projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
orangefs: sanitize ->llseek()
[deliverable/linux.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/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_CACHE_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_CACHE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_CACHE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
217 page_cache_release(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 > PAGE_CACHE_SIZE ||
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_CACHE_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_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_CACHE_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_CACHE_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_CACHE_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_CACHE_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 page_cache_release(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 page_cache_release(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 page_cache_release(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_CACHE_SIZE) / PAGE_CACHE_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_CACHE_SIZE) >>
1110 PAGE_CACHE_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_CACHE_SIZE) >>
1168 PAGE_CACHE_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_CACHE_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_CACHE_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_extent(inode, page_start);
2006 if (ordered) {
2007 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2008 page_end, &cached_state, GFP_NOFS);
2009 unlock_page(page);
2010 btrfs_start_ordered_extent(inode, ordered, 1);
2011 btrfs_put_ordered_extent(ordered);
2012 goto again;
2013 }
2014
2015 ret = btrfs_delalloc_reserve_space(inode, page_start,
2016 PAGE_CACHE_SIZE);
2017 if (ret) {
2018 mapping_set_error(page->mapping, ret);
2019 end_extent_writepage(page, ret, page_start, page_end);
2020 ClearPageChecked(page);
2021 goto out;
2022 }
2023
2024 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2025 ClearPageChecked(page);
2026 set_page_dirty(page);
2027 out:
2028 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2029 &cached_state, GFP_NOFS);
2030 out_page:
2031 unlock_page(page);
2032 page_cache_release(page);
2033 kfree(fixup);
2034 }
2035
2036 /*
2037 * There are a few paths in the higher layers of the kernel that directly
2038 * set the page dirty bit without asking the filesystem if it is a
2039 * good idea. This causes problems because we want to make sure COW
2040 * properly happens and the data=ordered rules are followed.
2041 *
2042 * In our case any range that doesn't have the ORDERED bit set
2043 * hasn't been properly setup for IO. We kick off an async process
2044 * to fix it up. The async helper will wait for ordered extents, set
2045 * the delalloc bit and make it safe to write the page.
2046 */
2047 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2048 {
2049 struct inode *inode = page->mapping->host;
2050 struct btrfs_writepage_fixup *fixup;
2051 struct btrfs_root *root = BTRFS_I(inode)->root;
2052
2053 /* this page is properly in the ordered list */
2054 if (TestClearPagePrivate2(page))
2055 return 0;
2056
2057 if (PageChecked(page))
2058 return -EAGAIN;
2059
2060 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2061 if (!fixup)
2062 return -EAGAIN;
2063
2064 SetPageChecked(page);
2065 page_cache_get(page);
2066 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2067 btrfs_writepage_fixup_worker, NULL, NULL);
2068 fixup->page = page;
2069 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2070 return -EBUSY;
2071 }
2072
2073 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2074 struct inode *inode, u64 file_pos,
2075 u64 disk_bytenr, u64 disk_num_bytes,
2076 u64 num_bytes, u64 ram_bytes,
2077 u8 compression, u8 encryption,
2078 u16 other_encoding, int extent_type)
2079 {
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
2081 struct btrfs_file_extent_item *fi;
2082 struct btrfs_path *path;
2083 struct extent_buffer *leaf;
2084 struct btrfs_key ins;
2085 int extent_inserted = 0;
2086 int ret;
2087
2088 path = btrfs_alloc_path();
2089 if (!path)
2090 return -ENOMEM;
2091
2092 /*
2093 * we may be replacing one extent in the tree with another.
2094 * The new extent is pinned in the extent map, and we don't want
2095 * to drop it from the cache until it is completely in the btree.
2096 *
2097 * So, tell btrfs_drop_extents to leave this extent in the cache.
2098 * the caller is expected to unpin it and allow it to be merged
2099 * with the others.
2100 */
2101 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2102 file_pos + num_bytes, NULL, 0,
2103 1, sizeof(*fi), &extent_inserted);
2104 if (ret)
2105 goto out;
2106
2107 if (!extent_inserted) {
2108 ins.objectid = btrfs_ino(inode);
2109 ins.offset = file_pos;
2110 ins.type = BTRFS_EXTENT_DATA_KEY;
2111
2112 path->leave_spinning = 1;
2113 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2114 sizeof(*fi));
2115 if (ret)
2116 goto out;
2117 }
2118 leaf = path->nodes[0];
2119 fi = btrfs_item_ptr(leaf, path->slots[0],
2120 struct btrfs_file_extent_item);
2121 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2122 btrfs_set_file_extent_type(leaf, fi, extent_type);
2123 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2124 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2125 btrfs_set_file_extent_offset(leaf, fi, 0);
2126 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2127 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2128 btrfs_set_file_extent_compression(leaf, fi, compression);
2129 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2130 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2131
2132 btrfs_mark_buffer_dirty(leaf);
2133 btrfs_release_path(path);
2134
2135 inode_add_bytes(inode, num_bytes);
2136
2137 ins.objectid = disk_bytenr;
2138 ins.offset = disk_num_bytes;
2139 ins.type = BTRFS_EXTENT_ITEM_KEY;
2140 ret = btrfs_alloc_reserved_file_extent(trans, root,
2141 root->root_key.objectid,
2142 btrfs_ino(inode), file_pos,
2143 ram_bytes, &ins);
2144 /*
2145 * Release the reserved range from inode dirty range map, as it is
2146 * already moved into delayed_ref_head
2147 */
2148 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2149 out:
2150 btrfs_free_path(path);
2151
2152 return ret;
2153 }
2154
2155 /* snapshot-aware defrag */
2156 struct sa_defrag_extent_backref {
2157 struct rb_node node;
2158 struct old_sa_defrag_extent *old;
2159 u64 root_id;
2160 u64 inum;
2161 u64 file_pos;
2162 u64 extent_offset;
2163 u64 num_bytes;
2164 u64 generation;
2165 };
2166
2167 struct old_sa_defrag_extent {
2168 struct list_head list;
2169 struct new_sa_defrag_extent *new;
2170
2171 u64 extent_offset;
2172 u64 bytenr;
2173 u64 offset;
2174 u64 len;
2175 int count;
2176 };
2177
2178 struct new_sa_defrag_extent {
2179 struct rb_root root;
2180 struct list_head head;
2181 struct btrfs_path *path;
2182 struct inode *inode;
2183 u64 file_pos;
2184 u64 len;
2185 u64 bytenr;
2186 u64 disk_len;
2187 u8 compress_type;
2188 };
2189
2190 static int backref_comp(struct sa_defrag_extent_backref *b1,
2191 struct sa_defrag_extent_backref *b2)
2192 {
2193 if (b1->root_id < b2->root_id)
2194 return -1;
2195 else if (b1->root_id > b2->root_id)
2196 return 1;
2197
2198 if (b1->inum < b2->inum)
2199 return -1;
2200 else if (b1->inum > b2->inum)
2201 return 1;
2202
2203 if (b1->file_pos < b2->file_pos)
2204 return -1;
2205 else if (b1->file_pos > b2->file_pos)
2206 return 1;
2207
2208 /*
2209 * [------------------------------] ===> (a range of space)
2210 * |<--->| |<---->| =============> (fs/file tree A)
2211 * |<---------------------------->| ===> (fs/file tree B)
2212 *
2213 * A range of space can refer to two file extents in one tree while
2214 * refer to only one file extent in another tree.
2215 *
2216 * So we may process a disk offset more than one time(two extents in A)
2217 * and locate at the same extent(one extent in B), then insert two same
2218 * backrefs(both refer to the extent in B).
2219 */
2220 return 0;
2221 }
2222
2223 static void backref_insert(struct rb_root *root,
2224 struct sa_defrag_extent_backref *backref)
2225 {
2226 struct rb_node **p = &root->rb_node;
2227 struct rb_node *parent = NULL;
2228 struct sa_defrag_extent_backref *entry;
2229 int ret;
2230
2231 while (*p) {
2232 parent = *p;
2233 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2234
2235 ret = backref_comp(backref, entry);
2236 if (ret < 0)
2237 p = &(*p)->rb_left;
2238 else
2239 p = &(*p)->rb_right;
2240 }
2241
2242 rb_link_node(&backref->node, parent, p);
2243 rb_insert_color(&backref->node, root);
2244 }
2245
2246 /*
2247 * Note the backref might has changed, and in this case we just return 0.
2248 */
2249 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2250 void *ctx)
2251 {
2252 struct btrfs_file_extent_item *extent;
2253 struct btrfs_fs_info *fs_info;
2254 struct old_sa_defrag_extent *old = ctx;
2255 struct new_sa_defrag_extent *new = old->new;
2256 struct btrfs_path *path = new->path;
2257 struct btrfs_key key;
2258 struct btrfs_root *root;
2259 struct sa_defrag_extent_backref *backref;
2260 struct extent_buffer *leaf;
2261 struct inode *inode = new->inode;
2262 int slot;
2263 int ret;
2264 u64 extent_offset;
2265 u64 num_bytes;
2266
2267 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2268 inum == btrfs_ino(inode))
2269 return 0;
2270
2271 key.objectid = root_id;
2272 key.type = BTRFS_ROOT_ITEM_KEY;
2273 key.offset = (u64)-1;
2274
2275 fs_info = BTRFS_I(inode)->root->fs_info;
2276 root = btrfs_read_fs_root_no_name(fs_info, &key);
2277 if (IS_ERR(root)) {
2278 if (PTR_ERR(root) == -ENOENT)
2279 return 0;
2280 WARN_ON(1);
2281 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2282 inum, offset, root_id);
2283 return PTR_ERR(root);
2284 }
2285
2286 key.objectid = inum;
2287 key.type = BTRFS_EXTENT_DATA_KEY;
2288 if (offset > (u64)-1 << 32)
2289 key.offset = 0;
2290 else
2291 key.offset = offset;
2292
2293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2294 if (WARN_ON(ret < 0))
2295 return ret;
2296 ret = 0;
2297
2298 while (1) {
2299 cond_resched();
2300
2301 leaf = path->nodes[0];
2302 slot = path->slots[0];
2303
2304 if (slot >= btrfs_header_nritems(leaf)) {
2305 ret = btrfs_next_leaf(root, path);
2306 if (ret < 0) {
2307 goto out;
2308 } else if (ret > 0) {
2309 ret = 0;
2310 goto out;
2311 }
2312 continue;
2313 }
2314
2315 path->slots[0]++;
2316
2317 btrfs_item_key_to_cpu(leaf, &key, slot);
2318
2319 if (key.objectid > inum)
2320 goto out;
2321
2322 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2323 continue;
2324
2325 extent = btrfs_item_ptr(leaf, slot,
2326 struct btrfs_file_extent_item);
2327
2328 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2329 continue;
2330
2331 /*
2332 * 'offset' refers to the exact key.offset,
2333 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2334 * (key.offset - extent_offset).
2335 */
2336 if (key.offset != offset)
2337 continue;
2338
2339 extent_offset = btrfs_file_extent_offset(leaf, extent);
2340 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2341
2342 if (extent_offset >= old->extent_offset + old->offset +
2343 old->len || extent_offset + num_bytes <=
2344 old->extent_offset + old->offset)
2345 continue;
2346 break;
2347 }
2348
2349 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2350 if (!backref) {
2351 ret = -ENOENT;
2352 goto out;
2353 }
2354
2355 backref->root_id = root_id;
2356 backref->inum = inum;
2357 backref->file_pos = offset;
2358 backref->num_bytes = num_bytes;
2359 backref->extent_offset = extent_offset;
2360 backref->generation = btrfs_file_extent_generation(leaf, extent);
2361 backref->old = old;
2362 backref_insert(&new->root, backref);
2363 old->count++;
2364 out:
2365 btrfs_release_path(path);
2366 WARN_ON(ret);
2367 return ret;
2368 }
2369
2370 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2371 struct new_sa_defrag_extent *new)
2372 {
2373 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2374 struct old_sa_defrag_extent *old, *tmp;
2375 int ret;
2376
2377 new->path = path;
2378
2379 list_for_each_entry_safe(old, tmp, &new->head, list) {
2380 ret = iterate_inodes_from_logical(old->bytenr +
2381 old->extent_offset, fs_info,
2382 path, record_one_backref,
2383 old);
2384 if (ret < 0 && ret != -ENOENT)
2385 return false;
2386
2387 /* no backref to be processed for this extent */
2388 if (!old->count) {
2389 list_del(&old->list);
2390 kfree(old);
2391 }
2392 }
2393
2394 if (list_empty(&new->head))
2395 return false;
2396
2397 return true;
2398 }
2399
2400 static int relink_is_mergable(struct extent_buffer *leaf,
2401 struct btrfs_file_extent_item *fi,
2402 struct new_sa_defrag_extent *new)
2403 {
2404 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2405 return 0;
2406
2407 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2408 return 0;
2409
2410 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2411 return 0;
2412
2413 if (btrfs_file_extent_encryption(leaf, fi) ||
2414 btrfs_file_extent_other_encoding(leaf, fi))
2415 return 0;
2416
2417 return 1;
2418 }
2419
2420 /*
2421 * Note the backref might has changed, and in this case we just return 0.
2422 */
2423 static noinline int relink_extent_backref(struct btrfs_path *path,
2424 struct sa_defrag_extent_backref *prev,
2425 struct sa_defrag_extent_backref *backref)
2426 {
2427 struct btrfs_file_extent_item *extent;
2428 struct btrfs_file_extent_item *item;
2429 struct btrfs_ordered_extent *ordered;
2430 struct btrfs_trans_handle *trans;
2431 struct btrfs_fs_info *fs_info;
2432 struct btrfs_root *root;
2433 struct btrfs_key key;
2434 struct extent_buffer *leaf;
2435 struct old_sa_defrag_extent *old = backref->old;
2436 struct new_sa_defrag_extent *new = old->new;
2437 struct inode *src_inode = new->inode;
2438 struct inode *inode;
2439 struct extent_state *cached = NULL;
2440 int ret = 0;
2441 u64 start;
2442 u64 len;
2443 u64 lock_start;
2444 u64 lock_end;
2445 bool merge = false;
2446 int index;
2447
2448 if (prev && prev->root_id == backref->root_id &&
2449 prev->inum == backref->inum &&
2450 prev->file_pos + prev->num_bytes == backref->file_pos)
2451 merge = true;
2452
2453 /* step 1: get root */
2454 key.objectid = backref->root_id;
2455 key.type = BTRFS_ROOT_ITEM_KEY;
2456 key.offset = (u64)-1;
2457
2458 fs_info = BTRFS_I(src_inode)->root->fs_info;
2459 index = srcu_read_lock(&fs_info->subvol_srcu);
2460
2461 root = btrfs_read_fs_root_no_name(fs_info, &key);
2462 if (IS_ERR(root)) {
2463 srcu_read_unlock(&fs_info->subvol_srcu, index);
2464 if (PTR_ERR(root) == -ENOENT)
2465 return 0;
2466 return PTR_ERR(root);
2467 }
2468
2469 if (btrfs_root_readonly(root)) {
2470 srcu_read_unlock(&fs_info->subvol_srcu, index);
2471 return 0;
2472 }
2473
2474 /* step 2: get inode */
2475 key.objectid = backref->inum;
2476 key.type = BTRFS_INODE_ITEM_KEY;
2477 key.offset = 0;
2478
2479 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2480 if (IS_ERR(inode)) {
2481 srcu_read_unlock(&fs_info->subvol_srcu, index);
2482 return 0;
2483 }
2484
2485 srcu_read_unlock(&fs_info->subvol_srcu, index);
2486
2487 /* step 3: relink backref */
2488 lock_start = backref->file_pos;
2489 lock_end = backref->file_pos + backref->num_bytes - 1;
2490 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2491 &cached);
2492
2493 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2494 if (ordered) {
2495 btrfs_put_ordered_extent(ordered);
2496 goto out_unlock;
2497 }
2498
2499 trans = btrfs_join_transaction(root);
2500 if (IS_ERR(trans)) {
2501 ret = PTR_ERR(trans);
2502 goto out_unlock;
2503 }
2504
2505 key.objectid = backref->inum;
2506 key.type = BTRFS_EXTENT_DATA_KEY;
2507 key.offset = backref->file_pos;
2508
2509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510 if (ret < 0) {
2511 goto out_free_path;
2512 } else if (ret > 0) {
2513 ret = 0;
2514 goto out_free_path;
2515 }
2516
2517 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2518 struct btrfs_file_extent_item);
2519
2520 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2521 backref->generation)
2522 goto out_free_path;
2523
2524 btrfs_release_path(path);
2525
2526 start = backref->file_pos;
2527 if (backref->extent_offset < old->extent_offset + old->offset)
2528 start += old->extent_offset + old->offset -
2529 backref->extent_offset;
2530
2531 len = min(backref->extent_offset + backref->num_bytes,
2532 old->extent_offset + old->offset + old->len);
2533 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2534
2535 ret = btrfs_drop_extents(trans, root, inode, start,
2536 start + len, 1);
2537 if (ret)
2538 goto out_free_path;
2539 again:
2540 key.objectid = btrfs_ino(inode);
2541 key.type = BTRFS_EXTENT_DATA_KEY;
2542 key.offset = start;
2543
2544 path->leave_spinning = 1;
2545 if (merge) {
2546 struct btrfs_file_extent_item *fi;
2547 u64 extent_len;
2548 struct btrfs_key found_key;
2549
2550 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2551 if (ret < 0)
2552 goto out_free_path;
2553
2554 path->slots[0]--;
2555 leaf = path->nodes[0];
2556 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2557
2558 fi = btrfs_item_ptr(leaf, path->slots[0],
2559 struct btrfs_file_extent_item);
2560 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2561
2562 if (extent_len + found_key.offset == start &&
2563 relink_is_mergable(leaf, fi, new)) {
2564 btrfs_set_file_extent_num_bytes(leaf, fi,
2565 extent_len + len);
2566 btrfs_mark_buffer_dirty(leaf);
2567 inode_add_bytes(inode, len);
2568
2569 ret = 1;
2570 goto out_free_path;
2571 } else {
2572 merge = false;
2573 btrfs_release_path(path);
2574 goto again;
2575 }
2576 }
2577
2578 ret = btrfs_insert_empty_item(trans, root, path, &key,
2579 sizeof(*extent));
2580 if (ret) {
2581 btrfs_abort_transaction(trans, root, ret);
2582 goto out_free_path;
2583 }
2584
2585 leaf = path->nodes[0];
2586 item = btrfs_item_ptr(leaf, path->slots[0],
2587 struct btrfs_file_extent_item);
2588 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2589 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2590 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2591 btrfs_set_file_extent_num_bytes(leaf, item, len);
2592 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2593 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2594 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2595 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2596 btrfs_set_file_extent_encryption(leaf, item, 0);
2597 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2598
2599 btrfs_mark_buffer_dirty(leaf);
2600 inode_add_bytes(inode, len);
2601 btrfs_release_path(path);
2602
2603 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2604 new->disk_len, 0,
2605 backref->root_id, backref->inum,
2606 new->file_pos); /* start - extent_offset */
2607 if (ret) {
2608 btrfs_abort_transaction(trans, root, ret);
2609 goto out_free_path;
2610 }
2611
2612 ret = 1;
2613 out_free_path:
2614 btrfs_release_path(path);
2615 path->leave_spinning = 0;
2616 btrfs_end_transaction(trans, root);
2617 out_unlock:
2618 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2619 &cached, GFP_NOFS);
2620 iput(inode);
2621 return ret;
2622 }
2623
2624 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2625 {
2626 struct old_sa_defrag_extent *old, *tmp;
2627
2628 if (!new)
2629 return;
2630
2631 list_for_each_entry_safe(old, tmp, &new->head, list) {
2632 kfree(old);
2633 }
2634 kfree(new);
2635 }
2636
2637 static void relink_file_extents(struct new_sa_defrag_extent *new)
2638 {
2639 struct btrfs_path *path;
2640 struct sa_defrag_extent_backref *backref;
2641 struct sa_defrag_extent_backref *prev = NULL;
2642 struct inode *inode;
2643 struct btrfs_root *root;
2644 struct rb_node *node;
2645 int ret;
2646
2647 inode = new->inode;
2648 root = BTRFS_I(inode)->root;
2649
2650 path = btrfs_alloc_path();
2651 if (!path)
2652 return;
2653
2654 if (!record_extent_backrefs(path, new)) {
2655 btrfs_free_path(path);
2656 goto out;
2657 }
2658 btrfs_release_path(path);
2659
2660 while (1) {
2661 node = rb_first(&new->root);
2662 if (!node)
2663 break;
2664 rb_erase(node, &new->root);
2665
2666 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2667
2668 ret = relink_extent_backref(path, prev, backref);
2669 WARN_ON(ret < 0);
2670
2671 kfree(prev);
2672
2673 if (ret == 1)
2674 prev = backref;
2675 else
2676 prev = NULL;
2677 cond_resched();
2678 }
2679 kfree(prev);
2680
2681 btrfs_free_path(path);
2682 out:
2683 free_sa_defrag_extent(new);
2684
2685 atomic_dec(&root->fs_info->defrag_running);
2686 wake_up(&root->fs_info->transaction_wait);
2687 }
2688
2689 static struct new_sa_defrag_extent *
2690 record_old_file_extents(struct inode *inode,
2691 struct btrfs_ordered_extent *ordered)
2692 {
2693 struct btrfs_root *root = BTRFS_I(inode)->root;
2694 struct btrfs_path *path;
2695 struct btrfs_key key;
2696 struct old_sa_defrag_extent *old;
2697 struct new_sa_defrag_extent *new;
2698 int ret;
2699
2700 new = kmalloc(sizeof(*new), GFP_NOFS);
2701 if (!new)
2702 return NULL;
2703
2704 new->inode = inode;
2705 new->file_pos = ordered->file_offset;
2706 new->len = ordered->len;
2707 new->bytenr = ordered->start;
2708 new->disk_len = ordered->disk_len;
2709 new->compress_type = ordered->compress_type;
2710 new->root = RB_ROOT;
2711 INIT_LIST_HEAD(&new->head);
2712
2713 path = btrfs_alloc_path();
2714 if (!path)
2715 goto out_kfree;
2716
2717 key.objectid = btrfs_ino(inode);
2718 key.type = BTRFS_EXTENT_DATA_KEY;
2719 key.offset = new->file_pos;
2720
2721 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2722 if (ret < 0)
2723 goto out_free_path;
2724 if (ret > 0 && path->slots[0] > 0)
2725 path->slots[0]--;
2726
2727 /* find out all the old extents for the file range */
2728 while (1) {
2729 struct btrfs_file_extent_item *extent;
2730 struct extent_buffer *l;
2731 int slot;
2732 u64 num_bytes;
2733 u64 offset;
2734 u64 end;
2735 u64 disk_bytenr;
2736 u64 extent_offset;
2737
2738 l = path->nodes[0];
2739 slot = path->slots[0];
2740
2741 if (slot >= btrfs_header_nritems(l)) {
2742 ret = btrfs_next_leaf(root, path);
2743 if (ret < 0)
2744 goto out_free_path;
2745 else if (ret > 0)
2746 break;
2747 continue;
2748 }
2749
2750 btrfs_item_key_to_cpu(l, &key, slot);
2751
2752 if (key.objectid != btrfs_ino(inode))
2753 break;
2754 if (key.type != BTRFS_EXTENT_DATA_KEY)
2755 break;
2756 if (key.offset >= new->file_pos + new->len)
2757 break;
2758
2759 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2760
2761 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2762 if (key.offset + num_bytes < new->file_pos)
2763 goto next;
2764
2765 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2766 if (!disk_bytenr)
2767 goto next;
2768
2769 extent_offset = btrfs_file_extent_offset(l, extent);
2770
2771 old = kmalloc(sizeof(*old), GFP_NOFS);
2772 if (!old)
2773 goto out_free_path;
2774
2775 offset = max(new->file_pos, key.offset);
2776 end = min(new->file_pos + new->len, key.offset + num_bytes);
2777
2778 old->bytenr = disk_bytenr;
2779 old->extent_offset = extent_offset;
2780 old->offset = offset - key.offset;
2781 old->len = end - offset;
2782 old->new = new;
2783 old->count = 0;
2784 list_add_tail(&old->list, &new->head);
2785 next:
2786 path->slots[0]++;
2787 cond_resched();
2788 }
2789
2790 btrfs_free_path(path);
2791 atomic_inc(&root->fs_info->defrag_running);
2792
2793 return new;
2794
2795 out_free_path:
2796 btrfs_free_path(path);
2797 out_kfree:
2798 free_sa_defrag_extent(new);
2799 return NULL;
2800 }
2801
2802 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2803 u64 start, u64 len)
2804 {
2805 struct btrfs_block_group_cache *cache;
2806
2807 cache = btrfs_lookup_block_group(root->fs_info, start);
2808 ASSERT(cache);
2809
2810 spin_lock(&cache->lock);
2811 cache->delalloc_bytes -= len;
2812 spin_unlock(&cache->lock);
2813
2814 btrfs_put_block_group(cache);
2815 }
2816
2817 /* as ordered data IO finishes, this gets called so we can finish
2818 * an ordered extent if the range of bytes in the file it covers are
2819 * fully written.
2820 */
2821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2822 {
2823 struct inode *inode = ordered_extent->inode;
2824 struct btrfs_root *root = BTRFS_I(inode)->root;
2825 struct btrfs_trans_handle *trans = NULL;
2826 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2827 struct extent_state *cached_state = NULL;
2828 struct new_sa_defrag_extent *new = NULL;
2829 int compress_type = 0;
2830 int ret = 0;
2831 u64 logical_len = ordered_extent->len;
2832 bool nolock;
2833 bool truncated = false;
2834
2835 nolock = btrfs_is_free_space_inode(inode);
2836
2837 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2838 ret = -EIO;
2839 goto out;
2840 }
2841
2842 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2843 ordered_extent->file_offset +
2844 ordered_extent->len - 1);
2845
2846 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2847 truncated = true;
2848 logical_len = ordered_extent->truncated_len;
2849 /* Truncated the entire extent, don't bother adding */
2850 if (!logical_len)
2851 goto out;
2852 }
2853
2854 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2855 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2856
2857 /*
2858 * For mwrite(mmap + memset to write) case, we still reserve
2859 * space for NOCOW range.
2860 * As NOCOW won't cause a new delayed ref, just free the space
2861 */
2862 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2863 ordered_extent->len);
2864 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2865 if (nolock)
2866 trans = btrfs_join_transaction_nolock(root);
2867 else
2868 trans = btrfs_join_transaction(root);
2869 if (IS_ERR(trans)) {
2870 ret = PTR_ERR(trans);
2871 trans = NULL;
2872 goto out;
2873 }
2874 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2875 ret = btrfs_update_inode_fallback(trans, root, inode);
2876 if (ret) /* -ENOMEM or corruption */
2877 btrfs_abort_transaction(trans, root, ret);
2878 goto out;
2879 }
2880
2881 lock_extent_bits(io_tree, ordered_extent->file_offset,
2882 ordered_extent->file_offset + ordered_extent->len - 1,
2883 &cached_state);
2884
2885 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2886 ordered_extent->file_offset + ordered_extent->len - 1,
2887 EXTENT_DEFRAG, 1, cached_state);
2888 if (ret) {
2889 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2890 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2891 /* the inode is shared */
2892 new = record_old_file_extents(inode, ordered_extent);
2893
2894 clear_extent_bit(io_tree, ordered_extent->file_offset,
2895 ordered_extent->file_offset + ordered_extent->len - 1,
2896 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2897 }
2898
2899 if (nolock)
2900 trans = btrfs_join_transaction_nolock(root);
2901 else
2902 trans = btrfs_join_transaction(root);
2903 if (IS_ERR(trans)) {
2904 ret = PTR_ERR(trans);
2905 trans = NULL;
2906 goto out_unlock;
2907 }
2908
2909 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2910
2911 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2912 compress_type = ordered_extent->compress_type;
2913 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2914 BUG_ON(compress_type);
2915 ret = btrfs_mark_extent_written(trans, inode,
2916 ordered_extent->file_offset,
2917 ordered_extent->file_offset +
2918 logical_len);
2919 } else {
2920 BUG_ON(root == root->fs_info->tree_root);
2921 ret = insert_reserved_file_extent(trans, inode,
2922 ordered_extent->file_offset,
2923 ordered_extent->start,
2924 ordered_extent->disk_len,
2925 logical_len, logical_len,
2926 compress_type, 0, 0,
2927 BTRFS_FILE_EXTENT_REG);
2928 if (!ret)
2929 btrfs_release_delalloc_bytes(root,
2930 ordered_extent->start,
2931 ordered_extent->disk_len);
2932 }
2933 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2934 ordered_extent->file_offset, ordered_extent->len,
2935 trans->transid);
2936 if (ret < 0) {
2937 btrfs_abort_transaction(trans, root, ret);
2938 goto out_unlock;
2939 }
2940
2941 add_pending_csums(trans, inode, ordered_extent->file_offset,
2942 &ordered_extent->list);
2943
2944 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2945 ret = btrfs_update_inode_fallback(trans, root, inode);
2946 if (ret) { /* -ENOMEM or corruption */
2947 btrfs_abort_transaction(trans, root, ret);
2948 goto out_unlock;
2949 }
2950 ret = 0;
2951 out_unlock:
2952 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2953 ordered_extent->file_offset +
2954 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2955 out:
2956 if (root != root->fs_info->tree_root)
2957 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2958 if (trans)
2959 btrfs_end_transaction(trans, root);
2960
2961 if (ret || truncated) {
2962 u64 start, end;
2963
2964 if (truncated)
2965 start = ordered_extent->file_offset + logical_len;
2966 else
2967 start = ordered_extent->file_offset;
2968 end = ordered_extent->file_offset + ordered_extent->len - 1;
2969 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2970
2971 /* Drop the cache for the part of the extent we didn't write. */
2972 btrfs_drop_extent_cache(inode, start, end, 0);
2973
2974 /*
2975 * If the ordered extent had an IOERR or something else went
2976 * wrong we need to return the space for this ordered extent
2977 * back to the allocator. We only free the extent in the
2978 * truncated case if we didn't write out the extent at all.
2979 */
2980 if ((ret || !logical_len) &&
2981 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2982 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2983 btrfs_free_reserved_extent(root, ordered_extent->start,
2984 ordered_extent->disk_len, 1);
2985 }
2986
2987
2988 /*
2989 * This needs to be done to make sure anybody waiting knows we are done
2990 * updating everything for this ordered extent.
2991 */
2992 btrfs_remove_ordered_extent(inode, ordered_extent);
2993
2994 /* for snapshot-aware defrag */
2995 if (new) {
2996 if (ret) {
2997 free_sa_defrag_extent(new);
2998 atomic_dec(&root->fs_info->defrag_running);
2999 } else {
3000 relink_file_extents(new);
3001 }
3002 }
3003
3004 /* once for us */
3005 btrfs_put_ordered_extent(ordered_extent);
3006 /* once for the tree */
3007 btrfs_put_ordered_extent(ordered_extent);
3008
3009 return ret;
3010 }
3011
3012 static void finish_ordered_fn(struct btrfs_work *work)
3013 {
3014 struct btrfs_ordered_extent *ordered_extent;
3015 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3016 btrfs_finish_ordered_io(ordered_extent);
3017 }
3018
3019 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3020 struct extent_state *state, int uptodate)
3021 {
3022 struct inode *inode = page->mapping->host;
3023 struct btrfs_root *root = BTRFS_I(inode)->root;
3024 struct btrfs_ordered_extent *ordered_extent = NULL;
3025 struct btrfs_workqueue *wq;
3026 btrfs_work_func_t func;
3027
3028 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3029
3030 ClearPagePrivate2(page);
3031 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3032 end - start + 1, uptodate))
3033 return 0;
3034
3035 if (btrfs_is_free_space_inode(inode)) {
3036 wq = root->fs_info->endio_freespace_worker;
3037 func = btrfs_freespace_write_helper;
3038 } else {
3039 wq = root->fs_info->endio_write_workers;
3040 func = btrfs_endio_write_helper;
3041 }
3042
3043 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3044 NULL);
3045 btrfs_queue_work(wq, &ordered_extent->work);
3046
3047 return 0;
3048 }
3049
3050 static int __readpage_endio_check(struct inode *inode,
3051 struct btrfs_io_bio *io_bio,
3052 int icsum, struct page *page,
3053 int pgoff, u64 start, size_t len)
3054 {
3055 char *kaddr;
3056 u32 csum_expected;
3057 u32 csum = ~(u32)0;
3058
3059 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3060
3061 kaddr = kmap_atomic(page);
3062 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3063 btrfs_csum_final(csum, (char *)&csum);
3064 if (csum != csum_expected)
3065 goto zeroit;
3066
3067 kunmap_atomic(kaddr);
3068 return 0;
3069 zeroit:
3070 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3071 "csum failed ino %llu off %llu csum %u expected csum %u",
3072 btrfs_ino(inode), start, csum, csum_expected);
3073 memset(kaddr + pgoff, 1, len);
3074 flush_dcache_page(page);
3075 kunmap_atomic(kaddr);
3076 if (csum_expected == 0)
3077 return 0;
3078 return -EIO;
3079 }
3080
3081 /*
3082 * when reads are done, we need to check csums to verify the data is correct
3083 * if there's a match, we allow the bio to finish. If not, the code in
3084 * extent_io.c will try to find good copies for us.
3085 */
3086 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3087 u64 phy_offset, struct page *page,
3088 u64 start, u64 end, int mirror)
3089 {
3090 size_t offset = start - page_offset(page);
3091 struct inode *inode = page->mapping->host;
3092 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3093 struct btrfs_root *root = BTRFS_I(inode)->root;
3094
3095 if (PageChecked(page)) {
3096 ClearPageChecked(page);
3097 return 0;
3098 }
3099
3100 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3101 return 0;
3102
3103 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3104 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3105 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3106 GFP_NOFS);
3107 return 0;
3108 }
3109
3110 phy_offset >>= inode->i_sb->s_blocksize_bits;
3111 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112 start, (size_t)(end - start + 1));
3113 }
3114
3115 void btrfs_add_delayed_iput(struct inode *inode)
3116 {
3117 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3118 struct btrfs_inode *binode = BTRFS_I(inode);
3119
3120 if (atomic_add_unless(&inode->i_count, -1, 1))
3121 return;
3122
3123 spin_lock(&fs_info->delayed_iput_lock);
3124 if (binode->delayed_iput_count == 0) {
3125 ASSERT(list_empty(&binode->delayed_iput));
3126 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3127 } else {
3128 binode->delayed_iput_count++;
3129 }
3130 spin_unlock(&fs_info->delayed_iput_lock);
3131 }
3132
3133 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3134 {
3135 struct btrfs_fs_info *fs_info = root->fs_info;
3136
3137 spin_lock(&fs_info->delayed_iput_lock);
3138 while (!list_empty(&fs_info->delayed_iputs)) {
3139 struct btrfs_inode *inode;
3140
3141 inode = list_first_entry(&fs_info->delayed_iputs,
3142 struct btrfs_inode, delayed_iput);
3143 if (inode->delayed_iput_count) {
3144 inode->delayed_iput_count--;
3145 list_move_tail(&inode->delayed_iput,
3146 &fs_info->delayed_iputs);
3147 } else {
3148 list_del_init(&inode->delayed_iput);
3149 }
3150 spin_unlock(&fs_info->delayed_iput_lock);
3151 iput(&inode->vfs_inode);
3152 spin_lock(&fs_info->delayed_iput_lock);
3153 }
3154 spin_unlock(&fs_info->delayed_iput_lock);
3155 }
3156
3157 /*
3158 * This is called in transaction commit time. If there are no orphan
3159 * files in the subvolume, it removes orphan item and frees block_rsv
3160 * structure.
3161 */
3162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root)
3164 {
3165 struct btrfs_block_rsv *block_rsv;
3166 int ret;
3167
3168 if (atomic_read(&root->orphan_inodes) ||
3169 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3170 return;
3171
3172 spin_lock(&root->orphan_lock);
3173 if (atomic_read(&root->orphan_inodes)) {
3174 spin_unlock(&root->orphan_lock);
3175 return;
3176 }
3177
3178 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179 spin_unlock(&root->orphan_lock);
3180 return;
3181 }
3182
3183 block_rsv = root->orphan_block_rsv;
3184 root->orphan_block_rsv = NULL;
3185 spin_unlock(&root->orphan_lock);
3186
3187 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3188 btrfs_root_refs(&root->root_item) > 0) {
3189 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190 root->root_key.objectid);
3191 if (ret)
3192 btrfs_abort_transaction(trans, root, ret);
3193 else
3194 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3195 &root->state);
3196 }
3197
3198 if (block_rsv) {
3199 WARN_ON(block_rsv->size > 0);
3200 btrfs_free_block_rsv(root, block_rsv);
3201 }
3202 }
3203
3204 /*
3205 * This creates an orphan entry for the given inode in case something goes
3206 * wrong in the middle of an unlink/truncate.
3207 *
3208 * NOTE: caller of this function should reserve 5 units of metadata for
3209 * this function.
3210 */
3211 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3212 {
3213 struct btrfs_root *root = BTRFS_I(inode)->root;
3214 struct btrfs_block_rsv *block_rsv = NULL;
3215 int reserve = 0;
3216 int insert = 0;
3217 int ret;
3218
3219 if (!root->orphan_block_rsv) {
3220 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3221 if (!block_rsv)
3222 return -ENOMEM;
3223 }
3224
3225 spin_lock(&root->orphan_lock);
3226 if (!root->orphan_block_rsv) {
3227 root->orphan_block_rsv = block_rsv;
3228 } else if (block_rsv) {
3229 btrfs_free_block_rsv(root, block_rsv);
3230 block_rsv = NULL;
3231 }
3232
3233 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234 &BTRFS_I(inode)->runtime_flags)) {
3235 #if 0
3236 /*
3237 * For proper ENOSPC handling, we should do orphan
3238 * cleanup when mounting. But this introduces backward
3239 * compatibility issue.
3240 */
3241 if (!xchg(&root->orphan_item_inserted, 1))
3242 insert = 2;
3243 else
3244 insert = 1;
3245 #endif
3246 insert = 1;
3247 atomic_inc(&root->orphan_inodes);
3248 }
3249
3250 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251 &BTRFS_I(inode)->runtime_flags))
3252 reserve = 1;
3253 spin_unlock(&root->orphan_lock);
3254
3255 /* grab metadata reservation from transaction handle */
3256 if (reserve) {
3257 ret = btrfs_orphan_reserve_metadata(trans, inode);
3258 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3259 }
3260
3261 /* insert an orphan item to track this unlinked/truncated file */
3262 if (insert >= 1) {
3263 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3264 if (ret) {
3265 atomic_dec(&root->orphan_inodes);
3266 if (reserve) {
3267 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268 &BTRFS_I(inode)->runtime_flags);
3269 btrfs_orphan_release_metadata(inode);
3270 }
3271 if (ret != -EEXIST) {
3272 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273 &BTRFS_I(inode)->runtime_flags);
3274 btrfs_abort_transaction(trans, root, ret);
3275 return ret;
3276 }
3277 }
3278 ret = 0;
3279 }
3280
3281 /* insert an orphan item to track subvolume contains orphan files */
3282 if (insert >= 2) {
3283 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284 root->root_key.objectid);
3285 if (ret && ret != -EEXIST) {
3286 btrfs_abort_transaction(trans, root, ret);
3287 return ret;
3288 }
3289 }
3290 return 0;
3291 }
3292
3293 /*
3294 * We have done the truncate/delete so we can go ahead and remove the orphan
3295 * item for this particular inode.
3296 */
3297 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298 struct inode *inode)
3299 {
3300 struct btrfs_root *root = BTRFS_I(inode)->root;
3301 int delete_item = 0;
3302 int release_rsv = 0;
3303 int ret = 0;
3304
3305 spin_lock(&root->orphan_lock);
3306 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307 &BTRFS_I(inode)->runtime_flags))
3308 delete_item = 1;
3309
3310 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311 &BTRFS_I(inode)->runtime_flags))
3312 release_rsv = 1;
3313 spin_unlock(&root->orphan_lock);
3314
3315 if (delete_item) {
3316 atomic_dec(&root->orphan_inodes);
3317 if (trans)
3318 ret = btrfs_del_orphan_item(trans, root,
3319 btrfs_ino(inode));
3320 }
3321
3322 if (release_rsv)
3323 btrfs_orphan_release_metadata(inode);
3324
3325 return ret;
3326 }
3327
3328 /*
3329 * this cleans up any orphans that may be left on the list from the last use
3330 * of this root.
3331 */
3332 int btrfs_orphan_cleanup(struct btrfs_root *root)
3333 {
3334 struct btrfs_path *path;
3335 struct extent_buffer *leaf;
3336 struct btrfs_key key, found_key;
3337 struct btrfs_trans_handle *trans;
3338 struct inode *inode;
3339 u64 last_objectid = 0;
3340 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3341
3342 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3343 return 0;
3344
3345 path = btrfs_alloc_path();
3346 if (!path) {
3347 ret = -ENOMEM;
3348 goto out;
3349 }
3350 path->reada = READA_BACK;
3351
3352 key.objectid = BTRFS_ORPHAN_OBJECTID;
3353 key.type = BTRFS_ORPHAN_ITEM_KEY;
3354 key.offset = (u64)-1;
3355
3356 while (1) {
3357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3358 if (ret < 0)
3359 goto out;
3360
3361 /*
3362 * if ret == 0 means we found what we were searching for, which
3363 * is weird, but possible, so only screw with path if we didn't
3364 * find the key and see if we have stuff that matches
3365 */
3366 if (ret > 0) {
3367 ret = 0;
3368 if (path->slots[0] == 0)
3369 break;
3370 path->slots[0]--;
3371 }
3372
3373 /* pull out the item */
3374 leaf = path->nodes[0];
3375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3376
3377 /* make sure the item matches what we want */
3378 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3379 break;
3380 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3381 break;
3382
3383 /* release the path since we're done with it */
3384 btrfs_release_path(path);
3385
3386 /*
3387 * this is where we are basically btrfs_lookup, without the
3388 * crossing root thing. we store the inode number in the
3389 * offset of the orphan item.
3390 */
3391
3392 if (found_key.offset == last_objectid) {
3393 btrfs_err(root->fs_info,
3394 "Error removing orphan entry, stopping orphan cleanup");
3395 ret = -EINVAL;
3396 goto out;
3397 }
3398
3399 last_objectid = found_key.offset;
3400
3401 found_key.objectid = found_key.offset;
3402 found_key.type = BTRFS_INODE_ITEM_KEY;
3403 found_key.offset = 0;
3404 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3405 ret = PTR_ERR_OR_ZERO(inode);
3406 if (ret && ret != -ESTALE)
3407 goto out;
3408
3409 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410 struct btrfs_root *dead_root;
3411 struct btrfs_fs_info *fs_info = root->fs_info;
3412 int is_dead_root = 0;
3413
3414 /*
3415 * this is an orphan in the tree root. Currently these
3416 * could come from 2 sources:
3417 * a) a snapshot deletion in progress
3418 * b) a free space cache inode
3419 * We need to distinguish those two, as the snapshot
3420 * orphan must not get deleted.
3421 * find_dead_roots already ran before us, so if this
3422 * is a snapshot deletion, we should find the root
3423 * in the dead_roots list
3424 */
3425 spin_lock(&fs_info->trans_lock);
3426 list_for_each_entry(dead_root, &fs_info->dead_roots,
3427 root_list) {
3428 if (dead_root->root_key.objectid ==
3429 found_key.objectid) {
3430 is_dead_root = 1;
3431 break;
3432 }
3433 }
3434 spin_unlock(&fs_info->trans_lock);
3435 if (is_dead_root) {
3436 /* prevent this orphan from being found again */
3437 key.offset = found_key.objectid - 1;
3438 continue;
3439 }
3440 }
3441 /*
3442 * Inode is already gone but the orphan item is still there,
3443 * kill the orphan item.
3444 */
3445 if (ret == -ESTALE) {
3446 trans = btrfs_start_transaction(root, 1);
3447 if (IS_ERR(trans)) {
3448 ret = PTR_ERR(trans);
3449 goto out;
3450 }
3451 btrfs_debug(root->fs_info, "auto deleting %Lu",
3452 found_key.objectid);
3453 ret = btrfs_del_orphan_item(trans, root,
3454 found_key.objectid);
3455 btrfs_end_transaction(trans, root);
3456 if (ret)
3457 goto out;
3458 continue;
3459 }
3460
3461 /*
3462 * add this inode to the orphan list so btrfs_orphan_del does
3463 * the proper thing when we hit it
3464 */
3465 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466 &BTRFS_I(inode)->runtime_flags);
3467 atomic_inc(&root->orphan_inodes);
3468
3469 /* if we have links, this was a truncate, lets do that */
3470 if (inode->i_nlink) {
3471 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3472 iput(inode);
3473 continue;
3474 }
3475 nr_truncate++;
3476
3477 /* 1 for the orphan item deletion. */
3478 trans = btrfs_start_transaction(root, 1);
3479 if (IS_ERR(trans)) {
3480 iput(inode);
3481 ret = PTR_ERR(trans);
3482 goto out;
3483 }
3484 ret = btrfs_orphan_add(trans, inode);
3485 btrfs_end_transaction(trans, root);
3486 if (ret) {
3487 iput(inode);
3488 goto out;
3489 }
3490
3491 ret = btrfs_truncate(inode);
3492 if (ret)
3493 btrfs_orphan_del(NULL, inode);
3494 } else {
3495 nr_unlink++;
3496 }
3497
3498 /* this will do delete_inode and everything for us */
3499 iput(inode);
3500 if (ret)
3501 goto out;
3502 }
3503 /* release the path since we're done with it */
3504 btrfs_release_path(path);
3505
3506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3507
3508 if (root->orphan_block_rsv)
3509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3510 (u64)-1);
3511
3512 if (root->orphan_block_rsv ||
3513 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3514 trans = btrfs_join_transaction(root);
3515 if (!IS_ERR(trans))
3516 btrfs_end_transaction(trans, root);
3517 }
3518
3519 if (nr_unlink)
3520 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3521 if (nr_truncate)
3522 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3523
3524 out:
3525 if (ret)
3526 btrfs_err(root->fs_info,
3527 "could not do orphan cleanup %d", ret);
3528 btrfs_free_path(path);
3529 return ret;
3530 }
3531
3532 /*
3533 * very simple check to peek ahead in the leaf looking for xattrs. If we
3534 * don't find any xattrs, we know there can't be any acls.
3535 *
3536 * slot is the slot the inode is in, objectid is the objectid of the inode
3537 */
3538 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3539 int slot, u64 objectid,
3540 int *first_xattr_slot)
3541 {
3542 u32 nritems = btrfs_header_nritems(leaf);
3543 struct btrfs_key found_key;
3544 static u64 xattr_access = 0;
3545 static u64 xattr_default = 0;
3546 int scanned = 0;
3547
3548 if (!xattr_access) {
3549 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3550 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3551 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3552 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3553 }
3554
3555 slot++;
3556 *first_xattr_slot = -1;
3557 while (slot < nritems) {
3558 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3559
3560 /* we found a different objectid, there must not be acls */
3561 if (found_key.objectid != objectid)
3562 return 0;
3563
3564 /* we found an xattr, assume we've got an acl */
3565 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3566 if (*first_xattr_slot == -1)
3567 *first_xattr_slot = slot;
3568 if (found_key.offset == xattr_access ||
3569 found_key.offset == xattr_default)
3570 return 1;
3571 }
3572
3573 /*
3574 * we found a key greater than an xattr key, there can't
3575 * be any acls later on
3576 */
3577 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3578 return 0;
3579
3580 slot++;
3581 scanned++;
3582
3583 /*
3584 * it goes inode, inode backrefs, xattrs, extents,
3585 * so if there are a ton of hard links to an inode there can
3586 * be a lot of backrefs. Don't waste time searching too hard,
3587 * this is just an optimization
3588 */
3589 if (scanned >= 8)
3590 break;
3591 }
3592 /* we hit the end of the leaf before we found an xattr or
3593 * something larger than an xattr. We have to assume the inode
3594 * has acls
3595 */
3596 if (*first_xattr_slot == -1)
3597 *first_xattr_slot = slot;
3598 return 1;
3599 }
3600
3601 /*
3602 * read an inode from the btree into the in-memory inode
3603 */
3604 static void btrfs_read_locked_inode(struct inode *inode)
3605 {
3606 struct btrfs_path *path;
3607 struct extent_buffer *leaf;
3608 struct btrfs_inode_item *inode_item;
3609 struct btrfs_root *root = BTRFS_I(inode)->root;
3610 struct btrfs_key location;
3611 unsigned long ptr;
3612 int maybe_acls;
3613 u32 rdev;
3614 int ret;
3615 bool filled = false;
3616 int first_xattr_slot;
3617
3618 ret = btrfs_fill_inode(inode, &rdev);
3619 if (!ret)
3620 filled = true;
3621
3622 path = btrfs_alloc_path();
3623 if (!path)
3624 goto make_bad;
3625
3626 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3627
3628 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3629 if (ret)
3630 goto make_bad;
3631
3632 leaf = path->nodes[0];
3633
3634 if (filled)
3635 goto cache_index;
3636
3637 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638 struct btrfs_inode_item);
3639 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3640 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3641 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3643 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3644
3645 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3647
3648 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3650
3651 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3653
3654 BTRFS_I(inode)->i_otime.tv_sec =
3655 btrfs_timespec_sec(leaf, &inode_item->otime);
3656 BTRFS_I(inode)->i_otime.tv_nsec =
3657 btrfs_timespec_nsec(leaf, &inode_item->otime);
3658
3659 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3660 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3661 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3662
3663 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664 inode->i_generation = BTRFS_I(inode)->generation;
3665 inode->i_rdev = 0;
3666 rdev = btrfs_inode_rdev(leaf, inode_item);
3667
3668 BTRFS_I(inode)->index_cnt = (u64)-1;
3669 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3670
3671 cache_index:
3672 /*
3673 * If we were modified in the current generation and evicted from memory
3674 * and then re-read we need to do a full sync since we don't have any
3675 * idea about which extents were modified before we were evicted from
3676 * cache.
3677 *
3678 * This is required for both inode re-read from disk and delayed inode
3679 * in delayed_nodes_tree.
3680 */
3681 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683 &BTRFS_I(inode)->runtime_flags);
3684
3685 /*
3686 * We don't persist the id of the transaction where an unlink operation
3687 * against the inode was last made. So here we assume the inode might
3688 * have been evicted, and therefore the exact value of last_unlink_trans
3689 * lost, and set it to last_trans to avoid metadata inconsistencies
3690 * between the inode and its parent if the inode is fsync'ed and the log
3691 * replayed. For example, in the scenario:
3692 *
3693 * touch mydir/foo
3694 * ln mydir/foo mydir/bar
3695 * sync
3696 * unlink mydir/bar
3697 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3698 * xfs_io -c fsync mydir/foo
3699 * <power failure>
3700 * mount fs, triggers fsync log replay
3701 *
3702 * We must make sure that when we fsync our inode foo we also log its
3703 * parent inode, otherwise after log replay the parent still has the
3704 * dentry with the "bar" name but our inode foo has a link count of 1
3705 * and doesn't have an inode ref with the name "bar" anymore.
3706 *
3707 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708 * but it guarantees correctness at the expense of ocassional full
3709 * transaction commits on fsync if our inode is a directory, or if our
3710 * inode is not a directory, logging its parent unnecessarily.
3711 */
3712 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3713
3714 path->slots[0]++;
3715 if (inode->i_nlink != 1 ||
3716 path->slots[0] >= btrfs_header_nritems(leaf))
3717 goto cache_acl;
3718
3719 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720 if (location.objectid != btrfs_ino(inode))
3721 goto cache_acl;
3722
3723 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724 if (location.type == BTRFS_INODE_REF_KEY) {
3725 struct btrfs_inode_ref *ref;
3726
3727 ref = (struct btrfs_inode_ref *)ptr;
3728 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730 struct btrfs_inode_extref *extref;
3731
3732 extref = (struct btrfs_inode_extref *)ptr;
3733 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3734 extref);
3735 }
3736 cache_acl:
3737 /*
3738 * try to precache a NULL acl entry for files that don't have
3739 * any xattrs or acls
3740 */
3741 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3742 btrfs_ino(inode), &first_xattr_slot);
3743 if (first_xattr_slot != -1) {
3744 path->slots[0] = first_xattr_slot;
3745 ret = btrfs_load_inode_props(inode, path);
3746 if (ret)
3747 btrfs_err(root->fs_info,
3748 "error loading props for ino %llu (root %llu): %d",
3749 btrfs_ino(inode),
3750 root->root_key.objectid, ret);
3751 }
3752 btrfs_free_path(path);
3753
3754 if (!maybe_acls)
3755 cache_no_acl(inode);
3756
3757 switch (inode->i_mode & S_IFMT) {
3758 case S_IFREG:
3759 inode->i_mapping->a_ops = &btrfs_aops;
3760 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3761 inode->i_fop = &btrfs_file_operations;
3762 inode->i_op = &btrfs_file_inode_operations;
3763 break;
3764 case S_IFDIR:
3765 inode->i_fop = &btrfs_dir_file_operations;
3766 if (root == root->fs_info->tree_root)
3767 inode->i_op = &btrfs_dir_ro_inode_operations;
3768 else
3769 inode->i_op = &btrfs_dir_inode_operations;
3770 break;
3771 case S_IFLNK:
3772 inode->i_op = &btrfs_symlink_inode_operations;
3773 inode_nohighmem(inode);
3774 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3775 break;
3776 default:
3777 inode->i_op = &btrfs_special_inode_operations;
3778 init_special_inode(inode, inode->i_mode, rdev);
3779 break;
3780 }
3781
3782 btrfs_update_iflags(inode);
3783 return;
3784
3785 make_bad:
3786 btrfs_free_path(path);
3787 make_bad_inode(inode);
3788 }
3789
3790 /*
3791 * given a leaf and an inode, copy the inode fields into the leaf
3792 */
3793 static void fill_inode_item(struct btrfs_trans_handle *trans,
3794 struct extent_buffer *leaf,
3795 struct btrfs_inode_item *item,
3796 struct inode *inode)
3797 {
3798 struct btrfs_map_token token;
3799
3800 btrfs_init_map_token(&token);
3801
3802 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3803 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3804 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805 &token);
3806 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3807 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3808
3809 btrfs_set_token_timespec_sec(leaf, &item->atime,
3810 inode->i_atime.tv_sec, &token);
3811 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3812 inode->i_atime.tv_nsec, &token);
3813
3814 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3815 inode->i_mtime.tv_sec, &token);
3816 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3817 inode->i_mtime.tv_nsec, &token);
3818
3819 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3820 inode->i_ctime.tv_sec, &token);
3821 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3822 inode->i_ctime.tv_nsec, &token);
3823
3824 btrfs_set_token_timespec_sec(leaf, &item->otime,
3825 BTRFS_I(inode)->i_otime.tv_sec, &token);
3826 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3827 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828
3829 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830 &token);
3831 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832 &token);
3833 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3834 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3835 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3836 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3837 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3838 }
3839
3840 /*
3841 * copy everything in the in-memory inode into the btree.
3842 */
3843 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3844 struct btrfs_root *root, struct inode *inode)
3845 {
3846 struct btrfs_inode_item *inode_item;
3847 struct btrfs_path *path;
3848 struct extent_buffer *leaf;
3849 int ret;
3850
3851 path = btrfs_alloc_path();
3852 if (!path)
3853 return -ENOMEM;
3854
3855 path->leave_spinning = 1;
3856 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3857 1);
3858 if (ret) {
3859 if (ret > 0)
3860 ret = -ENOENT;
3861 goto failed;
3862 }
3863
3864 leaf = path->nodes[0];
3865 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3866 struct btrfs_inode_item);
3867
3868 fill_inode_item(trans, leaf, inode_item, inode);
3869 btrfs_mark_buffer_dirty(leaf);
3870 btrfs_set_inode_last_trans(trans, inode);
3871 ret = 0;
3872 failed:
3873 btrfs_free_path(path);
3874 return ret;
3875 }
3876
3877 /*
3878 * copy everything in the in-memory inode into the btree.
3879 */
3880 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3881 struct btrfs_root *root, struct inode *inode)
3882 {
3883 int ret;
3884
3885 /*
3886 * If the inode is a free space inode, we can deadlock during commit
3887 * if we put it into the delayed code.
3888 *
3889 * The data relocation inode should also be directly updated
3890 * without delay
3891 */
3892 if (!btrfs_is_free_space_inode(inode)
3893 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3894 && !root->fs_info->log_root_recovering) {
3895 btrfs_update_root_times(trans, root);
3896
3897 ret = btrfs_delayed_update_inode(trans, root, inode);
3898 if (!ret)
3899 btrfs_set_inode_last_trans(trans, inode);
3900 return ret;
3901 }
3902
3903 return btrfs_update_inode_item(trans, root, inode);
3904 }
3905
3906 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3907 struct btrfs_root *root,
3908 struct inode *inode)
3909 {
3910 int ret;
3911
3912 ret = btrfs_update_inode(trans, root, inode);
3913 if (ret == -ENOSPC)
3914 return btrfs_update_inode_item(trans, root, inode);
3915 return ret;
3916 }
3917
3918 /*
3919 * unlink helper that gets used here in inode.c and in the tree logging
3920 * recovery code. It remove a link in a directory with a given name, and
3921 * also drops the back refs in the inode to the directory
3922 */
3923 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3924 struct btrfs_root *root,
3925 struct inode *dir, struct inode *inode,
3926 const char *name, int name_len)
3927 {
3928 struct btrfs_path *path;
3929 int ret = 0;
3930 struct extent_buffer *leaf;
3931 struct btrfs_dir_item *di;
3932 struct btrfs_key key;
3933 u64 index;
3934 u64 ino = btrfs_ino(inode);
3935 u64 dir_ino = btrfs_ino(dir);
3936
3937 path = btrfs_alloc_path();
3938 if (!path) {
3939 ret = -ENOMEM;
3940 goto out;
3941 }
3942
3943 path->leave_spinning = 1;
3944 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3945 name, name_len, -1);
3946 if (IS_ERR(di)) {
3947 ret = PTR_ERR(di);
3948 goto err;
3949 }
3950 if (!di) {
3951 ret = -ENOENT;
3952 goto err;
3953 }
3954 leaf = path->nodes[0];
3955 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3956 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3957 if (ret)
3958 goto err;
3959 btrfs_release_path(path);
3960
3961 /*
3962 * If we don't have dir index, we have to get it by looking up
3963 * the inode ref, since we get the inode ref, remove it directly,
3964 * it is unnecessary to do delayed deletion.
3965 *
3966 * But if we have dir index, needn't search inode ref to get it.
3967 * Since the inode ref is close to the inode item, it is better
3968 * that we delay to delete it, and just do this deletion when
3969 * we update the inode item.
3970 */
3971 if (BTRFS_I(inode)->dir_index) {
3972 ret = btrfs_delayed_delete_inode_ref(inode);
3973 if (!ret) {
3974 index = BTRFS_I(inode)->dir_index;
3975 goto skip_backref;
3976 }
3977 }
3978
3979 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3980 dir_ino, &index);
3981 if (ret) {
3982 btrfs_info(root->fs_info,
3983 "failed to delete reference to %.*s, inode %llu parent %llu",
3984 name_len, name, ino, dir_ino);
3985 btrfs_abort_transaction(trans, root, ret);
3986 goto err;
3987 }
3988 skip_backref:
3989 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3990 if (ret) {
3991 btrfs_abort_transaction(trans, root, ret);
3992 goto err;
3993 }
3994
3995 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3996 inode, dir_ino);
3997 if (ret != 0 && ret != -ENOENT) {
3998 btrfs_abort_transaction(trans, root, ret);
3999 goto err;
4000 }
4001
4002 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4003 dir, index);
4004 if (ret == -ENOENT)
4005 ret = 0;
4006 else if (ret)
4007 btrfs_abort_transaction(trans, root, ret);
4008 err:
4009 btrfs_free_path(path);
4010 if (ret)
4011 goto out;
4012
4013 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4014 inode_inc_iversion(inode);
4015 inode_inc_iversion(dir);
4016 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4017 ret = btrfs_update_inode(trans, root, dir);
4018 out:
4019 return ret;
4020 }
4021
4022 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4023 struct btrfs_root *root,
4024 struct inode *dir, struct inode *inode,
4025 const char *name, int name_len)
4026 {
4027 int ret;
4028 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4029 if (!ret) {
4030 drop_nlink(inode);
4031 ret = btrfs_update_inode(trans, root, inode);
4032 }
4033 return ret;
4034 }
4035
4036 /*
4037 * helper to start transaction for unlink and rmdir.
4038 *
4039 * unlink and rmdir are special in btrfs, they do not always free space, so
4040 * if we cannot make our reservations the normal way try and see if there is
4041 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4042 * allow the unlink to occur.
4043 */
4044 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4045 {
4046 struct btrfs_root *root = BTRFS_I(dir)->root;
4047
4048 /*
4049 * 1 for the possible orphan item
4050 * 1 for the dir item
4051 * 1 for the dir index
4052 * 1 for the inode ref
4053 * 1 for the inode
4054 */
4055 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4056 }
4057
4058 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4059 {
4060 struct btrfs_root *root = BTRFS_I(dir)->root;
4061 struct btrfs_trans_handle *trans;
4062 struct inode *inode = d_inode(dentry);
4063 int ret;
4064
4065 trans = __unlink_start_trans(dir);
4066 if (IS_ERR(trans))
4067 return PTR_ERR(trans);
4068
4069 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4070
4071 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4072 dentry->d_name.name, dentry->d_name.len);
4073 if (ret)
4074 goto out;
4075
4076 if (inode->i_nlink == 0) {
4077 ret = btrfs_orphan_add(trans, inode);
4078 if (ret)
4079 goto out;
4080 }
4081
4082 out:
4083 btrfs_end_transaction(trans, root);
4084 btrfs_btree_balance_dirty(root);
4085 return ret;
4086 }
4087
4088 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4089 struct btrfs_root *root,
4090 struct inode *dir, u64 objectid,
4091 const char *name, int name_len)
4092 {
4093 struct btrfs_path *path;
4094 struct extent_buffer *leaf;
4095 struct btrfs_dir_item *di;
4096 struct btrfs_key key;
4097 u64 index;
4098 int ret;
4099 u64 dir_ino = btrfs_ino(dir);
4100
4101 path = btrfs_alloc_path();
4102 if (!path)
4103 return -ENOMEM;
4104
4105 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4106 name, name_len, -1);
4107 if (IS_ERR_OR_NULL(di)) {
4108 if (!di)
4109 ret = -ENOENT;
4110 else
4111 ret = PTR_ERR(di);
4112 goto out;
4113 }
4114
4115 leaf = path->nodes[0];
4116 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4117 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4118 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4119 if (ret) {
4120 btrfs_abort_transaction(trans, root, ret);
4121 goto out;
4122 }
4123 btrfs_release_path(path);
4124
4125 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4126 objectid, root->root_key.objectid,
4127 dir_ino, &index, name, name_len);
4128 if (ret < 0) {
4129 if (ret != -ENOENT) {
4130 btrfs_abort_transaction(trans, root, ret);
4131 goto out;
4132 }
4133 di = btrfs_search_dir_index_item(root, path, dir_ino,
4134 name, name_len);
4135 if (IS_ERR_OR_NULL(di)) {
4136 if (!di)
4137 ret = -ENOENT;
4138 else
4139 ret = PTR_ERR(di);
4140 btrfs_abort_transaction(trans, root, ret);
4141 goto out;
4142 }
4143
4144 leaf = path->nodes[0];
4145 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4146 btrfs_release_path(path);
4147 index = key.offset;
4148 }
4149 btrfs_release_path(path);
4150
4151 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4152 if (ret) {
4153 btrfs_abort_transaction(trans, root, ret);
4154 goto out;
4155 }
4156
4157 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4158 inode_inc_iversion(dir);
4159 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4160 ret = btrfs_update_inode_fallback(trans, root, dir);
4161 if (ret)
4162 btrfs_abort_transaction(trans, root, ret);
4163 out:
4164 btrfs_free_path(path);
4165 return ret;
4166 }
4167
4168 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4169 {
4170 struct inode *inode = d_inode(dentry);
4171 int err = 0;
4172 struct btrfs_root *root = BTRFS_I(dir)->root;
4173 struct btrfs_trans_handle *trans;
4174
4175 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4176 return -ENOTEMPTY;
4177 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4178 return -EPERM;
4179
4180 trans = __unlink_start_trans(dir);
4181 if (IS_ERR(trans))
4182 return PTR_ERR(trans);
4183
4184 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4185 err = btrfs_unlink_subvol(trans, root, dir,
4186 BTRFS_I(inode)->location.objectid,
4187 dentry->d_name.name,
4188 dentry->d_name.len);
4189 goto out;
4190 }
4191
4192 err = btrfs_orphan_add(trans, inode);
4193 if (err)
4194 goto out;
4195
4196 /* now the directory is empty */
4197 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4198 dentry->d_name.name, dentry->d_name.len);
4199 if (!err)
4200 btrfs_i_size_write(inode, 0);
4201 out:
4202 btrfs_end_transaction(trans, root);
4203 btrfs_btree_balance_dirty(root);
4204
4205 return err;
4206 }
4207
4208 static int truncate_space_check(struct btrfs_trans_handle *trans,
4209 struct btrfs_root *root,
4210 u64 bytes_deleted)
4211 {
4212 int ret;
4213
4214 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4215 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4216 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4217 if (!ret)
4218 trans->bytes_reserved += bytes_deleted;
4219 return ret;
4220
4221 }
4222
4223 static int truncate_inline_extent(struct inode *inode,
4224 struct btrfs_path *path,
4225 struct btrfs_key *found_key,
4226 const u64 item_end,
4227 const u64 new_size)
4228 {
4229 struct extent_buffer *leaf = path->nodes[0];
4230 int slot = path->slots[0];
4231 struct btrfs_file_extent_item *fi;
4232 u32 size = (u32)(new_size - found_key->offset);
4233 struct btrfs_root *root = BTRFS_I(inode)->root;
4234
4235 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4236
4237 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4238 loff_t offset = new_size;
4239 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4240
4241 /*
4242 * Zero out the remaining of the last page of our inline extent,
4243 * instead of directly truncating our inline extent here - that
4244 * would be much more complex (decompressing all the data, then
4245 * compressing the truncated data, which might be bigger than
4246 * the size of the inline extent, resize the extent, etc).
4247 * We release the path because to get the page we might need to
4248 * read the extent item from disk (data not in the page cache).
4249 */
4250 btrfs_release_path(path);
4251 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4252 }
4253
4254 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4255 size = btrfs_file_extent_calc_inline_size(size);
4256 btrfs_truncate_item(root, path, size, 1);
4257
4258 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4259 inode_sub_bytes(inode, item_end + 1 - new_size);
4260
4261 return 0;
4262 }
4263
4264 /*
4265 * this can truncate away extent items, csum items and directory items.
4266 * It starts at a high offset and removes keys until it can't find
4267 * any higher than new_size
4268 *
4269 * csum items that cross the new i_size are truncated to the new size
4270 * as well.
4271 *
4272 * min_type is the minimum key type to truncate down to. If set to 0, this
4273 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4274 */
4275 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4276 struct btrfs_root *root,
4277 struct inode *inode,
4278 u64 new_size, u32 min_type)
4279 {
4280 struct btrfs_path *path;
4281 struct extent_buffer *leaf;
4282 struct btrfs_file_extent_item *fi;
4283 struct btrfs_key key;
4284 struct btrfs_key found_key;
4285 u64 extent_start = 0;
4286 u64 extent_num_bytes = 0;
4287 u64 extent_offset = 0;
4288 u64 item_end = 0;
4289 u64 last_size = new_size;
4290 u32 found_type = (u8)-1;
4291 int found_extent;
4292 int del_item;
4293 int pending_del_nr = 0;
4294 int pending_del_slot = 0;
4295 int extent_type = -1;
4296 int ret;
4297 int err = 0;
4298 u64 ino = btrfs_ino(inode);
4299 u64 bytes_deleted = 0;
4300 bool be_nice = 0;
4301 bool should_throttle = 0;
4302 bool should_end = 0;
4303
4304 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4305
4306 /*
4307 * for non-free space inodes and ref cows, we want to back off from
4308 * time to time
4309 */
4310 if (!btrfs_is_free_space_inode(inode) &&
4311 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4312 be_nice = 1;
4313
4314 path = btrfs_alloc_path();
4315 if (!path)
4316 return -ENOMEM;
4317 path->reada = READA_BACK;
4318
4319 /*
4320 * We want to drop from the next block forward in case this new size is
4321 * not block aligned since we will be keeping the last block of the
4322 * extent just the way it is.
4323 */
4324 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4325 root == root->fs_info->tree_root)
4326 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4327 root->sectorsize), (u64)-1, 0);
4328
4329 /*
4330 * This function is also used to drop the items in the log tree before
4331 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4332 * it is used to drop the loged items. So we shouldn't kill the delayed
4333 * items.
4334 */
4335 if (min_type == 0 && root == BTRFS_I(inode)->root)
4336 btrfs_kill_delayed_inode_items(inode);
4337
4338 key.objectid = ino;
4339 key.offset = (u64)-1;
4340 key.type = (u8)-1;
4341
4342 search_again:
4343 /*
4344 * with a 16K leaf size and 128MB extents, you can actually queue
4345 * up a huge file in a single leaf. Most of the time that
4346 * bytes_deleted is > 0, it will be huge by the time we get here
4347 */
4348 if (be_nice && bytes_deleted > SZ_32M) {
4349 if (btrfs_should_end_transaction(trans, root)) {
4350 err = -EAGAIN;
4351 goto error;
4352 }
4353 }
4354
4355
4356 path->leave_spinning = 1;
4357 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4358 if (ret < 0) {
4359 err = ret;
4360 goto out;
4361 }
4362
4363 if (ret > 0) {
4364 /* there are no items in the tree for us to truncate, we're
4365 * done
4366 */
4367 if (path->slots[0] == 0)
4368 goto out;
4369 path->slots[0]--;
4370 }
4371
4372 while (1) {
4373 fi = NULL;
4374 leaf = path->nodes[0];
4375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4376 found_type = found_key.type;
4377
4378 if (found_key.objectid != ino)
4379 break;
4380
4381 if (found_type < min_type)
4382 break;
4383
4384 item_end = found_key.offset;
4385 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4386 fi = btrfs_item_ptr(leaf, path->slots[0],
4387 struct btrfs_file_extent_item);
4388 extent_type = btrfs_file_extent_type(leaf, fi);
4389 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4390 item_end +=
4391 btrfs_file_extent_num_bytes(leaf, fi);
4392 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4393 item_end += btrfs_file_extent_inline_len(leaf,
4394 path->slots[0], fi);
4395 }
4396 item_end--;
4397 }
4398 if (found_type > min_type) {
4399 del_item = 1;
4400 } else {
4401 if (item_end < new_size)
4402 break;
4403 if (found_key.offset >= new_size)
4404 del_item = 1;
4405 else
4406 del_item = 0;
4407 }
4408 found_extent = 0;
4409 /* FIXME, shrink the extent if the ref count is only 1 */
4410 if (found_type != BTRFS_EXTENT_DATA_KEY)
4411 goto delete;
4412
4413 if (del_item)
4414 last_size = found_key.offset;
4415 else
4416 last_size = new_size;
4417
4418 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4419 u64 num_dec;
4420 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4421 if (!del_item) {
4422 u64 orig_num_bytes =
4423 btrfs_file_extent_num_bytes(leaf, fi);
4424 extent_num_bytes = ALIGN(new_size -
4425 found_key.offset,
4426 root->sectorsize);
4427 btrfs_set_file_extent_num_bytes(leaf, fi,
4428 extent_num_bytes);
4429 num_dec = (orig_num_bytes -
4430 extent_num_bytes);
4431 if (test_bit(BTRFS_ROOT_REF_COWS,
4432 &root->state) &&
4433 extent_start != 0)
4434 inode_sub_bytes(inode, num_dec);
4435 btrfs_mark_buffer_dirty(leaf);
4436 } else {
4437 extent_num_bytes =
4438 btrfs_file_extent_disk_num_bytes(leaf,
4439 fi);
4440 extent_offset = found_key.offset -
4441 btrfs_file_extent_offset(leaf, fi);
4442
4443 /* FIXME blocksize != 4096 */
4444 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4445 if (extent_start != 0) {
4446 found_extent = 1;
4447 if (test_bit(BTRFS_ROOT_REF_COWS,
4448 &root->state))
4449 inode_sub_bytes(inode, num_dec);
4450 }
4451 }
4452 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4453 /*
4454 * we can't truncate inline items that have had
4455 * special encodings
4456 */
4457 if (!del_item &&
4458 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4459 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4460
4461 /*
4462 * Need to release path in order to truncate a
4463 * compressed extent. So delete any accumulated
4464 * extent items so far.
4465 */
4466 if (btrfs_file_extent_compression(leaf, fi) !=
4467 BTRFS_COMPRESS_NONE && pending_del_nr) {
4468 err = btrfs_del_items(trans, root, path,
4469 pending_del_slot,
4470 pending_del_nr);
4471 if (err) {
4472 btrfs_abort_transaction(trans,
4473 root,
4474 err);
4475 goto error;
4476 }
4477 pending_del_nr = 0;
4478 }
4479
4480 err = truncate_inline_extent(inode, path,
4481 &found_key,
4482 item_end,
4483 new_size);
4484 if (err) {
4485 btrfs_abort_transaction(trans,
4486 root, err);
4487 goto error;
4488 }
4489 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4490 &root->state)) {
4491 inode_sub_bytes(inode, item_end + 1 - new_size);
4492 }
4493 }
4494 delete:
4495 if (del_item) {
4496 if (!pending_del_nr) {
4497 /* no pending yet, add ourselves */
4498 pending_del_slot = path->slots[0];
4499 pending_del_nr = 1;
4500 } else if (pending_del_nr &&
4501 path->slots[0] + 1 == pending_del_slot) {
4502 /* hop on the pending chunk */
4503 pending_del_nr++;
4504 pending_del_slot = path->slots[0];
4505 } else {
4506 BUG();
4507 }
4508 } else {
4509 break;
4510 }
4511 should_throttle = 0;
4512
4513 if (found_extent &&
4514 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4515 root == root->fs_info->tree_root)) {
4516 btrfs_set_path_blocking(path);
4517 bytes_deleted += extent_num_bytes;
4518 ret = btrfs_free_extent(trans, root, extent_start,
4519 extent_num_bytes, 0,
4520 btrfs_header_owner(leaf),
4521 ino, extent_offset);
4522 BUG_ON(ret);
4523 if (btrfs_should_throttle_delayed_refs(trans, root))
4524 btrfs_async_run_delayed_refs(root,
4525 trans->delayed_ref_updates * 2, 0);
4526 if (be_nice) {
4527 if (truncate_space_check(trans, root,
4528 extent_num_bytes)) {
4529 should_end = 1;
4530 }
4531 if (btrfs_should_throttle_delayed_refs(trans,
4532 root)) {
4533 should_throttle = 1;
4534 }
4535 }
4536 }
4537
4538 if (found_type == BTRFS_INODE_ITEM_KEY)
4539 break;
4540
4541 if (path->slots[0] == 0 ||
4542 path->slots[0] != pending_del_slot ||
4543 should_throttle || should_end) {
4544 if (pending_del_nr) {
4545 ret = btrfs_del_items(trans, root, path,
4546 pending_del_slot,
4547 pending_del_nr);
4548 if (ret) {
4549 btrfs_abort_transaction(trans,
4550 root, ret);
4551 goto error;
4552 }
4553 pending_del_nr = 0;
4554 }
4555 btrfs_release_path(path);
4556 if (should_throttle) {
4557 unsigned long updates = trans->delayed_ref_updates;
4558 if (updates) {
4559 trans->delayed_ref_updates = 0;
4560 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4561 if (ret && !err)
4562 err = ret;
4563 }
4564 }
4565 /*
4566 * if we failed to refill our space rsv, bail out
4567 * and let the transaction restart
4568 */
4569 if (should_end) {
4570 err = -EAGAIN;
4571 goto error;
4572 }
4573 goto search_again;
4574 } else {
4575 path->slots[0]--;
4576 }
4577 }
4578 out:
4579 if (pending_del_nr) {
4580 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4581 pending_del_nr);
4582 if (ret)
4583 btrfs_abort_transaction(trans, root, ret);
4584 }
4585 error:
4586 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4587 btrfs_ordered_update_i_size(inode, last_size, NULL);
4588
4589 btrfs_free_path(path);
4590
4591 if (be_nice && bytes_deleted > SZ_32M) {
4592 unsigned long updates = trans->delayed_ref_updates;
4593 if (updates) {
4594 trans->delayed_ref_updates = 0;
4595 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4596 if (ret && !err)
4597 err = ret;
4598 }
4599 }
4600 return err;
4601 }
4602
4603 /*
4604 * btrfs_truncate_page - read, zero a chunk and write a page
4605 * @inode - inode that we're zeroing
4606 * @from - the offset to start zeroing
4607 * @len - the length to zero, 0 to zero the entire range respective to the
4608 * offset
4609 * @front - zero up to the offset instead of from the offset on
4610 *
4611 * This will find the page for the "from" offset and cow the page and zero the
4612 * part we want to zero. This is used with truncate and hole punching.
4613 */
4614 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4615 int front)
4616 {
4617 struct address_space *mapping = inode->i_mapping;
4618 struct btrfs_root *root = BTRFS_I(inode)->root;
4619 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4620 struct btrfs_ordered_extent *ordered;
4621 struct extent_state *cached_state = NULL;
4622 char *kaddr;
4623 u32 blocksize = root->sectorsize;
4624 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4625 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4626 struct page *page;
4627 gfp_t mask = btrfs_alloc_write_mask(mapping);
4628 int ret = 0;
4629 u64 page_start;
4630 u64 page_end;
4631
4632 if ((offset & (blocksize - 1)) == 0 &&
4633 (!len || ((len & (blocksize - 1)) == 0)))
4634 goto out;
4635 ret = btrfs_delalloc_reserve_space(inode,
4636 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4637 if (ret)
4638 goto out;
4639
4640 again:
4641 page = find_or_create_page(mapping, index, mask);
4642 if (!page) {
4643 btrfs_delalloc_release_space(inode,
4644 round_down(from, PAGE_CACHE_SIZE),
4645 PAGE_CACHE_SIZE);
4646 ret = -ENOMEM;
4647 goto out;
4648 }
4649
4650 page_start = page_offset(page);
4651 page_end = page_start + PAGE_CACHE_SIZE - 1;
4652
4653 if (!PageUptodate(page)) {
4654 ret = btrfs_readpage(NULL, page);
4655 lock_page(page);
4656 if (page->mapping != mapping) {
4657 unlock_page(page);
4658 page_cache_release(page);
4659 goto again;
4660 }
4661 if (!PageUptodate(page)) {
4662 ret = -EIO;
4663 goto out_unlock;
4664 }
4665 }
4666 wait_on_page_writeback(page);
4667
4668 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
4669 set_page_extent_mapped(page);
4670
4671 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4672 if (ordered) {
4673 unlock_extent_cached(io_tree, page_start, page_end,
4674 &cached_state, GFP_NOFS);
4675 unlock_page(page);
4676 page_cache_release(page);
4677 btrfs_start_ordered_extent(inode, ordered, 1);
4678 btrfs_put_ordered_extent(ordered);
4679 goto again;
4680 }
4681
4682 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4683 EXTENT_DIRTY | EXTENT_DELALLOC |
4684 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4685 0, 0, &cached_state, GFP_NOFS);
4686
4687 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4688 &cached_state);
4689 if (ret) {
4690 unlock_extent_cached(io_tree, page_start, page_end,
4691 &cached_state, GFP_NOFS);
4692 goto out_unlock;
4693 }
4694
4695 if (offset != PAGE_CACHE_SIZE) {
4696 if (!len)
4697 len = PAGE_CACHE_SIZE - offset;
4698 kaddr = kmap(page);
4699 if (front)
4700 memset(kaddr, 0, offset);
4701 else
4702 memset(kaddr + offset, 0, len);
4703 flush_dcache_page(page);
4704 kunmap(page);
4705 }
4706 ClearPageChecked(page);
4707 set_page_dirty(page);
4708 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4709 GFP_NOFS);
4710
4711 out_unlock:
4712 if (ret)
4713 btrfs_delalloc_release_space(inode, page_start,
4714 PAGE_CACHE_SIZE);
4715 unlock_page(page);
4716 page_cache_release(page);
4717 out:
4718 return ret;
4719 }
4720
4721 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4722 u64 offset, u64 len)
4723 {
4724 struct btrfs_trans_handle *trans;
4725 int ret;
4726
4727 /*
4728 * Still need to make sure the inode looks like it's been updated so
4729 * that any holes get logged if we fsync.
4730 */
4731 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4732 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4733 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4734 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4735 return 0;
4736 }
4737
4738 /*
4739 * 1 - for the one we're dropping
4740 * 1 - for the one we're adding
4741 * 1 - for updating the inode.
4742 */
4743 trans = btrfs_start_transaction(root, 3);
4744 if (IS_ERR(trans))
4745 return PTR_ERR(trans);
4746
4747 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4748 if (ret) {
4749 btrfs_abort_transaction(trans, root, ret);
4750 btrfs_end_transaction(trans, root);
4751 return ret;
4752 }
4753
4754 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4755 0, 0, len, 0, len, 0, 0, 0);
4756 if (ret)
4757 btrfs_abort_transaction(trans, root, ret);
4758 else
4759 btrfs_update_inode(trans, root, inode);
4760 btrfs_end_transaction(trans, root);
4761 return ret;
4762 }
4763
4764 /*
4765 * This function puts in dummy file extents for the area we're creating a hole
4766 * for. So if we are truncating this file to a larger size we need to insert
4767 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4768 * the range between oldsize and size
4769 */
4770 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4771 {
4772 struct btrfs_root *root = BTRFS_I(inode)->root;
4773 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4774 struct extent_map *em = NULL;
4775 struct extent_state *cached_state = NULL;
4776 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4777 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4778 u64 block_end = ALIGN(size, root->sectorsize);
4779 u64 last_byte;
4780 u64 cur_offset;
4781 u64 hole_size;
4782 int err = 0;
4783
4784 /*
4785 * If our size started in the middle of a page we need to zero out the
4786 * rest of the page before we expand the i_size, otherwise we could
4787 * expose stale data.
4788 */
4789 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4790 if (err)
4791 return err;
4792
4793 if (size <= hole_start)
4794 return 0;
4795
4796 while (1) {
4797 struct btrfs_ordered_extent *ordered;
4798
4799 lock_extent_bits(io_tree, hole_start, block_end - 1,
4800 &cached_state);
4801 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4802 block_end - hole_start);
4803 if (!ordered)
4804 break;
4805 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4806 &cached_state, GFP_NOFS);
4807 btrfs_start_ordered_extent(inode, ordered, 1);
4808 btrfs_put_ordered_extent(ordered);
4809 }
4810
4811 cur_offset = hole_start;
4812 while (1) {
4813 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4814 block_end - cur_offset, 0);
4815 if (IS_ERR(em)) {
4816 err = PTR_ERR(em);
4817 em = NULL;
4818 break;
4819 }
4820 last_byte = min(extent_map_end(em), block_end);
4821 last_byte = ALIGN(last_byte , root->sectorsize);
4822 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4823 struct extent_map *hole_em;
4824 hole_size = last_byte - cur_offset;
4825
4826 err = maybe_insert_hole(root, inode, cur_offset,
4827 hole_size);
4828 if (err)
4829 break;
4830 btrfs_drop_extent_cache(inode, cur_offset,
4831 cur_offset + hole_size - 1, 0);
4832 hole_em = alloc_extent_map();
4833 if (!hole_em) {
4834 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4835 &BTRFS_I(inode)->runtime_flags);
4836 goto next;
4837 }
4838 hole_em->start = cur_offset;
4839 hole_em->len = hole_size;
4840 hole_em->orig_start = cur_offset;
4841
4842 hole_em->block_start = EXTENT_MAP_HOLE;
4843 hole_em->block_len = 0;
4844 hole_em->orig_block_len = 0;
4845 hole_em->ram_bytes = hole_size;
4846 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4847 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4848 hole_em->generation = root->fs_info->generation;
4849
4850 while (1) {
4851 write_lock(&em_tree->lock);
4852 err = add_extent_mapping(em_tree, hole_em, 1);
4853 write_unlock(&em_tree->lock);
4854 if (err != -EEXIST)
4855 break;
4856 btrfs_drop_extent_cache(inode, cur_offset,
4857 cur_offset +
4858 hole_size - 1, 0);
4859 }
4860 free_extent_map(hole_em);
4861 }
4862 next:
4863 free_extent_map(em);
4864 em = NULL;
4865 cur_offset = last_byte;
4866 if (cur_offset >= block_end)
4867 break;
4868 }
4869 free_extent_map(em);
4870 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4871 GFP_NOFS);
4872 return err;
4873 }
4874
4875 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4876 {
4877 struct btrfs_root *root = BTRFS_I(inode)->root;
4878 struct btrfs_trans_handle *trans;
4879 loff_t oldsize = i_size_read(inode);
4880 loff_t newsize = attr->ia_size;
4881 int mask = attr->ia_valid;
4882 int ret;
4883
4884 /*
4885 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4886 * special case where we need to update the times despite not having
4887 * these flags set. For all other operations the VFS set these flags
4888 * explicitly if it wants a timestamp update.
4889 */
4890 if (newsize != oldsize) {
4891 inode_inc_iversion(inode);
4892 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4893 inode->i_ctime = inode->i_mtime =
4894 current_fs_time(inode->i_sb);
4895 }
4896
4897 if (newsize > oldsize) {
4898 truncate_pagecache(inode, newsize);
4899 /*
4900 * Don't do an expanding truncate while snapshoting is ongoing.
4901 * This is to ensure the snapshot captures a fully consistent
4902 * state of this file - if the snapshot captures this expanding
4903 * truncation, it must capture all writes that happened before
4904 * this truncation.
4905 */
4906 btrfs_wait_for_snapshot_creation(root);
4907 ret = btrfs_cont_expand(inode, oldsize, newsize);
4908 if (ret) {
4909 btrfs_end_write_no_snapshoting(root);
4910 return ret;
4911 }
4912
4913 trans = btrfs_start_transaction(root, 1);
4914 if (IS_ERR(trans)) {
4915 btrfs_end_write_no_snapshoting(root);
4916 return PTR_ERR(trans);
4917 }
4918
4919 i_size_write(inode, newsize);
4920 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4921 ret = btrfs_update_inode(trans, root, inode);
4922 btrfs_end_write_no_snapshoting(root);
4923 btrfs_end_transaction(trans, root);
4924 } else {
4925
4926 /*
4927 * We're truncating a file that used to have good data down to
4928 * zero. Make sure it gets into the ordered flush list so that
4929 * any new writes get down to disk quickly.
4930 */
4931 if (newsize == 0)
4932 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4933 &BTRFS_I(inode)->runtime_flags);
4934
4935 /*
4936 * 1 for the orphan item we're going to add
4937 * 1 for the orphan item deletion.
4938 */
4939 trans = btrfs_start_transaction(root, 2);
4940 if (IS_ERR(trans))
4941 return PTR_ERR(trans);
4942
4943 /*
4944 * We need to do this in case we fail at _any_ point during the
4945 * actual truncate. Once we do the truncate_setsize we could
4946 * invalidate pages which forces any outstanding ordered io to
4947 * be instantly completed which will give us extents that need
4948 * to be truncated. If we fail to get an orphan inode down we
4949 * could have left over extents that were never meant to live,
4950 * so we need to garuntee from this point on that everything
4951 * will be consistent.
4952 */
4953 ret = btrfs_orphan_add(trans, inode);
4954 btrfs_end_transaction(trans, root);
4955 if (ret)
4956 return ret;
4957
4958 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4959 truncate_setsize(inode, newsize);
4960
4961 /* Disable nonlocked read DIO to avoid the end less truncate */
4962 btrfs_inode_block_unlocked_dio(inode);
4963 inode_dio_wait(inode);
4964 btrfs_inode_resume_unlocked_dio(inode);
4965
4966 ret = btrfs_truncate(inode);
4967 if (ret && inode->i_nlink) {
4968 int err;
4969
4970 /*
4971 * failed to truncate, disk_i_size is only adjusted down
4972 * as we remove extents, so it should represent the true
4973 * size of the inode, so reset the in memory size and
4974 * delete our orphan entry.
4975 */
4976 trans = btrfs_join_transaction(root);
4977 if (IS_ERR(trans)) {
4978 btrfs_orphan_del(NULL, inode);
4979 return ret;
4980 }
4981 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4982 err = btrfs_orphan_del(trans, inode);
4983 if (err)
4984 btrfs_abort_transaction(trans, root, err);
4985 btrfs_end_transaction(trans, root);
4986 }
4987 }
4988
4989 return ret;
4990 }
4991
4992 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4993 {
4994 struct inode *inode = d_inode(dentry);
4995 struct btrfs_root *root = BTRFS_I(inode)->root;
4996 int err;
4997
4998 if (btrfs_root_readonly(root))
4999 return -EROFS;
5000
5001 err = inode_change_ok(inode, attr);
5002 if (err)
5003 return err;
5004
5005 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5006 err = btrfs_setsize(inode, attr);
5007 if (err)
5008 return err;
5009 }
5010
5011 if (attr->ia_valid) {
5012 setattr_copy(inode, attr);
5013 inode_inc_iversion(inode);
5014 err = btrfs_dirty_inode(inode);
5015
5016 if (!err && attr->ia_valid & ATTR_MODE)
5017 err = posix_acl_chmod(inode, inode->i_mode);
5018 }
5019
5020 return err;
5021 }
5022
5023 /*
5024 * While truncating the inode pages during eviction, we get the VFS calling
5025 * btrfs_invalidatepage() against each page of the inode. This is slow because
5026 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5027 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5028 * extent_state structures over and over, wasting lots of time.
5029 *
5030 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5031 * those expensive operations on a per page basis and do only the ordered io
5032 * finishing, while we release here the extent_map and extent_state structures,
5033 * without the excessive merging and splitting.
5034 */
5035 static void evict_inode_truncate_pages(struct inode *inode)
5036 {
5037 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5038 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5039 struct rb_node *node;
5040
5041 ASSERT(inode->i_state & I_FREEING);
5042 truncate_inode_pages_final(&inode->i_data);
5043
5044 write_lock(&map_tree->lock);
5045 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5046 struct extent_map *em;
5047
5048 node = rb_first(&map_tree->map);
5049 em = rb_entry(node, struct extent_map, rb_node);
5050 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5051 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5052 remove_extent_mapping(map_tree, em);
5053 free_extent_map(em);
5054 if (need_resched()) {
5055 write_unlock(&map_tree->lock);
5056 cond_resched();
5057 write_lock(&map_tree->lock);
5058 }
5059 }
5060 write_unlock(&map_tree->lock);
5061
5062 /*
5063 * Keep looping until we have no more ranges in the io tree.
5064 * We can have ongoing bios started by readpages (called from readahead)
5065 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5066 * still in progress (unlocked the pages in the bio but did not yet
5067 * unlocked the ranges in the io tree). Therefore this means some
5068 * ranges can still be locked and eviction started because before
5069 * submitting those bios, which are executed by a separate task (work
5070 * queue kthread), inode references (inode->i_count) were not taken
5071 * (which would be dropped in the end io callback of each bio).
5072 * Therefore here we effectively end up waiting for those bios and
5073 * anyone else holding locked ranges without having bumped the inode's
5074 * reference count - if we don't do it, when they access the inode's
5075 * io_tree to unlock a range it may be too late, leading to an
5076 * use-after-free issue.
5077 */
5078 spin_lock(&io_tree->lock);
5079 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5080 struct extent_state *state;
5081 struct extent_state *cached_state = NULL;
5082 u64 start;
5083 u64 end;
5084
5085 node = rb_first(&io_tree->state);
5086 state = rb_entry(node, struct extent_state, rb_node);
5087 start = state->start;
5088 end = state->end;
5089 spin_unlock(&io_tree->lock);
5090
5091 lock_extent_bits(io_tree, start, end, &cached_state);
5092
5093 /*
5094 * If still has DELALLOC flag, the extent didn't reach disk,
5095 * and its reserved space won't be freed by delayed_ref.
5096 * So we need to free its reserved space here.
5097 * (Refer to comment in btrfs_invalidatepage, case 2)
5098 *
5099 * Note, end is the bytenr of last byte, so we need + 1 here.
5100 */
5101 if (state->state & EXTENT_DELALLOC)
5102 btrfs_qgroup_free_data(inode, start, end - start + 1);
5103
5104 clear_extent_bit(io_tree, start, end,
5105 EXTENT_LOCKED | EXTENT_DIRTY |
5106 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5107 EXTENT_DEFRAG, 1, 1,
5108 &cached_state, GFP_NOFS);
5109
5110 cond_resched();
5111 spin_lock(&io_tree->lock);
5112 }
5113 spin_unlock(&io_tree->lock);
5114 }
5115
5116 void btrfs_evict_inode(struct inode *inode)
5117 {
5118 struct btrfs_trans_handle *trans;
5119 struct btrfs_root *root = BTRFS_I(inode)->root;
5120 struct btrfs_block_rsv *rsv, *global_rsv;
5121 int steal_from_global = 0;
5122 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5123 int ret;
5124
5125 trace_btrfs_inode_evict(inode);
5126
5127 evict_inode_truncate_pages(inode);
5128
5129 if (inode->i_nlink &&
5130 ((btrfs_root_refs(&root->root_item) != 0 &&
5131 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5132 btrfs_is_free_space_inode(inode)))
5133 goto no_delete;
5134
5135 if (is_bad_inode(inode)) {
5136 btrfs_orphan_del(NULL, inode);
5137 goto no_delete;
5138 }
5139 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5140 if (!special_file(inode->i_mode))
5141 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5142
5143 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5144
5145 if (root->fs_info->log_root_recovering) {
5146 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5147 &BTRFS_I(inode)->runtime_flags));
5148 goto no_delete;
5149 }
5150
5151 if (inode->i_nlink > 0) {
5152 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5153 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5154 goto no_delete;
5155 }
5156
5157 ret = btrfs_commit_inode_delayed_inode(inode);
5158 if (ret) {
5159 btrfs_orphan_del(NULL, inode);
5160 goto no_delete;
5161 }
5162
5163 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5164 if (!rsv) {
5165 btrfs_orphan_del(NULL, inode);
5166 goto no_delete;
5167 }
5168 rsv->size = min_size;
5169 rsv->failfast = 1;
5170 global_rsv = &root->fs_info->global_block_rsv;
5171
5172 btrfs_i_size_write(inode, 0);
5173
5174 /*
5175 * This is a bit simpler than btrfs_truncate since we've already
5176 * reserved our space for our orphan item in the unlink, so we just
5177 * need to reserve some slack space in case we add bytes and update
5178 * inode item when doing the truncate.
5179 */
5180 while (1) {
5181 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5182 BTRFS_RESERVE_FLUSH_LIMIT);
5183
5184 /*
5185 * Try and steal from the global reserve since we will
5186 * likely not use this space anyway, we want to try as
5187 * hard as possible to get this to work.
5188 */
5189 if (ret)
5190 steal_from_global++;
5191 else
5192 steal_from_global = 0;
5193 ret = 0;
5194
5195 /*
5196 * steal_from_global == 0: we reserved stuff, hooray!
5197 * steal_from_global == 1: we didn't reserve stuff, boo!
5198 * steal_from_global == 2: we've committed, still not a lot of
5199 * room but maybe we'll have room in the global reserve this
5200 * time.
5201 * steal_from_global == 3: abandon all hope!
5202 */
5203 if (steal_from_global > 2) {
5204 btrfs_warn(root->fs_info,
5205 "Could not get space for a delete, will truncate on mount %d",
5206 ret);
5207 btrfs_orphan_del(NULL, inode);
5208 btrfs_free_block_rsv(root, rsv);
5209 goto no_delete;
5210 }
5211
5212 trans = btrfs_join_transaction(root);
5213 if (IS_ERR(trans)) {
5214 btrfs_orphan_del(NULL, inode);
5215 btrfs_free_block_rsv(root, rsv);
5216 goto no_delete;
5217 }
5218
5219 /*
5220 * We can't just steal from the global reserve, we need tomake
5221 * sure there is room to do it, if not we need to commit and try
5222 * again.
5223 */
5224 if (steal_from_global) {
5225 if (!btrfs_check_space_for_delayed_refs(trans, root))
5226 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5227 min_size);
5228 else
5229 ret = -ENOSPC;
5230 }
5231
5232 /*
5233 * Couldn't steal from the global reserve, we have too much
5234 * pending stuff built up, commit the transaction and try it
5235 * again.
5236 */
5237 if (ret) {
5238 ret = btrfs_commit_transaction(trans, root);
5239 if (ret) {
5240 btrfs_orphan_del(NULL, inode);
5241 btrfs_free_block_rsv(root, rsv);
5242 goto no_delete;
5243 }
5244 continue;
5245 } else {
5246 steal_from_global = 0;
5247 }
5248
5249 trans->block_rsv = rsv;
5250
5251 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5252 if (ret != -ENOSPC && ret != -EAGAIN)
5253 break;
5254
5255 trans->block_rsv = &root->fs_info->trans_block_rsv;
5256 btrfs_end_transaction(trans, root);
5257 trans = NULL;
5258 btrfs_btree_balance_dirty(root);
5259 }
5260
5261 btrfs_free_block_rsv(root, rsv);
5262
5263 /*
5264 * Errors here aren't a big deal, it just means we leave orphan items
5265 * in the tree. They will be cleaned up on the next mount.
5266 */
5267 if (ret == 0) {
5268 trans->block_rsv = root->orphan_block_rsv;
5269 btrfs_orphan_del(trans, inode);
5270 } else {
5271 btrfs_orphan_del(NULL, inode);
5272 }
5273
5274 trans->block_rsv = &root->fs_info->trans_block_rsv;
5275 if (!(root == root->fs_info->tree_root ||
5276 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5277 btrfs_return_ino(root, btrfs_ino(inode));
5278
5279 btrfs_end_transaction(trans, root);
5280 btrfs_btree_balance_dirty(root);
5281 no_delete:
5282 btrfs_remove_delayed_node(inode);
5283 clear_inode(inode);
5284 }
5285
5286 /*
5287 * this returns the key found in the dir entry in the location pointer.
5288 * If no dir entries were found, location->objectid is 0.
5289 */
5290 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5291 struct btrfs_key *location)
5292 {
5293 const char *name = dentry->d_name.name;
5294 int namelen = dentry->d_name.len;
5295 struct btrfs_dir_item *di;
5296 struct btrfs_path *path;
5297 struct btrfs_root *root = BTRFS_I(dir)->root;
5298 int ret = 0;
5299
5300 path = btrfs_alloc_path();
5301 if (!path)
5302 return -ENOMEM;
5303
5304 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5305 namelen, 0);
5306 if (IS_ERR(di))
5307 ret = PTR_ERR(di);
5308
5309 if (IS_ERR_OR_NULL(di))
5310 goto out_err;
5311
5312 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5313 out:
5314 btrfs_free_path(path);
5315 return ret;
5316 out_err:
5317 location->objectid = 0;
5318 goto out;
5319 }
5320
5321 /*
5322 * when we hit a tree root in a directory, the btrfs part of the inode
5323 * needs to be changed to reflect the root directory of the tree root. This
5324 * is kind of like crossing a mount point.
5325 */
5326 static int fixup_tree_root_location(struct btrfs_root *root,
5327 struct inode *dir,
5328 struct dentry *dentry,
5329 struct btrfs_key *location,
5330 struct btrfs_root **sub_root)
5331 {
5332 struct btrfs_path *path;
5333 struct btrfs_root *new_root;
5334 struct btrfs_root_ref *ref;
5335 struct extent_buffer *leaf;
5336 struct btrfs_key key;
5337 int ret;
5338 int err = 0;
5339
5340 path = btrfs_alloc_path();
5341 if (!path) {
5342 err = -ENOMEM;
5343 goto out;
5344 }
5345
5346 err = -ENOENT;
5347 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5348 key.type = BTRFS_ROOT_REF_KEY;
5349 key.offset = location->objectid;
5350
5351 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5352 0, 0);
5353 if (ret) {
5354 if (ret < 0)
5355 err = ret;
5356 goto out;
5357 }
5358
5359 leaf = path->nodes[0];
5360 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5361 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5362 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5363 goto out;
5364
5365 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5366 (unsigned long)(ref + 1),
5367 dentry->d_name.len);
5368 if (ret)
5369 goto out;
5370
5371 btrfs_release_path(path);
5372
5373 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5374 if (IS_ERR(new_root)) {
5375 err = PTR_ERR(new_root);
5376 goto out;
5377 }
5378
5379 *sub_root = new_root;
5380 location->objectid = btrfs_root_dirid(&new_root->root_item);
5381 location->type = BTRFS_INODE_ITEM_KEY;
5382 location->offset = 0;
5383 err = 0;
5384 out:
5385 btrfs_free_path(path);
5386 return err;
5387 }
5388
5389 static void inode_tree_add(struct inode *inode)
5390 {
5391 struct btrfs_root *root = BTRFS_I(inode)->root;
5392 struct btrfs_inode *entry;
5393 struct rb_node **p;
5394 struct rb_node *parent;
5395 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5396 u64 ino = btrfs_ino(inode);
5397
5398 if (inode_unhashed(inode))
5399 return;
5400 parent = NULL;
5401 spin_lock(&root->inode_lock);
5402 p = &root->inode_tree.rb_node;
5403 while (*p) {
5404 parent = *p;
5405 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5406
5407 if (ino < btrfs_ino(&entry->vfs_inode))
5408 p = &parent->rb_left;
5409 else if (ino > btrfs_ino(&entry->vfs_inode))
5410 p = &parent->rb_right;
5411 else {
5412 WARN_ON(!(entry->vfs_inode.i_state &
5413 (I_WILL_FREE | I_FREEING)));
5414 rb_replace_node(parent, new, &root->inode_tree);
5415 RB_CLEAR_NODE(parent);
5416 spin_unlock(&root->inode_lock);
5417 return;
5418 }
5419 }
5420 rb_link_node(new, parent, p);
5421 rb_insert_color(new, &root->inode_tree);
5422 spin_unlock(&root->inode_lock);
5423 }
5424
5425 static void inode_tree_del(struct inode *inode)
5426 {
5427 struct btrfs_root *root = BTRFS_I(inode)->root;
5428 int empty = 0;
5429
5430 spin_lock(&root->inode_lock);
5431 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5432 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5433 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5434 empty = RB_EMPTY_ROOT(&root->inode_tree);
5435 }
5436 spin_unlock(&root->inode_lock);
5437
5438 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5439 synchronize_srcu(&root->fs_info->subvol_srcu);
5440 spin_lock(&root->inode_lock);
5441 empty = RB_EMPTY_ROOT(&root->inode_tree);
5442 spin_unlock(&root->inode_lock);
5443 if (empty)
5444 btrfs_add_dead_root(root);
5445 }
5446 }
5447
5448 void btrfs_invalidate_inodes(struct btrfs_root *root)
5449 {
5450 struct rb_node *node;
5451 struct rb_node *prev;
5452 struct btrfs_inode *entry;
5453 struct inode *inode;
5454 u64 objectid = 0;
5455
5456 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5457 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5458
5459 spin_lock(&root->inode_lock);
5460 again:
5461 node = root->inode_tree.rb_node;
5462 prev = NULL;
5463 while (node) {
5464 prev = node;
5465 entry = rb_entry(node, struct btrfs_inode, rb_node);
5466
5467 if (objectid < btrfs_ino(&entry->vfs_inode))
5468 node = node->rb_left;
5469 else if (objectid > btrfs_ino(&entry->vfs_inode))
5470 node = node->rb_right;
5471 else
5472 break;
5473 }
5474 if (!node) {
5475 while (prev) {
5476 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5477 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5478 node = prev;
5479 break;
5480 }
5481 prev = rb_next(prev);
5482 }
5483 }
5484 while (node) {
5485 entry = rb_entry(node, struct btrfs_inode, rb_node);
5486 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5487 inode = igrab(&entry->vfs_inode);
5488 if (inode) {
5489 spin_unlock(&root->inode_lock);
5490 if (atomic_read(&inode->i_count) > 1)
5491 d_prune_aliases(inode);
5492 /*
5493 * btrfs_drop_inode will have it removed from
5494 * the inode cache when its usage count
5495 * hits zero.
5496 */
5497 iput(inode);
5498 cond_resched();
5499 spin_lock(&root->inode_lock);
5500 goto again;
5501 }
5502
5503 if (cond_resched_lock(&root->inode_lock))
5504 goto again;
5505
5506 node = rb_next(node);
5507 }
5508 spin_unlock(&root->inode_lock);
5509 }
5510
5511 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5512 {
5513 struct btrfs_iget_args *args = p;
5514 inode->i_ino = args->location->objectid;
5515 memcpy(&BTRFS_I(inode)->location, args->location,
5516 sizeof(*args->location));
5517 BTRFS_I(inode)->root = args->root;
5518 return 0;
5519 }
5520
5521 static int btrfs_find_actor(struct inode *inode, void *opaque)
5522 {
5523 struct btrfs_iget_args *args = opaque;
5524 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5525 args->root == BTRFS_I(inode)->root;
5526 }
5527
5528 static struct inode *btrfs_iget_locked(struct super_block *s,
5529 struct btrfs_key *location,
5530 struct btrfs_root *root)
5531 {
5532 struct inode *inode;
5533 struct btrfs_iget_args args;
5534 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5535
5536 args.location = location;
5537 args.root = root;
5538
5539 inode = iget5_locked(s, hashval, btrfs_find_actor,
5540 btrfs_init_locked_inode,
5541 (void *)&args);
5542 return inode;
5543 }
5544
5545 /* Get an inode object given its location and corresponding root.
5546 * Returns in *is_new if the inode was read from disk
5547 */
5548 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5549 struct btrfs_root *root, int *new)
5550 {
5551 struct inode *inode;
5552
5553 inode = btrfs_iget_locked(s, location, root);
5554 if (!inode)
5555 return ERR_PTR(-ENOMEM);
5556
5557 if (inode->i_state & I_NEW) {
5558 btrfs_read_locked_inode(inode);
5559 if (!is_bad_inode(inode)) {
5560 inode_tree_add(inode);
5561 unlock_new_inode(inode);
5562 if (new)
5563 *new = 1;
5564 } else {
5565 unlock_new_inode(inode);
5566 iput(inode);
5567 inode = ERR_PTR(-ESTALE);
5568 }
5569 }
5570
5571 return inode;
5572 }
5573
5574 static struct inode *new_simple_dir(struct super_block *s,
5575 struct btrfs_key *key,
5576 struct btrfs_root *root)
5577 {
5578 struct inode *inode = new_inode(s);
5579
5580 if (!inode)
5581 return ERR_PTR(-ENOMEM);
5582
5583 BTRFS_I(inode)->root = root;
5584 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5585 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5586
5587 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5588 inode->i_op = &btrfs_dir_ro_inode_operations;
5589 inode->i_fop = &simple_dir_operations;
5590 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5591 inode->i_mtime = CURRENT_TIME;
5592 inode->i_atime = inode->i_mtime;
5593 inode->i_ctime = inode->i_mtime;
5594 BTRFS_I(inode)->i_otime = inode->i_mtime;
5595
5596 return inode;
5597 }
5598
5599 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5600 {
5601 struct inode *inode;
5602 struct btrfs_root *root = BTRFS_I(dir)->root;
5603 struct btrfs_root *sub_root = root;
5604 struct btrfs_key location;
5605 int index;
5606 int ret = 0;
5607
5608 if (dentry->d_name.len > BTRFS_NAME_LEN)
5609 return ERR_PTR(-ENAMETOOLONG);
5610
5611 ret = btrfs_inode_by_name(dir, dentry, &location);
5612 if (ret < 0)
5613 return ERR_PTR(ret);
5614
5615 if (location.objectid == 0)
5616 return ERR_PTR(-ENOENT);
5617
5618 if (location.type == BTRFS_INODE_ITEM_KEY) {
5619 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5620 return inode;
5621 }
5622
5623 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5624
5625 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5626 ret = fixup_tree_root_location(root, dir, dentry,
5627 &location, &sub_root);
5628 if (ret < 0) {
5629 if (ret != -ENOENT)
5630 inode = ERR_PTR(ret);
5631 else
5632 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5633 } else {
5634 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5635 }
5636 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5637
5638 if (!IS_ERR(inode) && root != sub_root) {
5639 down_read(&root->fs_info->cleanup_work_sem);
5640 if (!(inode->i_sb->s_flags & MS_RDONLY))
5641 ret = btrfs_orphan_cleanup(sub_root);
5642 up_read(&root->fs_info->cleanup_work_sem);
5643 if (ret) {
5644 iput(inode);
5645 inode = ERR_PTR(ret);
5646 }
5647 }
5648
5649 return inode;
5650 }
5651
5652 static int btrfs_dentry_delete(const struct dentry *dentry)
5653 {
5654 struct btrfs_root *root;
5655 struct inode *inode = d_inode(dentry);
5656
5657 if (!inode && !IS_ROOT(dentry))
5658 inode = d_inode(dentry->d_parent);
5659
5660 if (inode) {
5661 root = BTRFS_I(inode)->root;
5662 if (btrfs_root_refs(&root->root_item) == 0)
5663 return 1;
5664
5665 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5666 return 1;
5667 }
5668 return 0;
5669 }
5670
5671 static void btrfs_dentry_release(struct dentry *dentry)
5672 {
5673 kfree(dentry->d_fsdata);
5674 }
5675
5676 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5677 unsigned int flags)
5678 {
5679 struct inode *inode;
5680
5681 inode = btrfs_lookup_dentry(dir, dentry);
5682 if (IS_ERR(inode)) {
5683 if (PTR_ERR(inode) == -ENOENT)
5684 inode = NULL;
5685 else
5686 return ERR_CAST(inode);
5687 }
5688
5689 return d_splice_alias(inode, dentry);
5690 }
5691
5692 unsigned char btrfs_filetype_table[] = {
5693 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5694 };
5695
5696 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5697 {
5698 struct inode *inode = file_inode(file);
5699 struct btrfs_root *root = BTRFS_I(inode)->root;
5700 struct btrfs_item *item;
5701 struct btrfs_dir_item *di;
5702 struct btrfs_key key;
5703 struct btrfs_key found_key;
5704 struct btrfs_path *path;
5705 struct list_head ins_list;
5706 struct list_head del_list;
5707 int ret;
5708 struct extent_buffer *leaf;
5709 int slot;
5710 unsigned char d_type;
5711 int over = 0;
5712 u32 di_cur;
5713 u32 di_total;
5714 u32 di_len;
5715 int key_type = BTRFS_DIR_INDEX_KEY;
5716 char tmp_name[32];
5717 char *name_ptr;
5718 int name_len;
5719 int is_curr = 0; /* ctx->pos points to the current index? */
5720 bool emitted;
5721
5722 /* FIXME, use a real flag for deciding about the key type */
5723 if (root->fs_info->tree_root == root)
5724 key_type = BTRFS_DIR_ITEM_KEY;
5725
5726 if (!dir_emit_dots(file, ctx))
5727 return 0;
5728
5729 path = btrfs_alloc_path();
5730 if (!path)
5731 return -ENOMEM;
5732
5733 path->reada = READA_FORWARD;
5734
5735 if (key_type == BTRFS_DIR_INDEX_KEY) {
5736 INIT_LIST_HEAD(&ins_list);
5737 INIT_LIST_HEAD(&del_list);
5738 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5739 }
5740
5741 key.type = key_type;
5742 key.offset = ctx->pos;
5743 key.objectid = btrfs_ino(inode);
5744
5745 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5746 if (ret < 0)
5747 goto err;
5748
5749 emitted = false;
5750 while (1) {
5751 leaf = path->nodes[0];
5752 slot = path->slots[0];
5753 if (slot >= btrfs_header_nritems(leaf)) {
5754 ret = btrfs_next_leaf(root, path);
5755 if (ret < 0)
5756 goto err;
5757 else if (ret > 0)
5758 break;
5759 continue;
5760 }
5761
5762 item = btrfs_item_nr(slot);
5763 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5764
5765 if (found_key.objectid != key.objectid)
5766 break;
5767 if (found_key.type != key_type)
5768 break;
5769 if (found_key.offset < ctx->pos)
5770 goto next;
5771 if (key_type == BTRFS_DIR_INDEX_KEY &&
5772 btrfs_should_delete_dir_index(&del_list,
5773 found_key.offset))
5774 goto next;
5775
5776 ctx->pos = found_key.offset;
5777 is_curr = 1;
5778
5779 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5780 di_cur = 0;
5781 di_total = btrfs_item_size(leaf, item);
5782
5783 while (di_cur < di_total) {
5784 struct btrfs_key location;
5785
5786 if (verify_dir_item(root, leaf, di))
5787 break;
5788
5789 name_len = btrfs_dir_name_len(leaf, di);
5790 if (name_len <= sizeof(tmp_name)) {
5791 name_ptr = tmp_name;
5792 } else {
5793 name_ptr = kmalloc(name_len, GFP_NOFS);
5794 if (!name_ptr) {
5795 ret = -ENOMEM;
5796 goto err;
5797 }
5798 }
5799 read_extent_buffer(leaf, name_ptr,
5800 (unsigned long)(di + 1), name_len);
5801
5802 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5803 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5804
5805
5806 /* is this a reference to our own snapshot? If so
5807 * skip it.
5808 *
5809 * In contrast to old kernels, we insert the snapshot's
5810 * dir item and dir index after it has been created, so
5811 * we won't find a reference to our own snapshot. We
5812 * still keep the following code for backward
5813 * compatibility.
5814 */
5815 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5816 location.objectid == root->root_key.objectid) {
5817 over = 0;
5818 goto skip;
5819 }
5820 over = !dir_emit(ctx, name_ptr, name_len,
5821 location.objectid, d_type);
5822
5823 skip:
5824 if (name_ptr != tmp_name)
5825 kfree(name_ptr);
5826
5827 if (over)
5828 goto nopos;
5829 emitted = true;
5830 di_len = btrfs_dir_name_len(leaf, di) +
5831 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5832 di_cur += di_len;
5833 di = (struct btrfs_dir_item *)((char *)di + di_len);
5834 }
5835 next:
5836 path->slots[0]++;
5837 }
5838
5839 if (key_type == BTRFS_DIR_INDEX_KEY) {
5840 if (is_curr)
5841 ctx->pos++;
5842 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5843 if (ret)
5844 goto nopos;
5845 }
5846
5847 /*
5848 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5849 * it was was set to the termination value in previous call. We assume
5850 * that "." and ".." were emitted if we reach this point and set the
5851 * termination value as well for an empty directory.
5852 */
5853 if (ctx->pos > 2 && !emitted)
5854 goto nopos;
5855
5856 /* Reached end of directory/root. Bump pos past the last item. */
5857 ctx->pos++;
5858
5859 /*
5860 * Stop new entries from being returned after we return the last
5861 * entry.
5862 *
5863 * New directory entries are assigned a strictly increasing
5864 * offset. This means that new entries created during readdir
5865 * are *guaranteed* to be seen in the future by that readdir.
5866 * This has broken buggy programs which operate on names as
5867 * they're returned by readdir. Until we re-use freed offsets
5868 * we have this hack to stop new entries from being returned
5869 * under the assumption that they'll never reach this huge
5870 * offset.
5871 *
5872 * This is being careful not to overflow 32bit loff_t unless the
5873 * last entry requires it because doing so has broken 32bit apps
5874 * in the past.
5875 */
5876 if (key_type == BTRFS_DIR_INDEX_KEY) {
5877 if (ctx->pos >= INT_MAX)
5878 ctx->pos = LLONG_MAX;
5879 else
5880 ctx->pos = INT_MAX;
5881 }
5882 nopos:
5883 ret = 0;
5884 err:
5885 if (key_type == BTRFS_DIR_INDEX_KEY)
5886 btrfs_put_delayed_items(&ins_list, &del_list);
5887 btrfs_free_path(path);
5888 return ret;
5889 }
5890
5891 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5892 {
5893 struct btrfs_root *root = BTRFS_I(inode)->root;
5894 struct btrfs_trans_handle *trans;
5895 int ret = 0;
5896 bool nolock = false;
5897
5898 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5899 return 0;
5900
5901 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5902 nolock = true;
5903
5904 if (wbc->sync_mode == WB_SYNC_ALL) {
5905 if (nolock)
5906 trans = btrfs_join_transaction_nolock(root);
5907 else
5908 trans = btrfs_join_transaction(root);
5909 if (IS_ERR(trans))
5910 return PTR_ERR(trans);
5911 ret = btrfs_commit_transaction(trans, root);
5912 }
5913 return ret;
5914 }
5915
5916 /*
5917 * This is somewhat expensive, updating the tree every time the
5918 * inode changes. But, it is most likely to find the inode in cache.
5919 * FIXME, needs more benchmarking...there are no reasons other than performance
5920 * to keep or drop this code.
5921 */
5922 static int btrfs_dirty_inode(struct inode *inode)
5923 {
5924 struct btrfs_root *root = BTRFS_I(inode)->root;
5925 struct btrfs_trans_handle *trans;
5926 int ret;
5927
5928 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5929 return 0;
5930
5931 trans = btrfs_join_transaction(root);
5932 if (IS_ERR(trans))
5933 return PTR_ERR(trans);
5934
5935 ret = btrfs_update_inode(trans, root, inode);
5936 if (ret && ret == -ENOSPC) {
5937 /* whoops, lets try again with the full transaction */
5938 btrfs_end_transaction(trans, root);
5939 trans = btrfs_start_transaction(root, 1);
5940 if (IS_ERR(trans))
5941 return PTR_ERR(trans);
5942
5943 ret = btrfs_update_inode(trans, root, inode);
5944 }
5945 btrfs_end_transaction(trans, root);
5946 if (BTRFS_I(inode)->delayed_node)
5947 btrfs_balance_delayed_items(root);
5948
5949 return ret;
5950 }
5951
5952 /*
5953 * This is a copy of file_update_time. We need this so we can return error on
5954 * ENOSPC for updating the inode in the case of file write and mmap writes.
5955 */
5956 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5957 int flags)
5958 {
5959 struct btrfs_root *root = BTRFS_I(inode)->root;
5960
5961 if (btrfs_root_readonly(root))
5962 return -EROFS;
5963
5964 if (flags & S_VERSION)
5965 inode_inc_iversion(inode);
5966 if (flags & S_CTIME)
5967 inode->i_ctime = *now;
5968 if (flags & S_MTIME)
5969 inode->i_mtime = *now;
5970 if (flags & S_ATIME)
5971 inode->i_atime = *now;
5972 return btrfs_dirty_inode(inode);
5973 }
5974
5975 /*
5976 * find the highest existing sequence number in a directory
5977 * and then set the in-memory index_cnt variable to reflect
5978 * free sequence numbers
5979 */
5980 static int btrfs_set_inode_index_count(struct inode *inode)
5981 {
5982 struct btrfs_root *root = BTRFS_I(inode)->root;
5983 struct btrfs_key key, found_key;
5984 struct btrfs_path *path;
5985 struct extent_buffer *leaf;
5986 int ret;
5987
5988 key.objectid = btrfs_ino(inode);
5989 key.type = BTRFS_DIR_INDEX_KEY;
5990 key.offset = (u64)-1;
5991
5992 path = btrfs_alloc_path();
5993 if (!path)
5994 return -ENOMEM;
5995
5996 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5997 if (ret < 0)
5998 goto out;
5999 /* FIXME: we should be able to handle this */
6000 if (ret == 0)
6001 goto out;
6002 ret = 0;
6003
6004 /*
6005 * MAGIC NUMBER EXPLANATION:
6006 * since we search a directory based on f_pos we have to start at 2
6007 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6008 * else has to start at 2
6009 */
6010 if (path->slots[0] == 0) {
6011 BTRFS_I(inode)->index_cnt = 2;
6012 goto out;
6013 }
6014
6015 path->slots[0]--;
6016
6017 leaf = path->nodes[0];
6018 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6019
6020 if (found_key.objectid != btrfs_ino(inode) ||
6021 found_key.type != BTRFS_DIR_INDEX_KEY) {
6022 BTRFS_I(inode)->index_cnt = 2;
6023 goto out;
6024 }
6025
6026 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6027 out:
6028 btrfs_free_path(path);
6029 return ret;
6030 }
6031
6032 /*
6033 * helper to find a free sequence number in a given directory. This current
6034 * code is very simple, later versions will do smarter things in the btree
6035 */
6036 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6037 {
6038 int ret = 0;
6039
6040 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6041 ret = btrfs_inode_delayed_dir_index_count(dir);
6042 if (ret) {
6043 ret = btrfs_set_inode_index_count(dir);
6044 if (ret)
6045 return ret;
6046 }
6047 }
6048
6049 *index = BTRFS_I(dir)->index_cnt;
6050 BTRFS_I(dir)->index_cnt++;
6051
6052 return ret;
6053 }
6054
6055 static int btrfs_insert_inode_locked(struct inode *inode)
6056 {
6057 struct btrfs_iget_args args;
6058 args.location = &BTRFS_I(inode)->location;
6059 args.root = BTRFS_I(inode)->root;
6060
6061 return insert_inode_locked4(inode,
6062 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6063 btrfs_find_actor, &args);
6064 }
6065
6066 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6067 struct btrfs_root *root,
6068 struct inode *dir,
6069 const char *name, int name_len,
6070 u64 ref_objectid, u64 objectid,
6071 umode_t mode, u64 *index)
6072 {
6073 struct inode *inode;
6074 struct btrfs_inode_item *inode_item;
6075 struct btrfs_key *location;
6076 struct btrfs_path *path;
6077 struct btrfs_inode_ref *ref;
6078 struct btrfs_key key[2];
6079 u32 sizes[2];
6080 int nitems = name ? 2 : 1;
6081 unsigned long ptr;
6082 int ret;
6083
6084 path = btrfs_alloc_path();
6085 if (!path)
6086 return ERR_PTR(-ENOMEM);
6087
6088 inode = new_inode(root->fs_info->sb);
6089 if (!inode) {
6090 btrfs_free_path(path);
6091 return ERR_PTR(-ENOMEM);
6092 }
6093
6094 /*
6095 * O_TMPFILE, set link count to 0, so that after this point,
6096 * we fill in an inode item with the correct link count.
6097 */
6098 if (!name)
6099 set_nlink(inode, 0);
6100
6101 /*
6102 * we have to initialize this early, so we can reclaim the inode
6103 * number if we fail afterwards in this function.
6104 */
6105 inode->i_ino = objectid;
6106
6107 if (dir && name) {
6108 trace_btrfs_inode_request(dir);
6109
6110 ret = btrfs_set_inode_index(dir, index);
6111 if (ret) {
6112 btrfs_free_path(path);
6113 iput(inode);
6114 return ERR_PTR(ret);
6115 }
6116 } else if (dir) {
6117 *index = 0;
6118 }
6119 /*
6120 * index_cnt is ignored for everything but a dir,
6121 * btrfs_get_inode_index_count has an explanation for the magic
6122 * number
6123 */
6124 BTRFS_I(inode)->index_cnt = 2;
6125 BTRFS_I(inode)->dir_index = *index;
6126 BTRFS_I(inode)->root = root;
6127 BTRFS_I(inode)->generation = trans->transid;
6128 inode->i_generation = BTRFS_I(inode)->generation;
6129
6130 /*
6131 * We could have gotten an inode number from somebody who was fsynced
6132 * and then removed in this same transaction, so let's just set full
6133 * sync since it will be a full sync anyway and this will blow away the
6134 * old info in the log.
6135 */
6136 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6137
6138 key[0].objectid = objectid;
6139 key[0].type = BTRFS_INODE_ITEM_KEY;
6140 key[0].offset = 0;
6141
6142 sizes[0] = sizeof(struct btrfs_inode_item);
6143
6144 if (name) {
6145 /*
6146 * Start new inodes with an inode_ref. This is slightly more
6147 * efficient for small numbers of hard links since they will
6148 * be packed into one item. Extended refs will kick in if we
6149 * add more hard links than can fit in the ref item.
6150 */
6151 key[1].objectid = objectid;
6152 key[1].type = BTRFS_INODE_REF_KEY;
6153 key[1].offset = ref_objectid;
6154
6155 sizes[1] = name_len + sizeof(*ref);
6156 }
6157
6158 location = &BTRFS_I(inode)->location;
6159 location->objectid = objectid;
6160 location->offset = 0;
6161 location->type = BTRFS_INODE_ITEM_KEY;
6162
6163 ret = btrfs_insert_inode_locked(inode);
6164 if (ret < 0)
6165 goto fail;
6166
6167 path->leave_spinning = 1;
6168 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6169 if (ret != 0)
6170 goto fail_unlock;
6171
6172 inode_init_owner(inode, dir, mode);
6173 inode_set_bytes(inode, 0);
6174
6175 inode->i_mtime = CURRENT_TIME;
6176 inode->i_atime = inode->i_mtime;
6177 inode->i_ctime = inode->i_mtime;
6178 BTRFS_I(inode)->i_otime = inode->i_mtime;
6179
6180 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6181 struct btrfs_inode_item);
6182 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6183 sizeof(*inode_item));
6184 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6185
6186 if (name) {
6187 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6188 struct btrfs_inode_ref);
6189 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6190 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6191 ptr = (unsigned long)(ref + 1);
6192 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6193 }
6194
6195 btrfs_mark_buffer_dirty(path->nodes[0]);
6196 btrfs_free_path(path);
6197
6198 btrfs_inherit_iflags(inode, dir);
6199
6200 if (S_ISREG(mode)) {
6201 if (btrfs_test_opt(root, NODATASUM))
6202 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6203 if (btrfs_test_opt(root, NODATACOW))
6204 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6205 BTRFS_INODE_NODATASUM;
6206 }
6207
6208 inode_tree_add(inode);
6209
6210 trace_btrfs_inode_new(inode);
6211 btrfs_set_inode_last_trans(trans, inode);
6212
6213 btrfs_update_root_times(trans, root);
6214
6215 ret = btrfs_inode_inherit_props(trans, inode, dir);
6216 if (ret)
6217 btrfs_err(root->fs_info,
6218 "error inheriting props for ino %llu (root %llu): %d",
6219 btrfs_ino(inode), root->root_key.objectid, ret);
6220
6221 return inode;
6222
6223 fail_unlock:
6224 unlock_new_inode(inode);
6225 fail:
6226 if (dir && name)
6227 BTRFS_I(dir)->index_cnt--;
6228 btrfs_free_path(path);
6229 iput(inode);
6230 return ERR_PTR(ret);
6231 }
6232
6233 static inline u8 btrfs_inode_type(struct inode *inode)
6234 {
6235 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6236 }
6237
6238 /*
6239 * utility function to add 'inode' into 'parent_inode' with
6240 * a give name and a given sequence number.
6241 * if 'add_backref' is true, also insert a backref from the
6242 * inode to the parent directory.
6243 */
6244 int btrfs_add_link(struct btrfs_trans_handle *trans,
6245 struct inode *parent_inode, struct inode *inode,
6246 const char *name, int name_len, int add_backref, u64 index)
6247 {
6248 int ret = 0;
6249 struct btrfs_key key;
6250 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6251 u64 ino = btrfs_ino(inode);
6252 u64 parent_ino = btrfs_ino(parent_inode);
6253
6254 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6255 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6256 } else {
6257 key.objectid = ino;
6258 key.type = BTRFS_INODE_ITEM_KEY;
6259 key.offset = 0;
6260 }
6261
6262 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6263 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6264 key.objectid, root->root_key.objectid,
6265 parent_ino, index, name, name_len);
6266 } else if (add_backref) {
6267 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6268 parent_ino, index);
6269 }
6270
6271 /* Nothing to clean up yet */
6272 if (ret)
6273 return ret;
6274
6275 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6276 parent_inode, &key,
6277 btrfs_inode_type(inode), index);
6278 if (ret == -EEXIST || ret == -EOVERFLOW)
6279 goto fail_dir_item;
6280 else if (ret) {
6281 btrfs_abort_transaction(trans, root, ret);
6282 return ret;
6283 }
6284
6285 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6286 name_len * 2);
6287 inode_inc_iversion(parent_inode);
6288 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6289 ret = btrfs_update_inode(trans, root, parent_inode);
6290 if (ret)
6291 btrfs_abort_transaction(trans, root, ret);
6292 return ret;
6293
6294 fail_dir_item:
6295 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6296 u64 local_index;
6297 int err;
6298 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6299 key.objectid, root->root_key.objectid,
6300 parent_ino, &local_index, name, name_len);
6301
6302 } else if (add_backref) {
6303 u64 local_index;
6304 int err;
6305
6306 err = btrfs_del_inode_ref(trans, root, name, name_len,
6307 ino, parent_ino, &local_index);
6308 }
6309 return ret;
6310 }
6311
6312 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6313 struct inode *dir, struct dentry *dentry,
6314 struct inode *inode, int backref, u64 index)
6315 {
6316 int err = btrfs_add_link(trans, dir, inode,
6317 dentry->d_name.name, dentry->d_name.len,
6318 backref, index);
6319 if (err > 0)
6320 err = -EEXIST;
6321 return err;
6322 }
6323
6324 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6325 umode_t mode, dev_t rdev)
6326 {
6327 struct btrfs_trans_handle *trans;
6328 struct btrfs_root *root = BTRFS_I(dir)->root;
6329 struct inode *inode = NULL;
6330 int err;
6331 int drop_inode = 0;
6332 u64 objectid;
6333 u64 index = 0;
6334
6335 /*
6336 * 2 for inode item and ref
6337 * 2 for dir items
6338 * 1 for xattr if selinux is on
6339 */
6340 trans = btrfs_start_transaction(root, 5);
6341 if (IS_ERR(trans))
6342 return PTR_ERR(trans);
6343
6344 err = btrfs_find_free_ino(root, &objectid);
6345 if (err)
6346 goto out_unlock;
6347
6348 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6349 dentry->d_name.len, btrfs_ino(dir), objectid,
6350 mode, &index);
6351 if (IS_ERR(inode)) {
6352 err = PTR_ERR(inode);
6353 goto out_unlock;
6354 }
6355
6356 /*
6357 * If the active LSM wants to access the inode during
6358 * d_instantiate it needs these. Smack checks to see
6359 * if the filesystem supports xattrs by looking at the
6360 * ops vector.
6361 */
6362 inode->i_op = &btrfs_special_inode_operations;
6363 init_special_inode(inode, inode->i_mode, rdev);
6364
6365 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6366 if (err)
6367 goto out_unlock_inode;
6368
6369 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6370 if (err) {
6371 goto out_unlock_inode;
6372 } else {
6373 btrfs_update_inode(trans, root, inode);
6374 unlock_new_inode(inode);
6375 d_instantiate(dentry, inode);
6376 }
6377
6378 out_unlock:
6379 btrfs_end_transaction(trans, root);
6380 btrfs_balance_delayed_items(root);
6381 btrfs_btree_balance_dirty(root);
6382 if (drop_inode) {
6383 inode_dec_link_count(inode);
6384 iput(inode);
6385 }
6386 return err;
6387
6388 out_unlock_inode:
6389 drop_inode = 1;
6390 unlock_new_inode(inode);
6391 goto out_unlock;
6392
6393 }
6394
6395 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6396 umode_t mode, bool excl)
6397 {
6398 struct btrfs_trans_handle *trans;
6399 struct btrfs_root *root = BTRFS_I(dir)->root;
6400 struct inode *inode = NULL;
6401 int drop_inode_on_err = 0;
6402 int err;
6403 u64 objectid;
6404 u64 index = 0;
6405
6406 /*
6407 * 2 for inode item and ref
6408 * 2 for dir items
6409 * 1 for xattr if selinux is on
6410 */
6411 trans = btrfs_start_transaction(root, 5);
6412 if (IS_ERR(trans))
6413 return PTR_ERR(trans);
6414
6415 err = btrfs_find_free_ino(root, &objectid);
6416 if (err)
6417 goto out_unlock;
6418
6419 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6420 dentry->d_name.len, btrfs_ino(dir), objectid,
6421 mode, &index);
6422 if (IS_ERR(inode)) {
6423 err = PTR_ERR(inode);
6424 goto out_unlock;
6425 }
6426 drop_inode_on_err = 1;
6427 /*
6428 * If the active LSM wants to access the inode during
6429 * d_instantiate it needs these. Smack checks to see
6430 * if the filesystem supports xattrs by looking at the
6431 * ops vector.
6432 */
6433 inode->i_fop = &btrfs_file_operations;
6434 inode->i_op = &btrfs_file_inode_operations;
6435 inode->i_mapping->a_ops = &btrfs_aops;
6436
6437 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6438 if (err)
6439 goto out_unlock_inode;
6440
6441 err = btrfs_update_inode(trans, root, inode);
6442 if (err)
6443 goto out_unlock_inode;
6444
6445 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6446 if (err)
6447 goto out_unlock_inode;
6448
6449 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6450 unlock_new_inode(inode);
6451 d_instantiate(dentry, inode);
6452
6453 out_unlock:
6454 btrfs_end_transaction(trans, root);
6455 if (err && drop_inode_on_err) {
6456 inode_dec_link_count(inode);
6457 iput(inode);
6458 }
6459 btrfs_balance_delayed_items(root);
6460 btrfs_btree_balance_dirty(root);
6461 return err;
6462
6463 out_unlock_inode:
6464 unlock_new_inode(inode);
6465 goto out_unlock;
6466
6467 }
6468
6469 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6470 struct dentry *dentry)
6471 {
6472 struct btrfs_trans_handle *trans = NULL;
6473 struct btrfs_root *root = BTRFS_I(dir)->root;
6474 struct inode *inode = d_inode(old_dentry);
6475 u64 index;
6476 int err;
6477 int drop_inode = 0;
6478
6479 /* do not allow sys_link's with other subvols of the same device */
6480 if (root->objectid != BTRFS_I(inode)->root->objectid)
6481 return -EXDEV;
6482
6483 if (inode->i_nlink >= BTRFS_LINK_MAX)
6484 return -EMLINK;
6485
6486 err = btrfs_set_inode_index(dir, &index);
6487 if (err)
6488 goto fail;
6489
6490 /*
6491 * 2 items for inode and inode ref
6492 * 2 items for dir items
6493 * 1 item for parent inode
6494 */
6495 trans = btrfs_start_transaction(root, 5);
6496 if (IS_ERR(trans)) {
6497 err = PTR_ERR(trans);
6498 trans = NULL;
6499 goto fail;
6500 }
6501
6502 /* There are several dir indexes for this inode, clear the cache. */
6503 BTRFS_I(inode)->dir_index = 0ULL;
6504 inc_nlink(inode);
6505 inode_inc_iversion(inode);
6506 inode->i_ctime = CURRENT_TIME;
6507 ihold(inode);
6508 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6509
6510 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6511
6512 if (err) {
6513 drop_inode = 1;
6514 } else {
6515 struct dentry *parent = dentry->d_parent;
6516 err = btrfs_update_inode(trans, root, inode);
6517 if (err)
6518 goto fail;
6519 if (inode->i_nlink == 1) {
6520 /*
6521 * If new hard link count is 1, it's a file created
6522 * with open(2) O_TMPFILE flag.
6523 */
6524 err = btrfs_orphan_del(trans, inode);
6525 if (err)
6526 goto fail;
6527 }
6528 d_instantiate(dentry, inode);
6529 btrfs_log_new_name(trans, inode, NULL, parent);
6530 }
6531
6532 btrfs_balance_delayed_items(root);
6533 fail:
6534 if (trans)
6535 btrfs_end_transaction(trans, root);
6536 if (drop_inode) {
6537 inode_dec_link_count(inode);
6538 iput(inode);
6539 }
6540 btrfs_btree_balance_dirty(root);
6541 return err;
6542 }
6543
6544 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6545 {
6546 struct inode *inode = NULL;
6547 struct btrfs_trans_handle *trans;
6548 struct btrfs_root *root = BTRFS_I(dir)->root;
6549 int err = 0;
6550 int drop_on_err = 0;
6551 u64 objectid = 0;
6552 u64 index = 0;
6553
6554 /*
6555 * 2 items for inode and ref
6556 * 2 items for dir items
6557 * 1 for xattr if selinux is on
6558 */
6559 trans = btrfs_start_transaction(root, 5);
6560 if (IS_ERR(trans))
6561 return PTR_ERR(trans);
6562
6563 err = btrfs_find_free_ino(root, &objectid);
6564 if (err)
6565 goto out_fail;
6566
6567 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6568 dentry->d_name.len, btrfs_ino(dir), objectid,
6569 S_IFDIR | mode, &index);
6570 if (IS_ERR(inode)) {
6571 err = PTR_ERR(inode);
6572 goto out_fail;
6573 }
6574
6575 drop_on_err = 1;
6576 /* these must be set before we unlock the inode */
6577 inode->i_op = &btrfs_dir_inode_operations;
6578 inode->i_fop = &btrfs_dir_file_operations;
6579
6580 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6581 if (err)
6582 goto out_fail_inode;
6583
6584 btrfs_i_size_write(inode, 0);
6585 err = btrfs_update_inode(trans, root, inode);
6586 if (err)
6587 goto out_fail_inode;
6588
6589 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6590 dentry->d_name.len, 0, index);
6591 if (err)
6592 goto out_fail_inode;
6593
6594 d_instantiate(dentry, inode);
6595 /*
6596 * mkdir is special. We're unlocking after we call d_instantiate
6597 * to avoid a race with nfsd calling d_instantiate.
6598 */
6599 unlock_new_inode(inode);
6600 drop_on_err = 0;
6601
6602 out_fail:
6603 btrfs_end_transaction(trans, root);
6604 if (drop_on_err) {
6605 inode_dec_link_count(inode);
6606 iput(inode);
6607 }
6608 btrfs_balance_delayed_items(root);
6609 btrfs_btree_balance_dirty(root);
6610 return err;
6611
6612 out_fail_inode:
6613 unlock_new_inode(inode);
6614 goto out_fail;
6615 }
6616
6617 /* Find next extent map of a given extent map, caller needs to ensure locks */
6618 static struct extent_map *next_extent_map(struct extent_map *em)
6619 {
6620 struct rb_node *next;
6621
6622 next = rb_next(&em->rb_node);
6623 if (!next)
6624 return NULL;
6625 return container_of(next, struct extent_map, rb_node);
6626 }
6627
6628 static struct extent_map *prev_extent_map(struct extent_map *em)
6629 {
6630 struct rb_node *prev;
6631
6632 prev = rb_prev(&em->rb_node);
6633 if (!prev)
6634 return NULL;
6635 return container_of(prev, struct extent_map, rb_node);
6636 }
6637
6638 /* helper for btfs_get_extent. Given an existing extent in the tree,
6639 * the existing extent is the nearest extent to map_start,
6640 * and an extent that you want to insert, deal with overlap and insert
6641 * the best fitted new extent into the tree.
6642 */
6643 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6644 struct extent_map *existing,
6645 struct extent_map *em,
6646 u64 map_start)
6647 {
6648 struct extent_map *prev;
6649 struct extent_map *next;
6650 u64 start;
6651 u64 end;
6652 u64 start_diff;
6653
6654 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6655
6656 if (existing->start > map_start) {
6657 next = existing;
6658 prev = prev_extent_map(next);
6659 } else {
6660 prev = existing;
6661 next = next_extent_map(prev);
6662 }
6663
6664 start = prev ? extent_map_end(prev) : em->start;
6665 start = max_t(u64, start, em->start);
6666 end = next ? next->start : extent_map_end(em);
6667 end = min_t(u64, end, extent_map_end(em));
6668 start_diff = start - em->start;
6669 em->start = start;
6670 em->len = end - start;
6671 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6672 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6673 em->block_start += start_diff;
6674 em->block_len -= start_diff;
6675 }
6676 return add_extent_mapping(em_tree, em, 0);
6677 }
6678
6679 static noinline int uncompress_inline(struct btrfs_path *path,
6680 struct page *page,
6681 size_t pg_offset, u64 extent_offset,
6682 struct btrfs_file_extent_item *item)
6683 {
6684 int ret;
6685 struct extent_buffer *leaf = path->nodes[0];
6686 char *tmp;
6687 size_t max_size;
6688 unsigned long inline_size;
6689 unsigned long ptr;
6690 int compress_type;
6691
6692 WARN_ON(pg_offset != 0);
6693 compress_type = btrfs_file_extent_compression(leaf, item);
6694 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6695 inline_size = btrfs_file_extent_inline_item_len(leaf,
6696 btrfs_item_nr(path->slots[0]));
6697 tmp = kmalloc(inline_size, GFP_NOFS);
6698 if (!tmp)
6699 return -ENOMEM;
6700 ptr = btrfs_file_extent_inline_start(item);
6701
6702 read_extent_buffer(leaf, tmp, ptr, inline_size);
6703
6704 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6705 ret = btrfs_decompress(compress_type, tmp, page,
6706 extent_offset, inline_size, max_size);
6707 kfree(tmp);
6708 return ret;
6709 }
6710
6711 /*
6712 * a bit scary, this does extent mapping from logical file offset to the disk.
6713 * the ugly parts come from merging extents from the disk with the in-ram
6714 * representation. This gets more complex because of the data=ordered code,
6715 * where the in-ram extents might be locked pending data=ordered completion.
6716 *
6717 * This also copies inline extents directly into the page.
6718 */
6719
6720 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6721 size_t pg_offset, u64 start, u64 len,
6722 int create)
6723 {
6724 int ret;
6725 int err = 0;
6726 u64 extent_start = 0;
6727 u64 extent_end = 0;
6728 u64 objectid = btrfs_ino(inode);
6729 u32 found_type;
6730 struct btrfs_path *path = NULL;
6731 struct btrfs_root *root = BTRFS_I(inode)->root;
6732 struct btrfs_file_extent_item *item;
6733 struct extent_buffer *leaf;
6734 struct btrfs_key found_key;
6735 struct extent_map *em = NULL;
6736 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6737 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6738 struct btrfs_trans_handle *trans = NULL;
6739 const bool new_inline = !page || create;
6740
6741 again:
6742 read_lock(&em_tree->lock);
6743 em = lookup_extent_mapping(em_tree, start, len);
6744 if (em)
6745 em->bdev = root->fs_info->fs_devices->latest_bdev;
6746 read_unlock(&em_tree->lock);
6747
6748 if (em) {
6749 if (em->start > start || em->start + em->len <= start)
6750 free_extent_map(em);
6751 else if (em->block_start == EXTENT_MAP_INLINE && page)
6752 free_extent_map(em);
6753 else
6754 goto out;
6755 }
6756 em = alloc_extent_map();
6757 if (!em) {
6758 err = -ENOMEM;
6759 goto out;
6760 }
6761 em->bdev = root->fs_info->fs_devices->latest_bdev;
6762 em->start = EXTENT_MAP_HOLE;
6763 em->orig_start = EXTENT_MAP_HOLE;
6764 em->len = (u64)-1;
6765 em->block_len = (u64)-1;
6766
6767 if (!path) {
6768 path = btrfs_alloc_path();
6769 if (!path) {
6770 err = -ENOMEM;
6771 goto out;
6772 }
6773 /*
6774 * Chances are we'll be called again, so go ahead and do
6775 * readahead
6776 */
6777 path->reada = READA_FORWARD;
6778 }
6779
6780 ret = btrfs_lookup_file_extent(trans, root, path,
6781 objectid, start, trans != NULL);
6782 if (ret < 0) {
6783 err = ret;
6784 goto out;
6785 }
6786
6787 if (ret != 0) {
6788 if (path->slots[0] == 0)
6789 goto not_found;
6790 path->slots[0]--;
6791 }
6792
6793 leaf = path->nodes[0];
6794 item = btrfs_item_ptr(leaf, path->slots[0],
6795 struct btrfs_file_extent_item);
6796 /* are we inside the extent that was found? */
6797 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6798 found_type = found_key.type;
6799 if (found_key.objectid != objectid ||
6800 found_type != BTRFS_EXTENT_DATA_KEY) {
6801 /*
6802 * If we backup past the first extent we want to move forward
6803 * and see if there is an extent in front of us, otherwise we'll
6804 * say there is a hole for our whole search range which can
6805 * cause problems.
6806 */
6807 extent_end = start;
6808 goto next;
6809 }
6810
6811 found_type = btrfs_file_extent_type(leaf, item);
6812 extent_start = found_key.offset;
6813 if (found_type == BTRFS_FILE_EXTENT_REG ||
6814 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6815 extent_end = extent_start +
6816 btrfs_file_extent_num_bytes(leaf, item);
6817 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6818 size_t size;
6819 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6820 extent_end = ALIGN(extent_start + size, root->sectorsize);
6821 }
6822 next:
6823 if (start >= extent_end) {
6824 path->slots[0]++;
6825 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6826 ret = btrfs_next_leaf(root, path);
6827 if (ret < 0) {
6828 err = ret;
6829 goto out;
6830 }
6831 if (ret > 0)
6832 goto not_found;
6833 leaf = path->nodes[0];
6834 }
6835 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6836 if (found_key.objectid != objectid ||
6837 found_key.type != BTRFS_EXTENT_DATA_KEY)
6838 goto not_found;
6839 if (start + len <= found_key.offset)
6840 goto not_found;
6841 if (start > found_key.offset)
6842 goto next;
6843 em->start = start;
6844 em->orig_start = start;
6845 em->len = found_key.offset - start;
6846 goto not_found_em;
6847 }
6848
6849 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6850
6851 if (found_type == BTRFS_FILE_EXTENT_REG ||
6852 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6853 goto insert;
6854 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6855 unsigned long ptr;
6856 char *map;
6857 size_t size;
6858 size_t extent_offset;
6859 size_t copy_size;
6860
6861 if (new_inline)
6862 goto out;
6863
6864 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6865 extent_offset = page_offset(page) + pg_offset - extent_start;
6866 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6867 size - extent_offset);
6868 em->start = extent_start + extent_offset;
6869 em->len = ALIGN(copy_size, root->sectorsize);
6870 em->orig_block_len = em->len;
6871 em->orig_start = em->start;
6872 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6873 if (create == 0 && !PageUptodate(page)) {
6874 if (btrfs_file_extent_compression(leaf, item) !=
6875 BTRFS_COMPRESS_NONE) {
6876 ret = uncompress_inline(path, page, pg_offset,
6877 extent_offset, item);
6878 if (ret) {
6879 err = ret;
6880 goto out;
6881 }
6882 } else {
6883 map = kmap(page);
6884 read_extent_buffer(leaf, map + pg_offset, ptr,
6885 copy_size);
6886 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6887 memset(map + pg_offset + copy_size, 0,
6888 PAGE_CACHE_SIZE - pg_offset -
6889 copy_size);
6890 }
6891 kunmap(page);
6892 }
6893 flush_dcache_page(page);
6894 } else if (create && PageUptodate(page)) {
6895 BUG();
6896 if (!trans) {
6897 kunmap(page);
6898 free_extent_map(em);
6899 em = NULL;
6900
6901 btrfs_release_path(path);
6902 trans = btrfs_join_transaction(root);
6903
6904 if (IS_ERR(trans))
6905 return ERR_CAST(trans);
6906 goto again;
6907 }
6908 map = kmap(page);
6909 write_extent_buffer(leaf, map + pg_offset, ptr,
6910 copy_size);
6911 kunmap(page);
6912 btrfs_mark_buffer_dirty(leaf);
6913 }
6914 set_extent_uptodate(io_tree, em->start,
6915 extent_map_end(em) - 1, NULL, GFP_NOFS);
6916 goto insert;
6917 }
6918 not_found:
6919 em->start = start;
6920 em->orig_start = start;
6921 em->len = len;
6922 not_found_em:
6923 em->block_start = EXTENT_MAP_HOLE;
6924 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6925 insert:
6926 btrfs_release_path(path);
6927 if (em->start > start || extent_map_end(em) <= start) {
6928 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6929 em->start, em->len, start, len);
6930 err = -EIO;
6931 goto out;
6932 }
6933
6934 err = 0;
6935 write_lock(&em_tree->lock);
6936 ret = add_extent_mapping(em_tree, em, 0);
6937 /* it is possible that someone inserted the extent into the tree
6938 * while we had the lock dropped. It is also possible that
6939 * an overlapping map exists in the tree
6940 */
6941 if (ret == -EEXIST) {
6942 struct extent_map *existing;
6943
6944 ret = 0;
6945
6946 existing = search_extent_mapping(em_tree, start, len);
6947 /*
6948 * existing will always be non-NULL, since there must be
6949 * extent causing the -EEXIST.
6950 */
6951 if (start >= extent_map_end(existing) ||
6952 start <= existing->start) {
6953 /*
6954 * The existing extent map is the one nearest to
6955 * the [start, start + len) range which overlaps
6956 */
6957 err = merge_extent_mapping(em_tree, existing,
6958 em, start);
6959 free_extent_map(existing);
6960 if (err) {
6961 free_extent_map(em);
6962 em = NULL;
6963 }
6964 } else {
6965 free_extent_map(em);
6966 em = existing;
6967 err = 0;
6968 }
6969 }
6970 write_unlock(&em_tree->lock);
6971 out:
6972
6973 trace_btrfs_get_extent(root, em);
6974
6975 btrfs_free_path(path);
6976 if (trans) {
6977 ret = btrfs_end_transaction(trans, root);
6978 if (!err)
6979 err = ret;
6980 }
6981 if (err) {
6982 free_extent_map(em);
6983 return ERR_PTR(err);
6984 }
6985 BUG_ON(!em); /* Error is always set */
6986 return em;
6987 }
6988
6989 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6990 size_t pg_offset, u64 start, u64 len,
6991 int create)
6992 {
6993 struct extent_map *em;
6994 struct extent_map *hole_em = NULL;
6995 u64 range_start = start;
6996 u64 end;
6997 u64 found;
6998 u64 found_end;
6999 int err = 0;
7000
7001 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7002 if (IS_ERR(em))
7003 return em;
7004 if (em) {
7005 /*
7006 * if our em maps to
7007 * - a hole or
7008 * - a pre-alloc extent,
7009 * there might actually be delalloc bytes behind it.
7010 */
7011 if (em->block_start != EXTENT_MAP_HOLE &&
7012 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7013 return em;
7014 else
7015 hole_em = em;
7016 }
7017
7018 /* check to see if we've wrapped (len == -1 or similar) */
7019 end = start + len;
7020 if (end < start)
7021 end = (u64)-1;
7022 else
7023 end -= 1;
7024
7025 em = NULL;
7026
7027 /* ok, we didn't find anything, lets look for delalloc */
7028 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7029 end, len, EXTENT_DELALLOC, 1);
7030 found_end = range_start + found;
7031 if (found_end < range_start)
7032 found_end = (u64)-1;
7033
7034 /*
7035 * we didn't find anything useful, return
7036 * the original results from get_extent()
7037 */
7038 if (range_start > end || found_end <= start) {
7039 em = hole_em;
7040 hole_em = NULL;
7041 goto out;
7042 }
7043
7044 /* adjust the range_start to make sure it doesn't
7045 * go backwards from the start they passed in
7046 */
7047 range_start = max(start, range_start);
7048 found = found_end - range_start;
7049
7050 if (found > 0) {
7051 u64 hole_start = start;
7052 u64 hole_len = len;
7053
7054 em = alloc_extent_map();
7055 if (!em) {
7056 err = -ENOMEM;
7057 goto out;
7058 }
7059 /*
7060 * when btrfs_get_extent can't find anything it
7061 * returns one huge hole
7062 *
7063 * make sure what it found really fits our range, and
7064 * adjust to make sure it is based on the start from
7065 * the caller
7066 */
7067 if (hole_em) {
7068 u64 calc_end = extent_map_end(hole_em);
7069
7070 if (calc_end <= start || (hole_em->start > end)) {
7071 free_extent_map(hole_em);
7072 hole_em = NULL;
7073 } else {
7074 hole_start = max(hole_em->start, start);
7075 hole_len = calc_end - hole_start;
7076 }
7077 }
7078 em->bdev = NULL;
7079 if (hole_em && range_start > hole_start) {
7080 /* our hole starts before our delalloc, so we
7081 * have to return just the parts of the hole
7082 * that go until the delalloc starts
7083 */
7084 em->len = min(hole_len,
7085 range_start - hole_start);
7086 em->start = hole_start;
7087 em->orig_start = hole_start;
7088 /*
7089 * don't adjust block start at all,
7090 * it is fixed at EXTENT_MAP_HOLE
7091 */
7092 em->block_start = hole_em->block_start;
7093 em->block_len = hole_len;
7094 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7095 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7096 } else {
7097 em->start = range_start;
7098 em->len = found;
7099 em->orig_start = range_start;
7100 em->block_start = EXTENT_MAP_DELALLOC;
7101 em->block_len = found;
7102 }
7103 } else if (hole_em) {
7104 return hole_em;
7105 }
7106 out:
7107
7108 free_extent_map(hole_em);
7109 if (err) {
7110 free_extent_map(em);
7111 return ERR_PTR(err);
7112 }
7113 return em;
7114 }
7115
7116 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7117 u64 start, u64 len)
7118 {
7119 struct btrfs_root *root = BTRFS_I(inode)->root;
7120 struct extent_map *em;
7121 struct btrfs_key ins;
7122 u64 alloc_hint;
7123 int ret;
7124
7125 alloc_hint = get_extent_allocation_hint(inode, start, len);
7126 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7127 alloc_hint, &ins, 1, 1);
7128 if (ret)
7129 return ERR_PTR(ret);
7130
7131 /*
7132 * Create the ordered extent before the extent map. This is to avoid
7133 * races with the fast fsync path that would lead to it logging file
7134 * extent items that point to disk extents that were not yet written to.
7135 * The fast fsync path collects ordered extents into a local list and
7136 * then collects all the new extent maps, so we must create the ordered
7137 * extent first and make sure the fast fsync path collects any new
7138 * ordered extents after collecting new extent maps as well.
7139 * The fsync path simply can not rely on inode_dio_wait() because it
7140 * causes deadlock with AIO.
7141 */
7142 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7143 ins.offset, ins.offset, 0);
7144 if (ret) {
7145 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7146 return ERR_PTR(ret);
7147 }
7148
7149 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7150 ins.offset, ins.offset, ins.offset, 0);
7151 if (IS_ERR(em)) {
7152 struct btrfs_ordered_extent *oe;
7153
7154 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7155 oe = btrfs_lookup_ordered_extent(inode, start);
7156 ASSERT(oe);
7157 if (WARN_ON(!oe))
7158 return em;
7159 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7160 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7161 btrfs_remove_ordered_extent(inode, oe);
7162 /* Once for our lookup and once for the ordered extents tree. */
7163 btrfs_put_ordered_extent(oe);
7164 btrfs_put_ordered_extent(oe);
7165 }
7166 return em;
7167 }
7168
7169 /*
7170 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7171 * block must be cow'd
7172 */
7173 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7174 u64 *orig_start, u64 *orig_block_len,
7175 u64 *ram_bytes)
7176 {
7177 struct btrfs_trans_handle *trans;
7178 struct btrfs_path *path;
7179 int ret;
7180 struct extent_buffer *leaf;
7181 struct btrfs_root *root = BTRFS_I(inode)->root;
7182 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7183 struct btrfs_file_extent_item *fi;
7184 struct btrfs_key key;
7185 u64 disk_bytenr;
7186 u64 backref_offset;
7187 u64 extent_end;
7188 u64 num_bytes;
7189 int slot;
7190 int found_type;
7191 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7192
7193 path = btrfs_alloc_path();
7194 if (!path)
7195 return -ENOMEM;
7196
7197 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7198 offset, 0);
7199 if (ret < 0)
7200 goto out;
7201
7202 slot = path->slots[0];
7203 if (ret == 1) {
7204 if (slot == 0) {
7205 /* can't find the item, must cow */
7206 ret = 0;
7207 goto out;
7208 }
7209 slot--;
7210 }
7211 ret = 0;
7212 leaf = path->nodes[0];
7213 btrfs_item_key_to_cpu(leaf, &key, slot);
7214 if (key.objectid != btrfs_ino(inode) ||
7215 key.type != BTRFS_EXTENT_DATA_KEY) {
7216 /* not our file or wrong item type, must cow */
7217 goto out;
7218 }
7219
7220 if (key.offset > offset) {
7221 /* Wrong offset, must cow */
7222 goto out;
7223 }
7224
7225 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7226 found_type = btrfs_file_extent_type(leaf, fi);
7227 if (found_type != BTRFS_FILE_EXTENT_REG &&
7228 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7229 /* not a regular extent, must cow */
7230 goto out;
7231 }
7232
7233 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7234 goto out;
7235
7236 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7237 if (extent_end <= offset)
7238 goto out;
7239
7240 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7241 if (disk_bytenr == 0)
7242 goto out;
7243
7244 if (btrfs_file_extent_compression(leaf, fi) ||
7245 btrfs_file_extent_encryption(leaf, fi) ||
7246 btrfs_file_extent_other_encoding(leaf, fi))
7247 goto out;
7248
7249 backref_offset = btrfs_file_extent_offset(leaf, fi);
7250
7251 if (orig_start) {
7252 *orig_start = key.offset - backref_offset;
7253 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7254 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7255 }
7256
7257 if (btrfs_extent_readonly(root, disk_bytenr))
7258 goto out;
7259
7260 num_bytes = min(offset + *len, extent_end) - offset;
7261 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7262 u64 range_end;
7263
7264 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7265 ret = test_range_bit(io_tree, offset, range_end,
7266 EXTENT_DELALLOC, 0, NULL);
7267 if (ret) {
7268 ret = -EAGAIN;
7269 goto out;
7270 }
7271 }
7272
7273 btrfs_release_path(path);
7274
7275 /*
7276 * look for other files referencing this extent, if we
7277 * find any we must cow
7278 */
7279 trans = btrfs_join_transaction(root);
7280 if (IS_ERR(trans)) {
7281 ret = 0;
7282 goto out;
7283 }
7284
7285 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7286 key.offset - backref_offset, disk_bytenr);
7287 btrfs_end_transaction(trans, root);
7288 if (ret) {
7289 ret = 0;
7290 goto out;
7291 }
7292
7293 /*
7294 * adjust disk_bytenr and num_bytes to cover just the bytes
7295 * in this extent we are about to write. If there
7296 * are any csums in that range we have to cow in order
7297 * to keep the csums correct
7298 */
7299 disk_bytenr += backref_offset;
7300 disk_bytenr += offset - key.offset;
7301 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7302 goto out;
7303 /*
7304 * all of the above have passed, it is safe to overwrite this extent
7305 * without cow
7306 */
7307 *len = num_bytes;
7308 ret = 1;
7309 out:
7310 btrfs_free_path(path);
7311 return ret;
7312 }
7313
7314 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7315 {
7316 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7317 int found = false;
7318 void **pagep = NULL;
7319 struct page *page = NULL;
7320 int start_idx;
7321 int end_idx;
7322
7323 start_idx = start >> PAGE_CACHE_SHIFT;
7324
7325 /*
7326 * end is the last byte in the last page. end == start is legal
7327 */
7328 end_idx = end >> PAGE_CACHE_SHIFT;
7329
7330 rcu_read_lock();
7331
7332 /* Most of the code in this while loop is lifted from
7333 * find_get_page. It's been modified to begin searching from a
7334 * page and return just the first page found in that range. If the
7335 * found idx is less than or equal to the end idx then we know that
7336 * a page exists. If no pages are found or if those pages are
7337 * outside of the range then we're fine (yay!) */
7338 while (page == NULL &&
7339 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7340 page = radix_tree_deref_slot(pagep);
7341 if (unlikely(!page))
7342 break;
7343
7344 if (radix_tree_exception(page)) {
7345 if (radix_tree_deref_retry(page)) {
7346 page = NULL;
7347 continue;
7348 }
7349 /*
7350 * Otherwise, shmem/tmpfs must be storing a swap entry
7351 * here as an exceptional entry: so return it without
7352 * attempting to raise page count.
7353 */
7354 page = NULL;
7355 break; /* TODO: Is this relevant for this use case? */
7356 }
7357
7358 if (!page_cache_get_speculative(page)) {
7359 page = NULL;
7360 continue;
7361 }
7362
7363 /*
7364 * Has the page moved?
7365 * This is part of the lockless pagecache protocol. See
7366 * include/linux/pagemap.h for details.
7367 */
7368 if (unlikely(page != *pagep)) {
7369 page_cache_release(page);
7370 page = NULL;
7371 }
7372 }
7373
7374 if (page) {
7375 if (page->index <= end_idx)
7376 found = true;
7377 page_cache_release(page);
7378 }
7379
7380 rcu_read_unlock();
7381 return found;
7382 }
7383
7384 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7385 struct extent_state **cached_state, int writing)
7386 {
7387 struct btrfs_ordered_extent *ordered;
7388 int ret = 0;
7389
7390 while (1) {
7391 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7392 cached_state);
7393 /*
7394 * We're concerned with the entire range that we're going to be
7395 * doing DIO to, so we need to make sure theres no ordered
7396 * extents in this range.
7397 */
7398 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7399 lockend - lockstart + 1);
7400
7401 /*
7402 * We need to make sure there are no buffered pages in this
7403 * range either, we could have raced between the invalidate in
7404 * generic_file_direct_write and locking the extent. The
7405 * invalidate needs to happen so that reads after a write do not
7406 * get stale data.
7407 */
7408 if (!ordered &&
7409 (!writing ||
7410 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7411 break;
7412
7413 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7414 cached_state, GFP_NOFS);
7415
7416 if (ordered) {
7417 btrfs_start_ordered_extent(inode, ordered, 1);
7418 btrfs_put_ordered_extent(ordered);
7419 } else {
7420 /*
7421 * We could trigger writeback for this range (and wait
7422 * for it to complete) and then invalidate the pages for
7423 * this range (through invalidate_inode_pages2_range()),
7424 * but that can lead us to a deadlock with a concurrent
7425 * call to readpages() (a buffered read or a defrag call
7426 * triggered a readahead) on a page lock due to an
7427 * ordered dio extent we created before but did not have
7428 * yet a corresponding bio submitted (whence it can not
7429 * complete), which makes readpages() wait for that
7430 * ordered extent to complete while holding a lock on
7431 * that page.
7432 */
7433 ret = -ENOTBLK;
7434 break;
7435 }
7436
7437 cond_resched();
7438 }
7439
7440 return ret;
7441 }
7442
7443 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7444 u64 len, u64 orig_start,
7445 u64 block_start, u64 block_len,
7446 u64 orig_block_len, u64 ram_bytes,
7447 int type)
7448 {
7449 struct extent_map_tree *em_tree;
7450 struct extent_map *em;
7451 struct btrfs_root *root = BTRFS_I(inode)->root;
7452 int ret;
7453
7454 em_tree = &BTRFS_I(inode)->extent_tree;
7455 em = alloc_extent_map();
7456 if (!em)
7457 return ERR_PTR(-ENOMEM);
7458
7459 em->start = start;
7460 em->orig_start = orig_start;
7461 em->mod_start = start;
7462 em->mod_len = len;
7463 em->len = len;
7464 em->block_len = block_len;
7465 em->block_start = block_start;
7466 em->bdev = root->fs_info->fs_devices->latest_bdev;
7467 em->orig_block_len = orig_block_len;
7468 em->ram_bytes = ram_bytes;
7469 em->generation = -1;
7470 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7471 if (type == BTRFS_ORDERED_PREALLOC)
7472 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7473
7474 do {
7475 btrfs_drop_extent_cache(inode, em->start,
7476 em->start + em->len - 1, 0);
7477 write_lock(&em_tree->lock);
7478 ret = add_extent_mapping(em_tree, em, 1);
7479 write_unlock(&em_tree->lock);
7480 } while (ret == -EEXIST);
7481
7482 if (ret) {
7483 free_extent_map(em);
7484 return ERR_PTR(ret);
7485 }
7486
7487 return em;
7488 }
7489
7490 static void adjust_dio_outstanding_extents(struct inode *inode,
7491 struct btrfs_dio_data *dio_data,
7492 const u64 len)
7493 {
7494 unsigned num_extents;
7495
7496 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7497 BTRFS_MAX_EXTENT_SIZE);
7498 /*
7499 * If we have an outstanding_extents count still set then we're
7500 * within our reservation, otherwise we need to adjust our inode
7501 * counter appropriately.
7502 */
7503 if (dio_data->outstanding_extents) {
7504 dio_data->outstanding_extents -= num_extents;
7505 } else {
7506 spin_lock(&BTRFS_I(inode)->lock);
7507 BTRFS_I(inode)->outstanding_extents += num_extents;
7508 spin_unlock(&BTRFS_I(inode)->lock);
7509 }
7510 }
7511
7512 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7513 struct buffer_head *bh_result, int create)
7514 {
7515 struct extent_map *em;
7516 struct btrfs_root *root = BTRFS_I(inode)->root;
7517 struct extent_state *cached_state = NULL;
7518 struct btrfs_dio_data *dio_data = NULL;
7519 u64 start = iblock << inode->i_blkbits;
7520 u64 lockstart, lockend;
7521 u64 len = bh_result->b_size;
7522 int unlock_bits = EXTENT_LOCKED;
7523 int ret = 0;
7524
7525 if (create)
7526 unlock_bits |= EXTENT_DIRTY;
7527 else
7528 len = min_t(u64, len, root->sectorsize);
7529
7530 lockstart = start;
7531 lockend = start + len - 1;
7532
7533 if (current->journal_info) {
7534 /*
7535 * Need to pull our outstanding extents and set journal_info to NULL so
7536 * that anything that needs to check if there's a transction doesn't get
7537 * confused.
7538 */
7539 dio_data = current->journal_info;
7540 current->journal_info = NULL;
7541 }
7542
7543 /*
7544 * If this errors out it's because we couldn't invalidate pagecache for
7545 * this range and we need to fallback to buffered.
7546 */
7547 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7548 create)) {
7549 ret = -ENOTBLK;
7550 goto err;
7551 }
7552
7553 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7554 if (IS_ERR(em)) {
7555 ret = PTR_ERR(em);
7556 goto unlock_err;
7557 }
7558
7559 /*
7560 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7561 * io. INLINE is special, and we could probably kludge it in here, but
7562 * it's still buffered so for safety lets just fall back to the generic
7563 * buffered path.
7564 *
7565 * For COMPRESSED we _have_ to read the entire extent in so we can
7566 * decompress it, so there will be buffering required no matter what we
7567 * do, so go ahead and fallback to buffered.
7568 *
7569 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7570 * to buffered IO. Don't blame me, this is the price we pay for using
7571 * the generic code.
7572 */
7573 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7574 em->block_start == EXTENT_MAP_INLINE) {
7575 free_extent_map(em);
7576 ret = -ENOTBLK;
7577 goto unlock_err;
7578 }
7579
7580 /* Just a good old fashioned hole, return */
7581 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7582 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7583 free_extent_map(em);
7584 goto unlock_err;
7585 }
7586
7587 /*
7588 * We don't allocate a new extent in the following cases
7589 *
7590 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7591 * existing extent.
7592 * 2) The extent is marked as PREALLOC. We're good to go here and can
7593 * just use the extent.
7594 *
7595 */
7596 if (!create) {
7597 len = min(len, em->len - (start - em->start));
7598 lockstart = start + len;
7599 goto unlock;
7600 }
7601
7602 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7603 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7604 em->block_start != EXTENT_MAP_HOLE)) {
7605 int type;
7606 u64 block_start, orig_start, orig_block_len, ram_bytes;
7607
7608 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7609 type = BTRFS_ORDERED_PREALLOC;
7610 else
7611 type = BTRFS_ORDERED_NOCOW;
7612 len = min(len, em->len - (start - em->start));
7613 block_start = em->block_start + (start - em->start);
7614
7615 if (can_nocow_extent(inode, start, &len, &orig_start,
7616 &orig_block_len, &ram_bytes) == 1) {
7617 if (type == BTRFS_ORDERED_PREALLOC) {
7618 free_extent_map(em);
7619 em = create_pinned_em(inode, start, len,
7620 orig_start,
7621 block_start, len,
7622 orig_block_len,
7623 ram_bytes, type);
7624 if (IS_ERR(em)) {
7625 ret = PTR_ERR(em);
7626 goto unlock_err;
7627 }
7628 }
7629
7630 ret = btrfs_add_ordered_extent_dio(inode, start,
7631 block_start, len, len, type);
7632 if (ret) {
7633 free_extent_map(em);
7634 goto unlock_err;
7635 }
7636 goto unlock;
7637 }
7638 }
7639
7640 /*
7641 * this will cow the extent, reset the len in case we changed
7642 * it above
7643 */
7644 len = bh_result->b_size;
7645 free_extent_map(em);
7646 em = btrfs_new_extent_direct(inode, start, len);
7647 if (IS_ERR(em)) {
7648 ret = PTR_ERR(em);
7649 goto unlock_err;
7650 }
7651 len = min(len, em->len - (start - em->start));
7652 unlock:
7653 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7654 inode->i_blkbits;
7655 bh_result->b_size = len;
7656 bh_result->b_bdev = em->bdev;
7657 set_buffer_mapped(bh_result);
7658 if (create) {
7659 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7660 set_buffer_new(bh_result);
7661
7662 /*
7663 * Need to update the i_size under the extent lock so buffered
7664 * readers will get the updated i_size when we unlock.
7665 */
7666 if (start + len > i_size_read(inode))
7667 i_size_write(inode, start + len);
7668
7669 adjust_dio_outstanding_extents(inode, dio_data, len);
7670 btrfs_free_reserved_data_space(inode, start, len);
7671 WARN_ON(dio_data->reserve < len);
7672 dio_data->reserve -= len;
7673 dio_data->unsubmitted_oe_range_end = start + len;
7674 current->journal_info = dio_data;
7675 }
7676
7677 /*
7678 * In the case of write we need to clear and unlock the entire range,
7679 * in the case of read we need to unlock only the end area that we
7680 * aren't using if there is any left over space.
7681 */
7682 if (lockstart < lockend) {
7683 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7684 lockend, unlock_bits, 1, 0,
7685 &cached_state, GFP_NOFS);
7686 } else {
7687 free_extent_state(cached_state);
7688 }
7689
7690 free_extent_map(em);
7691
7692 return 0;
7693
7694 unlock_err:
7695 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7696 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7697 err:
7698 if (dio_data)
7699 current->journal_info = dio_data;
7700 /*
7701 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7702 * write less data then expected, so that we don't underflow our inode's
7703 * outstanding extents counter.
7704 */
7705 if (create && dio_data)
7706 adjust_dio_outstanding_extents(inode, dio_data, len);
7707
7708 return ret;
7709 }
7710
7711 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7712 int rw, int mirror_num)
7713 {
7714 struct btrfs_root *root = BTRFS_I(inode)->root;
7715 int ret;
7716
7717 BUG_ON(rw & REQ_WRITE);
7718
7719 bio_get(bio);
7720
7721 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7722 BTRFS_WQ_ENDIO_DIO_REPAIR);
7723 if (ret)
7724 goto err;
7725
7726 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7727 err:
7728 bio_put(bio);
7729 return ret;
7730 }
7731
7732 static int btrfs_check_dio_repairable(struct inode *inode,
7733 struct bio *failed_bio,
7734 struct io_failure_record *failrec,
7735 int failed_mirror)
7736 {
7737 int num_copies;
7738
7739 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7740 failrec->logical, failrec->len);
7741 if (num_copies == 1) {
7742 /*
7743 * we only have a single copy of the data, so don't bother with
7744 * all the retry and error correction code that follows. no
7745 * matter what the error is, it is very likely to persist.
7746 */
7747 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7748 num_copies, failrec->this_mirror, failed_mirror);
7749 return 0;
7750 }
7751
7752 failrec->failed_mirror = failed_mirror;
7753 failrec->this_mirror++;
7754 if (failrec->this_mirror == failed_mirror)
7755 failrec->this_mirror++;
7756
7757 if (failrec->this_mirror > num_copies) {
7758 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7759 num_copies, failrec->this_mirror, failed_mirror);
7760 return 0;
7761 }
7762
7763 return 1;
7764 }
7765
7766 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7767 struct page *page, u64 start, u64 end,
7768 int failed_mirror, bio_end_io_t *repair_endio,
7769 void *repair_arg)
7770 {
7771 struct io_failure_record *failrec;
7772 struct bio *bio;
7773 int isector;
7774 int read_mode;
7775 int ret;
7776
7777 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7778
7779 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7780 if (ret)
7781 return ret;
7782
7783 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7784 failed_mirror);
7785 if (!ret) {
7786 free_io_failure(inode, failrec);
7787 return -EIO;
7788 }
7789
7790 if (failed_bio->bi_vcnt > 1)
7791 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7792 else
7793 read_mode = READ_SYNC;
7794
7795 isector = start - btrfs_io_bio(failed_bio)->logical;
7796 isector >>= inode->i_sb->s_blocksize_bits;
7797 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7798 0, isector, repair_endio, repair_arg);
7799 if (!bio) {
7800 free_io_failure(inode, failrec);
7801 return -EIO;
7802 }
7803
7804 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7805 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7806 read_mode, failrec->this_mirror, failrec->in_validation);
7807
7808 ret = submit_dio_repair_bio(inode, bio, read_mode,
7809 failrec->this_mirror);
7810 if (ret) {
7811 free_io_failure(inode, failrec);
7812 bio_put(bio);
7813 }
7814
7815 return ret;
7816 }
7817
7818 struct btrfs_retry_complete {
7819 struct completion done;
7820 struct inode *inode;
7821 u64 start;
7822 int uptodate;
7823 };
7824
7825 static void btrfs_retry_endio_nocsum(struct bio *bio)
7826 {
7827 struct btrfs_retry_complete *done = bio->bi_private;
7828 struct bio_vec *bvec;
7829 int i;
7830
7831 if (bio->bi_error)
7832 goto end;
7833
7834 done->uptodate = 1;
7835 bio_for_each_segment_all(bvec, bio, i)
7836 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7837 end:
7838 complete(&done->done);
7839 bio_put(bio);
7840 }
7841
7842 static int __btrfs_correct_data_nocsum(struct inode *inode,
7843 struct btrfs_io_bio *io_bio)
7844 {
7845 struct bio_vec *bvec;
7846 struct btrfs_retry_complete done;
7847 u64 start;
7848 int i;
7849 int ret;
7850
7851 start = io_bio->logical;
7852 done.inode = inode;
7853
7854 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7855 try_again:
7856 done.uptodate = 0;
7857 done.start = start;
7858 init_completion(&done.done);
7859
7860 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7861 start + bvec->bv_len - 1,
7862 io_bio->mirror_num,
7863 btrfs_retry_endio_nocsum, &done);
7864 if (ret)
7865 return ret;
7866
7867 wait_for_completion(&done.done);
7868
7869 if (!done.uptodate) {
7870 /* We might have another mirror, so try again */
7871 goto try_again;
7872 }
7873
7874 start += bvec->bv_len;
7875 }
7876
7877 return 0;
7878 }
7879
7880 static void btrfs_retry_endio(struct bio *bio)
7881 {
7882 struct btrfs_retry_complete *done = bio->bi_private;
7883 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7884 struct bio_vec *bvec;
7885 int uptodate;
7886 int ret;
7887 int i;
7888
7889 if (bio->bi_error)
7890 goto end;
7891
7892 uptodate = 1;
7893 bio_for_each_segment_all(bvec, bio, i) {
7894 ret = __readpage_endio_check(done->inode, io_bio, i,
7895 bvec->bv_page, 0,
7896 done->start, bvec->bv_len);
7897 if (!ret)
7898 clean_io_failure(done->inode, done->start,
7899 bvec->bv_page, 0);
7900 else
7901 uptodate = 0;
7902 }
7903
7904 done->uptodate = uptodate;
7905 end:
7906 complete(&done->done);
7907 bio_put(bio);
7908 }
7909
7910 static int __btrfs_subio_endio_read(struct inode *inode,
7911 struct btrfs_io_bio *io_bio, int err)
7912 {
7913 struct bio_vec *bvec;
7914 struct btrfs_retry_complete done;
7915 u64 start;
7916 u64 offset = 0;
7917 int i;
7918 int ret;
7919
7920 err = 0;
7921 start = io_bio->logical;
7922 done.inode = inode;
7923
7924 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7925 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7926 0, start, bvec->bv_len);
7927 if (likely(!ret))
7928 goto next;
7929 try_again:
7930 done.uptodate = 0;
7931 done.start = start;
7932 init_completion(&done.done);
7933
7934 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7935 start + bvec->bv_len - 1,
7936 io_bio->mirror_num,
7937 btrfs_retry_endio, &done);
7938 if (ret) {
7939 err = ret;
7940 goto next;
7941 }
7942
7943 wait_for_completion(&done.done);
7944
7945 if (!done.uptodate) {
7946 /* We might have another mirror, so try again */
7947 goto try_again;
7948 }
7949 next:
7950 offset += bvec->bv_len;
7951 start += bvec->bv_len;
7952 }
7953
7954 return err;
7955 }
7956
7957 static int btrfs_subio_endio_read(struct inode *inode,
7958 struct btrfs_io_bio *io_bio, int err)
7959 {
7960 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7961
7962 if (skip_csum) {
7963 if (unlikely(err))
7964 return __btrfs_correct_data_nocsum(inode, io_bio);
7965 else
7966 return 0;
7967 } else {
7968 return __btrfs_subio_endio_read(inode, io_bio, err);
7969 }
7970 }
7971
7972 static void btrfs_endio_direct_read(struct bio *bio)
7973 {
7974 struct btrfs_dio_private *dip = bio->bi_private;
7975 struct inode *inode = dip->inode;
7976 struct bio *dio_bio;
7977 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7978 int err = bio->bi_error;
7979
7980 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7981 err = btrfs_subio_endio_read(inode, io_bio, err);
7982
7983 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7984 dip->logical_offset + dip->bytes - 1);
7985 dio_bio = dip->dio_bio;
7986
7987 kfree(dip);
7988
7989 dio_bio->bi_error = bio->bi_error;
7990 dio_end_io(dio_bio, bio->bi_error);
7991
7992 if (io_bio->end_io)
7993 io_bio->end_io(io_bio, err);
7994 bio_put(bio);
7995 }
7996
7997 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
7998 const u64 offset,
7999 const u64 bytes,
8000 const int uptodate)
8001 {
8002 struct btrfs_root *root = BTRFS_I(inode)->root;
8003 struct btrfs_ordered_extent *ordered = NULL;
8004 u64 ordered_offset = offset;
8005 u64 ordered_bytes = bytes;
8006 int ret;
8007
8008 again:
8009 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8010 &ordered_offset,
8011 ordered_bytes,
8012 uptodate);
8013 if (!ret)
8014 goto out_test;
8015
8016 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8017 finish_ordered_fn, NULL, NULL);
8018 btrfs_queue_work(root->fs_info->endio_write_workers,
8019 &ordered->work);
8020 out_test:
8021 /*
8022 * our bio might span multiple ordered extents. If we haven't
8023 * completed the accounting for the whole dio, go back and try again
8024 */
8025 if (ordered_offset < offset + bytes) {
8026 ordered_bytes = offset + bytes - ordered_offset;
8027 ordered = NULL;
8028 goto again;
8029 }
8030 }
8031
8032 static void btrfs_endio_direct_write(struct bio *bio)
8033 {
8034 struct btrfs_dio_private *dip = bio->bi_private;
8035 struct bio *dio_bio = dip->dio_bio;
8036
8037 btrfs_endio_direct_write_update_ordered(dip->inode,
8038 dip->logical_offset,
8039 dip->bytes,
8040 !bio->bi_error);
8041
8042 kfree(dip);
8043
8044 dio_bio->bi_error = bio->bi_error;
8045 dio_end_io(dio_bio, bio->bi_error);
8046 bio_put(bio);
8047 }
8048
8049 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8050 struct bio *bio, int mirror_num,
8051 unsigned long bio_flags, u64 offset)
8052 {
8053 int ret;
8054 struct btrfs_root *root = BTRFS_I(inode)->root;
8055 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8056 BUG_ON(ret); /* -ENOMEM */
8057 return 0;
8058 }
8059
8060 static void btrfs_end_dio_bio(struct bio *bio)
8061 {
8062 struct btrfs_dio_private *dip = bio->bi_private;
8063 int err = bio->bi_error;
8064
8065 if (err)
8066 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8067 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8068 btrfs_ino(dip->inode), bio->bi_rw,
8069 (unsigned long long)bio->bi_iter.bi_sector,
8070 bio->bi_iter.bi_size, err);
8071
8072 if (dip->subio_endio)
8073 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8074
8075 if (err) {
8076 dip->errors = 1;
8077
8078 /*
8079 * before atomic variable goto zero, we must make sure
8080 * dip->errors is perceived to be set.
8081 */
8082 smp_mb__before_atomic();
8083 }
8084
8085 /* if there are more bios still pending for this dio, just exit */
8086 if (!atomic_dec_and_test(&dip->pending_bios))
8087 goto out;
8088
8089 if (dip->errors) {
8090 bio_io_error(dip->orig_bio);
8091 } else {
8092 dip->dio_bio->bi_error = 0;
8093 bio_endio(dip->orig_bio);
8094 }
8095 out:
8096 bio_put(bio);
8097 }
8098
8099 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8100 u64 first_sector, gfp_t gfp_flags)
8101 {
8102 struct bio *bio;
8103 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8104 if (bio)
8105 bio_associate_current(bio);
8106 return bio;
8107 }
8108
8109 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8110 struct inode *inode,
8111 struct btrfs_dio_private *dip,
8112 struct bio *bio,
8113 u64 file_offset)
8114 {
8115 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8116 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8117 int ret;
8118
8119 /*
8120 * We load all the csum data we need when we submit
8121 * the first bio to reduce the csum tree search and
8122 * contention.
8123 */
8124 if (dip->logical_offset == file_offset) {
8125 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8126 file_offset);
8127 if (ret)
8128 return ret;
8129 }
8130
8131 if (bio == dip->orig_bio)
8132 return 0;
8133
8134 file_offset -= dip->logical_offset;
8135 file_offset >>= inode->i_sb->s_blocksize_bits;
8136 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8137
8138 return 0;
8139 }
8140
8141 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8142 int rw, u64 file_offset, int skip_sum,
8143 int async_submit)
8144 {
8145 struct btrfs_dio_private *dip = bio->bi_private;
8146 int write = rw & REQ_WRITE;
8147 struct btrfs_root *root = BTRFS_I(inode)->root;
8148 int ret;
8149
8150 if (async_submit)
8151 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8152
8153 bio_get(bio);
8154
8155 if (!write) {
8156 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8157 BTRFS_WQ_ENDIO_DATA);
8158 if (ret)
8159 goto err;
8160 }
8161
8162 if (skip_sum)
8163 goto map;
8164
8165 if (write && async_submit) {
8166 ret = btrfs_wq_submit_bio(root->fs_info,
8167 inode, rw, bio, 0, 0,
8168 file_offset,
8169 __btrfs_submit_bio_start_direct_io,
8170 __btrfs_submit_bio_done);
8171 goto err;
8172 } else if (write) {
8173 /*
8174 * If we aren't doing async submit, calculate the csum of the
8175 * bio now.
8176 */
8177 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8178 if (ret)
8179 goto err;
8180 } else {
8181 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8182 file_offset);
8183 if (ret)
8184 goto err;
8185 }
8186 map:
8187 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8188 err:
8189 bio_put(bio);
8190 return ret;
8191 }
8192
8193 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8194 int skip_sum)
8195 {
8196 struct inode *inode = dip->inode;
8197 struct btrfs_root *root = BTRFS_I(inode)->root;
8198 struct bio *bio;
8199 struct bio *orig_bio = dip->orig_bio;
8200 struct bio_vec *bvec = orig_bio->bi_io_vec;
8201 u64 start_sector = orig_bio->bi_iter.bi_sector;
8202 u64 file_offset = dip->logical_offset;
8203 u64 submit_len = 0;
8204 u64 map_length;
8205 int nr_pages = 0;
8206 int ret;
8207 int async_submit = 0;
8208
8209 map_length = orig_bio->bi_iter.bi_size;
8210 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8211 &map_length, NULL, 0);
8212 if (ret)
8213 return -EIO;
8214
8215 if (map_length >= orig_bio->bi_iter.bi_size) {
8216 bio = orig_bio;
8217 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8218 goto submit;
8219 }
8220
8221 /* async crcs make it difficult to collect full stripe writes. */
8222 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8223 async_submit = 0;
8224 else
8225 async_submit = 1;
8226
8227 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8228 if (!bio)
8229 return -ENOMEM;
8230
8231 bio->bi_private = dip;
8232 bio->bi_end_io = btrfs_end_dio_bio;
8233 btrfs_io_bio(bio)->logical = file_offset;
8234 atomic_inc(&dip->pending_bios);
8235
8236 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8237 if (map_length < submit_len + bvec->bv_len ||
8238 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8239 bvec->bv_offset) < bvec->bv_len) {
8240 /*
8241 * inc the count before we submit the bio so
8242 * we know the end IO handler won't happen before
8243 * we inc the count. Otherwise, the dip might get freed
8244 * before we're done setting it up
8245 */
8246 atomic_inc(&dip->pending_bios);
8247 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8248 file_offset, skip_sum,
8249 async_submit);
8250 if (ret) {
8251 bio_put(bio);
8252 atomic_dec(&dip->pending_bios);
8253 goto out_err;
8254 }
8255
8256 start_sector += submit_len >> 9;
8257 file_offset += submit_len;
8258
8259 submit_len = 0;
8260 nr_pages = 0;
8261
8262 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8263 start_sector, GFP_NOFS);
8264 if (!bio)
8265 goto out_err;
8266 bio->bi_private = dip;
8267 bio->bi_end_io = btrfs_end_dio_bio;
8268 btrfs_io_bio(bio)->logical = file_offset;
8269
8270 map_length = orig_bio->bi_iter.bi_size;
8271 ret = btrfs_map_block(root->fs_info, rw,
8272 start_sector << 9,
8273 &map_length, NULL, 0);
8274 if (ret) {
8275 bio_put(bio);
8276 goto out_err;
8277 }
8278 } else {
8279 submit_len += bvec->bv_len;
8280 nr_pages++;
8281 bvec++;
8282 }
8283 }
8284
8285 submit:
8286 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8287 async_submit);
8288 if (!ret)
8289 return 0;
8290
8291 bio_put(bio);
8292 out_err:
8293 dip->errors = 1;
8294 /*
8295 * before atomic variable goto zero, we must
8296 * make sure dip->errors is perceived to be set.
8297 */
8298 smp_mb__before_atomic();
8299 if (atomic_dec_and_test(&dip->pending_bios))
8300 bio_io_error(dip->orig_bio);
8301
8302 /* bio_end_io() will handle error, so we needn't return it */
8303 return 0;
8304 }
8305
8306 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8307 struct inode *inode, loff_t file_offset)
8308 {
8309 struct btrfs_dio_private *dip = NULL;
8310 struct bio *io_bio = NULL;
8311 struct btrfs_io_bio *btrfs_bio;
8312 int skip_sum;
8313 int write = rw & REQ_WRITE;
8314 int ret = 0;
8315
8316 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8317
8318 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8319 if (!io_bio) {
8320 ret = -ENOMEM;
8321 goto free_ordered;
8322 }
8323
8324 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8325 if (!dip) {
8326 ret = -ENOMEM;
8327 goto free_ordered;
8328 }
8329
8330 dip->private = dio_bio->bi_private;
8331 dip->inode = inode;
8332 dip->logical_offset = file_offset;
8333 dip->bytes = dio_bio->bi_iter.bi_size;
8334 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8335 io_bio->bi_private = dip;
8336 dip->orig_bio = io_bio;
8337 dip->dio_bio = dio_bio;
8338 atomic_set(&dip->pending_bios, 0);
8339 btrfs_bio = btrfs_io_bio(io_bio);
8340 btrfs_bio->logical = file_offset;
8341
8342 if (write) {
8343 io_bio->bi_end_io = btrfs_endio_direct_write;
8344 } else {
8345 io_bio->bi_end_io = btrfs_endio_direct_read;
8346 dip->subio_endio = btrfs_subio_endio_read;
8347 }
8348
8349 /*
8350 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8351 * even if we fail to submit a bio, because in such case we do the
8352 * corresponding error handling below and it must not be done a second
8353 * time by btrfs_direct_IO().
8354 */
8355 if (write) {
8356 struct btrfs_dio_data *dio_data = current->journal_info;
8357
8358 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8359 dip->bytes;
8360 dio_data->unsubmitted_oe_range_start =
8361 dio_data->unsubmitted_oe_range_end;
8362 }
8363
8364 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8365 if (!ret)
8366 return;
8367
8368 if (btrfs_bio->end_io)
8369 btrfs_bio->end_io(btrfs_bio, ret);
8370
8371 free_ordered:
8372 /*
8373 * If we arrived here it means either we failed to submit the dip
8374 * or we either failed to clone the dio_bio or failed to allocate the
8375 * dip. If we cloned the dio_bio and allocated the dip, we can just
8376 * call bio_endio against our io_bio so that we get proper resource
8377 * cleanup if we fail to submit the dip, otherwise, we must do the
8378 * same as btrfs_endio_direct_[write|read] because we can't call these
8379 * callbacks - they require an allocated dip and a clone of dio_bio.
8380 */
8381 if (io_bio && dip) {
8382 io_bio->bi_error = -EIO;
8383 bio_endio(io_bio);
8384 /*
8385 * The end io callbacks free our dip, do the final put on io_bio
8386 * and all the cleanup and final put for dio_bio (through
8387 * dio_end_io()).
8388 */
8389 dip = NULL;
8390 io_bio = NULL;
8391 } else {
8392 if (write)
8393 btrfs_endio_direct_write_update_ordered(inode,
8394 file_offset,
8395 dio_bio->bi_iter.bi_size,
8396 0);
8397 else
8398 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8399 file_offset + dio_bio->bi_iter.bi_size - 1);
8400
8401 dio_bio->bi_error = -EIO;
8402 /*
8403 * Releases and cleans up our dio_bio, no need to bio_put()
8404 * nor bio_endio()/bio_io_error() against dio_bio.
8405 */
8406 dio_end_io(dio_bio, ret);
8407 }
8408 if (io_bio)
8409 bio_put(io_bio);
8410 kfree(dip);
8411 }
8412
8413 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8414 const struct iov_iter *iter, loff_t offset)
8415 {
8416 int seg;
8417 int i;
8418 unsigned blocksize_mask = root->sectorsize - 1;
8419 ssize_t retval = -EINVAL;
8420
8421 if (offset & blocksize_mask)
8422 goto out;
8423
8424 if (iov_iter_alignment(iter) & blocksize_mask)
8425 goto out;
8426
8427 /* If this is a write we don't need to check anymore */
8428 if (iov_iter_rw(iter) == WRITE)
8429 return 0;
8430 /*
8431 * Check to make sure we don't have duplicate iov_base's in this
8432 * iovec, if so return EINVAL, otherwise we'll get csum errors
8433 * when reading back.
8434 */
8435 for (seg = 0; seg < iter->nr_segs; seg++) {
8436 for (i = seg + 1; i < iter->nr_segs; i++) {
8437 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8438 goto out;
8439 }
8440 }
8441 retval = 0;
8442 out:
8443 return retval;
8444 }
8445
8446 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8447 loff_t offset)
8448 {
8449 struct file *file = iocb->ki_filp;
8450 struct inode *inode = file->f_mapping->host;
8451 struct btrfs_root *root = BTRFS_I(inode)->root;
8452 struct btrfs_dio_data dio_data = { 0 };
8453 size_t count = 0;
8454 int flags = 0;
8455 bool wakeup = true;
8456 bool relock = false;
8457 ssize_t ret;
8458
8459 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8460 return 0;
8461
8462 inode_dio_begin(inode);
8463 smp_mb__after_atomic();
8464
8465 /*
8466 * The generic stuff only does filemap_write_and_wait_range, which
8467 * isn't enough if we've written compressed pages to this area, so
8468 * we need to flush the dirty pages again to make absolutely sure
8469 * that any outstanding dirty pages are on disk.
8470 */
8471 count = iov_iter_count(iter);
8472 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8473 &BTRFS_I(inode)->runtime_flags))
8474 filemap_fdatawrite_range(inode->i_mapping, offset,
8475 offset + count - 1);
8476
8477 if (iov_iter_rw(iter) == WRITE) {
8478 /*
8479 * If the write DIO is beyond the EOF, we need update
8480 * the isize, but it is protected by i_mutex. So we can
8481 * not unlock the i_mutex at this case.
8482 */
8483 if (offset + count <= inode->i_size) {
8484 inode_unlock(inode);
8485 relock = true;
8486 }
8487 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8488 if (ret)
8489 goto out;
8490 dio_data.outstanding_extents = div64_u64(count +
8491 BTRFS_MAX_EXTENT_SIZE - 1,
8492 BTRFS_MAX_EXTENT_SIZE);
8493
8494 /*
8495 * We need to know how many extents we reserved so that we can
8496 * do the accounting properly if we go over the number we
8497 * originally calculated. Abuse current->journal_info for this.
8498 */
8499 dio_data.reserve = round_up(count, root->sectorsize);
8500 dio_data.unsubmitted_oe_range_start = (u64)offset;
8501 dio_data.unsubmitted_oe_range_end = (u64)offset;
8502 current->journal_info = &dio_data;
8503 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8504 &BTRFS_I(inode)->runtime_flags)) {
8505 inode_dio_end(inode);
8506 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8507 wakeup = false;
8508 }
8509
8510 ret = __blockdev_direct_IO(iocb, inode,
8511 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8512 iter, offset, btrfs_get_blocks_direct, NULL,
8513 btrfs_submit_direct, flags);
8514 if (iov_iter_rw(iter) == WRITE) {
8515 current->journal_info = NULL;
8516 if (ret < 0 && ret != -EIOCBQUEUED) {
8517 if (dio_data.reserve)
8518 btrfs_delalloc_release_space(inode, offset,
8519 dio_data.reserve);
8520 /*
8521 * On error we might have left some ordered extents
8522 * without submitting corresponding bios for them, so
8523 * cleanup them up to avoid other tasks getting them
8524 * and waiting for them to complete forever.
8525 */
8526 if (dio_data.unsubmitted_oe_range_start <
8527 dio_data.unsubmitted_oe_range_end)
8528 btrfs_endio_direct_write_update_ordered(inode,
8529 dio_data.unsubmitted_oe_range_start,
8530 dio_data.unsubmitted_oe_range_end -
8531 dio_data.unsubmitted_oe_range_start,
8532 0);
8533 } else if (ret >= 0 && (size_t)ret < count)
8534 btrfs_delalloc_release_space(inode, offset,
8535 count - (size_t)ret);
8536 }
8537 out:
8538 if (wakeup)
8539 inode_dio_end(inode);
8540 if (relock)
8541 inode_lock(inode);
8542
8543 return ret;
8544 }
8545
8546 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8547
8548 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8549 __u64 start, __u64 len)
8550 {
8551 int ret;
8552
8553 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8554 if (ret)
8555 return ret;
8556
8557 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8558 }
8559
8560 int btrfs_readpage(struct file *file, struct page *page)
8561 {
8562 struct extent_io_tree *tree;
8563 tree = &BTRFS_I(page->mapping->host)->io_tree;
8564 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8565 }
8566
8567 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8568 {
8569 struct extent_io_tree *tree;
8570 struct inode *inode = page->mapping->host;
8571 int ret;
8572
8573 if (current->flags & PF_MEMALLOC) {
8574 redirty_page_for_writepage(wbc, page);
8575 unlock_page(page);
8576 return 0;
8577 }
8578
8579 /*
8580 * If we are under memory pressure we will call this directly from the
8581 * VM, we need to make sure we have the inode referenced for the ordered
8582 * extent. If not just return like we didn't do anything.
8583 */
8584 if (!igrab(inode)) {
8585 redirty_page_for_writepage(wbc, page);
8586 return AOP_WRITEPAGE_ACTIVATE;
8587 }
8588 tree = &BTRFS_I(page->mapping->host)->io_tree;
8589 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8590 btrfs_add_delayed_iput(inode);
8591 return ret;
8592 }
8593
8594 static int btrfs_writepages(struct address_space *mapping,
8595 struct writeback_control *wbc)
8596 {
8597 struct extent_io_tree *tree;
8598
8599 tree = &BTRFS_I(mapping->host)->io_tree;
8600 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8601 }
8602
8603 static int
8604 btrfs_readpages(struct file *file, struct address_space *mapping,
8605 struct list_head *pages, unsigned nr_pages)
8606 {
8607 struct extent_io_tree *tree;
8608 tree = &BTRFS_I(mapping->host)->io_tree;
8609 return extent_readpages(tree, mapping, pages, nr_pages,
8610 btrfs_get_extent);
8611 }
8612 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8613 {
8614 struct extent_io_tree *tree;
8615 struct extent_map_tree *map;
8616 int ret;
8617
8618 tree = &BTRFS_I(page->mapping->host)->io_tree;
8619 map = &BTRFS_I(page->mapping->host)->extent_tree;
8620 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8621 if (ret == 1) {
8622 ClearPagePrivate(page);
8623 set_page_private(page, 0);
8624 page_cache_release(page);
8625 }
8626 return ret;
8627 }
8628
8629 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8630 {
8631 if (PageWriteback(page) || PageDirty(page))
8632 return 0;
8633 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8634 }
8635
8636 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8637 unsigned int length)
8638 {
8639 struct inode *inode = page->mapping->host;
8640 struct extent_io_tree *tree;
8641 struct btrfs_ordered_extent *ordered;
8642 struct extent_state *cached_state = NULL;
8643 u64 page_start = page_offset(page);
8644 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8645 int inode_evicting = inode->i_state & I_FREEING;
8646
8647 /*
8648 * we have the page locked, so new writeback can't start,
8649 * and the dirty bit won't be cleared while we are here.
8650 *
8651 * Wait for IO on this page so that we can safely clear
8652 * the PagePrivate2 bit and do ordered accounting
8653 */
8654 wait_on_page_writeback(page);
8655
8656 tree = &BTRFS_I(inode)->io_tree;
8657 if (offset) {
8658 btrfs_releasepage(page, GFP_NOFS);
8659 return;
8660 }
8661
8662 if (!inode_evicting)
8663 lock_extent_bits(tree, page_start, page_end, &cached_state);
8664 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8665 if (ordered) {
8666 /*
8667 * IO on this page will never be started, so we need
8668 * to account for any ordered extents now
8669 */
8670 if (!inode_evicting)
8671 clear_extent_bit(tree, page_start, page_end,
8672 EXTENT_DIRTY | EXTENT_DELALLOC |
8673 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8674 EXTENT_DEFRAG, 1, 0, &cached_state,
8675 GFP_NOFS);
8676 /*
8677 * whoever cleared the private bit is responsible
8678 * for the finish_ordered_io
8679 */
8680 if (TestClearPagePrivate2(page)) {
8681 struct btrfs_ordered_inode_tree *tree;
8682 u64 new_len;
8683
8684 tree = &BTRFS_I(inode)->ordered_tree;
8685
8686 spin_lock_irq(&tree->lock);
8687 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8688 new_len = page_start - ordered->file_offset;
8689 if (new_len < ordered->truncated_len)
8690 ordered->truncated_len = new_len;
8691 spin_unlock_irq(&tree->lock);
8692
8693 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8694 page_start,
8695 PAGE_CACHE_SIZE, 1))
8696 btrfs_finish_ordered_io(ordered);
8697 }
8698 btrfs_put_ordered_extent(ordered);
8699 if (!inode_evicting) {
8700 cached_state = NULL;
8701 lock_extent_bits(tree, page_start, page_end,
8702 &cached_state);
8703 }
8704 }
8705
8706 /*
8707 * Qgroup reserved space handler
8708 * Page here will be either
8709 * 1) Already written to disk
8710 * In this case, its reserved space is released from data rsv map
8711 * and will be freed by delayed_ref handler finally.
8712 * So even we call qgroup_free_data(), it won't decrease reserved
8713 * space.
8714 * 2) Not written to disk
8715 * This means the reserved space should be freed here.
8716 */
8717 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8718 if (!inode_evicting) {
8719 clear_extent_bit(tree, page_start, page_end,
8720 EXTENT_LOCKED | EXTENT_DIRTY |
8721 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8722 EXTENT_DEFRAG, 1, 1,
8723 &cached_state, GFP_NOFS);
8724
8725 __btrfs_releasepage(page, GFP_NOFS);
8726 }
8727
8728 ClearPageChecked(page);
8729 if (PagePrivate(page)) {
8730 ClearPagePrivate(page);
8731 set_page_private(page, 0);
8732 page_cache_release(page);
8733 }
8734 }
8735
8736 /*
8737 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8738 * called from a page fault handler when a page is first dirtied. Hence we must
8739 * be careful to check for EOF conditions here. We set the page up correctly
8740 * for a written page which means we get ENOSPC checking when writing into
8741 * holes and correct delalloc and unwritten extent mapping on filesystems that
8742 * support these features.
8743 *
8744 * We are not allowed to take the i_mutex here so we have to play games to
8745 * protect against truncate races as the page could now be beyond EOF. Because
8746 * vmtruncate() writes the inode size before removing pages, once we have the
8747 * page lock we can determine safely if the page is beyond EOF. If it is not
8748 * beyond EOF, then the page is guaranteed safe against truncation until we
8749 * unlock the page.
8750 */
8751 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8752 {
8753 struct page *page = vmf->page;
8754 struct inode *inode = file_inode(vma->vm_file);
8755 struct btrfs_root *root = BTRFS_I(inode)->root;
8756 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8757 struct btrfs_ordered_extent *ordered;
8758 struct extent_state *cached_state = NULL;
8759 char *kaddr;
8760 unsigned long zero_start;
8761 loff_t size;
8762 int ret;
8763 int reserved = 0;
8764 u64 page_start;
8765 u64 page_end;
8766
8767 sb_start_pagefault(inode->i_sb);
8768 page_start = page_offset(page);
8769 page_end = page_start + PAGE_CACHE_SIZE - 1;
8770
8771 ret = btrfs_delalloc_reserve_space(inode, page_start,
8772 PAGE_CACHE_SIZE);
8773 if (!ret) {
8774 ret = file_update_time(vma->vm_file);
8775 reserved = 1;
8776 }
8777 if (ret) {
8778 if (ret == -ENOMEM)
8779 ret = VM_FAULT_OOM;
8780 else /* -ENOSPC, -EIO, etc */
8781 ret = VM_FAULT_SIGBUS;
8782 if (reserved)
8783 goto out;
8784 goto out_noreserve;
8785 }
8786
8787 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8788 again:
8789 lock_page(page);
8790 size = i_size_read(inode);
8791
8792 if ((page->mapping != inode->i_mapping) ||
8793 (page_start >= size)) {
8794 /* page got truncated out from underneath us */
8795 goto out_unlock;
8796 }
8797 wait_on_page_writeback(page);
8798
8799 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8800 set_page_extent_mapped(page);
8801
8802 /*
8803 * we can't set the delalloc bits if there are pending ordered
8804 * extents. Drop our locks and wait for them to finish
8805 */
8806 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8807 if (ordered) {
8808 unlock_extent_cached(io_tree, page_start, page_end,
8809 &cached_state, GFP_NOFS);
8810 unlock_page(page);
8811 btrfs_start_ordered_extent(inode, ordered, 1);
8812 btrfs_put_ordered_extent(ordered);
8813 goto again;
8814 }
8815
8816 /*
8817 * XXX - page_mkwrite gets called every time the page is dirtied, even
8818 * if it was already dirty, so for space accounting reasons we need to
8819 * clear any delalloc bits for the range we are fixing to save. There
8820 * is probably a better way to do this, but for now keep consistent with
8821 * prepare_pages in the normal write path.
8822 */
8823 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8824 EXTENT_DIRTY | EXTENT_DELALLOC |
8825 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8826 0, 0, &cached_state, GFP_NOFS);
8827
8828 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8829 &cached_state);
8830 if (ret) {
8831 unlock_extent_cached(io_tree, page_start, page_end,
8832 &cached_state, GFP_NOFS);
8833 ret = VM_FAULT_SIGBUS;
8834 goto out_unlock;
8835 }
8836 ret = 0;
8837
8838 /* page is wholly or partially inside EOF */
8839 if (page_start + PAGE_CACHE_SIZE > size)
8840 zero_start = size & ~PAGE_CACHE_MASK;
8841 else
8842 zero_start = PAGE_CACHE_SIZE;
8843
8844 if (zero_start != PAGE_CACHE_SIZE) {
8845 kaddr = kmap(page);
8846 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8847 flush_dcache_page(page);
8848 kunmap(page);
8849 }
8850 ClearPageChecked(page);
8851 set_page_dirty(page);
8852 SetPageUptodate(page);
8853
8854 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8855 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8856 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8857
8858 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8859
8860 out_unlock:
8861 if (!ret) {
8862 sb_end_pagefault(inode->i_sb);
8863 return VM_FAULT_LOCKED;
8864 }
8865 unlock_page(page);
8866 out:
8867 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8868 out_noreserve:
8869 sb_end_pagefault(inode->i_sb);
8870 return ret;
8871 }
8872
8873 static int btrfs_truncate(struct inode *inode)
8874 {
8875 struct btrfs_root *root = BTRFS_I(inode)->root;
8876 struct btrfs_block_rsv *rsv;
8877 int ret = 0;
8878 int err = 0;
8879 struct btrfs_trans_handle *trans;
8880 u64 mask = root->sectorsize - 1;
8881 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8882
8883 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8884 (u64)-1);
8885 if (ret)
8886 return ret;
8887
8888 /*
8889 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8890 * 3 things going on here
8891 *
8892 * 1) We need to reserve space for our orphan item and the space to
8893 * delete our orphan item. Lord knows we don't want to have a dangling
8894 * orphan item because we didn't reserve space to remove it.
8895 *
8896 * 2) We need to reserve space to update our inode.
8897 *
8898 * 3) We need to have something to cache all the space that is going to
8899 * be free'd up by the truncate operation, but also have some slack
8900 * space reserved in case it uses space during the truncate (thank you
8901 * very much snapshotting).
8902 *
8903 * And we need these to all be seperate. The fact is we can use alot of
8904 * space doing the truncate, and we have no earthly idea how much space
8905 * we will use, so we need the truncate reservation to be seperate so it
8906 * doesn't end up using space reserved for updating the inode or
8907 * removing the orphan item. We also need to be able to stop the
8908 * transaction and start a new one, which means we need to be able to
8909 * update the inode several times, and we have no idea of knowing how
8910 * many times that will be, so we can't just reserve 1 item for the
8911 * entirety of the opration, so that has to be done seperately as well.
8912 * Then there is the orphan item, which does indeed need to be held on
8913 * to for the whole operation, and we need nobody to touch this reserved
8914 * space except the orphan code.
8915 *
8916 * So that leaves us with
8917 *
8918 * 1) root->orphan_block_rsv - for the orphan deletion.
8919 * 2) rsv - for the truncate reservation, which we will steal from the
8920 * transaction reservation.
8921 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8922 * updating the inode.
8923 */
8924 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8925 if (!rsv)
8926 return -ENOMEM;
8927 rsv->size = min_size;
8928 rsv->failfast = 1;
8929
8930 /*
8931 * 1 for the truncate slack space
8932 * 1 for updating the inode.
8933 */
8934 trans = btrfs_start_transaction(root, 2);
8935 if (IS_ERR(trans)) {
8936 err = PTR_ERR(trans);
8937 goto out;
8938 }
8939
8940 /* Migrate the slack space for the truncate to our reserve */
8941 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8942 min_size);
8943 BUG_ON(ret);
8944
8945 /*
8946 * So if we truncate and then write and fsync we normally would just
8947 * write the extents that changed, which is a problem if we need to
8948 * first truncate that entire inode. So set this flag so we write out
8949 * all of the extents in the inode to the sync log so we're completely
8950 * safe.
8951 */
8952 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8953 trans->block_rsv = rsv;
8954
8955 while (1) {
8956 ret = btrfs_truncate_inode_items(trans, root, inode,
8957 inode->i_size,
8958 BTRFS_EXTENT_DATA_KEY);
8959 if (ret != -ENOSPC && ret != -EAGAIN) {
8960 err = ret;
8961 break;
8962 }
8963
8964 trans->block_rsv = &root->fs_info->trans_block_rsv;
8965 ret = btrfs_update_inode(trans, root, inode);
8966 if (ret) {
8967 err = ret;
8968 break;
8969 }
8970
8971 btrfs_end_transaction(trans, root);
8972 btrfs_btree_balance_dirty(root);
8973
8974 trans = btrfs_start_transaction(root, 2);
8975 if (IS_ERR(trans)) {
8976 ret = err = PTR_ERR(trans);
8977 trans = NULL;
8978 break;
8979 }
8980
8981 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8982 rsv, min_size);
8983 BUG_ON(ret); /* shouldn't happen */
8984 trans->block_rsv = rsv;
8985 }
8986
8987 if (ret == 0 && inode->i_nlink > 0) {
8988 trans->block_rsv = root->orphan_block_rsv;
8989 ret = btrfs_orphan_del(trans, inode);
8990 if (ret)
8991 err = ret;
8992 }
8993
8994 if (trans) {
8995 trans->block_rsv = &root->fs_info->trans_block_rsv;
8996 ret = btrfs_update_inode(trans, root, inode);
8997 if (ret && !err)
8998 err = ret;
8999
9000 ret = btrfs_end_transaction(trans, root);
9001 btrfs_btree_balance_dirty(root);
9002 }
9003
9004 out:
9005 btrfs_free_block_rsv(root, rsv);
9006
9007 if (ret && !err)
9008 err = ret;
9009
9010 return err;
9011 }
9012
9013 /*
9014 * create a new subvolume directory/inode (helper for the ioctl).
9015 */
9016 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9017 struct btrfs_root *new_root,
9018 struct btrfs_root *parent_root,
9019 u64 new_dirid)
9020 {
9021 struct inode *inode;
9022 int err;
9023 u64 index = 0;
9024
9025 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9026 new_dirid, new_dirid,
9027 S_IFDIR | (~current_umask() & S_IRWXUGO),
9028 &index);
9029 if (IS_ERR(inode))
9030 return PTR_ERR(inode);
9031 inode->i_op = &btrfs_dir_inode_operations;
9032 inode->i_fop = &btrfs_dir_file_operations;
9033
9034 set_nlink(inode, 1);
9035 btrfs_i_size_write(inode, 0);
9036 unlock_new_inode(inode);
9037
9038 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9039 if (err)
9040 btrfs_err(new_root->fs_info,
9041 "error inheriting subvolume %llu properties: %d",
9042 new_root->root_key.objectid, err);
9043
9044 err = btrfs_update_inode(trans, new_root, inode);
9045
9046 iput(inode);
9047 return err;
9048 }
9049
9050 struct inode *btrfs_alloc_inode(struct super_block *sb)
9051 {
9052 struct btrfs_inode *ei;
9053 struct inode *inode;
9054
9055 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9056 if (!ei)
9057 return NULL;
9058
9059 ei->root = NULL;
9060 ei->generation = 0;
9061 ei->last_trans = 0;
9062 ei->last_sub_trans = 0;
9063 ei->logged_trans = 0;
9064 ei->delalloc_bytes = 0;
9065 ei->defrag_bytes = 0;
9066 ei->disk_i_size = 0;
9067 ei->flags = 0;
9068 ei->csum_bytes = 0;
9069 ei->index_cnt = (u64)-1;
9070 ei->dir_index = 0;
9071 ei->last_unlink_trans = 0;
9072 ei->last_log_commit = 0;
9073 ei->delayed_iput_count = 0;
9074
9075 spin_lock_init(&ei->lock);
9076 ei->outstanding_extents = 0;
9077 ei->reserved_extents = 0;
9078
9079 ei->runtime_flags = 0;
9080 ei->force_compress = BTRFS_COMPRESS_NONE;
9081
9082 ei->delayed_node = NULL;
9083
9084 ei->i_otime.tv_sec = 0;
9085 ei->i_otime.tv_nsec = 0;
9086
9087 inode = &ei->vfs_inode;
9088 extent_map_tree_init(&ei->extent_tree);
9089 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9090 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9091 ei->io_tree.track_uptodate = 1;
9092 ei->io_failure_tree.track_uptodate = 1;
9093 atomic_set(&ei->sync_writers, 0);
9094 mutex_init(&ei->log_mutex);
9095 mutex_init(&ei->delalloc_mutex);
9096 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9097 INIT_LIST_HEAD(&ei->delalloc_inodes);
9098 INIT_LIST_HEAD(&ei->delayed_iput);
9099 RB_CLEAR_NODE(&ei->rb_node);
9100
9101 return inode;
9102 }
9103
9104 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9105 void btrfs_test_destroy_inode(struct inode *inode)
9106 {
9107 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9108 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9109 }
9110 #endif
9111
9112 static void btrfs_i_callback(struct rcu_head *head)
9113 {
9114 struct inode *inode = container_of(head, struct inode, i_rcu);
9115 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9116 }
9117
9118 void btrfs_destroy_inode(struct inode *inode)
9119 {
9120 struct btrfs_ordered_extent *ordered;
9121 struct btrfs_root *root = BTRFS_I(inode)->root;
9122
9123 WARN_ON(!hlist_empty(&inode->i_dentry));
9124 WARN_ON(inode->i_data.nrpages);
9125 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9126 WARN_ON(BTRFS_I(inode)->reserved_extents);
9127 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9128 WARN_ON(BTRFS_I(inode)->csum_bytes);
9129 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9130
9131 /*
9132 * This can happen where we create an inode, but somebody else also
9133 * created the same inode and we need to destroy the one we already
9134 * created.
9135 */
9136 if (!root)
9137 goto free;
9138
9139 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9140 &BTRFS_I(inode)->runtime_flags)) {
9141 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9142 btrfs_ino(inode));
9143 atomic_dec(&root->orphan_inodes);
9144 }
9145
9146 while (1) {
9147 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9148 if (!ordered)
9149 break;
9150 else {
9151 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9152 ordered->file_offset, ordered->len);
9153 btrfs_remove_ordered_extent(inode, ordered);
9154 btrfs_put_ordered_extent(ordered);
9155 btrfs_put_ordered_extent(ordered);
9156 }
9157 }
9158 btrfs_qgroup_check_reserved_leak(inode);
9159 inode_tree_del(inode);
9160 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9161 free:
9162 call_rcu(&inode->i_rcu, btrfs_i_callback);
9163 }
9164
9165 int btrfs_drop_inode(struct inode *inode)
9166 {
9167 struct btrfs_root *root = BTRFS_I(inode)->root;
9168
9169 if (root == NULL)
9170 return 1;
9171
9172 /* the snap/subvol tree is on deleting */
9173 if (btrfs_root_refs(&root->root_item) == 0)
9174 return 1;
9175 else
9176 return generic_drop_inode(inode);
9177 }
9178
9179 static void init_once(void *foo)
9180 {
9181 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9182
9183 inode_init_once(&ei->vfs_inode);
9184 }
9185
9186 void btrfs_destroy_cachep(void)
9187 {
9188 /*
9189 * Make sure all delayed rcu free inodes are flushed before we
9190 * destroy cache.
9191 */
9192 rcu_barrier();
9193 if (btrfs_inode_cachep)
9194 kmem_cache_destroy(btrfs_inode_cachep);
9195 if (btrfs_trans_handle_cachep)
9196 kmem_cache_destroy(btrfs_trans_handle_cachep);
9197 if (btrfs_transaction_cachep)
9198 kmem_cache_destroy(btrfs_transaction_cachep);
9199 if (btrfs_path_cachep)
9200 kmem_cache_destroy(btrfs_path_cachep);
9201 if (btrfs_free_space_cachep)
9202 kmem_cache_destroy(btrfs_free_space_cachep);
9203 }
9204
9205 int btrfs_init_cachep(void)
9206 {
9207 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9208 sizeof(struct btrfs_inode), 0,
9209 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9210 init_once);
9211 if (!btrfs_inode_cachep)
9212 goto fail;
9213
9214 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9215 sizeof(struct btrfs_trans_handle), 0,
9216 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9217 if (!btrfs_trans_handle_cachep)
9218 goto fail;
9219
9220 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9221 sizeof(struct btrfs_transaction), 0,
9222 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9223 if (!btrfs_transaction_cachep)
9224 goto fail;
9225
9226 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9227 sizeof(struct btrfs_path), 0,
9228 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9229 if (!btrfs_path_cachep)
9230 goto fail;
9231
9232 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9233 sizeof(struct btrfs_free_space), 0,
9234 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9235 if (!btrfs_free_space_cachep)
9236 goto fail;
9237
9238 return 0;
9239 fail:
9240 btrfs_destroy_cachep();
9241 return -ENOMEM;
9242 }
9243
9244 static int btrfs_getattr(struct vfsmount *mnt,
9245 struct dentry *dentry, struct kstat *stat)
9246 {
9247 u64 delalloc_bytes;
9248 struct inode *inode = d_inode(dentry);
9249 u32 blocksize = inode->i_sb->s_blocksize;
9250
9251 generic_fillattr(inode, stat);
9252 stat->dev = BTRFS_I(inode)->root->anon_dev;
9253 stat->blksize = PAGE_CACHE_SIZE;
9254
9255 spin_lock(&BTRFS_I(inode)->lock);
9256 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9257 spin_unlock(&BTRFS_I(inode)->lock);
9258 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9259 ALIGN(delalloc_bytes, blocksize)) >> 9;
9260 return 0;
9261 }
9262
9263 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9264 struct inode *new_dir, struct dentry *new_dentry)
9265 {
9266 struct btrfs_trans_handle *trans;
9267 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9268 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9269 struct inode *new_inode = d_inode(new_dentry);
9270 struct inode *old_inode = d_inode(old_dentry);
9271 struct timespec ctime = CURRENT_TIME;
9272 u64 index = 0;
9273 u64 root_objectid;
9274 int ret;
9275 u64 old_ino = btrfs_ino(old_inode);
9276
9277 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9278 return -EPERM;
9279
9280 /* we only allow rename subvolume link between subvolumes */
9281 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9282 return -EXDEV;
9283
9284 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9285 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9286 return -ENOTEMPTY;
9287
9288 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9289 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9290 return -ENOTEMPTY;
9291
9292
9293 /* check for collisions, even if the name isn't there */
9294 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9295 new_dentry->d_name.name,
9296 new_dentry->d_name.len);
9297
9298 if (ret) {
9299 if (ret == -EEXIST) {
9300 /* we shouldn't get
9301 * eexist without a new_inode */
9302 if (WARN_ON(!new_inode)) {
9303 return ret;
9304 }
9305 } else {
9306 /* maybe -EOVERFLOW */
9307 return ret;
9308 }
9309 }
9310 ret = 0;
9311
9312 /*
9313 * we're using rename to replace one file with another. Start IO on it
9314 * now so we don't add too much work to the end of the transaction
9315 */
9316 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9317 filemap_flush(old_inode->i_mapping);
9318
9319 /* close the racy window with snapshot create/destroy ioctl */
9320 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9321 down_read(&root->fs_info->subvol_sem);
9322 /*
9323 * We want to reserve the absolute worst case amount of items. So if
9324 * both inodes are subvols and we need to unlink them then that would
9325 * require 4 item modifications, but if they are both normal inodes it
9326 * would require 5 item modifications, so we'll assume their normal
9327 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9328 * should cover the worst case number of items we'll modify.
9329 */
9330 trans = btrfs_start_transaction(root, 11);
9331 if (IS_ERR(trans)) {
9332 ret = PTR_ERR(trans);
9333 goto out_notrans;
9334 }
9335
9336 if (dest != root)
9337 btrfs_record_root_in_trans(trans, dest);
9338
9339 ret = btrfs_set_inode_index(new_dir, &index);
9340 if (ret)
9341 goto out_fail;
9342
9343 BTRFS_I(old_inode)->dir_index = 0ULL;
9344 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9345 /* force full log commit if subvolume involved. */
9346 btrfs_set_log_full_commit(root->fs_info, trans);
9347 } else {
9348 ret = btrfs_insert_inode_ref(trans, dest,
9349 new_dentry->d_name.name,
9350 new_dentry->d_name.len,
9351 old_ino,
9352 btrfs_ino(new_dir), index);
9353 if (ret)
9354 goto out_fail;
9355 /*
9356 * this is an ugly little race, but the rename is required
9357 * to make sure that if we crash, the inode is either at the
9358 * old name or the new one. pinning the log transaction lets
9359 * us make sure we don't allow a log commit to come in after
9360 * we unlink the name but before we add the new name back in.
9361 */
9362 btrfs_pin_log_trans(root);
9363 }
9364
9365 inode_inc_iversion(old_dir);
9366 inode_inc_iversion(new_dir);
9367 inode_inc_iversion(old_inode);
9368 old_dir->i_ctime = old_dir->i_mtime = ctime;
9369 new_dir->i_ctime = new_dir->i_mtime = ctime;
9370 old_inode->i_ctime = ctime;
9371
9372 if (old_dentry->d_parent != new_dentry->d_parent)
9373 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9374
9375 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9376 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9377 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9378 old_dentry->d_name.name,
9379 old_dentry->d_name.len);
9380 } else {
9381 ret = __btrfs_unlink_inode(trans, root, old_dir,
9382 d_inode(old_dentry),
9383 old_dentry->d_name.name,
9384 old_dentry->d_name.len);
9385 if (!ret)
9386 ret = btrfs_update_inode(trans, root, old_inode);
9387 }
9388 if (ret) {
9389 btrfs_abort_transaction(trans, root, ret);
9390 goto out_fail;
9391 }
9392
9393 if (new_inode) {
9394 inode_inc_iversion(new_inode);
9395 new_inode->i_ctime = CURRENT_TIME;
9396 if (unlikely(btrfs_ino(new_inode) ==
9397 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9398 root_objectid = BTRFS_I(new_inode)->location.objectid;
9399 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9400 root_objectid,
9401 new_dentry->d_name.name,
9402 new_dentry->d_name.len);
9403 BUG_ON(new_inode->i_nlink == 0);
9404 } else {
9405 ret = btrfs_unlink_inode(trans, dest, new_dir,
9406 d_inode(new_dentry),
9407 new_dentry->d_name.name,
9408 new_dentry->d_name.len);
9409 }
9410 if (!ret && new_inode->i_nlink == 0)
9411 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9412 if (ret) {
9413 btrfs_abort_transaction(trans, root, ret);
9414 goto out_fail;
9415 }
9416 }
9417
9418 ret = btrfs_add_link(trans, new_dir, old_inode,
9419 new_dentry->d_name.name,
9420 new_dentry->d_name.len, 0, index);
9421 if (ret) {
9422 btrfs_abort_transaction(trans, root, ret);
9423 goto out_fail;
9424 }
9425
9426 if (old_inode->i_nlink == 1)
9427 BTRFS_I(old_inode)->dir_index = index;
9428
9429 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9430 struct dentry *parent = new_dentry->d_parent;
9431 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9432 btrfs_end_log_trans(root);
9433 }
9434 out_fail:
9435 btrfs_end_transaction(trans, root);
9436 out_notrans:
9437 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9438 up_read(&root->fs_info->subvol_sem);
9439
9440 return ret;
9441 }
9442
9443 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9444 struct inode *new_dir, struct dentry *new_dentry,
9445 unsigned int flags)
9446 {
9447 if (flags & ~RENAME_NOREPLACE)
9448 return -EINVAL;
9449
9450 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9451 }
9452
9453 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9454 {
9455 struct btrfs_delalloc_work *delalloc_work;
9456 struct inode *inode;
9457
9458 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9459 work);
9460 inode = delalloc_work->inode;
9461 filemap_flush(inode->i_mapping);
9462 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9463 &BTRFS_I(inode)->runtime_flags))
9464 filemap_flush(inode->i_mapping);
9465
9466 if (delalloc_work->delay_iput)
9467 btrfs_add_delayed_iput(inode);
9468 else
9469 iput(inode);
9470 complete(&delalloc_work->completion);
9471 }
9472
9473 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9474 int delay_iput)
9475 {
9476 struct btrfs_delalloc_work *work;
9477
9478 work = kmalloc(sizeof(*work), GFP_NOFS);
9479 if (!work)
9480 return NULL;
9481
9482 init_completion(&work->completion);
9483 INIT_LIST_HEAD(&work->list);
9484 work->inode = inode;
9485 work->delay_iput = delay_iput;
9486 WARN_ON_ONCE(!inode);
9487 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9488 btrfs_run_delalloc_work, NULL, NULL);
9489
9490 return work;
9491 }
9492
9493 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9494 {
9495 wait_for_completion(&work->completion);
9496 kfree(work);
9497 }
9498
9499 /*
9500 * some fairly slow code that needs optimization. This walks the list
9501 * of all the inodes with pending delalloc and forces them to disk.
9502 */
9503 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9504 int nr)
9505 {
9506 struct btrfs_inode *binode;
9507 struct inode *inode;
9508 struct btrfs_delalloc_work *work, *next;
9509 struct list_head works;
9510 struct list_head splice;
9511 int ret = 0;
9512
9513 INIT_LIST_HEAD(&works);
9514 INIT_LIST_HEAD(&splice);
9515
9516 mutex_lock(&root->delalloc_mutex);
9517 spin_lock(&root->delalloc_lock);
9518 list_splice_init(&root->delalloc_inodes, &splice);
9519 while (!list_empty(&splice)) {
9520 binode = list_entry(splice.next, struct btrfs_inode,
9521 delalloc_inodes);
9522
9523 list_move_tail(&binode->delalloc_inodes,
9524 &root->delalloc_inodes);
9525 inode = igrab(&binode->vfs_inode);
9526 if (!inode) {
9527 cond_resched_lock(&root->delalloc_lock);
9528 continue;
9529 }
9530 spin_unlock(&root->delalloc_lock);
9531
9532 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9533 if (!work) {
9534 if (delay_iput)
9535 btrfs_add_delayed_iput(inode);
9536 else
9537 iput(inode);
9538 ret = -ENOMEM;
9539 goto out;
9540 }
9541 list_add_tail(&work->list, &works);
9542 btrfs_queue_work(root->fs_info->flush_workers,
9543 &work->work);
9544 ret++;
9545 if (nr != -1 && ret >= nr)
9546 goto out;
9547 cond_resched();
9548 spin_lock(&root->delalloc_lock);
9549 }
9550 spin_unlock(&root->delalloc_lock);
9551
9552 out:
9553 list_for_each_entry_safe(work, next, &works, list) {
9554 list_del_init(&work->list);
9555 btrfs_wait_and_free_delalloc_work(work);
9556 }
9557
9558 if (!list_empty_careful(&splice)) {
9559 spin_lock(&root->delalloc_lock);
9560 list_splice_tail(&splice, &root->delalloc_inodes);
9561 spin_unlock(&root->delalloc_lock);
9562 }
9563 mutex_unlock(&root->delalloc_mutex);
9564 return ret;
9565 }
9566
9567 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9568 {
9569 int ret;
9570
9571 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9572 return -EROFS;
9573
9574 ret = __start_delalloc_inodes(root, delay_iput, -1);
9575 if (ret > 0)
9576 ret = 0;
9577 /*
9578 * the filemap_flush will queue IO into the worker threads, but
9579 * we have to make sure the IO is actually started and that
9580 * ordered extents get created before we return
9581 */
9582 atomic_inc(&root->fs_info->async_submit_draining);
9583 while (atomic_read(&root->fs_info->nr_async_submits) ||
9584 atomic_read(&root->fs_info->async_delalloc_pages)) {
9585 wait_event(root->fs_info->async_submit_wait,
9586 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9587 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9588 }
9589 atomic_dec(&root->fs_info->async_submit_draining);
9590 return ret;
9591 }
9592
9593 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9594 int nr)
9595 {
9596 struct btrfs_root *root;
9597 struct list_head splice;
9598 int ret;
9599
9600 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9601 return -EROFS;
9602
9603 INIT_LIST_HEAD(&splice);
9604
9605 mutex_lock(&fs_info->delalloc_root_mutex);
9606 spin_lock(&fs_info->delalloc_root_lock);
9607 list_splice_init(&fs_info->delalloc_roots, &splice);
9608 while (!list_empty(&splice) && nr) {
9609 root = list_first_entry(&splice, struct btrfs_root,
9610 delalloc_root);
9611 root = btrfs_grab_fs_root(root);
9612 BUG_ON(!root);
9613 list_move_tail(&root->delalloc_root,
9614 &fs_info->delalloc_roots);
9615 spin_unlock(&fs_info->delalloc_root_lock);
9616
9617 ret = __start_delalloc_inodes(root, delay_iput, nr);
9618 btrfs_put_fs_root(root);
9619 if (ret < 0)
9620 goto out;
9621
9622 if (nr != -1) {
9623 nr -= ret;
9624 WARN_ON(nr < 0);
9625 }
9626 spin_lock(&fs_info->delalloc_root_lock);
9627 }
9628 spin_unlock(&fs_info->delalloc_root_lock);
9629
9630 ret = 0;
9631 atomic_inc(&fs_info->async_submit_draining);
9632 while (atomic_read(&fs_info->nr_async_submits) ||
9633 atomic_read(&fs_info->async_delalloc_pages)) {
9634 wait_event(fs_info->async_submit_wait,
9635 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9636 atomic_read(&fs_info->async_delalloc_pages) == 0));
9637 }
9638 atomic_dec(&fs_info->async_submit_draining);
9639 out:
9640 if (!list_empty_careful(&splice)) {
9641 spin_lock(&fs_info->delalloc_root_lock);
9642 list_splice_tail(&splice, &fs_info->delalloc_roots);
9643 spin_unlock(&fs_info->delalloc_root_lock);
9644 }
9645 mutex_unlock(&fs_info->delalloc_root_mutex);
9646 return ret;
9647 }
9648
9649 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9650 const char *symname)
9651 {
9652 struct btrfs_trans_handle *trans;
9653 struct btrfs_root *root = BTRFS_I(dir)->root;
9654 struct btrfs_path *path;
9655 struct btrfs_key key;
9656 struct inode *inode = NULL;
9657 int err;
9658 int drop_inode = 0;
9659 u64 objectid;
9660 u64 index = 0;
9661 int name_len;
9662 int datasize;
9663 unsigned long ptr;
9664 struct btrfs_file_extent_item *ei;
9665 struct extent_buffer *leaf;
9666
9667 name_len = strlen(symname);
9668 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9669 return -ENAMETOOLONG;
9670
9671 /*
9672 * 2 items for inode item and ref
9673 * 2 items for dir items
9674 * 1 item for updating parent inode item
9675 * 1 item for the inline extent item
9676 * 1 item for xattr if selinux is on
9677 */
9678 trans = btrfs_start_transaction(root, 7);
9679 if (IS_ERR(trans))
9680 return PTR_ERR(trans);
9681
9682 err = btrfs_find_free_ino(root, &objectid);
9683 if (err)
9684 goto out_unlock;
9685
9686 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9687 dentry->d_name.len, btrfs_ino(dir), objectid,
9688 S_IFLNK|S_IRWXUGO, &index);
9689 if (IS_ERR(inode)) {
9690 err = PTR_ERR(inode);
9691 goto out_unlock;
9692 }
9693
9694 /*
9695 * If the active LSM wants to access the inode during
9696 * d_instantiate it needs these. Smack checks to see
9697 * if the filesystem supports xattrs by looking at the
9698 * ops vector.
9699 */
9700 inode->i_fop = &btrfs_file_operations;
9701 inode->i_op = &btrfs_file_inode_operations;
9702 inode->i_mapping->a_ops = &btrfs_aops;
9703 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9704
9705 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9706 if (err)
9707 goto out_unlock_inode;
9708
9709 path = btrfs_alloc_path();
9710 if (!path) {
9711 err = -ENOMEM;
9712 goto out_unlock_inode;
9713 }
9714 key.objectid = btrfs_ino(inode);
9715 key.offset = 0;
9716 key.type = BTRFS_EXTENT_DATA_KEY;
9717 datasize = btrfs_file_extent_calc_inline_size(name_len);
9718 err = btrfs_insert_empty_item(trans, root, path, &key,
9719 datasize);
9720 if (err) {
9721 btrfs_free_path(path);
9722 goto out_unlock_inode;
9723 }
9724 leaf = path->nodes[0];
9725 ei = btrfs_item_ptr(leaf, path->slots[0],
9726 struct btrfs_file_extent_item);
9727 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9728 btrfs_set_file_extent_type(leaf, ei,
9729 BTRFS_FILE_EXTENT_INLINE);
9730 btrfs_set_file_extent_encryption(leaf, ei, 0);
9731 btrfs_set_file_extent_compression(leaf, ei, 0);
9732 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9733 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9734
9735 ptr = btrfs_file_extent_inline_start(ei);
9736 write_extent_buffer(leaf, symname, ptr, name_len);
9737 btrfs_mark_buffer_dirty(leaf);
9738 btrfs_free_path(path);
9739
9740 inode->i_op = &btrfs_symlink_inode_operations;
9741 inode_nohighmem(inode);
9742 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9743 inode_set_bytes(inode, name_len);
9744 btrfs_i_size_write(inode, name_len);
9745 err = btrfs_update_inode(trans, root, inode);
9746 /*
9747 * Last step, add directory indexes for our symlink inode. This is the
9748 * last step to avoid extra cleanup of these indexes if an error happens
9749 * elsewhere above.
9750 */
9751 if (!err)
9752 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9753 if (err) {
9754 drop_inode = 1;
9755 goto out_unlock_inode;
9756 }
9757
9758 unlock_new_inode(inode);
9759 d_instantiate(dentry, inode);
9760
9761 out_unlock:
9762 btrfs_end_transaction(trans, root);
9763 if (drop_inode) {
9764 inode_dec_link_count(inode);
9765 iput(inode);
9766 }
9767 btrfs_btree_balance_dirty(root);
9768 return err;
9769
9770 out_unlock_inode:
9771 drop_inode = 1;
9772 unlock_new_inode(inode);
9773 goto out_unlock;
9774 }
9775
9776 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9777 u64 start, u64 num_bytes, u64 min_size,
9778 loff_t actual_len, u64 *alloc_hint,
9779 struct btrfs_trans_handle *trans)
9780 {
9781 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9782 struct extent_map *em;
9783 struct btrfs_root *root = BTRFS_I(inode)->root;
9784 struct btrfs_key ins;
9785 u64 cur_offset = start;
9786 u64 i_size;
9787 u64 cur_bytes;
9788 u64 last_alloc = (u64)-1;
9789 int ret = 0;
9790 bool own_trans = true;
9791
9792 if (trans)
9793 own_trans = false;
9794 while (num_bytes > 0) {
9795 if (own_trans) {
9796 trans = btrfs_start_transaction(root, 3);
9797 if (IS_ERR(trans)) {
9798 ret = PTR_ERR(trans);
9799 break;
9800 }
9801 }
9802
9803 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9804 cur_bytes = max(cur_bytes, min_size);
9805 /*
9806 * If we are severely fragmented we could end up with really
9807 * small allocations, so if the allocator is returning small
9808 * chunks lets make its job easier by only searching for those
9809 * sized chunks.
9810 */
9811 cur_bytes = min(cur_bytes, last_alloc);
9812 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9813 *alloc_hint, &ins, 1, 0);
9814 if (ret) {
9815 if (own_trans)
9816 btrfs_end_transaction(trans, root);
9817 break;
9818 }
9819
9820 last_alloc = ins.offset;
9821 ret = insert_reserved_file_extent(trans, inode,
9822 cur_offset, ins.objectid,
9823 ins.offset, ins.offset,
9824 ins.offset, 0, 0, 0,
9825 BTRFS_FILE_EXTENT_PREALLOC);
9826 if (ret) {
9827 btrfs_free_reserved_extent(root, ins.objectid,
9828 ins.offset, 0);
9829 btrfs_abort_transaction(trans, root, ret);
9830 if (own_trans)
9831 btrfs_end_transaction(trans, root);
9832 break;
9833 }
9834
9835 btrfs_drop_extent_cache(inode, cur_offset,
9836 cur_offset + ins.offset -1, 0);
9837
9838 em = alloc_extent_map();
9839 if (!em) {
9840 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9841 &BTRFS_I(inode)->runtime_flags);
9842 goto next;
9843 }
9844
9845 em->start = cur_offset;
9846 em->orig_start = cur_offset;
9847 em->len = ins.offset;
9848 em->block_start = ins.objectid;
9849 em->block_len = ins.offset;
9850 em->orig_block_len = ins.offset;
9851 em->ram_bytes = ins.offset;
9852 em->bdev = root->fs_info->fs_devices->latest_bdev;
9853 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9854 em->generation = trans->transid;
9855
9856 while (1) {
9857 write_lock(&em_tree->lock);
9858 ret = add_extent_mapping(em_tree, em, 1);
9859 write_unlock(&em_tree->lock);
9860 if (ret != -EEXIST)
9861 break;
9862 btrfs_drop_extent_cache(inode, cur_offset,
9863 cur_offset + ins.offset - 1,
9864 0);
9865 }
9866 free_extent_map(em);
9867 next:
9868 num_bytes -= ins.offset;
9869 cur_offset += ins.offset;
9870 *alloc_hint = ins.objectid + ins.offset;
9871
9872 inode_inc_iversion(inode);
9873 inode->i_ctime = CURRENT_TIME;
9874 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9875 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9876 (actual_len > inode->i_size) &&
9877 (cur_offset > inode->i_size)) {
9878 if (cur_offset > actual_len)
9879 i_size = actual_len;
9880 else
9881 i_size = cur_offset;
9882 i_size_write(inode, i_size);
9883 btrfs_ordered_update_i_size(inode, i_size, NULL);
9884 }
9885
9886 ret = btrfs_update_inode(trans, root, inode);
9887
9888 if (ret) {
9889 btrfs_abort_transaction(trans, root, ret);
9890 if (own_trans)
9891 btrfs_end_transaction(trans, root);
9892 break;
9893 }
9894
9895 if (own_trans)
9896 btrfs_end_transaction(trans, root);
9897 }
9898 return ret;
9899 }
9900
9901 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9902 u64 start, u64 num_bytes, u64 min_size,
9903 loff_t actual_len, u64 *alloc_hint)
9904 {
9905 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9906 min_size, actual_len, alloc_hint,
9907 NULL);
9908 }
9909
9910 int btrfs_prealloc_file_range_trans(struct inode *inode,
9911 struct btrfs_trans_handle *trans, int mode,
9912 u64 start, u64 num_bytes, u64 min_size,
9913 loff_t actual_len, u64 *alloc_hint)
9914 {
9915 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9916 min_size, actual_len, alloc_hint, trans);
9917 }
9918
9919 static int btrfs_set_page_dirty(struct page *page)
9920 {
9921 return __set_page_dirty_nobuffers(page);
9922 }
9923
9924 static int btrfs_permission(struct inode *inode, int mask)
9925 {
9926 struct btrfs_root *root = BTRFS_I(inode)->root;
9927 umode_t mode = inode->i_mode;
9928
9929 if (mask & MAY_WRITE &&
9930 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9931 if (btrfs_root_readonly(root))
9932 return -EROFS;
9933 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9934 return -EACCES;
9935 }
9936 return generic_permission(inode, mask);
9937 }
9938
9939 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9940 {
9941 struct btrfs_trans_handle *trans;
9942 struct btrfs_root *root = BTRFS_I(dir)->root;
9943 struct inode *inode = NULL;
9944 u64 objectid;
9945 u64 index;
9946 int ret = 0;
9947
9948 /*
9949 * 5 units required for adding orphan entry
9950 */
9951 trans = btrfs_start_transaction(root, 5);
9952 if (IS_ERR(trans))
9953 return PTR_ERR(trans);
9954
9955 ret = btrfs_find_free_ino(root, &objectid);
9956 if (ret)
9957 goto out;
9958
9959 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9960 btrfs_ino(dir), objectid, mode, &index);
9961 if (IS_ERR(inode)) {
9962 ret = PTR_ERR(inode);
9963 inode = NULL;
9964 goto out;
9965 }
9966
9967 inode->i_fop = &btrfs_file_operations;
9968 inode->i_op = &btrfs_file_inode_operations;
9969
9970 inode->i_mapping->a_ops = &btrfs_aops;
9971 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9972
9973 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9974 if (ret)
9975 goto out_inode;
9976
9977 ret = btrfs_update_inode(trans, root, inode);
9978 if (ret)
9979 goto out_inode;
9980 ret = btrfs_orphan_add(trans, inode);
9981 if (ret)
9982 goto out_inode;
9983
9984 /*
9985 * We set number of links to 0 in btrfs_new_inode(), and here we set
9986 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9987 * through:
9988 *
9989 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9990 */
9991 set_nlink(inode, 1);
9992 unlock_new_inode(inode);
9993 d_tmpfile(dentry, inode);
9994 mark_inode_dirty(inode);
9995
9996 out:
9997 btrfs_end_transaction(trans, root);
9998 if (ret)
9999 iput(inode);
10000 btrfs_balance_delayed_items(root);
10001 btrfs_btree_balance_dirty(root);
10002 return ret;
10003
10004 out_inode:
10005 unlock_new_inode(inode);
10006 goto out;
10007
10008 }
10009
10010 /* Inspired by filemap_check_errors() */
10011 int btrfs_inode_check_errors(struct inode *inode)
10012 {
10013 int ret = 0;
10014
10015 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10016 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10017 ret = -ENOSPC;
10018 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10019 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10020 ret = -EIO;
10021
10022 return ret;
10023 }
10024
10025 static const struct inode_operations btrfs_dir_inode_operations = {
10026 .getattr = btrfs_getattr,
10027 .lookup = btrfs_lookup,
10028 .create = btrfs_create,
10029 .unlink = btrfs_unlink,
10030 .link = btrfs_link,
10031 .mkdir = btrfs_mkdir,
10032 .rmdir = btrfs_rmdir,
10033 .rename2 = btrfs_rename2,
10034 .symlink = btrfs_symlink,
10035 .setattr = btrfs_setattr,
10036 .mknod = btrfs_mknod,
10037 .setxattr = btrfs_setxattr,
10038 .getxattr = generic_getxattr,
10039 .listxattr = btrfs_listxattr,
10040 .removexattr = btrfs_removexattr,
10041 .permission = btrfs_permission,
10042 .get_acl = btrfs_get_acl,
10043 .set_acl = btrfs_set_acl,
10044 .update_time = btrfs_update_time,
10045 .tmpfile = btrfs_tmpfile,
10046 };
10047 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10048 .lookup = btrfs_lookup,
10049 .permission = btrfs_permission,
10050 .get_acl = btrfs_get_acl,
10051 .set_acl = btrfs_set_acl,
10052 .update_time = btrfs_update_time,
10053 };
10054
10055 static const struct file_operations btrfs_dir_file_operations = {
10056 .llseek = generic_file_llseek,
10057 .read = generic_read_dir,
10058 .iterate = btrfs_real_readdir,
10059 .unlocked_ioctl = btrfs_ioctl,
10060 #ifdef CONFIG_COMPAT
10061 .compat_ioctl = btrfs_ioctl,
10062 #endif
10063 .release = btrfs_release_file,
10064 .fsync = btrfs_sync_file,
10065 };
10066
10067 static const struct extent_io_ops btrfs_extent_io_ops = {
10068 .fill_delalloc = run_delalloc_range,
10069 .submit_bio_hook = btrfs_submit_bio_hook,
10070 .merge_bio_hook = btrfs_merge_bio_hook,
10071 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10072 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10073 .writepage_start_hook = btrfs_writepage_start_hook,
10074 .set_bit_hook = btrfs_set_bit_hook,
10075 .clear_bit_hook = btrfs_clear_bit_hook,
10076 .merge_extent_hook = btrfs_merge_extent_hook,
10077 .split_extent_hook = btrfs_split_extent_hook,
10078 };
10079
10080 /*
10081 * btrfs doesn't support the bmap operation because swapfiles
10082 * use bmap to make a mapping of extents in the file. They assume
10083 * these extents won't change over the life of the file and they
10084 * use the bmap result to do IO directly to the drive.
10085 *
10086 * the btrfs bmap call would return logical addresses that aren't
10087 * suitable for IO and they also will change frequently as COW
10088 * operations happen. So, swapfile + btrfs == corruption.
10089 *
10090 * For now we're avoiding this by dropping bmap.
10091 */
10092 static const struct address_space_operations btrfs_aops = {
10093 .readpage = btrfs_readpage,
10094 .writepage = btrfs_writepage,
10095 .writepages = btrfs_writepages,
10096 .readpages = btrfs_readpages,
10097 .direct_IO = btrfs_direct_IO,
10098 .invalidatepage = btrfs_invalidatepage,
10099 .releasepage = btrfs_releasepage,
10100 .set_page_dirty = btrfs_set_page_dirty,
10101 .error_remove_page = generic_error_remove_page,
10102 };
10103
10104 static const struct address_space_operations btrfs_symlink_aops = {
10105 .readpage = btrfs_readpage,
10106 .writepage = btrfs_writepage,
10107 .invalidatepage = btrfs_invalidatepage,
10108 .releasepage = btrfs_releasepage,
10109 };
10110
10111 static const struct inode_operations btrfs_file_inode_operations = {
10112 .getattr = btrfs_getattr,
10113 .setattr = btrfs_setattr,
10114 .setxattr = btrfs_setxattr,
10115 .getxattr = generic_getxattr,
10116 .listxattr = btrfs_listxattr,
10117 .removexattr = btrfs_removexattr,
10118 .permission = btrfs_permission,
10119 .fiemap = btrfs_fiemap,
10120 .get_acl = btrfs_get_acl,
10121 .set_acl = btrfs_set_acl,
10122 .update_time = btrfs_update_time,
10123 };
10124 static const struct inode_operations btrfs_special_inode_operations = {
10125 .getattr = btrfs_getattr,
10126 .setattr = btrfs_setattr,
10127 .permission = btrfs_permission,
10128 .setxattr = btrfs_setxattr,
10129 .getxattr = generic_getxattr,
10130 .listxattr = btrfs_listxattr,
10131 .removexattr = btrfs_removexattr,
10132 .get_acl = btrfs_get_acl,
10133 .set_acl = btrfs_set_acl,
10134 .update_time = btrfs_update_time,
10135 };
10136 static const struct inode_operations btrfs_symlink_inode_operations = {
10137 .readlink = generic_readlink,
10138 .get_link = page_get_link,
10139 .getattr = btrfs_getattr,
10140 .setattr = btrfs_setattr,
10141 .permission = btrfs_permission,
10142 .setxattr = btrfs_setxattr,
10143 .getxattr = generic_getxattr,
10144 .listxattr = btrfs_listxattr,
10145 .removexattr = btrfs_removexattr,
10146 .update_time = btrfs_update_time,
10147 };
10148
10149 const struct dentry_operations btrfs_dentry_operations = {
10150 .d_delete = btrfs_dentry_delete,
10151 .d_release = btrfs_dentry_release,
10152 };
Linux kernel - Deliverables
This page took 0.359649 seconds and 5 git commands to generate.