
6775 goto out;
6776 }
6777 em->bdev = root->fs_info->fs_devices->latest_bdev;
6778 em->start = EXTENT_MAP_HOLE;
6779 em->orig_start = EXTENT_MAP_HOLE;
6780 em->len = (u64)-1;
6781 em->block_len = (u64)-1;
6782
6783 if (!path) {
6784 path = btrfs_alloc_path();
6785 if (!path) {
6786 err = -ENOMEM;
6787 goto out;
6788 }
6789 /*
6790 * Chances are we'll be called again, so go ahead and do
6791 * readahead
6792 */
6793 path->reada = READA_FORWARD;
6794 }
6795
6796 ret = btrfs_lookup_file_extent(trans, root, path,
6797 objectid, start, trans != NULL);
6798 if (ret < 0) {
6799 err = ret;
6800 goto out;
6801 }
6802
6803 if (ret != 0) {
6804 if (path->slots[0] == 0)
6805 goto not_found;
6806 path->slots[0]--;
6807 }
6808
6809 leaf = path->nodes[0];
6810 item = btrfs_item_ptr(leaf, path->slots[0],
6811 struct btrfs_file_extent_item);
6812 /* are we inside the extent that was found? */
6813 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6814 found_type = found_key.type;
6815 if (found_key.objectid != objectid ||
6816 found_type != BTRFS_EXTENT_DATA_KEY) {
6817 /*
6818 * If we backup past the first extent we want to move forward
6819 * and see if there is an extent in front of us, otherwise we'll
6820 * say there is a hole for our whole search range which can
6821 * cause problems.
6822 */
6823 extent_end = start;
6824 goto next;
6825 }
6826
6827 found_type = btrfs_file_extent_type(leaf, item);
6828 extent_start = found_key.offset;
6829 if (found_type == BTRFS_FILE_EXTENT_REG ||
6830 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6831 extent_end = extent_start +
6832 btrfs_file_extent_num_bytes(leaf, item);
6833 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6834 size_t size;
6835 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6836 extent_end = ALIGN(extent_start + size, root->sectorsize);
6837 }
6838 next:
6839 if (start >= extent_end) {
6840 path->slots[0]++;
6841 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6842 ret = btrfs_next_leaf(root, path);
6843 if (ret < 0) {
6844 err = ret;
6845 goto out;
6846 }
6847 if (ret > 0)
6848 goto not_found;
6849 leaf = path->nodes[0];
6850 }
6851 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6852 if (found_key.objectid != objectid ||
6853 found_key.type != BTRFS_EXTENT_DATA_KEY)
6854 goto not_found;
6855 if (start + len <= found_key.offset)
6856 goto not_found;
6857 if (start > found_key.offset)
6858 goto next;
6859 em->start = start;
6860 em->orig_start = start;
6861 em->len = found_key.offset - start;
6862 goto not_found_em;
6863 }
6864
6865 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6866
6867 if (found_type == BTRFS_FILE_EXTENT_REG ||
6868 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6869 goto insert;
6870 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6871 unsigned long ptr;
6872 char *map;
6873 size_t size;
6874 size_t extent_offset;
6875 size_t copy_size;
6876
6877 if (new_inline)
6878 goto out;
6879
6880 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6881 extent_offset = page_offset(page) + pg_offset - extent_start;
6882 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6883 size - extent_offset);
6884 em->start = extent_start + extent_offset;
6885 em->len = ALIGN(copy_size, root->sectorsize);
6886 em->orig_block_len = em->len;
6887 em->orig_start = em->start;
6888 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6889 if (create == 0 && !PageUptodate(page)) {
6890 if (btrfs_file_extent_compression(leaf, item) !=
6891 BTRFS_COMPRESS_NONE) {
6892 ret = uncompress_inline(path, page, pg_offset,
6893 extent_offset, item);
6894 if (ret) {
6895 err = ret;
6896 goto out;
6897 }
6898 } else {
6899 map = kmap(page);
6900 read_extent_buffer(leaf, map + pg_offset, ptr,
6901 copy_size);
6902 if (pg_offset + copy_size < PAGE_SIZE) {
6903 memset(map + pg_offset + copy_size, 0,
6904 PAGE_SIZE - pg_offset -
6905 copy_size);
6906 }
6907 kunmap(page);
6908 }
6909 flush_dcache_page(page);
6910 } else if (create && PageUptodate(page)) {
6911 BUG();
6912 if (!trans) {
6913 kunmap(page);
6914 free_extent_map(em);
6915 em = NULL;
6916
6917 btrfs_release_path(path);
6918 trans = btrfs_join_transaction(root);
6919
6920 if (IS_ERR(trans))
6921 return ERR_CAST(trans);
6922 goto again;
6923 }
6924 map = kmap(page);
6925 write_extent_buffer(leaf, map + pg_offset, ptr,
6926 copy_size);
6927 kunmap(page);
6928 btrfs_mark_buffer_dirty(leaf);
6929 }
6930 set_extent_uptodate(io_tree, em->start,
6931 extent_map_end(em) - 1, NULL, GFP_NOFS);
6932 goto insert;
6933 }
6934 not_found:
6935 em->start = start;
6936 em->orig_start = start;
6937 em->len = len;
6938 not_found_em:
6939 em->block_start = EXTENT_MAP_HOLE;
6940 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6941 insert:
6942 btrfs_release_path(path);
6943 if (em->start > start || extent_map_end(em) <= start) {
6944 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6945 em->start, em->len, start, len);
6946 err = -EIO;
6947 goto out;
6948 }
6949
6950 err = 0;
6951 write_lock(&em_tree->lock);
6952 ret = add_extent_mapping(em_tree, em, 0);
6953 /* it is possible that someone inserted the extent into the tree
6954 * while we had the lock dropped. It is also possible that
6955 * an overlapping map exists in the tree
6956 */
6957 if (ret == -EEXIST) {
6958 struct extent_map *existing;
6959
6960 ret = 0;
6961
6962 existing = search_extent_mapping(em_tree, start, len);
6963 /*
6964 * existing will always be non-NULL, since there must be
6965 * extent causing the -EEXIST.
6966 */
6967 if (start >= extent_map_end(existing) ||
6968 start <= existing->start) {
6969 /*
6970 * The existing extent map is the one nearest to
6971 * the [start, start + len) range which overlaps
6972 */
6973 err = merge_extent_mapping(em_tree, existing,
6974 em, start);
6975 free_extent_map(existing);
6976 if (err) {
6977 free_extent_map(em);
6978 em = NULL;
6979 }
6980 } else {
6981 free_extent_map(em);
6982 em = existing;
6983 err = 0;
6984 }
6985 }
6986 write_unlock(&em_tree->lock);
6987 out:
6988
6989 trace_btrfs_get_extent(root, em);
6990
6991 btrfs_free_path(path);
6992 if (trans) {
6993 ret = btrfs_end_transaction(trans, root);
6994 if (!err)
6995 err = ret;
6996 }
6997 if (err) {
6998 free_extent_map(em);
6999 return ERR_PTR(err);
7000 }
7001 BUG_ON(!em); /* Error is always set */
7002 return em;
7003 }
7004
7005 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7006 size_t pg_offset, u64 start, u64 len,
7007 int create)
7008 {
7009 struct extent_map *em;
7010 struct extent_map *hole_em = NULL;
7011 u64 range_start = start;
7012 u64 end;
7013 u64 found;
7014 u64 found_end;
7015 int err = 0;
7016
7017 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7018 if (IS_ERR(em))
7019 return em;
7020 if (em) {
7021 /*
7022 * if our em maps to
7023 * - a hole or
7024 * - a pre-alloc extent,
7025 * there might actually be delalloc bytes behind it.
7026 */
7027 if (em->block_start != EXTENT_MAP_HOLE &&
7028 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7029 return em;
7030 else
7031 hole_em = em;
7032 }
7033
7034 /* check to see if we've wrapped (len == -1 or similar) */
7035 end = start + len;
7036 if (end < start)
7037 end = (u64)-1;
7038 else
7039 end -= 1;
7040
7041 em = NULL;
7042
7043 /* ok, we didn't find anything, lets look for delalloc */
7044 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7045 end, len, EXTENT_DELALLOC, 1);
7046 found_end = range_start + found;
7047 if (found_end < range_start)
7048 found_end = (u64)-1;
7049
7050 /*
7051 * we didn't find anything useful, return
7052 * the original results from get_extent()
7053 */
7054 if (range_start > end || found_end <= start) {
7055 em = hole_em;
7056 hole_em = NULL;
7057 goto out;
7058 }
7059
7060 /* adjust the range_start to make sure it doesn't
7061 * go backwards from the start they passed in
7062 */
7063 range_start = max(start, range_start);
7064 found = found_end - range_start;
7065
7066 if (found > 0) {
7067 u64 hole_start = start;
7068 u64 hole_len = len;
7069
7070 em = alloc_extent_map();
7071 if (!em) {
7072 err = -ENOMEM;
7073 goto out;
7074 }
7075 /*
7076 * when btrfs_get_extent can't find anything it
7077 * returns one huge hole
7078 *
7079 * make sure what it found really fits our range, and
7080 * adjust to make sure it is based on the start from
7081 * the caller
7082 */
7083 if (hole_em) {
7084 u64 calc_end = extent_map_end(hole_em);
7085
7086 if (calc_end <= start || (hole_em->start > end)) {
7087 free_extent_map(hole_em);
7088 hole_em = NULL;
7089 } else {
7090 hole_start = max(hole_em->start, start);
7091 hole_len = calc_end - hole_start;
7092 }
7093 }
7094 em->bdev = NULL;
7095 if (hole_em && range_start > hole_start) {
7096 /* our hole starts before our delalloc, so we
7097 * have to return just the parts of the hole
7098 * that go until the delalloc starts
7099 */
7100 em->len = min(hole_len,
7101 range_start - hole_start);
7102 em->start = hole_start;
7103 em->orig_start = hole_start;
7104 /*
7105 * don't adjust block start at all,
7106 * it is fixed at EXTENT_MAP_HOLE
7107 */
7108 em->block_start = hole_em->block_start;
7109 em->block_len = hole_len;
7110 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7111 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7112 } else {
7113 em->start = range_start;
7114 em->len = found;
7115 em->orig_start = range_start;
7116 em->block_start = EXTENT_MAP_DELALLOC;
7117 em->block_len = found;
7118 }
7119 } else if (hole_em) {
7120 return hole_em;
7121 }
7122 out:
7123
7124 free_extent_map(hole_em);
7125 if (err) {
7126 free_extent_map(em);
7127 return ERR_PTR(err);
7128 }
7129 return em;
7130 }
7131
7132 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7133 u64 start, u64 len)
7134 {
7135 struct btrfs_root *root = BTRFS_I(inode)->root;
7136 struct extent_map *em;
7137 struct btrfs_key ins;
7138 u64 alloc_hint;
7139 int ret;
7140
7141 alloc_hint = get_extent_allocation_hint(inode, start, len);
7142 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7143 alloc_hint, &ins, 1, 1);
7144 if (ret)
7145 return ERR_PTR(ret);
7146
7147 /*
7148 * Create the ordered extent before the extent map. This is to avoid
7149 * races with the fast fsync path that would lead to it logging file
7150 * extent items that point to disk extents that were not yet written to.
7151 * The fast fsync path collects ordered extents into a local list and
7152 * then collects all the new extent maps, so we must create the ordered
7153 * extent first and make sure the fast fsync path collects any new
7154 * ordered extents after collecting new extent maps as well.
7155 * The fsync path simply can not rely on inode_dio_wait() because it
7156 * causes deadlock with AIO.
7157 */
7158 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7159 ins.offset, ins.offset, 0);
7160 if (ret) {
7161 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7162 return ERR_PTR(ret);
7163 }
7164
7165 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7166 ins.offset, ins.offset, ins.offset, 0);
7167 if (IS_ERR(em)) {
7168 struct btrfs_ordered_extent *oe;
7169
7170 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7171 oe = btrfs_lookup_ordered_extent(inode, start);
7172 ASSERT(oe);
7173 if (WARN_ON(!oe))
7174 return em;
7175 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7176 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7177 btrfs_remove_ordered_extent(inode, oe);
7178 /* Once for our lookup and once for the ordered extents tree. */
7179 btrfs_put_ordered_extent(oe);
7180 btrfs_put_ordered_extent(oe);
7181 }
7182 return em;
7183 }
7184
7185 /*
7186 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7187 * block must be cow'd
7188 */
7189 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7190 u64 *orig_start, u64 *orig_block_len,
7191 u64 *ram_bytes)
7192 {
7193 struct btrfs_trans_handle *trans;
7194 struct btrfs_path *path;
7195 int ret;
7196 struct extent_buffer *leaf;
7197 struct btrfs_root *root = BTRFS_I(inode)->root;
7198 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7199 struct btrfs_file_extent_item *fi;
7200 struct btrfs_key key;
7201 u64 disk_bytenr;
7202 u64 backref_offset;
7203 u64 extent_end;
7204 u64 num_bytes;
7205 int slot;
7206 int found_type;
7207 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7208
7209 path = btrfs_alloc_path();
7210 if (!path)
7211 return -ENOMEM;
7212
7213 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7214 offset, 0);
7215 if (ret < 0)
7216 goto out;
7217
7218 slot = path->slots[0];
7219 if (ret == 1) {
7220 if (slot == 0) {
7221 /* can't find the item, must cow */
7222 ret = 0;
7223 goto out;
7224 }
7225 slot--;
7226 }
7227 ret = 0;
7228 leaf = path->nodes[0];
7229 btrfs_item_key_to_cpu(leaf, &key, slot);
7230 if (key.objectid != btrfs_ino(inode) ||
7231 key.type != BTRFS_EXTENT_DATA_KEY) {
7232 /* not our file or wrong item type, must cow */
7233 goto out;
7234 }
7235
7236 if (key.offset > offset) {
7237 /* Wrong offset, must cow */
7238 goto out;
7239 }
7240
7241 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7242 found_type = btrfs_file_extent_type(leaf, fi);
7243 if (found_type != BTRFS_FILE_EXTENT_REG &&
7244 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7245 /* not a regular extent, must cow */
7246 goto out;
7247 }
7248
7249 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7250 goto out;
7251
7252 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7253 if (extent_end <= offset)
7254 goto out;
7255
7256 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7257 if (disk_bytenr == 0)
7258 goto out;
7259
7260 if (btrfs_file_extent_compression(leaf, fi) ||
7261 btrfs_file_extent_encryption(leaf, fi) ||
7262 btrfs_file_extent_other_encoding(leaf, fi))
7263 goto out;
7264
7265 backref_offset = btrfs_file_extent_offset(leaf, fi);
7266
7267 if (orig_start) {
7268 *orig_start = key.offset - backref_offset;
7269 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7270 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7271 }
7272
7273 if (btrfs_extent_readonly(root, disk_bytenr))
7274 goto out;
7275
7276 num_bytes = min(offset + *len, extent_end) - offset;
7277 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7278 u64 range_end;
7279
7280 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7281 ret = test_range_bit(io_tree, offset, range_end,
7282 EXTENT_DELALLOC, 0, NULL);
7283 if (ret) {
7284 ret = -EAGAIN;
7285 goto out;
7286 }
7287 }
7288
7289 btrfs_release_path(path);
7290
7291 /*
7292 * look for other files referencing this extent, if we
7293 * find any we must cow
7294 */
7295 trans = btrfs_join_transaction(root);
7296 if (IS_ERR(trans)) {
7297 ret = 0;
7298 goto out;
7299 }
7300
7301 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7302 key.offset - backref_offset, disk_bytenr);
7303 btrfs_end_transaction(trans, root);
7304 if (ret) {
7305 ret = 0;
7306 goto out;
7307 }
7308
7309 /*
7310 * adjust disk_bytenr and num_bytes to cover just the bytes
7311 * in this extent we are about to write. If there
7312 * are any csums in that range we have to cow in order
7313 * to keep the csums correct
7314 */
7315 disk_bytenr += backref_offset;
7316 disk_bytenr += offset - key.offset;
7317 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7318 goto out;
7319 /*
7320 * all of the above have passed, it is safe to overwrite this extent
7321 * without cow
7322 */
7323 *len = num_bytes;
7324 ret = 1;
7325 out:
7326 btrfs_free_path(path);
7327 return ret;
7328 }
7329
7330 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7331 {
7332 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7333 int found = false;
7334 void **pagep = NULL;
7335 struct page *page = NULL;
7336 int start_idx;
7337 int end_idx;
7338
7339 start_idx = start >> PAGE_SHIFT;
7340
7341 /*
7342 * end is the last byte in the last page. end == start is legal
7343 */
7344 end_idx = end >> PAGE_SHIFT;
7345
7346 rcu_read_lock();
7347
7348 /* Most of the code in this while loop is lifted from
7349 * find_get_page. It's been modified to begin searching from a
7350 * page and return just the first page found in that range. If the
7351 * found idx is less than or equal to the end idx then we know that
7352 * a page exists. If no pages are found or if those pages are
7353 * outside of the range then we're fine (yay!) */
7354 while (page == NULL &&
7355 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7356 page = radix_tree_deref_slot(pagep);
7357 if (unlikely(!page))
7358 break;
7359
7360 if (radix_tree_exception(page)) {
7361 if (radix_tree_deref_retry(page)) {
7362 page = NULL;
7363 continue;
7364 }
7365 /*
7366 * Otherwise, shmem/tmpfs must be storing a swap entry
7367 * here as an exceptional entry: so return it without
7368 * attempting to raise page count.
7369 */
7370 page = NULL;
7371 break; /* TODO: Is this relevant for this use case? */
7372 }
7373
7374 if (!page_cache_get_speculative(page)) {
7375 page = NULL;
7376 continue;
7377 }
7378
7379 /*
7380 * Has the page moved?
7381 * This is part of the lockless pagecache protocol. See
7382 * include/linux/pagemap.h for details.
7383 */
7384 if (unlikely(page != *pagep)) {
7385 put_page(page);
7386 page = NULL;
7387 }
7388 }
7389
7390 if (page) {
7391 if (page->index <= end_idx)
7392 found = true;
7393 put_page(page);
7394 }
7395
7396 rcu_read_unlock();
7397 return found;
7398 }
7399
7400 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7401 struct extent_state **cached_state, int writing)
7402 {
7403 struct btrfs_ordered_extent *ordered;
7404 int ret = 0;
7405
7406 while (1) {
7407 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7408 cached_state);
7409 /*
7410 * We're concerned with the entire range that we're going to be
7411 * doing DIO to, so we need to make sure theres no ordered
7412 * extents in this range.
7413 */
7414 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7415 lockend - lockstart + 1);
7416
7417 /*
7418 * We need to make sure there are no buffered pages in this
7419 * range either, we could have raced between the invalidate in
7420 * generic_file_direct_write and locking the extent. The
7421 * invalidate needs to happen so that reads after a write do not
7422 * get stale data.
7423 */
7424 if (!ordered &&
7425 (!writing ||
7426 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7427 break;
7428
7429 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7430 cached_state, GFP_NOFS);
7431
7432 if (ordered) {
7433 /*
7434 * If we are doing a DIO read and the ordered extent we
7435 * found is for a buffered write, we can not wait for it
7436 * to complete and retry, because if we do so we can
7437 * deadlock with concurrent buffered writes on page
7438 * locks. This happens only if our DIO read covers more
7439 * than one extent map, if at this point has already
7440 * created an ordered extent for a previous extent map
7441 * and locked its range in the inode's io tree, and a
7442 * concurrent write against that previous extent map's
7443 * range and this range started (we unlock the ranges
7444 * in the io tree only when the bios complete and
7445 * buffered writes always lock pages before attempting
7446 * to lock range in the io tree).
7447 */
7448 if (writing ||
7449 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7450 btrfs_start_ordered_extent(inode, ordered, 1);
7451 else
7452 ret = -ENOTBLK;
7453 btrfs_put_ordered_extent(ordered);
7454 } else {
7455 /*
7456 * We could trigger writeback for this range (and wait
7457 * for it to complete) and then invalidate the pages for
7458 * this range (through invalidate_inode_pages2_range()),
7459 * but that can lead us to a deadlock with a concurrent
7460 * call to readpages() (a buffered read or a defrag call
7461 * triggered a readahead) on a page lock due to an
7462 * ordered dio extent we created before but did not have
7463 * yet a corresponding bio submitted (whence it can not
7464 * complete), which makes readpages() wait for that
7465 * ordered extent to complete while holding a lock on
7466 * that page.
7467 */
7468 ret = -ENOTBLK;
7469 }
7470
7471 if (ret)
7472 break;
7473
7474 cond_resched();
7475 }
7476
7477 return ret;
7478 }
7479
7480 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7481 u64 len, u64 orig_start,
7482 u64 block_start, u64 block_len,
7483 u64 orig_block_len, u64 ram_bytes,
7484 int type)
7485 {
7486 struct extent_map_tree *em_tree;
7487 struct extent_map *em;
7488 struct btrfs_root *root = BTRFS_I(inode)->root;
7489 int ret;
7490
7491 em_tree = &BTRFS_I(inode)->extent_tree;
7492 em = alloc_extent_map();
7493 if (!em)
7494 return ERR_PTR(-ENOMEM);
7495
7496 em->start = start;
7497 em->orig_start = orig_start;
7498 em->mod_start = start;
7499 em->mod_len = len;
7500 em->len = len;
7501 em->block_len = block_len;
7502 em->block_start = block_start;
7503 em->bdev = root->fs_info->fs_devices->latest_bdev;
7504 em->orig_block_len = orig_block_len;
7505 em->ram_bytes = ram_bytes;
7506 em->generation = -1;
7507 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7508 if (type == BTRFS_ORDERED_PREALLOC)
7509 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7510
7511 do {
7512 btrfs_drop_extent_cache(inode, em->start,
7513 em->start + em->len - 1, 0);
7514 write_lock(&em_tree->lock);
7515 ret = add_extent_mapping(em_tree, em, 1);
7516 write_unlock(&em_tree->lock);
7517 } while (ret == -EEXIST);
7518
7519 if (ret) {
7520 free_extent_map(em);
7521 return ERR_PTR(ret);
7522 }
7523
7524 return em;
7525 }
7526
7527 static void adjust_dio_outstanding_extents(struct inode *inode,
7528 struct btrfs_dio_data *dio_data,
7529 const u64 len)
7530 {
7531 unsigned num_extents;
7532
7533 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7534 BTRFS_MAX_EXTENT_SIZE);
7535 /*
7536 * If we have an outstanding_extents count still set then we're
7537 * within our reservation, otherwise we need to adjust our inode
7538 * counter appropriately.
7539 */
7540 if (dio_data->outstanding_extents) {
7541 dio_data->outstanding_extents -= num_extents;
7542 } else {
7543 spin_lock(&BTRFS_I(inode)->lock);
7544 BTRFS_I(inode)->outstanding_extents += num_extents;
7545 spin_unlock(&BTRFS_I(inode)->lock);
7546 }
7547 }
7548
7549 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7550 struct buffer_head *bh_result, int create)
7551 {
7552 struct extent_map *em;
7553 struct btrfs_root *root = BTRFS_I(inode)->root;
7554 struct extent_state *cached_state = NULL;
7555 struct btrfs_dio_data *dio_data = NULL;
7556 u64 start = iblock << inode->i_blkbits;
7557 u64 lockstart, lockend;
7558 u64 len = bh_result->b_size;
7559 int unlock_bits = EXTENT_LOCKED;
7560 int ret = 0;
7561
7562 if (create)
7563 unlock_bits |= EXTENT_DIRTY;
7564 else
7565 len = min_t(u64, len, root->sectorsize);
7566
7567 lockstart = start;
7568 lockend = start + len - 1;
7569
7570 if (current->journal_info) {
7571 /*
7572 * Need to pull our outstanding extents and set journal_info to NULL so
7573 * that anything that needs to check if there's a transction doesn't get
7574 * confused.
7575 */
7576 dio_data = current->journal_info;
7577 current->journal_info = NULL;
7578 }
7579
7580 /*
7581 * If this errors out it's because we couldn't invalidate pagecache for
7582 * this range and we need to fallback to buffered.
7583 */
7584 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7585 create)) {
7586 ret = -ENOTBLK;
7587 goto err;
7588 }
7589
7590 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7591 if (IS_ERR(em)) {
7592 ret = PTR_ERR(em);
7593 goto unlock_err;
7594 }
7595
7596 /*
7597 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7598 * io. INLINE is special, and we could probably kludge it in here, but
7599 * it's still buffered so for safety lets just fall back to the generic
7600 * buffered path.
7601 *
7602 * For COMPRESSED we _have_ to read the entire extent in so we can
7603 * decompress it, so there will be buffering required no matter what we
7604 * do, so go ahead and fallback to buffered.
7605 *
7606 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7607 * to buffered IO. Don't blame me, this is the price we pay for using
7608 * the generic code.
7609 */
7610 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7611 em->block_start == EXTENT_MAP_INLINE) {
7612 free_extent_map(em);
7613 ret = -ENOTBLK;
7614 goto unlock_err;
7615 }
7616
7617 /* Just a good old fashioned hole, return */
7618 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7619 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7620 free_extent_map(em);
7621 goto unlock_err;
7622 }
7623
7624 /*
7625 * We don't allocate a new extent in the following cases
7626 *
7627 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7628 * existing extent.
7629 * 2) The extent is marked as PREALLOC. We're good to go here and can
7630 * just use the extent.
7631 *
7632 */
7633 if (!create) {
7634 len = min(len, em->len - (start - em->start));
7635 lockstart = start + len;
7636 goto unlock;
7637 }
7638
7639 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7640 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7641 em->block_start != EXTENT_MAP_HOLE)) {
7642 int type;
7643 u64 block_start, orig_start, orig_block_len, ram_bytes;
7644
7645 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7646 type = BTRFS_ORDERED_PREALLOC;
7647 else
7648 type = BTRFS_ORDERED_NOCOW;
7649 len = min(len, em->len - (start - em->start));
7650 block_start = em->block_start + (start - em->start);
7651
7652 if (can_nocow_extent(inode, start, &len, &orig_start,
7653 &orig_block_len, &ram_bytes) == 1) {
7654 if (type == BTRFS_ORDERED_PREALLOC) {
7655 free_extent_map(em);
7656 em = create_pinned_em(inode, start, len,
7657 orig_start,
7658 block_start, len,
7659 orig_block_len,
7660 ram_bytes, type);
7661 if (IS_ERR(em)) {
7662 ret = PTR_ERR(em);
7663 goto unlock_err;
7664 }
7665 }
7666
7667 ret = btrfs_add_ordered_extent_dio(inode, start,
7668 block_start, len, len, type);
7669 if (ret) {
7670 free_extent_map(em);
7671 goto unlock_err;
7672 }
7673 goto unlock;
7674 }
7675 }
7676
7677 /*
7678 * this will cow the extent, reset the len in case we changed
7679 * it above
7680 */
7681 len = bh_result->b_size;
7682 free_extent_map(em);
7683 em = btrfs_new_extent_direct(inode, start, len);
7684 if (IS_ERR(em)) {
7685 ret = PTR_ERR(em);
7686 goto unlock_err;
7687 }
7688 len = min(len, em->len - (start - em->start));
7689 unlock:
7690 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7691 inode->i_blkbits;
7692 bh_result->b_size = len;
7693 bh_result->b_bdev = em->bdev;
7694 set_buffer_mapped(bh_result);
7695 if (create) {
7696 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7697 set_buffer_new(bh_result);
7698
7699 /*
7700 * Need to update the i_size under the extent lock so buffered
7701 * readers will get the updated i_size when we unlock.
7702 */
7703 if (start + len > i_size_read(inode))
7704 i_size_write(inode, start + len);
7705
7706 adjust_dio_outstanding_extents(inode, dio_data, len);
7707 btrfs_free_reserved_data_space(inode, start, len);
7708 WARN_ON(dio_data->reserve < len);
7709 dio_data->reserve -= len;
7710 dio_data->unsubmitted_oe_range_end = start + len;
7711 current->journal_info = dio_data;
7712 }
7713
7714 /*
7715 * In the case of write we need to clear and unlock the entire range,
7716 * in the case of read we need to unlock only the end area that we
7717 * aren't using if there is any left over space.
7718 */
7719 if (lockstart < lockend) {
7720 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7721 lockend, unlock_bits, 1, 0,
7722 &cached_state, GFP_NOFS);
7723 } else {
7724 free_extent_state(cached_state);
7725 }
7726
7727 free_extent_map(em);
7728
7729 return 0;
7730
7731 unlock_err:
7732 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7733 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7734 err:
7735 if (dio_data)
7736 current->journal_info = dio_data;
7737 /*
7738 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7739 * write less data then expected, so that we don't underflow our inode's
7740 * outstanding extents counter.
7741 */
7742 if (create && dio_data)
7743 adjust_dio_outstanding_extents(inode, dio_data, len);
7744
7745 return ret;
7746 }
7747
7748 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7749 int rw, int mirror_num)
7750 {
7751 struct btrfs_root *root = BTRFS_I(inode)->root;
7752 int ret;
7753
7754 BUG_ON(rw & REQ_WRITE);
7755
7756 bio_get(bio);
7757
7758 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7759 BTRFS_WQ_ENDIO_DIO_REPAIR);
7760 if (ret)
7761 goto err;
7762
7763 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7764 err:
7765 bio_put(bio);
7766 return ret;
7767 }
7768
7769 static int btrfs_check_dio_repairable(struct inode *inode,
7770 struct bio *failed_bio,
7771 struct io_failure_record *failrec,
7772 int failed_mirror)
7773 {
7774 int num_copies;
7775
7776 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7777 failrec->logical, failrec->len);
7778 if (num_copies == 1) {
7779 /*
7780 * we only have a single copy of the data, so don't bother with
7781 * all the retry and error correction code that follows. no
7782 * matter what the error is, it is very likely to persist.
7783 */
7784 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7785 num_copies, failrec->this_mirror, failed_mirror);
7786 return 0;
7787 }
7788
7789 failrec->failed_mirror = failed_mirror;
7790 failrec->this_mirror++;
7791 if (failrec->this_mirror == failed_mirror)
7792 failrec->this_mirror++;
7793
7794 if (failrec->this_mirror > num_copies) {
7795 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7796 num_copies, failrec->this_mirror, failed_mirror);
7797 return 0;
7798 }
7799
7800 return 1;
7801 }
7802
7803 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7804 struct page *page, unsigned int pgoff,
7805 u64 start, u64 end, int failed_mirror,
7806 bio_end_io_t *repair_endio, void *repair_arg)
7807 {
7808 struct io_failure_record *failrec;
7809 struct bio *bio;
7810 int isector;
7811 int read_mode;
7812 int ret;
7813
7814 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7815
7816 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7817 if (ret)
7818 return ret;
7819
7820 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7821 failed_mirror);
7822 if (!ret) {
7823 free_io_failure(inode, failrec);
7824 return -EIO;
7825 }
7826
7827 if ((failed_bio->bi_vcnt > 1)
7828 || (failed_bio->bi_io_vec->bv_len
7829 > BTRFS_I(inode)->root->sectorsize))
7830 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7831 else
7832 read_mode = READ_SYNC;
7833
7834 isector = start - btrfs_io_bio(failed_bio)->logical;
7835 isector >>= inode->i_sb->s_blocksize_bits;
7836 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7837 pgoff, isector, repair_endio, repair_arg);
7838 if (!bio) {
7839 free_io_failure(inode, failrec);
7840 return -EIO;
7841 }
7842
7843 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7844 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7845 read_mode, failrec->this_mirror, failrec->in_validation);
7846
7847 ret = submit_dio_repair_bio(inode, bio, read_mode,
7848 failrec->this_mirror);
7849 if (ret) {
7850 free_io_failure(inode, failrec);
7851 bio_put(bio);
7852 }
7853
7854 return ret;
7855 }
7856
7857 struct btrfs_retry_complete {
7858 struct completion done;
7859 struct inode *inode;
7860 u64 start;
7861 int uptodate;
7862 };
7863
7864 static void btrfs_retry_endio_nocsum(struct bio *bio)
7865 {
7866 struct btrfs_retry_complete *done = bio->bi_private;
7867 struct inode *inode;
7868 struct bio_vec *bvec;
7869 int i;
7870
7871 if (bio->bi_error)
7872 goto end;
7873
7874 ASSERT(bio->bi_vcnt == 1);
7875 inode = bio->bi_io_vec->bv_page->mapping->host;
7876 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7877
7878 done->uptodate = 1;
7879 bio_for_each_segment_all(bvec, bio, i)
7880 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7881 end:
7882 complete(&done->done);
7883 bio_put(bio);
7884 }
7885
7886 static int __btrfs_correct_data_nocsum(struct inode *inode,
7887 struct btrfs_io_bio *io_bio)
7888 {
7889 struct btrfs_fs_info *fs_info;
7890 struct bio_vec *bvec;
7891 struct btrfs_retry_complete done;
7892 u64 start;
7893 unsigned int pgoff;
7894 u32 sectorsize;
7895 int nr_sectors;
7896 int i;
7897 int ret;
7898
7899 fs_info = BTRFS_I(inode)->root->fs_info;
7900 sectorsize = BTRFS_I(inode)->root->sectorsize;
7901
7902 start = io_bio->logical;
7903 done.inode = inode;
7904
7905 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7906 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7907 pgoff = bvec->bv_offset;
7908
7909 next_block_or_try_again:
7910 done.uptodate = 0;
7911 done.start = start;
7912 init_completion(&done.done);
7913
7914 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7915 pgoff, start, start + sectorsize - 1,
7916 io_bio->mirror_num,
7917 btrfs_retry_endio_nocsum, &done);
7918 if (ret)
7919 return ret;
7920
7921 wait_for_completion(&done.done);
7922
7923 if (!done.uptodate) {
7924 /* We might have another mirror, so try again */
7925 goto next_block_or_try_again;
7926 }
7927
7928 start += sectorsize;
7929
7930 if (nr_sectors--) {
7931 pgoff += sectorsize;
7932 goto next_block_or_try_again;
7933 }
7934 }
7935
7936 return 0;
7937 }
7938
7939 static void btrfs_retry_endio(struct bio *bio)
7940 {
7941 struct btrfs_retry_complete *done = bio->bi_private;
7942 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7943 struct inode *inode;
7944 struct bio_vec *bvec;
7945 u64 start;
7946 int uptodate;
7947 int ret;
7948 int i;
7949
7950 if (bio->bi_error)
7951 goto end;
7952
7953 uptodate = 1;
7954
7955 start = done->start;
7956
7957 ASSERT(bio->bi_vcnt == 1);
7958 inode = bio->bi_io_vec->bv_page->mapping->host;
7959 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7960
7961 bio_for_each_segment_all(bvec, bio, i) {
7962 ret = __readpage_endio_check(done->inode, io_bio, i,
7963 bvec->bv_page, bvec->bv_offset,
7964 done->start, bvec->bv_len);
7965 if (!ret)
7966 clean_io_failure(done->inode, done->start,
7967 bvec->bv_page, bvec->bv_offset);
7968 else
7969 uptodate = 0;
7970 }
7971
7972 done->uptodate = uptodate;
7973 end:
7974 complete(&done->done);
7975 bio_put(bio);
7976 }
7977
7978 static int __btrfs_subio_endio_read(struct inode *inode,
7979 struct btrfs_io_bio *io_bio, int err)
7980 {
7981 struct btrfs_fs_info *fs_info;
7982 struct bio_vec *bvec;
7983 struct btrfs_retry_complete done;
7984 u64 start;
7985 u64 offset = 0;
7986 u32 sectorsize;
7987 int nr_sectors;
7988 unsigned int pgoff;
7989 int csum_pos;
7990 int i;
7991 int ret;
7992
7993 fs_info = BTRFS_I(inode)->root->fs_info;
7994 sectorsize = BTRFS_I(inode)->root->sectorsize;
7995
7996 err = 0;
7997 start = io_bio->logical;
7998 done.inode = inode;
7999
8000 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8001 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8002
8003 pgoff = bvec->bv_offset;
8004 next_block:
8005 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8006 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8007 bvec->bv_page, pgoff, start,
8008 sectorsize);
8009 if (likely(!ret))
8010 goto next;
8011 try_again:
8012 done.uptodate = 0;
8013 done.start = start;
8014 init_completion(&done.done);
8015
8016 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8017 pgoff, start, start + sectorsize - 1,
8018 io_bio->mirror_num,
8019 btrfs_retry_endio, &done);
8020 if (ret) {
8021 err = ret;
8022 goto next;
8023 }
8024
8025 wait_for_completion(&done.done);
8026
8027 if (!done.uptodate) {
8028 /* We might have another mirror, so try again */
8029 goto try_again;
8030 }
8031 next:
8032 offset += sectorsize;
8033 start += sectorsize;
8034
8035 ASSERT(nr_sectors);
8036
8037 if (--nr_sectors) {
8038 pgoff += sectorsize;
8039 goto next_block;
8040 }
8041 }
8042
8043 return err;
8044 }
8045
8046 static int btrfs_subio_endio_read(struct inode *inode,
8047 struct btrfs_io_bio *io_bio, int err)
8048 {
8049 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8050
8051 if (skip_csum) {
8052 if (unlikely(err))
8053 return __btrfs_correct_data_nocsum(inode, io_bio);
8054 else
8055 return 0;
8056 } else {
8057 return __btrfs_subio_endio_read(inode, io_bio, err);
8058 }
8059 }
8060
8061 static void btrfs_endio_direct_read(struct bio *bio)
8062 {
8063 struct btrfs_dio_private *dip = bio->bi_private;
8064 struct inode *inode = dip->inode;
8065 struct bio *dio_bio;
8066 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8067 int err = bio->bi_error;
8068
8069 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8070 err = btrfs_subio_endio_read(inode, io_bio, err);
8071
8072 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8073 dip->logical_offset + dip->bytes - 1);
8074 dio_bio = dip->dio_bio;
8075
8076 kfree(dip);
8077
8078 dio_bio->bi_error = bio->bi_error;
8079 dio_end_io(dio_bio, bio->bi_error);
8080
8081 if (io_bio->end_io)
8082 io_bio->end_io(io_bio, err);
8083 bio_put(bio);
8084 }
8085
8086 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8087 const u64 offset,
8088 const u64 bytes,
8089 const int uptodate)
8090 {
8091 struct btrfs_root *root = BTRFS_I(inode)->root;
8092 struct btrfs_ordered_extent *ordered = NULL;
8093 u64 ordered_offset = offset;
8094 u64 ordered_bytes = bytes;
8095 int ret;
8096
8097 again:
8098 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8099 &ordered_offset,
8100 ordered_bytes,
8101 uptodate);
8102 if (!ret)
8103 goto out_test;
8104
8105 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8106 finish_ordered_fn, NULL, NULL);
8107 btrfs_queue_work(root->fs_info->endio_write_workers,
8108 &ordered->work);
8109 out_test:
8110 /*
8111 * our bio might span multiple ordered extents. If we haven't
8112 * completed the accounting for the whole dio, go back and try again
8113 */
8114 if (ordered_offset < offset + bytes) {
8115 ordered_bytes = offset + bytes - ordered_offset;
8116 ordered = NULL;
8117 goto again;
8118 }
8119 }
8120
8121 static void btrfs_endio_direct_write(struct bio *bio)
8122 {
8123 struct btrfs_dio_private *dip = bio->bi_private;
8124 struct bio *dio_bio = dip->dio_bio;
8125
8126 btrfs_endio_direct_write_update_ordered(dip->inode,
8127 dip->logical_offset,
8128 dip->bytes,
8129 !bio->bi_error);
8130
8131 kfree(dip);
8132
8133 dio_bio->bi_error = bio->bi_error;
8134 dio_end_io(dio_bio, bio->bi_error);
8135 bio_put(bio);
8136 }
8137
8138 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8139 struct bio *bio, int mirror_num,
8140 unsigned long bio_flags, u64 offset)
8141 {
8142 int ret;
8143 struct btrfs_root *root = BTRFS_I(inode)->root;
8144 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8145 BUG_ON(ret); /* -ENOMEM */
8146 return 0;
8147 }
8148
8149 static void btrfs_end_dio_bio(struct bio *bio)
8150 {
8151 struct btrfs_dio_private *dip = bio->bi_private;
8152 int err = bio->bi_error;
8153
8154 if (err)
8155 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8156 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8157 btrfs_ino(dip->inode), bio->bi_rw,
8158 (unsigned long long)bio->bi_iter.bi_sector,
8159 bio->bi_iter.bi_size, err);
8160
8161 if (dip->subio_endio)
8162 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8163
8164 if (err) {
8165 dip->errors = 1;
8166
8167 /*
8168 * before atomic variable goto zero, we must make sure
8169 * dip->errors is perceived to be set.
8170 */
8171 smp_mb__before_atomic();
8172 }
8173
8174 /* if there are more bios still pending for this dio, just exit */
8175 if (!atomic_dec_and_test(&dip->pending_bios))
8176 goto out;
8177
8178 if (dip->errors) {
8179 bio_io_error(dip->orig_bio);
8180 } else {
8181 dip->dio_bio->bi_error = 0;
8182 bio_endio(dip->orig_bio);
8183 }
8184 out:
8185 bio_put(bio);
8186 }
8187
8188 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8189 u64 first_sector, gfp_t gfp_flags)
8190 {
8191 struct bio *bio;
8192 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8193 if (bio)
8194 bio_associate_current(bio);
8195 return bio;
8196 }
8197
8198 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8199 struct inode *inode,
8200 struct btrfs_dio_private *dip,
8201 struct bio *bio,
8202 u64 file_offset)
8203 {
8204 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8205 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8206 int ret;
8207
8208 /*
8209 * We load all the csum data we need when we submit
8210 * the first bio to reduce the csum tree search and
8211 * contention.
8212 */
8213 if (dip->logical_offset == file_offset) {
8214 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8215 file_offset);
8216 if (ret)
8217 return ret;
8218 }
8219
8220 if (bio == dip->orig_bio)
8221 return 0;
8222
8223 file_offset -= dip->logical_offset;
8224 file_offset >>= inode->i_sb->s_blocksize_bits;
8225 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8226
8227 return 0;
8228 }
8229
8230 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8231 int rw, u64 file_offset, int skip_sum,
8232 int async_submit)
8233 {
8234 struct btrfs_dio_private *dip = bio->bi_private;
8235 int write = rw & REQ_WRITE;
8236 struct btrfs_root *root = BTRFS_I(inode)->root;
8237 int ret;
8238
8239 if (async_submit)
8240 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8241
8242 bio_get(bio);
8243
8244 if (!write) {
8245 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8246 BTRFS_WQ_ENDIO_DATA);
8247 if (ret)
8248 goto err;
8249 }
8250
8251 if (skip_sum)
8252 goto map;
8253
8254 if (write && async_submit) {
8255 ret = btrfs_wq_submit_bio(root->fs_info,
8256 inode, rw, bio, 0, 0,
8257 file_offset,
8258 __btrfs_submit_bio_start_direct_io,
8259 __btrfs_submit_bio_done);
8260 goto err;
8261 } else if (write) {
8262 /*
8263 * If we aren't doing async submit, calculate the csum of the
8264 * bio now.
8265 */
8266 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8267 if (ret)
8268 goto err;
8269 } else {
8270 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8271 file_offset);
8272 if (ret)
8273 goto err;
8274 }
8275 map:
8276 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8277 err:
8278 bio_put(bio);
8279 return ret;
8280 }
8281
8282 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8283 int skip_sum)
8284 {
8285 struct inode *inode = dip->inode;
8286 struct btrfs_root *root = BTRFS_I(inode)->root;
8287 struct bio *bio;
8288 struct bio *orig_bio = dip->orig_bio;
8289 struct bio_vec *bvec = orig_bio->bi_io_vec;
8290 u64 start_sector = orig_bio->bi_iter.bi_sector;
8291 u64 file_offset = dip->logical_offset;
8292 u64 submit_len = 0;
8293 u64 map_length;
8294 u32 blocksize = root->sectorsize;
8295 int async_submit = 0;
8296 int nr_sectors;
8297 int ret;
8298 int i;
8299
8300 map_length = orig_bio->bi_iter.bi_size;
8301 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8302 &map_length, NULL, 0);
8303 if (ret)
8304 return -EIO;
8305
8306 if (map_length >= orig_bio->bi_iter.bi_size) {
8307 bio = orig_bio;
8308 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8309 goto submit;
8310 }
8311
8312 /* async crcs make it difficult to collect full stripe writes. */
8313 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8314 async_submit = 0;
8315 else
8316 async_submit = 1;
8317
8318 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8319 if (!bio)
8320 return -ENOMEM;
8321
8322 bio->bi_private = dip;
8323 bio->bi_end_io = btrfs_end_dio_bio;
8324 btrfs_io_bio(bio)->logical = file_offset;
8325 atomic_inc(&dip->pending_bios);
8326
8327 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8328 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8329 i = 0;
8330 next_block:
8331 if (unlikely(map_length < submit_len + blocksize ||
8332 bio_add_page(bio, bvec->bv_page, blocksize,
8333 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8334 /*
8335 * inc the count before we submit the bio so
8336 * we know the end IO handler won't happen before
8337 * we inc the count. Otherwise, the dip might get freed
8338 * before we're done setting it up
8339 */
8340 atomic_inc(&dip->pending_bios);
8341 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8342 file_offset, skip_sum,
8343 async_submit);
8344 if (ret) {
8345 bio_put(bio);
8346 atomic_dec(&dip->pending_bios);
8347 goto out_err;
8348 }
8349
8350 start_sector += submit_len >> 9;
8351 file_offset += submit_len;
8352
8353 submit_len = 0;
8354
8355 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8356 start_sector, GFP_NOFS);
8357 if (!bio)
8358 goto out_err;
8359 bio->bi_private = dip;
8360 bio->bi_end_io = btrfs_end_dio_bio;
8361 btrfs_io_bio(bio)->logical = file_offset;
8362
8363 map_length = orig_bio->bi_iter.bi_size;
8364 ret = btrfs_map_block(root->fs_info, rw,
8365 start_sector << 9,
8366 &map_length, NULL, 0);
8367 if (ret) {
8368 bio_put(bio);
8369 goto out_err;
8370 }
8371
8372 goto next_block;
8373 } else {
8374 submit_len += blocksize;
8375 if (--nr_sectors) {
8376 i++;
8377 goto next_block;
8378 }
8379 bvec++;
8380 }
8381 }
8382
8383 submit:
8384 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8385 async_submit);
8386 if (!ret)
8387 return 0;
8388
8389 bio_put(bio);
8390 out_err:
8391 dip->errors = 1;
8392 /*
8393 * before atomic variable goto zero, we must
8394 * make sure dip->errors is perceived to be set.
8395 */
8396 smp_mb__before_atomic();
8397 if (atomic_dec_and_test(&dip->pending_bios))
8398 bio_io_error(dip->orig_bio);
8399
8400 /* bio_end_io() will handle error, so we needn't return it */
8401 return 0;
8402 }
8403
8404 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8405 struct inode *inode, loff_t file_offset)
8406 {
8407 struct btrfs_dio_private *dip = NULL;
8408 struct bio *io_bio = NULL;
8409 struct btrfs_io_bio *btrfs_bio;
8410 int skip_sum;
8411 int write = rw & REQ_WRITE;
8412 int ret = 0;
8413
8414 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8415
8416 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8417 if (!io_bio) {
8418 ret = -ENOMEM;
8419 goto free_ordered;
8420 }
8421
8422 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8423 if (!dip) {
8424 ret = -ENOMEM;
8425 goto free_ordered;
8426 }
8427
8428 dip->private = dio_bio->bi_private;
8429 dip->inode = inode;
8430 dip->logical_offset = file_offset;
8431 dip->bytes = dio_bio->bi_iter.bi_size;
8432 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8433 io_bio->bi_private = dip;
8434 dip->orig_bio = io_bio;
8435 dip->dio_bio = dio_bio;
8436 atomic_set(&dip->pending_bios, 0);
8437 btrfs_bio = btrfs_io_bio(io_bio);
8438 btrfs_bio->logical = file_offset;
8439
8440 if (write) {
8441 io_bio->bi_end_io = btrfs_endio_direct_write;
8442 } else {
8443 io_bio->bi_end_io = btrfs_endio_direct_read;
8444 dip->subio_endio = btrfs_subio_endio_read;
8445 }
8446
8447 /*
8448 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8449 * even if we fail to submit a bio, because in such case we do the
8450 * corresponding error handling below and it must not be done a second
8451 * time by btrfs_direct_IO().
8452 */
8453 if (write) {
8454 struct btrfs_dio_data *dio_data = current->journal_info;
8455
8456 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8457 dip->bytes;
8458 dio_data->unsubmitted_oe_range_start =
8459 dio_data->unsubmitted_oe_range_end;
8460 }
8461
8462 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8463 if (!ret)
8464 return;
8465
8466 if (btrfs_bio->end_io)
8467 btrfs_bio->end_io(btrfs_bio, ret);
8468
8469 free_ordered:
8470 /*
8471 * If we arrived here it means either we failed to submit the dip
8472 * or we either failed to clone the dio_bio or failed to allocate the
8473 * dip. If we cloned the dio_bio and allocated the dip, we can just
8474 * call bio_endio against our io_bio so that we get proper resource
8475 * cleanup if we fail to submit the dip, otherwise, we must do the
8476 * same as btrfs_endio_direct_[write|read] because we can't call these
8477 * callbacks - they require an allocated dip and a clone of dio_bio.
8478 */
8479 if (io_bio && dip) {
8480 io_bio->bi_error = -EIO;
8481 bio_endio(io_bio);
8482 /*
8483 * The end io callbacks free our dip, do the final put on io_bio
8484 * and all the cleanup and final put for dio_bio (through
8485 * dio_end_io()).
8486 */
8487 dip = NULL;
8488 io_bio = NULL;
8489 } else {
8490 if (write)
8491 btrfs_endio_direct_write_update_ordered(inode,
8492 file_offset,
8493 dio_bio->bi_iter.bi_size,
8494 0);
8495 else
8496 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8497 file_offset + dio_bio->bi_iter.bi_size - 1);
8498
8499 dio_bio->bi_error = -EIO;
8500 /*
8501 * Releases and cleans up our dio_bio, no need to bio_put()
8502 * nor bio_endio()/bio_io_error() against dio_bio.
8503 */
8504 dio_end_io(dio_bio, ret);
8505 }
8506 if (io_bio)
8507 bio_put(io_bio);
8508 kfree(dip);
8509 }
8510
8511 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8512 const struct iov_iter *iter, loff_t offset)
8513 {
8514 int seg;
8515 int i;
8516 unsigned blocksize_mask = root->sectorsize - 1;
8517 ssize_t retval = -EINVAL;
8518
8519 if (offset & blocksize_mask)
8520 goto out;
8521
8522 if (iov_iter_alignment(iter) & blocksize_mask)
8523 goto out;
8524
8525 /* If this is a write we don't need to check anymore */
8526 if (iov_iter_rw(iter) == WRITE)
8527 return 0;
8528 /*
8529 * Check to make sure we don't have duplicate iov_base's in this
8530 * iovec, if so return EINVAL, otherwise we'll get csum errors
8531 * when reading back.
8532 */
8533 for (seg = 0; seg < iter->nr_segs; seg++) {
8534 for (i = seg + 1; i < iter->nr_segs; i++) {
8535 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8536 goto out;
8537 }
8538 }
8539 retval = 0;
8540 out:
8541 return retval;
8542 }
8543
8544 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8545 loff_t offset)
8546 {
8547 struct file *file = iocb->ki_filp;
8548 struct inode *inode = file->f_mapping->host;
8549 struct btrfs_root *root = BTRFS_I(inode)->root;
8550 struct btrfs_dio_data dio_data = { 0 };
8551 size_t count = 0;
8552 int flags = 0;
8553 bool wakeup = true;
8554 bool relock = false;
8555 ssize_t ret;
8556
8557 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8558 return 0;
8559
8560 inode_dio_begin(inode);
8561 smp_mb__after_atomic();
8562
8563 /*
8564 * The generic stuff only does filemap_write_and_wait_range, which
8565 * isn't enough if we've written compressed pages to this area, so
8566 * we need to flush the dirty pages again to make absolutely sure
8567 * that any outstanding dirty pages are on disk.
8568 */
8569 count = iov_iter_count(iter);
8570 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8571 &BTRFS_I(inode)->runtime_flags))
8572 filemap_fdatawrite_range(inode->i_mapping, offset,
8573 offset + count - 1);
8574
8575 if (iov_iter_rw(iter) == WRITE) {
8576 /*
8577 * If the write DIO is beyond the EOF, we need update
8578 * the isize, but it is protected by i_mutex. So we can
8579 * not unlock the i_mutex at this case.
8580 */
8581 if (offset + count <= inode->i_size) {
8582 inode_unlock(inode);
8583 relock = true;
8584 }
8585 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8586 if (ret)
8587 goto out;
8588 dio_data.outstanding_extents = div64_u64(count +
8589 BTRFS_MAX_EXTENT_SIZE - 1,
8590 BTRFS_MAX_EXTENT_SIZE);
8591
8592 /*
8593 * We need to know how many extents we reserved so that we can
8594 * do the accounting properly if we go over the number we
8595 * originally calculated. Abuse current->journal_info for this.
8596 */
8597 dio_data.reserve = round_up(count, root->sectorsize);
8598 dio_data.unsubmitted_oe_range_start = (u64)offset;
8599 dio_data.unsubmitted_oe_range_end = (u64)offset;
8600 current->journal_info = &dio_data;
8601 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8602 &BTRFS_I(inode)->runtime_flags)) {
8603 inode_dio_end(inode);
8604 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8605 wakeup = false;
8606 }
8607
8608 ret = __blockdev_direct_IO(iocb, inode,
8609 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8610 iter, offset, btrfs_get_blocks_direct, NULL,
8611 btrfs_submit_direct, flags);
8612 if (iov_iter_rw(iter) == WRITE) {
8613 current->journal_info = NULL;
8614 if (ret < 0 && ret != -EIOCBQUEUED) {
8615 if (dio_data.reserve)
8616 btrfs_delalloc_release_space(inode, offset,
8617 dio_data.reserve);
8618 /*
8619 * On error we might have left some ordered extents
8620 * without submitting corresponding bios for them, so
8621 * cleanup them up to avoid other tasks getting them
8622 * and waiting for them to complete forever.
8623 */
8624 if (dio_data.unsubmitted_oe_range_start <
8625 dio_data.unsubmitted_oe_range_end)
8626 btrfs_endio_direct_write_update_ordered(inode,
8627 dio_data.unsubmitted_oe_range_start,
8628 dio_data.unsubmitted_oe_range_end -
8629 dio_data.unsubmitted_oe_range_start,
8630 0);
8631 } else if (ret >= 0 && (size_t)ret < count)
8632 btrfs_delalloc_release_space(inode, offset,
8633 count - (size_t)ret);
8634 }
8635 out:
8636 if (wakeup)
8637 inode_dio_end(inode);
8638 if (relock)
8639 inode_lock(inode);
8640
8641 return ret;
8642 }
8643
8644 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8645
8646 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8647 __u64 start, __u64 len)
8648 {
8649 int ret;
8650
8651 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8652 if (ret)
8653 return ret;
8654
8655 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8656 }
8657
8658 int btrfs_readpage(struct file *file, struct page *page)
8659 {
8660 struct extent_io_tree *tree;
8661 tree = &BTRFS_I(page->mapping->host)->io_tree;
8662 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8663 }
8664
8665 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8666 {
8667 struct extent_io_tree *tree;
8668 struct inode *inode = page->mapping->host;
8669 int ret;
8670
8671 if (current->flags & PF_MEMALLOC) {
8672 redirty_page_for_writepage(wbc, page);
8673 unlock_page(page);
8674 return 0;
8675 }
8676
8677 /*
8678 * If we are under memory pressure we will call this directly from the
8679 * VM, we need to make sure we have the inode referenced for the ordered
8680 * extent. If not just return like we didn't do anything.
8681 */
8682 if (!igrab(inode)) {
8683 redirty_page_for_writepage(wbc, page);
8684 return AOP_WRITEPAGE_ACTIVATE;
8685 }
8686 tree = &BTRFS_I(page->mapping->host)->io_tree;
8687 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8688 btrfs_add_delayed_iput(inode);
8689 return ret;
8690 }
8691
8692 static int btrfs_writepages(struct address_space *mapping,
8693 struct writeback_control *wbc)
8694 {
8695 struct extent_io_tree *tree;
8696
8697 tree = &BTRFS_I(mapping->host)->io_tree;
8698 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8699 }
8700
8701 static int
8702 btrfs_readpages(struct file *file, struct address_space *mapping,
8703 struct list_head *pages, unsigned nr_pages)
8704 {
8705 struct extent_io_tree *tree;
8706 tree = &BTRFS_I(mapping->host)->io_tree;
8707 return extent_readpages(tree, mapping, pages, nr_pages,
8708 btrfs_get_extent);
8709 }
8710 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8711 {
8712 struct extent_io_tree *tree;
8713 struct extent_map_tree *map;
8714 int ret;
8715
8716 tree = &BTRFS_I(page->mapping->host)->io_tree;
8717 map = &BTRFS_I(page->mapping->host)->extent_tree;
8718 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8719 if (ret == 1) {
8720 ClearPagePrivate(page);
8721 set_page_private(page, 0);
8722 put_page(page);
8723 }
8724 return ret;
8725 }
8726
8727 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8728 {
8729 if (PageWriteback(page) || PageDirty(page))
8730 return 0;
8731 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8732 }
8733
8734 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8735 unsigned int length)
8736 {
8737 struct inode *inode = page->mapping->host;
8738 struct extent_io_tree *tree;
8739 struct btrfs_ordered_extent *ordered;
8740 struct extent_state *cached_state = NULL;
8741 u64 page_start = page_offset(page);
8742 u64 page_end = page_start + PAGE_SIZE - 1;
8743 u64 start;
8744 u64 end;
8745 int inode_evicting = inode->i_state & I_FREEING;
8746
8747 /*
8748 * we have the page locked, so new writeback can't start,
8749 * and the dirty bit won't be cleared while we are here.
8750 *
8751 * Wait for IO on this page so that we can safely clear
8752 * the PagePrivate2 bit and do ordered accounting
8753 */
8754 wait_on_page_writeback(page);
8755
8756 tree = &BTRFS_I(inode)->io_tree;
8757 if (offset) {
8758 btrfs_releasepage(page, GFP_NOFS);
8759 return;
8760 }
8761
8762 if (!inode_evicting)
8763 lock_extent_bits(tree, page_start, page_end, &cached_state);
8764 again:
8765 start = page_start;
8766 ordered = btrfs_lookup_ordered_range(inode, start,
8767 page_end - start + 1);
8768 if (ordered) {
8769 end = min(page_end, ordered->file_offset + ordered->len - 1);
8770 /*
8771 * IO on this page will never be started, so we need
8772 * to account for any ordered extents now
8773 */
8774 if (!inode_evicting)
8775 clear_extent_bit(tree, start, end,
8776 EXTENT_DIRTY | EXTENT_DELALLOC |
8777 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8778 EXTENT_DEFRAG, 1, 0, &cached_state,
8779 GFP_NOFS);
8780 /*
8781 * whoever cleared the private bit is responsible
8782 * for the finish_ordered_io
8783 */
8784 if (TestClearPagePrivate2(page)) {
8785 struct btrfs_ordered_inode_tree *tree;
8786 u64 new_len;
8787
8788 tree = &BTRFS_I(inode)->ordered_tree;
8789
8790 spin_lock_irq(&tree->lock);
8791 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8792 new_len = start - ordered->file_offset;
8793 if (new_len < ordered->truncated_len)
8794 ordered->truncated_len = new_len;
8795 spin_unlock_irq(&tree->lock);
8796
8797 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8798 start,
8799 end - start + 1, 1))
8800 btrfs_finish_ordered_io(ordered);
8801 }
8802 btrfs_put_ordered_extent(ordered);
8803 if (!inode_evicting) {
8804 cached_state = NULL;
8805 lock_extent_bits(tree, start, end,
8806 &cached_state);
8807 }
8808
8809 start = end + 1;
8810 if (start < page_end)
8811 goto again;
8812 }
8813
8814 /*
8815 * Qgroup reserved space handler
8816 * Page here will be either
8817 * 1) Already written to disk
8818 * In this case, its reserved space is released from data rsv map
8819 * and will be freed by delayed_ref handler finally.
8820 * So even we call qgroup_free_data(), it won't decrease reserved
8821 * space.
8822 * 2) Not written to disk
8823 * This means the reserved space should be freed here.
8824 */
8825 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8826 if (!inode_evicting) {
8827 clear_extent_bit(tree, page_start, page_end,
8828 EXTENT_LOCKED | EXTENT_DIRTY |
8829 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8830 EXTENT_DEFRAG, 1, 1,
8831 &cached_state, GFP_NOFS);
8832
8833 __btrfs_releasepage(page, GFP_NOFS);
8834 }
8835
8836 ClearPageChecked(page);
8837 if (PagePrivate(page)) {
8838 ClearPagePrivate(page);
8839 set_page_private(page, 0);
8840 put_page(page);
8841 }
8842 }
8843
8844 /*
8845 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8846 * called from a page fault handler when a page is first dirtied. Hence we must
8847 * be careful to check for EOF conditions here. We set the page up correctly
8848 * for a written page which means we get ENOSPC checking when writing into
8849 * holes and correct delalloc and unwritten extent mapping on filesystems that
8850 * support these features.
8851 *
8852 * We are not allowed to take the i_mutex here so we have to play games to
8853 * protect against truncate races as the page could now be beyond EOF. Because
8854 * vmtruncate() writes the inode size before removing pages, once we have the
8855 * page lock we can determine safely if the page is beyond EOF. If it is not
8856 * beyond EOF, then the page is guaranteed safe against truncation until we
8857 * unlock the page.
8858 */
8859 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8860 {
8861 struct page *page = vmf->page;
8862 struct inode *inode = file_inode(vma->vm_file);
8863 struct btrfs_root *root = BTRFS_I(inode)->root;
8864 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8865 struct btrfs_ordered_extent *ordered;
8866 struct extent_state *cached_state = NULL;
8867 char *kaddr;
8868 unsigned long zero_start;
8869 loff_t size;
8870 int ret;
8871 int reserved = 0;
8872 u64 reserved_space;
8873 u64 page_start;
8874 u64 page_end;
8875 u64 end;
8876
8877 reserved_space = PAGE_SIZE;
8878
8879 sb_start_pagefault(inode->i_sb);
8880 page_start = page_offset(page);
8881 page_end = page_start + PAGE_SIZE - 1;
8882 end = page_end;
8883
8884 /*
8885 * Reserving delalloc space after obtaining the page lock can lead to
8886 * deadlock. For example, if a dirty page is locked by this function
8887 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8888 * dirty page write out, then the btrfs_writepage() function could
8889 * end up waiting indefinitely to get a lock on the page currently
8890 * being processed by btrfs_page_mkwrite() function.
8891 */
8892 ret = btrfs_delalloc_reserve_space(inode, page_start,
8893 reserved_space);
8894 if (!ret) {
8895 ret = file_update_time(vma->vm_file);
8896 reserved = 1;
8897 }
8898 if (ret) {
8899 if (ret == -ENOMEM)
8900 ret = VM_FAULT_OOM;
8901 else /* -ENOSPC, -EIO, etc */
8902 ret = VM_FAULT_SIGBUS;
8903 if (reserved)
8904 goto out;
8905 goto out_noreserve;
8906 }
8907
8908 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8909 again:
8910 lock_page(page);
8911 size = i_size_read(inode);
8912
8913 if ((page->mapping != inode->i_mapping) ||
8914 (page_start >= size)) {
8915 /* page got truncated out from underneath us */
8916 goto out_unlock;
8917 }
8918 wait_on_page_writeback(page);
8919
8920 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8921 set_page_extent_mapped(page);
8922
8923 /*
8924 * we can't set the delalloc bits if there are pending ordered
8925 * extents. Drop our locks and wait for them to finish
8926 */
8927 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8928 if (ordered) {
8929 unlock_extent_cached(io_tree, page_start, page_end,
8930 &cached_state, GFP_NOFS);
8931 unlock_page(page);
8932 btrfs_start_ordered_extent(inode, ordered, 1);
8933 btrfs_put_ordered_extent(ordered);
8934 goto again;
8935 }
8936
8937 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8938 reserved_space = round_up(size - page_start, root->sectorsize);
8939 if (reserved_space < PAGE_SIZE) {
8940 end = page_start + reserved_space - 1;
8941 spin_lock(&BTRFS_I(inode)->lock);
8942 BTRFS_I(inode)->outstanding_extents++;
8943 spin_unlock(&BTRFS_I(inode)->lock);
8944 btrfs_delalloc_release_space(inode, page_start,
8945 PAGE_SIZE - reserved_space);
8946 }
8947 }
8948
8949 /*
8950 * XXX - page_mkwrite gets called every time the page is dirtied, even
8951 * if it was already dirty, so for space accounting reasons we need to
8952 * clear any delalloc bits for the range we are fixing to save. There
8953 * is probably a better way to do this, but for now keep consistent with
8954 * prepare_pages in the normal write path.
8955 */
8956 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8957 EXTENT_DIRTY | EXTENT_DELALLOC |
8958 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8959 0, 0, &cached_state, GFP_NOFS);
8960
8961 ret = btrfs_set_extent_delalloc(inode, page_start, end,
8962 &cached_state);
8963 if (ret) {
8964 unlock_extent_cached(io_tree, page_start, page_end,
8965 &cached_state, GFP_NOFS);
8966 ret = VM_FAULT_SIGBUS;
8967 goto out_unlock;
8968 }
8969 ret = 0;
8970
8971 /* page is wholly or partially inside EOF */
8972 if (page_start + PAGE_SIZE > size)
8973 zero_start = size & ~PAGE_MASK;
8974 else
8975 zero_start = PAGE_SIZE;
8976
8977 if (zero_start != PAGE_SIZE) {
8978 kaddr = kmap(page);
8979 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8980 flush_dcache_page(page);
8981 kunmap(page);
8982 }
8983 ClearPageChecked(page);
8984 set_page_dirty(page);
8985 SetPageUptodate(page);
8986
8987 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8988 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8989 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8990
8991 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8992
8993 out_unlock:
8994 if (!ret) {
8995 sb_end_pagefault(inode->i_sb);
8996 return VM_FAULT_LOCKED;
8997 }
8998 unlock_page(page);
8999 out:
9000 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9001 out_noreserve:
9002 sb_end_pagefault(inode->i_sb);
9003 return ret;
9004 }
9005
9006 static int btrfs_truncate(struct inode *inode)
9007 {
9008 struct btrfs_root *root = BTRFS_I(inode)->root;
9009 struct btrfs_block_rsv *rsv;
9010 int ret = 0;
9011 int err = 0;
9012 struct btrfs_trans_handle *trans;
9013 u64 mask = root->sectorsize - 1;
9014 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9015
9016 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9017 (u64)-1);
9018 if (ret)
9019 return ret;
9020
9021 /*
9022 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
9023 * 3 things going on here
9024 *
9025 * 1) We need to reserve space for our orphan item and the space to
9026 * delete our orphan item. Lord knows we don't want to have a dangling
9027 * orphan item because we didn't reserve space to remove it.
9028 *
9029 * 2) We need to reserve space to update our inode.
9030 *
9031 * 3) We need to have something to cache all the space that is going to
9032 * be free'd up by the truncate operation, but also have some slack
9033 * space reserved in case it uses space during the truncate (thank you
9034 * very much snapshotting).
9035 *
9036 * And we need these to all be seperate. The fact is we can use alot of
9037 * space doing the truncate, and we have no earthly idea how much space
9038 * we will use, so we need the truncate reservation to be seperate so it
9039 * doesn't end up using space reserved for updating the inode or
9040 * removing the orphan item. We also need to be able to stop the
9041 * transaction and start a new one, which means we need to be able to
9042 * update the inode several times, and we have no idea of knowing how
9043 * many times that will be, so we can't just reserve 1 item for the
9044 * entirety of the opration, so that has to be done seperately as well.
9045 * Then there is the orphan item, which does indeed need to be held on
9046 * to for the whole operation, and we need nobody to touch this reserved
9047 * space except the orphan code.
9048 *
9049 * So that leaves us with
9050 *
9051 * 1) root->orphan_block_rsv - for the orphan deletion.
9052 * 2) rsv - for the truncate reservation, which we will steal from the
9053 * transaction reservation.
9054 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9055 * updating the inode.
9056 */
9057 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9058 if (!rsv)
9059 return -ENOMEM;
9060 rsv->size = min_size;
9061 rsv->failfast = 1;
9062
9063 /*
9064 * 1 for the truncate slack space
9065 * 1 for updating the inode.
9066 */
9067 trans = btrfs_start_transaction(root, 2);
9068 if (IS_ERR(trans)) {
9069 err = PTR_ERR(trans);
9070 goto out;
9071 }
9072
9073 /* Migrate the slack space for the truncate to our reserve */
9074 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9075 min_size);
9076 BUG_ON(ret);
9077
9078 /*
9079 * So if we truncate and then write and fsync we normally would just
9080 * write the extents that changed, which is a problem if we need to
9081 * first truncate that entire inode. So set this flag so we write out
9082 * all of the extents in the inode to the sync log so we're completely
9083 * safe.
9084 */
9085 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9086 trans->block_rsv = rsv;
9087
9088 while (1) {
9089 ret = btrfs_truncate_inode_items(trans, root, inode,
9090 inode->i_size,
9091 BTRFS_EXTENT_DATA_KEY);
9092 if (ret != -ENOSPC && ret != -EAGAIN) {
9093 err = ret;
9094 break;
9095 }
9096
9097 trans->block_rsv = &root->fs_info->trans_block_rsv;
9098 ret = btrfs_update_inode(trans, root, inode);
9099 if (ret) {
9100 err = ret;
9101 break;
9102 }
9103
9104 btrfs_end_transaction(trans, root);
9105 btrfs_btree_balance_dirty(root);
9106
9107 trans = btrfs_start_transaction(root, 2);
9108 if (IS_ERR(trans)) {
9109 ret = err = PTR_ERR(trans);
9110 trans = NULL;
9111 break;
9112 }
9113
9114 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9115 rsv, min_size);
9116 BUG_ON(ret); /* shouldn't happen */
9117 trans->block_rsv = rsv;
9118 }
9119
9120 if (ret == 0 && inode->i_nlink > 0) {
9121 trans->block_rsv = root->orphan_block_rsv;
9122 ret = btrfs_orphan_del(trans, inode);
9123 if (ret)
9124 err = ret;
9125 }
9126
9127 if (trans) {
9128 trans->block_rsv = &root->fs_info->trans_block_rsv;
9129 ret = btrfs_update_inode(trans, root, inode);
9130 if (ret && !err)
9131 err = ret;
9132
9133 ret = btrfs_end_transaction(trans, root);
9134 btrfs_btree_balance_dirty(root);
9135 }
9136
9137 out:
9138 btrfs_free_block_rsv(root, rsv);
9139
9140 if (ret && !err)
9141 err = ret;
9142
9143 return err;
9144 }
9145
9146 /*
9147 * create a new subvolume directory/inode (helper for the ioctl).
9148 */
9149 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9150 struct btrfs_root *new_root,
9151 struct btrfs_root *parent_root,
9152 u64 new_dirid)
9153 {
9154 struct inode *inode;
9155 int err;
9156 u64 index = 0;
9157
9158 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9159 new_dirid, new_dirid,
9160 S_IFDIR | (~current_umask() & S_IRWXUGO),
9161 &index);
9162 if (IS_ERR(inode))
9163 return PTR_ERR(inode);
9164 inode->i_op = &btrfs_dir_inode_operations;
9165 inode->i_fop = &btrfs_dir_file_operations;
9166
9167 set_nlink(inode, 1);
9168 btrfs_i_size_write(inode, 0);
9169 unlock_new_inode(inode);
9170
9171 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9172 if (err)
9173 btrfs_err(new_root->fs_info,
9174 "error inheriting subvolume %llu properties: %d",
9175 new_root->root_key.objectid, err);
9176
9177 err = btrfs_update_inode(trans, new_root, inode);
9178
9179 iput(inode);
9180 return err;
9181 }
9182
9183 struct inode *btrfs_alloc_inode(struct super_block *sb)
9184 {
9185 struct btrfs_inode *ei;
9186 struct inode *inode;
9187
9188 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9189 if (!ei)
9190 return NULL;
9191
9192 ei->root = NULL;
9193 ei->generation = 0;
9194 ei->last_trans = 0;
9195 ei->last_sub_trans = 0;
9196 ei->logged_trans = 0;
9197 ei->delalloc_bytes = 0;
9198 ei->defrag_bytes = 0;
9199 ei->disk_i_size = 0;
9200 ei->flags = 0;
9201 ei->csum_bytes = 0;
9202 ei->index_cnt = (u64)-1;
9203 ei->dir_index = 0;
9204 ei->last_unlink_trans = 0;
9205 ei->last_log_commit = 0;
9206 ei->delayed_iput_count = 0;
9207
9208 spin_lock_init(&ei->lock);
9209 ei->outstanding_extents = 0;
9210 ei->reserved_extents = 0;
9211
9212 ei->runtime_flags = 0;
9213 ei->force_compress = BTRFS_COMPRESS_NONE;
9214
9215 ei->delayed_node = NULL;
9216
9217 ei->i_otime.tv_sec = 0;
9218 ei->i_otime.tv_nsec = 0;
9219
9220 inode = &ei->vfs_inode;
9221 extent_map_tree_init(&ei->extent_tree);
9222 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9223 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9224 ei->io_tree.track_uptodate = 1;
9225 ei->io_failure_tree.track_uptodate = 1;
9226 atomic_set(&ei->sync_writers, 0);
9227 mutex_init(&ei->log_mutex);
9228 mutex_init(&ei->delalloc_mutex);
9229 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9230 INIT_LIST_HEAD(&ei->delalloc_inodes);
9231 INIT_LIST_HEAD(&ei->delayed_iput);
9232 RB_CLEAR_NODE(&ei->rb_node);
9233
9234 return inode;
9235 }
9236
9237 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9238 void btrfs_test_destroy_inode(struct inode *inode)
9239 {
9240 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9241 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9242 }
9243 #endif
9244
9245 static void btrfs_i_callback(struct rcu_head *head)
9246 {
9247 struct inode *inode = container_of(head, struct inode, i_rcu);
9248 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9249 }
9250
9251 void btrfs_destroy_inode(struct inode *inode)
9252 {
9253 struct btrfs_ordered_extent *ordered;
9254 struct btrfs_root *root = BTRFS_I(inode)->root;
9255
9256 WARN_ON(!hlist_empty(&inode->i_dentry));
9257 WARN_ON(inode->i_data.nrpages);
9258 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9259 WARN_ON(BTRFS_I(inode)->reserved_extents);
9260 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9261 WARN_ON(BTRFS_I(inode)->csum_bytes);
9262 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9263
9264 /*
9265 * This can happen where we create an inode, but somebody else also
9266 * created the same inode and we need to destroy the one we already
9267 * created.
9268 */
9269 if (!root)
9270 goto free;
9271
9272 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9273 &BTRFS_I(inode)->runtime_flags)) {
9274 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9275 btrfs_ino(inode));
9276 atomic_dec(&root->orphan_inodes);
9277 }
9278
9279 while (1) {
9280 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9281 if (!ordered)
9282 break;
9283 else {
9284 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9285 ordered->file_offset, ordered->len);
9286 btrfs_remove_ordered_extent(inode, ordered);
9287 btrfs_put_ordered_extent(ordered);
9288 btrfs_put_ordered_extent(ordered);
9289 }
9290 }
9291 btrfs_qgroup_check_reserved_leak(inode);
9292 inode_tree_del(inode);
9293 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9294 free:
9295 call_rcu(&inode->i_rcu, btrfs_i_callback);
9296 }
9297
9298 int btrfs_drop_inode(struct inode *inode)
9299 {
9300 struct btrfs_root *root = BTRFS_I(inode)->root;
9301
9302 if (root == NULL)
9303 return 1;
9304
9305 /* the snap/subvol tree is on deleting */
9306 if (btrfs_root_refs(&root->root_item) == 0)
9307 return 1;
9308 else
9309 return generic_drop_inode(inode);
9310 }
9311
9312 static void init_once(void *foo)
9313 {
9314 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9315
9316 inode_init_once(&ei->vfs_inode);
9317 }
9318
9319 void btrfs_destroy_cachep(void)
9320 {
9321 /*
9322 * Make sure all delayed rcu free inodes are flushed before we
9323 * destroy cache.
9324 */
9325 rcu_barrier();
9326 kmem_cache_destroy(btrfs_inode_cachep);
9327 kmem_cache_destroy(btrfs_trans_handle_cachep);
9328 kmem_cache_destroy(btrfs_transaction_cachep);
9329 kmem_cache_destroy(btrfs_path_cachep);
9330 kmem_cache_destroy(btrfs_free_space_cachep);
9331 }
9332
9333 int btrfs_init_cachep(void)
9334 {
9335 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9336 sizeof(struct btrfs_inode), 0,
9337 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9338 init_once);
9339 if (!btrfs_inode_cachep)
9340 goto fail;
9341
9342 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9343 sizeof(struct btrfs_trans_handle), 0,
9344 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9345 if (!btrfs_trans_handle_cachep)
9346 goto fail;
9347
9348 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9349 sizeof(struct btrfs_transaction), 0,
9350 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9351 if (!btrfs_transaction_cachep)
9352 goto fail;
9353
9354 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9355 sizeof(struct btrfs_path), 0,
9356 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9357 if (!btrfs_path_cachep)
9358 goto fail;
9359
9360 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9361 sizeof(struct btrfs_free_space), 0,
9362 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9363 if (!btrfs_free_space_cachep)
9364 goto fail;
9365
9366 return 0;
9367 fail:
9368 btrfs_destroy_cachep();
9369 return -ENOMEM;
9370 }
9371
9372 static int btrfs_getattr(struct vfsmount *mnt,
9373 struct dentry *dentry, struct kstat *stat)
9374 {
9375 u64 delalloc_bytes;
9376 struct inode *inode = d_inode(dentry);
9377 u32 blocksize = inode->i_sb->s_blocksize;
9378
9379 generic_fillattr(inode, stat);
9380 stat->dev = BTRFS_I(inode)->root->anon_dev;
9381
9382 spin_lock(&BTRFS_I(inode)->lock);
9383 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9384 spin_unlock(&BTRFS_I(inode)->lock);
9385 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9386 ALIGN(delalloc_bytes, blocksize)) >> 9;
9387 return 0;
9388 }
9389
9390 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9391 struct inode *new_dir, struct dentry *new_dentry)
9392 {
9393 struct btrfs_trans_handle *trans;
9394 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9395 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9396 struct inode *new_inode = d_inode(new_dentry);
9397 struct inode *old_inode = d_inode(old_dentry);
9398 u64 index = 0;
9399 u64 root_objectid;
9400 int ret;
9401 u64 old_ino = btrfs_ino(old_inode);
9402
9403 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9404 return -EPERM;
9405
9406 /* we only allow rename subvolume link between subvolumes */
9407 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9408 return -EXDEV;
9409
9410 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9411 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9412 return -ENOTEMPTY;
9413
9414 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9415 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9416 return -ENOTEMPTY;
9417
9418
9419 /* check for collisions, even if the name isn't there */
9420 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9421 new_dentry->d_name.name,
9422 new_dentry->d_name.len);
9423
9424 if (ret) {
9425 if (ret == -EEXIST) {
9426 /* we shouldn't get
9427 * eexist without a new_inode */
9428 if (WARN_ON(!new_inode)) {
9429 return ret;
9430 }
9431 } else {
9432 /* maybe -EOVERFLOW */
9433 return ret;
9434 }
9435 }
9436 ret = 0;
9437
9438 /*
9439 * we're using rename to replace one file with another. Start IO on it
9440 * now so we don't add too much work to the end of the transaction
9441 */
9442 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9443 filemap_flush(old_inode->i_mapping);
9444
9445 /* close the racy window with snapshot create/destroy ioctl */
9446 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9447 down_read(&root->fs_info->subvol_sem);
9448 /*
9449 * We want to reserve the absolute worst case amount of items. So if
9450 * both inodes are subvols and we need to unlink them then that would
9451 * require 4 item modifications, but if they are both normal inodes it
9452 * would require 5 item modifications, so we'll assume their normal
9453 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9454 * should cover the worst case number of items we'll modify.
9455 */
9456 trans = btrfs_start_transaction(root, 11);
9457 if (IS_ERR(trans)) {
9458 ret = PTR_ERR(trans);
9459 goto out_notrans;
9460 }
9461
9462 if (dest != root)
9463 btrfs_record_root_in_trans(trans, dest);
9464
9465 ret = btrfs_set_inode_index(new_dir, &index);
9466 if (ret)
9467 goto out_fail;
9468
9469 BTRFS_I(old_inode)->dir_index = 0ULL;
9470 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9471 /* force full log commit if subvolume involved. */
9472 btrfs_set_log_full_commit(root->fs_info, trans);
9473 } else {
9474 ret = btrfs_insert_inode_ref(trans, dest,
9475 new_dentry->d_name.name,
9476 new_dentry->d_name.len,
9477 old_ino,
9478 btrfs_ino(new_dir), index);
9479 if (ret)
9480 goto out_fail;
9481 /*
9482 * this is an ugly little race, but the rename is required
9483 * to make sure that if we crash, the inode is either at the
9484 * old name or the new one. pinning the log transaction lets
9485 * us make sure we don't allow a log commit to come in after
9486 * we unlink the name but before we add the new name back in.
9487 */
9488 btrfs_pin_log_trans(root);
9489 }
9490
9491 inode_inc_iversion(old_dir);
9492 inode_inc_iversion(new_dir);
9493 inode_inc_iversion(old_inode);
9494 old_dir->i_ctime = old_dir->i_mtime =
9495 new_dir->i_ctime = new_dir->i_mtime =
9496 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9497
9498 if (old_dentry->d_parent != new_dentry->d_parent)
9499 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9500
9501 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9502 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9503 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9504 old_dentry->d_name.name,
9505 old_dentry->d_name.len);
9506 } else {
9507 ret = __btrfs_unlink_inode(trans, root, old_dir,
9508 d_inode(old_dentry),
9509 old_dentry->d_name.name,
9510 old_dentry->d_name.len);
9511 if (!ret)
9512 ret = btrfs_update_inode(trans, root, old_inode);
9513 }
9514 if (ret) {
9515 btrfs_abort_transaction(trans, root, ret);
9516 goto out_fail;
9517 }
9518
9519 if (new_inode) {
9520 inode_inc_iversion(new_inode);
9521 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9522 if (unlikely(btrfs_ino(new_inode) ==
9523 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9524 root_objectid = BTRFS_I(new_inode)->location.objectid;
9525 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9526 root_objectid,
9527 new_dentry->d_name.name,
9528 new_dentry->d_name.len);
9529 BUG_ON(new_inode->i_nlink == 0);
9530 } else {
9531 ret = btrfs_unlink_inode(trans, dest, new_dir,
9532 d_inode(new_dentry),
9533 new_dentry->d_name.name,
9534 new_dentry->d_name.len);
9535 }
9536 if (!ret && new_inode->i_nlink == 0)
9537 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9538 if (ret) {
9539 btrfs_abort_transaction(trans, root, ret);
9540 goto out_fail;
9541 }
9542 }
9543
9544 ret = btrfs_add_link(trans, new_dir, old_inode,
9545 new_dentry->d_name.name,
9546 new_dentry->d_name.len, 0, index);
9547 if (ret) {
9548 btrfs_abort_transaction(trans, root, ret);
9549 goto out_fail;
9550 }
9551
9552 if (old_inode->i_nlink == 1)
9553 BTRFS_I(old_inode)->dir_index = index;
9554
9555 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9556 struct dentry *parent = new_dentry->d_parent;
9557 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9558 btrfs_end_log_trans(root);
9559 }
9560 out_fail:
9561 btrfs_end_transaction(trans, root);
9562 out_notrans:
9563 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9564 up_read(&root->fs_info->subvol_sem);
9565
9566 return ret;
9567 }
9568
9569 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9570 struct inode *new_dir, struct dentry *new_dentry,
9571 unsigned int flags)
9572 {
9573 if (flags & ~RENAME_NOREPLACE)
9574 return -EINVAL;
9575
9576 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9577 }
9578
9579 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9580 {
9581 struct btrfs_delalloc_work *delalloc_work;
9582 struct inode *inode;
9583
9584 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9585 work);
9586 inode = delalloc_work->inode;
9587 filemap_flush(inode->i_mapping);
9588 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9589 &BTRFS_I(inode)->runtime_flags))
9590 filemap_flush(inode->i_mapping);
9591
9592 if (delalloc_work->delay_iput)
9593 btrfs_add_delayed_iput(inode);
9594 else
9595 iput(inode);
9596 complete(&delalloc_work->completion);
9597 }
9598
9599 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9600 int delay_iput)
9601 {
9602 struct btrfs_delalloc_work *work;
9603
9604 work = kmalloc(sizeof(*work), GFP_NOFS);
9605 if (!work)
9606 return NULL;
9607
9608 init_completion(&work->completion);
9609 INIT_LIST_HEAD(&work->list);
9610 work->inode = inode;
9611 work->delay_iput = delay_iput;
9612 WARN_ON_ONCE(!inode);
9613 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9614 btrfs_run_delalloc_work, NULL, NULL);
9615
9616 return work;
9617 }
9618
9619 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9620 {
9621 wait_for_completion(&work->completion);
9622 kfree(work);
9623 }
9624
9625 /*
9626 * some fairly slow code that needs optimization. This walks the list
9627 * of all the inodes with pending delalloc and forces them to disk.
9628 */
9629 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9630 int nr)
9631 {
9632 struct btrfs_inode *binode;
9633 struct inode *inode;
9634 struct btrfs_delalloc_work *work, *next;
9635 struct list_head works;
9636 struct list_head splice;
9637 int ret = 0;
9638
9639 INIT_LIST_HEAD(&works);
9640 INIT_LIST_HEAD(&splice);
9641
9642 mutex_lock(&root->delalloc_mutex);
9643 spin_lock(&root->delalloc_lock);
9644 list_splice_init(&root->delalloc_inodes, &splice);
9645 while (!list_empty(&splice)) {
9646 binode = list_entry(splice.next, struct btrfs_inode,
9647 delalloc_inodes);
9648
9649 list_move_tail(&binode->delalloc_inodes,
9650 &root->delalloc_inodes);
9651 inode = igrab(&binode->vfs_inode);
9652 if (!inode) {
9653 cond_resched_lock(&root->delalloc_lock);
9654 continue;
9655 }
9656 spin_unlock(&root->delalloc_lock);
9657
9658 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9659 if (!work) {
9660 if (delay_iput)
9661 btrfs_add_delayed_iput(inode);
9662 else
9663 iput(inode);
9664 ret = -ENOMEM;
9665 goto out;
9666 }
9667 list_add_tail(&work->list, &works);
9668 btrfs_queue_work(root->fs_info->flush_workers,
9669 &work->work);
9670 ret++;
9671 if (nr != -1 && ret >= nr)
9672 goto out;
9673 cond_resched();
9674 spin_lock(&root->delalloc_lock);
9675 }
9676 spin_unlock(&root->delalloc_lock);
9677
9678 out:
9679 list_for_each_entry_safe(work, next, &works, list) {
9680 list_del_init(&work->list);
9681 btrfs_wait_and_free_delalloc_work(work);
9682 }
9683
9684 if (!list_empty_careful(&splice)) {
9685 spin_lock(&root->delalloc_lock);
9686 list_splice_tail(&splice, &root->delalloc_inodes);
9687 spin_unlock(&root->delalloc_lock);
9688 }
9689 mutex_unlock(&root->delalloc_mutex);
9690 return ret;
9691 }
9692
9693 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9694 {
9695 int ret;
9696
9697 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9698 return -EROFS;
9699
9700 ret = __start_delalloc_inodes(root, delay_iput, -1);
9701 if (ret > 0)
9702 ret = 0;
9703 /*
9704 * the filemap_flush will queue IO into the worker threads, but
9705 * we have to make sure the IO is actually started and that
9706 * ordered extents get created before we return
9707 */
9708 atomic_inc(&root->fs_info->async_submit_draining);
9709 while (atomic_read(&root->fs_info->nr_async_submits) ||
9710 atomic_read(&root->fs_info->async_delalloc_pages)) {
9711 wait_event(root->fs_info->async_submit_wait,
9712 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9713 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9714 }
9715 atomic_dec(&root->fs_info->async_submit_draining);
9716 return ret;
9717 }
9718
9719 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9720 int nr)
9721 {
9722 struct btrfs_root *root;
9723 struct list_head splice;
9724 int ret;
9725
9726 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9727 return -EROFS;
9728
9729 INIT_LIST_HEAD(&splice);
9730
9731 mutex_lock(&fs_info->delalloc_root_mutex);
9732 spin_lock(&fs_info->delalloc_root_lock);
9733 list_splice_init(&fs_info->delalloc_roots, &splice);
9734 while (!list_empty(&splice) && nr) {
9735 root = list_first_entry(&splice, struct btrfs_root,
9736 delalloc_root);
9737 root = btrfs_grab_fs_root(root);
9738 BUG_ON(!root);
9739 list_move_tail(&root->delalloc_root,
9740 &fs_info->delalloc_roots);
9741 spin_unlock(&fs_info->delalloc_root_lock);
9742
9743 ret = __start_delalloc_inodes(root, delay_iput, nr);
9744 btrfs_put_fs_root(root);
9745 if (ret < 0)
9746 goto out;
9747
9748 if (nr != -1) {
9749 nr -= ret;
9750 WARN_ON(nr < 0);
9751 }
9752 spin_lock(&fs_info->delalloc_root_lock);
9753 }
9754 spin_unlock(&fs_info->delalloc_root_lock);
9755
9756 ret = 0;
9757 atomic_inc(&fs_info->async_submit_draining);
9758 while (atomic_read(&fs_info->nr_async_submits) ||
9759 atomic_read(&fs_info->async_delalloc_pages)) {
9760 wait_event(fs_info->async_submit_wait,
9761 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9762 atomic_read(&fs_info->async_delalloc_pages) == 0));
9763 }
9764 atomic_dec(&fs_info->async_submit_draining);
9765 out:
9766 if (!list_empty_careful(&splice)) {
9767 spin_lock(&fs_info->delalloc_root_lock);
9768 list_splice_tail(&splice, &fs_info->delalloc_roots);
9769 spin_unlock(&fs_info->delalloc_root_lock);
9770 }
9771 mutex_unlock(&fs_info->delalloc_root_mutex);
9772 return ret;
9773 }
9774
9775 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9776 const char *symname)
9777 {
9778 struct btrfs_trans_handle *trans;
9779 struct btrfs_root *root = BTRFS_I(dir)->root;
9780 struct btrfs_path *path;
9781 struct btrfs_key key;
9782 struct inode *inode = NULL;
9783 int err;
9784 int drop_inode = 0;
9785 u64 objectid;
9786 u64 index = 0;
9787 int name_len;
9788 int datasize;
9789 unsigned long ptr;
9790 struct btrfs_file_extent_item *ei;
9791 struct extent_buffer *leaf;
9792
9793 name_len = strlen(symname);
9794 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9795 return -ENAMETOOLONG;
9796
9797 /*
9798 * 2 items for inode item and ref
9799 * 2 items for dir items
9800 * 1 item for updating parent inode item
9801 * 1 item for the inline extent item
9802 * 1 item for xattr if selinux is on
9803 */
9804 trans = btrfs_start_transaction(root, 7);
9805 if (IS_ERR(trans))
9806 return PTR_ERR(trans);
9807
9808 err = btrfs_find_free_ino(root, &objectid);
9809 if (err)
9810 goto out_unlock;
9811
9812 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9813 dentry->d_name.len, btrfs_ino(dir), objectid,
9814 S_IFLNK|S_IRWXUGO, &index);
9815 if (IS_ERR(inode)) {
9816 err = PTR_ERR(inode);
9817 goto out_unlock;
9818 }
9819
9820 /*
9821 * If the active LSM wants to access the inode during
9822 * d_instantiate it needs these. Smack checks to see
9823 * if the filesystem supports xattrs by looking at the
9824 * ops vector.
9825 */
9826 inode->i_fop = &btrfs_file_operations;
9827 inode->i_op = &btrfs_file_inode_operations;
9828 inode->i_mapping->a_ops = &btrfs_aops;
9829 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9830
9831 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9832 if (err)
9833 goto out_unlock_inode;
9834
9835 path = btrfs_alloc_path();
9836 if (!path) {
9837 err = -ENOMEM;
9838 goto out_unlock_inode;
9839 }
9840 key.objectid = btrfs_ino(inode);
9841 key.offset = 0;
9842 key.type = BTRFS_EXTENT_DATA_KEY;
9843 datasize = btrfs_file_extent_calc_inline_size(name_len);
9844 err = btrfs_insert_empty_item(trans, root, path, &key,
9845 datasize);
9846 if (err) {
9847 btrfs_free_path(path);
9848 goto out_unlock_inode;
9849 }
9850 leaf = path->nodes[0];
9851 ei = btrfs_item_ptr(leaf, path->slots[0],
9852 struct btrfs_file_extent_item);
9853 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9854 btrfs_set_file_extent_type(leaf, ei,
9855 BTRFS_FILE_EXTENT_INLINE);
9856 btrfs_set_file_extent_encryption(leaf, ei, 0);
9857 btrfs_set_file_extent_compression(leaf, ei, 0);
9858 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9859 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9860
9861 ptr = btrfs_file_extent_inline_start(ei);
9862 write_extent_buffer(leaf, symname, ptr, name_len);
9863 btrfs_mark_buffer_dirty(leaf);
9864 btrfs_free_path(path);
9865
9866 inode->i_op = &btrfs_symlink_inode_operations;
9867 inode_nohighmem(inode);
9868 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9869 inode_set_bytes(inode, name_len);
9870 btrfs_i_size_write(inode, name_len);
9871 err = btrfs_update_inode(trans, root, inode);
9872 /*
9873 * Last step, add directory indexes for our symlink inode. This is the
9874 * last step to avoid extra cleanup of these indexes if an error happens
9875 * elsewhere above.
9876 */
9877 if (!err)
9878 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9879 if (err) {
9880 drop_inode = 1;
9881 goto out_unlock_inode;
9882 }
9883
9884 unlock_new_inode(inode);
9885 d_instantiate(dentry, inode);
9886
9887 out_unlock:
9888 btrfs_end_transaction(trans, root);
9889 if (drop_inode) {
9890 inode_dec_link_count(inode);
9891 iput(inode);
9892 }
9893 btrfs_btree_balance_dirty(root);
9894 return err;
9895
9896 out_unlock_inode:
9897 drop_inode = 1;
9898 unlock_new_inode(inode);
9899 goto out_unlock;
9900 }
9901
9902 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9903 u64 start, u64 num_bytes, u64 min_size,
9904 loff_t actual_len, u64 *alloc_hint,
9905 struct btrfs_trans_handle *trans)
9906 {
9907 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9908 struct extent_map *em;
9909 struct btrfs_root *root = BTRFS_I(inode)->root;
9910 struct btrfs_key ins;
9911 u64 cur_offset = start;
9912 u64 i_size;
9913 u64 cur_bytes;
9914 u64 last_alloc = (u64)-1;
9915 int ret = 0;
9916 bool own_trans = true;
9917
9918 if (trans)
9919 own_trans = false;
9920 while (num_bytes > 0) {
9921 if (own_trans) {
9922 trans = btrfs_start_transaction(root, 3);
9923 if (IS_ERR(trans)) {
9924 ret = PTR_ERR(trans);
9925 break;
9926 }
9927 }
9928
9929 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9930 cur_bytes = max(cur_bytes, min_size);
9931 /*
9932 * If we are severely fragmented we could end up with really
9933 * small allocations, so if the allocator is returning small
9934 * chunks lets make its job easier by only searching for those
9935 * sized chunks.
9936 */
9937 cur_bytes = min(cur_bytes, last_alloc);
9938 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9939 *alloc_hint, &ins, 1, 0);
9940 if (ret) {
9941 if (own_trans)
9942 btrfs_end_transaction(trans, root);
9943 break;
9944 }
9945
9946 last_alloc = ins.offset;
9947 ret = insert_reserved_file_extent(trans, inode,
9948 cur_offset, ins.objectid,
9949 ins.offset, ins.offset,
9950 ins.offset, 0, 0, 0,
9951 BTRFS_FILE_EXTENT_PREALLOC);
9952 if (ret) {
9953 btrfs_free_reserved_extent(root, ins.objectid,
9954 ins.offset, 0);
9955 btrfs_abort_transaction(trans, root, ret);
9956 if (own_trans)
9957 btrfs_end_transaction(trans, root);
9958 break;
9959 }
9960
9961 btrfs_drop_extent_cache(inode, cur_offset,
9962 cur_offset + ins.offset -1, 0);
9963
9964 em = alloc_extent_map();
9965 if (!em) {
9966 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9967 &BTRFS_I(inode)->runtime_flags);
9968 goto next;
9969 }
9970
9971 em->start = cur_offset;
9972 em->orig_start = cur_offset;
9973 em->len = ins.offset;
9974 em->block_start = ins.objectid;
9975 em->block_len = ins.offset;
9976 em->orig_block_len = ins.offset;
9977 em->ram_bytes = ins.offset;
9978 em->bdev = root->fs_info->fs_devices->latest_bdev;
9979 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9980 em->generation = trans->transid;
9981
9982 while (1) {
9983 write_lock(&em_tree->lock);
9984 ret = add_extent_mapping(em_tree, em, 1);
9985 write_unlock(&em_tree->lock);
9986 if (ret != -EEXIST)
9987 break;
9988 btrfs_drop_extent_cache(inode, cur_offset,
9989 cur_offset + ins.offset - 1,
9990 0);
9991 }
9992 free_extent_map(em);
9993 next:
9994 num_bytes -= ins.offset;
9995 cur_offset += ins.offset;
9996 *alloc_hint = ins.objectid + ins.offset;
9997
9998 inode_inc_iversion(inode);
9999 inode->i_ctime = current_fs_time(inode->i_sb);
10000 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10001 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10002 (actual_len > inode->i_size) &&
10003 (cur_offset > inode->i_size)) {
10004 if (cur_offset > actual_len)
10005 i_size = actual_len;
10006 else
10007 i_size = cur_offset;
10008 i_size_write(inode, i_size);
10009 btrfs_ordered_update_i_size(inode, i_size, NULL);
10010 }
10011
10012 ret = btrfs_update_inode(trans, root, inode);
10013
10014 if (ret) {
10015 btrfs_abort_transaction(trans, root, ret);
10016 if (own_trans)
10017 btrfs_end_transaction(trans, root);
10018 break;
10019 }
10020
10021 if (own_trans)
10022 btrfs_end_transaction(trans, root);
10023 }
10024 return ret;
10025 }
10026
10027 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10028 u64 start, u64 num_bytes, u64 min_size,
10029 loff_t actual_len, u64 *alloc_hint)
10030 {
10031 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10032 min_size, actual_len, alloc_hint,
10033 NULL);
10034 }
10035
10036 int btrfs_prealloc_file_range_trans(struct inode *inode,
10037 struct btrfs_trans_handle *trans, int mode,
10038 u64 start, u64 num_bytes, u64 min_size,
10039 loff_t actual_len, u64 *alloc_hint)
10040 {
10041 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10042 min_size, actual_len, alloc_hint, trans);
10043 }
10044
10045 static int btrfs_set_page_dirty(struct page *page)
10046 {
10047 return __set_page_dirty_nobuffers(page);
10048 }
10049
10050 static int btrfs_permission(struct inode *inode, int mask)
10051 {
10052 struct btrfs_root *root = BTRFS_I(inode)->root;
10053 umode_t mode = inode->i_mode;
10054
10055 if (mask & MAY_WRITE &&
10056 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10057 if (btrfs_root_readonly(root))
10058 return -EROFS;
10059 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10060 return -EACCES;
10061 }
10062 return generic_permission(inode, mask);
10063 }
10064
10065 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10066 {
10067 struct btrfs_trans_handle *trans;
10068 struct btrfs_root *root = BTRFS_I(dir)->root;
10069 struct inode *inode = NULL;
10070 u64 objectid;
10071 u64 index;
10072 int ret = 0;
10073
10074 /*
10075 * 5 units required for adding orphan entry
10076 */
10077 trans = btrfs_start_transaction(root, 5);
10078 if (IS_ERR(trans))
10079 return PTR_ERR(trans);
10080
10081 ret = btrfs_find_free_ino(root, &objectid);
10082 if (ret)
10083 goto out;
10084
10085 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10086 btrfs_ino(dir), objectid, mode, &index);
10087 if (IS_ERR(inode)) {
10088 ret = PTR_ERR(inode);
10089 inode = NULL;
10090 goto out;
10091 }
10092
10093 inode->i_fop = &btrfs_file_operations;
10094 inode->i_op = &btrfs_file_inode_operations;
10095
10096 inode->i_mapping->a_ops = &btrfs_aops;
10097 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10098
10099 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10100 if (ret)
10101 goto out_inode;
10102
10103 ret = btrfs_update_inode(trans, root, inode);
10104 if (ret)
10105 goto out_inode;
10106 ret = btrfs_orphan_add(trans, inode);
10107 if (ret)
10108 goto out_inode;
10109
10110 /*
10111 * We set number of links to 0 in btrfs_new_inode(), and here we set
10112 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10113 * through:
10114 *
10115 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10116 */
10117 set_nlink(inode, 1);
10118 unlock_new_inode(inode);
10119 d_tmpfile(dentry, inode);
10120 mark_inode_dirty(inode);
10121
10122 out:
10123 btrfs_end_transaction(trans, root);
10124 if (ret)
10125 iput(inode);
10126 btrfs_balance_delayed_items(root);
10127 btrfs_btree_balance_dirty(root);
10128 return ret;
10129
10130 out_inode:
10131 unlock_new_inode(inode);
10132 goto out;
10133
10134 }
10135
10136 /* Inspired by filemap_check_errors() */
10137 int btrfs_inode_check_errors(struct inode *inode)
10138 {
10139 int ret = 0;
10140
10141 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10142 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10143 ret = -ENOSPC;
10144 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10145 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10146 ret = -EIO;
10147
10148 return ret;
10149 }
10150
10151 static const struct inode_operations btrfs_dir_inode_operations = {
10152 .getattr = btrfs_getattr,
10153 .lookup = btrfs_lookup,
10154 .create = btrfs_create,
10155 .unlink = btrfs_unlink,
10156 .link = btrfs_link,
10157 .mkdir = btrfs_mkdir,
10158 .rmdir = btrfs_rmdir,
10159 .rename2 = btrfs_rename2,
10160 .symlink = btrfs_symlink,
10161 .setattr = btrfs_setattr,
10162 .mknod = btrfs_mknod,
10163 .setxattr = btrfs_setxattr,
10164 .getxattr = generic_getxattr,
10165 .listxattr = btrfs_listxattr,
10166 .removexattr = btrfs_removexattr,
10167 .permission = btrfs_permission,
10168 .get_acl = btrfs_get_acl,
10169 .set_acl = btrfs_set_acl,
10170 .update_time = btrfs_update_time,
10171 .tmpfile = btrfs_tmpfile,
10172 };
10173 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10174 .lookup = btrfs_lookup,
10175 .permission = btrfs_permission,
10176 .get_acl = btrfs_get_acl,
10177 .set_acl = btrfs_set_acl,
10178 .update_time = btrfs_update_time,
10179 };
10180
10181 static const struct file_operations btrfs_dir_file_operations = {
10182 .llseek = generic_file_llseek,
10183 .read = generic_read_dir,
10184 .iterate = btrfs_real_readdir,
10185 .unlocked_ioctl = btrfs_ioctl,
10186 #ifdef CONFIG_COMPAT
10187 .compat_ioctl = btrfs_ioctl,
10188 #endif
10189 .release = btrfs_release_file,
10190 .fsync = btrfs_sync_file,
10191 };
10192
10193 static const struct extent_io_ops btrfs_extent_io_ops = {
10194 .fill_delalloc = run_delalloc_range,
10195 .submit_bio_hook = btrfs_submit_bio_hook,
10196 .merge_bio_hook = btrfs_merge_bio_hook,
10197 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10198 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10199 .writepage_start_hook = btrfs_writepage_start_hook,
10200 .set_bit_hook = btrfs_set_bit_hook,
10201 .clear_bit_hook = btrfs_clear_bit_hook,
10202 .merge_extent_hook = btrfs_merge_extent_hook,
10203 .split_extent_hook = btrfs_split_extent_hook,
10204 };
10205
10206 /*
10207 * btrfs doesn't support the bmap operation because swapfiles
10208 * use bmap to make a mapping of extents in the file. They assume
10209 * these extents won't change over the life of the file and they
10210 * use the bmap result to do IO directly to the drive.
10211 *
10212 * the btrfs bmap call would return logical addresses that aren't
10213 * suitable for IO and they also will change frequently as COW
10214 * operations happen. So, swapfile + btrfs == corruption.
10215 *
10216 * For now we're avoiding this by dropping bmap.
10217 */
10218 static const struct address_space_operations btrfs_aops = {
10219 .readpage = btrfs_readpage,
10220 .writepage = btrfs_writepage,
10221 .writepages = btrfs_writepages,
10222 .readpages = btrfs_readpages,
10223 .direct_IO = btrfs_direct_IO,
10224 .invalidatepage = btrfs_invalidatepage,
10225 .releasepage = btrfs_releasepage,
10226 .set_page_dirty = btrfs_set_page_dirty,
10227 .error_remove_page = generic_error_remove_page,
10228 };
10229
10230 static const struct address_space_operations btrfs_symlink_aops = {
10231 .readpage = btrfs_readpage,
10232 .writepage = btrfs_writepage,
10233 .invalidatepage = btrfs_invalidatepage,
10234 .releasepage = btrfs_releasepage,
10235 };
10236
10237 static const struct inode_operations btrfs_file_inode_operations = {
10238 .getattr = btrfs_getattr,
10239 .setattr = btrfs_setattr,
10240 .setxattr = btrfs_setxattr,
10241 .getxattr = generic_getxattr,
10242 .listxattr = btrfs_listxattr,
10243 .removexattr = btrfs_removexattr,
10244 .permission = btrfs_permission,
10245 .fiemap = btrfs_fiemap,
10246 .get_acl = btrfs_get_acl,
10247 .set_acl = btrfs_set_acl,
10248 .update_time = btrfs_update_time,
10249 };
10250 static const struct inode_operations btrfs_special_inode_operations = {
10251 .getattr = btrfs_getattr,
10252 .setattr = btrfs_setattr,
10253 .permission = btrfs_permission,
10254 .setxattr = btrfs_setxattr,
10255 .getxattr = generic_getxattr,
10256 .listxattr = btrfs_listxattr,
10257 .removexattr = btrfs_removexattr,
10258 .get_acl = btrfs_get_acl,
10259 .set_acl = btrfs_set_acl,
10260 .update_time = btrfs_update_time,
10261 };
10262 static const struct inode_operations btrfs_symlink_inode_operations = {
10263 .readlink = generic_readlink,
10264 .get_link = page_get_link,
10265 .getattr = btrfs_getattr,
10266 .setattr = btrfs_setattr,
10267 .permission = btrfs_permission,
10268 .setxattr = btrfs_setxattr,
10269 .getxattr = generic_getxattr,
10270 .listxattr = btrfs_listxattr,
10271 .removexattr = btrfs_removexattr,
10272 .update_time = btrfs_update_time,
10273 };
10274
10275 const struct dentry_operations btrfs_dentry_operations = {
10276 .d_delete = btrfs_dentry_delete,
10277 .d_release = btrfs_dentry_release,
10278 };
Ubuntu Artful kernel mirror