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 /* release the path since we're done with it */
2205 btrfs_release_path(path);
2207 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2209 if (root->orphan_block_rsv)
2210 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2213 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2214 trans = btrfs_join_transaction(root);
2216 btrfs_end_transaction(trans, root);
2220 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2222 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2226 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2227 btrfs_free_path(path);
2232 * very simple check to peek ahead in the leaf looking for xattrs. If we
2233 * don't find any xattrs, we know there can't be any acls.
2235 * slot is the slot the inode is in, objectid is the objectid of the inode
2237 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2238 int slot, u64 objectid)
2240 u32 nritems = btrfs_header_nritems(leaf);
2241 struct btrfs_key found_key;
2245 while (slot < nritems) {
2246 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2248 /* we found a different objectid, there must not be acls */
2249 if (found_key.objectid != objectid)
2252 /* we found an xattr, assume we've got an acl */
2253 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2257 * we found a key greater than an xattr key, there can't
2258 * be any acls later on
2260 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2267 * it goes inode, inode backrefs, xattrs, extents,
2268 * so if there are a ton of hard links to an inode there can
2269 * be a lot of backrefs. Don't waste time searching too hard,
2270 * this is just an optimization
2275 /* we hit the end of the leaf before we found an xattr or
2276 * something larger than an xattr. We have to assume the inode
2283 * read an inode from the btree into the in-memory inode
2285 static void btrfs_read_locked_inode(struct inode *inode)
2287 struct btrfs_path *path;
2288 struct extent_buffer *leaf;
2289 struct btrfs_inode_item *inode_item;
2290 struct btrfs_timespec *tspec;
2291 struct btrfs_root *root = BTRFS_I(inode)->root;
2292 struct btrfs_key location;
2296 bool filled = false;
2298 ret = btrfs_fill_inode(inode, &rdev);
2302 path = btrfs_alloc_path();
2306 path->leave_spinning = 1;
2307 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2309 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2313 leaf = path->nodes[0];
2318 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2319 struct btrfs_inode_item);
2320 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2321 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2322 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2323 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2324 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2326 tspec = btrfs_inode_atime(inode_item);
2327 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2328 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2330 tspec = btrfs_inode_mtime(inode_item);
2331 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2332 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2334 tspec = btrfs_inode_ctime(inode_item);
2335 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2336 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2338 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2339 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2340 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2341 inode->i_generation = BTRFS_I(inode)->generation;
2343 rdev = btrfs_inode_rdev(leaf, inode_item);
2345 BTRFS_I(inode)->index_cnt = (u64)-1;
2346 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2349 * try to precache a NULL acl entry for files that don't have
2350 * any xattrs or acls
2352 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2355 cache_no_acl(inode);
2357 btrfs_free_path(path);
2359 switch (inode->i_mode & S_IFMT) {
2361 inode->i_mapping->a_ops = &btrfs_aops;
2362 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2363 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2364 inode->i_fop = &btrfs_file_operations;
2365 inode->i_op = &btrfs_file_inode_operations;
2368 inode->i_fop = &btrfs_dir_file_operations;
2369 if (root == root->fs_info->tree_root)
2370 inode->i_op = &btrfs_dir_ro_inode_operations;
2372 inode->i_op = &btrfs_dir_inode_operations;
2375 inode->i_op = &btrfs_symlink_inode_operations;
2376 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2377 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2380 inode->i_op = &btrfs_special_inode_operations;
2381 init_special_inode(inode, inode->i_mode, rdev);
2385 btrfs_update_iflags(inode);
2389 btrfs_free_path(path);
2390 make_bad_inode(inode);
2394 * given a leaf and an inode, copy the inode fields into the leaf
2396 static void fill_inode_item(struct btrfs_trans_handle *trans,
2397 struct extent_buffer *leaf,
2398 struct btrfs_inode_item *item,
2399 struct inode *inode)
2401 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2402 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2403 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2404 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2405 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2407 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2408 inode->i_atime.tv_sec);
2409 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2410 inode->i_atime.tv_nsec);
2412 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2413 inode->i_mtime.tv_sec);
2414 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2415 inode->i_mtime.tv_nsec);
2417 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2418 inode->i_ctime.tv_sec);
2419 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2420 inode->i_ctime.tv_nsec);
2422 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2423 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2424 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2425 btrfs_set_inode_transid(leaf, item, trans->transid);
2426 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2427 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2428 btrfs_set_inode_block_group(leaf, item, 0);
2432 * copy everything in the in-memory inode into the btree.
2434 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2435 struct btrfs_root *root, struct inode *inode)
2437 struct btrfs_inode_item *inode_item;
2438 struct btrfs_path *path;
2439 struct extent_buffer *leaf;
2442 path = btrfs_alloc_path();
2446 path->leave_spinning = 1;
2447 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2455 btrfs_unlock_up_safe(path, 1);
2456 leaf = path->nodes[0];
2457 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2458 struct btrfs_inode_item);
2460 fill_inode_item(trans, leaf, inode_item, inode);
2461 btrfs_mark_buffer_dirty(leaf);
2462 btrfs_set_inode_last_trans(trans, inode);
2465 btrfs_free_path(path);
2470 * copy everything in the in-memory inode into the btree.
2472 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2473 struct btrfs_root *root, struct inode *inode)
2478 * If the inode is a free space inode, we can deadlock during commit
2479 * if we put it into the delayed code.
2481 * The data relocation inode should also be directly updated
2484 if (!btrfs_is_free_space_inode(root, inode)
2485 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2486 ret = btrfs_delayed_update_inode(trans, root, inode);
2488 btrfs_set_inode_last_trans(trans, inode);
2492 return btrfs_update_inode_item(trans, root, inode);
2495 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2496 struct btrfs_root *root, struct inode *inode)
2500 ret = btrfs_update_inode(trans, root, inode);
2502 return btrfs_update_inode_item(trans, root, inode);
2507 * unlink helper that gets used here in inode.c and in the tree logging
2508 * recovery code. It remove a link in a directory with a given name, and
2509 * also drops the back refs in the inode to the directory
2511 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2512 struct btrfs_root *root,
2513 struct inode *dir, struct inode *inode,
2514 const char *name, int name_len)
2516 struct btrfs_path *path;
2518 struct extent_buffer *leaf;
2519 struct btrfs_dir_item *di;
2520 struct btrfs_key key;
2522 u64 ino = btrfs_ino(inode);
2523 u64 dir_ino = btrfs_ino(dir);
2525 path = btrfs_alloc_path();
2531 path->leave_spinning = 1;
2532 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2533 name, name_len, -1);
2542 leaf = path->nodes[0];
2543 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2544 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2547 btrfs_release_path(path);
2549 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2552 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2553 "inode %llu parent %llu\n", name_len, name,
2554 (unsigned long long)ino, (unsigned long long)dir_ino);
2558 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2562 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2564 BUG_ON(ret != 0 && ret != -ENOENT);
2566 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2571 btrfs_free_path(path);
2575 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2576 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2577 btrfs_update_inode(trans, root, dir);
2582 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2583 struct btrfs_root *root,
2584 struct inode *dir, struct inode *inode,
2585 const char *name, int name_len)
2588 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2590 btrfs_drop_nlink(inode);
2591 ret = btrfs_update_inode(trans, root, inode);
2597 /* helper to check if there is any shared block in the path */
2598 static int check_path_shared(struct btrfs_root *root,
2599 struct btrfs_path *path)
2601 struct extent_buffer *eb;
2605 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2608 if (!path->nodes[level])
2610 eb = path->nodes[level];
2611 if (!btrfs_block_can_be_shared(root, eb))
2613 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2622 * helper to start transaction for unlink and rmdir.
2624 * unlink and rmdir are special in btrfs, they do not always free space.
2625 * so in enospc case, we should make sure they will free space before
2626 * allowing them to use the global metadata reservation.
2628 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2629 struct dentry *dentry)
2631 struct btrfs_trans_handle *trans;
2632 struct btrfs_root *root = BTRFS_I(dir)->root;
2633 struct btrfs_path *path;
2634 struct btrfs_inode_ref *ref;
2635 struct btrfs_dir_item *di;
2636 struct inode *inode = dentry->d_inode;
2641 u64 ino = btrfs_ino(inode);
2642 u64 dir_ino = btrfs_ino(dir);
2645 * 1 for the possible orphan item
2646 * 1 for the dir item
2647 * 1 for the dir index
2648 * 1 for the inode ref
2649 * 1 for the inode ref in the tree log
2650 * 2 for the dir entries in the log
2653 trans = btrfs_start_transaction(root, 8);
2654 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2657 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2658 return ERR_PTR(-ENOSPC);
2660 /* check if there is someone else holds reference */
2661 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2662 return ERR_PTR(-ENOSPC);
2664 if (atomic_read(&inode->i_count) > 2)
2665 return ERR_PTR(-ENOSPC);
2667 if (xchg(&root->fs_info->enospc_unlink, 1))
2668 return ERR_PTR(-ENOSPC);
2670 path = btrfs_alloc_path();
2672 root->fs_info->enospc_unlink = 0;
2673 return ERR_PTR(-ENOMEM);
2676 /* 1 for the orphan item */
2677 trans = btrfs_start_transaction(root, 1);
2678 if (IS_ERR(trans)) {
2679 btrfs_free_path(path);
2680 root->fs_info->enospc_unlink = 0;
2684 path->skip_locking = 1;
2685 path->search_commit_root = 1;
2687 ret = btrfs_lookup_inode(trans, root, path,
2688 &BTRFS_I(dir)->location, 0);
2694 if (check_path_shared(root, path))
2699 btrfs_release_path(path);
2701 ret = btrfs_lookup_inode(trans, root, path,
2702 &BTRFS_I(inode)->location, 0);
2708 if (check_path_shared(root, path))
2713 btrfs_release_path(path);
2715 if (ret == 0 && S_ISREG(inode->i_mode)) {
2716 ret = btrfs_lookup_file_extent(trans, root, path,
2723 if (check_path_shared(root, path))
2725 btrfs_release_path(path);
2733 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2734 dentry->d_name.name, dentry->d_name.len, 0);
2740 if (check_path_shared(root, path))
2746 btrfs_release_path(path);
2748 ref = btrfs_lookup_inode_ref(trans, root, path,
2749 dentry->d_name.name, dentry->d_name.len,
2756 if (check_path_shared(root, path))
2758 index = btrfs_inode_ref_index(path->nodes[0], ref);
2759 btrfs_release_path(path);
2762 * This is a commit root search, if we can lookup inode item and other
2763 * relative items in the commit root, it means the transaction of
2764 * dir/file creation has been committed, and the dir index item that we
2765 * delay to insert has also been inserted into the commit root. So
2766 * we needn't worry about the delayed insertion of the dir index item
2769 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2770 dentry->d_name.name, dentry->d_name.len, 0);
2775 BUG_ON(ret == -ENOENT);
2776 if (check_path_shared(root, path))
2781 btrfs_free_path(path);
2782 /* Migrate the orphan reservation over */
2784 err = btrfs_block_rsv_migrate(trans->block_rsv,
2785 &root->fs_info->global_block_rsv,
2786 trans->bytes_reserved);
2789 btrfs_end_transaction(trans, root);
2790 root->fs_info->enospc_unlink = 0;
2791 return ERR_PTR(err);
2794 trans->block_rsv = &root->fs_info->global_block_rsv;
2798 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2799 struct btrfs_root *root)
2801 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2802 btrfs_block_rsv_release(root, trans->block_rsv,
2803 trans->bytes_reserved);
2804 trans->block_rsv = &root->fs_info->trans_block_rsv;
2805 BUG_ON(!root->fs_info->enospc_unlink);
2806 root->fs_info->enospc_unlink = 0;
2808 btrfs_end_transaction_throttle(trans, root);
2811 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2813 struct btrfs_root *root = BTRFS_I(dir)->root;
2814 struct btrfs_trans_handle *trans;
2815 struct inode *inode = dentry->d_inode;
2817 unsigned long nr = 0;
2819 trans = __unlink_start_trans(dir, dentry);
2821 return PTR_ERR(trans);
2823 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2825 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2826 dentry->d_name.name, dentry->d_name.len);
2830 if (inode->i_nlink == 0) {
2831 ret = btrfs_orphan_add(trans, inode);
2837 nr = trans->blocks_used;
2838 __unlink_end_trans(trans, root);
2839 btrfs_btree_balance_dirty(root, nr);
2843 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *root,
2845 struct inode *dir, u64 objectid,
2846 const char *name, int name_len)
2848 struct btrfs_path *path;
2849 struct extent_buffer *leaf;
2850 struct btrfs_dir_item *di;
2851 struct btrfs_key key;
2854 u64 dir_ino = btrfs_ino(dir);
2856 path = btrfs_alloc_path();
2860 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2861 name, name_len, -1);
2862 BUG_ON(IS_ERR_OR_NULL(di));
2864 leaf = path->nodes[0];
2865 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2866 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2867 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2869 btrfs_release_path(path);
2871 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2872 objectid, root->root_key.objectid,
2873 dir_ino, &index, name, name_len);
2875 BUG_ON(ret != -ENOENT);
2876 di = btrfs_search_dir_index_item(root, path, dir_ino,
2878 BUG_ON(IS_ERR_OR_NULL(di));
2880 leaf = path->nodes[0];
2881 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2882 btrfs_release_path(path);
2885 btrfs_release_path(path);
2887 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2890 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2891 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2892 ret = btrfs_update_inode(trans, root, dir);
2895 btrfs_free_path(path);
2899 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2901 struct inode *inode = dentry->d_inode;
2903 struct btrfs_root *root = BTRFS_I(dir)->root;
2904 struct btrfs_trans_handle *trans;
2905 unsigned long nr = 0;
2907 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2908 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2911 trans = __unlink_start_trans(dir, dentry);
2913 return PTR_ERR(trans);
2915 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2916 err = btrfs_unlink_subvol(trans, root, dir,
2917 BTRFS_I(inode)->location.objectid,
2918 dentry->d_name.name,
2919 dentry->d_name.len);
2923 err = btrfs_orphan_add(trans, inode);
2927 /* now the directory is empty */
2928 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2929 dentry->d_name.name, dentry->d_name.len);
2931 btrfs_i_size_write(inode, 0);
2933 nr = trans->blocks_used;
2934 __unlink_end_trans(trans, root);
2935 btrfs_btree_balance_dirty(root, nr);
2941 * this can truncate away extent items, csum items and directory items.
2942 * It starts at a high offset and removes keys until it can't find
2943 * any higher than new_size
2945 * csum items that cross the new i_size are truncated to the new size
2948 * min_type is the minimum key type to truncate down to. If set to 0, this
2949 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2951 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2952 struct btrfs_root *root,
2953 struct inode *inode,
2954 u64 new_size, u32 min_type)
2956 struct btrfs_path *path;
2957 struct extent_buffer *leaf;
2958 struct btrfs_file_extent_item *fi;
2959 struct btrfs_key key;
2960 struct btrfs_key found_key;
2961 u64 extent_start = 0;
2962 u64 extent_num_bytes = 0;
2963 u64 extent_offset = 0;
2965 u64 mask = root->sectorsize - 1;
2966 u32 found_type = (u8)-1;
2969 int pending_del_nr = 0;
2970 int pending_del_slot = 0;
2971 int extent_type = -1;
2975 u64 ino = btrfs_ino(inode);
2977 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2979 path = btrfs_alloc_path();
2984 if (root->ref_cows || root == root->fs_info->tree_root)
2985 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2988 * This function is also used to drop the items in the log tree before
2989 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2990 * it is used to drop the loged items. So we shouldn't kill the delayed
2993 if (min_type == 0 && root == BTRFS_I(inode)->root)
2994 btrfs_kill_delayed_inode_items(inode);
2997 key.offset = (u64)-1;
3001 path->leave_spinning = 1;
3002 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3009 /* there are no items in the tree for us to truncate, we're
3012 if (path->slots[0] == 0)
3019 leaf = path->nodes[0];
3020 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3021 found_type = btrfs_key_type(&found_key);
3024 if (found_key.objectid != ino)
3027 if (found_type < min_type)
3030 item_end = found_key.offset;
3031 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3032 fi = btrfs_item_ptr(leaf, path->slots[0],
3033 struct btrfs_file_extent_item);
3034 extent_type = btrfs_file_extent_type(leaf, fi);
3035 encoding = btrfs_file_extent_compression(leaf, fi);
3036 encoding |= btrfs_file_extent_encryption(leaf, fi);
3037 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3039 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3041 btrfs_file_extent_num_bytes(leaf, fi);
3042 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3043 item_end += btrfs_file_extent_inline_len(leaf,
3048 if (found_type > min_type) {
3051 if (item_end < new_size)
3053 if (found_key.offset >= new_size)
3059 /* FIXME, shrink the extent if the ref count is only 1 */
3060 if (found_type != BTRFS_EXTENT_DATA_KEY)
3063 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3065 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3066 if (!del_item && !encoding) {
3067 u64 orig_num_bytes =
3068 btrfs_file_extent_num_bytes(leaf, fi);
3069 extent_num_bytes = new_size -
3070 found_key.offset + root->sectorsize - 1;
3071 extent_num_bytes = extent_num_bytes &
3072 ~((u64)root->sectorsize - 1);
3073 btrfs_set_file_extent_num_bytes(leaf, fi,
3075 num_dec = (orig_num_bytes -
3077 if (root->ref_cows && extent_start != 0)
3078 inode_sub_bytes(inode, num_dec);
3079 btrfs_mark_buffer_dirty(leaf);
3082 btrfs_file_extent_disk_num_bytes(leaf,
3084 extent_offset = found_key.offset -
3085 btrfs_file_extent_offset(leaf, fi);
3087 /* FIXME blocksize != 4096 */
3088 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3089 if (extent_start != 0) {
3092 inode_sub_bytes(inode, num_dec);
3095 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3097 * we can't truncate inline items that have had
3101 btrfs_file_extent_compression(leaf, fi) == 0 &&
3102 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3103 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3104 u32 size = new_size - found_key.offset;
3106 if (root->ref_cows) {
3107 inode_sub_bytes(inode, item_end + 1 -
3111 btrfs_file_extent_calc_inline_size(size);
3112 ret = btrfs_truncate_item(trans, root, path,
3114 } else if (root->ref_cows) {
3115 inode_sub_bytes(inode, item_end + 1 -
3121 if (!pending_del_nr) {
3122 /* no pending yet, add ourselves */
3123 pending_del_slot = path->slots[0];
3125 } else if (pending_del_nr &&
3126 path->slots[0] + 1 == pending_del_slot) {
3127 /* hop on the pending chunk */
3129 pending_del_slot = path->slots[0];
3136 if (found_extent && (root->ref_cows ||
3137 root == root->fs_info->tree_root)) {
3138 btrfs_set_path_blocking(path);
3139 ret = btrfs_free_extent(trans, root, extent_start,
3140 extent_num_bytes, 0,
3141 btrfs_header_owner(leaf),
3142 ino, extent_offset);
3146 if (found_type == BTRFS_INODE_ITEM_KEY)
3149 if (path->slots[0] == 0 ||
3150 path->slots[0] != pending_del_slot) {
3151 if (root->ref_cows &&
3152 BTRFS_I(inode)->location.objectid !=
3153 BTRFS_FREE_INO_OBJECTID) {
3157 if (pending_del_nr) {
3158 ret = btrfs_del_items(trans, root, path,
3164 btrfs_release_path(path);
3171 if (pending_del_nr) {
3172 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3176 btrfs_free_path(path);
3181 * taken from block_truncate_page, but does cow as it zeros out
3182 * any bytes left in the last page in the file.
3184 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3186 struct inode *inode = mapping->host;
3187 struct btrfs_root *root = BTRFS_I(inode)->root;
3188 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3189 struct btrfs_ordered_extent *ordered;
3190 struct extent_state *cached_state = NULL;
3192 u32 blocksize = root->sectorsize;
3193 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3194 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3196 gfp_t mask = btrfs_alloc_write_mask(mapping);
3201 if ((offset & (blocksize - 1)) == 0)
3203 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3209 page = find_or_create_page(mapping, index, mask);
3211 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3215 page_start = page_offset(page);
3216 page_end = page_start + PAGE_CACHE_SIZE - 1;
3218 if (!PageUptodate(page)) {
3219 ret = btrfs_readpage(NULL, page);
3221 if (page->mapping != mapping) {
3223 page_cache_release(page);
3226 if (!PageUptodate(page)) {
3231 wait_on_page_writeback(page);
3233 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3235 set_page_extent_mapped(page);
3237 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3239 unlock_extent_cached(io_tree, page_start, page_end,
3240 &cached_state, GFP_NOFS);
3242 page_cache_release(page);
3243 btrfs_start_ordered_extent(inode, ordered, 1);
3244 btrfs_put_ordered_extent(ordered);
3248 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3249 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3250 0, 0, &cached_state, GFP_NOFS);
3252 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3255 unlock_extent_cached(io_tree, page_start, page_end,
3256 &cached_state, GFP_NOFS);
3261 if (offset != PAGE_CACHE_SIZE) {
3263 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3264 flush_dcache_page(page);
3267 ClearPageChecked(page);
3268 set_page_dirty(page);
3269 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3274 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3276 page_cache_release(page);
3282 * This function puts in dummy file extents for the area we're creating a hole
3283 * for. So if we are truncating this file to a larger size we need to insert
3284 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3285 * the range between oldsize and size
3287 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3289 struct btrfs_trans_handle *trans;
3290 struct btrfs_root *root = BTRFS_I(inode)->root;
3291 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3292 struct extent_map *em = NULL;
3293 struct extent_state *cached_state = NULL;
3294 u64 mask = root->sectorsize - 1;
3295 u64 hole_start = (oldsize + mask) & ~mask;
3296 u64 block_end = (size + mask) & ~mask;
3302 if (size <= hole_start)
3306 struct btrfs_ordered_extent *ordered;
3307 btrfs_wait_ordered_range(inode, hole_start,
3308 block_end - hole_start);
3309 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3310 &cached_state, GFP_NOFS);
3311 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3314 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3315 &cached_state, GFP_NOFS);
3316 btrfs_put_ordered_extent(ordered);
3319 cur_offset = hole_start;
3321 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3322 block_end - cur_offset, 0);
3323 BUG_ON(IS_ERR_OR_NULL(em));
3324 last_byte = min(extent_map_end(em), block_end);
3325 last_byte = (last_byte + mask) & ~mask;
3326 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3328 hole_size = last_byte - cur_offset;
3330 trans = btrfs_start_transaction(root, 2);
3331 if (IS_ERR(trans)) {
3332 err = PTR_ERR(trans);
3336 err = btrfs_drop_extents(trans, inode, cur_offset,
3337 cur_offset + hole_size,
3340 btrfs_end_transaction(trans, root);
3344 err = btrfs_insert_file_extent(trans, root,
3345 btrfs_ino(inode), cur_offset, 0,
3346 0, hole_size, 0, hole_size,
3349 btrfs_end_transaction(trans, root);
3353 btrfs_drop_extent_cache(inode, hole_start,
3356 btrfs_end_transaction(trans, root);
3358 free_extent_map(em);
3360 cur_offset = last_byte;
3361 if (cur_offset >= block_end)
3365 free_extent_map(em);
3366 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3371 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3373 struct btrfs_root *root = BTRFS_I(inode)->root;
3374 struct btrfs_trans_handle *trans;
3375 loff_t oldsize = i_size_read(inode);
3378 if (newsize == oldsize)
3381 if (newsize > oldsize) {
3382 truncate_pagecache(inode, oldsize, newsize);
3383 ret = btrfs_cont_expand(inode, oldsize, newsize);
3387 trans = btrfs_start_transaction(root, 1);
3389 return PTR_ERR(trans);
3391 i_size_write(inode, newsize);
3392 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3393 ret = btrfs_update_inode(trans, root, inode);
3395 btrfs_end_transaction_throttle(trans, root);
3399 * We're truncating a file that used to have good data down to
3400 * zero. Make sure it gets into the ordered flush list so that
3401 * any new writes get down to disk quickly.
3404 BTRFS_I(inode)->ordered_data_close = 1;
3406 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3407 truncate_setsize(inode, newsize);
3408 ret = btrfs_truncate(inode);
3414 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3416 struct inode *inode = dentry->d_inode;
3417 struct btrfs_root *root = BTRFS_I(inode)->root;
3420 if (btrfs_root_readonly(root))
3423 err = inode_change_ok(inode, attr);
3427 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3428 err = btrfs_setsize(inode, attr->ia_size);
3433 if (attr->ia_valid) {
3434 setattr_copy(inode, attr);
3435 mark_inode_dirty(inode);
3437 if (attr->ia_valid & ATTR_MODE)
3438 err = btrfs_acl_chmod(inode);
3444 void btrfs_evict_inode(struct inode *inode)
3446 struct btrfs_trans_handle *trans;
3447 struct btrfs_root *root = BTRFS_I(inode)->root;
3448 struct btrfs_block_rsv *rsv, *global_rsv;
3449 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3453 trace_btrfs_inode_evict(inode);
3455 truncate_inode_pages(&inode->i_data, 0);
3456 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3457 btrfs_is_free_space_inode(root, inode)))
3460 if (is_bad_inode(inode)) {
3461 btrfs_orphan_del(NULL, inode);
3464 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3465 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3467 if (root->fs_info->log_root_recovering) {
3468 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3472 if (inode->i_nlink > 0) {
3473 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3477 rsv = btrfs_alloc_block_rsv(root);
3479 btrfs_orphan_del(NULL, inode);
3482 rsv->size = min_size;
3483 global_rsv = &root->fs_info->global_block_rsv;
3485 btrfs_i_size_write(inode, 0);
3488 * This is a bit simpler than btrfs_truncate since
3490 * 1) We've already reserved our space for our orphan item in the
3492 * 2) We're going to delete the inode item, so we don't need to update
3495 * So we just need to reserve some slack space in case we add bytes when
3496 * doing the truncate.
3499 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3502 * Try and steal from the global reserve since we will
3503 * likely not use this space anyway, we want to try as
3504 * hard as possible to get this to work.
3507 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3510 printk(KERN_WARNING "Could not get space for a "
3511 "delete, will truncate on mount %d\n", ret);
3512 btrfs_orphan_del(NULL, inode);
3513 btrfs_free_block_rsv(root, rsv);
3517 trans = btrfs_start_transaction(root, 0);
3518 if (IS_ERR(trans)) {
3519 btrfs_orphan_del(NULL, inode);
3520 btrfs_free_block_rsv(root, rsv);
3524 trans->block_rsv = rsv;
3526 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3530 nr = trans->blocks_used;
3531 btrfs_end_transaction(trans, root);
3533 btrfs_btree_balance_dirty(root, nr);
3536 btrfs_free_block_rsv(root, rsv);
3539 trans->block_rsv = root->orphan_block_rsv;
3540 ret = btrfs_orphan_del(trans, inode);
3544 trans->block_rsv = &root->fs_info->trans_block_rsv;
3545 if (!(root == root->fs_info->tree_root ||
3546 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3547 btrfs_return_ino(root, btrfs_ino(inode));
3549 nr = trans->blocks_used;
3550 btrfs_end_transaction(trans, root);
3551 btrfs_btree_balance_dirty(root, nr);
3553 end_writeback(inode);
3558 * this returns the key found in the dir entry in the location pointer.
3559 * If no dir entries were found, location->objectid is 0.
3561 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3562 struct btrfs_key *location)
3564 const char *name = dentry->d_name.name;
3565 int namelen = dentry->d_name.len;
3566 struct btrfs_dir_item *di;
3567 struct btrfs_path *path;
3568 struct btrfs_root *root = BTRFS_I(dir)->root;
3571 path = btrfs_alloc_path();
3575 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3580 if (IS_ERR_OR_NULL(di))
3583 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3585 btrfs_free_path(path);
3588 location->objectid = 0;
3593 * when we hit a tree root in a directory, the btrfs part of the inode
3594 * needs to be changed to reflect the root directory of the tree root. This
3595 * is kind of like crossing a mount point.
3597 static int fixup_tree_root_location(struct btrfs_root *root,
3599 struct dentry *dentry,
3600 struct btrfs_key *location,
3601 struct btrfs_root **sub_root)
3603 struct btrfs_path *path;
3604 struct btrfs_root *new_root;
3605 struct btrfs_root_ref *ref;
3606 struct extent_buffer *leaf;
3610 path = btrfs_alloc_path();
3617 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3618 BTRFS_I(dir)->root->root_key.objectid,
3619 location->objectid);
3626 leaf = path->nodes[0];
3627 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3628 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3629 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3632 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3633 (unsigned long)(ref + 1),
3634 dentry->d_name.len);
3638 btrfs_release_path(path);
3640 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3641 if (IS_ERR(new_root)) {
3642 err = PTR_ERR(new_root);
3646 if (btrfs_root_refs(&new_root->root_item) == 0) {
3651 *sub_root = new_root;
3652 location->objectid = btrfs_root_dirid(&new_root->root_item);
3653 location->type = BTRFS_INODE_ITEM_KEY;
3654 location->offset = 0;
3657 btrfs_free_path(path);
3661 static void inode_tree_add(struct inode *inode)
3663 struct btrfs_root *root = BTRFS_I(inode)->root;
3664 struct btrfs_inode *entry;
3666 struct rb_node *parent;
3667 u64 ino = btrfs_ino(inode);
3669 p = &root->inode_tree.rb_node;
3672 if (inode_unhashed(inode))
3675 spin_lock(&root->inode_lock);
3678 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3680 if (ino < btrfs_ino(&entry->vfs_inode))
3681 p = &parent->rb_left;
3682 else if (ino > btrfs_ino(&entry->vfs_inode))
3683 p = &parent->rb_right;
3685 WARN_ON(!(entry->vfs_inode.i_state &
3686 (I_WILL_FREE | I_FREEING)));
3687 rb_erase(parent, &root->inode_tree);
3688 RB_CLEAR_NODE(parent);
3689 spin_unlock(&root->inode_lock);
3693 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3694 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3695 spin_unlock(&root->inode_lock);
3698 static void inode_tree_del(struct inode *inode)
3700 struct btrfs_root *root = BTRFS_I(inode)->root;
3703 spin_lock(&root->inode_lock);
3704 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3705 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3706 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3707 empty = RB_EMPTY_ROOT(&root->inode_tree);
3709 spin_unlock(&root->inode_lock);
3712 * Free space cache has inodes in the tree root, but the tree root has a
3713 * root_refs of 0, so this could end up dropping the tree root as a
3714 * snapshot, so we need the extra !root->fs_info->tree_root check to
3715 * make sure we don't drop it.
3717 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3718 root != root->fs_info->tree_root) {
3719 synchronize_srcu(&root->fs_info->subvol_srcu);
3720 spin_lock(&root->inode_lock);
3721 empty = RB_EMPTY_ROOT(&root->inode_tree);
3722 spin_unlock(&root->inode_lock);
3724 btrfs_add_dead_root(root);
3728 int btrfs_invalidate_inodes(struct btrfs_root *root)
3730 struct rb_node *node;
3731 struct rb_node *prev;
3732 struct btrfs_inode *entry;
3733 struct inode *inode;
3736 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3738 spin_lock(&root->inode_lock);
3740 node = root->inode_tree.rb_node;
3744 entry = rb_entry(node, struct btrfs_inode, rb_node);
3746 if (objectid < btrfs_ino(&entry->vfs_inode))
3747 node = node->rb_left;
3748 else if (objectid > btrfs_ino(&entry->vfs_inode))
3749 node = node->rb_right;
3755 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3756 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3760 prev = rb_next(prev);
3764 entry = rb_entry(node, struct btrfs_inode, rb_node);
3765 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3766 inode = igrab(&entry->vfs_inode);
3768 spin_unlock(&root->inode_lock);
3769 if (atomic_read(&inode->i_count) > 1)
3770 d_prune_aliases(inode);
3772 * btrfs_drop_inode will have it removed from
3773 * the inode cache when its usage count
3778 spin_lock(&root->inode_lock);
3782 if (cond_resched_lock(&root->inode_lock))
3785 node = rb_next(node);
3787 spin_unlock(&root->inode_lock);
3791 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3793 struct btrfs_iget_args *args = p;
3794 inode->i_ino = args->ino;
3795 BTRFS_I(inode)->root = args->root;
3796 btrfs_set_inode_space_info(args->root, inode);
3800 static int btrfs_find_actor(struct inode *inode, void *opaque)
3802 struct btrfs_iget_args *args = opaque;
3803 return args->ino == btrfs_ino(inode) &&
3804 args->root == BTRFS_I(inode)->root;
3807 static struct inode *btrfs_iget_locked(struct super_block *s,
3809 struct btrfs_root *root)
3811 struct inode *inode;
3812 struct btrfs_iget_args args;
3813 args.ino = objectid;
3816 inode = iget5_locked(s, objectid, btrfs_find_actor,
3817 btrfs_init_locked_inode,
3822 /* Get an inode object given its location and corresponding root.
3823 * Returns in *is_new if the inode was read from disk
3825 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3826 struct btrfs_root *root, int *new)
3828 struct inode *inode;
3830 inode = btrfs_iget_locked(s, location->objectid, root);
3832 return ERR_PTR(-ENOMEM);
3834 if (inode->i_state & I_NEW) {
3835 BTRFS_I(inode)->root = root;
3836 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3837 btrfs_read_locked_inode(inode);
3838 if (!is_bad_inode(inode)) {
3839 inode_tree_add(inode);
3840 unlock_new_inode(inode);
3844 unlock_new_inode(inode);
3846 inode = ERR_PTR(-ESTALE);
3853 static struct inode *new_simple_dir(struct super_block *s,
3854 struct btrfs_key *key,
3855 struct btrfs_root *root)
3857 struct inode *inode = new_inode(s);
3860 return ERR_PTR(-ENOMEM);
3862 BTRFS_I(inode)->root = root;
3863 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3864 BTRFS_I(inode)->dummy_inode = 1;
3866 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3867 inode->i_op = &simple_dir_inode_operations;
3868 inode->i_fop = &simple_dir_operations;
3869 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3870 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3875 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3877 struct inode *inode;
3878 struct btrfs_root *root = BTRFS_I(dir)->root;
3879 struct btrfs_root *sub_root = root;
3880 struct btrfs_key location;
3884 if (dentry->d_name.len > BTRFS_NAME_LEN)
3885 return ERR_PTR(-ENAMETOOLONG);
3887 if (unlikely(d_need_lookup(dentry))) {
3888 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3889 kfree(dentry->d_fsdata);
3890 dentry->d_fsdata = NULL;
3891 /* This thing is hashed, drop it for now */
3894 ret = btrfs_inode_by_name(dir, dentry, &location);
3898 return ERR_PTR(ret);
3900 if (location.objectid == 0)
3903 if (location.type == BTRFS_INODE_ITEM_KEY) {
3904 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3908 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3910 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3911 ret = fixup_tree_root_location(root, dir, dentry,
3912 &location, &sub_root);
3915 inode = ERR_PTR(ret);
3917 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3919 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3921 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3923 if (!IS_ERR(inode) && root != sub_root) {
3924 down_read(&root->fs_info->cleanup_work_sem);
3925 if (!(inode->i_sb->s_flags & MS_RDONLY))
3926 ret = btrfs_orphan_cleanup(sub_root);
3927 up_read(&root->fs_info->cleanup_work_sem);
3929 inode = ERR_PTR(ret);
3935 static int btrfs_dentry_delete(const struct dentry *dentry)
3937 struct btrfs_root *root;
3939 if (!dentry->d_inode && !IS_ROOT(dentry))
3940 dentry = dentry->d_parent;
3942 if (dentry->d_inode) {
3943 root = BTRFS_I(dentry->d_inode)->root;
3944 if (btrfs_root_refs(&root->root_item) == 0)
3950 static void btrfs_dentry_release(struct dentry *dentry)
3952 if (dentry->d_fsdata)
3953 kfree(dentry->d_fsdata);
3956 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3957 struct nameidata *nd)
3961 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3962 if (unlikely(d_need_lookup(dentry))) {
3963 spin_lock(&dentry->d_lock);
3964 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
3965 spin_unlock(&dentry->d_lock);
3970 unsigned char btrfs_filetype_table[] = {
3971 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3974 static int btrfs_real_readdir(struct file *filp, void *dirent,
3977 struct inode *inode = filp->f_dentry->d_inode;
3978 struct btrfs_root *root = BTRFS_I(inode)->root;
3979 struct btrfs_item *item;
3980 struct btrfs_dir_item *di;
3981 struct btrfs_key key;
3982 struct btrfs_key found_key;
3983 struct btrfs_path *path;
3984 struct list_head ins_list;
3985 struct list_head del_list;
3988 struct extent_buffer *leaf;
3990 unsigned char d_type;
3995 int key_type = BTRFS_DIR_INDEX_KEY;
3999 int is_curr = 0; /* filp->f_pos points to the current index? */
4001 /* FIXME, use a real flag for deciding about the key type */
4002 if (root->fs_info->tree_root == root)
4003 key_type = BTRFS_DIR_ITEM_KEY;
4005 /* special case for "." */
4006 if (filp->f_pos == 0) {
4007 over = filldir(dirent, ".", 1,
4008 filp->f_pos, btrfs_ino(inode), DT_DIR);
4013 /* special case for .., just use the back ref */
4014 if (filp->f_pos == 1) {
4015 u64 pino = parent_ino(filp->f_path.dentry);
4016 over = filldir(dirent, "..", 2,
4017 filp->f_pos, pino, DT_DIR);
4022 path = btrfs_alloc_path();
4028 if (key_type == BTRFS_DIR_INDEX_KEY) {
4029 INIT_LIST_HEAD(&ins_list);
4030 INIT_LIST_HEAD(&del_list);
4031 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4034 btrfs_set_key_type(&key, key_type);
4035 key.offset = filp->f_pos;
4036 key.objectid = btrfs_ino(inode);
4038 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4043 leaf = path->nodes[0];
4044 slot = path->slots[0];
4045 if (slot >= btrfs_header_nritems(leaf)) {
4046 ret = btrfs_next_leaf(root, path);
4054 item = btrfs_item_nr(leaf, slot);
4055 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4057 if (found_key.objectid != key.objectid)
4059 if (btrfs_key_type(&found_key) != key_type)
4061 if (found_key.offset < filp->f_pos)
4063 if (key_type == BTRFS_DIR_INDEX_KEY &&
4064 btrfs_should_delete_dir_index(&del_list,
4068 filp->f_pos = found_key.offset;
4071 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4073 di_total = btrfs_item_size(leaf, item);
4075 while (di_cur < di_total) {
4076 struct btrfs_key location;
4079 if (verify_dir_item(root, leaf, di))
4082 name_len = btrfs_dir_name_len(leaf, di);
4083 if (name_len <= sizeof(tmp_name)) {
4084 name_ptr = tmp_name;
4086 name_ptr = kmalloc(name_len, GFP_NOFS);
4092 read_extent_buffer(leaf, name_ptr,
4093 (unsigned long)(di + 1), name_len);
4095 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4096 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4100 q.hash = full_name_hash(q.name, q.len);
4101 tmp = d_lookup(filp->f_dentry, &q);
4103 struct btrfs_key *newkey;
4105 newkey = kzalloc(sizeof(struct btrfs_key),
4109 tmp = d_alloc(filp->f_dentry, &q);
4115 memcpy(newkey, &location,
4116 sizeof(struct btrfs_key));
4117 tmp->d_fsdata = newkey;
4118 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4125 /* is this a reference to our own snapshot? If so
4128 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4129 location.objectid == root->root_key.objectid) {
4133 over = filldir(dirent, name_ptr, name_len,
4134 found_key.offset, location.objectid,
4138 if (name_ptr != tmp_name)
4143 di_len = btrfs_dir_name_len(leaf, di) +
4144 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4146 di = (struct btrfs_dir_item *)((char *)di + di_len);
4152 if (key_type == BTRFS_DIR_INDEX_KEY) {
4155 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4161 /* Reached end of directory/root. Bump pos past the last item. */
4162 if (key_type == BTRFS_DIR_INDEX_KEY)
4164 * 32-bit glibc will use getdents64, but then strtol -
4165 * so the last number we can serve is this.
4167 filp->f_pos = 0x7fffffff;
4173 if (key_type == BTRFS_DIR_INDEX_KEY)
4174 btrfs_put_delayed_items(&ins_list, &del_list);
4175 btrfs_free_path(path);
4179 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4181 struct btrfs_root *root = BTRFS_I(inode)->root;
4182 struct btrfs_trans_handle *trans;
4184 bool nolock = false;
4186 if (BTRFS_I(inode)->dummy_inode)
4189 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4192 if (wbc->sync_mode == WB_SYNC_ALL) {
4194 trans = btrfs_join_transaction_nolock(root);
4196 trans = btrfs_join_transaction(root);
4198 return PTR_ERR(trans);
4200 ret = btrfs_end_transaction_nolock(trans, root);
4202 ret = btrfs_commit_transaction(trans, root);
4208 * This is somewhat expensive, updating the tree every time the
4209 * inode changes. But, it is most likely to find the inode in cache.
4210 * FIXME, needs more benchmarking...there are no reasons other than performance
4211 * to keep or drop this code.
4213 void btrfs_dirty_inode(struct inode *inode, int flags)
4215 struct btrfs_root *root = BTRFS_I(inode)->root;
4216 struct btrfs_trans_handle *trans;
4219 if (BTRFS_I(inode)->dummy_inode)
4222 trans = btrfs_join_transaction(root);
4223 BUG_ON(IS_ERR(trans));
4225 ret = btrfs_update_inode(trans, root, inode);
4226 if (ret && ret == -ENOSPC) {
4227 /* whoops, lets try again with the full transaction */
4228 btrfs_end_transaction(trans, root);
4229 trans = btrfs_start_transaction(root, 1);
4230 if (IS_ERR(trans)) {
4231 printk_ratelimited(KERN_ERR "btrfs: fail to "
4232 "dirty inode %llu error %ld\n",
4233 (unsigned long long)btrfs_ino(inode),
4238 ret = btrfs_update_inode(trans, root, inode);
4240 printk_ratelimited(KERN_ERR "btrfs: fail to "
4241 "dirty inode %llu error %d\n",
4242 (unsigned long long)btrfs_ino(inode),
4246 btrfs_end_transaction(trans, root);
4247 if (BTRFS_I(inode)->delayed_node)
4248 btrfs_balance_delayed_items(root);
4252 * find the highest existing sequence number in a directory
4253 * and then set the in-memory index_cnt variable to reflect
4254 * free sequence numbers
4256 static int btrfs_set_inode_index_count(struct inode *inode)
4258 struct btrfs_root *root = BTRFS_I(inode)->root;
4259 struct btrfs_key key, found_key;
4260 struct btrfs_path *path;
4261 struct extent_buffer *leaf;
4264 key.objectid = btrfs_ino(inode);
4265 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4266 key.offset = (u64)-1;
4268 path = btrfs_alloc_path();
4272 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4275 /* FIXME: we should be able to handle this */
4281 * MAGIC NUMBER EXPLANATION:
4282 * since we search a directory based on f_pos we have to start at 2
4283 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4284 * else has to start at 2
4286 if (path->slots[0] == 0) {
4287 BTRFS_I(inode)->index_cnt = 2;
4293 leaf = path->nodes[0];
4294 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4296 if (found_key.objectid != btrfs_ino(inode) ||
4297 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4298 BTRFS_I(inode)->index_cnt = 2;
4302 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4304 btrfs_free_path(path);
4309 * helper to find a free sequence number in a given directory. This current
4310 * code is very simple, later versions will do smarter things in the btree
4312 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4316 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4317 ret = btrfs_inode_delayed_dir_index_count(dir);
4319 ret = btrfs_set_inode_index_count(dir);
4325 *index = BTRFS_I(dir)->index_cnt;
4326 BTRFS_I(dir)->index_cnt++;
4331 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4332 struct btrfs_root *root,
4334 const char *name, int name_len,
4335 u64 ref_objectid, u64 objectid, int mode,
4338 struct inode *inode;
4339 struct btrfs_inode_item *inode_item;
4340 struct btrfs_key *location;
4341 struct btrfs_path *path;
4342 struct btrfs_inode_ref *ref;
4343 struct btrfs_key key[2];
4349 path = btrfs_alloc_path();
4351 return ERR_PTR(-ENOMEM);
4353 inode = new_inode(root->fs_info->sb);
4355 btrfs_free_path(path);
4356 return ERR_PTR(-ENOMEM);
4360 * we have to initialize this early, so we can reclaim the inode
4361 * number if we fail afterwards in this function.
4363 inode->i_ino = objectid;
4366 trace_btrfs_inode_request(dir);
4368 ret = btrfs_set_inode_index(dir, index);
4370 btrfs_free_path(path);
4372 return ERR_PTR(ret);
4376 * index_cnt is ignored for everything but a dir,
4377 * btrfs_get_inode_index_count has an explanation for the magic
4380 BTRFS_I(inode)->index_cnt = 2;
4381 BTRFS_I(inode)->root = root;
4382 BTRFS_I(inode)->generation = trans->transid;
4383 inode->i_generation = BTRFS_I(inode)->generation;
4384 btrfs_set_inode_space_info(root, inode);
4391 key[0].objectid = objectid;
4392 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4395 key[1].objectid = objectid;
4396 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4397 key[1].offset = ref_objectid;
4399 sizes[0] = sizeof(struct btrfs_inode_item);
4400 sizes[1] = name_len + sizeof(*ref);
4402 path->leave_spinning = 1;
4403 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4407 inode_init_owner(inode, dir, mode);
4408 inode_set_bytes(inode, 0);
4409 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4410 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4411 struct btrfs_inode_item);
4412 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4414 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4415 struct btrfs_inode_ref);
4416 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4417 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4418 ptr = (unsigned long)(ref + 1);
4419 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4421 btrfs_mark_buffer_dirty(path->nodes[0]);
4422 btrfs_free_path(path);
4424 location = &BTRFS_I(inode)->location;
4425 location->objectid = objectid;
4426 location->offset = 0;
4427 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4429 btrfs_inherit_iflags(inode, dir);
4431 if (S_ISREG(mode)) {
4432 if (btrfs_test_opt(root, NODATASUM))
4433 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4434 if (btrfs_test_opt(root, NODATACOW) ||
4435 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4436 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4439 insert_inode_hash(inode);
4440 inode_tree_add(inode);
4442 trace_btrfs_inode_new(inode);
4443 btrfs_set_inode_last_trans(trans, inode);
4448 BTRFS_I(dir)->index_cnt--;
4449 btrfs_free_path(path);
4451 return ERR_PTR(ret);
4454 static inline u8 btrfs_inode_type(struct inode *inode)
4456 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4460 * utility function to add 'inode' into 'parent_inode' with
4461 * a give name and a given sequence number.
4462 * if 'add_backref' is true, also insert a backref from the
4463 * inode to the parent directory.
4465 int btrfs_add_link(struct btrfs_trans_handle *trans,
4466 struct inode *parent_inode, struct inode *inode,
4467 const char *name, int name_len, int add_backref, u64 index)
4470 struct btrfs_key key;
4471 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4472 u64 ino = btrfs_ino(inode);
4473 u64 parent_ino = btrfs_ino(parent_inode);
4475 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4476 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4479 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4483 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4484 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4485 key.objectid, root->root_key.objectid,
4486 parent_ino, index, name, name_len);
4487 } else if (add_backref) {
4488 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4493 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4495 btrfs_inode_type(inode), index);
4498 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4500 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4501 ret = btrfs_update_inode(trans, root, parent_inode);
4506 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4507 struct inode *dir, struct dentry *dentry,
4508 struct inode *inode, int backref, u64 index)
4510 int err = btrfs_add_link(trans, dir, inode,
4511 dentry->d_name.name, dentry->d_name.len,
4514 d_instantiate(dentry, inode);
4522 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4523 int mode, dev_t rdev)
4525 struct btrfs_trans_handle *trans;
4526 struct btrfs_root *root = BTRFS_I(dir)->root;
4527 struct inode *inode = NULL;
4531 unsigned long nr = 0;
4534 if (!new_valid_dev(rdev))
4538 * 2 for inode item and ref
4540 * 1 for xattr if selinux is on
4542 trans = btrfs_start_transaction(root, 5);
4544 return PTR_ERR(trans);
4546 err = btrfs_find_free_ino(root, &objectid);
4550 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4551 dentry->d_name.len, btrfs_ino(dir), objectid,
4553 if (IS_ERR(inode)) {
4554 err = PTR_ERR(inode);
4558 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4564 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4568 inode->i_op = &btrfs_special_inode_operations;
4569 init_special_inode(inode, inode->i_mode, rdev);
4570 btrfs_update_inode(trans, root, inode);
4573 nr = trans->blocks_used;
4574 btrfs_end_transaction_throttle(trans, root);
4575 btrfs_btree_balance_dirty(root, nr);
4577 inode_dec_link_count(inode);
4583 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4584 int mode, struct nameidata *nd)
4586 struct btrfs_trans_handle *trans;
4587 struct btrfs_root *root = BTRFS_I(dir)->root;
4588 struct inode *inode = NULL;
4591 unsigned long nr = 0;
4596 * 2 for inode item and ref
4598 * 1 for xattr if selinux is on
4600 trans = btrfs_start_transaction(root, 5);
4602 return PTR_ERR(trans);
4604 err = btrfs_find_free_ino(root, &objectid);
4608 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4609 dentry->d_name.len, btrfs_ino(dir), objectid,
4611 if (IS_ERR(inode)) {
4612 err = PTR_ERR(inode);
4616 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4622 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4626 inode->i_mapping->a_ops = &btrfs_aops;
4627 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4628 inode->i_fop = &btrfs_file_operations;
4629 inode->i_op = &btrfs_file_inode_operations;
4630 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4633 nr = trans->blocks_used;
4634 btrfs_end_transaction_throttle(trans, root);
4636 inode_dec_link_count(inode);
4639 btrfs_btree_balance_dirty(root, nr);
4643 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4644 struct dentry *dentry)
4646 struct btrfs_trans_handle *trans;
4647 struct btrfs_root *root = BTRFS_I(dir)->root;
4648 struct inode *inode = old_dentry->d_inode;
4650 unsigned long nr = 0;
4654 /* do not allow sys_link's with other subvols of the same device */
4655 if (root->objectid != BTRFS_I(inode)->root->objectid)
4658 if (inode->i_nlink == ~0U)
4661 err = btrfs_set_inode_index(dir, &index);
4666 * 2 items for inode and inode ref
4667 * 2 items for dir items
4668 * 1 item for parent inode
4670 trans = btrfs_start_transaction(root, 5);
4671 if (IS_ERR(trans)) {
4672 err = PTR_ERR(trans);
4676 btrfs_inc_nlink(inode);
4677 inode->i_ctime = CURRENT_TIME;
4680 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4685 struct dentry *parent = dentry->d_parent;
4686 err = btrfs_update_inode(trans, root, inode);
4688 btrfs_log_new_name(trans, inode, NULL, parent);
4691 nr = trans->blocks_used;
4692 btrfs_end_transaction_throttle(trans, root);
4695 inode_dec_link_count(inode);
4698 btrfs_btree_balance_dirty(root, nr);
4702 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4704 struct inode *inode = NULL;
4705 struct btrfs_trans_handle *trans;
4706 struct btrfs_root *root = BTRFS_I(dir)->root;
4708 int drop_on_err = 0;
4711 unsigned long nr = 1;
4714 * 2 items for inode and ref
4715 * 2 items for dir items
4716 * 1 for xattr if selinux is on
4718 trans = btrfs_start_transaction(root, 5);
4720 return PTR_ERR(trans);
4722 err = btrfs_find_free_ino(root, &objectid);
4726 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4727 dentry->d_name.len, btrfs_ino(dir), objectid,
4728 S_IFDIR | mode, &index);
4729 if (IS_ERR(inode)) {
4730 err = PTR_ERR(inode);
4736 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4740 inode->i_op = &btrfs_dir_inode_operations;
4741 inode->i_fop = &btrfs_dir_file_operations;
4743 btrfs_i_size_write(inode, 0);
4744 err = btrfs_update_inode(trans, root, inode);
4748 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4749 dentry->d_name.len, 0, index);
4753 d_instantiate(dentry, inode);
4757 nr = trans->blocks_used;
4758 btrfs_end_transaction_throttle(trans, root);
4761 btrfs_btree_balance_dirty(root, nr);
4765 /* helper for btfs_get_extent. Given an existing extent in the tree,
4766 * and an extent that you want to insert, deal with overlap and insert
4767 * the new extent into the tree.
4769 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4770 struct extent_map *existing,
4771 struct extent_map *em,
4772 u64 map_start, u64 map_len)
4776 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4777 start_diff = map_start - em->start;
4778 em->start = map_start;
4780 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4781 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4782 em->block_start += start_diff;
4783 em->block_len -= start_diff;
4785 return add_extent_mapping(em_tree, em);
4788 static noinline int uncompress_inline(struct btrfs_path *path,
4789 struct inode *inode, struct page *page,
4790 size_t pg_offset, u64 extent_offset,
4791 struct btrfs_file_extent_item *item)
4794 struct extent_buffer *leaf = path->nodes[0];
4797 unsigned long inline_size;
4801 WARN_ON(pg_offset != 0);
4802 compress_type = btrfs_file_extent_compression(leaf, item);
4803 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4804 inline_size = btrfs_file_extent_inline_item_len(leaf,
4805 btrfs_item_nr(leaf, path->slots[0]));
4806 tmp = kmalloc(inline_size, GFP_NOFS);
4809 ptr = btrfs_file_extent_inline_start(item);
4811 read_extent_buffer(leaf, tmp, ptr, inline_size);
4813 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4814 ret = btrfs_decompress(compress_type, tmp, page,
4815 extent_offset, inline_size, max_size);
4817 char *kaddr = kmap_atomic(page, KM_USER0);
4818 unsigned long copy_size = min_t(u64,
4819 PAGE_CACHE_SIZE - pg_offset,
4820 max_size - extent_offset);
4821 memset(kaddr + pg_offset, 0, copy_size);
4822 kunmap_atomic(kaddr, KM_USER0);
4829 * a bit scary, this does extent mapping from logical file offset to the disk.
4830 * the ugly parts come from merging extents from the disk with the in-ram
4831 * representation. This gets more complex because of the data=ordered code,
4832 * where the in-ram extents might be locked pending data=ordered completion.
4834 * This also copies inline extents directly into the page.
4837 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4838 size_t pg_offset, u64 start, u64 len,
4844 u64 extent_start = 0;
4846 u64 objectid = btrfs_ino(inode);
4848 struct btrfs_path *path = NULL;
4849 struct btrfs_root *root = BTRFS_I(inode)->root;
4850 struct btrfs_file_extent_item *item;
4851 struct extent_buffer *leaf;
4852 struct btrfs_key found_key;
4853 struct extent_map *em = NULL;
4854 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4855 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4856 struct btrfs_trans_handle *trans = NULL;
4860 read_lock(&em_tree->lock);
4861 em = lookup_extent_mapping(em_tree, start, len);
4863 em->bdev = root->fs_info->fs_devices->latest_bdev;
4864 read_unlock(&em_tree->lock);
4867 if (em->start > start || em->start + em->len <= start)
4868 free_extent_map(em);
4869 else if (em->block_start == EXTENT_MAP_INLINE && page)
4870 free_extent_map(em);
4874 em = alloc_extent_map();
4879 em->bdev = root->fs_info->fs_devices->latest_bdev;
4880 em->start = EXTENT_MAP_HOLE;
4881 em->orig_start = EXTENT_MAP_HOLE;
4883 em->block_len = (u64)-1;
4886 path = btrfs_alloc_path();
4892 * Chances are we'll be called again, so go ahead and do
4898 ret = btrfs_lookup_file_extent(trans, root, path,
4899 objectid, start, trans != NULL);
4906 if (path->slots[0] == 0)
4911 leaf = path->nodes[0];
4912 item = btrfs_item_ptr(leaf, path->slots[0],
4913 struct btrfs_file_extent_item);
4914 /* are we inside the extent that was found? */
4915 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4916 found_type = btrfs_key_type(&found_key);
4917 if (found_key.objectid != objectid ||
4918 found_type != BTRFS_EXTENT_DATA_KEY) {
4922 found_type = btrfs_file_extent_type(leaf, item);
4923 extent_start = found_key.offset;
4924 compress_type = btrfs_file_extent_compression(leaf, item);
4925 if (found_type == BTRFS_FILE_EXTENT_REG ||
4926 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4927 extent_end = extent_start +
4928 btrfs_file_extent_num_bytes(leaf, item);
4929 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4931 size = btrfs_file_extent_inline_len(leaf, item);
4932 extent_end = (extent_start + size + root->sectorsize - 1) &
4933 ~((u64)root->sectorsize - 1);
4936 if (start >= extent_end) {
4938 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4939 ret = btrfs_next_leaf(root, path);
4946 leaf = path->nodes[0];
4948 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4949 if (found_key.objectid != objectid ||
4950 found_key.type != BTRFS_EXTENT_DATA_KEY)
4952 if (start + len <= found_key.offset)
4955 em->len = found_key.offset - start;
4959 if (found_type == BTRFS_FILE_EXTENT_REG ||
4960 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4961 em->start = extent_start;
4962 em->len = extent_end - extent_start;
4963 em->orig_start = extent_start -
4964 btrfs_file_extent_offset(leaf, item);
4965 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4967 em->block_start = EXTENT_MAP_HOLE;
4970 if (compress_type != BTRFS_COMPRESS_NONE) {
4971 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4972 em->compress_type = compress_type;
4973 em->block_start = bytenr;
4974 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4977 bytenr += btrfs_file_extent_offset(leaf, item);
4978 em->block_start = bytenr;
4979 em->block_len = em->len;
4980 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4981 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4984 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4988 size_t extent_offset;
4991 em->block_start = EXTENT_MAP_INLINE;
4992 if (!page || create) {
4993 em->start = extent_start;
4994 em->len = extent_end - extent_start;
4998 size = btrfs_file_extent_inline_len(leaf, item);
4999 extent_offset = page_offset(page) + pg_offset - extent_start;
5000 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5001 size - extent_offset);
5002 em->start = extent_start + extent_offset;
5003 em->len = (copy_size + root->sectorsize - 1) &
5004 ~((u64)root->sectorsize - 1);
5005 em->orig_start = EXTENT_MAP_INLINE;
5006 if (compress_type) {
5007 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5008 em->compress_type = compress_type;
5010 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5011 if (create == 0 && !PageUptodate(page)) {
5012 if (btrfs_file_extent_compression(leaf, item) !=
5013 BTRFS_COMPRESS_NONE) {
5014 ret = uncompress_inline(path, inode, page,
5016 extent_offset, item);
5020 read_extent_buffer(leaf, map + pg_offset, ptr,
5022 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5023 memset(map + pg_offset + copy_size, 0,
5024 PAGE_CACHE_SIZE - pg_offset -
5029 flush_dcache_page(page);
5030 } else if (create && PageUptodate(page)) {
5034 free_extent_map(em);
5037 btrfs_release_path(path);
5038 trans = btrfs_join_transaction(root);
5041 return ERR_CAST(trans);
5045 write_extent_buffer(leaf, map + pg_offset, ptr,
5048 btrfs_mark_buffer_dirty(leaf);
5050 set_extent_uptodate(io_tree, em->start,
5051 extent_map_end(em) - 1, NULL, GFP_NOFS);
5054 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5061 em->block_start = EXTENT_MAP_HOLE;
5062 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5064 btrfs_release_path(path);
5065 if (em->start > start || extent_map_end(em) <= start) {
5066 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5067 "[%llu %llu]\n", (unsigned long long)em->start,
5068 (unsigned long long)em->len,
5069 (unsigned long long)start,
5070 (unsigned long long)len);
5076 write_lock(&em_tree->lock);
5077 ret = add_extent_mapping(em_tree, em);
5078 /* it is possible that someone inserted the extent into the tree
5079 * while we had the lock dropped. It is also possible that
5080 * an overlapping map exists in the tree
5082 if (ret == -EEXIST) {
5083 struct extent_map *existing;
5087 existing = lookup_extent_mapping(em_tree, start, len);
5088 if (existing && (existing->start > start ||
5089 existing->start + existing->len <= start)) {
5090 free_extent_map(existing);
5094 existing = lookup_extent_mapping(em_tree, em->start,
5097 err = merge_extent_mapping(em_tree, existing,
5100 free_extent_map(existing);
5102 free_extent_map(em);
5107 free_extent_map(em);
5111 free_extent_map(em);
5116 write_unlock(&em_tree->lock);
5119 trace_btrfs_get_extent(root, em);
5122 btrfs_free_path(path);
5124 ret = btrfs_end_transaction(trans, root);
5129 free_extent_map(em);
5130 return ERR_PTR(err);
5135 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5136 size_t pg_offset, u64 start, u64 len,
5139 struct extent_map *em;
5140 struct extent_map *hole_em = NULL;
5141 u64 range_start = start;
5147 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5152 * if our em maps to a hole, there might
5153 * actually be delalloc bytes behind it
5155 if (em->block_start != EXTENT_MAP_HOLE)
5161 /* check to see if we've wrapped (len == -1 or similar) */
5170 /* ok, we didn't find anything, lets look for delalloc */
5171 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5172 end, len, EXTENT_DELALLOC, 1);
5173 found_end = range_start + found;
5174 if (found_end < range_start)
5175 found_end = (u64)-1;
5178 * we didn't find anything useful, return
5179 * the original results from get_extent()
5181 if (range_start > end || found_end <= start) {
5187 /* adjust the range_start to make sure it doesn't
5188 * go backwards from the start they passed in
5190 range_start = max(start,range_start);
5191 found = found_end - range_start;
5194 u64 hole_start = start;
5197 em = alloc_extent_map();
5203 * when btrfs_get_extent can't find anything it
5204 * returns one huge hole
5206 * make sure what it found really fits our range, and
5207 * adjust to make sure it is based on the start from
5211 u64 calc_end = extent_map_end(hole_em);
5213 if (calc_end <= start || (hole_em->start > end)) {
5214 free_extent_map(hole_em);
5217 hole_start = max(hole_em->start, start);
5218 hole_len = calc_end - hole_start;
5222 if (hole_em && range_start > hole_start) {
5223 /* our hole starts before our delalloc, so we
5224 * have to return just the parts of the hole
5225 * that go until the delalloc starts
5227 em->len = min(hole_len,
5228 range_start - hole_start);
5229 em->start = hole_start;
5230 em->orig_start = hole_start;
5232 * don't adjust block start at all,
5233 * it is fixed at EXTENT_MAP_HOLE
5235 em->block_start = hole_em->block_start;
5236 em->block_len = hole_len;
5238 em->start = range_start;
5240 em->orig_start = range_start;
5241 em->block_start = EXTENT_MAP_DELALLOC;
5242 em->block_len = found;
5244 } else if (hole_em) {
5249 free_extent_map(hole_em);
5251 free_extent_map(em);
5252 return ERR_PTR(err);
5257 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5258 struct extent_map *em,
5261 struct btrfs_root *root = BTRFS_I(inode)->root;
5262 struct btrfs_trans_handle *trans;
5263 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5264 struct btrfs_key ins;
5267 bool insert = false;
5270 * Ok if the extent map we looked up is a hole and is for the exact
5271 * range we want, there is no reason to allocate a new one, however if
5272 * it is not right then we need to free this one and drop the cache for
5275 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5277 free_extent_map(em);
5280 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5283 trans = btrfs_join_transaction(root);
5285 return ERR_CAST(trans);
5287 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5288 btrfs_add_inode_defrag(trans, inode);
5290 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5292 alloc_hint = get_extent_allocation_hint(inode, start, len);
5293 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5294 alloc_hint, (u64)-1, &ins, 1);
5301 em = alloc_extent_map();
5303 em = ERR_PTR(-ENOMEM);
5309 em->orig_start = em->start;
5310 em->len = ins.offset;
5312 em->block_start = ins.objectid;
5313 em->block_len = ins.offset;
5314 em->bdev = root->fs_info->fs_devices->latest_bdev;
5317 * We need to do this because if we're using the original em we searched
5318 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5321 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5324 write_lock(&em_tree->lock);
5325 ret = add_extent_mapping(em_tree, em);
5326 write_unlock(&em_tree->lock);
5329 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5332 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5333 ins.offset, ins.offset, 0);
5335 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5339 btrfs_end_transaction(trans, root);
5344 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5345 * block must be cow'd
5347 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5348 struct inode *inode, u64 offset, u64 len)
5350 struct btrfs_path *path;
5352 struct extent_buffer *leaf;
5353 struct btrfs_root *root = BTRFS_I(inode)->root;
5354 struct btrfs_file_extent_item *fi;
5355 struct btrfs_key key;
5363 path = btrfs_alloc_path();
5367 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5372 slot = path->slots[0];
5375 /* can't find the item, must cow */
5382 leaf = path->nodes[0];
5383 btrfs_item_key_to_cpu(leaf, &key, slot);
5384 if (key.objectid != btrfs_ino(inode) ||
5385 key.type != BTRFS_EXTENT_DATA_KEY) {
5386 /* not our file or wrong item type, must cow */
5390 if (key.offset > offset) {
5391 /* Wrong offset, must cow */
5395 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5396 found_type = btrfs_file_extent_type(leaf, fi);
5397 if (found_type != BTRFS_FILE_EXTENT_REG &&
5398 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5399 /* not a regular extent, must cow */
5402 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5403 backref_offset = btrfs_file_extent_offset(leaf, fi);
5405 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5406 if (extent_end < offset + len) {
5407 /* extent doesn't include our full range, must cow */
5411 if (btrfs_extent_readonly(root, disk_bytenr))
5415 * look for other files referencing this extent, if we
5416 * find any we must cow
5418 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5419 key.offset - backref_offset, disk_bytenr))
5423 * adjust disk_bytenr and num_bytes to cover just the bytes
5424 * in this extent we are about to write. If there
5425 * are any csums in that range we have to cow in order
5426 * to keep the csums correct
5428 disk_bytenr += backref_offset;
5429 disk_bytenr += offset - key.offset;
5430 num_bytes = min(offset + len, extent_end) - offset;
5431 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5434 * all of the above have passed, it is safe to overwrite this extent
5439 btrfs_free_path(path);
5443 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5444 struct buffer_head *bh_result, int create)
5446 struct extent_map *em;
5447 struct btrfs_root *root = BTRFS_I(inode)->root;
5448 u64 start = iblock << inode->i_blkbits;
5449 u64 len = bh_result->b_size;
5450 struct btrfs_trans_handle *trans;
5452 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5457 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5458 * io. INLINE is special, and we could probably kludge it in here, but
5459 * it's still buffered so for safety lets just fall back to the generic
5462 * For COMPRESSED we _have_ to read the entire extent in so we can
5463 * decompress it, so there will be buffering required no matter what we
5464 * do, so go ahead and fallback to buffered.
5466 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5467 * to buffered IO. Don't blame me, this is the price we pay for using
5470 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5471 em->block_start == EXTENT_MAP_INLINE) {
5472 free_extent_map(em);
5476 /* Just a good old fashioned hole, return */
5477 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5478 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5479 free_extent_map(em);
5480 /* DIO will do one hole at a time, so just unlock a sector */
5481 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5482 start + root->sectorsize - 1, GFP_NOFS);
5487 * We don't allocate a new extent in the following cases
5489 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5491 * 2) The extent is marked as PREALLOC. We're good to go here and can
5492 * just use the extent.
5496 len = em->len - (start - em->start);
5500 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5501 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5502 em->block_start != EXTENT_MAP_HOLE)) {
5507 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5508 type = BTRFS_ORDERED_PREALLOC;
5510 type = BTRFS_ORDERED_NOCOW;
5511 len = min(len, em->len - (start - em->start));
5512 block_start = em->block_start + (start - em->start);
5515 * we're not going to log anything, but we do need
5516 * to make sure the current transaction stays open
5517 * while we look for nocow cross refs
5519 trans = btrfs_join_transaction(root);
5523 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5524 ret = btrfs_add_ordered_extent_dio(inode, start,
5525 block_start, len, len, type);
5526 btrfs_end_transaction(trans, root);
5528 free_extent_map(em);
5533 btrfs_end_transaction(trans, root);
5537 * this will cow the extent, reset the len in case we changed
5540 len = bh_result->b_size;
5541 em = btrfs_new_extent_direct(inode, em, start, len);
5544 len = min(len, em->len - (start - em->start));
5546 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5547 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5550 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5552 bh_result->b_size = len;
5553 bh_result->b_bdev = em->bdev;
5554 set_buffer_mapped(bh_result);
5555 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5556 set_buffer_new(bh_result);
5558 free_extent_map(em);
5563 struct btrfs_dio_private {
5564 struct inode *inode;
5571 /* number of bios pending for this dio */
5572 atomic_t pending_bios;
5577 struct bio *orig_bio;
5580 static void btrfs_endio_direct_read(struct bio *bio, int err)
5582 struct btrfs_dio_private *dip = bio->bi_private;
5583 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5584 struct bio_vec *bvec = bio->bi_io_vec;
5585 struct inode *inode = dip->inode;
5586 struct btrfs_root *root = BTRFS_I(inode)->root;
5588 u32 *private = dip->csums;
5590 start = dip->logical_offset;
5592 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5593 struct page *page = bvec->bv_page;
5596 unsigned long flags;
5598 local_irq_save(flags);
5599 kaddr = kmap_atomic(page, KM_IRQ0);
5600 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5601 csum, bvec->bv_len);
5602 btrfs_csum_final(csum, (char *)&csum);
5603 kunmap_atomic(kaddr, KM_IRQ0);
5604 local_irq_restore(flags);
5606 flush_dcache_page(bvec->bv_page);
5607 if (csum != *private) {
5608 printk(KERN_ERR "btrfs csum failed ino %llu off"
5609 " %llu csum %u private %u\n",
5610 (unsigned long long)btrfs_ino(inode),
5611 (unsigned long long)start,
5617 start += bvec->bv_len;
5620 } while (bvec <= bvec_end);
5622 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5623 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5624 bio->bi_private = dip->private;
5629 /* If we had a csum failure make sure to clear the uptodate flag */
5631 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5632 dio_end_io(bio, err);
5635 static void btrfs_endio_direct_write(struct bio *bio, int err)
5637 struct btrfs_dio_private *dip = bio->bi_private;
5638 struct inode *inode = dip->inode;
5639 struct btrfs_root *root = BTRFS_I(inode)->root;
5640 struct btrfs_trans_handle *trans;
5641 struct btrfs_ordered_extent *ordered = NULL;
5642 struct extent_state *cached_state = NULL;
5643 u64 ordered_offset = dip->logical_offset;
5644 u64 ordered_bytes = dip->bytes;
5650 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5658 trans = btrfs_join_transaction(root);
5659 if (IS_ERR(trans)) {
5663 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5665 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5666 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5668 err = btrfs_update_inode_fallback(trans, root, inode);
5672 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5673 ordered->file_offset + ordered->len - 1, 0,
5674 &cached_state, GFP_NOFS);
5676 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5677 ret = btrfs_mark_extent_written(trans, inode,
5678 ordered->file_offset,
5679 ordered->file_offset +
5686 ret = insert_reserved_file_extent(trans, inode,
5687 ordered->file_offset,
5693 BTRFS_FILE_EXTENT_REG);
5694 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5695 ordered->file_offset, ordered->len);
5703 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5704 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5705 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5706 btrfs_update_inode_fallback(trans, root, inode);
5709 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5710 ordered->file_offset + ordered->len - 1,
5711 &cached_state, GFP_NOFS);
5713 btrfs_delalloc_release_metadata(inode, ordered->len);
5714 btrfs_end_transaction(trans, root);
5715 ordered_offset = ordered->file_offset + ordered->len;
5716 btrfs_put_ordered_extent(ordered);
5717 btrfs_put_ordered_extent(ordered);
5721 * our bio might span multiple ordered extents. If we haven't
5722 * completed the accounting for the whole dio, go back and try again
5724 if (ordered_offset < dip->logical_offset + dip->bytes) {
5725 ordered_bytes = dip->logical_offset + dip->bytes -
5730 bio->bi_private = dip->private;
5735 /* If we had an error make sure to clear the uptodate flag */
5737 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5738 dio_end_io(bio, err);
5741 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5742 struct bio *bio, int mirror_num,
5743 unsigned long bio_flags, u64 offset)
5746 struct btrfs_root *root = BTRFS_I(inode)->root;
5747 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5752 static void btrfs_end_dio_bio(struct bio *bio, int err)
5754 struct btrfs_dio_private *dip = bio->bi_private;
5757 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5758 "sector %#Lx len %u err no %d\n",
5759 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5760 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5764 * before atomic variable goto zero, we must make sure
5765 * dip->errors is perceived to be set.
5767 smp_mb__before_atomic_dec();
5770 /* if there are more bios still pending for this dio, just exit */
5771 if (!atomic_dec_and_test(&dip->pending_bios))
5775 bio_io_error(dip->orig_bio);
5777 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5778 bio_endio(dip->orig_bio, 0);
5784 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5785 u64 first_sector, gfp_t gfp_flags)
5787 int nr_vecs = bio_get_nr_vecs(bdev);
5788 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5791 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5792 int rw, u64 file_offset, int skip_sum,
5793 u32 *csums, int async_submit)
5795 int write = rw & REQ_WRITE;
5796 struct btrfs_root *root = BTRFS_I(inode)->root;
5800 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5807 if (write && async_submit) {
5808 ret = btrfs_wq_submit_bio(root->fs_info,
5809 inode, rw, bio, 0, 0,
5811 __btrfs_submit_bio_start_direct_io,
5812 __btrfs_submit_bio_done);
5816 * If we aren't doing async submit, calculate the csum of the
5819 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5822 } else if (!skip_sum) {
5823 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5824 file_offset, csums);
5830 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5836 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5839 struct inode *inode = dip->inode;
5840 struct btrfs_root *root = BTRFS_I(inode)->root;
5841 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5843 struct bio *orig_bio = dip->orig_bio;
5844 struct bio_vec *bvec = orig_bio->bi_io_vec;
5845 u64 start_sector = orig_bio->bi_sector;
5846 u64 file_offset = dip->logical_offset;
5850 u32 *csums = dip->csums;
5852 int async_submit = 0;
5853 int write = rw & REQ_WRITE;
5855 map_length = orig_bio->bi_size;
5856 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5857 &map_length, NULL, 0);
5863 if (map_length >= orig_bio->bi_size) {
5869 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5872 bio->bi_private = dip;
5873 bio->bi_end_io = btrfs_end_dio_bio;
5874 atomic_inc(&dip->pending_bios);
5876 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5877 if (unlikely(map_length < submit_len + bvec->bv_len ||
5878 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5879 bvec->bv_offset) < bvec->bv_len)) {
5881 * inc the count before we submit the bio so
5882 * we know the end IO handler won't happen before
5883 * we inc the count. Otherwise, the dip might get freed
5884 * before we're done setting it up
5886 atomic_inc(&dip->pending_bios);
5887 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5888 file_offset, skip_sum,
5889 csums, async_submit);
5892 atomic_dec(&dip->pending_bios);
5896 /* Write's use the ordered csums */
5897 if (!write && !skip_sum)
5898 csums = csums + nr_pages;
5899 start_sector += submit_len >> 9;
5900 file_offset += submit_len;
5905 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5906 start_sector, GFP_NOFS);
5909 bio->bi_private = dip;
5910 bio->bi_end_io = btrfs_end_dio_bio;
5912 map_length = orig_bio->bi_size;
5913 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5914 &map_length, NULL, 0);
5920 submit_len += bvec->bv_len;
5927 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5928 csums, async_submit);
5936 * before atomic variable goto zero, we must
5937 * make sure dip->errors is perceived to be set.
5939 smp_mb__before_atomic_dec();
5940 if (atomic_dec_and_test(&dip->pending_bios))
5941 bio_io_error(dip->orig_bio);
5943 /* bio_end_io() will handle error, so we needn't return it */
5947 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5950 struct btrfs_root *root = BTRFS_I(inode)->root;
5951 struct btrfs_dio_private *dip;
5952 struct bio_vec *bvec = bio->bi_io_vec;
5954 int write = rw & REQ_WRITE;
5957 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5959 dip = kmalloc(sizeof(*dip), GFP_NOFS);
5966 /* Write's use the ordered csum stuff, so we don't need dip->csums */
5967 if (!write && !skip_sum) {
5968 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5976 dip->private = bio->bi_private;
5978 dip->logical_offset = file_offset;
5982 dip->bytes += bvec->bv_len;
5984 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5986 dip->disk_bytenr = (u64)bio->bi_sector << 9;
5987 bio->bi_private = dip;
5989 dip->orig_bio = bio;
5990 atomic_set(&dip->pending_bios, 0);
5993 bio->bi_end_io = btrfs_endio_direct_write;
5995 bio->bi_end_io = btrfs_endio_direct_read;
5997 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6002 * If this is a write, we need to clean up the reserved space and kill
6003 * the ordered extent.
6006 struct btrfs_ordered_extent *ordered;
6007 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6008 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6009 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6010 btrfs_free_reserved_extent(root, ordered->start,
6012 btrfs_put_ordered_extent(ordered);
6013 btrfs_put_ordered_extent(ordered);
6015 bio_endio(bio, ret);
6018 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6019 const struct iovec *iov, loff_t offset,
6020 unsigned long nr_segs)
6026 unsigned blocksize_mask = root->sectorsize - 1;
6027 ssize_t retval = -EINVAL;
6028 loff_t end = offset;
6030 if (offset & blocksize_mask)
6033 /* Check the memory alignment. Blocks cannot straddle pages */
6034 for (seg = 0; seg < nr_segs; seg++) {
6035 addr = (unsigned long)iov[seg].iov_base;
6036 size = iov[seg].iov_len;
6038 if ((addr & blocksize_mask) || (size & blocksize_mask))
6041 /* If this is a write we don't need to check anymore */
6046 * Check to make sure we don't have duplicate iov_base's in this
6047 * iovec, if so return EINVAL, otherwise we'll get csum errors
6048 * when reading back.
6050 for (i = seg + 1; i < nr_segs; i++) {
6051 if (iov[seg].iov_base == iov[i].iov_base)
6059 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6060 const struct iovec *iov, loff_t offset,
6061 unsigned long nr_segs)
6063 struct file *file = iocb->ki_filp;
6064 struct inode *inode = file->f_mapping->host;
6065 struct btrfs_ordered_extent *ordered;
6066 struct extent_state *cached_state = NULL;
6067 u64 lockstart, lockend;
6069 int writing = rw & WRITE;
6071 size_t count = iov_length(iov, nr_segs);
6073 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6079 lockend = offset + count - 1;
6082 ret = btrfs_delalloc_reserve_space(inode, count);
6088 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6089 0, &cached_state, GFP_NOFS);
6091 * We're concerned with the entire range that we're going to be
6092 * doing DIO to, so we need to make sure theres no ordered
6093 * extents in this range.
6095 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6096 lockend - lockstart + 1);
6099 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6100 &cached_state, GFP_NOFS);
6101 btrfs_start_ordered_extent(inode, ordered, 1);
6102 btrfs_put_ordered_extent(ordered);
6107 * we don't use btrfs_set_extent_delalloc because we don't want
6108 * the dirty or uptodate bits
6111 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6112 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6113 EXTENT_DELALLOC, 0, NULL, &cached_state,
6116 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6117 lockend, EXTENT_LOCKED | write_bits,
6118 1, 0, &cached_state, GFP_NOFS);
6123 free_extent_state(cached_state);
6124 cached_state = NULL;
6126 ret = __blockdev_direct_IO(rw, iocb, inode,
6127 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6128 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6129 btrfs_submit_direct, 0);
6131 if (ret < 0 && ret != -EIOCBQUEUED) {
6132 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6133 offset + iov_length(iov, nr_segs) - 1,
6134 EXTENT_LOCKED | write_bits, 1, 0,
6135 &cached_state, GFP_NOFS);
6136 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6138 * We're falling back to buffered, unlock the section we didn't
6141 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6142 offset + iov_length(iov, nr_segs) - 1,
6143 EXTENT_LOCKED | write_bits, 1, 0,
6144 &cached_state, GFP_NOFS);
6147 free_extent_state(cached_state);
6151 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6152 __u64 start, __u64 len)
6154 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6157 int btrfs_readpage(struct file *file, struct page *page)
6159 struct extent_io_tree *tree;
6160 tree = &BTRFS_I(page->mapping->host)->io_tree;
6161 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6164 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6166 struct extent_io_tree *tree;
6169 if (current->flags & PF_MEMALLOC) {
6170 redirty_page_for_writepage(wbc, page);
6174 tree = &BTRFS_I(page->mapping->host)->io_tree;
6175 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6178 int btrfs_writepages(struct address_space *mapping,
6179 struct writeback_control *wbc)
6181 struct extent_io_tree *tree;
6183 tree = &BTRFS_I(mapping->host)->io_tree;
6184 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6188 btrfs_readpages(struct file *file, struct address_space *mapping,
6189 struct list_head *pages, unsigned nr_pages)
6191 struct extent_io_tree *tree;
6192 tree = &BTRFS_I(mapping->host)->io_tree;
6193 return extent_readpages(tree, mapping, pages, nr_pages,
6196 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6198 struct extent_io_tree *tree;
6199 struct extent_map_tree *map;
6202 tree = &BTRFS_I(page->mapping->host)->io_tree;
6203 map = &BTRFS_I(page->mapping->host)->extent_tree;
6204 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6206 ClearPagePrivate(page);
6207 set_page_private(page, 0);
6208 page_cache_release(page);
6213 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6215 if (PageWriteback(page) || PageDirty(page))
6217 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6220 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6222 struct extent_io_tree *tree;
6223 struct btrfs_ordered_extent *ordered;
6224 struct extent_state *cached_state = NULL;
6225 u64 page_start = page_offset(page);
6226 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6230 * we have the page locked, so new writeback can't start,
6231 * and the dirty bit won't be cleared while we are here.
6233 * Wait for IO on this page so that we can safely clear
6234 * the PagePrivate2 bit and do ordered accounting
6236 wait_on_page_writeback(page);
6238 tree = &BTRFS_I(page->mapping->host)->io_tree;
6240 btrfs_releasepage(page, GFP_NOFS);
6243 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6245 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6249 * IO on this page will never be started, so we need
6250 * to account for any ordered extents now
6252 clear_extent_bit(tree, page_start, page_end,
6253 EXTENT_DIRTY | EXTENT_DELALLOC |
6254 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6255 &cached_state, GFP_NOFS);
6257 * whoever cleared the private bit is responsible
6258 * for the finish_ordered_io
6260 if (TestClearPagePrivate2(page)) {
6261 btrfs_finish_ordered_io(page->mapping->host,
6262 page_start, page_end);
6264 btrfs_put_ordered_extent(ordered);
6265 cached_state = NULL;
6266 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6269 clear_extent_bit(tree, page_start, page_end,
6270 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6271 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6272 __btrfs_releasepage(page, GFP_NOFS);
6274 ClearPageChecked(page);
6275 if (PagePrivate(page)) {
6276 ClearPagePrivate(page);
6277 set_page_private(page, 0);
6278 page_cache_release(page);
6283 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6284 * called from a page fault handler when a page is first dirtied. Hence we must
6285 * be careful to check for EOF conditions here. We set the page up correctly
6286 * for a written page which means we get ENOSPC checking when writing into
6287 * holes and correct delalloc and unwritten extent mapping on filesystems that
6288 * support these features.
6290 * We are not allowed to take the i_mutex here so we have to play games to
6291 * protect against truncate races as the page could now be beyond EOF. Because
6292 * vmtruncate() writes the inode size before removing pages, once we have the
6293 * page lock we can determine safely if the page is beyond EOF. If it is not
6294 * beyond EOF, then the page is guaranteed safe against truncation until we
6297 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6299 struct page *page = vmf->page;
6300 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6301 struct btrfs_root *root = BTRFS_I(inode)->root;
6302 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6303 struct btrfs_ordered_extent *ordered;
6304 struct extent_state *cached_state = NULL;
6306 unsigned long zero_start;
6312 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6316 else /* -ENOSPC, -EIO, etc */
6317 ret = VM_FAULT_SIGBUS;
6321 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6324 size = i_size_read(inode);
6325 page_start = page_offset(page);
6326 page_end = page_start + PAGE_CACHE_SIZE - 1;
6328 if ((page->mapping != inode->i_mapping) ||
6329 (page_start >= size)) {
6330 /* page got truncated out from underneath us */
6333 wait_on_page_writeback(page);
6335 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6337 set_page_extent_mapped(page);
6340 * we can't set the delalloc bits if there are pending ordered
6341 * extents. Drop our locks and wait for them to finish
6343 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6345 unlock_extent_cached(io_tree, page_start, page_end,
6346 &cached_state, GFP_NOFS);
6348 btrfs_start_ordered_extent(inode, ordered, 1);
6349 btrfs_put_ordered_extent(ordered);
6354 * XXX - page_mkwrite gets called every time the page is dirtied, even
6355 * if it was already dirty, so for space accounting reasons we need to
6356 * clear any delalloc bits for the range we are fixing to save. There
6357 * is probably a better way to do this, but for now keep consistent with
6358 * prepare_pages in the normal write path.
6360 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6361 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6362 0, 0, &cached_state, GFP_NOFS);
6364 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6367 unlock_extent_cached(io_tree, page_start, page_end,
6368 &cached_state, GFP_NOFS);
6369 ret = VM_FAULT_SIGBUS;
6374 /* page is wholly or partially inside EOF */
6375 if (page_start + PAGE_CACHE_SIZE > size)
6376 zero_start = size & ~PAGE_CACHE_MASK;
6378 zero_start = PAGE_CACHE_SIZE;
6380 if (zero_start != PAGE_CACHE_SIZE) {
6382 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6383 flush_dcache_page(page);
6386 ClearPageChecked(page);
6387 set_page_dirty(page);
6388 SetPageUptodate(page);
6390 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6391 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6393 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6397 return VM_FAULT_LOCKED;
6399 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6404 static int btrfs_truncate(struct inode *inode)
6406 struct btrfs_root *root = BTRFS_I(inode)->root;
6407 struct btrfs_block_rsv *rsv;
6410 struct btrfs_trans_handle *trans;
6412 u64 mask = root->sectorsize - 1;
6413 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6415 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6419 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6420 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6423 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6424 * 3 things going on here
6426 * 1) We need to reserve space for our orphan item and the space to
6427 * delete our orphan item. Lord knows we don't want to have a dangling
6428 * orphan item because we didn't reserve space to remove it.
6430 * 2) We need to reserve space to update our inode.
6432 * 3) We need to have something to cache all the space that is going to
6433 * be free'd up by the truncate operation, but also have some slack
6434 * space reserved in case it uses space during the truncate (thank you
6435 * very much snapshotting).
6437 * And we need these to all be seperate. The fact is we can use alot of
6438 * space doing the truncate, and we have no earthly idea how much space
6439 * we will use, so we need the truncate reservation to be seperate so it
6440 * doesn't end up using space reserved for updating the inode or
6441 * removing the orphan item. We also need to be able to stop the
6442 * transaction and start a new one, which means we need to be able to
6443 * update the inode several times, and we have no idea of knowing how
6444 * many times that will be, so we can't just reserve 1 item for the
6445 * entirety of the opration, so that has to be done seperately as well.
6446 * Then there is the orphan item, which does indeed need to be held on
6447 * to for the whole operation, and we need nobody to touch this reserved
6448 * space except the orphan code.
6450 * So that leaves us with
6452 * 1) root->orphan_block_rsv - for the orphan deletion.
6453 * 2) rsv - for the truncate reservation, which we will steal from the
6454 * transaction reservation.
6455 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6456 * updating the inode.
6458 rsv = btrfs_alloc_block_rsv(root);
6461 rsv->size = min_size;
6464 * 1 for the truncate slack space
6465 * 1 for the orphan item we're going to add
6466 * 1 for the orphan item deletion
6467 * 1 for updating the inode.
6469 trans = btrfs_start_transaction(root, 4);
6470 if (IS_ERR(trans)) {
6471 err = PTR_ERR(trans);
6475 /* Migrate the slack space for the truncate to our reserve */
6476 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6480 ret = btrfs_orphan_add(trans, inode);
6482 btrfs_end_transaction(trans, root);
6487 * setattr is responsible for setting the ordered_data_close flag,
6488 * but that is only tested during the last file release. That
6489 * could happen well after the next commit, leaving a great big
6490 * window where new writes may get lost if someone chooses to write
6491 * to this file after truncating to zero
6493 * The inode doesn't have any dirty data here, and so if we commit
6494 * this is a noop. If someone immediately starts writing to the inode
6495 * it is very likely we'll catch some of their writes in this
6496 * transaction, and the commit will find this file on the ordered
6497 * data list with good things to send down.
6499 * This is a best effort solution, there is still a window where
6500 * using truncate to replace the contents of the file will
6501 * end up with a zero length file after a crash.
6503 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6504 btrfs_add_ordered_operation(trans, root, inode);
6507 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6510 * This can only happen with the original transaction we
6511 * started above, every other time we shouldn't have a
6512 * transaction started yet.
6521 /* Just need the 1 for updating the inode */
6522 trans = btrfs_start_transaction(root, 1);
6523 if (IS_ERR(trans)) {
6524 err = PTR_ERR(trans);
6529 trans->block_rsv = rsv;
6531 ret = btrfs_truncate_inode_items(trans, root, inode,
6533 BTRFS_EXTENT_DATA_KEY);
6534 if (ret != -EAGAIN) {
6539 trans->block_rsv = &root->fs_info->trans_block_rsv;
6540 ret = btrfs_update_inode(trans, root, inode);
6546 nr = trans->blocks_used;
6547 btrfs_end_transaction(trans, root);
6549 btrfs_btree_balance_dirty(root, nr);
6552 if (ret == 0 && inode->i_nlink > 0) {
6553 trans->block_rsv = root->orphan_block_rsv;
6554 ret = btrfs_orphan_del(trans, inode);
6557 } else if (ret && inode->i_nlink > 0) {
6559 * Failed to do the truncate, remove us from the in memory
6562 ret = btrfs_orphan_del(NULL, inode);
6566 trans->block_rsv = &root->fs_info->trans_block_rsv;
6567 ret = btrfs_update_inode(trans, root, inode);
6571 nr = trans->blocks_used;
6572 ret = btrfs_end_transaction_throttle(trans, root);
6573 btrfs_btree_balance_dirty(root, nr);
6577 btrfs_free_block_rsv(root, rsv);
6586 * create a new subvolume directory/inode (helper for the ioctl).
6588 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6589 struct btrfs_root *new_root, u64 new_dirid)
6591 struct inode *inode;
6595 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6596 new_dirid, S_IFDIR | 0700, &index);
6598 return PTR_ERR(inode);
6599 inode->i_op = &btrfs_dir_inode_operations;
6600 inode->i_fop = &btrfs_dir_file_operations;
6603 btrfs_i_size_write(inode, 0);
6605 err = btrfs_update_inode(trans, new_root, inode);
6612 struct inode *btrfs_alloc_inode(struct super_block *sb)
6614 struct btrfs_inode *ei;
6615 struct inode *inode;
6617 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6622 ei->space_info = NULL;
6626 ei->last_sub_trans = 0;
6627 ei->logged_trans = 0;
6628 ei->delalloc_bytes = 0;
6629 ei->disk_i_size = 0;
6632 ei->index_cnt = (u64)-1;
6633 ei->last_unlink_trans = 0;
6635 spin_lock_init(&ei->lock);
6636 ei->outstanding_extents = 0;
6637 ei->reserved_extents = 0;
6639 ei->ordered_data_close = 0;
6640 ei->orphan_meta_reserved = 0;
6641 ei->dummy_inode = 0;
6643 ei->delalloc_meta_reserved = 0;
6644 ei->force_compress = BTRFS_COMPRESS_NONE;
6646 ei->delayed_node = NULL;
6648 inode = &ei->vfs_inode;
6649 extent_map_tree_init(&ei->extent_tree);
6650 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6651 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6652 mutex_init(&ei->log_mutex);
6653 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6654 INIT_LIST_HEAD(&ei->i_orphan);
6655 INIT_LIST_HEAD(&ei->delalloc_inodes);
6656 INIT_LIST_HEAD(&ei->ordered_operations);
6657 RB_CLEAR_NODE(&ei->rb_node);
6662 static void btrfs_i_callback(struct rcu_head *head)
6664 struct inode *inode = container_of(head, struct inode, i_rcu);
6665 INIT_LIST_HEAD(&inode->i_dentry);
6666 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6669 void btrfs_destroy_inode(struct inode *inode)
6671 struct btrfs_ordered_extent *ordered;
6672 struct btrfs_root *root = BTRFS_I(inode)->root;
6674 WARN_ON(!list_empty(&inode->i_dentry));
6675 WARN_ON(inode->i_data.nrpages);
6676 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6677 WARN_ON(BTRFS_I(inode)->reserved_extents);
6678 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6679 WARN_ON(BTRFS_I(inode)->csum_bytes);
6682 * This can happen where we create an inode, but somebody else also
6683 * created the same inode and we need to destroy the one we already
6690 * Make sure we're properly removed from the ordered operation
6694 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6695 spin_lock(&root->fs_info->ordered_extent_lock);
6696 list_del_init(&BTRFS_I(inode)->ordered_operations);
6697 spin_unlock(&root->fs_info->ordered_extent_lock);
6700 spin_lock(&root->orphan_lock);
6701 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6702 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6703 (unsigned long long)btrfs_ino(inode));
6704 list_del_init(&BTRFS_I(inode)->i_orphan);
6706 spin_unlock(&root->orphan_lock);
6709 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6713 printk(KERN_ERR "btrfs found ordered "
6714 "extent %llu %llu on inode cleanup\n",
6715 (unsigned long long)ordered->file_offset,
6716 (unsigned long long)ordered->len);
6717 btrfs_remove_ordered_extent(inode, ordered);
6718 btrfs_put_ordered_extent(ordered);
6719 btrfs_put_ordered_extent(ordered);
6722 inode_tree_del(inode);
6723 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6725 btrfs_remove_delayed_node(inode);
6726 call_rcu(&inode->i_rcu, btrfs_i_callback);
6729 int btrfs_drop_inode(struct inode *inode)
6731 struct btrfs_root *root = BTRFS_I(inode)->root;
6733 if (btrfs_root_refs(&root->root_item) == 0 &&
6734 !btrfs_is_free_space_inode(root, inode))
6737 return generic_drop_inode(inode);
6740 static void init_once(void *foo)
6742 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6744 inode_init_once(&ei->vfs_inode);
6747 void btrfs_destroy_cachep(void)
6749 if (btrfs_inode_cachep)
6750 kmem_cache_destroy(btrfs_inode_cachep);
6751 if (btrfs_trans_handle_cachep)
6752 kmem_cache_destroy(btrfs_trans_handle_cachep);
6753 if (btrfs_transaction_cachep)
6754 kmem_cache_destroy(btrfs_transaction_cachep);
6755 if (btrfs_path_cachep)
6756 kmem_cache_destroy(btrfs_path_cachep);
6757 if (btrfs_free_space_cachep)
6758 kmem_cache_destroy(btrfs_free_space_cachep);
6761 int btrfs_init_cachep(void)
6763 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6764 sizeof(struct btrfs_inode), 0,
6765 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6766 if (!btrfs_inode_cachep)
6769 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6770 sizeof(struct btrfs_trans_handle), 0,
6771 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6772 if (!btrfs_trans_handle_cachep)
6775 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6776 sizeof(struct btrfs_transaction), 0,
6777 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6778 if (!btrfs_transaction_cachep)
6781 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6782 sizeof(struct btrfs_path), 0,
6783 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6784 if (!btrfs_path_cachep)
6787 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6788 sizeof(struct btrfs_free_space), 0,
6789 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6790 if (!btrfs_free_space_cachep)
6795 btrfs_destroy_cachep();
6799 static int btrfs_getattr(struct vfsmount *mnt,
6800 struct dentry *dentry, struct kstat *stat)
6802 struct inode *inode = dentry->d_inode;
6803 u32 blocksize = inode->i_sb->s_blocksize;
6805 generic_fillattr(inode, stat);
6806 stat->dev = BTRFS_I(inode)->root->anon_dev;
6807 stat->blksize = PAGE_CACHE_SIZE;
6808 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6809 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6814 * If a file is moved, it will inherit the cow and compression flags of the new
6817 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6819 struct btrfs_inode *b_dir = BTRFS_I(dir);
6820 struct btrfs_inode *b_inode = BTRFS_I(inode);
6822 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6823 b_inode->flags |= BTRFS_INODE_NODATACOW;
6825 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6827 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6828 b_inode->flags |= BTRFS_INODE_COMPRESS;
6830 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6833 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6834 struct inode *new_dir, struct dentry *new_dentry)
6836 struct btrfs_trans_handle *trans;
6837 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6838 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6839 struct inode *new_inode = new_dentry->d_inode;
6840 struct inode *old_inode = old_dentry->d_inode;
6841 struct timespec ctime = CURRENT_TIME;
6845 u64 old_ino = btrfs_ino(old_inode);
6847 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6850 /* we only allow rename subvolume link between subvolumes */
6851 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6854 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6855 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6858 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6859 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6862 * we're using rename to replace one file with another.
6863 * and the replacement file is large. Start IO on it now so
6864 * we don't add too much work to the end of the transaction
6866 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6867 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6868 filemap_flush(old_inode->i_mapping);
6870 /* close the racy window with snapshot create/destroy ioctl */
6871 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6872 down_read(&root->fs_info->subvol_sem);
6874 * We want to reserve the absolute worst case amount of items. So if
6875 * both inodes are subvols and we need to unlink them then that would
6876 * require 4 item modifications, but if they are both normal inodes it
6877 * would require 5 item modifications, so we'll assume their normal
6878 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6879 * should cover the worst case number of items we'll modify.
6881 trans = btrfs_start_transaction(root, 20);
6882 if (IS_ERR(trans)) {
6883 ret = PTR_ERR(trans);
6888 btrfs_record_root_in_trans(trans, dest);
6890 ret = btrfs_set_inode_index(new_dir, &index);
6894 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6895 /* force full log commit if subvolume involved. */
6896 root->fs_info->last_trans_log_full_commit = trans->transid;
6898 ret = btrfs_insert_inode_ref(trans, dest,
6899 new_dentry->d_name.name,
6900 new_dentry->d_name.len,
6902 btrfs_ino(new_dir), index);
6906 * this is an ugly little race, but the rename is required
6907 * to make sure that if we crash, the inode is either at the
6908 * old name or the new one. pinning the log transaction lets
6909 * us make sure we don't allow a log commit to come in after
6910 * we unlink the name but before we add the new name back in.
6912 btrfs_pin_log_trans(root);
6915 * make sure the inode gets flushed if it is replacing
6918 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6919 btrfs_add_ordered_operation(trans, root, old_inode);
6921 old_dir->i_ctime = old_dir->i_mtime = ctime;
6922 new_dir->i_ctime = new_dir->i_mtime = ctime;
6923 old_inode->i_ctime = ctime;
6925 if (old_dentry->d_parent != new_dentry->d_parent)
6926 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6928 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6929 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6930 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6931 old_dentry->d_name.name,
6932 old_dentry->d_name.len);
6934 ret = __btrfs_unlink_inode(trans, root, old_dir,
6935 old_dentry->d_inode,
6936 old_dentry->d_name.name,
6937 old_dentry->d_name.len);
6939 ret = btrfs_update_inode(trans, root, old_inode);
6944 new_inode->i_ctime = CURRENT_TIME;
6945 if (unlikely(btrfs_ino(new_inode) ==
6946 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6947 root_objectid = BTRFS_I(new_inode)->location.objectid;
6948 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6950 new_dentry->d_name.name,
6951 new_dentry->d_name.len);
6952 BUG_ON(new_inode->i_nlink == 0);
6954 ret = btrfs_unlink_inode(trans, dest, new_dir,
6955 new_dentry->d_inode,
6956 new_dentry->d_name.name,
6957 new_dentry->d_name.len);
6960 if (new_inode->i_nlink == 0) {
6961 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6966 fixup_inode_flags(new_dir, old_inode);
6968 ret = btrfs_add_link(trans, new_dir, old_inode,
6969 new_dentry->d_name.name,
6970 new_dentry->d_name.len, 0, index);
6973 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
6974 struct dentry *parent = new_dentry->d_parent;
6975 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6976 btrfs_end_log_trans(root);
6979 btrfs_end_transaction_throttle(trans, root);
6981 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6982 up_read(&root->fs_info->subvol_sem);
6988 * some fairly slow code that needs optimization. This walks the list
6989 * of all the inodes with pending delalloc and forces them to disk.
6991 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6993 struct list_head *head = &root->fs_info->delalloc_inodes;
6994 struct btrfs_inode *binode;
6995 struct inode *inode;
6997 if (root->fs_info->sb->s_flags & MS_RDONLY)
7000 spin_lock(&root->fs_info->delalloc_lock);
7001 while (!list_empty(head)) {
7002 binode = list_entry(head->next, struct btrfs_inode,
7004 inode = igrab(&binode->vfs_inode);
7006 list_del_init(&binode->delalloc_inodes);
7007 spin_unlock(&root->fs_info->delalloc_lock);
7009 filemap_flush(inode->i_mapping);
7011 btrfs_add_delayed_iput(inode);
7016 spin_lock(&root->fs_info->delalloc_lock);
7018 spin_unlock(&root->fs_info->delalloc_lock);
7020 /* the filemap_flush will queue IO into the worker threads, but
7021 * we have to make sure the IO is actually started and that
7022 * ordered extents get created before we return
7024 atomic_inc(&root->fs_info->async_submit_draining);
7025 while (atomic_read(&root->fs_info->nr_async_submits) ||
7026 atomic_read(&root->fs_info->async_delalloc_pages)) {
7027 wait_event(root->fs_info->async_submit_wait,
7028 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7029 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7031 atomic_dec(&root->fs_info->async_submit_draining);
7035 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7036 const char *symname)
7038 struct btrfs_trans_handle *trans;
7039 struct btrfs_root *root = BTRFS_I(dir)->root;
7040 struct btrfs_path *path;
7041 struct btrfs_key key;
7042 struct inode *inode = NULL;
7050 struct btrfs_file_extent_item *ei;
7051 struct extent_buffer *leaf;
7052 unsigned long nr = 0;
7054 name_len = strlen(symname) + 1;
7055 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7056 return -ENAMETOOLONG;
7059 * 2 items for inode item and ref
7060 * 2 items for dir items
7061 * 1 item for xattr if selinux is on
7063 trans = btrfs_start_transaction(root, 5);
7065 return PTR_ERR(trans);
7067 err = btrfs_find_free_ino(root, &objectid);
7071 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7072 dentry->d_name.len, btrfs_ino(dir), objectid,
7073 S_IFLNK|S_IRWXUGO, &index);
7074 if (IS_ERR(inode)) {
7075 err = PTR_ERR(inode);
7079 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7085 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7089 inode->i_mapping->a_ops = &btrfs_aops;
7090 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7091 inode->i_fop = &btrfs_file_operations;
7092 inode->i_op = &btrfs_file_inode_operations;
7093 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7098 path = btrfs_alloc_path();
7104 key.objectid = btrfs_ino(inode);
7106 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7107 datasize = btrfs_file_extent_calc_inline_size(name_len);
7108 err = btrfs_insert_empty_item(trans, root, path, &key,
7112 btrfs_free_path(path);
7115 leaf = path->nodes[0];
7116 ei = btrfs_item_ptr(leaf, path->slots[0],
7117 struct btrfs_file_extent_item);
7118 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7119 btrfs_set_file_extent_type(leaf, ei,
7120 BTRFS_FILE_EXTENT_INLINE);
7121 btrfs_set_file_extent_encryption(leaf, ei, 0);
7122 btrfs_set_file_extent_compression(leaf, ei, 0);
7123 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7124 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7126 ptr = btrfs_file_extent_inline_start(ei);
7127 write_extent_buffer(leaf, symname, ptr, name_len);
7128 btrfs_mark_buffer_dirty(leaf);
7129 btrfs_free_path(path);
7131 inode->i_op = &btrfs_symlink_inode_operations;
7132 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7133 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7134 inode_set_bytes(inode, name_len);
7135 btrfs_i_size_write(inode, name_len - 1);
7136 err = btrfs_update_inode(trans, root, inode);
7141 nr = trans->blocks_used;
7142 btrfs_end_transaction_throttle(trans, root);
7144 inode_dec_link_count(inode);
7147 btrfs_btree_balance_dirty(root, nr);
7151 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7152 u64 start, u64 num_bytes, u64 min_size,
7153 loff_t actual_len, u64 *alloc_hint,
7154 struct btrfs_trans_handle *trans)
7156 struct btrfs_root *root = BTRFS_I(inode)->root;
7157 struct btrfs_key ins;
7158 u64 cur_offset = start;
7161 bool own_trans = true;
7165 while (num_bytes > 0) {
7167 trans = btrfs_start_transaction(root, 3);
7168 if (IS_ERR(trans)) {
7169 ret = PTR_ERR(trans);
7174 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7175 0, *alloc_hint, (u64)-1, &ins, 1);
7178 btrfs_end_transaction(trans, root);
7182 ret = insert_reserved_file_extent(trans, inode,
7183 cur_offset, ins.objectid,
7184 ins.offset, ins.offset,
7185 ins.offset, 0, 0, 0,
7186 BTRFS_FILE_EXTENT_PREALLOC);
7188 btrfs_drop_extent_cache(inode, cur_offset,
7189 cur_offset + ins.offset -1, 0);
7191 num_bytes -= ins.offset;
7192 cur_offset += ins.offset;
7193 *alloc_hint = ins.objectid + ins.offset;
7195 inode->i_ctime = CURRENT_TIME;
7196 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7197 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7198 (actual_len > inode->i_size) &&
7199 (cur_offset > inode->i_size)) {
7200 if (cur_offset > actual_len)
7201 i_size = actual_len;
7203 i_size = cur_offset;
7204 i_size_write(inode, i_size);
7205 btrfs_ordered_update_i_size(inode, i_size, NULL);
7208 ret = btrfs_update_inode(trans, root, inode);
7212 btrfs_end_transaction(trans, root);
7217 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7218 u64 start, u64 num_bytes, u64 min_size,
7219 loff_t actual_len, u64 *alloc_hint)
7221 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7222 min_size, actual_len, alloc_hint,
7226 int btrfs_prealloc_file_range_trans(struct inode *inode,
7227 struct btrfs_trans_handle *trans, int mode,
7228 u64 start, u64 num_bytes, u64 min_size,
7229 loff_t actual_len, u64 *alloc_hint)
7231 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7232 min_size, actual_len, alloc_hint, trans);
7235 static int btrfs_set_page_dirty(struct page *page)
7237 return __set_page_dirty_nobuffers(page);
7240 static int btrfs_permission(struct inode *inode, int mask)
7242 struct btrfs_root *root = BTRFS_I(inode)->root;
7243 umode_t mode = inode->i_mode;
7245 if (mask & MAY_WRITE &&
7246 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7247 if (btrfs_root_readonly(root))
7249 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7252 return generic_permission(inode, mask);
7255 static const struct inode_operations btrfs_dir_inode_operations = {
7256 .getattr = btrfs_getattr,
7257 .lookup = btrfs_lookup,
7258 .create = btrfs_create,
7259 .unlink = btrfs_unlink,
7261 .mkdir = btrfs_mkdir,
7262 .rmdir = btrfs_rmdir,
7263 .rename = btrfs_rename,
7264 .symlink = btrfs_symlink,
7265 .setattr = btrfs_setattr,
7266 .mknod = btrfs_mknod,
7267 .setxattr = btrfs_setxattr,
7268 .getxattr = btrfs_getxattr,
7269 .listxattr = btrfs_listxattr,
7270 .removexattr = btrfs_removexattr,
7271 .permission = btrfs_permission,
7272 .get_acl = btrfs_get_acl,
7274 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7275 .lookup = btrfs_lookup,
7276 .permission = btrfs_permission,
7277 .get_acl = btrfs_get_acl,
7280 static const struct file_operations btrfs_dir_file_operations = {
7281 .llseek = generic_file_llseek,
7282 .read = generic_read_dir,
7283 .readdir = btrfs_real_readdir,
7284 .unlocked_ioctl = btrfs_ioctl,
7285 #ifdef CONFIG_COMPAT
7286 .compat_ioctl = btrfs_ioctl,
7288 .release = btrfs_release_file,
7289 .fsync = btrfs_sync_file,
7292 static struct extent_io_ops btrfs_extent_io_ops = {
7293 .fill_delalloc = run_delalloc_range,
7294 .submit_bio_hook = btrfs_submit_bio_hook,
7295 .merge_bio_hook = btrfs_merge_bio_hook,
7296 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7297 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7298 .writepage_start_hook = btrfs_writepage_start_hook,
7299 .set_bit_hook = btrfs_set_bit_hook,
7300 .clear_bit_hook = btrfs_clear_bit_hook,
7301 .merge_extent_hook = btrfs_merge_extent_hook,
7302 .split_extent_hook = btrfs_split_extent_hook,
7306 * btrfs doesn't support the bmap operation because swapfiles
7307 * use bmap to make a mapping of extents in the file. They assume
7308 * these extents won't change over the life of the file and they
7309 * use the bmap result to do IO directly to the drive.
7311 * the btrfs bmap call would return logical addresses that aren't
7312 * suitable for IO and they also will change frequently as COW
7313 * operations happen. So, swapfile + btrfs == corruption.
7315 * For now we're avoiding this by dropping bmap.
7317 static const struct address_space_operations btrfs_aops = {
7318 .readpage = btrfs_readpage,
7319 .writepage = btrfs_writepage,
7320 .writepages = btrfs_writepages,
7321 .readpages = btrfs_readpages,
7322 .direct_IO = btrfs_direct_IO,
7323 .invalidatepage = btrfs_invalidatepage,
7324 .releasepage = btrfs_releasepage,
7325 .set_page_dirty = btrfs_set_page_dirty,
7326 .error_remove_page = generic_error_remove_page,
7329 static const struct address_space_operations btrfs_symlink_aops = {
7330 .readpage = btrfs_readpage,
7331 .writepage = btrfs_writepage,
7332 .invalidatepage = btrfs_invalidatepage,
7333 .releasepage = btrfs_releasepage,
7336 static const struct inode_operations btrfs_file_inode_operations = {
7337 .getattr = btrfs_getattr,
7338 .setattr = btrfs_setattr,
7339 .setxattr = btrfs_setxattr,
7340 .getxattr = btrfs_getxattr,
7341 .listxattr = btrfs_listxattr,
7342 .removexattr = btrfs_removexattr,
7343 .permission = btrfs_permission,
7344 .fiemap = btrfs_fiemap,
7345 .get_acl = btrfs_get_acl,
7347 static const struct inode_operations btrfs_special_inode_operations = {
7348 .getattr = btrfs_getattr,
7349 .setattr = btrfs_setattr,
7350 .permission = btrfs_permission,
7351 .setxattr = btrfs_setxattr,
7352 .getxattr = btrfs_getxattr,
7353 .listxattr = btrfs_listxattr,
7354 .removexattr = btrfs_removexattr,
7355 .get_acl = btrfs_get_acl,
7357 static const struct inode_operations btrfs_symlink_inode_operations = {
7358 .readlink = generic_readlink,
7359 .follow_link = page_follow_link_light,
7360 .put_link = page_put_link,
7361 .getattr = btrfs_getattr,
7362 .permission = btrfs_permission,
7363 .setxattr = btrfs_setxattr,
7364 .getxattr = btrfs_getxattr,
7365 .listxattr = btrfs_listxattr,
7366 .removexattr = btrfs_removexattr,
7367 .get_acl = btrfs_get_acl,
7370 const struct dentry_operations btrfs_dentry_operations = {
7371 .d_delete = btrfs_dentry_delete,
7372 .d_release = btrfs_dentry_release,