2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
44 #include "transaction.h"
45 #include "btrfs_inode.h"
47 #include "print-tree.h"
49 #include "ordered-data.h"
52 #include "compression.h"
54 #include "free-space-cache.h"
55 #include "inode-map.h"
57 struct btrfs_iget_args {
59 struct btrfs_root *root;
62 static const struct inode_operations btrfs_dir_inode_operations;
63 static const struct inode_operations btrfs_symlink_inode_operations;
64 static const struct inode_operations btrfs_dir_ro_inode_operations;
65 static const struct inode_operations btrfs_special_inode_operations;
66 static const struct inode_operations btrfs_file_inode_operations;
67 static const struct address_space_operations btrfs_aops;
68 static const struct address_space_operations btrfs_symlink_aops;
69 static const struct file_operations btrfs_dir_file_operations;
70 static struct extent_io_ops btrfs_extent_io_ops;
72 static struct kmem_cache *btrfs_inode_cachep;
73 struct kmem_cache *btrfs_trans_handle_cachep;
74 struct kmem_cache *btrfs_transaction_cachep;
75 struct kmem_cache *btrfs_path_cachep;
76 struct kmem_cache *btrfs_free_space_cachep;
79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
81 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
82 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
83 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
84 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
85 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
86 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
90 static int btrfs_truncate(struct inode *inode);
91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
92 static noinline int cow_file_range(struct inode *inode,
93 struct page *locked_page,
94 u64 start, u64 end, int *page_started,
95 unsigned long *nr_written, int unlock);
97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
98 struct inode *inode, struct inode *dir,
99 const struct qstr *qstr)
103 err = btrfs_init_acl(trans, inode, dir);
105 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
110 * this does all the hard work for inserting an inline extent into
111 * the btree. The caller should have done a btrfs_drop_extents so that
112 * no overlapping inline items exist in the btree
114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
115 struct btrfs_root *root, struct inode *inode,
116 u64 start, size_t size, size_t compressed_size,
118 struct page **compressed_pages)
120 struct btrfs_key key;
121 struct btrfs_path *path;
122 struct extent_buffer *leaf;
123 struct page *page = NULL;
126 struct btrfs_file_extent_item *ei;
129 size_t cur_size = size;
131 unsigned long offset;
133 if (compressed_size && compressed_pages)
134 cur_size = compressed_size;
136 path = btrfs_alloc_path();
140 path->leave_spinning = 1;
142 key.objectid = btrfs_ino(inode);
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
165 if (compress_type != BTRFS_COMPRESS_NONE) {
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
179 compressed_size -= cur_size;
181 btrfs_set_file_extent_compression(leaf, ei,
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
210 btrfs_free_path(path);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size, int compress_type,
224 struct page **compressed_pages)
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
232 u64 data_len = inline_len;
236 data_len = compressed_size;
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (actual_end & (root->sectorsize - 1)) == 0) ||
244 data_len > root->fs_info->max_inline) {
248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
256 compress_type, compressed_pages);
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
263 struct async_extent {
268 unsigned long nr_pages;
270 struct list_head list;
275 struct btrfs_root *root;
276 struct page *locked_page;
279 struct list_head extents;
280 struct btrfs_work work;
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
287 unsigned long nr_pages,
290 struct async_extent *async_extent;
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 BUG_ON(!async_extent);
294 async_extent->start = start;
295 async_extent->ram_size = ram_size;
296 async_extent->compressed_size = compressed_size;
297 async_extent->pages = pages;
298 async_extent->nr_pages = nr_pages;
299 async_extent->compress_type = compress_type;
300 list_add_tail(&async_extent->list, &cow->extents);
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
320 static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
323 struct async_cow *async_cow,
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
329 u64 blocksize = root->sectorsize;
331 u64 isize = i_size_read(inode);
333 struct page **pages = NULL;
334 unsigned long nr_pages;
335 unsigned long nr_pages_ret = 0;
336 unsigned long total_compressed = 0;
337 unsigned long total_in = 0;
338 unsigned long max_compressed = 128 * 1024;
339 unsigned long max_uncompressed = 128 * 1024;
342 int compress_type = root->fs_info->compress_type;
344 /* if this is a small write inside eof, kick off a defragbot */
345 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
346 btrfs_add_inode_defrag(NULL, inode);
348 actual_end = min_t(u64, isize, end + 1);
351 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
352 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
355 * we don't want to send crud past the end of i_size through
356 * compression, that's just a waste of CPU time. So, if the
357 * end of the file is before the start of our current
358 * requested range of bytes, we bail out to the uncompressed
359 * cleanup code that can deal with all of this.
361 * It isn't really the fastest way to fix things, but this is a
362 * very uncommon corner.
364 if (actual_end <= start)
365 goto cleanup_and_bail_uncompressed;
367 total_compressed = actual_end - start;
369 /* we want to make sure that amount of ram required to uncompress
370 * an extent is reasonable, so we limit the total size in ram
371 * of a compressed extent to 128k. This is a crucial number
372 * because it also controls how easily we can spread reads across
373 * cpus for decompression.
375 * We also want to make sure the amount of IO required to do
376 * a random read is reasonably small, so we limit the size of
377 * a compressed extent to 128k.
379 total_compressed = min(total_compressed, max_uncompressed);
380 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
381 num_bytes = max(blocksize, num_bytes);
386 * we do compression for mount -o compress and when the
387 * inode has not been flagged as nocompress. This flag can
388 * change at any time if we discover bad compression ratios.
390 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
391 (btrfs_test_opt(root, COMPRESS) ||
392 (BTRFS_I(inode)->force_compress) ||
393 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
395 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
397 /* just bail out to the uncompressed code */
401 if (BTRFS_I(inode)->force_compress)
402 compress_type = BTRFS_I(inode)->force_compress;
404 ret = btrfs_compress_pages(compress_type,
405 inode->i_mapping, start,
406 total_compressed, pages,
407 nr_pages, &nr_pages_ret,
413 unsigned long offset = total_compressed &
414 (PAGE_CACHE_SIZE - 1);
415 struct page *page = pages[nr_pages_ret - 1];
418 /* zero the tail end of the last page, we might be
419 * sending it down to disk
422 kaddr = kmap_atomic(page, KM_USER0);
423 memset(kaddr + offset, 0,
424 PAGE_CACHE_SIZE - offset);
425 kunmap_atomic(kaddr, KM_USER0);
432 trans = btrfs_join_transaction(root);
433 BUG_ON(IS_ERR(trans));
434 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
436 /* lets try to make an inline extent */
437 if (ret || total_in < (actual_end - start)) {
438 /* we didn't compress the entire range, try
439 * to make an uncompressed inline extent.
441 ret = cow_file_range_inline(trans, root, inode,
442 start, end, 0, 0, NULL);
444 /* try making a compressed inline extent */
445 ret = cow_file_range_inline(trans, root, inode,
448 compress_type, pages);
452 * inline extent creation worked, we don't need
453 * to create any more async work items. Unlock
454 * and free up our temp pages.
456 extent_clear_unlock_delalloc(inode,
457 &BTRFS_I(inode)->io_tree,
459 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
460 EXTENT_CLEAR_DELALLOC |
461 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
463 btrfs_end_transaction(trans, root);
466 btrfs_end_transaction(trans, root);
471 * we aren't doing an inline extent round the compressed size
472 * up to a block size boundary so the allocator does sane
475 total_compressed = (total_compressed + blocksize - 1) &
479 * one last check to make sure the compression is really a
480 * win, compare the page count read with the blocks on disk
482 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
483 ~(PAGE_CACHE_SIZE - 1);
484 if (total_compressed >= total_in) {
487 num_bytes = total_in;
490 if (!will_compress && pages) {
492 * the compression code ran but failed to make things smaller,
493 * free any pages it allocated and our page pointer array
495 for (i = 0; i < nr_pages_ret; i++) {
496 WARN_ON(pages[i]->mapping);
497 page_cache_release(pages[i]);
501 total_compressed = 0;
504 /* flag the file so we don't compress in the future */
505 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
506 !(BTRFS_I(inode)->force_compress)) {
507 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
513 /* the async work queues will take care of doing actual
514 * allocation on disk for these compressed pages,
515 * and will submit them to the elevator.
517 add_async_extent(async_cow, start, num_bytes,
518 total_compressed, pages, nr_pages_ret,
521 if (start + num_bytes < end) {
528 cleanup_and_bail_uncompressed:
530 * No compression, but we still need to write the pages in
531 * the file we've been given so far. redirty the locked
532 * page if it corresponds to our extent and set things up
533 * for the async work queue to run cow_file_range to do
534 * the normal delalloc dance
536 if (page_offset(locked_page) >= start &&
537 page_offset(locked_page) <= end) {
538 __set_page_dirty_nobuffers(locked_page);
539 /* unlocked later on in the async handlers */
541 add_async_extent(async_cow, start, end - start + 1,
542 0, NULL, 0, BTRFS_COMPRESS_NONE);
550 for (i = 0; i < nr_pages_ret; i++) {
551 WARN_ON(pages[i]->mapping);
552 page_cache_release(pages[i]);
560 * phase two of compressed writeback. This is the ordered portion
561 * of the code, which only gets called in the order the work was
562 * queued. We walk all the async extents created by compress_file_range
563 * and send them down to the disk.
565 static noinline int submit_compressed_extents(struct inode *inode,
566 struct async_cow *async_cow)
568 struct async_extent *async_extent;
570 struct btrfs_trans_handle *trans;
571 struct btrfs_key ins;
572 struct extent_map *em;
573 struct btrfs_root *root = BTRFS_I(inode)->root;
574 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
575 struct extent_io_tree *io_tree;
578 if (list_empty(&async_cow->extents))
582 while (!list_empty(&async_cow->extents)) {
583 async_extent = list_entry(async_cow->extents.next,
584 struct async_extent, list);
585 list_del(&async_extent->list);
587 io_tree = &BTRFS_I(inode)->io_tree;
590 /* did the compression code fall back to uncompressed IO? */
591 if (!async_extent->pages) {
592 int page_started = 0;
593 unsigned long nr_written = 0;
595 lock_extent(io_tree, async_extent->start,
596 async_extent->start +
597 async_extent->ram_size - 1, GFP_NOFS);
599 /* allocate blocks */
600 ret = cow_file_range(inode, async_cow->locked_page,
602 async_extent->start +
603 async_extent->ram_size - 1,
604 &page_started, &nr_written, 0);
607 * if page_started, cow_file_range inserted an
608 * inline extent and took care of all the unlocking
609 * and IO for us. Otherwise, we need to submit
610 * all those pages down to the drive.
612 if (!page_started && !ret)
613 extent_write_locked_range(io_tree,
614 inode, async_extent->start,
615 async_extent->start +
616 async_extent->ram_size - 1,
624 lock_extent(io_tree, async_extent->start,
625 async_extent->start + async_extent->ram_size - 1,
628 trans = btrfs_join_transaction(root);
629 BUG_ON(IS_ERR(trans));
630 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
631 ret = btrfs_reserve_extent(trans, root,
632 async_extent->compressed_size,
633 async_extent->compressed_size,
636 btrfs_end_transaction(trans, root);
640 for (i = 0; i < async_extent->nr_pages; i++) {
641 WARN_ON(async_extent->pages[i]->mapping);
642 page_cache_release(async_extent->pages[i]);
644 kfree(async_extent->pages);
645 async_extent->nr_pages = 0;
646 async_extent->pages = NULL;
647 unlock_extent(io_tree, async_extent->start,
648 async_extent->start +
649 async_extent->ram_size - 1, GFP_NOFS);
654 * here we're doing allocation and writeback of the
657 btrfs_drop_extent_cache(inode, async_extent->start,
658 async_extent->start +
659 async_extent->ram_size - 1, 0);
661 em = alloc_extent_map();
663 em->start = async_extent->start;
664 em->len = async_extent->ram_size;
665 em->orig_start = em->start;
667 em->block_start = ins.objectid;
668 em->block_len = ins.offset;
669 em->bdev = root->fs_info->fs_devices->latest_bdev;
670 em->compress_type = async_extent->compress_type;
671 set_bit(EXTENT_FLAG_PINNED, &em->flags);
672 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
675 write_lock(&em_tree->lock);
676 ret = add_extent_mapping(em_tree, em);
677 write_unlock(&em_tree->lock);
678 if (ret != -EEXIST) {
682 btrfs_drop_extent_cache(inode, async_extent->start,
683 async_extent->start +
684 async_extent->ram_size - 1, 0);
687 ret = btrfs_add_ordered_extent_compress(inode,
690 async_extent->ram_size,
692 BTRFS_ORDERED_COMPRESSED,
693 async_extent->compress_type);
697 * clear dirty, set writeback and unlock the pages.
699 extent_clear_unlock_delalloc(inode,
700 &BTRFS_I(inode)->io_tree,
702 async_extent->start +
703 async_extent->ram_size - 1,
704 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
705 EXTENT_CLEAR_UNLOCK |
706 EXTENT_CLEAR_DELALLOC |
707 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
709 ret = btrfs_submit_compressed_write(inode,
711 async_extent->ram_size,
713 ins.offset, async_extent->pages,
714 async_extent->nr_pages);
717 alloc_hint = ins.objectid + ins.offset;
725 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
728 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
729 struct extent_map *em;
732 read_lock(&em_tree->lock);
733 em = search_extent_mapping(em_tree, start, num_bytes);
736 * if block start isn't an actual block number then find the
737 * first block in this inode and use that as a hint. If that
738 * block is also bogus then just don't worry about it.
740 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
742 em = search_extent_mapping(em_tree, 0, 0);
743 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
744 alloc_hint = em->block_start;
748 alloc_hint = em->block_start;
752 read_unlock(&em_tree->lock);
758 * when extent_io.c finds a delayed allocation range in the file,
759 * the call backs end up in this code. The basic idea is to
760 * allocate extents on disk for the range, and create ordered data structs
761 * in ram to track those extents.
763 * locked_page is the page that writepage had locked already. We use
764 * it to make sure we don't do extra locks or unlocks.
766 * *page_started is set to one if we unlock locked_page and do everything
767 * required to start IO on it. It may be clean and already done with
770 static noinline int cow_file_range(struct inode *inode,
771 struct page *locked_page,
772 u64 start, u64 end, int *page_started,
773 unsigned long *nr_written,
776 struct btrfs_root *root = BTRFS_I(inode)->root;
777 struct btrfs_trans_handle *trans;
780 unsigned long ram_size;
783 u64 blocksize = root->sectorsize;
784 struct btrfs_key ins;
785 struct extent_map *em;
786 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
789 BUG_ON(btrfs_is_free_space_inode(root, inode));
790 trans = btrfs_join_transaction(root);
791 BUG_ON(IS_ERR(trans));
792 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
794 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
795 num_bytes = max(blocksize, num_bytes);
796 disk_num_bytes = num_bytes;
799 /* if this is a small write inside eof, kick off defrag */
800 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
801 btrfs_add_inode_defrag(trans, inode);
804 /* lets try to make an inline extent */
805 ret = cow_file_range_inline(trans, root, inode,
806 start, end, 0, 0, NULL);
808 extent_clear_unlock_delalloc(inode,
809 &BTRFS_I(inode)->io_tree,
811 EXTENT_CLEAR_UNLOCK_PAGE |
812 EXTENT_CLEAR_UNLOCK |
813 EXTENT_CLEAR_DELALLOC |
815 EXTENT_SET_WRITEBACK |
816 EXTENT_END_WRITEBACK);
818 *nr_written = *nr_written +
819 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
826 BUG_ON(disk_num_bytes >
827 btrfs_super_total_bytes(root->fs_info->super_copy));
829 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
830 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
832 while (disk_num_bytes > 0) {
835 cur_alloc_size = disk_num_bytes;
836 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
837 root->sectorsize, 0, alloc_hint,
841 em = alloc_extent_map();
844 em->orig_start = em->start;
845 ram_size = ins.offset;
846 em->len = ins.offset;
848 em->block_start = ins.objectid;
849 em->block_len = ins.offset;
850 em->bdev = root->fs_info->fs_devices->latest_bdev;
851 set_bit(EXTENT_FLAG_PINNED, &em->flags);
854 write_lock(&em_tree->lock);
855 ret = add_extent_mapping(em_tree, em);
856 write_unlock(&em_tree->lock);
857 if (ret != -EEXIST) {
861 btrfs_drop_extent_cache(inode, start,
862 start + ram_size - 1, 0);
865 cur_alloc_size = ins.offset;
866 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
867 ram_size, cur_alloc_size, 0);
870 if (root->root_key.objectid ==
871 BTRFS_DATA_RELOC_TREE_OBJECTID) {
872 ret = btrfs_reloc_clone_csums(inode, start,
877 if (disk_num_bytes < cur_alloc_size)
880 /* we're not doing compressed IO, don't unlock the first
881 * page (which the caller expects to stay locked), don't
882 * clear any dirty bits and don't set any writeback bits
884 * Do set the Private2 bit so we know this page was properly
885 * setup for writepage
887 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
888 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
891 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
892 start, start + ram_size - 1,
894 disk_num_bytes -= cur_alloc_size;
895 num_bytes -= cur_alloc_size;
896 alloc_hint = ins.objectid + ins.offset;
897 start += cur_alloc_size;
901 btrfs_end_transaction(trans, root);
907 * work queue call back to started compression on a file and pages
909 static noinline void async_cow_start(struct btrfs_work *work)
911 struct async_cow *async_cow;
913 async_cow = container_of(work, struct async_cow, work);
915 compress_file_range(async_cow->inode, async_cow->locked_page,
916 async_cow->start, async_cow->end, async_cow,
919 async_cow->inode = NULL;
923 * work queue call back to submit previously compressed pages
925 static noinline void async_cow_submit(struct btrfs_work *work)
927 struct async_cow *async_cow;
928 struct btrfs_root *root;
929 unsigned long nr_pages;
931 async_cow = container_of(work, struct async_cow, work);
933 root = async_cow->root;
934 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
937 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
939 if (atomic_read(&root->fs_info->async_delalloc_pages) <
941 waitqueue_active(&root->fs_info->async_submit_wait))
942 wake_up(&root->fs_info->async_submit_wait);
944 if (async_cow->inode)
945 submit_compressed_extents(async_cow->inode, async_cow);
948 static noinline void async_cow_free(struct btrfs_work *work)
950 struct async_cow *async_cow;
951 async_cow = container_of(work, struct async_cow, work);
955 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
956 u64 start, u64 end, int *page_started,
957 unsigned long *nr_written)
959 struct async_cow *async_cow;
960 struct btrfs_root *root = BTRFS_I(inode)->root;
961 unsigned long nr_pages;
963 int limit = 10 * 1024 * 1042;
965 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
966 1, 0, NULL, GFP_NOFS);
967 while (start < end) {
968 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
970 async_cow->inode = inode;
971 async_cow->root = root;
972 async_cow->locked_page = locked_page;
973 async_cow->start = start;
975 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
978 cur_end = min(end, start + 512 * 1024 - 1);
980 async_cow->end = cur_end;
981 INIT_LIST_HEAD(&async_cow->extents);
983 async_cow->work.func = async_cow_start;
984 async_cow->work.ordered_func = async_cow_submit;
985 async_cow->work.ordered_free = async_cow_free;
986 async_cow->work.flags = 0;
988 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
990 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
992 btrfs_queue_worker(&root->fs_info->delalloc_workers,
995 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
996 wait_event(root->fs_info->async_submit_wait,
997 (atomic_read(&root->fs_info->async_delalloc_pages) <
1001 while (atomic_read(&root->fs_info->async_submit_draining) &&
1002 atomic_read(&root->fs_info->async_delalloc_pages)) {
1003 wait_event(root->fs_info->async_submit_wait,
1004 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1008 *nr_written += nr_pages;
1009 start = cur_end + 1;
1015 static noinline int csum_exist_in_range(struct btrfs_root *root,
1016 u64 bytenr, u64 num_bytes)
1019 struct btrfs_ordered_sum *sums;
1022 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1023 bytenr + num_bytes - 1, &list, 0);
1024 if (ret == 0 && list_empty(&list))
1027 while (!list_empty(&list)) {
1028 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1029 list_del(&sums->list);
1036 * when nowcow writeback call back. This checks for snapshots or COW copies
1037 * of the extents that exist in the file, and COWs the file as required.
1039 * If no cow copies or snapshots exist, we write directly to the existing
1042 static noinline int run_delalloc_nocow(struct inode *inode,
1043 struct page *locked_page,
1044 u64 start, u64 end, int *page_started, int force,
1045 unsigned long *nr_written)
1047 struct btrfs_root *root = BTRFS_I(inode)->root;
1048 struct btrfs_trans_handle *trans;
1049 struct extent_buffer *leaf;
1050 struct btrfs_path *path;
1051 struct btrfs_file_extent_item *fi;
1052 struct btrfs_key found_key;
1065 u64 ino = btrfs_ino(inode);
1067 path = btrfs_alloc_path();
1071 nolock = btrfs_is_free_space_inode(root, inode);
1074 trans = btrfs_join_transaction_nolock(root);
1076 trans = btrfs_join_transaction(root);
1078 BUG_ON(IS_ERR(trans));
1079 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1081 cow_start = (u64)-1;
1084 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1087 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1088 leaf = path->nodes[0];
1089 btrfs_item_key_to_cpu(leaf, &found_key,
1090 path->slots[0] - 1);
1091 if (found_key.objectid == ino &&
1092 found_key.type == BTRFS_EXTENT_DATA_KEY)
1097 leaf = path->nodes[0];
1098 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1099 ret = btrfs_next_leaf(root, path);
1104 leaf = path->nodes[0];
1110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1112 if (found_key.objectid > ino ||
1113 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1114 found_key.offset > end)
1117 if (found_key.offset > cur_offset) {
1118 extent_end = found_key.offset;
1123 fi = btrfs_item_ptr(leaf, path->slots[0],
1124 struct btrfs_file_extent_item);
1125 extent_type = btrfs_file_extent_type(leaf, fi);
1127 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1128 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1129 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1130 extent_offset = btrfs_file_extent_offset(leaf, fi);
1131 extent_end = found_key.offset +
1132 btrfs_file_extent_num_bytes(leaf, fi);
1133 if (extent_end <= start) {
1137 if (disk_bytenr == 0)
1139 if (btrfs_file_extent_compression(leaf, fi) ||
1140 btrfs_file_extent_encryption(leaf, fi) ||
1141 btrfs_file_extent_other_encoding(leaf, fi))
1143 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1145 if (btrfs_extent_readonly(root, disk_bytenr))
1147 if (btrfs_cross_ref_exist(trans, root, ino,
1149 extent_offset, disk_bytenr))
1151 disk_bytenr += extent_offset;
1152 disk_bytenr += cur_offset - found_key.offset;
1153 num_bytes = min(end + 1, extent_end) - cur_offset;
1155 * force cow if csum exists in the range.
1156 * this ensure that csum for a given extent are
1157 * either valid or do not exist.
1159 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1162 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1163 extent_end = found_key.offset +
1164 btrfs_file_extent_inline_len(leaf, fi);
1165 extent_end = ALIGN(extent_end, root->sectorsize);
1170 if (extent_end <= start) {
1175 if (cow_start == (u64)-1)
1176 cow_start = cur_offset;
1177 cur_offset = extent_end;
1178 if (cur_offset > end)
1184 btrfs_release_path(path);
1185 if (cow_start != (u64)-1) {
1186 ret = cow_file_range(inode, locked_page, cow_start,
1187 found_key.offset - 1, page_started,
1190 cow_start = (u64)-1;
1193 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1194 struct extent_map *em;
1195 struct extent_map_tree *em_tree;
1196 em_tree = &BTRFS_I(inode)->extent_tree;
1197 em = alloc_extent_map();
1199 em->start = cur_offset;
1200 em->orig_start = em->start;
1201 em->len = num_bytes;
1202 em->block_len = num_bytes;
1203 em->block_start = disk_bytenr;
1204 em->bdev = root->fs_info->fs_devices->latest_bdev;
1205 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1207 write_lock(&em_tree->lock);
1208 ret = add_extent_mapping(em_tree, em);
1209 write_unlock(&em_tree->lock);
1210 if (ret != -EEXIST) {
1211 free_extent_map(em);
1214 btrfs_drop_extent_cache(inode, em->start,
1215 em->start + em->len - 1, 0);
1217 type = BTRFS_ORDERED_PREALLOC;
1219 type = BTRFS_ORDERED_NOCOW;
1222 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1223 num_bytes, num_bytes, type);
1226 if (root->root_key.objectid ==
1227 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1228 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1233 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1234 cur_offset, cur_offset + num_bytes - 1,
1235 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1236 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1237 EXTENT_SET_PRIVATE2);
1238 cur_offset = extent_end;
1239 if (cur_offset > end)
1242 btrfs_release_path(path);
1244 if (cur_offset <= end && cow_start == (u64)-1)
1245 cow_start = cur_offset;
1246 if (cow_start != (u64)-1) {
1247 ret = cow_file_range(inode, locked_page, cow_start, end,
1248 page_started, nr_written, 1);
1253 ret = btrfs_end_transaction_nolock(trans, root);
1256 ret = btrfs_end_transaction(trans, root);
1259 btrfs_free_path(path);
1264 * extent_io.c call back to do delayed allocation processing
1266 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1267 u64 start, u64 end, int *page_started,
1268 unsigned long *nr_written)
1271 struct btrfs_root *root = BTRFS_I(inode)->root;
1273 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1274 ret = run_delalloc_nocow(inode, locked_page, start, end,
1275 page_started, 1, nr_written);
1276 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1277 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278 page_started, 0, nr_written);
1279 else if (!btrfs_test_opt(root, COMPRESS) &&
1280 !(BTRFS_I(inode)->force_compress) &&
1281 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1282 ret = cow_file_range(inode, locked_page, start, end,
1283 page_started, nr_written, 1);
1285 ret = cow_file_range_async(inode, locked_page, start, end,
1286 page_started, nr_written);
1290 static void btrfs_split_extent_hook(struct inode *inode,
1291 struct extent_state *orig, u64 split)
1293 /* not delalloc, ignore it */
1294 if (!(orig->state & EXTENT_DELALLOC))
1297 spin_lock(&BTRFS_I(inode)->lock);
1298 BTRFS_I(inode)->outstanding_extents++;
1299 spin_unlock(&BTRFS_I(inode)->lock);
1303 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1304 * extents so we can keep track of new extents that are just merged onto old
1305 * extents, such as when we are doing sequential writes, so we can properly
1306 * account for the metadata space we'll need.
1308 static void btrfs_merge_extent_hook(struct inode *inode,
1309 struct extent_state *new,
1310 struct extent_state *other)
1312 /* not delalloc, ignore it */
1313 if (!(other->state & EXTENT_DELALLOC))
1316 spin_lock(&BTRFS_I(inode)->lock);
1317 BTRFS_I(inode)->outstanding_extents--;
1318 spin_unlock(&BTRFS_I(inode)->lock);
1322 * extent_io.c set_bit_hook, used to track delayed allocation
1323 * bytes in this file, and to maintain the list of inodes that
1324 * have pending delalloc work to be done.
1326 static void btrfs_set_bit_hook(struct inode *inode,
1327 struct extent_state *state, int *bits)
1331 * set_bit and clear bit hooks normally require _irqsave/restore
1332 * but in this case, we are only testing for the DELALLOC
1333 * bit, which is only set or cleared with irqs on
1335 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1336 struct btrfs_root *root = BTRFS_I(inode)->root;
1337 u64 len = state->end + 1 - state->start;
1338 bool do_list = !btrfs_is_free_space_inode(root, inode);
1340 if (*bits & EXTENT_FIRST_DELALLOC) {
1341 *bits &= ~EXTENT_FIRST_DELALLOC;
1343 spin_lock(&BTRFS_I(inode)->lock);
1344 BTRFS_I(inode)->outstanding_extents++;
1345 spin_unlock(&BTRFS_I(inode)->lock);
1348 spin_lock(&root->fs_info->delalloc_lock);
1349 BTRFS_I(inode)->delalloc_bytes += len;
1350 root->fs_info->delalloc_bytes += len;
1351 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1352 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1353 &root->fs_info->delalloc_inodes);
1355 spin_unlock(&root->fs_info->delalloc_lock);
1360 * extent_io.c clear_bit_hook, see set_bit_hook for why
1362 static void btrfs_clear_bit_hook(struct inode *inode,
1363 struct extent_state *state, int *bits)
1366 * set_bit and clear bit hooks normally require _irqsave/restore
1367 * but in this case, we are only testing for the DELALLOC
1368 * bit, which is only set or cleared with irqs on
1370 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371 struct btrfs_root *root = BTRFS_I(inode)->root;
1372 u64 len = state->end + 1 - state->start;
1373 bool do_list = !btrfs_is_free_space_inode(root, inode);
1375 if (*bits & EXTENT_FIRST_DELALLOC) {
1376 *bits &= ~EXTENT_FIRST_DELALLOC;
1377 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1378 spin_lock(&BTRFS_I(inode)->lock);
1379 BTRFS_I(inode)->outstanding_extents--;
1380 spin_unlock(&BTRFS_I(inode)->lock);
1383 if (*bits & EXTENT_DO_ACCOUNTING)
1384 btrfs_delalloc_release_metadata(inode, len);
1386 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1388 btrfs_free_reserved_data_space(inode, len);
1390 spin_lock(&root->fs_info->delalloc_lock);
1391 root->fs_info->delalloc_bytes -= len;
1392 BTRFS_I(inode)->delalloc_bytes -= len;
1394 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1395 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1396 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1398 spin_unlock(&root->fs_info->delalloc_lock);
1403 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1404 * we don't create bios that span stripes or chunks
1406 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1407 size_t size, struct bio *bio,
1408 unsigned long bio_flags)
1410 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1411 struct btrfs_mapping_tree *map_tree;
1412 u64 logical = (u64)bio->bi_sector << 9;
1417 if (bio_flags & EXTENT_BIO_COMPRESSED)
1420 length = bio->bi_size;
1421 map_tree = &root->fs_info->mapping_tree;
1422 map_length = length;
1423 ret = btrfs_map_block(map_tree, READ, logical,
1424 &map_length, NULL, 0);
1426 if (map_length < length + size)
1432 * in order to insert checksums into the metadata in large chunks,
1433 * we wait until bio submission time. All the pages in the bio are
1434 * checksummed and sums are attached onto the ordered extent record.
1436 * At IO completion time the cums attached on the ordered extent record
1437 * are inserted into the btree
1439 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1440 struct bio *bio, int mirror_num,
1441 unsigned long bio_flags,
1444 struct btrfs_root *root = BTRFS_I(inode)->root;
1447 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1453 * in order to insert checksums into the metadata in large chunks,
1454 * we wait until bio submission time. All the pages in the bio are
1455 * checksummed and sums are attached onto the ordered extent record.
1457 * At IO completion time the cums attached on the ordered extent record
1458 * are inserted into the btree
1460 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1461 int mirror_num, unsigned long bio_flags,
1464 struct btrfs_root *root = BTRFS_I(inode)->root;
1465 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1469 * extent_io.c submission hook. This does the right thing for csum calculation
1470 * on write, or reading the csums from the tree before a read
1472 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1473 int mirror_num, unsigned long bio_flags,
1476 struct btrfs_root *root = BTRFS_I(inode)->root;
1480 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1482 if (btrfs_is_free_space_inode(root, inode))
1483 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1485 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1488 if (!(rw & REQ_WRITE)) {
1489 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1490 return btrfs_submit_compressed_read(inode, bio,
1491 mirror_num, bio_flags);
1492 } else if (!skip_sum) {
1493 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1498 } else if (!skip_sum) {
1499 /* csum items have already been cloned */
1500 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1502 /* we're doing a write, do the async checksumming */
1503 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1504 inode, rw, bio, mirror_num,
1505 bio_flags, bio_offset,
1506 __btrfs_submit_bio_start,
1507 __btrfs_submit_bio_done);
1511 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1515 * given a list of ordered sums record them in the inode. This happens
1516 * at IO completion time based on sums calculated at bio submission time.
1518 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1519 struct inode *inode, u64 file_offset,
1520 struct list_head *list)
1522 struct btrfs_ordered_sum *sum;
1524 list_for_each_entry(sum, list, list) {
1525 btrfs_csum_file_blocks(trans,
1526 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1531 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1532 struct extent_state **cached_state)
1534 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1536 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1537 cached_state, GFP_NOFS);
1540 /* see btrfs_writepage_start_hook for details on why this is required */
1541 struct btrfs_writepage_fixup {
1543 struct btrfs_work work;
1546 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1548 struct btrfs_writepage_fixup *fixup;
1549 struct btrfs_ordered_extent *ordered;
1550 struct extent_state *cached_state = NULL;
1552 struct inode *inode;
1556 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1560 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1561 ClearPageChecked(page);
1565 inode = page->mapping->host;
1566 page_start = page_offset(page);
1567 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1569 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1570 &cached_state, GFP_NOFS);
1572 /* already ordered? We're done */
1573 if (PagePrivate2(page))
1576 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1578 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1579 page_end, &cached_state, GFP_NOFS);
1581 btrfs_start_ordered_extent(inode, ordered, 1);
1586 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1587 ClearPageChecked(page);
1589 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1590 &cached_state, GFP_NOFS);
1593 page_cache_release(page);
1598 * There are a few paths in the higher layers of the kernel that directly
1599 * set the page dirty bit without asking the filesystem if it is a
1600 * good idea. This causes problems because we want to make sure COW
1601 * properly happens and the data=ordered rules are followed.
1603 * In our case any range that doesn't have the ORDERED bit set
1604 * hasn't been properly setup for IO. We kick off an async process
1605 * to fix it up. The async helper will wait for ordered extents, set
1606 * the delalloc bit and make it safe to write the page.
1608 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1610 struct inode *inode = page->mapping->host;
1611 struct btrfs_writepage_fixup *fixup;
1612 struct btrfs_root *root = BTRFS_I(inode)->root;
1614 /* this page is properly in the ordered list */
1615 if (TestClearPagePrivate2(page))
1618 if (PageChecked(page))
1621 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1625 SetPageChecked(page);
1626 page_cache_get(page);
1627 fixup->work.func = btrfs_writepage_fixup_worker;
1629 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1633 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1634 struct inode *inode, u64 file_pos,
1635 u64 disk_bytenr, u64 disk_num_bytes,
1636 u64 num_bytes, u64 ram_bytes,
1637 u8 compression, u8 encryption,
1638 u16 other_encoding, int extent_type)
1640 struct btrfs_root *root = BTRFS_I(inode)->root;
1641 struct btrfs_file_extent_item *fi;
1642 struct btrfs_path *path;
1643 struct extent_buffer *leaf;
1644 struct btrfs_key ins;
1648 path = btrfs_alloc_path();
1652 path->leave_spinning = 1;
1655 * we may be replacing one extent in the tree with another.
1656 * The new extent is pinned in the extent map, and we don't want
1657 * to drop it from the cache until it is completely in the btree.
1659 * So, tell btrfs_drop_extents to leave this extent in the cache.
1660 * the caller is expected to unpin it and allow it to be merged
1663 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1667 ins.objectid = btrfs_ino(inode);
1668 ins.offset = file_pos;
1669 ins.type = BTRFS_EXTENT_DATA_KEY;
1670 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1672 leaf = path->nodes[0];
1673 fi = btrfs_item_ptr(leaf, path->slots[0],
1674 struct btrfs_file_extent_item);
1675 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1676 btrfs_set_file_extent_type(leaf, fi, extent_type);
1677 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1678 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1679 btrfs_set_file_extent_offset(leaf, fi, 0);
1680 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1681 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1682 btrfs_set_file_extent_compression(leaf, fi, compression);
1683 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1684 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1686 btrfs_unlock_up_safe(path, 1);
1687 btrfs_set_lock_blocking(leaf);
1689 btrfs_mark_buffer_dirty(leaf);
1691 inode_add_bytes(inode, num_bytes);
1693 ins.objectid = disk_bytenr;
1694 ins.offset = disk_num_bytes;
1695 ins.type = BTRFS_EXTENT_ITEM_KEY;
1696 ret = btrfs_alloc_reserved_file_extent(trans, root,
1697 root->root_key.objectid,
1698 btrfs_ino(inode), file_pos, &ins);
1700 btrfs_free_path(path);
1706 * helper function for btrfs_finish_ordered_io, this
1707 * just reads in some of the csum leaves to prime them into ram
1708 * before we start the transaction. It limits the amount of btree
1709 * reads required while inside the transaction.
1711 /* as ordered data IO finishes, this gets called so we can finish
1712 * an ordered extent if the range of bytes in the file it covers are
1715 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1717 struct btrfs_root *root = BTRFS_I(inode)->root;
1718 struct btrfs_trans_handle *trans = NULL;
1719 struct btrfs_ordered_extent *ordered_extent = NULL;
1720 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1721 struct extent_state *cached_state = NULL;
1722 int compress_type = 0;
1726 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1730 BUG_ON(!ordered_extent);
1732 nolock = btrfs_is_free_space_inode(root, inode);
1734 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1735 BUG_ON(!list_empty(&ordered_extent->list));
1736 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1739 trans = btrfs_join_transaction_nolock(root);
1741 trans = btrfs_join_transaction(root);
1742 BUG_ON(IS_ERR(trans));
1743 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1744 ret = btrfs_update_inode(trans, root, inode);
1750 lock_extent_bits(io_tree, ordered_extent->file_offset,
1751 ordered_extent->file_offset + ordered_extent->len - 1,
1752 0, &cached_state, GFP_NOFS);
1755 trans = btrfs_join_transaction_nolock(root);
1757 trans = btrfs_join_transaction(root);
1758 BUG_ON(IS_ERR(trans));
1759 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1761 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1762 compress_type = ordered_extent->compress_type;
1763 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1764 BUG_ON(compress_type);
1765 ret = btrfs_mark_extent_written(trans, inode,
1766 ordered_extent->file_offset,
1767 ordered_extent->file_offset +
1768 ordered_extent->len);
1771 BUG_ON(root == root->fs_info->tree_root);
1772 ret = insert_reserved_file_extent(trans, inode,
1773 ordered_extent->file_offset,
1774 ordered_extent->start,
1775 ordered_extent->disk_len,
1776 ordered_extent->len,
1777 ordered_extent->len,
1778 compress_type, 0, 0,
1779 BTRFS_FILE_EXTENT_REG);
1780 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1781 ordered_extent->file_offset,
1782 ordered_extent->len);
1785 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1786 ordered_extent->file_offset +
1787 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1789 add_pending_csums(trans, inode, ordered_extent->file_offset,
1790 &ordered_extent->list);
1792 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1793 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1794 ret = btrfs_update_inode(trans, root, inode);
1799 if (root != root->fs_info->tree_root)
1800 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1803 btrfs_end_transaction_nolock(trans, root);
1805 btrfs_end_transaction(trans, root);
1809 btrfs_put_ordered_extent(ordered_extent);
1810 /* once for the tree */
1811 btrfs_put_ordered_extent(ordered_extent);
1816 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1817 struct extent_state *state, int uptodate)
1819 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1821 ClearPagePrivate2(page);
1822 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1826 * When IO fails, either with EIO or csum verification fails, we
1827 * try other mirrors that might have a good copy of the data. This
1828 * io_failure_record is used to record state as we go through all the
1829 * mirrors. If another mirror has good data, the page is set up to date
1830 * and things continue. If a good mirror can't be found, the original
1831 * bio end_io callback is called to indicate things have failed.
1833 struct io_failure_record {
1838 unsigned long bio_flags;
1842 static int btrfs_io_failed_hook(struct bio *failed_bio,
1843 struct page *page, u64 start, u64 end,
1844 struct extent_state *state)
1846 struct io_failure_record *failrec = NULL;
1848 struct extent_map *em;
1849 struct inode *inode = page->mapping->host;
1850 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1851 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1858 ret = get_state_private(failure_tree, start, &private);
1860 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1863 failrec->start = start;
1864 failrec->len = end - start + 1;
1865 failrec->last_mirror = 0;
1866 failrec->bio_flags = 0;
1868 read_lock(&em_tree->lock);
1869 em = lookup_extent_mapping(em_tree, start, failrec->len);
1870 if (em->start > start || em->start + em->len < start) {
1871 free_extent_map(em);
1874 read_unlock(&em_tree->lock);
1876 if (IS_ERR_OR_NULL(em)) {
1880 logical = start - em->start;
1881 logical = em->block_start + logical;
1882 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1883 logical = em->block_start;
1884 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1885 extent_set_compress_type(&failrec->bio_flags,
1888 failrec->logical = logical;
1889 free_extent_map(em);
1890 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1891 EXTENT_DIRTY, GFP_NOFS);
1892 set_state_private(failure_tree, start,
1893 (u64)(unsigned long)failrec);
1895 failrec = (struct io_failure_record *)(unsigned long)private;
1897 num_copies = btrfs_num_copies(
1898 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1899 failrec->logical, failrec->len);
1900 failrec->last_mirror++;
1902 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1903 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1906 if (state && state->start != failrec->start)
1908 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1910 if (!state || failrec->last_mirror > num_copies) {
1911 set_state_private(failure_tree, failrec->start, 0);
1912 clear_extent_bits(failure_tree, failrec->start,
1913 failrec->start + failrec->len - 1,
1914 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1918 bio = bio_alloc(GFP_NOFS, 1);
1919 bio->bi_private = state;
1920 bio->bi_end_io = failed_bio->bi_end_io;
1921 bio->bi_sector = failrec->logical >> 9;
1922 bio->bi_bdev = failed_bio->bi_bdev;
1925 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1926 if (failed_bio->bi_rw & REQ_WRITE)
1931 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1932 failrec->last_mirror,
1933 failrec->bio_flags, 0);
1938 * each time an IO finishes, we do a fast check in the IO failure tree
1939 * to see if we need to process or clean up an io_failure_record
1941 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1944 u64 private_failure;
1945 struct io_failure_record *failure;
1949 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1950 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1951 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1952 start, &private_failure);
1954 failure = (struct io_failure_record *)(unsigned long)
1956 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1958 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1960 failure->start + failure->len - 1,
1961 EXTENT_DIRTY | EXTENT_LOCKED,
1970 * when reads are done, we need to check csums to verify the data is correct
1971 * if there's a match, we allow the bio to finish. If not, we go through
1972 * the io_failure_record routines to find good copies
1974 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1975 struct extent_state *state)
1977 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1978 struct inode *inode = page->mapping->host;
1979 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1981 u64 private = ~(u32)0;
1983 struct btrfs_root *root = BTRFS_I(inode)->root;
1986 if (PageChecked(page)) {
1987 ClearPageChecked(page);
1991 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1994 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1995 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1996 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2001 if (state && state->start == start) {
2002 private = state->private;
2005 ret = get_state_private(io_tree, start, &private);
2007 kaddr = kmap_atomic(page, KM_USER0);
2011 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2012 btrfs_csum_final(csum, (char *)&csum);
2013 if (csum != private)
2016 kunmap_atomic(kaddr, KM_USER0);
2018 /* if the io failure tree for this inode is non-empty,
2019 * check to see if we've recovered from a failed IO
2021 btrfs_clean_io_failures(inode, start);
2025 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2027 (unsigned long long)btrfs_ino(page->mapping->host),
2028 (unsigned long long)start, csum,
2029 (unsigned long long)private);
2030 memset(kaddr + offset, 1, end - start + 1);
2031 flush_dcache_page(page);
2032 kunmap_atomic(kaddr, KM_USER0);
2038 struct delayed_iput {
2039 struct list_head list;
2040 struct inode *inode;
2043 void btrfs_add_delayed_iput(struct inode *inode)
2045 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2046 struct delayed_iput *delayed;
2048 if (atomic_add_unless(&inode->i_count, -1, 1))
2051 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2052 delayed->inode = inode;
2054 spin_lock(&fs_info->delayed_iput_lock);
2055 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2056 spin_unlock(&fs_info->delayed_iput_lock);
2059 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2062 struct btrfs_fs_info *fs_info = root->fs_info;
2063 struct delayed_iput *delayed;
2066 spin_lock(&fs_info->delayed_iput_lock);
2067 empty = list_empty(&fs_info->delayed_iputs);
2068 spin_unlock(&fs_info->delayed_iput_lock);
2072 down_read(&root->fs_info->cleanup_work_sem);
2073 spin_lock(&fs_info->delayed_iput_lock);
2074 list_splice_init(&fs_info->delayed_iputs, &list);
2075 spin_unlock(&fs_info->delayed_iput_lock);
2077 while (!list_empty(&list)) {
2078 delayed = list_entry(list.next, struct delayed_iput, list);
2079 list_del(&delayed->list);
2080 iput(delayed->inode);
2083 up_read(&root->fs_info->cleanup_work_sem);
2086 enum btrfs_orphan_cleanup_state {
2087 ORPHAN_CLEANUP_STARTED = 1,
2088 ORPHAN_CLEANUP_DONE = 2,
2092 * This is called in transaction commmit time. If there are no orphan
2093 * files in the subvolume, it removes orphan item and frees block_rsv
2096 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *root)
2101 if (!list_empty(&root->orphan_list) ||
2102 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2105 if (root->orphan_item_inserted &&
2106 btrfs_root_refs(&root->root_item) > 0) {
2107 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2108 root->root_key.objectid);
2110 root->orphan_item_inserted = 0;
2113 if (root->orphan_block_rsv) {
2114 WARN_ON(root->orphan_block_rsv->size > 0);
2115 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2116 root->orphan_block_rsv = NULL;
2121 * This creates an orphan entry for the given inode in case something goes
2122 * wrong in the middle of an unlink/truncate.
2124 * NOTE: caller of this function should reserve 5 units of metadata for
2127 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2129 struct btrfs_root *root = BTRFS_I(inode)->root;
2130 struct btrfs_block_rsv *block_rsv = NULL;
2135 if (!root->orphan_block_rsv) {
2136 block_rsv = btrfs_alloc_block_rsv(root);
2141 spin_lock(&root->orphan_lock);
2142 if (!root->orphan_block_rsv) {
2143 root->orphan_block_rsv = block_rsv;
2144 } else if (block_rsv) {
2145 btrfs_free_block_rsv(root, block_rsv);
2149 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2150 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2153 * For proper ENOSPC handling, we should do orphan
2154 * cleanup when mounting. But this introduces backward
2155 * compatibility issue.
2157 if (!xchg(&root->orphan_item_inserted, 1))
2165 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2166 BTRFS_I(inode)->orphan_meta_reserved = 1;
2169 spin_unlock(&root->orphan_lock);
2171 /* grab metadata reservation from transaction handle */
2173 ret = btrfs_orphan_reserve_metadata(trans, inode);
2177 /* insert an orphan item to track this unlinked/truncated file */
2179 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2183 /* insert an orphan item to track subvolume contains orphan files */
2185 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2186 root->root_key.objectid);
2193 * We have done the truncate/delete so we can go ahead and remove the orphan
2194 * item for this particular inode.
2196 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2198 struct btrfs_root *root = BTRFS_I(inode)->root;
2199 int delete_item = 0;
2200 int release_rsv = 0;
2203 spin_lock(&root->orphan_lock);
2204 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2205 list_del_init(&BTRFS_I(inode)->i_orphan);
2209 if (BTRFS_I(inode)->orphan_meta_reserved) {
2210 BTRFS_I(inode)->orphan_meta_reserved = 0;
2213 spin_unlock(&root->orphan_lock);
2215 if (trans && delete_item) {
2216 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2221 btrfs_orphan_release_metadata(inode);
2227 * this cleans up any orphans that may be left on the list from the last use
2230 int btrfs_orphan_cleanup(struct btrfs_root *root)
2232 struct btrfs_path *path;
2233 struct extent_buffer *leaf;
2234 struct btrfs_key key, found_key;
2235 struct btrfs_trans_handle *trans;
2236 struct inode *inode;
2237 u64 last_objectid = 0;
2238 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2240 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2243 path = btrfs_alloc_path();
2250 key.objectid = BTRFS_ORPHAN_OBJECTID;
2251 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2252 key.offset = (u64)-1;
2255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2260 * if ret == 0 means we found what we were searching for, which
2261 * is weird, but possible, so only screw with path if we didn't
2262 * find the key and see if we have stuff that matches
2266 if (path->slots[0] == 0)
2271 /* pull out the item */
2272 leaf = path->nodes[0];
2273 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2275 /* make sure the item matches what we want */
2276 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2278 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2281 /* release the path since we're done with it */
2282 btrfs_release_path(path);
2285 * this is where we are basically btrfs_lookup, without the
2286 * crossing root thing. we store the inode number in the
2287 * offset of the orphan item.
2290 if (found_key.offset == last_objectid) {
2291 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2292 "stopping orphan cleanup\n");
2297 last_objectid = found_key.offset;
2299 found_key.objectid = found_key.offset;
2300 found_key.type = BTRFS_INODE_ITEM_KEY;
2301 found_key.offset = 0;
2302 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2303 ret = PTR_RET(inode);
2304 if (ret && ret != -ESTALE)
2308 * Inode is already gone but the orphan item is still there,
2309 * kill the orphan item.
2311 if (ret == -ESTALE) {
2312 trans = btrfs_start_transaction(root, 1);
2313 if (IS_ERR(trans)) {
2314 ret = PTR_ERR(trans);
2317 ret = btrfs_del_orphan_item(trans, root,
2318 found_key.objectid);
2320 btrfs_end_transaction(trans, root);
2325 * add this inode to the orphan list so btrfs_orphan_del does
2326 * the proper thing when we hit it
2328 spin_lock(&root->orphan_lock);
2329 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2330 spin_unlock(&root->orphan_lock);
2332 /* if we have links, this was a truncate, lets do that */
2333 if (inode->i_nlink) {
2334 if (!S_ISREG(inode->i_mode)) {
2340 ret = btrfs_truncate(inode);
2345 /* this will do delete_inode and everything for us */
2350 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2352 if (root->orphan_block_rsv)
2353 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2356 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2357 trans = btrfs_join_transaction(root);
2359 btrfs_end_transaction(trans, root);
2363 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2365 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2369 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2370 btrfs_free_path(path);
2375 * very simple check to peek ahead in the leaf looking for xattrs. If we
2376 * don't find any xattrs, we know there can't be any acls.
2378 * slot is the slot the inode is in, objectid is the objectid of the inode
2380 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2381 int slot, u64 objectid)
2383 u32 nritems = btrfs_header_nritems(leaf);
2384 struct btrfs_key found_key;
2388 while (slot < nritems) {
2389 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2391 /* we found a different objectid, there must not be acls */
2392 if (found_key.objectid != objectid)
2395 /* we found an xattr, assume we've got an acl */
2396 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2400 * we found a key greater than an xattr key, there can't
2401 * be any acls later on
2403 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2410 * it goes inode, inode backrefs, xattrs, extents,
2411 * so if there are a ton of hard links to an inode there can
2412 * be a lot of backrefs. Don't waste time searching too hard,
2413 * this is just an optimization
2418 /* we hit the end of the leaf before we found an xattr or
2419 * something larger than an xattr. We have to assume the inode
2426 * read an inode from the btree into the in-memory inode
2428 static void btrfs_read_locked_inode(struct inode *inode)
2430 struct btrfs_path *path;
2431 struct extent_buffer *leaf;
2432 struct btrfs_inode_item *inode_item;
2433 struct btrfs_timespec *tspec;
2434 struct btrfs_root *root = BTRFS_I(inode)->root;
2435 struct btrfs_key location;
2439 bool filled = false;
2441 ret = btrfs_fill_inode(inode, &rdev);
2445 path = btrfs_alloc_path();
2449 path->leave_spinning = 1;
2450 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2452 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2456 leaf = path->nodes[0];
2461 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2462 struct btrfs_inode_item);
2463 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2464 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2465 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2466 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2467 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2469 tspec = btrfs_inode_atime(inode_item);
2470 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2471 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2473 tspec = btrfs_inode_mtime(inode_item);
2474 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2475 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2477 tspec = btrfs_inode_ctime(inode_item);
2478 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2479 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2481 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2482 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2483 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2484 inode->i_generation = BTRFS_I(inode)->generation;
2486 rdev = btrfs_inode_rdev(leaf, inode_item);
2488 BTRFS_I(inode)->index_cnt = (u64)-1;
2489 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2492 * try to precache a NULL acl entry for files that don't have
2493 * any xattrs or acls
2495 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2498 cache_no_acl(inode);
2500 btrfs_free_path(path);
2502 switch (inode->i_mode & S_IFMT) {
2504 inode->i_mapping->a_ops = &btrfs_aops;
2505 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2506 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2507 inode->i_fop = &btrfs_file_operations;
2508 inode->i_op = &btrfs_file_inode_operations;
2511 inode->i_fop = &btrfs_dir_file_operations;
2512 if (root == root->fs_info->tree_root)
2513 inode->i_op = &btrfs_dir_ro_inode_operations;
2515 inode->i_op = &btrfs_dir_inode_operations;
2518 inode->i_op = &btrfs_symlink_inode_operations;
2519 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2520 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2523 inode->i_op = &btrfs_special_inode_operations;
2524 init_special_inode(inode, inode->i_mode, rdev);
2528 btrfs_update_iflags(inode);
2532 btrfs_free_path(path);
2533 make_bad_inode(inode);
2537 * given a leaf and an inode, copy the inode fields into the leaf
2539 static void fill_inode_item(struct btrfs_trans_handle *trans,
2540 struct extent_buffer *leaf,
2541 struct btrfs_inode_item *item,
2542 struct inode *inode)
2544 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2545 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2546 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2547 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2548 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2550 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2551 inode->i_atime.tv_sec);
2552 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2553 inode->i_atime.tv_nsec);
2555 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2556 inode->i_mtime.tv_sec);
2557 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2558 inode->i_mtime.tv_nsec);
2560 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2561 inode->i_ctime.tv_sec);
2562 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2563 inode->i_ctime.tv_nsec);
2565 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2566 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2567 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2568 btrfs_set_inode_transid(leaf, item, trans->transid);
2569 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2570 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2571 btrfs_set_inode_block_group(leaf, item, 0);
2575 * copy everything in the in-memory inode into the btree.
2577 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2578 struct btrfs_root *root, struct inode *inode)
2580 struct btrfs_inode_item *inode_item;
2581 struct btrfs_path *path;
2582 struct extent_buffer *leaf;
2586 * If the inode is a free space inode, we can deadlock during commit
2587 * if we put it into the delayed code.
2589 * The data relocation inode should also be directly updated
2592 if (!btrfs_is_free_space_inode(root, inode)
2593 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2594 ret = btrfs_delayed_update_inode(trans, root, inode);
2596 btrfs_set_inode_last_trans(trans, inode);
2600 path = btrfs_alloc_path();
2604 path->leave_spinning = 1;
2605 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2613 btrfs_unlock_up_safe(path, 1);
2614 leaf = path->nodes[0];
2615 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2616 struct btrfs_inode_item);
2618 fill_inode_item(trans, leaf, inode_item, inode);
2619 btrfs_mark_buffer_dirty(leaf);
2620 btrfs_set_inode_last_trans(trans, inode);
2623 btrfs_free_path(path);
2628 * unlink helper that gets used here in inode.c and in the tree logging
2629 * recovery code. It remove a link in a directory with a given name, and
2630 * also drops the back refs in the inode to the directory
2632 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2633 struct btrfs_root *root,
2634 struct inode *dir, struct inode *inode,
2635 const char *name, int name_len)
2637 struct btrfs_path *path;
2639 struct extent_buffer *leaf;
2640 struct btrfs_dir_item *di;
2641 struct btrfs_key key;
2643 u64 ino = btrfs_ino(inode);
2644 u64 dir_ino = btrfs_ino(dir);
2646 path = btrfs_alloc_path();
2652 path->leave_spinning = 1;
2653 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2654 name, name_len, -1);
2663 leaf = path->nodes[0];
2664 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2665 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2668 btrfs_release_path(path);
2670 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2673 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2674 "inode %llu parent %llu\n", name_len, name,
2675 (unsigned long long)ino, (unsigned long long)dir_ino);
2679 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2683 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2685 BUG_ON(ret != 0 && ret != -ENOENT);
2687 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2692 btrfs_free_path(path);
2696 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2697 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2698 btrfs_update_inode(trans, root, dir);
2703 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2704 struct btrfs_root *root,
2705 struct inode *dir, struct inode *inode,
2706 const char *name, int name_len)
2709 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2711 btrfs_drop_nlink(inode);
2712 ret = btrfs_update_inode(trans, root, inode);
2718 /* helper to check if there is any shared block in the path */
2719 static int check_path_shared(struct btrfs_root *root,
2720 struct btrfs_path *path)
2722 struct extent_buffer *eb;
2726 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2729 if (!path->nodes[level])
2731 eb = path->nodes[level];
2732 if (!btrfs_block_can_be_shared(root, eb))
2734 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2743 * helper to start transaction for unlink and rmdir.
2745 * unlink and rmdir are special in btrfs, they do not always free space.
2746 * so in enospc case, we should make sure they will free space before
2747 * allowing them to use the global metadata reservation.
2749 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2750 struct dentry *dentry)
2752 struct btrfs_trans_handle *trans;
2753 struct btrfs_root *root = BTRFS_I(dir)->root;
2754 struct btrfs_path *path;
2755 struct btrfs_inode_ref *ref;
2756 struct btrfs_dir_item *di;
2757 struct inode *inode = dentry->d_inode;
2762 u64 ino = btrfs_ino(inode);
2763 u64 dir_ino = btrfs_ino(dir);
2766 * 1 for the possible orphan item
2767 * 1 for the dir item
2768 * 1 for the dir index
2769 * 1 for the inode ref
2770 * 1 for the inode ref in the tree log
2771 * 2 for the dir entries in the log
2774 trans = btrfs_start_transaction(root, 8);
2775 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2778 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2779 return ERR_PTR(-ENOSPC);
2781 /* check if there is someone else holds reference */
2782 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2783 return ERR_PTR(-ENOSPC);
2785 if (atomic_read(&inode->i_count) > 2)
2786 return ERR_PTR(-ENOSPC);
2788 if (xchg(&root->fs_info->enospc_unlink, 1))
2789 return ERR_PTR(-ENOSPC);
2791 path = btrfs_alloc_path();
2793 root->fs_info->enospc_unlink = 0;
2794 return ERR_PTR(-ENOMEM);
2797 /* 1 for the orphan item */
2798 trans = btrfs_start_transaction(root, 1);
2799 if (IS_ERR(trans)) {
2800 btrfs_free_path(path);
2801 root->fs_info->enospc_unlink = 0;
2805 path->skip_locking = 1;
2806 path->search_commit_root = 1;
2808 ret = btrfs_lookup_inode(trans, root, path,
2809 &BTRFS_I(dir)->location, 0);
2815 if (check_path_shared(root, path))
2820 btrfs_release_path(path);
2822 ret = btrfs_lookup_inode(trans, root, path,
2823 &BTRFS_I(inode)->location, 0);
2829 if (check_path_shared(root, path))
2834 btrfs_release_path(path);
2836 if (ret == 0 && S_ISREG(inode->i_mode)) {
2837 ret = btrfs_lookup_file_extent(trans, root, path,
2844 if (check_path_shared(root, path))
2846 btrfs_release_path(path);
2854 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2855 dentry->d_name.name, dentry->d_name.len, 0);
2861 if (check_path_shared(root, path))
2867 btrfs_release_path(path);
2869 ref = btrfs_lookup_inode_ref(trans, root, path,
2870 dentry->d_name.name, dentry->d_name.len,
2877 if (check_path_shared(root, path))
2879 index = btrfs_inode_ref_index(path->nodes[0], ref);
2880 btrfs_release_path(path);
2883 * This is a commit root search, if we can lookup inode item and other
2884 * relative items in the commit root, it means the transaction of
2885 * dir/file creation has been committed, and the dir index item that we
2886 * delay to insert has also been inserted into the commit root. So
2887 * we needn't worry about the delayed insertion of the dir index item
2890 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2891 dentry->d_name.name, dentry->d_name.len, 0);
2896 BUG_ON(ret == -ENOENT);
2897 if (check_path_shared(root, path))
2902 btrfs_free_path(path);
2903 /* Migrate the orphan reservation over */
2905 err = btrfs_block_rsv_migrate(trans->block_rsv,
2906 &root->fs_info->global_block_rsv,
2907 trans->bytes_reserved);
2910 btrfs_end_transaction(trans, root);
2911 root->fs_info->enospc_unlink = 0;
2912 return ERR_PTR(err);
2915 trans->block_rsv = &root->fs_info->global_block_rsv;
2919 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2920 struct btrfs_root *root)
2922 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2923 btrfs_block_rsv_release(root, trans->block_rsv,
2924 trans->bytes_reserved);
2925 trans->block_rsv = &root->fs_info->trans_block_rsv;
2926 BUG_ON(!root->fs_info->enospc_unlink);
2927 root->fs_info->enospc_unlink = 0;
2929 btrfs_end_transaction_throttle(trans, root);
2932 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2934 struct btrfs_root *root = BTRFS_I(dir)->root;
2935 struct btrfs_trans_handle *trans;
2936 struct inode *inode = dentry->d_inode;
2938 unsigned long nr = 0;
2940 trans = __unlink_start_trans(dir, dentry);
2942 return PTR_ERR(trans);
2944 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2946 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2947 dentry->d_name.name, dentry->d_name.len);
2951 if (inode->i_nlink == 0) {
2952 ret = btrfs_orphan_add(trans, inode);
2958 nr = trans->blocks_used;
2959 __unlink_end_trans(trans, root);
2960 btrfs_btree_balance_dirty(root, nr);
2964 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2965 struct btrfs_root *root,
2966 struct inode *dir, u64 objectid,
2967 const char *name, int name_len)
2969 struct btrfs_path *path;
2970 struct extent_buffer *leaf;
2971 struct btrfs_dir_item *di;
2972 struct btrfs_key key;
2975 u64 dir_ino = btrfs_ino(dir);
2977 path = btrfs_alloc_path();
2981 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2982 name, name_len, -1);
2983 BUG_ON(IS_ERR_OR_NULL(di));
2985 leaf = path->nodes[0];
2986 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2987 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2988 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2990 btrfs_release_path(path);
2992 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2993 objectid, root->root_key.objectid,
2994 dir_ino, &index, name, name_len);
2996 BUG_ON(ret != -ENOENT);
2997 di = btrfs_search_dir_index_item(root, path, dir_ino,
2999 BUG_ON(IS_ERR_OR_NULL(di));
3001 leaf = path->nodes[0];
3002 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3003 btrfs_release_path(path);
3006 btrfs_release_path(path);
3008 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3011 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3012 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3013 ret = btrfs_update_inode(trans, root, dir);
3016 btrfs_free_path(path);
3020 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3022 struct inode *inode = dentry->d_inode;
3024 struct btrfs_root *root = BTRFS_I(dir)->root;
3025 struct btrfs_trans_handle *trans;
3026 unsigned long nr = 0;
3028 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3029 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3032 trans = __unlink_start_trans(dir, dentry);
3034 return PTR_ERR(trans);
3036 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3037 err = btrfs_unlink_subvol(trans, root, dir,
3038 BTRFS_I(inode)->location.objectid,
3039 dentry->d_name.name,
3040 dentry->d_name.len);
3044 err = btrfs_orphan_add(trans, inode);
3048 /* now the directory is empty */
3049 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3050 dentry->d_name.name, dentry->d_name.len);
3052 btrfs_i_size_write(inode, 0);
3054 nr = trans->blocks_used;
3055 __unlink_end_trans(trans, root);
3056 btrfs_btree_balance_dirty(root, nr);
3062 * this can truncate away extent items, csum items and directory items.
3063 * It starts at a high offset and removes keys until it can't find
3064 * any higher than new_size
3066 * csum items that cross the new i_size are truncated to the new size
3069 * min_type is the minimum key type to truncate down to. If set to 0, this
3070 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3072 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3073 struct btrfs_root *root,
3074 struct inode *inode,
3075 u64 new_size, u32 min_type)
3077 struct btrfs_path *path;
3078 struct extent_buffer *leaf;
3079 struct btrfs_file_extent_item *fi;
3080 struct btrfs_key key;
3081 struct btrfs_key found_key;
3082 u64 extent_start = 0;
3083 u64 extent_num_bytes = 0;
3084 u64 extent_offset = 0;
3086 u64 mask = root->sectorsize - 1;
3087 u32 found_type = (u8)-1;
3090 int pending_del_nr = 0;
3091 int pending_del_slot = 0;
3092 int extent_type = -1;
3096 u64 ino = btrfs_ino(inode);
3098 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3100 path = btrfs_alloc_path();
3105 if (root->ref_cows || root == root->fs_info->tree_root)
3106 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3109 * This function is also used to drop the items in the log tree before
3110 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3111 * it is used to drop the loged items. So we shouldn't kill the delayed
3114 if (min_type == 0 && root == BTRFS_I(inode)->root)
3115 btrfs_kill_delayed_inode_items(inode);
3118 key.offset = (u64)-1;
3122 path->leave_spinning = 1;
3123 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3130 /* there are no items in the tree for us to truncate, we're
3133 if (path->slots[0] == 0)
3140 leaf = path->nodes[0];
3141 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3142 found_type = btrfs_key_type(&found_key);
3145 if (found_key.objectid != ino)
3148 if (found_type < min_type)
3151 item_end = found_key.offset;
3152 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3153 fi = btrfs_item_ptr(leaf, path->slots[0],
3154 struct btrfs_file_extent_item);
3155 extent_type = btrfs_file_extent_type(leaf, fi);
3156 encoding = btrfs_file_extent_compression(leaf, fi);
3157 encoding |= btrfs_file_extent_encryption(leaf, fi);
3158 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3160 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3162 btrfs_file_extent_num_bytes(leaf, fi);
3163 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3164 item_end += btrfs_file_extent_inline_len(leaf,
3169 if (found_type > min_type) {
3172 if (item_end < new_size)
3174 if (found_key.offset >= new_size)
3180 /* FIXME, shrink the extent if the ref count is only 1 */
3181 if (found_type != BTRFS_EXTENT_DATA_KEY)
3184 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3186 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3187 if (!del_item && !encoding) {
3188 u64 orig_num_bytes =
3189 btrfs_file_extent_num_bytes(leaf, fi);
3190 extent_num_bytes = new_size -
3191 found_key.offset + root->sectorsize - 1;
3192 extent_num_bytes = extent_num_bytes &
3193 ~((u64)root->sectorsize - 1);
3194 btrfs_set_file_extent_num_bytes(leaf, fi,
3196 num_dec = (orig_num_bytes -
3198 if (root->ref_cows && extent_start != 0)
3199 inode_sub_bytes(inode, num_dec);
3200 btrfs_mark_buffer_dirty(leaf);
3203 btrfs_file_extent_disk_num_bytes(leaf,
3205 extent_offset = found_key.offset -
3206 btrfs_file_extent_offset(leaf, fi);
3208 /* FIXME blocksize != 4096 */
3209 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3210 if (extent_start != 0) {
3213 inode_sub_bytes(inode, num_dec);
3216 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3218 * we can't truncate inline items that have had
3222 btrfs_file_extent_compression(leaf, fi) == 0 &&
3223 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3224 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3225 u32 size = new_size - found_key.offset;
3227 if (root->ref_cows) {
3228 inode_sub_bytes(inode, item_end + 1 -
3232 btrfs_file_extent_calc_inline_size(size);
3233 ret = btrfs_truncate_item(trans, root, path,
3235 } else if (root->ref_cows) {
3236 inode_sub_bytes(inode, item_end + 1 -
3242 if (!pending_del_nr) {
3243 /* no pending yet, add ourselves */
3244 pending_del_slot = path->slots[0];
3246 } else if (pending_del_nr &&
3247 path->slots[0] + 1 == pending_del_slot) {
3248 /* hop on the pending chunk */
3250 pending_del_slot = path->slots[0];
3257 if (found_extent && (root->ref_cows ||
3258 root == root->fs_info->tree_root)) {
3259 btrfs_set_path_blocking(path);
3260 ret = btrfs_free_extent(trans, root, extent_start,
3261 extent_num_bytes, 0,
3262 btrfs_header_owner(leaf),
3263 ino, extent_offset);
3267 if (found_type == BTRFS_INODE_ITEM_KEY)
3270 if (path->slots[0] == 0 ||
3271 path->slots[0] != pending_del_slot) {
3272 if (root->ref_cows &&
3273 BTRFS_I(inode)->location.objectid !=
3274 BTRFS_FREE_INO_OBJECTID) {
3278 if (pending_del_nr) {
3279 ret = btrfs_del_items(trans, root, path,
3285 btrfs_release_path(path);
3292 if (pending_del_nr) {
3293 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3297 btrfs_free_path(path);
3302 * taken from block_truncate_page, but does cow as it zeros out
3303 * any bytes left in the last page in the file.
3305 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3307 struct inode *inode = mapping->host;
3308 struct btrfs_root *root = BTRFS_I(inode)->root;
3309 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3310 struct btrfs_ordered_extent *ordered;
3311 struct extent_state *cached_state = NULL;
3313 u32 blocksize = root->sectorsize;
3314 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3315 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3317 gfp_t mask = btrfs_alloc_write_mask(mapping);
3322 if ((offset & (blocksize - 1)) == 0)
3324 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3330 page = find_or_create_page(mapping, index, mask);
3332 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3336 page_start = page_offset(page);
3337 page_end = page_start + PAGE_CACHE_SIZE - 1;
3339 if (!PageUptodate(page)) {
3340 ret = btrfs_readpage(NULL, page);
3342 if (page->mapping != mapping) {
3344 page_cache_release(page);
3347 if (!PageUptodate(page)) {
3352 wait_on_page_writeback(page);
3354 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3356 set_page_extent_mapped(page);
3358 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3360 unlock_extent_cached(io_tree, page_start, page_end,
3361 &cached_state, GFP_NOFS);
3363 page_cache_release(page);
3364 btrfs_start_ordered_extent(inode, ordered, 1);
3365 btrfs_put_ordered_extent(ordered);
3369 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3370 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3371 0, 0, &cached_state, GFP_NOFS);
3373 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3376 unlock_extent_cached(io_tree, page_start, page_end,
3377 &cached_state, GFP_NOFS);
3382 if (offset != PAGE_CACHE_SIZE) {
3384 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3385 flush_dcache_page(page);
3388 ClearPageChecked(page);
3389 set_page_dirty(page);
3390 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3395 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3397 page_cache_release(page);
3403 * This function puts in dummy file extents for the area we're creating a hole
3404 * for. So if we are truncating this file to a larger size we need to insert
3405 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3406 * the range between oldsize and size
3408 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3410 struct btrfs_trans_handle *trans;
3411 struct btrfs_root *root = BTRFS_I(inode)->root;
3412 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3413 struct extent_map *em = NULL;
3414 struct extent_state *cached_state = NULL;
3415 u64 mask = root->sectorsize - 1;
3416 u64 hole_start = (oldsize + mask) & ~mask;
3417 u64 block_end = (size + mask) & ~mask;
3423 if (size <= hole_start)
3427 struct btrfs_ordered_extent *ordered;
3428 btrfs_wait_ordered_range(inode, hole_start,
3429 block_end - hole_start);
3430 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3431 &cached_state, GFP_NOFS);
3432 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3435 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3436 &cached_state, GFP_NOFS);
3437 btrfs_put_ordered_extent(ordered);
3440 cur_offset = hole_start;
3442 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3443 block_end - cur_offset, 0);
3444 BUG_ON(IS_ERR_OR_NULL(em));
3445 last_byte = min(extent_map_end(em), block_end);
3446 last_byte = (last_byte + mask) & ~mask;
3447 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3449 hole_size = last_byte - cur_offset;
3451 trans = btrfs_start_transaction(root, 2);
3452 if (IS_ERR(trans)) {
3453 err = PTR_ERR(trans);
3457 err = btrfs_drop_extents(trans, inode, cur_offset,
3458 cur_offset + hole_size,
3461 btrfs_end_transaction(trans, root);
3465 err = btrfs_insert_file_extent(trans, root,
3466 btrfs_ino(inode), cur_offset, 0,
3467 0, hole_size, 0, hole_size,
3470 btrfs_end_transaction(trans, root);
3474 btrfs_drop_extent_cache(inode, hole_start,
3477 btrfs_end_transaction(trans, root);
3479 free_extent_map(em);
3481 cur_offset = last_byte;
3482 if (cur_offset >= block_end)
3486 free_extent_map(em);
3487 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3492 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3494 loff_t oldsize = i_size_read(inode);
3497 if (newsize == oldsize)
3500 if (newsize > oldsize) {
3501 i_size_write(inode, newsize);
3502 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3503 truncate_pagecache(inode, oldsize, newsize);
3504 ret = btrfs_cont_expand(inode, oldsize, newsize);
3506 btrfs_setsize(inode, oldsize);
3510 mark_inode_dirty(inode);
3514 * We're truncating a file that used to have good data down to
3515 * zero. Make sure it gets into the ordered flush list so that
3516 * any new writes get down to disk quickly.
3519 BTRFS_I(inode)->ordered_data_close = 1;
3521 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3522 truncate_setsize(inode, newsize);
3523 ret = btrfs_truncate(inode);
3529 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3531 struct inode *inode = dentry->d_inode;
3532 struct btrfs_root *root = BTRFS_I(inode)->root;
3535 if (btrfs_root_readonly(root))
3538 err = inode_change_ok(inode, attr);
3542 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3543 err = btrfs_setsize(inode, attr->ia_size);
3548 if (attr->ia_valid) {
3549 setattr_copy(inode, attr);
3550 mark_inode_dirty(inode);
3552 if (attr->ia_valid & ATTR_MODE)
3553 err = btrfs_acl_chmod(inode);
3559 void btrfs_evict_inode(struct inode *inode)
3561 struct btrfs_trans_handle *trans;
3562 struct btrfs_root *root = BTRFS_I(inode)->root;
3563 struct btrfs_block_rsv *rsv, *global_rsv;
3564 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3568 trace_btrfs_inode_evict(inode);
3570 truncate_inode_pages(&inode->i_data, 0);
3571 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3572 btrfs_is_free_space_inode(root, inode)))
3575 if (is_bad_inode(inode)) {
3576 btrfs_orphan_del(NULL, inode);
3579 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3580 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3582 if (root->fs_info->log_root_recovering) {
3583 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3587 if (inode->i_nlink > 0) {
3588 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3592 rsv = btrfs_alloc_block_rsv(root);
3594 btrfs_orphan_del(NULL, inode);
3597 rsv->size = min_size;
3598 global_rsv = &root->fs_info->global_block_rsv;
3600 btrfs_i_size_write(inode, 0);
3603 * This is a bit simpler than btrfs_truncate since
3605 * 1) We've already reserved our space for our orphan item in the
3607 * 2) We're going to delete the inode item, so we don't need to update
3610 * So we just need to reserve some slack space in case we add bytes when
3611 * doing the truncate.
3614 ret = btrfs_block_rsv_refill(root, rsv, min_size);
3617 * Try and steal from the global reserve since we will
3618 * likely not use this space anyway, we want to try as
3619 * hard as possible to get this to work.
3622 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3625 printk(KERN_WARNING "Could not get space for a "
3626 "delete, will truncate on mount %d\n", ret);
3627 btrfs_orphan_del(NULL, inode);
3628 btrfs_free_block_rsv(root, rsv);
3632 trans = btrfs_start_transaction(root, 0);
3633 if (IS_ERR(trans)) {
3634 btrfs_orphan_del(NULL, inode);
3635 btrfs_free_block_rsv(root, rsv);
3639 trans->block_rsv = rsv;
3641 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3645 nr = trans->blocks_used;
3646 btrfs_end_transaction(trans, root);
3648 btrfs_btree_balance_dirty(root, nr);
3651 btrfs_free_block_rsv(root, rsv);
3654 trans->block_rsv = root->orphan_block_rsv;
3655 ret = btrfs_orphan_del(trans, inode);
3659 trans->block_rsv = &root->fs_info->trans_block_rsv;
3660 if (!(root == root->fs_info->tree_root ||
3661 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3662 btrfs_return_ino(root, btrfs_ino(inode));
3664 nr = trans->blocks_used;
3665 btrfs_end_transaction(trans, root);
3666 btrfs_btree_balance_dirty(root, nr);
3668 end_writeback(inode);
3673 * this returns the key found in the dir entry in the location pointer.
3674 * If no dir entries were found, location->objectid is 0.
3676 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3677 struct btrfs_key *location)
3679 const char *name = dentry->d_name.name;
3680 int namelen = dentry->d_name.len;
3681 struct btrfs_dir_item *di;
3682 struct btrfs_path *path;
3683 struct btrfs_root *root = BTRFS_I(dir)->root;
3686 path = btrfs_alloc_path();
3690 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3695 if (IS_ERR_OR_NULL(di))
3698 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3700 btrfs_free_path(path);
3703 location->objectid = 0;
3708 * when we hit a tree root in a directory, the btrfs part of the inode
3709 * needs to be changed to reflect the root directory of the tree root. This
3710 * is kind of like crossing a mount point.
3712 static int fixup_tree_root_location(struct btrfs_root *root,
3714 struct dentry *dentry,
3715 struct btrfs_key *location,
3716 struct btrfs_root **sub_root)
3718 struct btrfs_path *path;
3719 struct btrfs_root *new_root;
3720 struct btrfs_root_ref *ref;
3721 struct extent_buffer *leaf;
3725 path = btrfs_alloc_path();
3732 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3733 BTRFS_I(dir)->root->root_key.objectid,
3734 location->objectid);
3741 leaf = path->nodes[0];
3742 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3743 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3744 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3747 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3748 (unsigned long)(ref + 1),
3749 dentry->d_name.len);
3753 btrfs_release_path(path);
3755 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3756 if (IS_ERR(new_root)) {
3757 err = PTR_ERR(new_root);
3761 if (btrfs_root_refs(&new_root->root_item) == 0) {
3766 *sub_root = new_root;
3767 location->objectid = btrfs_root_dirid(&new_root->root_item);
3768 location->type = BTRFS_INODE_ITEM_KEY;
3769 location->offset = 0;
3772 btrfs_free_path(path);
3776 static void inode_tree_add(struct inode *inode)
3778 struct btrfs_root *root = BTRFS_I(inode)->root;
3779 struct btrfs_inode *entry;
3781 struct rb_node *parent;
3782 u64 ino = btrfs_ino(inode);
3784 p = &root->inode_tree.rb_node;
3787 if (inode_unhashed(inode))
3790 spin_lock(&root->inode_lock);
3793 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3795 if (ino < btrfs_ino(&entry->vfs_inode))
3796 p = &parent->rb_left;
3797 else if (ino > btrfs_ino(&entry->vfs_inode))
3798 p = &parent->rb_right;
3800 WARN_ON(!(entry->vfs_inode.i_state &
3801 (I_WILL_FREE | I_FREEING)));
3802 rb_erase(parent, &root->inode_tree);
3803 RB_CLEAR_NODE(parent);
3804 spin_unlock(&root->inode_lock);
3808 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3809 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3810 spin_unlock(&root->inode_lock);
3813 static void inode_tree_del(struct inode *inode)
3815 struct btrfs_root *root = BTRFS_I(inode)->root;
3818 spin_lock(&root->inode_lock);
3819 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3820 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3821 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3822 empty = RB_EMPTY_ROOT(&root->inode_tree);
3824 spin_unlock(&root->inode_lock);
3827 * Free space cache has inodes in the tree root, but the tree root has a
3828 * root_refs of 0, so this could end up dropping the tree root as a
3829 * snapshot, so we need the extra !root->fs_info->tree_root check to
3830 * make sure we don't drop it.
3832 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3833 root != root->fs_info->tree_root) {
3834 synchronize_srcu(&root->fs_info->subvol_srcu);
3835 spin_lock(&root->inode_lock);
3836 empty = RB_EMPTY_ROOT(&root->inode_tree);
3837 spin_unlock(&root->inode_lock);
3839 btrfs_add_dead_root(root);
3843 int btrfs_invalidate_inodes(struct btrfs_root *root)
3845 struct rb_node *node;
3846 struct rb_node *prev;
3847 struct btrfs_inode *entry;
3848 struct inode *inode;
3851 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3853 spin_lock(&root->inode_lock);
3855 node = root->inode_tree.rb_node;
3859 entry = rb_entry(node, struct btrfs_inode, rb_node);
3861 if (objectid < btrfs_ino(&entry->vfs_inode))
3862 node = node->rb_left;
3863 else if (objectid > btrfs_ino(&entry->vfs_inode))
3864 node = node->rb_right;
3870 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3871 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3875 prev = rb_next(prev);
3879 entry = rb_entry(node, struct btrfs_inode, rb_node);
3880 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3881 inode = igrab(&entry->vfs_inode);
3883 spin_unlock(&root->inode_lock);
3884 if (atomic_read(&inode->i_count) > 1)
3885 d_prune_aliases(inode);
3887 * btrfs_drop_inode will have it removed from
3888 * the inode cache when its usage count
3893 spin_lock(&root->inode_lock);
3897 if (cond_resched_lock(&root->inode_lock))
3900 node = rb_next(node);
3902 spin_unlock(&root->inode_lock);
3906 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3908 struct btrfs_iget_args *args = p;
3909 inode->i_ino = args->ino;
3910 BTRFS_I(inode)->root = args->root;
3911 btrfs_set_inode_space_info(args->root, inode);
3915 static int btrfs_find_actor(struct inode *inode, void *opaque)
3917 struct btrfs_iget_args *args = opaque;
3918 return args->ino == btrfs_ino(inode) &&
3919 args->root == BTRFS_I(inode)->root;
3922 static struct inode *btrfs_iget_locked(struct super_block *s,
3924 struct btrfs_root *root)
3926 struct inode *inode;
3927 struct btrfs_iget_args args;
3928 args.ino = objectid;
3931 inode = iget5_locked(s, objectid, btrfs_find_actor,
3932 btrfs_init_locked_inode,
3937 /* Get an inode object given its location and corresponding root.
3938 * Returns in *is_new if the inode was read from disk
3940 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3941 struct btrfs_root *root, int *new)
3943 struct inode *inode;
3945 inode = btrfs_iget_locked(s, location->objectid, root);
3947 return ERR_PTR(-ENOMEM);
3949 if (inode->i_state & I_NEW) {
3950 BTRFS_I(inode)->root = root;
3951 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3952 btrfs_read_locked_inode(inode);
3953 if (!is_bad_inode(inode)) {
3954 inode_tree_add(inode);
3955 unlock_new_inode(inode);
3959 unlock_new_inode(inode);
3961 inode = ERR_PTR(-ESTALE);
3968 static struct inode *new_simple_dir(struct super_block *s,
3969 struct btrfs_key *key,
3970 struct btrfs_root *root)
3972 struct inode *inode = new_inode(s);
3975 return ERR_PTR(-ENOMEM);
3977 BTRFS_I(inode)->root = root;
3978 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3979 BTRFS_I(inode)->dummy_inode = 1;
3981 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3982 inode->i_op = &simple_dir_inode_operations;
3983 inode->i_fop = &simple_dir_operations;
3984 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3985 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3990 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3992 struct inode *inode;
3993 struct btrfs_root *root = BTRFS_I(dir)->root;
3994 struct btrfs_root *sub_root = root;
3995 struct btrfs_key location;
3999 if (dentry->d_name.len > BTRFS_NAME_LEN)
4000 return ERR_PTR(-ENAMETOOLONG);
4002 if (unlikely(d_need_lookup(dentry))) {
4003 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4004 kfree(dentry->d_fsdata);
4005 dentry->d_fsdata = NULL;
4006 /* This thing is hashed, drop it for now */
4009 ret = btrfs_inode_by_name(dir, dentry, &location);
4013 return ERR_PTR(ret);
4015 if (location.objectid == 0)
4018 if (location.type == BTRFS_INODE_ITEM_KEY) {
4019 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4023 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4025 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4026 ret = fixup_tree_root_location(root, dir, dentry,
4027 &location, &sub_root);
4030 inode = ERR_PTR(ret);
4032 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4034 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4036 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4038 if (!IS_ERR(inode) && root != sub_root) {
4039 down_read(&root->fs_info->cleanup_work_sem);
4040 if (!(inode->i_sb->s_flags & MS_RDONLY))
4041 ret = btrfs_orphan_cleanup(sub_root);
4042 up_read(&root->fs_info->cleanup_work_sem);
4044 inode = ERR_PTR(ret);
4050 static int btrfs_dentry_delete(const struct dentry *dentry)
4052 struct btrfs_root *root;
4054 if (!dentry->d_inode && !IS_ROOT(dentry))
4055 dentry = dentry->d_parent;
4057 if (dentry->d_inode) {
4058 root = BTRFS_I(dentry->d_inode)->root;
4059 if (btrfs_root_refs(&root->root_item) == 0)
4065 static void btrfs_dentry_release(struct dentry *dentry)
4067 if (dentry->d_fsdata)
4068 kfree(dentry->d_fsdata);
4071 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4072 struct nameidata *nd)
4076 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4077 if (unlikely(d_need_lookup(dentry))) {
4078 spin_lock(&dentry->d_lock);
4079 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4080 spin_unlock(&dentry->d_lock);
4085 unsigned char btrfs_filetype_table[] = {
4086 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4089 static int btrfs_real_readdir(struct file *filp, void *dirent,
4092 struct inode *inode = filp->f_dentry->d_inode;
4093 struct btrfs_root *root = BTRFS_I(inode)->root;
4094 struct btrfs_item *item;
4095 struct btrfs_dir_item *di;
4096 struct btrfs_key key;
4097 struct btrfs_key found_key;
4098 struct btrfs_path *path;
4099 struct list_head ins_list;
4100 struct list_head del_list;
4103 struct extent_buffer *leaf;
4105 unsigned char d_type;
4110 int key_type = BTRFS_DIR_INDEX_KEY;
4114 int is_curr = 0; /* filp->f_pos points to the current index? */
4116 /* FIXME, use a real flag for deciding about the key type */
4117 if (root->fs_info->tree_root == root)
4118 key_type = BTRFS_DIR_ITEM_KEY;
4120 /* special case for "." */
4121 if (filp->f_pos == 0) {
4122 over = filldir(dirent, ".", 1,
4123 filp->f_pos, btrfs_ino(inode), DT_DIR);
4128 /* special case for .., just use the back ref */
4129 if (filp->f_pos == 1) {
4130 u64 pino = parent_ino(filp->f_path.dentry);
4131 over = filldir(dirent, "..", 2,
4132 filp->f_pos, pino, DT_DIR);
4137 path = btrfs_alloc_path();
4143 if (key_type == BTRFS_DIR_INDEX_KEY) {
4144 INIT_LIST_HEAD(&ins_list);
4145 INIT_LIST_HEAD(&del_list);
4146 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4149 btrfs_set_key_type(&key, key_type);
4150 key.offset = filp->f_pos;
4151 key.objectid = btrfs_ino(inode);
4153 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4158 leaf = path->nodes[0];
4159 slot = path->slots[0];
4160 if (slot >= btrfs_header_nritems(leaf)) {
4161 ret = btrfs_next_leaf(root, path);
4169 item = btrfs_item_nr(leaf, slot);
4170 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4172 if (found_key.objectid != key.objectid)
4174 if (btrfs_key_type(&found_key) != key_type)
4176 if (found_key.offset < filp->f_pos)
4178 if (key_type == BTRFS_DIR_INDEX_KEY &&
4179 btrfs_should_delete_dir_index(&del_list,
4183 filp->f_pos = found_key.offset;
4186 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4188 di_total = btrfs_item_size(leaf, item);
4190 while (di_cur < di_total) {
4191 struct btrfs_key location;
4194 if (verify_dir_item(root, leaf, di))
4197 name_len = btrfs_dir_name_len(leaf, di);
4198 if (name_len <= sizeof(tmp_name)) {
4199 name_ptr = tmp_name;
4201 name_ptr = kmalloc(name_len, GFP_NOFS);
4207 read_extent_buffer(leaf, name_ptr,
4208 (unsigned long)(di + 1), name_len);
4210 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4211 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4215 q.hash = full_name_hash(q.name, q.len);
4216 tmp = d_lookup(filp->f_dentry, &q);
4218 struct btrfs_key *newkey;
4220 newkey = kzalloc(sizeof(struct btrfs_key),
4224 tmp = d_alloc(filp->f_dentry, &q);
4230 memcpy(newkey, &location,
4231 sizeof(struct btrfs_key));
4232 tmp->d_fsdata = newkey;
4233 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4240 /* is this a reference to our own snapshot? If so
4243 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4244 location.objectid == root->root_key.objectid) {
4248 over = filldir(dirent, name_ptr, name_len,
4249 found_key.offset, location.objectid,
4253 if (name_ptr != tmp_name)
4258 di_len = btrfs_dir_name_len(leaf, di) +
4259 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4261 di = (struct btrfs_dir_item *)((char *)di + di_len);
4267 if (key_type == BTRFS_DIR_INDEX_KEY) {
4270 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4276 /* Reached end of directory/root. Bump pos past the last item. */
4277 if (key_type == BTRFS_DIR_INDEX_KEY)
4279 * 32-bit glibc will use getdents64, but then strtol -
4280 * so the last number we can serve is this.
4282 filp->f_pos = 0x7fffffff;
4288 if (key_type == BTRFS_DIR_INDEX_KEY)
4289 btrfs_put_delayed_items(&ins_list, &del_list);
4290 btrfs_free_path(path);
4294 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4296 struct btrfs_root *root = BTRFS_I(inode)->root;
4297 struct btrfs_trans_handle *trans;
4299 bool nolock = false;
4301 if (BTRFS_I(inode)->dummy_inode)
4304 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4307 if (wbc->sync_mode == WB_SYNC_ALL) {
4309 trans = btrfs_join_transaction_nolock(root);
4311 trans = btrfs_join_transaction(root);
4313 return PTR_ERR(trans);
4315 ret = btrfs_end_transaction_nolock(trans, root);
4317 ret = btrfs_commit_transaction(trans, root);
4323 * This is somewhat expensive, updating the tree every time the
4324 * inode changes. But, it is most likely to find the inode in cache.
4325 * FIXME, needs more benchmarking...there are no reasons other than performance
4326 * to keep or drop this code.
4328 void btrfs_dirty_inode(struct inode *inode, int flags)
4330 struct btrfs_root *root = BTRFS_I(inode)->root;
4331 struct btrfs_trans_handle *trans;
4334 if (BTRFS_I(inode)->dummy_inode)
4337 trans = btrfs_join_transaction(root);
4338 BUG_ON(IS_ERR(trans));
4340 ret = btrfs_update_inode(trans, root, inode);
4341 if (ret && ret == -ENOSPC) {
4342 /* whoops, lets try again with the full transaction */
4343 btrfs_end_transaction(trans, root);
4344 trans = btrfs_start_transaction(root, 1);
4345 if (IS_ERR(trans)) {
4346 printk_ratelimited(KERN_ERR "btrfs: fail to "
4347 "dirty inode %llu error %ld\n",
4348 (unsigned long long)btrfs_ino(inode),
4353 ret = btrfs_update_inode(trans, root, inode);
4355 printk_ratelimited(KERN_ERR "btrfs: fail to "
4356 "dirty inode %llu error %d\n",
4357 (unsigned long long)btrfs_ino(inode),
4361 btrfs_end_transaction(trans, root);
4362 if (BTRFS_I(inode)->delayed_node)
4363 btrfs_balance_delayed_items(root);
4367 * find the highest existing sequence number in a directory
4368 * and then set the in-memory index_cnt variable to reflect
4369 * free sequence numbers
4371 static int btrfs_set_inode_index_count(struct inode *inode)
4373 struct btrfs_root *root = BTRFS_I(inode)->root;
4374 struct btrfs_key key, found_key;
4375 struct btrfs_path *path;
4376 struct extent_buffer *leaf;
4379 key.objectid = btrfs_ino(inode);
4380 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4381 key.offset = (u64)-1;
4383 path = btrfs_alloc_path();
4387 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4390 /* FIXME: we should be able to handle this */
4396 * MAGIC NUMBER EXPLANATION:
4397 * since we search a directory based on f_pos we have to start at 2
4398 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4399 * else has to start at 2
4401 if (path->slots[0] == 0) {
4402 BTRFS_I(inode)->index_cnt = 2;
4408 leaf = path->nodes[0];
4409 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4411 if (found_key.objectid != btrfs_ino(inode) ||
4412 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4413 BTRFS_I(inode)->index_cnt = 2;
4417 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4419 btrfs_free_path(path);
4424 * helper to find a free sequence number in a given directory. This current
4425 * code is very simple, later versions will do smarter things in the btree
4427 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4431 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4432 ret = btrfs_inode_delayed_dir_index_count(dir);
4434 ret = btrfs_set_inode_index_count(dir);
4440 *index = BTRFS_I(dir)->index_cnt;
4441 BTRFS_I(dir)->index_cnt++;
4446 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4447 struct btrfs_root *root,
4449 const char *name, int name_len,
4450 u64 ref_objectid, u64 objectid, int mode,
4453 struct inode *inode;
4454 struct btrfs_inode_item *inode_item;
4455 struct btrfs_key *location;
4456 struct btrfs_path *path;
4457 struct btrfs_inode_ref *ref;
4458 struct btrfs_key key[2];
4464 path = btrfs_alloc_path();
4466 return ERR_PTR(-ENOMEM);
4468 inode = new_inode(root->fs_info->sb);
4470 btrfs_free_path(path);
4471 return ERR_PTR(-ENOMEM);
4475 * we have to initialize this early, so we can reclaim the inode
4476 * number if we fail afterwards in this function.
4478 inode->i_ino = objectid;
4481 trace_btrfs_inode_request(dir);
4483 ret = btrfs_set_inode_index(dir, index);
4485 btrfs_free_path(path);
4487 return ERR_PTR(ret);
4491 * index_cnt is ignored for everything but a dir,
4492 * btrfs_get_inode_index_count has an explanation for the magic
4495 BTRFS_I(inode)->index_cnt = 2;
4496 BTRFS_I(inode)->root = root;
4497 BTRFS_I(inode)->generation = trans->transid;
4498 inode->i_generation = BTRFS_I(inode)->generation;
4499 btrfs_set_inode_space_info(root, inode);
4506 key[0].objectid = objectid;
4507 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4510 key[1].objectid = objectid;
4511 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4512 key[1].offset = ref_objectid;
4514 sizes[0] = sizeof(struct btrfs_inode_item);
4515 sizes[1] = name_len + sizeof(*ref);
4517 path->leave_spinning = 1;
4518 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4522 inode_init_owner(inode, dir, mode);
4523 inode_set_bytes(inode, 0);
4524 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4525 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4526 struct btrfs_inode_item);
4527 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4529 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4530 struct btrfs_inode_ref);
4531 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4532 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4533 ptr = (unsigned long)(ref + 1);
4534 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4536 btrfs_mark_buffer_dirty(path->nodes[0]);
4537 btrfs_free_path(path);
4539 location = &BTRFS_I(inode)->location;
4540 location->objectid = objectid;
4541 location->offset = 0;
4542 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4544 btrfs_inherit_iflags(inode, dir);
4546 if (S_ISREG(mode)) {
4547 if (btrfs_test_opt(root, NODATASUM))
4548 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4549 if (btrfs_test_opt(root, NODATACOW) ||
4550 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4551 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4554 insert_inode_hash(inode);
4555 inode_tree_add(inode);
4557 trace_btrfs_inode_new(inode);
4558 btrfs_set_inode_last_trans(trans, inode);
4563 BTRFS_I(dir)->index_cnt--;
4564 btrfs_free_path(path);
4566 return ERR_PTR(ret);
4569 static inline u8 btrfs_inode_type(struct inode *inode)
4571 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4575 * utility function to add 'inode' into 'parent_inode' with
4576 * a give name and a given sequence number.
4577 * if 'add_backref' is true, also insert a backref from the
4578 * inode to the parent directory.
4580 int btrfs_add_link(struct btrfs_trans_handle *trans,
4581 struct inode *parent_inode, struct inode *inode,
4582 const char *name, int name_len, int add_backref, u64 index)
4585 struct btrfs_key key;
4586 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4587 u64 ino = btrfs_ino(inode);
4588 u64 parent_ino = btrfs_ino(parent_inode);
4590 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4591 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4594 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4598 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4599 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4600 key.objectid, root->root_key.objectid,
4601 parent_ino, index, name, name_len);
4602 } else if (add_backref) {
4603 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4608 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4610 btrfs_inode_type(inode), index);
4613 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4615 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4616 ret = btrfs_update_inode(trans, root, parent_inode);
4621 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4622 struct inode *dir, struct dentry *dentry,
4623 struct inode *inode, int backref, u64 index)
4625 int err = btrfs_add_link(trans, dir, inode,
4626 dentry->d_name.name, dentry->d_name.len,
4629 d_instantiate(dentry, inode);
4637 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4638 int mode, dev_t rdev)
4640 struct btrfs_trans_handle *trans;
4641 struct btrfs_root *root = BTRFS_I(dir)->root;
4642 struct inode *inode = NULL;
4646 unsigned long nr = 0;
4649 if (!new_valid_dev(rdev))
4653 * 2 for inode item and ref
4655 * 1 for xattr if selinux is on
4657 trans = btrfs_start_transaction(root, 5);
4659 return PTR_ERR(trans);
4661 err = btrfs_find_free_ino(root, &objectid);
4665 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4666 dentry->d_name.len, btrfs_ino(dir), objectid,
4668 if (IS_ERR(inode)) {
4669 err = PTR_ERR(inode);
4673 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4679 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4683 inode->i_op = &btrfs_special_inode_operations;
4684 init_special_inode(inode, inode->i_mode, rdev);
4685 btrfs_update_inode(trans, root, inode);
4688 nr = trans->blocks_used;
4689 btrfs_end_transaction_throttle(trans, root);
4690 btrfs_btree_balance_dirty(root, nr);
4692 inode_dec_link_count(inode);
4698 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4699 int mode, struct nameidata *nd)
4701 struct btrfs_trans_handle *trans;
4702 struct btrfs_root *root = BTRFS_I(dir)->root;
4703 struct inode *inode = NULL;
4706 unsigned long nr = 0;
4711 * 2 for inode item and ref
4713 * 1 for xattr if selinux is on
4715 trans = btrfs_start_transaction(root, 5);
4717 return PTR_ERR(trans);
4719 err = btrfs_find_free_ino(root, &objectid);
4723 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4724 dentry->d_name.len, btrfs_ino(dir), objectid,
4726 if (IS_ERR(inode)) {
4727 err = PTR_ERR(inode);
4731 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4737 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4741 inode->i_mapping->a_ops = &btrfs_aops;
4742 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4743 inode->i_fop = &btrfs_file_operations;
4744 inode->i_op = &btrfs_file_inode_operations;
4745 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4748 nr = trans->blocks_used;
4749 btrfs_end_transaction_throttle(trans, root);
4751 inode_dec_link_count(inode);
4754 btrfs_btree_balance_dirty(root, nr);
4758 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4759 struct dentry *dentry)
4761 struct btrfs_trans_handle *trans;
4762 struct btrfs_root *root = BTRFS_I(dir)->root;
4763 struct inode *inode = old_dentry->d_inode;
4765 unsigned long nr = 0;
4769 /* do not allow sys_link's with other subvols of the same device */
4770 if (root->objectid != BTRFS_I(inode)->root->objectid)
4773 if (inode->i_nlink == ~0U)
4776 err = btrfs_set_inode_index(dir, &index);
4781 * 2 items for inode and inode ref
4782 * 2 items for dir items
4783 * 1 item for parent inode
4785 trans = btrfs_start_transaction(root, 5);
4786 if (IS_ERR(trans)) {
4787 err = PTR_ERR(trans);
4791 btrfs_inc_nlink(inode);
4792 inode->i_ctime = CURRENT_TIME;
4795 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4800 struct dentry *parent = dentry->d_parent;
4801 err = btrfs_update_inode(trans, root, inode);
4803 btrfs_log_new_name(trans, inode, NULL, parent);
4806 nr = trans->blocks_used;
4807 btrfs_end_transaction_throttle(trans, root);
4810 inode_dec_link_count(inode);
4813 btrfs_btree_balance_dirty(root, nr);
4817 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4819 struct inode *inode = NULL;
4820 struct btrfs_trans_handle *trans;
4821 struct btrfs_root *root = BTRFS_I(dir)->root;
4823 int drop_on_err = 0;
4826 unsigned long nr = 1;
4829 * 2 items for inode and ref
4830 * 2 items for dir items
4831 * 1 for xattr if selinux is on
4833 trans = btrfs_start_transaction(root, 5);
4835 return PTR_ERR(trans);
4837 err = btrfs_find_free_ino(root, &objectid);
4841 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4842 dentry->d_name.len, btrfs_ino(dir), objectid,
4843 S_IFDIR | mode, &index);
4844 if (IS_ERR(inode)) {
4845 err = PTR_ERR(inode);
4851 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4855 inode->i_op = &btrfs_dir_inode_operations;
4856 inode->i_fop = &btrfs_dir_file_operations;
4858 btrfs_i_size_write(inode, 0);
4859 err = btrfs_update_inode(trans, root, inode);
4863 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4864 dentry->d_name.len, 0, index);
4868 d_instantiate(dentry, inode);
4872 nr = trans->blocks_used;
4873 btrfs_end_transaction_throttle(trans, root);
4876 btrfs_btree_balance_dirty(root, nr);
4880 /* helper for btfs_get_extent. Given an existing extent in the tree,
4881 * and an extent that you want to insert, deal with overlap and insert
4882 * the new extent into the tree.
4884 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4885 struct extent_map *existing,
4886 struct extent_map *em,
4887 u64 map_start, u64 map_len)
4891 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4892 start_diff = map_start - em->start;
4893 em->start = map_start;
4895 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4896 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4897 em->block_start += start_diff;
4898 em->block_len -= start_diff;
4900 return add_extent_mapping(em_tree, em);
4903 static noinline int uncompress_inline(struct btrfs_path *path,
4904 struct inode *inode, struct page *page,
4905 size_t pg_offset, u64 extent_offset,
4906 struct btrfs_file_extent_item *item)
4909 struct extent_buffer *leaf = path->nodes[0];
4912 unsigned long inline_size;
4916 WARN_ON(pg_offset != 0);
4917 compress_type = btrfs_file_extent_compression(leaf, item);
4918 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4919 inline_size = btrfs_file_extent_inline_item_len(leaf,
4920 btrfs_item_nr(leaf, path->slots[0]));
4921 tmp = kmalloc(inline_size, GFP_NOFS);
4924 ptr = btrfs_file_extent_inline_start(item);
4926 read_extent_buffer(leaf, tmp, ptr, inline_size);
4928 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4929 ret = btrfs_decompress(compress_type, tmp, page,
4930 extent_offset, inline_size, max_size);
4932 char *kaddr = kmap_atomic(page, KM_USER0);
4933 unsigned long copy_size = min_t(u64,
4934 PAGE_CACHE_SIZE - pg_offset,
4935 max_size - extent_offset);
4936 memset(kaddr + pg_offset, 0, copy_size);
4937 kunmap_atomic(kaddr, KM_USER0);
4944 * a bit scary, this does extent mapping from logical file offset to the disk.
4945 * the ugly parts come from merging extents from the disk with the in-ram
4946 * representation. This gets more complex because of the data=ordered code,
4947 * where the in-ram extents might be locked pending data=ordered completion.
4949 * This also copies inline extents directly into the page.
4952 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4953 size_t pg_offset, u64 start, u64 len,
4959 u64 extent_start = 0;
4961 u64 objectid = btrfs_ino(inode);
4963 struct btrfs_path *path = NULL;
4964 struct btrfs_root *root = BTRFS_I(inode)->root;
4965 struct btrfs_file_extent_item *item;
4966 struct extent_buffer *leaf;
4967 struct btrfs_key found_key;
4968 struct extent_map *em = NULL;
4969 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4970 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4971 struct btrfs_trans_handle *trans = NULL;
4975 read_lock(&em_tree->lock);
4976 em = lookup_extent_mapping(em_tree, start, len);
4978 em->bdev = root->fs_info->fs_devices->latest_bdev;
4979 read_unlock(&em_tree->lock);
4982 if (em->start > start || em->start + em->len <= start)
4983 free_extent_map(em);
4984 else if (em->block_start == EXTENT_MAP_INLINE && page)
4985 free_extent_map(em);
4989 em = alloc_extent_map();
4994 em->bdev = root->fs_info->fs_devices->latest_bdev;
4995 em->start = EXTENT_MAP_HOLE;
4996 em->orig_start = EXTENT_MAP_HOLE;
4998 em->block_len = (u64)-1;
5001 path = btrfs_alloc_path();
5007 * Chances are we'll be called again, so go ahead and do
5013 ret = btrfs_lookup_file_extent(trans, root, path,
5014 objectid, start, trans != NULL);
5021 if (path->slots[0] == 0)
5026 leaf = path->nodes[0];
5027 item = btrfs_item_ptr(leaf, path->slots[0],
5028 struct btrfs_file_extent_item);
5029 /* are we inside the extent that was found? */
5030 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5031 found_type = btrfs_key_type(&found_key);
5032 if (found_key.objectid != objectid ||
5033 found_type != BTRFS_EXTENT_DATA_KEY) {
5037 found_type = btrfs_file_extent_type(leaf, item);
5038 extent_start = found_key.offset;
5039 compress_type = btrfs_file_extent_compression(leaf, item);
5040 if (found_type == BTRFS_FILE_EXTENT_REG ||
5041 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5042 extent_end = extent_start +
5043 btrfs_file_extent_num_bytes(leaf, item);
5044 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5046 size = btrfs_file_extent_inline_len(leaf, item);
5047 extent_end = (extent_start + size + root->sectorsize - 1) &
5048 ~((u64)root->sectorsize - 1);
5051 if (start >= extent_end) {
5053 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5054 ret = btrfs_next_leaf(root, path);
5061 leaf = path->nodes[0];
5063 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5064 if (found_key.objectid != objectid ||
5065 found_key.type != BTRFS_EXTENT_DATA_KEY)
5067 if (start + len <= found_key.offset)
5070 em->len = found_key.offset - start;
5074 if (found_type == BTRFS_FILE_EXTENT_REG ||
5075 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5076 em->start = extent_start;
5077 em->len = extent_end - extent_start;
5078 em->orig_start = extent_start -
5079 btrfs_file_extent_offset(leaf, item);
5080 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5082 em->block_start = EXTENT_MAP_HOLE;
5085 if (compress_type != BTRFS_COMPRESS_NONE) {
5086 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5087 em->compress_type = compress_type;
5088 em->block_start = bytenr;
5089 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5092 bytenr += btrfs_file_extent_offset(leaf, item);
5093 em->block_start = bytenr;
5094 em->block_len = em->len;
5095 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5096 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5099 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5103 size_t extent_offset;
5106 em->block_start = EXTENT_MAP_INLINE;
5107 if (!page || create) {
5108 em->start = extent_start;
5109 em->len = extent_end - extent_start;
5113 size = btrfs_file_extent_inline_len(leaf, item);
5114 extent_offset = page_offset(page) + pg_offset - extent_start;
5115 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5116 size - extent_offset);
5117 em->start = extent_start + extent_offset;
5118 em->len = (copy_size + root->sectorsize - 1) &
5119 ~((u64)root->sectorsize - 1);
5120 em->orig_start = EXTENT_MAP_INLINE;
5121 if (compress_type) {
5122 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5123 em->compress_type = compress_type;
5125 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5126 if (create == 0 && !PageUptodate(page)) {
5127 if (btrfs_file_extent_compression(leaf, item) !=
5128 BTRFS_COMPRESS_NONE) {
5129 ret = uncompress_inline(path, inode, page,
5131 extent_offset, item);
5135 read_extent_buffer(leaf, map + pg_offset, ptr,
5137 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5138 memset(map + pg_offset + copy_size, 0,
5139 PAGE_CACHE_SIZE - pg_offset -
5144 flush_dcache_page(page);
5145 } else if (create && PageUptodate(page)) {
5149 free_extent_map(em);
5152 btrfs_release_path(path);
5153 trans = btrfs_join_transaction(root);
5156 return ERR_CAST(trans);
5160 write_extent_buffer(leaf, map + pg_offset, ptr,
5163 btrfs_mark_buffer_dirty(leaf);
5165 set_extent_uptodate(io_tree, em->start,
5166 extent_map_end(em) - 1, NULL, GFP_NOFS);
5169 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5176 em->block_start = EXTENT_MAP_HOLE;
5177 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5179 btrfs_release_path(path);
5180 if (em->start > start || extent_map_end(em) <= start) {
5181 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5182 "[%llu %llu]\n", (unsigned long long)em->start,
5183 (unsigned long long)em->len,
5184 (unsigned long long)start,
5185 (unsigned long long)len);
5191 write_lock(&em_tree->lock);
5192 ret = add_extent_mapping(em_tree, em);
5193 /* it is possible that someone inserted the extent into the tree
5194 * while we had the lock dropped. It is also possible that
5195 * an overlapping map exists in the tree
5197 if (ret == -EEXIST) {
5198 struct extent_map *existing;
5202 existing = lookup_extent_mapping(em_tree, start, len);
5203 if (existing && (existing->start > start ||
5204 existing->start + existing->len <= start)) {
5205 free_extent_map(existing);
5209 existing = lookup_extent_mapping(em_tree, em->start,
5212 err = merge_extent_mapping(em_tree, existing,
5215 free_extent_map(existing);
5217 free_extent_map(em);
5222 free_extent_map(em);
5226 free_extent_map(em);
5231 write_unlock(&em_tree->lock);
5234 trace_btrfs_get_extent(root, em);
5237 btrfs_free_path(path);
5239 ret = btrfs_end_transaction(trans, root);
5244 free_extent_map(em);
5245 return ERR_PTR(err);
5250 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5251 size_t pg_offset, u64 start, u64 len,
5254 struct extent_map *em;
5255 struct extent_map *hole_em = NULL;
5256 u64 range_start = start;
5262 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5267 * if our em maps to a hole, there might
5268 * actually be delalloc bytes behind it
5270 if (em->block_start != EXTENT_MAP_HOLE)
5276 /* check to see if we've wrapped (len == -1 or similar) */
5285 /* ok, we didn't find anything, lets look for delalloc */
5286 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5287 end, len, EXTENT_DELALLOC, 1);
5288 found_end = range_start + found;
5289 if (found_end < range_start)
5290 found_end = (u64)-1;
5293 * we didn't find anything useful, return
5294 * the original results from get_extent()
5296 if (range_start > end || found_end <= start) {
5302 /* adjust the range_start to make sure it doesn't
5303 * go backwards from the start they passed in
5305 range_start = max(start,range_start);
5306 found = found_end - range_start;
5309 u64 hole_start = start;
5312 em = alloc_extent_map();
5318 * when btrfs_get_extent can't find anything it
5319 * returns one huge hole
5321 * make sure what it found really fits our range, and
5322 * adjust to make sure it is based on the start from
5326 u64 calc_end = extent_map_end(hole_em);
5328 if (calc_end <= start || (hole_em->start > end)) {
5329 free_extent_map(hole_em);
5332 hole_start = max(hole_em->start, start);
5333 hole_len = calc_end - hole_start;
5337 if (hole_em && range_start > hole_start) {
5338 /* our hole starts before our delalloc, so we
5339 * have to return just the parts of the hole
5340 * that go until the delalloc starts
5342 em->len = min(hole_len,
5343 range_start - hole_start);
5344 em->start = hole_start;
5345 em->orig_start = hole_start;
5347 * don't adjust block start at all,
5348 * it is fixed at EXTENT_MAP_HOLE
5350 em->block_start = hole_em->block_start;
5351 em->block_len = hole_len;
5353 em->start = range_start;
5355 em->orig_start = range_start;
5356 em->block_start = EXTENT_MAP_DELALLOC;
5357 em->block_len = found;
5359 } else if (hole_em) {
5364 free_extent_map(hole_em);
5366 free_extent_map(em);
5367 return ERR_PTR(err);
5372 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5373 struct extent_map *em,
5376 struct btrfs_root *root = BTRFS_I(inode)->root;
5377 struct btrfs_trans_handle *trans;
5378 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5379 struct btrfs_key ins;
5382 bool insert = false;
5385 * Ok if the extent map we looked up is a hole and is for the exact
5386 * range we want, there is no reason to allocate a new one, however if
5387 * it is not right then we need to free this one and drop the cache for
5390 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5392 free_extent_map(em);
5395 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5398 trans = btrfs_join_transaction(root);
5400 return ERR_CAST(trans);
5402 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5403 btrfs_add_inode_defrag(trans, inode);
5405 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5407 alloc_hint = get_extent_allocation_hint(inode, start, len);
5408 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5409 alloc_hint, (u64)-1, &ins, 1);
5416 em = alloc_extent_map();
5418 em = ERR_PTR(-ENOMEM);
5424 em->orig_start = em->start;
5425 em->len = ins.offset;
5427 em->block_start = ins.objectid;
5428 em->block_len = ins.offset;
5429 em->bdev = root->fs_info->fs_devices->latest_bdev;
5432 * We need to do this because if we're using the original em we searched
5433 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5436 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5439 write_lock(&em_tree->lock);
5440 ret = add_extent_mapping(em_tree, em);
5441 write_unlock(&em_tree->lock);
5444 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5447 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5448 ins.offset, ins.offset, 0);
5450 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5454 btrfs_end_transaction(trans, root);
5459 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5460 * block must be cow'd
5462 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5463 struct inode *inode, u64 offset, u64 len)
5465 struct btrfs_path *path;
5467 struct extent_buffer *leaf;
5468 struct btrfs_root *root = BTRFS_I(inode)->root;
5469 struct btrfs_file_extent_item *fi;
5470 struct btrfs_key key;
5478 path = btrfs_alloc_path();
5482 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5487 slot = path->slots[0];
5490 /* can't find the item, must cow */
5497 leaf = path->nodes[0];
5498 btrfs_item_key_to_cpu(leaf, &key, slot);
5499 if (key.objectid != btrfs_ino(inode) ||
5500 key.type != BTRFS_EXTENT_DATA_KEY) {
5501 /* not our file or wrong item type, must cow */
5505 if (key.offset > offset) {
5506 /* Wrong offset, must cow */
5510 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5511 found_type = btrfs_file_extent_type(leaf, fi);
5512 if (found_type != BTRFS_FILE_EXTENT_REG &&
5513 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5514 /* not a regular extent, must cow */
5517 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5518 backref_offset = btrfs_file_extent_offset(leaf, fi);
5520 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5521 if (extent_end < offset + len) {
5522 /* extent doesn't include our full range, must cow */
5526 if (btrfs_extent_readonly(root, disk_bytenr))
5530 * look for other files referencing this extent, if we
5531 * find any we must cow
5533 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5534 key.offset - backref_offset, disk_bytenr))
5538 * adjust disk_bytenr and num_bytes to cover just the bytes
5539 * in this extent we are about to write. If there
5540 * are any csums in that range we have to cow in order
5541 * to keep the csums correct
5543 disk_bytenr += backref_offset;
5544 disk_bytenr += offset - key.offset;
5545 num_bytes = min(offset + len, extent_end) - offset;
5546 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5549 * all of the above have passed, it is safe to overwrite this extent
5554 btrfs_free_path(path);
5558 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5559 struct buffer_head *bh_result, int create)
5561 struct extent_map *em;
5562 struct btrfs_root *root = BTRFS_I(inode)->root;
5563 u64 start = iblock << inode->i_blkbits;
5564 u64 len = bh_result->b_size;
5565 struct btrfs_trans_handle *trans;
5567 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5572 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5573 * io. INLINE is special, and we could probably kludge it in here, but
5574 * it's still buffered so for safety lets just fall back to the generic
5577 * For COMPRESSED we _have_ to read the entire extent in so we can
5578 * decompress it, so there will be buffering required no matter what we
5579 * do, so go ahead and fallback to buffered.
5581 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5582 * to buffered IO. Don't blame me, this is the price we pay for using
5585 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5586 em->block_start == EXTENT_MAP_INLINE) {
5587 free_extent_map(em);
5591 /* Just a good old fashioned hole, return */
5592 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5593 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5594 free_extent_map(em);
5595 /* DIO will do one hole at a time, so just unlock a sector */
5596 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5597 start + root->sectorsize - 1, GFP_NOFS);
5602 * We don't allocate a new extent in the following cases
5604 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5606 * 2) The extent is marked as PREALLOC. We're good to go here and can
5607 * just use the extent.
5611 len = em->len - (start - em->start);
5615 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5616 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5617 em->block_start != EXTENT_MAP_HOLE)) {
5622 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5623 type = BTRFS_ORDERED_PREALLOC;
5625 type = BTRFS_ORDERED_NOCOW;
5626 len = min(len, em->len - (start - em->start));
5627 block_start = em->block_start + (start - em->start);
5630 * we're not going to log anything, but we do need
5631 * to make sure the current transaction stays open
5632 * while we look for nocow cross refs
5634 trans = btrfs_join_transaction(root);
5638 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5639 ret = btrfs_add_ordered_extent_dio(inode, start,
5640 block_start, len, len, type);
5641 btrfs_end_transaction(trans, root);
5643 free_extent_map(em);
5648 btrfs_end_transaction(trans, root);
5652 * this will cow the extent, reset the len in case we changed
5655 len = bh_result->b_size;
5656 em = btrfs_new_extent_direct(inode, em, start, len);
5659 len = min(len, em->len - (start - em->start));
5661 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5662 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5665 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5667 bh_result->b_size = len;
5668 bh_result->b_bdev = em->bdev;
5669 set_buffer_mapped(bh_result);
5670 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5671 set_buffer_new(bh_result);
5673 free_extent_map(em);
5678 struct btrfs_dio_private {
5679 struct inode *inode;
5686 /* number of bios pending for this dio */
5687 atomic_t pending_bios;
5692 struct bio *orig_bio;
5695 static void btrfs_endio_direct_read(struct bio *bio, int err)
5697 struct btrfs_dio_private *dip = bio->bi_private;
5698 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5699 struct bio_vec *bvec = bio->bi_io_vec;
5700 struct inode *inode = dip->inode;
5701 struct btrfs_root *root = BTRFS_I(inode)->root;
5703 u32 *private = dip->csums;
5705 start = dip->logical_offset;
5707 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5708 struct page *page = bvec->bv_page;
5711 unsigned long flags;
5713 local_irq_save(flags);
5714 kaddr = kmap_atomic(page, KM_IRQ0);
5715 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5716 csum, bvec->bv_len);
5717 btrfs_csum_final(csum, (char *)&csum);
5718 kunmap_atomic(kaddr, KM_IRQ0);
5719 local_irq_restore(flags);
5721 flush_dcache_page(bvec->bv_page);
5722 if (csum != *private) {
5723 printk(KERN_ERR "btrfs csum failed ino %llu off"
5724 " %llu csum %u private %u\n",
5725 (unsigned long long)btrfs_ino(inode),
5726 (unsigned long long)start,
5732 start += bvec->bv_len;
5735 } while (bvec <= bvec_end);
5737 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5738 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5739 bio->bi_private = dip->private;
5744 /* If we had a csum failure make sure to clear the uptodate flag */
5746 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5747 dio_end_io(bio, err);
5750 static void btrfs_endio_direct_write(struct bio *bio, int err)
5752 struct btrfs_dio_private *dip = bio->bi_private;
5753 struct inode *inode = dip->inode;
5754 struct btrfs_root *root = BTRFS_I(inode)->root;
5755 struct btrfs_trans_handle *trans;
5756 struct btrfs_ordered_extent *ordered = NULL;
5757 struct extent_state *cached_state = NULL;
5758 u64 ordered_offset = dip->logical_offset;
5759 u64 ordered_bytes = dip->bytes;
5765 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5773 trans = btrfs_join_transaction(root);
5774 if (IS_ERR(trans)) {
5778 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5780 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5781 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5783 err = btrfs_update_inode(trans, root, inode);
5787 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5788 ordered->file_offset + ordered->len - 1, 0,
5789 &cached_state, GFP_NOFS);
5791 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5792 ret = btrfs_mark_extent_written(trans, inode,
5793 ordered->file_offset,
5794 ordered->file_offset +
5801 ret = insert_reserved_file_extent(trans, inode,
5802 ordered->file_offset,
5808 BTRFS_FILE_EXTENT_REG);
5809 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5810 ordered->file_offset, ordered->len);
5818 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5819 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5820 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5821 btrfs_update_inode(trans, root, inode);
5824 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5825 ordered->file_offset + ordered->len - 1,
5826 &cached_state, GFP_NOFS);
5828 btrfs_delalloc_release_metadata(inode, ordered->len);
5829 btrfs_end_transaction(trans, root);
5830 ordered_offset = ordered->file_offset + ordered->len;
5831 btrfs_put_ordered_extent(ordered);
5832 btrfs_put_ordered_extent(ordered);
5836 * our bio might span multiple ordered extents. If we haven't
5837 * completed the accounting for the whole dio, go back and try again
5839 if (ordered_offset < dip->logical_offset + dip->bytes) {
5840 ordered_bytes = dip->logical_offset + dip->bytes -
5845 bio->bi_private = dip->private;
5850 /* If we had an error make sure to clear the uptodate flag */
5852 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5853 dio_end_io(bio, err);
5856 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5857 struct bio *bio, int mirror_num,
5858 unsigned long bio_flags, u64 offset)
5861 struct btrfs_root *root = BTRFS_I(inode)->root;
5862 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5867 static void btrfs_end_dio_bio(struct bio *bio, int err)
5869 struct btrfs_dio_private *dip = bio->bi_private;
5872 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5873 "sector %#Lx len %u err no %d\n",
5874 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5875 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5879 * before atomic variable goto zero, we must make sure
5880 * dip->errors is perceived to be set.
5882 smp_mb__before_atomic_dec();
5885 /* if there are more bios still pending for this dio, just exit */
5886 if (!atomic_dec_and_test(&dip->pending_bios))
5890 bio_io_error(dip->orig_bio);
5892 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5893 bio_endio(dip->orig_bio, 0);
5899 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5900 u64 first_sector, gfp_t gfp_flags)
5902 int nr_vecs = bio_get_nr_vecs(bdev);
5903 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5906 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5907 int rw, u64 file_offset, int skip_sum,
5908 u32 *csums, int async_submit)
5910 int write = rw & REQ_WRITE;
5911 struct btrfs_root *root = BTRFS_I(inode)->root;
5915 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5922 if (write && async_submit) {
5923 ret = btrfs_wq_submit_bio(root->fs_info,
5924 inode, rw, bio, 0, 0,
5926 __btrfs_submit_bio_start_direct_io,
5927 __btrfs_submit_bio_done);
5931 * If we aren't doing async submit, calculate the csum of the
5934 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5937 } else if (!skip_sum) {
5938 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5939 file_offset, csums);
5945 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5951 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5954 struct inode *inode = dip->inode;
5955 struct btrfs_root *root = BTRFS_I(inode)->root;
5956 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5958 struct bio *orig_bio = dip->orig_bio;
5959 struct bio_vec *bvec = orig_bio->bi_io_vec;
5960 u64 start_sector = orig_bio->bi_sector;
5961 u64 file_offset = dip->logical_offset;
5965 u32 *csums = dip->csums;
5967 int async_submit = 0;
5968 int write = rw & REQ_WRITE;
5970 map_length = orig_bio->bi_size;
5971 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5972 &map_length, NULL, 0);
5978 if (map_length >= orig_bio->bi_size) {
5984 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5987 bio->bi_private = dip;
5988 bio->bi_end_io = btrfs_end_dio_bio;
5989 atomic_inc(&dip->pending_bios);
5991 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5992 if (unlikely(map_length < submit_len + bvec->bv_len ||
5993 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5994 bvec->bv_offset) < bvec->bv_len)) {
5996 * inc the count before we submit the bio so
5997 * we know the end IO handler won't happen before
5998 * we inc the count. Otherwise, the dip might get freed
5999 * before we're done setting it up
6001 atomic_inc(&dip->pending_bios);
6002 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6003 file_offset, skip_sum,
6004 csums, async_submit);
6007 atomic_dec(&dip->pending_bios);
6011 /* Write's use the ordered csums */
6012 if (!write && !skip_sum)
6013 csums = csums + nr_pages;
6014 start_sector += submit_len >> 9;
6015 file_offset += submit_len;
6020 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6021 start_sector, GFP_NOFS);
6024 bio->bi_private = dip;
6025 bio->bi_end_io = btrfs_end_dio_bio;
6027 map_length = orig_bio->bi_size;
6028 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6029 &map_length, NULL, 0);
6035 submit_len += bvec->bv_len;
6042 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6043 csums, async_submit);
6051 * before atomic variable goto zero, we must
6052 * make sure dip->errors is perceived to be set.
6054 smp_mb__before_atomic_dec();
6055 if (atomic_dec_and_test(&dip->pending_bios))
6056 bio_io_error(dip->orig_bio);
6058 /* bio_end_io() will handle error, so we needn't return it */
6062 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6065 struct btrfs_root *root = BTRFS_I(inode)->root;
6066 struct btrfs_dio_private *dip;
6067 struct bio_vec *bvec = bio->bi_io_vec;
6069 int write = rw & REQ_WRITE;
6072 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6074 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6081 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6082 if (!write && !skip_sum) {
6083 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6091 dip->private = bio->bi_private;
6093 dip->logical_offset = file_offset;
6097 dip->bytes += bvec->bv_len;
6099 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6101 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6102 bio->bi_private = dip;
6104 dip->orig_bio = bio;
6105 atomic_set(&dip->pending_bios, 0);
6108 bio->bi_end_io = btrfs_endio_direct_write;
6110 bio->bi_end_io = btrfs_endio_direct_read;
6112 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6117 * If this is a write, we need to clean up the reserved space and kill
6118 * the ordered extent.
6121 struct btrfs_ordered_extent *ordered;
6122 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6123 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6124 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6125 btrfs_free_reserved_extent(root, ordered->start,
6127 btrfs_put_ordered_extent(ordered);
6128 btrfs_put_ordered_extent(ordered);
6130 bio_endio(bio, ret);
6133 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6134 const struct iovec *iov, loff_t offset,
6135 unsigned long nr_segs)
6141 unsigned blocksize_mask = root->sectorsize - 1;
6142 ssize_t retval = -EINVAL;
6143 loff_t end = offset;
6145 if (offset & blocksize_mask)
6148 /* Check the memory alignment. Blocks cannot straddle pages */
6149 for (seg = 0; seg < nr_segs; seg++) {
6150 addr = (unsigned long)iov[seg].iov_base;
6151 size = iov[seg].iov_len;
6153 if ((addr & blocksize_mask) || (size & blocksize_mask))
6156 /* If this is a write we don't need to check anymore */
6161 * Check to make sure we don't have duplicate iov_base's in this
6162 * iovec, if so return EINVAL, otherwise we'll get csum errors
6163 * when reading back.
6165 for (i = seg + 1; i < nr_segs; i++) {
6166 if (iov[seg].iov_base == iov[i].iov_base)
6174 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6175 const struct iovec *iov, loff_t offset,
6176 unsigned long nr_segs)
6178 struct file *file = iocb->ki_filp;
6179 struct inode *inode = file->f_mapping->host;
6180 struct btrfs_ordered_extent *ordered;
6181 struct extent_state *cached_state = NULL;
6182 u64 lockstart, lockend;
6184 int writing = rw & WRITE;
6186 size_t count = iov_length(iov, nr_segs);
6188 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6194 lockend = offset + count - 1;
6197 ret = btrfs_delalloc_reserve_space(inode, count);
6203 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6204 0, &cached_state, GFP_NOFS);
6206 * We're concerned with the entire range that we're going to be
6207 * doing DIO to, so we need to make sure theres no ordered
6208 * extents in this range.
6210 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6211 lockend - lockstart + 1);
6214 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6215 &cached_state, GFP_NOFS);
6216 btrfs_start_ordered_extent(inode, ordered, 1);
6217 btrfs_put_ordered_extent(ordered);
6222 * we don't use btrfs_set_extent_delalloc because we don't want
6223 * the dirty or uptodate bits
6226 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6227 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6228 EXTENT_DELALLOC, 0, NULL, &cached_state,
6231 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6232 lockend, EXTENT_LOCKED | write_bits,
6233 1, 0, &cached_state, GFP_NOFS);
6238 free_extent_state(cached_state);
6239 cached_state = NULL;
6241 ret = __blockdev_direct_IO(rw, iocb, inode,
6242 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6243 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6244 btrfs_submit_direct, 0);
6246 if (ret < 0 && ret != -EIOCBQUEUED) {
6247 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6248 offset + iov_length(iov, nr_segs) - 1,
6249 EXTENT_LOCKED | write_bits, 1, 0,
6250 &cached_state, GFP_NOFS);
6251 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6253 * We're falling back to buffered, unlock the section we didn't
6256 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6257 offset + iov_length(iov, nr_segs) - 1,
6258 EXTENT_LOCKED | write_bits, 1, 0,
6259 &cached_state, GFP_NOFS);
6262 free_extent_state(cached_state);
6266 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6267 __u64 start, __u64 len)
6269 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6272 int btrfs_readpage(struct file *file, struct page *page)
6274 struct extent_io_tree *tree;
6275 tree = &BTRFS_I(page->mapping->host)->io_tree;
6276 return extent_read_full_page(tree, page, btrfs_get_extent);
6279 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6281 struct extent_io_tree *tree;
6284 if (current->flags & PF_MEMALLOC) {
6285 redirty_page_for_writepage(wbc, page);
6289 tree = &BTRFS_I(page->mapping->host)->io_tree;
6290 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6293 int btrfs_writepages(struct address_space *mapping,
6294 struct writeback_control *wbc)
6296 struct extent_io_tree *tree;
6298 tree = &BTRFS_I(mapping->host)->io_tree;
6299 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6303 btrfs_readpages(struct file *file, struct address_space *mapping,
6304 struct list_head *pages, unsigned nr_pages)
6306 struct extent_io_tree *tree;
6307 tree = &BTRFS_I(mapping->host)->io_tree;
6308 return extent_readpages(tree, mapping, pages, nr_pages,
6311 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6313 struct extent_io_tree *tree;
6314 struct extent_map_tree *map;
6317 tree = &BTRFS_I(page->mapping->host)->io_tree;
6318 map = &BTRFS_I(page->mapping->host)->extent_tree;
6319 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6321 ClearPagePrivate(page);
6322 set_page_private(page, 0);
6323 page_cache_release(page);
6328 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6330 if (PageWriteback(page) || PageDirty(page))
6332 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6335 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6337 struct extent_io_tree *tree;
6338 struct btrfs_ordered_extent *ordered;
6339 struct extent_state *cached_state = NULL;
6340 u64 page_start = page_offset(page);
6341 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6345 * we have the page locked, so new writeback can't start,
6346 * and the dirty bit won't be cleared while we are here.
6348 * Wait for IO on this page so that we can safely clear
6349 * the PagePrivate2 bit and do ordered accounting
6351 wait_on_page_writeback(page);
6353 tree = &BTRFS_I(page->mapping->host)->io_tree;
6355 btrfs_releasepage(page, GFP_NOFS);
6358 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6360 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6364 * IO on this page will never be started, so we need
6365 * to account for any ordered extents now
6367 clear_extent_bit(tree, page_start, page_end,
6368 EXTENT_DIRTY | EXTENT_DELALLOC |
6369 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6370 &cached_state, GFP_NOFS);
6372 * whoever cleared the private bit is responsible
6373 * for the finish_ordered_io
6375 if (TestClearPagePrivate2(page)) {
6376 btrfs_finish_ordered_io(page->mapping->host,
6377 page_start, page_end);
6379 btrfs_put_ordered_extent(ordered);
6380 cached_state = NULL;
6381 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6384 clear_extent_bit(tree, page_start, page_end,
6385 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6386 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6387 __btrfs_releasepage(page, GFP_NOFS);
6389 ClearPageChecked(page);
6390 if (PagePrivate(page)) {
6391 ClearPagePrivate(page);
6392 set_page_private(page, 0);
6393 page_cache_release(page);
6398 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6399 * called from a page fault handler when a page is first dirtied. Hence we must
6400 * be careful to check for EOF conditions here. We set the page up correctly
6401 * for a written page which means we get ENOSPC checking when writing into
6402 * holes and correct delalloc and unwritten extent mapping on filesystems that
6403 * support these features.
6405 * We are not allowed to take the i_mutex here so we have to play games to
6406 * protect against truncate races as the page could now be beyond EOF. Because
6407 * vmtruncate() writes the inode size before removing pages, once we have the
6408 * page lock we can determine safely if the page is beyond EOF. If it is not
6409 * beyond EOF, then the page is guaranteed safe against truncation until we
6412 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6414 struct page *page = vmf->page;
6415 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6416 struct btrfs_root *root = BTRFS_I(inode)->root;
6417 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6418 struct btrfs_ordered_extent *ordered;
6419 struct extent_state *cached_state = NULL;
6421 unsigned long zero_start;
6427 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6431 else /* -ENOSPC, -EIO, etc */
6432 ret = VM_FAULT_SIGBUS;
6436 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6439 size = i_size_read(inode);
6440 page_start = page_offset(page);
6441 page_end = page_start + PAGE_CACHE_SIZE - 1;
6443 if ((page->mapping != inode->i_mapping) ||
6444 (page_start >= size)) {
6445 /* page got truncated out from underneath us */
6448 wait_on_page_writeback(page);
6450 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6452 set_page_extent_mapped(page);
6455 * we can't set the delalloc bits if there are pending ordered
6456 * extents. Drop our locks and wait for them to finish
6458 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6460 unlock_extent_cached(io_tree, page_start, page_end,
6461 &cached_state, GFP_NOFS);
6463 btrfs_start_ordered_extent(inode, ordered, 1);
6464 btrfs_put_ordered_extent(ordered);
6469 * XXX - page_mkwrite gets called every time the page is dirtied, even
6470 * if it was already dirty, so for space accounting reasons we need to
6471 * clear any delalloc bits for the range we are fixing to save. There
6472 * is probably a better way to do this, but for now keep consistent with
6473 * prepare_pages in the normal write path.
6475 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6476 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6477 0, 0, &cached_state, GFP_NOFS);
6479 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6482 unlock_extent_cached(io_tree, page_start, page_end,
6483 &cached_state, GFP_NOFS);
6484 ret = VM_FAULT_SIGBUS;
6489 /* page is wholly or partially inside EOF */
6490 if (page_start + PAGE_CACHE_SIZE > size)
6491 zero_start = size & ~PAGE_CACHE_MASK;
6493 zero_start = PAGE_CACHE_SIZE;
6495 if (zero_start != PAGE_CACHE_SIZE) {
6497 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6498 flush_dcache_page(page);
6501 ClearPageChecked(page);
6502 set_page_dirty(page);
6503 SetPageUptodate(page);
6505 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6506 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6508 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6512 return VM_FAULT_LOCKED;
6514 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6519 static int btrfs_truncate(struct inode *inode)
6521 struct btrfs_root *root = BTRFS_I(inode)->root;
6522 struct btrfs_block_rsv *rsv;
6525 struct btrfs_trans_handle *trans;
6527 u64 mask = root->sectorsize - 1;
6528 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6530 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6534 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6535 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6538 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6539 * 3 things going on here
6541 * 1) We need to reserve space for our orphan item and the space to
6542 * delete our orphan item. Lord knows we don't want to have a dangling
6543 * orphan item because we didn't reserve space to remove it.
6545 * 2) We need to reserve space to update our inode.
6547 * 3) We need to have something to cache all the space that is going to
6548 * be free'd up by the truncate operation, but also have some slack
6549 * space reserved in case it uses space during the truncate (thank you
6550 * very much snapshotting).
6552 * And we need these to all be seperate. The fact is we can use alot of
6553 * space doing the truncate, and we have no earthly idea how much space
6554 * we will use, so we need the truncate reservation to be seperate so it
6555 * doesn't end up using space reserved for updating the inode or
6556 * removing the orphan item. We also need to be able to stop the
6557 * transaction and start a new one, which means we need to be able to
6558 * update the inode several times, and we have no idea of knowing how
6559 * many times that will be, so we can't just reserve 1 item for the
6560 * entirety of the opration, so that has to be done seperately as well.
6561 * Then there is the orphan item, which does indeed need to be held on
6562 * to for the whole operation, and we need nobody to touch this reserved
6563 * space except the orphan code.
6565 * So that leaves us with
6567 * 1) root->orphan_block_rsv - for the orphan deletion.
6568 * 2) rsv - for the truncate reservation, which we will steal from the
6569 * transaction reservation.
6570 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6571 * updating the inode.
6573 rsv = btrfs_alloc_block_rsv(root);
6576 rsv->size = min_size;
6579 * 1 for the truncate slack space
6580 * 1 for the orphan item we're going to add
6581 * 1 for the orphan item deletion
6582 * 1 for updating the inode.
6584 trans = btrfs_start_transaction(root, 4);
6585 if (IS_ERR(trans)) {
6586 err = PTR_ERR(trans);
6590 /* Migrate the slack space for the truncate to our reserve */
6591 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6595 ret = btrfs_orphan_add(trans, inode);
6597 btrfs_end_transaction(trans, root);
6602 * setattr is responsible for setting the ordered_data_close flag,
6603 * but that is only tested during the last file release. That
6604 * could happen well after the next commit, leaving a great big
6605 * window where new writes may get lost if someone chooses to write
6606 * to this file after truncating to zero
6608 * The inode doesn't have any dirty data here, and so if we commit
6609 * this is a noop. If someone immediately starts writing to the inode
6610 * it is very likely we'll catch some of their writes in this
6611 * transaction, and the commit will find this file on the ordered
6612 * data list with good things to send down.
6614 * This is a best effort solution, there is still a window where
6615 * using truncate to replace the contents of the file will
6616 * end up with a zero length file after a crash.
6618 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6619 btrfs_add_ordered_operation(trans, root, inode);
6622 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6625 * This can only happen with the original transaction we
6626 * started above, every other time we shouldn't have a
6627 * transaction started yet.
6636 /* Just need the 1 for updating the inode */
6637 trans = btrfs_start_transaction(root, 1);
6638 if (IS_ERR(trans)) {
6639 err = PTR_ERR(trans);
6644 trans->block_rsv = rsv;
6646 ret = btrfs_truncate_inode_items(trans, root, inode,
6648 BTRFS_EXTENT_DATA_KEY);
6649 if (ret != -EAGAIN) {
6654 trans->block_rsv = &root->fs_info->trans_block_rsv;
6655 ret = btrfs_update_inode(trans, root, inode);
6661 nr = trans->blocks_used;
6662 btrfs_end_transaction(trans, root);
6664 btrfs_btree_balance_dirty(root, nr);
6667 if (ret == 0 && inode->i_nlink > 0) {
6668 trans->block_rsv = root->orphan_block_rsv;
6669 ret = btrfs_orphan_del(trans, inode);
6672 } else if (ret && inode->i_nlink > 0) {
6674 * Failed to do the truncate, remove us from the in memory
6677 ret = btrfs_orphan_del(NULL, inode);
6680 trans->block_rsv = &root->fs_info->trans_block_rsv;
6681 ret = btrfs_update_inode(trans, root, inode);
6685 nr = trans->blocks_used;
6686 ret = btrfs_end_transaction_throttle(trans, root);
6687 btrfs_btree_balance_dirty(root, nr);
6690 btrfs_free_block_rsv(root, rsv);
6699 * create a new subvolume directory/inode (helper for the ioctl).
6701 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6702 struct btrfs_root *new_root, u64 new_dirid)
6704 struct inode *inode;
6708 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6709 new_dirid, S_IFDIR | 0700, &index);
6711 return PTR_ERR(inode);
6712 inode->i_op = &btrfs_dir_inode_operations;
6713 inode->i_fop = &btrfs_dir_file_operations;
6716 btrfs_i_size_write(inode, 0);
6718 err = btrfs_update_inode(trans, new_root, inode);
6725 struct inode *btrfs_alloc_inode(struct super_block *sb)
6727 struct btrfs_inode *ei;
6728 struct inode *inode;
6730 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6735 ei->space_info = NULL;
6739 ei->last_sub_trans = 0;
6740 ei->logged_trans = 0;
6741 ei->delalloc_bytes = 0;
6742 ei->disk_i_size = 0;
6745 ei->index_cnt = (u64)-1;
6746 ei->last_unlink_trans = 0;
6748 spin_lock_init(&ei->lock);
6749 ei->outstanding_extents = 0;
6750 ei->reserved_extents = 0;
6752 ei->ordered_data_close = 0;
6753 ei->orphan_meta_reserved = 0;
6754 ei->dummy_inode = 0;
6756 ei->force_compress = BTRFS_COMPRESS_NONE;
6758 ei->delayed_node = NULL;
6760 inode = &ei->vfs_inode;
6761 extent_map_tree_init(&ei->extent_tree);
6762 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6763 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6764 mutex_init(&ei->log_mutex);
6765 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6766 INIT_LIST_HEAD(&ei->i_orphan);
6767 INIT_LIST_HEAD(&ei->delalloc_inodes);
6768 INIT_LIST_HEAD(&ei->ordered_operations);
6769 RB_CLEAR_NODE(&ei->rb_node);
6774 static void btrfs_i_callback(struct rcu_head *head)
6776 struct inode *inode = container_of(head, struct inode, i_rcu);
6777 INIT_LIST_HEAD(&inode->i_dentry);
6778 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6781 void btrfs_destroy_inode(struct inode *inode)
6783 struct btrfs_ordered_extent *ordered;
6784 struct btrfs_root *root = BTRFS_I(inode)->root;
6786 WARN_ON(!list_empty(&inode->i_dentry));
6787 WARN_ON(inode->i_data.nrpages);
6788 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6789 WARN_ON(BTRFS_I(inode)->reserved_extents);
6790 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6791 WARN_ON(BTRFS_I(inode)->csum_bytes);
6794 * This can happen where we create an inode, but somebody else also
6795 * created the same inode and we need to destroy the one we already
6802 * Make sure we're properly removed from the ordered operation
6806 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6807 spin_lock(&root->fs_info->ordered_extent_lock);
6808 list_del_init(&BTRFS_I(inode)->ordered_operations);
6809 spin_unlock(&root->fs_info->ordered_extent_lock);
6812 spin_lock(&root->orphan_lock);
6813 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6814 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6815 (unsigned long long)btrfs_ino(inode));
6816 list_del_init(&BTRFS_I(inode)->i_orphan);
6818 spin_unlock(&root->orphan_lock);
6821 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6825 printk(KERN_ERR "btrfs found ordered "
6826 "extent %llu %llu on inode cleanup\n",
6827 (unsigned long long)ordered->file_offset,
6828 (unsigned long long)ordered->len);
6829 btrfs_remove_ordered_extent(inode, ordered);
6830 btrfs_put_ordered_extent(ordered);
6831 btrfs_put_ordered_extent(ordered);
6834 inode_tree_del(inode);
6835 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6837 btrfs_remove_delayed_node(inode);
6838 call_rcu(&inode->i_rcu, btrfs_i_callback);
6841 int btrfs_drop_inode(struct inode *inode)
6843 struct btrfs_root *root = BTRFS_I(inode)->root;
6845 if (btrfs_root_refs(&root->root_item) == 0 &&
6846 !btrfs_is_free_space_inode(root, inode))
6849 return generic_drop_inode(inode);
6852 static void init_once(void *foo)
6854 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6856 inode_init_once(&ei->vfs_inode);
6859 void btrfs_destroy_cachep(void)
6861 if (btrfs_inode_cachep)
6862 kmem_cache_destroy(btrfs_inode_cachep);
6863 if (btrfs_trans_handle_cachep)
6864 kmem_cache_destroy(btrfs_trans_handle_cachep);
6865 if (btrfs_transaction_cachep)
6866 kmem_cache_destroy(btrfs_transaction_cachep);
6867 if (btrfs_path_cachep)
6868 kmem_cache_destroy(btrfs_path_cachep);
6869 if (btrfs_free_space_cachep)
6870 kmem_cache_destroy(btrfs_free_space_cachep);
6873 int btrfs_init_cachep(void)
6875 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6876 sizeof(struct btrfs_inode), 0,
6877 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6878 if (!btrfs_inode_cachep)
6881 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6882 sizeof(struct btrfs_trans_handle), 0,
6883 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6884 if (!btrfs_trans_handle_cachep)
6887 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6888 sizeof(struct btrfs_transaction), 0,
6889 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6890 if (!btrfs_transaction_cachep)
6893 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6894 sizeof(struct btrfs_path), 0,
6895 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6896 if (!btrfs_path_cachep)
6899 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6900 sizeof(struct btrfs_free_space), 0,
6901 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6902 if (!btrfs_free_space_cachep)
6907 btrfs_destroy_cachep();
6911 static int btrfs_getattr(struct vfsmount *mnt,
6912 struct dentry *dentry, struct kstat *stat)
6914 struct inode *inode = dentry->d_inode;
6915 generic_fillattr(inode, stat);
6916 stat->dev = BTRFS_I(inode)->root->anon_dev;
6917 stat->blksize = PAGE_CACHE_SIZE;
6918 stat->blocks = (inode_get_bytes(inode) +
6919 BTRFS_I(inode)->delalloc_bytes) >> 9;
6924 * If a file is moved, it will inherit the cow and compression flags of the new
6927 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6929 struct btrfs_inode *b_dir = BTRFS_I(dir);
6930 struct btrfs_inode *b_inode = BTRFS_I(inode);
6932 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6933 b_inode->flags |= BTRFS_INODE_NODATACOW;
6935 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6937 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6938 b_inode->flags |= BTRFS_INODE_COMPRESS;
6940 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6943 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6944 struct inode *new_dir, struct dentry *new_dentry)
6946 struct btrfs_trans_handle *trans;
6947 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6948 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6949 struct inode *new_inode = new_dentry->d_inode;
6950 struct inode *old_inode = old_dentry->d_inode;
6951 struct timespec ctime = CURRENT_TIME;
6955 u64 old_ino = btrfs_ino(old_inode);
6957 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6960 /* we only allow rename subvolume link between subvolumes */
6961 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6964 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6965 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6968 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6969 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6972 * we're using rename to replace one file with another.
6973 * and the replacement file is large. Start IO on it now so
6974 * we don't add too much work to the end of the transaction
6976 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6977 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6978 filemap_flush(old_inode->i_mapping);
6980 /* close the racy window with snapshot create/destroy ioctl */
6981 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6982 down_read(&root->fs_info->subvol_sem);
6984 * We want to reserve the absolute worst case amount of items. So if
6985 * both inodes are subvols and we need to unlink them then that would
6986 * require 4 item modifications, but if they are both normal inodes it
6987 * would require 5 item modifications, so we'll assume their normal
6988 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6989 * should cover the worst case number of items we'll modify.
6991 trans = btrfs_start_transaction(root, 20);
6992 if (IS_ERR(trans)) {
6993 ret = PTR_ERR(trans);
6998 btrfs_record_root_in_trans(trans, dest);
7000 ret = btrfs_set_inode_index(new_dir, &index);
7004 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7005 /* force full log commit if subvolume involved. */
7006 root->fs_info->last_trans_log_full_commit = trans->transid;
7008 ret = btrfs_insert_inode_ref(trans, dest,
7009 new_dentry->d_name.name,
7010 new_dentry->d_name.len,
7012 btrfs_ino(new_dir), index);
7016 * this is an ugly little race, but the rename is required
7017 * to make sure that if we crash, the inode is either at the
7018 * old name or the new one. pinning the log transaction lets
7019 * us make sure we don't allow a log commit to come in after
7020 * we unlink the name but before we add the new name back in.
7022 btrfs_pin_log_trans(root);
7025 * make sure the inode gets flushed if it is replacing
7028 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7029 btrfs_add_ordered_operation(trans, root, old_inode);
7031 old_dir->i_ctime = old_dir->i_mtime = ctime;
7032 new_dir->i_ctime = new_dir->i_mtime = ctime;
7033 old_inode->i_ctime = ctime;
7035 if (old_dentry->d_parent != new_dentry->d_parent)
7036 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7038 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7039 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7040 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7041 old_dentry->d_name.name,
7042 old_dentry->d_name.len);
7044 ret = __btrfs_unlink_inode(trans, root, old_dir,
7045 old_dentry->d_inode,
7046 old_dentry->d_name.name,
7047 old_dentry->d_name.len);
7049 ret = btrfs_update_inode(trans, root, old_inode);
7054 new_inode->i_ctime = CURRENT_TIME;
7055 if (unlikely(btrfs_ino(new_inode) ==
7056 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7057 root_objectid = BTRFS_I(new_inode)->location.objectid;
7058 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7060 new_dentry->d_name.name,
7061 new_dentry->d_name.len);
7062 BUG_ON(new_inode->i_nlink == 0);
7064 ret = btrfs_unlink_inode(trans, dest, new_dir,
7065 new_dentry->d_inode,
7066 new_dentry->d_name.name,
7067 new_dentry->d_name.len);
7070 if (new_inode->i_nlink == 0) {
7071 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7076 fixup_inode_flags(new_dir, old_inode);
7078 ret = btrfs_add_link(trans, new_dir, old_inode,
7079 new_dentry->d_name.name,
7080 new_dentry->d_name.len, 0, index);
7083 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7084 struct dentry *parent = new_dentry->d_parent;
7085 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7086 btrfs_end_log_trans(root);
7089 btrfs_end_transaction_throttle(trans, root);
7091 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7092 up_read(&root->fs_info->subvol_sem);
7098 * some fairly slow code that needs optimization. This walks the list
7099 * of all the inodes with pending delalloc and forces them to disk.
7101 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7103 struct list_head *head = &root->fs_info->delalloc_inodes;
7104 struct btrfs_inode *binode;
7105 struct inode *inode;
7107 if (root->fs_info->sb->s_flags & MS_RDONLY)
7110 spin_lock(&root->fs_info->delalloc_lock);
7111 while (!list_empty(head)) {
7112 binode = list_entry(head->next, struct btrfs_inode,
7114 inode = igrab(&binode->vfs_inode);
7116 list_del_init(&binode->delalloc_inodes);
7117 spin_unlock(&root->fs_info->delalloc_lock);
7119 filemap_flush(inode->i_mapping);
7121 btrfs_add_delayed_iput(inode);
7126 spin_lock(&root->fs_info->delalloc_lock);
7128 spin_unlock(&root->fs_info->delalloc_lock);
7130 /* the filemap_flush will queue IO into the worker threads, but
7131 * we have to make sure the IO is actually started and that
7132 * ordered extents get created before we return
7134 atomic_inc(&root->fs_info->async_submit_draining);
7135 while (atomic_read(&root->fs_info->nr_async_submits) ||
7136 atomic_read(&root->fs_info->async_delalloc_pages)) {
7137 wait_event(root->fs_info->async_submit_wait,
7138 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7139 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7141 atomic_dec(&root->fs_info->async_submit_draining);
7145 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7146 const char *symname)
7148 struct btrfs_trans_handle *trans;
7149 struct btrfs_root *root = BTRFS_I(dir)->root;
7150 struct btrfs_path *path;
7151 struct btrfs_key key;
7152 struct inode *inode = NULL;
7160 struct btrfs_file_extent_item *ei;
7161 struct extent_buffer *leaf;
7162 unsigned long nr = 0;
7164 name_len = strlen(symname) + 1;
7165 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7166 return -ENAMETOOLONG;
7169 * 2 items for inode item and ref
7170 * 2 items for dir items
7171 * 1 item for xattr if selinux is on
7173 trans = btrfs_start_transaction(root, 5);
7175 return PTR_ERR(trans);
7177 err = btrfs_find_free_ino(root, &objectid);
7181 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7182 dentry->d_name.len, btrfs_ino(dir), objectid,
7183 S_IFLNK|S_IRWXUGO, &index);
7184 if (IS_ERR(inode)) {
7185 err = PTR_ERR(inode);
7189 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7195 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7199 inode->i_mapping->a_ops = &btrfs_aops;
7200 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7201 inode->i_fop = &btrfs_file_operations;
7202 inode->i_op = &btrfs_file_inode_operations;
7203 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7208 path = btrfs_alloc_path();
7214 key.objectid = btrfs_ino(inode);
7216 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7217 datasize = btrfs_file_extent_calc_inline_size(name_len);
7218 err = btrfs_insert_empty_item(trans, root, path, &key,
7222 btrfs_free_path(path);
7225 leaf = path->nodes[0];
7226 ei = btrfs_item_ptr(leaf, path->slots[0],
7227 struct btrfs_file_extent_item);
7228 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7229 btrfs_set_file_extent_type(leaf, ei,
7230 BTRFS_FILE_EXTENT_INLINE);
7231 btrfs_set_file_extent_encryption(leaf, ei, 0);
7232 btrfs_set_file_extent_compression(leaf, ei, 0);
7233 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7234 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7236 ptr = btrfs_file_extent_inline_start(ei);
7237 write_extent_buffer(leaf, symname, ptr, name_len);
7238 btrfs_mark_buffer_dirty(leaf);
7239 btrfs_free_path(path);
7241 inode->i_op = &btrfs_symlink_inode_operations;
7242 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7243 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7244 inode_set_bytes(inode, name_len);
7245 btrfs_i_size_write(inode, name_len - 1);
7246 err = btrfs_update_inode(trans, root, inode);
7251 nr = trans->blocks_used;
7252 btrfs_end_transaction_throttle(trans, root);
7254 inode_dec_link_count(inode);
7257 btrfs_btree_balance_dirty(root, nr);
7261 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7262 u64 start, u64 num_bytes, u64 min_size,
7263 loff_t actual_len, u64 *alloc_hint,
7264 struct btrfs_trans_handle *trans)
7266 struct btrfs_root *root = BTRFS_I(inode)->root;
7267 struct btrfs_key ins;
7268 u64 cur_offset = start;
7271 bool own_trans = true;
7275 while (num_bytes > 0) {
7277 trans = btrfs_start_transaction(root, 3);
7278 if (IS_ERR(trans)) {
7279 ret = PTR_ERR(trans);
7284 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7285 0, *alloc_hint, (u64)-1, &ins, 1);
7288 btrfs_end_transaction(trans, root);
7292 ret = insert_reserved_file_extent(trans, inode,
7293 cur_offset, ins.objectid,
7294 ins.offset, ins.offset,
7295 ins.offset, 0, 0, 0,
7296 BTRFS_FILE_EXTENT_PREALLOC);
7298 btrfs_drop_extent_cache(inode, cur_offset,
7299 cur_offset + ins.offset -1, 0);
7301 num_bytes -= ins.offset;
7302 cur_offset += ins.offset;
7303 *alloc_hint = ins.objectid + ins.offset;
7305 inode->i_ctime = CURRENT_TIME;
7306 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7307 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7308 (actual_len > inode->i_size) &&
7309 (cur_offset > inode->i_size)) {
7310 if (cur_offset > actual_len)
7311 i_size = actual_len;
7313 i_size = cur_offset;
7314 i_size_write(inode, i_size);
7315 btrfs_ordered_update_i_size(inode, i_size, NULL);
7318 ret = btrfs_update_inode(trans, root, inode);
7322 btrfs_end_transaction(trans, root);
7327 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7328 u64 start, u64 num_bytes, u64 min_size,
7329 loff_t actual_len, u64 *alloc_hint)
7331 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7332 min_size, actual_len, alloc_hint,
7336 int btrfs_prealloc_file_range_trans(struct inode *inode,
7337 struct btrfs_trans_handle *trans, int mode,
7338 u64 start, u64 num_bytes, u64 min_size,
7339 loff_t actual_len, u64 *alloc_hint)
7341 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7342 min_size, actual_len, alloc_hint, trans);
7345 static int btrfs_set_page_dirty(struct page *page)
7347 return __set_page_dirty_nobuffers(page);
7350 static int btrfs_permission(struct inode *inode, int mask)
7352 struct btrfs_root *root = BTRFS_I(inode)->root;
7353 umode_t mode = inode->i_mode;
7355 if (mask & MAY_WRITE &&
7356 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7357 if (btrfs_root_readonly(root))
7359 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7362 return generic_permission(inode, mask);
7365 static const struct inode_operations btrfs_dir_inode_operations = {
7366 .getattr = btrfs_getattr,
7367 .lookup = btrfs_lookup,
7368 .create = btrfs_create,
7369 .unlink = btrfs_unlink,
7371 .mkdir = btrfs_mkdir,
7372 .rmdir = btrfs_rmdir,
7373 .rename = btrfs_rename,
7374 .symlink = btrfs_symlink,
7375 .setattr = btrfs_setattr,
7376 .mknod = btrfs_mknod,
7377 .setxattr = btrfs_setxattr,
7378 .getxattr = btrfs_getxattr,
7379 .listxattr = btrfs_listxattr,
7380 .removexattr = btrfs_removexattr,
7381 .permission = btrfs_permission,
7382 .get_acl = btrfs_get_acl,
7384 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7385 .lookup = btrfs_lookup,
7386 .permission = btrfs_permission,
7387 .get_acl = btrfs_get_acl,
7390 static const struct file_operations btrfs_dir_file_operations = {
7391 .llseek = generic_file_llseek,
7392 .read = generic_read_dir,
7393 .readdir = btrfs_real_readdir,
7394 .unlocked_ioctl = btrfs_ioctl,
7395 #ifdef CONFIG_COMPAT
7396 .compat_ioctl = btrfs_ioctl,
7398 .release = btrfs_release_file,
7399 .fsync = btrfs_sync_file,
7402 static struct extent_io_ops btrfs_extent_io_ops = {
7403 .fill_delalloc = run_delalloc_range,
7404 .submit_bio_hook = btrfs_submit_bio_hook,
7405 .merge_bio_hook = btrfs_merge_bio_hook,
7406 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7407 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7408 .writepage_start_hook = btrfs_writepage_start_hook,
7409 .readpage_io_failed_hook = btrfs_io_failed_hook,
7410 .set_bit_hook = btrfs_set_bit_hook,
7411 .clear_bit_hook = btrfs_clear_bit_hook,
7412 .merge_extent_hook = btrfs_merge_extent_hook,
7413 .split_extent_hook = btrfs_split_extent_hook,
7417 * btrfs doesn't support the bmap operation because swapfiles
7418 * use bmap to make a mapping of extents in the file. They assume
7419 * these extents won't change over the life of the file and they
7420 * use the bmap result to do IO directly to the drive.
7422 * the btrfs bmap call would return logical addresses that aren't
7423 * suitable for IO and they also will change frequently as COW
7424 * operations happen. So, swapfile + btrfs == corruption.
7426 * For now we're avoiding this by dropping bmap.
7428 static const struct address_space_operations btrfs_aops = {
7429 .readpage = btrfs_readpage,
7430 .writepage = btrfs_writepage,
7431 .writepages = btrfs_writepages,
7432 .readpages = btrfs_readpages,
7433 .direct_IO = btrfs_direct_IO,
7434 .invalidatepage = btrfs_invalidatepage,
7435 .releasepage = btrfs_releasepage,
7436 .set_page_dirty = btrfs_set_page_dirty,
7437 .error_remove_page = generic_error_remove_page,
7440 static const struct address_space_operations btrfs_symlink_aops = {
7441 .readpage = btrfs_readpage,
7442 .writepage = btrfs_writepage,
7443 .invalidatepage = btrfs_invalidatepage,
7444 .releasepage = btrfs_releasepage,
7447 static const struct inode_operations btrfs_file_inode_operations = {
7448 .getattr = btrfs_getattr,
7449 .setattr = btrfs_setattr,
7450 .setxattr = btrfs_setxattr,
7451 .getxattr = btrfs_getxattr,
7452 .listxattr = btrfs_listxattr,
7453 .removexattr = btrfs_removexattr,
7454 .permission = btrfs_permission,
7455 .fiemap = btrfs_fiemap,
7456 .get_acl = btrfs_get_acl,
7458 static const struct inode_operations btrfs_special_inode_operations = {
7459 .getattr = btrfs_getattr,
7460 .setattr = btrfs_setattr,
7461 .permission = btrfs_permission,
7462 .setxattr = btrfs_setxattr,
7463 .getxattr = btrfs_getxattr,
7464 .listxattr = btrfs_listxattr,
7465 .removexattr = btrfs_removexattr,
7466 .get_acl = btrfs_get_acl,
7468 static const struct inode_operations btrfs_symlink_inode_operations = {
7469 .readlink = generic_readlink,
7470 .follow_link = page_follow_link_light,
7471 .put_link = page_put_link,
7472 .getattr = btrfs_getattr,
7473 .permission = btrfs_permission,
7474 .setxattr = btrfs_setxattr,
7475 .getxattr = btrfs_getxattr,
7476 .listxattr = btrfs_listxattr,
7477 .removexattr = btrfs_removexattr,
7478 .get_acl = btrfs_get_acl,
7481 const struct dentry_operations btrfs_dentry_operations = {
7482 .d_delete = btrfs_dentry_delete,
7483 .d_release = btrfs_dentry_release,