]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/btrfs/inode.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


d117ab295b4a489faf4d3da596d22eec2a26642a
[mirror_ubuntu-bionic-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 <linux/magic.h>
46 #include <asm/unaligned.h>
47 #include "ctree.h"
48 #include "disk-io.h"
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
53 #include "xattr.h"
54 #include "tree-log.h"
55 #include "volumes.h"
56 #include "compression.h"
57 #include "locking.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
60 #include "backref.h"
61 #include "hash.h"
62 #include "props.h"
63 #include "qgroup.h"
64 #include "dedupe.h"
65
66 struct btrfs_iget_args {
67 struct btrfs_key *location;
68 struct btrfs_root *root;
69 };
70
71 struct btrfs_dio_data {
72 u64 reserve;
73 u64 unsubmitted_oe_range_start;
74 u64 unsubmitted_oe_range_end;
75 int overwrite;
76 };
77
78 static const struct inode_operations btrfs_dir_inode_operations;
79 static const struct inode_operations btrfs_symlink_inode_operations;
80 static const struct inode_operations btrfs_dir_ro_inode_operations;
81 static const struct inode_operations btrfs_special_inode_operations;
82 static const struct inode_operations btrfs_file_inode_operations;
83 static const struct address_space_operations btrfs_aops;
84 static const struct address_space_operations btrfs_symlink_aops;
85 static const struct file_operations btrfs_dir_file_operations;
86 static const struct extent_io_ops btrfs_extent_io_ops;
87
88 static struct kmem_cache *btrfs_inode_cachep;
89 struct kmem_cache *btrfs_trans_handle_cachep;
90 struct kmem_cache *btrfs_path_cachep;
91 struct kmem_cache *btrfs_free_space_cachep;
92
93 #define S_SHIFT 12
94 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
95 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
96 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
97 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
98 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
99 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
100 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
101 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
102 };
103
104 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
105 static int btrfs_truncate(struct inode *inode);
106 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
107 static noinline int cow_file_range(struct inode *inode,
108 struct page *locked_page,
109 u64 start, u64 end, u64 delalloc_end,
110 int *page_started, unsigned long *nr_written,
111 int unlock, struct btrfs_dedupe_hash *hash);
112 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
113 u64 orig_start, u64 block_start,
114 u64 block_len, u64 orig_block_len,
115 u64 ram_bytes, int compress_type,
116 int type);
117
118 static void __endio_write_update_ordered(struct inode *inode,
119 const u64 offset, const u64 bytes,
120 const bool uptodate);
121
122 /*
123 * Cleanup all submitted ordered extents in specified range to handle errors
124 * from the fill_dellaloc() callback.
125 *
126 * NOTE: caller must ensure that when an error happens, it can not call
127 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
128 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
129 * to be released, which we want to happen only when finishing the ordered
130 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
131 * fill_delalloc() callback already does proper cleanup for the first page of
132 * the range, that is, it invokes the callback writepage_end_io_hook() for the
133 * range of the first page.
134 */
135 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
136 const u64 offset,
137 const u64 bytes)
138 {
139 unsigned long index = offset >> PAGE_SHIFT;
140 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
141 struct page *page;
142
143 while (index <= end_index) {
144 page = find_get_page(inode->i_mapping, index);
145 index++;
146 if (!page)
147 continue;
148 ClearPagePrivate2(page);
149 put_page(page);
150 }
151 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
152 bytes - PAGE_SIZE, false);
153 }
154
155 static int btrfs_dirty_inode(struct inode *inode);
156
157 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
158 void btrfs_test_inode_set_ops(struct inode *inode)
159 {
160 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
161 }
162 #endif
163
164 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
165 struct inode *inode, struct inode *dir,
166 const struct qstr *qstr)
167 {
168 int err;
169
170 err = btrfs_init_acl(trans, inode, dir);
171 if (!err)
172 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
173 return err;
174 }
175
176 /*
177 * this does all the hard work for inserting an inline extent into
178 * the btree. The caller should have done a btrfs_drop_extents so that
179 * no overlapping inline items exist in the btree
180 */
181 static int insert_inline_extent(struct btrfs_trans_handle *trans,
182 struct btrfs_path *path, int extent_inserted,
183 struct btrfs_root *root, struct inode *inode,
184 u64 start, size_t size, size_t compressed_size,
185 int compress_type,
186 struct page **compressed_pages)
187 {
188 struct extent_buffer *leaf;
189 struct page *page = NULL;
190 char *kaddr;
191 unsigned long ptr;
192 struct btrfs_file_extent_item *ei;
193 int ret;
194 size_t cur_size = size;
195 unsigned long offset;
196
197 if (compressed_size && compressed_pages)
198 cur_size = compressed_size;
199
200 inode_add_bytes(inode, size);
201
202 if (!extent_inserted) {
203 struct btrfs_key key;
204 size_t datasize;
205
206 key.objectid = btrfs_ino(BTRFS_I(inode));
207 key.offset = start;
208 key.type = BTRFS_EXTENT_DATA_KEY;
209
210 datasize = btrfs_file_extent_calc_inline_size(cur_size);
211 path->leave_spinning = 1;
212 ret = btrfs_insert_empty_item(trans, root, path, &key,
213 datasize);
214 if (ret)
215 goto fail;
216 }
217 leaf = path->nodes[0];
218 ei = btrfs_item_ptr(leaf, path->slots[0],
219 struct btrfs_file_extent_item);
220 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
221 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
222 btrfs_set_file_extent_encryption(leaf, ei, 0);
223 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
224 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
225 ptr = btrfs_file_extent_inline_start(ei);
226
227 if (compress_type != BTRFS_COMPRESS_NONE) {
228 struct page *cpage;
229 int i = 0;
230 while (compressed_size > 0) {
231 cpage = compressed_pages[i];
232 cur_size = min_t(unsigned long, compressed_size,
233 PAGE_SIZE);
234
235 kaddr = kmap_atomic(cpage);
236 write_extent_buffer(leaf, kaddr, ptr, cur_size);
237 kunmap_atomic(kaddr);
238
239 i++;
240 ptr += cur_size;
241 compressed_size -= cur_size;
242 }
243 btrfs_set_file_extent_compression(leaf, ei,
244 compress_type);
245 } else {
246 page = find_get_page(inode->i_mapping,
247 start >> PAGE_SHIFT);
248 btrfs_set_file_extent_compression(leaf, ei, 0);
249 kaddr = kmap_atomic(page);
250 offset = start & (PAGE_SIZE - 1);
251 write_extent_buffer(leaf, kaddr + offset, ptr, size);
252 kunmap_atomic(kaddr);
253 put_page(page);
254 }
255 btrfs_mark_buffer_dirty(leaf);
256 btrfs_release_path(path);
257
258 /*
259 * we're an inline extent, so nobody can
260 * extend the file past i_size without locking
261 * a page we already have locked.
262 *
263 * We must do any isize and inode updates
264 * before we unlock the pages. Otherwise we
265 * could end up racing with unlink.
266 */
267 BTRFS_I(inode)->disk_i_size = inode->i_size;
268 ret = btrfs_update_inode(trans, root, inode);
269
270 fail:
271 return ret;
272 }
273
274
275 /*
276 * conditionally insert an inline extent into the file. This
277 * does the checks required to make sure the data is small enough
278 * to fit as an inline extent.
279 */
280 static noinline int cow_file_range_inline(struct btrfs_root *root,
281 struct inode *inode, u64 start,
282 u64 end, size_t compressed_size,
283 int compress_type,
284 struct page **compressed_pages)
285 {
286 struct btrfs_fs_info *fs_info = root->fs_info;
287 struct btrfs_trans_handle *trans;
288 u64 isize = i_size_read(inode);
289 u64 actual_end = min(end + 1, isize);
290 u64 inline_len = actual_end - start;
291 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
292 u64 data_len = inline_len;
293 int ret;
294 struct btrfs_path *path;
295 int extent_inserted = 0;
296 u32 extent_item_size;
297
298 if (compressed_size)
299 data_len = compressed_size;
300
301 if (start > 0 ||
302 actual_end > fs_info->sectorsize ||
303 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
304 (!compressed_size &&
305 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
306 end + 1 < isize ||
307 data_len > fs_info->max_inline) {
308 return 1;
309 }
310
311 path = btrfs_alloc_path();
312 if (!path)
313 return -ENOMEM;
314
315 trans = btrfs_join_transaction(root);
316 if (IS_ERR(trans)) {
317 btrfs_free_path(path);
318 return PTR_ERR(trans);
319 }
320 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
321
322 if (compressed_size && compressed_pages)
323 extent_item_size = btrfs_file_extent_calc_inline_size(
324 compressed_size);
325 else
326 extent_item_size = btrfs_file_extent_calc_inline_size(
327 inline_len);
328
329 ret = __btrfs_drop_extents(trans, root, inode, path,
330 start, aligned_end, NULL,
331 1, 1, extent_item_size, &extent_inserted);
332 if (ret) {
333 btrfs_abort_transaction(trans, ret);
334 goto out;
335 }
336
337 if (isize > actual_end)
338 inline_len = min_t(u64, isize, actual_end);
339 ret = insert_inline_extent(trans, path, extent_inserted,
340 root, inode, start,
341 inline_len, compressed_size,
342 compress_type, compressed_pages);
343 if (ret && ret != -ENOSPC) {
344 btrfs_abort_transaction(trans, ret);
345 goto out;
346 } else if (ret == -ENOSPC) {
347 ret = 1;
348 goto out;
349 }
350
351 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
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 unsigned int write_flags;
383 struct list_head extents;
384 struct btrfs_work work;
385 };
386
387 static noinline int add_async_extent(struct async_cow *cow,
388 u64 start, u64 ram_size,
389 u64 compressed_size,
390 struct page **pages,
391 unsigned long nr_pages,
392 int compress_type)
393 {
394 struct async_extent *async_extent;
395
396 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
397 BUG_ON(!async_extent); /* -ENOMEM */
398 async_extent->start = start;
399 async_extent->ram_size = ram_size;
400 async_extent->compressed_size = compressed_size;
401 async_extent->pages = pages;
402 async_extent->nr_pages = nr_pages;
403 async_extent->compress_type = compress_type;
404 list_add_tail(&async_extent->list, &cow->extents);
405 return 0;
406 }
407
408 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
409 {
410 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
411
412 /* force compress */
413 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
414 return 1;
415 /* defrag ioctl */
416 if (BTRFS_I(inode)->defrag_compress)
417 return 1;
418 /* bad compression ratios */
419 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
420 return 0;
421 if (btrfs_test_opt(fs_info, COMPRESS) ||
422 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
423 BTRFS_I(inode)->prop_compress)
424 return btrfs_compress_heuristic(inode, start, end);
425 return 0;
426 }
427
428 static inline void inode_should_defrag(struct btrfs_inode *inode,
429 u64 start, u64 end, u64 num_bytes, u64 small_write)
430 {
431 /* If this is a small write inside eof, kick off a defrag */
432 if (num_bytes < small_write &&
433 (start > 0 || end + 1 < inode->disk_i_size))
434 btrfs_add_inode_defrag(NULL, inode);
435 }
436
437 /*
438 * we create compressed extents in two phases. The first
439 * phase compresses a range of pages that have already been
440 * locked (both pages and state bits are locked).
441 *
442 * This is done inside an ordered work queue, and the compression
443 * is spread across many cpus. The actual IO submission is step
444 * two, and the ordered work queue takes care of making sure that
445 * happens in the same order things were put onto the queue by
446 * writepages and friends.
447 *
448 * If this code finds it can't get good compression, it puts an
449 * entry onto the work queue to write the uncompressed bytes. This
450 * makes sure that both compressed inodes and uncompressed inodes
451 * are written in the same order that the flusher thread sent them
452 * down.
453 */
454 static noinline void compress_file_range(struct inode *inode,
455 struct page *locked_page,
456 u64 start, u64 end,
457 struct async_cow *async_cow,
458 int *num_added)
459 {
460 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
461 struct btrfs_root *root = BTRFS_I(inode)->root;
462 u64 blocksize = fs_info->sectorsize;
463 u64 actual_end;
464 u64 isize = i_size_read(inode);
465 int ret = 0;
466 struct page **pages = NULL;
467 unsigned long nr_pages;
468 unsigned long total_compressed = 0;
469 unsigned long total_in = 0;
470 int i;
471 int will_compress;
472 int compress_type = fs_info->compress_type;
473 int redirty = 0;
474
475 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
476 SZ_16K);
477
478 actual_end = min_t(u64, isize, end + 1);
479 again:
480 will_compress = 0;
481 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
482 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
483 nr_pages = min_t(unsigned long, nr_pages,
484 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
485
486 /*
487 * we don't want to send crud past the end of i_size through
488 * compression, that's just a waste of CPU time. So, if the
489 * end of the file is before the start of our current
490 * requested range of bytes, we bail out to the uncompressed
491 * cleanup code that can deal with all of this.
492 *
493 * It isn't really the fastest way to fix things, but this is a
494 * very uncommon corner.
495 */
496 if (actual_end <= start)
497 goto cleanup_and_bail_uncompressed;
498
499 total_compressed = actual_end - start;
500
501 /*
502 * skip compression for a small file range(<=blocksize) that
503 * isn't an inline extent, since it doesn't save disk space at all.
504 */
505 if (total_compressed <= blocksize &&
506 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
507 goto cleanup_and_bail_uncompressed;
508
509 total_compressed = min_t(unsigned long, total_compressed,
510 BTRFS_MAX_UNCOMPRESSED);
511 total_in = 0;
512 ret = 0;
513
514 /*
515 * we do compression for mount -o compress and when the
516 * inode has not been flagged as nocompress. This flag can
517 * change at any time if we discover bad compression ratios.
518 */
519 if (inode_need_compress(inode, start, end)) {
520 WARN_ON(pages);
521 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
522 if (!pages) {
523 /* just bail out to the uncompressed code */
524 goto cont;
525 }
526
527 if (BTRFS_I(inode)->defrag_compress)
528 compress_type = BTRFS_I(inode)->defrag_compress;
529 else if (BTRFS_I(inode)->prop_compress)
530 compress_type = BTRFS_I(inode)->prop_compress;
531
532 /*
533 * we need to call clear_page_dirty_for_io on each
534 * page in the range. Otherwise applications with the file
535 * mmap'd can wander in and change the page contents while
536 * we are compressing them.
537 *
538 * If the compression fails for any reason, we set the pages
539 * dirty again later on.
540 */
541 extent_range_clear_dirty_for_io(inode, start, end);
542 redirty = 1;
543
544 /* Compression level is applied here and only here */
545 ret = btrfs_compress_pages(
546 compress_type | (fs_info->compress_level << 4),
547 inode->i_mapping, start,
548 pages,
549 &nr_pages,
550 &total_in,
551 &total_compressed);
552
553 if (!ret) {
554 unsigned long offset = total_compressed &
555 (PAGE_SIZE - 1);
556 struct page *page = pages[nr_pages - 1];
557 char *kaddr;
558
559 /* zero the tail end of the last page, we might be
560 * sending it down to disk
561 */
562 if (offset) {
563 kaddr = kmap_atomic(page);
564 memset(kaddr + offset, 0,
565 PAGE_SIZE - offset);
566 kunmap_atomic(kaddr);
567 }
568 will_compress = 1;
569 }
570 }
571 cont:
572 if (start == 0) {
573 /* lets try to make an inline extent */
574 if (ret || total_in < actual_end) {
575 /* we didn't compress the entire range, try
576 * to make an uncompressed inline extent.
577 */
578 ret = cow_file_range_inline(root, inode, start, end,
579 0, BTRFS_COMPRESS_NONE, NULL);
580 } else {
581 /* try making a compressed inline extent */
582 ret = cow_file_range_inline(root, inode, start, end,
583 total_compressed,
584 compress_type, pages);
585 }
586 if (ret <= 0) {
587 unsigned long clear_flags = EXTENT_DELALLOC |
588 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
589 EXTENT_DO_ACCOUNTING;
590 unsigned long page_error_op;
591
592 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
593
594 /*
595 * inline extent creation worked or returned error,
596 * we don't need to create any more async work items.
597 * Unlock and free up our temp pages.
598 *
599 * We use DO_ACCOUNTING here because we need the
600 * delalloc_release_metadata to be done _after_ we drop
601 * our outstanding extent for clearing delalloc for this
602 * range.
603 */
604 extent_clear_unlock_delalloc(inode, start, end, end,
605 NULL, clear_flags,
606 PAGE_UNLOCK |
607 PAGE_CLEAR_DIRTY |
608 PAGE_SET_WRITEBACK |
609 page_error_op |
610 PAGE_END_WRITEBACK);
611 goto free_pages_out;
612 }
613 }
614
615 if (will_compress) {
616 /*
617 * we aren't doing an inline extent round the compressed size
618 * up to a block size boundary so the allocator does sane
619 * things
620 */
621 total_compressed = ALIGN(total_compressed, blocksize);
622
623 /*
624 * one last check to make sure the compression is really a
625 * win, compare the page count read with the blocks on disk,
626 * compression must free at least one sector size
627 */
628 total_in = ALIGN(total_in, PAGE_SIZE);
629 if (total_compressed + blocksize <= total_in) {
630 *num_added += 1;
631
632 /*
633 * The async work queues will take care of doing actual
634 * allocation on disk for these compressed pages, and
635 * will submit them to the elevator.
636 */
637 add_async_extent(async_cow, start, total_in,
638 total_compressed, pages, nr_pages,
639 compress_type);
640
641 if (start + total_in < end) {
642 start += total_in;
643 pages = NULL;
644 cond_resched();
645 goto again;
646 }
647 return;
648 }
649 }
650 if (pages) {
651 /*
652 * the compression code ran but failed to make things smaller,
653 * free any pages it allocated and our page pointer array
654 */
655 for (i = 0; i < nr_pages; i++) {
656 WARN_ON(pages[i]->mapping);
657 put_page(pages[i]);
658 }
659 kfree(pages);
660 pages = NULL;
661 total_compressed = 0;
662 nr_pages = 0;
663
664 /* flag the file so we don't compress in the future */
665 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
666 !(BTRFS_I(inode)->prop_compress)) {
667 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
668 }
669 }
670 cleanup_and_bail_uncompressed:
671 /*
672 * No compression, but we still need to write the pages in the file
673 * we've been given so far. redirty the locked page if it corresponds
674 * to our extent and set things up for the async work queue to run
675 * cow_file_range to do the normal delalloc dance.
676 */
677 if (page_offset(locked_page) >= start &&
678 page_offset(locked_page) <= end)
679 __set_page_dirty_nobuffers(locked_page);
680 /* unlocked later on in the async handlers */
681
682 if (redirty)
683 extent_range_redirty_for_io(inode, start, end);
684 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
685 BTRFS_COMPRESS_NONE);
686 *num_added += 1;
687
688 return;
689
690 free_pages_out:
691 for (i = 0; i < nr_pages; i++) {
692 WARN_ON(pages[i]->mapping);
693 put_page(pages[i]);
694 }
695 kfree(pages);
696 }
697
698 static void free_async_extent_pages(struct async_extent *async_extent)
699 {
700 int i;
701
702 if (!async_extent->pages)
703 return;
704
705 for (i = 0; i < async_extent->nr_pages; i++) {
706 WARN_ON(async_extent->pages[i]->mapping);
707 put_page(async_extent->pages[i]);
708 }
709 kfree(async_extent->pages);
710 async_extent->nr_pages = 0;
711 async_extent->pages = NULL;
712 }
713
714 /*
715 * phase two of compressed writeback. This is the ordered portion
716 * of the code, which only gets called in the order the work was
717 * queued. We walk all the async extents created by compress_file_range
718 * and send them down to the disk.
719 */
720 static noinline void submit_compressed_extents(struct inode *inode,
721 struct async_cow *async_cow)
722 {
723 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
724 struct async_extent *async_extent;
725 u64 alloc_hint = 0;
726 struct btrfs_key ins;
727 struct extent_map *em;
728 struct btrfs_root *root = BTRFS_I(inode)->root;
729 struct extent_io_tree *io_tree;
730 int ret = 0;
731
732 again:
733 while (!list_empty(&async_cow->extents)) {
734 async_extent = list_entry(async_cow->extents.next,
735 struct async_extent, list);
736 list_del(&async_extent->list);
737
738 io_tree = &BTRFS_I(inode)->io_tree;
739
740 retry:
741 /* did the compression code fall back to uncompressed IO? */
742 if (!async_extent->pages) {
743 int page_started = 0;
744 unsigned long nr_written = 0;
745
746 lock_extent(io_tree, async_extent->start,
747 async_extent->start +
748 async_extent->ram_size - 1);
749
750 /* allocate blocks */
751 ret = cow_file_range(inode, async_cow->locked_page,
752 async_extent->start,
753 async_extent->start +
754 async_extent->ram_size - 1,
755 async_extent->start +
756 async_extent->ram_size - 1,
757 &page_started, &nr_written, 0,
758 NULL);
759
760 /* JDM XXX */
761
762 /*
763 * if page_started, cow_file_range inserted an
764 * inline extent and took care of all the unlocking
765 * and IO for us. Otherwise, we need to submit
766 * all those pages down to the drive.
767 */
768 if (!page_started && !ret)
769 extent_write_locked_range(io_tree,
770 inode, async_extent->start,
771 async_extent->start +
772 async_extent->ram_size - 1,
773 btrfs_get_extent,
774 WB_SYNC_ALL);
775 else if (ret)
776 unlock_page(async_cow->locked_page);
777 kfree(async_extent);
778 cond_resched();
779 continue;
780 }
781
782 lock_extent(io_tree, async_extent->start,
783 async_extent->start + async_extent->ram_size - 1);
784
785 ret = btrfs_reserve_extent(root, async_extent->ram_size,
786 async_extent->compressed_size,
787 async_extent->compressed_size,
788 0, alloc_hint, &ins, 1, 1);
789 if (ret) {
790 free_async_extent_pages(async_extent);
791
792 if (ret == -ENOSPC) {
793 unlock_extent(io_tree, async_extent->start,
794 async_extent->start +
795 async_extent->ram_size - 1);
796
797 /*
798 * we need to redirty the pages if we decide to
799 * fallback to uncompressed IO, otherwise we
800 * will not submit these pages down to lower
801 * layers.
802 */
803 extent_range_redirty_for_io(inode,
804 async_extent->start,
805 async_extent->start +
806 async_extent->ram_size - 1);
807
808 goto retry;
809 }
810 goto out_free;
811 }
812 /*
813 * here we're doing allocation and writeback of the
814 * compressed pages
815 */
816 em = create_io_em(inode, async_extent->start,
817 async_extent->ram_size, /* len */
818 async_extent->start, /* orig_start */
819 ins.objectid, /* block_start */
820 ins.offset, /* block_len */
821 ins.offset, /* orig_block_len */
822 async_extent->ram_size, /* ram_bytes */
823 async_extent->compress_type,
824 BTRFS_ORDERED_COMPRESSED);
825 if (IS_ERR(em))
826 /* ret value is not necessary due to void function */
827 goto out_free_reserve;
828 free_extent_map(em);
829
830 ret = btrfs_add_ordered_extent_compress(inode,
831 async_extent->start,
832 ins.objectid,
833 async_extent->ram_size,
834 ins.offset,
835 BTRFS_ORDERED_COMPRESSED,
836 async_extent->compress_type);
837 if (ret) {
838 btrfs_drop_extent_cache(BTRFS_I(inode),
839 async_extent->start,
840 async_extent->start +
841 async_extent->ram_size - 1, 0);
842 goto out_free_reserve;
843 }
844 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
845
846 /*
847 * clear dirty, set writeback and unlock the pages.
848 */
849 extent_clear_unlock_delalloc(inode, async_extent->start,
850 async_extent->start +
851 async_extent->ram_size - 1,
852 async_extent->start +
853 async_extent->ram_size - 1,
854 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
855 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
856 PAGE_SET_WRITEBACK);
857 if (btrfs_submit_compressed_write(inode,
858 async_extent->start,
859 async_extent->ram_size,
860 ins.objectid,
861 ins.offset, async_extent->pages,
862 async_extent->nr_pages,
863 async_cow->write_flags)) {
864 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
865 struct page *p = async_extent->pages[0];
866 const u64 start = async_extent->start;
867 const u64 end = start + async_extent->ram_size - 1;
868
869 p->mapping = inode->i_mapping;
870 tree->ops->writepage_end_io_hook(p, start, end,
871 NULL, 0);
872 p->mapping = NULL;
873 extent_clear_unlock_delalloc(inode, start, end, end,
874 NULL, 0,
875 PAGE_END_WRITEBACK |
876 PAGE_SET_ERROR);
877 free_async_extent_pages(async_extent);
878 }
879 alloc_hint = ins.objectid + ins.offset;
880 kfree(async_extent);
881 cond_resched();
882 }
883 return;
884 out_free_reserve:
885 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
886 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
887 out_free:
888 extent_clear_unlock_delalloc(inode, async_extent->start,
889 async_extent->start +
890 async_extent->ram_size - 1,
891 async_extent->start +
892 async_extent->ram_size - 1,
893 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
894 EXTENT_DELALLOC_NEW |
895 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
896 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
897 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
898 PAGE_SET_ERROR);
899 free_async_extent_pages(async_extent);
900 kfree(async_extent);
901 goto again;
902 }
903
904 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
905 u64 num_bytes)
906 {
907 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
908 struct extent_map *em;
909 u64 alloc_hint = 0;
910
911 read_lock(&em_tree->lock);
912 em = search_extent_mapping(em_tree, start, num_bytes);
913 if (em) {
914 /*
915 * if block start isn't an actual block number then find the
916 * first block in this inode and use that as a hint. If that
917 * block is also bogus then just don't worry about it.
918 */
919 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
920 free_extent_map(em);
921 em = search_extent_mapping(em_tree, 0, 0);
922 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
923 alloc_hint = em->block_start;
924 if (em)
925 free_extent_map(em);
926 } else {
927 alloc_hint = em->block_start;
928 free_extent_map(em);
929 }
930 }
931 read_unlock(&em_tree->lock);
932
933 return alloc_hint;
934 }
935
936 /*
937 * when extent_io.c finds a delayed allocation range in the file,
938 * the call backs end up in this code. The basic idea is to
939 * allocate extents on disk for the range, and create ordered data structs
940 * in ram to track those extents.
941 *
942 * locked_page is the page that writepage had locked already. We use
943 * it to make sure we don't do extra locks or unlocks.
944 *
945 * *page_started is set to one if we unlock locked_page and do everything
946 * required to start IO on it. It may be clean and already done with
947 * IO when we return.
948 */
949 static noinline int cow_file_range(struct inode *inode,
950 struct page *locked_page,
951 u64 start, u64 end, u64 delalloc_end,
952 int *page_started, unsigned long *nr_written,
953 int unlock, struct btrfs_dedupe_hash *hash)
954 {
955 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
956 struct btrfs_root *root = BTRFS_I(inode)->root;
957 u64 alloc_hint = 0;
958 u64 num_bytes;
959 unsigned long ram_size;
960 u64 disk_num_bytes;
961 u64 cur_alloc_size = 0;
962 u64 blocksize = fs_info->sectorsize;
963 struct btrfs_key ins;
964 struct extent_map *em;
965 unsigned clear_bits;
966 unsigned long page_ops;
967 bool extent_reserved = false;
968 int ret = 0;
969
970 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
971 WARN_ON_ONCE(1);
972 ret = -EINVAL;
973 goto out_unlock;
974 }
975
976 num_bytes = ALIGN(end - start + 1, blocksize);
977 num_bytes = max(blocksize, num_bytes);
978 disk_num_bytes = num_bytes;
979
980 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
981
982 if (start == 0) {
983 /* lets try to make an inline extent */
984 ret = cow_file_range_inline(root, inode, start, end, 0,
985 BTRFS_COMPRESS_NONE, NULL);
986 if (ret == 0) {
987 /*
988 * We use DO_ACCOUNTING here because we need the
989 * delalloc_release_metadata to be run _after_ we drop
990 * our outstanding extent for clearing delalloc for this
991 * range.
992 */
993 extent_clear_unlock_delalloc(inode, start, end,
994 delalloc_end, NULL,
995 EXTENT_LOCKED | EXTENT_DELALLOC |
996 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
997 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
998 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
999 PAGE_END_WRITEBACK);
1000 *nr_written = *nr_written +
1001 (end - start + PAGE_SIZE) / PAGE_SIZE;
1002 *page_started = 1;
1003 goto out;
1004 } else if (ret < 0) {
1005 goto out_unlock;
1006 }
1007 }
1008
1009 BUG_ON(disk_num_bytes >
1010 btrfs_super_total_bytes(fs_info->super_copy));
1011
1012 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1013 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1014 start + num_bytes - 1, 0);
1015
1016 while (disk_num_bytes > 0) {
1017 cur_alloc_size = disk_num_bytes;
1018 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1019 fs_info->sectorsize, 0, alloc_hint,
1020 &ins, 1, 1);
1021 if (ret < 0)
1022 goto out_unlock;
1023 cur_alloc_size = ins.offset;
1024 extent_reserved = true;
1025
1026 ram_size = ins.offset;
1027 em = create_io_em(inode, start, ins.offset, /* len */
1028 start, /* orig_start */
1029 ins.objectid, /* block_start */
1030 ins.offset, /* block_len */
1031 ins.offset, /* orig_block_len */
1032 ram_size, /* ram_bytes */
1033 BTRFS_COMPRESS_NONE, /* compress_type */
1034 BTRFS_ORDERED_REGULAR /* type */);
1035 if (IS_ERR(em))
1036 goto out_reserve;
1037 free_extent_map(em);
1038
1039 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1040 ram_size, cur_alloc_size, 0);
1041 if (ret)
1042 goto out_drop_extent_cache;
1043
1044 if (root->root_key.objectid ==
1045 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1046 ret = btrfs_reloc_clone_csums(inode, start,
1047 cur_alloc_size);
1048 /*
1049 * Only drop cache here, and process as normal.
1050 *
1051 * We must not allow extent_clear_unlock_delalloc()
1052 * at out_unlock label to free meta of this ordered
1053 * extent, as its meta should be freed by
1054 * btrfs_finish_ordered_io().
1055 *
1056 * So we must continue until @start is increased to
1057 * skip current ordered extent.
1058 */
1059 if (ret)
1060 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1061 start + ram_size - 1, 0);
1062 }
1063
1064 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1065
1066 /* we're not doing compressed IO, don't unlock the first
1067 * page (which the caller expects to stay locked), don't
1068 * clear any dirty bits and don't set any writeback bits
1069 *
1070 * Do set the Private2 bit so we know this page was properly
1071 * setup for writepage
1072 */
1073 page_ops = unlock ? PAGE_UNLOCK : 0;
1074 page_ops |= PAGE_SET_PRIVATE2;
1075
1076 extent_clear_unlock_delalloc(inode, start,
1077 start + ram_size - 1,
1078 delalloc_end, locked_page,
1079 EXTENT_LOCKED | EXTENT_DELALLOC,
1080 page_ops);
1081 if (disk_num_bytes < cur_alloc_size)
1082 disk_num_bytes = 0;
1083 else
1084 disk_num_bytes -= cur_alloc_size;
1085 num_bytes -= cur_alloc_size;
1086 alloc_hint = ins.objectid + ins.offset;
1087 start += cur_alloc_size;
1088 extent_reserved = false;
1089
1090 /*
1091 * btrfs_reloc_clone_csums() error, since start is increased
1092 * extent_clear_unlock_delalloc() at out_unlock label won't
1093 * free metadata of current ordered extent, we're OK to exit.
1094 */
1095 if (ret)
1096 goto out_unlock;
1097 }
1098 out:
1099 return ret;
1100
1101 out_drop_extent_cache:
1102 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1103 out_reserve:
1104 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1105 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1106 out_unlock:
1107 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1108 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1109 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1110 PAGE_END_WRITEBACK;
1111 /*
1112 * If we reserved an extent for our delalloc range (or a subrange) and
1113 * failed to create the respective ordered extent, then it means that
1114 * when we reserved the extent we decremented the extent's size from
1115 * the data space_info's bytes_may_use counter and incremented the
1116 * space_info's bytes_reserved counter by the same amount. We must make
1117 * sure extent_clear_unlock_delalloc() does not try to decrement again
1118 * the data space_info's bytes_may_use counter, therefore we do not pass
1119 * it the flag EXTENT_CLEAR_DATA_RESV.
1120 */
1121 if (extent_reserved) {
1122 extent_clear_unlock_delalloc(inode, start,
1123 start + cur_alloc_size,
1124 start + cur_alloc_size,
1125 locked_page,
1126 clear_bits,
1127 page_ops);
1128 start += cur_alloc_size;
1129 if (start >= end)
1130 goto out;
1131 }
1132 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1133 locked_page,
1134 clear_bits | EXTENT_CLEAR_DATA_RESV,
1135 page_ops);
1136 goto out;
1137 }
1138
1139 /*
1140 * work queue call back to started compression on a file and pages
1141 */
1142 static noinline void async_cow_start(struct btrfs_work *work)
1143 {
1144 struct async_cow *async_cow;
1145 int num_added = 0;
1146 async_cow = container_of(work, struct async_cow, work);
1147
1148 compress_file_range(async_cow->inode, async_cow->locked_page,
1149 async_cow->start, async_cow->end, async_cow,
1150 &num_added);
1151 if (num_added == 0) {
1152 btrfs_add_delayed_iput(async_cow->inode);
1153 async_cow->inode = NULL;
1154 }
1155 }
1156
1157 /*
1158 * work queue call back to submit previously compressed pages
1159 */
1160 static noinline void async_cow_submit(struct btrfs_work *work)
1161 {
1162 struct btrfs_fs_info *fs_info;
1163 struct async_cow *async_cow;
1164 struct btrfs_root *root;
1165 unsigned long nr_pages;
1166
1167 async_cow = container_of(work, struct async_cow, work);
1168
1169 root = async_cow->root;
1170 fs_info = root->fs_info;
1171 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1172 PAGE_SHIFT;
1173
1174 /*
1175 * atomic_sub_return implies a barrier for waitqueue_active
1176 */
1177 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1178 5 * SZ_1M &&
1179 waitqueue_active(&fs_info->async_submit_wait))
1180 wake_up(&fs_info->async_submit_wait);
1181
1182 if (async_cow->inode)
1183 submit_compressed_extents(async_cow->inode, async_cow);
1184 }
1185
1186 static noinline void async_cow_free(struct btrfs_work *work)
1187 {
1188 struct async_cow *async_cow;
1189 async_cow = container_of(work, struct async_cow, work);
1190 if (async_cow->inode)
1191 btrfs_add_delayed_iput(async_cow->inode);
1192 kfree(async_cow);
1193 }
1194
1195 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1196 u64 start, u64 end, int *page_started,
1197 unsigned long *nr_written,
1198 unsigned int write_flags)
1199 {
1200 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1201 struct async_cow *async_cow;
1202 struct btrfs_root *root = BTRFS_I(inode)->root;
1203 unsigned long nr_pages;
1204 u64 cur_end;
1205
1206 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1207 1, 0, NULL, GFP_NOFS);
1208 while (start < end) {
1209 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1210 BUG_ON(!async_cow); /* -ENOMEM */
1211 async_cow->inode = igrab(inode);
1212 async_cow->root = root;
1213 async_cow->locked_page = locked_page;
1214 async_cow->start = start;
1215 async_cow->write_flags = write_flags;
1216
1217 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1218 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1219 cur_end = end;
1220 else
1221 cur_end = min(end, start + SZ_512K - 1);
1222
1223 async_cow->end = cur_end;
1224 INIT_LIST_HEAD(&async_cow->extents);
1225
1226 btrfs_init_work(&async_cow->work,
1227 btrfs_delalloc_helper,
1228 async_cow_start, async_cow_submit,
1229 async_cow_free);
1230
1231 nr_pages = (cur_end - start + PAGE_SIZE) >>
1232 PAGE_SHIFT;
1233 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1234
1235 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1236
1237 *nr_written += nr_pages;
1238 start = cur_end + 1;
1239 }
1240 *page_started = 1;
1241 return 0;
1242 }
1243
1244 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1245 u64 bytenr, u64 num_bytes)
1246 {
1247 int ret;
1248 struct btrfs_ordered_sum *sums;
1249 LIST_HEAD(list);
1250
1251 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1252 bytenr + num_bytes - 1, &list, 0);
1253 if (ret == 0 && list_empty(&list))
1254 return 0;
1255
1256 while (!list_empty(&list)) {
1257 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1258 list_del(&sums->list);
1259 kfree(sums);
1260 }
1261 if (ret < 0)
1262 return ret;
1263 return 1;
1264 }
1265
1266 /*
1267 * when nowcow writeback call back. This checks for snapshots or COW copies
1268 * of the extents that exist in the file, and COWs the file as required.
1269 *
1270 * If no cow copies or snapshots exist, we write directly to the existing
1271 * blocks on disk
1272 */
1273 static noinline int run_delalloc_nocow(struct inode *inode,
1274 struct page *locked_page,
1275 u64 start, u64 end, int *page_started, int force,
1276 unsigned long *nr_written)
1277 {
1278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1279 struct btrfs_root *root = BTRFS_I(inode)->root;
1280 struct extent_buffer *leaf;
1281 struct btrfs_path *path;
1282 struct btrfs_file_extent_item *fi;
1283 struct btrfs_key found_key;
1284 struct extent_map *em;
1285 u64 cow_start;
1286 u64 cur_offset;
1287 u64 extent_end;
1288 u64 extent_offset;
1289 u64 disk_bytenr;
1290 u64 num_bytes;
1291 u64 disk_num_bytes;
1292 u64 ram_bytes;
1293 int extent_type;
1294 int ret, err;
1295 int type;
1296 int nocow;
1297 int check_prev = 1;
1298 bool nolock;
1299 u64 ino = btrfs_ino(BTRFS_I(inode));
1300
1301 path = btrfs_alloc_path();
1302 if (!path) {
1303 extent_clear_unlock_delalloc(inode, start, end, end,
1304 locked_page,
1305 EXTENT_LOCKED | EXTENT_DELALLOC |
1306 EXTENT_DO_ACCOUNTING |
1307 EXTENT_DEFRAG, PAGE_UNLOCK |
1308 PAGE_CLEAR_DIRTY |
1309 PAGE_SET_WRITEBACK |
1310 PAGE_END_WRITEBACK);
1311 return -ENOMEM;
1312 }
1313
1314 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1315
1316 cow_start = (u64)-1;
1317 cur_offset = start;
1318 while (1) {
1319 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1320 cur_offset, 0);
1321 if (ret < 0)
1322 goto error;
1323 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1324 leaf = path->nodes[0];
1325 btrfs_item_key_to_cpu(leaf, &found_key,
1326 path->slots[0] - 1);
1327 if (found_key.objectid == ino &&
1328 found_key.type == BTRFS_EXTENT_DATA_KEY)
1329 path->slots[0]--;
1330 }
1331 check_prev = 0;
1332 next_slot:
1333 leaf = path->nodes[0];
1334 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1335 ret = btrfs_next_leaf(root, path);
1336 if (ret < 0) {
1337 if (cow_start != (u64)-1)
1338 cur_offset = cow_start;
1339 goto error;
1340 }
1341 if (ret > 0)
1342 break;
1343 leaf = path->nodes[0];
1344 }
1345
1346 nocow = 0;
1347 disk_bytenr = 0;
1348 num_bytes = 0;
1349 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1350
1351 if (found_key.objectid > ino)
1352 break;
1353 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1354 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1355 path->slots[0]++;
1356 goto next_slot;
1357 }
1358 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1359 found_key.offset > end)
1360 break;
1361
1362 if (found_key.offset > cur_offset) {
1363 extent_end = found_key.offset;
1364 extent_type = 0;
1365 goto out_check;
1366 }
1367
1368 fi = btrfs_item_ptr(leaf, path->slots[0],
1369 struct btrfs_file_extent_item);
1370 extent_type = btrfs_file_extent_type(leaf, fi);
1371
1372 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1373 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1374 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1375 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1376 extent_offset = btrfs_file_extent_offset(leaf, fi);
1377 extent_end = found_key.offset +
1378 btrfs_file_extent_num_bytes(leaf, fi);
1379 disk_num_bytes =
1380 btrfs_file_extent_disk_num_bytes(leaf, fi);
1381 if (extent_end <= start) {
1382 path->slots[0]++;
1383 goto next_slot;
1384 }
1385 if (disk_bytenr == 0)
1386 goto out_check;
1387 if (btrfs_file_extent_compression(leaf, fi) ||
1388 btrfs_file_extent_encryption(leaf, fi) ||
1389 btrfs_file_extent_other_encoding(leaf, fi))
1390 goto out_check;
1391 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1392 goto out_check;
1393 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1394 goto out_check;
1395 ret = btrfs_cross_ref_exist(root, ino,
1396 found_key.offset -
1397 extent_offset, disk_bytenr);
1398 if (ret) {
1399 /*
1400 * ret could be -EIO if the above fails to read
1401 * metadata.
1402 */
1403 if (ret < 0) {
1404 if (cow_start != (u64)-1)
1405 cur_offset = cow_start;
1406 goto error;
1407 }
1408
1409 WARN_ON_ONCE(nolock);
1410 goto out_check;
1411 }
1412 disk_bytenr += extent_offset;
1413 disk_bytenr += cur_offset - found_key.offset;
1414 num_bytes = min(end + 1, extent_end) - cur_offset;
1415 /*
1416 * if there are pending snapshots for this root,
1417 * we fall into common COW way.
1418 */
1419 if (!nolock) {
1420 err = btrfs_start_write_no_snapshotting(root);
1421 if (!err)
1422 goto out_check;
1423 }
1424 /*
1425 * force cow if csum exists in the range.
1426 * this ensure that csum for a given extent are
1427 * either valid or do not exist.
1428 */
1429 ret = csum_exist_in_range(fs_info, disk_bytenr,
1430 num_bytes);
1431 if (ret) {
1432 if (!nolock)
1433 btrfs_end_write_no_snapshotting(root);
1434
1435 /*
1436 * ret could be -EIO if the above fails to read
1437 * metadata.
1438 */
1439 if (ret < 0) {
1440 if (cow_start != (u64)-1)
1441 cur_offset = cow_start;
1442 goto error;
1443 }
1444 WARN_ON_ONCE(nolock);
1445 goto out_check;
1446 }
1447 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1448 if (!nolock)
1449 btrfs_end_write_no_snapshotting(root);
1450 goto out_check;
1451 }
1452 nocow = 1;
1453 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1454 extent_end = found_key.offset +
1455 btrfs_file_extent_inline_len(leaf,
1456 path->slots[0], fi);
1457 extent_end = ALIGN(extent_end,
1458 fs_info->sectorsize);
1459 } else {
1460 BUG_ON(1);
1461 }
1462 out_check:
1463 if (extent_end <= start) {
1464 path->slots[0]++;
1465 if (!nolock && nocow)
1466 btrfs_end_write_no_snapshotting(root);
1467 if (nocow)
1468 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1469 goto next_slot;
1470 }
1471 if (!nocow) {
1472 if (cow_start == (u64)-1)
1473 cow_start = cur_offset;
1474 cur_offset = extent_end;
1475 if (cur_offset > end)
1476 break;
1477 path->slots[0]++;
1478 goto next_slot;
1479 }
1480
1481 btrfs_release_path(path);
1482 if (cow_start != (u64)-1) {
1483 ret = cow_file_range(inode, locked_page,
1484 cow_start, found_key.offset - 1,
1485 end, page_started, nr_written, 1,
1486 NULL);
1487 if (ret) {
1488 if (!nolock && nocow)
1489 btrfs_end_write_no_snapshotting(root);
1490 if (nocow)
1491 btrfs_dec_nocow_writers(fs_info,
1492 disk_bytenr);
1493 goto error;
1494 }
1495 cow_start = (u64)-1;
1496 }
1497
1498 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1499 u64 orig_start = found_key.offset - extent_offset;
1500
1501 em = create_io_em(inode, cur_offset, num_bytes,
1502 orig_start,
1503 disk_bytenr, /* block_start */
1504 num_bytes, /* block_len */
1505 disk_num_bytes, /* orig_block_len */
1506 ram_bytes, BTRFS_COMPRESS_NONE,
1507 BTRFS_ORDERED_PREALLOC);
1508 if (IS_ERR(em)) {
1509 if (!nolock && nocow)
1510 btrfs_end_write_no_snapshotting(root);
1511 if (nocow)
1512 btrfs_dec_nocow_writers(fs_info,
1513 disk_bytenr);
1514 ret = PTR_ERR(em);
1515 goto error;
1516 }
1517 free_extent_map(em);
1518 }
1519
1520 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1521 type = BTRFS_ORDERED_PREALLOC;
1522 } else {
1523 type = BTRFS_ORDERED_NOCOW;
1524 }
1525
1526 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1527 num_bytes, num_bytes, type);
1528 if (nocow)
1529 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1530 BUG_ON(ret); /* -ENOMEM */
1531
1532 if (root->root_key.objectid ==
1533 BTRFS_DATA_RELOC_TREE_OBJECTID)
1534 /*
1535 * Error handled later, as we must prevent
1536 * extent_clear_unlock_delalloc() in error handler
1537 * from freeing metadata of created ordered extent.
1538 */
1539 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1540 num_bytes);
1541
1542 extent_clear_unlock_delalloc(inode, cur_offset,
1543 cur_offset + num_bytes - 1, end,
1544 locked_page, EXTENT_LOCKED |
1545 EXTENT_DELALLOC |
1546 EXTENT_CLEAR_DATA_RESV,
1547 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1548
1549 if (!nolock && nocow)
1550 btrfs_end_write_no_snapshotting(root);
1551 cur_offset = extent_end;
1552
1553 /*
1554 * btrfs_reloc_clone_csums() error, now we're OK to call error
1555 * handler, as metadata for created ordered extent will only
1556 * be freed by btrfs_finish_ordered_io().
1557 */
1558 if (ret)
1559 goto error;
1560 if (cur_offset > end)
1561 break;
1562 }
1563 btrfs_release_path(path);
1564
1565 if (cur_offset <= end && cow_start == (u64)-1) {
1566 cow_start = cur_offset;
1567 cur_offset = end;
1568 }
1569
1570 if (cow_start != (u64)-1) {
1571 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1572 page_started, nr_written, 1, NULL);
1573 if (ret)
1574 goto error;
1575 }
1576
1577 error:
1578 if (ret && cur_offset < end)
1579 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1580 locked_page, EXTENT_LOCKED |
1581 EXTENT_DELALLOC | EXTENT_DEFRAG |
1582 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1583 PAGE_CLEAR_DIRTY |
1584 PAGE_SET_WRITEBACK |
1585 PAGE_END_WRITEBACK);
1586 btrfs_free_path(path);
1587 return ret;
1588 }
1589
1590 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1591 {
1592
1593 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1594 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1595 return 0;
1596
1597 /*
1598 * @defrag_bytes is a hint value, no spinlock held here,
1599 * if is not zero, it means the file is defragging.
1600 * Force cow if given extent needs to be defragged.
1601 */
1602 if (BTRFS_I(inode)->defrag_bytes &&
1603 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1604 EXTENT_DEFRAG, 0, NULL))
1605 return 1;
1606
1607 return 0;
1608 }
1609
1610 /*
1611 * extent_io.c call back to do delayed allocation processing
1612 */
1613 static int run_delalloc_range(void *private_data, struct page *locked_page,
1614 u64 start, u64 end, int *page_started,
1615 unsigned long *nr_written,
1616 struct writeback_control *wbc)
1617 {
1618 struct inode *inode = private_data;
1619 int ret;
1620 int force_cow = need_force_cow(inode, start, end);
1621 unsigned int write_flags = wbc_to_write_flags(wbc);
1622
1623 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1624 ret = run_delalloc_nocow(inode, locked_page, start, end,
1625 page_started, 1, nr_written);
1626 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1627 ret = run_delalloc_nocow(inode, locked_page, start, end,
1628 page_started, 0, nr_written);
1629 } else if (!inode_need_compress(inode, start, end)) {
1630 ret = cow_file_range(inode, locked_page, start, end, end,
1631 page_started, nr_written, 1, NULL);
1632 } else {
1633 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1634 &BTRFS_I(inode)->runtime_flags);
1635 ret = cow_file_range_async(inode, locked_page, start, end,
1636 page_started, nr_written,
1637 write_flags);
1638 }
1639 if (ret)
1640 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1641 return ret;
1642 }
1643
1644 static void btrfs_split_extent_hook(void *private_data,
1645 struct extent_state *orig, u64 split)
1646 {
1647 struct inode *inode = private_data;
1648 u64 size;
1649
1650 /* not delalloc, ignore it */
1651 if (!(orig->state & EXTENT_DELALLOC))
1652 return;
1653
1654 size = orig->end - orig->start + 1;
1655 if (size > BTRFS_MAX_EXTENT_SIZE) {
1656 u32 num_extents;
1657 u64 new_size;
1658
1659 /*
1660 * See the explanation in btrfs_merge_extent_hook, the same
1661 * applies here, just in reverse.
1662 */
1663 new_size = orig->end - split + 1;
1664 num_extents = count_max_extents(new_size);
1665 new_size = split - orig->start;
1666 num_extents += count_max_extents(new_size);
1667 if (count_max_extents(size) >= num_extents)
1668 return;
1669 }
1670
1671 spin_lock(&BTRFS_I(inode)->lock);
1672 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1673 spin_unlock(&BTRFS_I(inode)->lock);
1674 }
1675
1676 /*
1677 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1678 * extents so we can keep track of new extents that are just merged onto old
1679 * extents, such as when we are doing sequential writes, so we can properly
1680 * account for the metadata space we'll need.
1681 */
1682 static void btrfs_merge_extent_hook(void *private_data,
1683 struct extent_state *new,
1684 struct extent_state *other)
1685 {
1686 struct inode *inode = private_data;
1687 u64 new_size, old_size;
1688 u32 num_extents;
1689
1690 /* not delalloc, ignore it */
1691 if (!(other->state & EXTENT_DELALLOC))
1692 return;
1693
1694 if (new->start > other->start)
1695 new_size = new->end - other->start + 1;
1696 else
1697 new_size = other->end - new->start + 1;
1698
1699 /* we're not bigger than the max, unreserve the space and go */
1700 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1701 spin_lock(&BTRFS_I(inode)->lock);
1702 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1703 spin_unlock(&BTRFS_I(inode)->lock);
1704 return;
1705 }
1706
1707 /*
1708 * We have to add up either side to figure out how many extents were
1709 * accounted for before we merged into one big extent. If the number of
1710 * extents we accounted for is <= the amount we need for the new range
1711 * then we can return, otherwise drop. Think of it like this
1712 *
1713 * [ 4k][MAX_SIZE]
1714 *
1715 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1716 * need 2 outstanding extents, on one side we have 1 and the other side
1717 * we have 1 so they are == and we can return. But in this case
1718 *
1719 * [MAX_SIZE+4k][MAX_SIZE+4k]
1720 *
1721 * Each range on their own accounts for 2 extents, but merged together
1722 * they are only 3 extents worth of accounting, so we need to drop in
1723 * this case.
1724 */
1725 old_size = other->end - other->start + 1;
1726 num_extents = count_max_extents(old_size);
1727 old_size = new->end - new->start + 1;
1728 num_extents += count_max_extents(old_size);
1729 if (count_max_extents(new_size) >= num_extents)
1730 return;
1731
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1734 spin_unlock(&BTRFS_I(inode)->lock);
1735 }
1736
1737 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1738 struct inode *inode)
1739 {
1740 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1741
1742 spin_lock(&root->delalloc_lock);
1743 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1744 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1745 &root->delalloc_inodes);
1746 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1747 &BTRFS_I(inode)->runtime_flags);
1748 root->nr_delalloc_inodes++;
1749 if (root->nr_delalloc_inodes == 1) {
1750 spin_lock(&fs_info->delalloc_root_lock);
1751 BUG_ON(!list_empty(&root->delalloc_root));
1752 list_add_tail(&root->delalloc_root,
1753 &fs_info->delalloc_roots);
1754 spin_unlock(&fs_info->delalloc_root_lock);
1755 }
1756 }
1757 spin_unlock(&root->delalloc_lock);
1758 }
1759
1760
1761 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1762 struct btrfs_inode *inode)
1763 {
1764 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1765
1766 if (!list_empty(&inode->delalloc_inodes)) {
1767 list_del_init(&inode->delalloc_inodes);
1768 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1769 &inode->runtime_flags);
1770 root->nr_delalloc_inodes--;
1771 if (!root->nr_delalloc_inodes) {
1772 spin_lock(&fs_info->delalloc_root_lock);
1773 BUG_ON(list_empty(&root->delalloc_root));
1774 list_del_init(&root->delalloc_root);
1775 spin_unlock(&fs_info->delalloc_root_lock);
1776 }
1777 }
1778 }
1779
1780 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1781 struct btrfs_inode *inode)
1782 {
1783 spin_lock(&root->delalloc_lock);
1784 __btrfs_del_delalloc_inode(root, inode);
1785 spin_unlock(&root->delalloc_lock);
1786 }
1787
1788 /*
1789 * extent_io.c set_bit_hook, used to track delayed allocation
1790 * bytes in this file, and to maintain the list of inodes that
1791 * have pending delalloc work to be done.
1792 */
1793 static void btrfs_set_bit_hook(void *private_data,
1794 struct extent_state *state, unsigned *bits)
1795 {
1796 struct inode *inode = private_data;
1797
1798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1799
1800 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1801 WARN_ON(1);
1802 /*
1803 * set_bit and clear bit hooks normally require _irqsave/restore
1804 * but in this case, we are only testing for the DELALLOC
1805 * bit, which is only set or cleared with irqs on
1806 */
1807 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1808 struct btrfs_root *root = BTRFS_I(inode)->root;
1809 u64 len = state->end + 1 - state->start;
1810 u32 num_extents = count_max_extents(len);
1811 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1812
1813 spin_lock(&BTRFS_I(inode)->lock);
1814 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1815 spin_unlock(&BTRFS_I(inode)->lock);
1816
1817 /* For sanity tests */
1818 if (btrfs_is_testing(fs_info))
1819 return;
1820
1821 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1822 fs_info->delalloc_batch);
1823 spin_lock(&BTRFS_I(inode)->lock);
1824 BTRFS_I(inode)->delalloc_bytes += len;
1825 if (*bits & EXTENT_DEFRAG)
1826 BTRFS_I(inode)->defrag_bytes += len;
1827 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1828 &BTRFS_I(inode)->runtime_flags))
1829 btrfs_add_delalloc_inodes(root, inode);
1830 spin_unlock(&BTRFS_I(inode)->lock);
1831 }
1832
1833 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1834 (*bits & EXTENT_DELALLOC_NEW)) {
1835 spin_lock(&BTRFS_I(inode)->lock);
1836 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1837 state->start;
1838 spin_unlock(&BTRFS_I(inode)->lock);
1839 }
1840 }
1841
1842 /*
1843 * extent_io.c clear_bit_hook, see set_bit_hook for why
1844 */
1845 static void btrfs_clear_bit_hook(void *private_data,
1846 struct extent_state *state,
1847 unsigned *bits)
1848 {
1849 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1850 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1851 u64 len = state->end + 1 - state->start;
1852 u32 num_extents = count_max_extents(len);
1853
1854 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1855 spin_lock(&inode->lock);
1856 inode->defrag_bytes -= len;
1857 spin_unlock(&inode->lock);
1858 }
1859
1860 /*
1861 * set_bit and clear bit hooks normally require _irqsave/restore
1862 * but in this case, we are only testing for the DELALLOC
1863 * bit, which is only set or cleared with irqs on
1864 */
1865 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1866 struct btrfs_root *root = inode->root;
1867 bool do_list = !btrfs_is_free_space_inode(inode);
1868
1869 spin_lock(&inode->lock);
1870 btrfs_mod_outstanding_extents(inode, -num_extents);
1871 spin_unlock(&inode->lock);
1872
1873 /*
1874 * We don't reserve metadata space for space cache inodes so we
1875 * don't need to call dellalloc_release_metadata if there is an
1876 * error.
1877 */
1878 if (*bits & EXTENT_CLEAR_META_RESV &&
1879 root != fs_info->tree_root)
1880 btrfs_delalloc_release_metadata(inode, len);
1881
1882 /* For sanity tests. */
1883 if (btrfs_is_testing(fs_info))
1884 return;
1885
1886 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1887 do_list && !(state->state & EXTENT_NORESERVE) &&
1888 (*bits & EXTENT_CLEAR_DATA_RESV))
1889 btrfs_free_reserved_data_space_noquota(
1890 &inode->vfs_inode,
1891 state->start, len);
1892
1893 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1894 fs_info->delalloc_batch);
1895 spin_lock(&inode->lock);
1896 inode->delalloc_bytes -= len;
1897 if (do_list && inode->delalloc_bytes == 0 &&
1898 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1899 &inode->runtime_flags))
1900 btrfs_del_delalloc_inode(root, inode);
1901 spin_unlock(&inode->lock);
1902 }
1903
1904 if ((state->state & EXTENT_DELALLOC_NEW) &&
1905 (*bits & EXTENT_DELALLOC_NEW)) {
1906 spin_lock(&inode->lock);
1907 ASSERT(inode->new_delalloc_bytes >= len);
1908 inode->new_delalloc_bytes -= len;
1909 spin_unlock(&inode->lock);
1910 }
1911 }
1912
1913 /*
1914 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1915 * we don't create bios that span stripes or chunks
1916 *
1917 * return 1 if page cannot be merged to bio
1918 * return 0 if page can be merged to bio
1919 * return error otherwise
1920 */
1921 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1922 size_t size, struct bio *bio,
1923 unsigned long bio_flags)
1924 {
1925 struct inode *inode = page->mapping->host;
1926 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1927 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1928 u64 length = 0;
1929 u64 map_length;
1930 int ret;
1931
1932 if (bio_flags & EXTENT_BIO_COMPRESSED)
1933 return 0;
1934
1935 length = bio->bi_iter.bi_size;
1936 map_length = length;
1937 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1938 NULL, 0);
1939 if (ret < 0)
1940 return ret;
1941 if (map_length < length + size)
1942 return 1;
1943 return 0;
1944 }
1945
1946 /*
1947 * in order to insert checksums into the metadata in large chunks,
1948 * we wait until bio submission time. All the pages in the bio are
1949 * checksummed and sums are attached onto the ordered extent record.
1950 *
1951 * At IO completion time the cums attached on the ordered extent record
1952 * are inserted into the btree
1953 */
1954 static blk_status_t __btrfs_submit_bio_start(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 blk_status_t ret = 0;
1960
1961 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1962 BUG_ON(ret); /* -ENOMEM */
1963 return 0;
1964 }
1965
1966 /*
1967 * in order to insert checksums into the metadata in large chunks,
1968 * we wait until bio submission time. All the pages in the bio are
1969 * checksummed and sums are attached onto the ordered extent record.
1970 *
1971 * At IO completion time the cums attached on the ordered extent record
1972 * are inserted into the btree
1973 */
1974 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1975 int mirror_num, unsigned long bio_flags,
1976 u64 bio_offset)
1977 {
1978 struct inode *inode = private_data;
1979 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1980 blk_status_t ret;
1981
1982 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1983 if (ret) {
1984 bio->bi_status = ret;
1985 bio_endio(bio);
1986 }
1987 return ret;
1988 }
1989
1990 /*
1991 * extent_io.c submission hook. This does the right thing for csum calculation
1992 * on write, or reading the csums from the tree before a read
1993 */
1994 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1995 int mirror_num, unsigned long bio_flags,
1996 u64 bio_offset)
1997 {
1998 struct inode *inode = private_data;
1999 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2000 struct btrfs_root *root = BTRFS_I(inode)->root;
2001 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2002 blk_status_t ret = 0;
2003 int skip_sum;
2004 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2005
2006 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2007
2008 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2009 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2010
2011 if (bio_op(bio) != REQ_OP_WRITE) {
2012 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2013 if (ret)
2014 goto out;
2015
2016 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2017 ret = btrfs_submit_compressed_read(inode, bio,
2018 mirror_num,
2019 bio_flags);
2020 goto out;
2021 } else if (!skip_sum) {
2022 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2023 if (ret)
2024 goto out;
2025 }
2026 goto mapit;
2027 } else if (async && !skip_sum) {
2028 /* csum items have already been cloned */
2029 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2030 goto mapit;
2031 /* we're doing a write, do the async checksumming */
2032 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2033 bio_offset, inode,
2034 __btrfs_submit_bio_start,
2035 __btrfs_submit_bio_done);
2036 goto out;
2037 } else if (!skip_sum) {
2038 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2039 if (ret)
2040 goto out;
2041 }
2042
2043 mapit:
2044 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2045
2046 out:
2047 if (ret) {
2048 bio->bi_status = ret;
2049 bio_endio(bio);
2050 }
2051 return ret;
2052 }
2053
2054 /*
2055 * given a list of ordered sums record them in the inode. This happens
2056 * at IO completion time based on sums calculated at bio submission time.
2057 */
2058 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2059 struct inode *inode, struct list_head *list)
2060 {
2061 struct btrfs_ordered_sum *sum;
2062
2063 list_for_each_entry(sum, list, list) {
2064 trans->adding_csums = 1;
2065 btrfs_csum_file_blocks(trans,
2066 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2067 trans->adding_csums = 0;
2068 }
2069 return 0;
2070 }
2071
2072 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2073 unsigned int extra_bits,
2074 struct extent_state **cached_state, int dedupe)
2075 {
2076 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2077 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2078 extra_bits, cached_state);
2079 }
2080
2081 /* see btrfs_writepage_start_hook for details on why this is required */
2082 struct btrfs_writepage_fixup {
2083 struct page *page;
2084 struct btrfs_work work;
2085 };
2086
2087 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2088 {
2089 struct btrfs_writepage_fixup *fixup;
2090 struct btrfs_ordered_extent *ordered;
2091 struct extent_state *cached_state = NULL;
2092 struct extent_changeset *data_reserved = NULL;
2093 struct page *page;
2094 struct inode *inode;
2095 u64 page_start;
2096 u64 page_end;
2097 int ret;
2098
2099 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2100 page = fixup->page;
2101 again:
2102 lock_page(page);
2103 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2104 ClearPageChecked(page);
2105 goto out_page;
2106 }
2107
2108 inode = page->mapping->host;
2109 page_start = page_offset(page);
2110 page_end = page_offset(page) + PAGE_SIZE - 1;
2111
2112 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2113 &cached_state);
2114
2115 /* already ordered? We're done */
2116 if (PagePrivate2(page))
2117 goto out;
2118
2119 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2120 PAGE_SIZE);
2121 if (ordered) {
2122 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2123 page_end, &cached_state, GFP_NOFS);
2124 unlock_page(page);
2125 btrfs_start_ordered_extent(inode, ordered, 1);
2126 btrfs_put_ordered_extent(ordered);
2127 goto again;
2128 }
2129
2130 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2131 PAGE_SIZE);
2132 if (ret) {
2133 mapping_set_error(page->mapping, ret);
2134 end_extent_writepage(page, ret, page_start, page_end);
2135 ClearPageChecked(page);
2136 goto out;
2137 }
2138
2139 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2140 &cached_state, 0);
2141 if (ret) {
2142 mapping_set_error(page->mapping, ret);
2143 end_extent_writepage(page, ret, page_start, page_end);
2144 ClearPageChecked(page);
2145 goto out;
2146 }
2147
2148 ClearPageChecked(page);
2149 set_page_dirty(page);
2150 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2151 out:
2152 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2153 &cached_state, GFP_NOFS);
2154 out_page:
2155 unlock_page(page);
2156 put_page(page);
2157 kfree(fixup);
2158 extent_changeset_free(data_reserved);
2159 }
2160
2161 /*
2162 * There are a few paths in the higher layers of the kernel that directly
2163 * set the page dirty bit without asking the filesystem if it is a
2164 * good idea. This causes problems because we want to make sure COW
2165 * properly happens and the data=ordered rules are followed.
2166 *
2167 * In our case any range that doesn't have the ORDERED bit set
2168 * hasn't been properly setup for IO. We kick off an async process
2169 * to fix it up. The async helper will wait for ordered extents, set
2170 * the delalloc bit and make it safe to write the page.
2171 */
2172 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2173 {
2174 struct inode *inode = page->mapping->host;
2175 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2176 struct btrfs_writepage_fixup *fixup;
2177
2178 /* this page is properly in the ordered list */
2179 if (TestClearPagePrivate2(page))
2180 return 0;
2181
2182 if (PageChecked(page))
2183 return -EAGAIN;
2184
2185 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2186 if (!fixup)
2187 return -EAGAIN;
2188
2189 SetPageChecked(page);
2190 get_page(page);
2191 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2192 btrfs_writepage_fixup_worker, NULL, NULL);
2193 fixup->page = page;
2194 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2195 return -EBUSY;
2196 }
2197
2198 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2199 struct inode *inode, u64 file_pos,
2200 u64 disk_bytenr, u64 disk_num_bytes,
2201 u64 num_bytes, u64 ram_bytes,
2202 u8 compression, u8 encryption,
2203 u16 other_encoding, int extent_type)
2204 {
2205 struct btrfs_root *root = BTRFS_I(inode)->root;
2206 struct btrfs_file_extent_item *fi;
2207 struct btrfs_path *path;
2208 struct extent_buffer *leaf;
2209 struct btrfs_key ins;
2210 u64 qg_released;
2211 int extent_inserted = 0;
2212 int ret;
2213
2214 path = btrfs_alloc_path();
2215 if (!path)
2216 return -ENOMEM;
2217
2218 /*
2219 * we may be replacing one extent in the tree with another.
2220 * The new extent is pinned in the extent map, and we don't want
2221 * to drop it from the cache until it is completely in the btree.
2222 *
2223 * So, tell btrfs_drop_extents to leave this extent in the cache.
2224 * the caller is expected to unpin it and allow it to be merged
2225 * with the others.
2226 */
2227 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2228 file_pos + num_bytes, NULL, 0,
2229 1, sizeof(*fi), &extent_inserted);
2230 if (ret)
2231 goto out;
2232
2233 if (!extent_inserted) {
2234 ins.objectid = btrfs_ino(BTRFS_I(inode));
2235 ins.offset = file_pos;
2236 ins.type = BTRFS_EXTENT_DATA_KEY;
2237
2238 path->leave_spinning = 1;
2239 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2240 sizeof(*fi));
2241 if (ret)
2242 goto out;
2243 }
2244 leaf = path->nodes[0];
2245 fi = btrfs_item_ptr(leaf, path->slots[0],
2246 struct btrfs_file_extent_item);
2247 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2248 btrfs_set_file_extent_type(leaf, fi, extent_type);
2249 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2250 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2251 btrfs_set_file_extent_offset(leaf, fi, 0);
2252 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2253 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2254 btrfs_set_file_extent_compression(leaf, fi, compression);
2255 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2256 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2257
2258 btrfs_mark_buffer_dirty(leaf);
2259 btrfs_release_path(path);
2260
2261 inode_add_bytes(inode, num_bytes);
2262
2263 ins.objectid = disk_bytenr;
2264 ins.offset = disk_num_bytes;
2265 ins.type = BTRFS_EXTENT_ITEM_KEY;
2266
2267 /*
2268 * Release the reserved range from inode dirty range map, as it is
2269 * already moved into delayed_ref_head
2270 */
2271 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2272 if (ret < 0)
2273 goto out;
2274 qg_released = ret;
2275 ret = btrfs_alloc_reserved_file_extent(trans, root,
2276 btrfs_ino(BTRFS_I(inode)),
2277 file_pos, qg_released, &ins);
2278 out:
2279 btrfs_free_path(path);
2280
2281 return ret;
2282 }
2283
2284 /* snapshot-aware defrag */
2285 struct sa_defrag_extent_backref {
2286 struct rb_node node;
2287 struct old_sa_defrag_extent *old;
2288 u64 root_id;
2289 u64 inum;
2290 u64 file_pos;
2291 u64 extent_offset;
2292 u64 num_bytes;
2293 u64 generation;
2294 };
2295
2296 struct old_sa_defrag_extent {
2297 struct list_head list;
2298 struct new_sa_defrag_extent *new;
2299
2300 u64 extent_offset;
2301 u64 bytenr;
2302 u64 offset;
2303 u64 len;
2304 int count;
2305 };
2306
2307 struct new_sa_defrag_extent {
2308 struct rb_root root;
2309 struct list_head head;
2310 struct btrfs_path *path;
2311 struct inode *inode;
2312 u64 file_pos;
2313 u64 len;
2314 u64 bytenr;
2315 u64 disk_len;
2316 u8 compress_type;
2317 };
2318
2319 static int backref_comp(struct sa_defrag_extent_backref *b1,
2320 struct sa_defrag_extent_backref *b2)
2321 {
2322 if (b1->root_id < b2->root_id)
2323 return -1;
2324 else if (b1->root_id > b2->root_id)
2325 return 1;
2326
2327 if (b1->inum < b2->inum)
2328 return -1;
2329 else if (b1->inum > b2->inum)
2330 return 1;
2331
2332 if (b1->file_pos < b2->file_pos)
2333 return -1;
2334 else if (b1->file_pos > b2->file_pos)
2335 return 1;
2336
2337 /*
2338 * [------------------------------] ===> (a range of space)
2339 * |<--->| |<---->| =============> (fs/file tree A)
2340 * |<---------------------------->| ===> (fs/file tree B)
2341 *
2342 * A range of space can refer to two file extents in one tree while
2343 * refer to only one file extent in another tree.
2344 *
2345 * So we may process a disk offset more than one time(two extents in A)
2346 * and locate at the same extent(one extent in B), then insert two same
2347 * backrefs(both refer to the extent in B).
2348 */
2349 return 0;
2350 }
2351
2352 static void backref_insert(struct rb_root *root,
2353 struct sa_defrag_extent_backref *backref)
2354 {
2355 struct rb_node **p = &root->rb_node;
2356 struct rb_node *parent = NULL;
2357 struct sa_defrag_extent_backref *entry;
2358 int ret;
2359
2360 while (*p) {
2361 parent = *p;
2362 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2363
2364 ret = backref_comp(backref, entry);
2365 if (ret < 0)
2366 p = &(*p)->rb_left;
2367 else
2368 p = &(*p)->rb_right;
2369 }
2370
2371 rb_link_node(&backref->node, parent, p);
2372 rb_insert_color(&backref->node, root);
2373 }
2374
2375 /*
2376 * Note the backref might has changed, and in this case we just return 0.
2377 */
2378 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2379 void *ctx)
2380 {
2381 struct btrfs_file_extent_item *extent;
2382 struct old_sa_defrag_extent *old = ctx;
2383 struct new_sa_defrag_extent *new = old->new;
2384 struct btrfs_path *path = new->path;
2385 struct btrfs_key key;
2386 struct btrfs_root *root;
2387 struct sa_defrag_extent_backref *backref;
2388 struct extent_buffer *leaf;
2389 struct inode *inode = new->inode;
2390 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2391 int slot;
2392 int ret;
2393 u64 extent_offset;
2394 u64 num_bytes;
2395
2396 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2397 inum == btrfs_ino(BTRFS_I(inode)))
2398 return 0;
2399
2400 key.objectid = root_id;
2401 key.type = BTRFS_ROOT_ITEM_KEY;
2402 key.offset = (u64)-1;
2403
2404 root = btrfs_read_fs_root_no_name(fs_info, &key);
2405 if (IS_ERR(root)) {
2406 if (PTR_ERR(root) == -ENOENT)
2407 return 0;
2408 WARN_ON(1);
2409 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2410 inum, offset, root_id);
2411 return PTR_ERR(root);
2412 }
2413
2414 key.objectid = inum;
2415 key.type = BTRFS_EXTENT_DATA_KEY;
2416 if (offset > (u64)-1 << 32)
2417 key.offset = 0;
2418 else
2419 key.offset = offset;
2420
2421 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2422 if (WARN_ON(ret < 0))
2423 return ret;
2424 ret = 0;
2425
2426 while (1) {
2427 cond_resched();
2428
2429 leaf = path->nodes[0];
2430 slot = path->slots[0];
2431
2432 if (slot >= btrfs_header_nritems(leaf)) {
2433 ret = btrfs_next_leaf(root, path);
2434 if (ret < 0) {
2435 goto out;
2436 } else if (ret > 0) {
2437 ret = 0;
2438 goto out;
2439 }
2440 continue;
2441 }
2442
2443 path->slots[0]++;
2444
2445 btrfs_item_key_to_cpu(leaf, &key, slot);
2446
2447 if (key.objectid > inum)
2448 goto out;
2449
2450 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2451 continue;
2452
2453 extent = btrfs_item_ptr(leaf, slot,
2454 struct btrfs_file_extent_item);
2455
2456 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2457 continue;
2458
2459 /*
2460 * 'offset' refers to the exact key.offset,
2461 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2462 * (key.offset - extent_offset).
2463 */
2464 if (key.offset != offset)
2465 continue;
2466
2467 extent_offset = btrfs_file_extent_offset(leaf, extent);
2468 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2469
2470 if (extent_offset >= old->extent_offset + old->offset +
2471 old->len || extent_offset + num_bytes <=
2472 old->extent_offset + old->offset)
2473 continue;
2474 break;
2475 }
2476
2477 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2478 if (!backref) {
2479 ret = -ENOENT;
2480 goto out;
2481 }
2482
2483 backref->root_id = root_id;
2484 backref->inum = inum;
2485 backref->file_pos = offset;
2486 backref->num_bytes = num_bytes;
2487 backref->extent_offset = extent_offset;
2488 backref->generation = btrfs_file_extent_generation(leaf, extent);
2489 backref->old = old;
2490 backref_insert(&new->root, backref);
2491 old->count++;
2492 out:
2493 btrfs_release_path(path);
2494 WARN_ON(ret);
2495 return ret;
2496 }
2497
2498 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2499 struct new_sa_defrag_extent *new)
2500 {
2501 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2502 struct old_sa_defrag_extent *old, *tmp;
2503 int ret;
2504
2505 new->path = path;
2506
2507 list_for_each_entry_safe(old, tmp, &new->head, list) {
2508 ret = iterate_inodes_from_logical(old->bytenr +
2509 old->extent_offset, fs_info,
2510 path, record_one_backref,
2511 old, false);
2512 if (ret < 0 && ret != -ENOENT)
2513 return false;
2514
2515 /* no backref to be processed for this extent */
2516 if (!old->count) {
2517 list_del(&old->list);
2518 kfree(old);
2519 }
2520 }
2521
2522 if (list_empty(&new->head))
2523 return false;
2524
2525 return true;
2526 }
2527
2528 static int relink_is_mergable(struct extent_buffer *leaf,
2529 struct btrfs_file_extent_item *fi,
2530 struct new_sa_defrag_extent *new)
2531 {
2532 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2533 return 0;
2534
2535 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2536 return 0;
2537
2538 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2539 return 0;
2540
2541 if (btrfs_file_extent_encryption(leaf, fi) ||
2542 btrfs_file_extent_other_encoding(leaf, fi))
2543 return 0;
2544
2545 return 1;
2546 }
2547
2548 /*
2549 * Note the backref might has changed, and in this case we just return 0.
2550 */
2551 static noinline int relink_extent_backref(struct btrfs_path *path,
2552 struct sa_defrag_extent_backref *prev,
2553 struct sa_defrag_extent_backref *backref)
2554 {
2555 struct btrfs_file_extent_item *extent;
2556 struct btrfs_file_extent_item *item;
2557 struct btrfs_ordered_extent *ordered;
2558 struct btrfs_trans_handle *trans;
2559 struct btrfs_root *root;
2560 struct btrfs_key key;
2561 struct extent_buffer *leaf;
2562 struct old_sa_defrag_extent *old = backref->old;
2563 struct new_sa_defrag_extent *new = old->new;
2564 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2565 struct inode *inode;
2566 struct extent_state *cached = NULL;
2567 int ret = 0;
2568 u64 start;
2569 u64 len;
2570 u64 lock_start;
2571 u64 lock_end;
2572 bool merge = false;
2573 int index;
2574
2575 if (prev && prev->root_id == backref->root_id &&
2576 prev->inum == backref->inum &&
2577 prev->file_pos + prev->num_bytes == backref->file_pos)
2578 merge = true;
2579
2580 /* step 1: get root */
2581 key.objectid = backref->root_id;
2582 key.type = BTRFS_ROOT_ITEM_KEY;
2583 key.offset = (u64)-1;
2584
2585 index = srcu_read_lock(&fs_info->subvol_srcu);
2586
2587 root = btrfs_read_fs_root_no_name(fs_info, &key);
2588 if (IS_ERR(root)) {
2589 srcu_read_unlock(&fs_info->subvol_srcu, index);
2590 if (PTR_ERR(root) == -ENOENT)
2591 return 0;
2592 return PTR_ERR(root);
2593 }
2594
2595 if (btrfs_root_readonly(root)) {
2596 srcu_read_unlock(&fs_info->subvol_srcu, index);
2597 return 0;
2598 }
2599
2600 /* step 2: get inode */
2601 key.objectid = backref->inum;
2602 key.type = BTRFS_INODE_ITEM_KEY;
2603 key.offset = 0;
2604
2605 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2606 if (IS_ERR(inode)) {
2607 srcu_read_unlock(&fs_info->subvol_srcu, index);
2608 return 0;
2609 }
2610
2611 srcu_read_unlock(&fs_info->subvol_srcu, index);
2612
2613 /* step 3: relink backref */
2614 lock_start = backref->file_pos;
2615 lock_end = backref->file_pos + backref->num_bytes - 1;
2616 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2617 &cached);
2618
2619 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2620 if (ordered) {
2621 btrfs_put_ordered_extent(ordered);
2622 goto out_unlock;
2623 }
2624
2625 trans = btrfs_join_transaction(root);
2626 if (IS_ERR(trans)) {
2627 ret = PTR_ERR(trans);
2628 goto out_unlock;
2629 }
2630
2631 key.objectid = backref->inum;
2632 key.type = BTRFS_EXTENT_DATA_KEY;
2633 key.offset = backref->file_pos;
2634
2635 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2636 if (ret < 0) {
2637 goto out_free_path;
2638 } else if (ret > 0) {
2639 ret = 0;
2640 goto out_free_path;
2641 }
2642
2643 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2644 struct btrfs_file_extent_item);
2645
2646 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2647 backref->generation)
2648 goto out_free_path;
2649
2650 btrfs_release_path(path);
2651
2652 start = backref->file_pos;
2653 if (backref->extent_offset < old->extent_offset + old->offset)
2654 start += old->extent_offset + old->offset -
2655 backref->extent_offset;
2656
2657 len = min(backref->extent_offset + backref->num_bytes,
2658 old->extent_offset + old->offset + old->len);
2659 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2660
2661 ret = btrfs_drop_extents(trans, root, inode, start,
2662 start + len, 1);
2663 if (ret)
2664 goto out_free_path;
2665 again:
2666 key.objectid = btrfs_ino(BTRFS_I(inode));
2667 key.type = BTRFS_EXTENT_DATA_KEY;
2668 key.offset = start;
2669
2670 path->leave_spinning = 1;
2671 if (merge) {
2672 struct btrfs_file_extent_item *fi;
2673 u64 extent_len;
2674 struct btrfs_key found_key;
2675
2676 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2677 if (ret < 0)
2678 goto out_free_path;
2679
2680 path->slots[0]--;
2681 leaf = path->nodes[0];
2682 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2683
2684 fi = btrfs_item_ptr(leaf, path->slots[0],
2685 struct btrfs_file_extent_item);
2686 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2687
2688 if (extent_len + found_key.offset == start &&
2689 relink_is_mergable(leaf, fi, new)) {
2690 btrfs_set_file_extent_num_bytes(leaf, fi,
2691 extent_len + len);
2692 btrfs_mark_buffer_dirty(leaf);
2693 inode_add_bytes(inode, len);
2694
2695 ret = 1;
2696 goto out_free_path;
2697 } else {
2698 merge = false;
2699 btrfs_release_path(path);
2700 goto again;
2701 }
2702 }
2703
2704 ret = btrfs_insert_empty_item(trans, root, path, &key,
2705 sizeof(*extent));
2706 if (ret) {
2707 btrfs_abort_transaction(trans, ret);
2708 goto out_free_path;
2709 }
2710
2711 leaf = path->nodes[0];
2712 item = btrfs_item_ptr(leaf, path->slots[0],
2713 struct btrfs_file_extent_item);
2714 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2715 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2716 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2717 btrfs_set_file_extent_num_bytes(leaf, item, len);
2718 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2719 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2720 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2721 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2722 btrfs_set_file_extent_encryption(leaf, item, 0);
2723 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2724
2725 btrfs_mark_buffer_dirty(leaf);
2726 inode_add_bytes(inode, len);
2727 btrfs_release_path(path);
2728
2729 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2730 new->disk_len, 0,
2731 backref->root_id, backref->inum,
2732 new->file_pos); /* start - extent_offset */
2733 if (ret) {
2734 btrfs_abort_transaction(trans, ret);
2735 goto out_free_path;
2736 }
2737
2738 ret = 1;
2739 out_free_path:
2740 btrfs_release_path(path);
2741 path->leave_spinning = 0;
2742 btrfs_end_transaction(trans);
2743 out_unlock:
2744 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2745 &cached, GFP_NOFS);
2746 iput(inode);
2747 return ret;
2748 }
2749
2750 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2751 {
2752 struct old_sa_defrag_extent *old, *tmp;
2753
2754 if (!new)
2755 return;
2756
2757 list_for_each_entry_safe(old, tmp, &new->head, list) {
2758 kfree(old);
2759 }
2760 kfree(new);
2761 }
2762
2763 static void relink_file_extents(struct new_sa_defrag_extent *new)
2764 {
2765 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2766 struct btrfs_path *path;
2767 struct sa_defrag_extent_backref *backref;
2768 struct sa_defrag_extent_backref *prev = NULL;
2769 struct inode *inode;
2770 struct btrfs_root *root;
2771 struct rb_node *node;
2772 int ret;
2773
2774 inode = new->inode;
2775 root = BTRFS_I(inode)->root;
2776
2777 path = btrfs_alloc_path();
2778 if (!path)
2779 return;
2780
2781 if (!record_extent_backrefs(path, new)) {
2782 btrfs_free_path(path);
2783 goto out;
2784 }
2785 btrfs_release_path(path);
2786
2787 while (1) {
2788 node = rb_first(&new->root);
2789 if (!node)
2790 break;
2791 rb_erase(node, &new->root);
2792
2793 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2794
2795 ret = relink_extent_backref(path, prev, backref);
2796 WARN_ON(ret < 0);
2797
2798 kfree(prev);
2799
2800 if (ret == 1)
2801 prev = backref;
2802 else
2803 prev = NULL;
2804 cond_resched();
2805 }
2806 kfree(prev);
2807
2808 btrfs_free_path(path);
2809 out:
2810 free_sa_defrag_extent(new);
2811
2812 atomic_dec(&fs_info->defrag_running);
2813 wake_up(&fs_info->transaction_wait);
2814 }
2815
2816 static struct new_sa_defrag_extent *
2817 record_old_file_extents(struct inode *inode,
2818 struct btrfs_ordered_extent *ordered)
2819 {
2820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2821 struct btrfs_root *root = BTRFS_I(inode)->root;
2822 struct btrfs_path *path;
2823 struct btrfs_key key;
2824 struct old_sa_defrag_extent *old;
2825 struct new_sa_defrag_extent *new;
2826 int ret;
2827
2828 new = kmalloc(sizeof(*new), GFP_NOFS);
2829 if (!new)
2830 return NULL;
2831
2832 new->inode = inode;
2833 new->file_pos = ordered->file_offset;
2834 new->len = ordered->len;
2835 new->bytenr = ordered->start;
2836 new->disk_len = ordered->disk_len;
2837 new->compress_type = ordered->compress_type;
2838 new->root = RB_ROOT;
2839 INIT_LIST_HEAD(&new->head);
2840
2841 path = btrfs_alloc_path();
2842 if (!path)
2843 goto out_kfree;
2844
2845 key.objectid = btrfs_ino(BTRFS_I(inode));
2846 key.type = BTRFS_EXTENT_DATA_KEY;
2847 key.offset = new->file_pos;
2848
2849 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2850 if (ret < 0)
2851 goto out_free_path;
2852 if (ret > 0 && path->slots[0] > 0)
2853 path->slots[0]--;
2854
2855 /* find out all the old extents for the file range */
2856 while (1) {
2857 struct btrfs_file_extent_item *extent;
2858 struct extent_buffer *l;
2859 int slot;
2860 u64 num_bytes;
2861 u64 offset;
2862 u64 end;
2863 u64 disk_bytenr;
2864 u64 extent_offset;
2865
2866 l = path->nodes[0];
2867 slot = path->slots[0];
2868
2869 if (slot >= btrfs_header_nritems(l)) {
2870 ret = btrfs_next_leaf(root, path);
2871 if (ret < 0)
2872 goto out_free_path;
2873 else if (ret > 0)
2874 break;
2875 continue;
2876 }
2877
2878 btrfs_item_key_to_cpu(l, &key, slot);
2879
2880 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2881 break;
2882 if (key.type != BTRFS_EXTENT_DATA_KEY)
2883 break;
2884 if (key.offset >= new->file_pos + new->len)
2885 break;
2886
2887 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2888
2889 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2890 if (key.offset + num_bytes < new->file_pos)
2891 goto next;
2892
2893 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2894 if (!disk_bytenr)
2895 goto next;
2896
2897 extent_offset = btrfs_file_extent_offset(l, extent);
2898
2899 old = kmalloc(sizeof(*old), GFP_NOFS);
2900 if (!old)
2901 goto out_free_path;
2902
2903 offset = max(new->file_pos, key.offset);
2904 end = min(new->file_pos + new->len, key.offset + num_bytes);
2905
2906 old->bytenr = disk_bytenr;
2907 old->extent_offset = extent_offset;
2908 old->offset = offset - key.offset;
2909 old->len = end - offset;
2910 old->new = new;
2911 old->count = 0;
2912 list_add_tail(&old->list, &new->head);
2913 next:
2914 path->slots[0]++;
2915 cond_resched();
2916 }
2917
2918 btrfs_free_path(path);
2919 atomic_inc(&fs_info->defrag_running);
2920
2921 return new;
2922
2923 out_free_path:
2924 btrfs_free_path(path);
2925 out_kfree:
2926 free_sa_defrag_extent(new);
2927 return NULL;
2928 }
2929
2930 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2931 u64 start, u64 len)
2932 {
2933 struct btrfs_block_group_cache *cache;
2934
2935 cache = btrfs_lookup_block_group(fs_info, start);
2936 ASSERT(cache);
2937
2938 spin_lock(&cache->lock);
2939 cache->delalloc_bytes -= len;
2940 spin_unlock(&cache->lock);
2941
2942 btrfs_put_block_group(cache);
2943 }
2944
2945 /* as ordered data IO finishes, this gets called so we can finish
2946 * an ordered extent if the range of bytes in the file it covers are
2947 * fully written.
2948 */
2949 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2950 {
2951 struct inode *inode = ordered_extent->inode;
2952 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2953 struct btrfs_root *root = BTRFS_I(inode)->root;
2954 struct btrfs_trans_handle *trans = NULL;
2955 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2956 struct extent_state *cached_state = NULL;
2957 struct new_sa_defrag_extent *new = NULL;
2958 int compress_type = 0;
2959 int ret = 0;
2960 u64 logical_len = ordered_extent->len;
2961 bool nolock;
2962 bool truncated = false;
2963 bool range_locked = false;
2964 bool clear_new_delalloc_bytes = false;
2965
2966 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2967 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2968 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2969 clear_new_delalloc_bytes = true;
2970
2971 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2972
2973 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2974 ret = -EIO;
2975 goto out;
2976 }
2977
2978 btrfs_free_io_failure_record(BTRFS_I(inode),
2979 ordered_extent->file_offset,
2980 ordered_extent->file_offset +
2981 ordered_extent->len - 1);
2982
2983 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2984 truncated = true;
2985 logical_len = ordered_extent->truncated_len;
2986 /* Truncated the entire extent, don't bother adding */
2987 if (!logical_len)
2988 goto out;
2989 }
2990
2991 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2992 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2993
2994 /*
2995 * For mwrite(mmap + memset to write) case, we still reserve
2996 * space for NOCOW range.
2997 * As NOCOW won't cause a new delayed ref, just free the space
2998 */
2999 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3000 ordered_extent->len);
3001 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3002 if (nolock)
3003 trans = btrfs_join_transaction_nolock(root);
3004 else
3005 trans = btrfs_join_transaction(root);
3006 if (IS_ERR(trans)) {
3007 ret = PTR_ERR(trans);
3008 trans = NULL;
3009 goto out;
3010 }
3011 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3012 ret = btrfs_update_inode_fallback(trans, root, inode);
3013 if (ret) /* -ENOMEM or corruption */
3014 btrfs_abort_transaction(trans, ret);
3015 goto out;
3016 }
3017
3018 range_locked = true;
3019 lock_extent_bits(io_tree, ordered_extent->file_offset,
3020 ordered_extent->file_offset + ordered_extent->len - 1,
3021 &cached_state);
3022
3023 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3024 ordered_extent->file_offset + ordered_extent->len - 1,
3025 EXTENT_DEFRAG, 0, cached_state);
3026 if (ret) {
3027 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3028 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3029 /* the inode is shared */
3030 new = record_old_file_extents(inode, ordered_extent);
3031
3032 clear_extent_bit(io_tree, ordered_extent->file_offset,
3033 ordered_extent->file_offset + ordered_extent->len - 1,
3034 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
3035 }
3036
3037 if (nolock)
3038 trans = btrfs_join_transaction_nolock(root);
3039 else
3040 trans = btrfs_join_transaction(root);
3041 if (IS_ERR(trans)) {
3042 ret = PTR_ERR(trans);
3043 trans = NULL;
3044 goto out;
3045 }
3046
3047 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3048
3049 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3050 compress_type = ordered_extent->compress_type;
3051 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3052 BUG_ON(compress_type);
3053 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3054 ordered_extent->len);
3055 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3056 ordered_extent->file_offset,
3057 ordered_extent->file_offset +
3058 logical_len);
3059 } else {
3060 BUG_ON(root == fs_info->tree_root);
3061 ret = insert_reserved_file_extent(trans, inode,
3062 ordered_extent->file_offset,
3063 ordered_extent->start,
3064 ordered_extent->disk_len,
3065 logical_len, logical_len,
3066 compress_type, 0, 0,
3067 BTRFS_FILE_EXTENT_REG);
3068 if (!ret)
3069 btrfs_release_delalloc_bytes(fs_info,
3070 ordered_extent->start,
3071 ordered_extent->disk_len);
3072 }
3073 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3074 ordered_extent->file_offset, ordered_extent->len,
3075 trans->transid);
3076 if (ret < 0) {
3077 btrfs_abort_transaction(trans, ret);
3078 goto out;
3079 }
3080
3081 add_pending_csums(trans, inode, &ordered_extent->list);
3082
3083 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3084 ret = btrfs_update_inode_fallback(trans, root, inode);
3085 if (ret) { /* -ENOMEM or corruption */
3086 btrfs_abort_transaction(trans, ret);
3087 goto out;
3088 }
3089 ret = 0;
3090 out:
3091 if (range_locked || clear_new_delalloc_bytes) {
3092 unsigned int clear_bits = 0;
3093
3094 if (range_locked)
3095 clear_bits |= EXTENT_LOCKED;
3096 if (clear_new_delalloc_bytes)
3097 clear_bits |= EXTENT_DELALLOC_NEW;
3098 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3099 ordered_extent->file_offset,
3100 ordered_extent->file_offset +
3101 ordered_extent->len - 1,
3102 clear_bits,
3103 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3104 0, &cached_state, GFP_NOFS);
3105 }
3106
3107 if (trans)
3108 btrfs_end_transaction(trans);
3109
3110 if (ret || truncated) {
3111 u64 start, end;
3112
3113 if (truncated)
3114 start = ordered_extent->file_offset + logical_len;
3115 else
3116 start = ordered_extent->file_offset;
3117 end = ordered_extent->file_offset + ordered_extent->len - 1;
3118 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3119
3120 /* Drop the cache for the part of the extent we didn't write. */
3121 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3122
3123 /*
3124 * If the ordered extent had an IOERR or something else went
3125 * wrong we need to return the space for this ordered extent
3126 * back to the allocator. We only free the extent in the
3127 * truncated case if we didn't write out the extent at all.
3128 */
3129 if ((ret || !logical_len) &&
3130 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3131 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3132 btrfs_free_reserved_extent(fs_info,
3133 ordered_extent->start,
3134 ordered_extent->disk_len, 1);
3135 }
3136
3137
3138 /*
3139 * This needs to be done to make sure anybody waiting knows we are done
3140 * updating everything for this ordered extent.
3141 */
3142 btrfs_remove_ordered_extent(inode, ordered_extent);
3143
3144 /* for snapshot-aware defrag */
3145 if (new) {
3146 if (ret) {
3147 free_sa_defrag_extent(new);
3148 atomic_dec(&fs_info->defrag_running);
3149 } else {
3150 relink_file_extents(new);
3151 }
3152 }
3153
3154 /* once for us */
3155 btrfs_put_ordered_extent(ordered_extent);
3156 /* once for the tree */
3157 btrfs_put_ordered_extent(ordered_extent);
3158
3159 return ret;
3160 }
3161
3162 static void finish_ordered_fn(struct btrfs_work *work)
3163 {
3164 struct btrfs_ordered_extent *ordered_extent;
3165 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3166 btrfs_finish_ordered_io(ordered_extent);
3167 }
3168
3169 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3170 struct extent_state *state, int uptodate)
3171 {
3172 struct inode *inode = page->mapping->host;
3173 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3174 struct btrfs_ordered_extent *ordered_extent = NULL;
3175 struct btrfs_workqueue *wq;
3176 btrfs_work_func_t func;
3177
3178 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3179
3180 ClearPagePrivate2(page);
3181 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3182 end - start + 1, uptodate))
3183 return;
3184
3185 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3186 wq = fs_info->endio_freespace_worker;
3187 func = btrfs_freespace_write_helper;
3188 } else {
3189 wq = fs_info->endio_write_workers;
3190 func = btrfs_endio_write_helper;
3191 }
3192
3193 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3194 NULL);
3195 btrfs_queue_work(wq, &ordered_extent->work);
3196 }
3197
3198 static int __readpage_endio_check(struct inode *inode,
3199 struct btrfs_io_bio *io_bio,
3200 int icsum, struct page *page,
3201 int pgoff, u64 start, size_t len)
3202 {
3203 char *kaddr;
3204 u32 csum_expected;
3205 u32 csum = ~(u32)0;
3206
3207 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3208
3209 kaddr = kmap_atomic(page);
3210 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3211 btrfs_csum_final(csum, (u8 *)&csum);
3212 if (csum != csum_expected)
3213 goto zeroit;
3214
3215 kunmap_atomic(kaddr);
3216 return 0;
3217 zeroit:
3218 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3219 io_bio->mirror_num);
3220 memset(kaddr + pgoff, 1, len);
3221 flush_dcache_page(page);
3222 kunmap_atomic(kaddr);
3223 return -EIO;
3224 }
3225
3226 /*
3227 * when reads are done, we need to check csums to verify the data is correct
3228 * if there's a match, we allow the bio to finish. If not, the code in
3229 * extent_io.c will try to find good copies for us.
3230 */
3231 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3232 u64 phy_offset, struct page *page,
3233 u64 start, u64 end, int mirror)
3234 {
3235 size_t offset = start - page_offset(page);
3236 struct inode *inode = page->mapping->host;
3237 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3238 struct btrfs_root *root = BTRFS_I(inode)->root;
3239
3240 if (PageChecked(page)) {
3241 ClearPageChecked(page);
3242 return 0;
3243 }
3244
3245 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3246 return 0;
3247
3248 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3249 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3250 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3251 return 0;
3252 }
3253
3254 phy_offset >>= inode->i_sb->s_blocksize_bits;
3255 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3256 start, (size_t)(end - start + 1));
3257 }
3258
3259 void btrfs_add_delayed_iput(struct inode *inode)
3260 {
3261 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3262 struct btrfs_inode *binode = BTRFS_I(inode);
3263
3264 if (atomic_add_unless(&inode->i_count, -1, 1))
3265 return;
3266
3267 spin_lock(&fs_info->delayed_iput_lock);
3268 if (binode->delayed_iput_count == 0) {
3269 ASSERT(list_empty(&binode->delayed_iput));
3270 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3271 } else {
3272 binode->delayed_iput_count++;
3273 }
3274 spin_unlock(&fs_info->delayed_iput_lock);
3275 }
3276
3277 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3278 {
3279
3280 spin_lock(&fs_info->delayed_iput_lock);
3281 while (!list_empty(&fs_info->delayed_iputs)) {
3282 struct btrfs_inode *inode;
3283
3284 inode = list_first_entry(&fs_info->delayed_iputs,
3285 struct btrfs_inode, delayed_iput);
3286 if (inode->delayed_iput_count) {
3287 inode->delayed_iput_count--;
3288 list_move_tail(&inode->delayed_iput,
3289 &fs_info->delayed_iputs);
3290 } else {
3291 list_del_init(&inode->delayed_iput);
3292 }
3293 spin_unlock(&fs_info->delayed_iput_lock);
3294 iput(&inode->vfs_inode);
3295 spin_lock(&fs_info->delayed_iput_lock);
3296 }
3297 spin_unlock(&fs_info->delayed_iput_lock);
3298 }
3299
3300 /*
3301 * This is called in transaction commit time. If there are no orphan
3302 * files in the subvolume, it removes orphan item and frees block_rsv
3303 * structure.
3304 */
3305 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3306 struct btrfs_root *root)
3307 {
3308 struct btrfs_fs_info *fs_info = root->fs_info;
3309 struct btrfs_block_rsv *block_rsv;
3310 int ret;
3311
3312 if (atomic_read(&root->orphan_inodes) ||
3313 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3314 return;
3315
3316 spin_lock(&root->orphan_lock);
3317 if (atomic_read(&root->orphan_inodes)) {
3318 spin_unlock(&root->orphan_lock);
3319 return;
3320 }
3321
3322 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3323 spin_unlock(&root->orphan_lock);
3324 return;
3325 }
3326
3327 block_rsv = root->orphan_block_rsv;
3328 root->orphan_block_rsv = NULL;
3329 spin_unlock(&root->orphan_lock);
3330
3331 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3332 btrfs_root_refs(&root->root_item) > 0) {
3333 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3334 root->root_key.objectid);
3335 if (ret)
3336 btrfs_abort_transaction(trans, ret);
3337 else
3338 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3339 &root->state);
3340 }
3341
3342 if (block_rsv) {
3343 WARN_ON(block_rsv->size > 0);
3344 btrfs_free_block_rsv(fs_info, block_rsv);
3345 }
3346 }
3347
3348 /*
3349 * This creates an orphan entry for the given inode in case something goes
3350 * wrong in the middle of an unlink/truncate.
3351 *
3352 * NOTE: caller of this function should reserve 5 units of metadata for
3353 * this function.
3354 */
3355 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3356 struct btrfs_inode *inode)
3357 {
3358 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3359 struct btrfs_root *root = inode->root;
3360 struct btrfs_block_rsv *block_rsv = NULL;
3361 int reserve = 0;
3362 int insert = 0;
3363 int ret;
3364
3365 if (!root->orphan_block_rsv) {
3366 block_rsv = btrfs_alloc_block_rsv(fs_info,
3367 BTRFS_BLOCK_RSV_TEMP);
3368 if (!block_rsv)
3369 return -ENOMEM;
3370 }
3371
3372 spin_lock(&root->orphan_lock);
3373 if (!root->orphan_block_rsv) {
3374 root->orphan_block_rsv = block_rsv;
3375 } else if (block_rsv) {
3376 btrfs_free_block_rsv(fs_info, block_rsv);
3377 block_rsv = NULL;
3378 }
3379
3380 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3381 &inode->runtime_flags)) {
3382 #if 0
3383 /*
3384 * For proper ENOSPC handling, we should do orphan
3385 * cleanup when mounting. But this introduces backward
3386 * compatibility issue.
3387 */
3388 if (!xchg(&root->orphan_item_inserted, 1))
3389 insert = 2;
3390 else
3391 insert = 1;
3392 #endif
3393 insert = 1;
3394 atomic_inc(&root->orphan_inodes);
3395 }
3396
3397 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3398 &inode->runtime_flags))
3399 reserve = 1;
3400 spin_unlock(&root->orphan_lock);
3401
3402 /* grab metadata reservation from transaction handle */
3403 if (reserve) {
3404 ret = btrfs_orphan_reserve_metadata(trans, inode);
3405 ASSERT(!ret);
3406 if (ret) {
3407 /*
3408 * dec doesn't need spin_lock as ->orphan_block_rsv
3409 * would be released only if ->orphan_inodes is
3410 * zero.
3411 */
3412 atomic_dec(&root->orphan_inodes);
3413 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3414 &inode->runtime_flags);
3415 if (insert)
3416 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3417 &inode->runtime_flags);
3418 return ret;
3419 }
3420 }
3421
3422 /* insert an orphan item to track this unlinked/truncated file */
3423 if (insert >= 1) {
3424 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3425 if (ret) {
3426 if (reserve) {
3427 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3428 &inode->runtime_flags);
3429 btrfs_orphan_release_metadata(inode);
3430 }
3431 /*
3432 * btrfs_orphan_commit_root may race with us and set
3433 * ->orphan_block_rsv to zero, in order to avoid that,
3434 * decrease ->orphan_inodes after everything is done.
3435 */
3436 atomic_dec(&root->orphan_inodes);
3437 if (ret != -EEXIST) {
3438 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3439 &inode->runtime_flags);
3440 btrfs_abort_transaction(trans, ret);
3441 return ret;
3442 }
3443 }
3444 ret = 0;
3445 }
3446
3447 /* insert an orphan item to track subvolume contains orphan files */
3448 if (insert >= 2) {
3449 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3450 root->root_key.objectid);
3451 if (ret && ret != -EEXIST) {
3452 btrfs_abort_transaction(trans, ret);
3453 return ret;
3454 }
3455 }
3456 return 0;
3457 }
3458
3459 /*
3460 * We have done the truncate/delete so we can go ahead and remove the orphan
3461 * item for this particular inode.
3462 */
3463 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3464 struct btrfs_inode *inode)
3465 {
3466 struct btrfs_root *root = inode->root;
3467 int delete_item = 0;
3468 int ret = 0;
3469
3470 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3471 &inode->runtime_flags))
3472 delete_item = 1;
3473
3474 if (delete_item && trans)
3475 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3476
3477 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3478 &inode->runtime_flags))
3479 btrfs_orphan_release_metadata(inode);
3480
3481 /*
3482 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3483 * to zero, in order to avoid that, decrease ->orphan_inodes after
3484 * everything is done.
3485 */
3486 if (delete_item)
3487 atomic_dec(&root->orphan_inodes);
3488
3489 return ret;
3490 }
3491
3492 /*
3493 * this cleans up any orphans that may be left on the list from the last use
3494 * of this root.
3495 */
3496 int btrfs_orphan_cleanup(struct btrfs_root *root)
3497 {
3498 struct btrfs_fs_info *fs_info = root->fs_info;
3499 struct btrfs_path *path;
3500 struct extent_buffer *leaf;
3501 struct btrfs_key key, found_key;
3502 struct btrfs_trans_handle *trans;
3503 struct inode *inode;
3504 u64 last_objectid = 0;
3505 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3506
3507 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3508 return 0;
3509
3510 path = btrfs_alloc_path();
3511 if (!path) {
3512 ret = -ENOMEM;
3513 goto out;
3514 }
3515 path->reada = READA_BACK;
3516
3517 key.objectid = BTRFS_ORPHAN_OBJECTID;
3518 key.type = BTRFS_ORPHAN_ITEM_KEY;
3519 key.offset = (u64)-1;
3520
3521 while (1) {
3522 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3523 if (ret < 0)
3524 goto out;
3525
3526 /*
3527 * if ret == 0 means we found what we were searching for, which
3528 * is weird, but possible, so only screw with path if we didn't
3529 * find the key and see if we have stuff that matches
3530 */
3531 if (ret > 0) {
3532 ret = 0;
3533 if (path->slots[0] == 0)
3534 break;
3535 path->slots[0]--;
3536 }
3537
3538 /* pull out the item */
3539 leaf = path->nodes[0];
3540 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3541
3542 /* make sure the item matches what we want */
3543 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3544 break;
3545 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3546 break;
3547
3548 /* release the path since we're done with it */
3549 btrfs_release_path(path);
3550
3551 /*
3552 * this is where we are basically btrfs_lookup, without the
3553 * crossing root thing. we store the inode number in the
3554 * offset of the orphan item.
3555 */
3556
3557 if (found_key.offset == last_objectid) {
3558 btrfs_err(fs_info,
3559 "Error removing orphan entry, stopping orphan cleanup");
3560 ret = -EINVAL;
3561 goto out;
3562 }
3563
3564 last_objectid = found_key.offset;
3565
3566 found_key.objectid = found_key.offset;
3567 found_key.type = BTRFS_INODE_ITEM_KEY;
3568 found_key.offset = 0;
3569 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3570 ret = PTR_ERR_OR_ZERO(inode);
3571 if (ret && ret != -ENOENT)
3572 goto out;
3573
3574 if (ret == -ENOENT && root == fs_info->tree_root) {
3575 struct btrfs_root *dead_root;
3576 struct btrfs_fs_info *fs_info = root->fs_info;
3577 int is_dead_root = 0;
3578
3579 /*
3580 * this is an orphan in the tree root. Currently these
3581 * could come from 2 sources:
3582 * a) a snapshot deletion in progress
3583 * b) a free space cache inode
3584 * We need to distinguish those two, as the snapshot
3585 * orphan must not get deleted.
3586 * find_dead_roots already ran before us, so if this
3587 * is a snapshot deletion, we should find the root
3588 * in the dead_roots list
3589 */
3590 spin_lock(&fs_info->trans_lock);
3591 list_for_each_entry(dead_root, &fs_info->dead_roots,
3592 root_list) {
3593 if (dead_root->root_key.objectid ==
3594 found_key.objectid) {
3595 is_dead_root = 1;
3596 break;
3597 }
3598 }
3599 spin_unlock(&fs_info->trans_lock);
3600 if (is_dead_root) {
3601 /* prevent this orphan from being found again */
3602 key.offset = found_key.objectid - 1;
3603 continue;
3604 }
3605 }
3606 /*
3607 * Inode is already gone but the orphan item is still there,
3608 * kill the orphan item.
3609 */
3610 if (ret == -ENOENT) {
3611 trans = btrfs_start_transaction(root, 1);
3612 if (IS_ERR(trans)) {
3613 ret = PTR_ERR(trans);
3614 goto out;
3615 }
3616 btrfs_debug(fs_info, "auto deleting %Lu",
3617 found_key.objectid);
3618 ret = btrfs_del_orphan_item(trans, root,
3619 found_key.objectid);
3620 btrfs_end_transaction(trans);
3621 if (ret)
3622 goto out;
3623 continue;
3624 }
3625
3626 /*
3627 * add this inode to the orphan list so btrfs_orphan_del does
3628 * the proper thing when we hit it
3629 */
3630 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3631 &BTRFS_I(inode)->runtime_flags);
3632 atomic_inc(&root->orphan_inodes);
3633
3634 /* if we have links, this was a truncate, lets do that */
3635 if (inode->i_nlink) {
3636 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3637 iput(inode);
3638 continue;
3639 }
3640 nr_truncate++;
3641
3642 /* 1 for the orphan item deletion. */
3643 trans = btrfs_start_transaction(root, 1);
3644 if (IS_ERR(trans)) {
3645 iput(inode);
3646 ret = PTR_ERR(trans);
3647 goto out;
3648 }
3649 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3650 btrfs_end_transaction(trans);
3651 if (ret) {
3652 iput(inode);
3653 goto out;
3654 }
3655
3656 ret = btrfs_truncate(inode);
3657 if (ret)
3658 btrfs_orphan_del(NULL, BTRFS_I(inode));
3659 } else {
3660 nr_unlink++;
3661 }
3662
3663 /* this will do delete_inode and everything for us */
3664 iput(inode);
3665 if (ret)
3666 goto out;
3667 }
3668 /* release the path since we're done with it */
3669 btrfs_release_path(path);
3670
3671 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3672
3673 if (root->orphan_block_rsv)
3674 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3675 (u64)-1);
3676
3677 if (root->orphan_block_rsv ||
3678 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3679 trans = btrfs_join_transaction(root);
3680 if (!IS_ERR(trans))
3681 btrfs_end_transaction(trans);
3682 }
3683
3684 if (nr_unlink)
3685 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3686 if (nr_truncate)
3687 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3688
3689 out:
3690 if (ret)
3691 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3692 btrfs_free_path(path);
3693 return ret;
3694 }
3695
3696 /*
3697 * very simple check to peek ahead in the leaf looking for xattrs. If we
3698 * don't find any xattrs, we know there can't be any acls.
3699 *
3700 * slot is the slot the inode is in, objectid is the objectid of the inode
3701 */
3702 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3703 int slot, u64 objectid,
3704 int *first_xattr_slot)
3705 {
3706 u32 nritems = btrfs_header_nritems(leaf);
3707 struct btrfs_key found_key;
3708 static u64 xattr_access = 0;
3709 static u64 xattr_default = 0;
3710 int scanned = 0;
3711
3712 if (!xattr_access) {
3713 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3714 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3715 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3716 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3717 }
3718
3719 slot++;
3720 *first_xattr_slot = -1;
3721 while (slot < nritems) {
3722 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3723
3724 /* we found a different objectid, there must not be acls */
3725 if (found_key.objectid != objectid)
3726 return 0;
3727
3728 /* we found an xattr, assume we've got an acl */
3729 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3730 if (*first_xattr_slot == -1)
3731 *first_xattr_slot = slot;
3732 if (found_key.offset == xattr_access ||
3733 found_key.offset == xattr_default)
3734 return 1;
3735 }
3736
3737 /*
3738 * we found a key greater than an xattr key, there can't
3739 * be any acls later on
3740 */
3741 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3742 return 0;
3743
3744 slot++;
3745 scanned++;
3746
3747 /*
3748 * it goes inode, inode backrefs, xattrs, extents,
3749 * so if there are a ton of hard links to an inode there can
3750 * be a lot of backrefs. Don't waste time searching too hard,
3751 * this is just an optimization
3752 */
3753 if (scanned >= 8)
3754 break;
3755 }
3756 /* we hit the end of the leaf before we found an xattr or
3757 * something larger than an xattr. We have to assume the inode
3758 * has acls
3759 */
3760 if (*first_xattr_slot == -1)
3761 *first_xattr_slot = slot;
3762 return 1;
3763 }
3764
3765 /*
3766 * read an inode from the btree into the in-memory inode
3767 */
3768 static int btrfs_read_locked_inode(struct inode *inode)
3769 {
3770 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3771 struct btrfs_path *path;
3772 struct extent_buffer *leaf;
3773 struct btrfs_inode_item *inode_item;
3774 struct btrfs_root *root = BTRFS_I(inode)->root;
3775 struct btrfs_key location;
3776 unsigned long ptr;
3777 int maybe_acls;
3778 u32 rdev;
3779 int ret;
3780 bool filled = false;
3781 int first_xattr_slot;
3782
3783 ret = btrfs_fill_inode(inode, &rdev);
3784 if (!ret)
3785 filled = true;
3786
3787 path = btrfs_alloc_path();
3788 if (!path) {
3789 ret = -ENOMEM;
3790 goto make_bad;
3791 }
3792
3793 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3794
3795 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3796 if (ret) {
3797 if (ret > 0)
3798 ret = -ENOENT;
3799 goto make_bad;
3800 }
3801
3802 leaf = path->nodes[0];
3803
3804 if (filled)
3805 goto cache_index;
3806
3807 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3808 struct btrfs_inode_item);
3809 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3810 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3811 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3812 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3813 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3814
3815 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3816 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3817
3818 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3819 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3820
3821 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3822 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3823
3824 BTRFS_I(inode)->i_otime.tv_sec =
3825 btrfs_timespec_sec(leaf, &inode_item->otime);
3826 BTRFS_I(inode)->i_otime.tv_nsec =
3827 btrfs_timespec_nsec(leaf, &inode_item->otime);
3828
3829 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3830 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3831 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3832
3833 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3834 inode->i_generation = BTRFS_I(inode)->generation;
3835 inode->i_rdev = 0;
3836 rdev = btrfs_inode_rdev(leaf, inode_item);
3837
3838 BTRFS_I(inode)->index_cnt = (u64)-1;
3839 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3840
3841 cache_index:
3842 /*
3843 * If we were modified in the current generation and evicted from memory
3844 * and then re-read we need to do a full sync since we don't have any
3845 * idea about which extents were modified before we were evicted from
3846 * cache.
3847 *
3848 * This is required for both inode re-read from disk and delayed inode
3849 * in delayed_nodes_tree.
3850 */
3851 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3852 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3853 &BTRFS_I(inode)->runtime_flags);
3854
3855 /*
3856 * We don't persist the id of the transaction where an unlink operation
3857 * against the inode was last made. So here we assume the inode might
3858 * have been evicted, and therefore the exact value of last_unlink_trans
3859 * lost, and set it to last_trans to avoid metadata inconsistencies
3860 * between the inode and its parent if the inode is fsync'ed and the log
3861 * replayed. For example, in the scenario:
3862 *
3863 * touch mydir/foo
3864 * ln mydir/foo mydir/bar
3865 * sync
3866 * unlink mydir/bar
3867 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3868 * xfs_io -c fsync mydir/foo
3869 * <power failure>
3870 * mount fs, triggers fsync log replay
3871 *
3872 * We must make sure that when we fsync our inode foo we also log its
3873 * parent inode, otherwise after log replay the parent still has the
3874 * dentry with the "bar" name but our inode foo has a link count of 1
3875 * and doesn't have an inode ref with the name "bar" anymore.
3876 *
3877 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3878 * but it guarantees correctness at the expense of occasional full
3879 * transaction commits on fsync if our inode is a directory, or if our
3880 * inode is not a directory, logging its parent unnecessarily.
3881 */
3882 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3883
3884 path->slots[0]++;
3885 if (inode->i_nlink != 1 ||
3886 path->slots[0] >= btrfs_header_nritems(leaf))
3887 goto cache_acl;
3888
3889 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3890 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3891 goto cache_acl;
3892
3893 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3894 if (location.type == BTRFS_INODE_REF_KEY) {
3895 struct btrfs_inode_ref *ref;
3896
3897 ref = (struct btrfs_inode_ref *)ptr;
3898 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3899 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3900 struct btrfs_inode_extref *extref;
3901
3902 extref = (struct btrfs_inode_extref *)ptr;
3903 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3904 extref);
3905 }
3906 cache_acl:
3907 /*
3908 * try to precache a NULL acl entry for files that don't have
3909 * any xattrs or acls
3910 */
3911 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3912 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3913 if (first_xattr_slot != -1) {
3914 path->slots[0] = first_xattr_slot;
3915 ret = btrfs_load_inode_props(inode, path);
3916 if (ret)
3917 btrfs_err(fs_info,
3918 "error loading props for ino %llu (root %llu): %d",
3919 btrfs_ino(BTRFS_I(inode)),
3920 root->root_key.objectid, ret);
3921 }
3922 btrfs_free_path(path);
3923
3924 if (!maybe_acls)
3925 cache_no_acl(inode);
3926
3927 switch (inode->i_mode & S_IFMT) {
3928 case S_IFREG:
3929 inode->i_mapping->a_ops = &btrfs_aops;
3930 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3931 inode->i_fop = &btrfs_file_operations;
3932 inode->i_op = &btrfs_file_inode_operations;
3933 break;
3934 case S_IFDIR:
3935 inode->i_fop = &btrfs_dir_file_operations;
3936 inode->i_op = &btrfs_dir_inode_operations;
3937 break;
3938 case S_IFLNK:
3939 inode->i_op = &btrfs_symlink_inode_operations;
3940 inode_nohighmem(inode);
3941 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3942 break;
3943 default:
3944 inode->i_op = &btrfs_special_inode_operations;
3945 init_special_inode(inode, inode->i_mode, rdev);
3946 break;
3947 }
3948
3949 btrfs_update_iflags(inode);
3950 return 0;
3951
3952 make_bad:
3953 btrfs_free_path(path);
3954 make_bad_inode(inode);
3955 return ret;
3956 }
3957
3958 /*
3959 * given a leaf and an inode, copy the inode fields into the leaf
3960 */
3961 static void fill_inode_item(struct btrfs_trans_handle *trans,
3962 struct extent_buffer *leaf,
3963 struct btrfs_inode_item *item,
3964 struct inode *inode)
3965 {
3966 struct btrfs_map_token token;
3967
3968 btrfs_init_map_token(&token);
3969
3970 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3971 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3972 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3973 &token);
3974 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3975 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3976
3977 btrfs_set_token_timespec_sec(leaf, &item->atime,
3978 inode->i_atime.tv_sec, &token);
3979 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3980 inode->i_atime.tv_nsec, &token);
3981
3982 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3983 inode->i_mtime.tv_sec, &token);
3984 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3985 inode->i_mtime.tv_nsec, &token);
3986
3987 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3988 inode->i_ctime.tv_sec, &token);
3989 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3990 inode->i_ctime.tv_nsec, &token);
3991
3992 btrfs_set_token_timespec_sec(leaf, &item->otime,
3993 BTRFS_I(inode)->i_otime.tv_sec, &token);
3994 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3995 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3996
3997 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3998 &token);
3999 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
4000 &token);
4001 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
4002 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
4003 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
4004 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
4005 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
4006 }
4007
4008 /*
4009 * copy everything in the in-memory inode into the btree.
4010 */
4011 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4012 struct btrfs_root *root, struct inode *inode)
4013 {
4014 struct btrfs_inode_item *inode_item;
4015 struct btrfs_path *path;
4016 struct extent_buffer *leaf;
4017 int ret;
4018
4019 path = btrfs_alloc_path();
4020 if (!path)
4021 return -ENOMEM;
4022
4023 path->leave_spinning = 1;
4024 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4025 1);
4026 if (ret) {
4027 if (ret > 0)
4028 ret = -ENOENT;
4029 goto failed;
4030 }
4031
4032 leaf = path->nodes[0];
4033 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4034 struct btrfs_inode_item);
4035
4036 fill_inode_item(trans, leaf, inode_item, inode);
4037 btrfs_mark_buffer_dirty(leaf);
4038 btrfs_set_inode_last_trans(trans, inode);
4039 ret = 0;
4040 failed:
4041 btrfs_free_path(path);
4042 return ret;
4043 }
4044
4045 /*
4046 * copy everything in the in-memory inode into the btree.
4047 */
4048 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4049 struct btrfs_root *root, struct inode *inode)
4050 {
4051 struct btrfs_fs_info *fs_info = root->fs_info;
4052 int ret;
4053
4054 /*
4055 * If the inode is a free space inode, we can deadlock during commit
4056 * if we put it into the delayed code.
4057 *
4058 * The data relocation inode should also be directly updated
4059 * without delay
4060 */
4061 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4062 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4063 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4064 btrfs_update_root_times(trans, root);
4065
4066 ret = btrfs_delayed_update_inode(trans, root, inode);
4067 if (!ret)
4068 btrfs_set_inode_last_trans(trans, inode);
4069 return ret;
4070 }
4071
4072 return btrfs_update_inode_item(trans, root, inode);
4073 }
4074
4075 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4076 struct btrfs_root *root,
4077 struct inode *inode)
4078 {
4079 int ret;
4080
4081 ret = btrfs_update_inode(trans, root, inode);
4082 if (ret == -ENOSPC)
4083 return btrfs_update_inode_item(trans, root, inode);
4084 return ret;
4085 }
4086
4087 /*
4088 * unlink helper that gets used here in inode.c and in the tree logging
4089 * recovery code. It remove a link in a directory with a given name, and
4090 * also drops the back refs in the inode to the directory
4091 */
4092 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4093 struct btrfs_root *root,
4094 struct btrfs_inode *dir,
4095 struct btrfs_inode *inode,
4096 const char *name, int name_len)
4097 {
4098 struct btrfs_fs_info *fs_info = root->fs_info;
4099 struct btrfs_path *path;
4100 int ret = 0;
4101 struct extent_buffer *leaf;
4102 struct btrfs_dir_item *di;
4103 struct btrfs_key key;
4104 u64 index;
4105 u64 ino = btrfs_ino(inode);
4106 u64 dir_ino = btrfs_ino(dir);
4107
4108 path = btrfs_alloc_path();
4109 if (!path) {
4110 ret = -ENOMEM;
4111 goto out;
4112 }
4113
4114 path->leave_spinning = 1;
4115 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4116 name, name_len, -1);
4117 if (IS_ERR(di)) {
4118 ret = PTR_ERR(di);
4119 goto err;
4120 }
4121 if (!di) {
4122 ret = -ENOENT;
4123 goto err;
4124 }
4125 leaf = path->nodes[0];
4126 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4127 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4128 if (ret)
4129 goto err;
4130 btrfs_release_path(path);
4131
4132 /*
4133 * If we don't have dir index, we have to get it by looking up
4134 * the inode ref, since we get the inode ref, remove it directly,
4135 * it is unnecessary to do delayed deletion.
4136 *
4137 * But if we have dir index, needn't search inode ref to get it.
4138 * Since the inode ref is close to the inode item, it is better
4139 * that we delay to delete it, and just do this deletion when
4140 * we update the inode item.
4141 */
4142 if (inode->dir_index) {
4143 ret = btrfs_delayed_delete_inode_ref(inode);
4144 if (!ret) {
4145 index = inode->dir_index;
4146 goto skip_backref;
4147 }
4148 }
4149
4150 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4151 dir_ino, &index);
4152 if (ret) {
4153 btrfs_info(fs_info,
4154 "failed to delete reference to %.*s, inode %llu parent %llu",
4155 name_len, name, ino, dir_ino);
4156 btrfs_abort_transaction(trans, ret);
4157 goto err;
4158 }
4159 skip_backref:
4160 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4161 if (ret) {
4162 btrfs_abort_transaction(trans, ret);
4163 goto err;
4164 }
4165
4166 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4167 dir_ino);
4168 if (ret != 0 && ret != -ENOENT) {
4169 btrfs_abort_transaction(trans, ret);
4170 goto err;
4171 }
4172
4173 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4174 index);
4175 if (ret == -ENOENT)
4176 ret = 0;
4177 else if (ret)
4178 btrfs_abort_transaction(trans, ret);
4179 err:
4180 btrfs_free_path(path);
4181 if (ret)
4182 goto out;
4183
4184 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4185 inode_inc_iversion(&inode->vfs_inode);
4186 inode_inc_iversion(&dir->vfs_inode);
4187 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4188 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4189 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4190 out:
4191 return ret;
4192 }
4193
4194 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4195 struct btrfs_root *root,
4196 struct btrfs_inode *dir, struct btrfs_inode *inode,
4197 const char *name, int name_len)
4198 {
4199 int ret;
4200 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4201 if (!ret) {
4202 drop_nlink(&inode->vfs_inode);
4203 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4204 }
4205 return ret;
4206 }
4207
4208 /*
4209 * helper to start transaction for unlink and rmdir.
4210 *
4211 * unlink and rmdir are special in btrfs, they do not always free space, so
4212 * if we cannot make our reservations the normal way try and see if there is
4213 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4214 * allow the unlink to occur.
4215 */
4216 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4217 {
4218 struct btrfs_root *root = BTRFS_I(dir)->root;
4219
4220 /*
4221 * 1 for the possible orphan item
4222 * 1 for the dir item
4223 * 1 for the dir index
4224 * 1 for the inode ref
4225 * 1 for the inode
4226 */
4227 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4228 }
4229
4230 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4231 {
4232 struct btrfs_root *root = BTRFS_I(dir)->root;
4233 struct btrfs_trans_handle *trans;
4234 struct inode *inode = d_inode(dentry);
4235 int ret;
4236
4237 trans = __unlink_start_trans(dir);
4238 if (IS_ERR(trans))
4239 return PTR_ERR(trans);
4240
4241 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4242 0);
4243
4244 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4245 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4246 dentry->d_name.len);
4247 if (ret)
4248 goto out;
4249
4250 if (inode->i_nlink == 0) {
4251 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4252 if (ret)
4253 goto out;
4254 }
4255
4256 out:
4257 btrfs_end_transaction(trans);
4258 btrfs_btree_balance_dirty(root->fs_info);
4259 return ret;
4260 }
4261
4262 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4263 struct btrfs_root *root,
4264 struct inode *dir, u64 objectid,
4265 const char *name, int name_len)
4266 {
4267 struct btrfs_fs_info *fs_info = root->fs_info;
4268 struct btrfs_path *path;
4269 struct extent_buffer *leaf;
4270 struct btrfs_dir_item *di;
4271 struct btrfs_key key;
4272 u64 index;
4273 int ret;
4274 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4275
4276 path = btrfs_alloc_path();
4277 if (!path)
4278 return -ENOMEM;
4279
4280 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4281 name, name_len, -1);
4282 if (IS_ERR_OR_NULL(di)) {
4283 if (!di)
4284 ret = -ENOENT;
4285 else
4286 ret = PTR_ERR(di);
4287 goto out;
4288 }
4289
4290 leaf = path->nodes[0];
4291 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4292 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4293 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4294 if (ret) {
4295 btrfs_abort_transaction(trans, ret);
4296 goto out;
4297 }
4298 btrfs_release_path(path);
4299
4300 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4301 root->root_key.objectid, dir_ino,
4302 &index, name, name_len);
4303 if (ret < 0) {
4304 if (ret != -ENOENT) {
4305 btrfs_abort_transaction(trans, ret);
4306 goto out;
4307 }
4308 di = btrfs_search_dir_index_item(root, path, dir_ino,
4309 name, name_len);
4310 if (IS_ERR_OR_NULL(di)) {
4311 if (!di)
4312 ret = -ENOENT;
4313 else
4314 ret = PTR_ERR(di);
4315 btrfs_abort_transaction(trans, ret);
4316 goto out;
4317 }
4318
4319 leaf = path->nodes[0];
4320 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4321 btrfs_release_path(path);
4322 index = key.offset;
4323 }
4324 btrfs_release_path(path);
4325
4326 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4327 if (ret) {
4328 btrfs_abort_transaction(trans, ret);
4329 goto out;
4330 }
4331
4332 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4333 inode_inc_iversion(dir);
4334 dir->i_mtime = dir->i_ctime = current_time(dir);
4335 ret = btrfs_update_inode_fallback(trans, root, dir);
4336 if (ret)
4337 btrfs_abort_transaction(trans, ret);
4338 out:
4339 btrfs_free_path(path);
4340 return ret;
4341 }
4342
4343 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4344 {
4345 struct inode *inode = d_inode(dentry);
4346 int err = 0;
4347 struct btrfs_root *root = BTRFS_I(dir)->root;
4348 struct btrfs_trans_handle *trans;
4349 u64 last_unlink_trans;
4350
4351 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4352 return -ENOTEMPTY;
4353 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4354 return -EPERM;
4355
4356 trans = __unlink_start_trans(dir);
4357 if (IS_ERR(trans))
4358 return PTR_ERR(trans);
4359
4360 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4361 err = btrfs_unlink_subvol(trans, root, dir,
4362 BTRFS_I(inode)->location.objectid,
4363 dentry->d_name.name,
4364 dentry->d_name.len);
4365 goto out;
4366 }
4367
4368 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4369 if (err)
4370 goto out;
4371
4372 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4373
4374 /* now the directory is empty */
4375 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4376 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4377 dentry->d_name.len);
4378 if (!err) {
4379 btrfs_i_size_write(BTRFS_I(inode), 0);
4380 /*
4381 * Propagate the last_unlink_trans value of the deleted dir to
4382 * its parent directory. This is to prevent an unrecoverable
4383 * log tree in the case we do something like this:
4384 * 1) create dir foo
4385 * 2) create snapshot under dir foo
4386 * 3) delete the snapshot
4387 * 4) rmdir foo
4388 * 5) mkdir foo
4389 * 6) fsync foo or some file inside foo
4390 */
4391 if (last_unlink_trans >= trans->transid)
4392 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4393 }
4394 out:
4395 btrfs_end_transaction(trans);
4396 btrfs_btree_balance_dirty(root->fs_info);
4397
4398 return err;
4399 }
4400
4401 static int truncate_space_check(struct btrfs_trans_handle *trans,
4402 struct btrfs_root *root,
4403 u64 bytes_deleted)
4404 {
4405 struct btrfs_fs_info *fs_info = root->fs_info;
4406 int ret;
4407
4408 /*
4409 * This is only used to apply pressure to the enospc system, we don't
4410 * intend to use this reservation at all.
4411 */
4412 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4413 bytes_deleted *= fs_info->nodesize;
4414 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4415 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4416 if (!ret) {
4417 trace_btrfs_space_reservation(fs_info, "transaction",
4418 trans->transid,
4419 bytes_deleted, 1);
4420 trans->bytes_reserved += bytes_deleted;
4421 }
4422 return ret;
4423
4424 }
4425
4426 /*
4427 * Return this if we need to call truncate_block for the last bit of the
4428 * truncate.
4429 */
4430 #define NEED_TRUNCATE_BLOCK 1
4431
4432 /*
4433 * this can truncate away extent items, csum items and directory items.
4434 * It starts at a high offset and removes keys until it can't find
4435 * any higher than new_size
4436 *
4437 * csum items that cross the new i_size are truncated to the new size
4438 * as well.
4439 *
4440 * min_type is the minimum key type to truncate down to. If set to 0, this
4441 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4442 */
4443 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4444 struct btrfs_root *root,
4445 struct inode *inode,
4446 u64 new_size, u32 min_type)
4447 {
4448 struct btrfs_fs_info *fs_info = root->fs_info;
4449 struct btrfs_path *path;
4450 struct extent_buffer *leaf;
4451 struct btrfs_file_extent_item *fi;
4452 struct btrfs_key key;
4453 struct btrfs_key found_key;
4454 u64 extent_start = 0;
4455 u64 extent_num_bytes = 0;
4456 u64 extent_offset = 0;
4457 u64 item_end = 0;
4458 u64 last_size = new_size;
4459 u32 found_type = (u8)-1;
4460 int found_extent;
4461 int del_item;
4462 int pending_del_nr = 0;
4463 int pending_del_slot = 0;
4464 int extent_type = -1;
4465 int ret;
4466 int err = 0;
4467 u64 ino = btrfs_ino(BTRFS_I(inode));
4468 u64 bytes_deleted = 0;
4469 bool be_nice = false;
4470 bool should_throttle = false;
4471 bool should_end = false;
4472
4473 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4474
4475 /*
4476 * for non-free space inodes and ref cows, we want to back off from
4477 * time to time
4478 */
4479 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4480 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4481 be_nice = true;
4482
4483 path = btrfs_alloc_path();
4484 if (!path)
4485 return -ENOMEM;
4486 path->reada = READA_BACK;
4487
4488 /*
4489 * We want to drop from the next block forward in case this new size is
4490 * not block aligned since we will be keeping the last block of the
4491 * extent just the way it is.
4492 */
4493 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4494 root == fs_info->tree_root)
4495 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4496 fs_info->sectorsize),
4497 (u64)-1, 0);
4498
4499 /*
4500 * This function is also used to drop the items in the log tree before
4501 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4502 * it is used to drop the loged items. So we shouldn't kill the delayed
4503 * items.
4504 */
4505 if (min_type == 0 && root == BTRFS_I(inode)->root)
4506 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4507
4508 key.objectid = ino;
4509 key.offset = (u64)-1;
4510 key.type = (u8)-1;
4511
4512 search_again:
4513 /*
4514 * with a 16K leaf size and 128MB extents, you can actually queue
4515 * up a huge file in a single leaf. Most of the time that
4516 * bytes_deleted is > 0, it will be huge by the time we get here
4517 */
4518 if (be_nice && bytes_deleted > SZ_32M) {
4519 if (btrfs_should_end_transaction(trans)) {
4520 err = -EAGAIN;
4521 goto error;
4522 }
4523 }
4524
4525
4526 path->leave_spinning = 1;
4527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4528 if (ret < 0) {
4529 err = ret;
4530 goto out;
4531 }
4532
4533 if (ret > 0) {
4534 /* there are no items in the tree for us to truncate, we're
4535 * done
4536 */
4537 if (path->slots[0] == 0)
4538 goto out;
4539 path->slots[0]--;
4540 }
4541
4542 while (1) {
4543 fi = NULL;
4544 leaf = path->nodes[0];
4545 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4546 found_type = found_key.type;
4547
4548 if (found_key.objectid != ino)
4549 break;
4550
4551 if (found_type < min_type)
4552 break;
4553
4554 item_end = found_key.offset;
4555 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4556 fi = btrfs_item_ptr(leaf, path->slots[0],
4557 struct btrfs_file_extent_item);
4558 extent_type = btrfs_file_extent_type(leaf, fi);
4559 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4560 item_end +=
4561 btrfs_file_extent_num_bytes(leaf, fi);
4562
4563 trace_btrfs_truncate_show_fi_regular(
4564 BTRFS_I(inode), leaf, fi,
4565 found_key.offset);
4566 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4567 item_end += btrfs_file_extent_inline_len(leaf,
4568 path->slots[0], fi);
4569
4570 trace_btrfs_truncate_show_fi_inline(
4571 BTRFS_I(inode), leaf, fi, path->slots[0],
4572 found_key.offset);
4573 }
4574 item_end--;
4575 }
4576 if (found_type > min_type) {
4577 del_item = 1;
4578 } else {
4579 if (item_end < new_size)
4580 break;
4581 if (found_key.offset >= new_size)
4582 del_item = 1;
4583 else
4584 del_item = 0;
4585 }
4586 found_extent = 0;
4587 /* FIXME, shrink the extent if the ref count is only 1 */
4588 if (found_type != BTRFS_EXTENT_DATA_KEY)
4589 goto delete;
4590
4591 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4592 u64 num_dec;
4593 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4594 if (!del_item) {
4595 u64 orig_num_bytes =
4596 btrfs_file_extent_num_bytes(leaf, fi);
4597 extent_num_bytes = ALIGN(new_size -
4598 found_key.offset,
4599 fs_info->sectorsize);
4600 btrfs_set_file_extent_num_bytes(leaf, fi,
4601 extent_num_bytes);
4602 num_dec = (orig_num_bytes -
4603 extent_num_bytes);
4604 if (test_bit(BTRFS_ROOT_REF_COWS,
4605 &root->state) &&
4606 extent_start != 0)
4607 inode_sub_bytes(inode, num_dec);
4608 btrfs_mark_buffer_dirty(leaf);
4609 } else {
4610 extent_num_bytes =
4611 btrfs_file_extent_disk_num_bytes(leaf,
4612 fi);
4613 extent_offset = found_key.offset -
4614 btrfs_file_extent_offset(leaf, fi);
4615
4616 /* FIXME blocksize != 4096 */
4617 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4618 if (extent_start != 0) {
4619 found_extent = 1;
4620 if (test_bit(BTRFS_ROOT_REF_COWS,
4621 &root->state))
4622 inode_sub_bytes(inode, num_dec);
4623 }
4624 }
4625 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4626 /*
4627 * we can't truncate inline items that have had
4628 * special encodings
4629 */
4630 if (!del_item &&
4631 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4632 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4633 btrfs_file_extent_compression(leaf, fi) == 0) {
4634 u32 size = (u32)(new_size - found_key.offset);
4635
4636 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4637 size = btrfs_file_extent_calc_inline_size(size);
4638 btrfs_truncate_item(root->fs_info, path, size, 1);
4639 } else if (!del_item) {
4640 /*
4641 * We have to bail so the last_size is set to
4642 * just before this extent.
4643 */
4644 err = NEED_TRUNCATE_BLOCK;
4645 break;
4646 }
4647
4648 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4649 inode_sub_bytes(inode, item_end + 1 - new_size);
4650 }
4651 delete:
4652 if (del_item)
4653 last_size = found_key.offset;
4654 else
4655 last_size = new_size;
4656 if (del_item) {
4657 if (!pending_del_nr) {
4658 /* no pending yet, add ourselves */
4659 pending_del_slot = path->slots[0];
4660 pending_del_nr = 1;
4661 } else if (pending_del_nr &&
4662 path->slots[0] + 1 == pending_del_slot) {
4663 /* hop on the pending chunk */
4664 pending_del_nr++;
4665 pending_del_slot = path->slots[0];
4666 } else {
4667 BUG();
4668 }
4669 } else {
4670 break;
4671 }
4672 should_throttle = false;
4673
4674 if (found_extent &&
4675 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4676 root == fs_info->tree_root)) {
4677 btrfs_set_path_blocking(path);
4678 bytes_deleted += extent_num_bytes;
4679 ret = btrfs_free_extent(trans, root, extent_start,
4680 extent_num_bytes, 0,
4681 btrfs_header_owner(leaf),
4682 ino, extent_offset);
4683 BUG_ON(ret);
4684 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4685 btrfs_async_run_delayed_refs(fs_info,
4686 trans->delayed_ref_updates * 2,
4687 trans->transid, 0);
4688 if (be_nice) {
4689 if (truncate_space_check(trans, root,
4690 extent_num_bytes)) {
4691 should_end = true;
4692 }
4693 if (btrfs_should_throttle_delayed_refs(trans,
4694 fs_info))
4695 should_throttle = true;
4696 }
4697 }
4698
4699 if (found_type == BTRFS_INODE_ITEM_KEY)
4700 break;
4701
4702 if (path->slots[0] == 0 ||
4703 path->slots[0] != pending_del_slot ||
4704 should_throttle || should_end) {
4705 if (pending_del_nr) {
4706 ret = btrfs_del_items(trans, root, path,
4707 pending_del_slot,
4708 pending_del_nr);
4709 if (ret) {
4710 btrfs_abort_transaction(trans, ret);
4711 goto error;
4712 }
4713 pending_del_nr = 0;
4714 }
4715 btrfs_release_path(path);
4716 if (should_throttle) {
4717 unsigned long updates = trans->delayed_ref_updates;
4718 if (updates) {
4719 trans->delayed_ref_updates = 0;
4720 ret = btrfs_run_delayed_refs(trans,
4721 fs_info,
4722 updates * 2);
4723 if (ret && !err)
4724 err = ret;
4725 }
4726 }
4727 /*
4728 * if we failed to refill our space rsv, bail out
4729 * and let the transaction restart
4730 */
4731 if (should_end) {
4732 err = -EAGAIN;
4733 goto error;
4734 }
4735 goto search_again;
4736 } else {
4737 path->slots[0]--;
4738 }
4739 }
4740 out:
4741 if (pending_del_nr) {
4742 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4743 pending_del_nr);
4744 if (ret)
4745 btrfs_abort_transaction(trans, ret);
4746 }
4747 error:
4748 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4749 ASSERT(last_size >= new_size);
4750 if (!err && last_size > new_size)
4751 last_size = new_size;
4752 btrfs_ordered_update_i_size(inode, last_size, NULL);
4753 }
4754
4755 btrfs_free_path(path);
4756
4757 if (be_nice && bytes_deleted > SZ_32M) {
4758 unsigned long updates = trans->delayed_ref_updates;
4759 if (updates) {
4760 trans->delayed_ref_updates = 0;
4761 ret = btrfs_run_delayed_refs(trans, fs_info,
4762 updates * 2);
4763 if (ret && !err)
4764 err = ret;
4765 }
4766 }
4767 return err;
4768 }
4769
4770 /*
4771 * btrfs_truncate_block - read, zero a chunk and write a block
4772 * @inode - inode that we're zeroing
4773 * @from - the offset to start zeroing
4774 * @len - the length to zero, 0 to zero the entire range respective to the
4775 * offset
4776 * @front - zero up to the offset instead of from the offset on
4777 *
4778 * This will find the block for the "from" offset and cow the block and zero the
4779 * part we want to zero. This is used with truncate and hole punching.
4780 */
4781 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4782 int front)
4783 {
4784 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4785 struct address_space *mapping = inode->i_mapping;
4786 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4787 struct btrfs_ordered_extent *ordered;
4788 struct extent_state *cached_state = NULL;
4789 struct extent_changeset *data_reserved = NULL;
4790 char *kaddr;
4791 u32 blocksize = fs_info->sectorsize;
4792 pgoff_t index = from >> PAGE_SHIFT;
4793 unsigned offset = from & (blocksize - 1);
4794 struct page *page;
4795 gfp_t mask = btrfs_alloc_write_mask(mapping);
4796 int ret = 0;
4797 u64 block_start;
4798 u64 block_end;
4799
4800 if ((offset & (blocksize - 1)) == 0 &&
4801 (!len || ((len & (blocksize - 1)) == 0)))
4802 goto out;
4803
4804 block_start = round_down(from, blocksize);
4805 block_end = block_start + blocksize - 1;
4806
4807 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4808 block_start, blocksize);
4809 if (ret)
4810 goto out;
4811
4812 again:
4813 page = find_or_create_page(mapping, index, mask);
4814 if (!page) {
4815 btrfs_delalloc_release_space(inode, data_reserved,
4816 block_start, blocksize);
4817 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4818 ret = -ENOMEM;
4819 goto out;
4820 }
4821
4822 if (!PageUptodate(page)) {
4823 ret = btrfs_readpage(NULL, page);
4824 lock_page(page);
4825 if (page->mapping != mapping) {
4826 unlock_page(page);
4827 put_page(page);
4828 goto again;
4829 }
4830 if (!PageUptodate(page)) {
4831 ret = -EIO;
4832 goto out_unlock;
4833 }
4834 }
4835 wait_on_page_writeback(page);
4836
4837 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4838 set_page_extent_mapped(page);
4839
4840 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4841 if (ordered) {
4842 unlock_extent_cached(io_tree, block_start, block_end,
4843 &cached_state, GFP_NOFS);
4844 unlock_page(page);
4845 put_page(page);
4846 btrfs_start_ordered_extent(inode, ordered, 1);
4847 btrfs_put_ordered_extent(ordered);
4848 goto again;
4849 }
4850
4851 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4852 EXTENT_DIRTY | EXTENT_DELALLOC |
4853 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4854 0, 0, &cached_state, GFP_NOFS);
4855
4856 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4857 &cached_state, 0);
4858 if (ret) {
4859 unlock_extent_cached(io_tree, block_start, block_end,
4860 &cached_state, GFP_NOFS);
4861 goto out_unlock;
4862 }
4863
4864 if (offset != blocksize) {
4865 if (!len)
4866 len = blocksize - offset;
4867 kaddr = kmap(page);
4868 if (front)
4869 memset(kaddr + (block_start - page_offset(page)),
4870 0, offset);
4871 else
4872 memset(kaddr + (block_start - page_offset(page)) + offset,
4873 0, len);
4874 flush_dcache_page(page);
4875 kunmap(page);
4876 }
4877 ClearPageChecked(page);
4878 set_page_dirty(page);
4879 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4880 GFP_NOFS);
4881
4882 out_unlock:
4883 if (ret)
4884 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4885 blocksize);
4886 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4887 unlock_page(page);
4888 put_page(page);
4889 out:
4890 extent_changeset_free(data_reserved);
4891 return ret;
4892 }
4893
4894 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4895 u64 offset, u64 len)
4896 {
4897 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4898 struct btrfs_trans_handle *trans;
4899 int ret;
4900
4901 /*
4902 * Still need to make sure the inode looks like it's been updated so
4903 * that any holes get logged if we fsync.
4904 */
4905 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4906 BTRFS_I(inode)->last_trans = fs_info->generation;
4907 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4908 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4909 return 0;
4910 }
4911
4912 /*
4913 * 1 - for the one we're dropping
4914 * 1 - for the one we're adding
4915 * 1 - for updating the inode.
4916 */
4917 trans = btrfs_start_transaction(root, 3);
4918 if (IS_ERR(trans))
4919 return PTR_ERR(trans);
4920
4921 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4922 if (ret) {
4923 btrfs_abort_transaction(trans, ret);
4924 btrfs_end_transaction(trans);
4925 return ret;
4926 }
4927
4928 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4929 offset, 0, 0, len, 0, len, 0, 0, 0);
4930 if (ret)
4931 btrfs_abort_transaction(trans, ret);
4932 else
4933 btrfs_update_inode(trans, root, inode);
4934 btrfs_end_transaction(trans);
4935 return ret;
4936 }
4937
4938 /*
4939 * This function puts in dummy file extents for the area we're creating a hole
4940 * for. So if we are truncating this file to a larger size we need to insert
4941 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4942 * the range between oldsize and size
4943 */
4944 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4945 {
4946 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4947 struct btrfs_root *root = BTRFS_I(inode)->root;
4948 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4949 struct extent_map *em = NULL;
4950 struct extent_state *cached_state = NULL;
4951 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4952 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4953 u64 block_end = ALIGN(size, fs_info->sectorsize);
4954 u64 last_byte;
4955 u64 cur_offset;
4956 u64 hole_size;
4957 int err = 0;
4958
4959 /*
4960 * If our size started in the middle of a block we need to zero out the
4961 * rest of the block before we expand the i_size, otherwise we could
4962 * expose stale data.
4963 */
4964 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4965 if (err)
4966 return err;
4967
4968 if (size <= hole_start)
4969 return 0;
4970
4971 while (1) {
4972 struct btrfs_ordered_extent *ordered;
4973
4974 lock_extent_bits(io_tree, hole_start, block_end - 1,
4975 &cached_state);
4976 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4977 block_end - hole_start);
4978 if (!ordered)
4979 break;
4980 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4981 &cached_state, GFP_NOFS);
4982 btrfs_start_ordered_extent(inode, ordered, 1);
4983 btrfs_put_ordered_extent(ordered);
4984 }
4985
4986 cur_offset = hole_start;
4987 while (1) {
4988 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4989 block_end - cur_offset, 0);
4990 if (IS_ERR(em)) {
4991 err = PTR_ERR(em);
4992 em = NULL;
4993 break;
4994 }
4995 last_byte = min(extent_map_end(em), block_end);
4996 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4997 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4998 struct extent_map *hole_em;
4999 hole_size = last_byte - cur_offset;
5000
5001 err = maybe_insert_hole(root, inode, cur_offset,
5002 hole_size);
5003 if (err)
5004 break;
5005 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5006 cur_offset + hole_size - 1, 0);
5007 hole_em = alloc_extent_map();
5008 if (!hole_em) {
5009 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5010 &BTRFS_I(inode)->runtime_flags);
5011 goto next;
5012 }
5013 hole_em->start = cur_offset;
5014 hole_em->len = hole_size;
5015 hole_em->orig_start = cur_offset;
5016
5017 hole_em->block_start = EXTENT_MAP_HOLE;
5018 hole_em->block_len = 0;
5019 hole_em->orig_block_len = 0;
5020 hole_em->ram_bytes = hole_size;
5021 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5022 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5023 hole_em->generation = fs_info->generation;
5024
5025 while (1) {
5026 write_lock(&em_tree->lock);
5027 err = add_extent_mapping(em_tree, hole_em, 1);
5028 write_unlock(&em_tree->lock);
5029 if (err != -EEXIST)
5030 break;
5031 btrfs_drop_extent_cache(BTRFS_I(inode),
5032 cur_offset,
5033 cur_offset +
5034 hole_size - 1, 0);
5035 }
5036 free_extent_map(hole_em);
5037 }
5038 next:
5039 free_extent_map(em);
5040 em = NULL;
5041 cur_offset = last_byte;
5042 if (cur_offset >= block_end)
5043 break;
5044 }
5045 free_extent_map(em);
5046 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5047 GFP_NOFS);
5048 return err;
5049 }
5050
5051 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5052 {
5053 struct btrfs_root *root = BTRFS_I(inode)->root;
5054 struct btrfs_trans_handle *trans;
5055 loff_t oldsize = i_size_read(inode);
5056 loff_t newsize = attr->ia_size;
5057 int mask = attr->ia_valid;
5058 int ret;
5059
5060 /*
5061 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5062 * special case where we need to update the times despite not having
5063 * these flags set. For all other operations the VFS set these flags
5064 * explicitly if it wants a timestamp update.
5065 */
5066 if (newsize != oldsize) {
5067 inode_inc_iversion(inode);
5068 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5069 inode->i_ctime = inode->i_mtime =
5070 current_time(inode);
5071 }
5072
5073 if (newsize > oldsize) {
5074 /*
5075 * Don't do an expanding truncate while snapshotting is ongoing.
5076 * This is to ensure the snapshot captures a fully consistent
5077 * state of this file - if the snapshot captures this expanding
5078 * truncation, it must capture all writes that happened before
5079 * this truncation.
5080 */
5081 btrfs_wait_for_snapshot_creation(root);
5082 ret = btrfs_cont_expand(inode, oldsize, newsize);
5083 if (ret) {
5084 btrfs_end_write_no_snapshotting(root);
5085 return ret;
5086 }
5087
5088 trans = btrfs_start_transaction(root, 1);
5089 if (IS_ERR(trans)) {
5090 btrfs_end_write_no_snapshotting(root);
5091 return PTR_ERR(trans);
5092 }
5093
5094 i_size_write(inode, newsize);
5095 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5096 pagecache_isize_extended(inode, oldsize, newsize);
5097 ret = btrfs_update_inode(trans, root, inode);
5098 btrfs_end_write_no_snapshotting(root);
5099 btrfs_end_transaction(trans);
5100 } else {
5101
5102 /*
5103 * We're truncating a file that used to have good data down to
5104 * zero. Make sure it gets into the ordered flush list so that
5105 * any new writes get down to disk quickly.
5106 */
5107 if (newsize == 0)
5108 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5109 &BTRFS_I(inode)->runtime_flags);
5110
5111 /*
5112 * 1 for the orphan item we're going to add
5113 * 1 for the orphan item deletion.
5114 */
5115 trans = btrfs_start_transaction(root, 2);
5116 if (IS_ERR(trans))
5117 return PTR_ERR(trans);
5118
5119 /*
5120 * We need to do this in case we fail at _any_ point during the
5121 * actual truncate. Once we do the truncate_setsize we could
5122 * invalidate pages which forces any outstanding ordered io to
5123 * be instantly completed which will give us extents that need
5124 * to be truncated. If we fail to get an orphan inode down we
5125 * could have left over extents that were never meant to live,
5126 * so we need to guarantee from this point on that everything
5127 * will be consistent.
5128 */
5129 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5130 btrfs_end_transaction(trans);
5131 if (ret)
5132 return ret;
5133
5134 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5135 truncate_setsize(inode, newsize);
5136
5137 /* Disable nonlocked read DIO to avoid the end less truncate */
5138 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5139 inode_dio_wait(inode);
5140 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5141
5142 ret = btrfs_truncate(inode);
5143 if (ret && inode->i_nlink) {
5144 int err;
5145
5146 /* To get a stable disk_i_size */
5147 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5148 if (err) {
5149 btrfs_orphan_del(NULL, BTRFS_I(inode));
5150 return err;
5151 }
5152
5153 /*
5154 * failed to truncate, disk_i_size is only adjusted down
5155 * as we remove extents, so it should represent the true
5156 * size of the inode, so reset the in memory size and
5157 * delete our orphan entry.
5158 */
5159 trans = btrfs_join_transaction(root);
5160 if (IS_ERR(trans)) {
5161 btrfs_orphan_del(NULL, BTRFS_I(inode));
5162 return ret;
5163 }
5164 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5165 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5166 if (err)
5167 btrfs_abort_transaction(trans, err);
5168 btrfs_end_transaction(trans);
5169 }
5170 }
5171
5172 return ret;
5173 }
5174
5175 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5176 {
5177 struct inode *inode = d_inode(dentry);
5178 struct btrfs_root *root = BTRFS_I(inode)->root;
5179 int err;
5180
5181 if (btrfs_root_readonly(root))
5182 return -EROFS;
5183
5184 err = setattr_prepare(dentry, attr);
5185 if (err)
5186 return err;
5187
5188 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5189 err = btrfs_setsize(inode, attr);
5190 if (err)
5191 return err;
5192 }
5193
5194 if (attr->ia_valid) {
5195 setattr_copy(inode, attr);
5196 inode_inc_iversion(inode);
5197 err = btrfs_dirty_inode(inode);
5198
5199 if (!err && attr->ia_valid & ATTR_MODE)
5200 err = posix_acl_chmod(inode, inode->i_mode);
5201 }
5202
5203 return err;
5204 }
5205
5206 /*
5207 * While truncating the inode pages during eviction, we get the VFS calling
5208 * btrfs_invalidatepage() against each page of the inode. This is slow because
5209 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5210 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5211 * extent_state structures over and over, wasting lots of time.
5212 *
5213 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5214 * those expensive operations on a per page basis and do only the ordered io
5215 * finishing, while we release here the extent_map and extent_state structures,
5216 * without the excessive merging and splitting.
5217 */
5218 static void evict_inode_truncate_pages(struct inode *inode)
5219 {
5220 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5221 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5222 struct rb_node *node;
5223
5224 ASSERT(inode->i_state & I_FREEING);
5225 truncate_inode_pages_final(&inode->i_data);
5226
5227 write_lock(&map_tree->lock);
5228 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5229 struct extent_map *em;
5230
5231 node = rb_first(&map_tree->map);
5232 em = rb_entry(node, struct extent_map, rb_node);
5233 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5234 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5235 remove_extent_mapping(map_tree, em);
5236 free_extent_map(em);
5237 if (need_resched()) {
5238 write_unlock(&map_tree->lock);
5239 cond_resched();
5240 write_lock(&map_tree->lock);
5241 }
5242 }
5243 write_unlock(&map_tree->lock);
5244
5245 /*
5246 * Keep looping until we have no more ranges in the io tree.
5247 * We can have ongoing bios started by readpages (called from readahead)
5248 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5249 * still in progress (unlocked the pages in the bio but did not yet
5250 * unlocked the ranges in the io tree). Therefore this means some
5251 * ranges can still be locked and eviction started because before
5252 * submitting those bios, which are executed by a separate task (work
5253 * queue kthread), inode references (inode->i_count) were not taken
5254 * (which would be dropped in the end io callback of each bio).
5255 * Therefore here we effectively end up waiting for those bios and
5256 * anyone else holding locked ranges without having bumped the inode's
5257 * reference count - if we don't do it, when they access the inode's
5258 * io_tree to unlock a range it may be too late, leading to an
5259 * use-after-free issue.
5260 */
5261 spin_lock(&io_tree->lock);
5262 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5263 struct extent_state *state;
5264 struct extent_state *cached_state = NULL;
5265 u64 start;
5266 u64 end;
5267
5268 node = rb_first(&io_tree->state);
5269 state = rb_entry(node, struct extent_state, rb_node);
5270 start = state->start;
5271 end = state->end;
5272 spin_unlock(&io_tree->lock);
5273
5274 lock_extent_bits(io_tree, start, end, &cached_state);
5275
5276 /*
5277 * If still has DELALLOC flag, the extent didn't reach disk,
5278 * and its reserved space won't be freed by delayed_ref.
5279 * So we need to free its reserved space here.
5280 * (Refer to comment in btrfs_invalidatepage, case 2)
5281 *
5282 * Note, end is the bytenr of last byte, so we need + 1 here.
5283 */
5284 if (state->state & EXTENT_DELALLOC)
5285 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5286
5287 clear_extent_bit(io_tree, start, end,
5288 EXTENT_LOCKED | EXTENT_DIRTY |
5289 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5290 EXTENT_DEFRAG, 1, 1,
5291 &cached_state, GFP_NOFS);
5292
5293 cond_resched();
5294 spin_lock(&io_tree->lock);
5295 }
5296 spin_unlock(&io_tree->lock);
5297 }
5298
5299 void btrfs_evict_inode(struct inode *inode)
5300 {
5301 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5302 struct btrfs_trans_handle *trans;
5303 struct btrfs_root *root = BTRFS_I(inode)->root;
5304 struct btrfs_block_rsv *rsv, *global_rsv;
5305 int steal_from_global = 0;
5306 u64 min_size;
5307 int ret;
5308
5309 trace_btrfs_inode_evict(inode);
5310
5311 if (!root) {
5312 clear_inode(inode);
5313 return;
5314 }
5315
5316 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5317
5318 evict_inode_truncate_pages(inode);
5319
5320 if (inode->i_nlink &&
5321 ((btrfs_root_refs(&root->root_item) != 0 &&
5322 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5323 btrfs_is_free_space_inode(BTRFS_I(inode))))
5324 goto no_delete;
5325
5326 if (is_bad_inode(inode)) {
5327 btrfs_orphan_del(NULL, BTRFS_I(inode));
5328 goto no_delete;
5329 }
5330 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5331 if (!special_file(inode->i_mode))
5332 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5333
5334 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5335
5336 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5337 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5338 &BTRFS_I(inode)->runtime_flags));
5339 goto no_delete;
5340 }
5341
5342 if (inode->i_nlink > 0) {
5343 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5344 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5345 goto no_delete;
5346 }
5347
5348 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5349 if (ret) {
5350 btrfs_orphan_del(NULL, BTRFS_I(inode));
5351 goto no_delete;
5352 }
5353
5354 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5355 if (!rsv) {
5356 btrfs_orphan_del(NULL, BTRFS_I(inode));
5357 goto no_delete;
5358 }
5359 rsv->size = min_size;
5360 rsv->failfast = 1;
5361 global_rsv = &fs_info->global_block_rsv;
5362
5363 btrfs_i_size_write(BTRFS_I(inode), 0);
5364
5365 /*
5366 * This is a bit simpler than btrfs_truncate since we've already
5367 * reserved our space for our orphan item in the unlink, so we just
5368 * need to reserve some slack space in case we add bytes and update
5369 * inode item when doing the truncate.
5370 */
5371 while (1) {
5372 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5373 BTRFS_RESERVE_FLUSH_LIMIT);
5374
5375 /*
5376 * Try and steal from the global reserve since we will
5377 * likely not use this space anyway, we want to try as
5378 * hard as possible to get this to work.
5379 */
5380 if (ret)
5381 steal_from_global++;
5382 else
5383 steal_from_global = 0;
5384 ret = 0;
5385
5386 /*
5387 * steal_from_global == 0: we reserved stuff, hooray!
5388 * steal_from_global == 1: we didn't reserve stuff, boo!
5389 * steal_from_global == 2: we've committed, still not a lot of
5390 * room but maybe we'll have room in the global reserve this
5391 * time.
5392 * steal_from_global == 3: abandon all hope!
5393 */
5394 if (steal_from_global > 2) {
5395 btrfs_warn(fs_info,
5396 "Could not get space for a delete, will truncate on mount %d",
5397 ret);
5398 btrfs_orphan_del(NULL, BTRFS_I(inode));
5399 btrfs_free_block_rsv(fs_info, rsv);
5400 goto no_delete;
5401 }
5402
5403 trans = btrfs_join_transaction(root);
5404 if (IS_ERR(trans)) {
5405 btrfs_orphan_del(NULL, BTRFS_I(inode));
5406 btrfs_free_block_rsv(fs_info, rsv);
5407 goto no_delete;
5408 }
5409
5410 /*
5411 * We can't just steal from the global reserve, we need to make
5412 * sure there is room to do it, if not we need to commit and try
5413 * again.
5414 */
5415 if (steal_from_global) {
5416 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5417 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5418 min_size, 0);
5419 else
5420 ret = -ENOSPC;
5421 }
5422
5423 /*
5424 * Couldn't steal from the global reserve, we have too much
5425 * pending stuff built up, commit the transaction and try it
5426 * again.
5427 */
5428 if (ret) {
5429 ret = btrfs_commit_transaction(trans);
5430 if (ret) {
5431 btrfs_orphan_del(NULL, BTRFS_I(inode));
5432 btrfs_free_block_rsv(fs_info, rsv);
5433 goto no_delete;
5434 }
5435 continue;
5436 } else {
5437 steal_from_global = 0;
5438 }
5439
5440 trans->block_rsv = rsv;
5441
5442 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5443 if (ret != -ENOSPC && ret != -EAGAIN)
5444 break;
5445
5446 trans->block_rsv = &fs_info->trans_block_rsv;
5447 btrfs_end_transaction(trans);
5448 trans = NULL;
5449 btrfs_btree_balance_dirty(fs_info);
5450 }
5451
5452 btrfs_free_block_rsv(fs_info, rsv);
5453
5454 /*
5455 * Errors here aren't a big deal, it just means we leave orphan items
5456 * in the tree. They will be cleaned up on the next mount.
5457 */
5458 if (ret == 0) {
5459 trans->block_rsv = root->orphan_block_rsv;
5460 btrfs_orphan_del(trans, BTRFS_I(inode));
5461 } else {
5462 btrfs_orphan_del(NULL, BTRFS_I(inode));
5463 }
5464
5465 trans->block_rsv = &fs_info->trans_block_rsv;
5466 if (!(root == fs_info->tree_root ||
5467 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5468 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5469
5470 btrfs_end_transaction(trans);
5471 btrfs_btree_balance_dirty(fs_info);
5472 no_delete:
5473 btrfs_remove_delayed_node(BTRFS_I(inode));
5474 clear_inode(inode);
5475 }
5476
5477 /*
5478 * this returns the key found in the dir entry in the location pointer.
5479 * If no dir entries were found, location->objectid is 0.
5480 */
5481 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5482 struct btrfs_key *location)
5483 {
5484 const char *name = dentry->d_name.name;
5485 int namelen = dentry->d_name.len;
5486 struct btrfs_dir_item *di;
5487 struct btrfs_path *path;
5488 struct btrfs_root *root = BTRFS_I(dir)->root;
5489 int ret = 0;
5490
5491 path = btrfs_alloc_path();
5492 if (!path)
5493 return -ENOMEM;
5494
5495 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5496 name, namelen, 0);
5497 if (IS_ERR(di))
5498 ret = PTR_ERR(di);
5499
5500 if (IS_ERR_OR_NULL(di))
5501 goto out_err;
5502
5503 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5504 if (location->type != BTRFS_INODE_ITEM_KEY &&
5505 location->type != BTRFS_ROOT_ITEM_KEY) {
5506 btrfs_warn(root->fs_info,
5507 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5508 __func__, name, btrfs_ino(BTRFS_I(dir)),
5509 location->objectid, location->type, location->offset);
5510 goto out_err;
5511 }
5512 out:
5513 btrfs_free_path(path);
5514 return ret;
5515 out_err:
5516 location->objectid = 0;
5517 goto out;
5518 }
5519
5520 /*
5521 * when we hit a tree root in a directory, the btrfs part of the inode
5522 * needs to be changed to reflect the root directory of the tree root. This
5523 * is kind of like crossing a mount point.
5524 */
5525 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5526 struct inode *dir,
5527 struct dentry *dentry,
5528 struct btrfs_key *location,
5529 struct btrfs_root **sub_root)
5530 {
5531 struct btrfs_path *path;
5532 struct btrfs_root *new_root;
5533 struct btrfs_root_ref *ref;
5534 struct extent_buffer *leaf;
5535 struct btrfs_key key;
5536 int ret;
5537 int err = 0;
5538
5539 path = btrfs_alloc_path();
5540 if (!path) {
5541 err = -ENOMEM;
5542 goto out;
5543 }
5544
5545 err = -ENOENT;
5546 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5547 key.type = BTRFS_ROOT_REF_KEY;
5548 key.offset = location->objectid;
5549
5550 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5551 if (ret) {
5552 if (ret < 0)
5553 err = ret;
5554 goto out;
5555 }
5556
5557 leaf = path->nodes[0];
5558 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5559 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5560 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5561 goto out;
5562
5563 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5564 (unsigned long)(ref + 1),
5565 dentry->d_name.len);
5566 if (ret)
5567 goto out;
5568
5569 btrfs_release_path(path);
5570
5571 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5572 if (IS_ERR(new_root)) {
5573 err = PTR_ERR(new_root);
5574 goto out;
5575 }
5576
5577 *sub_root = new_root;
5578 location->objectid = btrfs_root_dirid(&new_root->root_item);
5579 location->type = BTRFS_INODE_ITEM_KEY;
5580 location->offset = 0;
5581 err = 0;
5582 out:
5583 btrfs_free_path(path);
5584 return err;
5585 }
5586
5587 static void inode_tree_add(struct inode *inode)
5588 {
5589 struct btrfs_root *root = BTRFS_I(inode)->root;
5590 struct btrfs_inode *entry;
5591 struct rb_node **p;
5592 struct rb_node *parent;
5593 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5594 u64 ino = btrfs_ino(BTRFS_I(inode));
5595
5596 if (inode_unhashed(inode))
5597 return;
5598 parent = NULL;
5599 spin_lock(&root->inode_lock);
5600 p = &root->inode_tree.rb_node;
5601 while (*p) {
5602 parent = *p;
5603 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5604
5605 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5606 p = &parent->rb_left;
5607 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5608 p = &parent->rb_right;
5609 else {
5610 WARN_ON(!(entry->vfs_inode.i_state &
5611 (I_WILL_FREE | I_FREEING)));
5612 rb_replace_node(parent, new, &root->inode_tree);
5613 RB_CLEAR_NODE(parent);
5614 spin_unlock(&root->inode_lock);
5615 return;
5616 }
5617 }
5618 rb_link_node(new, parent, p);
5619 rb_insert_color(new, &root->inode_tree);
5620 spin_unlock(&root->inode_lock);
5621 }
5622
5623 static void inode_tree_del(struct inode *inode)
5624 {
5625 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5626 struct btrfs_root *root = BTRFS_I(inode)->root;
5627 int empty = 0;
5628
5629 spin_lock(&root->inode_lock);
5630 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5631 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5632 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5633 empty = RB_EMPTY_ROOT(&root->inode_tree);
5634 }
5635 spin_unlock(&root->inode_lock);
5636
5637 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5638 synchronize_srcu(&fs_info->subvol_srcu);
5639 spin_lock(&root->inode_lock);
5640 empty = RB_EMPTY_ROOT(&root->inode_tree);
5641 spin_unlock(&root->inode_lock);
5642 if (empty)
5643 btrfs_add_dead_root(root);
5644 }
5645 }
5646
5647 void btrfs_invalidate_inodes(struct btrfs_root *root)
5648 {
5649 struct btrfs_fs_info *fs_info = root->fs_info;
5650 struct rb_node *node;
5651 struct rb_node *prev;
5652 struct btrfs_inode *entry;
5653 struct inode *inode;
5654 u64 objectid = 0;
5655
5656 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5657 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5658
5659 spin_lock(&root->inode_lock);
5660 again:
5661 node = root->inode_tree.rb_node;
5662 prev = NULL;
5663 while (node) {
5664 prev = node;
5665 entry = rb_entry(node, struct btrfs_inode, rb_node);
5666
5667 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5668 node = node->rb_left;
5669 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5670 node = node->rb_right;
5671 else
5672 break;
5673 }
5674 if (!node) {
5675 while (prev) {
5676 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5677 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5678 node = prev;
5679 break;
5680 }
5681 prev = rb_next(prev);
5682 }
5683 }
5684 while (node) {
5685 entry = rb_entry(node, struct btrfs_inode, rb_node);
5686 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5687 inode = igrab(&entry->vfs_inode);
5688 if (inode) {
5689 spin_unlock(&root->inode_lock);
5690 if (atomic_read(&inode->i_count) > 1)
5691 d_prune_aliases(inode);
5692 /*
5693 * btrfs_drop_inode will have it removed from
5694 * the inode cache when its usage count
5695 * hits zero.
5696 */
5697 iput(inode);
5698 cond_resched();
5699 spin_lock(&root->inode_lock);
5700 goto again;
5701 }
5702
5703 if (cond_resched_lock(&root->inode_lock))
5704 goto again;
5705
5706 node = rb_next(node);
5707 }
5708 spin_unlock(&root->inode_lock);
5709 }
5710
5711 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5712 {
5713 struct btrfs_iget_args *args = p;
5714 inode->i_ino = args->location->objectid;
5715 memcpy(&BTRFS_I(inode)->location, args->location,
5716 sizeof(*args->location));
5717 BTRFS_I(inode)->root = args->root;
5718 return 0;
5719 }
5720
5721 static int btrfs_find_actor(struct inode *inode, void *opaque)
5722 {
5723 struct btrfs_iget_args *args = opaque;
5724 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5725 args->root == BTRFS_I(inode)->root;
5726 }
5727
5728 static struct inode *btrfs_iget_locked(struct super_block *s,
5729 struct btrfs_key *location,
5730 struct btrfs_root *root)
5731 {
5732 struct inode *inode;
5733 struct btrfs_iget_args args;
5734 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5735
5736 args.location = location;
5737 args.root = root;
5738
5739 inode = iget5_locked(s, hashval, btrfs_find_actor,
5740 btrfs_init_locked_inode,
5741 (void *)&args);
5742 return inode;
5743 }
5744
5745 /* Get an inode object given its location and corresponding root.
5746 * Returns in *is_new if the inode was read from disk
5747 */
5748 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5749 struct btrfs_root *root, int *new)
5750 {
5751 struct inode *inode;
5752
5753 inode = btrfs_iget_locked(s, location, root);
5754 if (!inode)
5755 return ERR_PTR(-ENOMEM);
5756
5757 if (inode->i_state & I_NEW) {
5758 int ret;
5759
5760 ret = btrfs_read_locked_inode(inode);
5761 if (!is_bad_inode(inode)) {
5762 inode_tree_add(inode);
5763 unlock_new_inode(inode);
5764 if (new)
5765 *new = 1;
5766 } else {
5767 unlock_new_inode(inode);
5768 iput(inode);
5769 ASSERT(ret < 0);
5770 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5771 }
5772 }
5773
5774 return inode;
5775 }
5776
5777 static struct inode *new_simple_dir(struct super_block *s,
5778 struct btrfs_key *key,
5779 struct btrfs_root *root)
5780 {
5781 struct inode *inode = new_inode(s);
5782
5783 if (!inode)
5784 return ERR_PTR(-ENOMEM);
5785
5786 BTRFS_I(inode)->root = root;
5787 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5788 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5789
5790 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5791 inode->i_op = &btrfs_dir_ro_inode_operations;
5792 inode->i_opflags &= ~IOP_XATTR;
5793 inode->i_fop = &simple_dir_operations;
5794 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5795 inode->i_mtime = current_time(inode);
5796 inode->i_atime = inode->i_mtime;
5797 inode->i_ctime = inode->i_mtime;
5798 BTRFS_I(inode)->i_otime = inode->i_mtime;
5799
5800 return inode;
5801 }
5802
5803 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5804 {
5805 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5806 struct inode *inode;
5807 struct btrfs_root *root = BTRFS_I(dir)->root;
5808 struct btrfs_root *sub_root = root;
5809 struct btrfs_key location;
5810 int index;
5811 int ret = 0;
5812
5813 if (dentry->d_name.len > BTRFS_NAME_LEN)
5814 return ERR_PTR(-ENAMETOOLONG);
5815
5816 ret = btrfs_inode_by_name(dir, dentry, &location);
5817 if (ret < 0)
5818 return ERR_PTR(ret);
5819
5820 if (location.objectid == 0)
5821 return ERR_PTR(-ENOENT);
5822
5823 if (location.type == BTRFS_INODE_ITEM_KEY) {
5824 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5825 return inode;
5826 }
5827
5828 index = srcu_read_lock(&fs_info->subvol_srcu);
5829 ret = fixup_tree_root_location(fs_info, dir, dentry,
5830 &location, &sub_root);
5831 if (ret < 0) {
5832 if (ret != -ENOENT)
5833 inode = ERR_PTR(ret);
5834 else
5835 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5836 } else {
5837 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5838 }
5839 srcu_read_unlock(&fs_info->subvol_srcu, index);
5840
5841 if (!IS_ERR(inode) && root != sub_root) {
5842 down_read(&fs_info->cleanup_work_sem);
5843 if (!sb_rdonly(inode->i_sb))
5844 ret = btrfs_orphan_cleanup(sub_root);
5845 up_read(&fs_info->cleanup_work_sem);
5846 if (ret) {
5847 iput(inode);
5848 inode = ERR_PTR(ret);
5849 }
5850 }
5851
5852 return inode;
5853 }
5854
5855 static int btrfs_dentry_delete(const struct dentry *dentry)
5856 {
5857 struct btrfs_root *root;
5858 struct inode *inode = d_inode(dentry);
5859
5860 if (!inode && !IS_ROOT(dentry))
5861 inode = d_inode(dentry->d_parent);
5862
5863 if (inode) {
5864 root = BTRFS_I(inode)->root;
5865 if (btrfs_root_refs(&root->root_item) == 0)
5866 return 1;
5867
5868 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5869 return 1;
5870 }
5871 return 0;
5872 }
5873
5874 static void btrfs_dentry_release(struct dentry *dentry)
5875 {
5876 kfree(dentry->d_fsdata);
5877 }
5878
5879 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5880 unsigned int flags)
5881 {
5882 struct inode *inode;
5883
5884 inode = btrfs_lookup_dentry(dir, dentry);
5885 if (IS_ERR(inode)) {
5886 if (PTR_ERR(inode) == -ENOENT)
5887 inode = NULL;
5888 else
5889 return ERR_CAST(inode);
5890 }
5891
5892 return d_splice_alias(inode, dentry);
5893 }
5894
5895 unsigned char btrfs_filetype_table[] = {
5896 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5897 };
5898
5899 /*
5900 * All this infrastructure exists because dir_emit can fault, and we are holding
5901 * the tree lock when doing readdir. For now just allocate a buffer and copy
5902 * our information into that, and then dir_emit from the buffer. This is
5903 * similar to what NFS does, only we don't keep the buffer around in pagecache
5904 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5905 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5906 * tree lock.
5907 */
5908 static int btrfs_opendir(struct inode *inode, struct file *file)
5909 {
5910 struct btrfs_file_private *private;
5911
5912 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5913 if (!private)
5914 return -ENOMEM;
5915 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5916 if (!private->filldir_buf) {
5917 kfree(private);
5918 return -ENOMEM;
5919 }
5920 file->private_data = private;
5921 return 0;
5922 }
5923
5924 struct dir_entry {
5925 u64 ino;
5926 u64 offset;
5927 unsigned type;
5928 int name_len;
5929 };
5930
5931 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5932 {
5933 while (entries--) {
5934 struct dir_entry *entry = addr;
5935 char *name = (char *)(entry + 1);
5936
5937 ctx->pos = get_unaligned(&entry->offset);
5938 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5939 get_unaligned(&entry->ino),
5940 get_unaligned(&entry->type)))
5941 return 1;
5942 addr += sizeof(struct dir_entry) +
5943 get_unaligned(&entry->name_len);
5944 ctx->pos++;
5945 }
5946 return 0;
5947 }
5948
5949 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5950 {
5951 struct inode *inode = file_inode(file);
5952 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5953 struct btrfs_root *root = BTRFS_I(inode)->root;
5954 struct btrfs_file_private *private = file->private_data;
5955 struct btrfs_dir_item *di;
5956 struct btrfs_key key;
5957 struct btrfs_key found_key;
5958 struct btrfs_path *path;
5959 void *addr;
5960 struct list_head ins_list;
5961 struct list_head del_list;
5962 int ret;
5963 struct extent_buffer *leaf;
5964 int slot;
5965 char *name_ptr;
5966 int name_len;
5967 int entries = 0;
5968 int total_len = 0;
5969 bool put = false;
5970 struct btrfs_key location;
5971
5972 if (!dir_emit_dots(file, ctx))
5973 return 0;
5974
5975 path = btrfs_alloc_path();
5976 if (!path)
5977 return -ENOMEM;
5978
5979 addr = private->filldir_buf;
5980 path->reada = READA_FORWARD;
5981
5982 INIT_LIST_HEAD(&ins_list);
5983 INIT_LIST_HEAD(&del_list);
5984 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5985
5986 again:
5987 key.type = BTRFS_DIR_INDEX_KEY;
5988 key.offset = ctx->pos;
5989 key.objectid = btrfs_ino(BTRFS_I(inode));
5990
5991 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5992 if (ret < 0)
5993 goto err;
5994
5995 while (1) {
5996 struct dir_entry *entry;
5997
5998 leaf = path->nodes[0];
5999 slot = path->slots[0];
6000 if (slot >= btrfs_header_nritems(leaf)) {
6001 ret = btrfs_next_leaf(root, path);
6002 if (ret < 0)
6003 goto err;
6004 else if (ret > 0)
6005 break;
6006 continue;
6007 }
6008
6009 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6010
6011 if (found_key.objectid != key.objectid)
6012 break;
6013 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6014 break;
6015 if (found_key.offset < ctx->pos)
6016 goto next;
6017 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6018 goto next;
6019 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6020 if (verify_dir_item(fs_info, leaf, slot, di))
6021 goto next;
6022
6023 name_len = btrfs_dir_name_len(leaf, di);
6024 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6025 PAGE_SIZE) {
6026 btrfs_release_path(path);
6027 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6028 if (ret)
6029 goto nopos;
6030 addr = private->filldir_buf;
6031 entries = 0;
6032 total_len = 0;
6033 goto again;
6034 }
6035
6036 entry = addr;
6037 put_unaligned(name_len, &entry->name_len);
6038 name_ptr = (char *)(entry + 1);
6039 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6040 name_len);
6041 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
6042 &entry->type);
6043 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6044 put_unaligned(location.objectid, &entry->ino);
6045 put_unaligned(found_key.offset, &entry->offset);
6046 entries++;
6047 addr += sizeof(struct dir_entry) + name_len;
6048 total_len += sizeof(struct dir_entry) + name_len;
6049 next:
6050 path->slots[0]++;
6051 }
6052 btrfs_release_path(path);
6053
6054 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6055 if (ret)
6056 goto nopos;
6057
6058 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6059 if (ret)
6060 goto nopos;
6061
6062 /*
6063 * Stop new entries from being returned after we return the last
6064 * entry.
6065 *
6066 * New directory entries are assigned a strictly increasing
6067 * offset. This means that new entries created during readdir
6068 * are *guaranteed* to be seen in the future by that readdir.
6069 * This has broken buggy programs which operate on names as
6070 * they're returned by readdir. Until we re-use freed offsets
6071 * we have this hack to stop new entries from being returned
6072 * under the assumption that they'll never reach this huge
6073 * offset.
6074 *
6075 * This is being careful not to overflow 32bit loff_t unless the
6076 * last entry requires it because doing so has broken 32bit apps
6077 * in the past.
6078 */
6079 if (ctx->pos >= INT_MAX)
6080 ctx->pos = LLONG_MAX;
6081 else
6082 ctx->pos = INT_MAX;
6083 nopos:
6084 ret = 0;
6085 err:
6086 if (put)
6087 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6088 btrfs_free_path(path);
6089 return ret;
6090 }
6091
6092 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6093 {
6094 struct btrfs_root *root = BTRFS_I(inode)->root;
6095 struct btrfs_trans_handle *trans;
6096 int ret = 0;
6097 bool nolock = false;
6098
6099 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6100 return 0;
6101
6102 if (btrfs_fs_closing(root->fs_info) &&
6103 btrfs_is_free_space_inode(BTRFS_I(inode)))
6104 nolock = true;
6105
6106 if (wbc->sync_mode == WB_SYNC_ALL) {
6107 if (nolock)
6108 trans = btrfs_join_transaction_nolock(root);
6109 else
6110 trans = btrfs_join_transaction(root);
6111 if (IS_ERR(trans))
6112 return PTR_ERR(trans);
6113 ret = btrfs_commit_transaction(trans);
6114 }
6115 return ret;
6116 }
6117
6118 /*
6119 * This is somewhat expensive, updating the tree every time the
6120 * inode changes. But, it is most likely to find the inode in cache.
6121 * FIXME, needs more benchmarking...there are no reasons other than performance
6122 * to keep or drop this code.
6123 */
6124 static int btrfs_dirty_inode(struct inode *inode)
6125 {
6126 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6127 struct btrfs_root *root = BTRFS_I(inode)->root;
6128 struct btrfs_trans_handle *trans;
6129 int ret;
6130
6131 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6132 return 0;
6133
6134 trans = btrfs_join_transaction(root);
6135 if (IS_ERR(trans))
6136 return PTR_ERR(trans);
6137
6138 ret = btrfs_update_inode(trans, root, inode);
6139 if (ret && ret == -ENOSPC) {
6140 /* whoops, lets try again with the full transaction */
6141 btrfs_end_transaction(trans);
6142 trans = btrfs_start_transaction(root, 1);
6143 if (IS_ERR(trans))
6144 return PTR_ERR(trans);
6145
6146 ret = btrfs_update_inode(trans, root, inode);
6147 }
6148 btrfs_end_transaction(trans);
6149 if (BTRFS_I(inode)->delayed_node)
6150 btrfs_balance_delayed_items(fs_info);
6151
6152 return ret;
6153 }
6154
6155 /*
6156 * This is a copy of file_update_time. We need this so we can return error on
6157 * ENOSPC for updating the inode in the case of file write and mmap writes.
6158 */
6159 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6160 int flags)
6161 {
6162 struct btrfs_root *root = BTRFS_I(inode)->root;
6163
6164 if (btrfs_root_readonly(root))
6165 return -EROFS;
6166
6167 if (flags & S_VERSION)
6168 inode_inc_iversion(inode);
6169 if (flags & S_CTIME)
6170 inode->i_ctime = *now;
6171 if (flags & S_MTIME)
6172 inode->i_mtime = *now;
6173 if (flags & S_ATIME)
6174 inode->i_atime = *now;
6175 return btrfs_dirty_inode(inode);
6176 }
6177
6178 /*
6179 * find the highest existing sequence number in a directory
6180 * and then set the in-memory index_cnt variable to reflect
6181 * free sequence numbers
6182 */
6183 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6184 {
6185 struct btrfs_root *root = inode->root;
6186 struct btrfs_key key, found_key;
6187 struct btrfs_path *path;
6188 struct extent_buffer *leaf;
6189 int ret;
6190
6191 key.objectid = btrfs_ino(inode);
6192 key.type = BTRFS_DIR_INDEX_KEY;
6193 key.offset = (u64)-1;
6194
6195 path = btrfs_alloc_path();
6196 if (!path)
6197 return -ENOMEM;
6198
6199 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6200 if (ret < 0)
6201 goto out;
6202 /* FIXME: we should be able to handle this */
6203 if (ret == 0)
6204 goto out;
6205 ret = 0;
6206
6207 /*
6208 * MAGIC NUMBER EXPLANATION:
6209 * since we search a directory based on f_pos we have to start at 2
6210 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6211 * else has to start at 2
6212 */
6213 if (path->slots[0] == 0) {
6214 inode->index_cnt = 2;
6215 goto out;
6216 }
6217
6218 path->slots[0]--;
6219
6220 leaf = path->nodes[0];
6221 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6222
6223 if (found_key.objectid != btrfs_ino(inode) ||
6224 found_key.type != BTRFS_DIR_INDEX_KEY) {
6225 inode->index_cnt = 2;
6226 goto out;
6227 }
6228
6229 inode->index_cnt = found_key.offset + 1;
6230 out:
6231 btrfs_free_path(path);
6232 return ret;
6233 }
6234
6235 /*
6236 * helper to find a free sequence number in a given directory. This current
6237 * code is very simple, later versions will do smarter things in the btree
6238 */
6239 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6240 {
6241 int ret = 0;
6242
6243 if (dir->index_cnt == (u64)-1) {
6244 ret = btrfs_inode_delayed_dir_index_count(dir);
6245 if (ret) {
6246 ret = btrfs_set_inode_index_count(dir);
6247 if (ret)
6248 return ret;
6249 }
6250 }
6251
6252 *index = dir->index_cnt;
6253 dir->index_cnt++;
6254
6255 return ret;
6256 }
6257
6258 static int btrfs_insert_inode_locked(struct inode *inode)
6259 {
6260 struct btrfs_iget_args args;
6261 args.location = &BTRFS_I(inode)->location;
6262 args.root = BTRFS_I(inode)->root;
6263
6264 return insert_inode_locked4(inode,
6265 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6266 btrfs_find_actor, &args);
6267 }
6268
6269 /*
6270 * Inherit flags from the parent inode.
6271 *
6272 * Currently only the compression flags and the cow flags are inherited.
6273 */
6274 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6275 {
6276 unsigned int flags;
6277
6278 if (!dir)
6279 return;
6280
6281 flags = BTRFS_I(dir)->flags;
6282
6283 if (flags & BTRFS_INODE_NOCOMPRESS) {
6284 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6285 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6286 } else if (flags & BTRFS_INODE_COMPRESS) {
6287 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6288 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6289 }
6290
6291 if (flags & BTRFS_INODE_NODATACOW) {
6292 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6293 if (S_ISREG(inode->i_mode))
6294 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6295 }
6296
6297 btrfs_update_iflags(inode);
6298 }
6299
6300 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6301 struct btrfs_root *root,
6302 struct inode *dir,
6303 const char *name, int name_len,
6304 u64 ref_objectid, u64 objectid,
6305 umode_t mode, u64 *index)
6306 {
6307 struct btrfs_fs_info *fs_info = root->fs_info;
6308 struct inode *inode;
6309 struct btrfs_inode_item *inode_item;
6310 struct btrfs_key *location;
6311 struct btrfs_path *path;
6312 struct btrfs_inode_ref *ref;
6313 struct btrfs_key key[2];
6314 u32 sizes[2];
6315 int nitems = name ? 2 : 1;
6316 unsigned long ptr;
6317 int ret;
6318
6319 path = btrfs_alloc_path();
6320 if (!path)
6321 return ERR_PTR(-ENOMEM);
6322
6323 inode = new_inode(fs_info->sb);
6324 if (!inode) {
6325 btrfs_free_path(path);
6326 return ERR_PTR(-ENOMEM);
6327 }
6328
6329 /*
6330 * O_TMPFILE, set link count to 0, so that after this point,
6331 * we fill in an inode item with the correct link count.
6332 */
6333 if (!name)
6334 set_nlink(inode, 0);
6335
6336 /*
6337 * we have to initialize this early, so we can reclaim the inode
6338 * number if we fail afterwards in this function.
6339 */
6340 inode->i_ino = objectid;
6341
6342 if (dir && name) {
6343 trace_btrfs_inode_request(dir);
6344
6345 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6346 if (ret) {
6347 btrfs_free_path(path);
6348 iput(inode);
6349 return ERR_PTR(ret);
6350 }
6351 } else if (dir) {
6352 *index = 0;
6353 }
6354 /*
6355 * index_cnt is ignored for everything but a dir,
6356 * btrfs_get_inode_index_count has an explanation for the magic
6357 * number
6358 */
6359 BTRFS_I(inode)->index_cnt = 2;
6360 BTRFS_I(inode)->dir_index = *index;
6361 BTRFS_I(inode)->root = root;
6362 BTRFS_I(inode)->generation = trans->transid;
6363 inode->i_generation = BTRFS_I(inode)->generation;
6364
6365 /*
6366 * We could have gotten an inode number from somebody who was fsynced
6367 * and then removed in this same transaction, so let's just set full
6368 * sync since it will be a full sync anyway and this will blow away the
6369 * old info in the log.
6370 */
6371 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6372
6373 key[0].objectid = objectid;
6374 key[0].type = BTRFS_INODE_ITEM_KEY;
6375 key[0].offset = 0;
6376
6377 sizes[0] = sizeof(struct btrfs_inode_item);
6378
6379 if (name) {
6380 /*
6381 * Start new inodes with an inode_ref. This is slightly more
6382 * efficient for small numbers of hard links since they will
6383 * be packed into one item. Extended refs will kick in if we
6384 * add more hard links than can fit in the ref item.
6385 */
6386 key[1].objectid = objectid;
6387 key[1].type = BTRFS_INODE_REF_KEY;
6388 key[1].offset = ref_objectid;
6389
6390 sizes[1] = name_len + sizeof(*ref);
6391 }
6392
6393 location = &BTRFS_I(inode)->location;
6394 location->objectid = objectid;
6395 location->offset = 0;
6396 location->type = BTRFS_INODE_ITEM_KEY;
6397
6398 ret = btrfs_insert_inode_locked(inode);
6399 if (ret < 0)
6400 goto fail;
6401
6402 path->leave_spinning = 1;
6403 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6404 if (ret != 0)
6405 goto fail_unlock;
6406
6407 inode_init_owner(inode, dir, mode);
6408 inode_set_bytes(inode, 0);
6409
6410 inode->i_mtime = current_time(inode);
6411 inode->i_atime = inode->i_mtime;
6412 inode->i_ctime = inode->i_mtime;
6413 BTRFS_I(inode)->i_otime = inode->i_mtime;
6414
6415 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6416 struct btrfs_inode_item);
6417 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6418 sizeof(*inode_item));
6419 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6420
6421 if (name) {
6422 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6423 struct btrfs_inode_ref);
6424 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6425 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6426 ptr = (unsigned long)(ref + 1);
6427 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6428 }
6429
6430 btrfs_mark_buffer_dirty(path->nodes[0]);
6431 btrfs_free_path(path);
6432
6433 btrfs_inherit_iflags(inode, dir);
6434
6435 if (S_ISREG(mode)) {
6436 if (btrfs_test_opt(fs_info, NODATASUM))
6437 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6438 if (btrfs_test_opt(fs_info, NODATACOW))
6439 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6440 BTRFS_INODE_NODATASUM;
6441 }
6442
6443 inode_tree_add(inode);
6444
6445 trace_btrfs_inode_new(inode);
6446 btrfs_set_inode_last_trans(trans, inode);
6447
6448 btrfs_update_root_times(trans, root);
6449
6450 ret = btrfs_inode_inherit_props(trans, inode, dir);
6451 if (ret)
6452 btrfs_err(fs_info,
6453 "error inheriting props for ino %llu (root %llu): %d",
6454 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6455
6456 return inode;
6457
6458 fail_unlock:
6459 unlock_new_inode(inode);
6460 fail:
6461 if (dir && name)
6462 BTRFS_I(dir)->index_cnt--;
6463 btrfs_free_path(path);
6464 iput(inode);
6465 return ERR_PTR(ret);
6466 }
6467
6468 static inline u8 btrfs_inode_type(struct inode *inode)
6469 {
6470 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6471 }
6472
6473 /*
6474 * utility function to add 'inode' into 'parent_inode' with
6475 * a give name and a given sequence number.
6476 * if 'add_backref' is true, also insert a backref from the
6477 * inode to the parent directory.
6478 */
6479 int btrfs_add_link(struct btrfs_trans_handle *trans,
6480 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6481 const char *name, int name_len, int add_backref, u64 index)
6482 {
6483 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6484 int ret = 0;
6485 struct btrfs_key key;
6486 struct btrfs_root *root = parent_inode->root;
6487 u64 ino = btrfs_ino(inode);
6488 u64 parent_ino = btrfs_ino(parent_inode);
6489
6490 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6491 memcpy(&key, &inode->root->root_key, sizeof(key));
6492 } else {
6493 key.objectid = ino;
6494 key.type = BTRFS_INODE_ITEM_KEY;
6495 key.offset = 0;
6496 }
6497
6498 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6499 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6500 root->root_key.objectid, parent_ino,
6501 index, name, name_len);
6502 } else if (add_backref) {
6503 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6504 parent_ino, index);
6505 }
6506
6507 /* Nothing to clean up yet */
6508 if (ret)
6509 return ret;
6510
6511 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6512 parent_inode, &key,
6513 btrfs_inode_type(&inode->vfs_inode), index);
6514 if (ret == -EEXIST || ret == -EOVERFLOW)
6515 goto fail_dir_item;
6516 else if (ret) {
6517 btrfs_abort_transaction(trans, ret);
6518 return ret;
6519 }
6520
6521 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6522 name_len * 2);
6523 inode_inc_iversion(&parent_inode->vfs_inode);
6524 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6525 current_time(&parent_inode->vfs_inode);
6526 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6527 if (ret)
6528 btrfs_abort_transaction(trans, ret);
6529 return ret;
6530
6531 fail_dir_item:
6532 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6533 u64 local_index;
6534 int err;
6535 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6536 root->root_key.objectid, parent_ino,
6537 &local_index, name, name_len);
6538
6539 } else if (add_backref) {
6540 u64 local_index;
6541 int err;
6542
6543 err = btrfs_del_inode_ref(trans, root, name, name_len,
6544 ino, parent_ino, &local_index);
6545 }
6546 return ret;
6547 }
6548
6549 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6550 struct btrfs_inode *dir, struct dentry *dentry,
6551 struct btrfs_inode *inode, int backref, u64 index)
6552 {
6553 int err = btrfs_add_link(trans, dir, inode,
6554 dentry->d_name.name, dentry->d_name.len,
6555 backref, index);
6556 if (err > 0)
6557 err = -EEXIST;
6558 return err;
6559 }
6560
6561 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6562 umode_t mode, dev_t rdev)
6563 {
6564 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6565 struct btrfs_trans_handle *trans;
6566 struct btrfs_root *root = BTRFS_I(dir)->root;
6567 struct inode *inode = NULL;
6568 int err;
6569 int drop_inode = 0;
6570 u64 objectid;
6571 u64 index = 0;
6572
6573 /*
6574 * 2 for inode item and ref
6575 * 2 for dir items
6576 * 1 for xattr if selinux is on
6577 */
6578 trans = btrfs_start_transaction(root, 5);
6579 if (IS_ERR(trans))
6580 return PTR_ERR(trans);
6581
6582 err = btrfs_find_free_ino(root, &objectid);
6583 if (err)
6584 goto out_unlock;
6585
6586 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6587 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6588 mode, &index);
6589 if (IS_ERR(inode)) {
6590 err = PTR_ERR(inode);
6591 goto out_unlock;
6592 }
6593
6594 /*
6595 * If the active LSM wants to access the inode during
6596 * d_instantiate it needs these. Smack checks to see
6597 * if the filesystem supports xattrs by looking at the
6598 * ops vector.
6599 */
6600 inode->i_op = &btrfs_special_inode_operations;
6601 init_special_inode(inode, inode->i_mode, rdev);
6602
6603 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6604 if (err)
6605 goto out_unlock_inode;
6606
6607 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6608 0, index);
6609 if (err) {
6610 goto out_unlock_inode;
6611 } else {
6612 btrfs_update_inode(trans, root, inode);
6613 d_instantiate_new(dentry, inode);
6614 }
6615
6616 out_unlock:
6617 btrfs_end_transaction(trans);
6618 btrfs_balance_delayed_items(fs_info);
6619 btrfs_btree_balance_dirty(fs_info);
6620 if (drop_inode) {
6621 inode_dec_link_count(inode);
6622 iput(inode);
6623 }
6624 return err;
6625
6626 out_unlock_inode:
6627 drop_inode = 1;
6628 unlock_new_inode(inode);
6629 goto out_unlock;
6630
6631 }
6632
6633 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6634 umode_t mode, bool excl)
6635 {
6636 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6637 struct btrfs_trans_handle *trans;
6638 struct btrfs_root *root = BTRFS_I(dir)->root;
6639 struct inode *inode = NULL;
6640 int drop_inode_on_err = 0;
6641 int err;
6642 u64 objectid;
6643 u64 index = 0;
6644
6645 /*
6646 * 2 for inode item and ref
6647 * 2 for dir items
6648 * 1 for xattr if selinux is on
6649 */
6650 trans = btrfs_start_transaction(root, 5);
6651 if (IS_ERR(trans))
6652 return PTR_ERR(trans);
6653
6654 err = btrfs_find_free_ino(root, &objectid);
6655 if (err)
6656 goto out_unlock;
6657
6658 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6659 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6660 mode, &index);
6661 if (IS_ERR(inode)) {
6662 err = PTR_ERR(inode);
6663 goto out_unlock;
6664 }
6665 drop_inode_on_err = 1;
6666 /*
6667 * If the active LSM wants to access the inode during
6668 * d_instantiate it needs these. Smack checks to see
6669 * if the filesystem supports xattrs by looking at the
6670 * ops vector.
6671 */
6672 inode->i_fop = &btrfs_file_operations;
6673 inode->i_op = &btrfs_file_inode_operations;
6674 inode->i_mapping->a_ops = &btrfs_aops;
6675
6676 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6677 if (err)
6678 goto out_unlock_inode;
6679
6680 err = btrfs_update_inode(trans, root, inode);
6681 if (err)
6682 goto out_unlock_inode;
6683
6684 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6685 0, index);
6686 if (err)
6687 goto out_unlock_inode;
6688
6689 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6690 d_instantiate_new(dentry, inode);
6691
6692 out_unlock:
6693 btrfs_end_transaction(trans);
6694 if (err && drop_inode_on_err) {
6695 inode_dec_link_count(inode);
6696 iput(inode);
6697 }
6698 btrfs_balance_delayed_items(fs_info);
6699 btrfs_btree_balance_dirty(fs_info);
6700 return err;
6701
6702 out_unlock_inode:
6703 unlock_new_inode(inode);
6704 goto out_unlock;
6705
6706 }
6707
6708 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6709 struct dentry *dentry)
6710 {
6711 struct btrfs_trans_handle *trans = NULL;
6712 struct btrfs_root *root = BTRFS_I(dir)->root;
6713 struct inode *inode = d_inode(old_dentry);
6714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6715 u64 index;
6716 int err;
6717 int drop_inode = 0;
6718
6719 /* do not allow sys_link's with other subvols of the same device */
6720 if (root->objectid != BTRFS_I(inode)->root->objectid)
6721 return -EXDEV;
6722
6723 if (inode->i_nlink >= BTRFS_LINK_MAX)
6724 return -EMLINK;
6725
6726 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6727 if (err)
6728 goto fail;
6729
6730 /*
6731 * 2 items for inode and inode ref
6732 * 2 items for dir items
6733 * 1 item for parent inode
6734 */
6735 trans = btrfs_start_transaction(root, 5);
6736 if (IS_ERR(trans)) {
6737 err = PTR_ERR(trans);
6738 trans = NULL;
6739 goto fail;
6740 }
6741
6742 /* There are several dir indexes for this inode, clear the cache. */
6743 BTRFS_I(inode)->dir_index = 0ULL;
6744 inc_nlink(inode);
6745 inode_inc_iversion(inode);
6746 inode->i_ctime = current_time(inode);
6747 ihold(inode);
6748 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6749
6750 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6751 1, index);
6752
6753 if (err) {
6754 drop_inode = 1;
6755 } else {
6756 struct dentry *parent = dentry->d_parent;
6757 err = btrfs_update_inode(trans, root, inode);
6758 if (err)
6759 goto fail;
6760 if (inode->i_nlink == 1) {
6761 /*
6762 * If new hard link count is 1, it's a file created
6763 * with open(2) O_TMPFILE flag.
6764 */
6765 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6766 if (err)
6767 goto fail;
6768 }
6769 d_instantiate(dentry, inode);
6770 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6771 }
6772
6773 btrfs_balance_delayed_items(fs_info);
6774 fail:
6775 if (trans)
6776 btrfs_end_transaction(trans);
6777 if (drop_inode) {
6778 inode_dec_link_count(inode);
6779 iput(inode);
6780 }
6781 btrfs_btree_balance_dirty(fs_info);
6782 return err;
6783 }
6784
6785 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6786 {
6787 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6788 struct inode *inode = NULL;
6789 struct btrfs_trans_handle *trans;
6790 struct btrfs_root *root = BTRFS_I(dir)->root;
6791 int err = 0;
6792 int drop_on_err = 0;
6793 u64 objectid = 0;
6794 u64 index = 0;
6795
6796 /*
6797 * 2 items for inode and ref
6798 * 2 items for dir items
6799 * 1 for xattr if selinux is on
6800 */
6801 trans = btrfs_start_transaction(root, 5);
6802 if (IS_ERR(trans))
6803 return PTR_ERR(trans);
6804
6805 err = btrfs_find_free_ino(root, &objectid);
6806 if (err)
6807 goto out_fail;
6808
6809 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6810 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6811 S_IFDIR | mode, &index);
6812 if (IS_ERR(inode)) {
6813 err = PTR_ERR(inode);
6814 goto out_fail;
6815 }
6816
6817 drop_on_err = 1;
6818 /* these must be set before we unlock the inode */
6819 inode->i_op = &btrfs_dir_inode_operations;
6820 inode->i_fop = &btrfs_dir_file_operations;
6821
6822 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6823 if (err)
6824 goto out_fail_inode;
6825
6826 btrfs_i_size_write(BTRFS_I(inode), 0);
6827 err = btrfs_update_inode(trans, root, inode);
6828 if (err)
6829 goto out_fail_inode;
6830
6831 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6832 dentry->d_name.name,
6833 dentry->d_name.len, 0, index);
6834 if (err)
6835 goto out_fail_inode;
6836
6837 d_instantiate_new(dentry, inode);
6838 drop_on_err = 0;
6839
6840 out_fail:
6841 btrfs_end_transaction(trans);
6842 if (drop_on_err) {
6843 inode_dec_link_count(inode);
6844 iput(inode);
6845 }
6846 btrfs_balance_delayed_items(fs_info);
6847 btrfs_btree_balance_dirty(fs_info);
6848 return err;
6849
6850 out_fail_inode:
6851 unlock_new_inode(inode);
6852 goto out_fail;
6853 }
6854
6855 /* Find next extent map of a given extent map, caller needs to ensure locks */
6856 static struct extent_map *next_extent_map(struct extent_map *em)
6857 {
6858 struct rb_node *next;
6859
6860 next = rb_next(&em->rb_node);
6861 if (!next)
6862 return NULL;
6863 return container_of(next, struct extent_map, rb_node);
6864 }
6865
6866 static struct extent_map *prev_extent_map(struct extent_map *em)
6867 {
6868 struct rb_node *prev;
6869
6870 prev = rb_prev(&em->rb_node);
6871 if (!prev)
6872 return NULL;
6873 return container_of(prev, struct extent_map, rb_node);
6874 }
6875
6876 /* helper for btfs_get_extent. Given an existing extent in the tree,
6877 * the existing extent is the nearest extent to map_start,
6878 * and an extent that you want to insert, deal with overlap and insert
6879 * the best fitted new extent into the tree.
6880 */
6881 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6882 struct extent_map *existing,
6883 struct extent_map *em,
6884 u64 map_start)
6885 {
6886 struct extent_map *prev;
6887 struct extent_map *next;
6888 u64 start;
6889 u64 end;
6890 u64 start_diff;
6891
6892 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6893
6894 if (existing->start > map_start) {
6895 next = existing;
6896 prev = prev_extent_map(next);
6897 } else {
6898 prev = existing;
6899 next = next_extent_map(prev);
6900 }
6901
6902 start = prev ? extent_map_end(prev) : em->start;
6903 start = max_t(u64, start, em->start);
6904 end = next ? next->start : extent_map_end(em);
6905 end = min_t(u64, end, extent_map_end(em));
6906 start_diff = start - em->start;
6907 em->start = start;
6908 em->len = end - start;
6909 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6910 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6911 em->block_start += start_diff;
6912 em->block_len -= start_diff;
6913 }
6914 return add_extent_mapping(em_tree, em, 0);
6915 }
6916
6917 static noinline int uncompress_inline(struct btrfs_path *path,
6918 struct page *page,
6919 size_t pg_offset, u64 extent_offset,
6920 struct btrfs_file_extent_item *item)
6921 {
6922 int ret;
6923 struct extent_buffer *leaf = path->nodes[0];
6924 char *tmp;
6925 size_t max_size;
6926 unsigned long inline_size;
6927 unsigned long ptr;
6928 int compress_type;
6929
6930 WARN_ON(pg_offset != 0);
6931 compress_type = btrfs_file_extent_compression(leaf, item);
6932 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6933 inline_size = btrfs_file_extent_inline_item_len(leaf,
6934 btrfs_item_nr(path->slots[0]));
6935 tmp = kmalloc(inline_size, GFP_NOFS);
6936 if (!tmp)
6937 return -ENOMEM;
6938 ptr = btrfs_file_extent_inline_start(item);
6939
6940 read_extent_buffer(leaf, tmp, ptr, inline_size);
6941
6942 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6943 ret = btrfs_decompress(compress_type, tmp, page,
6944 extent_offset, inline_size, max_size);
6945
6946 /*
6947 * decompression code contains a memset to fill in any space between the end
6948 * of the uncompressed data and the end of max_size in case the decompressed
6949 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6950 * the end of an inline extent and the beginning of the next block, so we
6951 * cover that region here.
6952 */
6953
6954 if (max_size + pg_offset < PAGE_SIZE) {
6955 char *map = kmap(page);
6956 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6957 kunmap(page);
6958 }
6959 kfree(tmp);
6960 return ret;
6961 }
6962
6963 /*
6964 * a bit scary, this does extent mapping from logical file offset to the disk.
6965 * the ugly parts come from merging extents from the disk with the in-ram
6966 * representation. This gets more complex because of the data=ordered code,
6967 * where the in-ram extents might be locked pending data=ordered completion.
6968 *
6969 * This also copies inline extents directly into the page.
6970 */
6971 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6972 struct page *page,
6973 size_t pg_offset, u64 start, u64 len,
6974 int create)
6975 {
6976 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6977 int ret;
6978 int err = 0;
6979 u64 extent_start = 0;
6980 u64 extent_end = 0;
6981 u64 objectid = btrfs_ino(inode);
6982 u32 found_type;
6983 struct btrfs_path *path = NULL;
6984 struct btrfs_root *root = inode->root;
6985 struct btrfs_file_extent_item *item;
6986 struct extent_buffer *leaf;
6987 struct btrfs_key found_key;
6988 struct extent_map *em = NULL;
6989 struct extent_map_tree *em_tree = &inode->extent_tree;
6990 struct extent_io_tree *io_tree = &inode->io_tree;
6991 struct btrfs_trans_handle *trans = NULL;
6992 const bool new_inline = !page || create;
6993
6994 again:
6995 read_lock(&em_tree->lock);
6996 em = lookup_extent_mapping(em_tree, start, len);
6997 if (em)
6998 em->bdev = fs_info->fs_devices->latest_bdev;
6999 read_unlock(&em_tree->lock);
7000
7001 if (em) {
7002 if (em->start > start || em->start + em->len <= start)
7003 free_extent_map(em);
7004 else if (em->block_start == EXTENT_MAP_INLINE && page)
7005 free_extent_map(em);
7006 else
7007 goto out;
7008 }
7009 em = alloc_extent_map();
7010 if (!em) {
7011 err = -ENOMEM;
7012 goto out;
7013 }
7014 em->bdev = fs_info->fs_devices->latest_bdev;
7015 em->start = EXTENT_MAP_HOLE;
7016 em->orig_start = EXTENT_MAP_HOLE;
7017 em->len = (u64)-1;
7018 em->block_len = (u64)-1;
7019
7020 if (!path) {
7021 path = btrfs_alloc_path();
7022 if (!path) {
7023 err = -ENOMEM;
7024 goto out;
7025 }
7026 /*
7027 * Chances are we'll be called again, so go ahead and do
7028 * readahead
7029 */
7030 path->reada = READA_FORWARD;
7031 }
7032
7033 ret = btrfs_lookup_file_extent(trans, root, path,
7034 objectid, start, trans != NULL);
7035 if (ret < 0) {
7036 err = ret;
7037 goto out;
7038 }
7039
7040 if (ret != 0) {
7041 if (path->slots[0] == 0)
7042 goto not_found;
7043 path->slots[0]--;
7044 }
7045
7046 leaf = path->nodes[0];
7047 item = btrfs_item_ptr(leaf, path->slots[0],
7048 struct btrfs_file_extent_item);
7049 /* are we inside the extent that was found? */
7050 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7051 found_type = found_key.type;
7052 if (found_key.objectid != objectid ||
7053 found_type != BTRFS_EXTENT_DATA_KEY) {
7054 /*
7055 * If we backup past the first extent we want to move forward
7056 * and see if there is an extent in front of us, otherwise we'll
7057 * say there is a hole for our whole search range which can
7058 * cause problems.
7059 */
7060 extent_end = start;
7061 goto next;
7062 }
7063
7064 found_type = btrfs_file_extent_type(leaf, item);
7065 extent_start = found_key.offset;
7066 if (found_type == BTRFS_FILE_EXTENT_REG ||
7067 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7068 extent_end = extent_start +
7069 btrfs_file_extent_num_bytes(leaf, item);
7070
7071 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7072 extent_start);
7073 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7074 size_t size;
7075 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7076 extent_end = ALIGN(extent_start + size,
7077 fs_info->sectorsize);
7078
7079 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7080 path->slots[0],
7081 extent_start);
7082 }
7083 next:
7084 if (start >= extent_end) {
7085 path->slots[0]++;
7086 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7087 ret = btrfs_next_leaf(root, path);
7088 if (ret < 0) {
7089 err = ret;
7090 goto out;
7091 }
7092 if (ret > 0)
7093 goto not_found;
7094 leaf = path->nodes[0];
7095 }
7096 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7097 if (found_key.objectid != objectid ||
7098 found_key.type != BTRFS_EXTENT_DATA_KEY)
7099 goto not_found;
7100 if (start + len <= found_key.offset)
7101 goto not_found;
7102 if (start > found_key.offset)
7103 goto next;
7104 em->start = start;
7105 em->orig_start = start;
7106 em->len = found_key.offset - start;
7107 goto not_found_em;
7108 }
7109
7110 btrfs_extent_item_to_extent_map(inode, path, item,
7111 new_inline, em);
7112
7113 if (found_type == BTRFS_FILE_EXTENT_REG ||
7114 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7115 goto insert;
7116 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7117 unsigned long ptr;
7118 char *map;
7119 size_t size;
7120 size_t extent_offset;
7121 size_t copy_size;
7122
7123 if (new_inline)
7124 goto out;
7125
7126 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7127 extent_offset = page_offset(page) + pg_offset - extent_start;
7128 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7129 size - extent_offset);
7130 em->start = extent_start + extent_offset;
7131 em->len = ALIGN(copy_size, fs_info->sectorsize);
7132 em->orig_block_len = em->len;
7133 em->orig_start = em->start;
7134 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7135 if (create == 0 && !PageUptodate(page)) {
7136 if (btrfs_file_extent_compression(leaf, item) !=
7137 BTRFS_COMPRESS_NONE) {
7138 ret = uncompress_inline(path, page, pg_offset,
7139 extent_offset, item);
7140 if (ret) {
7141 err = ret;
7142 goto out;
7143 }
7144 } else {
7145 map = kmap(page);
7146 read_extent_buffer(leaf, map + pg_offset, ptr,
7147 copy_size);
7148 if (pg_offset + copy_size < PAGE_SIZE) {
7149 memset(map + pg_offset + copy_size, 0,
7150 PAGE_SIZE - pg_offset -
7151 copy_size);
7152 }
7153 kunmap(page);
7154 }
7155 flush_dcache_page(page);
7156 } else if (create && PageUptodate(page)) {
7157 BUG();
7158 if (!trans) {
7159 kunmap(page);
7160 free_extent_map(em);
7161 em = NULL;
7162
7163 btrfs_release_path(path);
7164 trans = btrfs_join_transaction(root);
7165
7166 if (IS_ERR(trans))
7167 return ERR_CAST(trans);
7168 goto again;
7169 }
7170 map = kmap(page);
7171 write_extent_buffer(leaf, map + pg_offset, ptr,
7172 copy_size);
7173 kunmap(page);
7174 btrfs_mark_buffer_dirty(leaf);
7175 }
7176 set_extent_uptodate(io_tree, em->start,
7177 extent_map_end(em) - 1, NULL, GFP_NOFS);
7178 goto insert;
7179 }
7180 not_found:
7181 em->start = start;
7182 em->orig_start = start;
7183 em->len = len;
7184 not_found_em:
7185 em->block_start = EXTENT_MAP_HOLE;
7186 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7187 insert:
7188 btrfs_release_path(path);
7189 if (em->start > start || extent_map_end(em) <= start) {
7190 btrfs_err(fs_info,
7191 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7192 em->start, em->len, start, len);
7193 err = -EIO;
7194 goto out;
7195 }
7196
7197 err = 0;
7198 write_lock(&em_tree->lock);
7199 ret = add_extent_mapping(em_tree, em, 0);
7200 /* it is possible that someone inserted the extent into the tree
7201 * while we had the lock dropped. It is also possible that
7202 * an overlapping map exists in the tree
7203 */
7204 if (ret == -EEXIST) {
7205 struct extent_map *existing;
7206
7207 ret = 0;
7208
7209 existing = search_extent_mapping(em_tree, start, len);
7210 /*
7211 * existing will always be non-NULL, since there must be
7212 * extent causing the -EEXIST.
7213 */
7214 if (start >= existing->start &&
7215 start < extent_map_end(existing)) {
7216 free_extent_map(em);
7217 em = existing;
7218 err = 0;
7219 } else {
7220 /*
7221 * The existing extent map is the one nearest to
7222 * the [start, start + len) range which overlaps
7223 */
7224 err = merge_extent_mapping(em_tree, existing,
7225 em, start);
7226 free_extent_map(existing);
7227 if (err) {
7228 free_extent_map(em);
7229 em = NULL;
7230 }
7231 }
7232 }
7233 write_unlock(&em_tree->lock);
7234 out:
7235
7236 trace_btrfs_get_extent(root, inode, em);
7237
7238 btrfs_free_path(path);
7239 if (trans) {
7240 ret = btrfs_end_transaction(trans);
7241 if (!err)
7242 err = ret;
7243 }
7244 if (err) {
7245 free_extent_map(em);
7246 return ERR_PTR(err);
7247 }
7248 BUG_ON(!em); /* Error is always set */
7249 return em;
7250 }
7251
7252 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7253 struct page *page,
7254 size_t pg_offset, u64 start, u64 len,
7255 int create)
7256 {
7257 struct extent_map *em;
7258 struct extent_map *hole_em = NULL;
7259 u64 range_start = start;
7260 u64 end;
7261 u64 found;
7262 u64 found_end;
7263 int err = 0;
7264
7265 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7266 if (IS_ERR(em))
7267 return em;
7268 /*
7269 * If our em maps to:
7270 * - a hole or
7271 * - a pre-alloc extent,
7272 * there might actually be delalloc bytes behind it.
7273 */
7274 if (em->block_start != EXTENT_MAP_HOLE &&
7275 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7276 return em;
7277 else
7278 hole_em = em;
7279
7280 /* check to see if we've wrapped (len == -1 or similar) */
7281 end = start + len;
7282 if (end < start)
7283 end = (u64)-1;
7284 else
7285 end -= 1;
7286
7287 em = NULL;
7288
7289 /* ok, we didn't find anything, lets look for delalloc */
7290 found = count_range_bits(&inode->io_tree, &range_start,
7291 end, len, EXTENT_DELALLOC, 1);
7292 found_end = range_start + found;
7293 if (found_end < range_start)
7294 found_end = (u64)-1;
7295
7296 /*
7297 * we didn't find anything useful, return
7298 * the original results from get_extent()
7299 */
7300 if (range_start > end || found_end <= start) {
7301 em = hole_em;
7302 hole_em = NULL;
7303 goto out;
7304 }
7305
7306 /* adjust the range_start to make sure it doesn't
7307 * go backwards from the start they passed in
7308 */
7309 range_start = max(start, range_start);
7310 found = found_end - range_start;
7311
7312 if (found > 0) {
7313 u64 hole_start = start;
7314 u64 hole_len = len;
7315
7316 em = alloc_extent_map();
7317 if (!em) {
7318 err = -ENOMEM;
7319 goto out;
7320 }
7321 /*
7322 * when btrfs_get_extent can't find anything it
7323 * returns one huge hole
7324 *
7325 * make sure what it found really fits our range, and
7326 * adjust to make sure it is based on the start from
7327 * the caller
7328 */
7329 if (hole_em) {
7330 u64 calc_end = extent_map_end(hole_em);
7331
7332 if (calc_end <= start || (hole_em->start > end)) {
7333 free_extent_map(hole_em);
7334 hole_em = NULL;
7335 } else {
7336 hole_start = max(hole_em->start, start);
7337 hole_len = calc_end - hole_start;
7338 }
7339 }
7340 em->bdev = NULL;
7341 if (hole_em && range_start > hole_start) {
7342 /* our hole starts before our delalloc, so we
7343 * have to return just the parts of the hole
7344 * that go until the delalloc starts
7345 */
7346 em->len = min(hole_len,
7347 range_start - hole_start);
7348 em->start = hole_start;
7349 em->orig_start = hole_start;
7350 /*
7351 * don't adjust block start at all,
7352 * it is fixed at EXTENT_MAP_HOLE
7353 */
7354 em->block_start = hole_em->block_start;
7355 em->block_len = hole_len;
7356 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7357 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7358 } else {
7359 em->start = range_start;
7360 em->len = found;
7361 em->orig_start = range_start;
7362 em->block_start = EXTENT_MAP_DELALLOC;
7363 em->block_len = found;
7364 }
7365 } else if (hole_em) {
7366 return hole_em;
7367 }
7368 out:
7369
7370 free_extent_map(hole_em);
7371 if (err) {
7372 free_extent_map(em);
7373 return ERR_PTR(err);
7374 }
7375 return em;
7376 }
7377
7378 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7379 const u64 start,
7380 const u64 len,
7381 const u64 orig_start,
7382 const u64 block_start,
7383 const u64 block_len,
7384 const u64 orig_block_len,
7385 const u64 ram_bytes,
7386 const int type)
7387 {
7388 struct extent_map *em = NULL;
7389 int ret;
7390
7391 if (type != BTRFS_ORDERED_NOCOW) {
7392 em = create_io_em(inode, start, len, orig_start,
7393 block_start, block_len, orig_block_len,
7394 ram_bytes,
7395 BTRFS_COMPRESS_NONE, /* compress_type */
7396 type);
7397 if (IS_ERR(em))
7398 goto out;
7399 }
7400 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7401 len, block_len, type);
7402 if (ret) {
7403 if (em) {
7404 free_extent_map(em);
7405 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7406 start + len - 1, 0);
7407 }
7408 em = ERR_PTR(ret);
7409 }
7410 out:
7411
7412 return em;
7413 }
7414
7415 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7416 u64 start, u64 len)
7417 {
7418 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7419 struct btrfs_root *root = BTRFS_I(inode)->root;
7420 struct extent_map *em;
7421 struct btrfs_key ins;
7422 u64 alloc_hint;
7423 int ret;
7424
7425 alloc_hint = get_extent_allocation_hint(inode, start, len);
7426 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7427 0, alloc_hint, &ins, 1, 1);
7428 if (ret)
7429 return ERR_PTR(ret);
7430
7431 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7432 ins.objectid, ins.offset, ins.offset,
7433 ins.offset, BTRFS_ORDERED_REGULAR);
7434 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7435 if (IS_ERR(em))
7436 btrfs_free_reserved_extent(fs_info, ins.objectid,
7437 ins.offset, 1);
7438
7439 return em;
7440 }
7441
7442 /*
7443 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7444 * block must be cow'd
7445 */
7446 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7447 u64 *orig_start, u64 *orig_block_len,
7448 u64 *ram_bytes)
7449 {
7450 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7451 struct btrfs_path *path;
7452 int ret;
7453 struct extent_buffer *leaf;
7454 struct btrfs_root *root = BTRFS_I(inode)->root;
7455 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7456 struct btrfs_file_extent_item *fi;
7457 struct btrfs_key key;
7458 u64 disk_bytenr;
7459 u64 backref_offset;
7460 u64 extent_end;
7461 u64 num_bytes;
7462 int slot;
7463 int found_type;
7464 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7465
7466 path = btrfs_alloc_path();
7467 if (!path)
7468 return -ENOMEM;
7469
7470 ret = btrfs_lookup_file_extent(NULL, root, path,
7471 btrfs_ino(BTRFS_I(inode)), offset, 0);
7472 if (ret < 0)
7473 goto out;
7474
7475 slot = path->slots[0];
7476 if (ret == 1) {
7477 if (slot == 0) {
7478 /* can't find the item, must cow */
7479 ret = 0;
7480 goto out;
7481 }
7482 slot--;
7483 }
7484 ret = 0;
7485 leaf = path->nodes[0];
7486 btrfs_item_key_to_cpu(leaf, &key, slot);
7487 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7488 key.type != BTRFS_EXTENT_DATA_KEY) {
7489 /* not our file or wrong item type, must cow */
7490 goto out;
7491 }
7492
7493 if (key.offset > offset) {
7494 /* Wrong offset, must cow */
7495 goto out;
7496 }
7497
7498 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7499 found_type = btrfs_file_extent_type(leaf, fi);
7500 if (found_type != BTRFS_FILE_EXTENT_REG &&
7501 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7502 /* not a regular extent, must cow */
7503 goto out;
7504 }
7505
7506 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7507 goto out;
7508
7509 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7510 if (extent_end <= offset)
7511 goto out;
7512
7513 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7514 if (disk_bytenr == 0)
7515 goto out;
7516
7517 if (btrfs_file_extent_compression(leaf, fi) ||
7518 btrfs_file_extent_encryption(leaf, fi) ||
7519 btrfs_file_extent_other_encoding(leaf, fi))
7520 goto out;
7521
7522 backref_offset = btrfs_file_extent_offset(leaf, fi);
7523
7524 if (orig_start) {
7525 *orig_start = key.offset - backref_offset;
7526 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7527 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7528 }
7529
7530 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7531 goto out;
7532
7533 num_bytes = min(offset + *len, extent_end) - offset;
7534 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7535 u64 range_end;
7536
7537 range_end = round_up(offset + num_bytes,
7538 root->fs_info->sectorsize) - 1;
7539 ret = test_range_bit(io_tree, offset, range_end,
7540 EXTENT_DELALLOC, 0, NULL);
7541 if (ret) {
7542 ret = -EAGAIN;
7543 goto out;
7544 }
7545 }
7546
7547 btrfs_release_path(path);
7548
7549 /*
7550 * look for other files referencing this extent, if we
7551 * find any we must cow
7552 */
7553
7554 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7555 key.offset - backref_offset, disk_bytenr);
7556 if (ret) {
7557 ret = 0;
7558 goto out;
7559 }
7560
7561 /*
7562 * adjust disk_bytenr and num_bytes to cover just the bytes
7563 * in this extent we are about to write. If there
7564 * are any csums in that range we have to cow in order
7565 * to keep the csums correct
7566 */
7567 disk_bytenr += backref_offset;
7568 disk_bytenr += offset - key.offset;
7569 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7570 goto out;
7571 /*
7572 * all of the above have passed, it is safe to overwrite this extent
7573 * without cow
7574 */
7575 *len = num_bytes;
7576 ret = 1;
7577 out:
7578 btrfs_free_path(path);
7579 return ret;
7580 }
7581
7582 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7583 {
7584 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7585 bool found = false;
7586 void **pagep = NULL;
7587 struct page *page = NULL;
7588 unsigned long start_idx;
7589 unsigned long end_idx;
7590
7591 start_idx = start >> PAGE_SHIFT;
7592
7593 /*
7594 * end is the last byte in the last page. end == start is legal
7595 */
7596 end_idx = end >> PAGE_SHIFT;
7597
7598 rcu_read_lock();
7599
7600 /* Most of the code in this while loop is lifted from
7601 * find_get_page. It's been modified to begin searching from a
7602 * page and return just the first page found in that range. If the
7603 * found idx is less than or equal to the end idx then we know that
7604 * a page exists. If no pages are found or if those pages are
7605 * outside of the range then we're fine (yay!) */
7606 while (page == NULL &&
7607 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7608 page = radix_tree_deref_slot(pagep);
7609 if (unlikely(!page))
7610 break;
7611
7612 if (radix_tree_exception(page)) {
7613 if (radix_tree_deref_retry(page)) {
7614 page = NULL;
7615 continue;
7616 }
7617 /*
7618 * Otherwise, shmem/tmpfs must be storing a swap entry
7619 * here as an exceptional entry: so return it without
7620 * attempting to raise page count.
7621 */
7622 page = NULL;
7623 break; /* TODO: Is this relevant for this use case? */
7624 }
7625
7626 if (!page_cache_get_speculative(page)) {
7627 page = NULL;
7628 continue;
7629 }
7630
7631 /*
7632 * Has the page moved?
7633 * This is part of the lockless pagecache protocol. See
7634 * include/linux/pagemap.h for details.
7635 */
7636 if (unlikely(page != *pagep)) {
7637 put_page(page);
7638 page = NULL;
7639 }
7640 }
7641
7642 if (page) {
7643 if (page->index <= end_idx)
7644 found = true;
7645 put_page(page);
7646 }
7647
7648 rcu_read_unlock();
7649 return found;
7650 }
7651
7652 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7653 struct extent_state **cached_state, int writing)
7654 {
7655 struct btrfs_ordered_extent *ordered;
7656 int ret = 0;
7657
7658 while (1) {
7659 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7660 cached_state);
7661 /*
7662 * We're concerned with the entire range that we're going to be
7663 * doing DIO to, so we need to make sure there's no ordered
7664 * extents in this range.
7665 */
7666 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7667 lockend - lockstart + 1);
7668
7669 /*
7670 * We need to make sure there are no buffered pages in this
7671 * range either, we could have raced between the invalidate in
7672 * generic_file_direct_write and locking the extent. The
7673 * invalidate needs to happen so that reads after a write do not
7674 * get stale data.
7675 */
7676 if (!ordered &&
7677 (!writing ||
7678 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7679 break;
7680
7681 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7682 cached_state, GFP_NOFS);
7683
7684 if (ordered) {
7685 /*
7686 * If we are doing a DIO read and the ordered extent we
7687 * found is for a buffered write, we can not wait for it
7688 * to complete and retry, because if we do so we can
7689 * deadlock with concurrent buffered writes on page
7690 * locks. This happens only if our DIO read covers more
7691 * than one extent map, if at this point has already
7692 * created an ordered extent for a previous extent map
7693 * and locked its range in the inode's io tree, and a
7694 * concurrent write against that previous extent map's
7695 * range and this range started (we unlock the ranges
7696 * in the io tree only when the bios complete and
7697 * buffered writes always lock pages before attempting
7698 * to lock range in the io tree).
7699 */
7700 if (writing ||
7701 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7702 btrfs_start_ordered_extent(inode, ordered, 1);
7703 else
7704 ret = -ENOTBLK;
7705 btrfs_put_ordered_extent(ordered);
7706 } else {
7707 /*
7708 * We could trigger writeback for this range (and wait
7709 * for it to complete) and then invalidate the pages for
7710 * this range (through invalidate_inode_pages2_range()),
7711 * but that can lead us to a deadlock with a concurrent
7712 * call to readpages() (a buffered read or a defrag call
7713 * triggered a readahead) on a page lock due to an
7714 * ordered dio extent we created before but did not have
7715 * yet a corresponding bio submitted (whence it can not
7716 * complete), which makes readpages() wait for that
7717 * ordered extent to complete while holding a lock on
7718 * that page.
7719 */
7720 ret = -ENOTBLK;
7721 }
7722
7723 if (ret)
7724 break;
7725
7726 cond_resched();
7727 }
7728
7729 return ret;
7730 }
7731
7732 /* The callers of this must take lock_extent() */
7733 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7734 u64 orig_start, u64 block_start,
7735 u64 block_len, u64 orig_block_len,
7736 u64 ram_bytes, int compress_type,
7737 int type)
7738 {
7739 struct extent_map_tree *em_tree;
7740 struct extent_map *em;
7741 struct btrfs_root *root = BTRFS_I(inode)->root;
7742 int ret;
7743
7744 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7745 type == BTRFS_ORDERED_COMPRESSED ||
7746 type == BTRFS_ORDERED_NOCOW ||
7747 type == BTRFS_ORDERED_REGULAR);
7748
7749 em_tree = &BTRFS_I(inode)->extent_tree;
7750 em = alloc_extent_map();
7751 if (!em)
7752 return ERR_PTR(-ENOMEM);
7753
7754 em->start = start;
7755 em->orig_start = orig_start;
7756 em->len = len;
7757 em->block_len = block_len;
7758 em->block_start = block_start;
7759 em->bdev = root->fs_info->fs_devices->latest_bdev;
7760 em->orig_block_len = orig_block_len;
7761 em->ram_bytes = ram_bytes;
7762 em->generation = -1;
7763 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7764 if (type == BTRFS_ORDERED_PREALLOC) {
7765 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7766 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7767 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7768 em->compress_type = compress_type;
7769 }
7770
7771 do {
7772 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7773 em->start + em->len - 1, 0);
7774 write_lock(&em_tree->lock);
7775 ret = add_extent_mapping(em_tree, em, 1);
7776 write_unlock(&em_tree->lock);
7777 /*
7778 * The caller has taken lock_extent(), who could race with us
7779 * to add em?
7780 */
7781 } while (ret == -EEXIST);
7782
7783 if (ret) {
7784 free_extent_map(em);
7785 return ERR_PTR(ret);
7786 }
7787
7788 /* em got 2 refs now, callers needs to do free_extent_map once. */
7789 return em;
7790 }
7791
7792 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7793 struct buffer_head *bh_result, int create)
7794 {
7795 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7796 struct extent_map *em;
7797 struct extent_state *cached_state = NULL;
7798 struct btrfs_dio_data *dio_data = NULL;
7799 u64 start = iblock << inode->i_blkbits;
7800 u64 lockstart, lockend;
7801 u64 len = bh_result->b_size;
7802 int unlock_bits = EXTENT_LOCKED;
7803 int ret = 0;
7804
7805 if (create)
7806 unlock_bits |= EXTENT_DIRTY;
7807 else
7808 len = min_t(u64, len, fs_info->sectorsize);
7809
7810 lockstart = start;
7811 lockend = start + len - 1;
7812
7813 if (current->journal_info) {
7814 /*
7815 * Need to pull our outstanding extents and set journal_info to NULL so
7816 * that anything that needs to check if there's a transaction doesn't get
7817 * confused.
7818 */
7819 dio_data = current->journal_info;
7820 current->journal_info = NULL;
7821 }
7822
7823 /*
7824 * If this errors out it's because we couldn't invalidate pagecache for
7825 * this range and we need to fallback to buffered.
7826 */
7827 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7828 create)) {
7829 ret = -ENOTBLK;
7830 goto err;
7831 }
7832
7833 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7834 if (IS_ERR(em)) {
7835 ret = PTR_ERR(em);
7836 goto unlock_err;
7837 }
7838
7839 /*
7840 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7841 * io. INLINE is special, and we could probably kludge it in here, but
7842 * it's still buffered so for safety lets just fall back to the generic
7843 * buffered path.
7844 *
7845 * For COMPRESSED we _have_ to read the entire extent in so we can
7846 * decompress it, so there will be buffering required no matter what we
7847 * do, so go ahead and fallback to buffered.
7848 *
7849 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7850 * to buffered IO. Don't blame me, this is the price we pay for using
7851 * the generic code.
7852 */
7853 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7854 em->block_start == EXTENT_MAP_INLINE) {
7855 free_extent_map(em);
7856 ret = -ENOTBLK;
7857 goto unlock_err;
7858 }
7859
7860 /* Just a good old fashioned hole, return */
7861 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7862 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7863 free_extent_map(em);
7864 goto unlock_err;
7865 }
7866
7867 /*
7868 * We don't allocate a new extent in the following cases
7869 *
7870 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7871 * existing extent.
7872 * 2) The extent is marked as PREALLOC. We're good to go here and can
7873 * just use the extent.
7874 *
7875 */
7876 if (!create) {
7877 len = min(len, em->len - (start - em->start));
7878 lockstart = start + len;
7879 goto unlock;
7880 }
7881
7882 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7883 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7884 em->block_start != EXTENT_MAP_HOLE)) {
7885 int type;
7886 u64 block_start, orig_start, orig_block_len, ram_bytes;
7887
7888 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7889 type = BTRFS_ORDERED_PREALLOC;
7890 else
7891 type = BTRFS_ORDERED_NOCOW;
7892 len = min(len, em->len - (start - em->start));
7893 block_start = em->block_start + (start - em->start);
7894
7895 if (can_nocow_extent(inode, start, &len, &orig_start,
7896 &orig_block_len, &ram_bytes) == 1 &&
7897 btrfs_inc_nocow_writers(fs_info, block_start)) {
7898 struct extent_map *em2;
7899
7900 em2 = btrfs_create_dio_extent(inode, start, len,
7901 orig_start, block_start,
7902 len, orig_block_len,
7903 ram_bytes, type);
7904 btrfs_dec_nocow_writers(fs_info, block_start);
7905 if (type == BTRFS_ORDERED_PREALLOC) {
7906 free_extent_map(em);
7907 em = em2;
7908 }
7909 if (em2 && IS_ERR(em2)) {
7910 ret = PTR_ERR(em2);
7911 goto unlock_err;
7912 }
7913 /*
7914 * For inode marked NODATACOW or extent marked PREALLOC,
7915 * use the existing or preallocated extent, so does not
7916 * need to adjust btrfs_space_info's bytes_may_use.
7917 */
7918 btrfs_free_reserved_data_space_noquota(inode,
7919 start, len);
7920 goto unlock;
7921 }
7922 }
7923
7924 /*
7925 * this will cow the extent, reset the len in case we changed
7926 * it above
7927 */
7928 len = bh_result->b_size;
7929 free_extent_map(em);
7930 em = btrfs_new_extent_direct(inode, start, len);
7931 if (IS_ERR(em)) {
7932 ret = PTR_ERR(em);
7933 goto unlock_err;
7934 }
7935 len = min(len, em->len - (start - em->start));
7936 unlock:
7937 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7938 inode->i_blkbits;
7939 bh_result->b_size = len;
7940 bh_result->b_bdev = em->bdev;
7941 set_buffer_mapped(bh_result);
7942 if (create) {
7943 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7944 set_buffer_new(bh_result);
7945
7946 /*
7947 * Need to update the i_size under the extent lock so buffered
7948 * readers will get the updated i_size when we unlock.
7949 */
7950 if (!dio_data->overwrite && start + len > i_size_read(inode))
7951 i_size_write(inode, start + len);
7952
7953 WARN_ON(dio_data->reserve < len);
7954 dio_data->reserve -= len;
7955 dio_data->unsubmitted_oe_range_end = start + len;
7956 current->journal_info = dio_data;
7957 }
7958
7959 /*
7960 * In the case of write we need to clear and unlock the entire range,
7961 * in the case of read we need to unlock only the end area that we
7962 * aren't using if there is any left over space.
7963 */
7964 if (lockstart < lockend) {
7965 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7966 lockend, unlock_bits, 1, 0,
7967 &cached_state, GFP_NOFS);
7968 } else {
7969 free_extent_state(cached_state);
7970 }
7971
7972 free_extent_map(em);
7973
7974 return 0;
7975
7976 unlock_err:
7977 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7978 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7979 err:
7980 if (dio_data)
7981 current->journal_info = dio_data;
7982 return ret;
7983 }
7984
7985 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7986 struct bio *bio,
7987 int mirror_num)
7988 {
7989 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7990 blk_status_t ret;
7991
7992 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7993
7994 bio_get(bio);
7995
7996 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7997 if (ret)
7998 goto err;
7999
8000 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
8001 err:
8002 bio_put(bio);
8003 return ret;
8004 }
8005
8006 static int btrfs_check_dio_repairable(struct inode *inode,
8007 struct bio *failed_bio,
8008 struct io_failure_record *failrec,
8009 int failed_mirror)
8010 {
8011 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8012 int num_copies;
8013
8014 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
8015 if (num_copies == 1) {
8016 /*
8017 * we only have a single copy of the data, so don't bother with
8018 * all the retry and error correction code that follows. no
8019 * matter what the error is, it is very likely to persist.
8020 */
8021 btrfs_debug(fs_info,
8022 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
8023 num_copies, failrec->this_mirror, failed_mirror);
8024 return 0;
8025 }
8026
8027 failrec->failed_mirror = failed_mirror;
8028 failrec->this_mirror++;
8029 if (failrec->this_mirror == failed_mirror)
8030 failrec->this_mirror++;
8031
8032 if (failrec->this_mirror > num_copies) {
8033 btrfs_debug(fs_info,
8034 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8035 num_copies, failrec->this_mirror, failed_mirror);
8036 return 0;
8037 }
8038
8039 return 1;
8040 }
8041
8042 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
8043 struct page *page, unsigned int pgoff,
8044 u64 start, u64 end, int failed_mirror,
8045 bio_end_io_t *repair_endio, void *repair_arg)
8046 {
8047 struct io_failure_record *failrec;
8048 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8049 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8050 struct bio *bio;
8051 int isector;
8052 unsigned int read_mode = 0;
8053 int segs;
8054 int ret;
8055 blk_status_t status;
8056
8057 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8058
8059 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8060 if (ret)
8061 return errno_to_blk_status(ret);
8062
8063 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8064 failed_mirror);
8065 if (!ret) {
8066 free_io_failure(failure_tree, io_tree, failrec);
8067 return BLK_STS_IOERR;
8068 }
8069
8070 segs = bio_segments(failed_bio);
8071 if (segs > 1 ||
8072 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8073 read_mode |= REQ_FAILFAST_DEV;
8074
8075 isector = start - btrfs_io_bio(failed_bio)->logical;
8076 isector >>= inode->i_sb->s_blocksize_bits;
8077 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8078 pgoff, isector, repair_endio, repair_arg);
8079 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8080
8081 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8082 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8083 read_mode, failrec->this_mirror, failrec->in_validation);
8084
8085 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8086 if (status) {
8087 free_io_failure(failure_tree, io_tree, failrec);
8088 bio_put(bio);
8089 }
8090
8091 return status;
8092 }
8093
8094 struct btrfs_retry_complete {
8095 struct completion done;
8096 struct inode *inode;
8097 u64 start;
8098 int uptodate;
8099 };
8100
8101 static void btrfs_retry_endio_nocsum(struct bio *bio)
8102 {
8103 struct btrfs_retry_complete *done = bio->bi_private;
8104 struct inode *inode = done->inode;
8105 struct bio_vec *bvec;
8106 struct extent_io_tree *io_tree, *failure_tree;
8107 int i;
8108
8109 if (bio->bi_status)
8110 goto end;
8111
8112 ASSERT(bio->bi_vcnt == 1);
8113 io_tree = &BTRFS_I(inode)->io_tree;
8114 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8115 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8116
8117 done->uptodate = 1;
8118 ASSERT(!bio_flagged(bio, BIO_CLONED));
8119 bio_for_each_segment_all(bvec, bio, i)
8120 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8121 io_tree, done->start, bvec->bv_page,
8122 btrfs_ino(BTRFS_I(inode)), 0);
8123 end:
8124 complete(&done->done);
8125 bio_put(bio);
8126 }
8127
8128 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8129 struct btrfs_io_bio *io_bio)
8130 {
8131 struct btrfs_fs_info *fs_info;
8132 struct bio_vec bvec;
8133 struct bvec_iter iter;
8134 struct btrfs_retry_complete done;
8135 u64 start;
8136 unsigned int pgoff;
8137 u32 sectorsize;
8138 int nr_sectors;
8139 blk_status_t ret;
8140 blk_status_t err = BLK_STS_OK;
8141
8142 fs_info = BTRFS_I(inode)->root->fs_info;
8143 sectorsize = fs_info->sectorsize;
8144
8145 start = io_bio->logical;
8146 done.inode = inode;
8147 io_bio->bio.bi_iter = io_bio->iter;
8148
8149 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8150 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8151 pgoff = bvec.bv_offset;
8152
8153 next_block_or_try_again:
8154 done.uptodate = 0;
8155 done.start = start;
8156 init_completion(&done.done);
8157
8158 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8159 pgoff, start, start + sectorsize - 1,
8160 io_bio->mirror_num,
8161 btrfs_retry_endio_nocsum, &done);
8162 if (ret) {
8163 err = ret;
8164 goto next;
8165 }
8166
8167 wait_for_completion_io(&done.done);
8168
8169 if (!done.uptodate) {
8170 /* We might have another mirror, so try again */
8171 goto next_block_or_try_again;
8172 }
8173
8174 next:
8175 start += sectorsize;
8176
8177 nr_sectors--;
8178 if (nr_sectors) {
8179 pgoff += sectorsize;
8180 ASSERT(pgoff < PAGE_SIZE);
8181 goto next_block_or_try_again;
8182 }
8183 }
8184
8185 return err;
8186 }
8187
8188 static void btrfs_retry_endio(struct bio *bio)
8189 {
8190 struct btrfs_retry_complete *done = bio->bi_private;
8191 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8192 struct extent_io_tree *io_tree, *failure_tree;
8193 struct inode *inode = done->inode;
8194 struct bio_vec *bvec;
8195 int uptodate;
8196 int ret;
8197 int i;
8198
8199 if (bio->bi_status)
8200 goto end;
8201
8202 uptodate = 1;
8203
8204 ASSERT(bio->bi_vcnt == 1);
8205 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8206
8207 io_tree = &BTRFS_I(inode)->io_tree;
8208 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8209
8210 ASSERT(!bio_flagged(bio, BIO_CLONED));
8211 bio_for_each_segment_all(bvec, bio, i) {
8212 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8213 bvec->bv_offset, done->start,
8214 bvec->bv_len);
8215 if (!ret)
8216 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8217 failure_tree, io_tree, done->start,
8218 bvec->bv_page,
8219 btrfs_ino(BTRFS_I(inode)),
8220 bvec->bv_offset);
8221 else
8222 uptodate = 0;
8223 }
8224
8225 done->uptodate = uptodate;
8226 end:
8227 complete(&done->done);
8228 bio_put(bio);
8229 }
8230
8231 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8232 struct btrfs_io_bio *io_bio, blk_status_t err)
8233 {
8234 struct btrfs_fs_info *fs_info;
8235 struct bio_vec bvec;
8236 struct bvec_iter iter;
8237 struct btrfs_retry_complete done;
8238 u64 start;
8239 u64 offset = 0;
8240 u32 sectorsize;
8241 int nr_sectors;
8242 unsigned int pgoff;
8243 int csum_pos;
8244 bool uptodate = (err == 0);
8245 int ret;
8246 blk_status_t status;
8247
8248 fs_info = BTRFS_I(inode)->root->fs_info;
8249 sectorsize = fs_info->sectorsize;
8250
8251 err = BLK_STS_OK;
8252 start = io_bio->logical;
8253 done.inode = inode;
8254 io_bio->bio.bi_iter = io_bio->iter;
8255
8256 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8257 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8258
8259 pgoff = bvec.bv_offset;
8260 next_block:
8261 if (uptodate) {
8262 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8263 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8264 bvec.bv_page, pgoff, start, sectorsize);
8265 if (likely(!ret))
8266 goto next;
8267 }
8268 try_again:
8269 done.uptodate = 0;
8270 done.start = start;
8271 init_completion(&done.done);
8272
8273 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8274 pgoff, start, start + sectorsize - 1,
8275 io_bio->mirror_num, btrfs_retry_endio,
8276 &done);
8277 if (status) {
8278 err = status;
8279 goto next;
8280 }
8281
8282 wait_for_completion_io(&done.done);
8283
8284 if (!done.uptodate) {
8285 /* We might have another mirror, so try again */
8286 goto try_again;
8287 }
8288 next:
8289 offset += sectorsize;
8290 start += sectorsize;
8291
8292 ASSERT(nr_sectors);
8293
8294 nr_sectors--;
8295 if (nr_sectors) {
8296 pgoff += sectorsize;
8297 ASSERT(pgoff < PAGE_SIZE);
8298 goto next_block;
8299 }
8300 }
8301
8302 return err;
8303 }
8304
8305 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8306 struct btrfs_io_bio *io_bio, blk_status_t err)
8307 {
8308 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8309
8310 if (skip_csum) {
8311 if (unlikely(err))
8312 return __btrfs_correct_data_nocsum(inode, io_bio);
8313 else
8314 return BLK_STS_OK;
8315 } else {
8316 return __btrfs_subio_endio_read(inode, io_bio, err);
8317 }
8318 }
8319
8320 static void btrfs_endio_direct_read(struct bio *bio)
8321 {
8322 struct btrfs_dio_private *dip = bio->bi_private;
8323 struct inode *inode = dip->inode;
8324 struct bio *dio_bio;
8325 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8326 blk_status_t err = bio->bi_status;
8327
8328 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8329 err = btrfs_subio_endio_read(inode, io_bio, err);
8330
8331 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8332 dip->logical_offset + dip->bytes - 1);
8333 dio_bio = dip->dio_bio;
8334
8335 kfree(dip);
8336
8337 dio_bio->bi_status = err;
8338 dio_end_io(dio_bio);
8339
8340 if (io_bio->end_io)
8341 io_bio->end_io(io_bio, blk_status_to_errno(err));
8342 bio_put(bio);
8343 }
8344
8345 static void __endio_write_update_ordered(struct inode *inode,
8346 const u64 offset, const u64 bytes,
8347 const bool uptodate)
8348 {
8349 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8350 struct btrfs_ordered_extent *ordered = NULL;
8351 struct btrfs_workqueue *wq;
8352 btrfs_work_func_t func;
8353 u64 ordered_offset = offset;
8354 u64 ordered_bytes = bytes;
8355 u64 last_offset;
8356 int ret;
8357
8358 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8359 wq = fs_info->endio_freespace_worker;
8360 func = btrfs_freespace_write_helper;
8361 } else {
8362 wq = fs_info->endio_write_workers;
8363 func = btrfs_endio_write_helper;
8364 }
8365
8366 again:
8367 last_offset = ordered_offset;
8368 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8369 &ordered_offset,
8370 ordered_bytes,
8371 uptodate);
8372 if (!ret)
8373 goto out_test;
8374
8375 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8376 btrfs_queue_work(wq, &ordered->work);
8377 out_test:
8378 /*
8379 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8380 * in the range, we can exit.
8381 */
8382 if (ordered_offset == last_offset)
8383 return;
8384 /*
8385 * our bio might span multiple ordered extents. If we haven't
8386 * completed the accounting for the whole dio, go back and try again
8387 */
8388 if (ordered_offset < offset + bytes) {
8389 ordered_bytes = offset + bytes - ordered_offset;
8390 ordered = NULL;
8391 goto again;
8392 }
8393 }
8394
8395 static void btrfs_endio_direct_write(struct bio *bio)
8396 {
8397 struct btrfs_dio_private *dip = bio->bi_private;
8398 struct bio *dio_bio = dip->dio_bio;
8399
8400 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8401 dip->bytes, !bio->bi_status);
8402
8403 kfree(dip);
8404
8405 dio_bio->bi_status = bio->bi_status;
8406 dio_end_io(dio_bio);
8407 bio_put(bio);
8408 }
8409
8410 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8411 struct bio *bio, int mirror_num,
8412 unsigned long bio_flags, u64 offset)
8413 {
8414 struct inode *inode = private_data;
8415 blk_status_t ret;
8416 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8417 BUG_ON(ret); /* -ENOMEM */
8418 return 0;
8419 }
8420
8421 static void btrfs_end_dio_bio(struct bio *bio)
8422 {
8423 struct btrfs_dio_private *dip = bio->bi_private;
8424 blk_status_t err = bio->bi_status;
8425
8426 if (err)
8427 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8428 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8429 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8430 bio->bi_opf,
8431 (unsigned long long)bio->bi_iter.bi_sector,
8432 bio->bi_iter.bi_size, err);
8433
8434 if (dip->subio_endio)
8435 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8436
8437 if (err) {
8438 dip->errors = 1;
8439
8440 /*
8441 * before atomic variable goto zero, we must make sure
8442 * dip->errors is perceived to be set.
8443 */
8444 smp_mb__before_atomic();
8445 }
8446
8447 /* if there are more bios still pending for this dio, just exit */
8448 if (!atomic_dec_and_test(&dip->pending_bios))
8449 goto out;
8450
8451 if (dip->errors) {
8452 bio_io_error(dip->orig_bio);
8453 } else {
8454 dip->dio_bio->bi_status = BLK_STS_OK;
8455 bio_endio(dip->orig_bio);
8456 }
8457 out:
8458 bio_put(bio);
8459 }
8460
8461 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8462 struct btrfs_dio_private *dip,
8463 struct bio *bio,
8464 u64 file_offset)
8465 {
8466 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8467 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8468 blk_status_t ret;
8469
8470 /*
8471 * We load all the csum data we need when we submit
8472 * the first bio to reduce the csum tree search and
8473 * contention.
8474 */
8475 if (dip->logical_offset == file_offset) {
8476 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8477 file_offset);
8478 if (ret)
8479 return ret;
8480 }
8481
8482 if (bio == dip->orig_bio)
8483 return 0;
8484
8485 file_offset -= dip->logical_offset;
8486 file_offset >>= inode->i_sb->s_blocksize_bits;
8487 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8488
8489 return 0;
8490 }
8491
8492 static inline blk_status_t
8493 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8494 int async_submit)
8495 {
8496 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8497 struct btrfs_dio_private *dip = bio->bi_private;
8498 bool write = bio_op(bio) == REQ_OP_WRITE;
8499 blk_status_t ret;
8500
8501 if (async_submit)
8502 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8503
8504 bio_get(bio);
8505
8506 if (!write) {
8507 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8508 if (ret)
8509 goto err;
8510 }
8511
8512 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8513 goto map;
8514
8515 if (write && async_submit) {
8516 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8517 file_offset, inode,
8518 __btrfs_submit_bio_start_direct_io,
8519 __btrfs_submit_bio_done);
8520 goto err;
8521 } else if (write) {
8522 /*
8523 * If we aren't doing async submit, calculate the csum of the
8524 * bio now.
8525 */
8526 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8527 if (ret)
8528 goto err;
8529 } else {
8530 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8531 file_offset);
8532 if (ret)
8533 goto err;
8534 }
8535 map:
8536 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8537 err:
8538 bio_put(bio);
8539 return ret;
8540 }
8541
8542 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8543 {
8544 struct inode *inode = dip->inode;
8545 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8546 struct bio *bio;
8547 struct bio *orig_bio = dip->orig_bio;
8548 u64 start_sector = orig_bio->bi_iter.bi_sector;
8549 u64 file_offset = dip->logical_offset;
8550 u64 map_length;
8551 int async_submit = 0;
8552 u64 submit_len;
8553 int clone_offset = 0;
8554 int clone_len;
8555 int ret;
8556 blk_status_t status;
8557
8558 map_length = orig_bio->bi_iter.bi_size;
8559 submit_len = map_length;
8560 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8561 &map_length, NULL, 0);
8562 if (ret)
8563 return -EIO;
8564
8565 if (map_length >= submit_len) {
8566 bio = orig_bio;
8567 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8568 goto submit;
8569 }
8570
8571 /* async crcs make it difficult to collect full stripe writes. */
8572 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8573 async_submit = 0;
8574 else
8575 async_submit = 1;
8576
8577 /* bio split */
8578 ASSERT(map_length <= INT_MAX);
8579 atomic_inc(&dip->pending_bios);
8580 do {
8581 clone_len = min_t(int, submit_len, map_length);
8582
8583 /*
8584 * This will never fail as it's passing GPF_NOFS and
8585 * the allocation is backed by btrfs_bioset.
8586 */
8587 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8588 clone_len);
8589 bio->bi_private = dip;
8590 bio->bi_end_io = btrfs_end_dio_bio;
8591 btrfs_io_bio(bio)->logical = file_offset;
8592
8593 ASSERT(submit_len >= clone_len);
8594 submit_len -= clone_len;
8595 if (submit_len == 0)
8596 break;
8597
8598 /*
8599 * Increase the count before we submit the bio so we know
8600 * the end IO handler won't happen before we increase the
8601 * count. Otherwise, the dip might get freed before we're
8602 * done setting it up.
8603 */
8604 atomic_inc(&dip->pending_bios);
8605
8606 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
8607 async_submit);
8608 if (status) {
8609 bio_put(bio);
8610 atomic_dec(&dip->pending_bios);
8611 goto out_err;
8612 }
8613
8614 clone_offset += clone_len;
8615 start_sector += clone_len >> 9;
8616 file_offset += clone_len;
8617
8618 map_length = submit_len;
8619 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8620 start_sector << 9, &map_length, NULL, 0);
8621 if (ret)
8622 goto out_err;
8623 } while (submit_len > 0);
8624
8625 submit:
8626 status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8627 if (!status)
8628 return 0;
8629
8630 bio_put(bio);
8631 out_err:
8632 dip->errors = 1;
8633 /*
8634 * before atomic variable goto zero, we must
8635 * make sure dip->errors is perceived to be set.
8636 */
8637 smp_mb__before_atomic();
8638 if (atomic_dec_and_test(&dip->pending_bios))
8639 bio_io_error(dip->orig_bio);
8640
8641 /* bio_end_io() will handle error, so we needn't return it */
8642 return 0;
8643 }
8644
8645 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8646 loff_t file_offset)
8647 {
8648 struct btrfs_dio_private *dip = NULL;
8649 struct bio *bio = NULL;
8650 struct btrfs_io_bio *io_bio;
8651 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8652 int ret = 0;
8653
8654 bio = btrfs_bio_clone(dio_bio);
8655
8656 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8657 if (!dip) {
8658 ret = -ENOMEM;
8659 goto free_ordered;
8660 }
8661
8662 dip->private = dio_bio->bi_private;
8663 dip->inode = inode;
8664 dip->logical_offset = file_offset;
8665 dip->bytes = dio_bio->bi_iter.bi_size;
8666 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8667 bio->bi_private = dip;
8668 dip->orig_bio = bio;
8669 dip->dio_bio = dio_bio;
8670 atomic_set(&dip->pending_bios, 0);
8671 io_bio = btrfs_io_bio(bio);
8672 io_bio->logical = file_offset;
8673
8674 if (write) {
8675 bio->bi_end_io = btrfs_endio_direct_write;
8676 } else {
8677 bio->bi_end_io = btrfs_endio_direct_read;
8678 dip->subio_endio = btrfs_subio_endio_read;
8679 }
8680
8681 /*
8682 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8683 * even if we fail to submit a bio, because in such case we do the
8684 * corresponding error handling below and it must not be done a second
8685 * time by btrfs_direct_IO().
8686 */
8687 if (write) {
8688 struct btrfs_dio_data *dio_data = current->journal_info;
8689
8690 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8691 dip->bytes;
8692 dio_data->unsubmitted_oe_range_start =
8693 dio_data->unsubmitted_oe_range_end;
8694 }
8695
8696 ret = btrfs_submit_direct_hook(dip);
8697 if (!ret)
8698 return;
8699
8700 if (io_bio->end_io)
8701 io_bio->end_io(io_bio, ret);
8702
8703 free_ordered:
8704 /*
8705 * If we arrived here it means either we failed to submit the dip
8706 * or we either failed to clone the dio_bio or failed to allocate the
8707 * dip. If we cloned the dio_bio and allocated the dip, we can just
8708 * call bio_endio against our io_bio so that we get proper resource
8709 * cleanup if we fail to submit the dip, otherwise, we must do the
8710 * same as btrfs_endio_direct_[write|read] because we can't call these
8711 * callbacks - they require an allocated dip and a clone of dio_bio.
8712 */
8713 if (bio && dip) {
8714 bio_io_error(bio);
8715 /*
8716 * The end io callbacks free our dip, do the final put on bio
8717 * and all the cleanup and final put for dio_bio (through
8718 * dio_end_io()).
8719 */
8720 dip = NULL;
8721 bio = NULL;
8722 } else {
8723 if (write)
8724 __endio_write_update_ordered(inode,
8725 file_offset,
8726 dio_bio->bi_iter.bi_size,
8727 false);
8728 else
8729 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8730 file_offset + dio_bio->bi_iter.bi_size - 1);
8731
8732 dio_bio->bi_status = BLK_STS_IOERR;
8733 /*
8734 * Releases and cleans up our dio_bio, no need to bio_put()
8735 * nor bio_endio()/bio_io_error() against dio_bio.
8736 */
8737 dio_end_io(dio_bio);
8738 }
8739 if (bio)
8740 bio_put(bio);
8741 kfree(dip);
8742 }
8743
8744 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8745 const struct iov_iter *iter, loff_t offset)
8746 {
8747 int seg;
8748 int i;
8749 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8750 ssize_t retval = -EINVAL;
8751
8752 if (offset & blocksize_mask)
8753 goto out;
8754
8755 if (iov_iter_alignment(iter) & blocksize_mask)
8756 goto out;
8757
8758 /* If this is a write we don't need to check anymore */
8759 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8760 return 0;
8761 /*
8762 * Check to make sure we don't have duplicate iov_base's in this
8763 * iovec, if so return EINVAL, otherwise we'll get csum errors
8764 * when reading back.
8765 */
8766 for (seg = 0; seg < iter->nr_segs; seg++) {
8767 for (i = seg + 1; i < iter->nr_segs; i++) {
8768 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8769 goto out;
8770 }
8771 }
8772 retval = 0;
8773 out:
8774 return retval;
8775 }
8776
8777 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8778 {
8779 struct file *file = iocb->ki_filp;
8780 struct inode *inode = file->f_mapping->host;
8781 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8782 struct btrfs_dio_data dio_data = { 0 };
8783 struct extent_changeset *data_reserved = NULL;
8784 loff_t offset = iocb->ki_pos;
8785 size_t count = 0;
8786 int flags = 0;
8787 bool wakeup = true;
8788 bool relock = false;
8789 ssize_t ret;
8790
8791 if (check_direct_IO(fs_info, iter, offset))
8792 return 0;
8793
8794 inode_dio_begin(inode);
8795
8796 /*
8797 * The generic stuff only does filemap_write_and_wait_range, which
8798 * isn't enough if we've written compressed pages to this area, so
8799 * we need to flush the dirty pages again to make absolutely sure
8800 * that any outstanding dirty pages are on disk.
8801 */
8802 count = iov_iter_count(iter);
8803 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8804 &BTRFS_I(inode)->runtime_flags))
8805 filemap_fdatawrite_range(inode->i_mapping, offset,
8806 offset + count - 1);
8807
8808 if (iov_iter_rw(iter) == WRITE) {
8809 /*
8810 * If the write DIO is beyond the EOF, we need update
8811 * the isize, but it is protected by i_mutex. So we can
8812 * not unlock the i_mutex at this case.
8813 */
8814 if (offset + count <= inode->i_size) {
8815 dio_data.overwrite = 1;
8816 inode_unlock(inode);
8817 relock = true;
8818 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8819 ret = -EAGAIN;
8820 goto out;
8821 }
8822 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8823 offset, count);
8824 if (ret)
8825 goto out;
8826
8827 /*
8828 * We need to know how many extents we reserved so that we can
8829 * do the accounting properly if we go over the number we
8830 * originally calculated. Abuse current->journal_info for this.
8831 */
8832 dio_data.reserve = round_up(count,
8833 fs_info->sectorsize);
8834 dio_data.unsubmitted_oe_range_start = (u64)offset;
8835 dio_data.unsubmitted_oe_range_end = (u64)offset;
8836 current->journal_info = &dio_data;
8837 down_read(&BTRFS_I(inode)->dio_sem);
8838 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8839 &BTRFS_I(inode)->runtime_flags)) {
8840 inode_dio_end(inode);
8841 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8842 wakeup = false;
8843 }
8844
8845 ret = __blockdev_direct_IO(iocb, inode,
8846 fs_info->fs_devices->latest_bdev,
8847 iter, btrfs_get_blocks_direct, NULL,
8848 btrfs_submit_direct, flags);
8849 if (iov_iter_rw(iter) == WRITE) {
8850 up_read(&BTRFS_I(inode)->dio_sem);
8851 current->journal_info = NULL;
8852 if (ret < 0 && ret != -EIOCBQUEUED) {
8853 if (dio_data.reserve)
8854 btrfs_delalloc_release_space(inode, data_reserved,
8855 offset, dio_data.reserve);
8856 /*
8857 * On error we might have left some ordered extents
8858 * without submitting corresponding bios for them, so
8859 * cleanup them up to avoid other tasks getting them
8860 * and waiting for them to complete forever.
8861 */
8862 if (dio_data.unsubmitted_oe_range_start <
8863 dio_data.unsubmitted_oe_range_end)
8864 __endio_write_update_ordered(inode,
8865 dio_data.unsubmitted_oe_range_start,
8866 dio_data.unsubmitted_oe_range_end -
8867 dio_data.unsubmitted_oe_range_start,
8868 false);
8869 } else if (ret >= 0 && (size_t)ret < count)
8870 btrfs_delalloc_release_space(inode, data_reserved,
8871 offset, count - (size_t)ret);
8872 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8873 }
8874 out:
8875 if (wakeup)
8876 inode_dio_end(inode);
8877 if (relock)
8878 inode_lock(inode);
8879
8880 extent_changeset_free(data_reserved);
8881 return ret;
8882 }
8883
8884 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8885
8886 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8887 __u64 start, __u64 len)
8888 {
8889 int ret;
8890
8891 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8892 if (ret)
8893 return ret;
8894
8895 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8896 }
8897
8898 int btrfs_readpage(struct file *file, struct page *page)
8899 {
8900 struct extent_io_tree *tree;
8901 tree = &BTRFS_I(page->mapping->host)->io_tree;
8902 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8903 }
8904
8905 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8906 {
8907 struct extent_io_tree *tree;
8908 struct inode *inode = page->mapping->host;
8909 int ret;
8910
8911 if (current->flags & PF_MEMALLOC) {
8912 redirty_page_for_writepage(wbc, page);
8913 unlock_page(page);
8914 return 0;
8915 }
8916
8917 /*
8918 * If we are under memory pressure we will call this directly from the
8919 * VM, we need to make sure we have the inode referenced for the ordered
8920 * extent. If not just return like we didn't do anything.
8921 */
8922 if (!igrab(inode)) {
8923 redirty_page_for_writepage(wbc, page);
8924 return AOP_WRITEPAGE_ACTIVATE;
8925 }
8926 tree = &BTRFS_I(page->mapping->host)->io_tree;
8927 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8928 btrfs_add_delayed_iput(inode);
8929 return ret;
8930 }
8931
8932 static int btrfs_writepages(struct address_space *mapping,
8933 struct writeback_control *wbc)
8934 {
8935 struct extent_io_tree *tree;
8936
8937 tree = &BTRFS_I(mapping->host)->io_tree;
8938 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8939 }
8940
8941 static int
8942 btrfs_readpages(struct file *file, struct address_space *mapping,
8943 struct list_head *pages, unsigned nr_pages)
8944 {
8945 struct extent_io_tree *tree;
8946 tree = &BTRFS_I(mapping->host)->io_tree;
8947 return extent_readpages(tree, mapping, pages, nr_pages,
8948 btrfs_get_extent);
8949 }
8950 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8951 {
8952 struct extent_io_tree *tree;
8953 struct extent_map_tree *map;
8954 int ret;
8955
8956 tree = &BTRFS_I(page->mapping->host)->io_tree;
8957 map = &BTRFS_I(page->mapping->host)->extent_tree;
8958 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8959 if (ret == 1) {
8960 ClearPagePrivate(page);
8961 set_page_private(page, 0);
8962 put_page(page);
8963 }
8964 return ret;
8965 }
8966
8967 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8968 {
8969 if (PageWriteback(page) || PageDirty(page))
8970 return 0;
8971 return __btrfs_releasepage(page, gfp_flags);
8972 }
8973
8974 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8975 unsigned int length)
8976 {
8977 struct inode *inode = page->mapping->host;
8978 struct extent_io_tree *tree;
8979 struct btrfs_ordered_extent *ordered;
8980 struct extent_state *cached_state = NULL;
8981 u64 page_start = page_offset(page);
8982 u64 page_end = page_start + PAGE_SIZE - 1;
8983 u64 start;
8984 u64 end;
8985 int inode_evicting = inode->i_state & I_FREEING;
8986
8987 /*
8988 * we have the page locked, so new writeback can't start,
8989 * and the dirty bit won't be cleared while we are here.
8990 *
8991 * Wait for IO on this page so that we can safely clear
8992 * the PagePrivate2 bit and do ordered accounting
8993 */
8994 wait_on_page_writeback(page);
8995
8996 tree = &BTRFS_I(inode)->io_tree;
8997 if (offset) {
8998 btrfs_releasepage(page, GFP_NOFS);
8999 return;
9000 }
9001
9002 if (!inode_evicting)
9003 lock_extent_bits(tree, page_start, page_end, &cached_state);
9004 again:
9005 start = page_start;
9006 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
9007 page_end - start + 1);
9008 if (ordered) {
9009 end = min(page_end, ordered->file_offset + ordered->len - 1);
9010 /*
9011 * IO on this page will never be started, so we need
9012 * to account for any ordered extents now
9013 */
9014 if (!inode_evicting)
9015 clear_extent_bit(tree, start, end,
9016 EXTENT_DIRTY | EXTENT_DELALLOC |
9017 EXTENT_DELALLOC_NEW |
9018 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
9019 EXTENT_DEFRAG, 1, 0, &cached_state,
9020 GFP_NOFS);
9021 /*
9022 * whoever cleared the private bit is responsible
9023 * for the finish_ordered_io
9024 */
9025 if (TestClearPagePrivate2(page)) {
9026 struct btrfs_ordered_inode_tree *tree;
9027 u64 new_len;
9028
9029 tree = &BTRFS_I(inode)->ordered_tree;
9030
9031 spin_lock_irq(&tree->lock);
9032 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
9033 new_len = start - ordered->file_offset;
9034 if (new_len < ordered->truncated_len)
9035 ordered->truncated_len = new_len;
9036 spin_unlock_irq(&tree->lock);
9037
9038 if (btrfs_dec_test_ordered_pending(inode, &ordered,
9039 start,
9040 end - start + 1, 1))
9041 btrfs_finish_ordered_io(ordered);
9042 }
9043 btrfs_put_ordered_extent(ordered);
9044 if (!inode_evicting) {
9045 cached_state = NULL;
9046 lock_extent_bits(tree, start, end,
9047 &cached_state);
9048 }
9049
9050 start = end + 1;
9051 if (start < page_end)
9052 goto again;
9053 }
9054
9055 /*
9056 * Qgroup reserved space handler
9057 * Page here will be either
9058 * 1) Already written to disk
9059 * In this case, its reserved space is released from data rsv map
9060 * and will be freed by delayed_ref handler finally.
9061 * So even we call qgroup_free_data(), it won't decrease reserved
9062 * space.
9063 * 2) Not written to disk
9064 * This means the reserved space should be freed here. However,
9065 * if a truncate invalidates the page (by clearing PageDirty)
9066 * and the page is accounted for while allocating extent
9067 * in btrfs_check_data_free_space() we let delayed_ref to
9068 * free the entire extent.
9069 */
9070 if (PageDirty(page))
9071 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9072 if (!inode_evicting) {
9073 clear_extent_bit(tree, page_start, page_end,
9074 EXTENT_LOCKED | EXTENT_DIRTY |
9075 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9076 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9077 &cached_state, GFP_NOFS);
9078
9079 __btrfs_releasepage(page, GFP_NOFS);
9080 }
9081
9082 ClearPageChecked(page);
9083 if (PagePrivate(page)) {
9084 ClearPagePrivate(page);
9085 set_page_private(page, 0);
9086 put_page(page);
9087 }
9088 }
9089
9090 /*
9091 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9092 * called from a page fault handler when a page is first dirtied. Hence we must
9093 * be careful to check for EOF conditions here. We set the page up correctly
9094 * for a written page which means we get ENOSPC checking when writing into
9095 * holes and correct delalloc and unwritten extent mapping on filesystems that
9096 * support these features.
9097 *
9098 * We are not allowed to take the i_mutex here so we have to play games to
9099 * protect against truncate races as the page could now be beyond EOF. Because
9100 * vmtruncate() writes the inode size before removing pages, once we have the
9101 * page lock we can determine safely if the page is beyond EOF. If it is not
9102 * beyond EOF, then the page is guaranteed safe against truncation until we
9103 * unlock the page.
9104 */
9105 int btrfs_page_mkwrite(struct vm_fault *vmf)
9106 {
9107 struct page *page = vmf->page;
9108 struct inode *inode = file_inode(vmf->vma->vm_file);
9109 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9110 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9111 struct btrfs_ordered_extent *ordered;
9112 struct extent_state *cached_state = NULL;
9113 struct extent_changeset *data_reserved = NULL;
9114 char *kaddr;
9115 unsigned long zero_start;
9116 loff_t size;
9117 int ret;
9118 int reserved = 0;
9119 u64 reserved_space;
9120 u64 page_start;
9121 u64 page_end;
9122 u64 end;
9123
9124 reserved_space = PAGE_SIZE;
9125
9126 sb_start_pagefault(inode->i_sb);
9127 page_start = page_offset(page);
9128 page_end = page_start + PAGE_SIZE - 1;
9129 end = page_end;
9130
9131 /*
9132 * Reserving delalloc space after obtaining the page lock can lead to
9133 * deadlock. For example, if a dirty page is locked by this function
9134 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9135 * dirty page write out, then the btrfs_writepage() function could
9136 * end up waiting indefinitely to get a lock on the page currently
9137 * being processed by btrfs_page_mkwrite() function.
9138 */
9139 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9140 reserved_space);
9141 if (!ret) {
9142 ret = file_update_time(vmf->vma->vm_file);
9143 reserved = 1;
9144 }
9145 if (ret) {
9146 if (ret == -ENOMEM)
9147 ret = VM_FAULT_OOM;
9148 else /* -ENOSPC, -EIO, etc */
9149 ret = VM_FAULT_SIGBUS;
9150 if (reserved)
9151 goto out;
9152 goto out_noreserve;
9153 }
9154
9155 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9156 again:
9157 lock_page(page);
9158 size = i_size_read(inode);
9159
9160 if ((page->mapping != inode->i_mapping) ||
9161 (page_start >= size)) {
9162 /* page got truncated out from underneath us */
9163 goto out_unlock;
9164 }
9165 wait_on_page_writeback(page);
9166
9167 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9168 set_page_extent_mapped(page);
9169
9170 /*
9171 * we can't set the delalloc bits if there are pending ordered
9172 * extents. Drop our locks and wait for them to finish
9173 */
9174 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9175 PAGE_SIZE);
9176 if (ordered) {
9177 unlock_extent_cached(io_tree, page_start, page_end,
9178 &cached_state, GFP_NOFS);
9179 unlock_page(page);
9180 btrfs_start_ordered_extent(inode, ordered, 1);
9181 btrfs_put_ordered_extent(ordered);
9182 goto again;
9183 }
9184
9185 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9186 reserved_space = round_up(size - page_start,
9187 fs_info->sectorsize);
9188 if (reserved_space < PAGE_SIZE) {
9189 end = page_start + reserved_space - 1;
9190 btrfs_delalloc_release_space(inode, data_reserved,
9191 page_start, PAGE_SIZE - reserved_space);
9192 }
9193 }
9194
9195 /*
9196 * page_mkwrite gets called when the page is firstly dirtied after it's
9197 * faulted in, but write(2) could also dirty a page and set delalloc
9198 * bits, thus in this case for space account reason, we still need to
9199 * clear any delalloc bits within this page range since we have to
9200 * reserve data&meta space before lock_page() (see above comments).
9201 */
9202 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9203 EXTENT_DIRTY | EXTENT_DELALLOC |
9204 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9205 0, 0, &cached_state, GFP_NOFS);
9206
9207 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
9208 &cached_state, 0);
9209 if (ret) {
9210 unlock_extent_cached(io_tree, page_start, page_end,
9211 &cached_state, GFP_NOFS);
9212 ret = VM_FAULT_SIGBUS;
9213 goto out_unlock;
9214 }
9215 ret = 0;
9216
9217 /* page is wholly or partially inside EOF */
9218 if (page_start + PAGE_SIZE > size)
9219 zero_start = size & ~PAGE_MASK;
9220 else
9221 zero_start = PAGE_SIZE;
9222
9223 if (zero_start != PAGE_SIZE) {
9224 kaddr = kmap(page);
9225 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9226 flush_dcache_page(page);
9227 kunmap(page);
9228 }
9229 ClearPageChecked(page);
9230 set_page_dirty(page);
9231 SetPageUptodate(page);
9232
9233 BTRFS_I(inode)->last_trans = fs_info->generation;
9234 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9235 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9236
9237 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9238
9239 out_unlock:
9240 if (!ret) {
9241 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9242 sb_end_pagefault(inode->i_sb);
9243 extent_changeset_free(data_reserved);
9244 return VM_FAULT_LOCKED;
9245 }
9246 unlock_page(page);
9247 out:
9248 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9249 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9250 reserved_space);
9251 out_noreserve:
9252 sb_end_pagefault(inode->i_sb);
9253 extent_changeset_free(data_reserved);
9254 return ret;
9255 }
9256
9257 static int btrfs_truncate(struct inode *inode)
9258 {
9259 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9260 struct btrfs_root *root = BTRFS_I(inode)->root;
9261 struct btrfs_block_rsv *rsv;
9262 int ret = 0;
9263 int err = 0;
9264 struct btrfs_trans_handle *trans;
9265 u64 mask = fs_info->sectorsize - 1;
9266 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9267
9268 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9269 (u64)-1);
9270 if (ret)
9271 return ret;
9272
9273 /*
9274 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9275 * 3 things going on here
9276 *
9277 * 1) We need to reserve space for our orphan item and the space to
9278 * delete our orphan item. Lord knows we don't want to have a dangling
9279 * orphan item because we didn't reserve space to remove it.
9280 *
9281 * 2) We need to reserve space to update our inode.
9282 *
9283 * 3) We need to have something to cache all the space that is going to
9284 * be free'd up by the truncate operation, but also have some slack
9285 * space reserved in case it uses space during the truncate (thank you
9286 * very much snapshotting).
9287 *
9288 * And we need these to all be separate. The fact is we can use a lot of
9289 * space doing the truncate, and we have no earthly idea how much space
9290 * we will use, so we need the truncate reservation to be separate so it
9291 * doesn't end up using space reserved for updating the inode or
9292 * removing the orphan item. We also need to be able to stop the
9293 * transaction and start a new one, which means we need to be able to
9294 * update the inode several times, and we have no idea of knowing how
9295 * many times that will be, so we can't just reserve 1 item for the
9296 * entirety of the operation, so that has to be done separately as well.
9297 * Then there is the orphan item, which does indeed need to be held on
9298 * to for the whole operation, and we need nobody to touch this reserved
9299 * space except the orphan code.
9300 *
9301 * So that leaves us with
9302 *
9303 * 1) root->orphan_block_rsv - for the orphan deletion.
9304 * 2) rsv - for the truncate reservation, which we will steal from the
9305 * transaction reservation.
9306 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9307 * updating the inode.
9308 */
9309 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9310 if (!rsv)
9311 return -ENOMEM;
9312 rsv->size = min_size;
9313 rsv->failfast = 1;
9314
9315 /*
9316 * 1 for the truncate slack space
9317 * 1 for updating the inode.
9318 */
9319 trans = btrfs_start_transaction(root, 2);
9320 if (IS_ERR(trans)) {
9321 err = PTR_ERR(trans);
9322 goto out;
9323 }
9324
9325 /* Migrate the slack space for the truncate to our reserve */
9326 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9327 min_size, 0);
9328 BUG_ON(ret);
9329
9330 /*
9331 * So if we truncate and then write and fsync we normally would just
9332 * write the extents that changed, which is a problem if we need to
9333 * first truncate that entire inode. So set this flag so we write out
9334 * all of the extents in the inode to the sync log so we're completely
9335 * safe.
9336 */
9337 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9338 trans->block_rsv = rsv;
9339
9340 while (1) {
9341 ret = btrfs_truncate_inode_items(trans, root, inode,
9342 inode->i_size,
9343 BTRFS_EXTENT_DATA_KEY);
9344 trans->block_rsv = &fs_info->trans_block_rsv;
9345 if (ret != -ENOSPC && ret != -EAGAIN) {
9346 err = ret;
9347 break;
9348 }
9349
9350 ret = btrfs_update_inode(trans, root, inode);
9351 if (ret) {
9352 err = ret;
9353 break;
9354 }
9355
9356 btrfs_end_transaction(trans);
9357 btrfs_btree_balance_dirty(fs_info);
9358
9359 trans = btrfs_start_transaction(root, 2);
9360 if (IS_ERR(trans)) {
9361 ret = err = PTR_ERR(trans);
9362 trans = NULL;
9363 break;
9364 }
9365
9366 btrfs_block_rsv_release(fs_info, rsv, -1);
9367 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9368 rsv, min_size, 0);
9369 BUG_ON(ret); /* shouldn't happen */
9370 trans->block_rsv = rsv;
9371 }
9372
9373 /*
9374 * We can't call btrfs_truncate_block inside a trans handle as we could
9375 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9376 * we've truncated everything except the last little bit, and can do
9377 * btrfs_truncate_block and then update the disk_i_size.
9378 */
9379 if (ret == NEED_TRUNCATE_BLOCK) {
9380 btrfs_end_transaction(trans);
9381 btrfs_btree_balance_dirty(fs_info);
9382
9383 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9384 if (ret)
9385 goto out;
9386 trans = btrfs_start_transaction(root, 1);
9387 if (IS_ERR(trans)) {
9388 ret = PTR_ERR(trans);
9389 goto out;
9390 }
9391 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9392 }
9393
9394 if (ret == 0 && inode->i_nlink > 0) {
9395 trans->block_rsv = root->orphan_block_rsv;
9396 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9397 if (ret)
9398 err = ret;
9399 }
9400
9401 if (trans) {
9402 trans->block_rsv = &fs_info->trans_block_rsv;
9403 ret = btrfs_update_inode(trans, root, inode);
9404 if (ret && !err)
9405 err = ret;
9406
9407 ret = btrfs_end_transaction(trans);
9408 btrfs_btree_balance_dirty(fs_info);
9409 }
9410 out:
9411 btrfs_free_block_rsv(fs_info, rsv);
9412
9413 if (ret && !err)
9414 err = ret;
9415
9416 return err;
9417 }
9418
9419 /*
9420 * create a new subvolume directory/inode (helper for the ioctl).
9421 */
9422 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9423 struct btrfs_root *new_root,
9424 struct btrfs_root *parent_root,
9425 u64 new_dirid)
9426 {
9427 struct inode *inode;
9428 int err;
9429 u64 index = 0;
9430
9431 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9432 new_dirid, new_dirid,
9433 S_IFDIR | (~current_umask() & S_IRWXUGO),
9434 &index);
9435 if (IS_ERR(inode))
9436 return PTR_ERR(inode);
9437 inode->i_op = &btrfs_dir_inode_operations;
9438 inode->i_fop = &btrfs_dir_file_operations;
9439
9440 set_nlink(inode, 1);
9441 btrfs_i_size_write(BTRFS_I(inode), 0);
9442 unlock_new_inode(inode);
9443
9444 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9445 if (err)
9446 btrfs_err(new_root->fs_info,
9447 "error inheriting subvolume %llu properties: %d",
9448 new_root->root_key.objectid, err);
9449
9450 err = btrfs_update_inode(trans, new_root, inode);
9451
9452 iput(inode);
9453 return err;
9454 }
9455
9456 struct inode *btrfs_alloc_inode(struct super_block *sb)
9457 {
9458 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9459 struct btrfs_inode *ei;
9460 struct inode *inode;
9461
9462 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9463 if (!ei)
9464 return NULL;
9465
9466 ei->root = NULL;
9467 ei->generation = 0;
9468 ei->last_trans = 0;
9469 ei->last_sub_trans = 0;
9470 ei->logged_trans = 0;
9471 ei->delalloc_bytes = 0;
9472 ei->new_delalloc_bytes = 0;
9473 ei->defrag_bytes = 0;
9474 ei->disk_i_size = 0;
9475 ei->flags = 0;
9476 ei->csum_bytes = 0;
9477 ei->index_cnt = (u64)-1;
9478 ei->dir_index = 0;
9479 ei->last_unlink_trans = 0;
9480 ei->last_log_commit = 0;
9481 ei->delayed_iput_count = 0;
9482
9483 spin_lock_init(&ei->lock);
9484 ei->outstanding_extents = 0;
9485 if (sb->s_magic != BTRFS_TEST_MAGIC)
9486 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9487 BTRFS_BLOCK_RSV_DELALLOC);
9488 ei->runtime_flags = 0;
9489 ei->prop_compress = BTRFS_COMPRESS_NONE;
9490 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9491
9492 ei->delayed_node = NULL;
9493
9494 ei->i_otime.tv_sec = 0;
9495 ei->i_otime.tv_nsec = 0;
9496
9497 inode = &ei->vfs_inode;
9498 extent_map_tree_init(&ei->extent_tree);
9499 extent_io_tree_init(&ei->io_tree, inode);
9500 extent_io_tree_init(&ei->io_failure_tree, inode);
9501 ei->io_tree.track_uptodate = 1;
9502 ei->io_failure_tree.track_uptodate = 1;
9503 atomic_set(&ei->sync_writers, 0);
9504 mutex_init(&ei->log_mutex);
9505 mutex_init(&ei->delalloc_mutex);
9506 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9507 INIT_LIST_HEAD(&ei->delalloc_inodes);
9508 INIT_LIST_HEAD(&ei->delayed_iput);
9509 RB_CLEAR_NODE(&ei->rb_node);
9510 init_rwsem(&ei->dio_sem);
9511
9512 return inode;
9513 }
9514
9515 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9516 void btrfs_test_destroy_inode(struct inode *inode)
9517 {
9518 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9519 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9520 }
9521 #endif
9522
9523 static void btrfs_i_callback(struct rcu_head *head)
9524 {
9525 struct inode *inode = container_of(head, struct inode, i_rcu);
9526 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9527 }
9528
9529 void btrfs_destroy_inode(struct inode *inode)
9530 {
9531 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9532 struct btrfs_ordered_extent *ordered;
9533 struct btrfs_root *root = BTRFS_I(inode)->root;
9534
9535 WARN_ON(!hlist_empty(&inode->i_dentry));
9536 WARN_ON(inode->i_data.nrpages);
9537 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9538 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9539 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9540 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9541 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9542 WARN_ON(BTRFS_I(inode)->csum_bytes);
9543 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9544
9545 /*
9546 * This can happen where we create an inode, but somebody else also
9547 * created the same inode and we need to destroy the one we already
9548 * created.
9549 */
9550 if (!root)
9551 goto free;
9552
9553 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9554 &BTRFS_I(inode)->runtime_flags)) {
9555 btrfs_info(fs_info, "inode %llu still on the orphan list",
9556 btrfs_ino(BTRFS_I(inode)));
9557 atomic_dec(&root->orphan_inodes);
9558 }
9559
9560 while (1) {
9561 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9562 if (!ordered)
9563 break;
9564 else {
9565 btrfs_err(fs_info,
9566 "found ordered extent %llu %llu on inode cleanup",
9567 ordered->file_offset, ordered->len);
9568 btrfs_remove_ordered_extent(inode, ordered);
9569 btrfs_put_ordered_extent(ordered);
9570 btrfs_put_ordered_extent(ordered);
9571 }
9572 }
9573 btrfs_qgroup_check_reserved_leak(inode);
9574 inode_tree_del(inode);
9575 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9576 free:
9577 call_rcu(&inode->i_rcu, btrfs_i_callback);
9578 }
9579
9580 int btrfs_drop_inode(struct inode *inode)
9581 {
9582 struct btrfs_root *root = BTRFS_I(inode)->root;
9583
9584 if (root == NULL)
9585 return 1;
9586
9587 /* the snap/subvol tree is on deleting */
9588 if (btrfs_root_refs(&root->root_item) == 0)
9589 return 1;
9590 else
9591 return generic_drop_inode(inode);
9592 }
9593
9594 static void init_once(void *foo)
9595 {
9596 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9597
9598 inode_init_once(&ei->vfs_inode);
9599 }
9600
9601 void btrfs_destroy_cachep(void)
9602 {
9603 /*
9604 * Make sure all delayed rcu free inodes are flushed before we
9605 * destroy cache.
9606 */
9607 rcu_barrier();
9608 kmem_cache_destroy(btrfs_inode_cachep);
9609 kmem_cache_destroy(btrfs_trans_handle_cachep);
9610 kmem_cache_destroy(btrfs_path_cachep);
9611 kmem_cache_destroy(btrfs_free_space_cachep);
9612 }
9613
9614 int btrfs_init_cachep(void)
9615 {
9616 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9617 sizeof(struct btrfs_inode), 0,
9618 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9619 init_once);
9620 if (!btrfs_inode_cachep)
9621 goto fail;
9622
9623 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9624 sizeof(struct btrfs_trans_handle), 0,
9625 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9626 if (!btrfs_trans_handle_cachep)
9627 goto fail;
9628
9629 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9630 sizeof(struct btrfs_path), 0,
9631 SLAB_MEM_SPREAD, NULL);
9632 if (!btrfs_path_cachep)
9633 goto fail;
9634
9635 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9636 sizeof(struct btrfs_free_space), 0,
9637 SLAB_MEM_SPREAD, NULL);
9638 if (!btrfs_free_space_cachep)
9639 goto fail;
9640
9641 return 0;
9642 fail:
9643 btrfs_destroy_cachep();
9644 return -ENOMEM;
9645 }
9646
9647 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9648 u32 request_mask, unsigned int flags)
9649 {
9650 u64 delalloc_bytes;
9651 struct inode *inode = d_inode(path->dentry);
9652 u32 blocksize = inode->i_sb->s_blocksize;
9653 u32 bi_flags = BTRFS_I(inode)->flags;
9654
9655 stat->result_mask |= STATX_BTIME;
9656 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9657 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9658 if (bi_flags & BTRFS_INODE_APPEND)
9659 stat->attributes |= STATX_ATTR_APPEND;
9660 if (bi_flags & BTRFS_INODE_COMPRESS)
9661 stat->attributes |= STATX_ATTR_COMPRESSED;
9662 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9663 stat->attributes |= STATX_ATTR_IMMUTABLE;
9664 if (bi_flags & BTRFS_INODE_NODUMP)
9665 stat->attributes |= STATX_ATTR_NODUMP;
9666
9667 stat->attributes_mask |= (STATX_ATTR_APPEND |
9668 STATX_ATTR_COMPRESSED |
9669 STATX_ATTR_IMMUTABLE |
9670 STATX_ATTR_NODUMP);
9671
9672 generic_fillattr(inode, stat);
9673 stat->dev = BTRFS_I(inode)->root->anon_dev;
9674
9675 spin_lock(&BTRFS_I(inode)->lock);
9676 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9677 spin_unlock(&BTRFS_I(inode)->lock);
9678 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9679 ALIGN(delalloc_bytes, blocksize)) >> 9;
9680 return 0;
9681 }
9682
9683 static int btrfs_rename_exchange(struct inode *old_dir,
9684 struct dentry *old_dentry,
9685 struct inode *new_dir,
9686 struct dentry *new_dentry)
9687 {
9688 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9689 struct btrfs_trans_handle *trans;
9690 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9691 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9692 struct inode *new_inode = new_dentry->d_inode;
9693 struct inode *old_inode = old_dentry->d_inode;
9694 struct timespec ctime = current_time(old_inode);
9695 struct dentry *parent;
9696 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9697 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9698 u64 old_idx = 0;
9699 u64 new_idx = 0;
9700 u64 root_objectid;
9701 int ret;
9702 bool root_log_pinned = false;
9703 bool dest_log_pinned = false;
9704
9705 /* we only allow rename subvolume link between subvolumes */
9706 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9707 return -EXDEV;
9708
9709 /* close the race window with snapshot create/destroy ioctl */
9710 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9711 down_read(&fs_info->subvol_sem);
9712 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9713 down_read(&fs_info->subvol_sem);
9714
9715 /*
9716 * We want to reserve the absolute worst case amount of items. So if
9717 * both inodes are subvols and we need to unlink them then that would
9718 * require 4 item modifications, but if they are both normal inodes it
9719 * would require 5 item modifications, so we'll assume their normal
9720 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9721 * should cover the worst case number of items we'll modify.
9722 */
9723 trans = btrfs_start_transaction(root, 12);
9724 if (IS_ERR(trans)) {
9725 ret = PTR_ERR(trans);
9726 goto out_notrans;
9727 }
9728
9729 /*
9730 * We need to find a free sequence number both in the source and
9731 * in the destination directory for the exchange.
9732 */
9733 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9734 if (ret)
9735 goto out_fail;
9736 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9737 if (ret)
9738 goto out_fail;
9739
9740 BTRFS_I(old_inode)->dir_index = 0ULL;
9741 BTRFS_I(new_inode)->dir_index = 0ULL;
9742
9743 /* Reference for the source. */
9744 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9745 /* force full log commit if subvolume involved. */
9746 btrfs_set_log_full_commit(fs_info, trans);
9747 } else {
9748 btrfs_pin_log_trans(root);
9749 root_log_pinned = true;
9750 ret = btrfs_insert_inode_ref(trans, dest,
9751 new_dentry->d_name.name,
9752 new_dentry->d_name.len,
9753 old_ino,
9754 btrfs_ino(BTRFS_I(new_dir)),
9755 old_idx);
9756 if (ret)
9757 goto out_fail;
9758 }
9759
9760 /* And now for the dest. */
9761 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9762 /* force full log commit if subvolume involved. */
9763 btrfs_set_log_full_commit(fs_info, trans);
9764 } else {
9765 btrfs_pin_log_trans(dest);
9766 dest_log_pinned = true;
9767 ret = btrfs_insert_inode_ref(trans, root,
9768 old_dentry->d_name.name,
9769 old_dentry->d_name.len,
9770 new_ino,
9771 btrfs_ino(BTRFS_I(old_dir)),
9772 new_idx);
9773 if (ret)
9774 goto out_fail;
9775 }
9776
9777 /* Update inode version and ctime/mtime. */
9778 inode_inc_iversion(old_dir);
9779 inode_inc_iversion(new_dir);
9780 inode_inc_iversion(old_inode);
9781 inode_inc_iversion(new_inode);
9782 old_dir->i_ctime = old_dir->i_mtime = ctime;
9783 new_dir->i_ctime = new_dir->i_mtime = ctime;
9784 old_inode->i_ctime = ctime;
9785 new_inode->i_ctime = ctime;
9786
9787 if (old_dentry->d_parent != new_dentry->d_parent) {
9788 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9789 BTRFS_I(old_inode), 1);
9790 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9791 BTRFS_I(new_inode), 1);
9792 }
9793
9794 /* src is a subvolume */
9795 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9796 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9797 ret = btrfs_unlink_subvol(trans, root, old_dir,
9798 root_objectid,
9799 old_dentry->d_name.name,
9800 old_dentry->d_name.len);
9801 } else { /* src is an inode */
9802 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9803 BTRFS_I(old_dentry->d_inode),
9804 old_dentry->d_name.name,
9805 old_dentry->d_name.len);
9806 if (!ret)
9807 ret = btrfs_update_inode(trans, root, old_inode);
9808 }
9809 if (ret) {
9810 btrfs_abort_transaction(trans, ret);
9811 goto out_fail;
9812 }
9813
9814 /* dest is a subvolume */
9815 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9816 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9817 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9818 root_objectid,
9819 new_dentry->d_name.name,
9820 new_dentry->d_name.len);
9821 } else { /* dest is an inode */
9822 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9823 BTRFS_I(new_dentry->d_inode),
9824 new_dentry->d_name.name,
9825 new_dentry->d_name.len);
9826 if (!ret)
9827 ret = btrfs_update_inode(trans, dest, new_inode);
9828 }
9829 if (ret) {
9830 btrfs_abort_transaction(trans, ret);
9831 goto out_fail;
9832 }
9833
9834 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9835 new_dentry->d_name.name,
9836 new_dentry->d_name.len, 0, old_idx);
9837 if (ret) {
9838 btrfs_abort_transaction(trans, ret);
9839 goto out_fail;
9840 }
9841
9842 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9843 old_dentry->d_name.name,
9844 old_dentry->d_name.len, 0, new_idx);
9845 if (ret) {
9846 btrfs_abort_transaction(trans, ret);
9847 goto out_fail;
9848 }
9849
9850 if (old_inode->i_nlink == 1)
9851 BTRFS_I(old_inode)->dir_index = old_idx;
9852 if (new_inode->i_nlink == 1)
9853 BTRFS_I(new_inode)->dir_index = new_idx;
9854
9855 if (root_log_pinned) {
9856 parent = new_dentry->d_parent;
9857 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9858 parent);
9859 btrfs_end_log_trans(root);
9860 root_log_pinned = false;
9861 }
9862 if (dest_log_pinned) {
9863 parent = old_dentry->d_parent;
9864 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9865 parent);
9866 btrfs_end_log_trans(dest);
9867 dest_log_pinned = false;
9868 }
9869 out_fail:
9870 /*
9871 * If we have pinned a log and an error happened, we unpin tasks
9872 * trying to sync the log and force them to fallback to a transaction
9873 * commit if the log currently contains any of the inodes involved in
9874 * this rename operation (to ensure we do not persist a log with an
9875 * inconsistent state for any of these inodes or leading to any
9876 * inconsistencies when replayed). If the transaction was aborted, the
9877 * abortion reason is propagated to userspace when attempting to commit
9878 * the transaction. If the log does not contain any of these inodes, we
9879 * allow the tasks to sync it.
9880 */
9881 if (ret && (root_log_pinned || dest_log_pinned)) {
9882 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9883 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9884 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9885 (new_inode &&
9886 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9887 btrfs_set_log_full_commit(fs_info, trans);
9888
9889 if (root_log_pinned) {
9890 btrfs_end_log_trans(root);
9891 root_log_pinned = false;
9892 }
9893 if (dest_log_pinned) {
9894 btrfs_end_log_trans(dest);
9895 dest_log_pinned = false;
9896 }
9897 }
9898 ret = btrfs_end_transaction(trans);
9899 out_notrans:
9900 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9901 up_read(&fs_info->subvol_sem);
9902 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9903 up_read(&fs_info->subvol_sem);
9904
9905 return ret;
9906 }
9907
9908 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9909 struct btrfs_root *root,
9910 struct inode *dir,
9911 struct dentry *dentry)
9912 {
9913 int ret;
9914 struct inode *inode;
9915 u64 objectid;
9916 u64 index;
9917
9918 ret = btrfs_find_free_ino(root, &objectid);
9919 if (ret)
9920 return ret;
9921
9922 inode = btrfs_new_inode(trans, root, dir,
9923 dentry->d_name.name,
9924 dentry->d_name.len,
9925 btrfs_ino(BTRFS_I(dir)),
9926 objectid,
9927 S_IFCHR | WHITEOUT_MODE,
9928 &index);
9929
9930 if (IS_ERR(inode)) {
9931 ret = PTR_ERR(inode);
9932 return ret;
9933 }
9934
9935 inode->i_op = &btrfs_special_inode_operations;
9936 init_special_inode(inode, inode->i_mode,
9937 WHITEOUT_DEV);
9938
9939 ret = btrfs_init_inode_security(trans, inode, dir,
9940 &dentry->d_name);
9941 if (ret)
9942 goto out;
9943
9944 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9945 BTRFS_I(inode), 0, index);
9946 if (ret)
9947 goto out;
9948
9949 ret = btrfs_update_inode(trans, root, inode);
9950 out:
9951 unlock_new_inode(inode);
9952 if (ret)
9953 inode_dec_link_count(inode);
9954 iput(inode);
9955
9956 return ret;
9957 }
9958
9959 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9960 struct inode *new_dir, struct dentry *new_dentry,
9961 unsigned int flags)
9962 {
9963 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9964 struct btrfs_trans_handle *trans;
9965 unsigned int trans_num_items;
9966 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9967 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9968 struct inode *new_inode = d_inode(new_dentry);
9969 struct inode *old_inode = d_inode(old_dentry);
9970 u64 index = 0;
9971 u64 root_objectid;
9972 int ret;
9973 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9974 bool log_pinned = false;
9975
9976 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9977 return -EPERM;
9978
9979 /* we only allow rename subvolume link between subvolumes */
9980 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9981 return -EXDEV;
9982
9983 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9984 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9985 return -ENOTEMPTY;
9986
9987 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9988 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9989 return -ENOTEMPTY;
9990
9991
9992 /* check for collisions, even if the name isn't there */
9993 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9994 new_dentry->d_name.name,
9995 new_dentry->d_name.len);
9996
9997 if (ret) {
9998 if (ret == -EEXIST) {
9999 /* we shouldn't get
10000 * eexist without a new_inode */
10001 if (WARN_ON(!new_inode)) {
10002 return ret;
10003 }
10004 } else {
10005 /* maybe -EOVERFLOW */
10006 return ret;
10007 }
10008 }
10009 ret = 0;
10010
10011 /*
10012 * we're using rename to replace one file with another. Start IO on it
10013 * now so we don't add too much work to the end of the transaction
10014 */
10015 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
10016 filemap_flush(old_inode->i_mapping);
10017
10018 /* close the racy window with snapshot create/destroy ioctl */
10019 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10020 down_read(&fs_info->subvol_sem);
10021 /*
10022 * We want to reserve the absolute worst case amount of items. So if
10023 * both inodes are subvols and we need to unlink them then that would
10024 * require 4 item modifications, but if they are both normal inodes it
10025 * would require 5 item modifications, so we'll assume they are normal
10026 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
10027 * should cover the worst case number of items we'll modify.
10028 * If our rename has the whiteout flag, we need more 5 units for the
10029 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
10030 * when selinux is enabled).
10031 */
10032 trans_num_items = 11;
10033 if (flags & RENAME_WHITEOUT)
10034 trans_num_items += 5;
10035 trans = btrfs_start_transaction(root, trans_num_items);
10036 if (IS_ERR(trans)) {
10037 ret = PTR_ERR(trans);
10038 goto out_notrans;
10039 }
10040
10041 if (dest != root)
10042 btrfs_record_root_in_trans(trans, dest);
10043
10044 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
10045 if (ret)
10046 goto out_fail;
10047
10048 BTRFS_I(old_inode)->dir_index = 0ULL;
10049 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10050 /* force full log commit if subvolume involved. */
10051 btrfs_set_log_full_commit(fs_info, trans);
10052 } else {
10053 btrfs_pin_log_trans(root);
10054 log_pinned = true;
10055 ret = btrfs_insert_inode_ref(trans, dest,
10056 new_dentry->d_name.name,
10057 new_dentry->d_name.len,
10058 old_ino,
10059 btrfs_ino(BTRFS_I(new_dir)), index);
10060 if (ret)
10061 goto out_fail;
10062 }
10063
10064 inode_inc_iversion(old_dir);
10065 inode_inc_iversion(new_dir);
10066 inode_inc_iversion(old_inode);
10067 old_dir->i_ctime = old_dir->i_mtime =
10068 new_dir->i_ctime = new_dir->i_mtime =
10069 old_inode->i_ctime = current_time(old_dir);
10070
10071 if (old_dentry->d_parent != new_dentry->d_parent)
10072 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10073 BTRFS_I(old_inode), 1);
10074
10075 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10076 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10077 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10078 old_dentry->d_name.name,
10079 old_dentry->d_name.len);
10080 } else {
10081 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10082 BTRFS_I(d_inode(old_dentry)),
10083 old_dentry->d_name.name,
10084 old_dentry->d_name.len);
10085 if (!ret)
10086 ret = btrfs_update_inode(trans, root, old_inode);
10087 }
10088 if (ret) {
10089 btrfs_abort_transaction(trans, ret);
10090 goto out_fail;
10091 }
10092
10093 if (new_inode) {
10094 inode_inc_iversion(new_inode);
10095 new_inode->i_ctime = current_time(new_inode);
10096 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10097 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10098 root_objectid = BTRFS_I(new_inode)->location.objectid;
10099 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10100 root_objectid,
10101 new_dentry->d_name.name,
10102 new_dentry->d_name.len);
10103 BUG_ON(new_inode->i_nlink == 0);
10104 } else {
10105 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10106 BTRFS_I(d_inode(new_dentry)),
10107 new_dentry->d_name.name,
10108 new_dentry->d_name.len);
10109 }
10110 if (!ret && new_inode->i_nlink == 0)
10111 ret = btrfs_orphan_add(trans,
10112 BTRFS_I(d_inode(new_dentry)));
10113 if (ret) {
10114 btrfs_abort_transaction(trans, ret);
10115 goto out_fail;
10116 }
10117 }
10118
10119 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10120 new_dentry->d_name.name,
10121 new_dentry->d_name.len, 0, index);
10122 if (ret) {
10123 btrfs_abort_transaction(trans, ret);
10124 goto out_fail;
10125 }
10126
10127 if (old_inode->i_nlink == 1)
10128 BTRFS_I(old_inode)->dir_index = index;
10129
10130 if (log_pinned) {
10131 struct dentry *parent = new_dentry->d_parent;
10132
10133 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10134 parent);
10135 btrfs_end_log_trans(root);
10136 log_pinned = false;
10137 }
10138
10139 if (flags & RENAME_WHITEOUT) {
10140 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10141 old_dentry);
10142
10143 if (ret) {
10144 btrfs_abort_transaction(trans, ret);
10145 goto out_fail;
10146 }
10147 }
10148 out_fail:
10149 /*
10150 * If we have pinned the log and an error happened, we unpin tasks
10151 * trying to sync the log and force them to fallback to a transaction
10152 * commit if the log currently contains any of the inodes involved in
10153 * this rename operation (to ensure we do not persist a log with an
10154 * inconsistent state for any of these inodes or leading to any
10155 * inconsistencies when replayed). If the transaction was aborted, the
10156 * abortion reason is propagated to userspace when attempting to commit
10157 * the transaction. If the log does not contain any of these inodes, we
10158 * allow the tasks to sync it.
10159 */
10160 if (ret && log_pinned) {
10161 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10162 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10163 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10164 (new_inode &&
10165 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10166 btrfs_set_log_full_commit(fs_info, trans);
10167
10168 btrfs_end_log_trans(root);
10169 log_pinned = false;
10170 }
10171 btrfs_end_transaction(trans);
10172 out_notrans:
10173 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10174 up_read(&fs_info->subvol_sem);
10175
10176 return ret;
10177 }
10178
10179 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10180 struct inode *new_dir, struct dentry *new_dentry,
10181 unsigned int flags)
10182 {
10183 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10184 return -EINVAL;
10185
10186 if (flags & RENAME_EXCHANGE)
10187 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10188 new_dentry);
10189
10190 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10191 }
10192
10193 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10194 {
10195 struct btrfs_delalloc_work *delalloc_work;
10196 struct inode *inode;
10197
10198 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10199 work);
10200 inode = delalloc_work->inode;
10201 filemap_flush(inode->i_mapping);
10202 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10203 &BTRFS_I(inode)->runtime_flags))
10204 filemap_flush(inode->i_mapping);
10205
10206 if (delalloc_work->delay_iput)
10207 btrfs_add_delayed_iput(inode);
10208 else
10209 iput(inode);
10210 complete(&delalloc_work->completion);
10211 }
10212
10213 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10214 int delay_iput)
10215 {
10216 struct btrfs_delalloc_work *work;
10217
10218 work = kmalloc(sizeof(*work), GFP_NOFS);
10219 if (!work)
10220 return NULL;
10221
10222 init_completion(&work->completion);
10223 INIT_LIST_HEAD(&work->list);
10224 work->inode = inode;
10225 work->delay_iput = delay_iput;
10226 WARN_ON_ONCE(!inode);
10227 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10228 btrfs_run_delalloc_work, NULL, NULL);
10229
10230 return work;
10231 }
10232
10233 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10234 {
10235 wait_for_completion(&work->completion);
10236 kfree(work);
10237 }
10238
10239 /*
10240 * some fairly slow code that needs optimization. This walks the list
10241 * of all the inodes with pending delalloc and forces them to disk.
10242 */
10243 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10244 int nr)
10245 {
10246 struct btrfs_inode *binode;
10247 struct inode *inode;
10248 struct btrfs_delalloc_work *work, *next;
10249 struct list_head works;
10250 struct list_head splice;
10251 int ret = 0;
10252
10253 INIT_LIST_HEAD(&works);
10254 INIT_LIST_HEAD(&splice);
10255
10256 mutex_lock(&root->delalloc_mutex);
10257 spin_lock(&root->delalloc_lock);
10258 list_splice_init(&root->delalloc_inodes, &splice);
10259 while (!list_empty(&splice)) {
10260 binode = list_entry(splice.next, struct btrfs_inode,
10261 delalloc_inodes);
10262
10263 list_move_tail(&binode->delalloc_inodes,
10264 &root->delalloc_inodes);
10265 inode = igrab(&binode->vfs_inode);
10266 if (!inode) {
10267 cond_resched_lock(&root->delalloc_lock);
10268 continue;
10269 }
10270 spin_unlock(&root->delalloc_lock);
10271
10272 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10273 if (!work) {
10274 if (delay_iput)
10275 btrfs_add_delayed_iput(inode);
10276 else
10277 iput(inode);
10278 ret = -ENOMEM;
10279 goto out;
10280 }
10281 list_add_tail(&work->list, &works);
10282 btrfs_queue_work(root->fs_info->flush_workers,
10283 &work->work);
10284 ret++;
10285 if (nr != -1 && ret >= nr)
10286 goto out;
10287 cond_resched();
10288 spin_lock(&root->delalloc_lock);
10289 }
10290 spin_unlock(&root->delalloc_lock);
10291
10292 out:
10293 list_for_each_entry_safe(work, next, &works, list) {
10294 list_del_init(&work->list);
10295 btrfs_wait_and_free_delalloc_work(work);
10296 }
10297
10298 if (!list_empty_careful(&splice)) {
10299 spin_lock(&root->delalloc_lock);
10300 list_splice_tail(&splice, &root->delalloc_inodes);
10301 spin_unlock(&root->delalloc_lock);
10302 }
10303 mutex_unlock(&root->delalloc_mutex);
10304 return ret;
10305 }
10306
10307 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10308 {
10309 struct btrfs_fs_info *fs_info = root->fs_info;
10310 int ret;
10311
10312 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10313 return -EROFS;
10314
10315 ret = __start_delalloc_inodes(root, delay_iput, -1);
10316 if (ret > 0)
10317 ret = 0;
10318 return ret;
10319 }
10320
10321 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10322 int nr)
10323 {
10324 struct btrfs_root *root;
10325 struct list_head splice;
10326 int ret;
10327
10328 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10329 return -EROFS;
10330
10331 INIT_LIST_HEAD(&splice);
10332
10333 mutex_lock(&fs_info->delalloc_root_mutex);
10334 spin_lock(&fs_info->delalloc_root_lock);
10335 list_splice_init(&fs_info->delalloc_roots, &splice);
10336 while (!list_empty(&splice) && nr) {
10337 root = list_first_entry(&splice, struct btrfs_root,
10338 delalloc_root);
10339 root = btrfs_grab_fs_root(root);
10340 BUG_ON(!root);
10341 list_move_tail(&root->delalloc_root,
10342 &fs_info->delalloc_roots);
10343 spin_unlock(&fs_info->delalloc_root_lock);
10344
10345 ret = __start_delalloc_inodes(root, delay_iput, nr);
10346 btrfs_put_fs_root(root);
10347 if (ret < 0)
10348 goto out;
10349
10350 if (nr != -1) {
10351 nr -= ret;
10352 WARN_ON(nr < 0);
10353 }
10354 spin_lock(&fs_info->delalloc_root_lock);
10355 }
10356 spin_unlock(&fs_info->delalloc_root_lock);
10357
10358 ret = 0;
10359 out:
10360 if (!list_empty_careful(&splice)) {
10361 spin_lock(&fs_info->delalloc_root_lock);
10362 list_splice_tail(&splice, &fs_info->delalloc_roots);
10363 spin_unlock(&fs_info->delalloc_root_lock);
10364 }
10365 mutex_unlock(&fs_info->delalloc_root_mutex);
10366 return ret;
10367 }
10368
10369 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10370 const char *symname)
10371 {
10372 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10373 struct btrfs_trans_handle *trans;
10374 struct btrfs_root *root = BTRFS_I(dir)->root;
10375 struct btrfs_path *path;
10376 struct btrfs_key key;
10377 struct inode *inode = NULL;
10378 int err;
10379 int drop_inode = 0;
10380 u64 objectid;
10381 u64 index = 0;
10382 int name_len;
10383 int datasize;
10384 unsigned long ptr;
10385 struct btrfs_file_extent_item *ei;
10386 struct extent_buffer *leaf;
10387
10388 name_len = strlen(symname);
10389 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10390 return -ENAMETOOLONG;
10391
10392 /*
10393 * 2 items for inode item and ref
10394 * 2 items for dir items
10395 * 1 item for updating parent inode item
10396 * 1 item for the inline extent item
10397 * 1 item for xattr if selinux is on
10398 */
10399 trans = btrfs_start_transaction(root, 7);
10400 if (IS_ERR(trans))
10401 return PTR_ERR(trans);
10402
10403 err = btrfs_find_free_ino(root, &objectid);
10404 if (err)
10405 goto out_unlock;
10406
10407 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10408 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10409 objectid, S_IFLNK|S_IRWXUGO, &index);
10410 if (IS_ERR(inode)) {
10411 err = PTR_ERR(inode);
10412 goto out_unlock;
10413 }
10414
10415 /*
10416 * If the active LSM wants to access the inode during
10417 * d_instantiate it needs these. Smack checks to see
10418 * if the filesystem supports xattrs by looking at the
10419 * ops vector.
10420 */
10421 inode->i_fop = &btrfs_file_operations;
10422 inode->i_op = &btrfs_file_inode_operations;
10423 inode->i_mapping->a_ops = &btrfs_aops;
10424 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10425
10426 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10427 if (err)
10428 goto out_unlock_inode;
10429
10430 path = btrfs_alloc_path();
10431 if (!path) {
10432 err = -ENOMEM;
10433 goto out_unlock_inode;
10434 }
10435 key.objectid = btrfs_ino(BTRFS_I(inode));
10436 key.offset = 0;
10437 key.type = BTRFS_EXTENT_DATA_KEY;
10438 datasize = btrfs_file_extent_calc_inline_size(name_len);
10439 err = btrfs_insert_empty_item(trans, root, path, &key,
10440 datasize);
10441 if (err) {
10442 btrfs_free_path(path);
10443 goto out_unlock_inode;
10444 }
10445 leaf = path->nodes[0];
10446 ei = btrfs_item_ptr(leaf, path->slots[0],
10447 struct btrfs_file_extent_item);
10448 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10449 btrfs_set_file_extent_type(leaf, ei,
10450 BTRFS_FILE_EXTENT_INLINE);
10451 btrfs_set_file_extent_encryption(leaf, ei, 0);
10452 btrfs_set_file_extent_compression(leaf, ei, 0);
10453 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10454 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10455
10456 ptr = btrfs_file_extent_inline_start(ei);
10457 write_extent_buffer(leaf, symname, ptr, name_len);
10458 btrfs_mark_buffer_dirty(leaf);
10459 btrfs_free_path(path);
10460
10461 inode->i_op = &btrfs_symlink_inode_operations;
10462 inode_nohighmem(inode);
10463 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10464 inode_set_bytes(inode, name_len);
10465 btrfs_i_size_write(BTRFS_I(inode), name_len);
10466 err = btrfs_update_inode(trans, root, inode);
10467 /*
10468 * Last step, add directory indexes for our symlink inode. This is the
10469 * last step to avoid extra cleanup of these indexes if an error happens
10470 * elsewhere above.
10471 */
10472 if (!err)
10473 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10474 BTRFS_I(inode), 0, index);
10475 if (err) {
10476 drop_inode = 1;
10477 goto out_unlock_inode;
10478 }
10479
10480 d_instantiate_new(dentry, inode);
10481
10482 out_unlock:
10483 btrfs_end_transaction(trans);
10484 if (drop_inode) {
10485 inode_dec_link_count(inode);
10486 iput(inode);
10487 }
10488 btrfs_btree_balance_dirty(fs_info);
10489 return err;
10490
10491 out_unlock_inode:
10492 drop_inode = 1;
10493 unlock_new_inode(inode);
10494 goto out_unlock;
10495 }
10496
10497 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10498 u64 start, u64 num_bytes, u64 min_size,
10499 loff_t actual_len, u64 *alloc_hint,
10500 struct btrfs_trans_handle *trans)
10501 {
10502 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10503 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10504 struct extent_map *em;
10505 struct btrfs_root *root = BTRFS_I(inode)->root;
10506 struct btrfs_key ins;
10507 u64 cur_offset = start;
10508 u64 i_size;
10509 u64 cur_bytes;
10510 u64 last_alloc = (u64)-1;
10511 int ret = 0;
10512 bool own_trans = true;
10513 u64 end = start + num_bytes - 1;
10514
10515 if (trans)
10516 own_trans = false;
10517 while (num_bytes > 0) {
10518 if (own_trans) {
10519 trans = btrfs_start_transaction(root, 3);
10520 if (IS_ERR(trans)) {
10521 ret = PTR_ERR(trans);
10522 break;
10523 }
10524 }
10525
10526 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10527 cur_bytes = max(cur_bytes, min_size);
10528 /*
10529 * If we are severely fragmented we could end up with really
10530 * small allocations, so if the allocator is returning small
10531 * chunks lets make its job easier by only searching for those
10532 * sized chunks.
10533 */
10534 cur_bytes = min(cur_bytes, last_alloc);
10535 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10536 min_size, 0, *alloc_hint, &ins, 1, 0);
10537 if (ret) {
10538 if (own_trans)
10539 btrfs_end_transaction(trans);
10540 break;
10541 }
10542 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10543
10544 last_alloc = ins.offset;
10545 ret = insert_reserved_file_extent(trans, inode,
10546 cur_offset, ins.objectid,
10547 ins.offset, ins.offset,
10548 ins.offset, 0, 0, 0,
10549 BTRFS_FILE_EXTENT_PREALLOC);
10550 if (ret) {
10551 btrfs_free_reserved_extent(fs_info, ins.objectid,
10552 ins.offset, 0);
10553 btrfs_abort_transaction(trans, ret);
10554 if (own_trans)
10555 btrfs_end_transaction(trans);
10556 break;
10557 }
10558
10559 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10560 cur_offset + ins.offset -1, 0);
10561
10562 em = alloc_extent_map();
10563 if (!em) {
10564 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10565 &BTRFS_I(inode)->runtime_flags);
10566 goto next;
10567 }
10568
10569 em->start = cur_offset;
10570 em->orig_start = cur_offset;
10571 em->len = ins.offset;
10572 em->block_start = ins.objectid;
10573 em->block_len = ins.offset;
10574 em->orig_block_len = ins.offset;
10575 em->ram_bytes = ins.offset;
10576 em->bdev = fs_info->fs_devices->latest_bdev;
10577 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10578 em->generation = trans->transid;
10579
10580 while (1) {
10581 write_lock(&em_tree->lock);
10582 ret = add_extent_mapping(em_tree, em, 1);
10583 write_unlock(&em_tree->lock);
10584 if (ret != -EEXIST)
10585 break;
10586 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10587 cur_offset + ins.offset - 1,
10588 0);
10589 }
10590 free_extent_map(em);
10591 next:
10592 num_bytes -= ins.offset;
10593 cur_offset += ins.offset;
10594 *alloc_hint = ins.objectid + ins.offset;
10595
10596 inode_inc_iversion(inode);
10597 inode->i_ctime = current_time(inode);
10598 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10599 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10600 (actual_len > inode->i_size) &&
10601 (cur_offset > inode->i_size)) {
10602 if (cur_offset > actual_len)
10603 i_size = actual_len;
10604 else
10605 i_size = cur_offset;
10606 i_size_write(inode, i_size);
10607 btrfs_ordered_update_i_size(inode, i_size, NULL);
10608 }
10609
10610 ret = btrfs_update_inode(trans, root, inode);
10611
10612 if (ret) {
10613 btrfs_abort_transaction(trans, ret);
10614 if (own_trans)
10615 btrfs_end_transaction(trans);
10616 break;
10617 }
10618
10619 if (own_trans)
10620 btrfs_end_transaction(trans);
10621 }
10622 if (cur_offset < end)
10623 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10624 end - cur_offset + 1);
10625 return ret;
10626 }
10627
10628 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10629 u64 start, u64 num_bytes, u64 min_size,
10630 loff_t actual_len, u64 *alloc_hint)
10631 {
10632 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10633 min_size, actual_len, alloc_hint,
10634 NULL);
10635 }
10636
10637 int btrfs_prealloc_file_range_trans(struct inode *inode,
10638 struct btrfs_trans_handle *trans, int mode,
10639 u64 start, u64 num_bytes, u64 min_size,
10640 loff_t actual_len, u64 *alloc_hint)
10641 {
10642 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10643 min_size, actual_len, alloc_hint, trans);
10644 }
10645
10646 static int btrfs_set_page_dirty(struct page *page)
10647 {
10648 return __set_page_dirty_nobuffers(page);
10649 }
10650
10651 static int btrfs_permission(struct inode *inode, int mask)
10652 {
10653 struct btrfs_root *root = BTRFS_I(inode)->root;
10654 umode_t mode = inode->i_mode;
10655
10656 if (mask & MAY_WRITE &&
10657 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10658 if (btrfs_root_readonly(root))
10659 return -EROFS;
10660 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10661 return -EACCES;
10662 }
10663 return generic_permission(inode, mask);
10664 }
10665
10666 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10667 {
10668 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10669 struct btrfs_trans_handle *trans;
10670 struct btrfs_root *root = BTRFS_I(dir)->root;
10671 struct inode *inode = NULL;
10672 u64 objectid;
10673 u64 index;
10674 int ret = 0;
10675
10676 /*
10677 * 5 units required for adding orphan entry
10678 */
10679 trans = btrfs_start_transaction(root, 5);
10680 if (IS_ERR(trans))
10681 return PTR_ERR(trans);
10682
10683 ret = btrfs_find_free_ino(root, &objectid);
10684 if (ret)
10685 goto out;
10686
10687 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10688 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10689 if (IS_ERR(inode)) {
10690 ret = PTR_ERR(inode);
10691 inode = NULL;
10692 goto out;
10693 }
10694
10695 inode->i_fop = &btrfs_file_operations;
10696 inode->i_op = &btrfs_file_inode_operations;
10697
10698 inode->i_mapping->a_ops = &btrfs_aops;
10699 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10700
10701 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10702 if (ret)
10703 goto out_inode;
10704
10705 ret = btrfs_update_inode(trans, root, inode);
10706 if (ret)
10707 goto out_inode;
10708 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10709 if (ret)
10710 goto out_inode;
10711
10712 /*
10713 * We set number of links to 0 in btrfs_new_inode(), and here we set
10714 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10715 * through:
10716 *
10717 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10718 */
10719 set_nlink(inode, 1);
10720 unlock_new_inode(inode);
10721 d_tmpfile(dentry, inode);
10722 mark_inode_dirty(inode);
10723
10724 out:
10725 btrfs_end_transaction(trans);
10726 if (ret)
10727 iput(inode);
10728 btrfs_balance_delayed_items(fs_info);
10729 btrfs_btree_balance_dirty(fs_info);
10730 return ret;
10731
10732 out_inode:
10733 unlock_new_inode(inode);
10734 goto out;
10735
10736 }
10737
10738 __attribute__((const))
10739 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10740 {
10741 return -EAGAIN;
10742 }
10743
10744 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10745 {
10746 struct inode *inode = private_data;
10747 return btrfs_sb(inode->i_sb);
10748 }
10749
10750 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10751 u64 start, u64 end)
10752 {
10753 struct inode *inode = private_data;
10754 u64 isize;
10755
10756 isize = i_size_read(inode);
10757 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10758 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10759 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10760 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10761 }
10762 }
10763
10764 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10765 {
10766 struct inode *inode = private_data;
10767 unsigned long index = start >> PAGE_SHIFT;
10768 unsigned long end_index = end >> PAGE_SHIFT;
10769 struct page *page;
10770
10771 while (index <= end_index) {
10772 page = find_get_page(inode->i_mapping, index);
10773 ASSERT(page); /* Pages should be in the extent_io_tree */
10774 set_page_writeback(page);
10775 put_page(page);
10776 index++;
10777 }
10778 }
10779
10780 static const struct inode_operations btrfs_dir_inode_operations = {
10781 .getattr = btrfs_getattr,
10782 .lookup = btrfs_lookup,
10783 .create = btrfs_create,
10784 .unlink = btrfs_unlink,
10785 .link = btrfs_link,
10786 .mkdir = btrfs_mkdir,
10787 .rmdir = btrfs_rmdir,
10788 .rename = btrfs_rename2,
10789 .symlink = btrfs_symlink,
10790 .setattr = btrfs_setattr,
10791 .mknod = btrfs_mknod,
10792 .listxattr = btrfs_listxattr,
10793 .permission = btrfs_permission,
10794 .get_acl = btrfs_get_acl,
10795 .set_acl = btrfs_set_acl,
10796 .update_time = btrfs_update_time,
10797 .tmpfile = btrfs_tmpfile,
10798 };
10799 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10800 .lookup = btrfs_lookup,
10801 .permission = btrfs_permission,
10802 .update_time = btrfs_update_time,
10803 };
10804
10805 static const struct file_operations btrfs_dir_file_operations = {
10806 .llseek = generic_file_llseek,
10807 .read = generic_read_dir,
10808 .iterate_shared = btrfs_real_readdir,
10809 .open = btrfs_opendir,
10810 .unlocked_ioctl = btrfs_ioctl,
10811 #ifdef CONFIG_COMPAT
10812 .compat_ioctl = btrfs_compat_ioctl,
10813 #endif
10814 .release = btrfs_release_file,
10815 .fsync = btrfs_sync_file,
10816 };
10817
10818 static const struct extent_io_ops btrfs_extent_io_ops = {
10819 /* mandatory callbacks */
10820 .submit_bio_hook = btrfs_submit_bio_hook,
10821 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10822 .merge_bio_hook = btrfs_merge_bio_hook,
10823 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10824 .tree_fs_info = iotree_fs_info,
10825 .set_range_writeback = btrfs_set_range_writeback,
10826
10827 /* optional callbacks */
10828 .fill_delalloc = run_delalloc_range,
10829 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10830 .writepage_start_hook = btrfs_writepage_start_hook,
10831 .set_bit_hook = btrfs_set_bit_hook,
10832 .clear_bit_hook = btrfs_clear_bit_hook,
10833 .merge_extent_hook = btrfs_merge_extent_hook,
10834 .split_extent_hook = btrfs_split_extent_hook,
10835 .check_extent_io_range = btrfs_check_extent_io_range,
10836 };
10837
10838 /*
10839 * btrfs doesn't support the bmap operation because swapfiles
10840 * use bmap to make a mapping of extents in the file. They assume
10841 * these extents won't change over the life of the file and they
10842 * use the bmap result to do IO directly to the drive.
10843 *
10844 * the btrfs bmap call would return logical addresses that aren't
10845 * suitable for IO and they also will change frequently as COW
10846 * operations happen. So, swapfile + btrfs == corruption.
10847 *
10848 * For now we're avoiding this by dropping bmap.
10849 */
10850 static const struct address_space_operations btrfs_aops = {
10851 .readpage = btrfs_readpage,
10852 .writepage = btrfs_writepage,
10853 .writepages = btrfs_writepages,
10854 .readpages = btrfs_readpages,
10855 .direct_IO = btrfs_direct_IO,
10856 .invalidatepage = btrfs_invalidatepage,
10857 .releasepage = btrfs_releasepage,
10858 .set_page_dirty = btrfs_set_page_dirty,
10859 .error_remove_page = generic_error_remove_page,
10860 };
10861
10862 static const struct address_space_operations btrfs_symlink_aops = {
10863 .readpage = btrfs_readpage,
10864 .writepage = btrfs_writepage,
10865 .invalidatepage = btrfs_invalidatepage,
10866 .releasepage = btrfs_releasepage,
10867 };
10868
10869 static const struct inode_operations btrfs_file_inode_operations = {
10870 .getattr = btrfs_getattr,
10871 .setattr = btrfs_setattr,
10872 .listxattr = btrfs_listxattr,
10873 .permission = btrfs_permission,
10874 .fiemap = btrfs_fiemap,
10875 .get_acl = btrfs_get_acl,
10876 .set_acl = btrfs_set_acl,
10877 .update_time = btrfs_update_time,
10878 };
10879 static const struct inode_operations btrfs_special_inode_operations = {
10880 .getattr = btrfs_getattr,
10881 .setattr = btrfs_setattr,
10882 .permission = btrfs_permission,
10883 .listxattr = btrfs_listxattr,
10884 .get_acl = btrfs_get_acl,
10885 .set_acl = btrfs_set_acl,
10886 .update_time = btrfs_update_time,
10887 };
10888 static const struct inode_operations btrfs_symlink_inode_operations = {
10889 .get_link = page_get_link,
10890 .getattr = btrfs_getattr,
10891 .setattr = btrfs_setattr,
10892 .permission = btrfs_permission,
10893 .listxattr = btrfs_listxattr,
10894 .update_time = btrfs_update_time,
10895 };
10896
10897 const struct dentry_operations btrfs_dentry_operations = {
10898 .d_delete = btrfs_dentry_delete,
10899 .d_release = btrfs_dentry_release,
10900 };
ubuntu-bionic-kernel mirror