1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
23 #include <linux/slab.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <asm/byteorder.h>
27 #include <linux/swap.h>
28 #include <linux/pipe_fs_i.h>
29 #include <linux/mpage.h>
30 #include <linux/quotaops.h>
32 #define MLOG_MASK_PREFIX ML_FILE_IO
33 #include <cluster/masklog.h>
40 #include "extent_map.h"
47 #include "refcounttree.h"
49 #include "buffer_head_io.h"
51 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
52 struct buffer_head
*bh_result
, int create
)
56 struct ocfs2_dinode
*fe
= NULL
;
57 struct buffer_head
*bh
= NULL
;
58 struct buffer_head
*buffer_cache_bh
= NULL
;
59 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
62 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
63 (unsigned long long)iblock
, bh_result
, create
);
65 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
67 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
68 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
69 (unsigned long long)iblock
);
73 status
= ocfs2_read_inode_block(inode
, &bh
);
78 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
80 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
81 le32_to_cpu(fe
->i_clusters
))) {
82 mlog(ML_ERROR
, "block offset is outside the allocated size: "
83 "%llu\n", (unsigned long long)iblock
);
87 /* We don't use the page cache to create symlink data, so if
88 * need be, copy it over from the buffer cache. */
89 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
90 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
92 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
93 if (!buffer_cache_bh
) {
94 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
98 /* we haven't locked out transactions, so a commit
99 * could've happened. Since we've got a reference on
100 * the bh, even if it commits while we're doing the
101 * copy, the data is still good. */
102 if (buffer_jbd(buffer_cache_bh
)
103 && ocfs2_inode_is_new(inode
)) {
104 kaddr
= kmap_atomic(bh_result
->b_page
, KM_USER0
);
106 mlog(ML_ERROR
, "couldn't kmap!\n");
109 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
110 buffer_cache_bh
->b_data
,
112 kunmap_atomic(kaddr
, KM_USER0
);
113 set_buffer_uptodate(bh_result
);
115 brelse(buffer_cache_bh
);
118 map_bh(bh_result
, inode
->i_sb
,
119 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
130 int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
131 struct buffer_head
*bh_result
, int create
)
134 unsigned int ext_flags
;
135 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
136 u64 p_blkno
, count
, past_eof
;
137 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
139 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
140 (unsigned long long)iblock
, bh_result
, create
);
142 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
143 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
144 inode
, inode
->i_ino
);
146 if (S_ISLNK(inode
->i_mode
)) {
147 /* this always does I/O for some reason. */
148 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
152 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
155 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
156 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
157 (unsigned long long)p_blkno
);
161 if (max_blocks
< count
)
165 * ocfs2 never allocates in this function - the only time we
166 * need to use BH_New is when we're extending i_size on a file
167 * system which doesn't support holes, in which case BH_New
168 * allows block_prepare_write() to zero.
170 * If we see this on a sparse file system, then a truncate has
171 * raced us and removed the cluster. In this case, we clear
172 * the buffers dirty and uptodate bits and let the buffer code
173 * ignore it as a hole.
175 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
176 clear_buffer_dirty(bh_result
);
177 clear_buffer_uptodate(bh_result
);
181 /* Treat the unwritten extent as a hole for zeroing purposes. */
182 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
183 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
185 bh_result
->b_size
= count
<< inode
->i_blkbits
;
187 if (!ocfs2_sparse_alloc(osb
)) {
191 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
192 (unsigned long long)iblock
,
193 (unsigned long long)p_blkno
,
194 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
195 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
200 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
201 mlog(0, "Inode %lu, past_eof = %llu\n", inode
->i_ino
,
202 (unsigned long long)past_eof
);
204 if (create
&& (iblock
>= past_eof
))
205 set_buffer_new(bh_result
);
216 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
217 struct buffer_head
*di_bh
)
221 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
223 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
224 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
225 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
229 size
= i_size_read(inode
);
231 if (size
> PAGE_CACHE_SIZE
||
232 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
233 ocfs2_error(inode
->i_sb
,
234 "Inode %llu has with inline data has bad size: %Lu",
235 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
236 (unsigned long long)size
);
240 kaddr
= kmap_atomic(page
, KM_USER0
);
242 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
243 /* Clear the remaining part of the page */
244 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
245 flush_dcache_page(page
);
246 kunmap_atomic(kaddr
, KM_USER0
);
248 SetPageUptodate(page
);
253 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
256 struct buffer_head
*di_bh
= NULL
;
258 BUG_ON(!PageLocked(page
));
259 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
261 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
267 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
275 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
277 struct inode
*inode
= page
->mapping
->host
;
278 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
279 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
282 mlog_entry("(0x%p, %lu)\n", file
, (page
? page
->index
: 0));
284 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
286 if (ret
== AOP_TRUNCATED_PAGE
)
292 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
293 ret
= AOP_TRUNCATED_PAGE
;
294 goto out_inode_unlock
;
298 * i_size might have just been updated as we grabed the meta lock. We
299 * might now be discovering a truncate that hit on another node.
300 * block_read_full_page->get_block freaks out if it is asked to read
301 * beyond the end of a file, so we check here. Callers
302 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
303 * and notice that the page they just read isn't needed.
305 * XXX sys_readahead() seems to get that wrong?
307 if (start
>= i_size_read(inode
)) {
308 zero_user(page
, 0, PAGE_SIZE
);
309 SetPageUptodate(page
);
314 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
315 ret
= ocfs2_readpage_inline(inode
, page
);
317 ret
= block_read_full_page(page
, ocfs2_get_block
);
321 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
323 ocfs2_inode_unlock(inode
, 0);
332 * This is used only for read-ahead. Failures or difficult to handle
333 * situations are safe to ignore.
335 * Right now, we don't bother with BH_Boundary - in-inode extent lists
336 * are quite large (243 extents on 4k blocks), so most inodes don't
337 * grow out to a tree. If need be, detecting boundary extents could
338 * trivially be added in a future version of ocfs2_get_block().
340 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
341 struct list_head
*pages
, unsigned nr_pages
)
344 struct inode
*inode
= mapping
->host
;
345 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
350 * Use the nonblocking flag for the dlm code to avoid page
351 * lock inversion, but don't bother with retrying.
353 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
357 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
358 ocfs2_inode_unlock(inode
, 0);
363 * Don't bother with inline-data. There isn't anything
364 * to read-ahead in that case anyway...
366 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
370 * Check whether a remote node truncated this file - we just
371 * drop out in that case as it's not worth handling here.
373 last
= list_entry(pages
->prev
, struct page
, lru
);
374 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
375 if (start
>= i_size_read(inode
))
378 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
381 up_read(&oi
->ip_alloc_sem
);
382 ocfs2_inode_unlock(inode
, 0);
387 /* Note: Because we don't support holes, our allocation has
388 * already happened (allocation writes zeros to the file data)
389 * so we don't have to worry about ordered writes in
392 * ->writepage is called during the process of invalidating the page cache
393 * during blocked lock processing. It can't block on any cluster locks
394 * to during block mapping. It's relying on the fact that the block
395 * mapping can't have disappeared under the dirty pages that it is
396 * being asked to write back.
398 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
402 mlog_entry("(0x%p)\n", page
);
404 ret
= block_write_full_page(page
, ocfs2_get_block
, wbc
);
412 * This is called from ocfs2_write_zero_page() which has handled it's
413 * own cluster locking and has ensured allocation exists for those
414 * blocks to be written.
416 int ocfs2_prepare_write_nolock(struct inode
*inode
, struct page
*page
,
417 unsigned from
, unsigned to
)
421 ret
= block_prepare_write(page
, from
, to
, ocfs2_get_block
);
426 /* Taken from ext3. We don't necessarily need the full blown
427 * functionality yet, but IMHO it's better to cut and paste the whole
428 * thing so we can avoid introducing our own bugs (and easily pick up
429 * their fixes when they happen) --Mark */
430 int walk_page_buffers( handle_t
*handle
,
431 struct buffer_head
*head
,
435 int (*fn
)( handle_t
*handle
,
436 struct buffer_head
*bh
))
438 struct buffer_head
*bh
;
439 unsigned block_start
, block_end
;
440 unsigned blocksize
= head
->b_size
;
442 struct buffer_head
*next
;
444 for ( bh
= head
, block_start
= 0;
445 ret
== 0 && (bh
!= head
|| !block_start
);
446 block_start
= block_end
, bh
= next
)
448 next
= bh
->b_this_page
;
449 block_end
= block_start
+ blocksize
;
450 if (block_end
<= from
|| block_start
>= to
) {
451 if (partial
&& !buffer_uptodate(bh
))
455 err
= (*fn
)(handle
, bh
);
462 handle_t
*ocfs2_start_walk_page_trans(struct inode
*inode
,
467 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
471 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
472 if (IS_ERR(handle
)) {
478 if (ocfs2_should_order_data(inode
)) {
479 ret
= ocfs2_jbd2_file_inode(handle
, inode
);
486 ocfs2_commit_trans(osb
, handle
);
487 handle
= ERR_PTR(ret
);
492 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
497 struct inode
*inode
= mapping
->host
;
499 mlog_entry("(block = %llu)\n", (unsigned long long)block
);
501 /* We don't need to lock journal system files, since they aren't
502 * accessed concurrently from multiple nodes.
504 if (!INODE_JOURNAL(inode
)) {
505 err
= ocfs2_inode_lock(inode
, NULL
, 0);
511 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
514 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
515 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
518 if (!INODE_JOURNAL(inode
)) {
519 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
520 ocfs2_inode_unlock(inode
, 0);
524 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
525 (unsigned long long)block
);
531 status
= err
? 0 : p_blkno
;
533 mlog_exit((int)status
);
539 * TODO: Make this into a generic get_blocks function.
541 * From do_direct_io in direct-io.c:
542 * "So what we do is to permit the ->get_blocks function to populate
543 * bh.b_size with the size of IO which is permitted at this offset and
546 * This function is called directly from get_more_blocks in direct-io.c.
548 * called like this: dio->get_blocks(dio->inode, fs_startblk,
549 * fs_count, map_bh, dio->rw == WRITE);
551 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
552 struct buffer_head
*bh_result
, int create
)
555 u64 p_blkno
, inode_blocks
, contig_blocks
;
556 unsigned int ext_flags
;
557 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
558 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
560 /* This function won't even be called if the request isn't all
561 * nicely aligned and of the right size, so there's no need
562 * for us to check any of that. */
564 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
567 * Any write past EOF is not allowed because we'd be extending.
569 if (create
&& (iblock
+ max_blocks
) > inode_blocks
) {
574 /* This figures out the size of the next contiguous block, and
575 * our logical offset */
576 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
577 &contig_blocks
, &ext_flags
);
579 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
580 (unsigned long long)iblock
);
585 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)) && !p_blkno
&& create
) {
586 ocfs2_error(inode
->i_sb
,
587 "Inode %llu has a hole at block %llu\n",
588 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
589 (unsigned long long)iblock
);
594 /* We should already CoW the refcounted extent. */
595 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
597 * get_more_blocks() expects us to describe a hole by clearing
598 * the mapped bit on bh_result().
600 * Consider an unwritten extent as a hole.
602 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
603 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
606 * ocfs2_prepare_inode_for_write() should have caught
607 * the case where we'd be filling a hole and triggered
608 * a buffered write instead.
616 clear_buffer_mapped(bh_result
);
619 /* make sure we don't map more than max_blocks blocks here as
620 that's all the kernel will handle at this point. */
621 if (max_blocks
< contig_blocks
)
622 contig_blocks
= max_blocks
;
623 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
629 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
630 * particularly interested in the aio/dio case. Like the core uses
631 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
632 * truncation on another.
634 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
639 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
642 /* this io's submitter should not have unlocked this before we could */
643 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
645 ocfs2_iocb_clear_rw_locked(iocb
);
647 level
= ocfs2_iocb_rw_locked_level(iocb
);
649 up_read(&inode
->i_alloc_sem
);
650 ocfs2_rw_unlock(inode
, level
);
654 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
655 * from ext3. PageChecked() bits have been removed as OCFS2 does not
656 * do journalled data.
658 static void ocfs2_invalidatepage(struct page
*page
, unsigned long offset
)
660 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
662 jbd2_journal_invalidatepage(journal
, page
, offset
);
665 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
667 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
669 if (!page_has_buffers(page
))
671 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
674 static ssize_t
ocfs2_direct_IO(int rw
,
676 const struct iovec
*iov
,
678 unsigned long nr_segs
)
680 struct file
*file
= iocb
->ki_filp
;
681 struct inode
*inode
= file
->f_path
.dentry
->d_inode
->i_mapping
->host
;
687 * Fallback to buffered I/O if we see an inode without
690 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
693 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
694 inode
->i_sb
->s_bdev
, iov
, offset
,
696 ocfs2_direct_IO_get_blocks
,
703 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
708 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
710 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
713 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
715 cluster_start
= cpos
% cpp
;
716 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
718 cluster_end
= cluster_start
+ osb
->s_clustersize
;
721 BUG_ON(cluster_start
> PAGE_SIZE
);
722 BUG_ON(cluster_end
> PAGE_SIZE
);
725 *start
= cluster_start
;
731 * 'from' and 'to' are the region in the page to avoid zeroing.
733 * If pagesize > clustersize, this function will avoid zeroing outside
734 * of the cluster boundary.
736 * from == to == 0 is code for "zero the entire cluster region"
738 static void ocfs2_clear_page_regions(struct page
*page
,
739 struct ocfs2_super
*osb
, u32 cpos
,
740 unsigned from
, unsigned to
)
743 unsigned int cluster_start
, cluster_end
;
745 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
747 kaddr
= kmap_atomic(page
, KM_USER0
);
750 if (from
> cluster_start
)
751 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
752 if (to
< cluster_end
)
753 memset(kaddr
+ to
, 0, cluster_end
- to
);
755 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
758 kunmap_atomic(kaddr
, KM_USER0
);
762 * Nonsparse file systems fully allocate before we get to the write
763 * code. This prevents ocfs2_write() from tagging the write as an
764 * allocating one, which means ocfs2_map_page_blocks() might try to
765 * read-in the blocks at the tail of our file. Avoid reading them by
766 * testing i_size against each block offset.
768 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
769 unsigned int block_start
)
771 u64 offset
= page_offset(page
) + block_start
;
773 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
776 if (i_size_read(inode
) > offset
)
783 * Some of this taken from block_prepare_write(). We already have our
784 * mapping by now though, and the entire write will be allocating or
785 * it won't, so not much need to use BH_New.
787 * This will also skip zeroing, which is handled externally.
789 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
790 struct inode
*inode
, unsigned int from
,
791 unsigned int to
, int new)
794 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
795 unsigned int block_end
, block_start
;
796 unsigned int bsize
= 1 << inode
->i_blkbits
;
798 if (!page_has_buffers(page
))
799 create_empty_buffers(page
, bsize
, 0);
801 head
= page_buffers(page
);
802 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
803 bh
= bh
->b_this_page
, block_start
+= bsize
) {
804 block_end
= block_start
+ bsize
;
806 clear_buffer_new(bh
);
809 * Ignore blocks outside of our i/o range -
810 * they may belong to unallocated clusters.
812 if (block_start
>= to
|| block_end
<= from
) {
813 if (PageUptodate(page
))
814 set_buffer_uptodate(bh
);
819 * For an allocating write with cluster size >= page
820 * size, we always write the entire page.
825 if (!buffer_mapped(bh
)) {
826 map_bh(bh
, inode
->i_sb
, *p_blkno
);
827 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
830 if (PageUptodate(page
)) {
831 if (!buffer_uptodate(bh
))
832 set_buffer_uptodate(bh
);
833 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
835 ocfs2_should_read_blk(inode
, page
, block_start
) &&
836 (block_start
< from
|| block_end
> to
)) {
837 ll_rw_block(READ
, 1, &bh
);
841 *p_blkno
= *p_blkno
+ 1;
845 * If we issued read requests - let them complete.
847 while(wait_bh
> wait
) {
848 wait_on_buffer(*--wait_bh
);
849 if (!buffer_uptodate(*wait_bh
))
853 if (ret
== 0 || !new)
857 * If we get -EIO above, zero out any newly allocated blocks
858 * to avoid exposing stale data.
863 block_end
= block_start
+ bsize
;
864 if (block_end
<= from
)
866 if (block_start
>= to
)
869 zero_user(page
, block_start
, bh
->b_size
);
870 set_buffer_uptodate(bh
);
871 mark_buffer_dirty(bh
);
874 block_start
= block_end
;
875 bh
= bh
->b_this_page
;
876 } while (bh
!= head
);
881 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
882 #define OCFS2_MAX_CTXT_PAGES 1
884 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
887 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
890 * Describe the state of a single cluster to be written to.
892 struct ocfs2_write_cluster_desc
{
896 * Give this a unique field because c_phys eventually gets
900 unsigned c_unwritten
;
901 unsigned c_needs_zero
;
904 struct ocfs2_write_ctxt
{
905 /* Logical cluster position / len of write */
909 /* First cluster allocated in a nonsparse extend */
910 u32 w_first_new_cpos
;
912 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
915 * This is true if page_size > cluster_size.
917 * It triggers a set of special cases during write which might
918 * have to deal with allocating writes to partial pages.
920 unsigned int w_large_pages
;
923 * Pages involved in this write.
925 * w_target_page is the page being written to by the user.
927 * w_pages is an array of pages which always contains
928 * w_target_page, and in the case of an allocating write with
929 * page_size < cluster size, it will contain zero'd and mapped
930 * pages adjacent to w_target_page which need to be written
931 * out in so that future reads from that region will get
934 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
935 unsigned int w_num_pages
;
936 struct page
*w_target_page
;
939 * ocfs2_write_end() uses this to know what the real range to
940 * write in the target should be.
942 unsigned int w_target_from
;
943 unsigned int w_target_to
;
946 * We could use journal_current_handle() but this is cleaner,
951 struct buffer_head
*w_di_bh
;
953 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
956 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
960 for(i
= 0; i
< num_pages
; i
++) {
962 unlock_page(pages
[i
]);
963 mark_page_accessed(pages
[i
]);
964 page_cache_release(pages
[i
]);
969 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
971 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
977 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
978 struct ocfs2_super
*osb
, loff_t pos
,
979 unsigned len
, struct buffer_head
*di_bh
)
982 struct ocfs2_write_ctxt
*wc
;
984 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
988 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
989 wc
->w_first_new_cpos
= UINT_MAX
;
990 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
991 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
995 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
996 wc
->w_large_pages
= 1;
998 wc
->w_large_pages
= 0;
1000 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
1008 * If a page has any new buffers, zero them out here, and mark them uptodate
1009 * and dirty so they'll be written out (in order to prevent uninitialised
1010 * block data from leaking). And clear the new bit.
1012 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1014 unsigned int block_start
, block_end
;
1015 struct buffer_head
*head
, *bh
;
1017 BUG_ON(!PageLocked(page
));
1018 if (!page_has_buffers(page
))
1021 bh
= head
= page_buffers(page
);
1024 block_end
= block_start
+ bh
->b_size
;
1026 if (buffer_new(bh
)) {
1027 if (block_end
> from
&& block_start
< to
) {
1028 if (!PageUptodate(page
)) {
1029 unsigned start
, end
;
1031 start
= max(from
, block_start
);
1032 end
= min(to
, block_end
);
1034 zero_user_segment(page
, start
, end
);
1035 set_buffer_uptodate(bh
);
1038 clear_buffer_new(bh
);
1039 mark_buffer_dirty(bh
);
1043 block_start
= block_end
;
1044 bh
= bh
->b_this_page
;
1045 } while (bh
!= head
);
1049 * Only called when we have a failure during allocating write to write
1050 * zero's to the newly allocated region.
1052 static void ocfs2_write_failure(struct inode
*inode
,
1053 struct ocfs2_write_ctxt
*wc
,
1054 loff_t user_pos
, unsigned user_len
)
1057 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1058 to
= user_pos
+ user_len
;
1059 struct page
*tmppage
;
1061 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1063 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1064 tmppage
= wc
->w_pages
[i
];
1066 if (page_has_buffers(tmppage
)) {
1067 if (ocfs2_should_order_data(inode
))
1068 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1070 block_commit_write(tmppage
, from
, to
);
1075 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1076 struct ocfs2_write_ctxt
*wc
,
1077 struct page
*page
, u32 cpos
,
1078 loff_t user_pos
, unsigned user_len
,
1082 unsigned int map_from
= 0, map_to
= 0;
1083 unsigned int cluster_start
, cluster_end
;
1084 unsigned int user_data_from
= 0, user_data_to
= 0;
1086 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1087 &cluster_start
, &cluster_end
);
1089 if (page
== wc
->w_target_page
) {
1090 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1091 map_to
= map_from
+ user_len
;
1094 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1095 cluster_start
, cluster_end
,
1098 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1099 map_from
, map_to
, new);
1105 user_data_from
= map_from
;
1106 user_data_to
= map_to
;
1108 map_from
= cluster_start
;
1109 map_to
= cluster_end
;
1113 * If we haven't allocated the new page yet, we
1114 * shouldn't be writing it out without copying user
1115 * data. This is likely a math error from the caller.
1119 map_from
= cluster_start
;
1120 map_to
= cluster_end
;
1122 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1123 cluster_start
, cluster_end
, new);
1131 * Parts of newly allocated pages need to be zero'd.
1133 * Above, we have also rewritten 'to' and 'from' - as far as
1134 * the rest of the function is concerned, the entire cluster
1135 * range inside of a page needs to be written.
1137 * We can skip this if the page is up to date - it's already
1138 * been zero'd from being read in as a hole.
1140 if (new && !PageUptodate(page
))
1141 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1142 cpos
, user_data_from
, user_data_to
);
1144 flush_dcache_page(page
);
1151 * This function will only grab one clusters worth of pages.
1153 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1154 struct ocfs2_write_ctxt
*wc
,
1155 u32 cpos
, loff_t user_pos
, int new,
1156 struct page
*mmap_page
)
1159 unsigned long start
, target_index
, index
;
1160 struct inode
*inode
= mapping
->host
;
1162 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1165 * Figure out how many pages we'll be manipulating here. For
1166 * non allocating write, we just change the one
1167 * page. Otherwise, we'll need a whole clusters worth.
1170 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1171 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1173 wc
->w_num_pages
= 1;
1174 start
= target_index
;
1177 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1180 if (index
== target_index
&& mmap_page
) {
1182 * ocfs2_pagemkwrite() is a little different
1183 * and wants us to directly use the page
1186 lock_page(mmap_page
);
1188 if (mmap_page
->mapping
!= mapping
) {
1189 unlock_page(mmap_page
);
1191 * Sanity check - the locking in
1192 * ocfs2_pagemkwrite() should ensure
1193 * that this code doesn't trigger.
1200 page_cache_get(mmap_page
);
1201 wc
->w_pages
[i
] = mmap_page
;
1203 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1205 if (!wc
->w_pages
[i
]) {
1212 if (index
== target_index
)
1213 wc
->w_target_page
= wc
->w_pages
[i
];
1220 * Prepare a single cluster for write one cluster into the file.
1222 static int ocfs2_write_cluster(struct address_space
*mapping
,
1223 u32 phys
, unsigned int unwritten
,
1224 unsigned int should_zero
,
1225 struct ocfs2_alloc_context
*data_ac
,
1226 struct ocfs2_alloc_context
*meta_ac
,
1227 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1228 loff_t user_pos
, unsigned user_len
)
1231 u64 v_blkno
, p_blkno
;
1232 struct inode
*inode
= mapping
->host
;
1233 struct ocfs2_extent_tree et
;
1235 new = phys
== 0 ? 1 : 0;
1240 * This is safe to call with the page locks - it won't take
1241 * any additional semaphores or cluster locks.
1244 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1245 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1246 wc
->w_handle
, data_ac
,
1249 * This shouldn't happen because we must have already
1250 * calculated the correct meta data allocation required. The
1251 * internal tree allocation code should know how to increase
1252 * transaction credits itself.
1254 * If need be, we could handle -EAGAIN for a
1255 * RESTART_TRANS here.
1257 mlog_bug_on_msg(ret
== -EAGAIN
,
1258 "Inode %llu: EAGAIN return during allocation.\n",
1259 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1264 } else if (unwritten
) {
1265 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1267 ret
= ocfs2_mark_extent_written(inode
, &et
,
1268 wc
->w_handle
, cpos
, 1, phys
,
1269 meta_ac
, &wc
->w_dealloc
);
1277 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1279 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1282 * The only reason this should fail is due to an inability to
1283 * find the extent added.
1285 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1288 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1289 "at logical block %llu",
1290 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1291 (unsigned long long)v_blkno
);
1295 BUG_ON(p_blkno
== 0);
1297 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1300 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1301 wc
->w_pages
[i
], cpos
,
1312 * We only have cleanup to do in case of allocating write.
1315 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1322 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1323 struct ocfs2_alloc_context
*data_ac
,
1324 struct ocfs2_alloc_context
*meta_ac
,
1325 struct ocfs2_write_ctxt
*wc
,
1326 loff_t pos
, unsigned len
)
1330 unsigned int local_len
= len
;
1331 struct ocfs2_write_cluster_desc
*desc
;
1332 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1334 for (i
= 0; i
< wc
->w_clen
; i
++) {
1335 desc
= &wc
->w_desc
[i
];
1338 * We have to make sure that the total write passed in
1339 * doesn't extend past a single cluster.
1342 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1343 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1344 local_len
= osb
->s_clustersize
- cluster_off
;
1346 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1350 wc
, desc
->c_cpos
, pos
, local_len
);
1366 * ocfs2_write_end() wants to know which parts of the target page it
1367 * should complete the write on. It's easiest to compute them ahead of
1368 * time when a more complete view of the write is available.
1370 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1371 struct ocfs2_write_ctxt
*wc
,
1372 loff_t pos
, unsigned len
, int alloc
)
1374 struct ocfs2_write_cluster_desc
*desc
;
1376 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1377 wc
->w_target_to
= wc
->w_target_from
+ len
;
1383 * Allocating write - we may have different boundaries based
1384 * on page size and cluster size.
1386 * NOTE: We can no longer compute one value from the other as
1387 * the actual write length and user provided length may be
1391 if (wc
->w_large_pages
) {
1393 * We only care about the 1st and last cluster within
1394 * our range and whether they should be zero'd or not. Either
1395 * value may be extended out to the start/end of a
1396 * newly allocated cluster.
1398 desc
= &wc
->w_desc
[0];
1399 if (desc
->c_needs_zero
)
1400 ocfs2_figure_cluster_boundaries(osb
,
1405 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1406 if (desc
->c_needs_zero
)
1407 ocfs2_figure_cluster_boundaries(osb
,
1412 wc
->w_target_from
= 0;
1413 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1418 * Populate each single-cluster write descriptor in the write context
1419 * with information about the i/o to be done.
1421 * Returns the number of clusters that will have to be allocated, as
1422 * well as a worst case estimate of the number of extent records that
1423 * would have to be created during a write to an unwritten region.
1425 static int ocfs2_populate_write_desc(struct inode
*inode
,
1426 struct ocfs2_write_ctxt
*wc
,
1427 unsigned int *clusters_to_alloc
,
1428 unsigned int *extents_to_split
)
1431 struct ocfs2_write_cluster_desc
*desc
;
1432 unsigned int num_clusters
= 0;
1433 unsigned int ext_flags
= 0;
1437 *clusters_to_alloc
= 0;
1438 *extents_to_split
= 0;
1440 for (i
= 0; i
< wc
->w_clen
; i
++) {
1441 desc
= &wc
->w_desc
[i
];
1442 desc
->c_cpos
= wc
->w_cpos
+ i
;
1444 if (num_clusters
== 0) {
1446 * Need to look up the next extent record.
1448 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1449 &num_clusters
, &ext_flags
);
1455 /* We should already CoW the refcountd extent. */
1456 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1459 * Assume worst case - that we're writing in
1460 * the middle of the extent.
1462 * We can assume that the write proceeds from
1463 * left to right, in which case the extent
1464 * insert code is smart enough to coalesce the
1465 * next splits into the previous records created.
1467 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1468 *extents_to_split
= *extents_to_split
+ 2;
1471 * Only increment phys if it doesn't describe
1478 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1479 * file that got extended. w_first_new_cpos tells us
1480 * where the newly allocated clusters are so we can
1483 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1485 desc
->c_needs_zero
= 1;
1488 desc
->c_phys
= phys
;
1491 desc
->c_needs_zero
= 1;
1492 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1495 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1496 desc
->c_unwritten
= 1;
1497 desc
->c_needs_zero
= 1;
1508 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1509 struct inode
*inode
,
1510 struct ocfs2_write_ctxt
*wc
)
1513 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1516 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1518 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1525 * If we don't set w_num_pages then this page won't get unlocked
1526 * and freed on cleanup of the write context.
1528 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1529 wc
->w_num_pages
= 1;
1531 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1532 if (IS_ERR(handle
)) {
1533 ret
= PTR_ERR(handle
);
1538 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1539 OCFS2_JOURNAL_ACCESS_WRITE
);
1541 ocfs2_commit_trans(osb
, handle
);
1547 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1548 ocfs2_set_inode_data_inline(inode
, di
);
1550 if (!PageUptodate(page
)) {
1551 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1553 ocfs2_commit_trans(osb
, handle
);
1559 wc
->w_handle
= handle
;
1564 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1566 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1568 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1573 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1574 struct inode
*inode
, loff_t pos
,
1575 unsigned len
, struct page
*mmap_page
,
1576 struct ocfs2_write_ctxt
*wc
)
1578 int ret
, written
= 0;
1579 loff_t end
= pos
+ len
;
1580 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1581 struct ocfs2_dinode
*di
= NULL
;
1583 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1584 (unsigned long long)oi
->ip_blkno
, len
, (unsigned long long)pos
,
1585 oi
->ip_dyn_features
);
1588 * Handle inodes which already have inline data 1st.
1590 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1591 if (mmap_page
== NULL
&&
1592 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1593 goto do_inline_write
;
1596 * The write won't fit - we have to give this inode an
1597 * inline extent list now.
1599 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1606 * Check whether the inode can accept inline data.
1608 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1612 * Check whether the write can fit.
1614 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1616 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1620 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1627 * This signals to the caller that the data can be written
1632 return written
? written
: ret
;
1636 * This function only does anything for file systems which can't
1637 * handle sparse files.
1639 * What we want to do here is fill in any hole between the current end
1640 * of allocation and the end of our write. That way the rest of the
1641 * write path can treat it as an non-allocating write, which has no
1642 * special case code for sparse/nonsparse files.
1644 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
, loff_t pos
,
1646 struct ocfs2_write_ctxt
*wc
)
1649 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1650 loff_t newsize
= pos
+ len
;
1652 if (ocfs2_sparse_alloc(osb
))
1655 if (newsize
<= i_size_read(inode
))
1658 ret
= ocfs2_extend_no_holes(inode
, newsize
, pos
);
1662 wc
->w_first_new_cpos
=
1663 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1668 int ocfs2_write_begin_nolock(struct address_space
*mapping
,
1669 loff_t pos
, unsigned len
, unsigned flags
,
1670 struct page
**pagep
, void **fsdata
,
1671 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1673 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1674 unsigned int clusters_to_alloc
, extents_to_split
;
1675 struct ocfs2_write_ctxt
*wc
;
1676 struct inode
*inode
= mapping
->host
;
1677 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1678 struct ocfs2_dinode
*di
;
1679 struct ocfs2_alloc_context
*data_ac
= NULL
;
1680 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1682 struct ocfs2_extent_tree et
;
1684 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1690 if (ocfs2_supports_inline_data(osb
)) {
1691 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1703 ret
= ocfs2_expand_nonsparse_inode(inode
, pos
, len
, wc
);
1709 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1713 } else if (ret
== 1) {
1714 ret
= ocfs2_refcount_cow(inode
, di_bh
,
1715 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1722 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1729 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1732 * We set w_target_from, w_target_to here so that
1733 * ocfs2_write_end() knows which range in the target page to
1734 * write out. An allocation requires that we write the entire
1737 if (clusters_to_alloc
|| extents_to_split
) {
1739 * XXX: We are stretching the limits of
1740 * ocfs2_lock_allocators(). It greatly over-estimates
1741 * the work to be done.
1743 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1744 " clusters_to_add = %u, extents_to_split = %u\n",
1745 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1746 (long long)i_size_read(inode
), le32_to_cpu(di
->i_clusters
),
1747 clusters_to_alloc
, extents_to_split
);
1749 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1751 ret
= ocfs2_lock_allocators(inode
, &et
,
1752 clusters_to_alloc
, extents_to_split
,
1753 &data_ac
, &meta_ac
);
1759 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1766 * We have to zero sparse allocated clusters, unwritten extent clusters,
1767 * and non-sparse clusters we just extended. For non-sparse writes,
1768 * we know zeros will only be needed in the first and/or last cluster.
1770 if (clusters_to_alloc
|| extents_to_split
||
1771 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
1772 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
1773 cluster_of_pages
= 1;
1775 cluster_of_pages
= 0;
1777 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1779 handle
= ocfs2_start_trans(osb
, credits
);
1780 if (IS_ERR(handle
)) {
1781 ret
= PTR_ERR(handle
);
1786 wc
->w_handle
= handle
;
1788 if (clusters_to_alloc
&& vfs_dq_alloc_space_nodirty(inode
,
1789 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
))) {
1794 * We don't want this to fail in ocfs2_write_end(), so do it
1797 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1798 OCFS2_JOURNAL_ACCESS_WRITE
);
1805 * Fill our page array first. That way we've grabbed enough so
1806 * that we can zero and flush if we error after adding the
1809 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
,
1810 cluster_of_pages
, mmap_page
);
1816 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1824 ocfs2_free_alloc_context(data_ac
);
1826 ocfs2_free_alloc_context(meta_ac
);
1829 *pagep
= wc
->w_target_page
;
1833 if (clusters_to_alloc
)
1834 vfs_dq_free_space(inode
,
1835 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1837 ocfs2_commit_trans(osb
, handle
);
1840 ocfs2_free_write_ctxt(wc
);
1843 ocfs2_free_alloc_context(data_ac
);
1845 ocfs2_free_alloc_context(meta_ac
);
1849 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1850 loff_t pos
, unsigned len
, unsigned flags
,
1851 struct page
**pagep
, void **fsdata
)
1854 struct buffer_head
*di_bh
= NULL
;
1855 struct inode
*inode
= mapping
->host
;
1857 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1864 * Take alloc sem here to prevent concurrent lookups. That way
1865 * the mapping, zeroing and tree manipulation within
1866 * ocfs2_write() will be safe against ->readpage(). This
1867 * should also serve to lock out allocation from a shared
1870 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1872 ret
= ocfs2_write_begin_nolock(mapping
, pos
, len
, flags
, pagep
,
1873 fsdata
, di_bh
, NULL
);
1884 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1887 ocfs2_inode_unlock(inode
, 1);
1892 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1893 unsigned len
, unsigned *copied
,
1894 struct ocfs2_dinode
*di
,
1895 struct ocfs2_write_ctxt
*wc
)
1899 if (unlikely(*copied
< len
)) {
1900 if (!PageUptodate(wc
->w_target_page
)) {
1906 kaddr
= kmap_atomic(wc
->w_target_page
, KM_USER0
);
1907 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1908 kunmap_atomic(kaddr
, KM_USER0
);
1910 mlog(0, "Data written to inode at offset %llu. "
1911 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1912 (unsigned long long)pos
, *copied
,
1913 le16_to_cpu(di
->id2
.i_data
.id_count
),
1914 le16_to_cpu(di
->i_dyn_features
));
1917 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1918 loff_t pos
, unsigned len
, unsigned copied
,
1919 struct page
*page
, void *fsdata
)
1922 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1923 struct inode
*inode
= mapping
->host
;
1924 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1925 struct ocfs2_write_ctxt
*wc
= fsdata
;
1926 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1927 handle_t
*handle
= wc
->w_handle
;
1928 struct page
*tmppage
;
1930 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1931 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1932 goto out_write_size
;
1935 if (unlikely(copied
< len
)) {
1936 if (!PageUptodate(wc
->w_target_page
))
1939 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
1942 flush_dcache_page(wc
->w_target_page
);
1944 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1945 tmppage
= wc
->w_pages
[i
];
1947 if (tmppage
== wc
->w_target_page
) {
1948 from
= wc
->w_target_from
;
1949 to
= wc
->w_target_to
;
1951 BUG_ON(from
> PAGE_CACHE_SIZE
||
1952 to
> PAGE_CACHE_SIZE
||
1956 * Pages adjacent to the target (if any) imply
1957 * a hole-filling write in which case we want
1958 * to flush their entire range.
1961 to
= PAGE_CACHE_SIZE
;
1964 if (page_has_buffers(tmppage
)) {
1965 if (ocfs2_should_order_data(inode
))
1966 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1967 block_commit_write(tmppage
, from
, to
);
1973 if (pos
> inode
->i_size
) {
1974 i_size_write(inode
, pos
);
1975 mark_inode_dirty(inode
);
1977 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1978 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
1979 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1980 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1981 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1982 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
1984 ocfs2_commit_trans(osb
, handle
);
1986 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
1988 ocfs2_free_write_ctxt(wc
);
1993 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
1994 loff_t pos
, unsigned len
, unsigned copied
,
1995 struct page
*page
, void *fsdata
)
1998 struct inode
*inode
= mapping
->host
;
2000 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2002 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2003 ocfs2_inode_unlock(inode
, 1);
2008 const struct address_space_operations ocfs2_aops
= {
2009 .readpage
= ocfs2_readpage
,
2010 .readpages
= ocfs2_readpages
,
2011 .writepage
= ocfs2_writepage
,
2012 .write_begin
= ocfs2_write_begin
,
2013 .write_end
= ocfs2_write_end
,
2015 .sync_page
= block_sync_page
,
2016 .direct_IO
= ocfs2_direct_IO
,
2017 .invalidatepage
= ocfs2_invalidatepage
,
2018 .releasepage
= ocfs2_releasepage
,
2019 .migratepage
= buffer_migrate_page
,
2020 .is_partially_uptodate
= block_is_partially_uptodate
,