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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 static struct inode_operations btrfs_dir_inode_operations;
60 static struct inode_operations btrfs_symlink_inode_operations;
61 static struct inode_operations btrfs_dir_ro_inode_operations;
62 static struct inode_operations btrfs_special_inode_operations;
63 static struct inode_operations btrfs_file_inode_operations;
64 static struct address_space_operations btrfs_aops;
65 static struct address_space_operations btrfs_symlink_aops;
66 static struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
92 static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
96 err = btrfs_init_acl(inode, dir);
98 err = btrfs_xattr_security_init(inode, dir);
103 * this does all the hard work for inserting an inline extent into
104 * the btree. The caller should have done a btrfs_drop_extents so that
105 * no overlapping inline items exist in the btree
107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
108 struct btrfs_root *root, struct inode *inode,
109 u64 start, size_t size, size_t compressed_size,
110 struct page **compressed_pages)
112 struct btrfs_key key;
113 struct btrfs_path *path;
114 struct extent_buffer *leaf;
115 struct page *page = NULL;
118 struct btrfs_file_extent_item *ei;
121 size_t cur_size = size;
123 unsigned long offset;
124 int use_compress = 0;
126 if (compressed_size && compressed_pages) {
128 cur_size = compressed_size;
131 path = btrfs_alloc_path();
135 path->leave_spinning = 1;
136 btrfs_set_trans_block_group(trans, inode);
138 key.objectid = inode->i_ino;
140 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
141 datasize = btrfs_file_extent_calc_inline_size(cur_size);
143 inode_add_bytes(inode, size);
144 ret = btrfs_insert_empty_item(trans, root, path, &key,
151 leaf = path->nodes[0];
152 ei = btrfs_item_ptr(leaf, path->slots[0],
153 struct btrfs_file_extent_item);
154 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156 btrfs_set_file_extent_encryption(leaf, ei, 0);
157 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159 ptr = btrfs_file_extent_inline_start(ei);
164 while (compressed_size > 0) {
165 cpage = compressed_pages[i];
166 cur_size = min_t(unsigned long, compressed_size,
169 kaddr = kmap_atomic(cpage, KM_USER0);
170 write_extent_buffer(leaf, kaddr, ptr, cur_size);
171 kunmap_atomic(kaddr, KM_USER0);
175 compressed_size -= cur_size;
177 btrfs_set_file_extent_compression(leaf, ei,
178 BTRFS_COMPRESS_ZLIB);
180 page = find_get_page(inode->i_mapping,
181 start >> PAGE_CACHE_SHIFT);
182 btrfs_set_file_extent_compression(leaf, ei, 0);
183 kaddr = kmap_atomic(page, KM_USER0);
184 offset = start & (PAGE_CACHE_SIZE - 1);
185 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186 kunmap_atomic(kaddr, KM_USER0);
187 page_cache_release(page);
189 btrfs_mark_buffer_dirty(leaf);
190 btrfs_free_path(path);
192 BTRFS_I(inode)->disk_i_size = inode->i_size;
193 btrfs_update_inode(trans, root, inode);
196 btrfs_free_path(path);
202 * conditionally insert an inline extent into the file. This
203 * does the checks required to make sure the data is small enough
204 * to fit as an inline extent.
206 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
207 struct btrfs_root *root,
208 struct inode *inode, u64 start, u64 end,
209 size_t compressed_size,
210 struct page **compressed_pages)
212 u64 isize = i_size_read(inode);
213 u64 actual_end = min(end + 1, isize);
214 u64 inline_len = actual_end - start;
215 u64 aligned_end = (end + root->sectorsize - 1) &
216 ~((u64)root->sectorsize - 1);
218 u64 data_len = inline_len;
222 data_len = compressed_size;
225 actual_end >= PAGE_CACHE_SIZE ||
226 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
228 (actual_end & (root->sectorsize - 1)) == 0) ||
230 data_len > root->fs_info->max_inline) {
234 ret = btrfs_drop_extents(trans, root, inode, start,
235 aligned_end, aligned_end, start,
239 if (isize > actual_end)
240 inline_len = min_t(u64, isize, actual_end);
241 ret = insert_inline_extent(trans, root, inode, start,
242 inline_len, compressed_size,
245 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
249 struct async_extent {
254 unsigned long nr_pages;
255 struct list_head list;
260 struct btrfs_root *root;
261 struct page *locked_page;
264 struct list_head extents;
265 struct btrfs_work work;
268 static noinline int add_async_extent(struct async_cow *cow,
269 u64 start, u64 ram_size,
272 unsigned long nr_pages)
274 struct async_extent *async_extent;
276 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
277 async_extent->start = start;
278 async_extent->ram_size = ram_size;
279 async_extent->compressed_size = compressed_size;
280 async_extent->pages = pages;
281 async_extent->nr_pages = nr_pages;
282 list_add_tail(&async_extent->list, &cow->extents);
287 * we create compressed extents in two phases. The first
288 * phase compresses a range of pages that have already been
289 * locked (both pages and state bits are locked).
291 * This is done inside an ordered work queue, and the compression
292 * is spread across many cpus. The actual IO submission is step
293 * two, and the ordered work queue takes care of making sure that
294 * happens in the same order things were put onto the queue by
295 * writepages and friends.
297 * If this code finds it can't get good compression, it puts an
298 * entry onto the work queue to write the uncompressed bytes. This
299 * makes sure that both compressed inodes and uncompressed inodes
300 * are written in the same order that pdflush sent them down.
302 static noinline int compress_file_range(struct inode *inode,
303 struct page *locked_page,
305 struct async_cow *async_cow,
308 struct btrfs_root *root = BTRFS_I(inode)->root;
309 struct btrfs_trans_handle *trans;
313 u64 blocksize = root->sectorsize;
315 u64 isize = i_size_read(inode);
317 struct page **pages = NULL;
318 unsigned long nr_pages;
319 unsigned long nr_pages_ret = 0;
320 unsigned long total_compressed = 0;
321 unsigned long total_in = 0;
322 unsigned long max_compressed = 128 * 1024;
323 unsigned long max_uncompressed = 128 * 1024;
329 actual_end = min_t(u64, isize, end + 1);
332 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
333 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
336 * we don't want to send crud past the end of i_size through
337 * compression, that's just a waste of CPU time. So, if the
338 * end of the file is before the start of our current
339 * requested range of bytes, we bail out to the uncompressed
340 * cleanup code that can deal with all of this.
342 * It isn't really the fastest way to fix things, but this is a
343 * very uncommon corner.
345 if (actual_end <= start)
346 goto cleanup_and_bail_uncompressed;
348 total_compressed = actual_end - start;
350 /* we want to make sure that amount of ram required to uncompress
351 * an extent is reasonable, so we limit the total size in ram
352 * of a compressed extent to 128k. This is a crucial number
353 * because it also controls how easily we can spread reads across
354 * cpus for decompression.
356 * We also want to make sure the amount of IO required to do
357 * a random read is reasonably small, so we limit the size of
358 * a compressed extent to 128k.
360 total_compressed = min(total_compressed, max_uncompressed);
361 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
362 num_bytes = max(blocksize, num_bytes);
363 disk_num_bytes = num_bytes;
368 * we do compression for mount -o compress and when the
369 * inode has not been flagged as nocompress. This flag can
370 * change at any time if we discover bad compression ratios.
372 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
373 btrfs_test_opt(root, COMPRESS)) {
375 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
377 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
378 total_compressed, pages,
379 nr_pages, &nr_pages_ret,
385 unsigned long offset = total_compressed &
386 (PAGE_CACHE_SIZE - 1);
387 struct page *page = pages[nr_pages_ret - 1];
390 /* zero the tail end of the last page, we might be
391 * sending it down to disk
394 kaddr = kmap_atomic(page, KM_USER0);
395 memset(kaddr + offset, 0,
396 PAGE_CACHE_SIZE - offset);
397 kunmap_atomic(kaddr, KM_USER0);
403 trans = btrfs_join_transaction(root, 1);
405 btrfs_set_trans_block_group(trans, inode);
407 /* lets try to make an inline extent */
408 if (ret || total_in < (actual_end - start)) {
409 /* we didn't compress the entire range, try
410 * to make an uncompressed inline extent.
412 ret = cow_file_range_inline(trans, root, inode,
413 start, end, 0, NULL);
415 /* try making a compressed inline extent */
416 ret = cow_file_range_inline(trans, root, inode,
418 total_compressed, pages);
420 btrfs_end_transaction(trans, root);
423 * inline extent creation worked, we don't need
424 * to create any more async work items. Unlock
425 * and free up our temp pages.
427 extent_clear_unlock_delalloc(inode,
428 &BTRFS_I(inode)->io_tree,
429 start, end, NULL, 1, 0,
438 * we aren't doing an inline extent round the compressed size
439 * up to a block size boundary so the allocator does sane
442 total_compressed = (total_compressed + blocksize - 1) &
446 * one last check to make sure the compression is really a
447 * win, compare the page count read with the blocks on disk
449 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
450 ~(PAGE_CACHE_SIZE - 1);
451 if (total_compressed >= total_in) {
454 disk_num_bytes = total_compressed;
455 num_bytes = total_in;
458 if (!will_compress && pages) {
460 * the compression code ran but failed to make things smaller,
461 * free any pages it allocated and our page pointer array
463 for (i = 0; i < nr_pages_ret; i++) {
464 WARN_ON(pages[i]->mapping);
465 page_cache_release(pages[i]);
469 total_compressed = 0;
472 /* flag the file so we don't compress in the future */
473 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
478 /* the async work queues will take care of doing actual
479 * allocation on disk for these compressed pages,
480 * and will submit them to the elevator.
482 add_async_extent(async_cow, start, num_bytes,
483 total_compressed, pages, nr_pages_ret);
485 if (start + num_bytes < end && start + num_bytes < actual_end) {
492 cleanup_and_bail_uncompressed:
494 * No compression, but we still need to write the pages in
495 * the file we've been given so far. redirty the locked
496 * page if it corresponds to our extent and set things up
497 * for the async work queue to run cow_file_range to do
498 * the normal delalloc dance
500 if (page_offset(locked_page) >= start &&
501 page_offset(locked_page) <= end) {
502 __set_page_dirty_nobuffers(locked_page);
503 /* unlocked later on in the async handlers */
505 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
513 for (i = 0; i < nr_pages_ret; i++) {
514 WARN_ON(pages[i]->mapping);
515 page_cache_release(pages[i]);
523 * phase two of compressed writeback. This is the ordered portion
524 * of the code, which only gets called in the order the work was
525 * queued. We walk all the async extents created by compress_file_range
526 * and send them down to the disk.
528 static noinline int submit_compressed_extents(struct inode *inode,
529 struct async_cow *async_cow)
531 struct async_extent *async_extent;
533 struct btrfs_trans_handle *trans;
534 struct btrfs_key ins;
535 struct extent_map *em;
536 struct btrfs_root *root = BTRFS_I(inode)->root;
537 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538 struct extent_io_tree *io_tree;
541 if (list_empty(&async_cow->extents))
544 trans = btrfs_join_transaction(root, 1);
546 while (!list_empty(&async_cow->extents)) {
547 async_extent = list_entry(async_cow->extents.next,
548 struct async_extent, list);
549 list_del(&async_extent->list);
551 io_tree = &BTRFS_I(inode)->io_tree;
553 /* did the compression code fall back to uncompressed IO? */
554 if (!async_extent->pages) {
555 int page_started = 0;
556 unsigned long nr_written = 0;
558 lock_extent(io_tree, async_extent->start,
559 async_extent->start +
560 async_extent->ram_size - 1, GFP_NOFS);
562 /* allocate blocks */
563 cow_file_range(inode, async_cow->locked_page,
565 async_extent->start +
566 async_extent->ram_size - 1,
567 &page_started, &nr_written, 0);
570 * if page_started, cow_file_range inserted an
571 * inline extent and took care of all the unlocking
572 * and IO for us. Otherwise, we need to submit
573 * all those pages down to the drive.
576 extent_write_locked_range(io_tree,
577 inode, async_extent->start,
578 async_extent->start +
579 async_extent->ram_size - 1,
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start + async_extent->ram_size - 1,
591 * here we're doing allocation and writeback of the
594 btrfs_drop_extent_cache(inode, async_extent->start,
595 async_extent->start +
596 async_extent->ram_size - 1, 0);
598 ret = btrfs_reserve_extent(trans, root,
599 async_extent->compressed_size,
600 async_extent->compressed_size,
604 em = alloc_extent_map(GFP_NOFS);
605 em->start = async_extent->start;
606 em->len = async_extent->ram_size;
607 em->orig_start = em->start;
609 em->block_start = ins.objectid;
610 em->block_len = ins.offset;
611 em->bdev = root->fs_info->fs_devices->latest_bdev;
612 set_bit(EXTENT_FLAG_PINNED, &em->flags);
613 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
616 write_lock(&em_tree->lock);
617 ret = add_extent_mapping(em_tree, em);
618 write_unlock(&em_tree->lock);
619 if (ret != -EEXIST) {
623 btrfs_drop_extent_cache(inode, async_extent->start,
624 async_extent->start +
625 async_extent->ram_size - 1, 0);
628 ret = btrfs_add_ordered_extent(inode, async_extent->start,
630 async_extent->ram_size,
632 BTRFS_ORDERED_COMPRESSED);
635 btrfs_end_transaction(trans, root);
638 * clear dirty, set writeback and unlock the pages.
640 extent_clear_unlock_delalloc(inode,
641 &BTRFS_I(inode)->io_tree,
643 async_extent->start +
644 async_extent->ram_size - 1,
645 NULL, 1, 1, 0, 1, 1, 0, 0);
647 ret = btrfs_submit_compressed_write(inode,
649 async_extent->ram_size,
651 ins.offset, async_extent->pages,
652 async_extent->nr_pages);
655 trans = btrfs_join_transaction(root, 1);
656 alloc_hint = ins.objectid + ins.offset;
661 btrfs_end_transaction(trans, root);
666 * when extent_io.c finds a delayed allocation range in the file,
667 * the call backs end up in this code. The basic idea is to
668 * allocate extents on disk for the range, and create ordered data structs
669 * in ram to track those extents.
671 * locked_page is the page that writepage had locked already. We use
672 * it to make sure we don't do extra locks or unlocks.
674 * *page_started is set to one if we unlock locked_page and do everything
675 * required to start IO on it. It may be clean and already done with
678 static noinline int cow_file_range(struct inode *inode,
679 struct page *locked_page,
680 u64 start, u64 end, int *page_started,
681 unsigned long *nr_written,
684 struct btrfs_root *root = BTRFS_I(inode)->root;
685 struct btrfs_trans_handle *trans;
688 unsigned long ram_size;
691 u64 blocksize = root->sectorsize;
693 u64 isize = i_size_read(inode);
694 struct btrfs_key ins;
695 struct extent_map *em;
696 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
699 trans = btrfs_join_transaction(root, 1);
701 btrfs_set_trans_block_group(trans, inode);
703 actual_end = min_t(u64, isize, end + 1);
705 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
706 num_bytes = max(blocksize, num_bytes);
707 disk_num_bytes = num_bytes;
711 /* lets try to make an inline extent */
712 ret = cow_file_range_inline(trans, root, inode,
713 start, end, 0, NULL);
715 extent_clear_unlock_delalloc(inode,
716 &BTRFS_I(inode)->io_tree,
717 start, end, NULL, 1, 1,
719 *nr_written = *nr_written +
720 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
727 BUG_ON(disk_num_bytes >
728 btrfs_super_total_bytes(&root->fs_info->super_copy));
730 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
732 while (disk_num_bytes > 0) {
733 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
734 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
735 root->sectorsize, 0, alloc_hint,
739 em = alloc_extent_map(GFP_NOFS);
741 em->orig_start = em->start;
743 ram_size = ins.offset;
744 em->len = ins.offset;
746 em->block_start = ins.objectid;
747 em->block_len = ins.offset;
748 em->bdev = root->fs_info->fs_devices->latest_bdev;
749 set_bit(EXTENT_FLAG_PINNED, &em->flags);
752 write_lock(&em_tree->lock);
753 ret = add_extent_mapping(em_tree, em);
754 write_unlock(&em_tree->lock);
755 if (ret != -EEXIST) {
759 btrfs_drop_extent_cache(inode, start,
760 start + ram_size - 1, 0);
763 cur_alloc_size = ins.offset;
764 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
765 ram_size, cur_alloc_size, 0);
768 if (root->root_key.objectid ==
769 BTRFS_DATA_RELOC_TREE_OBJECTID) {
770 ret = btrfs_reloc_clone_csums(inode, start,
775 if (disk_num_bytes < cur_alloc_size)
778 /* we're not doing compressed IO, don't unlock the first
779 * page (which the caller expects to stay locked), don't
780 * clear any dirty bits and don't set any writeback bits
782 * Do set the Private2 bit so we know this page was properly
783 * setup for writepage
785 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
786 start, start + ram_size - 1,
787 locked_page, unlock, 1,
789 disk_num_bytes -= cur_alloc_size;
790 num_bytes -= cur_alloc_size;
791 alloc_hint = ins.objectid + ins.offset;
792 start += cur_alloc_size;
796 btrfs_end_transaction(trans, root);
802 * work queue call back to started compression on a file and pages
804 static noinline void async_cow_start(struct btrfs_work *work)
806 struct async_cow *async_cow;
808 async_cow = container_of(work, struct async_cow, work);
810 compress_file_range(async_cow->inode, async_cow->locked_page,
811 async_cow->start, async_cow->end, async_cow,
814 async_cow->inode = NULL;
818 * work queue call back to submit previously compressed pages
820 static noinline void async_cow_submit(struct btrfs_work *work)
822 struct async_cow *async_cow;
823 struct btrfs_root *root;
824 unsigned long nr_pages;
826 async_cow = container_of(work, struct async_cow, work);
828 root = async_cow->root;
829 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
832 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
834 if (atomic_read(&root->fs_info->async_delalloc_pages) <
836 waitqueue_active(&root->fs_info->async_submit_wait))
837 wake_up(&root->fs_info->async_submit_wait);
839 if (async_cow->inode)
840 submit_compressed_extents(async_cow->inode, async_cow);
843 static noinline void async_cow_free(struct btrfs_work *work)
845 struct async_cow *async_cow;
846 async_cow = container_of(work, struct async_cow, work);
850 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
851 u64 start, u64 end, int *page_started,
852 unsigned long *nr_written)
854 struct async_cow *async_cow;
855 struct btrfs_root *root = BTRFS_I(inode)->root;
856 unsigned long nr_pages;
858 int limit = 10 * 1024 * 1042;
860 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
861 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
862 while (start < end) {
863 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
864 async_cow->inode = inode;
865 async_cow->root = root;
866 async_cow->locked_page = locked_page;
867 async_cow->start = start;
869 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
872 cur_end = min(end, start + 512 * 1024 - 1);
874 async_cow->end = cur_end;
875 INIT_LIST_HEAD(&async_cow->extents);
877 async_cow->work.func = async_cow_start;
878 async_cow->work.ordered_func = async_cow_submit;
879 async_cow->work.ordered_free = async_cow_free;
880 async_cow->work.flags = 0;
882 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
884 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
886 btrfs_queue_worker(&root->fs_info->delalloc_workers,
889 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
890 wait_event(root->fs_info->async_submit_wait,
891 (atomic_read(&root->fs_info->async_delalloc_pages) <
895 while (atomic_read(&root->fs_info->async_submit_draining) &&
896 atomic_read(&root->fs_info->async_delalloc_pages)) {
897 wait_event(root->fs_info->async_submit_wait,
898 (atomic_read(&root->fs_info->async_delalloc_pages) ==
902 *nr_written += nr_pages;
909 static noinline int csum_exist_in_range(struct btrfs_root *root,
910 u64 bytenr, u64 num_bytes)
913 struct btrfs_ordered_sum *sums;
916 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
917 bytenr + num_bytes - 1, &list);
918 if (ret == 0 && list_empty(&list))
921 while (!list_empty(&list)) {
922 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
923 list_del(&sums->list);
930 * when nowcow writeback call back. This checks for snapshots or COW copies
931 * of the extents that exist in the file, and COWs the file as required.
933 * If no cow copies or snapshots exist, we write directly to the existing
936 static noinline int run_delalloc_nocow(struct inode *inode,
937 struct page *locked_page,
938 u64 start, u64 end, int *page_started, int force,
939 unsigned long *nr_written)
941 struct btrfs_root *root = BTRFS_I(inode)->root;
942 struct btrfs_trans_handle *trans;
943 struct extent_buffer *leaf;
944 struct btrfs_path *path;
945 struct btrfs_file_extent_item *fi;
946 struct btrfs_key found_key;
959 path = btrfs_alloc_path();
961 trans = btrfs_join_transaction(root, 1);
967 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
970 if (ret > 0 && path->slots[0] > 0 && check_prev) {
971 leaf = path->nodes[0];
972 btrfs_item_key_to_cpu(leaf, &found_key,
974 if (found_key.objectid == inode->i_ino &&
975 found_key.type == BTRFS_EXTENT_DATA_KEY)
980 leaf = path->nodes[0];
981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982 ret = btrfs_next_leaf(root, path);
987 leaf = path->nodes[0];
993 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
995 if (found_key.objectid > inode->i_ino ||
996 found_key.type > BTRFS_EXTENT_DATA_KEY ||
997 found_key.offset > end)
1000 if (found_key.offset > cur_offset) {
1001 extent_end = found_key.offset;
1005 fi = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_file_extent_item);
1007 extent_type = btrfs_file_extent_type(leaf, fi);
1009 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1010 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1011 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1012 extent_offset = btrfs_file_extent_offset(leaf, fi);
1013 extent_end = found_key.offset +
1014 btrfs_file_extent_num_bytes(leaf, fi);
1015 if (extent_end <= start) {
1019 if (disk_bytenr == 0)
1021 if (btrfs_file_extent_compression(leaf, fi) ||
1022 btrfs_file_extent_encryption(leaf, fi) ||
1023 btrfs_file_extent_other_encoding(leaf, fi))
1025 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1027 if (btrfs_extent_readonly(root, disk_bytenr))
1029 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1031 extent_offset, disk_bytenr))
1033 disk_bytenr += extent_offset;
1034 disk_bytenr += cur_offset - found_key.offset;
1035 num_bytes = min(end + 1, extent_end) - cur_offset;
1037 * force cow if csum exists in the range.
1038 * this ensure that csum for a given extent are
1039 * either valid or do not exist.
1041 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1044 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1045 extent_end = found_key.offset +
1046 btrfs_file_extent_inline_len(leaf, fi);
1047 extent_end = ALIGN(extent_end, root->sectorsize);
1052 if (extent_end <= start) {
1057 if (cow_start == (u64)-1)
1058 cow_start = cur_offset;
1059 cur_offset = extent_end;
1060 if (cur_offset > end)
1066 btrfs_release_path(root, path);
1067 if (cow_start != (u64)-1) {
1068 ret = cow_file_range(inode, locked_page, cow_start,
1069 found_key.offset - 1, page_started,
1072 cow_start = (u64)-1;
1075 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1076 struct extent_map *em;
1077 struct extent_map_tree *em_tree;
1078 em_tree = &BTRFS_I(inode)->extent_tree;
1079 em = alloc_extent_map(GFP_NOFS);
1080 em->start = cur_offset;
1081 em->orig_start = em->start;
1082 em->len = num_bytes;
1083 em->block_len = num_bytes;
1084 em->block_start = disk_bytenr;
1085 em->bdev = root->fs_info->fs_devices->latest_bdev;
1086 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1088 write_lock(&em_tree->lock);
1089 ret = add_extent_mapping(em_tree, em);
1090 write_unlock(&em_tree->lock);
1091 if (ret != -EEXIST) {
1092 free_extent_map(em);
1095 btrfs_drop_extent_cache(inode, em->start,
1096 em->start + em->len - 1, 0);
1098 type = BTRFS_ORDERED_PREALLOC;
1100 type = BTRFS_ORDERED_NOCOW;
1103 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1104 num_bytes, num_bytes, type);
1107 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1108 cur_offset, cur_offset + num_bytes - 1,
1109 locked_page, 1, 1, 1, 0, 0, 0, 1);
1110 cur_offset = extent_end;
1111 if (cur_offset > end)
1114 btrfs_release_path(root, path);
1116 if (cur_offset <= end && cow_start == (u64)-1)
1117 cow_start = cur_offset;
1118 if (cow_start != (u64)-1) {
1119 ret = cow_file_range(inode, locked_page, cow_start, end,
1120 page_started, nr_written, 1);
1124 ret = btrfs_end_transaction(trans, root);
1126 btrfs_free_path(path);
1131 * extent_io.c call back to do delayed allocation processing
1133 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1140 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1141 ret = run_delalloc_nocow(inode, locked_page, start, end,
1142 page_started, 1, nr_written);
1143 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1144 ret = run_delalloc_nocow(inode, locked_page, start, end,
1145 page_started, 0, nr_written);
1146 else if (!btrfs_test_opt(root, COMPRESS))
1147 ret = cow_file_range(inode, locked_page, start, end,
1148 page_started, nr_written, 1);
1150 ret = cow_file_range_async(inode, locked_page, start, end,
1151 page_started, nr_written);
1156 * extent_io.c set_bit_hook, used to track delayed allocation
1157 * bytes in this file, and to maintain the list of inodes that
1158 * have pending delalloc work to be done.
1160 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1161 unsigned long old, unsigned long bits)
1164 * set_bit and clear bit hooks normally require _irqsave/restore
1165 * but in this case, we are only testeing for the DELALLOC
1166 * bit, which is only set or cleared with irqs on
1168 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1169 struct btrfs_root *root = BTRFS_I(inode)->root;
1170 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1171 spin_lock(&root->fs_info->delalloc_lock);
1172 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1173 root->fs_info->delalloc_bytes += end - start + 1;
1174 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1175 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1176 &root->fs_info->delalloc_inodes);
1178 spin_unlock(&root->fs_info->delalloc_lock);
1184 * extent_io.c clear_bit_hook, see set_bit_hook for why
1186 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1187 unsigned long old, unsigned long bits)
1190 * set_bit and clear bit hooks normally require _irqsave/restore
1191 * but in this case, we are only testeing for the DELALLOC
1192 * bit, which is only set or cleared with irqs on
1194 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1195 struct btrfs_root *root = BTRFS_I(inode)->root;
1197 spin_lock(&root->fs_info->delalloc_lock);
1198 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1199 printk(KERN_INFO "btrfs warning: delalloc account "
1201 (unsigned long long)end - start + 1,
1202 (unsigned long long)
1203 root->fs_info->delalloc_bytes);
1204 btrfs_delalloc_free_space(root, inode, (u64)-1);
1205 root->fs_info->delalloc_bytes = 0;
1206 BTRFS_I(inode)->delalloc_bytes = 0;
1208 btrfs_delalloc_free_space(root, inode,
1210 root->fs_info->delalloc_bytes -= end - start + 1;
1211 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1213 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1214 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1215 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1217 spin_unlock(&root->fs_info->delalloc_lock);
1223 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1224 * we don't create bios that span stripes or chunks
1226 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1227 size_t size, struct bio *bio,
1228 unsigned long bio_flags)
1230 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1231 struct btrfs_mapping_tree *map_tree;
1232 u64 logical = (u64)bio->bi_sector << 9;
1237 if (bio_flags & EXTENT_BIO_COMPRESSED)
1240 length = bio->bi_size;
1241 map_tree = &root->fs_info->mapping_tree;
1242 map_length = length;
1243 ret = btrfs_map_block(map_tree, READ, logical,
1244 &map_length, NULL, 0);
1246 if (map_length < length + size)
1252 * in order to insert checksums into the metadata in large chunks,
1253 * we wait until bio submission time. All the pages in the bio are
1254 * checksummed and sums are attached onto the ordered extent record.
1256 * At IO completion time the cums attached on the ordered extent record
1257 * are inserted into the btree
1259 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1260 struct bio *bio, int mirror_num,
1261 unsigned long bio_flags)
1263 struct btrfs_root *root = BTRFS_I(inode)->root;
1266 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1272 * in order to insert checksums into the metadata in large chunks,
1273 * we wait until bio submission time. All the pages in the bio are
1274 * checksummed and sums are attached onto the ordered extent record.
1276 * At IO completion time the cums attached on the ordered extent record
1277 * are inserted into the btree
1279 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1280 int mirror_num, unsigned long bio_flags)
1282 struct btrfs_root *root = BTRFS_I(inode)->root;
1283 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1287 * extent_io.c submission hook. This does the right thing for csum calculation
1288 * on write, or reading the csums from the tree before a read
1290 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1291 int mirror_num, unsigned long bio_flags)
1293 struct btrfs_root *root = BTRFS_I(inode)->root;
1297 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1299 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1302 if (!(rw & (1 << BIO_RW))) {
1303 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1304 return btrfs_submit_compressed_read(inode, bio,
1305 mirror_num, bio_flags);
1306 } else if (!skip_sum)
1307 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1309 } else if (!skip_sum) {
1310 /* csum items have already been cloned */
1311 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1313 /* we're doing a write, do the async checksumming */
1314 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1315 inode, rw, bio, mirror_num,
1316 bio_flags, __btrfs_submit_bio_start,
1317 __btrfs_submit_bio_done);
1321 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1325 * given a list of ordered sums record them in the inode. This happens
1326 * at IO completion time based on sums calculated at bio submission time.
1328 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1329 struct inode *inode, u64 file_offset,
1330 struct list_head *list)
1332 struct btrfs_ordered_sum *sum;
1334 btrfs_set_trans_block_group(trans, inode);
1336 list_for_each_entry(sum, list, list) {
1337 btrfs_csum_file_blocks(trans,
1338 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1343 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1345 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1347 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1351 /* see btrfs_writepage_start_hook for details on why this is required */
1352 struct btrfs_writepage_fixup {
1354 struct btrfs_work work;
1357 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1359 struct btrfs_writepage_fixup *fixup;
1360 struct btrfs_ordered_extent *ordered;
1362 struct inode *inode;
1366 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1370 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1371 ClearPageChecked(page);
1375 inode = page->mapping->host;
1376 page_start = page_offset(page);
1377 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1379 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1381 /* already ordered? We're done */
1382 if (PagePrivate2(page))
1385 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1387 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1388 page_end, GFP_NOFS);
1390 btrfs_start_ordered_extent(inode, ordered, 1);
1394 btrfs_set_extent_delalloc(inode, page_start, page_end);
1395 ClearPageChecked(page);
1397 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1400 page_cache_release(page);
1404 * There are a few paths in the higher layers of the kernel that directly
1405 * set the page dirty bit without asking the filesystem if it is a
1406 * good idea. This causes problems because we want to make sure COW
1407 * properly happens and the data=ordered rules are followed.
1409 * In our case any range that doesn't have the ORDERED bit set
1410 * hasn't been properly setup for IO. We kick off an async process
1411 * to fix it up. The async helper will wait for ordered extents, set
1412 * the delalloc bit and make it safe to write the page.
1414 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1416 struct inode *inode = page->mapping->host;
1417 struct btrfs_writepage_fixup *fixup;
1418 struct btrfs_root *root = BTRFS_I(inode)->root;
1420 /* this page is properly in the ordered list */
1421 if (TestClearPagePrivate2(page))
1424 if (PageChecked(page))
1427 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1431 SetPageChecked(page);
1432 page_cache_get(page);
1433 fixup->work.func = btrfs_writepage_fixup_worker;
1435 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440 struct inode *inode, u64 file_pos,
1441 u64 disk_bytenr, u64 disk_num_bytes,
1442 u64 num_bytes, u64 ram_bytes,
1444 u8 compression, u8 encryption,
1445 u16 other_encoding, int extent_type)
1447 struct btrfs_root *root = BTRFS_I(inode)->root;
1448 struct btrfs_file_extent_item *fi;
1449 struct btrfs_path *path;
1450 struct extent_buffer *leaf;
1451 struct btrfs_key ins;
1455 path = btrfs_alloc_path();
1458 path->leave_spinning = 1;
1461 * we may be replacing one extent in the tree with another.
1462 * The new extent is pinned in the extent map, and we don't want
1463 * to drop it from the cache until it is completely in the btree.
1465 * So, tell btrfs_drop_extents to leave this extent in the cache.
1466 * the caller is expected to unpin it and allow it to be merged
1469 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1470 file_pos + num_bytes, locked_end,
1471 file_pos, &hint, 0);
1474 ins.objectid = inode->i_ino;
1475 ins.offset = file_pos;
1476 ins.type = BTRFS_EXTENT_DATA_KEY;
1477 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1479 leaf = path->nodes[0];
1480 fi = btrfs_item_ptr(leaf, path->slots[0],
1481 struct btrfs_file_extent_item);
1482 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1483 btrfs_set_file_extent_type(leaf, fi, extent_type);
1484 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1485 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1486 btrfs_set_file_extent_offset(leaf, fi, 0);
1487 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1488 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1489 btrfs_set_file_extent_compression(leaf, fi, compression);
1490 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1491 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1493 btrfs_unlock_up_safe(path, 1);
1494 btrfs_set_lock_blocking(leaf);
1496 btrfs_mark_buffer_dirty(leaf);
1498 inode_add_bytes(inode, num_bytes);
1500 ins.objectid = disk_bytenr;
1501 ins.offset = disk_num_bytes;
1502 ins.type = BTRFS_EXTENT_ITEM_KEY;
1503 ret = btrfs_alloc_reserved_file_extent(trans, root,
1504 root->root_key.objectid,
1505 inode->i_ino, file_pos, &ins);
1507 btrfs_free_path(path);
1513 * helper function for btrfs_finish_ordered_io, this
1514 * just reads in some of the csum leaves to prime them into ram
1515 * before we start the transaction. It limits the amount of btree
1516 * reads required while inside the transaction.
1518 static noinline void reada_csum(struct btrfs_root *root,
1519 struct btrfs_path *path,
1520 struct btrfs_ordered_extent *ordered_extent)
1522 struct btrfs_ordered_sum *sum;
1525 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1527 bytenr = sum->sums[0].bytenr;
1530 * we don't care about the results, the point of this search is
1531 * just to get the btree leaves into ram
1533 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1536 /* as ordered data IO finishes, this gets called so we can finish
1537 * an ordered extent if the range of bytes in the file it covers are
1540 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1542 struct btrfs_root *root = BTRFS_I(inode)->root;
1543 struct btrfs_trans_handle *trans;
1544 struct btrfs_ordered_extent *ordered_extent = NULL;
1545 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1546 struct btrfs_path *path;
1550 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1555 * before we join the transaction, try to do some of our IO.
1556 * This will limit the amount of IO that we have to do with
1557 * the transaction running. We're unlikely to need to do any
1558 * IO if the file extents are new, the disk_i_size checks
1559 * covers the most common case.
1561 if (start < BTRFS_I(inode)->disk_i_size) {
1562 path = btrfs_alloc_path();
1564 ret = btrfs_lookup_file_extent(NULL, root, path,
1567 ordered_extent = btrfs_lookup_ordered_extent(inode,
1569 if (!list_empty(&ordered_extent->list)) {
1570 btrfs_release_path(root, path);
1571 reada_csum(root, path, ordered_extent);
1573 btrfs_free_path(path);
1577 trans = btrfs_join_transaction(root, 1);
1579 if (!ordered_extent)
1580 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1581 BUG_ON(!ordered_extent);
1582 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1585 lock_extent(io_tree, ordered_extent->file_offset,
1586 ordered_extent->file_offset + ordered_extent->len - 1,
1589 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1591 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1593 ret = btrfs_mark_extent_written(trans, root, inode,
1594 ordered_extent->file_offset,
1595 ordered_extent->file_offset +
1596 ordered_extent->len);
1599 ret = insert_reserved_file_extent(trans, inode,
1600 ordered_extent->file_offset,
1601 ordered_extent->start,
1602 ordered_extent->disk_len,
1603 ordered_extent->len,
1604 ordered_extent->len,
1605 ordered_extent->file_offset +
1606 ordered_extent->len,
1608 BTRFS_FILE_EXTENT_REG);
1609 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1610 ordered_extent->file_offset,
1611 ordered_extent->len);
1614 unlock_extent(io_tree, ordered_extent->file_offset,
1615 ordered_extent->file_offset + ordered_extent->len - 1,
1618 add_pending_csums(trans, inode, ordered_extent->file_offset,
1619 &ordered_extent->list);
1621 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1622 btrfs_ordered_update_i_size(inode, ordered_extent);
1623 btrfs_update_inode(trans, root, inode);
1624 btrfs_remove_ordered_extent(inode, ordered_extent);
1625 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1628 btrfs_put_ordered_extent(ordered_extent);
1629 /* once for the tree */
1630 btrfs_put_ordered_extent(ordered_extent);
1632 btrfs_end_transaction(trans, root);
1636 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1637 struct extent_state *state, int uptodate)
1639 ClearPagePrivate2(page);
1640 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1644 * When IO fails, either with EIO or csum verification fails, we
1645 * try other mirrors that might have a good copy of the data. This
1646 * io_failure_record is used to record state as we go through all the
1647 * mirrors. If another mirror has good data, the page is set up to date
1648 * and things continue. If a good mirror can't be found, the original
1649 * bio end_io callback is called to indicate things have failed.
1651 struct io_failure_record {
1656 unsigned long bio_flags;
1660 static int btrfs_io_failed_hook(struct bio *failed_bio,
1661 struct page *page, u64 start, u64 end,
1662 struct extent_state *state)
1664 struct io_failure_record *failrec = NULL;
1666 struct extent_map *em;
1667 struct inode *inode = page->mapping->host;
1668 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1669 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1676 ret = get_state_private(failure_tree, start, &private);
1678 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1681 failrec->start = start;
1682 failrec->len = end - start + 1;
1683 failrec->last_mirror = 0;
1684 failrec->bio_flags = 0;
1686 read_lock(&em_tree->lock);
1687 em = lookup_extent_mapping(em_tree, start, failrec->len);
1688 if (em->start > start || em->start + em->len < start) {
1689 free_extent_map(em);
1692 read_unlock(&em_tree->lock);
1694 if (!em || IS_ERR(em)) {
1698 logical = start - em->start;
1699 logical = em->block_start + logical;
1700 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1701 logical = em->block_start;
1702 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1704 failrec->logical = logical;
1705 free_extent_map(em);
1706 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1707 EXTENT_DIRTY, GFP_NOFS);
1708 set_state_private(failure_tree, start,
1709 (u64)(unsigned long)failrec);
1711 failrec = (struct io_failure_record *)(unsigned long)private;
1713 num_copies = btrfs_num_copies(
1714 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1715 failrec->logical, failrec->len);
1716 failrec->last_mirror++;
1718 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1719 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1722 if (state && state->start != failrec->start)
1724 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1726 if (!state || failrec->last_mirror > num_copies) {
1727 set_state_private(failure_tree, failrec->start, 0);
1728 clear_extent_bits(failure_tree, failrec->start,
1729 failrec->start + failrec->len - 1,
1730 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1734 bio = bio_alloc(GFP_NOFS, 1);
1735 bio->bi_private = state;
1736 bio->bi_end_io = failed_bio->bi_end_io;
1737 bio->bi_sector = failrec->logical >> 9;
1738 bio->bi_bdev = failed_bio->bi_bdev;
1741 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1742 if (failed_bio->bi_rw & (1 << BIO_RW))
1747 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1748 failrec->last_mirror,
1749 failrec->bio_flags);
1754 * each time an IO finishes, we do a fast check in the IO failure tree
1755 * to see if we need to process or clean up an io_failure_record
1757 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1760 u64 private_failure;
1761 struct io_failure_record *failure;
1765 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1766 (u64)-1, 1, EXTENT_DIRTY)) {
1767 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1768 start, &private_failure);
1770 failure = (struct io_failure_record *)(unsigned long)
1772 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1774 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1776 failure->start + failure->len - 1,
1777 EXTENT_DIRTY | EXTENT_LOCKED,
1786 * when reads are done, we need to check csums to verify the data is correct
1787 * if there's a match, we allow the bio to finish. If not, we go through
1788 * the io_failure_record routines to find good copies
1790 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1791 struct extent_state *state)
1793 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1794 struct inode *inode = page->mapping->host;
1795 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1797 u64 private = ~(u32)0;
1799 struct btrfs_root *root = BTRFS_I(inode)->root;
1802 if (PageChecked(page)) {
1803 ClearPageChecked(page);
1807 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1810 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1811 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1812 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1817 if (state && state->start == start) {
1818 private = state->private;
1821 ret = get_state_private(io_tree, start, &private);
1823 kaddr = kmap_atomic(page, KM_USER0);
1827 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1828 btrfs_csum_final(csum, (char *)&csum);
1829 if (csum != private)
1832 kunmap_atomic(kaddr, KM_USER0);
1834 /* if the io failure tree for this inode is non-empty,
1835 * check to see if we've recovered from a failed IO
1837 btrfs_clean_io_failures(inode, start);
1841 if (printk_ratelimit()) {
1842 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1843 "private %llu\n", page->mapping->host->i_ino,
1844 (unsigned long long)start, csum,
1845 (unsigned long long)private);
1847 memset(kaddr + offset, 1, end - start + 1);
1848 flush_dcache_page(page);
1849 kunmap_atomic(kaddr, KM_USER0);
1856 * This creates an orphan entry for the given inode in case something goes
1857 * wrong in the middle of an unlink/truncate.
1859 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1864 spin_lock(&root->list_lock);
1866 /* already on the orphan list, we're good */
1867 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1868 spin_unlock(&root->list_lock);
1872 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1874 spin_unlock(&root->list_lock);
1877 * insert an orphan item to track this unlinked/truncated file
1879 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1885 * We have done the truncate/delete so we can go ahead and remove the orphan
1886 * item for this particular inode.
1888 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1890 struct btrfs_root *root = BTRFS_I(inode)->root;
1893 spin_lock(&root->list_lock);
1895 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1896 spin_unlock(&root->list_lock);
1900 list_del_init(&BTRFS_I(inode)->i_orphan);
1902 spin_unlock(&root->list_lock);
1906 spin_unlock(&root->list_lock);
1908 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1914 * this cleans up any orphans that may be left on the list from the last use
1917 void btrfs_orphan_cleanup(struct btrfs_root *root)
1919 struct btrfs_path *path;
1920 struct extent_buffer *leaf;
1921 struct btrfs_item *item;
1922 struct btrfs_key key, found_key;
1923 struct btrfs_trans_handle *trans;
1924 struct inode *inode;
1925 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1927 path = btrfs_alloc_path();
1932 key.objectid = BTRFS_ORPHAN_OBJECTID;
1933 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1934 key.offset = (u64)-1;
1938 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1940 printk(KERN_ERR "Error searching slot for orphan: %d"
1946 * if ret == 0 means we found what we were searching for, which
1947 * is weird, but possible, so only screw with path if we didnt
1948 * find the key and see if we have stuff that matches
1951 if (path->slots[0] == 0)
1956 /* pull out the item */
1957 leaf = path->nodes[0];
1958 item = btrfs_item_nr(leaf, path->slots[0]);
1959 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1961 /* make sure the item matches what we want */
1962 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1964 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1967 /* release the path since we're done with it */
1968 btrfs_release_path(root, path);
1971 * this is where we are basically btrfs_lookup, without the
1972 * crossing root thing. we store the inode number in the
1973 * offset of the orphan item.
1975 found_key.objectid = found_key.offset;
1976 found_key.type = BTRFS_INODE_ITEM_KEY;
1977 found_key.offset = 0;
1978 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1983 * add this inode to the orphan list so btrfs_orphan_del does
1984 * the proper thing when we hit it
1986 spin_lock(&root->list_lock);
1987 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1988 spin_unlock(&root->list_lock);
1991 * if this is a bad inode, means we actually succeeded in
1992 * removing the inode, but not the orphan record, which means
1993 * we need to manually delete the orphan since iput will just
1994 * do a destroy_inode
1996 if (is_bad_inode(inode)) {
1997 trans = btrfs_start_transaction(root, 1);
1998 btrfs_orphan_del(trans, inode);
1999 btrfs_end_transaction(trans, root);
2004 /* if we have links, this was a truncate, lets do that */
2005 if (inode->i_nlink) {
2007 btrfs_truncate(inode);
2012 /* this will do delete_inode and everything for us */
2017 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2019 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2021 btrfs_free_path(path);
2025 * very simple check to peek ahead in the leaf looking for xattrs. If we
2026 * don't find any xattrs, we know there can't be any acls.
2028 * slot is the slot the inode is in, objectid is the objectid of the inode
2030 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2031 int slot, u64 objectid)
2033 u32 nritems = btrfs_header_nritems(leaf);
2034 struct btrfs_key found_key;
2038 while (slot < nritems) {
2039 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2041 /* we found a different objectid, there must not be acls */
2042 if (found_key.objectid != objectid)
2045 /* we found an xattr, assume we've got an acl */
2046 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2050 * we found a key greater than an xattr key, there can't
2051 * be any acls later on
2053 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2060 * it goes inode, inode backrefs, xattrs, extents,
2061 * so if there are a ton of hard links to an inode there can
2062 * be a lot of backrefs. Don't waste time searching too hard,
2063 * this is just an optimization
2068 /* we hit the end of the leaf before we found an xattr or
2069 * something larger than an xattr. We have to assume the inode
2076 * read an inode from the btree into the in-memory inode
2078 static void btrfs_read_locked_inode(struct inode *inode)
2080 struct btrfs_path *path;
2081 struct extent_buffer *leaf;
2082 struct btrfs_inode_item *inode_item;
2083 struct btrfs_timespec *tspec;
2084 struct btrfs_root *root = BTRFS_I(inode)->root;
2085 struct btrfs_key location;
2087 u64 alloc_group_block;
2091 path = btrfs_alloc_path();
2093 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2095 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2099 leaf = path->nodes[0];
2100 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2101 struct btrfs_inode_item);
2103 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2104 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2105 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2106 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2107 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2109 tspec = btrfs_inode_atime(inode_item);
2110 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2111 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2113 tspec = btrfs_inode_mtime(inode_item);
2114 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2115 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2117 tspec = btrfs_inode_ctime(inode_item);
2118 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2119 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2121 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2122 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2123 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2124 inode->i_generation = BTRFS_I(inode)->generation;
2126 rdev = btrfs_inode_rdev(leaf, inode_item);
2128 BTRFS_I(inode)->index_cnt = (u64)-1;
2129 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2131 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2134 * try to precache a NULL acl entry for files that don't have
2135 * any xattrs or acls
2137 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2139 BTRFS_I(inode)->i_acl = NULL;
2140 BTRFS_I(inode)->i_default_acl = NULL;
2143 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2144 alloc_group_block, 0);
2145 btrfs_free_path(path);
2148 switch (inode->i_mode & S_IFMT) {
2150 inode->i_mapping->a_ops = &btrfs_aops;
2151 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2152 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2153 inode->i_fop = &btrfs_file_operations;
2154 inode->i_op = &btrfs_file_inode_operations;
2157 inode->i_fop = &btrfs_dir_file_operations;
2158 if (root == root->fs_info->tree_root)
2159 inode->i_op = &btrfs_dir_ro_inode_operations;
2161 inode->i_op = &btrfs_dir_inode_operations;
2164 inode->i_op = &btrfs_symlink_inode_operations;
2165 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2166 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2169 inode->i_op = &btrfs_special_inode_operations;
2170 init_special_inode(inode, inode->i_mode, rdev);
2174 btrfs_update_iflags(inode);
2178 btrfs_free_path(path);
2179 make_bad_inode(inode);
2183 * given a leaf and an inode, copy the inode fields into the leaf
2185 static void fill_inode_item(struct btrfs_trans_handle *trans,
2186 struct extent_buffer *leaf,
2187 struct btrfs_inode_item *item,
2188 struct inode *inode)
2190 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2191 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2192 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2193 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2194 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2196 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2197 inode->i_atime.tv_sec);
2198 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2199 inode->i_atime.tv_nsec);
2201 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2202 inode->i_mtime.tv_sec);
2203 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2204 inode->i_mtime.tv_nsec);
2206 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2207 inode->i_ctime.tv_sec);
2208 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2209 inode->i_ctime.tv_nsec);
2211 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2212 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2213 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2214 btrfs_set_inode_transid(leaf, item, trans->transid);
2215 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2216 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2217 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2221 * copy everything in the in-memory inode into the btree.
2223 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2224 struct btrfs_root *root, struct inode *inode)
2226 struct btrfs_inode_item *inode_item;
2227 struct btrfs_path *path;
2228 struct extent_buffer *leaf;
2231 path = btrfs_alloc_path();
2233 path->leave_spinning = 1;
2234 ret = btrfs_lookup_inode(trans, root, path,
2235 &BTRFS_I(inode)->location, 1);
2242 btrfs_unlock_up_safe(path, 1);
2243 leaf = path->nodes[0];
2244 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2245 struct btrfs_inode_item);
2247 fill_inode_item(trans, leaf, inode_item, inode);
2248 btrfs_mark_buffer_dirty(leaf);
2249 btrfs_set_inode_last_trans(trans, inode);
2252 btrfs_free_path(path);
2258 * unlink helper that gets used here in inode.c and in the tree logging
2259 * recovery code. It remove a link in a directory with a given name, and
2260 * also drops the back refs in the inode to the directory
2262 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2263 struct btrfs_root *root,
2264 struct inode *dir, struct inode *inode,
2265 const char *name, int name_len)
2267 struct btrfs_path *path;
2269 struct extent_buffer *leaf;
2270 struct btrfs_dir_item *di;
2271 struct btrfs_key key;
2274 path = btrfs_alloc_path();
2280 path->leave_spinning = 1;
2281 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2282 name, name_len, -1);
2291 leaf = path->nodes[0];
2292 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2293 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2296 btrfs_release_path(root, path);
2298 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2300 dir->i_ino, &index);
2302 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2303 "inode %lu parent %lu\n", name_len, name,
2304 inode->i_ino, dir->i_ino);
2308 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2309 index, name, name_len, -1);
2318 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2319 btrfs_release_path(root, path);
2321 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2323 BUG_ON(ret != 0 && ret != -ENOENT);
2325 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2329 btrfs_free_path(path);
2333 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2334 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2335 btrfs_update_inode(trans, root, dir);
2336 btrfs_drop_nlink(inode);
2337 ret = btrfs_update_inode(trans, root, inode);
2338 dir->i_sb->s_dirt = 1;
2343 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2345 struct btrfs_root *root;
2346 struct btrfs_trans_handle *trans;
2347 struct inode *inode = dentry->d_inode;
2349 unsigned long nr = 0;
2351 root = BTRFS_I(dir)->root;
2353 trans = btrfs_start_transaction(root, 1);
2355 btrfs_set_trans_block_group(trans, dir);
2357 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2359 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2360 dentry->d_name.name, dentry->d_name.len);
2362 if (inode->i_nlink == 0)
2363 ret = btrfs_orphan_add(trans, inode);
2365 nr = trans->blocks_used;
2367 btrfs_end_transaction_throttle(trans, root);
2368 btrfs_btree_balance_dirty(root, nr);
2372 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2374 struct inode *inode = dentry->d_inode;
2377 struct btrfs_root *root = BTRFS_I(dir)->root;
2378 struct btrfs_trans_handle *trans;
2379 unsigned long nr = 0;
2382 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2383 * the root of a subvolume or snapshot
2385 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2386 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2390 trans = btrfs_start_transaction(root, 1);
2391 btrfs_set_trans_block_group(trans, dir);
2393 err = btrfs_orphan_add(trans, inode);
2397 /* now the directory is empty */
2398 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2399 dentry->d_name.name, dentry->d_name.len);
2401 btrfs_i_size_write(inode, 0);
2404 nr = trans->blocks_used;
2405 ret = btrfs_end_transaction_throttle(trans, root);
2406 btrfs_btree_balance_dirty(root, nr);
2415 * when truncating bytes in a file, it is possible to avoid reading
2416 * the leaves that contain only checksum items. This can be the
2417 * majority of the IO required to delete a large file, but it must
2418 * be done carefully.
2420 * The keys in the level just above the leaves are checked to make sure
2421 * the lowest key in a given leaf is a csum key, and starts at an offset
2422 * after the new size.
2424 * Then the key for the next leaf is checked to make sure it also has
2425 * a checksum item for the same file. If it does, we know our target leaf
2426 * contains only checksum items, and it can be safely freed without reading
2429 * This is just an optimization targeted at large files. It may do
2430 * nothing. It will return 0 unless things went badly.
2432 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2433 struct btrfs_root *root,
2434 struct btrfs_path *path,
2435 struct inode *inode, u64 new_size)
2437 struct btrfs_key key;
2440 struct btrfs_key found_key;
2441 struct btrfs_key other_key;
2442 struct btrfs_leaf_ref *ref;
2446 path->lowest_level = 1;
2447 key.objectid = inode->i_ino;
2448 key.type = BTRFS_CSUM_ITEM_KEY;
2449 key.offset = new_size;
2451 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2455 if (path->nodes[1] == NULL) {
2460 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2461 nritems = btrfs_header_nritems(path->nodes[1]);
2466 if (path->slots[1] >= nritems)
2469 /* did we find a key greater than anything we want to delete? */
2470 if (found_key.objectid > inode->i_ino ||
2471 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2474 /* we check the next key in the node to make sure the leave contains
2475 * only checksum items. This comparison doesn't work if our
2476 * leaf is the last one in the node
2478 if (path->slots[1] + 1 >= nritems) {
2480 /* search forward from the last key in the node, this
2481 * will bring us into the next node in the tree
2483 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2485 /* unlikely, but we inc below, so check to be safe */
2486 if (found_key.offset == (u64)-1)
2489 /* search_forward needs a path with locks held, do the
2490 * search again for the original key. It is possible
2491 * this will race with a balance and return a path that
2492 * we could modify, but this drop is just an optimization
2493 * and is allowed to miss some leaves.
2495 btrfs_release_path(root, path);
2498 /* setup a max key for search_forward */
2499 other_key.offset = (u64)-1;
2500 other_key.type = key.type;
2501 other_key.objectid = key.objectid;
2503 path->keep_locks = 1;
2504 ret = btrfs_search_forward(root, &found_key, &other_key,
2506 path->keep_locks = 0;
2507 if (ret || found_key.objectid != key.objectid ||
2508 found_key.type != key.type) {
2513 key.offset = found_key.offset;
2514 btrfs_release_path(root, path);
2519 /* we know there's one more slot after us in the tree,
2520 * read that key so we can verify it is also a checksum item
2522 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2524 if (found_key.objectid < inode->i_ino)
2527 if (found_key.type != key.type || found_key.offset < new_size)
2531 * if the key for the next leaf isn't a csum key from this objectid,
2532 * we can't be sure there aren't good items inside this leaf.
2535 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2538 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2539 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2541 * it is safe to delete this leaf, it contains only
2542 * csum items from this inode at an offset >= new_size
2544 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2547 if (root->ref_cows && leaf_gen < trans->transid) {
2548 ref = btrfs_alloc_leaf_ref(root, 0);
2550 ref->root_gen = root->root_key.offset;
2551 ref->bytenr = leaf_start;
2553 ref->generation = leaf_gen;
2556 btrfs_sort_leaf_ref(ref);
2558 ret = btrfs_add_leaf_ref(root, ref, 0);
2560 btrfs_free_leaf_ref(root, ref);
2566 btrfs_release_path(root, path);
2568 if (other_key.objectid == inode->i_ino &&
2569 other_key.type == key.type && other_key.offset > key.offset) {
2570 key.offset = other_key.offset;
2576 /* fixup any changes we've made to the path */
2577 path->lowest_level = 0;
2578 path->keep_locks = 0;
2579 btrfs_release_path(root, path);
2586 * this can truncate away extent items, csum items and directory items.
2587 * It starts at a high offset and removes keys until it can't find
2588 * any higher than new_size
2590 * csum items that cross the new i_size are truncated to the new size
2593 * min_type is the minimum key type to truncate down to. If set to 0, this
2594 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2596 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2597 struct btrfs_root *root,
2598 struct inode *inode,
2599 u64 new_size, u32 min_type)
2602 struct btrfs_path *path;
2603 struct btrfs_key key;
2604 struct btrfs_key found_key;
2605 u32 found_type = (u8)-1;
2606 struct extent_buffer *leaf;
2607 struct btrfs_file_extent_item *fi;
2608 u64 extent_start = 0;
2609 u64 extent_num_bytes = 0;
2610 u64 extent_offset = 0;
2614 int pending_del_nr = 0;
2615 int pending_del_slot = 0;
2616 int extent_type = -1;
2618 u64 mask = root->sectorsize - 1;
2621 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2622 path = btrfs_alloc_path();
2626 /* FIXME, add redo link to tree so we don't leak on crash */
2627 key.objectid = inode->i_ino;
2628 key.offset = (u64)-1;
2632 path->leave_spinning = 1;
2633 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2638 /* there are no items in the tree for us to truncate, we're
2641 if (path->slots[0] == 0) {
2650 leaf = path->nodes[0];
2651 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2652 found_type = btrfs_key_type(&found_key);
2655 if (found_key.objectid != inode->i_ino)
2658 if (found_type < min_type)
2661 item_end = found_key.offset;
2662 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2663 fi = btrfs_item_ptr(leaf, path->slots[0],
2664 struct btrfs_file_extent_item);
2665 extent_type = btrfs_file_extent_type(leaf, fi);
2666 encoding = btrfs_file_extent_compression(leaf, fi);
2667 encoding |= btrfs_file_extent_encryption(leaf, fi);
2668 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2670 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2672 btrfs_file_extent_num_bytes(leaf, fi);
2673 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2674 item_end += btrfs_file_extent_inline_len(leaf,
2679 if (item_end < new_size) {
2680 if (found_type == BTRFS_DIR_ITEM_KEY)
2681 found_type = BTRFS_INODE_ITEM_KEY;
2682 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2683 found_type = BTRFS_EXTENT_DATA_KEY;
2684 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2685 found_type = BTRFS_XATTR_ITEM_KEY;
2686 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2687 found_type = BTRFS_INODE_REF_KEY;
2688 else if (found_type)
2692 btrfs_set_key_type(&key, found_type);
2695 if (found_key.offset >= new_size)
2701 /* FIXME, shrink the extent if the ref count is only 1 */
2702 if (found_type != BTRFS_EXTENT_DATA_KEY)
2705 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2707 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2708 if (!del_item && !encoding) {
2709 u64 orig_num_bytes =
2710 btrfs_file_extent_num_bytes(leaf, fi);
2711 extent_num_bytes = new_size -
2712 found_key.offset + root->sectorsize - 1;
2713 extent_num_bytes = extent_num_bytes &
2714 ~((u64)root->sectorsize - 1);
2715 btrfs_set_file_extent_num_bytes(leaf, fi,
2717 num_dec = (orig_num_bytes -
2719 if (root->ref_cows && extent_start != 0)
2720 inode_sub_bytes(inode, num_dec);
2721 btrfs_mark_buffer_dirty(leaf);
2724 btrfs_file_extent_disk_num_bytes(leaf,
2726 extent_offset = found_key.offset -
2727 btrfs_file_extent_offset(leaf, fi);
2729 /* FIXME blocksize != 4096 */
2730 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2731 if (extent_start != 0) {
2734 inode_sub_bytes(inode, num_dec);
2737 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2739 * we can't truncate inline items that have had
2743 btrfs_file_extent_compression(leaf, fi) == 0 &&
2744 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2745 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2746 u32 size = new_size - found_key.offset;
2748 if (root->ref_cows) {
2749 inode_sub_bytes(inode, item_end + 1 -
2753 btrfs_file_extent_calc_inline_size(size);
2754 ret = btrfs_truncate_item(trans, root, path,
2757 } else if (root->ref_cows) {
2758 inode_sub_bytes(inode, item_end + 1 -
2764 if (!pending_del_nr) {
2765 /* no pending yet, add ourselves */
2766 pending_del_slot = path->slots[0];
2768 } else if (pending_del_nr &&
2769 path->slots[0] + 1 == pending_del_slot) {
2770 /* hop on the pending chunk */
2772 pending_del_slot = path->slots[0];
2779 if (found_extent && root->ref_cows) {
2780 btrfs_set_path_blocking(path);
2781 ret = btrfs_free_extent(trans, root, extent_start,
2782 extent_num_bytes, 0,
2783 btrfs_header_owner(leaf),
2784 inode->i_ino, extent_offset);
2788 if (path->slots[0] == 0) {
2791 btrfs_release_path(root, path);
2792 if (found_type == BTRFS_INODE_ITEM_KEY)
2798 if (pending_del_nr &&
2799 path->slots[0] + 1 != pending_del_slot) {
2800 struct btrfs_key debug;
2802 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2804 ret = btrfs_del_items(trans, root, path,
2809 btrfs_release_path(root, path);
2810 if (found_type == BTRFS_INODE_ITEM_KEY)
2817 if (pending_del_nr) {
2818 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2821 btrfs_free_path(path);
2822 inode->i_sb->s_dirt = 1;
2827 * taken from block_truncate_page, but does cow as it zeros out
2828 * any bytes left in the last page in the file.
2830 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2832 struct inode *inode = mapping->host;
2833 struct btrfs_root *root = BTRFS_I(inode)->root;
2834 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2835 struct btrfs_ordered_extent *ordered;
2837 u32 blocksize = root->sectorsize;
2838 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2839 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2845 if ((offset & (blocksize - 1)) == 0)
2850 page = grab_cache_page(mapping, index);
2854 page_start = page_offset(page);
2855 page_end = page_start + PAGE_CACHE_SIZE - 1;
2857 if (!PageUptodate(page)) {
2858 ret = btrfs_readpage(NULL, page);
2860 if (page->mapping != mapping) {
2862 page_cache_release(page);
2865 if (!PageUptodate(page)) {
2870 wait_on_page_writeback(page);
2872 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2873 set_page_extent_mapped(page);
2875 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2877 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2879 page_cache_release(page);
2880 btrfs_start_ordered_extent(inode, ordered, 1);
2881 btrfs_put_ordered_extent(ordered);
2885 btrfs_set_extent_delalloc(inode, page_start, page_end);
2887 if (offset != PAGE_CACHE_SIZE) {
2889 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2890 flush_dcache_page(page);
2893 ClearPageChecked(page);
2894 set_page_dirty(page);
2895 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2899 page_cache_release(page);
2904 int btrfs_cont_expand(struct inode *inode, loff_t size)
2906 struct btrfs_trans_handle *trans;
2907 struct btrfs_root *root = BTRFS_I(inode)->root;
2908 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2909 struct extent_map *em;
2910 u64 mask = root->sectorsize - 1;
2911 u64 hole_start = (inode->i_size + mask) & ~mask;
2912 u64 block_end = (size + mask) & ~mask;
2918 if (size <= hole_start)
2921 err = btrfs_check_metadata_free_space(root);
2925 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2928 struct btrfs_ordered_extent *ordered;
2929 btrfs_wait_ordered_range(inode, hole_start,
2930 block_end - hole_start);
2931 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2932 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2935 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2936 btrfs_put_ordered_extent(ordered);
2939 trans = btrfs_start_transaction(root, 1);
2940 btrfs_set_trans_block_group(trans, inode);
2942 cur_offset = hole_start;
2944 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2945 block_end - cur_offset, 0);
2946 BUG_ON(IS_ERR(em) || !em);
2947 last_byte = min(extent_map_end(em), block_end);
2948 last_byte = (last_byte + mask) & ~mask;
2949 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2951 hole_size = last_byte - cur_offset;
2952 err = btrfs_drop_extents(trans, root, inode,
2954 cur_offset + hole_size,
2956 cur_offset, &hint_byte, 1);
2959 err = btrfs_insert_file_extent(trans, root,
2960 inode->i_ino, cur_offset, 0,
2961 0, hole_size, 0, hole_size,
2963 btrfs_drop_extent_cache(inode, hole_start,
2966 free_extent_map(em);
2967 cur_offset = last_byte;
2968 if (err || cur_offset >= block_end)
2972 btrfs_end_transaction(trans, root);
2973 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2977 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2979 struct inode *inode = dentry->d_inode;
2982 err = inode_change_ok(inode, attr);
2986 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2987 if (attr->ia_size > inode->i_size) {
2988 err = btrfs_cont_expand(inode, attr->ia_size);
2991 } else if (inode->i_size > 0 &&
2992 attr->ia_size == 0) {
2994 /* we're truncating a file that used to have good
2995 * data down to zero. Make sure it gets into
2996 * the ordered flush list so that any new writes
2997 * get down to disk quickly.
2999 BTRFS_I(inode)->ordered_data_close = 1;
3003 err = inode_setattr(inode, attr);
3005 if (!err && ((attr->ia_valid & ATTR_MODE)))
3006 err = btrfs_acl_chmod(inode);
3010 void btrfs_delete_inode(struct inode *inode)
3012 struct btrfs_trans_handle *trans;
3013 struct btrfs_root *root = BTRFS_I(inode)->root;
3017 truncate_inode_pages(&inode->i_data, 0);
3018 if (is_bad_inode(inode)) {
3019 btrfs_orphan_del(NULL, inode);
3022 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3024 btrfs_i_size_write(inode, 0);
3025 trans = btrfs_join_transaction(root, 1);
3027 btrfs_set_trans_block_group(trans, inode);
3028 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3030 btrfs_orphan_del(NULL, inode);
3031 goto no_delete_lock;
3034 btrfs_orphan_del(trans, inode);
3036 nr = trans->blocks_used;
3039 btrfs_end_transaction(trans, root);
3040 btrfs_btree_balance_dirty(root, nr);
3044 nr = trans->blocks_used;
3045 btrfs_end_transaction(trans, root);
3046 btrfs_btree_balance_dirty(root, nr);
3052 * this returns the key found in the dir entry in the location pointer.
3053 * If no dir entries were found, location->objectid is 0.
3055 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3056 struct btrfs_key *location)
3058 const char *name = dentry->d_name.name;
3059 int namelen = dentry->d_name.len;
3060 struct btrfs_dir_item *di;
3061 struct btrfs_path *path;
3062 struct btrfs_root *root = BTRFS_I(dir)->root;
3065 path = btrfs_alloc_path();
3068 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3073 if (!di || IS_ERR(di))
3076 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3078 btrfs_free_path(path);
3081 location->objectid = 0;
3086 * when we hit a tree root in a directory, the btrfs part of the inode
3087 * needs to be changed to reflect the root directory of the tree root. This
3088 * is kind of like crossing a mount point.
3090 static int fixup_tree_root_location(struct btrfs_root *root,
3091 struct btrfs_key *location,
3092 struct btrfs_root **sub_root,
3093 struct dentry *dentry)
3095 struct btrfs_root_item *ri;
3097 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3099 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3102 *sub_root = btrfs_read_fs_root(root->fs_info, location,
3103 dentry->d_name.name,
3104 dentry->d_name.len);
3105 if (IS_ERR(*sub_root))
3106 return PTR_ERR(*sub_root);
3108 ri = &(*sub_root)->root_item;
3109 location->objectid = btrfs_root_dirid(ri);
3110 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3111 location->offset = 0;
3116 static void inode_tree_add(struct inode *inode)
3118 struct btrfs_root *root = BTRFS_I(inode)->root;
3119 struct btrfs_inode *entry;
3120 struct rb_node **p = &root->inode_tree.rb_node;
3121 struct rb_node *parent = NULL;
3123 spin_lock(&root->inode_lock);
3126 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3128 if (inode->i_ino < entry->vfs_inode.i_ino)
3130 else if (inode->i_ino > entry->vfs_inode.i_ino)
3131 p = &(*p)->rb_right;
3133 WARN_ON(!(entry->vfs_inode.i_state &
3134 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3138 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3139 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3140 spin_unlock(&root->inode_lock);
3143 static void inode_tree_del(struct inode *inode)
3145 struct btrfs_root *root = BTRFS_I(inode)->root;
3147 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3148 spin_lock(&root->inode_lock);
3149 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3150 spin_unlock(&root->inode_lock);
3151 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3155 static noinline void init_btrfs_i(struct inode *inode)
3157 struct btrfs_inode *bi = BTRFS_I(inode);
3159 bi->i_acl = BTRFS_ACL_NOT_CACHED;
3160 bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
3165 bi->logged_trans = 0;
3166 bi->delalloc_bytes = 0;
3167 bi->reserved_bytes = 0;
3168 bi->disk_i_size = 0;
3170 bi->index_cnt = (u64)-1;
3171 bi->last_unlink_trans = 0;
3172 bi->ordered_data_close = 0;
3173 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3174 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3175 inode->i_mapping, GFP_NOFS);
3176 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3177 inode->i_mapping, GFP_NOFS);
3178 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3179 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3180 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3181 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3182 mutex_init(&BTRFS_I(inode)->extent_mutex);
3183 mutex_init(&BTRFS_I(inode)->log_mutex);
3186 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3188 struct btrfs_iget_args *args = p;
3189 inode->i_ino = args->ino;
3190 init_btrfs_i(inode);
3191 BTRFS_I(inode)->root = args->root;
3192 btrfs_set_inode_space_info(args->root, inode);
3196 static int btrfs_find_actor(struct inode *inode, void *opaque)
3198 struct btrfs_iget_args *args = opaque;
3199 return args->ino == inode->i_ino &&
3200 args->root == BTRFS_I(inode)->root;
3203 static struct inode *btrfs_iget_locked(struct super_block *s,
3205 struct btrfs_root *root)
3207 struct inode *inode;
3208 struct btrfs_iget_args args;
3209 args.ino = objectid;
3212 inode = iget5_locked(s, objectid, btrfs_find_actor,
3213 btrfs_init_locked_inode,
3218 /* Get an inode object given its location and corresponding root.
3219 * Returns in *is_new if the inode was read from disk
3221 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3222 struct btrfs_root *root)
3224 struct inode *inode;
3226 inode = btrfs_iget_locked(s, location->objectid, root);
3228 return ERR_PTR(-ENOMEM);
3230 if (inode->i_state & I_NEW) {
3231 BTRFS_I(inode)->root = root;
3232 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3233 btrfs_read_locked_inode(inode);
3235 inode_tree_add(inode);
3236 unlock_new_inode(inode);
3242 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3244 struct inode *inode;
3245 struct btrfs_inode *bi = BTRFS_I(dir);
3246 struct btrfs_root *root = bi->root;
3247 struct btrfs_root *sub_root = root;
3248 struct btrfs_key location;
3251 if (dentry->d_name.len > BTRFS_NAME_LEN)
3252 return ERR_PTR(-ENAMETOOLONG);
3254 ret = btrfs_inode_by_name(dir, dentry, &location);
3257 return ERR_PTR(ret);
3260 if (location.objectid) {
3261 ret = fixup_tree_root_location(root, &location, &sub_root,
3264 return ERR_PTR(ret);
3266 return ERR_PTR(-ENOENT);
3267 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3269 return ERR_CAST(inode);
3274 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3275 struct nameidata *nd)
3277 struct inode *inode;
3279 if (dentry->d_name.len > BTRFS_NAME_LEN)
3280 return ERR_PTR(-ENAMETOOLONG);
3282 inode = btrfs_lookup_dentry(dir, dentry);
3284 return ERR_CAST(inode);
3286 return d_splice_alias(inode, dentry);
3289 static unsigned char btrfs_filetype_table[] = {
3290 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3293 static int btrfs_real_readdir(struct file *filp, void *dirent,
3296 struct inode *inode = filp->f_dentry->d_inode;
3297 struct btrfs_root *root = BTRFS_I(inode)->root;
3298 struct btrfs_item *item;
3299 struct btrfs_dir_item *di;
3300 struct btrfs_key key;
3301 struct btrfs_key found_key;
3302 struct btrfs_path *path;
3305 struct extent_buffer *leaf;
3308 unsigned char d_type;
3313 int key_type = BTRFS_DIR_INDEX_KEY;
3318 /* FIXME, use a real flag for deciding about the key type */
3319 if (root->fs_info->tree_root == root)
3320 key_type = BTRFS_DIR_ITEM_KEY;
3322 /* special case for "." */
3323 if (filp->f_pos == 0) {
3324 over = filldir(dirent, ".", 1,
3331 /* special case for .., just use the back ref */
3332 if (filp->f_pos == 1) {
3333 u64 pino = parent_ino(filp->f_path.dentry);
3334 over = filldir(dirent, "..", 2,
3340 path = btrfs_alloc_path();
3343 btrfs_set_key_type(&key, key_type);
3344 key.offset = filp->f_pos;
3345 key.objectid = inode->i_ino;
3347 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3353 leaf = path->nodes[0];
3354 nritems = btrfs_header_nritems(leaf);
3355 slot = path->slots[0];
3356 if (advance || slot >= nritems) {
3357 if (slot >= nritems - 1) {
3358 ret = btrfs_next_leaf(root, path);
3361 leaf = path->nodes[0];
3362 nritems = btrfs_header_nritems(leaf);
3363 slot = path->slots[0];
3371 item = btrfs_item_nr(leaf, slot);
3372 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3374 if (found_key.objectid != key.objectid)
3376 if (btrfs_key_type(&found_key) != key_type)
3378 if (found_key.offset < filp->f_pos)
3381 filp->f_pos = found_key.offset;
3383 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3385 di_total = btrfs_item_size(leaf, item);
3387 while (di_cur < di_total) {
3388 struct btrfs_key location;
3390 name_len = btrfs_dir_name_len(leaf, di);
3391 if (name_len <= sizeof(tmp_name)) {
3392 name_ptr = tmp_name;
3394 name_ptr = kmalloc(name_len, GFP_NOFS);
3400 read_extent_buffer(leaf, name_ptr,
3401 (unsigned long)(di + 1), name_len);
3403 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3404 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3406 /* is this a reference to our own snapshot? If so
3409 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3410 location.objectid == root->root_key.objectid) {
3414 over = filldir(dirent, name_ptr, name_len,
3415 found_key.offset, location.objectid,
3419 if (name_ptr != tmp_name)
3424 di_len = btrfs_dir_name_len(leaf, di) +
3425 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3427 di = (struct btrfs_dir_item *)((char *)di + di_len);
3431 /* Reached end of directory/root. Bump pos past the last item. */
3432 if (key_type == BTRFS_DIR_INDEX_KEY)
3433 filp->f_pos = INT_LIMIT(off_t);
3439 btrfs_free_path(path);
3443 int btrfs_write_inode(struct inode *inode, int wait)
3445 struct btrfs_root *root = BTRFS_I(inode)->root;
3446 struct btrfs_trans_handle *trans;
3449 if (root->fs_info->btree_inode == inode)
3453 trans = btrfs_join_transaction(root, 1);
3454 btrfs_set_trans_block_group(trans, inode);
3455 ret = btrfs_commit_transaction(trans, root);
3461 * This is somewhat expensive, updating the tree every time the
3462 * inode changes. But, it is most likely to find the inode in cache.
3463 * FIXME, needs more benchmarking...there are no reasons other than performance
3464 * to keep or drop this code.
3466 void btrfs_dirty_inode(struct inode *inode)
3468 struct btrfs_root *root = BTRFS_I(inode)->root;
3469 struct btrfs_trans_handle *trans;
3471 trans = btrfs_join_transaction(root, 1);
3472 btrfs_set_trans_block_group(trans, inode);
3473 btrfs_update_inode(trans, root, inode);
3474 btrfs_end_transaction(trans, root);
3478 * find the highest existing sequence number in a directory
3479 * and then set the in-memory index_cnt variable to reflect
3480 * free sequence numbers
3482 static int btrfs_set_inode_index_count(struct inode *inode)
3484 struct btrfs_root *root = BTRFS_I(inode)->root;
3485 struct btrfs_key key, found_key;
3486 struct btrfs_path *path;
3487 struct extent_buffer *leaf;
3490 key.objectid = inode->i_ino;
3491 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3492 key.offset = (u64)-1;
3494 path = btrfs_alloc_path();
3498 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3501 /* FIXME: we should be able to handle this */
3507 * MAGIC NUMBER EXPLANATION:
3508 * since we search a directory based on f_pos we have to start at 2
3509 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3510 * else has to start at 2
3512 if (path->slots[0] == 0) {
3513 BTRFS_I(inode)->index_cnt = 2;
3519 leaf = path->nodes[0];
3520 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3522 if (found_key.objectid != inode->i_ino ||
3523 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3524 BTRFS_I(inode)->index_cnt = 2;
3528 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3530 btrfs_free_path(path);
3535 * helper to find a free sequence number in a given directory. This current
3536 * code is very simple, later versions will do smarter things in the btree
3538 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3542 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3543 ret = btrfs_set_inode_index_count(dir);
3548 *index = BTRFS_I(dir)->index_cnt;
3549 BTRFS_I(dir)->index_cnt++;
3554 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3555 struct btrfs_root *root,
3557 const char *name, int name_len,
3558 u64 ref_objectid, u64 objectid,
3559 u64 alloc_hint, int mode, u64 *index)
3561 struct inode *inode;
3562 struct btrfs_inode_item *inode_item;
3563 struct btrfs_key *location;
3564 struct btrfs_path *path;
3565 struct btrfs_inode_ref *ref;
3566 struct btrfs_key key[2];
3572 path = btrfs_alloc_path();
3575 inode = new_inode(root->fs_info->sb);
3577 return ERR_PTR(-ENOMEM);
3580 ret = btrfs_set_inode_index(dir, index);
3583 return ERR_PTR(ret);
3587 * index_cnt is ignored for everything but a dir,
3588 * btrfs_get_inode_index_count has an explanation for the magic
3591 init_btrfs_i(inode);
3592 BTRFS_I(inode)->index_cnt = 2;
3593 BTRFS_I(inode)->root = root;
3594 BTRFS_I(inode)->generation = trans->transid;
3595 btrfs_set_inode_space_info(root, inode);
3601 BTRFS_I(inode)->block_group =
3602 btrfs_find_block_group(root, 0, alloc_hint, owner);
3604 key[0].objectid = objectid;
3605 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3608 key[1].objectid = objectid;
3609 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3610 key[1].offset = ref_objectid;
3612 sizes[0] = sizeof(struct btrfs_inode_item);
3613 sizes[1] = name_len + sizeof(*ref);
3615 path->leave_spinning = 1;
3616 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3620 if (objectid > root->highest_inode)
3621 root->highest_inode = objectid;
3623 inode->i_uid = current_fsuid();
3625 if (dir && (dir->i_mode & S_ISGID)) {
3626 inode->i_gid = dir->i_gid;
3630 inode->i_gid = current_fsgid();
3632 inode->i_mode = mode;
3633 inode->i_ino = objectid;
3634 inode_set_bytes(inode, 0);
3635 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3636 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3637 struct btrfs_inode_item);
3638 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3640 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3641 struct btrfs_inode_ref);
3642 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3643 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3644 ptr = (unsigned long)(ref + 1);
3645 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3647 btrfs_mark_buffer_dirty(path->nodes[0]);
3648 btrfs_free_path(path);
3650 location = &BTRFS_I(inode)->location;
3651 location->objectid = objectid;
3652 location->offset = 0;
3653 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3655 btrfs_inherit_iflags(inode, dir);
3657 if ((mode & S_IFREG)) {
3658 if (btrfs_test_opt(root, NODATASUM))
3659 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
3660 if (btrfs_test_opt(root, NODATACOW))
3661 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
3664 insert_inode_hash(inode);
3665 inode_tree_add(inode);
3669 BTRFS_I(dir)->index_cnt--;
3670 btrfs_free_path(path);
3672 return ERR_PTR(ret);
3675 static inline u8 btrfs_inode_type(struct inode *inode)
3677 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3681 * utility function to add 'inode' into 'parent_inode' with
3682 * a give name and a given sequence number.
3683 * if 'add_backref' is true, also insert a backref from the
3684 * inode to the parent directory.
3686 int btrfs_add_link(struct btrfs_trans_handle *trans,
3687 struct inode *parent_inode, struct inode *inode,
3688 const char *name, int name_len, int add_backref, u64 index)
3691 struct btrfs_key key;
3692 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3694 key.objectid = inode->i_ino;
3695 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3698 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3699 parent_inode->i_ino,
3700 &key, btrfs_inode_type(inode),
3704 ret = btrfs_insert_inode_ref(trans, root,
3707 parent_inode->i_ino,
3710 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3712 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3713 ret = btrfs_update_inode(trans, root, parent_inode);
3718 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3719 struct dentry *dentry, struct inode *inode,
3720 int backref, u64 index)
3722 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3723 inode, dentry->d_name.name,
3724 dentry->d_name.len, backref, index);
3726 d_instantiate(dentry, inode);
3734 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3735 int mode, dev_t rdev)
3737 struct btrfs_trans_handle *trans;
3738 struct btrfs_root *root = BTRFS_I(dir)->root;
3739 struct inode *inode = NULL;
3743 unsigned long nr = 0;
3746 if (!new_valid_dev(rdev))
3749 err = btrfs_check_metadata_free_space(root);
3753 trans = btrfs_start_transaction(root, 1);
3754 btrfs_set_trans_block_group(trans, dir);
3756 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3762 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3764 dentry->d_parent->d_inode->i_ino, objectid,
3765 BTRFS_I(dir)->block_group, mode, &index);
3766 err = PTR_ERR(inode);
3770 err = btrfs_init_inode_security(inode, dir);
3776 btrfs_set_trans_block_group(trans, inode);
3777 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3781 inode->i_op = &btrfs_special_inode_operations;
3782 init_special_inode(inode, inode->i_mode, rdev);
3783 btrfs_update_inode(trans, root, inode);
3785 dir->i_sb->s_dirt = 1;
3786 btrfs_update_inode_block_group(trans, inode);
3787 btrfs_update_inode_block_group(trans, dir);
3789 nr = trans->blocks_used;
3790 btrfs_end_transaction_throttle(trans, root);
3793 inode_dec_link_count(inode);
3796 btrfs_btree_balance_dirty(root, nr);
3800 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3801 int mode, struct nameidata *nd)
3803 struct btrfs_trans_handle *trans;
3804 struct btrfs_root *root = BTRFS_I(dir)->root;
3805 struct inode *inode = NULL;
3808 unsigned long nr = 0;
3812 err = btrfs_check_metadata_free_space(root);
3815 trans = btrfs_start_transaction(root, 1);
3816 btrfs_set_trans_block_group(trans, dir);
3818 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3824 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3826 dentry->d_parent->d_inode->i_ino,
3827 objectid, BTRFS_I(dir)->block_group, mode,
3829 err = PTR_ERR(inode);
3833 err = btrfs_init_inode_security(inode, dir);
3839 btrfs_set_trans_block_group(trans, inode);
3840 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3844 inode->i_mapping->a_ops = &btrfs_aops;
3845 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3846 inode->i_fop = &btrfs_file_operations;
3847 inode->i_op = &btrfs_file_inode_operations;
3848 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3850 dir->i_sb->s_dirt = 1;
3851 btrfs_update_inode_block_group(trans, inode);
3852 btrfs_update_inode_block_group(trans, dir);
3854 nr = trans->blocks_used;
3855 btrfs_end_transaction_throttle(trans, root);
3858 inode_dec_link_count(inode);
3861 btrfs_btree_balance_dirty(root, nr);
3865 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3866 struct dentry *dentry)
3868 struct btrfs_trans_handle *trans;
3869 struct btrfs_root *root = BTRFS_I(dir)->root;
3870 struct inode *inode = old_dentry->d_inode;
3872 unsigned long nr = 0;
3876 if (inode->i_nlink == 0)
3879 btrfs_inc_nlink(inode);
3880 err = btrfs_check_metadata_free_space(root);
3883 err = btrfs_set_inode_index(dir, &index);
3887 trans = btrfs_start_transaction(root, 1);
3889 btrfs_set_trans_block_group(trans, dir);
3890 atomic_inc(&inode->i_count);
3892 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3897 dir->i_sb->s_dirt = 1;
3898 btrfs_update_inode_block_group(trans, dir);
3899 err = btrfs_update_inode(trans, root, inode);
3904 nr = trans->blocks_used;
3906 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3907 btrfs_end_transaction_throttle(trans, root);
3910 inode_dec_link_count(inode);
3913 btrfs_btree_balance_dirty(root, nr);
3917 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3919 struct inode *inode = NULL;
3920 struct btrfs_trans_handle *trans;
3921 struct btrfs_root *root = BTRFS_I(dir)->root;
3923 int drop_on_err = 0;
3926 unsigned long nr = 1;
3928 err = btrfs_check_metadata_free_space(root);
3932 trans = btrfs_start_transaction(root, 1);
3933 btrfs_set_trans_block_group(trans, dir);
3935 if (IS_ERR(trans)) {
3936 err = PTR_ERR(trans);
3940 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3946 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3948 dentry->d_parent->d_inode->i_ino, objectid,
3949 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3951 if (IS_ERR(inode)) {
3952 err = PTR_ERR(inode);
3958 err = btrfs_init_inode_security(inode, dir);
3962 inode->i_op = &btrfs_dir_inode_operations;
3963 inode->i_fop = &btrfs_dir_file_operations;
3964 btrfs_set_trans_block_group(trans, inode);
3966 btrfs_i_size_write(inode, 0);
3967 err = btrfs_update_inode(trans, root, inode);
3971 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3972 inode, dentry->d_name.name,
3973 dentry->d_name.len, 0, index);
3977 d_instantiate(dentry, inode);
3979 dir->i_sb->s_dirt = 1;
3980 btrfs_update_inode_block_group(trans, inode);
3981 btrfs_update_inode_block_group(trans, dir);
3984 nr = trans->blocks_used;
3985 btrfs_end_transaction_throttle(trans, root);
3990 btrfs_btree_balance_dirty(root, nr);
3994 /* helper for btfs_get_extent. Given an existing extent in the tree,
3995 * and an extent that you want to insert, deal with overlap and insert
3996 * the new extent into the tree.
3998 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3999 struct extent_map *existing,
4000 struct extent_map *em,
4001 u64 map_start, u64 map_len)
4005 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4006 start_diff = map_start - em->start;
4007 em->start = map_start;
4009 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4010 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4011 em->block_start += start_diff;
4012 em->block_len -= start_diff;
4014 return add_extent_mapping(em_tree, em);
4017 static noinline int uncompress_inline(struct btrfs_path *path,
4018 struct inode *inode, struct page *page,
4019 size_t pg_offset, u64 extent_offset,
4020 struct btrfs_file_extent_item *item)
4023 struct extent_buffer *leaf = path->nodes[0];
4026 unsigned long inline_size;
4029 WARN_ON(pg_offset != 0);
4030 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4031 inline_size = btrfs_file_extent_inline_item_len(leaf,
4032 btrfs_item_nr(leaf, path->slots[0]));
4033 tmp = kmalloc(inline_size, GFP_NOFS);
4034 ptr = btrfs_file_extent_inline_start(item);
4036 read_extent_buffer(leaf, tmp, ptr, inline_size);
4038 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4039 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4040 inline_size, max_size);
4042 char *kaddr = kmap_atomic(page, KM_USER0);
4043 unsigned long copy_size = min_t(u64,
4044 PAGE_CACHE_SIZE - pg_offset,
4045 max_size - extent_offset);
4046 memset(kaddr + pg_offset, 0, copy_size);
4047 kunmap_atomic(kaddr, KM_USER0);
4054 * a bit scary, this does extent mapping from logical file offset to the disk.
4055 * the ugly parts come from merging extents from the disk with the in-ram
4056 * representation. This gets more complex because of the data=ordered code,
4057 * where the in-ram extents might be locked pending data=ordered completion.
4059 * This also copies inline extents directly into the page.
4062 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4063 size_t pg_offset, u64 start, u64 len,
4069 u64 extent_start = 0;
4071 u64 objectid = inode->i_ino;
4073 struct btrfs_path *path = NULL;
4074 struct btrfs_root *root = BTRFS_I(inode)->root;
4075 struct btrfs_file_extent_item *item;
4076 struct extent_buffer *leaf;
4077 struct btrfs_key found_key;
4078 struct extent_map *em = NULL;
4079 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4081 struct btrfs_trans_handle *trans = NULL;
4085 read_lock(&em_tree->lock);
4086 em = lookup_extent_mapping(em_tree, start, len);
4088 em->bdev = root->fs_info->fs_devices->latest_bdev;
4089 read_unlock(&em_tree->lock);
4092 if (em->start > start || em->start + em->len <= start)
4093 free_extent_map(em);
4094 else if (em->block_start == EXTENT_MAP_INLINE && page)
4095 free_extent_map(em);
4099 em = alloc_extent_map(GFP_NOFS);
4104 em->bdev = root->fs_info->fs_devices->latest_bdev;
4105 em->start = EXTENT_MAP_HOLE;
4106 em->orig_start = EXTENT_MAP_HOLE;
4108 em->block_len = (u64)-1;
4111 path = btrfs_alloc_path();
4115 ret = btrfs_lookup_file_extent(trans, root, path,
4116 objectid, start, trans != NULL);
4123 if (path->slots[0] == 0)
4128 leaf = path->nodes[0];
4129 item = btrfs_item_ptr(leaf, path->slots[0],
4130 struct btrfs_file_extent_item);
4131 /* are we inside the extent that was found? */
4132 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4133 found_type = btrfs_key_type(&found_key);
4134 if (found_key.objectid != objectid ||
4135 found_type != BTRFS_EXTENT_DATA_KEY) {
4139 found_type = btrfs_file_extent_type(leaf, item);
4140 extent_start = found_key.offset;
4141 compressed = btrfs_file_extent_compression(leaf, item);
4142 if (found_type == BTRFS_FILE_EXTENT_REG ||
4143 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4144 extent_end = extent_start +
4145 btrfs_file_extent_num_bytes(leaf, item);
4146 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4148 size = btrfs_file_extent_inline_len(leaf, item);
4149 extent_end = (extent_start + size + root->sectorsize - 1) &
4150 ~((u64)root->sectorsize - 1);
4153 if (start >= extent_end) {
4155 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4156 ret = btrfs_next_leaf(root, path);
4163 leaf = path->nodes[0];
4165 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4166 if (found_key.objectid != objectid ||
4167 found_key.type != BTRFS_EXTENT_DATA_KEY)
4169 if (start + len <= found_key.offset)
4172 em->len = found_key.offset - start;
4176 if (found_type == BTRFS_FILE_EXTENT_REG ||
4177 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4178 em->start = extent_start;
4179 em->len = extent_end - extent_start;
4180 em->orig_start = extent_start -
4181 btrfs_file_extent_offset(leaf, item);
4182 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4184 em->block_start = EXTENT_MAP_HOLE;
4188 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4189 em->block_start = bytenr;
4190 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4193 bytenr += btrfs_file_extent_offset(leaf, item);
4194 em->block_start = bytenr;
4195 em->block_len = em->len;
4196 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4197 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4200 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4204 size_t extent_offset;
4207 em->block_start = EXTENT_MAP_INLINE;
4208 if (!page || create) {
4209 em->start = extent_start;
4210 em->len = extent_end - extent_start;
4214 size = btrfs_file_extent_inline_len(leaf, item);
4215 extent_offset = page_offset(page) + pg_offset - extent_start;
4216 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4217 size - extent_offset);
4218 em->start = extent_start + extent_offset;
4219 em->len = (copy_size + root->sectorsize - 1) &
4220 ~((u64)root->sectorsize - 1);
4221 em->orig_start = EXTENT_MAP_INLINE;
4223 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4224 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4225 if (create == 0 && !PageUptodate(page)) {
4226 if (btrfs_file_extent_compression(leaf, item) ==
4227 BTRFS_COMPRESS_ZLIB) {
4228 ret = uncompress_inline(path, inode, page,
4230 extent_offset, item);
4234 read_extent_buffer(leaf, map + pg_offset, ptr,
4238 flush_dcache_page(page);
4239 } else if (create && PageUptodate(page)) {
4242 free_extent_map(em);
4244 btrfs_release_path(root, path);
4245 trans = btrfs_join_transaction(root, 1);
4249 write_extent_buffer(leaf, map + pg_offset, ptr,
4252 btrfs_mark_buffer_dirty(leaf);
4254 set_extent_uptodate(io_tree, em->start,
4255 extent_map_end(em) - 1, GFP_NOFS);
4258 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4265 em->block_start = EXTENT_MAP_HOLE;
4266 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4268 btrfs_release_path(root, path);
4269 if (em->start > start || extent_map_end(em) <= start) {
4270 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4271 "[%llu %llu]\n", (unsigned long long)em->start,
4272 (unsigned long long)em->len,
4273 (unsigned long long)start,
4274 (unsigned long long)len);
4280 write_lock(&em_tree->lock);
4281 ret = add_extent_mapping(em_tree, em);
4282 /* it is possible that someone inserted the extent into the tree
4283 * while we had the lock dropped. It is also possible that
4284 * an overlapping map exists in the tree
4286 if (ret == -EEXIST) {
4287 struct extent_map *existing;
4291 existing = lookup_extent_mapping(em_tree, start, len);
4292 if (existing && (existing->start > start ||
4293 existing->start + existing->len <= start)) {
4294 free_extent_map(existing);
4298 existing = lookup_extent_mapping(em_tree, em->start,
4301 err = merge_extent_mapping(em_tree, existing,
4304 free_extent_map(existing);
4306 free_extent_map(em);
4311 free_extent_map(em);
4315 free_extent_map(em);
4320 write_unlock(&em_tree->lock);
4323 btrfs_free_path(path);
4325 ret = btrfs_end_transaction(trans, root);
4330 free_extent_map(em);
4331 return ERR_PTR(err);
4336 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4337 const struct iovec *iov, loff_t offset,
4338 unsigned long nr_segs)
4343 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4344 __u64 start, __u64 len)
4346 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4349 int btrfs_readpage(struct file *file, struct page *page)
4351 struct extent_io_tree *tree;
4352 tree = &BTRFS_I(page->mapping->host)->io_tree;
4353 return extent_read_full_page(tree, page, btrfs_get_extent);
4356 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4358 struct extent_io_tree *tree;
4361 if (current->flags & PF_MEMALLOC) {
4362 redirty_page_for_writepage(wbc, page);
4366 tree = &BTRFS_I(page->mapping->host)->io_tree;
4367 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4370 int btrfs_writepages(struct address_space *mapping,
4371 struct writeback_control *wbc)
4373 struct extent_io_tree *tree;
4375 tree = &BTRFS_I(mapping->host)->io_tree;
4376 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4380 btrfs_readpages(struct file *file, struct address_space *mapping,
4381 struct list_head *pages, unsigned nr_pages)
4383 struct extent_io_tree *tree;
4384 tree = &BTRFS_I(mapping->host)->io_tree;
4385 return extent_readpages(tree, mapping, pages, nr_pages,
4388 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4390 struct extent_io_tree *tree;
4391 struct extent_map_tree *map;
4394 tree = &BTRFS_I(page->mapping->host)->io_tree;
4395 map = &BTRFS_I(page->mapping->host)->extent_tree;
4396 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4398 ClearPagePrivate(page);
4399 set_page_private(page, 0);
4400 page_cache_release(page);
4405 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4407 if (PageWriteback(page) || PageDirty(page))
4409 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4412 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4414 struct extent_io_tree *tree;
4415 struct btrfs_ordered_extent *ordered;
4416 u64 page_start = page_offset(page);
4417 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4421 * we have the page locked, so new writeback can't start,
4422 * and the dirty bit won't be cleared while we are here.
4424 * Wait for IO on this page so that we can safely clear
4425 * the PagePrivate2 bit and do ordered accounting
4427 wait_on_page_writeback(page);
4429 tree = &BTRFS_I(page->mapping->host)->io_tree;
4431 btrfs_releasepage(page, GFP_NOFS);
4434 lock_extent(tree, page_start, page_end, GFP_NOFS);
4435 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4439 * IO on this page will never be started, so we need
4440 * to account for any ordered extents now
4442 clear_extent_bit(tree, page_start, page_end,
4443 EXTENT_DIRTY | EXTENT_DELALLOC |
4444 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
4446 * whoever cleared the private bit is responsible
4447 * for the finish_ordered_io
4449 if (TestClearPagePrivate2(page)) {
4450 btrfs_finish_ordered_io(page->mapping->host,
4451 page_start, page_end);
4453 btrfs_put_ordered_extent(ordered);
4454 lock_extent(tree, page_start, page_end, GFP_NOFS);
4456 clear_extent_bit(tree, page_start, page_end,
4457 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
4458 1, 1, NULL, GFP_NOFS);
4459 __btrfs_releasepage(page, GFP_NOFS);
4461 ClearPageChecked(page);
4462 if (PagePrivate(page)) {
4463 ClearPagePrivate(page);
4464 set_page_private(page, 0);
4465 page_cache_release(page);
4470 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4471 * called from a page fault handler when a page is first dirtied. Hence we must
4472 * be careful to check for EOF conditions here. We set the page up correctly
4473 * for a written page which means we get ENOSPC checking when writing into
4474 * holes and correct delalloc and unwritten extent mapping on filesystems that
4475 * support these features.
4477 * We are not allowed to take the i_mutex here so we have to play games to
4478 * protect against truncate races as the page could now be beyond EOF. Because
4479 * vmtruncate() writes the inode size before removing pages, once we have the
4480 * page lock we can determine safely if the page is beyond EOF. If it is not
4481 * beyond EOF, then the page is guaranteed safe against truncation until we
4484 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4486 struct page *page = vmf->page;
4487 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4488 struct btrfs_root *root = BTRFS_I(inode)->root;
4489 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4490 struct btrfs_ordered_extent *ordered;
4492 unsigned long zero_start;
4498 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4502 else /* -ENOSPC, -EIO, etc */
4503 ret = VM_FAULT_SIGBUS;
4507 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4510 size = i_size_read(inode);
4511 page_start = page_offset(page);
4512 page_end = page_start + PAGE_CACHE_SIZE - 1;
4514 if ((page->mapping != inode->i_mapping) ||
4515 (page_start >= size)) {
4516 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4517 /* page got truncated out from underneath us */
4520 wait_on_page_writeback(page);
4522 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4523 set_page_extent_mapped(page);
4526 * we can't set the delalloc bits if there are pending ordered
4527 * extents. Drop our locks and wait for them to finish
4529 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4531 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4533 btrfs_start_ordered_extent(inode, ordered, 1);
4534 btrfs_put_ordered_extent(ordered);
4538 btrfs_set_extent_delalloc(inode, page_start, page_end);
4541 /* page is wholly or partially inside EOF */
4542 if (page_start + PAGE_CACHE_SIZE > size)
4543 zero_start = size & ~PAGE_CACHE_MASK;
4545 zero_start = PAGE_CACHE_SIZE;
4547 if (zero_start != PAGE_CACHE_SIZE) {
4549 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4550 flush_dcache_page(page);
4553 ClearPageChecked(page);
4554 set_page_dirty(page);
4555 SetPageUptodate(page);
4557 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4558 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4562 return VM_FAULT_LOCKED;
4568 static void btrfs_truncate(struct inode *inode)
4570 struct btrfs_root *root = BTRFS_I(inode)->root;
4572 struct btrfs_trans_handle *trans;
4574 u64 mask = root->sectorsize - 1;
4576 if (!S_ISREG(inode->i_mode))
4578 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4581 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4582 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4584 trans = btrfs_start_transaction(root, 1);
4587 * setattr is responsible for setting the ordered_data_close flag,
4588 * but that is only tested during the last file release. That
4589 * could happen well after the next commit, leaving a great big
4590 * window where new writes may get lost if someone chooses to write
4591 * to this file after truncating to zero
4593 * The inode doesn't have any dirty data here, and so if we commit
4594 * this is a noop. If someone immediately starts writing to the inode
4595 * it is very likely we'll catch some of their writes in this
4596 * transaction, and the commit will find this file on the ordered
4597 * data list with good things to send down.
4599 * This is a best effort solution, there is still a window where
4600 * using truncate to replace the contents of the file will
4601 * end up with a zero length file after a crash.
4603 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4604 btrfs_add_ordered_operation(trans, root, inode);
4606 btrfs_set_trans_block_group(trans, inode);
4607 btrfs_i_size_write(inode, inode->i_size);
4609 ret = btrfs_orphan_add(trans, inode);
4612 /* FIXME, add redo link to tree so we don't leak on crash */
4613 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4614 BTRFS_EXTENT_DATA_KEY);
4615 btrfs_update_inode(trans, root, inode);
4617 ret = btrfs_orphan_del(trans, inode);
4621 nr = trans->blocks_used;
4622 ret = btrfs_end_transaction_throttle(trans, root);
4624 btrfs_btree_balance_dirty(root, nr);
4628 * create a new subvolume directory/inode (helper for the ioctl).
4630 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4631 struct btrfs_root *new_root, struct dentry *dentry,
4632 u64 new_dirid, u64 alloc_hint)
4634 struct inode *inode;
4638 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4639 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4641 return PTR_ERR(inode);
4642 inode->i_op = &btrfs_dir_inode_operations;
4643 inode->i_fop = &btrfs_dir_file_operations;
4646 btrfs_i_size_write(inode, 0);
4648 error = btrfs_update_inode(trans, new_root, inode);
4652 d_instantiate(dentry, inode);
4656 /* helper function for file defrag and space balancing. This
4657 * forces readahead on a given range of bytes in an inode
4659 unsigned long btrfs_force_ra(struct address_space *mapping,
4660 struct file_ra_state *ra, struct file *file,
4661 pgoff_t offset, pgoff_t last_index)
4663 pgoff_t req_size = last_index - offset + 1;
4665 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4666 return offset + req_size;
4669 struct inode *btrfs_alloc_inode(struct super_block *sb)
4671 struct btrfs_inode *ei;
4673 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4677 ei->logged_trans = 0;
4678 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4679 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4680 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4681 INIT_LIST_HEAD(&ei->i_orphan);
4682 INIT_LIST_HEAD(&ei->ordered_operations);
4683 return &ei->vfs_inode;
4686 void btrfs_destroy_inode(struct inode *inode)
4688 struct btrfs_ordered_extent *ordered;
4689 struct btrfs_root *root = BTRFS_I(inode)->root;
4691 WARN_ON(!list_empty(&inode->i_dentry));
4692 WARN_ON(inode->i_data.nrpages);
4694 if (BTRFS_I(inode)->i_acl &&
4695 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4696 posix_acl_release(BTRFS_I(inode)->i_acl);
4697 if (BTRFS_I(inode)->i_default_acl &&
4698 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4699 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4702 * Make sure we're properly removed from the ordered operation
4706 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4707 spin_lock(&root->fs_info->ordered_extent_lock);
4708 list_del_init(&BTRFS_I(inode)->ordered_operations);
4709 spin_unlock(&root->fs_info->ordered_extent_lock);
4712 spin_lock(&root->list_lock);
4713 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4714 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4715 " list\n", inode->i_ino);
4718 spin_unlock(&root->list_lock);
4721 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4725 printk(KERN_ERR "btrfs found ordered "
4726 "extent %llu %llu on inode cleanup\n",
4727 (unsigned long long)ordered->file_offset,
4728 (unsigned long long)ordered->len);
4729 btrfs_remove_ordered_extent(inode, ordered);
4730 btrfs_put_ordered_extent(ordered);
4731 btrfs_put_ordered_extent(ordered);
4734 inode_tree_del(inode);
4735 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4736 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4739 static void init_once(void *foo)
4741 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4743 inode_init_once(&ei->vfs_inode);
4746 void btrfs_destroy_cachep(void)
4748 if (btrfs_inode_cachep)
4749 kmem_cache_destroy(btrfs_inode_cachep);
4750 if (btrfs_trans_handle_cachep)
4751 kmem_cache_destroy(btrfs_trans_handle_cachep);
4752 if (btrfs_transaction_cachep)
4753 kmem_cache_destroy(btrfs_transaction_cachep);
4754 if (btrfs_path_cachep)
4755 kmem_cache_destroy(btrfs_path_cachep);
4758 int btrfs_init_cachep(void)
4760 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4761 sizeof(struct btrfs_inode), 0,
4762 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4763 if (!btrfs_inode_cachep)
4766 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4767 sizeof(struct btrfs_trans_handle), 0,
4768 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4769 if (!btrfs_trans_handle_cachep)
4772 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4773 sizeof(struct btrfs_transaction), 0,
4774 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4775 if (!btrfs_transaction_cachep)
4778 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4779 sizeof(struct btrfs_path), 0,
4780 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4781 if (!btrfs_path_cachep)
4786 btrfs_destroy_cachep();
4790 static int btrfs_getattr(struct vfsmount *mnt,
4791 struct dentry *dentry, struct kstat *stat)
4793 struct inode *inode = dentry->d_inode;
4794 generic_fillattr(inode, stat);
4795 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4796 stat->blksize = PAGE_CACHE_SIZE;
4797 stat->blocks = (inode_get_bytes(inode) +
4798 BTRFS_I(inode)->delalloc_bytes) >> 9;
4802 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4803 struct inode *new_dir, struct dentry *new_dentry)
4805 struct btrfs_trans_handle *trans;
4806 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4807 struct inode *new_inode = new_dentry->d_inode;
4808 struct inode *old_inode = old_dentry->d_inode;
4809 struct timespec ctime = CURRENT_TIME;
4813 /* we're not allowed to rename between subvolumes */
4814 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4815 BTRFS_I(new_dir)->root->root_key.objectid)
4818 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4819 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4823 /* to rename a snapshot or subvolume, we need to juggle the
4824 * backrefs. This isn't coded yet
4826 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4829 ret = btrfs_check_metadata_free_space(root);
4834 * we're using rename to replace one file with another.
4835 * and the replacement file is large. Start IO on it now so
4836 * we don't add too much work to the end of the transaction
4838 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
4839 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4840 filemap_flush(old_inode->i_mapping);
4842 trans = btrfs_start_transaction(root, 1);
4845 * make sure the inode gets flushed if it is replacing
4848 if (new_inode && new_inode->i_size &&
4849 old_inode && S_ISREG(old_inode->i_mode)) {
4850 btrfs_add_ordered_operation(trans, root, old_inode);
4854 * this is an ugly little race, but the rename is required to make
4855 * sure that if we crash, the inode is either at the old name
4856 * or the new one. pinning the log transaction lets us make sure
4857 * we don't allow a log commit to come in after we unlink the
4858 * name but before we add the new name back in.
4860 btrfs_pin_log_trans(root);
4862 btrfs_set_trans_block_group(trans, new_dir);
4864 btrfs_inc_nlink(old_dentry->d_inode);
4865 old_dir->i_ctime = old_dir->i_mtime = ctime;
4866 new_dir->i_ctime = new_dir->i_mtime = ctime;
4867 old_inode->i_ctime = ctime;
4869 if (old_dentry->d_parent != new_dentry->d_parent)
4870 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4872 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4873 old_dentry->d_name.name,
4874 old_dentry->d_name.len);
4879 new_inode->i_ctime = CURRENT_TIME;
4880 ret = btrfs_unlink_inode(trans, root, new_dir,
4881 new_dentry->d_inode,
4882 new_dentry->d_name.name,
4883 new_dentry->d_name.len);
4886 if (new_inode->i_nlink == 0) {
4887 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4893 ret = btrfs_set_inode_index(new_dir, &index);
4897 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4898 old_inode, new_dentry->d_name.name,
4899 new_dentry->d_name.len, 1, index);
4903 btrfs_log_new_name(trans, old_inode, old_dir,
4904 new_dentry->d_parent);
4907 /* this btrfs_end_log_trans just allows the current
4908 * log-sub transaction to complete
4910 btrfs_end_log_trans(root);
4911 btrfs_end_transaction_throttle(trans, root);
4917 * some fairly slow code that needs optimization. This walks the list
4918 * of all the inodes with pending delalloc and forces them to disk.
4920 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4922 struct list_head *head = &root->fs_info->delalloc_inodes;
4923 struct btrfs_inode *binode;
4924 struct inode *inode;
4926 if (root->fs_info->sb->s_flags & MS_RDONLY)
4929 spin_lock(&root->fs_info->delalloc_lock);
4930 while (!list_empty(head)) {
4931 binode = list_entry(head->next, struct btrfs_inode,
4933 inode = igrab(&binode->vfs_inode);
4935 list_del_init(&binode->delalloc_inodes);
4936 spin_unlock(&root->fs_info->delalloc_lock);
4938 filemap_flush(inode->i_mapping);
4942 spin_lock(&root->fs_info->delalloc_lock);
4944 spin_unlock(&root->fs_info->delalloc_lock);
4946 /* the filemap_flush will queue IO into the worker threads, but
4947 * we have to make sure the IO is actually started and that
4948 * ordered extents get created before we return
4950 atomic_inc(&root->fs_info->async_submit_draining);
4951 while (atomic_read(&root->fs_info->nr_async_submits) ||
4952 atomic_read(&root->fs_info->async_delalloc_pages)) {
4953 wait_event(root->fs_info->async_submit_wait,
4954 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4955 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4957 atomic_dec(&root->fs_info->async_submit_draining);
4961 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4962 const char *symname)
4964 struct btrfs_trans_handle *trans;
4965 struct btrfs_root *root = BTRFS_I(dir)->root;
4966 struct btrfs_path *path;
4967 struct btrfs_key key;
4968 struct inode *inode = NULL;
4976 struct btrfs_file_extent_item *ei;
4977 struct extent_buffer *leaf;
4978 unsigned long nr = 0;
4980 name_len = strlen(symname) + 1;
4981 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4982 return -ENAMETOOLONG;
4984 err = btrfs_check_metadata_free_space(root);
4988 trans = btrfs_start_transaction(root, 1);
4989 btrfs_set_trans_block_group(trans, dir);
4991 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4997 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4999 dentry->d_parent->d_inode->i_ino, objectid,
5000 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5002 err = PTR_ERR(inode);
5006 err = btrfs_init_inode_security(inode, dir);
5012 btrfs_set_trans_block_group(trans, inode);
5013 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5017 inode->i_mapping->a_ops = &btrfs_aops;
5018 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5019 inode->i_fop = &btrfs_file_operations;
5020 inode->i_op = &btrfs_file_inode_operations;
5021 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5023 dir->i_sb->s_dirt = 1;
5024 btrfs_update_inode_block_group(trans, inode);
5025 btrfs_update_inode_block_group(trans, dir);
5029 path = btrfs_alloc_path();
5031 key.objectid = inode->i_ino;
5033 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5034 datasize = btrfs_file_extent_calc_inline_size(name_len);
5035 err = btrfs_insert_empty_item(trans, root, path, &key,
5041 leaf = path->nodes[0];
5042 ei = btrfs_item_ptr(leaf, path->slots[0],
5043 struct btrfs_file_extent_item);
5044 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5045 btrfs_set_file_extent_type(leaf, ei,
5046 BTRFS_FILE_EXTENT_INLINE);
5047 btrfs_set_file_extent_encryption(leaf, ei, 0);
5048 btrfs_set_file_extent_compression(leaf, ei, 0);
5049 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5050 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5052 ptr = btrfs_file_extent_inline_start(ei);
5053 write_extent_buffer(leaf, symname, ptr, name_len);
5054 btrfs_mark_buffer_dirty(leaf);
5055 btrfs_free_path(path);
5057 inode->i_op = &btrfs_symlink_inode_operations;
5058 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5059 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5060 inode_set_bytes(inode, name_len);
5061 btrfs_i_size_write(inode, name_len - 1);
5062 err = btrfs_update_inode(trans, root, inode);
5067 nr = trans->blocks_used;
5068 btrfs_end_transaction_throttle(trans, root);
5071 inode_dec_link_count(inode);
5074 btrfs_btree_balance_dirty(root, nr);
5078 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5079 struct inode *inode, u64 start, u64 end,
5080 u64 locked_end, u64 alloc_hint, int mode)
5082 struct btrfs_root *root = BTRFS_I(inode)->root;
5083 struct btrfs_key ins;
5085 u64 cur_offset = start;
5086 u64 num_bytes = end - start;
5089 while (num_bytes > 0) {
5090 alloc_size = min(num_bytes, root->fs_info->max_extent);
5091 ret = btrfs_reserve_extent(trans, root, alloc_size,
5092 root->sectorsize, 0, alloc_hint,
5098 ret = insert_reserved_file_extent(trans, inode,
5099 cur_offset, ins.objectid,
5100 ins.offset, ins.offset,
5101 ins.offset, locked_end,
5103 BTRFS_FILE_EXTENT_PREALLOC);
5105 btrfs_drop_extent_cache(inode, cur_offset,
5106 cur_offset + ins.offset -1, 0);
5107 num_bytes -= ins.offset;
5108 cur_offset += ins.offset;
5109 alloc_hint = ins.objectid + ins.offset;
5112 if (cur_offset > start) {
5113 inode->i_ctime = CURRENT_TIME;
5114 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5115 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5116 cur_offset > i_size_read(inode))
5117 btrfs_i_size_write(inode, cur_offset);
5118 ret = btrfs_update_inode(trans, root, inode);
5125 static long btrfs_fallocate(struct inode *inode, int mode,
5126 loff_t offset, loff_t len)
5134 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5135 struct extent_map *em;
5136 struct btrfs_trans_handle *trans;
5137 struct btrfs_root *root;
5140 alloc_start = offset & ~mask;
5141 alloc_end = (offset + len + mask) & ~mask;
5144 * wait for ordered IO before we have any locks. We'll loop again
5145 * below with the locks held.
5147 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5149 mutex_lock(&inode->i_mutex);
5150 if (alloc_start > inode->i_size) {
5151 ret = btrfs_cont_expand(inode, alloc_start);
5156 root = BTRFS_I(inode)->root;
5158 ret = btrfs_check_data_free_space(root, inode,
5159 alloc_end - alloc_start);
5163 locked_end = alloc_end - 1;
5165 struct btrfs_ordered_extent *ordered;
5167 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5173 /* the extent lock is ordered inside the running
5176 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5178 ordered = btrfs_lookup_first_ordered_extent(inode,
5181 ordered->file_offset + ordered->len > alloc_start &&
5182 ordered->file_offset < alloc_end) {
5183 btrfs_put_ordered_extent(ordered);
5184 unlock_extent(&BTRFS_I(inode)->io_tree,
5185 alloc_start, locked_end, GFP_NOFS);
5186 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5189 * we can't wait on the range with the transaction
5190 * running or with the extent lock held
5192 btrfs_wait_ordered_range(inode, alloc_start,
5193 alloc_end - alloc_start);
5196 btrfs_put_ordered_extent(ordered);
5201 cur_offset = alloc_start;
5203 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5204 alloc_end - cur_offset, 0);
5205 BUG_ON(IS_ERR(em) || !em);
5206 last_byte = min(extent_map_end(em), alloc_end);
5207 last_byte = (last_byte + mask) & ~mask;
5208 if (em->block_start == EXTENT_MAP_HOLE) {
5209 ret = prealloc_file_range(trans, inode, cur_offset,
5210 last_byte, locked_end + 1,
5213 free_extent_map(em);
5217 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5218 alloc_hint = em->block_start;
5219 free_extent_map(em);
5221 cur_offset = last_byte;
5222 if (cur_offset >= alloc_end) {
5227 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5230 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5232 btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5234 mutex_unlock(&inode->i_mutex);
5238 static int btrfs_set_page_dirty(struct page *page)
5240 return __set_page_dirty_nobuffers(page);
5243 static int btrfs_permission(struct inode *inode, int mask)
5245 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5247 return generic_permission(inode, mask, btrfs_check_acl);
5250 static struct inode_operations btrfs_dir_inode_operations = {
5251 .getattr = btrfs_getattr,
5252 .lookup = btrfs_lookup,
5253 .create = btrfs_create,
5254 .unlink = btrfs_unlink,
5256 .mkdir = btrfs_mkdir,
5257 .rmdir = btrfs_rmdir,
5258 .rename = btrfs_rename,
5259 .symlink = btrfs_symlink,
5260 .setattr = btrfs_setattr,
5261 .mknod = btrfs_mknod,
5262 .setxattr = btrfs_setxattr,
5263 .getxattr = btrfs_getxattr,
5264 .listxattr = btrfs_listxattr,
5265 .removexattr = btrfs_removexattr,
5266 .permission = btrfs_permission,
5268 static struct inode_operations btrfs_dir_ro_inode_operations = {
5269 .lookup = btrfs_lookup,
5270 .permission = btrfs_permission,
5272 static struct file_operations btrfs_dir_file_operations = {
5273 .llseek = generic_file_llseek,
5274 .read = generic_read_dir,
5275 .readdir = btrfs_real_readdir,
5276 .unlocked_ioctl = btrfs_ioctl,
5277 #ifdef CONFIG_COMPAT
5278 .compat_ioctl = btrfs_ioctl,
5280 .release = btrfs_release_file,
5281 .fsync = btrfs_sync_file,
5284 static struct extent_io_ops btrfs_extent_io_ops = {
5285 .fill_delalloc = run_delalloc_range,
5286 .submit_bio_hook = btrfs_submit_bio_hook,
5287 .merge_bio_hook = btrfs_merge_bio_hook,
5288 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5289 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5290 .writepage_start_hook = btrfs_writepage_start_hook,
5291 .readpage_io_failed_hook = btrfs_io_failed_hook,
5292 .set_bit_hook = btrfs_set_bit_hook,
5293 .clear_bit_hook = btrfs_clear_bit_hook,
5297 * btrfs doesn't support the bmap operation because swapfiles
5298 * use bmap to make a mapping of extents in the file. They assume
5299 * these extents won't change over the life of the file and they
5300 * use the bmap result to do IO directly to the drive.
5302 * the btrfs bmap call would return logical addresses that aren't
5303 * suitable for IO and they also will change frequently as COW
5304 * operations happen. So, swapfile + btrfs == corruption.
5306 * For now we're avoiding this by dropping bmap.
5308 static struct address_space_operations btrfs_aops = {
5309 .readpage = btrfs_readpage,
5310 .writepage = btrfs_writepage,
5311 .writepages = btrfs_writepages,
5312 .readpages = btrfs_readpages,
5313 .sync_page = block_sync_page,
5314 .direct_IO = btrfs_direct_IO,
5315 .invalidatepage = btrfs_invalidatepage,
5316 .releasepage = btrfs_releasepage,
5317 .set_page_dirty = btrfs_set_page_dirty,
5320 static struct address_space_operations btrfs_symlink_aops = {
5321 .readpage = btrfs_readpage,
5322 .writepage = btrfs_writepage,
5323 .invalidatepage = btrfs_invalidatepage,
5324 .releasepage = btrfs_releasepage,
5327 static struct inode_operations btrfs_file_inode_operations = {
5328 .truncate = btrfs_truncate,
5329 .getattr = btrfs_getattr,
5330 .setattr = btrfs_setattr,
5331 .setxattr = btrfs_setxattr,
5332 .getxattr = btrfs_getxattr,
5333 .listxattr = btrfs_listxattr,
5334 .removexattr = btrfs_removexattr,
5335 .permission = btrfs_permission,
5336 .fallocate = btrfs_fallocate,
5337 .fiemap = btrfs_fiemap,
5339 static struct inode_operations btrfs_special_inode_operations = {
5340 .getattr = btrfs_getattr,
5341 .setattr = btrfs_setattr,
5342 .permission = btrfs_permission,
5343 .setxattr = btrfs_setxattr,
5344 .getxattr = btrfs_getxattr,
5345 .listxattr = btrfs_listxattr,
5346 .removexattr = btrfs_removexattr,
5348 static struct inode_operations btrfs_symlink_inode_operations = {
5349 .readlink = generic_readlink,
5350 .follow_link = page_follow_link_light,
5351 .put_link = page_put_link,
5352 .permission = btrfs_permission,
5353 .setxattr = btrfs_setxattr,
5354 .getxattr = btrfs_getxattr,
5355 .listxattr = btrfs_listxattr,
5356 .removexattr = btrfs_removexattr,