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>
31 #define MLOG_MASK_PREFIX ML_FILE_IO
32 #include <cluster/masklog.h>
39 #include "extent_map.h"
47 #include "buffer_head_io.h"
49 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
50 struct buffer_head
*bh_result
, int create
)
54 struct ocfs2_dinode
*fe
= NULL
;
55 struct buffer_head
*bh
= NULL
;
56 struct buffer_head
*buffer_cache_bh
= NULL
;
57 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
60 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
61 (unsigned long long)iblock
, bh_result
, create
);
63 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
65 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
66 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
67 (unsigned long long)iblock
);
71 status
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
),
72 OCFS2_I(inode
)->ip_blkno
,
73 &bh
, OCFS2_BH_CACHED
, inode
);
78 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
80 if (!OCFS2_IS_VALID_DINODE(fe
)) {
81 mlog(ML_ERROR
, "Invalid dinode #%llu: signature = %.*s\n",
82 (unsigned long long)le64_to_cpu(fe
->i_blkno
), 7,
87 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
88 le32_to_cpu(fe
->i_clusters
))) {
89 mlog(ML_ERROR
, "block offset is outside the allocated size: "
90 "%llu\n", (unsigned long long)iblock
);
94 /* We don't use the page cache to create symlink data, so if
95 * need be, copy it over from the buffer cache. */
96 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
97 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
99 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
100 if (!buffer_cache_bh
) {
101 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
105 /* we haven't locked out transactions, so a commit
106 * could've happened. Since we've got a reference on
107 * the bh, even if it commits while we're doing the
108 * copy, the data is still good. */
109 if (buffer_jbd(buffer_cache_bh
)
110 && ocfs2_inode_is_new(inode
)) {
111 kaddr
= kmap_atomic(bh_result
->b_page
, KM_USER0
);
113 mlog(ML_ERROR
, "couldn't kmap!\n");
116 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
117 buffer_cache_bh
->b_data
,
119 kunmap_atomic(kaddr
, KM_USER0
);
120 set_buffer_uptodate(bh_result
);
122 brelse(buffer_cache_bh
);
125 map_bh(bh_result
, inode
->i_sb
,
126 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
138 static int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
139 struct buffer_head
*bh_result
, int create
)
142 unsigned int ext_flags
;
143 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
144 u64 p_blkno
, count
, past_eof
;
145 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
147 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
148 (unsigned long long)iblock
, bh_result
, create
);
150 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
151 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
152 inode
, inode
->i_ino
);
154 if (S_ISLNK(inode
->i_mode
)) {
155 /* this always does I/O for some reason. */
156 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
160 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
163 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
164 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
165 (unsigned long long)p_blkno
);
169 if (max_blocks
< count
)
173 * ocfs2 never allocates in this function - the only time we
174 * need to use BH_New is when we're extending i_size on a file
175 * system which doesn't support holes, in which case BH_New
176 * allows block_prepare_write() to zero.
178 * If we see this on a sparse file system, then a truncate has
179 * raced us and removed the cluster. In this case, we clear
180 * the buffers dirty and uptodate bits and let the buffer code
181 * ignore it as a hole.
183 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
184 clear_buffer_dirty(bh_result
);
185 clear_buffer_uptodate(bh_result
);
189 /* Treat the unwritten extent as a hole for zeroing purposes. */
190 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
191 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
193 bh_result
->b_size
= count
<< inode
->i_blkbits
;
195 if (!ocfs2_sparse_alloc(osb
)) {
199 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
200 (unsigned long long)iblock
,
201 (unsigned long long)p_blkno
,
202 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
203 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
207 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
208 mlog(0, "Inode %lu, past_eof = %llu\n", inode
->i_ino
,
209 (unsigned long long)past_eof
);
211 if (create
&& (iblock
>= past_eof
))
212 set_buffer_new(bh_result
);
223 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
224 struct buffer_head
*di_bh
)
228 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
230 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
231 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
232 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
236 size
= i_size_read(inode
);
238 if (size
> PAGE_CACHE_SIZE
||
239 size
> ocfs2_max_inline_data(inode
->i_sb
)) {
240 ocfs2_error(inode
->i_sb
,
241 "Inode %llu has with inline data has bad size: %Lu",
242 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
243 (unsigned long long)size
);
247 kaddr
= kmap_atomic(page
, KM_USER0
);
249 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
250 /* Clear the remaining part of the page */
251 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
252 flush_dcache_page(page
);
253 kunmap_atomic(kaddr
, KM_USER0
);
255 SetPageUptodate(page
);
260 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
263 struct buffer_head
*di_bh
= NULL
;
264 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
266 BUG_ON(!PageLocked(page
));
267 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
269 ret
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &di_bh
,
270 OCFS2_BH_CACHED
, inode
);
276 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
284 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
286 struct inode
*inode
= page
->mapping
->host
;
287 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
288 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
291 mlog_entry("(0x%p, %lu)\n", file
, (page
? page
->index
: 0));
293 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
295 if (ret
== AOP_TRUNCATED_PAGE
)
301 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
302 ret
= AOP_TRUNCATED_PAGE
;
303 goto out_inode_unlock
;
307 * i_size might have just been updated as we grabed the meta lock. We
308 * might now be discovering a truncate that hit on another node.
309 * block_read_full_page->get_block freaks out if it is asked to read
310 * beyond the end of a file, so we check here. Callers
311 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
312 * and notice that the page they just read isn't needed.
314 * XXX sys_readahead() seems to get that wrong?
316 if (start
>= i_size_read(inode
)) {
317 zero_user(page
, 0, PAGE_SIZE
);
318 SetPageUptodate(page
);
323 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
324 ret
= ocfs2_readpage_inline(inode
, page
);
326 ret
= block_read_full_page(page
, ocfs2_get_block
);
330 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
332 ocfs2_inode_unlock(inode
, 0);
341 * This is used only for read-ahead. Failures or difficult to handle
342 * situations are safe to ignore.
344 * Right now, we don't bother with BH_Boundary - in-inode extent lists
345 * are quite large (243 extents on 4k blocks), so most inodes don't
346 * grow out to a tree. If need be, detecting boundary extents could
347 * trivially be added in a future version of ocfs2_get_block().
349 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
350 struct list_head
*pages
, unsigned nr_pages
)
353 struct inode
*inode
= mapping
->host
;
354 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
359 * Use the nonblocking flag for the dlm code to avoid page
360 * lock inversion, but don't bother with retrying.
362 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
366 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
367 ocfs2_inode_unlock(inode
, 0);
372 * Don't bother with inline-data. There isn't anything
373 * to read-ahead in that case anyway...
375 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
379 * Check whether a remote node truncated this file - we just
380 * drop out in that case as it's not worth handling here.
382 last
= list_entry(pages
->prev
, struct page
, lru
);
383 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
384 if (start
>= i_size_read(inode
))
387 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
390 up_read(&oi
->ip_alloc_sem
);
391 ocfs2_inode_unlock(inode
, 0);
396 /* Note: Because we don't support holes, our allocation has
397 * already happened (allocation writes zeros to the file data)
398 * so we don't have to worry about ordered writes in
401 * ->writepage is called during the process of invalidating the page cache
402 * during blocked lock processing. It can't block on any cluster locks
403 * to during block mapping. It's relying on the fact that the block
404 * mapping can't have disappeared under the dirty pages that it is
405 * being asked to write back.
407 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
411 mlog_entry("(0x%p)\n", page
);
413 ret
= block_write_full_page(page
, ocfs2_get_block
, wbc
);
421 * This is called from ocfs2_write_zero_page() which has handled it's
422 * own cluster locking and has ensured allocation exists for those
423 * blocks to be written.
425 int ocfs2_prepare_write_nolock(struct inode
*inode
, struct page
*page
,
426 unsigned from
, unsigned to
)
430 ret
= block_prepare_write(page
, from
, to
, ocfs2_get_block
);
435 /* Taken from ext3. We don't necessarily need the full blown
436 * functionality yet, but IMHO it's better to cut and paste the whole
437 * thing so we can avoid introducing our own bugs (and easily pick up
438 * their fixes when they happen) --Mark */
439 int walk_page_buffers( handle_t
*handle
,
440 struct buffer_head
*head
,
444 int (*fn
)( handle_t
*handle
,
445 struct buffer_head
*bh
))
447 struct buffer_head
*bh
;
448 unsigned block_start
, block_end
;
449 unsigned blocksize
= head
->b_size
;
451 struct buffer_head
*next
;
453 for ( bh
= head
, block_start
= 0;
454 ret
== 0 && (bh
!= head
|| !block_start
);
455 block_start
= block_end
, bh
= next
)
457 next
= bh
->b_this_page
;
458 block_end
= block_start
+ blocksize
;
459 if (block_end
<= from
|| block_start
>= to
) {
460 if (partial
&& !buffer_uptodate(bh
))
464 err
= (*fn
)(handle
, bh
);
471 handle_t
*ocfs2_start_walk_page_trans(struct inode
*inode
,
476 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
480 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
481 if (IS_ERR(handle
)) {
487 if (ocfs2_should_order_data(inode
)) {
488 ret
= ocfs2_jbd2_file_inode(handle
, inode
);
489 #ifdef CONFIG_OCFS2_COMPAT_JBD
490 ret
= walk_page_buffers(handle
,
493 ocfs2_journal_dirty_data
);
501 ocfs2_commit_trans(osb
, handle
);
502 handle
= ERR_PTR(ret
);
507 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
512 struct inode
*inode
= mapping
->host
;
514 mlog_entry("(block = %llu)\n", (unsigned long long)block
);
516 /* We don't need to lock journal system files, since they aren't
517 * accessed concurrently from multiple nodes.
519 if (!INODE_JOURNAL(inode
)) {
520 err
= ocfs2_inode_lock(inode
, NULL
, 0);
526 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
529 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
530 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
533 if (!INODE_JOURNAL(inode
)) {
534 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
535 ocfs2_inode_unlock(inode
, 0);
539 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
540 (unsigned long long)block
);
546 status
= err
? 0 : p_blkno
;
548 mlog_exit((int)status
);
554 * TODO: Make this into a generic get_blocks function.
556 * From do_direct_io in direct-io.c:
557 * "So what we do is to permit the ->get_blocks function to populate
558 * bh.b_size with the size of IO which is permitted at this offset and
561 * This function is called directly from get_more_blocks in direct-io.c.
563 * called like this: dio->get_blocks(dio->inode, fs_startblk,
564 * fs_count, map_bh, dio->rw == WRITE);
566 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
567 struct buffer_head
*bh_result
, int create
)
570 u64 p_blkno
, inode_blocks
, contig_blocks
;
571 unsigned int ext_flags
;
572 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
573 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
575 /* This function won't even be called if the request isn't all
576 * nicely aligned and of the right size, so there's no need
577 * for us to check any of that. */
579 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
582 * Any write past EOF is not allowed because we'd be extending.
584 if (create
&& (iblock
+ max_blocks
) > inode_blocks
) {
589 /* This figures out the size of the next contiguous block, and
590 * our logical offset */
591 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
592 &contig_blocks
, &ext_flags
);
594 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
595 (unsigned long long)iblock
);
600 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)) && !p_blkno
&& create
) {
601 ocfs2_error(inode
->i_sb
,
602 "Inode %llu has a hole at block %llu\n",
603 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
604 (unsigned long long)iblock
);
610 * get_more_blocks() expects us to describe a hole by clearing
611 * the mapped bit on bh_result().
613 * Consider an unwritten extent as a hole.
615 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
616 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
619 * ocfs2_prepare_inode_for_write() should have caught
620 * the case where we'd be filling a hole and triggered
621 * a buffered write instead.
629 clear_buffer_mapped(bh_result
);
632 /* make sure we don't map more than max_blocks blocks here as
633 that's all the kernel will handle at this point. */
634 if (max_blocks
< contig_blocks
)
635 contig_blocks
= max_blocks
;
636 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
642 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
643 * particularly interested in the aio/dio case. Like the core uses
644 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
645 * truncation on another.
647 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
652 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
655 /* this io's submitter should not have unlocked this before we could */
656 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
658 ocfs2_iocb_clear_rw_locked(iocb
);
660 level
= ocfs2_iocb_rw_locked_level(iocb
);
662 up_read(&inode
->i_alloc_sem
);
663 ocfs2_rw_unlock(inode
, level
);
667 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
668 * from ext3. PageChecked() bits have been removed as OCFS2 does not
669 * do journalled data.
671 static void ocfs2_invalidatepage(struct page
*page
, unsigned long offset
)
673 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
675 jbd2_journal_invalidatepage(journal
, page
, offset
);
678 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
680 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
682 if (!page_has_buffers(page
))
684 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
687 static ssize_t
ocfs2_direct_IO(int rw
,
689 const struct iovec
*iov
,
691 unsigned long nr_segs
)
693 struct file
*file
= iocb
->ki_filp
;
694 struct inode
*inode
= file
->f_path
.dentry
->d_inode
->i_mapping
->host
;
700 * Fallback to buffered I/O if we see an inode without
703 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
706 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
707 inode
->i_sb
->s_bdev
, iov
, offset
,
709 ocfs2_direct_IO_get_blocks
,
716 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
721 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
723 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
726 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
728 cluster_start
= cpos
% cpp
;
729 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
731 cluster_end
= cluster_start
+ osb
->s_clustersize
;
734 BUG_ON(cluster_start
> PAGE_SIZE
);
735 BUG_ON(cluster_end
> PAGE_SIZE
);
738 *start
= cluster_start
;
744 * 'from' and 'to' are the region in the page to avoid zeroing.
746 * If pagesize > clustersize, this function will avoid zeroing outside
747 * of the cluster boundary.
749 * from == to == 0 is code for "zero the entire cluster region"
751 static void ocfs2_clear_page_regions(struct page
*page
,
752 struct ocfs2_super
*osb
, u32 cpos
,
753 unsigned from
, unsigned to
)
756 unsigned int cluster_start
, cluster_end
;
758 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
760 kaddr
= kmap_atomic(page
, KM_USER0
);
763 if (from
> cluster_start
)
764 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
765 if (to
< cluster_end
)
766 memset(kaddr
+ to
, 0, cluster_end
- to
);
768 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
771 kunmap_atomic(kaddr
, KM_USER0
);
775 * Nonsparse file systems fully allocate before we get to the write
776 * code. This prevents ocfs2_write() from tagging the write as an
777 * allocating one, which means ocfs2_map_page_blocks() might try to
778 * read-in the blocks at the tail of our file. Avoid reading them by
779 * testing i_size against each block offset.
781 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
782 unsigned int block_start
)
784 u64 offset
= page_offset(page
) + block_start
;
786 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
789 if (i_size_read(inode
) > offset
)
796 * Some of this taken from block_prepare_write(). We already have our
797 * mapping by now though, and the entire write will be allocating or
798 * it won't, so not much need to use BH_New.
800 * This will also skip zeroing, which is handled externally.
802 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
803 struct inode
*inode
, unsigned int from
,
804 unsigned int to
, int new)
807 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
808 unsigned int block_end
, block_start
;
809 unsigned int bsize
= 1 << inode
->i_blkbits
;
811 if (!page_has_buffers(page
))
812 create_empty_buffers(page
, bsize
, 0);
814 head
= page_buffers(page
);
815 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
816 bh
= bh
->b_this_page
, block_start
+= bsize
) {
817 block_end
= block_start
+ bsize
;
819 clear_buffer_new(bh
);
822 * Ignore blocks outside of our i/o range -
823 * they may belong to unallocated clusters.
825 if (block_start
>= to
|| block_end
<= from
) {
826 if (PageUptodate(page
))
827 set_buffer_uptodate(bh
);
832 * For an allocating write with cluster size >= page
833 * size, we always write the entire page.
838 if (!buffer_mapped(bh
)) {
839 map_bh(bh
, inode
->i_sb
, *p_blkno
);
840 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
843 if (PageUptodate(page
)) {
844 if (!buffer_uptodate(bh
))
845 set_buffer_uptodate(bh
);
846 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
848 ocfs2_should_read_blk(inode
, page
, block_start
) &&
849 (block_start
< from
|| block_end
> to
)) {
850 ll_rw_block(READ
, 1, &bh
);
854 *p_blkno
= *p_blkno
+ 1;
858 * If we issued read requests - let them complete.
860 while(wait_bh
> wait
) {
861 wait_on_buffer(*--wait_bh
);
862 if (!buffer_uptodate(*wait_bh
))
866 if (ret
== 0 || !new)
870 * If we get -EIO above, zero out any newly allocated blocks
871 * to avoid exposing stale data.
876 block_end
= block_start
+ bsize
;
877 if (block_end
<= from
)
879 if (block_start
>= to
)
882 zero_user(page
, block_start
, bh
->b_size
);
883 set_buffer_uptodate(bh
);
884 mark_buffer_dirty(bh
);
887 block_start
= block_end
;
888 bh
= bh
->b_this_page
;
889 } while (bh
!= head
);
894 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
895 #define OCFS2_MAX_CTXT_PAGES 1
897 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
900 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
903 * Describe the state of a single cluster to be written to.
905 struct ocfs2_write_cluster_desc
{
909 * Give this a unique field because c_phys eventually gets
913 unsigned c_unwritten
;
916 static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc
*d
)
918 return d
->c_new
|| d
->c_unwritten
;
921 struct ocfs2_write_ctxt
{
922 /* Logical cluster position / len of write */
926 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
929 * This is true if page_size > cluster_size.
931 * It triggers a set of special cases during write which might
932 * have to deal with allocating writes to partial pages.
934 unsigned int w_large_pages
;
937 * Pages involved in this write.
939 * w_target_page is the page being written to by the user.
941 * w_pages is an array of pages which always contains
942 * w_target_page, and in the case of an allocating write with
943 * page_size < cluster size, it will contain zero'd and mapped
944 * pages adjacent to w_target_page which need to be written
945 * out in so that future reads from that region will get
948 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
949 unsigned int w_num_pages
;
950 struct page
*w_target_page
;
953 * ocfs2_write_end() uses this to know what the real range to
954 * write in the target should be.
956 unsigned int w_target_from
;
957 unsigned int w_target_to
;
960 * We could use journal_current_handle() but this is cleaner,
965 struct buffer_head
*w_di_bh
;
967 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
970 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
974 for(i
= 0; i
< num_pages
; i
++) {
976 unlock_page(pages
[i
]);
977 mark_page_accessed(pages
[i
]);
978 page_cache_release(pages
[i
]);
983 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
985 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
991 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
992 struct ocfs2_super
*osb
, loff_t pos
,
993 unsigned len
, struct buffer_head
*di_bh
)
996 struct ocfs2_write_ctxt
*wc
;
998 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
1002 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
1003 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
1004 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
1006 wc
->w_di_bh
= di_bh
;
1008 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
1009 wc
->w_large_pages
= 1;
1011 wc
->w_large_pages
= 0;
1013 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
1021 * If a page has any new buffers, zero them out here, and mark them uptodate
1022 * and dirty so they'll be written out (in order to prevent uninitialised
1023 * block data from leaking). And clear the new bit.
1025 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1027 unsigned int block_start
, block_end
;
1028 struct buffer_head
*head
, *bh
;
1030 BUG_ON(!PageLocked(page
));
1031 if (!page_has_buffers(page
))
1034 bh
= head
= page_buffers(page
);
1037 block_end
= block_start
+ bh
->b_size
;
1039 if (buffer_new(bh
)) {
1040 if (block_end
> from
&& block_start
< to
) {
1041 if (!PageUptodate(page
)) {
1042 unsigned start
, end
;
1044 start
= max(from
, block_start
);
1045 end
= min(to
, block_end
);
1047 zero_user_segment(page
, start
, end
);
1048 set_buffer_uptodate(bh
);
1051 clear_buffer_new(bh
);
1052 mark_buffer_dirty(bh
);
1056 block_start
= block_end
;
1057 bh
= bh
->b_this_page
;
1058 } while (bh
!= head
);
1062 * Only called when we have a failure during allocating write to write
1063 * zero's to the newly allocated region.
1065 static void ocfs2_write_failure(struct inode
*inode
,
1066 struct ocfs2_write_ctxt
*wc
,
1067 loff_t user_pos
, unsigned user_len
)
1070 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1071 to
= user_pos
+ user_len
;
1072 struct page
*tmppage
;
1074 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1076 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1077 tmppage
= wc
->w_pages
[i
];
1079 if (page_has_buffers(tmppage
)) {
1080 if (ocfs2_should_order_data(inode
)) {
1081 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1082 #ifdef CONFIG_OCFS2_COMPAT_JBD
1083 walk_page_buffers(wc
->w_handle
,
1084 page_buffers(tmppage
),
1086 ocfs2_journal_dirty_data
);
1090 block_commit_write(tmppage
, from
, to
);
1095 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1096 struct ocfs2_write_ctxt
*wc
,
1097 struct page
*page
, u32 cpos
,
1098 loff_t user_pos
, unsigned user_len
,
1102 unsigned int map_from
= 0, map_to
= 0;
1103 unsigned int cluster_start
, cluster_end
;
1104 unsigned int user_data_from
= 0, user_data_to
= 0;
1106 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1107 &cluster_start
, &cluster_end
);
1109 if (page
== wc
->w_target_page
) {
1110 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1111 map_to
= map_from
+ user_len
;
1114 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1115 cluster_start
, cluster_end
,
1118 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1119 map_from
, map_to
, new);
1125 user_data_from
= map_from
;
1126 user_data_to
= map_to
;
1128 map_from
= cluster_start
;
1129 map_to
= cluster_end
;
1133 * If we haven't allocated the new page yet, we
1134 * shouldn't be writing it out without copying user
1135 * data. This is likely a math error from the caller.
1139 map_from
= cluster_start
;
1140 map_to
= cluster_end
;
1142 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1143 cluster_start
, cluster_end
, new);
1151 * Parts of newly allocated pages need to be zero'd.
1153 * Above, we have also rewritten 'to' and 'from' - as far as
1154 * the rest of the function is concerned, the entire cluster
1155 * range inside of a page needs to be written.
1157 * We can skip this if the page is up to date - it's already
1158 * been zero'd from being read in as a hole.
1160 if (new && !PageUptodate(page
))
1161 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1162 cpos
, user_data_from
, user_data_to
);
1164 flush_dcache_page(page
);
1171 * This function will only grab one clusters worth of pages.
1173 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1174 struct ocfs2_write_ctxt
*wc
,
1175 u32 cpos
, loff_t user_pos
, int new,
1176 struct page
*mmap_page
)
1179 unsigned long start
, target_index
, index
;
1180 struct inode
*inode
= mapping
->host
;
1182 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1185 * Figure out how many pages we'll be manipulating here. For
1186 * non allocating write, we just change the one
1187 * page. Otherwise, we'll need a whole clusters worth.
1190 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1191 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1193 wc
->w_num_pages
= 1;
1194 start
= target_index
;
1197 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1200 if (index
== target_index
&& mmap_page
) {
1202 * ocfs2_pagemkwrite() is a little different
1203 * and wants us to directly use the page
1206 lock_page(mmap_page
);
1208 if (mmap_page
->mapping
!= mapping
) {
1209 unlock_page(mmap_page
);
1211 * Sanity check - the locking in
1212 * ocfs2_pagemkwrite() should ensure
1213 * that this code doesn't trigger.
1220 page_cache_get(mmap_page
);
1221 wc
->w_pages
[i
] = mmap_page
;
1223 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1225 if (!wc
->w_pages
[i
]) {
1232 if (index
== target_index
)
1233 wc
->w_target_page
= wc
->w_pages
[i
];
1240 * Prepare a single cluster for write one cluster into the file.
1242 static int ocfs2_write_cluster(struct address_space
*mapping
,
1243 u32 phys
, unsigned int unwritten
,
1244 struct ocfs2_alloc_context
*data_ac
,
1245 struct ocfs2_alloc_context
*meta_ac
,
1246 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1247 loff_t user_pos
, unsigned user_len
)
1249 int ret
, i
, new, should_zero
= 0;
1250 u64 v_blkno
, p_blkno
;
1251 struct inode
*inode
= mapping
->host
;
1252 struct ocfs2_extent_tree et
;
1254 new = phys
== 0 ? 1 : 0;
1255 if (new || unwritten
)
1262 * This is safe to call with the page locks - it won't take
1263 * any additional semaphores or cluster locks.
1266 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1267 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1268 wc
->w_handle
, data_ac
,
1271 * This shouldn't happen because we must have already
1272 * calculated the correct meta data allocation required. The
1273 * internal tree allocation code should know how to increase
1274 * transaction credits itself.
1276 * If need be, we could handle -EAGAIN for a
1277 * RESTART_TRANS here.
1279 mlog_bug_on_msg(ret
== -EAGAIN
,
1280 "Inode %llu: EAGAIN return during allocation.\n",
1281 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1286 } else if (unwritten
) {
1287 ocfs2_init_dinode_extent_tree(&et
, inode
, wc
->w_di_bh
);
1288 ret
= ocfs2_mark_extent_written(inode
, &et
,
1289 wc
->w_handle
, cpos
, 1, phys
,
1290 meta_ac
, &wc
->w_dealloc
);
1298 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1300 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1303 * The only reason this should fail is due to an inability to
1304 * find the extent added.
1306 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1309 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1310 "at logical block %llu",
1311 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1312 (unsigned long long)v_blkno
);
1316 BUG_ON(p_blkno
== 0);
1318 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1321 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1322 wc
->w_pages
[i
], cpos
,
1333 * We only have cleanup to do in case of allocating write.
1336 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1343 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1344 struct ocfs2_alloc_context
*data_ac
,
1345 struct ocfs2_alloc_context
*meta_ac
,
1346 struct ocfs2_write_ctxt
*wc
,
1347 loff_t pos
, unsigned len
)
1351 unsigned int local_len
= len
;
1352 struct ocfs2_write_cluster_desc
*desc
;
1353 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1355 for (i
= 0; i
< wc
->w_clen
; i
++) {
1356 desc
= &wc
->w_desc
[i
];
1359 * We have to make sure that the total write passed in
1360 * doesn't extend past a single cluster.
1363 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1364 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1365 local_len
= osb
->s_clustersize
- cluster_off
;
1367 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1368 desc
->c_unwritten
, data_ac
, meta_ac
,
1369 wc
, desc
->c_cpos
, pos
, local_len
);
1385 * ocfs2_write_end() wants to know which parts of the target page it
1386 * should complete the write on. It's easiest to compute them ahead of
1387 * time when a more complete view of the write is available.
1389 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1390 struct ocfs2_write_ctxt
*wc
,
1391 loff_t pos
, unsigned len
, int alloc
)
1393 struct ocfs2_write_cluster_desc
*desc
;
1395 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1396 wc
->w_target_to
= wc
->w_target_from
+ len
;
1402 * Allocating write - we may have different boundaries based
1403 * on page size and cluster size.
1405 * NOTE: We can no longer compute one value from the other as
1406 * the actual write length and user provided length may be
1410 if (wc
->w_large_pages
) {
1412 * We only care about the 1st and last cluster within
1413 * our range and whether they should be zero'd or not. Either
1414 * value may be extended out to the start/end of a
1415 * newly allocated cluster.
1417 desc
= &wc
->w_desc
[0];
1418 if (ocfs2_should_zero_cluster(desc
))
1419 ocfs2_figure_cluster_boundaries(osb
,
1424 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1425 if (ocfs2_should_zero_cluster(desc
))
1426 ocfs2_figure_cluster_boundaries(osb
,
1431 wc
->w_target_from
= 0;
1432 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1437 * Populate each single-cluster write descriptor in the write context
1438 * with information about the i/o to be done.
1440 * Returns the number of clusters that will have to be allocated, as
1441 * well as a worst case estimate of the number of extent records that
1442 * would have to be created during a write to an unwritten region.
1444 static int ocfs2_populate_write_desc(struct inode
*inode
,
1445 struct ocfs2_write_ctxt
*wc
,
1446 unsigned int *clusters_to_alloc
,
1447 unsigned int *extents_to_split
)
1450 struct ocfs2_write_cluster_desc
*desc
;
1451 unsigned int num_clusters
= 0;
1452 unsigned int ext_flags
= 0;
1456 *clusters_to_alloc
= 0;
1457 *extents_to_split
= 0;
1459 for (i
= 0; i
< wc
->w_clen
; i
++) {
1460 desc
= &wc
->w_desc
[i
];
1461 desc
->c_cpos
= wc
->w_cpos
+ i
;
1463 if (num_clusters
== 0) {
1465 * Need to look up the next extent record.
1467 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1468 &num_clusters
, &ext_flags
);
1475 * Assume worst case - that we're writing in
1476 * the middle of the extent.
1478 * We can assume that the write proceeds from
1479 * left to right, in which case the extent
1480 * insert code is smart enough to coalesce the
1481 * next splits into the previous records created.
1483 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1484 *extents_to_split
= *extents_to_split
+ 2;
1487 * Only increment phys if it doesn't describe
1493 desc
->c_phys
= phys
;
1496 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1498 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1499 desc
->c_unwritten
= 1;
1509 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1510 struct inode
*inode
,
1511 struct ocfs2_write_ctxt
*wc
)
1514 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1517 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1519 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1526 * If we don't set w_num_pages then this page won't get unlocked
1527 * and freed on cleanup of the write context.
1529 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1530 wc
->w_num_pages
= 1;
1532 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1533 if (IS_ERR(handle
)) {
1534 ret
= PTR_ERR(handle
);
1539 ret
= ocfs2_journal_access(handle
, inode
, wc
->w_di_bh
,
1540 OCFS2_JOURNAL_ACCESS_WRITE
);
1542 ocfs2_commit_trans(osb
, handle
);
1548 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1549 ocfs2_set_inode_data_inline(inode
, di
);
1551 if (!PageUptodate(page
)) {
1552 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1554 ocfs2_commit_trans(osb
, handle
);
1560 wc
->w_handle
= handle
;
1565 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1567 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1569 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1574 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1575 struct inode
*inode
, loff_t pos
,
1576 unsigned len
, struct page
*mmap_page
,
1577 struct ocfs2_write_ctxt
*wc
)
1579 int ret
, written
= 0;
1580 loff_t end
= pos
+ len
;
1581 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
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 if (mmap_page
|| end
> ocfs2_max_inline_data(inode
->i_sb
))
1618 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1625 * This signals to the caller that the data can be written
1630 return written
? written
: ret
;
1634 * This function only does anything for file systems which can't
1635 * handle sparse files.
1637 * What we want to do here is fill in any hole between the current end
1638 * of allocation and the end of our write. That way the rest of the
1639 * write path can treat it as an non-allocating write, which has no
1640 * special case code for sparse/nonsparse files.
1642 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
, loff_t pos
,
1644 struct ocfs2_write_ctxt
*wc
)
1647 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1648 loff_t newsize
= pos
+ len
;
1650 if (ocfs2_sparse_alloc(osb
))
1653 if (newsize
<= i_size_read(inode
))
1656 ret
= ocfs2_extend_no_holes(inode
, newsize
, newsize
- len
);
1663 int ocfs2_write_begin_nolock(struct address_space
*mapping
,
1664 loff_t pos
, unsigned len
, unsigned flags
,
1665 struct page
**pagep
, void **fsdata
,
1666 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1668 int ret
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1669 unsigned int clusters_to_alloc
, extents_to_split
;
1670 struct ocfs2_write_ctxt
*wc
;
1671 struct inode
*inode
= mapping
->host
;
1672 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1673 struct ocfs2_dinode
*di
;
1674 struct ocfs2_alloc_context
*data_ac
= NULL
;
1675 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1677 struct ocfs2_extent_tree et
;
1679 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1685 if (ocfs2_supports_inline_data(osb
)) {
1686 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1698 ret
= ocfs2_expand_nonsparse_inode(inode
, pos
, len
, wc
);
1704 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1711 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1714 * We set w_target_from, w_target_to here so that
1715 * ocfs2_write_end() knows which range in the target page to
1716 * write out. An allocation requires that we write the entire
1719 if (clusters_to_alloc
|| extents_to_split
) {
1721 * XXX: We are stretching the limits of
1722 * ocfs2_lock_allocators(). It greatly over-estimates
1723 * the work to be done.
1725 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1726 " clusters_to_add = %u, extents_to_split = %u\n",
1727 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1728 (long long)i_size_read(inode
), le32_to_cpu(di
->i_clusters
),
1729 clusters_to_alloc
, extents_to_split
);
1731 ocfs2_init_dinode_extent_tree(&et
, inode
, wc
->w_di_bh
);
1732 ret
= ocfs2_lock_allocators(inode
, &et
,
1733 clusters_to_alloc
, extents_to_split
,
1734 &data_ac
, &meta_ac
);
1740 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1746 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
,
1747 clusters_to_alloc
+ extents_to_split
);
1749 handle
= ocfs2_start_trans(osb
, credits
);
1750 if (IS_ERR(handle
)) {
1751 ret
= PTR_ERR(handle
);
1756 wc
->w_handle
= handle
;
1759 * We don't want this to fail in ocfs2_write_end(), so do it
1762 ret
= ocfs2_journal_access(handle
, inode
, wc
->w_di_bh
,
1763 OCFS2_JOURNAL_ACCESS_WRITE
);
1770 * Fill our page array first. That way we've grabbed enough so
1771 * that we can zero and flush if we error after adding the
1774 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
,
1775 clusters_to_alloc
+ extents_to_split
,
1782 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1790 ocfs2_free_alloc_context(data_ac
);
1792 ocfs2_free_alloc_context(meta_ac
);
1795 *pagep
= wc
->w_target_page
;
1799 ocfs2_commit_trans(osb
, handle
);
1802 ocfs2_free_write_ctxt(wc
);
1805 ocfs2_free_alloc_context(data_ac
);
1807 ocfs2_free_alloc_context(meta_ac
);
1811 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1812 loff_t pos
, unsigned len
, unsigned flags
,
1813 struct page
**pagep
, void **fsdata
)
1816 struct buffer_head
*di_bh
= NULL
;
1817 struct inode
*inode
= mapping
->host
;
1819 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1826 * Take alloc sem here to prevent concurrent lookups. That way
1827 * the mapping, zeroing and tree manipulation within
1828 * ocfs2_write() will be safe against ->readpage(). This
1829 * should also serve to lock out allocation from a shared
1832 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1834 ret
= ocfs2_write_begin_nolock(mapping
, pos
, len
, flags
, pagep
,
1835 fsdata
, di_bh
, NULL
);
1846 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1849 ocfs2_inode_unlock(inode
, 1);
1854 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1855 unsigned len
, unsigned *copied
,
1856 struct ocfs2_dinode
*di
,
1857 struct ocfs2_write_ctxt
*wc
)
1861 if (unlikely(*copied
< len
)) {
1862 if (!PageUptodate(wc
->w_target_page
)) {
1868 kaddr
= kmap_atomic(wc
->w_target_page
, KM_USER0
);
1869 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1870 kunmap_atomic(kaddr
, KM_USER0
);
1872 mlog(0, "Data written to inode at offset %llu. "
1873 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1874 (unsigned long long)pos
, *copied
,
1875 le16_to_cpu(di
->id2
.i_data
.id_count
),
1876 le16_to_cpu(di
->i_dyn_features
));
1879 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1880 loff_t pos
, unsigned len
, unsigned copied
,
1881 struct page
*page
, void *fsdata
)
1884 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1885 struct inode
*inode
= mapping
->host
;
1886 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1887 struct ocfs2_write_ctxt
*wc
= fsdata
;
1888 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1889 handle_t
*handle
= wc
->w_handle
;
1890 struct page
*tmppage
;
1892 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1893 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1894 goto out_write_size
;
1897 if (unlikely(copied
< len
)) {
1898 if (!PageUptodate(wc
->w_target_page
))
1901 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
1904 flush_dcache_page(wc
->w_target_page
);
1906 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1907 tmppage
= wc
->w_pages
[i
];
1909 if (tmppage
== wc
->w_target_page
) {
1910 from
= wc
->w_target_from
;
1911 to
= wc
->w_target_to
;
1913 BUG_ON(from
> PAGE_CACHE_SIZE
||
1914 to
> PAGE_CACHE_SIZE
||
1918 * Pages adjacent to the target (if any) imply
1919 * a hole-filling write in which case we want
1920 * to flush their entire range.
1923 to
= PAGE_CACHE_SIZE
;
1926 if (page_has_buffers(tmppage
)) {
1927 if (ocfs2_should_order_data(inode
)) {
1928 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1929 #ifdef CONFIG_OCFS2_COMPAT_JBD
1930 walk_page_buffers(wc
->w_handle
,
1931 page_buffers(tmppage
),
1933 ocfs2_journal_dirty_data
);
1936 block_commit_write(tmppage
, from
, to
);
1942 if (pos
> inode
->i_size
) {
1943 i_size_write(inode
, pos
);
1944 mark_inode_dirty(inode
);
1946 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1947 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
1948 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1949 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1950 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1951 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
1953 ocfs2_commit_trans(osb
, handle
);
1955 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
1957 ocfs2_free_write_ctxt(wc
);
1962 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
1963 loff_t pos
, unsigned len
, unsigned copied
,
1964 struct page
*page
, void *fsdata
)
1967 struct inode
*inode
= mapping
->host
;
1969 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
1971 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1972 ocfs2_inode_unlock(inode
, 1);
1977 const struct address_space_operations ocfs2_aops
= {
1978 .readpage
= ocfs2_readpage
,
1979 .readpages
= ocfs2_readpages
,
1980 .writepage
= ocfs2_writepage
,
1981 .write_begin
= ocfs2_write_begin
,
1982 .write_end
= ocfs2_write_end
,
1984 .sync_page
= block_sync_page
,
1985 .direct_IO
= ocfs2_direct_IO
,
1986 .invalidatepage
= ocfs2_invalidatepage
,
1987 .releasepage
= ocfs2_releasepage
,
1988 .migratepage
= buffer_migrate_page
,