?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: fix xattr loss after power failure
[linux/fpc-iii.git] / fs / btrfs / inode.c
blobd241285a0d2a8dea0ff858f58a5590c18298f2b0
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
10 #include <linux/fs.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/mpage.h>
18 #include <linux/swap.h>
19 #include <linux/writeback.h>
20 #include <linux/compat.h>
21 #include <linux/bit_spinlock.h>
22 #include <linux/xattr.h>
23 #include <linux/posix_acl.h>
24 #include <linux/falloc.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/mount.h>
28 #include <linux/btrfs.h>
29 #include <linux/blkdev.h>
30 #include <linux/posix_acl_xattr.h>
31 #include <linux/uio.h>
32 #include <linux/magic.h>
33 #include <linux/iversion.h>
34 #include <asm/unaligned.h>
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
40 #include "ordered-data.h"
41 #include "xattr.h"
42 #include "tree-log.h"
43 #include "volumes.h"
44 #include "compression.h"
45 #include "locking.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "backref.h"
49 #include "props.h"
50 #include "qgroup.h"
51 #include "dedupe.h"
52
53 struct btrfs_iget_args {
54 struct btrfs_key *location;
55 struct btrfs_root *root;
56 };
57
58 struct btrfs_dio_data {
59 u64 reserve;
60 u64 unsubmitted_oe_range_start;
61 u64 unsubmitted_oe_range_end;
62 int overwrite;
63 };
64
65 static const struct inode_operations btrfs_dir_inode_operations;
66 static const struct inode_operations btrfs_symlink_inode_operations;
67 static const struct inode_operations btrfs_dir_ro_inode_operations;
68 static const struct inode_operations btrfs_special_inode_operations;
69 static const struct inode_operations btrfs_file_inode_operations;
70 static const struct address_space_operations btrfs_aops;
71 static const struct address_space_operations btrfs_symlink_aops;
72 static const struct file_operations btrfs_dir_file_operations;
73 static const struct extent_io_ops btrfs_extent_io_ops;
74
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79
80 #define S_SHIFT 12
81 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
89 };
90
91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
92 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, u64 delalloc_end,
97 int *page_started, unsigned long *nr_written,
98 int unlock, struct btrfs_dedupe_hash *hash);
99 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
100 u64 orig_start, u64 block_start,
101 u64 block_len, u64 orig_block_len,
102 u64 ram_bytes, int compress_type,
103 int type);
104
105 static void __endio_write_update_ordered(struct inode *inode,
106 const u64 offset, const u64 bytes,
107 const bool uptodate);
108
109 /*
110 * Cleanup all submitted ordered extents in specified range to handle errors
111 * from the fill_dellaloc() callback.
112 *
113 * NOTE: caller must ensure that when an error happens, it can not call
114 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
115 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
116 * to be released, which we want to happen only when finishing the ordered
117 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
118 * fill_delalloc() callback already does proper cleanup for the first page of
119 * the range, that is, it invokes the callback writepage_end_io_hook() for the
120 * range of the first page.
121 */
122 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
123 const u64 offset,
124 const u64 bytes)
125 {
126 unsigned long index = offset >> PAGE_SHIFT;
127 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
128 struct page *page;
129
130 while (index <= end_index) {
131 page = find_get_page(inode->i_mapping, index);
132 index++;
133 if (!page)
134 continue;
135 ClearPagePrivate2(page);
136 put_page(page);
137 }
138 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139 bytes - PAGE_SIZE, false);
140 }
141
142 static int btrfs_dirty_inode(struct inode *inode);
143
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
146 {
147 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
148 }
149 #endif
150
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 struct inode *inode, struct inode *dir,
153 const struct qstr *qstr)
154 {
155 int err;
156
157 err = btrfs_init_acl(trans, inode, dir);
158 if (!err)
159 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
160 return err;
161 }
162
163 /*
164 * this does all the hard work for inserting an inline extent into
165 * the btree. The caller should have done a btrfs_drop_extents so that
166 * no overlapping inline items exist in the btree
167 */
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 struct btrfs_path *path, int extent_inserted,
170 struct btrfs_root *root, struct inode *inode,
171 u64 start, size_t size, size_t compressed_size,
172 int compress_type,
173 struct page **compressed_pages)
174 {
175 struct extent_buffer *leaf;
176 struct page *page = NULL;
177 char *kaddr;
178 unsigned long ptr;
179 struct btrfs_file_extent_item *ei;
180 int ret;
181 size_t cur_size = size;
182 unsigned long offset;
183
184 if (compressed_size && compressed_pages)
185 cur_size = compressed_size;
186
187 inode_add_bytes(inode, size);
188
189 if (!extent_inserted) {
190 struct btrfs_key key;
191 size_t datasize;
192
193 key.objectid = btrfs_ino(BTRFS_I(inode));
194 key.offset = start;
195 key.type = BTRFS_EXTENT_DATA_KEY;
196
197 datasize = btrfs_file_extent_calc_inline_size(cur_size);
198 path->leave_spinning = 1;
199 ret = btrfs_insert_empty_item(trans, root, path, &key,
200 datasize);
201 if (ret)
202 goto fail;
203 }
204 leaf = path->nodes[0];
205 ei = btrfs_item_ptr(leaf, path->slots[0],
206 struct btrfs_file_extent_item);
207 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209 btrfs_set_file_extent_encryption(leaf, ei, 0);
210 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212 ptr = btrfs_file_extent_inline_start(ei);
213
214 if (compress_type != BTRFS_COMPRESS_NONE) {
215 struct page *cpage;
216 int i = 0;
217 while (compressed_size > 0) {
218 cpage = compressed_pages[i];
219 cur_size = min_t(unsigned long, compressed_size,
220 PAGE_SIZE);
221
222 kaddr = kmap_atomic(cpage);
223 write_extent_buffer(leaf, kaddr, ptr, cur_size);
224 kunmap_atomic(kaddr);
225
226 i++;
227 ptr += cur_size;
228 compressed_size -= cur_size;
229 }
230 btrfs_set_file_extent_compression(leaf, ei,
231 compress_type);
232 } else {
233 page = find_get_page(inode->i_mapping,
234 start >> PAGE_SHIFT);
235 btrfs_set_file_extent_compression(leaf, ei, 0);
236 kaddr = kmap_atomic(page);
237 offset = start & (PAGE_SIZE - 1);
238 write_extent_buffer(leaf, kaddr + offset, ptr, size);
239 kunmap_atomic(kaddr);
240 put_page(page);
241 }
242 btrfs_mark_buffer_dirty(leaf);
243 btrfs_release_path(path);
244
245 /*
246 * we're an inline extent, so nobody can
247 * extend the file past i_size without locking
248 * a page we already have locked.
249 *
250 * We must do any isize and inode updates
251 * before we unlock the pages. Otherwise we
252 * could end up racing with unlink.
253 */
254 BTRFS_I(inode)->disk_i_size = inode->i_size;
255 ret = btrfs_update_inode(trans, root, inode);
256
257 fail:
258 return ret;
259 }
260
261
262 /*
263 * conditionally insert an inline extent into the file. This
264 * does the checks required to make sure the data is small enough
265 * to fit as an inline extent.
266 */
267 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
268 u64 end, size_t compressed_size,
269 int compress_type,
270 struct page **compressed_pages)
271 {
272 struct btrfs_root *root = BTRFS_I(inode)->root;
273 struct btrfs_fs_info *fs_info = root->fs_info;
274 struct btrfs_trans_handle *trans;
275 u64 isize = i_size_read(inode);
276 u64 actual_end = min(end + 1, isize);
277 u64 inline_len = actual_end - start;
278 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279 u64 data_len = inline_len;
280 int ret;
281 struct btrfs_path *path;
282 int extent_inserted = 0;
283 u32 extent_item_size;
284
285 if (compressed_size)
286 data_len = compressed_size;
287
288 if (start > 0 ||
289 actual_end > fs_info->sectorsize ||
290 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
291 (!compressed_size &&
292 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
293 end + 1 < isize ||
294 data_len > fs_info->max_inline) {
295 return 1;
296 }
297
298 path = btrfs_alloc_path();
299 if (!path)
300 return -ENOMEM;
301
302 trans = btrfs_join_transaction(root);
303 if (IS_ERR(trans)) {
304 btrfs_free_path(path);
305 return PTR_ERR(trans);
306 }
307 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
308
309 if (compressed_size && compressed_pages)
310 extent_item_size = btrfs_file_extent_calc_inline_size(
311 compressed_size);
312 else
313 extent_item_size = btrfs_file_extent_calc_inline_size(
314 inline_len);
315
316 ret = __btrfs_drop_extents(trans, root, inode, path,
317 start, aligned_end, NULL,
318 1, 1, extent_item_size, &extent_inserted);
319 if (ret) {
320 btrfs_abort_transaction(trans, ret);
321 goto out;
322 }
323
324 if (isize > actual_end)
325 inline_len = min_t(u64, isize, actual_end);
326 ret = insert_inline_extent(trans, path, extent_inserted,
327 root, inode, start,
328 inline_len, compressed_size,
329 compress_type, compressed_pages);
330 if (ret && ret != -ENOSPC) {
331 btrfs_abort_transaction(trans, ret);
332 goto out;
333 } else if (ret == -ENOSPC) {
334 ret = 1;
335 goto out;
336 }
337
338 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
340 out:
341 /*
342 * Don't forget to free the reserved space, as for inlined extent
343 * it won't count as data extent, free them directly here.
344 * And at reserve time, it's always aligned to page size, so
345 * just free one page here.
346 */
347 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
348 btrfs_free_path(path);
349 btrfs_end_transaction(trans);
350 return ret;
351 }
352
353 struct async_extent {
354 u64 start;
355 u64 ram_size;
356 u64 compressed_size;
357 struct page **pages;
358 unsigned long nr_pages;
359 int compress_type;
360 struct list_head list;
361 };
362
363 struct async_cow {
364 struct inode *inode;
365 struct btrfs_root *root;
366 struct page *locked_page;
367 u64 start;
368 u64 end;
369 unsigned int write_flags;
370 struct list_head extents;
371 struct btrfs_work work;
372 };
373
374 static noinline int add_async_extent(struct async_cow *cow,
375 u64 start, u64 ram_size,
376 u64 compressed_size,
377 struct page **pages,
378 unsigned long nr_pages,
379 int compress_type)
380 {
381 struct async_extent *async_extent;
382
383 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384 BUG_ON(!async_extent); /* -ENOMEM */
385 async_extent->start = start;
386 async_extent->ram_size = ram_size;
387 async_extent->compressed_size = compressed_size;
388 async_extent->pages = pages;
389 async_extent->nr_pages = nr_pages;
390 async_extent->compress_type = compress_type;
391 list_add_tail(&async_extent->list, &cow->extents);
392 return 0;
393 }
394
395 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
396 {
397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
398
399 /* force compress */
400 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
401 return 1;
402 /* defrag ioctl */
403 if (BTRFS_I(inode)->defrag_compress)
404 return 1;
405 /* bad compression ratios */
406 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
407 return 0;
408 if (btrfs_test_opt(fs_info, COMPRESS) ||
409 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
410 BTRFS_I(inode)->prop_compress)
411 return btrfs_compress_heuristic(inode, start, end);
412 return 0;
413 }
414
415 static inline void inode_should_defrag(struct btrfs_inode *inode,
416 u64 start, u64 end, u64 num_bytes, u64 small_write)
417 {
418 /* If this is a small write inside eof, kick off a defrag */
419 if (num_bytes < small_write &&
420 (start > 0 || end + 1 < inode->disk_i_size))
421 btrfs_add_inode_defrag(NULL, inode);
422 }
423
424 /*
425 * we create compressed extents in two phases. The first
426 * phase compresses a range of pages that have already been
427 * locked (both pages and state bits are locked).
428 *
429 * This is done inside an ordered work queue, and the compression
430 * is spread across many cpus. The actual IO submission is step
431 * two, and the ordered work queue takes care of making sure that
432 * happens in the same order things were put onto the queue by
433 * writepages and friends.
434 *
435 * If this code finds it can't get good compression, it puts an
436 * entry onto the work queue to write the uncompressed bytes. This
437 * makes sure that both compressed inodes and uncompressed inodes
438 * are written in the same order that the flusher thread sent them
439 * down.
440 */
441 static noinline void compress_file_range(struct inode *inode,
442 struct page *locked_page,
443 u64 start, u64 end,
444 struct async_cow *async_cow,
445 int *num_added)
446 {
447 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
448 u64 blocksize = fs_info->sectorsize;
449 u64 actual_end;
450 u64 isize = i_size_read(inode);
451 int ret = 0;
452 struct page **pages = NULL;
453 unsigned long nr_pages;
454 unsigned long total_compressed = 0;
455 unsigned long total_in = 0;
456 int i;
457 int will_compress;
458 int compress_type = fs_info->compress_type;
459 int redirty = 0;
460
461 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
462 SZ_16K);
463
464 actual_end = min_t(u64, isize, end + 1);
465 again:
466 will_compress = 0;
467 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
468 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
469 nr_pages = min_t(unsigned long, nr_pages,
470 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
471
472 /*
473 * we don't want to send crud past the end of i_size through
474 * compression, that's just a waste of CPU time. So, if the
475 * end of the file is before the start of our current
476 * requested range of bytes, we bail out to the uncompressed
477 * cleanup code that can deal with all of this.
478 *
479 * It isn't really the fastest way to fix things, but this is a
480 * very uncommon corner.
481 */
482 if (actual_end <= start)
483 goto cleanup_and_bail_uncompressed;
484
485 total_compressed = actual_end - start;
486
487 /*
488 * skip compression for a small file range(<=blocksize) that
489 * isn't an inline extent, since it doesn't save disk space at all.
490 */
491 if (total_compressed <= blocksize &&
492 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
493 goto cleanup_and_bail_uncompressed;
494
495 total_compressed = min_t(unsigned long, total_compressed,
496 BTRFS_MAX_UNCOMPRESSED);
497 total_in = 0;
498 ret = 0;
499
500 /*
501 * we do compression for mount -o compress and when the
502 * inode has not been flagged as nocompress. This flag can
503 * change at any time if we discover bad compression ratios.
504 */
505 if (inode_need_compress(inode, start, end)) {
506 WARN_ON(pages);
507 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
508 if (!pages) {
509 /* just bail out to the uncompressed code */
510 goto cont;
511 }
512
513 if (BTRFS_I(inode)->defrag_compress)
514 compress_type = BTRFS_I(inode)->defrag_compress;
515 else if (BTRFS_I(inode)->prop_compress)
516 compress_type = BTRFS_I(inode)->prop_compress;
517
518 /*
519 * we need to call clear_page_dirty_for_io on each
520 * page in the range. Otherwise applications with the file
521 * mmap'd can wander in and change the page contents while
522 * we are compressing them.
523 *
524 * If the compression fails for any reason, we set the pages
525 * dirty again later on.
526 *
527 * Note that the remaining part is redirtied, the start pointer
528 * has moved, the end is the original one.
529 */
530 if (!redirty) {
531 extent_range_clear_dirty_for_io(inode, start, end);
532 redirty = 1;
533 }
534
535 /* Compression level is applied here and only here */
536 ret = btrfs_compress_pages(
537 compress_type | (fs_info->compress_level << 4),
538 inode->i_mapping, start,
539 pages,
540 &nr_pages,
541 &total_in,
542 &total_compressed);
543
544 if (!ret) {
545 unsigned long offset = total_compressed &
546 (PAGE_SIZE - 1);
547 struct page *page = pages[nr_pages - 1];
548 char *kaddr;
549
550 /* zero the tail end of the last page, we might be
551 * sending it down to disk
552 */
553 if (offset) {
554 kaddr = kmap_atomic(page);
555 memset(kaddr + offset, 0,
556 PAGE_SIZE - offset);
557 kunmap_atomic(kaddr);
558 }
559 will_compress = 1;
560 }
561 }
562 cont:
563 if (start == 0) {
564 /* lets try to make an inline extent */
565 if (ret || total_in < actual_end) {
566 /* we didn't compress the entire range, try
567 * to make an uncompressed inline extent.
568 */
569 ret = cow_file_range_inline(inode, start, end, 0,
570 BTRFS_COMPRESS_NONE, NULL);
571 } else {
572 /* try making a compressed inline extent */
573 ret = cow_file_range_inline(inode, start, end,
574 total_compressed,
575 compress_type, pages);
576 }
577 if (ret <= 0) {
578 unsigned long clear_flags = EXTENT_DELALLOC |
579 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
580 EXTENT_DO_ACCOUNTING;
581 unsigned long page_error_op;
582
583 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
584
585 /*
586 * inline extent creation worked or returned error,
587 * we don't need to create any more async work items.
588 * Unlock and free up our temp pages.
589 *
590 * We use DO_ACCOUNTING here because we need the
591 * delalloc_release_metadata to be done _after_ we drop
592 * our outstanding extent for clearing delalloc for this
593 * range.
594 */
595 extent_clear_unlock_delalloc(inode, start, end, end,
596 NULL, clear_flags,
597 PAGE_UNLOCK |
598 PAGE_CLEAR_DIRTY |
599 PAGE_SET_WRITEBACK |
600 page_error_op |
601 PAGE_END_WRITEBACK);
602 goto free_pages_out;
603 }
604 }
605
606 if (will_compress) {
607 /*
608 * we aren't doing an inline extent round the compressed size
609 * up to a block size boundary so the allocator does sane
610 * things
611 */
612 total_compressed = ALIGN(total_compressed, blocksize);
613
614 /*
615 * one last check to make sure the compression is really a
616 * win, compare the page count read with the blocks on disk,
617 * compression must free at least one sector size
618 */
619 total_in = ALIGN(total_in, PAGE_SIZE);
620 if (total_compressed + blocksize <= total_in) {
621 *num_added += 1;
622
623 /*
624 * The async work queues will take care of doing actual
625 * allocation on disk for these compressed pages, and
626 * will submit them to the elevator.
627 */
628 add_async_extent(async_cow, start, total_in,
629 total_compressed, pages, nr_pages,
630 compress_type);
631
632 if (start + total_in < end) {
633 start += total_in;
634 pages = NULL;
635 cond_resched();
636 goto again;
637 }
638 return;
639 }
640 }
641 if (pages) {
642 /*
643 * the compression code ran but failed to make things smaller,
644 * free any pages it allocated and our page pointer array
645 */
646 for (i = 0; i < nr_pages; i++) {
647 WARN_ON(pages[i]->mapping);
648 put_page(pages[i]);
649 }
650 kfree(pages);
651 pages = NULL;
652 total_compressed = 0;
653 nr_pages = 0;
654
655 /* flag the file so we don't compress in the future */
656 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
657 !(BTRFS_I(inode)->prop_compress)) {
658 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
659 }
660 }
661 cleanup_and_bail_uncompressed:
662 /*
663 * No compression, but we still need to write the pages in the file
664 * we've been given so far. redirty the locked page if it corresponds
665 * to our extent and set things up for the async work queue to run
666 * cow_file_range to do the normal delalloc dance.
667 */
668 if (page_offset(locked_page) >= start &&
669 page_offset(locked_page) <= end)
670 __set_page_dirty_nobuffers(locked_page);
671 /* unlocked later on in the async handlers */
672
673 if (redirty)
674 extent_range_redirty_for_io(inode, start, end);
675 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
676 BTRFS_COMPRESS_NONE);
677 *num_added += 1;
678
679 return;
680
681 free_pages_out:
682 for (i = 0; i < nr_pages; i++) {
683 WARN_ON(pages[i]->mapping);
684 put_page(pages[i]);
685 }
686 kfree(pages);
687 }
688
689 static void free_async_extent_pages(struct async_extent *async_extent)
690 {
691 int i;
692
693 if (!async_extent->pages)
694 return;
695
696 for (i = 0; i < async_extent->nr_pages; i++) {
697 WARN_ON(async_extent->pages[i]->mapping);
698 put_page(async_extent->pages[i]);
699 }
700 kfree(async_extent->pages);
701 async_extent->nr_pages = 0;
702 async_extent->pages = NULL;
703 }
704
705 /*
706 * phase two of compressed writeback. This is the ordered portion
707 * of the code, which only gets called in the order the work was
708 * queued. We walk all the async extents created by compress_file_range
709 * and send them down to the disk.
710 */
711 static noinline void submit_compressed_extents(struct inode *inode,
712 struct async_cow *async_cow)
713 {
714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
715 struct async_extent *async_extent;
716 u64 alloc_hint = 0;
717 struct btrfs_key ins;
718 struct extent_map *em;
719 struct btrfs_root *root = BTRFS_I(inode)->root;
720 struct extent_io_tree *io_tree;
721 int ret = 0;
722
723 again:
724 while (!list_empty(&async_cow->extents)) {
725 async_extent = list_entry(async_cow->extents.next,
726 struct async_extent, list);
727 list_del(&async_extent->list);
728
729 io_tree = &BTRFS_I(inode)->io_tree;
730
731 retry:
732 /* did the compression code fall back to uncompressed IO? */
733 if (!async_extent->pages) {
734 int page_started = 0;
735 unsigned long nr_written = 0;
736
737 lock_extent(io_tree, async_extent->start,
738 async_extent->start +
739 async_extent->ram_size - 1);
740
741 /* allocate blocks */
742 ret = cow_file_range(inode, async_cow->locked_page,
743 async_extent->start,
744 async_extent->start +
745 async_extent->ram_size - 1,
746 async_extent->start +
747 async_extent->ram_size - 1,
748 &page_started, &nr_written, 0,
749 NULL);
750
751 /* JDM XXX */
752
753 /*
754 * if page_started, cow_file_range inserted an
755 * inline extent and took care of all the unlocking
756 * and IO for us. Otherwise, we need to submit
757 * all those pages down to the drive.
758 */
759 if (!page_started && !ret)
760 extent_write_locked_range(inode,
761 async_extent->start,
762 async_extent->start +
763 async_extent->ram_size - 1,
764 WB_SYNC_ALL);
765 else if (ret)
766 unlock_page(async_cow->locked_page);
767 kfree(async_extent);
768 cond_resched();
769 continue;
770 }
771
772 lock_extent(io_tree, async_extent->start,
773 async_extent->start + async_extent->ram_size - 1);
774
775 ret = btrfs_reserve_extent(root, async_extent->ram_size,
776 async_extent->compressed_size,
777 async_extent->compressed_size,
778 0, alloc_hint, &ins, 1, 1);
779 if (ret) {
780 free_async_extent_pages(async_extent);
781
782 if (ret == -ENOSPC) {
783 unlock_extent(io_tree, async_extent->start,
784 async_extent->start +
785 async_extent->ram_size - 1);
786
787 /*
788 * we need to redirty the pages if we decide to
789 * fallback to uncompressed IO, otherwise we
790 * will not submit these pages down to lower
791 * layers.
792 */
793 extent_range_redirty_for_io(inode,
794 async_extent->start,
795 async_extent->start +
796 async_extent->ram_size - 1);
797
798 goto retry;
799 }
800 goto out_free;
801 }
802 /*
803 * here we're doing allocation and writeback of the
804 * compressed pages
805 */
806 em = create_io_em(inode, async_extent->start,
807 async_extent->ram_size, /* len */
808 async_extent->start, /* orig_start */
809 ins.objectid, /* block_start */
810 ins.offset, /* block_len */
811 ins.offset, /* orig_block_len */
812 async_extent->ram_size, /* ram_bytes */
813 async_extent->compress_type,
814 BTRFS_ORDERED_COMPRESSED);
815 if (IS_ERR(em))
816 /* ret value is not necessary due to void function */
817 goto out_free_reserve;
818 free_extent_map(em);
819
820 ret = btrfs_add_ordered_extent_compress(inode,
821 async_extent->start,
822 ins.objectid,
823 async_extent->ram_size,
824 ins.offset,
825 BTRFS_ORDERED_COMPRESSED,
826 async_extent->compress_type);
827 if (ret) {
828 btrfs_drop_extent_cache(BTRFS_I(inode),
829 async_extent->start,
830 async_extent->start +
831 async_extent->ram_size - 1, 0);
832 goto out_free_reserve;
833 }
834 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
835
836 /*
837 * clear dirty, set writeback and unlock the pages.
838 */
839 extent_clear_unlock_delalloc(inode, async_extent->start,
840 async_extent->start +
841 async_extent->ram_size - 1,
842 async_extent->start +
843 async_extent->ram_size - 1,
844 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
845 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
846 PAGE_SET_WRITEBACK);
847 if (btrfs_submit_compressed_write(inode,
848 async_extent->start,
849 async_extent->ram_size,
850 ins.objectid,
851 ins.offset, async_extent->pages,
852 async_extent->nr_pages,
853 async_cow->write_flags)) {
854 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
855 struct page *p = async_extent->pages[0];
856 const u64 start = async_extent->start;
857 const u64 end = start + async_extent->ram_size - 1;
858
859 p->mapping = inode->i_mapping;
860 tree->ops->writepage_end_io_hook(p, start, end,
861 NULL, 0);
862 p->mapping = NULL;
863 extent_clear_unlock_delalloc(inode, start, end, end,
864 NULL, 0,
865 PAGE_END_WRITEBACK |
866 PAGE_SET_ERROR);
867 free_async_extent_pages(async_extent);
868 }
869 alloc_hint = ins.objectid + ins.offset;
870 kfree(async_extent);
871 cond_resched();
872 }
873 return;
874 out_free_reserve:
875 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
876 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
877 out_free:
878 extent_clear_unlock_delalloc(inode, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1,
881 async_extent->start +
882 async_extent->ram_size - 1,
883 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
884 EXTENT_DELALLOC_NEW |
885 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
886 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
887 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
888 PAGE_SET_ERROR);
889 free_async_extent_pages(async_extent);
890 kfree(async_extent);
891 goto again;
892 }
893
894 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
895 u64 num_bytes)
896 {
897 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
898 struct extent_map *em;
899 u64 alloc_hint = 0;
900
901 read_lock(&em_tree->lock);
902 em = search_extent_mapping(em_tree, start, num_bytes);
903 if (em) {
904 /*
905 * if block start isn't an actual block number then find the
906 * first block in this inode and use that as a hint. If that
907 * block is also bogus then just don't worry about it.
908 */
909 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
910 free_extent_map(em);
911 em = search_extent_mapping(em_tree, 0, 0);
912 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
913 alloc_hint = em->block_start;
914 if (em)
915 free_extent_map(em);
916 } else {
917 alloc_hint = em->block_start;
918 free_extent_map(em);
919 }
920 }
921 read_unlock(&em_tree->lock);
922
923 return alloc_hint;
924 }
925
926 /*
927 * when extent_io.c finds a delayed allocation range in the file,
928 * the call backs end up in this code. The basic idea is to
929 * allocate extents on disk for the range, and create ordered data structs
930 * in ram to track those extents.
931 *
932 * locked_page is the page that writepage had locked already. We use
933 * it to make sure we don't do extra locks or unlocks.
934 *
935 * *page_started is set to one if we unlock locked_page and do everything
936 * required to start IO on it. It may be clean and already done with
937 * IO when we return.
938 */
939 static noinline int cow_file_range(struct inode *inode,
940 struct page *locked_page,
941 u64 start, u64 end, u64 delalloc_end,
942 int *page_started, unsigned long *nr_written,
943 int unlock, struct btrfs_dedupe_hash *hash)
944 {
945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
946 struct btrfs_root *root = BTRFS_I(inode)->root;
947 u64 alloc_hint = 0;
948 u64 num_bytes;
949 unsigned long ram_size;
950 u64 cur_alloc_size = 0;
951 u64 blocksize = fs_info->sectorsize;
952 struct btrfs_key ins;
953 struct extent_map *em;
954 unsigned clear_bits;
955 unsigned long page_ops;
956 bool extent_reserved = false;
957 int ret = 0;
958
959 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
960 WARN_ON_ONCE(1);
961 ret = -EINVAL;
962 goto out_unlock;
963 }
964
965 num_bytes = ALIGN(end - start + 1, blocksize);
966 num_bytes = max(blocksize, num_bytes);
967 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
968
969 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
970
971 if (start == 0) {
972 /* lets try to make an inline extent */
973 ret = cow_file_range_inline(inode, start, end, 0,
974 BTRFS_COMPRESS_NONE, NULL);
975 if (ret == 0) {
976 /*
977 * We use DO_ACCOUNTING here because we need the
978 * delalloc_release_metadata to be run _after_ we drop
979 * our outstanding extent for clearing delalloc for this
980 * range.
981 */
982 extent_clear_unlock_delalloc(inode, start, end,
983 delalloc_end, NULL,
984 EXTENT_LOCKED | EXTENT_DELALLOC |
985 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
986 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
987 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
988 PAGE_END_WRITEBACK);
989 *nr_written = *nr_written +
990 (end - start + PAGE_SIZE) / PAGE_SIZE;
991 *page_started = 1;
992 goto out;
993 } else if (ret < 0) {
994 goto out_unlock;
995 }
996 }
997
998 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
999 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1000 start + num_bytes - 1, 0);
1001
1002 while (num_bytes > 0) {
1003 cur_alloc_size = num_bytes;
1004 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1005 fs_info->sectorsize, 0, alloc_hint,
1006 &ins, 1, 1);
1007 if (ret < 0)
1008 goto out_unlock;
1009 cur_alloc_size = ins.offset;
1010 extent_reserved = true;
1011
1012 ram_size = ins.offset;
1013 em = create_io_em(inode, start, ins.offset, /* len */
1014 start, /* orig_start */
1015 ins.objectid, /* block_start */
1016 ins.offset, /* block_len */
1017 ins.offset, /* orig_block_len */
1018 ram_size, /* ram_bytes */
1019 BTRFS_COMPRESS_NONE, /* compress_type */
1020 BTRFS_ORDERED_REGULAR /* type */);
1021 if (IS_ERR(em))
1022 goto out_reserve;
1023 free_extent_map(em);
1024
1025 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1026 ram_size, cur_alloc_size, 0);
1027 if (ret)
1028 goto out_drop_extent_cache;
1029
1030 if (root->root_key.objectid ==
1031 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1032 ret = btrfs_reloc_clone_csums(inode, start,
1033 cur_alloc_size);
1034 /*
1035 * Only drop cache here, and process as normal.
1036 *
1037 * We must not allow extent_clear_unlock_delalloc()
1038 * at out_unlock label to free meta of this ordered
1039 * extent, as its meta should be freed by
1040 * btrfs_finish_ordered_io().
1041 *
1042 * So we must continue until @start is increased to
1043 * skip current ordered extent.
1044 */
1045 if (ret)
1046 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1047 start + ram_size - 1, 0);
1048 }
1049
1050 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1051
1052 /* we're not doing compressed IO, don't unlock the first
1053 * page (which the caller expects to stay locked), don't
1054 * clear any dirty bits and don't set any writeback bits
1055 *
1056 * Do set the Private2 bit so we know this page was properly
1057 * setup for writepage
1058 */
1059 page_ops = unlock ? PAGE_UNLOCK : 0;
1060 page_ops |= PAGE_SET_PRIVATE2;
1061
1062 extent_clear_unlock_delalloc(inode, start,
1063 start + ram_size - 1,
1064 delalloc_end, locked_page,
1065 EXTENT_LOCKED | EXTENT_DELALLOC,
1066 page_ops);
1067 if (num_bytes < cur_alloc_size)
1068 num_bytes = 0;
1069 else
1070 num_bytes -= cur_alloc_size;
1071 alloc_hint = ins.objectid + ins.offset;
1072 start += cur_alloc_size;
1073 extent_reserved = false;
1074
1075 /*
1076 * btrfs_reloc_clone_csums() error, since start is increased
1077 * extent_clear_unlock_delalloc() at out_unlock label won't
1078 * free metadata of current ordered extent, we're OK to exit.
1079 */
1080 if (ret)
1081 goto out_unlock;
1082 }
1083 out:
1084 return ret;
1085
1086 out_drop_extent_cache:
1087 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1088 out_reserve:
1089 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1090 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1091 out_unlock:
1092 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1093 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1094 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1095 PAGE_END_WRITEBACK;
1096 /*
1097 * If we reserved an extent for our delalloc range (or a subrange) and
1098 * failed to create the respective ordered extent, then it means that
1099 * when we reserved the extent we decremented the extent's size from
1100 * the data space_info's bytes_may_use counter and incremented the
1101 * space_info's bytes_reserved counter by the same amount. We must make
1102 * sure extent_clear_unlock_delalloc() does not try to decrement again
1103 * the data space_info's bytes_may_use counter, therefore we do not pass
1104 * it the flag EXTENT_CLEAR_DATA_RESV.
1105 */
1106 if (extent_reserved) {
1107 extent_clear_unlock_delalloc(inode, start,
1108 start + cur_alloc_size,
1109 start + cur_alloc_size,
1110 locked_page,
1111 clear_bits,
1112 page_ops);
1113 start += cur_alloc_size;
1114 if (start >= end)
1115 goto out;
1116 }
1117 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1118 locked_page,
1119 clear_bits | EXTENT_CLEAR_DATA_RESV,
1120 page_ops);
1121 goto out;
1122 }
1123
1124 /*
1125 * work queue call back to started compression on a file and pages
1126 */
1127 static noinline void async_cow_start(struct btrfs_work *work)
1128 {
1129 struct async_cow *async_cow;
1130 int num_added = 0;
1131 async_cow = container_of(work, struct async_cow, work);
1132
1133 compress_file_range(async_cow->inode, async_cow->locked_page,
1134 async_cow->start, async_cow->end, async_cow,
1135 &num_added);
1136 if (num_added == 0) {
1137 btrfs_add_delayed_iput(async_cow->inode);
1138 async_cow->inode = NULL;
1139 }
1140 }
1141
1142 /*
1143 * work queue call back to submit previously compressed pages
1144 */
1145 static noinline void async_cow_submit(struct btrfs_work *work)
1146 {
1147 struct btrfs_fs_info *fs_info;
1148 struct async_cow *async_cow;
1149 struct btrfs_root *root;
1150 unsigned long nr_pages;
1151
1152 async_cow = container_of(work, struct async_cow, work);
1153
1154 root = async_cow->root;
1155 fs_info = root->fs_info;
1156 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1157 PAGE_SHIFT;
1158
1159 /*
1160 * atomic_sub_return implies a barrier for waitqueue_active
1161 */
1162 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1163 5 * SZ_1M &&
1164 waitqueue_active(&fs_info->async_submit_wait))
1165 wake_up(&fs_info->async_submit_wait);
1166
1167 if (async_cow->inode)
1168 submit_compressed_extents(async_cow->inode, async_cow);
1169 }
1170
1171 static noinline void async_cow_free(struct btrfs_work *work)
1172 {
1173 struct async_cow *async_cow;
1174 async_cow = container_of(work, struct async_cow, work);
1175 if (async_cow->inode)
1176 btrfs_add_delayed_iput(async_cow->inode);
1177 kfree(async_cow);
1178 }
1179
1180 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1181 u64 start, u64 end, int *page_started,
1182 unsigned long *nr_written,
1183 unsigned int write_flags)
1184 {
1185 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1186 struct async_cow *async_cow;
1187 struct btrfs_root *root = BTRFS_I(inode)->root;
1188 unsigned long nr_pages;
1189 u64 cur_end;
1190
1191 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1192 1, 0, NULL);
1193 while (start < end) {
1194 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1195 BUG_ON(!async_cow); /* -ENOMEM */
1196 async_cow->inode = igrab(inode);
1197 async_cow->root = root;
1198 async_cow->locked_page = locked_page;
1199 async_cow->start = start;
1200 async_cow->write_flags = write_flags;
1201
1202 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1203 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1204 cur_end = end;
1205 else
1206 cur_end = min(end, start + SZ_512K - 1);
1207
1208 async_cow->end = cur_end;
1209 INIT_LIST_HEAD(&async_cow->extents);
1210
1211 btrfs_init_work(&async_cow->work,
1212 btrfs_delalloc_helper,
1213 async_cow_start, async_cow_submit,
1214 async_cow_free);
1215
1216 nr_pages = (cur_end - start + PAGE_SIZE) >>
1217 PAGE_SHIFT;
1218 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1219
1220 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1221
1222 *nr_written += nr_pages;
1223 start = cur_end + 1;
1224 }
1225 *page_started = 1;
1226 return 0;
1227 }
1228
1229 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1230 u64 bytenr, u64 num_bytes)
1231 {
1232 int ret;
1233 struct btrfs_ordered_sum *sums;
1234 LIST_HEAD(list);
1235
1236 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1237 bytenr + num_bytes - 1, &list, 0);
1238 if (ret == 0 && list_empty(&list))
1239 return 0;
1240
1241 while (!list_empty(&list)) {
1242 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1243 list_del(&sums->list);
1244 kfree(sums);
1245 }
1246 if (ret < 0)
1247 return ret;
1248 return 1;
1249 }
1250
1251 /*
1252 * when nowcow writeback call back. This checks for snapshots or COW copies
1253 * of the extents that exist in the file, and COWs the file as required.
1254 *
1255 * If no cow copies or snapshots exist, we write directly to the existing
1256 * blocks on disk
1257 */
1258 static noinline int run_delalloc_nocow(struct inode *inode,
1259 struct page *locked_page,
1260 u64 start, u64 end, int *page_started, int force,
1261 unsigned long *nr_written)
1262 {
1263 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1264 struct btrfs_root *root = BTRFS_I(inode)->root;
1265 struct extent_buffer *leaf;
1266 struct btrfs_path *path;
1267 struct btrfs_file_extent_item *fi;
1268 struct btrfs_key found_key;
1269 struct extent_map *em;
1270 u64 cow_start;
1271 u64 cur_offset;
1272 u64 extent_end;
1273 u64 extent_offset;
1274 u64 disk_bytenr;
1275 u64 num_bytes;
1276 u64 disk_num_bytes;
1277 u64 ram_bytes;
1278 int extent_type;
1279 int ret, err;
1280 int type;
1281 int nocow;
1282 int check_prev = 1;
1283 bool nolock;
1284 u64 ino = btrfs_ino(BTRFS_I(inode));
1285
1286 path = btrfs_alloc_path();
1287 if (!path) {
1288 extent_clear_unlock_delalloc(inode, start, end, end,
1289 locked_page,
1290 EXTENT_LOCKED | EXTENT_DELALLOC |
1291 EXTENT_DO_ACCOUNTING |
1292 EXTENT_DEFRAG, PAGE_UNLOCK |
1293 PAGE_CLEAR_DIRTY |
1294 PAGE_SET_WRITEBACK |
1295 PAGE_END_WRITEBACK);
1296 return -ENOMEM;
1297 }
1298
1299 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1300
1301 cow_start = (u64)-1;
1302 cur_offset = start;
1303 while (1) {
1304 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1305 cur_offset, 0);
1306 if (ret < 0)
1307 goto error;
1308 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1309 leaf = path->nodes[0];
1310 btrfs_item_key_to_cpu(leaf, &found_key,
1311 path->slots[0] - 1);
1312 if (found_key.objectid == ino &&
1313 found_key.type == BTRFS_EXTENT_DATA_KEY)
1314 path->slots[0]--;
1315 }
1316 check_prev = 0;
1317 next_slot:
1318 leaf = path->nodes[0];
1319 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1320 ret = btrfs_next_leaf(root, path);
1321 if (ret < 0) {
1322 if (cow_start != (u64)-1)
1323 cur_offset = cow_start;
1324 goto error;
1325 }
1326 if (ret > 0)
1327 break;
1328 leaf = path->nodes[0];
1329 }
1330
1331 nocow = 0;
1332 disk_bytenr = 0;
1333 num_bytes = 0;
1334 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1335
1336 if (found_key.objectid > ino)
1337 break;
1338 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1339 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1340 path->slots[0]++;
1341 goto next_slot;
1342 }
1343 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1344 found_key.offset > end)
1345 break;
1346
1347 if (found_key.offset > cur_offset) {
1348 extent_end = found_key.offset;
1349 extent_type = 0;
1350 goto out_check;
1351 }
1352
1353 fi = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_file_extent_item);
1355 extent_type = btrfs_file_extent_type(leaf, fi);
1356
1357 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1358 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1359 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1360 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1361 extent_offset = btrfs_file_extent_offset(leaf, fi);
1362 extent_end = found_key.offset +
1363 btrfs_file_extent_num_bytes(leaf, fi);
1364 disk_num_bytes =
1365 btrfs_file_extent_disk_num_bytes(leaf, fi);
1366 if (extent_end <= start) {
1367 path->slots[0]++;
1368 goto next_slot;
1369 }
1370 if (disk_bytenr == 0)
1371 goto out_check;
1372 if (btrfs_file_extent_compression(leaf, fi) ||
1373 btrfs_file_extent_encryption(leaf, fi) ||
1374 btrfs_file_extent_other_encoding(leaf, fi))
1375 goto out_check;
1376 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1377 goto out_check;
1378 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1379 goto out_check;
1380 ret = btrfs_cross_ref_exist(root, ino,
1381 found_key.offset -
1382 extent_offset, disk_bytenr);
1383 if (ret) {
1384 /*
1385 * ret could be -EIO if the above fails to read
1386 * metadata.
1387 */
1388 if (ret < 0) {
1389 if (cow_start != (u64)-1)
1390 cur_offset = cow_start;
1391 goto error;
1392 }
1393
1394 WARN_ON_ONCE(nolock);
1395 goto out_check;
1396 }
1397 disk_bytenr += extent_offset;
1398 disk_bytenr += cur_offset - found_key.offset;
1399 num_bytes = min(end + 1, extent_end) - cur_offset;
1400 /*
1401 * if there are pending snapshots for this root,
1402 * we fall into common COW way.
1403 */
1404 if (!nolock) {
1405 err = btrfs_start_write_no_snapshotting(root);
1406 if (!err)
1407 goto out_check;
1408 }
1409 /*
1410 * force cow if csum exists in the range.
1411 * this ensure that csum for a given extent are
1412 * either valid or do not exist.
1413 */
1414 ret = csum_exist_in_range(fs_info, disk_bytenr,
1415 num_bytes);
1416 if (ret) {
1417 if (!nolock)
1418 btrfs_end_write_no_snapshotting(root);
1419
1420 /*
1421 * ret could be -EIO if the above fails to read
1422 * metadata.
1423 */
1424 if (ret < 0) {
1425 if (cow_start != (u64)-1)
1426 cur_offset = cow_start;
1427 goto error;
1428 }
1429 WARN_ON_ONCE(nolock);
1430 goto out_check;
1431 }
1432 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1433 if (!nolock)
1434 btrfs_end_write_no_snapshotting(root);
1435 goto out_check;
1436 }
1437 nocow = 1;
1438 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1439 extent_end = found_key.offset +
1440 btrfs_file_extent_inline_len(leaf,
1441 path->slots[0], fi);
1442 extent_end = ALIGN(extent_end,
1443 fs_info->sectorsize);
1444 } else {
1445 BUG_ON(1);
1446 }
1447 out_check:
1448 if (extent_end <= start) {
1449 path->slots[0]++;
1450 if (!nolock && nocow)
1451 btrfs_end_write_no_snapshotting(root);
1452 if (nocow)
1453 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1454 goto next_slot;
1455 }
1456 if (!nocow) {
1457 if (cow_start == (u64)-1)
1458 cow_start = cur_offset;
1459 cur_offset = extent_end;
1460 if (cur_offset > end)
1461 break;
1462 path->slots[0]++;
1463 goto next_slot;
1464 }
1465
1466 btrfs_release_path(path);
1467 if (cow_start != (u64)-1) {
1468 ret = cow_file_range(inode, locked_page,
1469 cow_start, found_key.offset - 1,
1470 end, page_started, nr_written, 1,
1471 NULL);
1472 if (ret) {
1473 if (!nolock && nocow)
1474 btrfs_end_write_no_snapshotting(root);
1475 if (nocow)
1476 btrfs_dec_nocow_writers(fs_info,
1477 disk_bytenr);
1478 goto error;
1479 }
1480 cow_start = (u64)-1;
1481 }
1482
1483 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1484 u64 orig_start = found_key.offset - extent_offset;
1485
1486 em = create_io_em(inode, cur_offset, num_bytes,
1487 orig_start,
1488 disk_bytenr, /* block_start */
1489 num_bytes, /* block_len */
1490 disk_num_bytes, /* orig_block_len */
1491 ram_bytes, BTRFS_COMPRESS_NONE,
1492 BTRFS_ORDERED_PREALLOC);
1493 if (IS_ERR(em)) {
1494 if (!nolock && nocow)
1495 btrfs_end_write_no_snapshotting(root);
1496 if (nocow)
1497 btrfs_dec_nocow_writers(fs_info,
1498 disk_bytenr);
1499 ret = PTR_ERR(em);
1500 goto error;
1501 }
1502 free_extent_map(em);
1503 }
1504
1505 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1506 type = BTRFS_ORDERED_PREALLOC;
1507 } else {
1508 type = BTRFS_ORDERED_NOCOW;
1509 }
1510
1511 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1512 num_bytes, num_bytes, type);
1513 if (nocow)
1514 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1515 BUG_ON(ret); /* -ENOMEM */
1516
1517 if (root->root_key.objectid ==
1518 BTRFS_DATA_RELOC_TREE_OBJECTID)
1519 /*
1520 * Error handled later, as we must prevent
1521 * extent_clear_unlock_delalloc() in error handler
1522 * from freeing metadata of created ordered extent.
1523 */
1524 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1525 num_bytes);
1526
1527 extent_clear_unlock_delalloc(inode, cur_offset,
1528 cur_offset + num_bytes - 1, end,
1529 locked_page, EXTENT_LOCKED |
1530 EXTENT_DELALLOC |
1531 EXTENT_CLEAR_DATA_RESV,
1532 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1533
1534 if (!nolock && nocow)
1535 btrfs_end_write_no_snapshotting(root);
1536 cur_offset = extent_end;
1537
1538 /*
1539 * btrfs_reloc_clone_csums() error, now we're OK to call error
1540 * handler, as metadata for created ordered extent will only
1541 * be freed by btrfs_finish_ordered_io().
1542 */
1543 if (ret)
1544 goto error;
1545 if (cur_offset > end)
1546 break;
1547 }
1548 btrfs_release_path(path);
1549
1550 if (cur_offset <= end && cow_start == (u64)-1) {
1551 cow_start = cur_offset;
1552 cur_offset = end;
1553 }
1554
1555 if (cow_start != (u64)-1) {
1556 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1557 page_started, nr_written, 1, NULL);
1558 if (ret)
1559 goto error;
1560 }
1561
1562 error:
1563 if (ret && cur_offset < end)
1564 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1565 locked_page, EXTENT_LOCKED |
1566 EXTENT_DELALLOC | EXTENT_DEFRAG |
1567 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1568 PAGE_CLEAR_DIRTY |
1569 PAGE_SET_WRITEBACK |
1570 PAGE_END_WRITEBACK);
1571 btrfs_free_path(path);
1572 return ret;
1573 }
1574
1575 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1576 {
1577
1578 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1579 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1580 return 0;
1581
1582 /*
1583 * @defrag_bytes is a hint value, no spinlock held here,
1584 * if is not zero, it means the file is defragging.
1585 * Force cow if given extent needs to be defragged.
1586 */
1587 if (BTRFS_I(inode)->defrag_bytes &&
1588 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1589 EXTENT_DEFRAG, 0, NULL))
1590 return 1;
1591
1592 return 0;
1593 }
1594
1595 /*
1596 * extent_io.c call back to do delayed allocation processing
1597 */
1598 static int run_delalloc_range(void *private_data, struct page *locked_page,
1599 u64 start, u64 end, int *page_started,
1600 unsigned long *nr_written,
1601 struct writeback_control *wbc)
1602 {
1603 struct inode *inode = private_data;
1604 int ret;
1605 int force_cow = need_force_cow(inode, start, end);
1606 unsigned int write_flags = wbc_to_write_flags(wbc);
1607
1608 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1609 ret = run_delalloc_nocow(inode, locked_page, start, end,
1610 page_started, 1, nr_written);
1611 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1612 ret = run_delalloc_nocow(inode, locked_page, start, end,
1613 page_started, 0, nr_written);
1614 } else if (!inode_need_compress(inode, start, end)) {
1615 ret = cow_file_range(inode, locked_page, start, end, end,
1616 page_started, nr_written, 1, NULL);
1617 } else {
1618 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1619 &BTRFS_I(inode)->runtime_flags);
1620 ret = cow_file_range_async(inode, locked_page, start, end,
1621 page_started, nr_written,
1622 write_flags);
1623 }
1624 if (ret)
1625 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1626 return ret;
1627 }
1628
1629 static void btrfs_split_extent_hook(void *private_data,
1630 struct extent_state *orig, u64 split)
1631 {
1632 struct inode *inode = private_data;
1633 u64 size;
1634
1635 /* not delalloc, ignore it */
1636 if (!(orig->state & EXTENT_DELALLOC))
1637 return;
1638
1639 size = orig->end - orig->start + 1;
1640 if (size > BTRFS_MAX_EXTENT_SIZE) {
1641 u32 num_extents;
1642 u64 new_size;
1643
1644 /*
1645 * See the explanation in btrfs_merge_extent_hook, the same
1646 * applies here, just in reverse.
1647 */
1648 new_size = orig->end - split + 1;
1649 num_extents = count_max_extents(new_size);
1650 new_size = split - orig->start;
1651 num_extents += count_max_extents(new_size);
1652 if (count_max_extents(size) >= num_extents)
1653 return;
1654 }
1655
1656 spin_lock(&BTRFS_I(inode)->lock);
1657 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1658 spin_unlock(&BTRFS_I(inode)->lock);
1659 }
1660
1661 /*
1662 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1663 * extents so we can keep track of new extents that are just merged onto old
1664 * extents, such as when we are doing sequential writes, so we can properly
1665 * account for the metadata space we'll need.
1666 */
1667 static void btrfs_merge_extent_hook(void *private_data,
1668 struct extent_state *new,
1669 struct extent_state *other)
1670 {
1671 struct inode *inode = private_data;
1672 u64 new_size, old_size;
1673 u32 num_extents;
1674
1675 /* not delalloc, ignore it */
1676 if (!(other->state & EXTENT_DELALLOC))
1677 return;
1678
1679 if (new->start > other->start)
1680 new_size = new->end - other->start + 1;
1681 else
1682 new_size = other->end - new->start + 1;
1683
1684 /* we're not bigger than the max, unreserve the space and go */
1685 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1686 spin_lock(&BTRFS_I(inode)->lock);
1687 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1688 spin_unlock(&BTRFS_I(inode)->lock);
1689 return;
1690 }
1691
1692 /*
1693 * We have to add up either side to figure out how many extents were
1694 * accounted for before we merged into one big extent. If the number of
1695 * extents we accounted for is <= the amount we need for the new range
1696 * then we can return, otherwise drop. Think of it like this
1697 *
1698 * [ 4k][MAX_SIZE]
1699 *
1700 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1701 * need 2 outstanding extents, on one side we have 1 and the other side
1702 * we have 1 so they are == and we can return. But in this case
1703 *
1704 * [MAX_SIZE+4k][MAX_SIZE+4k]
1705 *
1706 * Each range on their own accounts for 2 extents, but merged together
1707 * they are only 3 extents worth of accounting, so we need to drop in
1708 * this case.
1709 */
1710 old_size = other->end - other->start + 1;
1711 num_extents = count_max_extents(old_size);
1712 old_size = new->end - new->start + 1;
1713 num_extents += count_max_extents(old_size);
1714 if (count_max_extents(new_size) >= num_extents)
1715 return;
1716
1717 spin_lock(&BTRFS_I(inode)->lock);
1718 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1719 spin_unlock(&BTRFS_I(inode)->lock);
1720 }
1721
1722 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1723 struct inode *inode)
1724 {
1725 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1726
1727 spin_lock(&root->delalloc_lock);
1728 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1729 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1730 &root->delalloc_inodes);
1731 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1732 &BTRFS_I(inode)->runtime_flags);
1733 root->nr_delalloc_inodes++;
1734 if (root->nr_delalloc_inodes == 1) {
1735 spin_lock(&fs_info->delalloc_root_lock);
1736 BUG_ON(!list_empty(&root->delalloc_root));
1737 list_add_tail(&root->delalloc_root,
1738 &fs_info->delalloc_roots);
1739 spin_unlock(&fs_info->delalloc_root_lock);
1740 }
1741 }
1742 spin_unlock(&root->delalloc_lock);
1743 }
1744
1745 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1746 struct btrfs_inode *inode)
1747 {
1748 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1749
1750 spin_lock(&root->delalloc_lock);
1751 if (!list_empty(&inode->delalloc_inodes)) {
1752 list_del_init(&inode->delalloc_inodes);
1753 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1754 &inode->runtime_flags);
1755 root->nr_delalloc_inodes--;
1756 if (!root->nr_delalloc_inodes) {
1757 spin_lock(&fs_info->delalloc_root_lock);
1758 BUG_ON(list_empty(&root->delalloc_root));
1759 list_del_init(&root->delalloc_root);
1760 spin_unlock(&fs_info->delalloc_root_lock);
1761 }
1762 }
1763 spin_unlock(&root->delalloc_lock);
1764 }
1765
1766 /*
1767 * extent_io.c set_bit_hook, used to track delayed allocation
1768 * bytes in this file, and to maintain the list of inodes that
1769 * have pending delalloc work to be done.
1770 */
1771 static void btrfs_set_bit_hook(void *private_data,
1772 struct extent_state *state, unsigned *bits)
1773 {
1774 struct inode *inode = private_data;
1775
1776 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1777
1778 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1779 WARN_ON(1);
1780 /*
1781 * set_bit and clear bit hooks normally require _irqsave/restore
1782 * but in this case, we are only testing for the DELALLOC
1783 * bit, which is only set or cleared with irqs on
1784 */
1785 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1786 struct btrfs_root *root = BTRFS_I(inode)->root;
1787 u64 len = state->end + 1 - state->start;
1788 u32 num_extents = count_max_extents(len);
1789 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1790
1791 spin_lock(&BTRFS_I(inode)->lock);
1792 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1793 spin_unlock(&BTRFS_I(inode)->lock);
1794
1795 /* For sanity tests */
1796 if (btrfs_is_testing(fs_info))
1797 return;
1798
1799 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1800 fs_info->delalloc_batch);
1801 spin_lock(&BTRFS_I(inode)->lock);
1802 BTRFS_I(inode)->delalloc_bytes += len;
1803 if (*bits & EXTENT_DEFRAG)
1804 BTRFS_I(inode)->defrag_bytes += len;
1805 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1806 &BTRFS_I(inode)->runtime_flags))
1807 btrfs_add_delalloc_inodes(root, inode);
1808 spin_unlock(&BTRFS_I(inode)->lock);
1809 }
1810
1811 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1812 (*bits & EXTENT_DELALLOC_NEW)) {
1813 spin_lock(&BTRFS_I(inode)->lock);
1814 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1815 state->start;
1816 spin_unlock(&BTRFS_I(inode)->lock);
1817 }
1818 }
1819
1820 /*
1821 * extent_io.c clear_bit_hook, see set_bit_hook for why
1822 */
1823 static void btrfs_clear_bit_hook(void *private_data,
1824 struct extent_state *state,
1825 unsigned *bits)
1826 {
1827 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1828 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1829 u64 len = state->end + 1 - state->start;
1830 u32 num_extents = count_max_extents(len);
1831
1832 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1833 spin_lock(&inode->lock);
1834 inode->defrag_bytes -= len;
1835 spin_unlock(&inode->lock);
1836 }
1837
1838 /*
1839 * set_bit and clear bit hooks normally require _irqsave/restore
1840 * but in this case, we are only testing for the DELALLOC
1841 * bit, which is only set or cleared with irqs on
1842 */
1843 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1844 struct btrfs_root *root = inode->root;
1845 bool do_list = !btrfs_is_free_space_inode(inode);
1846
1847 spin_lock(&inode->lock);
1848 btrfs_mod_outstanding_extents(inode, -num_extents);
1849 spin_unlock(&inode->lock);
1850
1851 /*
1852 * We don't reserve metadata space for space cache inodes so we
1853 * don't need to call dellalloc_release_metadata if there is an
1854 * error.
1855 */
1856 if (*bits & EXTENT_CLEAR_META_RESV &&
1857 root != fs_info->tree_root)
1858 btrfs_delalloc_release_metadata(inode, len, false);
1859
1860 /* For sanity tests. */
1861 if (btrfs_is_testing(fs_info))
1862 return;
1863
1864 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1865 do_list && !(state->state & EXTENT_NORESERVE) &&
1866 (*bits & EXTENT_CLEAR_DATA_RESV))
1867 btrfs_free_reserved_data_space_noquota(
1868 &inode->vfs_inode,
1869 state->start, len);
1870
1871 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1872 fs_info->delalloc_batch);
1873 spin_lock(&inode->lock);
1874 inode->delalloc_bytes -= len;
1875 if (do_list && inode->delalloc_bytes == 0 &&
1876 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1877 &inode->runtime_flags))
1878 btrfs_del_delalloc_inode(root, inode);
1879 spin_unlock(&inode->lock);
1880 }
1881
1882 if ((state->state & EXTENT_DELALLOC_NEW) &&
1883 (*bits & EXTENT_DELALLOC_NEW)) {
1884 spin_lock(&inode->lock);
1885 ASSERT(inode->new_delalloc_bytes >= len);
1886 inode->new_delalloc_bytes -= len;
1887 spin_unlock(&inode->lock);
1888 }
1889 }
1890
1891 /*
1892 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1893 * we don't create bios that span stripes or chunks
1894 *
1895 * return 1 if page cannot be merged to bio
1896 * return 0 if page can be merged to bio
1897 * return error otherwise
1898 */
1899 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1900 size_t size, struct bio *bio,
1901 unsigned long bio_flags)
1902 {
1903 struct inode *inode = page->mapping->host;
1904 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1905 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1906 u64 length = 0;
1907 u64 map_length;
1908 int ret;
1909
1910 if (bio_flags & EXTENT_BIO_COMPRESSED)
1911 return 0;
1912
1913 length = bio->bi_iter.bi_size;
1914 map_length = length;
1915 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1916 NULL, 0);
1917 if (ret < 0)
1918 return ret;
1919 if (map_length < length + size)
1920 return 1;
1921 return 0;
1922 }
1923
1924 /*
1925 * in order to insert checksums into the metadata in large chunks,
1926 * we wait until bio submission time. All the pages in the bio are
1927 * checksummed and sums are attached onto the ordered extent record.
1928 *
1929 * At IO completion time the cums attached on the ordered extent record
1930 * are inserted into the btree
1931 */
1932 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1933 u64 bio_offset)
1934 {
1935 struct inode *inode = private_data;
1936 blk_status_t ret = 0;
1937
1938 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1939 BUG_ON(ret); /* -ENOMEM */
1940 return 0;
1941 }
1942
1943 /*
1944 * in order to insert checksums into the metadata in large chunks,
1945 * we wait until bio submission time. All the pages in the bio are
1946 * checksummed and sums are attached onto the ordered extent record.
1947 *
1948 * At IO completion time the cums attached on the ordered extent record
1949 * are inserted into the btree
1950 */
1951 static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1952 int mirror_num)
1953 {
1954 struct inode *inode = private_data;
1955 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1956 blk_status_t ret;
1957
1958 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1959 if (ret) {
1960 bio->bi_status = ret;
1961 bio_endio(bio);
1962 }
1963 return ret;
1964 }
1965
1966 /*
1967 * extent_io.c submission hook. This does the right thing for csum calculation
1968 * on write, or reading the csums from the tree before a read.
1969 *
1970 * Rules about async/sync submit,
1971 * a) read: sync submit
1972 *
1973 * b) write without checksum: sync submit
1974 *
1975 * c) write with checksum:
1976 * c-1) if bio is issued by fsync: sync submit
1977 * (sync_writers != 0)
1978 *
1979 * c-2) if root is reloc root: sync submit
1980 * (only in case of buffered IO)
1981 *
1982 * c-3) otherwise: async submit
1983 */
1984 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1985 int mirror_num, unsigned long bio_flags,
1986 u64 bio_offset)
1987 {
1988 struct inode *inode = private_data;
1989 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1990 struct btrfs_root *root = BTRFS_I(inode)->root;
1991 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1992 blk_status_t ret = 0;
1993 int skip_sum;
1994 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1995
1996 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1997
1998 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1999 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2000
2001 if (bio_op(bio) != REQ_OP_WRITE) {
2002 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2003 if (ret)
2004 goto out;
2005
2006 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2007 ret = btrfs_submit_compressed_read(inode, bio,
2008 mirror_num,
2009 bio_flags);
2010 goto out;
2011 } else if (!skip_sum) {
2012 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2013 if (ret)
2014 goto out;
2015 }
2016 goto mapit;
2017 } else if (async && !skip_sum) {
2018 /* csum items have already been cloned */
2019 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2020 goto mapit;
2021 /* we're doing a write, do the async checksumming */
2022 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2023 bio_offset, inode,
2024 btrfs_submit_bio_start,
2025 btrfs_submit_bio_done);
2026 goto out;
2027 } else if (!skip_sum) {
2028 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2029 if (ret)
2030 goto out;
2031 }
2032
2033 mapit:
2034 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2035
2036 out:
2037 if (ret) {
2038 bio->bi_status = ret;
2039 bio_endio(bio);
2040 }
2041 return ret;
2042 }
2043
2044 /*
2045 * given a list of ordered sums record them in the inode. This happens
2046 * at IO completion time based on sums calculated at bio submission time.
2047 */
2048 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2049 struct inode *inode, struct list_head *list)
2050 {
2051 struct btrfs_ordered_sum *sum;
2052 int ret;
2053
2054 list_for_each_entry(sum, list, list) {
2055 trans->adding_csums = true;
2056 ret = btrfs_csum_file_blocks(trans,
2057 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2058 trans->adding_csums = false;
2059 if (ret)
2060 return ret;
2061 }
2062 return 0;
2063 }
2064
2065 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2066 unsigned int extra_bits,
2067 struct extent_state **cached_state, int dedupe)
2068 {
2069 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2070 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2071 extra_bits, cached_state);
2072 }
2073
2074 /* see btrfs_writepage_start_hook for details on why this is required */
2075 struct btrfs_writepage_fixup {
2076 struct page *page;
2077 struct btrfs_work work;
2078 };
2079
2080 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2081 {
2082 struct btrfs_writepage_fixup *fixup;
2083 struct btrfs_ordered_extent *ordered;
2084 struct extent_state *cached_state = NULL;
2085 struct extent_changeset *data_reserved = NULL;
2086 struct page *page;
2087 struct inode *inode;
2088 u64 page_start;
2089 u64 page_end;
2090 int ret;
2091
2092 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2093 page = fixup->page;
2094 again:
2095 lock_page(page);
2096 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2097 ClearPageChecked(page);
2098 goto out_page;
2099 }
2100
2101 inode = page->mapping->host;
2102 page_start = page_offset(page);
2103 page_end = page_offset(page) + PAGE_SIZE - 1;
2104
2105 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2106 &cached_state);
2107
2108 /* already ordered? We're done */
2109 if (PagePrivate2(page))
2110 goto out;
2111
2112 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2113 PAGE_SIZE);
2114 if (ordered) {
2115 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2116 page_end, &cached_state);
2117 unlock_page(page);
2118 btrfs_start_ordered_extent(inode, ordered, 1);
2119 btrfs_put_ordered_extent(ordered);
2120 goto again;
2121 }
2122
2123 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2124 PAGE_SIZE);
2125 if (ret) {
2126 mapping_set_error(page->mapping, ret);
2127 end_extent_writepage(page, ret, page_start, page_end);
2128 ClearPageChecked(page);
2129 goto out;
2130 }
2131
2132 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2133 &cached_state, 0);
2134 if (ret) {
2135 mapping_set_error(page->mapping, ret);
2136 end_extent_writepage(page, ret, page_start, page_end);
2137 ClearPageChecked(page);
2138 goto out;
2139 }
2140
2141 ClearPageChecked(page);
2142 set_page_dirty(page);
2143 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2144 out:
2145 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2146 &cached_state);
2147 out_page:
2148 unlock_page(page);
2149 put_page(page);
2150 kfree(fixup);
2151 extent_changeset_free(data_reserved);
2152 }
2153
2154 /*
2155 * There are a few paths in the higher layers of the kernel that directly
2156 * set the page dirty bit without asking the filesystem if it is a
2157 * good idea. This causes problems because we want to make sure COW
2158 * properly happens and the data=ordered rules are followed.
2159 *
2160 * In our case any range that doesn't have the ORDERED bit set
2161 * hasn't been properly setup for IO. We kick off an async process
2162 * to fix it up. The async helper will wait for ordered extents, set
2163 * the delalloc bit and make it safe to write the page.
2164 */
2165 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2166 {
2167 struct inode *inode = page->mapping->host;
2168 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2169 struct btrfs_writepage_fixup *fixup;
2170
2171 /* this page is properly in the ordered list */
2172 if (TestClearPagePrivate2(page))
2173 return 0;
2174
2175 if (PageChecked(page))
2176 return -EAGAIN;
2177
2178 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2179 if (!fixup)
2180 return -EAGAIN;
2181
2182 SetPageChecked(page);
2183 get_page(page);
2184 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2185 btrfs_writepage_fixup_worker, NULL, NULL);
2186 fixup->page = page;
2187 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2188 return -EBUSY;
2189 }
2190
2191 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2192 struct inode *inode, u64 file_pos,
2193 u64 disk_bytenr, u64 disk_num_bytes,
2194 u64 num_bytes, u64 ram_bytes,
2195 u8 compression, u8 encryption,
2196 u16 other_encoding, int extent_type)
2197 {
2198 struct btrfs_root *root = BTRFS_I(inode)->root;
2199 struct btrfs_file_extent_item *fi;
2200 struct btrfs_path *path;
2201 struct extent_buffer *leaf;
2202 struct btrfs_key ins;
2203 u64 qg_released;
2204 int extent_inserted = 0;
2205 int ret;
2206
2207 path = btrfs_alloc_path();
2208 if (!path)
2209 return -ENOMEM;
2210
2211 /*
2212 * we may be replacing one extent in the tree with another.
2213 * The new extent is pinned in the extent map, and we don't want
2214 * to drop it from the cache until it is completely in the btree.
2215 *
2216 * So, tell btrfs_drop_extents to leave this extent in the cache.
2217 * the caller is expected to unpin it and allow it to be merged
2218 * with the others.
2219 */
2220 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2221 file_pos + num_bytes, NULL, 0,
2222 1, sizeof(*fi), &extent_inserted);
2223 if (ret)
2224 goto out;
2225
2226 if (!extent_inserted) {
2227 ins.objectid = btrfs_ino(BTRFS_I(inode));
2228 ins.offset = file_pos;
2229 ins.type = BTRFS_EXTENT_DATA_KEY;
2230
2231 path->leave_spinning = 1;
2232 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2233 sizeof(*fi));
2234 if (ret)
2235 goto out;
2236 }
2237 leaf = path->nodes[0];
2238 fi = btrfs_item_ptr(leaf, path->slots[0],
2239 struct btrfs_file_extent_item);
2240 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2241 btrfs_set_file_extent_type(leaf, fi, extent_type);
2242 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2243 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2244 btrfs_set_file_extent_offset(leaf, fi, 0);
2245 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2246 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2247 btrfs_set_file_extent_compression(leaf, fi, compression);
2248 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2249 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2250
2251 btrfs_mark_buffer_dirty(leaf);
2252 btrfs_release_path(path);
2253
2254 inode_add_bytes(inode, num_bytes);
2255
2256 ins.objectid = disk_bytenr;
2257 ins.offset = disk_num_bytes;
2258 ins.type = BTRFS_EXTENT_ITEM_KEY;
2259
2260 /*
2261 * Release the reserved range from inode dirty range map, as it is
2262 * already moved into delayed_ref_head
2263 */
2264 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2265 if (ret < 0)
2266 goto out;
2267 qg_released = ret;
2268 ret = btrfs_alloc_reserved_file_extent(trans, root,
2269 btrfs_ino(BTRFS_I(inode)),
2270 file_pos, qg_released, &ins);
2271 out:
2272 btrfs_free_path(path);
2273
2274 return ret;
2275 }
2276
2277 /* snapshot-aware defrag */
2278 struct sa_defrag_extent_backref {
2279 struct rb_node node;
2280 struct old_sa_defrag_extent *old;
2281 u64 root_id;
2282 u64 inum;
2283 u64 file_pos;
2284 u64 extent_offset;
2285 u64 num_bytes;
2286 u64 generation;
2287 };
2288
2289 struct old_sa_defrag_extent {
2290 struct list_head list;
2291 struct new_sa_defrag_extent *new;
2292
2293 u64 extent_offset;
2294 u64 bytenr;
2295 u64 offset;
2296 u64 len;
2297 int count;
2298 };
2299
2300 struct new_sa_defrag_extent {
2301 struct rb_root root;
2302 struct list_head head;
2303 struct btrfs_path *path;
2304 struct inode *inode;
2305 u64 file_pos;
2306 u64 len;
2307 u64 bytenr;
2308 u64 disk_len;
2309 u8 compress_type;
2310 };
2311
2312 static int backref_comp(struct sa_defrag_extent_backref *b1,
2313 struct sa_defrag_extent_backref *b2)
2314 {
2315 if (b1->root_id < b2->root_id)
2316 return -1;
2317 else if (b1->root_id > b2->root_id)
2318 return 1;
2319
2320 if (b1->inum < b2->inum)
2321 return -1;
2322 else if (b1->inum > b2->inum)
2323 return 1;
2324
2325 if (b1->file_pos < b2->file_pos)
2326 return -1;
2327 else if (b1->file_pos > b2->file_pos)
2328 return 1;
2329
2330 /*
2331 * [------------------------------] ===> (a range of space)
2332 * |<--->| |<---->| =============> (fs/file tree A)
2333 * |<---------------------------->| ===> (fs/file tree B)
2334 *
2335 * A range of space can refer to two file extents in one tree while
2336 * refer to only one file extent in another tree.
2337 *
2338 * So we may process a disk offset more than one time(two extents in A)
2339 * and locate at the same extent(one extent in B), then insert two same
2340 * backrefs(both refer to the extent in B).
2341 */
2342 return 0;
2343 }
2344
2345 static void backref_insert(struct rb_root *root,
2346 struct sa_defrag_extent_backref *backref)
2347 {
2348 struct rb_node **p = &root->rb_node;
2349 struct rb_node *parent = NULL;
2350 struct sa_defrag_extent_backref *entry;
2351 int ret;
2352
2353 while (*p) {
2354 parent = *p;
2355 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2356
2357 ret = backref_comp(backref, entry);
2358 if (ret < 0)
2359 p = &(*p)->rb_left;
2360 else
2361 p = &(*p)->rb_right;
2362 }
2363
2364 rb_link_node(&backref->node, parent, p);
2365 rb_insert_color(&backref->node, root);
2366 }
2367
2368 /*
2369 * Note the backref might has changed, and in this case we just return 0.
2370 */
2371 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2372 void *ctx)
2373 {
2374 struct btrfs_file_extent_item *extent;
2375 struct old_sa_defrag_extent *old = ctx;
2376 struct new_sa_defrag_extent *new = old->new;
2377 struct btrfs_path *path = new->path;
2378 struct btrfs_key key;
2379 struct btrfs_root *root;
2380 struct sa_defrag_extent_backref *backref;
2381 struct extent_buffer *leaf;
2382 struct inode *inode = new->inode;
2383 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2384 int slot;
2385 int ret;
2386 u64 extent_offset;
2387 u64 num_bytes;
2388
2389 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2390 inum == btrfs_ino(BTRFS_I(inode)))
2391 return 0;
2392
2393 key.objectid = root_id;
2394 key.type = BTRFS_ROOT_ITEM_KEY;
2395 key.offset = (u64)-1;
2396
2397 root = btrfs_read_fs_root_no_name(fs_info, &key);
2398 if (IS_ERR(root)) {
2399 if (PTR_ERR(root) == -ENOENT)
2400 return 0;
2401 WARN_ON(1);
2402 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2403 inum, offset, root_id);
2404 return PTR_ERR(root);
2405 }
2406
2407 key.objectid = inum;
2408 key.type = BTRFS_EXTENT_DATA_KEY;
2409 if (offset > (u64)-1 << 32)
2410 key.offset = 0;
2411 else
2412 key.offset = offset;
2413
2414 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2415 if (WARN_ON(ret < 0))
2416 return ret;
2417 ret = 0;
2418
2419 while (1) {
2420 cond_resched();
2421
2422 leaf = path->nodes[0];
2423 slot = path->slots[0];
2424
2425 if (slot >= btrfs_header_nritems(leaf)) {
2426 ret = btrfs_next_leaf(root, path);
2427 if (ret < 0) {
2428 goto out;
2429 } else if (ret > 0) {
2430 ret = 0;
2431 goto out;
2432 }
2433 continue;
2434 }
2435
2436 path->slots[0]++;
2437
2438 btrfs_item_key_to_cpu(leaf, &key, slot);
2439
2440 if (key.objectid > inum)
2441 goto out;
2442
2443 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2444 continue;
2445
2446 extent = btrfs_item_ptr(leaf, slot,
2447 struct btrfs_file_extent_item);
2448
2449 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2450 continue;
2451
2452 /*
2453 * 'offset' refers to the exact key.offset,
2454 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2455 * (key.offset - extent_offset).
2456 */
2457 if (key.offset != offset)
2458 continue;
2459
2460 extent_offset = btrfs_file_extent_offset(leaf, extent);
2461 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2462
2463 if (extent_offset >= old->extent_offset + old->offset +
2464 old->len || extent_offset + num_bytes <=
2465 old->extent_offset + old->offset)
2466 continue;
2467 break;
2468 }
2469
2470 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2471 if (!backref) {
2472 ret = -ENOENT;
2473 goto out;
2474 }
2475
2476 backref->root_id = root_id;
2477 backref->inum = inum;
2478 backref->file_pos = offset;
2479 backref->num_bytes = num_bytes;
2480 backref->extent_offset = extent_offset;
2481 backref->generation = btrfs_file_extent_generation(leaf, extent);
2482 backref->old = old;
2483 backref_insert(&new->root, backref);
2484 old->count++;
2485 out:
2486 btrfs_release_path(path);
2487 WARN_ON(ret);
2488 return ret;
2489 }
2490
2491 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2492 struct new_sa_defrag_extent *new)
2493 {
2494 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2495 struct old_sa_defrag_extent *old, *tmp;
2496 int ret;
2497
2498 new->path = path;
2499
2500 list_for_each_entry_safe(old, tmp, &new->head, list) {
2501 ret = iterate_inodes_from_logical(old->bytenr +
2502 old->extent_offset, fs_info,
2503 path, record_one_backref,
2504 old, false);
2505 if (ret < 0 && ret != -ENOENT)
2506 return false;
2507
2508 /* no backref to be processed for this extent */
2509 if (!old->count) {
2510 list_del(&old->list);
2511 kfree(old);
2512 }
2513 }
2514
2515 if (list_empty(&new->head))
2516 return false;
2517
2518 return true;
2519 }
2520
2521 static int relink_is_mergable(struct extent_buffer *leaf,
2522 struct btrfs_file_extent_item *fi,
2523 struct new_sa_defrag_extent *new)
2524 {
2525 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2526 return 0;
2527
2528 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2529 return 0;
2530
2531 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2532 return 0;
2533
2534 if (btrfs_file_extent_encryption(leaf, fi) ||
2535 btrfs_file_extent_other_encoding(leaf, fi))
2536 return 0;
2537
2538 return 1;
2539 }
2540
2541 /*
2542 * Note the backref might has changed, and in this case we just return 0.
2543 */
2544 static noinline int relink_extent_backref(struct btrfs_path *path,
2545 struct sa_defrag_extent_backref *prev,
2546 struct sa_defrag_extent_backref *backref)
2547 {
2548 struct btrfs_file_extent_item *extent;
2549 struct btrfs_file_extent_item *item;
2550 struct btrfs_ordered_extent *ordered;
2551 struct btrfs_trans_handle *trans;
2552 struct btrfs_root *root;
2553 struct btrfs_key key;
2554 struct extent_buffer *leaf;
2555 struct old_sa_defrag_extent *old = backref->old;
2556 struct new_sa_defrag_extent *new = old->new;
2557 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2558 struct inode *inode;
2559 struct extent_state *cached = NULL;
2560 int ret = 0;
2561 u64 start;
2562 u64 len;
2563 u64 lock_start;
2564 u64 lock_end;
2565 bool merge = false;
2566 int index;
2567
2568 if (prev && prev->root_id == backref->root_id &&
2569 prev->inum == backref->inum &&
2570 prev->file_pos + prev->num_bytes == backref->file_pos)
2571 merge = true;
2572
2573 /* step 1: get root */
2574 key.objectid = backref->root_id;
2575 key.type = BTRFS_ROOT_ITEM_KEY;
2576 key.offset = (u64)-1;
2577
2578 index = srcu_read_lock(&fs_info->subvol_srcu);
2579
2580 root = btrfs_read_fs_root_no_name(fs_info, &key);
2581 if (IS_ERR(root)) {
2582 srcu_read_unlock(&fs_info->subvol_srcu, index);
2583 if (PTR_ERR(root) == -ENOENT)
2584 return 0;
2585 return PTR_ERR(root);
2586 }
2587
2588 if (btrfs_root_readonly(root)) {
2589 srcu_read_unlock(&fs_info->subvol_srcu, index);
2590 return 0;
2591 }
2592
2593 /* step 2: get inode */
2594 key.objectid = backref->inum;
2595 key.type = BTRFS_INODE_ITEM_KEY;
2596 key.offset = 0;
2597
2598 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2599 if (IS_ERR(inode)) {
2600 srcu_read_unlock(&fs_info->subvol_srcu, index);
2601 return 0;
2602 }
2603
2604 srcu_read_unlock(&fs_info->subvol_srcu, index);
2605
2606 /* step 3: relink backref */
2607 lock_start = backref->file_pos;
2608 lock_end = backref->file_pos + backref->num_bytes - 1;
2609 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2610 &cached);
2611
2612 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2613 if (ordered) {
2614 btrfs_put_ordered_extent(ordered);
2615 goto out_unlock;
2616 }
2617
2618 trans = btrfs_join_transaction(root);
2619 if (IS_ERR(trans)) {
2620 ret = PTR_ERR(trans);
2621 goto out_unlock;
2622 }
2623
2624 key.objectid = backref->inum;
2625 key.type = BTRFS_EXTENT_DATA_KEY;
2626 key.offset = backref->file_pos;
2627
2628 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2629 if (ret < 0) {
2630 goto out_free_path;
2631 } else if (ret > 0) {
2632 ret = 0;
2633 goto out_free_path;
2634 }
2635
2636 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2637 struct btrfs_file_extent_item);
2638
2639 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2640 backref->generation)
2641 goto out_free_path;
2642
2643 btrfs_release_path(path);
2644
2645 start = backref->file_pos;
2646 if (backref->extent_offset < old->extent_offset + old->offset)
2647 start += old->extent_offset + old->offset -
2648 backref->extent_offset;
2649
2650 len = min(backref->extent_offset + backref->num_bytes,
2651 old->extent_offset + old->offset + old->len);
2652 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2653
2654 ret = btrfs_drop_extents(trans, root, inode, start,
2655 start + len, 1);
2656 if (ret)
2657 goto out_free_path;
2658 again:
2659 key.objectid = btrfs_ino(BTRFS_I(inode));
2660 key.type = BTRFS_EXTENT_DATA_KEY;
2661 key.offset = start;
2662
2663 path->leave_spinning = 1;
2664 if (merge) {
2665 struct btrfs_file_extent_item *fi;
2666 u64 extent_len;
2667 struct btrfs_key found_key;
2668
2669 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2670 if (ret < 0)
2671 goto out_free_path;
2672
2673 path->slots[0]--;
2674 leaf = path->nodes[0];
2675 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2676
2677 fi = btrfs_item_ptr(leaf, path->slots[0],
2678 struct btrfs_file_extent_item);
2679 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2680
2681 if (extent_len + found_key.offset == start &&
2682 relink_is_mergable(leaf, fi, new)) {
2683 btrfs_set_file_extent_num_bytes(leaf, fi,
2684 extent_len + len);
2685 btrfs_mark_buffer_dirty(leaf);
2686 inode_add_bytes(inode, len);
2687
2688 ret = 1;
2689 goto out_free_path;
2690 } else {
2691 merge = false;
2692 btrfs_release_path(path);
2693 goto again;
2694 }
2695 }
2696
2697 ret = btrfs_insert_empty_item(trans, root, path, &key,
2698 sizeof(*extent));
2699 if (ret) {
2700 btrfs_abort_transaction(trans, ret);
2701 goto out_free_path;
2702 }
2703
2704 leaf = path->nodes[0];
2705 item = btrfs_item_ptr(leaf, path->slots[0],
2706 struct btrfs_file_extent_item);
2707 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2708 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2709 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2710 btrfs_set_file_extent_num_bytes(leaf, item, len);
2711 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2712 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2713 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2714 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2715 btrfs_set_file_extent_encryption(leaf, item, 0);
2716 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2717
2718 btrfs_mark_buffer_dirty(leaf);
2719 inode_add_bytes(inode, len);
2720 btrfs_release_path(path);
2721
2722 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2723 new->disk_len, 0,
2724 backref->root_id, backref->inum,
2725 new->file_pos); /* start - extent_offset */
2726 if (ret) {
2727 btrfs_abort_transaction(trans, ret);
2728 goto out_free_path;
2729 }
2730
2731 ret = 1;
2732 out_free_path:
2733 btrfs_release_path(path);
2734 path->leave_spinning = 0;
2735 btrfs_end_transaction(trans);
2736 out_unlock:
2737 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2738 &cached);
2739 iput(inode);
2740 return ret;
2741 }
2742
2743 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2744 {
2745 struct old_sa_defrag_extent *old, *tmp;
2746
2747 if (!new)
2748 return;
2749
2750 list_for_each_entry_safe(old, tmp, &new->head, list) {
2751 kfree(old);
2752 }
2753 kfree(new);
2754 }
2755
2756 static void relink_file_extents(struct new_sa_defrag_extent *new)
2757 {
2758 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2759 struct btrfs_path *path;
2760 struct sa_defrag_extent_backref *backref;
2761 struct sa_defrag_extent_backref *prev = NULL;
2762 struct inode *inode;
2763 struct rb_node *node;
2764 int ret;
2765
2766 inode = new->inode;
2767
2768 path = btrfs_alloc_path();
2769 if (!path)
2770 return;
2771
2772 if (!record_extent_backrefs(path, new)) {
2773 btrfs_free_path(path);
2774 goto out;
2775 }
2776 btrfs_release_path(path);
2777
2778 while (1) {
2779 node = rb_first(&new->root);
2780 if (!node)
2781 break;
2782 rb_erase(node, &new->root);
2783
2784 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2785
2786 ret = relink_extent_backref(path, prev, backref);
2787 WARN_ON(ret < 0);
2788
2789 kfree(prev);
2790
2791 if (ret == 1)
2792 prev = backref;
2793 else
2794 prev = NULL;
2795 cond_resched();
2796 }
2797 kfree(prev);
2798
2799 btrfs_free_path(path);
2800 out:
2801 free_sa_defrag_extent(new);
2802
2803 atomic_dec(&fs_info->defrag_running);
2804 wake_up(&fs_info->transaction_wait);
2805 }
2806
2807 static struct new_sa_defrag_extent *
2808 record_old_file_extents(struct inode *inode,
2809 struct btrfs_ordered_extent *ordered)
2810 {
2811 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2812 struct btrfs_root *root = BTRFS_I(inode)->root;
2813 struct btrfs_path *path;
2814 struct btrfs_key key;
2815 struct old_sa_defrag_extent *old;
2816 struct new_sa_defrag_extent *new;
2817 int ret;
2818
2819 new = kmalloc(sizeof(*new), GFP_NOFS);
2820 if (!new)
2821 return NULL;
2822
2823 new->inode = inode;
2824 new->file_pos = ordered->file_offset;
2825 new->len = ordered->len;
2826 new->bytenr = ordered->start;
2827 new->disk_len = ordered->disk_len;
2828 new->compress_type = ordered->compress_type;
2829 new->root = RB_ROOT;
2830 INIT_LIST_HEAD(&new->head);
2831
2832 path = btrfs_alloc_path();
2833 if (!path)
2834 goto out_kfree;
2835
2836 key.objectid = btrfs_ino(BTRFS_I(inode));
2837 key.type = BTRFS_EXTENT_DATA_KEY;
2838 key.offset = new->file_pos;
2839
2840 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2841 if (ret < 0)
2842 goto out_free_path;
2843 if (ret > 0 && path->slots[0] > 0)
2844 path->slots[0]--;
2845
2846 /* find out all the old extents for the file range */
2847 while (1) {
2848 struct btrfs_file_extent_item *extent;
2849 struct extent_buffer *l;
2850 int slot;
2851 u64 num_bytes;
2852 u64 offset;
2853 u64 end;
2854 u64 disk_bytenr;
2855 u64 extent_offset;
2856
2857 l = path->nodes[0];
2858 slot = path->slots[0];
2859
2860 if (slot >= btrfs_header_nritems(l)) {
2861 ret = btrfs_next_leaf(root, path);
2862 if (ret < 0)
2863 goto out_free_path;
2864 else if (ret > 0)
2865 break;
2866 continue;
2867 }
2868
2869 btrfs_item_key_to_cpu(l, &key, slot);
2870
2871 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2872 break;
2873 if (key.type != BTRFS_EXTENT_DATA_KEY)
2874 break;
2875 if (key.offset >= new->file_pos + new->len)
2876 break;
2877
2878 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2879
2880 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2881 if (key.offset + num_bytes < new->file_pos)
2882 goto next;
2883
2884 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2885 if (!disk_bytenr)
2886 goto next;
2887
2888 extent_offset = btrfs_file_extent_offset(l, extent);
2889
2890 old = kmalloc(sizeof(*old), GFP_NOFS);
2891 if (!old)
2892 goto out_free_path;
2893
2894 offset = max(new->file_pos, key.offset);
2895 end = min(new->file_pos + new->len, key.offset + num_bytes);
2896
2897 old->bytenr = disk_bytenr;
2898 old->extent_offset = extent_offset;
2899 old->offset = offset - key.offset;
2900 old->len = end - offset;
2901 old->new = new;
2902 old->count = 0;
2903 list_add_tail(&old->list, &new->head);
2904 next:
2905 path->slots[0]++;
2906 cond_resched();
2907 }
2908
2909 btrfs_free_path(path);
2910 atomic_inc(&fs_info->defrag_running);
2911
2912 return new;
2913
2914 out_free_path:
2915 btrfs_free_path(path);
2916 out_kfree:
2917 free_sa_defrag_extent(new);
2918 return NULL;
2919 }
2920
2921 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2922 u64 start, u64 len)
2923 {
2924 struct btrfs_block_group_cache *cache;
2925
2926 cache = btrfs_lookup_block_group(fs_info, start);
2927 ASSERT(cache);
2928
2929 spin_lock(&cache->lock);
2930 cache->delalloc_bytes -= len;
2931 spin_unlock(&cache->lock);
2932
2933 btrfs_put_block_group(cache);
2934 }
2935
2936 /* as ordered data IO finishes, this gets called so we can finish
2937 * an ordered extent if the range of bytes in the file it covers are
2938 * fully written.
2939 */
2940 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2941 {
2942 struct inode *inode = ordered_extent->inode;
2943 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2944 struct btrfs_root *root = BTRFS_I(inode)->root;
2945 struct btrfs_trans_handle *trans = NULL;
2946 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2947 struct extent_state *cached_state = NULL;
2948 struct new_sa_defrag_extent *new = NULL;
2949 int compress_type = 0;
2950 int ret = 0;
2951 u64 logical_len = ordered_extent->len;
2952 bool nolock;
2953 bool truncated = false;
2954 bool range_locked = false;
2955 bool clear_new_delalloc_bytes = false;
2956
2957 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2958 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2959 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2960 clear_new_delalloc_bytes = true;
2961
2962 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2963
2964 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2965 ret = -EIO;
2966 goto out;
2967 }
2968
2969 btrfs_free_io_failure_record(BTRFS_I(inode),
2970 ordered_extent->file_offset,
2971 ordered_extent->file_offset +
2972 ordered_extent->len - 1);
2973
2974 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2975 truncated = true;
2976 logical_len = ordered_extent->truncated_len;
2977 /* Truncated the entire extent, don't bother adding */
2978 if (!logical_len)
2979 goto out;
2980 }
2981
2982 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2983 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2984
2985 /*
2986 * For mwrite(mmap + memset to write) case, we still reserve
2987 * space for NOCOW range.
2988 * As NOCOW won't cause a new delayed ref, just free the space
2989 */
2990 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2991 ordered_extent->len);
2992 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2993 if (nolock)
2994 trans = btrfs_join_transaction_nolock(root);
2995 else
2996 trans = btrfs_join_transaction(root);
2997 if (IS_ERR(trans)) {
2998 ret = PTR_ERR(trans);
2999 trans = NULL;
3000 goto out;
3001 }
3002 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3003 ret = btrfs_update_inode_fallback(trans, root, inode);
3004 if (ret) /* -ENOMEM or corruption */
3005 btrfs_abort_transaction(trans, ret);
3006 goto out;
3007 }
3008
3009 range_locked = true;
3010 lock_extent_bits(io_tree, ordered_extent->file_offset,
3011 ordered_extent->file_offset + ordered_extent->len - 1,
3012 &cached_state);
3013
3014 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3015 ordered_extent->file_offset + ordered_extent->len - 1,
3016 EXTENT_DEFRAG, 0, cached_state);
3017 if (ret) {
3018 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3019 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3020 /* the inode is shared */
3021 new = record_old_file_extents(inode, ordered_extent);
3022
3023 clear_extent_bit(io_tree, ordered_extent->file_offset,
3024 ordered_extent->file_offset + ordered_extent->len - 1,
3025 EXTENT_DEFRAG, 0, 0, &cached_state);
3026 }
3027
3028 if (nolock)
3029 trans = btrfs_join_transaction_nolock(root);
3030 else
3031 trans = btrfs_join_transaction(root);
3032 if (IS_ERR(trans)) {
3033 ret = PTR_ERR(trans);
3034 trans = NULL;
3035 goto out;
3036 }
3037
3038 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3039
3040 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3041 compress_type = ordered_extent->compress_type;
3042 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3043 BUG_ON(compress_type);
3044 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3045 ordered_extent->len);
3046 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3047 ordered_extent->file_offset,
3048 ordered_extent->file_offset +
3049 logical_len);
3050 } else {
3051 BUG_ON(root == fs_info->tree_root);
3052 ret = insert_reserved_file_extent(trans, inode,
3053 ordered_extent->file_offset,
3054 ordered_extent->start,
3055 ordered_extent->disk_len,
3056 logical_len, logical_len,
3057 compress_type, 0, 0,
3058 BTRFS_FILE_EXTENT_REG);
3059 if (!ret)
3060 btrfs_release_delalloc_bytes(fs_info,
3061 ordered_extent->start,
3062 ordered_extent->disk_len);
3063 }
3064 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3065 ordered_extent->file_offset, ordered_extent->len,
3066 trans->transid);
3067 if (ret < 0) {
3068 btrfs_abort_transaction(trans, ret);
3069 goto out;
3070 }
3071
3072 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3073 if (ret) {
3074 btrfs_abort_transaction(trans, ret);
3075 goto out;
3076 }
3077
3078 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3079 ret = btrfs_update_inode_fallback(trans, root, inode);
3080 if (ret) { /* -ENOMEM or corruption */
3081 btrfs_abort_transaction(trans, ret);
3082 goto out;
3083 }
3084 ret = 0;
3085 out:
3086 if (range_locked || clear_new_delalloc_bytes) {
3087 unsigned int clear_bits = 0;
3088
3089 if (range_locked)
3090 clear_bits |= EXTENT_LOCKED;
3091 if (clear_new_delalloc_bytes)
3092 clear_bits |= EXTENT_DELALLOC_NEW;
3093 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3094 ordered_extent->file_offset,
3095 ordered_extent->file_offset +
3096 ordered_extent->len - 1,
3097 clear_bits,
3098 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3099 0, &cached_state);
3100 }
3101
3102 if (trans)
3103 btrfs_end_transaction(trans);
3104
3105 if (ret || truncated) {
3106 u64 start, end;
3107
3108 if (truncated)
3109 start = ordered_extent->file_offset + logical_len;
3110 else
3111 start = ordered_extent->file_offset;
3112 end = ordered_extent->file_offset + ordered_extent->len - 1;
3113 clear_extent_uptodate(io_tree, start, end, NULL);
3114
3115 /* Drop the cache for the part of the extent we didn't write. */
3116 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3117
3118 /*
3119 * If the ordered extent had an IOERR or something else went
3120 * wrong we need to return the space for this ordered extent
3121 * back to the allocator. We only free the extent in the
3122 * truncated case if we didn't write out the extent at all.
3123 */
3124 if ((ret || !logical_len) &&
3125 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3126 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3127 btrfs_free_reserved_extent(fs_info,
3128 ordered_extent->start,
3129 ordered_extent->disk_len, 1);
3130 }
3131
3132
3133 /*
3134 * This needs to be done to make sure anybody waiting knows we are done
3135 * updating everything for this ordered extent.
3136 */
3137 btrfs_remove_ordered_extent(inode, ordered_extent);
3138
3139 /* for snapshot-aware defrag */
3140 if (new) {
3141 if (ret) {
3142 free_sa_defrag_extent(new);
3143 atomic_dec(&fs_info->defrag_running);
3144 } else {
3145 relink_file_extents(new);
3146 }
3147 }
3148
3149 /* once for us */
3150 btrfs_put_ordered_extent(ordered_extent);
3151 /* once for the tree */
3152 btrfs_put_ordered_extent(ordered_extent);
3153
3154 return ret;
3155 }
3156
3157 static void finish_ordered_fn(struct btrfs_work *work)
3158 {
3159 struct btrfs_ordered_extent *ordered_extent;
3160 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3161 btrfs_finish_ordered_io(ordered_extent);
3162 }
3163
3164 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3165 struct extent_state *state, int uptodate)
3166 {
3167 struct inode *inode = page->mapping->host;
3168 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3169 struct btrfs_ordered_extent *ordered_extent = NULL;
3170 struct btrfs_workqueue *wq;
3171 btrfs_work_func_t func;
3172
3173 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3174
3175 ClearPagePrivate2(page);
3176 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3177 end - start + 1, uptodate))
3178 return;
3179
3180 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3181 wq = fs_info->endio_freespace_worker;
3182 func = btrfs_freespace_write_helper;
3183 } else {
3184 wq = fs_info->endio_write_workers;
3185 func = btrfs_endio_write_helper;
3186 }
3187
3188 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3189 NULL);
3190 btrfs_queue_work(wq, &ordered_extent->work);
3191 }
3192
3193 static int __readpage_endio_check(struct inode *inode,
3194 struct btrfs_io_bio *io_bio,
3195 int icsum, struct page *page,
3196 int pgoff, u64 start, size_t len)
3197 {
3198 char *kaddr;
3199 u32 csum_expected;
3200 u32 csum = ~(u32)0;
3201
3202 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3203
3204 kaddr = kmap_atomic(page);
3205 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3206 btrfs_csum_final(csum, (u8 *)&csum);
3207 if (csum != csum_expected)
3208 goto zeroit;
3209
3210 kunmap_atomic(kaddr);
3211 return 0;
3212 zeroit:
3213 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3214 io_bio->mirror_num);
3215 memset(kaddr + pgoff, 1, len);
3216 flush_dcache_page(page);
3217 kunmap_atomic(kaddr);
3218 return -EIO;
3219 }
3220
3221 /*
3222 * when reads are done, we need to check csums to verify the data is correct
3223 * if there's a match, we allow the bio to finish. If not, the code in
3224 * extent_io.c will try to find good copies for us.
3225 */
3226 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3227 u64 phy_offset, struct page *page,
3228 u64 start, u64 end, int mirror)
3229 {
3230 size_t offset = start - page_offset(page);
3231 struct inode *inode = page->mapping->host;
3232 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3233 struct btrfs_root *root = BTRFS_I(inode)->root;
3234
3235 if (PageChecked(page)) {
3236 ClearPageChecked(page);
3237 return 0;
3238 }
3239
3240 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3241 return 0;
3242
3243 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3244 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3245 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3246 return 0;
3247 }
3248
3249 phy_offset >>= inode->i_sb->s_blocksize_bits;
3250 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3251 start, (size_t)(end - start + 1));
3252 }
3253
3254 /*
3255 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3256 *
3257 * @inode: The inode we want to perform iput on
3258 *
3259 * This function uses the generic vfs_inode::i_count to track whether we should
3260 * just decrement it (in case it's > 1) or if this is the last iput then link
3261 * the inode to the delayed iput machinery. Delayed iputs are processed at
3262 * transaction commit time/superblock commit/cleaner kthread.
3263 */
3264 void btrfs_add_delayed_iput(struct inode *inode)
3265 {
3266 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3267 struct btrfs_inode *binode = BTRFS_I(inode);
3268
3269 if (atomic_add_unless(&inode->i_count, -1, 1))
3270 return;
3271
3272 spin_lock(&fs_info->delayed_iput_lock);
3273 ASSERT(list_empty(&binode->delayed_iput));
3274 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3275 spin_unlock(&fs_info->delayed_iput_lock);
3276 }
3277
3278 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3279 {
3280
3281 spin_lock(&fs_info->delayed_iput_lock);
3282 while (!list_empty(&fs_info->delayed_iputs)) {
3283 struct btrfs_inode *inode;
3284
3285 inode = list_first_entry(&fs_info->delayed_iputs,
3286 struct btrfs_inode, delayed_iput);
3287 list_del_init(&inode->delayed_iput);
3288 spin_unlock(&fs_info->delayed_iput_lock);
3289 iput(&inode->vfs_inode);
3290 spin_lock(&fs_info->delayed_iput_lock);
3291 }
3292 spin_unlock(&fs_info->delayed_iput_lock);
3293 }
3294
3295 /*
3296 * This is called in transaction commit time. If there are no orphan
3297 * files in the subvolume, it removes orphan item and frees block_rsv
3298 * structure.
3299 */
3300 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3301 struct btrfs_root *root)
3302 {
3303 struct btrfs_fs_info *fs_info = root->fs_info;
3304 struct btrfs_block_rsv *block_rsv;
3305 int ret;
3306
3307 if (atomic_read(&root->orphan_inodes) ||
3308 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3309 return;
3310
3311 spin_lock(&root->orphan_lock);
3312 if (atomic_read(&root->orphan_inodes)) {
3313 spin_unlock(&root->orphan_lock);
3314 return;
3315 }
3316
3317 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3318 spin_unlock(&root->orphan_lock);
3319 return;
3320 }
3321
3322 block_rsv = root->orphan_block_rsv;
3323 root->orphan_block_rsv = NULL;
3324 spin_unlock(&root->orphan_lock);
3325
3326 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3327 btrfs_root_refs(&root->root_item) > 0) {
3328 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3329 root->root_key.objectid);
3330 if (ret)
3331 btrfs_abort_transaction(trans, ret);
3332 else
3333 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3334 &root->state);
3335 }
3336
3337 if (block_rsv) {
3338 WARN_ON(block_rsv->size > 0);
3339 btrfs_free_block_rsv(fs_info, block_rsv);
3340 }
3341 }
3342
3343 /*
3344 * This creates an orphan entry for the given inode in case something goes
3345 * wrong in the middle of an unlink/truncate.
3346 *
3347 * NOTE: caller of this function should reserve 5 units of metadata for
3348 * this function.
3349 */
3350 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3351 struct btrfs_inode *inode)
3352 {
3353 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3354 struct btrfs_root *root = inode->root;
3355 struct btrfs_block_rsv *block_rsv = NULL;
3356 int reserve = 0;
3357 bool insert = false;
3358 int ret;
3359
3360 if (!root->orphan_block_rsv) {
3361 block_rsv = btrfs_alloc_block_rsv(fs_info,
3362 BTRFS_BLOCK_RSV_TEMP);
3363 if (!block_rsv)
3364 return -ENOMEM;
3365 }
3366
3367 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3368 &inode->runtime_flags))
3369 insert = true;
3370
3371 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3372 &inode->runtime_flags))
3373 reserve = 1;
3374
3375 spin_lock(&root->orphan_lock);
3376 /* If someone has created ->orphan_block_rsv, be happy to use it. */
3377 if (!root->orphan_block_rsv) {
3378 root->orphan_block_rsv = block_rsv;
3379 } else if (block_rsv) {
3380 btrfs_free_block_rsv(fs_info, block_rsv);
3381 block_rsv = NULL;
3382 }
3383
3384 if (insert)
3385 atomic_inc(&root->orphan_inodes);
3386 spin_unlock(&root->orphan_lock);
3387
3388 /* grab metadata reservation from transaction handle */
3389 if (reserve) {
3390 ret = btrfs_orphan_reserve_metadata(trans, inode);
3391 ASSERT(!ret);
3392 if (ret) {
3393 /*
3394 * dec doesn't need spin_lock as ->orphan_block_rsv
3395 * would be released only if ->orphan_inodes is
3396 * zero.
3397 */
3398 atomic_dec(&root->orphan_inodes);
3399 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3400 &inode->runtime_flags);
3401 if (insert)
3402 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3403 &inode->runtime_flags);
3404 return ret;
3405 }
3406 }
3407
3408 /* insert an orphan item to track this unlinked/truncated file */
3409 if (insert) {
3410 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3411 if (ret) {
3412 if (reserve) {
3413 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3414 &inode->runtime_flags);
3415 btrfs_orphan_release_metadata(inode);
3416 }
3417 /*
3418 * btrfs_orphan_commit_root may race with us and set
3419 * ->orphan_block_rsv to zero, in order to avoid that,
3420 * decrease ->orphan_inodes after everything is done.
3421 */
3422 atomic_dec(&root->orphan_inodes);
3423 if (ret != -EEXIST) {
3424 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3425 &inode->runtime_flags);
3426 btrfs_abort_transaction(trans, ret);
3427 return ret;
3428 }
3429 }
3430 ret = 0;
3431 }
3432
3433 return 0;
3434 }
3435
3436 /*
3437 * We have done the truncate/delete so we can go ahead and remove the orphan
3438 * item for this particular inode.
3439 */
3440 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3441 struct btrfs_inode *inode)
3442 {
3443 struct btrfs_root *root = inode->root;
3444 int delete_item = 0;
3445 int ret = 0;
3446
3447 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3448 &inode->runtime_flags))
3449 delete_item = 1;
3450
3451 if (delete_item && trans)
3452 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3453
3454 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3455 &inode->runtime_flags))
3456 btrfs_orphan_release_metadata(inode);
3457
3458 /*
3459 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3460 * to zero, in order to avoid that, decrease ->orphan_inodes after
3461 * everything is done.
3462 */
3463 if (delete_item)
3464 atomic_dec(&root->orphan_inodes);
3465
3466 return ret;
3467 }
3468
3469 /*
3470 * this cleans up any orphans that may be left on the list from the last use
3471 * of this root.
3472 */
3473 int btrfs_orphan_cleanup(struct btrfs_root *root)
3474 {
3475 struct btrfs_fs_info *fs_info = root->fs_info;
3476 struct btrfs_path *path;
3477 struct extent_buffer *leaf;
3478 struct btrfs_key key, found_key;
3479 struct btrfs_trans_handle *trans;
3480 struct inode *inode;
3481 u64 last_objectid = 0;
3482 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3483
3484 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3485 return 0;
3486
3487 path = btrfs_alloc_path();
3488 if (!path) {
3489 ret = -ENOMEM;
3490 goto out;
3491 }
3492 path->reada = READA_BACK;
3493
3494 key.objectid = BTRFS_ORPHAN_OBJECTID;
3495 key.type = BTRFS_ORPHAN_ITEM_KEY;
3496 key.offset = (u64)-1;
3497
3498 while (1) {
3499 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3500 if (ret < 0)
3501 goto out;
3502
3503 /*
3504 * if ret == 0 means we found what we were searching for, which
3505 * is weird, but possible, so only screw with path if we didn't
3506 * find the key and see if we have stuff that matches
3507 */
3508 if (ret > 0) {
3509 ret = 0;
3510 if (path->slots[0] == 0)
3511 break;
3512 path->slots[0]--;
3513 }
3514
3515 /* pull out the item */
3516 leaf = path->nodes[0];
3517 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3518
3519 /* make sure the item matches what we want */
3520 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3521 break;
3522 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3523 break;
3524
3525 /* release the path since we're done with it */
3526 btrfs_release_path(path);
3527
3528 /*
3529 * this is where we are basically btrfs_lookup, without the
3530 * crossing root thing. we store the inode number in the
3531 * offset of the orphan item.
3532 */
3533
3534 if (found_key.offset == last_objectid) {
3535 btrfs_err(fs_info,
3536 "Error removing orphan entry, stopping orphan cleanup");
3537 ret = -EINVAL;
3538 goto out;
3539 }
3540
3541 last_objectid = found_key.offset;
3542
3543 found_key.objectid = found_key.offset;
3544 found_key.type = BTRFS_INODE_ITEM_KEY;
3545 found_key.offset = 0;
3546 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3547 ret = PTR_ERR_OR_ZERO(inode);
3548 if (ret && ret != -ENOENT)
3549 goto out;
3550
3551 if (ret == -ENOENT && root == fs_info->tree_root) {
3552 struct btrfs_root *dead_root;
3553 struct btrfs_fs_info *fs_info = root->fs_info;
3554 int is_dead_root = 0;
3555
3556 /*
3557 * this is an orphan in the tree root. Currently these
3558 * could come from 2 sources:
3559 * a) a snapshot deletion in progress
3560 * b) a free space cache inode
3561 * We need to distinguish those two, as the snapshot
3562 * orphan must not get deleted.
3563 * find_dead_roots already ran before us, so if this
3564 * is a snapshot deletion, we should find the root
3565 * in the dead_roots list
3566 */
3567 spin_lock(&fs_info->trans_lock);
3568 list_for_each_entry(dead_root, &fs_info->dead_roots,
3569 root_list) {
3570 if (dead_root->root_key.objectid ==
3571 found_key.objectid) {
3572 is_dead_root = 1;
3573 break;
3574 }
3575 }
3576 spin_unlock(&fs_info->trans_lock);
3577 if (is_dead_root) {
3578 /* prevent this orphan from being found again */
3579 key.offset = found_key.objectid - 1;
3580 continue;
3581 }
3582 }
3583 /*
3584 * Inode is already gone but the orphan item is still there,
3585 * kill the orphan item.
3586 */
3587 if (ret == -ENOENT) {
3588 trans = btrfs_start_transaction(root, 1);
3589 if (IS_ERR(trans)) {
3590 ret = PTR_ERR(trans);
3591 goto out;
3592 }
3593 btrfs_debug(fs_info, "auto deleting %Lu",
3594 found_key.objectid);
3595 ret = btrfs_del_orphan_item(trans, root,
3596 found_key.objectid);
3597 btrfs_end_transaction(trans);
3598 if (ret)
3599 goto out;
3600 continue;
3601 }
3602
3603 /*
3604 * add this inode to the orphan list so btrfs_orphan_del does
3605 * the proper thing when we hit it
3606 */
3607 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3608 &BTRFS_I(inode)->runtime_flags);
3609 atomic_inc(&root->orphan_inodes);
3610
3611 /* if we have links, this was a truncate, lets do that */
3612 if (inode->i_nlink) {
3613 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3614 iput(inode);
3615 continue;
3616 }
3617 nr_truncate++;
3618
3619 /* 1 for the orphan item deletion. */
3620 trans = btrfs_start_transaction(root, 1);
3621 if (IS_ERR(trans)) {
3622 iput(inode);
3623 ret = PTR_ERR(trans);
3624 goto out;
3625 }
3626 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3627 btrfs_end_transaction(trans);
3628 if (ret) {
3629 iput(inode);
3630 goto out;
3631 }
3632
3633 ret = btrfs_truncate(inode, false);
3634 if (ret)
3635 btrfs_orphan_del(NULL, BTRFS_I(inode));
3636 } else {
3637 nr_unlink++;
3638 }
3639
3640 /* this will do delete_inode and everything for us */
3641 iput(inode);
3642 if (ret)
3643 goto out;
3644 }
3645 /* release the path since we're done with it */
3646 btrfs_release_path(path);
3647
3648 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3649
3650 if (root->orphan_block_rsv)
3651 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3652 (u64)-1);
3653
3654 if (root->orphan_block_rsv ||
3655 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3656 trans = btrfs_join_transaction(root);
3657 if (!IS_ERR(trans))
3658 btrfs_end_transaction(trans);
3659 }
3660
3661 if (nr_unlink)
3662 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3663 if (nr_truncate)
3664 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3665
3666 out:
3667 if (ret)
3668 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3669 btrfs_free_path(path);
3670 return ret;
3671 }
3672
3673 /*
3674 * very simple check to peek ahead in the leaf looking for xattrs. If we
3675 * don't find any xattrs, we know there can't be any acls.
3676 *
3677 * slot is the slot the inode is in, objectid is the objectid of the inode
3678 */
3679 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3680 int slot, u64 objectid,
3681 int *first_xattr_slot)
3682 {
3683 u32 nritems = btrfs_header_nritems(leaf);
3684 struct btrfs_key found_key;
3685 static u64 xattr_access = 0;
3686 static u64 xattr_default = 0;
3687 int scanned = 0;
3688
3689 if (!xattr_access) {
3690 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3691 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3692 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3693 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3694 }
3695
3696 slot++;
3697 *first_xattr_slot = -1;
3698 while (slot < nritems) {
3699 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3700
3701 /* we found a different objectid, there must not be acls */
3702 if (found_key.objectid != objectid)
3703 return 0;
3704
3705 /* we found an xattr, assume we've got an acl */
3706 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3707 if (*first_xattr_slot == -1)
3708 *first_xattr_slot = slot;
3709 if (found_key.offset == xattr_access ||
3710 found_key.offset == xattr_default)
3711 return 1;
3712 }
3713
3714 /*
3715 * we found a key greater than an xattr key, there can't
3716 * be any acls later on
3717 */
3718 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3719 return 0;
3720
3721 slot++;
3722 scanned++;
3723
3724 /*
3725 * it goes inode, inode backrefs, xattrs, extents,
3726 * so if there are a ton of hard links to an inode there can
3727 * be a lot of backrefs. Don't waste time searching too hard,
3728 * this is just an optimization
3729 */
3730 if (scanned >= 8)
3731 break;
3732 }
3733 /* we hit the end of the leaf before we found an xattr or
3734 * something larger than an xattr. We have to assume the inode
3735 * has acls
3736 */
3737 if (*first_xattr_slot == -1)
3738 *first_xattr_slot = slot;
3739 return 1;
3740 }
3741
3742 /*
3743 * read an inode from the btree into the in-memory inode
3744 */
3745 static int btrfs_read_locked_inode(struct inode *inode)
3746 {
3747 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3748 struct btrfs_path *path;
3749 struct extent_buffer *leaf;
3750 struct btrfs_inode_item *inode_item;
3751 struct btrfs_root *root = BTRFS_I(inode)->root;
3752 struct btrfs_key location;
3753 unsigned long ptr;
3754 int maybe_acls;
3755 u32 rdev;
3756 int ret;
3757 bool filled = false;
3758 int first_xattr_slot;
3759
3760 ret = btrfs_fill_inode(inode, &rdev);
3761 if (!ret)
3762 filled = true;
3763
3764 path = btrfs_alloc_path();
3765 if (!path) {
3766 ret = -ENOMEM;
3767 goto make_bad;
3768 }
3769
3770 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3771
3772 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3773 if (ret) {
3774 if (ret > 0)
3775 ret = -ENOENT;
3776 goto make_bad;
3777 }
3778
3779 leaf = path->nodes[0];
3780
3781 if (filled)
3782 goto cache_index;
3783
3784 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3785 struct btrfs_inode_item);
3786 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3787 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3788 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3789 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3790 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3791
3792 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3793 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3794
3795 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3796 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3797
3798 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3799 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3800
3801 BTRFS_I(inode)->i_otime.tv_sec =
3802 btrfs_timespec_sec(leaf, &inode_item->otime);
3803 BTRFS_I(inode)->i_otime.tv_nsec =
3804 btrfs_timespec_nsec(leaf, &inode_item->otime);
3805
3806 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3807 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3808 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3809
3810 inode_set_iversion_queried(inode,
3811 btrfs_inode_sequence(leaf, inode_item));
3812 inode->i_generation = BTRFS_I(inode)->generation;
3813 inode->i_rdev = 0;
3814 rdev = btrfs_inode_rdev(leaf, inode_item);
3815
3816 BTRFS_I(inode)->index_cnt = (u64)-1;
3817 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3818
3819 cache_index:
3820 /*
3821 * If we were modified in the current generation and evicted from memory
3822 * and then re-read we need to do a full sync since we don't have any
3823 * idea about which extents were modified before we were evicted from
3824 * cache.
3825 *
3826 * This is required for both inode re-read from disk and delayed inode
3827 * in delayed_nodes_tree.
3828 */
3829 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3830 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3831 &BTRFS_I(inode)->runtime_flags);
3832
3833 /*
3834 * We don't persist the id of the transaction where an unlink operation
3835 * against the inode was last made. So here we assume the inode might
3836 * have been evicted, and therefore the exact value of last_unlink_trans
3837 * lost, and set it to last_trans to avoid metadata inconsistencies
3838 * between the inode and its parent if the inode is fsync'ed and the log
3839 * replayed. For example, in the scenario:
3840 *
3841 * touch mydir/foo
3842 * ln mydir/foo mydir/bar
3843 * sync
3844 * unlink mydir/bar
3845 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3846 * xfs_io -c fsync mydir/foo
3847 * <power failure>
3848 * mount fs, triggers fsync log replay
3849 *
3850 * We must make sure that when we fsync our inode foo we also log its
3851 * parent inode, otherwise after log replay the parent still has the
3852 * dentry with the "bar" name but our inode foo has a link count of 1
3853 * and doesn't have an inode ref with the name "bar" anymore.
3854 *
3855 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3856 * but it guarantees correctness at the expense of occasional full
3857 * transaction commits on fsync if our inode is a directory, or if our
3858 * inode is not a directory, logging its parent unnecessarily.
3859 */
3860 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3861
3862 path->slots[0]++;
3863 if (inode->i_nlink != 1 ||
3864 path->slots[0] >= btrfs_header_nritems(leaf))
3865 goto cache_acl;
3866
3867 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3868 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3869 goto cache_acl;
3870
3871 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3872 if (location.type == BTRFS_INODE_REF_KEY) {
3873 struct btrfs_inode_ref *ref;
3874
3875 ref = (struct btrfs_inode_ref *)ptr;
3876 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3877 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3878 struct btrfs_inode_extref *extref;
3879
3880 extref = (struct btrfs_inode_extref *)ptr;
3881 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3882 extref);
3883 }
3884 cache_acl:
3885 /*
3886 * try to precache a NULL acl entry for files that don't have
3887 * any xattrs or acls
3888 */
3889 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3890 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3891 if (first_xattr_slot != -1) {
3892 path->slots[0] = first_xattr_slot;
3893 ret = btrfs_load_inode_props(inode, path);
3894 if (ret)
3895 btrfs_err(fs_info,
3896 "error loading props for ino %llu (root %llu): %d",
3897 btrfs_ino(BTRFS_I(inode)),
3898 root->root_key.objectid, ret);
3899 }
3900 btrfs_free_path(path);
3901
3902 if (!maybe_acls)
3903 cache_no_acl(inode);
3904
3905 switch (inode->i_mode & S_IFMT) {
3906 case S_IFREG:
3907 inode->i_mapping->a_ops = &btrfs_aops;
3908 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3909 inode->i_fop = &btrfs_file_operations;
3910 inode->i_op = &btrfs_file_inode_operations;
3911 break;
3912 case S_IFDIR:
3913 inode->i_fop = &btrfs_dir_file_operations;
3914 inode->i_op = &btrfs_dir_inode_operations;
3915 break;
3916 case S_IFLNK:
3917 inode->i_op = &btrfs_symlink_inode_operations;
3918 inode_nohighmem(inode);
3919 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3920 break;
3921 default:
3922 inode->i_op = &btrfs_special_inode_operations;
3923 init_special_inode(inode, inode->i_mode, rdev);
3924 break;
3925 }
3926
3927 btrfs_update_iflags(inode);
3928 return 0;
3929
3930 make_bad:
3931 btrfs_free_path(path);
3932 make_bad_inode(inode);
3933 return ret;
3934 }
3935
3936 /*
3937 * given a leaf and an inode, copy the inode fields into the leaf
3938 */
3939 static void fill_inode_item(struct btrfs_trans_handle *trans,
3940 struct extent_buffer *leaf,
3941 struct btrfs_inode_item *item,
3942 struct inode *inode)
3943 {
3944 struct btrfs_map_token token;
3945
3946 btrfs_init_map_token(&token);
3947
3948 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3949 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3950 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3951 &token);
3952 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3953 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3954
3955 btrfs_set_token_timespec_sec(leaf, &item->atime,
3956 inode->i_atime.tv_sec, &token);
3957 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3958 inode->i_atime.tv_nsec, &token);
3959
3960 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3961 inode->i_mtime.tv_sec, &token);
3962 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3963 inode->i_mtime.tv_nsec, &token);
3964
3965 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3966 inode->i_ctime.tv_sec, &token);
3967 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3968 inode->i_ctime.tv_nsec, &token);
3969
3970 btrfs_set_token_timespec_sec(leaf, &item->otime,
3971 BTRFS_I(inode)->i_otime.tv_sec, &token);
3972 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3973 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3974
3975 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3976 &token);
3977 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3978 &token);
3979 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3980 &token);
3981 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3982 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3983 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3984 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3985 }
3986
3987 /*
3988 * copy everything in the in-memory inode into the btree.
3989 */
3990 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3991 struct btrfs_root *root, struct inode *inode)
3992 {
3993 struct btrfs_inode_item *inode_item;
3994 struct btrfs_path *path;
3995 struct extent_buffer *leaf;
3996 int ret;
3997
3998 path = btrfs_alloc_path();
3999 if (!path)
4000 return -ENOMEM;
4001
4002 path->leave_spinning = 1;
4003 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4004 1);
4005 if (ret) {
4006 if (ret > 0)
4007 ret = -ENOENT;
4008 goto failed;
4009 }
4010
4011 leaf = path->nodes[0];
4012 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4013 struct btrfs_inode_item);
4014
4015 fill_inode_item(trans, leaf, inode_item, inode);
4016 btrfs_mark_buffer_dirty(leaf);
4017 btrfs_set_inode_last_trans(trans, inode);
4018 ret = 0;
4019 failed:
4020 btrfs_free_path(path);
4021 return ret;
4022 }
4023
4024 /*
4025 * copy everything in the in-memory inode into the btree.
4026 */
4027 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4028 struct btrfs_root *root, struct inode *inode)
4029 {
4030 struct btrfs_fs_info *fs_info = root->fs_info;
4031 int ret;
4032
4033 /*
4034 * If the inode is a free space inode, we can deadlock during commit
4035 * if we put it into the delayed code.
4036 *
4037 * The data relocation inode should also be directly updated
4038 * without delay
4039 */
4040 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4041 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4042 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4043 btrfs_update_root_times(trans, root);
4044
4045 ret = btrfs_delayed_update_inode(trans, root, inode);
4046 if (!ret)
4047 btrfs_set_inode_last_trans(trans, inode);
4048 return ret;
4049 }
4050
4051 return btrfs_update_inode_item(trans, root, inode);
4052 }
4053
4054 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4055 struct btrfs_root *root,
4056 struct inode *inode)
4057 {
4058 int ret;
4059
4060 ret = btrfs_update_inode(trans, root, inode);
4061 if (ret == -ENOSPC)
4062 return btrfs_update_inode_item(trans, root, inode);
4063 return ret;
4064 }
4065
4066 /*
4067 * unlink helper that gets used here in inode.c and in the tree logging
4068 * recovery code. It remove a link in a directory with a given name, and
4069 * also drops the back refs in the inode to the directory
4070 */
4071 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4072 struct btrfs_root *root,
4073 struct btrfs_inode *dir,
4074 struct btrfs_inode *inode,
4075 const char *name, int name_len)
4076 {
4077 struct btrfs_fs_info *fs_info = root->fs_info;
4078 struct btrfs_path *path;
4079 int ret = 0;
4080 struct extent_buffer *leaf;
4081 struct btrfs_dir_item *di;
4082 struct btrfs_key key;
4083 u64 index;
4084 u64 ino = btrfs_ino(inode);
4085 u64 dir_ino = btrfs_ino(dir);
4086
4087 path = btrfs_alloc_path();
4088 if (!path) {
4089 ret = -ENOMEM;
4090 goto out;
4091 }
4092
4093 path->leave_spinning = 1;
4094 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4095 name, name_len, -1);
4096 if (IS_ERR(di)) {
4097 ret = PTR_ERR(di);
4098 goto err;
4099 }
4100 if (!di) {
4101 ret = -ENOENT;
4102 goto err;
4103 }
4104 leaf = path->nodes[0];
4105 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4106 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4107 if (ret)
4108 goto err;
4109 btrfs_release_path(path);
4110
4111 /*
4112 * If we don't have dir index, we have to get it by looking up
4113 * the inode ref, since we get the inode ref, remove it directly,
4114 * it is unnecessary to do delayed deletion.
4115 *
4116 * But if we have dir index, needn't search inode ref to get it.
4117 * Since the inode ref is close to the inode item, it is better
4118 * that we delay to delete it, and just do this deletion when
4119 * we update the inode item.
4120 */
4121 if (inode->dir_index) {
4122 ret = btrfs_delayed_delete_inode_ref(inode);
4123 if (!ret) {
4124 index = inode->dir_index;
4125 goto skip_backref;
4126 }
4127 }
4128
4129 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4130 dir_ino, &index);
4131 if (ret) {
4132 btrfs_info(fs_info,
4133 "failed to delete reference to %.*s, inode %llu parent %llu",
4134 name_len, name, ino, dir_ino);
4135 btrfs_abort_transaction(trans, ret);
4136 goto err;
4137 }
4138 skip_backref:
4139 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4140 if (ret) {
4141 btrfs_abort_transaction(trans, ret);
4142 goto err;
4143 }
4144
4145 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4146 dir_ino);
4147 if (ret != 0 && ret != -ENOENT) {
4148 btrfs_abort_transaction(trans, ret);
4149 goto err;
4150 }
4151
4152 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4153 index);
4154 if (ret == -ENOENT)
4155 ret = 0;
4156 else if (ret)
4157 btrfs_abort_transaction(trans, ret);
4158 err:
4159 btrfs_free_path(path);
4160 if (ret)
4161 goto out;
4162
4163 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4164 inode_inc_iversion(&inode->vfs_inode);
4165 inode_inc_iversion(&dir->vfs_inode);
4166 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4167 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4168 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4169 out:
4170 return ret;
4171 }
4172
4173 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4174 struct btrfs_root *root,
4175 struct btrfs_inode *dir, struct btrfs_inode *inode,
4176 const char *name, int name_len)
4177 {
4178 int ret;
4179 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4180 if (!ret) {
4181 drop_nlink(&inode->vfs_inode);
4182 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4183 }
4184 return ret;
4185 }
4186
4187 /*
4188 * helper to start transaction for unlink and rmdir.
4189 *
4190 * unlink and rmdir are special in btrfs, they do not always free space, so
4191 * if we cannot make our reservations the normal way try and see if there is
4192 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4193 * allow the unlink to occur.
4194 */
4195 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4196 {
4197 struct btrfs_root *root = BTRFS_I(dir)->root;
4198
4199 /*
4200 * 1 for the possible orphan item
4201 * 1 for the dir item
4202 * 1 for the dir index
4203 * 1 for the inode ref
4204 * 1 for the inode
4205 */
4206 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4207 }
4208
4209 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4210 {
4211 struct btrfs_root *root = BTRFS_I(dir)->root;
4212 struct btrfs_trans_handle *trans;
4213 struct inode *inode = d_inode(dentry);
4214 int ret;
4215
4216 trans = __unlink_start_trans(dir);
4217 if (IS_ERR(trans))
4218 return PTR_ERR(trans);
4219
4220 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4221 0);
4222
4223 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4224 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4225 dentry->d_name.len);
4226 if (ret)
4227 goto out;
4228
4229 if (inode->i_nlink == 0) {
4230 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4231 if (ret)
4232 goto out;
4233 }
4234
4235 out:
4236 btrfs_end_transaction(trans);
4237 btrfs_btree_balance_dirty(root->fs_info);
4238 return ret;
4239 }
4240
4241 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4242 struct btrfs_root *root,
4243 struct inode *dir, u64 objectid,
4244 const char *name, int name_len)
4245 {
4246 struct btrfs_fs_info *fs_info = root->fs_info;
4247 struct btrfs_path *path;
4248 struct extent_buffer *leaf;
4249 struct btrfs_dir_item *di;
4250 struct btrfs_key key;
4251 u64 index;
4252 int ret;
4253 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4254
4255 path = btrfs_alloc_path();
4256 if (!path)
4257 return -ENOMEM;
4258
4259 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4260 name, name_len, -1);
4261 if (IS_ERR_OR_NULL(di)) {
4262 if (!di)
4263 ret = -ENOENT;
4264 else
4265 ret = PTR_ERR(di);
4266 goto out;
4267 }
4268
4269 leaf = path->nodes[0];
4270 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4271 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4272 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4273 if (ret) {
4274 btrfs_abort_transaction(trans, ret);
4275 goto out;
4276 }
4277 btrfs_release_path(path);
4278
4279 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4280 root->root_key.objectid, dir_ino,
4281 &index, name, name_len);
4282 if (ret < 0) {
4283 if (ret != -ENOENT) {
4284 btrfs_abort_transaction(trans, ret);
4285 goto out;
4286 }
4287 di = btrfs_search_dir_index_item(root, path, dir_ino,
4288 name, name_len);
4289 if (IS_ERR_OR_NULL(di)) {
4290 if (!di)
4291 ret = -ENOENT;
4292 else
4293 ret = PTR_ERR(di);
4294 btrfs_abort_transaction(trans, ret);
4295 goto out;
4296 }
4297
4298 leaf = path->nodes[0];
4299 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4300 btrfs_release_path(path);
4301 index = key.offset;
4302 }
4303 btrfs_release_path(path);
4304
4305 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4306 if (ret) {
4307 btrfs_abort_transaction(trans, ret);
4308 goto out;
4309 }
4310
4311 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4312 inode_inc_iversion(dir);
4313 dir->i_mtime = dir->i_ctime = current_time(dir);
4314 ret = btrfs_update_inode_fallback(trans, root, dir);
4315 if (ret)
4316 btrfs_abort_transaction(trans, ret);
4317 out:
4318 btrfs_free_path(path);
4319 return ret;
4320 }
4321
4322 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4323 {
4324 struct inode *inode = d_inode(dentry);
4325 int err = 0;
4326 struct btrfs_root *root = BTRFS_I(dir)->root;
4327 struct btrfs_trans_handle *trans;
4328 u64 last_unlink_trans;
4329
4330 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4331 return -ENOTEMPTY;
4332 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4333 return -EPERM;
4334
4335 trans = __unlink_start_trans(dir);
4336 if (IS_ERR(trans))
4337 return PTR_ERR(trans);
4338
4339 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4340 err = btrfs_unlink_subvol(trans, root, dir,
4341 BTRFS_I(inode)->location.objectid,
4342 dentry->d_name.name,
4343 dentry->d_name.len);
4344 goto out;
4345 }
4346
4347 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4348 if (err)
4349 goto out;
4350
4351 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4352
4353 /* now the directory is empty */
4354 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4355 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4356 dentry->d_name.len);
4357 if (!err) {
4358 btrfs_i_size_write(BTRFS_I(inode), 0);
4359 /*
4360 * Propagate the last_unlink_trans value of the deleted dir to
4361 * its parent directory. This is to prevent an unrecoverable
4362 * log tree in the case we do something like this:
4363 * 1) create dir foo
4364 * 2) create snapshot under dir foo
4365 * 3) delete the snapshot
4366 * 4) rmdir foo
4367 * 5) mkdir foo
4368 * 6) fsync foo or some file inside foo
4369 */
4370 if (last_unlink_trans >= trans->transid)
4371 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4372 }
4373 out:
4374 btrfs_end_transaction(trans);
4375 btrfs_btree_balance_dirty(root->fs_info);
4376
4377 return err;
4378 }
4379
4380 static int truncate_space_check(struct btrfs_trans_handle *trans,
4381 struct btrfs_root *root,
4382 u64 bytes_deleted)
4383 {
4384 struct btrfs_fs_info *fs_info = root->fs_info;
4385 int ret;
4386
4387 /*
4388 * This is only used to apply pressure to the enospc system, we don't
4389 * intend to use this reservation at all.
4390 */
4391 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4392 bytes_deleted *= fs_info->nodesize;
4393 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4394 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4395 if (!ret) {
4396 trace_btrfs_space_reservation(fs_info, "transaction",
4397 trans->transid,
4398 bytes_deleted, 1);
4399 trans->bytes_reserved += bytes_deleted;
4400 }
4401 return ret;
4402
4403 }
4404
4405 /*
4406 * Return this if we need to call truncate_block for the last bit of the
4407 * truncate.
4408 */
4409 #define NEED_TRUNCATE_BLOCK 1
4410
4411 /*
4412 * this can truncate away extent items, csum items and directory items.
4413 * It starts at a high offset and removes keys until it can't find
4414 * any higher than new_size
4415 *
4416 * csum items that cross the new i_size are truncated to the new size
4417 * as well.
4418 *
4419 * min_type is the minimum key type to truncate down to. If set to 0, this
4420 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4421 */
4422 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4423 struct btrfs_root *root,
4424 struct inode *inode,
4425 u64 new_size, u32 min_type)
4426 {
4427 struct btrfs_fs_info *fs_info = root->fs_info;
4428 struct btrfs_path *path;
4429 struct extent_buffer *leaf;
4430 struct btrfs_file_extent_item *fi;
4431 struct btrfs_key key;
4432 struct btrfs_key found_key;
4433 u64 extent_start = 0;
4434 u64 extent_num_bytes = 0;
4435 u64 extent_offset = 0;
4436 u64 item_end = 0;
4437 u64 last_size = new_size;
4438 u32 found_type = (u8)-1;
4439 int found_extent;
4440 int del_item;
4441 int pending_del_nr = 0;
4442 int pending_del_slot = 0;
4443 int extent_type = -1;
4444 int ret;
4445 int err = 0;
4446 u64 ino = btrfs_ino(BTRFS_I(inode));
4447 u64 bytes_deleted = 0;
4448 bool be_nice = false;
4449 bool should_throttle = false;
4450 bool should_end = false;
4451
4452 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4453
4454 /*
4455 * for non-free space inodes and ref cows, we want to back off from
4456 * time to time
4457 */
4458 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4459 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4460 be_nice = true;
4461
4462 path = btrfs_alloc_path();
4463 if (!path)
4464 return -ENOMEM;
4465 path->reada = READA_BACK;
4466
4467 /*
4468 * We want to drop from the next block forward in case this new size is
4469 * not block aligned since we will be keeping the last block of the
4470 * extent just the way it is.
4471 */
4472 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4473 root == fs_info->tree_root)
4474 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4475 fs_info->sectorsize),
4476 (u64)-1, 0);
4477
4478 /*
4479 * This function is also used to drop the items in the log tree before
4480 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4481 * it is used to drop the loged items. So we shouldn't kill the delayed
4482 * items.
4483 */
4484 if (min_type == 0 && root == BTRFS_I(inode)->root)
4485 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4486
4487 key.objectid = ino;
4488 key.offset = (u64)-1;
4489 key.type = (u8)-1;
4490
4491 search_again:
4492 /*
4493 * with a 16K leaf size and 128MB extents, you can actually queue
4494 * up a huge file in a single leaf. Most of the time that
4495 * bytes_deleted is > 0, it will be huge by the time we get here
4496 */
4497 if (be_nice && bytes_deleted > SZ_32M) {
4498 if (btrfs_should_end_transaction(trans)) {
4499 err = -EAGAIN;
4500 goto error;
4501 }
4502 }
4503
4504
4505 path->leave_spinning = 1;
4506 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4507 if (ret < 0) {
4508 err = ret;
4509 goto out;
4510 }
4511
4512 if (ret > 0) {
4513 /* there are no items in the tree for us to truncate, we're
4514 * done
4515 */
4516 if (path->slots[0] == 0)
4517 goto out;
4518 path->slots[0]--;
4519 }
4520
4521 while (1) {
4522 fi = NULL;
4523 leaf = path->nodes[0];
4524 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4525 found_type = found_key.type;
4526
4527 if (found_key.objectid != ino)
4528 break;
4529
4530 if (found_type < min_type)
4531 break;
4532
4533 item_end = found_key.offset;
4534 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4535 fi = btrfs_item_ptr(leaf, path->slots[0],
4536 struct btrfs_file_extent_item);
4537 extent_type = btrfs_file_extent_type(leaf, fi);
4538 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4539 item_end +=
4540 btrfs_file_extent_num_bytes(leaf, fi);
4541
4542 trace_btrfs_truncate_show_fi_regular(
4543 BTRFS_I(inode), leaf, fi,
4544 found_key.offset);
4545 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4546 item_end += btrfs_file_extent_inline_len(leaf,
4547 path->slots[0], fi);
4548
4549 trace_btrfs_truncate_show_fi_inline(
4550 BTRFS_I(inode), leaf, fi, path->slots[0],
4551 found_key.offset);
4552 }
4553 item_end--;
4554 }
4555 if (found_type > min_type) {
4556 del_item = 1;
4557 } else {
4558 if (item_end < new_size)
4559 break;
4560 if (found_key.offset >= new_size)
4561 del_item = 1;
4562 else
4563 del_item = 0;
4564 }
4565 found_extent = 0;
4566 /* FIXME, shrink the extent if the ref count is only 1 */
4567 if (found_type != BTRFS_EXTENT_DATA_KEY)
4568 goto delete;
4569
4570 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4571 u64 num_dec;
4572 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4573 if (!del_item) {
4574 u64 orig_num_bytes =
4575 btrfs_file_extent_num_bytes(leaf, fi);
4576 extent_num_bytes = ALIGN(new_size -
4577 found_key.offset,
4578 fs_info->sectorsize);
4579 btrfs_set_file_extent_num_bytes(leaf, fi,
4580 extent_num_bytes);
4581 num_dec = (orig_num_bytes -
4582 extent_num_bytes);
4583 if (test_bit(BTRFS_ROOT_REF_COWS,
4584 &root->state) &&
4585 extent_start != 0)
4586 inode_sub_bytes(inode, num_dec);
4587 btrfs_mark_buffer_dirty(leaf);
4588 } else {
4589 extent_num_bytes =
4590 btrfs_file_extent_disk_num_bytes(leaf,
4591 fi);
4592 extent_offset = found_key.offset -
4593 btrfs_file_extent_offset(leaf, fi);
4594
4595 /* FIXME blocksize != 4096 */
4596 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4597 if (extent_start != 0) {
4598 found_extent = 1;
4599 if (test_bit(BTRFS_ROOT_REF_COWS,
4600 &root->state))
4601 inode_sub_bytes(inode, num_dec);
4602 }
4603 }
4604 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4605 /*
4606 * we can't truncate inline items that have had
4607 * special encodings
4608 */
4609 if (!del_item &&
4610 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4611 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4612 btrfs_file_extent_compression(leaf, fi) == 0) {
4613 u32 size = (u32)(new_size - found_key.offset);
4614
4615 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4616 size = btrfs_file_extent_calc_inline_size(size);
4617 btrfs_truncate_item(root->fs_info, path, size, 1);
4618 } else if (!del_item) {
4619 /*
4620 * We have to bail so the last_size is set to
4621 * just before this extent.
4622 */
4623 err = NEED_TRUNCATE_BLOCK;
4624 break;
4625 }
4626
4627 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4628 inode_sub_bytes(inode, item_end + 1 - new_size);
4629 }
4630 delete:
4631 if (del_item)
4632 last_size = found_key.offset;
4633 else
4634 last_size = new_size;
4635 if (del_item) {
4636 if (!pending_del_nr) {
4637 /* no pending yet, add ourselves */
4638 pending_del_slot = path->slots[0];
4639 pending_del_nr = 1;
4640 } else if (pending_del_nr &&
4641 path->slots[0] + 1 == pending_del_slot) {
4642 /* hop on the pending chunk */
4643 pending_del_nr++;
4644 pending_del_slot = path->slots[0];
4645 } else {
4646 BUG();
4647 }
4648 } else {
4649 break;
4650 }
4651 should_throttle = false;
4652
4653 if (found_extent &&
4654 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4655 root == fs_info->tree_root)) {
4656 btrfs_set_path_blocking(path);
4657 bytes_deleted += extent_num_bytes;
4658 ret = btrfs_free_extent(trans, root, extent_start,
4659 extent_num_bytes, 0,
4660 btrfs_header_owner(leaf),
4661 ino, extent_offset);
4662 BUG_ON(ret);
4663 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4664 btrfs_async_run_delayed_refs(fs_info,
4665 trans->delayed_ref_updates * 2,
4666 trans->transid, 0);
4667 if (be_nice) {
4668 if (truncate_space_check(trans, root,
4669 extent_num_bytes)) {
4670 should_end = true;
4671 }
4672 if (btrfs_should_throttle_delayed_refs(trans,
4673 fs_info))
4674 should_throttle = true;
4675 }
4676 }
4677
4678 if (found_type == BTRFS_INODE_ITEM_KEY)
4679 break;
4680
4681 if (path->slots[0] == 0 ||
4682 path->slots[0] != pending_del_slot ||
4683 should_throttle || should_end) {
4684 if (pending_del_nr) {
4685 ret = btrfs_del_items(trans, root, path,
4686 pending_del_slot,
4687 pending_del_nr);
4688 if (ret) {
4689 btrfs_abort_transaction(trans, ret);
4690 goto error;
4691 }
4692 pending_del_nr = 0;
4693 }
4694 btrfs_release_path(path);
4695 if (should_throttle) {
4696 unsigned long updates = trans->delayed_ref_updates;
4697 if (updates) {
4698 trans->delayed_ref_updates = 0;
4699 ret = btrfs_run_delayed_refs(trans,
4700 updates * 2);
4701 if (ret && !err)
4702 err = ret;
4703 }
4704 }
4705 /*
4706 * if we failed to refill our space rsv, bail out
4707 * and let the transaction restart
4708 */
4709 if (should_end) {
4710 err = -EAGAIN;
4711 goto error;
4712 }
4713 goto search_again;
4714 } else {
4715 path->slots[0]--;
4716 }
4717 }
4718 out:
4719 if (pending_del_nr) {
4720 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4721 pending_del_nr);
4722 if (ret)
4723 btrfs_abort_transaction(trans, ret);
4724 }
4725 error:
4726 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4727 ASSERT(last_size >= new_size);
4728 if (!err && last_size > new_size)
4729 last_size = new_size;
4730 btrfs_ordered_update_i_size(inode, last_size, NULL);
4731 }
4732
4733 btrfs_free_path(path);
4734
4735 if (be_nice && bytes_deleted > SZ_32M) {
4736 unsigned long updates = trans->delayed_ref_updates;
4737 if (updates) {
4738 trans->delayed_ref_updates = 0;
4739 ret = btrfs_run_delayed_refs(trans, updates * 2);
4740 if (ret && !err)
4741 err = ret;
4742 }
4743 }
4744 return err;
4745 }
4746
4747 /*
4748 * btrfs_truncate_block - read, zero a chunk and write a block
4749 * @inode - inode that we're zeroing
4750 * @from - the offset to start zeroing
4751 * @len - the length to zero, 0 to zero the entire range respective to the
4752 * offset
4753 * @front - zero up to the offset instead of from the offset on
4754 *
4755 * This will find the block for the "from" offset and cow the block and zero the
4756 * part we want to zero. This is used with truncate and hole punching.
4757 */
4758 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4759 int front)
4760 {
4761 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4762 struct address_space *mapping = inode->i_mapping;
4763 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4764 struct btrfs_ordered_extent *ordered;
4765 struct extent_state *cached_state = NULL;
4766 struct extent_changeset *data_reserved = NULL;
4767 char *kaddr;
4768 u32 blocksize = fs_info->sectorsize;
4769 pgoff_t index = from >> PAGE_SHIFT;
4770 unsigned offset = from & (blocksize - 1);
4771 struct page *page;
4772 gfp_t mask = btrfs_alloc_write_mask(mapping);
4773 int ret = 0;
4774 u64 block_start;
4775 u64 block_end;
4776
4777 if (IS_ALIGNED(offset, blocksize) &&
4778 (!len || IS_ALIGNED(len, blocksize)))
4779 goto out;
4780
4781 block_start = round_down(from, blocksize);
4782 block_end = block_start + blocksize - 1;
4783
4784 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4785 block_start, blocksize);
4786 if (ret)
4787 goto out;
4788
4789 again:
4790 page = find_or_create_page(mapping, index, mask);
4791 if (!page) {
4792 btrfs_delalloc_release_space(inode, data_reserved,
4793 block_start, blocksize, true);
4794 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4795 ret = -ENOMEM;
4796 goto out;
4797 }
4798
4799 if (!PageUptodate(page)) {
4800 ret = btrfs_readpage(NULL, page);
4801 lock_page(page);
4802 if (page->mapping != mapping) {
4803 unlock_page(page);
4804 put_page(page);
4805 goto again;
4806 }
4807 if (!PageUptodate(page)) {
4808 ret = -EIO;
4809 goto out_unlock;
4810 }
4811 }
4812 wait_on_page_writeback(page);
4813
4814 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4815 set_page_extent_mapped(page);
4816
4817 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4818 if (ordered) {
4819 unlock_extent_cached(io_tree, block_start, block_end,
4820 &cached_state);
4821 unlock_page(page);
4822 put_page(page);
4823 btrfs_start_ordered_extent(inode, ordered, 1);
4824 btrfs_put_ordered_extent(ordered);
4825 goto again;
4826 }
4827
4828 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4829 EXTENT_DIRTY | EXTENT_DELALLOC |
4830 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4831 0, 0, &cached_state);
4832
4833 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4834 &cached_state, 0);
4835 if (ret) {
4836 unlock_extent_cached(io_tree, block_start, block_end,
4837 &cached_state);
4838 goto out_unlock;
4839 }
4840
4841 if (offset != blocksize) {
4842 if (!len)
4843 len = blocksize - offset;
4844 kaddr = kmap(page);
4845 if (front)
4846 memset(kaddr + (block_start - page_offset(page)),
4847 0, offset);
4848 else
4849 memset(kaddr + (block_start - page_offset(page)) + offset,
4850 0, len);
4851 flush_dcache_page(page);
4852 kunmap(page);
4853 }
4854 ClearPageChecked(page);
4855 set_page_dirty(page);
4856 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4857
4858 out_unlock:
4859 if (ret)
4860 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4861 blocksize, true);
4862 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4863 unlock_page(page);
4864 put_page(page);
4865 out:
4866 extent_changeset_free(data_reserved);
4867 return ret;
4868 }
4869
4870 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4871 u64 offset, u64 len)
4872 {
4873 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4874 struct btrfs_trans_handle *trans;
4875 int ret;
4876
4877 /*
4878 * Still need to make sure the inode looks like it's been updated so
4879 * that any holes get logged if we fsync.
4880 */
4881 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4882 BTRFS_I(inode)->last_trans = fs_info->generation;
4883 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4884 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4885 return 0;
4886 }
4887
4888 /*
4889 * 1 - for the one we're dropping
4890 * 1 - for the one we're adding
4891 * 1 - for updating the inode.
4892 */
4893 trans = btrfs_start_transaction(root, 3);
4894 if (IS_ERR(trans))
4895 return PTR_ERR(trans);
4896
4897 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4898 if (ret) {
4899 btrfs_abort_transaction(trans, ret);
4900 btrfs_end_transaction(trans);
4901 return ret;
4902 }
4903
4904 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4905 offset, 0, 0, len, 0, len, 0, 0, 0);
4906 if (ret)
4907 btrfs_abort_transaction(trans, ret);
4908 else
4909 btrfs_update_inode(trans, root, inode);
4910 btrfs_end_transaction(trans);
4911 return ret;
4912 }
4913
4914 /*
4915 * This function puts in dummy file extents for the area we're creating a hole
4916 * for. So if we are truncating this file to a larger size we need to insert
4917 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4918 * the range between oldsize and size
4919 */
4920 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4921 {
4922 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4923 struct btrfs_root *root = BTRFS_I(inode)->root;
4924 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4925 struct extent_map *em = NULL;
4926 struct extent_state *cached_state = NULL;
4927 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4928 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4929 u64 block_end = ALIGN(size, fs_info->sectorsize);
4930 u64 last_byte;
4931 u64 cur_offset;
4932 u64 hole_size;
4933 int err = 0;
4934
4935 /*
4936 * If our size started in the middle of a block we need to zero out the
4937 * rest of the block before we expand the i_size, otherwise we could
4938 * expose stale data.
4939 */
4940 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4941 if (err)
4942 return err;
4943
4944 if (size <= hole_start)
4945 return 0;
4946
4947 while (1) {
4948 struct btrfs_ordered_extent *ordered;
4949
4950 lock_extent_bits(io_tree, hole_start, block_end - 1,
4951 &cached_state);
4952 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4953 block_end - hole_start);
4954 if (!ordered)
4955 break;
4956 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4957 &cached_state);
4958 btrfs_start_ordered_extent(inode, ordered, 1);
4959 btrfs_put_ordered_extent(ordered);
4960 }
4961
4962 cur_offset = hole_start;
4963 while (1) {
4964 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4965 block_end - cur_offset, 0);
4966 if (IS_ERR(em)) {
4967 err = PTR_ERR(em);
4968 em = NULL;
4969 break;
4970 }
4971 last_byte = min(extent_map_end(em), block_end);
4972 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4973 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4974 struct extent_map *hole_em;
4975 hole_size = last_byte - cur_offset;
4976
4977 err = maybe_insert_hole(root, inode, cur_offset,
4978 hole_size);
4979 if (err)
4980 break;
4981 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4982 cur_offset + hole_size - 1, 0);
4983 hole_em = alloc_extent_map();
4984 if (!hole_em) {
4985 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4986 &BTRFS_I(inode)->runtime_flags);
4987 goto next;
4988 }
4989 hole_em->start = cur_offset;
4990 hole_em->len = hole_size;
4991 hole_em->orig_start = cur_offset;
4992
4993 hole_em->block_start = EXTENT_MAP_HOLE;
4994 hole_em->block_len = 0;
4995 hole_em->orig_block_len = 0;
4996 hole_em->ram_bytes = hole_size;
4997 hole_em->bdev = fs_info->fs_devices->latest_bdev;
4998 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4999 hole_em->generation = fs_info->generation;
5000
5001 while (1) {
5002 write_lock(&em_tree->lock);
5003 err = add_extent_mapping(em_tree, hole_em, 1);
5004 write_unlock(&em_tree->lock);
5005 if (err != -EEXIST)
5006 break;
5007 btrfs_drop_extent_cache(BTRFS_I(inode),
5008 cur_offset,
5009 cur_offset +
5010 hole_size - 1, 0);
5011 }
5012 free_extent_map(hole_em);
5013 }
5014 next:
5015 free_extent_map(em);
5016 em = NULL;
5017 cur_offset = last_byte;
5018 if (cur_offset >= block_end)
5019 break;
5020 }
5021 free_extent_map(em);
5022 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5023 return err;
5024 }
5025
5026 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5027 {
5028 struct btrfs_root *root = BTRFS_I(inode)->root;
5029 struct btrfs_trans_handle *trans;
5030 loff_t oldsize = i_size_read(inode);
5031 loff_t newsize = attr->ia_size;
5032 int mask = attr->ia_valid;
5033 int ret;
5034
5035 /*
5036 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5037 * special case where we need to update the times despite not having
5038 * these flags set. For all other operations the VFS set these flags
5039 * explicitly if it wants a timestamp update.
5040 */
5041 if (newsize != oldsize) {
5042 inode_inc_iversion(inode);
5043 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5044 inode->i_ctime = inode->i_mtime =
5045 current_time(inode);
5046 }
5047
5048 if (newsize > oldsize) {
5049 /*
5050 * Don't do an expanding truncate while snapshotting is ongoing.
5051 * This is to ensure the snapshot captures a fully consistent
5052 * state of this file - if the snapshot captures this expanding
5053 * truncation, it must capture all writes that happened before
5054 * this truncation.
5055 */
5056 btrfs_wait_for_snapshot_creation(root);
5057 ret = btrfs_cont_expand(inode, oldsize, newsize);
5058 if (ret) {
5059 btrfs_end_write_no_snapshotting(root);
5060 return ret;
5061 }
5062
5063 trans = btrfs_start_transaction(root, 1);
5064 if (IS_ERR(trans)) {
5065 btrfs_end_write_no_snapshotting(root);
5066 return PTR_ERR(trans);
5067 }
5068
5069 i_size_write(inode, newsize);
5070 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5071 pagecache_isize_extended(inode, oldsize, newsize);
5072 ret = btrfs_update_inode(trans, root, inode);
5073 btrfs_end_write_no_snapshotting(root);
5074 btrfs_end_transaction(trans);
5075 } else {
5076
5077 /*
5078 * We're truncating a file that used to have good data down to
5079 * zero. Make sure it gets into the ordered flush list so that
5080 * any new writes get down to disk quickly.
5081 */
5082 if (newsize == 0)
5083 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5084 &BTRFS_I(inode)->runtime_flags);
5085
5086 /*
5087 * 1 for the orphan item we're going to add
5088 * 1 for the orphan item deletion.
5089 */
5090 trans = btrfs_start_transaction(root, 2);
5091 if (IS_ERR(trans))
5092 return PTR_ERR(trans);
5093
5094 /*
5095 * We need to do this in case we fail at _any_ point during the
5096 * actual truncate. Once we do the truncate_setsize we could
5097 * invalidate pages which forces any outstanding ordered io to
5098 * be instantly completed which will give us extents that need
5099 * to be truncated. If we fail to get an orphan inode down we
5100 * could have left over extents that were never meant to live,
5101 * so we need to guarantee from this point on that everything
5102 * will be consistent.
5103 */
5104 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5105 btrfs_end_transaction(trans);
5106 if (ret)
5107 return ret;
5108
5109 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5110 truncate_setsize(inode, newsize);
5111
5112 /* Disable nonlocked read DIO to avoid the end less truncate */
5113 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5114 inode_dio_wait(inode);
5115 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5116
5117 ret = btrfs_truncate(inode, newsize == oldsize);
5118 if (ret && inode->i_nlink) {
5119 int err;
5120
5121 /* To get a stable disk_i_size */
5122 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5123 if (err) {
5124 btrfs_orphan_del(NULL, BTRFS_I(inode));
5125 return err;
5126 }
5127
5128 /*
5129 * failed to truncate, disk_i_size is only adjusted down
5130 * as we remove extents, so it should represent the true
5131 * size of the inode, so reset the in memory size and
5132 * delete our orphan entry.
5133 */
5134 trans = btrfs_join_transaction(root);
5135 if (IS_ERR(trans)) {
5136 btrfs_orphan_del(NULL, BTRFS_I(inode));
5137 return ret;
5138 }
5139 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5140 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5141 if (err)
5142 btrfs_abort_transaction(trans, err);
5143 btrfs_end_transaction(trans);
5144 }
5145 }
5146
5147 return ret;
5148 }
5149
5150 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5151 {
5152 struct inode *inode = d_inode(dentry);
5153 struct btrfs_root *root = BTRFS_I(inode)->root;
5154 int err;
5155
5156 if (btrfs_root_readonly(root))
5157 return -EROFS;
5158
5159 err = setattr_prepare(dentry, attr);
5160 if (err)
5161 return err;
5162
5163 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5164 err = btrfs_setsize(inode, attr);
5165 if (err)
5166 return err;
5167 }
5168
5169 if (attr->ia_valid) {
5170 setattr_copy(inode, attr);
5171 inode_inc_iversion(inode);
5172 err = btrfs_dirty_inode(inode);
5173
5174 if (!err && attr->ia_valid & ATTR_MODE)
5175 err = posix_acl_chmod(inode, inode->i_mode);
5176 }
5177
5178 return err;
5179 }
5180
5181 /*
5182 * While truncating the inode pages during eviction, we get the VFS calling
5183 * btrfs_invalidatepage() against each page of the inode. This is slow because
5184 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5185 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5186 * extent_state structures over and over, wasting lots of time.
5187 *
5188 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5189 * those expensive operations on a per page basis and do only the ordered io
5190 * finishing, while we release here the extent_map and extent_state structures,
5191 * without the excessive merging and splitting.
5192 */
5193 static void evict_inode_truncate_pages(struct inode *inode)
5194 {
5195 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5196 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5197 struct rb_node *node;
5198
5199 ASSERT(inode->i_state & I_FREEING);
5200 truncate_inode_pages_final(&inode->i_data);
5201
5202 write_lock(&map_tree->lock);
5203 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5204 struct extent_map *em;
5205
5206 node = rb_first(&map_tree->map);
5207 em = rb_entry(node, struct extent_map, rb_node);
5208 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5209 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5210 remove_extent_mapping(map_tree, em);
5211 free_extent_map(em);
5212 if (need_resched()) {
5213 write_unlock(&map_tree->lock);
5214 cond_resched();
5215 write_lock(&map_tree->lock);
5216 }
5217 }
5218 write_unlock(&map_tree->lock);
5219
5220 /*
5221 * Keep looping until we have no more ranges in the io tree.
5222 * We can have ongoing bios started by readpages (called from readahead)
5223 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5224 * still in progress (unlocked the pages in the bio but did not yet
5225 * unlocked the ranges in the io tree). Therefore this means some
5226 * ranges can still be locked and eviction started because before
5227 * submitting those bios, which are executed by a separate task (work
5228 * queue kthread), inode references (inode->i_count) were not taken
5229 * (which would be dropped in the end io callback of each bio).
5230 * Therefore here we effectively end up waiting for those bios and
5231 * anyone else holding locked ranges without having bumped the inode's
5232 * reference count - if we don't do it, when they access the inode's
5233 * io_tree to unlock a range it may be too late, leading to an
5234 * use-after-free issue.
5235 */
5236 spin_lock(&io_tree->lock);
5237 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5238 struct extent_state *state;
5239 struct extent_state *cached_state = NULL;
5240 u64 start;
5241 u64 end;
5242
5243 node = rb_first(&io_tree->state);
5244 state = rb_entry(node, struct extent_state, rb_node);
5245 start = state->start;
5246 end = state->end;
5247 spin_unlock(&io_tree->lock);
5248
5249 lock_extent_bits(io_tree, start, end, &cached_state);
5250
5251 /*
5252 * If still has DELALLOC flag, the extent didn't reach disk,
5253 * and its reserved space won't be freed by delayed_ref.
5254 * So we need to free its reserved space here.
5255 * (Refer to comment in btrfs_invalidatepage, case 2)
5256 *
5257 * Note, end is the bytenr of last byte, so we need + 1 here.
5258 */
5259 if (state->state & EXTENT_DELALLOC)
5260 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5261
5262 clear_extent_bit(io_tree, start, end,
5263 EXTENT_LOCKED | EXTENT_DIRTY |
5264 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5265 EXTENT_DEFRAG, 1, 1, &cached_state);
5266
5267 cond_resched();
5268 spin_lock(&io_tree->lock);
5269 }
5270 spin_unlock(&io_tree->lock);
5271 }
5272
5273 void btrfs_evict_inode(struct inode *inode)
5274 {
5275 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5276 struct btrfs_trans_handle *trans;
5277 struct btrfs_root *root = BTRFS_I(inode)->root;
5278 struct btrfs_block_rsv *rsv, *global_rsv;
5279 int steal_from_global = 0;
5280 u64 min_size;
5281 int ret;
5282
5283 trace_btrfs_inode_evict(inode);
5284
5285 if (!root) {
5286 clear_inode(inode);
5287 return;
5288 }
5289
5290 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5291
5292 evict_inode_truncate_pages(inode);
5293
5294 if (inode->i_nlink &&
5295 ((btrfs_root_refs(&root->root_item) != 0 &&
5296 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5297 btrfs_is_free_space_inode(BTRFS_I(inode))))
5298 goto no_delete;
5299
5300 if (is_bad_inode(inode)) {
5301 btrfs_orphan_del(NULL, BTRFS_I(inode));
5302 goto no_delete;
5303 }
5304 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5305 if (!special_file(inode->i_mode))
5306 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5307
5308 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5309
5310 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5311 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5312 &BTRFS_I(inode)->runtime_flags));
5313 goto no_delete;
5314 }
5315
5316 if (inode->i_nlink > 0) {
5317 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5318 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5319 goto no_delete;
5320 }
5321
5322 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5323 if (ret) {
5324 btrfs_orphan_del(NULL, BTRFS_I(inode));
5325 goto no_delete;
5326 }
5327
5328 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5329 if (!rsv) {
5330 btrfs_orphan_del(NULL, BTRFS_I(inode));
5331 goto no_delete;
5332 }
5333 rsv->size = min_size;
5334 rsv->failfast = 1;
5335 global_rsv = &fs_info->global_block_rsv;
5336
5337 btrfs_i_size_write(BTRFS_I(inode), 0);
5338
5339 /*
5340 * This is a bit simpler than btrfs_truncate since we've already
5341 * reserved our space for our orphan item in the unlink, so we just
5342 * need to reserve some slack space in case we add bytes and update
5343 * inode item when doing the truncate.
5344 */
5345 while (1) {
5346 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5347 BTRFS_RESERVE_FLUSH_LIMIT);
5348
5349 /*
5350 * Try and steal from the global reserve since we will
5351 * likely not use this space anyway, we want to try as
5352 * hard as possible to get this to work.
5353 */
5354 if (ret)
5355 steal_from_global++;
5356 else
5357 steal_from_global = 0;
5358 ret = 0;
5359
5360 /*
5361 * steal_from_global == 0: we reserved stuff, hooray!
5362 * steal_from_global == 1: we didn't reserve stuff, boo!
5363 * steal_from_global == 2: we've committed, still not a lot of
5364 * room but maybe we'll have room in the global reserve this
5365 * time.
5366 * steal_from_global == 3: abandon all hope!
5367 */
5368 if (steal_from_global > 2) {
5369 btrfs_warn(fs_info,
5370 "Could not get space for a delete, will truncate on mount %d",
5371 ret);
5372 btrfs_orphan_del(NULL, BTRFS_I(inode));
5373 btrfs_free_block_rsv(fs_info, rsv);
5374 goto no_delete;
5375 }
5376
5377 trans = btrfs_join_transaction(root);
5378 if (IS_ERR(trans)) {
5379 btrfs_orphan_del(NULL, BTRFS_I(inode));
5380 btrfs_free_block_rsv(fs_info, rsv);
5381 goto no_delete;
5382 }
5383
5384 /*
5385 * We can't just steal from the global reserve, we need to make
5386 * sure there is room to do it, if not we need to commit and try
5387 * again.
5388 */
5389 if (steal_from_global) {
5390 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5391 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5392 min_size, 0);
5393 else
5394 ret = -ENOSPC;
5395 }
5396
5397 /*
5398 * Couldn't steal from the global reserve, we have too much
5399 * pending stuff built up, commit the transaction and try it
5400 * again.
5401 */
5402 if (ret) {
5403 ret = btrfs_commit_transaction(trans);
5404 if (ret) {
5405 btrfs_orphan_del(NULL, BTRFS_I(inode));
5406 btrfs_free_block_rsv(fs_info, rsv);
5407 goto no_delete;
5408 }
5409 continue;
5410 } else {
5411 steal_from_global = 0;
5412 }
5413
5414 trans->block_rsv = rsv;
5415
5416 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5417 if (ret != -ENOSPC && ret != -EAGAIN)
5418 break;
5419
5420 trans->block_rsv = &fs_info->trans_block_rsv;
5421 btrfs_end_transaction(trans);
5422 trans = NULL;
5423 btrfs_btree_balance_dirty(fs_info);
5424 }
5425
5426 btrfs_free_block_rsv(fs_info, rsv);
5427
5428 /*
5429 * Errors here aren't a big deal, it just means we leave orphan items
5430 * in the tree. They will be cleaned up on the next mount.
5431 */
5432 if (ret == 0) {
5433 trans->block_rsv = root->orphan_block_rsv;
5434 btrfs_orphan_del(trans, BTRFS_I(inode));
5435 } else {
5436 btrfs_orphan_del(NULL, BTRFS_I(inode));
5437 }
5438
5439 trans->block_rsv = &fs_info->trans_block_rsv;
5440 if (!(root == fs_info->tree_root ||
5441 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5442 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5443
5444 btrfs_end_transaction(trans);
5445 btrfs_btree_balance_dirty(fs_info);
5446 no_delete:
5447 btrfs_remove_delayed_node(BTRFS_I(inode));
5448 clear_inode(inode);
5449 }
5450
5451 /*
5452 * this returns the key found in the dir entry in the location pointer.
5453 * If no dir entries were found, returns -ENOENT.
5454 * If found a corrupted location in dir entry, returns -EUCLEAN.
5455 */
5456 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5457 struct btrfs_key *location)
5458 {
5459 const char *name = dentry->d_name.name;
5460 int namelen = dentry->d_name.len;
5461 struct btrfs_dir_item *di;
5462 struct btrfs_path *path;
5463 struct btrfs_root *root = BTRFS_I(dir)->root;
5464 int ret = 0;
5465
5466 path = btrfs_alloc_path();
5467 if (!path)
5468 return -ENOMEM;
5469
5470 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5471 name, namelen, 0);
5472 if (!di) {
5473 ret = -ENOENT;
5474 goto out;
5475 }
5476 if (IS_ERR(di)) {
5477 ret = PTR_ERR(di);
5478 goto out;
5479 }
5480
5481 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5482 if (location->type != BTRFS_INODE_ITEM_KEY &&
5483 location->type != BTRFS_ROOT_ITEM_KEY) {
5484 ret = -EUCLEAN;
5485 btrfs_warn(root->fs_info,
5486 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5487 __func__, name, btrfs_ino(BTRFS_I(dir)),
5488 location->objectid, location->type, location->offset);
5489 }
5490 out:
5491 btrfs_free_path(path);
5492 return ret;
5493 }
5494
5495 /*
5496 * when we hit a tree root in a directory, the btrfs part of the inode
5497 * needs to be changed to reflect the root directory of the tree root. This
5498 * is kind of like crossing a mount point.
5499 */
5500 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5501 struct inode *dir,
5502 struct dentry *dentry,
5503 struct btrfs_key *location,
5504 struct btrfs_root **sub_root)
5505 {
5506 struct btrfs_path *path;
5507 struct btrfs_root *new_root;
5508 struct btrfs_root_ref *ref;
5509 struct extent_buffer *leaf;
5510 struct btrfs_key key;
5511 int ret;
5512 int err = 0;
5513
5514 path = btrfs_alloc_path();
5515 if (!path) {
5516 err = -ENOMEM;
5517 goto out;
5518 }
5519
5520 err = -ENOENT;
5521 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5522 key.type = BTRFS_ROOT_REF_KEY;
5523 key.offset = location->objectid;
5524
5525 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5526 if (ret) {
5527 if (ret < 0)
5528 err = ret;
5529 goto out;
5530 }
5531
5532 leaf = path->nodes[0];
5533 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5534 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5535 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5536 goto out;
5537
5538 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5539 (unsigned long)(ref + 1),
5540 dentry->d_name.len);
5541 if (ret)
5542 goto out;
5543
5544 btrfs_release_path(path);
5545
5546 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5547 if (IS_ERR(new_root)) {
5548 err = PTR_ERR(new_root);
5549 goto out;
5550 }
5551
5552 *sub_root = new_root;
5553 location->objectid = btrfs_root_dirid(&new_root->root_item);
5554 location->type = BTRFS_INODE_ITEM_KEY;
5555 location->offset = 0;
5556 err = 0;
5557 out:
5558 btrfs_free_path(path);
5559 return err;
5560 }
5561
5562 static void inode_tree_add(struct inode *inode)
5563 {
5564 struct btrfs_root *root = BTRFS_I(inode)->root;
5565 struct btrfs_inode *entry;
5566 struct rb_node **p;
5567 struct rb_node *parent;
5568 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5569 u64 ino = btrfs_ino(BTRFS_I(inode));
5570
5571 if (inode_unhashed(inode))
5572 return;
5573 parent = NULL;
5574 spin_lock(&root->inode_lock);
5575 p = &root->inode_tree.rb_node;
5576 while (*p) {
5577 parent = *p;
5578 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5579
5580 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5581 p = &parent->rb_left;
5582 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5583 p = &parent->rb_right;
5584 else {
5585 WARN_ON(!(entry->vfs_inode.i_state &
5586 (I_WILL_FREE | I_FREEING)));
5587 rb_replace_node(parent, new, &root->inode_tree);
5588 RB_CLEAR_NODE(parent);
5589 spin_unlock(&root->inode_lock);
5590 return;
5591 }
5592 }
5593 rb_link_node(new, parent, p);
5594 rb_insert_color(new, &root->inode_tree);
5595 spin_unlock(&root->inode_lock);
5596 }
5597
5598 static void inode_tree_del(struct inode *inode)
5599 {
5600 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5601 struct btrfs_root *root = BTRFS_I(inode)->root;
5602 int empty = 0;
5603
5604 spin_lock(&root->inode_lock);
5605 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5606 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5607 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5608 empty = RB_EMPTY_ROOT(&root->inode_tree);
5609 }
5610 spin_unlock(&root->inode_lock);
5611
5612 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5613 synchronize_srcu(&fs_info->subvol_srcu);
5614 spin_lock(&root->inode_lock);
5615 empty = RB_EMPTY_ROOT(&root->inode_tree);
5616 spin_unlock(&root->inode_lock);
5617 if (empty)
5618 btrfs_add_dead_root(root);
5619 }
5620 }
5621
5622 void btrfs_invalidate_inodes(struct btrfs_root *root)
5623 {
5624 struct btrfs_fs_info *fs_info = root->fs_info;
5625 struct rb_node *node;
5626 struct rb_node *prev;
5627 struct btrfs_inode *entry;
5628 struct inode *inode;
5629 u64 objectid = 0;
5630
5631 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5632 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5633
5634 spin_lock(&root->inode_lock);
5635 again:
5636 node = root->inode_tree.rb_node;
5637 prev = NULL;
5638 while (node) {
5639 prev = node;
5640 entry = rb_entry(node, struct btrfs_inode, rb_node);
5641
5642 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5643 node = node->rb_left;
5644 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5645 node = node->rb_right;
5646 else
5647 break;
5648 }
5649 if (!node) {
5650 while (prev) {
5651 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5652 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5653 node = prev;
5654 break;
5655 }
5656 prev = rb_next(prev);
5657 }
5658 }
5659 while (node) {
5660 entry = rb_entry(node, struct btrfs_inode, rb_node);
5661 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5662 inode = igrab(&entry->vfs_inode);
5663 if (inode) {
5664 spin_unlock(&root->inode_lock);
5665 if (atomic_read(&inode->i_count) > 1)
5666 d_prune_aliases(inode);
5667 /*
5668 * btrfs_drop_inode will have it removed from
5669 * the inode cache when its usage count
5670 * hits zero.
5671 */
5672 iput(inode);
5673 cond_resched();
5674 spin_lock(&root->inode_lock);
5675 goto again;
5676 }
5677
5678 if (cond_resched_lock(&root->inode_lock))
5679 goto again;
5680
5681 node = rb_next(node);
5682 }
5683 spin_unlock(&root->inode_lock);
5684 }
5685
5686 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5687 {
5688 struct btrfs_iget_args *args = p;
5689 inode->i_ino = args->location->objectid;
5690 memcpy(&BTRFS_I(inode)->location, args->location,
5691 sizeof(*args->location));
5692 BTRFS_I(inode)->root = args->root;
5693 return 0;
5694 }
5695
5696 static int btrfs_find_actor(struct inode *inode, void *opaque)
5697 {
5698 struct btrfs_iget_args *args = opaque;
5699 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5700 args->root == BTRFS_I(inode)->root;
5701 }
5702
5703 static struct inode *btrfs_iget_locked(struct super_block *s,
5704 struct btrfs_key *location,
5705 struct btrfs_root *root)
5706 {
5707 struct inode *inode;
5708 struct btrfs_iget_args args;
5709 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5710
5711 args.location = location;
5712 args.root = root;
5713
5714 inode = iget5_locked(s, hashval, btrfs_find_actor,
5715 btrfs_init_locked_inode,
5716 (void *)&args);
5717 return inode;
5718 }
5719
5720 /* Get an inode object given its location and corresponding root.
5721 * Returns in *is_new if the inode was read from disk
5722 */
5723 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5724 struct btrfs_root *root, int *new)
5725 {
5726 struct inode *inode;
5727
5728 inode = btrfs_iget_locked(s, location, root);
5729 if (!inode)
5730 return ERR_PTR(-ENOMEM);
5731
5732 if (inode->i_state & I_NEW) {
5733 int ret;
5734
5735 ret = btrfs_read_locked_inode(inode);
5736 if (!is_bad_inode(inode)) {
5737 inode_tree_add(inode);
5738 unlock_new_inode(inode);
5739 if (new)
5740 *new = 1;
5741 } else {
5742 unlock_new_inode(inode);
5743 iput(inode);
5744 ASSERT(ret < 0);
5745 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5746 }
5747 }
5748
5749 return inode;
5750 }
5751
5752 static struct inode *new_simple_dir(struct super_block *s,
5753 struct btrfs_key *key,
5754 struct btrfs_root *root)
5755 {
5756 struct inode *inode = new_inode(s);
5757
5758 if (!inode)
5759 return ERR_PTR(-ENOMEM);
5760
5761 BTRFS_I(inode)->root = root;
5762 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5763 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5764
5765 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5766 inode->i_op = &btrfs_dir_ro_inode_operations;
5767 inode->i_opflags &= ~IOP_XATTR;
5768 inode->i_fop = &simple_dir_operations;
5769 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5770 inode->i_mtime = current_time(inode);
5771 inode->i_atime = inode->i_mtime;
5772 inode->i_ctime = inode->i_mtime;
5773 BTRFS_I(inode)->i_otime = inode->i_mtime;
5774
5775 return inode;
5776 }
5777
5778 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5779 {
5780 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5781 struct inode *inode;
5782 struct btrfs_root *root = BTRFS_I(dir)->root;
5783 struct btrfs_root *sub_root = root;
5784 struct btrfs_key location;
5785 int index;
5786 int ret = 0;
5787
5788 if (dentry->d_name.len > BTRFS_NAME_LEN)
5789 return ERR_PTR(-ENAMETOOLONG);
5790
5791 ret = btrfs_inode_by_name(dir, dentry, &location);
5792 if (ret < 0)
5793 return ERR_PTR(ret);
5794
5795 if (location.type == BTRFS_INODE_ITEM_KEY) {
5796 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5797 return inode;
5798 }
5799
5800 index = srcu_read_lock(&fs_info->subvol_srcu);
5801 ret = fixup_tree_root_location(fs_info, dir, dentry,
5802 &location, &sub_root);
5803 if (ret < 0) {
5804 if (ret != -ENOENT)
5805 inode = ERR_PTR(ret);
5806 else
5807 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5808 } else {
5809 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5810 }
5811 srcu_read_unlock(&fs_info->subvol_srcu, index);
5812
5813 if (!IS_ERR(inode) && root != sub_root) {
5814 down_read(&fs_info->cleanup_work_sem);
5815 if (!sb_rdonly(inode->i_sb))
5816 ret = btrfs_orphan_cleanup(sub_root);
5817 up_read(&fs_info->cleanup_work_sem);
5818 if (ret) {
5819 iput(inode);
5820 inode = ERR_PTR(ret);
5821 }
5822 }
5823
5824 return inode;
5825 }
5826
5827 static int btrfs_dentry_delete(const struct dentry *dentry)
5828 {
5829 struct btrfs_root *root;
5830 struct inode *inode = d_inode(dentry);
5831
5832 if (!inode && !IS_ROOT(dentry))
5833 inode = d_inode(dentry->d_parent);
5834
5835 if (inode) {
5836 root = BTRFS_I(inode)->root;
5837 if (btrfs_root_refs(&root->root_item) == 0)
5838 return 1;
5839
5840 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5841 return 1;
5842 }
5843 return 0;
5844 }
5845
5846 static void btrfs_dentry_release(struct dentry *dentry)
5847 {
5848 kfree(dentry->d_fsdata);
5849 }
5850
5851 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5852 unsigned int flags)
5853 {
5854 struct inode *inode;
5855
5856 inode = btrfs_lookup_dentry(dir, dentry);
5857 if (IS_ERR(inode)) {
5858 if (PTR_ERR(inode) == -ENOENT)
5859 inode = NULL;
5860 else
5861 return ERR_CAST(inode);
5862 }
5863
5864 return d_splice_alias(inode, dentry);
5865 }
5866
5867 unsigned char btrfs_filetype_table[] = {
5868 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5869 };
5870
5871 /*
5872 * All this infrastructure exists because dir_emit can fault, and we are holding
5873 * the tree lock when doing readdir. For now just allocate a buffer and copy
5874 * our information into that, and then dir_emit from the buffer. This is
5875 * similar to what NFS does, only we don't keep the buffer around in pagecache
5876 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5877 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5878 * tree lock.
5879 */
5880 static int btrfs_opendir(struct inode *inode, struct file *file)
5881 {
5882 struct btrfs_file_private *private;
5883
5884 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5885 if (!private)
5886 return -ENOMEM;
5887 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5888 if (!private->filldir_buf) {
5889 kfree(private);
5890 return -ENOMEM;
5891 }
5892 file->private_data = private;
5893 return 0;
5894 }
5895
5896 struct dir_entry {
5897 u64 ino;
5898 u64 offset;
5899 unsigned type;
5900 int name_len;
5901 };
5902
5903 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5904 {
5905 while (entries--) {
5906 struct dir_entry *entry = addr;
5907 char *name = (char *)(entry + 1);
5908
5909 ctx->pos = get_unaligned(&entry->offset);
5910 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5911 get_unaligned(&entry->ino),
5912 get_unaligned(&entry->type)))
5913 return 1;
5914 addr += sizeof(struct dir_entry) +
5915 get_unaligned(&entry->name_len);
5916 ctx->pos++;
5917 }
5918 return 0;
5919 }
5920
5921 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5922 {
5923 struct inode *inode = file_inode(file);
5924 struct btrfs_root *root = BTRFS_I(inode)->root;
5925 struct btrfs_file_private *private = file->private_data;
5926 struct btrfs_dir_item *di;
5927 struct btrfs_key key;
5928 struct btrfs_key found_key;
5929 struct btrfs_path *path;
5930 void *addr;
5931 struct list_head ins_list;
5932 struct list_head del_list;
5933 int ret;
5934 struct extent_buffer *leaf;
5935 int slot;
5936 char *name_ptr;
5937 int name_len;
5938 int entries = 0;
5939 int total_len = 0;
5940 bool put = false;
5941 struct btrfs_key location;
5942
5943 if (!dir_emit_dots(file, ctx))
5944 return 0;
5945
5946 path = btrfs_alloc_path();
5947 if (!path)
5948 return -ENOMEM;
5949
5950 addr = private->filldir_buf;
5951 path->reada = READA_FORWARD;
5952
5953 INIT_LIST_HEAD(&ins_list);
5954 INIT_LIST_HEAD(&del_list);
5955 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5956
5957 again:
5958 key.type = BTRFS_DIR_INDEX_KEY;
5959 key.offset = ctx->pos;
5960 key.objectid = btrfs_ino(BTRFS_I(inode));
5961
5962 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5963 if (ret < 0)
5964 goto err;
5965
5966 while (1) {
5967 struct dir_entry *entry;
5968
5969 leaf = path->nodes[0];
5970 slot = path->slots[0];
5971 if (slot >= btrfs_header_nritems(leaf)) {
5972 ret = btrfs_next_leaf(root, path);
5973 if (ret < 0)
5974 goto err;
5975 else if (ret > 0)
5976 break;
5977 continue;
5978 }
5979
5980 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5981
5982 if (found_key.objectid != key.objectid)
5983 break;
5984 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5985 break;
5986 if (found_key.offset < ctx->pos)
5987 goto next;
5988 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5989 goto next;
5990 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5991 name_len = btrfs_dir_name_len(leaf, di);
5992 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5993 PAGE_SIZE) {
5994 btrfs_release_path(path);
5995 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5996 if (ret)
5997 goto nopos;
5998 addr = private->filldir_buf;
5999 entries = 0;
6000 total_len = 0;
6001 goto again;
6002 }
6003
6004 entry = addr;
6005 put_unaligned(name_len, &entry->name_len);
6006 name_ptr = (char *)(entry + 1);
6007 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6008 name_len);
6009 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
6010 &entry->type);
6011 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6012 put_unaligned(location.objectid, &entry->ino);
6013 put_unaligned(found_key.offset, &entry->offset);
6014 entries++;
6015 addr += sizeof(struct dir_entry) + name_len;
6016 total_len += sizeof(struct dir_entry) + name_len;
6017 next:
6018 path->slots[0]++;
6019 }
6020 btrfs_release_path(path);
6021
6022 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6023 if (ret)
6024 goto nopos;
6025
6026 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6027 if (ret)
6028 goto nopos;
6029
6030 /*
6031 * Stop new entries from being returned after we return the last
6032 * entry.
6033 *
6034 * New directory entries are assigned a strictly increasing
6035 * offset. This means that new entries created during readdir
6036 * are *guaranteed* to be seen in the future by that readdir.
6037 * This has broken buggy programs which operate on names as
6038 * they're returned by readdir. Until we re-use freed offsets
6039 * we have this hack to stop new entries from being returned
6040 * under the assumption that they'll never reach this huge
6041 * offset.
6042 *
6043 * This is being careful not to overflow 32bit loff_t unless the
6044 * last entry requires it because doing so has broken 32bit apps
6045 * in the past.
6046 */
6047 if (ctx->pos >= INT_MAX)
6048 ctx->pos = LLONG_MAX;
6049 else
6050 ctx->pos = INT_MAX;
6051 nopos:
6052 ret = 0;
6053 err:
6054 if (put)
6055 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6056 btrfs_free_path(path);
6057 return ret;
6058 }
6059
6060 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6061 {
6062 struct btrfs_root *root = BTRFS_I(inode)->root;
6063 struct btrfs_trans_handle *trans;
6064 int ret = 0;
6065 bool nolock = false;
6066
6067 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6068 return 0;
6069
6070 if (btrfs_fs_closing(root->fs_info) &&
6071 btrfs_is_free_space_inode(BTRFS_I(inode)))
6072 nolock = true;
6073
6074 if (wbc->sync_mode == WB_SYNC_ALL) {
6075 if (nolock)
6076 trans = btrfs_join_transaction_nolock(root);
6077 else
6078 trans = btrfs_join_transaction(root);
6079 if (IS_ERR(trans))
6080 return PTR_ERR(trans);
6081 ret = btrfs_commit_transaction(trans);
6082 }
6083 return ret;
6084 }
6085
6086 /*
6087 * This is somewhat expensive, updating the tree every time the
6088 * inode changes. But, it is most likely to find the inode in cache.
6089 * FIXME, needs more benchmarking...there are no reasons other than performance
6090 * to keep or drop this code.
6091 */
6092 static int btrfs_dirty_inode(struct inode *inode)
6093 {
6094 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6095 struct btrfs_root *root = BTRFS_I(inode)->root;
6096 struct btrfs_trans_handle *trans;
6097 int ret;
6098
6099 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6100 return 0;
6101
6102 trans = btrfs_join_transaction(root);
6103 if (IS_ERR(trans))
6104 return PTR_ERR(trans);
6105
6106 ret = btrfs_update_inode(trans, root, inode);
6107 if (ret && ret == -ENOSPC) {
6108 /* whoops, lets try again with the full transaction */
6109 btrfs_end_transaction(trans);
6110 trans = btrfs_start_transaction(root, 1);
6111 if (IS_ERR(trans))
6112 return PTR_ERR(trans);
6113
6114 ret = btrfs_update_inode(trans, root, inode);
6115 }
6116 btrfs_end_transaction(trans);
6117 if (BTRFS_I(inode)->delayed_node)
6118 btrfs_balance_delayed_items(fs_info);
6119
6120 return ret;
6121 }
6122
6123 /*
6124 * This is a copy of file_update_time. We need this so we can return error on
6125 * ENOSPC for updating the inode in the case of file write and mmap writes.
6126 */
6127 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6128 int flags)
6129 {
6130 struct btrfs_root *root = BTRFS_I(inode)->root;
6131 bool dirty = flags & ~S_VERSION;
6132
6133 if (btrfs_root_readonly(root))
6134 return -EROFS;
6135
6136 if (flags & S_VERSION)
6137 dirty |= inode_maybe_inc_iversion(inode, dirty);
6138 if (flags & S_CTIME)
6139 inode->i_ctime = *now;
6140 if (flags & S_MTIME)
6141 inode->i_mtime = *now;
6142 if (flags & S_ATIME)
6143 inode->i_atime = *now;
6144 return dirty ? btrfs_dirty_inode(inode) : 0;
6145 }
6146
6147 /*
6148 * find the highest existing sequence number in a directory
6149 * and then set the in-memory index_cnt variable to reflect
6150 * free sequence numbers
6151 */
6152 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6153 {
6154 struct btrfs_root *root = inode->root;
6155 struct btrfs_key key, found_key;
6156 struct btrfs_path *path;
6157 struct extent_buffer *leaf;
6158 int ret;
6159
6160 key.objectid = btrfs_ino(inode);
6161 key.type = BTRFS_DIR_INDEX_KEY;
6162 key.offset = (u64)-1;
6163
6164 path = btrfs_alloc_path();
6165 if (!path)
6166 return -ENOMEM;
6167
6168 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6169 if (ret < 0)
6170 goto out;
6171 /* FIXME: we should be able to handle this */
6172 if (ret == 0)
6173 goto out;
6174 ret = 0;
6175
6176 /*
6177 * MAGIC NUMBER EXPLANATION:
6178 * since we search a directory based on f_pos we have to start at 2
6179 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6180 * else has to start at 2
6181 */
6182 if (path->slots[0] == 0) {
6183 inode->index_cnt = 2;
6184 goto out;
6185 }
6186
6187 path->slots[0]--;
6188
6189 leaf = path->nodes[0];
6190 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6191
6192 if (found_key.objectid != btrfs_ino(inode) ||
6193 found_key.type != BTRFS_DIR_INDEX_KEY) {
6194 inode->index_cnt = 2;
6195 goto out;
6196 }
6197
6198 inode->index_cnt = found_key.offset + 1;
6199 out:
6200 btrfs_free_path(path);
6201 return ret;
6202 }
6203
6204 /*
6205 * helper to find a free sequence number in a given directory. This current
6206 * code is very simple, later versions will do smarter things in the btree
6207 */
6208 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6209 {
6210 int ret = 0;
6211
6212 if (dir->index_cnt == (u64)-1) {
6213 ret = btrfs_inode_delayed_dir_index_count(dir);
6214 if (ret) {
6215 ret = btrfs_set_inode_index_count(dir);
6216 if (ret)
6217 return ret;
6218 }
6219 }
6220
6221 *index = dir->index_cnt;
6222 dir->index_cnt++;
6223
6224 return ret;
6225 }
6226
6227 static int btrfs_insert_inode_locked(struct inode *inode)
6228 {
6229 struct btrfs_iget_args args;
6230 args.location = &BTRFS_I(inode)->location;
6231 args.root = BTRFS_I(inode)->root;
6232
6233 return insert_inode_locked4(inode,
6234 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6235 btrfs_find_actor, &args);
6236 }
6237
6238 /*
6239 * Inherit flags from the parent inode.
6240 *
6241 * Currently only the compression flags and the cow flags are inherited.
6242 */
6243 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6244 {
6245 unsigned int flags;
6246
6247 if (!dir)
6248 return;
6249
6250 flags = BTRFS_I(dir)->flags;
6251
6252 if (flags & BTRFS_INODE_NOCOMPRESS) {
6253 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6254 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6255 } else if (flags & BTRFS_INODE_COMPRESS) {
6256 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6257 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6258 }
6259
6260 if (flags & BTRFS_INODE_NODATACOW) {
6261 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6262 if (S_ISREG(inode->i_mode))
6263 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6264 }
6265
6266 btrfs_update_iflags(inode);
6267 }
6268
6269 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6270 struct btrfs_root *root,
6271 struct inode *dir,
6272 const char *name, int name_len,
6273 u64 ref_objectid, u64 objectid,
6274 umode_t mode, u64 *index)
6275 {
6276 struct btrfs_fs_info *fs_info = root->fs_info;
6277 struct inode *inode;
6278 struct btrfs_inode_item *inode_item;
6279 struct btrfs_key *location;
6280 struct btrfs_path *path;
6281 struct btrfs_inode_ref *ref;
6282 struct btrfs_key key[2];
6283 u32 sizes[2];
6284 int nitems = name ? 2 : 1;
6285 unsigned long ptr;
6286 int ret;
6287
6288 path = btrfs_alloc_path();
6289 if (!path)
6290 return ERR_PTR(-ENOMEM);
6291
6292 inode = new_inode(fs_info->sb);
6293 if (!inode) {
6294 btrfs_free_path(path);
6295 return ERR_PTR(-ENOMEM);
6296 }
6297
6298 /*
6299 * O_TMPFILE, set link count to 0, so that after this point,
6300 * we fill in an inode item with the correct link count.
6301 */
6302 if (!name)
6303 set_nlink(inode, 0);
6304
6305 /*
6306 * we have to initialize this early, so we can reclaim the inode
6307 * number if we fail afterwards in this function.
6308 */
6309 inode->i_ino = objectid;
6310
6311 if (dir && name) {
6312 trace_btrfs_inode_request(dir);
6313
6314 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6315 if (ret) {
6316 btrfs_free_path(path);
6317 iput(inode);
6318 return ERR_PTR(ret);
6319 }
6320 } else if (dir) {
6321 *index = 0;
6322 }
6323 /*
6324 * index_cnt is ignored for everything but a dir,
6325 * btrfs_set_inode_index_count has an explanation for the magic
6326 * number
6327 */
6328 BTRFS_I(inode)->index_cnt = 2;
6329 BTRFS_I(inode)->dir_index = *index;
6330 BTRFS_I(inode)->root = root;
6331 BTRFS_I(inode)->generation = trans->transid;
6332 inode->i_generation = BTRFS_I(inode)->generation;
6333
6334 /*
6335 * We could have gotten an inode number from somebody who was fsynced
6336 * and then removed in this same transaction, so let's just set full
6337 * sync since it will be a full sync anyway and this will blow away the
6338 * old info in the log.
6339 */
6340 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6341
6342 key[0].objectid = objectid;
6343 key[0].type = BTRFS_INODE_ITEM_KEY;
6344 key[0].offset = 0;
6345
6346 sizes[0] = sizeof(struct btrfs_inode_item);
6347
6348 if (name) {
6349 /*
6350 * Start new inodes with an inode_ref. This is slightly more
6351 * efficient for small numbers of hard links since they will
6352 * be packed into one item. Extended refs will kick in if we
6353 * add more hard links than can fit in the ref item.
6354 */
6355 key[1].objectid = objectid;
6356 key[1].type = BTRFS_INODE_REF_KEY;
6357 key[1].offset = ref_objectid;
6358
6359 sizes[1] = name_len + sizeof(*ref);
6360 }
6361
6362 location = &BTRFS_I(inode)->location;
6363 location->objectid = objectid;
6364 location->offset = 0;
6365 location->type = BTRFS_INODE_ITEM_KEY;
6366
6367 ret = btrfs_insert_inode_locked(inode);
6368 if (ret < 0)
6369 goto fail;
6370
6371 path->leave_spinning = 1;
6372 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6373 if (ret != 0)
6374 goto fail_unlock;
6375
6376 inode_init_owner(inode, dir, mode);
6377 inode_set_bytes(inode, 0);
6378
6379 inode->i_mtime = current_time(inode);
6380 inode->i_atime = inode->i_mtime;
6381 inode->i_ctime = inode->i_mtime;
6382 BTRFS_I(inode)->i_otime = inode->i_mtime;
6383
6384 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6385 struct btrfs_inode_item);
6386 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6387 sizeof(*inode_item));
6388 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6389
6390 if (name) {
6391 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6392 struct btrfs_inode_ref);
6393 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6394 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6395 ptr = (unsigned long)(ref + 1);
6396 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6397 }
6398
6399 btrfs_mark_buffer_dirty(path->nodes[0]);
6400 btrfs_free_path(path);
6401
6402 btrfs_inherit_iflags(inode, dir);
6403
6404 if (S_ISREG(mode)) {
6405 if (btrfs_test_opt(fs_info, NODATASUM))
6406 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6407 if (btrfs_test_opt(fs_info, NODATACOW))
6408 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6409 BTRFS_INODE_NODATASUM;
6410 }
6411
6412 inode_tree_add(inode);
6413
6414 trace_btrfs_inode_new(inode);
6415 btrfs_set_inode_last_trans(trans, inode);
6416
6417 btrfs_update_root_times(trans, root);
6418
6419 ret = btrfs_inode_inherit_props(trans, inode, dir);
6420 if (ret)
6421 btrfs_err(fs_info,
6422 "error inheriting props for ino %llu (root %llu): %d",
6423 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6424
6425 return inode;
6426
6427 fail_unlock:
6428 unlock_new_inode(inode);
6429 fail:
6430 if (dir && name)
6431 BTRFS_I(dir)->index_cnt--;
6432 btrfs_free_path(path);
6433 iput(inode);
6434 return ERR_PTR(ret);
6435 }
6436
6437 static inline u8 btrfs_inode_type(struct inode *inode)
6438 {
6439 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6440 }
6441
6442 /*
6443 * utility function to add 'inode' into 'parent_inode' with
6444 * a give name and a given sequence number.
6445 * if 'add_backref' is true, also insert a backref from the
6446 * inode to the parent directory.
6447 */
6448 int btrfs_add_link(struct btrfs_trans_handle *trans,
6449 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6450 const char *name, int name_len, int add_backref, u64 index)
6451 {
6452 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6453 int ret = 0;
6454 struct btrfs_key key;
6455 struct btrfs_root *root = parent_inode->root;
6456 u64 ino = btrfs_ino(inode);
6457 u64 parent_ino = btrfs_ino(parent_inode);
6458
6459 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6460 memcpy(&key, &inode->root->root_key, sizeof(key));
6461 } else {
6462 key.objectid = ino;
6463 key.type = BTRFS_INODE_ITEM_KEY;
6464 key.offset = 0;
6465 }
6466
6467 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6468 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6469 root->root_key.objectid, parent_ino,
6470 index, name, name_len);
6471 } else if (add_backref) {
6472 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6473 parent_ino, index);
6474 }
6475
6476 /* Nothing to clean up yet */
6477 if (ret)
6478 return ret;
6479
6480 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6481 parent_inode, &key,
6482 btrfs_inode_type(&inode->vfs_inode), index);
6483 if (ret == -EEXIST || ret == -EOVERFLOW)
6484 goto fail_dir_item;
6485 else if (ret) {
6486 btrfs_abort_transaction(trans, ret);
6487 return ret;
6488 }
6489
6490 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6491 name_len * 2);
6492 inode_inc_iversion(&parent_inode->vfs_inode);
6493 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6494 current_time(&parent_inode->vfs_inode);
6495 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6496 if (ret)
6497 btrfs_abort_transaction(trans, ret);
6498 return ret;
6499
6500 fail_dir_item:
6501 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6502 u64 local_index;
6503 int err;
6504 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6505 root->root_key.objectid, parent_ino,
6506 &local_index, name, name_len);
6507
6508 } else if (add_backref) {
6509 u64 local_index;
6510 int err;
6511
6512 err = btrfs_del_inode_ref(trans, root, name, name_len,
6513 ino, parent_ino, &local_index);
6514 }
6515 return ret;
6516 }
6517
6518 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6519 struct btrfs_inode *dir, struct dentry *dentry,
6520 struct btrfs_inode *inode, int backref, u64 index)
6521 {
6522 int err = btrfs_add_link(trans, dir, inode,
6523 dentry->d_name.name, dentry->d_name.len,
6524 backref, index);
6525 if (err > 0)
6526 err = -EEXIST;
6527 return err;
6528 }
6529
6530 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6531 umode_t mode, dev_t rdev)
6532 {
6533 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6534 struct btrfs_trans_handle *trans;
6535 struct btrfs_root *root = BTRFS_I(dir)->root;
6536 struct inode *inode = NULL;
6537 int err;
6538 int drop_inode = 0;
6539 u64 objectid;
6540 u64 index = 0;
6541
6542 /*
6543 * 2 for inode item and ref
6544 * 2 for dir items
6545 * 1 for xattr if selinux is on
6546 */
6547 trans = btrfs_start_transaction(root, 5);
6548 if (IS_ERR(trans))
6549 return PTR_ERR(trans);
6550
6551 err = btrfs_find_free_ino(root, &objectid);
6552 if (err)
6553 goto out_unlock;
6554
6555 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6556 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6557 mode, &index);
6558 if (IS_ERR(inode)) {
6559 err = PTR_ERR(inode);
6560 goto out_unlock;
6561 }
6562
6563 /*
6564 * If the active LSM wants to access the inode during
6565 * d_instantiate it needs these. Smack checks to see
6566 * if the filesystem supports xattrs by looking at the
6567 * ops vector.
6568 */
6569 inode->i_op = &btrfs_special_inode_operations;
6570 init_special_inode(inode, inode->i_mode, rdev);
6571
6572 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6573 if (err)
6574 goto out_unlock_inode;
6575
6576 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6577 0, index);
6578 if (err) {
6579 goto out_unlock_inode;
6580 } else {
6581 btrfs_update_inode(trans, root, inode);
6582 unlock_new_inode(inode);
6583 d_instantiate(dentry, inode);
6584 }
6585
6586 out_unlock:
6587 btrfs_end_transaction(trans);
6588 btrfs_btree_balance_dirty(fs_info);
6589 if (drop_inode) {
6590 inode_dec_link_count(inode);
6591 iput(inode);
6592 }
6593 return err;
6594
6595 out_unlock_inode:
6596 drop_inode = 1;
6597 unlock_new_inode(inode);
6598 goto out_unlock;
6599
6600 }
6601
6602 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6603 umode_t mode, bool excl)
6604 {
6605 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6606 struct btrfs_trans_handle *trans;
6607 struct btrfs_root *root = BTRFS_I(dir)->root;
6608 struct inode *inode = NULL;
6609 int drop_inode_on_err = 0;
6610 int err;
6611 u64 objectid;
6612 u64 index = 0;
6613
6614 /*
6615 * 2 for inode item and ref
6616 * 2 for dir items
6617 * 1 for xattr if selinux is on
6618 */
6619 trans = btrfs_start_transaction(root, 5);
6620 if (IS_ERR(trans))
6621 return PTR_ERR(trans);
6622
6623 err = btrfs_find_free_ino(root, &objectid);
6624 if (err)
6625 goto out_unlock;
6626
6627 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6628 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6629 mode, &index);
6630 if (IS_ERR(inode)) {
6631 err = PTR_ERR(inode);
6632 goto out_unlock;
6633 }
6634 drop_inode_on_err = 1;
6635 /*
6636 * If the active LSM wants to access the inode during
6637 * d_instantiate it needs these. Smack checks to see
6638 * if the filesystem supports xattrs by looking at the
6639 * ops vector.
6640 */
6641 inode->i_fop = &btrfs_file_operations;
6642 inode->i_op = &btrfs_file_inode_operations;
6643 inode->i_mapping->a_ops = &btrfs_aops;
6644
6645 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6646 if (err)
6647 goto out_unlock_inode;
6648
6649 err = btrfs_update_inode(trans, root, inode);
6650 if (err)
6651 goto out_unlock_inode;
6652
6653 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6654 0, index);
6655 if (err)
6656 goto out_unlock_inode;
6657
6658 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6659 unlock_new_inode(inode);
6660 d_instantiate(dentry, inode);
6661
6662 out_unlock:
6663 btrfs_end_transaction(trans);
6664 if (err && drop_inode_on_err) {
6665 inode_dec_link_count(inode);
6666 iput(inode);
6667 }
6668 btrfs_btree_balance_dirty(fs_info);
6669 return err;
6670
6671 out_unlock_inode:
6672 unlock_new_inode(inode);
6673 goto out_unlock;
6674
6675 }
6676
6677 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6678 struct dentry *dentry)
6679 {
6680 struct btrfs_trans_handle *trans = NULL;
6681 struct btrfs_root *root = BTRFS_I(dir)->root;
6682 struct inode *inode = d_inode(old_dentry);
6683 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6684 u64 index;
6685 int err;
6686 int drop_inode = 0;
6687
6688 /* do not allow sys_link's with other subvols of the same device */
6689 if (root->objectid != BTRFS_I(inode)->root->objectid)
6690 return -EXDEV;
6691
6692 if (inode->i_nlink >= BTRFS_LINK_MAX)
6693 return -EMLINK;
6694
6695 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6696 if (err)
6697 goto fail;
6698
6699 /*
6700 * 2 items for inode and inode ref
6701 * 2 items for dir items
6702 * 1 item for parent inode
6703 */
6704 trans = btrfs_start_transaction(root, 5);
6705 if (IS_ERR(trans)) {
6706 err = PTR_ERR(trans);
6707 trans = NULL;
6708 goto fail;
6709 }
6710
6711 /* There are several dir indexes for this inode, clear the cache. */
6712 BTRFS_I(inode)->dir_index = 0ULL;
6713 inc_nlink(inode);
6714 inode_inc_iversion(inode);
6715 inode->i_ctime = current_time(inode);
6716 ihold(inode);
6717 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6718
6719 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6720 1, index);
6721
6722 if (err) {
6723 drop_inode = 1;
6724 } else {
6725 struct dentry *parent = dentry->d_parent;
6726 err = btrfs_update_inode(trans, root, inode);
6727 if (err)
6728 goto fail;
6729 if (inode->i_nlink == 1) {
6730 /*
6731 * If new hard link count is 1, it's a file created
6732 * with open(2) O_TMPFILE flag.
6733 */
6734 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6735 if (err)
6736 goto fail;
6737 }
6738 d_instantiate(dentry, inode);
6739 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6740 }
6741
6742 fail:
6743 if (trans)
6744 btrfs_end_transaction(trans);
6745 if (drop_inode) {
6746 inode_dec_link_count(inode);
6747 iput(inode);
6748 }
6749 btrfs_btree_balance_dirty(fs_info);
6750 return err;
6751 }
6752
6753 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6754 {
6755 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6756 struct inode *inode = NULL;
6757 struct btrfs_trans_handle *trans;
6758 struct btrfs_root *root = BTRFS_I(dir)->root;
6759 int err = 0;
6760 int drop_on_err = 0;
6761 u64 objectid = 0;
6762 u64 index = 0;
6763
6764 /*
6765 * 2 items for inode and ref
6766 * 2 items for dir items
6767 * 1 for xattr if selinux is on
6768 */
6769 trans = btrfs_start_transaction(root, 5);
6770 if (IS_ERR(trans))
6771 return PTR_ERR(trans);
6772
6773 err = btrfs_find_free_ino(root, &objectid);
6774 if (err)
6775 goto out_fail;
6776
6777 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6778 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6779 S_IFDIR | mode, &index);
6780 if (IS_ERR(inode)) {
6781 err = PTR_ERR(inode);
6782 goto out_fail;
6783 }
6784
6785 drop_on_err = 1;
6786 /* these must be set before we unlock the inode */
6787 inode->i_op = &btrfs_dir_inode_operations;
6788 inode->i_fop = &btrfs_dir_file_operations;
6789
6790 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6791 if (err)
6792 goto out_fail_inode;
6793
6794 btrfs_i_size_write(BTRFS_I(inode), 0);
6795 err = btrfs_update_inode(trans, root, inode);
6796 if (err)
6797 goto out_fail_inode;
6798
6799 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6800 dentry->d_name.name,
6801 dentry->d_name.len, 0, index);
6802 if (err)
6803 goto out_fail_inode;
6804
6805 d_instantiate(dentry, inode);
6806 /*
6807 * mkdir is special. We're unlocking after we call d_instantiate
6808 * to avoid a race with nfsd calling d_instantiate.
6809 */
6810 unlock_new_inode(inode);
6811 drop_on_err = 0;
6812
6813 out_fail:
6814 btrfs_end_transaction(trans);
6815 if (drop_on_err) {
6816 inode_dec_link_count(inode);
6817 iput(inode);
6818 }
6819 btrfs_btree_balance_dirty(fs_info);
6820 return err;
6821
6822 out_fail_inode:
6823 unlock_new_inode(inode);
6824 goto out_fail;
6825 }
6826
6827 static noinline int uncompress_inline(struct btrfs_path *path,
6828 struct page *page,
6829 size_t pg_offset, u64 extent_offset,
6830 struct btrfs_file_extent_item *item)
6831 {
6832 int ret;
6833 struct extent_buffer *leaf = path->nodes[0];
6834 char *tmp;
6835 size_t max_size;
6836 unsigned long inline_size;
6837 unsigned long ptr;
6838 int compress_type;
6839
6840 WARN_ON(pg_offset != 0);
6841 compress_type = btrfs_file_extent_compression(leaf, item);
6842 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6843 inline_size = btrfs_file_extent_inline_item_len(leaf,
6844 btrfs_item_nr(path->slots[0]));
6845 tmp = kmalloc(inline_size, GFP_NOFS);
6846 if (!tmp)
6847 return -ENOMEM;
6848 ptr = btrfs_file_extent_inline_start(item);
6849
6850 read_extent_buffer(leaf, tmp, ptr, inline_size);
6851
6852 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6853 ret = btrfs_decompress(compress_type, tmp, page,
6854 extent_offset, inline_size, max_size);
6855
6856 /*
6857 * decompression code contains a memset to fill in any space between the end
6858 * of the uncompressed data and the end of max_size in case the decompressed
6859 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6860 * the end of an inline extent and the beginning of the next block, so we
6861 * cover that region here.
6862 */
6863
6864 if (max_size + pg_offset < PAGE_SIZE) {
6865 char *map = kmap(page);
6866 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6867 kunmap(page);
6868 }
6869 kfree(tmp);
6870 return ret;
6871 }
6872
6873 /*
6874 * a bit scary, this does extent mapping from logical file offset to the disk.
6875 * the ugly parts come from merging extents from the disk with the in-ram
6876 * representation. This gets more complex because of the data=ordered code,
6877 * where the in-ram extents might be locked pending data=ordered completion.
6878 *
6879 * This also copies inline extents directly into the page.
6880 */
6881 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6882 struct page *page,
6883 size_t pg_offset, u64 start, u64 len,
6884 int create)
6885 {
6886 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6887 int ret;
6888 int err = 0;
6889 u64 extent_start = 0;
6890 u64 extent_end = 0;
6891 u64 objectid = btrfs_ino(inode);
6892 u32 found_type;
6893 struct btrfs_path *path = NULL;
6894 struct btrfs_root *root = inode->root;
6895 struct btrfs_file_extent_item *item;
6896 struct extent_buffer *leaf;
6897 struct btrfs_key found_key;
6898 struct extent_map *em = NULL;
6899 struct extent_map_tree *em_tree = &inode->extent_tree;
6900 struct extent_io_tree *io_tree = &inode->io_tree;
6901 const bool new_inline = !page || create;
6902
6903 read_lock(&em_tree->lock);
6904 em = lookup_extent_mapping(em_tree, start, len);
6905 if (em)
6906 em->bdev = fs_info->fs_devices->latest_bdev;
6907 read_unlock(&em_tree->lock);
6908
6909 if (em) {
6910 if (em->start > start || em->start + em->len <= start)
6911 free_extent_map(em);
6912 else if (em->block_start == EXTENT_MAP_INLINE && page)
6913 free_extent_map(em);
6914 else
6915 goto out;
6916 }
6917 em = alloc_extent_map();
6918 if (!em) {
6919 err = -ENOMEM;
6920 goto out;
6921 }
6922 em->bdev = fs_info->fs_devices->latest_bdev;
6923 em->start = EXTENT_MAP_HOLE;
6924 em->orig_start = EXTENT_MAP_HOLE;
6925 em->len = (u64)-1;
6926 em->block_len = (u64)-1;
6927
6928 if (!path) {
6929 path = btrfs_alloc_path();
6930 if (!path) {
6931 err = -ENOMEM;
6932 goto out;
6933 }
6934 /*
6935 * Chances are we'll be called again, so go ahead and do
6936 * readahead
6937 */
6938 path->reada = READA_FORWARD;
6939 }
6940
6941 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6942 if (ret < 0) {
6943 err = ret;
6944 goto out;
6945 }
6946
6947 if (ret != 0) {
6948 if (path->slots[0] == 0)
6949 goto not_found;
6950 path->slots[0]--;
6951 }
6952
6953 leaf = path->nodes[0];
6954 item = btrfs_item_ptr(leaf, path->slots[0],
6955 struct btrfs_file_extent_item);
6956 /* are we inside the extent that was found? */
6957 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6958 found_type = found_key.type;
6959 if (found_key.objectid != objectid ||
6960 found_type != BTRFS_EXTENT_DATA_KEY) {
6961 /*
6962 * If we backup past the first extent we want to move forward
6963 * and see if there is an extent in front of us, otherwise we'll
6964 * say there is a hole for our whole search range which can
6965 * cause problems.
6966 */
6967 extent_end = start;
6968 goto next;
6969 }
6970
6971 found_type = btrfs_file_extent_type(leaf, item);
6972 extent_start = found_key.offset;
6973 if (found_type == BTRFS_FILE_EXTENT_REG ||
6974 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6975 extent_end = extent_start +
6976 btrfs_file_extent_num_bytes(leaf, item);
6977
6978 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6979 extent_start);
6980 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6981 size_t size;
6982 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6983 extent_end = ALIGN(extent_start + size,
6984 fs_info->sectorsize);
6985
6986 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6987 path->slots[0],
6988 extent_start);
6989 }
6990 next:
6991 if (start >= extent_end) {
6992 path->slots[0]++;
6993 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6994 ret = btrfs_next_leaf(root, path);
6995 if (ret < 0) {
6996 err = ret;
6997 goto out;
6998 }
6999 if (ret > 0)
7000 goto not_found;
7001 leaf = path->nodes[0];
7002 }
7003 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7004 if (found_key.objectid != objectid ||
7005 found_key.type != BTRFS_EXTENT_DATA_KEY)
7006 goto not_found;
7007 if (start + len <= found_key.offset)
7008 goto not_found;
7009 if (start > found_key.offset)
7010 goto next;
7011 em->start = start;
7012 em->orig_start = start;
7013 em->len = found_key.offset - start;
7014 goto not_found_em;
7015 }
7016
7017 btrfs_extent_item_to_extent_map(inode, path, item,
7018 new_inline, em);
7019
7020 if (found_type == BTRFS_FILE_EXTENT_REG ||
7021 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7022 goto insert;
7023 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7024 unsigned long ptr;
7025 char *map;
7026 size_t size;
7027 size_t extent_offset;
7028 size_t copy_size;
7029
7030 if (new_inline)
7031 goto out;
7032
7033 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7034 extent_offset = page_offset(page) + pg_offset - extent_start;
7035 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7036 size - extent_offset);
7037 em->start = extent_start + extent_offset;
7038 em->len = ALIGN(copy_size, fs_info->sectorsize);
7039 em->orig_block_len = em->len;
7040 em->orig_start = em->start;
7041 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7042 if (!PageUptodate(page)) {
7043 if (btrfs_file_extent_compression(leaf, item) !=
7044 BTRFS_COMPRESS_NONE) {
7045 ret = uncompress_inline(path, page, pg_offset,
7046 extent_offset, item);
7047 if (ret) {
7048 err = ret;
7049 goto out;
7050 }
7051 } else {
7052 map = kmap(page);
7053 read_extent_buffer(leaf, map + pg_offset, ptr,
7054 copy_size);
7055 if (pg_offset + copy_size < PAGE_SIZE) {
7056 memset(map + pg_offset + copy_size, 0,
7057 PAGE_SIZE - pg_offset -
7058 copy_size);
7059 }
7060 kunmap(page);
7061 }
7062 flush_dcache_page(page);
7063 }
7064 set_extent_uptodate(io_tree, em->start,
7065 extent_map_end(em) - 1, NULL, GFP_NOFS);
7066 goto insert;
7067 }
7068 not_found:
7069 em->start = start;
7070 em->orig_start = start;
7071 em->len = len;
7072 not_found_em:
7073 em->block_start = EXTENT_MAP_HOLE;
7074 insert:
7075 btrfs_release_path(path);
7076 if (em->start > start || extent_map_end(em) <= start) {
7077 btrfs_err(fs_info,
7078 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7079 em->start, em->len, start, len);
7080 err = -EIO;
7081 goto out;
7082 }
7083
7084 err = 0;
7085 write_lock(&em_tree->lock);
7086 err = btrfs_add_extent_mapping(em_tree, &em, start, len);
7087 write_unlock(&em_tree->lock);
7088 out:
7089
7090 trace_btrfs_get_extent(root, inode, em);
7091
7092 btrfs_free_path(path);
7093 if (err) {
7094 free_extent_map(em);
7095 return ERR_PTR(err);
7096 }
7097 BUG_ON(!em); /* Error is always set */
7098 return em;
7099 }
7100
7101 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7102 struct page *page,
7103 size_t pg_offset, u64 start, u64 len,
7104 int create)
7105 {
7106 struct extent_map *em;
7107 struct extent_map *hole_em = NULL;
7108 u64 range_start = start;
7109 u64 end;
7110 u64 found;
7111 u64 found_end;
7112 int err = 0;
7113
7114 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7115 if (IS_ERR(em))
7116 return em;
7117 /*
7118 * If our em maps to:
7119 * - a hole or
7120 * - a pre-alloc extent,
7121 * there might actually be delalloc bytes behind it.
7122 */
7123 if (em->block_start != EXTENT_MAP_HOLE &&
7124 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7125 return em;
7126 else
7127 hole_em = em;
7128
7129 /* check to see if we've wrapped (len == -1 or similar) */
7130 end = start + len;
7131 if (end < start)
7132 end = (u64)-1;
7133 else
7134 end -= 1;
7135
7136 em = NULL;
7137
7138 /* ok, we didn't find anything, lets look for delalloc */
7139 found = count_range_bits(&inode->io_tree, &range_start,
7140 end, len, EXTENT_DELALLOC, 1);
7141 found_end = range_start + found;
7142 if (found_end < range_start)
7143 found_end = (u64)-1;
7144
7145 /*
7146 * we didn't find anything useful, return
7147 * the original results from get_extent()
7148 */
7149 if (range_start > end || found_end <= start) {
7150 em = hole_em;
7151 hole_em = NULL;
7152 goto out;
7153 }
7154
7155 /* adjust the range_start to make sure it doesn't
7156 * go backwards from the start they passed in
7157 */
7158 range_start = max(start, range_start);
7159 found = found_end - range_start;
7160
7161 if (found > 0) {
7162 u64 hole_start = start;
7163 u64 hole_len = len;
7164
7165 em = alloc_extent_map();
7166 if (!em) {
7167 err = -ENOMEM;
7168 goto out;
7169 }
7170 /*
7171 * when btrfs_get_extent can't find anything it
7172 * returns one huge hole
7173 *
7174 * make sure what it found really fits our range, and
7175 * adjust to make sure it is based on the start from
7176 * the caller
7177 */
7178 if (hole_em) {
7179 u64 calc_end = extent_map_end(hole_em);
7180
7181 if (calc_end <= start || (hole_em->start > end)) {
7182 free_extent_map(hole_em);
7183 hole_em = NULL;
7184 } else {
7185 hole_start = max(hole_em->start, start);
7186 hole_len = calc_end - hole_start;
7187 }
7188 }
7189 em->bdev = NULL;
7190 if (hole_em && range_start > hole_start) {
7191 /* our hole starts before our delalloc, so we
7192 * have to return just the parts of the hole
7193 * that go until the delalloc starts
7194 */
7195 em->len = min(hole_len,
7196 range_start - hole_start);
7197 em->start = hole_start;
7198 em->orig_start = hole_start;
7199 /*
7200 * don't adjust block start at all,
7201 * it is fixed at EXTENT_MAP_HOLE
7202 */
7203 em->block_start = hole_em->block_start;
7204 em->block_len = hole_len;
7205 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7206 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7207 } else {
7208 em->start = range_start;
7209 em->len = found;
7210 em->orig_start = range_start;
7211 em->block_start = EXTENT_MAP_DELALLOC;
7212 em->block_len = found;
7213 }
7214 } else {
7215 return hole_em;
7216 }
7217 out:
7218
7219 free_extent_map(hole_em);
7220 if (err) {
7221 free_extent_map(em);
7222 return ERR_PTR(err);
7223 }
7224 return em;
7225 }
7226
7227 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7228 const u64 start,
7229 const u64 len,
7230 const u64 orig_start,
7231 const u64 block_start,
7232 const u64 block_len,
7233 const u64 orig_block_len,
7234 const u64 ram_bytes,
7235 const int type)
7236 {
7237 struct extent_map *em = NULL;
7238 int ret;
7239
7240 if (type != BTRFS_ORDERED_NOCOW) {
7241 em = create_io_em(inode, start, len, orig_start,
7242 block_start, block_len, orig_block_len,
7243 ram_bytes,
7244 BTRFS_COMPRESS_NONE, /* compress_type */
7245 type);
7246 if (IS_ERR(em))
7247 goto out;
7248 }
7249 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7250 len, block_len, type);
7251 if (ret) {
7252 if (em) {
7253 free_extent_map(em);
7254 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7255 start + len - 1, 0);
7256 }
7257 em = ERR_PTR(ret);
7258 }
7259 out:
7260
7261 return em;
7262 }
7263
7264 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7265 u64 start, u64 len)
7266 {
7267 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7268 struct btrfs_root *root = BTRFS_I(inode)->root;
7269 struct extent_map *em;
7270 struct btrfs_key ins;
7271 u64 alloc_hint;
7272 int ret;
7273
7274 alloc_hint = get_extent_allocation_hint(inode, start, len);
7275 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7276 0, alloc_hint, &ins, 1, 1);
7277 if (ret)
7278 return ERR_PTR(ret);
7279
7280 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7281 ins.objectid, ins.offset, ins.offset,
7282 ins.offset, BTRFS_ORDERED_REGULAR);
7283 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7284 if (IS_ERR(em))
7285 btrfs_free_reserved_extent(fs_info, ins.objectid,
7286 ins.offset, 1);
7287
7288 return em;
7289 }
7290
7291 /*
7292 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7293 * block must be cow'd
7294 */
7295 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7296 u64 *orig_start, u64 *orig_block_len,
7297 u64 *ram_bytes)
7298 {
7299 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7300 struct btrfs_path *path;
7301 int ret;
7302 struct extent_buffer *leaf;
7303 struct btrfs_root *root = BTRFS_I(inode)->root;
7304 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7305 struct btrfs_file_extent_item *fi;
7306 struct btrfs_key key;
7307 u64 disk_bytenr;
7308 u64 backref_offset;
7309 u64 extent_end;
7310 u64 num_bytes;
7311 int slot;
7312 int found_type;
7313 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7314
7315 path = btrfs_alloc_path();
7316 if (!path)
7317 return -ENOMEM;
7318
7319 ret = btrfs_lookup_file_extent(NULL, root, path,
7320 btrfs_ino(BTRFS_I(inode)), offset, 0);
7321 if (ret < 0)
7322 goto out;
7323
7324 slot = path->slots[0];
7325 if (ret == 1) {
7326 if (slot == 0) {
7327 /* can't find the item, must cow */
7328 ret = 0;
7329 goto out;
7330 }
7331 slot--;
7332 }
7333 ret = 0;
7334 leaf = path->nodes[0];
7335 btrfs_item_key_to_cpu(leaf, &key, slot);
7336 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7337 key.type != BTRFS_EXTENT_DATA_KEY) {
7338 /* not our file or wrong item type, must cow */
7339 goto out;
7340 }
7341
7342 if (key.offset > offset) {
7343 /* Wrong offset, must cow */
7344 goto out;
7345 }
7346
7347 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7348 found_type = btrfs_file_extent_type(leaf, fi);
7349 if (found_type != BTRFS_FILE_EXTENT_REG &&
7350 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7351 /* not a regular extent, must cow */
7352 goto out;
7353 }
7354
7355 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7356 goto out;
7357
7358 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7359 if (extent_end <= offset)
7360 goto out;
7361
7362 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7363 if (disk_bytenr == 0)
7364 goto out;
7365
7366 if (btrfs_file_extent_compression(leaf, fi) ||
7367 btrfs_file_extent_encryption(leaf, fi) ||
7368 btrfs_file_extent_other_encoding(leaf, fi))
7369 goto out;
7370
7371 backref_offset = btrfs_file_extent_offset(leaf, fi);
7372
7373 if (orig_start) {
7374 *orig_start = key.offset - backref_offset;
7375 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7376 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7377 }
7378
7379 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7380 goto out;
7381
7382 num_bytes = min(offset + *len, extent_end) - offset;
7383 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7384 u64 range_end;
7385
7386 range_end = round_up(offset + num_bytes,
7387 root->fs_info->sectorsize) - 1;
7388 ret = test_range_bit(io_tree, offset, range_end,
7389 EXTENT_DELALLOC, 0, NULL);
7390 if (ret) {
7391 ret = -EAGAIN;
7392 goto out;
7393 }
7394 }
7395
7396 btrfs_release_path(path);
7397
7398 /*
7399 * look for other files referencing this extent, if we
7400 * find any we must cow
7401 */
7402
7403 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7404 key.offset - backref_offset, disk_bytenr);
7405 if (ret) {
7406 ret = 0;
7407 goto out;
7408 }
7409
7410 /*
7411 * adjust disk_bytenr and num_bytes to cover just the bytes
7412 * in this extent we are about to write. If there
7413 * are any csums in that range we have to cow in order
7414 * to keep the csums correct
7415 */
7416 disk_bytenr += backref_offset;
7417 disk_bytenr += offset - key.offset;
7418 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7419 goto out;
7420 /*
7421 * all of the above have passed, it is safe to overwrite this extent
7422 * without cow
7423 */
7424 *len = num_bytes;
7425 ret = 1;
7426 out:
7427 btrfs_free_path(path);
7428 return ret;
7429 }
7430
7431 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7432 struct extent_state **cached_state, int writing)
7433 {
7434 struct btrfs_ordered_extent *ordered;
7435 int ret = 0;
7436
7437 while (1) {
7438 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7439 cached_state);
7440 /*
7441 * We're concerned with the entire range that we're going to be
7442 * doing DIO to, so we need to make sure there's no ordered
7443 * extents in this range.
7444 */
7445 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7446 lockend - lockstart + 1);
7447
7448 /*
7449 * We need to make sure there are no buffered pages in this
7450 * range either, we could have raced between the invalidate in
7451 * generic_file_direct_write and locking the extent. The
7452 * invalidate needs to happen so that reads after a write do not
7453 * get stale data.
7454 */
7455 if (!ordered &&
7456 (!writing || !filemap_range_has_page(inode->i_mapping,
7457 lockstart, lockend)))
7458 break;
7459
7460 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7461 cached_state);
7462
7463 if (ordered) {
7464 /*
7465 * If we are doing a DIO read and the ordered extent we
7466 * found is for a buffered write, we can not wait for it
7467 * to complete and retry, because if we do so we can
7468 * deadlock with concurrent buffered writes on page
7469 * locks. This happens only if our DIO read covers more
7470 * than one extent map, if at this point has already
7471 * created an ordered extent for a previous extent map
7472 * and locked its range in the inode's io tree, and a
7473 * concurrent write against that previous extent map's
7474 * range and this range started (we unlock the ranges
7475 * in the io tree only when the bios complete and
7476 * buffered writes always lock pages before attempting
7477 * to lock range in the io tree).
7478 */
7479 if (writing ||
7480 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7481 btrfs_start_ordered_extent(inode, ordered, 1);
7482 else
7483 ret = -ENOTBLK;
7484 btrfs_put_ordered_extent(ordered);
7485 } else {
7486 /*
7487 * We could trigger writeback for this range (and wait
7488 * for it to complete) and then invalidate the pages for
7489 * this range (through invalidate_inode_pages2_range()),
7490 * but that can lead us to a deadlock with a concurrent
7491 * call to readpages() (a buffered read or a defrag call
7492 * triggered a readahead) on a page lock due to an
7493 * ordered dio extent we created before but did not have
7494 * yet a corresponding bio submitted (whence it can not
7495 * complete), which makes readpages() wait for that
7496 * ordered extent to complete while holding a lock on
7497 * that page.
7498 */
7499 ret = -ENOTBLK;
7500 }
7501
7502 if (ret)
7503 break;
7504
7505 cond_resched();
7506 }
7507
7508 return ret;
7509 }
7510
7511 /* The callers of this must take lock_extent() */
7512 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7513 u64 orig_start, u64 block_start,
7514 u64 block_len, u64 orig_block_len,
7515 u64 ram_bytes, int compress_type,
7516 int type)
7517 {
7518 struct extent_map_tree *em_tree;
7519 struct extent_map *em;
7520 struct btrfs_root *root = BTRFS_I(inode)->root;
7521 int ret;
7522
7523 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7524 type == BTRFS_ORDERED_COMPRESSED ||
7525 type == BTRFS_ORDERED_NOCOW ||
7526 type == BTRFS_ORDERED_REGULAR);
7527
7528 em_tree = &BTRFS_I(inode)->extent_tree;
7529 em = alloc_extent_map();
7530 if (!em)
7531 return ERR_PTR(-ENOMEM);
7532
7533 em->start = start;
7534 em->orig_start = orig_start;
7535 em->len = len;
7536 em->block_len = block_len;
7537 em->block_start = block_start;
7538 em->bdev = root->fs_info->fs_devices->latest_bdev;
7539 em->orig_block_len = orig_block_len;
7540 em->ram_bytes = ram_bytes;
7541 em->generation = -1;
7542 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7543 if (type == BTRFS_ORDERED_PREALLOC) {
7544 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7545 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7546 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7547 em->compress_type = compress_type;
7548 }
7549
7550 do {
7551 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7552 em->start + em->len - 1, 0);
7553 write_lock(&em_tree->lock);
7554 ret = add_extent_mapping(em_tree, em, 1);
7555 write_unlock(&em_tree->lock);
7556 /*
7557 * The caller has taken lock_extent(), who could race with us
7558 * to add em?
7559 */
7560 } while (ret == -EEXIST);
7561
7562 if (ret) {
7563 free_extent_map(em);
7564 return ERR_PTR(ret);
7565 }
7566
7567 /* em got 2 refs now, callers needs to do free_extent_map once. */
7568 return em;
7569 }
7570
7571 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7572 struct buffer_head *bh_result, int create)
7573 {
7574 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7575 struct extent_map *em;
7576 struct extent_state *cached_state = NULL;
7577 struct btrfs_dio_data *dio_data = NULL;
7578 u64 start = iblock << inode->i_blkbits;
7579 u64 lockstart, lockend;
7580 u64 len = bh_result->b_size;
7581 int unlock_bits = EXTENT_LOCKED;
7582 int ret = 0;
7583
7584 if (create)
7585 unlock_bits |= EXTENT_DIRTY;
7586 else
7587 len = min_t(u64, len, fs_info->sectorsize);
7588
7589 lockstart = start;
7590 lockend = start + len - 1;
7591
7592 if (current->journal_info) {
7593 /*
7594 * Need to pull our outstanding extents and set journal_info to NULL so
7595 * that anything that needs to check if there's a transaction doesn't get
7596 * confused.
7597 */
7598 dio_data = current->journal_info;
7599 current->journal_info = NULL;
7600 }
7601
7602 /*
7603 * If this errors out it's because we couldn't invalidate pagecache for
7604 * this range and we need to fallback to buffered.
7605 */
7606 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7607 create)) {
7608 ret = -ENOTBLK;
7609 goto err;
7610 }
7611
7612 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7613 if (IS_ERR(em)) {
7614 ret = PTR_ERR(em);
7615 goto unlock_err;
7616 }
7617
7618 /*
7619 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7620 * io. INLINE is special, and we could probably kludge it in here, but
7621 * it's still buffered so for safety lets just fall back to the generic
7622 * buffered path.
7623 *
7624 * For COMPRESSED we _have_ to read the entire extent in so we can
7625 * decompress it, so there will be buffering required no matter what we
7626 * do, so go ahead and fallback to buffered.
7627 *
7628 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7629 * to buffered IO. Don't blame me, this is the price we pay for using
7630 * the generic code.
7631 */
7632 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7633 em->block_start == EXTENT_MAP_INLINE) {
7634 free_extent_map(em);
7635 ret = -ENOTBLK;
7636 goto unlock_err;
7637 }
7638
7639 /* Just a good old fashioned hole, return */
7640 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7641 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7642 free_extent_map(em);
7643 goto unlock_err;
7644 }
7645
7646 /*
7647 * We don't allocate a new extent in the following cases
7648 *
7649 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7650 * existing extent.
7651 * 2) The extent is marked as PREALLOC. We're good to go here and can
7652 * just use the extent.
7653 *
7654 */
7655 if (!create) {
7656 len = min(len, em->len - (start - em->start));
7657 lockstart = start + len;
7658 goto unlock;
7659 }
7660
7661 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7662 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7663 em->block_start != EXTENT_MAP_HOLE)) {
7664 int type;
7665 u64 block_start, orig_start, orig_block_len, ram_bytes;
7666
7667 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7668 type = BTRFS_ORDERED_PREALLOC;
7669 else
7670 type = BTRFS_ORDERED_NOCOW;
7671 len = min(len, em->len - (start - em->start));
7672 block_start = em->block_start + (start - em->start);
7673
7674 if (can_nocow_extent(inode, start, &len, &orig_start,
7675 &orig_block_len, &ram_bytes) == 1 &&
7676 btrfs_inc_nocow_writers(fs_info, block_start)) {
7677 struct extent_map *em2;
7678
7679 em2 = btrfs_create_dio_extent(inode, start, len,
7680 orig_start, block_start,
7681 len, orig_block_len,
7682 ram_bytes, type);
7683 btrfs_dec_nocow_writers(fs_info, block_start);
7684 if (type == BTRFS_ORDERED_PREALLOC) {
7685 free_extent_map(em);
7686 em = em2;
7687 }
7688 if (em2 && IS_ERR(em2)) {
7689 ret = PTR_ERR(em2);
7690 goto unlock_err;
7691 }
7692 /*
7693 * For inode marked NODATACOW or extent marked PREALLOC,
7694 * use the existing or preallocated extent, so does not
7695 * need to adjust btrfs_space_info's bytes_may_use.
7696 */
7697 btrfs_free_reserved_data_space_noquota(inode,
7698 start, len);
7699 goto unlock;
7700 }
7701 }
7702
7703 /*
7704 * this will cow the extent, reset the len in case we changed
7705 * it above
7706 */
7707 len = bh_result->b_size;
7708 free_extent_map(em);
7709 em = btrfs_new_extent_direct(inode, start, len);
7710 if (IS_ERR(em)) {
7711 ret = PTR_ERR(em);
7712 goto unlock_err;
7713 }
7714 len = min(len, em->len - (start - em->start));
7715 unlock:
7716 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7717 inode->i_blkbits;
7718 bh_result->b_size = len;
7719 bh_result->b_bdev = em->bdev;
7720 set_buffer_mapped(bh_result);
7721 if (create) {
7722 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7723 set_buffer_new(bh_result);
7724
7725 /*
7726 * Need to update the i_size under the extent lock so buffered
7727 * readers will get the updated i_size when we unlock.
7728 */
7729 if (!dio_data->overwrite && start + len > i_size_read(inode))
7730 i_size_write(inode, start + len);
7731
7732 WARN_ON(dio_data->reserve < len);
7733 dio_data->reserve -= len;
7734 dio_data->unsubmitted_oe_range_end = start + len;
7735 current->journal_info = dio_data;
7736 }
7737
7738 /*
7739 * In the case of write we need to clear and unlock the entire range,
7740 * in the case of read we need to unlock only the end area that we
7741 * aren't using if there is any left over space.
7742 */
7743 if (lockstart < lockend) {
7744 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7745 lockend, unlock_bits, 1, 0,
7746 &cached_state);
7747 } else {
7748 free_extent_state(cached_state);
7749 }
7750
7751 free_extent_map(em);
7752
7753 return 0;
7754
7755 unlock_err:
7756 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7757 unlock_bits, 1, 0, &cached_state);
7758 err:
7759 if (dio_data)
7760 current->journal_info = dio_data;
7761 return ret;
7762 }
7763
7764 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7765 struct bio *bio,
7766 int mirror_num)
7767 {
7768 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7769 blk_status_t ret;
7770
7771 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7772
7773 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7774 if (ret)
7775 return ret;
7776
7777 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7778
7779 return ret;
7780 }
7781
7782 static int btrfs_check_dio_repairable(struct inode *inode,
7783 struct bio *failed_bio,
7784 struct io_failure_record *failrec,
7785 int failed_mirror)
7786 {
7787 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7788 int num_copies;
7789
7790 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7791 if (num_copies == 1) {
7792 /*
7793 * we only have a single copy of the data, so don't bother with
7794 * all the retry and error correction code that follows. no
7795 * matter what the error is, it is very likely to persist.
7796 */
7797 btrfs_debug(fs_info,
7798 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7799 num_copies, failrec->this_mirror, failed_mirror);
7800 return 0;
7801 }
7802
7803 failrec->failed_mirror = failed_mirror;
7804 failrec->this_mirror++;
7805 if (failrec->this_mirror == failed_mirror)
7806 failrec->this_mirror++;
7807
7808 if (failrec->this_mirror > num_copies) {
7809 btrfs_debug(fs_info,
7810 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7811 num_copies, failrec->this_mirror, failed_mirror);
7812 return 0;
7813 }
7814
7815 return 1;
7816 }
7817
7818 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7819 struct page *page, unsigned int pgoff,
7820 u64 start, u64 end, int failed_mirror,
7821 bio_end_io_t *repair_endio, void *repair_arg)
7822 {
7823 struct io_failure_record *failrec;
7824 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7825 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7826 struct bio *bio;
7827 int isector;
7828 unsigned int read_mode = 0;
7829 int segs;
7830 int ret;
7831 blk_status_t status;
7832 struct bio_vec bvec;
7833
7834 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7835
7836 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7837 if (ret)
7838 return errno_to_blk_status(ret);
7839
7840 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7841 failed_mirror);
7842 if (!ret) {
7843 free_io_failure(failure_tree, io_tree, failrec);
7844 return BLK_STS_IOERR;
7845 }
7846
7847 segs = bio_segments(failed_bio);
7848 bio_get_first_bvec(failed_bio, &bvec);
7849 if (segs > 1 ||
7850 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7851 read_mode |= REQ_FAILFAST_DEV;
7852
7853 isector = start - btrfs_io_bio(failed_bio)->logical;
7854 isector >>= inode->i_sb->s_blocksize_bits;
7855 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7856 pgoff, isector, repair_endio, repair_arg);
7857 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7858
7859 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7860 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7861 read_mode, failrec->this_mirror, failrec->in_validation);
7862
7863 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7864 if (status) {
7865 free_io_failure(failure_tree, io_tree, failrec);
7866 bio_put(bio);
7867 }
7868
7869 return status;
7870 }
7871
7872 struct btrfs_retry_complete {
7873 struct completion done;
7874 struct inode *inode;
7875 u64 start;
7876 int uptodate;
7877 };
7878
7879 static void btrfs_retry_endio_nocsum(struct bio *bio)
7880 {
7881 struct btrfs_retry_complete *done = bio->bi_private;
7882 struct inode *inode = done->inode;
7883 struct bio_vec *bvec;
7884 struct extent_io_tree *io_tree, *failure_tree;
7885 int i;
7886
7887 if (bio->bi_status)
7888 goto end;
7889
7890 ASSERT(bio->bi_vcnt == 1);
7891 io_tree = &BTRFS_I(inode)->io_tree;
7892 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7893 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7894
7895 done->uptodate = 1;
7896 ASSERT(!bio_flagged(bio, BIO_CLONED));
7897 bio_for_each_segment_all(bvec, bio, i)
7898 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7899 io_tree, done->start, bvec->bv_page,
7900 btrfs_ino(BTRFS_I(inode)), 0);
7901 end:
7902 complete(&done->done);
7903 bio_put(bio);
7904 }
7905
7906 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7907 struct btrfs_io_bio *io_bio)
7908 {
7909 struct btrfs_fs_info *fs_info;
7910 struct bio_vec bvec;
7911 struct bvec_iter iter;
7912 struct btrfs_retry_complete done;
7913 u64 start;
7914 unsigned int pgoff;
7915 u32 sectorsize;
7916 int nr_sectors;
7917 blk_status_t ret;
7918 blk_status_t err = BLK_STS_OK;
7919
7920 fs_info = BTRFS_I(inode)->root->fs_info;
7921 sectorsize = fs_info->sectorsize;
7922
7923 start = io_bio->logical;
7924 done.inode = inode;
7925 io_bio->bio.bi_iter = io_bio->iter;
7926
7927 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7928 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7929 pgoff = bvec.bv_offset;
7930
7931 next_block_or_try_again:
7932 done.uptodate = 0;
7933 done.start = start;
7934 init_completion(&done.done);
7935
7936 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7937 pgoff, start, start + sectorsize - 1,
7938 io_bio->mirror_num,
7939 btrfs_retry_endio_nocsum, &done);
7940 if (ret) {
7941 err = ret;
7942 goto next;
7943 }
7944
7945 wait_for_completion_io(&done.done);
7946
7947 if (!done.uptodate) {
7948 /* We might have another mirror, so try again */
7949 goto next_block_or_try_again;
7950 }
7951
7952 next:
7953 start += sectorsize;
7954
7955 nr_sectors--;
7956 if (nr_sectors) {
7957 pgoff += sectorsize;
7958 ASSERT(pgoff < PAGE_SIZE);
7959 goto next_block_or_try_again;
7960 }
7961 }
7962
7963 return err;
7964 }
7965
7966 static void btrfs_retry_endio(struct bio *bio)
7967 {
7968 struct btrfs_retry_complete *done = bio->bi_private;
7969 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7970 struct extent_io_tree *io_tree, *failure_tree;
7971 struct inode *inode = done->inode;
7972 struct bio_vec *bvec;
7973 int uptodate;
7974 int ret;
7975 int i;
7976
7977 if (bio->bi_status)
7978 goto end;
7979
7980 uptodate = 1;
7981
7982 ASSERT(bio->bi_vcnt == 1);
7983 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7984
7985 io_tree = &BTRFS_I(inode)->io_tree;
7986 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7987
7988 ASSERT(!bio_flagged(bio, BIO_CLONED));
7989 bio_for_each_segment_all(bvec, bio, i) {
7990 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7991 bvec->bv_offset, done->start,
7992 bvec->bv_len);
7993 if (!ret)
7994 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7995 failure_tree, io_tree, done->start,
7996 bvec->bv_page,
7997 btrfs_ino(BTRFS_I(inode)),
7998 bvec->bv_offset);
7999 else
8000 uptodate = 0;
8001 }
8002
8003 done->uptodate = uptodate;
8004 end:
8005 complete(&done->done);
8006 bio_put(bio);
8007 }
8008
8009 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8010 struct btrfs_io_bio *io_bio, blk_status_t err)
8011 {
8012 struct btrfs_fs_info *fs_info;
8013 struct bio_vec bvec;
8014 struct bvec_iter iter;
8015 struct btrfs_retry_complete done;
8016 u64 start;
8017 u64 offset = 0;
8018 u32 sectorsize;
8019 int nr_sectors;
8020 unsigned int pgoff;
8021 int csum_pos;
8022 bool uptodate = (err == 0);
8023 int ret;
8024 blk_status_t status;
8025
8026 fs_info = BTRFS_I(inode)->root->fs_info;
8027 sectorsize = fs_info->sectorsize;
8028
8029 err = BLK_STS_OK;
8030 start = io_bio->logical;
8031 done.inode = inode;
8032 io_bio->bio.bi_iter = io_bio->iter;
8033
8034 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8035 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8036
8037 pgoff = bvec.bv_offset;
8038 next_block:
8039 if (uptodate) {
8040 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8041 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8042 bvec.bv_page, pgoff, start, sectorsize);
8043 if (likely(!ret))
8044 goto next;
8045 }
8046 try_again:
8047 done.uptodate = 0;
8048 done.start = start;
8049 init_completion(&done.done);
8050
8051 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8052 pgoff, start, start + sectorsize - 1,
8053 io_bio->mirror_num, btrfs_retry_endio,
8054 &done);
8055 if (status) {
8056 err = status;
8057 goto next;
8058 }
8059
8060 wait_for_completion_io(&done.done);
8061
8062 if (!done.uptodate) {
8063 /* We might have another mirror, so try again */
8064 goto try_again;
8065 }
8066 next:
8067 offset += sectorsize;
8068 start += sectorsize;
8069
8070 ASSERT(nr_sectors);
8071
8072 nr_sectors--;
8073 if (nr_sectors) {
8074 pgoff += sectorsize;
8075 ASSERT(pgoff < PAGE_SIZE);
8076 goto next_block;
8077 }
8078 }
8079
8080 return err;
8081 }
8082
8083 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8084 struct btrfs_io_bio *io_bio, blk_status_t err)
8085 {
8086 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8087
8088 if (skip_csum) {
8089 if (unlikely(err))
8090 return __btrfs_correct_data_nocsum(inode, io_bio);
8091 else
8092 return BLK_STS_OK;
8093 } else {
8094 return __btrfs_subio_endio_read(inode, io_bio, err);
8095 }
8096 }
8097
8098 static void btrfs_endio_direct_read(struct bio *bio)
8099 {
8100 struct btrfs_dio_private *dip = bio->bi_private;
8101 struct inode *inode = dip->inode;
8102 struct bio *dio_bio;
8103 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8104 blk_status_t err = bio->bi_status;
8105
8106 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8107 err = btrfs_subio_endio_read(inode, io_bio, err);
8108
8109 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8110 dip->logical_offset + dip->bytes - 1);
8111 dio_bio = dip->dio_bio;
8112
8113 kfree(dip);
8114
8115 dio_bio->bi_status = err;
8116 dio_end_io(dio_bio);
8117
8118 if (io_bio->end_io)
8119 io_bio->end_io(io_bio, blk_status_to_errno(err));
8120 bio_put(bio);
8121 }
8122
8123 static void __endio_write_update_ordered(struct inode *inode,
8124 const u64 offset, const u64 bytes,
8125 const bool uptodate)
8126 {
8127 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8128 struct btrfs_ordered_extent *ordered = NULL;
8129 struct btrfs_workqueue *wq;
8130 btrfs_work_func_t func;
8131 u64 ordered_offset = offset;
8132 u64 ordered_bytes = bytes;
8133 u64 last_offset;
8134 int ret;
8135
8136 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8137 wq = fs_info->endio_freespace_worker;
8138 func = btrfs_freespace_write_helper;
8139 } else {
8140 wq = fs_info->endio_write_workers;
8141 func = btrfs_endio_write_helper;
8142 }
8143
8144 again:
8145 last_offset = ordered_offset;
8146 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8147 &ordered_offset,
8148 ordered_bytes,
8149 uptodate);
8150 if (!ret)
8151 goto out_test;
8152
8153 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8154 btrfs_queue_work(wq, &ordered->work);
8155 out_test:
8156 /*
8157 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8158 * in the range, we can exit.
8159 */
8160 if (ordered_offset == last_offset)
8161 return;
8162 /*
8163 * our bio might span multiple ordered extents. If we haven't
8164 * completed the accounting for the whole dio, go back and try again
8165 */
8166 if (ordered_offset < offset + bytes) {
8167 ordered_bytes = offset + bytes - ordered_offset;
8168 ordered = NULL;
8169 goto again;
8170 }
8171 }
8172
8173 static void btrfs_endio_direct_write(struct bio *bio)
8174 {
8175 struct btrfs_dio_private *dip = bio->bi_private;
8176 struct bio *dio_bio = dip->dio_bio;
8177
8178 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8179 dip->bytes, !bio->bi_status);
8180
8181 kfree(dip);
8182
8183 dio_bio->bi_status = bio->bi_status;
8184 dio_end_io(dio_bio);
8185 bio_put(bio);
8186 }
8187
8188 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8189 struct bio *bio, u64 offset)
8190 {
8191 struct inode *inode = private_data;
8192 blk_status_t ret;
8193 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8194 BUG_ON(ret); /* -ENOMEM */
8195 return 0;
8196 }
8197
8198 static void btrfs_end_dio_bio(struct bio *bio)
8199 {
8200 struct btrfs_dio_private *dip = bio->bi_private;
8201 blk_status_t err = bio->bi_status;
8202
8203 if (err)
8204 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8205 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8206 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8207 bio->bi_opf,
8208 (unsigned long long)bio->bi_iter.bi_sector,
8209 bio->bi_iter.bi_size, err);
8210
8211 if (dip->subio_endio)
8212 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8213
8214 if (err) {
8215 /*
8216 * We want to perceive the errors flag being set before
8217 * decrementing the reference count. We don't need a barrier
8218 * since atomic operations with a return value are fully
8219 * ordered as per atomic_t.txt
8220 */
8221 dip->errors = 1;
8222 }
8223
8224 /* if there are more bios still pending for this dio, just exit */
8225 if (!atomic_dec_and_test(&dip->pending_bios))
8226 goto out;
8227
8228 if (dip->errors) {
8229 bio_io_error(dip->orig_bio);
8230 } else {
8231 dip->dio_bio->bi_status = BLK_STS_OK;
8232 bio_endio(dip->orig_bio);
8233 }
8234 out:
8235 bio_put(bio);
8236 }
8237
8238 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8239 struct btrfs_dio_private *dip,
8240 struct bio *bio,
8241 u64 file_offset)
8242 {
8243 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8244 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8245 blk_status_t ret;
8246
8247 /*
8248 * We load all the csum data we need when we submit
8249 * the first bio to reduce the csum tree search and
8250 * contention.
8251 */
8252 if (dip->logical_offset == file_offset) {
8253 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8254 file_offset);
8255 if (ret)
8256 return ret;
8257 }
8258
8259 if (bio == dip->orig_bio)
8260 return 0;
8261
8262 file_offset -= dip->logical_offset;
8263 file_offset >>= inode->i_sb->s_blocksize_bits;
8264 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8265
8266 return 0;
8267 }
8268
8269 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8270 struct inode *inode, u64 file_offset, int async_submit)
8271 {
8272 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8273 struct btrfs_dio_private *dip = bio->bi_private;
8274 bool write = bio_op(bio) == REQ_OP_WRITE;
8275 blk_status_t ret;
8276
8277 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8278 if (async_submit)
8279 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8280
8281 if (!write) {
8282 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8283 if (ret)
8284 goto err;
8285 }
8286
8287 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8288 goto map;
8289
8290 if (write && async_submit) {
8291 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8292 file_offset, inode,
8293 btrfs_submit_bio_start_direct_io,
8294 btrfs_submit_bio_done);
8295 goto err;
8296 } else if (write) {
8297 /*
8298 * If we aren't doing async submit, calculate the csum of the
8299 * bio now.
8300 */
8301 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8302 if (ret)
8303 goto err;
8304 } else {
8305 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8306 file_offset);
8307 if (ret)
8308 goto err;
8309 }
8310 map:
8311 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8312 err:
8313 return ret;
8314 }
8315
8316 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8317 {
8318 struct inode *inode = dip->inode;
8319 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8320 struct bio *bio;
8321 struct bio *orig_bio = dip->orig_bio;
8322 u64 start_sector = orig_bio->bi_iter.bi_sector;
8323 u64 file_offset = dip->logical_offset;
8324 u64 map_length;
8325 int async_submit = 0;
8326 u64 submit_len;
8327 int clone_offset = 0;
8328 int clone_len;
8329 int ret;
8330 blk_status_t status;
8331
8332 map_length = orig_bio->bi_iter.bi_size;
8333 submit_len = map_length;
8334 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8335 &map_length, NULL, 0);
8336 if (ret)
8337 return -EIO;
8338
8339 if (map_length >= submit_len) {
8340 bio = orig_bio;
8341 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8342 goto submit;
8343 }
8344
8345 /* async crcs make it difficult to collect full stripe writes. */
8346 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8347 async_submit = 0;
8348 else
8349 async_submit = 1;
8350
8351 /* bio split */
8352 ASSERT(map_length <= INT_MAX);
8353 atomic_inc(&dip->pending_bios);
8354 do {
8355 clone_len = min_t(int, submit_len, map_length);
8356
8357 /*
8358 * This will never fail as it's passing GPF_NOFS and
8359 * the allocation is backed by btrfs_bioset.
8360 */
8361 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8362 clone_len);
8363 bio->bi_private = dip;
8364 bio->bi_end_io = btrfs_end_dio_bio;
8365 btrfs_io_bio(bio)->logical = file_offset;
8366
8367 ASSERT(submit_len >= clone_len);
8368 submit_len -= clone_len;
8369 if (submit_len == 0)
8370 break;
8371
8372 /*
8373 * Increase the count before we submit the bio so we know
8374 * the end IO handler won't happen before we increase the
8375 * count. Otherwise, the dip might get freed before we're
8376 * done setting it up.
8377 */
8378 atomic_inc(&dip->pending_bios);
8379
8380 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8381 async_submit);
8382 if (status) {
8383 bio_put(bio);
8384 atomic_dec(&dip->pending_bios);
8385 goto out_err;
8386 }
8387
8388 clone_offset += clone_len;
8389 start_sector += clone_len >> 9;
8390 file_offset += clone_len;
8391
8392 map_length = submit_len;
8393 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8394 start_sector << 9, &map_length, NULL, 0);
8395 if (ret)
8396 goto out_err;
8397 } while (submit_len > 0);
8398
8399 submit:
8400 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8401 if (!status)
8402 return 0;
8403
8404 bio_put(bio);
8405 out_err:
8406 dip->errors = 1;
8407 /*
8408 * Before atomic variable goto zero, we must make sure dip->errors is
8409 * perceived to be set. This ordering is ensured by the fact that an
8410 * atomic operations with a return value are fully ordered as per
8411 * atomic_t.txt
8412 */
8413 if (atomic_dec_and_test(&dip->pending_bios))
8414 bio_io_error(dip->orig_bio);
8415
8416 /* bio_end_io() will handle error, so we needn't return it */
8417 return 0;
8418 }
8419
8420 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8421 loff_t file_offset)
8422 {
8423 struct btrfs_dio_private *dip = NULL;
8424 struct bio *bio = NULL;
8425 struct btrfs_io_bio *io_bio;
8426 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8427 int ret = 0;
8428
8429 bio = btrfs_bio_clone(dio_bio);
8430
8431 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8432 if (!dip) {
8433 ret = -ENOMEM;
8434 goto free_ordered;
8435 }
8436
8437 dip->private = dio_bio->bi_private;
8438 dip->inode = inode;
8439 dip->logical_offset = file_offset;
8440 dip->bytes = dio_bio->bi_iter.bi_size;
8441 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8442 bio->bi_private = dip;
8443 dip->orig_bio = bio;
8444 dip->dio_bio = dio_bio;
8445 atomic_set(&dip->pending_bios, 0);
8446 io_bio = btrfs_io_bio(bio);
8447 io_bio->logical = file_offset;
8448
8449 if (write) {
8450 bio->bi_end_io = btrfs_endio_direct_write;
8451 } else {
8452 bio->bi_end_io = btrfs_endio_direct_read;
8453 dip->subio_endio = btrfs_subio_endio_read;
8454 }
8455
8456 /*
8457 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8458 * even if we fail to submit a bio, because in such case we do the
8459 * corresponding error handling below and it must not be done a second
8460 * time by btrfs_direct_IO().
8461 */
8462 if (write) {
8463 struct btrfs_dio_data *dio_data = current->journal_info;
8464
8465 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8466 dip->bytes;
8467 dio_data->unsubmitted_oe_range_start =
8468 dio_data->unsubmitted_oe_range_end;
8469 }
8470
8471 ret = btrfs_submit_direct_hook(dip);
8472 if (!ret)
8473 return;
8474
8475 if (io_bio->end_io)
8476 io_bio->end_io(io_bio, ret);
8477
8478 free_ordered:
8479 /*
8480 * If we arrived here it means either we failed to submit the dip
8481 * or we either failed to clone the dio_bio or failed to allocate the
8482 * dip. If we cloned the dio_bio and allocated the dip, we can just
8483 * call bio_endio against our io_bio so that we get proper resource
8484 * cleanup if we fail to submit the dip, otherwise, we must do the
8485 * same as btrfs_endio_direct_[write|read] because we can't call these
8486 * callbacks - they require an allocated dip and a clone of dio_bio.
8487 */
8488 if (bio && dip) {
8489 bio_io_error(bio);
8490 /*
8491 * The end io callbacks free our dip, do the final put on bio
8492 * and all the cleanup and final put for dio_bio (through
8493 * dio_end_io()).
8494 */
8495 dip = NULL;
8496 bio = NULL;
8497 } else {
8498 if (write)
8499 __endio_write_update_ordered(inode,
8500 file_offset,
8501 dio_bio->bi_iter.bi_size,
8502 false);
8503 else
8504 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8505 file_offset + dio_bio->bi_iter.bi_size - 1);
8506
8507 dio_bio->bi_status = BLK_STS_IOERR;
8508 /*
8509 * Releases and cleans up our dio_bio, no need to bio_put()
8510 * nor bio_endio()/bio_io_error() against dio_bio.
8511 */
8512 dio_end_io(dio_bio);
8513 }
8514 if (bio)
8515 bio_put(bio);
8516 kfree(dip);
8517 }
8518
8519 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8520 const struct iov_iter *iter, loff_t offset)
8521 {
8522 int seg;
8523 int i;
8524 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8525 ssize_t retval = -EINVAL;
8526
8527 if (offset & blocksize_mask)
8528 goto out;
8529
8530 if (iov_iter_alignment(iter) & blocksize_mask)
8531 goto out;
8532
8533 /* If this is a write we don't need to check anymore */
8534 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8535 return 0;
8536 /*
8537 * Check to make sure we don't have duplicate iov_base's in this
8538 * iovec, if so return EINVAL, otherwise we'll get csum errors
8539 * when reading back.
8540 */
8541 for (seg = 0; seg < iter->nr_segs; seg++) {
8542 for (i = seg + 1; i < iter->nr_segs; i++) {
8543 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8544 goto out;
8545 }
8546 }
8547 retval = 0;
8548 out:
8549 return retval;
8550 }
8551
8552 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8553 {
8554 struct file *file = iocb->ki_filp;
8555 struct inode *inode = file->f_mapping->host;
8556 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8557 struct btrfs_dio_data dio_data = { 0 };
8558 struct extent_changeset *data_reserved = NULL;
8559 loff_t offset = iocb->ki_pos;
8560 size_t count = 0;
8561 int flags = 0;
8562 bool wakeup = true;
8563 bool relock = false;
8564 ssize_t ret;
8565
8566 if (check_direct_IO(fs_info, iter, offset))
8567 return 0;
8568
8569 inode_dio_begin(inode);
8570
8571 /*
8572 * The generic stuff only does filemap_write_and_wait_range, which
8573 * isn't enough if we've written compressed pages to this area, so
8574 * we need to flush the dirty pages again to make absolutely sure
8575 * that any outstanding dirty pages are on disk.
8576 */
8577 count = iov_iter_count(iter);
8578 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8579 &BTRFS_I(inode)->runtime_flags))
8580 filemap_fdatawrite_range(inode->i_mapping, offset,
8581 offset + count - 1);
8582
8583 if (iov_iter_rw(iter) == WRITE) {
8584 /*
8585 * If the write DIO is beyond the EOF, we need update
8586 * the isize, but it is protected by i_mutex. So we can
8587 * not unlock the i_mutex at this case.
8588 */
8589 if (offset + count <= inode->i_size) {
8590 dio_data.overwrite = 1;
8591 inode_unlock(inode);
8592 relock = true;
8593 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8594 ret = -EAGAIN;
8595 goto out;
8596 }
8597 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8598 offset, count);
8599 if (ret)
8600 goto out;
8601
8602 /*
8603 * We need to know how many extents we reserved so that we can
8604 * do the accounting properly if we go over the number we
8605 * originally calculated. Abuse current->journal_info for this.
8606 */
8607 dio_data.reserve = round_up(count,
8608 fs_info->sectorsize);
8609 dio_data.unsubmitted_oe_range_start = (u64)offset;
8610 dio_data.unsubmitted_oe_range_end = (u64)offset;
8611 current->journal_info = &dio_data;
8612 down_read(&BTRFS_I(inode)->dio_sem);
8613 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8614 &BTRFS_I(inode)->runtime_flags)) {
8615 inode_dio_end(inode);
8616 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8617 wakeup = false;
8618 }
8619
8620 ret = __blockdev_direct_IO(iocb, inode,
8621 fs_info->fs_devices->latest_bdev,
8622 iter, btrfs_get_blocks_direct, NULL,
8623 btrfs_submit_direct, flags);
8624 if (iov_iter_rw(iter) == WRITE) {
8625 up_read(&BTRFS_I(inode)->dio_sem);
8626 current->journal_info = NULL;
8627 if (ret < 0 && ret != -EIOCBQUEUED) {
8628 if (dio_data.reserve)
8629 btrfs_delalloc_release_space(inode, data_reserved,
8630 offset, dio_data.reserve, true);
8631 /*
8632 * On error we might have left some ordered extents
8633 * without submitting corresponding bios for them, so
8634 * cleanup them up to avoid other tasks getting them
8635 * and waiting for them to complete forever.
8636 */
8637 if (dio_data.unsubmitted_oe_range_start <
8638 dio_data.unsubmitted_oe_range_end)
8639 __endio_write_update_ordered(inode,
8640 dio_data.unsubmitted_oe_range_start,
8641 dio_data.unsubmitted_oe_range_end -
8642 dio_data.unsubmitted_oe_range_start,
8643 false);
8644 } else if (ret >= 0 && (size_t)ret < count)
8645 btrfs_delalloc_release_space(inode, data_reserved,
8646 offset, count - (size_t)ret, true);
8647 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8648 }
8649 out:
8650 if (wakeup)
8651 inode_dio_end(inode);
8652 if (relock)
8653 inode_lock(inode);
8654
8655 extent_changeset_free(data_reserved);
8656 return ret;
8657 }
8658
8659 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8660
8661 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8662 __u64 start, __u64 len)
8663 {
8664 int ret;
8665
8666 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8667 if (ret)
8668 return ret;
8669
8670 return extent_fiemap(inode, fieinfo, start, len);
8671 }
8672
8673 int btrfs_readpage(struct file *file, struct page *page)
8674 {
8675 struct extent_io_tree *tree;
8676 tree = &BTRFS_I(page->mapping->host)->io_tree;
8677 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8678 }
8679
8680 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8681 {
8682 struct inode *inode = page->mapping->host;
8683 int ret;
8684
8685 if (current->flags & PF_MEMALLOC) {
8686 redirty_page_for_writepage(wbc, page);
8687 unlock_page(page);
8688 return 0;
8689 }
8690
8691 /*
8692 * If we are under memory pressure we will call this directly from the
8693 * VM, we need to make sure we have the inode referenced for the ordered
8694 * extent. If not just return like we didn't do anything.
8695 */
8696 if (!igrab(inode)) {
8697 redirty_page_for_writepage(wbc, page);
8698 return AOP_WRITEPAGE_ACTIVATE;
8699 }
8700 ret = extent_write_full_page(page, wbc);
8701 btrfs_add_delayed_iput(inode);
8702 return ret;
8703 }
8704
8705 static int btrfs_writepages(struct address_space *mapping,
8706 struct writeback_control *wbc)
8707 {
8708 struct extent_io_tree *tree;
8709
8710 tree = &BTRFS_I(mapping->host)->io_tree;
8711 return extent_writepages(tree, mapping, wbc);
8712 }
8713
8714 static int
8715 btrfs_readpages(struct file *file, struct address_space *mapping,
8716 struct list_head *pages, unsigned nr_pages)
8717 {
8718 struct extent_io_tree *tree;
8719 tree = &BTRFS_I(mapping->host)->io_tree;
8720 return extent_readpages(tree, mapping, pages, nr_pages);
8721 }
8722 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8723 {
8724 struct extent_io_tree *tree;
8725 struct extent_map_tree *map;
8726 int ret;
8727
8728 tree = &BTRFS_I(page->mapping->host)->io_tree;
8729 map = &BTRFS_I(page->mapping->host)->extent_tree;
8730 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8731 if (ret == 1) {
8732 ClearPagePrivate(page);
8733 set_page_private(page, 0);
8734 put_page(page);
8735 }
8736 return ret;
8737 }
8738
8739 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8740 {
8741 if (PageWriteback(page) || PageDirty(page))
8742 return 0;
8743 return __btrfs_releasepage(page, gfp_flags);
8744 }
8745
8746 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8747 unsigned int length)
8748 {
8749 struct inode *inode = page->mapping->host;
8750 struct extent_io_tree *tree;
8751 struct btrfs_ordered_extent *ordered;
8752 struct extent_state *cached_state = NULL;
8753 u64 page_start = page_offset(page);
8754 u64 page_end = page_start + PAGE_SIZE - 1;
8755 u64 start;
8756 u64 end;
8757 int inode_evicting = inode->i_state & I_FREEING;
8758
8759 /*
8760 * we have the page locked, so new writeback can't start,
8761 * and the dirty bit won't be cleared while we are here.
8762 *
8763 * Wait for IO on this page so that we can safely clear
8764 * the PagePrivate2 bit and do ordered accounting
8765 */
8766 wait_on_page_writeback(page);
8767
8768 tree = &BTRFS_I(inode)->io_tree;
8769 if (offset) {
8770 btrfs_releasepage(page, GFP_NOFS);
8771 return;
8772 }
8773
8774 if (!inode_evicting)
8775 lock_extent_bits(tree, page_start, page_end, &cached_state);
8776 again:
8777 start = page_start;
8778 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8779 page_end - start + 1);
8780 if (ordered) {
8781 end = min(page_end, ordered->file_offset + ordered->len - 1);
8782 /*
8783 * IO on this page will never be started, so we need
8784 * to account for any ordered extents now
8785 */
8786 if (!inode_evicting)
8787 clear_extent_bit(tree, start, end,
8788 EXTENT_DIRTY | EXTENT_DELALLOC |
8789 EXTENT_DELALLOC_NEW |
8790 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8791 EXTENT_DEFRAG, 1, 0, &cached_state);
8792 /*
8793 * whoever cleared the private bit is responsible
8794 * for the finish_ordered_io
8795 */
8796 if (TestClearPagePrivate2(page)) {
8797 struct btrfs_ordered_inode_tree *tree;
8798 u64 new_len;
8799
8800 tree = &BTRFS_I(inode)->ordered_tree;
8801
8802 spin_lock_irq(&tree->lock);
8803 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8804 new_len = start - ordered->file_offset;
8805 if (new_len < ordered->truncated_len)
8806 ordered->truncated_len = new_len;
8807 spin_unlock_irq(&tree->lock);
8808
8809 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8810 start,
8811 end - start + 1, 1))
8812 btrfs_finish_ordered_io(ordered);
8813 }
8814 btrfs_put_ordered_extent(ordered);
8815 if (!inode_evicting) {
8816 cached_state = NULL;
8817 lock_extent_bits(tree, start, end,
8818 &cached_state);
8819 }
8820
8821 start = end + 1;
8822 if (start < page_end)
8823 goto again;
8824 }
8825
8826 /*
8827 * Qgroup reserved space handler
8828 * Page here will be either
8829 * 1) Already written to disk
8830 * In this case, its reserved space is released from data rsv map
8831 * and will be freed by delayed_ref handler finally.
8832 * So even we call qgroup_free_data(), it won't decrease reserved
8833 * space.
8834 * 2) Not written to disk
8835 * This means the reserved space should be freed here. However,
8836 * if a truncate invalidates the page (by clearing PageDirty)
8837 * and the page is accounted for while allocating extent
8838 * in btrfs_check_data_free_space() we let delayed_ref to
8839 * free the entire extent.
8840 */
8841 if (PageDirty(page))
8842 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8843 if (!inode_evicting) {
8844 clear_extent_bit(tree, page_start, page_end,
8845 EXTENT_LOCKED | EXTENT_DIRTY |
8846 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8847 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8848 &cached_state);
8849
8850 __btrfs_releasepage(page, GFP_NOFS);
8851 }
8852
8853 ClearPageChecked(page);
8854 if (PagePrivate(page)) {
8855 ClearPagePrivate(page);
8856 set_page_private(page, 0);
8857 put_page(page);
8858 }
8859 }
8860
8861 /*
8862 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8863 * called from a page fault handler when a page is first dirtied. Hence we must
8864 * be careful to check for EOF conditions here. We set the page up correctly
8865 * for a written page which means we get ENOSPC checking when writing into
8866 * holes and correct delalloc and unwritten extent mapping on filesystems that
8867 * support these features.
8868 *
8869 * We are not allowed to take the i_mutex here so we have to play games to
8870 * protect against truncate races as the page could now be beyond EOF. Because
8871 * vmtruncate() writes the inode size before removing pages, once we have the
8872 * page lock we can determine safely if the page is beyond EOF. If it is not
8873 * beyond EOF, then the page is guaranteed safe against truncation until we
8874 * unlock the page.
8875 */
8876 int btrfs_page_mkwrite(struct vm_fault *vmf)
8877 {
8878 struct page *page = vmf->page;
8879 struct inode *inode = file_inode(vmf->vma->vm_file);
8880 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8881 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8882 struct btrfs_ordered_extent *ordered;
8883 struct extent_state *cached_state = NULL;
8884 struct extent_changeset *data_reserved = NULL;
8885 char *kaddr;
8886 unsigned long zero_start;
8887 loff_t size;
8888 int ret;
8889 int reserved = 0;
8890 u64 reserved_space;
8891 u64 page_start;
8892 u64 page_end;
8893 u64 end;
8894
8895 reserved_space = PAGE_SIZE;
8896
8897 sb_start_pagefault(inode->i_sb);
8898 page_start = page_offset(page);
8899 page_end = page_start + PAGE_SIZE - 1;
8900 end = page_end;
8901
8902 /*
8903 * Reserving delalloc space after obtaining the page lock can lead to
8904 * deadlock. For example, if a dirty page is locked by this function
8905 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8906 * dirty page write out, then the btrfs_writepage() function could
8907 * end up waiting indefinitely to get a lock on the page currently
8908 * being processed by btrfs_page_mkwrite() function.
8909 */
8910 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8911 reserved_space);
8912 if (!ret) {
8913 ret = file_update_time(vmf->vma->vm_file);
8914 reserved = 1;
8915 }
8916 if (ret) {
8917 if (ret == -ENOMEM)
8918 ret = VM_FAULT_OOM;
8919 else /* -ENOSPC, -EIO, etc */
8920 ret = VM_FAULT_SIGBUS;
8921 if (reserved)
8922 goto out;
8923 goto out_noreserve;
8924 }
8925
8926 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8927 again:
8928 lock_page(page);
8929 size = i_size_read(inode);
8930
8931 if ((page->mapping != inode->i_mapping) ||
8932 (page_start >= size)) {
8933 /* page got truncated out from underneath us */
8934 goto out_unlock;
8935 }
8936 wait_on_page_writeback(page);
8937
8938 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8939 set_page_extent_mapped(page);
8940
8941 /*
8942 * we can't set the delalloc bits if there are pending ordered
8943 * extents. Drop our locks and wait for them to finish
8944 */
8945 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8946 PAGE_SIZE);
8947 if (ordered) {
8948 unlock_extent_cached(io_tree, page_start, page_end,
8949 &cached_state);
8950 unlock_page(page);
8951 btrfs_start_ordered_extent(inode, ordered, 1);
8952 btrfs_put_ordered_extent(ordered);
8953 goto again;
8954 }
8955
8956 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8957 reserved_space = round_up(size - page_start,
8958 fs_info->sectorsize);
8959 if (reserved_space < PAGE_SIZE) {
8960 end = page_start + reserved_space - 1;
8961 btrfs_delalloc_release_space(inode, data_reserved,
8962 page_start, PAGE_SIZE - reserved_space,
8963 true);
8964 }
8965 }
8966
8967 /*
8968 * page_mkwrite gets called when the page is firstly dirtied after it's
8969 * faulted in, but write(2) could also dirty a page and set delalloc
8970 * bits, thus in this case for space account reason, we still need to
8971 * clear any delalloc bits within this page range since we have to
8972 * reserve data&meta space before lock_page() (see above comments).
8973 */
8974 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8975 EXTENT_DIRTY | EXTENT_DELALLOC |
8976 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8977 0, 0, &cached_state);
8978
8979 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8980 &cached_state, 0);
8981 if (ret) {
8982 unlock_extent_cached(io_tree, page_start, page_end,
8983 &cached_state);
8984 ret = VM_FAULT_SIGBUS;
8985 goto out_unlock;
8986 }
8987 ret = 0;
8988
8989 /* page is wholly or partially inside EOF */
8990 if (page_start + PAGE_SIZE > size)
8991 zero_start = size & ~PAGE_MASK;
8992 else
8993 zero_start = PAGE_SIZE;
8994
8995 if (zero_start != PAGE_SIZE) {
8996 kaddr = kmap(page);
8997 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8998 flush_dcache_page(page);
8999 kunmap(page);
9000 }
9001 ClearPageChecked(page);
9002 set_page_dirty(page);
9003 SetPageUptodate(page);
9004
9005 BTRFS_I(inode)->last_trans = fs_info->generation;
9006 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9007 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9008
9009 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9010
9011 out_unlock:
9012 if (!ret) {
9013 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
9014 sb_end_pagefault(inode->i_sb);
9015 extent_changeset_free(data_reserved);
9016 return VM_FAULT_LOCKED;
9017 }
9018 unlock_page(page);
9019 out:
9020 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
9021 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9022 reserved_space, (ret != 0));
9023 out_noreserve:
9024 sb_end_pagefault(inode->i_sb);
9025 extent_changeset_free(data_reserved);
9026 return ret;
9027 }
9028
9029 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
9030 {
9031 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9032 struct btrfs_root *root = BTRFS_I(inode)->root;
9033 struct btrfs_block_rsv *rsv;
9034 int ret = 0;
9035 int err = 0;
9036 struct btrfs_trans_handle *trans;
9037 u64 mask = fs_info->sectorsize - 1;
9038 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9039
9040 if (!skip_writeback) {
9041 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9042 (u64)-1);
9043 if (ret)
9044 return ret;
9045 }
9046
9047 /*
9048 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9049 * 3 things going on here
9050 *
9051 * 1) We need to reserve space for our orphan item and the space to
9052 * delete our orphan item. Lord knows we don't want to have a dangling
9053 * orphan item because we didn't reserve space to remove it.
9054 *
9055 * 2) We need to reserve space to update our inode.
9056 *
9057 * 3) We need to have something to cache all the space that is going to
9058 * be free'd up by the truncate operation, but also have some slack
9059 * space reserved in case it uses space during the truncate (thank you
9060 * very much snapshotting).
9061 *
9062 * And we need these to all be separate. The fact is we can use a lot of
9063 * space doing the truncate, and we have no earthly idea how much space
9064 * we will use, so we need the truncate reservation to be separate so it
9065 * doesn't end up using space reserved for updating the inode or
9066 * removing the orphan item. We also need to be able to stop the
9067 * transaction and start a new one, which means we need to be able to
9068 * update the inode several times, and we have no idea of knowing how
9069 * many times that will be, so we can't just reserve 1 item for the
9070 * entirety of the operation, so that has to be done separately as well.
9071 * Then there is the orphan item, which does indeed need to be held on
9072 * to for the whole operation, and we need nobody to touch this reserved
9073 * space except the orphan code.
9074 *
9075 * So that leaves us with
9076 *
9077 * 1) root->orphan_block_rsv - for the orphan deletion.
9078 * 2) rsv - for the truncate reservation, which we will steal from the
9079 * transaction reservation.
9080 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9081 * updating the inode.
9082 */
9083 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9084 if (!rsv)
9085 return -ENOMEM;
9086 rsv->size = min_size;
9087 rsv->failfast = 1;
9088
9089 /*
9090 * 1 for the truncate slack space
9091 * 1 for updating the inode.
9092 */
9093 trans = btrfs_start_transaction(root, 2);
9094 if (IS_ERR(trans)) {
9095 err = PTR_ERR(trans);
9096 goto out;
9097 }
9098
9099 /* Migrate the slack space for the truncate to our reserve */
9100 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9101 min_size, 0);
9102 BUG_ON(ret);
9103
9104 /*
9105 * So if we truncate and then write and fsync we normally would just
9106 * write the extents that changed, which is a problem if we need to
9107 * first truncate that entire inode. So set this flag so we write out
9108 * all of the extents in the inode to the sync log so we're completely
9109 * safe.
9110 */
9111 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9112 trans->block_rsv = rsv;
9113
9114 while (1) {
9115 ret = btrfs_truncate_inode_items(trans, root, inode,
9116 inode->i_size,
9117 BTRFS_EXTENT_DATA_KEY);
9118 trans->block_rsv = &fs_info->trans_block_rsv;
9119 if (ret != -ENOSPC && ret != -EAGAIN) {
9120 err = ret;
9121 break;
9122 }
9123
9124 ret = btrfs_update_inode(trans, root, inode);
9125 if (ret) {
9126 err = ret;
9127 break;
9128 }
9129
9130 btrfs_end_transaction(trans);
9131 btrfs_btree_balance_dirty(fs_info);
9132
9133 trans = btrfs_start_transaction(root, 2);
9134 if (IS_ERR(trans)) {
9135 ret = err = PTR_ERR(trans);
9136 trans = NULL;
9137 break;
9138 }
9139
9140 btrfs_block_rsv_release(fs_info, rsv, -1);
9141 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9142 rsv, min_size, 0);
9143 BUG_ON(ret); /* shouldn't happen */
9144 trans->block_rsv = rsv;
9145 }
9146
9147 /*
9148 * We can't call btrfs_truncate_block inside a trans handle as we could
9149 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9150 * we've truncated everything except the last little bit, and can do
9151 * btrfs_truncate_block and then update the disk_i_size.
9152 */
9153 if (ret == NEED_TRUNCATE_BLOCK) {
9154 btrfs_end_transaction(trans);
9155 btrfs_btree_balance_dirty(fs_info);
9156
9157 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9158 if (ret)
9159 goto out;
9160 trans = btrfs_start_transaction(root, 1);
9161 if (IS_ERR(trans)) {
9162 ret = PTR_ERR(trans);
9163 goto out;
9164 }
9165 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9166 }
9167
9168 if (ret == 0 && inode->i_nlink > 0) {
9169 trans->block_rsv = root->orphan_block_rsv;
9170 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9171 if (ret)
9172 err = ret;
9173 }
9174
9175 if (trans) {
9176 trans->block_rsv = &fs_info->trans_block_rsv;
9177 ret = btrfs_update_inode(trans, root, inode);
9178 if (ret && !err)
9179 err = ret;
9180
9181 ret = btrfs_end_transaction(trans);
9182 btrfs_btree_balance_dirty(fs_info);
9183 }
9184 out:
9185 btrfs_free_block_rsv(fs_info, rsv);
9186
9187 if (ret && !err)
9188 err = ret;
9189
9190 return err;
9191 }
9192
9193 /*
9194 * create a new subvolume directory/inode (helper for the ioctl).
9195 */
9196 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9197 struct btrfs_root *new_root,
9198 struct btrfs_root *parent_root,
9199 u64 new_dirid)
9200 {
9201 struct inode *inode;
9202 int err;
9203 u64 index = 0;
9204
9205 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9206 new_dirid, new_dirid,
9207 S_IFDIR | (~current_umask() & S_IRWXUGO),
9208 &index);
9209 if (IS_ERR(inode))
9210 return PTR_ERR(inode);
9211 inode->i_op = &btrfs_dir_inode_operations;
9212 inode->i_fop = &btrfs_dir_file_operations;
9213
9214 set_nlink(inode, 1);
9215 btrfs_i_size_write(BTRFS_I(inode), 0);
9216 unlock_new_inode(inode);
9217
9218 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9219 if (err)
9220 btrfs_err(new_root->fs_info,
9221 "error inheriting subvolume %llu properties: %d",
9222 new_root->root_key.objectid, err);
9223
9224 err = btrfs_update_inode(trans, new_root, inode);
9225
9226 iput(inode);
9227 return err;
9228 }
9229
9230 struct inode *btrfs_alloc_inode(struct super_block *sb)
9231 {
9232 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9233 struct btrfs_inode *ei;
9234 struct inode *inode;
9235
9236 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9237 if (!ei)
9238 return NULL;
9239
9240 ei->root = NULL;
9241 ei->generation = 0;
9242 ei->last_trans = 0;
9243 ei->last_sub_trans = 0;
9244 ei->logged_trans = 0;
9245 ei->delalloc_bytes = 0;
9246 ei->new_delalloc_bytes = 0;
9247 ei->defrag_bytes = 0;
9248 ei->disk_i_size = 0;
9249 ei->flags = 0;
9250 ei->csum_bytes = 0;
9251 ei->index_cnt = (u64)-1;
9252 ei->dir_index = 0;
9253 ei->last_unlink_trans = 0;
9254 ei->last_log_commit = 0;
9255
9256 spin_lock_init(&ei->lock);
9257 ei->outstanding_extents = 0;
9258 if (sb->s_magic != BTRFS_TEST_MAGIC)
9259 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9260 BTRFS_BLOCK_RSV_DELALLOC);
9261 ei->runtime_flags = 0;
9262 ei->prop_compress = BTRFS_COMPRESS_NONE;
9263 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9264
9265 ei->delayed_node = NULL;
9266
9267 ei->i_otime.tv_sec = 0;
9268 ei->i_otime.tv_nsec = 0;
9269
9270 inode = &ei->vfs_inode;
9271 extent_map_tree_init(&ei->extent_tree);
9272 extent_io_tree_init(&ei->io_tree, inode);
9273 extent_io_tree_init(&ei->io_failure_tree, inode);
9274 ei->io_tree.track_uptodate = 1;
9275 ei->io_failure_tree.track_uptodate = 1;
9276 atomic_set(&ei->sync_writers, 0);
9277 mutex_init(&ei->log_mutex);
9278 mutex_init(&ei->delalloc_mutex);
9279 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9280 INIT_LIST_HEAD(&ei->delalloc_inodes);
9281 INIT_LIST_HEAD(&ei->delayed_iput);
9282 RB_CLEAR_NODE(&ei->rb_node);
9283 init_rwsem(&ei->dio_sem);
9284
9285 return inode;
9286 }
9287
9288 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9289 void btrfs_test_destroy_inode(struct inode *inode)
9290 {
9291 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9292 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9293 }
9294 #endif
9295
9296 static void btrfs_i_callback(struct rcu_head *head)
9297 {
9298 struct inode *inode = container_of(head, struct inode, i_rcu);
9299 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9300 }
9301
9302 void btrfs_destroy_inode(struct inode *inode)
9303 {
9304 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9305 struct btrfs_ordered_extent *ordered;
9306 struct btrfs_root *root = BTRFS_I(inode)->root;
9307
9308 WARN_ON(!hlist_empty(&inode->i_dentry));
9309 WARN_ON(inode->i_data.nrpages);
9310 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9311 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9312 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9313 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9314 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9315 WARN_ON(BTRFS_I(inode)->csum_bytes);
9316 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9317
9318 /*
9319 * This can happen where we create an inode, but somebody else also
9320 * created the same inode and we need to destroy the one we already
9321 * created.
9322 */
9323 if (!root)
9324 goto free;
9325
9326 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9327 &BTRFS_I(inode)->runtime_flags)) {
9328 btrfs_info(fs_info, "inode %llu still on the orphan list",
9329 btrfs_ino(BTRFS_I(inode)));
9330 atomic_dec(&root->orphan_inodes);
9331 }
9332
9333 while (1) {
9334 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9335 if (!ordered)
9336 break;
9337 else {
9338 btrfs_err(fs_info,
9339 "found ordered extent %llu %llu on inode cleanup",
9340 ordered->file_offset, ordered->len);
9341 btrfs_remove_ordered_extent(inode, ordered);
9342 btrfs_put_ordered_extent(ordered);
9343 btrfs_put_ordered_extent(ordered);
9344 }
9345 }
9346 btrfs_qgroup_check_reserved_leak(inode);
9347 inode_tree_del(inode);
9348 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9349 free:
9350 call_rcu(&inode->i_rcu, btrfs_i_callback);
9351 }
9352
9353 int btrfs_drop_inode(struct inode *inode)
9354 {
9355 struct btrfs_root *root = BTRFS_I(inode)->root;
9356
9357 if (root == NULL)
9358 return 1;
9359
9360 /* the snap/subvol tree is on deleting */
9361 if (btrfs_root_refs(&root->root_item) == 0)
9362 return 1;
9363 else
9364 return generic_drop_inode(inode);
9365 }
9366
9367 static void init_once(void *foo)
9368 {
9369 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9370
9371 inode_init_once(&ei->vfs_inode);
9372 }
9373
9374 void __cold btrfs_destroy_cachep(void)
9375 {
9376 /*
9377 * Make sure all delayed rcu free inodes are flushed before we
9378 * destroy cache.
9379 */
9380 rcu_barrier();
9381 kmem_cache_destroy(btrfs_inode_cachep);
9382 kmem_cache_destroy(btrfs_trans_handle_cachep);
9383 kmem_cache_destroy(btrfs_path_cachep);
9384 kmem_cache_destroy(btrfs_free_space_cachep);
9385 }
9386
9387 int __init btrfs_init_cachep(void)
9388 {
9389 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9390 sizeof(struct btrfs_inode), 0,
9391 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9392 init_once);
9393 if (!btrfs_inode_cachep)
9394 goto fail;
9395
9396 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9397 sizeof(struct btrfs_trans_handle), 0,
9398 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9399 if (!btrfs_trans_handle_cachep)
9400 goto fail;
9401
9402 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9403 sizeof(struct btrfs_path), 0,
9404 SLAB_MEM_SPREAD, NULL);
9405 if (!btrfs_path_cachep)
9406 goto fail;
9407
9408 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9409 sizeof(struct btrfs_free_space), 0,
9410 SLAB_MEM_SPREAD, NULL);
9411 if (!btrfs_free_space_cachep)
9412 goto fail;
9413
9414 return 0;
9415 fail:
9416 btrfs_destroy_cachep();
9417 return -ENOMEM;
9418 }
9419
9420 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9421 u32 request_mask, unsigned int flags)
9422 {
9423 u64 delalloc_bytes;
9424 struct inode *inode = d_inode(path->dentry);
9425 u32 blocksize = inode->i_sb->s_blocksize;
9426 u32 bi_flags = BTRFS_I(inode)->flags;
9427
9428 stat->result_mask |= STATX_BTIME;
9429 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9430 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9431 if (bi_flags & BTRFS_INODE_APPEND)
9432 stat->attributes |= STATX_ATTR_APPEND;
9433 if (bi_flags & BTRFS_INODE_COMPRESS)
9434 stat->attributes |= STATX_ATTR_COMPRESSED;
9435 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9436 stat->attributes |= STATX_ATTR_IMMUTABLE;
9437 if (bi_flags & BTRFS_INODE_NODUMP)
9438 stat->attributes |= STATX_ATTR_NODUMP;
9439
9440 stat->attributes_mask |= (STATX_ATTR_APPEND |
9441 STATX_ATTR_COMPRESSED |
9442 STATX_ATTR_IMMUTABLE |
9443 STATX_ATTR_NODUMP);
9444
9445 generic_fillattr(inode, stat);
9446 stat->dev = BTRFS_I(inode)->root->anon_dev;
9447
9448 spin_lock(&BTRFS_I(inode)->lock);
9449 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9450 spin_unlock(&BTRFS_I(inode)->lock);
9451 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9452 ALIGN(delalloc_bytes, blocksize)) >> 9;
9453 return 0;
9454 }
9455
9456 static int btrfs_rename_exchange(struct inode *old_dir,
9457 struct dentry *old_dentry,
9458 struct inode *new_dir,
9459 struct dentry *new_dentry)
9460 {
9461 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9462 struct btrfs_trans_handle *trans;
9463 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9464 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9465 struct inode *new_inode = new_dentry->d_inode;
9466 struct inode *old_inode = old_dentry->d_inode;
9467 struct timespec ctime = current_time(old_inode);
9468 struct dentry *parent;
9469 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9470 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9471 u64 old_idx = 0;
9472 u64 new_idx = 0;
9473 u64 root_objectid;
9474 int ret;
9475 bool root_log_pinned = false;
9476 bool dest_log_pinned = false;
9477
9478 /* we only allow rename subvolume link between subvolumes */
9479 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9480 return -EXDEV;
9481
9482 /* close the race window with snapshot create/destroy ioctl */
9483 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9484 down_read(&fs_info->subvol_sem);
9485 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9486 down_read(&fs_info->subvol_sem);
9487
9488 /*
9489 * We want to reserve the absolute worst case amount of items. So if
9490 * both inodes are subvols and we need to unlink them then that would
9491 * require 4 item modifications, but if they are both normal inodes it
9492 * would require 5 item modifications, so we'll assume their normal
9493 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9494 * should cover the worst case number of items we'll modify.
9495 */
9496 trans = btrfs_start_transaction(root, 12);
9497 if (IS_ERR(trans)) {
9498 ret = PTR_ERR(trans);
9499 goto out_notrans;
9500 }
9501
9502 /*
9503 * We need to find a free sequence number both in the source and
9504 * in the destination directory for the exchange.
9505 */
9506 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9507 if (ret)
9508 goto out_fail;
9509 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9510 if (ret)
9511 goto out_fail;
9512
9513 BTRFS_I(old_inode)->dir_index = 0ULL;
9514 BTRFS_I(new_inode)->dir_index = 0ULL;
9515
9516 /* Reference for the source. */
9517 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9518 /* force full log commit if subvolume involved. */
9519 btrfs_set_log_full_commit(fs_info, trans);
9520 } else {
9521 btrfs_pin_log_trans(root);
9522 root_log_pinned = true;
9523 ret = btrfs_insert_inode_ref(trans, dest,
9524 new_dentry->d_name.name,
9525 new_dentry->d_name.len,
9526 old_ino,
9527 btrfs_ino(BTRFS_I(new_dir)),
9528 old_idx);
9529 if (ret)
9530 goto out_fail;
9531 }
9532
9533 /* And now for the dest. */
9534 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9535 /* force full log commit if subvolume involved. */
9536 btrfs_set_log_full_commit(fs_info, trans);
9537 } else {
9538 btrfs_pin_log_trans(dest);
9539 dest_log_pinned = true;
9540 ret = btrfs_insert_inode_ref(trans, root,
9541 old_dentry->d_name.name,
9542 old_dentry->d_name.len,
9543 new_ino,
9544 btrfs_ino(BTRFS_I(old_dir)),
9545 new_idx);
9546 if (ret)
9547 goto out_fail;
9548 }
9549
9550 /* Update inode version and ctime/mtime. */
9551 inode_inc_iversion(old_dir);
9552 inode_inc_iversion(new_dir);
9553 inode_inc_iversion(old_inode);
9554 inode_inc_iversion(new_inode);
9555 old_dir->i_ctime = old_dir->i_mtime = ctime;
9556 new_dir->i_ctime = new_dir->i_mtime = ctime;
9557 old_inode->i_ctime = ctime;
9558 new_inode->i_ctime = ctime;
9559
9560 if (old_dentry->d_parent != new_dentry->d_parent) {
9561 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9562 BTRFS_I(old_inode), 1);
9563 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9564 BTRFS_I(new_inode), 1);
9565 }
9566
9567 /* src is a subvolume */
9568 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9569 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9570 ret = btrfs_unlink_subvol(trans, root, old_dir,
9571 root_objectid,
9572 old_dentry->d_name.name,
9573 old_dentry->d_name.len);
9574 } else { /* src is an inode */
9575 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9576 BTRFS_I(old_dentry->d_inode),
9577 old_dentry->d_name.name,
9578 old_dentry->d_name.len);
9579 if (!ret)
9580 ret = btrfs_update_inode(trans, root, old_inode);
9581 }
9582 if (ret) {
9583 btrfs_abort_transaction(trans, ret);
9584 goto out_fail;
9585 }
9586
9587 /* dest is a subvolume */
9588 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9589 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9590 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9591 root_objectid,
9592 new_dentry->d_name.name,
9593 new_dentry->d_name.len);
9594 } else { /* dest is an inode */
9595 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9596 BTRFS_I(new_dentry->d_inode),
9597 new_dentry->d_name.name,
9598 new_dentry->d_name.len);
9599 if (!ret)
9600 ret = btrfs_update_inode(trans, dest, new_inode);
9601 }
9602 if (ret) {
9603 btrfs_abort_transaction(trans, ret);
9604 goto out_fail;
9605 }
9606
9607 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9608 new_dentry->d_name.name,
9609 new_dentry->d_name.len, 0, old_idx);
9610 if (ret) {
9611 btrfs_abort_transaction(trans, ret);
9612 goto out_fail;
9613 }
9614
9615 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9616 old_dentry->d_name.name,
9617 old_dentry->d_name.len, 0, new_idx);
9618 if (ret) {
9619 btrfs_abort_transaction(trans, ret);
9620 goto out_fail;
9621 }
9622
9623 if (old_inode->i_nlink == 1)
9624 BTRFS_I(old_inode)->dir_index = old_idx;
9625 if (new_inode->i_nlink == 1)
9626 BTRFS_I(new_inode)->dir_index = new_idx;
9627
9628 if (root_log_pinned) {
9629 parent = new_dentry->d_parent;
9630 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9631 parent);
9632 btrfs_end_log_trans(root);
9633 root_log_pinned = false;
9634 }
9635 if (dest_log_pinned) {
9636 parent = old_dentry->d_parent;
9637 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9638 parent);
9639 btrfs_end_log_trans(dest);
9640 dest_log_pinned = false;
9641 }
9642 out_fail:
9643 /*
9644 * If we have pinned a log and an error happened, we unpin tasks
9645 * trying to sync the log and force them to fallback to a transaction
9646 * commit if the log currently contains any of the inodes involved in
9647 * this rename operation (to ensure we do not persist a log with an
9648 * inconsistent state for any of these inodes or leading to any
9649 * inconsistencies when replayed). If the transaction was aborted, the
9650 * abortion reason is propagated to userspace when attempting to commit
9651 * the transaction. If the log does not contain any of these inodes, we
9652 * allow the tasks to sync it.
9653 */
9654 if (ret && (root_log_pinned || dest_log_pinned)) {
9655 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9656 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9657 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9658 (new_inode &&
9659 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9660 btrfs_set_log_full_commit(fs_info, trans);
9661
9662 if (root_log_pinned) {
9663 btrfs_end_log_trans(root);
9664 root_log_pinned = false;
9665 }
9666 if (dest_log_pinned) {
9667 btrfs_end_log_trans(dest);
9668 dest_log_pinned = false;
9669 }
9670 }
9671 ret = btrfs_end_transaction(trans);
9672 out_notrans:
9673 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9674 up_read(&fs_info->subvol_sem);
9675 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9676 up_read(&fs_info->subvol_sem);
9677
9678 return ret;
9679 }
9680
9681 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9682 struct btrfs_root *root,
9683 struct inode *dir,
9684 struct dentry *dentry)
9685 {
9686 int ret;
9687 struct inode *inode;
9688 u64 objectid;
9689 u64 index;
9690
9691 ret = btrfs_find_free_ino(root, &objectid);
9692 if (ret)
9693 return ret;
9694
9695 inode = btrfs_new_inode(trans, root, dir,
9696 dentry->d_name.name,
9697 dentry->d_name.len,
9698 btrfs_ino(BTRFS_I(dir)),
9699 objectid,
9700 S_IFCHR | WHITEOUT_MODE,
9701 &index);
9702
9703 if (IS_ERR(inode)) {
9704 ret = PTR_ERR(inode);
9705 return ret;
9706 }
9707
9708 inode->i_op = &btrfs_special_inode_operations;
9709 init_special_inode(inode, inode->i_mode,
9710 WHITEOUT_DEV);
9711
9712 ret = btrfs_init_inode_security(trans, inode, dir,
9713 &dentry->d_name);
9714 if (ret)
9715 goto out;
9716
9717 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9718 BTRFS_I(inode), 0, index);
9719 if (ret)
9720 goto out;
9721
9722 ret = btrfs_update_inode(trans, root, inode);
9723 out:
9724 unlock_new_inode(inode);
9725 if (ret)
9726 inode_dec_link_count(inode);
9727 iput(inode);
9728
9729 return ret;
9730 }
9731
9732 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9733 struct inode *new_dir, struct dentry *new_dentry,
9734 unsigned int flags)
9735 {
9736 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9737 struct btrfs_trans_handle *trans;
9738 unsigned int trans_num_items;
9739 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9740 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9741 struct inode *new_inode = d_inode(new_dentry);
9742 struct inode *old_inode = d_inode(old_dentry);
9743 u64 index = 0;
9744 u64 root_objectid;
9745 int ret;
9746 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9747 bool log_pinned = false;
9748
9749 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9750 return -EPERM;
9751
9752 /* we only allow rename subvolume link between subvolumes */
9753 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9754 return -EXDEV;
9755
9756 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9757 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9758 return -ENOTEMPTY;
9759
9760 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9761 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9762 return -ENOTEMPTY;
9763
9764
9765 /* check for collisions, even if the name isn't there */
9766 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9767 new_dentry->d_name.name,
9768 new_dentry->d_name.len);
9769
9770 if (ret) {
9771 if (ret == -EEXIST) {
9772 /* we shouldn't get
9773 * eexist without a new_inode */
9774 if (WARN_ON(!new_inode)) {
9775 return ret;
9776 }
9777 } else {
9778 /* maybe -EOVERFLOW */
9779 return ret;
9780 }
9781 }
9782 ret = 0;
9783
9784 /*
9785 * we're using rename to replace one file with another. Start IO on it
9786 * now so we don't add too much work to the end of the transaction
9787 */
9788 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9789 filemap_flush(old_inode->i_mapping);
9790
9791 /* close the racy window with snapshot create/destroy ioctl */
9792 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9793 down_read(&fs_info->subvol_sem);
9794 /*
9795 * We want to reserve the absolute worst case amount of items. So if
9796 * both inodes are subvols and we need to unlink them then that would
9797 * require 4 item modifications, but if they are both normal inodes it
9798 * would require 5 item modifications, so we'll assume they are normal
9799 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9800 * should cover the worst case number of items we'll modify.
9801 * If our rename has the whiteout flag, we need more 5 units for the
9802 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9803 * when selinux is enabled).
9804 */
9805 trans_num_items = 11;
9806 if (flags & RENAME_WHITEOUT)
9807 trans_num_items += 5;
9808 trans = btrfs_start_transaction(root, trans_num_items);
9809 if (IS_ERR(trans)) {
9810 ret = PTR_ERR(trans);
9811 goto out_notrans;
9812 }
9813
9814 if (dest != root)
9815 btrfs_record_root_in_trans(trans, dest);
9816
9817 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9818 if (ret)
9819 goto out_fail;
9820
9821 BTRFS_I(old_inode)->dir_index = 0ULL;
9822 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9823 /* force full log commit if subvolume involved. */
9824 btrfs_set_log_full_commit(fs_info, trans);
9825 } else {
9826 btrfs_pin_log_trans(root);
9827 log_pinned = true;
9828 ret = btrfs_insert_inode_ref(trans, dest,
9829 new_dentry->d_name.name,
9830 new_dentry->d_name.len,
9831 old_ino,
9832 btrfs_ino(BTRFS_I(new_dir)), index);
9833 if (ret)
9834 goto out_fail;
9835 }
9836
9837 inode_inc_iversion(old_dir);
9838 inode_inc_iversion(new_dir);
9839 inode_inc_iversion(old_inode);
9840 old_dir->i_ctime = old_dir->i_mtime =
9841 new_dir->i_ctime = new_dir->i_mtime =
9842 old_inode->i_ctime = current_time(old_dir);
9843
9844 if (old_dentry->d_parent != new_dentry->d_parent)
9845 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9846 BTRFS_I(old_inode), 1);
9847
9848 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9849 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9850 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9851 old_dentry->d_name.name,
9852 old_dentry->d_name.len);
9853 } else {
9854 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9855 BTRFS_I(d_inode(old_dentry)),
9856 old_dentry->d_name.name,
9857 old_dentry->d_name.len);
9858 if (!ret)
9859 ret = btrfs_update_inode(trans, root, old_inode);
9860 }
9861 if (ret) {
9862 btrfs_abort_transaction(trans, ret);
9863 goto out_fail;
9864 }
9865
9866 if (new_inode) {
9867 inode_inc_iversion(new_inode);
9868 new_inode->i_ctime = current_time(new_inode);
9869 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9870 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9871 root_objectid = BTRFS_I(new_inode)->location.objectid;
9872 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9873 root_objectid,
9874 new_dentry->d_name.name,
9875 new_dentry->d_name.len);
9876 BUG_ON(new_inode->i_nlink == 0);
9877 } else {
9878 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9879 BTRFS_I(d_inode(new_dentry)),
9880 new_dentry->d_name.name,
9881 new_dentry->d_name.len);
9882 }
9883 if (!ret && new_inode->i_nlink == 0)
9884 ret = btrfs_orphan_add(trans,
9885 BTRFS_I(d_inode(new_dentry)));
9886 if (ret) {
9887 btrfs_abort_transaction(trans, ret);
9888 goto out_fail;
9889 }
9890 }
9891
9892 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9893 new_dentry->d_name.name,
9894 new_dentry->d_name.len, 0, index);
9895 if (ret) {
9896 btrfs_abort_transaction(trans, ret);
9897 goto out_fail;
9898 }
9899
9900 if (old_inode->i_nlink == 1)
9901 BTRFS_I(old_inode)->dir_index = index;
9902
9903 if (log_pinned) {
9904 struct dentry *parent = new_dentry->d_parent;
9905
9906 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9907 parent);
9908 btrfs_end_log_trans(root);
9909 log_pinned = false;
9910 }
9911
9912 if (flags & RENAME_WHITEOUT) {
9913 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9914 old_dentry);
9915
9916 if (ret) {
9917 btrfs_abort_transaction(trans, ret);
9918 goto out_fail;
9919 }
9920 }
9921 out_fail:
9922 /*
9923 * If we have pinned the log and an error happened, we unpin tasks
9924 * trying to sync the log and force them to fallback to a transaction
9925 * commit if the log currently contains any of the inodes involved in
9926 * this rename operation (to ensure we do not persist a log with an
9927 * inconsistent state for any of these inodes or leading to any
9928 * inconsistencies when replayed). If the transaction was aborted, the
9929 * abortion reason is propagated to userspace when attempting to commit
9930 * the transaction. If the log does not contain any of these inodes, we
9931 * allow the tasks to sync it.
9932 */
9933 if (ret && log_pinned) {
9934 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9935 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9936 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9937 (new_inode &&
9938 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9939 btrfs_set_log_full_commit(fs_info, trans);
9940
9941 btrfs_end_log_trans(root);
9942 log_pinned = false;
9943 }
9944 btrfs_end_transaction(trans);
9945 out_notrans:
9946 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9947 up_read(&fs_info->subvol_sem);
9948
9949 return ret;
9950 }
9951
9952 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9953 struct inode *new_dir, struct dentry *new_dentry,
9954 unsigned int flags)
9955 {
9956 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9957 return -EINVAL;
9958
9959 if (flags & RENAME_EXCHANGE)
9960 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9961 new_dentry);
9962
9963 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9964 }
9965
9966 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9967 {
9968 struct btrfs_delalloc_work *delalloc_work;
9969 struct inode *inode;
9970
9971 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9972 work);
9973 inode = delalloc_work->inode;
9974 filemap_flush(inode->i_mapping);
9975 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9976 &BTRFS_I(inode)->runtime_flags))
9977 filemap_flush(inode->i_mapping);
9978
9979 if (delalloc_work->delay_iput)
9980 btrfs_add_delayed_iput(inode);
9981 else
9982 iput(inode);
9983 complete(&delalloc_work->completion);
9984 }
9985
9986 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9987 int delay_iput)
9988 {
9989 struct btrfs_delalloc_work *work;
9990
9991 work = kmalloc(sizeof(*work), GFP_NOFS);
9992 if (!work)
9993 return NULL;
9994
9995 init_completion(&work->completion);
9996 INIT_LIST_HEAD(&work->list);
9997 work->inode = inode;
9998 work->delay_iput = delay_iput;
9999 WARN_ON_ONCE(!inode);
10000 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10001 btrfs_run_delalloc_work, NULL, NULL);
10002
10003 return work;
10004 }
10005
10006 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10007 {
10008 wait_for_completion(&work->completion);
10009 kfree(work);
10010 }
10011
10012 /*
10013 * some fairly slow code that needs optimization. This walks the list
10014 * of all the inodes with pending delalloc and forces them to disk.
10015 */
10016 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10017 int nr)
10018 {
10019 struct btrfs_inode *binode;
10020 struct inode *inode;
10021 struct btrfs_delalloc_work *work, *next;
10022 struct list_head works;
10023 struct list_head splice;
10024 int ret = 0;
10025
10026 INIT_LIST_HEAD(&works);
10027 INIT_LIST_HEAD(&splice);
10028
10029 mutex_lock(&root->delalloc_mutex);
10030 spin_lock(&root->delalloc_lock);
10031 list_splice_init(&root->delalloc_inodes, &splice);
10032 while (!list_empty(&splice)) {
10033 binode = list_entry(splice.next, struct btrfs_inode,
10034 delalloc_inodes);
10035
10036 list_move_tail(&binode->delalloc_inodes,
10037 &root->delalloc_inodes);
10038 inode = igrab(&binode->vfs_inode);
10039 if (!inode) {
10040 cond_resched_lock(&root->delalloc_lock);
10041 continue;
10042 }
10043 spin_unlock(&root->delalloc_lock);
10044
10045 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10046 if (!work) {
10047 if (delay_iput)
10048 btrfs_add_delayed_iput(inode);
10049 else
10050 iput(inode);
10051 ret = -ENOMEM;
10052 goto out;
10053 }
10054 list_add_tail(&work->list, &works);
10055 btrfs_queue_work(root->fs_info->flush_workers,
10056 &work->work);
10057 ret++;
10058 if (nr != -1 && ret >= nr)
10059 goto out;
10060 cond_resched();
10061 spin_lock(&root->delalloc_lock);
10062 }
10063 spin_unlock(&root->delalloc_lock);
10064
10065 out:
10066 list_for_each_entry_safe(work, next, &works, list) {
10067 list_del_init(&work->list);
10068 btrfs_wait_and_free_delalloc_work(work);
10069 }
10070
10071 if (!list_empty_careful(&splice)) {
10072 spin_lock(&root->delalloc_lock);
10073 list_splice_tail(&splice, &root->delalloc_inodes);
10074 spin_unlock(&root->delalloc_lock);
10075 }
10076 mutex_unlock(&root->delalloc_mutex);
10077 return ret;
10078 }
10079
10080 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10081 {
10082 struct btrfs_fs_info *fs_info = root->fs_info;
10083 int ret;
10084
10085 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10086 return -EROFS;
10087
10088 ret = __start_delalloc_inodes(root, delay_iput, -1);
10089 if (ret > 0)
10090 ret = 0;
10091 return ret;
10092 }
10093
10094 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10095 int nr)
10096 {
10097 struct btrfs_root *root;
10098 struct list_head splice;
10099 int ret;
10100
10101 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10102 return -EROFS;
10103
10104 INIT_LIST_HEAD(&splice);
10105
10106 mutex_lock(&fs_info->delalloc_root_mutex);
10107 spin_lock(&fs_info->delalloc_root_lock);
10108 list_splice_init(&fs_info->delalloc_roots, &splice);
10109 while (!list_empty(&splice) && nr) {
10110 root = list_first_entry(&splice, struct btrfs_root,
10111 delalloc_root);
10112 root = btrfs_grab_fs_root(root);
10113 BUG_ON(!root);
10114 list_move_tail(&root->delalloc_root,
10115 &fs_info->delalloc_roots);
10116 spin_unlock(&fs_info->delalloc_root_lock);
10117
10118 ret = __start_delalloc_inodes(root, delay_iput, nr);
10119 btrfs_put_fs_root(root);
10120 if (ret < 0)
10121 goto out;
10122
10123 if (nr != -1) {
10124 nr -= ret;
10125 WARN_ON(nr < 0);
10126 }
10127 spin_lock(&fs_info->delalloc_root_lock);
10128 }
10129 spin_unlock(&fs_info->delalloc_root_lock);
10130
10131 ret = 0;
10132 out:
10133 if (!list_empty_careful(&splice)) {
10134 spin_lock(&fs_info->delalloc_root_lock);
10135 list_splice_tail(&splice, &fs_info->delalloc_roots);
10136 spin_unlock(&fs_info->delalloc_root_lock);
10137 }
10138 mutex_unlock(&fs_info->delalloc_root_mutex);
10139 return ret;
10140 }
10141
10142 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10143 const char *symname)
10144 {
10145 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10146 struct btrfs_trans_handle *trans;
10147 struct btrfs_root *root = BTRFS_I(dir)->root;
10148 struct btrfs_path *path;
10149 struct btrfs_key key;
10150 struct inode *inode = NULL;
10151 int err;
10152 int drop_inode = 0;
10153 u64 objectid;
10154 u64 index = 0;
10155 int name_len;
10156 int datasize;
10157 unsigned long ptr;
10158 struct btrfs_file_extent_item *ei;
10159 struct extent_buffer *leaf;
10160
10161 name_len = strlen(symname);
10162 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10163 return -ENAMETOOLONG;
10164
10165 /*
10166 * 2 items for inode item and ref
10167 * 2 items for dir items
10168 * 1 item for updating parent inode item
10169 * 1 item for the inline extent item
10170 * 1 item for xattr if selinux is on
10171 */
10172 trans = btrfs_start_transaction(root, 7);
10173 if (IS_ERR(trans))
10174 return PTR_ERR(trans);
10175
10176 err = btrfs_find_free_ino(root, &objectid);
10177 if (err)
10178 goto out_unlock;
10179
10180 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10181 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10182 objectid, S_IFLNK|S_IRWXUGO, &index);
10183 if (IS_ERR(inode)) {
10184 err = PTR_ERR(inode);
10185 goto out_unlock;
10186 }
10187
10188 /*
10189 * If the active LSM wants to access the inode during
10190 * d_instantiate it needs these. Smack checks to see
10191 * if the filesystem supports xattrs by looking at the
10192 * ops vector.
10193 */
10194 inode->i_fop = &btrfs_file_operations;
10195 inode->i_op = &btrfs_file_inode_operations;
10196 inode->i_mapping->a_ops = &btrfs_aops;
10197 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10198
10199 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10200 if (err)
10201 goto out_unlock_inode;
10202
10203 path = btrfs_alloc_path();
10204 if (!path) {
10205 err = -ENOMEM;
10206 goto out_unlock_inode;
10207 }
10208 key.objectid = btrfs_ino(BTRFS_I(inode));
10209 key.offset = 0;
10210 key.type = BTRFS_EXTENT_DATA_KEY;
10211 datasize = btrfs_file_extent_calc_inline_size(name_len);
10212 err = btrfs_insert_empty_item(trans, root, path, &key,
10213 datasize);
10214 if (err) {
10215 btrfs_free_path(path);
10216 goto out_unlock_inode;
10217 }
10218 leaf = path->nodes[0];
10219 ei = btrfs_item_ptr(leaf, path->slots[0],
10220 struct btrfs_file_extent_item);
10221 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10222 btrfs_set_file_extent_type(leaf, ei,
10223 BTRFS_FILE_EXTENT_INLINE);
10224 btrfs_set_file_extent_encryption(leaf, ei, 0);
10225 btrfs_set_file_extent_compression(leaf, ei, 0);
10226 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10227 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10228
10229 ptr = btrfs_file_extent_inline_start(ei);
10230 write_extent_buffer(leaf, symname, ptr, name_len);
10231 btrfs_mark_buffer_dirty(leaf);
10232 btrfs_free_path(path);
10233
10234 inode->i_op = &btrfs_symlink_inode_operations;
10235 inode_nohighmem(inode);
10236 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10237 inode_set_bytes(inode, name_len);
10238 btrfs_i_size_write(BTRFS_I(inode), name_len);
10239 err = btrfs_update_inode(trans, root, inode);
10240 /*
10241 * Last step, add directory indexes for our symlink inode. This is the
10242 * last step to avoid extra cleanup of these indexes if an error happens
10243 * elsewhere above.
10244 */
10245 if (!err)
10246 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10247 BTRFS_I(inode), 0, index);
10248 if (err) {
10249 drop_inode = 1;
10250 goto out_unlock_inode;
10251 }
10252
10253 unlock_new_inode(inode);
10254 d_instantiate(dentry, inode);
10255
10256 out_unlock:
10257 btrfs_end_transaction(trans);
10258 if (drop_inode) {
10259 inode_dec_link_count(inode);
10260 iput(inode);
10261 }
10262 btrfs_btree_balance_dirty(fs_info);
10263 return err;
10264
10265 out_unlock_inode:
10266 drop_inode = 1;
10267 unlock_new_inode(inode);
10268 goto out_unlock;
10269 }
10270
10271 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10272 u64 start, u64 num_bytes, u64 min_size,
10273 loff_t actual_len, u64 *alloc_hint,
10274 struct btrfs_trans_handle *trans)
10275 {
10276 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10277 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10278 struct extent_map *em;
10279 struct btrfs_root *root = BTRFS_I(inode)->root;
10280 struct btrfs_key ins;
10281 u64 cur_offset = start;
10282 u64 i_size;
10283 u64 cur_bytes;
10284 u64 last_alloc = (u64)-1;
10285 int ret = 0;
10286 bool own_trans = true;
10287 u64 end = start + num_bytes - 1;
10288
10289 if (trans)
10290 own_trans = false;
10291 while (num_bytes > 0) {
10292 if (own_trans) {
10293 trans = btrfs_start_transaction(root, 3);
10294 if (IS_ERR(trans)) {
10295 ret = PTR_ERR(trans);
10296 break;
10297 }
10298 }
10299
10300 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10301 cur_bytes = max(cur_bytes, min_size);
10302 /*
10303 * If we are severely fragmented we could end up with really
10304 * small allocations, so if the allocator is returning small
10305 * chunks lets make its job easier by only searching for those
10306 * sized chunks.
10307 */
10308 cur_bytes = min(cur_bytes, last_alloc);
10309 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10310 min_size, 0, *alloc_hint, &ins, 1, 0);
10311 if (ret) {
10312 if (own_trans)
10313 btrfs_end_transaction(trans);
10314 break;
10315 }
10316 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10317
10318 last_alloc = ins.offset;
10319 ret = insert_reserved_file_extent(trans, inode,
10320 cur_offset, ins.objectid,
10321 ins.offset, ins.offset,
10322 ins.offset, 0, 0, 0,
10323 BTRFS_FILE_EXTENT_PREALLOC);
10324 if (ret) {
10325 btrfs_free_reserved_extent(fs_info, ins.objectid,
10326 ins.offset, 0);
10327 btrfs_abort_transaction(trans, ret);
10328 if (own_trans)
10329 btrfs_end_transaction(trans);
10330 break;
10331 }
10332
10333 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10334 cur_offset + ins.offset -1, 0);
10335
10336 em = alloc_extent_map();
10337 if (!em) {
10338 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10339 &BTRFS_I(inode)->runtime_flags);
10340 goto next;
10341 }
10342
10343 em->start = cur_offset;
10344 em->orig_start = cur_offset;
10345 em->len = ins.offset;
10346 em->block_start = ins.objectid;
10347 em->block_len = ins.offset;
10348 em->orig_block_len = ins.offset;
10349 em->ram_bytes = ins.offset;
10350 em->bdev = fs_info->fs_devices->latest_bdev;
10351 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10352 em->generation = trans->transid;
10353
10354 while (1) {
10355 write_lock(&em_tree->lock);
10356 ret = add_extent_mapping(em_tree, em, 1);
10357 write_unlock(&em_tree->lock);
10358 if (ret != -EEXIST)
10359 break;
10360 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10361 cur_offset + ins.offset - 1,
10362 0);
10363 }
10364 free_extent_map(em);
10365 next:
10366 num_bytes -= ins.offset;
10367 cur_offset += ins.offset;
10368 *alloc_hint = ins.objectid + ins.offset;
10369
10370 inode_inc_iversion(inode);
10371 inode->i_ctime = current_time(inode);
10372 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10373 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10374 (actual_len > inode->i_size) &&
10375 (cur_offset > inode->i_size)) {
10376 if (cur_offset > actual_len)
10377 i_size = actual_len;
10378 else
10379 i_size = cur_offset;
10380 i_size_write(inode, i_size);
10381 btrfs_ordered_update_i_size(inode, i_size, NULL);
10382 }
10383
10384 ret = btrfs_update_inode(trans, root, inode);
10385
10386 if (ret) {
10387 btrfs_abort_transaction(trans, ret);
10388 if (own_trans)
10389 btrfs_end_transaction(trans);
10390 break;
10391 }
10392
10393 if (own_trans)
10394 btrfs_end_transaction(trans);
10395 }
10396 if (cur_offset < end)
10397 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10398 end - cur_offset + 1);
10399 return ret;
10400 }
10401
10402 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10403 u64 start, u64 num_bytes, u64 min_size,
10404 loff_t actual_len, u64 *alloc_hint)
10405 {
10406 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10407 min_size, actual_len, alloc_hint,
10408 NULL);
10409 }
10410
10411 int btrfs_prealloc_file_range_trans(struct inode *inode,
10412 struct btrfs_trans_handle *trans, int mode,
10413 u64 start, u64 num_bytes, u64 min_size,
10414 loff_t actual_len, u64 *alloc_hint)
10415 {
10416 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10417 min_size, actual_len, alloc_hint, trans);
10418 }
10419
10420 static int btrfs_set_page_dirty(struct page *page)
10421 {
10422 return __set_page_dirty_nobuffers(page);
10423 }
10424
10425 static int btrfs_permission(struct inode *inode, int mask)
10426 {
10427 struct btrfs_root *root = BTRFS_I(inode)->root;
10428 umode_t mode = inode->i_mode;
10429
10430 if (mask & MAY_WRITE &&
10431 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10432 if (btrfs_root_readonly(root))
10433 return -EROFS;
10434 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10435 return -EACCES;
10436 }
10437 return generic_permission(inode, mask);
10438 }
10439
10440 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10441 {
10442 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10443 struct btrfs_trans_handle *trans;
10444 struct btrfs_root *root = BTRFS_I(dir)->root;
10445 struct inode *inode = NULL;
10446 u64 objectid;
10447 u64 index;
10448 int ret = 0;
10449
10450 /*
10451 * 5 units required for adding orphan entry
10452 */
10453 trans = btrfs_start_transaction(root, 5);
10454 if (IS_ERR(trans))
10455 return PTR_ERR(trans);
10456
10457 ret = btrfs_find_free_ino(root, &objectid);
10458 if (ret)
10459 goto out;
10460
10461 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10462 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10463 if (IS_ERR(inode)) {
10464 ret = PTR_ERR(inode);
10465 inode = NULL;
10466 goto out;
10467 }
10468
10469 inode->i_fop = &btrfs_file_operations;
10470 inode->i_op = &btrfs_file_inode_operations;
10471
10472 inode->i_mapping->a_ops = &btrfs_aops;
10473 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10474
10475 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10476 if (ret)
10477 goto out_inode;
10478
10479 ret = btrfs_update_inode(trans, root, inode);
10480 if (ret)
10481 goto out_inode;
10482 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10483 if (ret)
10484 goto out_inode;
10485
10486 /*
10487 * We set number of links to 0 in btrfs_new_inode(), and here we set
10488 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10489 * through:
10490 *
10491 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10492 */
10493 set_nlink(inode, 1);
10494 unlock_new_inode(inode);
10495 d_tmpfile(dentry, inode);
10496 mark_inode_dirty(inode);
10497
10498 out:
10499 btrfs_end_transaction(trans);
10500 if (ret)
10501 iput(inode);
10502 btrfs_btree_balance_dirty(fs_info);
10503 return ret;
10504
10505 out_inode:
10506 unlock_new_inode(inode);
10507 goto out;
10508
10509 }
10510
10511 __attribute__((const))
10512 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10513 {
10514 return -EAGAIN;
10515 }
10516
10517 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10518 {
10519 struct inode *inode = private_data;
10520 return btrfs_sb(inode->i_sb);
10521 }
10522
10523 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10524 u64 start, u64 end)
10525 {
10526 struct inode *inode = private_data;
10527 u64 isize;
10528
10529 isize = i_size_read(inode);
10530 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10531 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10532 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10533 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10534 }
10535 }
10536
10537 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10538 {
10539 struct inode *inode = private_data;
10540 unsigned long index = start >> PAGE_SHIFT;
10541 unsigned long end_index = end >> PAGE_SHIFT;
10542 struct page *page;
10543
10544 while (index <= end_index) {
10545 page = find_get_page(inode->i_mapping, index);
10546 ASSERT(page); /* Pages should be in the extent_io_tree */
10547 set_page_writeback(page);
10548 put_page(page);
10549 index++;
10550 }
10551 }
10552
10553 static const struct inode_operations btrfs_dir_inode_operations = {
10554 .getattr = btrfs_getattr,
10555 .lookup = btrfs_lookup,
10556 .create = btrfs_create,
10557 .unlink = btrfs_unlink,
10558 .link = btrfs_link,
10559 .mkdir = btrfs_mkdir,
10560 .rmdir = btrfs_rmdir,
10561 .rename = btrfs_rename2,
10562 .symlink = btrfs_symlink,
10563 .setattr = btrfs_setattr,
10564 .mknod = btrfs_mknod,
10565 .listxattr = btrfs_listxattr,
10566 .permission = btrfs_permission,
10567 .get_acl = btrfs_get_acl,
10568 .set_acl = btrfs_set_acl,
10569 .update_time = btrfs_update_time,
10570 .tmpfile = btrfs_tmpfile,
10571 };
10572 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10573 .lookup = btrfs_lookup,
10574 .permission = btrfs_permission,
10575 .update_time = btrfs_update_time,
10576 };
10577
10578 static const struct file_operations btrfs_dir_file_operations = {
10579 .llseek = generic_file_llseek,
10580 .read = generic_read_dir,
10581 .iterate_shared = btrfs_real_readdir,
10582 .open = btrfs_opendir,
10583 .unlocked_ioctl = btrfs_ioctl,
10584 #ifdef CONFIG_COMPAT
10585 .compat_ioctl = btrfs_compat_ioctl,
10586 #endif
10587 .release = btrfs_release_file,
10588 .fsync = btrfs_sync_file,
10589 };
10590
10591 static const struct extent_io_ops btrfs_extent_io_ops = {
10592 /* mandatory callbacks */
10593 .submit_bio_hook = btrfs_submit_bio_hook,
10594 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10595 .merge_bio_hook = btrfs_merge_bio_hook,
10596 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10597 .tree_fs_info = iotree_fs_info,
10598 .set_range_writeback = btrfs_set_range_writeback,
10599
10600 /* optional callbacks */
10601 .fill_delalloc = run_delalloc_range,
10602 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10603 .writepage_start_hook = btrfs_writepage_start_hook,
10604 .set_bit_hook = btrfs_set_bit_hook,
10605 .clear_bit_hook = btrfs_clear_bit_hook,
10606 .merge_extent_hook = btrfs_merge_extent_hook,
10607 .split_extent_hook = btrfs_split_extent_hook,
10608 .check_extent_io_range = btrfs_check_extent_io_range,
10609 };
10610
10611 /*
10612 * btrfs doesn't support the bmap operation because swapfiles
10613 * use bmap to make a mapping of extents in the file. They assume
10614 * these extents won't change over the life of the file and they
10615 * use the bmap result to do IO directly to the drive.
10616 *
10617 * the btrfs bmap call would return logical addresses that aren't
10618 * suitable for IO and they also will change frequently as COW
10619 * operations happen. So, swapfile + btrfs == corruption.
10620 *
10621 * For now we're avoiding this by dropping bmap.
10622 */
10623 static const struct address_space_operations btrfs_aops = {
10624 .readpage = btrfs_readpage,
10625 .writepage = btrfs_writepage,
10626 .writepages = btrfs_writepages,
10627 .readpages = btrfs_readpages,
10628 .direct_IO = btrfs_direct_IO,
10629 .invalidatepage = btrfs_invalidatepage,
10630 .releasepage = btrfs_releasepage,
10631 .set_page_dirty = btrfs_set_page_dirty,
10632 .error_remove_page = generic_error_remove_page,
10633 };
10634
10635 static const struct address_space_operations btrfs_symlink_aops = {
10636 .readpage = btrfs_readpage,
10637 .writepage = btrfs_writepage,
10638 .invalidatepage = btrfs_invalidatepage,
10639 .releasepage = btrfs_releasepage,
10640 };
10641
10642 static const struct inode_operations btrfs_file_inode_operations = {
10643 .getattr = btrfs_getattr,
10644 .setattr = btrfs_setattr,
10645 .listxattr = btrfs_listxattr,
10646 .permission = btrfs_permission,
10647 .fiemap = btrfs_fiemap,
10648 .get_acl = btrfs_get_acl,
10649 .set_acl = btrfs_set_acl,
10650 .update_time = btrfs_update_time,
10651 };
10652 static const struct inode_operations btrfs_special_inode_operations = {
10653 .getattr = btrfs_getattr,
10654 .setattr = btrfs_setattr,
10655 .permission = btrfs_permission,
10656 .listxattr = btrfs_listxattr,
10657 .get_acl = btrfs_get_acl,
10658 .set_acl = btrfs_set_acl,
10659 .update_time = btrfs_update_time,
10660 };
10661 static const struct inode_operations btrfs_symlink_inode_operations = {
10662 .get_link = page_get_link,
10663 .getattr = btrfs_getattr,
10664 .setattr = btrfs_setattr,
10665 .permission = btrfs_permission,
10666 .listxattr = btrfs_listxattr,
10667 .update_time = btrfs_update_time,
10668 };
10669
10670 const struct dentry_operations btrfs_dentry_operations = {
10671 .d_delete = btrfs_dentry_delete,
10672 .d_release = btrfs_dentry_release,
10673 };
Linux with added FPC-III support