1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
23 #include <linux/slab.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <asm/byteorder.h>
27 #include <linux/swap.h>
28 #include <linux/pipe_fs_i.h>
29 #include <linux/mpage.h>
30 #include <linux/quotaops.h>
32 #define MLOG_MASK_PREFIX ML_FILE_IO
33 #include <cluster/masklog.h>
40 #include "extent_map.h"
47 #include "refcounttree.h"
49 #include "buffer_head_io.h"
51 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
52 struct buffer_head
*bh_result
, int create
)
56 struct ocfs2_dinode
*fe
= NULL
;
57 struct buffer_head
*bh
= NULL
;
58 struct buffer_head
*buffer_cache_bh
= NULL
;
59 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
62 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
63 (unsigned long long)iblock
, bh_result
, create
);
65 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
67 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
68 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
69 (unsigned long long)iblock
);
73 status
= ocfs2_read_inode_block(inode
, &bh
);
78 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
80 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
81 le32_to_cpu(fe
->i_clusters
))) {
82 mlog(ML_ERROR
, "block offset is outside the allocated size: "
83 "%llu\n", (unsigned long long)iblock
);
87 /* We don't use the page cache to create symlink data, so if
88 * need be, copy it over from the buffer cache. */
89 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
90 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
92 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
93 if (!buffer_cache_bh
) {
94 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
98 /* we haven't locked out transactions, so a commit
99 * could've happened. Since we've got a reference on
100 * the bh, even if it commits while we're doing the
101 * copy, the data is still good. */
102 if (buffer_jbd(buffer_cache_bh
)
103 && ocfs2_inode_is_new(inode
)) {
104 kaddr
= kmap_atomic(bh_result
->b_page
, KM_USER0
);
106 mlog(ML_ERROR
, "couldn't kmap!\n");
109 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
110 buffer_cache_bh
->b_data
,
112 kunmap_atomic(kaddr
, KM_USER0
);
113 set_buffer_uptodate(bh_result
);
115 brelse(buffer_cache_bh
);
118 map_bh(bh_result
, inode
->i_sb
,
119 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
130 int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
131 struct buffer_head
*bh_result
, int create
)
134 unsigned int ext_flags
;
135 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
136 u64 p_blkno
, count
, past_eof
;
137 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
139 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
140 (unsigned long long)iblock
, bh_result
, create
);
142 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
143 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
144 inode
, inode
->i_ino
);
146 if (S_ISLNK(inode
->i_mode
)) {
147 /* this always does I/O for some reason. */
148 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
152 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
155 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
156 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
157 (unsigned long long)p_blkno
);
161 if (max_blocks
< count
)
165 * ocfs2 never allocates in this function - the only time we
166 * need to use BH_New is when we're extending i_size on a file
167 * system which doesn't support holes, in which case BH_New
168 * allows block_prepare_write() to zero.
170 * If we see this on a sparse file system, then a truncate has
171 * raced us and removed the cluster. In this case, we clear
172 * the buffers dirty and uptodate bits and let the buffer code
173 * ignore it as a hole.
175 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
176 clear_buffer_dirty(bh_result
);
177 clear_buffer_uptodate(bh_result
);
181 /* Treat the unwritten extent as a hole for zeroing purposes. */
182 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
183 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
185 bh_result
->b_size
= count
<< inode
->i_blkbits
;
187 if (!ocfs2_sparse_alloc(osb
)) {
191 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
192 (unsigned long long)iblock
,
193 (unsigned long long)p_blkno
,
194 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
195 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
200 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
201 mlog(0, "Inode %lu, past_eof = %llu\n", inode
->i_ino
,
202 (unsigned long long)past_eof
);
204 if (create
&& (iblock
>= past_eof
))
205 set_buffer_new(bh_result
);
216 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
217 struct buffer_head
*di_bh
)
221 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
223 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
224 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
225 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
229 size
= i_size_read(inode
);
231 if (size
> PAGE_CACHE_SIZE
||
232 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
233 ocfs2_error(inode
->i_sb
,
234 "Inode %llu has with inline data has bad size: %Lu",
235 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
236 (unsigned long long)size
);
240 kaddr
= kmap_atomic(page
, KM_USER0
);
242 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
243 /* Clear the remaining part of the page */
244 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
245 flush_dcache_page(page
);
246 kunmap_atomic(kaddr
, KM_USER0
);
248 SetPageUptodate(page
);
253 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
256 struct buffer_head
*di_bh
= NULL
;
258 BUG_ON(!PageLocked(page
));
259 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
261 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
267 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
275 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
277 struct inode
*inode
= page
->mapping
->host
;
278 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
279 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
282 mlog_entry("(0x%p, %lu)\n", file
, (page
? page
->index
: 0));
284 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
286 if (ret
== AOP_TRUNCATED_PAGE
)
292 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
293 ret
= AOP_TRUNCATED_PAGE
;
294 goto out_inode_unlock
;
298 * i_size might have just been updated as we grabed the meta lock. We
299 * might now be discovering a truncate that hit on another node.
300 * block_read_full_page->get_block freaks out if it is asked to read
301 * beyond the end of a file, so we check here. Callers
302 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
303 * and notice that the page they just read isn't needed.
305 * XXX sys_readahead() seems to get that wrong?
307 if (start
>= i_size_read(inode
)) {
308 zero_user(page
, 0, PAGE_SIZE
);
309 SetPageUptodate(page
);
314 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
315 ret
= ocfs2_readpage_inline(inode
, page
);
317 ret
= block_read_full_page(page
, ocfs2_get_block
);
321 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
323 ocfs2_inode_unlock(inode
, 0);
332 * This is used only for read-ahead. Failures or difficult to handle
333 * situations are safe to ignore.
335 * Right now, we don't bother with BH_Boundary - in-inode extent lists
336 * are quite large (243 extents on 4k blocks), so most inodes don't
337 * grow out to a tree. If need be, detecting boundary extents could
338 * trivially be added in a future version of ocfs2_get_block().
340 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
341 struct list_head
*pages
, unsigned nr_pages
)
344 struct inode
*inode
= mapping
->host
;
345 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
350 * Use the nonblocking flag for the dlm code to avoid page
351 * lock inversion, but don't bother with retrying.
353 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
357 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
358 ocfs2_inode_unlock(inode
, 0);
363 * Don't bother with inline-data. There isn't anything
364 * to read-ahead in that case anyway...
366 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
370 * Check whether a remote node truncated this file - we just
371 * drop out in that case as it's not worth handling here.
373 last
= list_entry(pages
->prev
, struct page
, lru
);
374 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
375 if (start
>= i_size_read(inode
))
378 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
381 up_read(&oi
->ip_alloc_sem
);
382 ocfs2_inode_unlock(inode
, 0);
387 /* Note: Because we don't support holes, our allocation has
388 * already happened (allocation writes zeros to the file data)
389 * so we don't have to worry about ordered writes in
392 * ->writepage is called during the process of invalidating the page cache
393 * during blocked lock processing. It can't block on any cluster locks
394 * to during block mapping. It's relying on the fact that the block
395 * mapping can't have disappeared under the dirty pages that it is
396 * being asked to write back.
398 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
402 mlog_entry("(0x%p)\n", page
);
404 ret
= block_write_full_page(page
, ocfs2_get_block
, wbc
);
412 * This is called from ocfs2_write_zero_page() which has handled it's
413 * own cluster locking and has ensured allocation exists for those
414 * blocks to be written.
416 int ocfs2_prepare_write_nolock(struct inode
*inode
, struct page
*page
,
417 unsigned from
, unsigned to
)
421 ret
= block_prepare_write(page
, from
, to
, ocfs2_get_block
);
426 /* Taken from ext3. We don't necessarily need the full blown
427 * functionality yet, but IMHO it's better to cut and paste the whole
428 * thing so we can avoid introducing our own bugs (and easily pick up
429 * their fixes when they happen) --Mark */
430 int walk_page_buffers( handle_t
*handle
,
431 struct buffer_head
*head
,
435 int (*fn
)( handle_t
*handle
,
436 struct buffer_head
*bh
))
438 struct buffer_head
*bh
;
439 unsigned block_start
, block_end
;
440 unsigned blocksize
= head
->b_size
;
442 struct buffer_head
*next
;
444 for ( bh
= head
, block_start
= 0;
445 ret
== 0 && (bh
!= head
|| !block_start
);
446 block_start
= block_end
, bh
= next
)
448 next
= bh
->b_this_page
;
449 block_end
= block_start
+ blocksize
;
450 if (block_end
<= from
|| block_start
>= to
) {
451 if (partial
&& !buffer_uptodate(bh
))
455 err
= (*fn
)(handle
, bh
);
462 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
467 struct inode
*inode
= mapping
->host
;
469 mlog_entry("(block = %llu)\n", (unsigned long long)block
);
471 /* We don't need to lock journal system files, since they aren't
472 * accessed concurrently from multiple nodes.
474 if (!INODE_JOURNAL(inode
)) {
475 err
= ocfs2_inode_lock(inode
, NULL
, 0);
481 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
484 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
485 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
488 if (!INODE_JOURNAL(inode
)) {
489 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
490 ocfs2_inode_unlock(inode
, 0);
494 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
495 (unsigned long long)block
);
501 status
= err
? 0 : p_blkno
;
503 mlog_exit((int)status
);
509 * TODO: Make this into a generic get_blocks function.
511 * From do_direct_io in direct-io.c:
512 * "So what we do is to permit the ->get_blocks function to populate
513 * bh.b_size with the size of IO which is permitted at this offset and
516 * This function is called directly from get_more_blocks in direct-io.c.
518 * called like this: dio->get_blocks(dio->inode, fs_startblk,
519 * fs_count, map_bh, dio->rw == WRITE);
521 * Note that we never bother to allocate blocks here, and thus ignore the
524 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
525 struct buffer_head
*bh_result
, int create
)
528 u64 p_blkno
, inode_blocks
, contig_blocks
;
529 unsigned int ext_flags
;
530 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
531 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
533 /* This function won't even be called if the request isn't all
534 * nicely aligned and of the right size, so there's no need
535 * for us to check any of that. */
537 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
539 /* This figures out the size of the next contiguous block, and
540 * our logical offset */
541 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
542 &contig_blocks
, &ext_flags
);
544 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
545 (unsigned long long)iblock
);
550 /* We should already CoW the refcounted extent in case of create. */
551 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
554 * get_more_blocks() expects us to describe a hole by clearing
555 * the mapped bit on bh_result().
557 * Consider an unwritten extent as a hole.
559 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
560 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
562 clear_buffer_mapped(bh_result
);
564 /* make sure we don't map more than max_blocks blocks here as
565 that's all the kernel will handle at this point. */
566 if (max_blocks
< contig_blocks
)
567 contig_blocks
= max_blocks
;
568 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
574 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
575 * particularly interested in the aio/dio case. Like the core uses
576 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
577 * truncation on another.
579 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
584 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
587 /* this io's submitter should not have unlocked this before we could */
588 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
590 ocfs2_iocb_clear_rw_locked(iocb
);
592 level
= ocfs2_iocb_rw_locked_level(iocb
);
594 up_read(&inode
->i_alloc_sem
);
595 ocfs2_rw_unlock(inode
, level
);
599 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
600 * from ext3. PageChecked() bits have been removed as OCFS2 does not
601 * do journalled data.
603 static void ocfs2_invalidatepage(struct page
*page
, unsigned long offset
)
605 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
607 jbd2_journal_invalidatepage(journal
, page
, offset
);
610 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
612 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
614 if (!page_has_buffers(page
))
616 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
619 static ssize_t
ocfs2_direct_IO(int rw
,
621 const struct iovec
*iov
,
623 unsigned long nr_segs
)
625 struct file
*file
= iocb
->ki_filp
;
626 struct inode
*inode
= file
->f_path
.dentry
->d_inode
->i_mapping
->host
;
632 * Fallback to buffered I/O if we see an inode without
635 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
638 /* Fallback to buffered I/O if we are appending. */
639 if (i_size_read(inode
) <= offset
)
642 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
643 inode
->i_sb
->s_bdev
, iov
, offset
,
645 ocfs2_direct_IO_get_blocks
,
652 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
657 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
659 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
662 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
664 cluster_start
= cpos
% cpp
;
665 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
667 cluster_end
= cluster_start
+ osb
->s_clustersize
;
670 BUG_ON(cluster_start
> PAGE_SIZE
);
671 BUG_ON(cluster_end
> PAGE_SIZE
);
674 *start
= cluster_start
;
680 * 'from' and 'to' are the region in the page to avoid zeroing.
682 * If pagesize > clustersize, this function will avoid zeroing outside
683 * of the cluster boundary.
685 * from == to == 0 is code for "zero the entire cluster region"
687 static void ocfs2_clear_page_regions(struct page
*page
,
688 struct ocfs2_super
*osb
, u32 cpos
,
689 unsigned from
, unsigned to
)
692 unsigned int cluster_start
, cluster_end
;
694 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
696 kaddr
= kmap_atomic(page
, KM_USER0
);
699 if (from
> cluster_start
)
700 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
701 if (to
< cluster_end
)
702 memset(kaddr
+ to
, 0, cluster_end
- to
);
704 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
707 kunmap_atomic(kaddr
, KM_USER0
);
711 * Nonsparse file systems fully allocate before we get to the write
712 * code. This prevents ocfs2_write() from tagging the write as an
713 * allocating one, which means ocfs2_map_page_blocks() might try to
714 * read-in the blocks at the tail of our file. Avoid reading them by
715 * testing i_size against each block offset.
717 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
718 unsigned int block_start
)
720 u64 offset
= page_offset(page
) + block_start
;
722 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
725 if (i_size_read(inode
) > offset
)
732 * Some of this taken from block_prepare_write(). We already have our
733 * mapping by now though, and the entire write will be allocating or
734 * it won't, so not much need to use BH_New.
736 * This will also skip zeroing, which is handled externally.
738 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
739 struct inode
*inode
, unsigned int from
,
740 unsigned int to
, int new)
743 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
744 unsigned int block_end
, block_start
;
745 unsigned int bsize
= 1 << inode
->i_blkbits
;
747 if (!page_has_buffers(page
))
748 create_empty_buffers(page
, bsize
, 0);
750 head
= page_buffers(page
);
751 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
752 bh
= bh
->b_this_page
, block_start
+= bsize
) {
753 block_end
= block_start
+ bsize
;
755 clear_buffer_new(bh
);
758 * Ignore blocks outside of our i/o range -
759 * they may belong to unallocated clusters.
761 if (block_start
>= to
|| block_end
<= from
) {
762 if (PageUptodate(page
))
763 set_buffer_uptodate(bh
);
768 * For an allocating write with cluster size >= page
769 * size, we always write the entire page.
774 if (!buffer_mapped(bh
)) {
775 map_bh(bh
, inode
->i_sb
, *p_blkno
);
776 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
779 if (PageUptodate(page
)) {
780 if (!buffer_uptodate(bh
))
781 set_buffer_uptodate(bh
);
782 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
784 ocfs2_should_read_blk(inode
, page
, block_start
) &&
785 (block_start
< from
|| block_end
> to
)) {
786 ll_rw_block(READ
, 1, &bh
);
790 *p_blkno
= *p_blkno
+ 1;
794 * If we issued read requests - let them complete.
796 while(wait_bh
> wait
) {
797 wait_on_buffer(*--wait_bh
);
798 if (!buffer_uptodate(*wait_bh
))
802 if (ret
== 0 || !new)
806 * If we get -EIO above, zero out any newly allocated blocks
807 * to avoid exposing stale data.
812 block_end
= block_start
+ bsize
;
813 if (block_end
<= from
)
815 if (block_start
>= to
)
818 zero_user(page
, block_start
, bh
->b_size
);
819 set_buffer_uptodate(bh
);
820 mark_buffer_dirty(bh
);
823 block_start
= block_end
;
824 bh
= bh
->b_this_page
;
825 } while (bh
!= head
);
830 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
831 #define OCFS2_MAX_CTXT_PAGES 1
833 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
836 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
839 * Describe the state of a single cluster to be written to.
841 struct ocfs2_write_cluster_desc
{
845 * Give this a unique field because c_phys eventually gets
849 unsigned c_unwritten
;
850 unsigned c_needs_zero
;
853 struct ocfs2_write_ctxt
{
854 /* Logical cluster position / len of write */
858 /* First cluster allocated in a nonsparse extend */
859 u32 w_first_new_cpos
;
861 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
864 * This is true if page_size > cluster_size.
866 * It triggers a set of special cases during write which might
867 * have to deal with allocating writes to partial pages.
869 unsigned int w_large_pages
;
872 * Pages involved in this write.
874 * w_target_page is the page being written to by the user.
876 * w_pages is an array of pages which always contains
877 * w_target_page, and in the case of an allocating write with
878 * page_size < cluster size, it will contain zero'd and mapped
879 * pages adjacent to w_target_page which need to be written
880 * out in so that future reads from that region will get
883 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
884 unsigned int w_num_pages
;
885 struct page
*w_target_page
;
888 * ocfs2_write_end() uses this to know what the real range to
889 * write in the target should be.
891 unsigned int w_target_from
;
892 unsigned int w_target_to
;
895 * We could use journal_current_handle() but this is cleaner,
900 struct buffer_head
*w_di_bh
;
902 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
905 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
909 for(i
= 0; i
< num_pages
; i
++) {
911 unlock_page(pages
[i
]);
912 mark_page_accessed(pages
[i
]);
913 page_cache_release(pages
[i
]);
918 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
920 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
926 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
927 struct ocfs2_super
*osb
, loff_t pos
,
928 unsigned len
, struct buffer_head
*di_bh
)
931 struct ocfs2_write_ctxt
*wc
;
933 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
937 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
938 wc
->w_first_new_cpos
= UINT_MAX
;
939 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
940 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
944 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
945 wc
->w_large_pages
= 1;
947 wc
->w_large_pages
= 0;
949 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
957 * If a page has any new buffers, zero them out here, and mark them uptodate
958 * and dirty so they'll be written out (in order to prevent uninitialised
959 * block data from leaking). And clear the new bit.
961 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
963 unsigned int block_start
, block_end
;
964 struct buffer_head
*head
, *bh
;
966 BUG_ON(!PageLocked(page
));
967 if (!page_has_buffers(page
))
970 bh
= head
= page_buffers(page
);
973 block_end
= block_start
+ bh
->b_size
;
975 if (buffer_new(bh
)) {
976 if (block_end
> from
&& block_start
< to
) {
977 if (!PageUptodate(page
)) {
980 start
= max(from
, block_start
);
981 end
= min(to
, block_end
);
983 zero_user_segment(page
, start
, end
);
984 set_buffer_uptodate(bh
);
987 clear_buffer_new(bh
);
988 mark_buffer_dirty(bh
);
992 block_start
= block_end
;
993 bh
= bh
->b_this_page
;
994 } while (bh
!= head
);
998 * Only called when we have a failure during allocating write to write
999 * zero's to the newly allocated region.
1001 static void ocfs2_write_failure(struct inode
*inode
,
1002 struct ocfs2_write_ctxt
*wc
,
1003 loff_t user_pos
, unsigned user_len
)
1006 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1007 to
= user_pos
+ user_len
;
1008 struct page
*tmppage
;
1010 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1012 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1013 tmppage
= wc
->w_pages
[i
];
1015 if (page_has_buffers(tmppage
)) {
1016 if (ocfs2_should_order_data(inode
))
1017 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1019 block_commit_write(tmppage
, from
, to
);
1024 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1025 struct ocfs2_write_ctxt
*wc
,
1026 struct page
*page
, u32 cpos
,
1027 loff_t user_pos
, unsigned user_len
,
1031 unsigned int map_from
= 0, map_to
= 0;
1032 unsigned int cluster_start
, cluster_end
;
1033 unsigned int user_data_from
= 0, user_data_to
= 0;
1035 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1036 &cluster_start
, &cluster_end
);
1038 if (page
== wc
->w_target_page
) {
1039 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1040 map_to
= map_from
+ user_len
;
1043 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1044 cluster_start
, cluster_end
,
1047 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1048 map_from
, map_to
, new);
1054 user_data_from
= map_from
;
1055 user_data_to
= map_to
;
1057 map_from
= cluster_start
;
1058 map_to
= cluster_end
;
1062 * If we haven't allocated the new page yet, we
1063 * shouldn't be writing it out without copying user
1064 * data. This is likely a math error from the caller.
1068 map_from
= cluster_start
;
1069 map_to
= cluster_end
;
1071 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1072 cluster_start
, cluster_end
, new);
1080 * Parts of newly allocated pages need to be zero'd.
1082 * Above, we have also rewritten 'to' and 'from' - as far as
1083 * the rest of the function is concerned, the entire cluster
1084 * range inside of a page needs to be written.
1086 * We can skip this if the page is up to date - it's already
1087 * been zero'd from being read in as a hole.
1089 if (new && !PageUptodate(page
))
1090 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1091 cpos
, user_data_from
, user_data_to
);
1093 flush_dcache_page(page
);
1100 * This function will only grab one clusters worth of pages.
1102 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1103 struct ocfs2_write_ctxt
*wc
,
1104 u32 cpos
, loff_t user_pos
,
1105 unsigned user_len
, int new,
1106 struct page
*mmap_page
)
1109 unsigned long start
, target_index
, end_index
, index
;
1110 struct inode
*inode
= mapping
->host
;
1113 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1116 * Figure out how many pages we'll be manipulating here. For
1117 * non allocating write, we just change the one
1118 * page. Otherwise, we'll need a whole clusters worth. If we're
1119 * writing past i_size, we only need enough pages to cover the
1120 * last page of the write.
1123 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1124 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1126 * We need the index *past* the last page we could possibly
1127 * touch. This is the page past the end of the write or
1128 * i_size, whichever is greater.
1130 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1131 BUG_ON(last_byte
< 1);
1132 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1133 if ((start
+ wc
->w_num_pages
) > end_index
)
1134 wc
->w_num_pages
= end_index
- start
;
1136 wc
->w_num_pages
= 1;
1137 start
= target_index
;
1140 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1143 if (index
== target_index
&& mmap_page
) {
1145 * ocfs2_pagemkwrite() is a little different
1146 * and wants us to directly use the page
1149 lock_page(mmap_page
);
1151 if (mmap_page
->mapping
!= mapping
) {
1152 unlock_page(mmap_page
);
1154 * Sanity check - the locking in
1155 * ocfs2_pagemkwrite() should ensure
1156 * that this code doesn't trigger.
1163 page_cache_get(mmap_page
);
1164 wc
->w_pages
[i
] = mmap_page
;
1166 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1168 if (!wc
->w_pages
[i
]) {
1175 if (index
== target_index
)
1176 wc
->w_target_page
= wc
->w_pages
[i
];
1183 * Prepare a single cluster for write one cluster into the file.
1185 static int ocfs2_write_cluster(struct address_space
*mapping
,
1186 u32 phys
, unsigned int unwritten
,
1187 unsigned int should_zero
,
1188 struct ocfs2_alloc_context
*data_ac
,
1189 struct ocfs2_alloc_context
*meta_ac
,
1190 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1191 loff_t user_pos
, unsigned user_len
)
1194 u64 v_blkno
, p_blkno
;
1195 struct inode
*inode
= mapping
->host
;
1196 struct ocfs2_extent_tree et
;
1198 new = phys
== 0 ? 1 : 0;
1203 * This is safe to call with the page locks - it won't take
1204 * any additional semaphores or cluster locks.
1207 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1208 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1209 wc
->w_handle
, data_ac
,
1212 * This shouldn't happen because we must have already
1213 * calculated the correct meta data allocation required. The
1214 * internal tree allocation code should know how to increase
1215 * transaction credits itself.
1217 * If need be, we could handle -EAGAIN for a
1218 * RESTART_TRANS here.
1220 mlog_bug_on_msg(ret
== -EAGAIN
,
1221 "Inode %llu: EAGAIN return during allocation.\n",
1222 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1227 } else if (unwritten
) {
1228 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1230 ret
= ocfs2_mark_extent_written(inode
, &et
,
1231 wc
->w_handle
, cpos
, 1, phys
,
1232 meta_ac
, &wc
->w_dealloc
);
1240 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1242 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1245 * The only reason this should fail is due to an inability to
1246 * find the extent added.
1248 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1251 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1252 "at logical block %llu",
1253 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1254 (unsigned long long)v_blkno
);
1258 BUG_ON(p_blkno
== 0);
1260 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1263 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1264 wc
->w_pages
[i
], cpos
,
1275 * We only have cleanup to do in case of allocating write.
1278 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1285 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1286 struct ocfs2_alloc_context
*data_ac
,
1287 struct ocfs2_alloc_context
*meta_ac
,
1288 struct ocfs2_write_ctxt
*wc
,
1289 loff_t pos
, unsigned len
)
1293 unsigned int local_len
= len
;
1294 struct ocfs2_write_cluster_desc
*desc
;
1295 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1297 for (i
= 0; i
< wc
->w_clen
; i
++) {
1298 desc
= &wc
->w_desc
[i
];
1301 * We have to make sure that the total write passed in
1302 * doesn't extend past a single cluster.
1305 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1306 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1307 local_len
= osb
->s_clustersize
- cluster_off
;
1309 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1313 wc
, desc
->c_cpos
, pos
, local_len
);
1329 * ocfs2_write_end() wants to know which parts of the target page it
1330 * should complete the write on. It's easiest to compute them ahead of
1331 * time when a more complete view of the write is available.
1333 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1334 struct ocfs2_write_ctxt
*wc
,
1335 loff_t pos
, unsigned len
, int alloc
)
1337 struct ocfs2_write_cluster_desc
*desc
;
1339 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1340 wc
->w_target_to
= wc
->w_target_from
+ len
;
1346 * Allocating write - we may have different boundaries based
1347 * on page size and cluster size.
1349 * NOTE: We can no longer compute one value from the other as
1350 * the actual write length and user provided length may be
1354 if (wc
->w_large_pages
) {
1356 * We only care about the 1st and last cluster within
1357 * our range and whether they should be zero'd or not. Either
1358 * value may be extended out to the start/end of a
1359 * newly allocated cluster.
1361 desc
= &wc
->w_desc
[0];
1362 if (desc
->c_needs_zero
)
1363 ocfs2_figure_cluster_boundaries(osb
,
1368 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1369 if (desc
->c_needs_zero
)
1370 ocfs2_figure_cluster_boundaries(osb
,
1375 wc
->w_target_from
= 0;
1376 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1381 * Populate each single-cluster write descriptor in the write context
1382 * with information about the i/o to be done.
1384 * Returns the number of clusters that will have to be allocated, as
1385 * well as a worst case estimate of the number of extent records that
1386 * would have to be created during a write to an unwritten region.
1388 static int ocfs2_populate_write_desc(struct inode
*inode
,
1389 struct ocfs2_write_ctxt
*wc
,
1390 unsigned int *clusters_to_alloc
,
1391 unsigned int *extents_to_split
)
1394 struct ocfs2_write_cluster_desc
*desc
;
1395 unsigned int num_clusters
= 0;
1396 unsigned int ext_flags
= 0;
1400 *clusters_to_alloc
= 0;
1401 *extents_to_split
= 0;
1403 for (i
= 0; i
< wc
->w_clen
; i
++) {
1404 desc
= &wc
->w_desc
[i
];
1405 desc
->c_cpos
= wc
->w_cpos
+ i
;
1407 if (num_clusters
== 0) {
1409 * Need to look up the next extent record.
1411 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1412 &num_clusters
, &ext_flags
);
1418 /* We should already CoW the refcountd extent. */
1419 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1422 * Assume worst case - that we're writing in
1423 * the middle of the extent.
1425 * We can assume that the write proceeds from
1426 * left to right, in which case the extent
1427 * insert code is smart enough to coalesce the
1428 * next splits into the previous records created.
1430 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1431 *extents_to_split
= *extents_to_split
+ 2;
1434 * Only increment phys if it doesn't describe
1441 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1442 * file that got extended. w_first_new_cpos tells us
1443 * where the newly allocated clusters are so we can
1446 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1448 desc
->c_needs_zero
= 1;
1451 desc
->c_phys
= phys
;
1454 desc
->c_needs_zero
= 1;
1455 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1458 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1459 desc
->c_unwritten
= 1;
1460 desc
->c_needs_zero
= 1;
1471 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1472 struct inode
*inode
,
1473 struct ocfs2_write_ctxt
*wc
)
1476 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1479 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1481 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1488 * If we don't set w_num_pages then this page won't get unlocked
1489 * and freed on cleanup of the write context.
1491 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1492 wc
->w_num_pages
= 1;
1494 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1495 if (IS_ERR(handle
)) {
1496 ret
= PTR_ERR(handle
);
1501 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1502 OCFS2_JOURNAL_ACCESS_WRITE
);
1504 ocfs2_commit_trans(osb
, handle
);
1510 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1511 ocfs2_set_inode_data_inline(inode
, di
);
1513 if (!PageUptodate(page
)) {
1514 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1516 ocfs2_commit_trans(osb
, handle
);
1522 wc
->w_handle
= handle
;
1527 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1529 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1531 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1536 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1537 struct inode
*inode
, loff_t pos
,
1538 unsigned len
, struct page
*mmap_page
,
1539 struct ocfs2_write_ctxt
*wc
)
1541 int ret
, written
= 0;
1542 loff_t end
= pos
+ len
;
1543 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1544 struct ocfs2_dinode
*di
= NULL
;
1546 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1547 (unsigned long long)oi
->ip_blkno
, len
, (unsigned long long)pos
,
1548 oi
->ip_dyn_features
);
1551 * Handle inodes which already have inline data 1st.
1553 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1554 if (mmap_page
== NULL
&&
1555 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1556 goto do_inline_write
;
1559 * The write won't fit - we have to give this inode an
1560 * inline extent list now.
1562 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1569 * Check whether the inode can accept inline data.
1571 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1575 * Check whether the write can fit.
1577 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1579 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1583 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1590 * This signals to the caller that the data can be written
1595 return written
? written
: ret
;
1599 * This function only does anything for file systems which can't
1600 * handle sparse files.
1602 * What we want to do here is fill in any hole between the current end
1603 * of allocation and the end of our write. That way the rest of the
1604 * write path can treat it as an non-allocating write, which has no
1605 * special case code for sparse/nonsparse files.
1607 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
, loff_t pos
,
1609 struct ocfs2_write_ctxt
*wc
)
1612 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1613 loff_t newsize
= pos
+ len
;
1615 if (ocfs2_sparse_alloc(osb
))
1618 if (newsize
<= i_size_read(inode
))
1621 ret
= ocfs2_extend_no_holes(inode
, newsize
, pos
);
1625 wc
->w_first_new_cpos
=
1626 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1631 int ocfs2_write_begin_nolock(struct address_space
*mapping
,
1632 loff_t pos
, unsigned len
, unsigned flags
,
1633 struct page
**pagep
, void **fsdata
,
1634 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1636 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1637 unsigned int clusters_to_alloc
, extents_to_split
;
1638 struct ocfs2_write_ctxt
*wc
;
1639 struct inode
*inode
= mapping
->host
;
1640 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1641 struct ocfs2_dinode
*di
;
1642 struct ocfs2_alloc_context
*data_ac
= NULL
;
1643 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1645 struct ocfs2_extent_tree et
;
1647 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1653 if (ocfs2_supports_inline_data(osb
)) {
1654 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1666 ret
= ocfs2_expand_nonsparse_inode(inode
, pos
, len
, wc
);
1672 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1676 } else if (ret
== 1) {
1677 ret
= ocfs2_refcount_cow(inode
, di_bh
,
1678 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1685 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1692 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1695 * We set w_target_from, w_target_to here so that
1696 * ocfs2_write_end() knows which range in the target page to
1697 * write out. An allocation requires that we write the entire
1700 if (clusters_to_alloc
|| extents_to_split
) {
1702 * XXX: We are stretching the limits of
1703 * ocfs2_lock_allocators(). It greatly over-estimates
1704 * the work to be done.
1706 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1707 " clusters_to_add = %u, extents_to_split = %u\n",
1708 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1709 (long long)i_size_read(inode
), le32_to_cpu(di
->i_clusters
),
1710 clusters_to_alloc
, extents_to_split
);
1712 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1714 ret
= ocfs2_lock_allocators(inode
, &et
,
1715 clusters_to_alloc
, extents_to_split
,
1716 &data_ac
, &meta_ac
);
1722 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1729 * We have to zero sparse allocated clusters, unwritten extent clusters,
1730 * and non-sparse clusters we just extended. For non-sparse writes,
1731 * we know zeros will only be needed in the first and/or last cluster.
1733 if (clusters_to_alloc
|| extents_to_split
||
1734 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
1735 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
1736 cluster_of_pages
= 1;
1738 cluster_of_pages
= 0;
1740 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1742 handle
= ocfs2_start_trans(osb
, credits
);
1743 if (IS_ERR(handle
)) {
1744 ret
= PTR_ERR(handle
);
1749 wc
->w_handle
= handle
;
1751 if (clusters_to_alloc
) {
1752 ret
= dquot_alloc_space_nodirty(inode
,
1753 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1758 * We don't want this to fail in ocfs2_write_end(), so do it
1761 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1762 OCFS2_JOURNAL_ACCESS_WRITE
);
1769 * Fill our page array first. That way we've grabbed enough so
1770 * that we can zero and flush if we error after adding the
1773 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
1774 cluster_of_pages
, mmap_page
);
1780 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1788 ocfs2_free_alloc_context(data_ac
);
1790 ocfs2_free_alloc_context(meta_ac
);
1793 *pagep
= wc
->w_target_page
;
1797 if (clusters_to_alloc
)
1798 dquot_free_space(inode
,
1799 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1801 ocfs2_commit_trans(osb
, handle
);
1804 ocfs2_free_write_ctxt(wc
);
1807 ocfs2_free_alloc_context(data_ac
);
1809 ocfs2_free_alloc_context(meta_ac
);
1813 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1814 loff_t pos
, unsigned len
, unsigned flags
,
1815 struct page
**pagep
, void **fsdata
)
1818 struct buffer_head
*di_bh
= NULL
;
1819 struct inode
*inode
= mapping
->host
;
1821 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1828 * Take alloc sem here to prevent concurrent lookups. That way
1829 * the mapping, zeroing and tree manipulation within
1830 * ocfs2_write() will be safe against ->readpage(). This
1831 * should also serve to lock out allocation from a shared
1834 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1836 ret
= ocfs2_write_begin_nolock(mapping
, pos
, len
, flags
, pagep
,
1837 fsdata
, di_bh
, NULL
);
1848 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1851 ocfs2_inode_unlock(inode
, 1);
1856 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1857 unsigned len
, unsigned *copied
,
1858 struct ocfs2_dinode
*di
,
1859 struct ocfs2_write_ctxt
*wc
)
1863 if (unlikely(*copied
< len
)) {
1864 if (!PageUptodate(wc
->w_target_page
)) {
1870 kaddr
= kmap_atomic(wc
->w_target_page
, KM_USER0
);
1871 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1872 kunmap_atomic(kaddr
, KM_USER0
);
1874 mlog(0, "Data written to inode at offset %llu. "
1875 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1876 (unsigned long long)pos
, *copied
,
1877 le16_to_cpu(di
->id2
.i_data
.id_count
),
1878 le16_to_cpu(di
->i_dyn_features
));
1881 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1882 loff_t pos
, unsigned len
, unsigned copied
,
1883 struct page
*page
, void *fsdata
)
1886 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1887 struct inode
*inode
= mapping
->host
;
1888 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1889 struct ocfs2_write_ctxt
*wc
= fsdata
;
1890 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1891 handle_t
*handle
= wc
->w_handle
;
1892 struct page
*tmppage
;
1894 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1895 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1896 goto out_write_size
;
1899 if (unlikely(copied
< len
)) {
1900 if (!PageUptodate(wc
->w_target_page
))
1903 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
1906 flush_dcache_page(wc
->w_target_page
);
1908 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1909 tmppage
= wc
->w_pages
[i
];
1911 if (tmppage
== wc
->w_target_page
) {
1912 from
= wc
->w_target_from
;
1913 to
= wc
->w_target_to
;
1915 BUG_ON(from
> PAGE_CACHE_SIZE
||
1916 to
> PAGE_CACHE_SIZE
||
1920 * Pages adjacent to the target (if any) imply
1921 * a hole-filling write in which case we want
1922 * to flush their entire range.
1925 to
= PAGE_CACHE_SIZE
;
1928 if (page_has_buffers(tmppage
)) {
1929 if (ocfs2_should_order_data(inode
))
1930 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1931 block_commit_write(tmppage
, from
, to
);
1937 if (pos
> inode
->i_size
) {
1938 i_size_write(inode
, pos
);
1939 mark_inode_dirty(inode
);
1941 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1942 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
1943 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1944 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1945 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1946 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
1948 ocfs2_commit_trans(osb
, handle
);
1950 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
1952 ocfs2_free_write_ctxt(wc
);
1957 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
1958 loff_t pos
, unsigned len
, unsigned copied
,
1959 struct page
*page
, void *fsdata
)
1962 struct inode
*inode
= mapping
->host
;
1964 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
1966 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1967 ocfs2_inode_unlock(inode
, 1);
1972 const struct address_space_operations ocfs2_aops
= {
1973 .readpage
= ocfs2_readpage
,
1974 .readpages
= ocfs2_readpages
,
1975 .writepage
= ocfs2_writepage
,
1976 .write_begin
= ocfs2_write_begin
,
1977 .write_end
= ocfs2_write_end
,
1979 .sync_page
= block_sync_page
,
1980 .direct_IO
= ocfs2_direct_IO
,
1981 .invalidatepage
= ocfs2_invalidatepage
,
1982 .releasepage
= ocfs2_releasepage
,
1983 .migratepage
= buffer_migrate_page
,
1984 .is_partially_uptodate
= block_is_partially_uptodate
,
1985 .error_remove_page
= generic_error_remove_page
,