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(0, "(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
);
129 int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
130 struct buffer_head
*bh_result
, int create
)
133 unsigned int ext_flags
;
134 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
135 u64 p_blkno
, count
, past_eof
;
136 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
138 mlog(0, "(0x%p, %llu, 0x%p, %d)\n", inode
,
139 (unsigned long long)iblock
, bh_result
, create
);
141 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
142 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
143 inode
, inode
->i_ino
);
145 if (S_ISLNK(inode
->i_mode
)) {
146 /* this always does I/O for some reason. */
147 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
151 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
154 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
155 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
156 (unsigned long long)p_blkno
);
160 if (max_blocks
< count
)
164 * ocfs2 never allocates in this function - the only time we
165 * need to use BH_New is when we're extending i_size on a file
166 * system which doesn't support holes, in which case BH_New
167 * allows __block_write_begin() to zero.
169 * If we see this on a sparse file system, then a truncate has
170 * raced us and removed the cluster. In this case, we clear
171 * the buffers dirty and uptodate bits and let the buffer code
172 * ignore it as a hole.
174 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
175 clear_buffer_dirty(bh_result
);
176 clear_buffer_uptodate(bh_result
);
180 /* Treat the unwritten extent as a hole for zeroing purposes. */
181 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
182 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
184 bh_result
->b_size
= count
<< inode
->i_blkbits
;
186 if (!ocfs2_sparse_alloc(osb
)) {
190 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
191 (unsigned long long)iblock
,
192 (unsigned long long)p_blkno
,
193 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
194 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
);
203 if (create
&& (iblock
>= past_eof
))
204 set_buffer_new(bh_result
);
213 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
214 struct buffer_head
*di_bh
)
218 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
220 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
221 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
222 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
226 size
= i_size_read(inode
);
228 if (size
> PAGE_CACHE_SIZE
||
229 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
230 ocfs2_error(inode
->i_sb
,
231 "Inode %llu has with inline data has bad size: %Lu",
232 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
233 (unsigned long long)size
);
237 kaddr
= kmap_atomic(page
, KM_USER0
);
239 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
240 /* Clear the remaining part of the page */
241 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
242 flush_dcache_page(page
);
243 kunmap_atomic(kaddr
, KM_USER0
);
245 SetPageUptodate(page
);
250 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
253 struct buffer_head
*di_bh
= NULL
;
255 BUG_ON(!PageLocked(page
));
256 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
258 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
264 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
272 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
274 struct inode
*inode
= page
->mapping
->host
;
275 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
276 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
279 mlog(0, "(0x%p, %lu)\n", file
, (page
? page
->index
: 0));
281 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
283 if (ret
== AOP_TRUNCATED_PAGE
)
289 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
290 ret
= AOP_TRUNCATED_PAGE
;
291 goto out_inode_unlock
;
295 * i_size might have just been updated as we grabed the meta lock. We
296 * might now be discovering a truncate that hit on another node.
297 * block_read_full_page->get_block freaks out if it is asked to read
298 * beyond the end of a file, so we check here. Callers
299 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
300 * and notice that the page they just read isn't needed.
302 * XXX sys_readahead() seems to get that wrong?
304 if (start
>= i_size_read(inode
)) {
305 zero_user(page
, 0, PAGE_SIZE
);
306 SetPageUptodate(page
);
311 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
312 ret
= ocfs2_readpage_inline(inode
, page
);
314 ret
= block_read_full_page(page
, ocfs2_get_block
);
318 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
320 ocfs2_inode_unlock(inode
, 0);
328 * This is used only for read-ahead. Failures or difficult to handle
329 * situations are safe to ignore.
331 * Right now, we don't bother with BH_Boundary - in-inode extent lists
332 * are quite large (243 extents on 4k blocks), so most inodes don't
333 * grow out to a tree. If need be, detecting boundary extents could
334 * trivially be added in a future version of ocfs2_get_block().
336 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
337 struct list_head
*pages
, unsigned nr_pages
)
340 struct inode
*inode
= mapping
->host
;
341 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
346 * Use the nonblocking flag for the dlm code to avoid page
347 * lock inversion, but don't bother with retrying.
349 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
353 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
354 ocfs2_inode_unlock(inode
, 0);
359 * Don't bother with inline-data. There isn't anything
360 * to read-ahead in that case anyway...
362 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
366 * Check whether a remote node truncated this file - we just
367 * drop out in that case as it's not worth handling here.
369 last
= list_entry(pages
->prev
, struct page
, lru
);
370 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
371 if (start
>= i_size_read(inode
))
374 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
377 up_read(&oi
->ip_alloc_sem
);
378 ocfs2_inode_unlock(inode
, 0);
383 /* Note: Because we don't support holes, our allocation has
384 * already happened (allocation writes zeros to the file data)
385 * so we don't have to worry about ordered writes in
388 * ->writepage is called during the process of invalidating the page cache
389 * during blocked lock processing. It can't block on any cluster locks
390 * to during block mapping. It's relying on the fact that the block
391 * mapping can't have disappeared under the dirty pages that it is
392 * being asked to write back.
394 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
398 mlog(0, "(0x%p)\n", page
);
400 ret
= block_write_full_page(page
, ocfs2_get_block
, wbc
);
405 /* Taken from ext3. We don't necessarily need the full blown
406 * functionality yet, but IMHO it's better to cut and paste the whole
407 * thing so we can avoid introducing our own bugs (and easily pick up
408 * their fixes when they happen) --Mark */
409 int walk_page_buffers( handle_t
*handle
,
410 struct buffer_head
*head
,
414 int (*fn
)( handle_t
*handle
,
415 struct buffer_head
*bh
))
417 struct buffer_head
*bh
;
418 unsigned block_start
, block_end
;
419 unsigned blocksize
= head
->b_size
;
421 struct buffer_head
*next
;
423 for ( bh
= head
, block_start
= 0;
424 ret
== 0 && (bh
!= head
|| !block_start
);
425 block_start
= block_end
, bh
= next
)
427 next
= bh
->b_this_page
;
428 block_end
= block_start
+ blocksize
;
429 if (block_end
<= from
|| block_start
>= to
) {
430 if (partial
&& !buffer_uptodate(bh
))
434 err
= (*fn
)(handle
, bh
);
441 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
446 struct inode
*inode
= mapping
->host
;
448 mlog(0, "(block = %llu)\n", (unsigned long long)block
);
450 /* We don't need to lock journal system files, since they aren't
451 * accessed concurrently from multiple nodes.
453 if (!INODE_JOURNAL(inode
)) {
454 err
= ocfs2_inode_lock(inode
, NULL
, 0);
460 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
463 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
464 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
467 if (!INODE_JOURNAL(inode
)) {
468 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
469 ocfs2_inode_unlock(inode
, 0);
473 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
474 (unsigned long long)block
);
480 status
= err
? 0 : p_blkno
;
486 * TODO: Make this into a generic get_blocks function.
488 * From do_direct_io in direct-io.c:
489 * "So what we do is to permit the ->get_blocks function to populate
490 * bh.b_size with the size of IO which is permitted at this offset and
493 * This function is called directly from get_more_blocks in direct-io.c.
495 * called like this: dio->get_blocks(dio->inode, fs_startblk,
496 * fs_count, map_bh, dio->rw == WRITE);
498 * Note that we never bother to allocate blocks here, and thus ignore the
501 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
502 struct buffer_head
*bh_result
, int create
)
505 u64 p_blkno
, inode_blocks
, contig_blocks
;
506 unsigned int ext_flags
;
507 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
508 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
510 /* This function won't even be called if the request isn't all
511 * nicely aligned and of the right size, so there's no need
512 * for us to check any of that. */
514 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
516 /* This figures out the size of the next contiguous block, and
517 * our logical offset */
518 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
519 &contig_blocks
, &ext_flags
);
521 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
522 (unsigned long long)iblock
);
527 /* We should already CoW the refcounted extent in case of create. */
528 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
531 * get_more_blocks() expects us to describe a hole by clearing
532 * the mapped bit on bh_result().
534 * Consider an unwritten extent as a hole.
536 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
537 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
539 clear_buffer_mapped(bh_result
);
541 /* make sure we don't map more than max_blocks blocks here as
542 that's all the kernel will handle at this point. */
543 if (max_blocks
< contig_blocks
)
544 contig_blocks
= max_blocks
;
545 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
551 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
552 * particularly interested in the aio/dio case. Like the core uses
553 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
554 * truncation on another.
556 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
563 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
566 /* this io's submitter should not have unlocked this before we could */
567 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
569 if (ocfs2_iocb_is_sem_locked(iocb
)) {
570 up_read(&inode
->i_alloc_sem
);
571 ocfs2_iocb_clear_sem_locked(iocb
);
574 ocfs2_iocb_clear_rw_locked(iocb
);
576 level
= ocfs2_iocb_rw_locked_level(iocb
);
577 ocfs2_rw_unlock(inode
, level
);
580 aio_complete(iocb
, ret
, 0);
584 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
585 * from ext3. PageChecked() bits have been removed as OCFS2 does not
586 * do journalled data.
588 static void ocfs2_invalidatepage(struct page
*page
, unsigned long offset
)
590 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
592 jbd2_journal_invalidatepage(journal
, page
, offset
);
595 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
597 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
599 if (!page_has_buffers(page
))
601 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
604 static ssize_t
ocfs2_direct_IO(int rw
,
606 const struct iovec
*iov
,
608 unsigned long nr_segs
)
610 struct file
*file
= iocb
->ki_filp
;
611 struct inode
*inode
= file
->f_path
.dentry
->d_inode
->i_mapping
->host
;
614 * Fallback to buffered I/O if we see an inode without
617 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
620 /* Fallback to buffered I/O if we are appending. */
621 if (i_size_read(inode
) <= offset
)
624 return __blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
625 iov
, offset
, nr_segs
,
626 ocfs2_direct_IO_get_blocks
,
627 ocfs2_dio_end_io
, NULL
, 0);
630 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
635 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
637 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
640 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
642 cluster_start
= cpos
% cpp
;
643 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
645 cluster_end
= cluster_start
+ osb
->s_clustersize
;
648 BUG_ON(cluster_start
> PAGE_SIZE
);
649 BUG_ON(cluster_end
> PAGE_SIZE
);
652 *start
= cluster_start
;
658 * 'from' and 'to' are the region in the page to avoid zeroing.
660 * If pagesize > clustersize, this function will avoid zeroing outside
661 * of the cluster boundary.
663 * from == to == 0 is code for "zero the entire cluster region"
665 static void ocfs2_clear_page_regions(struct page
*page
,
666 struct ocfs2_super
*osb
, u32 cpos
,
667 unsigned from
, unsigned to
)
670 unsigned int cluster_start
, cluster_end
;
672 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
674 kaddr
= kmap_atomic(page
, KM_USER0
);
677 if (from
> cluster_start
)
678 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
679 if (to
< cluster_end
)
680 memset(kaddr
+ to
, 0, cluster_end
- to
);
682 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
685 kunmap_atomic(kaddr
, KM_USER0
);
689 * Nonsparse file systems fully allocate before we get to the write
690 * code. This prevents ocfs2_write() from tagging the write as an
691 * allocating one, which means ocfs2_map_page_blocks() might try to
692 * read-in the blocks at the tail of our file. Avoid reading them by
693 * testing i_size against each block offset.
695 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
696 unsigned int block_start
)
698 u64 offset
= page_offset(page
) + block_start
;
700 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
703 if (i_size_read(inode
) > offset
)
710 * Some of this taken from __block_write_begin(). We already have our
711 * mapping by now though, and the entire write will be allocating or
712 * it won't, so not much need to use BH_New.
714 * This will also skip zeroing, which is handled externally.
716 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
717 struct inode
*inode
, unsigned int from
,
718 unsigned int to
, int new)
721 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
722 unsigned int block_end
, block_start
;
723 unsigned int bsize
= 1 << inode
->i_blkbits
;
725 if (!page_has_buffers(page
))
726 create_empty_buffers(page
, bsize
, 0);
728 head
= page_buffers(page
);
729 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
730 bh
= bh
->b_this_page
, block_start
+= bsize
) {
731 block_end
= block_start
+ bsize
;
733 clear_buffer_new(bh
);
736 * Ignore blocks outside of our i/o range -
737 * they may belong to unallocated clusters.
739 if (block_start
>= to
|| block_end
<= from
) {
740 if (PageUptodate(page
))
741 set_buffer_uptodate(bh
);
746 * For an allocating write with cluster size >= page
747 * size, we always write the entire page.
752 if (!buffer_mapped(bh
)) {
753 map_bh(bh
, inode
->i_sb
, *p_blkno
);
754 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
757 if (PageUptodate(page
)) {
758 if (!buffer_uptodate(bh
))
759 set_buffer_uptodate(bh
);
760 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
762 ocfs2_should_read_blk(inode
, page
, block_start
) &&
763 (block_start
< from
|| block_end
> to
)) {
764 ll_rw_block(READ
, 1, &bh
);
768 *p_blkno
= *p_blkno
+ 1;
772 * If we issued read requests - let them complete.
774 while(wait_bh
> wait
) {
775 wait_on_buffer(*--wait_bh
);
776 if (!buffer_uptodate(*wait_bh
))
780 if (ret
== 0 || !new)
784 * If we get -EIO above, zero out any newly allocated blocks
785 * to avoid exposing stale data.
790 block_end
= block_start
+ bsize
;
791 if (block_end
<= from
)
793 if (block_start
>= to
)
796 zero_user(page
, block_start
, bh
->b_size
);
797 set_buffer_uptodate(bh
);
798 mark_buffer_dirty(bh
);
801 block_start
= block_end
;
802 bh
= bh
->b_this_page
;
803 } while (bh
!= head
);
808 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
809 #define OCFS2_MAX_CTXT_PAGES 1
811 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
814 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
817 * Describe the state of a single cluster to be written to.
819 struct ocfs2_write_cluster_desc
{
823 * Give this a unique field because c_phys eventually gets
827 unsigned c_unwritten
;
828 unsigned c_needs_zero
;
831 struct ocfs2_write_ctxt
{
832 /* Logical cluster position / len of write */
836 /* First cluster allocated in a nonsparse extend */
837 u32 w_first_new_cpos
;
839 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
842 * This is true if page_size > cluster_size.
844 * It triggers a set of special cases during write which might
845 * have to deal with allocating writes to partial pages.
847 unsigned int w_large_pages
;
850 * Pages involved in this write.
852 * w_target_page is the page being written to by the user.
854 * w_pages is an array of pages which always contains
855 * w_target_page, and in the case of an allocating write with
856 * page_size < cluster size, it will contain zero'd and mapped
857 * pages adjacent to w_target_page which need to be written
858 * out in so that future reads from that region will get
861 unsigned int w_num_pages
;
862 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
863 struct page
*w_target_page
;
866 * ocfs2_write_end() uses this to know what the real range to
867 * write in the target should be.
869 unsigned int w_target_from
;
870 unsigned int w_target_to
;
873 * We could use journal_current_handle() but this is cleaner,
878 struct buffer_head
*w_di_bh
;
880 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
883 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
887 for(i
= 0; i
< num_pages
; i
++) {
889 unlock_page(pages
[i
]);
890 mark_page_accessed(pages
[i
]);
891 page_cache_release(pages
[i
]);
896 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
898 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
904 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
905 struct ocfs2_super
*osb
, loff_t pos
,
906 unsigned len
, struct buffer_head
*di_bh
)
909 struct ocfs2_write_ctxt
*wc
;
911 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
915 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
916 wc
->w_first_new_cpos
= UINT_MAX
;
917 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
918 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
922 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
923 wc
->w_large_pages
= 1;
925 wc
->w_large_pages
= 0;
927 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
935 * If a page has any new buffers, zero them out here, and mark them uptodate
936 * and dirty so they'll be written out (in order to prevent uninitialised
937 * block data from leaking). And clear the new bit.
939 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
941 unsigned int block_start
, block_end
;
942 struct buffer_head
*head
, *bh
;
944 BUG_ON(!PageLocked(page
));
945 if (!page_has_buffers(page
))
948 bh
= head
= page_buffers(page
);
951 block_end
= block_start
+ bh
->b_size
;
953 if (buffer_new(bh
)) {
954 if (block_end
> from
&& block_start
< to
) {
955 if (!PageUptodate(page
)) {
958 start
= max(from
, block_start
);
959 end
= min(to
, block_end
);
961 zero_user_segment(page
, start
, end
);
962 set_buffer_uptodate(bh
);
965 clear_buffer_new(bh
);
966 mark_buffer_dirty(bh
);
970 block_start
= block_end
;
971 bh
= bh
->b_this_page
;
972 } while (bh
!= head
);
976 * Only called when we have a failure during allocating write to write
977 * zero's to the newly allocated region.
979 static void ocfs2_write_failure(struct inode
*inode
,
980 struct ocfs2_write_ctxt
*wc
,
981 loff_t user_pos
, unsigned user_len
)
984 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
985 to
= user_pos
+ user_len
;
986 struct page
*tmppage
;
988 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
990 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
991 tmppage
= wc
->w_pages
[i
];
993 if (page_has_buffers(tmppage
)) {
994 if (ocfs2_should_order_data(inode
))
995 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
997 block_commit_write(tmppage
, from
, to
);
1002 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1003 struct ocfs2_write_ctxt
*wc
,
1004 struct page
*page
, u32 cpos
,
1005 loff_t user_pos
, unsigned user_len
,
1009 unsigned int map_from
= 0, map_to
= 0;
1010 unsigned int cluster_start
, cluster_end
;
1011 unsigned int user_data_from
= 0, user_data_to
= 0;
1013 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1014 &cluster_start
, &cluster_end
);
1016 if (page
== wc
->w_target_page
) {
1017 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1018 map_to
= map_from
+ user_len
;
1021 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1022 cluster_start
, cluster_end
,
1025 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1026 map_from
, map_to
, new);
1032 user_data_from
= map_from
;
1033 user_data_to
= map_to
;
1035 map_from
= cluster_start
;
1036 map_to
= cluster_end
;
1040 * If we haven't allocated the new page yet, we
1041 * shouldn't be writing it out without copying user
1042 * data. This is likely a math error from the caller.
1046 map_from
= cluster_start
;
1047 map_to
= cluster_end
;
1049 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1050 cluster_start
, cluster_end
, new);
1058 * Parts of newly allocated pages need to be zero'd.
1060 * Above, we have also rewritten 'to' and 'from' - as far as
1061 * the rest of the function is concerned, the entire cluster
1062 * range inside of a page needs to be written.
1064 * We can skip this if the page is up to date - it's already
1065 * been zero'd from being read in as a hole.
1067 if (new && !PageUptodate(page
))
1068 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1069 cpos
, user_data_from
, user_data_to
);
1071 flush_dcache_page(page
);
1078 * This function will only grab one clusters worth of pages.
1080 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1081 struct ocfs2_write_ctxt
*wc
,
1082 u32 cpos
, loff_t user_pos
,
1083 unsigned user_len
, int new,
1084 struct page
*mmap_page
)
1087 unsigned long start
, target_index
, end_index
, index
;
1088 struct inode
*inode
= mapping
->host
;
1091 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1094 * Figure out how many pages we'll be manipulating here. For
1095 * non allocating write, we just change the one
1096 * page. Otherwise, we'll need a whole clusters worth. If we're
1097 * writing past i_size, we only need enough pages to cover the
1098 * last page of the write.
1101 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1102 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1104 * We need the index *past* the last page we could possibly
1105 * touch. This is the page past the end of the write or
1106 * i_size, whichever is greater.
1108 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1109 BUG_ON(last_byte
< 1);
1110 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1111 if ((start
+ wc
->w_num_pages
) > end_index
)
1112 wc
->w_num_pages
= end_index
- start
;
1114 wc
->w_num_pages
= 1;
1115 start
= target_index
;
1118 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1121 if (index
== target_index
&& mmap_page
) {
1123 * ocfs2_pagemkwrite() is a little different
1124 * and wants us to directly use the page
1127 lock_page(mmap_page
);
1129 if (mmap_page
->mapping
!= mapping
) {
1130 unlock_page(mmap_page
);
1132 * Sanity check - the locking in
1133 * ocfs2_pagemkwrite() should ensure
1134 * that this code doesn't trigger.
1141 page_cache_get(mmap_page
);
1142 wc
->w_pages
[i
] = mmap_page
;
1144 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1146 if (!wc
->w_pages
[i
]) {
1153 if (index
== target_index
)
1154 wc
->w_target_page
= wc
->w_pages
[i
];
1161 * Prepare a single cluster for write one cluster into the file.
1163 static int ocfs2_write_cluster(struct address_space
*mapping
,
1164 u32 phys
, unsigned int unwritten
,
1165 unsigned int should_zero
,
1166 struct ocfs2_alloc_context
*data_ac
,
1167 struct ocfs2_alloc_context
*meta_ac
,
1168 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1169 loff_t user_pos
, unsigned user_len
)
1172 u64 v_blkno
, p_blkno
;
1173 struct inode
*inode
= mapping
->host
;
1174 struct ocfs2_extent_tree et
;
1176 new = phys
== 0 ? 1 : 0;
1181 * This is safe to call with the page locks - it won't take
1182 * any additional semaphores or cluster locks.
1185 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1186 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1187 wc
->w_handle
, data_ac
,
1190 * This shouldn't happen because we must have already
1191 * calculated the correct meta data allocation required. The
1192 * internal tree allocation code should know how to increase
1193 * transaction credits itself.
1195 * If need be, we could handle -EAGAIN for a
1196 * RESTART_TRANS here.
1198 mlog_bug_on_msg(ret
== -EAGAIN
,
1199 "Inode %llu: EAGAIN return during allocation.\n",
1200 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1205 } else if (unwritten
) {
1206 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1208 ret
= ocfs2_mark_extent_written(inode
, &et
,
1209 wc
->w_handle
, cpos
, 1, phys
,
1210 meta_ac
, &wc
->w_dealloc
);
1218 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1220 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1223 * The only reason this should fail is due to an inability to
1224 * find the extent added.
1226 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1229 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1230 "at logical block %llu",
1231 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1232 (unsigned long long)v_blkno
);
1236 BUG_ON(p_blkno
== 0);
1238 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1241 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1242 wc
->w_pages
[i
], cpos
,
1253 * We only have cleanup to do in case of allocating write.
1256 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1263 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1264 struct ocfs2_alloc_context
*data_ac
,
1265 struct ocfs2_alloc_context
*meta_ac
,
1266 struct ocfs2_write_ctxt
*wc
,
1267 loff_t pos
, unsigned len
)
1271 unsigned int local_len
= len
;
1272 struct ocfs2_write_cluster_desc
*desc
;
1273 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1275 for (i
= 0; i
< wc
->w_clen
; i
++) {
1276 desc
= &wc
->w_desc
[i
];
1279 * We have to make sure that the total write passed in
1280 * doesn't extend past a single cluster.
1283 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1284 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1285 local_len
= osb
->s_clustersize
- cluster_off
;
1287 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1291 wc
, desc
->c_cpos
, pos
, local_len
);
1307 * ocfs2_write_end() wants to know which parts of the target page it
1308 * should complete the write on. It's easiest to compute them ahead of
1309 * time when a more complete view of the write is available.
1311 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1312 struct ocfs2_write_ctxt
*wc
,
1313 loff_t pos
, unsigned len
, int alloc
)
1315 struct ocfs2_write_cluster_desc
*desc
;
1317 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1318 wc
->w_target_to
= wc
->w_target_from
+ len
;
1324 * Allocating write - we may have different boundaries based
1325 * on page size and cluster size.
1327 * NOTE: We can no longer compute one value from the other as
1328 * the actual write length and user provided length may be
1332 if (wc
->w_large_pages
) {
1334 * We only care about the 1st and last cluster within
1335 * our range and whether they should be zero'd or not. Either
1336 * value may be extended out to the start/end of a
1337 * newly allocated cluster.
1339 desc
= &wc
->w_desc
[0];
1340 if (desc
->c_needs_zero
)
1341 ocfs2_figure_cluster_boundaries(osb
,
1346 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1347 if (desc
->c_needs_zero
)
1348 ocfs2_figure_cluster_boundaries(osb
,
1353 wc
->w_target_from
= 0;
1354 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1359 * Populate each single-cluster write descriptor in the write context
1360 * with information about the i/o to be done.
1362 * Returns the number of clusters that will have to be allocated, as
1363 * well as a worst case estimate of the number of extent records that
1364 * would have to be created during a write to an unwritten region.
1366 static int ocfs2_populate_write_desc(struct inode
*inode
,
1367 struct ocfs2_write_ctxt
*wc
,
1368 unsigned int *clusters_to_alloc
,
1369 unsigned int *extents_to_split
)
1372 struct ocfs2_write_cluster_desc
*desc
;
1373 unsigned int num_clusters
= 0;
1374 unsigned int ext_flags
= 0;
1378 *clusters_to_alloc
= 0;
1379 *extents_to_split
= 0;
1381 for (i
= 0; i
< wc
->w_clen
; i
++) {
1382 desc
= &wc
->w_desc
[i
];
1383 desc
->c_cpos
= wc
->w_cpos
+ i
;
1385 if (num_clusters
== 0) {
1387 * Need to look up the next extent record.
1389 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1390 &num_clusters
, &ext_flags
);
1396 /* We should already CoW the refcountd extent. */
1397 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1400 * Assume worst case - that we're writing in
1401 * the middle of the extent.
1403 * We can assume that the write proceeds from
1404 * left to right, in which case the extent
1405 * insert code is smart enough to coalesce the
1406 * next splits into the previous records created.
1408 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1409 *extents_to_split
= *extents_to_split
+ 2;
1412 * Only increment phys if it doesn't describe
1419 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1420 * file that got extended. w_first_new_cpos tells us
1421 * where the newly allocated clusters are so we can
1424 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1426 desc
->c_needs_zero
= 1;
1429 desc
->c_phys
= phys
;
1432 desc
->c_needs_zero
= 1;
1433 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1436 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1437 desc
->c_unwritten
= 1;
1438 desc
->c_needs_zero
= 1;
1449 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1450 struct inode
*inode
,
1451 struct ocfs2_write_ctxt
*wc
)
1454 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1457 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1459 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1466 * If we don't set w_num_pages then this page won't get unlocked
1467 * and freed on cleanup of the write context.
1469 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1470 wc
->w_num_pages
= 1;
1472 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1473 if (IS_ERR(handle
)) {
1474 ret
= PTR_ERR(handle
);
1479 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1480 OCFS2_JOURNAL_ACCESS_WRITE
);
1482 ocfs2_commit_trans(osb
, handle
);
1488 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1489 ocfs2_set_inode_data_inline(inode
, di
);
1491 if (!PageUptodate(page
)) {
1492 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1494 ocfs2_commit_trans(osb
, handle
);
1500 wc
->w_handle
= handle
;
1505 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1507 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1509 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1514 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1515 struct inode
*inode
, loff_t pos
,
1516 unsigned len
, struct page
*mmap_page
,
1517 struct ocfs2_write_ctxt
*wc
)
1519 int ret
, written
= 0;
1520 loff_t end
= pos
+ len
;
1521 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1522 struct ocfs2_dinode
*di
= NULL
;
1524 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1525 (unsigned long long)oi
->ip_blkno
, len
, (unsigned long long)pos
,
1526 oi
->ip_dyn_features
);
1529 * Handle inodes which already have inline data 1st.
1531 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1532 if (mmap_page
== NULL
&&
1533 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1534 goto do_inline_write
;
1537 * The write won't fit - we have to give this inode an
1538 * inline extent list now.
1540 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1547 * Check whether the inode can accept inline data.
1549 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1553 * Check whether the write can fit.
1555 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1557 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1561 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1568 * This signals to the caller that the data can be written
1573 return written
? written
: ret
;
1577 * This function only does anything for file systems which can't
1578 * handle sparse files.
1580 * What we want to do here is fill in any hole between the current end
1581 * of allocation and the end of our write. That way the rest of the
1582 * write path can treat it as an non-allocating write, which has no
1583 * special case code for sparse/nonsparse files.
1585 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1586 struct buffer_head
*di_bh
,
1587 loff_t pos
, unsigned len
,
1588 struct ocfs2_write_ctxt
*wc
)
1591 loff_t newsize
= pos
+ len
;
1593 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1595 if (newsize
<= i_size_read(inode
))
1598 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1602 wc
->w_first_new_cpos
=
1603 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1608 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
1613 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1614 if (pos
> i_size_read(inode
))
1615 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
1621 * Try to flush truncate logs if we can free enough clusters from it.
1622 * As for return value, "< 0" means error, "0" no space and "1" means
1623 * we have freed enough spaces and let the caller try to allocate again.
1625 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
1626 unsigned int needed
)
1630 unsigned int truncated_clusters
;
1632 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
1633 truncated_clusters
= osb
->truncated_clusters
;
1634 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
1637 * Check whether we can succeed in allocating if we free
1640 if (truncated_clusters
< needed
)
1643 ret
= ocfs2_flush_truncate_log(osb
);
1649 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
1650 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
1657 int ocfs2_write_begin_nolock(struct file
*filp
,
1658 struct address_space
*mapping
,
1659 loff_t pos
, unsigned len
, unsigned flags
,
1660 struct page
**pagep
, void **fsdata
,
1661 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1663 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1664 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
1665 struct ocfs2_write_ctxt
*wc
;
1666 struct inode
*inode
= mapping
->host
;
1667 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1668 struct ocfs2_dinode
*di
;
1669 struct ocfs2_alloc_context
*data_ac
= NULL
;
1670 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1672 struct ocfs2_extent_tree et
;
1673 int try_free
= 1, ret1
;
1676 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1682 if (ocfs2_supports_inline_data(osb
)) {
1683 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1695 if (ocfs2_sparse_alloc(osb
))
1696 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
1698 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
1705 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1709 } else if (ret
== 1) {
1710 clusters_need
= wc
->w_clen
;
1711 ret
= ocfs2_refcount_cow(inode
, filp
, di_bh
,
1712 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1719 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1725 clusters_need
+= clusters_to_alloc
;
1727 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1730 * We set w_target_from, w_target_to here so that
1731 * ocfs2_write_end() knows which range in the target page to
1732 * write out. An allocation requires that we write the entire
1735 if (clusters_to_alloc
|| extents_to_split
) {
1737 * XXX: We are stretching the limits of
1738 * ocfs2_lock_allocators(). It greatly over-estimates
1739 * the work to be done.
1741 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1742 " clusters_to_add = %u, extents_to_split = %u\n",
1743 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1744 (long long)i_size_read(inode
), le32_to_cpu(di
->i_clusters
),
1745 clusters_to_alloc
, extents_to_split
);
1747 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1749 ret
= ocfs2_lock_allocators(inode
, &et
,
1750 clusters_to_alloc
, extents_to_split
,
1751 &data_ac
, &meta_ac
);
1758 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
1760 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1767 * We have to zero sparse allocated clusters, unwritten extent clusters,
1768 * and non-sparse clusters we just extended. For non-sparse writes,
1769 * we know zeros will only be needed in the first and/or last cluster.
1771 if (clusters_to_alloc
|| extents_to_split
||
1772 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
1773 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
1774 cluster_of_pages
= 1;
1776 cluster_of_pages
= 0;
1778 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1780 handle
= ocfs2_start_trans(osb
, credits
);
1781 if (IS_ERR(handle
)) {
1782 ret
= PTR_ERR(handle
);
1787 wc
->w_handle
= handle
;
1789 if (clusters_to_alloc
) {
1790 ret
= dquot_alloc_space_nodirty(inode
,
1791 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1796 * We don't want this to fail in ocfs2_write_end(), so do it
1799 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1800 OCFS2_JOURNAL_ACCESS_WRITE
);
1807 * Fill our page array first. That way we've grabbed enough so
1808 * that we can zero and flush if we error after adding the
1811 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
1812 cluster_of_pages
, mmap_page
);
1818 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1826 ocfs2_free_alloc_context(data_ac
);
1828 ocfs2_free_alloc_context(meta_ac
);
1831 *pagep
= wc
->w_target_page
;
1835 if (clusters_to_alloc
)
1836 dquot_free_space(inode
,
1837 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1839 ocfs2_commit_trans(osb
, handle
);
1842 ocfs2_free_write_ctxt(wc
);
1845 ocfs2_free_alloc_context(data_ac
);
1847 ocfs2_free_alloc_context(meta_ac
);
1849 if (ret
== -ENOSPC
&& try_free
) {
1851 * Try to free some truncate log so that we can have enough
1852 * clusters to allocate.
1856 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
1867 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1868 loff_t pos
, unsigned len
, unsigned flags
,
1869 struct page
**pagep
, void **fsdata
)
1872 struct buffer_head
*di_bh
= NULL
;
1873 struct inode
*inode
= mapping
->host
;
1875 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1882 * Take alloc sem here to prevent concurrent lookups. That way
1883 * the mapping, zeroing and tree manipulation within
1884 * ocfs2_write() will be safe against ->readpage(). This
1885 * should also serve to lock out allocation from a shared
1888 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1890 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
1891 fsdata
, di_bh
, NULL
);
1902 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1905 ocfs2_inode_unlock(inode
, 1);
1910 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1911 unsigned len
, unsigned *copied
,
1912 struct ocfs2_dinode
*di
,
1913 struct ocfs2_write_ctxt
*wc
)
1917 if (unlikely(*copied
< len
)) {
1918 if (!PageUptodate(wc
->w_target_page
)) {
1924 kaddr
= kmap_atomic(wc
->w_target_page
, KM_USER0
);
1925 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1926 kunmap_atomic(kaddr
, KM_USER0
);
1928 mlog(0, "Data written to inode at offset %llu. "
1929 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1930 (unsigned long long)pos
, *copied
,
1931 le16_to_cpu(di
->id2
.i_data
.id_count
),
1932 le16_to_cpu(di
->i_dyn_features
));
1935 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1936 loff_t pos
, unsigned len
, unsigned copied
,
1937 struct page
*page
, void *fsdata
)
1940 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1941 struct inode
*inode
= mapping
->host
;
1942 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1943 struct ocfs2_write_ctxt
*wc
= fsdata
;
1944 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1945 handle_t
*handle
= wc
->w_handle
;
1946 struct page
*tmppage
;
1948 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1949 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1950 goto out_write_size
;
1953 if (unlikely(copied
< len
)) {
1954 if (!PageUptodate(wc
->w_target_page
))
1957 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
1960 flush_dcache_page(wc
->w_target_page
);
1962 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1963 tmppage
= wc
->w_pages
[i
];
1965 if (tmppage
== wc
->w_target_page
) {
1966 from
= wc
->w_target_from
;
1967 to
= wc
->w_target_to
;
1969 BUG_ON(from
> PAGE_CACHE_SIZE
||
1970 to
> PAGE_CACHE_SIZE
||
1974 * Pages adjacent to the target (if any) imply
1975 * a hole-filling write in which case we want
1976 * to flush their entire range.
1979 to
= PAGE_CACHE_SIZE
;
1982 if (page_has_buffers(tmppage
)) {
1983 if (ocfs2_should_order_data(inode
))
1984 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1985 block_commit_write(tmppage
, from
, to
);
1991 if (pos
> inode
->i_size
) {
1992 i_size_write(inode
, pos
);
1993 mark_inode_dirty(inode
);
1995 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1996 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
1997 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1998 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1999 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2000 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2002 ocfs2_commit_trans(osb
, handle
);
2004 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2006 ocfs2_free_write_ctxt(wc
);
2011 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2012 loff_t pos
, unsigned len
, unsigned copied
,
2013 struct page
*page
, void *fsdata
)
2016 struct inode
*inode
= mapping
->host
;
2018 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2020 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2021 ocfs2_inode_unlock(inode
, 1);
2026 const struct address_space_operations ocfs2_aops
= {
2027 .readpage
= ocfs2_readpage
,
2028 .readpages
= ocfs2_readpages
,
2029 .writepage
= ocfs2_writepage
,
2030 .write_begin
= ocfs2_write_begin
,
2031 .write_end
= ocfs2_write_end
,
2033 .sync_page
= block_sync_page
,
2034 .direct_IO
= ocfs2_direct_IO
,
2035 .invalidatepage
= ocfs2_invalidatepage
,
2036 .releasepage
= ocfs2_releasepage
,
2037 .migratepage
= buffer_migrate_page
,
2038 .is_partially_uptodate
= block_is_partially_uptodate
,
2039 .error_remove_page
= generic_error_remove_page
,