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"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93 struct inode *inode, struct inode *dir,
94 const struct qstr *qstr)
98 err = btrfs_init_acl(trans, inode, dir);
100 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
105 * this does all the hard work for inserting an inline extent into
106 * the btree. The caller should have done a btrfs_drop_extents so that
107 * no overlapping inline items exist in the btree
109 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
110 struct btrfs_root *root, struct inode *inode,
111 u64 start, size_t size, size_t compressed_size,
112 struct page **compressed_pages)
114 struct btrfs_key key;
115 struct btrfs_path *path;
116 struct extent_buffer *leaf;
117 struct page *page = NULL;
120 struct btrfs_file_extent_item *ei;
123 size_t cur_size = size;
125 unsigned long offset;
126 int compress_type = BTRFS_COMPRESS_NONE;
128 if (compressed_size && compressed_pages) {
129 compress_type = root->fs_info->compress_type;
130 cur_size = compressed_size;
133 path = btrfs_alloc_path();
137 path->leave_spinning = 1;
138 btrfs_set_trans_block_group(trans, inode);
140 key.objectid = inode->i_ino;
142 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143 datasize = btrfs_file_extent_calc_inline_size(cur_size);
145 inode_add_bytes(inode, size);
146 ret = btrfs_insert_empty_item(trans, root, path, &key,
153 leaf = path->nodes[0];
154 ei = btrfs_item_ptr(leaf, path->slots[0],
155 struct btrfs_file_extent_item);
156 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158 btrfs_set_file_extent_encryption(leaf, ei, 0);
159 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161 ptr = btrfs_file_extent_inline_start(ei);
163 if (compress_type != BTRFS_COMPRESS_NONE) {
166 while (compressed_size > 0) {
167 cpage = compressed_pages[i];
168 cur_size = min_t(unsigned long, compressed_size,
171 kaddr = kmap_atomic(cpage, KM_USER0);
172 write_extent_buffer(leaf, kaddr, ptr, cur_size);
173 kunmap_atomic(kaddr, KM_USER0);
177 compressed_size -= cur_size;
179 btrfs_set_file_extent_compression(leaf, ei,
182 page = find_get_page(inode->i_mapping,
183 start >> PAGE_CACHE_SHIFT);
184 btrfs_set_file_extent_compression(leaf, ei, 0);
185 kaddr = kmap_atomic(page, KM_USER0);
186 offset = start & (PAGE_CACHE_SIZE - 1);
187 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188 kunmap_atomic(kaddr, KM_USER0);
189 page_cache_release(page);
191 btrfs_mark_buffer_dirty(leaf);
192 btrfs_free_path(path);
195 * we're an inline extent, so nobody can
196 * extend the file past i_size without locking
197 * a page we already have locked.
199 * We must do any isize and inode updates
200 * before we unlock the pages. Otherwise we
201 * could end up racing with unlink.
203 BTRFS_I(inode)->disk_i_size = inode->i_size;
204 btrfs_update_inode(trans, root, inode);
208 btrfs_free_path(path);
214 * conditionally insert an inline extent into the file. This
215 * does the checks required to make sure the data is small enough
216 * to fit as an inline extent.
218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
219 struct btrfs_root *root,
220 struct inode *inode, u64 start, u64 end,
221 size_t compressed_size,
222 struct page **compressed_pages)
224 u64 isize = i_size_read(inode);
225 u64 actual_end = min(end + 1, isize);
226 u64 inline_len = actual_end - start;
227 u64 aligned_end = (end + root->sectorsize - 1) &
228 ~((u64)root->sectorsize - 1);
230 u64 data_len = inline_len;
234 data_len = compressed_size;
237 actual_end >= PAGE_CACHE_SIZE ||
238 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
240 (actual_end & (root->sectorsize - 1)) == 0) ||
242 data_len > root->fs_info->max_inline) {
246 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
250 if (isize > actual_end)
251 inline_len = min_t(u64, isize, actual_end);
252 ret = insert_inline_extent(trans, root, inode, start,
253 inline_len, compressed_size,
256 btrfs_delalloc_release_metadata(inode, end + 1 - start);
257 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
261 struct async_extent {
266 unsigned long nr_pages;
268 struct list_head list;
273 struct btrfs_root *root;
274 struct page *locked_page;
277 struct list_head extents;
278 struct btrfs_work work;
281 static noinline int add_async_extent(struct async_cow *cow,
282 u64 start, u64 ram_size,
285 unsigned long nr_pages,
288 struct async_extent *async_extent;
290 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
291 async_extent->start = start;
292 async_extent->ram_size = ram_size;
293 async_extent->compressed_size = compressed_size;
294 async_extent->pages = pages;
295 async_extent->nr_pages = nr_pages;
296 async_extent->compress_type = compress_type;
297 list_add_tail(&async_extent->list, &cow->extents);
302 * we create compressed extents in two phases. The first
303 * phase compresses a range of pages that have already been
304 * locked (both pages and state bits are locked).
306 * This is done inside an ordered work queue, and the compression
307 * is spread across many cpus. The actual IO submission is step
308 * two, and the ordered work queue takes care of making sure that
309 * happens in the same order things were put onto the queue by
310 * writepages and friends.
312 * If this code finds it can't get good compression, it puts an
313 * entry onto the work queue to write the uncompressed bytes. This
314 * makes sure that both compressed inodes and uncompressed inodes
315 * are written in the same order that pdflush sent them down.
317 static noinline int compress_file_range(struct inode *inode,
318 struct page *locked_page,
320 struct async_cow *async_cow,
323 struct btrfs_root *root = BTRFS_I(inode)->root;
324 struct btrfs_trans_handle *trans;
326 u64 blocksize = root->sectorsize;
328 u64 isize = i_size_read(inode);
330 struct page **pages = NULL;
331 unsigned long nr_pages;
332 unsigned long nr_pages_ret = 0;
333 unsigned long total_compressed = 0;
334 unsigned long total_in = 0;
335 unsigned long max_compressed = 128 * 1024;
336 unsigned long max_uncompressed = 128 * 1024;
339 int compress_type = root->fs_info->compress_type;
341 actual_end = min_t(u64, isize, end + 1);
344 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
345 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
348 * we don't want to send crud past the end of i_size through
349 * compression, that's just a waste of CPU time. So, if the
350 * end of the file is before the start of our current
351 * requested range of bytes, we bail out to the uncompressed
352 * cleanup code that can deal with all of this.
354 * It isn't really the fastest way to fix things, but this is a
355 * very uncommon corner.
357 if (actual_end <= start)
358 goto cleanup_and_bail_uncompressed;
360 total_compressed = actual_end - start;
362 /* we want to make sure that amount of ram required to uncompress
363 * an extent is reasonable, so we limit the total size in ram
364 * of a compressed extent to 128k. This is a crucial number
365 * because it also controls how easily we can spread reads across
366 * cpus for decompression.
368 * We also want to make sure the amount of IO required to do
369 * a random read is reasonably small, so we limit the size of
370 * a compressed extent to 128k.
372 total_compressed = min(total_compressed, max_uncompressed);
373 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
374 num_bytes = max(blocksize, num_bytes);
379 * we do compression for mount -o compress and when the
380 * inode has not been flagged as nocompress. This flag can
381 * change at any time if we discover bad compression ratios.
383 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
384 (btrfs_test_opt(root, COMPRESS) ||
385 (BTRFS_I(inode)->force_compress))) {
387 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
389 if (BTRFS_I(inode)->force_compress)
390 compress_type = BTRFS_I(inode)->force_compress;
392 ret = btrfs_compress_pages(compress_type,
393 inode->i_mapping, start,
394 total_compressed, pages,
395 nr_pages, &nr_pages_ret,
401 unsigned long offset = total_compressed &
402 (PAGE_CACHE_SIZE - 1);
403 struct page *page = pages[nr_pages_ret - 1];
406 /* zero the tail end of the last page, we might be
407 * sending it down to disk
410 kaddr = kmap_atomic(page, KM_USER0);
411 memset(kaddr + offset, 0,
412 PAGE_CACHE_SIZE - offset);
413 kunmap_atomic(kaddr, KM_USER0);
419 trans = btrfs_join_transaction(root, 1);
420 BUG_ON(IS_ERR(trans));
421 btrfs_set_trans_block_group(trans, inode);
422 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
424 /* lets try to make an inline extent */
425 if (ret || total_in < (actual_end - start)) {
426 /* we didn't compress the entire range, try
427 * to make an uncompressed inline extent.
429 ret = cow_file_range_inline(trans, root, inode,
430 start, end, 0, NULL);
432 /* try making a compressed inline extent */
433 ret = cow_file_range_inline(trans, root, inode,
435 total_compressed, pages);
439 * inline extent creation worked, we don't need
440 * to create any more async work items. Unlock
441 * and free up our temp pages.
443 extent_clear_unlock_delalloc(inode,
444 &BTRFS_I(inode)->io_tree,
446 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
447 EXTENT_CLEAR_DELALLOC |
448 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
450 btrfs_end_transaction(trans, root);
453 btrfs_end_transaction(trans, root);
458 * we aren't doing an inline extent round the compressed size
459 * up to a block size boundary so the allocator does sane
462 total_compressed = (total_compressed + blocksize - 1) &
466 * one last check to make sure the compression is really a
467 * win, compare the page count read with the blocks on disk
469 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
470 ~(PAGE_CACHE_SIZE - 1);
471 if (total_compressed >= total_in) {
474 num_bytes = total_in;
477 if (!will_compress && pages) {
479 * the compression code ran but failed to make things smaller,
480 * free any pages it allocated and our page pointer array
482 for (i = 0; i < nr_pages_ret; i++) {
483 WARN_ON(pages[i]->mapping);
484 page_cache_release(pages[i]);
488 total_compressed = 0;
491 /* flag the file so we don't compress in the future */
492 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
493 !(BTRFS_I(inode)->force_compress)) {
494 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
500 /* the async work queues will take care of doing actual
501 * allocation on disk for these compressed pages,
502 * and will submit them to the elevator.
504 add_async_extent(async_cow, start, num_bytes,
505 total_compressed, pages, nr_pages_ret,
508 if (start + num_bytes < end) {
515 cleanup_and_bail_uncompressed:
517 * No compression, but we still need to write the pages in
518 * the file we've been given so far. redirty the locked
519 * page if it corresponds to our extent and set things up
520 * for the async work queue to run cow_file_range to do
521 * the normal delalloc dance
523 if (page_offset(locked_page) >= start &&
524 page_offset(locked_page) <= end) {
525 __set_page_dirty_nobuffers(locked_page);
526 /* unlocked later on in the async handlers */
528 add_async_extent(async_cow, start, end - start + 1,
529 0, NULL, 0, BTRFS_COMPRESS_NONE);
537 for (i = 0; i < nr_pages_ret; i++) {
538 WARN_ON(pages[i]->mapping);
539 page_cache_release(pages[i]);
547 * phase two of compressed writeback. This is the ordered portion
548 * of the code, which only gets called in the order the work was
549 * queued. We walk all the async extents created by compress_file_range
550 * and send them down to the disk.
552 static noinline int submit_compressed_extents(struct inode *inode,
553 struct async_cow *async_cow)
555 struct async_extent *async_extent;
557 struct btrfs_trans_handle *trans;
558 struct btrfs_key ins;
559 struct extent_map *em;
560 struct btrfs_root *root = BTRFS_I(inode)->root;
561 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
562 struct extent_io_tree *io_tree;
565 if (list_empty(&async_cow->extents))
569 while (!list_empty(&async_cow->extents)) {
570 async_extent = list_entry(async_cow->extents.next,
571 struct async_extent, list);
572 list_del(&async_extent->list);
574 io_tree = &BTRFS_I(inode)->io_tree;
577 /* did the compression code fall back to uncompressed IO? */
578 if (!async_extent->pages) {
579 int page_started = 0;
580 unsigned long nr_written = 0;
582 lock_extent(io_tree, async_extent->start,
583 async_extent->start +
584 async_extent->ram_size - 1, GFP_NOFS);
586 /* allocate blocks */
587 ret = cow_file_range(inode, async_cow->locked_page,
589 async_extent->start +
590 async_extent->ram_size - 1,
591 &page_started, &nr_written, 0);
594 * if page_started, cow_file_range inserted an
595 * inline extent and took care of all the unlocking
596 * and IO for us. Otherwise, we need to submit
597 * all those pages down to the drive.
599 if (!page_started && !ret)
600 extent_write_locked_range(io_tree,
601 inode, async_extent->start,
602 async_extent->start +
603 async_extent->ram_size - 1,
611 lock_extent(io_tree, async_extent->start,
612 async_extent->start + async_extent->ram_size - 1,
615 trans = btrfs_join_transaction(root, 1);
616 BUG_ON(IS_ERR(trans));
617 ret = btrfs_reserve_extent(trans, root,
618 async_extent->compressed_size,
619 async_extent->compressed_size,
622 btrfs_end_transaction(trans, root);
626 for (i = 0; i < async_extent->nr_pages; i++) {
627 WARN_ON(async_extent->pages[i]->mapping);
628 page_cache_release(async_extent->pages[i]);
630 kfree(async_extent->pages);
631 async_extent->nr_pages = 0;
632 async_extent->pages = NULL;
633 unlock_extent(io_tree, async_extent->start,
634 async_extent->start +
635 async_extent->ram_size - 1, GFP_NOFS);
640 * here we're doing allocation and writeback of the
643 btrfs_drop_extent_cache(inode, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1, 0);
647 em = alloc_extent_map(GFP_NOFS);
649 em->start = async_extent->start;
650 em->len = async_extent->ram_size;
651 em->orig_start = em->start;
653 em->block_start = ins.objectid;
654 em->block_len = ins.offset;
655 em->bdev = root->fs_info->fs_devices->latest_bdev;
656 em->compress_type = async_extent->compress_type;
657 set_bit(EXTENT_FLAG_PINNED, &em->flags);
658 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
661 write_lock(&em_tree->lock);
662 ret = add_extent_mapping(em_tree, em);
663 write_unlock(&em_tree->lock);
664 if (ret != -EEXIST) {
668 btrfs_drop_extent_cache(inode, async_extent->start,
669 async_extent->start +
670 async_extent->ram_size - 1, 0);
673 ret = btrfs_add_ordered_extent_compress(inode,
676 async_extent->ram_size,
678 BTRFS_ORDERED_COMPRESSED,
679 async_extent->compress_type);
683 * clear dirty, set writeback and unlock the pages.
685 extent_clear_unlock_delalloc(inode,
686 &BTRFS_I(inode)->io_tree,
688 async_extent->start +
689 async_extent->ram_size - 1,
690 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
691 EXTENT_CLEAR_UNLOCK |
692 EXTENT_CLEAR_DELALLOC |
693 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
695 ret = btrfs_submit_compressed_write(inode,
697 async_extent->ram_size,
699 ins.offset, async_extent->pages,
700 async_extent->nr_pages);
703 alloc_hint = ins.objectid + ins.offset;
711 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
714 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
715 struct extent_map *em;
718 read_lock(&em_tree->lock);
719 em = search_extent_mapping(em_tree, start, num_bytes);
722 * if block start isn't an actual block number then find the
723 * first block in this inode and use that as a hint. If that
724 * block is also bogus then just don't worry about it.
726 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
728 em = search_extent_mapping(em_tree, 0, 0);
729 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
730 alloc_hint = em->block_start;
734 alloc_hint = em->block_start;
738 read_unlock(&em_tree->lock);
744 * when extent_io.c finds a delayed allocation range in the file,
745 * the call backs end up in this code. The basic idea is to
746 * allocate extents on disk for the range, and create ordered data structs
747 * in ram to track those extents.
749 * locked_page is the page that writepage had locked already. We use
750 * it to make sure we don't do extra locks or unlocks.
752 * *page_started is set to one if we unlock locked_page and do everything
753 * required to start IO on it. It may be clean and already done with
756 static noinline int cow_file_range(struct inode *inode,
757 struct page *locked_page,
758 u64 start, u64 end, int *page_started,
759 unsigned long *nr_written,
762 struct btrfs_root *root = BTRFS_I(inode)->root;
763 struct btrfs_trans_handle *trans;
766 unsigned long ram_size;
769 u64 blocksize = root->sectorsize;
770 struct btrfs_key ins;
771 struct extent_map *em;
772 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
775 BUG_ON(root == root->fs_info->tree_root);
776 trans = btrfs_join_transaction(root, 1);
777 BUG_ON(IS_ERR(trans));
778 btrfs_set_trans_block_group(trans, inode);
779 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
781 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
782 num_bytes = max(blocksize, num_bytes);
783 disk_num_bytes = num_bytes;
787 /* lets try to make an inline extent */
788 ret = cow_file_range_inline(trans, root, inode,
789 start, end, 0, NULL);
791 extent_clear_unlock_delalloc(inode,
792 &BTRFS_I(inode)->io_tree,
794 EXTENT_CLEAR_UNLOCK_PAGE |
795 EXTENT_CLEAR_UNLOCK |
796 EXTENT_CLEAR_DELALLOC |
798 EXTENT_SET_WRITEBACK |
799 EXTENT_END_WRITEBACK);
801 *nr_written = *nr_written +
802 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
809 BUG_ON(disk_num_bytes >
810 btrfs_super_total_bytes(&root->fs_info->super_copy));
812 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
813 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
815 while (disk_num_bytes > 0) {
818 cur_alloc_size = disk_num_bytes;
819 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
820 root->sectorsize, 0, alloc_hint,
824 em = alloc_extent_map(GFP_NOFS);
827 em->orig_start = em->start;
828 ram_size = ins.offset;
829 em->len = ins.offset;
831 em->block_start = ins.objectid;
832 em->block_len = ins.offset;
833 em->bdev = root->fs_info->fs_devices->latest_bdev;
834 set_bit(EXTENT_FLAG_PINNED, &em->flags);
837 write_lock(&em_tree->lock);
838 ret = add_extent_mapping(em_tree, em);
839 write_unlock(&em_tree->lock);
840 if (ret != -EEXIST) {
844 btrfs_drop_extent_cache(inode, start,
845 start + ram_size - 1, 0);
848 cur_alloc_size = ins.offset;
849 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
850 ram_size, cur_alloc_size, 0);
853 if (root->root_key.objectid ==
854 BTRFS_DATA_RELOC_TREE_OBJECTID) {
855 ret = btrfs_reloc_clone_csums(inode, start,
860 if (disk_num_bytes < cur_alloc_size)
863 /* we're not doing compressed IO, don't unlock the first
864 * page (which the caller expects to stay locked), don't
865 * clear any dirty bits and don't set any writeback bits
867 * Do set the Private2 bit so we know this page was properly
868 * setup for writepage
870 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
871 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
874 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
875 start, start + ram_size - 1,
877 disk_num_bytes -= cur_alloc_size;
878 num_bytes -= cur_alloc_size;
879 alloc_hint = ins.objectid + ins.offset;
880 start += cur_alloc_size;
884 btrfs_end_transaction(trans, root);
890 * work queue call back to started compression on a file and pages
892 static noinline void async_cow_start(struct btrfs_work *work)
894 struct async_cow *async_cow;
896 async_cow = container_of(work, struct async_cow, work);
898 compress_file_range(async_cow->inode, async_cow->locked_page,
899 async_cow->start, async_cow->end, async_cow,
902 async_cow->inode = NULL;
906 * work queue call back to submit previously compressed pages
908 static noinline void async_cow_submit(struct btrfs_work *work)
910 struct async_cow *async_cow;
911 struct btrfs_root *root;
912 unsigned long nr_pages;
914 async_cow = container_of(work, struct async_cow, work);
916 root = async_cow->root;
917 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
920 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
922 if (atomic_read(&root->fs_info->async_delalloc_pages) <
924 waitqueue_active(&root->fs_info->async_submit_wait))
925 wake_up(&root->fs_info->async_submit_wait);
927 if (async_cow->inode)
928 submit_compressed_extents(async_cow->inode, async_cow);
931 static noinline void async_cow_free(struct btrfs_work *work)
933 struct async_cow *async_cow;
934 async_cow = container_of(work, struct async_cow, work);
938 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
939 u64 start, u64 end, int *page_started,
940 unsigned long *nr_written)
942 struct async_cow *async_cow;
943 struct btrfs_root *root = BTRFS_I(inode)->root;
944 unsigned long nr_pages;
946 int limit = 10 * 1024 * 1042;
948 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
949 1, 0, NULL, GFP_NOFS);
950 while (start < end) {
951 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
952 async_cow->inode = inode;
953 async_cow->root = root;
954 async_cow->locked_page = locked_page;
955 async_cow->start = start;
957 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
960 cur_end = min(end, start + 512 * 1024 - 1);
962 async_cow->end = cur_end;
963 INIT_LIST_HEAD(&async_cow->extents);
965 async_cow->work.func = async_cow_start;
966 async_cow->work.ordered_func = async_cow_submit;
967 async_cow->work.ordered_free = async_cow_free;
968 async_cow->work.flags = 0;
970 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
972 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
974 btrfs_queue_worker(&root->fs_info->delalloc_workers,
977 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
978 wait_event(root->fs_info->async_submit_wait,
979 (atomic_read(&root->fs_info->async_delalloc_pages) <
983 while (atomic_read(&root->fs_info->async_submit_draining) &&
984 atomic_read(&root->fs_info->async_delalloc_pages)) {
985 wait_event(root->fs_info->async_submit_wait,
986 (atomic_read(&root->fs_info->async_delalloc_pages) ==
990 *nr_written += nr_pages;
997 static noinline int csum_exist_in_range(struct btrfs_root *root,
998 u64 bytenr, u64 num_bytes)
1001 struct btrfs_ordered_sum *sums;
1004 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1005 bytenr + num_bytes - 1, &list);
1006 if (ret == 0 && list_empty(&list))
1009 while (!list_empty(&list)) {
1010 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1011 list_del(&sums->list);
1018 * when nowcow writeback call back. This checks for snapshots or COW copies
1019 * of the extents that exist in the file, and COWs the file as required.
1021 * If no cow copies or snapshots exist, we write directly to the existing
1024 static noinline int run_delalloc_nocow(struct inode *inode,
1025 struct page *locked_page,
1026 u64 start, u64 end, int *page_started, int force,
1027 unsigned long *nr_written)
1029 struct btrfs_root *root = BTRFS_I(inode)->root;
1030 struct btrfs_trans_handle *trans;
1031 struct extent_buffer *leaf;
1032 struct btrfs_path *path;
1033 struct btrfs_file_extent_item *fi;
1034 struct btrfs_key found_key;
1046 bool nolock = false;
1048 path = btrfs_alloc_path();
1050 if (root == root->fs_info->tree_root) {
1052 trans = btrfs_join_transaction_nolock(root, 1);
1054 trans = btrfs_join_transaction(root, 1);
1056 BUG_ON(IS_ERR(trans));
1058 cow_start = (u64)-1;
1061 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1064 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1065 leaf = path->nodes[0];
1066 btrfs_item_key_to_cpu(leaf, &found_key,
1067 path->slots[0] - 1);
1068 if (found_key.objectid == inode->i_ino &&
1069 found_key.type == BTRFS_EXTENT_DATA_KEY)
1074 leaf = path->nodes[0];
1075 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1076 ret = btrfs_next_leaf(root, path);
1081 leaf = path->nodes[0];
1087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1089 if (found_key.objectid > inode->i_ino ||
1090 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1091 found_key.offset > end)
1094 if (found_key.offset > cur_offset) {
1095 extent_end = found_key.offset;
1100 fi = btrfs_item_ptr(leaf, path->slots[0],
1101 struct btrfs_file_extent_item);
1102 extent_type = btrfs_file_extent_type(leaf, fi);
1104 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1105 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1106 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1107 extent_offset = btrfs_file_extent_offset(leaf, fi);
1108 extent_end = found_key.offset +
1109 btrfs_file_extent_num_bytes(leaf, fi);
1110 if (extent_end <= start) {
1114 if (disk_bytenr == 0)
1116 if (btrfs_file_extent_compression(leaf, fi) ||
1117 btrfs_file_extent_encryption(leaf, fi) ||
1118 btrfs_file_extent_other_encoding(leaf, fi))
1120 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1122 if (btrfs_extent_readonly(root, disk_bytenr))
1124 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1126 extent_offset, disk_bytenr))
1128 disk_bytenr += extent_offset;
1129 disk_bytenr += cur_offset - found_key.offset;
1130 num_bytes = min(end + 1, extent_end) - cur_offset;
1132 * force cow if csum exists in the range.
1133 * this ensure that csum for a given extent are
1134 * either valid or do not exist.
1136 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1139 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1140 extent_end = found_key.offset +
1141 btrfs_file_extent_inline_len(leaf, fi);
1142 extent_end = ALIGN(extent_end, root->sectorsize);
1147 if (extent_end <= start) {
1152 if (cow_start == (u64)-1)
1153 cow_start = cur_offset;
1154 cur_offset = extent_end;
1155 if (cur_offset > end)
1161 btrfs_release_path(root, path);
1162 if (cow_start != (u64)-1) {
1163 ret = cow_file_range(inode, locked_page, cow_start,
1164 found_key.offset - 1, page_started,
1167 cow_start = (u64)-1;
1170 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1171 struct extent_map *em;
1172 struct extent_map_tree *em_tree;
1173 em_tree = &BTRFS_I(inode)->extent_tree;
1174 em = alloc_extent_map(GFP_NOFS);
1176 em->start = cur_offset;
1177 em->orig_start = em->start;
1178 em->len = num_bytes;
1179 em->block_len = num_bytes;
1180 em->block_start = disk_bytenr;
1181 em->bdev = root->fs_info->fs_devices->latest_bdev;
1182 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1184 write_lock(&em_tree->lock);
1185 ret = add_extent_mapping(em_tree, em);
1186 write_unlock(&em_tree->lock);
1187 if (ret != -EEXIST) {
1188 free_extent_map(em);
1191 btrfs_drop_extent_cache(inode, em->start,
1192 em->start + em->len - 1, 0);
1194 type = BTRFS_ORDERED_PREALLOC;
1196 type = BTRFS_ORDERED_NOCOW;
1199 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1200 num_bytes, num_bytes, type);
1203 if (root->root_key.objectid ==
1204 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1205 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1210 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1211 cur_offset, cur_offset + num_bytes - 1,
1212 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1213 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1214 EXTENT_SET_PRIVATE2);
1215 cur_offset = extent_end;
1216 if (cur_offset > end)
1219 btrfs_release_path(root, path);
1221 if (cur_offset <= end && cow_start == (u64)-1)
1222 cow_start = cur_offset;
1223 if (cow_start != (u64)-1) {
1224 ret = cow_file_range(inode, locked_page, cow_start, end,
1225 page_started, nr_written, 1);
1230 ret = btrfs_end_transaction_nolock(trans, root);
1233 ret = btrfs_end_transaction(trans, root);
1236 btrfs_free_path(path);
1241 * extent_io.c call back to do delayed allocation processing
1243 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1244 u64 start, u64 end, int *page_started,
1245 unsigned long *nr_written)
1248 struct btrfs_root *root = BTRFS_I(inode)->root;
1250 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1251 ret = run_delalloc_nocow(inode, locked_page, start, end,
1252 page_started, 1, nr_written);
1253 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1254 ret = run_delalloc_nocow(inode, locked_page, start, end,
1255 page_started, 0, nr_written);
1256 else if (!btrfs_test_opt(root, COMPRESS) &&
1257 !(BTRFS_I(inode)->force_compress))
1258 ret = cow_file_range(inode, locked_page, start, end,
1259 page_started, nr_written, 1);
1261 ret = cow_file_range_async(inode, locked_page, start, end,
1262 page_started, nr_written);
1266 static int btrfs_split_extent_hook(struct inode *inode,
1267 struct extent_state *orig, u64 split)
1269 /* not delalloc, ignore it */
1270 if (!(orig->state & EXTENT_DELALLOC))
1273 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1278 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1279 * extents so we can keep track of new extents that are just merged onto old
1280 * extents, such as when we are doing sequential writes, so we can properly
1281 * account for the metadata space we'll need.
1283 static int btrfs_merge_extent_hook(struct inode *inode,
1284 struct extent_state *new,
1285 struct extent_state *other)
1287 /* not delalloc, ignore it */
1288 if (!(other->state & EXTENT_DELALLOC))
1291 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1296 * extent_io.c set_bit_hook, used to track delayed allocation
1297 * bytes in this file, and to maintain the list of inodes that
1298 * have pending delalloc work to be done.
1300 static int btrfs_set_bit_hook(struct inode *inode,
1301 struct extent_state *state, int *bits)
1305 * set_bit and clear bit hooks normally require _irqsave/restore
1306 * but in this case, we are only testeing for the DELALLOC
1307 * bit, which is only set or cleared with irqs on
1309 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1310 struct btrfs_root *root = BTRFS_I(inode)->root;
1311 u64 len = state->end + 1 - state->start;
1312 int do_list = (root->root_key.objectid !=
1313 BTRFS_ROOT_TREE_OBJECTID);
1315 if (*bits & EXTENT_FIRST_DELALLOC)
1316 *bits &= ~EXTENT_FIRST_DELALLOC;
1318 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1320 spin_lock(&root->fs_info->delalloc_lock);
1321 BTRFS_I(inode)->delalloc_bytes += len;
1322 root->fs_info->delalloc_bytes += len;
1323 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1324 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1325 &root->fs_info->delalloc_inodes);
1327 spin_unlock(&root->fs_info->delalloc_lock);
1333 * extent_io.c clear_bit_hook, see set_bit_hook for why
1335 static int btrfs_clear_bit_hook(struct inode *inode,
1336 struct extent_state *state, int *bits)
1339 * set_bit and clear bit hooks normally require _irqsave/restore
1340 * but in this case, we are only testeing for the DELALLOC
1341 * bit, which is only set or cleared with irqs on
1343 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1344 struct btrfs_root *root = BTRFS_I(inode)->root;
1345 u64 len = state->end + 1 - state->start;
1346 int do_list = (root->root_key.objectid !=
1347 BTRFS_ROOT_TREE_OBJECTID);
1349 if (*bits & EXTENT_FIRST_DELALLOC)
1350 *bits &= ~EXTENT_FIRST_DELALLOC;
1351 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1352 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1354 if (*bits & EXTENT_DO_ACCOUNTING)
1355 btrfs_delalloc_release_metadata(inode, len);
1357 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1359 btrfs_free_reserved_data_space(inode, len);
1361 spin_lock(&root->fs_info->delalloc_lock);
1362 root->fs_info->delalloc_bytes -= len;
1363 BTRFS_I(inode)->delalloc_bytes -= len;
1365 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1366 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1367 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1369 spin_unlock(&root->fs_info->delalloc_lock);
1375 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1376 * we don't create bios that span stripes or chunks
1378 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1379 size_t size, struct bio *bio,
1380 unsigned long bio_flags)
1382 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1383 struct btrfs_mapping_tree *map_tree;
1384 u64 logical = (u64)bio->bi_sector << 9;
1389 if (bio_flags & EXTENT_BIO_COMPRESSED)
1392 length = bio->bi_size;
1393 map_tree = &root->fs_info->mapping_tree;
1394 map_length = length;
1395 ret = btrfs_map_block(map_tree, READ, logical,
1396 &map_length, NULL, 0);
1398 if (map_length < length + size)
1404 * in order to insert checksums into the metadata in large chunks,
1405 * we wait until bio submission time. All the pages in the bio are
1406 * checksummed and sums are attached onto the ordered extent record.
1408 * At IO completion time the cums attached on the ordered extent record
1409 * are inserted into the btree
1411 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1412 struct bio *bio, int mirror_num,
1413 unsigned long bio_flags,
1416 struct btrfs_root *root = BTRFS_I(inode)->root;
1419 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1425 * in order to insert checksums into the metadata in large chunks,
1426 * we wait until bio submission time. All the pages in the bio are
1427 * checksummed and sums are attached onto the ordered extent record.
1429 * At IO completion time the cums attached on the ordered extent record
1430 * are inserted into the btree
1432 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1433 int mirror_num, unsigned long bio_flags,
1436 struct btrfs_root *root = BTRFS_I(inode)->root;
1437 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1441 * extent_io.c submission hook. This does the right thing for csum calculation
1442 * on write, or reading the csums from the tree before a read
1444 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1445 int mirror_num, unsigned long bio_flags,
1448 struct btrfs_root *root = BTRFS_I(inode)->root;
1452 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1454 if (root == root->fs_info->tree_root)
1455 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1457 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1460 if (!(rw & REQ_WRITE)) {
1461 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1462 return btrfs_submit_compressed_read(inode, bio,
1463 mirror_num, bio_flags);
1464 } else if (!skip_sum)
1465 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1467 } else if (!skip_sum) {
1468 /* csum items have already been cloned */
1469 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1471 /* we're doing a write, do the async checksumming */
1472 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1473 inode, rw, bio, mirror_num,
1474 bio_flags, bio_offset,
1475 __btrfs_submit_bio_start,
1476 __btrfs_submit_bio_done);
1480 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1484 * given a list of ordered sums record them in the inode. This happens
1485 * at IO completion time based on sums calculated at bio submission time.
1487 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1488 struct inode *inode, u64 file_offset,
1489 struct list_head *list)
1491 struct btrfs_ordered_sum *sum;
1493 btrfs_set_trans_block_group(trans, inode);
1495 list_for_each_entry(sum, list, list) {
1496 btrfs_csum_file_blocks(trans,
1497 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1502 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1503 struct extent_state **cached_state)
1505 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1507 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1508 cached_state, GFP_NOFS);
1511 /* see btrfs_writepage_start_hook for details on why this is required */
1512 struct btrfs_writepage_fixup {
1514 struct btrfs_work work;
1517 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1519 struct btrfs_writepage_fixup *fixup;
1520 struct btrfs_ordered_extent *ordered;
1521 struct extent_state *cached_state = NULL;
1523 struct inode *inode;
1527 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1531 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1532 ClearPageChecked(page);
1536 inode = page->mapping->host;
1537 page_start = page_offset(page);
1538 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1540 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1541 &cached_state, GFP_NOFS);
1543 /* already ordered? We're done */
1544 if (PagePrivate2(page))
1547 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1549 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1550 page_end, &cached_state, GFP_NOFS);
1552 btrfs_start_ordered_extent(inode, ordered, 1);
1557 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1558 ClearPageChecked(page);
1560 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1561 &cached_state, GFP_NOFS);
1564 page_cache_release(page);
1569 * There are a few paths in the higher layers of the kernel that directly
1570 * set the page dirty bit without asking the filesystem if it is a
1571 * good idea. This causes problems because we want to make sure COW
1572 * properly happens and the data=ordered rules are followed.
1574 * In our case any range that doesn't have the ORDERED bit set
1575 * hasn't been properly setup for IO. We kick off an async process
1576 * to fix it up. The async helper will wait for ordered extents, set
1577 * the delalloc bit and make it safe to write the page.
1579 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1581 struct inode *inode = page->mapping->host;
1582 struct btrfs_writepage_fixup *fixup;
1583 struct btrfs_root *root = BTRFS_I(inode)->root;
1585 /* this page is properly in the ordered list */
1586 if (TestClearPagePrivate2(page))
1589 if (PageChecked(page))
1592 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1596 SetPageChecked(page);
1597 page_cache_get(page);
1598 fixup->work.func = btrfs_writepage_fixup_worker;
1600 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1604 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1605 struct inode *inode, u64 file_pos,
1606 u64 disk_bytenr, u64 disk_num_bytes,
1607 u64 num_bytes, u64 ram_bytes,
1608 u8 compression, u8 encryption,
1609 u16 other_encoding, int extent_type)
1611 struct btrfs_root *root = BTRFS_I(inode)->root;
1612 struct btrfs_file_extent_item *fi;
1613 struct btrfs_path *path;
1614 struct extent_buffer *leaf;
1615 struct btrfs_key ins;
1619 path = btrfs_alloc_path();
1622 path->leave_spinning = 1;
1625 * we may be replacing one extent in the tree with another.
1626 * The new extent is pinned in the extent map, and we don't want
1627 * to drop it from the cache until it is completely in the btree.
1629 * So, tell btrfs_drop_extents to leave this extent in the cache.
1630 * the caller is expected to unpin it and allow it to be merged
1633 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1637 ins.objectid = inode->i_ino;
1638 ins.offset = file_pos;
1639 ins.type = BTRFS_EXTENT_DATA_KEY;
1640 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1642 leaf = path->nodes[0];
1643 fi = btrfs_item_ptr(leaf, path->slots[0],
1644 struct btrfs_file_extent_item);
1645 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1646 btrfs_set_file_extent_type(leaf, fi, extent_type);
1647 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1648 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1649 btrfs_set_file_extent_offset(leaf, fi, 0);
1650 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1651 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1652 btrfs_set_file_extent_compression(leaf, fi, compression);
1653 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1654 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1656 btrfs_unlock_up_safe(path, 1);
1657 btrfs_set_lock_blocking(leaf);
1659 btrfs_mark_buffer_dirty(leaf);
1661 inode_add_bytes(inode, num_bytes);
1663 ins.objectid = disk_bytenr;
1664 ins.offset = disk_num_bytes;
1665 ins.type = BTRFS_EXTENT_ITEM_KEY;
1666 ret = btrfs_alloc_reserved_file_extent(trans, root,
1667 root->root_key.objectid,
1668 inode->i_ino, file_pos, &ins);
1670 btrfs_free_path(path);
1676 * helper function for btrfs_finish_ordered_io, this
1677 * just reads in some of the csum leaves to prime them into ram
1678 * before we start the transaction. It limits the amount of btree
1679 * reads required while inside the transaction.
1681 /* as ordered data IO finishes, this gets called so we can finish
1682 * an ordered extent if the range of bytes in the file it covers are
1685 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1687 struct btrfs_root *root = BTRFS_I(inode)->root;
1688 struct btrfs_trans_handle *trans = NULL;
1689 struct btrfs_ordered_extent *ordered_extent = NULL;
1690 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1691 struct extent_state *cached_state = NULL;
1692 int compress_type = 0;
1694 bool nolock = false;
1696 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1700 BUG_ON(!ordered_extent);
1702 nolock = (root == root->fs_info->tree_root);
1704 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1705 BUG_ON(!list_empty(&ordered_extent->list));
1706 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1709 trans = btrfs_join_transaction_nolock(root, 1);
1711 trans = btrfs_join_transaction(root, 1);
1712 BUG_ON(IS_ERR(trans));
1713 btrfs_set_trans_block_group(trans, inode);
1714 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1715 ret = btrfs_update_inode(trans, root, inode);
1721 lock_extent_bits(io_tree, ordered_extent->file_offset,
1722 ordered_extent->file_offset + ordered_extent->len - 1,
1723 0, &cached_state, GFP_NOFS);
1726 trans = btrfs_join_transaction_nolock(root, 1);
1728 trans = btrfs_join_transaction(root, 1);
1729 BUG_ON(IS_ERR(trans));
1730 btrfs_set_trans_block_group(trans, inode);
1731 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1733 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1734 compress_type = ordered_extent->compress_type;
1735 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1736 BUG_ON(compress_type);
1737 ret = btrfs_mark_extent_written(trans, inode,
1738 ordered_extent->file_offset,
1739 ordered_extent->file_offset +
1740 ordered_extent->len);
1743 BUG_ON(root == root->fs_info->tree_root);
1744 ret = insert_reserved_file_extent(trans, inode,
1745 ordered_extent->file_offset,
1746 ordered_extent->start,
1747 ordered_extent->disk_len,
1748 ordered_extent->len,
1749 ordered_extent->len,
1750 compress_type, 0, 0,
1751 BTRFS_FILE_EXTENT_REG);
1752 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1753 ordered_extent->file_offset,
1754 ordered_extent->len);
1757 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1758 ordered_extent->file_offset +
1759 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1761 add_pending_csums(trans, inode, ordered_extent->file_offset,
1762 &ordered_extent->list);
1764 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1765 ret = btrfs_update_inode(trans, root, inode);
1770 btrfs_end_transaction_nolock(trans, root);
1772 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1774 btrfs_end_transaction(trans, root);
1778 btrfs_put_ordered_extent(ordered_extent);
1779 /* once for the tree */
1780 btrfs_put_ordered_extent(ordered_extent);
1785 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1786 struct extent_state *state, int uptodate)
1788 ClearPagePrivate2(page);
1789 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1793 * When IO fails, either with EIO or csum verification fails, we
1794 * try other mirrors that might have a good copy of the data. This
1795 * io_failure_record is used to record state as we go through all the
1796 * mirrors. If another mirror has good data, the page is set up to date
1797 * and things continue. If a good mirror can't be found, the original
1798 * bio end_io callback is called to indicate things have failed.
1800 struct io_failure_record {
1805 unsigned long bio_flags;
1809 static int btrfs_io_failed_hook(struct bio *failed_bio,
1810 struct page *page, u64 start, u64 end,
1811 struct extent_state *state)
1813 struct io_failure_record *failrec = NULL;
1815 struct extent_map *em;
1816 struct inode *inode = page->mapping->host;
1817 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1818 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1825 ret = get_state_private(failure_tree, start, &private);
1827 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1830 failrec->start = start;
1831 failrec->len = end - start + 1;
1832 failrec->last_mirror = 0;
1833 failrec->bio_flags = 0;
1835 read_lock(&em_tree->lock);
1836 em = lookup_extent_mapping(em_tree, start, failrec->len);
1837 if (em->start > start || em->start + em->len < start) {
1838 free_extent_map(em);
1841 read_unlock(&em_tree->lock);
1843 if (!em || IS_ERR(em)) {
1847 logical = start - em->start;
1848 logical = em->block_start + logical;
1849 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1850 logical = em->block_start;
1851 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1852 extent_set_compress_type(&failrec->bio_flags,
1855 failrec->logical = logical;
1856 free_extent_map(em);
1857 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1858 EXTENT_DIRTY, GFP_NOFS);
1859 set_state_private(failure_tree, start,
1860 (u64)(unsigned long)failrec);
1862 failrec = (struct io_failure_record *)(unsigned long)private;
1864 num_copies = btrfs_num_copies(
1865 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1866 failrec->logical, failrec->len);
1867 failrec->last_mirror++;
1869 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1870 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1873 if (state && state->start != failrec->start)
1875 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1877 if (!state || failrec->last_mirror > num_copies) {
1878 set_state_private(failure_tree, failrec->start, 0);
1879 clear_extent_bits(failure_tree, failrec->start,
1880 failrec->start + failrec->len - 1,
1881 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1885 bio = bio_alloc(GFP_NOFS, 1);
1886 bio->bi_private = state;
1887 bio->bi_end_io = failed_bio->bi_end_io;
1888 bio->bi_sector = failrec->logical >> 9;
1889 bio->bi_bdev = failed_bio->bi_bdev;
1892 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1893 if (failed_bio->bi_rw & REQ_WRITE)
1898 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1899 failrec->last_mirror,
1900 failrec->bio_flags, 0);
1905 * each time an IO finishes, we do a fast check in the IO failure tree
1906 * to see if we need to process or clean up an io_failure_record
1908 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1911 u64 private_failure;
1912 struct io_failure_record *failure;
1916 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1917 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1918 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1919 start, &private_failure);
1921 failure = (struct io_failure_record *)(unsigned long)
1923 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1925 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1927 failure->start + failure->len - 1,
1928 EXTENT_DIRTY | EXTENT_LOCKED,
1937 * when reads are done, we need to check csums to verify the data is correct
1938 * if there's a match, we allow the bio to finish. If not, we go through
1939 * the io_failure_record routines to find good copies
1941 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1942 struct extent_state *state)
1944 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1945 struct inode *inode = page->mapping->host;
1946 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1948 u64 private = ~(u32)0;
1950 struct btrfs_root *root = BTRFS_I(inode)->root;
1953 if (PageChecked(page)) {
1954 ClearPageChecked(page);
1958 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1961 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1962 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1963 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1968 if (state && state->start == start) {
1969 private = state->private;
1972 ret = get_state_private(io_tree, start, &private);
1974 kaddr = kmap_atomic(page, KM_USER0);
1978 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1979 btrfs_csum_final(csum, (char *)&csum);
1980 if (csum != private)
1983 kunmap_atomic(kaddr, KM_USER0);
1985 /* if the io failure tree for this inode is non-empty,
1986 * check to see if we've recovered from a failed IO
1988 btrfs_clean_io_failures(inode, start);
1992 if (printk_ratelimit()) {
1993 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1994 "private %llu\n", page->mapping->host->i_ino,
1995 (unsigned long long)start, csum,
1996 (unsigned long long)private);
1998 memset(kaddr + offset, 1, end - start + 1);
1999 flush_dcache_page(page);
2000 kunmap_atomic(kaddr, KM_USER0);
2006 struct delayed_iput {
2007 struct list_head list;
2008 struct inode *inode;
2011 void btrfs_add_delayed_iput(struct inode *inode)
2013 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2014 struct delayed_iput *delayed;
2016 if (atomic_add_unless(&inode->i_count, -1, 1))
2019 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2020 delayed->inode = inode;
2022 spin_lock(&fs_info->delayed_iput_lock);
2023 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2024 spin_unlock(&fs_info->delayed_iput_lock);
2027 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2030 struct btrfs_fs_info *fs_info = root->fs_info;
2031 struct delayed_iput *delayed;
2034 spin_lock(&fs_info->delayed_iput_lock);
2035 empty = list_empty(&fs_info->delayed_iputs);
2036 spin_unlock(&fs_info->delayed_iput_lock);
2040 down_read(&root->fs_info->cleanup_work_sem);
2041 spin_lock(&fs_info->delayed_iput_lock);
2042 list_splice_init(&fs_info->delayed_iputs, &list);
2043 spin_unlock(&fs_info->delayed_iput_lock);
2045 while (!list_empty(&list)) {
2046 delayed = list_entry(list.next, struct delayed_iput, list);
2047 list_del(&delayed->list);
2048 iput(delayed->inode);
2051 up_read(&root->fs_info->cleanup_work_sem);
2055 * calculate extra metadata reservation when snapshotting a subvolume
2056 * contains orphan files.
2058 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2059 struct btrfs_pending_snapshot *pending,
2060 u64 *bytes_to_reserve)
2062 struct btrfs_root *root;
2063 struct btrfs_block_rsv *block_rsv;
2067 root = pending->root;
2068 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2071 block_rsv = root->orphan_block_rsv;
2073 /* orphan block reservation for the snapshot */
2074 num_bytes = block_rsv->size;
2077 * after the snapshot is created, COWing tree blocks may use more
2078 * space than it frees. So we should make sure there is enough
2081 index = trans->transid & 0x1;
2082 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2083 num_bytes += block_rsv->size -
2084 (block_rsv->reserved + block_rsv->freed[index]);
2087 *bytes_to_reserve += num_bytes;
2090 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2091 struct btrfs_pending_snapshot *pending)
2093 struct btrfs_root *root = pending->root;
2094 struct btrfs_root *snap = pending->snap;
2095 struct btrfs_block_rsv *block_rsv;
2100 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2103 /* refill source subvolume's orphan block reservation */
2104 block_rsv = root->orphan_block_rsv;
2105 index = trans->transid & 0x1;
2106 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2107 num_bytes = block_rsv->size -
2108 (block_rsv->reserved + block_rsv->freed[index]);
2109 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2110 root->orphan_block_rsv,
2115 /* setup orphan block reservation for the snapshot */
2116 block_rsv = btrfs_alloc_block_rsv(snap);
2119 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2120 snap->orphan_block_rsv = block_rsv;
2122 num_bytes = root->orphan_block_rsv->size;
2123 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2124 block_rsv, num_bytes);
2128 /* insert orphan item for the snapshot */
2129 WARN_ON(!root->orphan_item_inserted);
2130 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2131 snap->root_key.objectid);
2133 snap->orphan_item_inserted = 1;
2137 enum btrfs_orphan_cleanup_state {
2138 ORPHAN_CLEANUP_STARTED = 1,
2139 ORPHAN_CLEANUP_DONE = 2,
2143 * This is called in transaction commmit time. If there are no orphan
2144 * files in the subvolume, it removes orphan item and frees block_rsv
2147 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2148 struct btrfs_root *root)
2152 if (!list_empty(&root->orphan_list) ||
2153 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2156 if (root->orphan_item_inserted &&
2157 btrfs_root_refs(&root->root_item) > 0) {
2158 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2159 root->root_key.objectid);
2161 root->orphan_item_inserted = 0;
2164 if (root->orphan_block_rsv) {
2165 WARN_ON(root->orphan_block_rsv->size > 0);
2166 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2167 root->orphan_block_rsv = NULL;
2172 * This creates an orphan entry for the given inode in case something goes
2173 * wrong in the middle of an unlink/truncate.
2175 * NOTE: caller of this function should reserve 5 units of metadata for
2178 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2180 struct btrfs_root *root = BTRFS_I(inode)->root;
2181 struct btrfs_block_rsv *block_rsv = NULL;
2186 if (!root->orphan_block_rsv) {
2187 block_rsv = btrfs_alloc_block_rsv(root);
2191 spin_lock(&root->orphan_lock);
2192 if (!root->orphan_block_rsv) {
2193 root->orphan_block_rsv = block_rsv;
2194 } else if (block_rsv) {
2195 btrfs_free_block_rsv(root, block_rsv);
2199 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2200 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2203 * For proper ENOSPC handling, we should do orphan
2204 * cleanup when mounting. But this introduces backward
2205 * compatibility issue.
2207 if (!xchg(&root->orphan_item_inserted, 1))
2214 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2217 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2218 BTRFS_I(inode)->orphan_meta_reserved = 1;
2221 spin_unlock(&root->orphan_lock);
2224 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2226 /* grab metadata reservation from transaction handle */
2228 ret = btrfs_orphan_reserve_metadata(trans, inode);
2232 /* insert an orphan item to track this unlinked/truncated file */
2234 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2238 /* insert an orphan item to track subvolume contains orphan files */
2240 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2241 root->root_key.objectid);
2248 * We have done the truncate/delete so we can go ahead and remove the orphan
2249 * item for this particular inode.
2251 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2253 struct btrfs_root *root = BTRFS_I(inode)->root;
2254 int delete_item = 0;
2255 int release_rsv = 0;
2258 spin_lock(&root->orphan_lock);
2259 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2260 list_del_init(&BTRFS_I(inode)->i_orphan);
2264 if (BTRFS_I(inode)->orphan_meta_reserved) {
2265 BTRFS_I(inode)->orphan_meta_reserved = 0;
2268 spin_unlock(&root->orphan_lock);
2270 if (trans && delete_item) {
2271 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2276 btrfs_orphan_release_metadata(inode);
2282 * this cleans up any orphans that may be left on the list from the last use
2285 void btrfs_orphan_cleanup(struct btrfs_root *root)
2287 struct btrfs_path *path;
2288 struct extent_buffer *leaf;
2289 struct btrfs_key key, found_key;
2290 struct btrfs_trans_handle *trans;
2291 struct inode *inode;
2292 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2294 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2297 path = btrfs_alloc_path();
2301 key.objectid = BTRFS_ORPHAN_OBJECTID;
2302 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2303 key.offset = (u64)-1;
2306 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2308 printk(KERN_ERR "Error searching slot for orphan: %d"
2314 * if ret == 0 means we found what we were searching for, which
2315 * is weird, but possible, so only screw with path if we didnt
2316 * find the key and see if we have stuff that matches
2319 if (path->slots[0] == 0)
2324 /* pull out the item */
2325 leaf = path->nodes[0];
2326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2328 /* make sure the item matches what we want */
2329 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2331 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2334 /* release the path since we're done with it */
2335 btrfs_release_path(root, path);
2338 * this is where we are basically btrfs_lookup, without the
2339 * crossing root thing. we store the inode number in the
2340 * offset of the orphan item.
2342 found_key.objectid = found_key.offset;
2343 found_key.type = BTRFS_INODE_ITEM_KEY;
2344 found_key.offset = 0;
2345 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2346 BUG_ON(IS_ERR(inode));
2349 * add this inode to the orphan list so btrfs_orphan_del does
2350 * the proper thing when we hit it
2352 spin_lock(&root->orphan_lock);
2353 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2354 spin_unlock(&root->orphan_lock);
2357 * if this is a bad inode, means we actually succeeded in
2358 * removing the inode, but not the orphan record, which means
2359 * we need to manually delete the orphan since iput will just
2360 * do a destroy_inode
2362 if (is_bad_inode(inode)) {
2363 trans = btrfs_start_transaction(root, 0);
2364 BUG_ON(IS_ERR(trans));
2365 btrfs_orphan_del(trans, inode);
2366 btrfs_end_transaction(trans, root);
2371 /* if we have links, this was a truncate, lets do that */
2372 if (inode->i_nlink) {
2374 btrfs_truncate(inode);
2379 /* this will do delete_inode and everything for us */
2382 btrfs_free_path(path);
2384 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2386 if (root->orphan_block_rsv)
2387 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2390 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2391 trans = btrfs_join_transaction(root, 1);
2392 BUG_ON(IS_ERR(trans));
2393 btrfs_end_transaction(trans, root);
2397 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2399 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2403 * very simple check to peek ahead in the leaf looking for xattrs. If we
2404 * don't find any xattrs, we know there can't be any acls.
2406 * slot is the slot the inode is in, objectid is the objectid of the inode
2408 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2409 int slot, u64 objectid)
2411 u32 nritems = btrfs_header_nritems(leaf);
2412 struct btrfs_key found_key;
2416 while (slot < nritems) {
2417 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2419 /* we found a different objectid, there must not be acls */
2420 if (found_key.objectid != objectid)
2423 /* we found an xattr, assume we've got an acl */
2424 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2428 * we found a key greater than an xattr key, there can't
2429 * be any acls later on
2431 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2438 * it goes inode, inode backrefs, xattrs, extents,
2439 * so if there are a ton of hard links to an inode there can
2440 * be a lot of backrefs. Don't waste time searching too hard,
2441 * this is just an optimization
2446 /* we hit the end of the leaf before we found an xattr or
2447 * something larger than an xattr. We have to assume the inode
2454 * read an inode from the btree into the in-memory inode
2456 static void btrfs_read_locked_inode(struct inode *inode)
2458 struct btrfs_path *path;
2459 struct extent_buffer *leaf;
2460 struct btrfs_inode_item *inode_item;
2461 struct btrfs_timespec *tspec;
2462 struct btrfs_root *root = BTRFS_I(inode)->root;
2463 struct btrfs_key location;
2465 u64 alloc_group_block;
2469 path = btrfs_alloc_path();
2471 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2473 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2477 leaf = path->nodes[0];
2478 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2479 struct btrfs_inode_item);
2481 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2482 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2483 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2484 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2485 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2487 tspec = btrfs_inode_atime(inode_item);
2488 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2489 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2491 tspec = btrfs_inode_mtime(inode_item);
2492 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2493 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2495 tspec = btrfs_inode_ctime(inode_item);
2496 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2497 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2499 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2500 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2501 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2502 inode->i_generation = BTRFS_I(inode)->generation;
2504 rdev = btrfs_inode_rdev(leaf, inode_item);
2506 BTRFS_I(inode)->index_cnt = (u64)-1;
2507 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2509 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2512 * try to precache a NULL acl entry for files that don't have
2513 * any xattrs or acls
2515 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2517 cache_no_acl(inode);
2519 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2520 alloc_group_block, 0);
2521 btrfs_free_path(path);
2524 switch (inode->i_mode & S_IFMT) {
2526 inode->i_mapping->a_ops = &btrfs_aops;
2527 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2528 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2529 inode->i_fop = &btrfs_file_operations;
2530 inode->i_op = &btrfs_file_inode_operations;
2533 inode->i_fop = &btrfs_dir_file_operations;
2534 if (root == root->fs_info->tree_root)
2535 inode->i_op = &btrfs_dir_ro_inode_operations;
2537 inode->i_op = &btrfs_dir_inode_operations;
2540 inode->i_op = &btrfs_symlink_inode_operations;
2541 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2542 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2545 inode->i_op = &btrfs_special_inode_operations;
2546 init_special_inode(inode, inode->i_mode, rdev);
2550 btrfs_update_iflags(inode);
2554 btrfs_free_path(path);
2555 make_bad_inode(inode);
2559 * given a leaf and an inode, copy the inode fields into the leaf
2561 static void fill_inode_item(struct btrfs_trans_handle *trans,
2562 struct extent_buffer *leaf,
2563 struct btrfs_inode_item *item,
2564 struct inode *inode)
2566 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2567 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2568 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2569 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2570 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2572 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2573 inode->i_atime.tv_sec);
2574 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2575 inode->i_atime.tv_nsec);
2577 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2578 inode->i_mtime.tv_sec);
2579 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2580 inode->i_mtime.tv_nsec);
2582 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2583 inode->i_ctime.tv_sec);
2584 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2585 inode->i_ctime.tv_nsec);
2587 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2588 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2589 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2590 btrfs_set_inode_transid(leaf, item, trans->transid);
2591 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2592 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2593 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2597 * copy everything in the in-memory inode into the btree.
2599 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2600 struct btrfs_root *root, struct inode *inode)
2602 struct btrfs_inode_item *inode_item;
2603 struct btrfs_path *path;
2604 struct extent_buffer *leaf;
2607 path = btrfs_alloc_path();
2609 path->leave_spinning = 1;
2610 ret = btrfs_lookup_inode(trans, root, path,
2611 &BTRFS_I(inode)->location, 1);
2618 btrfs_unlock_up_safe(path, 1);
2619 leaf = path->nodes[0];
2620 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2621 struct btrfs_inode_item);
2623 fill_inode_item(trans, leaf, inode_item, inode);
2624 btrfs_mark_buffer_dirty(leaf);
2625 btrfs_set_inode_last_trans(trans, inode);
2628 btrfs_free_path(path);
2634 * unlink helper that gets used here in inode.c and in the tree logging
2635 * recovery code. It remove a link in a directory with a given name, and
2636 * also drops the back refs in the inode to the directory
2638 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2639 struct btrfs_root *root,
2640 struct inode *dir, struct inode *inode,
2641 const char *name, int name_len)
2643 struct btrfs_path *path;
2645 struct extent_buffer *leaf;
2646 struct btrfs_dir_item *di;
2647 struct btrfs_key key;
2650 path = btrfs_alloc_path();
2656 path->leave_spinning = 1;
2657 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2658 name, name_len, -1);
2667 leaf = path->nodes[0];
2668 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2669 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2672 btrfs_release_path(root, path);
2674 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2676 dir->i_ino, &index);
2678 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2679 "inode %lu parent %lu\n", name_len, name,
2680 inode->i_ino, dir->i_ino);
2684 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2685 index, name, name_len, -1);
2694 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2695 btrfs_release_path(root, path);
2697 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2699 BUG_ON(ret != 0 && ret != -ENOENT);
2701 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2706 btrfs_free_path(path);
2710 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2711 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2712 btrfs_update_inode(trans, root, dir);
2713 btrfs_drop_nlink(inode);
2714 ret = btrfs_update_inode(trans, root, inode);
2719 /* helper to check if there is any shared block in the path */
2720 static int check_path_shared(struct btrfs_root *root,
2721 struct btrfs_path *path)
2723 struct extent_buffer *eb;
2727 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2730 if (!path->nodes[level])
2732 eb = path->nodes[level];
2733 if (!btrfs_block_can_be_shared(root, eb))
2735 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2744 * helper to start transaction for unlink and rmdir.
2746 * unlink and rmdir are special in btrfs, they do not always free space.
2747 * so in enospc case, we should make sure they will free space before
2748 * allowing them to use the global metadata reservation.
2750 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2751 struct dentry *dentry)
2753 struct btrfs_trans_handle *trans;
2754 struct btrfs_root *root = BTRFS_I(dir)->root;
2755 struct btrfs_path *path;
2756 struct btrfs_inode_ref *ref;
2757 struct btrfs_dir_item *di;
2758 struct inode *inode = dentry->d_inode;
2764 trans = btrfs_start_transaction(root, 10);
2765 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2768 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2769 return ERR_PTR(-ENOSPC);
2771 /* check if there is someone else holds reference */
2772 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2773 return ERR_PTR(-ENOSPC);
2775 if (atomic_read(&inode->i_count) > 2)
2776 return ERR_PTR(-ENOSPC);
2778 if (xchg(&root->fs_info->enospc_unlink, 1))
2779 return ERR_PTR(-ENOSPC);
2781 path = btrfs_alloc_path();
2783 root->fs_info->enospc_unlink = 0;
2784 return ERR_PTR(-ENOMEM);
2787 trans = btrfs_start_transaction(root, 0);
2788 if (IS_ERR(trans)) {
2789 btrfs_free_path(path);
2790 root->fs_info->enospc_unlink = 0;
2794 path->skip_locking = 1;
2795 path->search_commit_root = 1;
2797 ret = btrfs_lookup_inode(trans, root, path,
2798 &BTRFS_I(dir)->location, 0);
2804 if (check_path_shared(root, path))
2809 btrfs_release_path(root, path);
2811 ret = btrfs_lookup_inode(trans, root, path,
2812 &BTRFS_I(inode)->location, 0);
2818 if (check_path_shared(root, path))
2823 btrfs_release_path(root, path);
2825 if (ret == 0 && S_ISREG(inode->i_mode)) {
2826 ret = btrfs_lookup_file_extent(trans, root, path,
2827 inode->i_ino, (u64)-1, 0);
2833 if (check_path_shared(root, path))
2835 btrfs_release_path(root, path);
2843 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2844 dentry->d_name.name, dentry->d_name.len, 0);
2850 if (check_path_shared(root, path))
2856 btrfs_release_path(root, path);
2858 ref = btrfs_lookup_inode_ref(trans, root, path,
2859 dentry->d_name.name, dentry->d_name.len,
2860 inode->i_ino, dir->i_ino, 0);
2866 if (check_path_shared(root, path))
2868 index = btrfs_inode_ref_index(path->nodes[0], ref);
2869 btrfs_release_path(root, path);
2871 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2872 dentry->d_name.name, dentry->d_name.len, 0);
2877 BUG_ON(ret == -ENOENT);
2878 if (check_path_shared(root, path))
2883 btrfs_free_path(path);
2885 btrfs_end_transaction(trans, root);
2886 root->fs_info->enospc_unlink = 0;
2887 return ERR_PTR(err);
2890 trans->block_rsv = &root->fs_info->global_block_rsv;
2894 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2895 struct btrfs_root *root)
2897 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2898 BUG_ON(!root->fs_info->enospc_unlink);
2899 root->fs_info->enospc_unlink = 0;
2901 btrfs_end_transaction_throttle(trans, root);
2904 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2906 struct btrfs_root *root = BTRFS_I(dir)->root;
2907 struct btrfs_trans_handle *trans;
2908 struct inode *inode = dentry->d_inode;
2910 unsigned long nr = 0;
2912 trans = __unlink_start_trans(dir, dentry);
2914 return PTR_ERR(trans);
2916 btrfs_set_trans_block_group(trans, dir);
2918 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2920 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2921 dentry->d_name.name, dentry->d_name.len);
2924 if (inode->i_nlink == 0) {
2925 ret = btrfs_orphan_add(trans, inode);
2929 nr = trans->blocks_used;
2930 __unlink_end_trans(trans, root);
2931 btrfs_btree_balance_dirty(root, nr);
2935 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2936 struct btrfs_root *root,
2937 struct inode *dir, u64 objectid,
2938 const char *name, int name_len)
2940 struct btrfs_path *path;
2941 struct extent_buffer *leaf;
2942 struct btrfs_dir_item *di;
2943 struct btrfs_key key;
2947 path = btrfs_alloc_path();
2951 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2952 name, name_len, -1);
2953 BUG_ON(!di || IS_ERR(di));
2955 leaf = path->nodes[0];
2956 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2957 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2958 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2960 btrfs_release_path(root, path);
2962 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2963 objectid, root->root_key.objectid,
2964 dir->i_ino, &index, name, name_len);
2966 BUG_ON(ret != -ENOENT);
2967 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2969 BUG_ON(!di || IS_ERR(di));
2971 leaf = path->nodes[0];
2972 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2973 btrfs_release_path(root, path);
2977 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2978 index, name, name_len, -1);
2979 BUG_ON(!di || IS_ERR(di));
2981 leaf = path->nodes[0];
2982 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2983 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2984 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2986 btrfs_release_path(root, path);
2988 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2989 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2990 ret = btrfs_update_inode(trans, root, dir);
2993 btrfs_free_path(path);
2997 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2999 struct inode *inode = dentry->d_inode;
3001 struct btrfs_root *root = BTRFS_I(dir)->root;
3002 struct btrfs_trans_handle *trans;
3003 unsigned long nr = 0;
3005 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3006 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3009 trans = __unlink_start_trans(dir, dentry);
3011 return PTR_ERR(trans);
3013 btrfs_set_trans_block_group(trans, dir);
3015 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3016 err = btrfs_unlink_subvol(trans, root, dir,
3017 BTRFS_I(inode)->location.objectid,
3018 dentry->d_name.name,
3019 dentry->d_name.len);
3023 err = btrfs_orphan_add(trans, inode);
3027 /* now the directory is empty */
3028 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3029 dentry->d_name.name, dentry->d_name.len);
3031 btrfs_i_size_write(inode, 0);
3033 nr = trans->blocks_used;
3034 __unlink_end_trans(trans, root);
3035 btrfs_btree_balance_dirty(root, nr);
3042 * when truncating bytes in a file, it is possible to avoid reading
3043 * the leaves that contain only checksum items. This can be the
3044 * majority of the IO required to delete a large file, but it must
3045 * be done carefully.
3047 * The keys in the level just above the leaves are checked to make sure
3048 * the lowest key in a given leaf is a csum key, and starts at an offset
3049 * after the new size.
3051 * Then the key for the next leaf is checked to make sure it also has
3052 * a checksum item for the same file. If it does, we know our target leaf
3053 * contains only checksum items, and it can be safely freed without reading
3056 * This is just an optimization targeted at large files. It may do
3057 * nothing. It will return 0 unless things went badly.
3059 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3060 struct btrfs_root *root,
3061 struct btrfs_path *path,
3062 struct inode *inode, u64 new_size)
3064 struct btrfs_key key;
3067 struct btrfs_key found_key;
3068 struct btrfs_key other_key;
3069 struct btrfs_leaf_ref *ref;
3073 path->lowest_level = 1;
3074 key.objectid = inode->i_ino;
3075 key.type = BTRFS_CSUM_ITEM_KEY;
3076 key.offset = new_size;
3078 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3082 if (path->nodes[1] == NULL) {
3087 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3088 nritems = btrfs_header_nritems(path->nodes[1]);
3093 if (path->slots[1] >= nritems)
3096 /* did we find a key greater than anything we want to delete? */
3097 if (found_key.objectid > inode->i_ino ||
3098 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3101 /* we check the next key in the node to make sure the leave contains
3102 * only checksum items. This comparison doesn't work if our
3103 * leaf is the last one in the node
3105 if (path->slots[1] + 1 >= nritems) {
3107 /* search forward from the last key in the node, this
3108 * will bring us into the next node in the tree
3110 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3112 /* unlikely, but we inc below, so check to be safe */
3113 if (found_key.offset == (u64)-1)
3116 /* search_forward needs a path with locks held, do the
3117 * search again for the original key. It is possible
3118 * this will race with a balance and return a path that
3119 * we could modify, but this drop is just an optimization
3120 * and is allowed to miss some leaves.
3122 btrfs_release_path(root, path);
3125 /* setup a max key for search_forward */
3126 other_key.offset = (u64)-1;
3127 other_key.type = key.type;
3128 other_key.objectid = key.objectid;
3130 path->keep_locks = 1;
3131 ret = btrfs_search_forward(root, &found_key, &other_key,
3133 path->keep_locks = 0;
3134 if (ret || found_key.objectid != key.objectid ||
3135 found_key.type != key.type) {
3140 key.offset = found_key.offset;
3141 btrfs_release_path(root, path);
3146 /* we know there's one more slot after us in the tree,
3147 * read that key so we can verify it is also a checksum item
3149 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3151 if (found_key.objectid < inode->i_ino)
3154 if (found_key.type != key.type || found_key.offset < new_size)
3158 * if the key for the next leaf isn't a csum key from this objectid,
3159 * we can't be sure there aren't good items inside this leaf.
3162 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3165 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3166 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3168 * it is safe to delete this leaf, it contains only
3169 * csum items from this inode at an offset >= new_size
3171 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3174 if (root->ref_cows && leaf_gen < trans->transid) {
3175 ref = btrfs_alloc_leaf_ref(root, 0);
3177 ref->root_gen = root->root_key.offset;
3178 ref->bytenr = leaf_start;
3180 ref->generation = leaf_gen;
3183 btrfs_sort_leaf_ref(ref);
3185 ret = btrfs_add_leaf_ref(root, ref, 0);
3187 btrfs_free_leaf_ref(root, ref);
3193 btrfs_release_path(root, path);
3195 if (other_key.objectid == inode->i_ino &&
3196 other_key.type == key.type && other_key.offset > key.offset) {
3197 key.offset = other_key.offset;
3203 /* fixup any changes we've made to the path */
3204 path->lowest_level = 0;
3205 path->keep_locks = 0;
3206 btrfs_release_path(root, path);
3213 * this can truncate away extent items, csum items and directory items.
3214 * It starts at a high offset and removes keys until it can't find
3215 * any higher than new_size
3217 * csum items that cross the new i_size are truncated to the new size
3220 * min_type is the minimum key type to truncate down to. If set to 0, this
3221 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3223 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3224 struct btrfs_root *root,
3225 struct inode *inode,
3226 u64 new_size, u32 min_type)
3228 struct btrfs_path *path;
3229 struct extent_buffer *leaf;
3230 struct btrfs_file_extent_item *fi;
3231 struct btrfs_key key;
3232 struct btrfs_key found_key;
3233 u64 extent_start = 0;
3234 u64 extent_num_bytes = 0;
3235 u64 extent_offset = 0;
3237 u64 mask = root->sectorsize - 1;
3238 u32 found_type = (u8)-1;
3241 int pending_del_nr = 0;
3242 int pending_del_slot = 0;
3243 int extent_type = -1;
3248 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3250 if (root->ref_cows || root == root->fs_info->tree_root)
3251 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3253 path = btrfs_alloc_path();
3257 key.objectid = inode->i_ino;
3258 key.offset = (u64)-1;
3262 path->leave_spinning = 1;
3263 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3270 /* there are no items in the tree for us to truncate, we're
3273 if (path->slots[0] == 0)
3280 leaf = path->nodes[0];
3281 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3282 found_type = btrfs_key_type(&found_key);
3285 if (found_key.objectid != inode->i_ino)
3288 if (found_type < min_type)
3291 item_end = found_key.offset;
3292 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3293 fi = btrfs_item_ptr(leaf, path->slots[0],
3294 struct btrfs_file_extent_item);
3295 extent_type = btrfs_file_extent_type(leaf, fi);
3296 encoding = btrfs_file_extent_compression(leaf, fi);
3297 encoding |= btrfs_file_extent_encryption(leaf, fi);
3298 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3300 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3302 btrfs_file_extent_num_bytes(leaf, fi);
3303 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3304 item_end += btrfs_file_extent_inline_len(leaf,
3309 if (found_type > min_type) {
3312 if (item_end < new_size)
3314 if (found_key.offset >= new_size)
3320 /* FIXME, shrink the extent if the ref count is only 1 */
3321 if (found_type != BTRFS_EXTENT_DATA_KEY)
3324 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3326 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3327 if (!del_item && !encoding) {
3328 u64 orig_num_bytes =
3329 btrfs_file_extent_num_bytes(leaf, fi);
3330 extent_num_bytes = new_size -
3331 found_key.offset + root->sectorsize - 1;
3332 extent_num_bytes = extent_num_bytes &
3333 ~((u64)root->sectorsize - 1);
3334 btrfs_set_file_extent_num_bytes(leaf, fi,
3336 num_dec = (orig_num_bytes -
3338 if (root->ref_cows && extent_start != 0)
3339 inode_sub_bytes(inode, num_dec);
3340 btrfs_mark_buffer_dirty(leaf);
3343 btrfs_file_extent_disk_num_bytes(leaf,
3345 extent_offset = found_key.offset -
3346 btrfs_file_extent_offset(leaf, fi);
3348 /* FIXME blocksize != 4096 */
3349 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3350 if (extent_start != 0) {
3353 inode_sub_bytes(inode, num_dec);
3356 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3358 * we can't truncate inline items that have had
3362 btrfs_file_extent_compression(leaf, fi) == 0 &&
3363 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3364 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3365 u32 size = new_size - found_key.offset;
3367 if (root->ref_cows) {
3368 inode_sub_bytes(inode, item_end + 1 -
3372 btrfs_file_extent_calc_inline_size(size);
3373 ret = btrfs_truncate_item(trans, root, path,
3376 } else if (root->ref_cows) {
3377 inode_sub_bytes(inode, item_end + 1 -
3383 if (!pending_del_nr) {
3384 /* no pending yet, add ourselves */
3385 pending_del_slot = path->slots[0];
3387 } else if (pending_del_nr &&
3388 path->slots[0] + 1 == pending_del_slot) {
3389 /* hop on the pending chunk */
3391 pending_del_slot = path->slots[0];
3398 if (found_extent && (root->ref_cows ||
3399 root == root->fs_info->tree_root)) {
3400 btrfs_set_path_blocking(path);
3401 ret = btrfs_free_extent(trans, root, extent_start,
3402 extent_num_bytes, 0,
3403 btrfs_header_owner(leaf),
3404 inode->i_ino, extent_offset);
3408 if (found_type == BTRFS_INODE_ITEM_KEY)
3411 if (path->slots[0] == 0 ||
3412 path->slots[0] != pending_del_slot) {
3413 if (root->ref_cows) {
3417 if (pending_del_nr) {
3418 ret = btrfs_del_items(trans, root, path,
3424 btrfs_release_path(root, path);
3431 if (pending_del_nr) {
3432 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3436 btrfs_free_path(path);
3441 * taken from block_truncate_page, but does cow as it zeros out
3442 * any bytes left in the last page in the file.
3444 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3446 struct inode *inode = mapping->host;
3447 struct btrfs_root *root = BTRFS_I(inode)->root;
3448 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3449 struct btrfs_ordered_extent *ordered;
3450 struct extent_state *cached_state = NULL;
3452 u32 blocksize = root->sectorsize;
3453 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3454 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3460 if ((offset & (blocksize - 1)) == 0)
3462 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3468 page = grab_cache_page(mapping, index);
3470 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3474 page_start = page_offset(page);
3475 page_end = page_start + PAGE_CACHE_SIZE - 1;
3477 if (!PageUptodate(page)) {
3478 ret = btrfs_readpage(NULL, page);
3480 if (page->mapping != mapping) {
3482 page_cache_release(page);
3485 if (!PageUptodate(page)) {
3490 wait_on_page_writeback(page);
3492 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3494 set_page_extent_mapped(page);
3496 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3498 unlock_extent_cached(io_tree, page_start, page_end,
3499 &cached_state, GFP_NOFS);
3501 page_cache_release(page);
3502 btrfs_start_ordered_extent(inode, ordered, 1);
3503 btrfs_put_ordered_extent(ordered);
3507 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3508 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3509 0, 0, &cached_state, GFP_NOFS);
3511 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3514 unlock_extent_cached(io_tree, page_start, page_end,
3515 &cached_state, GFP_NOFS);
3520 if (offset != PAGE_CACHE_SIZE) {
3522 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3523 flush_dcache_page(page);
3526 ClearPageChecked(page);
3527 set_page_dirty(page);
3528 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3533 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3535 page_cache_release(page);
3540 int btrfs_cont_expand(struct inode *inode, loff_t size)
3542 struct btrfs_trans_handle *trans;
3543 struct btrfs_root *root = BTRFS_I(inode)->root;
3544 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3545 struct extent_map *em = NULL;
3546 struct extent_state *cached_state = NULL;
3547 u64 mask = root->sectorsize - 1;
3548 u64 hole_start = (inode->i_size + mask) & ~mask;
3549 u64 block_end = (size + mask) & ~mask;
3555 if (size <= hole_start)
3559 struct btrfs_ordered_extent *ordered;
3560 btrfs_wait_ordered_range(inode, hole_start,
3561 block_end - hole_start);
3562 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3563 &cached_state, GFP_NOFS);
3564 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3567 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3568 &cached_state, GFP_NOFS);
3569 btrfs_put_ordered_extent(ordered);
3572 cur_offset = hole_start;
3574 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3575 block_end - cur_offset, 0);
3576 BUG_ON(IS_ERR(em) || !em);
3577 last_byte = min(extent_map_end(em), block_end);
3578 last_byte = (last_byte + mask) & ~mask;
3579 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3581 hole_size = last_byte - cur_offset;
3583 trans = btrfs_start_transaction(root, 2);
3584 if (IS_ERR(trans)) {
3585 err = PTR_ERR(trans);
3588 btrfs_set_trans_block_group(trans, inode);
3590 err = btrfs_drop_extents(trans, inode, cur_offset,
3591 cur_offset + hole_size,
3595 err = btrfs_insert_file_extent(trans, root,
3596 inode->i_ino, cur_offset, 0,
3597 0, hole_size, 0, hole_size,
3601 btrfs_drop_extent_cache(inode, hole_start,
3604 btrfs_end_transaction(trans, root);
3606 free_extent_map(em);
3608 cur_offset = last_byte;
3609 if (cur_offset >= block_end)
3613 free_extent_map(em);
3614 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3619 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3621 struct btrfs_root *root = BTRFS_I(inode)->root;
3622 struct btrfs_trans_handle *trans;
3626 if (attr->ia_size == inode->i_size)
3629 if (attr->ia_size > inode->i_size) {
3630 unsigned long limit;
3631 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3632 if (attr->ia_size > inode->i_sb->s_maxbytes)
3634 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3635 send_sig(SIGXFSZ, current, 0);
3640 trans = btrfs_start_transaction(root, 5);
3642 return PTR_ERR(trans);
3644 btrfs_set_trans_block_group(trans, inode);
3646 ret = btrfs_orphan_add(trans, inode);
3649 nr = trans->blocks_used;
3650 btrfs_end_transaction(trans, root);
3651 btrfs_btree_balance_dirty(root, nr);
3653 if (attr->ia_size > inode->i_size) {
3654 ret = btrfs_cont_expand(inode, attr->ia_size);
3656 btrfs_truncate(inode);
3660 i_size_write(inode, attr->ia_size);
3661 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3663 trans = btrfs_start_transaction(root, 0);
3664 BUG_ON(IS_ERR(trans));
3665 btrfs_set_trans_block_group(trans, inode);
3666 trans->block_rsv = root->orphan_block_rsv;
3667 BUG_ON(!trans->block_rsv);
3669 ret = btrfs_update_inode(trans, root, inode);
3671 if (inode->i_nlink > 0) {
3672 ret = btrfs_orphan_del(trans, inode);
3675 nr = trans->blocks_used;
3676 btrfs_end_transaction(trans, root);
3677 btrfs_btree_balance_dirty(root, nr);
3682 * We're truncating a file that used to have good data down to
3683 * zero. Make sure it gets into the ordered flush list so that
3684 * any new writes get down to disk quickly.
3686 if (attr->ia_size == 0)
3687 BTRFS_I(inode)->ordered_data_close = 1;
3689 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3690 ret = vmtruncate(inode, attr->ia_size);
3696 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3698 struct inode *inode = dentry->d_inode;
3699 struct btrfs_root *root = BTRFS_I(inode)->root;
3702 if (btrfs_root_readonly(root))
3705 err = inode_change_ok(inode, attr);
3709 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3710 err = btrfs_setattr_size(inode, attr);
3715 if (attr->ia_valid) {
3716 setattr_copy(inode, attr);
3717 mark_inode_dirty(inode);
3719 if (attr->ia_valid & ATTR_MODE)
3720 err = btrfs_acl_chmod(inode);
3726 void btrfs_evict_inode(struct inode *inode)
3728 struct btrfs_trans_handle *trans;
3729 struct btrfs_root *root = BTRFS_I(inode)->root;
3733 truncate_inode_pages(&inode->i_data, 0);
3734 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3735 root == root->fs_info->tree_root))
3738 if (is_bad_inode(inode)) {
3739 btrfs_orphan_del(NULL, inode);
3742 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3743 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3745 if (root->fs_info->log_root_recovering) {
3746 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3750 if (inode->i_nlink > 0) {
3751 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3755 btrfs_i_size_write(inode, 0);
3758 trans = btrfs_start_transaction(root, 0);
3759 BUG_ON(IS_ERR(trans));
3760 btrfs_set_trans_block_group(trans, inode);
3761 trans->block_rsv = root->orphan_block_rsv;
3763 ret = btrfs_block_rsv_check(trans, root,
3764 root->orphan_block_rsv, 0, 5);
3766 BUG_ON(ret != -EAGAIN);
3767 ret = btrfs_commit_transaction(trans, root);
3772 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3776 nr = trans->blocks_used;
3777 btrfs_end_transaction(trans, root);
3779 btrfs_btree_balance_dirty(root, nr);
3784 ret = btrfs_orphan_del(trans, inode);
3788 nr = trans->blocks_used;
3789 btrfs_end_transaction(trans, root);
3790 btrfs_btree_balance_dirty(root, nr);
3792 end_writeback(inode);
3797 * this returns the key found in the dir entry in the location pointer.
3798 * If no dir entries were found, location->objectid is 0.
3800 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3801 struct btrfs_key *location)
3803 const char *name = dentry->d_name.name;
3804 int namelen = dentry->d_name.len;
3805 struct btrfs_dir_item *di;
3806 struct btrfs_path *path;
3807 struct btrfs_root *root = BTRFS_I(dir)->root;
3810 path = btrfs_alloc_path();
3813 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3818 if (!di || IS_ERR(di))
3821 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3823 btrfs_free_path(path);
3826 location->objectid = 0;
3831 * when we hit a tree root in a directory, the btrfs part of the inode
3832 * needs to be changed to reflect the root directory of the tree root. This
3833 * is kind of like crossing a mount point.
3835 static int fixup_tree_root_location(struct btrfs_root *root,
3837 struct dentry *dentry,
3838 struct btrfs_key *location,
3839 struct btrfs_root **sub_root)
3841 struct btrfs_path *path;
3842 struct btrfs_root *new_root;
3843 struct btrfs_root_ref *ref;
3844 struct extent_buffer *leaf;
3848 path = btrfs_alloc_path();
3855 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3856 BTRFS_I(dir)->root->root_key.objectid,
3857 location->objectid);
3864 leaf = path->nodes[0];
3865 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3866 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3867 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3870 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3871 (unsigned long)(ref + 1),
3872 dentry->d_name.len);
3876 btrfs_release_path(root->fs_info->tree_root, path);
3878 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3879 if (IS_ERR(new_root)) {
3880 err = PTR_ERR(new_root);
3884 if (btrfs_root_refs(&new_root->root_item) == 0) {
3889 *sub_root = new_root;
3890 location->objectid = btrfs_root_dirid(&new_root->root_item);
3891 location->type = BTRFS_INODE_ITEM_KEY;
3892 location->offset = 0;
3895 btrfs_free_path(path);
3899 static void inode_tree_add(struct inode *inode)
3901 struct btrfs_root *root = BTRFS_I(inode)->root;
3902 struct btrfs_inode *entry;
3904 struct rb_node *parent;
3906 p = &root->inode_tree.rb_node;
3909 if (inode_unhashed(inode))
3912 spin_lock(&root->inode_lock);
3915 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3917 if (inode->i_ino < entry->vfs_inode.i_ino)
3918 p = &parent->rb_left;
3919 else if (inode->i_ino > entry->vfs_inode.i_ino)
3920 p = &parent->rb_right;
3922 WARN_ON(!(entry->vfs_inode.i_state &
3923 (I_WILL_FREE | I_FREEING)));
3924 rb_erase(parent, &root->inode_tree);
3925 RB_CLEAR_NODE(parent);
3926 spin_unlock(&root->inode_lock);
3930 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3931 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3932 spin_unlock(&root->inode_lock);
3935 static void inode_tree_del(struct inode *inode)
3937 struct btrfs_root *root = BTRFS_I(inode)->root;
3940 spin_lock(&root->inode_lock);
3941 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3942 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3943 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3944 empty = RB_EMPTY_ROOT(&root->inode_tree);
3946 spin_unlock(&root->inode_lock);
3949 * Free space cache has inodes in the tree root, but the tree root has a
3950 * root_refs of 0, so this could end up dropping the tree root as a
3951 * snapshot, so we need the extra !root->fs_info->tree_root check to
3952 * make sure we don't drop it.
3954 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3955 root != root->fs_info->tree_root) {
3956 synchronize_srcu(&root->fs_info->subvol_srcu);
3957 spin_lock(&root->inode_lock);
3958 empty = RB_EMPTY_ROOT(&root->inode_tree);
3959 spin_unlock(&root->inode_lock);
3961 btrfs_add_dead_root(root);
3965 int btrfs_invalidate_inodes(struct btrfs_root *root)
3967 struct rb_node *node;
3968 struct rb_node *prev;
3969 struct btrfs_inode *entry;
3970 struct inode *inode;
3973 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3975 spin_lock(&root->inode_lock);
3977 node = root->inode_tree.rb_node;
3981 entry = rb_entry(node, struct btrfs_inode, rb_node);
3983 if (objectid < entry->vfs_inode.i_ino)
3984 node = node->rb_left;
3985 else if (objectid > entry->vfs_inode.i_ino)
3986 node = node->rb_right;
3992 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3993 if (objectid <= entry->vfs_inode.i_ino) {
3997 prev = rb_next(prev);
4001 entry = rb_entry(node, struct btrfs_inode, rb_node);
4002 objectid = entry->vfs_inode.i_ino + 1;
4003 inode = igrab(&entry->vfs_inode);
4005 spin_unlock(&root->inode_lock);
4006 if (atomic_read(&inode->i_count) > 1)
4007 d_prune_aliases(inode);
4009 * btrfs_drop_inode will have it removed from
4010 * the inode cache when its usage count
4015 spin_lock(&root->inode_lock);
4019 if (cond_resched_lock(&root->inode_lock))
4022 node = rb_next(node);
4024 spin_unlock(&root->inode_lock);
4028 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4030 struct btrfs_iget_args *args = p;
4031 inode->i_ino = args->ino;
4032 BTRFS_I(inode)->root = args->root;
4033 btrfs_set_inode_space_info(args->root, inode);
4037 static int btrfs_find_actor(struct inode *inode, void *opaque)
4039 struct btrfs_iget_args *args = opaque;
4040 return args->ino == inode->i_ino &&
4041 args->root == BTRFS_I(inode)->root;
4044 static struct inode *btrfs_iget_locked(struct super_block *s,
4046 struct btrfs_root *root)
4048 struct inode *inode;
4049 struct btrfs_iget_args args;
4050 args.ino = objectid;
4053 inode = iget5_locked(s, objectid, btrfs_find_actor,
4054 btrfs_init_locked_inode,
4059 /* Get an inode object given its location and corresponding root.
4060 * Returns in *is_new if the inode was read from disk
4062 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4063 struct btrfs_root *root, int *new)
4065 struct inode *inode;
4067 inode = btrfs_iget_locked(s, location->objectid, root);
4069 return ERR_PTR(-ENOMEM);
4071 if (inode->i_state & I_NEW) {
4072 BTRFS_I(inode)->root = root;
4073 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4074 btrfs_read_locked_inode(inode);
4076 inode_tree_add(inode);
4077 unlock_new_inode(inode);
4085 static struct inode *new_simple_dir(struct super_block *s,
4086 struct btrfs_key *key,
4087 struct btrfs_root *root)
4089 struct inode *inode = new_inode(s);
4092 return ERR_PTR(-ENOMEM);
4094 BTRFS_I(inode)->root = root;
4095 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4096 BTRFS_I(inode)->dummy_inode = 1;
4098 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4099 inode->i_op = &simple_dir_inode_operations;
4100 inode->i_fop = &simple_dir_operations;
4101 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4102 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4107 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4109 struct inode *inode;
4110 struct btrfs_root *root = BTRFS_I(dir)->root;
4111 struct btrfs_root *sub_root = root;
4112 struct btrfs_key location;
4116 if (dentry->d_name.len > BTRFS_NAME_LEN)
4117 return ERR_PTR(-ENAMETOOLONG);
4119 ret = btrfs_inode_by_name(dir, dentry, &location);
4122 return ERR_PTR(ret);
4124 if (location.objectid == 0)
4127 if (location.type == BTRFS_INODE_ITEM_KEY) {
4128 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4132 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4134 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4135 ret = fixup_tree_root_location(root, dir, dentry,
4136 &location, &sub_root);
4139 inode = ERR_PTR(ret);
4141 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4143 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4145 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4147 if (!IS_ERR(inode) && root != sub_root) {
4148 down_read(&root->fs_info->cleanup_work_sem);
4149 if (!(inode->i_sb->s_flags & MS_RDONLY))
4150 btrfs_orphan_cleanup(sub_root);
4151 up_read(&root->fs_info->cleanup_work_sem);
4157 static int btrfs_dentry_delete(const struct dentry *dentry)
4159 struct btrfs_root *root;
4161 if (!dentry->d_inode && !IS_ROOT(dentry))
4162 dentry = dentry->d_parent;
4164 if (dentry->d_inode) {
4165 root = BTRFS_I(dentry->d_inode)->root;
4166 if (btrfs_root_refs(&root->root_item) == 0)
4172 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4173 struct nameidata *nd)
4175 struct inode *inode;
4177 inode = btrfs_lookup_dentry(dir, dentry);
4179 return ERR_CAST(inode);
4181 return d_splice_alias(inode, dentry);
4184 static unsigned char btrfs_filetype_table[] = {
4185 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4188 static int btrfs_real_readdir(struct file *filp, void *dirent,
4191 struct inode *inode = filp->f_dentry->d_inode;
4192 struct btrfs_root *root = BTRFS_I(inode)->root;
4193 struct btrfs_item *item;
4194 struct btrfs_dir_item *di;
4195 struct btrfs_key key;
4196 struct btrfs_key found_key;
4197 struct btrfs_path *path;
4200 struct extent_buffer *leaf;
4203 unsigned char d_type;
4208 int key_type = BTRFS_DIR_INDEX_KEY;
4213 /* FIXME, use a real flag for deciding about the key type */
4214 if (root->fs_info->tree_root == root)
4215 key_type = BTRFS_DIR_ITEM_KEY;
4217 /* special case for "." */
4218 if (filp->f_pos == 0) {
4219 over = filldir(dirent, ".", 1,
4226 /* special case for .., just use the back ref */
4227 if (filp->f_pos == 1) {
4228 u64 pino = parent_ino(filp->f_path.dentry);
4229 over = filldir(dirent, "..", 2,
4235 path = btrfs_alloc_path();
4238 btrfs_set_key_type(&key, key_type);
4239 key.offset = filp->f_pos;
4240 key.objectid = inode->i_ino;
4242 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4248 leaf = path->nodes[0];
4249 nritems = btrfs_header_nritems(leaf);
4250 slot = path->slots[0];
4251 if (advance || slot >= nritems) {
4252 if (slot >= nritems - 1) {
4253 ret = btrfs_next_leaf(root, path);
4256 leaf = path->nodes[0];
4257 nritems = btrfs_header_nritems(leaf);
4258 slot = path->slots[0];
4266 item = btrfs_item_nr(leaf, slot);
4267 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4269 if (found_key.objectid != key.objectid)
4271 if (btrfs_key_type(&found_key) != key_type)
4273 if (found_key.offset < filp->f_pos)
4276 filp->f_pos = found_key.offset;
4278 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4280 di_total = btrfs_item_size(leaf, item);
4282 while (di_cur < di_total) {
4283 struct btrfs_key location;
4285 name_len = btrfs_dir_name_len(leaf, di);
4286 if (name_len <= sizeof(tmp_name)) {
4287 name_ptr = tmp_name;
4289 name_ptr = kmalloc(name_len, GFP_NOFS);
4295 read_extent_buffer(leaf, name_ptr,
4296 (unsigned long)(di + 1), name_len);
4298 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4299 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4301 /* is this a reference to our own snapshot? If so
4304 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4305 location.objectid == root->root_key.objectid) {
4309 over = filldir(dirent, name_ptr, name_len,
4310 found_key.offset, location.objectid,
4314 if (name_ptr != tmp_name)
4319 di_len = btrfs_dir_name_len(leaf, di) +
4320 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4322 di = (struct btrfs_dir_item *)((char *)di + di_len);
4326 /* Reached end of directory/root. Bump pos past the last item. */
4327 if (key_type == BTRFS_DIR_INDEX_KEY)
4329 * 32-bit glibc will use getdents64, but then strtol -
4330 * so the last number we can serve is this.
4332 filp->f_pos = 0x7fffffff;
4338 btrfs_free_path(path);
4342 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4344 struct btrfs_root *root = BTRFS_I(inode)->root;
4345 struct btrfs_trans_handle *trans;
4347 bool nolock = false;
4349 if (BTRFS_I(inode)->dummy_inode)
4353 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4355 if (wbc->sync_mode == WB_SYNC_ALL) {
4357 trans = btrfs_join_transaction_nolock(root, 1);
4359 trans = btrfs_join_transaction(root, 1);
4361 return PTR_ERR(trans);
4362 btrfs_set_trans_block_group(trans, inode);
4364 ret = btrfs_end_transaction_nolock(trans, root);
4366 ret = btrfs_commit_transaction(trans, root);
4372 * This is somewhat expensive, updating the tree every time the
4373 * inode changes. But, it is most likely to find the inode in cache.
4374 * FIXME, needs more benchmarking...there are no reasons other than performance
4375 * to keep or drop this code.
4377 void btrfs_dirty_inode(struct inode *inode)
4379 struct btrfs_root *root = BTRFS_I(inode)->root;
4380 struct btrfs_trans_handle *trans;
4383 if (BTRFS_I(inode)->dummy_inode)
4386 trans = btrfs_join_transaction(root, 1);
4387 BUG_ON(IS_ERR(trans));
4388 btrfs_set_trans_block_group(trans, inode);
4390 ret = btrfs_update_inode(trans, root, inode);
4391 if (ret && ret == -ENOSPC) {
4392 /* whoops, lets try again with the full transaction */
4393 btrfs_end_transaction(trans, root);
4394 trans = btrfs_start_transaction(root, 1);
4395 if (IS_ERR(trans)) {
4396 if (printk_ratelimit()) {
4397 printk(KERN_ERR "btrfs: fail to "
4398 "dirty inode %lu error %ld\n",
4399 inode->i_ino, PTR_ERR(trans));
4403 btrfs_set_trans_block_group(trans, inode);
4405 ret = btrfs_update_inode(trans, root, inode);
4407 if (printk_ratelimit()) {
4408 printk(KERN_ERR "btrfs: fail to "
4409 "dirty inode %lu error %d\n",
4414 btrfs_end_transaction(trans, root);
4418 * find the highest existing sequence number in a directory
4419 * and then set the in-memory index_cnt variable to reflect
4420 * free sequence numbers
4422 static int btrfs_set_inode_index_count(struct inode *inode)
4424 struct btrfs_root *root = BTRFS_I(inode)->root;
4425 struct btrfs_key key, found_key;
4426 struct btrfs_path *path;
4427 struct extent_buffer *leaf;
4430 key.objectid = inode->i_ino;
4431 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4432 key.offset = (u64)-1;
4434 path = btrfs_alloc_path();
4438 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4441 /* FIXME: we should be able to handle this */
4447 * MAGIC NUMBER EXPLANATION:
4448 * since we search a directory based on f_pos we have to start at 2
4449 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4450 * else has to start at 2
4452 if (path->slots[0] == 0) {
4453 BTRFS_I(inode)->index_cnt = 2;
4459 leaf = path->nodes[0];
4460 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4462 if (found_key.objectid != inode->i_ino ||
4463 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4464 BTRFS_I(inode)->index_cnt = 2;
4468 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4470 btrfs_free_path(path);
4475 * helper to find a free sequence number in a given directory. This current
4476 * code is very simple, later versions will do smarter things in the btree
4478 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4482 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4483 ret = btrfs_set_inode_index_count(dir);
4488 *index = BTRFS_I(dir)->index_cnt;
4489 BTRFS_I(dir)->index_cnt++;
4494 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4495 struct btrfs_root *root,
4497 const char *name, int name_len,
4498 u64 ref_objectid, u64 objectid,
4499 u64 alloc_hint, int mode, u64 *index)
4501 struct inode *inode;
4502 struct btrfs_inode_item *inode_item;
4503 struct btrfs_key *location;
4504 struct btrfs_path *path;
4505 struct btrfs_inode_ref *ref;
4506 struct btrfs_key key[2];
4512 path = btrfs_alloc_path();
4515 inode = new_inode(root->fs_info->sb);
4517 return ERR_PTR(-ENOMEM);
4520 ret = btrfs_set_inode_index(dir, index);
4523 return ERR_PTR(ret);
4527 * index_cnt is ignored for everything but a dir,
4528 * btrfs_get_inode_index_count has an explanation for the magic
4531 BTRFS_I(inode)->index_cnt = 2;
4532 BTRFS_I(inode)->root = root;
4533 BTRFS_I(inode)->generation = trans->transid;
4534 inode->i_generation = BTRFS_I(inode)->generation;
4535 btrfs_set_inode_space_info(root, inode);
4541 BTRFS_I(inode)->block_group =
4542 btrfs_find_block_group(root, 0, alloc_hint, owner);
4544 key[0].objectid = objectid;
4545 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4548 key[1].objectid = objectid;
4549 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4550 key[1].offset = ref_objectid;
4552 sizes[0] = sizeof(struct btrfs_inode_item);
4553 sizes[1] = name_len + sizeof(*ref);
4555 path->leave_spinning = 1;
4556 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4560 inode_init_owner(inode, dir, mode);
4561 inode->i_ino = objectid;
4562 inode_set_bytes(inode, 0);
4563 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4564 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4565 struct btrfs_inode_item);
4566 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4568 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4569 struct btrfs_inode_ref);
4570 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4571 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4572 ptr = (unsigned long)(ref + 1);
4573 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4575 btrfs_mark_buffer_dirty(path->nodes[0]);
4576 btrfs_free_path(path);
4578 location = &BTRFS_I(inode)->location;
4579 location->objectid = objectid;
4580 location->offset = 0;
4581 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4583 btrfs_inherit_iflags(inode, dir);
4585 if ((mode & S_IFREG)) {
4586 if (btrfs_test_opt(root, NODATASUM))
4587 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4588 if (btrfs_test_opt(root, NODATACOW))
4589 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4592 insert_inode_hash(inode);
4593 inode_tree_add(inode);
4597 BTRFS_I(dir)->index_cnt--;
4598 btrfs_free_path(path);
4600 return ERR_PTR(ret);
4603 static inline u8 btrfs_inode_type(struct inode *inode)
4605 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4609 * utility function to add 'inode' into 'parent_inode' with
4610 * a give name and a given sequence number.
4611 * if 'add_backref' is true, also insert a backref from the
4612 * inode to the parent directory.
4614 int btrfs_add_link(struct btrfs_trans_handle *trans,
4615 struct inode *parent_inode, struct inode *inode,
4616 const char *name, int name_len, int add_backref, u64 index)
4619 struct btrfs_key key;
4620 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4622 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4623 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4625 key.objectid = inode->i_ino;
4626 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4630 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4631 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4632 key.objectid, root->root_key.objectid,
4633 parent_inode->i_ino,
4634 index, name, name_len);
4635 } else if (add_backref) {
4636 ret = btrfs_insert_inode_ref(trans, root,
4637 name, name_len, inode->i_ino,
4638 parent_inode->i_ino, index);
4642 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4643 parent_inode->i_ino, &key,
4644 btrfs_inode_type(inode), index);
4647 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4649 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4650 ret = btrfs_update_inode(trans, root, parent_inode);
4655 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4656 struct inode *dir, struct dentry *dentry,
4657 struct inode *inode, int backref, u64 index)
4659 int err = btrfs_add_link(trans, dir, inode,
4660 dentry->d_name.name, dentry->d_name.len,
4663 d_instantiate(dentry, inode);
4671 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4672 int mode, dev_t rdev)
4674 struct btrfs_trans_handle *trans;
4675 struct btrfs_root *root = BTRFS_I(dir)->root;
4676 struct inode *inode = NULL;
4680 unsigned long nr = 0;
4683 if (!new_valid_dev(rdev))
4686 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4691 * 2 for inode item and ref
4693 * 1 for xattr if selinux is on
4695 trans = btrfs_start_transaction(root, 5);
4697 return PTR_ERR(trans);
4699 btrfs_set_trans_block_group(trans, dir);
4701 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4702 dentry->d_name.len, dir->i_ino, objectid,
4703 BTRFS_I(dir)->block_group, mode, &index);
4704 err = PTR_ERR(inode);
4708 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4714 btrfs_set_trans_block_group(trans, inode);
4715 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4719 inode->i_op = &btrfs_special_inode_operations;
4720 init_special_inode(inode, inode->i_mode, rdev);
4721 btrfs_update_inode(trans, root, inode);
4723 btrfs_update_inode_block_group(trans, inode);
4724 btrfs_update_inode_block_group(trans, dir);
4726 nr = trans->blocks_used;
4727 btrfs_end_transaction_throttle(trans, root);
4728 btrfs_btree_balance_dirty(root, nr);
4730 inode_dec_link_count(inode);
4736 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4737 int mode, struct nameidata *nd)
4739 struct btrfs_trans_handle *trans;
4740 struct btrfs_root *root = BTRFS_I(dir)->root;
4741 struct inode *inode = NULL;
4744 unsigned long nr = 0;
4748 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4752 * 2 for inode item and ref
4754 * 1 for xattr if selinux is on
4756 trans = btrfs_start_transaction(root, 5);
4758 return PTR_ERR(trans);
4760 btrfs_set_trans_block_group(trans, dir);
4762 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4763 dentry->d_name.len, dir->i_ino, objectid,
4764 BTRFS_I(dir)->block_group, mode, &index);
4765 err = PTR_ERR(inode);
4769 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4775 btrfs_set_trans_block_group(trans, inode);
4776 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4780 inode->i_mapping->a_ops = &btrfs_aops;
4781 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4782 inode->i_fop = &btrfs_file_operations;
4783 inode->i_op = &btrfs_file_inode_operations;
4784 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4786 btrfs_update_inode_block_group(trans, inode);
4787 btrfs_update_inode_block_group(trans, dir);
4789 nr = trans->blocks_used;
4790 btrfs_end_transaction_throttle(trans, root);
4792 inode_dec_link_count(inode);
4795 btrfs_btree_balance_dirty(root, nr);
4799 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4800 struct dentry *dentry)
4802 struct btrfs_trans_handle *trans;
4803 struct btrfs_root *root = BTRFS_I(dir)->root;
4804 struct inode *inode = old_dentry->d_inode;
4806 unsigned long nr = 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, &dentry->d_name);
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 void 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 if (!S_ISREG(inode->i_mode)) {
6490 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6494 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6495 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6497 trans = btrfs_start_transaction(root, 0);
6498 BUG_ON(IS_ERR(trans));
6499 btrfs_set_trans_block_group(trans, inode);
6500 trans->block_rsv = root->orphan_block_rsv;
6503 * setattr is responsible for setting the ordered_data_close flag,
6504 * but that is only tested during the last file release. That
6505 * could happen well after the next commit, leaving a great big
6506 * window where new writes may get lost if someone chooses to write
6507 * to this file after truncating to zero
6509 * The inode doesn't have any dirty data here, and so if we commit
6510 * this is a noop. If someone immediately starts writing to the inode
6511 * it is very likely we'll catch some of their writes in this
6512 * transaction, and the commit will find this file on the ordered
6513 * data list with good things to send down.
6515 * This is a best effort solution, there is still a window where
6516 * using truncate to replace the contents of the file will
6517 * end up with a zero length file after a crash.
6519 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6520 btrfs_add_ordered_operation(trans, root, inode);
6524 trans = btrfs_start_transaction(root, 0);
6525 BUG_ON(IS_ERR(trans));
6526 btrfs_set_trans_block_group(trans, inode);
6527 trans->block_rsv = root->orphan_block_rsv;
6530 ret = btrfs_block_rsv_check(trans, root,
6531 root->orphan_block_rsv, 0, 5);
6533 BUG_ON(ret != -EAGAIN);
6534 ret = btrfs_commit_transaction(trans, root);
6540 ret = btrfs_truncate_inode_items(trans, root, inode,
6542 BTRFS_EXTENT_DATA_KEY);
6546 ret = btrfs_update_inode(trans, root, inode);
6549 nr = trans->blocks_used;
6550 btrfs_end_transaction(trans, root);
6552 btrfs_btree_balance_dirty(root, nr);
6555 if (ret == 0 && inode->i_nlink > 0) {
6556 ret = btrfs_orphan_del(trans, inode);
6560 ret = btrfs_update_inode(trans, root, inode);
6563 nr = trans->blocks_used;
6564 ret = btrfs_end_transaction_throttle(trans, root);
6566 btrfs_btree_balance_dirty(root, nr);
6570 * create a new subvolume directory/inode (helper for the ioctl).
6572 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6573 struct btrfs_root *new_root,
6574 u64 new_dirid, u64 alloc_hint)
6576 struct inode *inode;
6580 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6581 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6583 return PTR_ERR(inode);
6584 inode->i_op = &btrfs_dir_inode_operations;
6585 inode->i_fop = &btrfs_dir_file_operations;
6588 btrfs_i_size_write(inode, 0);
6590 err = btrfs_update_inode(trans, new_root, inode);
6597 /* helper function for file defrag and space balancing. This
6598 * forces readahead on a given range of bytes in an inode
6600 unsigned long btrfs_force_ra(struct address_space *mapping,
6601 struct file_ra_state *ra, struct file *file,
6602 pgoff_t offset, pgoff_t last_index)
6604 pgoff_t req_size = last_index - offset + 1;
6606 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6607 return offset + req_size;
6610 struct inode *btrfs_alloc_inode(struct super_block *sb)
6612 struct btrfs_inode *ei;
6613 struct inode *inode;
6615 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6620 ei->space_info = NULL;
6624 ei->last_sub_trans = 0;
6625 ei->logged_trans = 0;
6626 ei->delalloc_bytes = 0;
6627 ei->reserved_bytes = 0;
6628 ei->disk_i_size = 0;
6630 ei->index_cnt = (u64)-1;
6631 ei->last_unlink_trans = 0;
6633 spin_lock_init(&ei->accounting_lock);
6634 atomic_set(&ei->outstanding_extents, 0);
6635 ei->reserved_extents = 0;
6637 ei->ordered_data_close = 0;
6638 ei->orphan_meta_reserved = 0;
6639 ei->dummy_inode = 0;
6640 ei->force_compress = BTRFS_COMPRESS_NONE;
6642 inode = &ei->vfs_inode;
6643 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6644 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6645 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6646 mutex_init(&ei->log_mutex);
6647 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6648 INIT_LIST_HEAD(&ei->i_orphan);
6649 INIT_LIST_HEAD(&ei->delalloc_inodes);
6650 INIT_LIST_HEAD(&ei->ordered_operations);
6651 RB_CLEAR_NODE(&ei->rb_node);
6656 static void btrfs_i_callback(struct rcu_head *head)
6658 struct inode *inode = container_of(head, struct inode, i_rcu);
6659 INIT_LIST_HEAD(&inode->i_dentry);
6660 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6663 void btrfs_destroy_inode(struct inode *inode)
6665 struct btrfs_ordered_extent *ordered;
6666 struct btrfs_root *root = BTRFS_I(inode)->root;
6668 WARN_ON(!list_empty(&inode->i_dentry));
6669 WARN_ON(inode->i_data.nrpages);
6670 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6671 WARN_ON(BTRFS_I(inode)->reserved_extents);
6674 * This can happen where we create an inode, but somebody else also
6675 * created the same inode and we need to destroy the one we already
6682 * Make sure we're properly removed from the ordered operation
6686 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6687 spin_lock(&root->fs_info->ordered_extent_lock);
6688 list_del_init(&BTRFS_I(inode)->ordered_operations);
6689 spin_unlock(&root->fs_info->ordered_extent_lock);
6692 if (root == root->fs_info->tree_root) {
6693 struct btrfs_block_group_cache *block_group;
6695 block_group = btrfs_lookup_block_group(root->fs_info,
6696 BTRFS_I(inode)->block_group);
6697 if (block_group && block_group->inode == inode) {
6698 spin_lock(&block_group->lock);
6699 block_group->inode = NULL;
6700 spin_unlock(&block_group->lock);
6701 btrfs_put_block_group(block_group);
6702 } else if (block_group) {
6703 btrfs_put_block_group(block_group);
6707 spin_lock(&root->orphan_lock);
6708 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6709 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6711 list_del_init(&BTRFS_I(inode)->i_orphan);
6713 spin_unlock(&root->orphan_lock);
6716 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6720 printk(KERN_ERR "btrfs found ordered "
6721 "extent %llu %llu on inode cleanup\n",
6722 (unsigned long long)ordered->file_offset,
6723 (unsigned long long)ordered->len);
6724 btrfs_remove_ordered_extent(inode, ordered);
6725 btrfs_put_ordered_extent(ordered);
6726 btrfs_put_ordered_extent(ordered);
6729 inode_tree_del(inode);
6730 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6732 call_rcu(&inode->i_rcu, btrfs_i_callback);
6735 int btrfs_drop_inode(struct inode *inode)
6737 struct btrfs_root *root = BTRFS_I(inode)->root;
6739 if (btrfs_root_refs(&root->root_item) == 0 &&
6740 root != root->fs_info->tree_root)
6743 return generic_drop_inode(inode);
6746 static void init_once(void *foo)
6748 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6750 inode_init_once(&ei->vfs_inode);
6753 void btrfs_destroy_cachep(void)
6755 if (btrfs_inode_cachep)
6756 kmem_cache_destroy(btrfs_inode_cachep);
6757 if (btrfs_trans_handle_cachep)
6758 kmem_cache_destroy(btrfs_trans_handle_cachep);
6759 if (btrfs_transaction_cachep)
6760 kmem_cache_destroy(btrfs_transaction_cachep);
6761 if (btrfs_path_cachep)
6762 kmem_cache_destroy(btrfs_path_cachep);
6765 int btrfs_init_cachep(void)
6767 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6768 sizeof(struct btrfs_inode), 0,
6769 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6770 if (!btrfs_inode_cachep)
6773 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6774 sizeof(struct btrfs_trans_handle), 0,
6775 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6776 if (!btrfs_trans_handle_cachep)
6779 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6780 sizeof(struct btrfs_transaction), 0,
6781 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6782 if (!btrfs_transaction_cachep)
6785 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6786 sizeof(struct btrfs_path), 0,
6787 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6788 if (!btrfs_path_cachep)
6793 btrfs_destroy_cachep();
6797 static int btrfs_getattr(struct vfsmount *mnt,
6798 struct dentry *dentry, struct kstat *stat)
6800 struct inode *inode = dentry->d_inode;
6801 generic_fillattr(inode, stat);
6802 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6803 stat->blksize = PAGE_CACHE_SIZE;
6804 stat->blocks = (inode_get_bytes(inode) +
6805 BTRFS_I(inode)->delalloc_bytes) >> 9;
6809 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6810 struct inode *new_dir, struct dentry *new_dentry)
6812 struct btrfs_trans_handle *trans;
6813 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6814 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6815 struct inode *new_inode = new_dentry->d_inode;
6816 struct inode *old_inode = old_dentry->d_inode;
6817 struct timespec ctime = CURRENT_TIME;
6822 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6825 /* we only allow rename subvolume link between subvolumes */
6826 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6829 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6830 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6833 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6834 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6837 * we're using rename to replace one file with another.
6838 * and the replacement file is large. Start IO on it now so
6839 * we don't add too much work to the end of the transaction
6841 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6842 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6843 filemap_flush(old_inode->i_mapping);
6845 /* close the racy window with snapshot create/destroy ioctl */
6846 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6847 down_read(&root->fs_info->subvol_sem);
6849 * We want to reserve the absolute worst case amount of items. So if
6850 * both inodes are subvols and we need to unlink them then that would
6851 * require 4 item modifications, but if they are both normal inodes it
6852 * would require 5 item modifications, so we'll assume their normal
6853 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6854 * should cover the worst case number of items we'll modify.
6856 trans = btrfs_start_transaction(root, 20);
6858 return PTR_ERR(trans);
6860 btrfs_set_trans_block_group(trans, new_dir);
6863 btrfs_record_root_in_trans(trans, dest);
6865 ret = btrfs_set_inode_index(new_dir, &index);
6869 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6870 /* force full log commit if subvolume involved. */
6871 root->fs_info->last_trans_log_full_commit = trans->transid;
6873 ret = btrfs_insert_inode_ref(trans, dest,
6874 new_dentry->d_name.name,
6875 new_dentry->d_name.len,
6877 new_dir->i_ino, index);
6881 * this is an ugly little race, but the rename is required
6882 * to make sure that if we crash, the inode is either at the
6883 * old name or the new one. pinning the log transaction lets
6884 * us make sure we don't allow a log commit to come in after
6885 * we unlink the name but before we add the new name back in.
6887 btrfs_pin_log_trans(root);
6890 * make sure the inode gets flushed if it is replacing
6893 if (new_inode && new_inode->i_size &&
6894 old_inode && S_ISREG(old_inode->i_mode)) {
6895 btrfs_add_ordered_operation(trans, root, old_inode);
6898 old_dir->i_ctime = old_dir->i_mtime = ctime;
6899 new_dir->i_ctime = new_dir->i_mtime = ctime;
6900 old_inode->i_ctime = ctime;
6902 if (old_dentry->d_parent != new_dentry->d_parent)
6903 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6905 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6906 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6907 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6908 old_dentry->d_name.name,
6909 old_dentry->d_name.len);
6911 btrfs_inc_nlink(old_dentry->d_inode);
6912 ret = btrfs_unlink_inode(trans, root, old_dir,
6913 old_dentry->d_inode,
6914 old_dentry->d_name.name,
6915 old_dentry->d_name.len);
6920 new_inode->i_ctime = CURRENT_TIME;
6921 if (unlikely(new_inode->i_ino ==
6922 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6923 root_objectid = BTRFS_I(new_inode)->location.objectid;
6924 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6926 new_dentry->d_name.name,
6927 new_dentry->d_name.len);
6928 BUG_ON(new_inode->i_nlink == 0);
6930 ret = btrfs_unlink_inode(trans, dest, new_dir,
6931 new_dentry->d_inode,
6932 new_dentry->d_name.name,
6933 new_dentry->d_name.len);
6936 if (new_inode->i_nlink == 0) {
6937 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6942 ret = btrfs_add_link(trans, new_dir, old_inode,
6943 new_dentry->d_name.name,
6944 new_dentry->d_name.len, 0, index);
6947 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6948 struct dentry *parent = dget_parent(new_dentry);
6949 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6951 btrfs_end_log_trans(root);
6954 btrfs_end_transaction_throttle(trans, root);
6956 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6957 up_read(&root->fs_info->subvol_sem);
6963 * some fairly slow code that needs optimization. This walks the list
6964 * of all the inodes with pending delalloc and forces them to disk.
6966 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6968 struct list_head *head = &root->fs_info->delalloc_inodes;
6969 struct btrfs_inode *binode;
6970 struct inode *inode;
6972 if (root->fs_info->sb->s_flags & MS_RDONLY)
6975 spin_lock(&root->fs_info->delalloc_lock);
6976 while (!list_empty(head)) {
6977 binode = list_entry(head->next, struct btrfs_inode,
6979 inode = igrab(&binode->vfs_inode);
6981 list_del_init(&binode->delalloc_inodes);
6982 spin_unlock(&root->fs_info->delalloc_lock);
6984 filemap_flush(inode->i_mapping);
6986 btrfs_add_delayed_iput(inode);
6991 spin_lock(&root->fs_info->delalloc_lock);
6993 spin_unlock(&root->fs_info->delalloc_lock);
6995 /* the filemap_flush will queue IO into the worker threads, but
6996 * we have to make sure the IO is actually started and that
6997 * ordered extents get created before we return
6999 atomic_inc(&root->fs_info->async_submit_draining);
7000 while (atomic_read(&root->fs_info->nr_async_submits) ||
7001 atomic_read(&root->fs_info->async_delalloc_pages)) {
7002 wait_event(root->fs_info->async_submit_wait,
7003 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7004 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7006 atomic_dec(&root->fs_info->async_submit_draining);
7010 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7013 struct btrfs_inode *binode;
7014 struct inode *inode = NULL;
7016 spin_lock(&root->fs_info->delalloc_lock);
7017 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7018 binode = list_entry(root->fs_info->delalloc_inodes.next,
7019 struct btrfs_inode, delalloc_inodes);
7020 inode = igrab(&binode->vfs_inode);
7022 list_move_tail(&binode->delalloc_inodes,
7023 &root->fs_info->delalloc_inodes);
7027 list_del_init(&binode->delalloc_inodes);
7028 cond_resched_lock(&root->fs_info->delalloc_lock);
7030 spin_unlock(&root->fs_info->delalloc_lock);
7034 filemap_write_and_wait(inode->i_mapping);
7036 * We have to do this because compression doesn't
7037 * actually set PG_writeback until it submits the pages
7038 * for IO, which happens in an async thread, so we could
7039 * race and not actually wait for any writeback pages
7040 * because they've not been submitted yet. Technically
7041 * this could still be the case for the ordered stuff
7042 * since the async thread may not have started to do its
7043 * work yet. If this becomes the case then we need to
7044 * figure out a way to make sure that in writepage we
7045 * wait for any async pages to be submitted before
7046 * returning so that fdatawait does what its supposed to
7049 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7051 filemap_flush(inode->i_mapping);
7054 btrfs_add_delayed_iput(inode);
7062 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7063 const char *symname)
7065 struct btrfs_trans_handle *trans;
7066 struct btrfs_root *root = BTRFS_I(dir)->root;
7067 struct btrfs_path *path;
7068 struct btrfs_key key;
7069 struct inode *inode = NULL;
7077 struct btrfs_file_extent_item *ei;
7078 struct extent_buffer *leaf;
7079 unsigned long nr = 0;
7081 name_len = strlen(symname) + 1;
7082 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7083 return -ENAMETOOLONG;
7085 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7089 * 2 items for inode item and ref
7090 * 2 items for dir items
7091 * 1 item for xattr if selinux is on
7093 trans = btrfs_start_transaction(root, 5);
7095 return PTR_ERR(trans);
7097 btrfs_set_trans_block_group(trans, dir);
7099 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7100 dentry->d_name.len, dir->i_ino, objectid,
7101 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7103 err = PTR_ERR(inode);
7107 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7113 btrfs_set_trans_block_group(trans, inode);
7114 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7118 inode->i_mapping->a_ops = &btrfs_aops;
7119 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7120 inode->i_fop = &btrfs_file_operations;
7121 inode->i_op = &btrfs_file_inode_operations;
7122 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7124 btrfs_update_inode_block_group(trans, inode);
7125 btrfs_update_inode_block_group(trans, dir);
7129 path = btrfs_alloc_path();
7131 key.objectid = inode->i_ino;
7133 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7134 datasize = btrfs_file_extent_calc_inline_size(name_len);
7135 err = btrfs_insert_empty_item(trans, root, path, &key,
7141 leaf = path->nodes[0];
7142 ei = btrfs_item_ptr(leaf, path->slots[0],
7143 struct btrfs_file_extent_item);
7144 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7145 btrfs_set_file_extent_type(leaf, ei,
7146 BTRFS_FILE_EXTENT_INLINE);
7147 btrfs_set_file_extent_encryption(leaf, ei, 0);
7148 btrfs_set_file_extent_compression(leaf, ei, 0);
7149 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7150 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7152 ptr = btrfs_file_extent_inline_start(ei);
7153 write_extent_buffer(leaf, symname, ptr, name_len);
7154 btrfs_mark_buffer_dirty(leaf);
7155 btrfs_free_path(path);
7157 inode->i_op = &btrfs_symlink_inode_operations;
7158 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7159 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7160 inode_set_bytes(inode, name_len);
7161 btrfs_i_size_write(inode, name_len - 1);
7162 err = btrfs_update_inode(trans, root, inode);
7167 nr = trans->blocks_used;
7168 btrfs_end_transaction_throttle(trans, root);
7170 inode_dec_link_count(inode);
7173 btrfs_btree_balance_dirty(root, nr);
7177 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7178 u64 start, u64 num_bytes, u64 min_size,
7179 loff_t actual_len, u64 *alloc_hint,
7180 struct btrfs_trans_handle *trans)
7182 struct btrfs_root *root = BTRFS_I(inode)->root;
7183 struct btrfs_key ins;
7184 u64 cur_offset = start;
7187 bool own_trans = true;
7191 while (num_bytes > 0) {
7193 trans = btrfs_start_transaction(root, 3);
7194 if (IS_ERR(trans)) {
7195 ret = PTR_ERR(trans);
7200 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7201 0, *alloc_hint, (u64)-1, &ins, 1);
7204 btrfs_end_transaction(trans, root);
7208 ret = insert_reserved_file_extent(trans, inode,
7209 cur_offset, ins.objectid,
7210 ins.offset, ins.offset,
7211 ins.offset, 0, 0, 0,
7212 BTRFS_FILE_EXTENT_PREALLOC);
7214 btrfs_drop_extent_cache(inode, cur_offset,
7215 cur_offset + ins.offset -1, 0);
7217 num_bytes -= ins.offset;
7218 cur_offset += ins.offset;
7219 *alloc_hint = ins.objectid + ins.offset;
7221 inode->i_ctime = CURRENT_TIME;
7222 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7223 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7224 (actual_len > inode->i_size) &&
7225 (cur_offset > inode->i_size)) {
7226 if (cur_offset > actual_len)
7227 i_size = actual_len;
7229 i_size = cur_offset;
7230 i_size_write(inode, i_size);
7231 btrfs_ordered_update_i_size(inode, i_size, NULL);
7234 ret = btrfs_update_inode(trans, root, inode);
7238 btrfs_end_transaction(trans, root);
7243 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7244 u64 start, u64 num_bytes, u64 min_size,
7245 loff_t actual_len, u64 *alloc_hint)
7247 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7248 min_size, actual_len, alloc_hint,
7252 int btrfs_prealloc_file_range_trans(struct inode *inode,
7253 struct btrfs_trans_handle *trans, int mode,
7254 u64 start, u64 num_bytes, u64 min_size,
7255 loff_t actual_len, u64 *alloc_hint)
7257 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7258 min_size, actual_len, alloc_hint, trans);
7261 static int btrfs_set_page_dirty(struct page *page)
7263 return __set_page_dirty_nobuffers(page);
7266 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7268 struct btrfs_root *root = BTRFS_I(inode)->root;
7270 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7272 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7274 return generic_permission(inode, mask, flags, btrfs_check_acl);
7277 static const struct inode_operations btrfs_dir_inode_operations = {
7278 .getattr = btrfs_getattr,
7279 .lookup = btrfs_lookup,
7280 .create = btrfs_create,
7281 .unlink = btrfs_unlink,
7283 .mkdir = btrfs_mkdir,
7284 .rmdir = btrfs_rmdir,
7285 .rename = btrfs_rename,
7286 .symlink = btrfs_symlink,
7287 .setattr = btrfs_setattr,
7288 .mknod = btrfs_mknod,
7289 .setxattr = btrfs_setxattr,
7290 .getxattr = btrfs_getxattr,
7291 .listxattr = btrfs_listxattr,
7292 .removexattr = btrfs_removexattr,
7293 .permission = btrfs_permission,
7295 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7296 .lookup = btrfs_lookup,
7297 .permission = btrfs_permission,
7300 static const struct file_operations btrfs_dir_file_operations = {
7301 .llseek = generic_file_llseek,
7302 .read = generic_read_dir,
7303 .readdir = btrfs_real_readdir,
7304 .unlocked_ioctl = btrfs_ioctl,
7305 #ifdef CONFIG_COMPAT
7306 .compat_ioctl = btrfs_ioctl,
7308 .release = btrfs_release_file,
7309 .fsync = btrfs_sync_file,
7312 static struct extent_io_ops btrfs_extent_io_ops = {
7313 .fill_delalloc = run_delalloc_range,
7314 .submit_bio_hook = btrfs_submit_bio_hook,
7315 .merge_bio_hook = btrfs_merge_bio_hook,
7316 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7317 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7318 .writepage_start_hook = btrfs_writepage_start_hook,
7319 .readpage_io_failed_hook = btrfs_io_failed_hook,
7320 .set_bit_hook = btrfs_set_bit_hook,
7321 .clear_bit_hook = btrfs_clear_bit_hook,
7322 .merge_extent_hook = btrfs_merge_extent_hook,
7323 .split_extent_hook = btrfs_split_extent_hook,
7327 * btrfs doesn't support the bmap operation because swapfiles
7328 * use bmap to make a mapping of extents in the file. They assume
7329 * these extents won't change over the life of the file and they
7330 * use the bmap result to do IO directly to the drive.
7332 * the btrfs bmap call would return logical addresses that aren't
7333 * suitable for IO and they also will change frequently as COW
7334 * operations happen. So, swapfile + btrfs == corruption.
7336 * For now we're avoiding this by dropping bmap.
7338 static const struct address_space_operations btrfs_aops = {
7339 .readpage = btrfs_readpage,
7340 .writepage = btrfs_writepage,
7341 .writepages = btrfs_writepages,
7342 .readpages = btrfs_readpages,
7343 .direct_IO = btrfs_direct_IO,
7344 .invalidatepage = btrfs_invalidatepage,
7345 .releasepage = btrfs_releasepage,
7346 .set_page_dirty = btrfs_set_page_dirty,
7347 .error_remove_page = generic_error_remove_page,
7350 static const struct address_space_operations btrfs_symlink_aops = {
7351 .readpage = btrfs_readpage,
7352 .writepage = btrfs_writepage,
7353 .invalidatepage = btrfs_invalidatepage,
7354 .releasepage = btrfs_releasepage,
7357 static const struct inode_operations btrfs_file_inode_operations = {
7358 .truncate = btrfs_truncate,
7359 .getattr = btrfs_getattr,
7360 .setattr = btrfs_setattr,
7361 .setxattr = btrfs_setxattr,
7362 .getxattr = btrfs_getxattr,
7363 .listxattr = btrfs_listxattr,
7364 .removexattr = btrfs_removexattr,
7365 .permission = btrfs_permission,
7366 .fiemap = btrfs_fiemap,
7368 static const struct inode_operations btrfs_special_inode_operations = {
7369 .getattr = btrfs_getattr,
7370 .setattr = btrfs_setattr,
7371 .permission = btrfs_permission,
7372 .setxattr = btrfs_setxattr,
7373 .getxattr = btrfs_getxattr,
7374 .listxattr = btrfs_listxattr,
7375 .removexattr = btrfs_removexattr,
7377 static const struct inode_operations btrfs_symlink_inode_operations = {
7378 .readlink = generic_readlink,
7379 .follow_link = page_follow_link_light,
7380 .put_link = page_put_link,
7381 .getattr = btrfs_getattr,
7382 .permission = btrfs_permission,
7383 .setxattr = btrfs_setxattr,
7384 .getxattr = btrfs_getxattr,
7385 .listxattr = btrfs_listxattr,
7386 .removexattr = btrfs_removexattr,
7389 const struct dentry_operations btrfs_dentry_operations = {
7390 .d_delete = btrfs_dentry_delete,