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>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir)
100 err = btrfs_init_acl(trans, inode, dir);
102 err = btrfs_xattr_security_init(trans, inode, dir);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root, struct inode *inode,
113 u64 start, size_t size, size_t compressed_size,
114 struct page **compressed_pages)
116 struct btrfs_key key;
117 struct btrfs_path *path;
118 struct extent_buffer *leaf;
119 struct page *page = NULL;
122 struct btrfs_file_extent_item *ei;
125 size_t cur_size = size;
127 unsigned long offset;
128 int compress_type = BTRFS_COMPRESS_NONE;
130 if (compressed_size && compressed_pages) {
131 compress_type = root->fs_info->compress_type;
132 cur_size = compressed_size;
135 path = btrfs_alloc_path();
139 path->leave_spinning = 1;
140 btrfs_set_trans_block_group(trans, inode);
142 key.objectid = inode->i_ino;
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
165 if (compress_type != BTRFS_COMPRESS_NONE) {
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
179 compressed_size -= cur_size;
181 btrfs_set_file_extent_compression(leaf, ei,
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
210 btrfs_free_path(path);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size,
224 struct page **compressed_pages)
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
232 u64 data_len = inline_len;
236 data_len = compressed_size;
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (actual_end & (root->sectorsize - 1)) == 0) ||
244 data_len > root->fs_info->max_inline) {
248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
263 struct async_extent {
268 unsigned long nr_pages;
270 struct list_head list;
275 struct btrfs_root *root;
276 struct page *locked_page;
279 struct list_head extents;
280 struct btrfs_work work;
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
287 unsigned long nr_pages,
290 struct async_extent *async_extent;
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 async_extent->start = start;
294 async_extent->ram_size = ram_size;
295 async_extent->compressed_size = compressed_size;
296 async_extent->pages = pages;
297 async_extent->nr_pages = nr_pages;
298 async_extent->compress_type = compress_type;
299 list_add_tail(&async_extent->list, &cow->extents);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline int compress_file_range(struct inode *inode,
320 struct page *locked_page,
322 struct async_cow *async_cow,
325 struct btrfs_root *root = BTRFS_I(inode)->root;
326 struct btrfs_trans_handle *trans;
328 u64 blocksize = root->sectorsize;
330 u64 isize = i_size_read(inode);
332 struct page **pages = NULL;
333 unsigned long nr_pages;
334 unsigned long nr_pages_ret = 0;
335 unsigned long total_compressed = 0;
336 unsigned long total_in = 0;
337 unsigned long max_compressed = 128 * 1024;
338 unsigned long max_uncompressed = 128 * 1024;
341 int compress_type = root->fs_info->compress_type;
343 actual_end = min_t(u64, isize, end + 1);
346 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
347 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
350 * we don't want to send crud past the end of i_size through
351 * compression, that's just a waste of CPU time. So, if the
352 * end of the file is before the start of our current
353 * requested range of bytes, we bail out to the uncompressed
354 * cleanup code that can deal with all of this.
356 * It isn't really the fastest way to fix things, but this is a
357 * very uncommon corner.
359 if (actual_end <= start)
360 goto cleanup_and_bail_uncompressed;
362 total_compressed = actual_end - start;
364 /* we want to make sure that amount of ram required to uncompress
365 * an extent is reasonable, so we limit the total size in ram
366 * of a compressed extent to 128k. This is a crucial number
367 * because it also controls how easily we can spread reads across
368 * cpus for decompression.
370 * We also want to make sure the amount of IO required to do
371 * a random read is reasonably small, so we limit the size of
372 * a compressed extent to 128k.
374 total_compressed = min(total_compressed, max_uncompressed);
375 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376 num_bytes = max(blocksize, num_bytes);
381 * we do compression for mount -o compress and when the
382 * inode has not been flagged as nocompress. This flag can
383 * change at any time if we discover bad compression ratios.
385 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
386 (btrfs_test_opt(root, COMPRESS) ||
387 (BTRFS_I(inode)->force_compress))) {
389 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
391 if (BTRFS_I(inode)->force_compress)
392 compress_type = BTRFS_I(inode)->force_compress;
394 ret = btrfs_compress_pages(compress_type,
395 inode->i_mapping, start,
396 total_compressed, pages,
397 nr_pages, &nr_pages_ret,
403 unsigned long offset = total_compressed &
404 (PAGE_CACHE_SIZE - 1);
405 struct page *page = pages[nr_pages_ret - 1];
408 /* zero the tail end of the last page, we might be
409 * sending it down to disk
412 kaddr = kmap_atomic(page, KM_USER0);
413 memset(kaddr + offset, 0,
414 PAGE_CACHE_SIZE - offset);
415 kunmap_atomic(kaddr, KM_USER0);
421 trans = btrfs_join_transaction(root, 1);
422 BUG_ON(IS_ERR(trans));
423 btrfs_set_trans_block_group(trans, inode);
424 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
426 /* lets try to make an inline extent */
427 if (ret || total_in < (actual_end - start)) {
428 /* we didn't compress the entire range, try
429 * to make an uncompressed inline extent.
431 ret = cow_file_range_inline(trans, root, inode,
432 start, end, 0, NULL);
434 /* try making a compressed inline extent */
435 ret = cow_file_range_inline(trans, root, inode,
437 total_compressed, pages);
441 * inline extent creation worked, we don't need
442 * to create any more async work items. Unlock
443 * and free up our temp pages.
445 extent_clear_unlock_delalloc(inode,
446 &BTRFS_I(inode)->io_tree,
448 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
449 EXTENT_CLEAR_DELALLOC |
450 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
452 btrfs_end_transaction(trans, root);
455 btrfs_end_transaction(trans, root);
460 * we aren't doing an inline extent round the compressed size
461 * up to a block size boundary so the allocator does sane
464 total_compressed = (total_compressed + blocksize - 1) &
468 * one last check to make sure the compression is really a
469 * win, compare the page count read with the blocks on disk
471 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
472 ~(PAGE_CACHE_SIZE - 1);
473 if (total_compressed >= total_in) {
476 num_bytes = total_in;
479 if (!will_compress && pages) {
481 * the compression code ran but failed to make things smaller,
482 * free any pages it allocated and our page pointer array
484 for (i = 0; i < nr_pages_ret; i++) {
485 WARN_ON(pages[i]->mapping);
486 page_cache_release(pages[i]);
490 total_compressed = 0;
493 /* flag the file so we don't compress in the future */
494 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
495 !(BTRFS_I(inode)->force_compress)) {
496 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
502 /* the async work queues will take care of doing actual
503 * allocation on disk for these compressed pages,
504 * and will submit them to the elevator.
506 add_async_extent(async_cow, start, num_bytes,
507 total_compressed, pages, nr_pages_ret,
510 if (start + num_bytes < end) {
517 cleanup_and_bail_uncompressed:
519 * No compression, but we still need to write the pages in
520 * the file we've been given so far. redirty the locked
521 * page if it corresponds to our extent and set things up
522 * for the async work queue to run cow_file_range to do
523 * the normal delalloc dance
525 if (page_offset(locked_page) >= start &&
526 page_offset(locked_page) <= end) {
527 __set_page_dirty_nobuffers(locked_page);
528 /* unlocked later on in the async handlers */
530 add_async_extent(async_cow, start, end - start + 1,
531 0, NULL, 0, BTRFS_COMPRESS_NONE);
539 for (i = 0; i < nr_pages_ret; i++) {
540 WARN_ON(pages[i]->mapping);
541 page_cache_release(pages[i]);
549 * phase two of compressed writeback. This is the ordered portion
550 * of the code, which only gets called in the order the work was
551 * queued. We walk all the async extents created by compress_file_range
552 * and send them down to the disk.
554 static noinline int submit_compressed_extents(struct inode *inode,
555 struct async_cow *async_cow)
557 struct async_extent *async_extent;
559 struct btrfs_trans_handle *trans;
560 struct btrfs_key ins;
561 struct extent_map *em;
562 struct btrfs_root *root = BTRFS_I(inode)->root;
563 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
564 struct extent_io_tree *io_tree;
567 if (list_empty(&async_cow->extents))
571 while (!list_empty(&async_cow->extents)) {
572 async_extent = list_entry(async_cow->extents.next,
573 struct async_extent, list);
574 list_del(&async_extent->list);
576 io_tree = &BTRFS_I(inode)->io_tree;
579 /* did the compression code fall back to uncompressed IO? */
580 if (!async_extent->pages) {
581 int page_started = 0;
582 unsigned long nr_written = 0;
584 lock_extent(io_tree, async_extent->start,
585 async_extent->start +
586 async_extent->ram_size - 1, GFP_NOFS);
588 /* allocate blocks */
589 ret = cow_file_range(inode, async_cow->locked_page,
591 async_extent->start +
592 async_extent->ram_size - 1,
593 &page_started, &nr_written, 0);
596 * if page_started, cow_file_range inserted an
597 * inline extent and took care of all the unlocking
598 * and IO for us. Otherwise, we need to submit
599 * all those pages down to the drive.
601 if (!page_started && !ret)
602 extent_write_locked_range(io_tree,
603 inode, async_extent->start,
604 async_extent->start +
605 async_extent->ram_size - 1,
613 lock_extent(io_tree, async_extent->start,
614 async_extent->start + async_extent->ram_size - 1,
617 trans = btrfs_join_transaction(root, 1);
618 BUG_ON(IS_ERR(trans));
619 ret = btrfs_reserve_extent(trans, root,
620 async_extent->compressed_size,
621 async_extent->compressed_size,
624 btrfs_end_transaction(trans, root);
628 for (i = 0; i < async_extent->nr_pages; i++) {
629 WARN_ON(async_extent->pages[i]->mapping);
630 page_cache_release(async_extent->pages[i]);
632 kfree(async_extent->pages);
633 async_extent->nr_pages = 0;
634 async_extent->pages = NULL;
635 unlock_extent(io_tree, async_extent->start,
636 async_extent->start +
637 async_extent->ram_size - 1, GFP_NOFS);
642 * here we're doing allocation and writeback of the
645 btrfs_drop_extent_cache(inode, async_extent->start,
646 async_extent->start +
647 async_extent->ram_size - 1, 0);
649 em = alloc_extent_map(GFP_NOFS);
651 em->start = async_extent->start;
652 em->len = async_extent->ram_size;
653 em->orig_start = em->start;
655 em->block_start = ins.objectid;
656 em->block_len = ins.offset;
657 em->bdev = root->fs_info->fs_devices->latest_bdev;
658 em->compress_type = async_extent->compress_type;
659 set_bit(EXTENT_FLAG_PINNED, &em->flags);
660 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
663 write_lock(&em_tree->lock);
664 ret = add_extent_mapping(em_tree, em);
665 write_unlock(&em_tree->lock);
666 if (ret != -EEXIST) {
670 btrfs_drop_extent_cache(inode, async_extent->start,
671 async_extent->start +
672 async_extent->ram_size - 1, 0);
675 ret = btrfs_add_ordered_extent_compress(inode,
678 async_extent->ram_size,
680 BTRFS_ORDERED_COMPRESSED,
681 async_extent->compress_type);
685 * clear dirty, set writeback and unlock the pages.
687 extent_clear_unlock_delalloc(inode,
688 &BTRFS_I(inode)->io_tree,
690 async_extent->start +
691 async_extent->ram_size - 1,
692 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
693 EXTENT_CLEAR_UNLOCK |
694 EXTENT_CLEAR_DELALLOC |
695 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
697 ret = btrfs_submit_compressed_write(inode,
699 async_extent->ram_size,
701 ins.offset, async_extent->pages,
702 async_extent->nr_pages);
705 alloc_hint = ins.objectid + ins.offset;
713 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
716 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
717 struct extent_map *em;
720 read_lock(&em_tree->lock);
721 em = search_extent_mapping(em_tree, start, num_bytes);
724 * if block start isn't an actual block number then find the
725 * first block in this inode and use that as a hint. If that
726 * block is also bogus then just don't worry about it.
728 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
730 em = search_extent_mapping(em_tree, 0, 0);
731 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
732 alloc_hint = em->block_start;
736 alloc_hint = em->block_start;
740 read_unlock(&em_tree->lock);
746 * when extent_io.c finds a delayed allocation range in the file,
747 * the call backs end up in this code. The basic idea is to
748 * allocate extents on disk for the range, and create ordered data structs
749 * in ram to track those extents.
751 * locked_page is the page that writepage had locked already. We use
752 * it to make sure we don't do extra locks or unlocks.
754 * *page_started is set to one if we unlock locked_page and do everything
755 * required to start IO on it. It may be clean and already done with
758 static noinline int cow_file_range(struct inode *inode,
759 struct page *locked_page,
760 u64 start, u64 end, int *page_started,
761 unsigned long *nr_written,
764 struct btrfs_root *root = BTRFS_I(inode)->root;
765 struct btrfs_trans_handle *trans;
768 unsigned long ram_size;
771 u64 blocksize = root->sectorsize;
772 struct btrfs_key ins;
773 struct extent_map *em;
774 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
777 BUG_ON(root == root->fs_info->tree_root);
778 trans = btrfs_join_transaction(root, 1);
779 BUG_ON(IS_ERR(trans));
780 btrfs_set_trans_block_group(trans, inode);
781 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
783 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
784 num_bytes = max(blocksize, num_bytes);
785 disk_num_bytes = num_bytes;
789 /* lets try to make an inline extent */
790 ret = cow_file_range_inline(trans, root, inode,
791 start, end, 0, NULL);
793 extent_clear_unlock_delalloc(inode,
794 &BTRFS_I(inode)->io_tree,
796 EXTENT_CLEAR_UNLOCK_PAGE |
797 EXTENT_CLEAR_UNLOCK |
798 EXTENT_CLEAR_DELALLOC |
800 EXTENT_SET_WRITEBACK |
801 EXTENT_END_WRITEBACK);
803 *nr_written = *nr_written +
804 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
811 BUG_ON(disk_num_bytes >
812 btrfs_super_total_bytes(&root->fs_info->super_copy));
814 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
815 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
817 while (disk_num_bytes > 0) {
820 cur_alloc_size = disk_num_bytes;
821 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
822 root->sectorsize, 0, alloc_hint,
826 em = alloc_extent_map(GFP_NOFS);
829 em->orig_start = em->start;
830 ram_size = ins.offset;
831 em->len = ins.offset;
833 em->block_start = ins.objectid;
834 em->block_len = ins.offset;
835 em->bdev = root->fs_info->fs_devices->latest_bdev;
836 set_bit(EXTENT_FLAG_PINNED, &em->flags);
839 write_lock(&em_tree->lock);
840 ret = add_extent_mapping(em_tree, em);
841 write_unlock(&em_tree->lock);
842 if (ret != -EEXIST) {
846 btrfs_drop_extent_cache(inode, start,
847 start + ram_size - 1, 0);
850 cur_alloc_size = ins.offset;
851 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
852 ram_size, cur_alloc_size, 0);
855 if (root->root_key.objectid ==
856 BTRFS_DATA_RELOC_TREE_OBJECTID) {
857 ret = btrfs_reloc_clone_csums(inode, start,
862 if (disk_num_bytes < cur_alloc_size)
865 /* we're not doing compressed IO, don't unlock the first
866 * page (which the caller expects to stay locked), don't
867 * clear any dirty bits and don't set any writeback bits
869 * Do set the Private2 bit so we know this page was properly
870 * setup for writepage
872 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
873 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
876 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
877 start, start + ram_size - 1,
879 disk_num_bytes -= cur_alloc_size;
880 num_bytes -= cur_alloc_size;
881 alloc_hint = ins.objectid + ins.offset;
882 start += cur_alloc_size;
886 btrfs_end_transaction(trans, root);
892 * work queue call back to started compression on a file and pages
894 static noinline void async_cow_start(struct btrfs_work *work)
896 struct async_cow *async_cow;
898 async_cow = container_of(work, struct async_cow, work);
900 compress_file_range(async_cow->inode, async_cow->locked_page,
901 async_cow->start, async_cow->end, async_cow,
904 async_cow->inode = NULL;
908 * work queue call back to submit previously compressed pages
910 static noinline void async_cow_submit(struct btrfs_work *work)
912 struct async_cow *async_cow;
913 struct btrfs_root *root;
914 unsigned long nr_pages;
916 async_cow = container_of(work, struct async_cow, work);
918 root = async_cow->root;
919 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
922 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
924 if (atomic_read(&root->fs_info->async_delalloc_pages) <
926 waitqueue_active(&root->fs_info->async_submit_wait))
927 wake_up(&root->fs_info->async_submit_wait);
929 if (async_cow->inode)
930 submit_compressed_extents(async_cow->inode, async_cow);
933 static noinline void async_cow_free(struct btrfs_work *work)
935 struct async_cow *async_cow;
936 async_cow = container_of(work, struct async_cow, work);
940 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
941 u64 start, u64 end, int *page_started,
942 unsigned long *nr_written)
944 struct async_cow *async_cow;
945 struct btrfs_root *root = BTRFS_I(inode)->root;
946 unsigned long nr_pages;
948 int limit = 10 * 1024 * 1042;
950 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
951 1, 0, NULL, GFP_NOFS);
952 while (start < end) {
953 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
954 async_cow->inode = inode;
955 async_cow->root = root;
956 async_cow->locked_page = locked_page;
957 async_cow->start = start;
959 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
962 cur_end = min(end, start + 512 * 1024 - 1);
964 async_cow->end = cur_end;
965 INIT_LIST_HEAD(&async_cow->extents);
967 async_cow->work.func = async_cow_start;
968 async_cow->work.ordered_func = async_cow_submit;
969 async_cow->work.ordered_free = async_cow_free;
970 async_cow->work.flags = 0;
972 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
974 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
976 btrfs_queue_worker(&root->fs_info->delalloc_workers,
979 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
980 wait_event(root->fs_info->async_submit_wait,
981 (atomic_read(&root->fs_info->async_delalloc_pages) <
985 while (atomic_read(&root->fs_info->async_submit_draining) &&
986 atomic_read(&root->fs_info->async_delalloc_pages)) {
987 wait_event(root->fs_info->async_submit_wait,
988 (atomic_read(&root->fs_info->async_delalloc_pages) ==
992 *nr_written += nr_pages;
999 static noinline int csum_exist_in_range(struct btrfs_root *root,
1000 u64 bytenr, u64 num_bytes)
1003 struct btrfs_ordered_sum *sums;
1006 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1007 bytenr + num_bytes - 1, &list);
1008 if (ret == 0 && list_empty(&list))
1011 while (!list_empty(&list)) {
1012 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1013 list_del(&sums->list);
1020 * when nowcow writeback call back. This checks for snapshots or COW copies
1021 * of the extents that exist in the file, and COWs the file as required.
1023 * If no cow copies or snapshots exist, we write directly to the existing
1026 static noinline int run_delalloc_nocow(struct inode *inode,
1027 struct page *locked_page,
1028 u64 start, u64 end, int *page_started, int force,
1029 unsigned long *nr_written)
1031 struct btrfs_root *root = BTRFS_I(inode)->root;
1032 struct btrfs_trans_handle *trans;
1033 struct extent_buffer *leaf;
1034 struct btrfs_path *path;
1035 struct btrfs_file_extent_item *fi;
1036 struct btrfs_key found_key;
1048 bool nolock = false;
1050 path = btrfs_alloc_path();
1052 if (root == root->fs_info->tree_root) {
1054 trans = btrfs_join_transaction_nolock(root, 1);
1056 trans = btrfs_join_transaction(root, 1);
1058 BUG_ON(IS_ERR(trans));
1060 cow_start = (u64)-1;
1063 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1066 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1067 leaf = path->nodes[0];
1068 btrfs_item_key_to_cpu(leaf, &found_key,
1069 path->slots[0] - 1);
1070 if (found_key.objectid == inode->i_ino &&
1071 found_key.type == BTRFS_EXTENT_DATA_KEY)
1076 leaf = path->nodes[0];
1077 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1078 ret = btrfs_next_leaf(root, path);
1083 leaf = path->nodes[0];
1089 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1091 if (found_key.objectid > inode->i_ino ||
1092 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1093 found_key.offset > end)
1096 if (found_key.offset > cur_offset) {
1097 extent_end = found_key.offset;
1102 fi = btrfs_item_ptr(leaf, path->slots[0],
1103 struct btrfs_file_extent_item);
1104 extent_type = btrfs_file_extent_type(leaf, fi);
1106 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1107 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1108 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1109 extent_offset = btrfs_file_extent_offset(leaf, fi);
1110 extent_end = found_key.offset +
1111 btrfs_file_extent_num_bytes(leaf, fi);
1112 if (extent_end <= start) {
1116 if (disk_bytenr == 0)
1118 if (btrfs_file_extent_compression(leaf, fi) ||
1119 btrfs_file_extent_encryption(leaf, fi) ||
1120 btrfs_file_extent_other_encoding(leaf, fi))
1122 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1124 if (btrfs_extent_readonly(root, disk_bytenr))
1126 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1128 extent_offset, disk_bytenr))
1130 disk_bytenr += extent_offset;
1131 disk_bytenr += cur_offset - found_key.offset;
1132 num_bytes = min(end + 1, extent_end) - cur_offset;
1134 * force cow if csum exists in the range.
1135 * this ensure that csum for a given extent are
1136 * either valid or do not exist.
1138 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1141 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1142 extent_end = found_key.offset +
1143 btrfs_file_extent_inline_len(leaf, fi);
1144 extent_end = ALIGN(extent_end, root->sectorsize);
1149 if (extent_end <= start) {
1154 if (cow_start == (u64)-1)
1155 cow_start = cur_offset;
1156 cur_offset = extent_end;
1157 if (cur_offset > end)
1163 btrfs_release_path(root, path);
1164 if (cow_start != (u64)-1) {
1165 ret = cow_file_range(inode, locked_page, cow_start,
1166 found_key.offset - 1, page_started,
1169 cow_start = (u64)-1;
1172 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1173 struct extent_map *em;
1174 struct extent_map_tree *em_tree;
1175 em_tree = &BTRFS_I(inode)->extent_tree;
1176 em = alloc_extent_map(GFP_NOFS);
1178 em->start = cur_offset;
1179 em->orig_start = em->start;
1180 em->len = num_bytes;
1181 em->block_len = num_bytes;
1182 em->block_start = disk_bytenr;
1183 em->bdev = root->fs_info->fs_devices->latest_bdev;
1184 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1186 write_lock(&em_tree->lock);
1187 ret = add_extent_mapping(em_tree, em);
1188 write_unlock(&em_tree->lock);
1189 if (ret != -EEXIST) {
1190 free_extent_map(em);
1193 btrfs_drop_extent_cache(inode, em->start,
1194 em->start + em->len - 1, 0);
1196 type = BTRFS_ORDERED_PREALLOC;
1198 type = BTRFS_ORDERED_NOCOW;
1201 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1202 num_bytes, num_bytes, type);
1205 if (root->root_key.objectid ==
1206 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1207 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1212 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1213 cur_offset, cur_offset + num_bytes - 1,
1214 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1215 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1216 EXTENT_SET_PRIVATE2);
1217 cur_offset = extent_end;
1218 if (cur_offset > end)
1221 btrfs_release_path(root, path);
1223 if (cur_offset <= end && cow_start == (u64)-1)
1224 cow_start = cur_offset;
1225 if (cow_start != (u64)-1) {
1226 ret = cow_file_range(inode, locked_page, cow_start, end,
1227 page_started, nr_written, 1);
1232 ret = btrfs_end_transaction_nolock(trans, root);
1235 ret = btrfs_end_transaction(trans, root);
1238 btrfs_free_path(path);
1243 * extent_io.c call back to do delayed allocation processing
1245 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1246 u64 start, u64 end, int *page_started,
1247 unsigned long *nr_written)
1250 struct btrfs_root *root = BTRFS_I(inode)->root;
1252 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1253 ret = run_delalloc_nocow(inode, locked_page, start, end,
1254 page_started, 1, nr_written);
1255 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1256 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257 page_started, 0, nr_written);
1258 else if (!btrfs_test_opt(root, COMPRESS) &&
1259 !(BTRFS_I(inode)->force_compress))
1260 ret = cow_file_range(inode, locked_page, start, end,
1261 page_started, nr_written, 1);
1263 ret = cow_file_range_async(inode, locked_page, start, end,
1264 page_started, nr_written);
1268 static int btrfs_split_extent_hook(struct inode *inode,
1269 struct extent_state *orig, u64 split)
1271 /* not delalloc, ignore it */
1272 if (!(orig->state & EXTENT_DELALLOC))
1275 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1280 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1281 * extents so we can keep track of new extents that are just merged onto old
1282 * extents, such as when we are doing sequential writes, so we can properly
1283 * account for the metadata space we'll need.
1285 static int btrfs_merge_extent_hook(struct inode *inode,
1286 struct extent_state *new,
1287 struct extent_state *other)
1289 /* not delalloc, ignore it */
1290 if (!(other->state & EXTENT_DELALLOC))
1293 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1298 * extent_io.c set_bit_hook, used to track delayed allocation
1299 * bytes in this file, and to maintain the list of inodes that
1300 * have pending delalloc work to be done.
1302 static int btrfs_set_bit_hook(struct inode *inode,
1303 struct extent_state *state, int *bits)
1307 * set_bit and clear bit hooks normally require _irqsave/restore
1308 * but in this case, we are only testeing for the DELALLOC
1309 * bit, which is only set or cleared with irqs on
1311 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1312 struct btrfs_root *root = BTRFS_I(inode)->root;
1313 u64 len = state->end + 1 - state->start;
1314 int do_list = (root->root_key.objectid !=
1315 BTRFS_ROOT_TREE_OBJECTID);
1317 if (*bits & EXTENT_FIRST_DELALLOC)
1318 *bits &= ~EXTENT_FIRST_DELALLOC;
1320 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1322 spin_lock(&root->fs_info->delalloc_lock);
1323 BTRFS_I(inode)->delalloc_bytes += len;
1324 root->fs_info->delalloc_bytes += len;
1325 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1326 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1327 &root->fs_info->delalloc_inodes);
1329 spin_unlock(&root->fs_info->delalloc_lock);
1335 * extent_io.c clear_bit_hook, see set_bit_hook for why
1337 static int btrfs_clear_bit_hook(struct inode *inode,
1338 struct extent_state *state, int *bits)
1341 * set_bit and clear bit hooks normally require _irqsave/restore
1342 * but in this case, we are only testeing for the DELALLOC
1343 * bit, which is only set or cleared with irqs on
1345 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1346 struct btrfs_root *root = BTRFS_I(inode)->root;
1347 u64 len = state->end + 1 - state->start;
1348 int do_list = (root->root_key.objectid !=
1349 BTRFS_ROOT_TREE_OBJECTID);
1351 if (*bits & EXTENT_FIRST_DELALLOC)
1352 *bits &= ~EXTENT_FIRST_DELALLOC;
1353 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1354 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1356 if (*bits & EXTENT_DO_ACCOUNTING)
1357 btrfs_delalloc_release_metadata(inode, len);
1359 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1361 btrfs_free_reserved_data_space(inode, len);
1363 spin_lock(&root->fs_info->delalloc_lock);
1364 root->fs_info->delalloc_bytes -= len;
1365 BTRFS_I(inode)->delalloc_bytes -= len;
1367 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1368 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1369 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1371 spin_unlock(&root->fs_info->delalloc_lock);
1377 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1378 * we don't create bios that span stripes or chunks
1380 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1381 size_t size, struct bio *bio,
1382 unsigned long bio_flags)
1384 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1385 struct btrfs_mapping_tree *map_tree;
1386 u64 logical = (u64)bio->bi_sector << 9;
1391 if (bio_flags & EXTENT_BIO_COMPRESSED)
1394 length = bio->bi_size;
1395 map_tree = &root->fs_info->mapping_tree;
1396 map_length = length;
1397 ret = btrfs_map_block(map_tree, READ, logical,
1398 &map_length, NULL, 0);
1400 if (map_length < length + size)
1406 * in order to insert checksums into the metadata in large chunks,
1407 * we wait until bio submission time. All the pages in the bio are
1408 * checksummed and sums are attached onto the ordered extent record.
1410 * At IO completion time the cums attached on the ordered extent record
1411 * are inserted into the btree
1413 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1414 struct bio *bio, int mirror_num,
1415 unsigned long bio_flags,
1418 struct btrfs_root *root = BTRFS_I(inode)->root;
1421 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1427 * in order to insert checksums into the metadata in large chunks,
1428 * we wait until bio submission time. All the pages in the bio are
1429 * checksummed and sums are attached onto the ordered extent record.
1431 * At IO completion time the cums attached on the ordered extent record
1432 * are inserted into the btree
1434 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1435 int mirror_num, unsigned long bio_flags,
1438 struct btrfs_root *root = BTRFS_I(inode)->root;
1439 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1443 * extent_io.c submission hook. This does the right thing for csum calculation
1444 * on write, or reading the csums from the tree before a read
1446 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1447 int mirror_num, unsigned long bio_flags,
1450 struct btrfs_root *root = BTRFS_I(inode)->root;
1454 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1456 if (root == root->fs_info->tree_root)
1457 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1459 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1462 if (!(rw & REQ_WRITE)) {
1463 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1464 return btrfs_submit_compressed_read(inode, bio,
1465 mirror_num, bio_flags);
1466 } else if (!skip_sum)
1467 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1469 } else if (!skip_sum) {
1470 /* csum items have already been cloned */
1471 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1473 /* we're doing a write, do the async checksumming */
1474 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1475 inode, rw, bio, mirror_num,
1476 bio_flags, bio_offset,
1477 __btrfs_submit_bio_start,
1478 __btrfs_submit_bio_done);
1482 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1486 * given a list of ordered sums record them in the inode. This happens
1487 * at IO completion time based on sums calculated at bio submission time.
1489 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1490 struct inode *inode, u64 file_offset,
1491 struct list_head *list)
1493 struct btrfs_ordered_sum *sum;
1495 btrfs_set_trans_block_group(trans, inode);
1497 list_for_each_entry(sum, list, list) {
1498 btrfs_csum_file_blocks(trans,
1499 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1504 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1505 struct extent_state **cached_state)
1507 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1509 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1510 cached_state, GFP_NOFS);
1513 /* see btrfs_writepage_start_hook for details on why this is required */
1514 struct btrfs_writepage_fixup {
1516 struct btrfs_work work;
1519 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1521 struct btrfs_writepage_fixup *fixup;
1522 struct btrfs_ordered_extent *ordered;
1523 struct extent_state *cached_state = NULL;
1525 struct inode *inode;
1529 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1533 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1534 ClearPageChecked(page);
1538 inode = page->mapping->host;
1539 page_start = page_offset(page);
1540 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1542 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1543 &cached_state, GFP_NOFS);
1545 /* already ordered? We're done */
1546 if (PagePrivate2(page))
1549 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1551 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1552 page_end, &cached_state, GFP_NOFS);
1554 btrfs_start_ordered_extent(inode, ordered, 1);
1559 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1560 ClearPageChecked(page);
1562 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1563 &cached_state, GFP_NOFS);
1566 page_cache_release(page);
1571 * There are a few paths in the higher layers of the kernel that directly
1572 * set the page dirty bit without asking the filesystem if it is a
1573 * good idea. This causes problems because we want to make sure COW
1574 * properly happens and the data=ordered rules are followed.
1576 * In our case any range that doesn't have the ORDERED bit set
1577 * hasn't been properly setup for IO. We kick off an async process
1578 * to fix it up. The async helper will wait for ordered extents, set
1579 * the delalloc bit and make it safe to write the page.
1581 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1583 struct inode *inode = page->mapping->host;
1584 struct btrfs_writepage_fixup *fixup;
1585 struct btrfs_root *root = BTRFS_I(inode)->root;
1587 /* this page is properly in the ordered list */
1588 if (TestClearPagePrivate2(page))
1591 if (PageChecked(page))
1594 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1598 SetPageChecked(page);
1599 page_cache_get(page);
1600 fixup->work.func = btrfs_writepage_fixup_worker;
1602 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1606 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1607 struct inode *inode, u64 file_pos,
1608 u64 disk_bytenr, u64 disk_num_bytes,
1609 u64 num_bytes, u64 ram_bytes,
1610 u8 compression, u8 encryption,
1611 u16 other_encoding, int extent_type)
1613 struct btrfs_root *root = BTRFS_I(inode)->root;
1614 struct btrfs_file_extent_item *fi;
1615 struct btrfs_path *path;
1616 struct extent_buffer *leaf;
1617 struct btrfs_key ins;
1621 path = btrfs_alloc_path();
1624 path->leave_spinning = 1;
1627 * we may be replacing one extent in the tree with another.
1628 * The new extent is pinned in the extent map, and we don't want
1629 * to drop it from the cache until it is completely in the btree.
1631 * So, tell btrfs_drop_extents to leave this extent in the cache.
1632 * the caller is expected to unpin it and allow it to be merged
1635 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1639 ins.objectid = inode->i_ino;
1640 ins.offset = file_pos;
1641 ins.type = BTRFS_EXTENT_DATA_KEY;
1642 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1644 leaf = path->nodes[0];
1645 fi = btrfs_item_ptr(leaf, path->slots[0],
1646 struct btrfs_file_extent_item);
1647 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1648 btrfs_set_file_extent_type(leaf, fi, extent_type);
1649 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1650 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1651 btrfs_set_file_extent_offset(leaf, fi, 0);
1652 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1653 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1654 btrfs_set_file_extent_compression(leaf, fi, compression);
1655 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1656 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1658 btrfs_unlock_up_safe(path, 1);
1659 btrfs_set_lock_blocking(leaf);
1661 btrfs_mark_buffer_dirty(leaf);
1663 inode_add_bytes(inode, num_bytes);
1665 ins.objectid = disk_bytenr;
1666 ins.offset = disk_num_bytes;
1667 ins.type = BTRFS_EXTENT_ITEM_KEY;
1668 ret = btrfs_alloc_reserved_file_extent(trans, root,
1669 root->root_key.objectid,
1670 inode->i_ino, file_pos, &ins);
1672 btrfs_free_path(path);
1678 * helper function for btrfs_finish_ordered_io, this
1679 * just reads in some of the csum leaves to prime them into ram
1680 * before we start the transaction. It limits the amount of btree
1681 * reads required while inside the transaction.
1683 /* as ordered data IO finishes, this gets called so we can finish
1684 * an ordered extent if the range of bytes in the file it covers are
1687 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1689 struct btrfs_root *root = BTRFS_I(inode)->root;
1690 struct btrfs_trans_handle *trans = NULL;
1691 struct btrfs_ordered_extent *ordered_extent = NULL;
1692 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1693 struct extent_state *cached_state = NULL;
1694 int compress_type = 0;
1696 bool nolock = false;
1698 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1702 BUG_ON(!ordered_extent);
1704 nolock = (root == root->fs_info->tree_root);
1706 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1707 BUG_ON(!list_empty(&ordered_extent->list));
1708 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1711 trans = btrfs_join_transaction_nolock(root, 1);
1713 trans = btrfs_join_transaction(root, 1);
1714 BUG_ON(IS_ERR(trans));
1715 btrfs_set_trans_block_group(trans, inode);
1716 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1717 ret = btrfs_update_inode(trans, root, inode);
1723 lock_extent_bits(io_tree, ordered_extent->file_offset,
1724 ordered_extent->file_offset + ordered_extent->len - 1,
1725 0, &cached_state, GFP_NOFS);
1728 trans = btrfs_join_transaction_nolock(root, 1);
1730 trans = btrfs_join_transaction(root, 1);
1731 BUG_ON(IS_ERR(trans));
1732 btrfs_set_trans_block_group(trans, inode);
1733 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1735 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1736 compress_type = ordered_extent->compress_type;
1737 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1738 BUG_ON(compress_type);
1739 ret = btrfs_mark_extent_written(trans, inode,
1740 ordered_extent->file_offset,
1741 ordered_extent->file_offset +
1742 ordered_extent->len);
1745 BUG_ON(root == root->fs_info->tree_root);
1746 ret = insert_reserved_file_extent(trans, inode,
1747 ordered_extent->file_offset,
1748 ordered_extent->start,
1749 ordered_extent->disk_len,
1750 ordered_extent->len,
1751 ordered_extent->len,
1752 compress_type, 0, 0,
1753 BTRFS_FILE_EXTENT_REG);
1754 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1755 ordered_extent->file_offset,
1756 ordered_extent->len);
1759 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1760 ordered_extent->file_offset +
1761 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1763 add_pending_csums(trans, inode, ordered_extent->file_offset,
1764 &ordered_extent->list);
1766 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1767 ret = btrfs_update_inode(trans, root, inode);
1772 btrfs_end_transaction_nolock(trans, root);
1774 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1776 btrfs_end_transaction(trans, root);
1780 btrfs_put_ordered_extent(ordered_extent);
1781 /* once for the tree */
1782 btrfs_put_ordered_extent(ordered_extent);
1787 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1788 struct extent_state *state, int uptodate)
1790 ClearPagePrivate2(page);
1791 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1795 * When IO fails, either with EIO or csum verification fails, we
1796 * try other mirrors that might have a good copy of the data. This
1797 * io_failure_record is used to record state as we go through all the
1798 * mirrors. If another mirror has good data, the page is set up to date
1799 * and things continue. If a good mirror can't be found, the original
1800 * bio end_io callback is called to indicate things have failed.
1802 struct io_failure_record {
1807 unsigned long bio_flags;
1811 static int btrfs_io_failed_hook(struct bio *failed_bio,
1812 struct page *page, u64 start, u64 end,
1813 struct extent_state *state)
1815 struct io_failure_record *failrec = NULL;
1817 struct extent_map *em;
1818 struct inode *inode = page->mapping->host;
1819 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1820 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1827 ret = get_state_private(failure_tree, start, &private);
1829 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1832 failrec->start = start;
1833 failrec->len = end - start + 1;
1834 failrec->last_mirror = 0;
1835 failrec->bio_flags = 0;
1837 read_lock(&em_tree->lock);
1838 em = lookup_extent_mapping(em_tree, start, failrec->len);
1839 if (em->start > start || em->start + em->len < start) {
1840 free_extent_map(em);
1843 read_unlock(&em_tree->lock);
1845 if (!em || IS_ERR(em)) {
1849 logical = start - em->start;
1850 logical = em->block_start + logical;
1851 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1852 logical = em->block_start;
1853 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1854 extent_set_compress_type(&failrec->bio_flags,
1857 failrec->logical = logical;
1858 free_extent_map(em);
1859 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1860 EXTENT_DIRTY, GFP_NOFS);
1861 set_state_private(failure_tree, start,
1862 (u64)(unsigned long)failrec);
1864 failrec = (struct io_failure_record *)(unsigned long)private;
1866 num_copies = btrfs_num_copies(
1867 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1868 failrec->logical, failrec->len);
1869 failrec->last_mirror++;
1871 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1872 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1875 if (state && state->start != failrec->start)
1877 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1879 if (!state || failrec->last_mirror > num_copies) {
1880 set_state_private(failure_tree, failrec->start, 0);
1881 clear_extent_bits(failure_tree, failrec->start,
1882 failrec->start + failrec->len - 1,
1883 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1887 bio = bio_alloc(GFP_NOFS, 1);
1888 bio->bi_private = state;
1889 bio->bi_end_io = failed_bio->bi_end_io;
1890 bio->bi_sector = failrec->logical >> 9;
1891 bio->bi_bdev = failed_bio->bi_bdev;
1894 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1895 if (failed_bio->bi_rw & REQ_WRITE)
1900 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1901 failrec->last_mirror,
1902 failrec->bio_flags, 0);
1907 * each time an IO finishes, we do a fast check in the IO failure tree
1908 * to see if we need to process or clean up an io_failure_record
1910 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1913 u64 private_failure;
1914 struct io_failure_record *failure;
1918 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1919 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1920 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1921 start, &private_failure);
1923 failure = (struct io_failure_record *)(unsigned long)
1925 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1927 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1929 failure->start + failure->len - 1,
1930 EXTENT_DIRTY | EXTENT_LOCKED,
1939 * when reads are done, we need to check csums to verify the data is correct
1940 * if there's a match, we allow the bio to finish. If not, we go through
1941 * the io_failure_record routines to find good copies
1943 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1944 struct extent_state *state)
1946 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1947 struct inode *inode = page->mapping->host;
1948 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1950 u64 private = ~(u32)0;
1952 struct btrfs_root *root = BTRFS_I(inode)->root;
1955 if (PageChecked(page)) {
1956 ClearPageChecked(page);
1960 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1963 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1964 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1965 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1970 if (state && state->start == start) {
1971 private = state->private;
1974 ret = get_state_private(io_tree, start, &private);
1976 kaddr = kmap_atomic(page, KM_USER0);
1980 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1981 btrfs_csum_final(csum, (char *)&csum);
1982 if (csum != private)
1985 kunmap_atomic(kaddr, KM_USER0);
1987 /* if the io failure tree for this inode is non-empty,
1988 * check to see if we've recovered from a failed IO
1990 btrfs_clean_io_failures(inode, start);
1994 if (printk_ratelimit()) {
1995 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1996 "private %llu\n", page->mapping->host->i_ino,
1997 (unsigned long long)start, csum,
1998 (unsigned long long)private);
2000 memset(kaddr + offset, 1, end - start + 1);
2001 flush_dcache_page(page);
2002 kunmap_atomic(kaddr, KM_USER0);
2008 struct delayed_iput {
2009 struct list_head list;
2010 struct inode *inode;
2013 void btrfs_add_delayed_iput(struct inode *inode)
2015 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2016 struct delayed_iput *delayed;
2018 if (atomic_add_unless(&inode->i_count, -1, 1))
2021 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2022 delayed->inode = inode;
2024 spin_lock(&fs_info->delayed_iput_lock);
2025 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2026 spin_unlock(&fs_info->delayed_iput_lock);
2029 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2032 struct btrfs_fs_info *fs_info = root->fs_info;
2033 struct delayed_iput *delayed;
2036 spin_lock(&fs_info->delayed_iput_lock);
2037 empty = list_empty(&fs_info->delayed_iputs);
2038 spin_unlock(&fs_info->delayed_iput_lock);
2042 down_read(&root->fs_info->cleanup_work_sem);
2043 spin_lock(&fs_info->delayed_iput_lock);
2044 list_splice_init(&fs_info->delayed_iputs, &list);
2045 spin_unlock(&fs_info->delayed_iput_lock);
2047 while (!list_empty(&list)) {
2048 delayed = list_entry(list.next, struct delayed_iput, list);
2049 list_del(&delayed->list);
2050 iput(delayed->inode);
2053 up_read(&root->fs_info->cleanup_work_sem);
2057 * calculate extra metadata reservation when snapshotting a subvolume
2058 * contains orphan files.
2060 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2061 struct btrfs_pending_snapshot *pending,
2062 u64 *bytes_to_reserve)
2064 struct btrfs_root *root;
2065 struct btrfs_block_rsv *block_rsv;
2069 root = pending->root;
2070 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2073 block_rsv = root->orphan_block_rsv;
2075 /* orphan block reservation for the snapshot */
2076 num_bytes = block_rsv->size;
2079 * after the snapshot is created, COWing tree blocks may use more
2080 * space than it frees. So we should make sure there is enough
2083 index = trans->transid & 0x1;
2084 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2085 num_bytes += block_rsv->size -
2086 (block_rsv->reserved + block_rsv->freed[index]);
2089 *bytes_to_reserve += num_bytes;
2092 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2093 struct btrfs_pending_snapshot *pending)
2095 struct btrfs_root *root = pending->root;
2096 struct btrfs_root *snap = pending->snap;
2097 struct btrfs_block_rsv *block_rsv;
2102 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2105 /* refill source subvolume's orphan block reservation */
2106 block_rsv = root->orphan_block_rsv;
2107 index = trans->transid & 0x1;
2108 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2109 num_bytes = block_rsv->size -
2110 (block_rsv->reserved + block_rsv->freed[index]);
2111 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2112 root->orphan_block_rsv,
2117 /* setup orphan block reservation for the snapshot */
2118 block_rsv = btrfs_alloc_block_rsv(snap);
2121 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2122 snap->orphan_block_rsv = block_rsv;
2124 num_bytes = root->orphan_block_rsv->size;
2125 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2126 block_rsv, num_bytes);
2130 /* insert orphan item for the snapshot */
2131 WARN_ON(!root->orphan_item_inserted);
2132 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2133 snap->root_key.objectid);
2135 snap->orphan_item_inserted = 1;
2139 enum btrfs_orphan_cleanup_state {
2140 ORPHAN_CLEANUP_STARTED = 1,
2141 ORPHAN_CLEANUP_DONE = 2,
2145 * This is called in transaction commmit time. If there are no orphan
2146 * files in the subvolume, it removes orphan item and frees block_rsv
2149 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2150 struct btrfs_root *root)
2154 if (!list_empty(&root->orphan_list) ||
2155 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2158 if (root->orphan_item_inserted &&
2159 btrfs_root_refs(&root->root_item) > 0) {
2160 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2161 root->root_key.objectid);
2163 root->orphan_item_inserted = 0;
2166 if (root->orphan_block_rsv) {
2167 WARN_ON(root->orphan_block_rsv->size > 0);
2168 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2169 root->orphan_block_rsv = NULL;
2174 * This creates an orphan entry for the given inode in case something goes
2175 * wrong in the middle of an unlink/truncate.
2177 * NOTE: caller of this function should reserve 5 units of metadata for
2180 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2182 struct btrfs_root *root = BTRFS_I(inode)->root;
2183 struct btrfs_block_rsv *block_rsv = NULL;
2188 if (!root->orphan_block_rsv) {
2189 block_rsv = btrfs_alloc_block_rsv(root);
2193 spin_lock(&root->orphan_lock);
2194 if (!root->orphan_block_rsv) {
2195 root->orphan_block_rsv = block_rsv;
2196 } else if (block_rsv) {
2197 btrfs_free_block_rsv(root, block_rsv);
2201 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2202 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2205 * For proper ENOSPC handling, we should do orphan
2206 * cleanup when mounting. But this introduces backward
2207 * compatibility issue.
2209 if (!xchg(&root->orphan_item_inserted, 1))
2216 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2219 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2220 BTRFS_I(inode)->orphan_meta_reserved = 1;
2223 spin_unlock(&root->orphan_lock);
2226 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2228 /* grab metadata reservation from transaction handle */
2230 ret = btrfs_orphan_reserve_metadata(trans, inode);
2234 /* insert an orphan item to track this unlinked/truncated file */
2236 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2240 /* insert an orphan item to track subvolume contains orphan files */
2242 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2243 root->root_key.objectid);
2250 * We have done the truncate/delete so we can go ahead and remove the orphan
2251 * item for this particular inode.
2253 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2255 struct btrfs_root *root = BTRFS_I(inode)->root;
2256 int delete_item = 0;
2257 int release_rsv = 0;
2260 spin_lock(&root->orphan_lock);
2261 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2262 list_del_init(&BTRFS_I(inode)->i_orphan);
2266 if (BTRFS_I(inode)->orphan_meta_reserved) {
2267 BTRFS_I(inode)->orphan_meta_reserved = 0;
2270 spin_unlock(&root->orphan_lock);
2272 if (trans && delete_item) {
2273 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2278 btrfs_orphan_release_metadata(inode);
2284 * this cleans up any orphans that may be left on the list from the last use
2287 void btrfs_orphan_cleanup(struct btrfs_root *root)
2289 struct btrfs_path *path;
2290 struct extent_buffer *leaf;
2291 struct btrfs_key key, found_key;
2292 struct btrfs_trans_handle *trans;
2293 struct inode *inode;
2294 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2296 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2299 path = btrfs_alloc_path();
2303 key.objectid = BTRFS_ORPHAN_OBJECTID;
2304 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2305 key.offset = (u64)-1;
2308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2310 printk(KERN_ERR "Error searching slot for orphan: %d"
2316 * if ret == 0 means we found what we were searching for, which
2317 * is weird, but possible, so only screw with path if we didnt
2318 * find the key and see if we have stuff that matches
2321 if (path->slots[0] == 0)
2326 /* pull out the item */
2327 leaf = path->nodes[0];
2328 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2330 /* make sure the item matches what we want */
2331 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2333 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2336 /* release the path since we're done with it */
2337 btrfs_release_path(root, path);
2340 * this is where we are basically btrfs_lookup, without the
2341 * crossing root thing. we store the inode number in the
2342 * offset of the orphan item.
2344 found_key.objectid = found_key.offset;
2345 found_key.type = BTRFS_INODE_ITEM_KEY;
2346 found_key.offset = 0;
2347 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2348 BUG_ON(IS_ERR(inode));
2351 * add this inode to the orphan list so btrfs_orphan_del does
2352 * the proper thing when we hit it
2354 spin_lock(&root->orphan_lock);
2355 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2356 spin_unlock(&root->orphan_lock);
2359 * if this is a bad inode, means we actually succeeded in
2360 * removing the inode, but not the orphan record, which means
2361 * we need to manually delete the orphan since iput will just
2362 * do a destroy_inode
2364 if (is_bad_inode(inode)) {
2365 trans = btrfs_start_transaction(root, 0);
2366 BUG_ON(IS_ERR(trans));
2367 btrfs_orphan_del(trans, inode);
2368 btrfs_end_transaction(trans, root);
2373 /* if we have links, this was a truncate, lets do that */
2374 if (inode->i_nlink) {
2375 if (!S_ISREG(inode->i_mode)) {
2381 btrfs_truncate(inode);
2386 /* this will do delete_inode and everything for us */
2389 btrfs_free_path(path);
2391 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2393 if (root->orphan_block_rsv)
2394 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2397 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2398 trans = btrfs_join_transaction(root, 1);
2399 BUG_ON(IS_ERR(trans));
2400 btrfs_end_transaction(trans, root);
2404 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2406 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2410 * very simple check to peek ahead in the leaf looking for xattrs. If we
2411 * don't find any xattrs, we know there can't be any acls.
2413 * slot is the slot the inode is in, objectid is the objectid of the inode
2415 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2416 int slot, u64 objectid)
2418 u32 nritems = btrfs_header_nritems(leaf);
2419 struct btrfs_key found_key;
2423 while (slot < nritems) {
2424 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2426 /* we found a different objectid, there must not be acls */
2427 if (found_key.objectid != objectid)
2430 /* we found an xattr, assume we've got an acl */
2431 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2435 * we found a key greater than an xattr key, there can't
2436 * be any acls later on
2438 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2445 * it goes inode, inode backrefs, xattrs, extents,
2446 * so if there are a ton of hard links to an inode there can
2447 * be a lot of backrefs. Don't waste time searching too hard,
2448 * this is just an optimization
2453 /* we hit the end of the leaf before we found an xattr or
2454 * something larger than an xattr. We have to assume the inode
2461 * read an inode from the btree into the in-memory inode
2463 static void btrfs_read_locked_inode(struct inode *inode)
2465 struct btrfs_path *path;
2466 struct extent_buffer *leaf;
2467 struct btrfs_inode_item *inode_item;
2468 struct btrfs_timespec *tspec;
2469 struct btrfs_root *root = BTRFS_I(inode)->root;
2470 struct btrfs_key location;
2472 u64 alloc_group_block;
2476 path = btrfs_alloc_path();
2478 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2480 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2484 leaf = path->nodes[0];
2485 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2486 struct btrfs_inode_item);
2488 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2489 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2490 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2491 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2492 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2494 tspec = btrfs_inode_atime(inode_item);
2495 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2496 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2498 tspec = btrfs_inode_mtime(inode_item);
2499 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2500 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2502 tspec = btrfs_inode_ctime(inode_item);
2503 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2504 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2506 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2507 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2508 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2509 inode->i_generation = BTRFS_I(inode)->generation;
2511 rdev = btrfs_inode_rdev(leaf, inode_item);
2513 BTRFS_I(inode)->index_cnt = (u64)-1;
2514 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2516 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2519 * try to precache a NULL acl entry for files that don't have
2520 * any xattrs or acls
2522 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2524 cache_no_acl(inode);
2526 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2527 alloc_group_block, 0);
2528 btrfs_free_path(path);
2531 switch (inode->i_mode & S_IFMT) {
2533 inode->i_mapping->a_ops = &btrfs_aops;
2534 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2535 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2536 inode->i_fop = &btrfs_file_operations;
2537 inode->i_op = &btrfs_file_inode_operations;
2540 inode->i_fop = &btrfs_dir_file_operations;
2541 if (root == root->fs_info->tree_root)
2542 inode->i_op = &btrfs_dir_ro_inode_operations;
2544 inode->i_op = &btrfs_dir_inode_operations;
2547 inode->i_op = &btrfs_symlink_inode_operations;
2548 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2549 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2552 inode->i_op = &btrfs_special_inode_operations;
2553 init_special_inode(inode, inode->i_mode, rdev);
2557 btrfs_update_iflags(inode);
2561 btrfs_free_path(path);
2562 make_bad_inode(inode);
2566 * given a leaf and an inode, copy the inode fields into the leaf
2568 static void fill_inode_item(struct btrfs_trans_handle *trans,
2569 struct extent_buffer *leaf,
2570 struct btrfs_inode_item *item,
2571 struct inode *inode)
2573 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2574 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2575 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2576 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2577 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2579 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2580 inode->i_atime.tv_sec);
2581 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2582 inode->i_atime.tv_nsec);
2584 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2585 inode->i_mtime.tv_sec);
2586 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2587 inode->i_mtime.tv_nsec);
2589 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2590 inode->i_ctime.tv_sec);
2591 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2592 inode->i_ctime.tv_nsec);
2594 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2595 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2596 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2597 btrfs_set_inode_transid(leaf, item, trans->transid);
2598 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2599 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2600 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2604 * copy everything in the in-memory inode into the btree.
2606 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2607 struct btrfs_root *root, struct inode *inode)
2609 struct btrfs_inode_item *inode_item;
2610 struct btrfs_path *path;
2611 struct extent_buffer *leaf;
2614 path = btrfs_alloc_path();
2616 path->leave_spinning = 1;
2617 ret = btrfs_lookup_inode(trans, root, path,
2618 &BTRFS_I(inode)->location, 1);
2625 btrfs_unlock_up_safe(path, 1);
2626 leaf = path->nodes[0];
2627 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2628 struct btrfs_inode_item);
2630 fill_inode_item(trans, leaf, inode_item, inode);
2631 btrfs_mark_buffer_dirty(leaf);
2632 btrfs_set_inode_last_trans(trans, inode);
2635 btrfs_free_path(path);
2641 * unlink helper that gets used here in inode.c and in the tree logging
2642 * recovery code. It remove a link in a directory with a given name, and
2643 * also drops the back refs in the inode to the directory
2645 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2646 struct btrfs_root *root,
2647 struct inode *dir, struct inode *inode,
2648 const char *name, int name_len)
2650 struct btrfs_path *path;
2652 struct extent_buffer *leaf;
2653 struct btrfs_dir_item *di;
2654 struct btrfs_key key;
2657 path = btrfs_alloc_path();
2663 path->leave_spinning = 1;
2664 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2665 name, name_len, -1);
2674 leaf = path->nodes[0];
2675 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2676 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2679 btrfs_release_path(root, path);
2681 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2683 dir->i_ino, &index);
2685 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2686 "inode %lu parent %lu\n", name_len, name,
2687 inode->i_ino, dir->i_ino);
2691 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2692 index, name, name_len, -1);
2701 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2702 btrfs_release_path(root, path);
2704 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2706 BUG_ON(ret != 0 && ret != -ENOENT);
2708 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2713 btrfs_free_path(path);
2717 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2718 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2719 btrfs_update_inode(trans, root, dir);
2720 btrfs_drop_nlink(inode);
2721 ret = btrfs_update_inode(trans, root, inode);
2726 /* helper to check if there is any shared block in the path */
2727 static int check_path_shared(struct btrfs_root *root,
2728 struct btrfs_path *path)
2730 struct extent_buffer *eb;
2734 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2737 if (!path->nodes[level])
2739 eb = path->nodes[level];
2740 if (!btrfs_block_can_be_shared(root, eb))
2742 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2751 * helper to start transaction for unlink and rmdir.
2753 * unlink and rmdir are special in btrfs, they do not always free space.
2754 * so in enospc case, we should make sure they will free space before
2755 * allowing them to use the global metadata reservation.
2757 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2758 struct dentry *dentry)
2760 struct btrfs_trans_handle *trans;
2761 struct btrfs_root *root = BTRFS_I(dir)->root;
2762 struct btrfs_path *path;
2763 struct btrfs_inode_ref *ref;
2764 struct btrfs_dir_item *di;
2765 struct inode *inode = dentry->d_inode;
2771 trans = btrfs_start_transaction(root, 10);
2772 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2775 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2776 return ERR_PTR(-ENOSPC);
2778 /* check if there is someone else holds reference */
2779 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2780 return ERR_PTR(-ENOSPC);
2782 if (atomic_read(&inode->i_count) > 2)
2783 return ERR_PTR(-ENOSPC);
2785 if (xchg(&root->fs_info->enospc_unlink, 1))
2786 return ERR_PTR(-ENOSPC);
2788 path = btrfs_alloc_path();
2790 root->fs_info->enospc_unlink = 0;
2791 return ERR_PTR(-ENOMEM);
2794 trans = btrfs_start_transaction(root, 0);
2795 if (IS_ERR(trans)) {
2796 btrfs_free_path(path);
2797 root->fs_info->enospc_unlink = 0;
2801 path->skip_locking = 1;
2802 path->search_commit_root = 1;
2804 ret = btrfs_lookup_inode(trans, root, path,
2805 &BTRFS_I(dir)->location, 0);
2811 if (check_path_shared(root, path))
2816 btrfs_release_path(root, path);
2818 ret = btrfs_lookup_inode(trans, root, path,
2819 &BTRFS_I(inode)->location, 0);
2825 if (check_path_shared(root, path))
2830 btrfs_release_path(root, path);
2832 if (ret == 0 && S_ISREG(inode->i_mode)) {
2833 ret = btrfs_lookup_file_extent(trans, root, path,
2834 inode->i_ino, (u64)-1, 0);
2840 if (check_path_shared(root, path))
2842 btrfs_release_path(root, path);
2850 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2851 dentry->d_name.name, dentry->d_name.len, 0);
2857 if (check_path_shared(root, path))
2863 btrfs_release_path(root, path);
2865 ref = btrfs_lookup_inode_ref(trans, root, path,
2866 dentry->d_name.name, dentry->d_name.len,
2867 inode->i_ino, dir->i_ino, 0);
2873 if (check_path_shared(root, path))
2875 index = btrfs_inode_ref_index(path->nodes[0], ref);
2876 btrfs_release_path(root, path);
2878 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2879 dentry->d_name.name, dentry->d_name.len, 0);
2884 BUG_ON(ret == -ENOENT);
2885 if (check_path_shared(root, path))
2890 btrfs_free_path(path);
2892 btrfs_end_transaction(trans, root);
2893 root->fs_info->enospc_unlink = 0;
2894 return ERR_PTR(err);
2897 trans->block_rsv = &root->fs_info->global_block_rsv;
2901 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2902 struct btrfs_root *root)
2904 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2905 BUG_ON(!root->fs_info->enospc_unlink);
2906 root->fs_info->enospc_unlink = 0;
2908 btrfs_end_transaction_throttle(trans, root);
2911 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2913 struct btrfs_root *root = BTRFS_I(dir)->root;
2914 struct btrfs_trans_handle *trans;
2915 struct inode *inode = dentry->d_inode;
2917 unsigned long nr = 0;
2919 trans = __unlink_start_trans(dir, dentry);
2921 return PTR_ERR(trans);
2923 btrfs_set_trans_block_group(trans, dir);
2925 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2927 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2928 dentry->d_name.name, dentry->d_name.len);
2931 if (inode->i_nlink == 0) {
2932 ret = btrfs_orphan_add(trans, inode);
2936 nr = trans->blocks_used;
2937 __unlink_end_trans(trans, root);
2938 btrfs_btree_balance_dirty(root, nr);
2942 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2943 struct btrfs_root *root,
2944 struct inode *dir, u64 objectid,
2945 const char *name, int name_len)
2947 struct btrfs_path *path;
2948 struct extent_buffer *leaf;
2949 struct btrfs_dir_item *di;
2950 struct btrfs_key key;
2954 path = btrfs_alloc_path();
2958 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2959 name, name_len, -1);
2960 BUG_ON(!di || IS_ERR(di));
2962 leaf = path->nodes[0];
2963 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2964 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2965 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2967 btrfs_release_path(root, path);
2969 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2970 objectid, root->root_key.objectid,
2971 dir->i_ino, &index, name, name_len);
2973 BUG_ON(ret != -ENOENT);
2974 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2976 BUG_ON(!di || IS_ERR(di));
2978 leaf = path->nodes[0];
2979 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2980 btrfs_release_path(root, path);
2984 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2985 index, name, name_len, -1);
2986 BUG_ON(!di || IS_ERR(di));
2988 leaf = path->nodes[0];
2989 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2990 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2991 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2993 btrfs_release_path(root, path);
2995 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2996 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2997 ret = btrfs_update_inode(trans, root, dir);
3000 btrfs_free_path(path);
3004 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3006 struct inode *inode = dentry->d_inode;
3008 struct btrfs_root *root = BTRFS_I(dir)->root;
3009 struct btrfs_trans_handle *trans;
3010 unsigned long nr = 0;
3012 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3013 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3016 trans = __unlink_start_trans(dir, dentry);
3018 return PTR_ERR(trans);
3020 btrfs_set_trans_block_group(trans, dir);
3022 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3023 err = btrfs_unlink_subvol(trans, root, dir,
3024 BTRFS_I(inode)->location.objectid,
3025 dentry->d_name.name,
3026 dentry->d_name.len);
3030 err = btrfs_orphan_add(trans, inode);
3034 /* now the directory is empty */
3035 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3036 dentry->d_name.name, dentry->d_name.len);
3038 btrfs_i_size_write(inode, 0);
3040 nr = trans->blocks_used;
3041 __unlink_end_trans(trans, root);
3042 btrfs_btree_balance_dirty(root, nr);
3049 * when truncating bytes in a file, it is possible to avoid reading
3050 * the leaves that contain only checksum items. This can be the
3051 * majority of the IO required to delete a large file, but it must
3052 * be done carefully.
3054 * The keys in the level just above the leaves are checked to make sure
3055 * the lowest key in a given leaf is a csum key, and starts at an offset
3056 * after the new size.
3058 * Then the key for the next leaf is checked to make sure it also has
3059 * a checksum item for the same file. If it does, we know our target leaf
3060 * contains only checksum items, and it can be safely freed without reading
3063 * This is just an optimization targeted at large files. It may do
3064 * nothing. It will return 0 unless things went badly.
3066 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3067 struct btrfs_root *root,
3068 struct btrfs_path *path,
3069 struct inode *inode, u64 new_size)
3071 struct btrfs_key key;
3074 struct btrfs_key found_key;
3075 struct btrfs_key other_key;
3076 struct btrfs_leaf_ref *ref;
3080 path->lowest_level = 1;
3081 key.objectid = inode->i_ino;
3082 key.type = BTRFS_CSUM_ITEM_KEY;
3083 key.offset = new_size;
3085 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3089 if (path->nodes[1] == NULL) {
3094 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3095 nritems = btrfs_header_nritems(path->nodes[1]);
3100 if (path->slots[1] >= nritems)
3103 /* did we find a key greater than anything we want to delete? */
3104 if (found_key.objectid > inode->i_ino ||
3105 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3108 /* we check the next key in the node to make sure the leave contains
3109 * only checksum items. This comparison doesn't work if our
3110 * leaf is the last one in the node
3112 if (path->slots[1] + 1 >= nritems) {
3114 /* search forward from the last key in the node, this
3115 * will bring us into the next node in the tree
3117 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3119 /* unlikely, but we inc below, so check to be safe */
3120 if (found_key.offset == (u64)-1)
3123 /* search_forward needs a path with locks held, do the
3124 * search again for the original key. It is possible
3125 * this will race with a balance and return a path that
3126 * we could modify, but this drop is just an optimization
3127 * and is allowed to miss some leaves.
3129 btrfs_release_path(root, path);
3132 /* setup a max key for search_forward */
3133 other_key.offset = (u64)-1;
3134 other_key.type = key.type;
3135 other_key.objectid = key.objectid;
3137 path->keep_locks = 1;
3138 ret = btrfs_search_forward(root, &found_key, &other_key,
3140 path->keep_locks = 0;
3141 if (ret || found_key.objectid != key.objectid ||
3142 found_key.type != key.type) {
3147 key.offset = found_key.offset;
3148 btrfs_release_path(root, path);
3153 /* we know there's one more slot after us in the tree,
3154 * read that key so we can verify it is also a checksum item
3156 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3158 if (found_key.objectid < inode->i_ino)
3161 if (found_key.type != key.type || found_key.offset < new_size)
3165 * if the key for the next leaf isn't a csum key from this objectid,
3166 * we can't be sure there aren't good items inside this leaf.
3169 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3172 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3173 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3175 * it is safe to delete this leaf, it contains only
3176 * csum items from this inode at an offset >= new_size
3178 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3181 if (root->ref_cows && leaf_gen < trans->transid) {
3182 ref = btrfs_alloc_leaf_ref(root, 0);
3184 ref->root_gen = root->root_key.offset;
3185 ref->bytenr = leaf_start;
3187 ref->generation = leaf_gen;
3190 btrfs_sort_leaf_ref(ref);
3192 ret = btrfs_add_leaf_ref(root, ref, 0);
3194 btrfs_free_leaf_ref(root, ref);
3200 btrfs_release_path(root, path);
3202 if (other_key.objectid == inode->i_ino &&
3203 other_key.type == key.type && other_key.offset > key.offset) {
3204 key.offset = other_key.offset;
3210 /* fixup any changes we've made to the path */
3211 path->lowest_level = 0;
3212 path->keep_locks = 0;
3213 btrfs_release_path(root, path);
3220 * this can truncate away extent items, csum items and directory items.
3221 * It starts at a high offset and removes keys until it can't find
3222 * any higher than new_size
3224 * csum items that cross the new i_size are truncated to the new size
3227 * min_type is the minimum key type to truncate down to. If set to 0, this
3228 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3230 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3231 struct btrfs_root *root,
3232 struct inode *inode,
3233 u64 new_size, u32 min_type)
3235 struct btrfs_path *path;
3236 struct extent_buffer *leaf;
3237 struct btrfs_file_extent_item *fi;
3238 struct btrfs_key key;
3239 struct btrfs_key found_key;
3240 u64 extent_start = 0;
3241 u64 extent_num_bytes = 0;
3242 u64 extent_offset = 0;
3244 u64 mask = root->sectorsize - 1;
3245 u32 found_type = (u8)-1;
3248 int pending_del_nr = 0;
3249 int pending_del_slot = 0;
3250 int extent_type = -1;
3255 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3257 if (root->ref_cows || root == root->fs_info->tree_root)
3258 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3260 path = btrfs_alloc_path();
3264 key.objectid = inode->i_ino;
3265 key.offset = (u64)-1;
3269 path->leave_spinning = 1;
3270 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3277 /* there are no items in the tree for us to truncate, we're
3280 if (path->slots[0] == 0)
3287 leaf = path->nodes[0];
3288 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3289 found_type = btrfs_key_type(&found_key);
3292 if (found_key.objectid != inode->i_ino)
3295 if (found_type < min_type)
3298 item_end = found_key.offset;
3299 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3300 fi = btrfs_item_ptr(leaf, path->slots[0],
3301 struct btrfs_file_extent_item);
3302 extent_type = btrfs_file_extent_type(leaf, fi);
3303 encoding = btrfs_file_extent_compression(leaf, fi);
3304 encoding |= btrfs_file_extent_encryption(leaf, fi);
3305 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3307 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3309 btrfs_file_extent_num_bytes(leaf, fi);
3310 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3311 item_end += btrfs_file_extent_inline_len(leaf,
3316 if (found_type > min_type) {
3319 if (item_end < new_size)
3321 if (found_key.offset >= new_size)
3327 /* FIXME, shrink the extent if the ref count is only 1 */
3328 if (found_type != BTRFS_EXTENT_DATA_KEY)
3331 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3333 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3334 if (!del_item && !encoding) {
3335 u64 orig_num_bytes =
3336 btrfs_file_extent_num_bytes(leaf, fi);
3337 extent_num_bytes = new_size -
3338 found_key.offset + root->sectorsize - 1;
3339 extent_num_bytes = extent_num_bytes &
3340 ~((u64)root->sectorsize - 1);
3341 btrfs_set_file_extent_num_bytes(leaf, fi,
3343 num_dec = (orig_num_bytes -
3345 if (root->ref_cows && extent_start != 0)
3346 inode_sub_bytes(inode, num_dec);
3347 btrfs_mark_buffer_dirty(leaf);
3350 btrfs_file_extent_disk_num_bytes(leaf,
3352 extent_offset = found_key.offset -
3353 btrfs_file_extent_offset(leaf, fi);
3355 /* FIXME blocksize != 4096 */
3356 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3357 if (extent_start != 0) {
3360 inode_sub_bytes(inode, num_dec);
3363 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3365 * we can't truncate inline items that have had
3369 btrfs_file_extent_compression(leaf, fi) == 0 &&
3370 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3371 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3372 u32 size = new_size - found_key.offset;
3374 if (root->ref_cows) {
3375 inode_sub_bytes(inode, item_end + 1 -
3379 btrfs_file_extent_calc_inline_size(size);
3380 ret = btrfs_truncate_item(trans, root, path,
3383 } else if (root->ref_cows) {
3384 inode_sub_bytes(inode, item_end + 1 -
3390 if (!pending_del_nr) {
3391 /* no pending yet, add ourselves */
3392 pending_del_slot = path->slots[0];
3394 } else if (pending_del_nr &&
3395 path->slots[0] + 1 == pending_del_slot) {
3396 /* hop on the pending chunk */
3398 pending_del_slot = path->slots[0];
3405 if (found_extent && (root->ref_cows ||
3406 root == root->fs_info->tree_root)) {
3407 btrfs_set_path_blocking(path);
3408 ret = btrfs_free_extent(trans, root, extent_start,
3409 extent_num_bytes, 0,
3410 btrfs_header_owner(leaf),
3411 inode->i_ino, extent_offset);
3415 if (found_type == BTRFS_INODE_ITEM_KEY)
3418 if (path->slots[0] == 0 ||
3419 path->slots[0] != pending_del_slot) {
3420 if (root->ref_cows) {
3424 if (pending_del_nr) {
3425 ret = btrfs_del_items(trans, root, path,
3431 btrfs_release_path(root, path);
3438 if (pending_del_nr) {
3439 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3443 btrfs_free_path(path);
3448 * taken from block_truncate_page, but does cow as it zeros out
3449 * any bytes left in the last page in the file.
3451 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3453 struct inode *inode = mapping->host;
3454 struct btrfs_root *root = BTRFS_I(inode)->root;
3455 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3456 struct btrfs_ordered_extent *ordered;
3457 struct extent_state *cached_state = NULL;
3459 u32 blocksize = root->sectorsize;
3460 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3461 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3467 if ((offset & (blocksize - 1)) == 0)
3469 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3475 page = grab_cache_page(mapping, index);
3477 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3481 page_start = page_offset(page);
3482 page_end = page_start + PAGE_CACHE_SIZE - 1;
3484 if (!PageUptodate(page)) {
3485 ret = btrfs_readpage(NULL, page);
3487 if (page->mapping != mapping) {
3489 page_cache_release(page);
3492 if (!PageUptodate(page)) {
3497 wait_on_page_writeback(page);
3499 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3501 set_page_extent_mapped(page);
3503 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3505 unlock_extent_cached(io_tree, page_start, page_end,
3506 &cached_state, GFP_NOFS);
3508 page_cache_release(page);
3509 btrfs_start_ordered_extent(inode, ordered, 1);
3510 btrfs_put_ordered_extent(ordered);
3514 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3515 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3516 0, 0, &cached_state, GFP_NOFS);
3518 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3521 unlock_extent_cached(io_tree, page_start, page_end,
3522 &cached_state, GFP_NOFS);
3527 if (offset != PAGE_CACHE_SIZE) {
3529 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3530 flush_dcache_page(page);
3533 ClearPageChecked(page);
3534 set_page_dirty(page);
3535 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3540 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3542 page_cache_release(page);
3547 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3549 struct btrfs_trans_handle *trans;
3550 struct btrfs_root *root = BTRFS_I(inode)->root;
3551 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3552 struct extent_map *em = NULL;
3553 struct extent_state *cached_state = NULL;
3554 u64 mask = root->sectorsize - 1;
3555 u64 hole_start = (oldsize + mask) & ~mask;
3556 u64 block_end = (size + mask) & ~mask;
3562 if (size <= hole_start)
3566 struct btrfs_ordered_extent *ordered;
3567 btrfs_wait_ordered_range(inode, hole_start,
3568 block_end - hole_start);
3569 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3570 &cached_state, GFP_NOFS);
3571 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3574 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3575 &cached_state, GFP_NOFS);
3576 btrfs_put_ordered_extent(ordered);
3579 cur_offset = hole_start;
3581 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3582 block_end - cur_offset, 0);
3583 BUG_ON(IS_ERR(em) || !em);
3584 last_byte = min(extent_map_end(em), block_end);
3585 last_byte = (last_byte + mask) & ~mask;
3586 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3588 hole_size = last_byte - cur_offset;
3590 trans = btrfs_start_transaction(root, 2);
3591 if (IS_ERR(trans)) {
3592 err = PTR_ERR(trans);
3595 btrfs_set_trans_block_group(trans, inode);
3597 err = btrfs_drop_extents(trans, inode, cur_offset,
3598 cur_offset + hole_size,
3602 err = btrfs_insert_file_extent(trans, root,
3603 inode->i_ino, cur_offset, 0,
3604 0, hole_size, 0, hole_size,
3608 btrfs_drop_extent_cache(inode, hole_start,
3611 btrfs_end_transaction(trans, root);
3613 free_extent_map(em);
3615 cur_offset = last_byte;
3616 if (cur_offset >= block_end)
3620 free_extent_map(em);
3621 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3626 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3628 struct btrfs_root *root = BTRFS_I(inode)->root;
3629 struct btrfs_trans_handle *trans;
3630 loff_t oldsize = i_size_read(inode);
3634 if (newsize == oldsize)
3637 trans = btrfs_start_transaction(root, 5);
3639 return PTR_ERR(trans);
3641 btrfs_set_trans_block_group(trans, inode);
3643 ret = btrfs_orphan_add(trans, inode);
3646 nr = trans->blocks_used;
3647 btrfs_end_transaction(trans, root);
3648 btrfs_btree_balance_dirty(root, nr);
3650 if (newsize > oldsize) {
3651 i_size_write(inode, newsize);
3652 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3653 truncate_pagecache(inode, oldsize, newsize);
3654 ret = btrfs_cont_expand(inode, oldsize, newsize);
3656 btrfs_setsize(inode, oldsize);
3660 trans = btrfs_start_transaction(root, 0);
3661 BUG_ON(IS_ERR(trans));
3662 btrfs_set_trans_block_group(trans, inode);
3663 trans->block_rsv = root->orphan_block_rsv;
3664 BUG_ON(!trans->block_rsv);
3666 ret = btrfs_update_inode(trans, root, inode);
3668 if (inode->i_nlink > 0) {
3669 ret = btrfs_orphan_del(trans, inode);
3672 nr = trans->blocks_used;
3673 btrfs_end_transaction(trans, root);
3674 btrfs_btree_balance_dirty(root, nr);
3678 * We're truncating a file that used to have good data down to
3679 * zero. Make sure it gets into the ordered flush list so that
3680 * any new writes get down to disk quickly.
3683 BTRFS_I(inode)->ordered_data_close = 1;
3685 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3686 truncate_setsize(inode, newsize);
3687 ret = btrfs_truncate(inode);
3693 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3695 struct inode *inode = dentry->d_inode;
3696 struct btrfs_root *root = BTRFS_I(inode)->root;
3699 if (btrfs_root_readonly(root))
3702 err = inode_change_ok(inode, attr);
3706 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3707 err = btrfs_setsize(inode, attr->ia_size);
3712 if (attr->ia_valid) {
3713 setattr_copy(inode, attr);
3714 mark_inode_dirty(inode);
3716 if (attr->ia_valid & ATTR_MODE)
3717 err = btrfs_acl_chmod(inode);
3723 void btrfs_evict_inode(struct inode *inode)
3725 struct btrfs_trans_handle *trans;
3726 struct btrfs_root *root = BTRFS_I(inode)->root;
3730 truncate_inode_pages(&inode->i_data, 0);
3731 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3732 root == root->fs_info->tree_root))
3735 if (is_bad_inode(inode)) {
3736 btrfs_orphan_del(NULL, inode);
3739 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3740 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3742 if (root->fs_info->log_root_recovering) {
3743 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3747 if (inode->i_nlink > 0) {
3748 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3752 btrfs_i_size_write(inode, 0);
3755 trans = btrfs_start_transaction(root, 0);
3756 BUG_ON(IS_ERR(trans));
3757 btrfs_set_trans_block_group(trans, inode);
3758 trans->block_rsv = root->orphan_block_rsv;
3760 ret = btrfs_block_rsv_check(trans, root,
3761 root->orphan_block_rsv, 0, 5);
3763 BUG_ON(ret != -EAGAIN);
3764 ret = btrfs_commit_transaction(trans, root);
3769 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3773 nr = trans->blocks_used;
3774 btrfs_end_transaction(trans, root);
3776 btrfs_btree_balance_dirty(root, nr);
3781 ret = btrfs_orphan_del(trans, inode);
3785 nr = trans->blocks_used;
3786 btrfs_end_transaction(trans, root);
3787 btrfs_btree_balance_dirty(root, nr);
3789 end_writeback(inode);
3794 * this returns the key found in the dir entry in the location pointer.
3795 * If no dir entries were found, location->objectid is 0.
3797 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3798 struct btrfs_key *location)
3800 const char *name = dentry->d_name.name;
3801 int namelen = dentry->d_name.len;
3802 struct btrfs_dir_item *di;
3803 struct btrfs_path *path;
3804 struct btrfs_root *root = BTRFS_I(dir)->root;
3807 path = btrfs_alloc_path();
3810 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3815 if (!di || IS_ERR(di))
3818 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3820 btrfs_free_path(path);
3823 location->objectid = 0;
3828 * when we hit a tree root in a directory, the btrfs part of the inode
3829 * needs to be changed to reflect the root directory of the tree root. This
3830 * is kind of like crossing a mount point.
3832 static int fixup_tree_root_location(struct btrfs_root *root,
3834 struct dentry *dentry,
3835 struct btrfs_key *location,
3836 struct btrfs_root **sub_root)
3838 struct btrfs_path *path;
3839 struct btrfs_root *new_root;
3840 struct btrfs_root_ref *ref;
3841 struct extent_buffer *leaf;
3845 path = btrfs_alloc_path();
3852 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3853 BTRFS_I(dir)->root->root_key.objectid,
3854 location->objectid);
3861 leaf = path->nodes[0];
3862 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3863 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3864 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3867 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3868 (unsigned long)(ref + 1),
3869 dentry->d_name.len);
3873 btrfs_release_path(root->fs_info->tree_root, path);
3875 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3876 if (IS_ERR(new_root)) {
3877 err = PTR_ERR(new_root);
3881 if (btrfs_root_refs(&new_root->root_item) == 0) {
3886 *sub_root = new_root;
3887 location->objectid = btrfs_root_dirid(&new_root->root_item);
3888 location->type = BTRFS_INODE_ITEM_KEY;
3889 location->offset = 0;
3892 btrfs_free_path(path);
3896 static void inode_tree_add(struct inode *inode)
3898 struct btrfs_root *root = BTRFS_I(inode)->root;
3899 struct btrfs_inode *entry;
3901 struct rb_node *parent;
3903 p = &root->inode_tree.rb_node;
3906 if (inode_unhashed(inode))
3909 spin_lock(&root->inode_lock);
3912 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3914 if (inode->i_ino < entry->vfs_inode.i_ino)
3915 p = &parent->rb_left;
3916 else if (inode->i_ino > entry->vfs_inode.i_ino)
3917 p = &parent->rb_right;
3919 WARN_ON(!(entry->vfs_inode.i_state &
3920 (I_WILL_FREE | I_FREEING)));
3921 rb_erase(parent, &root->inode_tree);
3922 RB_CLEAR_NODE(parent);
3923 spin_unlock(&root->inode_lock);
3927 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3928 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3929 spin_unlock(&root->inode_lock);
3932 static void inode_tree_del(struct inode *inode)
3934 struct btrfs_root *root = BTRFS_I(inode)->root;
3937 spin_lock(&root->inode_lock);
3938 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3939 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3940 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3941 empty = RB_EMPTY_ROOT(&root->inode_tree);
3943 spin_unlock(&root->inode_lock);
3946 * Free space cache has inodes in the tree root, but the tree root has a
3947 * root_refs of 0, so this could end up dropping the tree root as a
3948 * snapshot, so we need the extra !root->fs_info->tree_root check to
3949 * make sure we don't drop it.
3951 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3952 root != root->fs_info->tree_root) {
3953 synchronize_srcu(&root->fs_info->subvol_srcu);
3954 spin_lock(&root->inode_lock);
3955 empty = RB_EMPTY_ROOT(&root->inode_tree);
3956 spin_unlock(&root->inode_lock);
3958 btrfs_add_dead_root(root);
3962 int btrfs_invalidate_inodes(struct btrfs_root *root)
3964 struct rb_node *node;
3965 struct rb_node *prev;
3966 struct btrfs_inode *entry;
3967 struct inode *inode;
3970 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3972 spin_lock(&root->inode_lock);
3974 node = root->inode_tree.rb_node;
3978 entry = rb_entry(node, struct btrfs_inode, rb_node);
3980 if (objectid < entry->vfs_inode.i_ino)
3981 node = node->rb_left;
3982 else if (objectid > entry->vfs_inode.i_ino)
3983 node = node->rb_right;
3989 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3990 if (objectid <= entry->vfs_inode.i_ino) {
3994 prev = rb_next(prev);
3998 entry = rb_entry(node, struct btrfs_inode, rb_node);
3999 objectid = entry->vfs_inode.i_ino + 1;
4000 inode = igrab(&entry->vfs_inode);
4002 spin_unlock(&root->inode_lock);
4003 if (atomic_read(&inode->i_count) > 1)
4004 d_prune_aliases(inode);
4006 * btrfs_drop_inode will have it removed from
4007 * the inode cache when its usage count
4012 spin_lock(&root->inode_lock);
4016 if (cond_resched_lock(&root->inode_lock))
4019 node = rb_next(node);
4021 spin_unlock(&root->inode_lock);
4025 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4027 struct btrfs_iget_args *args = p;
4028 inode->i_ino = args->ino;
4029 BTRFS_I(inode)->root = args->root;
4030 btrfs_set_inode_space_info(args->root, inode);
4034 static int btrfs_find_actor(struct inode *inode, void *opaque)
4036 struct btrfs_iget_args *args = opaque;
4037 return args->ino == inode->i_ino &&
4038 args->root == BTRFS_I(inode)->root;
4041 static struct inode *btrfs_iget_locked(struct super_block *s,
4043 struct btrfs_root *root)
4045 struct inode *inode;
4046 struct btrfs_iget_args args;
4047 args.ino = objectid;
4050 inode = iget5_locked(s, objectid, btrfs_find_actor,
4051 btrfs_init_locked_inode,
4056 /* Get an inode object given its location and corresponding root.
4057 * Returns in *is_new if the inode was read from disk
4059 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4060 struct btrfs_root *root, int *new)
4062 struct inode *inode;
4064 inode = btrfs_iget_locked(s, location->objectid, root);
4066 return ERR_PTR(-ENOMEM);
4068 if (inode->i_state & I_NEW) {
4069 BTRFS_I(inode)->root = root;
4070 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4071 btrfs_read_locked_inode(inode);
4073 inode_tree_add(inode);
4074 unlock_new_inode(inode);
4082 static struct inode *new_simple_dir(struct super_block *s,
4083 struct btrfs_key *key,
4084 struct btrfs_root *root)
4086 struct inode *inode = new_inode(s);
4089 return ERR_PTR(-ENOMEM);
4091 BTRFS_I(inode)->root = root;
4092 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4093 BTRFS_I(inode)->dummy_inode = 1;
4095 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4096 inode->i_op = &simple_dir_inode_operations;
4097 inode->i_fop = &simple_dir_operations;
4098 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4099 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4104 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4106 struct inode *inode;
4107 struct btrfs_root *root = BTRFS_I(dir)->root;
4108 struct btrfs_root *sub_root = root;
4109 struct btrfs_key location;
4113 if (dentry->d_name.len > BTRFS_NAME_LEN)
4114 return ERR_PTR(-ENAMETOOLONG);
4116 ret = btrfs_inode_by_name(dir, dentry, &location);
4119 return ERR_PTR(ret);
4121 if (location.objectid == 0)
4124 if (location.type == BTRFS_INODE_ITEM_KEY) {
4125 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4129 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4131 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4132 ret = fixup_tree_root_location(root, dir, dentry,
4133 &location, &sub_root);
4136 inode = ERR_PTR(ret);
4138 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4140 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4142 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4144 if (!IS_ERR(inode) && root != sub_root) {
4145 down_read(&root->fs_info->cleanup_work_sem);
4146 if (!(inode->i_sb->s_flags & MS_RDONLY))
4147 btrfs_orphan_cleanup(sub_root);
4148 up_read(&root->fs_info->cleanup_work_sem);
4154 static int btrfs_dentry_delete(const struct dentry *dentry)
4156 struct btrfs_root *root;
4158 if (!dentry->d_inode && !IS_ROOT(dentry))
4159 dentry = dentry->d_parent;
4161 if (dentry->d_inode) {
4162 root = BTRFS_I(dentry->d_inode)->root;
4163 if (btrfs_root_refs(&root->root_item) == 0)
4169 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4170 struct nameidata *nd)
4172 struct inode *inode;
4174 inode = btrfs_lookup_dentry(dir, dentry);
4176 return ERR_CAST(inode);
4178 return d_splice_alias(inode, dentry);
4181 static unsigned char btrfs_filetype_table[] = {
4182 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4185 static int btrfs_real_readdir(struct file *filp, void *dirent,
4188 struct inode *inode = filp->f_dentry->d_inode;
4189 struct btrfs_root *root = BTRFS_I(inode)->root;
4190 struct btrfs_item *item;
4191 struct btrfs_dir_item *di;
4192 struct btrfs_key key;
4193 struct btrfs_key found_key;
4194 struct btrfs_path *path;
4197 struct extent_buffer *leaf;
4200 unsigned char d_type;
4205 int key_type = BTRFS_DIR_INDEX_KEY;
4210 /* FIXME, use a real flag for deciding about the key type */
4211 if (root->fs_info->tree_root == root)
4212 key_type = BTRFS_DIR_ITEM_KEY;
4214 /* special case for "." */
4215 if (filp->f_pos == 0) {
4216 over = filldir(dirent, ".", 1,
4223 /* special case for .., just use the back ref */
4224 if (filp->f_pos == 1) {
4225 u64 pino = parent_ino(filp->f_path.dentry);
4226 over = filldir(dirent, "..", 2,
4232 path = btrfs_alloc_path();
4235 btrfs_set_key_type(&key, key_type);
4236 key.offset = filp->f_pos;
4237 key.objectid = inode->i_ino;
4239 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4245 leaf = path->nodes[0];
4246 nritems = btrfs_header_nritems(leaf);
4247 slot = path->slots[0];
4248 if (advance || slot >= nritems) {
4249 if (slot >= nritems - 1) {
4250 ret = btrfs_next_leaf(root, path);
4253 leaf = path->nodes[0];
4254 nritems = btrfs_header_nritems(leaf);
4255 slot = path->slots[0];
4263 item = btrfs_item_nr(leaf, slot);
4264 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4266 if (found_key.objectid != key.objectid)
4268 if (btrfs_key_type(&found_key) != key_type)
4270 if (found_key.offset < filp->f_pos)
4273 filp->f_pos = found_key.offset;
4275 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4277 di_total = btrfs_item_size(leaf, item);
4279 while (di_cur < di_total) {
4280 struct btrfs_key location;
4282 name_len = btrfs_dir_name_len(leaf, di);
4283 if (name_len <= sizeof(tmp_name)) {
4284 name_ptr = tmp_name;
4286 name_ptr = kmalloc(name_len, GFP_NOFS);
4292 read_extent_buffer(leaf, name_ptr,
4293 (unsigned long)(di + 1), name_len);
4295 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4296 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4298 /* is this a reference to our own snapshot? If so
4301 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4302 location.objectid == root->root_key.objectid) {
4306 over = filldir(dirent, name_ptr, name_len,
4307 found_key.offset, location.objectid,
4311 if (name_ptr != tmp_name)
4316 di_len = btrfs_dir_name_len(leaf, di) +
4317 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4319 di = (struct btrfs_dir_item *)((char *)di + di_len);
4323 /* Reached end of directory/root. Bump pos past the last item. */
4324 if (key_type == BTRFS_DIR_INDEX_KEY)
4326 * 32-bit glibc will use getdents64, but then strtol -
4327 * so the last number we can serve is this.
4329 filp->f_pos = 0x7fffffff;
4335 btrfs_free_path(path);
4339 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4341 struct btrfs_root *root = BTRFS_I(inode)->root;
4342 struct btrfs_trans_handle *trans;
4344 bool nolock = false;
4346 if (BTRFS_I(inode)->dummy_inode)
4350 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4352 if (wbc->sync_mode == WB_SYNC_ALL) {
4354 trans = btrfs_join_transaction_nolock(root, 1);
4356 trans = btrfs_join_transaction(root, 1);
4358 return PTR_ERR(trans);
4359 btrfs_set_trans_block_group(trans, inode);
4361 ret = btrfs_end_transaction_nolock(trans, root);
4363 ret = btrfs_commit_transaction(trans, root);
4369 * This is somewhat expensive, updating the tree every time the
4370 * inode changes. But, it is most likely to find the inode in cache.
4371 * FIXME, needs more benchmarking...there are no reasons other than performance
4372 * to keep or drop this code.
4374 void btrfs_dirty_inode(struct inode *inode)
4376 struct btrfs_root *root = BTRFS_I(inode)->root;
4377 struct btrfs_trans_handle *trans;
4380 if (BTRFS_I(inode)->dummy_inode)
4383 trans = btrfs_join_transaction(root, 1);
4384 BUG_ON(IS_ERR(trans));
4385 btrfs_set_trans_block_group(trans, inode);
4387 ret = btrfs_update_inode(trans, root, inode);
4388 if (ret && ret == -ENOSPC) {
4389 /* whoops, lets try again with the full transaction */
4390 btrfs_end_transaction(trans, root);
4391 trans = btrfs_start_transaction(root, 1);
4392 if (IS_ERR(trans)) {
4393 if (printk_ratelimit()) {
4394 printk(KERN_ERR "btrfs: fail to "
4395 "dirty inode %lu error %ld\n",
4396 inode->i_ino, PTR_ERR(trans));
4400 btrfs_set_trans_block_group(trans, inode);
4402 ret = btrfs_update_inode(trans, root, inode);
4404 if (printk_ratelimit()) {
4405 printk(KERN_ERR "btrfs: fail to "
4406 "dirty inode %lu error %d\n",
4411 btrfs_end_transaction(trans, root);
4415 * find the highest existing sequence number in a directory
4416 * and then set the in-memory index_cnt variable to reflect
4417 * free sequence numbers
4419 static int btrfs_set_inode_index_count(struct inode *inode)
4421 struct btrfs_root *root = BTRFS_I(inode)->root;
4422 struct btrfs_key key, found_key;
4423 struct btrfs_path *path;
4424 struct extent_buffer *leaf;
4427 key.objectid = inode->i_ino;
4428 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4429 key.offset = (u64)-1;
4431 path = btrfs_alloc_path();
4435 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4438 /* FIXME: we should be able to handle this */
4444 * MAGIC NUMBER EXPLANATION:
4445 * since we search a directory based on f_pos we have to start at 2
4446 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4447 * else has to start at 2
4449 if (path->slots[0] == 0) {
4450 BTRFS_I(inode)->index_cnt = 2;
4456 leaf = path->nodes[0];
4457 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4459 if (found_key.objectid != inode->i_ino ||
4460 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4461 BTRFS_I(inode)->index_cnt = 2;
4465 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4467 btrfs_free_path(path);
4472 * helper to find a free sequence number in a given directory. This current
4473 * code is very simple, later versions will do smarter things in the btree
4475 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4479 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4480 ret = btrfs_set_inode_index_count(dir);
4485 *index = BTRFS_I(dir)->index_cnt;
4486 BTRFS_I(dir)->index_cnt++;
4491 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4492 struct btrfs_root *root,
4494 const char *name, int name_len,
4495 u64 ref_objectid, u64 objectid,
4496 u64 alloc_hint, int mode, u64 *index)
4498 struct inode *inode;
4499 struct btrfs_inode_item *inode_item;
4500 struct btrfs_key *location;
4501 struct btrfs_path *path;
4502 struct btrfs_inode_ref *ref;
4503 struct btrfs_key key[2];
4509 path = btrfs_alloc_path();
4512 inode = new_inode(root->fs_info->sb);
4514 return ERR_PTR(-ENOMEM);
4517 ret = btrfs_set_inode_index(dir, index);
4520 return ERR_PTR(ret);
4524 * index_cnt is ignored for everything but a dir,
4525 * btrfs_get_inode_index_count has an explanation for the magic
4528 BTRFS_I(inode)->index_cnt = 2;
4529 BTRFS_I(inode)->root = root;
4530 BTRFS_I(inode)->generation = trans->transid;
4531 inode->i_generation = BTRFS_I(inode)->generation;
4532 btrfs_set_inode_space_info(root, inode);
4538 BTRFS_I(inode)->block_group =
4539 btrfs_find_block_group(root, 0, alloc_hint, owner);
4541 key[0].objectid = objectid;
4542 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4545 key[1].objectid = objectid;
4546 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4547 key[1].offset = ref_objectid;
4549 sizes[0] = sizeof(struct btrfs_inode_item);
4550 sizes[1] = name_len + sizeof(*ref);
4552 path->leave_spinning = 1;
4553 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4557 inode_init_owner(inode, dir, mode);
4558 inode->i_ino = objectid;
4559 inode_set_bytes(inode, 0);
4560 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4561 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4562 struct btrfs_inode_item);
4563 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4565 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4566 struct btrfs_inode_ref);
4567 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4568 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4569 ptr = (unsigned long)(ref + 1);
4570 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4572 btrfs_mark_buffer_dirty(path->nodes[0]);
4573 btrfs_free_path(path);
4575 location = &BTRFS_I(inode)->location;
4576 location->objectid = objectid;
4577 location->offset = 0;
4578 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4580 btrfs_inherit_iflags(inode, dir);
4582 if ((mode & S_IFREG)) {
4583 if (btrfs_test_opt(root, NODATASUM))
4584 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4585 if (btrfs_test_opt(root, NODATACOW))
4586 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4589 insert_inode_hash(inode);
4590 inode_tree_add(inode);
4594 BTRFS_I(dir)->index_cnt--;
4595 btrfs_free_path(path);
4597 return ERR_PTR(ret);
4600 static inline u8 btrfs_inode_type(struct inode *inode)
4602 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4606 * utility function to add 'inode' into 'parent_inode' with
4607 * a give name and a given sequence number.
4608 * if 'add_backref' is true, also insert a backref from the
4609 * inode to the parent directory.
4611 int btrfs_add_link(struct btrfs_trans_handle *trans,
4612 struct inode *parent_inode, struct inode *inode,
4613 const char *name, int name_len, int add_backref, u64 index)
4616 struct btrfs_key key;
4617 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4619 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4620 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4622 key.objectid = inode->i_ino;
4623 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4627 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4628 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4629 key.objectid, root->root_key.objectid,
4630 parent_inode->i_ino,
4631 index, name, name_len);
4632 } else if (add_backref) {
4633 ret = btrfs_insert_inode_ref(trans, root,
4634 name, name_len, inode->i_ino,
4635 parent_inode->i_ino, index);
4639 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4640 parent_inode->i_ino, &key,
4641 btrfs_inode_type(inode), index);
4644 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4646 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4647 ret = btrfs_update_inode(trans, root, parent_inode);
4652 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4653 struct inode *dir, struct dentry *dentry,
4654 struct inode *inode, int backref, u64 index)
4656 int err = btrfs_add_link(trans, dir, inode,
4657 dentry->d_name.name, dentry->d_name.len,
4660 d_instantiate(dentry, inode);
4668 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4669 int mode, dev_t rdev)
4671 struct btrfs_trans_handle *trans;
4672 struct btrfs_root *root = BTRFS_I(dir)->root;
4673 struct inode *inode = NULL;
4677 unsigned long nr = 0;
4680 if (!new_valid_dev(rdev))
4683 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4688 * 2 for inode item and ref
4690 * 1 for xattr if selinux is on
4692 trans = btrfs_start_transaction(root, 5);
4694 return PTR_ERR(trans);
4696 btrfs_set_trans_block_group(trans, dir);
4698 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4699 dentry->d_name.len, dir->i_ino, objectid,
4700 BTRFS_I(dir)->block_group, mode, &index);
4701 err = PTR_ERR(inode);
4705 err = btrfs_init_inode_security(trans, inode, dir);
4711 btrfs_set_trans_block_group(trans, inode);
4712 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4716 inode->i_op = &btrfs_special_inode_operations;
4717 init_special_inode(inode, inode->i_mode, rdev);
4718 btrfs_update_inode(trans, root, inode);
4720 btrfs_update_inode_block_group(trans, inode);
4721 btrfs_update_inode_block_group(trans, dir);
4723 nr = trans->blocks_used;
4724 btrfs_end_transaction_throttle(trans, root);
4725 btrfs_btree_balance_dirty(root, nr);
4727 inode_dec_link_count(inode);
4733 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4734 int mode, struct nameidata *nd)
4736 struct btrfs_trans_handle *trans;
4737 struct btrfs_root *root = BTRFS_I(dir)->root;
4738 struct inode *inode = NULL;
4741 unsigned long nr = 0;
4745 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4749 * 2 for inode item and ref
4751 * 1 for xattr if selinux is on
4753 trans = btrfs_start_transaction(root, 5);
4755 return PTR_ERR(trans);
4757 btrfs_set_trans_block_group(trans, dir);
4759 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4760 dentry->d_name.len, dir->i_ino, objectid,
4761 BTRFS_I(dir)->block_group, mode, &index);
4762 err = PTR_ERR(inode);
4766 err = btrfs_init_inode_security(trans, inode, dir);
4772 btrfs_set_trans_block_group(trans, inode);
4773 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4777 inode->i_mapping->a_ops = &btrfs_aops;
4778 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4779 inode->i_fop = &btrfs_file_operations;
4780 inode->i_op = &btrfs_file_inode_operations;
4781 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4783 btrfs_update_inode_block_group(trans, inode);
4784 btrfs_update_inode_block_group(trans, dir);
4786 nr = trans->blocks_used;
4787 btrfs_end_transaction_throttle(trans, root);
4789 inode_dec_link_count(inode);
4792 btrfs_btree_balance_dirty(root, nr);
4796 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4797 struct dentry *dentry)
4799 struct btrfs_trans_handle *trans;
4800 struct btrfs_root *root = BTRFS_I(dir)->root;
4801 struct inode *inode = old_dentry->d_inode;
4803 unsigned long nr = 0;
4807 if (inode->i_nlink == 0)
4810 /* do not allow sys_link's with other subvols of the same device */
4811 if (root->objectid != BTRFS_I(inode)->root->objectid)
4814 btrfs_inc_nlink(inode);
4815 inode->i_ctime = CURRENT_TIME;
4817 err = btrfs_set_inode_index(dir, &index);
4822 * 2 items for inode and inode ref
4823 * 2 items for dir items
4824 * 1 item for parent inode
4826 trans = btrfs_start_transaction(root, 5);
4827 if (IS_ERR(trans)) {
4828 err = PTR_ERR(trans);
4832 btrfs_set_trans_block_group(trans, dir);
4835 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4840 struct dentry *parent = dget_parent(dentry);
4841 btrfs_update_inode_block_group(trans, dir);
4842 err = btrfs_update_inode(trans, root, inode);
4844 btrfs_log_new_name(trans, inode, NULL, parent);
4848 nr = trans->blocks_used;
4849 btrfs_end_transaction_throttle(trans, root);
4852 inode_dec_link_count(inode);
4855 btrfs_btree_balance_dirty(root, nr);
4859 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4861 struct inode *inode = NULL;
4862 struct btrfs_trans_handle *trans;
4863 struct btrfs_root *root = BTRFS_I(dir)->root;
4865 int drop_on_err = 0;
4868 unsigned long nr = 1;
4870 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4875 * 2 items for inode and ref
4876 * 2 items for dir items
4877 * 1 for xattr if selinux is on
4879 trans = btrfs_start_transaction(root, 5);
4881 return PTR_ERR(trans);
4882 btrfs_set_trans_block_group(trans, dir);
4884 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4885 dentry->d_name.len, dir->i_ino, objectid,
4886 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4888 if (IS_ERR(inode)) {
4889 err = PTR_ERR(inode);
4895 err = btrfs_init_inode_security(trans, inode, dir);
4899 inode->i_op = &btrfs_dir_inode_operations;
4900 inode->i_fop = &btrfs_dir_file_operations;
4901 btrfs_set_trans_block_group(trans, inode);
4903 btrfs_i_size_write(inode, 0);
4904 err = btrfs_update_inode(trans, root, inode);
4908 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4909 dentry->d_name.len, 0, index);
4913 d_instantiate(dentry, inode);
4915 btrfs_update_inode_block_group(trans, inode);
4916 btrfs_update_inode_block_group(trans, dir);
4919 nr = trans->blocks_used;
4920 btrfs_end_transaction_throttle(trans, root);
4923 btrfs_btree_balance_dirty(root, nr);
4927 /* helper for btfs_get_extent. Given an existing extent in the tree,
4928 * and an extent that you want to insert, deal with overlap and insert
4929 * the new extent into the tree.
4931 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4932 struct extent_map *existing,
4933 struct extent_map *em,
4934 u64 map_start, u64 map_len)
4938 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4939 start_diff = map_start - em->start;
4940 em->start = map_start;
4942 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4943 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4944 em->block_start += start_diff;
4945 em->block_len -= start_diff;
4947 return add_extent_mapping(em_tree, em);
4950 static noinline int uncompress_inline(struct btrfs_path *path,
4951 struct inode *inode, struct page *page,
4952 size_t pg_offset, u64 extent_offset,
4953 struct btrfs_file_extent_item *item)
4956 struct extent_buffer *leaf = path->nodes[0];
4959 unsigned long inline_size;
4963 WARN_ON(pg_offset != 0);
4964 compress_type = btrfs_file_extent_compression(leaf, item);
4965 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4966 inline_size = btrfs_file_extent_inline_item_len(leaf,
4967 btrfs_item_nr(leaf, path->slots[0]));
4968 tmp = kmalloc(inline_size, GFP_NOFS);
4969 ptr = btrfs_file_extent_inline_start(item);
4971 read_extent_buffer(leaf, tmp, ptr, inline_size);
4973 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4974 ret = btrfs_decompress(compress_type, tmp, page,
4975 extent_offset, inline_size, max_size);
4977 char *kaddr = kmap_atomic(page, KM_USER0);
4978 unsigned long copy_size = min_t(u64,
4979 PAGE_CACHE_SIZE - pg_offset,
4980 max_size - extent_offset);
4981 memset(kaddr + pg_offset, 0, copy_size);
4982 kunmap_atomic(kaddr, KM_USER0);
4989 * a bit scary, this does extent mapping from logical file offset to the disk.
4990 * the ugly parts come from merging extents from the disk with the in-ram
4991 * representation. This gets more complex because of the data=ordered code,
4992 * where the in-ram extents might be locked pending data=ordered completion.
4994 * This also copies inline extents directly into the page.
4997 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4998 size_t pg_offset, u64 start, u64 len,
5004 u64 extent_start = 0;
5006 u64 objectid = inode->i_ino;
5008 struct btrfs_path *path = NULL;
5009 struct btrfs_root *root = BTRFS_I(inode)->root;
5010 struct btrfs_file_extent_item *item;
5011 struct extent_buffer *leaf;
5012 struct btrfs_key found_key;
5013 struct extent_map *em = NULL;
5014 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5015 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5016 struct btrfs_trans_handle *trans = NULL;
5020 read_lock(&em_tree->lock);
5021 em = lookup_extent_mapping(em_tree, start, len);
5023 em->bdev = root->fs_info->fs_devices->latest_bdev;
5024 read_unlock(&em_tree->lock);
5027 if (em->start > start || em->start + em->len <= start)
5028 free_extent_map(em);
5029 else if (em->block_start == EXTENT_MAP_INLINE && page)
5030 free_extent_map(em);
5034 em = alloc_extent_map(GFP_NOFS);
5039 em->bdev = root->fs_info->fs_devices->latest_bdev;
5040 em->start = EXTENT_MAP_HOLE;
5041 em->orig_start = EXTENT_MAP_HOLE;
5043 em->block_len = (u64)-1;
5046 path = btrfs_alloc_path();
5050 ret = btrfs_lookup_file_extent(trans, root, path,
5051 objectid, start, trans != NULL);
5058 if (path->slots[0] == 0)
5063 leaf = path->nodes[0];
5064 item = btrfs_item_ptr(leaf, path->slots[0],
5065 struct btrfs_file_extent_item);
5066 /* are we inside the extent that was found? */
5067 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5068 found_type = btrfs_key_type(&found_key);
5069 if (found_key.objectid != objectid ||
5070 found_type != BTRFS_EXTENT_DATA_KEY) {
5074 found_type = btrfs_file_extent_type(leaf, item);
5075 extent_start = found_key.offset;
5076 compress_type = btrfs_file_extent_compression(leaf, item);
5077 if (found_type == BTRFS_FILE_EXTENT_REG ||
5078 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5079 extent_end = extent_start +
5080 btrfs_file_extent_num_bytes(leaf, item);
5081 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5083 size = btrfs_file_extent_inline_len(leaf, item);
5084 extent_end = (extent_start + size + root->sectorsize - 1) &
5085 ~((u64)root->sectorsize - 1);
5088 if (start >= extent_end) {
5090 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5091 ret = btrfs_next_leaf(root, path);
5098 leaf = path->nodes[0];
5100 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5101 if (found_key.objectid != objectid ||
5102 found_key.type != BTRFS_EXTENT_DATA_KEY)
5104 if (start + len <= found_key.offset)
5107 em->len = found_key.offset - start;
5111 if (found_type == BTRFS_FILE_EXTENT_REG ||
5112 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5113 em->start = extent_start;
5114 em->len = extent_end - extent_start;
5115 em->orig_start = extent_start -
5116 btrfs_file_extent_offset(leaf, item);
5117 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5119 em->block_start = EXTENT_MAP_HOLE;
5122 if (compress_type != BTRFS_COMPRESS_NONE) {
5123 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5124 em->compress_type = compress_type;
5125 em->block_start = bytenr;
5126 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5129 bytenr += btrfs_file_extent_offset(leaf, item);
5130 em->block_start = bytenr;
5131 em->block_len = em->len;
5132 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5133 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5136 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5140 size_t extent_offset;
5143 em->block_start = EXTENT_MAP_INLINE;
5144 if (!page || create) {
5145 em->start = extent_start;
5146 em->len = extent_end - extent_start;
5150 size = btrfs_file_extent_inline_len(leaf, item);
5151 extent_offset = page_offset(page) + pg_offset - extent_start;
5152 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5153 size - extent_offset);
5154 em->start = extent_start + extent_offset;
5155 em->len = (copy_size + root->sectorsize - 1) &
5156 ~((u64)root->sectorsize - 1);
5157 em->orig_start = EXTENT_MAP_INLINE;
5158 if (compress_type) {
5159 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5160 em->compress_type = compress_type;
5162 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5163 if (create == 0 && !PageUptodate(page)) {
5164 if (btrfs_file_extent_compression(leaf, item) !=
5165 BTRFS_COMPRESS_NONE) {
5166 ret = uncompress_inline(path, inode, page,
5168 extent_offset, item);
5172 read_extent_buffer(leaf, map + pg_offset, ptr,
5174 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5175 memset(map + pg_offset + copy_size, 0,
5176 PAGE_CACHE_SIZE - pg_offset -
5181 flush_dcache_page(page);
5182 } else if (create && PageUptodate(page)) {
5186 free_extent_map(em);
5188 btrfs_release_path(root, path);
5189 trans = btrfs_join_transaction(root, 1);
5191 return ERR_CAST(trans);
5195 write_extent_buffer(leaf, map + pg_offset, ptr,
5198 btrfs_mark_buffer_dirty(leaf);
5200 set_extent_uptodate(io_tree, em->start,
5201 extent_map_end(em) - 1, GFP_NOFS);
5204 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5211 em->block_start = EXTENT_MAP_HOLE;
5212 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5214 btrfs_release_path(root, path);
5215 if (em->start > start || extent_map_end(em) <= start) {
5216 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5217 "[%llu %llu]\n", (unsigned long long)em->start,
5218 (unsigned long long)em->len,
5219 (unsigned long long)start,
5220 (unsigned long long)len);
5226 write_lock(&em_tree->lock);
5227 ret = add_extent_mapping(em_tree, em);
5228 /* it is possible that someone inserted the extent into the tree
5229 * while we had the lock dropped. It is also possible that
5230 * an overlapping map exists in the tree
5232 if (ret == -EEXIST) {
5233 struct extent_map *existing;
5237 existing = lookup_extent_mapping(em_tree, start, len);
5238 if (existing && (existing->start > start ||
5239 existing->start + existing->len <= start)) {
5240 free_extent_map(existing);
5244 existing = lookup_extent_mapping(em_tree, em->start,
5247 err = merge_extent_mapping(em_tree, existing,
5250 free_extent_map(existing);
5252 free_extent_map(em);
5257 free_extent_map(em);
5261 free_extent_map(em);
5266 write_unlock(&em_tree->lock);
5269 btrfs_free_path(path);
5271 ret = btrfs_end_transaction(trans, root);
5276 free_extent_map(em);
5277 return ERR_PTR(err);
5282 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5283 size_t pg_offset, u64 start, u64 len,
5286 struct extent_map *em;
5287 struct extent_map *hole_em = NULL;
5288 u64 range_start = start;
5294 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5299 * if our em maps to a hole, there might
5300 * actually be delalloc bytes behind it
5302 if (em->block_start != EXTENT_MAP_HOLE)
5308 /* check to see if we've wrapped (len == -1 or similar) */
5317 /* ok, we didn't find anything, lets look for delalloc */
5318 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5319 end, len, EXTENT_DELALLOC, 1);
5320 found_end = range_start + found;
5321 if (found_end < range_start)
5322 found_end = (u64)-1;
5325 * we didn't find anything useful, return
5326 * the original results from get_extent()
5328 if (range_start > end || found_end <= start) {
5334 /* adjust the range_start to make sure it doesn't
5335 * go backwards from the start they passed in
5337 range_start = max(start,range_start);
5338 found = found_end - range_start;
5341 u64 hole_start = start;
5344 em = alloc_extent_map(GFP_NOFS);
5350 * when btrfs_get_extent can't find anything it
5351 * returns one huge hole
5353 * make sure what it found really fits our range, and
5354 * adjust to make sure it is based on the start from
5358 u64 calc_end = extent_map_end(hole_em);
5360 if (calc_end <= start || (hole_em->start > end)) {
5361 free_extent_map(hole_em);
5364 hole_start = max(hole_em->start, start);
5365 hole_len = calc_end - hole_start;
5369 if (hole_em && range_start > hole_start) {
5370 /* our hole starts before our delalloc, so we
5371 * have to return just the parts of the hole
5372 * that go until the delalloc starts
5374 em->len = min(hole_len,
5375 range_start - hole_start);
5376 em->start = hole_start;
5377 em->orig_start = hole_start;
5379 * don't adjust block start at all,
5380 * it is fixed at EXTENT_MAP_HOLE
5382 em->block_start = hole_em->block_start;
5383 em->block_len = hole_len;
5385 em->start = range_start;
5387 em->orig_start = range_start;
5388 em->block_start = EXTENT_MAP_DELALLOC;
5389 em->block_len = found;
5391 } else if (hole_em) {
5396 free_extent_map(hole_em);
5398 free_extent_map(em);
5399 return ERR_PTR(err);
5404 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5407 struct btrfs_root *root = BTRFS_I(inode)->root;
5408 struct btrfs_trans_handle *trans;
5409 struct extent_map *em;
5410 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5411 struct btrfs_key ins;
5415 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5417 trans = btrfs_join_transaction(root, 0);
5419 return ERR_CAST(trans);
5421 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5423 alloc_hint = get_extent_allocation_hint(inode, start, len);
5424 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5425 alloc_hint, (u64)-1, &ins, 1);
5431 em = alloc_extent_map(GFP_NOFS);
5433 em = ERR_PTR(-ENOMEM);
5438 em->orig_start = em->start;
5439 em->len = ins.offset;
5441 em->block_start = ins.objectid;
5442 em->block_len = ins.offset;
5443 em->bdev = root->fs_info->fs_devices->latest_bdev;
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, inode->i_ino,
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 != inode->i_ino ||
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, inode->i_ino,
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, GFP_NOFS);
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, 0);
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 free_extent_map(em);
5665 em = btrfs_new_extent_direct(inode, start, len);
5668 len = min(len, em->len - (start - em->start));
5670 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5671 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5674 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5676 bh_result->b_size = len;
5677 bh_result->b_bdev = em->bdev;
5678 set_buffer_mapped(bh_result);
5679 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5680 set_buffer_new(bh_result);
5682 free_extent_map(em);
5687 struct btrfs_dio_private {
5688 struct inode *inode;
5695 /* number of bios pending for this dio */
5696 atomic_t pending_bios;
5701 struct bio *orig_bio;
5704 static void btrfs_endio_direct_read(struct bio *bio, int err)
5706 struct btrfs_dio_private *dip = bio->bi_private;
5707 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5708 struct bio_vec *bvec = bio->bi_io_vec;
5709 struct inode *inode = dip->inode;
5710 struct btrfs_root *root = BTRFS_I(inode)->root;
5712 u32 *private = dip->csums;
5714 start = dip->logical_offset;
5716 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5717 struct page *page = bvec->bv_page;
5720 unsigned long flags;
5722 local_irq_save(flags);
5723 kaddr = kmap_atomic(page, KM_IRQ0);
5724 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5725 csum, bvec->bv_len);
5726 btrfs_csum_final(csum, (char *)&csum);
5727 kunmap_atomic(kaddr, KM_IRQ0);
5728 local_irq_restore(flags);
5730 flush_dcache_page(bvec->bv_page);
5731 if (csum != *private) {
5732 printk(KERN_ERR "btrfs csum failed ino %lu off"
5733 " %llu csum %u private %u\n",
5734 inode->i_ino, (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, GFP_NOFS);
5747 bio->bi_private = dip->private;
5751 dio_end_io(bio, err);
5754 static void btrfs_endio_direct_write(struct bio *bio, int err)
5756 struct btrfs_dio_private *dip = bio->bi_private;
5757 struct inode *inode = dip->inode;
5758 struct btrfs_root *root = BTRFS_I(inode)->root;
5759 struct btrfs_trans_handle *trans;
5760 struct btrfs_ordered_extent *ordered = NULL;
5761 struct extent_state *cached_state = NULL;
5762 u64 ordered_offset = dip->logical_offset;
5763 u64 ordered_bytes = dip->bytes;
5769 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5777 trans = btrfs_join_transaction(root, 1);
5778 if (IS_ERR(trans)) {
5782 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5784 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5785 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5787 ret = btrfs_update_inode(trans, root, inode);
5792 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5793 ordered->file_offset + ordered->len - 1, 0,
5794 &cached_state, GFP_NOFS);
5796 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5797 ret = btrfs_mark_extent_written(trans, inode,
5798 ordered->file_offset,
5799 ordered->file_offset +
5806 ret = insert_reserved_file_extent(trans, inode,
5807 ordered->file_offset,
5813 BTRFS_FILE_EXTENT_REG);
5814 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5815 ordered->file_offset, ordered->len);
5823 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5824 btrfs_ordered_update_i_size(inode, 0, ordered);
5825 btrfs_update_inode(trans, root, inode);
5827 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5828 ordered->file_offset + ordered->len - 1,
5829 &cached_state, GFP_NOFS);
5831 btrfs_delalloc_release_metadata(inode, ordered->len);
5832 btrfs_end_transaction(trans, root);
5833 ordered_offset = ordered->file_offset + ordered->len;
5834 btrfs_put_ordered_extent(ordered);
5835 btrfs_put_ordered_extent(ordered);
5839 * our bio might span multiple ordered extents. If we haven't
5840 * completed the accounting for the whole dio, go back and try again
5842 if (ordered_offset < dip->logical_offset + dip->bytes) {
5843 ordered_bytes = dip->logical_offset + dip->bytes -
5848 bio->bi_private = dip->private;
5852 dio_end_io(bio, err);
5855 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5856 struct bio *bio, int mirror_num,
5857 unsigned long bio_flags, u64 offset)
5860 struct btrfs_root *root = BTRFS_I(inode)->root;
5861 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5866 static void btrfs_end_dio_bio(struct bio *bio, int err)
5868 struct btrfs_dio_private *dip = bio->bi_private;
5871 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5872 "sector %#Lx len %u err no %d\n",
5873 dip->inode->i_ino, bio->bi_rw,
5874 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5878 * before atomic variable goto zero, we must make sure
5879 * dip->errors is perceived to be set.
5881 smp_mb__before_atomic_dec();
5884 /* if there are more bios still pending for this dio, just exit */
5885 if (!atomic_dec_and_test(&dip->pending_bios))
5889 bio_io_error(dip->orig_bio);
5891 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5892 bio_endio(dip->orig_bio, 0);
5898 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5899 u64 first_sector, gfp_t gfp_flags)
5901 int nr_vecs = bio_get_nr_vecs(bdev);
5902 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5905 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5906 int rw, u64 file_offset, int skip_sum,
5909 int write = rw & REQ_WRITE;
5910 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5918 if (write && !skip_sum) {
5919 ret = btrfs_wq_submit_bio(root->fs_info,
5920 inode, rw, bio, 0, 0,
5922 __btrfs_submit_bio_start_direct_io,
5923 __btrfs_submit_bio_done);
5925 } else if (!skip_sum)
5926 btrfs_lookup_bio_sums_dio(root, inode, bio,
5927 file_offset, csums);
5929 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5935 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5938 struct inode *inode = dip->inode;
5939 struct btrfs_root *root = BTRFS_I(inode)->root;
5940 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5942 struct bio *orig_bio = dip->orig_bio;
5943 struct bio_vec *bvec = orig_bio->bi_io_vec;
5944 u64 start_sector = orig_bio->bi_sector;
5945 u64 file_offset = dip->logical_offset;
5949 u32 *csums = dip->csums;
5952 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5955 bio->bi_private = dip;
5956 bio->bi_end_io = btrfs_end_dio_bio;
5957 atomic_inc(&dip->pending_bios);
5959 map_length = orig_bio->bi_size;
5960 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5961 &map_length, NULL, 0);
5967 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5968 if (unlikely(map_length < submit_len + bvec->bv_len ||
5969 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5970 bvec->bv_offset) < bvec->bv_len)) {
5972 * inc the count before we submit the bio so
5973 * we know the end IO handler won't happen before
5974 * we inc the count. Otherwise, the dip might get freed
5975 * before we're done setting it up
5977 atomic_inc(&dip->pending_bios);
5978 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5979 file_offset, skip_sum,
5983 atomic_dec(&dip->pending_bios);
5988 csums = csums + nr_pages;
5989 start_sector += submit_len >> 9;
5990 file_offset += submit_len;
5995 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5996 start_sector, GFP_NOFS);
5999 bio->bi_private = dip;
6000 bio->bi_end_io = btrfs_end_dio_bio;
6002 map_length = orig_bio->bi_size;
6003 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6004 &map_length, NULL, 0);
6010 submit_len += bvec->bv_len;
6016 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6025 * before atomic variable goto zero, we must
6026 * make sure dip->errors is perceived to be set.
6028 smp_mb__before_atomic_dec();
6029 if (atomic_dec_and_test(&dip->pending_bios))
6030 bio_io_error(dip->orig_bio);
6032 /* bio_end_io() will handle error, so we needn't return it */
6036 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6039 struct btrfs_root *root = BTRFS_I(inode)->root;
6040 struct btrfs_dio_private *dip;
6041 struct bio_vec *bvec = bio->bi_io_vec;
6043 int write = rw & REQ_WRITE;
6046 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6048 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6056 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6064 dip->private = bio->bi_private;
6066 dip->logical_offset = file_offset;
6070 dip->bytes += bvec->bv_len;
6072 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6074 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6075 bio->bi_private = dip;
6077 dip->orig_bio = bio;
6078 atomic_set(&dip->pending_bios, 0);
6081 bio->bi_end_io = btrfs_endio_direct_write;
6083 bio->bi_end_io = btrfs_endio_direct_read;
6085 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6090 * If this is a write, we need to clean up the reserved space and kill
6091 * the ordered extent.
6094 struct btrfs_ordered_extent *ordered;
6095 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6096 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6097 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6098 btrfs_free_reserved_extent(root, ordered->start,
6100 btrfs_put_ordered_extent(ordered);
6101 btrfs_put_ordered_extent(ordered);
6103 bio_endio(bio, ret);
6106 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6107 const struct iovec *iov, loff_t offset,
6108 unsigned long nr_segs)
6113 unsigned blocksize_mask = root->sectorsize - 1;
6114 ssize_t retval = -EINVAL;
6115 loff_t end = offset;
6117 if (offset & blocksize_mask)
6120 /* Check the memory alignment. Blocks cannot straddle pages */
6121 for (seg = 0; seg < nr_segs; seg++) {
6122 addr = (unsigned long)iov[seg].iov_base;
6123 size = iov[seg].iov_len;
6125 if ((addr & blocksize_mask) || (size & blocksize_mask))
6132 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6133 const struct iovec *iov, loff_t offset,
6134 unsigned long nr_segs)
6136 struct file *file = iocb->ki_filp;
6137 struct inode *inode = file->f_mapping->host;
6138 struct btrfs_ordered_extent *ordered;
6139 struct extent_state *cached_state = NULL;
6140 u64 lockstart, lockend;
6142 int writing = rw & WRITE;
6144 size_t count = iov_length(iov, nr_segs);
6146 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6152 lockend = offset + count - 1;
6155 ret = btrfs_delalloc_reserve_space(inode, count);
6161 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6162 0, &cached_state, GFP_NOFS);
6164 * We're concerned with the entire range that we're going to be
6165 * doing DIO to, so we need to make sure theres no ordered
6166 * extents in this range.
6168 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6169 lockend - lockstart + 1);
6172 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6173 &cached_state, GFP_NOFS);
6174 btrfs_start_ordered_extent(inode, ordered, 1);
6175 btrfs_put_ordered_extent(ordered);
6180 * we don't use btrfs_set_extent_delalloc because we don't want
6181 * the dirty or uptodate bits
6184 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6185 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6186 EXTENT_DELALLOC, 0, NULL, &cached_state,
6189 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6190 lockend, EXTENT_LOCKED | write_bits,
6191 1, 0, &cached_state, GFP_NOFS);
6196 free_extent_state(cached_state);
6197 cached_state = NULL;
6199 ret = __blockdev_direct_IO(rw, iocb, inode,
6200 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6201 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6202 btrfs_submit_direct, 0);
6204 if (ret < 0 && ret != -EIOCBQUEUED) {
6205 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6206 offset + iov_length(iov, nr_segs) - 1,
6207 EXTENT_LOCKED | write_bits, 1, 0,
6208 &cached_state, GFP_NOFS);
6209 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6211 * We're falling back to buffered, unlock the section we didn't
6214 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6215 offset + iov_length(iov, nr_segs) - 1,
6216 EXTENT_LOCKED | write_bits, 1, 0,
6217 &cached_state, GFP_NOFS);
6220 free_extent_state(cached_state);
6224 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6225 __u64 start, __u64 len)
6227 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6230 int btrfs_readpage(struct file *file, struct page *page)
6232 struct extent_io_tree *tree;
6233 tree = &BTRFS_I(page->mapping->host)->io_tree;
6234 return extent_read_full_page(tree, page, btrfs_get_extent);
6237 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6239 struct extent_io_tree *tree;
6242 if (current->flags & PF_MEMALLOC) {
6243 redirty_page_for_writepage(wbc, page);
6247 tree = &BTRFS_I(page->mapping->host)->io_tree;
6248 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6251 int btrfs_writepages(struct address_space *mapping,
6252 struct writeback_control *wbc)
6254 struct extent_io_tree *tree;
6256 tree = &BTRFS_I(mapping->host)->io_tree;
6257 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6261 btrfs_readpages(struct file *file, struct address_space *mapping,
6262 struct list_head *pages, unsigned nr_pages)
6264 struct extent_io_tree *tree;
6265 tree = &BTRFS_I(mapping->host)->io_tree;
6266 return extent_readpages(tree, mapping, pages, nr_pages,
6269 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6271 struct extent_io_tree *tree;
6272 struct extent_map_tree *map;
6275 tree = &BTRFS_I(page->mapping->host)->io_tree;
6276 map = &BTRFS_I(page->mapping->host)->extent_tree;
6277 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6279 ClearPagePrivate(page);
6280 set_page_private(page, 0);
6281 page_cache_release(page);
6286 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6288 if (PageWriteback(page) || PageDirty(page))
6290 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6293 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6295 struct extent_io_tree *tree;
6296 struct btrfs_ordered_extent *ordered;
6297 struct extent_state *cached_state = NULL;
6298 u64 page_start = page_offset(page);
6299 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6303 * we have the page locked, so new writeback can't start,
6304 * and the dirty bit won't be cleared while we are here.
6306 * Wait for IO on this page so that we can safely clear
6307 * the PagePrivate2 bit and do ordered accounting
6309 wait_on_page_writeback(page);
6311 tree = &BTRFS_I(page->mapping->host)->io_tree;
6313 btrfs_releasepage(page, GFP_NOFS);
6316 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6318 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6322 * IO on this page will never be started, so we need
6323 * to account for any ordered extents now
6325 clear_extent_bit(tree, page_start, page_end,
6326 EXTENT_DIRTY | EXTENT_DELALLOC |
6327 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6328 &cached_state, GFP_NOFS);
6330 * whoever cleared the private bit is responsible
6331 * for the finish_ordered_io
6333 if (TestClearPagePrivate2(page)) {
6334 btrfs_finish_ordered_io(page->mapping->host,
6335 page_start, page_end);
6337 btrfs_put_ordered_extent(ordered);
6338 cached_state = NULL;
6339 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6342 clear_extent_bit(tree, page_start, page_end,
6343 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6344 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6345 __btrfs_releasepage(page, GFP_NOFS);
6347 ClearPageChecked(page);
6348 if (PagePrivate(page)) {
6349 ClearPagePrivate(page);
6350 set_page_private(page, 0);
6351 page_cache_release(page);
6356 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6357 * called from a page fault handler when a page is first dirtied. Hence we must
6358 * be careful to check for EOF conditions here. We set the page up correctly
6359 * for a written page which means we get ENOSPC checking when writing into
6360 * holes and correct delalloc and unwritten extent mapping on filesystems that
6361 * support these features.
6363 * We are not allowed to take the i_mutex here so we have to play games to
6364 * protect against truncate races as the page could now be beyond EOF. Because
6365 * vmtruncate() writes the inode size before removing pages, once we have the
6366 * page lock we can determine safely if the page is beyond EOF. If it is not
6367 * beyond EOF, then the page is guaranteed safe against truncation until we
6370 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6372 struct page *page = vmf->page;
6373 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6374 struct btrfs_root *root = BTRFS_I(inode)->root;
6375 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6376 struct btrfs_ordered_extent *ordered;
6377 struct extent_state *cached_state = NULL;
6379 unsigned long zero_start;
6385 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6389 else /* -ENOSPC, -EIO, etc */
6390 ret = VM_FAULT_SIGBUS;
6394 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6397 size = i_size_read(inode);
6398 page_start = page_offset(page);
6399 page_end = page_start + PAGE_CACHE_SIZE - 1;
6401 if ((page->mapping != inode->i_mapping) ||
6402 (page_start >= size)) {
6403 /* page got truncated out from underneath us */
6406 wait_on_page_writeback(page);
6408 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6410 set_page_extent_mapped(page);
6413 * we can't set the delalloc bits if there are pending ordered
6414 * extents. Drop our locks and wait for them to finish
6416 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6418 unlock_extent_cached(io_tree, page_start, page_end,
6419 &cached_state, GFP_NOFS);
6421 btrfs_start_ordered_extent(inode, ordered, 1);
6422 btrfs_put_ordered_extent(ordered);
6427 * XXX - page_mkwrite gets called every time the page is dirtied, even
6428 * if it was already dirty, so for space accounting reasons we need to
6429 * clear any delalloc bits for the range we are fixing to save. There
6430 * is probably a better way to do this, but for now keep consistent with
6431 * prepare_pages in the normal write path.
6433 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6434 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6435 0, 0, &cached_state, GFP_NOFS);
6437 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6440 unlock_extent_cached(io_tree, page_start, page_end,
6441 &cached_state, GFP_NOFS);
6442 ret = VM_FAULT_SIGBUS;
6447 /* page is wholly or partially inside EOF */
6448 if (page_start + PAGE_CACHE_SIZE > size)
6449 zero_start = size & ~PAGE_CACHE_MASK;
6451 zero_start = PAGE_CACHE_SIZE;
6453 if (zero_start != PAGE_CACHE_SIZE) {
6455 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6456 flush_dcache_page(page);
6459 ClearPageChecked(page);
6460 set_page_dirty(page);
6461 SetPageUptodate(page);
6463 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6464 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6466 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6470 return VM_FAULT_LOCKED;
6472 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6477 static int btrfs_truncate(struct inode *inode)
6479 struct btrfs_root *root = BTRFS_I(inode)->root;
6481 struct btrfs_trans_handle *trans;
6483 u64 mask = root->sectorsize - 1;
6485 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6489 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6490 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6492 trans = btrfs_start_transaction(root, 0);
6493 BUG_ON(IS_ERR(trans));
6494 btrfs_set_trans_block_group(trans, inode);
6495 trans->block_rsv = root->orphan_block_rsv;
6498 * setattr is responsible for setting the ordered_data_close flag,
6499 * but that is only tested during the last file release. That
6500 * could happen well after the next commit, leaving a great big
6501 * window where new writes may get lost if someone chooses to write
6502 * to this file after truncating to zero
6504 * The inode doesn't have any dirty data here, and so if we commit
6505 * this is a noop. If someone immediately starts writing to the inode
6506 * it is very likely we'll catch some of their writes in this
6507 * transaction, and the commit will find this file on the ordered
6508 * data list with good things to send down.
6510 * This is a best effort solution, there is still a window where
6511 * using truncate to replace the contents of the file will
6512 * end up with a zero length file after a crash.
6514 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6515 btrfs_add_ordered_operation(trans, root, inode);
6519 trans = btrfs_start_transaction(root, 0);
6520 BUG_ON(IS_ERR(trans));
6521 btrfs_set_trans_block_group(trans, inode);
6522 trans->block_rsv = root->orphan_block_rsv;
6525 ret = btrfs_block_rsv_check(trans, root,
6526 root->orphan_block_rsv, 0, 5);
6528 BUG_ON(ret != -EAGAIN);
6529 ret = btrfs_commit_transaction(trans, root);
6535 ret = btrfs_truncate_inode_items(trans, root, inode,
6537 BTRFS_EXTENT_DATA_KEY);
6541 ret = btrfs_update_inode(trans, root, inode);
6544 nr = trans->blocks_used;
6545 btrfs_end_transaction(trans, root);
6547 btrfs_btree_balance_dirty(root, nr);
6550 if (ret == 0 && inode->i_nlink > 0) {
6551 ret = btrfs_orphan_del(trans, inode);
6555 ret = btrfs_update_inode(trans, root, inode);
6558 nr = trans->blocks_used;
6559 ret = btrfs_end_transaction_throttle(trans, root);
6561 btrfs_btree_balance_dirty(root, nr);
6567 * create a new subvolume directory/inode (helper for the ioctl).
6569 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6570 struct btrfs_root *new_root,
6571 u64 new_dirid, u64 alloc_hint)
6573 struct inode *inode;
6577 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6578 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6580 return PTR_ERR(inode);
6581 inode->i_op = &btrfs_dir_inode_operations;
6582 inode->i_fop = &btrfs_dir_file_operations;
6585 btrfs_i_size_write(inode, 0);
6587 err = btrfs_update_inode(trans, new_root, inode);
6594 /* helper function for file defrag and space balancing. This
6595 * forces readahead on a given range of bytes in an inode
6597 unsigned long btrfs_force_ra(struct address_space *mapping,
6598 struct file_ra_state *ra, struct file *file,
6599 pgoff_t offset, pgoff_t last_index)
6601 pgoff_t req_size = last_index - offset + 1;
6603 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6604 return offset + req_size;
6607 struct inode *btrfs_alloc_inode(struct super_block *sb)
6609 struct btrfs_inode *ei;
6610 struct inode *inode;
6612 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6617 ei->space_info = NULL;
6621 ei->last_sub_trans = 0;
6622 ei->logged_trans = 0;
6623 ei->delalloc_bytes = 0;
6624 ei->reserved_bytes = 0;
6625 ei->disk_i_size = 0;
6627 ei->index_cnt = (u64)-1;
6628 ei->last_unlink_trans = 0;
6630 atomic_set(&ei->outstanding_extents, 0);
6631 atomic_set(&ei->reserved_extents, 0);
6633 ei->ordered_data_close = 0;
6634 ei->orphan_meta_reserved = 0;
6635 ei->dummy_inode = 0;
6636 ei->force_compress = BTRFS_COMPRESS_NONE;
6638 inode = &ei->vfs_inode;
6639 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6640 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6641 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6642 mutex_init(&ei->log_mutex);
6643 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6644 INIT_LIST_HEAD(&ei->i_orphan);
6645 INIT_LIST_HEAD(&ei->delalloc_inodes);
6646 INIT_LIST_HEAD(&ei->ordered_operations);
6647 RB_CLEAR_NODE(&ei->rb_node);
6652 static void btrfs_i_callback(struct rcu_head *head)
6654 struct inode *inode = container_of(head, struct inode, i_rcu);
6655 INIT_LIST_HEAD(&inode->i_dentry);
6656 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6659 void btrfs_destroy_inode(struct inode *inode)
6661 struct btrfs_ordered_extent *ordered;
6662 struct btrfs_root *root = BTRFS_I(inode)->root;
6664 WARN_ON(!list_empty(&inode->i_dentry));
6665 WARN_ON(inode->i_data.nrpages);
6666 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6667 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6670 * This can happen where we create an inode, but somebody else also
6671 * created the same inode and we need to destroy the one we already
6678 * Make sure we're properly removed from the ordered operation
6682 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6683 spin_lock(&root->fs_info->ordered_extent_lock);
6684 list_del_init(&BTRFS_I(inode)->ordered_operations);
6685 spin_unlock(&root->fs_info->ordered_extent_lock);
6688 if (root == root->fs_info->tree_root) {
6689 struct btrfs_block_group_cache *block_group;
6691 block_group = btrfs_lookup_block_group(root->fs_info,
6692 BTRFS_I(inode)->block_group);
6693 if (block_group && block_group->inode == inode) {
6694 spin_lock(&block_group->lock);
6695 block_group->inode = NULL;
6696 spin_unlock(&block_group->lock);
6697 btrfs_put_block_group(block_group);
6698 } else if (block_group) {
6699 btrfs_put_block_group(block_group);
6703 spin_lock(&root->orphan_lock);
6704 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6705 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6707 list_del_init(&BTRFS_I(inode)->i_orphan);
6709 spin_unlock(&root->orphan_lock);
6712 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6716 printk(KERN_ERR "btrfs found ordered "
6717 "extent %llu %llu on inode cleanup\n",
6718 (unsigned long long)ordered->file_offset,
6719 (unsigned long long)ordered->len);
6720 btrfs_remove_ordered_extent(inode, ordered);
6721 btrfs_put_ordered_extent(ordered);
6722 btrfs_put_ordered_extent(ordered);
6725 inode_tree_del(inode);
6726 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6728 call_rcu(&inode->i_rcu, btrfs_i_callback);
6731 int btrfs_drop_inode(struct inode *inode)
6733 struct btrfs_root *root = BTRFS_I(inode)->root;
6735 if (btrfs_root_refs(&root->root_item) == 0 &&
6736 root != root->fs_info->tree_root)
6739 return generic_drop_inode(inode);
6742 static void init_once(void *foo)
6744 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6746 inode_init_once(&ei->vfs_inode);
6749 void btrfs_destroy_cachep(void)
6751 if (btrfs_inode_cachep)
6752 kmem_cache_destroy(btrfs_inode_cachep);
6753 if (btrfs_trans_handle_cachep)
6754 kmem_cache_destroy(btrfs_trans_handle_cachep);
6755 if (btrfs_transaction_cachep)
6756 kmem_cache_destroy(btrfs_transaction_cachep);
6757 if (btrfs_path_cachep)
6758 kmem_cache_destroy(btrfs_path_cachep);
6759 if (btrfs_free_space_cachep)
6760 kmem_cache_destroy(btrfs_free_space_cachep);
6763 int btrfs_init_cachep(void)
6765 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6766 sizeof(struct btrfs_inode), 0,
6767 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6768 if (!btrfs_inode_cachep)
6771 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6772 sizeof(struct btrfs_trans_handle), 0,
6773 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6774 if (!btrfs_trans_handle_cachep)
6777 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6778 sizeof(struct btrfs_transaction), 0,
6779 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6780 if (!btrfs_transaction_cachep)
6783 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6784 sizeof(struct btrfs_path), 0,
6785 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6786 if (!btrfs_path_cachep)
6789 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6790 sizeof(struct btrfs_free_space), 0,
6791 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6792 if (!btrfs_free_space_cachep)
6797 btrfs_destroy_cachep();
6801 static int btrfs_getattr(struct vfsmount *mnt,
6802 struct dentry *dentry, struct kstat *stat)
6804 struct inode *inode = dentry->d_inode;
6805 generic_fillattr(inode, stat);
6806 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6807 stat->blksize = PAGE_CACHE_SIZE;
6808 stat->blocks = (inode_get_bytes(inode) +
6809 BTRFS_I(inode)->delalloc_bytes) >> 9;
6813 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6814 struct inode *new_dir, struct dentry *new_dentry)
6816 struct btrfs_trans_handle *trans;
6817 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6818 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6819 struct inode *new_inode = new_dentry->d_inode;
6820 struct inode *old_inode = old_dentry->d_inode;
6821 struct timespec ctime = CURRENT_TIME;
6826 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6829 /* we only allow rename subvolume link between subvolumes */
6830 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6833 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6834 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6837 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6838 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6841 * we're using rename to replace one file with another.
6842 * and the replacement file is large. Start IO on it now so
6843 * we don't add too much work to the end of the transaction
6845 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6846 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6847 filemap_flush(old_inode->i_mapping);
6849 /* close the racy window with snapshot create/destroy ioctl */
6850 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6851 down_read(&root->fs_info->subvol_sem);
6853 * We want to reserve the absolute worst case amount of items. So if
6854 * both inodes are subvols and we need to unlink them then that would
6855 * require 4 item modifications, but if they are both normal inodes it
6856 * would require 5 item modifications, so we'll assume their normal
6857 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6858 * should cover the worst case number of items we'll modify.
6860 trans = btrfs_start_transaction(root, 20);
6862 return PTR_ERR(trans);
6864 btrfs_set_trans_block_group(trans, new_dir);
6867 btrfs_record_root_in_trans(trans, dest);
6869 ret = btrfs_set_inode_index(new_dir, &index);
6873 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6874 /* force full log commit if subvolume involved. */
6875 root->fs_info->last_trans_log_full_commit = trans->transid;
6877 ret = btrfs_insert_inode_ref(trans, dest,
6878 new_dentry->d_name.name,
6879 new_dentry->d_name.len,
6881 new_dir->i_ino, index);
6885 * this is an ugly little race, but the rename is required
6886 * to make sure that if we crash, the inode is either at the
6887 * old name or the new one. pinning the log transaction lets
6888 * us make sure we don't allow a log commit to come in after
6889 * we unlink the name but before we add the new name back in.
6891 btrfs_pin_log_trans(root);
6894 * make sure the inode gets flushed if it is replacing
6897 if (new_inode && new_inode->i_size &&
6898 old_inode && S_ISREG(old_inode->i_mode)) {
6899 btrfs_add_ordered_operation(trans, root, old_inode);
6902 old_dir->i_ctime = old_dir->i_mtime = ctime;
6903 new_dir->i_ctime = new_dir->i_mtime = ctime;
6904 old_inode->i_ctime = ctime;
6906 if (old_dentry->d_parent != new_dentry->d_parent)
6907 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6909 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6910 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6911 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6912 old_dentry->d_name.name,
6913 old_dentry->d_name.len);
6915 btrfs_inc_nlink(old_dentry->d_inode);
6916 ret = btrfs_unlink_inode(trans, root, old_dir,
6917 old_dentry->d_inode,
6918 old_dentry->d_name.name,
6919 old_dentry->d_name.len);
6924 new_inode->i_ctime = CURRENT_TIME;
6925 if (unlikely(new_inode->i_ino ==
6926 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6927 root_objectid = BTRFS_I(new_inode)->location.objectid;
6928 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6930 new_dentry->d_name.name,
6931 new_dentry->d_name.len);
6932 BUG_ON(new_inode->i_nlink == 0);
6934 ret = btrfs_unlink_inode(trans, dest, new_dir,
6935 new_dentry->d_inode,
6936 new_dentry->d_name.name,
6937 new_dentry->d_name.len);
6940 if (new_inode->i_nlink == 0) {
6941 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6946 ret = btrfs_add_link(trans, new_dir, old_inode,
6947 new_dentry->d_name.name,
6948 new_dentry->d_name.len, 0, index);
6951 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6952 struct dentry *parent = dget_parent(new_dentry);
6953 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6955 btrfs_end_log_trans(root);
6958 btrfs_end_transaction_throttle(trans, root);
6960 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6961 up_read(&root->fs_info->subvol_sem);
6967 * some fairly slow code that needs optimization. This walks the list
6968 * of all the inodes with pending delalloc and forces them to disk.
6970 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6972 struct list_head *head = &root->fs_info->delalloc_inodes;
6973 struct btrfs_inode *binode;
6974 struct inode *inode;
6976 if (root->fs_info->sb->s_flags & MS_RDONLY)
6979 spin_lock(&root->fs_info->delalloc_lock);
6980 while (!list_empty(head)) {
6981 binode = list_entry(head->next, struct btrfs_inode,
6983 inode = igrab(&binode->vfs_inode);
6985 list_del_init(&binode->delalloc_inodes);
6986 spin_unlock(&root->fs_info->delalloc_lock);
6988 filemap_flush(inode->i_mapping);
6990 btrfs_add_delayed_iput(inode);
6995 spin_lock(&root->fs_info->delalloc_lock);
6997 spin_unlock(&root->fs_info->delalloc_lock);
6999 /* the filemap_flush will queue IO into the worker threads, but
7000 * we have to make sure the IO is actually started and that
7001 * ordered extents get created before we return
7003 atomic_inc(&root->fs_info->async_submit_draining);
7004 while (atomic_read(&root->fs_info->nr_async_submits) ||
7005 atomic_read(&root->fs_info->async_delalloc_pages)) {
7006 wait_event(root->fs_info->async_submit_wait,
7007 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7008 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7010 atomic_dec(&root->fs_info->async_submit_draining);
7014 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7017 struct btrfs_inode *binode;
7018 struct inode *inode = NULL;
7020 spin_lock(&root->fs_info->delalloc_lock);
7021 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7022 binode = list_entry(root->fs_info->delalloc_inodes.next,
7023 struct btrfs_inode, delalloc_inodes);
7024 inode = igrab(&binode->vfs_inode);
7026 list_move_tail(&binode->delalloc_inodes,
7027 &root->fs_info->delalloc_inodes);
7031 list_del_init(&binode->delalloc_inodes);
7032 cond_resched_lock(&root->fs_info->delalloc_lock);
7034 spin_unlock(&root->fs_info->delalloc_lock);
7038 filemap_write_and_wait(inode->i_mapping);
7040 * We have to do this because compression doesn't
7041 * actually set PG_writeback until it submits the pages
7042 * for IO, which happens in an async thread, so we could
7043 * race and not actually wait for any writeback pages
7044 * because they've not been submitted yet. Technically
7045 * this could still be the case for the ordered stuff
7046 * since the async thread may not have started to do its
7047 * work yet. If this becomes the case then we need to
7048 * figure out a way to make sure that in writepage we
7049 * wait for any async pages to be submitted before
7050 * returning so that fdatawait does what its supposed to
7053 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7055 filemap_flush(inode->i_mapping);
7058 btrfs_add_delayed_iput(inode);
7066 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7067 const char *symname)
7069 struct btrfs_trans_handle *trans;
7070 struct btrfs_root *root = BTRFS_I(dir)->root;
7071 struct btrfs_path *path;
7072 struct btrfs_key key;
7073 struct inode *inode = NULL;
7081 struct btrfs_file_extent_item *ei;
7082 struct extent_buffer *leaf;
7083 unsigned long nr = 0;
7085 name_len = strlen(symname) + 1;
7086 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7087 return -ENAMETOOLONG;
7089 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7093 * 2 items for inode item and ref
7094 * 2 items for dir items
7095 * 1 item for xattr if selinux is on
7097 trans = btrfs_start_transaction(root, 5);
7099 return PTR_ERR(trans);
7101 btrfs_set_trans_block_group(trans, dir);
7103 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7104 dentry->d_name.len, dir->i_ino, objectid,
7105 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7107 err = PTR_ERR(inode);
7111 err = btrfs_init_inode_security(trans, inode, dir);
7117 btrfs_set_trans_block_group(trans, inode);
7118 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7122 inode->i_mapping->a_ops = &btrfs_aops;
7123 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7124 inode->i_fop = &btrfs_file_operations;
7125 inode->i_op = &btrfs_file_inode_operations;
7126 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7128 btrfs_update_inode_block_group(trans, inode);
7129 btrfs_update_inode_block_group(trans, dir);
7133 path = btrfs_alloc_path();
7135 key.objectid = inode->i_ino;
7137 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7138 datasize = btrfs_file_extent_calc_inline_size(name_len);
7139 err = btrfs_insert_empty_item(trans, root, path, &key,
7145 leaf = path->nodes[0];
7146 ei = btrfs_item_ptr(leaf, path->slots[0],
7147 struct btrfs_file_extent_item);
7148 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7149 btrfs_set_file_extent_type(leaf, ei,
7150 BTRFS_FILE_EXTENT_INLINE);
7151 btrfs_set_file_extent_encryption(leaf, ei, 0);
7152 btrfs_set_file_extent_compression(leaf, ei, 0);
7153 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7154 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7156 ptr = btrfs_file_extent_inline_start(ei);
7157 write_extent_buffer(leaf, symname, ptr, name_len);
7158 btrfs_mark_buffer_dirty(leaf);
7159 btrfs_free_path(path);
7161 inode->i_op = &btrfs_symlink_inode_operations;
7162 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7163 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7164 inode_set_bytes(inode, name_len);
7165 btrfs_i_size_write(inode, name_len - 1);
7166 err = btrfs_update_inode(trans, root, inode);
7171 nr = trans->blocks_used;
7172 btrfs_end_transaction_throttle(trans, root);
7174 inode_dec_link_count(inode);
7177 btrfs_btree_balance_dirty(root, nr);
7181 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7182 u64 start, u64 num_bytes, u64 min_size,
7183 loff_t actual_len, u64 *alloc_hint,
7184 struct btrfs_trans_handle *trans)
7186 struct btrfs_root *root = BTRFS_I(inode)->root;
7187 struct btrfs_key ins;
7188 u64 cur_offset = start;
7191 bool own_trans = true;
7195 while (num_bytes > 0) {
7197 trans = btrfs_start_transaction(root, 3);
7198 if (IS_ERR(trans)) {
7199 ret = PTR_ERR(trans);
7204 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7205 0, *alloc_hint, (u64)-1, &ins, 1);
7208 btrfs_end_transaction(trans, root);
7212 ret = insert_reserved_file_extent(trans, inode,
7213 cur_offset, ins.objectid,
7214 ins.offset, ins.offset,
7215 ins.offset, 0, 0, 0,
7216 BTRFS_FILE_EXTENT_PREALLOC);
7218 btrfs_drop_extent_cache(inode, cur_offset,
7219 cur_offset + ins.offset -1, 0);
7221 num_bytes -= ins.offset;
7222 cur_offset += ins.offset;
7223 *alloc_hint = ins.objectid + ins.offset;
7225 inode->i_ctime = CURRENT_TIME;
7226 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7227 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7228 (actual_len > inode->i_size) &&
7229 (cur_offset > inode->i_size)) {
7230 if (cur_offset > actual_len)
7231 i_size = actual_len;
7233 i_size = cur_offset;
7234 i_size_write(inode, i_size);
7235 btrfs_ordered_update_i_size(inode, i_size, NULL);
7238 ret = btrfs_update_inode(trans, root, inode);
7242 btrfs_end_transaction(trans, root);
7247 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7248 u64 start, u64 num_bytes, u64 min_size,
7249 loff_t actual_len, u64 *alloc_hint)
7251 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7252 min_size, actual_len, alloc_hint,
7256 int btrfs_prealloc_file_range_trans(struct inode *inode,
7257 struct btrfs_trans_handle *trans, int mode,
7258 u64 start, u64 num_bytes, u64 min_size,
7259 loff_t actual_len, u64 *alloc_hint)
7261 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7262 min_size, actual_len, alloc_hint, trans);
7265 static int btrfs_set_page_dirty(struct page *page)
7267 return __set_page_dirty_nobuffers(page);
7270 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7272 struct btrfs_root *root = BTRFS_I(inode)->root;
7274 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7276 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7278 return generic_permission(inode, mask, flags, btrfs_check_acl);
7281 static const struct inode_operations btrfs_dir_inode_operations = {
7282 .getattr = btrfs_getattr,
7283 .lookup = btrfs_lookup,
7284 .create = btrfs_create,
7285 .unlink = btrfs_unlink,
7287 .mkdir = btrfs_mkdir,
7288 .rmdir = btrfs_rmdir,
7289 .rename = btrfs_rename,
7290 .symlink = btrfs_symlink,
7291 .setattr = btrfs_setattr,
7292 .mknod = btrfs_mknod,
7293 .setxattr = btrfs_setxattr,
7294 .getxattr = btrfs_getxattr,
7295 .listxattr = btrfs_listxattr,
7296 .removexattr = btrfs_removexattr,
7297 .permission = btrfs_permission,
7299 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7300 .lookup = btrfs_lookup,
7301 .permission = btrfs_permission,
7304 static const struct file_operations btrfs_dir_file_operations = {
7305 .llseek = generic_file_llseek,
7306 .read = generic_read_dir,
7307 .readdir = btrfs_real_readdir,
7308 .unlocked_ioctl = btrfs_ioctl,
7309 #ifdef CONFIG_COMPAT
7310 .compat_ioctl = btrfs_ioctl,
7312 .release = btrfs_release_file,
7313 .fsync = btrfs_sync_file,
7316 static struct extent_io_ops btrfs_extent_io_ops = {
7317 .fill_delalloc = run_delalloc_range,
7318 .submit_bio_hook = btrfs_submit_bio_hook,
7319 .merge_bio_hook = btrfs_merge_bio_hook,
7320 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7321 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7322 .writepage_start_hook = btrfs_writepage_start_hook,
7323 .readpage_io_failed_hook = btrfs_io_failed_hook,
7324 .set_bit_hook = btrfs_set_bit_hook,
7325 .clear_bit_hook = btrfs_clear_bit_hook,
7326 .merge_extent_hook = btrfs_merge_extent_hook,
7327 .split_extent_hook = btrfs_split_extent_hook,
7331 * btrfs doesn't support the bmap operation because swapfiles
7332 * use bmap to make a mapping of extents in the file. They assume
7333 * these extents won't change over the life of the file and they
7334 * use the bmap result to do IO directly to the drive.
7336 * the btrfs bmap call would return logical addresses that aren't
7337 * suitable for IO and they also will change frequently as COW
7338 * operations happen. So, swapfile + btrfs == corruption.
7340 * For now we're avoiding this by dropping bmap.
7342 static const struct address_space_operations btrfs_aops = {
7343 .readpage = btrfs_readpage,
7344 .writepage = btrfs_writepage,
7345 .writepages = btrfs_writepages,
7346 .readpages = btrfs_readpages,
7347 .sync_page = block_sync_page,
7348 .direct_IO = btrfs_direct_IO,
7349 .invalidatepage = btrfs_invalidatepage,
7350 .releasepage = btrfs_releasepage,
7351 .set_page_dirty = btrfs_set_page_dirty,
7352 .error_remove_page = generic_error_remove_page,
7355 static const struct address_space_operations btrfs_symlink_aops = {
7356 .readpage = btrfs_readpage,
7357 .writepage = btrfs_writepage,
7358 .invalidatepage = btrfs_invalidatepage,
7359 .releasepage = btrfs_releasepage,
7362 static const struct inode_operations btrfs_file_inode_operations = {
7363 .getattr = btrfs_getattr,
7364 .setattr = btrfs_setattr,
7365 .setxattr = btrfs_setxattr,
7366 .getxattr = btrfs_getxattr,
7367 .listxattr = btrfs_listxattr,
7368 .removexattr = btrfs_removexattr,
7369 .permission = btrfs_permission,
7370 .fiemap = btrfs_fiemap,
7372 static const struct inode_operations btrfs_special_inode_operations = {
7373 .getattr = btrfs_getattr,
7374 .setattr = btrfs_setattr,
7375 .permission = btrfs_permission,
7376 .setxattr = btrfs_setxattr,
7377 .getxattr = btrfs_getxattr,
7378 .listxattr = btrfs_listxattr,
7379 .removexattr = btrfs_removexattr,
7381 static const struct inode_operations btrfs_symlink_inode_operations = {
7382 .readlink = generic_readlink,
7383 .follow_link = page_follow_link_light,
7384 .put_link = page_put_link,
7385 .getattr = btrfs_getattr,
7386 .permission = btrfs_permission,
7387 .setxattr = btrfs_setxattr,
7388 .getxattr = btrfs_getxattr,
7389 .listxattr = btrfs_listxattr,
7390 .removexattr = btrfs_removexattr,
7393 const struct dentry_operations btrfs_dentry_operations = {
7394 .d_delete = btrfs_dentry_delete,