]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/inode.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: clear ordered flag on cleaning up ordered extents
[mirror_ubuntu-artful-kernel.git] / fs / btrfs / inode.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
51 #include "xattr.h"
52 #include "tree-log.h"
53 #include "volumes.h"
54 #include "compression.h"
55 #include "locking.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
58 #include "backref.h"
59 #include "hash.h"
60 #include "props.h"
61 #include "qgroup.h"
62 #include "dedupe.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74 int overwrite;
75 };
76
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
86
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
91
92 #define S_SHIFT 12
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101 };
102
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
112 u64 orig_start, u64 block_start,
113 u64 block_len, u64 orig_block_len,
114 u64 ram_bytes, int compress_type,
115 int type);
116
117 static void __endio_write_update_ordered(struct inode *inode,
118 const u64 offset, const u64 bytes,
119 const bool uptodate);
120
121 /*
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
124 *
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
133 */
134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
135 const u64 offset,
136 const u64 bytes)
137 {
138 unsigned long index = offset >> PAGE_SHIFT;
139 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
140 struct page *page;
141
142 while (index <= end_index) {
143 page = find_get_page(inode->i_mapping, index);
144 index++;
145 if (!page)
146 continue;
147 ClearPagePrivate2(page);
148 put_page(page);
149 }
150 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
151 bytes - PAGE_SIZE, false);
152 }
153
154 static int btrfs_dirty_inode(struct inode *inode);
155
156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
157 void btrfs_test_inode_set_ops(struct inode *inode)
158 {
159 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
160 }
161 #endif
162
163 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
164 struct inode *inode, struct inode *dir,
165 const struct qstr *qstr)
166 {
167 int err;
168
169 err = btrfs_init_acl(trans, inode, dir);
170 if (!err)
171 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
172 return err;
173 }
174
175 /*
176 * this does all the hard work for inserting an inline extent into
177 * the btree. The caller should have done a btrfs_drop_extents so that
178 * no overlapping inline items exist in the btree
179 */
180 static int insert_inline_extent(struct btrfs_trans_handle *trans,
181 struct btrfs_path *path, int extent_inserted,
182 struct btrfs_root *root, struct inode *inode,
183 u64 start, size_t size, size_t compressed_size,
184 int compress_type,
185 struct page **compressed_pages)
186 {
187 struct extent_buffer *leaf;
188 struct page *page = NULL;
189 char *kaddr;
190 unsigned long ptr;
191 struct btrfs_file_extent_item *ei;
192 int ret;
193 size_t cur_size = size;
194 unsigned long offset;
195
196 if (compressed_size && compressed_pages)
197 cur_size = compressed_size;
198
199 inode_add_bytes(inode, size);
200
201 if (!extent_inserted) {
202 struct btrfs_key key;
203 size_t datasize;
204
205 key.objectid = btrfs_ino(BTRFS_I(inode));
206 key.offset = start;
207 key.type = BTRFS_EXTENT_DATA_KEY;
208
209 datasize = btrfs_file_extent_calc_inline_size(cur_size);
210 path->leave_spinning = 1;
211 ret = btrfs_insert_empty_item(trans, root, path, &key,
212 datasize);
213 if (ret)
214 goto fail;
215 }
216 leaf = path->nodes[0];
217 ei = btrfs_item_ptr(leaf, path->slots[0],
218 struct btrfs_file_extent_item);
219 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
220 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
221 btrfs_set_file_extent_encryption(leaf, ei, 0);
222 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
223 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
224 ptr = btrfs_file_extent_inline_start(ei);
225
226 if (compress_type != BTRFS_COMPRESS_NONE) {
227 struct page *cpage;
228 int i = 0;
229 while (compressed_size > 0) {
230 cpage = compressed_pages[i];
231 cur_size = min_t(unsigned long, compressed_size,
232 PAGE_SIZE);
233
234 kaddr = kmap_atomic(cpage);
235 write_extent_buffer(leaf, kaddr, ptr, cur_size);
236 kunmap_atomic(kaddr);
237
238 i++;
239 ptr += cur_size;
240 compressed_size -= cur_size;
241 }
242 btrfs_set_file_extent_compression(leaf, ei,
243 compress_type);
244 } else {
245 page = find_get_page(inode->i_mapping,
246 start >> PAGE_SHIFT);
247 btrfs_set_file_extent_compression(leaf, ei, 0);
248 kaddr = kmap_atomic(page);
249 offset = start & (PAGE_SIZE - 1);
250 write_extent_buffer(leaf, kaddr + offset, ptr, size);
251 kunmap_atomic(kaddr);
252 put_page(page);
253 }
254 btrfs_mark_buffer_dirty(leaf);
255 btrfs_release_path(path);
256
257 /*
258 * we're an inline extent, so nobody can
259 * extend the file past i_size without locking
260 * a page we already have locked.
261 *
262 * We must do any isize and inode updates
263 * before we unlock the pages. Otherwise we
264 * could end up racing with unlink.
265 */
266 BTRFS_I(inode)->disk_i_size = inode->i_size;
267 ret = btrfs_update_inode(trans, root, inode);
268
269 fail:
270 return ret;
271 }
272
273
274 /*
275 * conditionally insert an inline extent into the file. This
276 * does the checks required to make sure the data is small enough
277 * to fit as an inline extent.
278 */
279 static noinline int cow_file_range_inline(struct btrfs_root *root,
280 struct inode *inode, u64 start,
281 u64 end, size_t compressed_size,
282 int compress_type,
283 struct page **compressed_pages)
284 {
285 struct btrfs_fs_info *fs_info = root->fs_info;
286 struct btrfs_trans_handle *trans;
287 u64 isize = i_size_read(inode);
288 u64 actual_end = min(end + 1, isize);
289 u64 inline_len = actual_end - start;
290 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
291 u64 data_len = inline_len;
292 int ret;
293 struct btrfs_path *path;
294 int extent_inserted = 0;
295 u32 extent_item_size;
296
297 if (compressed_size)
298 data_len = compressed_size;
299
300 if (start > 0 ||
301 actual_end > fs_info->sectorsize ||
302 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
303 (!compressed_size &&
304 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
305 end + 1 < isize ||
306 data_len > fs_info->max_inline) {
307 return 1;
308 }
309
310 path = btrfs_alloc_path();
311 if (!path)
312 return -ENOMEM;
313
314 trans = btrfs_join_transaction(root);
315 if (IS_ERR(trans)) {
316 btrfs_free_path(path);
317 return PTR_ERR(trans);
318 }
319 trans->block_rsv = &fs_info->delalloc_block_rsv;
320
321 if (compressed_size && compressed_pages)
322 extent_item_size = btrfs_file_extent_calc_inline_size(
323 compressed_size);
324 else
325 extent_item_size = btrfs_file_extent_calc_inline_size(
326 inline_len);
327
328 ret = __btrfs_drop_extents(trans, root, inode, path,
329 start, aligned_end, NULL,
330 1, 1, extent_item_size, &extent_inserted);
331 if (ret) {
332 btrfs_abort_transaction(trans, ret);
333 goto out;
334 }
335
336 if (isize > actual_end)
337 inline_len = min_t(u64, isize, actual_end);
338 ret = insert_inline_extent(trans, path, extent_inserted,
339 root, inode, start,
340 inline_len, compressed_size,
341 compress_type, compressed_pages);
342 if (ret && ret != -ENOSPC) {
343 btrfs_abort_transaction(trans, ret);
344 goto out;
345 } else if (ret == -ENOSPC) {
346 ret = 1;
347 goto out;
348 }
349
350 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
351 btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
352 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
353 out:
354 /*
355 * Don't forget to free the reserved space, as for inlined extent
356 * it won't count as data extent, free them directly here.
357 * And at reserve time, it's always aligned to page size, so
358 * just free one page here.
359 */
360 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
361 btrfs_free_path(path);
362 btrfs_end_transaction(trans);
363 return ret;
364 }
365
366 struct async_extent {
367 u64 start;
368 u64 ram_size;
369 u64 compressed_size;
370 struct page **pages;
371 unsigned long nr_pages;
372 int compress_type;
373 struct list_head list;
374 };
375
376 struct async_cow {
377 struct inode *inode;
378 struct btrfs_root *root;
379 struct page *locked_page;
380 u64 start;
381 u64 end;
382 struct list_head extents;
383 struct btrfs_work work;
384 };
385
386 static noinline int add_async_extent(struct async_cow *cow,
387 u64 start, u64 ram_size,
388 u64 compressed_size,
389 struct page **pages,
390 unsigned long nr_pages,
391 int compress_type)
392 {
393 struct async_extent *async_extent;
394
395 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
396 BUG_ON(!async_extent); /* -ENOMEM */
397 async_extent->start = start;
398 async_extent->ram_size = ram_size;
399 async_extent->compressed_size = compressed_size;
400 async_extent->pages = pages;
401 async_extent->nr_pages = nr_pages;
402 async_extent->compress_type = compress_type;
403 list_add_tail(&async_extent->list, &cow->extents);
404 return 0;
405 }
406
407 static inline int inode_need_compress(struct inode *inode)
408 {
409 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
410
411 /* force compress */
412 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
413 return 1;
414 /* bad compression ratios */
415 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
416 return 0;
417 if (btrfs_test_opt(fs_info, COMPRESS) ||
418 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
419 BTRFS_I(inode)->force_compress)
420 return 1;
421 return 0;
422 }
423
424 static inline void inode_should_defrag(struct btrfs_inode *inode,
425 u64 start, u64 end, u64 num_bytes, u64 small_write)
426 {
427 /* If this is a small write inside eof, kick off a defrag */
428 if (num_bytes < small_write &&
429 (start > 0 || end + 1 < inode->disk_i_size))
430 btrfs_add_inode_defrag(NULL, inode);
431 }
432
433 /*
434 * we create compressed extents in two phases. The first
435 * phase compresses a range of pages that have already been
436 * locked (both pages and state bits are locked).
437 *
438 * This is done inside an ordered work queue, and the compression
439 * is spread across many cpus. The actual IO submission is step
440 * two, and the ordered work queue takes care of making sure that
441 * happens in the same order things were put onto the queue by
442 * writepages and friends.
443 *
444 * If this code finds it can't get good compression, it puts an
445 * entry onto the work queue to write the uncompressed bytes. This
446 * makes sure that both compressed inodes and uncompressed inodes
447 * are written in the same order that the flusher thread sent them
448 * down.
449 */
450 static noinline void compress_file_range(struct inode *inode,
451 struct page *locked_page,
452 u64 start, u64 end,
453 struct async_cow *async_cow,
454 int *num_added)
455 {
456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
457 struct btrfs_root *root = BTRFS_I(inode)->root;
458 u64 num_bytes;
459 u64 blocksize = fs_info->sectorsize;
460 u64 actual_end;
461 u64 isize = i_size_read(inode);
462 int ret = 0;
463 struct page **pages = NULL;
464 unsigned long nr_pages;
465 unsigned long total_compressed = 0;
466 unsigned long total_in = 0;
467 int i;
468 int will_compress;
469 int compress_type = fs_info->compress_type;
470 int redirty = 0;
471
472 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
473 SZ_16K);
474
475 actual_end = min_t(u64, isize, end + 1);
476 again:
477 will_compress = 0;
478 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
479 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
480 nr_pages = min_t(unsigned long, nr_pages,
481 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
482
483 /*
484 * we don't want to send crud past the end of i_size through
485 * compression, that's just a waste of CPU time. So, if the
486 * end of the file is before the start of our current
487 * requested range of bytes, we bail out to the uncompressed
488 * cleanup code that can deal with all of this.
489 *
490 * It isn't really the fastest way to fix things, but this is a
491 * very uncommon corner.
492 */
493 if (actual_end <= start)
494 goto cleanup_and_bail_uncompressed;
495
496 total_compressed = actual_end - start;
497
498 /*
499 * skip compression for a small file range(<=blocksize) that
500 * isn't an inline extent, since it doesn't save disk space at all.
501 */
502 if (total_compressed <= blocksize &&
503 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
504 goto cleanup_and_bail_uncompressed;
505
506 total_compressed = min_t(unsigned long, total_compressed,
507 BTRFS_MAX_UNCOMPRESSED);
508 num_bytes = ALIGN(end - start + 1, blocksize);
509 num_bytes = max(blocksize, num_bytes);
510 total_in = 0;
511 ret = 0;
512
513 /*
514 * we do compression for mount -o compress and when the
515 * inode has not been flagged as nocompress. This flag can
516 * change at any time if we discover bad compression ratios.
517 */
518 if (inode_need_compress(inode)) {
519 WARN_ON(pages);
520 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
521 if (!pages) {
522 /* just bail out to the uncompressed code */
523 goto cont;
524 }
525
526 if (BTRFS_I(inode)->force_compress)
527 compress_type = BTRFS_I(inode)->force_compress;
528
529 /*
530 * we need to call clear_page_dirty_for_io on each
531 * page in the range. Otherwise applications with the file
532 * mmap'd can wander in and change the page contents while
533 * we are compressing them.
534 *
535 * If the compression fails for any reason, we set the pages
536 * dirty again later on.
537 */
538 extent_range_clear_dirty_for_io(inode, start, end);
539 redirty = 1;
540 ret = btrfs_compress_pages(compress_type,
541 inode->i_mapping, start,
542 pages,
543 &nr_pages,
544 &total_in,
545 &total_compressed);
546
547 if (!ret) {
548 unsigned long offset = total_compressed &
549 (PAGE_SIZE - 1);
550 struct page *page = pages[nr_pages - 1];
551 char *kaddr;
552
553 /* zero the tail end of the last page, we might be
554 * sending it down to disk
555 */
556 if (offset) {
557 kaddr = kmap_atomic(page);
558 memset(kaddr + offset, 0,
559 PAGE_SIZE - offset);
560 kunmap_atomic(kaddr);
561 }
562 will_compress = 1;
563 }
564 }
565 cont:
566 if (start == 0) {
567 /* lets try to make an inline extent */
568 if (ret || total_in < (actual_end - start)) {
569 /* we didn't compress the entire range, try
570 * to make an uncompressed inline extent.
571 */
572 ret = cow_file_range_inline(root, inode, start, end,
573 0, BTRFS_COMPRESS_NONE, NULL);
574 } else {
575 /* try making a compressed inline extent */
576 ret = cow_file_range_inline(root, inode, start, end,
577 total_compressed,
578 compress_type, pages);
579 }
580 if (ret <= 0) {
581 unsigned long clear_flags = EXTENT_DELALLOC |
582 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG;
583 unsigned long page_error_op;
584
585 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
586 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
587
588 /*
589 * inline extent creation worked or returned error,
590 * we don't need to create any more async work items.
591 * Unlock and free up our temp pages.
592 */
593 extent_clear_unlock_delalloc(inode, start, end, end,
594 NULL, clear_flags,
595 PAGE_UNLOCK |
596 PAGE_CLEAR_DIRTY |
597 PAGE_SET_WRITEBACK |
598 page_error_op |
599 PAGE_END_WRITEBACK);
600 if (ret == 0)
601 btrfs_free_reserved_data_space_noquota(inode,
602 start,
603 end - start + 1);
604 goto free_pages_out;
605 }
606 }
607
608 if (will_compress) {
609 /*
610 * we aren't doing an inline extent round the compressed size
611 * up to a block size boundary so the allocator does sane
612 * things
613 */
614 total_compressed = ALIGN(total_compressed, blocksize);
615
616 /*
617 * one last check to make sure the compression is really a
618 * win, compare the page count read with the blocks on disk,
619 * compression must free at least one sector size
620 */
621 total_in = ALIGN(total_in, PAGE_SIZE);
622 if (total_compressed + blocksize <= total_in) {
623 num_bytes = total_in;
624 *num_added += 1;
625
626 /*
627 * The async work queues will take care of doing actual
628 * allocation on disk for these compressed pages, and
629 * will submit them to the elevator.
630 */
631 add_async_extent(async_cow, start, num_bytes,
632 total_compressed, pages, nr_pages,
633 compress_type);
634
635 if (start + num_bytes < end) {
636 start += num_bytes;
637 pages = NULL;
638 cond_resched();
639 goto again;
640 }
641 return;
642 }
643 }
644 if (pages) {
645 /*
646 * the compression code ran but failed to make things smaller,
647 * free any pages it allocated and our page pointer array
648 */
649 for (i = 0; i < nr_pages; i++) {
650 WARN_ON(pages[i]->mapping);
651 put_page(pages[i]);
652 }
653 kfree(pages);
654 pages = NULL;
655 total_compressed = 0;
656 nr_pages = 0;
657
658 /* flag the file so we don't compress in the future */
659 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
660 !(BTRFS_I(inode)->force_compress)) {
661 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
662 }
663 }
664 cleanup_and_bail_uncompressed:
665 /*
666 * No compression, but we still need to write the pages in the file
667 * we've been given so far. redirty the locked page if it corresponds
668 * to our extent and set things up for the async work queue to run
669 * cow_file_range to do the normal delalloc dance.
670 */
671 if (page_offset(locked_page) >= start &&
672 page_offset(locked_page) <= end)
673 __set_page_dirty_nobuffers(locked_page);
674 /* unlocked later on in the async handlers */
675
676 if (redirty)
677 extent_range_redirty_for_io(inode, start, end);
678 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
679 BTRFS_COMPRESS_NONE);
680 *num_added += 1;
681
682 return;
683
684 free_pages_out:
685 for (i = 0; i < nr_pages; i++) {
686 WARN_ON(pages[i]->mapping);
687 put_page(pages[i]);
688 }
689 kfree(pages);
690 }
691
692 static void free_async_extent_pages(struct async_extent *async_extent)
693 {
694 int i;
695
696 if (!async_extent->pages)
697 return;
698
699 for (i = 0; i < async_extent->nr_pages; i++) {
700 WARN_ON(async_extent->pages[i]->mapping);
701 put_page(async_extent->pages[i]);
702 }
703 kfree(async_extent->pages);
704 async_extent->nr_pages = 0;
705 async_extent->pages = NULL;
706 }
707
708 /*
709 * phase two of compressed writeback. This is the ordered portion
710 * of the code, which only gets called in the order the work was
711 * queued. We walk all the async extents created by compress_file_range
712 * and send them down to the disk.
713 */
714 static noinline void submit_compressed_extents(struct inode *inode,
715 struct async_cow *async_cow)
716 {
717 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
718 struct async_extent *async_extent;
719 u64 alloc_hint = 0;
720 struct btrfs_key ins;
721 struct extent_map *em;
722 struct btrfs_root *root = BTRFS_I(inode)->root;
723 struct extent_io_tree *io_tree;
724 int ret = 0;
725
726 again:
727 while (!list_empty(&async_cow->extents)) {
728 async_extent = list_entry(async_cow->extents.next,
729 struct async_extent, list);
730 list_del(&async_extent->list);
731
732 io_tree = &BTRFS_I(inode)->io_tree;
733
734 retry:
735 /* did the compression code fall back to uncompressed IO? */
736 if (!async_extent->pages) {
737 int page_started = 0;
738 unsigned long nr_written = 0;
739
740 lock_extent(io_tree, async_extent->start,
741 async_extent->start +
742 async_extent->ram_size - 1);
743
744 /* allocate blocks */
745 ret = cow_file_range(inode, async_cow->locked_page,
746 async_extent->start,
747 async_extent->start +
748 async_extent->ram_size - 1,
749 async_extent->start +
750 async_extent->ram_size - 1,
751 &page_started, &nr_written, 0,
752 NULL);
753
754 /* JDM XXX */
755
756 /*
757 * if page_started, cow_file_range inserted an
758 * inline extent and took care of all the unlocking
759 * and IO for us. Otherwise, we need to submit
760 * all those pages down to the drive.
761 */
762 if (!page_started && !ret)
763 extent_write_locked_range(io_tree,
764 inode, async_extent->start,
765 async_extent->start +
766 async_extent->ram_size - 1,
767 btrfs_get_extent,
768 WB_SYNC_ALL);
769 else if (ret)
770 unlock_page(async_cow->locked_page);
771 kfree(async_extent);
772 cond_resched();
773 continue;
774 }
775
776 lock_extent(io_tree, async_extent->start,
777 async_extent->start + async_extent->ram_size - 1);
778
779 ret = btrfs_reserve_extent(root, async_extent->ram_size,
780 async_extent->compressed_size,
781 async_extent->compressed_size,
782 0, alloc_hint, &ins, 1, 1);
783 if (ret) {
784 free_async_extent_pages(async_extent);
785
786 if (ret == -ENOSPC) {
787 unlock_extent(io_tree, async_extent->start,
788 async_extent->start +
789 async_extent->ram_size - 1);
790
791 /*
792 * we need to redirty the pages if we decide to
793 * fallback to uncompressed IO, otherwise we
794 * will not submit these pages down to lower
795 * layers.
796 */
797 extent_range_redirty_for_io(inode,
798 async_extent->start,
799 async_extent->start +
800 async_extent->ram_size - 1);
801
802 goto retry;
803 }
804 goto out_free;
805 }
806 /*
807 * here we're doing allocation and writeback of the
808 * compressed pages
809 */
810 em = create_io_em(inode, async_extent->start,
811 async_extent->ram_size, /* len */
812 async_extent->start, /* orig_start */
813 ins.objectid, /* block_start */
814 ins.offset, /* block_len */
815 ins.offset, /* orig_block_len */
816 async_extent->ram_size, /* ram_bytes */
817 async_extent->compress_type,
818 BTRFS_ORDERED_COMPRESSED);
819 if (IS_ERR(em))
820 /* ret value is not necessary due to void function */
821 goto out_free_reserve;
822 free_extent_map(em);
823
824 ret = btrfs_add_ordered_extent_compress(inode,
825 async_extent->start,
826 ins.objectid,
827 async_extent->ram_size,
828 ins.offset,
829 BTRFS_ORDERED_COMPRESSED,
830 async_extent->compress_type);
831 if (ret) {
832 btrfs_drop_extent_cache(BTRFS_I(inode),
833 async_extent->start,
834 async_extent->start +
835 async_extent->ram_size - 1, 0);
836 goto out_free_reserve;
837 }
838 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
839
840 /*
841 * clear dirty, set writeback and unlock the pages.
842 */
843 extent_clear_unlock_delalloc(inode, async_extent->start,
844 async_extent->start +
845 async_extent->ram_size - 1,
846 async_extent->start +
847 async_extent->ram_size - 1,
848 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
849 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
850 PAGE_SET_WRITEBACK);
851 if (btrfs_submit_compressed_write(inode,
852 async_extent->start,
853 async_extent->ram_size,
854 ins.objectid,
855 ins.offset, async_extent->pages,
856 async_extent->nr_pages)) {
857 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
858 struct page *p = async_extent->pages[0];
859 const u64 start = async_extent->start;
860 const u64 end = start + async_extent->ram_size - 1;
861
862 p->mapping = inode->i_mapping;
863 tree->ops->writepage_end_io_hook(p, start, end,
864 NULL, 0);
865 p->mapping = NULL;
866 extent_clear_unlock_delalloc(inode, start, end, end,
867 NULL, 0,
868 PAGE_END_WRITEBACK |
869 PAGE_SET_ERROR);
870 free_async_extent_pages(async_extent);
871 }
872 alloc_hint = ins.objectid + ins.offset;
873 kfree(async_extent);
874 cond_resched();
875 }
876 return;
877 out_free_reserve:
878 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
879 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
880 out_free:
881 extent_clear_unlock_delalloc(inode, async_extent->start,
882 async_extent->start +
883 async_extent->ram_size - 1,
884 async_extent->start +
885 async_extent->ram_size - 1,
886 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
887 EXTENT_DELALLOC_NEW |
888 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
889 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
890 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
891 PAGE_SET_ERROR);
892 free_async_extent_pages(async_extent);
893 kfree(async_extent);
894 goto again;
895 }
896
897 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
898 u64 num_bytes)
899 {
900 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
901 struct extent_map *em;
902 u64 alloc_hint = 0;
903
904 read_lock(&em_tree->lock);
905 em = search_extent_mapping(em_tree, start, num_bytes);
906 if (em) {
907 /*
908 * if block start isn't an actual block number then find the
909 * first block in this inode and use that as a hint. If that
910 * block is also bogus then just don't worry about it.
911 */
912 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
913 free_extent_map(em);
914 em = search_extent_mapping(em_tree, 0, 0);
915 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
916 alloc_hint = em->block_start;
917 if (em)
918 free_extent_map(em);
919 } else {
920 alloc_hint = em->block_start;
921 free_extent_map(em);
922 }
923 }
924 read_unlock(&em_tree->lock);
925
926 return alloc_hint;
927 }
928
929 /*
930 * when extent_io.c finds a delayed allocation range in the file,
931 * the call backs end up in this code. The basic idea is to
932 * allocate extents on disk for the range, and create ordered data structs
933 * in ram to track those extents.
934 *
935 * locked_page is the page that writepage had locked already. We use
936 * it to make sure we don't do extra locks or unlocks.
937 *
938 * *page_started is set to one if we unlock locked_page and do everything
939 * required to start IO on it. It may be clean and already done with
940 * IO when we return.
941 */
942 static noinline int cow_file_range(struct inode *inode,
943 struct page *locked_page,
944 u64 start, u64 end, u64 delalloc_end,
945 int *page_started, unsigned long *nr_written,
946 int unlock, struct btrfs_dedupe_hash *hash)
947 {
948 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
949 struct btrfs_root *root = BTRFS_I(inode)->root;
950 u64 alloc_hint = 0;
951 u64 num_bytes;
952 unsigned long ram_size;
953 u64 disk_num_bytes;
954 u64 cur_alloc_size = 0;
955 u64 blocksize = fs_info->sectorsize;
956 struct btrfs_key ins;
957 struct extent_map *em;
958 unsigned clear_bits;
959 unsigned long page_ops;
960 bool extent_reserved = false;
961 int ret = 0;
962
963 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
964 WARN_ON_ONCE(1);
965 ret = -EINVAL;
966 goto out_unlock;
967 }
968
969 num_bytes = ALIGN(end - start + 1, blocksize);
970 num_bytes = max(blocksize, num_bytes);
971 disk_num_bytes = num_bytes;
972
973 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
974
975 if (start == 0) {
976 /* lets try to make an inline extent */
977 ret = cow_file_range_inline(root, inode, start, end, 0,
978 BTRFS_COMPRESS_NONE, NULL);
979 if (ret == 0) {
980 extent_clear_unlock_delalloc(inode, start, end,
981 delalloc_end, NULL,
982 EXTENT_LOCKED | EXTENT_DELALLOC |
983 EXTENT_DELALLOC_NEW |
984 EXTENT_DEFRAG, PAGE_UNLOCK |
985 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
986 PAGE_END_WRITEBACK);
987 btrfs_free_reserved_data_space_noquota(inode, start,
988 end - start + 1);
989 *nr_written = *nr_written +
990 (end - start + PAGE_SIZE) / PAGE_SIZE;
991 *page_started = 1;
992 goto out;
993 } else if (ret < 0) {
994 goto out_unlock;
995 }
996 }
997
998 BUG_ON(disk_num_bytes >
999 btrfs_super_total_bytes(fs_info->super_copy));
1000
1001 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1002 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1003 start + num_bytes - 1, 0);
1004
1005 while (disk_num_bytes > 0) {
1006 cur_alloc_size = disk_num_bytes;
1007 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1008 fs_info->sectorsize, 0, alloc_hint,
1009 &ins, 1, 1);
1010 if (ret < 0)
1011 goto out_unlock;
1012 cur_alloc_size = ins.offset;
1013 extent_reserved = true;
1014
1015 ram_size = ins.offset;
1016 em = create_io_em(inode, start, ins.offset, /* len */
1017 start, /* orig_start */
1018 ins.objectid, /* block_start */
1019 ins.offset, /* block_len */
1020 ins.offset, /* orig_block_len */
1021 ram_size, /* ram_bytes */
1022 BTRFS_COMPRESS_NONE, /* compress_type */
1023 BTRFS_ORDERED_REGULAR /* type */);
1024 if (IS_ERR(em))
1025 goto out_reserve;
1026 free_extent_map(em);
1027
1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029 ram_size, cur_alloc_size, 0);
1030 if (ret)
1031 goto out_drop_extent_cache;
1032
1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1036 cur_alloc_size);
1037 /*
1038 * Only drop cache here, and process as normal.
1039 *
1040 * We must not allow extent_clear_unlock_delalloc()
1041 * at out_unlock label to free meta of this ordered
1042 * extent, as its meta should be freed by
1043 * btrfs_finish_ordered_io().
1044 *
1045 * So we must continue until @start is increased to
1046 * skip current ordered extent.
1047 */
1048 if (ret)
1049 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1050 start + ram_size - 1, 0);
1051 }
1052
1053 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1054
1055 /* we're not doing compressed IO, don't unlock the first
1056 * page (which the caller expects to stay locked), don't
1057 * clear any dirty bits and don't set any writeback bits
1058 *
1059 * Do set the Private2 bit so we know this page was properly
1060 * setup for writepage
1061 */
1062 page_ops = unlock ? PAGE_UNLOCK : 0;
1063 page_ops |= PAGE_SET_PRIVATE2;
1064
1065 extent_clear_unlock_delalloc(inode, start,
1066 start + ram_size - 1,
1067 delalloc_end, locked_page,
1068 EXTENT_LOCKED | EXTENT_DELALLOC,
1069 page_ops);
1070 if (disk_num_bytes < cur_alloc_size)
1071 disk_num_bytes = 0;
1072 else
1073 disk_num_bytes -= cur_alloc_size;
1074 num_bytes -= cur_alloc_size;
1075 alloc_hint = ins.objectid + ins.offset;
1076 start += cur_alloc_size;
1077 extent_reserved = false;
1078
1079 /*
1080 * btrfs_reloc_clone_csums() error, since start is increased
1081 * extent_clear_unlock_delalloc() at out_unlock label won't
1082 * free metadata of current ordered extent, we're OK to exit.
1083 */
1084 if (ret)
1085 goto out_unlock;
1086 }
1087 out:
1088 return ret;
1089
1090 out_drop_extent_cache:
1091 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1092 out_reserve:
1093 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1094 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1095 out_unlock:
1096 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1097 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1098 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1099 PAGE_END_WRITEBACK;
1100 /*
1101 * If we reserved an extent for our delalloc range (or a subrange) and
1102 * failed to create the respective ordered extent, then it means that
1103 * when we reserved the extent we decremented the extent's size from
1104 * the data space_info's bytes_may_use counter and incremented the
1105 * space_info's bytes_reserved counter by the same amount. We must make
1106 * sure extent_clear_unlock_delalloc() does not try to decrement again
1107 * the data space_info's bytes_may_use counter, therefore we do not pass
1108 * it the flag EXTENT_CLEAR_DATA_RESV.
1109 */
1110 if (extent_reserved) {
1111 extent_clear_unlock_delalloc(inode, start,
1112 start + cur_alloc_size,
1113 start + cur_alloc_size,
1114 locked_page,
1115 clear_bits,
1116 page_ops);
1117 start += cur_alloc_size;
1118 if (start >= end)
1119 goto out;
1120 }
1121 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1122 locked_page,
1123 clear_bits | EXTENT_CLEAR_DATA_RESV,
1124 page_ops);
1125 goto out;
1126 }
1127
1128 /*
1129 * work queue call back to started compression on a file and pages
1130 */
1131 static noinline void async_cow_start(struct btrfs_work *work)
1132 {
1133 struct async_cow *async_cow;
1134 int num_added = 0;
1135 async_cow = container_of(work, struct async_cow, work);
1136
1137 compress_file_range(async_cow->inode, async_cow->locked_page,
1138 async_cow->start, async_cow->end, async_cow,
1139 &num_added);
1140 if (num_added == 0) {
1141 btrfs_add_delayed_iput(async_cow->inode);
1142 async_cow->inode = NULL;
1143 }
1144 }
1145
1146 /*
1147 * work queue call back to submit previously compressed pages
1148 */
1149 static noinline void async_cow_submit(struct btrfs_work *work)
1150 {
1151 struct btrfs_fs_info *fs_info;
1152 struct async_cow *async_cow;
1153 struct btrfs_root *root;
1154 unsigned long nr_pages;
1155
1156 async_cow = container_of(work, struct async_cow, work);
1157
1158 root = async_cow->root;
1159 fs_info = root->fs_info;
1160 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1161 PAGE_SHIFT;
1162
1163 /*
1164 * atomic_sub_return implies a barrier for waitqueue_active
1165 */
1166 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1167 5 * SZ_1M &&
1168 waitqueue_active(&fs_info->async_submit_wait))
1169 wake_up(&fs_info->async_submit_wait);
1170
1171 if (async_cow->inode)
1172 submit_compressed_extents(async_cow->inode, async_cow);
1173 }
1174
1175 static noinline void async_cow_free(struct btrfs_work *work)
1176 {
1177 struct async_cow *async_cow;
1178 async_cow = container_of(work, struct async_cow, work);
1179 if (async_cow->inode)
1180 btrfs_add_delayed_iput(async_cow->inode);
1181 kfree(async_cow);
1182 }
1183
1184 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1185 u64 start, u64 end, int *page_started,
1186 unsigned long *nr_written)
1187 {
1188 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1189 struct async_cow *async_cow;
1190 struct btrfs_root *root = BTRFS_I(inode)->root;
1191 unsigned long nr_pages;
1192 u64 cur_end;
1193
1194 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1195 1, 0, NULL, GFP_NOFS);
1196 while (start < end) {
1197 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1198 BUG_ON(!async_cow); /* -ENOMEM */
1199 async_cow->inode = igrab(inode);
1200 async_cow->root = root;
1201 async_cow->locked_page = locked_page;
1202 async_cow->start = start;
1203
1204 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1205 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1206 cur_end = end;
1207 else
1208 cur_end = min(end, start + SZ_512K - 1);
1209
1210 async_cow->end = cur_end;
1211 INIT_LIST_HEAD(&async_cow->extents);
1212
1213 btrfs_init_work(&async_cow->work,
1214 btrfs_delalloc_helper,
1215 async_cow_start, async_cow_submit,
1216 async_cow_free);
1217
1218 nr_pages = (cur_end - start + PAGE_SIZE) >>
1219 PAGE_SHIFT;
1220 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1221
1222 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1223
1224 while (atomic_read(&fs_info->async_submit_draining) &&
1225 atomic_read(&fs_info->async_delalloc_pages)) {
1226 wait_event(fs_info->async_submit_wait,
1227 (atomic_read(&fs_info->async_delalloc_pages) ==
1228 0));
1229 }
1230
1231 *nr_written += nr_pages;
1232 start = cur_end + 1;
1233 }
1234 *page_started = 1;
1235 return 0;
1236 }
1237
1238 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1239 u64 bytenr, u64 num_bytes)
1240 {
1241 int ret;
1242 struct btrfs_ordered_sum *sums;
1243 LIST_HEAD(list);
1244
1245 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1246 bytenr + num_bytes - 1, &list, 0);
1247 if (ret == 0 && list_empty(&list))
1248 return 0;
1249
1250 while (!list_empty(&list)) {
1251 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1252 list_del(&sums->list);
1253 kfree(sums);
1254 }
1255 return 1;
1256 }
1257
1258 /*
1259 * when nowcow writeback call back. This checks for snapshots or COW copies
1260 * of the extents that exist in the file, and COWs the file as required.
1261 *
1262 * If no cow copies or snapshots exist, we write directly to the existing
1263 * blocks on disk
1264 */
1265 static noinline int run_delalloc_nocow(struct inode *inode,
1266 struct page *locked_page,
1267 u64 start, u64 end, int *page_started, int force,
1268 unsigned long *nr_written)
1269 {
1270 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1271 struct btrfs_root *root = BTRFS_I(inode)->root;
1272 struct extent_buffer *leaf;
1273 struct btrfs_path *path;
1274 struct btrfs_file_extent_item *fi;
1275 struct btrfs_key found_key;
1276 struct extent_map *em;
1277 u64 cow_start;
1278 u64 cur_offset;
1279 u64 extent_end;
1280 u64 extent_offset;
1281 u64 disk_bytenr;
1282 u64 num_bytes;
1283 u64 disk_num_bytes;
1284 u64 ram_bytes;
1285 int extent_type;
1286 int ret, err;
1287 int type;
1288 int nocow;
1289 int check_prev = 1;
1290 bool nolock;
1291 u64 ino = btrfs_ino(BTRFS_I(inode));
1292
1293 path = btrfs_alloc_path();
1294 if (!path) {
1295 extent_clear_unlock_delalloc(inode, start, end, end,
1296 locked_page,
1297 EXTENT_LOCKED | EXTENT_DELALLOC |
1298 EXTENT_DO_ACCOUNTING |
1299 EXTENT_DEFRAG, PAGE_UNLOCK |
1300 PAGE_CLEAR_DIRTY |
1301 PAGE_SET_WRITEBACK |
1302 PAGE_END_WRITEBACK);
1303 return -ENOMEM;
1304 }
1305
1306 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1307
1308 cow_start = (u64)-1;
1309 cur_offset = start;
1310 while (1) {
1311 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1312 cur_offset, 0);
1313 if (ret < 0)
1314 goto error;
1315 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1316 leaf = path->nodes[0];
1317 btrfs_item_key_to_cpu(leaf, &found_key,
1318 path->slots[0] - 1);
1319 if (found_key.objectid == ino &&
1320 found_key.type == BTRFS_EXTENT_DATA_KEY)
1321 path->slots[0]--;
1322 }
1323 check_prev = 0;
1324 next_slot:
1325 leaf = path->nodes[0];
1326 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1327 ret = btrfs_next_leaf(root, path);
1328 if (ret < 0)
1329 goto error;
1330 if (ret > 0)
1331 break;
1332 leaf = path->nodes[0];
1333 }
1334
1335 nocow = 0;
1336 disk_bytenr = 0;
1337 num_bytes = 0;
1338 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1339
1340 if (found_key.objectid > ino)
1341 break;
1342 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1343 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1344 path->slots[0]++;
1345 goto next_slot;
1346 }
1347 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1348 found_key.offset > end)
1349 break;
1350
1351 if (found_key.offset > cur_offset) {
1352 extent_end = found_key.offset;
1353 extent_type = 0;
1354 goto out_check;
1355 }
1356
1357 fi = btrfs_item_ptr(leaf, path->slots[0],
1358 struct btrfs_file_extent_item);
1359 extent_type = btrfs_file_extent_type(leaf, fi);
1360
1361 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1362 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1363 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1364 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1365 extent_offset = btrfs_file_extent_offset(leaf, fi);
1366 extent_end = found_key.offset +
1367 btrfs_file_extent_num_bytes(leaf, fi);
1368 disk_num_bytes =
1369 btrfs_file_extent_disk_num_bytes(leaf, fi);
1370 if (extent_end <= start) {
1371 path->slots[0]++;
1372 goto next_slot;
1373 }
1374 if (disk_bytenr == 0)
1375 goto out_check;
1376 if (btrfs_file_extent_compression(leaf, fi) ||
1377 btrfs_file_extent_encryption(leaf, fi) ||
1378 btrfs_file_extent_other_encoding(leaf, fi))
1379 goto out_check;
1380 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1381 goto out_check;
1382 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1383 goto out_check;
1384 if (btrfs_cross_ref_exist(root, ino,
1385 found_key.offset -
1386 extent_offset, disk_bytenr))
1387 goto out_check;
1388 disk_bytenr += extent_offset;
1389 disk_bytenr += cur_offset - found_key.offset;
1390 num_bytes = min(end + 1, extent_end) - cur_offset;
1391 /*
1392 * if there are pending snapshots for this root,
1393 * we fall into common COW way.
1394 */
1395 if (!nolock) {
1396 err = btrfs_start_write_no_snapshoting(root);
1397 if (!err)
1398 goto out_check;
1399 }
1400 /*
1401 * force cow if csum exists in the range.
1402 * this ensure that csum for a given extent are
1403 * either valid or do not exist.
1404 */
1405 if (csum_exist_in_range(fs_info, disk_bytenr,
1406 num_bytes)) {
1407 if (!nolock)
1408 btrfs_end_write_no_snapshoting(root);
1409 goto out_check;
1410 }
1411 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1412 if (!nolock)
1413 btrfs_end_write_no_snapshoting(root);
1414 goto out_check;
1415 }
1416 nocow = 1;
1417 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1418 extent_end = found_key.offset +
1419 btrfs_file_extent_inline_len(leaf,
1420 path->slots[0], fi);
1421 extent_end = ALIGN(extent_end,
1422 fs_info->sectorsize);
1423 } else {
1424 BUG_ON(1);
1425 }
1426 out_check:
1427 if (extent_end <= start) {
1428 path->slots[0]++;
1429 if (!nolock && nocow)
1430 btrfs_end_write_no_snapshoting(root);
1431 if (nocow)
1432 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1433 goto next_slot;
1434 }
1435 if (!nocow) {
1436 if (cow_start == (u64)-1)
1437 cow_start = cur_offset;
1438 cur_offset = extent_end;
1439 if (cur_offset > end)
1440 break;
1441 path->slots[0]++;
1442 goto next_slot;
1443 }
1444
1445 btrfs_release_path(path);
1446 if (cow_start != (u64)-1) {
1447 ret = cow_file_range(inode, locked_page,
1448 cow_start, found_key.offset - 1,
1449 end, page_started, nr_written, 1,
1450 NULL);
1451 if (ret) {
1452 if (!nolock && nocow)
1453 btrfs_end_write_no_snapshoting(root);
1454 if (nocow)
1455 btrfs_dec_nocow_writers(fs_info,
1456 disk_bytenr);
1457 goto error;
1458 }
1459 cow_start = (u64)-1;
1460 }
1461
1462 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1463 u64 orig_start = found_key.offset - extent_offset;
1464
1465 em = create_io_em(inode, cur_offset, num_bytes,
1466 orig_start,
1467 disk_bytenr, /* block_start */
1468 num_bytes, /* block_len */
1469 disk_num_bytes, /* orig_block_len */
1470 ram_bytes, BTRFS_COMPRESS_NONE,
1471 BTRFS_ORDERED_PREALLOC);
1472 if (IS_ERR(em)) {
1473 if (!nolock && nocow)
1474 btrfs_end_write_no_snapshoting(root);
1475 if (nocow)
1476 btrfs_dec_nocow_writers(fs_info,
1477 disk_bytenr);
1478 ret = PTR_ERR(em);
1479 goto error;
1480 }
1481 free_extent_map(em);
1482 }
1483
1484 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1485 type = BTRFS_ORDERED_PREALLOC;
1486 } else {
1487 type = BTRFS_ORDERED_NOCOW;
1488 }
1489
1490 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1491 num_bytes, num_bytes, type);
1492 if (nocow)
1493 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1494 BUG_ON(ret); /* -ENOMEM */
1495
1496 if (root->root_key.objectid ==
1497 BTRFS_DATA_RELOC_TREE_OBJECTID)
1498 /*
1499 * Error handled later, as we must prevent
1500 * extent_clear_unlock_delalloc() in error handler
1501 * from freeing metadata of created ordered extent.
1502 */
1503 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1504 num_bytes);
1505
1506 extent_clear_unlock_delalloc(inode, cur_offset,
1507 cur_offset + num_bytes - 1, end,
1508 locked_page, EXTENT_LOCKED |
1509 EXTENT_DELALLOC |
1510 EXTENT_CLEAR_DATA_RESV,
1511 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1512
1513 if (!nolock && nocow)
1514 btrfs_end_write_no_snapshoting(root);
1515 cur_offset = extent_end;
1516
1517 /*
1518 * btrfs_reloc_clone_csums() error, now we're OK to call error
1519 * handler, as metadata for created ordered extent will only
1520 * be freed by btrfs_finish_ordered_io().
1521 */
1522 if (ret)
1523 goto error;
1524 if (cur_offset > end)
1525 break;
1526 }
1527 btrfs_release_path(path);
1528
1529 if (cur_offset <= end && cow_start == (u64)-1) {
1530 cow_start = cur_offset;
1531 cur_offset = end;
1532 }
1533
1534 if (cow_start != (u64)-1) {
1535 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1536 page_started, nr_written, 1, NULL);
1537 if (ret)
1538 goto error;
1539 }
1540
1541 error:
1542 if (ret && cur_offset < end)
1543 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1544 locked_page, EXTENT_LOCKED |
1545 EXTENT_DELALLOC | EXTENT_DEFRAG |
1546 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1547 PAGE_CLEAR_DIRTY |
1548 PAGE_SET_WRITEBACK |
1549 PAGE_END_WRITEBACK);
1550 btrfs_free_path(path);
1551 return ret;
1552 }
1553
1554 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1555 {
1556
1557 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1558 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1559 return 0;
1560
1561 /*
1562 * @defrag_bytes is a hint value, no spinlock held here,
1563 * if is not zero, it means the file is defragging.
1564 * Force cow if given extent needs to be defragged.
1565 */
1566 if (BTRFS_I(inode)->defrag_bytes &&
1567 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1568 EXTENT_DEFRAG, 0, NULL))
1569 return 1;
1570
1571 return 0;
1572 }
1573
1574 /*
1575 * extent_io.c call back to do delayed allocation processing
1576 */
1577 static int run_delalloc_range(void *private_data, struct page *locked_page,
1578 u64 start, u64 end, int *page_started,
1579 unsigned long *nr_written)
1580 {
1581 struct inode *inode = private_data;
1582 int ret;
1583 int force_cow = need_force_cow(inode, start, end);
1584
1585 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1586 ret = run_delalloc_nocow(inode, locked_page, start, end,
1587 page_started, 1, nr_written);
1588 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1589 ret = run_delalloc_nocow(inode, locked_page, start, end,
1590 page_started, 0, nr_written);
1591 } else if (!inode_need_compress(inode)) {
1592 ret = cow_file_range(inode, locked_page, start, end, end,
1593 page_started, nr_written, 1, NULL);
1594 } else {
1595 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1596 &BTRFS_I(inode)->runtime_flags);
1597 ret = cow_file_range_async(inode, locked_page, start, end,
1598 page_started, nr_written);
1599 }
1600 if (ret)
1601 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1602 return ret;
1603 }
1604
1605 static void btrfs_split_extent_hook(void *private_data,
1606 struct extent_state *orig, u64 split)
1607 {
1608 struct inode *inode = private_data;
1609 u64 size;
1610
1611 /* not delalloc, ignore it */
1612 if (!(orig->state & EXTENT_DELALLOC))
1613 return;
1614
1615 size = orig->end - orig->start + 1;
1616 if (size > BTRFS_MAX_EXTENT_SIZE) {
1617 u32 num_extents;
1618 u64 new_size;
1619
1620 /*
1621 * See the explanation in btrfs_merge_extent_hook, the same
1622 * applies here, just in reverse.
1623 */
1624 new_size = orig->end - split + 1;
1625 num_extents = count_max_extents(new_size);
1626 new_size = split - orig->start;
1627 num_extents += count_max_extents(new_size);
1628 if (count_max_extents(size) >= num_extents)
1629 return;
1630 }
1631
1632 spin_lock(&BTRFS_I(inode)->lock);
1633 BTRFS_I(inode)->outstanding_extents++;
1634 spin_unlock(&BTRFS_I(inode)->lock);
1635 }
1636
1637 /*
1638 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1639 * extents so we can keep track of new extents that are just merged onto old
1640 * extents, such as when we are doing sequential writes, so we can properly
1641 * account for the metadata space we'll need.
1642 */
1643 static void btrfs_merge_extent_hook(void *private_data,
1644 struct extent_state *new,
1645 struct extent_state *other)
1646 {
1647 struct inode *inode = private_data;
1648 u64 new_size, old_size;
1649 u32 num_extents;
1650
1651 /* not delalloc, ignore it */
1652 if (!(other->state & EXTENT_DELALLOC))
1653 return;
1654
1655 if (new->start > other->start)
1656 new_size = new->end - other->start + 1;
1657 else
1658 new_size = other->end - new->start + 1;
1659
1660 /* we're not bigger than the max, unreserve the space and go */
1661 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1662 spin_lock(&BTRFS_I(inode)->lock);
1663 BTRFS_I(inode)->outstanding_extents--;
1664 spin_unlock(&BTRFS_I(inode)->lock);
1665 return;
1666 }
1667
1668 /*
1669 * We have to add up either side to figure out how many extents were
1670 * accounted for before we merged into one big extent. If the number of
1671 * extents we accounted for is <= the amount we need for the new range
1672 * then we can return, otherwise drop. Think of it like this
1673 *
1674 * [ 4k][MAX_SIZE]
1675 *
1676 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1677 * need 2 outstanding extents, on one side we have 1 and the other side
1678 * we have 1 so they are == and we can return. But in this case
1679 *
1680 * [MAX_SIZE+4k][MAX_SIZE+4k]
1681 *
1682 * Each range on their own accounts for 2 extents, but merged together
1683 * they are only 3 extents worth of accounting, so we need to drop in
1684 * this case.
1685 */
1686 old_size = other->end - other->start + 1;
1687 num_extents = count_max_extents(old_size);
1688 old_size = new->end - new->start + 1;
1689 num_extents += count_max_extents(old_size);
1690 if (count_max_extents(new_size) >= num_extents)
1691 return;
1692
1693 spin_lock(&BTRFS_I(inode)->lock);
1694 BTRFS_I(inode)->outstanding_extents--;
1695 spin_unlock(&BTRFS_I(inode)->lock);
1696 }
1697
1698 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1699 struct inode *inode)
1700 {
1701 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1702
1703 spin_lock(&root->delalloc_lock);
1704 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1705 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1706 &root->delalloc_inodes);
1707 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1708 &BTRFS_I(inode)->runtime_flags);
1709 root->nr_delalloc_inodes++;
1710 if (root->nr_delalloc_inodes == 1) {
1711 spin_lock(&fs_info->delalloc_root_lock);
1712 BUG_ON(!list_empty(&root->delalloc_root));
1713 list_add_tail(&root->delalloc_root,
1714 &fs_info->delalloc_roots);
1715 spin_unlock(&fs_info->delalloc_root_lock);
1716 }
1717 }
1718 spin_unlock(&root->delalloc_lock);
1719 }
1720
1721 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1722 struct btrfs_inode *inode)
1723 {
1724 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1725
1726 spin_lock(&root->delalloc_lock);
1727 if (!list_empty(&inode->delalloc_inodes)) {
1728 list_del_init(&inode->delalloc_inodes);
1729 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1730 &inode->runtime_flags);
1731 root->nr_delalloc_inodes--;
1732 if (!root->nr_delalloc_inodes) {
1733 spin_lock(&fs_info->delalloc_root_lock);
1734 BUG_ON(list_empty(&root->delalloc_root));
1735 list_del_init(&root->delalloc_root);
1736 spin_unlock(&fs_info->delalloc_root_lock);
1737 }
1738 }
1739 spin_unlock(&root->delalloc_lock);
1740 }
1741
1742 /*
1743 * extent_io.c set_bit_hook, used to track delayed allocation
1744 * bytes in this file, and to maintain the list of inodes that
1745 * have pending delalloc work to be done.
1746 */
1747 static void btrfs_set_bit_hook(void *private_data,
1748 struct extent_state *state, unsigned *bits)
1749 {
1750 struct inode *inode = private_data;
1751
1752 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1753
1754 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1755 WARN_ON(1);
1756 /*
1757 * set_bit and clear bit hooks normally require _irqsave/restore
1758 * but in this case, we are only testing for the DELALLOC
1759 * bit, which is only set or cleared with irqs on
1760 */
1761 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1762 struct btrfs_root *root = BTRFS_I(inode)->root;
1763 u64 len = state->end + 1 - state->start;
1764 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1765
1766 if (*bits & EXTENT_FIRST_DELALLOC) {
1767 *bits &= ~EXTENT_FIRST_DELALLOC;
1768 } else {
1769 spin_lock(&BTRFS_I(inode)->lock);
1770 BTRFS_I(inode)->outstanding_extents++;
1771 spin_unlock(&BTRFS_I(inode)->lock);
1772 }
1773
1774 /* For sanity tests */
1775 if (btrfs_is_testing(fs_info))
1776 return;
1777
1778 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1779 fs_info->delalloc_batch);
1780 spin_lock(&BTRFS_I(inode)->lock);
1781 BTRFS_I(inode)->delalloc_bytes += len;
1782 if (*bits & EXTENT_DEFRAG)
1783 BTRFS_I(inode)->defrag_bytes += len;
1784 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1785 &BTRFS_I(inode)->runtime_flags))
1786 btrfs_add_delalloc_inodes(root, inode);
1787 spin_unlock(&BTRFS_I(inode)->lock);
1788 }
1789
1790 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1791 (*bits & EXTENT_DELALLOC_NEW)) {
1792 spin_lock(&BTRFS_I(inode)->lock);
1793 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1794 state->start;
1795 spin_unlock(&BTRFS_I(inode)->lock);
1796 }
1797 }
1798
1799 /*
1800 * extent_io.c clear_bit_hook, see set_bit_hook for why
1801 */
1802 static void btrfs_clear_bit_hook(void *private_data,
1803 struct extent_state *state,
1804 unsigned *bits)
1805 {
1806 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1807 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1808 u64 len = state->end + 1 - state->start;
1809 u32 num_extents = count_max_extents(len);
1810
1811 spin_lock(&inode->lock);
1812 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1813 inode->defrag_bytes -= len;
1814 spin_unlock(&inode->lock);
1815
1816 /*
1817 * set_bit and clear bit hooks normally require _irqsave/restore
1818 * but in this case, we are only testing for the DELALLOC
1819 * bit, which is only set or cleared with irqs on
1820 */
1821 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1822 struct btrfs_root *root = inode->root;
1823 bool do_list = !btrfs_is_free_space_inode(inode);
1824
1825 if (*bits & EXTENT_FIRST_DELALLOC) {
1826 *bits &= ~EXTENT_FIRST_DELALLOC;
1827 } else if (!(*bits & EXTENT_CLEAR_META_RESV)) {
1828 spin_lock(&inode->lock);
1829 inode->outstanding_extents -= num_extents;
1830 spin_unlock(&inode->lock);
1831 }
1832
1833 /*
1834 * We don't reserve metadata space for space cache inodes so we
1835 * don't need to call dellalloc_release_metadata if there is an
1836 * error.
1837 */
1838 if (*bits & EXTENT_CLEAR_META_RESV &&
1839 root != fs_info->tree_root)
1840 btrfs_delalloc_release_metadata(inode, len);
1841
1842 /* For sanity tests. */
1843 if (btrfs_is_testing(fs_info))
1844 return;
1845
1846 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1847 do_list && !(state->state & EXTENT_NORESERVE) &&
1848 (*bits & EXTENT_CLEAR_DATA_RESV))
1849 btrfs_free_reserved_data_space_noquota(
1850 &inode->vfs_inode,
1851 state->start, len);
1852
1853 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1854 fs_info->delalloc_batch);
1855 spin_lock(&inode->lock);
1856 inode->delalloc_bytes -= len;
1857 if (do_list && inode->delalloc_bytes == 0 &&
1858 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1859 &inode->runtime_flags))
1860 btrfs_del_delalloc_inode(root, inode);
1861 spin_unlock(&inode->lock);
1862 }
1863
1864 if ((state->state & EXTENT_DELALLOC_NEW) &&
1865 (*bits & EXTENT_DELALLOC_NEW)) {
1866 spin_lock(&inode->lock);
1867 ASSERT(inode->new_delalloc_bytes >= len);
1868 inode->new_delalloc_bytes -= len;
1869 spin_unlock(&inode->lock);
1870 }
1871 }
1872
1873 /*
1874 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1875 * we don't create bios that span stripes or chunks
1876 *
1877 * return 1 if page cannot be merged to bio
1878 * return 0 if page can be merged to bio
1879 * return error otherwise
1880 */
1881 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1882 size_t size, struct bio *bio,
1883 unsigned long bio_flags)
1884 {
1885 struct inode *inode = page->mapping->host;
1886 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1887 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1888 u64 length = 0;
1889 u64 map_length;
1890 int ret;
1891
1892 if (bio_flags & EXTENT_BIO_COMPRESSED)
1893 return 0;
1894
1895 length = bio->bi_iter.bi_size;
1896 map_length = length;
1897 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1898 NULL, 0);
1899 if (ret < 0)
1900 return ret;
1901 if (map_length < length + size)
1902 return 1;
1903 return 0;
1904 }
1905
1906 /*
1907 * in order to insert checksums into the metadata in large chunks,
1908 * we wait until bio submission time. All the pages in the bio are
1909 * checksummed and sums are attached onto the ordered extent record.
1910 *
1911 * At IO completion time the cums attached on the ordered extent record
1912 * are inserted into the btree
1913 */
1914 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1915 int mirror_num, unsigned long bio_flags,
1916 u64 bio_offset)
1917 {
1918 struct inode *inode = private_data;
1919 blk_status_t ret = 0;
1920
1921 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1922 BUG_ON(ret); /* -ENOMEM */
1923 return 0;
1924 }
1925
1926 /*
1927 * in order to insert checksums into the metadata in large chunks,
1928 * we wait until bio submission time. All the pages in the bio are
1929 * checksummed and sums are attached onto the ordered extent record.
1930 *
1931 * At IO completion time the cums attached on the ordered extent record
1932 * are inserted into the btree
1933 */
1934 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1935 int mirror_num, unsigned long bio_flags,
1936 u64 bio_offset)
1937 {
1938 struct inode *inode = private_data;
1939 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1940 blk_status_t ret;
1941
1942 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1943 if (ret) {
1944 bio->bi_status = ret;
1945 bio_endio(bio);
1946 }
1947 return ret;
1948 }
1949
1950 /*
1951 * extent_io.c submission hook. This does the right thing for csum calculation
1952 * on write, or reading the csums from the tree before a read
1953 */
1954 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1955 int mirror_num, unsigned long bio_flags,
1956 u64 bio_offset)
1957 {
1958 struct inode *inode = private_data;
1959 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1960 struct btrfs_root *root = BTRFS_I(inode)->root;
1961 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1962 blk_status_t ret = 0;
1963 int skip_sum;
1964 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1965
1966 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1967
1968 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1969 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1970
1971 if (bio_op(bio) != REQ_OP_WRITE) {
1972 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1973 if (ret)
1974 goto out;
1975
1976 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1977 ret = btrfs_submit_compressed_read(inode, bio,
1978 mirror_num,
1979 bio_flags);
1980 goto out;
1981 } else if (!skip_sum) {
1982 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1983 if (ret)
1984 goto out;
1985 }
1986 goto mapit;
1987 } else if (async && !skip_sum) {
1988 /* csum items have already been cloned */
1989 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1990 goto mapit;
1991 /* we're doing a write, do the async checksumming */
1992 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1993 bio_offset, inode,
1994 __btrfs_submit_bio_start,
1995 __btrfs_submit_bio_done);
1996 goto out;
1997 } else if (!skip_sum) {
1998 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1999 if (ret)
2000 goto out;
2001 }
2002
2003 mapit:
2004 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2005
2006 out:
2007 if (ret) {
2008 bio->bi_status = ret;
2009 bio_endio(bio);
2010 }
2011 return ret;
2012 }
2013
2014 /*
2015 * given a list of ordered sums record them in the inode. This happens
2016 * at IO completion time based on sums calculated at bio submission time.
2017 */
2018 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2019 struct inode *inode, struct list_head *list)
2020 {
2021 struct btrfs_ordered_sum *sum;
2022
2023 list_for_each_entry(sum, list, list) {
2024 trans->adding_csums = 1;
2025 btrfs_csum_file_blocks(trans,
2026 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2027 trans->adding_csums = 0;
2028 }
2029 return 0;
2030 }
2031
2032 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2033 struct extent_state **cached_state, int dedupe)
2034 {
2035 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2036 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2037 cached_state);
2038 }
2039
2040 /* see btrfs_writepage_start_hook for details on why this is required */
2041 struct btrfs_writepage_fixup {
2042 struct page *page;
2043 struct btrfs_work work;
2044 };
2045
2046 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2047 {
2048 struct btrfs_writepage_fixup *fixup;
2049 struct btrfs_ordered_extent *ordered;
2050 struct extent_state *cached_state = NULL;
2051 struct extent_changeset *data_reserved = NULL;
2052 struct page *page;
2053 struct inode *inode;
2054 u64 page_start;
2055 u64 page_end;
2056 int ret;
2057
2058 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2059 page = fixup->page;
2060 again:
2061 lock_page(page);
2062 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2063 ClearPageChecked(page);
2064 goto out_page;
2065 }
2066
2067 inode = page->mapping->host;
2068 page_start = page_offset(page);
2069 page_end = page_offset(page) + PAGE_SIZE - 1;
2070
2071 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2072 &cached_state);
2073
2074 /* already ordered? We're done */
2075 if (PagePrivate2(page))
2076 goto out;
2077
2078 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2079 PAGE_SIZE);
2080 if (ordered) {
2081 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2082 page_end, &cached_state, GFP_NOFS);
2083 unlock_page(page);
2084 btrfs_start_ordered_extent(inode, ordered, 1);
2085 btrfs_put_ordered_extent(ordered);
2086 goto again;
2087 }
2088
2089 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2090 PAGE_SIZE);
2091 if (ret) {
2092 mapping_set_error(page->mapping, ret);
2093 end_extent_writepage(page, ret, page_start, page_end);
2094 ClearPageChecked(page);
2095 goto out;
2096 }
2097
2098 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2099 0);
2100 ClearPageChecked(page);
2101 set_page_dirty(page);
2102 out:
2103 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2104 &cached_state, GFP_NOFS);
2105 out_page:
2106 unlock_page(page);
2107 put_page(page);
2108 kfree(fixup);
2109 extent_changeset_free(data_reserved);
2110 }
2111
2112 /*
2113 * There are a few paths in the higher layers of the kernel that directly
2114 * set the page dirty bit without asking the filesystem if it is a
2115 * good idea. This causes problems because we want to make sure COW
2116 * properly happens and the data=ordered rules are followed.
2117 *
2118 * In our case any range that doesn't have the ORDERED bit set
2119 * hasn't been properly setup for IO. We kick off an async process
2120 * to fix it up. The async helper will wait for ordered extents, set
2121 * the delalloc bit and make it safe to write the page.
2122 */
2123 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2124 {
2125 struct inode *inode = page->mapping->host;
2126 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2127 struct btrfs_writepage_fixup *fixup;
2128
2129 /* this page is properly in the ordered list */
2130 if (TestClearPagePrivate2(page))
2131 return 0;
2132
2133 if (PageChecked(page))
2134 return -EAGAIN;
2135
2136 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2137 if (!fixup)
2138 return -EAGAIN;
2139
2140 SetPageChecked(page);
2141 get_page(page);
2142 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2143 btrfs_writepage_fixup_worker, NULL, NULL);
2144 fixup->page = page;
2145 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2146 return -EBUSY;
2147 }
2148
2149 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2150 struct inode *inode, u64 file_pos,
2151 u64 disk_bytenr, u64 disk_num_bytes,
2152 u64 num_bytes, u64 ram_bytes,
2153 u8 compression, u8 encryption,
2154 u16 other_encoding, int extent_type)
2155 {
2156 struct btrfs_root *root = BTRFS_I(inode)->root;
2157 struct btrfs_file_extent_item *fi;
2158 struct btrfs_path *path;
2159 struct extent_buffer *leaf;
2160 struct btrfs_key ins;
2161 u64 qg_released;
2162 int extent_inserted = 0;
2163 int ret;
2164
2165 path = btrfs_alloc_path();
2166 if (!path)
2167 return -ENOMEM;
2168
2169 /*
2170 * we may be replacing one extent in the tree with another.
2171 * The new extent is pinned in the extent map, and we don't want
2172 * to drop it from the cache until it is completely in the btree.
2173 *
2174 * So, tell btrfs_drop_extents to leave this extent in the cache.
2175 * the caller is expected to unpin it and allow it to be merged
2176 * with the others.
2177 */
2178 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2179 file_pos + num_bytes, NULL, 0,
2180 1, sizeof(*fi), &extent_inserted);
2181 if (ret)
2182 goto out;
2183
2184 if (!extent_inserted) {
2185 ins.objectid = btrfs_ino(BTRFS_I(inode));
2186 ins.offset = file_pos;
2187 ins.type = BTRFS_EXTENT_DATA_KEY;
2188
2189 path->leave_spinning = 1;
2190 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2191 sizeof(*fi));
2192 if (ret)
2193 goto out;
2194 }
2195 leaf = path->nodes[0];
2196 fi = btrfs_item_ptr(leaf, path->slots[0],
2197 struct btrfs_file_extent_item);
2198 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2199 btrfs_set_file_extent_type(leaf, fi, extent_type);
2200 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2201 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2202 btrfs_set_file_extent_offset(leaf, fi, 0);
2203 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2204 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2205 btrfs_set_file_extent_compression(leaf, fi, compression);
2206 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2207 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2208
2209 btrfs_mark_buffer_dirty(leaf);
2210 btrfs_release_path(path);
2211
2212 inode_add_bytes(inode, num_bytes);
2213
2214 ins.objectid = disk_bytenr;
2215 ins.offset = disk_num_bytes;
2216 ins.type = BTRFS_EXTENT_ITEM_KEY;
2217
2218 /*
2219 * Release the reserved range from inode dirty range map, as it is
2220 * already moved into delayed_ref_head
2221 */
2222 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2223 if (ret < 0)
2224 goto out;
2225 qg_released = ret;
2226 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2227 btrfs_ino(BTRFS_I(inode)), file_pos, qg_released, &ins);
2228 out:
2229 btrfs_free_path(path);
2230
2231 return ret;
2232 }
2233
2234 /* snapshot-aware defrag */
2235 struct sa_defrag_extent_backref {
2236 struct rb_node node;
2237 struct old_sa_defrag_extent *old;
2238 u64 root_id;
2239 u64 inum;
2240 u64 file_pos;
2241 u64 extent_offset;
2242 u64 num_bytes;
2243 u64 generation;
2244 };
2245
2246 struct old_sa_defrag_extent {
2247 struct list_head list;
2248 struct new_sa_defrag_extent *new;
2249
2250 u64 extent_offset;
2251 u64 bytenr;
2252 u64 offset;
2253 u64 len;
2254 int count;
2255 };
2256
2257 struct new_sa_defrag_extent {
2258 struct rb_root root;
2259 struct list_head head;
2260 struct btrfs_path *path;
2261 struct inode *inode;
2262 u64 file_pos;
2263 u64 len;
2264 u64 bytenr;
2265 u64 disk_len;
2266 u8 compress_type;
2267 };
2268
2269 static int backref_comp(struct sa_defrag_extent_backref *b1,
2270 struct sa_defrag_extent_backref *b2)
2271 {
2272 if (b1->root_id < b2->root_id)
2273 return -1;
2274 else if (b1->root_id > b2->root_id)
2275 return 1;
2276
2277 if (b1->inum < b2->inum)
2278 return -1;
2279 else if (b1->inum > b2->inum)
2280 return 1;
2281
2282 if (b1->file_pos < b2->file_pos)
2283 return -1;
2284 else if (b1->file_pos > b2->file_pos)
2285 return 1;
2286
2287 /*
2288 * [------------------------------] ===> (a range of space)
2289 * |<--->| |<---->| =============> (fs/file tree A)
2290 * |<---------------------------->| ===> (fs/file tree B)
2291 *
2292 * A range of space can refer to two file extents in one tree while
2293 * refer to only one file extent in another tree.
2294 *
2295 * So we may process a disk offset more than one time(two extents in A)
2296 * and locate at the same extent(one extent in B), then insert two same
2297 * backrefs(both refer to the extent in B).
2298 */
2299 return 0;
2300 }
2301
2302 static void backref_insert(struct rb_root *root,
2303 struct sa_defrag_extent_backref *backref)
2304 {
2305 struct rb_node **p = &root->rb_node;
2306 struct rb_node *parent = NULL;
2307 struct sa_defrag_extent_backref *entry;
2308 int ret;
2309
2310 while (*p) {
2311 parent = *p;
2312 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2313
2314 ret = backref_comp(backref, entry);
2315 if (ret < 0)
2316 p = &(*p)->rb_left;
2317 else
2318 p = &(*p)->rb_right;
2319 }
2320
2321 rb_link_node(&backref->node, parent, p);
2322 rb_insert_color(&backref->node, root);
2323 }
2324
2325 /*
2326 * Note the backref might has changed, and in this case we just return 0.
2327 */
2328 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2329 void *ctx)
2330 {
2331 struct btrfs_file_extent_item *extent;
2332 struct old_sa_defrag_extent *old = ctx;
2333 struct new_sa_defrag_extent *new = old->new;
2334 struct btrfs_path *path = new->path;
2335 struct btrfs_key key;
2336 struct btrfs_root *root;
2337 struct sa_defrag_extent_backref *backref;
2338 struct extent_buffer *leaf;
2339 struct inode *inode = new->inode;
2340 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2341 int slot;
2342 int ret;
2343 u64 extent_offset;
2344 u64 num_bytes;
2345
2346 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2347 inum == btrfs_ino(BTRFS_I(inode)))
2348 return 0;
2349
2350 key.objectid = root_id;
2351 key.type = BTRFS_ROOT_ITEM_KEY;
2352 key.offset = (u64)-1;
2353
2354 root = btrfs_read_fs_root_no_name(fs_info, &key);
2355 if (IS_ERR(root)) {
2356 if (PTR_ERR(root) == -ENOENT)
2357 return 0;
2358 WARN_ON(1);
2359 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2360 inum, offset, root_id);
2361 return PTR_ERR(root);
2362 }
2363
2364 key.objectid = inum;
2365 key.type = BTRFS_EXTENT_DATA_KEY;
2366 if (offset > (u64)-1 << 32)
2367 key.offset = 0;
2368 else
2369 key.offset = offset;
2370
2371 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2372 if (WARN_ON(ret < 0))
2373 return ret;
2374 ret = 0;
2375
2376 while (1) {
2377 cond_resched();
2378
2379 leaf = path->nodes[0];
2380 slot = path->slots[0];
2381
2382 if (slot >= btrfs_header_nritems(leaf)) {
2383 ret = btrfs_next_leaf(root, path);
2384 if (ret < 0) {
2385 goto out;
2386 } else if (ret > 0) {
2387 ret = 0;
2388 goto out;
2389 }
2390 continue;
2391 }
2392
2393 path->slots[0]++;
2394
2395 btrfs_item_key_to_cpu(leaf, &key, slot);
2396
2397 if (key.objectid > inum)
2398 goto out;
2399
2400 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2401 continue;
2402
2403 extent = btrfs_item_ptr(leaf, slot,
2404 struct btrfs_file_extent_item);
2405
2406 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2407 continue;
2408
2409 /*
2410 * 'offset' refers to the exact key.offset,
2411 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2412 * (key.offset - extent_offset).
2413 */
2414 if (key.offset != offset)
2415 continue;
2416
2417 extent_offset = btrfs_file_extent_offset(leaf, extent);
2418 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2419
2420 if (extent_offset >= old->extent_offset + old->offset +
2421 old->len || extent_offset + num_bytes <=
2422 old->extent_offset + old->offset)
2423 continue;
2424 break;
2425 }
2426
2427 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2428 if (!backref) {
2429 ret = -ENOENT;
2430 goto out;
2431 }
2432
2433 backref->root_id = root_id;
2434 backref->inum = inum;
2435 backref->file_pos = offset;
2436 backref->num_bytes = num_bytes;
2437 backref->extent_offset = extent_offset;
2438 backref->generation = btrfs_file_extent_generation(leaf, extent);
2439 backref->old = old;
2440 backref_insert(&new->root, backref);
2441 old->count++;
2442 out:
2443 btrfs_release_path(path);
2444 WARN_ON(ret);
2445 return ret;
2446 }
2447
2448 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2449 struct new_sa_defrag_extent *new)
2450 {
2451 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2452 struct old_sa_defrag_extent *old, *tmp;
2453 int ret;
2454
2455 new->path = path;
2456
2457 list_for_each_entry_safe(old, tmp, &new->head, list) {
2458 ret = iterate_inodes_from_logical(old->bytenr +
2459 old->extent_offset, fs_info,
2460 path, record_one_backref,
2461 old);
2462 if (ret < 0 && ret != -ENOENT)
2463 return false;
2464
2465 /* no backref to be processed for this extent */
2466 if (!old->count) {
2467 list_del(&old->list);
2468 kfree(old);
2469 }
2470 }
2471
2472 if (list_empty(&new->head))
2473 return false;
2474
2475 return true;
2476 }
2477
2478 static int relink_is_mergable(struct extent_buffer *leaf,
2479 struct btrfs_file_extent_item *fi,
2480 struct new_sa_defrag_extent *new)
2481 {
2482 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2483 return 0;
2484
2485 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2486 return 0;
2487
2488 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2489 return 0;
2490
2491 if (btrfs_file_extent_encryption(leaf, fi) ||
2492 btrfs_file_extent_other_encoding(leaf, fi))
2493 return 0;
2494
2495 return 1;
2496 }
2497
2498 /*
2499 * Note the backref might has changed, and in this case we just return 0.
2500 */
2501 static noinline int relink_extent_backref(struct btrfs_path *path,
2502 struct sa_defrag_extent_backref *prev,
2503 struct sa_defrag_extent_backref *backref)
2504 {
2505 struct btrfs_file_extent_item *extent;
2506 struct btrfs_file_extent_item *item;
2507 struct btrfs_ordered_extent *ordered;
2508 struct btrfs_trans_handle *trans;
2509 struct btrfs_root *root;
2510 struct btrfs_key key;
2511 struct extent_buffer *leaf;
2512 struct old_sa_defrag_extent *old = backref->old;
2513 struct new_sa_defrag_extent *new = old->new;
2514 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2515 struct inode *inode;
2516 struct extent_state *cached = NULL;
2517 int ret = 0;
2518 u64 start;
2519 u64 len;
2520 u64 lock_start;
2521 u64 lock_end;
2522 bool merge = false;
2523 int index;
2524
2525 if (prev && prev->root_id == backref->root_id &&
2526 prev->inum == backref->inum &&
2527 prev->file_pos + prev->num_bytes == backref->file_pos)
2528 merge = true;
2529
2530 /* step 1: get root */
2531 key.objectid = backref->root_id;
2532 key.type = BTRFS_ROOT_ITEM_KEY;
2533 key.offset = (u64)-1;
2534
2535 index = srcu_read_lock(&fs_info->subvol_srcu);
2536
2537 root = btrfs_read_fs_root_no_name(fs_info, &key);
2538 if (IS_ERR(root)) {
2539 srcu_read_unlock(&fs_info->subvol_srcu, index);
2540 if (PTR_ERR(root) == -ENOENT)
2541 return 0;
2542 return PTR_ERR(root);
2543 }
2544
2545 if (btrfs_root_readonly(root)) {
2546 srcu_read_unlock(&fs_info->subvol_srcu, index);
2547 return 0;
2548 }
2549
2550 /* step 2: get inode */
2551 key.objectid = backref->inum;
2552 key.type = BTRFS_INODE_ITEM_KEY;
2553 key.offset = 0;
2554
2555 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2556 if (IS_ERR(inode)) {
2557 srcu_read_unlock(&fs_info->subvol_srcu, index);
2558 return 0;
2559 }
2560
2561 srcu_read_unlock(&fs_info->subvol_srcu, index);
2562
2563 /* step 3: relink backref */
2564 lock_start = backref->file_pos;
2565 lock_end = backref->file_pos + backref->num_bytes - 1;
2566 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2567 &cached);
2568
2569 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2570 if (ordered) {
2571 btrfs_put_ordered_extent(ordered);
2572 goto out_unlock;
2573 }
2574
2575 trans = btrfs_join_transaction(root);
2576 if (IS_ERR(trans)) {
2577 ret = PTR_ERR(trans);
2578 goto out_unlock;
2579 }
2580
2581 key.objectid = backref->inum;
2582 key.type = BTRFS_EXTENT_DATA_KEY;
2583 key.offset = backref->file_pos;
2584
2585 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2586 if (ret < 0) {
2587 goto out_free_path;
2588 } else if (ret > 0) {
2589 ret = 0;
2590 goto out_free_path;
2591 }
2592
2593 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2594 struct btrfs_file_extent_item);
2595
2596 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2597 backref->generation)
2598 goto out_free_path;
2599
2600 btrfs_release_path(path);
2601
2602 start = backref->file_pos;
2603 if (backref->extent_offset < old->extent_offset + old->offset)
2604 start += old->extent_offset + old->offset -
2605 backref->extent_offset;
2606
2607 len = min(backref->extent_offset + backref->num_bytes,
2608 old->extent_offset + old->offset + old->len);
2609 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2610
2611 ret = btrfs_drop_extents(trans, root, inode, start,
2612 start + len, 1);
2613 if (ret)
2614 goto out_free_path;
2615 again:
2616 key.objectid = btrfs_ino(BTRFS_I(inode));
2617 key.type = BTRFS_EXTENT_DATA_KEY;
2618 key.offset = start;
2619
2620 path->leave_spinning = 1;
2621 if (merge) {
2622 struct btrfs_file_extent_item *fi;
2623 u64 extent_len;
2624 struct btrfs_key found_key;
2625
2626 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2627 if (ret < 0)
2628 goto out_free_path;
2629
2630 path->slots[0]--;
2631 leaf = path->nodes[0];
2632 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2633
2634 fi = btrfs_item_ptr(leaf, path->slots[0],
2635 struct btrfs_file_extent_item);
2636 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2637
2638 if (extent_len + found_key.offset == start &&
2639 relink_is_mergable(leaf, fi, new)) {
2640 btrfs_set_file_extent_num_bytes(leaf, fi,
2641 extent_len + len);
2642 btrfs_mark_buffer_dirty(leaf);
2643 inode_add_bytes(inode, len);
2644
2645 ret = 1;
2646 goto out_free_path;
2647 } else {
2648 merge = false;
2649 btrfs_release_path(path);
2650 goto again;
2651 }
2652 }
2653
2654 ret = btrfs_insert_empty_item(trans, root, path, &key,
2655 sizeof(*extent));
2656 if (ret) {
2657 btrfs_abort_transaction(trans, ret);
2658 goto out_free_path;
2659 }
2660
2661 leaf = path->nodes[0];
2662 item = btrfs_item_ptr(leaf, path->slots[0],
2663 struct btrfs_file_extent_item);
2664 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2665 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2666 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2667 btrfs_set_file_extent_num_bytes(leaf, item, len);
2668 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2669 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2670 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2671 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2672 btrfs_set_file_extent_encryption(leaf, item, 0);
2673 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2674
2675 btrfs_mark_buffer_dirty(leaf);
2676 inode_add_bytes(inode, len);
2677 btrfs_release_path(path);
2678
2679 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2680 new->disk_len, 0,
2681 backref->root_id, backref->inum,
2682 new->file_pos); /* start - extent_offset */
2683 if (ret) {
2684 btrfs_abort_transaction(trans, ret);
2685 goto out_free_path;
2686 }
2687
2688 ret = 1;
2689 out_free_path:
2690 btrfs_release_path(path);
2691 path->leave_spinning = 0;
2692 btrfs_end_transaction(trans);
2693 out_unlock:
2694 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2695 &cached, GFP_NOFS);
2696 iput(inode);
2697 return ret;
2698 }
2699
2700 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2701 {
2702 struct old_sa_defrag_extent *old, *tmp;
2703
2704 if (!new)
2705 return;
2706
2707 list_for_each_entry_safe(old, tmp, &new->head, list) {
2708 kfree(old);
2709 }
2710 kfree(new);
2711 }
2712
2713 static void relink_file_extents(struct new_sa_defrag_extent *new)
2714 {
2715 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2716 struct btrfs_path *path;
2717 struct sa_defrag_extent_backref *backref;
2718 struct sa_defrag_extent_backref *prev = NULL;
2719 struct inode *inode;
2720 struct btrfs_root *root;
2721 struct rb_node *node;
2722 int ret;
2723
2724 inode = new->inode;
2725 root = BTRFS_I(inode)->root;
2726
2727 path = btrfs_alloc_path();
2728 if (!path)
2729 return;
2730
2731 if (!record_extent_backrefs(path, new)) {
2732 btrfs_free_path(path);
2733 goto out;
2734 }
2735 btrfs_release_path(path);
2736
2737 while (1) {
2738 node = rb_first(&new->root);
2739 if (!node)
2740 break;
2741 rb_erase(node, &new->root);
2742
2743 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2744
2745 ret = relink_extent_backref(path, prev, backref);
2746 WARN_ON(ret < 0);
2747
2748 kfree(prev);
2749
2750 if (ret == 1)
2751 prev = backref;
2752 else
2753 prev = NULL;
2754 cond_resched();
2755 }
2756 kfree(prev);
2757
2758 btrfs_free_path(path);
2759 out:
2760 free_sa_defrag_extent(new);
2761
2762 atomic_dec(&fs_info->defrag_running);
2763 wake_up(&fs_info->transaction_wait);
2764 }
2765
2766 static struct new_sa_defrag_extent *
2767 record_old_file_extents(struct inode *inode,
2768 struct btrfs_ordered_extent *ordered)
2769 {
2770 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2771 struct btrfs_root *root = BTRFS_I(inode)->root;
2772 struct btrfs_path *path;
2773 struct btrfs_key key;
2774 struct old_sa_defrag_extent *old;
2775 struct new_sa_defrag_extent *new;
2776 int ret;
2777
2778 new = kmalloc(sizeof(*new), GFP_NOFS);
2779 if (!new)
2780 return NULL;
2781
2782 new->inode = inode;
2783 new->file_pos = ordered->file_offset;
2784 new->len = ordered->len;
2785 new->bytenr = ordered->start;
2786 new->disk_len = ordered->disk_len;
2787 new->compress_type = ordered->compress_type;
2788 new->root = RB_ROOT;
2789 INIT_LIST_HEAD(&new->head);
2790
2791 path = btrfs_alloc_path();
2792 if (!path)
2793 goto out_kfree;
2794
2795 key.objectid = btrfs_ino(BTRFS_I(inode));
2796 key.type = BTRFS_EXTENT_DATA_KEY;
2797 key.offset = new->file_pos;
2798
2799 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2800 if (ret < 0)
2801 goto out_free_path;
2802 if (ret > 0 && path->slots[0] > 0)
2803 path->slots[0]--;
2804
2805 /* find out all the old extents for the file range */
2806 while (1) {
2807 struct btrfs_file_extent_item *extent;
2808 struct extent_buffer *l;
2809 int slot;
2810 u64 num_bytes;
2811 u64 offset;
2812 u64 end;
2813 u64 disk_bytenr;
2814 u64 extent_offset;
2815
2816 l = path->nodes[0];
2817 slot = path->slots[0];
2818
2819 if (slot >= btrfs_header_nritems(l)) {
2820 ret = btrfs_next_leaf(root, path);
2821 if (ret < 0)
2822 goto out_free_path;
2823 else if (ret > 0)
2824 break;
2825 continue;
2826 }
2827
2828 btrfs_item_key_to_cpu(l, &key, slot);
2829
2830 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2831 break;
2832 if (key.type != BTRFS_EXTENT_DATA_KEY)
2833 break;
2834 if (key.offset >= new->file_pos + new->len)
2835 break;
2836
2837 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2838
2839 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2840 if (key.offset + num_bytes < new->file_pos)
2841 goto next;
2842
2843 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2844 if (!disk_bytenr)
2845 goto next;
2846
2847 extent_offset = btrfs_file_extent_offset(l, extent);
2848
2849 old = kmalloc(sizeof(*old), GFP_NOFS);
2850 if (!old)
2851 goto out_free_path;
2852
2853 offset = max(new->file_pos, key.offset);
2854 end = min(new->file_pos + new->len, key.offset + num_bytes);
2855
2856 old->bytenr = disk_bytenr;
2857 old->extent_offset = extent_offset;
2858 old->offset = offset - key.offset;
2859 old->len = end - offset;
2860 old->new = new;
2861 old->count = 0;
2862 list_add_tail(&old->list, &new->head);
2863 next:
2864 path->slots[0]++;
2865 cond_resched();
2866 }
2867
2868 btrfs_free_path(path);
2869 atomic_inc(&fs_info->defrag_running);
2870
2871 return new;
2872
2873 out_free_path:
2874 btrfs_free_path(path);
2875 out_kfree:
2876 free_sa_defrag_extent(new);
2877 return NULL;
2878 }
2879
2880 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2881 u64 start, u64 len)
2882 {
2883 struct btrfs_block_group_cache *cache;
2884
2885 cache = btrfs_lookup_block_group(fs_info, start);
2886 ASSERT(cache);
2887
2888 spin_lock(&cache->lock);
2889 cache->delalloc_bytes -= len;
2890 spin_unlock(&cache->lock);
2891
2892 btrfs_put_block_group(cache);
2893 }
2894
2895 /* as ordered data IO finishes, this gets called so we can finish
2896 * an ordered extent if the range of bytes in the file it covers are
2897 * fully written.
2898 */
2899 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2900 {
2901 struct inode *inode = ordered_extent->inode;
2902 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2903 struct btrfs_root *root = BTRFS_I(inode)->root;
2904 struct btrfs_trans_handle *trans = NULL;
2905 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2906 struct extent_state *cached_state = NULL;
2907 struct new_sa_defrag_extent *new = NULL;
2908 int compress_type = 0;
2909 int ret = 0;
2910 u64 logical_len = ordered_extent->len;
2911 bool nolock;
2912 bool truncated = false;
2913 bool range_locked = false;
2914 bool clear_new_delalloc_bytes = false;
2915
2916 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2917 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2918 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2919 clear_new_delalloc_bytes = true;
2920
2921 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2922
2923 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2924 ret = -EIO;
2925 goto out;
2926 }
2927
2928 btrfs_free_io_failure_record(BTRFS_I(inode),
2929 ordered_extent->file_offset,
2930 ordered_extent->file_offset +
2931 ordered_extent->len - 1);
2932
2933 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2934 truncated = true;
2935 logical_len = ordered_extent->truncated_len;
2936 /* Truncated the entire extent, don't bother adding */
2937 if (!logical_len)
2938 goto out;
2939 }
2940
2941 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2942 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2943
2944 /*
2945 * For mwrite(mmap + memset to write) case, we still reserve
2946 * space for NOCOW range.
2947 * As NOCOW won't cause a new delayed ref, just free the space
2948 */
2949 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2950 ordered_extent->len);
2951 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2952 if (nolock)
2953 trans = btrfs_join_transaction_nolock(root);
2954 else
2955 trans = btrfs_join_transaction(root);
2956 if (IS_ERR(trans)) {
2957 ret = PTR_ERR(trans);
2958 trans = NULL;
2959 goto out;
2960 }
2961 trans->block_rsv = &fs_info->delalloc_block_rsv;
2962 ret = btrfs_update_inode_fallback(trans, root, inode);
2963 if (ret) /* -ENOMEM or corruption */
2964 btrfs_abort_transaction(trans, ret);
2965 goto out;
2966 }
2967
2968 range_locked = true;
2969 lock_extent_bits(io_tree, ordered_extent->file_offset,
2970 ordered_extent->file_offset + ordered_extent->len - 1,
2971 &cached_state);
2972
2973 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2974 ordered_extent->file_offset + ordered_extent->len - 1,
2975 EXTENT_DEFRAG, 0, cached_state);
2976 if (ret) {
2977 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2978 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2979 /* the inode is shared */
2980 new = record_old_file_extents(inode, ordered_extent);
2981
2982 clear_extent_bit(io_tree, ordered_extent->file_offset,
2983 ordered_extent->file_offset + ordered_extent->len - 1,
2984 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2985 }
2986
2987 if (nolock)
2988 trans = btrfs_join_transaction_nolock(root);
2989 else
2990 trans = btrfs_join_transaction(root);
2991 if (IS_ERR(trans)) {
2992 ret = PTR_ERR(trans);
2993 trans = NULL;
2994 goto out;
2995 }
2996
2997 trans->block_rsv = &fs_info->delalloc_block_rsv;
2998
2999 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3000 compress_type = ordered_extent->compress_type;
3001 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3002 BUG_ON(compress_type);
3003 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3004 ordered_extent->file_offset,
3005 ordered_extent->file_offset +
3006 logical_len);
3007 } else {
3008 BUG_ON(root == fs_info->tree_root);
3009 ret = insert_reserved_file_extent(trans, inode,
3010 ordered_extent->file_offset,
3011 ordered_extent->start,
3012 ordered_extent->disk_len,
3013 logical_len, logical_len,
3014 compress_type, 0, 0,
3015 BTRFS_FILE_EXTENT_REG);
3016 if (!ret)
3017 btrfs_release_delalloc_bytes(fs_info,
3018 ordered_extent->start,
3019 ordered_extent->disk_len);
3020 }
3021 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3022 ordered_extent->file_offset, ordered_extent->len,
3023 trans->transid);
3024 if (ret < 0) {
3025 btrfs_abort_transaction(trans, ret);
3026 goto out;
3027 }
3028
3029 add_pending_csums(trans, inode, &ordered_extent->list);
3030
3031 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3032 ret = btrfs_update_inode_fallback(trans, root, inode);
3033 if (ret) { /* -ENOMEM or corruption */
3034 btrfs_abort_transaction(trans, ret);
3035 goto out;
3036 }
3037 ret = 0;
3038 out:
3039 if (range_locked || clear_new_delalloc_bytes) {
3040 unsigned int clear_bits = 0;
3041
3042 if (range_locked)
3043 clear_bits |= EXTENT_LOCKED;
3044 if (clear_new_delalloc_bytes)
3045 clear_bits |= EXTENT_DELALLOC_NEW;
3046 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3047 ordered_extent->file_offset,
3048 ordered_extent->file_offset +
3049 ordered_extent->len - 1,
3050 clear_bits,
3051 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3052 0, &cached_state, GFP_NOFS);
3053 }
3054
3055 if (root != fs_info->tree_root)
3056 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3057 ordered_extent->len);
3058 if (trans)
3059 btrfs_end_transaction(trans);
3060
3061 if (ret || truncated) {
3062 u64 start, end;
3063
3064 if (truncated)
3065 start = ordered_extent->file_offset + logical_len;
3066 else
3067 start = ordered_extent->file_offset;
3068 end = ordered_extent->file_offset + ordered_extent->len - 1;
3069 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3070
3071 /* Drop the cache for the part of the extent we didn't write. */
3072 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3073
3074 /*
3075 * If the ordered extent had an IOERR or something else went
3076 * wrong we need to return the space for this ordered extent
3077 * back to the allocator. We only free the extent in the
3078 * truncated case if we didn't write out the extent at all.
3079 */
3080 if ((ret || !logical_len) &&
3081 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3082 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3083 btrfs_free_reserved_extent(fs_info,
3084 ordered_extent->start,
3085 ordered_extent->disk_len, 1);
3086 }
3087
3088
3089 /*
3090 * This needs to be done to make sure anybody waiting knows we are done
3091 * updating everything for this ordered extent.
3092 */
3093 btrfs_remove_ordered_extent(inode, ordered_extent);
3094
3095 /* for snapshot-aware defrag */
3096 if (new) {
3097 if (ret) {
3098 free_sa_defrag_extent(new);
3099 atomic_dec(&fs_info->defrag_running);
3100 } else {
3101 relink_file_extents(new);
3102 }
3103 }
3104
3105 /* once for us */
3106 btrfs_put_ordered_extent(ordered_extent);
3107 /* once for the tree */
3108 btrfs_put_ordered_extent(ordered_extent);
3109
3110 return ret;
3111 }
3112
3113 static void finish_ordered_fn(struct btrfs_work *work)
3114 {
3115 struct btrfs_ordered_extent *ordered_extent;
3116 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3117 btrfs_finish_ordered_io(ordered_extent);
3118 }
3119
3120 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3121 struct extent_state *state, int uptodate)
3122 {
3123 struct inode *inode = page->mapping->host;
3124 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3125 struct btrfs_ordered_extent *ordered_extent = NULL;
3126 struct btrfs_workqueue *wq;
3127 btrfs_work_func_t func;
3128
3129 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3130
3131 ClearPagePrivate2(page);
3132 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3133 end - start + 1, uptodate))
3134 return;
3135
3136 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3137 wq = fs_info->endio_freespace_worker;
3138 func = btrfs_freespace_write_helper;
3139 } else {
3140 wq = fs_info->endio_write_workers;
3141 func = btrfs_endio_write_helper;
3142 }
3143
3144 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3145 NULL);
3146 btrfs_queue_work(wq, &ordered_extent->work);
3147 }
3148
3149 static int __readpage_endio_check(struct inode *inode,
3150 struct btrfs_io_bio *io_bio,
3151 int icsum, struct page *page,
3152 int pgoff, u64 start, size_t len)
3153 {
3154 char *kaddr;
3155 u32 csum_expected;
3156 u32 csum = ~(u32)0;
3157
3158 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3159
3160 kaddr = kmap_atomic(page);
3161 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3162 btrfs_csum_final(csum, (u8 *)&csum);
3163 if (csum != csum_expected)
3164 goto zeroit;
3165
3166 kunmap_atomic(kaddr);
3167 return 0;
3168 zeroit:
3169 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3170 io_bio->mirror_num);
3171 memset(kaddr + pgoff, 1, len);
3172 flush_dcache_page(page);
3173 kunmap_atomic(kaddr);
3174 if (csum_expected == 0)
3175 return 0;
3176 return -EIO;
3177 }
3178
3179 /*
3180 * when reads are done, we need to check csums to verify the data is correct
3181 * if there's a match, we allow the bio to finish. If not, the code in
3182 * extent_io.c will try to find good copies for us.
3183 */
3184 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3185 u64 phy_offset, struct page *page,
3186 u64 start, u64 end, int mirror)
3187 {
3188 size_t offset = start - page_offset(page);
3189 struct inode *inode = page->mapping->host;
3190 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3191 struct btrfs_root *root = BTRFS_I(inode)->root;
3192
3193 if (PageChecked(page)) {
3194 ClearPageChecked(page);
3195 return 0;
3196 }
3197
3198 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3199 return 0;
3200
3201 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3202 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3203 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3204 return 0;
3205 }
3206
3207 phy_offset >>= inode->i_sb->s_blocksize_bits;
3208 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3209 start, (size_t)(end - start + 1));
3210 }
3211
3212 void btrfs_add_delayed_iput(struct inode *inode)
3213 {
3214 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3215 struct btrfs_inode *binode = BTRFS_I(inode);
3216
3217 if (atomic_add_unless(&inode->i_count, -1, 1))
3218 return;
3219
3220 spin_lock(&fs_info->delayed_iput_lock);
3221 if (binode->delayed_iput_count == 0) {
3222 ASSERT(list_empty(&binode->delayed_iput));
3223 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3224 } else {
3225 binode->delayed_iput_count++;
3226 }
3227 spin_unlock(&fs_info->delayed_iput_lock);
3228 }
3229
3230 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3231 {
3232
3233 spin_lock(&fs_info->delayed_iput_lock);
3234 while (!list_empty(&fs_info->delayed_iputs)) {
3235 struct btrfs_inode *inode;
3236
3237 inode = list_first_entry(&fs_info->delayed_iputs,
3238 struct btrfs_inode, delayed_iput);
3239 if (inode->delayed_iput_count) {
3240 inode->delayed_iput_count--;
3241 list_move_tail(&inode->delayed_iput,
3242 &fs_info->delayed_iputs);
3243 } else {
3244 list_del_init(&inode->delayed_iput);
3245 }
3246 spin_unlock(&fs_info->delayed_iput_lock);
3247 iput(&inode->vfs_inode);
3248 spin_lock(&fs_info->delayed_iput_lock);
3249 }
3250 spin_unlock(&fs_info->delayed_iput_lock);
3251 }
3252
3253 /*
3254 * This is called in transaction commit time. If there are no orphan
3255 * files in the subvolume, it removes orphan item and frees block_rsv
3256 * structure.
3257 */
3258 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root)
3260 {
3261 struct btrfs_fs_info *fs_info = root->fs_info;
3262 struct btrfs_block_rsv *block_rsv;
3263 int ret;
3264
3265 if (atomic_read(&root->orphan_inodes) ||
3266 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3267 return;
3268
3269 spin_lock(&root->orphan_lock);
3270 if (atomic_read(&root->orphan_inodes)) {
3271 spin_unlock(&root->orphan_lock);
3272 return;
3273 }
3274
3275 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3276 spin_unlock(&root->orphan_lock);
3277 return;
3278 }
3279
3280 block_rsv = root->orphan_block_rsv;
3281 root->orphan_block_rsv = NULL;
3282 spin_unlock(&root->orphan_lock);
3283
3284 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3285 btrfs_root_refs(&root->root_item) > 0) {
3286 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3287 root->root_key.objectid);
3288 if (ret)
3289 btrfs_abort_transaction(trans, ret);
3290 else
3291 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3292 &root->state);
3293 }
3294
3295 if (block_rsv) {
3296 WARN_ON(block_rsv->size > 0);
3297 btrfs_free_block_rsv(fs_info, block_rsv);
3298 }
3299 }
3300
3301 /*
3302 * This creates an orphan entry for the given inode in case something goes
3303 * wrong in the middle of an unlink/truncate.
3304 *
3305 * NOTE: caller of this function should reserve 5 units of metadata for
3306 * this function.
3307 */
3308 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3309 struct btrfs_inode *inode)
3310 {
3311 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3312 struct btrfs_root *root = inode->root;
3313 struct btrfs_block_rsv *block_rsv = NULL;
3314 int reserve = 0;
3315 int insert = 0;
3316 int ret;
3317
3318 if (!root->orphan_block_rsv) {
3319 block_rsv = btrfs_alloc_block_rsv(fs_info,
3320 BTRFS_BLOCK_RSV_TEMP);
3321 if (!block_rsv)
3322 return -ENOMEM;
3323 }
3324
3325 spin_lock(&root->orphan_lock);
3326 if (!root->orphan_block_rsv) {
3327 root->orphan_block_rsv = block_rsv;
3328 } else if (block_rsv) {
3329 btrfs_free_block_rsv(fs_info, block_rsv);
3330 block_rsv = NULL;
3331 }
3332
3333 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3334 &inode->runtime_flags)) {
3335 #if 0
3336 /*
3337 * For proper ENOSPC handling, we should do orphan
3338 * cleanup when mounting. But this introduces backward
3339 * compatibility issue.
3340 */
3341 if (!xchg(&root->orphan_item_inserted, 1))
3342 insert = 2;
3343 else
3344 insert = 1;
3345 #endif
3346 insert = 1;
3347 atomic_inc(&root->orphan_inodes);
3348 }
3349
3350 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3351 &inode->runtime_flags))
3352 reserve = 1;
3353 spin_unlock(&root->orphan_lock);
3354
3355 /* grab metadata reservation from transaction handle */
3356 if (reserve) {
3357 ret = btrfs_orphan_reserve_metadata(trans, inode);
3358 ASSERT(!ret);
3359 if (ret) {
3360 atomic_dec(&root->orphan_inodes);
3361 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3362 &inode->runtime_flags);
3363 if (insert)
3364 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3365 &inode->runtime_flags);
3366 return ret;
3367 }
3368 }
3369
3370 /* insert an orphan item to track this unlinked/truncated file */
3371 if (insert >= 1) {
3372 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3373 if (ret) {
3374 atomic_dec(&root->orphan_inodes);
3375 if (reserve) {
3376 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3377 &inode->runtime_flags);
3378 btrfs_orphan_release_metadata(inode);
3379 }
3380 if (ret != -EEXIST) {
3381 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3382 &inode->runtime_flags);
3383 btrfs_abort_transaction(trans, ret);
3384 return ret;
3385 }
3386 }
3387 ret = 0;
3388 }
3389
3390 /* insert an orphan item to track subvolume contains orphan files */
3391 if (insert >= 2) {
3392 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3393 root->root_key.objectid);
3394 if (ret && ret != -EEXIST) {
3395 btrfs_abort_transaction(trans, ret);
3396 return ret;
3397 }
3398 }
3399 return 0;
3400 }
3401
3402 /*
3403 * We have done the truncate/delete so we can go ahead and remove the orphan
3404 * item for this particular inode.
3405 */
3406 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3407 struct btrfs_inode *inode)
3408 {
3409 struct btrfs_root *root = inode->root;
3410 int delete_item = 0;
3411 int release_rsv = 0;
3412 int ret = 0;
3413
3414 spin_lock(&root->orphan_lock);
3415 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3416 &inode->runtime_flags))
3417 delete_item = 1;
3418
3419 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3420 &inode->runtime_flags))
3421 release_rsv = 1;
3422 spin_unlock(&root->orphan_lock);
3423
3424 if (delete_item) {
3425 atomic_dec(&root->orphan_inodes);
3426 if (trans)
3427 ret = btrfs_del_orphan_item(trans, root,
3428 btrfs_ino(inode));
3429 }
3430
3431 if (release_rsv)
3432 btrfs_orphan_release_metadata(inode);
3433
3434 return ret;
3435 }
3436
3437 /*
3438 * this cleans up any orphans that may be left on the list from the last use
3439 * of this root.
3440 */
3441 int btrfs_orphan_cleanup(struct btrfs_root *root)
3442 {
3443 struct btrfs_fs_info *fs_info = root->fs_info;
3444 struct btrfs_path *path;
3445 struct extent_buffer *leaf;
3446 struct btrfs_key key, found_key;
3447 struct btrfs_trans_handle *trans;
3448 struct inode *inode;
3449 u64 last_objectid = 0;
3450 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3451
3452 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3453 return 0;
3454
3455 path = btrfs_alloc_path();
3456 if (!path) {
3457 ret = -ENOMEM;
3458 goto out;
3459 }
3460 path->reada = READA_BACK;
3461
3462 key.objectid = BTRFS_ORPHAN_OBJECTID;
3463 key.type = BTRFS_ORPHAN_ITEM_KEY;
3464 key.offset = (u64)-1;
3465
3466 while (1) {
3467 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3468 if (ret < 0)
3469 goto out;
3470
3471 /*
3472 * if ret == 0 means we found what we were searching for, which
3473 * is weird, but possible, so only screw with path if we didn't
3474 * find the key and see if we have stuff that matches
3475 */
3476 if (ret > 0) {
3477 ret = 0;
3478 if (path->slots[0] == 0)
3479 break;
3480 path->slots[0]--;
3481 }
3482
3483 /* pull out the item */
3484 leaf = path->nodes[0];
3485 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3486
3487 /* make sure the item matches what we want */
3488 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3489 break;
3490 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3491 break;
3492
3493 /* release the path since we're done with it */
3494 btrfs_release_path(path);
3495
3496 /*
3497 * this is where we are basically btrfs_lookup, without the
3498 * crossing root thing. we store the inode number in the
3499 * offset of the orphan item.
3500 */
3501
3502 if (found_key.offset == last_objectid) {
3503 btrfs_err(fs_info,
3504 "Error removing orphan entry, stopping orphan cleanup");
3505 ret = -EINVAL;
3506 goto out;
3507 }
3508
3509 last_objectid = found_key.offset;
3510
3511 found_key.objectid = found_key.offset;
3512 found_key.type = BTRFS_INODE_ITEM_KEY;
3513 found_key.offset = 0;
3514 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3515 ret = PTR_ERR_OR_ZERO(inode);
3516 if (ret && ret != -ENOENT)
3517 goto out;
3518
3519 if (ret == -ENOENT && root == fs_info->tree_root) {
3520 struct btrfs_root *dead_root;
3521 struct btrfs_fs_info *fs_info = root->fs_info;
3522 int is_dead_root = 0;
3523
3524 /*
3525 * this is an orphan in the tree root. Currently these
3526 * could come from 2 sources:
3527 * a) a snapshot deletion in progress
3528 * b) a free space cache inode
3529 * We need to distinguish those two, as the snapshot
3530 * orphan must not get deleted.
3531 * find_dead_roots already ran before us, so if this
3532 * is a snapshot deletion, we should find the root
3533 * in the dead_roots list
3534 */
3535 spin_lock(&fs_info->trans_lock);
3536 list_for_each_entry(dead_root, &fs_info->dead_roots,
3537 root_list) {
3538 if (dead_root->root_key.objectid ==
3539 found_key.objectid) {
3540 is_dead_root = 1;
3541 break;
3542 }
3543 }
3544 spin_unlock(&fs_info->trans_lock);
3545 if (is_dead_root) {
3546 /* prevent this orphan from being found again */
3547 key.offset = found_key.objectid - 1;
3548 continue;
3549 }
3550 }
3551 /*
3552 * Inode is already gone but the orphan item is still there,
3553 * kill the orphan item.
3554 */
3555 if (ret == -ENOENT) {
3556 trans = btrfs_start_transaction(root, 1);
3557 if (IS_ERR(trans)) {
3558 ret = PTR_ERR(trans);
3559 goto out;
3560 }
3561 btrfs_debug(fs_info, "auto deleting %Lu",
3562 found_key.objectid);
3563 ret = btrfs_del_orphan_item(trans, root,
3564 found_key.objectid);
3565 btrfs_end_transaction(trans);
3566 if (ret)
3567 goto out;
3568 continue;
3569 }
3570
3571 /*
3572 * add this inode to the orphan list so btrfs_orphan_del does
3573 * the proper thing when we hit it
3574 */
3575 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3576 &BTRFS_I(inode)->runtime_flags);
3577 atomic_inc(&root->orphan_inodes);
3578
3579 /* if we have links, this was a truncate, lets do that */
3580 if (inode->i_nlink) {
3581 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3582 iput(inode);
3583 continue;
3584 }
3585 nr_truncate++;
3586
3587 /* 1 for the orphan item deletion. */
3588 trans = btrfs_start_transaction(root, 1);
3589 if (IS_ERR(trans)) {
3590 iput(inode);
3591 ret = PTR_ERR(trans);
3592 goto out;
3593 }
3594 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3595 btrfs_end_transaction(trans);
3596 if (ret) {
3597 iput(inode);
3598 goto out;
3599 }
3600
3601 ret = btrfs_truncate(inode);
3602 if (ret)
3603 btrfs_orphan_del(NULL, BTRFS_I(inode));
3604 } else {
3605 nr_unlink++;
3606 }
3607
3608 /* this will do delete_inode and everything for us */
3609 iput(inode);
3610 if (ret)
3611 goto out;
3612 }
3613 /* release the path since we're done with it */
3614 btrfs_release_path(path);
3615
3616 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3617
3618 if (root->orphan_block_rsv)
3619 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3620 (u64)-1);
3621
3622 if (root->orphan_block_rsv ||
3623 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3624 trans = btrfs_join_transaction(root);
3625 if (!IS_ERR(trans))
3626 btrfs_end_transaction(trans);
3627 }
3628
3629 if (nr_unlink)
3630 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3631 if (nr_truncate)
3632 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3633
3634 out:
3635 if (ret)
3636 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3637 btrfs_free_path(path);
3638 return ret;
3639 }
3640
3641 /*
3642 * very simple check to peek ahead in the leaf looking for xattrs. If we
3643 * don't find any xattrs, we know there can't be any acls.
3644 *
3645 * slot is the slot the inode is in, objectid is the objectid of the inode
3646 */
3647 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3648 int slot, u64 objectid,
3649 int *first_xattr_slot)
3650 {
3651 u32 nritems = btrfs_header_nritems(leaf);
3652 struct btrfs_key found_key;
3653 static u64 xattr_access = 0;
3654 static u64 xattr_default = 0;
3655 int scanned = 0;
3656
3657 if (!xattr_access) {
3658 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3659 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3660 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3661 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3662 }
3663
3664 slot++;
3665 *first_xattr_slot = -1;
3666 while (slot < nritems) {
3667 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3668
3669 /* we found a different objectid, there must not be acls */
3670 if (found_key.objectid != objectid)
3671 return 0;
3672
3673 /* we found an xattr, assume we've got an acl */
3674 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3675 if (*first_xattr_slot == -1)
3676 *first_xattr_slot = slot;
3677 if (found_key.offset == xattr_access ||
3678 found_key.offset == xattr_default)
3679 return 1;
3680 }
3681
3682 /*
3683 * we found a key greater than an xattr key, there can't
3684 * be any acls later on
3685 */
3686 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3687 return 0;
3688
3689 slot++;
3690 scanned++;
3691
3692 /*
3693 * it goes inode, inode backrefs, xattrs, extents,
3694 * so if there are a ton of hard links to an inode there can
3695 * be a lot of backrefs. Don't waste time searching too hard,
3696 * this is just an optimization
3697 */
3698 if (scanned >= 8)
3699 break;
3700 }
3701 /* we hit the end of the leaf before we found an xattr or
3702 * something larger than an xattr. We have to assume the inode
3703 * has acls
3704 */
3705 if (*first_xattr_slot == -1)
3706 *first_xattr_slot = slot;
3707 return 1;
3708 }
3709
3710 /*
3711 * read an inode from the btree into the in-memory inode
3712 */
3713 static int btrfs_read_locked_inode(struct inode *inode)
3714 {
3715 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3716 struct btrfs_path *path;
3717 struct extent_buffer *leaf;
3718 struct btrfs_inode_item *inode_item;
3719 struct btrfs_root *root = BTRFS_I(inode)->root;
3720 struct btrfs_key location;
3721 unsigned long ptr;
3722 int maybe_acls;
3723 u32 rdev;
3724 int ret;
3725 bool filled = false;
3726 int first_xattr_slot;
3727
3728 ret = btrfs_fill_inode(inode, &rdev);
3729 if (!ret)
3730 filled = true;
3731
3732 path = btrfs_alloc_path();
3733 if (!path) {
3734 ret = -ENOMEM;
3735 goto make_bad;
3736 }
3737
3738 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3739
3740 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3741 if (ret) {
3742 if (ret > 0)
3743 ret = -ENOENT;
3744 goto make_bad;
3745 }
3746
3747 leaf = path->nodes[0];
3748
3749 if (filled)
3750 goto cache_index;
3751
3752 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3753 struct btrfs_inode_item);
3754 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3755 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3756 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3757 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3758 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3759
3760 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3761 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3762
3763 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3764 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3765
3766 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3767 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3768
3769 BTRFS_I(inode)->i_otime.tv_sec =
3770 btrfs_timespec_sec(leaf, &inode_item->otime);
3771 BTRFS_I(inode)->i_otime.tv_nsec =
3772 btrfs_timespec_nsec(leaf, &inode_item->otime);
3773
3774 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3775 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3776 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3777
3778 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3779 inode->i_generation = BTRFS_I(inode)->generation;
3780 inode->i_rdev = 0;
3781 rdev = btrfs_inode_rdev(leaf, inode_item);
3782
3783 BTRFS_I(inode)->index_cnt = (u64)-1;
3784 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3785
3786 cache_index:
3787 /*
3788 * If we were modified in the current generation and evicted from memory
3789 * and then re-read we need to do a full sync since we don't have any
3790 * idea about which extents were modified before we were evicted from
3791 * cache.
3792 *
3793 * This is required for both inode re-read from disk and delayed inode
3794 * in delayed_nodes_tree.
3795 */
3796 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3797 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3798 &BTRFS_I(inode)->runtime_flags);
3799
3800 /*
3801 * We don't persist the id of the transaction where an unlink operation
3802 * against the inode was last made. So here we assume the inode might
3803 * have been evicted, and therefore the exact value of last_unlink_trans
3804 * lost, and set it to last_trans to avoid metadata inconsistencies
3805 * between the inode and its parent if the inode is fsync'ed and the log
3806 * replayed. For example, in the scenario:
3807 *
3808 * touch mydir/foo
3809 * ln mydir/foo mydir/bar
3810 * sync
3811 * unlink mydir/bar
3812 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3813 * xfs_io -c fsync mydir/foo
3814 * <power failure>
3815 * mount fs, triggers fsync log replay
3816 *
3817 * We must make sure that when we fsync our inode foo we also log its
3818 * parent inode, otherwise after log replay the parent still has the
3819 * dentry with the "bar" name but our inode foo has a link count of 1
3820 * and doesn't have an inode ref with the name "bar" anymore.
3821 *
3822 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3823 * but it guarantees correctness at the expense of occasional full
3824 * transaction commits on fsync if our inode is a directory, or if our
3825 * inode is not a directory, logging its parent unnecessarily.
3826 */
3827 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3828
3829 path->slots[0]++;
3830 if (inode->i_nlink != 1 ||
3831 path->slots[0] >= btrfs_header_nritems(leaf))
3832 goto cache_acl;
3833
3834 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3835 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3836 goto cache_acl;
3837
3838 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3839 if (location.type == BTRFS_INODE_REF_KEY) {
3840 struct btrfs_inode_ref *ref;
3841
3842 ref = (struct btrfs_inode_ref *)ptr;
3843 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3844 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3845 struct btrfs_inode_extref *extref;
3846
3847 extref = (struct btrfs_inode_extref *)ptr;
3848 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3849 extref);
3850 }
3851 cache_acl:
3852 /*
3853 * try to precache a NULL acl entry for files that don't have
3854 * any xattrs or acls
3855 */
3856 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3857 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3858 if (first_xattr_slot != -1) {
3859 path->slots[0] = first_xattr_slot;
3860 ret = btrfs_load_inode_props(inode, path);
3861 if (ret)
3862 btrfs_err(fs_info,
3863 "error loading props for ino %llu (root %llu): %d",
3864 btrfs_ino(BTRFS_I(inode)),
3865 root->root_key.objectid, ret);
3866 }
3867 btrfs_free_path(path);
3868
3869 if (!maybe_acls)
3870 cache_no_acl(inode);
3871
3872 switch (inode->i_mode & S_IFMT) {
3873 case S_IFREG:
3874 inode->i_mapping->a_ops = &btrfs_aops;
3875 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3876 inode->i_fop = &btrfs_file_operations;
3877 inode->i_op = &btrfs_file_inode_operations;
3878 break;
3879 case S_IFDIR:
3880 inode->i_fop = &btrfs_dir_file_operations;
3881 inode->i_op = &btrfs_dir_inode_operations;
3882 break;
3883 case S_IFLNK:
3884 inode->i_op = &btrfs_symlink_inode_operations;
3885 inode_nohighmem(inode);
3886 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3887 break;
3888 default:
3889 inode->i_op = &btrfs_special_inode_operations;
3890 init_special_inode(inode, inode->i_mode, rdev);
3891 break;
3892 }
3893
3894 btrfs_update_iflags(inode);
3895 return 0;
3896
3897 make_bad:
3898 btrfs_free_path(path);
3899 make_bad_inode(inode);
3900 return ret;
3901 }
3902
3903 /*
3904 * given a leaf and an inode, copy the inode fields into the leaf
3905 */
3906 static void fill_inode_item(struct btrfs_trans_handle *trans,
3907 struct extent_buffer *leaf,
3908 struct btrfs_inode_item *item,
3909 struct inode *inode)
3910 {
3911 struct btrfs_map_token token;
3912
3913 btrfs_init_map_token(&token);
3914
3915 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3916 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3917 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3918 &token);
3919 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3920 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3921
3922 btrfs_set_token_timespec_sec(leaf, &item->atime,
3923 inode->i_atime.tv_sec, &token);
3924 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3925 inode->i_atime.tv_nsec, &token);
3926
3927 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3928 inode->i_mtime.tv_sec, &token);
3929 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3930 inode->i_mtime.tv_nsec, &token);
3931
3932 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3933 inode->i_ctime.tv_sec, &token);
3934 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3935 inode->i_ctime.tv_nsec, &token);
3936
3937 btrfs_set_token_timespec_sec(leaf, &item->otime,
3938 BTRFS_I(inode)->i_otime.tv_sec, &token);
3939 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3940 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3941
3942 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3943 &token);
3944 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3945 &token);
3946 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3947 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3948 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3949 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3950 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3951 }
3952
3953 /*
3954 * copy everything in the in-memory inode into the btree.
3955 */
3956 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3957 struct btrfs_root *root, struct inode *inode)
3958 {
3959 struct btrfs_inode_item *inode_item;
3960 struct btrfs_path *path;
3961 struct extent_buffer *leaf;
3962 int ret;
3963
3964 path = btrfs_alloc_path();
3965 if (!path)
3966 return -ENOMEM;
3967
3968 path->leave_spinning = 1;
3969 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3970 1);
3971 if (ret) {
3972 if (ret > 0)
3973 ret = -ENOENT;
3974 goto failed;
3975 }
3976
3977 leaf = path->nodes[0];
3978 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3979 struct btrfs_inode_item);
3980
3981 fill_inode_item(trans, leaf, inode_item, inode);
3982 btrfs_mark_buffer_dirty(leaf);
3983 btrfs_set_inode_last_trans(trans, inode);
3984 ret = 0;
3985 failed:
3986 btrfs_free_path(path);
3987 return ret;
3988 }
3989
3990 /*
3991 * copy everything in the in-memory inode into the btree.
3992 */
3993 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3994 struct btrfs_root *root, struct inode *inode)
3995 {
3996 struct btrfs_fs_info *fs_info = root->fs_info;
3997 int ret;
3998
3999 /*
4000 * If the inode is a free space inode, we can deadlock during commit
4001 * if we put it into the delayed code.
4002 *
4003 * The data relocation inode should also be directly updated
4004 * without delay
4005 */
4006 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4007 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4008 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4009 btrfs_update_root_times(trans, root);
4010
4011 ret = btrfs_delayed_update_inode(trans, root, inode);
4012 if (!ret)
4013 btrfs_set_inode_last_trans(trans, inode);
4014 return ret;
4015 }
4016
4017 return btrfs_update_inode_item(trans, root, inode);
4018 }
4019
4020 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4021 struct btrfs_root *root,
4022 struct inode *inode)
4023 {
4024 int ret;
4025
4026 ret = btrfs_update_inode(trans, root, inode);
4027 if (ret == -ENOSPC)
4028 return btrfs_update_inode_item(trans, root, inode);
4029 return ret;
4030 }
4031
4032 /*
4033 * unlink helper that gets used here in inode.c and in the tree logging
4034 * recovery code. It remove a link in a directory with a given name, and
4035 * also drops the back refs in the inode to the directory
4036 */
4037 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4038 struct btrfs_root *root,
4039 struct btrfs_inode *dir,
4040 struct btrfs_inode *inode,
4041 const char *name, int name_len)
4042 {
4043 struct btrfs_fs_info *fs_info = root->fs_info;
4044 struct btrfs_path *path;
4045 int ret = 0;
4046 struct extent_buffer *leaf;
4047 struct btrfs_dir_item *di;
4048 struct btrfs_key key;
4049 u64 index;
4050 u64 ino = btrfs_ino(inode);
4051 u64 dir_ino = btrfs_ino(dir);
4052
4053 path = btrfs_alloc_path();
4054 if (!path) {
4055 ret = -ENOMEM;
4056 goto out;
4057 }
4058
4059 path->leave_spinning = 1;
4060 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4061 name, name_len, -1);
4062 if (IS_ERR(di)) {
4063 ret = PTR_ERR(di);
4064 goto err;
4065 }
4066 if (!di) {
4067 ret = -ENOENT;
4068 goto err;
4069 }
4070 leaf = path->nodes[0];
4071 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4072 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4073 if (ret)
4074 goto err;
4075 btrfs_release_path(path);
4076
4077 /*
4078 * If we don't have dir index, we have to get it by looking up
4079 * the inode ref, since we get the inode ref, remove it directly,
4080 * it is unnecessary to do delayed deletion.
4081 *
4082 * But if we have dir index, needn't search inode ref to get it.
4083 * Since the inode ref is close to the inode item, it is better
4084 * that we delay to delete it, and just do this deletion when
4085 * we update the inode item.
4086 */
4087 if (inode->dir_index) {
4088 ret = btrfs_delayed_delete_inode_ref(inode);
4089 if (!ret) {
4090 index = inode->dir_index;
4091 goto skip_backref;
4092 }
4093 }
4094
4095 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4096 dir_ino, &index);
4097 if (ret) {
4098 btrfs_info(fs_info,
4099 "failed to delete reference to %.*s, inode %llu parent %llu",
4100 name_len, name, ino, dir_ino);
4101 btrfs_abort_transaction(trans, ret);
4102 goto err;
4103 }
4104 skip_backref:
4105 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4106 if (ret) {
4107 btrfs_abort_transaction(trans, ret);
4108 goto err;
4109 }
4110
4111 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4112 dir_ino);
4113 if (ret != 0 && ret != -ENOENT) {
4114 btrfs_abort_transaction(trans, ret);
4115 goto err;
4116 }
4117
4118 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4119 index);
4120 if (ret == -ENOENT)
4121 ret = 0;
4122 else if (ret)
4123 btrfs_abort_transaction(trans, ret);
4124 err:
4125 btrfs_free_path(path);
4126 if (ret)
4127 goto out;
4128
4129 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4130 inode_inc_iversion(&inode->vfs_inode);
4131 inode_inc_iversion(&dir->vfs_inode);
4132 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4133 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4134 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4135 out:
4136 return ret;
4137 }
4138
4139 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4140 struct btrfs_root *root,
4141 struct btrfs_inode *dir, struct btrfs_inode *inode,
4142 const char *name, int name_len)
4143 {
4144 int ret;
4145 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4146 if (!ret) {
4147 drop_nlink(&inode->vfs_inode);
4148 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4149 }
4150 return ret;
4151 }
4152
4153 /*
4154 * helper to start transaction for unlink and rmdir.
4155 *
4156 * unlink and rmdir are special in btrfs, they do not always free space, so
4157 * if we cannot make our reservations the normal way try and see if there is
4158 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4159 * allow the unlink to occur.
4160 */
4161 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4162 {
4163 struct btrfs_root *root = BTRFS_I(dir)->root;
4164
4165 /*
4166 * 1 for the possible orphan item
4167 * 1 for the dir item
4168 * 1 for the dir index
4169 * 1 for the inode ref
4170 * 1 for the inode
4171 */
4172 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4173 }
4174
4175 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4176 {
4177 struct btrfs_root *root = BTRFS_I(dir)->root;
4178 struct btrfs_trans_handle *trans;
4179 struct inode *inode = d_inode(dentry);
4180 int ret;
4181
4182 trans = __unlink_start_trans(dir);
4183 if (IS_ERR(trans))
4184 return PTR_ERR(trans);
4185
4186 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4187 0);
4188
4189 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4190 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4191 dentry->d_name.len);
4192 if (ret)
4193 goto out;
4194
4195 if (inode->i_nlink == 0) {
4196 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4197 if (ret)
4198 goto out;
4199 }
4200
4201 out:
4202 btrfs_end_transaction(trans);
4203 btrfs_btree_balance_dirty(root->fs_info);
4204 return ret;
4205 }
4206
4207 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root,
4209 struct inode *dir, u64 objectid,
4210 const char *name, int name_len)
4211 {
4212 struct btrfs_fs_info *fs_info = root->fs_info;
4213 struct btrfs_path *path;
4214 struct extent_buffer *leaf;
4215 struct btrfs_dir_item *di;
4216 struct btrfs_key key;
4217 u64 index;
4218 int ret;
4219 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4220
4221 path = btrfs_alloc_path();
4222 if (!path)
4223 return -ENOMEM;
4224
4225 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4226 name, name_len, -1);
4227 if (IS_ERR_OR_NULL(di)) {
4228 if (!di)
4229 ret = -ENOENT;
4230 else
4231 ret = PTR_ERR(di);
4232 goto out;
4233 }
4234
4235 leaf = path->nodes[0];
4236 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4237 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4238 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4239 if (ret) {
4240 btrfs_abort_transaction(trans, ret);
4241 goto out;
4242 }
4243 btrfs_release_path(path);
4244
4245 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4246 root->root_key.objectid, dir_ino,
4247 &index, name, name_len);
4248 if (ret < 0) {
4249 if (ret != -ENOENT) {
4250 btrfs_abort_transaction(trans, ret);
4251 goto out;
4252 }
4253 di = btrfs_search_dir_index_item(root, path, dir_ino,
4254 name, name_len);
4255 if (IS_ERR_OR_NULL(di)) {
4256 if (!di)
4257 ret = -ENOENT;
4258 else
4259 ret = PTR_ERR(di);
4260 btrfs_abort_transaction(trans, ret);
4261 goto out;
4262 }
4263
4264 leaf = path->nodes[0];
4265 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4266 btrfs_release_path(path);
4267 index = key.offset;
4268 }
4269 btrfs_release_path(path);
4270
4271 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4272 if (ret) {
4273 btrfs_abort_transaction(trans, ret);
4274 goto out;
4275 }
4276
4277 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4278 inode_inc_iversion(dir);
4279 dir->i_mtime = dir->i_ctime = current_time(dir);
4280 ret = btrfs_update_inode_fallback(trans, root, dir);
4281 if (ret)
4282 btrfs_abort_transaction(trans, ret);
4283 out:
4284 btrfs_free_path(path);
4285 return ret;
4286 }
4287
4288 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4289 {
4290 struct inode *inode = d_inode(dentry);
4291 int err = 0;
4292 struct btrfs_root *root = BTRFS_I(dir)->root;
4293 struct btrfs_trans_handle *trans;
4294 u64 last_unlink_trans;
4295
4296 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4297 return -ENOTEMPTY;
4298 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4299 return -EPERM;
4300
4301 trans = __unlink_start_trans(dir);
4302 if (IS_ERR(trans))
4303 return PTR_ERR(trans);
4304
4305 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4306 err = btrfs_unlink_subvol(trans, root, dir,
4307 BTRFS_I(inode)->location.objectid,
4308 dentry->d_name.name,
4309 dentry->d_name.len);
4310 goto out;
4311 }
4312
4313 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4314 if (err)
4315 goto out;
4316
4317 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4318
4319 /* now the directory is empty */
4320 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4321 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4322 dentry->d_name.len);
4323 if (!err) {
4324 btrfs_i_size_write(BTRFS_I(inode), 0);
4325 /*
4326 * Propagate the last_unlink_trans value of the deleted dir to
4327 * its parent directory. This is to prevent an unrecoverable
4328 * log tree in the case we do something like this:
4329 * 1) create dir foo
4330 * 2) create snapshot under dir foo
4331 * 3) delete the snapshot
4332 * 4) rmdir foo
4333 * 5) mkdir foo
4334 * 6) fsync foo or some file inside foo
4335 */
4336 if (last_unlink_trans >= trans->transid)
4337 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4338 }
4339 out:
4340 btrfs_end_transaction(trans);
4341 btrfs_btree_balance_dirty(root->fs_info);
4342
4343 return err;
4344 }
4345
4346 static int truncate_space_check(struct btrfs_trans_handle *trans,
4347 struct btrfs_root *root,
4348 u64 bytes_deleted)
4349 {
4350 struct btrfs_fs_info *fs_info = root->fs_info;
4351 int ret;
4352
4353 /*
4354 * This is only used to apply pressure to the enospc system, we don't
4355 * intend to use this reservation at all.
4356 */
4357 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4358 bytes_deleted *= fs_info->nodesize;
4359 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4360 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4361 if (!ret) {
4362 trace_btrfs_space_reservation(fs_info, "transaction",
4363 trans->transid,
4364 bytes_deleted, 1);
4365 trans->bytes_reserved += bytes_deleted;
4366 }
4367 return ret;
4368
4369 }
4370
4371 static int truncate_inline_extent(struct inode *inode,
4372 struct btrfs_path *path,
4373 struct btrfs_key *found_key,
4374 const u64 item_end,
4375 const u64 new_size)
4376 {
4377 struct extent_buffer *leaf = path->nodes[0];
4378 int slot = path->slots[0];
4379 struct btrfs_file_extent_item *fi;
4380 u32 size = (u32)(new_size - found_key->offset);
4381 struct btrfs_root *root = BTRFS_I(inode)->root;
4382
4383 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4384
4385 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4386 loff_t offset = new_size;
4387 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4388
4389 /*
4390 * Zero out the remaining of the last page of our inline extent,
4391 * instead of directly truncating our inline extent here - that
4392 * would be much more complex (decompressing all the data, then
4393 * compressing the truncated data, which might be bigger than
4394 * the size of the inline extent, resize the extent, etc).
4395 * We release the path because to get the page we might need to
4396 * read the extent item from disk (data not in the page cache).
4397 */
4398 btrfs_release_path(path);
4399 return btrfs_truncate_block(inode, offset, page_end - offset,
4400 0);
4401 }
4402
4403 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4404 size = btrfs_file_extent_calc_inline_size(size);
4405 btrfs_truncate_item(root->fs_info, path, size, 1);
4406
4407 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4408 inode_sub_bytes(inode, item_end + 1 - new_size);
4409
4410 return 0;
4411 }
4412
4413 /*
4414 * this can truncate away extent items, csum items and directory items.
4415 * It starts at a high offset and removes keys until it can't find
4416 * any higher than new_size
4417 *
4418 * csum items that cross the new i_size are truncated to the new size
4419 * as well.
4420 *
4421 * min_type is the minimum key type to truncate down to. If set to 0, this
4422 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4423 */
4424 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4425 struct btrfs_root *root,
4426 struct inode *inode,
4427 u64 new_size, u32 min_type)
4428 {
4429 struct btrfs_fs_info *fs_info = root->fs_info;
4430 struct btrfs_path *path;
4431 struct extent_buffer *leaf;
4432 struct btrfs_file_extent_item *fi;
4433 struct btrfs_key key;
4434 struct btrfs_key found_key;
4435 u64 extent_start = 0;
4436 u64 extent_num_bytes = 0;
4437 u64 extent_offset = 0;
4438 u64 item_end = 0;
4439 u64 last_size = new_size;
4440 u32 found_type = (u8)-1;
4441 int found_extent;
4442 int del_item;
4443 int pending_del_nr = 0;
4444 int pending_del_slot = 0;
4445 int extent_type = -1;
4446 int ret;
4447 int err = 0;
4448 u64 ino = btrfs_ino(BTRFS_I(inode));
4449 u64 bytes_deleted = 0;
4450 bool be_nice = 0;
4451 bool should_throttle = 0;
4452 bool should_end = 0;
4453
4454 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4455
4456 /*
4457 * for non-free space inodes and ref cows, we want to back off from
4458 * time to time
4459 */
4460 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4461 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4462 be_nice = 1;
4463
4464 path = btrfs_alloc_path();
4465 if (!path)
4466 return -ENOMEM;
4467 path->reada = READA_BACK;
4468
4469 /*
4470 * We want to drop from the next block forward in case this new size is
4471 * not block aligned since we will be keeping the last block of the
4472 * extent just the way it is.
4473 */
4474 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4475 root == fs_info->tree_root)
4476 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4477 fs_info->sectorsize),
4478 (u64)-1, 0);
4479
4480 /*
4481 * This function is also used to drop the items in the log tree before
4482 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4483 * it is used to drop the loged items. So we shouldn't kill the delayed
4484 * items.
4485 */
4486 if (min_type == 0 && root == BTRFS_I(inode)->root)
4487 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4488
4489 key.objectid = ino;
4490 key.offset = (u64)-1;
4491 key.type = (u8)-1;
4492
4493 search_again:
4494 /*
4495 * with a 16K leaf size and 128MB extents, you can actually queue
4496 * up a huge file in a single leaf. Most of the time that
4497 * bytes_deleted is > 0, it will be huge by the time we get here
4498 */
4499 if (be_nice && bytes_deleted > SZ_32M) {
4500 if (btrfs_should_end_transaction(trans)) {
4501 err = -EAGAIN;
4502 goto error;
4503 }
4504 }
4505
4506
4507 path->leave_spinning = 1;
4508 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4509 if (ret < 0) {
4510 err = ret;
4511 goto out;
4512 }
4513
4514 if (ret > 0) {
4515 /* there are no items in the tree for us to truncate, we're
4516 * done
4517 */
4518 if (path->slots[0] == 0)
4519 goto out;
4520 path->slots[0]--;
4521 }
4522
4523 while (1) {
4524 fi = NULL;
4525 leaf = path->nodes[0];
4526 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4527 found_type = found_key.type;
4528
4529 if (found_key.objectid != ino)
4530 break;
4531
4532 if (found_type < min_type)
4533 break;
4534
4535 item_end = found_key.offset;
4536 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4537 fi = btrfs_item_ptr(leaf, path->slots[0],
4538 struct btrfs_file_extent_item);
4539 extent_type = btrfs_file_extent_type(leaf, fi);
4540 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4541 item_end +=
4542 btrfs_file_extent_num_bytes(leaf, fi);
4543
4544 trace_btrfs_truncate_show_fi_regular(
4545 BTRFS_I(inode), leaf, fi,
4546 found_key.offset);
4547 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4548 item_end += btrfs_file_extent_inline_len(leaf,
4549 path->slots[0], fi);
4550
4551 trace_btrfs_truncate_show_fi_inline(
4552 BTRFS_I(inode), leaf, fi, path->slots[0],
4553 found_key.offset);
4554 }
4555 item_end--;
4556 }
4557 if (found_type > min_type) {
4558 del_item = 1;
4559 } else {
4560 if (item_end < new_size)
4561 break;
4562 if (found_key.offset >= new_size)
4563 del_item = 1;
4564 else
4565 del_item = 0;
4566 }
4567 found_extent = 0;
4568 /* FIXME, shrink the extent if the ref count is only 1 */
4569 if (found_type != BTRFS_EXTENT_DATA_KEY)
4570 goto delete;
4571
4572 if (del_item)
4573 last_size = found_key.offset;
4574 else
4575 last_size = new_size;
4576
4577 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4578 u64 num_dec;
4579 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4580 if (!del_item) {
4581 u64 orig_num_bytes =
4582 btrfs_file_extent_num_bytes(leaf, fi);
4583 extent_num_bytes = ALIGN(new_size -
4584 found_key.offset,
4585 fs_info->sectorsize);
4586 btrfs_set_file_extent_num_bytes(leaf, fi,
4587 extent_num_bytes);
4588 num_dec = (orig_num_bytes -
4589 extent_num_bytes);
4590 if (test_bit(BTRFS_ROOT_REF_COWS,
4591 &root->state) &&
4592 extent_start != 0)
4593 inode_sub_bytes(inode, num_dec);
4594 btrfs_mark_buffer_dirty(leaf);
4595 } else {
4596 extent_num_bytes =
4597 btrfs_file_extent_disk_num_bytes(leaf,
4598 fi);
4599 extent_offset = found_key.offset -
4600 btrfs_file_extent_offset(leaf, fi);
4601
4602 /* FIXME blocksize != 4096 */
4603 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4604 if (extent_start != 0) {
4605 found_extent = 1;
4606 if (test_bit(BTRFS_ROOT_REF_COWS,
4607 &root->state))
4608 inode_sub_bytes(inode, num_dec);
4609 }
4610 }
4611 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4612 /*
4613 * we can't truncate inline items that have had
4614 * special encodings
4615 */
4616 if (!del_item &&
4617 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4618 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4619
4620 /*
4621 * Need to release path in order to truncate a
4622 * compressed extent. So delete any accumulated
4623 * extent items so far.
4624 */
4625 if (btrfs_file_extent_compression(leaf, fi) !=
4626 BTRFS_COMPRESS_NONE && pending_del_nr) {
4627 err = btrfs_del_items(trans, root, path,
4628 pending_del_slot,
4629 pending_del_nr);
4630 if (err) {
4631 btrfs_abort_transaction(trans,
4632 err);
4633 goto error;
4634 }
4635 pending_del_nr = 0;
4636 }
4637
4638 err = truncate_inline_extent(inode, path,
4639 &found_key,
4640 item_end,
4641 new_size);
4642 if (err) {
4643 btrfs_abort_transaction(trans, err);
4644 goto error;
4645 }
4646 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4647 &root->state)) {
4648 inode_sub_bytes(inode, item_end + 1 - new_size);
4649 }
4650 }
4651 delete:
4652 if (del_item) {
4653 if (!pending_del_nr) {
4654 /* no pending yet, add ourselves */
4655 pending_del_slot = path->slots[0];
4656 pending_del_nr = 1;
4657 } else if (pending_del_nr &&
4658 path->slots[0] + 1 == pending_del_slot) {
4659 /* hop on the pending chunk */
4660 pending_del_nr++;
4661 pending_del_slot = path->slots[0];
4662 } else {
4663 BUG();
4664 }
4665 } else {
4666 break;
4667 }
4668 should_throttle = 0;
4669
4670 if (found_extent &&
4671 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4672 root == fs_info->tree_root)) {
4673 btrfs_set_path_blocking(path);
4674 bytes_deleted += extent_num_bytes;
4675 ret = btrfs_free_extent(trans, fs_info, extent_start,
4676 extent_num_bytes, 0,
4677 btrfs_header_owner(leaf),
4678 ino, extent_offset);
4679 BUG_ON(ret);
4680 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4681 btrfs_async_run_delayed_refs(fs_info,
4682 trans->delayed_ref_updates * 2,
4683 trans->transid, 0);
4684 if (be_nice) {
4685 if (truncate_space_check(trans, root,
4686 extent_num_bytes)) {
4687 should_end = 1;
4688 }
4689 if (btrfs_should_throttle_delayed_refs(trans,
4690 fs_info))
4691 should_throttle = 1;
4692 }
4693 }
4694
4695 if (found_type == BTRFS_INODE_ITEM_KEY)
4696 break;
4697
4698 if (path->slots[0] == 0 ||
4699 path->slots[0] != pending_del_slot ||
4700 should_throttle || should_end) {
4701 if (pending_del_nr) {
4702 ret = btrfs_del_items(trans, root, path,
4703 pending_del_slot,
4704 pending_del_nr);
4705 if (ret) {
4706 btrfs_abort_transaction(trans, ret);
4707 goto error;
4708 }
4709 pending_del_nr = 0;
4710 }
4711 btrfs_release_path(path);
4712 if (should_throttle) {
4713 unsigned long updates = trans->delayed_ref_updates;
4714 if (updates) {
4715 trans->delayed_ref_updates = 0;
4716 ret = btrfs_run_delayed_refs(trans,
4717 fs_info,
4718 updates * 2);
4719 if (ret && !err)
4720 err = ret;
4721 }
4722 }
4723 /*
4724 * if we failed to refill our space rsv, bail out
4725 * and let the transaction restart
4726 */
4727 if (should_end) {
4728 err = -EAGAIN;
4729 goto error;
4730 }
4731 goto search_again;
4732 } else {
4733 path->slots[0]--;
4734 }
4735 }
4736 out:
4737 if (pending_del_nr) {
4738 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4739 pending_del_nr);
4740 if (ret)
4741 btrfs_abort_transaction(trans, ret);
4742 }
4743 error:
4744 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4745 ASSERT(last_size >= new_size);
4746 if (!err && last_size > new_size)
4747 last_size = new_size;
4748 btrfs_ordered_update_i_size(inode, last_size, NULL);
4749 }
4750
4751 btrfs_free_path(path);
4752
4753 if (be_nice && bytes_deleted > SZ_32M) {
4754 unsigned long updates = trans->delayed_ref_updates;
4755 if (updates) {
4756 trans->delayed_ref_updates = 0;
4757 ret = btrfs_run_delayed_refs(trans, fs_info,
4758 updates * 2);
4759 if (ret && !err)
4760 err = ret;
4761 }
4762 }
4763 return err;
4764 }
4765
4766 /*
4767 * btrfs_truncate_block - read, zero a chunk and write a block
4768 * @inode - inode that we're zeroing
4769 * @from - the offset to start zeroing
4770 * @len - the length to zero, 0 to zero the entire range respective to the
4771 * offset
4772 * @front - zero up to the offset instead of from the offset on
4773 *
4774 * This will find the block for the "from" offset and cow the block and zero the
4775 * part we want to zero. This is used with truncate and hole punching.
4776 */
4777 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4778 int front)
4779 {
4780 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4781 struct address_space *mapping = inode->i_mapping;
4782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4783 struct btrfs_ordered_extent *ordered;
4784 struct extent_state *cached_state = NULL;
4785 struct extent_changeset *data_reserved = NULL;
4786 char *kaddr;
4787 u32 blocksize = fs_info->sectorsize;
4788 pgoff_t index = from >> PAGE_SHIFT;
4789 unsigned offset = from & (blocksize - 1);
4790 struct page *page;
4791 gfp_t mask = btrfs_alloc_write_mask(mapping);
4792 int ret = 0;
4793 u64 block_start;
4794 u64 block_end;
4795
4796 if ((offset & (blocksize - 1)) == 0 &&
4797 (!len || ((len & (blocksize - 1)) == 0)))
4798 goto out;
4799
4800 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4801 round_down(from, blocksize), blocksize);
4802 if (ret)
4803 goto out;
4804
4805 again:
4806 page = find_or_create_page(mapping, index, mask);
4807 if (!page) {
4808 btrfs_delalloc_release_space(inode, data_reserved,
4809 round_down(from, blocksize),
4810 blocksize);
4811 ret = -ENOMEM;
4812 goto out;
4813 }
4814
4815 block_start = round_down(from, blocksize);
4816 block_end = block_start + blocksize - 1;
4817
4818 if (!PageUptodate(page)) {
4819 ret = btrfs_readpage(NULL, page);
4820 lock_page(page);
4821 if (page->mapping != mapping) {
4822 unlock_page(page);
4823 put_page(page);
4824 goto again;
4825 }
4826 if (!PageUptodate(page)) {
4827 ret = -EIO;
4828 goto out_unlock;
4829 }
4830 }
4831 wait_on_page_writeback(page);
4832
4833 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4834 set_page_extent_mapped(page);
4835
4836 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4837 if (ordered) {
4838 unlock_extent_cached(io_tree, block_start, block_end,
4839 &cached_state, GFP_NOFS);
4840 unlock_page(page);
4841 put_page(page);
4842 btrfs_start_ordered_extent(inode, ordered, 1);
4843 btrfs_put_ordered_extent(ordered);
4844 goto again;
4845 }
4846
4847 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4848 EXTENT_DIRTY | EXTENT_DELALLOC |
4849 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4850 0, 0, &cached_state, GFP_NOFS);
4851
4852 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4853 &cached_state, 0);
4854 if (ret) {
4855 unlock_extent_cached(io_tree, block_start, block_end,
4856 &cached_state, GFP_NOFS);
4857 goto out_unlock;
4858 }
4859
4860 if (offset != blocksize) {
4861 if (!len)
4862 len = blocksize - offset;
4863 kaddr = kmap(page);
4864 if (front)
4865 memset(kaddr + (block_start - page_offset(page)),
4866 0, offset);
4867 else
4868 memset(kaddr + (block_start - page_offset(page)) + offset,
4869 0, len);
4870 flush_dcache_page(page);
4871 kunmap(page);
4872 }
4873 ClearPageChecked(page);
4874 set_page_dirty(page);
4875 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4876 GFP_NOFS);
4877
4878 out_unlock:
4879 if (ret)
4880 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4881 blocksize);
4882 unlock_page(page);
4883 put_page(page);
4884 out:
4885 extent_changeset_free(data_reserved);
4886 return ret;
4887 }
4888
4889 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4890 u64 offset, u64 len)
4891 {
4892 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4893 struct btrfs_trans_handle *trans;
4894 int ret;
4895
4896 /*
4897 * Still need to make sure the inode looks like it's been updated so
4898 * that any holes get logged if we fsync.
4899 */
4900 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4901 BTRFS_I(inode)->last_trans = fs_info->generation;
4902 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4903 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4904 return 0;
4905 }
4906
4907 /*
4908 * 1 - for the one we're dropping
4909 * 1 - for the one we're adding
4910 * 1 - for updating the inode.
4911 */
4912 trans = btrfs_start_transaction(root, 3);
4913 if (IS_ERR(trans))
4914 return PTR_ERR(trans);
4915
4916 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4917 if (ret) {
4918 btrfs_abort_transaction(trans, ret);
4919 btrfs_end_transaction(trans);
4920 return ret;
4921 }
4922
4923 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4924 offset, 0, 0, len, 0, len, 0, 0, 0);
4925 if (ret)
4926 btrfs_abort_transaction(trans, ret);
4927 else
4928 btrfs_update_inode(trans, root, inode);
4929 btrfs_end_transaction(trans);
4930 return ret;
4931 }
4932
4933 /*
4934 * This function puts in dummy file extents for the area we're creating a hole
4935 * for. So if we are truncating this file to a larger size we need to insert
4936 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4937 * the range between oldsize and size
4938 */
4939 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4940 {
4941 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4942 struct btrfs_root *root = BTRFS_I(inode)->root;
4943 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4944 struct extent_map *em = NULL;
4945 struct extent_state *cached_state = NULL;
4946 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4947 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4948 u64 block_end = ALIGN(size, fs_info->sectorsize);
4949 u64 last_byte;
4950 u64 cur_offset;
4951 u64 hole_size;
4952 int err = 0;
4953
4954 /*
4955 * If our size started in the middle of a block we need to zero out the
4956 * rest of the block before we expand the i_size, otherwise we could
4957 * expose stale data.
4958 */
4959 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4960 if (err)
4961 return err;
4962
4963 if (size <= hole_start)
4964 return 0;
4965
4966 while (1) {
4967 struct btrfs_ordered_extent *ordered;
4968
4969 lock_extent_bits(io_tree, hole_start, block_end - 1,
4970 &cached_state);
4971 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4972 block_end - hole_start);
4973 if (!ordered)
4974 break;
4975 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4976 &cached_state, GFP_NOFS);
4977 btrfs_start_ordered_extent(inode, ordered, 1);
4978 btrfs_put_ordered_extent(ordered);
4979 }
4980
4981 cur_offset = hole_start;
4982 while (1) {
4983 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4984 block_end - cur_offset, 0);
4985 if (IS_ERR(em)) {
4986 err = PTR_ERR(em);
4987 em = NULL;
4988 break;
4989 }
4990 last_byte = min(extent_map_end(em), block_end);
4991 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4992 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4993 struct extent_map *hole_em;
4994 hole_size = last_byte - cur_offset;
4995
4996 err = maybe_insert_hole(root, inode, cur_offset,
4997 hole_size);
4998 if (err)
4999 break;
5000 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5001 cur_offset + hole_size - 1, 0);
5002 hole_em = alloc_extent_map();
5003 if (!hole_em) {
5004 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5005 &BTRFS_I(inode)->runtime_flags);
5006 goto next;
5007 }
5008 hole_em->start = cur_offset;
5009 hole_em->len = hole_size;
5010 hole_em->orig_start = cur_offset;
5011
5012 hole_em->block_start = EXTENT_MAP_HOLE;
5013 hole_em->block_len = 0;
5014 hole_em->orig_block_len = 0;
5015 hole_em->ram_bytes = hole_size;
5016 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5017 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5018 hole_em->generation = fs_info->generation;
5019
5020 while (1) {
5021 write_lock(&em_tree->lock);
5022 err = add_extent_mapping(em_tree, hole_em, 1);
5023 write_unlock(&em_tree->lock);
5024 if (err != -EEXIST)
5025 break;
5026 btrfs_drop_extent_cache(BTRFS_I(inode),
5027 cur_offset,
5028 cur_offset +
5029 hole_size - 1, 0);
5030 }
5031 free_extent_map(hole_em);
5032 }
5033 next:
5034 free_extent_map(em);
5035 em = NULL;
5036 cur_offset = last_byte;
5037 if (cur_offset >= block_end)
5038 break;
5039 }
5040 free_extent_map(em);
5041 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5042 GFP_NOFS);
5043 return err;
5044 }
5045
5046 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5047 {
5048 struct btrfs_root *root = BTRFS_I(inode)->root;
5049 struct btrfs_trans_handle *trans;
5050 loff_t oldsize = i_size_read(inode);
5051 loff_t newsize = attr->ia_size;
5052 int mask = attr->ia_valid;
5053 int ret;
5054
5055 /*
5056 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5057 * special case where we need to update the times despite not having
5058 * these flags set. For all other operations the VFS set these flags
5059 * explicitly if it wants a timestamp update.
5060 */
5061 if (newsize != oldsize) {
5062 inode_inc_iversion(inode);
5063 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5064 inode->i_ctime = inode->i_mtime =
5065 current_time(inode);
5066 }
5067
5068 if (newsize > oldsize) {
5069 /*
5070 * Don't do an expanding truncate while snapshoting is ongoing.
5071 * This is to ensure the snapshot captures a fully consistent
5072 * state of this file - if the snapshot captures this expanding
5073 * truncation, it must capture all writes that happened before
5074 * this truncation.
5075 */
5076 btrfs_wait_for_snapshot_creation(root);
5077 ret = btrfs_cont_expand(inode, oldsize, newsize);
5078 if (ret) {
5079 btrfs_end_write_no_snapshoting(root);
5080 return ret;
5081 }
5082
5083 trans = btrfs_start_transaction(root, 1);
5084 if (IS_ERR(trans)) {
5085 btrfs_end_write_no_snapshoting(root);
5086 return PTR_ERR(trans);
5087 }
5088
5089 i_size_write(inode, newsize);
5090 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5091 pagecache_isize_extended(inode, oldsize, newsize);
5092 ret = btrfs_update_inode(trans, root, inode);
5093 btrfs_end_write_no_snapshoting(root);
5094 btrfs_end_transaction(trans);
5095 } else {
5096
5097 /*
5098 * We're truncating a file that used to have good data down to
5099 * zero. Make sure it gets into the ordered flush list so that
5100 * any new writes get down to disk quickly.
5101 */
5102 if (newsize == 0)
5103 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5104 &BTRFS_I(inode)->runtime_flags);
5105
5106 /*
5107 * 1 for the orphan item we're going to add
5108 * 1 for the orphan item deletion.
5109 */
5110 trans = btrfs_start_transaction(root, 2);
5111 if (IS_ERR(trans))
5112 return PTR_ERR(trans);
5113
5114 /*
5115 * We need to do this in case we fail at _any_ point during the
5116 * actual truncate. Once we do the truncate_setsize we could
5117 * invalidate pages which forces any outstanding ordered io to
5118 * be instantly completed which will give us extents that need
5119 * to be truncated. If we fail to get an orphan inode down we
5120 * could have left over extents that were never meant to live,
5121 * so we need to guarantee from this point on that everything
5122 * will be consistent.
5123 */
5124 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5125 btrfs_end_transaction(trans);
5126 if (ret)
5127 return ret;
5128
5129 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5130 truncate_setsize(inode, newsize);
5131
5132 /* Disable nonlocked read DIO to avoid the end less truncate */
5133 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5134 inode_dio_wait(inode);
5135 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5136
5137 ret = btrfs_truncate(inode);
5138 if (ret && inode->i_nlink) {
5139 int err;
5140
5141 /* To get a stable disk_i_size */
5142 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5143 if (err) {
5144 btrfs_orphan_del(NULL, BTRFS_I(inode));
5145 return err;
5146 }
5147
5148 /*
5149 * failed to truncate, disk_i_size is only adjusted down
5150 * as we remove extents, so it should represent the true
5151 * size of the inode, so reset the in memory size and
5152 * delete our orphan entry.
5153 */
5154 trans = btrfs_join_transaction(root);
5155 if (IS_ERR(trans)) {
5156 btrfs_orphan_del(NULL, BTRFS_I(inode));
5157 return ret;
5158 }
5159 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5160 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5161 if (err)
5162 btrfs_abort_transaction(trans, err);
5163 btrfs_end_transaction(trans);
5164 }
5165 }
5166
5167 return ret;
5168 }
5169
5170 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5171 {
5172 struct inode *inode = d_inode(dentry);
5173 struct btrfs_root *root = BTRFS_I(inode)->root;
5174 int err;
5175
5176 if (btrfs_root_readonly(root))
5177 return -EROFS;
5178
5179 err = setattr_prepare(dentry, attr);
5180 if (err)
5181 return err;
5182
5183 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5184 err = btrfs_setsize(inode, attr);
5185 if (err)
5186 return err;
5187 }
5188
5189 if (attr->ia_valid) {
5190 setattr_copy(inode, attr);
5191 inode_inc_iversion(inode);
5192 err = btrfs_dirty_inode(inode);
5193
5194 if (!err && attr->ia_valid & ATTR_MODE)
5195 err = posix_acl_chmod(inode, inode->i_mode);
5196 }
5197
5198 return err;
5199 }
5200
5201 /*
5202 * While truncating the inode pages during eviction, we get the VFS calling
5203 * btrfs_invalidatepage() against each page of the inode. This is slow because
5204 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5205 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5206 * extent_state structures over and over, wasting lots of time.
5207 *
5208 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5209 * those expensive operations on a per page basis and do only the ordered io
5210 * finishing, while we release here the extent_map and extent_state structures,
5211 * without the excessive merging and splitting.
5212 */
5213 static void evict_inode_truncate_pages(struct inode *inode)
5214 {
5215 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5216 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5217 struct rb_node *node;
5218
5219 ASSERT(inode->i_state & I_FREEING);
5220 truncate_inode_pages_final(&inode->i_data);
5221
5222 write_lock(&map_tree->lock);
5223 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5224 struct extent_map *em;
5225
5226 node = rb_first(&map_tree->map);
5227 em = rb_entry(node, struct extent_map, rb_node);
5228 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5229 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5230 remove_extent_mapping(map_tree, em);
5231 free_extent_map(em);
5232 if (need_resched()) {
5233 write_unlock(&map_tree->lock);
5234 cond_resched();
5235 write_lock(&map_tree->lock);
5236 }
5237 }
5238 write_unlock(&map_tree->lock);
5239
5240 /*
5241 * Keep looping until we have no more ranges in the io tree.
5242 * We can have ongoing bios started by readpages (called from readahead)
5243 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5244 * still in progress (unlocked the pages in the bio but did not yet
5245 * unlocked the ranges in the io tree). Therefore this means some
5246 * ranges can still be locked and eviction started because before
5247 * submitting those bios, which are executed by a separate task (work
5248 * queue kthread), inode references (inode->i_count) were not taken
5249 * (which would be dropped in the end io callback of each bio).
5250 * Therefore here we effectively end up waiting for those bios and
5251 * anyone else holding locked ranges without having bumped the inode's
5252 * reference count - if we don't do it, when they access the inode's
5253 * io_tree to unlock a range it may be too late, leading to an
5254 * use-after-free issue.
5255 */
5256 spin_lock(&io_tree->lock);
5257 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5258 struct extent_state *state;
5259 struct extent_state *cached_state = NULL;
5260 u64 start;
5261 u64 end;
5262
5263 node = rb_first(&io_tree->state);
5264 state = rb_entry(node, struct extent_state, rb_node);
5265 start = state->start;
5266 end = state->end;
5267 spin_unlock(&io_tree->lock);
5268
5269 lock_extent_bits(io_tree, start, end, &cached_state);
5270
5271 /*
5272 * If still has DELALLOC flag, the extent didn't reach disk,
5273 * and its reserved space won't be freed by delayed_ref.
5274 * So we need to free its reserved space here.
5275 * (Refer to comment in btrfs_invalidatepage, case 2)
5276 *
5277 * Note, end is the bytenr of last byte, so we need + 1 here.
5278 */
5279 if (state->state & EXTENT_DELALLOC)
5280 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5281
5282 clear_extent_bit(io_tree, start, end,
5283 EXTENT_LOCKED | EXTENT_DIRTY |
5284 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5285 EXTENT_DEFRAG, 1, 1,
5286 &cached_state, GFP_NOFS);
5287
5288 cond_resched();
5289 spin_lock(&io_tree->lock);
5290 }
5291 spin_unlock(&io_tree->lock);
5292 }
5293
5294 void btrfs_evict_inode(struct inode *inode)
5295 {
5296 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5297 struct btrfs_trans_handle *trans;
5298 struct btrfs_root *root = BTRFS_I(inode)->root;
5299 struct btrfs_block_rsv *rsv, *global_rsv;
5300 int steal_from_global = 0;
5301 u64 min_size;
5302 int ret;
5303
5304 trace_btrfs_inode_evict(inode);
5305
5306 if (!root) {
5307 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5308 return;
5309 }
5310
5311 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5312
5313 evict_inode_truncate_pages(inode);
5314
5315 if (inode->i_nlink &&
5316 ((btrfs_root_refs(&root->root_item) != 0 &&
5317 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5318 btrfs_is_free_space_inode(BTRFS_I(inode))))
5319 goto no_delete;
5320
5321 if (is_bad_inode(inode)) {
5322 btrfs_orphan_del(NULL, BTRFS_I(inode));
5323 goto no_delete;
5324 }
5325 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5326 if (!special_file(inode->i_mode))
5327 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5328
5329 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5330
5331 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5332 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5333 &BTRFS_I(inode)->runtime_flags));
5334 goto no_delete;
5335 }
5336
5337 if (inode->i_nlink > 0) {
5338 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5339 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5340 goto no_delete;
5341 }
5342
5343 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5344 if (ret) {
5345 btrfs_orphan_del(NULL, BTRFS_I(inode));
5346 goto no_delete;
5347 }
5348
5349 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5350 if (!rsv) {
5351 btrfs_orphan_del(NULL, BTRFS_I(inode));
5352 goto no_delete;
5353 }
5354 rsv->size = min_size;
5355 rsv->failfast = 1;
5356 global_rsv = &fs_info->global_block_rsv;
5357
5358 btrfs_i_size_write(BTRFS_I(inode), 0);
5359
5360 /*
5361 * This is a bit simpler than btrfs_truncate since we've already
5362 * reserved our space for our orphan item in the unlink, so we just
5363 * need to reserve some slack space in case we add bytes and update
5364 * inode item when doing the truncate.
5365 */
5366 while (1) {
5367 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5368 BTRFS_RESERVE_FLUSH_LIMIT);
5369
5370 /*
5371 * Try and steal from the global reserve since we will
5372 * likely not use this space anyway, we want to try as
5373 * hard as possible to get this to work.
5374 */
5375 if (ret)
5376 steal_from_global++;
5377 else
5378 steal_from_global = 0;
5379 ret = 0;
5380
5381 /*
5382 * steal_from_global == 0: we reserved stuff, hooray!
5383 * steal_from_global == 1: we didn't reserve stuff, boo!
5384 * steal_from_global == 2: we've committed, still not a lot of
5385 * room but maybe we'll have room in the global reserve this
5386 * time.
5387 * steal_from_global == 3: abandon all hope!
5388 */
5389 if (steal_from_global > 2) {
5390 btrfs_warn(fs_info,
5391 "Could not get space for a delete, will truncate on mount %d",
5392 ret);
5393 btrfs_orphan_del(NULL, BTRFS_I(inode));
5394 btrfs_free_block_rsv(fs_info, rsv);
5395 goto no_delete;
5396 }
5397
5398 trans = btrfs_join_transaction(root);
5399 if (IS_ERR(trans)) {
5400 btrfs_orphan_del(NULL, BTRFS_I(inode));
5401 btrfs_free_block_rsv(fs_info, rsv);
5402 goto no_delete;
5403 }
5404
5405 /*
5406 * We can't just steal from the global reserve, we need to make
5407 * sure there is room to do it, if not we need to commit and try
5408 * again.
5409 */
5410 if (steal_from_global) {
5411 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5412 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5413 min_size, 0);
5414 else
5415 ret = -ENOSPC;
5416 }
5417
5418 /*
5419 * Couldn't steal from the global reserve, we have too much
5420 * pending stuff built up, commit the transaction and try it
5421 * again.
5422 */
5423 if (ret) {
5424 ret = btrfs_commit_transaction(trans);
5425 if (ret) {
5426 btrfs_orphan_del(NULL, BTRFS_I(inode));
5427 btrfs_free_block_rsv(fs_info, rsv);
5428 goto no_delete;
5429 }
5430 continue;
5431 } else {
5432 steal_from_global = 0;
5433 }
5434
5435 trans->block_rsv = rsv;
5436
5437 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5438 if (ret != -ENOSPC && ret != -EAGAIN)
5439 break;
5440
5441 trans->block_rsv = &fs_info->trans_block_rsv;
5442 btrfs_end_transaction(trans);
5443 trans = NULL;
5444 btrfs_btree_balance_dirty(fs_info);
5445 }
5446
5447 btrfs_free_block_rsv(fs_info, rsv);
5448
5449 /*
5450 * Errors here aren't a big deal, it just means we leave orphan items
5451 * in the tree. They will be cleaned up on the next mount.
5452 */
5453 if (ret == 0) {
5454 trans->block_rsv = root->orphan_block_rsv;
5455 btrfs_orphan_del(trans, BTRFS_I(inode));
5456 } else {
5457 btrfs_orphan_del(NULL, BTRFS_I(inode));
5458 }
5459
5460 trans->block_rsv = &fs_info->trans_block_rsv;
5461 if (!(root == fs_info->tree_root ||
5462 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5463 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5464
5465 btrfs_end_transaction(trans);
5466 btrfs_btree_balance_dirty(fs_info);
5467 no_delete:
5468 btrfs_remove_delayed_node(BTRFS_I(inode));
5469 clear_inode(inode);
5470 }
5471
5472 /*
5473 * this returns the key found in the dir entry in the location pointer.
5474 * If no dir entries were found, location->objectid is 0.
5475 */
5476 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5477 struct btrfs_key *location)
5478 {
5479 const char *name = dentry->d_name.name;
5480 int namelen = dentry->d_name.len;
5481 struct btrfs_dir_item *di;
5482 struct btrfs_path *path;
5483 struct btrfs_root *root = BTRFS_I(dir)->root;
5484 int ret = 0;
5485
5486 path = btrfs_alloc_path();
5487 if (!path)
5488 return -ENOMEM;
5489
5490 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5491 name, namelen, 0);
5492 if (IS_ERR(di))
5493 ret = PTR_ERR(di);
5494
5495 if (IS_ERR_OR_NULL(di))
5496 goto out_err;
5497
5498 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5499 out:
5500 btrfs_free_path(path);
5501 return ret;
5502 out_err:
5503 location->objectid = 0;
5504 goto out;
5505 }
5506
5507 /*
5508 * when we hit a tree root in a directory, the btrfs part of the inode
5509 * needs to be changed to reflect the root directory of the tree root. This
5510 * is kind of like crossing a mount point.
5511 */
5512 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5513 struct inode *dir,
5514 struct dentry *dentry,
5515 struct btrfs_key *location,
5516 struct btrfs_root **sub_root)
5517 {
5518 struct btrfs_path *path;
5519 struct btrfs_root *new_root;
5520 struct btrfs_root_ref *ref;
5521 struct extent_buffer *leaf;
5522 struct btrfs_key key;
5523 int ret;
5524 int err = 0;
5525
5526 path = btrfs_alloc_path();
5527 if (!path) {
5528 err = -ENOMEM;
5529 goto out;
5530 }
5531
5532 err = -ENOENT;
5533 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5534 key.type = BTRFS_ROOT_REF_KEY;
5535 key.offset = location->objectid;
5536
5537 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5538 if (ret) {
5539 if (ret < 0)
5540 err = ret;
5541 goto out;
5542 }
5543
5544 leaf = path->nodes[0];
5545 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5546 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5547 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5548 goto out;
5549
5550 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5551 (unsigned long)(ref + 1),
5552 dentry->d_name.len);
5553 if (ret)
5554 goto out;
5555
5556 btrfs_release_path(path);
5557
5558 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5559 if (IS_ERR(new_root)) {
5560 err = PTR_ERR(new_root);
5561 goto out;
5562 }
5563
5564 *sub_root = new_root;
5565 location->objectid = btrfs_root_dirid(&new_root->root_item);
5566 location->type = BTRFS_INODE_ITEM_KEY;
5567 location->offset = 0;
5568 err = 0;
5569 out:
5570 btrfs_free_path(path);
5571 return err;
5572 }
5573
5574 static void inode_tree_add(struct inode *inode)
5575 {
5576 struct btrfs_root *root = BTRFS_I(inode)->root;
5577 struct btrfs_inode *entry;
5578 struct rb_node **p;
5579 struct rb_node *parent;
5580 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5581 u64 ino = btrfs_ino(BTRFS_I(inode));
5582
5583 if (inode_unhashed(inode))
5584 return;
5585 parent = NULL;
5586 spin_lock(&root->inode_lock);
5587 p = &root->inode_tree.rb_node;
5588 while (*p) {
5589 parent = *p;
5590 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5591
5592 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5593 p = &parent->rb_left;
5594 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5595 p = &parent->rb_right;
5596 else {
5597 WARN_ON(!(entry->vfs_inode.i_state &
5598 (I_WILL_FREE | I_FREEING)));
5599 rb_replace_node(parent, new, &root->inode_tree);
5600 RB_CLEAR_NODE(parent);
5601 spin_unlock(&root->inode_lock);
5602 return;
5603 }
5604 }
5605 rb_link_node(new, parent, p);
5606 rb_insert_color(new, &root->inode_tree);
5607 spin_unlock(&root->inode_lock);
5608 }
5609
5610 static void inode_tree_del(struct inode *inode)
5611 {
5612 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5613 struct btrfs_root *root = BTRFS_I(inode)->root;
5614 int empty = 0;
5615
5616 spin_lock(&root->inode_lock);
5617 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5618 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5619 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5620 empty = RB_EMPTY_ROOT(&root->inode_tree);
5621 }
5622 spin_unlock(&root->inode_lock);
5623
5624 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5625 synchronize_srcu(&fs_info->subvol_srcu);
5626 spin_lock(&root->inode_lock);
5627 empty = RB_EMPTY_ROOT(&root->inode_tree);
5628 spin_unlock(&root->inode_lock);
5629 if (empty)
5630 btrfs_add_dead_root(root);
5631 }
5632 }
5633
5634 void btrfs_invalidate_inodes(struct btrfs_root *root)
5635 {
5636 struct btrfs_fs_info *fs_info = root->fs_info;
5637 struct rb_node *node;
5638 struct rb_node *prev;
5639 struct btrfs_inode *entry;
5640 struct inode *inode;
5641 u64 objectid = 0;
5642
5643 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5644 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5645
5646 spin_lock(&root->inode_lock);
5647 again:
5648 node = root->inode_tree.rb_node;
5649 prev = NULL;
5650 while (node) {
5651 prev = node;
5652 entry = rb_entry(node, struct btrfs_inode, rb_node);
5653
5654 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5655 node = node->rb_left;
5656 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5657 node = node->rb_right;
5658 else
5659 break;
5660 }
5661 if (!node) {
5662 while (prev) {
5663 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5664 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5665 node = prev;
5666 break;
5667 }
5668 prev = rb_next(prev);
5669 }
5670 }
5671 while (node) {
5672 entry = rb_entry(node, struct btrfs_inode, rb_node);
5673 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5674 inode = igrab(&entry->vfs_inode);
5675 if (inode) {
5676 spin_unlock(&root->inode_lock);
5677 if (atomic_read(&inode->i_count) > 1)
5678 d_prune_aliases(inode);
5679 /*
5680 * btrfs_drop_inode will have it removed from
5681 * the inode cache when its usage count
5682 * hits zero.
5683 */
5684 iput(inode);
5685 cond_resched();
5686 spin_lock(&root->inode_lock);
5687 goto again;
5688 }
5689
5690 if (cond_resched_lock(&root->inode_lock))
5691 goto again;
5692
5693 node = rb_next(node);
5694 }
5695 spin_unlock(&root->inode_lock);
5696 }
5697
5698 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5699 {
5700 struct btrfs_iget_args *args = p;
5701 inode->i_ino = args->location->objectid;
5702 memcpy(&BTRFS_I(inode)->location, args->location,
5703 sizeof(*args->location));
5704 BTRFS_I(inode)->root = args->root;
5705 return 0;
5706 }
5707
5708 static int btrfs_find_actor(struct inode *inode, void *opaque)
5709 {
5710 struct btrfs_iget_args *args = opaque;
5711 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5712 args->root == BTRFS_I(inode)->root;
5713 }
5714
5715 static struct inode *btrfs_iget_locked(struct super_block *s,
5716 struct btrfs_key *location,
5717 struct btrfs_root *root)
5718 {
5719 struct inode *inode;
5720 struct btrfs_iget_args args;
5721 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5722
5723 args.location = location;
5724 args.root = root;
5725
5726 inode = iget5_locked(s, hashval, btrfs_find_actor,
5727 btrfs_init_locked_inode,
5728 (void *)&args);
5729 return inode;
5730 }
5731
5732 /* Get an inode object given its location and corresponding root.
5733 * Returns in *is_new if the inode was read from disk
5734 */
5735 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5736 struct btrfs_root *root, int *new)
5737 {
5738 struct inode *inode;
5739
5740 inode = btrfs_iget_locked(s, location, root);
5741 if (!inode)
5742 return ERR_PTR(-ENOMEM);
5743
5744 if (inode->i_state & I_NEW) {
5745 int ret;
5746
5747 ret = btrfs_read_locked_inode(inode);
5748 if (!is_bad_inode(inode)) {
5749 inode_tree_add(inode);
5750 unlock_new_inode(inode);
5751 if (new)
5752 *new = 1;
5753 } else {
5754 unlock_new_inode(inode);
5755 iput(inode);
5756 ASSERT(ret < 0);
5757 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5758 }
5759 }
5760
5761 return inode;
5762 }
5763
5764 static struct inode *new_simple_dir(struct super_block *s,
5765 struct btrfs_key *key,
5766 struct btrfs_root *root)
5767 {
5768 struct inode *inode = new_inode(s);
5769
5770 if (!inode)
5771 return ERR_PTR(-ENOMEM);
5772
5773 BTRFS_I(inode)->root = root;
5774 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5775 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5776
5777 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5778 inode->i_op = &btrfs_dir_ro_inode_operations;
5779 inode->i_opflags &= ~IOP_XATTR;
5780 inode->i_fop = &simple_dir_operations;
5781 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5782 inode->i_mtime = current_time(inode);
5783 inode->i_atime = inode->i_mtime;
5784 inode->i_ctime = inode->i_mtime;
5785 BTRFS_I(inode)->i_otime = inode->i_mtime;
5786
5787 return inode;
5788 }
5789
5790 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5791 {
5792 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5793 struct inode *inode;
5794 struct btrfs_root *root = BTRFS_I(dir)->root;
5795 struct btrfs_root *sub_root = root;
5796 struct btrfs_key location;
5797 int index;
5798 int ret = 0;
5799
5800 if (dentry->d_name.len > BTRFS_NAME_LEN)
5801 return ERR_PTR(-ENAMETOOLONG);
5802
5803 ret = btrfs_inode_by_name(dir, dentry, &location);
5804 if (ret < 0)
5805 return ERR_PTR(ret);
5806
5807 if (location.objectid == 0)
5808 return ERR_PTR(-ENOENT);
5809
5810 if (location.type == BTRFS_INODE_ITEM_KEY) {
5811 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5812 return inode;
5813 }
5814
5815 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5816
5817 index = srcu_read_lock(&fs_info->subvol_srcu);
5818 ret = fixup_tree_root_location(fs_info, dir, dentry,
5819 &location, &sub_root);
5820 if (ret < 0) {
5821 if (ret != -ENOENT)
5822 inode = ERR_PTR(ret);
5823 else
5824 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5825 } else {
5826 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5827 }
5828 srcu_read_unlock(&fs_info->subvol_srcu, index);
5829
5830 if (!IS_ERR(inode) && root != sub_root) {
5831 down_read(&fs_info->cleanup_work_sem);
5832 if (!(inode->i_sb->s_flags & MS_RDONLY))
5833 ret = btrfs_orphan_cleanup(sub_root);
5834 up_read(&fs_info->cleanup_work_sem);
5835 if (ret) {
5836 iput(inode);
5837 inode = ERR_PTR(ret);
5838 }
5839 }
5840
5841 return inode;
5842 }
5843
5844 static int btrfs_dentry_delete(const struct dentry *dentry)
5845 {
5846 struct btrfs_root *root;
5847 struct inode *inode = d_inode(dentry);
5848
5849 if (!inode && !IS_ROOT(dentry))
5850 inode = d_inode(dentry->d_parent);
5851
5852 if (inode) {
5853 root = BTRFS_I(inode)->root;
5854 if (btrfs_root_refs(&root->root_item) == 0)
5855 return 1;
5856
5857 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5858 return 1;
5859 }
5860 return 0;
5861 }
5862
5863 static void btrfs_dentry_release(struct dentry *dentry)
5864 {
5865 kfree(dentry->d_fsdata);
5866 }
5867
5868 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5869 unsigned int flags)
5870 {
5871 struct inode *inode;
5872
5873 inode = btrfs_lookup_dentry(dir, dentry);
5874 if (IS_ERR(inode)) {
5875 if (PTR_ERR(inode) == -ENOENT)
5876 inode = NULL;
5877 else
5878 return ERR_CAST(inode);
5879 }
5880
5881 return d_splice_alias(inode, dentry);
5882 }
5883
5884 unsigned char btrfs_filetype_table[] = {
5885 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5886 };
5887
5888 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5889 {
5890 struct inode *inode = file_inode(file);
5891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5892 struct btrfs_root *root = BTRFS_I(inode)->root;
5893 struct btrfs_dir_item *di;
5894 struct btrfs_key key;
5895 struct btrfs_key found_key;
5896 struct btrfs_path *path;
5897 struct list_head ins_list;
5898 struct list_head del_list;
5899 int ret;
5900 struct extent_buffer *leaf;
5901 int slot;
5902 unsigned char d_type;
5903 int over = 0;
5904 char tmp_name[32];
5905 char *name_ptr;
5906 int name_len;
5907 bool put = false;
5908 struct btrfs_key location;
5909
5910 if (!dir_emit_dots(file, ctx))
5911 return 0;
5912
5913 path = btrfs_alloc_path();
5914 if (!path)
5915 return -ENOMEM;
5916
5917 path->reada = READA_FORWARD;
5918
5919 INIT_LIST_HEAD(&ins_list);
5920 INIT_LIST_HEAD(&del_list);
5921 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5922
5923 key.type = BTRFS_DIR_INDEX_KEY;
5924 key.offset = ctx->pos;
5925 key.objectid = btrfs_ino(BTRFS_I(inode));
5926
5927 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5928 if (ret < 0)
5929 goto err;
5930
5931 while (1) {
5932 leaf = path->nodes[0];
5933 slot = path->slots[0];
5934 if (slot >= btrfs_header_nritems(leaf)) {
5935 ret = btrfs_next_leaf(root, path);
5936 if (ret < 0)
5937 goto err;
5938 else if (ret > 0)
5939 break;
5940 continue;
5941 }
5942
5943 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5944
5945 if (found_key.objectid != key.objectid)
5946 break;
5947 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5948 break;
5949 if (found_key.offset < ctx->pos)
5950 goto next;
5951 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5952 goto next;
5953
5954 ctx->pos = found_key.offset;
5955
5956 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5957 if (verify_dir_item(fs_info, leaf, slot, di))
5958 goto next;
5959
5960 name_len = btrfs_dir_name_len(leaf, di);
5961 if (name_len <= sizeof(tmp_name)) {
5962 name_ptr = tmp_name;
5963 } else {
5964 name_ptr = kmalloc(name_len, GFP_KERNEL);
5965 if (!name_ptr) {
5966 ret = -ENOMEM;
5967 goto err;
5968 }
5969 }
5970 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5971 name_len);
5972
5973 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5974 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5975
5976 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5977 d_type);
5978
5979 if (name_ptr != tmp_name)
5980 kfree(name_ptr);
5981
5982 if (over)
5983 goto nopos;
5984 ctx->pos++;
5985 next:
5986 path->slots[0]++;
5987 }
5988
5989 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5990 if (ret)
5991 goto nopos;
5992
5993 /*
5994 * Stop new entries from being returned after we return the last
5995 * entry.
5996 *
5997 * New directory entries are assigned a strictly increasing
5998 * offset. This means that new entries created during readdir
5999 * are *guaranteed* to be seen in the future by that readdir.
6000 * This has broken buggy programs which operate on names as
6001 * they're returned by readdir. Until we re-use freed offsets
6002 * we have this hack to stop new entries from being returned
6003 * under the assumption that they'll never reach this huge
6004 * offset.
6005 *
6006 * This is being careful not to overflow 32bit loff_t unless the
6007 * last entry requires it because doing so has broken 32bit apps
6008 * in the past.
6009 */
6010 if (ctx->pos >= INT_MAX)
6011 ctx->pos = LLONG_MAX;
6012 else
6013 ctx->pos = INT_MAX;
6014 nopos:
6015 ret = 0;
6016 err:
6017 if (put)
6018 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6019 btrfs_free_path(path);
6020 return ret;
6021 }
6022
6023 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6024 {
6025 struct btrfs_root *root = BTRFS_I(inode)->root;
6026 struct btrfs_trans_handle *trans;
6027 int ret = 0;
6028 bool nolock = false;
6029
6030 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6031 return 0;
6032
6033 if (btrfs_fs_closing(root->fs_info) &&
6034 btrfs_is_free_space_inode(BTRFS_I(inode)))
6035 nolock = true;
6036
6037 if (wbc->sync_mode == WB_SYNC_ALL) {
6038 if (nolock)
6039 trans = btrfs_join_transaction_nolock(root);
6040 else
6041 trans = btrfs_join_transaction(root);
6042 if (IS_ERR(trans))
6043 return PTR_ERR(trans);
6044 ret = btrfs_commit_transaction(trans);
6045 }
6046 return ret;
6047 }
6048
6049 /*
6050 * This is somewhat expensive, updating the tree every time the
6051 * inode changes. But, it is most likely to find the inode in cache.
6052 * FIXME, needs more benchmarking...there are no reasons other than performance
6053 * to keep or drop this code.
6054 */
6055 static int btrfs_dirty_inode(struct inode *inode)
6056 {
6057 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6058 struct btrfs_root *root = BTRFS_I(inode)->root;
6059 struct btrfs_trans_handle *trans;
6060 int ret;
6061
6062 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6063 return 0;
6064
6065 trans = btrfs_join_transaction(root);
6066 if (IS_ERR(trans))
6067 return PTR_ERR(trans);
6068
6069 ret = btrfs_update_inode(trans, root, inode);
6070 if (ret && ret == -ENOSPC) {
6071 /* whoops, lets try again with the full transaction */
6072 btrfs_end_transaction(trans);
6073 trans = btrfs_start_transaction(root, 1);
6074 if (IS_ERR(trans))
6075 return PTR_ERR(trans);
6076
6077 ret = btrfs_update_inode(trans, root, inode);
6078 }
6079 btrfs_end_transaction(trans);
6080 if (BTRFS_I(inode)->delayed_node)
6081 btrfs_balance_delayed_items(fs_info);
6082
6083 return ret;
6084 }
6085
6086 /*
6087 * This is a copy of file_update_time. We need this so we can return error on
6088 * ENOSPC for updating the inode in the case of file write and mmap writes.
6089 */
6090 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6091 int flags)
6092 {
6093 struct btrfs_root *root = BTRFS_I(inode)->root;
6094
6095 if (btrfs_root_readonly(root))
6096 return -EROFS;
6097
6098 if (flags & S_VERSION)
6099 inode_inc_iversion(inode);
6100 if (flags & S_CTIME)
6101 inode->i_ctime = *now;
6102 if (flags & S_MTIME)
6103 inode->i_mtime = *now;
6104 if (flags & S_ATIME)
6105 inode->i_atime = *now;
6106 return btrfs_dirty_inode(inode);
6107 }
6108
6109 /*
6110 * find the highest existing sequence number in a directory
6111 * and then set the in-memory index_cnt variable to reflect
6112 * free sequence numbers
6113 */
6114 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6115 {
6116 struct btrfs_root *root = inode->root;
6117 struct btrfs_key key, found_key;
6118 struct btrfs_path *path;
6119 struct extent_buffer *leaf;
6120 int ret;
6121
6122 key.objectid = btrfs_ino(inode);
6123 key.type = BTRFS_DIR_INDEX_KEY;
6124 key.offset = (u64)-1;
6125
6126 path = btrfs_alloc_path();
6127 if (!path)
6128 return -ENOMEM;
6129
6130 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6131 if (ret < 0)
6132 goto out;
6133 /* FIXME: we should be able to handle this */
6134 if (ret == 0)
6135 goto out;
6136 ret = 0;
6137
6138 /*
6139 * MAGIC NUMBER EXPLANATION:
6140 * since we search a directory based on f_pos we have to start at 2
6141 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6142 * else has to start at 2
6143 */
6144 if (path->slots[0] == 0) {
6145 inode->index_cnt = 2;
6146 goto out;
6147 }
6148
6149 path->slots[0]--;
6150
6151 leaf = path->nodes[0];
6152 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6153
6154 if (found_key.objectid != btrfs_ino(inode) ||
6155 found_key.type != BTRFS_DIR_INDEX_KEY) {
6156 inode->index_cnt = 2;
6157 goto out;
6158 }
6159
6160 inode->index_cnt = found_key.offset + 1;
6161 out:
6162 btrfs_free_path(path);
6163 return ret;
6164 }
6165
6166 /*
6167 * helper to find a free sequence number in a given directory. This current
6168 * code is very simple, later versions will do smarter things in the btree
6169 */
6170 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6171 {
6172 int ret = 0;
6173
6174 if (dir->index_cnt == (u64)-1) {
6175 ret = btrfs_inode_delayed_dir_index_count(dir);
6176 if (ret) {
6177 ret = btrfs_set_inode_index_count(dir);
6178 if (ret)
6179 return ret;
6180 }
6181 }
6182
6183 *index = dir->index_cnt;
6184 dir->index_cnt++;
6185
6186 return ret;
6187 }
6188
6189 static int btrfs_insert_inode_locked(struct inode *inode)
6190 {
6191 struct btrfs_iget_args args;
6192 args.location = &BTRFS_I(inode)->location;
6193 args.root = BTRFS_I(inode)->root;
6194
6195 return insert_inode_locked4(inode,
6196 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6197 btrfs_find_actor, &args);
6198 }
6199
6200 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6201 struct btrfs_root *root,
6202 struct inode *dir,
6203 const char *name, int name_len,
6204 u64 ref_objectid, u64 objectid,
6205 umode_t mode, u64 *index)
6206 {
6207 struct btrfs_fs_info *fs_info = root->fs_info;
6208 struct inode *inode;
6209 struct btrfs_inode_item *inode_item;
6210 struct btrfs_key *location;
6211 struct btrfs_path *path;
6212 struct btrfs_inode_ref *ref;
6213 struct btrfs_key key[2];
6214 u32 sizes[2];
6215 int nitems = name ? 2 : 1;
6216 unsigned long ptr;
6217 int ret;
6218
6219 path = btrfs_alloc_path();
6220 if (!path)
6221 return ERR_PTR(-ENOMEM);
6222
6223 inode = new_inode(fs_info->sb);
6224 if (!inode) {
6225 btrfs_free_path(path);
6226 return ERR_PTR(-ENOMEM);
6227 }
6228
6229 /*
6230 * O_TMPFILE, set link count to 0, so that after this point,
6231 * we fill in an inode item with the correct link count.
6232 */
6233 if (!name)
6234 set_nlink(inode, 0);
6235
6236 /*
6237 * we have to initialize this early, so we can reclaim the inode
6238 * number if we fail afterwards in this function.
6239 */
6240 inode->i_ino = objectid;
6241
6242 if (dir && name) {
6243 trace_btrfs_inode_request(dir);
6244
6245 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6246 if (ret) {
6247 btrfs_free_path(path);
6248 iput(inode);
6249 return ERR_PTR(ret);
6250 }
6251 } else if (dir) {
6252 *index = 0;
6253 }
6254 /*
6255 * index_cnt is ignored for everything but a dir,
6256 * btrfs_get_inode_index_count has an explanation for the magic
6257 * number
6258 */
6259 BTRFS_I(inode)->index_cnt = 2;
6260 BTRFS_I(inode)->dir_index = *index;
6261 BTRFS_I(inode)->root = root;
6262 BTRFS_I(inode)->generation = trans->transid;
6263 inode->i_generation = BTRFS_I(inode)->generation;
6264
6265 /*
6266 * We could have gotten an inode number from somebody who was fsynced
6267 * and then removed in this same transaction, so let's just set full
6268 * sync since it will be a full sync anyway and this will blow away the
6269 * old info in the log.
6270 */
6271 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6272
6273 key[0].objectid = objectid;
6274 key[0].type = BTRFS_INODE_ITEM_KEY;
6275 key[0].offset = 0;
6276
6277 sizes[0] = sizeof(struct btrfs_inode_item);
6278
6279 if (name) {
6280 /*
6281 * Start new inodes with an inode_ref. This is slightly more
6282 * efficient for small numbers of hard links since they will
6283 * be packed into one item. Extended refs will kick in if we
6284 * add more hard links than can fit in the ref item.
6285 */
6286 key[1].objectid = objectid;
6287 key[1].type = BTRFS_INODE_REF_KEY;
6288 key[1].offset = ref_objectid;
6289
6290 sizes[1] = name_len + sizeof(*ref);
6291 }
6292
6293 location = &BTRFS_I(inode)->location;
6294 location->objectid = objectid;
6295 location->offset = 0;
6296 location->type = BTRFS_INODE_ITEM_KEY;
6297
6298 ret = btrfs_insert_inode_locked(inode);
6299 if (ret < 0)
6300 goto fail;
6301
6302 path->leave_spinning = 1;
6303 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6304 if (ret != 0)
6305 goto fail_unlock;
6306
6307 inode_init_owner(inode, dir, mode);
6308 inode_set_bytes(inode, 0);
6309
6310 inode->i_mtime = current_time(inode);
6311 inode->i_atime = inode->i_mtime;
6312 inode->i_ctime = inode->i_mtime;
6313 BTRFS_I(inode)->i_otime = inode->i_mtime;
6314
6315 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6316 struct btrfs_inode_item);
6317 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6318 sizeof(*inode_item));
6319 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6320
6321 if (name) {
6322 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6323 struct btrfs_inode_ref);
6324 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6325 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6326 ptr = (unsigned long)(ref + 1);
6327 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6328 }
6329
6330 btrfs_mark_buffer_dirty(path->nodes[0]);
6331 btrfs_free_path(path);
6332
6333 btrfs_inherit_iflags(inode, dir);
6334
6335 if (S_ISREG(mode)) {
6336 if (btrfs_test_opt(fs_info, NODATASUM))
6337 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6338 if (btrfs_test_opt(fs_info, NODATACOW))
6339 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6340 BTRFS_INODE_NODATASUM;
6341 }
6342
6343 inode_tree_add(inode);
6344
6345 trace_btrfs_inode_new(inode);
6346 btrfs_set_inode_last_trans(trans, inode);
6347
6348 btrfs_update_root_times(trans, root);
6349
6350 ret = btrfs_inode_inherit_props(trans, inode, dir);
6351 if (ret)
6352 btrfs_err(fs_info,
6353 "error inheriting props for ino %llu (root %llu): %d",
6354 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6355
6356 return inode;
6357
6358 fail_unlock:
6359 unlock_new_inode(inode);
6360 fail:
6361 if (dir && name)
6362 BTRFS_I(dir)->index_cnt--;
6363 btrfs_free_path(path);
6364 iput(inode);
6365 return ERR_PTR(ret);
6366 }
6367
6368 static inline u8 btrfs_inode_type(struct inode *inode)
6369 {
6370 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6371 }
6372
6373 /*
6374 * utility function to add 'inode' into 'parent_inode' with
6375 * a give name and a given sequence number.
6376 * if 'add_backref' is true, also insert a backref from the
6377 * inode to the parent directory.
6378 */
6379 int btrfs_add_link(struct btrfs_trans_handle *trans,
6380 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6381 const char *name, int name_len, int add_backref, u64 index)
6382 {
6383 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6384 int ret = 0;
6385 struct btrfs_key key;
6386 struct btrfs_root *root = parent_inode->root;
6387 u64 ino = btrfs_ino(inode);
6388 u64 parent_ino = btrfs_ino(parent_inode);
6389
6390 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6391 memcpy(&key, &inode->root->root_key, sizeof(key));
6392 } else {
6393 key.objectid = ino;
6394 key.type = BTRFS_INODE_ITEM_KEY;
6395 key.offset = 0;
6396 }
6397
6398 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6399 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6400 root->root_key.objectid, parent_ino,
6401 index, name, name_len);
6402 } else if (add_backref) {
6403 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6404 parent_ino, index);
6405 }
6406
6407 /* Nothing to clean up yet */
6408 if (ret)
6409 return ret;
6410
6411 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6412 parent_inode, &key,
6413 btrfs_inode_type(&inode->vfs_inode), index);
6414 if (ret == -EEXIST || ret == -EOVERFLOW)
6415 goto fail_dir_item;
6416 else if (ret) {
6417 btrfs_abort_transaction(trans, ret);
6418 return ret;
6419 }
6420
6421 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6422 name_len * 2);
6423 inode_inc_iversion(&parent_inode->vfs_inode);
6424 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6425 current_time(&parent_inode->vfs_inode);
6426 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6427 if (ret)
6428 btrfs_abort_transaction(trans, ret);
6429 return ret;
6430
6431 fail_dir_item:
6432 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6433 u64 local_index;
6434 int err;
6435 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6436 root->root_key.objectid, parent_ino,
6437 &local_index, name, name_len);
6438
6439 } else if (add_backref) {
6440 u64 local_index;
6441 int err;
6442
6443 err = btrfs_del_inode_ref(trans, root, name, name_len,
6444 ino, parent_ino, &local_index);
6445 }
6446 return ret;
6447 }
6448
6449 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6450 struct btrfs_inode *dir, struct dentry *dentry,
6451 struct btrfs_inode *inode, int backref, u64 index)
6452 {
6453 int err = btrfs_add_link(trans, dir, inode,
6454 dentry->d_name.name, dentry->d_name.len,
6455 backref, index);
6456 if (err > 0)
6457 err = -EEXIST;
6458 return err;
6459 }
6460
6461 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6462 umode_t mode, dev_t rdev)
6463 {
6464 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6465 struct btrfs_trans_handle *trans;
6466 struct btrfs_root *root = BTRFS_I(dir)->root;
6467 struct inode *inode = NULL;
6468 int err;
6469 int drop_inode = 0;
6470 u64 objectid;
6471 u64 index = 0;
6472
6473 /*
6474 * 2 for inode item and ref
6475 * 2 for dir items
6476 * 1 for xattr if selinux is on
6477 */
6478 trans = btrfs_start_transaction(root, 5);
6479 if (IS_ERR(trans))
6480 return PTR_ERR(trans);
6481
6482 err = btrfs_find_free_ino(root, &objectid);
6483 if (err)
6484 goto out_unlock;
6485
6486 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6487 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6488 mode, &index);
6489 if (IS_ERR(inode)) {
6490 err = PTR_ERR(inode);
6491 goto out_unlock;
6492 }
6493
6494 /*
6495 * If the active LSM wants to access the inode during
6496 * d_instantiate it needs these. Smack checks to see
6497 * if the filesystem supports xattrs by looking at the
6498 * ops vector.
6499 */
6500 inode->i_op = &btrfs_special_inode_operations;
6501 init_special_inode(inode, inode->i_mode, rdev);
6502
6503 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6504 if (err)
6505 goto out_unlock_inode;
6506
6507 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6508 0, index);
6509 if (err) {
6510 goto out_unlock_inode;
6511 } else {
6512 btrfs_update_inode(trans, root, inode);
6513 unlock_new_inode(inode);
6514 d_instantiate(dentry, inode);
6515 }
6516
6517 out_unlock:
6518 btrfs_end_transaction(trans);
6519 btrfs_balance_delayed_items(fs_info);
6520 btrfs_btree_balance_dirty(fs_info);
6521 if (drop_inode) {
6522 inode_dec_link_count(inode);
6523 iput(inode);
6524 }
6525 return err;
6526
6527 out_unlock_inode:
6528 drop_inode = 1;
6529 unlock_new_inode(inode);
6530 goto out_unlock;
6531
6532 }
6533
6534 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6535 umode_t mode, bool excl)
6536 {
6537 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6538 struct btrfs_trans_handle *trans;
6539 struct btrfs_root *root = BTRFS_I(dir)->root;
6540 struct inode *inode = NULL;
6541 int drop_inode_on_err = 0;
6542 int err;
6543 u64 objectid;
6544 u64 index = 0;
6545
6546 /*
6547 * 2 for inode item and ref
6548 * 2 for dir items
6549 * 1 for xattr if selinux is on
6550 */
6551 trans = btrfs_start_transaction(root, 5);
6552 if (IS_ERR(trans))
6553 return PTR_ERR(trans);
6554
6555 err = btrfs_find_free_ino(root, &objectid);
6556 if (err)
6557 goto out_unlock;
6558
6559 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6560 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6561 mode, &index);
6562 if (IS_ERR(inode)) {
6563 err = PTR_ERR(inode);
6564 goto out_unlock;
6565 }
6566 drop_inode_on_err = 1;
6567 /*
6568 * If the active LSM wants to access the inode during
6569 * d_instantiate it needs these. Smack checks to see
6570 * if the filesystem supports xattrs by looking at the
6571 * ops vector.
6572 */
6573 inode->i_fop = &btrfs_file_operations;
6574 inode->i_op = &btrfs_file_inode_operations;
6575 inode->i_mapping->a_ops = &btrfs_aops;
6576
6577 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6578 if (err)
6579 goto out_unlock_inode;
6580
6581 err = btrfs_update_inode(trans, root, inode);
6582 if (err)
6583 goto out_unlock_inode;
6584
6585 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6586 0, index);
6587 if (err)
6588 goto out_unlock_inode;
6589
6590 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6591 unlock_new_inode(inode);
6592 d_instantiate(dentry, inode);
6593
6594 out_unlock:
6595 btrfs_end_transaction(trans);
6596 if (err && drop_inode_on_err) {
6597 inode_dec_link_count(inode);
6598 iput(inode);
6599 }
6600 btrfs_balance_delayed_items(fs_info);
6601 btrfs_btree_balance_dirty(fs_info);
6602 return err;
6603
6604 out_unlock_inode:
6605 unlock_new_inode(inode);
6606 goto out_unlock;
6607
6608 }
6609
6610 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6611 struct dentry *dentry)
6612 {
6613 struct btrfs_trans_handle *trans = NULL;
6614 struct btrfs_root *root = BTRFS_I(dir)->root;
6615 struct inode *inode = d_inode(old_dentry);
6616 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6617 u64 index;
6618 int err;
6619 int drop_inode = 0;
6620
6621 /* do not allow sys_link's with other subvols of the same device */
6622 if (root->objectid != BTRFS_I(inode)->root->objectid)
6623 return -EXDEV;
6624
6625 if (inode->i_nlink >= BTRFS_LINK_MAX)
6626 return -EMLINK;
6627
6628 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6629 if (err)
6630 goto fail;
6631
6632 /*
6633 * 2 items for inode and inode ref
6634 * 2 items for dir items
6635 * 1 item for parent inode
6636 */
6637 trans = btrfs_start_transaction(root, 5);
6638 if (IS_ERR(trans)) {
6639 err = PTR_ERR(trans);
6640 trans = NULL;
6641 goto fail;
6642 }
6643
6644 /* There are several dir indexes for this inode, clear the cache. */
6645 BTRFS_I(inode)->dir_index = 0ULL;
6646 inc_nlink(inode);
6647 inode_inc_iversion(inode);
6648 inode->i_ctime = current_time(inode);
6649 ihold(inode);
6650 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6651
6652 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6653 1, index);
6654
6655 if (err) {
6656 drop_inode = 1;
6657 } else {
6658 struct dentry *parent = dentry->d_parent;
6659 err = btrfs_update_inode(trans, root, inode);
6660 if (err)
6661 goto fail;
6662 if (inode->i_nlink == 1) {
6663 /*
6664 * If new hard link count is 1, it's a file created
6665 * with open(2) O_TMPFILE flag.
6666 */
6667 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6668 if (err)
6669 goto fail;
6670 }
6671 d_instantiate(dentry, inode);
6672 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6673 }
6674
6675 btrfs_balance_delayed_items(fs_info);
6676 fail:
6677 if (trans)
6678 btrfs_end_transaction(trans);
6679 if (drop_inode) {
6680 inode_dec_link_count(inode);
6681 iput(inode);
6682 }
6683 btrfs_btree_balance_dirty(fs_info);
6684 return err;
6685 }
6686
6687 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6688 {
6689 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6690 struct inode *inode = NULL;
6691 struct btrfs_trans_handle *trans;
6692 struct btrfs_root *root = BTRFS_I(dir)->root;
6693 int err = 0;
6694 int drop_on_err = 0;
6695 u64 objectid = 0;
6696 u64 index = 0;
6697
6698 /*
6699 * 2 items for inode and ref
6700 * 2 items for dir items
6701 * 1 for xattr if selinux is on
6702 */
6703 trans = btrfs_start_transaction(root, 5);
6704 if (IS_ERR(trans))
6705 return PTR_ERR(trans);
6706
6707 err = btrfs_find_free_ino(root, &objectid);
6708 if (err)
6709 goto out_fail;
6710
6711 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6712 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6713 S_IFDIR | mode, &index);
6714 if (IS_ERR(inode)) {
6715 err = PTR_ERR(inode);
6716 goto out_fail;
6717 }
6718
6719 drop_on_err = 1;
6720 /* these must be set before we unlock the inode */
6721 inode->i_op = &btrfs_dir_inode_operations;
6722 inode->i_fop = &btrfs_dir_file_operations;
6723
6724 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6725 if (err)
6726 goto out_fail_inode;
6727
6728 btrfs_i_size_write(BTRFS_I(inode), 0);
6729 err = btrfs_update_inode(trans, root, inode);
6730 if (err)
6731 goto out_fail_inode;
6732
6733 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6734 dentry->d_name.name,
6735 dentry->d_name.len, 0, index);
6736 if (err)
6737 goto out_fail_inode;
6738
6739 d_instantiate(dentry, inode);
6740 /*
6741 * mkdir is special. We're unlocking after we call d_instantiate
6742 * to avoid a race with nfsd calling d_instantiate.
6743 */
6744 unlock_new_inode(inode);
6745 drop_on_err = 0;
6746
6747 out_fail:
6748 btrfs_end_transaction(trans);
6749 if (drop_on_err) {
6750 inode_dec_link_count(inode);
6751 iput(inode);
6752 }
6753 btrfs_balance_delayed_items(fs_info);
6754 btrfs_btree_balance_dirty(fs_info);
6755 return err;
6756
6757 out_fail_inode:
6758 unlock_new_inode(inode);
6759 goto out_fail;
6760 }
6761
6762 /* Find next extent map of a given extent map, caller needs to ensure locks */
6763 static struct extent_map *next_extent_map(struct extent_map *em)
6764 {
6765 struct rb_node *next;
6766
6767 next = rb_next(&em->rb_node);
6768 if (!next)
6769 return NULL;
6770 return container_of(next, struct extent_map, rb_node);
6771 }
6772
6773 static struct extent_map *prev_extent_map(struct extent_map *em)
6774 {
6775 struct rb_node *prev;
6776
6777 prev = rb_prev(&em->rb_node);
6778 if (!prev)
6779 return NULL;
6780 return container_of(prev, struct extent_map, rb_node);
6781 }
6782
6783 /* helper for btfs_get_extent. Given an existing extent in the tree,
6784 * the existing extent is the nearest extent to map_start,
6785 * and an extent that you want to insert, deal with overlap and insert
6786 * the best fitted new extent into the tree.
6787 */
6788 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6789 struct extent_map *existing,
6790 struct extent_map *em,
6791 u64 map_start)
6792 {
6793 struct extent_map *prev;
6794 struct extent_map *next;
6795 u64 start;
6796 u64 end;
6797 u64 start_diff;
6798
6799 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6800
6801 if (existing->start > map_start) {
6802 next = existing;
6803 prev = prev_extent_map(next);
6804 } else {
6805 prev = existing;
6806 next = next_extent_map(prev);
6807 }
6808
6809 start = prev ? extent_map_end(prev) : em->start;
6810 start = max_t(u64, start, em->start);
6811 end = next ? next->start : extent_map_end(em);
6812 end = min_t(u64, end, extent_map_end(em));
6813 start_diff = start - em->start;
6814 em->start = start;
6815 em->len = end - start;
6816 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6817 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6818 em->block_start += start_diff;
6819 em->block_len -= start_diff;
6820 }
6821 return add_extent_mapping(em_tree, em, 0);
6822 }
6823
6824 static noinline int uncompress_inline(struct btrfs_path *path,
6825 struct page *page,
6826 size_t pg_offset, u64 extent_offset,
6827 struct btrfs_file_extent_item *item)
6828 {
6829 int ret;
6830 struct extent_buffer *leaf = path->nodes[0];
6831 char *tmp;
6832 size_t max_size;
6833 unsigned long inline_size;
6834 unsigned long ptr;
6835 int compress_type;
6836
6837 WARN_ON(pg_offset != 0);
6838 compress_type = btrfs_file_extent_compression(leaf, item);
6839 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6840 inline_size = btrfs_file_extent_inline_item_len(leaf,
6841 btrfs_item_nr(path->slots[0]));
6842 tmp = kmalloc(inline_size, GFP_NOFS);
6843 if (!tmp)
6844 return -ENOMEM;
6845 ptr = btrfs_file_extent_inline_start(item);
6846
6847 read_extent_buffer(leaf, tmp, ptr, inline_size);
6848
6849 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6850 ret = btrfs_decompress(compress_type, tmp, page,
6851 extent_offset, inline_size, max_size);
6852
6853 /*
6854 * decompression code contains a memset to fill in any space between the end
6855 * of the uncompressed data and the end of max_size in case the decompressed
6856 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6857 * the end of an inline extent and the beginning of the next block, so we
6858 * cover that region here.
6859 */
6860
6861 if (max_size + pg_offset < PAGE_SIZE) {
6862 char *map = kmap(page);
6863 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6864 kunmap(page);
6865 }
6866 kfree(tmp);
6867 return ret;
6868 }
6869
6870 /*
6871 * a bit scary, this does extent mapping from logical file offset to the disk.
6872 * the ugly parts come from merging extents from the disk with the in-ram
6873 * representation. This gets more complex because of the data=ordered code,
6874 * where the in-ram extents might be locked pending data=ordered completion.
6875 *
6876 * This also copies inline extents directly into the page.
6877 */
6878 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6879 struct page *page,
6880 size_t pg_offset, u64 start, u64 len,
6881 int create)
6882 {
6883 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6884 int ret;
6885 int err = 0;
6886 u64 extent_start = 0;
6887 u64 extent_end = 0;
6888 u64 objectid = btrfs_ino(inode);
6889 u32 found_type;
6890 struct btrfs_path *path = NULL;
6891 struct btrfs_root *root = inode->root;
6892 struct btrfs_file_extent_item *item;
6893 struct extent_buffer *leaf;
6894 struct btrfs_key found_key;
6895 struct extent_map *em = NULL;
6896 struct extent_map_tree *em_tree = &inode->extent_tree;
6897 struct extent_io_tree *io_tree = &inode->io_tree;
6898 struct btrfs_trans_handle *trans = NULL;
6899 const bool new_inline = !page || create;
6900
6901 again:
6902 read_lock(&em_tree->lock);
6903 em = lookup_extent_mapping(em_tree, start, len);
6904 if (em)
6905 em->bdev = fs_info->fs_devices->latest_bdev;
6906 read_unlock(&em_tree->lock);
6907
6908 if (em) {
6909 if (em->start > start || em->start + em->len <= start)
6910 free_extent_map(em);
6911 else if (em->block_start == EXTENT_MAP_INLINE && page)
6912 free_extent_map(em);
6913 else
6914 goto out;
6915 }
6916 em = alloc_extent_map();
6917 if (!em) {
6918 err = -ENOMEM;
6919 goto out;
6920 }
6921 em->bdev = fs_info->fs_devices->latest_bdev;
6922 em->start = EXTENT_MAP_HOLE;
6923 em->orig_start = EXTENT_MAP_HOLE;
6924 em->len = (u64)-1;
6925 em->block_len = (u64)-1;
6926
6927 if (!path) {
6928 path = btrfs_alloc_path();
6929 if (!path) {
6930 err = -ENOMEM;
6931 goto out;
6932 }
6933 /*
6934 * Chances are we'll be called again, so go ahead and do
6935 * readahead
6936 */
6937 path->reada = READA_FORWARD;
6938 }
6939
6940 ret = btrfs_lookup_file_extent(trans, root, path,
6941 objectid, start, trans != NULL);
6942 if (ret < 0) {
6943 err = ret;
6944 goto out;
6945 }
6946
6947 if (ret != 0) {
6948 if (path->slots[0] == 0)
6949 goto not_found;
6950 path->slots[0]--;
6951 }
6952
6953 leaf = path->nodes[0];
6954 item = btrfs_item_ptr(leaf, path->slots[0],
6955 struct btrfs_file_extent_item);
6956 /* are we inside the extent that was found? */
6957 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6958 found_type = found_key.type;
6959 if (found_key.objectid != objectid ||
6960 found_type != BTRFS_EXTENT_DATA_KEY) {
6961 /*
6962 * If we backup past the first extent we want to move forward
6963 * and see if there is an extent in front of us, otherwise we'll
6964 * say there is a hole for our whole search range which can
6965 * cause problems.
6966 */
6967 extent_end = start;
6968 goto next;
6969 }
6970
6971 found_type = btrfs_file_extent_type(leaf, item);
6972 extent_start = found_key.offset;
6973 if (found_type == BTRFS_FILE_EXTENT_REG ||
6974 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6975 extent_end = extent_start +
6976 btrfs_file_extent_num_bytes(leaf, item);
6977
6978 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6979 extent_start);
6980 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6981 size_t size;
6982 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6983 extent_end = ALIGN(extent_start + size,
6984 fs_info->sectorsize);
6985
6986 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6987 path->slots[0],
6988 extent_start);
6989 }
6990 next:
6991 if (start >= extent_end) {
6992 path->slots[0]++;
6993 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6994 ret = btrfs_next_leaf(root, path);
6995 if (ret < 0) {
6996 err = ret;
6997 goto out;
6998 }
6999 if (ret > 0)
7000 goto not_found;
7001 leaf = path->nodes[0];
7002 }
7003 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7004 if (found_key.objectid != objectid ||
7005 found_key.type != BTRFS_EXTENT_DATA_KEY)
7006 goto not_found;
7007 if (start + len <= found_key.offset)
7008 goto not_found;
7009 if (start > found_key.offset)
7010 goto next;
7011 em->start = start;
7012 em->orig_start = start;
7013 em->len = found_key.offset - start;
7014 goto not_found_em;
7015 }
7016
7017 btrfs_extent_item_to_extent_map(inode, path, item,
7018 new_inline, em);
7019
7020 if (found_type == BTRFS_FILE_EXTENT_REG ||
7021 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7022 goto insert;
7023 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7024 unsigned long ptr;
7025 char *map;
7026 size_t size;
7027 size_t extent_offset;
7028 size_t copy_size;
7029
7030 if (new_inline)
7031 goto out;
7032
7033 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7034 extent_offset = page_offset(page) + pg_offset - extent_start;
7035 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7036 size - extent_offset);
7037 em->start = extent_start + extent_offset;
7038 em->len = ALIGN(copy_size, fs_info->sectorsize);
7039 em->orig_block_len = em->len;
7040 em->orig_start = em->start;
7041 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7042 if (create == 0 && !PageUptodate(page)) {
7043 if (btrfs_file_extent_compression(leaf, item) !=
7044 BTRFS_COMPRESS_NONE) {
7045 ret = uncompress_inline(path, page, pg_offset,
7046 extent_offset, item);
7047 if (ret) {
7048 err = ret;
7049 goto out;
7050 }
7051 } else {
7052 map = kmap(page);
7053 read_extent_buffer(leaf, map + pg_offset, ptr,
7054 copy_size);
7055 if (pg_offset + copy_size < PAGE_SIZE) {
7056 memset(map + pg_offset + copy_size, 0,
7057 PAGE_SIZE - pg_offset -
7058 copy_size);
7059 }
7060 kunmap(page);
7061 }
7062 flush_dcache_page(page);
7063 } else if (create && PageUptodate(page)) {
7064 BUG();
7065 if (!trans) {
7066 kunmap(page);
7067 free_extent_map(em);
7068 em = NULL;
7069
7070 btrfs_release_path(path);
7071 trans = btrfs_join_transaction(root);
7072
7073 if (IS_ERR(trans))
7074 return ERR_CAST(trans);
7075 goto again;
7076 }
7077 map = kmap(page);
7078 write_extent_buffer(leaf, map + pg_offset, ptr,
7079 copy_size);
7080 kunmap(page);
7081 btrfs_mark_buffer_dirty(leaf);
7082 }
7083 set_extent_uptodate(io_tree, em->start,
7084 extent_map_end(em) - 1, NULL, GFP_NOFS);
7085 goto insert;
7086 }
7087 not_found:
7088 em->start = start;
7089 em->orig_start = start;
7090 em->len = len;
7091 not_found_em:
7092 em->block_start = EXTENT_MAP_HOLE;
7093 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7094 insert:
7095 btrfs_release_path(path);
7096 if (em->start > start || extent_map_end(em) <= start) {
7097 btrfs_err(fs_info,
7098 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7099 em->start, em->len, start, len);
7100 err = -EIO;
7101 goto out;
7102 }
7103
7104 err = 0;
7105 write_lock(&em_tree->lock);
7106 ret = add_extent_mapping(em_tree, em, 0);
7107 /* it is possible that someone inserted the extent into the tree
7108 * while we had the lock dropped. It is also possible that
7109 * an overlapping map exists in the tree
7110 */
7111 if (ret == -EEXIST) {
7112 struct extent_map *existing;
7113
7114 ret = 0;
7115
7116 existing = search_extent_mapping(em_tree, start, len);
7117 /*
7118 * existing will always be non-NULL, since there must be
7119 * extent causing the -EEXIST.
7120 */
7121 if (existing->start == em->start &&
7122 extent_map_end(existing) >= extent_map_end(em) &&
7123 em->block_start == existing->block_start) {
7124 /*
7125 * The existing extent map already encompasses the
7126 * entire extent map we tried to add.
7127 */
7128 free_extent_map(em);
7129 em = existing;
7130 err = 0;
7131
7132 } else if (start >= extent_map_end(existing) ||
7133 start <= existing->start) {
7134 /*
7135 * The existing extent map is the one nearest to
7136 * the [start, start + len) range which overlaps
7137 */
7138 err = merge_extent_mapping(em_tree, existing,
7139 em, start);
7140 free_extent_map(existing);
7141 if (err) {
7142 free_extent_map(em);
7143 em = NULL;
7144 }
7145 } else {
7146 free_extent_map(em);
7147 em = existing;
7148 err = 0;
7149 }
7150 }
7151 write_unlock(&em_tree->lock);
7152 out:
7153
7154 trace_btrfs_get_extent(root, inode, em);
7155
7156 btrfs_free_path(path);
7157 if (trans) {
7158 ret = btrfs_end_transaction(trans);
7159 if (!err)
7160 err = ret;
7161 }
7162 if (err) {
7163 free_extent_map(em);
7164 return ERR_PTR(err);
7165 }
7166 BUG_ON(!em); /* Error is always set */
7167 return em;
7168 }
7169
7170 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7171 struct page *page,
7172 size_t pg_offset, u64 start, u64 len,
7173 int create)
7174 {
7175 struct extent_map *em;
7176 struct extent_map *hole_em = NULL;
7177 u64 range_start = start;
7178 u64 end;
7179 u64 found;
7180 u64 found_end;
7181 int err = 0;
7182
7183 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7184 if (IS_ERR(em))
7185 return em;
7186 /*
7187 * If our em maps to:
7188 * - a hole or
7189 * - a pre-alloc extent,
7190 * there might actually be delalloc bytes behind it.
7191 */
7192 if (em->block_start != EXTENT_MAP_HOLE &&
7193 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7194 return em;
7195 else
7196 hole_em = em;
7197
7198 /* check to see if we've wrapped (len == -1 or similar) */
7199 end = start + len;
7200 if (end < start)
7201 end = (u64)-1;
7202 else
7203 end -= 1;
7204
7205 em = NULL;
7206
7207 /* ok, we didn't find anything, lets look for delalloc */
7208 found = count_range_bits(&inode->io_tree, &range_start,
7209 end, len, EXTENT_DELALLOC, 1);
7210 found_end = range_start + found;
7211 if (found_end < range_start)
7212 found_end = (u64)-1;
7213
7214 /*
7215 * we didn't find anything useful, return
7216 * the original results from get_extent()
7217 */
7218 if (range_start > end || found_end <= start) {
7219 em = hole_em;
7220 hole_em = NULL;
7221 goto out;
7222 }
7223
7224 /* adjust the range_start to make sure it doesn't
7225 * go backwards from the start they passed in
7226 */
7227 range_start = max(start, range_start);
7228 found = found_end - range_start;
7229
7230 if (found > 0) {
7231 u64 hole_start = start;
7232 u64 hole_len = len;
7233
7234 em = alloc_extent_map();
7235 if (!em) {
7236 err = -ENOMEM;
7237 goto out;
7238 }
7239 /*
7240 * when btrfs_get_extent can't find anything it
7241 * returns one huge hole
7242 *
7243 * make sure what it found really fits our range, and
7244 * adjust to make sure it is based on the start from
7245 * the caller
7246 */
7247 if (hole_em) {
7248 u64 calc_end = extent_map_end(hole_em);
7249
7250 if (calc_end <= start || (hole_em->start > end)) {
7251 free_extent_map(hole_em);
7252 hole_em = NULL;
7253 } else {
7254 hole_start = max(hole_em->start, start);
7255 hole_len = calc_end - hole_start;
7256 }
7257 }
7258 em->bdev = NULL;
7259 if (hole_em && range_start > hole_start) {
7260 /* our hole starts before our delalloc, so we
7261 * have to return just the parts of the hole
7262 * that go until the delalloc starts
7263 */
7264 em->len = min(hole_len,
7265 range_start - hole_start);
7266 em->start = hole_start;
7267 em->orig_start = hole_start;
7268 /*
7269 * don't adjust block start at all,
7270 * it is fixed at EXTENT_MAP_HOLE
7271 */
7272 em->block_start = hole_em->block_start;
7273 em->block_len = hole_len;
7274 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7275 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7276 } else {
7277 em->start = range_start;
7278 em->len = found;
7279 em->orig_start = range_start;
7280 em->block_start = EXTENT_MAP_DELALLOC;
7281 em->block_len = found;
7282 }
7283 } else if (hole_em) {
7284 return hole_em;
7285 }
7286 out:
7287
7288 free_extent_map(hole_em);
7289 if (err) {
7290 free_extent_map(em);
7291 return ERR_PTR(err);
7292 }
7293 return em;
7294 }
7295
7296 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7297 const u64 start,
7298 const u64 len,
7299 const u64 orig_start,
7300 const u64 block_start,
7301 const u64 block_len,
7302 const u64 orig_block_len,
7303 const u64 ram_bytes,
7304 const int type)
7305 {
7306 struct extent_map *em = NULL;
7307 int ret;
7308
7309 if (type != BTRFS_ORDERED_NOCOW) {
7310 em = create_io_em(inode, start, len, orig_start,
7311 block_start, block_len, orig_block_len,
7312 ram_bytes,
7313 BTRFS_COMPRESS_NONE, /* compress_type */
7314 type);
7315 if (IS_ERR(em))
7316 goto out;
7317 }
7318 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7319 len, block_len, type);
7320 if (ret) {
7321 if (em) {
7322 free_extent_map(em);
7323 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7324 start + len - 1, 0);
7325 }
7326 em = ERR_PTR(ret);
7327 }
7328 out:
7329
7330 return em;
7331 }
7332
7333 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7334 u64 start, u64 len)
7335 {
7336 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7337 struct btrfs_root *root = BTRFS_I(inode)->root;
7338 struct extent_map *em;
7339 struct btrfs_key ins;
7340 u64 alloc_hint;
7341 int ret;
7342
7343 alloc_hint = get_extent_allocation_hint(inode, start, len);
7344 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7345 0, alloc_hint, &ins, 1, 1);
7346 if (ret)
7347 return ERR_PTR(ret);
7348
7349 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7350 ins.objectid, ins.offset, ins.offset,
7351 ins.offset, BTRFS_ORDERED_REGULAR);
7352 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7353 if (IS_ERR(em))
7354 btrfs_free_reserved_extent(fs_info, ins.objectid,
7355 ins.offset, 1);
7356
7357 return em;
7358 }
7359
7360 /*
7361 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7362 * block must be cow'd
7363 */
7364 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7365 u64 *orig_start, u64 *orig_block_len,
7366 u64 *ram_bytes)
7367 {
7368 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7369 struct btrfs_path *path;
7370 int ret;
7371 struct extent_buffer *leaf;
7372 struct btrfs_root *root = BTRFS_I(inode)->root;
7373 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7374 struct btrfs_file_extent_item *fi;
7375 struct btrfs_key key;
7376 u64 disk_bytenr;
7377 u64 backref_offset;
7378 u64 extent_end;
7379 u64 num_bytes;
7380 int slot;
7381 int found_type;
7382 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7383
7384 path = btrfs_alloc_path();
7385 if (!path)
7386 return -ENOMEM;
7387
7388 ret = btrfs_lookup_file_extent(NULL, root, path,
7389 btrfs_ino(BTRFS_I(inode)), offset, 0);
7390 if (ret < 0)
7391 goto out;
7392
7393 slot = path->slots[0];
7394 if (ret == 1) {
7395 if (slot == 0) {
7396 /* can't find the item, must cow */
7397 ret = 0;
7398 goto out;
7399 }
7400 slot--;
7401 }
7402 ret = 0;
7403 leaf = path->nodes[0];
7404 btrfs_item_key_to_cpu(leaf, &key, slot);
7405 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7406 key.type != BTRFS_EXTENT_DATA_KEY) {
7407 /* not our file or wrong item type, must cow */
7408 goto out;
7409 }
7410
7411 if (key.offset > offset) {
7412 /* Wrong offset, must cow */
7413 goto out;
7414 }
7415
7416 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7417 found_type = btrfs_file_extent_type(leaf, fi);
7418 if (found_type != BTRFS_FILE_EXTENT_REG &&
7419 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7420 /* not a regular extent, must cow */
7421 goto out;
7422 }
7423
7424 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7425 goto out;
7426
7427 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7428 if (extent_end <= offset)
7429 goto out;
7430
7431 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7432 if (disk_bytenr == 0)
7433 goto out;
7434
7435 if (btrfs_file_extent_compression(leaf, fi) ||
7436 btrfs_file_extent_encryption(leaf, fi) ||
7437 btrfs_file_extent_other_encoding(leaf, fi))
7438 goto out;
7439
7440 backref_offset = btrfs_file_extent_offset(leaf, fi);
7441
7442 if (orig_start) {
7443 *orig_start = key.offset - backref_offset;
7444 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7445 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7446 }
7447
7448 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7449 goto out;
7450
7451 num_bytes = min(offset + *len, extent_end) - offset;
7452 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7453 u64 range_end;
7454
7455 range_end = round_up(offset + num_bytes,
7456 root->fs_info->sectorsize) - 1;
7457 ret = test_range_bit(io_tree, offset, range_end,
7458 EXTENT_DELALLOC, 0, NULL);
7459 if (ret) {
7460 ret = -EAGAIN;
7461 goto out;
7462 }
7463 }
7464
7465 btrfs_release_path(path);
7466
7467 /*
7468 * look for other files referencing this extent, if we
7469 * find any we must cow
7470 */
7471
7472 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7473 key.offset - backref_offset, disk_bytenr);
7474 if (ret) {
7475 ret = 0;
7476 goto out;
7477 }
7478
7479 /*
7480 * adjust disk_bytenr and num_bytes to cover just the bytes
7481 * in this extent we are about to write. If there
7482 * are any csums in that range we have to cow in order
7483 * to keep the csums correct
7484 */
7485 disk_bytenr += backref_offset;
7486 disk_bytenr += offset - key.offset;
7487 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7488 goto out;
7489 /*
7490 * all of the above have passed, it is safe to overwrite this extent
7491 * without cow
7492 */
7493 *len = num_bytes;
7494 ret = 1;
7495 out:
7496 btrfs_free_path(path);
7497 return ret;
7498 }
7499
7500 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7501 {
7502 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7503 bool found = false;
7504 void **pagep = NULL;
7505 struct page *page = NULL;
7506 unsigned long start_idx;
7507 unsigned long end_idx;
7508
7509 start_idx = start >> PAGE_SHIFT;
7510
7511 /*
7512 * end is the last byte in the last page. end == start is legal
7513 */
7514 end_idx = end >> PAGE_SHIFT;
7515
7516 rcu_read_lock();
7517
7518 /* Most of the code in this while loop is lifted from
7519 * find_get_page. It's been modified to begin searching from a
7520 * page and return just the first page found in that range. If the
7521 * found idx is less than or equal to the end idx then we know that
7522 * a page exists. If no pages are found or if those pages are
7523 * outside of the range then we're fine (yay!) */
7524 while (page == NULL &&
7525 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7526 page = radix_tree_deref_slot(pagep);
7527 if (unlikely(!page))
7528 break;
7529
7530 if (radix_tree_exception(page)) {
7531 if (radix_tree_deref_retry(page)) {
7532 page = NULL;
7533 continue;
7534 }
7535 /*
7536 * Otherwise, shmem/tmpfs must be storing a swap entry
7537 * here as an exceptional entry: so return it without
7538 * attempting to raise page count.
7539 */
7540 page = NULL;
7541 break; /* TODO: Is this relevant for this use case? */
7542 }
7543
7544 if (!page_cache_get_speculative(page)) {
7545 page = NULL;
7546 continue;
7547 }
7548
7549 /*
7550 * Has the page moved?
7551 * This is part of the lockless pagecache protocol. See
7552 * include/linux/pagemap.h for details.
7553 */
7554 if (unlikely(page != *pagep)) {
7555 put_page(page);
7556 page = NULL;
7557 }
7558 }
7559
7560 if (page) {
7561 if (page->index <= end_idx)
7562 found = true;
7563 put_page(page);
7564 }
7565
7566 rcu_read_unlock();
7567 return found;
7568 }
7569
7570 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7571 struct extent_state **cached_state, int writing)
7572 {
7573 struct btrfs_ordered_extent *ordered;
7574 int ret = 0;
7575
7576 while (1) {
7577 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7578 cached_state);
7579 /*
7580 * We're concerned with the entire range that we're going to be
7581 * doing DIO to, so we need to make sure there's no ordered
7582 * extents in this range.
7583 */
7584 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7585 lockend - lockstart + 1);
7586
7587 /*
7588 * We need to make sure there are no buffered pages in this
7589 * range either, we could have raced between the invalidate in
7590 * generic_file_direct_write and locking the extent. The
7591 * invalidate needs to happen so that reads after a write do not
7592 * get stale data.
7593 */
7594 if (!ordered &&
7595 (!writing ||
7596 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7597 break;
7598
7599 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7600 cached_state, GFP_NOFS);
7601
7602 if (ordered) {
7603 /*
7604 * If we are doing a DIO read and the ordered extent we
7605 * found is for a buffered write, we can not wait for it
7606 * to complete and retry, because if we do so we can
7607 * deadlock with concurrent buffered writes on page
7608 * locks. This happens only if our DIO read covers more
7609 * than one extent map, if at this point has already
7610 * created an ordered extent for a previous extent map
7611 * and locked its range in the inode's io tree, and a
7612 * concurrent write against that previous extent map's
7613 * range and this range started (we unlock the ranges
7614 * in the io tree only when the bios complete and
7615 * buffered writes always lock pages before attempting
7616 * to lock range in the io tree).
7617 */
7618 if (writing ||
7619 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7620 btrfs_start_ordered_extent(inode, ordered, 1);
7621 else
7622 ret = -ENOTBLK;
7623 btrfs_put_ordered_extent(ordered);
7624 } else {
7625 /*
7626 * We could trigger writeback for this range (and wait
7627 * for it to complete) and then invalidate the pages for
7628 * this range (through invalidate_inode_pages2_range()),
7629 * but that can lead us to a deadlock with a concurrent
7630 * call to readpages() (a buffered read or a defrag call
7631 * triggered a readahead) on a page lock due to an
7632 * ordered dio extent we created before but did not have
7633 * yet a corresponding bio submitted (whence it can not
7634 * complete), which makes readpages() wait for that
7635 * ordered extent to complete while holding a lock on
7636 * that page.
7637 */
7638 ret = -ENOTBLK;
7639 }
7640
7641 if (ret)
7642 break;
7643
7644 cond_resched();
7645 }
7646
7647 return ret;
7648 }
7649
7650 /* The callers of this must take lock_extent() */
7651 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7652 u64 orig_start, u64 block_start,
7653 u64 block_len, u64 orig_block_len,
7654 u64 ram_bytes, int compress_type,
7655 int type)
7656 {
7657 struct extent_map_tree *em_tree;
7658 struct extent_map *em;
7659 struct btrfs_root *root = BTRFS_I(inode)->root;
7660 int ret;
7661
7662 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7663 type == BTRFS_ORDERED_COMPRESSED ||
7664 type == BTRFS_ORDERED_NOCOW ||
7665 type == BTRFS_ORDERED_REGULAR);
7666
7667 em_tree = &BTRFS_I(inode)->extent_tree;
7668 em = alloc_extent_map();
7669 if (!em)
7670 return ERR_PTR(-ENOMEM);
7671
7672 em->start = start;
7673 em->orig_start = orig_start;
7674 em->len = len;
7675 em->block_len = block_len;
7676 em->block_start = block_start;
7677 em->bdev = root->fs_info->fs_devices->latest_bdev;
7678 em->orig_block_len = orig_block_len;
7679 em->ram_bytes = ram_bytes;
7680 em->generation = -1;
7681 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7682 if (type == BTRFS_ORDERED_PREALLOC) {
7683 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7684 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7685 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7686 em->compress_type = compress_type;
7687 }
7688
7689 do {
7690 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7691 em->start + em->len - 1, 0);
7692 write_lock(&em_tree->lock);
7693 ret = add_extent_mapping(em_tree, em, 1);
7694 write_unlock(&em_tree->lock);
7695 /*
7696 * The caller has taken lock_extent(), who could race with us
7697 * to add em?
7698 */
7699 } while (ret == -EEXIST);
7700
7701 if (ret) {
7702 free_extent_map(em);
7703 return ERR_PTR(ret);
7704 }
7705
7706 /* em got 2 refs now, callers needs to do free_extent_map once. */
7707 return em;
7708 }
7709
7710 static void adjust_dio_outstanding_extents(struct inode *inode,
7711 struct btrfs_dio_data *dio_data,
7712 const u64 len)
7713 {
7714 unsigned num_extents = count_max_extents(len);
7715
7716 /*
7717 * If we have an outstanding_extents count still set then we're
7718 * within our reservation, otherwise we need to adjust our inode
7719 * counter appropriately.
7720 */
7721 if (dio_data->outstanding_extents >= num_extents) {
7722 dio_data->outstanding_extents -= num_extents;
7723 } else {
7724 /*
7725 * If dio write length has been split due to no large enough
7726 * contiguous space, we need to compensate our inode counter
7727 * appropriately.
7728 */
7729 u64 num_needed = num_extents - dio_data->outstanding_extents;
7730
7731 spin_lock(&BTRFS_I(inode)->lock);
7732 BTRFS_I(inode)->outstanding_extents += num_needed;
7733 spin_unlock(&BTRFS_I(inode)->lock);
7734 }
7735 }
7736
7737 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7738 struct buffer_head *bh_result, int create)
7739 {
7740 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7741 struct extent_map *em;
7742 struct extent_state *cached_state = NULL;
7743 struct btrfs_dio_data *dio_data = NULL;
7744 u64 start = iblock << inode->i_blkbits;
7745 u64 lockstart, lockend;
7746 u64 len = bh_result->b_size;
7747 int unlock_bits = EXTENT_LOCKED;
7748 int ret = 0;
7749
7750 if (create)
7751 unlock_bits |= EXTENT_DIRTY;
7752 else
7753 len = min_t(u64, len, fs_info->sectorsize);
7754
7755 lockstart = start;
7756 lockend = start + len - 1;
7757
7758 if (current->journal_info) {
7759 /*
7760 * Need to pull our outstanding extents and set journal_info to NULL so
7761 * that anything that needs to check if there's a transaction doesn't get
7762 * confused.
7763 */
7764 dio_data = current->journal_info;
7765 current->journal_info = NULL;
7766 }
7767
7768 /*
7769 * If this errors out it's because we couldn't invalidate pagecache for
7770 * this range and we need to fallback to buffered.
7771 */
7772 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7773 create)) {
7774 ret = -ENOTBLK;
7775 goto err;
7776 }
7777
7778 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7779 if (IS_ERR(em)) {
7780 ret = PTR_ERR(em);
7781 goto unlock_err;
7782 }
7783
7784 /*
7785 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7786 * io. INLINE is special, and we could probably kludge it in here, but
7787 * it's still buffered so for safety lets just fall back to the generic
7788 * buffered path.
7789 *
7790 * For COMPRESSED we _have_ to read the entire extent in so we can
7791 * decompress it, so there will be buffering required no matter what we
7792 * do, so go ahead and fallback to buffered.
7793 *
7794 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7795 * to buffered IO. Don't blame me, this is the price we pay for using
7796 * the generic code.
7797 */
7798 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7799 em->block_start == EXTENT_MAP_INLINE) {
7800 free_extent_map(em);
7801 ret = -ENOTBLK;
7802 goto unlock_err;
7803 }
7804
7805 /* Just a good old fashioned hole, return */
7806 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7807 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7808 free_extent_map(em);
7809 goto unlock_err;
7810 }
7811
7812 /*
7813 * We don't allocate a new extent in the following cases
7814 *
7815 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7816 * existing extent.
7817 * 2) The extent is marked as PREALLOC. We're good to go here and can
7818 * just use the extent.
7819 *
7820 */
7821 if (!create) {
7822 len = min(len, em->len - (start - em->start));
7823 lockstart = start + len;
7824 goto unlock;
7825 }
7826
7827 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7828 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7829 em->block_start != EXTENT_MAP_HOLE)) {
7830 int type;
7831 u64 block_start, orig_start, orig_block_len, ram_bytes;
7832
7833 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7834 type = BTRFS_ORDERED_PREALLOC;
7835 else
7836 type = BTRFS_ORDERED_NOCOW;
7837 len = min(len, em->len - (start - em->start));
7838 block_start = em->block_start + (start - em->start);
7839
7840 if (can_nocow_extent(inode, start, &len, &orig_start,
7841 &orig_block_len, &ram_bytes) == 1 &&
7842 btrfs_inc_nocow_writers(fs_info, block_start)) {
7843 struct extent_map *em2;
7844
7845 em2 = btrfs_create_dio_extent(inode, start, len,
7846 orig_start, block_start,
7847 len, orig_block_len,
7848 ram_bytes, type);
7849 btrfs_dec_nocow_writers(fs_info, block_start);
7850 if (type == BTRFS_ORDERED_PREALLOC) {
7851 free_extent_map(em);
7852 em = em2;
7853 }
7854 if (em2 && IS_ERR(em2)) {
7855 ret = PTR_ERR(em2);
7856 goto unlock_err;
7857 }
7858 /*
7859 * For inode marked NODATACOW or extent marked PREALLOC,
7860 * use the existing or preallocated extent, so does not
7861 * need to adjust btrfs_space_info's bytes_may_use.
7862 */
7863 btrfs_free_reserved_data_space_noquota(inode,
7864 start, len);
7865 goto unlock;
7866 }
7867 }
7868
7869 /*
7870 * this will cow the extent, reset the len in case we changed
7871 * it above
7872 */
7873 len = bh_result->b_size;
7874 free_extent_map(em);
7875 em = btrfs_new_extent_direct(inode, start, len);
7876 if (IS_ERR(em)) {
7877 ret = PTR_ERR(em);
7878 goto unlock_err;
7879 }
7880 len = min(len, em->len - (start - em->start));
7881 unlock:
7882 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7883 inode->i_blkbits;
7884 bh_result->b_size = len;
7885 bh_result->b_bdev = em->bdev;
7886 set_buffer_mapped(bh_result);
7887 if (create) {
7888 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7889 set_buffer_new(bh_result);
7890
7891 /*
7892 * Need to update the i_size under the extent lock so buffered
7893 * readers will get the updated i_size when we unlock.
7894 */
7895 if (!dio_data->overwrite && start + len > i_size_read(inode))
7896 i_size_write(inode, start + len);
7897
7898 adjust_dio_outstanding_extents(inode, dio_data, len);
7899 WARN_ON(dio_data->reserve < len);
7900 dio_data->reserve -= len;
7901 dio_data->unsubmitted_oe_range_end = start + len;
7902 current->journal_info = dio_data;
7903 }
7904
7905 /*
7906 * In the case of write we need to clear and unlock the entire range,
7907 * in the case of read we need to unlock only the end area that we
7908 * aren't using if there is any left over space.
7909 */
7910 if (lockstart < lockend) {
7911 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7912 lockend, unlock_bits, 1, 0,
7913 &cached_state, GFP_NOFS);
7914 } else {
7915 free_extent_state(cached_state);
7916 }
7917
7918 free_extent_map(em);
7919
7920 return 0;
7921
7922 unlock_err:
7923 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7924 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7925 err:
7926 if (dio_data)
7927 current->journal_info = dio_data;
7928 /*
7929 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7930 * write less data then expected, so that we don't underflow our inode's
7931 * outstanding extents counter.
7932 */
7933 if (create && dio_data)
7934 adjust_dio_outstanding_extents(inode, dio_data, len);
7935
7936 return ret;
7937 }
7938
7939 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7940 struct bio *bio,
7941 int mirror_num)
7942 {
7943 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7944 blk_status_t ret;
7945
7946 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7947
7948 bio_get(bio);
7949
7950 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7951 if (ret)
7952 goto err;
7953
7954 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7955 err:
7956 bio_put(bio);
7957 return ret;
7958 }
7959
7960 static int btrfs_check_dio_repairable(struct inode *inode,
7961 struct bio *failed_bio,
7962 struct io_failure_record *failrec,
7963 int failed_mirror)
7964 {
7965 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7966 int num_copies;
7967
7968 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7969 if (num_copies == 1) {
7970 /*
7971 * we only have a single copy of the data, so don't bother with
7972 * all the retry and error correction code that follows. no
7973 * matter what the error is, it is very likely to persist.
7974 */
7975 btrfs_debug(fs_info,
7976 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7977 num_copies, failrec->this_mirror, failed_mirror);
7978 return 0;
7979 }
7980
7981 failrec->failed_mirror = failed_mirror;
7982 failrec->this_mirror++;
7983 if (failrec->this_mirror == failed_mirror)
7984 failrec->this_mirror++;
7985
7986 if (failrec->this_mirror > num_copies) {
7987 btrfs_debug(fs_info,
7988 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7989 num_copies, failrec->this_mirror, failed_mirror);
7990 return 0;
7991 }
7992
7993 return 1;
7994 }
7995
7996 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7997 struct page *page, unsigned int pgoff,
7998 u64 start, u64 end, int failed_mirror,
7999 bio_end_io_t *repair_endio, void *repair_arg)
8000 {
8001 struct io_failure_record *failrec;
8002 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8003 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8004 struct bio *bio;
8005 int isector;
8006 int read_mode = 0;
8007 int segs;
8008 int ret;
8009 blk_status_t status;
8010
8011 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8012
8013 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8014 if (ret)
8015 return errno_to_blk_status(ret);
8016
8017 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8018 failed_mirror);
8019 if (!ret) {
8020 free_io_failure(failure_tree, io_tree, failrec);
8021 return BLK_STS_IOERR;
8022 }
8023
8024 segs = bio_segments(failed_bio);
8025 if (segs > 1 ||
8026 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8027 read_mode |= REQ_FAILFAST_DEV;
8028
8029 isector = start - btrfs_io_bio(failed_bio)->logical;
8030 isector >>= inode->i_sb->s_blocksize_bits;
8031 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8032 pgoff, isector, repair_endio, repair_arg);
8033 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8034
8035 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8036 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
8037 read_mode, failrec->this_mirror, failrec->in_validation);
8038
8039 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8040 if (status) {
8041 free_io_failure(failure_tree, io_tree, failrec);
8042 bio_put(bio);
8043 }
8044
8045 return status;
8046 }
8047
8048 struct btrfs_retry_complete {
8049 struct completion done;
8050 struct inode *inode;
8051 u64 start;
8052 int uptodate;
8053 };
8054
8055 static void btrfs_retry_endio_nocsum(struct bio *bio)
8056 {
8057 struct btrfs_retry_complete *done = bio->bi_private;
8058 struct inode *inode = done->inode;
8059 struct bio_vec *bvec;
8060 struct extent_io_tree *io_tree, *failure_tree;
8061 int i;
8062
8063 if (bio->bi_status)
8064 goto end;
8065
8066 ASSERT(bio->bi_vcnt == 1);
8067 io_tree = &BTRFS_I(inode)->io_tree;
8068 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8069 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8070
8071 done->uptodate = 1;
8072 ASSERT(!bio_flagged(bio, BIO_CLONED));
8073 bio_for_each_segment_all(bvec, bio, i)
8074 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8075 io_tree, done->start, bvec->bv_page,
8076 btrfs_ino(BTRFS_I(inode)), 0);
8077 end:
8078 complete(&done->done);
8079 bio_put(bio);
8080 }
8081
8082 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8083 struct btrfs_io_bio *io_bio)
8084 {
8085 struct btrfs_fs_info *fs_info;
8086 struct bio_vec bvec;
8087 struct bvec_iter iter;
8088 struct btrfs_retry_complete done;
8089 u64 start;
8090 unsigned int pgoff;
8091 u32 sectorsize;
8092 int nr_sectors;
8093 blk_status_t ret;
8094 blk_status_t err = BLK_STS_OK;
8095
8096 fs_info = BTRFS_I(inode)->root->fs_info;
8097 sectorsize = fs_info->sectorsize;
8098
8099 start = io_bio->logical;
8100 done.inode = inode;
8101 io_bio->bio.bi_iter = io_bio->iter;
8102
8103 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8104 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8105 pgoff = bvec.bv_offset;
8106
8107 next_block_or_try_again:
8108 done.uptodate = 0;
8109 done.start = start;
8110 init_completion(&done.done);
8111
8112 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8113 pgoff, start, start + sectorsize - 1,
8114 io_bio->mirror_num,
8115 btrfs_retry_endio_nocsum, &done);
8116 if (ret) {
8117 err = ret;
8118 goto next;
8119 }
8120
8121 wait_for_completion(&done.done);
8122
8123 if (!done.uptodate) {
8124 /* We might have another mirror, so try again */
8125 goto next_block_or_try_again;
8126 }
8127
8128 next:
8129 start += sectorsize;
8130
8131 nr_sectors--;
8132 if (nr_sectors) {
8133 pgoff += sectorsize;
8134 ASSERT(pgoff < PAGE_SIZE);
8135 goto next_block_or_try_again;
8136 }
8137 }
8138
8139 return err;
8140 }
8141
8142 static void btrfs_retry_endio(struct bio *bio)
8143 {
8144 struct btrfs_retry_complete *done = bio->bi_private;
8145 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8146 struct extent_io_tree *io_tree, *failure_tree;
8147 struct inode *inode = done->inode;
8148 struct bio_vec *bvec;
8149 int uptodate;
8150 int ret;
8151 int i;
8152
8153 if (bio->bi_status)
8154 goto end;
8155
8156 uptodate = 1;
8157
8158 ASSERT(bio->bi_vcnt == 1);
8159 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8160
8161 io_tree = &BTRFS_I(inode)->io_tree;
8162 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8163
8164 ASSERT(!bio_flagged(bio, BIO_CLONED));
8165 bio_for_each_segment_all(bvec, bio, i) {
8166 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8167 bvec->bv_offset, done->start,
8168 bvec->bv_len);
8169 if (!ret)
8170 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8171 failure_tree, io_tree, done->start,
8172 bvec->bv_page,
8173 btrfs_ino(BTRFS_I(inode)),
8174 bvec->bv_offset);
8175 else
8176 uptodate = 0;
8177 }
8178
8179 done->uptodate = uptodate;
8180 end:
8181 complete(&done->done);
8182 bio_put(bio);
8183 }
8184
8185 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8186 struct btrfs_io_bio *io_bio, blk_status_t err)
8187 {
8188 struct btrfs_fs_info *fs_info;
8189 struct bio_vec bvec;
8190 struct bvec_iter iter;
8191 struct btrfs_retry_complete done;
8192 u64 start;
8193 u64 offset = 0;
8194 u32 sectorsize;
8195 int nr_sectors;
8196 unsigned int pgoff;
8197 int csum_pos;
8198 bool uptodate = (err == 0);
8199 int ret;
8200 blk_status_t status;
8201
8202 fs_info = BTRFS_I(inode)->root->fs_info;
8203 sectorsize = fs_info->sectorsize;
8204
8205 err = BLK_STS_OK;
8206 start = io_bio->logical;
8207 done.inode = inode;
8208 io_bio->bio.bi_iter = io_bio->iter;
8209
8210 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8211 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8212
8213 pgoff = bvec.bv_offset;
8214 next_block:
8215 if (uptodate) {
8216 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8217 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8218 bvec.bv_page, pgoff, start, sectorsize);
8219 if (likely(!ret))
8220 goto next;
8221 }
8222 try_again:
8223 done.uptodate = 0;
8224 done.start = start;
8225 init_completion(&done.done);
8226
8227 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8228 pgoff, start, start + sectorsize - 1,
8229 io_bio->mirror_num, btrfs_retry_endio,
8230 &done);
8231 if (status) {
8232 err = status;
8233 goto next;
8234 }
8235
8236 wait_for_completion(&done.done);
8237
8238 if (!done.uptodate) {
8239 /* We might have another mirror, so try again */
8240 goto try_again;
8241 }
8242 next:
8243 offset += sectorsize;
8244 start += sectorsize;
8245
8246 ASSERT(nr_sectors);
8247
8248 nr_sectors--;
8249 if (nr_sectors) {
8250 pgoff += sectorsize;
8251 ASSERT(pgoff < PAGE_SIZE);
8252 goto next_block;
8253 }
8254 }
8255
8256 return err;
8257 }
8258
8259 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8260 struct btrfs_io_bio *io_bio, blk_status_t err)
8261 {
8262 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8263
8264 if (skip_csum) {
8265 if (unlikely(err))
8266 return __btrfs_correct_data_nocsum(inode, io_bio);
8267 else
8268 return BLK_STS_OK;
8269 } else {
8270 return __btrfs_subio_endio_read(inode, io_bio, err);
8271 }
8272 }
8273
8274 static void btrfs_endio_direct_read(struct bio *bio)
8275 {
8276 struct btrfs_dio_private *dip = bio->bi_private;
8277 struct inode *inode = dip->inode;
8278 struct bio *dio_bio;
8279 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8280 blk_status_t err = bio->bi_status;
8281
8282 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) {
8283 err = btrfs_subio_endio_read(inode, io_bio, err);
8284 if (!err)
8285 bio->bi_status = 0;
8286 }
8287
8288 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8289 dip->logical_offset + dip->bytes - 1);
8290 dio_bio = dip->dio_bio;
8291
8292 kfree(dip);
8293
8294 dio_bio->bi_status = bio->bi_status;
8295 dio_end_io(dio_bio);
8296
8297 if (io_bio->end_io)
8298 io_bio->end_io(io_bio, blk_status_to_errno(err));
8299 bio_put(bio);
8300 }
8301
8302 static void __endio_write_update_ordered(struct inode *inode,
8303 const u64 offset, const u64 bytes,
8304 const bool uptodate)
8305 {
8306 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8307 struct btrfs_ordered_extent *ordered = NULL;
8308 struct btrfs_workqueue *wq;
8309 btrfs_work_func_t func;
8310 u64 ordered_offset = offset;
8311 u64 ordered_bytes = bytes;
8312 int ret;
8313
8314 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8315 wq = fs_info->endio_freespace_worker;
8316 func = btrfs_freespace_write_helper;
8317 } else {
8318 wq = fs_info->endio_write_workers;
8319 func = btrfs_endio_write_helper;
8320 }
8321
8322 again:
8323 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8324 &ordered_offset,
8325 ordered_bytes,
8326 uptodate);
8327 if (!ret)
8328 goto out_test;
8329
8330 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8331 btrfs_queue_work(wq, &ordered->work);
8332 out_test:
8333 /*
8334 * our bio might span multiple ordered extents. If we haven't
8335 * completed the accounting for the whole dio, go back and try again
8336 */
8337 if (ordered_offset < offset + bytes) {
8338 ordered_bytes = offset + bytes - ordered_offset;
8339 ordered = NULL;
8340 goto again;
8341 }
8342 }
8343
8344 static void btrfs_endio_direct_write(struct bio *bio)
8345 {
8346 struct btrfs_dio_private *dip = bio->bi_private;
8347 struct bio *dio_bio = dip->dio_bio;
8348
8349 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8350 dip->bytes, !bio->bi_status);
8351
8352 kfree(dip);
8353
8354 dio_bio->bi_status = bio->bi_status;
8355 dio_end_io(dio_bio);
8356 bio_put(bio);
8357 }
8358
8359 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8360 struct bio *bio, int mirror_num,
8361 unsigned long bio_flags, u64 offset)
8362 {
8363 struct inode *inode = private_data;
8364 blk_status_t ret;
8365 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8366 BUG_ON(ret); /* -ENOMEM */
8367 return 0;
8368 }
8369
8370 static void btrfs_end_dio_bio(struct bio *bio)
8371 {
8372 struct btrfs_dio_private *dip = bio->bi_private;
8373 blk_status_t err = bio->bi_status;
8374
8375 if (err)
8376 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8377 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8378 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8379 bio->bi_opf,
8380 (unsigned long long)bio->bi_iter.bi_sector,
8381 bio->bi_iter.bi_size, err);
8382
8383 if (dip->subio_endio)
8384 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8385
8386 if (err) {
8387 dip->errors = 1;
8388
8389 /*
8390 * before atomic variable goto zero, we must make sure
8391 * dip->errors is perceived to be set.
8392 */
8393 smp_mb__before_atomic();
8394 }
8395
8396 /* if there are more bios still pending for this dio, just exit */
8397 if (!atomic_dec_and_test(&dip->pending_bios))
8398 goto out;
8399
8400 if (dip->errors) {
8401 bio_io_error(dip->orig_bio);
8402 } else {
8403 dip->dio_bio->bi_status = 0;
8404 bio_endio(dip->orig_bio);
8405 }
8406 out:
8407 bio_put(bio);
8408 }
8409
8410 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8411 struct btrfs_dio_private *dip,
8412 struct bio *bio,
8413 u64 file_offset)
8414 {
8415 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8416 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8417 blk_status_t ret;
8418
8419 /*
8420 * We load all the csum data we need when we submit
8421 * the first bio to reduce the csum tree search and
8422 * contention.
8423 */
8424 if (dip->logical_offset == file_offset) {
8425 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8426 file_offset);
8427 if (ret)
8428 return ret;
8429 }
8430
8431 if (bio == dip->orig_bio)
8432 return 0;
8433
8434 file_offset -= dip->logical_offset;
8435 file_offset >>= inode->i_sb->s_blocksize_bits;
8436 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8437
8438 return 0;
8439 }
8440
8441 static inline blk_status_t
8442 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8443 int skip_sum, int async_submit)
8444 {
8445 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8446 struct btrfs_dio_private *dip = bio->bi_private;
8447 bool write = bio_op(bio) == REQ_OP_WRITE;
8448 blk_status_t ret;
8449
8450 if (async_submit)
8451 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8452
8453 bio_get(bio);
8454
8455 if (!write) {
8456 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8457 if (ret)
8458 goto err;
8459 }
8460
8461 if (skip_sum)
8462 goto map;
8463
8464 if (write && async_submit) {
8465 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8466 file_offset, inode,
8467 __btrfs_submit_bio_start_direct_io,
8468 __btrfs_submit_bio_done);
8469 goto err;
8470 } else if (write) {
8471 /*
8472 * If we aren't doing async submit, calculate the csum of the
8473 * bio now.
8474 */
8475 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8476 if (ret)
8477 goto err;
8478 } else {
8479 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8480 file_offset);
8481 if (ret)
8482 goto err;
8483 }
8484 map:
8485 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8486 err:
8487 bio_put(bio);
8488 return ret;
8489 }
8490
8491 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8492 int skip_sum)
8493 {
8494 struct inode *inode = dip->inode;
8495 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8496 struct bio *bio;
8497 struct bio *orig_bio = dip->orig_bio;
8498 u64 start_sector = orig_bio->bi_iter.bi_sector;
8499 u64 file_offset = dip->logical_offset;
8500 u64 map_length;
8501 int async_submit = 0;
8502 u64 submit_len;
8503 int clone_offset = 0;
8504 int clone_len;
8505 int ret;
8506 blk_status_t status;
8507
8508 map_length = orig_bio->bi_iter.bi_size;
8509 submit_len = map_length;
8510 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8511 &map_length, NULL, 0);
8512 if (ret)
8513 return -EIO;
8514
8515 if (map_length >= submit_len) {
8516 bio = orig_bio;
8517 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8518 goto submit;
8519 }
8520
8521 /* async crcs make it difficult to collect full stripe writes. */
8522 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8523 async_submit = 0;
8524 else
8525 async_submit = 1;
8526
8527 /* bio split */
8528 ASSERT(map_length <= INT_MAX);
8529 atomic_inc(&dip->pending_bios);
8530 do {
8531 clone_len = min_t(int, submit_len, map_length);
8532
8533 /*
8534 * This will never fail as it's passing GPF_NOFS and
8535 * the allocation is backed by btrfs_bioset.
8536 */
8537 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8538 clone_len);
8539 bio->bi_private = dip;
8540 bio->bi_end_io = btrfs_end_dio_bio;
8541 btrfs_io_bio(bio)->logical = file_offset;
8542
8543 ASSERT(submit_len >= clone_len);
8544 submit_len -= clone_len;
8545 if (submit_len == 0)
8546 break;
8547
8548 /*
8549 * Increase the count before we submit the bio so we know
8550 * the end IO handler won't happen before we increase the
8551 * count. Otherwise, the dip might get freed before we're
8552 * done setting it up.
8553 */
8554 atomic_inc(&dip->pending_bios);
8555
8556 status = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8557 async_submit);
8558 if (status) {
8559 bio_put(bio);
8560 atomic_dec(&dip->pending_bios);
8561 goto out_err;
8562 }
8563
8564 clone_offset += clone_len;
8565 start_sector += clone_len >> 9;
8566 file_offset += clone_len;
8567
8568 map_length = submit_len;
8569 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8570 start_sector << 9, &map_length, NULL, 0);
8571 if (ret)
8572 goto out_err;
8573 } while (submit_len > 0);
8574
8575 submit:
8576 status = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8577 async_submit);
8578 if (!status)
8579 return 0;
8580
8581 bio_put(bio);
8582 out_err:
8583 dip->errors = 1;
8584 /*
8585 * before atomic variable goto zero, we must
8586 * make sure dip->errors is perceived to be set.
8587 */
8588 smp_mb__before_atomic();
8589 if (atomic_dec_and_test(&dip->pending_bios))
8590 bio_io_error(dip->orig_bio);
8591
8592 /* bio_end_io() will handle error, so we needn't return it */
8593 return 0;
8594 }
8595
8596 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8597 loff_t file_offset)
8598 {
8599 struct btrfs_dio_private *dip = NULL;
8600 struct bio *bio = NULL;
8601 struct btrfs_io_bio *io_bio;
8602 int skip_sum;
8603 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8604 int ret = 0;
8605
8606 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8607
8608 bio = btrfs_bio_clone(dio_bio);
8609
8610 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8611 if (!dip) {
8612 ret = -ENOMEM;
8613 goto free_ordered;
8614 }
8615
8616 dip->private = dio_bio->bi_private;
8617 dip->inode = inode;
8618 dip->logical_offset = file_offset;
8619 dip->bytes = dio_bio->bi_iter.bi_size;
8620 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8621 bio->bi_private = dip;
8622 dip->orig_bio = bio;
8623 dip->dio_bio = dio_bio;
8624 atomic_set(&dip->pending_bios, 0);
8625 io_bio = btrfs_io_bio(bio);
8626 io_bio->logical = file_offset;
8627
8628 if (write) {
8629 bio->bi_end_io = btrfs_endio_direct_write;
8630 } else {
8631 bio->bi_end_io = btrfs_endio_direct_read;
8632 dip->subio_endio = btrfs_subio_endio_read;
8633 }
8634
8635 /*
8636 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8637 * even if we fail to submit a bio, because in such case we do the
8638 * corresponding error handling below and it must not be done a second
8639 * time by btrfs_direct_IO().
8640 */
8641 if (write) {
8642 struct btrfs_dio_data *dio_data = current->journal_info;
8643
8644 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8645 dip->bytes;
8646 dio_data->unsubmitted_oe_range_start =
8647 dio_data->unsubmitted_oe_range_end;
8648 }
8649
8650 ret = btrfs_submit_direct_hook(dip, skip_sum);
8651 if (!ret)
8652 return;
8653
8654 if (io_bio->end_io)
8655 io_bio->end_io(io_bio, ret);
8656
8657 free_ordered:
8658 /*
8659 * If we arrived here it means either we failed to submit the dip
8660 * or we either failed to clone the dio_bio or failed to allocate the
8661 * dip. If we cloned the dio_bio and allocated the dip, we can just
8662 * call bio_endio against our io_bio so that we get proper resource
8663 * cleanup if we fail to submit the dip, otherwise, we must do the
8664 * same as btrfs_endio_direct_[write|read] because we can't call these
8665 * callbacks - they require an allocated dip and a clone of dio_bio.
8666 */
8667 if (bio && dip) {
8668 bio_io_error(bio);
8669 /*
8670 * The end io callbacks free our dip, do the final put on bio
8671 * and all the cleanup and final put for dio_bio (through
8672 * dio_end_io()).
8673 */
8674 dip = NULL;
8675 bio = NULL;
8676 } else {
8677 if (write)
8678 __endio_write_update_ordered(inode,
8679 file_offset,
8680 dio_bio->bi_iter.bi_size,
8681 false);
8682 else
8683 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8684 file_offset + dio_bio->bi_iter.bi_size - 1);
8685
8686 dio_bio->bi_status = BLK_STS_IOERR;
8687 /*
8688 * Releases and cleans up our dio_bio, no need to bio_put()
8689 * nor bio_endio()/bio_io_error() against dio_bio.
8690 */
8691 dio_end_io(dio_bio);
8692 }
8693 if (bio)
8694 bio_put(bio);
8695 kfree(dip);
8696 }
8697
8698 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8699 struct kiocb *iocb,
8700 const struct iov_iter *iter, loff_t offset)
8701 {
8702 int seg;
8703 int i;
8704 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8705 ssize_t retval = -EINVAL;
8706
8707 if (offset & blocksize_mask)
8708 goto out;
8709
8710 if (iov_iter_alignment(iter) & blocksize_mask)
8711 goto out;
8712
8713 /* If this is a write we don't need to check anymore */
8714 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8715 return 0;
8716 /*
8717 * Check to make sure we don't have duplicate iov_base's in this
8718 * iovec, if so return EINVAL, otherwise we'll get csum errors
8719 * when reading back.
8720 */
8721 for (seg = 0; seg < iter->nr_segs; seg++) {
8722 for (i = seg + 1; i < iter->nr_segs; i++) {
8723 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8724 goto out;
8725 }
8726 }
8727 retval = 0;
8728 out:
8729 return retval;
8730 }
8731
8732 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8733 {
8734 struct file *file = iocb->ki_filp;
8735 struct inode *inode = file->f_mapping->host;
8736 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8737 struct btrfs_dio_data dio_data = { 0 };
8738 struct extent_changeset *data_reserved = NULL;
8739 loff_t offset = iocb->ki_pos;
8740 size_t count = 0;
8741 int flags = 0;
8742 bool wakeup = true;
8743 bool relock = false;
8744 ssize_t ret;
8745
8746 if (check_direct_IO(fs_info, iocb, iter, offset))
8747 return 0;
8748
8749 inode_dio_begin(inode);
8750 smp_mb__after_atomic();
8751
8752 /*
8753 * The generic stuff only does filemap_write_and_wait_range, which
8754 * isn't enough if we've written compressed pages to this area, so
8755 * we need to flush the dirty pages again to make absolutely sure
8756 * that any outstanding dirty pages are on disk.
8757 */
8758 count = iov_iter_count(iter);
8759 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8760 &BTRFS_I(inode)->runtime_flags))
8761 filemap_fdatawrite_range(inode->i_mapping, offset,
8762 offset + count - 1);
8763
8764 if (iov_iter_rw(iter) == WRITE) {
8765 /*
8766 * If the write DIO is beyond the EOF, we need update
8767 * the isize, but it is protected by i_mutex. So we can
8768 * not unlock the i_mutex at this case.
8769 */
8770 if (offset + count <= inode->i_size) {
8771 dio_data.overwrite = 1;
8772 inode_unlock(inode);
8773 relock = true;
8774 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8775 ret = -EAGAIN;
8776 goto out;
8777 }
8778 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8779 offset, count);
8780 if (ret)
8781 goto out;
8782 dio_data.outstanding_extents = count_max_extents(count);
8783
8784 /*
8785 * We need to know how many extents we reserved so that we can
8786 * do the accounting properly if we go over the number we
8787 * originally calculated. Abuse current->journal_info for this.
8788 */
8789 dio_data.reserve = round_up(count,
8790 fs_info->sectorsize);
8791 dio_data.unsubmitted_oe_range_start = (u64)offset;
8792 dio_data.unsubmitted_oe_range_end = (u64)offset;
8793 current->journal_info = &dio_data;
8794 down_read(&BTRFS_I(inode)->dio_sem);
8795 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8796 &BTRFS_I(inode)->runtime_flags)) {
8797 inode_dio_end(inode);
8798 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8799 wakeup = false;
8800 }
8801
8802 ret = __blockdev_direct_IO(iocb, inode,
8803 fs_info->fs_devices->latest_bdev,
8804 iter, btrfs_get_blocks_direct, NULL,
8805 btrfs_submit_direct, flags);
8806 if (iov_iter_rw(iter) == WRITE) {
8807 up_read(&BTRFS_I(inode)->dio_sem);
8808 current->journal_info = NULL;
8809 if (ret < 0 && ret != -EIOCBQUEUED) {
8810 if (dio_data.reserve)
8811 btrfs_delalloc_release_space(inode, data_reserved,
8812 offset, dio_data.reserve);
8813 /*
8814 * On error we might have left some ordered extents
8815 * without submitting corresponding bios for them, so
8816 * cleanup them up to avoid other tasks getting them
8817 * and waiting for them to complete forever.
8818 */
8819 if (dio_data.unsubmitted_oe_range_start <
8820 dio_data.unsubmitted_oe_range_end)
8821 __endio_write_update_ordered(inode,
8822 dio_data.unsubmitted_oe_range_start,
8823 dio_data.unsubmitted_oe_range_end -
8824 dio_data.unsubmitted_oe_range_start,
8825 false);
8826 } else if (ret >= 0 && (size_t)ret < count)
8827 btrfs_delalloc_release_space(inode, data_reserved,
8828 offset, count - (size_t)ret);
8829 }
8830 out:
8831 if (wakeup)
8832 inode_dio_end(inode);
8833 if (relock)
8834 inode_lock(inode);
8835
8836 extent_changeset_free(data_reserved);
8837 return ret;
8838 }
8839
8840 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8841
8842 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8843 __u64 start, __u64 len)
8844 {
8845 int ret;
8846
8847 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8848 if (ret)
8849 return ret;
8850
8851 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8852 }
8853
8854 int btrfs_readpage(struct file *file, struct page *page)
8855 {
8856 struct extent_io_tree *tree;
8857 tree = &BTRFS_I(page->mapping->host)->io_tree;
8858 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8859 }
8860
8861 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8862 {
8863 struct extent_io_tree *tree;
8864 struct inode *inode = page->mapping->host;
8865 int ret;
8866
8867 if (current->flags & PF_MEMALLOC) {
8868 redirty_page_for_writepage(wbc, page);
8869 unlock_page(page);
8870 return 0;
8871 }
8872
8873 /*
8874 * If we are under memory pressure we will call this directly from the
8875 * VM, we need to make sure we have the inode referenced for the ordered
8876 * extent. If not just return like we didn't do anything.
8877 */
8878 if (!igrab(inode)) {
8879 redirty_page_for_writepage(wbc, page);
8880 return AOP_WRITEPAGE_ACTIVATE;
8881 }
8882 tree = &BTRFS_I(page->mapping->host)->io_tree;
8883 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8884 btrfs_add_delayed_iput(inode);
8885 return ret;
8886 }
8887
8888 static int btrfs_writepages(struct address_space *mapping,
8889 struct writeback_control *wbc)
8890 {
8891 struct extent_io_tree *tree;
8892
8893 tree = &BTRFS_I(mapping->host)->io_tree;
8894 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8895 }
8896
8897 static int
8898 btrfs_readpages(struct file *file, struct address_space *mapping,
8899 struct list_head *pages, unsigned nr_pages)
8900 {
8901 struct extent_io_tree *tree;
8902 tree = &BTRFS_I(mapping->host)->io_tree;
8903 return extent_readpages(tree, mapping, pages, nr_pages,
8904 btrfs_get_extent);
8905 }
8906 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8907 {
8908 struct extent_io_tree *tree;
8909 struct extent_map_tree *map;
8910 int ret;
8911
8912 tree = &BTRFS_I(page->mapping->host)->io_tree;
8913 map = &BTRFS_I(page->mapping->host)->extent_tree;
8914 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8915 if (ret == 1) {
8916 ClearPagePrivate(page);
8917 set_page_private(page, 0);
8918 put_page(page);
8919 }
8920 return ret;
8921 }
8922
8923 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8924 {
8925 if (PageWriteback(page) || PageDirty(page))
8926 return 0;
8927 return __btrfs_releasepage(page, gfp_flags);
8928 }
8929
8930 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8931 unsigned int length)
8932 {
8933 struct inode *inode = page->mapping->host;
8934 struct extent_io_tree *tree;
8935 struct btrfs_ordered_extent *ordered;
8936 struct extent_state *cached_state = NULL;
8937 u64 page_start = page_offset(page);
8938 u64 page_end = page_start + PAGE_SIZE - 1;
8939 u64 start;
8940 u64 end;
8941 int inode_evicting = inode->i_state & I_FREEING;
8942
8943 /*
8944 * we have the page locked, so new writeback can't start,
8945 * and the dirty bit won't be cleared while we are here.
8946 *
8947 * Wait for IO on this page so that we can safely clear
8948 * the PagePrivate2 bit and do ordered accounting
8949 */
8950 wait_on_page_writeback(page);
8951
8952 tree = &BTRFS_I(inode)->io_tree;
8953 if (offset) {
8954 btrfs_releasepage(page, GFP_NOFS);
8955 return;
8956 }
8957
8958 if (!inode_evicting)
8959 lock_extent_bits(tree, page_start, page_end, &cached_state);
8960 again:
8961 start = page_start;
8962 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8963 page_end - start + 1);
8964 if (ordered) {
8965 end = min(page_end, ordered->file_offset + ordered->len - 1);
8966 /*
8967 * IO on this page will never be started, so we need
8968 * to account for any ordered extents now
8969 */
8970 if (!inode_evicting)
8971 clear_extent_bit(tree, start, end,
8972 EXTENT_DIRTY | EXTENT_DELALLOC |
8973 EXTENT_DELALLOC_NEW |
8974 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8975 EXTENT_DEFRAG, 1, 0, &cached_state,
8976 GFP_NOFS);
8977 /*
8978 * whoever cleared the private bit is responsible
8979 * for the finish_ordered_io
8980 */
8981 if (TestClearPagePrivate2(page)) {
8982 struct btrfs_ordered_inode_tree *tree;
8983 u64 new_len;
8984
8985 tree = &BTRFS_I(inode)->ordered_tree;
8986
8987 spin_lock_irq(&tree->lock);
8988 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8989 new_len = start - ordered->file_offset;
8990 if (new_len < ordered->truncated_len)
8991 ordered->truncated_len = new_len;
8992 spin_unlock_irq(&tree->lock);
8993
8994 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8995 start,
8996 end - start + 1, 1))
8997 btrfs_finish_ordered_io(ordered);
8998 }
8999 btrfs_put_ordered_extent(ordered);
9000 if (!inode_evicting) {
9001 cached_state = NULL;
9002 lock_extent_bits(tree, start, end,
9003 &cached_state);
9004 }
9005
9006 start = end + 1;
9007 if (start < page_end)
9008 goto again;
9009 }
9010
9011 /*
9012 * Qgroup reserved space handler
9013 * Page here will be either
9014 * 1) Already written to disk
9015 * In this case, its reserved space is released from data rsv map
9016 * and will be freed by delayed_ref handler finally.
9017 * So even we call qgroup_free_data(), it won't decrease reserved
9018 * space.
9019 * 2) Not written to disk
9020 * This means the reserved space should be freed here. However,
9021 * if a truncate invalidates the page (by clearing PageDirty)
9022 * and the page is accounted for while allocating extent
9023 * in btrfs_check_data_free_space() we let delayed_ref to
9024 * free the entire extent.
9025 */
9026 if (PageDirty(page))
9027 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9028 if (!inode_evicting) {
9029 clear_extent_bit(tree, page_start, page_end,
9030 EXTENT_LOCKED | EXTENT_DIRTY |
9031 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9032 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9033 &cached_state, GFP_NOFS);
9034
9035 __btrfs_releasepage(page, GFP_NOFS);
9036 }
9037
9038 ClearPageChecked(page);
9039 if (PagePrivate(page)) {
9040 ClearPagePrivate(page);
9041 set_page_private(page, 0);
9042 put_page(page);
9043 }
9044 }
9045
9046 /*
9047 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9048 * called from a page fault handler when a page is first dirtied. Hence we must
9049 * be careful to check for EOF conditions here. We set the page up correctly
9050 * for a written page which means we get ENOSPC checking when writing into
9051 * holes and correct delalloc and unwritten extent mapping on filesystems that
9052 * support these features.
9053 *
9054 * We are not allowed to take the i_mutex here so we have to play games to
9055 * protect against truncate races as the page could now be beyond EOF. Because
9056 * vmtruncate() writes the inode size before removing pages, once we have the
9057 * page lock we can determine safely if the page is beyond EOF. If it is not
9058 * beyond EOF, then the page is guaranteed safe against truncation until we
9059 * unlock the page.
9060 */
9061 int btrfs_page_mkwrite(struct vm_fault *vmf)
9062 {
9063 struct page *page = vmf->page;
9064 struct inode *inode = file_inode(vmf->vma->vm_file);
9065 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9066 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9067 struct btrfs_ordered_extent *ordered;
9068 struct extent_state *cached_state = NULL;
9069 struct extent_changeset *data_reserved = NULL;
9070 char *kaddr;
9071 unsigned long zero_start;
9072 loff_t size;
9073 int ret;
9074 int reserved = 0;
9075 u64 reserved_space;
9076 u64 page_start;
9077 u64 page_end;
9078 u64 end;
9079
9080 reserved_space = PAGE_SIZE;
9081
9082 sb_start_pagefault(inode->i_sb);
9083 page_start = page_offset(page);
9084 page_end = page_start + PAGE_SIZE - 1;
9085 end = page_end;
9086
9087 /*
9088 * Reserving delalloc space after obtaining the page lock can lead to
9089 * deadlock. For example, if a dirty page is locked by this function
9090 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9091 * dirty page write out, then the btrfs_writepage() function could
9092 * end up waiting indefinitely to get a lock on the page currently
9093 * being processed by btrfs_page_mkwrite() function.
9094 */
9095 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9096 reserved_space);
9097 if (!ret) {
9098 ret = file_update_time(vmf->vma->vm_file);
9099 reserved = 1;
9100 }
9101 if (ret) {
9102 if (ret == -ENOMEM)
9103 ret = VM_FAULT_OOM;
9104 else /* -ENOSPC, -EIO, etc */
9105 ret = VM_FAULT_SIGBUS;
9106 if (reserved)
9107 goto out;
9108 goto out_noreserve;
9109 }
9110
9111 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9112 again:
9113 lock_page(page);
9114 size = i_size_read(inode);
9115
9116 if ((page->mapping != inode->i_mapping) ||
9117 (page_start >= size)) {
9118 /* page got truncated out from underneath us */
9119 goto out_unlock;
9120 }
9121 wait_on_page_writeback(page);
9122
9123 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9124 set_page_extent_mapped(page);
9125
9126 /*
9127 * we can't set the delalloc bits if there are pending ordered
9128 * extents. Drop our locks and wait for them to finish
9129 */
9130 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9131 PAGE_SIZE);
9132 if (ordered) {
9133 unlock_extent_cached(io_tree, page_start, page_end,
9134 &cached_state, GFP_NOFS);
9135 unlock_page(page);
9136 btrfs_start_ordered_extent(inode, ordered, 1);
9137 btrfs_put_ordered_extent(ordered);
9138 goto again;
9139 }
9140
9141 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9142 reserved_space = round_up(size - page_start,
9143 fs_info->sectorsize);
9144 if (reserved_space < PAGE_SIZE) {
9145 end = page_start + reserved_space - 1;
9146 spin_lock(&BTRFS_I(inode)->lock);
9147 BTRFS_I(inode)->outstanding_extents++;
9148 spin_unlock(&BTRFS_I(inode)->lock);
9149 btrfs_delalloc_release_space(inode, data_reserved,
9150 page_start, PAGE_SIZE - reserved_space);
9151 }
9152 }
9153
9154 /*
9155 * page_mkwrite gets called when the page is firstly dirtied after it's
9156 * faulted in, but write(2) could also dirty a page and set delalloc
9157 * bits, thus in this case for space account reason, we still need to
9158 * clear any delalloc bits within this page range since we have to
9159 * reserve data&meta space before lock_page() (see above comments).
9160 */
9161 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9162 EXTENT_DIRTY | EXTENT_DELALLOC |
9163 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9164 0, 0, &cached_state, GFP_NOFS);
9165
9166 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9167 &cached_state, 0);
9168 if (ret) {
9169 unlock_extent_cached(io_tree, page_start, page_end,
9170 &cached_state, GFP_NOFS);
9171 ret = VM_FAULT_SIGBUS;
9172 goto out_unlock;
9173 }
9174 ret = 0;
9175
9176 /* page is wholly or partially inside EOF */
9177 if (page_start + PAGE_SIZE > size)
9178 zero_start = size & ~PAGE_MASK;
9179 else
9180 zero_start = PAGE_SIZE;
9181
9182 if (zero_start != PAGE_SIZE) {
9183 kaddr = kmap(page);
9184 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9185 flush_dcache_page(page);
9186 kunmap(page);
9187 }
9188 ClearPageChecked(page);
9189 set_page_dirty(page);
9190 SetPageUptodate(page);
9191
9192 BTRFS_I(inode)->last_trans = fs_info->generation;
9193 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9194 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9195
9196 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9197
9198 out_unlock:
9199 if (!ret) {
9200 sb_end_pagefault(inode->i_sb);
9201 extent_changeset_free(data_reserved);
9202 return VM_FAULT_LOCKED;
9203 }
9204 unlock_page(page);
9205 out:
9206 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9207 reserved_space);
9208 out_noreserve:
9209 sb_end_pagefault(inode->i_sb);
9210 extent_changeset_free(data_reserved);
9211 return ret;
9212 }
9213
9214 static int btrfs_truncate(struct inode *inode)
9215 {
9216 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9217 struct btrfs_root *root = BTRFS_I(inode)->root;
9218 struct btrfs_block_rsv *rsv;
9219 int ret = 0;
9220 int err = 0;
9221 struct btrfs_trans_handle *trans;
9222 u64 mask = fs_info->sectorsize - 1;
9223 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9224
9225 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9226 (u64)-1);
9227 if (ret)
9228 return ret;
9229
9230 /*
9231 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9232 * 3 things going on here
9233 *
9234 * 1) We need to reserve space for our orphan item and the space to
9235 * delete our orphan item. Lord knows we don't want to have a dangling
9236 * orphan item because we didn't reserve space to remove it.
9237 *
9238 * 2) We need to reserve space to update our inode.
9239 *
9240 * 3) We need to have something to cache all the space that is going to
9241 * be free'd up by the truncate operation, but also have some slack
9242 * space reserved in case it uses space during the truncate (thank you
9243 * very much snapshotting).
9244 *
9245 * And we need these to all be separate. The fact is we can use a lot of
9246 * space doing the truncate, and we have no earthly idea how much space
9247 * we will use, so we need the truncate reservation to be separate so it
9248 * doesn't end up using space reserved for updating the inode or
9249 * removing the orphan item. We also need to be able to stop the
9250 * transaction and start a new one, which means we need to be able to
9251 * update the inode several times, and we have no idea of knowing how
9252 * many times that will be, so we can't just reserve 1 item for the
9253 * entirety of the operation, so that has to be done separately as well.
9254 * Then there is the orphan item, which does indeed need to be held on
9255 * to for the whole operation, and we need nobody to touch this reserved
9256 * space except the orphan code.
9257 *
9258 * So that leaves us with
9259 *
9260 * 1) root->orphan_block_rsv - for the orphan deletion.
9261 * 2) rsv - for the truncate reservation, which we will steal from the
9262 * transaction reservation.
9263 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9264 * updating the inode.
9265 */
9266 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9267 if (!rsv)
9268 return -ENOMEM;
9269 rsv->size = min_size;
9270 rsv->failfast = 1;
9271
9272 /*
9273 * 1 for the truncate slack space
9274 * 1 for updating the inode.
9275 */
9276 trans = btrfs_start_transaction(root, 2);
9277 if (IS_ERR(trans)) {
9278 err = PTR_ERR(trans);
9279 goto out;
9280 }
9281
9282 /* Migrate the slack space for the truncate to our reserve */
9283 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9284 min_size, 0);
9285 BUG_ON(ret);
9286
9287 /*
9288 * So if we truncate and then write and fsync we normally would just
9289 * write the extents that changed, which is a problem if we need to
9290 * first truncate that entire inode. So set this flag so we write out
9291 * all of the extents in the inode to the sync log so we're completely
9292 * safe.
9293 */
9294 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9295 trans->block_rsv = rsv;
9296
9297 while (1) {
9298 ret = btrfs_truncate_inode_items(trans, root, inode,
9299 inode->i_size,
9300 BTRFS_EXTENT_DATA_KEY);
9301 if (ret != -ENOSPC && ret != -EAGAIN) {
9302 err = ret;
9303 break;
9304 }
9305
9306 trans->block_rsv = &fs_info->trans_block_rsv;
9307 ret = btrfs_update_inode(trans, root, inode);
9308 if (ret) {
9309 err = ret;
9310 break;
9311 }
9312
9313 btrfs_end_transaction(trans);
9314 btrfs_btree_balance_dirty(fs_info);
9315
9316 trans = btrfs_start_transaction(root, 2);
9317 if (IS_ERR(trans)) {
9318 ret = err = PTR_ERR(trans);
9319 trans = NULL;
9320 break;
9321 }
9322
9323 btrfs_block_rsv_release(fs_info, rsv, -1);
9324 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9325 rsv, min_size, 0);
9326 BUG_ON(ret); /* shouldn't happen */
9327 trans->block_rsv = rsv;
9328 }
9329
9330 if (ret == 0 && inode->i_nlink > 0) {
9331 trans->block_rsv = root->orphan_block_rsv;
9332 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9333 if (ret)
9334 err = ret;
9335 }
9336
9337 if (trans) {
9338 trans->block_rsv = &fs_info->trans_block_rsv;
9339 ret = btrfs_update_inode(trans, root, inode);
9340 if (ret && !err)
9341 err = ret;
9342
9343 ret = btrfs_end_transaction(trans);
9344 btrfs_btree_balance_dirty(fs_info);
9345 }
9346 out:
9347 btrfs_free_block_rsv(fs_info, rsv);
9348
9349 if (ret && !err)
9350 err = ret;
9351
9352 return err;
9353 }
9354
9355 /*
9356 * create a new subvolume directory/inode (helper for the ioctl).
9357 */
9358 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9359 struct btrfs_root *new_root,
9360 struct btrfs_root *parent_root,
9361 u64 new_dirid)
9362 {
9363 struct inode *inode;
9364 int err;
9365 u64 index = 0;
9366
9367 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9368 new_dirid, new_dirid,
9369 S_IFDIR | (~current_umask() & S_IRWXUGO),
9370 &index);
9371 if (IS_ERR(inode))
9372 return PTR_ERR(inode);
9373 inode->i_op = &btrfs_dir_inode_operations;
9374 inode->i_fop = &btrfs_dir_file_operations;
9375
9376 set_nlink(inode, 1);
9377 btrfs_i_size_write(BTRFS_I(inode), 0);
9378 unlock_new_inode(inode);
9379
9380 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9381 if (err)
9382 btrfs_err(new_root->fs_info,
9383 "error inheriting subvolume %llu properties: %d",
9384 new_root->root_key.objectid, err);
9385
9386 err = btrfs_update_inode(trans, new_root, inode);
9387
9388 iput(inode);
9389 return err;
9390 }
9391
9392 struct inode *btrfs_alloc_inode(struct super_block *sb)
9393 {
9394 struct btrfs_inode *ei;
9395 struct inode *inode;
9396
9397 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9398 if (!ei)
9399 return NULL;
9400
9401 ei->root = NULL;
9402 ei->generation = 0;
9403 ei->last_trans = 0;
9404 ei->last_sub_trans = 0;
9405 ei->logged_trans = 0;
9406 ei->delalloc_bytes = 0;
9407 ei->new_delalloc_bytes = 0;
9408 ei->defrag_bytes = 0;
9409 ei->disk_i_size = 0;
9410 ei->flags = 0;
9411 ei->csum_bytes = 0;
9412 ei->index_cnt = (u64)-1;
9413 ei->dir_index = 0;
9414 ei->last_unlink_trans = 0;
9415 ei->last_log_commit = 0;
9416 ei->delayed_iput_count = 0;
9417
9418 spin_lock_init(&ei->lock);
9419 ei->outstanding_extents = 0;
9420 ei->reserved_extents = 0;
9421
9422 ei->runtime_flags = 0;
9423 ei->force_compress = BTRFS_COMPRESS_NONE;
9424
9425 ei->delayed_node = NULL;
9426
9427 ei->i_otime.tv_sec = 0;
9428 ei->i_otime.tv_nsec = 0;
9429
9430 inode = &ei->vfs_inode;
9431 extent_map_tree_init(&ei->extent_tree);
9432 extent_io_tree_init(&ei->io_tree, inode);
9433 extent_io_tree_init(&ei->io_failure_tree, inode);
9434 ei->io_tree.track_uptodate = 1;
9435 ei->io_failure_tree.track_uptodate = 1;
9436 atomic_set(&ei->sync_writers, 0);
9437 mutex_init(&ei->log_mutex);
9438 mutex_init(&ei->delalloc_mutex);
9439 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9440 INIT_LIST_HEAD(&ei->delalloc_inodes);
9441 INIT_LIST_HEAD(&ei->delayed_iput);
9442 RB_CLEAR_NODE(&ei->rb_node);
9443 init_rwsem(&ei->dio_sem);
9444
9445 return inode;
9446 }
9447
9448 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9449 void btrfs_test_destroy_inode(struct inode *inode)
9450 {
9451 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9452 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9453 }
9454 #endif
9455
9456 static void btrfs_i_callback(struct rcu_head *head)
9457 {
9458 struct inode *inode = container_of(head, struct inode, i_rcu);
9459 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9460 }
9461
9462 void btrfs_destroy_inode(struct inode *inode)
9463 {
9464 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9465 struct btrfs_ordered_extent *ordered;
9466 struct btrfs_root *root = BTRFS_I(inode)->root;
9467
9468 WARN_ON(!hlist_empty(&inode->i_dentry));
9469 WARN_ON(inode->i_data.nrpages);
9470 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9471 WARN_ON(BTRFS_I(inode)->reserved_extents);
9472 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9473 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9474 WARN_ON(BTRFS_I(inode)->csum_bytes);
9475 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9476
9477 /*
9478 * This can happen where we create an inode, but somebody else also
9479 * created the same inode and we need to destroy the one we already
9480 * created.
9481 */
9482 if (!root)
9483 goto free;
9484
9485 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9486 &BTRFS_I(inode)->runtime_flags)) {
9487 btrfs_info(fs_info, "inode %llu still on the orphan list",
9488 btrfs_ino(BTRFS_I(inode)));
9489 atomic_dec(&root->orphan_inodes);
9490 }
9491
9492 while (1) {
9493 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9494 if (!ordered)
9495 break;
9496 else {
9497 btrfs_err(fs_info,
9498 "found ordered extent %llu %llu on inode cleanup",
9499 ordered->file_offset, ordered->len);
9500 btrfs_remove_ordered_extent(inode, ordered);
9501 btrfs_put_ordered_extent(ordered);
9502 btrfs_put_ordered_extent(ordered);
9503 }
9504 }
9505 btrfs_qgroup_check_reserved_leak(inode);
9506 inode_tree_del(inode);
9507 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9508 free:
9509 call_rcu(&inode->i_rcu, btrfs_i_callback);
9510 }
9511
9512 int btrfs_drop_inode(struct inode *inode)
9513 {
9514 struct btrfs_root *root = BTRFS_I(inode)->root;
9515
9516 if (root == NULL)
9517 return 1;
9518
9519 /* the snap/subvol tree is on deleting */
9520 if (btrfs_root_refs(&root->root_item) == 0)
9521 return 1;
9522 else
9523 return generic_drop_inode(inode);
9524 }
9525
9526 static void init_once(void *foo)
9527 {
9528 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9529
9530 inode_init_once(&ei->vfs_inode);
9531 }
9532
9533 void btrfs_destroy_cachep(void)
9534 {
9535 /*
9536 * Make sure all delayed rcu free inodes are flushed before we
9537 * destroy cache.
9538 */
9539 rcu_barrier();
9540 kmem_cache_destroy(btrfs_inode_cachep);
9541 kmem_cache_destroy(btrfs_trans_handle_cachep);
9542 kmem_cache_destroy(btrfs_path_cachep);
9543 kmem_cache_destroy(btrfs_free_space_cachep);
9544 }
9545
9546 int btrfs_init_cachep(void)
9547 {
9548 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9549 sizeof(struct btrfs_inode), 0,
9550 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9551 init_once);
9552 if (!btrfs_inode_cachep)
9553 goto fail;
9554
9555 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9556 sizeof(struct btrfs_trans_handle), 0,
9557 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9558 if (!btrfs_trans_handle_cachep)
9559 goto fail;
9560
9561 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9562 sizeof(struct btrfs_path), 0,
9563 SLAB_MEM_SPREAD, NULL);
9564 if (!btrfs_path_cachep)
9565 goto fail;
9566
9567 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9568 sizeof(struct btrfs_free_space), 0,
9569 SLAB_MEM_SPREAD, NULL);
9570 if (!btrfs_free_space_cachep)
9571 goto fail;
9572
9573 return 0;
9574 fail:
9575 btrfs_destroy_cachep();
9576 return -ENOMEM;
9577 }
9578
9579 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9580 u32 request_mask, unsigned int flags)
9581 {
9582 u64 delalloc_bytes;
9583 struct inode *inode = d_inode(path->dentry);
9584 u32 blocksize = inode->i_sb->s_blocksize;
9585 u32 bi_flags = BTRFS_I(inode)->flags;
9586
9587 stat->result_mask |= STATX_BTIME;
9588 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9589 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9590 if (bi_flags & BTRFS_INODE_APPEND)
9591 stat->attributes |= STATX_ATTR_APPEND;
9592 if (bi_flags & BTRFS_INODE_COMPRESS)
9593 stat->attributes |= STATX_ATTR_COMPRESSED;
9594 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9595 stat->attributes |= STATX_ATTR_IMMUTABLE;
9596 if (bi_flags & BTRFS_INODE_NODUMP)
9597 stat->attributes |= STATX_ATTR_NODUMP;
9598
9599 stat->attributes_mask |= (STATX_ATTR_APPEND |
9600 STATX_ATTR_COMPRESSED |
9601 STATX_ATTR_IMMUTABLE |
9602 STATX_ATTR_NODUMP);
9603
9604 generic_fillattr(inode, stat);
9605 stat->dev = BTRFS_I(inode)->root->anon_dev;
9606
9607 spin_lock(&BTRFS_I(inode)->lock);
9608 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9609 spin_unlock(&BTRFS_I(inode)->lock);
9610 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9611 ALIGN(delalloc_bytes, blocksize)) >> 9;
9612 return 0;
9613 }
9614
9615 static int btrfs_rename_exchange(struct inode *old_dir,
9616 struct dentry *old_dentry,
9617 struct inode *new_dir,
9618 struct dentry *new_dentry)
9619 {
9620 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9621 struct btrfs_trans_handle *trans;
9622 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9623 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9624 struct inode *new_inode = new_dentry->d_inode;
9625 struct inode *old_inode = old_dentry->d_inode;
9626 struct timespec ctime = current_time(old_inode);
9627 struct dentry *parent;
9628 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9629 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9630 u64 old_idx = 0;
9631 u64 new_idx = 0;
9632 u64 root_objectid;
9633 int ret;
9634 bool root_log_pinned = false;
9635 bool dest_log_pinned = false;
9636
9637 /* we only allow rename subvolume link between subvolumes */
9638 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9639 return -EXDEV;
9640
9641 /* close the race window with snapshot create/destroy ioctl */
9642 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9643 down_read(&fs_info->subvol_sem);
9644 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9645 down_read(&fs_info->subvol_sem);
9646
9647 /*
9648 * We want to reserve the absolute worst case amount of items. So if
9649 * both inodes are subvols and we need to unlink them then that would
9650 * require 4 item modifications, but if they are both normal inodes it
9651 * would require 5 item modifications, so we'll assume their normal
9652 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9653 * should cover the worst case number of items we'll modify.
9654 */
9655 trans = btrfs_start_transaction(root, 12);
9656 if (IS_ERR(trans)) {
9657 ret = PTR_ERR(trans);
9658 goto out_notrans;
9659 }
9660
9661 /*
9662 * We need to find a free sequence number both in the source and
9663 * in the destination directory for the exchange.
9664 */
9665 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9666 if (ret)
9667 goto out_fail;
9668 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9669 if (ret)
9670 goto out_fail;
9671
9672 BTRFS_I(old_inode)->dir_index = 0ULL;
9673 BTRFS_I(new_inode)->dir_index = 0ULL;
9674
9675 /* Reference for the source. */
9676 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9677 /* force full log commit if subvolume involved. */
9678 btrfs_set_log_full_commit(fs_info, trans);
9679 } else {
9680 btrfs_pin_log_trans(root);
9681 root_log_pinned = true;
9682 ret = btrfs_insert_inode_ref(trans, dest,
9683 new_dentry->d_name.name,
9684 new_dentry->d_name.len,
9685 old_ino,
9686 btrfs_ino(BTRFS_I(new_dir)),
9687 old_idx);
9688 if (ret)
9689 goto out_fail;
9690 }
9691
9692 /* And now for the dest. */
9693 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9694 /* force full log commit if subvolume involved. */
9695 btrfs_set_log_full_commit(fs_info, trans);
9696 } else {
9697 btrfs_pin_log_trans(dest);
9698 dest_log_pinned = true;
9699 ret = btrfs_insert_inode_ref(trans, root,
9700 old_dentry->d_name.name,
9701 old_dentry->d_name.len,
9702 new_ino,
9703 btrfs_ino(BTRFS_I(old_dir)),
9704 new_idx);
9705 if (ret)
9706 goto out_fail;
9707 }
9708
9709 /* Update inode version and ctime/mtime. */
9710 inode_inc_iversion(old_dir);
9711 inode_inc_iversion(new_dir);
9712 inode_inc_iversion(old_inode);
9713 inode_inc_iversion(new_inode);
9714 old_dir->i_ctime = old_dir->i_mtime = ctime;
9715 new_dir->i_ctime = new_dir->i_mtime = ctime;
9716 old_inode->i_ctime = ctime;
9717 new_inode->i_ctime = ctime;
9718
9719 if (old_dentry->d_parent != new_dentry->d_parent) {
9720 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9721 BTRFS_I(old_inode), 1);
9722 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9723 BTRFS_I(new_inode), 1);
9724 }
9725
9726 /* src is a subvolume */
9727 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9728 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9729 ret = btrfs_unlink_subvol(trans, root, old_dir,
9730 root_objectid,
9731 old_dentry->d_name.name,
9732 old_dentry->d_name.len);
9733 } else { /* src is an inode */
9734 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9735 BTRFS_I(old_dentry->d_inode),
9736 old_dentry->d_name.name,
9737 old_dentry->d_name.len);
9738 if (!ret)
9739 ret = btrfs_update_inode(trans, root, old_inode);
9740 }
9741 if (ret) {
9742 btrfs_abort_transaction(trans, ret);
9743 goto out_fail;
9744 }
9745
9746 /* dest is a subvolume */
9747 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9748 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9749 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9750 root_objectid,
9751 new_dentry->d_name.name,
9752 new_dentry->d_name.len);
9753 } else { /* dest is an inode */
9754 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9755 BTRFS_I(new_dentry->d_inode),
9756 new_dentry->d_name.name,
9757 new_dentry->d_name.len);
9758 if (!ret)
9759 ret = btrfs_update_inode(trans, dest, new_inode);
9760 }
9761 if (ret) {
9762 btrfs_abort_transaction(trans, ret);
9763 goto out_fail;
9764 }
9765
9766 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9767 new_dentry->d_name.name,
9768 new_dentry->d_name.len, 0, old_idx);
9769 if (ret) {
9770 btrfs_abort_transaction(trans, ret);
9771 goto out_fail;
9772 }
9773
9774 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9775 old_dentry->d_name.name,
9776 old_dentry->d_name.len, 0, new_idx);
9777 if (ret) {
9778 btrfs_abort_transaction(trans, ret);
9779 goto out_fail;
9780 }
9781
9782 if (old_inode->i_nlink == 1)
9783 BTRFS_I(old_inode)->dir_index = old_idx;
9784 if (new_inode->i_nlink == 1)
9785 BTRFS_I(new_inode)->dir_index = new_idx;
9786
9787 if (root_log_pinned) {
9788 parent = new_dentry->d_parent;
9789 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9790 parent);
9791 btrfs_end_log_trans(root);
9792 root_log_pinned = false;
9793 }
9794 if (dest_log_pinned) {
9795 parent = old_dentry->d_parent;
9796 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9797 parent);
9798 btrfs_end_log_trans(dest);
9799 dest_log_pinned = false;
9800 }
9801 out_fail:
9802 /*
9803 * If we have pinned a log and an error happened, we unpin tasks
9804 * trying to sync the log and force them to fallback to a transaction
9805 * commit if the log currently contains any of the inodes involved in
9806 * this rename operation (to ensure we do not persist a log with an
9807 * inconsistent state for any of these inodes or leading to any
9808 * inconsistencies when replayed). If the transaction was aborted, the
9809 * abortion reason is propagated to userspace when attempting to commit
9810 * the transaction. If the log does not contain any of these inodes, we
9811 * allow the tasks to sync it.
9812 */
9813 if (ret && (root_log_pinned || dest_log_pinned)) {
9814 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9815 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9816 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9817 (new_inode &&
9818 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9819 btrfs_set_log_full_commit(fs_info, trans);
9820
9821 if (root_log_pinned) {
9822 btrfs_end_log_trans(root);
9823 root_log_pinned = false;
9824 }
9825 if (dest_log_pinned) {
9826 btrfs_end_log_trans(dest);
9827 dest_log_pinned = false;
9828 }
9829 }
9830 ret = btrfs_end_transaction(trans);
9831 out_notrans:
9832 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9833 up_read(&fs_info->subvol_sem);
9834 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9835 up_read(&fs_info->subvol_sem);
9836
9837 return ret;
9838 }
9839
9840 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9841 struct btrfs_root *root,
9842 struct inode *dir,
9843 struct dentry *dentry)
9844 {
9845 int ret;
9846 struct inode *inode;
9847 u64 objectid;
9848 u64 index;
9849
9850 ret = btrfs_find_free_ino(root, &objectid);
9851 if (ret)
9852 return ret;
9853
9854 inode = btrfs_new_inode(trans, root, dir,
9855 dentry->d_name.name,
9856 dentry->d_name.len,
9857 btrfs_ino(BTRFS_I(dir)),
9858 objectid,
9859 S_IFCHR | WHITEOUT_MODE,
9860 &index);
9861
9862 if (IS_ERR(inode)) {
9863 ret = PTR_ERR(inode);
9864 return ret;
9865 }
9866
9867 inode->i_op = &btrfs_special_inode_operations;
9868 init_special_inode(inode, inode->i_mode,
9869 WHITEOUT_DEV);
9870
9871 ret = btrfs_init_inode_security(trans, inode, dir,
9872 &dentry->d_name);
9873 if (ret)
9874 goto out;
9875
9876 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9877 BTRFS_I(inode), 0, index);
9878 if (ret)
9879 goto out;
9880
9881 ret = btrfs_update_inode(trans, root, inode);
9882 out:
9883 unlock_new_inode(inode);
9884 if (ret)
9885 inode_dec_link_count(inode);
9886 iput(inode);
9887
9888 return ret;
9889 }
9890
9891 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9892 struct inode *new_dir, struct dentry *new_dentry,
9893 unsigned int flags)
9894 {
9895 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9896 struct btrfs_trans_handle *trans;
9897 unsigned int trans_num_items;
9898 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9899 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9900 struct inode *new_inode = d_inode(new_dentry);
9901 struct inode *old_inode = d_inode(old_dentry);
9902 u64 index = 0;
9903 u64 root_objectid;
9904 int ret;
9905 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9906 bool log_pinned = false;
9907
9908 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9909 return -EPERM;
9910
9911 /* we only allow rename subvolume link between subvolumes */
9912 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9913 return -EXDEV;
9914
9915 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9916 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9917 return -ENOTEMPTY;
9918
9919 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9920 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9921 return -ENOTEMPTY;
9922
9923
9924 /* check for collisions, even if the name isn't there */
9925 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9926 new_dentry->d_name.name,
9927 new_dentry->d_name.len);
9928
9929 if (ret) {
9930 if (ret == -EEXIST) {
9931 /* we shouldn't get
9932 * eexist without a new_inode */
9933 if (WARN_ON(!new_inode)) {
9934 return ret;
9935 }
9936 } else {
9937 /* maybe -EOVERFLOW */
9938 return ret;
9939 }
9940 }
9941 ret = 0;
9942
9943 /*
9944 * we're using rename to replace one file with another. Start IO on it
9945 * now so we don't add too much work to the end of the transaction
9946 */
9947 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9948 filemap_flush(old_inode->i_mapping);
9949
9950 /* close the racy window with snapshot create/destroy ioctl */
9951 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9952 down_read(&fs_info->subvol_sem);
9953 /*
9954 * We want to reserve the absolute worst case amount of items. So if
9955 * both inodes are subvols and we need to unlink them then that would
9956 * require 4 item modifications, but if they are both normal inodes it
9957 * would require 5 item modifications, so we'll assume they are normal
9958 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9959 * should cover the worst case number of items we'll modify.
9960 * If our rename has the whiteout flag, we need more 5 units for the
9961 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9962 * when selinux is enabled).
9963 */
9964 trans_num_items = 11;
9965 if (flags & RENAME_WHITEOUT)
9966 trans_num_items += 5;
9967 trans = btrfs_start_transaction(root, trans_num_items);
9968 if (IS_ERR(trans)) {
9969 ret = PTR_ERR(trans);
9970 goto out_notrans;
9971 }
9972
9973 if (dest != root)
9974 btrfs_record_root_in_trans(trans, dest);
9975
9976 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9977 if (ret)
9978 goto out_fail;
9979
9980 BTRFS_I(old_inode)->dir_index = 0ULL;
9981 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9982 /* force full log commit if subvolume involved. */
9983 btrfs_set_log_full_commit(fs_info, trans);
9984 } else {
9985 btrfs_pin_log_trans(root);
9986 log_pinned = true;
9987 ret = btrfs_insert_inode_ref(trans, dest,
9988 new_dentry->d_name.name,
9989 new_dentry->d_name.len,
9990 old_ino,
9991 btrfs_ino(BTRFS_I(new_dir)), index);
9992 if (ret)
9993 goto out_fail;
9994 }
9995
9996 inode_inc_iversion(old_dir);
9997 inode_inc_iversion(new_dir);
9998 inode_inc_iversion(old_inode);
9999 old_dir->i_ctime = old_dir->i_mtime =
10000 new_dir->i_ctime = new_dir->i_mtime =
10001 old_inode->i_ctime = current_time(old_dir);
10002
10003 if (old_dentry->d_parent != new_dentry->d_parent)
10004 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10005 BTRFS_I(old_inode), 1);
10006
10007 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10008 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10009 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10010 old_dentry->d_name.name,
10011 old_dentry->d_name.len);
10012 } else {
10013 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10014 BTRFS_I(d_inode(old_dentry)),
10015 old_dentry->d_name.name,
10016 old_dentry->d_name.len);
10017 if (!ret)
10018 ret = btrfs_update_inode(trans, root, old_inode);
10019 }
10020 if (ret) {
10021 btrfs_abort_transaction(trans, ret);
10022 goto out_fail;
10023 }
10024
10025 if (new_inode) {
10026 inode_inc_iversion(new_inode);
10027 new_inode->i_ctime = current_time(new_inode);
10028 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10029 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10030 root_objectid = BTRFS_I(new_inode)->location.objectid;
10031 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10032 root_objectid,
10033 new_dentry->d_name.name,
10034 new_dentry->d_name.len);
10035 BUG_ON(new_inode->i_nlink == 0);
10036 } else {
10037 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10038 BTRFS_I(d_inode(new_dentry)),
10039 new_dentry->d_name.name,
10040 new_dentry->d_name.len);
10041 }
10042 if (!ret && new_inode->i_nlink == 0)
10043 ret = btrfs_orphan_add(trans,
10044 BTRFS_I(d_inode(new_dentry)));
10045 if (ret) {
10046 btrfs_abort_transaction(trans, ret);
10047 goto out_fail;
10048 }
10049 }
10050
10051 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10052 new_dentry->d_name.name,
10053 new_dentry->d_name.len, 0, index);
10054 if (ret) {
10055 btrfs_abort_transaction(trans, ret);
10056 goto out_fail;
10057 }
10058
10059 if (old_inode->i_nlink == 1)
10060 BTRFS_I(old_inode)->dir_index = index;
10061
10062 if (log_pinned) {
10063 struct dentry *parent = new_dentry->d_parent;
10064
10065 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10066 parent);
10067 btrfs_end_log_trans(root);
10068 log_pinned = false;
10069 }
10070
10071 if (flags & RENAME_WHITEOUT) {
10072 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10073 old_dentry);
10074
10075 if (ret) {
10076 btrfs_abort_transaction(trans, ret);
10077 goto out_fail;
10078 }
10079 }
10080 out_fail:
10081 /*
10082 * If we have pinned the log and an error happened, we unpin tasks
10083 * trying to sync the log and force them to fallback to a transaction
10084 * commit if the log currently contains any of the inodes involved in
10085 * this rename operation (to ensure we do not persist a log with an
10086 * inconsistent state for any of these inodes or leading to any
10087 * inconsistencies when replayed). If the transaction was aborted, the
10088 * abortion reason is propagated to userspace when attempting to commit
10089 * the transaction. If the log does not contain any of these inodes, we
10090 * allow the tasks to sync it.
10091 */
10092 if (ret && log_pinned) {
10093 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10094 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10095 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10096 (new_inode &&
10097 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10098 btrfs_set_log_full_commit(fs_info, trans);
10099
10100 btrfs_end_log_trans(root);
10101 log_pinned = false;
10102 }
10103 btrfs_end_transaction(trans);
10104 out_notrans:
10105 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10106 up_read(&fs_info->subvol_sem);
10107
10108 return ret;
10109 }
10110
10111 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10112 struct inode *new_dir, struct dentry *new_dentry,
10113 unsigned int flags)
10114 {
10115 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10116 return -EINVAL;
10117
10118 if (flags & RENAME_EXCHANGE)
10119 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10120 new_dentry);
10121
10122 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10123 }
10124
10125 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10126 {
10127 struct btrfs_delalloc_work *delalloc_work;
10128 struct inode *inode;
10129
10130 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10131 work);
10132 inode = delalloc_work->inode;
10133 filemap_flush(inode->i_mapping);
10134 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10135 &BTRFS_I(inode)->runtime_flags))
10136 filemap_flush(inode->i_mapping);
10137
10138 if (delalloc_work->delay_iput)
10139 btrfs_add_delayed_iput(inode);
10140 else
10141 iput(inode);
10142 complete(&delalloc_work->completion);
10143 }
10144
10145 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10146 int delay_iput)
10147 {
10148 struct btrfs_delalloc_work *work;
10149
10150 work = kmalloc(sizeof(*work), GFP_NOFS);
10151 if (!work)
10152 return NULL;
10153
10154 init_completion(&work->completion);
10155 INIT_LIST_HEAD(&work->list);
10156 work->inode = inode;
10157 work->delay_iput = delay_iput;
10158 WARN_ON_ONCE(!inode);
10159 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10160 btrfs_run_delalloc_work, NULL, NULL);
10161
10162 return work;
10163 }
10164
10165 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10166 {
10167 wait_for_completion(&work->completion);
10168 kfree(work);
10169 }
10170
10171 /*
10172 * some fairly slow code that needs optimization. This walks the list
10173 * of all the inodes with pending delalloc and forces them to disk.
10174 */
10175 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10176 int nr)
10177 {
10178 struct btrfs_inode *binode;
10179 struct inode *inode;
10180 struct btrfs_delalloc_work *work, *next;
10181 struct list_head works;
10182 struct list_head splice;
10183 int ret = 0;
10184
10185 INIT_LIST_HEAD(&works);
10186 INIT_LIST_HEAD(&splice);
10187
10188 mutex_lock(&root->delalloc_mutex);
10189 spin_lock(&root->delalloc_lock);
10190 list_splice_init(&root->delalloc_inodes, &splice);
10191 while (!list_empty(&splice)) {
10192 binode = list_entry(splice.next, struct btrfs_inode,
10193 delalloc_inodes);
10194
10195 list_move_tail(&binode->delalloc_inodes,
10196 &root->delalloc_inodes);
10197 inode = igrab(&binode->vfs_inode);
10198 if (!inode) {
10199 cond_resched_lock(&root->delalloc_lock);
10200 continue;
10201 }
10202 spin_unlock(&root->delalloc_lock);
10203
10204 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10205 if (!work) {
10206 if (delay_iput)
10207 btrfs_add_delayed_iput(inode);
10208 else
10209 iput(inode);
10210 ret = -ENOMEM;
10211 goto out;
10212 }
10213 list_add_tail(&work->list, &works);
10214 btrfs_queue_work(root->fs_info->flush_workers,
10215 &work->work);
10216 ret++;
10217 if (nr != -1 && ret >= nr)
10218 goto out;
10219 cond_resched();
10220 spin_lock(&root->delalloc_lock);
10221 }
10222 spin_unlock(&root->delalloc_lock);
10223
10224 out:
10225 list_for_each_entry_safe(work, next, &works, list) {
10226 list_del_init(&work->list);
10227 btrfs_wait_and_free_delalloc_work(work);
10228 }
10229
10230 if (!list_empty_careful(&splice)) {
10231 spin_lock(&root->delalloc_lock);
10232 list_splice_tail(&splice, &root->delalloc_inodes);
10233 spin_unlock(&root->delalloc_lock);
10234 }
10235 mutex_unlock(&root->delalloc_mutex);
10236 return ret;
10237 }
10238
10239 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10240 {
10241 struct btrfs_fs_info *fs_info = root->fs_info;
10242 int ret;
10243
10244 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10245 return -EROFS;
10246
10247 ret = __start_delalloc_inodes(root, delay_iput, -1);
10248 if (ret > 0)
10249 ret = 0;
10250 /*
10251 * the filemap_flush will queue IO into the worker threads, but
10252 * we have to make sure the IO is actually started and that
10253 * ordered extents get created before we return
10254 */
10255 atomic_inc(&fs_info->async_submit_draining);
10256 while (atomic_read(&fs_info->nr_async_submits) ||
10257 atomic_read(&fs_info->async_delalloc_pages)) {
10258 wait_event(fs_info->async_submit_wait,
10259 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10260 atomic_read(&fs_info->async_delalloc_pages) == 0));
10261 }
10262 atomic_dec(&fs_info->async_submit_draining);
10263 return ret;
10264 }
10265
10266 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10267 int nr)
10268 {
10269 struct btrfs_root *root;
10270 struct list_head splice;
10271 int ret;
10272
10273 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10274 return -EROFS;
10275
10276 INIT_LIST_HEAD(&splice);
10277
10278 mutex_lock(&fs_info->delalloc_root_mutex);
10279 spin_lock(&fs_info->delalloc_root_lock);
10280 list_splice_init(&fs_info->delalloc_roots, &splice);
10281 while (!list_empty(&splice) && nr) {
10282 root = list_first_entry(&splice, struct btrfs_root,
10283 delalloc_root);
10284 root = btrfs_grab_fs_root(root);
10285 BUG_ON(!root);
10286 list_move_tail(&root->delalloc_root,
10287 &fs_info->delalloc_roots);
10288 spin_unlock(&fs_info->delalloc_root_lock);
10289
10290 ret = __start_delalloc_inodes(root, delay_iput, nr);
10291 btrfs_put_fs_root(root);
10292 if (ret < 0)
10293 goto out;
10294
10295 if (nr != -1) {
10296 nr -= ret;
10297 WARN_ON(nr < 0);
10298 }
10299 spin_lock(&fs_info->delalloc_root_lock);
10300 }
10301 spin_unlock(&fs_info->delalloc_root_lock);
10302
10303 ret = 0;
10304 atomic_inc(&fs_info->async_submit_draining);
10305 while (atomic_read(&fs_info->nr_async_submits) ||
10306 atomic_read(&fs_info->async_delalloc_pages)) {
10307 wait_event(fs_info->async_submit_wait,
10308 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10309 atomic_read(&fs_info->async_delalloc_pages) == 0));
10310 }
10311 atomic_dec(&fs_info->async_submit_draining);
10312 out:
10313 if (!list_empty_careful(&splice)) {
10314 spin_lock(&fs_info->delalloc_root_lock);
10315 list_splice_tail(&splice, &fs_info->delalloc_roots);
10316 spin_unlock(&fs_info->delalloc_root_lock);
10317 }
10318 mutex_unlock(&fs_info->delalloc_root_mutex);
10319 return ret;
10320 }
10321
10322 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10323 const char *symname)
10324 {
10325 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10326 struct btrfs_trans_handle *trans;
10327 struct btrfs_root *root = BTRFS_I(dir)->root;
10328 struct btrfs_path *path;
10329 struct btrfs_key key;
10330 struct inode *inode = NULL;
10331 int err;
10332 int drop_inode = 0;
10333 u64 objectid;
10334 u64 index = 0;
10335 int name_len;
10336 int datasize;
10337 unsigned long ptr;
10338 struct btrfs_file_extent_item *ei;
10339 struct extent_buffer *leaf;
10340
10341 name_len = strlen(symname);
10342 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10343 return -ENAMETOOLONG;
10344
10345 /*
10346 * 2 items for inode item and ref
10347 * 2 items for dir items
10348 * 1 item for updating parent inode item
10349 * 1 item for the inline extent item
10350 * 1 item for xattr if selinux is on
10351 */
10352 trans = btrfs_start_transaction(root, 7);
10353 if (IS_ERR(trans))
10354 return PTR_ERR(trans);
10355
10356 err = btrfs_find_free_ino(root, &objectid);
10357 if (err)
10358 goto out_unlock;
10359
10360 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10361 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10362 objectid, S_IFLNK|S_IRWXUGO, &index);
10363 if (IS_ERR(inode)) {
10364 err = PTR_ERR(inode);
10365 goto out_unlock;
10366 }
10367
10368 /*
10369 * If the active LSM wants to access the inode during
10370 * d_instantiate it needs these. Smack checks to see
10371 * if the filesystem supports xattrs by looking at the
10372 * ops vector.
10373 */
10374 inode->i_fop = &btrfs_file_operations;
10375 inode->i_op = &btrfs_file_inode_operations;
10376 inode->i_mapping->a_ops = &btrfs_aops;
10377 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10378
10379 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10380 if (err)
10381 goto out_unlock_inode;
10382
10383 path = btrfs_alloc_path();
10384 if (!path) {
10385 err = -ENOMEM;
10386 goto out_unlock_inode;
10387 }
10388 key.objectid = btrfs_ino(BTRFS_I(inode));
10389 key.offset = 0;
10390 key.type = BTRFS_EXTENT_DATA_KEY;
10391 datasize = btrfs_file_extent_calc_inline_size(name_len);
10392 err = btrfs_insert_empty_item(trans, root, path, &key,
10393 datasize);
10394 if (err) {
10395 btrfs_free_path(path);
10396 goto out_unlock_inode;
10397 }
10398 leaf = path->nodes[0];
10399 ei = btrfs_item_ptr(leaf, path->slots[0],
10400 struct btrfs_file_extent_item);
10401 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10402 btrfs_set_file_extent_type(leaf, ei,
10403 BTRFS_FILE_EXTENT_INLINE);
10404 btrfs_set_file_extent_encryption(leaf, ei, 0);
10405 btrfs_set_file_extent_compression(leaf, ei, 0);
10406 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10407 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10408
10409 ptr = btrfs_file_extent_inline_start(ei);
10410 write_extent_buffer(leaf, symname, ptr, name_len);
10411 btrfs_mark_buffer_dirty(leaf);
10412 btrfs_free_path(path);
10413
10414 inode->i_op = &btrfs_symlink_inode_operations;
10415 inode_nohighmem(inode);
10416 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10417 inode_set_bytes(inode, name_len);
10418 btrfs_i_size_write(BTRFS_I(inode), name_len);
10419 err = btrfs_update_inode(trans, root, inode);
10420 /*
10421 * Last step, add directory indexes for our symlink inode. This is the
10422 * last step to avoid extra cleanup of these indexes if an error happens
10423 * elsewhere above.
10424 */
10425 if (!err)
10426 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10427 BTRFS_I(inode), 0, index);
10428 if (err) {
10429 drop_inode = 1;
10430 goto out_unlock_inode;
10431 }
10432
10433 unlock_new_inode(inode);
10434 d_instantiate(dentry, inode);
10435
10436 out_unlock:
10437 btrfs_end_transaction(trans);
10438 if (drop_inode) {
10439 inode_dec_link_count(inode);
10440 iput(inode);
10441 }
10442 btrfs_btree_balance_dirty(fs_info);
10443 return err;
10444
10445 out_unlock_inode:
10446 drop_inode = 1;
10447 unlock_new_inode(inode);
10448 goto out_unlock;
10449 }
10450
10451 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10452 u64 start, u64 num_bytes, u64 min_size,
10453 loff_t actual_len, u64 *alloc_hint,
10454 struct btrfs_trans_handle *trans)
10455 {
10456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10457 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10458 struct extent_map *em;
10459 struct btrfs_root *root = BTRFS_I(inode)->root;
10460 struct btrfs_key ins;
10461 u64 cur_offset = start;
10462 u64 i_size;
10463 u64 cur_bytes;
10464 u64 last_alloc = (u64)-1;
10465 int ret = 0;
10466 bool own_trans = true;
10467 u64 end = start + num_bytes - 1;
10468
10469 if (trans)
10470 own_trans = false;
10471 while (num_bytes > 0) {
10472 if (own_trans) {
10473 trans = btrfs_start_transaction(root, 3);
10474 if (IS_ERR(trans)) {
10475 ret = PTR_ERR(trans);
10476 break;
10477 }
10478 }
10479
10480 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10481 cur_bytes = max(cur_bytes, min_size);
10482 /*
10483 * If we are severely fragmented we could end up with really
10484 * small allocations, so if the allocator is returning small
10485 * chunks lets make its job easier by only searching for those
10486 * sized chunks.
10487 */
10488 cur_bytes = min(cur_bytes, last_alloc);
10489 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10490 min_size, 0, *alloc_hint, &ins, 1, 0);
10491 if (ret) {
10492 if (own_trans)
10493 btrfs_end_transaction(trans);
10494 break;
10495 }
10496 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10497
10498 last_alloc = ins.offset;
10499 ret = insert_reserved_file_extent(trans, inode,
10500 cur_offset, ins.objectid,
10501 ins.offset, ins.offset,
10502 ins.offset, 0, 0, 0,
10503 BTRFS_FILE_EXTENT_PREALLOC);
10504 if (ret) {
10505 btrfs_free_reserved_extent(fs_info, ins.objectid,
10506 ins.offset, 0);
10507 btrfs_abort_transaction(trans, ret);
10508 if (own_trans)
10509 btrfs_end_transaction(trans);
10510 break;
10511 }
10512
10513 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10514 cur_offset + ins.offset -1, 0);
10515
10516 em = alloc_extent_map();
10517 if (!em) {
10518 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10519 &BTRFS_I(inode)->runtime_flags);
10520 goto next;
10521 }
10522
10523 em->start = cur_offset;
10524 em->orig_start = cur_offset;
10525 em->len = ins.offset;
10526 em->block_start = ins.objectid;
10527 em->block_len = ins.offset;
10528 em->orig_block_len = ins.offset;
10529 em->ram_bytes = ins.offset;
10530 em->bdev = fs_info->fs_devices->latest_bdev;
10531 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10532 em->generation = trans->transid;
10533
10534 while (1) {
10535 write_lock(&em_tree->lock);
10536 ret = add_extent_mapping(em_tree, em, 1);
10537 write_unlock(&em_tree->lock);
10538 if (ret != -EEXIST)
10539 break;
10540 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10541 cur_offset + ins.offset - 1,
10542 0);
10543 }
10544 free_extent_map(em);
10545 next:
10546 num_bytes -= ins.offset;
10547 cur_offset += ins.offset;
10548 *alloc_hint = ins.objectid + ins.offset;
10549
10550 inode_inc_iversion(inode);
10551 inode->i_ctime = current_time(inode);
10552 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10553 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10554 (actual_len > inode->i_size) &&
10555 (cur_offset > inode->i_size)) {
10556 if (cur_offset > actual_len)
10557 i_size = actual_len;
10558 else
10559 i_size = cur_offset;
10560 i_size_write(inode, i_size);
10561 btrfs_ordered_update_i_size(inode, i_size, NULL);
10562 }
10563
10564 ret = btrfs_update_inode(trans, root, inode);
10565
10566 if (ret) {
10567 btrfs_abort_transaction(trans, ret);
10568 if (own_trans)
10569 btrfs_end_transaction(trans);
10570 break;
10571 }
10572
10573 if (own_trans)
10574 btrfs_end_transaction(trans);
10575 }
10576 if (cur_offset < end)
10577 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10578 end - cur_offset + 1);
10579 return ret;
10580 }
10581
10582 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10583 u64 start, u64 num_bytes, u64 min_size,
10584 loff_t actual_len, u64 *alloc_hint)
10585 {
10586 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10587 min_size, actual_len, alloc_hint,
10588 NULL);
10589 }
10590
10591 int btrfs_prealloc_file_range_trans(struct inode *inode,
10592 struct btrfs_trans_handle *trans, int mode,
10593 u64 start, u64 num_bytes, u64 min_size,
10594 loff_t actual_len, u64 *alloc_hint)
10595 {
10596 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10597 min_size, actual_len, alloc_hint, trans);
10598 }
10599
10600 static int btrfs_set_page_dirty(struct page *page)
10601 {
10602 return __set_page_dirty_nobuffers(page);
10603 }
10604
10605 static int btrfs_permission(struct inode *inode, int mask)
10606 {
10607 struct btrfs_root *root = BTRFS_I(inode)->root;
10608 umode_t mode = inode->i_mode;
10609
10610 if (mask & MAY_WRITE &&
10611 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10612 if (btrfs_root_readonly(root))
10613 return -EROFS;
10614 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10615 return -EACCES;
10616 }
10617 return generic_permission(inode, mask);
10618 }
10619
10620 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10621 {
10622 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10623 struct btrfs_trans_handle *trans;
10624 struct btrfs_root *root = BTRFS_I(dir)->root;
10625 struct inode *inode = NULL;
10626 u64 objectid;
10627 u64 index;
10628 int ret = 0;
10629
10630 /*
10631 * 5 units required for adding orphan entry
10632 */
10633 trans = btrfs_start_transaction(root, 5);
10634 if (IS_ERR(trans))
10635 return PTR_ERR(trans);
10636
10637 ret = btrfs_find_free_ino(root, &objectid);
10638 if (ret)
10639 goto out;
10640
10641 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10642 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10643 if (IS_ERR(inode)) {
10644 ret = PTR_ERR(inode);
10645 inode = NULL;
10646 goto out;
10647 }
10648
10649 inode->i_fop = &btrfs_file_operations;
10650 inode->i_op = &btrfs_file_inode_operations;
10651
10652 inode->i_mapping->a_ops = &btrfs_aops;
10653 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10654
10655 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10656 if (ret)
10657 goto out_inode;
10658
10659 ret = btrfs_update_inode(trans, root, inode);
10660 if (ret)
10661 goto out_inode;
10662 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10663 if (ret)
10664 goto out_inode;
10665
10666 /*
10667 * We set number of links to 0 in btrfs_new_inode(), and here we set
10668 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10669 * through:
10670 *
10671 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10672 */
10673 set_nlink(inode, 1);
10674 unlock_new_inode(inode);
10675 d_tmpfile(dentry, inode);
10676 mark_inode_dirty(inode);
10677
10678 out:
10679 btrfs_end_transaction(trans);
10680 if (ret)
10681 iput(inode);
10682 btrfs_balance_delayed_items(fs_info);
10683 btrfs_btree_balance_dirty(fs_info);
10684 return ret;
10685
10686 out_inode:
10687 unlock_new_inode(inode);
10688 goto out;
10689
10690 }
10691
10692 __attribute__((const))
10693 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10694 {
10695 return -EAGAIN;
10696 }
10697
10698 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10699 {
10700 struct inode *inode = private_data;
10701 return btrfs_sb(inode->i_sb);
10702 }
10703
10704 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10705 u64 start, u64 end)
10706 {
10707 struct inode *inode = private_data;
10708 u64 isize;
10709
10710 isize = i_size_read(inode);
10711 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10712 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10713 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10714 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10715 }
10716 }
10717
10718 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10719 {
10720 struct inode *inode = private_data;
10721 unsigned long index = start >> PAGE_SHIFT;
10722 unsigned long end_index = end >> PAGE_SHIFT;
10723 struct page *page;
10724
10725 while (index <= end_index) {
10726 page = find_get_page(inode->i_mapping, index);
10727 ASSERT(page); /* Pages should be in the extent_io_tree */
10728 set_page_writeback(page);
10729 put_page(page);
10730 index++;
10731 }
10732 }
10733
10734 static const struct inode_operations btrfs_dir_inode_operations = {
10735 .getattr = btrfs_getattr,
10736 .lookup = btrfs_lookup,
10737 .create = btrfs_create,
10738 .unlink = btrfs_unlink,
10739 .link = btrfs_link,
10740 .mkdir = btrfs_mkdir,
10741 .rmdir = btrfs_rmdir,
10742 .rename = btrfs_rename2,
10743 .symlink = btrfs_symlink,
10744 .setattr = btrfs_setattr,
10745 .mknod = btrfs_mknod,
10746 .listxattr = btrfs_listxattr,
10747 .permission = btrfs_permission,
10748 .get_acl = btrfs_get_acl,
10749 .set_acl = btrfs_set_acl,
10750 .update_time = btrfs_update_time,
10751 .tmpfile = btrfs_tmpfile,
10752 };
10753 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10754 .lookup = btrfs_lookup,
10755 .permission = btrfs_permission,
10756 .update_time = btrfs_update_time,
10757 };
10758
10759 static const struct file_operations btrfs_dir_file_operations = {
10760 .llseek = generic_file_llseek,
10761 .read = generic_read_dir,
10762 .iterate_shared = btrfs_real_readdir,
10763 .unlocked_ioctl = btrfs_ioctl,
10764 #ifdef CONFIG_COMPAT
10765 .compat_ioctl = btrfs_compat_ioctl,
10766 #endif
10767 .release = btrfs_release_file,
10768 .fsync = btrfs_sync_file,
10769 };
10770
10771 static const struct extent_io_ops btrfs_extent_io_ops = {
10772 /* mandatory callbacks */
10773 .submit_bio_hook = btrfs_submit_bio_hook,
10774 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10775 .merge_bio_hook = btrfs_merge_bio_hook,
10776 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10777 .tree_fs_info = iotree_fs_info,
10778 .set_range_writeback = btrfs_set_range_writeback,
10779
10780 /* optional callbacks */
10781 .fill_delalloc = run_delalloc_range,
10782 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10783 .writepage_start_hook = btrfs_writepage_start_hook,
10784 .set_bit_hook = btrfs_set_bit_hook,
10785 .clear_bit_hook = btrfs_clear_bit_hook,
10786 .merge_extent_hook = btrfs_merge_extent_hook,
10787 .split_extent_hook = btrfs_split_extent_hook,
10788 .check_extent_io_range = btrfs_check_extent_io_range,
10789 };
10790
10791 /*
10792 * btrfs doesn't support the bmap operation because swapfiles
10793 * use bmap to make a mapping of extents in the file. They assume
10794 * these extents won't change over the life of the file and they
10795 * use the bmap result to do IO directly to the drive.
10796 *
10797 * the btrfs bmap call would return logical addresses that aren't
10798 * suitable for IO and they also will change frequently as COW
10799 * operations happen. So, swapfile + btrfs == corruption.
10800 *
10801 * For now we're avoiding this by dropping bmap.
10802 */
10803 static const struct address_space_operations btrfs_aops = {
10804 .readpage = btrfs_readpage,
10805 .writepage = btrfs_writepage,
10806 .writepages = btrfs_writepages,
10807 .readpages = btrfs_readpages,
10808 .direct_IO = btrfs_direct_IO,
10809 .invalidatepage = btrfs_invalidatepage,
10810 .releasepage = btrfs_releasepage,
10811 .set_page_dirty = btrfs_set_page_dirty,
10812 .error_remove_page = generic_error_remove_page,
10813 };
10814
10815 static const struct address_space_operations btrfs_symlink_aops = {
10816 .readpage = btrfs_readpage,
10817 .writepage = btrfs_writepage,
10818 .invalidatepage = btrfs_invalidatepage,
10819 .releasepage = btrfs_releasepage,
10820 };
10821
10822 static const struct inode_operations btrfs_file_inode_operations = {
10823 .getattr = btrfs_getattr,
10824 .setattr = btrfs_setattr,
10825 .listxattr = btrfs_listxattr,
10826 .permission = btrfs_permission,
10827 .fiemap = btrfs_fiemap,
10828 .get_acl = btrfs_get_acl,
10829 .set_acl = btrfs_set_acl,
10830 .update_time = btrfs_update_time,
10831 };
10832 static const struct inode_operations btrfs_special_inode_operations = {
10833 .getattr = btrfs_getattr,
10834 .setattr = btrfs_setattr,
10835 .permission = btrfs_permission,
10836 .listxattr = btrfs_listxattr,
10837 .get_acl = btrfs_get_acl,
10838 .set_acl = btrfs_set_acl,
10839 .update_time = btrfs_update_time,
10840 };
10841 static const struct inode_operations btrfs_symlink_inode_operations = {
10842 .get_link = page_get_link,
10843 .getattr = btrfs_getattr,
10844 .setattr = btrfs_setattr,
10845 .permission = btrfs_permission,
10846 .listxattr = btrfs_listxattr,
10847 .update_time = btrfs_update_time,
10848 };
10849
10850 const struct dentry_operations btrfs_dentry_operations = {
10851 .d_delete = btrfs_dentry_delete,
10852 .d_release = btrfs_dentry_release,
10853 };
Ubuntu Artful kernel mirror