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"
48 #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);
96 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct inode *inode);
99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
100 struct inode *inode, struct inode *dir,
101 const struct qstr *qstr)
105 err = btrfs_init_acl(trans, inode, dir);
107 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
112 * this does all the hard work for inserting an inline extent into
113 * the btree. The caller should have done a btrfs_drop_extents so that
114 * no overlapping inline items exist in the btree
116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
117 struct btrfs_root *root, struct inode *inode,
118 u64 start, size_t size, size_t compressed_size,
120 struct page **compressed_pages)
122 struct btrfs_key key;
123 struct btrfs_path *path;
124 struct extent_buffer *leaf;
125 struct page *page = NULL;
128 struct btrfs_file_extent_item *ei;
131 size_t cur_size = size;
133 unsigned long offset;
135 if (compressed_size && compressed_pages)
136 cur_size = compressed_size;
138 path = btrfs_alloc_path();
142 path->leave_spinning = 1;
144 key.objectid = btrfs_ino(inode);
146 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
147 datasize = btrfs_file_extent_calc_inline_size(cur_size);
149 inode_add_bytes(inode, size);
150 ret = btrfs_insert_empty_item(trans, root, path, &key,
157 leaf = path->nodes[0];
158 ei = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162 btrfs_set_file_extent_encryption(leaf, ei, 0);
163 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165 ptr = btrfs_file_extent_inline_start(ei);
167 if (compress_type != BTRFS_COMPRESS_NONE) {
170 while (compressed_size > 0) {
171 cpage = compressed_pages[i];
172 cur_size = min_t(unsigned long, compressed_size,
175 kaddr = kmap_atomic(cpage, KM_USER0);
176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
177 kunmap_atomic(kaddr, KM_USER0);
181 compressed_size -= cur_size;
183 btrfs_set_file_extent_compression(leaf, ei,
186 page = find_get_page(inode->i_mapping,
187 start >> PAGE_CACHE_SHIFT);
188 btrfs_set_file_extent_compression(leaf, ei, 0);
189 kaddr = kmap_atomic(page, KM_USER0);
190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192 kunmap_atomic(kaddr, KM_USER0);
193 page_cache_release(page);
195 btrfs_mark_buffer_dirty(leaf);
196 btrfs_free_path(path);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode)->disk_i_size = inode->i_size;
208 btrfs_update_inode(trans, root, inode);
212 btrfs_free_path(path);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
225 size_t compressed_size, int compress_type,
226 struct page **compressed_pages)
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
234 u64 data_len = inline_len;
238 data_len = compressed_size;
241 actual_end >= PAGE_CACHE_SIZE ||
242 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244 (actual_end & (root->sectorsize - 1)) == 0) ||
246 data_len > root->fs_info->max_inline) {
250 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
254 if (isize > actual_end)
255 inline_len = min_t(u64, isize, actual_end);
256 ret = insert_inline_extent(trans, root, inode, start,
257 inline_len, compressed_size,
258 compress_type, compressed_pages);
260 btrfs_delalloc_release_metadata(inode, end + 1 - start);
261 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
265 struct async_extent {
270 unsigned long nr_pages;
272 struct list_head list;
277 struct btrfs_root *root;
278 struct page *locked_page;
281 struct list_head extents;
282 struct btrfs_work work;
285 static noinline int add_async_extent(struct async_cow *cow,
286 u64 start, u64 ram_size,
289 unsigned long nr_pages,
292 struct async_extent *async_extent;
294 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
295 BUG_ON(!async_extent);
296 async_extent->start = start;
297 async_extent->ram_size = ram_size;
298 async_extent->compressed_size = compressed_size;
299 async_extent->pages = pages;
300 async_extent->nr_pages = nr_pages;
301 async_extent->compress_type = compress_type;
302 list_add_tail(&async_extent->list, &cow->extents);
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
322 static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
325 struct async_cow *async_cow,
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
331 u64 blocksize = root->sectorsize;
333 u64 isize = i_size_read(inode);
335 struct page **pages = NULL;
336 unsigned long nr_pages;
337 unsigned long nr_pages_ret = 0;
338 unsigned long total_compressed = 0;
339 unsigned long total_in = 0;
340 unsigned long max_compressed = 128 * 1024;
341 unsigned long max_uncompressed = 128 * 1024;
344 int compress_type = root->fs_info->compress_type;
346 /* if this is a small write inside eof, kick off a defragbot */
347 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
348 btrfs_add_inode_defrag(NULL, inode);
350 actual_end = min_t(u64, isize, end + 1);
353 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
354 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
357 * we don't want to send crud past the end of i_size through
358 * compression, that's just a waste of CPU time. So, if the
359 * end of the file is before the start of our current
360 * requested range of bytes, we bail out to the uncompressed
361 * cleanup code that can deal with all of this.
363 * It isn't really the fastest way to fix things, but this is a
364 * very uncommon corner.
366 if (actual_end <= start)
367 goto cleanup_and_bail_uncompressed;
369 total_compressed = actual_end - start;
371 /* we want to make sure that amount of ram required to uncompress
372 * an extent is reasonable, so we limit the total size in ram
373 * of a compressed extent to 128k. This is a crucial number
374 * because it also controls how easily we can spread reads across
375 * cpus for decompression.
377 * We also want to make sure the amount of IO required to do
378 * a random read is reasonably small, so we limit the size of
379 * a compressed extent to 128k.
381 total_compressed = min(total_compressed, max_uncompressed);
382 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
383 num_bytes = max(blocksize, num_bytes);
388 * we do compression for mount -o compress and when the
389 * inode has not been flagged as nocompress. This flag can
390 * change at any time if we discover bad compression ratios.
392 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
393 (btrfs_test_opt(root, COMPRESS) ||
394 (BTRFS_I(inode)->force_compress) ||
395 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
397 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
399 /* just bail out to the uncompressed code */
403 if (BTRFS_I(inode)->force_compress)
404 compress_type = BTRFS_I(inode)->force_compress;
406 ret = btrfs_compress_pages(compress_type,
407 inode->i_mapping, start,
408 total_compressed, pages,
409 nr_pages, &nr_pages_ret,
415 unsigned long offset = total_compressed &
416 (PAGE_CACHE_SIZE - 1);
417 struct page *page = pages[nr_pages_ret - 1];
420 /* zero the tail end of the last page, we might be
421 * sending it down to disk
424 kaddr = kmap_atomic(page, KM_USER0);
425 memset(kaddr + offset, 0,
426 PAGE_CACHE_SIZE - offset);
427 kunmap_atomic(kaddr, KM_USER0);
434 trans = btrfs_join_transaction(root);
435 BUG_ON(IS_ERR(trans));
436 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
438 /* lets try to make an inline extent */
439 if (ret || total_in < (actual_end - start)) {
440 /* we didn't compress the entire range, try
441 * to make an uncompressed inline extent.
443 ret = cow_file_range_inline(trans, root, inode,
444 start, end, 0, 0, NULL);
446 /* try making a compressed inline extent */
447 ret = cow_file_range_inline(trans, root, inode,
450 compress_type, pages);
454 * inline extent creation worked, we don't need
455 * to create any more async work items. Unlock
456 * and free up our temp pages.
458 extent_clear_unlock_delalloc(inode,
459 &BTRFS_I(inode)->io_tree,
461 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
462 EXTENT_CLEAR_DELALLOC |
463 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
465 btrfs_end_transaction(trans, root);
468 btrfs_end_transaction(trans, root);
473 * we aren't doing an inline extent round the compressed size
474 * up to a block size boundary so the allocator does sane
477 total_compressed = (total_compressed + blocksize - 1) &
481 * one last check to make sure the compression is really a
482 * win, compare the page count read with the blocks on disk
484 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
485 ~(PAGE_CACHE_SIZE - 1);
486 if (total_compressed >= total_in) {
489 num_bytes = total_in;
492 if (!will_compress && pages) {
494 * the compression code ran but failed to make things smaller,
495 * free any pages it allocated and our page pointer array
497 for (i = 0; i < nr_pages_ret; i++) {
498 WARN_ON(pages[i]->mapping);
499 page_cache_release(pages[i]);
503 total_compressed = 0;
506 /* flag the file so we don't compress in the future */
507 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
508 !(BTRFS_I(inode)->force_compress)) {
509 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
515 /* the async work queues will take care of doing actual
516 * allocation on disk for these compressed pages,
517 * and will submit them to the elevator.
519 add_async_extent(async_cow, start, num_bytes,
520 total_compressed, pages, nr_pages_ret,
523 if (start + num_bytes < end) {
530 cleanup_and_bail_uncompressed:
532 * No compression, but we still need to write the pages in
533 * the file we've been given so far. redirty the locked
534 * page if it corresponds to our extent and set things up
535 * for the async work queue to run cow_file_range to do
536 * the normal delalloc dance
538 if (page_offset(locked_page) >= start &&
539 page_offset(locked_page) <= end) {
540 __set_page_dirty_nobuffers(locked_page);
541 /* unlocked later on in the async handlers */
543 add_async_extent(async_cow, start, end - start + 1,
544 0, NULL, 0, BTRFS_COMPRESS_NONE);
552 for (i = 0; i < nr_pages_ret; i++) {
553 WARN_ON(pages[i]->mapping);
554 page_cache_release(pages[i]);
562 * phase two of compressed writeback. This is the ordered portion
563 * of the code, which only gets called in the order the work was
564 * queued. We walk all the async extents created by compress_file_range
565 * and send them down to the disk.
567 static noinline int submit_compressed_extents(struct inode *inode,
568 struct async_cow *async_cow)
570 struct async_extent *async_extent;
572 struct btrfs_trans_handle *trans;
573 struct btrfs_key ins;
574 struct extent_map *em;
575 struct btrfs_root *root = BTRFS_I(inode)->root;
576 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
577 struct extent_io_tree *io_tree;
580 if (list_empty(&async_cow->extents))
584 while (!list_empty(&async_cow->extents)) {
585 async_extent = list_entry(async_cow->extents.next,
586 struct async_extent, list);
587 list_del(&async_extent->list);
589 io_tree = &BTRFS_I(inode)->io_tree;
592 /* did the compression code fall back to uncompressed IO? */
593 if (!async_extent->pages) {
594 int page_started = 0;
595 unsigned long nr_written = 0;
597 lock_extent(io_tree, async_extent->start,
598 async_extent->start +
599 async_extent->ram_size - 1, GFP_NOFS);
601 /* allocate blocks */
602 ret = cow_file_range(inode, async_cow->locked_page,
604 async_extent->start +
605 async_extent->ram_size - 1,
606 &page_started, &nr_written, 0);
609 * if page_started, cow_file_range inserted an
610 * inline extent and took care of all the unlocking
611 * and IO for us. Otherwise, we need to submit
612 * all those pages down to the drive.
614 if (!page_started && !ret)
615 extent_write_locked_range(io_tree,
616 inode, async_extent->start,
617 async_extent->start +
618 async_extent->ram_size - 1,
626 lock_extent(io_tree, async_extent->start,
627 async_extent->start + async_extent->ram_size - 1,
630 trans = btrfs_join_transaction(root);
631 BUG_ON(IS_ERR(trans));
632 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
633 ret = btrfs_reserve_extent(trans, root,
634 async_extent->compressed_size,
635 async_extent->compressed_size,
638 btrfs_end_transaction(trans, root);
642 for (i = 0; i < async_extent->nr_pages; i++) {
643 WARN_ON(async_extent->pages[i]->mapping);
644 page_cache_release(async_extent->pages[i]);
646 kfree(async_extent->pages);
647 async_extent->nr_pages = 0;
648 async_extent->pages = NULL;
649 unlock_extent(io_tree, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1, GFP_NOFS);
656 * here we're doing allocation and writeback of the
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
663 em = alloc_extent_map();
665 em->start = async_extent->start;
666 em->len = async_extent->ram_size;
667 em->orig_start = em->start;
669 em->block_start = ins.objectid;
670 em->block_len = ins.offset;
671 em->bdev = root->fs_info->fs_devices->latest_bdev;
672 em->compress_type = async_extent->compress_type;
673 set_bit(EXTENT_FLAG_PINNED, &em->flags);
674 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
677 write_lock(&em_tree->lock);
678 ret = add_extent_mapping(em_tree, em);
679 write_unlock(&em_tree->lock);
680 if (ret != -EEXIST) {
684 btrfs_drop_extent_cache(inode, async_extent->start,
685 async_extent->start +
686 async_extent->ram_size - 1, 0);
689 ret = btrfs_add_ordered_extent_compress(inode,
692 async_extent->ram_size,
694 BTRFS_ORDERED_COMPRESSED,
695 async_extent->compress_type);
699 * clear dirty, set writeback and unlock the pages.
701 extent_clear_unlock_delalloc(inode,
702 &BTRFS_I(inode)->io_tree,
704 async_extent->start +
705 async_extent->ram_size - 1,
706 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
707 EXTENT_CLEAR_UNLOCK |
708 EXTENT_CLEAR_DELALLOC |
709 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
711 ret = btrfs_submit_compressed_write(inode,
713 async_extent->ram_size,
715 ins.offset, async_extent->pages,
716 async_extent->nr_pages);
719 alloc_hint = ins.objectid + ins.offset;
727 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
730 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731 struct extent_map *em;
734 read_lock(&em_tree->lock);
735 em = search_extent_mapping(em_tree, start, num_bytes);
738 * if block start isn't an actual block number then find the
739 * first block in this inode and use that as a hint. If that
740 * block is also bogus then just don't worry about it.
742 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
744 em = search_extent_mapping(em_tree, 0, 0);
745 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
746 alloc_hint = em->block_start;
750 alloc_hint = em->block_start;
754 read_unlock(&em_tree->lock);
760 * when extent_io.c finds a delayed allocation range in the file,
761 * the call backs end up in this code. The basic idea is to
762 * allocate extents on disk for the range, and create ordered data structs
763 * in ram to track those extents.
765 * locked_page is the page that writepage had locked already. We use
766 * it to make sure we don't do extra locks or unlocks.
768 * *page_started is set to one if we unlock locked_page and do everything
769 * required to start IO on it. It may be clean and already done with
772 static noinline int cow_file_range(struct inode *inode,
773 struct page *locked_page,
774 u64 start, u64 end, int *page_started,
775 unsigned long *nr_written,
778 struct btrfs_root *root = BTRFS_I(inode)->root;
779 struct btrfs_trans_handle *trans;
782 unsigned long ram_size;
785 u64 blocksize = root->sectorsize;
786 struct btrfs_key ins;
787 struct extent_map *em;
788 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
791 BUG_ON(btrfs_is_free_space_inode(root, inode));
792 trans = btrfs_join_transaction(root);
793 BUG_ON(IS_ERR(trans));
794 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
796 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
797 num_bytes = max(blocksize, num_bytes);
798 disk_num_bytes = num_bytes;
801 /* if this is a small write inside eof, kick off defrag */
802 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
803 btrfs_add_inode_defrag(trans, inode);
806 /* lets try to make an inline extent */
807 ret = cow_file_range_inline(trans, root, inode,
808 start, end, 0, 0, NULL);
810 extent_clear_unlock_delalloc(inode,
811 &BTRFS_I(inode)->io_tree,
813 EXTENT_CLEAR_UNLOCK_PAGE |
814 EXTENT_CLEAR_UNLOCK |
815 EXTENT_CLEAR_DELALLOC |
817 EXTENT_SET_WRITEBACK |
818 EXTENT_END_WRITEBACK);
820 *nr_written = *nr_written +
821 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
828 BUG_ON(disk_num_bytes >
829 btrfs_super_total_bytes(root->fs_info->super_copy));
831 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
832 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
834 while (disk_num_bytes > 0) {
837 cur_alloc_size = disk_num_bytes;
838 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
839 root->sectorsize, 0, alloc_hint,
843 em = alloc_extent_map();
846 em->orig_start = em->start;
847 ram_size = ins.offset;
848 em->len = ins.offset;
850 em->block_start = ins.objectid;
851 em->block_len = ins.offset;
852 em->bdev = root->fs_info->fs_devices->latest_bdev;
853 set_bit(EXTENT_FLAG_PINNED, &em->flags);
856 write_lock(&em_tree->lock);
857 ret = add_extent_mapping(em_tree, em);
858 write_unlock(&em_tree->lock);
859 if (ret != -EEXIST) {
863 btrfs_drop_extent_cache(inode, start,
864 start + ram_size - 1, 0);
867 cur_alloc_size = ins.offset;
868 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
869 ram_size, cur_alloc_size, 0);
872 if (root->root_key.objectid ==
873 BTRFS_DATA_RELOC_TREE_OBJECTID) {
874 ret = btrfs_reloc_clone_csums(inode, start,
879 if (disk_num_bytes < cur_alloc_size)
882 /* we're not doing compressed IO, don't unlock the first
883 * page (which the caller expects to stay locked), don't
884 * clear any dirty bits and don't set any writeback bits
886 * Do set the Private2 bit so we know this page was properly
887 * setup for writepage
889 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
890 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
893 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
894 start, start + ram_size - 1,
896 disk_num_bytes -= cur_alloc_size;
897 num_bytes -= cur_alloc_size;
898 alloc_hint = ins.objectid + ins.offset;
899 start += cur_alloc_size;
903 btrfs_end_transaction(trans, root);
909 * work queue call back to started compression on a file and pages
911 static noinline void async_cow_start(struct btrfs_work *work)
913 struct async_cow *async_cow;
915 async_cow = container_of(work, struct async_cow, work);
917 compress_file_range(async_cow->inode, async_cow->locked_page,
918 async_cow->start, async_cow->end, async_cow,
921 async_cow->inode = NULL;
925 * work queue call back to submit previously compressed pages
927 static noinline void async_cow_submit(struct btrfs_work *work)
929 struct async_cow *async_cow;
930 struct btrfs_root *root;
931 unsigned long nr_pages;
933 async_cow = container_of(work, struct async_cow, work);
935 root = async_cow->root;
936 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
939 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
941 if (atomic_read(&root->fs_info->async_delalloc_pages) <
943 waitqueue_active(&root->fs_info->async_submit_wait))
944 wake_up(&root->fs_info->async_submit_wait);
946 if (async_cow->inode)
947 submit_compressed_extents(async_cow->inode, async_cow);
950 static noinline void async_cow_free(struct btrfs_work *work)
952 struct async_cow *async_cow;
953 async_cow = container_of(work, struct async_cow, work);
957 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
958 u64 start, u64 end, int *page_started,
959 unsigned long *nr_written)
961 struct async_cow *async_cow;
962 struct btrfs_root *root = BTRFS_I(inode)->root;
963 unsigned long nr_pages;
965 int limit = 10 * 1024 * 1042;
967 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
968 1, 0, NULL, GFP_NOFS);
969 while (start < end) {
970 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
972 async_cow->inode = inode;
973 async_cow->root = root;
974 async_cow->locked_page = locked_page;
975 async_cow->start = start;
977 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
980 cur_end = min(end, start + 512 * 1024 - 1);
982 async_cow->end = cur_end;
983 INIT_LIST_HEAD(&async_cow->extents);
985 async_cow->work.func = async_cow_start;
986 async_cow->work.ordered_func = async_cow_submit;
987 async_cow->work.ordered_free = async_cow_free;
988 async_cow->work.flags = 0;
990 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
992 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
994 btrfs_queue_worker(&root->fs_info->delalloc_workers,
997 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
998 wait_event(root->fs_info->async_submit_wait,
999 (atomic_read(&root->fs_info->async_delalloc_pages) <
1003 while (atomic_read(&root->fs_info->async_submit_draining) &&
1004 atomic_read(&root->fs_info->async_delalloc_pages)) {
1005 wait_event(root->fs_info->async_submit_wait,
1006 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010 *nr_written += nr_pages;
1011 start = cur_end + 1;
1017 static noinline int csum_exist_in_range(struct btrfs_root *root,
1018 u64 bytenr, u64 num_bytes)
1021 struct btrfs_ordered_sum *sums;
1024 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1025 bytenr + num_bytes - 1, &list, 0);
1026 if (ret == 0 && list_empty(&list))
1029 while (!list_empty(&list)) {
1030 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031 list_del(&sums->list);
1038 * when nowcow writeback call back. This checks for snapshots or COW copies
1039 * of the extents that exist in the file, and COWs the file as required.
1041 * If no cow copies or snapshots exist, we write directly to the existing
1044 static noinline int run_delalloc_nocow(struct inode *inode,
1045 struct page *locked_page,
1046 u64 start, u64 end, int *page_started, int force,
1047 unsigned long *nr_written)
1049 struct btrfs_root *root = BTRFS_I(inode)->root;
1050 struct btrfs_trans_handle *trans;
1051 struct extent_buffer *leaf;
1052 struct btrfs_path *path;
1053 struct btrfs_file_extent_item *fi;
1054 struct btrfs_key found_key;
1067 u64 ino = btrfs_ino(inode);
1069 path = btrfs_alloc_path();
1073 nolock = btrfs_is_free_space_inode(root, inode);
1076 trans = btrfs_join_transaction_nolock(root);
1078 trans = btrfs_join_transaction(root);
1080 BUG_ON(IS_ERR(trans));
1081 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083 cow_start = (u64)-1;
1086 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0] - 1);
1093 if (found_key.objectid == ino &&
1094 found_key.type == BTRFS_EXTENT_DATA_KEY)
1099 leaf = path->nodes[0];
1100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101 ret = btrfs_next_leaf(root, path);
1106 leaf = path->nodes[0];
1112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114 if (found_key.objectid > ino ||
1115 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116 found_key.offset > end)
1119 if (found_key.offset > cur_offset) {
1120 extent_end = found_key.offset;
1125 fi = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_file_extent_item);
1127 extent_type = btrfs_file_extent_type(leaf, fi);
1129 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1131 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132 extent_offset = btrfs_file_extent_offset(leaf, fi);
1133 extent_end = found_key.offset +
1134 btrfs_file_extent_num_bytes(leaf, fi);
1135 if (extent_end <= start) {
1139 if (disk_bytenr == 0)
1141 if (btrfs_file_extent_compression(leaf, fi) ||
1142 btrfs_file_extent_encryption(leaf, fi) ||
1143 btrfs_file_extent_other_encoding(leaf, fi))
1145 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147 if (btrfs_extent_readonly(root, disk_bytenr))
1149 if (btrfs_cross_ref_exist(trans, root, ino,
1151 extent_offset, disk_bytenr))
1153 disk_bytenr += extent_offset;
1154 disk_bytenr += cur_offset - found_key.offset;
1155 num_bytes = min(end + 1, extent_end) - cur_offset;
1157 * force cow if csum exists in the range.
1158 * this ensure that csum for a given extent are
1159 * either valid or do not exist.
1161 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165 extent_end = found_key.offset +
1166 btrfs_file_extent_inline_len(leaf, fi);
1167 extent_end = ALIGN(extent_end, root->sectorsize);
1172 if (extent_end <= start) {
1177 if (cow_start == (u64)-1)
1178 cow_start = cur_offset;
1179 cur_offset = extent_end;
1180 if (cur_offset > end)
1186 btrfs_release_path(path);
1187 if (cow_start != (u64)-1) {
1188 ret = cow_file_range(inode, locked_page, cow_start,
1189 found_key.offset - 1, page_started,
1192 cow_start = (u64)-1;
1195 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196 struct extent_map *em;
1197 struct extent_map_tree *em_tree;
1198 em_tree = &BTRFS_I(inode)->extent_tree;
1199 em = alloc_extent_map();
1201 em->start = cur_offset;
1202 em->orig_start = em->start;
1203 em->len = num_bytes;
1204 em->block_len = num_bytes;
1205 em->block_start = disk_bytenr;
1206 em->bdev = root->fs_info->fs_devices->latest_bdev;
1207 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209 write_lock(&em_tree->lock);
1210 ret = add_extent_mapping(em_tree, em);
1211 write_unlock(&em_tree->lock);
1212 if (ret != -EEXIST) {
1213 free_extent_map(em);
1216 btrfs_drop_extent_cache(inode, em->start,
1217 em->start + em->len - 1, 0);
1219 type = BTRFS_ORDERED_PREALLOC;
1221 type = BTRFS_ORDERED_NOCOW;
1224 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1225 num_bytes, num_bytes, type);
1228 if (root->root_key.objectid ==
1229 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1235 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1236 cur_offset, cur_offset + num_bytes - 1,
1237 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239 EXTENT_SET_PRIVATE2);
1240 cur_offset = extent_end;
1241 if (cur_offset > end)
1244 btrfs_release_path(path);
1246 if (cur_offset <= end && cow_start == (u64)-1)
1247 cow_start = cur_offset;
1248 if (cow_start != (u64)-1) {
1249 ret = cow_file_range(inode, locked_page, cow_start, end,
1250 page_started, nr_written, 1);
1255 ret = btrfs_end_transaction_nolock(trans, root);
1258 ret = btrfs_end_transaction(trans, root);
1261 btrfs_free_path(path);
1266 * extent_io.c call back to do delayed allocation processing
1268 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1269 u64 start, u64 end, int *page_started,
1270 unsigned long *nr_written)
1273 struct btrfs_root *root = BTRFS_I(inode)->root;
1275 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1276 ret = run_delalloc_nocow(inode, locked_page, start, end,
1277 page_started, 1, nr_written);
1278 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1279 ret = run_delalloc_nocow(inode, locked_page, start, end,
1280 page_started, 0, nr_written);
1281 else if (!btrfs_test_opt(root, COMPRESS) &&
1282 !(BTRFS_I(inode)->force_compress) &&
1283 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1284 ret = cow_file_range(inode, locked_page, start, end,
1285 page_started, nr_written, 1);
1287 ret = cow_file_range_async(inode, locked_page, start, end,
1288 page_started, nr_written);
1292 static void btrfs_split_extent_hook(struct inode *inode,
1293 struct extent_state *orig, u64 split)
1295 /* not delalloc, ignore it */
1296 if (!(orig->state & EXTENT_DELALLOC))
1299 spin_lock(&BTRFS_I(inode)->lock);
1300 BTRFS_I(inode)->outstanding_extents++;
1301 spin_unlock(&BTRFS_I(inode)->lock);
1305 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306 * extents so we can keep track of new extents that are just merged onto old
1307 * extents, such as when we are doing sequential writes, so we can properly
1308 * account for the metadata space we'll need.
1310 static void btrfs_merge_extent_hook(struct inode *inode,
1311 struct extent_state *new,
1312 struct extent_state *other)
1314 /* not delalloc, ignore it */
1315 if (!(other->state & EXTENT_DELALLOC))
1318 spin_lock(&BTRFS_I(inode)->lock);
1319 BTRFS_I(inode)->outstanding_extents--;
1320 spin_unlock(&BTRFS_I(inode)->lock);
1324 * extent_io.c set_bit_hook, used to track delayed allocation
1325 * bytes in this file, and to maintain the list of inodes that
1326 * have pending delalloc work to be done.
1328 static void btrfs_set_bit_hook(struct inode *inode,
1329 struct extent_state *state, int *bits)
1333 * set_bit and clear bit hooks normally require _irqsave/restore
1334 * but in this case, we are only testing for the DELALLOC
1335 * bit, which is only set or cleared with irqs on
1337 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338 struct btrfs_root *root = BTRFS_I(inode)->root;
1339 u64 len = state->end + 1 - state->start;
1340 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342 if (*bits & EXTENT_FIRST_DELALLOC) {
1343 *bits &= ~EXTENT_FIRST_DELALLOC;
1345 spin_lock(&BTRFS_I(inode)->lock);
1346 BTRFS_I(inode)->outstanding_extents++;
1347 spin_unlock(&BTRFS_I(inode)->lock);
1350 spin_lock(&root->fs_info->delalloc_lock);
1351 BTRFS_I(inode)->delalloc_bytes += len;
1352 root->fs_info->delalloc_bytes += len;
1353 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1354 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355 &root->fs_info->delalloc_inodes);
1357 spin_unlock(&root->fs_info->delalloc_lock);
1362 * extent_io.c clear_bit_hook, see set_bit_hook for why
1364 static void btrfs_clear_bit_hook(struct inode *inode,
1365 struct extent_state *state, int *bits)
1368 * set_bit and clear bit hooks normally require _irqsave/restore
1369 * but in this case, we are only testing for the DELALLOC
1370 * bit, which is only set or cleared with irqs on
1372 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1373 struct btrfs_root *root = BTRFS_I(inode)->root;
1374 u64 len = state->end + 1 - state->start;
1375 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377 if (*bits & EXTENT_FIRST_DELALLOC) {
1378 *bits &= ~EXTENT_FIRST_DELALLOC;
1379 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380 spin_lock(&BTRFS_I(inode)->lock);
1381 BTRFS_I(inode)->outstanding_extents--;
1382 spin_unlock(&BTRFS_I(inode)->lock);
1385 if (*bits & EXTENT_DO_ACCOUNTING)
1386 btrfs_delalloc_release_metadata(inode, len);
1388 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390 btrfs_free_reserved_data_space(inode, len);
1392 spin_lock(&root->fs_info->delalloc_lock);
1393 root->fs_info->delalloc_bytes -= len;
1394 BTRFS_I(inode)->delalloc_bytes -= len;
1396 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1397 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400 spin_unlock(&root->fs_info->delalloc_lock);
1405 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406 * we don't create bios that span stripes or chunks
1408 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1409 size_t size, struct bio *bio,
1410 unsigned long bio_flags)
1412 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413 struct btrfs_mapping_tree *map_tree;
1414 u64 logical = (u64)bio->bi_sector << 9;
1419 if (bio_flags & EXTENT_BIO_COMPRESSED)
1422 length = bio->bi_size;
1423 map_tree = &root->fs_info->mapping_tree;
1424 map_length = length;
1425 ret = btrfs_map_block(map_tree, READ, logical,
1426 &map_length, NULL, 0);
1428 if (map_length < length + size)
1434 * in order to insert checksums into the metadata in large chunks,
1435 * we wait until bio submission time. All the pages in the bio are
1436 * checksummed and sums are attached onto the ordered extent record.
1438 * At IO completion time the cums attached on the ordered extent record
1439 * are inserted into the btree
1441 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442 struct bio *bio, int mirror_num,
1443 unsigned long bio_flags,
1446 struct btrfs_root *root = BTRFS_I(inode)->root;
1449 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1455 * in order to insert checksums into the metadata in large chunks,
1456 * we wait until bio submission time. All the pages in the bio are
1457 * checksummed and sums are attached onto the ordered extent record.
1459 * At IO completion time the cums attached on the ordered extent record
1460 * are inserted into the btree
1462 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1463 int mirror_num, unsigned long bio_flags,
1466 struct btrfs_root *root = BTRFS_I(inode)->root;
1467 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1471 * extent_io.c submission hook. This does the right thing for csum calculation
1472 * on write, or reading the csums from the tree before a read
1474 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1475 int mirror_num, unsigned long bio_flags,
1478 struct btrfs_root *root = BTRFS_I(inode)->root;
1482 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1484 if (btrfs_is_free_space_inode(root, inode))
1485 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1487 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1490 if (!(rw & REQ_WRITE)) {
1491 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1492 return btrfs_submit_compressed_read(inode, bio,
1493 mirror_num, bio_flags);
1494 } else if (!skip_sum) {
1495 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1500 } else if (!skip_sum) {
1501 /* csum items have already been cloned */
1502 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1504 /* we're doing a write, do the async checksumming */
1505 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1506 inode, rw, bio, mirror_num,
1507 bio_flags, bio_offset,
1508 __btrfs_submit_bio_start,
1509 __btrfs_submit_bio_done);
1513 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1517 * given a list of ordered sums record them in the inode. This happens
1518 * at IO completion time based on sums calculated at bio submission time.
1520 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1521 struct inode *inode, u64 file_offset,
1522 struct list_head *list)
1524 struct btrfs_ordered_sum *sum;
1526 list_for_each_entry(sum, list, list) {
1527 btrfs_csum_file_blocks(trans,
1528 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1533 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534 struct extent_state **cached_state)
1536 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1538 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1539 cached_state, GFP_NOFS);
1542 /* see btrfs_writepage_start_hook for details on why this is required */
1543 struct btrfs_writepage_fixup {
1545 struct btrfs_work work;
1548 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1550 struct btrfs_writepage_fixup *fixup;
1551 struct btrfs_ordered_extent *ordered;
1552 struct extent_state *cached_state = NULL;
1554 struct inode *inode;
1558 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1562 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1563 ClearPageChecked(page);
1567 inode = page->mapping->host;
1568 page_start = page_offset(page);
1569 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1571 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1572 &cached_state, GFP_NOFS);
1574 /* already ordered? We're done */
1575 if (PagePrivate2(page))
1578 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1580 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1581 page_end, &cached_state, GFP_NOFS);
1583 btrfs_start_ordered_extent(inode, ordered, 1);
1588 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1589 ClearPageChecked(page);
1591 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1592 &cached_state, GFP_NOFS);
1595 page_cache_release(page);
1600 * There are a few paths in the higher layers of the kernel that directly
1601 * set the page dirty bit without asking the filesystem if it is a
1602 * good idea. This causes problems because we want to make sure COW
1603 * properly happens and the data=ordered rules are followed.
1605 * In our case any range that doesn't have the ORDERED bit set
1606 * hasn't been properly setup for IO. We kick off an async process
1607 * to fix it up. The async helper will wait for ordered extents, set
1608 * the delalloc bit and make it safe to write the page.
1610 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1612 struct inode *inode = page->mapping->host;
1613 struct btrfs_writepage_fixup *fixup;
1614 struct btrfs_root *root = BTRFS_I(inode)->root;
1616 /* this page is properly in the ordered list */
1617 if (TestClearPagePrivate2(page))
1620 if (PageChecked(page))
1623 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1627 SetPageChecked(page);
1628 page_cache_get(page);
1629 fixup->work.func = btrfs_writepage_fixup_worker;
1631 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1635 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1636 struct inode *inode, u64 file_pos,
1637 u64 disk_bytenr, u64 disk_num_bytes,
1638 u64 num_bytes, u64 ram_bytes,
1639 u8 compression, u8 encryption,
1640 u16 other_encoding, int extent_type)
1642 struct btrfs_root *root = BTRFS_I(inode)->root;
1643 struct btrfs_file_extent_item *fi;
1644 struct btrfs_path *path;
1645 struct extent_buffer *leaf;
1646 struct btrfs_key ins;
1650 path = btrfs_alloc_path();
1654 path->leave_spinning = 1;
1657 * we may be replacing one extent in the tree with another.
1658 * The new extent is pinned in the extent map, and we don't want
1659 * to drop it from the cache until it is completely in the btree.
1661 * So, tell btrfs_drop_extents to leave this extent in the cache.
1662 * the caller is expected to unpin it and allow it to be merged
1665 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1669 ins.objectid = btrfs_ino(inode);
1670 ins.offset = file_pos;
1671 ins.type = BTRFS_EXTENT_DATA_KEY;
1672 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1674 leaf = path->nodes[0];
1675 fi = btrfs_item_ptr(leaf, path->slots[0],
1676 struct btrfs_file_extent_item);
1677 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1678 btrfs_set_file_extent_type(leaf, fi, extent_type);
1679 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1680 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1681 btrfs_set_file_extent_offset(leaf, fi, 0);
1682 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1683 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1684 btrfs_set_file_extent_compression(leaf, fi, compression);
1685 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1686 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1688 btrfs_unlock_up_safe(path, 1);
1689 btrfs_set_lock_blocking(leaf);
1691 btrfs_mark_buffer_dirty(leaf);
1693 inode_add_bytes(inode, num_bytes);
1695 ins.objectid = disk_bytenr;
1696 ins.offset = disk_num_bytes;
1697 ins.type = BTRFS_EXTENT_ITEM_KEY;
1698 ret = btrfs_alloc_reserved_file_extent(trans, root,
1699 root->root_key.objectid,
1700 btrfs_ino(inode), file_pos, &ins);
1702 btrfs_free_path(path);
1708 * helper function for btrfs_finish_ordered_io, this
1709 * just reads in some of the csum leaves to prime them into ram
1710 * before we start the transaction. It limits the amount of btree
1711 * reads required while inside the transaction.
1713 /* as ordered data IO finishes, this gets called so we can finish
1714 * an ordered extent if the range of bytes in the file it covers are
1717 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1719 struct btrfs_root *root = BTRFS_I(inode)->root;
1720 struct btrfs_trans_handle *trans = NULL;
1721 struct btrfs_ordered_extent *ordered_extent = NULL;
1722 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1723 struct extent_state *cached_state = NULL;
1724 int compress_type = 0;
1728 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1732 BUG_ON(!ordered_extent);
1734 nolock = btrfs_is_free_space_inode(root, inode);
1736 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1737 BUG_ON(!list_empty(&ordered_extent->list));
1738 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1741 trans = btrfs_join_transaction_nolock(root);
1743 trans = btrfs_join_transaction(root);
1744 BUG_ON(IS_ERR(trans));
1745 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1746 ret = btrfs_update_inode_fallback(trans, root, inode);
1752 lock_extent_bits(io_tree, ordered_extent->file_offset,
1753 ordered_extent->file_offset + ordered_extent->len - 1,
1754 0, &cached_state, GFP_NOFS);
1757 trans = btrfs_join_transaction_nolock(root);
1759 trans = btrfs_join_transaction(root);
1760 BUG_ON(IS_ERR(trans));
1761 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1763 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1764 compress_type = ordered_extent->compress_type;
1765 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1766 BUG_ON(compress_type);
1767 ret = btrfs_mark_extent_written(trans, inode,
1768 ordered_extent->file_offset,
1769 ordered_extent->file_offset +
1770 ordered_extent->len);
1773 BUG_ON(root == root->fs_info->tree_root);
1774 ret = insert_reserved_file_extent(trans, inode,
1775 ordered_extent->file_offset,
1776 ordered_extent->start,
1777 ordered_extent->disk_len,
1778 ordered_extent->len,
1779 ordered_extent->len,
1780 compress_type, 0, 0,
1781 BTRFS_FILE_EXTENT_REG);
1782 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1783 ordered_extent->file_offset,
1784 ordered_extent->len);
1787 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1788 ordered_extent->file_offset +
1789 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1791 add_pending_csums(trans, inode, ordered_extent->file_offset,
1792 &ordered_extent->list);
1794 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1795 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1796 ret = btrfs_update_inode_fallback(trans, root, inode);
1801 if (root != root->fs_info->tree_root)
1802 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1805 btrfs_end_transaction_nolock(trans, root);
1807 btrfs_end_transaction(trans, root);
1811 btrfs_put_ordered_extent(ordered_extent);
1812 /* once for the tree */
1813 btrfs_put_ordered_extent(ordered_extent);
1818 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1819 struct extent_state *state, int uptodate)
1821 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1823 ClearPagePrivate2(page);
1824 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1828 * when reads are done, we need to check csums to verify the data is correct
1829 * if there's a match, we allow the bio to finish. If not, the code in
1830 * extent_io.c will try to find good copies for us.
1832 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1833 struct extent_state *state)
1835 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1836 struct inode *inode = page->mapping->host;
1837 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1839 u64 private = ~(u32)0;
1841 struct btrfs_root *root = BTRFS_I(inode)->root;
1844 if (PageChecked(page)) {
1845 ClearPageChecked(page);
1849 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1852 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1853 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1854 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1859 if (state && state->start == start) {
1860 private = state->private;
1863 ret = get_state_private(io_tree, start, &private);
1865 kaddr = kmap_atomic(page, KM_USER0);
1869 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1870 btrfs_csum_final(csum, (char *)&csum);
1871 if (csum != private)
1874 kunmap_atomic(kaddr, KM_USER0);
1879 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1881 (unsigned long long)btrfs_ino(page->mapping->host),
1882 (unsigned long long)start, csum,
1883 (unsigned long long)private);
1884 memset(kaddr + offset, 1, end - start + 1);
1885 flush_dcache_page(page);
1886 kunmap_atomic(kaddr, KM_USER0);
1892 struct delayed_iput {
1893 struct list_head list;
1894 struct inode *inode;
1897 void btrfs_add_delayed_iput(struct inode *inode)
1899 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1900 struct delayed_iput *delayed;
1902 if (atomic_add_unless(&inode->i_count, -1, 1))
1905 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1906 delayed->inode = inode;
1908 spin_lock(&fs_info->delayed_iput_lock);
1909 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1910 spin_unlock(&fs_info->delayed_iput_lock);
1913 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1916 struct btrfs_fs_info *fs_info = root->fs_info;
1917 struct delayed_iput *delayed;
1920 spin_lock(&fs_info->delayed_iput_lock);
1921 empty = list_empty(&fs_info->delayed_iputs);
1922 spin_unlock(&fs_info->delayed_iput_lock);
1926 down_read(&root->fs_info->cleanup_work_sem);
1927 spin_lock(&fs_info->delayed_iput_lock);
1928 list_splice_init(&fs_info->delayed_iputs, &list);
1929 spin_unlock(&fs_info->delayed_iput_lock);
1931 while (!list_empty(&list)) {
1932 delayed = list_entry(list.next, struct delayed_iput, list);
1933 list_del(&delayed->list);
1934 iput(delayed->inode);
1937 up_read(&root->fs_info->cleanup_work_sem);
1940 enum btrfs_orphan_cleanup_state {
1941 ORPHAN_CLEANUP_STARTED = 1,
1942 ORPHAN_CLEANUP_DONE = 2,
1946 * This is called in transaction commmit time. If there are no orphan
1947 * files in the subvolume, it removes orphan item and frees block_rsv
1950 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1951 struct btrfs_root *root)
1955 if (!list_empty(&root->orphan_list) ||
1956 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1959 if (root->orphan_item_inserted &&
1960 btrfs_root_refs(&root->root_item) > 0) {
1961 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1962 root->root_key.objectid);
1964 root->orphan_item_inserted = 0;
1967 if (root->orphan_block_rsv) {
1968 WARN_ON(root->orphan_block_rsv->size > 0);
1969 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1970 root->orphan_block_rsv = NULL;
1975 * This creates an orphan entry for the given inode in case something goes
1976 * wrong in the middle of an unlink/truncate.
1978 * NOTE: caller of this function should reserve 5 units of metadata for
1981 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1983 struct btrfs_root *root = BTRFS_I(inode)->root;
1984 struct btrfs_block_rsv *block_rsv = NULL;
1989 if (!root->orphan_block_rsv) {
1990 block_rsv = btrfs_alloc_block_rsv(root);
1995 spin_lock(&root->orphan_lock);
1996 if (!root->orphan_block_rsv) {
1997 root->orphan_block_rsv = block_rsv;
1998 } else if (block_rsv) {
1999 btrfs_free_block_rsv(root, block_rsv);
2003 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2004 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2007 * For proper ENOSPC handling, we should do orphan
2008 * cleanup when mounting. But this introduces backward
2009 * compatibility issue.
2011 if (!xchg(&root->orphan_item_inserted, 1))
2019 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2020 BTRFS_I(inode)->orphan_meta_reserved = 1;
2023 spin_unlock(&root->orphan_lock);
2025 /* grab metadata reservation from transaction handle */
2027 ret = btrfs_orphan_reserve_metadata(trans, inode);
2031 /* insert an orphan item to track this unlinked/truncated file */
2033 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2037 /* insert an orphan item to track subvolume contains orphan files */
2039 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2040 root->root_key.objectid);
2047 * We have done the truncate/delete so we can go ahead and remove the orphan
2048 * item for this particular inode.
2050 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
2053 int delete_item = 0;
2054 int release_rsv = 0;
2057 spin_lock(&root->orphan_lock);
2058 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2059 list_del_init(&BTRFS_I(inode)->i_orphan);
2063 if (BTRFS_I(inode)->orphan_meta_reserved) {
2064 BTRFS_I(inode)->orphan_meta_reserved = 0;
2067 spin_unlock(&root->orphan_lock);
2069 if (trans && delete_item) {
2070 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2075 btrfs_orphan_release_metadata(inode);
2081 * this cleans up any orphans that may be left on the list from the last use
2084 int btrfs_orphan_cleanup(struct btrfs_root *root)
2086 struct btrfs_path *path;
2087 struct extent_buffer *leaf;
2088 struct btrfs_key key, found_key;
2089 struct btrfs_trans_handle *trans;
2090 struct inode *inode;
2091 u64 last_objectid = 0;
2092 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2094 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2097 path = btrfs_alloc_path();
2104 key.objectid = BTRFS_ORPHAN_OBJECTID;
2105 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2106 key.offset = (u64)-1;
2109 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2114 * if ret == 0 means we found what we were searching for, which
2115 * is weird, but possible, so only screw with path if we didn't
2116 * find the key and see if we have stuff that matches
2120 if (path->slots[0] == 0)
2125 /* pull out the item */
2126 leaf = path->nodes[0];
2127 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2129 /* make sure the item matches what we want */
2130 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2132 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2135 /* release the path since we're done with it */
2136 btrfs_release_path(path);
2139 * this is where we are basically btrfs_lookup, without the
2140 * crossing root thing. we store the inode number in the
2141 * offset of the orphan item.
2144 if (found_key.offset == last_objectid) {
2145 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2146 "stopping orphan cleanup\n");
2151 last_objectid = found_key.offset;
2153 found_key.objectid = found_key.offset;
2154 found_key.type = BTRFS_INODE_ITEM_KEY;
2155 found_key.offset = 0;
2156 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2157 ret = PTR_RET(inode);
2158 if (ret && ret != -ESTALE)
2162 * Inode is already gone but the orphan item is still there,
2163 * kill the orphan item.
2165 if (ret == -ESTALE) {
2166 trans = btrfs_start_transaction(root, 1);
2167 if (IS_ERR(trans)) {
2168 ret = PTR_ERR(trans);
2171 ret = btrfs_del_orphan_item(trans, root,
2172 found_key.objectid);
2174 btrfs_end_transaction(trans, root);
2179 * add this inode to the orphan list so btrfs_orphan_del does
2180 * the proper thing when we hit it
2182 spin_lock(&root->orphan_lock);
2183 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2184 spin_unlock(&root->orphan_lock);
2186 /* if we have links, this was a truncate, lets do that */
2187 if (inode->i_nlink) {
2188 if (!S_ISREG(inode->i_mode)) {
2194 ret = btrfs_truncate(inode);
2199 /* this will do delete_inode and everything for us */
2204 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2206 if (root->orphan_block_rsv)
2207 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2210 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2211 trans = btrfs_join_transaction(root);
2213 btrfs_end_transaction(trans, root);
2217 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2219 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2223 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2224 btrfs_free_path(path);
2229 * very simple check to peek ahead in the leaf looking for xattrs. If we
2230 * don't find any xattrs, we know there can't be any acls.
2232 * slot is the slot the inode is in, objectid is the objectid of the inode
2234 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2235 int slot, u64 objectid)
2237 u32 nritems = btrfs_header_nritems(leaf);
2238 struct btrfs_key found_key;
2242 while (slot < nritems) {
2243 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2245 /* we found a different objectid, there must not be acls */
2246 if (found_key.objectid != objectid)
2249 /* we found an xattr, assume we've got an acl */
2250 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2254 * we found a key greater than an xattr key, there can't
2255 * be any acls later on
2257 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2264 * it goes inode, inode backrefs, xattrs, extents,
2265 * so if there are a ton of hard links to an inode there can
2266 * be a lot of backrefs. Don't waste time searching too hard,
2267 * this is just an optimization
2272 /* we hit the end of the leaf before we found an xattr or
2273 * something larger than an xattr. We have to assume the inode
2280 * read an inode from the btree into the in-memory inode
2282 static void btrfs_read_locked_inode(struct inode *inode)
2284 struct btrfs_path *path;
2285 struct extent_buffer *leaf;
2286 struct btrfs_inode_item *inode_item;
2287 struct btrfs_timespec *tspec;
2288 struct btrfs_root *root = BTRFS_I(inode)->root;
2289 struct btrfs_key location;
2293 bool filled = false;
2295 ret = btrfs_fill_inode(inode, &rdev);
2299 path = btrfs_alloc_path();
2303 path->leave_spinning = 1;
2304 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2306 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2310 leaf = path->nodes[0];
2315 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2316 struct btrfs_inode_item);
2317 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2318 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2319 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2320 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2321 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2323 tspec = btrfs_inode_atime(inode_item);
2324 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2325 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2327 tspec = btrfs_inode_mtime(inode_item);
2328 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2329 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2331 tspec = btrfs_inode_ctime(inode_item);
2332 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2333 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2335 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2336 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2337 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2338 inode->i_generation = BTRFS_I(inode)->generation;
2340 rdev = btrfs_inode_rdev(leaf, inode_item);
2342 BTRFS_I(inode)->index_cnt = (u64)-1;
2343 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2346 * try to precache a NULL acl entry for files that don't have
2347 * any xattrs or acls
2349 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2352 cache_no_acl(inode);
2354 btrfs_free_path(path);
2356 switch (inode->i_mode & S_IFMT) {
2358 inode->i_mapping->a_ops = &btrfs_aops;
2359 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2360 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2361 inode->i_fop = &btrfs_file_operations;
2362 inode->i_op = &btrfs_file_inode_operations;
2365 inode->i_fop = &btrfs_dir_file_operations;
2366 if (root == root->fs_info->tree_root)
2367 inode->i_op = &btrfs_dir_ro_inode_operations;
2369 inode->i_op = &btrfs_dir_inode_operations;
2372 inode->i_op = &btrfs_symlink_inode_operations;
2373 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2374 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2377 inode->i_op = &btrfs_special_inode_operations;
2378 init_special_inode(inode, inode->i_mode, rdev);
2382 btrfs_update_iflags(inode);
2386 btrfs_free_path(path);
2387 make_bad_inode(inode);
2391 * given a leaf and an inode, copy the inode fields into the leaf
2393 static void fill_inode_item(struct btrfs_trans_handle *trans,
2394 struct extent_buffer *leaf,
2395 struct btrfs_inode_item *item,
2396 struct inode *inode)
2398 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2399 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2400 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2401 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2402 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2404 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2405 inode->i_atime.tv_sec);
2406 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2407 inode->i_atime.tv_nsec);
2409 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2410 inode->i_mtime.tv_sec);
2411 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2412 inode->i_mtime.tv_nsec);
2414 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2415 inode->i_ctime.tv_sec);
2416 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2417 inode->i_ctime.tv_nsec);
2419 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2420 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2421 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2422 btrfs_set_inode_transid(leaf, item, trans->transid);
2423 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2424 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2425 btrfs_set_inode_block_group(leaf, item, 0);
2429 * copy everything in the in-memory inode into the btree.
2431 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2432 struct btrfs_root *root, struct inode *inode)
2434 struct btrfs_inode_item *inode_item;
2435 struct btrfs_path *path;
2436 struct extent_buffer *leaf;
2439 path = btrfs_alloc_path();
2443 path->leave_spinning = 1;
2444 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2452 btrfs_unlock_up_safe(path, 1);
2453 leaf = path->nodes[0];
2454 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2455 struct btrfs_inode_item);
2457 fill_inode_item(trans, leaf, inode_item, inode);
2458 btrfs_mark_buffer_dirty(leaf);
2459 btrfs_set_inode_last_trans(trans, inode);
2462 btrfs_free_path(path);
2467 * copy everything in the in-memory inode into the btree.
2469 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2470 struct btrfs_root *root, struct inode *inode)
2475 * If the inode is a free space inode, we can deadlock during commit
2476 * if we put it into the delayed code.
2478 * The data relocation inode should also be directly updated
2481 if (!btrfs_is_free_space_inode(root, inode)
2482 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2483 ret = btrfs_delayed_update_inode(trans, root, inode);
2485 btrfs_set_inode_last_trans(trans, inode);
2489 return btrfs_update_inode_item(trans, root, inode);
2492 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2493 struct btrfs_root *root, struct inode *inode)
2497 ret = btrfs_update_inode(trans, root, inode);
2499 return btrfs_update_inode_item(trans, root, inode);
2504 * unlink helper that gets used here in inode.c and in the tree logging
2505 * recovery code. It remove a link in a directory with a given name, and
2506 * also drops the back refs in the inode to the directory
2508 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 struct inode *dir, struct inode *inode,
2511 const char *name, int name_len)
2513 struct btrfs_path *path;
2515 struct extent_buffer *leaf;
2516 struct btrfs_dir_item *di;
2517 struct btrfs_key key;
2519 u64 ino = btrfs_ino(inode);
2520 u64 dir_ino = btrfs_ino(dir);
2522 path = btrfs_alloc_path();
2528 path->leave_spinning = 1;
2529 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2530 name, name_len, -1);
2539 leaf = path->nodes[0];
2540 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2541 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2544 btrfs_release_path(path);
2546 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2549 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2550 "inode %llu parent %llu\n", name_len, name,
2551 (unsigned long long)ino, (unsigned long long)dir_ino);
2555 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2559 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2561 BUG_ON(ret != 0 && ret != -ENOENT);
2563 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2568 btrfs_free_path(path);
2572 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2573 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2574 btrfs_update_inode(trans, root, dir);
2579 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2580 struct btrfs_root *root,
2581 struct inode *dir, struct inode *inode,
2582 const char *name, int name_len)
2585 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2587 btrfs_drop_nlink(inode);
2588 ret = btrfs_update_inode(trans, root, inode);
2594 /* helper to check if there is any shared block in the path */
2595 static int check_path_shared(struct btrfs_root *root,
2596 struct btrfs_path *path)
2598 struct extent_buffer *eb;
2602 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2605 if (!path->nodes[level])
2607 eb = path->nodes[level];
2608 if (!btrfs_block_can_be_shared(root, eb))
2610 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2619 * helper to start transaction for unlink and rmdir.
2621 * unlink and rmdir are special in btrfs, they do not always free space.
2622 * so in enospc case, we should make sure they will free space before
2623 * allowing them to use the global metadata reservation.
2625 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2626 struct dentry *dentry)
2628 struct btrfs_trans_handle *trans;
2629 struct btrfs_root *root = BTRFS_I(dir)->root;
2630 struct btrfs_path *path;
2631 struct btrfs_inode_ref *ref;
2632 struct btrfs_dir_item *di;
2633 struct inode *inode = dentry->d_inode;
2638 u64 ino = btrfs_ino(inode);
2639 u64 dir_ino = btrfs_ino(dir);
2642 * 1 for the possible orphan item
2643 * 1 for the dir item
2644 * 1 for the dir index
2645 * 1 for the inode ref
2646 * 1 for the inode ref in the tree log
2647 * 2 for the dir entries in the log
2650 trans = btrfs_start_transaction(root, 8);
2651 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2654 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2655 return ERR_PTR(-ENOSPC);
2657 /* check if there is someone else holds reference */
2658 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2659 return ERR_PTR(-ENOSPC);
2661 if (atomic_read(&inode->i_count) > 2)
2662 return ERR_PTR(-ENOSPC);
2664 if (xchg(&root->fs_info->enospc_unlink, 1))
2665 return ERR_PTR(-ENOSPC);
2667 path = btrfs_alloc_path();
2669 root->fs_info->enospc_unlink = 0;
2670 return ERR_PTR(-ENOMEM);
2673 /* 1 for the orphan item */
2674 trans = btrfs_start_transaction(root, 1);
2675 if (IS_ERR(trans)) {
2676 btrfs_free_path(path);
2677 root->fs_info->enospc_unlink = 0;
2681 path->skip_locking = 1;
2682 path->search_commit_root = 1;
2684 ret = btrfs_lookup_inode(trans, root, path,
2685 &BTRFS_I(dir)->location, 0);
2691 if (check_path_shared(root, path))
2696 btrfs_release_path(path);
2698 ret = btrfs_lookup_inode(trans, root, path,
2699 &BTRFS_I(inode)->location, 0);
2705 if (check_path_shared(root, path))
2710 btrfs_release_path(path);
2712 if (ret == 0 && S_ISREG(inode->i_mode)) {
2713 ret = btrfs_lookup_file_extent(trans, root, path,
2720 if (check_path_shared(root, path))
2722 btrfs_release_path(path);
2730 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2731 dentry->d_name.name, dentry->d_name.len, 0);
2737 if (check_path_shared(root, path))
2743 btrfs_release_path(path);
2745 ref = btrfs_lookup_inode_ref(trans, root, path,
2746 dentry->d_name.name, dentry->d_name.len,
2753 if (check_path_shared(root, path))
2755 index = btrfs_inode_ref_index(path->nodes[0], ref);
2756 btrfs_release_path(path);
2759 * This is a commit root search, if we can lookup inode item and other
2760 * relative items in the commit root, it means the transaction of
2761 * dir/file creation has been committed, and the dir index item that we
2762 * delay to insert has also been inserted into the commit root. So
2763 * we needn't worry about the delayed insertion of the dir index item
2766 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2767 dentry->d_name.name, dentry->d_name.len, 0);
2772 BUG_ON(ret == -ENOENT);
2773 if (check_path_shared(root, path))
2778 btrfs_free_path(path);
2779 /* Migrate the orphan reservation over */
2781 err = btrfs_block_rsv_migrate(trans->block_rsv,
2782 &root->fs_info->global_block_rsv,
2783 trans->bytes_reserved);
2786 btrfs_end_transaction(trans, root);
2787 root->fs_info->enospc_unlink = 0;
2788 return ERR_PTR(err);
2791 trans->block_rsv = &root->fs_info->global_block_rsv;
2795 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2796 struct btrfs_root *root)
2798 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2799 btrfs_block_rsv_release(root, trans->block_rsv,
2800 trans->bytes_reserved);
2801 trans->block_rsv = &root->fs_info->trans_block_rsv;
2802 BUG_ON(!root->fs_info->enospc_unlink);
2803 root->fs_info->enospc_unlink = 0;
2805 btrfs_end_transaction_throttle(trans, root);
2808 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2810 struct btrfs_root *root = BTRFS_I(dir)->root;
2811 struct btrfs_trans_handle *trans;
2812 struct inode *inode = dentry->d_inode;
2814 unsigned long nr = 0;
2816 trans = __unlink_start_trans(dir, dentry);
2818 return PTR_ERR(trans);
2820 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2822 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2823 dentry->d_name.name, dentry->d_name.len);
2827 if (inode->i_nlink == 0) {
2828 ret = btrfs_orphan_add(trans, inode);
2834 nr = trans->blocks_used;
2835 __unlink_end_trans(trans, root);
2836 btrfs_btree_balance_dirty(root, nr);
2840 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2841 struct btrfs_root *root,
2842 struct inode *dir, u64 objectid,
2843 const char *name, int name_len)
2845 struct btrfs_path *path;
2846 struct extent_buffer *leaf;
2847 struct btrfs_dir_item *di;
2848 struct btrfs_key key;
2851 u64 dir_ino = btrfs_ino(dir);
2853 path = btrfs_alloc_path();
2857 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2858 name, name_len, -1);
2859 BUG_ON(IS_ERR_OR_NULL(di));
2861 leaf = path->nodes[0];
2862 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2863 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2864 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2866 btrfs_release_path(path);
2868 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2869 objectid, root->root_key.objectid,
2870 dir_ino, &index, name, name_len);
2872 BUG_ON(ret != -ENOENT);
2873 di = btrfs_search_dir_index_item(root, path, dir_ino,
2875 BUG_ON(IS_ERR_OR_NULL(di));
2877 leaf = path->nodes[0];
2878 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2879 btrfs_release_path(path);
2882 btrfs_release_path(path);
2884 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2887 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2888 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2889 ret = btrfs_update_inode(trans, root, dir);
2892 btrfs_free_path(path);
2896 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2898 struct inode *inode = dentry->d_inode;
2900 struct btrfs_root *root = BTRFS_I(dir)->root;
2901 struct btrfs_trans_handle *trans;
2902 unsigned long nr = 0;
2904 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2905 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2908 trans = __unlink_start_trans(dir, dentry);
2910 return PTR_ERR(trans);
2912 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2913 err = btrfs_unlink_subvol(trans, root, dir,
2914 BTRFS_I(inode)->location.objectid,
2915 dentry->d_name.name,
2916 dentry->d_name.len);
2920 err = btrfs_orphan_add(trans, inode);
2924 /* now the directory is empty */
2925 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2926 dentry->d_name.name, dentry->d_name.len);
2928 btrfs_i_size_write(inode, 0);
2930 nr = trans->blocks_used;
2931 __unlink_end_trans(trans, root);
2932 btrfs_btree_balance_dirty(root, nr);
2938 * this can truncate away extent items, csum items and directory items.
2939 * It starts at a high offset and removes keys until it can't find
2940 * any higher than new_size
2942 * csum items that cross the new i_size are truncated to the new size
2945 * min_type is the minimum key type to truncate down to. If set to 0, this
2946 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2948 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2949 struct btrfs_root *root,
2950 struct inode *inode,
2951 u64 new_size, u32 min_type)
2953 struct btrfs_path *path;
2954 struct extent_buffer *leaf;
2955 struct btrfs_file_extent_item *fi;
2956 struct btrfs_key key;
2957 struct btrfs_key found_key;
2958 u64 extent_start = 0;
2959 u64 extent_num_bytes = 0;
2960 u64 extent_offset = 0;
2962 u64 mask = root->sectorsize - 1;
2963 u32 found_type = (u8)-1;
2966 int pending_del_nr = 0;
2967 int pending_del_slot = 0;
2968 int extent_type = -1;
2972 u64 ino = btrfs_ino(inode);
2974 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2976 path = btrfs_alloc_path();
2981 if (root->ref_cows || root == root->fs_info->tree_root)
2982 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2985 * This function is also used to drop the items in the log tree before
2986 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2987 * it is used to drop the loged items. So we shouldn't kill the delayed
2990 if (min_type == 0 && root == BTRFS_I(inode)->root)
2991 btrfs_kill_delayed_inode_items(inode);
2994 key.offset = (u64)-1;
2998 path->leave_spinning = 1;
2999 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3006 /* there are no items in the tree for us to truncate, we're
3009 if (path->slots[0] == 0)
3016 leaf = path->nodes[0];
3017 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3018 found_type = btrfs_key_type(&found_key);
3021 if (found_key.objectid != ino)
3024 if (found_type < min_type)
3027 item_end = found_key.offset;
3028 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3029 fi = btrfs_item_ptr(leaf, path->slots[0],
3030 struct btrfs_file_extent_item);
3031 extent_type = btrfs_file_extent_type(leaf, fi);
3032 encoding = btrfs_file_extent_compression(leaf, fi);
3033 encoding |= btrfs_file_extent_encryption(leaf, fi);
3034 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3036 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3038 btrfs_file_extent_num_bytes(leaf, fi);
3039 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3040 item_end += btrfs_file_extent_inline_len(leaf,
3045 if (found_type > min_type) {
3048 if (item_end < new_size)
3050 if (found_key.offset >= new_size)
3056 /* FIXME, shrink the extent if the ref count is only 1 */
3057 if (found_type != BTRFS_EXTENT_DATA_KEY)
3060 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3062 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3063 if (!del_item && !encoding) {
3064 u64 orig_num_bytes =
3065 btrfs_file_extent_num_bytes(leaf, fi);
3066 extent_num_bytes = new_size -
3067 found_key.offset + root->sectorsize - 1;
3068 extent_num_bytes = extent_num_bytes &
3069 ~((u64)root->sectorsize - 1);
3070 btrfs_set_file_extent_num_bytes(leaf, fi,
3072 num_dec = (orig_num_bytes -
3074 if (root->ref_cows && extent_start != 0)
3075 inode_sub_bytes(inode, num_dec);
3076 btrfs_mark_buffer_dirty(leaf);
3079 btrfs_file_extent_disk_num_bytes(leaf,
3081 extent_offset = found_key.offset -
3082 btrfs_file_extent_offset(leaf, fi);
3084 /* FIXME blocksize != 4096 */
3085 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3086 if (extent_start != 0) {
3089 inode_sub_bytes(inode, num_dec);
3092 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3094 * we can't truncate inline items that have had
3098 btrfs_file_extent_compression(leaf, fi) == 0 &&
3099 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3100 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3101 u32 size = new_size - found_key.offset;
3103 if (root->ref_cows) {
3104 inode_sub_bytes(inode, item_end + 1 -
3108 btrfs_file_extent_calc_inline_size(size);
3109 ret = btrfs_truncate_item(trans, root, path,
3111 } else if (root->ref_cows) {
3112 inode_sub_bytes(inode, item_end + 1 -
3118 if (!pending_del_nr) {
3119 /* no pending yet, add ourselves */
3120 pending_del_slot = path->slots[0];
3122 } else if (pending_del_nr &&
3123 path->slots[0] + 1 == pending_del_slot) {
3124 /* hop on the pending chunk */
3126 pending_del_slot = path->slots[0];
3133 if (found_extent && (root->ref_cows ||
3134 root == root->fs_info->tree_root)) {
3135 btrfs_set_path_blocking(path);
3136 ret = btrfs_free_extent(trans, root, extent_start,
3137 extent_num_bytes, 0,
3138 btrfs_header_owner(leaf),
3139 ino, extent_offset);
3143 if (found_type == BTRFS_INODE_ITEM_KEY)
3146 if (path->slots[0] == 0 ||
3147 path->slots[0] != pending_del_slot) {
3148 if (root->ref_cows &&
3149 BTRFS_I(inode)->location.objectid !=
3150 BTRFS_FREE_INO_OBJECTID) {
3154 if (pending_del_nr) {
3155 ret = btrfs_del_items(trans, root, path,
3161 btrfs_release_path(path);
3168 if (pending_del_nr) {
3169 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3173 btrfs_free_path(path);
3178 * taken from block_truncate_page, but does cow as it zeros out
3179 * any bytes left in the last page in the file.
3181 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3183 struct inode *inode = mapping->host;
3184 struct btrfs_root *root = BTRFS_I(inode)->root;
3185 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3186 struct btrfs_ordered_extent *ordered;
3187 struct extent_state *cached_state = NULL;
3189 u32 blocksize = root->sectorsize;
3190 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3191 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3193 gfp_t mask = btrfs_alloc_write_mask(mapping);
3198 if ((offset & (blocksize - 1)) == 0)
3200 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3206 page = find_or_create_page(mapping, index, mask);
3208 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3212 page_start = page_offset(page);
3213 page_end = page_start + PAGE_CACHE_SIZE - 1;
3215 if (!PageUptodate(page)) {
3216 ret = btrfs_readpage(NULL, page);
3218 if (page->mapping != mapping) {
3220 page_cache_release(page);
3223 if (!PageUptodate(page)) {
3228 wait_on_page_writeback(page);
3230 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3232 set_page_extent_mapped(page);
3234 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3236 unlock_extent_cached(io_tree, page_start, page_end,
3237 &cached_state, GFP_NOFS);
3239 page_cache_release(page);
3240 btrfs_start_ordered_extent(inode, ordered, 1);
3241 btrfs_put_ordered_extent(ordered);
3245 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3246 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3247 0, 0, &cached_state, GFP_NOFS);
3249 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3252 unlock_extent_cached(io_tree, page_start, page_end,
3253 &cached_state, GFP_NOFS);
3258 if (offset != PAGE_CACHE_SIZE) {
3260 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3261 flush_dcache_page(page);
3264 ClearPageChecked(page);
3265 set_page_dirty(page);
3266 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3271 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3273 page_cache_release(page);
3279 * This function puts in dummy file extents for the area we're creating a hole
3280 * for. So if we are truncating this file to a larger size we need to insert
3281 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3282 * the range between oldsize and size
3284 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3286 struct btrfs_trans_handle *trans;
3287 struct btrfs_root *root = BTRFS_I(inode)->root;
3288 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3289 struct extent_map *em = NULL;
3290 struct extent_state *cached_state = NULL;
3291 u64 mask = root->sectorsize - 1;
3292 u64 hole_start = (oldsize + mask) & ~mask;
3293 u64 block_end = (size + mask) & ~mask;
3299 if (size <= hole_start)
3303 struct btrfs_ordered_extent *ordered;
3304 btrfs_wait_ordered_range(inode, hole_start,
3305 block_end - hole_start);
3306 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3307 &cached_state, GFP_NOFS);
3308 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3311 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3312 &cached_state, GFP_NOFS);
3313 btrfs_put_ordered_extent(ordered);
3316 cur_offset = hole_start;
3318 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3319 block_end - cur_offset, 0);
3320 BUG_ON(IS_ERR_OR_NULL(em));
3321 last_byte = min(extent_map_end(em), block_end);
3322 last_byte = (last_byte + mask) & ~mask;
3323 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3325 hole_size = last_byte - cur_offset;
3327 trans = btrfs_start_transaction(root, 2);
3328 if (IS_ERR(trans)) {
3329 err = PTR_ERR(trans);
3333 err = btrfs_drop_extents(trans, inode, cur_offset,
3334 cur_offset + hole_size,
3337 btrfs_end_transaction(trans, root);
3341 err = btrfs_insert_file_extent(trans, root,
3342 btrfs_ino(inode), cur_offset, 0,
3343 0, hole_size, 0, hole_size,
3346 btrfs_end_transaction(trans, root);
3350 btrfs_drop_extent_cache(inode, hole_start,
3353 btrfs_end_transaction(trans, root);
3355 free_extent_map(em);
3357 cur_offset = last_byte;
3358 if (cur_offset >= block_end)
3362 free_extent_map(em);
3363 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3368 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3370 loff_t oldsize = i_size_read(inode);
3373 if (newsize == oldsize)
3376 if (newsize > oldsize) {
3377 i_size_write(inode, newsize);
3378 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3379 truncate_pagecache(inode, oldsize, newsize);
3380 ret = btrfs_cont_expand(inode, oldsize, newsize);
3382 btrfs_setsize(inode, oldsize);
3386 mark_inode_dirty(inode);
3390 * We're truncating a file that used to have good data down to
3391 * zero. Make sure it gets into the ordered flush list so that
3392 * any new writes get down to disk quickly.
3395 BTRFS_I(inode)->ordered_data_close = 1;
3397 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3398 truncate_setsize(inode, newsize);
3399 ret = btrfs_truncate(inode);
3405 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3407 struct inode *inode = dentry->d_inode;
3408 struct btrfs_root *root = BTRFS_I(inode)->root;
3411 if (btrfs_root_readonly(root))
3414 err = inode_change_ok(inode, attr);
3418 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3419 err = btrfs_setsize(inode, attr->ia_size);
3424 if (attr->ia_valid) {
3425 setattr_copy(inode, attr);
3426 mark_inode_dirty(inode);
3428 if (attr->ia_valid & ATTR_MODE)
3429 err = btrfs_acl_chmod(inode);
3435 void btrfs_evict_inode(struct inode *inode)
3437 struct btrfs_trans_handle *trans;
3438 struct btrfs_root *root = BTRFS_I(inode)->root;
3439 struct btrfs_block_rsv *rsv, *global_rsv;
3440 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3444 trace_btrfs_inode_evict(inode);
3446 truncate_inode_pages(&inode->i_data, 0);
3447 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3448 btrfs_is_free_space_inode(root, inode)))
3451 if (is_bad_inode(inode)) {
3452 btrfs_orphan_del(NULL, inode);
3455 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3456 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3458 if (root->fs_info->log_root_recovering) {
3459 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3463 if (inode->i_nlink > 0) {
3464 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3468 rsv = btrfs_alloc_block_rsv(root);
3470 btrfs_orphan_del(NULL, inode);
3473 rsv->size = min_size;
3474 global_rsv = &root->fs_info->global_block_rsv;
3476 btrfs_i_size_write(inode, 0);
3479 * This is a bit simpler than btrfs_truncate since
3481 * 1) We've already reserved our space for our orphan item in the
3483 * 2) We're going to delete the inode item, so we don't need to update
3486 * So we just need to reserve some slack space in case we add bytes when
3487 * doing the truncate.
3490 ret = btrfs_block_rsv_refill(root, rsv, min_size);
3493 * Try and steal from the global reserve since we will
3494 * likely not use this space anyway, we want to try as
3495 * hard as possible to get this to work.
3498 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3501 printk(KERN_WARNING "Could not get space for a "
3502 "delete, will truncate on mount %d\n", ret);
3503 btrfs_orphan_del(NULL, inode);
3504 btrfs_free_block_rsv(root, rsv);
3508 trans = btrfs_start_transaction(root, 0);
3509 if (IS_ERR(trans)) {
3510 btrfs_orphan_del(NULL, inode);
3511 btrfs_free_block_rsv(root, rsv);
3515 trans->block_rsv = rsv;
3517 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3521 nr = trans->blocks_used;
3522 btrfs_end_transaction(trans, root);
3524 btrfs_btree_balance_dirty(root, nr);
3527 btrfs_free_block_rsv(root, rsv);
3530 trans->block_rsv = root->orphan_block_rsv;
3531 ret = btrfs_orphan_del(trans, inode);
3535 trans->block_rsv = &root->fs_info->trans_block_rsv;
3536 if (!(root == root->fs_info->tree_root ||
3537 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3538 btrfs_return_ino(root, btrfs_ino(inode));
3540 nr = trans->blocks_used;
3541 btrfs_end_transaction(trans, root);
3542 btrfs_btree_balance_dirty(root, nr);
3544 end_writeback(inode);
3549 * this returns the key found in the dir entry in the location pointer.
3550 * If no dir entries were found, location->objectid is 0.
3552 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3553 struct btrfs_key *location)
3555 const char *name = dentry->d_name.name;
3556 int namelen = dentry->d_name.len;
3557 struct btrfs_dir_item *di;
3558 struct btrfs_path *path;
3559 struct btrfs_root *root = BTRFS_I(dir)->root;
3562 path = btrfs_alloc_path();
3566 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3571 if (IS_ERR_OR_NULL(di))
3574 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3576 btrfs_free_path(path);
3579 location->objectid = 0;
3584 * when we hit a tree root in a directory, the btrfs part of the inode
3585 * needs to be changed to reflect the root directory of the tree root. This
3586 * is kind of like crossing a mount point.
3588 static int fixup_tree_root_location(struct btrfs_root *root,
3590 struct dentry *dentry,
3591 struct btrfs_key *location,
3592 struct btrfs_root **sub_root)
3594 struct btrfs_path *path;
3595 struct btrfs_root *new_root;
3596 struct btrfs_root_ref *ref;
3597 struct extent_buffer *leaf;
3601 path = btrfs_alloc_path();
3608 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3609 BTRFS_I(dir)->root->root_key.objectid,
3610 location->objectid);
3617 leaf = path->nodes[0];
3618 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3619 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3620 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3623 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3624 (unsigned long)(ref + 1),
3625 dentry->d_name.len);
3629 btrfs_release_path(path);
3631 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3632 if (IS_ERR(new_root)) {
3633 err = PTR_ERR(new_root);
3637 if (btrfs_root_refs(&new_root->root_item) == 0) {
3642 *sub_root = new_root;
3643 location->objectid = btrfs_root_dirid(&new_root->root_item);
3644 location->type = BTRFS_INODE_ITEM_KEY;
3645 location->offset = 0;
3648 btrfs_free_path(path);
3652 static void inode_tree_add(struct inode *inode)
3654 struct btrfs_root *root = BTRFS_I(inode)->root;
3655 struct btrfs_inode *entry;
3657 struct rb_node *parent;
3658 u64 ino = btrfs_ino(inode);
3660 p = &root->inode_tree.rb_node;
3663 if (inode_unhashed(inode))
3666 spin_lock(&root->inode_lock);
3669 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3671 if (ino < btrfs_ino(&entry->vfs_inode))
3672 p = &parent->rb_left;
3673 else if (ino > btrfs_ino(&entry->vfs_inode))
3674 p = &parent->rb_right;
3676 WARN_ON(!(entry->vfs_inode.i_state &
3677 (I_WILL_FREE | I_FREEING)));
3678 rb_erase(parent, &root->inode_tree);
3679 RB_CLEAR_NODE(parent);
3680 spin_unlock(&root->inode_lock);
3684 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3685 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3686 spin_unlock(&root->inode_lock);
3689 static void inode_tree_del(struct inode *inode)
3691 struct btrfs_root *root = BTRFS_I(inode)->root;
3694 spin_lock(&root->inode_lock);
3695 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3696 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3697 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3698 empty = RB_EMPTY_ROOT(&root->inode_tree);
3700 spin_unlock(&root->inode_lock);
3703 * Free space cache has inodes in the tree root, but the tree root has a
3704 * root_refs of 0, so this could end up dropping the tree root as a
3705 * snapshot, so we need the extra !root->fs_info->tree_root check to
3706 * make sure we don't drop it.
3708 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3709 root != root->fs_info->tree_root) {
3710 synchronize_srcu(&root->fs_info->subvol_srcu);
3711 spin_lock(&root->inode_lock);
3712 empty = RB_EMPTY_ROOT(&root->inode_tree);
3713 spin_unlock(&root->inode_lock);
3715 btrfs_add_dead_root(root);
3719 int btrfs_invalidate_inodes(struct btrfs_root *root)
3721 struct rb_node *node;
3722 struct rb_node *prev;
3723 struct btrfs_inode *entry;
3724 struct inode *inode;
3727 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3729 spin_lock(&root->inode_lock);
3731 node = root->inode_tree.rb_node;
3735 entry = rb_entry(node, struct btrfs_inode, rb_node);
3737 if (objectid < btrfs_ino(&entry->vfs_inode))
3738 node = node->rb_left;
3739 else if (objectid > btrfs_ino(&entry->vfs_inode))
3740 node = node->rb_right;
3746 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3747 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3751 prev = rb_next(prev);
3755 entry = rb_entry(node, struct btrfs_inode, rb_node);
3756 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3757 inode = igrab(&entry->vfs_inode);
3759 spin_unlock(&root->inode_lock);
3760 if (atomic_read(&inode->i_count) > 1)
3761 d_prune_aliases(inode);
3763 * btrfs_drop_inode will have it removed from
3764 * the inode cache when its usage count
3769 spin_lock(&root->inode_lock);
3773 if (cond_resched_lock(&root->inode_lock))
3776 node = rb_next(node);
3778 spin_unlock(&root->inode_lock);
3782 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3784 struct btrfs_iget_args *args = p;
3785 inode->i_ino = args->ino;
3786 BTRFS_I(inode)->root = args->root;
3787 btrfs_set_inode_space_info(args->root, inode);
3791 static int btrfs_find_actor(struct inode *inode, void *opaque)
3793 struct btrfs_iget_args *args = opaque;
3794 return args->ino == btrfs_ino(inode) &&
3795 args->root == BTRFS_I(inode)->root;
3798 static struct inode *btrfs_iget_locked(struct super_block *s,
3800 struct btrfs_root *root)
3802 struct inode *inode;
3803 struct btrfs_iget_args args;
3804 args.ino = objectid;
3807 inode = iget5_locked(s, objectid, btrfs_find_actor,
3808 btrfs_init_locked_inode,
3813 /* Get an inode object given its location and corresponding root.
3814 * Returns in *is_new if the inode was read from disk
3816 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3817 struct btrfs_root *root, int *new)
3819 struct inode *inode;
3821 inode = btrfs_iget_locked(s, location->objectid, root);
3823 return ERR_PTR(-ENOMEM);
3825 if (inode->i_state & I_NEW) {
3826 BTRFS_I(inode)->root = root;
3827 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3828 btrfs_read_locked_inode(inode);
3829 if (!is_bad_inode(inode)) {
3830 inode_tree_add(inode);
3831 unlock_new_inode(inode);
3835 unlock_new_inode(inode);
3837 inode = ERR_PTR(-ESTALE);
3844 static struct inode *new_simple_dir(struct super_block *s,
3845 struct btrfs_key *key,
3846 struct btrfs_root *root)
3848 struct inode *inode = new_inode(s);
3851 return ERR_PTR(-ENOMEM);
3853 BTRFS_I(inode)->root = root;
3854 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3855 BTRFS_I(inode)->dummy_inode = 1;
3857 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3858 inode->i_op = &simple_dir_inode_operations;
3859 inode->i_fop = &simple_dir_operations;
3860 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3861 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3866 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3868 struct inode *inode;
3869 struct btrfs_root *root = BTRFS_I(dir)->root;
3870 struct btrfs_root *sub_root = root;
3871 struct btrfs_key location;
3875 if (dentry->d_name.len > BTRFS_NAME_LEN)
3876 return ERR_PTR(-ENAMETOOLONG);
3878 if (unlikely(d_need_lookup(dentry))) {
3879 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3880 kfree(dentry->d_fsdata);
3881 dentry->d_fsdata = NULL;
3882 /* This thing is hashed, drop it for now */
3885 ret = btrfs_inode_by_name(dir, dentry, &location);
3889 return ERR_PTR(ret);
3891 if (location.objectid == 0)
3894 if (location.type == BTRFS_INODE_ITEM_KEY) {
3895 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3899 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3901 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3902 ret = fixup_tree_root_location(root, dir, dentry,
3903 &location, &sub_root);
3906 inode = ERR_PTR(ret);
3908 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3910 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3912 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3914 if (!IS_ERR(inode) && root != sub_root) {
3915 down_read(&root->fs_info->cleanup_work_sem);
3916 if (!(inode->i_sb->s_flags & MS_RDONLY))
3917 ret = btrfs_orphan_cleanup(sub_root);
3918 up_read(&root->fs_info->cleanup_work_sem);
3920 inode = ERR_PTR(ret);
3926 static int btrfs_dentry_delete(const struct dentry *dentry)
3928 struct btrfs_root *root;
3930 if (!dentry->d_inode && !IS_ROOT(dentry))
3931 dentry = dentry->d_parent;
3933 if (dentry->d_inode) {
3934 root = BTRFS_I(dentry->d_inode)->root;
3935 if (btrfs_root_refs(&root->root_item) == 0)
3941 static void btrfs_dentry_release(struct dentry *dentry)
3943 if (dentry->d_fsdata)
3944 kfree(dentry->d_fsdata);
3947 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3948 struct nameidata *nd)
3952 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3953 if (unlikely(d_need_lookup(dentry))) {
3954 spin_lock(&dentry->d_lock);
3955 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
3956 spin_unlock(&dentry->d_lock);
3961 unsigned char btrfs_filetype_table[] = {
3962 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3965 static int btrfs_real_readdir(struct file *filp, void *dirent,
3968 struct inode *inode = filp->f_dentry->d_inode;
3969 struct btrfs_root *root = BTRFS_I(inode)->root;
3970 struct btrfs_item *item;
3971 struct btrfs_dir_item *di;
3972 struct btrfs_key key;
3973 struct btrfs_key found_key;
3974 struct btrfs_path *path;
3975 struct list_head ins_list;
3976 struct list_head del_list;
3979 struct extent_buffer *leaf;
3981 unsigned char d_type;
3986 int key_type = BTRFS_DIR_INDEX_KEY;
3990 int is_curr = 0; /* filp->f_pos points to the current index? */
3992 /* FIXME, use a real flag for deciding about the key type */
3993 if (root->fs_info->tree_root == root)
3994 key_type = BTRFS_DIR_ITEM_KEY;
3996 /* special case for "." */
3997 if (filp->f_pos == 0) {
3998 over = filldir(dirent, ".", 1,
3999 filp->f_pos, btrfs_ino(inode), DT_DIR);
4004 /* special case for .., just use the back ref */
4005 if (filp->f_pos == 1) {
4006 u64 pino = parent_ino(filp->f_path.dentry);
4007 over = filldir(dirent, "..", 2,
4008 filp->f_pos, pino, DT_DIR);
4013 path = btrfs_alloc_path();
4019 if (key_type == BTRFS_DIR_INDEX_KEY) {
4020 INIT_LIST_HEAD(&ins_list);
4021 INIT_LIST_HEAD(&del_list);
4022 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4025 btrfs_set_key_type(&key, key_type);
4026 key.offset = filp->f_pos;
4027 key.objectid = btrfs_ino(inode);
4029 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4034 leaf = path->nodes[0];
4035 slot = path->slots[0];
4036 if (slot >= btrfs_header_nritems(leaf)) {
4037 ret = btrfs_next_leaf(root, path);
4045 item = btrfs_item_nr(leaf, slot);
4046 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4048 if (found_key.objectid != key.objectid)
4050 if (btrfs_key_type(&found_key) != key_type)
4052 if (found_key.offset < filp->f_pos)
4054 if (key_type == BTRFS_DIR_INDEX_KEY &&
4055 btrfs_should_delete_dir_index(&del_list,
4059 filp->f_pos = found_key.offset;
4062 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4064 di_total = btrfs_item_size(leaf, item);
4066 while (di_cur < di_total) {
4067 struct btrfs_key location;
4070 if (verify_dir_item(root, leaf, di))
4073 name_len = btrfs_dir_name_len(leaf, di);
4074 if (name_len <= sizeof(tmp_name)) {
4075 name_ptr = tmp_name;
4077 name_ptr = kmalloc(name_len, GFP_NOFS);
4083 read_extent_buffer(leaf, name_ptr,
4084 (unsigned long)(di + 1), name_len);
4086 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4087 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4091 q.hash = full_name_hash(q.name, q.len);
4092 tmp = d_lookup(filp->f_dentry, &q);
4094 struct btrfs_key *newkey;
4096 newkey = kzalloc(sizeof(struct btrfs_key),
4100 tmp = d_alloc(filp->f_dentry, &q);
4106 memcpy(newkey, &location,
4107 sizeof(struct btrfs_key));
4108 tmp->d_fsdata = newkey;
4109 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4116 /* is this a reference to our own snapshot? If so
4119 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4120 location.objectid == root->root_key.objectid) {
4124 over = filldir(dirent, name_ptr, name_len,
4125 found_key.offset, location.objectid,
4129 if (name_ptr != tmp_name)
4134 di_len = btrfs_dir_name_len(leaf, di) +
4135 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4137 di = (struct btrfs_dir_item *)((char *)di + di_len);
4143 if (key_type == BTRFS_DIR_INDEX_KEY) {
4146 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4152 /* Reached end of directory/root. Bump pos past the last item. */
4153 if (key_type == BTRFS_DIR_INDEX_KEY)
4155 * 32-bit glibc will use getdents64, but then strtol -
4156 * so the last number we can serve is this.
4158 filp->f_pos = 0x7fffffff;
4164 if (key_type == BTRFS_DIR_INDEX_KEY)
4165 btrfs_put_delayed_items(&ins_list, &del_list);
4166 btrfs_free_path(path);
4170 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4172 struct btrfs_root *root = BTRFS_I(inode)->root;
4173 struct btrfs_trans_handle *trans;
4175 bool nolock = false;
4177 if (BTRFS_I(inode)->dummy_inode)
4180 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4183 if (wbc->sync_mode == WB_SYNC_ALL) {
4185 trans = btrfs_join_transaction_nolock(root);
4187 trans = btrfs_join_transaction(root);
4189 return PTR_ERR(trans);
4191 ret = btrfs_end_transaction_nolock(trans, root);
4193 ret = btrfs_commit_transaction(trans, root);
4199 * This is somewhat expensive, updating the tree every time the
4200 * inode changes. But, it is most likely to find the inode in cache.
4201 * FIXME, needs more benchmarking...there are no reasons other than performance
4202 * to keep or drop this code.
4204 void btrfs_dirty_inode(struct inode *inode, int flags)
4206 struct btrfs_root *root = BTRFS_I(inode)->root;
4207 struct btrfs_trans_handle *trans;
4210 if (BTRFS_I(inode)->dummy_inode)
4213 trans = btrfs_join_transaction(root);
4214 BUG_ON(IS_ERR(trans));
4216 ret = btrfs_update_inode(trans, root, inode);
4217 if (ret && ret == -ENOSPC) {
4218 /* whoops, lets try again with the full transaction */
4219 btrfs_end_transaction(trans, root);
4220 trans = btrfs_start_transaction(root, 1);
4221 if (IS_ERR(trans)) {
4222 printk_ratelimited(KERN_ERR "btrfs: fail to "
4223 "dirty inode %llu error %ld\n",
4224 (unsigned long long)btrfs_ino(inode),
4229 ret = btrfs_update_inode(trans, root, inode);
4231 printk_ratelimited(KERN_ERR "btrfs: fail to "
4232 "dirty inode %llu error %d\n",
4233 (unsigned long long)btrfs_ino(inode),
4237 btrfs_end_transaction(trans, root);
4238 if (BTRFS_I(inode)->delayed_node)
4239 btrfs_balance_delayed_items(root);
4243 * find the highest existing sequence number in a directory
4244 * and then set the in-memory index_cnt variable to reflect
4245 * free sequence numbers
4247 static int btrfs_set_inode_index_count(struct inode *inode)
4249 struct btrfs_root *root = BTRFS_I(inode)->root;
4250 struct btrfs_key key, found_key;
4251 struct btrfs_path *path;
4252 struct extent_buffer *leaf;
4255 key.objectid = btrfs_ino(inode);
4256 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4257 key.offset = (u64)-1;
4259 path = btrfs_alloc_path();
4263 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4266 /* FIXME: we should be able to handle this */
4272 * MAGIC NUMBER EXPLANATION:
4273 * since we search a directory based on f_pos we have to start at 2
4274 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4275 * else has to start at 2
4277 if (path->slots[0] == 0) {
4278 BTRFS_I(inode)->index_cnt = 2;
4284 leaf = path->nodes[0];
4285 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4287 if (found_key.objectid != btrfs_ino(inode) ||
4288 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4289 BTRFS_I(inode)->index_cnt = 2;
4293 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4295 btrfs_free_path(path);
4300 * helper to find a free sequence number in a given directory. This current
4301 * code is very simple, later versions will do smarter things in the btree
4303 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4307 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4308 ret = btrfs_inode_delayed_dir_index_count(dir);
4310 ret = btrfs_set_inode_index_count(dir);
4316 *index = BTRFS_I(dir)->index_cnt;
4317 BTRFS_I(dir)->index_cnt++;
4322 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4323 struct btrfs_root *root,
4325 const char *name, int name_len,
4326 u64 ref_objectid, u64 objectid, int mode,
4329 struct inode *inode;
4330 struct btrfs_inode_item *inode_item;
4331 struct btrfs_key *location;
4332 struct btrfs_path *path;
4333 struct btrfs_inode_ref *ref;
4334 struct btrfs_key key[2];
4340 path = btrfs_alloc_path();
4342 return ERR_PTR(-ENOMEM);
4344 inode = new_inode(root->fs_info->sb);
4346 btrfs_free_path(path);
4347 return ERR_PTR(-ENOMEM);
4351 * we have to initialize this early, so we can reclaim the inode
4352 * number if we fail afterwards in this function.
4354 inode->i_ino = objectid;
4357 trace_btrfs_inode_request(dir);
4359 ret = btrfs_set_inode_index(dir, index);
4361 btrfs_free_path(path);
4363 return ERR_PTR(ret);
4367 * index_cnt is ignored for everything but a dir,
4368 * btrfs_get_inode_index_count has an explanation for the magic
4371 BTRFS_I(inode)->index_cnt = 2;
4372 BTRFS_I(inode)->root = root;
4373 BTRFS_I(inode)->generation = trans->transid;
4374 inode->i_generation = BTRFS_I(inode)->generation;
4375 btrfs_set_inode_space_info(root, inode);
4382 key[0].objectid = objectid;
4383 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4386 key[1].objectid = objectid;
4387 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4388 key[1].offset = ref_objectid;
4390 sizes[0] = sizeof(struct btrfs_inode_item);
4391 sizes[1] = name_len + sizeof(*ref);
4393 path->leave_spinning = 1;
4394 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4398 inode_init_owner(inode, dir, mode);
4399 inode_set_bytes(inode, 0);
4400 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4401 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4402 struct btrfs_inode_item);
4403 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4405 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4406 struct btrfs_inode_ref);
4407 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4408 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4409 ptr = (unsigned long)(ref + 1);
4410 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4412 btrfs_mark_buffer_dirty(path->nodes[0]);
4413 btrfs_free_path(path);
4415 location = &BTRFS_I(inode)->location;
4416 location->objectid = objectid;
4417 location->offset = 0;
4418 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4420 btrfs_inherit_iflags(inode, dir);
4422 if (S_ISREG(mode)) {
4423 if (btrfs_test_opt(root, NODATASUM))
4424 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4425 if (btrfs_test_opt(root, NODATACOW) ||
4426 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4427 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4430 insert_inode_hash(inode);
4431 inode_tree_add(inode);
4433 trace_btrfs_inode_new(inode);
4434 btrfs_set_inode_last_trans(trans, inode);
4439 BTRFS_I(dir)->index_cnt--;
4440 btrfs_free_path(path);
4442 return ERR_PTR(ret);
4445 static inline u8 btrfs_inode_type(struct inode *inode)
4447 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4451 * utility function to add 'inode' into 'parent_inode' with
4452 * a give name and a given sequence number.
4453 * if 'add_backref' is true, also insert a backref from the
4454 * inode to the parent directory.
4456 int btrfs_add_link(struct btrfs_trans_handle *trans,
4457 struct inode *parent_inode, struct inode *inode,
4458 const char *name, int name_len, int add_backref, u64 index)
4461 struct btrfs_key key;
4462 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4463 u64 ino = btrfs_ino(inode);
4464 u64 parent_ino = btrfs_ino(parent_inode);
4466 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4467 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4470 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4474 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4475 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4476 key.objectid, root->root_key.objectid,
4477 parent_ino, index, name, name_len);
4478 } else if (add_backref) {
4479 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4484 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4486 btrfs_inode_type(inode), index);
4489 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4491 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4492 ret = btrfs_update_inode(trans, root, parent_inode);
4497 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4498 struct inode *dir, struct dentry *dentry,
4499 struct inode *inode, int backref, u64 index)
4501 int err = btrfs_add_link(trans, dir, inode,
4502 dentry->d_name.name, dentry->d_name.len,
4505 d_instantiate(dentry, inode);
4513 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4514 int mode, dev_t rdev)
4516 struct btrfs_trans_handle *trans;
4517 struct btrfs_root *root = BTRFS_I(dir)->root;
4518 struct inode *inode = NULL;
4522 unsigned long nr = 0;
4525 if (!new_valid_dev(rdev))
4529 * 2 for inode item and ref
4531 * 1 for xattr if selinux is on
4533 trans = btrfs_start_transaction(root, 5);
4535 return PTR_ERR(trans);
4537 err = btrfs_find_free_ino(root, &objectid);
4541 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4542 dentry->d_name.len, btrfs_ino(dir), objectid,
4544 if (IS_ERR(inode)) {
4545 err = PTR_ERR(inode);
4549 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4555 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4559 inode->i_op = &btrfs_special_inode_operations;
4560 init_special_inode(inode, inode->i_mode, rdev);
4561 btrfs_update_inode(trans, root, inode);
4564 nr = trans->blocks_used;
4565 btrfs_end_transaction_throttle(trans, root);
4566 btrfs_btree_balance_dirty(root, nr);
4568 inode_dec_link_count(inode);
4574 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4575 int mode, struct nameidata *nd)
4577 struct btrfs_trans_handle *trans;
4578 struct btrfs_root *root = BTRFS_I(dir)->root;
4579 struct inode *inode = NULL;
4582 unsigned long nr = 0;
4587 * 2 for inode item and ref
4589 * 1 for xattr if selinux is on
4591 trans = btrfs_start_transaction(root, 5);
4593 return PTR_ERR(trans);
4595 err = btrfs_find_free_ino(root, &objectid);
4599 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4600 dentry->d_name.len, btrfs_ino(dir), objectid,
4602 if (IS_ERR(inode)) {
4603 err = PTR_ERR(inode);
4607 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4613 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4617 inode->i_mapping->a_ops = &btrfs_aops;
4618 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4619 inode->i_fop = &btrfs_file_operations;
4620 inode->i_op = &btrfs_file_inode_operations;
4621 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4624 nr = trans->blocks_used;
4625 btrfs_end_transaction_throttle(trans, root);
4627 inode_dec_link_count(inode);
4630 btrfs_btree_balance_dirty(root, nr);
4634 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4635 struct dentry *dentry)
4637 struct btrfs_trans_handle *trans;
4638 struct btrfs_root *root = BTRFS_I(dir)->root;
4639 struct inode *inode = old_dentry->d_inode;
4641 unsigned long nr = 0;
4645 /* do not allow sys_link's with other subvols of the same device */
4646 if (root->objectid != BTRFS_I(inode)->root->objectid)
4649 if (inode->i_nlink == ~0U)
4652 err = btrfs_set_inode_index(dir, &index);
4657 * 2 items for inode and inode ref
4658 * 2 items for dir items
4659 * 1 item for parent inode
4661 trans = btrfs_start_transaction(root, 5);
4662 if (IS_ERR(trans)) {
4663 err = PTR_ERR(trans);
4667 btrfs_inc_nlink(inode);
4668 inode->i_ctime = CURRENT_TIME;
4671 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4676 struct dentry *parent = dentry->d_parent;
4677 err = btrfs_update_inode(trans, root, inode);
4679 btrfs_log_new_name(trans, inode, NULL, parent);
4682 nr = trans->blocks_used;
4683 btrfs_end_transaction_throttle(trans, root);
4686 inode_dec_link_count(inode);
4689 btrfs_btree_balance_dirty(root, nr);
4693 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4695 struct inode *inode = NULL;
4696 struct btrfs_trans_handle *trans;
4697 struct btrfs_root *root = BTRFS_I(dir)->root;
4699 int drop_on_err = 0;
4702 unsigned long nr = 1;
4705 * 2 items for inode and ref
4706 * 2 items for dir items
4707 * 1 for xattr if selinux is on
4709 trans = btrfs_start_transaction(root, 5);
4711 return PTR_ERR(trans);
4713 err = btrfs_find_free_ino(root, &objectid);
4717 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4718 dentry->d_name.len, btrfs_ino(dir), objectid,
4719 S_IFDIR | mode, &index);
4720 if (IS_ERR(inode)) {
4721 err = PTR_ERR(inode);
4727 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4731 inode->i_op = &btrfs_dir_inode_operations;
4732 inode->i_fop = &btrfs_dir_file_operations;
4734 btrfs_i_size_write(inode, 0);
4735 err = btrfs_update_inode(trans, root, inode);
4739 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4740 dentry->d_name.len, 0, index);
4744 d_instantiate(dentry, inode);
4748 nr = trans->blocks_used;
4749 btrfs_end_transaction_throttle(trans, root);
4752 btrfs_btree_balance_dirty(root, nr);
4756 /* helper for btfs_get_extent. Given an existing extent in the tree,
4757 * and an extent that you want to insert, deal with overlap and insert
4758 * the new extent into the tree.
4760 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4761 struct extent_map *existing,
4762 struct extent_map *em,
4763 u64 map_start, u64 map_len)
4767 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4768 start_diff = map_start - em->start;
4769 em->start = map_start;
4771 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4772 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4773 em->block_start += start_diff;
4774 em->block_len -= start_diff;
4776 return add_extent_mapping(em_tree, em);
4779 static noinline int uncompress_inline(struct btrfs_path *path,
4780 struct inode *inode, struct page *page,
4781 size_t pg_offset, u64 extent_offset,
4782 struct btrfs_file_extent_item *item)
4785 struct extent_buffer *leaf = path->nodes[0];
4788 unsigned long inline_size;
4792 WARN_ON(pg_offset != 0);
4793 compress_type = btrfs_file_extent_compression(leaf, item);
4794 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4795 inline_size = btrfs_file_extent_inline_item_len(leaf,
4796 btrfs_item_nr(leaf, path->slots[0]));
4797 tmp = kmalloc(inline_size, GFP_NOFS);
4800 ptr = btrfs_file_extent_inline_start(item);
4802 read_extent_buffer(leaf, tmp, ptr, inline_size);
4804 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4805 ret = btrfs_decompress(compress_type, tmp, page,
4806 extent_offset, inline_size, max_size);
4808 char *kaddr = kmap_atomic(page, KM_USER0);
4809 unsigned long copy_size = min_t(u64,
4810 PAGE_CACHE_SIZE - pg_offset,
4811 max_size - extent_offset);
4812 memset(kaddr + pg_offset, 0, copy_size);
4813 kunmap_atomic(kaddr, KM_USER0);
4820 * a bit scary, this does extent mapping from logical file offset to the disk.
4821 * the ugly parts come from merging extents from the disk with the in-ram
4822 * representation. This gets more complex because of the data=ordered code,
4823 * where the in-ram extents might be locked pending data=ordered completion.
4825 * This also copies inline extents directly into the page.
4828 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4829 size_t pg_offset, u64 start, u64 len,
4835 u64 extent_start = 0;
4837 u64 objectid = btrfs_ino(inode);
4839 struct btrfs_path *path = NULL;
4840 struct btrfs_root *root = BTRFS_I(inode)->root;
4841 struct btrfs_file_extent_item *item;
4842 struct extent_buffer *leaf;
4843 struct btrfs_key found_key;
4844 struct extent_map *em = NULL;
4845 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4846 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4847 struct btrfs_trans_handle *trans = NULL;
4851 read_lock(&em_tree->lock);
4852 em = lookup_extent_mapping(em_tree, start, len);
4854 em->bdev = root->fs_info->fs_devices->latest_bdev;
4855 read_unlock(&em_tree->lock);
4858 if (em->start > start || em->start + em->len <= start)
4859 free_extent_map(em);
4860 else if (em->block_start == EXTENT_MAP_INLINE && page)
4861 free_extent_map(em);
4865 em = alloc_extent_map();
4870 em->bdev = root->fs_info->fs_devices->latest_bdev;
4871 em->start = EXTENT_MAP_HOLE;
4872 em->orig_start = EXTENT_MAP_HOLE;
4874 em->block_len = (u64)-1;
4877 path = btrfs_alloc_path();
4883 * Chances are we'll be called again, so go ahead and do
4889 ret = btrfs_lookup_file_extent(trans, root, path,
4890 objectid, start, trans != NULL);
4897 if (path->slots[0] == 0)
4902 leaf = path->nodes[0];
4903 item = btrfs_item_ptr(leaf, path->slots[0],
4904 struct btrfs_file_extent_item);
4905 /* are we inside the extent that was found? */
4906 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4907 found_type = btrfs_key_type(&found_key);
4908 if (found_key.objectid != objectid ||
4909 found_type != BTRFS_EXTENT_DATA_KEY) {
4913 found_type = btrfs_file_extent_type(leaf, item);
4914 extent_start = found_key.offset;
4915 compress_type = btrfs_file_extent_compression(leaf, item);
4916 if (found_type == BTRFS_FILE_EXTENT_REG ||
4917 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4918 extent_end = extent_start +
4919 btrfs_file_extent_num_bytes(leaf, item);
4920 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4922 size = btrfs_file_extent_inline_len(leaf, item);
4923 extent_end = (extent_start + size + root->sectorsize - 1) &
4924 ~((u64)root->sectorsize - 1);
4927 if (start >= extent_end) {
4929 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4930 ret = btrfs_next_leaf(root, path);
4937 leaf = path->nodes[0];
4939 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4940 if (found_key.objectid != objectid ||
4941 found_key.type != BTRFS_EXTENT_DATA_KEY)
4943 if (start + len <= found_key.offset)
4946 em->len = found_key.offset - start;
4950 if (found_type == BTRFS_FILE_EXTENT_REG ||
4951 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4952 em->start = extent_start;
4953 em->len = extent_end - extent_start;
4954 em->orig_start = extent_start -
4955 btrfs_file_extent_offset(leaf, item);
4956 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4958 em->block_start = EXTENT_MAP_HOLE;
4961 if (compress_type != BTRFS_COMPRESS_NONE) {
4962 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4963 em->compress_type = compress_type;
4964 em->block_start = bytenr;
4965 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4968 bytenr += btrfs_file_extent_offset(leaf, item);
4969 em->block_start = bytenr;
4970 em->block_len = em->len;
4971 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4972 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4975 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4979 size_t extent_offset;
4982 em->block_start = EXTENT_MAP_INLINE;
4983 if (!page || create) {
4984 em->start = extent_start;
4985 em->len = extent_end - extent_start;
4989 size = btrfs_file_extent_inline_len(leaf, item);
4990 extent_offset = page_offset(page) + pg_offset - extent_start;
4991 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4992 size - extent_offset);
4993 em->start = extent_start + extent_offset;
4994 em->len = (copy_size + root->sectorsize - 1) &
4995 ~((u64)root->sectorsize - 1);
4996 em->orig_start = EXTENT_MAP_INLINE;
4997 if (compress_type) {
4998 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4999 em->compress_type = compress_type;
5001 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5002 if (create == 0 && !PageUptodate(page)) {
5003 if (btrfs_file_extent_compression(leaf, item) !=
5004 BTRFS_COMPRESS_NONE) {
5005 ret = uncompress_inline(path, inode, page,
5007 extent_offset, item);
5011 read_extent_buffer(leaf, map + pg_offset, ptr,
5013 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5014 memset(map + pg_offset + copy_size, 0,
5015 PAGE_CACHE_SIZE - pg_offset -
5020 flush_dcache_page(page);
5021 } else if (create && PageUptodate(page)) {
5025 free_extent_map(em);
5028 btrfs_release_path(path);
5029 trans = btrfs_join_transaction(root);
5032 return ERR_CAST(trans);
5036 write_extent_buffer(leaf, map + pg_offset, ptr,
5039 btrfs_mark_buffer_dirty(leaf);
5041 set_extent_uptodate(io_tree, em->start,
5042 extent_map_end(em) - 1, NULL, GFP_NOFS);
5045 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5052 em->block_start = EXTENT_MAP_HOLE;
5053 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5055 btrfs_release_path(path);
5056 if (em->start > start || extent_map_end(em) <= start) {
5057 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5058 "[%llu %llu]\n", (unsigned long long)em->start,
5059 (unsigned long long)em->len,
5060 (unsigned long long)start,
5061 (unsigned long long)len);
5067 write_lock(&em_tree->lock);
5068 ret = add_extent_mapping(em_tree, em);
5069 /* it is possible that someone inserted the extent into the tree
5070 * while we had the lock dropped. It is also possible that
5071 * an overlapping map exists in the tree
5073 if (ret == -EEXIST) {
5074 struct extent_map *existing;
5078 existing = lookup_extent_mapping(em_tree, start, len);
5079 if (existing && (existing->start > start ||
5080 existing->start + existing->len <= start)) {
5081 free_extent_map(existing);
5085 existing = lookup_extent_mapping(em_tree, em->start,
5088 err = merge_extent_mapping(em_tree, existing,
5091 free_extent_map(existing);
5093 free_extent_map(em);
5098 free_extent_map(em);
5102 free_extent_map(em);
5107 write_unlock(&em_tree->lock);
5110 trace_btrfs_get_extent(root, em);
5113 btrfs_free_path(path);
5115 ret = btrfs_end_transaction(trans, root);
5120 free_extent_map(em);
5121 return ERR_PTR(err);
5126 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5127 size_t pg_offset, u64 start, u64 len,
5130 struct extent_map *em;
5131 struct extent_map *hole_em = NULL;
5132 u64 range_start = start;
5138 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5143 * if our em maps to a hole, there might
5144 * actually be delalloc bytes behind it
5146 if (em->block_start != EXTENT_MAP_HOLE)
5152 /* check to see if we've wrapped (len == -1 or similar) */
5161 /* ok, we didn't find anything, lets look for delalloc */
5162 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5163 end, len, EXTENT_DELALLOC, 1);
5164 found_end = range_start + found;
5165 if (found_end < range_start)
5166 found_end = (u64)-1;
5169 * we didn't find anything useful, return
5170 * the original results from get_extent()
5172 if (range_start > end || found_end <= start) {
5178 /* adjust the range_start to make sure it doesn't
5179 * go backwards from the start they passed in
5181 range_start = max(start,range_start);
5182 found = found_end - range_start;
5185 u64 hole_start = start;
5188 em = alloc_extent_map();
5194 * when btrfs_get_extent can't find anything it
5195 * returns one huge hole
5197 * make sure what it found really fits our range, and
5198 * adjust to make sure it is based on the start from
5202 u64 calc_end = extent_map_end(hole_em);
5204 if (calc_end <= start || (hole_em->start > end)) {
5205 free_extent_map(hole_em);
5208 hole_start = max(hole_em->start, start);
5209 hole_len = calc_end - hole_start;
5213 if (hole_em && range_start > hole_start) {
5214 /* our hole starts before our delalloc, so we
5215 * have to return just the parts of the hole
5216 * that go until the delalloc starts
5218 em->len = min(hole_len,
5219 range_start - hole_start);
5220 em->start = hole_start;
5221 em->orig_start = hole_start;
5223 * don't adjust block start at all,
5224 * it is fixed at EXTENT_MAP_HOLE
5226 em->block_start = hole_em->block_start;
5227 em->block_len = hole_len;
5229 em->start = range_start;
5231 em->orig_start = range_start;
5232 em->block_start = EXTENT_MAP_DELALLOC;
5233 em->block_len = found;
5235 } else if (hole_em) {
5240 free_extent_map(hole_em);
5242 free_extent_map(em);
5243 return ERR_PTR(err);
5248 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5249 struct extent_map *em,
5252 struct btrfs_root *root = BTRFS_I(inode)->root;
5253 struct btrfs_trans_handle *trans;
5254 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5255 struct btrfs_key ins;
5258 bool insert = false;
5261 * Ok if the extent map we looked up is a hole and is for the exact
5262 * range we want, there is no reason to allocate a new one, however if
5263 * it is not right then we need to free this one and drop the cache for
5266 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5268 free_extent_map(em);
5271 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5274 trans = btrfs_join_transaction(root);
5276 return ERR_CAST(trans);
5278 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5279 btrfs_add_inode_defrag(trans, inode);
5281 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5283 alloc_hint = get_extent_allocation_hint(inode, start, len);
5284 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5285 alloc_hint, (u64)-1, &ins, 1);
5292 em = alloc_extent_map();
5294 em = ERR_PTR(-ENOMEM);
5300 em->orig_start = em->start;
5301 em->len = ins.offset;
5303 em->block_start = ins.objectid;
5304 em->block_len = ins.offset;
5305 em->bdev = root->fs_info->fs_devices->latest_bdev;
5308 * We need to do this because if we're using the original em we searched
5309 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5312 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5315 write_lock(&em_tree->lock);
5316 ret = add_extent_mapping(em_tree, em);
5317 write_unlock(&em_tree->lock);
5320 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5323 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5324 ins.offset, ins.offset, 0);
5326 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5330 btrfs_end_transaction(trans, root);
5335 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5336 * block must be cow'd
5338 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5339 struct inode *inode, u64 offset, u64 len)
5341 struct btrfs_path *path;
5343 struct extent_buffer *leaf;
5344 struct btrfs_root *root = BTRFS_I(inode)->root;
5345 struct btrfs_file_extent_item *fi;
5346 struct btrfs_key key;
5354 path = btrfs_alloc_path();
5358 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5363 slot = path->slots[0];
5366 /* can't find the item, must cow */
5373 leaf = path->nodes[0];
5374 btrfs_item_key_to_cpu(leaf, &key, slot);
5375 if (key.objectid != btrfs_ino(inode) ||
5376 key.type != BTRFS_EXTENT_DATA_KEY) {
5377 /* not our file or wrong item type, must cow */
5381 if (key.offset > offset) {
5382 /* Wrong offset, must cow */
5386 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5387 found_type = btrfs_file_extent_type(leaf, fi);
5388 if (found_type != BTRFS_FILE_EXTENT_REG &&
5389 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5390 /* not a regular extent, must cow */
5393 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5394 backref_offset = btrfs_file_extent_offset(leaf, fi);
5396 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5397 if (extent_end < offset + len) {
5398 /* extent doesn't include our full range, must cow */
5402 if (btrfs_extent_readonly(root, disk_bytenr))
5406 * look for other files referencing this extent, if we
5407 * find any we must cow
5409 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5410 key.offset - backref_offset, disk_bytenr))
5414 * adjust disk_bytenr and num_bytes to cover just the bytes
5415 * in this extent we are about to write. If there
5416 * are any csums in that range we have to cow in order
5417 * to keep the csums correct
5419 disk_bytenr += backref_offset;
5420 disk_bytenr += offset - key.offset;
5421 num_bytes = min(offset + len, extent_end) - offset;
5422 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5425 * all of the above have passed, it is safe to overwrite this extent
5430 btrfs_free_path(path);
5434 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5435 struct buffer_head *bh_result, int create)
5437 struct extent_map *em;
5438 struct btrfs_root *root = BTRFS_I(inode)->root;
5439 u64 start = iblock << inode->i_blkbits;
5440 u64 len = bh_result->b_size;
5441 struct btrfs_trans_handle *trans;
5443 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5448 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5449 * io. INLINE is special, and we could probably kludge it in here, but
5450 * it's still buffered so for safety lets just fall back to the generic
5453 * For COMPRESSED we _have_ to read the entire extent in so we can
5454 * decompress it, so there will be buffering required no matter what we
5455 * do, so go ahead and fallback to buffered.
5457 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5458 * to buffered IO. Don't blame me, this is the price we pay for using
5461 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5462 em->block_start == EXTENT_MAP_INLINE) {
5463 free_extent_map(em);
5467 /* Just a good old fashioned hole, return */
5468 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5469 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5470 free_extent_map(em);
5471 /* DIO will do one hole at a time, so just unlock a sector */
5472 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5473 start + root->sectorsize - 1, GFP_NOFS);
5478 * We don't allocate a new extent in the following cases
5480 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5482 * 2) The extent is marked as PREALLOC. We're good to go here and can
5483 * just use the extent.
5487 len = em->len - (start - em->start);
5491 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5492 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5493 em->block_start != EXTENT_MAP_HOLE)) {
5498 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5499 type = BTRFS_ORDERED_PREALLOC;
5501 type = BTRFS_ORDERED_NOCOW;
5502 len = min(len, em->len - (start - em->start));
5503 block_start = em->block_start + (start - em->start);
5506 * we're not going to log anything, but we do need
5507 * to make sure the current transaction stays open
5508 * while we look for nocow cross refs
5510 trans = btrfs_join_transaction(root);
5514 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5515 ret = btrfs_add_ordered_extent_dio(inode, start,
5516 block_start, len, len, type);
5517 btrfs_end_transaction(trans, root);
5519 free_extent_map(em);
5524 btrfs_end_transaction(trans, root);
5528 * this will cow the extent, reset the len in case we changed
5531 len = bh_result->b_size;
5532 em = btrfs_new_extent_direct(inode, em, start, len);
5535 len = min(len, em->len - (start - em->start));
5537 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5538 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5541 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5543 bh_result->b_size = len;
5544 bh_result->b_bdev = em->bdev;
5545 set_buffer_mapped(bh_result);
5546 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5547 set_buffer_new(bh_result);
5549 free_extent_map(em);
5554 struct btrfs_dio_private {
5555 struct inode *inode;
5562 /* number of bios pending for this dio */
5563 atomic_t pending_bios;
5568 struct bio *orig_bio;
5571 static void btrfs_endio_direct_read(struct bio *bio, int err)
5573 struct btrfs_dio_private *dip = bio->bi_private;
5574 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5575 struct bio_vec *bvec = bio->bi_io_vec;
5576 struct inode *inode = dip->inode;
5577 struct btrfs_root *root = BTRFS_I(inode)->root;
5579 u32 *private = dip->csums;
5581 start = dip->logical_offset;
5583 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5584 struct page *page = bvec->bv_page;
5587 unsigned long flags;
5589 local_irq_save(flags);
5590 kaddr = kmap_atomic(page, KM_IRQ0);
5591 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5592 csum, bvec->bv_len);
5593 btrfs_csum_final(csum, (char *)&csum);
5594 kunmap_atomic(kaddr, KM_IRQ0);
5595 local_irq_restore(flags);
5597 flush_dcache_page(bvec->bv_page);
5598 if (csum != *private) {
5599 printk(KERN_ERR "btrfs csum failed ino %llu off"
5600 " %llu csum %u private %u\n",
5601 (unsigned long long)btrfs_ino(inode),
5602 (unsigned long long)start,
5608 start += bvec->bv_len;
5611 } while (bvec <= bvec_end);
5613 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5614 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5615 bio->bi_private = dip->private;
5620 /* If we had a csum failure make sure to clear the uptodate flag */
5622 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5623 dio_end_io(bio, err);
5626 static void btrfs_endio_direct_write(struct bio *bio, int err)
5628 struct btrfs_dio_private *dip = bio->bi_private;
5629 struct inode *inode = dip->inode;
5630 struct btrfs_root *root = BTRFS_I(inode)->root;
5631 struct btrfs_trans_handle *trans;
5632 struct btrfs_ordered_extent *ordered = NULL;
5633 struct extent_state *cached_state = NULL;
5634 u64 ordered_offset = dip->logical_offset;
5635 u64 ordered_bytes = dip->bytes;
5641 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5649 trans = btrfs_join_transaction(root);
5650 if (IS_ERR(trans)) {
5654 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5656 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5657 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5659 err = btrfs_update_inode_fallback(trans, root, inode);
5663 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5664 ordered->file_offset + ordered->len - 1, 0,
5665 &cached_state, GFP_NOFS);
5667 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5668 ret = btrfs_mark_extent_written(trans, inode,
5669 ordered->file_offset,
5670 ordered->file_offset +
5677 ret = insert_reserved_file_extent(trans, inode,
5678 ordered->file_offset,
5684 BTRFS_FILE_EXTENT_REG);
5685 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5686 ordered->file_offset, ordered->len);
5694 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5695 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5696 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5697 btrfs_update_inode_fallback(trans, root, inode);
5700 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5701 ordered->file_offset + ordered->len - 1,
5702 &cached_state, GFP_NOFS);
5704 btrfs_delalloc_release_metadata(inode, ordered->len);
5705 btrfs_end_transaction(trans, root);
5706 ordered_offset = ordered->file_offset + ordered->len;
5707 btrfs_put_ordered_extent(ordered);
5708 btrfs_put_ordered_extent(ordered);
5712 * our bio might span multiple ordered extents. If we haven't
5713 * completed the accounting for the whole dio, go back and try again
5715 if (ordered_offset < dip->logical_offset + dip->bytes) {
5716 ordered_bytes = dip->logical_offset + dip->bytes -
5721 bio->bi_private = dip->private;
5726 /* If we had an error make sure to clear the uptodate flag */
5728 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5729 dio_end_io(bio, err);
5732 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5733 struct bio *bio, int mirror_num,
5734 unsigned long bio_flags, u64 offset)
5737 struct btrfs_root *root = BTRFS_I(inode)->root;
5738 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5743 static void btrfs_end_dio_bio(struct bio *bio, int err)
5745 struct btrfs_dio_private *dip = bio->bi_private;
5748 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5749 "sector %#Lx len %u err no %d\n",
5750 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5751 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5755 * before atomic variable goto zero, we must make sure
5756 * dip->errors is perceived to be set.
5758 smp_mb__before_atomic_dec();
5761 /* if there are more bios still pending for this dio, just exit */
5762 if (!atomic_dec_and_test(&dip->pending_bios))
5766 bio_io_error(dip->orig_bio);
5768 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5769 bio_endio(dip->orig_bio, 0);
5775 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5776 u64 first_sector, gfp_t gfp_flags)
5778 int nr_vecs = bio_get_nr_vecs(bdev);
5779 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5782 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5783 int rw, u64 file_offset, int skip_sum,
5784 u32 *csums, int async_submit)
5786 int write = rw & REQ_WRITE;
5787 struct btrfs_root *root = BTRFS_I(inode)->root;
5791 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5798 if (write && async_submit) {
5799 ret = btrfs_wq_submit_bio(root->fs_info,
5800 inode, rw, bio, 0, 0,
5802 __btrfs_submit_bio_start_direct_io,
5803 __btrfs_submit_bio_done);
5807 * If we aren't doing async submit, calculate the csum of the
5810 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5813 } else if (!skip_sum) {
5814 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5815 file_offset, csums);
5821 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5827 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5830 struct inode *inode = dip->inode;
5831 struct btrfs_root *root = BTRFS_I(inode)->root;
5832 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5834 struct bio *orig_bio = dip->orig_bio;
5835 struct bio_vec *bvec = orig_bio->bi_io_vec;
5836 u64 start_sector = orig_bio->bi_sector;
5837 u64 file_offset = dip->logical_offset;
5841 u32 *csums = dip->csums;
5843 int async_submit = 0;
5844 int write = rw & REQ_WRITE;
5846 map_length = orig_bio->bi_size;
5847 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5848 &map_length, NULL, 0);
5854 if (map_length >= orig_bio->bi_size) {
5860 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5863 bio->bi_private = dip;
5864 bio->bi_end_io = btrfs_end_dio_bio;
5865 atomic_inc(&dip->pending_bios);
5867 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5868 if (unlikely(map_length < submit_len + bvec->bv_len ||
5869 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5870 bvec->bv_offset) < bvec->bv_len)) {
5872 * inc the count before we submit the bio so
5873 * we know the end IO handler won't happen before
5874 * we inc the count. Otherwise, the dip might get freed
5875 * before we're done setting it up
5877 atomic_inc(&dip->pending_bios);
5878 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5879 file_offset, skip_sum,
5880 csums, async_submit);
5883 atomic_dec(&dip->pending_bios);
5887 /* Write's use the ordered csums */
5888 if (!write && !skip_sum)
5889 csums = csums + nr_pages;
5890 start_sector += submit_len >> 9;
5891 file_offset += submit_len;
5896 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5897 start_sector, GFP_NOFS);
5900 bio->bi_private = dip;
5901 bio->bi_end_io = btrfs_end_dio_bio;
5903 map_length = orig_bio->bi_size;
5904 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5905 &map_length, NULL, 0);
5911 submit_len += bvec->bv_len;
5918 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5919 csums, async_submit);
5927 * before atomic variable goto zero, we must
5928 * make sure dip->errors is perceived to be set.
5930 smp_mb__before_atomic_dec();
5931 if (atomic_dec_and_test(&dip->pending_bios))
5932 bio_io_error(dip->orig_bio);
5934 /* bio_end_io() will handle error, so we needn't return it */
5938 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5941 struct btrfs_root *root = BTRFS_I(inode)->root;
5942 struct btrfs_dio_private *dip;
5943 struct bio_vec *bvec = bio->bi_io_vec;
5945 int write = rw & REQ_WRITE;
5948 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5950 dip = kmalloc(sizeof(*dip), GFP_NOFS);
5957 /* Write's use the ordered csum stuff, so we don't need dip->csums */
5958 if (!write && !skip_sum) {
5959 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5967 dip->private = bio->bi_private;
5969 dip->logical_offset = file_offset;
5973 dip->bytes += bvec->bv_len;
5975 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5977 dip->disk_bytenr = (u64)bio->bi_sector << 9;
5978 bio->bi_private = dip;
5980 dip->orig_bio = bio;
5981 atomic_set(&dip->pending_bios, 0);
5984 bio->bi_end_io = btrfs_endio_direct_write;
5986 bio->bi_end_io = btrfs_endio_direct_read;
5988 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
5993 * If this is a write, we need to clean up the reserved space and kill
5994 * the ordered extent.
5997 struct btrfs_ordered_extent *ordered;
5998 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
5999 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6000 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6001 btrfs_free_reserved_extent(root, ordered->start,
6003 btrfs_put_ordered_extent(ordered);
6004 btrfs_put_ordered_extent(ordered);
6006 bio_endio(bio, ret);
6009 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6010 const struct iovec *iov, loff_t offset,
6011 unsigned long nr_segs)
6017 unsigned blocksize_mask = root->sectorsize - 1;
6018 ssize_t retval = -EINVAL;
6019 loff_t end = offset;
6021 if (offset & blocksize_mask)
6024 /* Check the memory alignment. Blocks cannot straddle pages */
6025 for (seg = 0; seg < nr_segs; seg++) {
6026 addr = (unsigned long)iov[seg].iov_base;
6027 size = iov[seg].iov_len;
6029 if ((addr & blocksize_mask) || (size & blocksize_mask))
6032 /* If this is a write we don't need to check anymore */
6037 * Check to make sure we don't have duplicate iov_base's in this
6038 * iovec, if so return EINVAL, otherwise we'll get csum errors
6039 * when reading back.
6041 for (i = seg + 1; i < nr_segs; i++) {
6042 if (iov[seg].iov_base == iov[i].iov_base)
6050 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6051 const struct iovec *iov, loff_t offset,
6052 unsigned long nr_segs)
6054 struct file *file = iocb->ki_filp;
6055 struct inode *inode = file->f_mapping->host;
6056 struct btrfs_ordered_extent *ordered;
6057 struct extent_state *cached_state = NULL;
6058 u64 lockstart, lockend;
6060 int writing = rw & WRITE;
6062 size_t count = iov_length(iov, nr_segs);
6064 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6070 lockend = offset + count - 1;
6073 ret = btrfs_delalloc_reserve_space(inode, count);
6079 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6080 0, &cached_state, GFP_NOFS);
6082 * We're concerned with the entire range that we're going to be
6083 * doing DIO to, so we need to make sure theres no ordered
6084 * extents in this range.
6086 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6087 lockend - lockstart + 1);
6090 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6091 &cached_state, GFP_NOFS);
6092 btrfs_start_ordered_extent(inode, ordered, 1);
6093 btrfs_put_ordered_extent(ordered);
6098 * we don't use btrfs_set_extent_delalloc because we don't want
6099 * the dirty or uptodate bits
6102 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6103 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6104 EXTENT_DELALLOC, 0, NULL, &cached_state,
6107 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6108 lockend, EXTENT_LOCKED | write_bits,
6109 1, 0, &cached_state, GFP_NOFS);
6114 free_extent_state(cached_state);
6115 cached_state = NULL;
6117 ret = __blockdev_direct_IO(rw, iocb, inode,
6118 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6119 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6120 btrfs_submit_direct, 0);
6122 if (ret < 0 && ret != -EIOCBQUEUED) {
6123 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6124 offset + iov_length(iov, nr_segs) - 1,
6125 EXTENT_LOCKED | write_bits, 1, 0,
6126 &cached_state, GFP_NOFS);
6127 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6129 * We're falling back to buffered, unlock the section we didn't
6132 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6133 offset + iov_length(iov, nr_segs) - 1,
6134 EXTENT_LOCKED | write_bits, 1, 0,
6135 &cached_state, GFP_NOFS);
6138 free_extent_state(cached_state);
6142 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6143 __u64 start, __u64 len)
6145 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6148 int btrfs_readpage(struct file *file, struct page *page)
6150 struct extent_io_tree *tree;
6151 tree = &BTRFS_I(page->mapping->host)->io_tree;
6152 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6155 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6157 struct extent_io_tree *tree;
6160 if (current->flags & PF_MEMALLOC) {
6161 redirty_page_for_writepage(wbc, page);
6165 tree = &BTRFS_I(page->mapping->host)->io_tree;
6166 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6169 int btrfs_writepages(struct address_space *mapping,
6170 struct writeback_control *wbc)
6172 struct extent_io_tree *tree;
6174 tree = &BTRFS_I(mapping->host)->io_tree;
6175 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6179 btrfs_readpages(struct file *file, struct address_space *mapping,
6180 struct list_head *pages, unsigned nr_pages)
6182 struct extent_io_tree *tree;
6183 tree = &BTRFS_I(mapping->host)->io_tree;
6184 return extent_readpages(tree, mapping, pages, nr_pages,
6187 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6189 struct extent_io_tree *tree;
6190 struct extent_map_tree *map;
6193 tree = &BTRFS_I(page->mapping->host)->io_tree;
6194 map = &BTRFS_I(page->mapping->host)->extent_tree;
6195 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6197 ClearPagePrivate(page);
6198 set_page_private(page, 0);
6199 page_cache_release(page);
6204 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6206 if (PageWriteback(page) || PageDirty(page))
6208 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6211 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6213 struct extent_io_tree *tree;
6214 struct btrfs_ordered_extent *ordered;
6215 struct extent_state *cached_state = NULL;
6216 u64 page_start = page_offset(page);
6217 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6221 * we have the page locked, so new writeback can't start,
6222 * and the dirty bit won't be cleared while we are here.
6224 * Wait for IO on this page so that we can safely clear
6225 * the PagePrivate2 bit and do ordered accounting
6227 wait_on_page_writeback(page);
6229 tree = &BTRFS_I(page->mapping->host)->io_tree;
6231 btrfs_releasepage(page, GFP_NOFS);
6234 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6236 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6240 * IO on this page will never be started, so we need
6241 * to account for any ordered extents now
6243 clear_extent_bit(tree, page_start, page_end,
6244 EXTENT_DIRTY | EXTENT_DELALLOC |
6245 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6246 &cached_state, GFP_NOFS);
6248 * whoever cleared the private bit is responsible
6249 * for the finish_ordered_io
6251 if (TestClearPagePrivate2(page)) {
6252 btrfs_finish_ordered_io(page->mapping->host,
6253 page_start, page_end);
6255 btrfs_put_ordered_extent(ordered);
6256 cached_state = NULL;
6257 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6260 clear_extent_bit(tree, page_start, page_end,
6261 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6262 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6263 __btrfs_releasepage(page, GFP_NOFS);
6265 ClearPageChecked(page);
6266 if (PagePrivate(page)) {
6267 ClearPagePrivate(page);
6268 set_page_private(page, 0);
6269 page_cache_release(page);
6274 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6275 * called from a page fault handler when a page is first dirtied. Hence we must
6276 * be careful to check for EOF conditions here. We set the page up correctly
6277 * for a written page which means we get ENOSPC checking when writing into
6278 * holes and correct delalloc and unwritten extent mapping on filesystems that
6279 * support these features.
6281 * We are not allowed to take the i_mutex here so we have to play games to
6282 * protect against truncate races as the page could now be beyond EOF. Because
6283 * vmtruncate() writes the inode size before removing pages, once we have the
6284 * page lock we can determine safely if the page is beyond EOF. If it is not
6285 * beyond EOF, then the page is guaranteed safe against truncation until we
6288 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6290 struct page *page = vmf->page;
6291 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6292 struct btrfs_root *root = BTRFS_I(inode)->root;
6293 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6294 struct btrfs_ordered_extent *ordered;
6295 struct extent_state *cached_state = NULL;
6297 unsigned long zero_start;
6303 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6307 else /* -ENOSPC, -EIO, etc */
6308 ret = VM_FAULT_SIGBUS;
6312 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6315 size = i_size_read(inode);
6316 page_start = page_offset(page);
6317 page_end = page_start + PAGE_CACHE_SIZE - 1;
6319 if ((page->mapping != inode->i_mapping) ||
6320 (page_start >= size)) {
6321 /* page got truncated out from underneath us */
6324 wait_on_page_writeback(page);
6326 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6328 set_page_extent_mapped(page);
6331 * we can't set the delalloc bits if there are pending ordered
6332 * extents. Drop our locks and wait for them to finish
6334 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6336 unlock_extent_cached(io_tree, page_start, page_end,
6337 &cached_state, GFP_NOFS);
6339 btrfs_start_ordered_extent(inode, ordered, 1);
6340 btrfs_put_ordered_extent(ordered);
6345 * XXX - page_mkwrite gets called every time the page is dirtied, even
6346 * if it was already dirty, so for space accounting reasons we need to
6347 * clear any delalloc bits for the range we are fixing to save. There
6348 * is probably a better way to do this, but for now keep consistent with
6349 * prepare_pages in the normal write path.
6351 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6352 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6353 0, 0, &cached_state, GFP_NOFS);
6355 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6358 unlock_extent_cached(io_tree, page_start, page_end,
6359 &cached_state, GFP_NOFS);
6360 ret = VM_FAULT_SIGBUS;
6365 /* page is wholly or partially inside EOF */
6366 if (page_start + PAGE_CACHE_SIZE > size)
6367 zero_start = size & ~PAGE_CACHE_MASK;
6369 zero_start = PAGE_CACHE_SIZE;
6371 if (zero_start != PAGE_CACHE_SIZE) {
6373 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6374 flush_dcache_page(page);
6377 ClearPageChecked(page);
6378 set_page_dirty(page);
6379 SetPageUptodate(page);
6381 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6382 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6384 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6388 return VM_FAULT_LOCKED;
6390 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6395 static int btrfs_truncate(struct inode *inode)
6397 struct btrfs_root *root = BTRFS_I(inode)->root;
6398 struct btrfs_block_rsv *rsv;
6401 struct btrfs_trans_handle *trans;
6403 u64 mask = root->sectorsize - 1;
6404 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6406 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6410 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6411 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6414 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6415 * 3 things going on here
6417 * 1) We need to reserve space for our orphan item and the space to
6418 * delete our orphan item. Lord knows we don't want to have a dangling
6419 * orphan item because we didn't reserve space to remove it.
6421 * 2) We need to reserve space to update our inode.
6423 * 3) We need to have something to cache all the space that is going to
6424 * be free'd up by the truncate operation, but also have some slack
6425 * space reserved in case it uses space during the truncate (thank you
6426 * very much snapshotting).
6428 * And we need these to all be seperate. The fact is we can use alot of
6429 * space doing the truncate, and we have no earthly idea how much space
6430 * we will use, so we need the truncate reservation to be seperate so it
6431 * doesn't end up using space reserved for updating the inode or
6432 * removing the orphan item. We also need to be able to stop the
6433 * transaction and start a new one, which means we need to be able to
6434 * update the inode several times, and we have no idea of knowing how
6435 * many times that will be, so we can't just reserve 1 item for the
6436 * entirety of the opration, so that has to be done seperately as well.
6437 * Then there is the orphan item, which does indeed need to be held on
6438 * to for the whole operation, and we need nobody to touch this reserved
6439 * space except the orphan code.
6441 * So that leaves us with
6443 * 1) root->orphan_block_rsv - for the orphan deletion.
6444 * 2) rsv - for the truncate reservation, which we will steal from the
6445 * transaction reservation.
6446 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6447 * updating the inode.
6449 rsv = btrfs_alloc_block_rsv(root);
6452 rsv->size = min_size;
6455 * 1 for the truncate slack space
6456 * 1 for the orphan item we're going to add
6457 * 1 for the orphan item deletion
6458 * 1 for updating the inode.
6460 trans = btrfs_start_transaction(root, 4);
6461 if (IS_ERR(trans)) {
6462 err = PTR_ERR(trans);
6466 /* Migrate the slack space for the truncate to our reserve */
6467 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6471 ret = btrfs_orphan_add(trans, inode);
6473 btrfs_end_transaction(trans, root);
6478 * setattr is responsible for setting the ordered_data_close flag,
6479 * but that is only tested during the last file release. That
6480 * could happen well after the next commit, leaving a great big
6481 * window where new writes may get lost if someone chooses to write
6482 * to this file after truncating to zero
6484 * The inode doesn't have any dirty data here, and so if we commit
6485 * this is a noop. If someone immediately starts writing to the inode
6486 * it is very likely we'll catch some of their writes in this
6487 * transaction, and the commit will find this file on the ordered
6488 * data list with good things to send down.
6490 * This is a best effort solution, there is still a window where
6491 * using truncate to replace the contents of the file will
6492 * end up with a zero length file after a crash.
6494 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6495 btrfs_add_ordered_operation(trans, root, inode);
6498 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6501 * This can only happen with the original transaction we
6502 * started above, every other time we shouldn't have a
6503 * transaction started yet.
6512 /* Just need the 1 for updating the inode */
6513 trans = btrfs_start_transaction(root, 1);
6514 if (IS_ERR(trans)) {
6515 err = PTR_ERR(trans);
6520 trans->block_rsv = rsv;
6522 ret = btrfs_truncate_inode_items(trans, root, inode,
6524 BTRFS_EXTENT_DATA_KEY);
6525 if (ret != -EAGAIN) {
6530 trans->block_rsv = &root->fs_info->trans_block_rsv;
6531 ret = btrfs_update_inode(trans, root, inode);
6537 nr = trans->blocks_used;
6538 btrfs_end_transaction(trans, root);
6540 btrfs_btree_balance_dirty(root, nr);
6543 if (ret == 0 && inode->i_nlink > 0) {
6544 trans->block_rsv = root->orphan_block_rsv;
6545 ret = btrfs_orphan_del(trans, inode);
6548 } else if (ret && inode->i_nlink > 0) {
6550 * Failed to do the truncate, remove us from the in memory
6553 ret = btrfs_orphan_del(NULL, inode);
6557 trans->block_rsv = &root->fs_info->trans_block_rsv;
6558 ret = btrfs_update_inode(trans, root, inode);
6562 nr = trans->blocks_used;
6563 ret = btrfs_end_transaction_throttle(trans, root);
6564 btrfs_btree_balance_dirty(root, nr);
6568 btrfs_free_block_rsv(root, rsv);
6577 * create a new subvolume directory/inode (helper for the ioctl).
6579 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6580 struct btrfs_root *new_root, u64 new_dirid)
6582 struct inode *inode;
6586 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6587 new_dirid, S_IFDIR | 0700, &index);
6589 return PTR_ERR(inode);
6590 inode->i_op = &btrfs_dir_inode_operations;
6591 inode->i_fop = &btrfs_dir_file_operations;
6594 btrfs_i_size_write(inode, 0);
6596 err = btrfs_update_inode(trans, new_root, inode);
6603 struct inode *btrfs_alloc_inode(struct super_block *sb)
6605 struct btrfs_inode *ei;
6606 struct inode *inode;
6608 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6613 ei->space_info = NULL;
6617 ei->last_sub_trans = 0;
6618 ei->logged_trans = 0;
6619 ei->delalloc_bytes = 0;
6620 ei->disk_i_size = 0;
6623 ei->index_cnt = (u64)-1;
6624 ei->last_unlink_trans = 0;
6626 spin_lock_init(&ei->lock);
6627 ei->outstanding_extents = 0;
6628 ei->reserved_extents = 0;
6630 ei->ordered_data_close = 0;
6631 ei->orphan_meta_reserved = 0;
6632 ei->dummy_inode = 0;
6634 ei->delalloc_meta_reserved = 0;
6635 ei->force_compress = BTRFS_COMPRESS_NONE;
6637 ei->delayed_node = NULL;
6639 inode = &ei->vfs_inode;
6640 extent_map_tree_init(&ei->extent_tree);
6641 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6642 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6643 mutex_init(&ei->log_mutex);
6644 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6645 INIT_LIST_HEAD(&ei->i_orphan);
6646 INIT_LIST_HEAD(&ei->delalloc_inodes);
6647 INIT_LIST_HEAD(&ei->ordered_operations);
6648 RB_CLEAR_NODE(&ei->rb_node);
6653 static void btrfs_i_callback(struct rcu_head *head)
6655 struct inode *inode = container_of(head, struct inode, i_rcu);
6656 INIT_LIST_HEAD(&inode->i_dentry);
6657 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6660 void btrfs_destroy_inode(struct inode *inode)
6662 struct btrfs_ordered_extent *ordered;
6663 struct btrfs_root *root = BTRFS_I(inode)->root;
6665 WARN_ON(!list_empty(&inode->i_dentry));
6666 WARN_ON(inode->i_data.nrpages);
6667 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6668 WARN_ON(BTRFS_I(inode)->reserved_extents);
6669 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6670 WARN_ON(BTRFS_I(inode)->csum_bytes);
6673 * This can happen where we create an inode, but somebody else also
6674 * created the same inode and we need to destroy the one we already
6681 * Make sure we're properly removed from the ordered operation
6685 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6686 spin_lock(&root->fs_info->ordered_extent_lock);
6687 list_del_init(&BTRFS_I(inode)->ordered_operations);
6688 spin_unlock(&root->fs_info->ordered_extent_lock);
6691 spin_lock(&root->orphan_lock);
6692 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6693 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6694 (unsigned long long)btrfs_ino(inode));
6695 list_del_init(&BTRFS_I(inode)->i_orphan);
6697 spin_unlock(&root->orphan_lock);
6700 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6704 printk(KERN_ERR "btrfs found ordered "
6705 "extent %llu %llu on inode cleanup\n",
6706 (unsigned long long)ordered->file_offset,
6707 (unsigned long long)ordered->len);
6708 btrfs_remove_ordered_extent(inode, ordered);
6709 btrfs_put_ordered_extent(ordered);
6710 btrfs_put_ordered_extent(ordered);
6713 inode_tree_del(inode);
6714 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6716 btrfs_remove_delayed_node(inode);
6717 call_rcu(&inode->i_rcu, btrfs_i_callback);
6720 int btrfs_drop_inode(struct inode *inode)
6722 struct btrfs_root *root = BTRFS_I(inode)->root;
6724 if (btrfs_root_refs(&root->root_item) == 0 &&
6725 !btrfs_is_free_space_inode(root, inode))
6728 return generic_drop_inode(inode);
6731 static void init_once(void *foo)
6733 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6735 inode_init_once(&ei->vfs_inode);
6738 void btrfs_destroy_cachep(void)
6740 if (btrfs_inode_cachep)
6741 kmem_cache_destroy(btrfs_inode_cachep);
6742 if (btrfs_trans_handle_cachep)
6743 kmem_cache_destroy(btrfs_trans_handle_cachep);
6744 if (btrfs_transaction_cachep)
6745 kmem_cache_destroy(btrfs_transaction_cachep);
6746 if (btrfs_path_cachep)
6747 kmem_cache_destroy(btrfs_path_cachep);
6748 if (btrfs_free_space_cachep)
6749 kmem_cache_destroy(btrfs_free_space_cachep);
6752 int btrfs_init_cachep(void)
6754 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6755 sizeof(struct btrfs_inode), 0,
6756 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6757 if (!btrfs_inode_cachep)
6760 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6761 sizeof(struct btrfs_trans_handle), 0,
6762 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6763 if (!btrfs_trans_handle_cachep)
6766 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6767 sizeof(struct btrfs_transaction), 0,
6768 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6769 if (!btrfs_transaction_cachep)
6772 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6773 sizeof(struct btrfs_path), 0,
6774 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6775 if (!btrfs_path_cachep)
6778 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6779 sizeof(struct btrfs_free_space), 0,
6780 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6781 if (!btrfs_free_space_cachep)
6786 btrfs_destroy_cachep();
6790 static int btrfs_getattr(struct vfsmount *mnt,
6791 struct dentry *dentry, struct kstat *stat)
6793 struct inode *inode = dentry->d_inode;
6794 generic_fillattr(inode, stat);
6795 stat->dev = BTRFS_I(inode)->root->anon_dev;
6796 stat->blksize = PAGE_CACHE_SIZE;
6797 stat->blocks = (inode_get_bytes(inode) +
6798 BTRFS_I(inode)->delalloc_bytes) >> 9;
6803 * If a file is moved, it will inherit the cow and compression flags of the new
6806 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6808 struct btrfs_inode *b_dir = BTRFS_I(dir);
6809 struct btrfs_inode *b_inode = BTRFS_I(inode);
6811 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6812 b_inode->flags |= BTRFS_INODE_NODATACOW;
6814 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6816 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6817 b_inode->flags |= BTRFS_INODE_COMPRESS;
6819 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6822 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6823 struct inode *new_dir, struct dentry *new_dentry)
6825 struct btrfs_trans_handle *trans;
6826 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6827 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6828 struct inode *new_inode = new_dentry->d_inode;
6829 struct inode *old_inode = old_dentry->d_inode;
6830 struct timespec ctime = CURRENT_TIME;
6834 u64 old_ino = btrfs_ino(old_inode);
6836 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6839 /* we only allow rename subvolume link between subvolumes */
6840 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6843 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6844 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6847 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6848 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6851 * we're using rename to replace one file with another.
6852 * and the replacement file is large. Start IO on it now so
6853 * we don't add too much work to the end of the transaction
6855 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6856 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6857 filemap_flush(old_inode->i_mapping);
6859 /* close the racy window with snapshot create/destroy ioctl */
6860 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6861 down_read(&root->fs_info->subvol_sem);
6863 * We want to reserve the absolute worst case amount of items. So if
6864 * both inodes are subvols and we need to unlink them then that would
6865 * require 4 item modifications, but if they are both normal inodes it
6866 * would require 5 item modifications, so we'll assume their normal
6867 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6868 * should cover the worst case number of items we'll modify.
6870 trans = btrfs_start_transaction(root, 20);
6871 if (IS_ERR(trans)) {
6872 ret = PTR_ERR(trans);
6877 btrfs_record_root_in_trans(trans, dest);
6879 ret = btrfs_set_inode_index(new_dir, &index);
6883 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6884 /* force full log commit if subvolume involved. */
6885 root->fs_info->last_trans_log_full_commit = trans->transid;
6887 ret = btrfs_insert_inode_ref(trans, dest,
6888 new_dentry->d_name.name,
6889 new_dentry->d_name.len,
6891 btrfs_ino(new_dir), index);
6895 * this is an ugly little race, but the rename is required
6896 * to make sure that if we crash, the inode is either at the
6897 * old name or the new one. pinning the log transaction lets
6898 * us make sure we don't allow a log commit to come in after
6899 * we unlink the name but before we add the new name back in.
6901 btrfs_pin_log_trans(root);
6904 * make sure the inode gets flushed if it is replacing
6907 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6908 btrfs_add_ordered_operation(trans, root, old_inode);
6910 old_dir->i_ctime = old_dir->i_mtime = ctime;
6911 new_dir->i_ctime = new_dir->i_mtime = ctime;
6912 old_inode->i_ctime = ctime;
6914 if (old_dentry->d_parent != new_dentry->d_parent)
6915 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6917 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6918 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6919 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6920 old_dentry->d_name.name,
6921 old_dentry->d_name.len);
6923 ret = __btrfs_unlink_inode(trans, root, old_dir,
6924 old_dentry->d_inode,
6925 old_dentry->d_name.name,
6926 old_dentry->d_name.len);
6928 ret = btrfs_update_inode(trans, root, old_inode);
6933 new_inode->i_ctime = CURRENT_TIME;
6934 if (unlikely(btrfs_ino(new_inode) ==
6935 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6936 root_objectid = BTRFS_I(new_inode)->location.objectid;
6937 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6939 new_dentry->d_name.name,
6940 new_dentry->d_name.len);
6941 BUG_ON(new_inode->i_nlink == 0);
6943 ret = btrfs_unlink_inode(trans, dest, new_dir,
6944 new_dentry->d_inode,
6945 new_dentry->d_name.name,
6946 new_dentry->d_name.len);
6949 if (new_inode->i_nlink == 0) {
6950 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6955 fixup_inode_flags(new_dir, old_inode);
6957 ret = btrfs_add_link(trans, new_dir, old_inode,
6958 new_dentry->d_name.name,
6959 new_dentry->d_name.len, 0, index);
6962 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
6963 struct dentry *parent = new_dentry->d_parent;
6964 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6965 btrfs_end_log_trans(root);
6968 btrfs_end_transaction_throttle(trans, root);
6970 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6971 up_read(&root->fs_info->subvol_sem);
6977 * some fairly slow code that needs optimization. This walks the list
6978 * of all the inodes with pending delalloc and forces them to disk.
6980 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6982 struct list_head *head = &root->fs_info->delalloc_inodes;
6983 struct btrfs_inode *binode;
6984 struct inode *inode;
6986 if (root->fs_info->sb->s_flags & MS_RDONLY)
6989 spin_lock(&root->fs_info->delalloc_lock);
6990 while (!list_empty(head)) {
6991 binode = list_entry(head->next, struct btrfs_inode,
6993 inode = igrab(&binode->vfs_inode);
6995 list_del_init(&binode->delalloc_inodes);
6996 spin_unlock(&root->fs_info->delalloc_lock);
6998 filemap_flush(inode->i_mapping);
7000 btrfs_add_delayed_iput(inode);
7005 spin_lock(&root->fs_info->delalloc_lock);
7007 spin_unlock(&root->fs_info->delalloc_lock);
7009 /* the filemap_flush will queue IO into the worker threads, but
7010 * we have to make sure the IO is actually started and that
7011 * ordered extents get created before we return
7013 atomic_inc(&root->fs_info->async_submit_draining);
7014 while (atomic_read(&root->fs_info->nr_async_submits) ||
7015 atomic_read(&root->fs_info->async_delalloc_pages)) {
7016 wait_event(root->fs_info->async_submit_wait,
7017 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7018 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7020 atomic_dec(&root->fs_info->async_submit_draining);
7024 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7025 const char *symname)
7027 struct btrfs_trans_handle *trans;
7028 struct btrfs_root *root = BTRFS_I(dir)->root;
7029 struct btrfs_path *path;
7030 struct btrfs_key key;
7031 struct inode *inode = NULL;
7039 struct btrfs_file_extent_item *ei;
7040 struct extent_buffer *leaf;
7041 unsigned long nr = 0;
7043 name_len = strlen(symname) + 1;
7044 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7045 return -ENAMETOOLONG;
7048 * 2 items for inode item and ref
7049 * 2 items for dir items
7050 * 1 item for xattr if selinux is on
7052 trans = btrfs_start_transaction(root, 5);
7054 return PTR_ERR(trans);
7056 err = btrfs_find_free_ino(root, &objectid);
7060 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7061 dentry->d_name.len, btrfs_ino(dir), objectid,
7062 S_IFLNK|S_IRWXUGO, &index);
7063 if (IS_ERR(inode)) {
7064 err = PTR_ERR(inode);
7068 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7074 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7078 inode->i_mapping->a_ops = &btrfs_aops;
7079 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7080 inode->i_fop = &btrfs_file_operations;
7081 inode->i_op = &btrfs_file_inode_operations;
7082 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7087 path = btrfs_alloc_path();
7093 key.objectid = btrfs_ino(inode);
7095 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7096 datasize = btrfs_file_extent_calc_inline_size(name_len);
7097 err = btrfs_insert_empty_item(trans, root, path, &key,
7101 btrfs_free_path(path);
7104 leaf = path->nodes[0];
7105 ei = btrfs_item_ptr(leaf, path->slots[0],
7106 struct btrfs_file_extent_item);
7107 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7108 btrfs_set_file_extent_type(leaf, ei,
7109 BTRFS_FILE_EXTENT_INLINE);
7110 btrfs_set_file_extent_encryption(leaf, ei, 0);
7111 btrfs_set_file_extent_compression(leaf, ei, 0);
7112 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7113 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7115 ptr = btrfs_file_extent_inline_start(ei);
7116 write_extent_buffer(leaf, symname, ptr, name_len);
7117 btrfs_mark_buffer_dirty(leaf);
7118 btrfs_free_path(path);
7120 inode->i_op = &btrfs_symlink_inode_operations;
7121 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7122 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7123 inode_set_bytes(inode, name_len);
7124 btrfs_i_size_write(inode, name_len - 1);
7125 err = btrfs_update_inode(trans, root, inode);
7130 nr = trans->blocks_used;
7131 btrfs_end_transaction_throttle(trans, root);
7133 inode_dec_link_count(inode);
7136 btrfs_btree_balance_dirty(root, nr);
7140 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7141 u64 start, u64 num_bytes, u64 min_size,
7142 loff_t actual_len, u64 *alloc_hint,
7143 struct btrfs_trans_handle *trans)
7145 struct btrfs_root *root = BTRFS_I(inode)->root;
7146 struct btrfs_key ins;
7147 u64 cur_offset = start;
7150 bool own_trans = true;
7154 while (num_bytes > 0) {
7156 trans = btrfs_start_transaction(root, 3);
7157 if (IS_ERR(trans)) {
7158 ret = PTR_ERR(trans);
7163 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7164 0, *alloc_hint, (u64)-1, &ins, 1);
7167 btrfs_end_transaction(trans, root);
7171 ret = insert_reserved_file_extent(trans, inode,
7172 cur_offset, ins.objectid,
7173 ins.offset, ins.offset,
7174 ins.offset, 0, 0, 0,
7175 BTRFS_FILE_EXTENT_PREALLOC);
7177 btrfs_drop_extent_cache(inode, cur_offset,
7178 cur_offset + ins.offset -1, 0);
7180 num_bytes -= ins.offset;
7181 cur_offset += ins.offset;
7182 *alloc_hint = ins.objectid + ins.offset;
7184 inode->i_ctime = CURRENT_TIME;
7185 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7186 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7187 (actual_len > inode->i_size) &&
7188 (cur_offset > inode->i_size)) {
7189 if (cur_offset > actual_len)
7190 i_size = actual_len;
7192 i_size = cur_offset;
7193 i_size_write(inode, i_size);
7194 btrfs_ordered_update_i_size(inode, i_size, NULL);
7197 ret = btrfs_update_inode(trans, root, inode);
7201 btrfs_end_transaction(trans, root);
7206 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7207 u64 start, u64 num_bytes, u64 min_size,
7208 loff_t actual_len, u64 *alloc_hint)
7210 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7211 min_size, actual_len, alloc_hint,
7215 int btrfs_prealloc_file_range_trans(struct inode *inode,
7216 struct btrfs_trans_handle *trans, int mode,
7217 u64 start, u64 num_bytes, u64 min_size,
7218 loff_t actual_len, u64 *alloc_hint)
7220 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7221 min_size, actual_len, alloc_hint, trans);
7224 static int btrfs_set_page_dirty(struct page *page)
7226 return __set_page_dirty_nobuffers(page);
7229 static int btrfs_permission(struct inode *inode, int mask)
7231 struct btrfs_root *root = BTRFS_I(inode)->root;
7232 umode_t mode = inode->i_mode;
7234 if (mask & MAY_WRITE &&
7235 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7236 if (btrfs_root_readonly(root))
7238 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7241 return generic_permission(inode, mask);
7244 static const struct inode_operations btrfs_dir_inode_operations = {
7245 .getattr = btrfs_getattr,
7246 .lookup = btrfs_lookup,
7247 .create = btrfs_create,
7248 .unlink = btrfs_unlink,
7250 .mkdir = btrfs_mkdir,
7251 .rmdir = btrfs_rmdir,
7252 .rename = btrfs_rename,
7253 .symlink = btrfs_symlink,
7254 .setattr = btrfs_setattr,
7255 .mknod = btrfs_mknod,
7256 .setxattr = btrfs_setxattr,
7257 .getxattr = btrfs_getxattr,
7258 .listxattr = btrfs_listxattr,
7259 .removexattr = btrfs_removexattr,
7260 .permission = btrfs_permission,
7261 .get_acl = btrfs_get_acl,
7263 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7264 .lookup = btrfs_lookup,
7265 .permission = btrfs_permission,
7266 .get_acl = btrfs_get_acl,
7269 static const struct file_operations btrfs_dir_file_operations = {
7270 .llseek = generic_file_llseek,
7271 .read = generic_read_dir,
7272 .readdir = btrfs_real_readdir,
7273 .unlocked_ioctl = btrfs_ioctl,
7274 #ifdef CONFIG_COMPAT
7275 .compat_ioctl = btrfs_ioctl,
7277 .release = btrfs_release_file,
7278 .fsync = btrfs_sync_file,
7281 static struct extent_io_ops btrfs_extent_io_ops = {
7282 .fill_delalloc = run_delalloc_range,
7283 .submit_bio_hook = btrfs_submit_bio_hook,
7284 .merge_bio_hook = btrfs_merge_bio_hook,
7285 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7286 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7287 .writepage_start_hook = btrfs_writepage_start_hook,
7288 .set_bit_hook = btrfs_set_bit_hook,
7289 .clear_bit_hook = btrfs_clear_bit_hook,
7290 .merge_extent_hook = btrfs_merge_extent_hook,
7291 .split_extent_hook = btrfs_split_extent_hook,
7295 * btrfs doesn't support the bmap operation because swapfiles
7296 * use bmap to make a mapping of extents in the file. They assume
7297 * these extents won't change over the life of the file and they
7298 * use the bmap result to do IO directly to the drive.
7300 * the btrfs bmap call would return logical addresses that aren't
7301 * suitable for IO and they also will change frequently as COW
7302 * operations happen. So, swapfile + btrfs == corruption.
7304 * For now we're avoiding this by dropping bmap.
7306 static const struct address_space_operations btrfs_aops = {
7307 .readpage = btrfs_readpage,
7308 .writepage = btrfs_writepage,
7309 .writepages = btrfs_writepages,
7310 .readpages = btrfs_readpages,
7311 .direct_IO = btrfs_direct_IO,
7312 .invalidatepage = btrfs_invalidatepage,
7313 .releasepage = btrfs_releasepage,
7314 .set_page_dirty = btrfs_set_page_dirty,
7315 .error_remove_page = generic_error_remove_page,
7318 static const struct address_space_operations btrfs_symlink_aops = {
7319 .readpage = btrfs_readpage,
7320 .writepage = btrfs_writepage,
7321 .invalidatepage = btrfs_invalidatepage,
7322 .releasepage = btrfs_releasepage,
7325 static const struct inode_operations btrfs_file_inode_operations = {
7326 .getattr = btrfs_getattr,
7327 .setattr = btrfs_setattr,
7328 .setxattr = btrfs_setxattr,
7329 .getxattr = btrfs_getxattr,
7330 .listxattr = btrfs_listxattr,
7331 .removexattr = btrfs_removexattr,
7332 .permission = btrfs_permission,
7333 .fiemap = btrfs_fiemap,
7334 .get_acl = btrfs_get_acl,
7336 static const struct inode_operations btrfs_special_inode_operations = {
7337 .getattr = btrfs_getattr,
7338 .setattr = btrfs_setattr,
7339 .permission = btrfs_permission,
7340 .setxattr = btrfs_setxattr,
7341 .getxattr = btrfs_getxattr,
7342 .listxattr = btrfs_listxattr,
7343 .removexattr = btrfs_removexattr,
7344 .get_acl = btrfs_get_acl,
7346 static const struct inode_operations btrfs_symlink_inode_operations = {
7347 .readlink = generic_readlink,
7348 .follow_link = page_follow_link_light,
7349 .put_link = page_put_link,
7350 .getattr = btrfs_getattr,
7351 .permission = btrfs_permission,
7352 .setxattr = btrfs_setxattr,
7353 .getxattr = btrfs_getxattr,
7354 .listxattr = btrfs_listxattr,
7355 .removexattr = btrfs_removexattr,
7356 .get_acl = btrfs_get_acl,
7359 const struct dentry_operations btrfs_dentry_operations = {
7360 .d_delete = btrfs_dentry_delete,
7361 .d_release = btrfs_dentry_release,