?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ocfs2: fix locking for res->tracking and dlm->tracking_list
[linux/fpc-iii.git] / fs / btrfs / inode.c
blobb895be3d43112f9487d6661c0bf4f4a3433f01c9
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 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_dir_ro_inode_operations;
72 static const struct inode_operations btrfs_special_inode_operations;
73 static const struct inode_operations btrfs_file_inode_operations;
74 static const struct address_space_operations btrfs_aops;
75 static const struct address_space_operations btrfs_symlink_aops;
76 static const struct file_operations btrfs_dir_file_operations;
77 static struct extent_io_ops btrfs_extent_io_ops;
78
79 static struct kmem_cache *btrfs_inode_cachep;
80 static struct kmem_cache *btrfs_delalloc_work_cachep;
81 struct kmem_cache *btrfs_trans_handle_cachep;
82 struct kmem_cache *btrfs_transaction_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
85
86 #define S_SHIFT 12
87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
95 };
96
97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
98 static int btrfs_truncate(struct inode *inode);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
100 static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
107 u64 orig_block_len, u64 ram_bytes,
108 int type);
109
110 static int btrfs_dirty_inode(struct inode *inode);
111
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode *inode)
114 {
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
116 }
117 #endif
118
119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
122 {
123 int err;
124
125 err = btrfs_init_acl(trans, inode, dir);
126 if (!err)
127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
128 return err;
129 }
130
131 /*
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
135 */
136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_path *path, int extent_inserted,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
140 int compress_type,
141 struct page **compressed_pages)
142 {
143 struct extent_buffer *leaf;
144 struct page *page = NULL;
145 char *kaddr;
146 unsigned long ptr;
147 struct btrfs_file_extent_item *ei;
148 int err = 0;
149 int ret;
150 size_t cur_size = size;
151 unsigned long offset;
152
153 if (compressed_size && compressed_pages)
154 cur_size = compressed_size;
155
156 inode_add_bytes(inode, size);
157
158 if (!extent_inserted) {
159 struct btrfs_key key;
160 size_t datasize;
161
162 key.objectid = btrfs_ino(inode);
163 key.offset = start;
164 key.type = BTRFS_EXTENT_DATA_KEY;
165
166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 datasize);
170 if (ret) {
171 err = ret;
172 goto fail;
173 }
174 }
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
184
185 if (compress_type != BTRFS_COMPRESS_NONE) {
186 struct page *cpage;
187 int i = 0;
188 while (compressed_size > 0) {
189 cpage = compressed_pages[i];
190 cur_size = min_t(unsigned long, compressed_size,
191 PAGE_CACHE_SIZE);
192
193 kaddr = kmap_atomic(cpage);
194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
195 kunmap_atomic(kaddr);
196
197 i++;
198 ptr += cur_size;
199 compressed_size -= cur_size;
200 }
201 btrfs_set_file_extent_compression(leaf, ei,
202 compress_type);
203 } else {
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
207 kaddr = kmap_atomic(page);
208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
210 kunmap_atomic(kaddr);
211 page_cache_release(page);
212 }
213 btrfs_mark_buffer_dirty(leaf);
214 btrfs_release_path(path);
215
216 /*
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
220 *
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
224 */
225 BTRFS_I(inode)->disk_i_size = inode->i_size;
226 ret = btrfs_update_inode(trans, root, inode);
227
228 return ret;
229 fail:
230 return err;
231 }
232
233
234 /*
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
238 */
239 static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
242 int compress_type,
243 struct page **compressed_pages)
244 {
245 struct btrfs_trans_handle *trans;
246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
249 u64 aligned_end = ALIGN(end, root->sectorsize);
250 u64 data_len = inline_len;
251 int ret;
252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
255
256 if (compressed_size)
257 data_len = compressed_size;
258
259 if (start > 0 ||
260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
262 (!compressed_size &&
263 (actual_end & (root->sectorsize - 1)) == 0) ||
264 end + 1 < isize ||
265 data_len > root->fs_info->max_inline) {
266 return 1;
267 }
268
269 path = btrfs_alloc_path();
270 if (!path)
271 return -ENOMEM;
272
273 trans = btrfs_join_transaction(root);
274 if (IS_ERR(trans)) {
275 btrfs_free_path(path);
276 return PTR_ERR(trans);
277 }
278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
279
280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
282 compressed_size);
283 else
284 extent_item_size = btrfs_file_extent_calc_inline_size(
285 inline_len);
286
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
290 if (ret) {
291 btrfs_abort_transaction(trans, root, ret);
292 goto out;
293 }
294
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
297 ret = insert_inline_extent(trans, path, extent_inserted,
298 root, inode, start,
299 inline_len, compressed_size,
300 compress_type, compressed_pages);
301 if (ret && ret != -ENOSPC) {
302 btrfs_abort_transaction(trans, root, ret);
303 goto out;
304 } else if (ret == -ENOSPC) {
305 ret = 1;
306 goto out;
307 }
308
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
312 out:
313 /*
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
318 */
319 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
320 btrfs_free_path(path);
321 btrfs_end_transaction(trans, root);
322 return ret;
323 }
324
325 struct async_extent {
326 u64 start;
327 u64 ram_size;
328 u64 compressed_size;
329 struct page **pages;
330 unsigned long nr_pages;
331 int compress_type;
332 struct list_head list;
333 };
334
335 struct async_cow {
336 struct inode *inode;
337 struct btrfs_root *root;
338 struct page *locked_page;
339 u64 start;
340 u64 end;
341 struct list_head extents;
342 struct btrfs_work work;
343 };
344
345 static noinline int add_async_extent(struct async_cow *cow,
346 u64 start, u64 ram_size,
347 u64 compressed_size,
348 struct page **pages,
349 unsigned long nr_pages,
350 int compress_type)
351 {
352 struct async_extent *async_extent;
353
354 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
355 BUG_ON(!async_extent); /* -ENOMEM */
356 async_extent->start = start;
357 async_extent->ram_size = ram_size;
358 async_extent->compressed_size = compressed_size;
359 async_extent->pages = pages;
360 async_extent->nr_pages = nr_pages;
361 async_extent->compress_type = compress_type;
362 list_add_tail(&async_extent->list, &cow->extents);
363 return 0;
364 }
365
366 static inline int inode_need_compress(struct inode *inode)
367 {
368 struct btrfs_root *root = BTRFS_I(inode)->root;
369
370 /* force compress */
371 if (btrfs_test_opt(root, FORCE_COMPRESS))
372 return 1;
373 /* bad compression ratios */
374 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
375 return 0;
376 if (btrfs_test_opt(root, COMPRESS) ||
377 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
378 BTRFS_I(inode)->force_compress)
379 return 1;
380 return 0;
381 }
382
383 /*
384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
387 *
388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
393 *
394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
397 * are written in the same order that the flusher thread sent them
398 * down.
399 */
400 static noinline void compress_file_range(struct inode *inode,
401 struct page *locked_page,
402 u64 start, u64 end,
403 struct async_cow *async_cow,
404 int *num_added)
405 {
406 struct btrfs_root *root = BTRFS_I(inode)->root;
407 u64 num_bytes;
408 u64 blocksize = root->sectorsize;
409 u64 actual_end;
410 u64 isize = i_size_read(inode);
411 int ret = 0;
412 struct page **pages = NULL;
413 unsigned long nr_pages;
414 unsigned long nr_pages_ret = 0;
415 unsigned long total_compressed = 0;
416 unsigned long total_in = 0;
417 unsigned long max_compressed = 128 * 1024;
418 unsigned long max_uncompressed = 128 * 1024;
419 int i;
420 int will_compress;
421 int compress_type = root->fs_info->compress_type;
422 int redirty = 0;
423
424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end - start + 1) < 16 * 1024 &&
426 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
427 btrfs_add_inode_defrag(NULL, inode);
428
429 actual_end = min_t(u64, isize, end + 1);
430 again:
431 will_compress = 0;
432 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
433 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
434
435 /*
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
441 *
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
444 */
445 if (actual_end <= start)
446 goto cleanup_and_bail_uncompressed;
447
448 total_compressed = actual_end - start;
449
450 /*
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
453 */
454 if (total_compressed <= blocksize &&
455 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
456 goto cleanup_and_bail_uncompressed;
457
458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
463 *
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
467 */
468 total_compressed = min(total_compressed, max_uncompressed);
469 num_bytes = ALIGN(end - start + 1, blocksize);
470 num_bytes = max(blocksize, num_bytes);
471 total_in = 0;
472 ret = 0;
473
474 /*
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
478 */
479 if (inode_need_compress(inode)) {
480 WARN_ON(pages);
481 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
482 if (!pages) {
483 /* just bail out to the uncompressed code */
484 goto cont;
485 }
486
487 if (BTRFS_I(inode)->force_compress)
488 compress_type = BTRFS_I(inode)->force_compress;
489
490 /*
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
495 *
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
498 */
499 extent_range_clear_dirty_for_io(inode, start, end);
500 redirty = 1;
501 ret = btrfs_compress_pages(compress_type,
502 inode->i_mapping, start,
503 total_compressed, pages,
504 nr_pages, &nr_pages_ret,
505 &total_in,
506 &total_compressed,
507 max_compressed);
508
509 if (!ret) {
510 unsigned long offset = total_compressed &
511 (PAGE_CACHE_SIZE - 1);
512 struct page *page = pages[nr_pages_ret - 1];
513 char *kaddr;
514
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
517 */
518 if (offset) {
519 kaddr = kmap_atomic(page);
520 memset(kaddr + offset, 0,
521 PAGE_CACHE_SIZE - offset);
522 kunmap_atomic(kaddr);
523 }
524 will_compress = 1;
525 }
526 }
527 cont:
528 if (start == 0) {
529 /* lets try to make an inline extent */
530 if (ret || total_in < (actual_end - start)) {
531 /* we didn't compress the entire range, try
532 * to make an uncompressed inline extent.
533 */
534 ret = cow_file_range_inline(root, inode, start, end,
535 0, 0, NULL);
536 } else {
537 /* try making a compressed inline extent */
538 ret = cow_file_range_inline(root, inode, start, end,
539 total_compressed,
540 compress_type, pages);
541 }
542 if (ret <= 0) {
543 unsigned long clear_flags = EXTENT_DELALLOC |
544 EXTENT_DEFRAG;
545 unsigned long page_error_op;
546
547 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
548 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
549
550 /*
551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
554 */
555 extent_clear_unlock_delalloc(inode, start, end, NULL,
556 clear_flags, PAGE_UNLOCK |
557 PAGE_CLEAR_DIRTY |
558 PAGE_SET_WRITEBACK |
559 page_error_op |
560 PAGE_END_WRITEBACK);
561 goto free_pages_out;
562 }
563 }
564
565 if (will_compress) {
566 /*
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
569 * things
570 */
571 total_compressed = ALIGN(total_compressed, blocksize);
572
573 /*
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
576 */
577 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
578 if (total_compressed >= total_in) {
579 will_compress = 0;
580 } else {
581 num_bytes = total_in;
582 }
583 }
584 if (!will_compress && pages) {
585 /*
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
588 */
589 for (i = 0; i < nr_pages_ret; i++) {
590 WARN_ON(pages[i]->mapping);
591 page_cache_release(pages[i]);
592 }
593 kfree(pages);
594 pages = NULL;
595 total_compressed = 0;
596 nr_pages_ret = 0;
597
598 /* flag the file so we don't compress in the future */
599 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
600 !(BTRFS_I(inode)->force_compress)) {
601 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
602 }
603 }
604 if (will_compress) {
605 *num_added += 1;
606
607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
610 */
611 add_async_extent(async_cow, start, num_bytes,
612 total_compressed, pages, nr_pages_ret,
613 compress_type);
614
615 if (start + num_bytes < end) {
616 start += num_bytes;
617 pages = NULL;
618 cond_resched();
619 goto again;
620 }
621 } else {
622 cleanup_and_bail_uncompressed:
623 /*
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
629 */
630 if (page_offset(locked_page) >= start &&
631 page_offset(locked_page) <= end) {
632 __set_page_dirty_nobuffers(locked_page);
633 /* unlocked later on in the async handlers */
634 }
635 if (redirty)
636 extent_range_redirty_for_io(inode, start, end);
637 add_async_extent(async_cow, start, end - start + 1,
638 0, NULL, 0, BTRFS_COMPRESS_NONE);
639 *num_added += 1;
640 }
641
642 return;
643
644 free_pages_out:
645 for (i = 0; i < nr_pages_ret; i++) {
646 WARN_ON(pages[i]->mapping);
647 page_cache_release(pages[i]);
648 }
649 kfree(pages);
650 }
651
652 static void free_async_extent_pages(struct async_extent *async_extent)
653 {
654 int i;
655
656 if (!async_extent->pages)
657 return;
658
659 for (i = 0; i < async_extent->nr_pages; i++) {
660 WARN_ON(async_extent->pages[i]->mapping);
661 page_cache_release(async_extent->pages[i]);
662 }
663 kfree(async_extent->pages);
664 async_extent->nr_pages = 0;
665 async_extent->pages = NULL;
666 }
667
668 /*
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
673 */
674 static noinline void submit_compressed_extents(struct inode *inode,
675 struct async_cow *async_cow)
676 {
677 struct async_extent *async_extent;
678 u64 alloc_hint = 0;
679 struct btrfs_key ins;
680 struct extent_map *em;
681 struct btrfs_root *root = BTRFS_I(inode)->root;
682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683 struct extent_io_tree *io_tree;
684 int ret = 0;
685
686 again:
687 while (!list_empty(&async_cow->extents)) {
688 async_extent = list_entry(async_cow->extents.next,
689 struct async_extent, list);
690 list_del(&async_extent->list);
691
692 io_tree = &BTRFS_I(inode)->io_tree;
693
694 retry:
695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent->pages) {
697 int page_started = 0;
698 unsigned long nr_written = 0;
699
700 lock_extent(io_tree, async_extent->start,
701 async_extent->start +
702 async_extent->ram_size - 1);
703
704 /* allocate blocks */
705 ret = cow_file_range(inode, async_cow->locked_page,
706 async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1,
709 &page_started, &nr_written, 0);
710
711 /* JDM XXX */
712
713 /*
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
718 */
719 if (!page_started && !ret)
720 extent_write_locked_range(io_tree,
721 inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1,
724 btrfs_get_extent,
725 WB_SYNC_ALL);
726 else if (ret)
727 unlock_page(async_cow->locked_page);
728 kfree(async_extent);
729 cond_resched();
730 continue;
731 }
732
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start + async_extent->ram_size - 1);
735
736 ret = btrfs_reserve_extent(root,
737 async_extent->compressed_size,
738 async_extent->compressed_size,
739 0, alloc_hint, &ins, 1, 1);
740 if (ret) {
741 free_async_extent_pages(async_extent);
742
743 if (ret == -ENOSPC) {
744 unlock_extent(io_tree, async_extent->start,
745 async_extent->start +
746 async_extent->ram_size - 1);
747
748 /*
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
752 * layers.
753 */
754 extent_range_redirty_for_io(inode,
755 async_extent->start,
756 async_extent->start +
757 async_extent->ram_size - 1);
758
759 goto retry;
760 }
761 goto out_free;
762 }
763 /*
764 * here we're doing allocation and writeback of the
765 * compressed pages
766 */
767 btrfs_drop_extent_cache(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1, 0);
770
771 em = alloc_extent_map();
772 if (!em) {
773 ret = -ENOMEM;
774 goto out_free_reserve;
775 }
776 em->start = async_extent->start;
777 em->len = async_extent->ram_size;
778 em->orig_start = em->start;
779 em->mod_start = em->start;
780 em->mod_len = em->len;
781
782 em->block_start = ins.objectid;
783 em->block_len = ins.offset;
784 em->orig_block_len = ins.offset;
785 em->ram_bytes = async_extent->ram_size;
786 em->bdev = root->fs_info->fs_devices->latest_bdev;
787 em->compress_type = async_extent->compress_type;
788 set_bit(EXTENT_FLAG_PINNED, &em->flags);
789 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
790 em->generation = -1;
791
792 while (1) {
793 write_lock(&em_tree->lock);
794 ret = add_extent_mapping(em_tree, em, 1);
795 write_unlock(&em_tree->lock);
796 if (ret != -EEXIST) {
797 free_extent_map(em);
798 break;
799 }
800 btrfs_drop_extent_cache(inode, async_extent->start,
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
803 }
804
805 if (ret)
806 goto out_free_reserve;
807
808 ret = btrfs_add_ordered_extent_compress(inode,
809 async_extent->start,
810 ins.objectid,
811 async_extent->ram_size,
812 ins.offset,
813 BTRFS_ORDERED_COMPRESSED,
814 async_extent->compress_type);
815 if (ret) {
816 btrfs_drop_extent_cache(inode, async_extent->start,
817 async_extent->start +
818 async_extent->ram_size - 1, 0);
819 goto out_free_reserve;
820 }
821
822 /*
823 * clear dirty, set writeback and unlock the pages.
824 */
825 extent_clear_unlock_delalloc(inode, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1,
828 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
829 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
830 PAGE_SET_WRITEBACK);
831 ret = btrfs_submit_compressed_write(inode,
832 async_extent->start,
833 async_extent->ram_size,
834 ins.objectid,
835 ins.offset, async_extent->pages,
836 async_extent->nr_pages);
837 if (ret) {
838 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
839 struct page *p = async_extent->pages[0];
840 const u64 start = async_extent->start;
841 const u64 end = start + async_extent->ram_size - 1;
842
843 p->mapping = inode->i_mapping;
844 tree->ops->writepage_end_io_hook(p, start, end,
845 NULL, 0);
846 p->mapping = NULL;
847 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
848 PAGE_END_WRITEBACK |
849 PAGE_SET_ERROR);
850 free_async_extent_pages(async_extent);
851 }
852 alloc_hint = ins.objectid + ins.offset;
853 kfree(async_extent);
854 cond_resched();
855 }
856 return;
857 out_free_reserve:
858 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
859 out_free:
860 extent_clear_unlock_delalloc(inode, async_extent->start,
861 async_extent->start +
862 async_extent->ram_size - 1,
863 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
864 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
865 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
866 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
867 PAGE_SET_ERROR);
868 free_async_extent_pages(async_extent);
869 kfree(async_extent);
870 goto again;
871 }
872
873 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
874 u64 num_bytes)
875 {
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
877 struct extent_map *em;
878 u64 alloc_hint = 0;
879
880 read_lock(&em_tree->lock);
881 em = search_extent_mapping(em_tree, start, num_bytes);
882 if (em) {
883 /*
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
887 */
888 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
889 free_extent_map(em);
890 em = search_extent_mapping(em_tree, 0, 0);
891 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
892 alloc_hint = em->block_start;
893 if (em)
894 free_extent_map(em);
895 } else {
896 alloc_hint = em->block_start;
897 free_extent_map(em);
898 }
899 }
900 read_unlock(&em_tree->lock);
901
902 return alloc_hint;
903 }
904
905 /*
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
910 *
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
913 *
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
916 * IO when we return.
917 */
918 static noinline int cow_file_range(struct inode *inode,
919 struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written,
922 int unlock)
923 {
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 u64 alloc_hint = 0;
926 u64 num_bytes;
927 unsigned long ram_size;
928 u64 disk_num_bytes;
929 u64 cur_alloc_size;
930 u64 blocksize = root->sectorsize;
931 struct btrfs_key ins;
932 struct extent_map *em;
933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
934 int ret = 0;
935
936 if (btrfs_is_free_space_inode(inode)) {
937 WARN_ON_ONCE(1);
938 ret = -EINVAL;
939 goto out_unlock;
940 }
941
942 num_bytes = ALIGN(end - start + 1, blocksize);
943 num_bytes = max(blocksize, num_bytes);
944 disk_num_bytes = num_bytes;
945
946 /* if this is a small write inside eof, kick off defrag */
947 if (num_bytes < 64 * 1024 &&
948 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
949 btrfs_add_inode_defrag(NULL, inode);
950
951 if (start == 0) {
952 /* lets try to make an inline extent */
953 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
954 NULL);
955 if (ret == 0) {
956 extent_clear_unlock_delalloc(inode, start, end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
958 EXTENT_DEFRAG, PAGE_UNLOCK |
959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
960 PAGE_END_WRITEBACK);
961
962 *nr_written = *nr_written +
963 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
964 *page_started = 1;
965 goto out;
966 } else if (ret < 0) {
967 goto out_unlock;
968 }
969 }
970
971 BUG_ON(disk_num_bytes >
972 btrfs_super_total_bytes(root->fs_info->super_copy));
973
974 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
975 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
976
977 while (disk_num_bytes > 0) {
978 unsigned long op;
979
980 cur_alloc_size = disk_num_bytes;
981 ret = btrfs_reserve_extent(root, cur_alloc_size,
982 root->sectorsize, 0, alloc_hint,
983 &ins, 1, 1);
984 if (ret < 0)
985 goto out_unlock;
986
987 em = alloc_extent_map();
988 if (!em) {
989 ret = -ENOMEM;
990 goto out_reserve;
991 }
992 em->start = start;
993 em->orig_start = em->start;
994 ram_size = ins.offset;
995 em->len = ins.offset;
996 em->mod_start = em->start;
997 em->mod_len = em->len;
998
999 em->block_start = ins.objectid;
1000 em->block_len = ins.offset;
1001 em->orig_block_len = ins.offset;
1002 em->ram_bytes = ram_size;
1003 em->bdev = root->fs_info->fs_devices->latest_bdev;
1004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1005 em->generation = -1;
1006
1007 while (1) {
1008 write_lock(&em_tree->lock);
1009 ret = add_extent_mapping(em_tree, em, 1);
1010 write_unlock(&em_tree->lock);
1011 if (ret != -EEXIST) {
1012 free_extent_map(em);
1013 break;
1014 }
1015 btrfs_drop_extent_cache(inode, start,
1016 start + ram_size - 1, 0);
1017 }
1018 if (ret)
1019 goto out_reserve;
1020
1021 cur_alloc_size = ins.offset;
1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1023 ram_size, cur_alloc_size, 0);
1024 if (ret)
1025 goto out_drop_extent_cache;
1026
1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1030 cur_alloc_size);
1031 if (ret)
1032 goto out_drop_extent_cache;
1033 }
1034
1035 if (disk_num_bytes < cur_alloc_size)
1036 break;
1037
1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
1041 *
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
1044 */
1045 op = unlock ? PAGE_UNLOCK : 0;
1046 op |= PAGE_SET_PRIVATE2;
1047
1048 extent_clear_unlock_delalloc(inode, start,
1049 start + ram_size - 1, locked_page,
1050 EXTENT_LOCKED | EXTENT_DELALLOC,
1051 op);
1052 disk_num_bytes -= cur_alloc_size;
1053 num_bytes -= cur_alloc_size;
1054 alloc_hint = ins.objectid + ins.offset;
1055 start += cur_alloc_size;
1056 }
1057 out:
1058 return ret;
1059
1060 out_drop_extent_cache:
1061 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1062 out_reserve:
1063 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1064 out_unlock:
1065 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1066 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1067 EXTENT_DELALLOC | EXTENT_DEFRAG,
1068 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1069 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1070 goto out;
1071 }
1072
1073 /*
1074 * work queue call back to started compression on a file and pages
1075 */
1076 static noinline void async_cow_start(struct btrfs_work *work)
1077 {
1078 struct async_cow *async_cow;
1079 int num_added = 0;
1080 async_cow = container_of(work, struct async_cow, work);
1081
1082 compress_file_range(async_cow->inode, async_cow->locked_page,
1083 async_cow->start, async_cow->end, async_cow,
1084 &num_added);
1085 if (num_added == 0) {
1086 btrfs_add_delayed_iput(async_cow->inode);
1087 async_cow->inode = NULL;
1088 }
1089 }
1090
1091 /*
1092 * work queue call back to submit previously compressed pages
1093 */
1094 static noinline void async_cow_submit(struct btrfs_work *work)
1095 {
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root;
1098 unsigned long nr_pages;
1099
1100 async_cow = container_of(work, struct async_cow, work);
1101
1102 root = async_cow->root;
1103 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1104 PAGE_CACHE_SHIFT;
1105
1106 /*
1107 * atomic_sub_return implies a barrier for waitqueue_active
1108 */
1109 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1110 5 * 1024 * 1024 &&
1111 waitqueue_active(&root->fs_info->async_submit_wait))
1112 wake_up(&root->fs_info->async_submit_wait);
1113
1114 if (async_cow->inode)
1115 submit_compressed_extents(async_cow->inode, async_cow);
1116 }
1117
1118 static noinline void async_cow_free(struct btrfs_work *work)
1119 {
1120 struct async_cow *async_cow;
1121 async_cow = container_of(work, struct async_cow, work);
1122 if (async_cow->inode)
1123 btrfs_add_delayed_iput(async_cow->inode);
1124 kfree(async_cow);
1125 }
1126
1127 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1128 u64 start, u64 end, int *page_started,
1129 unsigned long *nr_written)
1130 {
1131 struct async_cow *async_cow;
1132 struct btrfs_root *root = BTRFS_I(inode)->root;
1133 unsigned long nr_pages;
1134 u64 cur_end;
1135 int limit = 10 * 1024 * 1024;
1136
1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1138 1, 0, NULL, GFP_NOFS);
1139 while (start < end) {
1140 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1141 BUG_ON(!async_cow); /* -ENOMEM */
1142 async_cow->inode = igrab(inode);
1143 async_cow->root = root;
1144 async_cow->locked_page = locked_page;
1145 async_cow->start = start;
1146
1147 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1148 !btrfs_test_opt(root, FORCE_COMPRESS))
1149 cur_end = end;
1150 else
1151 cur_end = min(end, start + 512 * 1024 - 1);
1152
1153 async_cow->end = cur_end;
1154 INIT_LIST_HEAD(&async_cow->extents);
1155
1156 btrfs_init_work(&async_cow->work,
1157 btrfs_delalloc_helper,
1158 async_cow_start, async_cow_submit,
1159 async_cow_free);
1160
1161 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1162 PAGE_CACHE_SHIFT;
1163 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1164
1165 btrfs_queue_work(root->fs_info->delalloc_workers,
1166 &async_cow->work);
1167
1168 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1169 wait_event(root->fs_info->async_submit_wait,
1170 (atomic_read(&root->fs_info->async_delalloc_pages) <
1171 limit));
1172 }
1173
1174 while (atomic_read(&root->fs_info->async_submit_draining) &&
1175 atomic_read(&root->fs_info->async_delalloc_pages)) {
1176 wait_event(root->fs_info->async_submit_wait,
1177 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1178 0));
1179 }
1180
1181 *nr_written += nr_pages;
1182 start = cur_end + 1;
1183 }
1184 *page_started = 1;
1185 return 0;
1186 }
1187
1188 static noinline int csum_exist_in_range(struct btrfs_root *root,
1189 u64 bytenr, u64 num_bytes)
1190 {
1191 int ret;
1192 struct btrfs_ordered_sum *sums;
1193 LIST_HEAD(list);
1194
1195 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1196 bytenr + num_bytes - 1, &list, 0);
1197 if (ret == 0 && list_empty(&list))
1198 return 0;
1199
1200 while (!list_empty(&list)) {
1201 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1202 list_del(&sums->list);
1203 kfree(sums);
1204 }
1205 if (ret < 0)
1206 return ret;
1207 return 1;
1208 }
1209
1210 /*
1211 * when nowcow writeback call back. This checks for snapshots or COW copies
1212 * of the extents that exist in the file, and COWs the file as required.
1213 *
1214 * If no cow copies or snapshots exist, we write directly to the existing
1215 * blocks on disk
1216 */
1217 static noinline int run_delalloc_nocow(struct inode *inode,
1218 struct page *locked_page,
1219 u64 start, u64 end, int *page_started, int force,
1220 unsigned long *nr_written)
1221 {
1222 struct btrfs_root *root = BTRFS_I(inode)->root;
1223 struct btrfs_trans_handle *trans;
1224 struct extent_buffer *leaf;
1225 struct btrfs_path *path;
1226 struct btrfs_file_extent_item *fi;
1227 struct btrfs_key found_key;
1228 u64 cow_start;
1229 u64 cur_offset;
1230 u64 extent_end;
1231 u64 extent_offset;
1232 u64 disk_bytenr;
1233 u64 num_bytes;
1234 u64 disk_num_bytes;
1235 u64 ram_bytes;
1236 int extent_type;
1237 int ret, err;
1238 int type;
1239 int nocow;
1240 int check_prev = 1;
1241 bool nolock;
1242 u64 ino = btrfs_ino(inode);
1243
1244 path = btrfs_alloc_path();
1245 if (!path) {
1246 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1247 EXTENT_LOCKED | EXTENT_DELALLOC |
1248 EXTENT_DO_ACCOUNTING |
1249 EXTENT_DEFRAG, PAGE_UNLOCK |
1250 PAGE_CLEAR_DIRTY |
1251 PAGE_SET_WRITEBACK |
1252 PAGE_END_WRITEBACK);
1253 return -ENOMEM;
1254 }
1255
1256 nolock = btrfs_is_free_space_inode(inode);
1257
1258 if (nolock)
1259 trans = btrfs_join_transaction_nolock(root);
1260 else
1261 trans = btrfs_join_transaction(root);
1262
1263 if (IS_ERR(trans)) {
1264 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1265 EXTENT_LOCKED | EXTENT_DELALLOC |
1266 EXTENT_DO_ACCOUNTING |
1267 EXTENT_DEFRAG, PAGE_UNLOCK |
1268 PAGE_CLEAR_DIRTY |
1269 PAGE_SET_WRITEBACK |
1270 PAGE_END_WRITEBACK);
1271 btrfs_free_path(path);
1272 return PTR_ERR(trans);
1273 }
1274
1275 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1276
1277 cow_start = (u64)-1;
1278 cur_offset = start;
1279 while (1) {
1280 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1281 cur_offset, 0);
1282 if (ret < 0)
1283 goto error;
1284 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1285 leaf = path->nodes[0];
1286 btrfs_item_key_to_cpu(leaf, &found_key,
1287 path->slots[0] - 1);
1288 if (found_key.objectid == ino &&
1289 found_key.type == BTRFS_EXTENT_DATA_KEY)
1290 path->slots[0]--;
1291 }
1292 check_prev = 0;
1293 next_slot:
1294 leaf = path->nodes[0];
1295 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1296 ret = btrfs_next_leaf(root, path);
1297 if (ret < 0) {
1298 if (cow_start != (u64)-1)
1299 cur_offset = cow_start;
1300 goto error;
1301 }
1302 if (ret > 0)
1303 break;
1304 leaf = path->nodes[0];
1305 }
1306
1307 nocow = 0;
1308 disk_bytenr = 0;
1309 num_bytes = 0;
1310 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1311
1312 if (found_key.objectid > ino)
1313 break;
1314 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1315 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1316 path->slots[0]++;
1317 goto next_slot;
1318 }
1319 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1320 found_key.offset > end)
1321 break;
1322
1323 if (found_key.offset > cur_offset) {
1324 extent_end = found_key.offset;
1325 extent_type = 0;
1326 goto out_check;
1327 }
1328
1329 fi = btrfs_item_ptr(leaf, path->slots[0],
1330 struct btrfs_file_extent_item);
1331 extent_type = btrfs_file_extent_type(leaf, fi);
1332
1333 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1334 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1335 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1336 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1337 extent_offset = btrfs_file_extent_offset(leaf, fi);
1338 extent_end = found_key.offset +
1339 btrfs_file_extent_num_bytes(leaf, fi);
1340 disk_num_bytes =
1341 btrfs_file_extent_disk_num_bytes(leaf, fi);
1342 if (extent_end <= start) {
1343 path->slots[0]++;
1344 goto next_slot;
1345 }
1346 if (disk_bytenr == 0)
1347 goto out_check;
1348 if (btrfs_file_extent_compression(leaf, fi) ||
1349 btrfs_file_extent_encryption(leaf, fi) ||
1350 btrfs_file_extent_other_encoding(leaf, fi))
1351 goto out_check;
1352 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1353 goto out_check;
1354 if (btrfs_extent_readonly(root, disk_bytenr))
1355 goto out_check;
1356 ret = btrfs_cross_ref_exist(trans, root, ino,
1357 found_key.offset -
1358 extent_offset, disk_bytenr);
1359 if (ret) {
1360 /*
1361 * ret could be -EIO if the above fails to read
1362 * metadata.
1363 */
1364 if (ret < 0) {
1365 if (cow_start != (u64)-1)
1366 cur_offset = cow_start;
1367 goto error;
1368 }
1369
1370 WARN_ON_ONCE(nolock);
1371 goto out_check;
1372 }
1373 disk_bytenr += extent_offset;
1374 disk_bytenr += cur_offset - found_key.offset;
1375 num_bytes = min(end + 1, extent_end) - cur_offset;
1376 /*
1377 * if there are pending snapshots for this root,
1378 * we fall into common COW way.
1379 */
1380 if (!nolock) {
1381 err = btrfs_start_write_no_snapshoting(root);
1382 if (!err)
1383 goto out_check;
1384 }
1385 /*
1386 * force cow if csum exists in the range.
1387 * this ensure that csum for a given extent are
1388 * either valid or do not exist.
1389 */
1390 ret = csum_exist_in_range(root, disk_bytenr, num_bytes);
1391 if (ret) {
1392 /*
1393 * ret could be -EIO if the above fails to read
1394 * metadata.
1395 */
1396 if (ret < 0) {
1397 if (cow_start != (u64)-1)
1398 cur_offset = cow_start;
1399 goto error;
1400 }
1401 WARN_ON_ONCE(nolock);
1402 goto out_check;
1403 }
1404 nocow = 1;
1405 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1406 extent_end = found_key.offset +
1407 btrfs_file_extent_inline_len(leaf,
1408 path->slots[0], fi);
1409 extent_end = ALIGN(extent_end, root->sectorsize);
1410 } else {
1411 BUG_ON(1);
1412 }
1413 out_check:
1414 if (extent_end <= start) {
1415 path->slots[0]++;
1416 if (!nolock && nocow)
1417 btrfs_end_write_no_snapshoting(root);
1418 goto next_slot;
1419 }
1420 if (!nocow) {
1421 if (cow_start == (u64)-1)
1422 cow_start = cur_offset;
1423 cur_offset = extent_end;
1424 if (cur_offset > end)
1425 break;
1426 path->slots[0]++;
1427 goto next_slot;
1428 }
1429
1430 btrfs_release_path(path);
1431 if (cow_start != (u64)-1) {
1432 ret = cow_file_range(inode, locked_page,
1433 cow_start, found_key.offset - 1,
1434 page_started, nr_written, 1);
1435 if (ret) {
1436 if (!nolock && nocow)
1437 btrfs_end_write_no_snapshoting(root);
1438 goto error;
1439 }
1440 cow_start = (u64)-1;
1441 }
1442
1443 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1444 struct extent_map *em;
1445 struct extent_map_tree *em_tree;
1446 em_tree = &BTRFS_I(inode)->extent_tree;
1447 em = alloc_extent_map();
1448 BUG_ON(!em); /* -ENOMEM */
1449 em->start = cur_offset;
1450 em->orig_start = found_key.offset - extent_offset;
1451 em->len = num_bytes;
1452 em->block_len = num_bytes;
1453 em->block_start = disk_bytenr;
1454 em->orig_block_len = disk_num_bytes;
1455 em->ram_bytes = ram_bytes;
1456 em->bdev = root->fs_info->fs_devices->latest_bdev;
1457 em->mod_start = em->start;
1458 em->mod_len = em->len;
1459 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1460 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1461 em->generation = -1;
1462 while (1) {
1463 write_lock(&em_tree->lock);
1464 ret = add_extent_mapping(em_tree, em, 1);
1465 write_unlock(&em_tree->lock);
1466 if (ret != -EEXIST) {
1467 free_extent_map(em);
1468 break;
1469 }
1470 btrfs_drop_extent_cache(inode, em->start,
1471 em->start + em->len - 1, 0);
1472 }
1473 type = BTRFS_ORDERED_PREALLOC;
1474 } else {
1475 type = BTRFS_ORDERED_NOCOW;
1476 }
1477
1478 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1479 num_bytes, num_bytes, type);
1480 BUG_ON(ret); /* -ENOMEM */
1481
1482 if (root->root_key.objectid ==
1483 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1484 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1485 num_bytes);
1486 if (ret) {
1487 if (!nolock && nocow)
1488 btrfs_end_write_no_snapshoting(root);
1489 goto error;
1490 }
1491 }
1492
1493 extent_clear_unlock_delalloc(inode, cur_offset,
1494 cur_offset + num_bytes - 1,
1495 locked_page, EXTENT_LOCKED |
1496 EXTENT_DELALLOC, PAGE_UNLOCK |
1497 PAGE_SET_PRIVATE2);
1498 if (!nolock && nocow)
1499 btrfs_end_write_no_snapshoting(root);
1500 cur_offset = extent_end;
1501 if (cur_offset > end)
1502 break;
1503 }
1504 btrfs_release_path(path);
1505
1506 if (cur_offset <= end && cow_start == (u64)-1) {
1507 cow_start = cur_offset;
1508 cur_offset = end;
1509 }
1510
1511 if (cow_start != (u64)-1) {
1512 ret = cow_file_range(inode, locked_page, cow_start, end,
1513 page_started, nr_written, 1);
1514 if (ret)
1515 goto error;
1516 }
1517
1518 error:
1519 err = btrfs_end_transaction(trans, root);
1520 if (!ret)
1521 ret = err;
1522
1523 if (ret && cur_offset < end)
1524 extent_clear_unlock_delalloc(inode, cur_offset, end,
1525 locked_page, EXTENT_LOCKED |
1526 EXTENT_DELALLOC | EXTENT_DEFRAG |
1527 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1528 PAGE_CLEAR_DIRTY |
1529 PAGE_SET_WRITEBACK |
1530 PAGE_END_WRITEBACK);
1531 btrfs_free_path(path);
1532 return ret;
1533 }
1534
1535 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1536 {
1537
1538 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1539 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1540 return 0;
1541
1542 /*
1543 * @defrag_bytes is a hint value, no spinlock held here,
1544 * if is not zero, it means the file is defragging.
1545 * Force cow if given extent needs to be defragged.
1546 */
1547 if (BTRFS_I(inode)->defrag_bytes &&
1548 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1549 EXTENT_DEFRAG, 0, NULL))
1550 return 1;
1551
1552 return 0;
1553 }
1554
1555 /*
1556 * extent_io.c call back to do delayed allocation processing
1557 */
1558 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1559 u64 start, u64 end, int *page_started,
1560 unsigned long *nr_written)
1561 {
1562 int ret;
1563 int force_cow = need_force_cow(inode, start, end);
1564
1565 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1566 ret = run_delalloc_nocow(inode, locked_page, start, end,
1567 page_started, 1, nr_written);
1568 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1569 ret = run_delalloc_nocow(inode, locked_page, start, end,
1570 page_started, 0, nr_written);
1571 } else if (!inode_need_compress(inode)) {
1572 ret = cow_file_range(inode, locked_page, start, end,
1573 page_started, nr_written, 1);
1574 } else {
1575 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1576 &BTRFS_I(inode)->runtime_flags);
1577 ret = cow_file_range_async(inode, locked_page, start, end,
1578 page_started, nr_written);
1579 }
1580 return ret;
1581 }
1582
1583 static void btrfs_split_extent_hook(struct inode *inode,
1584 struct extent_state *orig, u64 split)
1585 {
1586 u64 size;
1587
1588 /* not delalloc, ignore it */
1589 if (!(orig->state & EXTENT_DELALLOC))
1590 return;
1591
1592 size = orig->end - orig->start + 1;
1593 if (size > BTRFS_MAX_EXTENT_SIZE) {
1594 u64 num_extents;
1595 u64 new_size;
1596
1597 /*
1598 * See the explanation in btrfs_merge_extent_hook, the same
1599 * applies here, just in reverse.
1600 */
1601 new_size = orig->end - split + 1;
1602 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1603 BTRFS_MAX_EXTENT_SIZE);
1604 new_size = split - orig->start;
1605 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1606 BTRFS_MAX_EXTENT_SIZE);
1607 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1608 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1609 return;
1610 }
1611
1612 spin_lock(&BTRFS_I(inode)->lock);
1613 BTRFS_I(inode)->outstanding_extents++;
1614 spin_unlock(&BTRFS_I(inode)->lock);
1615 }
1616
1617 /*
1618 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1619 * extents so we can keep track of new extents that are just merged onto old
1620 * extents, such as when we are doing sequential writes, so we can properly
1621 * account for the metadata space we'll need.
1622 */
1623 static void btrfs_merge_extent_hook(struct inode *inode,
1624 struct extent_state *new,
1625 struct extent_state *other)
1626 {
1627 u64 new_size, old_size;
1628 u64 num_extents;
1629
1630 /* not delalloc, ignore it */
1631 if (!(other->state & EXTENT_DELALLOC))
1632 return;
1633
1634 if (new->start > other->start)
1635 new_size = new->end - other->start + 1;
1636 else
1637 new_size = other->end - new->start + 1;
1638
1639 /* we're not bigger than the max, unreserve the space and go */
1640 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1641 spin_lock(&BTRFS_I(inode)->lock);
1642 BTRFS_I(inode)->outstanding_extents--;
1643 spin_unlock(&BTRFS_I(inode)->lock);
1644 return;
1645 }
1646
1647 /*
1648 * We have to add up either side to figure out how many extents were
1649 * accounted for before we merged into one big extent. If the number of
1650 * extents we accounted for is <= the amount we need for the new range
1651 * then we can return, otherwise drop. Think of it like this
1652 *
1653 * [ 4k][MAX_SIZE]
1654 *
1655 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1656 * need 2 outstanding extents, on one side we have 1 and the other side
1657 * we have 1 so they are == and we can return. But in this case
1658 *
1659 * [MAX_SIZE+4k][MAX_SIZE+4k]
1660 *
1661 * Each range on their own accounts for 2 extents, but merged together
1662 * they are only 3 extents worth of accounting, so we need to drop in
1663 * this case.
1664 */
1665 old_size = other->end - other->start + 1;
1666 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1667 BTRFS_MAX_EXTENT_SIZE);
1668 old_size = new->end - new->start + 1;
1669 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1670 BTRFS_MAX_EXTENT_SIZE);
1671
1672 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1673 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1674 return;
1675
1676 spin_lock(&BTRFS_I(inode)->lock);
1677 BTRFS_I(inode)->outstanding_extents--;
1678 spin_unlock(&BTRFS_I(inode)->lock);
1679 }
1680
1681 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1682 struct inode *inode)
1683 {
1684 spin_lock(&root->delalloc_lock);
1685 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1686 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1687 &root->delalloc_inodes);
1688 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1689 &BTRFS_I(inode)->runtime_flags);
1690 root->nr_delalloc_inodes++;
1691 if (root->nr_delalloc_inodes == 1) {
1692 spin_lock(&root->fs_info->delalloc_root_lock);
1693 BUG_ON(!list_empty(&root->delalloc_root));
1694 list_add_tail(&root->delalloc_root,
1695 &root->fs_info->delalloc_roots);
1696 spin_unlock(&root->fs_info->delalloc_root_lock);
1697 }
1698 }
1699 spin_unlock(&root->delalloc_lock);
1700 }
1701
1702 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1703 struct inode *inode)
1704 {
1705 spin_lock(&root->delalloc_lock);
1706 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1707 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1708 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1709 &BTRFS_I(inode)->runtime_flags);
1710 root->nr_delalloc_inodes--;
1711 if (!root->nr_delalloc_inodes) {
1712 spin_lock(&root->fs_info->delalloc_root_lock);
1713 BUG_ON(list_empty(&root->delalloc_root));
1714 list_del_init(&root->delalloc_root);
1715 spin_unlock(&root->fs_info->delalloc_root_lock);
1716 }
1717 }
1718 spin_unlock(&root->delalloc_lock);
1719 }
1720
1721 /*
1722 * extent_io.c set_bit_hook, used to track delayed allocation
1723 * bytes in this file, and to maintain the list of inodes that
1724 * have pending delalloc work to be done.
1725 */
1726 static void btrfs_set_bit_hook(struct inode *inode,
1727 struct extent_state *state, unsigned *bits)
1728 {
1729
1730 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1731 WARN_ON(1);
1732 /*
1733 * set_bit and clear bit hooks normally require _irqsave/restore
1734 * but in this case, we are only testing for the DELALLOC
1735 * bit, which is only set or cleared with irqs on
1736 */
1737 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1738 struct btrfs_root *root = BTRFS_I(inode)->root;
1739 u64 len = state->end + 1 - state->start;
1740 bool do_list = !btrfs_is_free_space_inode(inode);
1741
1742 if (*bits & EXTENT_FIRST_DELALLOC) {
1743 *bits &= ~EXTENT_FIRST_DELALLOC;
1744 } else {
1745 spin_lock(&BTRFS_I(inode)->lock);
1746 BTRFS_I(inode)->outstanding_extents++;
1747 spin_unlock(&BTRFS_I(inode)->lock);
1748 }
1749
1750 /* For sanity tests */
1751 if (btrfs_test_is_dummy_root(root))
1752 return;
1753
1754 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1755 root->fs_info->delalloc_batch);
1756 spin_lock(&BTRFS_I(inode)->lock);
1757 BTRFS_I(inode)->delalloc_bytes += len;
1758 if (*bits & EXTENT_DEFRAG)
1759 BTRFS_I(inode)->defrag_bytes += len;
1760 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1761 &BTRFS_I(inode)->runtime_flags))
1762 btrfs_add_delalloc_inodes(root, inode);
1763 spin_unlock(&BTRFS_I(inode)->lock);
1764 }
1765 }
1766
1767 /*
1768 * extent_io.c clear_bit_hook, see set_bit_hook for why
1769 */
1770 static void btrfs_clear_bit_hook(struct inode *inode,
1771 struct extent_state *state,
1772 unsigned *bits)
1773 {
1774 u64 len = state->end + 1 - state->start;
1775 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1776 BTRFS_MAX_EXTENT_SIZE);
1777
1778 spin_lock(&BTRFS_I(inode)->lock);
1779 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1780 BTRFS_I(inode)->defrag_bytes -= len;
1781 spin_unlock(&BTRFS_I(inode)->lock);
1782
1783 /*
1784 * set_bit and clear bit hooks normally require _irqsave/restore
1785 * but in this case, we are only testing for the DELALLOC
1786 * bit, which is only set or cleared with irqs on
1787 */
1788 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1789 struct btrfs_root *root = BTRFS_I(inode)->root;
1790 bool do_list = !btrfs_is_free_space_inode(inode);
1791
1792 if (*bits & EXTENT_FIRST_DELALLOC) {
1793 *bits &= ~EXTENT_FIRST_DELALLOC;
1794 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1795 spin_lock(&BTRFS_I(inode)->lock);
1796 BTRFS_I(inode)->outstanding_extents -= num_extents;
1797 spin_unlock(&BTRFS_I(inode)->lock);
1798 }
1799
1800 /*
1801 * We don't reserve metadata space for space cache inodes so we
1802 * don't need to call dellalloc_release_metadata if there is an
1803 * error.
1804 */
1805 if (*bits & EXTENT_DO_ACCOUNTING &&
1806 root != root->fs_info->tree_root)
1807 btrfs_delalloc_release_metadata(inode, len);
1808
1809 /* For sanity tests. */
1810 if (btrfs_test_is_dummy_root(root))
1811 return;
1812
1813 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1814 && do_list && !(state->state & EXTENT_NORESERVE))
1815 btrfs_free_reserved_data_space_noquota(inode,
1816 state->start, len);
1817
1818 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1819 root->fs_info->delalloc_batch);
1820 spin_lock(&BTRFS_I(inode)->lock);
1821 BTRFS_I(inode)->delalloc_bytes -= len;
1822 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1823 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1824 &BTRFS_I(inode)->runtime_flags))
1825 btrfs_del_delalloc_inode(root, inode);
1826 spin_unlock(&BTRFS_I(inode)->lock);
1827 }
1828 }
1829
1830 /*
1831 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1832 * we don't create bios that span stripes or chunks
1833 */
1834 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1835 size_t size, struct bio *bio,
1836 unsigned long bio_flags)
1837 {
1838 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1839 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1840 u64 length = 0;
1841 u64 map_length;
1842 int ret;
1843
1844 if (bio_flags & EXTENT_BIO_COMPRESSED)
1845 return 0;
1846
1847 length = bio->bi_iter.bi_size;
1848 map_length = length;
1849 ret = btrfs_map_block(root->fs_info, rw, logical,
1850 &map_length, NULL, 0);
1851 /* Will always return 0 with map_multi == NULL */
1852 BUG_ON(ret < 0);
1853 if (map_length < length + size)
1854 return 1;
1855 return 0;
1856 }
1857
1858 /*
1859 * in order to insert checksums into the metadata in large chunks,
1860 * we wait until bio submission time. All the pages in the bio are
1861 * checksummed and sums are attached onto the ordered extent record.
1862 *
1863 * At IO completion time the cums attached on the ordered extent record
1864 * are inserted into the btree
1865 */
1866 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1867 struct bio *bio, int mirror_num,
1868 unsigned long bio_flags,
1869 u64 bio_offset)
1870 {
1871 struct btrfs_root *root = BTRFS_I(inode)->root;
1872 int ret = 0;
1873
1874 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1875 BUG_ON(ret); /* -ENOMEM */
1876 return 0;
1877 }
1878
1879 /*
1880 * in order to insert checksums into the metadata in large chunks,
1881 * we wait until bio submission time. All the pages in the bio are
1882 * checksummed and sums are attached onto the ordered extent record.
1883 *
1884 * At IO completion time the cums attached on the ordered extent record
1885 * are inserted into the btree
1886 */
1887 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1888 int mirror_num, unsigned long bio_flags,
1889 u64 bio_offset)
1890 {
1891 struct btrfs_root *root = BTRFS_I(inode)->root;
1892 int ret;
1893
1894 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1895 if (ret) {
1896 bio->bi_error = ret;
1897 bio_endio(bio);
1898 }
1899 return ret;
1900 }
1901
1902 /*
1903 * extent_io.c submission hook. This does the right thing for csum calculation
1904 * on write, or reading the csums from the tree before a read
1905 */
1906 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1907 int mirror_num, unsigned long bio_flags,
1908 u64 bio_offset)
1909 {
1910 struct btrfs_root *root = BTRFS_I(inode)->root;
1911 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1912 int ret = 0;
1913 int skip_sum;
1914 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1915
1916 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1917
1918 if (btrfs_is_free_space_inode(inode))
1919 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1920
1921 if (!(rw & REQ_WRITE)) {
1922 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1923 if (ret)
1924 goto out;
1925
1926 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1927 ret = btrfs_submit_compressed_read(inode, bio,
1928 mirror_num,
1929 bio_flags);
1930 goto out;
1931 } else if (!skip_sum) {
1932 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1933 if (ret)
1934 goto out;
1935 }
1936 goto mapit;
1937 } else if (async && !skip_sum) {
1938 /* csum items have already been cloned */
1939 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1940 goto mapit;
1941 /* we're doing a write, do the async checksumming */
1942 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1943 inode, rw, bio, mirror_num,
1944 bio_flags, bio_offset,
1945 __btrfs_submit_bio_start,
1946 __btrfs_submit_bio_done);
1947 goto out;
1948 } else if (!skip_sum) {
1949 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1950 if (ret)
1951 goto out;
1952 }
1953
1954 mapit:
1955 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1956
1957 out:
1958 if (ret < 0) {
1959 bio->bi_error = ret;
1960 bio_endio(bio);
1961 }
1962 return ret;
1963 }
1964
1965 /*
1966 * given a list of ordered sums record them in the inode. This happens
1967 * at IO completion time based on sums calculated at bio submission time.
1968 */
1969 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1970 struct inode *inode, u64 file_offset,
1971 struct list_head *list)
1972 {
1973 struct btrfs_ordered_sum *sum;
1974
1975 list_for_each_entry(sum, list, list) {
1976 trans->adding_csums = 1;
1977 btrfs_csum_file_blocks(trans,
1978 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1979 trans->adding_csums = 0;
1980 }
1981 return 0;
1982 }
1983
1984 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1985 struct extent_state **cached_state)
1986 {
1987 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1988 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1989 cached_state, GFP_NOFS);
1990 }
1991
1992 /* see btrfs_writepage_start_hook for details on why this is required */
1993 struct btrfs_writepage_fixup {
1994 struct page *page;
1995 struct btrfs_work work;
1996 };
1997
1998 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1999 {
2000 struct btrfs_writepage_fixup *fixup;
2001 struct btrfs_ordered_extent *ordered;
2002 struct extent_state *cached_state = NULL;
2003 struct page *page;
2004 struct inode *inode;
2005 u64 page_start;
2006 u64 page_end;
2007 int ret;
2008
2009 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2010 page = fixup->page;
2011 again:
2012 lock_page(page);
2013 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2014 ClearPageChecked(page);
2015 goto out_page;
2016 }
2017
2018 inode = page->mapping->host;
2019 page_start = page_offset(page);
2020 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
2021
2022 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
2023 &cached_state);
2024
2025 /* already ordered? We're done */
2026 if (PagePrivate2(page))
2027 goto out;
2028
2029 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2030 if (ordered) {
2031 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2032 page_end, &cached_state, GFP_NOFS);
2033 unlock_page(page);
2034 btrfs_start_ordered_extent(inode, ordered, 1);
2035 btrfs_put_ordered_extent(ordered);
2036 goto again;
2037 }
2038
2039 ret = btrfs_delalloc_reserve_space(inode, page_start,
2040 PAGE_CACHE_SIZE);
2041 if (ret) {
2042 mapping_set_error(page->mapping, ret);
2043 end_extent_writepage(page, ret, page_start, page_end);
2044 ClearPageChecked(page);
2045 goto out;
2046 }
2047
2048 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
2049 &cached_state);
2050 if (ret) {
2051 mapping_set_error(page->mapping, ret);
2052 end_extent_writepage(page, ret, page_start, page_end);
2053 ClearPageChecked(page);
2054 goto out;
2055 }
2056
2057 ClearPageChecked(page);
2058 set_page_dirty(page);
2059 out:
2060 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2061 &cached_state, GFP_NOFS);
2062 out_page:
2063 unlock_page(page);
2064 page_cache_release(page);
2065 kfree(fixup);
2066 }
2067
2068 /*
2069 * There are a few paths in the higher layers of the kernel that directly
2070 * set the page dirty bit without asking the filesystem if it is a
2071 * good idea. This causes problems because we want to make sure COW
2072 * properly happens and the data=ordered rules are followed.
2073 *
2074 * In our case any range that doesn't have the ORDERED bit set
2075 * hasn't been properly setup for IO. We kick off an async process
2076 * to fix it up. The async helper will wait for ordered extents, set
2077 * the delalloc bit and make it safe to write the page.
2078 */
2079 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2080 {
2081 struct inode *inode = page->mapping->host;
2082 struct btrfs_writepage_fixup *fixup;
2083 struct btrfs_root *root = BTRFS_I(inode)->root;
2084
2085 /* this page is properly in the ordered list */
2086 if (TestClearPagePrivate2(page))
2087 return 0;
2088
2089 if (PageChecked(page))
2090 return -EAGAIN;
2091
2092 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2093 if (!fixup)
2094 return -EAGAIN;
2095
2096 SetPageChecked(page);
2097 page_cache_get(page);
2098 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2099 btrfs_writepage_fixup_worker, NULL, NULL);
2100 fixup->page = page;
2101 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2102 return -EBUSY;
2103 }
2104
2105 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2106 struct inode *inode, u64 file_pos,
2107 u64 disk_bytenr, u64 disk_num_bytes,
2108 u64 num_bytes, u64 ram_bytes,
2109 u8 compression, u8 encryption,
2110 u16 other_encoding, int extent_type)
2111 {
2112 struct btrfs_root *root = BTRFS_I(inode)->root;
2113 struct btrfs_file_extent_item *fi;
2114 struct btrfs_path *path;
2115 struct extent_buffer *leaf;
2116 struct btrfs_key ins;
2117 int extent_inserted = 0;
2118 int ret;
2119
2120 path = btrfs_alloc_path();
2121 if (!path)
2122 return -ENOMEM;
2123
2124 /*
2125 * we may be replacing one extent in the tree with another.
2126 * The new extent is pinned in the extent map, and we don't want
2127 * to drop it from the cache until it is completely in the btree.
2128 *
2129 * So, tell btrfs_drop_extents to leave this extent in the cache.
2130 * the caller is expected to unpin it and allow it to be merged
2131 * with the others.
2132 */
2133 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2134 file_pos + num_bytes, NULL, 0,
2135 1, sizeof(*fi), &extent_inserted);
2136 if (ret)
2137 goto out;
2138
2139 if (!extent_inserted) {
2140 ins.objectid = btrfs_ino(inode);
2141 ins.offset = file_pos;
2142 ins.type = BTRFS_EXTENT_DATA_KEY;
2143
2144 path->leave_spinning = 1;
2145 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2146 sizeof(*fi));
2147 if (ret)
2148 goto out;
2149 }
2150 leaf = path->nodes[0];
2151 fi = btrfs_item_ptr(leaf, path->slots[0],
2152 struct btrfs_file_extent_item);
2153 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2154 btrfs_set_file_extent_type(leaf, fi, extent_type);
2155 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2156 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2157 btrfs_set_file_extent_offset(leaf, fi, 0);
2158 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2159 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2160 btrfs_set_file_extent_compression(leaf, fi, compression);
2161 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2162 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2163
2164 btrfs_mark_buffer_dirty(leaf);
2165 btrfs_release_path(path);
2166
2167 inode_add_bytes(inode, num_bytes);
2168
2169 ins.objectid = disk_bytenr;
2170 ins.offset = disk_num_bytes;
2171 ins.type = BTRFS_EXTENT_ITEM_KEY;
2172 ret = btrfs_alloc_reserved_file_extent(trans, root,
2173 root->root_key.objectid,
2174 btrfs_ino(inode), file_pos,
2175 ram_bytes, &ins);
2176 /*
2177 * Release the reserved range from inode dirty range map, as it is
2178 * already moved into delayed_ref_head
2179 */
2180 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2181 out:
2182 btrfs_free_path(path);
2183
2184 return ret;
2185 }
2186
2187 /* snapshot-aware defrag */
2188 struct sa_defrag_extent_backref {
2189 struct rb_node node;
2190 struct old_sa_defrag_extent *old;
2191 u64 root_id;
2192 u64 inum;
2193 u64 file_pos;
2194 u64 extent_offset;
2195 u64 num_bytes;
2196 u64 generation;
2197 };
2198
2199 struct old_sa_defrag_extent {
2200 struct list_head list;
2201 struct new_sa_defrag_extent *new;
2202
2203 u64 extent_offset;
2204 u64 bytenr;
2205 u64 offset;
2206 u64 len;
2207 int count;
2208 };
2209
2210 struct new_sa_defrag_extent {
2211 struct rb_root root;
2212 struct list_head head;
2213 struct btrfs_path *path;
2214 struct inode *inode;
2215 u64 file_pos;
2216 u64 len;
2217 u64 bytenr;
2218 u64 disk_len;
2219 u8 compress_type;
2220 };
2221
2222 static int backref_comp(struct sa_defrag_extent_backref *b1,
2223 struct sa_defrag_extent_backref *b2)
2224 {
2225 if (b1->root_id < b2->root_id)
2226 return -1;
2227 else if (b1->root_id > b2->root_id)
2228 return 1;
2229
2230 if (b1->inum < b2->inum)
2231 return -1;
2232 else if (b1->inum > b2->inum)
2233 return 1;
2234
2235 if (b1->file_pos < b2->file_pos)
2236 return -1;
2237 else if (b1->file_pos > b2->file_pos)
2238 return 1;
2239
2240 /*
2241 * [------------------------------] ===> (a range of space)
2242 * |<--->| |<---->| =============> (fs/file tree A)
2243 * |<---------------------------->| ===> (fs/file tree B)
2244 *
2245 * A range of space can refer to two file extents in one tree while
2246 * refer to only one file extent in another tree.
2247 *
2248 * So we may process a disk offset more than one time(two extents in A)
2249 * and locate at the same extent(one extent in B), then insert two same
2250 * backrefs(both refer to the extent in B).
2251 */
2252 return 0;
2253 }
2254
2255 static void backref_insert(struct rb_root *root,
2256 struct sa_defrag_extent_backref *backref)
2257 {
2258 struct rb_node **p = &root->rb_node;
2259 struct rb_node *parent = NULL;
2260 struct sa_defrag_extent_backref *entry;
2261 int ret;
2262
2263 while (*p) {
2264 parent = *p;
2265 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2266
2267 ret = backref_comp(backref, entry);
2268 if (ret < 0)
2269 p = &(*p)->rb_left;
2270 else
2271 p = &(*p)->rb_right;
2272 }
2273
2274 rb_link_node(&backref->node, parent, p);
2275 rb_insert_color(&backref->node, root);
2276 }
2277
2278 /*
2279 * Note the backref might has changed, and in this case we just return 0.
2280 */
2281 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2282 void *ctx)
2283 {
2284 struct btrfs_file_extent_item *extent;
2285 struct btrfs_fs_info *fs_info;
2286 struct old_sa_defrag_extent *old = ctx;
2287 struct new_sa_defrag_extent *new = old->new;
2288 struct btrfs_path *path = new->path;
2289 struct btrfs_key key;
2290 struct btrfs_root *root;
2291 struct sa_defrag_extent_backref *backref;
2292 struct extent_buffer *leaf;
2293 struct inode *inode = new->inode;
2294 int slot;
2295 int ret;
2296 u64 extent_offset;
2297 u64 num_bytes;
2298
2299 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2300 inum == btrfs_ino(inode))
2301 return 0;
2302
2303 key.objectid = root_id;
2304 key.type = BTRFS_ROOT_ITEM_KEY;
2305 key.offset = (u64)-1;
2306
2307 fs_info = BTRFS_I(inode)->root->fs_info;
2308 root = btrfs_read_fs_root_no_name(fs_info, &key);
2309 if (IS_ERR(root)) {
2310 if (PTR_ERR(root) == -ENOENT)
2311 return 0;
2312 WARN_ON(1);
2313 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2314 inum, offset, root_id);
2315 return PTR_ERR(root);
2316 }
2317
2318 key.objectid = inum;
2319 key.type = BTRFS_EXTENT_DATA_KEY;
2320 if (offset > (u64)-1 << 32)
2321 key.offset = 0;
2322 else
2323 key.offset = offset;
2324
2325 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2326 if (WARN_ON(ret < 0))
2327 return ret;
2328 ret = 0;
2329
2330 while (1) {
2331 cond_resched();
2332
2333 leaf = path->nodes[0];
2334 slot = path->slots[0];
2335
2336 if (slot >= btrfs_header_nritems(leaf)) {
2337 ret = btrfs_next_leaf(root, path);
2338 if (ret < 0) {
2339 goto out;
2340 } else if (ret > 0) {
2341 ret = 0;
2342 goto out;
2343 }
2344 continue;
2345 }
2346
2347 path->slots[0]++;
2348
2349 btrfs_item_key_to_cpu(leaf, &key, slot);
2350
2351 if (key.objectid > inum)
2352 goto out;
2353
2354 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2355 continue;
2356
2357 extent = btrfs_item_ptr(leaf, slot,
2358 struct btrfs_file_extent_item);
2359
2360 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2361 continue;
2362
2363 /*
2364 * 'offset' refers to the exact key.offset,
2365 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2366 * (key.offset - extent_offset).
2367 */
2368 if (key.offset != offset)
2369 continue;
2370
2371 extent_offset = btrfs_file_extent_offset(leaf, extent);
2372 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2373
2374 if (extent_offset >= old->extent_offset + old->offset +
2375 old->len || extent_offset + num_bytes <=
2376 old->extent_offset + old->offset)
2377 continue;
2378 break;
2379 }
2380
2381 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2382 if (!backref) {
2383 ret = -ENOENT;
2384 goto out;
2385 }
2386
2387 backref->root_id = root_id;
2388 backref->inum = inum;
2389 backref->file_pos = offset;
2390 backref->num_bytes = num_bytes;
2391 backref->extent_offset = extent_offset;
2392 backref->generation = btrfs_file_extent_generation(leaf, extent);
2393 backref->old = old;
2394 backref_insert(&new->root, backref);
2395 old->count++;
2396 out:
2397 btrfs_release_path(path);
2398 WARN_ON(ret);
2399 return ret;
2400 }
2401
2402 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2403 struct new_sa_defrag_extent *new)
2404 {
2405 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2406 struct old_sa_defrag_extent *old, *tmp;
2407 int ret;
2408
2409 new->path = path;
2410
2411 list_for_each_entry_safe(old, tmp, &new->head, list) {
2412 ret = iterate_inodes_from_logical(old->bytenr +
2413 old->extent_offset, fs_info,
2414 path, record_one_backref,
2415 old);
2416 if (ret < 0 && ret != -ENOENT)
2417 return false;
2418
2419 /* no backref to be processed for this extent */
2420 if (!old->count) {
2421 list_del(&old->list);
2422 kfree(old);
2423 }
2424 }
2425
2426 if (list_empty(&new->head))
2427 return false;
2428
2429 return true;
2430 }
2431
2432 static int relink_is_mergable(struct extent_buffer *leaf,
2433 struct btrfs_file_extent_item *fi,
2434 struct new_sa_defrag_extent *new)
2435 {
2436 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2437 return 0;
2438
2439 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2440 return 0;
2441
2442 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2443 return 0;
2444
2445 if (btrfs_file_extent_encryption(leaf, fi) ||
2446 btrfs_file_extent_other_encoding(leaf, fi))
2447 return 0;
2448
2449 return 1;
2450 }
2451
2452 /*
2453 * Note the backref might has changed, and in this case we just return 0.
2454 */
2455 static noinline int relink_extent_backref(struct btrfs_path *path,
2456 struct sa_defrag_extent_backref *prev,
2457 struct sa_defrag_extent_backref *backref)
2458 {
2459 struct btrfs_file_extent_item *extent;
2460 struct btrfs_file_extent_item *item;
2461 struct btrfs_ordered_extent *ordered;
2462 struct btrfs_trans_handle *trans;
2463 struct btrfs_fs_info *fs_info;
2464 struct btrfs_root *root;
2465 struct btrfs_key key;
2466 struct extent_buffer *leaf;
2467 struct old_sa_defrag_extent *old = backref->old;
2468 struct new_sa_defrag_extent *new = old->new;
2469 struct inode *src_inode = new->inode;
2470 struct inode *inode;
2471 struct extent_state *cached = NULL;
2472 int ret = 0;
2473 u64 start;
2474 u64 len;
2475 u64 lock_start;
2476 u64 lock_end;
2477 bool merge = false;
2478 int index;
2479
2480 if (prev && prev->root_id == backref->root_id &&
2481 prev->inum == backref->inum &&
2482 prev->file_pos + prev->num_bytes == backref->file_pos)
2483 merge = true;
2484
2485 /* step 1: get root */
2486 key.objectid = backref->root_id;
2487 key.type = BTRFS_ROOT_ITEM_KEY;
2488 key.offset = (u64)-1;
2489
2490 fs_info = BTRFS_I(src_inode)->root->fs_info;
2491 index = srcu_read_lock(&fs_info->subvol_srcu);
2492
2493 root = btrfs_read_fs_root_no_name(fs_info, &key);
2494 if (IS_ERR(root)) {
2495 srcu_read_unlock(&fs_info->subvol_srcu, index);
2496 if (PTR_ERR(root) == -ENOENT)
2497 return 0;
2498 return PTR_ERR(root);
2499 }
2500
2501 if (btrfs_root_readonly(root)) {
2502 srcu_read_unlock(&fs_info->subvol_srcu, index);
2503 return 0;
2504 }
2505
2506 /* step 2: get inode */
2507 key.objectid = backref->inum;
2508 key.type = BTRFS_INODE_ITEM_KEY;
2509 key.offset = 0;
2510
2511 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2512 if (IS_ERR(inode)) {
2513 srcu_read_unlock(&fs_info->subvol_srcu, index);
2514 return 0;
2515 }
2516
2517 srcu_read_unlock(&fs_info->subvol_srcu, index);
2518
2519 /* step 3: relink backref */
2520 lock_start = backref->file_pos;
2521 lock_end = backref->file_pos + backref->num_bytes - 1;
2522 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2523 0, &cached);
2524
2525 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2526 if (ordered) {
2527 btrfs_put_ordered_extent(ordered);
2528 goto out_unlock;
2529 }
2530
2531 trans = btrfs_join_transaction(root);
2532 if (IS_ERR(trans)) {
2533 ret = PTR_ERR(trans);
2534 goto out_unlock;
2535 }
2536
2537 key.objectid = backref->inum;
2538 key.type = BTRFS_EXTENT_DATA_KEY;
2539 key.offset = backref->file_pos;
2540
2541 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2542 if (ret < 0) {
2543 goto out_free_path;
2544 } else if (ret > 0) {
2545 ret = 0;
2546 goto out_free_path;
2547 }
2548
2549 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2550 struct btrfs_file_extent_item);
2551
2552 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2553 backref->generation)
2554 goto out_free_path;
2555
2556 btrfs_release_path(path);
2557
2558 start = backref->file_pos;
2559 if (backref->extent_offset < old->extent_offset + old->offset)
2560 start += old->extent_offset + old->offset -
2561 backref->extent_offset;
2562
2563 len = min(backref->extent_offset + backref->num_bytes,
2564 old->extent_offset + old->offset + old->len);
2565 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2566
2567 ret = btrfs_drop_extents(trans, root, inode, start,
2568 start + len, 1);
2569 if (ret)
2570 goto out_free_path;
2571 again:
2572 key.objectid = btrfs_ino(inode);
2573 key.type = BTRFS_EXTENT_DATA_KEY;
2574 key.offset = start;
2575
2576 path->leave_spinning = 1;
2577 if (merge) {
2578 struct btrfs_file_extent_item *fi;
2579 u64 extent_len;
2580 struct btrfs_key found_key;
2581
2582 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2583 if (ret < 0)
2584 goto out_free_path;
2585
2586 path->slots[0]--;
2587 leaf = path->nodes[0];
2588 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2589
2590 fi = btrfs_item_ptr(leaf, path->slots[0],
2591 struct btrfs_file_extent_item);
2592 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2593
2594 if (extent_len + found_key.offset == start &&
2595 relink_is_mergable(leaf, fi, new)) {
2596 btrfs_set_file_extent_num_bytes(leaf, fi,
2597 extent_len + len);
2598 btrfs_mark_buffer_dirty(leaf);
2599 inode_add_bytes(inode, len);
2600
2601 ret = 1;
2602 goto out_free_path;
2603 } else {
2604 merge = false;
2605 btrfs_release_path(path);
2606 goto again;
2607 }
2608 }
2609
2610 ret = btrfs_insert_empty_item(trans, root, path, &key,
2611 sizeof(*extent));
2612 if (ret) {
2613 btrfs_abort_transaction(trans, root, ret);
2614 goto out_free_path;
2615 }
2616
2617 leaf = path->nodes[0];
2618 item = btrfs_item_ptr(leaf, path->slots[0],
2619 struct btrfs_file_extent_item);
2620 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2621 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2622 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2623 btrfs_set_file_extent_num_bytes(leaf, item, len);
2624 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2625 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2626 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2627 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2628 btrfs_set_file_extent_encryption(leaf, item, 0);
2629 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2630
2631 btrfs_mark_buffer_dirty(leaf);
2632 inode_add_bytes(inode, len);
2633 btrfs_release_path(path);
2634
2635 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2636 new->disk_len, 0,
2637 backref->root_id, backref->inum,
2638 new->file_pos); /* start - extent_offset */
2639 if (ret) {
2640 btrfs_abort_transaction(trans, root, ret);
2641 goto out_free_path;
2642 }
2643
2644 ret = 1;
2645 out_free_path:
2646 btrfs_release_path(path);
2647 path->leave_spinning = 0;
2648 btrfs_end_transaction(trans, root);
2649 out_unlock:
2650 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2651 &cached, GFP_NOFS);
2652 iput(inode);
2653 return ret;
2654 }
2655
2656 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2657 {
2658 struct old_sa_defrag_extent *old, *tmp;
2659
2660 if (!new)
2661 return;
2662
2663 list_for_each_entry_safe(old, tmp, &new->head, list) {
2664 kfree(old);
2665 }
2666 kfree(new);
2667 }
2668
2669 static void relink_file_extents(struct new_sa_defrag_extent *new)
2670 {
2671 struct btrfs_path *path;
2672 struct sa_defrag_extent_backref *backref;
2673 struct sa_defrag_extent_backref *prev = NULL;
2674 struct inode *inode;
2675 struct btrfs_root *root;
2676 struct rb_node *node;
2677 int ret;
2678
2679 inode = new->inode;
2680 root = BTRFS_I(inode)->root;
2681
2682 path = btrfs_alloc_path();
2683 if (!path)
2684 return;
2685
2686 if (!record_extent_backrefs(path, new)) {
2687 btrfs_free_path(path);
2688 goto out;
2689 }
2690 btrfs_release_path(path);
2691
2692 while (1) {
2693 node = rb_first(&new->root);
2694 if (!node)
2695 break;
2696 rb_erase(node, &new->root);
2697
2698 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2699
2700 ret = relink_extent_backref(path, prev, backref);
2701 WARN_ON(ret < 0);
2702
2703 kfree(prev);
2704
2705 if (ret == 1)
2706 prev = backref;
2707 else
2708 prev = NULL;
2709 cond_resched();
2710 }
2711 kfree(prev);
2712
2713 btrfs_free_path(path);
2714 out:
2715 free_sa_defrag_extent(new);
2716
2717 atomic_dec(&root->fs_info->defrag_running);
2718 wake_up(&root->fs_info->transaction_wait);
2719 }
2720
2721 static struct new_sa_defrag_extent *
2722 record_old_file_extents(struct inode *inode,
2723 struct btrfs_ordered_extent *ordered)
2724 {
2725 struct btrfs_root *root = BTRFS_I(inode)->root;
2726 struct btrfs_path *path;
2727 struct btrfs_key key;
2728 struct old_sa_defrag_extent *old;
2729 struct new_sa_defrag_extent *new;
2730 int ret;
2731
2732 new = kmalloc(sizeof(*new), GFP_NOFS);
2733 if (!new)
2734 return NULL;
2735
2736 new->inode = inode;
2737 new->file_pos = ordered->file_offset;
2738 new->len = ordered->len;
2739 new->bytenr = ordered->start;
2740 new->disk_len = ordered->disk_len;
2741 new->compress_type = ordered->compress_type;
2742 new->root = RB_ROOT;
2743 INIT_LIST_HEAD(&new->head);
2744
2745 path = btrfs_alloc_path();
2746 if (!path)
2747 goto out_kfree;
2748
2749 key.objectid = btrfs_ino(inode);
2750 key.type = BTRFS_EXTENT_DATA_KEY;
2751 key.offset = new->file_pos;
2752
2753 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2754 if (ret < 0)
2755 goto out_free_path;
2756 if (ret > 0 && path->slots[0] > 0)
2757 path->slots[0]--;
2758
2759 /* find out all the old extents for the file range */
2760 while (1) {
2761 struct btrfs_file_extent_item *extent;
2762 struct extent_buffer *l;
2763 int slot;
2764 u64 num_bytes;
2765 u64 offset;
2766 u64 end;
2767 u64 disk_bytenr;
2768 u64 extent_offset;
2769
2770 l = path->nodes[0];
2771 slot = path->slots[0];
2772
2773 if (slot >= btrfs_header_nritems(l)) {
2774 ret = btrfs_next_leaf(root, path);
2775 if (ret < 0)
2776 goto out_free_path;
2777 else if (ret > 0)
2778 break;
2779 continue;
2780 }
2781
2782 btrfs_item_key_to_cpu(l, &key, slot);
2783
2784 if (key.objectid != btrfs_ino(inode))
2785 break;
2786 if (key.type != BTRFS_EXTENT_DATA_KEY)
2787 break;
2788 if (key.offset >= new->file_pos + new->len)
2789 break;
2790
2791 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2792
2793 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2794 if (key.offset + num_bytes < new->file_pos)
2795 goto next;
2796
2797 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2798 if (!disk_bytenr)
2799 goto next;
2800
2801 extent_offset = btrfs_file_extent_offset(l, extent);
2802
2803 old = kmalloc(sizeof(*old), GFP_NOFS);
2804 if (!old)
2805 goto out_free_path;
2806
2807 offset = max(new->file_pos, key.offset);
2808 end = min(new->file_pos + new->len, key.offset + num_bytes);
2809
2810 old->bytenr = disk_bytenr;
2811 old->extent_offset = extent_offset;
2812 old->offset = offset - key.offset;
2813 old->len = end - offset;
2814 old->new = new;
2815 old->count = 0;
2816 list_add_tail(&old->list, &new->head);
2817 next:
2818 path->slots[0]++;
2819 cond_resched();
2820 }
2821
2822 btrfs_free_path(path);
2823 atomic_inc(&root->fs_info->defrag_running);
2824
2825 return new;
2826
2827 out_free_path:
2828 btrfs_free_path(path);
2829 out_kfree:
2830 free_sa_defrag_extent(new);
2831 return NULL;
2832 }
2833
2834 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2835 u64 start, u64 len)
2836 {
2837 struct btrfs_block_group_cache *cache;
2838
2839 cache = btrfs_lookup_block_group(root->fs_info, start);
2840 ASSERT(cache);
2841
2842 spin_lock(&cache->lock);
2843 cache->delalloc_bytes -= len;
2844 spin_unlock(&cache->lock);
2845
2846 btrfs_put_block_group(cache);
2847 }
2848
2849 /* as ordered data IO finishes, this gets called so we can finish
2850 * an ordered extent if the range of bytes in the file it covers are
2851 * fully written.
2852 */
2853 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2854 {
2855 struct inode *inode = ordered_extent->inode;
2856 struct btrfs_root *root = BTRFS_I(inode)->root;
2857 struct btrfs_trans_handle *trans = NULL;
2858 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2859 struct extent_state *cached_state = NULL;
2860 struct new_sa_defrag_extent *new = NULL;
2861 int compress_type = 0;
2862 int ret = 0;
2863 u64 logical_len = ordered_extent->len;
2864 bool nolock;
2865 bool truncated = false;
2866
2867 nolock = btrfs_is_free_space_inode(inode);
2868
2869 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2870 ret = -EIO;
2871 goto out;
2872 }
2873
2874 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2875 ordered_extent->file_offset +
2876 ordered_extent->len - 1);
2877
2878 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2879 truncated = true;
2880 logical_len = ordered_extent->truncated_len;
2881 /* Truncated the entire extent, don't bother adding */
2882 if (!logical_len)
2883 goto out;
2884 }
2885
2886 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2887 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2888
2889 /*
2890 * For mwrite(mmap + memset to write) case, we still reserve
2891 * space for NOCOW range.
2892 * As NOCOW won't cause a new delayed ref, just free the space
2893 */
2894 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2895 ordered_extent->len);
2896 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2897 if (nolock)
2898 trans = btrfs_join_transaction_nolock(root);
2899 else
2900 trans = btrfs_join_transaction(root);
2901 if (IS_ERR(trans)) {
2902 ret = PTR_ERR(trans);
2903 trans = NULL;
2904 goto out;
2905 }
2906 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2907 ret = btrfs_update_inode_fallback(trans, root, inode);
2908 if (ret) /* -ENOMEM or corruption */
2909 btrfs_abort_transaction(trans, root, ret);
2910 goto out;
2911 }
2912
2913 lock_extent_bits(io_tree, ordered_extent->file_offset,
2914 ordered_extent->file_offset + ordered_extent->len - 1,
2915 0, &cached_state);
2916
2917 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2918 ordered_extent->file_offset + ordered_extent->len - 1,
2919 EXTENT_DEFRAG, 1, cached_state);
2920 if (ret) {
2921 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2922 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2923 /* the inode is shared */
2924 new = record_old_file_extents(inode, ordered_extent);
2925
2926 clear_extent_bit(io_tree, ordered_extent->file_offset,
2927 ordered_extent->file_offset + ordered_extent->len - 1,
2928 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2929 }
2930
2931 if (nolock)
2932 trans = btrfs_join_transaction_nolock(root);
2933 else
2934 trans = btrfs_join_transaction(root);
2935 if (IS_ERR(trans)) {
2936 ret = PTR_ERR(trans);
2937 trans = NULL;
2938 goto out_unlock;
2939 }
2940
2941 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2942
2943 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2944 compress_type = ordered_extent->compress_type;
2945 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2946 BUG_ON(compress_type);
2947 ret = btrfs_mark_extent_written(trans, inode,
2948 ordered_extent->file_offset,
2949 ordered_extent->file_offset +
2950 logical_len);
2951 } else {
2952 BUG_ON(root == root->fs_info->tree_root);
2953 ret = insert_reserved_file_extent(trans, inode,
2954 ordered_extent->file_offset,
2955 ordered_extent->start,
2956 ordered_extent->disk_len,
2957 logical_len, logical_len,
2958 compress_type, 0, 0,
2959 BTRFS_FILE_EXTENT_REG);
2960 if (!ret)
2961 btrfs_release_delalloc_bytes(root,
2962 ordered_extent->start,
2963 ordered_extent->disk_len);
2964 }
2965 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2966 ordered_extent->file_offset, ordered_extent->len,
2967 trans->transid);
2968 if (ret < 0) {
2969 btrfs_abort_transaction(trans, root, ret);
2970 goto out_unlock;
2971 }
2972
2973 add_pending_csums(trans, inode, ordered_extent->file_offset,
2974 &ordered_extent->list);
2975
2976 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2977 ret = btrfs_update_inode_fallback(trans, root, inode);
2978 if (ret) { /* -ENOMEM or corruption */
2979 btrfs_abort_transaction(trans, root, ret);
2980 goto out_unlock;
2981 }
2982 ret = 0;
2983 out_unlock:
2984 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2985 ordered_extent->file_offset +
2986 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2987 out:
2988 if (root != root->fs_info->tree_root)
2989 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2990 if (trans)
2991 btrfs_end_transaction(trans, root);
2992
2993 if (ret || truncated) {
2994 u64 start, end;
2995
2996 if (truncated)
2997 start = ordered_extent->file_offset + logical_len;
2998 else
2999 start = ordered_extent->file_offset;
3000 end = ordered_extent->file_offset + ordered_extent->len - 1;
3001 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3002
3003 /* Drop the cache for the part of the extent we didn't write. */
3004 btrfs_drop_extent_cache(inode, start, end, 0);
3005
3006 /*
3007 * If the ordered extent had an IOERR or something else went
3008 * wrong we need to return the space for this ordered extent
3009 * back to the allocator. We only free the extent in the
3010 * truncated case if we didn't write out the extent at all.
3011 */
3012 if ((ret || !logical_len) &&
3013 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3014 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3015 btrfs_free_reserved_extent(root, ordered_extent->start,
3016 ordered_extent->disk_len, 1);
3017 }
3018
3019
3020 /*
3021 * This needs to be done to make sure anybody waiting knows we are done
3022 * updating everything for this ordered extent.
3023 */
3024 btrfs_remove_ordered_extent(inode, ordered_extent);
3025
3026 /* for snapshot-aware defrag */
3027 if (new) {
3028 if (ret) {
3029 free_sa_defrag_extent(new);
3030 atomic_dec(&root->fs_info->defrag_running);
3031 } else {
3032 relink_file_extents(new);
3033 }
3034 }
3035
3036 /* once for us */
3037 btrfs_put_ordered_extent(ordered_extent);
3038 /* once for the tree */
3039 btrfs_put_ordered_extent(ordered_extent);
3040
3041 return ret;
3042 }
3043
3044 static void finish_ordered_fn(struct btrfs_work *work)
3045 {
3046 struct btrfs_ordered_extent *ordered_extent;
3047 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3048 btrfs_finish_ordered_io(ordered_extent);
3049 }
3050
3051 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3052 struct extent_state *state, int uptodate)
3053 {
3054 struct inode *inode = page->mapping->host;
3055 struct btrfs_root *root = BTRFS_I(inode)->root;
3056 struct btrfs_ordered_extent *ordered_extent = NULL;
3057 struct btrfs_workqueue *wq;
3058 btrfs_work_func_t func;
3059
3060 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3061
3062 ClearPagePrivate2(page);
3063 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3064 end - start + 1, uptodate))
3065 return 0;
3066
3067 if (btrfs_is_free_space_inode(inode)) {
3068 wq = root->fs_info->endio_freespace_worker;
3069 func = btrfs_freespace_write_helper;
3070 } else {
3071 wq = root->fs_info->endio_write_workers;
3072 func = btrfs_endio_write_helper;
3073 }
3074
3075 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3076 NULL);
3077 btrfs_queue_work(wq, &ordered_extent->work);
3078
3079 return 0;
3080 }
3081
3082 static int __readpage_endio_check(struct inode *inode,
3083 struct btrfs_io_bio *io_bio,
3084 int icsum, struct page *page,
3085 int pgoff, u64 start, size_t len)
3086 {
3087 char *kaddr;
3088 u32 csum_expected;
3089 u32 csum = ~(u32)0;
3090
3091 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3092
3093 kaddr = kmap_atomic(page);
3094 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3095 btrfs_csum_final(csum, (char *)&csum);
3096 if (csum != csum_expected)
3097 goto zeroit;
3098
3099 kunmap_atomic(kaddr);
3100 return 0;
3101 zeroit:
3102 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3103 "csum failed ino %llu off %llu csum %u expected csum %u",
3104 btrfs_ino(inode), start, csum, csum_expected);
3105 memset(kaddr + pgoff, 1, len);
3106 flush_dcache_page(page);
3107 kunmap_atomic(kaddr);
3108 if (csum_expected == 0)
3109 return 0;
3110 return -EIO;
3111 }
3112
3113 /*
3114 * when reads are done, we need to check csums to verify the data is correct
3115 * if there's a match, we allow the bio to finish. If not, the code in
3116 * extent_io.c will try to find good copies for us.
3117 */
3118 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3119 u64 phy_offset, struct page *page,
3120 u64 start, u64 end, int mirror)
3121 {
3122 size_t offset = start - page_offset(page);
3123 struct inode *inode = page->mapping->host;
3124 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3125 struct btrfs_root *root = BTRFS_I(inode)->root;
3126
3127 if (PageChecked(page)) {
3128 ClearPageChecked(page);
3129 return 0;
3130 }
3131
3132 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3133 return 0;
3134
3135 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3136 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3137 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3138 GFP_NOFS);
3139 return 0;
3140 }
3141
3142 phy_offset >>= inode->i_sb->s_blocksize_bits;
3143 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3144 start, (size_t)(end - start + 1));
3145 }
3146
3147 struct delayed_iput {
3148 struct list_head list;
3149 struct inode *inode;
3150 };
3151
3152 /* JDM: If this is fs-wide, why can't we add a pointer to
3153 * btrfs_inode instead and avoid the allocation? */
3154 void btrfs_add_delayed_iput(struct inode *inode)
3155 {
3156 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3157 struct delayed_iput *delayed;
3158
3159 if (atomic_add_unless(&inode->i_count, -1, 1))
3160 return;
3161
3162 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3163 delayed->inode = inode;
3164
3165 spin_lock(&fs_info->delayed_iput_lock);
3166 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3167 spin_unlock(&fs_info->delayed_iput_lock);
3168 }
3169
3170 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3171 {
3172 LIST_HEAD(list);
3173 struct btrfs_fs_info *fs_info = root->fs_info;
3174 struct delayed_iput *delayed;
3175 int empty;
3176
3177 spin_lock(&fs_info->delayed_iput_lock);
3178 empty = list_empty(&fs_info->delayed_iputs);
3179 spin_unlock(&fs_info->delayed_iput_lock);
3180 if (empty)
3181 return;
3182
3183 spin_lock(&fs_info->delayed_iput_lock);
3184 list_splice_init(&fs_info->delayed_iputs, &list);
3185 spin_unlock(&fs_info->delayed_iput_lock);
3186
3187 while (!list_empty(&list)) {
3188 delayed = list_entry(list.next, struct delayed_iput, list);
3189 list_del(&delayed->list);
3190 iput(delayed->inode);
3191 kfree(delayed);
3192 }
3193 }
3194
3195 /*
3196 * This is called in transaction commit time. If there are no orphan
3197 * files in the subvolume, it removes orphan item and frees block_rsv
3198 * structure.
3199 */
3200 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3201 struct btrfs_root *root)
3202 {
3203 struct btrfs_block_rsv *block_rsv;
3204 int ret;
3205
3206 if (atomic_read(&root->orphan_inodes) ||
3207 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3208 return;
3209
3210 spin_lock(&root->orphan_lock);
3211 if (atomic_read(&root->orphan_inodes)) {
3212 spin_unlock(&root->orphan_lock);
3213 return;
3214 }
3215
3216 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3217 spin_unlock(&root->orphan_lock);
3218 return;
3219 }
3220
3221 block_rsv = root->orphan_block_rsv;
3222 root->orphan_block_rsv = NULL;
3223 spin_unlock(&root->orphan_lock);
3224
3225 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3226 btrfs_root_refs(&root->root_item) > 0) {
3227 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3228 root->root_key.objectid);
3229 if (ret)
3230 btrfs_abort_transaction(trans, root, ret);
3231 else
3232 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3233 &root->state);
3234 }
3235
3236 if (block_rsv) {
3237 WARN_ON(block_rsv->size > 0);
3238 btrfs_free_block_rsv(root, block_rsv);
3239 }
3240 }
3241
3242 /*
3243 * This creates an orphan entry for the given inode in case something goes
3244 * wrong in the middle of an unlink/truncate.
3245 *
3246 * NOTE: caller of this function should reserve 5 units of metadata for
3247 * this function.
3248 */
3249 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3250 {
3251 struct btrfs_root *root = BTRFS_I(inode)->root;
3252 struct btrfs_block_rsv *block_rsv = NULL;
3253 int reserve = 0;
3254 int insert = 0;
3255 int ret;
3256
3257 if (!root->orphan_block_rsv) {
3258 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3259 if (!block_rsv)
3260 return -ENOMEM;
3261 }
3262
3263 spin_lock(&root->orphan_lock);
3264 if (!root->orphan_block_rsv) {
3265 root->orphan_block_rsv = block_rsv;
3266 } else if (block_rsv) {
3267 btrfs_free_block_rsv(root, block_rsv);
3268 block_rsv = NULL;
3269 }
3270
3271 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3272 &BTRFS_I(inode)->runtime_flags)) {
3273 #if 0
3274 /*
3275 * For proper ENOSPC handling, we should do orphan
3276 * cleanup when mounting. But this introduces backward
3277 * compatibility issue.
3278 */
3279 if (!xchg(&root->orphan_item_inserted, 1))
3280 insert = 2;
3281 else
3282 insert = 1;
3283 #endif
3284 insert = 1;
3285 atomic_inc(&root->orphan_inodes);
3286 }
3287
3288 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3289 &BTRFS_I(inode)->runtime_flags))
3290 reserve = 1;
3291 spin_unlock(&root->orphan_lock);
3292
3293 /* grab metadata reservation from transaction handle */
3294 if (reserve) {
3295 ret = btrfs_orphan_reserve_metadata(trans, inode);
3296 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3297 }
3298
3299 /* insert an orphan item to track this unlinked/truncated file */
3300 if (insert >= 1) {
3301 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3302 if (ret) {
3303 atomic_dec(&root->orphan_inodes);
3304 if (reserve) {
3305 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3306 &BTRFS_I(inode)->runtime_flags);
3307 btrfs_orphan_release_metadata(inode);
3308 }
3309 if (ret != -EEXIST) {
3310 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3311 &BTRFS_I(inode)->runtime_flags);
3312 btrfs_abort_transaction(trans, root, ret);
3313 return ret;
3314 }
3315 }
3316 ret = 0;
3317 }
3318
3319 /* insert an orphan item to track subvolume contains orphan files */
3320 if (insert >= 2) {
3321 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3322 root->root_key.objectid);
3323 if (ret && ret != -EEXIST) {
3324 btrfs_abort_transaction(trans, root, ret);
3325 return ret;
3326 }
3327 }
3328 return 0;
3329 }
3330
3331 /*
3332 * We have done the truncate/delete so we can go ahead and remove the orphan
3333 * item for this particular inode.
3334 */
3335 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3336 struct inode *inode)
3337 {
3338 struct btrfs_root *root = BTRFS_I(inode)->root;
3339 int delete_item = 0;
3340 int release_rsv = 0;
3341 int ret = 0;
3342
3343 spin_lock(&root->orphan_lock);
3344 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3345 &BTRFS_I(inode)->runtime_flags))
3346 delete_item = 1;
3347
3348 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3349 &BTRFS_I(inode)->runtime_flags))
3350 release_rsv = 1;
3351 spin_unlock(&root->orphan_lock);
3352
3353 if (delete_item) {
3354 atomic_dec(&root->orphan_inodes);
3355 if (trans)
3356 ret = btrfs_del_orphan_item(trans, root,
3357 btrfs_ino(inode));
3358 }
3359
3360 if (release_rsv)
3361 btrfs_orphan_release_metadata(inode);
3362
3363 return ret;
3364 }
3365
3366 /*
3367 * this cleans up any orphans that may be left on the list from the last use
3368 * of this root.
3369 */
3370 int btrfs_orphan_cleanup(struct btrfs_root *root)
3371 {
3372 struct btrfs_path *path;
3373 struct extent_buffer *leaf;
3374 struct btrfs_key key, found_key;
3375 struct btrfs_trans_handle *trans;
3376 struct inode *inode;
3377 u64 last_objectid = 0;
3378 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3379
3380 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3381 return 0;
3382
3383 path = btrfs_alloc_path();
3384 if (!path) {
3385 ret = -ENOMEM;
3386 goto out;
3387 }
3388 path->reada = -1;
3389
3390 key.objectid = BTRFS_ORPHAN_OBJECTID;
3391 key.type = BTRFS_ORPHAN_ITEM_KEY;
3392 key.offset = (u64)-1;
3393
3394 while (1) {
3395 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3396 if (ret < 0)
3397 goto out;
3398
3399 /*
3400 * if ret == 0 means we found what we were searching for, which
3401 * is weird, but possible, so only screw with path if we didn't
3402 * find the key and see if we have stuff that matches
3403 */
3404 if (ret > 0) {
3405 ret = 0;
3406 if (path->slots[0] == 0)
3407 break;
3408 path->slots[0]--;
3409 }
3410
3411 /* pull out the item */
3412 leaf = path->nodes[0];
3413 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3414
3415 /* make sure the item matches what we want */
3416 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3417 break;
3418 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3419 break;
3420
3421 /* release the path since we're done with it */
3422 btrfs_release_path(path);
3423
3424 /*
3425 * this is where we are basically btrfs_lookup, without the
3426 * crossing root thing. we store the inode number in the
3427 * offset of the orphan item.
3428 */
3429
3430 if (found_key.offset == last_objectid) {
3431 btrfs_err(root->fs_info,
3432 "Error removing orphan entry, stopping orphan cleanup");
3433 ret = -EINVAL;
3434 goto out;
3435 }
3436
3437 last_objectid = found_key.offset;
3438
3439 found_key.objectid = found_key.offset;
3440 found_key.type = BTRFS_INODE_ITEM_KEY;
3441 found_key.offset = 0;
3442 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3443 ret = PTR_ERR_OR_ZERO(inode);
3444 if (ret && ret != -ESTALE)
3445 goto out;
3446
3447 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3448 struct btrfs_root *dead_root;
3449 struct btrfs_fs_info *fs_info = root->fs_info;
3450 int is_dead_root = 0;
3451
3452 /*
3453 * this is an orphan in the tree root. Currently these
3454 * could come from 2 sources:
3455 * a) a snapshot deletion in progress
3456 * b) a free space cache inode
3457 * We need to distinguish those two, as the snapshot
3458 * orphan must not get deleted.
3459 * find_dead_roots already ran before us, so if this
3460 * is a snapshot deletion, we should find the root
3461 * in the dead_roots list
3462 */
3463 spin_lock(&fs_info->trans_lock);
3464 list_for_each_entry(dead_root, &fs_info->dead_roots,
3465 root_list) {
3466 if (dead_root->root_key.objectid ==
3467 found_key.objectid) {
3468 is_dead_root = 1;
3469 break;
3470 }
3471 }
3472 spin_unlock(&fs_info->trans_lock);
3473 if (is_dead_root) {
3474 /* prevent this orphan from being found again */
3475 key.offset = found_key.objectid - 1;
3476 continue;
3477 }
3478 }
3479 /*
3480 * Inode is already gone but the orphan item is still there,
3481 * kill the orphan item.
3482 */
3483 if (ret == -ESTALE) {
3484 trans = btrfs_start_transaction(root, 1);
3485 if (IS_ERR(trans)) {
3486 ret = PTR_ERR(trans);
3487 goto out;
3488 }
3489 btrfs_debug(root->fs_info, "auto deleting %Lu",
3490 found_key.objectid);
3491 ret = btrfs_del_orphan_item(trans, root,
3492 found_key.objectid);
3493 btrfs_end_transaction(trans, root);
3494 if (ret)
3495 goto out;
3496 continue;
3497 }
3498
3499 /*
3500 * add this inode to the orphan list so btrfs_orphan_del does
3501 * the proper thing when we hit it
3502 */
3503 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3504 &BTRFS_I(inode)->runtime_flags);
3505 atomic_inc(&root->orphan_inodes);
3506
3507 /* if we have links, this was a truncate, lets do that */
3508 if (inode->i_nlink) {
3509 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3510 iput(inode);
3511 continue;
3512 }
3513 nr_truncate++;
3514
3515 /* 1 for the orphan item deletion. */
3516 trans = btrfs_start_transaction(root, 1);
3517 if (IS_ERR(trans)) {
3518 iput(inode);
3519 ret = PTR_ERR(trans);
3520 goto out;
3521 }
3522 ret = btrfs_orphan_add(trans, inode);
3523 btrfs_end_transaction(trans, root);
3524 if (ret) {
3525 iput(inode);
3526 goto out;
3527 }
3528
3529 ret = btrfs_truncate(inode);
3530 if (ret)
3531 btrfs_orphan_del(NULL, inode);
3532 } else {
3533 nr_unlink++;
3534 }
3535
3536 /* this will do delete_inode and everything for us */
3537 iput(inode);
3538 if (ret)
3539 goto out;
3540 }
3541 /* release the path since we're done with it */
3542 btrfs_release_path(path);
3543
3544 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3545
3546 if (root->orphan_block_rsv)
3547 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3548 (u64)-1);
3549
3550 if (root->orphan_block_rsv ||
3551 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3552 trans = btrfs_join_transaction(root);
3553 if (!IS_ERR(trans))
3554 btrfs_end_transaction(trans, root);
3555 }
3556
3557 if (nr_unlink)
3558 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3559 if (nr_truncate)
3560 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3561
3562 out:
3563 if (ret)
3564 btrfs_err(root->fs_info,
3565 "could not do orphan cleanup %d", ret);
3566 btrfs_free_path(path);
3567 return ret;
3568 }
3569
3570 /*
3571 * very simple check to peek ahead in the leaf looking for xattrs. If we
3572 * don't find any xattrs, we know there can't be any acls.
3573 *
3574 * slot is the slot the inode is in, objectid is the objectid of the inode
3575 */
3576 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3577 int slot, u64 objectid,
3578 int *first_xattr_slot)
3579 {
3580 u32 nritems = btrfs_header_nritems(leaf);
3581 struct btrfs_key found_key;
3582 static u64 xattr_access = 0;
3583 static u64 xattr_default = 0;
3584 int scanned = 0;
3585
3586 if (!xattr_access) {
3587 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3588 strlen(POSIX_ACL_XATTR_ACCESS));
3589 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3590 strlen(POSIX_ACL_XATTR_DEFAULT));
3591 }
3592
3593 slot++;
3594 *first_xattr_slot = -1;
3595 while (slot < nritems) {
3596 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3597
3598 /* we found a different objectid, there must not be acls */
3599 if (found_key.objectid != objectid)
3600 return 0;
3601
3602 /* we found an xattr, assume we've got an acl */
3603 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3604 if (*first_xattr_slot == -1)
3605 *first_xattr_slot = slot;
3606 if (found_key.offset == xattr_access ||
3607 found_key.offset == xattr_default)
3608 return 1;
3609 }
3610
3611 /*
3612 * we found a key greater than an xattr key, there can't
3613 * be any acls later on
3614 */
3615 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3616 return 0;
3617
3618 slot++;
3619 scanned++;
3620
3621 /*
3622 * it goes inode, inode backrefs, xattrs, extents,
3623 * so if there are a ton of hard links to an inode there can
3624 * be a lot of backrefs. Don't waste time searching too hard,
3625 * this is just an optimization
3626 */
3627 if (scanned >= 8)
3628 break;
3629 }
3630 /* we hit the end of the leaf before we found an xattr or
3631 * something larger than an xattr. We have to assume the inode
3632 * has acls
3633 */
3634 if (*first_xattr_slot == -1)
3635 *first_xattr_slot = slot;
3636 return 1;
3637 }
3638
3639 /*
3640 * read an inode from the btree into the in-memory inode
3641 */
3642 static void btrfs_read_locked_inode(struct inode *inode)
3643 {
3644 struct btrfs_path *path;
3645 struct extent_buffer *leaf;
3646 struct btrfs_inode_item *inode_item;
3647 struct btrfs_root *root = BTRFS_I(inode)->root;
3648 struct btrfs_key location;
3649 unsigned long ptr;
3650 int maybe_acls;
3651 u32 rdev;
3652 int ret;
3653 bool filled = false;
3654 int first_xattr_slot;
3655
3656 ret = btrfs_fill_inode(inode, &rdev);
3657 if (!ret)
3658 filled = true;
3659
3660 path = btrfs_alloc_path();
3661 if (!path)
3662 goto make_bad;
3663
3664 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3665
3666 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3667 if (ret)
3668 goto make_bad;
3669
3670 leaf = path->nodes[0];
3671
3672 if (filled)
3673 goto cache_index;
3674
3675 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3676 struct btrfs_inode_item);
3677 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3678 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3679 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3680 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3681 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3682
3683 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3684 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3685
3686 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3687 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3688
3689 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3690 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3691
3692 BTRFS_I(inode)->i_otime.tv_sec =
3693 btrfs_timespec_sec(leaf, &inode_item->otime);
3694 BTRFS_I(inode)->i_otime.tv_nsec =
3695 btrfs_timespec_nsec(leaf, &inode_item->otime);
3696
3697 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3698 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3699 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3700
3701 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3702 inode->i_generation = BTRFS_I(inode)->generation;
3703 inode->i_rdev = 0;
3704 rdev = btrfs_inode_rdev(leaf, inode_item);
3705
3706 BTRFS_I(inode)->index_cnt = (u64)-1;
3707 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3708
3709 cache_index:
3710 /*
3711 * If we were modified in the current generation and evicted from memory
3712 * and then re-read we need to do a full sync since we don't have any
3713 * idea about which extents were modified before we were evicted from
3714 * cache.
3715 *
3716 * This is required for both inode re-read from disk and delayed inode
3717 * in delayed_nodes_tree.
3718 */
3719 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3720 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3721 &BTRFS_I(inode)->runtime_flags);
3722
3723 /*
3724 * We don't persist the id of the transaction where an unlink operation
3725 * against the inode was last made. So here we assume the inode might
3726 * have been evicted, and therefore the exact value of last_unlink_trans
3727 * lost, and set it to last_trans to avoid metadata inconsistencies
3728 * between the inode and its parent if the inode is fsync'ed and the log
3729 * replayed. For example, in the scenario:
3730 *
3731 * touch mydir/foo
3732 * ln mydir/foo mydir/bar
3733 * sync
3734 * unlink mydir/bar
3735 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3736 * xfs_io -c fsync mydir/foo
3737 * <power failure>
3738 * mount fs, triggers fsync log replay
3739 *
3740 * We must make sure that when we fsync our inode foo we also log its
3741 * parent inode, otherwise after log replay the parent still has the
3742 * dentry with the "bar" name but our inode foo has a link count of 1
3743 * and doesn't have an inode ref with the name "bar" anymore.
3744 *
3745 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3746 * but it guarantees correctness at the expense of ocassional full
3747 * transaction commits on fsync if our inode is a directory, or if our
3748 * inode is not a directory, logging its parent unnecessarily.
3749 */
3750 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3751
3752 path->slots[0]++;
3753 if (inode->i_nlink != 1 ||
3754 path->slots[0] >= btrfs_header_nritems(leaf))
3755 goto cache_acl;
3756
3757 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3758 if (location.objectid != btrfs_ino(inode))
3759 goto cache_acl;
3760
3761 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3762 if (location.type == BTRFS_INODE_REF_KEY) {
3763 struct btrfs_inode_ref *ref;
3764
3765 ref = (struct btrfs_inode_ref *)ptr;
3766 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3767 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3768 struct btrfs_inode_extref *extref;
3769
3770 extref = (struct btrfs_inode_extref *)ptr;
3771 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3772 extref);
3773 }
3774 cache_acl:
3775 /*
3776 * try to precache a NULL acl entry for files that don't have
3777 * any xattrs or acls
3778 */
3779 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3780 btrfs_ino(inode), &first_xattr_slot);
3781 if (first_xattr_slot != -1) {
3782 path->slots[0] = first_xattr_slot;
3783 ret = btrfs_load_inode_props(inode, path);
3784 if (ret)
3785 btrfs_err(root->fs_info,
3786 "error loading props for ino %llu (root %llu): %d",
3787 btrfs_ino(inode),
3788 root->root_key.objectid, ret);
3789 }
3790 btrfs_free_path(path);
3791
3792 if (!maybe_acls)
3793 cache_no_acl(inode);
3794
3795 switch (inode->i_mode & S_IFMT) {
3796 case S_IFREG:
3797 inode->i_mapping->a_ops = &btrfs_aops;
3798 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3799 inode->i_fop = &btrfs_file_operations;
3800 inode->i_op = &btrfs_file_inode_operations;
3801 break;
3802 case S_IFDIR:
3803 inode->i_fop = &btrfs_dir_file_operations;
3804 if (root == root->fs_info->tree_root)
3805 inode->i_op = &btrfs_dir_ro_inode_operations;
3806 else
3807 inode->i_op = &btrfs_dir_inode_operations;
3808 break;
3809 case S_IFLNK:
3810 inode->i_op = &btrfs_symlink_inode_operations;
3811 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3812 break;
3813 default:
3814 inode->i_op = &btrfs_special_inode_operations;
3815 init_special_inode(inode, inode->i_mode, rdev);
3816 break;
3817 }
3818
3819 btrfs_update_iflags(inode);
3820 return;
3821
3822 make_bad:
3823 btrfs_free_path(path);
3824 make_bad_inode(inode);
3825 }
3826
3827 /*
3828 * given a leaf and an inode, copy the inode fields into the leaf
3829 */
3830 static void fill_inode_item(struct btrfs_trans_handle *trans,
3831 struct extent_buffer *leaf,
3832 struct btrfs_inode_item *item,
3833 struct inode *inode)
3834 {
3835 struct btrfs_map_token token;
3836
3837 btrfs_init_map_token(&token);
3838
3839 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3840 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3841 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3842 &token);
3843 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3844 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3845
3846 btrfs_set_token_timespec_sec(leaf, &item->atime,
3847 inode->i_atime.tv_sec, &token);
3848 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3849 inode->i_atime.tv_nsec, &token);
3850
3851 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3852 inode->i_mtime.tv_sec, &token);
3853 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3854 inode->i_mtime.tv_nsec, &token);
3855
3856 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3857 inode->i_ctime.tv_sec, &token);
3858 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3859 inode->i_ctime.tv_nsec, &token);
3860
3861 btrfs_set_token_timespec_sec(leaf, &item->otime,
3862 BTRFS_I(inode)->i_otime.tv_sec, &token);
3863 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3864 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3865
3866 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3867 &token);
3868 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3869 &token);
3870 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3871 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3872 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3873 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3874 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3875 }
3876
3877 /*
3878 * copy everything in the in-memory inode into the btree.
3879 */
3880 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3881 struct btrfs_root *root, struct inode *inode)
3882 {
3883 struct btrfs_inode_item *inode_item;
3884 struct btrfs_path *path;
3885 struct extent_buffer *leaf;
3886 int ret;
3887
3888 path = btrfs_alloc_path();
3889 if (!path)
3890 return -ENOMEM;
3891
3892 path->leave_spinning = 1;
3893 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3894 1);
3895 if (ret) {
3896 if (ret > 0)
3897 ret = -ENOENT;
3898 goto failed;
3899 }
3900
3901 leaf = path->nodes[0];
3902 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3903 struct btrfs_inode_item);
3904
3905 fill_inode_item(trans, leaf, inode_item, inode);
3906 btrfs_mark_buffer_dirty(leaf);
3907 btrfs_set_inode_last_trans(trans, inode);
3908 ret = 0;
3909 failed:
3910 btrfs_free_path(path);
3911 return ret;
3912 }
3913
3914 /*
3915 * copy everything in the in-memory inode into the btree.
3916 */
3917 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3918 struct btrfs_root *root, struct inode *inode)
3919 {
3920 int ret;
3921
3922 /*
3923 * If the inode is a free space inode, we can deadlock during commit
3924 * if we put it into the delayed code.
3925 *
3926 * The data relocation inode should also be directly updated
3927 * without delay
3928 */
3929 if (!btrfs_is_free_space_inode(inode)
3930 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3931 && !root->fs_info->log_root_recovering) {
3932 btrfs_update_root_times(trans, root);
3933
3934 ret = btrfs_delayed_update_inode(trans, root, inode);
3935 if (!ret)
3936 btrfs_set_inode_last_trans(trans, inode);
3937 return ret;
3938 }
3939
3940 return btrfs_update_inode_item(trans, root, inode);
3941 }
3942
3943 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3944 struct btrfs_root *root,
3945 struct inode *inode)
3946 {
3947 int ret;
3948
3949 ret = btrfs_update_inode(trans, root, inode);
3950 if (ret == -ENOSPC)
3951 return btrfs_update_inode_item(trans, root, inode);
3952 return ret;
3953 }
3954
3955 /*
3956 * unlink helper that gets used here in inode.c and in the tree logging
3957 * recovery code. It remove a link in a directory with a given name, and
3958 * also drops the back refs in the inode to the directory
3959 */
3960 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3961 struct btrfs_root *root,
3962 struct inode *dir, struct inode *inode,
3963 const char *name, int name_len)
3964 {
3965 struct btrfs_path *path;
3966 int ret = 0;
3967 struct extent_buffer *leaf;
3968 struct btrfs_dir_item *di;
3969 struct btrfs_key key;
3970 u64 index;
3971 u64 ino = btrfs_ino(inode);
3972 u64 dir_ino = btrfs_ino(dir);
3973
3974 path = btrfs_alloc_path();
3975 if (!path) {
3976 ret = -ENOMEM;
3977 goto out;
3978 }
3979
3980 path->leave_spinning = 1;
3981 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3982 name, name_len, -1);
3983 if (IS_ERR(di)) {
3984 ret = PTR_ERR(di);
3985 goto err;
3986 }
3987 if (!di) {
3988 ret = -ENOENT;
3989 goto err;
3990 }
3991 leaf = path->nodes[0];
3992 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3993 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3994 if (ret)
3995 goto err;
3996 btrfs_release_path(path);
3997
3998 /*
3999 * If we don't have dir index, we have to get it by looking up
4000 * the inode ref, since we get the inode ref, remove it directly,
4001 * it is unnecessary to do delayed deletion.
4002 *
4003 * But if we have dir index, needn't search inode ref to get it.
4004 * Since the inode ref is close to the inode item, it is better
4005 * that we delay to delete it, and just do this deletion when
4006 * we update the inode item.
4007 */
4008 if (BTRFS_I(inode)->dir_index) {
4009 ret = btrfs_delayed_delete_inode_ref(inode);
4010 if (!ret) {
4011 index = BTRFS_I(inode)->dir_index;
4012 goto skip_backref;
4013 }
4014 }
4015
4016 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4017 dir_ino, &index);
4018 if (ret) {
4019 btrfs_info(root->fs_info,
4020 "failed to delete reference to %.*s, inode %llu parent %llu",
4021 name_len, name, ino, dir_ino);
4022 btrfs_abort_transaction(trans, root, ret);
4023 goto err;
4024 }
4025 skip_backref:
4026 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4027 if (ret) {
4028 btrfs_abort_transaction(trans, root, ret);
4029 goto err;
4030 }
4031
4032 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4033 inode, dir_ino);
4034 if (ret != 0 && ret != -ENOENT) {
4035 btrfs_abort_transaction(trans, root, ret);
4036 goto err;
4037 }
4038
4039 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4040 dir, index);
4041 if (ret == -ENOENT)
4042 ret = 0;
4043 else if (ret)
4044 btrfs_abort_transaction(trans, root, ret);
4045 err:
4046 btrfs_free_path(path);
4047 if (ret)
4048 goto out;
4049
4050 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4051 inode_inc_iversion(inode);
4052 inode_inc_iversion(dir);
4053 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4054 ret = btrfs_update_inode(trans, root, dir);
4055 out:
4056 return ret;
4057 }
4058
4059 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4060 struct btrfs_root *root,
4061 struct inode *dir, struct inode *inode,
4062 const char *name, int name_len)
4063 {
4064 int ret;
4065 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4066 if (!ret) {
4067 drop_nlink(inode);
4068 ret = btrfs_update_inode(trans, root, inode);
4069 }
4070 return ret;
4071 }
4072
4073 /*
4074 * helper to start transaction for unlink and rmdir.
4075 *
4076 * unlink and rmdir are special in btrfs, they do not always free space, so
4077 * if we cannot make our reservations the normal way try and see if there is
4078 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4079 * allow the unlink to occur.
4080 */
4081 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4082 {
4083 struct btrfs_root *root = BTRFS_I(dir)->root;
4084
4085 /*
4086 * 1 for the possible orphan item
4087 * 1 for the dir item
4088 * 1 for the dir index
4089 * 1 for the inode ref
4090 * 1 for the inode
4091 */
4092 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4093 }
4094
4095 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4096 {
4097 struct btrfs_root *root = BTRFS_I(dir)->root;
4098 struct btrfs_trans_handle *trans;
4099 struct inode *inode = d_inode(dentry);
4100 int ret;
4101
4102 trans = __unlink_start_trans(dir);
4103 if (IS_ERR(trans))
4104 return PTR_ERR(trans);
4105
4106 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4107
4108 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4109 dentry->d_name.name, dentry->d_name.len);
4110 if (ret)
4111 goto out;
4112
4113 if (inode->i_nlink == 0) {
4114 ret = btrfs_orphan_add(trans, inode);
4115 if (ret)
4116 goto out;
4117 }
4118
4119 out:
4120 btrfs_end_transaction(trans, root);
4121 btrfs_btree_balance_dirty(root);
4122 return ret;
4123 }
4124
4125 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4126 struct btrfs_root *root,
4127 struct inode *dir, u64 objectid,
4128 const char *name, int name_len)
4129 {
4130 struct btrfs_path *path;
4131 struct extent_buffer *leaf;
4132 struct btrfs_dir_item *di;
4133 struct btrfs_key key;
4134 u64 index;
4135 int ret;
4136 u64 dir_ino = btrfs_ino(dir);
4137
4138 path = btrfs_alloc_path();
4139 if (!path)
4140 return -ENOMEM;
4141
4142 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4143 name, name_len, -1);
4144 if (IS_ERR_OR_NULL(di)) {
4145 if (!di)
4146 ret = -ENOENT;
4147 else
4148 ret = PTR_ERR(di);
4149 goto out;
4150 }
4151
4152 leaf = path->nodes[0];
4153 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4154 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4155 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4156 if (ret) {
4157 btrfs_abort_transaction(trans, root, ret);
4158 goto out;
4159 }
4160 btrfs_release_path(path);
4161
4162 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4163 objectid, root->root_key.objectid,
4164 dir_ino, &index, name, name_len);
4165 if (ret < 0) {
4166 if (ret != -ENOENT) {
4167 btrfs_abort_transaction(trans, root, ret);
4168 goto out;
4169 }
4170 di = btrfs_search_dir_index_item(root, path, dir_ino,
4171 name, name_len);
4172 if (IS_ERR_OR_NULL(di)) {
4173 if (!di)
4174 ret = -ENOENT;
4175 else
4176 ret = PTR_ERR(di);
4177 btrfs_abort_transaction(trans, root, ret);
4178 goto out;
4179 }
4180
4181 leaf = path->nodes[0];
4182 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4183 btrfs_release_path(path);
4184 index = key.offset;
4185 }
4186 btrfs_release_path(path);
4187
4188 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4189 if (ret) {
4190 btrfs_abort_transaction(trans, root, ret);
4191 goto out;
4192 }
4193
4194 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4195 inode_inc_iversion(dir);
4196 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4197 ret = btrfs_update_inode_fallback(trans, root, dir);
4198 if (ret)
4199 btrfs_abort_transaction(trans, root, ret);
4200 out:
4201 btrfs_free_path(path);
4202 return ret;
4203 }
4204
4205 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4206 {
4207 struct inode *inode = d_inode(dentry);
4208 int err = 0;
4209 struct btrfs_root *root = BTRFS_I(dir)->root;
4210 struct btrfs_trans_handle *trans;
4211
4212 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4213 return -ENOTEMPTY;
4214 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4215 return -EPERM;
4216
4217 trans = __unlink_start_trans(dir);
4218 if (IS_ERR(trans))
4219 return PTR_ERR(trans);
4220
4221 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4222 err = btrfs_unlink_subvol(trans, root, dir,
4223 BTRFS_I(inode)->location.objectid,
4224 dentry->d_name.name,
4225 dentry->d_name.len);
4226 goto out;
4227 }
4228
4229 err = btrfs_orphan_add(trans, inode);
4230 if (err)
4231 goto out;
4232
4233 /* now the directory is empty */
4234 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4235 dentry->d_name.name, dentry->d_name.len);
4236 if (!err)
4237 btrfs_i_size_write(inode, 0);
4238 out:
4239 btrfs_end_transaction(trans, root);
4240 btrfs_btree_balance_dirty(root);
4241
4242 return err;
4243 }
4244
4245 static int truncate_space_check(struct btrfs_trans_handle *trans,
4246 struct btrfs_root *root,
4247 u64 bytes_deleted)
4248 {
4249 int ret;
4250
4251 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4252 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4253 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4254 if (!ret)
4255 trans->bytes_reserved += bytes_deleted;
4256 return ret;
4257
4258 }
4259
4260 static int truncate_inline_extent(struct inode *inode,
4261 struct btrfs_path *path,
4262 struct btrfs_key *found_key,
4263 const u64 item_end,
4264 const u64 new_size)
4265 {
4266 struct extent_buffer *leaf = path->nodes[0];
4267 int slot = path->slots[0];
4268 struct btrfs_file_extent_item *fi;
4269 u32 size = (u32)(new_size - found_key->offset);
4270 struct btrfs_root *root = BTRFS_I(inode)->root;
4271
4272 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4273
4274 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4275 loff_t offset = new_size;
4276 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4277
4278 /*
4279 * Zero out the remaining of the last page of our inline extent,
4280 * instead of directly truncating our inline extent here - that
4281 * would be much more complex (decompressing all the data, then
4282 * compressing the truncated data, which might be bigger than
4283 * the size of the inline extent, resize the extent, etc).
4284 * We release the path because to get the page we might need to
4285 * read the extent item from disk (data not in the page cache).
4286 */
4287 btrfs_release_path(path);
4288 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4289 }
4290
4291 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4292 size = btrfs_file_extent_calc_inline_size(size);
4293 btrfs_truncate_item(root, path, size, 1);
4294
4295 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4296 inode_sub_bytes(inode, item_end + 1 - new_size);
4297
4298 return 0;
4299 }
4300
4301 /*
4302 * this can truncate away extent items, csum items and directory items.
4303 * It starts at a high offset and removes keys until it can't find
4304 * any higher than new_size
4305 *
4306 * csum items that cross the new i_size are truncated to the new size
4307 * as well.
4308 *
4309 * min_type is the minimum key type to truncate down to. If set to 0, this
4310 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4311 */
4312 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4313 struct btrfs_root *root,
4314 struct inode *inode,
4315 u64 new_size, u32 min_type)
4316 {
4317 struct btrfs_path *path;
4318 struct extent_buffer *leaf;
4319 struct btrfs_file_extent_item *fi;
4320 struct btrfs_key key;
4321 struct btrfs_key found_key;
4322 u64 extent_start = 0;
4323 u64 extent_num_bytes = 0;
4324 u64 extent_offset = 0;
4325 u64 item_end = 0;
4326 u64 last_size = new_size;
4327 u32 found_type = (u8)-1;
4328 int found_extent;
4329 int del_item;
4330 int pending_del_nr = 0;
4331 int pending_del_slot = 0;
4332 int extent_type = -1;
4333 int ret;
4334 int err = 0;
4335 u64 ino = btrfs_ino(inode);
4336 u64 bytes_deleted = 0;
4337 bool be_nice = 0;
4338 bool should_throttle = 0;
4339 bool should_end = 0;
4340
4341 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4342
4343 /*
4344 * for non-free space inodes and ref cows, we want to back off from
4345 * time to time
4346 */
4347 if (!btrfs_is_free_space_inode(inode) &&
4348 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4349 be_nice = 1;
4350
4351 path = btrfs_alloc_path();
4352 if (!path)
4353 return -ENOMEM;
4354 path->reada = -1;
4355
4356 /*
4357 * We want to drop from the next block forward in case this new size is
4358 * not block aligned since we will be keeping the last block of the
4359 * extent just the way it is.
4360 */
4361 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4362 root == root->fs_info->tree_root)
4363 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4364 root->sectorsize), (u64)-1, 0);
4365
4366 /*
4367 * This function is also used to drop the items in the log tree before
4368 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4369 * it is used to drop the loged items. So we shouldn't kill the delayed
4370 * items.
4371 */
4372 if (min_type == 0 && root == BTRFS_I(inode)->root)
4373 btrfs_kill_delayed_inode_items(inode);
4374
4375 key.objectid = ino;
4376 key.offset = (u64)-1;
4377 key.type = (u8)-1;
4378
4379 search_again:
4380 /*
4381 * with a 16K leaf size and 128MB extents, you can actually queue
4382 * up a huge file in a single leaf. Most of the time that
4383 * bytes_deleted is > 0, it will be huge by the time we get here
4384 */
4385 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4386 if (btrfs_should_end_transaction(trans, root)) {
4387 err = -EAGAIN;
4388 goto error;
4389 }
4390 }
4391
4392
4393 path->leave_spinning = 1;
4394 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4395 if (ret < 0) {
4396 err = ret;
4397 goto out;
4398 }
4399
4400 if (ret > 0) {
4401 /* there are no items in the tree for us to truncate, we're
4402 * done
4403 */
4404 if (path->slots[0] == 0)
4405 goto out;
4406 path->slots[0]--;
4407 }
4408
4409 while (1) {
4410 fi = NULL;
4411 leaf = path->nodes[0];
4412 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4413 found_type = found_key.type;
4414
4415 if (found_key.objectid != ino)
4416 break;
4417
4418 if (found_type < min_type)
4419 break;
4420
4421 item_end = found_key.offset;
4422 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4423 fi = btrfs_item_ptr(leaf, path->slots[0],
4424 struct btrfs_file_extent_item);
4425 extent_type = btrfs_file_extent_type(leaf, fi);
4426 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4427 item_end +=
4428 btrfs_file_extent_num_bytes(leaf, fi);
4429 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4430 item_end += btrfs_file_extent_inline_len(leaf,
4431 path->slots[0], fi);
4432 }
4433 item_end--;
4434 }
4435 if (found_type > min_type) {
4436 del_item = 1;
4437 } else {
4438 if (item_end < new_size) {
4439 /*
4440 * With NO_HOLES mode, for the following mapping
4441 *
4442 * [0-4k][hole][8k-12k]
4443 *
4444 * if truncating isize down to 6k, it ends up
4445 * isize being 8k.
4446 */
4447 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
4448 last_size = new_size;
4449 break;
4450 }
4451 if (found_key.offset >= new_size)
4452 del_item = 1;
4453 else
4454 del_item = 0;
4455 }
4456 found_extent = 0;
4457 /* FIXME, shrink the extent if the ref count is only 1 */
4458 if (found_type != BTRFS_EXTENT_DATA_KEY)
4459 goto delete;
4460
4461 if (del_item)
4462 last_size = found_key.offset;
4463 else
4464 last_size = new_size;
4465
4466 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4467 u64 num_dec;
4468 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4469 if (!del_item) {
4470 u64 orig_num_bytes =
4471 btrfs_file_extent_num_bytes(leaf, fi);
4472 extent_num_bytes = ALIGN(new_size -
4473 found_key.offset,
4474 root->sectorsize);
4475 btrfs_set_file_extent_num_bytes(leaf, fi,
4476 extent_num_bytes);
4477 num_dec = (orig_num_bytes -
4478 extent_num_bytes);
4479 if (test_bit(BTRFS_ROOT_REF_COWS,
4480 &root->state) &&
4481 extent_start != 0)
4482 inode_sub_bytes(inode, num_dec);
4483 btrfs_mark_buffer_dirty(leaf);
4484 } else {
4485 extent_num_bytes =
4486 btrfs_file_extent_disk_num_bytes(leaf,
4487 fi);
4488 extent_offset = found_key.offset -
4489 btrfs_file_extent_offset(leaf, fi);
4490
4491 /* FIXME blocksize != 4096 */
4492 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4493 if (extent_start != 0) {
4494 found_extent = 1;
4495 if (test_bit(BTRFS_ROOT_REF_COWS,
4496 &root->state))
4497 inode_sub_bytes(inode, num_dec);
4498 }
4499 }
4500 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4501 /*
4502 * we can't truncate inline items that have had
4503 * special encodings
4504 */
4505 if (!del_item &&
4506 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4507 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4508
4509 /*
4510 * Need to release path in order to truncate a
4511 * compressed extent. So delete any accumulated
4512 * extent items so far.
4513 */
4514 if (btrfs_file_extent_compression(leaf, fi) !=
4515 BTRFS_COMPRESS_NONE && pending_del_nr) {
4516 err = btrfs_del_items(trans, root, path,
4517 pending_del_slot,
4518 pending_del_nr);
4519 if (err) {
4520 btrfs_abort_transaction(trans,
4521 root,
4522 err);
4523 goto error;
4524 }
4525 pending_del_nr = 0;
4526 }
4527
4528 err = truncate_inline_extent(inode, path,
4529 &found_key,
4530 item_end,
4531 new_size);
4532 if (err) {
4533 btrfs_abort_transaction(trans,
4534 root, err);
4535 goto error;
4536 }
4537 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4538 &root->state)) {
4539 inode_sub_bytes(inode, item_end + 1 - new_size);
4540 }
4541 }
4542 delete:
4543 if (del_item) {
4544 if (!pending_del_nr) {
4545 /* no pending yet, add ourselves */
4546 pending_del_slot = path->slots[0];
4547 pending_del_nr = 1;
4548 } else if (pending_del_nr &&
4549 path->slots[0] + 1 == pending_del_slot) {
4550 /* hop on the pending chunk */
4551 pending_del_nr++;
4552 pending_del_slot = path->slots[0];
4553 } else {
4554 BUG();
4555 }
4556 } else {
4557 break;
4558 }
4559 should_throttle = 0;
4560
4561 if (found_extent &&
4562 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4563 root == root->fs_info->tree_root)) {
4564 btrfs_set_path_blocking(path);
4565 bytes_deleted += extent_num_bytes;
4566 ret = btrfs_free_extent(trans, root, extent_start,
4567 extent_num_bytes, 0,
4568 btrfs_header_owner(leaf),
4569 ino, extent_offset);
4570 BUG_ON(ret);
4571 if (btrfs_should_throttle_delayed_refs(trans, root))
4572 btrfs_async_run_delayed_refs(root,
4573 trans->delayed_ref_updates * 2, 0);
4574 if (be_nice) {
4575 if (truncate_space_check(trans, root,
4576 extent_num_bytes)) {
4577 should_end = 1;
4578 }
4579 if (btrfs_should_throttle_delayed_refs(trans,
4580 root)) {
4581 should_throttle = 1;
4582 }
4583 }
4584 }
4585
4586 if (found_type == BTRFS_INODE_ITEM_KEY)
4587 break;
4588
4589 if (path->slots[0] == 0 ||
4590 path->slots[0] != pending_del_slot ||
4591 should_throttle || should_end) {
4592 if (pending_del_nr) {
4593 ret = btrfs_del_items(trans, root, path,
4594 pending_del_slot,
4595 pending_del_nr);
4596 if (ret) {
4597 btrfs_abort_transaction(trans,
4598 root, ret);
4599 goto error;
4600 }
4601 pending_del_nr = 0;
4602 }
4603 btrfs_release_path(path);
4604 if (should_throttle) {
4605 unsigned long updates = trans->delayed_ref_updates;
4606 if (updates) {
4607 trans->delayed_ref_updates = 0;
4608 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4609 if (ret && !err)
4610 err = ret;
4611 }
4612 }
4613 /*
4614 * if we failed to refill our space rsv, bail out
4615 * and let the transaction restart
4616 */
4617 if (should_end) {
4618 err = -EAGAIN;
4619 goto error;
4620 }
4621 goto search_again;
4622 } else {
4623 path->slots[0]--;
4624 }
4625 }
4626 out:
4627 if (pending_del_nr) {
4628 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4629 pending_del_nr);
4630 if (ret)
4631 btrfs_abort_transaction(trans, root, ret);
4632 }
4633 error:
4634 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4635 btrfs_ordered_update_i_size(inode, last_size, NULL);
4636
4637 btrfs_free_path(path);
4638
4639 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4640 unsigned long updates = trans->delayed_ref_updates;
4641 if (updates) {
4642 trans->delayed_ref_updates = 0;
4643 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4644 if (ret && !err)
4645 err = ret;
4646 }
4647 }
4648 return err;
4649 }
4650
4651 /*
4652 * btrfs_truncate_page - read, zero a chunk and write a page
4653 * @inode - inode that we're zeroing
4654 * @from - the offset to start zeroing
4655 * @len - the length to zero, 0 to zero the entire range respective to the
4656 * offset
4657 * @front - zero up to the offset instead of from the offset on
4658 *
4659 * This will find the page for the "from" offset and cow the page and zero the
4660 * part we want to zero. This is used with truncate and hole punching.
4661 */
4662 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4663 int front)
4664 {
4665 struct address_space *mapping = inode->i_mapping;
4666 struct btrfs_root *root = BTRFS_I(inode)->root;
4667 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4668 struct btrfs_ordered_extent *ordered;
4669 struct extent_state *cached_state = NULL;
4670 char *kaddr;
4671 u32 blocksize = root->sectorsize;
4672 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4673 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4674 struct page *page;
4675 gfp_t mask = btrfs_alloc_write_mask(mapping);
4676 int ret = 0;
4677 u64 page_start;
4678 u64 page_end;
4679
4680 if ((offset & (blocksize - 1)) == 0 &&
4681 (!len || ((len & (blocksize - 1)) == 0)))
4682 goto out;
4683 ret = btrfs_delalloc_reserve_space(inode,
4684 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4685 if (ret)
4686 goto out;
4687
4688 again:
4689 page = find_or_create_page(mapping, index, mask);
4690 if (!page) {
4691 btrfs_delalloc_release_space(inode,
4692 round_down(from, PAGE_CACHE_SIZE),
4693 PAGE_CACHE_SIZE);
4694 ret = -ENOMEM;
4695 goto out;
4696 }
4697
4698 page_start = page_offset(page);
4699 page_end = page_start + PAGE_CACHE_SIZE - 1;
4700
4701 if (!PageUptodate(page)) {
4702 ret = btrfs_readpage(NULL, page);
4703 lock_page(page);
4704 if (page->mapping != mapping) {
4705 unlock_page(page);
4706 page_cache_release(page);
4707 goto again;
4708 }
4709 if (!PageUptodate(page)) {
4710 ret = -EIO;
4711 goto out_unlock;
4712 }
4713 }
4714 wait_on_page_writeback(page);
4715
4716 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4717 set_page_extent_mapped(page);
4718
4719 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4720 if (ordered) {
4721 unlock_extent_cached(io_tree, page_start, page_end,
4722 &cached_state, GFP_NOFS);
4723 unlock_page(page);
4724 page_cache_release(page);
4725 btrfs_start_ordered_extent(inode, ordered, 1);
4726 btrfs_put_ordered_extent(ordered);
4727 goto again;
4728 }
4729
4730 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4731 EXTENT_DIRTY | EXTENT_DELALLOC |
4732 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4733 0, 0, &cached_state, GFP_NOFS);
4734
4735 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4736 &cached_state);
4737 if (ret) {
4738 unlock_extent_cached(io_tree, page_start, page_end,
4739 &cached_state, GFP_NOFS);
4740 goto out_unlock;
4741 }
4742
4743 if (offset != PAGE_CACHE_SIZE) {
4744 if (!len)
4745 len = PAGE_CACHE_SIZE - offset;
4746 kaddr = kmap(page);
4747 if (front)
4748 memset(kaddr, 0, offset);
4749 else
4750 memset(kaddr + offset, 0, len);
4751 flush_dcache_page(page);
4752 kunmap(page);
4753 }
4754 ClearPageChecked(page);
4755 set_page_dirty(page);
4756 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4757 GFP_NOFS);
4758
4759 out_unlock:
4760 if (ret)
4761 btrfs_delalloc_release_space(inode, page_start,
4762 PAGE_CACHE_SIZE);
4763 unlock_page(page);
4764 page_cache_release(page);
4765 out:
4766 return ret;
4767 }
4768
4769 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4770 u64 offset, u64 len)
4771 {
4772 struct btrfs_trans_handle *trans;
4773 int ret;
4774
4775 /*
4776 * Still need to make sure the inode looks like it's been updated so
4777 * that any holes get logged if we fsync.
4778 */
4779 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4780 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4781 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4782 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4783 return 0;
4784 }
4785
4786 /*
4787 * 1 - for the one we're dropping
4788 * 1 - for the one we're adding
4789 * 1 - for updating the inode.
4790 */
4791 trans = btrfs_start_transaction(root, 3);
4792 if (IS_ERR(trans))
4793 return PTR_ERR(trans);
4794
4795 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4796 if (ret) {
4797 btrfs_abort_transaction(trans, root, ret);
4798 btrfs_end_transaction(trans, root);
4799 return ret;
4800 }
4801
4802 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4803 0, 0, len, 0, len, 0, 0, 0);
4804 if (ret)
4805 btrfs_abort_transaction(trans, root, ret);
4806 else
4807 btrfs_update_inode(trans, root, inode);
4808 btrfs_end_transaction(trans, root);
4809 return ret;
4810 }
4811
4812 /*
4813 * This function puts in dummy file extents for the area we're creating a hole
4814 * for. So if we are truncating this file to a larger size we need to insert
4815 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4816 * the range between oldsize and size
4817 */
4818 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4819 {
4820 struct btrfs_root *root = BTRFS_I(inode)->root;
4821 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4822 struct extent_map *em = NULL;
4823 struct extent_state *cached_state = NULL;
4824 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4825 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4826 u64 block_end = ALIGN(size, root->sectorsize);
4827 u64 last_byte;
4828 u64 cur_offset;
4829 u64 hole_size;
4830 int err = 0;
4831
4832 /*
4833 * If our size started in the middle of a page we need to zero out the
4834 * rest of the page before we expand the i_size, otherwise we could
4835 * expose stale data.
4836 */
4837 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4838 if (err)
4839 return err;
4840
4841 if (size <= hole_start)
4842 return 0;
4843
4844 while (1) {
4845 struct btrfs_ordered_extent *ordered;
4846
4847 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4848 &cached_state);
4849 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4850 block_end - hole_start);
4851 if (!ordered)
4852 break;
4853 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4854 &cached_state, GFP_NOFS);
4855 btrfs_start_ordered_extent(inode, ordered, 1);
4856 btrfs_put_ordered_extent(ordered);
4857 }
4858
4859 cur_offset = hole_start;
4860 while (1) {
4861 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4862 block_end - cur_offset, 0);
4863 if (IS_ERR(em)) {
4864 err = PTR_ERR(em);
4865 em = NULL;
4866 break;
4867 }
4868 last_byte = min(extent_map_end(em), block_end);
4869 last_byte = ALIGN(last_byte , root->sectorsize);
4870 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4871 struct extent_map *hole_em;
4872 hole_size = last_byte - cur_offset;
4873
4874 err = maybe_insert_hole(root, inode, cur_offset,
4875 hole_size);
4876 if (err)
4877 break;
4878 btrfs_drop_extent_cache(inode, cur_offset,
4879 cur_offset + hole_size - 1, 0);
4880 hole_em = alloc_extent_map();
4881 if (!hole_em) {
4882 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4883 &BTRFS_I(inode)->runtime_flags);
4884 goto next;
4885 }
4886 hole_em->start = cur_offset;
4887 hole_em->len = hole_size;
4888 hole_em->orig_start = cur_offset;
4889
4890 hole_em->block_start = EXTENT_MAP_HOLE;
4891 hole_em->block_len = 0;
4892 hole_em->orig_block_len = 0;
4893 hole_em->ram_bytes = hole_size;
4894 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4895 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4896 hole_em->generation = root->fs_info->generation;
4897
4898 while (1) {
4899 write_lock(&em_tree->lock);
4900 err = add_extent_mapping(em_tree, hole_em, 1);
4901 write_unlock(&em_tree->lock);
4902 if (err != -EEXIST)
4903 break;
4904 btrfs_drop_extent_cache(inode, cur_offset,
4905 cur_offset +
4906 hole_size - 1, 0);
4907 }
4908 free_extent_map(hole_em);
4909 }
4910 next:
4911 free_extent_map(em);
4912 em = NULL;
4913 cur_offset = last_byte;
4914 if (cur_offset >= block_end)
4915 break;
4916 }
4917 free_extent_map(em);
4918 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4919 GFP_NOFS);
4920 return err;
4921 }
4922
4923 static int wait_snapshoting_atomic_t(atomic_t *a)
4924 {
4925 schedule();
4926 return 0;
4927 }
4928
4929 static void wait_for_snapshot_creation(struct btrfs_root *root)
4930 {
4931 while (true) {
4932 int ret;
4933
4934 ret = btrfs_start_write_no_snapshoting(root);
4935 if (ret)
4936 break;
4937 wait_on_atomic_t(&root->will_be_snapshoted,
4938 wait_snapshoting_atomic_t,
4939 TASK_UNINTERRUPTIBLE);
4940 }
4941 }
4942
4943 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4944 {
4945 struct btrfs_root *root = BTRFS_I(inode)->root;
4946 struct btrfs_trans_handle *trans;
4947 loff_t oldsize = i_size_read(inode);
4948 loff_t newsize = attr->ia_size;
4949 int mask = attr->ia_valid;
4950 int ret;
4951
4952 /*
4953 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4954 * special case where we need to update the times despite not having
4955 * these flags set. For all other operations the VFS set these flags
4956 * explicitly if it wants a timestamp update.
4957 */
4958 if (newsize != oldsize) {
4959 inode_inc_iversion(inode);
4960 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4961 inode->i_ctime = inode->i_mtime =
4962 current_fs_time(inode->i_sb);
4963 }
4964
4965 if (newsize > oldsize) {
4966 truncate_pagecache(inode, newsize);
4967 /*
4968 * Don't do an expanding truncate while snapshoting is ongoing.
4969 * This is to ensure the snapshot captures a fully consistent
4970 * state of this file - if the snapshot captures this expanding
4971 * truncation, it must capture all writes that happened before
4972 * this truncation.
4973 */
4974 wait_for_snapshot_creation(root);
4975 ret = btrfs_cont_expand(inode, oldsize, newsize);
4976 if (ret) {
4977 btrfs_end_write_no_snapshoting(root);
4978 return ret;
4979 }
4980
4981 trans = btrfs_start_transaction(root, 1);
4982 if (IS_ERR(trans)) {
4983 btrfs_end_write_no_snapshoting(root);
4984 return PTR_ERR(trans);
4985 }
4986
4987 i_size_write(inode, newsize);
4988 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4989 ret = btrfs_update_inode(trans, root, inode);
4990 btrfs_end_write_no_snapshoting(root);
4991 btrfs_end_transaction(trans, root);
4992 } else {
4993
4994 /*
4995 * We're truncating a file that used to have good data down to
4996 * zero. Make sure it gets into the ordered flush list so that
4997 * any new writes get down to disk quickly.
4998 */
4999 if (newsize == 0)
5000 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5001 &BTRFS_I(inode)->runtime_flags);
5002
5003 /*
5004 * 1 for the orphan item we're going to add
5005 * 1 for the orphan item deletion.
5006 */
5007 trans = btrfs_start_transaction(root, 2);
5008 if (IS_ERR(trans))
5009 return PTR_ERR(trans);
5010
5011 /*
5012 * We need to do this in case we fail at _any_ point during the
5013 * actual truncate. Once we do the truncate_setsize we could
5014 * invalidate pages which forces any outstanding ordered io to
5015 * be instantly completed which will give us extents that need
5016 * to be truncated. If we fail to get an orphan inode down we
5017 * could have left over extents that were never meant to live,
5018 * so we need to garuntee from this point on that everything
5019 * will be consistent.
5020 */
5021 ret = btrfs_orphan_add(trans, inode);
5022 btrfs_end_transaction(trans, root);
5023 if (ret)
5024 return ret;
5025
5026 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5027 truncate_setsize(inode, newsize);
5028
5029 /* Disable nonlocked read DIO to avoid the end less truncate */
5030 btrfs_inode_block_unlocked_dio(inode);
5031 inode_dio_wait(inode);
5032 btrfs_inode_resume_unlocked_dio(inode);
5033
5034 ret = btrfs_truncate(inode);
5035 if (ret && inode->i_nlink) {
5036 int err;
5037
5038 /*
5039 * failed to truncate, disk_i_size is only adjusted down
5040 * as we remove extents, so it should represent the true
5041 * size of the inode, so reset the in memory size and
5042 * delete our orphan entry.
5043 */
5044 trans = btrfs_join_transaction(root);
5045 if (IS_ERR(trans)) {
5046 btrfs_orphan_del(NULL, inode);
5047 return ret;
5048 }
5049 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5050 err = btrfs_orphan_del(trans, inode);
5051 if (err)
5052 btrfs_abort_transaction(trans, root, err);
5053 btrfs_end_transaction(trans, root);
5054 }
5055 }
5056
5057 return ret;
5058 }
5059
5060 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5061 {
5062 struct inode *inode = d_inode(dentry);
5063 struct btrfs_root *root = BTRFS_I(inode)->root;
5064 int err;
5065
5066 if (btrfs_root_readonly(root))
5067 return -EROFS;
5068
5069 err = inode_change_ok(inode, attr);
5070 if (err)
5071 return err;
5072
5073 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5074 err = btrfs_setsize(inode, attr);
5075 if (err)
5076 return err;
5077 }
5078
5079 if (attr->ia_valid) {
5080 setattr_copy(inode, attr);
5081 inode_inc_iversion(inode);
5082 err = btrfs_dirty_inode(inode);
5083
5084 if (!err && attr->ia_valid & ATTR_MODE)
5085 err = posix_acl_chmod(inode, inode->i_mode);
5086 }
5087
5088 return err;
5089 }
5090
5091 /*
5092 * While truncating the inode pages during eviction, we get the VFS calling
5093 * btrfs_invalidatepage() against each page of the inode. This is slow because
5094 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5095 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5096 * extent_state structures over and over, wasting lots of time.
5097 *
5098 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5099 * those expensive operations on a per page basis and do only the ordered io
5100 * finishing, while we release here the extent_map and extent_state structures,
5101 * without the excessive merging and splitting.
5102 */
5103 static void evict_inode_truncate_pages(struct inode *inode)
5104 {
5105 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5106 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5107 struct rb_node *node;
5108
5109 ASSERT(inode->i_state & I_FREEING);
5110 truncate_inode_pages_final(&inode->i_data);
5111
5112 write_lock(&map_tree->lock);
5113 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5114 struct extent_map *em;
5115
5116 node = rb_first(&map_tree->map);
5117 em = rb_entry(node, struct extent_map, rb_node);
5118 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5119 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5120 remove_extent_mapping(map_tree, em);
5121 free_extent_map(em);
5122 if (need_resched()) {
5123 write_unlock(&map_tree->lock);
5124 cond_resched();
5125 write_lock(&map_tree->lock);
5126 }
5127 }
5128 write_unlock(&map_tree->lock);
5129
5130 /*
5131 * Keep looping until we have no more ranges in the io tree.
5132 * We can have ongoing bios started by readpages (called from readahead)
5133 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5134 * still in progress (unlocked the pages in the bio but did not yet
5135 * unlocked the ranges in the io tree). Therefore this means some
5136 * ranges can still be locked and eviction started because before
5137 * submitting those bios, which are executed by a separate task (work
5138 * queue kthread), inode references (inode->i_count) were not taken
5139 * (which would be dropped in the end io callback of each bio).
5140 * Therefore here we effectively end up waiting for those bios and
5141 * anyone else holding locked ranges without having bumped the inode's
5142 * reference count - if we don't do it, when they access the inode's
5143 * io_tree to unlock a range it may be too late, leading to an
5144 * use-after-free issue.
5145 */
5146 spin_lock(&io_tree->lock);
5147 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5148 struct extent_state *state;
5149 struct extent_state *cached_state = NULL;
5150 u64 start;
5151 u64 end;
5152
5153 node = rb_first(&io_tree->state);
5154 state = rb_entry(node, struct extent_state, rb_node);
5155 start = state->start;
5156 end = state->end;
5157 spin_unlock(&io_tree->lock);
5158
5159 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5160
5161 /*
5162 * If still has DELALLOC flag, the extent didn't reach disk,
5163 * and its reserved space won't be freed by delayed_ref.
5164 * So we need to free its reserved space here.
5165 * (Refer to comment in btrfs_invalidatepage, case 2)
5166 *
5167 * Note, end is the bytenr of last byte, so we need + 1 here.
5168 */
5169 if (state->state & EXTENT_DELALLOC)
5170 btrfs_qgroup_free_data(inode, start, end - start + 1);
5171
5172 clear_extent_bit(io_tree, start, end,
5173 EXTENT_LOCKED | EXTENT_DIRTY |
5174 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5175 EXTENT_DEFRAG, 1, 1,
5176 &cached_state, GFP_NOFS);
5177
5178 cond_resched();
5179 spin_lock(&io_tree->lock);
5180 }
5181 spin_unlock(&io_tree->lock);
5182 }
5183
5184 void btrfs_evict_inode(struct inode *inode)
5185 {
5186 struct btrfs_trans_handle *trans;
5187 struct btrfs_root *root = BTRFS_I(inode)->root;
5188 struct btrfs_block_rsv *rsv, *global_rsv;
5189 int steal_from_global = 0;
5190 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5191 int ret;
5192
5193 trace_btrfs_inode_evict(inode);
5194
5195 evict_inode_truncate_pages(inode);
5196
5197 if (inode->i_nlink &&
5198 ((btrfs_root_refs(&root->root_item) != 0 &&
5199 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5200 btrfs_is_free_space_inode(inode)))
5201 goto no_delete;
5202
5203 if (is_bad_inode(inode)) {
5204 btrfs_orphan_del(NULL, inode);
5205 goto no_delete;
5206 }
5207 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5208 if (!special_file(inode->i_mode))
5209 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5210
5211 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5212
5213 if (root->fs_info->log_root_recovering) {
5214 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5215 &BTRFS_I(inode)->runtime_flags));
5216 goto no_delete;
5217 }
5218
5219 if (inode->i_nlink > 0) {
5220 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5221 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5222 goto no_delete;
5223 }
5224
5225 ret = btrfs_commit_inode_delayed_inode(inode);
5226 if (ret) {
5227 btrfs_orphan_del(NULL, inode);
5228 goto no_delete;
5229 }
5230
5231 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5232 if (!rsv) {
5233 btrfs_orphan_del(NULL, inode);
5234 goto no_delete;
5235 }
5236 rsv->size = min_size;
5237 rsv->failfast = 1;
5238 global_rsv = &root->fs_info->global_block_rsv;
5239
5240 btrfs_i_size_write(inode, 0);
5241
5242 /*
5243 * This is a bit simpler than btrfs_truncate since we've already
5244 * reserved our space for our orphan item in the unlink, so we just
5245 * need to reserve some slack space in case we add bytes and update
5246 * inode item when doing the truncate.
5247 */
5248 while (1) {
5249 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5250 BTRFS_RESERVE_FLUSH_LIMIT);
5251
5252 /*
5253 * Try and steal from the global reserve since we will
5254 * likely not use this space anyway, we want to try as
5255 * hard as possible to get this to work.
5256 */
5257 if (ret)
5258 steal_from_global++;
5259 else
5260 steal_from_global = 0;
5261 ret = 0;
5262
5263 /*
5264 * steal_from_global == 0: we reserved stuff, hooray!
5265 * steal_from_global == 1: we didn't reserve stuff, boo!
5266 * steal_from_global == 2: we've committed, still not a lot of
5267 * room but maybe we'll have room in the global reserve this
5268 * time.
5269 * steal_from_global == 3: abandon all hope!
5270 */
5271 if (steal_from_global > 2) {
5272 btrfs_warn(root->fs_info,
5273 "Could not get space for a delete, will truncate on mount %d",
5274 ret);
5275 btrfs_orphan_del(NULL, inode);
5276 btrfs_free_block_rsv(root, rsv);
5277 goto no_delete;
5278 }
5279
5280 trans = btrfs_join_transaction(root);
5281 if (IS_ERR(trans)) {
5282 btrfs_orphan_del(NULL, inode);
5283 btrfs_free_block_rsv(root, rsv);
5284 goto no_delete;
5285 }
5286
5287 /*
5288 * We can't just steal from the global reserve, we need tomake
5289 * sure there is room to do it, if not we need to commit and try
5290 * again.
5291 */
5292 if (steal_from_global) {
5293 if (!btrfs_check_space_for_delayed_refs(trans, root))
5294 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5295 min_size);
5296 else
5297 ret = -ENOSPC;
5298 }
5299
5300 /*
5301 * Couldn't steal from the global reserve, we have too much
5302 * pending stuff built up, commit the transaction and try it
5303 * again.
5304 */
5305 if (ret) {
5306 ret = btrfs_commit_transaction(trans, root);
5307 if (ret) {
5308 btrfs_orphan_del(NULL, inode);
5309 btrfs_free_block_rsv(root, rsv);
5310 goto no_delete;
5311 }
5312 continue;
5313 } else {
5314 steal_from_global = 0;
5315 }
5316
5317 trans->block_rsv = rsv;
5318
5319 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5320 if (ret != -ENOSPC && ret != -EAGAIN)
5321 break;
5322
5323 trans->block_rsv = &root->fs_info->trans_block_rsv;
5324 btrfs_end_transaction(trans, root);
5325 trans = NULL;
5326 btrfs_btree_balance_dirty(root);
5327 }
5328
5329 btrfs_free_block_rsv(root, rsv);
5330
5331 /*
5332 * Errors here aren't a big deal, it just means we leave orphan items
5333 * in the tree. They will be cleaned up on the next mount.
5334 */
5335 if (ret == 0) {
5336 trans->block_rsv = root->orphan_block_rsv;
5337 btrfs_orphan_del(trans, inode);
5338 } else {
5339 btrfs_orphan_del(NULL, inode);
5340 }
5341
5342 trans->block_rsv = &root->fs_info->trans_block_rsv;
5343 if (!(root == root->fs_info->tree_root ||
5344 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5345 btrfs_return_ino(root, btrfs_ino(inode));
5346
5347 btrfs_end_transaction(trans, root);
5348 btrfs_btree_balance_dirty(root);
5349 no_delete:
5350 btrfs_remove_delayed_node(inode);
5351 clear_inode(inode);
5352 return;
5353 }
5354
5355 /*
5356 * this returns the key found in the dir entry in the location pointer.
5357 * If no dir entries were found, location->objectid is 0.
5358 */
5359 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5360 struct btrfs_key *location)
5361 {
5362 const char *name = dentry->d_name.name;
5363 int namelen = dentry->d_name.len;
5364 struct btrfs_dir_item *di;
5365 struct btrfs_path *path;
5366 struct btrfs_root *root = BTRFS_I(dir)->root;
5367 int ret = 0;
5368
5369 path = btrfs_alloc_path();
5370 if (!path)
5371 return -ENOMEM;
5372
5373 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5374 namelen, 0);
5375 if (IS_ERR(di))
5376 ret = PTR_ERR(di);
5377
5378 if (IS_ERR_OR_NULL(di))
5379 goto out_err;
5380
5381 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5382 out:
5383 btrfs_free_path(path);
5384 return ret;
5385 out_err:
5386 location->objectid = 0;
5387 goto out;
5388 }
5389
5390 /*
5391 * when we hit a tree root in a directory, the btrfs part of the inode
5392 * needs to be changed to reflect the root directory of the tree root. This
5393 * is kind of like crossing a mount point.
5394 */
5395 static int fixup_tree_root_location(struct btrfs_root *root,
5396 struct inode *dir,
5397 struct dentry *dentry,
5398 struct btrfs_key *location,
5399 struct btrfs_root **sub_root)
5400 {
5401 struct btrfs_path *path;
5402 struct btrfs_root *new_root;
5403 struct btrfs_root_ref *ref;
5404 struct extent_buffer *leaf;
5405 struct btrfs_key key;
5406 int ret;
5407 int err = 0;
5408
5409 path = btrfs_alloc_path();
5410 if (!path) {
5411 err = -ENOMEM;
5412 goto out;
5413 }
5414
5415 err = -ENOENT;
5416 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5417 key.type = BTRFS_ROOT_REF_KEY;
5418 key.offset = location->objectid;
5419
5420 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5421 0, 0);
5422 if (ret) {
5423 if (ret < 0)
5424 err = ret;
5425 goto out;
5426 }
5427
5428 leaf = path->nodes[0];
5429 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5430 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5431 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5432 goto out;
5433
5434 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5435 (unsigned long)(ref + 1),
5436 dentry->d_name.len);
5437 if (ret)
5438 goto out;
5439
5440 btrfs_release_path(path);
5441
5442 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5443 if (IS_ERR(new_root)) {
5444 err = PTR_ERR(new_root);
5445 goto out;
5446 }
5447
5448 *sub_root = new_root;
5449 location->objectid = btrfs_root_dirid(&new_root->root_item);
5450 location->type = BTRFS_INODE_ITEM_KEY;
5451 location->offset = 0;
5452 err = 0;
5453 out:
5454 btrfs_free_path(path);
5455 return err;
5456 }
5457
5458 static void inode_tree_add(struct inode *inode)
5459 {
5460 struct btrfs_root *root = BTRFS_I(inode)->root;
5461 struct btrfs_inode *entry;
5462 struct rb_node **p;
5463 struct rb_node *parent;
5464 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5465 u64 ino = btrfs_ino(inode);
5466
5467 if (inode_unhashed(inode))
5468 return;
5469 parent = NULL;
5470 spin_lock(&root->inode_lock);
5471 p = &root->inode_tree.rb_node;
5472 while (*p) {
5473 parent = *p;
5474 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5475
5476 if (ino < btrfs_ino(&entry->vfs_inode))
5477 p = &parent->rb_left;
5478 else if (ino > btrfs_ino(&entry->vfs_inode))
5479 p = &parent->rb_right;
5480 else {
5481 WARN_ON(!(entry->vfs_inode.i_state &
5482 (I_WILL_FREE | I_FREEING)));
5483 rb_replace_node(parent, new, &root->inode_tree);
5484 RB_CLEAR_NODE(parent);
5485 spin_unlock(&root->inode_lock);
5486 return;
5487 }
5488 }
5489 rb_link_node(new, parent, p);
5490 rb_insert_color(new, &root->inode_tree);
5491 spin_unlock(&root->inode_lock);
5492 }
5493
5494 static void inode_tree_del(struct inode *inode)
5495 {
5496 struct btrfs_root *root = BTRFS_I(inode)->root;
5497 int empty = 0;
5498
5499 spin_lock(&root->inode_lock);
5500 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5501 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5502 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5503 empty = RB_EMPTY_ROOT(&root->inode_tree);
5504 }
5505 spin_unlock(&root->inode_lock);
5506
5507 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5508 synchronize_srcu(&root->fs_info->subvol_srcu);
5509 spin_lock(&root->inode_lock);
5510 empty = RB_EMPTY_ROOT(&root->inode_tree);
5511 spin_unlock(&root->inode_lock);
5512 if (empty)
5513 btrfs_add_dead_root(root);
5514 }
5515 }
5516
5517 void btrfs_invalidate_inodes(struct btrfs_root *root)
5518 {
5519 struct rb_node *node;
5520 struct rb_node *prev;
5521 struct btrfs_inode *entry;
5522 struct inode *inode;
5523 u64 objectid = 0;
5524
5525 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5526 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5527
5528 spin_lock(&root->inode_lock);
5529 again:
5530 node = root->inode_tree.rb_node;
5531 prev = NULL;
5532 while (node) {
5533 prev = node;
5534 entry = rb_entry(node, struct btrfs_inode, rb_node);
5535
5536 if (objectid < btrfs_ino(&entry->vfs_inode))
5537 node = node->rb_left;
5538 else if (objectid > btrfs_ino(&entry->vfs_inode))
5539 node = node->rb_right;
5540 else
5541 break;
5542 }
5543 if (!node) {
5544 while (prev) {
5545 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5546 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5547 node = prev;
5548 break;
5549 }
5550 prev = rb_next(prev);
5551 }
5552 }
5553 while (node) {
5554 entry = rb_entry(node, struct btrfs_inode, rb_node);
5555 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5556 inode = igrab(&entry->vfs_inode);
5557 if (inode) {
5558 spin_unlock(&root->inode_lock);
5559 if (atomic_read(&inode->i_count) > 1)
5560 d_prune_aliases(inode);
5561 /*
5562 * btrfs_drop_inode will have it removed from
5563 * the inode cache when its usage count
5564 * hits zero.
5565 */
5566 iput(inode);
5567 cond_resched();
5568 spin_lock(&root->inode_lock);
5569 goto again;
5570 }
5571
5572 if (cond_resched_lock(&root->inode_lock))
5573 goto again;
5574
5575 node = rb_next(node);
5576 }
5577 spin_unlock(&root->inode_lock);
5578 }
5579
5580 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5581 {
5582 struct btrfs_iget_args *args = p;
5583 inode->i_ino = args->location->objectid;
5584 memcpy(&BTRFS_I(inode)->location, args->location,
5585 sizeof(*args->location));
5586 BTRFS_I(inode)->root = args->root;
5587 return 0;
5588 }
5589
5590 static int btrfs_find_actor(struct inode *inode, void *opaque)
5591 {
5592 struct btrfs_iget_args *args = opaque;
5593 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5594 args->root == BTRFS_I(inode)->root;
5595 }
5596
5597 static struct inode *btrfs_iget_locked(struct super_block *s,
5598 struct btrfs_key *location,
5599 struct btrfs_root *root)
5600 {
5601 struct inode *inode;
5602 struct btrfs_iget_args args;
5603 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5604
5605 args.location = location;
5606 args.root = root;
5607
5608 inode = iget5_locked(s, hashval, btrfs_find_actor,
5609 btrfs_init_locked_inode,
5610 (void *)&args);
5611 return inode;
5612 }
5613
5614 /* Get an inode object given its location and corresponding root.
5615 * Returns in *is_new if the inode was read from disk
5616 */
5617 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5618 struct btrfs_root *root, int *new)
5619 {
5620 struct inode *inode;
5621
5622 inode = btrfs_iget_locked(s, location, root);
5623 if (!inode)
5624 return ERR_PTR(-ENOMEM);
5625
5626 if (inode->i_state & I_NEW) {
5627 btrfs_read_locked_inode(inode);
5628 if (!is_bad_inode(inode)) {
5629 inode_tree_add(inode);
5630 unlock_new_inode(inode);
5631 if (new)
5632 *new = 1;
5633 } else {
5634 unlock_new_inode(inode);
5635 iput(inode);
5636 inode = ERR_PTR(-ESTALE);
5637 }
5638 }
5639
5640 return inode;
5641 }
5642
5643 static struct inode *new_simple_dir(struct super_block *s,
5644 struct btrfs_key *key,
5645 struct btrfs_root *root)
5646 {
5647 struct inode *inode = new_inode(s);
5648
5649 if (!inode)
5650 return ERR_PTR(-ENOMEM);
5651
5652 BTRFS_I(inode)->root = root;
5653 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5654 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5655
5656 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5657 inode->i_op = &btrfs_dir_ro_inode_operations;
5658 inode->i_fop = &simple_dir_operations;
5659 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5660 inode->i_mtime = CURRENT_TIME;
5661 inode->i_atime = inode->i_mtime;
5662 inode->i_ctime = inode->i_mtime;
5663 BTRFS_I(inode)->i_otime = inode->i_mtime;
5664
5665 return inode;
5666 }
5667
5668 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5669 {
5670 struct inode *inode;
5671 struct btrfs_root *root = BTRFS_I(dir)->root;
5672 struct btrfs_root *sub_root = root;
5673 struct btrfs_key location;
5674 int index;
5675 int ret = 0;
5676
5677 if (dentry->d_name.len > BTRFS_NAME_LEN)
5678 return ERR_PTR(-ENAMETOOLONG);
5679
5680 ret = btrfs_inode_by_name(dir, dentry, &location);
5681 if (ret < 0)
5682 return ERR_PTR(ret);
5683
5684 if (location.objectid == 0)
5685 return ERR_PTR(-ENOENT);
5686
5687 if (location.type == BTRFS_INODE_ITEM_KEY) {
5688 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5689 return inode;
5690 }
5691
5692 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5693
5694 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5695 ret = fixup_tree_root_location(root, dir, dentry,
5696 &location, &sub_root);
5697 if (ret < 0) {
5698 if (ret != -ENOENT)
5699 inode = ERR_PTR(ret);
5700 else
5701 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5702 } else {
5703 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5704 }
5705 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5706
5707 if (!IS_ERR(inode) && root != sub_root) {
5708 down_read(&root->fs_info->cleanup_work_sem);
5709 if (!(inode->i_sb->s_flags & MS_RDONLY))
5710 ret = btrfs_orphan_cleanup(sub_root);
5711 up_read(&root->fs_info->cleanup_work_sem);
5712 if (ret) {
5713 iput(inode);
5714 inode = ERR_PTR(ret);
5715 }
5716 }
5717
5718 return inode;
5719 }
5720
5721 static int btrfs_dentry_delete(const struct dentry *dentry)
5722 {
5723 struct btrfs_root *root;
5724 struct inode *inode = d_inode(dentry);
5725
5726 if (!inode && !IS_ROOT(dentry))
5727 inode = d_inode(dentry->d_parent);
5728
5729 if (inode) {
5730 root = BTRFS_I(inode)->root;
5731 if (btrfs_root_refs(&root->root_item) == 0)
5732 return 1;
5733
5734 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5735 return 1;
5736 }
5737 return 0;
5738 }
5739
5740 static void btrfs_dentry_release(struct dentry *dentry)
5741 {
5742 kfree(dentry->d_fsdata);
5743 }
5744
5745 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5746 unsigned int flags)
5747 {
5748 struct inode *inode;
5749
5750 inode = btrfs_lookup_dentry(dir, dentry);
5751 if (IS_ERR(inode)) {
5752 if (PTR_ERR(inode) == -ENOENT)
5753 inode = NULL;
5754 else
5755 return ERR_CAST(inode);
5756 }
5757
5758 return d_splice_alias(inode, dentry);
5759 }
5760
5761 unsigned char btrfs_filetype_table[] = {
5762 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5763 };
5764
5765 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5766 {
5767 struct inode *inode = file_inode(file);
5768 struct btrfs_root *root = BTRFS_I(inode)->root;
5769 struct btrfs_item *item;
5770 struct btrfs_dir_item *di;
5771 struct btrfs_key key;
5772 struct btrfs_key found_key;
5773 struct btrfs_path *path;
5774 struct list_head ins_list;
5775 struct list_head del_list;
5776 int ret;
5777 struct extent_buffer *leaf;
5778 int slot;
5779 unsigned char d_type;
5780 int over = 0;
5781 u32 di_cur;
5782 u32 di_total;
5783 u32 di_len;
5784 int key_type = BTRFS_DIR_INDEX_KEY;
5785 char tmp_name[32];
5786 char *name_ptr;
5787 int name_len;
5788 int is_curr = 0; /* ctx->pos points to the current index? */
5789 bool emitted;
5790
5791 /* FIXME, use a real flag for deciding about the key type */
5792 if (root->fs_info->tree_root == root)
5793 key_type = BTRFS_DIR_ITEM_KEY;
5794
5795 if (!dir_emit_dots(file, ctx))
5796 return 0;
5797
5798 path = btrfs_alloc_path();
5799 if (!path)
5800 return -ENOMEM;
5801
5802 path->reada = 1;
5803
5804 if (key_type == BTRFS_DIR_INDEX_KEY) {
5805 INIT_LIST_HEAD(&ins_list);
5806 INIT_LIST_HEAD(&del_list);
5807 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5808 }
5809
5810 key.type = key_type;
5811 key.offset = ctx->pos;
5812 key.objectid = btrfs_ino(inode);
5813
5814 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5815 if (ret < 0)
5816 goto err;
5817
5818 emitted = false;
5819 while (1) {
5820 leaf = path->nodes[0];
5821 slot = path->slots[0];
5822 if (slot >= btrfs_header_nritems(leaf)) {
5823 ret = btrfs_next_leaf(root, path);
5824 if (ret < 0)
5825 goto err;
5826 else if (ret > 0)
5827 break;
5828 continue;
5829 }
5830
5831 item = btrfs_item_nr(slot);
5832 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5833
5834 if (found_key.objectid != key.objectid)
5835 break;
5836 if (found_key.type != key_type)
5837 break;
5838 if (found_key.offset < ctx->pos)
5839 goto next;
5840 if (key_type == BTRFS_DIR_INDEX_KEY &&
5841 btrfs_should_delete_dir_index(&del_list,
5842 found_key.offset))
5843 goto next;
5844
5845 ctx->pos = found_key.offset;
5846 is_curr = 1;
5847
5848 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5849 di_cur = 0;
5850 di_total = btrfs_item_size(leaf, item);
5851
5852 while (di_cur < di_total) {
5853 struct btrfs_key location;
5854
5855 if (verify_dir_item(root, leaf, di))
5856 break;
5857
5858 name_len = btrfs_dir_name_len(leaf, di);
5859 if (name_len <= sizeof(tmp_name)) {
5860 name_ptr = tmp_name;
5861 } else {
5862 name_ptr = kmalloc(name_len, GFP_NOFS);
5863 if (!name_ptr) {
5864 ret = -ENOMEM;
5865 goto err;
5866 }
5867 }
5868 read_extent_buffer(leaf, name_ptr,
5869 (unsigned long)(di + 1), name_len);
5870
5871 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5872 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5873
5874
5875 /* is this a reference to our own snapshot? If so
5876 * skip it.
5877 *
5878 * In contrast to old kernels, we insert the snapshot's
5879 * dir item and dir index after it has been created, so
5880 * we won't find a reference to our own snapshot. We
5881 * still keep the following code for backward
5882 * compatibility.
5883 */
5884 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5885 location.objectid == root->root_key.objectid) {
5886 over = 0;
5887 goto skip;
5888 }
5889 over = !dir_emit(ctx, name_ptr, name_len,
5890 location.objectid, d_type);
5891
5892 skip:
5893 if (name_ptr != tmp_name)
5894 kfree(name_ptr);
5895
5896 if (over)
5897 goto nopos;
5898 emitted = true;
5899 di_len = btrfs_dir_name_len(leaf, di) +
5900 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5901 di_cur += di_len;
5902 di = (struct btrfs_dir_item *)((char *)di + di_len);
5903 }
5904 next:
5905 path->slots[0]++;
5906 }
5907
5908 if (key_type == BTRFS_DIR_INDEX_KEY) {
5909 if (is_curr)
5910 ctx->pos++;
5911 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5912 if (ret)
5913 goto nopos;
5914 }
5915
5916 /*
5917 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5918 * it was was set to the termination value in previous call. We assume
5919 * that "." and ".." were emitted if we reach this point and set the
5920 * termination value as well for an empty directory.
5921 */
5922 if (ctx->pos > 2 && !emitted)
5923 goto nopos;
5924
5925 /* Reached end of directory/root. Bump pos past the last item. */
5926 ctx->pos++;
5927
5928 /*
5929 * Stop new entries from being returned after we return the last
5930 * entry.
5931 *
5932 * New directory entries are assigned a strictly increasing
5933 * offset. This means that new entries created during readdir
5934 * are *guaranteed* to be seen in the future by that readdir.
5935 * This has broken buggy programs which operate on names as
5936 * they're returned by readdir. Until we re-use freed offsets
5937 * we have this hack to stop new entries from being returned
5938 * under the assumption that they'll never reach this huge
5939 * offset.
5940 *
5941 * This is being careful not to overflow 32bit loff_t unless the
5942 * last entry requires it because doing so has broken 32bit apps
5943 * in the past.
5944 */
5945 if (key_type == BTRFS_DIR_INDEX_KEY) {
5946 if (ctx->pos >= INT_MAX)
5947 ctx->pos = LLONG_MAX;
5948 else
5949 ctx->pos = INT_MAX;
5950 }
5951 nopos:
5952 ret = 0;
5953 err:
5954 if (key_type == BTRFS_DIR_INDEX_KEY)
5955 btrfs_put_delayed_items(&ins_list, &del_list);
5956 btrfs_free_path(path);
5957 return ret;
5958 }
5959
5960 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5961 {
5962 struct btrfs_root *root = BTRFS_I(inode)->root;
5963 struct btrfs_trans_handle *trans;
5964 int ret = 0;
5965 bool nolock = false;
5966
5967 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5968 return 0;
5969
5970 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5971 nolock = true;
5972
5973 if (wbc->sync_mode == WB_SYNC_ALL) {
5974 if (nolock)
5975 trans = btrfs_join_transaction_nolock(root);
5976 else
5977 trans = btrfs_join_transaction(root);
5978 if (IS_ERR(trans))
5979 return PTR_ERR(trans);
5980 ret = btrfs_commit_transaction(trans, root);
5981 }
5982 return ret;
5983 }
5984
5985 /*
5986 * This is somewhat expensive, updating the tree every time the
5987 * inode changes. But, it is most likely to find the inode in cache.
5988 * FIXME, needs more benchmarking...there are no reasons other than performance
5989 * to keep or drop this code.
5990 */
5991 static int btrfs_dirty_inode(struct inode *inode)
5992 {
5993 struct btrfs_root *root = BTRFS_I(inode)->root;
5994 struct btrfs_trans_handle *trans;
5995 int ret;
5996
5997 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5998 return 0;
5999
6000 trans = btrfs_join_transaction(root);
6001 if (IS_ERR(trans))
6002 return PTR_ERR(trans);
6003
6004 ret = btrfs_update_inode(trans, root, inode);
6005 if (ret && ret == -ENOSPC) {
6006 /* whoops, lets try again with the full transaction */
6007 btrfs_end_transaction(trans, root);
6008 trans = btrfs_start_transaction(root, 1);
6009 if (IS_ERR(trans))
6010 return PTR_ERR(trans);
6011
6012 ret = btrfs_update_inode(trans, root, inode);
6013 }
6014 btrfs_end_transaction(trans, root);
6015 if (BTRFS_I(inode)->delayed_node)
6016 btrfs_balance_delayed_items(root);
6017
6018 return ret;
6019 }
6020
6021 /*
6022 * This is a copy of file_update_time. We need this so we can return error on
6023 * ENOSPC for updating the inode in the case of file write and mmap writes.
6024 */
6025 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6026 int flags)
6027 {
6028 struct btrfs_root *root = BTRFS_I(inode)->root;
6029
6030 if (btrfs_root_readonly(root))
6031 return -EROFS;
6032
6033 if (flags & S_VERSION)
6034 inode_inc_iversion(inode);
6035 if (flags & S_CTIME)
6036 inode->i_ctime = *now;
6037 if (flags & S_MTIME)
6038 inode->i_mtime = *now;
6039 if (flags & S_ATIME)
6040 inode->i_atime = *now;
6041 return btrfs_dirty_inode(inode);
6042 }
6043
6044 /*
6045 * find the highest existing sequence number in a directory
6046 * and then set the in-memory index_cnt variable to reflect
6047 * free sequence numbers
6048 */
6049 static int btrfs_set_inode_index_count(struct inode *inode)
6050 {
6051 struct btrfs_root *root = BTRFS_I(inode)->root;
6052 struct btrfs_key key, found_key;
6053 struct btrfs_path *path;
6054 struct extent_buffer *leaf;
6055 int ret;
6056
6057 key.objectid = btrfs_ino(inode);
6058 key.type = BTRFS_DIR_INDEX_KEY;
6059 key.offset = (u64)-1;
6060
6061 path = btrfs_alloc_path();
6062 if (!path)
6063 return -ENOMEM;
6064
6065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6066 if (ret < 0)
6067 goto out;
6068 /* FIXME: we should be able to handle this */
6069 if (ret == 0)
6070 goto out;
6071 ret = 0;
6072
6073 /*
6074 * MAGIC NUMBER EXPLANATION:
6075 * since we search a directory based on f_pos we have to start at 2
6076 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6077 * else has to start at 2
6078 */
6079 if (path->slots[0] == 0) {
6080 BTRFS_I(inode)->index_cnt = 2;
6081 goto out;
6082 }
6083
6084 path->slots[0]--;
6085
6086 leaf = path->nodes[0];
6087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6088
6089 if (found_key.objectid != btrfs_ino(inode) ||
6090 found_key.type != BTRFS_DIR_INDEX_KEY) {
6091 BTRFS_I(inode)->index_cnt = 2;
6092 goto out;
6093 }
6094
6095 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6096 out:
6097 btrfs_free_path(path);
6098 return ret;
6099 }
6100
6101 /*
6102 * helper to find a free sequence number in a given directory. This current
6103 * code is very simple, later versions will do smarter things in the btree
6104 */
6105 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6106 {
6107 int ret = 0;
6108
6109 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6110 ret = btrfs_inode_delayed_dir_index_count(dir);
6111 if (ret) {
6112 ret = btrfs_set_inode_index_count(dir);
6113 if (ret)
6114 return ret;
6115 }
6116 }
6117
6118 *index = BTRFS_I(dir)->index_cnt;
6119 BTRFS_I(dir)->index_cnt++;
6120
6121 return ret;
6122 }
6123
6124 static int btrfs_insert_inode_locked(struct inode *inode)
6125 {
6126 struct btrfs_iget_args args;
6127 args.location = &BTRFS_I(inode)->location;
6128 args.root = BTRFS_I(inode)->root;
6129
6130 return insert_inode_locked4(inode,
6131 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6132 btrfs_find_actor, &args);
6133 }
6134
6135 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6136 struct btrfs_root *root,
6137 struct inode *dir,
6138 const char *name, int name_len,
6139 u64 ref_objectid, u64 objectid,
6140 umode_t mode, u64 *index)
6141 {
6142 struct inode *inode;
6143 struct btrfs_inode_item *inode_item;
6144 struct btrfs_key *location;
6145 struct btrfs_path *path;
6146 struct btrfs_inode_ref *ref;
6147 struct btrfs_key key[2];
6148 u32 sizes[2];
6149 int nitems = name ? 2 : 1;
6150 unsigned long ptr;
6151 int ret;
6152
6153 path = btrfs_alloc_path();
6154 if (!path)
6155 return ERR_PTR(-ENOMEM);
6156
6157 inode = new_inode(root->fs_info->sb);
6158 if (!inode) {
6159 btrfs_free_path(path);
6160 return ERR_PTR(-ENOMEM);
6161 }
6162
6163 /*
6164 * O_TMPFILE, set link count to 0, so that after this point,
6165 * we fill in an inode item with the correct link count.
6166 */
6167 if (!name)
6168 set_nlink(inode, 0);
6169
6170 /*
6171 * we have to initialize this early, so we can reclaim the inode
6172 * number if we fail afterwards in this function.
6173 */
6174 inode->i_ino = objectid;
6175
6176 if (dir && name) {
6177 trace_btrfs_inode_request(dir);
6178
6179 ret = btrfs_set_inode_index(dir, index);
6180 if (ret) {
6181 btrfs_free_path(path);
6182 iput(inode);
6183 return ERR_PTR(ret);
6184 }
6185 } else if (dir) {
6186 *index = 0;
6187 }
6188 /*
6189 * index_cnt is ignored for everything but a dir,
6190 * btrfs_get_inode_index_count has an explanation for the magic
6191 * number
6192 */
6193 BTRFS_I(inode)->index_cnt = 2;
6194 BTRFS_I(inode)->dir_index = *index;
6195 BTRFS_I(inode)->root = root;
6196 BTRFS_I(inode)->generation = trans->transid;
6197 inode->i_generation = BTRFS_I(inode)->generation;
6198
6199 /*
6200 * We could have gotten an inode number from somebody who was fsynced
6201 * and then removed in this same transaction, so let's just set full
6202 * sync since it will be a full sync anyway and this will blow away the
6203 * old info in the log.
6204 */
6205 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6206
6207 key[0].objectid = objectid;
6208 key[0].type = BTRFS_INODE_ITEM_KEY;
6209 key[0].offset = 0;
6210
6211 sizes[0] = sizeof(struct btrfs_inode_item);
6212
6213 if (name) {
6214 /*
6215 * Start new inodes with an inode_ref. This is slightly more
6216 * efficient for small numbers of hard links since they will
6217 * be packed into one item. Extended refs will kick in if we
6218 * add more hard links than can fit in the ref item.
6219 */
6220 key[1].objectid = objectid;
6221 key[1].type = BTRFS_INODE_REF_KEY;
6222 key[1].offset = ref_objectid;
6223
6224 sizes[1] = name_len + sizeof(*ref);
6225 }
6226
6227 location = &BTRFS_I(inode)->location;
6228 location->objectid = objectid;
6229 location->offset = 0;
6230 location->type = BTRFS_INODE_ITEM_KEY;
6231
6232 ret = btrfs_insert_inode_locked(inode);
6233 if (ret < 0)
6234 goto fail;
6235
6236 path->leave_spinning = 1;
6237 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6238 if (ret != 0)
6239 goto fail_unlock;
6240
6241 inode_init_owner(inode, dir, mode);
6242 inode_set_bytes(inode, 0);
6243
6244 inode->i_mtime = CURRENT_TIME;
6245 inode->i_atime = inode->i_mtime;
6246 inode->i_ctime = inode->i_mtime;
6247 BTRFS_I(inode)->i_otime = inode->i_mtime;
6248
6249 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6250 struct btrfs_inode_item);
6251 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6252 sizeof(*inode_item));
6253 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6254
6255 if (name) {
6256 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6257 struct btrfs_inode_ref);
6258 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6259 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6260 ptr = (unsigned long)(ref + 1);
6261 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6262 }
6263
6264 btrfs_mark_buffer_dirty(path->nodes[0]);
6265 btrfs_free_path(path);
6266
6267 btrfs_inherit_iflags(inode, dir);
6268
6269 if (S_ISREG(mode)) {
6270 if (btrfs_test_opt(root, NODATASUM))
6271 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6272 if (btrfs_test_opt(root, NODATACOW))
6273 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6274 BTRFS_INODE_NODATASUM;
6275 }
6276
6277 inode_tree_add(inode);
6278
6279 trace_btrfs_inode_new(inode);
6280 btrfs_set_inode_last_trans(trans, inode);
6281
6282 btrfs_update_root_times(trans, root);
6283
6284 ret = btrfs_inode_inherit_props(trans, inode, dir);
6285 if (ret)
6286 btrfs_err(root->fs_info,
6287 "error inheriting props for ino %llu (root %llu): %d",
6288 btrfs_ino(inode), root->root_key.objectid, ret);
6289
6290 return inode;
6291
6292 fail_unlock:
6293 unlock_new_inode(inode);
6294 fail:
6295 if (dir && name)
6296 BTRFS_I(dir)->index_cnt--;
6297 btrfs_free_path(path);
6298 iput(inode);
6299 return ERR_PTR(ret);
6300 }
6301
6302 static inline u8 btrfs_inode_type(struct inode *inode)
6303 {
6304 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6305 }
6306
6307 /*
6308 * utility function to add 'inode' into 'parent_inode' with
6309 * a give name and a given sequence number.
6310 * if 'add_backref' is true, also insert a backref from the
6311 * inode to the parent directory.
6312 */
6313 int btrfs_add_link(struct btrfs_trans_handle *trans,
6314 struct inode *parent_inode, struct inode *inode,
6315 const char *name, int name_len, int add_backref, u64 index)
6316 {
6317 int ret = 0;
6318 struct btrfs_key key;
6319 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6320 u64 ino = btrfs_ino(inode);
6321 u64 parent_ino = btrfs_ino(parent_inode);
6322
6323 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6324 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6325 } else {
6326 key.objectid = ino;
6327 key.type = BTRFS_INODE_ITEM_KEY;
6328 key.offset = 0;
6329 }
6330
6331 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6332 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6333 key.objectid, root->root_key.objectid,
6334 parent_ino, index, name, name_len);
6335 } else if (add_backref) {
6336 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6337 parent_ino, index);
6338 }
6339
6340 /* Nothing to clean up yet */
6341 if (ret)
6342 return ret;
6343
6344 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6345 parent_inode, &key,
6346 btrfs_inode_type(inode), index);
6347 if (ret == -EEXIST || ret == -EOVERFLOW)
6348 goto fail_dir_item;
6349 else if (ret) {
6350 btrfs_abort_transaction(trans, root, ret);
6351 return ret;
6352 }
6353
6354 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6355 name_len * 2);
6356 inode_inc_iversion(parent_inode);
6357 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6358 ret = btrfs_update_inode(trans, root, parent_inode);
6359 if (ret)
6360 btrfs_abort_transaction(trans, root, ret);
6361 return ret;
6362
6363 fail_dir_item:
6364 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6365 u64 local_index;
6366 int err;
6367 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6368 key.objectid, root->root_key.objectid,
6369 parent_ino, &local_index, name, name_len);
6370
6371 } else if (add_backref) {
6372 u64 local_index;
6373 int err;
6374
6375 err = btrfs_del_inode_ref(trans, root, name, name_len,
6376 ino, parent_ino, &local_index);
6377 }
6378 return ret;
6379 }
6380
6381 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6382 struct inode *dir, struct dentry *dentry,
6383 struct inode *inode, int backref, u64 index)
6384 {
6385 int err = btrfs_add_link(trans, dir, inode,
6386 dentry->d_name.name, dentry->d_name.len,
6387 backref, index);
6388 if (err > 0)
6389 err = -EEXIST;
6390 return err;
6391 }
6392
6393 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6394 umode_t mode, dev_t rdev)
6395 {
6396 struct btrfs_trans_handle *trans;
6397 struct btrfs_root *root = BTRFS_I(dir)->root;
6398 struct inode *inode = NULL;
6399 int err;
6400 int drop_inode = 0;
6401 u64 objectid;
6402 u64 index = 0;
6403
6404 /*
6405 * 2 for inode item and ref
6406 * 2 for dir items
6407 * 1 for xattr if selinux is on
6408 */
6409 trans = btrfs_start_transaction(root, 5);
6410 if (IS_ERR(trans))
6411 return PTR_ERR(trans);
6412
6413 err = btrfs_find_free_ino(root, &objectid);
6414 if (err)
6415 goto out_unlock;
6416
6417 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6418 dentry->d_name.len, btrfs_ino(dir), objectid,
6419 mode, &index);
6420 if (IS_ERR(inode)) {
6421 err = PTR_ERR(inode);
6422 goto out_unlock;
6423 }
6424
6425 /*
6426 * If the active LSM wants to access the inode during
6427 * d_instantiate it needs these. Smack checks to see
6428 * if the filesystem supports xattrs by looking at the
6429 * ops vector.
6430 */
6431 inode->i_op = &btrfs_special_inode_operations;
6432 init_special_inode(inode, inode->i_mode, rdev);
6433
6434 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6435 if (err)
6436 goto out_unlock_inode;
6437
6438 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6439 if (err) {
6440 goto out_unlock_inode;
6441 } else {
6442 btrfs_update_inode(trans, root, inode);
6443 d_instantiate_new(dentry, inode);
6444 }
6445
6446 out_unlock:
6447 btrfs_end_transaction(trans, root);
6448 btrfs_balance_delayed_items(root);
6449 btrfs_btree_balance_dirty(root);
6450 if (drop_inode) {
6451 inode_dec_link_count(inode);
6452 iput(inode);
6453 }
6454 return err;
6455
6456 out_unlock_inode:
6457 drop_inode = 1;
6458 unlock_new_inode(inode);
6459 goto out_unlock;
6460
6461 }
6462
6463 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6464 umode_t mode, bool excl)
6465 {
6466 struct btrfs_trans_handle *trans;
6467 struct btrfs_root *root = BTRFS_I(dir)->root;
6468 struct inode *inode = NULL;
6469 int drop_inode_on_err = 0;
6470 int err;
6471 u64 objectid;
6472 u64 index = 0;
6473
6474 /*
6475 * 2 for inode item and ref
6476 * 2 for dir items
6477 * 1 for xattr if selinux is on
6478 */
6479 trans = btrfs_start_transaction(root, 5);
6480 if (IS_ERR(trans))
6481 return PTR_ERR(trans);
6482
6483 err = btrfs_find_free_ino(root, &objectid);
6484 if (err)
6485 goto out_unlock;
6486
6487 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6488 dentry->d_name.len, btrfs_ino(dir), objectid,
6489 mode, &index);
6490 if (IS_ERR(inode)) {
6491 err = PTR_ERR(inode);
6492 goto out_unlock;
6493 }
6494 drop_inode_on_err = 1;
6495 /*
6496 * If the active LSM wants to access the inode during
6497 * d_instantiate it needs these. Smack checks to see
6498 * if the filesystem supports xattrs by looking at the
6499 * ops vector.
6500 */
6501 inode->i_fop = &btrfs_file_operations;
6502 inode->i_op = &btrfs_file_inode_operations;
6503 inode->i_mapping->a_ops = &btrfs_aops;
6504
6505 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6506 if (err)
6507 goto out_unlock_inode;
6508
6509 err = btrfs_update_inode(trans, root, inode);
6510 if (err)
6511 goto out_unlock_inode;
6512
6513 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6514 if (err)
6515 goto out_unlock_inode;
6516
6517 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6518 d_instantiate_new(dentry, inode);
6519
6520 out_unlock:
6521 btrfs_end_transaction(trans, root);
6522 if (err && drop_inode_on_err) {
6523 inode_dec_link_count(inode);
6524 iput(inode);
6525 }
6526 btrfs_balance_delayed_items(root);
6527 btrfs_btree_balance_dirty(root);
6528 return err;
6529
6530 out_unlock_inode:
6531 unlock_new_inode(inode);
6532 goto out_unlock;
6533
6534 }
6535
6536 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6537 struct dentry *dentry)
6538 {
6539 struct btrfs_trans_handle *trans = NULL;
6540 struct btrfs_root *root = BTRFS_I(dir)->root;
6541 struct inode *inode = d_inode(old_dentry);
6542 u64 index;
6543 int err;
6544 int drop_inode = 0;
6545
6546 /* do not allow sys_link's with other subvols of the same device */
6547 if (root->objectid != BTRFS_I(inode)->root->objectid)
6548 return -EXDEV;
6549
6550 if (inode->i_nlink >= BTRFS_LINK_MAX)
6551 return -EMLINK;
6552
6553 err = btrfs_set_inode_index(dir, &index);
6554 if (err)
6555 goto fail;
6556
6557 /*
6558 * 2 items for inode and inode ref
6559 * 2 items for dir items
6560 * 1 item for parent inode
6561 */
6562 trans = btrfs_start_transaction(root, 5);
6563 if (IS_ERR(trans)) {
6564 err = PTR_ERR(trans);
6565 trans = NULL;
6566 goto fail;
6567 }
6568
6569 /* There are several dir indexes for this inode, clear the cache. */
6570 BTRFS_I(inode)->dir_index = 0ULL;
6571 inc_nlink(inode);
6572 inode_inc_iversion(inode);
6573 inode->i_ctime = CURRENT_TIME;
6574 ihold(inode);
6575 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6576
6577 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6578
6579 if (err) {
6580 drop_inode = 1;
6581 } else {
6582 struct dentry *parent = dentry->d_parent;
6583 err = btrfs_update_inode(trans, root, inode);
6584 if (err)
6585 goto fail;
6586 if (inode->i_nlink == 1) {
6587 /*
6588 * If new hard link count is 1, it's a file created
6589 * with open(2) O_TMPFILE flag.
6590 */
6591 err = btrfs_orphan_del(trans, inode);
6592 if (err)
6593 goto fail;
6594 }
6595 d_instantiate(dentry, inode);
6596 btrfs_log_new_name(trans, inode, NULL, parent);
6597 }
6598
6599 btrfs_balance_delayed_items(root);
6600 fail:
6601 if (trans)
6602 btrfs_end_transaction(trans, root);
6603 if (drop_inode) {
6604 inode_dec_link_count(inode);
6605 iput(inode);
6606 }
6607 btrfs_btree_balance_dirty(root);
6608 return err;
6609 }
6610
6611 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6612 {
6613 struct inode *inode = NULL;
6614 struct btrfs_trans_handle *trans;
6615 struct btrfs_root *root = BTRFS_I(dir)->root;
6616 int err = 0;
6617 int drop_on_err = 0;
6618 u64 objectid = 0;
6619 u64 index = 0;
6620
6621 /*
6622 * 2 items for inode and ref
6623 * 2 items for dir items
6624 * 1 for xattr if selinux is on
6625 */
6626 trans = btrfs_start_transaction(root, 5);
6627 if (IS_ERR(trans))
6628 return PTR_ERR(trans);
6629
6630 err = btrfs_find_free_ino(root, &objectid);
6631 if (err)
6632 goto out_fail;
6633
6634 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6635 dentry->d_name.len, btrfs_ino(dir), objectid,
6636 S_IFDIR | mode, &index);
6637 if (IS_ERR(inode)) {
6638 err = PTR_ERR(inode);
6639 goto out_fail;
6640 }
6641
6642 drop_on_err = 1;
6643 /* these must be set before we unlock the inode */
6644 inode->i_op = &btrfs_dir_inode_operations;
6645 inode->i_fop = &btrfs_dir_file_operations;
6646
6647 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6648 if (err)
6649 goto out_fail_inode;
6650
6651 btrfs_i_size_write(inode, 0);
6652 err = btrfs_update_inode(trans, root, inode);
6653 if (err)
6654 goto out_fail_inode;
6655
6656 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6657 dentry->d_name.len, 0, index);
6658 if (err)
6659 goto out_fail_inode;
6660
6661 d_instantiate_new(dentry, inode);
6662 drop_on_err = 0;
6663
6664 out_fail:
6665 btrfs_end_transaction(trans, root);
6666 if (drop_on_err) {
6667 inode_dec_link_count(inode);
6668 iput(inode);
6669 }
6670 btrfs_balance_delayed_items(root);
6671 btrfs_btree_balance_dirty(root);
6672 return err;
6673
6674 out_fail_inode:
6675 unlock_new_inode(inode);
6676 goto out_fail;
6677 }
6678
6679 /* Find next extent map of a given extent map, caller needs to ensure locks */
6680 static struct extent_map *next_extent_map(struct extent_map *em)
6681 {
6682 struct rb_node *next;
6683
6684 next = rb_next(&em->rb_node);
6685 if (!next)
6686 return NULL;
6687 return container_of(next, struct extent_map, rb_node);
6688 }
6689
6690 static struct extent_map *prev_extent_map(struct extent_map *em)
6691 {
6692 struct rb_node *prev;
6693
6694 prev = rb_prev(&em->rb_node);
6695 if (!prev)
6696 return NULL;
6697 return container_of(prev, struct extent_map, rb_node);
6698 }
6699
6700 /* helper for btfs_get_extent. Given an existing extent in the tree,
6701 * the existing extent is the nearest extent to map_start,
6702 * and an extent that you want to insert, deal with overlap and insert
6703 * the best fitted new extent into the tree.
6704 */
6705 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6706 struct extent_map *existing,
6707 struct extent_map *em,
6708 u64 map_start)
6709 {
6710 struct extent_map *prev;
6711 struct extent_map *next;
6712 u64 start;
6713 u64 end;
6714 u64 start_diff;
6715
6716 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6717
6718 if (existing->start > map_start) {
6719 next = existing;
6720 prev = prev_extent_map(next);
6721 } else {
6722 prev = existing;
6723 next = next_extent_map(prev);
6724 }
6725
6726 start = prev ? extent_map_end(prev) : em->start;
6727 start = max_t(u64, start, em->start);
6728 end = next ? next->start : extent_map_end(em);
6729 end = min_t(u64, end, extent_map_end(em));
6730 start_diff = start - em->start;
6731 em->start = start;
6732 em->len = end - start;
6733 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6734 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6735 em->block_start += start_diff;
6736 em->block_len -= start_diff;
6737 }
6738 return add_extent_mapping(em_tree, em, 0);
6739 }
6740
6741 static noinline int uncompress_inline(struct btrfs_path *path,
6742 struct inode *inode, struct page *page,
6743 size_t pg_offset, u64 extent_offset,
6744 struct btrfs_file_extent_item *item)
6745 {
6746 int ret;
6747 struct extent_buffer *leaf = path->nodes[0];
6748 char *tmp;
6749 size_t max_size;
6750 unsigned long inline_size;
6751 unsigned long ptr;
6752 int compress_type;
6753
6754 WARN_ON(pg_offset != 0);
6755 compress_type = btrfs_file_extent_compression(leaf, item);
6756 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6757 inline_size = btrfs_file_extent_inline_item_len(leaf,
6758 btrfs_item_nr(path->slots[0]));
6759 tmp = kmalloc(inline_size, GFP_NOFS);
6760 if (!tmp)
6761 return -ENOMEM;
6762 ptr = btrfs_file_extent_inline_start(item);
6763
6764 read_extent_buffer(leaf, tmp, ptr, inline_size);
6765
6766 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6767 ret = btrfs_decompress(compress_type, tmp, page,
6768 extent_offset, inline_size, max_size);
6769
6770 /*
6771 * decompression code contains a memset to fill in any space between the end
6772 * of the uncompressed data and the end of max_size in case the decompressed
6773 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6774 * the end of an inline extent and the beginning of the next block, so we
6775 * cover that region here.
6776 */
6777
6778 if (max_size + pg_offset < PAGE_SIZE) {
6779 char *map = kmap(page);
6780 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6781 kunmap(page);
6782 }
6783 kfree(tmp);
6784 return ret;
6785 }
6786
6787 /*
6788 * a bit scary, this does extent mapping from logical file offset to the disk.
6789 * the ugly parts come from merging extents from the disk with the in-ram
6790 * representation. This gets more complex because of the data=ordered code,
6791 * where the in-ram extents might be locked pending data=ordered completion.
6792 *
6793 * This also copies inline extents directly into the page.
6794 */
6795
6796 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6797 size_t pg_offset, u64 start, u64 len,
6798 int create)
6799 {
6800 int ret;
6801 int err = 0;
6802 u64 extent_start = 0;
6803 u64 extent_end = 0;
6804 u64 objectid = btrfs_ino(inode);
6805 u32 found_type;
6806 struct btrfs_path *path = NULL;
6807 struct btrfs_root *root = BTRFS_I(inode)->root;
6808 struct btrfs_file_extent_item *item;
6809 struct extent_buffer *leaf;
6810 struct btrfs_key found_key;
6811 struct extent_map *em = NULL;
6812 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6813 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6814 struct btrfs_trans_handle *trans = NULL;
6815 const bool new_inline = !page || create;
6816
6817 again:
6818 read_lock(&em_tree->lock);
6819 em = lookup_extent_mapping(em_tree, start, len);
6820 if (em)
6821 em->bdev = root->fs_info->fs_devices->latest_bdev;
6822 read_unlock(&em_tree->lock);
6823
6824 if (em) {
6825 if (em->start > start || em->start + em->len <= start)
6826 free_extent_map(em);
6827 else if (em->block_start == EXTENT_MAP_INLINE && page)
6828 free_extent_map(em);
6829 else
6830 goto out;
6831 }
6832 em = alloc_extent_map();
6833 if (!em) {
6834 err = -ENOMEM;
6835 goto out;
6836 }
6837 em->bdev = root->fs_info->fs_devices->latest_bdev;
6838 em->start = EXTENT_MAP_HOLE;
6839 em->orig_start = EXTENT_MAP_HOLE;
6840 em->len = (u64)-1;
6841 em->block_len = (u64)-1;
6842
6843 if (!path) {
6844 path = btrfs_alloc_path();
6845 if (!path) {
6846 err = -ENOMEM;
6847 goto out;
6848 }
6849 /*
6850 * Chances are we'll be called again, so go ahead and do
6851 * readahead
6852 */
6853 path->reada = 1;
6854 }
6855
6856 ret = btrfs_lookup_file_extent(trans, root, path,
6857 objectid, start, trans != NULL);
6858 if (ret < 0) {
6859 err = ret;
6860 goto out;
6861 }
6862
6863 if (ret != 0) {
6864 if (path->slots[0] == 0)
6865 goto not_found;
6866 path->slots[0]--;
6867 }
6868
6869 leaf = path->nodes[0];
6870 item = btrfs_item_ptr(leaf, path->slots[0],
6871 struct btrfs_file_extent_item);
6872 /* are we inside the extent that was found? */
6873 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6874 found_type = found_key.type;
6875 if (found_key.objectid != objectid ||
6876 found_type != BTRFS_EXTENT_DATA_KEY) {
6877 /*
6878 * If we backup past the first extent we want to move forward
6879 * and see if there is an extent in front of us, otherwise we'll
6880 * say there is a hole for our whole search range which can
6881 * cause problems.
6882 */
6883 extent_end = start;
6884 goto next;
6885 }
6886
6887 found_type = btrfs_file_extent_type(leaf, item);
6888 extent_start = found_key.offset;
6889 if (found_type == BTRFS_FILE_EXTENT_REG ||
6890 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6891 extent_end = extent_start +
6892 btrfs_file_extent_num_bytes(leaf, item);
6893 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6894 size_t size;
6895 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6896 extent_end = ALIGN(extent_start + size, root->sectorsize);
6897 }
6898 next:
6899 if (start >= extent_end) {
6900 path->slots[0]++;
6901 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6902 ret = btrfs_next_leaf(root, path);
6903 if (ret < 0) {
6904 err = ret;
6905 goto out;
6906 }
6907 if (ret > 0)
6908 goto not_found;
6909 leaf = path->nodes[0];
6910 }
6911 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6912 if (found_key.objectid != objectid ||
6913 found_key.type != BTRFS_EXTENT_DATA_KEY)
6914 goto not_found;
6915 if (start + len <= found_key.offset)
6916 goto not_found;
6917 if (start > found_key.offset)
6918 goto next;
6919 em->start = start;
6920 em->orig_start = start;
6921 em->len = found_key.offset - start;
6922 goto not_found_em;
6923 }
6924
6925 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6926
6927 if (found_type == BTRFS_FILE_EXTENT_REG ||
6928 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6929 goto insert;
6930 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6931 unsigned long ptr;
6932 char *map;
6933 size_t size;
6934 size_t extent_offset;
6935 size_t copy_size;
6936
6937 if (new_inline)
6938 goto out;
6939
6940 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6941 extent_offset = page_offset(page) + pg_offset - extent_start;
6942 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6943 size - extent_offset);
6944 em->start = extent_start + extent_offset;
6945 em->len = ALIGN(copy_size, root->sectorsize);
6946 em->orig_block_len = em->len;
6947 em->orig_start = em->start;
6948 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6949 if (create == 0 && !PageUptodate(page)) {
6950 if (btrfs_file_extent_compression(leaf, item) !=
6951 BTRFS_COMPRESS_NONE) {
6952 ret = uncompress_inline(path, inode, page,
6953 pg_offset,
6954 extent_offset, item);
6955 if (ret) {
6956 err = ret;
6957 goto out;
6958 }
6959 } else {
6960 map = kmap(page);
6961 read_extent_buffer(leaf, map + pg_offset, ptr,
6962 copy_size);
6963 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6964 memset(map + pg_offset + copy_size, 0,
6965 PAGE_CACHE_SIZE - pg_offset -
6966 copy_size);
6967 }
6968 kunmap(page);
6969 }
6970 flush_dcache_page(page);
6971 } else if (create && PageUptodate(page)) {
6972 BUG();
6973 if (!trans) {
6974 kunmap(page);
6975 free_extent_map(em);
6976 em = NULL;
6977
6978 btrfs_release_path(path);
6979 trans = btrfs_join_transaction(root);
6980
6981 if (IS_ERR(trans))
6982 return ERR_CAST(trans);
6983 goto again;
6984 }
6985 map = kmap(page);
6986 write_extent_buffer(leaf, map + pg_offset, ptr,
6987 copy_size);
6988 kunmap(page);
6989 btrfs_mark_buffer_dirty(leaf);
6990 }
6991 set_extent_uptodate(io_tree, em->start,
6992 extent_map_end(em) - 1, NULL, GFP_NOFS);
6993 goto insert;
6994 }
6995 not_found:
6996 em->start = start;
6997 em->orig_start = start;
6998 em->len = len;
6999 not_found_em:
7000 em->block_start = EXTENT_MAP_HOLE;
7001 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7002 insert:
7003 btrfs_release_path(path);
7004 if (em->start > start || extent_map_end(em) <= start) {
7005 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
7006 em->start, em->len, start, len);
7007 err = -EIO;
7008 goto out;
7009 }
7010
7011 err = 0;
7012 write_lock(&em_tree->lock);
7013 ret = add_extent_mapping(em_tree, em, 0);
7014 /* it is possible that someone inserted the extent into the tree
7015 * while we had the lock dropped. It is also possible that
7016 * an overlapping map exists in the tree
7017 */
7018 if (ret == -EEXIST) {
7019 struct extent_map *existing;
7020
7021 ret = 0;
7022
7023 existing = search_extent_mapping(em_tree, start, len);
7024 /*
7025 * existing will always be non-NULL, since there must be
7026 * extent causing the -EEXIST.
7027 */
7028 if (start >= extent_map_end(existing) ||
7029 start <= existing->start) {
7030 /*
7031 * The existing extent map is the one nearest to
7032 * the [start, start + len) range which overlaps
7033 */
7034 err = merge_extent_mapping(em_tree, existing,
7035 em, start);
7036 free_extent_map(existing);
7037 if (err) {
7038 free_extent_map(em);
7039 em = NULL;
7040 }
7041 } else {
7042 free_extent_map(em);
7043 em = existing;
7044 err = 0;
7045 }
7046 }
7047 write_unlock(&em_tree->lock);
7048 out:
7049
7050 trace_btrfs_get_extent(root, em);
7051
7052 btrfs_free_path(path);
7053 if (trans) {
7054 ret = btrfs_end_transaction(trans, root);
7055 if (!err)
7056 err = ret;
7057 }
7058 if (err) {
7059 free_extent_map(em);
7060 return ERR_PTR(err);
7061 }
7062 BUG_ON(!em); /* Error is always set */
7063 return em;
7064 }
7065
7066 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7067 size_t pg_offset, u64 start, u64 len,
7068 int create)
7069 {
7070 struct extent_map *em;
7071 struct extent_map *hole_em = NULL;
7072 u64 range_start = start;
7073 u64 end;
7074 u64 found;
7075 u64 found_end;
7076 int err = 0;
7077
7078 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7079 if (IS_ERR(em))
7080 return em;
7081 if (em) {
7082 /*
7083 * if our em maps to
7084 * - a hole or
7085 * - a pre-alloc extent,
7086 * there might actually be delalloc bytes behind it.
7087 */
7088 if (em->block_start != EXTENT_MAP_HOLE &&
7089 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7090 return em;
7091 else
7092 hole_em = em;
7093 }
7094
7095 /* check to see if we've wrapped (len == -1 or similar) */
7096 end = start + len;
7097 if (end < start)
7098 end = (u64)-1;
7099 else
7100 end -= 1;
7101
7102 em = NULL;
7103
7104 /* ok, we didn't find anything, lets look for delalloc */
7105 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7106 end, len, EXTENT_DELALLOC, 1);
7107 found_end = range_start + found;
7108 if (found_end < range_start)
7109 found_end = (u64)-1;
7110
7111 /*
7112 * we didn't find anything useful, return
7113 * the original results from get_extent()
7114 */
7115 if (range_start > end || found_end <= start) {
7116 em = hole_em;
7117 hole_em = NULL;
7118 goto out;
7119 }
7120
7121 /* adjust the range_start to make sure it doesn't
7122 * go backwards from the start they passed in
7123 */
7124 range_start = max(start, range_start);
7125 found = found_end - range_start;
7126
7127 if (found > 0) {
7128 u64 hole_start = start;
7129 u64 hole_len = len;
7130
7131 em = alloc_extent_map();
7132 if (!em) {
7133 err = -ENOMEM;
7134 goto out;
7135 }
7136 /*
7137 * when btrfs_get_extent can't find anything it
7138 * returns one huge hole
7139 *
7140 * make sure what it found really fits our range, and
7141 * adjust to make sure it is based on the start from
7142 * the caller
7143 */
7144 if (hole_em) {
7145 u64 calc_end = extent_map_end(hole_em);
7146
7147 if (calc_end <= start || (hole_em->start > end)) {
7148 free_extent_map(hole_em);
7149 hole_em = NULL;
7150 } else {
7151 hole_start = max(hole_em->start, start);
7152 hole_len = calc_end - hole_start;
7153 }
7154 }
7155 em->bdev = NULL;
7156 if (hole_em && range_start > hole_start) {
7157 /* our hole starts before our delalloc, so we
7158 * have to return just the parts of the hole
7159 * that go until the delalloc starts
7160 */
7161 em->len = min(hole_len,
7162 range_start - hole_start);
7163 em->start = hole_start;
7164 em->orig_start = hole_start;
7165 /*
7166 * don't adjust block start at all,
7167 * it is fixed at EXTENT_MAP_HOLE
7168 */
7169 em->block_start = hole_em->block_start;
7170 em->block_len = hole_len;
7171 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7172 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7173 } else {
7174 em->start = range_start;
7175 em->len = found;
7176 em->orig_start = range_start;
7177 em->block_start = EXTENT_MAP_DELALLOC;
7178 em->block_len = found;
7179 }
7180 } else if (hole_em) {
7181 return hole_em;
7182 }
7183 out:
7184
7185 free_extent_map(hole_em);
7186 if (err) {
7187 free_extent_map(em);
7188 return ERR_PTR(err);
7189 }
7190 return em;
7191 }
7192
7193 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7194 u64 start, u64 len)
7195 {
7196 struct btrfs_root *root = BTRFS_I(inode)->root;
7197 struct extent_map *em;
7198 struct btrfs_key ins;
7199 u64 alloc_hint;
7200 int ret;
7201
7202 alloc_hint = get_extent_allocation_hint(inode, start, len);
7203 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7204 alloc_hint, &ins, 1, 1);
7205 if (ret)
7206 return ERR_PTR(ret);
7207
7208 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7209 ins.offset, ins.offset, ins.offset, 0);
7210 if (IS_ERR(em)) {
7211 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7212 return em;
7213 }
7214
7215 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7216 ins.offset, ins.offset, 0);
7217 if (ret) {
7218 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7219 free_extent_map(em);
7220 return ERR_PTR(ret);
7221 }
7222
7223 return em;
7224 }
7225
7226 /*
7227 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7228 * block must be cow'd
7229 */
7230 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7231 u64 *orig_start, u64 *orig_block_len,
7232 u64 *ram_bytes)
7233 {
7234 struct btrfs_trans_handle *trans;
7235 struct btrfs_path *path;
7236 int ret;
7237 struct extent_buffer *leaf;
7238 struct btrfs_root *root = BTRFS_I(inode)->root;
7239 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7240 struct btrfs_file_extent_item *fi;
7241 struct btrfs_key key;
7242 u64 disk_bytenr;
7243 u64 backref_offset;
7244 u64 extent_end;
7245 u64 num_bytes;
7246 int slot;
7247 int found_type;
7248 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7249
7250 path = btrfs_alloc_path();
7251 if (!path)
7252 return -ENOMEM;
7253
7254 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7255 offset, 0);
7256 if (ret < 0)
7257 goto out;
7258
7259 slot = path->slots[0];
7260 if (ret == 1) {
7261 if (slot == 0) {
7262 /* can't find the item, must cow */
7263 ret = 0;
7264 goto out;
7265 }
7266 slot--;
7267 }
7268 ret = 0;
7269 leaf = path->nodes[0];
7270 btrfs_item_key_to_cpu(leaf, &key, slot);
7271 if (key.objectid != btrfs_ino(inode) ||
7272 key.type != BTRFS_EXTENT_DATA_KEY) {
7273 /* not our file or wrong item type, must cow */
7274 goto out;
7275 }
7276
7277 if (key.offset > offset) {
7278 /* Wrong offset, must cow */
7279 goto out;
7280 }
7281
7282 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7283 found_type = btrfs_file_extent_type(leaf, fi);
7284 if (found_type != BTRFS_FILE_EXTENT_REG &&
7285 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7286 /* not a regular extent, must cow */
7287 goto out;
7288 }
7289
7290 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7291 goto out;
7292
7293 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7294 if (extent_end <= offset)
7295 goto out;
7296
7297 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7298 if (disk_bytenr == 0)
7299 goto out;
7300
7301 if (btrfs_file_extent_compression(leaf, fi) ||
7302 btrfs_file_extent_encryption(leaf, fi) ||
7303 btrfs_file_extent_other_encoding(leaf, fi))
7304 goto out;
7305
7306 backref_offset = btrfs_file_extent_offset(leaf, fi);
7307
7308 if (orig_start) {
7309 *orig_start = key.offset - backref_offset;
7310 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7311 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7312 }
7313
7314 if (btrfs_extent_readonly(root, disk_bytenr))
7315 goto out;
7316
7317 num_bytes = min(offset + *len, extent_end) - offset;
7318 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7319 u64 range_end;
7320
7321 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7322 ret = test_range_bit(io_tree, offset, range_end,
7323 EXTENT_DELALLOC, 0, NULL);
7324 if (ret) {
7325 ret = -EAGAIN;
7326 goto out;
7327 }
7328 }
7329
7330 btrfs_release_path(path);
7331
7332 /*
7333 * look for other files referencing this extent, if we
7334 * find any we must cow
7335 */
7336 trans = btrfs_join_transaction(root);
7337 if (IS_ERR(trans)) {
7338 ret = 0;
7339 goto out;
7340 }
7341
7342 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7343 key.offset - backref_offset, disk_bytenr);
7344 btrfs_end_transaction(trans, root);
7345 if (ret) {
7346 ret = 0;
7347 goto out;
7348 }
7349
7350 /*
7351 * adjust disk_bytenr and num_bytes to cover just the bytes
7352 * in this extent we are about to write. If there
7353 * are any csums in that range we have to cow in order
7354 * to keep the csums correct
7355 */
7356 disk_bytenr += backref_offset;
7357 disk_bytenr += offset - key.offset;
7358 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7359 goto out;
7360 /*
7361 * all of the above have passed, it is safe to overwrite this extent
7362 * without cow
7363 */
7364 *len = num_bytes;
7365 ret = 1;
7366 out:
7367 btrfs_free_path(path);
7368 return ret;
7369 }
7370
7371 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7372 {
7373 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7374 int found = false;
7375 void **pagep = NULL;
7376 struct page *page = NULL;
7377 unsigned long start_idx;
7378 unsigned long end_idx;
7379
7380 start_idx = start >> PAGE_CACHE_SHIFT;
7381
7382 /*
7383 * end is the last byte in the last page. end == start is legal
7384 */
7385 end_idx = end >> PAGE_CACHE_SHIFT;
7386
7387 rcu_read_lock();
7388
7389 /* Most of the code in this while loop is lifted from
7390 * find_get_page. It's been modified to begin searching from a
7391 * page and return just the first page found in that range. If the
7392 * found idx is less than or equal to the end idx then we know that
7393 * a page exists. If no pages are found or if those pages are
7394 * outside of the range then we're fine (yay!) */
7395 while (page == NULL &&
7396 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7397 page = radix_tree_deref_slot(pagep);
7398 if (unlikely(!page))
7399 break;
7400
7401 if (radix_tree_exception(page)) {
7402 if (radix_tree_deref_retry(page)) {
7403 page = NULL;
7404 continue;
7405 }
7406 /*
7407 * Otherwise, shmem/tmpfs must be storing a swap entry
7408 * here as an exceptional entry: so return it without
7409 * attempting to raise page count.
7410 */
7411 page = NULL;
7412 break; /* TODO: Is this relevant for this use case? */
7413 }
7414
7415 if (!page_cache_get_speculative(page)) {
7416 page = NULL;
7417 continue;
7418 }
7419
7420 /*
7421 * Has the page moved?
7422 * This is part of the lockless pagecache protocol. See
7423 * include/linux/pagemap.h for details.
7424 */
7425 if (unlikely(page != *pagep)) {
7426 page_cache_release(page);
7427 page = NULL;
7428 }
7429 }
7430
7431 if (page) {
7432 if (page->index <= end_idx)
7433 found = true;
7434 page_cache_release(page);
7435 }
7436
7437 rcu_read_unlock();
7438 return found;
7439 }
7440
7441 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7442 struct extent_state **cached_state, int writing)
7443 {
7444 struct btrfs_ordered_extent *ordered;
7445 int ret = 0;
7446
7447 while (1) {
7448 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7449 0, cached_state);
7450 /*
7451 * We're concerned with the entire range that we're going to be
7452 * doing DIO to, so we need to make sure theres no ordered
7453 * extents in this range.
7454 */
7455 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7456 lockend - lockstart + 1);
7457
7458 /*
7459 * We need to make sure there are no buffered pages in this
7460 * range either, we could have raced between the invalidate in
7461 * generic_file_direct_write and locking the extent. The
7462 * invalidate needs to happen so that reads after a write do not
7463 * get stale data.
7464 */
7465 if (!ordered &&
7466 (!writing ||
7467 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7468 break;
7469
7470 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7471 cached_state, GFP_NOFS);
7472
7473 if (ordered) {
7474 btrfs_start_ordered_extent(inode, ordered, 1);
7475 btrfs_put_ordered_extent(ordered);
7476 } else {
7477 /* Screw you mmap */
7478 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7479 if (ret)
7480 break;
7481 ret = filemap_fdatawait_range(inode->i_mapping,
7482 lockstart,
7483 lockend);
7484 if (ret)
7485 break;
7486
7487 /*
7488 * If we found a page that couldn't be invalidated just
7489 * fall back to buffered.
7490 */
7491 ret = invalidate_inode_pages2_range(inode->i_mapping,
7492 lockstart >> PAGE_CACHE_SHIFT,
7493 lockend >> PAGE_CACHE_SHIFT);
7494 if (ret)
7495 break;
7496 }
7497
7498 cond_resched();
7499 }
7500
7501 return ret;
7502 }
7503
7504 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7505 u64 len, u64 orig_start,
7506 u64 block_start, u64 block_len,
7507 u64 orig_block_len, u64 ram_bytes,
7508 int type)
7509 {
7510 struct extent_map_tree *em_tree;
7511 struct extent_map *em;
7512 struct btrfs_root *root = BTRFS_I(inode)->root;
7513 int ret;
7514
7515 em_tree = &BTRFS_I(inode)->extent_tree;
7516 em = alloc_extent_map();
7517 if (!em)
7518 return ERR_PTR(-ENOMEM);
7519
7520 em->start = start;
7521 em->orig_start = orig_start;
7522 em->mod_start = start;
7523 em->mod_len = len;
7524 em->len = len;
7525 em->block_len = block_len;
7526 em->block_start = block_start;
7527 em->bdev = root->fs_info->fs_devices->latest_bdev;
7528 em->orig_block_len = orig_block_len;
7529 em->ram_bytes = ram_bytes;
7530 em->generation = -1;
7531 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7532 if (type == BTRFS_ORDERED_PREALLOC)
7533 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7534
7535 do {
7536 btrfs_drop_extent_cache(inode, em->start,
7537 em->start + em->len - 1, 0);
7538 write_lock(&em_tree->lock);
7539 ret = add_extent_mapping(em_tree, em, 1);
7540 write_unlock(&em_tree->lock);
7541 } while (ret == -EEXIST);
7542
7543 if (ret) {
7544 free_extent_map(em);
7545 return ERR_PTR(ret);
7546 }
7547
7548 return em;
7549 }
7550
7551 struct btrfs_dio_data {
7552 u64 outstanding_extents;
7553 u64 reserve;
7554 };
7555
7556 static void adjust_dio_outstanding_extents(struct inode *inode,
7557 struct btrfs_dio_data *dio_data,
7558 const u64 len)
7559 {
7560 unsigned num_extents;
7561
7562 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7563 BTRFS_MAX_EXTENT_SIZE);
7564 /*
7565 * If we have an outstanding_extents count still set then we're
7566 * within our reservation, otherwise we need to adjust our inode
7567 * counter appropriately.
7568 */
7569 if (dio_data->outstanding_extents >= num_extents) {
7570 dio_data->outstanding_extents -= num_extents;
7571 } else {
7572 /*
7573 * If dio write length has been split due to no large enough
7574 * contiguous space, we need to compensate our inode counter
7575 * appropriately.
7576 */
7577 u64 num_needed = num_extents - dio_data->outstanding_extents;
7578
7579 spin_lock(&BTRFS_I(inode)->lock);
7580 BTRFS_I(inode)->outstanding_extents += num_needed;
7581 spin_unlock(&BTRFS_I(inode)->lock);
7582 }
7583 }
7584
7585 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7586 struct buffer_head *bh_result, int create)
7587 {
7588 struct extent_map *em;
7589 struct btrfs_root *root = BTRFS_I(inode)->root;
7590 struct extent_state *cached_state = NULL;
7591 struct btrfs_dio_data *dio_data = NULL;
7592 u64 start = iblock << inode->i_blkbits;
7593 u64 lockstart, lockend;
7594 u64 len = bh_result->b_size;
7595 int unlock_bits = EXTENT_LOCKED;
7596 int ret = 0;
7597
7598 if (create)
7599 unlock_bits |= EXTENT_DIRTY;
7600 else
7601 len = min_t(u64, len, root->sectorsize);
7602
7603 lockstart = start;
7604 lockend = start + len - 1;
7605
7606 if (current->journal_info) {
7607 /*
7608 * Need to pull our outstanding extents and set journal_info to NULL so
7609 * that anything that needs to check if there's a transction doesn't get
7610 * confused.
7611 */
7612 dio_data = current->journal_info;
7613 current->journal_info = NULL;
7614 }
7615
7616 /*
7617 * If this errors out it's because we couldn't invalidate pagecache for
7618 * this range and we need to fallback to buffered.
7619 */
7620 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7621 create)) {
7622 ret = -ENOTBLK;
7623 goto err;
7624 }
7625
7626 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7627 if (IS_ERR(em)) {
7628 ret = PTR_ERR(em);
7629 goto unlock_err;
7630 }
7631
7632 /*
7633 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7634 * io. INLINE is special, and we could probably kludge it in here, but
7635 * it's still buffered so for safety lets just fall back to the generic
7636 * buffered path.
7637 *
7638 * For COMPRESSED we _have_ to read the entire extent in so we can
7639 * decompress it, so there will be buffering required no matter what we
7640 * do, so go ahead and fallback to buffered.
7641 *
7642 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7643 * to buffered IO. Don't blame me, this is the price we pay for using
7644 * the generic code.
7645 */
7646 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7647 em->block_start == EXTENT_MAP_INLINE) {
7648 free_extent_map(em);
7649 ret = -ENOTBLK;
7650 goto unlock_err;
7651 }
7652
7653 /* Just a good old fashioned hole, return */
7654 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7655 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7656 free_extent_map(em);
7657 goto unlock_err;
7658 }
7659
7660 /*
7661 * We don't allocate a new extent in the following cases
7662 *
7663 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7664 * existing extent.
7665 * 2) The extent is marked as PREALLOC. We're good to go here and can
7666 * just use the extent.
7667 *
7668 */
7669 if (!create) {
7670 len = min(len, em->len - (start - em->start));
7671 lockstart = start + len;
7672 goto unlock;
7673 }
7674
7675 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7676 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7677 em->block_start != EXTENT_MAP_HOLE)) {
7678 int type;
7679 u64 block_start, orig_start, orig_block_len, ram_bytes;
7680
7681 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7682 type = BTRFS_ORDERED_PREALLOC;
7683 else
7684 type = BTRFS_ORDERED_NOCOW;
7685 len = min(len, em->len - (start - em->start));
7686 block_start = em->block_start + (start - em->start);
7687
7688 if (can_nocow_extent(inode, start, &len, &orig_start,
7689 &orig_block_len, &ram_bytes) == 1) {
7690 if (type == BTRFS_ORDERED_PREALLOC) {
7691 free_extent_map(em);
7692 em = create_pinned_em(inode, start, len,
7693 orig_start,
7694 block_start, len,
7695 orig_block_len,
7696 ram_bytes, type);
7697 if (IS_ERR(em)) {
7698 ret = PTR_ERR(em);
7699 goto unlock_err;
7700 }
7701 }
7702
7703 ret = btrfs_add_ordered_extent_dio(inode, start,
7704 block_start, len, len, type);
7705 if (ret) {
7706 free_extent_map(em);
7707 goto unlock_err;
7708 }
7709 goto unlock;
7710 }
7711 }
7712
7713 /*
7714 * this will cow the extent, reset the len in case we changed
7715 * it above
7716 */
7717 len = bh_result->b_size;
7718 free_extent_map(em);
7719 em = btrfs_new_extent_direct(inode, start, len);
7720 if (IS_ERR(em)) {
7721 ret = PTR_ERR(em);
7722 goto unlock_err;
7723 }
7724 len = min(len, em->len - (start - em->start));
7725 unlock:
7726 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7727 inode->i_blkbits;
7728 bh_result->b_size = len;
7729 bh_result->b_bdev = em->bdev;
7730 set_buffer_mapped(bh_result);
7731 if (create) {
7732 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7733 set_buffer_new(bh_result);
7734
7735 /*
7736 * Need to update the i_size under the extent lock so buffered
7737 * readers will get the updated i_size when we unlock.
7738 */
7739 if (start + len > i_size_read(inode))
7740 i_size_write(inode, start + len);
7741
7742 adjust_dio_outstanding_extents(inode, dio_data, len);
7743 btrfs_free_reserved_data_space(inode, start, len);
7744 WARN_ON(dio_data->reserve < len);
7745 dio_data->reserve -= len;
7746 current->journal_info = dio_data;
7747 }
7748
7749 /*
7750 * In the case of write we need to clear and unlock the entire range,
7751 * in the case of read we need to unlock only the end area that we
7752 * aren't using if there is any left over space.
7753 */
7754 if (lockstart < lockend) {
7755 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7756 lockend, unlock_bits, 1, 0,
7757 &cached_state, GFP_NOFS);
7758 } else {
7759 free_extent_state(cached_state);
7760 }
7761
7762 free_extent_map(em);
7763
7764 return 0;
7765
7766 unlock_err:
7767 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7768 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7769 err:
7770 if (dio_data)
7771 current->journal_info = dio_data;
7772 /*
7773 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7774 * write less data then expected, so that we don't underflow our inode's
7775 * outstanding extents counter.
7776 */
7777 if (create && dio_data)
7778 adjust_dio_outstanding_extents(inode, dio_data, len);
7779
7780 return ret;
7781 }
7782
7783 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7784 int rw, int mirror_num)
7785 {
7786 struct btrfs_root *root = BTRFS_I(inode)->root;
7787 int ret;
7788
7789 BUG_ON(rw & REQ_WRITE);
7790
7791 bio_get(bio);
7792
7793 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7794 BTRFS_WQ_ENDIO_DIO_REPAIR);
7795 if (ret)
7796 goto err;
7797
7798 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7799 err:
7800 bio_put(bio);
7801 return ret;
7802 }
7803
7804 static int btrfs_check_dio_repairable(struct inode *inode,
7805 struct bio *failed_bio,
7806 struct io_failure_record *failrec,
7807 int failed_mirror)
7808 {
7809 int num_copies;
7810
7811 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7812 failrec->logical, failrec->len);
7813 if (num_copies == 1) {
7814 /*
7815 * we only have a single copy of the data, so don't bother with
7816 * all the retry and error correction code that follows. no
7817 * matter what the error is, it is very likely to persist.
7818 */
7819 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7820 num_copies, failrec->this_mirror, failed_mirror);
7821 return 0;
7822 }
7823
7824 failrec->failed_mirror = failed_mirror;
7825 failrec->this_mirror++;
7826 if (failrec->this_mirror == failed_mirror)
7827 failrec->this_mirror++;
7828
7829 if (failrec->this_mirror > num_copies) {
7830 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7831 num_copies, failrec->this_mirror, failed_mirror);
7832 return 0;
7833 }
7834
7835 return 1;
7836 }
7837
7838 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7839 struct page *page, u64 start, u64 end,
7840 int failed_mirror, bio_end_io_t *repair_endio,
7841 void *repair_arg)
7842 {
7843 struct io_failure_record *failrec;
7844 struct bio *bio;
7845 int isector;
7846 int read_mode;
7847 int ret;
7848
7849 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7850
7851 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7852 if (ret)
7853 return ret;
7854
7855 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7856 failed_mirror);
7857 if (!ret) {
7858 free_io_failure(inode, failrec);
7859 return -EIO;
7860 }
7861
7862 if (failed_bio->bi_vcnt > 1)
7863 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7864 else
7865 read_mode = READ_SYNC;
7866
7867 isector = start - btrfs_io_bio(failed_bio)->logical;
7868 isector >>= inode->i_sb->s_blocksize_bits;
7869 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7870 0, isector, repair_endio, repair_arg);
7871 if (!bio) {
7872 free_io_failure(inode, failrec);
7873 return -EIO;
7874 }
7875
7876 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7877 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7878 read_mode, failrec->this_mirror, failrec->in_validation);
7879
7880 ret = submit_dio_repair_bio(inode, bio, read_mode,
7881 failrec->this_mirror);
7882 if (ret) {
7883 free_io_failure(inode, failrec);
7884 bio_put(bio);
7885 }
7886
7887 return ret;
7888 }
7889
7890 struct btrfs_retry_complete {
7891 struct completion done;
7892 struct inode *inode;
7893 u64 start;
7894 int uptodate;
7895 };
7896
7897 static void btrfs_retry_endio_nocsum(struct bio *bio)
7898 {
7899 struct btrfs_retry_complete *done = bio->bi_private;
7900 struct bio_vec *bvec;
7901 int i;
7902
7903 if (bio->bi_error)
7904 goto end;
7905
7906 done->uptodate = 1;
7907 bio_for_each_segment_all(bvec, bio, i)
7908 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7909 end:
7910 complete(&done->done);
7911 bio_put(bio);
7912 }
7913
7914 static int __btrfs_correct_data_nocsum(struct inode *inode,
7915 struct btrfs_io_bio *io_bio)
7916 {
7917 struct bio_vec *bvec;
7918 struct btrfs_retry_complete done;
7919 u64 start;
7920 int i;
7921 int ret;
7922
7923 start = io_bio->logical;
7924 done.inode = inode;
7925
7926 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7927 try_again:
7928 done.uptodate = 0;
7929 done.start = start;
7930 init_completion(&done.done);
7931
7932 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7933 start + bvec->bv_len - 1,
7934 io_bio->mirror_num,
7935 btrfs_retry_endio_nocsum, &done);
7936 if (ret)
7937 return ret;
7938
7939 wait_for_completion(&done.done);
7940
7941 if (!done.uptodate) {
7942 /* We might have another mirror, so try again */
7943 goto try_again;
7944 }
7945
7946 start += bvec->bv_len;
7947 }
7948
7949 return 0;
7950 }
7951
7952 static void btrfs_retry_endio(struct bio *bio)
7953 {
7954 struct btrfs_retry_complete *done = bio->bi_private;
7955 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7956 struct bio_vec *bvec;
7957 int uptodate;
7958 int ret;
7959 int i;
7960
7961 if (bio->bi_error)
7962 goto end;
7963
7964 uptodate = 1;
7965 bio_for_each_segment_all(bvec, bio, i) {
7966 ret = __readpage_endio_check(done->inode, io_bio, i,
7967 bvec->bv_page, 0,
7968 done->start, bvec->bv_len);
7969 if (!ret)
7970 clean_io_failure(done->inode, done->start,
7971 bvec->bv_page, 0);
7972 else
7973 uptodate = 0;
7974 }
7975
7976 done->uptodate = uptodate;
7977 end:
7978 complete(&done->done);
7979 bio_put(bio);
7980 }
7981
7982 static int __btrfs_subio_endio_read(struct inode *inode,
7983 struct btrfs_io_bio *io_bio, int err)
7984 {
7985 struct bio_vec *bvec;
7986 struct btrfs_retry_complete done;
7987 u64 start;
7988 u64 offset = 0;
7989 int i;
7990 int ret;
7991
7992 err = 0;
7993 start = io_bio->logical;
7994 done.inode = inode;
7995
7996 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7997 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7998 0, start, bvec->bv_len);
7999 if (likely(!ret))
8000 goto next;
8001 try_again:
8002 done.uptodate = 0;
8003 done.start = start;
8004 init_completion(&done.done);
8005
8006 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
8007 start + bvec->bv_len - 1,
8008 io_bio->mirror_num,
8009 btrfs_retry_endio, &done);
8010 if (ret) {
8011 err = ret;
8012 goto next;
8013 }
8014
8015 wait_for_completion(&done.done);
8016
8017 if (!done.uptodate) {
8018 /* We might have another mirror, so try again */
8019 goto try_again;
8020 }
8021 next:
8022 offset += bvec->bv_len;
8023 start += bvec->bv_len;
8024 }
8025
8026 return err;
8027 }
8028
8029 static int btrfs_subio_endio_read(struct inode *inode,
8030 struct btrfs_io_bio *io_bio, int err)
8031 {
8032 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8033
8034 if (skip_csum) {
8035 if (unlikely(err))
8036 return __btrfs_correct_data_nocsum(inode, io_bio);
8037 else
8038 return 0;
8039 } else {
8040 return __btrfs_subio_endio_read(inode, io_bio, err);
8041 }
8042 }
8043
8044 static void btrfs_endio_direct_read(struct bio *bio)
8045 {
8046 struct btrfs_dio_private *dip = bio->bi_private;
8047 struct inode *inode = dip->inode;
8048 struct bio *dio_bio;
8049 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8050 int err = bio->bi_error;
8051
8052 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8053 err = btrfs_subio_endio_read(inode, io_bio, err);
8054
8055 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8056 dip->logical_offset + dip->bytes - 1);
8057 dio_bio = dip->dio_bio;
8058
8059 kfree(dip);
8060
8061 dio_bio->bi_error = bio->bi_error;
8062 dio_end_io(dio_bio, bio->bi_error);
8063
8064 if (io_bio->end_io)
8065 io_bio->end_io(io_bio, err);
8066 bio_put(bio);
8067 }
8068
8069 static void btrfs_endio_direct_write(struct bio *bio)
8070 {
8071 struct btrfs_dio_private *dip = bio->bi_private;
8072 struct inode *inode = dip->inode;
8073 struct btrfs_root *root = BTRFS_I(inode)->root;
8074 struct btrfs_ordered_extent *ordered = NULL;
8075 u64 ordered_offset = dip->logical_offset;
8076 u64 ordered_bytes = dip->bytes;
8077 struct bio *dio_bio;
8078 int ret;
8079
8080 again:
8081 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8082 &ordered_offset,
8083 ordered_bytes,
8084 !bio->bi_error);
8085 if (!ret)
8086 goto out_test;
8087
8088 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8089 finish_ordered_fn, NULL, NULL);
8090 btrfs_queue_work(root->fs_info->endio_write_workers,
8091 &ordered->work);
8092 out_test:
8093 /*
8094 * our bio might span multiple ordered extents. If we haven't
8095 * completed the accounting for the whole dio, go back and try again
8096 */
8097 if (ordered_offset < dip->logical_offset + dip->bytes) {
8098 ordered_bytes = dip->logical_offset + dip->bytes -
8099 ordered_offset;
8100 ordered = NULL;
8101 goto again;
8102 }
8103 dio_bio = dip->dio_bio;
8104
8105 kfree(dip);
8106
8107 dio_bio->bi_error = bio->bi_error;
8108 dio_end_io(dio_bio, bio->bi_error);
8109 bio_put(bio);
8110 }
8111
8112 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8113 struct bio *bio, int mirror_num,
8114 unsigned long bio_flags, u64 offset)
8115 {
8116 int ret;
8117 struct btrfs_root *root = BTRFS_I(inode)->root;
8118 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8119 BUG_ON(ret); /* -ENOMEM */
8120 return 0;
8121 }
8122
8123 static void btrfs_end_dio_bio(struct bio *bio)
8124 {
8125 struct btrfs_dio_private *dip = bio->bi_private;
8126 int err = bio->bi_error;
8127
8128 if (err)
8129 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8130 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8131 btrfs_ino(dip->inode), bio->bi_rw,
8132 (unsigned long long)bio->bi_iter.bi_sector,
8133 bio->bi_iter.bi_size, err);
8134
8135 if (dip->subio_endio)
8136 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8137
8138 if (err) {
8139 dip->errors = 1;
8140
8141 /*
8142 * before atomic variable goto zero, we must make sure
8143 * dip->errors is perceived to be set.
8144 */
8145 smp_mb__before_atomic();
8146 }
8147
8148 /* if there are more bios still pending for this dio, just exit */
8149 if (!atomic_dec_and_test(&dip->pending_bios))
8150 goto out;
8151
8152 if (dip->errors) {
8153 bio_io_error(dip->orig_bio);
8154 } else {
8155 dip->dio_bio->bi_error = 0;
8156 bio_endio(dip->orig_bio);
8157 }
8158 out:
8159 bio_put(bio);
8160 }
8161
8162 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8163 u64 first_sector, gfp_t gfp_flags)
8164 {
8165 struct bio *bio;
8166 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8167 if (bio)
8168 bio_associate_current(bio);
8169 return bio;
8170 }
8171
8172 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8173 struct inode *inode,
8174 struct btrfs_dio_private *dip,
8175 struct bio *bio,
8176 u64 file_offset)
8177 {
8178 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8179 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8180 int ret;
8181
8182 /*
8183 * We load all the csum data we need when we submit
8184 * the first bio to reduce the csum tree search and
8185 * contention.
8186 */
8187 if (dip->logical_offset == file_offset) {
8188 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8189 file_offset);
8190 if (ret)
8191 return ret;
8192 }
8193
8194 if (bio == dip->orig_bio)
8195 return 0;
8196
8197 file_offset -= dip->logical_offset;
8198 file_offset >>= inode->i_sb->s_blocksize_bits;
8199 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8200
8201 return 0;
8202 }
8203
8204 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8205 int rw, u64 file_offset, int skip_sum,
8206 int async_submit)
8207 {
8208 struct btrfs_dio_private *dip = bio->bi_private;
8209 int write = rw & REQ_WRITE;
8210 struct btrfs_root *root = BTRFS_I(inode)->root;
8211 int ret;
8212
8213 if (async_submit)
8214 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8215
8216 bio_get(bio);
8217
8218 if (!write) {
8219 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8220 BTRFS_WQ_ENDIO_DATA);
8221 if (ret)
8222 goto err;
8223 }
8224
8225 if (skip_sum)
8226 goto map;
8227
8228 if (write && async_submit) {
8229 ret = btrfs_wq_submit_bio(root->fs_info,
8230 inode, rw, bio, 0, 0,
8231 file_offset,
8232 __btrfs_submit_bio_start_direct_io,
8233 __btrfs_submit_bio_done);
8234 goto err;
8235 } else if (write) {
8236 /*
8237 * If we aren't doing async submit, calculate the csum of the
8238 * bio now.
8239 */
8240 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8241 if (ret)
8242 goto err;
8243 } else {
8244 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8245 file_offset);
8246 if (ret)
8247 goto err;
8248 }
8249 map:
8250 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8251 err:
8252 bio_put(bio);
8253 return ret;
8254 }
8255
8256 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8257 int skip_sum)
8258 {
8259 struct inode *inode = dip->inode;
8260 struct btrfs_root *root = BTRFS_I(inode)->root;
8261 struct bio *bio;
8262 struct bio *orig_bio = dip->orig_bio;
8263 struct bio_vec *bvec = orig_bio->bi_io_vec;
8264 u64 start_sector = orig_bio->bi_iter.bi_sector;
8265 u64 file_offset = dip->logical_offset;
8266 u64 submit_len = 0;
8267 u64 map_length;
8268 int nr_pages = 0;
8269 int ret;
8270 int async_submit = 0;
8271
8272 map_length = orig_bio->bi_iter.bi_size;
8273 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8274 &map_length, NULL, 0);
8275 if (ret)
8276 return -EIO;
8277
8278 if (map_length >= orig_bio->bi_iter.bi_size) {
8279 bio = orig_bio;
8280 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8281 goto submit;
8282 }
8283
8284 /* async crcs make it difficult to collect full stripe writes. */
8285 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8286 async_submit = 0;
8287 else
8288 async_submit = 1;
8289
8290 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8291 if (!bio)
8292 return -ENOMEM;
8293
8294 bio->bi_private = dip;
8295 bio->bi_end_io = btrfs_end_dio_bio;
8296 btrfs_io_bio(bio)->logical = file_offset;
8297 atomic_inc(&dip->pending_bios);
8298
8299 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8300 if (map_length < submit_len + bvec->bv_len ||
8301 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8302 bvec->bv_offset) < bvec->bv_len) {
8303 /*
8304 * inc the count before we submit the bio so
8305 * we know the end IO handler won't happen before
8306 * we inc the count. Otherwise, the dip might get freed
8307 * before we're done setting it up
8308 */
8309 atomic_inc(&dip->pending_bios);
8310 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8311 file_offset, skip_sum,
8312 async_submit);
8313 if (ret) {
8314 bio_put(bio);
8315 atomic_dec(&dip->pending_bios);
8316 goto out_err;
8317 }
8318
8319 start_sector += submit_len >> 9;
8320 file_offset += submit_len;
8321
8322 submit_len = 0;
8323 nr_pages = 0;
8324
8325 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8326 start_sector, GFP_NOFS);
8327 if (!bio)
8328 goto out_err;
8329 bio->bi_private = dip;
8330 bio->bi_end_io = btrfs_end_dio_bio;
8331 btrfs_io_bio(bio)->logical = file_offset;
8332
8333 map_length = orig_bio->bi_iter.bi_size;
8334 ret = btrfs_map_block(root->fs_info, rw,
8335 start_sector << 9,
8336 &map_length, NULL, 0);
8337 if (ret) {
8338 bio_put(bio);
8339 goto out_err;
8340 }
8341 } else {
8342 submit_len += bvec->bv_len;
8343 nr_pages++;
8344 bvec++;
8345 }
8346 }
8347
8348 submit:
8349 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8350 async_submit);
8351 if (!ret)
8352 return 0;
8353
8354 bio_put(bio);
8355 out_err:
8356 dip->errors = 1;
8357 /*
8358 * before atomic variable goto zero, we must
8359 * make sure dip->errors is perceived to be set.
8360 */
8361 smp_mb__before_atomic();
8362 if (atomic_dec_and_test(&dip->pending_bios))
8363 bio_io_error(dip->orig_bio);
8364
8365 /* bio_end_io() will handle error, so we needn't return it */
8366 return 0;
8367 }
8368
8369 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8370 struct inode *inode, loff_t file_offset)
8371 {
8372 struct btrfs_dio_private *dip = NULL;
8373 struct bio *io_bio = NULL;
8374 struct btrfs_io_bio *btrfs_bio;
8375 int skip_sum;
8376 int write = rw & REQ_WRITE;
8377 int ret = 0;
8378
8379 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8380
8381 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8382 if (!io_bio) {
8383 ret = -ENOMEM;
8384 goto free_ordered;
8385 }
8386
8387 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8388 if (!dip) {
8389 ret = -ENOMEM;
8390 goto free_ordered;
8391 }
8392
8393 dip->private = dio_bio->bi_private;
8394 dip->inode = inode;
8395 dip->logical_offset = file_offset;
8396 dip->bytes = dio_bio->bi_iter.bi_size;
8397 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8398 io_bio->bi_private = dip;
8399 dip->orig_bio = io_bio;
8400 dip->dio_bio = dio_bio;
8401 atomic_set(&dip->pending_bios, 0);
8402 btrfs_bio = btrfs_io_bio(io_bio);
8403 btrfs_bio->logical = file_offset;
8404
8405 if (write) {
8406 io_bio->bi_end_io = btrfs_endio_direct_write;
8407 } else {
8408 io_bio->bi_end_io = btrfs_endio_direct_read;
8409 dip->subio_endio = btrfs_subio_endio_read;
8410 }
8411
8412 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8413 if (!ret)
8414 return;
8415
8416 if (btrfs_bio->end_io)
8417 btrfs_bio->end_io(btrfs_bio, ret);
8418
8419 free_ordered:
8420 /*
8421 * If we arrived here it means either we failed to submit the dip
8422 * or we either failed to clone the dio_bio or failed to allocate the
8423 * dip. If we cloned the dio_bio and allocated the dip, we can just
8424 * call bio_endio against our io_bio so that we get proper resource
8425 * cleanup if we fail to submit the dip, otherwise, we must do the
8426 * same as btrfs_endio_direct_[write|read] because we can't call these
8427 * callbacks - they require an allocated dip and a clone of dio_bio.
8428 */
8429 if (io_bio && dip) {
8430 io_bio->bi_error = -EIO;
8431 bio_endio(io_bio);
8432 /*
8433 * The end io callbacks free our dip, do the final put on io_bio
8434 * and all the cleanup and final put for dio_bio (through
8435 * dio_end_io()).
8436 */
8437 dip = NULL;
8438 io_bio = NULL;
8439 } else {
8440 if (write) {
8441 struct btrfs_ordered_extent *ordered;
8442
8443 ordered = btrfs_lookup_ordered_extent(inode,
8444 file_offset);
8445 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8446 /*
8447 * Decrements our ref on the ordered extent and removes
8448 * the ordered extent from the inode's ordered tree,
8449 * doing all the proper resource cleanup such as for the
8450 * reserved space and waking up any waiters for this
8451 * ordered extent (through btrfs_remove_ordered_extent).
8452 */
8453 btrfs_finish_ordered_io(ordered);
8454 } else {
8455 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8456 file_offset + dio_bio->bi_iter.bi_size - 1);
8457 }
8458 dio_bio->bi_error = -EIO;
8459 /*
8460 * Releases and cleans up our dio_bio, no need to bio_put()
8461 * nor bio_endio()/bio_io_error() against dio_bio.
8462 */
8463 dio_end_io(dio_bio, ret);
8464 }
8465 if (io_bio)
8466 bio_put(io_bio);
8467 kfree(dip);
8468 }
8469
8470 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8471 const struct iov_iter *iter, loff_t offset)
8472 {
8473 int seg;
8474 int i;
8475 unsigned blocksize_mask = root->sectorsize - 1;
8476 ssize_t retval = -EINVAL;
8477
8478 if (offset & blocksize_mask)
8479 goto out;
8480
8481 if (iov_iter_alignment(iter) & blocksize_mask)
8482 goto out;
8483
8484 /* If this is a write we don't need to check anymore */
8485 if (iov_iter_rw(iter) == WRITE)
8486 return 0;
8487 /*
8488 * Check to make sure we don't have duplicate iov_base's in this
8489 * iovec, if so return EINVAL, otherwise we'll get csum errors
8490 * when reading back.
8491 */
8492 for (seg = 0; seg < iter->nr_segs; seg++) {
8493 for (i = seg + 1; i < iter->nr_segs; i++) {
8494 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8495 goto out;
8496 }
8497 }
8498 retval = 0;
8499 out:
8500 return retval;
8501 }
8502
8503 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8504 loff_t offset)
8505 {
8506 struct file *file = iocb->ki_filp;
8507 struct inode *inode = file->f_mapping->host;
8508 struct btrfs_root *root = BTRFS_I(inode)->root;
8509 struct btrfs_dio_data dio_data = { 0 };
8510 size_t count = 0;
8511 int flags = 0;
8512 bool wakeup = true;
8513 bool relock = false;
8514 ssize_t ret;
8515
8516 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8517 return 0;
8518
8519 inode_dio_begin(inode);
8520 smp_mb__after_atomic();
8521
8522 /*
8523 * The generic stuff only does filemap_write_and_wait_range, which
8524 * isn't enough if we've written compressed pages to this area, so
8525 * we need to flush the dirty pages again to make absolutely sure
8526 * that any outstanding dirty pages are on disk.
8527 */
8528 count = iov_iter_count(iter);
8529 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8530 &BTRFS_I(inode)->runtime_flags))
8531 filemap_fdatawrite_range(inode->i_mapping, offset,
8532 offset + count - 1);
8533
8534 if (iov_iter_rw(iter) == WRITE) {
8535 /*
8536 * If the write DIO is beyond the EOF, we need update
8537 * the isize, but it is protected by i_mutex. So we can
8538 * not unlock the i_mutex at this case.
8539 */
8540 if (offset + count <= inode->i_size) {
8541 mutex_unlock(&inode->i_mutex);
8542 relock = true;
8543 }
8544 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8545 if (ret)
8546 goto out;
8547 dio_data.outstanding_extents = div64_u64(count +
8548 BTRFS_MAX_EXTENT_SIZE - 1,
8549 BTRFS_MAX_EXTENT_SIZE);
8550
8551 /*
8552 * We need to know how many extents we reserved so that we can
8553 * do the accounting properly if we go over the number we
8554 * originally calculated. Abuse current->journal_info for this.
8555 */
8556 dio_data.reserve = round_up(count, root->sectorsize);
8557 current->journal_info = &dio_data;
8558 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8559 &BTRFS_I(inode)->runtime_flags)) {
8560 inode_dio_end(inode);
8561 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8562 wakeup = false;
8563 }
8564
8565 ret = __blockdev_direct_IO(iocb, inode,
8566 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8567 iter, offset, btrfs_get_blocks_direct, NULL,
8568 btrfs_submit_direct, flags);
8569 if (iov_iter_rw(iter) == WRITE) {
8570 current->journal_info = NULL;
8571 if (ret < 0 && ret != -EIOCBQUEUED) {
8572 if (dio_data.reserve)
8573 btrfs_delalloc_release_space(inode, offset,
8574 dio_data.reserve);
8575 } else if (ret >= 0 && (size_t)ret < count)
8576 btrfs_delalloc_release_space(inode, offset,
8577 count - (size_t)ret);
8578 }
8579 out:
8580 if (wakeup)
8581 inode_dio_end(inode);
8582 if (relock)
8583 mutex_lock(&inode->i_mutex);
8584
8585 return ret;
8586 }
8587
8588 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8589
8590 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8591 __u64 start, __u64 len)
8592 {
8593 int ret;
8594
8595 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8596 if (ret)
8597 return ret;
8598
8599 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8600 }
8601
8602 int btrfs_readpage(struct file *file, struct page *page)
8603 {
8604 struct extent_io_tree *tree;
8605 tree = &BTRFS_I(page->mapping->host)->io_tree;
8606 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8607 }
8608
8609 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8610 {
8611 struct extent_io_tree *tree;
8612 struct inode *inode = page->mapping->host;
8613 int ret;
8614
8615 if (current->flags & PF_MEMALLOC) {
8616 redirty_page_for_writepage(wbc, page);
8617 unlock_page(page);
8618 return 0;
8619 }
8620
8621 /*
8622 * If we are under memory pressure we will call this directly from the
8623 * VM, we need to make sure we have the inode referenced for the ordered
8624 * extent. If not just return like we didn't do anything.
8625 */
8626 if (!igrab(inode)) {
8627 redirty_page_for_writepage(wbc, page);
8628 return AOP_WRITEPAGE_ACTIVATE;
8629 }
8630 tree = &BTRFS_I(page->mapping->host)->io_tree;
8631 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8632 btrfs_add_delayed_iput(inode);
8633 return ret;
8634 }
8635
8636 static int btrfs_writepages(struct address_space *mapping,
8637 struct writeback_control *wbc)
8638 {
8639 struct extent_io_tree *tree;
8640
8641 tree = &BTRFS_I(mapping->host)->io_tree;
8642 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8643 }
8644
8645 static int
8646 btrfs_readpages(struct file *file, struct address_space *mapping,
8647 struct list_head *pages, unsigned nr_pages)
8648 {
8649 struct extent_io_tree *tree;
8650 tree = &BTRFS_I(mapping->host)->io_tree;
8651 return extent_readpages(tree, mapping, pages, nr_pages,
8652 btrfs_get_extent);
8653 }
8654 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8655 {
8656 struct extent_io_tree *tree;
8657 struct extent_map_tree *map;
8658 int ret;
8659
8660 tree = &BTRFS_I(page->mapping->host)->io_tree;
8661 map = &BTRFS_I(page->mapping->host)->extent_tree;
8662 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8663 if (ret == 1) {
8664 ClearPagePrivate(page);
8665 set_page_private(page, 0);
8666 page_cache_release(page);
8667 }
8668 return ret;
8669 }
8670
8671 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8672 {
8673 if (PageWriteback(page) || PageDirty(page))
8674 return 0;
8675 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8676 }
8677
8678 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8679 unsigned int length)
8680 {
8681 struct inode *inode = page->mapping->host;
8682 struct extent_io_tree *tree;
8683 struct btrfs_ordered_extent *ordered;
8684 struct extent_state *cached_state = NULL;
8685 u64 page_start = page_offset(page);
8686 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8687 int inode_evicting = inode->i_state & I_FREEING;
8688
8689 /*
8690 * we have the page locked, so new writeback can't start,
8691 * and the dirty bit won't be cleared while we are here.
8692 *
8693 * Wait for IO on this page so that we can safely clear
8694 * the PagePrivate2 bit and do ordered accounting
8695 */
8696 wait_on_page_writeback(page);
8697
8698 tree = &BTRFS_I(inode)->io_tree;
8699 if (offset) {
8700 btrfs_releasepage(page, GFP_NOFS);
8701 return;
8702 }
8703
8704 if (!inode_evicting)
8705 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8706 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8707 if (ordered) {
8708 /*
8709 * IO on this page will never be started, so we need
8710 * to account for any ordered extents now
8711 */
8712 if (!inode_evicting)
8713 clear_extent_bit(tree, page_start, page_end,
8714 EXTENT_DIRTY | EXTENT_DELALLOC |
8715 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8716 EXTENT_DEFRAG, 1, 0, &cached_state,
8717 GFP_NOFS);
8718 /*
8719 * whoever cleared the private bit is responsible
8720 * for the finish_ordered_io
8721 */
8722 if (TestClearPagePrivate2(page)) {
8723 struct btrfs_ordered_inode_tree *tree;
8724 u64 new_len;
8725
8726 tree = &BTRFS_I(inode)->ordered_tree;
8727
8728 spin_lock_irq(&tree->lock);
8729 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8730 new_len = page_start - ordered->file_offset;
8731 if (new_len < ordered->truncated_len)
8732 ordered->truncated_len = new_len;
8733 spin_unlock_irq(&tree->lock);
8734
8735 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8736 page_start,
8737 PAGE_CACHE_SIZE, 1))
8738 btrfs_finish_ordered_io(ordered);
8739 }
8740 btrfs_put_ordered_extent(ordered);
8741 if (!inode_evicting) {
8742 cached_state = NULL;
8743 lock_extent_bits(tree, page_start, page_end, 0,
8744 &cached_state);
8745 }
8746 }
8747
8748 /*
8749 * Qgroup reserved space handler
8750 * Page here will be either
8751 * 1) Already written to disk
8752 * In this case, its reserved space is released from data rsv map
8753 * and will be freed by delayed_ref handler finally.
8754 * So even we call qgroup_free_data(), it won't decrease reserved
8755 * space.
8756 * 2) Not written to disk
8757 * This means the reserved space should be freed here. However,
8758 * if a truncate invalidates the page (by clearing PageDirty)
8759 * and the page is accounted for while allocating extent
8760 * in btrfs_check_data_free_space() we let delayed_ref to
8761 * free the entire extent.
8762 */
8763 if (PageDirty(page))
8764 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8765 if (!inode_evicting) {
8766 clear_extent_bit(tree, page_start, page_end,
8767 EXTENT_LOCKED | EXTENT_DIRTY |
8768 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8769 EXTENT_DEFRAG, 1, 1,
8770 &cached_state, GFP_NOFS);
8771
8772 __btrfs_releasepage(page, GFP_NOFS);
8773 }
8774
8775 ClearPageChecked(page);
8776 if (PagePrivate(page)) {
8777 ClearPagePrivate(page);
8778 set_page_private(page, 0);
8779 page_cache_release(page);
8780 }
8781 }
8782
8783 /*
8784 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8785 * called from a page fault handler when a page is first dirtied. Hence we must
8786 * be careful to check for EOF conditions here. We set the page up correctly
8787 * for a written page which means we get ENOSPC checking when writing into
8788 * holes and correct delalloc and unwritten extent mapping on filesystems that
8789 * support these features.
8790 *
8791 * We are not allowed to take the i_mutex here so we have to play games to
8792 * protect against truncate races as the page could now be beyond EOF. Because
8793 * vmtruncate() writes the inode size before removing pages, once we have the
8794 * page lock we can determine safely if the page is beyond EOF. If it is not
8795 * beyond EOF, then the page is guaranteed safe against truncation until we
8796 * unlock the page.
8797 */
8798 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8799 {
8800 struct page *page = vmf->page;
8801 struct inode *inode = file_inode(vma->vm_file);
8802 struct btrfs_root *root = BTRFS_I(inode)->root;
8803 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8804 struct btrfs_ordered_extent *ordered;
8805 struct extent_state *cached_state = NULL;
8806 char *kaddr;
8807 unsigned long zero_start;
8808 loff_t size;
8809 int ret;
8810 int reserved = 0;
8811 u64 page_start;
8812 u64 page_end;
8813
8814 sb_start_pagefault(inode->i_sb);
8815 page_start = page_offset(page);
8816 page_end = page_start + PAGE_CACHE_SIZE - 1;
8817
8818 ret = btrfs_delalloc_reserve_space(inode, page_start,
8819 PAGE_CACHE_SIZE);
8820 if (!ret) {
8821 ret = file_update_time(vma->vm_file);
8822 reserved = 1;
8823 }
8824 if (ret) {
8825 if (ret == -ENOMEM)
8826 ret = VM_FAULT_OOM;
8827 else /* -ENOSPC, -EIO, etc */
8828 ret = VM_FAULT_SIGBUS;
8829 if (reserved)
8830 goto out;
8831 goto out_noreserve;
8832 }
8833
8834 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8835 again:
8836 lock_page(page);
8837 size = i_size_read(inode);
8838
8839 if ((page->mapping != inode->i_mapping) ||
8840 (page_start >= size)) {
8841 /* page got truncated out from underneath us */
8842 goto out_unlock;
8843 }
8844 wait_on_page_writeback(page);
8845
8846 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8847 set_page_extent_mapped(page);
8848
8849 /*
8850 * we can't set the delalloc bits if there are pending ordered
8851 * extents. Drop our locks and wait for them to finish
8852 */
8853 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8854 if (ordered) {
8855 unlock_extent_cached(io_tree, page_start, page_end,
8856 &cached_state, GFP_NOFS);
8857 unlock_page(page);
8858 btrfs_start_ordered_extent(inode, ordered, 1);
8859 btrfs_put_ordered_extent(ordered);
8860 goto again;
8861 }
8862
8863 /*
8864 * XXX - page_mkwrite gets called every time the page is dirtied, even
8865 * if it was already dirty, so for space accounting reasons we need to
8866 * clear any delalloc bits for the range we are fixing to save. There
8867 * is probably a better way to do this, but for now keep consistent with
8868 * prepare_pages in the normal write path.
8869 */
8870 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8871 EXTENT_DIRTY | EXTENT_DELALLOC |
8872 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8873 0, 0, &cached_state, GFP_NOFS);
8874
8875 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8876 &cached_state);
8877 if (ret) {
8878 unlock_extent_cached(io_tree, page_start, page_end,
8879 &cached_state, GFP_NOFS);
8880 ret = VM_FAULT_SIGBUS;
8881 goto out_unlock;
8882 }
8883 ret = 0;
8884
8885 /* page is wholly or partially inside EOF */
8886 if (page_start + PAGE_CACHE_SIZE > size)
8887 zero_start = size & ~PAGE_CACHE_MASK;
8888 else
8889 zero_start = PAGE_CACHE_SIZE;
8890
8891 if (zero_start != PAGE_CACHE_SIZE) {
8892 kaddr = kmap(page);
8893 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8894 flush_dcache_page(page);
8895 kunmap(page);
8896 }
8897 ClearPageChecked(page);
8898 set_page_dirty(page);
8899 SetPageUptodate(page);
8900
8901 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8902 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8903 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8904
8905 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8906
8907 out_unlock:
8908 if (!ret) {
8909 sb_end_pagefault(inode->i_sb);
8910 return VM_FAULT_LOCKED;
8911 }
8912 unlock_page(page);
8913 out:
8914 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8915 out_noreserve:
8916 sb_end_pagefault(inode->i_sb);
8917 return ret;
8918 }
8919
8920 static int btrfs_truncate(struct inode *inode)
8921 {
8922 struct btrfs_root *root = BTRFS_I(inode)->root;
8923 struct btrfs_block_rsv *rsv;
8924 int ret = 0;
8925 int err = 0;
8926 struct btrfs_trans_handle *trans;
8927 u64 mask = root->sectorsize - 1;
8928 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8929
8930 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8931 (u64)-1);
8932 if (ret)
8933 return ret;
8934
8935 /*
8936 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8937 * 3 things going on here
8938 *
8939 * 1) We need to reserve space for our orphan item and the space to
8940 * delete our orphan item. Lord knows we don't want to have a dangling
8941 * orphan item because we didn't reserve space to remove it.
8942 *
8943 * 2) We need to reserve space to update our inode.
8944 *
8945 * 3) We need to have something to cache all the space that is going to
8946 * be free'd up by the truncate operation, but also have some slack
8947 * space reserved in case it uses space during the truncate (thank you
8948 * very much snapshotting).
8949 *
8950 * And we need these to all be seperate. The fact is we can use alot of
8951 * space doing the truncate, and we have no earthly idea how much space
8952 * we will use, so we need the truncate reservation to be seperate so it
8953 * doesn't end up using space reserved for updating the inode or
8954 * removing the orphan item. We also need to be able to stop the
8955 * transaction and start a new one, which means we need to be able to
8956 * update the inode several times, and we have no idea of knowing how
8957 * many times that will be, so we can't just reserve 1 item for the
8958 * entirety of the opration, so that has to be done seperately as well.
8959 * Then there is the orphan item, which does indeed need to be held on
8960 * to for the whole operation, and we need nobody to touch this reserved
8961 * space except the orphan code.
8962 *
8963 * So that leaves us with
8964 *
8965 * 1) root->orphan_block_rsv - for the orphan deletion.
8966 * 2) rsv - for the truncate reservation, which we will steal from the
8967 * transaction reservation.
8968 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8969 * updating the inode.
8970 */
8971 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8972 if (!rsv)
8973 return -ENOMEM;
8974 rsv->size = min_size;
8975 rsv->failfast = 1;
8976
8977 /*
8978 * 1 for the truncate slack space
8979 * 1 for updating the inode.
8980 */
8981 trans = btrfs_start_transaction(root, 2);
8982 if (IS_ERR(trans)) {
8983 err = PTR_ERR(trans);
8984 goto out;
8985 }
8986
8987 /* Migrate the slack space for the truncate to our reserve */
8988 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8989 min_size);
8990 BUG_ON(ret);
8991
8992 /*
8993 * So if we truncate and then write and fsync we normally would just
8994 * write the extents that changed, which is a problem if we need to
8995 * first truncate that entire inode. So set this flag so we write out
8996 * all of the extents in the inode to the sync log so we're completely
8997 * safe.
8998 */
8999 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9000 trans->block_rsv = rsv;
9001
9002 while (1) {
9003 ret = btrfs_truncate_inode_items(trans, root, inode,
9004 inode->i_size,
9005 BTRFS_EXTENT_DATA_KEY);
9006 if (ret != -ENOSPC && ret != -EAGAIN) {
9007 err = ret;
9008 break;
9009 }
9010
9011 trans->block_rsv = &root->fs_info->trans_block_rsv;
9012 ret = btrfs_update_inode(trans, root, inode);
9013 if (ret) {
9014 err = ret;
9015 break;
9016 }
9017
9018 btrfs_end_transaction(trans, root);
9019 btrfs_btree_balance_dirty(root);
9020
9021 trans = btrfs_start_transaction(root, 2);
9022 if (IS_ERR(trans)) {
9023 ret = err = PTR_ERR(trans);
9024 trans = NULL;
9025 break;
9026 }
9027
9028 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9029 rsv, min_size);
9030 BUG_ON(ret); /* shouldn't happen */
9031 trans->block_rsv = rsv;
9032 }
9033
9034 if (ret == 0 && inode->i_nlink > 0) {
9035 trans->block_rsv = root->orphan_block_rsv;
9036 ret = btrfs_orphan_del(trans, inode);
9037 if (ret)
9038 err = ret;
9039 }
9040
9041 if (trans) {
9042 trans->block_rsv = &root->fs_info->trans_block_rsv;
9043 ret = btrfs_update_inode(trans, root, inode);
9044 if (ret && !err)
9045 err = ret;
9046
9047 ret = btrfs_end_transaction(trans, root);
9048 btrfs_btree_balance_dirty(root);
9049 }
9050
9051 out:
9052 btrfs_free_block_rsv(root, rsv);
9053
9054 if (ret && !err)
9055 err = ret;
9056
9057 return err;
9058 }
9059
9060 /*
9061 * create a new subvolume directory/inode (helper for the ioctl).
9062 */
9063 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9064 struct btrfs_root *new_root,
9065 struct btrfs_root *parent_root,
9066 u64 new_dirid)
9067 {
9068 struct inode *inode;
9069 int err;
9070 u64 index = 0;
9071
9072 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9073 new_dirid, new_dirid,
9074 S_IFDIR | (~current_umask() & S_IRWXUGO),
9075 &index);
9076 if (IS_ERR(inode))
9077 return PTR_ERR(inode);
9078 inode->i_op = &btrfs_dir_inode_operations;
9079 inode->i_fop = &btrfs_dir_file_operations;
9080
9081 set_nlink(inode, 1);
9082 btrfs_i_size_write(inode, 0);
9083 unlock_new_inode(inode);
9084
9085 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9086 if (err)
9087 btrfs_err(new_root->fs_info,
9088 "error inheriting subvolume %llu properties: %d",
9089 new_root->root_key.objectid, err);
9090
9091 err = btrfs_update_inode(trans, new_root, inode);
9092
9093 iput(inode);
9094 return err;
9095 }
9096
9097 struct inode *btrfs_alloc_inode(struct super_block *sb)
9098 {
9099 struct btrfs_inode *ei;
9100 struct inode *inode;
9101
9102 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9103 if (!ei)
9104 return NULL;
9105
9106 ei->root = NULL;
9107 ei->generation = 0;
9108 ei->last_trans = 0;
9109 ei->last_sub_trans = 0;
9110 ei->logged_trans = 0;
9111 ei->delalloc_bytes = 0;
9112 ei->defrag_bytes = 0;
9113 ei->disk_i_size = 0;
9114 ei->flags = 0;
9115 ei->csum_bytes = 0;
9116 ei->index_cnt = (u64)-1;
9117 ei->dir_index = 0;
9118 ei->last_unlink_trans = 0;
9119 ei->last_log_commit = 0;
9120
9121 spin_lock_init(&ei->lock);
9122 ei->outstanding_extents = 0;
9123 ei->reserved_extents = 0;
9124
9125 ei->runtime_flags = 0;
9126 ei->force_compress = BTRFS_COMPRESS_NONE;
9127
9128 ei->delayed_node = NULL;
9129
9130 ei->i_otime.tv_sec = 0;
9131 ei->i_otime.tv_nsec = 0;
9132
9133 inode = &ei->vfs_inode;
9134 extent_map_tree_init(&ei->extent_tree);
9135 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9136 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9137 ei->io_tree.track_uptodate = 1;
9138 ei->io_failure_tree.track_uptodate = 1;
9139 atomic_set(&ei->sync_writers, 0);
9140 mutex_init(&ei->log_mutex);
9141 mutex_init(&ei->delalloc_mutex);
9142 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9143 INIT_LIST_HEAD(&ei->delalloc_inodes);
9144 RB_CLEAR_NODE(&ei->rb_node);
9145
9146 return inode;
9147 }
9148
9149 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9150 void btrfs_test_destroy_inode(struct inode *inode)
9151 {
9152 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9153 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9154 }
9155 #endif
9156
9157 static void btrfs_i_callback(struct rcu_head *head)
9158 {
9159 struct inode *inode = container_of(head, struct inode, i_rcu);
9160 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9161 }
9162
9163 void btrfs_destroy_inode(struct inode *inode)
9164 {
9165 struct btrfs_ordered_extent *ordered;
9166 struct btrfs_root *root = BTRFS_I(inode)->root;
9167
9168 WARN_ON(!hlist_empty(&inode->i_dentry));
9169 WARN_ON(inode->i_data.nrpages);
9170 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9171 WARN_ON(BTRFS_I(inode)->reserved_extents);
9172 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9173 WARN_ON(BTRFS_I(inode)->csum_bytes);
9174 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9175
9176 /*
9177 * This can happen where we create an inode, but somebody else also
9178 * created the same inode and we need to destroy the one we already
9179 * created.
9180 */
9181 if (!root)
9182 goto free;
9183
9184 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9185 &BTRFS_I(inode)->runtime_flags)) {
9186 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9187 btrfs_ino(inode));
9188 atomic_dec(&root->orphan_inodes);
9189 }
9190
9191 while (1) {
9192 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9193 if (!ordered)
9194 break;
9195 else {
9196 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9197 ordered->file_offset, ordered->len);
9198 btrfs_remove_ordered_extent(inode, ordered);
9199 btrfs_put_ordered_extent(ordered);
9200 btrfs_put_ordered_extent(ordered);
9201 }
9202 }
9203 btrfs_qgroup_check_reserved_leak(inode);
9204 inode_tree_del(inode);
9205 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9206 free:
9207 call_rcu(&inode->i_rcu, btrfs_i_callback);
9208 }
9209
9210 int btrfs_drop_inode(struct inode *inode)
9211 {
9212 struct btrfs_root *root = BTRFS_I(inode)->root;
9213
9214 if (root == NULL)
9215 return 1;
9216
9217 /* the snap/subvol tree is on deleting */
9218 if (btrfs_root_refs(&root->root_item) == 0)
9219 return 1;
9220 else
9221 return generic_drop_inode(inode);
9222 }
9223
9224 static void init_once(void *foo)
9225 {
9226 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9227
9228 inode_init_once(&ei->vfs_inode);
9229 }
9230
9231 void btrfs_destroy_cachep(void)
9232 {
9233 /*
9234 * Make sure all delayed rcu free inodes are flushed before we
9235 * destroy cache.
9236 */
9237 rcu_barrier();
9238 if (btrfs_inode_cachep)
9239 kmem_cache_destroy(btrfs_inode_cachep);
9240 if (btrfs_trans_handle_cachep)
9241 kmem_cache_destroy(btrfs_trans_handle_cachep);
9242 if (btrfs_transaction_cachep)
9243 kmem_cache_destroy(btrfs_transaction_cachep);
9244 if (btrfs_path_cachep)
9245 kmem_cache_destroy(btrfs_path_cachep);
9246 if (btrfs_free_space_cachep)
9247 kmem_cache_destroy(btrfs_free_space_cachep);
9248 if (btrfs_delalloc_work_cachep)
9249 kmem_cache_destroy(btrfs_delalloc_work_cachep);
9250 }
9251
9252 int btrfs_init_cachep(void)
9253 {
9254 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9255 sizeof(struct btrfs_inode), 0,
9256 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9257 if (!btrfs_inode_cachep)
9258 goto fail;
9259
9260 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9261 sizeof(struct btrfs_trans_handle), 0,
9262 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9263 if (!btrfs_trans_handle_cachep)
9264 goto fail;
9265
9266 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9267 sizeof(struct btrfs_transaction), 0,
9268 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9269 if (!btrfs_transaction_cachep)
9270 goto fail;
9271
9272 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9273 sizeof(struct btrfs_path), 0,
9274 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9275 if (!btrfs_path_cachep)
9276 goto fail;
9277
9278 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9279 sizeof(struct btrfs_free_space), 0,
9280 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9281 if (!btrfs_free_space_cachep)
9282 goto fail;
9283
9284 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9285 sizeof(struct btrfs_delalloc_work), 0,
9286 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9287 NULL);
9288 if (!btrfs_delalloc_work_cachep)
9289 goto fail;
9290
9291 return 0;
9292 fail:
9293 btrfs_destroy_cachep();
9294 return -ENOMEM;
9295 }
9296
9297 static int btrfs_getattr(struct vfsmount *mnt,
9298 struct dentry *dentry, struct kstat *stat)
9299 {
9300 u64 delalloc_bytes;
9301 struct inode *inode = d_inode(dentry);
9302 u32 blocksize = inode->i_sb->s_blocksize;
9303
9304 generic_fillattr(inode, stat);
9305 stat->dev = BTRFS_I(inode)->root->anon_dev;
9306 stat->blksize = PAGE_CACHE_SIZE;
9307
9308 spin_lock(&BTRFS_I(inode)->lock);
9309 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9310 spin_unlock(&BTRFS_I(inode)->lock);
9311 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9312 ALIGN(delalloc_bytes, blocksize)) >> 9;
9313 return 0;
9314 }
9315
9316 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9317 struct inode *new_dir, struct dentry *new_dentry)
9318 {
9319 struct btrfs_trans_handle *trans;
9320 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9321 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9322 struct inode *new_inode = d_inode(new_dentry);
9323 struct inode *old_inode = d_inode(old_dentry);
9324 struct timespec ctime = CURRENT_TIME;
9325 u64 index = 0;
9326 u64 root_objectid;
9327 int ret;
9328 u64 old_ino = btrfs_ino(old_inode);
9329
9330 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9331 return -EPERM;
9332
9333 /* we only allow rename subvolume link between subvolumes */
9334 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9335 return -EXDEV;
9336
9337 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9338 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9339 return -ENOTEMPTY;
9340
9341 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9342 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9343 return -ENOTEMPTY;
9344
9345
9346 /* check for collisions, even if the name isn't there */
9347 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9348 new_dentry->d_name.name,
9349 new_dentry->d_name.len);
9350
9351 if (ret) {
9352 if (ret == -EEXIST) {
9353 /* we shouldn't get
9354 * eexist without a new_inode */
9355 if (WARN_ON(!new_inode)) {
9356 return ret;
9357 }
9358 } else {
9359 /* maybe -EOVERFLOW */
9360 return ret;
9361 }
9362 }
9363 ret = 0;
9364
9365 /*
9366 * we're using rename to replace one file with another. Start IO on it
9367 * now so we don't add too much work to the end of the transaction
9368 */
9369 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9370 filemap_flush(old_inode->i_mapping);
9371
9372 /* close the racy window with snapshot create/destroy ioctl */
9373 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9374 down_read(&root->fs_info->subvol_sem);
9375 /*
9376 * We want to reserve the absolute worst case amount of items. So if
9377 * both inodes are subvols and we need to unlink them then that would
9378 * require 4 item modifications, but if they are both normal inodes it
9379 * would require 5 item modifications, so we'll assume their normal
9380 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9381 * should cover the worst case number of items we'll modify.
9382 */
9383 trans = btrfs_start_transaction(root, 11);
9384 if (IS_ERR(trans)) {
9385 ret = PTR_ERR(trans);
9386 goto out_notrans;
9387 }
9388
9389 if (dest != root)
9390 btrfs_record_root_in_trans(trans, dest);
9391
9392 ret = btrfs_set_inode_index(new_dir, &index);
9393 if (ret)
9394 goto out_fail;
9395
9396 BTRFS_I(old_inode)->dir_index = 0ULL;
9397 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9398 /* force full log commit if subvolume involved. */
9399 btrfs_set_log_full_commit(root->fs_info, trans);
9400 } else {
9401 ret = btrfs_insert_inode_ref(trans, dest,
9402 new_dentry->d_name.name,
9403 new_dentry->d_name.len,
9404 old_ino,
9405 btrfs_ino(new_dir), index);
9406 if (ret)
9407 goto out_fail;
9408 /*
9409 * this is an ugly little race, but the rename is required
9410 * to make sure that if we crash, the inode is either at the
9411 * old name or the new one. pinning the log transaction lets
9412 * us make sure we don't allow a log commit to come in after
9413 * we unlink the name but before we add the new name back in.
9414 */
9415 btrfs_pin_log_trans(root);
9416 }
9417
9418 inode_inc_iversion(old_dir);
9419 inode_inc_iversion(new_dir);
9420 inode_inc_iversion(old_inode);
9421 old_dir->i_ctime = old_dir->i_mtime = ctime;
9422 new_dir->i_ctime = new_dir->i_mtime = ctime;
9423 old_inode->i_ctime = ctime;
9424
9425 if (old_dentry->d_parent != new_dentry->d_parent)
9426 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9427
9428 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9429 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9430 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9431 old_dentry->d_name.name,
9432 old_dentry->d_name.len);
9433 } else {
9434 ret = __btrfs_unlink_inode(trans, root, old_dir,
9435 d_inode(old_dentry),
9436 old_dentry->d_name.name,
9437 old_dentry->d_name.len);
9438 if (!ret)
9439 ret = btrfs_update_inode(trans, root, old_inode);
9440 }
9441 if (ret) {
9442 btrfs_abort_transaction(trans, root, ret);
9443 goto out_fail;
9444 }
9445
9446 if (new_inode) {
9447 inode_inc_iversion(new_inode);
9448 new_inode->i_ctime = CURRENT_TIME;
9449 if (unlikely(btrfs_ino(new_inode) ==
9450 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9451 root_objectid = BTRFS_I(new_inode)->location.objectid;
9452 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9453 root_objectid,
9454 new_dentry->d_name.name,
9455 new_dentry->d_name.len);
9456 BUG_ON(new_inode->i_nlink == 0);
9457 } else {
9458 ret = btrfs_unlink_inode(trans, dest, new_dir,
9459 d_inode(new_dentry),
9460 new_dentry->d_name.name,
9461 new_dentry->d_name.len);
9462 }
9463 if (!ret && new_inode->i_nlink == 0)
9464 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9465 if (ret) {
9466 btrfs_abort_transaction(trans, root, ret);
9467 goto out_fail;
9468 }
9469 }
9470
9471 ret = btrfs_add_link(trans, new_dir, old_inode,
9472 new_dentry->d_name.name,
9473 new_dentry->d_name.len, 0, index);
9474 if (ret) {
9475 btrfs_abort_transaction(trans, root, ret);
9476 goto out_fail;
9477 }
9478
9479 if (old_inode->i_nlink == 1)
9480 BTRFS_I(old_inode)->dir_index = index;
9481
9482 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9483 struct dentry *parent = new_dentry->d_parent;
9484 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9485 btrfs_end_log_trans(root);
9486 }
9487 out_fail:
9488 btrfs_end_transaction(trans, root);
9489 out_notrans:
9490 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9491 up_read(&root->fs_info->subvol_sem);
9492
9493 return ret;
9494 }
9495
9496 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9497 struct inode *new_dir, struct dentry *new_dentry,
9498 unsigned int flags)
9499 {
9500 if (flags & ~RENAME_NOREPLACE)
9501 return -EINVAL;
9502
9503 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9504 }
9505
9506 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9507 {
9508 struct btrfs_delalloc_work *delalloc_work;
9509 struct inode *inode;
9510
9511 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9512 work);
9513 inode = delalloc_work->inode;
9514 if (delalloc_work->wait) {
9515 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9516 } else {
9517 filemap_flush(inode->i_mapping);
9518 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9519 &BTRFS_I(inode)->runtime_flags))
9520 filemap_flush(inode->i_mapping);
9521 }
9522
9523 if (delalloc_work->delay_iput)
9524 btrfs_add_delayed_iput(inode);
9525 else
9526 iput(inode);
9527 complete(&delalloc_work->completion);
9528 }
9529
9530 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9531 int wait, int delay_iput)
9532 {
9533 struct btrfs_delalloc_work *work;
9534
9535 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9536 if (!work)
9537 return NULL;
9538
9539 init_completion(&work->completion);
9540 INIT_LIST_HEAD(&work->list);
9541 work->inode = inode;
9542 work->wait = wait;
9543 work->delay_iput = delay_iput;
9544 WARN_ON_ONCE(!inode);
9545 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9546 btrfs_run_delalloc_work, NULL, NULL);
9547
9548 return work;
9549 }
9550
9551 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9552 {
9553 wait_for_completion(&work->completion);
9554 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9555 }
9556
9557 /*
9558 * some fairly slow code that needs optimization. This walks the list
9559 * of all the inodes with pending delalloc and forces them to disk.
9560 */
9561 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9562 int nr)
9563 {
9564 struct btrfs_inode *binode;
9565 struct inode *inode;
9566 struct btrfs_delalloc_work *work, *next;
9567 struct list_head works;
9568 struct list_head splice;
9569 int ret = 0;
9570
9571 INIT_LIST_HEAD(&works);
9572 INIT_LIST_HEAD(&splice);
9573
9574 mutex_lock(&root->delalloc_mutex);
9575 spin_lock(&root->delalloc_lock);
9576 list_splice_init(&root->delalloc_inodes, &splice);
9577 while (!list_empty(&splice)) {
9578 binode = list_entry(splice.next, struct btrfs_inode,
9579 delalloc_inodes);
9580
9581 list_move_tail(&binode->delalloc_inodes,
9582 &root->delalloc_inodes);
9583 inode = igrab(&binode->vfs_inode);
9584 if (!inode) {
9585 cond_resched_lock(&root->delalloc_lock);
9586 continue;
9587 }
9588 spin_unlock(&root->delalloc_lock);
9589
9590 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9591 if (!work) {
9592 if (delay_iput)
9593 btrfs_add_delayed_iput(inode);
9594 else
9595 iput(inode);
9596 ret = -ENOMEM;
9597 goto out;
9598 }
9599 list_add_tail(&work->list, &works);
9600 btrfs_queue_work(root->fs_info->flush_workers,
9601 &work->work);
9602 ret++;
9603 if (nr != -1 && ret >= nr)
9604 goto out;
9605 cond_resched();
9606 spin_lock(&root->delalloc_lock);
9607 }
9608 spin_unlock(&root->delalloc_lock);
9609
9610 out:
9611 list_for_each_entry_safe(work, next, &works, list) {
9612 list_del_init(&work->list);
9613 btrfs_wait_and_free_delalloc_work(work);
9614 }
9615
9616 if (!list_empty_careful(&splice)) {
9617 spin_lock(&root->delalloc_lock);
9618 list_splice_tail(&splice, &root->delalloc_inodes);
9619 spin_unlock(&root->delalloc_lock);
9620 }
9621 mutex_unlock(&root->delalloc_mutex);
9622 return ret;
9623 }
9624
9625 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9626 {
9627 int ret;
9628
9629 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9630 return -EROFS;
9631
9632 ret = __start_delalloc_inodes(root, delay_iput, -1);
9633 if (ret > 0)
9634 ret = 0;
9635 /*
9636 * the filemap_flush will queue IO into the worker threads, but
9637 * we have to make sure the IO is actually started and that
9638 * ordered extents get created before we return
9639 */
9640 atomic_inc(&root->fs_info->async_submit_draining);
9641 while (atomic_read(&root->fs_info->nr_async_submits) ||
9642 atomic_read(&root->fs_info->async_delalloc_pages)) {
9643 wait_event(root->fs_info->async_submit_wait,
9644 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9645 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9646 }
9647 atomic_dec(&root->fs_info->async_submit_draining);
9648 return ret;
9649 }
9650
9651 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9652 int nr)
9653 {
9654 struct btrfs_root *root;
9655 struct list_head splice;
9656 int ret;
9657
9658 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9659 return -EROFS;
9660
9661 INIT_LIST_HEAD(&splice);
9662
9663 mutex_lock(&fs_info->delalloc_root_mutex);
9664 spin_lock(&fs_info->delalloc_root_lock);
9665 list_splice_init(&fs_info->delalloc_roots, &splice);
9666 while (!list_empty(&splice) && nr) {
9667 root = list_first_entry(&splice, struct btrfs_root,
9668 delalloc_root);
9669 root = btrfs_grab_fs_root(root);
9670 BUG_ON(!root);
9671 list_move_tail(&root->delalloc_root,
9672 &fs_info->delalloc_roots);
9673 spin_unlock(&fs_info->delalloc_root_lock);
9674
9675 ret = __start_delalloc_inodes(root, delay_iput, nr);
9676 btrfs_put_fs_root(root);
9677 if (ret < 0)
9678 goto out;
9679
9680 if (nr != -1) {
9681 nr -= ret;
9682 WARN_ON(nr < 0);
9683 }
9684 spin_lock(&fs_info->delalloc_root_lock);
9685 }
9686 spin_unlock(&fs_info->delalloc_root_lock);
9687
9688 ret = 0;
9689 atomic_inc(&fs_info->async_submit_draining);
9690 while (atomic_read(&fs_info->nr_async_submits) ||
9691 atomic_read(&fs_info->async_delalloc_pages)) {
9692 wait_event(fs_info->async_submit_wait,
9693 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9694 atomic_read(&fs_info->async_delalloc_pages) == 0));
9695 }
9696 atomic_dec(&fs_info->async_submit_draining);
9697 out:
9698 if (!list_empty_careful(&splice)) {
9699 spin_lock(&fs_info->delalloc_root_lock);
9700 list_splice_tail(&splice, &fs_info->delalloc_roots);
9701 spin_unlock(&fs_info->delalloc_root_lock);
9702 }
9703 mutex_unlock(&fs_info->delalloc_root_mutex);
9704 return ret;
9705 }
9706
9707 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9708 const char *symname)
9709 {
9710 struct btrfs_trans_handle *trans;
9711 struct btrfs_root *root = BTRFS_I(dir)->root;
9712 struct btrfs_path *path;
9713 struct btrfs_key key;
9714 struct inode *inode = NULL;
9715 int err;
9716 int drop_inode = 0;
9717 u64 objectid;
9718 u64 index = 0;
9719 int name_len;
9720 int datasize;
9721 unsigned long ptr;
9722 struct btrfs_file_extent_item *ei;
9723 struct extent_buffer *leaf;
9724
9725 name_len = strlen(symname);
9726 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9727 return -ENAMETOOLONG;
9728
9729 /*
9730 * 2 items for inode item and ref
9731 * 2 items for dir items
9732 * 1 item for updating parent inode item
9733 * 1 item for the inline extent item
9734 * 1 item for xattr if selinux is on
9735 */
9736 trans = btrfs_start_transaction(root, 7);
9737 if (IS_ERR(trans))
9738 return PTR_ERR(trans);
9739
9740 err = btrfs_find_free_ino(root, &objectid);
9741 if (err)
9742 goto out_unlock;
9743
9744 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9745 dentry->d_name.len, btrfs_ino(dir), objectid,
9746 S_IFLNK|S_IRWXUGO, &index);
9747 if (IS_ERR(inode)) {
9748 err = PTR_ERR(inode);
9749 goto out_unlock;
9750 }
9751
9752 /*
9753 * If the active LSM wants to access the inode during
9754 * d_instantiate it needs these. Smack checks to see
9755 * if the filesystem supports xattrs by looking at the
9756 * ops vector.
9757 */
9758 inode->i_fop = &btrfs_file_operations;
9759 inode->i_op = &btrfs_file_inode_operations;
9760 inode->i_mapping->a_ops = &btrfs_aops;
9761 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9762
9763 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9764 if (err)
9765 goto out_unlock_inode;
9766
9767 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9768 if (err)
9769 goto out_unlock_inode;
9770
9771 path = btrfs_alloc_path();
9772 if (!path) {
9773 err = -ENOMEM;
9774 goto out_unlock_inode;
9775 }
9776 key.objectid = btrfs_ino(inode);
9777 key.offset = 0;
9778 key.type = BTRFS_EXTENT_DATA_KEY;
9779 datasize = btrfs_file_extent_calc_inline_size(name_len);
9780 err = btrfs_insert_empty_item(trans, root, path, &key,
9781 datasize);
9782 if (err) {
9783 btrfs_free_path(path);
9784 goto out_unlock_inode;
9785 }
9786 leaf = path->nodes[0];
9787 ei = btrfs_item_ptr(leaf, path->slots[0],
9788 struct btrfs_file_extent_item);
9789 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9790 btrfs_set_file_extent_type(leaf, ei,
9791 BTRFS_FILE_EXTENT_INLINE);
9792 btrfs_set_file_extent_encryption(leaf, ei, 0);
9793 btrfs_set_file_extent_compression(leaf, ei, 0);
9794 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9795 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9796
9797 ptr = btrfs_file_extent_inline_start(ei);
9798 write_extent_buffer(leaf, symname, ptr, name_len);
9799 btrfs_mark_buffer_dirty(leaf);
9800 btrfs_free_path(path);
9801
9802 inode->i_op = &btrfs_symlink_inode_operations;
9803 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9804 inode_set_bytes(inode, name_len);
9805 btrfs_i_size_write(inode, name_len);
9806 err = btrfs_update_inode(trans, root, inode);
9807 if (err) {
9808 drop_inode = 1;
9809 goto out_unlock_inode;
9810 }
9811
9812 d_instantiate_new(dentry, inode);
9813
9814 out_unlock:
9815 btrfs_end_transaction(trans, root);
9816 if (drop_inode) {
9817 inode_dec_link_count(inode);
9818 iput(inode);
9819 }
9820 btrfs_btree_balance_dirty(root);
9821 return err;
9822
9823 out_unlock_inode:
9824 drop_inode = 1;
9825 unlock_new_inode(inode);
9826 goto out_unlock;
9827 }
9828
9829 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9830 u64 start, u64 num_bytes, u64 min_size,
9831 loff_t actual_len, u64 *alloc_hint,
9832 struct btrfs_trans_handle *trans)
9833 {
9834 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9835 struct extent_map *em;
9836 struct btrfs_root *root = BTRFS_I(inode)->root;
9837 struct btrfs_key ins;
9838 u64 cur_offset = start;
9839 u64 i_size;
9840 u64 cur_bytes;
9841 u64 last_alloc = (u64)-1;
9842 int ret = 0;
9843 bool own_trans = true;
9844
9845 if (trans)
9846 own_trans = false;
9847 while (num_bytes > 0) {
9848 if (own_trans) {
9849 trans = btrfs_start_transaction(root, 3);
9850 if (IS_ERR(trans)) {
9851 ret = PTR_ERR(trans);
9852 break;
9853 }
9854 }
9855
9856 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9857 cur_bytes = max(cur_bytes, min_size);
9858 /*
9859 * If we are severely fragmented we could end up with really
9860 * small allocations, so if the allocator is returning small
9861 * chunks lets make its job easier by only searching for those
9862 * sized chunks.
9863 */
9864 cur_bytes = min(cur_bytes, last_alloc);
9865 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9866 *alloc_hint, &ins, 1, 0);
9867 if (ret) {
9868 if (own_trans)
9869 btrfs_end_transaction(trans, root);
9870 break;
9871 }
9872
9873 last_alloc = ins.offset;
9874 ret = insert_reserved_file_extent(trans, inode,
9875 cur_offset, ins.objectid,
9876 ins.offset, ins.offset,
9877 ins.offset, 0, 0, 0,
9878 BTRFS_FILE_EXTENT_PREALLOC);
9879 if (ret) {
9880 btrfs_free_reserved_extent(root, ins.objectid,
9881 ins.offset, 0);
9882 btrfs_abort_transaction(trans, root, ret);
9883 if (own_trans)
9884 btrfs_end_transaction(trans, root);
9885 break;
9886 }
9887
9888 btrfs_drop_extent_cache(inode, cur_offset,
9889 cur_offset + ins.offset -1, 0);
9890
9891 em = alloc_extent_map();
9892 if (!em) {
9893 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9894 &BTRFS_I(inode)->runtime_flags);
9895 goto next;
9896 }
9897
9898 em->start = cur_offset;
9899 em->orig_start = cur_offset;
9900 em->len = ins.offset;
9901 em->block_start = ins.objectid;
9902 em->block_len = ins.offset;
9903 em->orig_block_len = ins.offset;
9904 em->ram_bytes = ins.offset;
9905 em->bdev = root->fs_info->fs_devices->latest_bdev;
9906 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9907 em->generation = trans->transid;
9908
9909 while (1) {
9910 write_lock(&em_tree->lock);
9911 ret = add_extent_mapping(em_tree, em, 1);
9912 write_unlock(&em_tree->lock);
9913 if (ret != -EEXIST)
9914 break;
9915 btrfs_drop_extent_cache(inode, cur_offset,
9916 cur_offset + ins.offset - 1,
9917 0);
9918 }
9919 free_extent_map(em);
9920 next:
9921 num_bytes -= ins.offset;
9922 cur_offset += ins.offset;
9923 *alloc_hint = ins.objectid + ins.offset;
9924
9925 inode_inc_iversion(inode);
9926 inode->i_ctime = CURRENT_TIME;
9927 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9928 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9929 (actual_len > inode->i_size) &&
9930 (cur_offset > inode->i_size)) {
9931 if (cur_offset > actual_len)
9932 i_size = actual_len;
9933 else
9934 i_size = cur_offset;
9935 i_size_write(inode, i_size);
9936 btrfs_ordered_update_i_size(inode, i_size, NULL);
9937 }
9938
9939 ret = btrfs_update_inode(trans, root, inode);
9940
9941 if (ret) {
9942 btrfs_abort_transaction(trans, root, ret);
9943 if (own_trans)
9944 btrfs_end_transaction(trans, root);
9945 break;
9946 }
9947
9948 if (own_trans)
9949 btrfs_end_transaction(trans, root);
9950 }
9951 return ret;
9952 }
9953
9954 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9955 u64 start, u64 num_bytes, u64 min_size,
9956 loff_t actual_len, u64 *alloc_hint)
9957 {
9958 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9959 min_size, actual_len, alloc_hint,
9960 NULL);
9961 }
9962
9963 int btrfs_prealloc_file_range_trans(struct inode *inode,
9964 struct btrfs_trans_handle *trans, int mode,
9965 u64 start, u64 num_bytes, u64 min_size,
9966 loff_t actual_len, u64 *alloc_hint)
9967 {
9968 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9969 min_size, actual_len, alloc_hint, trans);
9970 }
9971
9972 static int btrfs_set_page_dirty(struct page *page)
9973 {
9974 return __set_page_dirty_nobuffers(page);
9975 }
9976
9977 static int btrfs_permission(struct inode *inode, int mask)
9978 {
9979 struct btrfs_root *root = BTRFS_I(inode)->root;
9980 umode_t mode = inode->i_mode;
9981
9982 if (mask & MAY_WRITE &&
9983 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9984 if (btrfs_root_readonly(root))
9985 return -EROFS;
9986 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9987 return -EACCES;
9988 }
9989 return generic_permission(inode, mask);
9990 }
9991
9992 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9993 {
9994 struct btrfs_trans_handle *trans;
9995 struct btrfs_root *root = BTRFS_I(dir)->root;
9996 struct inode *inode = NULL;
9997 u64 objectid;
9998 u64 index;
9999 int ret = 0;
10000
10001 /*
10002 * 5 units required for adding orphan entry
10003 */
10004 trans = btrfs_start_transaction(root, 5);
10005 if (IS_ERR(trans))
10006 return PTR_ERR(trans);
10007
10008 ret = btrfs_find_free_ino(root, &objectid);
10009 if (ret)
10010 goto out;
10011
10012 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10013 btrfs_ino(dir), objectid, mode, &index);
10014 if (IS_ERR(inode)) {
10015 ret = PTR_ERR(inode);
10016 inode = NULL;
10017 goto out;
10018 }
10019
10020 inode->i_fop = &btrfs_file_operations;
10021 inode->i_op = &btrfs_file_inode_operations;
10022
10023 inode->i_mapping->a_ops = &btrfs_aops;
10024 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10025
10026 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10027 if (ret)
10028 goto out_inode;
10029
10030 ret = btrfs_update_inode(trans, root, inode);
10031 if (ret)
10032 goto out_inode;
10033 ret = btrfs_orphan_add(trans, inode);
10034 if (ret)
10035 goto out_inode;
10036
10037 /*
10038 * We set number of links to 0 in btrfs_new_inode(), and here we set
10039 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10040 * through:
10041 *
10042 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10043 */
10044 set_nlink(inode, 1);
10045 unlock_new_inode(inode);
10046 d_tmpfile(dentry, inode);
10047 mark_inode_dirty(inode);
10048
10049 out:
10050 btrfs_end_transaction(trans, root);
10051 if (ret)
10052 iput(inode);
10053 btrfs_balance_delayed_items(root);
10054 btrfs_btree_balance_dirty(root);
10055 return ret;
10056
10057 out_inode:
10058 unlock_new_inode(inode);
10059 goto out;
10060
10061 }
10062
10063 /* Inspired by filemap_check_errors() */
10064 int btrfs_inode_check_errors(struct inode *inode)
10065 {
10066 int ret = 0;
10067
10068 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10069 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10070 ret = -ENOSPC;
10071 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10072 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10073 ret = -EIO;
10074
10075 return ret;
10076 }
10077
10078 static const struct inode_operations btrfs_dir_inode_operations = {
10079 .getattr = btrfs_getattr,
10080 .lookup = btrfs_lookup,
10081 .create = btrfs_create,
10082 .unlink = btrfs_unlink,
10083 .link = btrfs_link,
10084 .mkdir = btrfs_mkdir,
10085 .rmdir = btrfs_rmdir,
10086 .rename2 = btrfs_rename2,
10087 .symlink = btrfs_symlink,
10088 .setattr = btrfs_setattr,
10089 .mknod = btrfs_mknod,
10090 .setxattr = btrfs_setxattr,
10091 .getxattr = btrfs_getxattr,
10092 .listxattr = btrfs_listxattr,
10093 .removexattr = btrfs_removexattr,
10094 .permission = btrfs_permission,
10095 .get_acl = btrfs_get_acl,
10096 .set_acl = btrfs_set_acl,
10097 .update_time = btrfs_update_time,
10098 .tmpfile = btrfs_tmpfile,
10099 };
10100 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10101 .lookup = btrfs_lookup,
10102 .permission = btrfs_permission,
10103 .get_acl = btrfs_get_acl,
10104 .set_acl = btrfs_set_acl,
10105 .update_time = btrfs_update_time,
10106 };
10107
10108 static const struct file_operations btrfs_dir_file_operations = {
10109 .llseek = generic_file_llseek,
10110 .read = generic_read_dir,
10111 .iterate = btrfs_real_readdir,
10112 .unlocked_ioctl = btrfs_ioctl,
10113 #ifdef CONFIG_COMPAT
10114 .compat_ioctl = btrfs_ioctl,
10115 #endif
10116 .release = btrfs_release_file,
10117 .fsync = btrfs_sync_file,
10118 };
10119
10120 static struct extent_io_ops btrfs_extent_io_ops = {
10121 .fill_delalloc = run_delalloc_range,
10122 .submit_bio_hook = btrfs_submit_bio_hook,
10123 .merge_bio_hook = btrfs_merge_bio_hook,
10124 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10125 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10126 .writepage_start_hook = btrfs_writepage_start_hook,
10127 .set_bit_hook = btrfs_set_bit_hook,
10128 .clear_bit_hook = btrfs_clear_bit_hook,
10129 .merge_extent_hook = btrfs_merge_extent_hook,
10130 .split_extent_hook = btrfs_split_extent_hook,
10131 };
10132
10133 /*
10134 * btrfs doesn't support the bmap operation because swapfiles
10135 * use bmap to make a mapping of extents in the file. They assume
10136 * these extents won't change over the life of the file and they
10137 * use the bmap result to do IO directly to the drive.
10138 *
10139 * the btrfs bmap call would return logical addresses that aren't
10140 * suitable for IO and they also will change frequently as COW
10141 * operations happen. So, swapfile + btrfs == corruption.
10142 *
10143 * For now we're avoiding this by dropping bmap.
10144 */
10145 static const struct address_space_operations btrfs_aops = {
10146 .readpage = btrfs_readpage,
10147 .writepage = btrfs_writepage,
10148 .writepages = btrfs_writepages,
10149 .readpages = btrfs_readpages,
10150 .direct_IO = btrfs_direct_IO,
10151 .invalidatepage = btrfs_invalidatepage,
10152 .releasepage = btrfs_releasepage,
10153 .set_page_dirty = btrfs_set_page_dirty,
10154 .error_remove_page = generic_error_remove_page,
10155 };
10156
10157 static const struct address_space_operations btrfs_symlink_aops = {
10158 .readpage = btrfs_readpage,
10159 .writepage = btrfs_writepage,
10160 .invalidatepage = btrfs_invalidatepage,
10161 .releasepage = btrfs_releasepage,
10162 };
10163
10164 static const struct inode_operations btrfs_file_inode_operations = {
10165 .getattr = btrfs_getattr,
10166 .setattr = btrfs_setattr,
10167 .setxattr = btrfs_setxattr,
10168 .getxattr = btrfs_getxattr,
10169 .listxattr = btrfs_listxattr,
10170 .removexattr = btrfs_removexattr,
10171 .permission = btrfs_permission,
10172 .fiemap = btrfs_fiemap,
10173 .get_acl = btrfs_get_acl,
10174 .set_acl = btrfs_set_acl,
10175 .update_time = btrfs_update_time,
10176 };
10177 static const struct inode_operations btrfs_special_inode_operations = {
10178 .getattr = btrfs_getattr,
10179 .setattr = btrfs_setattr,
10180 .permission = btrfs_permission,
10181 .setxattr = btrfs_setxattr,
10182 .getxattr = btrfs_getxattr,
10183 .listxattr = btrfs_listxattr,
10184 .removexattr = btrfs_removexattr,
10185 .get_acl = btrfs_get_acl,
10186 .set_acl = btrfs_set_acl,
10187 .update_time = btrfs_update_time,
10188 };
10189 static const struct inode_operations btrfs_symlink_inode_operations = {
10190 .readlink = generic_readlink,
10191 .follow_link = page_follow_link_light,
10192 .put_link = page_put_link,
10193 .getattr = btrfs_getattr,
10194 .setattr = btrfs_setattr,
10195 .permission = btrfs_permission,
10196 .setxattr = btrfs_setxattr,
10197 .getxattr = btrfs_getxattr,
10198 .listxattr = btrfs_listxattr,
10199 .removexattr = btrfs_removexattr,
10200 .update_time = btrfs_update_time,
10201 };
10202
10203 const struct dentry_operations btrfs_dentry_operations = {
10204 .d_delete = btrfs_dentry_delete,
10205 .d_release = btrfs_dentry_release,
10206 };
Linux with added FPC-III support