2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
93 static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
106 err = btrfs_init_acl(trans, inode, dir);
108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
121 struct page **compressed_pages)
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
129 struct btrfs_file_extent_item *ei;
132 size_t cur_size = size;
134 unsigned long offset;
136 if (compressed_size && compressed_pages)
137 cur_size = compressed_size;
139 path = btrfs_alloc_path();
143 path->leave_spinning = 1;
145 key.objectid = btrfs_ino(inode);
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
158 leaf = path->nodes[0];
159 ei = btrfs_item_ptr(leaf, path->slots[0],
160 struct btrfs_file_extent_item);
161 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
162 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
163 btrfs_set_file_extent_encryption(leaf, ei, 0);
164 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
165 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
166 ptr = btrfs_file_extent_inline_start(ei);
168 if (compress_type != BTRFS_COMPRESS_NONE) {
171 while (compressed_size > 0) {
172 cpage = compressed_pages[i];
173 cur_size = min_t(unsigned long, compressed_size,
176 kaddr = kmap_atomic(cpage, KM_USER0);
177 write_extent_buffer(leaf, kaddr, ptr, cur_size);
178 kunmap_atomic(kaddr, KM_USER0);
182 compressed_size -= cur_size;
184 btrfs_set_file_extent_compression(leaf, ei,
187 page = find_get_page(inode->i_mapping,
188 start >> PAGE_CACHE_SHIFT);
189 btrfs_set_file_extent_compression(leaf, ei, 0);
190 kaddr = kmap_atomic(page, KM_USER0);
191 offset = start & (PAGE_CACHE_SIZE - 1);
192 write_extent_buffer(leaf, kaddr + offset, ptr, size);
193 kunmap_atomic(kaddr, KM_USER0);
194 page_cache_release(page);
196 btrfs_mark_buffer_dirty(leaf);
197 btrfs_free_path(path);
200 * we're an inline extent, so nobody can
201 * extend the file past i_size without locking
202 * a page we already have locked.
204 * We must do any isize and inode updates
205 * before we unlock the pages. Otherwise we
206 * could end up racing with unlink.
208 BTRFS_I(inode)->disk_i_size = inode->i_size;
209 btrfs_update_inode(trans, root, inode);
213 btrfs_free_path(path);
219 * conditionally insert an inline extent into the file. This
220 * does the checks required to make sure the data is small enough
221 * to fit as an inline extent.
223 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
224 struct btrfs_root *root,
225 struct inode *inode, u64 start, u64 end,
226 size_t compressed_size, int compress_type,
227 struct page **compressed_pages)
229 u64 isize = i_size_read(inode);
230 u64 actual_end = min(end + 1, isize);
231 u64 inline_len = actual_end - start;
232 u64 aligned_end = (end + root->sectorsize - 1) &
233 ~((u64)root->sectorsize - 1);
235 u64 data_len = inline_len;
239 data_len = compressed_size;
242 actual_end >= PAGE_CACHE_SIZE ||
243 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
245 (actual_end & (root->sectorsize - 1)) == 0) ||
247 data_len > root->fs_info->max_inline) {
251 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
259 compress_type, compressed_pages);
261 btrfs_delalloc_release_metadata(inode, end + 1 - start);
262 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
266 struct async_extent {
271 unsigned long nr_pages;
273 struct list_head list;
278 struct btrfs_root *root;
279 struct page *locked_page;
282 struct list_head extents;
283 struct btrfs_work work;
286 static noinline int add_async_extent(struct async_cow *cow,
287 u64 start, u64 ram_size,
290 unsigned long nr_pages,
293 struct async_extent *async_extent;
295 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
296 BUG_ON(!async_extent);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 async_extent->compress_type = compress_type;
303 list_add_tail(&async_extent->list, &cow->extents);
308 * we create compressed extents in two phases. The first
309 * phase compresses a range of pages that have already been
310 * locked (both pages and state bits are locked).
312 * This is done inside an ordered work queue, and the compression
313 * is spread across many cpus. The actual IO submission is step
314 * two, and the ordered work queue takes care of making sure that
315 * happens in the same order things were put onto the queue by
316 * writepages and friends.
318 * If this code finds it can't get good compression, it puts an
319 * entry onto the work queue to write the uncompressed bytes. This
320 * makes sure that both compressed inodes and uncompressed inodes
321 * are written in the same order that pdflush sent them down.
323 static noinline int compress_file_range(struct inode *inode,
324 struct page *locked_page,
326 struct async_cow *async_cow,
329 struct btrfs_root *root = BTRFS_I(inode)->root;
330 struct btrfs_trans_handle *trans;
332 u64 blocksize = root->sectorsize;
334 u64 isize = i_size_read(inode);
336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
342 unsigned long max_uncompressed = 128 * 1024;
345 int compress_type = root->fs_info->compress_type;
347 /* if this is a small write inside eof, kick off a defragbot */
348 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
349 btrfs_add_inode_defrag(NULL, inode);
351 actual_end = min_t(u64, isize, end + 1);
354 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
355 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
358 * we don't want to send crud past the end of i_size through
359 * compression, that's just a waste of CPU time. So, if the
360 * end of the file is before the start of our current
361 * requested range of bytes, we bail out to the uncompressed
362 * cleanup code that can deal with all of this.
364 * It isn't really the fastest way to fix things, but this is a
365 * very uncommon corner.
367 if (actual_end <= start)
368 goto cleanup_and_bail_uncompressed;
370 total_compressed = actual_end - start;
372 /* we want to make sure that amount of ram required to uncompress
373 * an extent is reasonable, so we limit the total size in ram
374 * of a compressed extent to 128k. This is a crucial number
375 * because it also controls how easily we can spread reads across
376 * cpus for decompression.
378 * We also want to make sure the amount of IO required to do
379 * a random read is reasonably small, so we limit the size of
380 * a compressed extent to 128k.
382 total_compressed = min(total_compressed, max_uncompressed);
383 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
384 num_bytes = max(blocksize, num_bytes);
389 * we do compression for mount -o compress and when the
390 * inode has not been flagged as nocompress. This flag can
391 * change at any time if we discover bad compression ratios.
393 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
394 (btrfs_test_opt(root, COMPRESS) ||
395 (BTRFS_I(inode)->force_compress) ||
396 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
398 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
400 /* just bail out to the uncompressed code */
404 if (BTRFS_I(inode)->force_compress)
405 compress_type = BTRFS_I(inode)->force_compress;
407 ret = btrfs_compress_pages(compress_type,
408 inode->i_mapping, start,
409 total_compressed, pages,
410 nr_pages, &nr_pages_ret,
416 unsigned long offset = total_compressed &
417 (PAGE_CACHE_SIZE - 1);
418 struct page *page = pages[nr_pages_ret - 1];
421 /* zero the tail end of the last page, we might be
422 * sending it down to disk
425 kaddr = kmap_atomic(page, KM_USER0);
426 memset(kaddr + offset, 0,
427 PAGE_CACHE_SIZE - offset);
428 kunmap_atomic(kaddr, KM_USER0);
435 trans = btrfs_join_transaction(root);
436 BUG_ON(IS_ERR(trans));
437 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
439 /* lets try to make an inline extent */
440 if (ret || total_in < (actual_end - start)) {
441 /* we didn't compress the entire range, try
442 * to make an uncompressed inline extent.
444 ret = cow_file_range_inline(trans, root, inode,
445 start, end, 0, 0, NULL);
447 /* try making a compressed inline extent */
448 ret = cow_file_range_inline(trans, root, inode,
451 compress_type, pages);
455 * inline extent creation worked, we don't need
456 * to create any more async work items. Unlock
457 * and free up our temp pages.
459 extent_clear_unlock_delalloc(inode,
460 &BTRFS_I(inode)->io_tree,
462 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
463 EXTENT_CLEAR_DELALLOC |
464 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
466 btrfs_end_transaction(trans, root);
469 btrfs_end_transaction(trans, root);
474 * we aren't doing an inline extent round the compressed size
475 * up to a block size boundary so the allocator does sane
478 total_compressed = (total_compressed + blocksize - 1) &
482 * one last check to make sure the compression is really a
483 * win, compare the page count read with the blocks on disk
485 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
486 ~(PAGE_CACHE_SIZE - 1);
487 if (total_compressed >= total_in) {
490 num_bytes = total_in;
493 if (!will_compress && pages) {
495 * the compression code ran but failed to make things smaller,
496 * free any pages it allocated and our page pointer array
498 for (i = 0; i < nr_pages_ret; i++) {
499 WARN_ON(pages[i]->mapping);
500 page_cache_release(pages[i]);
504 total_compressed = 0;
507 /* flag the file so we don't compress in the future */
508 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
509 !(BTRFS_I(inode)->force_compress)) {
510 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
516 /* the async work queues will take care of doing actual
517 * allocation on disk for these compressed pages,
518 * and will submit them to the elevator.
520 add_async_extent(async_cow, start, num_bytes,
521 total_compressed, pages, nr_pages_ret,
524 if (start + num_bytes < end) {
531 cleanup_and_bail_uncompressed:
533 * No compression, but we still need to write the pages in
534 * the file we've been given so far. redirty the locked
535 * page if it corresponds to our extent and set things up
536 * for the async work queue to run cow_file_range to do
537 * the normal delalloc dance
539 if (page_offset(locked_page) >= start &&
540 page_offset(locked_page) <= end) {
541 __set_page_dirty_nobuffers(locked_page);
542 /* unlocked later on in the async handlers */
544 add_async_extent(async_cow, start, end - start + 1,
545 0, NULL, 0, BTRFS_COMPRESS_NONE);
553 for (i = 0; i < nr_pages_ret; i++) {
554 WARN_ON(pages[i]->mapping);
555 page_cache_release(pages[i]);
563 * phase two of compressed writeback. This is the ordered portion
564 * of the code, which only gets called in the order the work was
565 * queued. We walk all the async extents created by compress_file_range
566 * and send them down to the disk.
568 static noinline int submit_compressed_extents(struct inode *inode,
569 struct async_cow *async_cow)
571 struct async_extent *async_extent;
573 struct btrfs_trans_handle *trans;
574 struct btrfs_key ins;
575 struct extent_map *em;
576 struct btrfs_root *root = BTRFS_I(inode)->root;
577 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
578 struct extent_io_tree *io_tree;
581 if (list_empty(&async_cow->extents))
585 while (!list_empty(&async_cow->extents)) {
586 async_extent = list_entry(async_cow->extents.next,
587 struct async_extent, list);
588 list_del(&async_extent->list);
590 io_tree = &BTRFS_I(inode)->io_tree;
593 /* did the compression code fall back to uncompressed IO? */
594 if (!async_extent->pages) {
595 int page_started = 0;
596 unsigned long nr_written = 0;
598 lock_extent(io_tree, async_extent->start,
599 async_extent->start +
600 async_extent->ram_size - 1);
602 /* allocate blocks */
603 ret = cow_file_range(inode, async_cow->locked_page,
605 async_extent->start +
606 async_extent->ram_size - 1,
607 &page_started, &nr_written, 0);
610 * if page_started, cow_file_range inserted an
611 * inline extent and took care of all the unlocking
612 * and IO for us. Otherwise, we need to submit
613 * all those pages down to the drive.
615 if (!page_started && !ret)
616 extent_write_locked_range(io_tree,
617 inode, async_extent->start,
618 async_extent->start +
619 async_extent->ram_size - 1,
627 lock_extent(io_tree, async_extent->start,
628 async_extent->start + async_extent->ram_size - 1);
630 trans = btrfs_join_transaction(root);
631 BUG_ON(IS_ERR(trans));
632 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
633 ret = btrfs_reserve_extent(trans, root,
634 async_extent->compressed_size,
635 async_extent->compressed_size,
638 btrfs_end_transaction(trans, root);
642 for (i = 0; i < async_extent->nr_pages; i++) {
643 WARN_ON(async_extent->pages[i]->mapping);
644 page_cache_release(async_extent->pages[i]);
646 kfree(async_extent->pages);
647 async_extent->nr_pages = 0;
648 async_extent->pages = NULL;
649 unlock_extent(io_tree, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1);
656 * here we're doing allocation and writeback of the
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
663 em = alloc_extent_map();
665 em->start = async_extent->start;
666 em->len = async_extent->ram_size;
667 em->orig_start = em->start;
669 em->block_start = ins.objectid;
670 em->block_len = ins.offset;
671 em->bdev = root->fs_info->fs_devices->latest_bdev;
672 em->compress_type = async_extent->compress_type;
673 set_bit(EXTENT_FLAG_PINNED, &em->flags);
674 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
677 write_lock(&em_tree->lock);
678 ret = add_extent_mapping(em_tree, em);
679 write_unlock(&em_tree->lock);
680 if (ret != -EEXIST) {
684 btrfs_drop_extent_cache(inode, async_extent->start,
685 async_extent->start +
686 async_extent->ram_size - 1, 0);
689 ret = btrfs_add_ordered_extent_compress(inode,
692 async_extent->ram_size,
694 BTRFS_ORDERED_COMPRESSED,
695 async_extent->compress_type);
699 * clear dirty, set writeback and unlock the pages.
701 extent_clear_unlock_delalloc(inode,
702 &BTRFS_I(inode)->io_tree,
704 async_extent->start +
705 async_extent->ram_size - 1,
706 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
707 EXTENT_CLEAR_UNLOCK |
708 EXTENT_CLEAR_DELALLOC |
709 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
711 ret = btrfs_submit_compressed_write(inode,
713 async_extent->ram_size,
715 ins.offset, async_extent->pages,
716 async_extent->nr_pages);
719 alloc_hint = ins.objectid + ins.offset;
727 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
730 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731 struct extent_map *em;
734 read_lock(&em_tree->lock);
735 em = search_extent_mapping(em_tree, start, num_bytes);
738 * if block start isn't an actual block number then find the
739 * first block in this inode and use that as a hint. If that
740 * block is also bogus then just don't worry about it.
742 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
744 em = search_extent_mapping(em_tree, 0, 0);
745 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
746 alloc_hint = em->block_start;
750 alloc_hint = em->block_start;
754 read_unlock(&em_tree->lock);
760 * when extent_io.c finds a delayed allocation range in the file,
761 * the call backs end up in this code. The basic idea is to
762 * allocate extents on disk for the range, and create ordered data structs
763 * in ram to track those extents.
765 * locked_page is the page that writepage had locked already. We use
766 * it to make sure we don't do extra locks or unlocks.
768 * *page_started is set to one if we unlock locked_page and do everything
769 * required to start IO on it. It may be clean and already done with
772 static noinline int cow_file_range(struct inode *inode,
773 struct page *locked_page,
774 u64 start, u64 end, int *page_started,
775 unsigned long *nr_written,
778 struct btrfs_root *root = BTRFS_I(inode)->root;
779 struct btrfs_trans_handle *trans;
782 unsigned long ram_size;
785 u64 blocksize = root->sectorsize;
786 struct btrfs_key ins;
787 struct extent_map *em;
788 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
791 BUG_ON(btrfs_is_free_space_inode(root, inode));
792 trans = btrfs_join_transaction(root);
793 BUG_ON(IS_ERR(trans));
794 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
796 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
797 num_bytes = max(blocksize, num_bytes);
798 disk_num_bytes = num_bytes;
801 /* if this is a small write inside eof, kick off defrag */
802 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
803 btrfs_add_inode_defrag(trans, inode);
806 /* lets try to make an inline extent */
807 ret = cow_file_range_inline(trans, root, inode,
808 start, end, 0, 0, NULL);
810 extent_clear_unlock_delalloc(inode,
811 &BTRFS_I(inode)->io_tree,
813 EXTENT_CLEAR_UNLOCK_PAGE |
814 EXTENT_CLEAR_UNLOCK |
815 EXTENT_CLEAR_DELALLOC |
817 EXTENT_SET_WRITEBACK |
818 EXTENT_END_WRITEBACK);
820 *nr_written = *nr_written +
821 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
828 BUG_ON(disk_num_bytes >
829 btrfs_super_total_bytes(root->fs_info->super_copy));
831 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
832 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
834 while (disk_num_bytes > 0) {
837 cur_alloc_size = disk_num_bytes;
838 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
839 root->sectorsize, 0, alloc_hint,
843 em = alloc_extent_map();
846 em->orig_start = em->start;
847 ram_size = ins.offset;
848 em->len = ins.offset;
850 em->block_start = ins.objectid;
851 em->block_len = ins.offset;
852 em->bdev = root->fs_info->fs_devices->latest_bdev;
853 set_bit(EXTENT_FLAG_PINNED, &em->flags);
856 write_lock(&em_tree->lock);
857 ret = add_extent_mapping(em_tree, em);
858 write_unlock(&em_tree->lock);
859 if (ret != -EEXIST) {
863 btrfs_drop_extent_cache(inode, start,
864 start + ram_size - 1, 0);
867 cur_alloc_size = ins.offset;
868 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
869 ram_size, cur_alloc_size, 0);
872 if (root->root_key.objectid ==
873 BTRFS_DATA_RELOC_TREE_OBJECTID) {
874 ret = btrfs_reloc_clone_csums(inode, start,
879 if (disk_num_bytes < cur_alloc_size)
882 /* we're not doing compressed IO, don't unlock the first
883 * page (which the caller expects to stay locked), don't
884 * clear any dirty bits and don't set any writeback bits
886 * Do set the Private2 bit so we know this page was properly
887 * setup for writepage
889 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
890 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
893 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
894 start, start + ram_size - 1,
896 disk_num_bytes -= cur_alloc_size;
897 num_bytes -= cur_alloc_size;
898 alloc_hint = ins.objectid + ins.offset;
899 start += cur_alloc_size;
903 btrfs_end_transaction(trans, root);
909 * work queue call back to started compression on a file and pages
911 static noinline void async_cow_start(struct btrfs_work *work)
913 struct async_cow *async_cow;
915 async_cow = container_of(work, struct async_cow, work);
917 compress_file_range(async_cow->inode, async_cow->locked_page,
918 async_cow->start, async_cow->end, async_cow,
921 async_cow->inode = NULL;
925 * work queue call back to submit previously compressed pages
927 static noinline void async_cow_submit(struct btrfs_work *work)
929 struct async_cow *async_cow;
930 struct btrfs_root *root;
931 unsigned long nr_pages;
933 async_cow = container_of(work, struct async_cow, work);
935 root = async_cow->root;
936 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
939 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
941 if (atomic_read(&root->fs_info->async_delalloc_pages) <
943 waitqueue_active(&root->fs_info->async_submit_wait))
944 wake_up(&root->fs_info->async_submit_wait);
946 if (async_cow->inode)
947 submit_compressed_extents(async_cow->inode, async_cow);
950 static noinline void async_cow_free(struct btrfs_work *work)
952 struct async_cow *async_cow;
953 async_cow = container_of(work, struct async_cow, work);
957 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
958 u64 start, u64 end, int *page_started,
959 unsigned long *nr_written)
961 struct async_cow *async_cow;
962 struct btrfs_root *root = BTRFS_I(inode)->root;
963 unsigned long nr_pages;
965 int limit = 10 * 1024 * 1042;
967 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
968 1, 0, NULL, GFP_NOFS);
969 while (start < end) {
970 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
972 async_cow->inode = inode;
973 async_cow->root = root;
974 async_cow->locked_page = locked_page;
975 async_cow->start = start;
977 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
980 cur_end = min(end, start + 512 * 1024 - 1);
982 async_cow->end = cur_end;
983 INIT_LIST_HEAD(&async_cow->extents);
985 async_cow->work.func = async_cow_start;
986 async_cow->work.ordered_func = async_cow_submit;
987 async_cow->work.ordered_free = async_cow_free;
988 async_cow->work.flags = 0;
990 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
992 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
994 btrfs_queue_worker(&root->fs_info->delalloc_workers,
997 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
998 wait_event(root->fs_info->async_submit_wait,
999 (atomic_read(&root->fs_info->async_delalloc_pages) <
1003 while (atomic_read(&root->fs_info->async_submit_draining) &&
1004 atomic_read(&root->fs_info->async_delalloc_pages)) {
1005 wait_event(root->fs_info->async_submit_wait,
1006 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010 *nr_written += nr_pages;
1011 start = cur_end + 1;
1017 static noinline int csum_exist_in_range(struct btrfs_root *root,
1018 u64 bytenr, u64 num_bytes)
1021 struct btrfs_ordered_sum *sums;
1024 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1025 bytenr + num_bytes - 1, &list, 0);
1026 if (ret == 0 && list_empty(&list))
1029 while (!list_empty(&list)) {
1030 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031 list_del(&sums->list);
1038 * when nowcow writeback call back. This checks for snapshots or COW copies
1039 * of the extents that exist in the file, and COWs the file as required.
1041 * If no cow copies or snapshots exist, we write directly to the existing
1044 static noinline int run_delalloc_nocow(struct inode *inode,
1045 struct page *locked_page,
1046 u64 start, u64 end, int *page_started, int force,
1047 unsigned long *nr_written)
1049 struct btrfs_root *root = BTRFS_I(inode)->root;
1050 struct btrfs_trans_handle *trans;
1051 struct extent_buffer *leaf;
1052 struct btrfs_path *path;
1053 struct btrfs_file_extent_item *fi;
1054 struct btrfs_key found_key;
1067 u64 ino = btrfs_ino(inode);
1069 path = btrfs_alloc_path();
1073 nolock = btrfs_is_free_space_inode(root, inode);
1076 trans = btrfs_join_transaction_nolock(root);
1078 trans = btrfs_join_transaction(root);
1080 BUG_ON(IS_ERR(trans));
1081 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083 cow_start = (u64)-1;
1086 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0] - 1);
1093 if (found_key.objectid == ino &&
1094 found_key.type == BTRFS_EXTENT_DATA_KEY)
1099 leaf = path->nodes[0];
1100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101 ret = btrfs_next_leaf(root, path);
1106 leaf = path->nodes[0];
1112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114 if (found_key.objectid > ino ||
1115 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116 found_key.offset > end)
1119 if (found_key.offset > cur_offset) {
1120 extent_end = found_key.offset;
1125 fi = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_file_extent_item);
1127 extent_type = btrfs_file_extent_type(leaf, fi);
1129 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1131 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132 extent_offset = btrfs_file_extent_offset(leaf, fi);
1133 extent_end = found_key.offset +
1134 btrfs_file_extent_num_bytes(leaf, fi);
1135 if (extent_end <= start) {
1139 if (disk_bytenr == 0)
1141 if (btrfs_file_extent_compression(leaf, fi) ||
1142 btrfs_file_extent_encryption(leaf, fi) ||
1143 btrfs_file_extent_other_encoding(leaf, fi))
1145 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147 if (btrfs_extent_readonly(root, disk_bytenr))
1149 if (btrfs_cross_ref_exist(trans, root, ino,
1151 extent_offset, disk_bytenr))
1153 disk_bytenr += extent_offset;
1154 disk_bytenr += cur_offset - found_key.offset;
1155 num_bytes = min(end + 1, extent_end) - cur_offset;
1157 * force cow if csum exists in the range.
1158 * this ensure that csum for a given extent are
1159 * either valid or do not exist.
1161 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165 extent_end = found_key.offset +
1166 btrfs_file_extent_inline_len(leaf, fi);
1167 extent_end = ALIGN(extent_end, root->sectorsize);
1172 if (extent_end <= start) {
1177 if (cow_start == (u64)-1)
1178 cow_start = cur_offset;
1179 cur_offset = extent_end;
1180 if (cur_offset > end)
1186 btrfs_release_path(path);
1187 if (cow_start != (u64)-1) {
1188 ret = cow_file_range(inode, locked_page, cow_start,
1189 found_key.offset - 1, page_started,
1192 cow_start = (u64)-1;
1195 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196 struct extent_map *em;
1197 struct extent_map_tree *em_tree;
1198 em_tree = &BTRFS_I(inode)->extent_tree;
1199 em = alloc_extent_map();
1201 em->start = cur_offset;
1202 em->orig_start = em->start;
1203 em->len = num_bytes;
1204 em->block_len = num_bytes;
1205 em->block_start = disk_bytenr;
1206 em->bdev = root->fs_info->fs_devices->latest_bdev;
1207 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209 write_lock(&em_tree->lock);
1210 ret = add_extent_mapping(em_tree, em);
1211 write_unlock(&em_tree->lock);
1212 if (ret != -EEXIST) {
1213 free_extent_map(em);
1216 btrfs_drop_extent_cache(inode, em->start,
1217 em->start + em->len - 1, 0);
1219 type = BTRFS_ORDERED_PREALLOC;
1221 type = BTRFS_ORDERED_NOCOW;
1224 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1225 num_bytes, num_bytes, type);
1228 if (root->root_key.objectid ==
1229 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1235 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1236 cur_offset, cur_offset + num_bytes - 1,
1237 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239 EXTENT_SET_PRIVATE2);
1240 cur_offset = extent_end;
1241 if (cur_offset > end)
1244 btrfs_release_path(path);
1246 if (cur_offset <= end && cow_start == (u64)-1)
1247 cow_start = cur_offset;
1248 if (cow_start != (u64)-1) {
1249 ret = cow_file_range(inode, locked_page, cow_start, end,
1250 page_started, nr_written, 1);
1255 ret = btrfs_end_transaction_nolock(trans, root);
1258 ret = btrfs_end_transaction(trans, root);
1261 btrfs_free_path(path);
1266 * extent_io.c call back to do delayed allocation processing
1268 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1269 u64 start, u64 end, int *page_started,
1270 unsigned long *nr_written)
1273 struct btrfs_root *root = BTRFS_I(inode)->root;
1275 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1276 ret = run_delalloc_nocow(inode, locked_page, start, end,
1277 page_started, 1, nr_written);
1278 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1279 ret = run_delalloc_nocow(inode, locked_page, start, end,
1280 page_started, 0, nr_written);
1281 else if (!btrfs_test_opt(root, COMPRESS) &&
1282 !(BTRFS_I(inode)->force_compress) &&
1283 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1284 ret = cow_file_range(inode, locked_page, start, end,
1285 page_started, nr_written, 1);
1287 ret = cow_file_range_async(inode, locked_page, start, end,
1288 page_started, nr_written);
1292 static void btrfs_split_extent_hook(struct inode *inode,
1293 struct extent_state *orig, u64 split)
1295 /* not delalloc, ignore it */
1296 if (!(orig->state & EXTENT_DELALLOC))
1299 spin_lock(&BTRFS_I(inode)->lock);
1300 BTRFS_I(inode)->outstanding_extents++;
1301 spin_unlock(&BTRFS_I(inode)->lock);
1305 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306 * extents so we can keep track of new extents that are just merged onto old
1307 * extents, such as when we are doing sequential writes, so we can properly
1308 * account for the metadata space we'll need.
1310 static void btrfs_merge_extent_hook(struct inode *inode,
1311 struct extent_state *new,
1312 struct extent_state *other)
1314 /* not delalloc, ignore it */
1315 if (!(other->state & EXTENT_DELALLOC))
1318 spin_lock(&BTRFS_I(inode)->lock);
1319 BTRFS_I(inode)->outstanding_extents--;
1320 spin_unlock(&BTRFS_I(inode)->lock);
1324 * extent_io.c set_bit_hook, used to track delayed allocation
1325 * bytes in this file, and to maintain the list of inodes that
1326 * have pending delalloc work to be done.
1328 static void btrfs_set_bit_hook(struct inode *inode,
1329 struct extent_state *state, int *bits)
1333 * set_bit and clear bit hooks normally require _irqsave/restore
1334 * but in this case, we are only testing for the DELALLOC
1335 * bit, which is only set or cleared with irqs on
1337 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338 struct btrfs_root *root = BTRFS_I(inode)->root;
1339 u64 len = state->end + 1 - state->start;
1340 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342 if (*bits & EXTENT_FIRST_DELALLOC) {
1343 *bits &= ~EXTENT_FIRST_DELALLOC;
1345 spin_lock(&BTRFS_I(inode)->lock);
1346 BTRFS_I(inode)->outstanding_extents++;
1347 spin_unlock(&BTRFS_I(inode)->lock);
1350 spin_lock(&root->fs_info->delalloc_lock);
1351 BTRFS_I(inode)->delalloc_bytes += len;
1352 root->fs_info->delalloc_bytes += len;
1353 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1354 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355 &root->fs_info->delalloc_inodes);
1357 spin_unlock(&root->fs_info->delalloc_lock);
1362 * extent_io.c clear_bit_hook, see set_bit_hook for why
1364 static void btrfs_clear_bit_hook(struct inode *inode,
1365 struct extent_state *state, int *bits)
1368 * set_bit and clear bit hooks normally require _irqsave/restore
1369 * but in this case, we are only testing for the DELALLOC
1370 * bit, which is only set or cleared with irqs on
1372 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1373 struct btrfs_root *root = BTRFS_I(inode)->root;
1374 u64 len = state->end + 1 - state->start;
1375 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377 if (*bits & EXTENT_FIRST_DELALLOC) {
1378 *bits &= ~EXTENT_FIRST_DELALLOC;
1379 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380 spin_lock(&BTRFS_I(inode)->lock);
1381 BTRFS_I(inode)->outstanding_extents--;
1382 spin_unlock(&BTRFS_I(inode)->lock);
1385 if (*bits & EXTENT_DO_ACCOUNTING)
1386 btrfs_delalloc_release_metadata(inode, len);
1388 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390 btrfs_free_reserved_data_space(inode, len);
1392 spin_lock(&root->fs_info->delalloc_lock);
1393 root->fs_info->delalloc_bytes -= len;
1394 BTRFS_I(inode)->delalloc_bytes -= len;
1396 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1397 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400 spin_unlock(&root->fs_info->delalloc_lock);
1405 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406 * we don't create bios that span stripes or chunks
1408 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1409 size_t size, struct bio *bio,
1410 unsigned long bio_flags)
1412 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413 struct btrfs_mapping_tree *map_tree;
1414 u64 logical = (u64)bio->bi_sector << 9;
1419 if (bio_flags & EXTENT_BIO_COMPRESSED)
1422 length = bio->bi_size;
1423 map_tree = &root->fs_info->mapping_tree;
1424 map_length = length;
1425 ret = btrfs_map_block(map_tree, READ, logical,
1426 &map_length, NULL, 0);
1427 /* Will always return 0 or 1 with map_multi == NULL */
1429 if (map_length < length + size)
1435 * in order to insert checksums into the metadata in large chunks,
1436 * we wait until bio submission time. All the pages in the bio are
1437 * checksummed and sums are attached onto the ordered extent record.
1439 * At IO completion time the cums attached on the ordered extent record
1440 * are inserted into the btree
1442 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443 struct bio *bio, int mirror_num,
1444 unsigned long bio_flags,
1447 struct btrfs_root *root = BTRFS_I(inode)->root;
1450 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1456 * in order to insert checksums into the metadata in large chunks,
1457 * we wait until bio submission time. All the pages in the bio are
1458 * checksummed and sums are attached onto the ordered extent record.
1460 * At IO completion time the cums attached on the ordered extent record
1461 * are inserted into the btree
1463 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1464 int mirror_num, unsigned long bio_flags,
1467 struct btrfs_root *root = BTRFS_I(inode)->root;
1468 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1472 * extent_io.c submission hook. This does the right thing for csum calculation
1473 * on write, or reading the csums from the tree before a read
1475 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1476 int mirror_num, unsigned long bio_flags,
1479 struct btrfs_root *root = BTRFS_I(inode)->root;
1484 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1486 if (btrfs_is_free_space_inode(root, inode))
1489 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1493 if (!(rw & REQ_WRITE)) {
1494 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1495 return btrfs_submit_compressed_read(inode, bio,
1496 mirror_num, bio_flags);
1497 } else if (!skip_sum) {
1498 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1503 } else if (!skip_sum) {
1504 /* csum items have already been cloned */
1505 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1507 /* we're doing a write, do the async checksumming */
1508 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1509 inode, rw, bio, mirror_num,
1510 bio_flags, bio_offset,
1511 __btrfs_submit_bio_start,
1512 __btrfs_submit_bio_done);
1516 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1520 * given a list of ordered sums record them in the inode. This happens
1521 * at IO completion time based on sums calculated at bio submission time.
1523 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1524 struct inode *inode, u64 file_offset,
1525 struct list_head *list)
1527 struct btrfs_ordered_sum *sum;
1529 list_for_each_entry(sum, list, list) {
1530 btrfs_csum_file_blocks(trans,
1531 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1536 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1537 struct extent_state **cached_state)
1539 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1541 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1542 cached_state, GFP_NOFS);
1545 /* see btrfs_writepage_start_hook for details on why this is required */
1546 struct btrfs_writepage_fixup {
1548 struct btrfs_work work;
1551 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1553 struct btrfs_writepage_fixup *fixup;
1554 struct btrfs_ordered_extent *ordered;
1555 struct extent_state *cached_state = NULL;
1557 struct inode *inode;
1562 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1566 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1567 ClearPageChecked(page);
1571 inode = page->mapping->host;
1572 page_start = page_offset(page);
1573 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1575 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1578 /* already ordered? We're done */
1579 if (PagePrivate2(page))
1582 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1584 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1585 page_end, &cached_state, GFP_NOFS);
1587 btrfs_start_ordered_extent(inode, ordered, 1);
1588 btrfs_put_ordered_extent(ordered);
1592 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1594 mapping_set_error(page->mapping, ret);
1595 end_extent_writepage(page, ret, page_start, page_end);
1596 ClearPageChecked(page);
1600 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1601 ClearPageChecked(page);
1602 set_page_dirty(page);
1604 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1605 &cached_state, GFP_NOFS);
1608 page_cache_release(page);
1613 * There are a few paths in the higher layers of the kernel that directly
1614 * set the page dirty bit without asking the filesystem if it is a
1615 * good idea. This causes problems because we want to make sure COW
1616 * properly happens and the data=ordered rules are followed.
1618 * In our case any range that doesn't have the ORDERED bit set
1619 * hasn't been properly setup for IO. We kick off an async process
1620 * to fix it up. The async helper will wait for ordered extents, set
1621 * the delalloc bit and make it safe to write the page.
1623 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1625 struct inode *inode = page->mapping->host;
1626 struct btrfs_writepage_fixup *fixup;
1627 struct btrfs_root *root = BTRFS_I(inode)->root;
1629 /* this page is properly in the ordered list */
1630 if (TestClearPagePrivate2(page))
1633 if (PageChecked(page))
1636 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1640 SetPageChecked(page);
1641 page_cache_get(page);
1642 fixup->work.func = btrfs_writepage_fixup_worker;
1644 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1648 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1649 struct inode *inode, u64 file_pos,
1650 u64 disk_bytenr, u64 disk_num_bytes,
1651 u64 num_bytes, u64 ram_bytes,
1652 u8 compression, u8 encryption,
1653 u16 other_encoding, int extent_type)
1655 struct btrfs_root *root = BTRFS_I(inode)->root;
1656 struct btrfs_file_extent_item *fi;
1657 struct btrfs_path *path;
1658 struct extent_buffer *leaf;
1659 struct btrfs_key ins;
1663 path = btrfs_alloc_path();
1667 path->leave_spinning = 1;
1670 * we may be replacing one extent in the tree with another.
1671 * The new extent is pinned in the extent map, and we don't want
1672 * to drop it from the cache until it is completely in the btree.
1674 * So, tell btrfs_drop_extents to leave this extent in the cache.
1675 * the caller is expected to unpin it and allow it to be merged
1678 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1682 ins.objectid = btrfs_ino(inode);
1683 ins.offset = file_pos;
1684 ins.type = BTRFS_EXTENT_DATA_KEY;
1685 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1687 leaf = path->nodes[0];
1688 fi = btrfs_item_ptr(leaf, path->slots[0],
1689 struct btrfs_file_extent_item);
1690 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1691 btrfs_set_file_extent_type(leaf, fi, extent_type);
1692 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1693 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1694 btrfs_set_file_extent_offset(leaf, fi, 0);
1695 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1696 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1697 btrfs_set_file_extent_compression(leaf, fi, compression);
1698 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1699 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1701 btrfs_unlock_up_safe(path, 1);
1702 btrfs_set_lock_blocking(leaf);
1704 btrfs_mark_buffer_dirty(leaf);
1706 inode_add_bytes(inode, num_bytes);
1708 ins.objectid = disk_bytenr;
1709 ins.offset = disk_num_bytes;
1710 ins.type = BTRFS_EXTENT_ITEM_KEY;
1711 ret = btrfs_alloc_reserved_file_extent(trans, root,
1712 root->root_key.objectid,
1713 btrfs_ino(inode), file_pos, &ins);
1715 btrfs_free_path(path);
1721 * helper function for btrfs_finish_ordered_io, this
1722 * just reads in some of the csum leaves to prime them into ram
1723 * before we start the transaction. It limits the amount of btree
1724 * reads required while inside the transaction.
1726 /* as ordered data IO finishes, this gets called so we can finish
1727 * an ordered extent if the range of bytes in the file it covers are
1730 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1732 struct btrfs_root *root = BTRFS_I(inode)->root;
1733 struct btrfs_trans_handle *trans = NULL;
1734 struct btrfs_ordered_extent *ordered_extent = NULL;
1735 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1736 struct extent_state *cached_state = NULL;
1737 int compress_type = 0;
1741 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1745 BUG_ON(!ordered_extent);
1747 nolock = btrfs_is_free_space_inode(root, inode);
1749 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1750 BUG_ON(!list_empty(&ordered_extent->list));
1751 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1754 trans = btrfs_join_transaction_nolock(root);
1756 trans = btrfs_join_transaction(root);
1757 BUG_ON(IS_ERR(trans));
1758 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1759 ret = btrfs_update_inode_fallback(trans, root, inode);
1765 lock_extent_bits(io_tree, ordered_extent->file_offset,
1766 ordered_extent->file_offset + ordered_extent->len - 1,
1770 trans = btrfs_join_transaction_nolock(root);
1772 trans = btrfs_join_transaction(root);
1773 BUG_ON(IS_ERR(trans));
1774 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1776 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1777 compress_type = ordered_extent->compress_type;
1778 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1779 BUG_ON(compress_type);
1780 ret = btrfs_mark_extent_written(trans, inode,
1781 ordered_extent->file_offset,
1782 ordered_extent->file_offset +
1783 ordered_extent->len);
1786 BUG_ON(root == root->fs_info->tree_root);
1787 ret = insert_reserved_file_extent(trans, inode,
1788 ordered_extent->file_offset,
1789 ordered_extent->start,
1790 ordered_extent->disk_len,
1791 ordered_extent->len,
1792 ordered_extent->len,
1793 compress_type, 0, 0,
1794 BTRFS_FILE_EXTENT_REG);
1795 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1796 ordered_extent->file_offset,
1797 ordered_extent->len);
1800 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1801 ordered_extent->file_offset +
1802 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1804 add_pending_csums(trans, inode, ordered_extent->file_offset,
1805 &ordered_extent->list);
1807 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1808 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1809 ret = btrfs_update_inode_fallback(trans, root, inode);
1814 if (root != root->fs_info->tree_root)
1815 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1818 btrfs_end_transaction_nolock(trans, root);
1820 btrfs_end_transaction(trans, root);
1824 btrfs_put_ordered_extent(ordered_extent);
1825 /* once for the tree */
1826 btrfs_put_ordered_extent(ordered_extent);
1831 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1832 struct extent_state *state, int uptodate)
1834 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1836 ClearPagePrivate2(page);
1837 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1841 * when reads are done, we need to check csums to verify the data is correct
1842 * if there's a match, we allow the bio to finish. If not, the code in
1843 * extent_io.c will try to find good copies for us.
1845 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1846 struct extent_state *state)
1848 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1849 struct inode *inode = page->mapping->host;
1850 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1852 u64 private = ~(u32)0;
1854 struct btrfs_root *root = BTRFS_I(inode)->root;
1857 if (PageChecked(page)) {
1858 ClearPageChecked(page);
1862 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1865 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1866 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1867 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1872 if (state && state->start == start) {
1873 private = state->private;
1876 ret = get_state_private(io_tree, start, &private);
1878 kaddr = kmap_atomic(page, KM_USER0);
1882 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1883 btrfs_csum_final(csum, (char *)&csum);
1884 if (csum != private)
1887 kunmap_atomic(kaddr, KM_USER0);
1892 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1894 (unsigned long long)btrfs_ino(page->mapping->host),
1895 (unsigned long long)start, csum,
1896 (unsigned long long)private);
1897 memset(kaddr + offset, 1, end - start + 1);
1898 flush_dcache_page(page);
1899 kunmap_atomic(kaddr, KM_USER0);
1905 struct delayed_iput {
1906 struct list_head list;
1907 struct inode *inode;
1910 void btrfs_add_delayed_iput(struct inode *inode)
1912 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1913 struct delayed_iput *delayed;
1915 if (atomic_add_unless(&inode->i_count, -1, 1))
1918 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1919 delayed->inode = inode;
1921 spin_lock(&fs_info->delayed_iput_lock);
1922 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1923 spin_unlock(&fs_info->delayed_iput_lock);
1926 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1929 struct btrfs_fs_info *fs_info = root->fs_info;
1930 struct delayed_iput *delayed;
1933 spin_lock(&fs_info->delayed_iput_lock);
1934 empty = list_empty(&fs_info->delayed_iputs);
1935 spin_unlock(&fs_info->delayed_iput_lock);
1939 down_read(&root->fs_info->cleanup_work_sem);
1940 spin_lock(&fs_info->delayed_iput_lock);
1941 list_splice_init(&fs_info->delayed_iputs, &list);
1942 spin_unlock(&fs_info->delayed_iput_lock);
1944 while (!list_empty(&list)) {
1945 delayed = list_entry(list.next, struct delayed_iput, list);
1946 list_del(&delayed->list);
1947 iput(delayed->inode);
1950 up_read(&root->fs_info->cleanup_work_sem);
1953 enum btrfs_orphan_cleanup_state {
1954 ORPHAN_CLEANUP_STARTED = 1,
1955 ORPHAN_CLEANUP_DONE = 2,
1959 * This is called in transaction commit time. If there are no orphan
1960 * files in the subvolume, it removes orphan item and frees block_rsv
1963 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1964 struct btrfs_root *root)
1966 struct btrfs_block_rsv *block_rsv;
1969 if (!list_empty(&root->orphan_list) ||
1970 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1973 spin_lock(&root->orphan_lock);
1974 if (!list_empty(&root->orphan_list)) {
1975 spin_unlock(&root->orphan_lock);
1979 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1980 spin_unlock(&root->orphan_lock);
1984 block_rsv = root->orphan_block_rsv;
1985 root->orphan_block_rsv = NULL;
1986 spin_unlock(&root->orphan_lock);
1988 if (root->orphan_item_inserted &&
1989 btrfs_root_refs(&root->root_item) > 0) {
1990 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1991 root->root_key.objectid);
1993 root->orphan_item_inserted = 0;
1997 WARN_ON(block_rsv->size > 0);
1998 btrfs_free_block_rsv(root, block_rsv);
2003 * This creates an orphan entry for the given inode in case something goes
2004 * wrong in the middle of an unlink/truncate.
2006 * NOTE: caller of this function should reserve 5 units of metadata for
2009 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2011 struct btrfs_root *root = BTRFS_I(inode)->root;
2012 struct btrfs_block_rsv *block_rsv = NULL;
2017 if (!root->orphan_block_rsv) {
2018 block_rsv = btrfs_alloc_block_rsv(root);
2023 spin_lock(&root->orphan_lock);
2024 if (!root->orphan_block_rsv) {
2025 root->orphan_block_rsv = block_rsv;
2026 } else if (block_rsv) {
2027 btrfs_free_block_rsv(root, block_rsv);
2031 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2032 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2035 * For proper ENOSPC handling, we should do orphan
2036 * cleanup when mounting. But this introduces backward
2037 * compatibility issue.
2039 if (!xchg(&root->orphan_item_inserted, 1))
2047 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2048 BTRFS_I(inode)->orphan_meta_reserved = 1;
2051 spin_unlock(&root->orphan_lock);
2053 /* grab metadata reservation from transaction handle */
2055 ret = btrfs_orphan_reserve_metadata(trans, inode);
2059 /* insert an orphan item to track this unlinked/truncated file */
2061 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2062 BUG_ON(ret && ret != -EEXIST);
2065 /* insert an orphan item to track subvolume contains orphan files */
2067 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2068 root->root_key.objectid);
2075 * We have done the truncate/delete so we can go ahead and remove the orphan
2076 * item for this particular inode.
2078 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
2081 int delete_item = 0;
2082 int release_rsv = 0;
2085 spin_lock(&root->orphan_lock);
2086 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2087 list_del_init(&BTRFS_I(inode)->i_orphan);
2091 if (BTRFS_I(inode)->orphan_meta_reserved) {
2092 BTRFS_I(inode)->orphan_meta_reserved = 0;
2095 spin_unlock(&root->orphan_lock);
2097 if (trans && delete_item) {
2098 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2103 btrfs_orphan_release_metadata(inode);
2109 * this cleans up any orphans that may be left on the list from the last use
2112 int btrfs_orphan_cleanup(struct btrfs_root *root)
2114 struct btrfs_path *path;
2115 struct extent_buffer *leaf;
2116 struct btrfs_key key, found_key;
2117 struct btrfs_trans_handle *trans;
2118 struct inode *inode;
2119 u64 last_objectid = 0;
2120 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2122 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2125 path = btrfs_alloc_path();
2132 key.objectid = BTRFS_ORPHAN_OBJECTID;
2133 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2134 key.offset = (u64)-1;
2137 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2142 * if ret == 0 means we found what we were searching for, which
2143 * is weird, but possible, so only screw with path if we didn't
2144 * find the key and see if we have stuff that matches
2148 if (path->slots[0] == 0)
2153 /* pull out the item */
2154 leaf = path->nodes[0];
2155 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2157 /* make sure the item matches what we want */
2158 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2160 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2163 /* release the path since we're done with it */
2164 btrfs_release_path(path);
2167 * this is where we are basically btrfs_lookup, without the
2168 * crossing root thing. we store the inode number in the
2169 * offset of the orphan item.
2172 if (found_key.offset == last_objectid) {
2173 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2174 "stopping orphan cleanup\n");
2179 last_objectid = found_key.offset;
2181 found_key.objectid = found_key.offset;
2182 found_key.type = BTRFS_INODE_ITEM_KEY;
2183 found_key.offset = 0;
2184 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2185 ret = PTR_RET(inode);
2186 if (ret && ret != -ESTALE)
2189 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2190 struct btrfs_root *dead_root;
2191 struct btrfs_fs_info *fs_info = root->fs_info;
2192 int is_dead_root = 0;
2195 * this is an orphan in the tree root. Currently these
2196 * could come from 2 sources:
2197 * a) a snapshot deletion in progress
2198 * b) a free space cache inode
2199 * We need to distinguish those two, as the snapshot
2200 * orphan must not get deleted.
2201 * find_dead_roots already ran before us, so if this
2202 * is a snapshot deletion, we should find the root
2203 * in the dead_roots list
2205 spin_lock(&fs_info->trans_lock);
2206 list_for_each_entry(dead_root, &fs_info->dead_roots,
2208 if (dead_root->root_key.objectid ==
2209 found_key.objectid) {
2214 spin_unlock(&fs_info->trans_lock);
2216 /* prevent this orphan from being found again */
2217 key.offset = found_key.objectid - 1;
2222 * Inode is already gone but the orphan item is still there,
2223 * kill the orphan item.
2225 if (ret == -ESTALE) {
2226 trans = btrfs_start_transaction(root, 1);
2227 if (IS_ERR(trans)) {
2228 ret = PTR_ERR(trans);
2231 ret = btrfs_del_orphan_item(trans, root,
2232 found_key.objectid);
2234 btrfs_end_transaction(trans, root);
2239 * add this inode to the orphan list so btrfs_orphan_del does
2240 * the proper thing when we hit it
2242 spin_lock(&root->orphan_lock);
2243 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2244 spin_unlock(&root->orphan_lock);
2246 /* if we have links, this was a truncate, lets do that */
2247 if (inode->i_nlink) {
2248 if (!S_ISREG(inode->i_mode)) {
2254 ret = btrfs_truncate(inode);
2259 /* this will do delete_inode and everything for us */
2264 /* release the path since we're done with it */
2265 btrfs_release_path(path);
2267 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2269 if (root->orphan_block_rsv)
2270 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2273 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2274 trans = btrfs_join_transaction(root);
2276 btrfs_end_transaction(trans, root);
2280 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2282 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2286 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2287 btrfs_free_path(path);
2292 * very simple check to peek ahead in the leaf looking for xattrs. If we
2293 * don't find any xattrs, we know there can't be any acls.
2295 * slot is the slot the inode is in, objectid is the objectid of the inode
2297 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2298 int slot, u64 objectid)
2300 u32 nritems = btrfs_header_nritems(leaf);
2301 struct btrfs_key found_key;
2305 while (slot < nritems) {
2306 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2308 /* we found a different objectid, there must not be acls */
2309 if (found_key.objectid != objectid)
2312 /* we found an xattr, assume we've got an acl */
2313 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2317 * we found a key greater than an xattr key, there can't
2318 * be any acls later on
2320 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2327 * it goes inode, inode backrefs, xattrs, extents,
2328 * so if there are a ton of hard links to an inode there can
2329 * be a lot of backrefs. Don't waste time searching too hard,
2330 * this is just an optimization
2335 /* we hit the end of the leaf before we found an xattr or
2336 * something larger than an xattr. We have to assume the inode
2343 * read an inode from the btree into the in-memory inode
2345 static void btrfs_read_locked_inode(struct inode *inode)
2347 struct btrfs_path *path;
2348 struct extent_buffer *leaf;
2349 struct btrfs_inode_item *inode_item;
2350 struct btrfs_timespec *tspec;
2351 struct btrfs_root *root = BTRFS_I(inode)->root;
2352 struct btrfs_key location;
2356 bool filled = false;
2358 ret = btrfs_fill_inode(inode, &rdev);
2362 path = btrfs_alloc_path();
2366 path->leave_spinning = 1;
2367 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2369 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2373 leaf = path->nodes[0];
2378 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2379 struct btrfs_inode_item);
2380 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2381 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2382 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2383 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2384 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2386 tspec = btrfs_inode_atime(inode_item);
2387 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2388 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2390 tspec = btrfs_inode_mtime(inode_item);
2391 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2392 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2394 tspec = btrfs_inode_ctime(inode_item);
2395 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2396 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2398 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2399 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2400 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2401 inode->i_generation = BTRFS_I(inode)->generation;
2403 rdev = btrfs_inode_rdev(leaf, inode_item);
2405 BTRFS_I(inode)->index_cnt = (u64)-1;
2406 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2409 * try to precache a NULL acl entry for files that don't have
2410 * any xattrs or acls
2412 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2415 cache_no_acl(inode);
2417 btrfs_free_path(path);
2419 switch (inode->i_mode & S_IFMT) {
2421 inode->i_mapping->a_ops = &btrfs_aops;
2422 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2423 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2424 inode->i_fop = &btrfs_file_operations;
2425 inode->i_op = &btrfs_file_inode_operations;
2428 inode->i_fop = &btrfs_dir_file_operations;
2429 if (root == root->fs_info->tree_root)
2430 inode->i_op = &btrfs_dir_ro_inode_operations;
2432 inode->i_op = &btrfs_dir_inode_operations;
2435 inode->i_op = &btrfs_symlink_inode_operations;
2436 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2437 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2440 inode->i_op = &btrfs_special_inode_operations;
2441 init_special_inode(inode, inode->i_mode, rdev);
2445 btrfs_update_iflags(inode);
2449 btrfs_free_path(path);
2450 make_bad_inode(inode);
2454 * given a leaf and an inode, copy the inode fields into the leaf
2456 static void fill_inode_item(struct btrfs_trans_handle *trans,
2457 struct extent_buffer *leaf,
2458 struct btrfs_inode_item *item,
2459 struct inode *inode)
2461 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2462 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2463 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2464 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2465 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2467 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2468 inode->i_atime.tv_sec);
2469 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2470 inode->i_atime.tv_nsec);
2472 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2473 inode->i_mtime.tv_sec);
2474 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2475 inode->i_mtime.tv_nsec);
2477 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2478 inode->i_ctime.tv_sec);
2479 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2480 inode->i_ctime.tv_nsec);
2482 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2483 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2484 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2485 btrfs_set_inode_transid(leaf, item, trans->transid);
2486 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2487 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2488 btrfs_set_inode_block_group(leaf, item, 0);
2492 * copy everything in the in-memory inode into the btree.
2494 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2495 struct btrfs_root *root, struct inode *inode)
2497 struct btrfs_inode_item *inode_item;
2498 struct btrfs_path *path;
2499 struct extent_buffer *leaf;
2502 path = btrfs_alloc_path();
2506 path->leave_spinning = 1;
2507 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2515 btrfs_unlock_up_safe(path, 1);
2516 leaf = path->nodes[0];
2517 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2518 struct btrfs_inode_item);
2520 fill_inode_item(trans, leaf, inode_item, inode);
2521 btrfs_mark_buffer_dirty(leaf);
2522 btrfs_set_inode_last_trans(trans, inode);
2525 btrfs_free_path(path);
2530 * copy everything in the in-memory inode into the btree.
2532 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2533 struct btrfs_root *root, struct inode *inode)
2538 * If the inode is a free space inode, we can deadlock during commit
2539 * if we put it into the delayed code.
2541 * The data relocation inode should also be directly updated
2544 if (!btrfs_is_free_space_inode(root, inode)
2545 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2546 ret = btrfs_delayed_update_inode(trans, root, inode);
2548 btrfs_set_inode_last_trans(trans, inode);
2552 return btrfs_update_inode_item(trans, root, inode);
2555 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2556 struct btrfs_root *root, struct inode *inode)
2560 ret = btrfs_update_inode(trans, root, inode);
2562 return btrfs_update_inode_item(trans, root, inode);
2567 * unlink helper that gets used here in inode.c and in the tree logging
2568 * recovery code. It remove a link in a directory with a given name, and
2569 * also drops the back refs in the inode to the directory
2571 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2572 struct btrfs_root *root,
2573 struct inode *dir, struct inode *inode,
2574 const char *name, int name_len)
2576 struct btrfs_path *path;
2578 struct extent_buffer *leaf;
2579 struct btrfs_dir_item *di;
2580 struct btrfs_key key;
2582 u64 ino = btrfs_ino(inode);
2583 u64 dir_ino = btrfs_ino(dir);
2585 path = btrfs_alloc_path();
2591 path->leave_spinning = 1;
2592 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2593 name, name_len, -1);
2602 leaf = path->nodes[0];
2603 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2604 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2607 btrfs_release_path(path);
2609 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2612 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2613 "inode %llu parent %llu\n", name_len, name,
2614 (unsigned long long)ino, (unsigned long long)dir_ino);
2618 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2622 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2624 BUG_ON(ret != 0 && ret != -ENOENT);
2626 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2631 btrfs_free_path(path);
2635 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2636 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2637 btrfs_update_inode(trans, root, dir);
2642 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2643 struct btrfs_root *root,
2644 struct inode *dir, struct inode *inode,
2645 const char *name, int name_len)
2648 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2650 btrfs_drop_nlink(inode);
2651 ret = btrfs_update_inode(trans, root, inode);
2657 /* helper to check if there is any shared block in the path */
2658 static int check_path_shared(struct btrfs_root *root,
2659 struct btrfs_path *path)
2661 struct extent_buffer *eb;
2665 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2668 if (!path->nodes[level])
2670 eb = path->nodes[level];
2671 if (!btrfs_block_can_be_shared(root, eb))
2673 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2682 * helper to start transaction for unlink and rmdir.
2684 * unlink and rmdir are special in btrfs, they do not always free space.
2685 * so in enospc case, we should make sure they will free space before
2686 * allowing them to use the global metadata reservation.
2688 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2689 struct dentry *dentry)
2691 struct btrfs_trans_handle *trans;
2692 struct btrfs_root *root = BTRFS_I(dir)->root;
2693 struct btrfs_path *path;
2694 struct btrfs_inode_ref *ref;
2695 struct btrfs_dir_item *di;
2696 struct inode *inode = dentry->d_inode;
2701 u64 ino = btrfs_ino(inode);
2702 u64 dir_ino = btrfs_ino(dir);
2705 * 1 for the possible orphan item
2706 * 1 for the dir item
2707 * 1 for the dir index
2708 * 1 for the inode ref
2709 * 1 for the inode ref in the tree log
2710 * 2 for the dir entries in the log
2713 trans = btrfs_start_transaction(root, 8);
2714 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2717 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2718 return ERR_PTR(-ENOSPC);
2720 /* check if there is someone else holds reference */
2721 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2722 return ERR_PTR(-ENOSPC);
2724 if (atomic_read(&inode->i_count) > 2)
2725 return ERR_PTR(-ENOSPC);
2727 if (xchg(&root->fs_info->enospc_unlink, 1))
2728 return ERR_PTR(-ENOSPC);
2730 path = btrfs_alloc_path();
2732 root->fs_info->enospc_unlink = 0;
2733 return ERR_PTR(-ENOMEM);
2736 /* 1 for the orphan item */
2737 trans = btrfs_start_transaction(root, 1);
2738 if (IS_ERR(trans)) {
2739 btrfs_free_path(path);
2740 root->fs_info->enospc_unlink = 0;
2744 path->skip_locking = 1;
2745 path->search_commit_root = 1;
2747 ret = btrfs_lookup_inode(trans, root, path,
2748 &BTRFS_I(dir)->location, 0);
2754 if (check_path_shared(root, path))
2759 btrfs_release_path(path);
2761 ret = btrfs_lookup_inode(trans, root, path,
2762 &BTRFS_I(inode)->location, 0);
2768 if (check_path_shared(root, path))
2773 btrfs_release_path(path);
2775 if (ret == 0 && S_ISREG(inode->i_mode)) {
2776 ret = btrfs_lookup_file_extent(trans, root, path,
2783 if (check_path_shared(root, path))
2785 btrfs_release_path(path);
2793 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2794 dentry->d_name.name, dentry->d_name.len, 0);
2800 if (check_path_shared(root, path))
2806 btrfs_release_path(path);
2808 ref = btrfs_lookup_inode_ref(trans, root, path,
2809 dentry->d_name.name, dentry->d_name.len,
2816 if (check_path_shared(root, path))
2818 index = btrfs_inode_ref_index(path->nodes[0], ref);
2819 btrfs_release_path(path);
2822 * This is a commit root search, if we can lookup inode item and other
2823 * relative items in the commit root, it means the transaction of
2824 * dir/file creation has been committed, and the dir index item that we
2825 * delay to insert has also been inserted into the commit root. So
2826 * we needn't worry about the delayed insertion of the dir index item
2829 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2830 dentry->d_name.name, dentry->d_name.len, 0);
2835 BUG_ON(ret == -ENOENT);
2836 if (check_path_shared(root, path))
2841 btrfs_free_path(path);
2842 /* Migrate the orphan reservation over */
2844 err = btrfs_block_rsv_migrate(trans->block_rsv,
2845 &root->fs_info->global_block_rsv,
2846 trans->bytes_reserved);
2849 btrfs_end_transaction(trans, root);
2850 root->fs_info->enospc_unlink = 0;
2851 return ERR_PTR(err);
2854 trans->block_rsv = &root->fs_info->global_block_rsv;
2858 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2859 struct btrfs_root *root)
2861 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2862 btrfs_block_rsv_release(root, trans->block_rsv,
2863 trans->bytes_reserved);
2864 trans->block_rsv = &root->fs_info->trans_block_rsv;
2865 BUG_ON(!root->fs_info->enospc_unlink);
2866 root->fs_info->enospc_unlink = 0;
2868 btrfs_end_transaction(trans, root);
2871 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2873 struct btrfs_root *root = BTRFS_I(dir)->root;
2874 struct btrfs_trans_handle *trans;
2875 struct inode *inode = dentry->d_inode;
2877 unsigned long nr = 0;
2879 trans = __unlink_start_trans(dir, dentry);
2881 return PTR_ERR(trans);
2883 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2885 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2886 dentry->d_name.name, dentry->d_name.len);
2890 if (inode->i_nlink == 0) {
2891 ret = btrfs_orphan_add(trans, inode);
2897 nr = trans->blocks_used;
2898 __unlink_end_trans(trans, root);
2899 btrfs_btree_balance_dirty(root, nr);
2903 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2904 struct btrfs_root *root,
2905 struct inode *dir, u64 objectid,
2906 const char *name, int name_len)
2908 struct btrfs_path *path;
2909 struct extent_buffer *leaf;
2910 struct btrfs_dir_item *di;
2911 struct btrfs_key key;
2914 u64 dir_ino = btrfs_ino(dir);
2916 path = btrfs_alloc_path();
2920 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2921 name, name_len, -1);
2922 BUG_ON(IS_ERR_OR_NULL(di));
2924 leaf = path->nodes[0];
2925 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2926 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2927 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2929 btrfs_release_path(path);
2931 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2932 objectid, root->root_key.objectid,
2933 dir_ino, &index, name, name_len);
2935 BUG_ON(ret != -ENOENT);
2936 di = btrfs_search_dir_index_item(root, path, dir_ino,
2938 BUG_ON(IS_ERR_OR_NULL(di));
2940 leaf = path->nodes[0];
2941 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2942 btrfs_release_path(path);
2945 btrfs_release_path(path);
2947 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2950 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2951 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2952 ret = btrfs_update_inode(trans, root, dir);
2955 btrfs_free_path(path);
2959 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2961 struct inode *inode = dentry->d_inode;
2963 struct btrfs_root *root = BTRFS_I(dir)->root;
2964 struct btrfs_trans_handle *trans;
2965 unsigned long nr = 0;
2967 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2968 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2971 trans = __unlink_start_trans(dir, dentry);
2973 return PTR_ERR(trans);
2975 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2976 err = btrfs_unlink_subvol(trans, root, dir,
2977 BTRFS_I(inode)->location.objectid,
2978 dentry->d_name.name,
2979 dentry->d_name.len);
2983 err = btrfs_orphan_add(trans, inode);
2987 /* now the directory is empty */
2988 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2989 dentry->d_name.name, dentry->d_name.len);
2991 btrfs_i_size_write(inode, 0);
2993 nr = trans->blocks_used;
2994 __unlink_end_trans(trans, root);
2995 btrfs_btree_balance_dirty(root, nr);
3001 * this can truncate away extent items, csum items and directory items.
3002 * It starts at a high offset and removes keys until it can't find
3003 * any higher than new_size
3005 * csum items that cross the new i_size are truncated to the new size
3008 * min_type is the minimum key type to truncate down to. If set to 0, this
3009 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3011 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3012 struct btrfs_root *root,
3013 struct inode *inode,
3014 u64 new_size, u32 min_type)
3016 struct btrfs_path *path;
3017 struct extent_buffer *leaf;
3018 struct btrfs_file_extent_item *fi;
3019 struct btrfs_key key;
3020 struct btrfs_key found_key;
3021 u64 extent_start = 0;
3022 u64 extent_num_bytes = 0;
3023 u64 extent_offset = 0;
3025 u64 mask = root->sectorsize - 1;
3026 u32 found_type = (u8)-1;
3029 int pending_del_nr = 0;
3030 int pending_del_slot = 0;
3031 int extent_type = -1;
3034 u64 ino = btrfs_ino(inode);
3036 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3038 path = btrfs_alloc_path();
3043 if (root->ref_cows || root == root->fs_info->tree_root)
3044 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3047 * This function is also used to drop the items in the log tree before
3048 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3049 * it is used to drop the loged items. So we shouldn't kill the delayed
3052 if (min_type == 0 && root == BTRFS_I(inode)->root)
3053 btrfs_kill_delayed_inode_items(inode);
3056 key.offset = (u64)-1;
3060 path->leave_spinning = 1;
3061 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3068 /* there are no items in the tree for us to truncate, we're
3071 if (path->slots[0] == 0)
3078 leaf = path->nodes[0];
3079 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3080 found_type = btrfs_key_type(&found_key);
3082 if (found_key.objectid != ino)
3085 if (found_type < min_type)
3088 item_end = found_key.offset;
3089 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3090 fi = btrfs_item_ptr(leaf, path->slots[0],
3091 struct btrfs_file_extent_item);
3092 extent_type = btrfs_file_extent_type(leaf, fi);
3093 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3095 btrfs_file_extent_num_bytes(leaf, fi);
3096 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3097 item_end += btrfs_file_extent_inline_len(leaf,
3102 if (found_type > min_type) {
3105 if (item_end < new_size)
3107 if (found_key.offset >= new_size)
3113 /* FIXME, shrink the extent if the ref count is only 1 */
3114 if (found_type != BTRFS_EXTENT_DATA_KEY)
3117 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3119 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3121 u64 orig_num_bytes =
3122 btrfs_file_extent_num_bytes(leaf, fi);
3123 extent_num_bytes = new_size -
3124 found_key.offset + root->sectorsize - 1;
3125 extent_num_bytes = extent_num_bytes &
3126 ~((u64)root->sectorsize - 1);
3127 btrfs_set_file_extent_num_bytes(leaf, fi,
3129 num_dec = (orig_num_bytes -
3131 if (root->ref_cows && extent_start != 0)
3132 inode_sub_bytes(inode, num_dec);
3133 btrfs_mark_buffer_dirty(leaf);
3136 btrfs_file_extent_disk_num_bytes(leaf,
3138 extent_offset = found_key.offset -
3139 btrfs_file_extent_offset(leaf, fi);
3141 /* FIXME blocksize != 4096 */
3142 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3143 if (extent_start != 0) {
3146 inode_sub_bytes(inode, num_dec);
3149 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3151 * we can't truncate inline items that have had
3155 btrfs_file_extent_compression(leaf, fi) == 0 &&
3156 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3157 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3158 u32 size = new_size - found_key.offset;
3160 if (root->ref_cows) {
3161 inode_sub_bytes(inode, item_end + 1 -
3165 btrfs_file_extent_calc_inline_size(size);
3166 btrfs_truncate_item(trans, root, path,
3168 } else if (root->ref_cows) {
3169 inode_sub_bytes(inode, item_end + 1 -
3175 if (!pending_del_nr) {
3176 /* no pending yet, add ourselves */
3177 pending_del_slot = path->slots[0];
3179 } else if (pending_del_nr &&
3180 path->slots[0] + 1 == pending_del_slot) {
3181 /* hop on the pending chunk */
3183 pending_del_slot = path->slots[0];
3190 if (found_extent && (root->ref_cows ||
3191 root == root->fs_info->tree_root)) {
3192 btrfs_set_path_blocking(path);
3193 ret = btrfs_free_extent(trans, root, extent_start,
3194 extent_num_bytes, 0,
3195 btrfs_header_owner(leaf),
3196 ino, extent_offset, 0);
3200 if (found_type == BTRFS_INODE_ITEM_KEY)
3203 if (path->slots[0] == 0 ||
3204 path->slots[0] != pending_del_slot) {
3205 if (root->ref_cows &&
3206 BTRFS_I(inode)->location.objectid !=
3207 BTRFS_FREE_INO_OBJECTID) {
3211 if (pending_del_nr) {
3212 ret = btrfs_del_items(trans, root, path,
3218 btrfs_release_path(path);
3225 if (pending_del_nr) {
3226 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3230 btrfs_free_path(path);
3235 * taken from block_truncate_page, but does cow as it zeros out
3236 * any bytes left in the last page in the file.
3238 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3240 struct inode *inode = mapping->host;
3241 struct btrfs_root *root = BTRFS_I(inode)->root;
3242 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3243 struct btrfs_ordered_extent *ordered;
3244 struct extent_state *cached_state = NULL;
3246 u32 blocksize = root->sectorsize;
3247 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3248 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3250 gfp_t mask = btrfs_alloc_write_mask(mapping);
3255 if ((offset & (blocksize - 1)) == 0)
3257 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3263 page = find_or_create_page(mapping, index, mask);
3265 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3269 page_start = page_offset(page);
3270 page_end = page_start + PAGE_CACHE_SIZE - 1;
3272 if (!PageUptodate(page)) {
3273 ret = btrfs_readpage(NULL, page);
3275 if (page->mapping != mapping) {
3277 page_cache_release(page);
3280 if (!PageUptodate(page)) {
3285 wait_on_page_writeback(page);
3287 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3288 set_page_extent_mapped(page);
3290 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3292 unlock_extent_cached(io_tree, page_start, page_end,
3293 &cached_state, GFP_NOFS);
3295 page_cache_release(page);
3296 btrfs_start_ordered_extent(inode, ordered, 1);
3297 btrfs_put_ordered_extent(ordered);
3301 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3302 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3303 0, 0, &cached_state, GFP_NOFS);
3305 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3308 unlock_extent_cached(io_tree, page_start, page_end,
3309 &cached_state, GFP_NOFS);
3314 if (offset != PAGE_CACHE_SIZE) {
3316 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3317 flush_dcache_page(page);
3320 ClearPageChecked(page);
3321 set_page_dirty(page);
3322 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3327 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3329 page_cache_release(page);
3335 * This function puts in dummy file extents for the area we're creating a hole
3336 * for. So if we are truncating this file to a larger size we need to insert
3337 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3338 * the range between oldsize and size
3340 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3342 struct btrfs_trans_handle *trans;
3343 struct btrfs_root *root = BTRFS_I(inode)->root;
3344 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3345 struct extent_map *em = NULL;
3346 struct extent_state *cached_state = NULL;
3347 u64 mask = root->sectorsize - 1;
3348 u64 hole_start = (oldsize + mask) & ~mask;
3349 u64 block_end = (size + mask) & ~mask;
3355 if (size <= hole_start)
3359 struct btrfs_ordered_extent *ordered;
3360 btrfs_wait_ordered_range(inode, hole_start,
3361 block_end - hole_start);
3362 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3364 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3367 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3368 &cached_state, GFP_NOFS);
3369 btrfs_put_ordered_extent(ordered);
3372 cur_offset = hole_start;
3374 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3375 block_end - cur_offset, 0);
3376 BUG_ON(IS_ERR_OR_NULL(em));
3377 last_byte = min(extent_map_end(em), block_end);
3378 last_byte = (last_byte + mask) & ~mask;
3379 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3381 hole_size = last_byte - cur_offset;
3383 trans = btrfs_start_transaction(root, 3);
3384 if (IS_ERR(trans)) {
3385 err = PTR_ERR(trans);
3389 err = btrfs_drop_extents(trans, inode, cur_offset,
3390 cur_offset + hole_size,
3393 btrfs_update_inode(trans, root, inode);
3394 btrfs_end_transaction(trans, root);
3398 err = btrfs_insert_file_extent(trans, root,
3399 btrfs_ino(inode), cur_offset, 0,
3400 0, hole_size, 0, hole_size,
3403 btrfs_update_inode(trans, root, inode);
3404 btrfs_end_transaction(trans, root);
3408 btrfs_drop_extent_cache(inode, hole_start,
3411 btrfs_update_inode(trans, root, inode);
3412 btrfs_end_transaction(trans, root);
3414 free_extent_map(em);
3416 cur_offset = last_byte;
3417 if (cur_offset >= block_end)
3421 free_extent_map(em);
3422 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3427 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3429 struct btrfs_root *root = BTRFS_I(inode)->root;
3430 struct btrfs_trans_handle *trans;
3431 loff_t oldsize = i_size_read(inode);
3434 if (newsize == oldsize)
3437 if (newsize > oldsize) {
3438 truncate_pagecache(inode, oldsize, newsize);
3439 ret = btrfs_cont_expand(inode, oldsize, newsize);
3443 trans = btrfs_start_transaction(root, 1);
3445 return PTR_ERR(trans);
3447 i_size_write(inode, newsize);
3448 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3449 ret = btrfs_update_inode(trans, root, inode);
3450 btrfs_end_transaction(trans, root);
3454 * We're truncating a file that used to have good data down to
3455 * zero. Make sure it gets into the ordered flush list so that
3456 * any new writes get down to disk quickly.
3459 BTRFS_I(inode)->ordered_data_close = 1;
3461 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3462 truncate_setsize(inode, newsize);
3463 ret = btrfs_truncate(inode);
3469 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3471 struct inode *inode = dentry->d_inode;
3472 struct btrfs_root *root = BTRFS_I(inode)->root;
3475 if (btrfs_root_readonly(root))
3478 err = inode_change_ok(inode, attr);
3482 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3483 err = btrfs_setsize(inode, attr->ia_size);
3488 if (attr->ia_valid) {
3489 setattr_copy(inode, attr);
3490 err = btrfs_dirty_inode(inode);
3492 if (!err && attr->ia_valid & ATTR_MODE)
3493 err = btrfs_acl_chmod(inode);
3499 void btrfs_evict_inode(struct inode *inode)
3501 struct btrfs_trans_handle *trans;
3502 struct btrfs_root *root = BTRFS_I(inode)->root;
3503 struct btrfs_block_rsv *rsv, *global_rsv;
3504 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3508 trace_btrfs_inode_evict(inode);
3510 truncate_inode_pages(&inode->i_data, 0);
3511 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3512 btrfs_is_free_space_inode(root, inode)))
3515 if (is_bad_inode(inode)) {
3516 btrfs_orphan_del(NULL, inode);
3519 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3520 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3522 if (root->fs_info->log_root_recovering) {
3523 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3527 if (inode->i_nlink > 0) {
3528 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3532 rsv = btrfs_alloc_block_rsv(root);
3534 btrfs_orphan_del(NULL, inode);
3537 rsv->size = min_size;
3538 global_rsv = &root->fs_info->global_block_rsv;
3540 btrfs_i_size_write(inode, 0);
3543 * This is a bit simpler than btrfs_truncate since
3545 * 1) We've already reserved our space for our orphan item in the
3547 * 2) We're going to delete the inode item, so we don't need to update
3550 * So we just need to reserve some slack space in case we add bytes when
3551 * doing the truncate.
3554 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3557 * Try and steal from the global reserve since we will
3558 * likely not use this space anyway, we want to try as
3559 * hard as possible to get this to work.
3562 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3565 printk(KERN_WARNING "Could not get space for a "
3566 "delete, will truncate on mount %d\n", ret);
3567 btrfs_orphan_del(NULL, inode);
3568 btrfs_free_block_rsv(root, rsv);
3572 trans = btrfs_start_transaction(root, 0);
3573 if (IS_ERR(trans)) {
3574 btrfs_orphan_del(NULL, inode);
3575 btrfs_free_block_rsv(root, rsv);
3579 trans->block_rsv = rsv;
3581 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3585 nr = trans->blocks_used;
3586 btrfs_end_transaction(trans, root);
3588 btrfs_btree_balance_dirty(root, nr);
3591 btrfs_free_block_rsv(root, rsv);
3594 trans->block_rsv = root->orphan_block_rsv;
3595 ret = btrfs_orphan_del(trans, inode);
3599 trans->block_rsv = &root->fs_info->trans_block_rsv;
3600 if (!(root == root->fs_info->tree_root ||
3601 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3602 btrfs_return_ino(root, btrfs_ino(inode));
3604 nr = trans->blocks_used;
3605 btrfs_end_transaction(trans, root);
3606 btrfs_btree_balance_dirty(root, nr);
3608 end_writeback(inode);
3613 * this returns the key found in the dir entry in the location pointer.
3614 * If no dir entries were found, location->objectid is 0.
3616 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3617 struct btrfs_key *location)
3619 const char *name = dentry->d_name.name;
3620 int namelen = dentry->d_name.len;
3621 struct btrfs_dir_item *di;
3622 struct btrfs_path *path;
3623 struct btrfs_root *root = BTRFS_I(dir)->root;
3626 path = btrfs_alloc_path();
3630 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3635 if (IS_ERR_OR_NULL(di))
3638 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3640 btrfs_free_path(path);
3643 location->objectid = 0;
3648 * when we hit a tree root in a directory, the btrfs part of the inode
3649 * needs to be changed to reflect the root directory of the tree root. This
3650 * is kind of like crossing a mount point.
3652 static int fixup_tree_root_location(struct btrfs_root *root,
3654 struct dentry *dentry,
3655 struct btrfs_key *location,
3656 struct btrfs_root **sub_root)
3658 struct btrfs_path *path;
3659 struct btrfs_root *new_root;
3660 struct btrfs_root_ref *ref;
3661 struct extent_buffer *leaf;
3665 path = btrfs_alloc_path();
3672 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3673 BTRFS_I(dir)->root->root_key.objectid,
3674 location->objectid);
3681 leaf = path->nodes[0];
3682 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3683 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3684 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3687 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3688 (unsigned long)(ref + 1),
3689 dentry->d_name.len);
3693 btrfs_release_path(path);
3695 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3696 if (IS_ERR(new_root)) {
3697 err = PTR_ERR(new_root);
3701 if (btrfs_root_refs(&new_root->root_item) == 0) {
3706 *sub_root = new_root;
3707 location->objectid = btrfs_root_dirid(&new_root->root_item);
3708 location->type = BTRFS_INODE_ITEM_KEY;
3709 location->offset = 0;
3712 btrfs_free_path(path);
3716 static void inode_tree_add(struct inode *inode)
3718 struct btrfs_root *root = BTRFS_I(inode)->root;
3719 struct btrfs_inode *entry;
3721 struct rb_node *parent;
3722 u64 ino = btrfs_ino(inode);
3724 p = &root->inode_tree.rb_node;
3727 if (inode_unhashed(inode))
3730 spin_lock(&root->inode_lock);
3733 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3735 if (ino < btrfs_ino(&entry->vfs_inode))
3736 p = &parent->rb_left;
3737 else if (ino > btrfs_ino(&entry->vfs_inode))
3738 p = &parent->rb_right;
3740 WARN_ON(!(entry->vfs_inode.i_state &
3741 (I_WILL_FREE | I_FREEING)));
3742 rb_erase(parent, &root->inode_tree);
3743 RB_CLEAR_NODE(parent);
3744 spin_unlock(&root->inode_lock);
3748 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3749 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3750 spin_unlock(&root->inode_lock);
3753 static void inode_tree_del(struct inode *inode)
3755 struct btrfs_root *root = BTRFS_I(inode)->root;
3758 spin_lock(&root->inode_lock);
3759 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3760 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3761 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3762 empty = RB_EMPTY_ROOT(&root->inode_tree);
3764 spin_unlock(&root->inode_lock);
3767 * Free space cache has inodes in the tree root, but the tree root has a
3768 * root_refs of 0, so this could end up dropping the tree root as a
3769 * snapshot, so we need the extra !root->fs_info->tree_root check to
3770 * make sure we don't drop it.
3772 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3773 root != root->fs_info->tree_root) {
3774 synchronize_srcu(&root->fs_info->subvol_srcu);
3775 spin_lock(&root->inode_lock);
3776 empty = RB_EMPTY_ROOT(&root->inode_tree);
3777 spin_unlock(&root->inode_lock);
3779 btrfs_add_dead_root(root);
3783 void btrfs_invalidate_inodes(struct btrfs_root *root)
3785 struct rb_node *node;
3786 struct rb_node *prev;
3787 struct btrfs_inode *entry;
3788 struct inode *inode;
3791 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3793 spin_lock(&root->inode_lock);
3795 node = root->inode_tree.rb_node;
3799 entry = rb_entry(node, struct btrfs_inode, rb_node);
3801 if (objectid < btrfs_ino(&entry->vfs_inode))
3802 node = node->rb_left;
3803 else if (objectid > btrfs_ino(&entry->vfs_inode))
3804 node = node->rb_right;
3810 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3811 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3815 prev = rb_next(prev);
3819 entry = rb_entry(node, struct btrfs_inode, rb_node);
3820 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3821 inode = igrab(&entry->vfs_inode);
3823 spin_unlock(&root->inode_lock);
3824 if (atomic_read(&inode->i_count) > 1)
3825 d_prune_aliases(inode);
3827 * btrfs_drop_inode will have it removed from
3828 * the inode cache when its usage count
3833 spin_lock(&root->inode_lock);
3837 if (cond_resched_lock(&root->inode_lock))
3840 node = rb_next(node);
3842 spin_unlock(&root->inode_lock);
3845 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3847 struct btrfs_iget_args *args = p;
3848 inode->i_ino = args->ino;
3849 BTRFS_I(inode)->root = args->root;
3850 btrfs_set_inode_space_info(args->root, inode);
3854 static int btrfs_find_actor(struct inode *inode, void *opaque)
3856 struct btrfs_iget_args *args = opaque;
3857 return args->ino == btrfs_ino(inode) &&
3858 args->root == BTRFS_I(inode)->root;
3861 static struct inode *btrfs_iget_locked(struct super_block *s,
3863 struct btrfs_root *root)
3865 struct inode *inode;
3866 struct btrfs_iget_args args;
3867 args.ino = objectid;
3870 inode = iget5_locked(s, objectid, btrfs_find_actor,
3871 btrfs_init_locked_inode,
3876 /* Get an inode object given its location and corresponding root.
3877 * Returns in *is_new if the inode was read from disk
3879 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3880 struct btrfs_root *root, int *new)
3882 struct inode *inode;
3884 inode = btrfs_iget_locked(s, location->objectid, root);
3886 return ERR_PTR(-ENOMEM);
3888 if (inode->i_state & I_NEW) {
3889 BTRFS_I(inode)->root = root;
3890 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3891 btrfs_read_locked_inode(inode);
3892 if (!is_bad_inode(inode)) {
3893 inode_tree_add(inode);
3894 unlock_new_inode(inode);
3898 unlock_new_inode(inode);
3900 inode = ERR_PTR(-ESTALE);
3907 static struct inode *new_simple_dir(struct super_block *s,
3908 struct btrfs_key *key,
3909 struct btrfs_root *root)
3911 struct inode *inode = new_inode(s);
3914 return ERR_PTR(-ENOMEM);
3916 BTRFS_I(inode)->root = root;
3917 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3918 BTRFS_I(inode)->dummy_inode = 1;
3920 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3921 inode->i_op = &simple_dir_inode_operations;
3922 inode->i_fop = &simple_dir_operations;
3923 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3924 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3929 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3931 struct inode *inode;
3932 struct btrfs_root *root = BTRFS_I(dir)->root;
3933 struct btrfs_root *sub_root = root;
3934 struct btrfs_key location;
3938 if (dentry->d_name.len > BTRFS_NAME_LEN)
3939 return ERR_PTR(-ENAMETOOLONG);
3941 if (unlikely(d_need_lookup(dentry))) {
3942 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3943 kfree(dentry->d_fsdata);
3944 dentry->d_fsdata = NULL;
3945 /* This thing is hashed, drop it for now */
3948 ret = btrfs_inode_by_name(dir, dentry, &location);
3952 return ERR_PTR(ret);
3954 if (location.objectid == 0)
3957 if (location.type == BTRFS_INODE_ITEM_KEY) {
3958 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3962 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3964 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3965 ret = fixup_tree_root_location(root, dir, dentry,
3966 &location, &sub_root);
3969 inode = ERR_PTR(ret);
3971 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3973 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3975 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3977 if (!IS_ERR(inode) && root != sub_root) {
3978 down_read(&root->fs_info->cleanup_work_sem);
3979 if (!(inode->i_sb->s_flags & MS_RDONLY))
3980 ret = btrfs_orphan_cleanup(sub_root);
3981 up_read(&root->fs_info->cleanup_work_sem);
3983 inode = ERR_PTR(ret);
3989 static int btrfs_dentry_delete(const struct dentry *dentry)
3991 struct btrfs_root *root;
3993 if (!dentry->d_inode && !IS_ROOT(dentry))
3994 dentry = dentry->d_parent;
3996 if (dentry->d_inode) {
3997 root = BTRFS_I(dentry->d_inode)->root;
3998 if (btrfs_root_refs(&root->root_item) == 0)
4004 static void btrfs_dentry_release(struct dentry *dentry)
4006 if (dentry->d_fsdata)
4007 kfree(dentry->d_fsdata);
4010 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4011 struct nameidata *nd)
4015 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4016 if (unlikely(d_need_lookup(dentry))) {
4017 spin_lock(&dentry->d_lock);
4018 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4019 spin_unlock(&dentry->d_lock);
4024 unsigned char btrfs_filetype_table[] = {
4025 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4028 static int btrfs_real_readdir(struct file *filp, void *dirent,
4031 struct inode *inode = filp->f_dentry->d_inode;
4032 struct btrfs_root *root = BTRFS_I(inode)->root;
4033 struct btrfs_item *item;
4034 struct btrfs_dir_item *di;
4035 struct btrfs_key key;
4036 struct btrfs_key found_key;
4037 struct btrfs_path *path;
4038 struct list_head ins_list;
4039 struct list_head del_list;
4042 struct extent_buffer *leaf;
4044 unsigned char d_type;
4049 int key_type = BTRFS_DIR_INDEX_KEY;
4053 int is_curr = 0; /* filp->f_pos points to the current index? */
4055 /* FIXME, use a real flag for deciding about the key type */
4056 if (root->fs_info->tree_root == root)
4057 key_type = BTRFS_DIR_ITEM_KEY;
4059 /* special case for "." */
4060 if (filp->f_pos == 0) {
4061 over = filldir(dirent, ".", 1,
4062 filp->f_pos, btrfs_ino(inode), DT_DIR);
4067 /* special case for .., just use the back ref */
4068 if (filp->f_pos == 1) {
4069 u64 pino = parent_ino(filp->f_path.dentry);
4070 over = filldir(dirent, "..", 2,
4071 filp->f_pos, pino, DT_DIR);
4076 path = btrfs_alloc_path();
4082 if (key_type == BTRFS_DIR_INDEX_KEY) {
4083 INIT_LIST_HEAD(&ins_list);
4084 INIT_LIST_HEAD(&del_list);
4085 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4088 btrfs_set_key_type(&key, key_type);
4089 key.offset = filp->f_pos;
4090 key.objectid = btrfs_ino(inode);
4092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4097 leaf = path->nodes[0];
4098 slot = path->slots[0];
4099 if (slot >= btrfs_header_nritems(leaf)) {
4100 ret = btrfs_next_leaf(root, path);
4108 item = btrfs_item_nr(leaf, slot);
4109 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4111 if (found_key.objectid != key.objectid)
4113 if (btrfs_key_type(&found_key) != key_type)
4115 if (found_key.offset < filp->f_pos)
4117 if (key_type == BTRFS_DIR_INDEX_KEY &&
4118 btrfs_should_delete_dir_index(&del_list,
4122 filp->f_pos = found_key.offset;
4125 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4127 di_total = btrfs_item_size(leaf, item);
4129 while (di_cur < di_total) {
4130 struct btrfs_key location;
4133 if (verify_dir_item(root, leaf, di))
4136 name_len = btrfs_dir_name_len(leaf, di);
4137 if (name_len <= sizeof(tmp_name)) {
4138 name_ptr = tmp_name;
4140 name_ptr = kmalloc(name_len, GFP_NOFS);
4146 read_extent_buffer(leaf, name_ptr,
4147 (unsigned long)(di + 1), name_len);
4149 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4150 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4154 q.hash = full_name_hash(q.name, q.len);
4155 tmp = d_lookup(filp->f_dentry, &q);
4157 struct btrfs_key *newkey;
4159 newkey = kzalloc(sizeof(struct btrfs_key),
4163 tmp = d_alloc(filp->f_dentry, &q);
4169 memcpy(newkey, &location,
4170 sizeof(struct btrfs_key));
4171 tmp->d_fsdata = newkey;
4172 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4179 /* is this a reference to our own snapshot? If so
4182 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4183 location.objectid == root->root_key.objectid) {
4187 over = filldir(dirent, name_ptr, name_len,
4188 found_key.offset, location.objectid,
4192 if (name_ptr != tmp_name)
4197 di_len = btrfs_dir_name_len(leaf, di) +
4198 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4200 di = (struct btrfs_dir_item *)((char *)di + di_len);
4206 if (key_type == BTRFS_DIR_INDEX_KEY) {
4209 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4215 /* Reached end of directory/root. Bump pos past the last item. */
4216 if (key_type == BTRFS_DIR_INDEX_KEY)
4218 * 32-bit glibc will use getdents64, but then strtol -
4219 * so the last number we can serve is this.
4221 filp->f_pos = 0x7fffffff;
4227 if (key_type == BTRFS_DIR_INDEX_KEY)
4228 btrfs_put_delayed_items(&ins_list, &del_list);
4229 btrfs_free_path(path);
4233 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4235 struct btrfs_root *root = BTRFS_I(inode)->root;
4236 struct btrfs_trans_handle *trans;
4238 bool nolock = false;
4240 if (BTRFS_I(inode)->dummy_inode)
4243 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4246 if (wbc->sync_mode == WB_SYNC_ALL) {
4248 trans = btrfs_join_transaction_nolock(root);
4250 trans = btrfs_join_transaction(root);
4252 return PTR_ERR(trans);
4254 ret = btrfs_end_transaction_nolock(trans, root);
4256 ret = btrfs_commit_transaction(trans, root);
4262 * This is somewhat expensive, updating the tree every time the
4263 * inode changes. But, it is most likely to find the inode in cache.
4264 * FIXME, needs more benchmarking...there are no reasons other than performance
4265 * to keep or drop this code.
4267 int btrfs_dirty_inode(struct inode *inode)
4269 struct btrfs_root *root = BTRFS_I(inode)->root;
4270 struct btrfs_trans_handle *trans;
4273 if (BTRFS_I(inode)->dummy_inode)
4276 trans = btrfs_join_transaction(root);
4278 return PTR_ERR(trans);
4280 ret = btrfs_update_inode(trans, root, inode);
4281 if (ret && ret == -ENOSPC) {
4282 /* whoops, lets try again with the full transaction */
4283 btrfs_end_transaction(trans, root);
4284 trans = btrfs_start_transaction(root, 1);
4286 return PTR_ERR(trans);
4288 ret = btrfs_update_inode(trans, root, inode);
4290 btrfs_end_transaction(trans, root);
4291 if (BTRFS_I(inode)->delayed_node)
4292 btrfs_balance_delayed_items(root);
4298 * This is a copy of file_update_time. We need this so we can return error on
4299 * ENOSPC for updating the inode in the case of file write and mmap writes.
4301 int btrfs_update_time(struct file *file)
4303 struct inode *inode = file->f_path.dentry->d_inode;
4304 struct timespec now;
4306 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4308 /* First try to exhaust all avenues to not sync */
4309 if (IS_NOCMTIME(inode))
4312 now = current_fs_time(inode->i_sb);
4313 if (!timespec_equal(&inode->i_mtime, &now))
4316 if (!timespec_equal(&inode->i_ctime, &now))
4319 if (IS_I_VERSION(inode))
4320 sync_it |= S_VERSION;
4325 /* Finally allowed to write? Takes lock. */
4326 if (mnt_want_write_file(file))
4329 /* Only change inode inside the lock region */
4330 if (sync_it & S_VERSION)
4331 inode_inc_iversion(inode);
4332 if (sync_it & S_CTIME)
4333 inode->i_ctime = now;
4334 if (sync_it & S_MTIME)
4335 inode->i_mtime = now;
4336 ret = btrfs_dirty_inode(inode);
4338 mark_inode_dirty_sync(inode);
4339 mnt_drop_write(file->f_path.mnt);
4344 * find the highest existing sequence number in a directory
4345 * and then set the in-memory index_cnt variable to reflect
4346 * free sequence numbers
4348 static int btrfs_set_inode_index_count(struct inode *inode)
4350 struct btrfs_root *root = BTRFS_I(inode)->root;
4351 struct btrfs_key key, found_key;
4352 struct btrfs_path *path;
4353 struct extent_buffer *leaf;
4356 key.objectid = btrfs_ino(inode);
4357 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4358 key.offset = (u64)-1;
4360 path = btrfs_alloc_path();
4364 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4367 /* FIXME: we should be able to handle this */
4373 * MAGIC NUMBER EXPLANATION:
4374 * since we search a directory based on f_pos we have to start at 2
4375 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4376 * else has to start at 2
4378 if (path->slots[0] == 0) {
4379 BTRFS_I(inode)->index_cnt = 2;
4385 leaf = path->nodes[0];
4386 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4388 if (found_key.objectid != btrfs_ino(inode) ||
4389 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4390 BTRFS_I(inode)->index_cnt = 2;
4394 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4396 btrfs_free_path(path);
4401 * helper to find a free sequence number in a given directory. This current
4402 * code is very simple, later versions will do smarter things in the btree
4404 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4408 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4409 ret = btrfs_inode_delayed_dir_index_count(dir);
4411 ret = btrfs_set_inode_index_count(dir);
4417 *index = BTRFS_I(dir)->index_cnt;
4418 BTRFS_I(dir)->index_cnt++;
4423 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4424 struct btrfs_root *root,
4426 const char *name, int name_len,
4427 u64 ref_objectid, u64 objectid,
4428 umode_t mode, u64 *index)
4430 struct inode *inode;
4431 struct btrfs_inode_item *inode_item;
4432 struct btrfs_key *location;
4433 struct btrfs_path *path;
4434 struct btrfs_inode_ref *ref;
4435 struct btrfs_key key[2];
4441 path = btrfs_alloc_path();
4443 return ERR_PTR(-ENOMEM);
4445 inode = new_inode(root->fs_info->sb);
4447 btrfs_free_path(path);
4448 return ERR_PTR(-ENOMEM);
4452 * we have to initialize this early, so we can reclaim the inode
4453 * number if we fail afterwards in this function.
4455 inode->i_ino = objectid;
4458 trace_btrfs_inode_request(dir);
4460 ret = btrfs_set_inode_index(dir, index);
4462 btrfs_free_path(path);
4464 return ERR_PTR(ret);
4468 * index_cnt is ignored for everything but a dir,
4469 * btrfs_get_inode_index_count has an explanation for the magic
4472 BTRFS_I(inode)->index_cnt = 2;
4473 BTRFS_I(inode)->root = root;
4474 BTRFS_I(inode)->generation = trans->transid;
4475 inode->i_generation = BTRFS_I(inode)->generation;
4476 btrfs_set_inode_space_info(root, inode);
4483 key[0].objectid = objectid;
4484 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4487 key[1].objectid = objectid;
4488 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4489 key[1].offset = ref_objectid;
4491 sizes[0] = sizeof(struct btrfs_inode_item);
4492 sizes[1] = name_len + sizeof(*ref);
4494 path->leave_spinning = 1;
4495 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4499 inode_init_owner(inode, dir, mode);
4500 inode_set_bytes(inode, 0);
4501 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4502 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4503 struct btrfs_inode_item);
4504 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4506 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4507 struct btrfs_inode_ref);
4508 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4509 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4510 ptr = (unsigned long)(ref + 1);
4511 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4513 btrfs_mark_buffer_dirty(path->nodes[0]);
4514 btrfs_free_path(path);
4516 location = &BTRFS_I(inode)->location;
4517 location->objectid = objectid;
4518 location->offset = 0;
4519 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4521 btrfs_inherit_iflags(inode, dir);
4523 if (S_ISREG(mode)) {
4524 if (btrfs_test_opt(root, NODATASUM))
4525 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4526 if (btrfs_test_opt(root, NODATACOW) ||
4527 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4528 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4531 insert_inode_hash(inode);
4532 inode_tree_add(inode);
4534 trace_btrfs_inode_new(inode);
4535 btrfs_set_inode_last_trans(trans, inode);
4540 BTRFS_I(dir)->index_cnt--;
4541 btrfs_free_path(path);
4543 return ERR_PTR(ret);
4546 static inline u8 btrfs_inode_type(struct inode *inode)
4548 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4552 * utility function to add 'inode' into 'parent_inode' with
4553 * a give name and a given sequence number.
4554 * if 'add_backref' is true, also insert a backref from the
4555 * inode to the parent directory.
4557 int btrfs_add_link(struct btrfs_trans_handle *trans,
4558 struct inode *parent_inode, struct inode *inode,
4559 const char *name, int name_len, int add_backref, u64 index)
4562 struct btrfs_key key;
4563 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4564 u64 ino = btrfs_ino(inode);
4565 u64 parent_ino = btrfs_ino(parent_inode);
4567 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4568 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4571 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4575 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4576 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4577 key.objectid, root->root_key.objectid,
4578 parent_ino, index, name, name_len);
4579 } else if (add_backref) {
4580 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4585 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4587 btrfs_inode_type(inode), index);
4591 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4593 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4594 ret = btrfs_update_inode(trans, root, parent_inode);
4599 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4602 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4603 key.objectid, root->root_key.objectid,
4604 parent_ino, &local_index, name, name_len);
4606 } else if (add_backref) {
4610 err = btrfs_del_inode_ref(trans, root, name, name_len,
4611 ino, parent_ino, &local_index);
4616 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4617 struct inode *dir, struct dentry *dentry,
4618 struct inode *inode, int backref, u64 index)
4620 int err = btrfs_add_link(trans, dir, inode,
4621 dentry->d_name.name, dentry->d_name.len,
4628 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4629 umode_t mode, dev_t rdev)
4631 struct btrfs_trans_handle *trans;
4632 struct btrfs_root *root = BTRFS_I(dir)->root;
4633 struct inode *inode = NULL;
4637 unsigned long nr = 0;
4640 if (!new_valid_dev(rdev))
4644 * 2 for inode item and ref
4646 * 1 for xattr if selinux is on
4648 trans = btrfs_start_transaction(root, 5);
4650 return PTR_ERR(trans);
4652 err = btrfs_find_free_ino(root, &objectid);
4656 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4657 dentry->d_name.len, btrfs_ino(dir), objectid,
4659 if (IS_ERR(inode)) {
4660 err = PTR_ERR(inode);
4664 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4671 * If the active LSM wants to access the inode during
4672 * d_instantiate it needs these. Smack checks to see
4673 * if the filesystem supports xattrs by looking at the
4677 inode->i_op = &btrfs_special_inode_operations;
4678 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4682 init_special_inode(inode, inode->i_mode, rdev);
4683 btrfs_update_inode(trans, root, inode);
4684 d_instantiate(dentry, inode);
4687 nr = trans->blocks_used;
4688 btrfs_end_transaction(trans, root);
4689 btrfs_btree_balance_dirty(root, nr);
4691 inode_dec_link_count(inode);
4697 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4698 umode_t mode, struct nameidata *nd)
4700 struct btrfs_trans_handle *trans;
4701 struct btrfs_root *root = BTRFS_I(dir)->root;
4702 struct inode *inode = NULL;
4705 unsigned long nr = 0;
4710 * 2 for inode item and ref
4712 * 1 for xattr if selinux is on
4714 trans = btrfs_start_transaction(root, 5);
4716 return PTR_ERR(trans);
4718 err = btrfs_find_free_ino(root, &objectid);
4722 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4723 dentry->d_name.len, btrfs_ino(dir), objectid,
4725 if (IS_ERR(inode)) {
4726 err = PTR_ERR(inode);
4730 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4737 * If the active LSM wants to access the inode during
4738 * d_instantiate it needs these. Smack checks to see
4739 * if the filesystem supports xattrs by looking at the
4742 inode->i_fop = &btrfs_file_operations;
4743 inode->i_op = &btrfs_file_inode_operations;
4745 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4749 inode->i_mapping->a_ops = &btrfs_aops;
4750 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4751 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4752 d_instantiate(dentry, inode);
4755 nr = trans->blocks_used;
4756 btrfs_end_transaction(trans, root);
4758 inode_dec_link_count(inode);
4761 btrfs_btree_balance_dirty(root, nr);
4765 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4766 struct dentry *dentry)
4768 struct btrfs_trans_handle *trans;
4769 struct btrfs_root *root = BTRFS_I(dir)->root;
4770 struct inode *inode = old_dentry->d_inode;
4772 unsigned long nr = 0;
4776 /* do not allow sys_link's with other subvols of the same device */
4777 if (root->objectid != BTRFS_I(inode)->root->objectid)
4780 if (inode->i_nlink == ~0U)
4783 err = btrfs_set_inode_index(dir, &index);
4788 * 2 items for inode and inode ref
4789 * 2 items for dir items
4790 * 1 item for parent inode
4792 trans = btrfs_start_transaction(root, 5);
4793 if (IS_ERR(trans)) {
4794 err = PTR_ERR(trans);
4798 btrfs_inc_nlink(inode);
4799 inode->i_ctime = CURRENT_TIME;
4802 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4807 struct dentry *parent = dentry->d_parent;
4808 err = btrfs_update_inode(trans, root, inode);
4810 d_instantiate(dentry, inode);
4811 btrfs_log_new_name(trans, inode, NULL, parent);
4814 nr = trans->blocks_used;
4815 btrfs_end_transaction(trans, root);
4818 inode_dec_link_count(inode);
4821 btrfs_btree_balance_dirty(root, nr);
4825 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4827 struct inode *inode = NULL;
4828 struct btrfs_trans_handle *trans;
4829 struct btrfs_root *root = BTRFS_I(dir)->root;
4831 int drop_on_err = 0;
4834 unsigned long nr = 1;
4837 * 2 items for inode and ref
4838 * 2 items for dir items
4839 * 1 for xattr if selinux is on
4841 trans = btrfs_start_transaction(root, 5);
4843 return PTR_ERR(trans);
4845 err = btrfs_find_free_ino(root, &objectid);
4849 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4850 dentry->d_name.len, btrfs_ino(dir), objectid,
4851 S_IFDIR | mode, &index);
4852 if (IS_ERR(inode)) {
4853 err = PTR_ERR(inode);
4859 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4863 inode->i_op = &btrfs_dir_inode_operations;
4864 inode->i_fop = &btrfs_dir_file_operations;
4866 btrfs_i_size_write(inode, 0);
4867 err = btrfs_update_inode(trans, root, inode);
4871 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4872 dentry->d_name.len, 0, index);
4876 d_instantiate(dentry, inode);
4880 nr = trans->blocks_used;
4881 btrfs_end_transaction(trans, root);
4884 btrfs_btree_balance_dirty(root, nr);
4888 /* helper for btfs_get_extent. Given an existing extent in the tree,
4889 * and an extent that you want to insert, deal with overlap and insert
4890 * the new extent into the tree.
4892 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4893 struct extent_map *existing,
4894 struct extent_map *em,
4895 u64 map_start, u64 map_len)
4899 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4900 start_diff = map_start - em->start;
4901 em->start = map_start;
4903 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4904 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4905 em->block_start += start_diff;
4906 em->block_len -= start_diff;
4908 return add_extent_mapping(em_tree, em);
4911 static noinline int uncompress_inline(struct btrfs_path *path,
4912 struct inode *inode, struct page *page,
4913 size_t pg_offset, u64 extent_offset,
4914 struct btrfs_file_extent_item *item)
4917 struct extent_buffer *leaf = path->nodes[0];
4920 unsigned long inline_size;
4924 WARN_ON(pg_offset != 0);
4925 compress_type = btrfs_file_extent_compression(leaf, item);
4926 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4927 inline_size = btrfs_file_extent_inline_item_len(leaf,
4928 btrfs_item_nr(leaf, path->slots[0]));
4929 tmp = kmalloc(inline_size, GFP_NOFS);
4932 ptr = btrfs_file_extent_inline_start(item);
4934 read_extent_buffer(leaf, tmp, ptr, inline_size);
4936 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4937 ret = btrfs_decompress(compress_type, tmp, page,
4938 extent_offset, inline_size, max_size);
4940 char *kaddr = kmap_atomic(page, KM_USER0);
4941 unsigned long copy_size = min_t(u64,
4942 PAGE_CACHE_SIZE - pg_offset,
4943 max_size - extent_offset);
4944 memset(kaddr + pg_offset, 0, copy_size);
4945 kunmap_atomic(kaddr, KM_USER0);
4952 * a bit scary, this does extent mapping from logical file offset to the disk.
4953 * the ugly parts come from merging extents from the disk with the in-ram
4954 * representation. This gets more complex because of the data=ordered code,
4955 * where the in-ram extents might be locked pending data=ordered completion.
4957 * This also copies inline extents directly into the page.
4960 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4961 size_t pg_offset, u64 start, u64 len,
4967 u64 extent_start = 0;
4969 u64 objectid = btrfs_ino(inode);
4971 struct btrfs_path *path = NULL;
4972 struct btrfs_root *root = BTRFS_I(inode)->root;
4973 struct btrfs_file_extent_item *item;
4974 struct extent_buffer *leaf;
4975 struct btrfs_key found_key;
4976 struct extent_map *em = NULL;
4977 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4978 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4979 struct btrfs_trans_handle *trans = NULL;
4983 read_lock(&em_tree->lock);
4984 em = lookup_extent_mapping(em_tree, start, len);
4986 em->bdev = root->fs_info->fs_devices->latest_bdev;
4987 read_unlock(&em_tree->lock);
4990 if (em->start > start || em->start + em->len <= start)
4991 free_extent_map(em);
4992 else if (em->block_start == EXTENT_MAP_INLINE && page)
4993 free_extent_map(em);
4997 em = alloc_extent_map();
5002 em->bdev = root->fs_info->fs_devices->latest_bdev;
5003 em->start = EXTENT_MAP_HOLE;
5004 em->orig_start = EXTENT_MAP_HOLE;
5006 em->block_len = (u64)-1;
5009 path = btrfs_alloc_path();
5015 * Chances are we'll be called again, so go ahead and do
5021 ret = btrfs_lookup_file_extent(trans, root, path,
5022 objectid, start, trans != NULL);
5029 if (path->slots[0] == 0)
5034 leaf = path->nodes[0];
5035 item = btrfs_item_ptr(leaf, path->slots[0],
5036 struct btrfs_file_extent_item);
5037 /* are we inside the extent that was found? */
5038 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5039 found_type = btrfs_key_type(&found_key);
5040 if (found_key.objectid != objectid ||
5041 found_type != BTRFS_EXTENT_DATA_KEY) {
5045 found_type = btrfs_file_extent_type(leaf, item);
5046 extent_start = found_key.offset;
5047 compress_type = btrfs_file_extent_compression(leaf, item);
5048 if (found_type == BTRFS_FILE_EXTENT_REG ||
5049 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5050 extent_end = extent_start +
5051 btrfs_file_extent_num_bytes(leaf, item);
5052 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5054 size = btrfs_file_extent_inline_len(leaf, item);
5055 extent_end = (extent_start + size + root->sectorsize - 1) &
5056 ~((u64)root->sectorsize - 1);
5059 if (start >= extent_end) {
5061 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5062 ret = btrfs_next_leaf(root, path);
5069 leaf = path->nodes[0];
5071 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5072 if (found_key.objectid != objectid ||
5073 found_key.type != BTRFS_EXTENT_DATA_KEY)
5075 if (start + len <= found_key.offset)
5078 em->len = found_key.offset - start;
5082 if (found_type == BTRFS_FILE_EXTENT_REG ||
5083 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5084 em->start = extent_start;
5085 em->len = extent_end - extent_start;
5086 em->orig_start = extent_start -
5087 btrfs_file_extent_offset(leaf, item);
5088 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5090 em->block_start = EXTENT_MAP_HOLE;
5093 if (compress_type != BTRFS_COMPRESS_NONE) {
5094 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5095 em->compress_type = compress_type;
5096 em->block_start = bytenr;
5097 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5100 bytenr += btrfs_file_extent_offset(leaf, item);
5101 em->block_start = bytenr;
5102 em->block_len = em->len;
5103 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5104 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5107 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5111 size_t extent_offset;
5114 em->block_start = EXTENT_MAP_INLINE;
5115 if (!page || create) {
5116 em->start = extent_start;
5117 em->len = extent_end - extent_start;
5121 size = btrfs_file_extent_inline_len(leaf, item);
5122 extent_offset = page_offset(page) + pg_offset - extent_start;
5123 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5124 size - extent_offset);
5125 em->start = extent_start + extent_offset;
5126 em->len = (copy_size + root->sectorsize - 1) &
5127 ~((u64)root->sectorsize - 1);
5128 em->orig_start = EXTENT_MAP_INLINE;
5129 if (compress_type) {
5130 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5131 em->compress_type = compress_type;
5133 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5134 if (create == 0 && !PageUptodate(page)) {
5135 if (btrfs_file_extent_compression(leaf, item) !=
5136 BTRFS_COMPRESS_NONE) {
5137 ret = uncompress_inline(path, inode, page,
5139 extent_offset, item);
5143 read_extent_buffer(leaf, map + pg_offset, ptr,
5145 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5146 memset(map + pg_offset + copy_size, 0,
5147 PAGE_CACHE_SIZE - pg_offset -
5152 flush_dcache_page(page);
5153 } else if (create && PageUptodate(page)) {
5157 free_extent_map(em);
5160 btrfs_release_path(path);
5161 trans = btrfs_join_transaction(root);
5164 return ERR_CAST(trans);
5168 write_extent_buffer(leaf, map + pg_offset, ptr,
5171 btrfs_mark_buffer_dirty(leaf);
5173 set_extent_uptodate(io_tree, em->start,
5174 extent_map_end(em) - 1, NULL, GFP_NOFS);
5177 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5184 em->block_start = EXTENT_MAP_HOLE;
5185 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5187 btrfs_release_path(path);
5188 if (em->start > start || extent_map_end(em) <= start) {
5189 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5190 "[%llu %llu]\n", (unsigned long long)em->start,
5191 (unsigned long long)em->len,
5192 (unsigned long long)start,
5193 (unsigned long long)len);
5199 write_lock(&em_tree->lock);
5200 ret = add_extent_mapping(em_tree, em);
5201 /* it is possible that someone inserted the extent into the tree
5202 * while we had the lock dropped. It is also possible that
5203 * an overlapping map exists in the tree
5205 if (ret == -EEXIST) {
5206 struct extent_map *existing;
5210 existing = lookup_extent_mapping(em_tree, start, len);
5211 if (existing && (existing->start > start ||
5212 existing->start + existing->len <= start)) {
5213 free_extent_map(existing);
5217 existing = lookup_extent_mapping(em_tree, em->start,
5220 err = merge_extent_mapping(em_tree, existing,
5223 free_extent_map(existing);
5225 free_extent_map(em);
5230 free_extent_map(em);
5234 free_extent_map(em);
5239 write_unlock(&em_tree->lock);
5242 trace_btrfs_get_extent(root, em);
5245 btrfs_free_path(path);
5247 ret = btrfs_end_transaction(trans, root);
5252 free_extent_map(em);
5253 return ERR_PTR(err);
5258 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5259 size_t pg_offset, u64 start, u64 len,
5262 struct extent_map *em;
5263 struct extent_map *hole_em = NULL;
5264 u64 range_start = start;
5270 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5275 * if our em maps to a hole, there might
5276 * actually be delalloc bytes behind it
5278 if (em->block_start != EXTENT_MAP_HOLE)
5284 /* check to see if we've wrapped (len == -1 or similar) */
5293 /* ok, we didn't find anything, lets look for delalloc */
5294 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5295 end, len, EXTENT_DELALLOC, 1);
5296 found_end = range_start + found;
5297 if (found_end < range_start)
5298 found_end = (u64)-1;
5301 * we didn't find anything useful, return
5302 * the original results from get_extent()
5304 if (range_start > end || found_end <= start) {
5310 /* adjust the range_start to make sure it doesn't
5311 * go backwards from the start they passed in
5313 range_start = max(start,range_start);
5314 found = found_end - range_start;
5317 u64 hole_start = start;
5320 em = alloc_extent_map();
5326 * when btrfs_get_extent can't find anything it
5327 * returns one huge hole
5329 * make sure what it found really fits our range, and
5330 * adjust to make sure it is based on the start from
5334 u64 calc_end = extent_map_end(hole_em);
5336 if (calc_end <= start || (hole_em->start > end)) {
5337 free_extent_map(hole_em);
5340 hole_start = max(hole_em->start, start);
5341 hole_len = calc_end - hole_start;
5345 if (hole_em && range_start > hole_start) {
5346 /* our hole starts before our delalloc, so we
5347 * have to return just the parts of the hole
5348 * that go until the delalloc starts
5350 em->len = min(hole_len,
5351 range_start - hole_start);
5352 em->start = hole_start;
5353 em->orig_start = hole_start;
5355 * don't adjust block start at all,
5356 * it is fixed at EXTENT_MAP_HOLE
5358 em->block_start = hole_em->block_start;
5359 em->block_len = hole_len;
5361 em->start = range_start;
5363 em->orig_start = range_start;
5364 em->block_start = EXTENT_MAP_DELALLOC;
5365 em->block_len = found;
5367 } else if (hole_em) {
5372 free_extent_map(hole_em);
5374 free_extent_map(em);
5375 return ERR_PTR(err);
5380 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5381 struct extent_map *em,
5384 struct btrfs_root *root = BTRFS_I(inode)->root;
5385 struct btrfs_trans_handle *trans;
5386 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5387 struct btrfs_key ins;
5390 bool insert = false;
5393 * Ok if the extent map we looked up is a hole and is for the exact
5394 * range we want, there is no reason to allocate a new one, however if
5395 * it is not right then we need to free this one and drop the cache for
5398 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5400 free_extent_map(em);
5403 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5406 trans = btrfs_join_transaction(root);
5408 return ERR_CAST(trans);
5410 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5411 btrfs_add_inode_defrag(trans, inode);
5413 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5415 alloc_hint = get_extent_allocation_hint(inode, start, len);
5416 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5417 alloc_hint, (u64)-1, &ins, 1);
5424 em = alloc_extent_map();
5426 em = ERR_PTR(-ENOMEM);
5432 em->orig_start = em->start;
5433 em->len = ins.offset;
5435 em->block_start = ins.objectid;
5436 em->block_len = ins.offset;
5437 em->bdev = root->fs_info->fs_devices->latest_bdev;
5440 * We need to do this because if we're using the original em we searched
5441 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5444 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5447 write_lock(&em_tree->lock);
5448 ret = add_extent_mapping(em_tree, em);
5449 write_unlock(&em_tree->lock);
5452 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5455 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5456 ins.offset, ins.offset, 0);
5458 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5462 btrfs_end_transaction(trans, root);
5467 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5468 * block must be cow'd
5470 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5471 struct inode *inode, u64 offset, u64 len)
5473 struct btrfs_path *path;
5475 struct extent_buffer *leaf;
5476 struct btrfs_root *root = BTRFS_I(inode)->root;
5477 struct btrfs_file_extent_item *fi;
5478 struct btrfs_key key;
5486 path = btrfs_alloc_path();
5490 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5495 slot = path->slots[0];
5498 /* can't find the item, must cow */
5505 leaf = path->nodes[0];
5506 btrfs_item_key_to_cpu(leaf, &key, slot);
5507 if (key.objectid != btrfs_ino(inode) ||
5508 key.type != BTRFS_EXTENT_DATA_KEY) {
5509 /* not our file or wrong item type, must cow */
5513 if (key.offset > offset) {
5514 /* Wrong offset, must cow */
5518 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5519 found_type = btrfs_file_extent_type(leaf, fi);
5520 if (found_type != BTRFS_FILE_EXTENT_REG &&
5521 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5522 /* not a regular extent, must cow */
5525 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5526 backref_offset = btrfs_file_extent_offset(leaf, fi);
5528 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5529 if (extent_end < offset + len) {
5530 /* extent doesn't include our full range, must cow */
5534 if (btrfs_extent_readonly(root, disk_bytenr))
5538 * look for other files referencing this extent, if we
5539 * find any we must cow
5541 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5542 key.offset - backref_offset, disk_bytenr))
5546 * adjust disk_bytenr and num_bytes to cover just the bytes
5547 * in this extent we are about to write. If there
5548 * are any csums in that range we have to cow in order
5549 * to keep the csums correct
5551 disk_bytenr += backref_offset;
5552 disk_bytenr += offset - key.offset;
5553 num_bytes = min(offset + len, extent_end) - offset;
5554 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5557 * all of the above have passed, it is safe to overwrite this extent
5562 btrfs_free_path(path);
5566 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5567 struct buffer_head *bh_result, int create)
5569 struct extent_map *em;
5570 struct btrfs_root *root = BTRFS_I(inode)->root;
5571 u64 start = iblock << inode->i_blkbits;
5572 u64 len = bh_result->b_size;
5573 struct btrfs_trans_handle *trans;
5575 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5580 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5581 * io. INLINE is special, and we could probably kludge it in here, but
5582 * it's still buffered so for safety lets just fall back to the generic
5585 * For COMPRESSED we _have_ to read the entire extent in so we can
5586 * decompress it, so there will be buffering required no matter what we
5587 * do, so go ahead and fallback to buffered.
5589 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5590 * to buffered IO. Don't blame me, this is the price we pay for using
5593 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5594 em->block_start == EXTENT_MAP_INLINE) {
5595 free_extent_map(em);
5599 /* Just a good old fashioned hole, return */
5600 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5601 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5602 free_extent_map(em);
5603 /* DIO will do one hole at a time, so just unlock a sector */
5604 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5605 start + root->sectorsize - 1);
5610 * We don't allocate a new extent in the following cases
5612 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5614 * 2) The extent is marked as PREALLOC. We're good to go here and can
5615 * just use the extent.
5619 len = em->len - (start - em->start);
5623 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5624 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5625 em->block_start != EXTENT_MAP_HOLE)) {
5630 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5631 type = BTRFS_ORDERED_PREALLOC;
5633 type = BTRFS_ORDERED_NOCOW;
5634 len = min(len, em->len - (start - em->start));
5635 block_start = em->block_start + (start - em->start);
5638 * we're not going to log anything, but we do need
5639 * to make sure the current transaction stays open
5640 * while we look for nocow cross refs
5642 trans = btrfs_join_transaction(root);
5646 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5647 ret = btrfs_add_ordered_extent_dio(inode, start,
5648 block_start, len, len, type);
5649 btrfs_end_transaction(trans, root);
5651 free_extent_map(em);
5656 btrfs_end_transaction(trans, root);
5660 * this will cow the extent, reset the len in case we changed
5663 len = bh_result->b_size;
5664 em = btrfs_new_extent_direct(inode, em, start, len);
5667 len = min(len, em->len - (start - em->start));
5669 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5670 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5673 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5675 bh_result->b_size = len;
5676 bh_result->b_bdev = em->bdev;
5677 set_buffer_mapped(bh_result);
5678 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5679 set_buffer_new(bh_result);
5681 free_extent_map(em);
5686 struct btrfs_dio_private {
5687 struct inode *inode;
5694 /* number of bios pending for this dio */
5695 atomic_t pending_bios;
5700 struct bio *orig_bio;
5703 static void btrfs_endio_direct_read(struct bio *bio, int err)
5705 struct btrfs_dio_private *dip = bio->bi_private;
5706 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5707 struct bio_vec *bvec = bio->bi_io_vec;
5708 struct inode *inode = dip->inode;
5709 struct btrfs_root *root = BTRFS_I(inode)->root;
5711 u32 *private = dip->csums;
5713 start = dip->logical_offset;
5715 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5716 struct page *page = bvec->bv_page;
5719 unsigned long flags;
5721 local_irq_save(flags);
5722 kaddr = kmap_atomic(page, KM_IRQ0);
5723 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5724 csum, bvec->bv_len);
5725 btrfs_csum_final(csum, (char *)&csum);
5726 kunmap_atomic(kaddr, KM_IRQ0);
5727 local_irq_restore(flags);
5729 flush_dcache_page(bvec->bv_page);
5730 if (csum != *private) {
5731 printk(KERN_ERR "btrfs csum failed ino %llu off"
5732 " %llu csum %u private %u\n",
5733 (unsigned long long)btrfs_ino(inode),
5734 (unsigned long long)start,
5740 start += bvec->bv_len;
5743 } while (bvec <= bvec_end);
5745 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5746 dip->logical_offset + dip->bytes - 1);
5747 bio->bi_private = dip->private;
5752 /* If we had a csum failure make sure to clear the uptodate flag */
5754 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5755 dio_end_io(bio, err);
5758 static void btrfs_endio_direct_write(struct bio *bio, int err)
5760 struct btrfs_dio_private *dip = bio->bi_private;
5761 struct inode *inode = dip->inode;
5762 struct btrfs_root *root = BTRFS_I(inode)->root;
5763 struct btrfs_trans_handle *trans;
5764 struct btrfs_ordered_extent *ordered = NULL;
5765 struct extent_state *cached_state = NULL;
5766 u64 ordered_offset = dip->logical_offset;
5767 u64 ordered_bytes = dip->bytes;
5773 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5781 trans = btrfs_join_transaction(root);
5782 if (IS_ERR(trans)) {
5786 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5788 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5789 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5791 err = btrfs_update_inode_fallback(trans, root, inode);
5795 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5796 ordered->file_offset + ordered->len - 1, 0,
5799 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5800 ret = btrfs_mark_extent_written(trans, inode,
5801 ordered->file_offset,
5802 ordered->file_offset +
5809 ret = insert_reserved_file_extent(trans, inode,
5810 ordered->file_offset,
5816 BTRFS_FILE_EXTENT_REG);
5817 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5818 ordered->file_offset, ordered->len);
5826 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5827 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5828 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5829 btrfs_update_inode_fallback(trans, root, inode);
5832 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5833 ordered->file_offset + ordered->len - 1,
5834 &cached_state, GFP_NOFS);
5836 btrfs_delalloc_release_metadata(inode, ordered->len);
5837 btrfs_end_transaction(trans, root);
5838 ordered_offset = ordered->file_offset + ordered->len;
5839 btrfs_put_ordered_extent(ordered);
5840 btrfs_put_ordered_extent(ordered);
5844 * our bio might span multiple ordered extents. If we haven't
5845 * completed the accounting for the whole dio, go back and try again
5847 if (ordered_offset < dip->logical_offset + dip->bytes) {
5848 ordered_bytes = dip->logical_offset + dip->bytes -
5853 bio->bi_private = dip->private;
5858 /* If we had an error make sure to clear the uptodate flag */
5860 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5861 dio_end_io(bio, err);
5864 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5865 struct bio *bio, int mirror_num,
5866 unsigned long bio_flags, u64 offset)
5869 struct btrfs_root *root = BTRFS_I(inode)->root;
5870 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5875 static void btrfs_end_dio_bio(struct bio *bio, int err)
5877 struct btrfs_dio_private *dip = bio->bi_private;
5880 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5881 "sector %#Lx len %u err no %d\n",
5882 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5883 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5887 * before atomic variable goto zero, we must make sure
5888 * dip->errors is perceived to be set.
5890 smp_mb__before_atomic_dec();
5893 /* if there are more bios still pending for this dio, just exit */
5894 if (!atomic_dec_and_test(&dip->pending_bios))
5898 bio_io_error(dip->orig_bio);
5900 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5901 bio_endio(dip->orig_bio, 0);
5907 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5908 u64 first_sector, gfp_t gfp_flags)
5910 int nr_vecs = bio_get_nr_vecs(bdev);
5911 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5914 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5915 int rw, u64 file_offset, int skip_sum,
5916 u32 *csums, int async_submit)
5918 int write = rw & REQ_WRITE;
5919 struct btrfs_root *root = BTRFS_I(inode)->root;
5923 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5930 if (write && async_submit) {
5931 ret = btrfs_wq_submit_bio(root->fs_info,
5932 inode, rw, bio, 0, 0,
5934 __btrfs_submit_bio_start_direct_io,
5935 __btrfs_submit_bio_done);
5939 * If we aren't doing async submit, calculate the csum of the
5942 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5945 } else if (!skip_sum) {
5946 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5947 file_offset, csums);
5953 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5959 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5962 struct inode *inode = dip->inode;
5963 struct btrfs_root *root = BTRFS_I(inode)->root;
5964 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5966 struct bio *orig_bio = dip->orig_bio;
5967 struct bio_vec *bvec = orig_bio->bi_io_vec;
5968 u64 start_sector = orig_bio->bi_sector;
5969 u64 file_offset = dip->logical_offset;
5973 u32 *csums = dip->csums;
5975 int async_submit = 0;
5976 int write = rw & REQ_WRITE;
5978 map_length = orig_bio->bi_size;
5979 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5980 &map_length, NULL, 0);
5986 if (map_length >= orig_bio->bi_size) {
5992 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5995 bio->bi_private = dip;
5996 bio->bi_end_io = btrfs_end_dio_bio;
5997 atomic_inc(&dip->pending_bios);
5999 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6000 if (unlikely(map_length < submit_len + bvec->bv_len ||
6001 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6002 bvec->bv_offset) < bvec->bv_len)) {
6004 * inc the count before we submit the bio so
6005 * we know the end IO handler won't happen before
6006 * we inc the count. Otherwise, the dip might get freed
6007 * before we're done setting it up
6009 atomic_inc(&dip->pending_bios);
6010 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6011 file_offset, skip_sum,
6012 csums, async_submit);
6015 atomic_dec(&dip->pending_bios);
6019 /* Write's use the ordered csums */
6020 if (!write && !skip_sum)
6021 csums = csums + nr_pages;
6022 start_sector += submit_len >> 9;
6023 file_offset += submit_len;
6028 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6029 start_sector, GFP_NOFS);
6032 bio->bi_private = dip;
6033 bio->bi_end_io = btrfs_end_dio_bio;
6035 map_length = orig_bio->bi_size;
6036 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6037 &map_length, NULL, 0);
6043 submit_len += bvec->bv_len;
6050 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6051 csums, async_submit);
6059 * before atomic variable goto zero, we must
6060 * make sure dip->errors is perceived to be set.
6062 smp_mb__before_atomic_dec();
6063 if (atomic_dec_and_test(&dip->pending_bios))
6064 bio_io_error(dip->orig_bio);
6066 /* bio_end_io() will handle error, so we needn't return it */
6070 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6073 struct btrfs_root *root = BTRFS_I(inode)->root;
6074 struct btrfs_dio_private *dip;
6075 struct bio_vec *bvec = bio->bi_io_vec;
6077 int write = rw & REQ_WRITE;
6080 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6082 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6089 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6090 if (!write && !skip_sum) {
6091 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6099 dip->private = bio->bi_private;
6101 dip->logical_offset = file_offset;
6105 dip->bytes += bvec->bv_len;
6107 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6109 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6110 bio->bi_private = dip;
6112 dip->orig_bio = bio;
6113 atomic_set(&dip->pending_bios, 0);
6116 bio->bi_end_io = btrfs_endio_direct_write;
6118 bio->bi_end_io = btrfs_endio_direct_read;
6120 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6125 * If this is a write, we need to clean up the reserved space and kill
6126 * the ordered extent.
6129 struct btrfs_ordered_extent *ordered;
6130 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6131 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6132 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6133 btrfs_free_reserved_extent(root, ordered->start,
6135 btrfs_put_ordered_extent(ordered);
6136 btrfs_put_ordered_extent(ordered);
6138 bio_endio(bio, ret);
6141 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6142 const struct iovec *iov, loff_t offset,
6143 unsigned long nr_segs)
6149 unsigned blocksize_mask = root->sectorsize - 1;
6150 ssize_t retval = -EINVAL;
6151 loff_t end = offset;
6153 if (offset & blocksize_mask)
6156 /* Check the memory alignment. Blocks cannot straddle pages */
6157 for (seg = 0; seg < nr_segs; seg++) {
6158 addr = (unsigned long)iov[seg].iov_base;
6159 size = iov[seg].iov_len;
6161 if ((addr & blocksize_mask) || (size & blocksize_mask))
6164 /* If this is a write we don't need to check anymore */
6169 * Check to make sure we don't have duplicate iov_base's in this
6170 * iovec, if so return EINVAL, otherwise we'll get csum errors
6171 * when reading back.
6173 for (i = seg + 1; i < nr_segs; i++) {
6174 if (iov[seg].iov_base == iov[i].iov_base)
6182 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6183 const struct iovec *iov, loff_t offset,
6184 unsigned long nr_segs)
6186 struct file *file = iocb->ki_filp;
6187 struct inode *inode = file->f_mapping->host;
6188 struct btrfs_ordered_extent *ordered;
6189 struct extent_state *cached_state = NULL;
6190 u64 lockstart, lockend;
6192 int writing = rw & WRITE;
6194 size_t count = iov_length(iov, nr_segs);
6196 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6202 lockend = offset + count - 1;
6205 ret = btrfs_delalloc_reserve_space(inode, count);
6211 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6214 * We're concerned with the entire range that we're going to be
6215 * doing DIO to, so we need to make sure theres no ordered
6216 * extents in this range.
6218 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6219 lockend - lockstart + 1);
6222 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6223 &cached_state, GFP_NOFS);
6224 btrfs_start_ordered_extent(inode, ordered, 1);
6225 btrfs_put_ordered_extent(ordered);
6230 * we don't use btrfs_set_extent_delalloc because we don't want
6231 * the dirty or uptodate bits
6234 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6235 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6236 EXTENT_DELALLOC, NULL, &cached_state,
6239 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6240 lockend, EXTENT_LOCKED | write_bits,
6241 1, 0, &cached_state, GFP_NOFS);
6246 free_extent_state(cached_state);
6247 cached_state = NULL;
6249 ret = __blockdev_direct_IO(rw, iocb, inode,
6250 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6251 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6252 btrfs_submit_direct, 0);
6254 if (ret < 0 && ret != -EIOCBQUEUED) {
6255 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6256 offset + iov_length(iov, nr_segs) - 1,
6257 EXTENT_LOCKED | write_bits, 1, 0,
6258 &cached_state, GFP_NOFS);
6259 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6261 * We're falling back to buffered, unlock the section we didn't
6264 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6265 offset + iov_length(iov, nr_segs) - 1,
6266 EXTENT_LOCKED | write_bits, 1, 0,
6267 &cached_state, GFP_NOFS);
6270 free_extent_state(cached_state);
6274 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6275 __u64 start, __u64 len)
6277 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6280 int btrfs_readpage(struct file *file, struct page *page)
6282 struct extent_io_tree *tree;
6283 tree = &BTRFS_I(page->mapping->host)->io_tree;
6284 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6287 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6289 struct extent_io_tree *tree;
6292 if (current->flags & PF_MEMALLOC) {
6293 redirty_page_for_writepage(wbc, page);
6297 tree = &BTRFS_I(page->mapping->host)->io_tree;
6298 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6301 int btrfs_writepages(struct address_space *mapping,
6302 struct writeback_control *wbc)
6304 struct extent_io_tree *tree;
6306 tree = &BTRFS_I(mapping->host)->io_tree;
6307 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6311 btrfs_readpages(struct file *file, struct address_space *mapping,
6312 struct list_head *pages, unsigned nr_pages)
6314 struct extent_io_tree *tree;
6315 tree = &BTRFS_I(mapping->host)->io_tree;
6316 return extent_readpages(tree, mapping, pages, nr_pages,
6319 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6321 struct extent_io_tree *tree;
6322 struct extent_map_tree *map;
6325 tree = &BTRFS_I(page->mapping->host)->io_tree;
6326 map = &BTRFS_I(page->mapping->host)->extent_tree;
6327 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6329 ClearPagePrivate(page);
6330 set_page_private(page, 0);
6331 page_cache_release(page);
6336 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6338 if (PageWriteback(page) || PageDirty(page))
6340 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6343 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6345 struct extent_io_tree *tree;
6346 struct btrfs_ordered_extent *ordered;
6347 struct extent_state *cached_state = NULL;
6348 u64 page_start = page_offset(page);
6349 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6353 * we have the page locked, so new writeback can't start,
6354 * and the dirty bit won't be cleared while we are here.
6356 * Wait for IO on this page so that we can safely clear
6357 * the PagePrivate2 bit and do ordered accounting
6359 wait_on_page_writeback(page);
6361 tree = &BTRFS_I(page->mapping->host)->io_tree;
6363 btrfs_releasepage(page, GFP_NOFS);
6366 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6367 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6371 * IO on this page will never be started, so we need
6372 * to account for any ordered extents now
6374 clear_extent_bit(tree, page_start, page_end,
6375 EXTENT_DIRTY | EXTENT_DELALLOC |
6376 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6377 &cached_state, GFP_NOFS);
6379 * whoever cleared the private bit is responsible
6380 * for the finish_ordered_io
6382 if (TestClearPagePrivate2(page)) {
6383 btrfs_finish_ordered_io(page->mapping->host,
6384 page_start, page_end);
6386 btrfs_put_ordered_extent(ordered);
6387 cached_state = NULL;
6388 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6390 clear_extent_bit(tree, page_start, page_end,
6391 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6392 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6393 __btrfs_releasepage(page, GFP_NOFS);
6395 ClearPageChecked(page);
6396 if (PagePrivate(page)) {
6397 ClearPagePrivate(page);
6398 set_page_private(page, 0);
6399 page_cache_release(page);
6404 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6405 * called from a page fault handler when a page is first dirtied. Hence we must
6406 * be careful to check for EOF conditions here. We set the page up correctly
6407 * for a written page which means we get ENOSPC checking when writing into
6408 * holes and correct delalloc and unwritten extent mapping on filesystems that
6409 * support these features.
6411 * We are not allowed to take the i_mutex here so we have to play games to
6412 * protect against truncate races as the page could now be beyond EOF. Because
6413 * vmtruncate() writes the inode size before removing pages, once we have the
6414 * page lock we can determine safely if the page is beyond EOF. If it is not
6415 * beyond EOF, then the page is guaranteed safe against truncation until we
6418 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6420 struct page *page = vmf->page;
6421 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6422 struct btrfs_root *root = BTRFS_I(inode)->root;
6423 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6424 struct btrfs_ordered_extent *ordered;
6425 struct extent_state *cached_state = NULL;
6427 unsigned long zero_start;
6434 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6436 ret = btrfs_update_time(vma->vm_file);
6442 else /* -ENOSPC, -EIO, etc */
6443 ret = VM_FAULT_SIGBUS;
6449 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6452 size = i_size_read(inode);
6453 page_start = page_offset(page);
6454 page_end = page_start + PAGE_CACHE_SIZE - 1;
6456 if ((page->mapping != inode->i_mapping) ||
6457 (page_start >= size)) {
6458 /* page got truncated out from underneath us */
6461 wait_on_page_writeback(page);
6463 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6464 set_page_extent_mapped(page);
6467 * we can't set the delalloc bits if there are pending ordered
6468 * extents. Drop our locks and wait for them to finish
6470 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6472 unlock_extent_cached(io_tree, page_start, page_end,
6473 &cached_state, GFP_NOFS);
6475 btrfs_start_ordered_extent(inode, ordered, 1);
6476 btrfs_put_ordered_extent(ordered);
6481 * XXX - page_mkwrite gets called every time the page is dirtied, even
6482 * if it was already dirty, so for space accounting reasons we need to
6483 * clear any delalloc bits for the range we are fixing to save. There
6484 * is probably a better way to do this, but for now keep consistent with
6485 * prepare_pages in the normal write path.
6487 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6488 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6489 0, 0, &cached_state, GFP_NOFS);
6491 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6494 unlock_extent_cached(io_tree, page_start, page_end,
6495 &cached_state, GFP_NOFS);
6496 ret = VM_FAULT_SIGBUS;
6501 /* page is wholly or partially inside EOF */
6502 if (page_start + PAGE_CACHE_SIZE > size)
6503 zero_start = size & ~PAGE_CACHE_MASK;
6505 zero_start = PAGE_CACHE_SIZE;
6507 if (zero_start != PAGE_CACHE_SIZE) {
6509 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6510 flush_dcache_page(page);
6513 ClearPageChecked(page);
6514 set_page_dirty(page);
6515 SetPageUptodate(page);
6517 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6518 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6520 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6524 return VM_FAULT_LOCKED;
6527 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6532 static int btrfs_truncate(struct inode *inode)
6534 struct btrfs_root *root = BTRFS_I(inode)->root;
6535 struct btrfs_block_rsv *rsv;
6538 struct btrfs_trans_handle *trans;
6540 u64 mask = root->sectorsize - 1;
6541 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6543 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6547 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6548 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6551 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6552 * 3 things going on here
6554 * 1) We need to reserve space for our orphan item and the space to
6555 * delete our orphan item. Lord knows we don't want to have a dangling
6556 * orphan item because we didn't reserve space to remove it.
6558 * 2) We need to reserve space to update our inode.
6560 * 3) We need to have something to cache all the space that is going to
6561 * be free'd up by the truncate operation, but also have some slack
6562 * space reserved in case it uses space during the truncate (thank you
6563 * very much snapshotting).
6565 * And we need these to all be seperate. The fact is we can use alot of
6566 * space doing the truncate, and we have no earthly idea how much space
6567 * we will use, so we need the truncate reservation to be seperate so it
6568 * doesn't end up using space reserved for updating the inode or
6569 * removing the orphan item. We also need to be able to stop the
6570 * transaction and start a new one, which means we need to be able to
6571 * update the inode several times, and we have no idea of knowing how
6572 * many times that will be, so we can't just reserve 1 item for the
6573 * entirety of the opration, so that has to be done seperately as well.
6574 * Then there is the orphan item, which does indeed need to be held on
6575 * to for the whole operation, and we need nobody to touch this reserved
6576 * space except the orphan code.
6578 * So that leaves us with
6580 * 1) root->orphan_block_rsv - for the orphan deletion.
6581 * 2) rsv - for the truncate reservation, which we will steal from the
6582 * transaction reservation.
6583 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6584 * updating the inode.
6586 rsv = btrfs_alloc_block_rsv(root);
6589 rsv->size = min_size;
6592 * 1 for the truncate slack space
6593 * 1 for the orphan item we're going to add
6594 * 1 for the orphan item deletion
6595 * 1 for updating the inode.
6597 trans = btrfs_start_transaction(root, 4);
6598 if (IS_ERR(trans)) {
6599 err = PTR_ERR(trans);
6603 /* Migrate the slack space for the truncate to our reserve */
6604 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6608 ret = btrfs_orphan_add(trans, inode);
6610 btrfs_end_transaction(trans, root);
6615 * setattr is responsible for setting the ordered_data_close flag,
6616 * but that is only tested during the last file release. That
6617 * could happen well after the next commit, leaving a great big
6618 * window where new writes may get lost if someone chooses to write
6619 * to this file after truncating to zero
6621 * The inode doesn't have any dirty data here, and so if we commit
6622 * this is a noop. If someone immediately starts writing to the inode
6623 * it is very likely we'll catch some of their writes in this
6624 * transaction, and the commit will find this file on the ordered
6625 * data list with good things to send down.
6627 * This is a best effort solution, there is still a window where
6628 * using truncate to replace the contents of the file will
6629 * end up with a zero length file after a crash.
6631 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6632 btrfs_add_ordered_operation(trans, root, inode);
6635 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6638 * This can only happen with the original transaction we
6639 * started above, every other time we shouldn't have a
6640 * transaction started yet.
6649 /* Just need the 1 for updating the inode */
6650 trans = btrfs_start_transaction(root, 1);
6651 if (IS_ERR(trans)) {
6652 ret = err = PTR_ERR(trans);
6658 trans->block_rsv = rsv;
6660 ret = btrfs_truncate_inode_items(trans, root, inode,
6662 BTRFS_EXTENT_DATA_KEY);
6663 if (ret != -EAGAIN) {
6668 trans->block_rsv = &root->fs_info->trans_block_rsv;
6669 ret = btrfs_update_inode(trans, root, inode);
6675 nr = trans->blocks_used;
6676 btrfs_end_transaction(trans, root);
6678 btrfs_btree_balance_dirty(root, nr);
6681 if (ret == 0 && inode->i_nlink > 0) {
6682 trans->block_rsv = root->orphan_block_rsv;
6683 ret = btrfs_orphan_del(trans, inode);
6686 } else if (ret && inode->i_nlink > 0) {
6688 * Failed to do the truncate, remove us from the in memory
6691 ret = btrfs_orphan_del(NULL, inode);
6695 trans->block_rsv = &root->fs_info->trans_block_rsv;
6696 ret = btrfs_update_inode(trans, root, inode);
6700 nr = trans->blocks_used;
6701 ret = btrfs_end_transaction(trans, root);
6702 btrfs_btree_balance_dirty(root, nr);
6706 btrfs_free_block_rsv(root, rsv);
6715 * create a new subvolume directory/inode (helper for the ioctl).
6717 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6718 struct btrfs_root *new_root, u64 new_dirid)
6720 struct inode *inode;
6724 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6725 new_dirid, new_dirid,
6726 S_IFDIR | (~current_umask() & S_IRWXUGO),
6729 return PTR_ERR(inode);
6730 inode->i_op = &btrfs_dir_inode_operations;
6731 inode->i_fop = &btrfs_dir_file_operations;
6733 set_nlink(inode, 1);
6734 btrfs_i_size_write(inode, 0);
6736 err = btrfs_update_inode(trans, new_root, inode);
6743 struct inode *btrfs_alloc_inode(struct super_block *sb)
6745 struct btrfs_inode *ei;
6746 struct inode *inode;
6748 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6753 ei->space_info = NULL;
6757 ei->last_sub_trans = 0;
6758 ei->logged_trans = 0;
6759 ei->delalloc_bytes = 0;
6760 ei->disk_i_size = 0;
6763 ei->index_cnt = (u64)-1;
6764 ei->last_unlink_trans = 0;
6766 spin_lock_init(&ei->lock);
6767 ei->outstanding_extents = 0;
6768 ei->reserved_extents = 0;
6770 ei->ordered_data_close = 0;
6771 ei->orphan_meta_reserved = 0;
6772 ei->dummy_inode = 0;
6774 ei->delalloc_meta_reserved = 0;
6775 ei->force_compress = BTRFS_COMPRESS_NONE;
6777 ei->delayed_node = NULL;
6779 inode = &ei->vfs_inode;
6780 extent_map_tree_init(&ei->extent_tree);
6781 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6782 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6783 mutex_init(&ei->log_mutex);
6784 mutex_init(&ei->delalloc_mutex);
6785 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6786 INIT_LIST_HEAD(&ei->i_orphan);
6787 INIT_LIST_HEAD(&ei->delalloc_inodes);
6788 INIT_LIST_HEAD(&ei->ordered_operations);
6789 RB_CLEAR_NODE(&ei->rb_node);
6794 static void btrfs_i_callback(struct rcu_head *head)
6796 struct inode *inode = container_of(head, struct inode, i_rcu);
6797 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6800 void btrfs_destroy_inode(struct inode *inode)
6802 struct btrfs_ordered_extent *ordered;
6803 struct btrfs_root *root = BTRFS_I(inode)->root;
6805 WARN_ON(!list_empty(&inode->i_dentry));
6806 WARN_ON(inode->i_data.nrpages);
6807 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6808 WARN_ON(BTRFS_I(inode)->reserved_extents);
6809 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6810 WARN_ON(BTRFS_I(inode)->csum_bytes);
6813 * This can happen where we create an inode, but somebody else also
6814 * created the same inode and we need to destroy the one we already
6821 * Make sure we're properly removed from the ordered operation
6825 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6826 spin_lock(&root->fs_info->ordered_extent_lock);
6827 list_del_init(&BTRFS_I(inode)->ordered_operations);
6828 spin_unlock(&root->fs_info->ordered_extent_lock);
6831 spin_lock(&root->orphan_lock);
6832 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6833 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6834 (unsigned long long)btrfs_ino(inode));
6835 list_del_init(&BTRFS_I(inode)->i_orphan);
6837 spin_unlock(&root->orphan_lock);
6840 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6844 printk(KERN_ERR "btrfs found ordered "
6845 "extent %llu %llu on inode cleanup\n",
6846 (unsigned long long)ordered->file_offset,
6847 (unsigned long long)ordered->len);
6848 btrfs_remove_ordered_extent(inode, ordered);
6849 btrfs_put_ordered_extent(ordered);
6850 btrfs_put_ordered_extent(ordered);
6853 inode_tree_del(inode);
6854 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6856 btrfs_remove_delayed_node(inode);
6857 call_rcu(&inode->i_rcu, btrfs_i_callback);
6860 int btrfs_drop_inode(struct inode *inode)
6862 struct btrfs_root *root = BTRFS_I(inode)->root;
6864 if (btrfs_root_refs(&root->root_item) == 0 &&
6865 !btrfs_is_free_space_inode(root, inode))
6868 return generic_drop_inode(inode);
6871 static void init_once(void *foo)
6873 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6875 inode_init_once(&ei->vfs_inode);
6878 void btrfs_destroy_cachep(void)
6880 if (btrfs_inode_cachep)
6881 kmem_cache_destroy(btrfs_inode_cachep);
6882 if (btrfs_trans_handle_cachep)
6883 kmem_cache_destroy(btrfs_trans_handle_cachep);
6884 if (btrfs_transaction_cachep)
6885 kmem_cache_destroy(btrfs_transaction_cachep);
6886 if (btrfs_path_cachep)
6887 kmem_cache_destroy(btrfs_path_cachep);
6888 if (btrfs_free_space_cachep)
6889 kmem_cache_destroy(btrfs_free_space_cachep);
6892 int btrfs_init_cachep(void)
6894 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6895 sizeof(struct btrfs_inode), 0,
6896 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6897 if (!btrfs_inode_cachep)
6900 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6901 sizeof(struct btrfs_trans_handle), 0,
6902 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6903 if (!btrfs_trans_handle_cachep)
6906 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6907 sizeof(struct btrfs_transaction), 0,
6908 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6909 if (!btrfs_transaction_cachep)
6912 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6913 sizeof(struct btrfs_path), 0,
6914 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6915 if (!btrfs_path_cachep)
6918 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6919 sizeof(struct btrfs_free_space), 0,
6920 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6921 if (!btrfs_free_space_cachep)
6926 btrfs_destroy_cachep();
6930 static int btrfs_getattr(struct vfsmount *mnt,
6931 struct dentry *dentry, struct kstat *stat)
6933 struct inode *inode = dentry->d_inode;
6934 u32 blocksize = inode->i_sb->s_blocksize;
6936 generic_fillattr(inode, stat);
6937 stat->dev = BTRFS_I(inode)->root->anon_dev;
6938 stat->blksize = PAGE_CACHE_SIZE;
6939 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6940 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6945 * If a file is moved, it will inherit the cow and compression flags of the new
6948 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6950 struct btrfs_inode *b_dir = BTRFS_I(dir);
6951 struct btrfs_inode *b_inode = BTRFS_I(inode);
6953 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6954 b_inode->flags |= BTRFS_INODE_NODATACOW;
6956 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6958 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6959 b_inode->flags |= BTRFS_INODE_COMPRESS;
6961 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6964 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6965 struct inode *new_dir, struct dentry *new_dentry)
6967 struct btrfs_trans_handle *trans;
6968 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6969 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6970 struct inode *new_inode = new_dentry->d_inode;
6971 struct inode *old_inode = old_dentry->d_inode;
6972 struct timespec ctime = CURRENT_TIME;
6976 u64 old_ino = btrfs_ino(old_inode);
6978 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6981 /* we only allow rename subvolume link between subvolumes */
6982 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6985 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6986 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6989 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6990 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6993 * we're using rename to replace one file with another.
6994 * and the replacement file is large. Start IO on it now so
6995 * we don't add too much work to the end of the transaction
6997 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6998 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6999 filemap_flush(old_inode->i_mapping);
7001 /* close the racy window with snapshot create/destroy ioctl */
7002 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7003 down_read(&root->fs_info->subvol_sem);
7005 * We want to reserve the absolute worst case amount of items. So if
7006 * both inodes are subvols and we need to unlink them then that would
7007 * require 4 item modifications, but if they are both normal inodes it
7008 * would require 5 item modifications, so we'll assume their normal
7009 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7010 * should cover the worst case number of items we'll modify.
7012 trans = btrfs_start_transaction(root, 20);
7013 if (IS_ERR(trans)) {
7014 ret = PTR_ERR(trans);
7019 btrfs_record_root_in_trans(trans, dest);
7021 ret = btrfs_set_inode_index(new_dir, &index);
7025 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7026 /* force full log commit if subvolume involved. */
7027 root->fs_info->last_trans_log_full_commit = trans->transid;
7029 ret = btrfs_insert_inode_ref(trans, dest,
7030 new_dentry->d_name.name,
7031 new_dentry->d_name.len,
7033 btrfs_ino(new_dir), index);
7037 * this is an ugly little race, but the rename is required
7038 * to make sure that if we crash, the inode is either at the
7039 * old name or the new one. pinning the log transaction lets
7040 * us make sure we don't allow a log commit to come in after
7041 * we unlink the name but before we add the new name back in.
7043 btrfs_pin_log_trans(root);
7046 * make sure the inode gets flushed if it is replacing
7049 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7050 btrfs_add_ordered_operation(trans, root, old_inode);
7052 old_dir->i_ctime = old_dir->i_mtime = ctime;
7053 new_dir->i_ctime = new_dir->i_mtime = ctime;
7054 old_inode->i_ctime = ctime;
7056 if (old_dentry->d_parent != new_dentry->d_parent)
7057 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7059 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7060 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7061 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7062 old_dentry->d_name.name,
7063 old_dentry->d_name.len);
7065 ret = __btrfs_unlink_inode(trans, root, old_dir,
7066 old_dentry->d_inode,
7067 old_dentry->d_name.name,
7068 old_dentry->d_name.len);
7070 ret = btrfs_update_inode(trans, root, old_inode);
7075 new_inode->i_ctime = CURRENT_TIME;
7076 if (unlikely(btrfs_ino(new_inode) ==
7077 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7078 root_objectid = BTRFS_I(new_inode)->location.objectid;
7079 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7081 new_dentry->d_name.name,
7082 new_dentry->d_name.len);
7083 BUG_ON(new_inode->i_nlink == 0);
7085 ret = btrfs_unlink_inode(trans, dest, new_dir,
7086 new_dentry->d_inode,
7087 new_dentry->d_name.name,
7088 new_dentry->d_name.len);
7091 if (new_inode->i_nlink == 0) {
7092 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7097 fixup_inode_flags(new_dir, old_inode);
7099 ret = btrfs_add_link(trans, new_dir, old_inode,
7100 new_dentry->d_name.name,
7101 new_dentry->d_name.len, 0, index);
7104 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7105 struct dentry *parent = new_dentry->d_parent;
7106 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7107 btrfs_end_log_trans(root);
7110 btrfs_end_transaction(trans, root);
7112 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7113 up_read(&root->fs_info->subvol_sem);
7119 * some fairly slow code that needs optimization. This walks the list
7120 * of all the inodes with pending delalloc and forces them to disk.
7122 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7124 struct list_head *head = &root->fs_info->delalloc_inodes;
7125 struct btrfs_inode *binode;
7126 struct inode *inode;
7128 if (root->fs_info->sb->s_flags & MS_RDONLY)
7131 spin_lock(&root->fs_info->delalloc_lock);
7132 while (!list_empty(head)) {
7133 binode = list_entry(head->next, struct btrfs_inode,
7135 inode = igrab(&binode->vfs_inode);
7137 list_del_init(&binode->delalloc_inodes);
7138 spin_unlock(&root->fs_info->delalloc_lock);
7140 filemap_flush(inode->i_mapping);
7142 btrfs_add_delayed_iput(inode);
7147 spin_lock(&root->fs_info->delalloc_lock);
7149 spin_unlock(&root->fs_info->delalloc_lock);
7151 /* the filemap_flush will queue IO into the worker threads, but
7152 * we have to make sure the IO is actually started and that
7153 * ordered extents get created before we return
7155 atomic_inc(&root->fs_info->async_submit_draining);
7156 while (atomic_read(&root->fs_info->nr_async_submits) ||
7157 atomic_read(&root->fs_info->async_delalloc_pages)) {
7158 wait_event(root->fs_info->async_submit_wait,
7159 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7160 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7162 atomic_dec(&root->fs_info->async_submit_draining);
7166 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7167 const char *symname)
7169 struct btrfs_trans_handle *trans;
7170 struct btrfs_root *root = BTRFS_I(dir)->root;
7171 struct btrfs_path *path;
7172 struct btrfs_key key;
7173 struct inode *inode = NULL;
7181 struct btrfs_file_extent_item *ei;
7182 struct extent_buffer *leaf;
7183 unsigned long nr = 0;
7185 name_len = strlen(symname) + 1;
7186 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7187 return -ENAMETOOLONG;
7190 * 2 items for inode item and ref
7191 * 2 items for dir items
7192 * 1 item for xattr if selinux is on
7194 trans = btrfs_start_transaction(root, 5);
7196 return PTR_ERR(trans);
7198 err = btrfs_find_free_ino(root, &objectid);
7202 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7203 dentry->d_name.len, btrfs_ino(dir), objectid,
7204 S_IFLNK|S_IRWXUGO, &index);
7205 if (IS_ERR(inode)) {
7206 err = PTR_ERR(inode);
7210 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7217 * If the active LSM wants to access the inode during
7218 * d_instantiate it needs these. Smack checks to see
7219 * if the filesystem supports xattrs by looking at the
7222 inode->i_fop = &btrfs_file_operations;
7223 inode->i_op = &btrfs_file_inode_operations;
7225 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7229 inode->i_mapping->a_ops = &btrfs_aops;
7230 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7231 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7236 path = btrfs_alloc_path();
7242 key.objectid = btrfs_ino(inode);
7244 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7245 datasize = btrfs_file_extent_calc_inline_size(name_len);
7246 err = btrfs_insert_empty_item(trans, root, path, &key,
7250 btrfs_free_path(path);
7253 leaf = path->nodes[0];
7254 ei = btrfs_item_ptr(leaf, path->slots[0],
7255 struct btrfs_file_extent_item);
7256 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7257 btrfs_set_file_extent_type(leaf, ei,
7258 BTRFS_FILE_EXTENT_INLINE);
7259 btrfs_set_file_extent_encryption(leaf, ei, 0);
7260 btrfs_set_file_extent_compression(leaf, ei, 0);
7261 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7262 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7264 ptr = btrfs_file_extent_inline_start(ei);
7265 write_extent_buffer(leaf, symname, ptr, name_len);
7266 btrfs_mark_buffer_dirty(leaf);
7267 btrfs_free_path(path);
7269 inode->i_op = &btrfs_symlink_inode_operations;
7270 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7271 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7272 inode_set_bytes(inode, name_len);
7273 btrfs_i_size_write(inode, name_len - 1);
7274 err = btrfs_update_inode(trans, root, inode);
7280 d_instantiate(dentry, inode);
7281 nr = trans->blocks_used;
7282 btrfs_end_transaction(trans, root);
7284 inode_dec_link_count(inode);
7287 btrfs_btree_balance_dirty(root, nr);
7291 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7292 u64 start, u64 num_bytes, u64 min_size,
7293 loff_t actual_len, u64 *alloc_hint,
7294 struct btrfs_trans_handle *trans)
7296 struct btrfs_root *root = BTRFS_I(inode)->root;
7297 struct btrfs_key ins;
7298 u64 cur_offset = start;
7301 bool own_trans = true;
7305 while (num_bytes > 0) {
7307 trans = btrfs_start_transaction(root, 3);
7308 if (IS_ERR(trans)) {
7309 ret = PTR_ERR(trans);
7314 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7315 0, *alloc_hint, (u64)-1, &ins, 1);
7318 btrfs_end_transaction(trans, root);
7322 ret = insert_reserved_file_extent(trans, inode,
7323 cur_offset, ins.objectid,
7324 ins.offset, ins.offset,
7325 ins.offset, 0, 0, 0,
7326 BTRFS_FILE_EXTENT_PREALLOC);
7328 btrfs_drop_extent_cache(inode, cur_offset,
7329 cur_offset + ins.offset -1, 0);
7331 num_bytes -= ins.offset;
7332 cur_offset += ins.offset;
7333 *alloc_hint = ins.objectid + ins.offset;
7335 inode->i_ctime = CURRENT_TIME;
7336 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7337 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7338 (actual_len > inode->i_size) &&
7339 (cur_offset > inode->i_size)) {
7340 if (cur_offset > actual_len)
7341 i_size = actual_len;
7343 i_size = cur_offset;
7344 i_size_write(inode, i_size);
7345 btrfs_ordered_update_i_size(inode, i_size, NULL);
7348 ret = btrfs_update_inode(trans, root, inode);
7352 btrfs_end_transaction(trans, root);
7357 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7358 u64 start, u64 num_bytes, u64 min_size,
7359 loff_t actual_len, u64 *alloc_hint)
7361 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7362 min_size, actual_len, alloc_hint,
7366 int btrfs_prealloc_file_range_trans(struct inode *inode,
7367 struct btrfs_trans_handle *trans, int mode,
7368 u64 start, u64 num_bytes, u64 min_size,
7369 loff_t actual_len, u64 *alloc_hint)
7371 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7372 min_size, actual_len, alloc_hint, trans);
7375 static int btrfs_set_page_dirty(struct page *page)
7377 return __set_page_dirty_nobuffers(page);
7380 static int btrfs_permission(struct inode *inode, int mask)
7382 struct btrfs_root *root = BTRFS_I(inode)->root;
7383 umode_t mode = inode->i_mode;
7385 if (mask & MAY_WRITE &&
7386 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7387 if (btrfs_root_readonly(root))
7389 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7392 return generic_permission(inode, mask);
7395 static const struct inode_operations btrfs_dir_inode_operations = {
7396 .getattr = btrfs_getattr,
7397 .lookup = btrfs_lookup,
7398 .create = btrfs_create,
7399 .unlink = btrfs_unlink,
7401 .mkdir = btrfs_mkdir,
7402 .rmdir = btrfs_rmdir,
7403 .rename = btrfs_rename,
7404 .symlink = btrfs_symlink,
7405 .setattr = btrfs_setattr,
7406 .mknod = btrfs_mknod,
7407 .setxattr = btrfs_setxattr,
7408 .getxattr = btrfs_getxattr,
7409 .listxattr = btrfs_listxattr,
7410 .removexattr = btrfs_removexattr,
7411 .permission = btrfs_permission,
7412 .get_acl = btrfs_get_acl,
7414 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7415 .lookup = btrfs_lookup,
7416 .permission = btrfs_permission,
7417 .get_acl = btrfs_get_acl,
7420 static const struct file_operations btrfs_dir_file_operations = {
7421 .llseek = generic_file_llseek,
7422 .read = generic_read_dir,
7423 .readdir = btrfs_real_readdir,
7424 .unlocked_ioctl = btrfs_ioctl,
7425 #ifdef CONFIG_COMPAT
7426 .compat_ioctl = btrfs_ioctl,
7428 .release = btrfs_release_file,
7429 .fsync = btrfs_sync_file,
7432 static struct extent_io_ops btrfs_extent_io_ops = {
7433 .fill_delalloc = run_delalloc_range,
7434 .submit_bio_hook = btrfs_submit_bio_hook,
7435 .merge_bio_hook = btrfs_merge_bio_hook,
7436 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7437 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7438 .writepage_start_hook = btrfs_writepage_start_hook,
7439 .set_bit_hook = btrfs_set_bit_hook,
7440 .clear_bit_hook = btrfs_clear_bit_hook,
7441 .merge_extent_hook = btrfs_merge_extent_hook,
7442 .split_extent_hook = btrfs_split_extent_hook,
7446 * btrfs doesn't support the bmap operation because swapfiles
7447 * use bmap to make a mapping of extents in the file. They assume
7448 * these extents won't change over the life of the file and they
7449 * use the bmap result to do IO directly to the drive.
7451 * the btrfs bmap call would return logical addresses that aren't
7452 * suitable for IO and they also will change frequently as COW
7453 * operations happen. So, swapfile + btrfs == corruption.
7455 * For now we're avoiding this by dropping bmap.
7457 static const struct address_space_operations btrfs_aops = {
7458 .readpage = btrfs_readpage,
7459 .writepage = btrfs_writepage,
7460 .writepages = btrfs_writepages,
7461 .readpages = btrfs_readpages,
7462 .direct_IO = btrfs_direct_IO,
7463 .invalidatepage = btrfs_invalidatepage,
7464 .releasepage = btrfs_releasepage,
7465 .set_page_dirty = btrfs_set_page_dirty,
7466 .error_remove_page = generic_error_remove_page,
7469 static const struct address_space_operations btrfs_symlink_aops = {
7470 .readpage = btrfs_readpage,
7471 .writepage = btrfs_writepage,
7472 .invalidatepage = btrfs_invalidatepage,
7473 .releasepage = btrfs_releasepage,
7476 static const struct inode_operations btrfs_file_inode_operations = {
7477 .getattr = btrfs_getattr,
7478 .setattr = btrfs_setattr,
7479 .setxattr = btrfs_setxattr,
7480 .getxattr = btrfs_getxattr,
7481 .listxattr = btrfs_listxattr,
7482 .removexattr = btrfs_removexattr,
7483 .permission = btrfs_permission,
7484 .fiemap = btrfs_fiemap,
7485 .get_acl = btrfs_get_acl,
7487 static const struct inode_operations btrfs_special_inode_operations = {
7488 .getattr = btrfs_getattr,
7489 .setattr = btrfs_setattr,
7490 .permission = btrfs_permission,
7491 .setxattr = btrfs_setxattr,
7492 .getxattr = btrfs_getxattr,
7493 .listxattr = btrfs_listxattr,
7494 .removexattr = btrfs_removexattr,
7495 .get_acl = btrfs_get_acl,
7497 static const struct inode_operations btrfs_symlink_inode_operations = {
7498 .readlink = generic_readlink,
7499 .follow_link = page_follow_link_light,
7500 .put_link = page_put_link,
7501 .getattr = btrfs_getattr,
7502 .setattr = btrfs_setattr,
7503 .permission = btrfs_permission,
7504 .setxattr = btrfs_setxattr,
7505 .getxattr = btrfs_getxattr,
7506 .listxattr = btrfs_listxattr,
7507 .removexattr = btrfs_removexattr,
7508 .get_acl = btrfs_get_acl,
7511 const struct dentry_operations btrfs_dentry_operations = {
7512 .d_delete = btrfs_dentry_delete,
7513 .d_release = btrfs_dentry_release,