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.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/version.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
41 static int btrfs_copy_from_user(loff_t pos
, int num_pages
, int write_bytes
,
42 struct page
**prepared_pages
,
43 const char __user
* buf
)
47 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
49 for (i
= 0; i
< num_pages
&& write_bytes
> 0; i
++, offset
= 0) {
50 size_t count
= min_t(size_t,
51 PAGE_CACHE_SIZE
- offset
, write_bytes
);
52 struct page
*page
= prepared_pages
[i
];
53 fault_in_pages_readable(buf
, count
);
55 /* Copy data from userspace to the current page */
57 page_fault
= __copy_from_user(page_address(page
) + offset
,
59 /* Flush processor's dcache for this page */
60 flush_dcache_page(page
);
68 return page_fault
? -EFAULT
: 0;
71 static void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
74 for (i
= 0; i
< num_pages
; i
++) {
77 unlock_page(pages
[i
]);
78 mark_page_accessed(pages
[i
]);
79 page_cache_release(pages
[i
]);
83 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
84 struct btrfs_root
*root
, struct inode
*inode
,
85 u64 offset
, size_t size
,
86 struct page
**pages
, size_t page_offset
,
90 struct btrfs_path
*path
;
91 struct extent_buffer
*leaf
;
94 struct btrfs_file_extent_item
*ei
;
102 path
= btrfs_alloc_path();
106 btrfs_set_trans_block_group(trans
, inode
);
108 key
.objectid
= inode
->i_ino
;
110 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
112 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
119 leaf
= path
->nodes
[0];
120 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
121 struct btrfs_file_extent_item
);
123 if (btrfs_file_extent_type(leaf
, ei
) !=
124 BTRFS_FILE_EXTENT_INLINE
) {
127 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
134 leaf
= path
->nodes
[0];
135 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
136 struct btrfs_file_extent_item
);
138 if (btrfs_file_extent_type(leaf
, ei
) !=
139 BTRFS_FILE_EXTENT_INLINE
) {
141 btrfs_print_leaf(root
, leaf
);
142 printk("found wasn't inline offset %Lu inode %lu\n",
143 offset
, inode
->i_ino
);
146 found_size
= btrfs_file_extent_inline_len(leaf
,
147 btrfs_item_nr(leaf
, path
->slots
[0]));
148 found_end
= key
.offset
+ found_size
;
150 if (found_end
< offset
+ size
) {
151 btrfs_release_path(root
, path
);
152 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
153 offset
+ size
- found_end
, 1);
155 ret
= btrfs_extend_item(trans
, root
, path
,
156 offset
+ size
- found_end
);
161 leaf
= path
->nodes
[0];
162 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
163 struct btrfs_file_extent_item
);
165 if (found_end
< offset
) {
166 ptr
= btrfs_file_extent_inline_start(ei
) + found_size
;
167 memset_extent_buffer(leaf
, 0, ptr
, offset
- found_end
);
171 btrfs_release_path(root
, path
);
172 datasize
= offset
+ size
- key
.offset
;
173 datasize
= btrfs_file_extent_calc_inline_size(datasize
);
174 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
178 printk("got bad ret %d\n", ret
);
181 leaf
= path
->nodes
[0];
182 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
183 struct btrfs_file_extent_item
);
184 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
185 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
187 ptr
= btrfs_file_extent_inline_start(ei
) + offset
- key
.offset
;
193 kaddr
= kmap_atomic(page
, KM_USER0
);
194 cur_size
= min_t(size_t, PAGE_CACHE_SIZE
- page_offset
, size
);
195 write_extent_buffer(leaf
, kaddr
+ page_offset
, ptr
, cur_size
);
196 kunmap_atomic(kaddr
, KM_USER0
);
200 if (i
>= num_pages
) {
201 printk("i %d num_pages %d\n", i
, num_pages
);
205 btrfs_mark_buffer_dirty(leaf
);
207 btrfs_free_path(path
);
211 static int dirty_and_release_pages(struct btrfs_trans_handle
*trans
,
212 struct btrfs_root
*root
,
221 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
222 struct extent_map
*em
;
223 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
227 u64 end_of_last_block
;
228 u64 end_pos
= pos
+ write_bytes
;
230 loff_t isize
= i_size_read(inode
);
232 em
= alloc_extent_map(GFP_NOFS
);
236 em
->bdev
= inode
->i_sb
->s_bdev
;
238 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
239 num_bytes
= (write_bytes
+ pos
- start_pos
+
240 root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
242 down_read(&BTRFS_I(inode
)->root
->snap_sem
);
243 end_of_last_block
= start_pos
+ num_bytes
- 1;
245 lock_extent(em_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
246 mutex_lock(&root
->fs_info
->fs_mutex
);
247 trans
= btrfs_start_transaction(root
, 1);
252 btrfs_set_trans_block_group(trans
, inode
);
253 inode
->i_blocks
+= num_bytes
>> 9;
256 if ((end_of_last_block
& 4095) == 0) {
257 printk("strange end of last %Lu %zu %Lu\n", start_pos
, write_bytes
, end_of_last_block
);
259 set_extent_uptodate(em_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
261 /* FIXME...EIEIO, ENOSPC and more */
263 /* insert any holes we need to create */
264 if (inode
->i_size
< start_pos
) {
265 u64 last_pos_in_file
;
267 u64 mask
= root
->sectorsize
- 1;
268 last_pos_in_file
= (isize
+ mask
) & ~mask
;
269 hole_size
= (start_pos
- last_pos_in_file
+ mask
) & ~mask
;
271 if (last_pos_in_file
< start_pos
) {
272 err
= btrfs_drop_extents(trans
, root
, inode
,
274 last_pos_in_file
+ hole_size
,
280 err
= btrfs_insert_file_extent(trans
, root
,
290 * either allocate an extent for the new bytes or setup the key
291 * to show we are doing inline data in the extent
293 inline_size
= end_pos
;
294 if (isize
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
295 inline_size
> 8192 ||
296 inline_size
>= BTRFS_MAX_INLINE_DATA_SIZE(root
)) {
299 for (i
= 0; i
< num_pages
; i
++) {
300 struct page
*p
= pages
[i
];
304 last_end
= pages
[num_pages
-1]->index
<< PAGE_CACHE_SHIFT
;
305 last_end
+= PAGE_CACHE_SIZE
- 1;
306 set_extent_delalloc(em_tree
, start_pos
, end_of_last_block
,
310 /* step one, delete the existing extents in this range */
311 aligned_end
= (pos
+ write_bytes
+ root
->sectorsize
- 1) &
312 ~((u64
)root
->sectorsize
- 1);
313 err
= btrfs_drop_extents(trans
, root
, inode
, start_pos
,
314 aligned_end
, end_pos
, &hint_byte
);
317 err
= insert_inline_extent(trans
, root
, inode
, start_pos
,
318 end_pos
- start_pos
, pages
, 0,
322 if (end_pos
> isize
) {
323 i_size_write(inode
, end_pos
);
324 btrfs_update_inode(trans
, root
, inode
);
327 err
= btrfs_end_transaction(trans
, root
);
329 mutex_unlock(&root
->fs_info
->fs_mutex
);
330 unlock_extent(em_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
332 up_read(&BTRFS_I(inode
)->root
->snap_sem
);
336 int btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
)
338 struct extent_map
*em
;
339 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
342 em
= lookup_extent_mapping(em_tree
, start
, end
);
345 remove_extent_mapping(em_tree
, em
);
348 /* once for the tree*/
355 * this is very complex, but the basic idea is to drop all extents
356 * in the range start - end. hint_block is filled in with a block number
357 * that would be a good hint to the block allocator for this file.
359 * If an extent intersects the range but is not entirely inside the range
360 * it is either truncated or split. Anything entirely inside the range
361 * is deleted from the tree.
363 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
364 struct btrfs_root
*root
, struct inode
*inode
,
365 u64 start
, u64 end
, u64 inline_end
, u64
*hint_byte
)
368 struct btrfs_key key
;
369 struct extent_buffer
*leaf
;
371 struct btrfs_file_extent_item
*extent
;
374 struct btrfs_file_extent_item old
;
375 struct btrfs_path
*path
;
376 u64 search_start
= start
;
383 btrfs_drop_extent_cache(inode
, start
, end
- 1);
385 path
= btrfs_alloc_path();
390 btrfs_release_path(root
, path
);
391 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
396 if (path
->slots
[0] == 0) {
408 leaf
= path
->nodes
[0];
409 slot
= path
->slots
[0];
411 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
412 if (key
.offset
>= end
|| key
.objectid
!= inode
->i_ino
) {
415 if (btrfs_key_type(&key
) > BTRFS_EXTENT_DATA_KEY
) {
419 search_start
= key
.offset
;
422 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
) {
423 extent
= btrfs_item_ptr(leaf
, slot
,
424 struct btrfs_file_extent_item
);
425 found_type
= btrfs_file_extent_type(leaf
, extent
);
426 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
427 extent_end
= key
.offset
+
428 btrfs_file_extent_num_bytes(leaf
, extent
);
430 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
431 struct btrfs_item
*item
;
432 item
= btrfs_item_nr(leaf
, slot
);
434 extent_end
= key
.offset
+
435 btrfs_file_extent_inline_len(leaf
, item
);
438 extent_end
= search_start
;
441 /* we found nothing we can drop */
442 if ((!found_extent
&& !found_inline
) ||
443 search_start
>= extent_end
) {
446 nritems
= btrfs_header_nritems(leaf
);
447 if (slot
>= nritems
- 1) {
448 nextret
= btrfs_next_leaf(root
, path
);
458 /* FIXME, there's only one inline extent allowed right now */
460 u64 mask
= root
->sectorsize
- 1;
461 search_start
= (extent_end
+ mask
) & ~mask
;
463 search_start
= extent_end
;
465 if (end
< extent_end
&& end
>= key
.offset
) {
468 btrfs_file_extent_disk_bytenr(leaf
, extent
);
470 btrfs_file_extent_disk_num_bytes(leaf
,
472 read_extent_buffer(leaf
, &old
,
473 (unsigned long)extent
,
475 if (disk_bytenr
!= 0) {
476 ret
= btrfs_inc_extent_ref(trans
, root
,
477 disk_bytenr
, disk_num_bytes
);
484 /* truncate existing extent */
485 if (start
> key
.offset
) {
489 WARN_ON(start
& (root
->sectorsize
- 1));
491 new_num
= start
- key
.offset
;
492 old_num
= btrfs_file_extent_num_bytes(leaf
,
495 btrfs_file_extent_disk_bytenr(leaf
,
497 if (btrfs_file_extent_disk_bytenr(leaf
,
500 (old_num
- new_num
) >> 9;
502 btrfs_set_file_extent_num_bytes(leaf
, extent
,
504 btrfs_mark_buffer_dirty(leaf
);
505 } else if (end
> extent_end
&&
506 key
.offset
< inline_end
&&
507 inline_end
< extent_end
) {
509 new_size
= btrfs_file_extent_calc_inline_size(
510 inline_end
- key
.offset
);
511 btrfs_truncate_item(trans
, root
, path
,
515 /* delete the entire extent */
518 u64 disk_num_bytes
= 0;
519 u64 extent_num_bytes
= 0;
522 btrfs_file_extent_disk_bytenr(leaf
,
525 btrfs_file_extent_disk_num_bytes(leaf
,
528 btrfs_file_extent_num_bytes(leaf
, extent
);
530 btrfs_file_extent_disk_bytenr(leaf
,
533 ret
= btrfs_del_item(trans
, root
, path
);
534 /* TODO update progress marker and return */
536 btrfs_release_path(root
, path
);
538 if (found_extent
&& disk_bytenr
!= 0) {
539 inode
->i_blocks
-= extent_num_bytes
>> 9;
540 ret
= btrfs_free_extent(trans
, root
,
546 if (!bookend
&& search_start
>= end
) {
553 /* create bookend, splitting the extent in two */
554 if (bookend
&& found_extent
) {
555 struct btrfs_key ins
;
556 ins
.objectid
= inode
->i_ino
;
558 btrfs_set_key_type(&ins
, BTRFS_EXTENT_DATA_KEY
);
559 btrfs_release_path(root
, path
);
560 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
563 leaf
= path
->nodes
[0];
565 btrfs_print_leaf(root
, leaf
);
566 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret
, ins
.objectid
, ins
.type
, ins
.offset
, start
, end
, key
.offset
, extent_end
, keep
);
569 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
570 struct btrfs_file_extent_item
);
571 write_extent_buffer(leaf
, &old
,
572 (unsigned long)extent
, sizeof(old
));
574 btrfs_set_file_extent_offset(leaf
, extent
,
575 le64_to_cpu(old
.offset
) + end
- key
.offset
);
576 WARN_ON(le64_to_cpu(old
.num_bytes
) <
578 btrfs_set_file_extent_num_bytes(leaf
, extent
,
580 btrfs_set_file_extent_type(leaf
, extent
,
581 BTRFS_FILE_EXTENT_REG
);
583 btrfs_mark_buffer_dirty(path
->nodes
[0]);
584 if (le64_to_cpu(old
.disk_bytenr
) != 0) {
586 btrfs_file_extent_num_bytes(leaf
,
594 btrfs_free_path(path
);
599 * this gets pages into the page cache and locks them down
601 static int prepare_pages(struct btrfs_root
*root
,
606 unsigned long first_index
,
607 unsigned long last_index
,
611 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
612 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
616 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
618 memset(pages
, 0, num_pages
* sizeof(struct page
*));
620 for (i
= 0; i
< num_pages
; i
++) {
621 pages
[i
] = grab_cache_page(inode
->i_mapping
, index
+ i
);
626 cancel_dirty_page(pages
[i
], PAGE_CACHE_SIZE
);
627 wait_on_page_writeback(pages
[i
]);
628 set_page_extent_mapped(pages
[i
]);
629 WARN_ON(!PageLocked(pages
[i
]));
634 static ssize_t
btrfs_file_write(struct file
*file
, const char __user
*buf
,
635 size_t count
, loff_t
*ppos
)
639 ssize_t num_written
= 0;
642 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
643 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
644 struct page
**pages
= NULL
;
646 struct page
*pinned
[2];
647 unsigned long first_index
;
648 unsigned long last_index
;
650 nrptrs
= min((count
+ PAGE_CACHE_SIZE
- 1) / PAGE_CACHE_SIZE
,
651 PAGE_CACHE_SIZE
/ (sizeof(struct page
*)));
654 if (file
->f_flags
& O_DIRECT
)
660 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
661 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
662 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
667 err
= remove_suid(file
->f_path
.dentry
);
670 file_update_time(file
);
672 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
674 mutex_lock(&inode
->i_mutex
);
675 first_index
= pos
>> PAGE_CACHE_SHIFT
;
676 last_index
= (pos
+ count
) >> PAGE_CACHE_SHIFT
;
679 * there are lots of better ways to do this, but this code
680 * makes sure the first and last page in the file range are
681 * up to date and ready for cow
683 if ((pos
& (PAGE_CACHE_SIZE
- 1))) {
684 pinned
[0] = grab_cache_page(inode
->i_mapping
, first_index
);
685 if (!PageUptodate(pinned
[0])) {
686 ret
= btrfs_readpage(NULL
, pinned
[0]);
688 wait_on_page_locked(pinned
[0]);
690 unlock_page(pinned
[0]);
693 if ((pos
+ count
) & (PAGE_CACHE_SIZE
- 1)) {
694 pinned
[1] = grab_cache_page(inode
->i_mapping
, last_index
);
695 if (!PageUptodate(pinned
[1])) {
696 ret
= btrfs_readpage(NULL
, pinned
[1]);
698 wait_on_page_locked(pinned
[1]);
700 unlock_page(pinned
[1]);
705 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
706 size_t write_bytes
= min(count
, nrptrs
*
707 (size_t)PAGE_CACHE_SIZE
-
709 size_t num_pages
= (write_bytes
+ PAGE_CACHE_SIZE
- 1) >>
712 WARN_ON(num_pages
> nrptrs
);
713 memset(pages
, 0, sizeof(pages
));
714 ret
= prepare_pages(root
, file
, pages
, num_pages
,
715 pos
, first_index
, last_index
,
720 ret
= btrfs_copy_from_user(pos
, num_pages
,
721 write_bytes
, pages
, buf
);
723 btrfs_drop_pages(pages
, num_pages
);
727 ret
= dirty_and_release_pages(NULL
, root
, file
, pages
,
728 num_pages
, pos
, write_bytes
);
729 btrfs_drop_pages(pages
, num_pages
);
734 count
-= write_bytes
;
736 num_written
+= write_bytes
;
738 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, num_pages
);
739 btrfs_btree_balance_dirty(root
, 1);
742 mutex_unlock(&inode
->i_mutex
);
746 page_cache_release(pinned
[0]);
748 page_cache_release(pinned
[1]);
751 if (num_written
> 0 && ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
752 err
= sync_page_range(inode
, inode
->i_mapping
,
753 start_pos
, num_written
);
757 current
->backing_dev_info
= NULL
;
758 return num_written
? num_written
: err
;
761 static int btrfs_sync_file(struct file
*file
,
762 struct dentry
*dentry
, int datasync
)
764 struct inode
*inode
= dentry
->d_inode
;
765 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
767 struct btrfs_trans_handle
*trans
;
770 * check the transaction that last modified this inode
771 * and see if its already been committed
773 mutex_lock(&root
->fs_info
->fs_mutex
);
774 if (!BTRFS_I(inode
)->last_trans
)
776 mutex_lock(&root
->fs_info
->trans_mutex
);
777 if (BTRFS_I(inode
)->last_trans
<=
778 root
->fs_info
->last_trans_committed
) {
779 BTRFS_I(inode
)->last_trans
= 0;
780 mutex_unlock(&root
->fs_info
->trans_mutex
);
783 mutex_unlock(&root
->fs_info
->trans_mutex
);
786 * ok we haven't committed the transaction yet, lets do a commit
788 trans
= btrfs_start_transaction(root
, 1);
793 ret
= btrfs_commit_transaction(trans
, root
);
795 mutex_unlock(&root
->fs_info
->fs_mutex
);
796 return ret
> 0 ? EIO
: ret
;
799 static struct vm_operations_struct btrfs_file_vm_ops
= {
800 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
801 .nopage
= filemap_nopage
,
802 .populate
= filemap_populate
,
804 .fault
= filemap_fault
,
806 .page_mkwrite
= btrfs_page_mkwrite
,
809 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
811 vma
->vm_ops
= &btrfs_file_vm_ops
;
816 struct file_operations btrfs_file_operations
= {
817 .llseek
= generic_file_llseek
,
818 .read
= do_sync_read
,
819 .aio_read
= generic_file_aio_read
,
820 .write
= btrfs_file_write
,
821 .mmap
= btrfs_file_mmap
,
822 .open
= generic_file_open
,
823 .fsync
= btrfs_sync_file
,
824 .unlocked_ioctl
= btrfs_ioctl
,
826 .compat_ioctl
= btrfs_ioctl
,