?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
crypto: s5p-sss - Fix kernel Oops in AES-ECB mode
[linux/fpc-iii.git] / fs / btrfs / inode.c
blobc04183cc21178d63f1217df0e28c438068184a81
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include <linux/magic.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61 #include "props.h"
62 #include "qgroup.h"
63 #include "dedupe.h"
64
65 struct btrfs_iget_args {
66 struct btrfs_key *location;
67 struct btrfs_root *root;
68 };
69
70 struct btrfs_dio_data {
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74 int overwrite;
75 };
76
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
86
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
91
92 #define S_SHIFT 12
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101 };
102
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
112 u64 orig_start, u64 block_start,
113 u64 block_len, u64 orig_block_len,
114 u64 ram_bytes, int compress_type,
115 int type);
116
117 static void __endio_write_update_ordered(struct inode *inode,
118 const u64 offset, const u64 bytes,
119 const bool uptodate);
120
121 /*
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
124 *
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
133 */
134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
135 const u64 offset,
136 const u64 bytes)
137 {
138 unsigned long index = offset >> PAGE_SHIFT;
139 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
140 struct page *page;
141
142 while (index <= end_index) {
143 page = find_get_page(inode->i_mapping, index);
144 index++;
145 if (!page)
146 continue;
147 ClearPagePrivate2(page);
148 put_page(page);
149 }
150 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
151 bytes - PAGE_SIZE, false);
152 }
153
154 static int btrfs_dirty_inode(struct inode *inode);
155
156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
157 void btrfs_test_inode_set_ops(struct inode *inode)
158 {
159 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
160 }
161 #endif
162
163 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
164 struct inode *inode, struct inode *dir,
165 const struct qstr *qstr)
166 {
167 int err;
168
169 err = btrfs_init_acl(trans, inode, dir);
170 if (!err)
171 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
172 return err;
173 }
174
175 /*
176 * this does all the hard work for inserting an inline extent into
177 * the btree. The caller should have done a btrfs_drop_extents so that
178 * no overlapping inline items exist in the btree
179 */
180 static int insert_inline_extent(struct btrfs_trans_handle *trans,
181 struct btrfs_path *path, int extent_inserted,
182 struct btrfs_root *root, struct inode *inode,
183 u64 start, size_t size, size_t compressed_size,
184 int compress_type,
185 struct page **compressed_pages)
186 {
187 struct extent_buffer *leaf;
188 struct page *page = NULL;
189 char *kaddr;
190 unsigned long ptr;
191 struct btrfs_file_extent_item *ei;
192 int ret;
193 size_t cur_size = size;
194 unsigned long offset;
195
196 if (compressed_size && compressed_pages)
197 cur_size = compressed_size;
198
199 inode_add_bytes(inode, size);
200
201 if (!extent_inserted) {
202 struct btrfs_key key;
203 size_t datasize;
204
205 key.objectid = btrfs_ino(BTRFS_I(inode));
206 key.offset = start;
207 key.type = BTRFS_EXTENT_DATA_KEY;
208
209 datasize = btrfs_file_extent_calc_inline_size(cur_size);
210 path->leave_spinning = 1;
211 ret = btrfs_insert_empty_item(trans, root, path, &key,
212 datasize);
213 if (ret)
214 goto fail;
215 }
216 leaf = path->nodes[0];
217 ei = btrfs_item_ptr(leaf, path->slots[0],
218 struct btrfs_file_extent_item);
219 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
220 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
221 btrfs_set_file_extent_encryption(leaf, ei, 0);
222 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
223 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
224 ptr = btrfs_file_extent_inline_start(ei);
225
226 if (compress_type != BTRFS_COMPRESS_NONE) {
227 struct page *cpage;
228 int i = 0;
229 while (compressed_size > 0) {
230 cpage = compressed_pages[i];
231 cur_size = min_t(unsigned long, compressed_size,
232 PAGE_SIZE);
233
234 kaddr = kmap_atomic(cpage);
235 write_extent_buffer(leaf, kaddr, ptr, cur_size);
236 kunmap_atomic(kaddr);
237
238 i++;
239 ptr += cur_size;
240 compressed_size -= cur_size;
241 }
242 btrfs_set_file_extent_compression(leaf, ei,
243 compress_type);
244 } else {
245 page = find_get_page(inode->i_mapping,
246 start >> PAGE_SHIFT);
247 btrfs_set_file_extent_compression(leaf, ei, 0);
248 kaddr = kmap_atomic(page);
249 offset = start & (PAGE_SIZE - 1);
250 write_extent_buffer(leaf, kaddr + offset, ptr, size);
251 kunmap_atomic(kaddr);
252 put_page(page);
253 }
254 btrfs_mark_buffer_dirty(leaf);
255 btrfs_release_path(path);
256
257 /*
258 * we're an inline extent, so nobody can
259 * extend the file past i_size without locking
260 * a page we already have locked.
261 *
262 * We must do any isize and inode updates
263 * before we unlock the pages. Otherwise we
264 * could end up racing with unlink.
265 */
266 BTRFS_I(inode)->disk_i_size = inode->i_size;
267 ret = btrfs_update_inode(trans, root, inode);
268
269 fail:
270 return ret;
271 }
272
273
274 /*
275 * conditionally insert an inline extent into the file. This
276 * does the checks required to make sure the data is small enough
277 * to fit as an inline extent.
278 */
279 static noinline int cow_file_range_inline(struct btrfs_root *root,
280 struct inode *inode, u64 start,
281 u64 end, size_t compressed_size,
282 int compress_type,
283 struct page **compressed_pages)
284 {
285 struct btrfs_fs_info *fs_info = root->fs_info;
286 struct btrfs_trans_handle *trans;
287 u64 isize = i_size_read(inode);
288 u64 actual_end = min(end + 1, isize);
289 u64 inline_len = actual_end - start;
290 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
291 u64 data_len = inline_len;
292 int ret;
293 struct btrfs_path *path;
294 int extent_inserted = 0;
295 u32 extent_item_size;
296
297 if (compressed_size)
298 data_len = compressed_size;
299
300 if (start > 0 ||
301 actual_end > fs_info->sectorsize ||
302 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
303 (!compressed_size &&
304 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
305 end + 1 < isize ||
306 data_len > fs_info->max_inline) {
307 return 1;
308 }
309
310 path = btrfs_alloc_path();
311 if (!path)
312 return -ENOMEM;
313
314 trans = btrfs_join_transaction(root);
315 if (IS_ERR(trans)) {
316 btrfs_free_path(path);
317 return PTR_ERR(trans);
318 }
319 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
320
321 if (compressed_size && compressed_pages)
322 extent_item_size = btrfs_file_extent_calc_inline_size(
323 compressed_size);
324 else
325 extent_item_size = btrfs_file_extent_calc_inline_size(
326 inline_len);
327
328 ret = __btrfs_drop_extents(trans, root, inode, path,
329 start, aligned_end, NULL,
330 1, 1, extent_item_size, &extent_inserted);
331 if (ret) {
332 btrfs_abort_transaction(trans, ret);
333 goto out;
334 }
335
336 if (isize > actual_end)
337 inline_len = min_t(u64, isize, actual_end);
338 ret = insert_inline_extent(trans, path, extent_inserted,
339 root, inode, start,
340 inline_len, compressed_size,
341 compress_type, compressed_pages);
342 if (ret && ret != -ENOSPC) {
343 btrfs_abort_transaction(trans, ret);
344 goto out;
345 } else if (ret == -ENOSPC) {
346 ret = 1;
347 goto out;
348 }
349
350 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
351 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
352 out:
353 /*
354 * Don't forget to free the reserved space, as for inlined extent
355 * it won't count as data extent, free them directly here.
356 * And at reserve time, it's always aligned to page size, so
357 * just free one page here.
358 */
359 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
360 btrfs_free_path(path);
361 btrfs_end_transaction(trans);
362 return ret;
363 }
364
365 struct async_extent {
366 u64 start;
367 u64 ram_size;
368 u64 compressed_size;
369 struct page **pages;
370 unsigned long nr_pages;
371 int compress_type;
372 struct list_head list;
373 };
374
375 struct async_cow {
376 struct inode *inode;
377 struct btrfs_root *root;
378 struct page *locked_page;
379 u64 start;
380 u64 end;
381 unsigned int write_flags;
382 struct list_head extents;
383 struct btrfs_work work;
384 };
385
386 static noinline int add_async_extent(struct async_cow *cow,
387 u64 start, u64 ram_size,
388 u64 compressed_size,
389 struct page **pages,
390 unsigned long nr_pages,
391 int compress_type)
392 {
393 struct async_extent *async_extent;
394
395 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
396 BUG_ON(!async_extent); /* -ENOMEM */
397 async_extent->start = start;
398 async_extent->ram_size = ram_size;
399 async_extent->compressed_size = compressed_size;
400 async_extent->pages = pages;
401 async_extent->nr_pages = nr_pages;
402 async_extent->compress_type = compress_type;
403 list_add_tail(&async_extent->list, &cow->extents);
404 return 0;
405 }
406
407 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
408 {
409 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
410
411 /* force compress */
412 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
413 return 1;
414 /* defrag ioctl */
415 if (BTRFS_I(inode)->defrag_compress)
416 return 1;
417 /* bad compression ratios */
418 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
419 return 0;
420 if (btrfs_test_opt(fs_info, COMPRESS) ||
421 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
422 BTRFS_I(inode)->prop_compress)
423 return btrfs_compress_heuristic(inode, start, end);
424 return 0;
425 }
426
427 static inline void inode_should_defrag(struct btrfs_inode *inode,
428 u64 start, u64 end, u64 num_bytes, u64 small_write)
429 {
430 /* If this is a small write inside eof, kick off a defrag */
431 if (num_bytes < small_write &&
432 (start > 0 || end + 1 < inode->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
434 }
435
436 /*
437 * we create compressed extents in two phases. The first
438 * phase compresses a range of pages that have already been
439 * locked (both pages and state bits are locked).
440 *
441 * This is done inside an ordered work queue, and the compression
442 * is spread across many cpus. The actual IO submission is step
443 * two, and the ordered work queue takes care of making sure that
444 * happens in the same order things were put onto the queue by
445 * writepages and friends.
446 *
447 * If this code finds it can't get good compression, it puts an
448 * entry onto the work queue to write the uncompressed bytes. This
449 * makes sure that both compressed inodes and uncompressed inodes
450 * are written in the same order that the flusher thread sent them
451 * down.
452 */
453 static noinline void compress_file_range(struct inode *inode,
454 struct page *locked_page,
455 u64 start, u64 end,
456 struct async_cow *async_cow,
457 int *num_added)
458 {
459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
460 struct btrfs_root *root = BTRFS_I(inode)->root;
461 u64 blocksize = fs_info->sectorsize;
462 u64 actual_end;
463 u64 isize = i_size_read(inode);
464 int ret = 0;
465 struct page **pages = NULL;
466 unsigned long nr_pages;
467 unsigned long total_compressed = 0;
468 unsigned long total_in = 0;
469 int i;
470 int will_compress;
471 int compress_type = fs_info->compress_type;
472 int redirty = 0;
473
474 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
475 SZ_16K);
476
477 actual_end = min_t(u64, isize, end + 1);
478 again:
479 will_compress = 0;
480 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
481 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
482 nr_pages = min_t(unsigned long, nr_pages,
483 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
484
485 /*
486 * we don't want to send crud past the end of i_size through
487 * compression, that's just a waste of CPU time. So, if the
488 * end of the file is before the start of our current
489 * requested range of bytes, we bail out to the uncompressed
490 * cleanup code that can deal with all of this.
491 *
492 * It isn't really the fastest way to fix things, but this is a
493 * very uncommon corner.
494 */
495 if (actual_end <= start)
496 goto cleanup_and_bail_uncompressed;
497
498 total_compressed = actual_end - start;
499
500 /*
501 * skip compression for a small file range(<=blocksize) that
502 * isn't an inline extent, since it doesn't save disk space at all.
503 */
504 if (total_compressed <= blocksize &&
505 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
506 goto cleanup_and_bail_uncompressed;
507
508 total_compressed = min_t(unsigned long, total_compressed,
509 BTRFS_MAX_UNCOMPRESSED);
510 total_in = 0;
511 ret = 0;
512
513 /*
514 * we do compression for mount -o compress and when the
515 * inode has not been flagged as nocompress. This flag can
516 * change at any time if we discover bad compression ratios.
517 */
518 if (inode_need_compress(inode, start, end)) {
519 WARN_ON(pages);
520 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
521 if (!pages) {
522 /* just bail out to the uncompressed code */
523 goto cont;
524 }
525
526 if (BTRFS_I(inode)->defrag_compress)
527 compress_type = BTRFS_I(inode)->defrag_compress;
528 else if (BTRFS_I(inode)->prop_compress)
529 compress_type = BTRFS_I(inode)->prop_compress;
530
531 /*
532 * we need to call clear_page_dirty_for_io on each
533 * page in the range. Otherwise applications with the file
534 * mmap'd can wander in and change the page contents while
535 * we are compressing them.
536 *
537 * If the compression fails for any reason, we set the pages
538 * dirty again later on.
539 */
540 extent_range_clear_dirty_for_io(inode, start, end);
541 redirty = 1;
542
543 /* Compression level is applied here and only here */
544 ret = btrfs_compress_pages(
545 compress_type | (fs_info->compress_level << 4),
546 inode->i_mapping, start,
547 pages,
548 &nr_pages,
549 &total_in,
550 &total_compressed);
551
552 if (!ret) {
553 unsigned long offset = total_compressed &
554 (PAGE_SIZE - 1);
555 struct page *page = pages[nr_pages - 1];
556 char *kaddr;
557
558 /* zero the tail end of the last page, we might be
559 * sending it down to disk
560 */
561 if (offset) {
562 kaddr = kmap_atomic(page);
563 memset(kaddr + offset, 0,
564 PAGE_SIZE - offset);
565 kunmap_atomic(kaddr);
566 }
567 will_compress = 1;
568 }
569 }
570 cont:
571 if (start == 0) {
572 /* lets try to make an inline extent */
573 if (ret || total_in < actual_end) {
574 /* we didn't compress the entire range, try
575 * to make an uncompressed inline extent.
576 */
577 ret = cow_file_range_inline(root, inode, start, end,
578 0, BTRFS_COMPRESS_NONE, NULL);
579 } else {
580 /* try making a compressed inline extent */
581 ret = cow_file_range_inline(root, inode, start, end,
582 total_compressed,
583 compress_type, pages);
584 }
585 if (ret <= 0) {
586 unsigned long clear_flags = EXTENT_DELALLOC |
587 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
588 EXTENT_DO_ACCOUNTING;
589 unsigned long page_error_op;
590
591 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
592
593 /*
594 * inline extent creation worked or returned error,
595 * we don't need to create any more async work items.
596 * Unlock and free up our temp pages.
597 *
598 * We use DO_ACCOUNTING here because we need the
599 * delalloc_release_metadata to be done _after_ we drop
600 * our outstanding extent for clearing delalloc for this
601 * range.
602 */
603 extent_clear_unlock_delalloc(inode, start, end, end,
604 NULL, clear_flags,
605 PAGE_UNLOCK |
606 PAGE_CLEAR_DIRTY |
607 PAGE_SET_WRITEBACK |
608 page_error_op |
609 PAGE_END_WRITEBACK);
610 goto free_pages_out;
611 }
612 }
613
614 if (will_compress) {
615 /*
616 * we aren't doing an inline extent round the compressed size
617 * up to a block size boundary so the allocator does sane
618 * things
619 */
620 total_compressed = ALIGN(total_compressed, blocksize);
621
622 /*
623 * one last check to make sure the compression is really a
624 * win, compare the page count read with the blocks on disk,
625 * compression must free at least one sector size
626 */
627 total_in = ALIGN(total_in, PAGE_SIZE);
628 if (total_compressed + blocksize <= total_in) {
629 *num_added += 1;
630
631 /*
632 * The async work queues will take care of doing actual
633 * allocation on disk for these compressed pages, and
634 * will submit them to the elevator.
635 */
636 add_async_extent(async_cow, start, total_in,
637 total_compressed, pages, nr_pages,
638 compress_type);
639
640 if (start + total_in < end) {
641 start += total_in;
642 pages = NULL;
643 cond_resched();
644 goto again;
645 }
646 return;
647 }
648 }
649 if (pages) {
650 /*
651 * the compression code ran but failed to make things smaller,
652 * free any pages it allocated and our page pointer array
653 */
654 for (i = 0; i < nr_pages; i++) {
655 WARN_ON(pages[i]->mapping);
656 put_page(pages[i]);
657 }
658 kfree(pages);
659 pages = NULL;
660 total_compressed = 0;
661 nr_pages = 0;
662
663 /* flag the file so we don't compress in the future */
664 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
665 !(BTRFS_I(inode)->prop_compress)) {
666 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
667 }
668 }
669 cleanup_and_bail_uncompressed:
670 /*
671 * No compression, but we still need to write the pages in the file
672 * we've been given so far. redirty the locked page if it corresponds
673 * to our extent and set things up for the async work queue to run
674 * cow_file_range to do the normal delalloc dance.
675 */
676 if (page_offset(locked_page) >= start &&
677 page_offset(locked_page) <= end)
678 __set_page_dirty_nobuffers(locked_page);
679 /* unlocked later on in the async handlers */
680
681 if (redirty)
682 extent_range_redirty_for_io(inode, start, end);
683 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
684 BTRFS_COMPRESS_NONE);
685 *num_added += 1;
686
687 return;
688
689 free_pages_out:
690 for (i = 0; i < nr_pages; i++) {
691 WARN_ON(pages[i]->mapping);
692 put_page(pages[i]);
693 }
694 kfree(pages);
695 }
696
697 static void free_async_extent_pages(struct async_extent *async_extent)
698 {
699 int i;
700
701 if (!async_extent->pages)
702 return;
703
704 for (i = 0; i < async_extent->nr_pages; i++) {
705 WARN_ON(async_extent->pages[i]->mapping);
706 put_page(async_extent->pages[i]);
707 }
708 kfree(async_extent->pages);
709 async_extent->nr_pages = 0;
710 async_extent->pages = NULL;
711 }
712
713 /*
714 * phase two of compressed writeback. This is the ordered portion
715 * of the code, which only gets called in the order the work was
716 * queued. We walk all the async extents created by compress_file_range
717 * and send them down to the disk.
718 */
719 static noinline void submit_compressed_extents(struct inode *inode,
720 struct async_cow *async_cow)
721 {
722 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
723 struct async_extent *async_extent;
724 u64 alloc_hint = 0;
725 struct btrfs_key ins;
726 struct extent_map *em;
727 struct btrfs_root *root = BTRFS_I(inode)->root;
728 struct extent_io_tree *io_tree;
729 int ret = 0;
730
731 again:
732 while (!list_empty(&async_cow->extents)) {
733 async_extent = list_entry(async_cow->extents.next,
734 struct async_extent, list);
735 list_del(&async_extent->list);
736
737 io_tree = &BTRFS_I(inode)->io_tree;
738
739 retry:
740 /* did the compression code fall back to uncompressed IO? */
741 if (!async_extent->pages) {
742 int page_started = 0;
743 unsigned long nr_written = 0;
744
745 lock_extent(io_tree, async_extent->start,
746 async_extent->start +
747 async_extent->ram_size - 1);
748
749 /* allocate blocks */
750 ret = cow_file_range(inode, async_cow->locked_page,
751 async_extent->start,
752 async_extent->start +
753 async_extent->ram_size - 1,
754 async_extent->start +
755 async_extent->ram_size - 1,
756 &page_started, &nr_written, 0,
757 NULL);
758
759 /* JDM XXX */
760
761 /*
762 * if page_started, cow_file_range inserted an
763 * inline extent and took care of all the unlocking
764 * and IO for us. Otherwise, we need to submit
765 * all those pages down to the drive.
766 */
767 if (!page_started && !ret)
768 extent_write_locked_range(io_tree,
769 inode, async_extent->start,
770 async_extent->start +
771 async_extent->ram_size - 1,
772 btrfs_get_extent,
773 WB_SYNC_ALL);
774 else if (ret)
775 unlock_page(async_cow->locked_page);
776 kfree(async_extent);
777 cond_resched();
778 continue;
779 }
780
781 lock_extent(io_tree, async_extent->start,
782 async_extent->start + async_extent->ram_size - 1);
783
784 ret = btrfs_reserve_extent(root, async_extent->ram_size,
785 async_extent->compressed_size,
786 async_extent->compressed_size,
787 0, alloc_hint, &ins, 1, 1);
788 if (ret) {
789 free_async_extent_pages(async_extent);
790
791 if (ret == -ENOSPC) {
792 unlock_extent(io_tree, async_extent->start,
793 async_extent->start +
794 async_extent->ram_size - 1);
795
796 /*
797 * we need to redirty the pages if we decide to
798 * fallback to uncompressed IO, otherwise we
799 * will not submit these pages down to lower
800 * layers.
801 */
802 extent_range_redirty_for_io(inode,
803 async_extent->start,
804 async_extent->start +
805 async_extent->ram_size - 1);
806
807 goto retry;
808 }
809 goto out_free;
810 }
811 /*
812 * here we're doing allocation and writeback of the
813 * compressed pages
814 */
815 em = create_io_em(inode, async_extent->start,
816 async_extent->ram_size, /* len */
817 async_extent->start, /* orig_start */
818 ins.objectid, /* block_start */
819 ins.offset, /* block_len */
820 ins.offset, /* orig_block_len */
821 async_extent->ram_size, /* ram_bytes */
822 async_extent->compress_type,
823 BTRFS_ORDERED_COMPRESSED);
824 if (IS_ERR(em))
825 /* ret value is not necessary due to void function */
826 goto out_free_reserve;
827 free_extent_map(em);
828
829 ret = btrfs_add_ordered_extent_compress(inode,
830 async_extent->start,
831 ins.objectid,
832 async_extent->ram_size,
833 ins.offset,
834 BTRFS_ORDERED_COMPRESSED,
835 async_extent->compress_type);
836 if (ret) {
837 btrfs_drop_extent_cache(BTRFS_I(inode),
838 async_extent->start,
839 async_extent->start +
840 async_extent->ram_size - 1, 0);
841 goto out_free_reserve;
842 }
843 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
844
845 /*
846 * clear dirty, set writeback and unlock the pages.
847 */
848 extent_clear_unlock_delalloc(inode, async_extent->start,
849 async_extent->start +
850 async_extent->ram_size - 1,
851 async_extent->start +
852 async_extent->ram_size - 1,
853 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
854 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
855 PAGE_SET_WRITEBACK);
856 if (btrfs_submit_compressed_write(inode,
857 async_extent->start,
858 async_extent->ram_size,
859 ins.objectid,
860 ins.offset, async_extent->pages,
861 async_extent->nr_pages,
862 async_cow->write_flags)) {
863 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
864 struct page *p = async_extent->pages[0];
865 const u64 start = async_extent->start;
866 const u64 end = start + async_extent->ram_size - 1;
867
868 p->mapping = inode->i_mapping;
869 tree->ops->writepage_end_io_hook(p, start, end,
870 NULL, 0);
871 p->mapping = NULL;
872 extent_clear_unlock_delalloc(inode, start, end, end,
873 NULL, 0,
874 PAGE_END_WRITEBACK |
875 PAGE_SET_ERROR);
876 free_async_extent_pages(async_extent);
877 }
878 alloc_hint = ins.objectid + ins.offset;
879 kfree(async_extent);
880 cond_resched();
881 }
882 return;
883 out_free_reserve:
884 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
885 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
886 out_free:
887 extent_clear_unlock_delalloc(inode, async_extent->start,
888 async_extent->start +
889 async_extent->ram_size - 1,
890 async_extent->start +
891 async_extent->ram_size - 1,
892 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
893 EXTENT_DELALLOC_NEW |
894 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
895 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
896 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
897 PAGE_SET_ERROR);
898 free_async_extent_pages(async_extent);
899 kfree(async_extent);
900 goto again;
901 }
902
903 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
904 u64 num_bytes)
905 {
906 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
907 struct extent_map *em;
908 u64 alloc_hint = 0;
909
910 read_lock(&em_tree->lock);
911 em = search_extent_mapping(em_tree, start, num_bytes);
912 if (em) {
913 /*
914 * if block start isn't an actual block number then find the
915 * first block in this inode and use that as a hint. If that
916 * block is also bogus then just don't worry about it.
917 */
918 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
919 free_extent_map(em);
920 em = search_extent_mapping(em_tree, 0, 0);
921 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
922 alloc_hint = em->block_start;
923 if (em)
924 free_extent_map(em);
925 } else {
926 alloc_hint = em->block_start;
927 free_extent_map(em);
928 }
929 }
930 read_unlock(&em_tree->lock);
931
932 return alloc_hint;
933 }
934
935 /*
936 * when extent_io.c finds a delayed allocation range in the file,
937 * the call backs end up in this code. The basic idea is to
938 * allocate extents on disk for the range, and create ordered data structs
939 * in ram to track those extents.
940 *
941 * locked_page is the page that writepage had locked already. We use
942 * it to make sure we don't do extra locks or unlocks.
943 *
944 * *page_started is set to one if we unlock locked_page and do everything
945 * required to start IO on it. It may be clean and already done with
946 * IO when we return.
947 */
948 static noinline int cow_file_range(struct inode *inode,
949 struct page *locked_page,
950 u64 start, u64 end, u64 delalloc_end,
951 int *page_started, unsigned long *nr_written,
952 int unlock, struct btrfs_dedupe_hash *hash)
953 {
954 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
955 struct btrfs_root *root = BTRFS_I(inode)->root;
956 u64 alloc_hint = 0;
957 u64 num_bytes;
958 unsigned long ram_size;
959 u64 disk_num_bytes;
960 u64 cur_alloc_size = 0;
961 u64 blocksize = fs_info->sectorsize;
962 struct btrfs_key ins;
963 struct extent_map *em;
964 unsigned clear_bits;
965 unsigned long page_ops;
966 bool extent_reserved = false;
967 int ret = 0;
968
969 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
970 WARN_ON_ONCE(1);
971 ret = -EINVAL;
972 goto out_unlock;
973 }
974
975 num_bytes = ALIGN(end - start + 1, blocksize);
976 num_bytes = max(blocksize, num_bytes);
977 disk_num_bytes = num_bytes;
978
979 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
980
981 if (start == 0) {
982 /* lets try to make an inline extent */
983 ret = cow_file_range_inline(root, inode, start, end, 0,
984 BTRFS_COMPRESS_NONE, NULL);
985 if (ret == 0) {
986 /*
987 * We use DO_ACCOUNTING here because we need the
988 * delalloc_release_metadata to be run _after_ we drop
989 * our outstanding extent for clearing delalloc for this
990 * range.
991 */
992 extent_clear_unlock_delalloc(inode, start, end,
993 delalloc_end, NULL,
994 EXTENT_LOCKED | EXTENT_DELALLOC |
995 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
996 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
997 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
998 PAGE_END_WRITEBACK);
999 *nr_written = *nr_written +
1000 (end - start + PAGE_SIZE) / PAGE_SIZE;
1001 *page_started = 1;
1002 goto out;
1003 } else if (ret < 0) {
1004 goto out_unlock;
1005 }
1006 }
1007
1008 BUG_ON(disk_num_bytes >
1009 btrfs_super_total_bytes(fs_info->super_copy));
1010
1011 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1012 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1013 start + num_bytes - 1, 0);
1014
1015 while (disk_num_bytes > 0) {
1016 cur_alloc_size = disk_num_bytes;
1017 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1018 fs_info->sectorsize, 0, alloc_hint,
1019 &ins, 1, 1);
1020 if (ret < 0)
1021 goto out_unlock;
1022 cur_alloc_size = ins.offset;
1023 extent_reserved = true;
1024
1025 ram_size = ins.offset;
1026 em = create_io_em(inode, start, ins.offset, /* len */
1027 start, /* orig_start */
1028 ins.objectid, /* block_start */
1029 ins.offset, /* block_len */
1030 ins.offset, /* orig_block_len */
1031 ram_size, /* ram_bytes */
1032 BTRFS_COMPRESS_NONE, /* compress_type */
1033 BTRFS_ORDERED_REGULAR /* type */);
1034 if (IS_ERR(em))
1035 goto out_reserve;
1036 free_extent_map(em);
1037
1038 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1039 ram_size, cur_alloc_size, 0);
1040 if (ret)
1041 goto out_drop_extent_cache;
1042
1043 if (root->root_key.objectid ==
1044 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1045 ret = btrfs_reloc_clone_csums(inode, start,
1046 cur_alloc_size);
1047 /*
1048 * Only drop cache here, and process as normal.
1049 *
1050 * We must not allow extent_clear_unlock_delalloc()
1051 * at out_unlock label to free meta of this ordered
1052 * extent, as its meta should be freed by
1053 * btrfs_finish_ordered_io().
1054 *
1055 * So we must continue until @start is increased to
1056 * skip current ordered extent.
1057 */
1058 if (ret)
1059 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1060 start + ram_size - 1, 0);
1061 }
1062
1063 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1064
1065 /* we're not doing compressed IO, don't unlock the first
1066 * page (which the caller expects to stay locked), don't
1067 * clear any dirty bits and don't set any writeback bits
1068 *
1069 * Do set the Private2 bit so we know this page was properly
1070 * setup for writepage
1071 */
1072 page_ops = unlock ? PAGE_UNLOCK : 0;
1073 page_ops |= PAGE_SET_PRIVATE2;
1074
1075 extent_clear_unlock_delalloc(inode, start,
1076 start + ram_size - 1,
1077 delalloc_end, locked_page,
1078 EXTENT_LOCKED | EXTENT_DELALLOC,
1079 page_ops);
1080 if (disk_num_bytes < cur_alloc_size)
1081 disk_num_bytes = 0;
1082 else
1083 disk_num_bytes -= cur_alloc_size;
1084 num_bytes -= cur_alloc_size;
1085 alloc_hint = ins.objectid + ins.offset;
1086 start += cur_alloc_size;
1087 extent_reserved = false;
1088
1089 /*
1090 * btrfs_reloc_clone_csums() error, since start is increased
1091 * extent_clear_unlock_delalloc() at out_unlock label won't
1092 * free metadata of current ordered extent, we're OK to exit.
1093 */
1094 if (ret)
1095 goto out_unlock;
1096 }
1097 out:
1098 return ret;
1099
1100 out_drop_extent_cache:
1101 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1102 out_reserve:
1103 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1104 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1105 out_unlock:
1106 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1107 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1108 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1109 PAGE_END_WRITEBACK;
1110 /*
1111 * If we reserved an extent for our delalloc range (or a subrange) and
1112 * failed to create the respective ordered extent, then it means that
1113 * when we reserved the extent we decremented the extent's size from
1114 * the data space_info's bytes_may_use counter and incremented the
1115 * space_info's bytes_reserved counter by the same amount. We must make
1116 * sure extent_clear_unlock_delalloc() does not try to decrement again
1117 * the data space_info's bytes_may_use counter, therefore we do not pass
1118 * it the flag EXTENT_CLEAR_DATA_RESV.
1119 */
1120 if (extent_reserved) {
1121 extent_clear_unlock_delalloc(inode, start,
1122 start + cur_alloc_size,
1123 start + cur_alloc_size,
1124 locked_page,
1125 clear_bits,
1126 page_ops);
1127 start += cur_alloc_size;
1128 if (start >= end)
1129 goto out;
1130 }
1131 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1132 locked_page,
1133 clear_bits | EXTENT_CLEAR_DATA_RESV,
1134 page_ops);
1135 goto out;
1136 }
1137
1138 /*
1139 * work queue call back to started compression on a file and pages
1140 */
1141 static noinline void async_cow_start(struct btrfs_work *work)
1142 {
1143 struct async_cow *async_cow;
1144 int num_added = 0;
1145 async_cow = container_of(work, struct async_cow, work);
1146
1147 compress_file_range(async_cow->inode, async_cow->locked_page,
1148 async_cow->start, async_cow->end, async_cow,
1149 &num_added);
1150 if (num_added == 0) {
1151 btrfs_add_delayed_iput(async_cow->inode);
1152 async_cow->inode = NULL;
1153 }
1154 }
1155
1156 /*
1157 * work queue call back to submit previously compressed pages
1158 */
1159 static noinline void async_cow_submit(struct btrfs_work *work)
1160 {
1161 struct btrfs_fs_info *fs_info;
1162 struct async_cow *async_cow;
1163 struct btrfs_root *root;
1164 unsigned long nr_pages;
1165
1166 async_cow = container_of(work, struct async_cow, work);
1167
1168 root = async_cow->root;
1169 fs_info = root->fs_info;
1170 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1171 PAGE_SHIFT;
1172
1173 /*
1174 * atomic_sub_return implies a barrier for waitqueue_active
1175 */
1176 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1177 5 * SZ_1M &&
1178 waitqueue_active(&fs_info->async_submit_wait))
1179 wake_up(&fs_info->async_submit_wait);
1180
1181 if (async_cow->inode)
1182 submit_compressed_extents(async_cow->inode, async_cow);
1183 }
1184
1185 static noinline void async_cow_free(struct btrfs_work *work)
1186 {
1187 struct async_cow *async_cow;
1188 async_cow = container_of(work, struct async_cow, work);
1189 if (async_cow->inode)
1190 btrfs_add_delayed_iput(async_cow->inode);
1191 kfree(async_cow);
1192 }
1193
1194 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1195 u64 start, u64 end, int *page_started,
1196 unsigned long *nr_written,
1197 unsigned int write_flags)
1198 {
1199 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1200 struct async_cow *async_cow;
1201 struct btrfs_root *root = BTRFS_I(inode)->root;
1202 unsigned long nr_pages;
1203 u64 cur_end;
1204
1205 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1206 1, 0, NULL, GFP_NOFS);
1207 while (start < end) {
1208 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1209 BUG_ON(!async_cow); /* -ENOMEM */
1210 async_cow->inode = igrab(inode);
1211 async_cow->root = root;
1212 async_cow->locked_page = locked_page;
1213 async_cow->start = start;
1214 async_cow->write_flags = write_flags;
1215
1216 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1217 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1218 cur_end = end;
1219 else
1220 cur_end = min(end, start + SZ_512K - 1);
1221
1222 async_cow->end = cur_end;
1223 INIT_LIST_HEAD(&async_cow->extents);
1224
1225 btrfs_init_work(&async_cow->work,
1226 btrfs_delalloc_helper,
1227 async_cow_start, async_cow_submit,
1228 async_cow_free);
1229
1230 nr_pages = (cur_end - start + PAGE_SIZE) >>
1231 PAGE_SHIFT;
1232 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1233
1234 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1235
1236 *nr_written += nr_pages;
1237 start = cur_end + 1;
1238 }
1239 *page_started = 1;
1240 return 0;
1241 }
1242
1243 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1244 u64 bytenr, u64 num_bytes)
1245 {
1246 int ret;
1247 struct btrfs_ordered_sum *sums;
1248 LIST_HEAD(list);
1249
1250 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1251 bytenr + num_bytes - 1, &list, 0);
1252 if (ret == 0 && list_empty(&list))
1253 return 0;
1254
1255 while (!list_empty(&list)) {
1256 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1257 list_del(&sums->list);
1258 kfree(sums);
1259 }
1260 return 1;
1261 }
1262
1263 /*
1264 * when nowcow writeback call back. This checks for snapshots or COW copies
1265 * of the extents that exist in the file, and COWs the file as required.
1266 *
1267 * If no cow copies or snapshots exist, we write directly to the existing
1268 * blocks on disk
1269 */
1270 static noinline int run_delalloc_nocow(struct inode *inode,
1271 struct page *locked_page,
1272 u64 start, u64 end, int *page_started, int force,
1273 unsigned long *nr_written)
1274 {
1275 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1276 struct btrfs_root *root = BTRFS_I(inode)->root;
1277 struct extent_buffer *leaf;
1278 struct btrfs_path *path;
1279 struct btrfs_file_extent_item *fi;
1280 struct btrfs_key found_key;
1281 struct extent_map *em;
1282 u64 cow_start;
1283 u64 cur_offset;
1284 u64 extent_end;
1285 u64 extent_offset;
1286 u64 disk_bytenr;
1287 u64 num_bytes;
1288 u64 disk_num_bytes;
1289 u64 ram_bytes;
1290 int extent_type;
1291 int ret, err;
1292 int type;
1293 int nocow;
1294 int check_prev = 1;
1295 bool nolock;
1296 u64 ino = btrfs_ino(BTRFS_I(inode));
1297
1298 path = btrfs_alloc_path();
1299 if (!path) {
1300 extent_clear_unlock_delalloc(inode, start, end, end,
1301 locked_page,
1302 EXTENT_LOCKED | EXTENT_DELALLOC |
1303 EXTENT_DO_ACCOUNTING |
1304 EXTENT_DEFRAG, PAGE_UNLOCK |
1305 PAGE_CLEAR_DIRTY |
1306 PAGE_SET_WRITEBACK |
1307 PAGE_END_WRITEBACK);
1308 return -ENOMEM;
1309 }
1310
1311 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1312
1313 cow_start = (u64)-1;
1314 cur_offset = start;
1315 while (1) {
1316 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1317 cur_offset, 0);
1318 if (ret < 0)
1319 goto error;
1320 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1321 leaf = path->nodes[0];
1322 btrfs_item_key_to_cpu(leaf, &found_key,
1323 path->slots[0] - 1);
1324 if (found_key.objectid == ino &&
1325 found_key.type == BTRFS_EXTENT_DATA_KEY)
1326 path->slots[0]--;
1327 }
1328 check_prev = 0;
1329 next_slot:
1330 leaf = path->nodes[0];
1331 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332 ret = btrfs_next_leaf(root, path);
1333 if (ret < 0) {
1334 if (cow_start != (u64)-1)
1335 cur_offset = cow_start;
1336 goto error;
1337 }
1338 if (ret > 0)
1339 break;
1340 leaf = path->nodes[0];
1341 }
1342
1343 nocow = 0;
1344 disk_bytenr = 0;
1345 num_bytes = 0;
1346 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1347
1348 if (found_key.objectid > ino)
1349 break;
1350 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1351 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1352 path->slots[0]++;
1353 goto next_slot;
1354 }
1355 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1356 found_key.offset > end)
1357 break;
1358
1359 if (found_key.offset > cur_offset) {
1360 extent_end = found_key.offset;
1361 extent_type = 0;
1362 goto out_check;
1363 }
1364
1365 fi = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_file_extent_item);
1367 extent_type = btrfs_file_extent_type(leaf, fi);
1368
1369 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1370 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1371 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1372 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1373 extent_offset = btrfs_file_extent_offset(leaf, fi);
1374 extent_end = found_key.offset +
1375 btrfs_file_extent_num_bytes(leaf, fi);
1376 disk_num_bytes =
1377 btrfs_file_extent_disk_num_bytes(leaf, fi);
1378 if (extent_end <= start) {
1379 path->slots[0]++;
1380 goto next_slot;
1381 }
1382 if (disk_bytenr == 0)
1383 goto out_check;
1384 if (btrfs_file_extent_compression(leaf, fi) ||
1385 btrfs_file_extent_encryption(leaf, fi) ||
1386 btrfs_file_extent_other_encoding(leaf, fi))
1387 goto out_check;
1388 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1389 goto out_check;
1390 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1391 goto out_check;
1392 if (btrfs_cross_ref_exist(root, ino,
1393 found_key.offset -
1394 extent_offset, disk_bytenr))
1395 goto out_check;
1396 disk_bytenr += extent_offset;
1397 disk_bytenr += cur_offset - found_key.offset;
1398 num_bytes = min(end + 1, extent_end) - cur_offset;
1399 /*
1400 * if there are pending snapshots for this root,
1401 * we fall into common COW way.
1402 */
1403 if (!nolock) {
1404 err = btrfs_start_write_no_snapshotting(root);
1405 if (!err)
1406 goto out_check;
1407 }
1408 /*
1409 * force cow if csum exists in the range.
1410 * this ensure that csum for a given extent are
1411 * either valid or do not exist.
1412 */
1413 if (csum_exist_in_range(fs_info, disk_bytenr,
1414 num_bytes)) {
1415 if (!nolock)
1416 btrfs_end_write_no_snapshotting(root);
1417 goto out_check;
1418 }
1419 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1420 if (!nolock)
1421 btrfs_end_write_no_snapshotting(root);
1422 goto out_check;
1423 }
1424 nocow = 1;
1425 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1426 extent_end = found_key.offset +
1427 btrfs_file_extent_inline_len(leaf,
1428 path->slots[0], fi);
1429 extent_end = ALIGN(extent_end,
1430 fs_info->sectorsize);
1431 } else {
1432 BUG_ON(1);
1433 }
1434 out_check:
1435 if (extent_end <= start) {
1436 path->slots[0]++;
1437 if (!nolock && nocow)
1438 btrfs_end_write_no_snapshotting(root);
1439 if (nocow)
1440 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1441 goto next_slot;
1442 }
1443 if (!nocow) {
1444 if (cow_start == (u64)-1)
1445 cow_start = cur_offset;
1446 cur_offset = extent_end;
1447 if (cur_offset > end)
1448 break;
1449 path->slots[0]++;
1450 goto next_slot;
1451 }
1452
1453 btrfs_release_path(path);
1454 if (cow_start != (u64)-1) {
1455 ret = cow_file_range(inode, locked_page,
1456 cow_start, found_key.offset - 1,
1457 end, page_started, nr_written, 1,
1458 NULL);
1459 if (ret) {
1460 if (!nolock && nocow)
1461 btrfs_end_write_no_snapshotting(root);
1462 if (nocow)
1463 btrfs_dec_nocow_writers(fs_info,
1464 disk_bytenr);
1465 goto error;
1466 }
1467 cow_start = (u64)-1;
1468 }
1469
1470 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1471 u64 orig_start = found_key.offset - extent_offset;
1472
1473 em = create_io_em(inode, cur_offset, num_bytes,
1474 orig_start,
1475 disk_bytenr, /* block_start */
1476 num_bytes, /* block_len */
1477 disk_num_bytes, /* orig_block_len */
1478 ram_bytes, BTRFS_COMPRESS_NONE,
1479 BTRFS_ORDERED_PREALLOC);
1480 if (IS_ERR(em)) {
1481 if (!nolock && nocow)
1482 btrfs_end_write_no_snapshotting(root);
1483 if (nocow)
1484 btrfs_dec_nocow_writers(fs_info,
1485 disk_bytenr);
1486 ret = PTR_ERR(em);
1487 goto error;
1488 }
1489 free_extent_map(em);
1490 }
1491
1492 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1493 type = BTRFS_ORDERED_PREALLOC;
1494 } else {
1495 type = BTRFS_ORDERED_NOCOW;
1496 }
1497
1498 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1499 num_bytes, num_bytes, type);
1500 if (nocow)
1501 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1502 BUG_ON(ret); /* -ENOMEM */
1503
1504 if (root->root_key.objectid ==
1505 BTRFS_DATA_RELOC_TREE_OBJECTID)
1506 /*
1507 * Error handled later, as we must prevent
1508 * extent_clear_unlock_delalloc() in error handler
1509 * from freeing metadata of created ordered extent.
1510 */
1511 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1512 num_bytes);
1513
1514 extent_clear_unlock_delalloc(inode, cur_offset,
1515 cur_offset + num_bytes - 1, end,
1516 locked_page, EXTENT_LOCKED |
1517 EXTENT_DELALLOC |
1518 EXTENT_CLEAR_DATA_RESV,
1519 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1520
1521 if (!nolock && nocow)
1522 btrfs_end_write_no_snapshotting(root);
1523 cur_offset = extent_end;
1524
1525 /*
1526 * btrfs_reloc_clone_csums() error, now we're OK to call error
1527 * handler, as metadata for created ordered extent will only
1528 * be freed by btrfs_finish_ordered_io().
1529 */
1530 if (ret)
1531 goto error;
1532 if (cur_offset > end)
1533 break;
1534 }
1535 btrfs_release_path(path);
1536
1537 if (cur_offset <= end && cow_start == (u64)-1) {
1538 cow_start = cur_offset;
1539 cur_offset = end;
1540 }
1541
1542 if (cow_start != (u64)-1) {
1543 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1544 page_started, nr_written, 1, NULL);
1545 if (ret)
1546 goto error;
1547 }
1548
1549 error:
1550 if (ret && cur_offset < end)
1551 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1552 locked_page, EXTENT_LOCKED |
1553 EXTENT_DELALLOC | EXTENT_DEFRAG |
1554 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1555 PAGE_CLEAR_DIRTY |
1556 PAGE_SET_WRITEBACK |
1557 PAGE_END_WRITEBACK);
1558 btrfs_free_path(path);
1559 return ret;
1560 }
1561
1562 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1563 {
1564
1565 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1566 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1567 return 0;
1568
1569 /*
1570 * @defrag_bytes is a hint value, no spinlock held here,
1571 * if is not zero, it means the file is defragging.
1572 * Force cow if given extent needs to be defragged.
1573 */
1574 if (BTRFS_I(inode)->defrag_bytes &&
1575 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1576 EXTENT_DEFRAG, 0, NULL))
1577 return 1;
1578
1579 return 0;
1580 }
1581
1582 /*
1583 * extent_io.c call back to do delayed allocation processing
1584 */
1585 static int run_delalloc_range(void *private_data, struct page *locked_page,
1586 u64 start, u64 end, int *page_started,
1587 unsigned long *nr_written,
1588 struct writeback_control *wbc)
1589 {
1590 struct inode *inode = private_data;
1591 int ret;
1592 int force_cow = need_force_cow(inode, start, end);
1593 unsigned int write_flags = wbc_to_write_flags(wbc);
1594
1595 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1596 ret = run_delalloc_nocow(inode, locked_page, start, end,
1597 page_started, 1, nr_written);
1598 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1599 ret = run_delalloc_nocow(inode, locked_page, start, end,
1600 page_started, 0, nr_written);
1601 } else if (!inode_need_compress(inode, start, end)) {
1602 ret = cow_file_range(inode, locked_page, start, end, end,
1603 page_started, nr_written, 1, NULL);
1604 } else {
1605 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1606 &BTRFS_I(inode)->runtime_flags);
1607 ret = cow_file_range_async(inode, locked_page, start, end,
1608 page_started, nr_written,
1609 write_flags);
1610 }
1611 if (ret)
1612 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1613 return ret;
1614 }
1615
1616 static void btrfs_split_extent_hook(void *private_data,
1617 struct extent_state *orig, u64 split)
1618 {
1619 struct inode *inode = private_data;
1620 u64 size;
1621
1622 /* not delalloc, ignore it */
1623 if (!(orig->state & EXTENT_DELALLOC))
1624 return;
1625
1626 size = orig->end - orig->start + 1;
1627 if (size > BTRFS_MAX_EXTENT_SIZE) {
1628 u32 num_extents;
1629 u64 new_size;
1630
1631 /*
1632 * See the explanation in btrfs_merge_extent_hook, the same
1633 * applies here, just in reverse.
1634 */
1635 new_size = orig->end - split + 1;
1636 num_extents = count_max_extents(new_size);
1637 new_size = split - orig->start;
1638 num_extents += count_max_extents(new_size);
1639 if (count_max_extents(size) >= num_extents)
1640 return;
1641 }
1642
1643 spin_lock(&BTRFS_I(inode)->lock);
1644 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1645 spin_unlock(&BTRFS_I(inode)->lock);
1646 }
1647
1648 /*
1649 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1650 * extents so we can keep track of new extents that are just merged onto old
1651 * extents, such as when we are doing sequential writes, so we can properly
1652 * account for the metadata space we'll need.
1653 */
1654 static void btrfs_merge_extent_hook(void *private_data,
1655 struct extent_state *new,
1656 struct extent_state *other)
1657 {
1658 struct inode *inode = private_data;
1659 u64 new_size, old_size;
1660 u32 num_extents;
1661
1662 /* not delalloc, ignore it */
1663 if (!(other->state & EXTENT_DELALLOC))
1664 return;
1665
1666 if (new->start > other->start)
1667 new_size = new->end - other->start + 1;
1668 else
1669 new_size = other->end - new->start + 1;
1670
1671 /* we're not bigger than the max, unreserve the space and go */
1672 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1673 spin_lock(&BTRFS_I(inode)->lock);
1674 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1675 spin_unlock(&BTRFS_I(inode)->lock);
1676 return;
1677 }
1678
1679 /*
1680 * We have to add up either side to figure out how many extents were
1681 * accounted for before we merged into one big extent. If the number of
1682 * extents we accounted for is <= the amount we need for the new range
1683 * then we can return, otherwise drop. Think of it like this
1684 *
1685 * [ 4k][MAX_SIZE]
1686 *
1687 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1688 * need 2 outstanding extents, on one side we have 1 and the other side
1689 * we have 1 so they are == and we can return. But in this case
1690 *
1691 * [MAX_SIZE+4k][MAX_SIZE+4k]
1692 *
1693 * Each range on their own accounts for 2 extents, but merged together
1694 * they are only 3 extents worth of accounting, so we need to drop in
1695 * this case.
1696 */
1697 old_size = other->end - other->start + 1;
1698 num_extents = count_max_extents(old_size);
1699 old_size = new->end - new->start + 1;
1700 num_extents += count_max_extents(old_size);
1701 if (count_max_extents(new_size) >= num_extents)
1702 return;
1703
1704 spin_lock(&BTRFS_I(inode)->lock);
1705 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1706 spin_unlock(&BTRFS_I(inode)->lock);
1707 }
1708
1709 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1710 struct inode *inode)
1711 {
1712 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1713
1714 spin_lock(&root->delalloc_lock);
1715 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1716 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1717 &root->delalloc_inodes);
1718 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1719 &BTRFS_I(inode)->runtime_flags);
1720 root->nr_delalloc_inodes++;
1721 if (root->nr_delalloc_inodes == 1) {
1722 spin_lock(&fs_info->delalloc_root_lock);
1723 BUG_ON(!list_empty(&root->delalloc_root));
1724 list_add_tail(&root->delalloc_root,
1725 &fs_info->delalloc_roots);
1726 spin_unlock(&fs_info->delalloc_root_lock);
1727 }
1728 }
1729 spin_unlock(&root->delalloc_lock);
1730 }
1731
1732 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1733 struct btrfs_inode *inode)
1734 {
1735 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1736
1737 spin_lock(&root->delalloc_lock);
1738 if (!list_empty(&inode->delalloc_inodes)) {
1739 list_del_init(&inode->delalloc_inodes);
1740 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1741 &inode->runtime_flags);
1742 root->nr_delalloc_inodes--;
1743 if (!root->nr_delalloc_inodes) {
1744 spin_lock(&fs_info->delalloc_root_lock);
1745 BUG_ON(list_empty(&root->delalloc_root));
1746 list_del_init(&root->delalloc_root);
1747 spin_unlock(&fs_info->delalloc_root_lock);
1748 }
1749 }
1750 spin_unlock(&root->delalloc_lock);
1751 }
1752
1753 /*
1754 * extent_io.c set_bit_hook, used to track delayed allocation
1755 * bytes in this file, and to maintain the list of inodes that
1756 * have pending delalloc work to be done.
1757 */
1758 static void btrfs_set_bit_hook(void *private_data,
1759 struct extent_state *state, unsigned *bits)
1760 {
1761 struct inode *inode = private_data;
1762
1763 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1764
1765 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1766 WARN_ON(1);
1767 /*
1768 * set_bit and clear bit hooks normally require _irqsave/restore
1769 * but in this case, we are only testing for the DELALLOC
1770 * bit, which is only set or cleared with irqs on
1771 */
1772 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1773 struct btrfs_root *root = BTRFS_I(inode)->root;
1774 u64 len = state->end + 1 - state->start;
1775 u32 num_extents = count_max_extents(len);
1776 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1777
1778 spin_lock(&BTRFS_I(inode)->lock);
1779 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1780 spin_unlock(&BTRFS_I(inode)->lock);
1781
1782 /* For sanity tests */
1783 if (btrfs_is_testing(fs_info))
1784 return;
1785
1786 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1787 fs_info->delalloc_batch);
1788 spin_lock(&BTRFS_I(inode)->lock);
1789 BTRFS_I(inode)->delalloc_bytes += len;
1790 if (*bits & EXTENT_DEFRAG)
1791 BTRFS_I(inode)->defrag_bytes += len;
1792 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1793 &BTRFS_I(inode)->runtime_flags))
1794 btrfs_add_delalloc_inodes(root, inode);
1795 spin_unlock(&BTRFS_I(inode)->lock);
1796 }
1797
1798 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1799 (*bits & EXTENT_DELALLOC_NEW)) {
1800 spin_lock(&BTRFS_I(inode)->lock);
1801 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1802 state->start;
1803 spin_unlock(&BTRFS_I(inode)->lock);
1804 }
1805 }
1806
1807 /*
1808 * extent_io.c clear_bit_hook, see set_bit_hook for why
1809 */
1810 static void btrfs_clear_bit_hook(void *private_data,
1811 struct extent_state *state,
1812 unsigned *bits)
1813 {
1814 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1815 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1816 u64 len = state->end + 1 - state->start;
1817 u32 num_extents = count_max_extents(len);
1818
1819 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1820 spin_lock(&inode->lock);
1821 inode->defrag_bytes -= len;
1822 spin_unlock(&inode->lock);
1823 }
1824
1825 /*
1826 * set_bit and clear bit hooks normally require _irqsave/restore
1827 * but in this case, we are only testing for the DELALLOC
1828 * bit, which is only set or cleared with irqs on
1829 */
1830 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1831 struct btrfs_root *root = inode->root;
1832 bool do_list = !btrfs_is_free_space_inode(inode);
1833
1834 spin_lock(&inode->lock);
1835 btrfs_mod_outstanding_extents(inode, -num_extents);
1836 spin_unlock(&inode->lock);
1837
1838 /*
1839 * We don't reserve metadata space for space cache inodes so we
1840 * don't need to call dellalloc_release_metadata if there is an
1841 * error.
1842 */
1843 if (*bits & EXTENT_CLEAR_META_RESV &&
1844 root != fs_info->tree_root)
1845 btrfs_delalloc_release_metadata(inode, len);
1846
1847 /* For sanity tests. */
1848 if (btrfs_is_testing(fs_info))
1849 return;
1850
1851 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1852 do_list && !(state->state & EXTENT_NORESERVE) &&
1853 (*bits & EXTENT_CLEAR_DATA_RESV))
1854 btrfs_free_reserved_data_space_noquota(
1855 &inode->vfs_inode,
1856 state->start, len);
1857
1858 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1859 fs_info->delalloc_batch);
1860 spin_lock(&inode->lock);
1861 inode->delalloc_bytes -= len;
1862 if (do_list && inode->delalloc_bytes == 0 &&
1863 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1864 &inode->runtime_flags))
1865 btrfs_del_delalloc_inode(root, inode);
1866 spin_unlock(&inode->lock);
1867 }
1868
1869 if ((state->state & EXTENT_DELALLOC_NEW) &&
1870 (*bits & EXTENT_DELALLOC_NEW)) {
1871 spin_lock(&inode->lock);
1872 ASSERT(inode->new_delalloc_bytes >= len);
1873 inode->new_delalloc_bytes -= len;
1874 spin_unlock(&inode->lock);
1875 }
1876 }
1877
1878 /*
1879 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1880 * we don't create bios that span stripes or chunks
1881 *
1882 * return 1 if page cannot be merged to bio
1883 * return 0 if page can be merged to bio
1884 * return error otherwise
1885 */
1886 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1887 size_t size, struct bio *bio,
1888 unsigned long bio_flags)
1889 {
1890 struct inode *inode = page->mapping->host;
1891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1892 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1893 u64 length = 0;
1894 u64 map_length;
1895 int ret;
1896
1897 if (bio_flags & EXTENT_BIO_COMPRESSED)
1898 return 0;
1899
1900 length = bio->bi_iter.bi_size;
1901 map_length = length;
1902 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1903 NULL, 0);
1904 if (ret < 0)
1905 return ret;
1906 if (map_length < length + size)
1907 return 1;
1908 return 0;
1909 }
1910
1911 /*
1912 * in order to insert checksums into the metadata in large chunks,
1913 * we wait until bio submission time. All the pages in the bio are
1914 * checksummed and sums are attached onto the ordered extent record.
1915 *
1916 * At IO completion time the cums attached on the ordered extent record
1917 * are inserted into the btree
1918 */
1919 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1920 int mirror_num, unsigned long bio_flags,
1921 u64 bio_offset)
1922 {
1923 struct inode *inode = private_data;
1924 blk_status_t ret = 0;
1925
1926 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1927 BUG_ON(ret); /* -ENOMEM */
1928 return 0;
1929 }
1930
1931 /*
1932 * in order to insert checksums into the metadata in large chunks,
1933 * we wait until bio submission time. All the pages in the bio are
1934 * checksummed and sums are attached onto the ordered extent record.
1935 *
1936 * At IO completion time the cums attached on the ordered extent record
1937 * are inserted into the btree
1938 */
1939 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1940 int mirror_num, unsigned long bio_flags,
1941 u64 bio_offset)
1942 {
1943 struct inode *inode = private_data;
1944 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1945 blk_status_t ret;
1946
1947 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1948 if (ret) {
1949 bio->bi_status = ret;
1950 bio_endio(bio);
1951 }
1952 return ret;
1953 }
1954
1955 /*
1956 * extent_io.c submission hook. This does the right thing for csum calculation
1957 * on write, or reading the csums from the tree before a read
1958 */
1959 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1960 int mirror_num, unsigned long bio_flags,
1961 u64 bio_offset)
1962 {
1963 struct inode *inode = private_data;
1964 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1965 struct btrfs_root *root = BTRFS_I(inode)->root;
1966 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1967 blk_status_t ret = 0;
1968 int skip_sum;
1969 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1970
1971 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1972
1973 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1974 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1975
1976 if (bio_op(bio) != REQ_OP_WRITE) {
1977 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1978 if (ret)
1979 goto out;
1980
1981 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1982 ret = btrfs_submit_compressed_read(inode, bio,
1983 mirror_num,
1984 bio_flags);
1985 goto out;
1986 } else if (!skip_sum) {
1987 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1988 if (ret)
1989 goto out;
1990 }
1991 goto mapit;
1992 } else if (async && !skip_sum) {
1993 /* csum items have already been cloned */
1994 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1995 goto mapit;
1996 /* we're doing a write, do the async checksumming */
1997 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1998 bio_offset, inode,
1999 __btrfs_submit_bio_start,
2000 __btrfs_submit_bio_done);
2001 goto out;
2002 } else if (!skip_sum) {
2003 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2004 if (ret)
2005 goto out;
2006 }
2007
2008 mapit:
2009 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2010
2011 out:
2012 if (ret) {
2013 bio->bi_status = ret;
2014 bio_endio(bio);
2015 }
2016 return ret;
2017 }
2018
2019 /*
2020 * given a list of ordered sums record them in the inode. This happens
2021 * at IO completion time based on sums calculated at bio submission time.
2022 */
2023 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2024 struct inode *inode, struct list_head *list)
2025 {
2026 struct btrfs_ordered_sum *sum;
2027
2028 list_for_each_entry(sum, list, list) {
2029 trans->adding_csums = 1;
2030 btrfs_csum_file_blocks(trans,
2031 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2032 trans->adding_csums = 0;
2033 }
2034 return 0;
2035 }
2036
2037 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2038 unsigned int extra_bits,
2039 struct extent_state **cached_state, int dedupe)
2040 {
2041 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2042 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2043 extra_bits, cached_state);
2044 }
2045
2046 /* see btrfs_writepage_start_hook for details on why this is required */
2047 struct btrfs_writepage_fixup {
2048 struct page *page;
2049 struct btrfs_work work;
2050 };
2051
2052 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2053 {
2054 struct btrfs_writepage_fixup *fixup;
2055 struct btrfs_ordered_extent *ordered;
2056 struct extent_state *cached_state = NULL;
2057 struct extent_changeset *data_reserved = NULL;
2058 struct page *page;
2059 struct inode *inode;
2060 u64 page_start;
2061 u64 page_end;
2062 int ret;
2063
2064 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2065 page = fixup->page;
2066 again:
2067 lock_page(page);
2068 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2069 ClearPageChecked(page);
2070 goto out_page;
2071 }
2072
2073 inode = page->mapping->host;
2074 page_start = page_offset(page);
2075 page_end = page_offset(page) + PAGE_SIZE - 1;
2076
2077 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2078 &cached_state);
2079
2080 /* already ordered? We're done */
2081 if (PagePrivate2(page))
2082 goto out;
2083
2084 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2085 PAGE_SIZE);
2086 if (ordered) {
2087 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2088 page_end, &cached_state, GFP_NOFS);
2089 unlock_page(page);
2090 btrfs_start_ordered_extent(inode, ordered, 1);
2091 btrfs_put_ordered_extent(ordered);
2092 goto again;
2093 }
2094
2095 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2096 PAGE_SIZE);
2097 if (ret) {
2098 mapping_set_error(page->mapping, ret);
2099 end_extent_writepage(page, ret, page_start, page_end);
2100 ClearPageChecked(page);
2101 goto out;
2102 }
2103
2104 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2105 &cached_state, 0);
2106 if (ret) {
2107 mapping_set_error(page->mapping, ret);
2108 end_extent_writepage(page, ret, page_start, page_end);
2109 ClearPageChecked(page);
2110 goto out;
2111 }
2112
2113 ClearPageChecked(page);
2114 set_page_dirty(page);
2115 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2116 out:
2117 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2118 &cached_state, GFP_NOFS);
2119 out_page:
2120 unlock_page(page);
2121 put_page(page);
2122 kfree(fixup);
2123 extent_changeset_free(data_reserved);
2124 }
2125
2126 /*
2127 * There are a few paths in the higher layers of the kernel that directly
2128 * set the page dirty bit without asking the filesystem if it is a
2129 * good idea. This causes problems because we want to make sure COW
2130 * properly happens and the data=ordered rules are followed.
2131 *
2132 * In our case any range that doesn't have the ORDERED bit set
2133 * hasn't been properly setup for IO. We kick off an async process
2134 * to fix it up. The async helper will wait for ordered extents, set
2135 * the delalloc bit and make it safe to write the page.
2136 */
2137 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2138 {
2139 struct inode *inode = page->mapping->host;
2140 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2141 struct btrfs_writepage_fixup *fixup;
2142
2143 /* this page is properly in the ordered list */
2144 if (TestClearPagePrivate2(page))
2145 return 0;
2146
2147 if (PageChecked(page))
2148 return -EAGAIN;
2149
2150 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2151 if (!fixup)
2152 return -EAGAIN;
2153
2154 SetPageChecked(page);
2155 get_page(page);
2156 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2157 btrfs_writepage_fixup_worker, NULL, NULL);
2158 fixup->page = page;
2159 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2160 return -EBUSY;
2161 }
2162
2163 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2164 struct inode *inode, u64 file_pos,
2165 u64 disk_bytenr, u64 disk_num_bytes,
2166 u64 num_bytes, u64 ram_bytes,
2167 u8 compression, u8 encryption,
2168 u16 other_encoding, int extent_type)
2169 {
2170 struct btrfs_root *root = BTRFS_I(inode)->root;
2171 struct btrfs_file_extent_item *fi;
2172 struct btrfs_path *path;
2173 struct extent_buffer *leaf;
2174 struct btrfs_key ins;
2175 u64 qg_released;
2176 int extent_inserted = 0;
2177 int ret;
2178
2179 path = btrfs_alloc_path();
2180 if (!path)
2181 return -ENOMEM;
2182
2183 /*
2184 * we may be replacing one extent in the tree with another.
2185 * The new extent is pinned in the extent map, and we don't want
2186 * to drop it from the cache until it is completely in the btree.
2187 *
2188 * So, tell btrfs_drop_extents to leave this extent in the cache.
2189 * the caller is expected to unpin it and allow it to be merged
2190 * with the others.
2191 */
2192 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2193 file_pos + num_bytes, NULL, 0,
2194 1, sizeof(*fi), &extent_inserted);
2195 if (ret)
2196 goto out;
2197
2198 if (!extent_inserted) {
2199 ins.objectid = btrfs_ino(BTRFS_I(inode));
2200 ins.offset = file_pos;
2201 ins.type = BTRFS_EXTENT_DATA_KEY;
2202
2203 path->leave_spinning = 1;
2204 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2205 sizeof(*fi));
2206 if (ret)
2207 goto out;
2208 }
2209 leaf = path->nodes[0];
2210 fi = btrfs_item_ptr(leaf, path->slots[0],
2211 struct btrfs_file_extent_item);
2212 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2213 btrfs_set_file_extent_type(leaf, fi, extent_type);
2214 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2215 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2216 btrfs_set_file_extent_offset(leaf, fi, 0);
2217 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2218 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2219 btrfs_set_file_extent_compression(leaf, fi, compression);
2220 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2221 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2222
2223 btrfs_mark_buffer_dirty(leaf);
2224 btrfs_release_path(path);
2225
2226 inode_add_bytes(inode, num_bytes);
2227
2228 ins.objectid = disk_bytenr;
2229 ins.offset = disk_num_bytes;
2230 ins.type = BTRFS_EXTENT_ITEM_KEY;
2231
2232 /*
2233 * Release the reserved range from inode dirty range map, as it is
2234 * already moved into delayed_ref_head
2235 */
2236 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2237 if (ret < 0)
2238 goto out;
2239 qg_released = ret;
2240 ret = btrfs_alloc_reserved_file_extent(trans, root,
2241 btrfs_ino(BTRFS_I(inode)),
2242 file_pos, qg_released, &ins);
2243 out:
2244 btrfs_free_path(path);
2245
2246 return ret;
2247 }
2248
2249 /* snapshot-aware defrag */
2250 struct sa_defrag_extent_backref {
2251 struct rb_node node;
2252 struct old_sa_defrag_extent *old;
2253 u64 root_id;
2254 u64 inum;
2255 u64 file_pos;
2256 u64 extent_offset;
2257 u64 num_bytes;
2258 u64 generation;
2259 };
2260
2261 struct old_sa_defrag_extent {
2262 struct list_head list;
2263 struct new_sa_defrag_extent *new;
2264
2265 u64 extent_offset;
2266 u64 bytenr;
2267 u64 offset;
2268 u64 len;
2269 int count;
2270 };
2271
2272 struct new_sa_defrag_extent {
2273 struct rb_root root;
2274 struct list_head head;
2275 struct btrfs_path *path;
2276 struct inode *inode;
2277 u64 file_pos;
2278 u64 len;
2279 u64 bytenr;
2280 u64 disk_len;
2281 u8 compress_type;
2282 };
2283
2284 static int backref_comp(struct sa_defrag_extent_backref *b1,
2285 struct sa_defrag_extent_backref *b2)
2286 {
2287 if (b1->root_id < b2->root_id)
2288 return -1;
2289 else if (b1->root_id > b2->root_id)
2290 return 1;
2291
2292 if (b1->inum < b2->inum)
2293 return -1;
2294 else if (b1->inum > b2->inum)
2295 return 1;
2296
2297 if (b1->file_pos < b2->file_pos)
2298 return -1;
2299 else if (b1->file_pos > b2->file_pos)
2300 return 1;
2301
2302 /*
2303 * [------------------------------] ===> (a range of space)
2304 * |<--->| |<---->| =============> (fs/file tree A)
2305 * |<---------------------------->| ===> (fs/file tree B)
2306 *
2307 * A range of space can refer to two file extents in one tree while
2308 * refer to only one file extent in another tree.
2309 *
2310 * So we may process a disk offset more than one time(two extents in A)
2311 * and locate at the same extent(one extent in B), then insert two same
2312 * backrefs(both refer to the extent in B).
2313 */
2314 return 0;
2315 }
2316
2317 static void backref_insert(struct rb_root *root,
2318 struct sa_defrag_extent_backref *backref)
2319 {
2320 struct rb_node **p = &root->rb_node;
2321 struct rb_node *parent = NULL;
2322 struct sa_defrag_extent_backref *entry;
2323 int ret;
2324
2325 while (*p) {
2326 parent = *p;
2327 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2328
2329 ret = backref_comp(backref, entry);
2330 if (ret < 0)
2331 p = &(*p)->rb_left;
2332 else
2333 p = &(*p)->rb_right;
2334 }
2335
2336 rb_link_node(&backref->node, parent, p);
2337 rb_insert_color(&backref->node, root);
2338 }
2339
2340 /*
2341 * Note the backref might has changed, and in this case we just return 0.
2342 */
2343 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2344 void *ctx)
2345 {
2346 struct btrfs_file_extent_item *extent;
2347 struct old_sa_defrag_extent *old = ctx;
2348 struct new_sa_defrag_extent *new = old->new;
2349 struct btrfs_path *path = new->path;
2350 struct btrfs_key key;
2351 struct btrfs_root *root;
2352 struct sa_defrag_extent_backref *backref;
2353 struct extent_buffer *leaf;
2354 struct inode *inode = new->inode;
2355 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2356 int slot;
2357 int ret;
2358 u64 extent_offset;
2359 u64 num_bytes;
2360
2361 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2362 inum == btrfs_ino(BTRFS_I(inode)))
2363 return 0;
2364
2365 key.objectid = root_id;
2366 key.type = BTRFS_ROOT_ITEM_KEY;
2367 key.offset = (u64)-1;
2368
2369 root = btrfs_read_fs_root_no_name(fs_info, &key);
2370 if (IS_ERR(root)) {
2371 if (PTR_ERR(root) == -ENOENT)
2372 return 0;
2373 WARN_ON(1);
2374 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2375 inum, offset, root_id);
2376 return PTR_ERR(root);
2377 }
2378
2379 key.objectid = inum;
2380 key.type = BTRFS_EXTENT_DATA_KEY;
2381 if (offset > (u64)-1 << 32)
2382 key.offset = 0;
2383 else
2384 key.offset = offset;
2385
2386 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2387 if (WARN_ON(ret < 0))
2388 return ret;
2389 ret = 0;
2390
2391 while (1) {
2392 cond_resched();
2393
2394 leaf = path->nodes[0];
2395 slot = path->slots[0];
2396
2397 if (slot >= btrfs_header_nritems(leaf)) {
2398 ret = btrfs_next_leaf(root, path);
2399 if (ret < 0) {
2400 goto out;
2401 } else if (ret > 0) {
2402 ret = 0;
2403 goto out;
2404 }
2405 continue;
2406 }
2407
2408 path->slots[0]++;
2409
2410 btrfs_item_key_to_cpu(leaf, &key, slot);
2411
2412 if (key.objectid > inum)
2413 goto out;
2414
2415 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2416 continue;
2417
2418 extent = btrfs_item_ptr(leaf, slot,
2419 struct btrfs_file_extent_item);
2420
2421 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2422 continue;
2423
2424 /*
2425 * 'offset' refers to the exact key.offset,
2426 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2427 * (key.offset - extent_offset).
2428 */
2429 if (key.offset != offset)
2430 continue;
2431
2432 extent_offset = btrfs_file_extent_offset(leaf, extent);
2433 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2434
2435 if (extent_offset >= old->extent_offset + old->offset +
2436 old->len || extent_offset + num_bytes <=
2437 old->extent_offset + old->offset)
2438 continue;
2439 break;
2440 }
2441
2442 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2443 if (!backref) {
2444 ret = -ENOENT;
2445 goto out;
2446 }
2447
2448 backref->root_id = root_id;
2449 backref->inum = inum;
2450 backref->file_pos = offset;
2451 backref->num_bytes = num_bytes;
2452 backref->extent_offset = extent_offset;
2453 backref->generation = btrfs_file_extent_generation(leaf, extent);
2454 backref->old = old;
2455 backref_insert(&new->root, backref);
2456 old->count++;
2457 out:
2458 btrfs_release_path(path);
2459 WARN_ON(ret);
2460 return ret;
2461 }
2462
2463 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2464 struct new_sa_defrag_extent *new)
2465 {
2466 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2467 struct old_sa_defrag_extent *old, *tmp;
2468 int ret;
2469
2470 new->path = path;
2471
2472 list_for_each_entry_safe(old, tmp, &new->head, list) {
2473 ret = iterate_inodes_from_logical(old->bytenr +
2474 old->extent_offset, fs_info,
2475 path, record_one_backref,
2476 old, false);
2477 if (ret < 0 && ret != -ENOENT)
2478 return false;
2479
2480 /* no backref to be processed for this extent */
2481 if (!old->count) {
2482 list_del(&old->list);
2483 kfree(old);
2484 }
2485 }
2486
2487 if (list_empty(&new->head))
2488 return false;
2489
2490 return true;
2491 }
2492
2493 static int relink_is_mergable(struct extent_buffer *leaf,
2494 struct btrfs_file_extent_item *fi,
2495 struct new_sa_defrag_extent *new)
2496 {
2497 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2498 return 0;
2499
2500 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2501 return 0;
2502
2503 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2504 return 0;
2505
2506 if (btrfs_file_extent_encryption(leaf, fi) ||
2507 btrfs_file_extent_other_encoding(leaf, fi))
2508 return 0;
2509
2510 return 1;
2511 }
2512
2513 /*
2514 * Note the backref might has changed, and in this case we just return 0.
2515 */
2516 static noinline int relink_extent_backref(struct btrfs_path *path,
2517 struct sa_defrag_extent_backref *prev,
2518 struct sa_defrag_extent_backref *backref)
2519 {
2520 struct btrfs_file_extent_item *extent;
2521 struct btrfs_file_extent_item *item;
2522 struct btrfs_ordered_extent *ordered;
2523 struct btrfs_trans_handle *trans;
2524 struct btrfs_root *root;
2525 struct btrfs_key key;
2526 struct extent_buffer *leaf;
2527 struct old_sa_defrag_extent *old = backref->old;
2528 struct new_sa_defrag_extent *new = old->new;
2529 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2530 struct inode *inode;
2531 struct extent_state *cached = NULL;
2532 int ret = 0;
2533 u64 start;
2534 u64 len;
2535 u64 lock_start;
2536 u64 lock_end;
2537 bool merge = false;
2538 int index;
2539
2540 if (prev && prev->root_id == backref->root_id &&
2541 prev->inum == backref->inum &&
2542 prev->file_pos + prev->num_bytes == backref->file_pos)
2543 merge = true;
2544
2545 /* step 1: get root */
2546 key.objectid = backref->root_id;
2547 key.type = BTRFS_ROOT_ITEM_KEY;
2548 key.offset = (u64)-1;
2549
2550 index = srcu_read_lock(&fs_info->subvol_srcu);
2551
2552 root = btrfs_read_fs_root_no_name(fs_info, &key);
2553 if (IS_ERR(root)) {
2554 srcu_read_unlock(&fs_info->subvol_srcu, index);
2555 if (PTR_ERR(root) == -ENOENT)
2556 return 0;
2557 return PTR_ERR(root);
2558 }
2559
2560 if (btrfs_root_readonly(root)) {
2561 srcu_read_unlock(&fs_info->subvol_srcu, index);
2562 return 0;
2563 }
2564
2565 /* step 2: get inode */
2566 key.objectid = backref->inum;
2567 key.type = BTRFS_INODE_ITEM_KEY;
2568 key.offset = 0;
2569
2570 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2571 if (IS_ERR(inode)) {
2572 srcu_read_unlock(&fs_info->subvol_srcu, index);
2573 return 0;
2574 }
2575
2576 srcu_read_unlock(&fs_info->subvol_srcu, index);
2577
2578 /* step 3: relink backref */
2579 lock_start = backref->file_pos;
2580 lock_end = backref->file_pos + backref->num_bytes - 1;
2581 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2582 &cached);
2583
2584 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2585 if (ordered) {
2586 btrfs_put_ordered_extent(ordered);
2587 goto out_unlock;
2588 }
2589
2590 trans = btrfs_join_transaction(root);
2591 if (IS_ERR(trans)) {
2592 ret = PTR_ERR(trans);
2593 goto out_unlock;
2594 }
2595
2596 key.objectid = backref->inum;
2597 key.type = BTRFS_EXTENT_DATA_KEY;
2598 key.offset = backref->file_pos;
2599
2600 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2601 if (ret < 0) {
2602 goto out_free_path;
2603 } else if (ret > 0) {
2604 ret = 0;
2605 goto out_free_path;
2606 }
2607
2608 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2609 struct btrfs_file_extent_item);
2610
2611 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2612 backref->generation)
2613 goto out_free_path;
2614
2615 btrfs_release_path(path);
2616
2617 start = backref->file_pos;
2618 if (backref->extent_offset < old->extent_offset + old->offset)
2619 start += old->extent_offset + old->offset -
2620 backref->extent_offset;
2621
2622 len = min(backref->extent_offset + backref->num_bytes,
2623 old->extent_offset + old->offset + old->len);
2624 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2625
2626 ret = btrfs_drop_extents(trans, root, inode, start,
2627 start + len, 1);
2628 if (ret)
2629 goto out_free_path;
2630 again:
2631 key.objectid = btrfs_ino(BTRFS_I(inode));
2632 key.type = BTRFS_EXTENT_DATA_KEY;
2633 key.offset = start;
2634
2635 path->leave_spinning = 1;
2636 if (merge) {
2637 struct btrfs_file_extent_item *fi;
2638 u64 extent_len;
2639 struct btrfs_key found_key;
2640
2641 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2642 if (ret < 0)
2643 goto out_free_path;
2644
2645 path->slots[0]--;
2646 leaf = path->nodes[0];
2647 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2648
2649 fi = btrfs_item_ptr(leaf, path->slots[0],
2650 struct btrfs_file_extent_item);
2651 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2652
2653 if (extent_len + found_key.offset == start &&
2654 relink_is_mergable(leaf, fi, new)) {
2655 btrfs_set_file_extent_num_bytes(leaf, fi,
2656 extent_len + len);
2657 btrfs_mark_buffer_dirty(leaf);
2658 inode_add_bytes(inode, len);
2659
2660 ret = 1;
2661 goto out_free_path;
2662 } else {
2663 merge = false;
2664 btrfs_release_path(path);
2665 goto again;
2666 }
2667 }
2668
2669 ret = btrfs_insert_empty_item(trans, root, path, &key,
2670 sizeof(*extent));
2671 if (ret) {
2672 btrfs_abort_transaction(trans, ret);
2673 goto out_free_path;
2674 }
2675
2676 leaf = path->nodes[0];
2677 item = btrfs_item_ptr(leaf, path->slots[0],
2678 struct btrfs_file_extent_item);
2679 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2680 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2681 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2682 btrfs_set_file_extent_num_bytes(leaf, item, len);
2683 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2684 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2685 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2686 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2687 btrfs_set_file_extent_encryption(leaf, item, 0);
2688 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2689
2690 btrfs_mark_buffer_dirty(leaf);
2691 inode_add_bytes(inode, len);
2692 btrfs_release_path(path);
2693
2694 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2695 new->disk_len, 0,
2696 backref->root_id, backref->inum,
2697 new->file_pos); /* start - extent_offset */
2698 if (ret) {
2699 btrfs_abort_transaction(trans, ret);
2700 goto out_free_path;
2701 }
2702
2703 ret = 1;
2704 out_free_path:
2705 btrfs_release_path(path);
2706 path->leave_spinning = 0;
2707 btrfs_end_transaction(trans);
2708 out_unlock:
2709 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2710 &cached, GFP_NOFS);
2711 iput(inode);
2712 return ret;
2713 }
2714
2715 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2716 {
2717 struct old_sa_defrag_extent *old, *tmp;
2718
2719 if (!new)
2720 return;
2721
2722 list_for_each_entry_safe(old, tmp, &new->head, list) {
2723 kfree(old);
2724 }
2725 kfree(new);
2726 }
2727
2728 static void relink_file_extents(struct new_sa_defrag_extent *new)
2729 {
2730 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2731 struct btrfs_path *path;
2732 struct sa_defrag_extent_backref *backref;
2733 struct sa_defrag_extent_backref *prev = NULL;
2734 struct inode *inode;
2735 struct btrfs_root *root;
2736 struct rb_node *node;
2737 int ret;
2738
2739 inode = new->inode;
2740 root = BTRFS_I(inode)->root;
2741
2742 path = btrfs_alloc_path();
2743 if (!path)
2744 return;
2745
2746 if (!record_extent_backrefs(path, new)) {
2747 btrfs_free_path(path);
2748 goto out;
2749 }
2750 btrfs_release_path(path);
2751
2752 while (1) {
2753 node = rb_first(&new->root);
2754 if (!node)
2755 break;
2756 rb_erase(node, &new->root);
2757
2758 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2759
2760 ret = relink_extent_backref(path, prev, backref);
2761 WARN_ON(ret < 0);
2762
2763 kfree(prev);
2764
2765 if (ret == 1)
2766 prev = backref;
2767 else
2768 prev = NULL;
2769 cond_resched();
2770 }
2771 kfree(prev);
2772
2773 btrfs_free_path(path);
2774 out:
2775 free_sa_defrag_extent(new);
2776
2777 atomic_dec(&fs_info->defrag_running);
2778 wake_up(&fs_info->transaction_wait);
2779 }
2780
2781 static struct new_sa_defrag_extent *
2782 record_old_file_extents(struct inode *inode,
2783 struct btrfs_ordered_extent *ordered)
2784 {
2785 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2786 struct btrfs_root *root = BTRFS_I(inode)->root;
2787 struct btrfs_path *path;
2788 struct btrfs_key key;
2789 struct old_sa_defrag_extent *old;
2790 struct new_sa_defrag_extent *new;
2791 int ret;
2792
2793 new = kmalloc(sizeof(*new), GFP_NOFS);
2794 if (!new)
2795 return NULL;
2796
2797 new->inode = inode;
2798 new->file_pos = ordered->file_offset;
2799 new->len = ordered->len;
2800 new->bytenr = ordered->start;
2801 new->disk_len = ordered->disk_len;
2802 new->compress_type = ordered->compress_type;
2803 new->root = RB_ROOT;
2804 INIT_LIST_HEAD(&new->head);
2805
2806 path = btrfs_alloc_path();
2807 if (!path)
2808 goto out_kfree;
2809
2810 key.objectid = btrfs_ino(BTRFS_I(inode));
2811 key.type = BTRFS_EXTENT_DATA_KEY;
2812 key.offset = new->file_pos;
2813
2814 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2815 if (ret < 0)
2816 goto out_free_path;
2817 if (ret > 0 && path->slots[0] > 0)
2818 path->slots[0]--;
2819
2820 /* find out all the old extents for the file range */
2821 while (1) {
2822 struct btrfs_file_extent_item *extent;
2823 struct extent_buffer *l;
2824 int slot;
2825 u64 num_bytes;
2826 u64 offset;
2827 u64 end;
2828 u64 disk_bytenr;
2829 u64 extent_offset;
2830
2831 l = path->nodes[0];
2832 slot = path->slots[0];
2833
2834 if (slot >= btrfs_header_nritems(l)) {
2835 ret = btrfs_next_leaf(root, path);
2836 if (ret < 0)
2837 goto out_free_path;
2838 else if (ret > 0)
2839 break;
2840 continue;
2841 }
2842
2843 btrfs_item_key_to_cpu(l, &key, slot);
2844
2845 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2846 break;
2847 if (key.type != BTRFS_EXTENT_DATA_KEY)
2848 break;
2849 if (key.offset >= new->file_pos + new->len)
2850 break;
2851
2852 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2853
2854 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2855 if (key.offset + num_bytes < new->file_pos)
2856 goto next;
2857
2858 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2859 if (!disk_bytenr)
2860 goto next;
2861
2862 extent_offset = btrfs_file_extent_offset(l, extent);
2863
2864 old = kmalloc(sizeof(*old), GFP_NOFS);
2865 if (!old)
2866 goto out_free_path;
2867
2868 offset = max(new->file_pos, key.offset);
2869 end = min(new->file_pos + new->len, key.offset + num_bytes);
2870
2871 old->bytenr = disk_bytenr;
2872 old->extent_offset = extent_offset;
2873 old->offset = offset - key.offset;
2874 old->len = end - offset;
2875 old->new = new;
2876 old->count = 0;
2877 list_add_tail(&old->list, &new->head);
2878 next:
2879 path->slots[0]++;
2880 cond_resched();
2881 }
2882
2883 btrfs_free_path(path);
2884 atomic_inc(&fs_info->defrag_running);
2885
2886 return new;
2887
2888 out_free_path:
2889 btrfs_free_path(path);
2890 out_kfree:
2891 free_sa_defrag_extent(new);
2892 return NULL;
2893 }
2894
2895 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2896 u64 start, u64 len)
2897 {
2898 struct btrfs_block_group_cache *cache;
2899
2900 cache = btrfs_lookup_block_group(fs_info, start);
2901 ASSERT(cache);
2902
2903 spin_lock(&cache->lock);
2904 cache->delalloc_bytes -= len;
2905 spin_unlock(&cache->lock);
2906
2907 btrfs_put_block_group(cache);
2908 }
2909
2910 /* as ordered data IO finishes, this gets called so we can finish
2911 * an ordered extent if the range of bytes in the file it covers are
2912 * fully written.
2913 */
2914 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2915 {
2916 struct inode *inode = ordered_extent->inode;
2917 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2918 struct btrfs_root *root = BTRFS_I(inode)->root;
2919 struct btrfs_trans_handle *trans = NULL;
2920 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2921 struct extent_state *cached_state = NULL;
2922 struct new_sa_defrag_extent *new = NULL;
2923 int compress_type = 0;
2924 int ret = 0;
2925 u64 logical_len = ordered_extent->len;
2926 bool nolock;
2927 bool truncated = false;
2928 bool range_locked = false;
2929 bool clear_new_delalloc_bytes = false;
2930
2931 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2932 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2933 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2934 clear_new_delalloc_bytes = true;
2935
2936 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2937
2938 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2939 ret = -EIO;
2940 goto out;
2941 }
2942
2943 btrfs_free_io_failure_record(BTRFS_I(inode),
2944 ordered_extent->file_offset,
2945 ordered_extent->file_offset +
2946 ordered_extent->len - 1);
2947
2948 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2949 truncated = true;
2950 logical_len = ordered_extent->truncated_len;
2951 /* Truncated the entire extent, don't bother adding */
2952 if (!logical_len)
2953 goto out;
2954 }
2955
2956 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2957 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2958
2959 /*
2960 * For mwrite(mmap + memset to write) case, we still reserve
2961 * space for NOCOW range.
2962 * As NOCOW won't cause a new delayed ref, just free the space
2963 */
2964 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2965 ordered_extent->len);
2966 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2967 if (nolock)
2968 trans = btrfs_join_transaction_nolock(root);
2969 else
2970 trans = btrfs_join_transaction(root);
2971 if (IS_ERR(trans)) {
2972 ret = PTR_ERR(trans);
2973 trans = NULL;
2974 goto out;
2975 }
2976 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2977 ret = btrfs_update_inode_fallback(trans, root, inode);
2978 if (ret) /* -ENOMEM or corruption */
2979 btrfs_abort_transaction(trans, ret);
2980 goto out;
2981 }
2982
2983 range_locked = true;
2984 lock_extent_bits(io_tree, ordered_extent->file_offset,
2985 ordered_extent->file_offset + ordered_extent->len - 1,
2986 &cached_state);
2987
2988 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2989 ordered_extent->file_offset + ordered_extent->len - 1,
2990 EXTENT_DEFRAG, 0, cached_state);
2991 if (ret) {
2992 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2993 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2994 /* the inode is shared */
2995 new = record_old_file_extents(inode, ordered_extent);
2996
2997 clear_extent_bit(io_tree, ordered_extent->file_offset,
2998 ordered_extent->file_offset + ordered_extent->len - 1,
2999 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
3000 }
3001
3002 if (nolock)
3003 trans = btrfs_join_transaction_nolock(root);
3004 else
3005 trans = btrfs_join_transaction(root);
3006 if (IS_ERR(trans)) {
3007 ret = PTR_ERR(trans);
3008 trans = NULL;
3009 goto out;
3010 }
3011
3012 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3013
3014 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3015 compress_type = ordered_extent->compress_type;
3016 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3017 BUG_ON(compress_type);
3018 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3019 ordered_extent->len);
3020 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3021 ordered_extent->file_offset,
3022 ordered_extent->file_offset +
3023 logical_len);
3024 } else {
3025 BUG_ON(root == fs_info->tree_root);
3026 ret = insert_reserved_file_extent(trans, inode,
3027 ordered_extent->file_offset,
3028 ordered_extent->start,
3029 ordered_extent->disk_len,
3030 logical_len, logical_len,
3031 compress_type, 0, 0,
3032 BTRFS_FILE_EXTENT_REG);
3033 if (!ret)
3034 btrfs_release_delalloc_bytes(fs_info,
3035 ordered_extent->start,
3036 ordered_extent->disk_len);
3037 }
3038 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3039 ordered_extent->file_offset, ordered_extent->len,
3040 trans->transid);
3041 if (ret < 0) {
3042 btrfs_abort_transaction(trans, ret);
3043 goto out;
3044 }
3045
3046 add_pending_csums(trans, inode, &ordered_extent->list);
3047
3048 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3049 ret = btrfs_update_inode_fallback(trans, root, inode);
3050 if (ret) { /* -ENOMEM or corruption */
3051 btrfs_abort_transaction(trans, ret);
3052 goto out;
3053 }
3054 ret = 0;
3055 out:
3056 if (range_locked || clear_new_delalloc_bytes) {
3057 unsigned int clear_bits = 0;
3058
3059 if (range_locked)
3060 clear_bits |= EXTENT_LOCKED;
3061 if (clear_new_delalloc_bytes)
3062 clear_bits |= EXTENT_DELALLOC_NEW;
3063 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3064 ordered_extent->file_offset,
3065 ordered_extent->file_offset +
3066 ordered_extent->len - 1,
3067 clear_bits,
3068 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3069 0, &cached_state, GFP_NOFS);
3070 }
3071
3072 if (trans)
3073 btrfs_end_transaction(trans);
3074
3075 if (ret || truncated) {
3076 u64 start, end;
3077
3078 if (truncated)
3079 start = ordered_extent->file_offset + logical_len;
3080 else
3081 start = ordered_extent->file_offset;
3082 end = ordered_extent->file_offset + ordered_extent->len - 1;
3083 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3084
3085 /* Drop the cache for the part of the extent we didn't write. */
3086 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3087
3088 /*
3089 * If the ordered extent had an IOERR or something else went
3090 * wrong we need to return the space for this ordered extent
3091 * back to the allocator. We only free the extent in the
3092 * truncated case if we didn't write out the extent at all.
3093 */
3094 if ((ret || !logical_len) &&
3095 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3096 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3097 btrfs_free_reserved_extent(fs_info,
3098 ordered_extent->start,
3099 ordered_extent->disk_len, 1);
3100 }
3101
3102
3103 /*
3104 * This needs to be done to make sure anybody waiting knows we are done
3105 * updating everything for this ordered extent.
3106 */
3107 btrfs_remove_ordered_extent(inode, ordered_extent);
3108
3109 /* for snapshot-aware defrag */
3110 if (new) {
3111 if (ret) {
3112 free_sa_defrag_extent(new);
3113 atomic_dec(&fs_info->defrag_running);
3114 } else {
3115 relink_file_extents(new);
3116 }
3117 }
3118
3119 /* once for us */
3120 btrfs_put_ordered_extent(ordered_extent);
3121 /* once for the tree */
3122 btrfs_put_ordered_extent(ordered_extent);
3123
3124 return ret;
3125 }
3126
3127 static void finish_ordered_fn(struct btrfs_work *work)
3128 {
3129 struct btrfs_ordered_extent *ordered_extent;
3130 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3131 btrfs_finish_ordered_io(ordered_extent);
3132 }
3133
3134 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3135 struct extent_state *state, int uptodate)
3136 {
3137 struct inode *inode = page->mapping->host;
3138 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3139 struct btrfs_ordered_extent *ordered_extent = NULL;
3140 struct btrfs_workqueue *wq;
3141 btrfs_work_func_t func;
3142
3143 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3144
3145 ClearPagePrivate2(page);
3146 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3147 end - start + 1, uptodate))
3148 return;
3149
3150 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3151 wq = fs_info->endio_freespace_worker;
3152 func = btrfs_freespace_write_helper;
3153 } else {
3154 wq = fs_info->endio_write_workers;
3155 func = btrfs_endio_write_helper;
3156 }
3157
3158 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3159 NULL);
3160 btrfs_queue_work(wq, &ordered_extent->work);
3161 }
3162
3163 static int __readpage_endio_check(struct inode *inode,
3164 struct btrfs_io_bio *io_bio,
3165 int icsum, struct page *page,
3166 int pgoff, u64 start, size_t len)
3167 {
3168 char *kaddr;
3169 u32 csum_expected;
3170 u32 csum = ~(u32)0;
3171
3172 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3173
3174 kaddr = kmap_atomic(page);
3175 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3176 btrfs_csum_final(csum, (u8 *)&csum);
3177 if (csum != csum_expected)
3178 goto zeroit;
3179
3180 kunmap_atomic(kaddr);
3181 return 0;
3182 zeroit:
3183 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3184 io_bio->mirror_num);
3185 memset(kaddr + pgoff, 1, len);
3186 flush_dcache_page(page);
3187 kunmap_atomic(kaddr);
3188 return -EIO;
3189 }
3190
3191 /*
3192 * when reads are done, we need to check csums to verify the data is correct
3193 * if there's a match, we allow the bio to finish. If not, the code in
3194 * extent_io.c will try to find good copies for us.
3195 */
3196 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3197 u64 phy_offset, struct page *page,
3198 u64 start, u64 end, int mirror)
3199 {
3200 size_t offset = start - page_offset(page);
3201 struct inode *inode = page->mapping->host;
3202 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3203 struct btrfs_root *root = BTRFS_I(inode)->root;
3204
3205 if (PageChecked(page)) {
3206 ClearPageChecked(page);
3207 return 0;
3208 }
3209
3210 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3211 return 0;
3212
3213 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3214 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3215 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3216 return 0;
3217 }
3218
3219 phy_offset >>= inode->i_sb->s_blocksize_bits;
3220 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3221 start, (size_t)(end - start + 1));
3222 }
3223
3224 void btrfs_add_delayed_iput(struct inode *inode)
3225 {
3226 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3227 struct btrfs_inode *binode = BTRFS_I(inode);
3228
3229 if (atomic_add_unless(&inode->i_count, -1, 1))
3230 return;
3231
3232 spin_lock(&fs_info->delayed_iput_lock);
3233 if (binode->delayed_iput_count == 0) {
3234 ASSERT(list_empty(&binode->delayed_iput));
3235 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3236 } else {
3237 binode->delayed_iput_count++;
3238 }
3239 spin_unlock(&fs_info->delayed_iput_lock);
3240 }
3241
3242 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3243 {
3244
3245 spin_lock(&fs_info->delayed_iput_lock);
3246 while (!list_empty(&fs_info->delayed_iputs)) {
3247 struct btrfs_inode *inode;
3248
3249 inode = list_first_entry(&fs_info->delayed_iputs,
3250 struct btrfs_inode, delayed_iput);
3251 if (inode->delayed_iput_count) {
3252 inode->delayed_iput_count--;
3253 list_move_tail(&inode->delayed_iput,
3254 &fs_info->delayed_iputs);
3255 } else {
3256 list_del_init(&inode->delayed_iput);
3257 }
3258 spin_unlock(&fs_info->delayed_iput_lock);
3259 iput(&inode->vfs_inode);
3260 spin_lock(&fs_info->delayed_iput_lock);
3261 }
3262 spin_unlock(&fs_info->delayed_iput_lock);
3263 }
3264
3265 /*
3266 * This is called in transaction commit time. If there are no orphan
3267 * files in the subvolume, it removes orphan item and frees block_rsv
3268 * structure.
3269 */
3270 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3271 struct btrfs_root *root)
3272 {
3273 struct btrfs_fs_info *fs_info = root->fs_info;
3274 struct btrfs_block_rsv *block_rsv;
3275 int ret;
3276
3277 if (atomic_read(&root->orphan_inodes) ||
3278 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3279 return;
3280
3281 spin_lock(&root->orphan_lock);
3282 if (atomic_read(&root->orphan_inodes)) {
3283 spin_unlock(&root->orphan_lock);
3284 return;
3285 }
3286
3287 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3288 spin_unlock(&root->orphan_lock);
3289 return;
3290 }
3291
3292 block_rsv = root->orphan_block_rsv;
3293 root->orphan_block_rsv = NULL;
3294 spin_unlock(&root->orphan_lock);
3295
3296 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3297 btrfs_root_refs(&root->root_item) > 0) {
3298 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3299 root->root_key.objectid);
3300 if (ret)
3301 btrfs_abort_transaction(trans, ret);
3302 else
3303 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3304 &root->state);
3305 }
3306
3307 if (block_rsv) {
3308 WARN_ON(block_rsv->size > 0);
3309 btrfs_free_block_rsv(fs_info, block_rsv);
3310 }
3311 }
3312
3313 /*
3314 * This creates an orphan entry for the given inode in case something goes
3315 * wrong in the middle of an unlink/truncate.
3316 *
3317 * NOTE: caller of this function should reserve 5 units of metadata for
3318 * this function.
3319 */
3320 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3321 struct btrfs_inode *inode)
3322 {
3323 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3324 struct btrfs_root *root = inode->root;
3325 struct btrfs_block_rsv *block_rsv = NULL;
3326 int reserve = 0;
3327 int insert = 0;
3328 int ret;
3329
3330 if (!root->orphan_block_rsv) {
3331 block_rsv = btrfs_alloc_block_rsv(fs_info,
3332 BTRFS_BLOCK_RSV_TEMP);
3333 if (!block_rsv)
3334 return -ENOMEM;
3335 }
3336
3337 spin_lock(&root->orphan_lock);
3338 if (!root->orphan_block_rsv) {
3339 root->orphan_block_rsv = block_rsv;
3340 } else if (block_rsv) {
3341 btrfs_free_block_rsv(fs_info, block_rsv);
3342 block_rsv = NULL;
3343 }
3344
3345 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3346 &inode->runtime_flags)) {
3347 #if 0
3348 /*
3349 * For proper ENOSPC handling, we should do orphan
3350 * cleanup when mounting. But this introduces backward
3351 * compatibility issue.
3352 */
3353 if (!xchg(&root->orphan_item_inserted, 1))
3354 insert = 2;
3355 else
3356 insert = 1;
3357 #endif
3358 insert = 1;
3359 atomic_inc(&root->orphan_inodes);
3360 }
3361
3362 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3363 &inode->runtime_flags))
3364 reserve = 1;
3365 spin_unlock(&root->orphan_lock);
3366
3367 /* grab metadata reservation from transaction handle */
3368 if (reserve) {
3369 ret = btrfs_orphan_reserve_metadata(trans, inode);
3370 ASSERT(!ret);
3371 if (ret) {
3372 /*
3373 * dec doesn't need spin_lock as ->orphan_block_rsv
3374 * would be released only if ->orphan_inodes is
3375 * zero.
3376 */
3377 atomic_dec(&root->orphan_inodes);
3378 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3379 &inode->runtime_flags);
3380 if (insert)
3381 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3382 &inode->runtime_flags);
3383 return ret;
3384 }
3385 }
3386
3387 /* insert an orphan item to track this unlinked/truncated file */
3388 if (insert >= 1) {
3389 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3390 if (ret) {
3391 if (reserve) {
3392 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3393 &inode->runtime_flags);
3394 btrfs_orphan_release_metadata(inode);
3395 }
3396 /*
3397 * btrfs_orphan_commit_root may race with us and set
3398 * ->orphan_block_rsv to zero, in order to avoid that,
3399 * decrease ->orphan_inodes after everything is done.
3400 */
3401 atomic_dec(&root->orphan_inodes);
3402 if (ret != -EEXIST) {
3403 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3404 &inode->runtime_flags);
3405 btrfs_abort_transaction(trans, ret);
3406 return ret;
3407 }
3408 }
3409 ret = 0;
3410 }
3411
3412 /* insert an orphan item to track subvolume contains orphan files */
3413 if (insert >= 2) {
3414 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3415 root->root_key.objectid);
3416 if (ret && ret != -EEXIST) {
3417 btrfs_abort_transaction(trans, ret);
3418 return ret;
3419 }
3420 }
3421 return 0;
3422 }
3423
3424 /*
3425 * We have done the truncate/delete so we can go ahead and remove the orphan
3426 * item for this particular inode.
3427 */
3428 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3429 struct btrfs_inode *inode)
3430 {
3431 struct btrfs_root *root = inode->root;
3432 int delete_item = 0;
3433 int ret = 0;
3434
3435 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3436 &inode->runtime_flags))
3437 delete_item = 1;
3438
3439 if (delete_item && trans)
3440 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3441
3442 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3443 &inode->runtime_flags))
3444 btrfs_orphan_release_metadata(inode);
3445
3446 /*
3447 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3448 * to zero, in order to avoid that, decrease ->orphan_inodes after
3449 * everything is done.
3450 */
3451 if (delete_item)
3452 atomic_dec(&root->orphan_inodes);
3453
3454 return ret;
3455 }
3456
3457 /*
3458 * this cleans up any orphans that may be left on the list from the last use
3459 * of this root.
3460 */
3461 int btrfs_orphan_cleanup(struct btrfs_root *root)
3462 {
3463 struct btrfs_fs_info *fs_info = root->fs_info;
3464 struct btrfs_path *path;
3465 struct extent_buffer *leaf;
3466 struct btrfs_key key, found_key;
3467 struct btrfs_trans_handle *trans;
3468 struct inode *inode;
3469 u64 last_objectid = 0;
3470 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3471
3472 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3473 return 0;
3474
3475 path = btrfs_alloc_path();
3476 if (!path) {
3477 ret = -ENOMEM;
3478 goto out;
3479 }
3480 path->reada = READA_BACK;
3481
3482 key.objectid = BTRFS_ORPHAN_OBJECTID;
3483 key.type = BTRFS_ORPHAN_ITEM_KEY;
3484 key.offset = (u64)-1;
3485
3486 while (1) {
3487 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3488 if (ret < 0)
3489 goto out;
3490
3491 /*
3492 * if ret == 0 means we found what we were searching for, which
3493 * is weird, but possible, so only screw with path if we didn't
3494 * find the key and see if we have stuff that matches
3495 */
3496 if (ret > 0) {
3497 ret = 0;
3498 if (path->slots[0] == 0)
3499 break;
3500 path->slots[0]--;
3501 }
3502
3503 /* pull out the item */
3504 leaf = path->nodes[0];
3505 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3506
3507 /* make sure the item matches what we want */
3508 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3509 break;
3510 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3511 break;
3512
3513 /* release the path since we're done with it */
3514 btrfs_release_path(path);
3515
3516 /*
3517 * this is where we are basically btrfs_lookup, without the
3518 * crossing root thing. we store the inode number in the
3519 * offset of the orphan item.
3520 */
3521
3522 if (found_key.offset == last_objectid) {
3523 btrfs_err(fs_info,
3524 "Error removing orphan entry, stopping orphan cleanup");
3525 ret = -EINVAL;
3526 goto out;
3527 }
3528
3529 last_objectid = found_key.offset;
3530
3531 found_key.objectid = found_key.offset;
3532 found_key.type = BTRFS_INODE_ITEM_KEY;
3533 found_key.offset = 0;
3534 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3535 ret = PTR_ERR_OR_ZERO(inode);
3536 if (ret && ret != -ENOENT)
3537 goto out;
3538
3539 if (ret == -ENOENT && root == fs_info->tree_root) {
3540 struct btrfs_root *dead_root;
3541 struct btrfs_fs_info *fs_info = root->fs_info;
3542 int is_dead_root = 0;
3543
3544 /*
3545 * this is an orphan in the tree root. Currently these
3546 * could come from 2 sources:
3547 * a) a snapshot deletion in progress
3548 * b) a free space cache inode
3549 * We need to distinguish those two, as the snapshot
3550 * orphan must not get deleted.
3551 * find_dead_roots already ran before us, so if this
3552 * is a snapshot deletion, we should find the root
3553 * in the dead_roots list
3554 */
3555 spin_lock(&fs_info->trans_lock);
3556 list_for_each_entry(dead_root, &fs_info->dead_roots,
3557 root_list) {
3558 if (dead_root->root_key.objectid ==
3559 found_key.objectid) {
3560 is_dead_root = 1;
3561 break;
3562 }
3563 }
3564 spin_unlock(&fs_info->trans_lock);
3565 if (is_dead_root) {
3566 /* prevent this orphan from being found again */
3567 key.offset = found_key.objectid - 1;
3568 continue;
3569 }
3570 }
3571 /*
3572 * Inode is already gone but the orphan item is still there,
3573 * kill the orphan item.
3574 */
3575 if (ret == -ENOENT) {
3576 trans = btrfs_start_transaction(root, 1);
3577 if (IS_ERR(trans)) {
3578 ret = PTR_ERR(trans);
3579 goto out;
3580 }
3581 btrfs_debug(fs_info, "auto deleting %Lu",
3582 found_key.objectid);
3583 ret = btrfs_del_orphan_item(trans, root,
3584 found_key.objectid);
3585 btrfs_end_transaction(trans);
3586 if (ret)
3587 goto out;
3588 continue;
3589 }
3590
3591 /*
3592 * add this inode to the orphan list so btrfs_orphan_del does
3593 * the proper thing when we hit it
3594 */
3595 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3596 &BTRFS_I(inode)->runtime_flags);
3597 atomic_inc(&root->orphan_inodes);
3598
3599 /* if we have links, this was a truncate, lets do that */
3600 if (inode->i_nlink) {
3601 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3602 iput(inode);
3603 continue;
3604 }
3605 nr_truncate++;
3606
3607 /* 1 for the orphan item deletion. */
3608 trans = btrfs_start_transaction(root, 1);
3609 if (IS_ERR(trans)) {
3610 iput(inode);
3611 ret = PTR_ERR(trans);
3612 goto out;
3613 }
3614 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3615 btrfs_end_transaction(trans);
3616 if (ret) {
3617 iput(inode);
3618 goto out;
3619 }
3620
3621 ret = btrfs_truncate(inode);
3622 if (ret)
3623 btrfs_orphan_del(NULL, BTRFS_I(inode));
3624 } else {
3625 nr_unlink++;
3626 }
3627
3628 /* this will do delete_inode and everything for us */
3629 iput(inode);
3630 if (ret)
3631 goto out;
3632 }
3633 /* release the path since we're done with it */
3634 btrfs_release_path(path);
3635
3636 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3637
3638 if (root->orphan_block_rsv)
3639 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3640 (u64)-1);
3641
3642 if (root->orphan_block_rsv ||
3643 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3644 trans = btrfs_join_transaction(root);
3645 if (!IS_ERR(trans))
3646 btrfs_end_transaction(trans);
3647 }
3648
3649 if (nr_unlink)
3650 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3651 if (nr_truncate)
3652 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3653
3654 out:
3655 if (ret)
3656 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3657 btrfs_free_path(path);
3658 return ret;
3659 }
3660
3661 /*
3662 * very simple check to peek ahead in the leaf looking for xattrs. If we
3663 * don't find any xattrs, we know there can't be any acls.
3664 *
3665 * slot is the slot the inode is in, objectid is the objectid of the inode
3666 */
3667 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3668 int slot, u64 objectid,
3669 int *first_xattr_slot)
3670 {
3671 u32 nritems = btrfs_header_nritems(leaf);
3672 struct btrfs_key found_key;
3673 static u64 xattr_access = 0;
3674 static u64 xattr_default = 0;
3675 int scanned = 0;
3676
3677 if (!xattr_access) {
3678 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3679 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3680 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3681 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3682 }
3683
3684 slot++;
3685 *first_xattr_slot = -1;
3686 while (slot < nritems) {
3687 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3688
3689 /* we found a different objectid, there must not be acls */
3690 if (found_key.objectid != objectid)
3691 return 0;
3692
3693 /* we found an xattr, assume we've got an acl */
3694 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3695 if (*first_xattr_slot == -1)
3696 *first_xattr_slot = slot;
3697 if (found_key.offset == xattr_access ||
3698 found_key.offset == xattr_default)
3699 return 1;
3700 }
3701
3702 /*
3703 * we found a key greater than an xattr key, there can't
3704 * be any acls later on
3705 */
3706 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3707 return 0;
3708
3709 slot++;
3710 scanned++;
3711
3712 /*
3713 * it goes inode, inode backrefs, xattrs, extents,
3714 * so if there are a ton of hard links to an inode there can
3715 * be a lot of backrefs. Don't waste time searching too hard,
3716 * this is just an optimization
3717 */
3718 if (scanned >= 8)
3719 break;
3720 }
3721 /* we hit the end of the leaf before we found an xattr or
3722 * something larger than an xattr. We have to assume the inode
3723 * has acls
3724 */
3725 if (*first_xattr_slot == -1)
3726 *first_xattr_slot = slot;
3727 return 1;
3728 }
3729
3730 /*
3731 * read an inode from the btree into the in-memory inode
3732 */
3733 static int btrfs_read_locked_inode(struct inode *inode)
3734 {
3735 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3736 struct btrfs_path *path;
3737 struct extent_buffer *leaf;
3738 struct btrfs_inode_item *inode_item;
3739 struct btrfs_root *root = BTRFS_I(inode)->root;
3740 struct btrfs_key location;
3741 unsigned long ptr;
3742 int maybe_acls;
3743 u32 rdev;
3744 int ret;
3745 bool filled = false;
3746 int first_xattr_slot;
3747
3748 ret = btrfs_fill_inode(inode, &rdev);
3749 if (!ret)
3750 filled = true;
3751
3752 path = btrfs_alloc_path();
3753 if (!path) {
3754 ret = -ENOMEM;
3755 goto make_bad;
3756 }
3757
3758 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3759
3760 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3761 if (ret) {
3762 if (ret > 0)
3763 ret = -ENOENT;
3764 goto make_bad;
3765 }
3766
3767 leaf = path->nodes[0];
3768
3769 if (filled)
3770 goto cache_index;
3771
3772 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3773 struct btrfs_inode_item);
3774 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3775 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3776 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3777 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3778 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3779
3780 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3781 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3782
3783 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3784 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3785
3786 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3787 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3788
3789 BTRFS_I(inode)->i_otime.tv_sec =
3790 btrfs_timespec_sec(leaf, &inode_item->otime);
3791 BTRFS_I(inode)->i_otime.tv_nsec =
3792 btrfs_timespec_nsec(leaf, &inode_item->otime);
3793
3794 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3795 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3796 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3797
3798 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3799 inode->i_generation = BTRFS_I(inode)->generation;
3800 inode->i_rdev = 0;
3801 rdev = btrfs_inode_rdev(leaf, inode_item);
3802
3803 BTRFS_I(inode)->index_cnt = (u64)-1;
3804 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3805
3806 cache_index:
3807 /*
3808 * If we were modified in the current generation and evicted from memory
3809 * and then re-read we need to do a full sync since we don't have any
3810 * idea about which extents were modified before we were evicted from
3811 * cache.
3812 *
3813 * This is required for both inode re-read from disk and delayed inode
3814 * in delayed_nodes_tree.
3815 */
3816 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3817 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3818 &BTRFS_I(inode)->runtime_flags);
3819
3820 /*
3821 * We don't persist the id of the transaction where an unlink operation
3822 * against the inode was last made. So here we assume the inode might
3823 * have been evicted, and therefore the exact value of last_unlink_trans
3824 * lost, and set it to last_trans to avoid metadata inconsistencies
3825 * between the inode and its parent if the inode is fsync'ed and the log
3826 * replayed. For example, in the scenario:
3827 *
3828 * touch mydir/foo
3829 * ln mydir/foo mydir/bar
3830 * sync
3831 * unlink mydir/bar
3832 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3833 * xfs_io -c fsync mydir/foo
3834 * <power failure>
3835 * mount fs, triggers fsync log replay
3836 *
3837 * We must make sure that when we fsync our inode foo we also log its
3838 * parent inode, otherwise after log replay the parent still has the
3839 * dentry with the "bar" name but our inode foo has a link count of 1
3840 * and doesn't have an inode ref with the name "bar" anymore.
3841 *
3842 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3843 * but it guarantees correctness at the expense of occasional full
3844 * transaction commits on fsync if our inode is a directory, or if our
3845 * inode is not a directory, logging its parent unnecessarily.
3846 */
3847 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3848
3849 path->slots[0]++;
3850 if (inode->i_nlink != 1 ||
3851 path->slots[0] >= btrfs_header_nritems(leaf))
3852 goto cache_acl;
3853
3854 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3855 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3856 goto cache_acl;
3857
3858 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3859 if (location.type == BTRFS_INODE_REF_KEY) {
3860 struct btrfs_inode_ref *ref;
3861
3862 ref = (struct btrfs_inode_ref *)ptr;
3863 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3864 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3865 struct btrfs_inode_extref *extref;
3866
3867 extref = (struct btrfs_inode_extref *)ptr;
3868 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3869 extref);
3870 }
3871 cache_acl:
3872 /*
3873 * try to precache a NULL acl entry for files that don't have
3874 * any xattrs or acls
3875 */
3876 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3877 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3878 if (first_xattr_slot != -1) {
3879 path->slots[0] = first_xattr_slot;
3880 ret = btrfs_load_inode_props(inode, path);
3881 if (ret)
3882 btrfs_err(fs_info,
3883 "error loading props for ino %llu (root %llu): %d",
3884 btrfs_ino(BTRFS_I(inode)),
3885 root->root_key.objectid, ret);
3886 }
3887 btrfs_free_path(path);
3888
3889 if (!maybe_acls)
3890 cache_no_acl(inode);
3891
3892 switch (inode->i_mode & S_IFMT) {
3893 case S_IFREG:
3894 inode->i_mapping->a_ops = &btrfs_aops;
3895 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3896 inode->i_fop = &btrfs_file_operations;
3897 inode->i_op = &btrfs_file_inode_operations;
3898 break;
3899 case S_IFDIR:
3900 inode->i_fop = &btrfs_dir_file_operations;
3901 inode->i_op = &btrfs_dir_inode_operations;
3902 break;
3903 case S_IFLNK:
3904 inode->i_op = &btrfs_symlink_inode_operations;
3905 inode_nohighmem(inode);
3906 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3907 break;
3908 default:
3909 inode->i_op = &btrfs_special_inode_operations;
3910 init_special_inode(inode, inode->i_mode, rdev);
3911 break;
3912 }
3913
3914 btrfs_update_iflags(inode);
3915 return 0;
3916
3917 make_bad:
3918 btrfs_free_path(path);
3919 make_bad_inode(inode);
3920 return ret;
3921 }
3922
3923 /*
3924 * given a leaf and an inode, copy the inode fields into the leaf
3925 */
3926 static void fill_inode_item(struct btrfs_trans_handle *trans,
3927 struct extent_buffer *leaf,
3928 struct btrfs_inode_item *item,
3929 struct inode *inode)
3930 {
3931 struct btrfs_map_token token;
3932
3933 btrfs_init_map_token(&token);
3934
3935 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3936 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3937 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3938 &token);
3939 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3940 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3941
3942 btrfs_set_token_timespec_sec(leaf, &item->atime,
3943 inode->i_atime.tv_sec, &token);
3944 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3945 inode->i_atime.tv_nsec, &token);
3946
3947 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3948 inode->i_mtime.tv_sec, &token);
3949 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3950 inode->i_mtime.tv_nsec, &token);
3951
3952 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3953 inode->i_ctime.tv_sec, &token);
3954 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3955 inode->i_ctime.tv_nsec, &token);
3956
3957 btrfs_set_token_timespec_sec(leaf, &item->otime,
3958 BTRFS_I(inode)->i_otime.tv_sec, &token);
3959 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3960 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3961
3962 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3963 &token);
3964 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3965 &token);
3966 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3967 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3968 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3969 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3970 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3971 }
3972
3973 /*
3974 * copy everything in the in-memory inode into the btree.
3975 */
3976 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3977 struct btrfs_root *root, struct inode *inode)
3978 {
3979 struct btrfs_inode_item *inode_item;
3980 struct btrfs_path *path;
3981 struct extent_buffer *leaf;
3982 int ret;
3983
3984 path = btrfs_alloc_path();
3985 if (!path)
3986 return -ENOMEM;
3987
3988 path->leave_spinning = 1;
3989 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3990 1);
3991 if (ret) {
3992 if (ret > 0)
3993 ret = -ENOENT;
3994 goto failed;
3995 }
3996
3997 leaf = path->nodes[0];
3998 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3999 struct btrfs_inode_item);
4000
4001 fill_inode_item(trans, leaf, inode_item, inode);
4002 btrfs_mark_buffer_dirty(leaf);
4003 btrfs_set_inode_last_trans(trans, inode);
4004 ret = 0;
4005 failed:
4006 btrfs_free_path(path);
4007 return ret;
4008 }
4009
4010 /*
4011 * copy everything in the in-memory inode into the btree.
4012 */
4013 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4014 struct btrfs_root *root, struct inode *inode)
4015 {
4016 struct btrfs_fs_info *fs_info = root->fs_info;
4017 int ret;
4018
4019 /*
4020 * If the inode is a free space inode, we can deadlock during commit
4021 * if we put it into the delayed code.
4022 *
4023 * The data relocation inode should also be directly updated
4024 * without delay
4025 */
4026 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4027 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4028 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4029 btrfs_update_root_times(trans, root);
4030
4031 ret = btrfs_delayed_update_inode(trans, root, inode);
4032 if (!ret)
4033 btrfs_set_inode_last_trans(trans, inode);
4034 return ret;
4035 }
4036
4037 return btrfs_update_inode_item(trans, root, inode);
4038 }
4039
4040 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4041 struct btrfs_root *root,
4042 struct inode *inode)
4043 {
4044 int ret;
4045
4046 ret = btrfs_update_inode(trans, root, inode);
4047 if (ret == -ENOSPC)
4048 return btrfs_update_inode_item(trans, root, inode);
4049 return ret;
4050 }
4051
4052 /*
4053 * unlink helper that gets used here in inode.c and in the tree logging
4054 * recovery code. It remove a link in a directory with a given name, and
4055 * also drops the back refs in the inode to the directory
4056 */
4057 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4058 struct btrfs_root *root,
4059 struct btrfs_inode *dir,
4060 struct btrfs_inode *inode,
4061 const char *name, int name_len)
4062 {
4063 struct btrfs_fs_info *fs_info = root->fs_info;
4064 struct btrfs_path *path;
4065 int ret = 0;
4066 struct extent_buffer *leaf;
4067 struct btrfs_dir_item *di;
4068 struct btrfs_key key;
4069 u64 index;
4070 u64 ino = btrfs_ino(inode);
4071 u64 dir_ino = btrfs_ino(dir);
4072
4073 path = btrfs_alloc_path();
4074 if (!path) {
4075 ret = -ENOMEM;
4076 goto out;
4077 }
4078
4079 path->leave_spinning = 1;
4080 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4081 name, name_len, -1);
4082 if (IS_ERR(di)) {
4083 ret = PTR_ERR(di);
4084 goto err;
4085 }
4086 if (!di) {
4087 ret = -ENOENT;
4088 goto err;
4089 }
4090 leaf = path->nodes[0];
4091 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4092 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4093 if (ret)
4094 goto err;
4095 btrfs_release_path(path);
4096
4097 /*
4098 * If we don't have dir index, we have to get it by looking up
4099 * the inode ref, since we get the inode ref, remove it directly,
4100 * it is unnecessary to do delayed deletion.
4101 *
4102 * But if we have dir index, needn't search inode ref to get it.
4103 * Since the inode ref is close to the inode item, it is better
4104 * that we delay to delete it, and just do this deletion when
4105 * we update the inode item.
4106 */
4107 if (inode->dir_index) {
4108 ret = btrfs_delayed_delete_inode_ref(inode);
4109 if (!ret) {
4110 index = inode->dir_index;
4111 goto skip_backref;
4112 }
4113 }
4114
4115 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4116 dir_ino, &index);
4117 if (ret) {
4118 btrfs_info(fs_info,
4119 "failed to delete reference to %.*s, inode %llu parent %llu",
4120 name_len, name, ino, dir_ino);
4121 btrfs_abort_transaction(trans, ret);
4122 goto err;
4123 }
4124 skip_backref:
4125 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4126 if (ret) {
4127 btrfs_abort_transaction(trans, ret);
4128 goto err;
4129 }
4130
4131 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4132 dir_ino);
4133 if (ret != 0 && ret != -ENOENT) {
4134 btrfs_abort_transaction(trans, ret);
4135 goto err;
4136 }
4137
4138 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4139 index);
4140 if (ret == -ENOENT)
4141 ret = 0;
4142 else if (ret)
4143 btrfs_abort_transaction(trans, ret);
4144 err:
4145 btrfs_free_path(path);
4146 if (ret)
4147 goto out;
4148
4149 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4150 inode_inc_iversion(&inode->vfs_inode);
4151 inode_inc_iversion(&dir->vfs_inode);
4152 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4153 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4154 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4155 out:
4156 return ret;
4157 }
4158
4159 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4160 struct btrfs_root *root,
4161 struct btrfs_inode *dir, struct btrfs_inode *inode,
4162 const char *name, int name_len)
4163 {
4164 int ret;
4165 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4166 if (!ret) {
4167 drop_nlink(&inode->vfs_inode);
4168 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4169 }
4170 return ret;
4171 }
4172
4173 /*
4174 * helper to start transaction for unlink and rmdir.
4175 *
4176 * unlink and rmdir are special in btrfs, they do not always free space, so
4177 * if we cannot make our reservations the normal way try and see if there is
4178 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4179 * allow the unlink to occur.
4180 */
4181 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4182 {
4183 struct btrfs_root *root = BTRFS_I(dir)->root;
4184
4185 /*
4186 * 1 for the possible orphan item
4187 * 1 for the dir item
4188 * 1 for the dir index
4189 * 1 for the inode ref
4190 * 1 for the inode
4191 */
4192 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4193 }
4194
4195 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4196 {
4197 struct btrfs_root *root = BTRFS_I(dir)->root;
4198 struct btrfs_trans_handle *trans;
4199 struct inode *inode = d_inode(dentry);
4200 int ret;
4201
4202 trans = __unlink_start_trans(dir);
4203 if (IS_ERR(trans))
4204 return PTR_ERR(trans);
4205
4206 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4207 0);
4208
4209 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4210 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4211 dentry->d_name.len);
4212 if (ret)
4213 goto out;
4214
4215 if (inode->i_nlink == 0) {
4216 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4217 if (ret)
4218 goto out;
4219 }
4220
4221 out:
4222 btrfs_end_transaction(trans);
4223 btrfs_btree_balance_dirty(root->fs_info);
4224 return ret;
4225 }
4226
4227 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4228 struct btrfs_root *root,
4229 struct inode *dir, u64 objectid,
4230 const char *name, int name_len)
4231 {
4232 struct btrfs_fs_info *fs_info = root->fs_info;
4233 struct btrfs_path *path;
4234 struct extent_buffer *leaf;
4235 struct btrfs_dir_item *di;
4236 struct btrfs_key key;
4237 u64 index;
4238 int ret;
4239 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4240
4241 path = btrfs_alloc_path();
4242 if (!path)
4243 return -ENOMEM;
4244
4245 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4246 name, name_len, -1);
4247 if (IS_ERR_OR_NULL(di)) {
4248 if (!di)
4249 ret = -ENOENT;
4250 else
4251 ret = PTR_ERR(di);
4252 goto out;
4253 }
4254
4255 leaf = path->nodes[0];
4256 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4257 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4258 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4259 if (ret) {
4260 btrfs_abort_transaction(trans, ret);
4261 goto out;
4262 }
4263 btrfs_release_path(path);
4264
4265 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4266 root->root_key.objectid, dir_ino,
4267 &index, name, name_len);
4268 if (ret < 0) {
4269 if (ret != -ENOENT) {
4270 btrfs_abort_transaction(trans, ret);
4271 goto out;
4272 }
4273 di = btrfs_search_dir_index_item(root, path, dir_ino,
4274 name, name_len);
4275 if (IS_ERR_OR_NULL(di)) {
4276 if (!di)
4277 ret = -ENOENT;
4278 else
4279 ret = PTR_ERR(di);
4280 btrfs_abort_transaction(trans, ret);
4281 goto out;
4282 }
4283
4284 leaf = path->nodes[0];
4285 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4286 btrfs_release_path(path);
4287 index = key.offset;
4288 }
4289 btrfs_release_path(path);
4290
4291 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4292 if (ret) {
4293 btrfs_abort_transaction(trans, ret);
4294 goto out;
4295 }
4296
4297 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4298 inode_inc_iversion(dir);
4299 dir->i_mtime = dir->i_ctime = current_time(dir);
4300 ret = btrfs_update_inode_fallback(trans, root, dir);
4301 if (ret)
4302 btrfs_abort_transaction(trans, ret);
4303 out:
4304 btrfs_free_path(path);
4305 return ret;
4306 }
4307
4308 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4309 {
4310 struct inode *inode = d_inode(dentry);
4311 int err = 0;
4312 struct btrfs_root *root = BTRFS_I(dir)->root;
4313 struct btrfs_trans_handle *trans;
4314 u64 last_unlink_trans;
4315
4316 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4317 return -ENOTEMPTY;
4318 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4319 return -EPERM;
4320
4321 trans = __unlink_start_trans(dir);
4322 if (IS_ERR(trans))
4323 return PTR_ERR(trans);
4324
4325 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4326 err = btrfs_unlink_subvol(trans, root, dir,
4327 BTRFS_I(inode)->location.objectid,
4328 dentry->d_name.name,
4329 dentry->d_name.len);
4330 goto out;
4331 }
4332
4333 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4334 if (err)
4335 goto out;
4336
4337 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4338
4339 /* now the directory is empty */
4340 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4341 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4342 dentry->d_name.len);
4343 if (!err) {
4344 btrfs_i_size_write(BTRFS_I(inode), 0);
4345 /*
4346 * Propagate the last_unlink_trans value of the deleted dir to
4347 * its parent directory. This is to prevent an unrecoverable
4348 * log tree in the case we do something like this:
4349 * 1) create dir foo
4350 * 2) create snapshot under dir foo
4351 * 3) delete the snapshot
4352 * 4) rmdir foo
4353 * 5) mkdir foo
4354 * 6) fsync foo or some file inside foo
4355 */
4356 if (last_unlink_trans >= trans->transid)
4357 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4358 }
4359 out:
4360 btrfs_end_transaction(trans);
4361 btrfs_btree_balance_dirty(root->fs_info);
4362
4363 return err;
4364 }
4365
4366 static int truncate_space_check(struct btrfs_trans_handle *trans,
4367 struct btrfs_root *root,
4368 u64 bytes_deleted)
4369 {
4370 struct btrfs_fs_info *fs_info = root->fs_info;
4371 int ret;
4372
4373 /*
4374 * This is only used to apply pressure to the enospc system, we don't
4375 * intend to use this reservation at all.
4376 */
4377 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4378 bytes_deleted *= fs_info->nodesize;
4379 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4380 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4381 if (!ret) {
4382 trace_btrfs_space_reservation(fs_info, "transaction",
4383 trans->transid,
4384 bytes_deleted, 1);
4385 trans->bytes_reserved += bytes_deleted;
4386 }
4387 return ret;
4388
4389 }
4390
4391 /*
4392 * Return this if we need to call truncate_block for the last bit of the
4393 * truncate.
4394 */
4395 #define NEED_TRUNCATE_BLOCK 1
4396
4397 /*
4398 * this can truncate away extent items, csum items and directory items.
4399 * It starts at a high offset and removes keys until it can't find
4400 * any higher than new_size
4401 *
4402 * csum items that cross the new i_size are truncated to the new size
4403 * as well.
4404 *
4405 * min_type is the minimum key type to truncate down to. If set to 0, this
4406 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4407 */
4408 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4409 struct btrfs_root *root,
4410 struct inode *inode,
4411 u64 new_size, u32 min_type)
4412 {
4413 struct btrfs_fs_info *fs_info = root->fs_info;
4414 struct btrfs_path *path;
4415 struct extent_buffer *leaf;
4416 struct btrfs_file_extent_item *fi;
4417 struct btrfs_key key;
4418 struct btrfs_key found_key;
4419 u64 extent_start = 0;
4420 u64 extent_num_bytes = 0;
4421 u64 extent_offset = 0;
4422 u64 item_end = 0;
4423 u64 last_size = new_size;
4424 u32 found_type = (u8)-1;
4425 int found_extent;
4426 int del_item;
4427 int pending_del_nr = 0;
4428 int pending_del_slot = 0;
4429 int extent_type = -1;
4430 int ret;
4431 int err = 0;
4432 u64 ino = btrfs_ino(BTRFS_I(inode));
4433 u64 bytes_deleted = 0;
4434 bool be_nice = false;
4435 bool should_throttle = false;
4436 bool should_end = false;
4437
4438 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4439
4440 /*
4441 * for non-free space inodes and ref cows, we want to back off from
4442 * time to time
4443 */
4444 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4445 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4446 be_nice = true;
4447
4448 path = btrfs_alloc_path();
4449 if (!path)
4450 return -ENOMEM;
4451 path->reada = READA_BACK;
4452
4453 /*
4454 * We want to drop from the next block forward in case this new size is
4455 * not block aligned since we will be keeping the last block of the
4456 * extent just the way it is.
4457 */
4458 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4459 root == fs_info->tree_root)
4460 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4461 fs_info->sectorsize),
4462 (u64)-1, 0);
4463
4464 /*
4465 * This function is also used to drop the items in the log tree before
4466 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4467 * it is used to drop the loged items. So we shouldn't kill the delayed
4468 * items.
4469 */
4470 if (min_type == 0 && root == BTRFS_I(inode)->root)
4471 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4472
4473 key.objectid = ino;
4474 key.offset = (u64)-1;
4475 key.type = (u8)-1;
4476
4477 search_again:
4478 /*
4479 * with a 16K leaf size and 128MB extents, you can actually queue
4480 * up a huge file in a single leaf. Most of the time that
4481 * bytes_deleted is > 0, it will be huge by the time we get here
4482 */
4483 if (be_nice && bytes_deleted > SZ_32M) {
4484 if (btrfs_should_end_transaction(trans)) {
4485 err = -EAGAIN;
4486 goto error;
4487 }
4488 }
4489
4490
4491 path->leave_spinning = 1;
4492 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4493 if (ret < 0) {
4494 err = ret;
4495 goto out;
4496 }
4497
4498 if (ret > 0) {
4499 /* there are no items in the tree for us to truncate, we're
4500 * done
4501 */
4502 if (path->slots[0] == 0)
4503 goto out;
4504 path->slots[0]--;
4505 }
4506
4507 while (1) {
4508 fi = NULL;
4509 leaf = path->nodes[0];
4510 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4511 found_type = found_key.type;
4512
4513 if (found_key.objectid != ino)
4514 break;
4515
4516 if (found_type < min_type)
4517 break;
4518
4519 item_end = found_key.offset;
4520 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4521 fi = btrfs_item_ptr(leaf, path->slots[0],
4522 struct btrfs_file_extent_item);
4523 extent_type = btrfs_file_extent_type(leaf, fi);
4524 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4525 item_end +=
4526 btrfs_file_extent_num_bytes(leaf, fi);
4527
4528 trace_btrfs_truncate_show_fi_regular(
4529 BTRFS_I(inode), leaf, fi,
4530 found_key.offset);
4531 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4532 item_end += btrfs_file_extent_inline_len(leaf,
4533 path->slots[0], fi);
4534
4535 trace_btrfs_truncate_show_fi_inline(
4536 BTRFS_I(inode), leaf, fi, path->slots[0],
4537 found_key.offset);
4538 }
4539 item_end--;
4540 }
4541 if (found_type > min_type) {
4542 del_item = 1;
4543 } else {
4544 if (item_end < new_size)
4545 break;
4546 if (found_key.offset >= new_size)
4547 del_item = 1;
4548 else
4549 del_item = 0;
4550 }
4551 found_extent = 0;
4552 /* FIXME, shrink the extent if the ref count is only 1 */
4553 if (found_type != BTRFS_EXTENT_DATA_KEY)
4554 goto delete;
4555
4556 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4557 u64 num_dec;
4558 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4559 if (!del_item) {
4560 u64 orig_num_bytes =
4561 btrfs_file_extent_num_bytes(leaf, fi);
4562 extent_num_bytes = ALIGN(new_size -
4563 found_key.offset,
4564 fs_info->sectorsize);
4565 btrfs_set_file_extent_num_bytes(leaf, fi,
4566 extent_num_bytes);
4567 num_dec = (orig_num_bytes -
4568 extent_num_bytes);
4569 if (test_bit(BTRFS_ROOT_REF_COWS,
4570 &root->state) &&
4571 extent_start != 0)
4572 inode_sub_bytes(inode, num_dec);
4573 btrfs_mark_buffer_dirty(leaf);
4574 } else {
4575 extent_num_bytes =
4576 btrfs_file_extent_disk_num_bytes(leaf,
4577 fi);
4578 extent_offset = found_key.offset -
4579 btrfs_file_extent_offset(leaf, fi);
4580
4581 /* FIXME blocksize != 4096 */
4582 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4583 if (extent_start != 0) {
4584 found_extent = 1;
4585 if (test_bit(BTRFS_ROOT_REF_COWS,
4586 &root->state))
4587 inode_sub_bytes(inode, num_dec);
4588 }
4589 }
4590 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4591 /*
4592 * we can't truncate inline items that have had
4593 * special encodings
4594 */
4595 if (!del_item &&
4596 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4597 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4598 btrfs_file_extent_compression(leaf, fi) == 0) {
4599 u32 size = (u32)(new_size - found_key.offset);
4600
4601 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4602 size = btrfs_file_extent_calc_inline_size(size);
4603 btrfs_truncate_item(root->fs_info, path, size, 1);
4604 } else if (!del_item) {
4605 /*
4606 * We have to bail so the last_size is set to
4607 * just before this extent.
4608 */
4609 err = NEED_TRUNCATE_BLOCK;
4610 break;
4611 }
4612
4613 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4614 inode_sub_bytes(inode, item_end + 1 - new_size);
4615 }
4616 delete:
4617 if (del_item)
4618 last_size = found_key.offset;
4619 else
4620 last_size = new_size;
4621 if (del_item) {
4622 if (!pending_del_nr) {
4623 /* no pending yet, add ourselves */
4624 pending_del_slot = path->slots[0];
4625 pending_del_nr = 1;
4626 } else if (pending_del_nr &&
4627 path->slots[0] + 1 == pending_del_slot) {
4628 /* hop on the pending chunk */
4629 pending_del_nr++;
4630 pending_del_slot = path->slots[0];
4631 } else {
4632 BUG();
4633 }
4634 } else {
4635 break;
4636 }
4637 should_throttle = false;
4638
4639 if (found_extent &&
4640 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4641 root == fs_info->tree_root)) {
4642 btrfs_set_path_blocking(path);
4643 bytes_deleted += extent_num_bytes;
4644 ret = btrfs_free_extent(trans, root, extent_start,
4645 extent_num_bytes, 0,
4646 btrfs_header_owner(leaf),
4647 ino, extent_offset);
4648 BUG_ON(ret);
4649 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4650 btrfs_async_run_delayed_refs(fs_info,
4651 trans->delayed_ref_updates * 2,
4652 trans->transid, 0);
4653 if (be_nice) {
4654 if (truncate_space_check(trans, root,
4655 extent_num_bytes)) {
4656 should_end = true;
4657 }
4658 if (btrfs_should_throttle_delayed_refs(trans,
4659 fs_info))
4660 should_throttle = true;
4661 }
4662 }
4663
4664 if (found_type == BTRFS_INODE_ITEM_KEY)
4665 break;
4666
4667 if (path->slots[0] == 0 ||
4668 path->slots[0] != pending_del_slot ||
4669 should_throttle || should_end) {
4670 if (pending_del_nr) {
4671 ret = btrfs_del_items(trans, root, path,
4672 pending_del_slot,
4673 pending_del_nr);
4674 if (ret) {
4675 btrfs_abort_transaction(trans, ret);
4676 goto error;
4677 }
4678 pending_del_nr = 0;
4679 }
4680 btrfs_release_path(path);
4681 if (should_throttle) {
4682 unsigned long updates = trans->delayed_ref_updates;
4683 if (updates) {
4684 trans->delayed_ref_updates = 0;
4685 ret = btrfs_run_delayed_refs(trans,
4686 fs_info,
4687 updates * 2);
4688 if (ret && !err)
4689 err = ret;
4690 }
4691 }
4692 /*
4693 * if we failed to refill our space rsv, bail out
4694 * and let the transaction restart
4695 */
4696 if (should_end) {
4697 err = -EAGAIN;
4698 goto error;
4699 }
4700 goto search_again;
4701 } else {
4702 path->slots[0]--;
4703 }
4704 }
4705 out:
4706 if (pending_del_nr) {
4707 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4708 pending_del_nr);
4709 if (ret)
4710 btrfs_abort_transaction(trans, ret);
4711 }
4712 error:
4713 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4714 ASSERT(last_size >= new_size);
4715 if (!err && last_size > new_size)
4716 last_size = new_size;
4717 btrfs_ordered_update_i_size(inode, last_size, NULL);
4718 }
4719
4720 btrfs_free_path(path);
4721
4722 if (be_nice && bytes_deleted > SZ_32M) {
4723 unsigned long updates = trans->delayed_ref_updates;
4724 if (updates) {
4725 trans->delayed_ref_updates = 0;
4726 ret = btrfs_run_delayed_refs(trans, fs_info,
4727 updates * 2);
4728 if (ret && !err)
4729 err = ret;
4730 }
4731 }
4732 return err;
4733 }
4734
4735 /*
4736 * btrfs_truncate_block - read, zero a chunk and write a block
4737 * @inode - inode that we're zeroing
4738 * @from - the offset to start zeroing
4739 * @len - the length to zero, 0 to zero the entire range respective to the
4740 * offset
4741 * @front - zero up to the offset instead of from the offset on
4742 *
4743 * This will find the block for the "from" offset and cow the block and zero the
4744 * part we want to zero. This is used with truncate and hole punching.
4745 */
4746 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4747 int front)
4748 {
4749 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4750 struct address_space *mapping = inode->i_mapping;
4751 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4752 struct btrfs_ordered_extent *ordered;
4753 struct extent_state *cached_state = NULL;
4754 struct extent_changeset *data_reserved = NULL;
4755 char *kaddr;
4756 u32 blocksize = fs_info->sectorsize;
4757 pgoff_t index = from >> PAGE_SHIFT;
4758 unsigned offset = from & (blocksize - 1);
4759 struct page *page;
4760 gfp_t mask = btrfs_alloc_write_mask(mapping);
4761 int ret = 0;
4762 u64 block_start;
4763 u64 block_end;
4764
4765 if ((offset & (blocksize - 1)) == 0 &&
4766 (!len || ((len & (blocksize - 1)) == 0)))
4767 goto out;
4768
4769 block_start = round_down(from, blocksize);
4770 block_end = block_start + blocksize - 1;
4771
4772 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4773 block_start, blocksize);
4774 if (ret)
4775 goto out;
4776
4777 again:
4778 page = find_or_create_page(mapping, index, mask);
4779 if (!page) {
4780 btrfs_delalloc_release_space(inode, data_reserved,
4781 block_start, blocksize);
4782 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4783 ret = -ENOMEM;
4784 goto out;
4785 }
4786
4787 if (!PageUptodate(page)) {
4788 ret = btrfs_readpage(NULL, page);
4789 lock_page(page);
4790 if (page->mapping != mapping) {
4791 unlock_page(page);
4792 put_page(page);
4793 goto again;
4794 }
4795 if (!PageUptodate(page)) {
4796 ret = -EIO;
4797 goto out_unlock;
4798 }
4799 }
4800 wait_on_page_writeback(page);
4801
4802 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4803 set_page_extent_mapped(page);
4804
4805 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4806 if (ordered) {
4807 unlock_extent_cached(io_tree, block_start, block_end,
4808 &cached_state, GFP_NOFS);
4809 unlock_page(page);
4810 put_page(page);
4811 btrfs_start_ordered_extent(inode, ordered, 1);
4812 btrfs_put_ordered_extent(ordered);
4813 goto again;
4814 }
4815
4816 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4817 EXTENT_DIRTY | EXTENT_DELALLOC |
4818 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4819 0, 0, &cached_state, GFP_NOFS);
4820
4821 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4822 &cached_state, 0);
4823 if (ret) {
4824 unlock_extent_cached(io_tree, block_start, block_end,
4825 &cached_state, GFP_NOFS);
4826 goto out_unlock;
4827 }
4828
4829 if (offset != blocksize) {
4830 if (!len)
4831 len = blocksize - offset;
4832 kaddr = kmap(page);
4833 if (front)
4834 memset(kaddr + (block_start - page_offset(page)),
4835 0, offset);
4836 else
4837 memset(kaddr + (block_start - page_offset(page)) + offset,
4838 0, len);
4839 flush_dcache_page(page);
4840 kunmap(page);
4841 }
4842 ClearPageChecked(page);
4843 set_page_dirty(page);
4844 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4845 GFP_NOFS);
4846
4847 out_unlock:
4848 if (ret)
4849 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4850 blocksize);
4851 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4852 unlock_page(page);
4853 put_page(page);
4854 out:
4855 extent_changeset_free(data_reserved);
4856 return ret;
4857 }
4858
4859 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4860 u64 offset, u64 len)
4861 {
4862 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4863 struct btrfs_trans_handle *trans;
4864 int ret;
4865
4866 /*
4867 * Still need to make sure the inode looks like it's been updated so
4868 * that any holes get logged if we fsync.
4869 */
4870 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4871 BTRFS_I(inode)->last_trans = fs_info->generation;
4872 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4873 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4874 return 0;
4875 }
4876
4877 /*
4878 * 1 - for the one we're dropping
4879 * 1 - for the one we're adding
4880 * 1 - for updating the inode.
4881 */
4882 trans = btrfs_start_transaction(root, 3);
4883 if (IS_ERR(trans))
4884 return PTR_ERR(trans);
4885
4886 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4887 if (ret) {
4888 btrfs_abort_transaction(trans, ret);
4889 btrfs_end_transaction(trans);
4890 return ret;
4891 }
4892
4893 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4894 offset, 0, 0, len, 0, len, 0, 0, 0);
4895 if (ret)
4896 btrfs_abort_transaction(trans, ret);
4897 else
4898 btrfs_update_inode(trans, root, inode);
4899 btrfs_end_transaction(trans);
4900 return ret;
4901 }
4902
4903 /*
4904 * This function puts in dummy file extents for the area we're creating a hole
4905 * for. So if we are truncating this file to a larger size we need to insert
4906 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4907 * the range between oldsize and size
4908 */
4909 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4910 {
4911 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4912 struct btrfs_root *root = BTRFS_I(inode)->root;
4913 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4914 struct extent_map *em = NULL;
4915 struct extent_state *cached_state = NULL;
4916 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4917 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4918 u64 block_end = ALIGN(size, fs_info->sectorsize);
4919 u64 last_byte;
4920 u64 cur_offset;
4921 u64 hole_size;
4922 int err = 0;
4923
4924 /*
4925 * If our size started in the middle of a block we need to zero out the
4926 * rest of the block before we expand the i_size, otherwise we could
4927 * expose stale data.
4928 */
4929 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4930 if (err)
4931 return err;
4932
4933 if (size <= hole_start)
4934 return 0;
4935
4936 while (1) {
4937 struct btrfs_ordered_extent *ordered;
4938
4939 lock_extent_bits(io_tree, hole_start, block_end - 1,
4940 &cached_state);
4941 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4942 block_end - hole_start);
4943 if (!ordered)
4944 break;
4945 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4946 &cached_state, GFP_NOFS);
4947 btrfs_start_ordered_extent(inode, ordered, 1);
4948 btrfs_put_ordered_extent(ordered);
4949 }
4950
4951 cur_offset = hole_start;
4952 while (1) {
4953 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4954 block_end - cur_offset, 0);
4955 if (IS_ERR(em)) {
4956 err = PTR_ERR(em);
4957 em = NULL;
4958 break;
4959 }
4960 last_byte = min(extent_map_end(em), block_end);
4961 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4962 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4963 struct extent_map *hole_em;
4964 hole_size = last_byte - cur_offset;
4965
4966 err = maybe_insert_hole(root, inode, cur_offset,
4967 hole_size);
4968 if (err)
4969 break;
4970 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4971 cur_offset + hole_size - 1, 0);
4972 hole_em = alloc_extent_map();
4973 if (!hole_em) {
4974 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4975 &BTRFS_I(inode)->runtime_flags);
4976 goto next;
4977 }
4978 hole_em->start = cur_offset;
4979 hole_em->len = hole_size;
4980 hole_em->orig_start = cur_offset;
4981
4982 hole_em->block_start = EXTENT_MAP_HOLE;
4983 hole_em->block_len = 0;
4984 hole_em->orig_block_len = 0;
4985 hole_em->ram_bytes = hole_size;
4986 hole_em->bdev = fs_info->fs_devices->latest_bdev;
4987 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4988 hole_em->generation = fs_info->generation;
4989
4990 while (1) {
4991 write_lock(&em_tree->lock);
4992 err = add_extent_mapping(em_tree, hole_em, 1);
4993 write_unlock(&em_tree->lock);
4994 if (err != -EEXIST)
4995 break;
4996 btrfs_drop_extent_cache(BTRFS_I(inode),
4997 cur_offset,
4998 cur_offset +
4999 hole_size - 1, 0);
5000 }
5001 free_extent_map(hole_em);
5002 }
5003 next:
5004 free_extent_map(em);
5005 em = NULL;
5006 cur_offset = last_byte;
5007 if (cur_offset >= block_end)
5008 break;
5009 }
5010 free_extent_map(em);
5011 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5012 GFP_NOFS);
5013 return err;
5014 }
5015
5016 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5017 {
5018 struct btrfs_root *root = BTRFS_I(inode)->root;
5019 struct btrfs_trans_handle *trans;
5020 loff_t oldsize = i_size_read(inode);
5021 loff_t newsize = attr->ia_size;
5022 int mask = attr->ia_valid;
5023 int ret;
5024
5025 /*
5026 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5027 * special case where we need to update the times despite not having
5028 * these flags set. For all other operations the VFS set these flags
5029 * explicitly if it wants a timestamp update.
5030 */
5031 if (newsize != oldsize) {
5032 inode_inc_iversion(inode);
5033 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5034 inode->i_ctime = inode->i_mtime =
5035 current_time(inode);
5036 }
5037
5038 if (newsize > oldsize) {
5039 /*
5040 * Don't do an expanding truncate while snapshotting is ongoing.
5041 * This is to ensure the snapshot captures a fully consistent
5042 * state of this file - if the snapshot captures this expanding
5043 * truncation, it must capture all writes that happened before
5044 * this truncation.
5045 */
5046 btrfs_wait_for_snapshot_creation(root);
5047 ret = btrfs_cont_expand(inode, oldsize, newsize);
5048 if (ret) {
5049 btrfs_end_write_no_snapshotting(root);
5050 return ret;
5051 }
5052
5053 trans = btrfs_start_transaction(root, 1);
5054 if (IS_ERR(trans)) {
5055 btrfs_end_write_no_snapshotting(root);
5056 return PTR_ERR(trans);
5057 }
5058
5059 i_size_write(inode, newsize);
5060 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5061 pagecache_isize_extended(inode, oldsize, newsize);
5062 ret = btrfs_update_inode(trans, root, inode);
5063 btrfs_end_write_no_snapshotting(root);
5064 btrfs_end_transaction(trans);
5065 } else {
5066
5067 /*
5068 * We're truncating a file that used to have good data down to
5069 * zero. Make sure it gets into the ordered flush list so that
5070 * any new writes get down to disk quickly.
5071 */
5072 if (newsize == 0)
5073 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5074 &BTRFS_I(inode)->runtime_flags);
5075
5076 /*
5077 * 1 for the orphan item we're going to add
5078 * 1 for the orphan item deletion.
5079 */
5080 trans = btrfs_start_transaction(root, 2);
5081 if (IS_ERR(trans))
5082 return PTR_ERR(trans);
5083
5084 /*
5085 * We need to do this in case we fail at _any_ point during the
5086 * actual truncate. Once we do the truncate_setsize we could
5087 * invalidate pages which forces any outstanding ordered io to
5088 * be instantly completed which will give us extents that need
5089 * to be truncated. If we fail to get an orphan inode down we
5090 * could have left over extents that were never meant to live,
5091 * so we need to guarantee from this point on that everything
5092 * will be consistent.
5093 */
5094 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5095 btrfs_end_transaction(trans);
5096 if (ret)
5097 return ret;
5098
5099 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5100 truncate_setsize(inode, newsize);
5101
5102 /* Disable nonlocked read DIO to avoid the end less truncate */
5103 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5104 inode_dio_wait(inode);
5105 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5106
5107 ret = btrfs_truncate(inode);
5108 if (ret && inode->i_nlink) {
5109 int err;
5110
5111 /* To get a stable disk_i_size */
5112 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5113 if (err) {
5114 btrfs_orphan_del(NULL, BTRFS_I(inode));
5115 return err;
5116 }
5117
5118 /*
5119 * failed to truncate, disk_i_size is only adjusted down
5120 * as we remove extents, so it should represent the true
5121 * size of the inode, so reset the in memory size and
5122 * delete our orphan entry.
5123 */
5124 trans = btrfs_join_transaction(root);
5125 if (IS_ERR(trans)) {
5126 btrfs_orphan_del(NULL, BTRFS_I(inode));
5127 return ret;
5128 }
5129 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5130 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5131 if (err)
5132 btrfs_abort_transaction(trans, err);
5133 btrfs_end_transaction(trans);
5134 }
5135 }
5136
5137 return ret;
5138 }
5139
5140 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5141 {
5142 struct inode *inode = d_inode(dentry);
5143 struct btrfs_root *root = BTRFS_I(inode)->root;
5144 int err;
5145
5146 if (btrfs_root_readonly(root))
5147 return -EROFS;
5148
5149 err = setattr_prepare(dentry, attr);
5150 if (err)
5151 return err;
5152
5153 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5154 err = btrfs_setsize(inode, attr);
5155 if (err)
5156 return err;
5157 }
5158
5159 if (attr->ia_valid) {
5160 setattr_copy(inode, attr);
5161 inode_inc_iversion(inode);
5162 err = btrfs_dirty_inode(inode);
5163
5164 if (!err && attr->ia_valid & ATTR_MODE)
5165 err = posix_acl_chmod(inode, inode->i_mode);
5166 }
5167
5168 return err;
5169 }
5170
5171 /*
5172 * While truncating the inode pages during eviction, we get the VFS calling
5173 * btrfs_invalidatepage() against each page of the inode. This is slow because
5174 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5175 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5176 * extent_state structures over and over, wasting lots of time.
5177 *
5178 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5179 * those expensive operations on a per page basis and do only the ordered io
5180 * finishing, while we release here the extent_map and extent_state structures,
5181 * without the excessive merging and splitting.
5182 */
5183 static void evict_inode_truncate_pages(struct inode *inode)
5184 {
5185 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5186 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5187 struct rb_node *node;
5188
5189 ASSERT(inode->i_state & I_FREEING);
5190 truncate_inode_pages_final(&inode->i_data);
5191
5192 write_lock(&map_tree->lock);
5193 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5194 struct extent_map *em;
5195
5196 node = rb_first(&map_tree->map);
5197 em = rb_entry(node, struct extent_map, rb_node);
5198 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5199 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5200 remove_extent_mapping(map_tree, em);
5201 free_extent_map(em);
5202 if (need_resched()) {
5203 write_unlock(&map_tree->lock);
5204 cond_resched();
5205 write_lock(&map_tree->lock);
5206 }
5207 }
5208 write_unlock(&map_tree->lock);
5209
5210 /*
5211 * Keep looping until we have no more ranges in the io tree.
5212 * We can have ongoing bios started by readpages (called from readahead)
5213 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5214 * still in progress (unlocked the pages in the bio but did not yet
5215 * unlocked the ranges in the io tree). Therefore this means some
5216 * ranges can still be locked and eviction started because before
5217 * submitting those bios, which are executed by a separate task (work
5218 * queue kthread), inode references (inode->i_count) were not taken
5219 * (which would be dropped in the end io callback of each bio).
5220 * Therefore here we effectively end up waiting for those bios and
5221 * anyone else holding locked ranges without having bumped the inode's
5222 * reference count - if we don't do it, when they access the inode's
5223 * io_tree to unlock a range it may be too late, leading to an
5224 * use-after-free issue.
5225 */
5226 spin_lock(&io_tree->lock);
5227 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5228 struct extent_state *state;
5229 struct extent_state *cached_state = NULL;
5230 u64 start;
5231 u64 end;
5232
5233 node = rb_first(&io_tree->state);
5234 state = rb_entry(node, struct extent_state, rb_node);
5235 start = state->start;
5236 end = state->end;
5237 spin_unlock(&io_tree->lock);
5238
5239 lock_extent_bits(io_tree, start, end, &cached_state);
5240
5241 /*
5242 * If still has DELALLOC flag, the extent didn't reach disk,
5243 * and its reserved space won't be freed by delayed_ref.
5244 * So we need to free its reserved space here.
5245 * (Refer to comment in btrfs_invalidatepage, case 2)
5246 *
5247 * Note, end is the bytenr of last byte, so we need + 1 here.
5248 */
5249 if (state->state & EXTENT_DELALLOC)
5250 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5251
5252 clear_extent_bit(io_tree, start, end,
5253 EXTENT_LOCKED | EXTENT_DIRTY |
5254 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5255 EXTENT_DEFRAG, 1, 1,
5256 &cached_state, GFP_NOFS);
5257
5258 cond_resched();
5259 spin_lock(&io_tree->lock);
5260 }
5261 spin_unlock(&io_tree->lock);
5262 }
5263
5264 void btrfs_evict_inode(struct inode *inode)
5265 {
5266 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5267 struct btrfs_trans_handle *trans;
5268 struct btrfs_root *root = BTRFS_I(inode)->root;
5269 struct btrfs_block_rsv *rsv, *global_rsv;
5270 int steal_from_global = 0;
5271 u64 min_size;
5272 int ret;
5273
5274 trace_btrfs_inode_evict(inode);
5275
5276 if (!root) {
5277 clear_inode(inode);
5278 return;
5279 }
5280
5281 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5282
5283 evict_inode_truncate_pages(inode);
5284
5285 if (inode->i_nlink &&
5286 ((btrfs_root_refs(&root->root_item) != 0 &&
5287 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5288 btrfs_is_free_space_inode(BTRFS_I(inode))))
5289 goto no_delete;
5290
5291 if (is_bad_inode(inode)) {
5292 btrfs_orphan_del(NULL, BTRFS_I(inode));
5293 goto no_delete;
5294 }
5295 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5296 if (!special_file(inode->i_mode))
5297 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5298
5299 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5300
5301 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5302 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5303 &BTRFS_I(inode)->runtime_flags));
5304 goto no_delete;
5305 }
5306
5307 if (inode->i_nlink > 0) {
5308 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5309 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5310 goto no_delete;
5311 }
5312
5313 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5314 if (ret) {
5315 btrfs_orphan_del(NULL, BTRFS_I(inode));
5316 goto no_delete;
5317 }
5318
5319 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5320 if (!rsv) {
5321 btrfs_orphan_del(NULL, BTRFS_I(inode));
5322 goto no_delete;
5323 }
5324 rsv->size = min_size;
5325 rsv->failfast = 1;
5326 global_rsv = &fs_info->global_block_rsv;
5327
5328 btrfs_i_size_write(BTRFS_I(inode), 0);
5329
5330 /*
5331 * This is a bit simpler than btrfs_truncate since we've already
5332 * reserved our space for our orphan item in the unlink, so we just
5333 * need to reserve some slack space in case we add bytes and update
5334 * inode item when doing the truncate.
5335 */
5336 while (1) {
5337 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5338 BTRFS_RESERVE_FLUSH_LIMIT);
5339
5340 /*
5341 * Try and steal from the global reserve since we will
5342 * likely not use this space anyway, we want to try as
5343 * hard as possible to get this to work.
5344 */
5345 if (ret)
5346 steal_from_global++;
5347 else
5348 steal_from_global = 0;
5349 ret = 0;
5350
5351 /*
5352 * steal_from_global == 0: we reserved stuff, hooray!
5353 * steal_from_global == 1: we didn't reserve stuff, boo!
5354 * steal_from_global == 2: we've committed, still not a lot of
5355 * room but maybe we'll have room in the global reserve this
5356 * time.
5357 * steal_from_global == 3: abandon all hope!
5358 */
5359 if (steal_from_global > 2) {
5360 btrfs_warn(fs_info,
5361 "Could not get space for a delete, will truncate on mount %d",
5362 ret);
5363 btrfs_orphan_del(NULL, BTRFS_I(inode));
5364 btrfs_free_block_rsv(fs_info, rsv);
5365 goto no_delete;
5366 }
5367
5368 trans = btrfs_join_transaction(root);
5369 if (IS_ERR(trans)) {
5370 btrfs_orphan_del(NULL, BTRFS_I(inode));
5371 btrfs_free_block_rsv(fs_info, rsv);
5372 goto no_delete;
5373 }
5374
5375 /*
5376 * We can't just steal from the global reserve, we need to make
5377 * sure there is room to do it, if not we need to commit and try
5378 * again.
5379 */
5380 if (steal_from_global) {
5381 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5382 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5383 min_size, 0);
5384 else
5385 ret = -ENOSPC;
5386 }
5387
5388 /*
5389 * Couldn't steal from the global reserve, we have too much
5390 * pending stuff built up, commit the transaction and try it
5391 * again.
5392 */
5393 if (ret) {
5394 ret = btrfs_commit_transaction(trans);
5395 if (ret) {
5396 btrfs_orphan_del(NULL, BTRFS_I(inode));
5397 btrfs_free_block_rsv(fs_info, rsv);
5398 goto no_delete;
5399 }
5400 continue;
5401 } else {
5402 steal_from_global = 0;
5403 }
5404
5405 trans->block_rsv = rsv;
5406
5407 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5408 if (ret != -ENOSPC && ret != -EAGAIN)
5409 break;
5410
5411 trans->block_rsv = &fs_info->trans_block_rsv;
5412 btrfs_end_transaction(trans);
5413 trans = NULL;
5414 btrfs_btree_balance_dirty(fs_info);
5415 }
5416
5417 btrfs_free_block_rsv(fs_info, rsv);
5418
5419 /*
5420 * Errors here aren't a big deal, it just means we leave orphan items
5421 * in the tree. They will be cleaned up on the next mount.
5422 */
5423 if (ret == 0) {
5424 trans->block_rsv = root->orphan_block_rsv;
5425 btrfs_orphan_del(trans, BTRFS_I(inode));
5426 } else {
5427 btrfs_orphan_del(NULL, BTRFS_I(inode));
5428 }
5429
5430 trans->block_rsv = &fs_info->trans_block_rsv;
5431 if (!(root == fs_info->tree_root ||
5432 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5433 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5434
5435 btrfs_end_transaction(trans);
5436 btrfs_btree_balance_dirty(fs_info);
5437 no_delete:
5438 btrfs_remove_delayed_node(BTRFS_I(inode));
5439 clear_inode(inode);
5440 }
5441
5442 /*
5443 * this returns the key found in the dir entry in the location pointer.
5444 * If no dir entries were found, location->objectid is 0.
5445 */
5446 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5447 struct btrfs_key *location)
5448 {
5449 const char *name = dentry->d_name.name;
5450 int namelen = dentry->d_name.len;
5451 struct btrfs_dir_item *di;
5452 struct btrfs_path *path;
5453 struct btrfs_root *root = BTRFS_I(dir)->root;
5454 int ret = 0;
5455
5456 path = btrfs_alloc_path();
5457 if (!path)
5458 return -ENOMEM;
5459
5460 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5461 name, namelen, 0);
5462 if (IS_ERR(di))
5463 ret = PTR_ERR(di);
5464
5465 if (IS_ERR_OR_NULL(di))
5466 goto out_err;
5467
5468 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5469 if (location->type != BTRFS_INODE_ITEM_KEY &&
5470 location->type != BTRFS_ROOT_ITEM_KEY) {
5471 btrfs_warn(root->fs_info,
5472 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5473 __func__, name, btrfs_ino(BTRFS_I(dir)),
5474 location->objectid, location->type, location->offset);
5475 goto out_err;
5476 }
5477 out:
5478 btrfs_free_path(path);
5479 return ret;
5480 out_err:
5481 location->objectid = 0;
5482 goto out;
5483 }
5484
5485 /*
5486 * when we hit a tree root in a directory, the btrfs part of the inode
5487 * needs to be changed to reflect the root directory of the tree root. This
5488 * is kind of like crossing a mount point.
5489 */
5490 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5491 struct inode *dir,
5492 struct dentry *dentry,
5493 struct btrfs_key *location,
5494 struct btrfs_root **sub_root)
5495 {
5496 struct btrfs_path *path;
5497 struct btrfs_root *new_root;
5498 struct btrfs_root_ref *ref;
5499 struct extent_buffer *leaf;
5500 struct btrfs_key key;
5501 int ret;
5502 int err = 0;
5503
5504 path = btrfs_alloc_path();
5505 if (!path) {
5506 err = -ENOMEM;
5507 goto out;
5508 }
5509
5510 err = -ENOENT;
5511 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5512 key.type = BTRFS_ROOT_REF_KEY;
5513 key.offset = location->objectid;
5514
5515 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5516 if (ret) {
5517 if (ret < 0)
5518 err = ret;
5519 goto out;
5520 }
5521
5522 leaf = path->nodes[0];
5523 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5524 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5525 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5526 goto out;
5527
5528 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5529 (unsigned long)(ref + 1),
5530 dentry->d_name.len);
5531 if (ret)
5532 goto out;
5533
5534 btrfs_release_path(path);
5535
5536 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5537 if (IS_ERR(new_root)) {
5538 err = PTR_ERR(new_root);
5539 goto out;
5540 }
5541
5542 *sub_root = new_root;
5543 location->objectid = btrfs_root_dirid(&new_root->root_item);
5544 location->type = BTRFS_INODE_ITEM_KEY;
5545 location->offset = 0;
5546 err = 0;
5547 out:
5548 btrfs_free_path(path);
5549 return err;
5550 }
5551
5552 static void inode_tree_add(struct inode *inode)
5553 {
5554 struct btrfs_root *root = BTRFS_I(inode)->root;
5555 struct btrfs_inode *entry;
5556 struct rb_node **p;
5557 struct rb_node *parent;
5558 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5559 u64 ino = btrfs_ino(BTRFS_I(inode));
5560
5561 if (inode_unhashed(inode))
5562 return;
5563 parent = NULL;
5564 spin_lock(&root->inode_lock);
5565 p = &root->inode_tree.rb_node;
5566 while (*p) {
5567 parent = *p;
5568 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5569
5570 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5571 p = &parent->rb_left;
5572 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5573 p = &parent->rb_right;
5574 else {
5575 WARN_ON(!(entry->vfs_inode.i_state &
5576 (I_WILL_FREE | I_FREEING)));
5577 rb_replace_node(parent, new, &root->inode_tree);
5578 RB_CLEAR_NODE(parent);
5579 spin_unlock(&root->inode_lock);
5580 return;
5581 }
5582 }
5583 rb_link_node(new, parent, p);
5584 rb_insert_color(new, &root->inode_tree);
5585 spin_unlock(&root->inode_lock);
5586 }
5587
5588 static void inode_tree_del(struct inode *inode)
5589 {
5590 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5591 struct btrfs_root *root = BTRFS_I(inode)->root;
5592 int empty = 0;
5593
5594 spin_lock(&root->inode_lock);
5595 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5596 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5597 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5598 empty = RB_EMPTY_ROOT(&root->inode_tree);
5599 }
5600 spin_unlock(&root->inode_lock);
5601
5602 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5603 synchronize_srcu(&fs_info->subvol_srcu);
5604 spin_lock(&root->inode_lock);
5605 empty = RB_EMPTY_ROOT(&root->inode_tree);
5606 spin_unlock(&root->inode_lock);
5607 if (empty)
5608 btrfs_add_dead_root(root);
5609 }
5610 }
5611
5612 void btrfs_invalidate_inodes(struct btrfs_root *root)
5613 {
5614 struct btrfs_fs_info *fs_info = root->fs_info;
5615 struct rb_node *node;
5616 struct rb_node *prev;
5617 struct btrfs_inode *entry;
5618 struct inode *inode;
5619 u64 objectid = 0;
5620
5621 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5622 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5623
5624 spin_lock(&root->inode_lock);
5625 again:
5626 node = root->inode_tree.rb_node;
5627 prev = NULL;
5628 while (node) {
5629 prev = node;
5630 entry = rb_entry(node, struct btrfs_inode, rb_node);
5631
5632 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5633 node = node->rb_left;
5634 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5635 node = node->rb_right;
5636 else
5637 break;
5638 }
5639 if (!node) {
5640 while (prev) {
5641 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5642 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5643 node = prev;
5644 break;
5645 }
5646 prev = rb_next(prev);
5647 }
5648 }
5649 while (node) {
5650 entry = rb_entry(node, struct btrfs_inode, rb_node);
5651 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5652 inode = igrab(&entry->vfs_inode);
5653 if (inode) {
5654 spin_unlock(&root->inode_lock);
5655 if (atomic_read(&inode->i_count) > 1)
5656 d_prune_aliases(inode);
5657 /*
5658 * btrfs_drop_inode will have it removed from
5659 * the inode cache when its usage count
5660 * hits zero.
5661 */
5662 iput(inode);
5663 cond_resched();
5664 spin_lock(&root->inode_lock);
5665 goto again;
5666 }
5667
5668 if (cond_resched_lock(&root->inode_lock))
5669 goto again;
5670
5671 node = rb_next(node);
5672 }
5673 spin_unlock(&root->inode_lock);
5674 }
5675
5676 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5677 {
5678 struct btrfs_iget_args *args = p;
5679 inode->i_ino = args->location->objectid;
5680 memcpy(&BTRFS_I(inode)->location, args->location,
5681 sizeof(*args->location));
5682 BTRFS_I(inode)->root = args->root;
5683 return 0;
5684 }
5685
5686 static int btrfs_find_actor(struct inode *inode, void *opaque)
5687 {
5688 struct btrfs_iget_args *args = opaque;
5689 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5690 args->root == BTRFS_I(inode)->root;
5691 }
5692
5693 static struct inode *btrfs_iget_locked(struct super_block *s,
5694 struct btrfs_key *location,
5695 struct btrfs_root *root)
5696 {
5697 struct inode *inode;
5698 struct btrfs_iget_args args;
5699 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5700
5701 args.location = location;
5702 args.root = root;
5703
5704 inode = iget5_locked(s, hashval, btrfs_find_actor,
5705 btrfs_init_locked_inode,
5706 (void *)&args);
5707 return inode;
5708 }
5709
5710 /* Get an inode object given its location and corresponding root.
5711 * Returns in *is_new if the inode was read from disk
5712 */
5713 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5714 struct btrfs_root *root, int *new)
5715 {
5716 struct inode *inode;
5717
5718 inode = btrfs_iget_locked(s, location, root);
5719 if (!inode)
5720 return ERR_PTR(-ENOMEM);
5721
5722 if (inode->i_state & I_NEW) {
5723 int ret;
5724
5725 ret = btrfs_read_locked_inode(inode);
5726 if (!is_bad_inode(inode)) {
5727 inode_tree_add(inode);
5728 unlock_new_inode(inode);
5729 if (new)
5730 *new = 1;
5731 } else {
5732 unlock_new_inode(inode);
5733 iput(inode);
5734 ASSERT(ret < 0);
5735 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5736 }
5737 }
5738
5739 return inode;
5740 }
5741
5742 static struct inode *new_simple_dir(struct super_block *s,
5743 struct btrfs_key *key,
5744 struct btrfs_root *root)
5745 {
5746 struct inode *inode = new_inode(s);
5747
5748 if (!inode)
5749 return ERR_PTR(-ENOMEM);
5750
5751 BTRFS_I(inode)->root = root;
5752 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5753 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5754
5755 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5756 inode->i_op = &btrfs_dir_ro_inode_operations;
5757 inode->i_opflags &= ~IOP_XATTR;
5758 inode->i_fop = &simple_dir_operations;
5759 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5760 inode->i_mtime = current_time(inode);
5761 inode->i_atime = inode->i_mtime;
5762 inode->i_ctime = inode->i_mtime;
5763 BTRFS_I(inode)->i_otime = inode->i_mtime;
5764
5765 return inode;
5766 }
5767
5768 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5769 {
5770 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5771 struct inode *inode;
5772 struct btrfs_root *root = BTRFS_I(dir)->root;
5773 struct btrfs_root *sub_root = root;
5774 struct btrfs_key location;
5775 int index;
5776 int ret = 0;
5777
5778 if (dentry->d_name.len > BTRFS_NAME_LEN)
5779 return ERR_PTR(-ENAMETOOLONG);
5780
5781 ret = btrfs_inode_by_name(dir, dentry, &location);
5782 if (ret < 0)
5783 return ERR_PTR(ret);
5784
5785 if (location.objectid == 0)
5786 return ERR_PTR(-ENOENT);
5787
5788 if (location.type == BTRFS_INODE_ITEM_KEY) {
5789 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5790 return inode;
5791 }
5792
5793 index = srcu_read_lock(&fs_info->subvol_srcu);
5794 ret = fixup_tree_root_location(fs_info, dir, dentry,
5795 &location, &sub_root);
5796 if (ret < 0) {
5797 if (ret != -ENOENT)
5798 inode = ERR_PTR(ret);
5799 else
5800 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5801 } else {
5802 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5803 }
5804 srcu_read_unlock(&fs_info->subvol_srcu, index);
5805
5806 if (!IS_ERR(inode) && root != sub_root) {
5807 down_read(&fs_info->cleanup_work_sem);
5808 if (!sb_rdonly(inode->i_sb))
5809 ret = btrfs_orphan_cleanup(sub_root);
5810 up_read(&fs_info->cleanup_work_sem);
5811 if (ret) {
5812 iput(inode);
5813 inode = ERR_PTR(ret);
5814 }
5815 }
5816
5817 return inode;
5818 }
5819
5820 static int btrfs_dentry_delete(const struct dentry *dentry)
5821 {
5822 struct btrfs_root *root;
5823 struct inode *inode = d_inode(dentry);
5824
5825 if (!inode && !IS_ROOT(dentry))
5826 inode = d_inode(dentry->d_parent);
5827
5828 if (inode) {
5829 root = BTRFS_I(inode)->root;
5830 if (btrfs_root_refs(&root->root_item) == 0)
5831 return 1;
5832
5833 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5834 return 1;
5835 }
5836 return 0;
5837 }
5838
5839 static void btrfs_dentry_release(struct dentry *dentry)
5840 {
5841 kfree(dentry->d_fsdata);
5842 }
5843
5844 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5845 unsigned int flags)
5846 {
5847 struct inode *inode;
5848
5849 inode = btrfs_lookup_dentry(dir, dentry);
5850 if (IS_ERR(inode)) {
5851 if (PTR_ERR(inode) == -ENOENT)
5852 inode = NULL;
5853 else
5854 return ERR_CAST(inode);
5855 }
5856
5857 return d_splice_alias(inode, dentry);
5858 }
5859
5860 unsigned char btrfs_filetype_table[] = {
5861 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5862 };
5863
5864 /*
5865 * All this infrastructure exists because dir_emit can fault, and we are holding
5866 * the tree lock when doing readdir. For now just allocate a buffer and copy
5867 * our information into that, and then dir_emit from the buffer. This is
5868 * similar to what NFS does, only we don't keep the buffer around in pagecache
5869 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5870 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5871 * tree lock.
5872 */
5873 static int btrfs_opendir(struct inode *inode, struct file *file)
5874 {
5875 struct btrfs_file_private *private;
5876
5877 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5878 if (!private)
5879 return -ENOMEM;
5880 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5881 if (!private->filldir_buf) {
5882 kfree(private);
5883 return -ENOMEM;
5884 }
5885 file->private_data = private;
5886 return 0;
5887 }
5888
5889 struct dir_entry {
5890 u64 ino;
5891 u64 offset;
5892 unsigned type;
5893 int name_len;
5894 };
5895
5896 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5897 {
5898 while (entries--) {
5899 struct dir_entry *entry = addr;
5900 char *name = (char *)(entry + 1);
5901
5902 ctx->pos = entry->offset;
5903 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5904 entry->type))
5905 return 1;
5906 addr += sizeof(struct dir_entry) + entry->name_len;
5907 ctx->pos++;
5908 }
5909 return 0;
5910 }
5911
5912 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5913 {
5914 struct inode *inode = file_inode(file);
5915 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5916 struct btrfs_root *root = BTRFS_I(inode)->root;
5917 struct btrfs_file_private *private = file->private_data;
5918 struct btrfs_dir_item *di;
5919 struct btrfs_key key;
5920 struct btrfs_key found_key;
5921 struct btrfs_path *path;
5922 void *addr;
5923 struct list_head ins_list;
5924 struct list_head del_list;
5925 int ret;
5926 struct extent_buffer *leaf;
5927 int slot;
5928 char *name_ptr;
5929 int name_len;
5930 int entries = 0;
5931 int total_len = 0;
5932 bool put = false;
5933 struct btrfs_key location;
5934
5935 if (!dir_emit_dots(file, ctx))
5936 return 0;
5937
5938 path = btrfs_alloc_path();
5939 if (!path)
5940 return -ENOMEM;
5941
5942 addr = private->filldir_buf;
5943 path->reada = READA_FORWARD;
5944
5945 INIT_LIST_HEAD(&ins_list);
5946 INIT_LIST_HEAD(&del_list);
5947 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5948
5949 again:
5950 key.type = BTRFS_DIR_INDEX_KEY;
5951 key.offset = ctx->pos;
5952 key.objectid = btrfs_ino(BTRFS_I(inode));
5953
5954 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5955 if (ret < 0)
5956 goto err;
5957
5958 while (1) {
5959 struct dir_entry *entry;
5960
5961 leaf = path->nodes[0];
5962 slot = path->slots[0];
5963 if (slot >= btrfs_header_nritems(leaf)) {
5964 ret = btrfs_next_leaf(root, path);
5965 if (ret < 0)
5966 goto err;
5967 else if (ret > 0)
5968 break;
5969 continue;
5970 }
5971
5972 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5973
5974 if (found_key.objectid != key.objectid)
5975 break;
5976 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5977 break;
5978 if (found_key.offset < ctx->pos)
5979 goto next;
5980 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5981 goto next;
5982 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5983 if (verify_dir_item(fs_info, leaf, slot, di))
5984 goto next;
5985
5986 name_len = btrfs_dir_name_len(leaf, di);
5987 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5988 PAGE_SIZE) {
5989 btrfs_release_path(path);
5990 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5991 if (ret)
5992 goto nopos;
5993 addr = private->filldir_buf;
5994 entries = 0;
5995 total_len = 0;
5996 goto again;
5997 }
5998
5999 entry = addr;
6000 entry->name_len = name_len;
6001 name_ptr = (char *)(entry + 1);
6002 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6003 name_len);
6004 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
6005 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6006 entry->ino = location.objectid;
6007 entry->offset = found_key.offset;
6008 entries++;
6009 addr += sizeof(struct dir_entry) + name_len;
6010 total_len += sizeof(struct dir_entry) + name_len;
6011 next:
6012 path->slots[0]++;
6013 }
6014 btrfs_release_path(path);
6015
6016 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6017 if (ret)
6018 goto nopos;
6019
6020 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6021 if (ret)
6022 goto nopos;
6023
6024 /*
6025 * Stop new entries from being returned after we return the last
6026 * entry.
6027 *
6028 * New directory entries are assigned a strictly increasing
6029 * offset. This means that new entries created during readdir
6030 * are *guaranteed* to be seen in the future by that readdir.
6031 * This has broken buggy programs which operate on names as
6032 * they're returned by readdir. Until we re-use freed offsets
6033 * we have this hack to stop new entries from being returned
6034 * under the assumption that they'll never reach this huge
6035 * offset.
6036 *
6037 * This is being careful not to overflow 32bit loff_t unless the
6038 * last entry requires it because doing so has broken 32bit apps
6039 * in the past.
6040 */
6041 if (ctx->pos >= INT_MAX)
6042 ctx->pos = LLONG_MAX;
6043 else
6044 ctx->pos = INT_MAX;
6045 nopos:
6046 ret = 0;
6047 err:
6048 if (put)
6049 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6050 btrfs_free_path(path);
6051 return ret;
6052 }
6053
6054 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6055 {
6056 struct btrfs_root *root = BTRFS_I(inode)->root;
6057 struct btrfs_trans_handle *trans;
6058 int ret = 0;
6059 bool nolock = false;
6060
6061 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6062 return 0;
6063
6064 if (btrfs_fs_closing(root->fs_info) &&
6065 btrfs_is_free_space_inode(BTRFS_I(inode)))
6066 nolock = true;
6067
6068 if (wbc->sync_mode == WB_SYNC_ALL) {
6069 if (nolock)
6070 trans = btrfs_join_transaction_nolock(root);
6071 else
6072 trans = btrfs_join_transaction(root);
6073 if (IS_ERR(trans))
6074 return PTR_ERR(trans);
6075 ret = btrfs_commit_transaction(trans);
6076 }
6077 return ret;
6078 }
6079
6080 /*
6081 * This is somewhat expensive, updating the tree every time the
6082 * inode changes. But, it is most likely to find the inode in cache.
6083 * FIXME, needs more benchmarking...there are no reasons other than performance
6084 * to keep or drop this code.
6085 */
6086 static int btrfs_dirty_inode(struct inode *inode)
6087 {
6088 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6089 struct btrfs_root *root = BTRFS_I(inode)->root;
6090 struct btrfs_trans_handle *trans;
6091 int ret;
6092
6093 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6094 return 0;
6095
6096 trans = btrfs_join_transaction(root);
6097 if (IS_ERR(trans))
6098 return PTR_ERR(trans);
6099
6100 ret = btrfs_update_inode(trans, root, inode);
6101 if (ret && ret == -ENOSPC) {
6102 /* whoops, lets try again with the full transaction */
6103 btrfs_end_transaction(trans);
6104 trans = btrfs_start_transaction(root, 1);
6105 if (IS_ERR(trans))
6106 return PTR_ERR(trans);
6107
6108 ret = btrfs_update_inode(trans, root, inode);
6109 }
6110 btrfs_end_transaction(trans);
6111 if (BTRFS_I(inode)->delayed_node)
6112 btrfs_balance_delayed_items(fs_info);
6113
6114 return ret;
6115 }
6116
6117 /*
6118 * This is a copy of file_update_time. We need this so we can return error on
6119 * ENOSPC for updating the inode in the case of file write and mmap writes.
6120 */
6121 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6122 int flags)
6123 {
6124 struct btrfs_root *root = BTRFS_I(inode)->root;
6125
6126 if (btrfs_root_readonly(root))
6127 return -EROFS;
6128
6129 if (flags & S_VERSION)
6130 inode_inc_iversion(inode);
6131 if (flags & S_CTIME)
6132 inode->i_ctime = *now;
6133 if (flags & S_MTIME)
6134 inode->i_mtime = *now;
6135 if (flags & S_ATIME)
6136 inode->i_atime = *now;
6137 return btrfs_dirty_inode(inode);
6138 }
6139
6140 /*
6141 * find the highest existing sequence number in a directory
6142 * and then set the in-memory index_cnt variable to reflect
6143 * free sequence numbers
6144 */
6145 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6146 {
6147 struct btrfs_root *root = inode->root;
6148 struct btrfs_key key, found_key;
6149 struct btrfs_path *path;
6150 struct extent_buffer *leaf;
6151 int ret;
6152
6153 key.objectid = btrfs_ino(inode);
6154 key.type = BTRFS_DIR_INDEX_KEY;
6155 key.offset = (u64)-1;
6156
6157 path = btrfs_alloc_path();
6158 if (!path)
6159 return -ENOMEM;
6160
6161 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6162 if (ret < 0)
6163 goto out;
6164 /* FIXME: we should be able to handle this */
6165 if (ret == 0)
6166 goto out;
6167 ret = 0;
6168
6169 /*
6170 * MAGIC NUMBER EXPLANATION:
6171 * since we search a directory based on f_pos we have to start at 2
6172 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6173 * else has to start at 2
6174 */
6175 if (path->slots[0] == 0) {
6176 inode->index_cnt = 2;
6177 goto out;
6178 }
6179
6180 path->slots[0]--;
6181
6182 leaf = path->nodes[0];
6183 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6184
6185 if (found_key.objectid != btrfs_ino(inode) ||
6186 found_key.type != BTRFS_DIR_INDEX_KEY) {
6187 inode->index_cnt = 2;
6188 goto out;
6189 }
6190
6191 inode->index_cnt = found_key.offset + 1;
6192 out:
6193 btrfs_free_path(path);
6194 return ret;
6195 }
6196
6197 /*
6198 * helper to find a free sequence number in a given directory. This current
6199 * code is very simple, later versions will do smarter things in the btree
6200 */
6201 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6202 {
6203 int ret = 0;
6204
6205 if (dir->index_cnt == (u64)-1) {
6206 ret = btrfs_inode_delayed_dir_index_count(dir);
6207 if (ret) {
6208 ret = btrfs_set_inode_index_count(dir);
6209 if (ret)
6210 return ret;
6211 }
6212 }
6213
6214 *index = dir->index_cnt;
6215 dir->index_cnt++;
6216
6217 return ret;
6218 }
6219
6220 static int btrfs_insert_inode_locked(struct inode *inode)
6221 {
6222 struct btrfs_iget_args args;
6223 args.location = &BTRFS_I(inode)->location;
6224 args.root = BTRFS_I(inode)->root;
6225
6226 return insert_inode_locked4(inode,
6227 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6228 btrfs_find_actor, &args);
6229 }
6230
6231 /*
6232 * Inherit flags from the parent inode.
6233 *
6234 * Currently only the compression flags and the cow flags are inherited.
6235 */
6236 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6237 {
6238 unsigned int flags;
6239
6240 if (!dir)
6241 return;
6242
6243 flags = BTRFS_I(dir)->flags;
6244
6245 if (flags & BTRFS_INODE_NOCOMPRESS) {
6246 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6247 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6248 } else if (flags & BTRFS_INODE_COMPRESS) {
6249 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6250 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6251 }
6252
6253 if (flags & BTRFS_INODE_NODATACOW) {
6254 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6255 if (S_ISREG(inode->i_mode))
6256 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6257 }
6258
6259 btrfs_update_iflags(inode);
6260 }
6261
6262 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6263 struct btrfs_root *root,
6264 struct inode *dir,
6265 const char *name, int name_len,
6266 u64 ref_objectid, u64 objectid,
6267 umode_t mode, u64 *index)
6268 {
6269 struct btrfs_fs_info *fs_info = root->fs_info;
6270 struct inode *inode;
6271 struct btrfs_inode_item *inode_item;
6272 struct btrfs_key *location;
6273 struct btrfs_path *path;
6274 struct btrfs_inode_ref *ref;
6275 struct btrfs_key key[2];
6276 u32 sizes[2];
6277 int nitems = name ? 2 : 1;
6278 unsigned long ptr;
6279 int ret;
6280
6281 path = btrfs_alloc_path();
6282 if (!path)
6283 return ERR_PTR(-ENOMEM);
6284
6285 inode = new_inode(fs_info->sb);
6286 if (!inode) {
6287 btrfs_free_path(path);
6288 return ERR_PTR(-ENOMEM);
6289 }
6290
6291 /*
6292 * O_TMPFILE, set link count to 0, so that after this point,
6293 * we fill in an inode item with the correct link count.
6294 */
6295 if (!name)
6296 set_nlink(inode, 0);
6297
6298 /*
6299 * we have to initialize this early, so we can reclaim the inode
6300 * number if we fail afterwards in this function.
6301 */
6302 inode->i_ino = objectid;
6303
6304 if (dir && name) {
6305 trace_btrfs_inode_request(dir);
6306
6307 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6308 if (ret) {
6309 btrfs_free_path(path);
6310 iput(inode);
6311 return ERR_PTR(ret);
6312 }
6313 } else if (dir) {
6314 *index = 0;
6315 }
6316 /*
6317 * index_cnt is ignored for everything but a dir,
6318 * btrfs_get_inode_index_count has an explanation for the magic
6319 * number
6320 */
6321 BTRFS_I(inode)->index_cnt = 2;
6322 BTRFS_I(inode)->dir_index = *index;
6323 BTRFS_I(inode)->root = root;
6324 BTRFS_I(inode)->generation = trans->transid;
6325 inode->i_generation = BTRFS_I(inode)->generation;
6326
6327 /*
6328 * We could have gotten an inode number from somebody who was fsynced
6329 * and then removed in this same transaction, so let's just set full
6330 * sync since it will be a full sync anyway and this will blow away the
6331 * old info in the log.
6332 */
6333 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6334
6335 key[0].objectid = objectid;
6336 key[0].type = BTRFS_INODE_ITEM_KEY;
6337 key[0].offset = 0;
6338
6339 sizes[0] = sizeof(struct btrfs_inode_item);
6340
6341 if (name) {
6342 /*
6343 * Start new inodes with an inode_ref. This is slightly more
6344 * efficient for small numbers of hard links since they will
6345 * be packed into one item. Extended refs will kick in if we
6346 * add more hard links than can fit in the ref item.
6347 */
6348 key[1].objectid = objectid;
6349 key[1].type = BTRFS_INODE_REF_KEY;
6350 key[1].offset = ref_objectid;
6351
6352 sizes[1] = name_len + sizeof(*ref);
6353 }
6354
6355 location = &BTRFS_I(inode)->location;
6356 location->objectid = objectid;
6357 location->offset = 0;
6358 location->type = BTRFS_INODE_ITEM_KEY;
6359
6360 ret = btrfs_insert_inode_locked(inode);
6361 if (ret < 0)
6362 goto fail;
6363
6364 path->leave_spinning = 1;
6365 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6366 if (ret != 0)
6367 goto fail_unlock;
6368
6369 inode_init_owner(inode, dir, mode);
6370 inode_set_bytes(inode, 0);
6371
6372 inode->i_mtime = current_time(inode);
6373 inode->i_atime = inode->i_mtime;
6374 inode->i_ctime = inode->i_mtime;
6375 BTRFS_I(inode)->i_otime = inode->i_mtime;
6376
6377 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6378 struct btrfs_inode_item);
6379 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6380 sizeof(*inode_item));
6381 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6382
6383 if (name) {
6384 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6385 struct btrfs_inode_ref);
6386 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6387 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6388 ptr = (unsigned long)(ref + 1);
6389 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6390 }
6391
6392 btrfs_mark_buffer_dirty(path->nodes[0]);
6393 btrfs_free_path(path);
6394
6395 btrfs_inherit_iflags(inode, dir);
6396
6397 if (S_ISREG(mode)) {
6398 if (btrfs_test_opt(fs_info, NODATASUM))
6399 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6400 if (btrfs_test_opt(fs_info, NODATACOW))
6401 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6402 BTRFS_INODE_NODATASUM;
6403 }
6404
6405 inode_tree_add(inode);
6406
6407 trace_btrfs_inode_new(inode);
6408 btrfs_set_inode_last_trans(trans, inode);
6409
6410 btrfs_update_root_times(trans, root);
6411
6412 ret = btrfs_inode_inherit_props(trans, inode, dir);
6413 if (ret)
6414 btrfs_err(fs_info,
6415 "error inheriting props for ino %llu (root %llu): %d",
6416 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6417
6418 return inode;
6419
6420 fail_unlock:
6421 unlock_new_inode(inode);
6422 fail:
6423 if (dir && name)
6424 BTRFS_I(dir)->index_cnt--;
6425 btrfs_free_path(path);
6426 iput(inode);
6427 return ERR_PTR(ret);
6428 }
6429
6430 static inline u8 btrfs_inode_type(struct inode *inode)
6431 {
6432 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6433 }
6434
6435 /*
6436 * utility function to add 'inode' into 'parent_inode' with
6437 * a give name and a given sequence number.
6438 * if 'add_backref' is true, also insert a backref from the
6439 * inode to the parent directory.
6440 */
6441 int btrfs_add_link(struct btrfs_trans_handle *trans,
6442 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6443 const char *name, int name_len, int add_backref, u64 index)
6444 {
6445 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6446 int ret = 0;
6447 struct btrfs_key key;
6448 struct btrfs_root *root = parent_inode->root;
6449 u64 ino = btrfs_ino(inode);
6450 u64 parent_ino = btrfs_ino(parent_inode);
6451
6452 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6453 memcpy(&key, &inode->root->root_key, sizeof(key));
6454 } else {
6455 key.objectid = ino;
6456 key.type = BTRFS_INODE_ITEM_KEY;
6457 key.offset = 0;
6458 }
6459
6460 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6461 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6462 root->root_key.objectid, parent_ino,
6463 index, name, name_len);
6464 } else if (add_backref) {
6465 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6466 parent_ino, index);
6467 }
6468
6469 /* Nothing to clean up yet */
6470 if (ret)
6471 return ret;
6472
6473 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6474 parent_inode, &key,
6475 btrfs_inode_type(&inode->vfs_inode), index);
6476 if (ret == -EEXIST || ret == -EOVERFLOW)
6477 goto fail_dir_item;
6478 else if (ret) {
6479 btrfs_abort_transaction(trans, ret);
6480 return ret;
6481 }
6482
6483 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6484 name_len * 2);
6485 inode_inc_iversion(&parent_inode->vfs_inode);
6486 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6487 current_time(&parent_inode->vfs_inode);
6488 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6489 if (ret)
6490 btrfs_abort_transaction(trans, ret);
6491 return ret;
6492
6493 fail_dir_item:
6494 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6495 u64 local_index;
6496 int err;
6497 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6498 root->root_key.objectid, parent_ino,
6499 &local_index, name, name_len);
6500
6501 } else if (add_backref) {
6502 u64 local_index;
6503 int err;
6504
6505 err = btrfs_del_inode_ref(trans, root, name, name_len,
6506 ino, parent_ino, &local_index);
6507 }
6508 return ret;
6509 }
6510
6511 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6512 struct btrfs_inode *dir, struct dentry *dentry,
6513 struct btrfs_inode *inode, int backref, u64 index)
6514 {
6515 int err = btrfs_add_link(trans, dir, inode,
6516 dentry->d_name.name, dentry->d_name.len,
6517 backref, index);
6518 if (err > 0)
6519 err = -EEXIST;
6520 return err;
6521 }
6522
6523 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6524 umode_t mode, dev_t rdev)
6525 {
6526 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6527 struct btrfs_trans_handle *trans;
6528 struct btrfs_root *root = BTRFS_I(dir)->root;
6529 struct inode *inode = NULL;
6530 int err;
6531 int drop_inode = 0;
6532 u64 objectid;
6533 u64 index = 0;
6534
6535 /*
6536 * 2 for inode item and ref
6537 * 2 for dir items
6538 * 1 for xattr if selinux is on
6539 */
6540 trans = btrfs_start_transaction(root, 5);
6541 if (IS_ERR(trans))
6542 return PTR_ERR(trans);
6543
6544 err = btrfs_find_free_ino(root, &objectid);
6545 if (err)
6546 goto out_unlock;
6547
6548 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6549 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6550 mode, &index);
6551 if (IS_ERR(inode)) {
6552 err = PTR_ERR(inode);
6553 goto out_unlock;
6554 }
6555
6556 /*
6557 * If the active LSM wants to access the inode during
6558 * d_instantiate it needs these. Smack checks to see
6559 * if the filesystem supports xattrs by looking at the
6560 * ops vector.
6561 */
6562 inode->i_op = &btrfs_special_inode_operations;
6563 init_special_inode(inode, inode->i_mode, rdev);
6564
6565 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6566 if (err)
6567 goto out_unlock_inode;
6568
6569 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6570 0, index);
6571 if (err) {
6572 goto out_unlock_inode;
6573 } else {
6574 btrfs_update_inode(trans, root, inode);
6575 unlock_new_inode(inode);
6576 d_instantiate(dentry, inode);
6577 }
6578
6579 out_unlock:
6580 btrfs_end_transaction(trans);
6581 btrfs_balance_delayed_items(fs_info);
6582 btrfs_btree_balance_dirty(fs_info);
6583 if (drop_inode) {
6584 inode_dec_link_count(inode);
6585 iput(inode);
6586 }
6587 return err;
6588
6589 out_unlock_inode:
6590 drop_inode = 1;
6591 unlock_new_inode(inode);
6592 goto out_unlock;
6593
6594 }
6595
6596 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6597 umode_t mode, bool excl)
6598 {
6599 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6600 struct btrfs_trans_handle *trans;
6601 struct btrfs_root *root = BTRFS_I(dir)->root;
6602 struct inode *inode = NULL;
6603 int drop_inode_on_err = 0;
6604 int err;
6605 u64 objectid;
6606 u64 index = 0;
6607
6608 /*
6609 * 2 for inode item and ref
6610 * 2 for dir items
6611 * 1 for xattr if selinux is on
6612 */
6613 trans = btrfs_start_transaction(root, 5);
6614 if (IS_ERR(trans))
6615 return PTR_ERR(trans);
6616
6617 err = btrfs_find_free_ino(root, &objectid);
6618 if (err)
6619 goto out_unlock;
6620
6621 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6622 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6623 mode, &index);
6624 if (IS_ERR(inode)) {
6625 err = PTR_ERR(inode);
6626 goto out_unlock;
6627 }
6628 drop_inode_on_err = 1;
6629 /*
6630 * If the active LSM wants to access the inode during
6631 * d_instantiate it needs these. Smack checks to see
6632 * if the filesystem supports xattrs by looking at the
6633 * ops vector.
6634 */
6635 inode->i_fop = &btrfs_file_operations;
6636 inode->i_op = &btrfs_file_inode_operations;
6637 inode->i_mapping->a_ops = &btrfs_aops;
6638
6639 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6640 if (err)
6641 goto out_unlock_inode;
6642
6643 err = btrfs_update_inode(trans, root, inode);
6644 if (err)
6645 goto out_unlock_inode;
6646
6647 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6648 0, index);
6649 if (err)
6650 goto out_unlock_inode;
6651
6652 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6653 unlock_new_inode(inode);
6654 d_instantiate(dentry, inode);
6655
6656 out_unlock:
6657 btrfs_end_transaction(trans);
6658 if (err && drop_inode_on_err) {
6659 inode_dec_link_count(inode);
6660 iput(inode);
6661 }
6662 btrfs_balance_delayed_items(fs_info);
6663 btrfs_btree_balance_dirty(fs_info);
6664 return err;
6665
6666 out_unlock_inode:
6667 unlock_new_inode(inode);
6668 goto out_unlock;
6669
6670 }
6671
6672 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6673 struct dentry *dentry)
6674 {
6675 struct btrfs_trans_handle *trans = NULL;
6676 struct btrfs_root *root = BTRFS_I(dir)->root;
6677 struct inode *inode = d_inode(old_dentry);
6678 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6679 u64 index;
6680 int err;
6681 int drop_inode = 0;
6682
6683 /* do not allow sys_link's with other subvols of the same device */
6684 if (root->objectid != BTRFS_I(inode)->root->objectid)
6685 return -EXDEV;
6686
6687 if (inode->i_nlink >= BTRFS_LINK_MAX)
6688 return -EMLINK;
6689
6690 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6691 if (err)
6692 goto fail;
6693
6694 /*
6695 * 2 items for inode and inode ref
6696 * 2 items for dir items
6697 * 1 item for parent inode
6698 */
6699 trans = btrfs_start_transaction(root, 5);
6700 if (IS_ERR(trans)) {
6701 err = PTR_ERR(trans);
6702 trans = NULL;
6703 goto fail;
6704 }
6705
6706 /* There are several dir indexes for this inode, clear the cache. */
6707 BTRFS_I(inode)->dir_index = 0ULL;
6708 inc_nlink(inode);
6709 inode_inc_iversion(inode);
6710 inode->i_ctime = current_time(inode);
6711 ihold(inode);
6712 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6713
6714 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6715 1, index);
6716
6717 if (err) {
6718 drop_inode = 1;
6719 } else {
6720 struct dentry *parent = dentry->d_parent;
6721 err = btrfs_update_inode(trans, root, inode);
6722 if (err)
6723 goto fail;
6724 if (inode->i_nlink == 1) {
6725 /*
6726 * If new hard link count is 1, it's a file created
6727 * with open(2) O_TMPFILE flag.
6728 */
6729 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6730 if (err)
6731 goto fail;
6732 }
6733 d_instantiate(dentry, inode);
6734 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6735 }
6736
6737 btrfs_balance_delayed_items(fs_info);
6738 fail:
6739 if (trans)
6740 btrfs_end_transaction(trans);
6741 if (drop_inode) {
6742 inode_dec_link_count(inode);
6743 iput(inode);
6744 }
6745 btrfs_btree_balance_dirty(fs_info);
6746 return err;
6747 }
6748
6749 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6750 {
6751 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6752 struct inode *inode = NULL;
6753 struct btrfs_trans_handle *trans;
6754 struct btrfs_root *root = BTRFS_I(dir)->root;
6755 int err = 0;
6756 int drop_on_err = 0;
6757 u64 objectid = 0;
6758 u64 index = 0;
6759
6760 /*
6761 * 2 items for inode and ref
6762 * 2 items for dir items
6763 * 1 for xattr if selinux is on
6764 */
6765 trans = btrfs_start_transaction(root, 5);
6766 if (IS_ERR(trans))
6767 return PTR_ERR(trans);
6768
6769 err = btrfs_find_free_ino(root, &objectid);
6770 if (err)
6771 goto out_fail;
6772
6773 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6774 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6775 S_IFDIR | mode, &index);
6776 if (IS_ERR(inode)) {
6777 err = PTR_ERR(inode);
6778 goto out_fail;
6779 }
6780
6781 drop_on_err = 1;
6782 /* these must be set before we unlock the inode */
6783 inode->i_op = &btrfs_dir_inode_operations;
6784 inode->i_fop = &btrfs_dir_file_operations;
6785
6786 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6787 if (err)
6788 goto out_fail_inode;
6789
6790 btrfs_i_size_write(BTRFS_I(inode), 0);
6791 err = btrfs_update_inode(trans, root, inode);
6792 if (err)
6793 goto out_fail_inode;
6794
6795 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6796 dentry->d_name.name,
6797 dentry->d_name.len, 0, index);
6798 if (err)
6799 goto out_fail_inode;
6800
6801 d_instantiate(dentry, inode);
6802 /*
6803 * mkdir is special. We're unlocking after we call d_instantiate
6804 * to avoid a race with nfsd calling d_instantiate.
6805 */
6806 unlock_new_inode(inode);
6807 drop_on_err = 0;
6808
6809 out_fail:
6810 btrfs_end_transaction(trans);
6811 if (drop_on_err) {
6812 inode_dec_link_count(inode);
6813 iput(inode);
6814 }
6815 btrfs_balance_delayed_items(fs_info);
6816 btrfs_btree_balance_dirty(fs_info);
6817 return err;
6818
6819 out_fail_inode:
6820 unlock_new_inode(inode);
6821 goto out_fail;
6822 }
6823
6824 /* Find next extent map of a given extent map, caller needs to ensure locks */
6825 static struct extent_map *next_extent_map(struct extent_map *em)
6826 {
6827 struct rb_node *next;
6828
6829 next = rb_next(&em->rb_node);
6830 if (!next)
6831 return NULL;
6832 return container_of(next, struct extent_map, rb_node);
6833 }
6834
6835 static struct extent_map *prev_extent_map(struct extent_map *em)
6836 {
6837 struct rb_node *prev;
6838
6839 prev = rb_prev(&em->rb_node);
6840 if (!prev)
6841 return NULL;
6842 return container_of(prev, struct extent_map, rb_node);
6843 }
6844
6845 /* helper for btfs_get_extent. Given an existing extent in the tree,
6846 * the existing extent is the nearest extent to map_start,
6847 * and an extent that you want to insert, deal with overlap and insert
6848 * the best fitted new extent into the tree.
6849 */
6850 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6851 struct extent_map *existing,
6852 struct extent_map *em,
6853 u64 map_start)
6854 {
6855 struct extent_map *prev;
6856 struct extent_map *next;
6857 u64 start;
6858 u64 end;
6859 u64 start_diff;
6860
6861 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6862
6863 if (existing->start > map_start) {
6864 next = existing;
6865 prev = prev_extent_map(next);
6866 } else {
6867 prev = existing;
6868 next = next_extent_map(prev);
6869 }
6870
6871 start = prev ? extent_map_end(prev) : em->start;
6872 start = max_t(u64, start, em->start);
6873 end = next ? next->start : extent_map_end(em);
6874 end = min_t(u64, end, extent_map_end(em));
6875 start_diff = start - em->start;
6876 em->start = start;
6877 em->len = end - start;
6878 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6879 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6880 em->block_start += start_diff;
6881 em->block_len -= start_diff;
6882 }
6883 return add_extent_mapping(em_tree, em, 0);
6884 }
6885
6886 static noinline int uncompress_inline(struct btrfs_path *path,
6887 struct page *page,
6888 size_t pg_offset, u64 extent_offset,
6889 struct btrfs_file_extent_item *item)
6890 {
6891 int ret;
6892 struct extent_buffer *leaf = path->nodes[0];
6893 char *tmp;
6894 size_t max_size;
6895 unsigned long inline_size;
6896 unsigned long ptr;
6897 int compress_type;
6898
6899 WARN_ON(pg_offset != 0);
6900 compress_type = btrfs_file_extent_compression(leaf, item);
6901 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6902 inline_size = btrfs_file_extent_inline_item_len(leaf,
6903 btrfs_item_nr(path->slots[0]));
6904 tmp = kmalloc(inline_size, GFP_NOFS);
6905 if (!tmp)
6906 return -ENOMEM;
6907 ptr = btrfs_file_extent_inline_start(item);
6908
6909 read_extent_buffer(leaf, tmp, ptr, inline_size);
6910
6911 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6912 ret = btrfs_decompress(compress_type, tmp, page,
6913 extent_offset, inline_size, max_size);
6914
6915 /*
6916 * decompression code contains a memset to fill in any space between the end
6917 * of the uncompressed data and the end of max_size in case the decompressed
6918 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6919 * the end of an inline extent and the beginning of the next block, so we
6920 * cover that region here.
6921 */
6922
6923 if (max_size + pg_offset < PAGE_SIZE) {
6924 char *map = kmap(page);
6925 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6926 kunmap(page);
6927 }
6928 kfree(tmp);
6929 return ret;
6930 }
6931
6932 /*
6933 * a bit scary, this does extent mapping from logical file offset to the disk.
6934 * the ugly parts come from merging extents from the disk with the in-ram
6935 * representation. This gets more complex because of the data=ordered code,
6936 * where the in-ram extents might be locked pending data=ordered completion.
6937 *
6938 * This also copies inline extents directly into the page.
6939 */
6940 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6941 struct page *page,
6942 size_t pg_offset, u64 start, u64 len,
6943 int create)
6944 {
6945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6946 int ret;
6947 int err = 0;
6948 u64 extent_start = 0;
6949 u64 extent_end = 0;
6950 u64 objectid = btrfs_ino(inode);
6951 u32 found_type;
6952 struct btrfs_path *path = NULL;
6953 struct btrfs_root *root = inode->root;
6954 struct btrfs_file_extent_item *item;
6955 struct extent_buffer *leaf;
6956 struct btrfs_key found_key;
6957 struct extent_map *em = NULL;
6958 struct extent_map_tree *em_tree = &inode->extent_tree;
6959 struct extent_io_tree *io_tree = &inode->io_tree;
6960 struct btrfs_trans_handle *trans = NULL;
6961 const bool new_inline = !page || create;
6962
6963 again:
6964 read_lock(&em_tree->lock);
6965 em = lookup_extent_mapping(em_tree, start, len);
6966 if (em)
6967 em->bdev = fs_info->fs_devices->latest_bdev;
6968 read_unlock(&em_tree->lock);
6969
6970 if (em) {
6971 if (em->start > start || em->start + em->len <= start)
6972 free_extent_map(em);
6973 else if (em->block_start == EXTENT_MAP_INLINE && page)
6974 free_extent_map(em);
6975 else
6976 goto out;
6977 }
6978 em = alloc_extent_map();
6979 if (!em) {
6980 err = -ENOMEM;
6981 goto out;
6982 }
6983 em->bdev = fs_info->fs_devices->latest_bdev;
6984 em->start = EXTENT_MAP_HOLE;
6985 em->orig_start = EXTENT_MAP_HOLE;
6986 em->len = (u64)-1;
6987 em->block_len = (u64)-1;
6988
6989 if (!path) {
6990 path = btrfs_alloc_path();
6991 if (!path) {
6992 err = -ENOMEM;
6993 goto out;
6994 }
6995 /*
6996 * Chances are we'll be called again, so go ahead and do
6997 * readahead
6998 */
6999 path->reada = READA_FORWARD;
7000 }
7001
7002 ret = btrfs_lookup_file_extent(trans, root, path,
7003 objectid, start, trans != NULL);
7004 if (ret < 0) {
7005 err = ret;
7006 goto out;
7007 }
7008
7009 if (ret != 0) {
7010 if (path->slots[0] == 0)
7011 goto not_found;
7012 path->slots[0]--;
7013 }
7014
7015 leaf = path->nodes[0];
7016 item = btrfs_item_ptr(leaf, path->slots[0],
7017 struct btrfs_file_extent_item);
7018 /* are we inside the extent that was found? */
7019 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7020 found_type = found_key.type;
7021 if (found_key.objectid != objectid ||
7022 found_type != BTRFS_EXTENT_DATA_KEY) {
7023 /*
7024 * If we backup past the first extent we want to move forward
7025 * and see if there is an extent in front of us, otherwise we'll
7026 * say there is a hole for our whole search range which can
7027 * cause problems.
7028 */
7029 extent_end = start;
7030 goto next;
7031 }
7032
7033 found_type = btrfs_file_extent_type(leaf, item);
7034 extent_start = found_key.offset;
7035 if (found_type == BTRFS_FILE_EXTENT_REG ||
7036 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7037 extent_end = extent_start +
7038 btrfs_file_extent_num_bytes(leaf, item);
7039
7040 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7041 extent_start);
7042 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7043 size_t size;
7044 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7045 extent_end = ALIGN(extent_start + size,
7046 fs_info->sectorsize);
7047
7048 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7049 path->slots[0],
7050 extent_start);
7051 }
7052 next:
7053 if (start >= extent_end) {
7054 path->slots[0]++;
7055 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7056 ret = btrfs_next_leaf(root, path);
7057 if (ret < 0) {
7058 err = ret;
7059 goto out;
7060 }
7061 if (ret > 0)
7062 goto not_found;
7063 leaf = path->nodes[0];
7064 }
7065 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7066 if (found_key.objectid != objectid ||
7067 found_key.type != BTRFS_EXTENT_DATA_KEY)
7068 goto not_found;
7069 if (start + len <= found_key.offset)
7070 goto not_found;
7071 if (start > found_key.offset)
7072 goto next;
7073 em->start = start;
7074 em->orig_start = start;
7075 em->len = found_key.offset - start;
7076 goto not_found_em;
7077 }
7078
7079 btrfs_extent_item_to_extent_map(inode, path, item,
7080 new_inline, em);
7081
7082 if (found_type == BTRFS_FILE_EXTENT_REG ||
7083 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7084 goto insert;
7085 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7086 unsigned long ptr;
7087 char *map;
7088 size_t size;
7089 size_t extent_offset;
7090 size_t copy_size;
7091
7092 if (new_inline)
7093 goto out;
7094
7095 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7096 extent_offset = page_offset(page) + pg_offset - extent_start;
7097 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7098 size - extent_offset);
7099 em->start = extent_start + extent_offset;
7100 em->len = ALIGN(copy_size, fs_info->sectorsize);
7101 em->orig_block_len = em->len;
7102 em->orig_start = em->start;
7103 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7104 if (create == 0 && !PageUptodate(page)) {
7105 if (btrfs_file_extent_compression(leaf, item) !=
7106 BTRFS_COMPRESS_NONE) {
7107 ret = uncompress_inline(path, page, pg_offset,
7108 extent_offset, item);
7109 if (ret) {
7110 err = ret;
7111 goto out;
7112 }
7113 } else {
7114 map = kmap(page);
7115 read_extent_buffer(leaf, map + pg_offset, ptr,
7116 copy_size);
7117 if (pg_offset + copy_size < PAGE_SIZE) {
7118 memset(map + pg_offset + copy_size, 0,
7119 PAGE_SIZE - pg_offset -
7120 copy_size);
7121 }
7122 kunmap(page);
7123 }
7124 flush_dcache_page(page);
7125 } else if (create && PageUptodate(page)) {
7126 BUG();
7127 if (!trans) {
7128 kunmap(page);
7129 free_extent_map(em);
7130 em = NULL;
7131
7132 btrfs_release_path(path);
7133 trans = btrfs_join_transaction(root);
7134
7135 if (IS_ERR(trans))
7136 return ERR_CAST(trans);
7137 goto again;
7138 }
7139 map = kmap(page);
7140 write_extent_buffer(leaf, map + pg_offset, ptr,
7141 copy_size);
7142 kunmap(page);
7143 btrfs_mark_buffer_dirty(leaf);
7144 }
7145 set_extent_uptodate(io_tree, em->start,
7146 extent_map_end(em) - 1, NULL, GFP_NOFS);
7147 goto insert;
7148 }
7149 not_found:
7150 em->start = start;
7151 em->orig_start = start;
7152 em->len = len;
7153 not_found_em:
7154 em->block_start = EXTENT_MAP_HOLE;
7155 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7156 insert:
7157 btrfs_release_path(path);
7158 if (em->start > start || extent_map_end(em) <= start) {
7159 btrfs_err(fs_info,
7160 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7161 em->start, em->len, start, len);
7162 err = -EIO;
7163 goto out;
7164 }
7165
7166 err = 0;
7167 write_lock(&em_tree->lock);
7168 ret = add_extent_mapping(em_tree, em, 0);
7169 /* it is possible that someone inserted the extent into the tree
7170 * while we had the lock dropped. It is also possible that
7171 * an overlapping map exists in the tree
7172 */
7173 if (ret == -EEXIST) {
7174 struct extent_map *existing;
7175
7176 ret = 0;
7177
7178 existing = search_extent_mapping(em_tree, start, len);
7179 /*
7180 * existing will always be non-NULL, since there must be
7181 * extent causing the -EEXIST.
7182 */
7183 if (existing->start == em->start &&
7184 extent_map_end(existing) >= extent_map_end(em) &&
7185 em->block_start == existing->block_start) {
7186 /*
7187 * The existing extent map already encompasses the
7188 * entire extent map we tried to add.
7189 */
7190 free_extent_map(em);
7191 em = existing;
7192 err = 0;
7193
7194 } else if (start >= extent_map_end(existing) ||
7195 start <= existing->start) {
7196 /*
7197 * The existing extent map is the one nearest to
7198 * the [start, start + len) range which overlaps
7199 */
7200 err = merge_extent_mapping(em_tree, existing,
7201 em, start);
7202 free_extent_map(existing);
7203 if (err) {
7204 free_extent_map(em);
7205 em = NULL;
7206 }
7207 } else {
7208 free_extent_map(em);
7209 em = existing;
7210 err = 0;
7211 }
7212 }
7213 write_unlock(&em_tree->lock);
7214 out:
7215
7216 trace_btrfs_get_extent(root, inode, em);
7217
7218 btrfs_free_path(path);
7219 if (trans) {
7220 ret = btrfs_end_transaction(trans);
7221 if (!err)
7222 err = ret;
7223 }
7224 if (err) {
7225 free_extent_map(em);
7226 return ERR_PTR(err);
7227 }
7228 BUG_ON(!em); /* Error is always set */
7229 return em;
7230 }
7231
7232 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7233 struct page *page,
7234 size_t pg_offset, u64 start, u64 len,
7235 int create)
7236 {
7237 struct extent_map *em;
7238 struct extent_map *hole_em = NULL;
7239 u64 range_start = start;
7240 u64 end;
7241 u64 found;
7242 u64 found_end;
7243 int err = 0;
7244
7245 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7246 if (IS_ERR(em))
7247 return em;
7248 /*
7249 * If our em maps to:
7250 * - a hole or
7251 * - a pre-alloc extent,
7252 * there might actually be delalloc bytes behind it.
7253 */
7254 if (em->block_start != EXTENT_MAP_HOLE &&
7255 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7256 return em;
7257 else
7258 hole_em = em;
7259
7260 /* check to see if we've wrapped (len == -1 or similar) */
7261 end = start + len;
7262 if (end < start)
7263 end = (u64)-1;
7264 else
7265 end -= 1;
7266
7267 em = NULL;
7268
7269 /* ok, we didn't find anything, lets look for delalloc */
7270 found = count_range_bits(&inode->io_tree, &range_start,
7271 end, len, EXTENT_DELALLOC, 1);
7272 found_end = range_start + found;
7273 if (found_end < range_start)
7274 found_end = (u64)-1;
7275
7276 /*
7277 * we didn't find anything useful, return
7278 * the original results from get_extent()
7279 */
7280 if (range_start > end || found_end <= start) {
7281 em = hole_em;
7282 hole_em = NULL;
7283 goto out;
7284 }
7285
7286 /* adjust the range_start to make sure it doesn't
7287 * go backwards from the start they passed in
7288 */
7289 range_start = max(start, range_start);
7290 found = found_end - range_start;
7291
7292 if (found > 0) {
7293 u64 hole_start = start;
7294 u64 hole_len = len;
7295
7296 em = alloc_extent_map();
7297 if (!em) {
7298 err = -ENOMEM;
7299 goto out;
7300 }
7301 /*
7302 * when btrfs_get_extent can't find anything it
7303 * returns one huge hole
7304 *
7305 * make sure what it found really fits our range, and
7306 * adjust to make sure it is based on the start from
7307 * the caller
7308 */
7309 if (hole_em) {
7310 u64 calc_end = extent_map_end(hole_em);
7311
7312 if (calc_end <= start || (hole_em->start > end)) {
7313 free_extent_map(hole_em);
7314 hole_em = NULL;
7315 } else {
7316 hole_start = max(hole_em->start, start);
7317 hole_len = calc_end - hole_start;
7318 }
7319 }
7320 em->bdev = NULL;
7321 if (hole_em && range_start > hole_start) {
7322 /* our hole starts before our delalloc, so we
7323 * have to return just the parts of the hole
7324 * that go until the delalloc starts
7325 */
7326 em->len = min(hole_len,
7327 range_start - hole_start);
7328 em->start = hole_start;
7329 em->orig_start = hole_start;
7330 /*
7331 * don't adjust block start at all,
7332 * it is fixed at EXTENT_MAP_HOLE
7333 */
7334 em->block_start = hole_em->block_start;
7335 em->block_len = hole_len;
7336 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7337 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7338 } else {
7339 em->start = range_start;
7340 em->len = found;
7341 em->orig_start = range_start;
7342 em->block_start = EXTENT_MAP_DELALLOC;
7343 em->block_len = found;
7344 }
7345 } else if (hole_em) {
7346 return hole_em;
7347 }
7348 out:
7349
7350 free_extent_map(hole_em);
7351 if (err) {
7352 free_extent_map(em);
7353 return ERR_PTR(err);
7354 }
7355 return em;
7356 }
7357
7358 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7359 const u64 start,
7360 const u64 len,
7361 const u64 orig_start,
7362 const u64 block_start,
7363 const u64 block_len,
7364 const u64 orig_block_len,
7365 const u64 ram_bytes,
7366 const int type)
7367 {
7368 struct extent_map *em = NULL;
7369 int ret;
7370
7371 if (type != BTRFS_ORDERED_NOCOW) {
7372 em = create_io_em(inode, start, len, orig_start,
7373 block_start, block_len, orig_block_len,
7374 ram_bytes,
7375 BTRFS_COMPRESS_NONE, /* compress_type */
7376 type);
7377 if (IS_ERR(em))
7378 goto out;
7379 }
7380 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7381 len, block_len, type);
7382 if (ret) {
7383 if (em) {
7384 free_extent_map(em);
7385 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7386 start + len - 1, 0);
7387 }
7388 em = ERR_PTR(ret);
7389 }
7390 out:
7391
7392 return em;
7393 }
7394
7395 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7396 u64 start, u64 len)
7397 {
7398 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7399 struct btrfs_root *root = BTRFS_I(inode)->root;
7400 struct extent_map *em;
7401 struct btrfs_key ins;
7402 u64 alloc_hint;
7403 int ret;
7404
7405 alloc_hint = get_extent_allocation_hint(inode, start, len);
7406 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7407 0, alloc_hint, &ins, 1, 1);
7408 if (ret)
7409 return ERR_PTR(ret);
7410
7411 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7412 ins.objectid, ins.offset, ins.offset,
7413 ins.offset, BTRFS_ORDERED_REGULAR);
7414 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7415 if (IS_ERR(em))
7416 btrfs_free_reserved_extent(fs_info, ins.objectid,
7417 ins.offset, 1);
7418
7419 return em;
7420 }
7421
7422 /*
7423 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7424 * block must be cow'd
7425 */
7426 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7427 u64 *orig_start, u64 *orig_block_len,
7428 u64 *ram_bytes)
7429 {
7430 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7431 struct btrfs_path *path;
7432 int ret;
7433 struct extent_buffer *leaf;
7434 struct btrfs_root *root = BTRFS_I(inode)->root;
7435 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7436 struct btrfs_file_extent_item *fi;
7437 struct btrfs_key key;
7438 u64 disk_bytenr;
7439 u64 backref_offset;
7440 u64 extent_end;
7441 u64 num_bytes;
7442 int slot;
7443 int found_type;
7444 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7445
7446 path = btrfs_alloc_path();
7447 if (!path)
7448 return -ENOMEM;
7449
7450 ret = btrfs_lookup_file_extent(NULL, root, path,
7451 btrfs_ino(BTRFS_I(inode)), offset, 0);
7452 if (ret < 0)
7453 goto out;
7454
7455 slot = path->slots[0];
7456 if (ret == 1) {
7457 if (slot == 0) {
7458 /* can't find the item, must cow */
7459 ret = 0;
7460 goto out;
7461 }
7462 slot--;
7463 }
7464 ret = 0;
7465 leaf = path->nodes[0];
7466 btrfs_item_key_to_cpu(leaf, &key, slot);
7467 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7468 key.type != BTRFS_EXTENT_DATA_KEY) {
7469 /* not our file or wrong item type, must cow */
7470 goto out;
7471 }
7472
7473 if (key.offset > offset) {
7474 /* Wrong offset, must cow */
7475 goto out;
7476 }
7477
7478 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7479 found_type = btrfs_file_extent_type(leaf, fi);
7480 if (found_type != BTRFS_FILE_EXTENT_REG &&
7481 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7482 /* not a regular extent, must cow */
7483 goto out;
7484 }
7485
7486 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7487 goto out;
7488
7489 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7490 if (extent_end <= offset)
7491 goto out;
7492
7493 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7494 if (disk_bytenr == 0)
7495 goto out;
7496
7497 if (btrfs_file_extent_compression(leaf, fi) ||
7498 btrfs_file_extent_encryption(leaf, fi) ||
7499 btrfs_file_extent_other_encoding(leaf, fi))
7500 goto out;
7501
7502 backref_offset = btrfs_file_extent_offset(leaf, fi);
7503
7504 if (orig_start) {
7505 *orig_start = key.offset - backref_offset;
7506 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7507 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7508 }
7509
7510 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7511 goto out;
7512
7513 num_bytes = min(offset + *len, extent_end) - offset;
7514 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7515 u64 range_end;
7516
7517 range_end = round_up(offset + num_bytes,
7518 root->fs_info->sectorsize) - 1;
7519 ret = test_range_bit(io_tree, offset, range_end,
7520 EXTENT_DELALLOC, 0, NULL);
7521 if (ret) {
7522 ret = -EAGAIN;
7523 goto out;
7524 }
7525 }
7526
7527 btrfs_release_path(path);
7528
7529 /*
7530 * look for other files referencing this extent, if we
7531 * find any we must cow
7532 */
7533
7534 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7535 key.offset - backref_offset, disk_bytenr);
7536 if (ret) {
7537 ret = 0;
7538 goto out;
7539 }
7540
7541 /*
7542 * adjust disk_bytenr and num_bytes to cover just the bytes
7543 * in this extent we are about to write. If there
7544 * are any csums in that range we have to cow in order
7545 * to keep the csums correct
7546 */
7547 disk_bytenr += backref_offset;
7548 disk_bytenr += offset - key.offset;
7549 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7550 goto out;
7551 /*
7552 * all of the above have passed, it is safe to overwrite this extent
7553 * without cow
7554 */
7555 *len = num_bytes;
7556 ret = 1;
7557 out:
7558 btrfs_free_path(path);
7559 return ret;
7560 }
7561
7562 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7563 {
7564 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7565 bool found = false;
7566 void **pagep = NULL;
7567 struct page *page = NULL;
7568 unsigned long start_idx;
7569 unsigned long end_idx;
7570
7571 start_idx = start >> PAGE_SHIFT;
7572
7573 /*
7574 * end is the last byte in the last page. end == start is legal
7575 */
7576 end_idx = end >> PAGE_SHIFT;
7577
7578 rcu_read_lock();
7579
7580 /* Most of the code in this while loop is lifted from
7581 * find_get_page. It's been modified to begin searching from a
7582 * page and return just the first page found in that range. If the
7583 * found idx is less than or equal to the end idx then we know that
7584 * a page exists. If no pages are found or if those pages are
7585 * outside of the range then we're fine (yay!) */
7586 while (page == NULL &&
7587 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7588 page = radix_tree_deref_slot(pagep);
7589 if (unlikely(!page))
7590 break;
7591
7592 if (radix_tree_exception(page)) {
7593 if (radix_tree_deref_retry(page)) {
7594 page = NULL;
7595 continue;
7596 }
7597 /*
7598 * Otherwise, shmem/tmpfs must be storing a swap entry
7599 * here as an exceptional entry: so return it without
7600 * attempting to raise page count.
7601 */
7602 page = NULL;
7603 break; /* TODO: Is this relevant for this use case? */
7604 }
7605
7606 if (!page_cache_get_speculative(page)) {
7607 page = NULL;
7608 continue;
7609 }
7610
7611 /*
7612 * Has the page moved?
7613 * This is part of the lockless pagecache protocol. See
7614 * include/linux/pagemap.h for details.
7615 */
7616 if (unlikely(page != *pagep)) {
7617 put_page(page);
7618 page = NULL;
7619 }
7620 }
7621
7622 if (page) {
7623 if (page->index <= end_idx)
7624 found = true;
7625 put_page(page);
7626 }
7627
7628 rcu_read_unlock();
7629 return found;
7630 }
7631
7632 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7633 struct extent_state **cached_state, int writing)
7634 {
7635 struct btrfs_ordered_extent *ordered;
7636 int ret = 0;
7637
7638 while (1) {
7639 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7640 cached_state);
7641 /*
7642 * We're concerned with the entire range that we're going to be
7643 * doing DIO to, so we need to make sure there's no ordered
7644 * extents in this range.
7645 */
7646 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7647 lockend - lockstart + 1);
7648
7649 /*
7650 * We need to make sure there are no buffered pages in this
7651 * range either, we could have raced between the invalidate in
7652 * generic_file_direct_write and locking the extent. The
7653 * invalidate needs to happen so that reads after a write do not
7654 * get stale data.
7655 */
7656 if (!ordered &&
7657 (!writing ||
7658 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7659 break;
7660
7661 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7662 cached_state, GFP_NOFS);
7663
7664 if (ordered) {
7665 /*
7666 * If we are doing a DIO read and the ordered extent we
7667 * found is for a buffered write, we can not wait for it
7668 * to complete and retry, because if we do so we can
7669 * deadlock with concurrent buffered writes on page
7670 * locks. This happens only if our DIO read covers more
7671 * than one extent map, if at this point has already
7672 * created an ordered extent for a previous extent map
7673 * and locked its range in the inode's io tree, and a
7674 * concurrent write against that previous extent map's
7675 * range and this range started (we unlock the ranges
7676 * in the io tree only when the bios complete and
7677 * buffered writes always lock pages before attempting
7678 * to lock range in the io tree).
7679 */
7680 if (writing ||
7681 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7682 btrfs_start_ordered_extent(inode, ordered, 1);
7683 else
7684 ret = -ENOTBLK;
7685 btrfs_put_ordered_extent(ordered);
7686 } else {
7687 /*
7688 * We could trigger writeback for this range (and wait
7689 * for it to complete) and then invalidate the pages for
7690 * this range (through invalidate_inode_pages2_range()),
7691 * but that can lead us to a deadlock with a concurrent
7692 * call to readpages() (a buffered read or a defrag call
7693 * triggered a readahead) on a page lock due to an
7694 * ordered dio extent we created before but did not have
7695 * yet a corresponding bio submitted (whence it can not
7696 * complete), which makes readpages() wait for that
7697 * ordered extent to complete while holding a lock on
7698 * that page.
7699 */
7700 ret = -ENOTBLK;
7701 }
7702
7703 if (ret)
7704 break;
7705
7706 cond_resched();
7707 }
7708
7709 return ret;
7710 }
7711
7712 /* The callers of this must take lock_extent() */
7713 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7714 u64 orig_start, u64 block_start,
7715 u64 block_len, u64 orig_block_len,
7716 u64 ram_bytes, int compress_type,
7717 int type)
7718 {
7719 struct extent_map_tree *em_tree;
7720 struct extent_map *em;
7721 struct btrfs_root *root = BTRFS_I(inode)->root;
7722 int ret;
7723
7724 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7725 type == BTRFS_ORDERED_COMPRESSED ||
7726 type == BTRFS_ORDERED_NOCOW ||
7727 type == BTRFS_ORDERED_REGULAR);
7728
7729 em_tree = &BTRFS_I(inode)->extent_tree;
7730 em = alloc_extent_map();
7731 if (!em)
7732 return ERR_PTR(-ENOMEM);
7733
7734 em->start = start;
7735 em->orig_start = orig_start;
7736 em->len = len;
7737 em->block_len = block_len;
7738 em->block_start = block_start;
7739 em->bdev = root->fs_info->fs_devices->latest_bdev;
7740 em->orig_block_len = orig_block_len;
7741 em->ram_bytes = ram_bytes;
7742 em->generation = -1;
7743 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7744 if (type == BTRFS_ORDERED_PREALLOC) {
7745 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7746 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7747 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7748 em->compress_type = compress_type;
7749 }
7750
7751 do {
7752 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7753 em->start + em->len - 1, 0);
7754 write_lock(&em_tree->lock);
7755 ret = add_extent_mapping(em_tree, em, 1);
7756 write_unlock(&em_tree->lock);
7757 /*
7758 * The caller has taken lock_extent(), who could race with us
7759 * to add em?
7760 */
7761 } while (ret == -EEXIST);
7762
7763 if (ret) {
7764 free_extent_map(em);
7765 return ERR_PTR(ret);
7766 }
7767
7768 /* em got 2 refs now, callers needs to do free_extent_map once. */
7769 return em;
7770 }
7771
7772 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7773 struct buffer_head *bh_result, int create)
7774 {
7775 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7776 struct extent_map *em;
7777 struct extent_state *cached_state = NULL;
7778 struct btrfs_dio_data *dio_data = NULL;
7779 u64 start = iblock << inode->i_blkbits;
7780 u64 lockstart, lockend;
7781 u64 len = bh_result->b_size;
7782 int unlock_bits = EXTENT_LOCKED;
7783 int ret = 0;
7784
7785 if (create)
7786 unlock_bits |= EXTENT_DIRTY;
7787 else
7788 len = min_t(u64, len, fs_info->sectorsize);
7789
7790 lockstart = start;
7791 lockend = start + len - 1;
7792
7793 if (current->journal_info) {
7794 /*
7795 * Need to pull our outstanding extents and set journal_info to NULL so
7796 * that anything that needs to check if there's a transaction doesn't get
7797 * confused.
7798 */
7799 dio_data = current->journal_info;
7800 current->journal_info = NULL;
7801 }
7802
7803 /*
7804 * If this errors out it's because we couldn't invalidate pagecache for
7805 * this range and we need to fallback to buffered.
7806 */
7807 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7808 create)) {
7809 ret = -ENOTBLK;
7810 goto err;
7811 }
7812
7813 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7814 if (IS_ERR(em)) {
7815 ret = PTR_ERR(em);
7816 goto unlock_err;
7817 }
7818
7819 /*
7820 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7821 * io. INLINE is special, and we could probably kludge it in here, but
7822 * it's still buffered so for safety lets just fall back to the generic
7823 * buffered path.
7824 *
7825 * For COMPRESSED we _have_ to read the entire extent in so we can
7826 * decompress it, so there will be buffering required no matter what we
7827 * do, so go ahead and fallback to buffered.
7828 *
7829 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7830 * to buffered IO. Don't blame me, this is the price we pay for using
7831 * the generic code.
7832 */
7833 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7834 em->block_start == EXTENT_MAP_INLINE) {
7835 free_extent_map(em);
7836 ret = -ENOTBLK;
7837 goto unlock_err;
7838 }
7839
7840 /* Just a good old fashioned hole, return */
7841 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7842 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7843 free_extent_map(em);
7844 goto unlock_err;
7845 }
7846
7847 /*
7848 * We don't allocate a new extent in the following cases
7849 *
7850 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7851 * existing extent.
7852 * 2) The extent is marked as PREALLOC. We're good to go here and can
7853 * just use the extent.
7854 *
7855 */
7856 if (!create) {
7857 len = min(len, em->len - (start - em->start));
7858 lockstart = start + len;
7859 goto unlock;
7860 }
7861
7862 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7863 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7864 em->block_start != EXTENT_MAP_HOLE)) {
7865 int type;
7866 u64 block_start, orig_start, orig_block_len, ram_bytes;
7867
7868 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7869 type = BTRFS_ORDERED_PREALLOC;
7870 else
7871 type = BTRFS_ORDERED_NOCOW;
7872 len = min(len, em->len - (start - em->start));
7873 block_start = em->block_start + (start - em->start);
7874
7875 if (can_nocow_extent(inode, start, &len, &orig_start,
7876 &orig_block_len, &ram_bytes) == 1 &&
7877 btrfs_inc_nocow_writers(fs_info, block_start)) {
7878 struct extent_map *em2;
7879
7880 em2 = btrfs_create_dio_extent(inode, start, len,
7881 orig_start, block_start,
7882 len, orig_block_len,
7883 ram_bytes, type);
7884 btrfs_dec_nocow_writers(fs_info, block_start);
7885 if (type == BTRFS_ORDERED_PREALLOC) {
7886 free_extent_map(em);
7887 em = em2;
7888 }
7889 if (em2 && IS_ERR(em2)) {
7890 ret = PTR_ERR(em2);
7891 goto unlock_err;
7892 }
7893 /*
7894 * For inode marked NODATACOW or extent marked PREALLOC,
7895 * use the existing or preallocated extent, so does not
7896 * need to adjust btrfs_space_info's bytes_may_use.
7897 */
7898 btrfs_free_reserved_data_space_noquota(inode,
7899 start, len);
7900 goto unlock;
7901 }
7902 }
7903
7904 /*
7905 * this will cow the extent, reset the len in case we changed
7906 * it above
7907 */
7908 len = bh_result->b_size;
7909 free_extent_map(em);
7910 em = btrfs_new_extent_direct(inode, start, len);
7911 if (IS_ERR(em)) {
7912 ret = PTR_ERR(em);
7913 goto unlock_err;
7914 }
7915 len = min(len, em->len - (start - em->start));
7916 unlock:
7917 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7918 inode->i_blkbits;
7919 bh_result->b_size = len;
7920 bh_result->b_bdev = em->bdev;
7921 set_buffer_mapped(bh_result);
7922 if (create) {
7923 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7924 set_buffer_new(bh_result);
7925
7926 /*
7927 * Need to update the i_size under the extent lock so buffered
7928 * readers will get the updated i_size when we unlock.
7929 */
7930 if (!dio_data->overwrite && start + len > i_size_read(inode))
7931 i_size_write(inode, start + len);
7932
7933 WARN_ON(dio_data->reserve < len);
7934 dio_data->reserve -= len;
7935 dio_data->unsubmitted_oe_range_end = start + len;
7936 current->journal_info = dio_data;
7937 }
7938
7939 /*
7940 * In the case of write we need to clear and unlock the entire range,
7941 * in the case of read we need to unlock only the end area that we
7942 * aren't using if there is any left over space.
7943 */
7944 if (lockstart < lockend) {
7945 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7946 lockend, unlock_bits, 1, 0,
7947 &cached_state, GFP_NOFS);
7948 } else {
7949 free_extent_state(cached_state);
7950 }
7951
7952 free_extent_map(em);
7953
7954 return 0;
7955
7956 unlock_err:
7957 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7958 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7959 err:
7960 if (dio_data)
7961 current->journal_info = dio_data;
7962 return ret;
7963 }
7964
7965 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7966 struct bio *bio,
7967 int mirror_num)
7968 {
7969 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7970 blk_status_t ret;
7971
7972 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7973
7974 bio_get(bio);
7975
7976 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7977 if (ret)
7978 goto err;
7979
7980 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7981 err:
7982 bio_put(bio);
7983 return ret;
7984 }
7985
7986 static int btrfs_check_dio_repairable(struct inode *inode,
7987 struct bio *failed_bio,
7988 struct io_failure_record *failrec,
7989 int failed_mirror)
7990 {
7991 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7992 int num_copies;
7993
7994 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7995 if (num_copies == 1) {
7996 /*
7997 * we only have a single copy of the data, so don't bother with
7998 * all the retry and error correction code that follows. no
7999 * matter what the error is, it is very likely to persist.
8000 */
8001 btrfs_debug(fs_info,
8002 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
8003 num_copies, failrec->this_mirror, failed_mirror);
8004 return 0;
8005 }
8006
8007 failrec->failed_mirror = failed_mirror;
8008 failrec->this_mirror++;
8009 if (failrec->this_mirror == failed_mirror)
8010 failrec->this_mirror++;
8011
8012 if (failrec->this_mirror > num_copies) {
8013 btrfs_debug(fs_info,
8014 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8015 num_copies, failrec->this_mirror, failed_mirror);
8016 return 0;
8017 }
8018
8019 return 1;
8020 }
8021
8022 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
8023 struct page *page, unsigned int pgoff,
8024 u64 start, u64 end, int failed_mirror,
8025 bio_end_io_t *repair_endio, void *repair_arg)
8026 {
8027 struct io_failure_record *failrec;
8028 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8029 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8030 struct bio *bio;
8031 int isector;
8032 unsigned int read_mode = 0;
8033 int segs;
8034 int ret;
8035 blk_status_t status;
8036
8037 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8038
8039 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8040 if (ret)
8041 return errno_to_blk_status(ret);
8042
8043 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8044 failed_mirror);
8045 if (!ret) {
8046 free_io_failure(failure_tree, io_tree, failrec);
8047 return BLK_STS_IOERR;
8048 }
8049
8050 segs = bio_segments(failed_bio);
8051 if (segs > 1 ||
8052 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8053 read_mode |= REQ_FAILFAST_DEV;
8054
8055 isector = start - btrfs_io_bio(failed_bio)->logical;
8056 isector >>= inode->i_sb->s_blocksize_bits;
8057 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8058 pgoff, isector, repair_endio, repair_arg);
8059 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8060
8061 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8062 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8063 read_mode, failrec->this_mirror, failrec->in_validation);
8064
8065 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8066 if (status) {
8067 free_io_failure(failure_tree, io_tree, failrec);
8068 bio_put(bio);
8069 }
8070
8071 return status;
8072 }
8073
8074 struct btrfs_retry_complete {
8075 struct completion done;
8076 struct inode *inode;
8077 u64 start;
8078 int uptodate;
8079 };
8080
8081 static void btrfs_retry_endio_nocsum(struct bio *bio)
8082 {
8083 struct btrfs_retry_complete *done = bio->bi_private;
8084 struct inode *inode = done->inode;
8085 struct bio_vec *bvec;
8086 struct extent_io_tree *io_tree, *failure_tree;
8087 int i;
8088
8089 if (bio->bi_status)
8090 goto end;
8091
8092 ASSERT(bio->bi_vcnt == 1);
8093 io_tree = &BTRFS_I(inode)->io_tree;
8094 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8095 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8096
8097 done->uptodate = 1;
8098 ASSERT(!bio_flagged(bio, BIO_CLONED));
8099 bio_for_each_segment_all(bvec, bio, i)
8100 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8101 io_tree, done->start, bvec->bv_page,
8102 btrfs_ino(BTRFS_I(inode)), 0);
8103 end:
8104 complete(&done->done);
8105 bio_put(bio);
8106 }
8107
8108 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8109 struct btrfs_io_bio *io_bio)
8110 {
8111 struct btrfs_fs_info *fs_info;
8112 struct bio_vec bvec;
8113 struct bvec_iter iter;
8114 struct btrfs_retry_complete done;
8115 u64 start;
8116 unsigned int pgoff;
8117 u32 sectorsize;
8118 int nr_sectors;
8119 blk_status_t ret;
8120 blk_status_t err = BLK_STS_OK;
8121
8122 fs_info = BTRFS_I(inode)->root->fs_info;
8123 sectorsize = fs_info->sectorsize;
8124
8125 start = io_bio->logical;
8126 done.inode = inode;
8127 io_bio->bio.bi_iter = io_bio->iter;
8128
8129 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8130 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8131 pgoff = bvec.bv_offset;
8132
8133 next_block_or_try_again:
8134 done.uptodate = 0;
8135 done.start = start;
8136 init_completion(&done.done);
8137
8138 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8139 pgoff, start, start + sectorsize - 1,
8140 io_bio->mirror_num,
8141 btrfs_retry_endio_nocsum, &done);
8142 if (ret) {
8143 err = ret;
8144 goto next;
8145 }
8146
8147 wait_for_completion_io(&done.done);
8148
8149 if (!done.uptodate) {
8150 /* We might have another mirror, so try again */
8151 goto next_block_or_try_again;
8152 }
8153
8154 next:
8155 start += sectorsize;
8156
8157 nr_sectors--;
8158 if (nr_sectors) {
8159 pgoff += sectorsize;
8160 ASSERT(pgoff < PAGE_SIZE);
8161 goto next_block_or_try_again;
8162 }
8163 }
8164
8165 return err;
8166 }
8167
8168 static void btrfs_retry_endio(struct bio *bio)
8169 {
8170 struct btrfs_retry_complete *done = bio->bi_private;
8171 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8172 struct extent_io_tree *io_tree, *failure_tree;
8173 struct inode *inode = done->inode;
8174 struct bio_vec *bvec;
8175 int uptodate;
8176 int ret;
8177 int i;
8178
8179 if (bio->bi_status)
8180 goto end;
8181
8182 uptodate = 1;
8183
8184 ASSERT(bio->bi_vcnt == 1);
8185 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8186
8187 io_tree = &BTRFS_I(inode)->io_tree;
8188 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8189
8190 ASSERT(!bio_flagged(bio, BIO_CLONED));
8191 bio_for_each_segment_all(bvec, bio, i) {
8192 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8193 bvec->bv_offset, done->start,
8194 bvec->bv_len);
8195 if (!ret)
8196 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8197 failure_tree, io_tree, done->start,
8198 bvec->bv_page,
8199 btrfs_ino(BTRFS_I(inode)),
8200 bvec->bv_offset);
8201 else
8202 uptodate = 0;
8203 }
8204
8205 done->uptodate = uptodate;
8206 end:
8207 complete(&done->done);
8208 bio_put(bio);
8209 }
8210
8211 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8212 struct btrfs_io_bio *io_bio, blk_status_t err)
8213 {
8214 struct btrfs_fs_info *fs_info;
8215 struct bio_vec bvec;
8216 struct bvec_iter iter;
8217 struct btrfs_retry_complete done;
8218 u64 start;
8219 u64 offset = 0;
8220 u32 sectorsize;
8221 int nr_sectors;
8222 unsigned int pgoff;
8223 int csum_pos;
8224 bool uptodate = (err == 0);
8225 int ret;
8226 blk_status_t status;
8227
8228 fs_info = BTRFS_I(inode)->root->fs_info;
8229 sectorsize = fs_info->sectorsize;
8230
8231 err = BLK_STS_OK;
8232 start = io_bio->logical;
8233 done.inode = inode;
8234 io_bio->bio.bi_iter = io_bio->iter;
8235
8236 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8237 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8238
8239 pgoff = bvec.bv_offset;
8240 next_block:
8241 if (uptodate) {
8242 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8243 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8244 bvec.bv_page, pgoff, start, sectorsize);
8245 if (likely(!ret))
8246 goto next;
8247 }
8248 try_again:
8249 done.uptodate = 0;
8250 done.start = start;
8251 init_completion(&done.done);
8252
8253 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8254 pgoff, start, start + sectorsize - 1,
8255 io_bio->mirror_num, btrfs_retry_endio,
8256 &done);
8257 if (status) {
8258 err = status;
8259 goto next;
8260 }
8261
8262 wait_for_completion_io(&done.done);
8263
8264 if (!done.uptodate) {
8265 /* We might have another mirror, so try again */
8266 goto try_again;
8267 }
8268 next:
8269 offset += sectorsize;
8270 start += sectorsize;
8271
8272 ASSERT(nr_sectors);
8273
8274 nr_sectors--;
8275 if (nr_sectors) {
8276 pgoff += sectorsize;
8277 ASSERT(pgoff < PAGE_SIZE);
8278 goto next_block;
8279 }
8280 }
8281
8282 return err;
8283 }
8284
8285 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8286 struct btrfs_io_bio *io_bio, blk_status_t err)
8287 {
8288 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8289
8290 if (skip_csum) {
8291 if (unlikely(err))
8292 return __btrfs_correct_data_nocsum(inode, io_bio);
8293 else
8294 return BLK_STS_OK;
8295 } else {
8296 return __btrfs_subio_endio_read(inode, io_bio, err);
8297 }
8298 }
8299
8300 static void btrfs_endio_direct_read(struct bio *bio)
8301 {
8302 struct btrfs_dio_private *dip = bio->bi_private;
8303 struct inode *inode = dip->inode;
8304 struct bio *dio_bio;
8305 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8306 blk_status_t err = bio->bi_status;
8307
8308 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8309 err = btrfs_subio_endio_read(inode, io_bio, err);
8310
8311 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8312 dip->logical_offset + dip->bytes - 1);
8313 dio_bio = dip->dio_bio;
8314
8315 kfree(dip);
8316
8317 dio_bio->bi_status = err;
8318 dio_end_io(dio_bio);
8319
8320 if (io_bio->end_io)
8321 io_bio->end_io(io_bio, blk_status_to_errno(err));
8322 bio_put(bio);
8323 }
8324
8325 static void __endio_write_update_ordered(struct inode *inode,
8326 const u64 offset, const u64 bytes,
8327 const bool uptodate)
8328 {
8329 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8330 struct btrfs_ordered_extent *ordered = NULL;
8331 struct btrfs_workqueue *wq;
8332 btrfs_work_func_t func;
8333 u64 ordered_offset = offset;
8334 u64 ordered_bytes = bytes;
8335 u64 last_offset;
8336 int ret;
8337
8338 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8339 wq = fs_info->endio_freespace_worker;
8340 func = btrfs_freespace_write_helper;
8341 } else {
8342 wq = fs_info->endio_write_workers;
8343 func = btrfs_endio_write_helper;
8344 }
8345
8346 again:
8347 last_offset = ordered_offset;
8348 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8349 &ordered_offset,
8350 ordered_bytes,
8351 uptodate);
8352 if (!ret)
8353 goto out_test;
8354
8355 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8356 btrfs_queue_work(wq, &ordered->work);
8357 out_test:
8358 /*
8359 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8360 * in the range, we can exit.
8361 */
8362 if (ordered_offset == last_offset)
8363 return;
8364 /*
8365 * our bio might span multiple ordered extents. If we haven't
8366 * completed the accounting for the whole dio, go back and try again
8367 */
8368 if (ordered_offset < offset + bytes) {
8369 ordered_bytes = offset + bytes - ordered_offset;
8370 ordered = NULL;
8371 goto again;
8372 }
8373 }
8374
8375 static void btrfs_endio_direct_write(struct bio *bio)
8376 {
8377 struct btrfs_dio_private *dip = bio->bi_private;
8378 struct bio *dio_bio = dip->dio_bio;
8379
8380 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8381 dip->bytes, !bio->bi_status);
8382
8383 kfree(dip);
8384
8385 dio_bio->bi_status = bio->bi_status;
8386 dio_end_io(dio_bio);
8387 bio_put(bio);
8388 }
8389
8390 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8391 struct bio *bio, int mirror_num,
8392 unsigned long bio_flags, u64 offset)
8393 {
8394 struct inode *inode = private_data;
8395 blk_status_t ret;
8396 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8397 BUG_ON(ret); /* -ENOMEM */
8398 return 0;
8399 }
8400
8401 static void btrfs_end_dio_bio(struct bio *bio)
8402 {
8403 struct btrfs_dio_private *dip = bio->bi_private;
8404 blk_status_t err = bio->bi_status;
8405
8406 if (err)
8407 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8408 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8409 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8410 bio->bi_opf,
8411 (unsigned long long)bio->bi_iter.bi_sector,
8412 bio->bi_iter.bi_size, err);
8413
8414 if (dip->subio_endio)
8415 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8416
8417 if (err) {
8418 dip->errors = 1;
8419
8420 /*
8421 * before atomic variable goto zero, we must make sure
8422 * dip->errors is perceived to be set.
8423 */
8424 smp_mb__before_atomic();
8425 }
8426
8427 /* if there are more bios still pending for this dio, just exit */
8428 if (!atomic_dec_and_test(&dip->pending_bios))
8429 goto out;
8430
8431 if (dip->errors) {
8432 bio_io_error(dip->orig_bio);
8433 } else {
8434 dip->dio_bio->bi_status = BLK_STS_OK;
8435 bio_endio(dip->orig_bio);
8436 }
8437 out:
8438 bio_put(bio);
8439 }
8440
8441 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8442 struct btrfs_dio_private *dip,
8443 struct bio *bio,
8444 u64 file_offset)
8445 {
8446 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8447 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8448 blk_status_t ret;
8449
8450 /*
8451 * We load all the csum data we need when we submit
8452 * the first bio to reduce the csum tree search and
8453 * contention.
8454 */
8455 if (dip->logical_offset == file_offset) {
8456 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8457 file_offset);
8458 if (ret)
8459 return ret;
8460 }
8461
8462 if (bio == dip->orig_bio)
8463 return 0;
8464
8465 file_offset -= dip->logical_offset;
8466 file_offset >>= inode->i_sb->s_blocksize_bits;
8467 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8468
8469 return 0;
8470 }
8471
8472 static inline blk_status_t
8473 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8474 int async_submit)
8475 {
8476 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8477 struct btrfs_dio_private *dip = bio->bi_private;
8478 bool write = bio_op(bio) == REQ_OP_WRITE;
8479 blk_status_t ret;
8480
8481 if (async_submit)
8482 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8483
8484 bio_get(bio);
8485
8486 if (!write) {
8487 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8488 if (ret)
8489 goto err;
8490 }
8491
8492 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8493 goto map;
8494
8495 if (write && async_submit) {
8496 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8497 file_offset, inode,
8498 __btrfs_submit_bio_start_direct_io,
8499 __btrfs_submit_bio_done);
8500 goto err;
8501 } else if (write) {
8502 /*
8503 * If we aren't doing async submit, calculate the csum of the
8504 * bio now.
8505 */
8506 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8507 if (ret)
8508 goto err;
8509 } else {
8510 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8511 file_offset);
8512 if (ret)
8513 goto err;
8514 }
8515 map:
8516 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8517 err:
8518 bio_put(bio);
8519 return ret;
8520 }
8521
8522 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8523 {
8524 struct inode *inode = dip->inode;
8525 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8526 struct bio *bio;
8527 struct bio *orig_bio = dip->orig_bio;
8528 u64 start_sector = orig_bio->bi_iter.bi_sector;
8529 u64 file_offset = dip->logical_offset;
8530 u64 map_length;
8531 int async_submit = 0;
8532 u64 submit_len;
8533 int clone_offset = 0;
8534 int clone_len;
8535 int ret;
8536 blk_status_t status;
8537
8538 map_length = orig_bio->bi_iter.bi_size;
8539 submit_len = map_length;
8540 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8541 &map_length, NULL, 0);
8542 if (ret)
8543 return -EIO;
8544
8545 if (map_length >= submit_len) {
8546 bio = orig_bio;
8547 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8548 goto submit;
8549 }
8550
8551 /* async crcs make it difficult to collect full stripe writes. */
8552 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8553 async_submit = 0;
8554 else
8555 async_submit = 1;
8556
8557 /* bio split */
8558 ASSERT(map_length <= INT_MAX);
8559 atomic_inc(&dip->pending_bios);
8560 do {
8561 clone_len = min_t(int, submit_len, map_length);
8562
8563 /*
8564 * This will never fail as it's passing GPF_NOFS and
8565 * the allocation is backed by btrfs_bioset.
8566 */
8567 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8568 clone_len);
8569 bio->bi_private = dip;
8570 bio->bi_end_io = btrfs_end_dio_bio;
8571 btrfs_io_bio(bio)->logical = file_offset;
8572
8573 ASSERT(submit_len >= clone_len);
8574 submit_len -= clone_len;
8575 if (submit_len == 0)
8576 break;
8577
8578 /*
8579 * Increase the count before we submit the bio so we know
8580 * the end IO handler won't happen before we increase the
8581 * count. Otherwise, the dip might get freed before we're
8582 * done setting it up.
8583 */
8584 atomic_inc(&dip->pending_bios);
8585
8586 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
8587 async_submit);
8588 if (status) {
8589 bio_put(bio);
8590 atomic_dec(&dip->pending_bios);
8591 goto out_err;
8592 }
8593
8594 clone_offset += clone_len;
8595 start_sector += clone_len >> 9;
8596 file_offset += clone_len;
8597
8598 map_length = submit_len;
8599 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8600 start_sector << 9, &map_length, NULL, 0);
8601 if (ret)
8602 goto out_err;
8603 } while (submit_len > 0);
8604
8605 submit:
8606 status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8607 if (!status)
8608 return 0;
8609
8610 bio_put(bio);
8611 out_err:
8612 dip->errors = 1;
8613 /*
8614 * before atomic variable goto zero, we must
8615 * make sure dip->errors is perceived to be set.
8616 */
8617 smp_mb__before_atomic();
8618 if (atomic_dec_and_test(&dip->pending_bios))
8619 bio_io_error(dip->orig_bio);
8620
8621 /* bio_end_io() will handle error, so we needn't return it */
8622 return 0;
8623 }
8624
8625 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8626 loff_t file_offset)
8627 {
8628 struct btrfs_dio_private *dip = NULL;
8629 struct bio *bio = NULL;
8630 struct btrfs_io_bio *io_bio;
8631 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8632 int ret = 0;
8633
8634 bio = btrfs_bio_clone(dio_bio);
8635
8636 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8637 if (!dip) {
8638 ret = -ENOMEM;
8639 goto free_ordered;
8640 }
8641
8642 dip->private = dio_bio->bi_private;
8643 dip->inode = inode;
8644 dip->logical_offset = file_offset;
8645 dip->bytes = dio_bio->bi_iter.bi_size;
8646 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8647 bio->bi_private = dip;
8648 dip->orig_bio = bio;
8649 dip->dio_bio = dio_bio;
8650 atomic_set(&dip->pending_bios, 0);
8651 io_bio = btrfs_io_bio(bio);
8652 io_bio->logical = file_offset;
8653
8654 if (write) {
8655 bio->bi_end_io = btrfs_endio_direct_write;
8656 } else {
8657 bio->bi_end_io = btrfs_endio_direct_read;
8658 dip->subio_endio = btrfs_subio_endio_read;
8659 }
8660
8661 /*
8662 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8663 * even if we fail to submit a bio, because in such case we do the
8664 * corresponding error handling below and it must not be done a second
8665 * time by btrfs_direct_IO().
8666 */
8667 if (write) {
8668 struct btrfs_dio_data *dio_data = current->journal_info;
8669
8670 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8671 dip->bytes;
8672 dio_data->unsubmitted_oe_range_start =
8673 dio_data->unsubmitted_oe_range_end;
8674 }
8675
8676 ret = btrfs_submit_direct_hook(dip);
8677 if (!ret)
8678 return;
8679
8680 if (io_bio->end_io)
8681 io_bio->end_io(io_bio, ret);
8682
8683 free_ordered:
8684 /*
8685 * If we arrived here it means either we failed to submit the dip
8686 * or we either failed to clone the dio_bio or failed to allocate the
8687 * dip. If we cloned the dio_bio and allocated the dip, we can just
8688 * call bio_endio against our io_bio so that we get proper resource
8689 * cleanup if we fail to submit the dip, otherwise, we must do the
8690 * same as btrfs_endio_direct_[write|read] because we can't call these
8691 * callbacks - they require an allocated dip and a clone of dio_bio.
8692 */
8693 if (bio && dip) {
8694 bio_io_error(bio);
8695 /*
8696 * The end io callbacks free our dip, do the final put on bio
8697 * and all the cleanup and final put for dio_bio (through
8698 * dio_end_io()).
8699 */
8700 dip = NULL;
8701 bio = NULL;
8702 } else {
8703 if (write)
8704 __endio_write_update_ordered(inode,
8705 file_offset,
8706 dio_bio->bi_iter.bi_size,
8707 false);
8708 else
8709 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8710 file_offset + dio_bio->bi_iter.bi_size - 1);
8711
8712 dio_bio->bi_status = BLK_STS_IOERR;
8713 /*
8714 * Releases and cleans up our dio_bio, no need to bio_put()
8715 * nor bio_endio()/bio_io_error() against dio_bio.
8716 */
8717 dio_end_io(dio_bio);
8718 }
8719 if (bio)
8720 bio_put(bio);
8721 kfree(dip);
8722 }
8723
8724 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8725 const struct iov_iter *iter, loff_t offset)
8726 {
8727 int seg;
8728 int i;
8729 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8730 ssize_t retval = -EINVAL;
8731
8732 if (offset & blocksize_mask)
8733 goto out;
8734
8735 if (iov_iter_alignment(iter) & blocksize_mask)
8736 goto out;
8737
8738 /* If this is a write we don't need to check anymore */
8739 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8740 return 0;
8741 /*
8742 * Check to make sure we don't have duplicate iov_base's in this
8743 * iovec, if so return EINVAL, otherwise we'll get csum errors
8744 * when reading back.
8745 */
8746 for (seg = 0; seg < iter->nr_segs; seg++) {
8747 for (i = seg + 1; i < iter->nr_segs; i++) {
8748 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8749 goto out;
8750 }
8751 }
8752 retval = 0;
8753 out:
8754 return retval;
8755 }
8756
8757 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8758 {
8759 struct file *file = iocb->ki_filp;
8760 struct inode *inode = file->f_mapping->host;
8761 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8762 struct btrfs_dio_data dio_data = { 0 };
8763 struct extent_changeset *data_reserved = NULL;
8764 loff_t offset = iocb->ki_pos;
8765 size_t count = 0;
8766 int flags = 0;
8767 bool wakeup = true;
8768 bool relock = false;
8769 ssize_t ret;
8770
8771 if (check_direct_IO(fs_info, iter, offset))
8772 return 0;
8773
8774 inode_dio_begin(inode);
8775
8776 /*
8777 * The generic stuff only does filemap_write_and_wait_range, which
8778 * isn't enough if we've written compressed pages to this area, so
8779 * we need to flush the dirty pages again to make absolutely sure
8780 * that any outstanding dirty pages are on disk.
8781 */
8782 count = iov_iter_count(iter);
8783 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8784 &BTRFS_I(inode)->runtime_flags))
8785 filemap_fdatawrite_range(inode->i_mapping, offset,
8786 offset + count - 1);
8787
8788 if (iov_iter_rw(iter) == WRITE) {
8789 /*
8790 * If the write DIO is beyond the EOF, we need update
8791 * the isize, but it is protected by i_mutex. So we can
8792 * not unlock the i_mutex at this case.
8793 */
8794 if (offset + count <= inode->i_size) {
8795 dio_data.overwrite = 1;
8796 inode_unlock(inode);
8797 relock = true;
8798 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8799 ret = -EAGAIN;
8800 goto out;
8801 }
8802 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8803 offset, count);
8804 if (ret)
8805 goto out;
8806
8807 /*
8808 * We need to know how many extents we reserved so that we can
8809 * do the accounting properly if we go over the number we
8810 * originally calculated. Abuse current->journal_info for this.
8811 */
8812 dio_data.reserve = round_up(count,
8813 fs_info->sectorsize);
8814 dio_data.unsubmitted_oe_range_start = (u64)offset;
8815 dio_data.unsubmitted_oe_range_end = (u64)offset;
8816 current->journal_info = &dio_data;
8817 down_read(&BTRFS_I(inode)->dio_sem);
8818 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8819 &BTRFS_I(inode)->runtime_flags)) {
8820 inode_dio_end(inode);
8821 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8822 wakeup = false;
8823 }
8824
8825 ret = __blockdev_direct_IO(iocb, inode,
8826 fs_info->fs_devices->latest_bdev,
8827 iter, btrfs_get_blocks_direct, NULL,
8828 btrfs_submit_direct, flags);
8829 if (iov_iter_rw(iter) == WRITE) {
8830 up_read(&BTRFS_I(inode)->dio_sem);
8831 current->journal_info = NULL;
8832 if (ret < 0 && ret != -EIOCBQUEUED) {
8833 if (dio_data.reserve)
8834 btrfs_delalloc_release_space(inode, data_reserved,
8835 offset, dio_data.reserve);
8836 /*
8837 * On error we might have left some ordered extents
8838 * without submitting corresponding bios for them, so
8839 * cleanup them up to avoid other tasks getting them
8840 * and waiting for them to complete forever.
8841 */
8842 if (dio_data.unsubmitted_oe_range_start <
8843 dio_data.unsubmitted_oe_range_end)
8844 __endio_write_update_ordered(inode,
8845 dio_data.unsubmitted_oe_range_start,
8846 dio_data.unsubmitted_oe_range_end -
8847 dio_data.unsubmitted_oe_range_start,
8848 false);
8849 } else if (ret >= 0 && (size_t)ret < count)
8850 btrfs_delalloc_release_space(inode, data_reserved,
8851 offset, count - (size_t)ret);
8852 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8853 }
8854 out:
8855 if (wakeup)
8856 inode_dio_end(inode);
8857 if (relock)
8858 inode_lock(inode);
8859
8860 extent_changeset_free(data_reserved);
8861 return ret;
8862 }
8863
8864 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8865
8866 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8867 __u64 start, __u64 len)
8868 {
8869 int ret;
8870
8871 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8872 if (ret)
8873 return ret;
8874
8875 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8876 }
8877
8878 int btrfs_readpage(struct file *file, struct page *page)
8879 {
8880 struct extent_io_tree *tree;
8881 tree = &BTRFS_I(page->mapping->host)->io_tree;
8882 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8883 }
8884
8885 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8886 {
8887 struct extent_io_tree *tree;
8888 struct inode *inode = page->mapping->host;
8889 int ret;
8890
8891 if (current->flags & PF_MEMALLOC) {
8892 redirty_page_for_writepage(wbc, page);
8893 unlock_page(page);
8894 return 0;
8895 }
8896
8897 /*
8898 * If we are under memory pressure we will call this directly from the
8899 * VM, we need to make sure we have the inode referenced for the ordered
8900 * extent. If not just return like we didn't do anything.
8901 */
8902 if (!igrab(inode)) {
8903 redirty_page_for_writepage(wbc, page);
8904 return AOP_WRITEPAGE_ACTIVATE;
8905 }
8906 tree = &BTRFS_I(page->mapping->host)->io_tree;
8907 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8908 btrfs_add_delayed_iput(inode);
8909 return ret;
8910 }
8911
8912 static int btrfs_writepages(struct address_space *mapping,
8913 struct writeback_control *wbc)
8914 {
8915 struct extent_io_tree *tree;
8916
8917 tree = &BTRFS_I(mapping->host)->io_tree;
8918 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8919 }
8920
8921 static int
8922 btrfs_readpages(struct file *file, struct address_space *mapping,
8923 struct list_head *pages, unsigned nr_pages)
8924 {
8925 struct extent_io_tree *tree;
8926 tree = &BTRFS_I(mapping->host)->io_tree;
8927 return extent_readpages(tree, mapping, pages, nr_pages,
8928 btrfs_get_extent);
8929 }
8930 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8931 {
8932 struct extent_io_tree *tree;
8933 struct extent_map_tree *map;
8934 int ret;
8935
8936 tree = &BTRFS_I(page->mapping->host)->io_tree;
8937 map = &BTRFS_I(page->mapping->host)->extent_tree;
8938 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8939 if (ret == 1) {
8940 ClearPagePrivate(page);
8941 set_page_private(page, 0);
8942 put_page(page);
8943 }
8944 return ret;
8945 }
8946
8947 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8948 {
8949 if (PageWriteback(page) || PageDirty(page))
8950 return 0;
8951 return __btrfs_releasepage(page, gfp_flags);
8952 }
8953
8954 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8955 unsigned int length)
8956 {
8957 struct inode *inode = page->mapping->host;
8958 struct extent_io_tree *tree;
8959 struct btrfs_ordered_extent *ordered;
8960 struct extent_state *cached_state = NULL;
8961 u64 page_start = page_offset(page);
8962 u64 page_end = page_start + PAGE_SIZE - 1;
8963 u64 start;
8964 u64 end;
8965 int inode_evicting = inode->i_state & I_FREEING;
8966
8967 /*
8968 * we have the page locked, so new writeback can't start,
8969 * and the dirty bit won't be cleared while we are here.
8970 *
8971 * Wait for IO on this page so that we can safely clear
8972 * the PagePrivate2 bit and do ordered accounting
8973 */
8974 wait_on_page_writeback(page);
8975
8976 tree = &BTRFS_I(inode)->io_tree;
8977 if (offset) {
8978 btrfs_releasepage(page, GFP_NOFS);
8979 return;
8980 }
8981
8982 if (!inode_evicting)
8983 lock_extent_bits(tree, page_start, page_end, &cached_state);
8984 again:
8985 start = page_start;
8986 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8987 page_end - start + 1);
8988 if (ordered) {
8989 end = min(page_end, ordered->file_offset + ordered->len - 1);
8990 /*
8991 * IO on this page will never be started, so we need
8992 * to account for any ordered extents now
8993 */
8994 if (!inode_evicting)
8995 clear_extent_bit(tree, start, end,
8996 EXTENT_DIRTY | EXTENT_DELALLOC |
8997 EXTENT_DELALLOC_NEW |
8998 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8999 EXTENT_DEFRAG, 1, 0, &cached_state,
9000 GFP_NOFS);
9001 /*
9002 * whoever cleared the private bit is responsible
9003 * for the finish_ordered_io
9004 */
9005 if (TestClearPagePrivate2(page)) {
9006 struct btrfs_ordered_inode_tree *tree;
9007 u64 new_len;
9008
9009 tree = &BTRFS_I(inode)->ordered_tree;
9010
9011 spin_lock_irq(&tree->lock);
9012 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
9013 new_len = start - ordered->file_offset;
9014 if (new_len < ordered->truncated_len)
9015 ordered->truncated_len = new_len;
9016 spin_unlock_irq(&tree->lock);
9017
9018 if (btrfs_dec_test_ordered_pending(inode, &ordered,
9019 start,
9020 end - start + 1, 1))
9021 btrfs_finish_ordered_io(ordered);
9022 }
9023 btrfs_put_ordered_extent(ordered);
9024 if (!inode_evicting) {
9025 cached_state = NULL;
9026 lock_extent_bits(tree, start, end,
9027 &cached_state);
9028 }
9029
9030 start = end + 1;
9031 if (start < page_end)
9032 goto again;
9033 }
9034
9035 /*
9036 * Qgroup reserved space handler
9037 * Page here will be either
9038 * 1) Already written to disk
9039 * In this case, its reserved space is released from data rsv map
9040 * and will be freed by delayed_ref handler finally.
9041 * So even we call qgroup_free_data(), it won't decrease reserved
9042 * space.
9043 * 2) Not written to disk
9044 * This means the reserved space should be freed here. However,
9045 * if a truncate invalidates the page (by clearing PageDirty)
9046 * and the page is accounted for while allocating extent
9047 * in btrfs_check_data_free_space() we let delayed_ref to
9048 * free the entire extent.
9049 */
9050 if (PageDirty(page))
9051 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9052 if (!inode_evicting) {
9053 clear_extent_bit(tree, page_start, page_end,
9054 EXTENT_LOCKED | EXTENT_DIRTY |
9055 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9056 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9057 &cached_state, GFP_NOFS);
9058
9059 __btrfs_releasepage(page, GFP_NOFS);
9060 }
9061
9062 ClearPageChecked(page);
9063 if (PagePrivate(page)) {
9064 ClearPagePrivate(page);
9065 set_page_private(page, 0);
9066 put_page(page);
9067 }
9068 }
9069
9070 /*
9071 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9072 * called from a page fault handler when a page is first dirtied. Hence we must
9073 * be careful to check for EOF conditions here. We set the page up correctly
9074 * for a written page which means we get ENOSPC checking when writing into
9075 * holes and correct delalloc and unwritten extent mapping on filesystems that
9076 * support these features.
9077 *
9078 * We are not allowed to take the i_mutex here so we have to play games to
9079 * protect against truncate races as the page could now be beyond EOF. Because
9080 * vmtruncate() writes the inode size before removing pages, once we have the
9081 * page lock we can determine safely if the page is beyond EOF. If it is not
9082 * beyond EOF, then the page is guaranteed safe against truncation until we
9083 * unlock the page.
9084 */
9085 int btrfs_page_mkwrite(struct vm_fault *vmf)
9086 {
9087 struct page *page = vmf->page;
9088 struct inode *inode = file_inode(vmf->vma->vm_file);
9089 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9090 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9091 struct btrfs_ordered_extent *ordered;
9092 struct extent_state *cached_state = NULL;
9093 struct extent_changeset *data_reserved = NULL;
9094 char *kaddr;
9095 unsigned long zero_start;
9096 loff_t size;
9097 int ret;
9098 int reserved = 0;
9099 u64 reserved_space;
9100 u64 page_start;
9101 u64 page_end;
9102 u64 end;
9103
9104 reserved_space = PAGE_SIZE;
9105
9106 sb_start_pagefault(inode->i_sb);
9107 page_start = page_offset(page);
9108 page_end = page_start + PAGE_SIZE - 1;
9109 end = page_end;
9110
9111 /*
9112 * Reserving delalloc space after obtaining the page lock can lead to
9113 * deadlock. For example, if a dirty page is locked by this function
9114 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9115 * dirty page write out, then the btrfs_writepage() function could
9116 * end up waiting indefinitely to get a lock on the page currently
9117 * being processed by btrfs_page_mkwrite() function.
9118 */
9119 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9120 reserved_space);
9121 if (!ret) {
9122 ret = file_update_time(vmf->vma->vm_file);
9123 reserved = 1;
9124 }
9125 if (ret) {
9126 if (ret == -ENOMEM)
9127 ret = VM_FAULT_OOM;
9128 else /* -ENOSPC, -EIO, etc */
9129 ret = VM_FAULT_SIGBUS;
9130 if (reserved)
9131 goto out;
9132 goto out_noreserve;
9133 }
9134
9135 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9136 again:
9137 lock_page(page);
9138 size = i_size_read(inode);
9139
9140 if ((page->mapping != inode->i_mapping) ||
9141 (page_start >= size)) {
9142 /* page got truncated out from underneath us */
9143 goto out_unlock;
9144 }
9145 wait_on_page_writeback(page);
9146
9147 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9148 set_page_extent_mapped(page);
9149
9150 /*
9151 * we can't set the delalloc bits if there are pending ordered
9152 * extents. Drop our locks and wait for them to finish
9153 */
9154 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9155 PAGE_SIZE);
9156 if (ordered) {
9157 unlock_extent_cached(io_tree, page_start, page_end,
9158 &cached_state, GFP_NOFS);
9159 unlock_page(page);
9160 btrfs_start_ordered_extent(inode, ordered, 1);
9161 btrfs_put_ordered_extent(ordered);
9162 goto again;
9163 }
9164
9165 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9166 reserved_space = round_up(size - page_start,
9167 fs_info->sectorsize);
9168 if (reserved_space < PAGE_SIZE) {
9169 end = page_start + reserved_space - 1;
9170 btrfs_delalloc_release_space(inode, data_reserved,
9171 page_start, PAGE_SIZE - reserved_space);
9172 }
9173 }
9174
9175 /*
9176 * page_mkwrite gets called when the page is firstly dirtied after it's
9177 * faulted in, but write(2) could also dirty a page and set delalloc
9178 * bits, thus in this case for space account reason, we still need to
9179 * clear any delalloc bits within this page range since we have to
9180 * reserve data&meta space before lock_page() (see above comments).
9181 */
9182 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9183 EXTENT_DIRTY | EXTENT_DELALLOC |
9184 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9185 0, 0, &cached_state, GFP_NOFS);
9186
9187 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
9188 &cached_state, 0);
9189 if (ret) {
9190 unlock_extent_cached(io_tree, page_start, page_end,
9191 &cached_state, GFP_NOFS);
9192 ret = VM_FAULT_SIGBUS;
9193 goto out_unlock;
9194 }
9195 ret = 0;
9196
9197 /* page is wholly or partially inside EOF */
9198 if (page_start + PAGE_SIZE > size)
9199 zero_start = size & ~PAGE_MASK;
9200 else
9201 zero_start = PAGE_SIZE;
9202
9203 if (zero_start != PAGE_SIZE) {
9204 kaddr = kmap(page);
9205 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9206 flush_dcache_page(page);
9207 kunmap(page);
9208 }
9209 ClearPageChecked(page);
9210 set_page_dirty(page);
9211 SetPageUptodate(page);
9212
9213 BTRFS_I(inode)->last_trans = fs_info->generation;
9214 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9215 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9216
9217 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9218
9219 out_unlock:
9220 if (!ret) {
9221 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9222 sb_end_pagefault(inode->i_sb);
9223 extent_changeset_free(data_reserved);
9224 return VM_FAULT_LOCKED;
9225 }
9226 unlock_page(page);
9227 out:
9228 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9229 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9230 reserved_space);
9231 out_noreserve:
9232 sb_end_pagefault(inode->i_sb);
9233 extent_changeset_free(data_reserved);
9234 return ret;
9235 }
9236
9237 static int btrfs_truncate(struct inode *inode)
9238 {
9239 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9240 struct btrfs_root *root = BTRFS_I(inode)->root;
9241 struct btrfs_block_rsv *rsv;
9242 int ret = 0;
9243 int err = 0;
9244 struct btrfs_trans_handle *trans;
9245 u64 mask = fs_info->sectorsize - 1;
9246 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9247
9248 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9249 (u64)-1);
9250 if (ret)
9251 return ret;
9252
9253 /*
9254 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9255 * 3 things going on here
9256 *
9257 * 1) We need to reserve space for our orphan item and the space to
9258 * delete our orphan item. Lord knows we don't want to have a dangling
9259 * orphan item because we didn't reserve space to remove it.
9260 *
9261 * 2) We need to reserve space to update our inode.
9262 *
9263 * 3) We need to have something to cache all the space that is going to
9264 * be free'd up by the truncate operation, but also have some slack
9265 * space reserved in case it uses space during the truncate (thank you
9266 * very much snapshotting).
9267 *
9268 * And we need these to all be separate. The fact is we can use a lot of
9269 * space doing the truncate, and we have no earthly idea how much space
9270 * we will use, so we need the truncate reservation to be separate so it
9271 * doesn't end up using space reserved for updating the inode or
9272 * removing the orphan item. We also need to be able to stop the
9273 * transaction and start a new one, which means we need to be able to
9274 * update the inode several times, and we have no idea of knowing how
9275 * many times that will be, so we can't just reserve 1 item for the
9276 * entirety of the operation, so that has to be done separately as well.
9277 * Then there is the orphan item, which does indeed need to be held on
9278 * to for the whole operation, and we need nobody to touch this reserved
9279 * space except the orphan code.
9280 *
9281 * So that leaves us with
9282 *
9283 * 1) root->orphan_block_rsv - for the orphan deletion.
9284 * 2) rsv - for the truncate reservation, which we will steal from the
9285 * transaction reservation.
9286 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9287 * updating the inode.
9288 */
9289 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9290 if (!rsv)
9291 return -ENOMEM;
9292 rsv->size = min_size;
9293 rsv->failfast = 1;
9294
9295 /*
9296 * 1 for the truncate slack space
9297 * 1 for updating the inode.
9298 */
9299 trans = btrfs_start_transaction(root, 2);
9300 if (IS_ERR(trans)) {
9301 err = PTR_ERR(trans);
9302 goto out;
9303 }
9304
9305 /* Migrate the slack space for the truncate to our reserve */
9306 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9307 min_size, 0);
9308 BUG_ON(ret);
9309
9310 /*
9311 * So if we truncate and then write and fsync we normally would just
9312 * write the extents that changed, which is a problem if we need to
9313 * first truncate that entire inode. So set this flag so we write out
9314 * all of the extents in the inode to the sync log so we're completely
9315 * safe.
9316 */
9317 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9318 trans->block_rsv = rsv;
9319
9320 while (1) {
9321 ret = btrfs_truncate_inode_items(trans, root, inode,
9322 inode->i_size,
9323 BTRFS_EXTENT_DATA_KEY);
9324 trans->block_rsv = &fs_info->trans_block_rsv;
9325 if (ret != -ENOSPC && ret != -EAGAIN) {
9326 err = ret;
9327 break;
9328 }
9329
9330 ret = btrfs_update_inode(trans, root, inode);
9331 if (ret) {
9332 err = ret;
9333 break;
9334 }
9335
9336 btrfs_end_transaction(trans);
9337 btrfs_btree_balance_dirty(fs_info);
9338
9339 trans = btrfs_start_transaction(root, 2);
9340 if (IS_ERR(trans)) {
9341 ret = err = PTR_ERR(trans);
9342 trans = NULL;
9343 break;
9344 }
9345
9346 btrfs_block_rsv_release(fs_info, rsv, -1);
9347 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9348 rsv, min_size, 0);
9349 BUG_ON(ret); /* shouldn't happen */
9350 trans->block_rsv = rsv;
9351 }
9352
9353 /*
9354 * We can't call btrfs_truncate_block inside a trans handle as we could
9355 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9356 * we've truncated everything except the last little bit, and can do
9357 * btrfs_truncate_block and then update the disk_i_size.
9358 */
9359 if (ret == NEED_TRUNCATE_BLOCK) {
9360 btrfs_end_transaction(trans);
9361 btrfs_btree_balance_dirty(fs_info);
9362
9363 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9364 if (ret)
9365 goto out;
9366 trans = btrfs_start_transaction(root, 1);
9367 if (IS_ERR(trans)) {
9368 ret = PTR_ERR(trans);
9369 goto out;
9370 }
9371 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9372 }
9373
9374 if (ret == 0 && inode->i_nlink > 0) {
9375 trans->block_rsv = root->orphan_block_rsv;
9376 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9377 if (ret)
9378 err = ret;
9379 }
9380
9381 if (trans) {
9382 trans->block_rsv = &fs_info->trans_block_rsv;
9383 ret = btrfs_update_inode(trans, root, inode);
9384 if (ret && !err)
9385 err = ret;
9386
9387 ret = btrfs_end_transaction(trans);
9388 btrfs_btree_balance_dirty(fs_info);
9389 }
9390 out:
9391 btrfs_free_block_rsv(fs_info, rsv);
9392
9393 if (ret && !err)
9394 err = ret;
9395
9396 return err;
9397 }
9398
9399 /*
9400 * create a new subvolume directory/inode (helper for the ioctl).
9401 */
9402 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9403 struct btrfs_root *new_root,
9404 struct btrfs_root *parent_root,
9405 u64 new_dirid)
9406 {
9407 struct inode *inode;
9408 int err;
9409 u64 index = 0;
9410
9411 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9412 new_dirid, new_dirid,
9413 S_IFDIR | (~current_umask() & S_IRWXUGO),
9414 &index);
9415 if (IS_ERR(inode))
9416 return PTR_ERR(inode);
9417 inode->i_op = &btrfs_dir_inode_operations;
9418 inode->i_fop = &btrfs_dir_file_operations;
9419
9420 set_nlink(inode, 1);
9421 btrfs_i_size_write(BTRFS_I(inode), 0);
9422 unlock_new_inode(inode);
9423
9424 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9425 if (err)
9426 btrfs_err(new_root->fs_info,
9427 "error inheriting subvolume %llu properties: %d",
9428 new_root->root_key.objectid, err);
9429
9430 err = btrfs_update_inode(trans, new_root, inode);
9431
9432 iput(inode);
9433 return err;
9434 }
9435
9436 struct inode *btrfs_alloc_inode(struct super_block *sb)
9437 {
9438 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9439 struct btrfs_inode *ei;
9440 struct inode *inode;
9441
9442 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9443 if (!ei)
9444 return NULL;
9445
9446 ei->root = NULL;
9447 ei->generation = 0;
9448 ei->last_trans = 0;
9449 ei->last_sub_trans = 0;
9450 ei->logged_trans = 0;
9451 ei->delalloc_bytes = 0;
9452 ei->new_delalloc_bytes = 0;
9453 ei->defrag_bytes = 0;
9454 ei->disk_i_size = 0;
9455 ei->flags = 0;
9456 ei->csum_bytes = 0;
9457 ei->index_cnt = (u64)-1;
9458 ei->dir_index = 0;
9459 ei->last_unlink_trans = 0;
9460 ei->last_log_commit = 0;
9461 ei->delayed_iput_count = 0;
9462
9463 spin_lock_init(&ei->lock);
9464 ei->outstanding_extents = 0;
9465 if (sb->s_magic != BTRFS_TEST_MAGIC)
9466 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9467 BTRFS_BLOCK_RSV_DELALLOC);
9468 ei->runtime_flags = 0;
9469 ei->prop_compress = BTRFS_COMPRESS_NONE;
9470 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9471
9472 ei->delayed_node = NULL;
9473
9474 ei->i_otime.tv_sec = 0;
9475 ei->i_otime.tv_nsec = 0;
9476
9477 inode = &ei->vfs_inode;
9478 extent_map_tree_init(&ei->extent_tree);
9479 extent_io_tree_init(&ei->io_tree, inode);
9480 extent_io_tree_init(&ei->io_failure_tree, inode);
9481 ei->io_tree.track_uptodate = 1;
9482 ei->io_failure_tree.track_uptodate = 1;
9483 atomic_set(&ei->sync_writers, 0);
9484 mutex_init(&ei->log_mutex);
9485 mutex_init(&ei->delalloc_mutex);
9486 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9487 INIT_LIST_HEAD(&ei->delalloc_inodes);
9488 INIT_LIST_HEAD(&ei->delayed_iput);
9489 RB_CLEAR_NODE(&ei->rb_node);
9490 init_rwsem(&ei->dio_sem);
9491
9492 return inode;
9493 }
9494
9495 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9496 void btrfs_test_destroy_inode(struct inode *inode)
9497 {
9498 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9499 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9500 }
9501 #endif
9502
9503 static void btrfs_i_callback(struct rcu_head *head)
9504 {
9505 struct inode *inode = container_of(head, struct inode, i_rcu);
9506 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9507 }
9508
9509 void btrfs_destroy_inode(struct inode *inode)
9510 {
9511 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9512 struct btrfs_ordered_extent *ordered;
9513 struct btrfs_root *root = BTRFS_I(inode)->root;
9514
9515 WARN_ON(!hlist_empty(&inode->i_dentry));
9516 WARN_ON(inode->i_data.nrpages);
9517 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9518 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9519 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9520 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9521 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9522 WARN_ON(BTRFS_I(inode)->csum_bytes);
9523 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9524
9525 /*
9526 * This can happen where we create an inode, but somebody else also
9527 * created the same inode and we need to destroy the one we already
9528 * created.
9529 */
9530 if (!root)
9531 goto free;
9532
9533 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9534 &BTRFS_I(inode)->runtime_flags)) {
9535 btrfs_info(fs_info, "inode %llu still on the orphan list",
9536 btrfs_ino(BTRFS_I(inode)));
9537 atomic_dec(&root->orphan_inodes);
9538 }
9539
9540 while (1) {
9541 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9542 if (!ordered)
9543 break;
9544 else {
9545 btrfs_err(fs_info,
9546 "found ordered extent %llu %llu on inode cleanup",
9547 ordered->file_offset, ordered->len);
9548 btrfs_remove_ordered_extent(inode, ordered);
9549 btrfs_put_ordered_extent(ordered);
9550 btrfs_put_ordered_extent(ordered);
9551 }
9552 }
9553 btrfs_qgroup_check_reserved_leak(inode);
9554 inode_tree_del(inode);
9555 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9556 free:
9557 call_rcu(&inode->i_rcu, btrfs_i_callback);
9558 }
9559
9560 int btrfs_drop_inode(struct inode *inode)
9561 {
9562 struct btrfs_root *root = BTRFS_I(inode)->root;
9563
9564 if (root == NULL)
9565 return 1;
9566
9567 /* the snap/subvol tree is on deleting */
9568 if (btrfs_root_refs(&root->root_item) == 0)
9569 return 1;
9570 else
9571 return generic_drop_inode(inode);
9572 }
9573
9574 static void init_once(void *foo)
9575 {
9576 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9577
9578 inode_init_once(&ei->vfs_inode);
9579 }
9580
9581 void btrfs_destroy_cachep(void)
9582 {
9583 /*
9584 * Make sure all delayed rcu free inodes are flushed before we
9585 * destroy cache.
9586 */
9587 rcu_barrier();
9588 kmem_cache_destroy(btrfs_inode_cachep);
9589 kmem_cache_destroy(btrfs_trans_handle_cachep);
9590 kmem_cache_destroy(btrfs_path_cachep);
9591 kmem_cache_destroy(btrfs_free_space_cachep);
9592 }
9593
9594 int btrfs_init_cachep(void)
9595 {
9596 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9597 sizeof(struct btrfs_inode), 0,
9598 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9599 init_once);
9600 if (!btrfs_inode_cachep)
9601 goto fail;
9602
9603 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9604 sizeof(struct btrfs_trans_handle), 0,
9605 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9606 if (!btrfs_trans_handle_cachep)
9607 goto fail;
9608
9609 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9610 sizeof(struct btrfs_path), 0,
9611 SLAB_MEM_SPREAD, NULL);
9612 if (!btrfs_path_cachep)
9613 goto fail;
9614
9615 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9616 sizeof(struct btrfs_free_space), 0,
9617 SLAB_MEM_SPREAD, NULL);
9618 if (!btrfs_free_space_cachep)
9619 goto fail;
9620
9621 return 0;
9622 fail:
9623 btrfs_destroy_cachep();
9624 return -ENOMEM;
9625 }
9626
9627 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9628 u32 request_mask, unsigned int flags)
9629 {
9630 u64 delalloc_bytes;
9631 struct inode *inode = d_inode(path->dentry);
9632 u32 blocksize = inode->i_sb->s_blocksize;
9633 u32 bi_flags = BTRFS_I(inode)->flags;
9634
9635 stat->result_mask |= STATX_BTIME;
9636 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9637 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9638 if (bi_flags & BTRFS_INODE_APPEND)
9639 stat->attributes |= STATX_ATTR_APPEND;
9640 if (bi_flags & BTRFS_INODE_COMPRESS)
9641 stat->attributes |= STATX_ATTR_COMPRESSED;
9642 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9643 stat->attributes |= STATX_ATTR_IMMUTABLE;
9644 if (bi_flags & BTRFS_INODE_NODUMP)
9645 stat->attributes |= STATX_ATTR_NODUMP;
9646
9647 stat->attributes_mask |= (STATX_ATTR_APPEND |
9648 STATX_ATTR_COMPRESSED |
9649 STATX_ATTR_IMMUTABLE |
9650 STATX_ATTR_NODUMP);
9651
9652 generic_fillattr(inode, stat);
9653 stat->dev = BTRFS_I(inode)->root->anon_dev;
9654
9655 spin_lock(&BTRFS_I(inode)->lock);
9656 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9657 spin_unlock(&BTRFS_I(inode)->lock);
9658 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9659 ALIGN(delalloc_bytes, blocksize)) >> 9;
9660 return 0;
9661 }
9662
9663 static int btrfs_rename_exchange(struct inode *old_dir,
9664 struct dentry *old_dentry,
9665 struct inode *new_dir,
9666 struct dentry *new_dentry)
9667 {
9668 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9669 struct btrfs_trans_handle *trans;
9670 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9671 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9672 struct inode *new_inode = new_dentry->d_inode;
9673 struct inode *old_inode = old_dentry->d_inode;
9674 struct timespec ctime = current_time(old_inode);
9675 struct dentry *parent;
9676 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9677 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9678 u64 old_idx = 0;
9679 u64 new_idx = 0;
9680 u64 root_objectid;
9681 int ret;
9682 bool root_log_pinned = false;
9683 bool dest_log_pinned = false;
9684
9685 /* we only allow rename subvolume link between subvolumes */
9686 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9687 return -EXDEV;
9688
9689 /* close the race window with snapshot create/destroy ioctl */
9690 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9691 down_read(&fs_info->subvol_sem);
9692 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9693 down_read(&fs_info->subvol_sem);
9694
9695 /*
9696 * We want to reserve the absolute worst case amount of items. So if
9697 * both inodes are subvols and we need to unlink them then that would
9698 * require 4 item modifications, but if they are both normal inodes it
9699 * would require 5 item modifications, so we'll assume their normal
9700 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9701 * should cover the worst case number of items we'll modify.
9702 */
9703 trans = btrfs_start_transaction(root, 12);
9704 if (IS_ERR(trans)) {
9705 ret = PTR_ERR(trans);
9706 goto out_notrans;
9707 }
9708
9709 /*
9710 * We need to find a free sequence number both in the source and
9711 * in the destination directory for the exchange.
9712 */
9713 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9714 if (ret)
9715 goto out_fail;
9716 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9717 if (ret)
9718 goto out_fail;
9719
9720 BTRFS_I(old_inode)->dir_index = 0ULL;
9721 BTRFS_I(new_inode)->dir_index = 0ULL;
9722
9723 /* Reference for the source. */
9724 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9725 /* force full log commit if subvolume involved. */
9726 btrfs_set_log_full_commit(fs_info, trans);
9727 } else {
9728 btrfs_pin_log_trans(root);
9729 root_log_pinned = true;
9730 ret = btrfs_insert_inode_ref(trans, dest,
9731 new_dentry->d_name.name,
9732 new_dentry->d_name.len,
9733 old_ino,
9734 btrfs_ino(BTRFS_I(new_dir)),
9735 old_idx);
9736 if (ret)
9737 goto out_fail;
9738 }
9739
9740 /* And now for the dest. */
9741 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9742 /* force full log commit if subvolume involved. */
9743 btrfs_set_log_full_commit(fs_info, trans);
9744 } else {
9745 btrfs_pin_log_trans(dest);
9746 dest_log_pinned = true;
9747 ret = btrfs_insert_inode_ref(trans, root,
9748 old_dentry->d_name.name,
9749 old_dentry->d_name.len,
9750 new_ino,
9751 btrfs_ino(BTRFS_I(old_dir)),
9752 new_idx);
9753 if (ret)
9754 goto out_fail;
9755 }
9756
9757 /* Update inode version and ctime/mtime. */
9758 inode_inc_iversion(old_dir);
9759 inode_inc_iversion(new_dir);
9760 inode_inc_iversion(old_inode);
9761 inode_inc_iversion(new_inode);
9762 old_dir->i_ctime = old_dir->i_mtime = ctime;
9763 new_dir->i_ctime = new_dir->i_mtime = ctime;
9764 old_inode->i_ctime = ctime;
9765 new_inode->i_ctime = ctime;
9766
9767 if (old_dentry->d_parent != new_dentry->d_parent) {
9768 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9769 BTRFS_I(old_inode), 1);
9770 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9771 BTRFS_I(new_inode), 1);
9772 }
9773
9774 /* src is a subvolume */
9775 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9776 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9777 ret = btrfs_unlink_subvol(trans, root, old_dir,
9778 root_objectid,
9779 old_dentry->d_name.name,
9780 old_dentry->d_name.len);
9781 } else { /* src is an inode */
9782 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9783 BTRFS_I(old_dentry->d_inode),
9784 old_dentry->d_name.name,
9785 old_dentry->d_name.len);
9786 if (!ret)
9787 ret = btrfs_update_inode(trans, root, old_inode);
9788 }
9789 if (ret) {
9790 btrfs_abort_transaction(trans, ret);
9791 goto out_fail;
9792 }
9793
9794 /* dest is a subvolume */
9795 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9796 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9797 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9798 root_objectid,
9799 new_dentry->d_name.name,
9800 new_dentry->d_name.len);
9801 } else { /* dest is an inode */
9802 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9803 BTRFS_I(new_dentry->d_inode),
9804 new_dentry->d_name.name,
9805 new_dentry->d_name.len);
9806 if (!ret)
9807 ret = btrfs_update_inode(trans, dest, new_inode);
9808 }
9809 if (ret) {
9810 btrfs_abort_transaction(trans, ret);
9811 goto out_fail;
9812 }
9813
9814 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9815 new_dentry->d_name.name,
9816 new_dentry->d_name.len, 0, old_idx);
9817 if (ret) {
9818 btrfs_abort_transaction(trans, ret);
9819 goto out_fail;
9820 }
9821
9822 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9823 old_dentry->d_name.name,
9824 old_dentry->d_name.len, 0, new_idx);
9825 if (ret) {
9826 btrfs_abort_transaction(trans, ret);
9827 goto out_fail;
9828 }
9829
9830 if (old_inode->i_nlink == 1)
9831 BTRFS_I(old_inode)->dir_index = old_idx;
9832 if (new_inode->i_nlink == 1)
9833 BTRFS_I(new_inode)->dir_index = new_idx;
9834
9835 if (root_log_pinned) {
9836 parent = new_dentry->d_parent;
9837 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9838 parent);
9839 btrfs_end_log_trans(root);
9840 root_log_pinned = false;
9841 }
9842 if (dest_log_pinned) {
9843 parent = old_dentry->d_parent;
9844 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9845 parent);
9846 btrfs_end_log_trans(dest);
9847 dest_log_pinned = false;
9848 }
9849 out_fail:
9850 /*
9851 * If we have pinned a log and an error happened, we unpin tasks
9852 * trying to sync the log and force them to fallback to a transaction
9853 * commit if the log currently contains any of the inodes involved in
9854 * this rename operation (to ensure we do not persist a log with an
9855 * inconsistent state for any of these inodes or leading to any
9856 * inconsistencies when replayed). If the transaction was aborted, the
9857 * abortion reason is propagated to userspace when attempting to commit
9858 * the transaction. If the log does not contain any of these inodes, we
9859 * allow the tasks to sync it.
9860 */
9861 if (ret && (root_log_pinned || dest_log_pinned)) {
9862 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9863 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9864 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9865 (new_inode &&
9866 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9867 btrfs_set_log_full_commit(fs_info, trans);
9868
9869 if (root_log_pinned) {
9870 btrfs_end_log_trans(root);
9871 root_log_pinned = false;
9872 }
9873 if (dest_log_pinned) {
9874 btrfs_end_log_trans(dest);
9875 dest_log_pinned = false;
9876 }
9877 }
9878 ret = btrfs_end_transaction(trans);
9879 out_notrans:
9880 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9881 up_read(&fs_info->subvol_sem);
9882 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9883 up_read(&fs_info->subvol_sem);
9884
9885 return ret;
9886 }
9887
9888 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9889 struct btrfs_root *root,
9890 struct inode *dir,
9891 struct dentry *dentry)
9892 {
9893 int ret;
9894 struct inode *inode;
9895 u64 objectid;
9896 u64 index;
9897
9898 ret = btrfs_find_free_ino(root, &objectid);
9899 if (ret)
9900 return ret;
9901
9902 inode = btrfs_new_inode(trans, root, dir,
9903 dentry->d_name.name,
9904 dentry->d_name.len,
9905 btrfs_ino(BTRFS_I(dir)),
9906 objectid,
9907 S_IFCHR | WHITEOUT_MODE,
9908 &index);
9909
9910 if (IS_ERR(inode)) {
9911 ret = PTR_ERR(inode);
9912 return ret;
9913 }
9914
9915 inode->i_op = &btrfs_special_inode_operations;
9916 init_special_inode(inode, inode->i_mode,
9917 WHITEOUT_DEV);
9918
9919 ret = btrfs_init_inode_security(trans, inode, dir,
9920 &dentry->d_name);
9921 if (ret)
9922 goto out;
9923
9924 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9925 BTRFS_I(inode), 0, index);
9926 if (ret)
9927 goto out;
9928
9929 ret = btrfs_update_inode(trans, root, inode);
9930 out:
9931 unlock_new_inode(inode);
9932 if (ret)
9933 inode_dec_link_count(inode);
9934 iput(inode);
9935
9936 return ret;
9937 }
9938
9939 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9940 struct inode *new_dir, struct dentry *new_dentry,
9941 unsigned int flags)
9942 {
9943 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9944 struct btrfs_trans_handle *trans;
9945 unsigned int trans_num_items;
9946 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9947 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9948 struct inode *new_inode = d_inode(new_dentry);
9949 struct inode *old_inode = d_inode(old_dentry);
9950 u64 index = 0;
9951 u64 root_objectid;
9952 int ret;
9953 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9954 bool log_pinned = false;
9955
9956 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9957 return -EPERM;
9958
9959 /* we only allow rename subvolume link between subvolumes */
9960 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9961 return -EXDEV;
9962
9963 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9964 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9965 return -ENOTEMPTY;
9966
9967 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9968 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9969 return -ENOTEMPTY;
9970
9971
9972 /* check for collisions, even if the name isn't there */
9973 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9974 new_dentry->d_name.name,
9975 new_dentry->d_name.len);
9976
9977 if (ret) {
9978 if (ret == -EEXIST) {
9979 /* we shouldn't get
9980 * eexist without a new_inode */
9981 if (WARN_ON(!new_inode)) {
9982 return ret;
9983 }
9984 } else {
9985 /* maybe -EOVERFLOW */
9986 return ret;
9987 }
9988 }
9989 ret = 0;
9990
9991 /*
9992 * we're using rename to replace one file with another. Start IO on it
9993 * now so we don't add too much work to the end of the transaction
9994 */
9995 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9996 filemap_flush(old_inode->i_mapping);
9997
9998 /* close the racy window with snapshot create/destroy ioctl */
9999 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10000 down_read(&fs_info->subvol_sem);
10001 /*
10002 * We want to reserve the absolute worst case amount of items. So if
10003 * both inodes are subvols and we need to unlink them then that would
10004 * require 4 item modifications, but if they are both normal inodes it
10005 * would require 5 item modifications, so we'll assume they are normal
10006 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
10007 * should cover the worst case number of items we'll modify.
10008 * If our rename has the whiteout flag, we need more 5 units for the
10009 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
10010 * when selinux is enabled).
10011 */
10012 trans_num_items = 11;
10013 if (flags & RENAME_WHITEOUT)
10014 trans_num_items += 5;
10015 trans = btrfs_start_transaction(root, trans_num_items);
10016 if (IS_ERR(trans)) {
10017 ret = PTR_ERR(trans);
10018 goto out_notrans;
10019 }
10020
10021 if (dest != root)
10022 btrfs_record_root_in_trans(trans, dest);
10023
10024 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
10025 if (ret)
10026 goto out_fail;
10027
10028 BTRFS_I(old_inode)->dir_index = 0ULL;
10029 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10030 /* force full log commit if subvolume involved. */
10031 btrfs_set_log_full_commit(fs_info, trans);
10032 } else {
10033 btrfs_pin_log_trans(root);
10034 log_pinned = true;
10035 ret = btrfs_insert_inode_ref(trans, dest,
10036 new_dentry->d_name.name,
10037 new_dentry->d_name.len,
10038 old_ino,
10039 btrfs_ino(BTRFS_I(new_dir)), index);
10040 if (ret)
10041 goto out_fail;
10042 }
10043
10044 inode_inc_iversion(old_dir);
10045 inode_inc_iversion(new_dir);
10046 inode_inc_iversion(old_inode);
10047 old_dir->i_ctime = old_dir->i_mtime =
10048 new_dir->i_ctime = new_dir->i_mtime =
10049 old_inode->i_ctime = current_time(old_dir);
10050
10051 if (old_dentry->d_parent != new_dentry->d_parent)
10052 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10053 BTRFS_I(old_inode), 1);
10054
10055 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10056 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10057 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10058 old_dentry->d_name.name,
10059 old_dentry->d_name.len);
10060 } else {
10061 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10062 BTRFS_I(d_inode(old_dentry)),
10063 old_dentry->d_name.name,
10064 old_dentry->d_name.len);
10065 if (!ret)
10066 ret = btrfs_update_inode(trans, root, old_inode);
10067 }
10068 if (ret) {
10069 btrfs_abort_transaction(trans, ret);
10070 goto out_fail;
10071 }
10072
10073 if (new_inode) {
10074 inode_inc_iversion(new_inode);
10075 new_inode->i_ctime = current_time(new_inode);
10076 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10077 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10078 root_objectid = BTRFS_I(new_inode)->location.objectid;
10079 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10080 root_objectid,
10081 new_dentry->d_name.name,
10082 new_dentry->d_name.len);
10083 BUG_ON(new_inode->i_nlink == 0);
10084 } else {
10085 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10086 BTRFS_I(d_inode(new_dentry)),
10087 new_dentry->d_name.name,
10088 new_dentry->d_name.len);
10089 }
10090 if (!ret && new_inode->i_nlink == 0)
10091 ret = btrfs_orphan_add(trans,
10092 BTRFS_I(d_inode(new_dentry)));
10093 if (ret) {
10094 btrfs_abort_transaction(trans, ret);
10095 goto out_fail;
10096 }
10097 }
10098
10099 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10100 new_dentry->d_name.name,
10101 new_dentry->d_name.len, 0, index);
10102 if (ret) {
10103 btrfs_abort_transaction(trans, ret);
10104 goto out_fail;
10105 }
10106
10107 if (old_inode->i_nlink == 1)
10108 BTRFS_I(old_inode)->dir_index = index;
10109
10110 if (log_pinned) {
10111 struct dentry *parent = new_dentry->d_parent;
10112
10113 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10114 parent);
10115 btrfs_end_log_trans(root);
10116 log_pinned = false;
10117 }
10118
10119 if (flags & RENAME_WHITEOUT) {
10120 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10121 old_dentry);
10122
10123 if (ret) {
10124 btrfs_abort_transaction(trans, ret);
10125 goto out_fail;
10126 }
10127 }
10128 out_fail:
10129 /*
10130 * If we have pinned the log and an error happened, we unpin tasks
10131 * trying to sync the log and force them to fallback to a transaction
10132 * commit if the log currently contains any of the inodes involved in
10133 * this rename operation (to ensure we do not persist a log with an
10134 * inconsistent state for any of these inodes or leading to any
10135 * inconsistencies when replayed). If the transaction was aborted, the
10136 * abortion reason is propagated to userspace when attempting to commit
10137 * the transaction. If the log does not contain any of these inodes, we
10138 * allow the tasks to sync it.
10139 */
10140 if (ret && log_pinned) {
10141 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10142 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10143 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10144 (new_inode &&
10145 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10146 btrfs_set_log_full_commit(fs_info, trans);
10147
10148 btrfs_end_log_trans(root);
10149 log_pinned = false;
10150 }
10151 btrfs_end_transaction(trans);
10152 out_notrans:
10153 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10154 up_read(&fs_info->subvol_sem);
10155
10156 return ret;
10157 }
10158
10159 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10160 struct inode *new_dir, struct dentry *new_dentry,
10161 unsigned int flags)
10162 {
10163 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10164 return -EINVAL;
10165
10166 if (flags & RENAME_EXCHANGE)
10167 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10168 new_dentry);
10169
10170 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10171 }
10172
10173 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10174 {
10175 struct btrfs_delalloc_work *delalloc_work;
10176 struct inode *inode;
10177
10178 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10179 work);
10180 inode = delalloc_work->inode;
10181 filemap_flush(inode->i_mapping);
10182 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10183 &BTRFS_I(inode)->runtime_flags))
10184 filemap_flush(inode->i_mapping);
10185
10186 if (delalloc_work->delay_iput)
10187 btrfs_add_delayed_iput(inode);
10188 else
10189 iput(inode);
10190 complete(&delalloc_work->completion);
10191 }
10192
10193 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10194 int delay_iput)
10195 {
10196 struct btrfs_delalloc_work *work;
10197
10198 work = kmalloc(sizeof(*work), GFP_NOFS);
10199 if (!work)
10200 return NULL;
10201
10202 init_completion(&work->completion);
10203 INIT_LIST_HEAD(&work->list);
10204 work->inode = inode;
10205 work->delay_iput = delay_iput;
10206 WARN_ON_ONCE(!inode);
10207 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10208 btrfs_run_delalloc_work, NULL, NULL);
10209
10210 return work;
10211 }
10212
10213 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10214 {
10215 wait_for_completion(&work->completion);
10216 kfree(work);
10217 }
10218
10219 /*
10220 * some fairly slow code that needs optimization. This walks the list
10221 * of all the inodes with pending delalloc and forces them to disk.
10222 */
10223 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10224 int nr)
10225 {
10226 struct btrfs_inode *binode;
10227 struct inode *inode;
10228 struct btrfs_delalloc_work *work, *next;
10229 struct list_head works;
10230 struct list_head splice;
10231 int ret = 0;
10232
10233 INIT_LIST_HEAD(&works);
10234 INIT_LIST_HEAD(&splice);
10235
10236 mutex_lock(&root->delalloc_mutex);
10237 spin_lock(&root->delalloc_lock);
10238 list_splice_init(&root->delalloc_inodes, &splice);
10239 while (!list_empty(&splice)) {
10240 binode = list_entry(splice.next, struct btrfs_inode,
10241 delalloc_inodes);
10242
10243 list_move_tail(&binode->delalloc_inodes,
10244 &root->delalloc_inodes);
10245 inode = igrab(&binode->vfs_inode);
10246 if (!inode) {
10247 cond_resched_lock(&root->delalloc_lock);
10248 continue;
10249 }
10250 spin_unlock(&root->delalloc_lock);
10251
10252 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10253 if (!work) {
10254 if (delay_iput)
10255 btrfs_add_delayed_iput(inode);
10256 else
10257 iput(inode);
10258 ret = -ENOMEM;
10259 goto out;
10260 }
10261 list_add_tail(&work->list, &works);
10262 btrfs_queue_work(root->fs_info->flush_workers,
10263 &work->work);
10264 ret++;
10265 if (nr != -1 && ret >= nr)
10266 goto out;
10267 cond_resched();
10268 spin_lock(&root->delalloc_lock);
10269 }
10270 spin_unlock(&root->delalloc_lock);
10271
10272 out:
10273 list_for_each_entry_safe(work, next, &works, list) {
10274 list_del_init(&work->list);
10275 btrfs_wait_and_free_delalloc_work(work);
10276 }
10277
10278 if (!list_empty_careful(&splice)) {
10279 spin_lock(&root->delalloc_lock);
10280 list_splice_tail(&splice, &root->delalloc_inodes);
10281 spin_unlock(&root->delalloc_lock);
10282 }
10283 mutex_unlock(&root->delalloc_mutex);
10284 return ret;
10285 }
10286
10287 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10288 {
10289 struct btrfs_fs_info *fs_info = root->fs_info;
10290 int ret;
10291
10292 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10293 return -EROFS;
10294
10295 ret = __start_delalloc_inodes(root, delay_iput, -1);
10296 if (ret > 0)
10297 ret = 0;
10298 return ret;
10299 }
10300
10301 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10302 int nr)
10303 {
10304 struct btrfs_root *root;
10305 struct list_head splice;
10306 int ret;
10307
10308 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10309 return -EROFS;
10310
10311 INIT_LIST_HEAD(&splice);
10312
10313 mutex_lock(&fs_info->delalloc_root_mutex);
10314 spin_lock(&fs_info->delalloc_root_lock);
10315 list_splice_init(&fs_info->delalloc_roots, &splice);
10316 while (!list_empty(&splice) && nr) {
10317 root = list_first_entry(&splice, struct btrfs_root,
10318 delalloc_root);
10319 root = btrfs_grab_fs_root(root);
10320 BUG_ON(!root);
10321 list_move_tail(&root->delalloc_root,
10322 &fs_info->delalloc_roots);
10323 spin_unlock(&fs_info->delalloc_root_lock);
10324
10325 ret = __start_delalloc_inodes(root, delay_iput, nr);
10326 btrfs_put_fs_root(root);
10327 if (ret < 0)
10328 goto out;
10329
10330 if (nr != -1) {
10331 nr -= ret;
10332 WARN_ON(nr < 0);
10333 }
10334 spin_lock(&fs_info->delalloc_root_lock);
10335 }
10336 spin_unlock(&fs_info->delalloc_root_lock);
10337
10338 ret = 0;
10339 out:
10340 if (!list_empty_careful(&splice)) {
10341 spin_lock(&fs_info->delalloc_root_lock);
10342 list_splice_tail(&splice, &fs_info->delalloc_roots);
10343 spin_unlock(&fs_info->delalloc_root_lock);
10344 }
10345 mutex_unlock(&fs_info->delalloc_root_mutex);
10346 return ret;
10347 }
10348
10349 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10350 const char *symname)
10351 {
10352 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10353 struct btrfs_trans_handle *trans;
10354 struct btrfs_root *root = BTRFS_I(dir)->root;
10355 struct btrfs_path *path;
10356 struct btrfs_key key;
10357 struct inode *inode = NULL;
10358 int err;
10359 int drop_inode = 0;
10360 u64 objectid;
10361 u64 index = 0;
10362 int name_len;
10363 int datasize;
10364 unsigned long ptr;
10365 struct btrfs_file_extent_item *ei;
10366 struct extent_buffer *leaf;
10367
10368 name_len = strlen(symname);
10369 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10370 return -ENAMETOOLONG;
10371
10372 /*
10373 * 2 items for inode item and ref
10374 * 2 items for dir items
10375 * 1 item for updating parent inode item
10376 * 1 item for the inline extent item
10377 * 1 item for xattr if selinux is on
10378 */
10379 trans = btrfs_start_transaction(root, 7);
10380 if (IS_ERR(trans))
10381 return PTR_ERR(trans);
10382
10383 err = btrfs_find_free_ino(root, &objectid);
10384 if (err)
10385 goto out_unlock;
10386
10387 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10388 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10389 objectid, S_IFLNK|S_IRWXUGO, &index);
10390 if (IS_ERR(inode)) {
10391 err = PTR_ERR(inode);
10392 goto out_unlock;
10393 }
10394
10395 /*
10396 * If the active LSM wants to access the inode during
10397 * d_instantiate it needs these. Smack checks to see
10398 * if the filesystem supports xattrs by looking at the
10399 * ops vector.
10400 */
10401 inode->i_fop = &btrfs_file_operations;
10402 inode->i_op = &btrfs_file_inode_operations;
10403 inode->i_mapping->a_ops = &btrfs_aops;
10404 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10405
10406 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10407 if (err)
10408 goto out_unlock_inode;
10409
10410 path = btrfs_alloc_path();
10411 if (!path) {
10412 err = -ENOMEM;
10413 goto out_unlock_inode;
10414 }
10415 key.objectid = btrfs_ino(BTRFS_I(inode));
10416 key.offset = 0;
10417 key.type = BTRFS_EXTENT_DATA_KEY;
10418 datasize = btrfs_file_extent_calc_inline_size(name_len);
10419 err = btrfs_insert_empty_item(trans, root, path, &key,
10420 datasize);
10421 if (err) {
10422 btrfs_free_path(path);
10423 goto out_unlock_inode;
10424 }
10425 leaf = path->nodes[0];
10426 ei = btrfs_item_ptr(leaf, path->slots[0],
10427 struct btrfs_file_extent_item);
10428 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10429 btrfs_set_file_extent_type(leaf, ei,
10430 BTRFS_FILE_EXTENT_INLINE);
10431 btrfs_set_file_extent_encryption(leaf, ei, 0);
10432 btrfs_set_file_extent_compression(leaf, ei, 0);
10433 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10434 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10435
10436 ptr = btrfs_file_extent_inline_start(ei);
10437 write_extent_buffer(leaf, symname, ptr, name_len);
10438 btrfs_mark_buffer_dirty(leaf);
10439 btrfs_free_path(path);
10440
10441 inode->i_op = &btrfs_symlink_inode_operations;
10442 inode_nohighmem(inode);
10443 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10444 inode_set_bytes(inode, name_len);
10445 btrfs_i_size_write(BTRFS_I(inode), name_len);
10446 err = btrfs_update_inode(trans, root, inode);
10447 /*
10448 * Last step, add directory indexes for our symlink inode. This is the
10449 * last step to avoid extra cleanup of these indexes if an error happens
10450 * elsewhere above.
10451 */
10452 if (!err)
10453 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10454 BTRFS_I(inode), 0, index);
10455 if (err) {
10456 drop_inode = 1;
10457 goto out_unlock_inode;
10458 }
10459
10460 unlock_new_inode(inode);
10461 d_instantiate(dentry, inode);
10462
10463 out_unlock:
10464 btrfs_end_transaction(trans);
10465 if (drop_inode) {
10466 inode_dec_link_count(inode);
10467 iput(inode);
10468 }
10469 btrfs_btree_balance_dirty(fs_info);
10470 return err;
10471
10472 out_unlock_inode:
10473 drop_inode = 1;
10474 unlock_new_inode(inode);
10475 goto out_unlock;
10476 }
10477
10478 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10479 u64 start, u64 num_bytes, u64 min_size,
10480 loff_t actual_len, u64 *alloc_hint,
10481 struct btrfs_trans_handle *trans)
10482 {
10483 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10484 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10485 struct extent_map *em;
10486 struct btrfs_root *root = BTRFS_I(inode)->root;
10487 struct btrfs_key ins;
10488 u64 cur_offset = start;
10489 u64 i_size;
10490 u64 cur_bytes;
10491 u64 last_alloc = (u64)-1;
10492 int ret = 0;
10493 bool own_trans = true;
10494 u64 end = start + num_bytes - 1;
10495
10496 if (trans)
10497 own_trans = false;
10498 while (num_bytes > 0) {
10499 if (own_trans) {
10500 trans = btrfs_start_transaction(root, 3);
10501 if (IS_ERR(trans)) {
10502 ret = PTR_ERR(trans);
10503 break;
10504 }
10505 }
10506
10507 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10508 cur_bytes = max(cur_bytes, min_size);
10509 /*
10510 * If we are severely fragmented we could end up with really
10511 * small allocations, so if the allocator is returning small
10512 * chunks lets make its job easier by only searching for those
10513 * sized chunks.
10514 */
10515 cur_bytes = min(cur_bytes, last_alloc);
10516 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10517 min_size, 0, *alloc_hint, &ins, 1, 0);
10518 if (ret) {
10519 if (own_trans)
10520 btrfs_end_transaction(trans);
10521 break;
10522 }
10523 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10524
10525 last_alloc = ins.offset;
10526 ret = insert_reserved_file_extent(trans, inode,
10527 cur_offset, ins.objectid,
10528 ins.offset, ins.offset,
10529 ins.offset, 0, 0, 0,
10530 BTRFS_FILE_EXTENT_PREALLOC);
10531 if (ret) {
10532 btrfs_free_reserved_extent(fs_info, ins.objectid,
10533 ins.offset, 0);
10534 btrfs_abort_transaction(trans, ret);
10535 if (own_trans)
10536 btrfs_end_transaction(trans);
10537 break;
10538 }
10539
10540 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10541 cur_offset + ins.offset -1, 0);
10542
10543 em = alloc_extent_map();
10544 if (!em) {
10545 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10546 &BTRFS_I(inode)->runtime_flags);
10547 goto next;
10548 }
10549
10550 em->start = cur_offset;
10551 em->orig_start = cur_offset;
10552 em->len = ins.offset;
10553 em->block_start = ins.objectid;
10554 em->block_len = ins.offset;
10555 em->orig_block_len = ins.offset;
10556 em->ram_bytes = ins.offset;
10557 em->bdev = fs_info->fs_devices->latest_bdev;
10558 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10559 em->generation = trans->transid;
10560
10561 while (1) {
10562 write_lock(&em_tree->lock);
10563 ret = add_extent_mapping(em_tree, em, 1);
10564 write_unlock(&em_tree->lock);
10565 if (ret != -EEXIST)
10566 break;
10567 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10568 cur_offset + ins.offset - 1,
10569 0);
10570 }
10571 free_extent_map(em);
10572 next:
10573 num_bytes -= ins.offset;
10574 cur_offset += ins.offset;
10575 *alloc_hint = ins.objectid + ins.offset;
10576
10577 inode_inc_iversion(inode);
10578 inode->i_ctime = current_time(inode);
10579 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10580 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10581 (actual_len > inode->i_size) &&
10582 (cur_offset > inode->i_size)) {
10583 if (cur_offset > actual_len)
10584 i_size = actual_len;
10585 else
10586 i_size = cur_offset;
10587 i_size_write(inode, i_size);
10588 btrfs_ordered_update_i_size(inode, i_size, NULL);
10589 }
10590
10591 ret = btrfs_update_inode(trans, root, inode);
10592
10593 if (ret) {
10594 btrfs_abort_transaction(trans, ret);
10595 if (own_trans)
10596 btrfs_end_transaction(trans);
10597 break;
10598 }
10599
10600 if (own_trans)
10601 btrfs_end_transaction(trans);
10602 }
10603 if (cur_offset < end)
10604 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10605 end - cur_offset + 1);
10606 return ret;
10607 }
10608
10609 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10610 u64 start, u64 num_bytes, u64 min_size,
10611 loff_t actual_len, u64 *alloc_hint)
10612 {
10613 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10614 min_size, actual_len, alloc_hint,
10615 NULL);
10616 }
10617
10618 int btrfs_prealloc_file_range_trans(struct inode *inode,
10619 struct btrfs_trans_handle *trans, int mode,
10620 u64 start, u64 num_bytes, u64 min_size,
10621 loff_t actual_len, u64 *alloc_hint)
10622 {
10623 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10624 min_size, actual_len, alloc_hint, trans);
10625 }
10626
10627 static int btrfs_set_page_dirty(struct page *page)
10628 {
10629 return __set_page_dirty_nobuffers(page);
10630 }
10631
10632 static int btrfs_permission(struct inode *inode, int mask)
10633 {
10634 struct btrfs_root *root = BTRFS_I(inode)->root;
10635 umode_t mode = inode->i_mode;
10636
10637 if (mask & MAY_WRITE &&
10638 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10639 if (btrfs_root_readonly(root))
10640 return -EROFS;
10641 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10642 return -EACCES;
10643 }
10644 return generic_permission(inode, mask);
10645 }
10646
10647 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10648 {
10649 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10650 struct btrfs_trans_handle *trans;
10651 struct btrfs_root *root = BTRFS_I(dir)->root;
10652 struct inode *inode = NULL;
10653 u64 objectid;
10654 u64 index;
10655 int ret = 0;
10656
10657 /*
10658 * 5 units required for adding orphan entry
10659 */
10660 trans = btrfs_start_transaction(root, 5);
10661 if (IS_ERR(trans))
10662 return PTR_ERR(trans);
10663
10664 ret = btrfs_find_free_ino(root, &objectid);
10665 if (ret)
10666 goto out;
10667
10668 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10669 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10670 if (IS_ERR(inode)) {
10671 ret = PTR_ERR(inode);
10672 inode = NULL;
10673 goto out;
10674 }
10675
10676 inode->i_fop = &btrfs_file_operations;
10677 inode->i_op = &btrfs_file_inode_operations;
10678
10679 inode->i_mapping->a_ops = &btrfs_aops;
10680 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10681
10682 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10683 if (ret)
10684 goto out_inode;
10685
10686 ret = btrfs_update_inode(trans, root, inode);
10687 if (ret)
10688 goto out_inode;
10689 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10690 if (ret)
10691 goto out_inode;
10692
10693 /*
10694 * We set number of links to 0 in btrfs_new_inode(), and here we set
10695 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10696 * through:
10697 *
10698 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10699 */
10700 set_nlink(inode, 1);
10701 unlock_new_inode(inode);
10702 d_tmpfile(dentry, inode);
10703 mark_inode_dirty(inode);
10704
10705 out:
10706 btrfs_end_transaction(trans);
10707 if (ret)
10708 iput(inode);
10709 btrfs_balance_delayed_items(fs_info);
10710 btrfs_btree_balance_dirty(fs_info);
10711 return ret;
10712
10713 out_inode:
10714 unlock_new_inode(inode);
10715 goto out;
10716
10717 }
10718
10719 __attribute__((const))
10720 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10721 {
10722 return -EAGAIN;
10723 }
10724
10725 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10726 {
10727 struct inode *inode = private_data;
10728 return btrfs_sb(inode->i_sb);
10729 }
10730
10731 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10732 u64 start, u64 end)
10733 {
10734 struct inode *inode = private_data;
10735 u64 isize;
10736
10737 isize = i_size_read(inode);
10738 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10739 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10740 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10741 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10742 }
10743 }
10744
10745 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10746 {
10747 struct inode *inode = private_data;
10748 unsigned long index = start >> PAGE_SHIFT;
10749 unsigned long end_index = end >> PAGE_SHIFT;
10750 struct page *page;
10751
10752 while (index <= end_index) {
10753 page = find_get_page(inode->i_mapping, index);
10754 ASSERT(page); /* Pages should be in the extent_io_tree */
10755 set_page_writeback(page);
10756 put_page(page);
10757 index++;
10758 }
10759 }
10760
10761 static const struct inode_operations btrfs_dir_inode_operations = {
10762 .getattr = btrfs_getattr,
10763 .lookup = btrfs_lookup,
10764 .create = btrfs_create,
10765 .unlink = btrfs_unlink,
10766 .link = btrfs_link,
10767 .mkdir = btrfs_mkdir,
10768 .rmdir = btrfs_rmdir,
10769 .rename = btrfs_rename2,
10770 .symlink = btrfs_symlink,
10771 .setattr = btrfs_setattr,
10772 .mknod = btrfs_mknod,
10773 .listxattr = btrfs_listxattr,
10774 .permission = btrfs_permission,
10775 .get_acl = btrfs_get_acl,
10776 .set_acl = btrfs_set_acl,
10777 .update_time = btrfs_update_time,
10778 .tmpfile = btrfs_tmpfile,
10779 };
10780 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10781 .lookup = btrfs_lookup,
10782 .permission = btrfs_permission,
10783 .update_time = btrfs_update_time,
10784 };
10785
10786 static const struct file_operations btrfs_dir_file_operations = {
10787 .llseek = generic_file_llseek,
10788 .read = generic_read_dir,
10789 .iterate_shared = btrfs_real_readdir,
10790 .open = btrfs_opendir,
10791 .unlocked_ioctl = btrfs_ioctl,
10792 #ifdef CONFIG_COMPAT
10793 .compat_ioctl = btrfs_compat_ioctl,
10794 #endif
10795 .release = btrfs_release_file,
10796 .fsync = btrfs_sync_file,
10797 };
10798
10799 static const struct extent_io_ops btrfs_extent_io_ops = {
10800 /* mandatory callbacks */
10801 .submit_bio_hook = btrfs_submit_bio_hook,
10802 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10803 .merge_bio_hook = btrfs_merge_bio_hook,
10804 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10805 .tree_fs_info = iotree_fs_info,
10806 .set_range_writeback = btrfs_set_range_writeback,
10807
10808 /* optional callbacks */
10809 .fill_delalloc = run_delalloc_range,
10810 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10811 .writepage_start_hook = btrfs_writepage_start_hook,
10812 .set_bit_hook = btrfs_set_bit_hook,
10813 .clear_bit_hook = btrfs_clear_bit_hook,
10814 .merge_extent_hook = btrfs_merge_extent_hook,
10815 .split_extent_hook = btrfs_split_extent_hook,
10816 .check_extent_io_range = btrfs_check_extent_io_range,
10817 };
10818
10819 /*
10820 * btrfs doesn't support the bmap operation because swapfiles
10821 * use bmap to make a mapping of extents in the file. They assume
10822 * these extents won't change over the life of the file and they
10823 * use the bmap result to do IO directly to the drive.
10824 *
10825 * the btrfs bmap call would return logical addresses that aren't
10826 * suitable for IO and they also will change frequently as COW
10827 * operations happen. So, swapfile + btrfs == corruption.
10828 *
10829 * For now we're avoiding this by dropping bmap.
10830 */
10831 static const struct address_space_operations btrfs_aops = {
10832 .readpage = btrfs_readpage,
10833 .writepage = btrfs_writepage,
10834 .writepages = btrfs_writepages,
10835 .readpages = btrfs_readpages,
10836 .direct_IO = btrfs_direct_IO,
10837 .invalidatepage = btrfs_invalidatepage,
10838 .releasepage = btrfs_releasepage,
10839 .set_page_dirty = btrfs_set_page_dirty,
10840 .error_remove_page = generic_error_remove_page,
10841 };
10842
10843 static const struct address_space_operations btrfs_symlink_aops = {
10844 .readpage = btrfs_readpage,
10845 .writepage = btrfs_writepage,
10846 .invalidatepage = btrfs_invalidatepage,
10847 .releasepage = btrfs_releasepage,
10848 };
10849
10850 static const struct inode_operations btrfs_file_inode_operations = {
10851 .getattr = btrfs_getattr,
10852 .setattr = btrfs_setattr,
10853 .listxattr = btrfs_listxattr,
10854 .permission = btrfs_permission,
10855 .fiemap = btrfs_fiemap,
10856 .get_acl = btrfs_get_acl,
10857 .set_acl = btrfs_set_acl,
10858 .update_time = btrfs_update_time,
10859 };
10860 static const struct inode_operations btrfs_special_inode_operations = {
10861 .getattr = btrfs_getattr,
10862 .setattr = btrfs_setattr,
10863 .permission = btrfs_permission,
10864 .listxattr = btrfs_listxattr,
10865 .get_acl = btrfs_get_acl,
10866 .set_acl = btrfs_set_acl,
10867 .update_time = btrfs_update_time,
10868 };
10869 static const struct inode_operations btrfs_symlink_inode_operations = {
10870 .get_link = page_get_link,
10871 .getattr = btrfs_getattr,
10872 .setattr = btrfs_setattr,
10873 .permission = btrfs_permission,
10874 .listxattr = btrfs_listxattr,
10875 .update_time = btrfs_update_time,
10876 };
10877
10878 const struct dentry_operations btrfs_dentry_operations = {
10879 .d_delete = btrfs_dentry_delete,
10880 .d_release = btrfs_dentry_release,
10881 };
Linux with added FPC-III support