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 #include <cluster/masklog.h>
39 #include "extent_map.h"
46 #include "refcounttree.h"
47 #include "ocfs2_trace.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 trace_ocfs2_symlink_get_block(
63 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
64 (unsigned long long)iblock
, bh_result
, create
);
66 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
68 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
69 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
70 (unsigned long long)iblock
);
74 status
= ocfs2_read_inode_block(inode
, &bh
);
79 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
81 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
82 le32_to_cpu(fe
->i_clusters
))) {
84 mlog(ML_ERROR
, "block offset is outside the allocated size: "
85 "%llu\n", (unsigned long long)iblock
);
89 /* We don't use the page cache to create symlink data, so if
90 * need be, copy it over from the buffer cache. */
91 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
92 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
94 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
95 if (!buffer_cache_bh
) {
97 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
101 /* we haven't locked out transactions, so a commit
102 * could've happened. Since we've got a reference on
103 * the bh, even if it commits while we're doing the
104 * copy, the data is still good. */
105 if (buffer_jbd(buffer_cache_bh
)
106 && ocfs2_inode_is_new(inode
)) {
107 kaddr
= kmap_atomic(bh_result
->b_page
);
109 mlog(ML_ERROR
, "couldn't kmap!\n");
112 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
113 buffer_cache_bh
->b_data
,
115 kunmap_atomic(kaddr
);
116 set_buffer_uptodate(bh_result
);
118 brelse(buffer_cache_bh
);
121 map_bh(bh_result
, inode
->i_sb
,
122 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
132 int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
133 struct buffer_head
*bh_result
, int create
)
136 unsigned int ext_flags
;
137 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
138 u64 p_blkno
, count
, past_eof
;
139 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
141 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
142 (unsigned long long)iblock
, bh_result
, create
);
144 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
145 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
146 inode
, inode
->i_ino
);
148 if (S_ISLNK(inode
->i_mode
)) {
149 /* this always does I/O for some reason. */
150 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
154 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
157 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
158 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
159 (unsigned long long)p_blkno
);
163 if (max_blocks
< count
)
167 * ocfs2 never allocates in this function - the only time we
168 * need to use BH_New is when we're extending i_size on a file
169 * system which doesn't support holes, in which case BH_New
170 * allows __block_write_begin() to zero.
172 * If we see this on a sparse file system, then a truncate has
173 * raced us and removed the cluster. In this case, we clear
174 * the buffers dirty and uptodate bits and let the buffer code
175 * ignore it as a hole.
177 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
178 clear_buffer_dirty(bh_result
);
179 clear_buffer_uptodate(bh_result
);
183 /* Treat the unwritten extent as a hole for zeroing purposes. */
184 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
185 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
187 bh_result
->b_size
= count
<< inode
->i_blkbits
;
189 if (!ocfs2_sparse_alloc(osb
)) {
193 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
194 (unsigned long long)iblock
,
195 (unsigned long long)p_blkno
,
196 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
197 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
203 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
205 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
206 (unsigned long long)past_eof
);
207 if (create
&& (iblock
>= past_eof
))
208 set_buffer_new(bh_result
);
217 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
218 struct buffer_head
*di_bh
)
222 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
224 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
225 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
226 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
230 size
= i_size_read(inode
);
232 if (size
> PAGE_CACHE_SIZE
||
233 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
234 ocfs2_error(inode
->i_sb
,
235 "Inode %llu has with inline data has bad size: %Lu",
236 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
237 (unsigned long long)size
);
241 kaddr
= kmap_atomic(page
);
243 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
244 /* Clear the remaining part of the page */
245 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
246 flush_dcache_page(page
);
247 kunmap_atomic(kaddr
);
249 SetPageUptodate(page
);
254 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
257 struct buffer_head
*di_bh
= NULL
;
259 BUG_ON(!PageLocked(page
));
260 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
262 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
268 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
276 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
278 struct inode
*inode
= page
->mapping
->host
;
279 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
280 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
283 trace_ocfs2_readpage((unsigned long long)oi
->ip_blkno
,
284 (page
? page
->index
: 0));
286 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
288 if (ret
== AOP_TRUNCATED_PAGE
)
294 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
296 * Unlock the page and cycle ip_alloc_sem so that we don't
297 * busyloop waiting for ip_alloc_sem to unlock
299 ret
= AOP_TRUNCATED_PAGE
;
302 down_read(&oi
->ip_alloc_sem
);
303 up_read(&oi
->ip_alloc_sem
);
304 goto out_inode_unlock
;
308 * i_size might have just been updated as we grabed the meta lock. We
309 * might now be discovering a truncate that hit on another node.
310 * block_read_full_page->get_block freaks out if it is asked to read
311 * beyond the end of a file, so we check here. Callers
312 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
313 * and notice that the page they just read isn't needed.
315 * XXX sys_readahead() seems to get that wrong?
317 if (start
>= i_size_read(inode
)) {
318 zero_user(page
, 0, PAGE_SIZE
);
319 SetPageUptodate(page
);
324 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
325 ret
= ocfs2_readpage_inline(inode
, page
);
327 ret
= block_read_full_page(page
, ocfs2_get_block
);
331 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
333 ocfs2_inode_unlock(inode
, 0);
341 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore.
344 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of ocfs2_get_block().
349 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
350 struct list_head
*pages
, unsigned nr_pages
)
353 struct inode
*inode
= mapping
->host
;
354 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
359 * Use the nonblocking flag for the dlm code to avoid page
360 * lock inversion, but don't bother with retrying.
362 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
366 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
367 ocfs2_inode_unlock(inode
, 0);
372 * Don't bother with inline-data. There isn't anything
373 * to read-ahead in that case anyway...
375 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
379 * Check whether a remote node truncated this file - we just
380 * drop out in that case as it's not worth handling here.
382 last
= list_entry(pages
->prev
, struct page
, lru
);
383 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
384 if (start
>= i_size_read(inode
))
387 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
390 up_read(&oi
->ip_alloc_sem
);
391 ocfs2_inode_unlock(inode
, 0);
396 /* Note: Because we don't support holes, our allocation has
397 * already happened (allocation writes zeros to the file data)
398 * so we don't have to worry about ordered writes in
401 * ->writepage is called during the process of invalidating the page cache
402 * during blocked lock processing. It can't block on any cluster locks
403 * to during block mapping. It's relying on the fact that the block
404 * mapping can't have disappeared under the dirty pages that it is
405 * being asked to write back.
407 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
409 trace_ocfs2_writepage(
410 (unsigned long long)OCFS2_I(page
->mapping
->host
)->ip_blkno
,
413 return block_write_full_page(page
, ocfs2_get_block
, wbc
);
416 /* Taken from ext3. We don't necessarily need the full blown
417 * functionality yet, but IMHO it's better to cut and paste the whole
418 * thing so we can avoid introducing our own bugs (and easily pick up
419 * their fixes when they happen) --Mark */
420 int walk_page_buffers( handle_t
*handle
,
421 struct buffer_head
*head
,
425 int (*fn
)( handle_t
*handle
,
426 struct buffer_head
*bh
))
428 struct buffer_head
*bh
;
429 unsigned block_start
, block_end
;
430 unsigned blocksize
= head
->b_size
;
432 struct buffer_head
*next
;
434 for ( bh
= head
, block_start
= 0;
435 ret
== 0 && (bh
!= head
|| !block_start
);
436 block_start
= block_end
, bh
= next
)
438 next
= bh
->b_this_page
;
439 block_end
= block_start
+ blocksize
;
440 if (block_end
<= from
|| block_start
>= to
) {
441 if (partial
&& !buffer_uptodate(bh
))
445 err
= (*fn
)(handle
, bh
);
452 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
457 struct inode
*inode
= mapping
->host
;
459 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
460 (unsigned long long)block
);
462 /* We don't need to lock journal system files, since they aren't
463 * accessed concurrently from multiple nodes.
465 if (!INODE_JOURNAL(inode
)) {
466 err
= ocfs2_inode_lock(inode
, NULL
, 0);
472 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
475 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
476 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
479 if (!INODE_JOURNAL(inode
)) {
480 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
481 ocfs2_inode_unlock(inode
, 0);
485 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
486 (unsigned long long)block
);
492 status
= err
? 0 : p_blkno
;
498 * TODO: Make this into a generic get_blocks function.
500 * From do_direct_io in direct-io.c:
501 * "So what we do is to permit the ->get_blocks function to populate
502 * bh.b_size with the size of IO which is permitted at this offset and
505 * This function is called directly from get_more_blocks in direct-io.c.
507 * called like this: dio->get_blocks(dio->inode, fs_startblk,
508 * fs_count, map_bh, dio->rw == WRITE);
510 * Note that we never bother to allocate blocks here, and thus ignore the
513 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
514 struct buffer_head
*bh_result
, int create
)
517 u64 p_blkno
, inode_blocks
, contig_blocks
;
518 unsigned int ext_flags
;
519 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
520 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
522 /* This function won't even be called if the request isn't all
523 * nicely aligned and of the right size, so there's no need
524 * for us to check any of that. */
526 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
528 /* This figures out the size of the next contiguous block, and
529 * our logical offset */
530 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
531 &contig_blocks
, &ext_flags
);
533 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
534 (unsigned long long)iblock
);
539 /* We should already CoW the refcounted extent in case of create. */
540 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
543 * get_more_blocks() expects us to describe a hole by clearing
544 * the mapped bit on bh_result().
546 * Consider an unwritten extent as a hole.
548 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
549 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
551 clear_buffer_mapped(bh_result
);
553 /* make sure we don't map more than max_blocks blocks here as
554 that's all the kernel will handle at this point. */
555 if (max_blocks
< contig_blocks
)
556 contig_blocks
= max_blocks
;
557 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
563 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
564 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
565 * to protect io on one node from truncation on another.
567 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
572 struct inode
*inode
= file_inode(iocb
->ki_filp
);
575 /* this io's submitter should not have unlocked this before we could */
576 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
578 if (ocfs2_iocb_is_sem_locked(iocb
))
579 ocfs2_iocb_clear_sem_locked(iocb
);
581 if (ocfs2_iocb_is_unaligned_aio(iocb
)) {
582 ocfs2_iocb_clear_unaligned_aio(iocb
);
584 mutex_unlock(&OCFS2_I(inode
)->ip_unaligned_aio
);
587 ocfs2_iocb_clear_rw_locked(iocb
);
589 level
= ocfs2_iocb_rw_locked_level(iocb
);
590 ocfs2_rw_unlock(inode
, level
);
593 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
595 if (!page_has_buffers(page
))
597 return try_to_free_buffers(page
);
600 static ssize_t
ocfs2_direct_IO(int rw
,
602 const struct iovec
*iov
,
604 unsigned long nr_segs
)
606 struct file
*file
= iocb
->ki_filp
;
607 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
610 * Fallback to buffered I/O if we see an inode without
613 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
616 /* Fallback to buffered I/O if we are appending. */
617 if (i_size_read(inode
) <= offset
)
620 return __blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
621 iov
, offset
, nr_segs
,
622 ocfs2_direct_IO_get_blocks
,
623 ocfs2_dio_end_io
, NULL
, 0);
626 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
631 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
633 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
636 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
638 cluster_start
= cpos
% cpp
;
639 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
641 cluster_end
= cluster_start
+ osb
->s_clustersize
;
644 BUG_ON(cluster_start
> PAGE_SIZE
);
645 BUG_ON(cluster_end
> PAGE_SIZE
);
648 *start
= cluster_start
;
654 * 'from' and 'to' are the region in the page to avoid zeroing.
656 * If pagesize > clustersize, this function will avoid zeroing outside
657 * of the cluster boundary.
659 * from == to == 0 is code for "zero the entire cluster region"
661 static void ocfs2_clear_page_regions(struct page
*page
,
662 struct ocfs2_super
*osb
, u32 cpos
,
663 unsigned from
, unsigned to
)
666 unsigned int cluster_start
, cluster_end
;
668 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
670 kaddr
= kmap_atomic(page
);
673 if (from
> cluster_start
)
674 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
675 if (to
< cluster_end
)
676 memset(kaddr
+ to
, 0, cluster_end
- to
);
678 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
681 kunmap_atomic(kaddr
);
685 * Nonsparse file systems fully allocate before we get to the write
686 * code. This prevents ocfs2_write() from tagging the write as an
687 * allocating one, which means ocfs2_map_page_blocks() might try to
688 * read-in the blocks at the tail of our file. Avoid reading them by
689 * testing i_size against each block offset.
691 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
692 unsigned int block_start
)
694 u64 offset
= page_offset(page
) + block_start
;
696 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
699 if (i_size_read(inode
) > offset
)
706 * Some of this taken from __block_write_begin(). We already have our
707 * mapping by now though, and the entire write will be allocating or
708 * it won't, so not much need to use BH_New.
710 * This will also skip zeroing, which is handled externally.
712 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
713 struct inode
*inode
, unsigned int from
,
714 unsigned int to
, int new)
717 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
718 unsigned int block_end
, block_start
;
719 unsigned int bsize
= 1 << inode
->i_blkbits
;
721 if (!page_has_buffers(page
))
722 create_empty_buffers(page
, bsize
, 0);
724 head
= page_buffers(page
);
725 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
726 bh
= bh
->b_this_page
, block_start
+= bsize
) {
727 block_end
= block_start
+ bsize
;
729 clear_buffer_new(bh
);
732 * Ignore blocks outside of our i/o range -
733 * they may belong to unallocated clusters.
735 if (block_start
>= to
|| block_end
<= from
) {
736 if (PageUptodate(page
))
737 set_buffer_uptodate(bh
);
742 * For an allocating write with cluster size >= page
743 * size, we always write the entire page.
748 if (!buffer_mapped(bh
)) {
749 map_bh(bh
, inode
->i_sb
, *p_blkno
);
750 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
753 if (PageUptodate(page
)) {
754 if (!buffer_uptodate(bh
))
755 set_buffer_uptodate(bh
);
756 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
758 ocfs2_should_read_blk(inode
, page
, block_start
) &&
759 (block_start
< from
|| block_end
> to
)) {
760 ll_rw_block(READ
, 1, &bh
);
764 *p_blkno
= *p_blkno
+ 1;
768 * If we issued read requests - let them complete.
770 while(wait_bh
> wait
) {
771 wait_on_buffer(*--wait_bh
);
772 if (!buffer_uptodate(*wait_bh
))
776 if (ret
== 0 || !new)
780 * If we get -EIO above, zero out any newly allocated blocks
781 * to avoid exposing stale data.
786 block_end
= block_start
+ bsize
;
787 if (block_end
<= from
)
789 if (block_start
>= to
)
792 zero_user(page
, block_start
, bh
->b_size
);
793 set_buffer_uptodate(bh
);
794 mark_buffer_dirty(bh
);
797 block_start
= block_end
;
798 bh
= bh
->b_this_page
;
799 } while (bh
!= head
);
804 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
805 #define OCFS2_MAX_CTXT_PAGES 1
807 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
810 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
813 * Describe the state of a single cluster to be written to.
815 struct ocfs2_write_cluster_desc
{
819 * Give this a unique field because c_phys eventually gets
823 unsigned c_unwritten
;
824 unsigned c_needs_zero
;
827 struct ocfs2_write_ctxt
{
828 /* Logical cluster position / len of write */
832 /* First cluster allocated in a nonsparse extend */
833 u32 w_first_new_cpos
;
835 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
838 * This is true if page_size > cluster_size.
840 * It triggers a set of special cases during write which might
841 * have to deal with allocating writes to partial pages.
843 unsigned int w_large_pages
;
846 * Pages involved in this write.
848 * w_target_page is the page being written to by the user.
850 * w_pages is an array of pages which always contains
851 * w_target_page, and in the case of an allocating write with
852 * page_size < cluster size, it will contain zero'd and mapped
853 * pages adjacent to w_target_page which need to be written
854 * out in so that future reads from that region will get
857 unsigned int w_num_pages
;
858 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
859 struct page
*w_target_page
;
862 * w_target_locked is used for page_mkwrite path indicating no unlocking
863 * against w_target_page in ocfs2_write_end_nolock.
865 unsigned int w_target_locked
:1;
868 * ocfs2_write_end() uses this to know what the real range to
869 * write in the target should be.
871 unsigned int w_target_from
;
872 unsigned int w_target_to
;
875 * We could use journal_current_handle() but this is cleaner,
880 struct buffer_head
*w_di_bh
;
882 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
885 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
889 for(i
= 0; i
< num_pages
; i
++) {
891 unlock_page(pages
[i
]);
892 mark_page_accessed(pages
[i
]);
893 page_cache_release(pages
[i
]);
898 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
903 * w_target_locked is only set to true in the page_mkwrite() case.
904 * The intent is to allow us to lock the target page from write_begin()
905 * to write_end(). The caller must hold a ref on w_target_page.
907 if (wc
->w_target_locked
) {
908 BUG_ON(!wc
->w_target_page
);
909 for (i
= 0; i
< wc
->w_num_pages
; i
++) {
910 if (wc
->w_target_page
== wc
->w_pages
[i
]) {
911 wc
->w_pages
[i
] = NULL
;
915 mark_page_accessed(wc
->w_target_page
);
916 page_cache_release(wc
->w_target_page
);
918 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
924 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
925 struct ocfs2_super
*osb
, loff_t pos
,
926 unsigned len
, struct buffer_head
*di_bh
)
929 struct ocfs2_write_ctxt
*wc
;
931 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
935 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
936 wc
->w_first_new_cpos
= UINT_MAX
;
937 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
938 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
942 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
943 wc
->w_large_pages
= 1;
945 wc
->w_large_pages
= 0;
947 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
955 * If a page has any new buffers, zero them out here, and mark them uptodate
956 * and dirty so they'll be written out (in order to prevent uninitialised
957 * block data from leaking). And clear the new bit.
959 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
961 unsigned int block_start
, block_end
;
962 struct buffer_head
*head
, *bh
;
964 BUG_ON(!PageLocked(page
));
965 if (!page_has_buffers(page
))
968 bh
= head
= page_buffers(page
);
971 block_end
= block_start
+ bh
->b_size
;
973 if (buffer_new(bh
)) {
974 if (block_end
> from
&& block_start
< to
) {
975 if (!PageUptodate(page
)) {
978 start
= max(from
, block_start
);
979 end
= min(to
, block_end
);
981 zero_user_segment(page
, start
, end
);
982 set_buffer_uptodate(bh
);
985 clear_buffer_new(bh
);
986 mark_buffer_dirty(bh
);
990 block_start
= block_end
;
991 bh
= bh
->b_this_page
;
992 } while (bh
!= head
);
996 * Only called when we have a failure during allocating write to write
997 * zero's to the newly allocated region.
999 static void ocfs2_write_failure(struct inode
*inode
,
1000 struct ocfs2_write_ctxt
*wc
,
1001 loff_t user_pos
, unsigned user_len
)
1004 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1005 to
= user_pos
+ user_len
;
1006 struct page
*tmppage
;
1008 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1010 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1011 tmppage
= wc
->w_pages
[i
];
1013 if (page_has_buffers(tmppage
)) {
1014 if (ocfs2_should_order_data(inode
))
1015 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1017 block_commit_write(tmppage
, from
, to
);
1022 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1023 struct ocfs2_write_ctxt
*wc
,
1024 struct page
*page
, u32 cpos
,
1025 loff_t user_pos
, unsigned user_len
,
1029 unsigned int map_from
= 0, map_to
= 0;
1030 unsigned int cluster_start
, cluster_end
;
1031 unsigned int user_data_from
= 0, user_data_to
= 0;
1033 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1034 &cluster_start
, &cluster_end
);
1036 /* treat the write as new if the a hole/lseek spanned across
1037 * the page boundary.
1039 new = new | ((i_size_read(inode
) <= page_offset(page
)) &&
1040 (page_offset(page
) <= user_pos
));
1042 if (page
== wc
->w_target_page
) {
1043 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1044 map_to
= map_from
+ user_len
;
1047 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1048 cluster_start
, cluster_end
,
1051 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1052 map_from
, map_to
, new);
1058 user_data_from
= map_from
;
1059 user_data_to
= map_to
;
1061 map_from
= cluster_start
;
1062 map_to
= cluster_end
;
1066 * If we haven't allocated the new page yet, we
1067 * shouldn't be writing it out without copying user
1068 * data. This is likely a math error from the caller.
1072 map_from
= cluster_start
;
1073 map_to
= cluster_end
;
1075 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1076 cluster_start
, cluster_end
, new);
1084 * Parts of newly allocated pages need to be zero'd.
1086 * Above, we have also rewritten 'to' and 'from' - as far as
1087 * the rest of the function is concerned, the entire cluster
1088 * range inside of a page needs to be written.
1090 * We can skip this if the page is up to date - it's already
1091 * been zero'd from being read in as a hole.
1093 if (new && !PageUptodate(page
))
1094 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1095 cpos
, user_data_from
, user_data_to
);
1097 flush_dcache_page(page
);
1104 * This function will only grab one clusters worth of pages.
1106 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1107 struct ocfs2_write_ctxt
*wc
,
1108 u32 cpos
, loff_t user_pos
,
1109 unsigned user_len
, int new,
1110 struct page
*mmap_page
)
1113 unsigned long start
, target_index
, end_index
, index
;
1114 struct inode
*inode
= mapping
->host
;
1117 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1120 * Figure out how many pages we'll be manipulating here. For
1121 * non allocating write, we just change the one
1122 * page. Otherwise, we'll need a whole clusters worth. If we're
1123 * writing past i_size, we only need enough pages to cover the
1124 * last page of the write.
1127 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1128 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1130 * We need the index *past* the last page we could possibly
1131 * touch. This is the page past the end of the write or
1132 * i_size, whichever is greater.
1134 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1135 BUG_ON(last_byte
< 1);
1136 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1137 if ((start
+ wc
->w_num_pages
) > end_index
)
1138 wc
->w_num_pages
= end_index
- start
;
1140 wc
->w_num_pages
= 1;
1141 start
= target_index
;
1144 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1147 if (index
== target_index
&& mmap_page
) {
1149 * ocfs2_pagemkwrite() is a little different
1150 * and wants us to directly use the page
1153 lock_page(mmap_page
);
1155 /* Exit and let the caller retry */
1156 if (mmap_page
->mapping
!= mapping
) {
1157 WARN_ON(mmap_page
->mapping
);
1158 unlock_page(mmap_page
);
1163 page_cache_get(mmap_page
);
1164 wc
->w_pages
[i
] = mmap_page
;
1165 wc
->w_target_locked
= true;
1167 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1169 if (!wc
->w_pages
[i
]) {
1175 wait_for_stable_page(wc
->w_pages
[i
]);
1177 if (index
== target_index
)
1178 wc
->w_target_page
= wc
->w_pages
[i
];
1182 wc
->w_target_locked
= false;
1187 * Prepare a single cluster for write one cluster into the file.
1189 static int ocfs2_write_cluster(struct address_space
*mapping
,
1190 u32 phys
, unsigned int unwritten
,
1191 unsigned int should_zero
,
1192 struct ocfs2_alloc_context
*data_ac
,
1193 struct ocfs2_alloc_context
*meta_ac
,
1194 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1195 loff_t user_pos
, unsigned user_len
)
1198 u64 v_blkno
, p_blkno
;
1199 struct inode
*inode
= mapping
->host
;
1200 struct ocfs2_extent_tree et
;
1202 new = phys
== 0 ? 1 : 0;
1207 * This is safe to call with the page locks - it won't take
1208 * any additional semaphores or cluster locks.
1211 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1212 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1213 wc
->w_handle
, data_ac
,
1216 * This shouldn't happen because we must have already
1217 * calculated the correct meta data allocation required. The
1218 * internal tree allocation code should know how to increase
1219 * transaction credits itself.
1221 * If need be, we could handle -EAGAIN for a
1222 * RESTART_TRANS here.
1224 mlog_bug_on_msg(ret
== -EAGAIN
,
1225 "Inode %llu: EAGAIN return during allocation.\n",
1226 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1231 } else if (unwritten
) {
1232 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1234 ret
= ocfs2_mark_extent_written(inode
, &et
,
1235 wc
->w_handle
, cpos
, 1, phys
,
1236 meta_ac
, &wc
->w_dealloc
);
1244 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1246 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1249 * The only reason this should fail is due to an inability to
1250 * find the extent added.
1252 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1255 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1256 "at logical block %llu",
1257 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1258 (unsigned long long)v_blkno
);
1262 BUG_ON(p_blkno
== 0);
1264 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1267 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1268 wc
->w_pages
[i
], cpos
,
1279 * We only have cleanup to do in case of allocating write.
1282 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1289 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1290 struct ocfs2_alloc_context
*data_ac
,
1291 struct ocfs2_alloc_context
*meta_ac
,
1292 struct ocfs2_write_ctxt
*wc
,
1293 loff_t pos
, unsigned len
)
1297 unsigned int local_len
= len
;
1298 struct ocfs2_write_cluster_desc
*desc
;
1299 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1301 for (i
= 0; i
< wc
->w_clen
; i
++) {
1302 desc
= &wc
->w_desc
[i
];
1305 * We have to make sure that the total write passed in
1306 * doesn't extend past a single cluster.
1309 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1310 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1311 local_len
= osb
->s_clustersize
- cluster_off
;
1313 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1317 wc
, desc
->c_cpos
, pos
, local_len
);
1333 * ocfs2_write_end() wants to know which parts of the target page it
1334 * should complete the write on. It's easiest to compute them ahead of
1335 * time when a more complete view of the write is available.
1337 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1338 struct ocfs2_write_ctxt
*wc
,
1339 loff_t pos
, unsigned len
, int alloc
)
1341 struct ocfs2_write_cluster_desc
*desc
;
1343 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1344 wc
->w_target_to
= wc
->w_target_from
+ len
;
1350 * Allocating write - we may have different boundaries based
1351 * on page size and cluster size.
1353 * NOTE: We can no longer compute one value from the other as
1354 * the actual write length and user provided length may be
1358 if (wc
->w_large_pages
) {
1360 * We only care about the 1st and last cluster within
1361 * our range and whether they should be zero'd or not. Either
1362 * value may be extended out to the start/end of a
1363 * newly allocated cluster.
1365 desc
= &wc
->w_desc
[0];
1366 if (desc
->c_needs_zero
)
1367 ocfs2_figure_cluster_boundaries(osb
,
1372 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1373 if (desc
->c_needs_zero
)
1374 ocfs2_figure_cluster_boundaries(osb
,
1379 wc
->w_target_from
= 0;
1380 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1385 * Populate each single-cluster write descriptor in the write context
1386 * with information about the i/o to be done.
1388 * Returns the number of clusters that will have to be allocated, as
1389 * well as a worst case estimate of the number of extent records that
1390 * would have to be created during a write to an unwritten region.
1392 static int ocfs2_populate_write_desc(struct inode
*inode
,
1393 struct ocfs2_write_ctxt
*wc
,
1394 unsigned int *clusters_to_alloc
,
1395 unsigned int *extents_to_split
)
1398 struct ocfs2_write_cluster_desc
*desc
;
1399 unsigned int num_clusters
= 0;
1400 unsigned int ext_flags
= 0;
1404 *clusters_to_alloc
= 0;
1405 *extents_to_split
= 0;
1407 for (i
= 0; i
< wc
->w_clen
; i
++) {
1408 desc
= &wc
->w_desc
[i
];
1409 desc
->c_cpos
= wc
->w_cpos
+ i
;
1411 if (num_clusters
== 0) {
1413 * Need to look up the next extent record.
1415 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1416 &num_clusters
, &ext_flags
);
1422 /* We should already CoW the refcountd extent. */
1423 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1426 * Assume worst case - that we're writing in
1427 * the middle of the extent.
1429 * We can assume that the write proceeds from
1430 * left to right, in which case the extent
1431 * insert code is smart enough to coalesce the
1432 * next splits into the previous records created.
1434 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1435 *extents_to_split
= *extents_to_split
+ 2;
1438 * Only increment phys if it doesn't describe
1445 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1446 * file that got extended. w_first_new_cpos tells us
1447 * where the newly allocated clusters are so we can
1450 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1452 desc
->c_needs_zero
= 1;
1455 desc
->c_phys
= phys
;
1458 desc
->c_needs_zero
= 1;
1459 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1462 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1463 desc
->c_unwritten
= 1;
1464 desc
->c_needs_zero
= 1;
1475 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1476 struct inode
*inode
,
1477 struct ocfs2_write_ctxt
*wc
)
1480 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1483 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1485 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1492 * If we don't set w_num_pages then this page won't get unlocked
1493 * and freed on cleanup of the write context.
1495 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1496 wc
->w_num_pages
= 1;
1498 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1499 if (IS_ERR(handle
)) {
1500 ret
= PTR_ERR(handle
);
1505 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1506 OCFS2_JOURNAL_ACCESS_WRITE
);
1508 ocfs2_commit_trans(osb
, handle
);
1514 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1515 ocfs2_set_inode_data_inline(inode
, di
);
1517 if (!PageUptodate(page
)) {
1518 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1520 ocfs2_commit_trans(osb
, handle
);
1526 wc
->w_handle
= handle
;
1531 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1533 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1535 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1540 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1541 struct inode
*inode
, loff_t pos
,
1542 unsigned len
, struct page
*mmap_page
,
1543 struct ocfs2_write_ctxt
*wc
)
1545 int ret
, written
= 0;
1546 loff_t end
= pos
+ len
;
1547 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1548 struct ocfs2_dinode
*di
= NULL
;
1550 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi
->ip_blkno
,
1551 len
, (unsigned long long)pos
,
1552 oi
->ip_dyn_features
);
1555 * Handle inodes which already have inline data 1st.
1557 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1558 if (mmap_page
== NULL
&&
1559 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1560 goto do_inline_write
;
1563 * The write won't fit - we have to give this inode an
1564 * inline extent list now.
1566 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1573 * Check whether the inode can accept inline data.
1575 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1579 * Check whether the write can fit.
1581 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1583 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1587 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1594 * This signals to the caller that the data can be written
1599 return written
? written
: ret
;
1603 * This function only does anything for file systems which can't
1604 * handle sparse files.
1606 * What we want to do here is fill in any hole between the current end
1607 * of allocation and the end of our write. That way the rest of the
1608 * write path can treat it as an non-allocating write, which has no
1609 * special case code for sparse/nonsparse files.
1611 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1612 struct buffer_head
*di_bh
,
1613 loff_t pos
, unsigned len
,
1614 struct ocfs2_write_ctxt
*wc
)
1617 loff_t newsize
= pos
+ len
;
1619 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1621 if (newsize
<= i_size_read(inode
))
1624 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1628 wc
->w_first_new_cpos
=
1629 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1634 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
1639 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1640 if (pos
> i_size_read(inode
))
1641 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
1647 * Try to flush truncate logs if we can free enough clusters from it.
1648 * As for return value, "< 0" means error, "0" no space and "1" means
1649 * we have freed enough spaces and let the caller try to allocate again.
1651 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
1652 unsigned int needed
)
1656 unsigned int truncated_clusters
;
1658 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
1659 truncated_clusters
= osb
->truncated_clusters
;
1660 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
1663 * Check whether we can succeed in allocating if we free
1666 if (truncated_clusters
< needed
)
1669 ret
= ocfs2_flush_truncate_log(osb
);
1675 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
1676 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
1683 int ocfs2_write_begin_nolock(struct file
*filp
,
1684 struct address_space
*mapping
,
1685 loff_t pos
, unsigned len
, unsigned flags
,
1686 struct page
**pagep
, void **fsdata
,
1687 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1689 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1690 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
1691 struct ocfs2_write_ctxt
*wc
;
1692 struct inode
*inode
= mapping
->host
;
1693 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1694 struct ocfs2_dinode
*di
;
1695 struct ocfs2_alloc_context
*data_ac
= NULL
;
1696 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1698 struct ocfs2_extent_tree et
;
1699 int try_free
= 1, ret1
;
1702 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1708 if (ocfs2_supports_inline_data(osb
)) {
1709 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1721 if (ocfs2_sparse_alloc(osb
))
1722 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
1724 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
1731 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1735 } else if (ret
== 1) {
1736 clusters_need
= wc
->w_clen
;
1737 ret
= ocfs2_refcount_cow(inode
, di_bh
,
1738 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1745 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1751 clusters_need
+= clusters_to_alloc
;
1753 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1755 trace_ocfs2_write_begin_nolock(
1756 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1757 (long long)i_size_read(inode
),
1758 le32_to_cpu(di
->i_clusters
),
1759 pos
, len
, flags
, mmap_page
,
1760 clusters_to_alloc
, extents_to_split
);
1763 * We set w_target_from, w_target_to here so that
1764 * ocfs2_write_end() knows which range in the target page to
1765 * write out. An allocation requires that we write the entire
1768 if (clusters_to_alloc
|| extents_to_split
) {
1770 * XXX: We are stretching the limits of
1771 * ocfs2_lock_allocators(). It greatly over-estimates
1772 * the work to be done.
1774 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1776 ret
= ocfs2_lock_allocators(inode
, &et
,
1777 clusters_to_alloc
, extents_to_split
,
1778 &data_ac
, &meta_ac
);
1785 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
1787 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1793 * We have to zero sparse allocated clusters, unwritten extent clusters,
1794 * and non-sparse clusters we just extended. For non-sparse writes,
1795 * we know zeros will only be needed in the first and/or last cluster.
1797 if (clusters_to_alloc
|| extents_to_split
||
1798 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
1799 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
1800 cluster_of_pages
= 1;
1802 cluster_of_pages
= 0;
1804 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1806 handle
= ocfs2_start_trans(osb
, credits
);
1807 if (IS_ERR(handle
)) {
1808 ret
= PTR_ERR(handle
);
1813 wc
->w_handle
= handle
;
1815 if (clusters_to_alloc
) {
1816 ret
= dquot_alloc_space_nodirty(inode
,
1817 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1822 * We don't want this to fail in ocfs2_write_end(), so do it
1825 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1826 OCFS2_JOURNAL_ACCESS_WRITE
);
1833 * Fill our page array first. That way we've grabbed enough so
1834 * that we can zero and flush if we error after adding the
1837 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
1838 cluster_of_pages
, mmap_page
);
1839 if (ret
&& ret
!= -EAGAIN
) {
1845 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
1846 * the target page. In this case, we exit with no error and no target
1847 * page. This will trigger the caller, page_mkwrite(), to re-try
1850 if (ret
== -EAGAIN
) {
1851 BUG_ON(wc
->w_target_page
);
1856 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1864 ocfs2_free_alloc_context(data_ac
);
1866 ocfs2_free_alloc_context(meta_ac
);
1869 *pagep
= wc
->w_target_page
;
1873 if (clusters_to_alloc
)
1874 dquot_free_space(inode
,
1875 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1877 ocfs2_commit_trans(osb
, handle
);
1880 ocfs2_free_write_ctxt(wc
);
1883 ocfs2_free_alloc_context(data_ac
);
1887 ocfs2_free_alloc_context(meta_ac
);
1891 if (ret
== -ENOSPC
&& try_free
) {
1893 * Try to free some truncate log so that we can have enough
1894 * clusters to allocate.
1898 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
1909 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1910 loff_t pos
, unsigned len
, unsigned flags
,
1911 struct page
**pagep
, void **fsdata
)
1914 struct buffer_head
*di_bh
= NULL
;
1915 struct inode
*inode
= mapping
->host
;
1917 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1924 * Take alloc sem here to prevent concurrent lookups. That way
1925 * the mapping, zeroing and tree manipulation within
1926 * ocfs2_write() will be safe against ->readpage(). This
1927 * should also serve to lock out allocation from a shared
1930 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1932 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
1933 fsdata
, di_bh
, NULL
);
1944 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1947 ocfs2_inode_unlock(inode
, 1);
1952 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1953 unsigned len
, unsigned *copied
,
1954 struct ocfs2_dinode
*di
,
1955 struct ocfs2_write_ctxt
*wc
)
1959 if (unlikely(*copied
< len
)) {
1960 if (!PageUptodate(wc
->w_target_page
)) {
1966 kaddr
= kmap_atomic(wc
->w_target_page
);
1967 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1968 kunmap_atomic(kaddr
);
1970 trace_ocfs2_write_end_inline(
1971 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1972 (unsigned long long)pos
, *copied
,
1973 le16_to_cpu(di
->id2
.i_data
.id_count
),
1974 le16_to_cpu(di
->i_dyn_features
));
1977 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1978 loff_t pos
, unsigned len
, unsigned copied
,
1979 struct page
*page
, void *fsdata
)
1982 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1983 struct inode
*inode
= mapping
->host
;
1984 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1985 struct ocfs2_write_ctxt
*wc
= fsdata
;
1986 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1987 handle_t
*handle
= wc
->w_handle
;
1988 struct page
*tmppage
;
1990 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1991 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1992 goto out_write_size
;
1995 if (unlikely(copied
< len
)) {
1996 if (!PageUptodate(wc
->w_target_page
))
1999 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
2002 flush_dcache_page(wc
->w_target_page
);
2004 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
2005 tmppage
= wc
->w_pages
[i
];
2007 if (tmppage
== wc
->w_target_page
) {
2008 from
= wc
->w_target_from
;
2009 to
= wc
->w_target_to
;
2011 BUG_ON(from
> PAGE_CACHE_SIZE
||
2012 to
> PAGE_CACHE_SIZE
||
2016 * Pages adjacent to the target (if any) imply
2017 * a hole-filling write in which case we want
2018 * to flush their entire range.
2021 to
= PAGE_CACHE_SIZE
;
2024 if (page_has_buffers(tmppage
)) {
2025 if (ocfs2_should_order_data(inode
))
2026 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
2027 block_commit_write(tmppage
, from
, to
);
2033 if (pos
> i_size_read(inode
)) {
2034 i_size_write(inode
, pos
);
2035 mark_inode_dirty(inode
);
2037 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
2038 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
2039 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2040 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
2041 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2042 ocfs2_update_inode_fsync_trans(handle
, inode
, 1);
2043 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2045 ocfs2_commit_trans(osb
, handle
);
2047 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2049 ocfs2_free_write_ctxt(wc
);
2054 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2055 loff_t pos
, unsigned len
, unsigned copied
,
2056 struct page
*page
, void *fsdata
)
2059 struct inode
*inode
= mapping
->host
;
2061 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2063 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2064 ocfs2_inode_unlock(inode
, 1);
2069 const struct address_space_operations ocfs2_aops
= {
2070 .readpage
= ocfs2_readpage
,
2071 .readpages
= ocfs2_readpages
,
2072 .writepage
= ocfs2_writepage
,
2073 .write_begin
= ocfs2_write_begin
,
2074 .write_end
= ocfs2_write_end
,
2076 .direct_IO
= ocfs2_direct_IO
,
2077 .invalidatepage
= block_invalidatepage
,
2078 .releasepage
= ocfs2_releasepage
,
2079 .migratepage
= buffer_migrate_page
,
2080 .is_partially_uptodate
= block_is_partially_uptodate
,
2081 .error_remove_page
= generic_error_remove_page
,