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"
48 #include "buffer_head_io.h"
50 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
51 struct buffer_head
*bh_result
, int create
)
55 struct ocfs2_dinode
*fe
= NULL
;
56 struct buffer_head
*bh
= NULL
;
57 struct buffer_head
*buffer_cache_bh
= NULL
;
58 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
61 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
62 (unsigned long long)iblock
, bh_result
, create
);
64 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
66 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
67 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
68 (unsigned long long)iblock
);
72 status
= ocfs2_read_inode_block(inode
, &bh
);
77 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
79 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
80 le32_to_cpu(fe
->i_clusters
))) {
81 mlog(ML_ERROR
, "block offset is outside the allocated size: "
82 "%llu\n", (unsigned long long)iblock
);
86 /* We don't use the page cache to create symlink data, so if
87 * need be, copy it over from the buffer cache. */
88 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
89 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
91 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
92 if (!buffer_cache_bh
) {
93 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
97 /* we haven't locked out transactions, so a commit
98 * could've happened. Since we've got a reference on
99 * the bh, even if it commits while we're doing the
100 * copy, the data is still good. */
101 if (buffer_jbd(buffer_cache_bh
)
102 && ocfs2_inode_is_new(inode
)) {
103 kaddr
= kmap_atomic(bh_result
->b_page
, KM_USER0
);
105 mlog(ML_ERROR
, "couldn't kmap!\n");
108 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
109 buffer_cache_bh
->b_data
,
111 kunmap_atomic(kaddr
, KM_USER0
);
112 set_buffer_uptodate(bh_result
);
114 brelse(buffer_cache_bh
);
117 map_bh(bh_result
, inode
->i_sb
,
118 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
129 static int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
130 struct buffer_head
*bh_result
, int create
)
133 unsigned int ext_flags
;
134 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
135 u64 p_blkno
, count
, past_eof
;
136 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
138 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode
,
139 (unsigned long long)iblock
, bh_result
, create
);
141 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
142 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
143 inode
, inode
->i_ino
);
145 if (S_ISLNK(inode
->i_mode
)) {
146 /* this always does I/O for some reason. */
147 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
151 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
154 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
155 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
156 (unsigned long long)p_blkno
);
160 if (max_blocks
< count
)
164 * ocfs2 never allocates in this function - the only time we
165 * need to use BH_New is when we're extending i_size on a file
166 * system which doesn't support holes, in which case BH_New
167 * allows block_prepare_write() to zero.
169 * If we see this on a sparse file system, then a truncate has
170 * raced us and removed the cluster. In this case, we clear
171 * the buffers dirty and uptodate bits and let the buffer code
172 * ignore it as a hole.
174 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
175 clear_buffer_dirty(bh_result
);
176 clear_buffer_uptodate(bh_result
);
180 /* Treat the unwritten extent as a hole for zeroing purposes. */
181 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
182 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
184 bh_result
->b_size
= count
<< inode
->i_blkbits
;
186 if (!ocfs2_sparse_alloc(osb
)) {
190 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
191 (unsigned long long)iblock
,
192 (unsigned long long)p_blkno
,
193 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
194 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
198 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
199 mlog(0, "Inode %lu, past_eof = %llu\n", inode
->i_ino
,
200 (unsigned long long)past_eof
);
202 if (create
&& (iblock
>= past_eof
))
203 set_buffer_new(bh_result
);
214 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
215 struct buffer_head
*di_bh
)
219 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
221 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
222 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
223 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
227 size
= i_size_read(inode
);
229 if (size
> PAGE_CACHE_SIZE
||
230 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
231 ocfs2_error(inode
->i_sb
,
232 "Inode %llu has with inline data has bad size: %Lu",
233 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
234 (unsigned long long)size
);
238 kaddr
= kmap_atomic(page
, KM_USER0
);
240 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
241 /* Clear the remaining part of the page */
242 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
243 flush_dcache_page(page
);
244 kunmap_atomic(kaddr
, KM_USER0
);
246 SetPageUptodate(page
);
251 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
254 struct buffer_head
*di_bh
= NULL
;
256 BUG_ON(!PageLocked(page
));
257 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
259 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
265 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
273 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
275 struct inode
*inode
= page
->mapping
->host
;
276 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
277 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
280 mlog_entry("(0x%p, %lu)\n", file
, (page
? page
->index
: 0));
282 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
284 if (ret
== AOP_TRUNCATED_PAGE
)
290 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
291 ret
= AOP_TRUNCATED_PAGE
;
292 goto out_inode_unlock
;
296 * i_size might have just been updated as we grabed the meta lock. We
297 * might now be discovering a truncate that hit on another node.
298 * block_read_full_page->get_block freaks out if it is asked to read
299 * beyond the end of a file, so we check here. Callers
300 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
301 * and notice that the page they just read isn't needed.
303 * XXX sys_readahead() seems to get that wrong?
305 if (start
>= i_size_read(inode
)) {
306 zero_user(page
, 0, PAGE_SIZE
);
307 SetPageUptodate(page
);
312 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
313 ret
= ocfs2_readpage_inline(inode
, page
);
315 ret
= block_read_full_page(page
, ocfs2_get_block
);
319 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
321 ocfs2_inode_unlock(inode
, 0);
330 * This is used only for read-ahead. Failures or difficult to handle
331 * situations are safe to ignore.
333 * Right now, we don't bother with BH_Boundary - in-inode extent lists
334 * are quite large (243 extents on 4k blocks), so most inodes don't
335 * grow out to a tree. If need be, detecting boundary extents could
336 * trivially be added in a future version of ocfs2_get_block().
338 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
339 struct list_head
*pages
, unsigned nr_pages
)
342 struct inode
*inode
= mapping
->host
;
343 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
348 * Use the nonblocking flag for the dlm code to avoid page
349 * lock inversion, but don't bother with retrying.
351 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
355 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
356 ocfs2_inode_unlock(inode
, 0);
361 * Don't bother with inline-data. There isn't anything
362 * to read-ahead in that case anyway...
364 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
368 * Check whether a remote node truncated this file - we just
369 * drop out in that case as it's not worth handling here.
371 last
= list_entry(pages
->prev
, struct page
, lru
);
372 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
373 if (start
>= i_size_read(inode
))
376 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
379 up_read(&oi
->ip_alloc_sem
);
380 ocfs2_inode_unlock(inode
, 0);
385 /* Note: Because we don't support holes, our allocation has
386 * already happened (allocation writes zeros to the file data)
387 * so we don't have to worry about ordered writes in
390 * ->writepage is called during the process of invalidating the page cache
391 * during blocked lock processing. It can't block on any cluster locks
392 * to during block mapping. It's relying on the fact that the block
393 * mapping can't have disappeared under the dirty pages that it is
394 * being asked to write back.
396 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
400 mlog_entry("(0x%p)\n", page
);
402 ret
= block_write_full_page(page
, ocfs2_get_block
, wbc
);
410 * This is called from ocfs2_write_zero_page() which has handled it's
411 * own cluster locking and has ensured allocation exists for those
412 * blocks to be written.
414 int ocfs2_prepare_write_nolock(struct inode
*inode
, struct page
*page
,
415 unsigned from
, unsigned to
)
419 ret
= block_prepare_write(page
, from
, to
, ocfs2_get_block
);
424 /* Taken from ext3. We don't necessarily need the full blown
425 * functionality yet, but IMHO it's better to cut and paste the whole
426 * thing so we can avoid introducing our own bugs (and easily pick up
427 * their fixes when they happen) --Mark */
428 int walk_page_buffers( handle_t
*handle
,
429 struct buffer_head
*head
,
433 int (*fn
)( handle_t
*handle
,
434 struct buffer_head
*bh
))
436 struct buffer_head
*bh
;
437 unsigned block_start
, block_end
;
438 unsigned blocksize
= head
->b_size
;
440 struct buffer_head
*next
;
442 for ( bh
= head
, block_start
= 0;
443 ret
== 0 && (bh
!= head
|| !block_start
);
444 block_start
= block_end
, bh
= next
)
446 next
= bh
->b_this_page
;
447 block_end
= block_start
+ blocksize
;
448 if (block_end
<= from
|| block_start
>= to
) {
449 if (partial
&& !buffer_uptodate(bh
))
453 err
= (*fn
)(handle
, bh
);
460 handle_t
*ocfs2_start_walk_page_trans(struct inode
*inode
,
465 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
469 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
470 if (IS_ERR(handle
)) {
476 if (ocfs2_should_order_data(inode
)) {
477 ret
= ocfs2_jbd2_file_inode(handle
, inode
);
484 ocfs2_commit_trans(osb
, handle
);
485 handle
= ERR_PTR(ret
);
490 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
495 struct inode
*inode
= mapping
->host
;
497 mlog_entry("(block = %llu)\n", (unsigned long long)block
);
499 /* We don't need to lock journal system files, since they aren't
500 * accessed concurrently from multiple nodes.
502 if (!INODE_JOURNAL(inode
)) {
503 err
= ocfs2_inode_lock(inode
, NULL
, 0);
509 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
512 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
513 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
516 if (!INODE_JOURNAL(inode
)) {
517 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
518 ocfs2_inode_unlock(inode
, 0);
522 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
523 (unsigned long long)block
);
529 status
= err
? 0 : p_blkno
;
531 mlog_exit((int)status
);
537 * TODO: Make this into a generic get_blocks function.
539 * From do_direct_io in direct-io.c:
540 * "So what we do is to permit the ->get_blocks function to populate
541 * bh.b_size with the size of IO which is permitted at this offset and
544 * This function is called directly from get_more_blocks in direct-io.c.
546 * called like this: dio->get_blocks(dio->inode, fs_startblk,
547 * fs_count, map_bh, dio->rw == WRITE);
549 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
550 struct buffer_head
*bh_result
, int create
)
553 u64 p_blkno
, inode_blocks
, contig_blocks
;
554 unsigned int ext_flags
;
555 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
556 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
558 /* This function won't even be called if the request isn't all
559 * nicely aligned and of the right size, so there's no need
560 * for us to check any of that. */
562 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
565 * Any write past EOF is not allowed because we'd be extending.
567 if (create
&& (iblock
+ max_blocks
) > inode_blocks
) {
572 /* This figures out the size of the next contiguous block, and
573 * our logical offset */
574 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
575 &contig_blocks
, &ext_flags
);
577 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
578 (unsigned long long)iblock
);
583 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)) && !p_blkno
&& create
) {
584 ocfs2_error(inode
->i_sb
,
585 "Inode %llu has a hole at block %llu\n",
586 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
587 (unsigned long long)iblock
);
593 * get_more_blocks() expects us to describe a hole by clearing
594 * the mapped bit on bh_result().
596 * Consider an unwritten extent as a hole.
598 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
599 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
602 * ocfs2_prepare_inode_for_write() should have caught
603 * the case where we'd be filling a hole and triggered
604 * a buffered write instead.
612 clear_buffer_mapped(bh_result
);
615 /* make sure we don't map more than max_blocks blocks here as
616 that's all the kernel will handle at this point. */
617 if (max_blocks
< contig_blocks
)
618 contig_blocks
= max_blocks
;
619 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
625 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
626 * particularly interested in the aio/dio case. Like the core uses
627 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
628 * truncation on another.
630 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
635 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
638 /* this io's submitter should not have unlocked this before we could */
639 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
641 ocfs2_iocb_clear_rw_locked(iocb
);
643 level
= ocfs2_iocb_rw_locked_level(iocb
);
645 up_read(&inode
->i_alloc_sem
);
646 ocfs2_rw_unlock(inode
, level
);
650 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
651 * from ext3. PageChecked() bits have been removed as OCFS2 does not
652 * do journalled data.
654 static void ocfs2_invalidatepage(struct page
*page
, unsigned long offset
)
656 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
658 jbd2_journal_invalidatepage(journal
, page
, offset
);
661 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
663 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
665 if (!page_has_buffers(page
))
667 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
670 static ssize_t
ocfs2_direct_IO(int rw
,
672 const struct iovec
*iov
,
674 unsigned long nr_segs
)
676 struct file
*file
= iocb
->ki_filp
;
677 struct inode
*inode
= file
->f_path
.dentry
->d_inode
->i_mapping
->host
;
683 * Fallback to buffered I/O if we see an inode without
686 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
689 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
690 inode
->i_sb
->s_bdev
, iov
, offset
,
692 ocfs2_direct_IO_get_blocks
,
699 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
704 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
706 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
709 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
711 cluster_start
= cpos
% cpp
;
712 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
714 cluster_end
= cluster_start
+ osb
->s_clustersize
;
717 BUG_ON(cluster_start
> PAGE_SIZE
);
718 BUG_ON(cluster_end
> PAGE_SIZE
);
721 *start
= cluster_start
;
727 * 'from' and 'to' are the region in the page to avoid zeroing.
729 * If pagesize > clustersize, this function will avoid zeroing outside
730 * of the cluster boundary.
732 * from == to == 0 is code for "zero the entire cluster region"
734 static void ocfs2_clear_page_regions(struct page
*page
,
735 struct ocfs2_super
*osb
, u32 cpos
,
736 unsigned from
, unsigned to
)
739 unsigned int cluster_start
, cluster_end
;
741 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
743 kaddr
= kmap_atomic(page
, KM_USER0
);
746 if (from
> cluster_start
)
747 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
748 if (to
< cluster_end
)
749 memset(kaddr
+ to
, 0, cluster_end
- to
);
751 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
754 kunmap_atomic(kaddr
, KM_USER0
);
758 * Nonsparse file systems fully allocate before we get to the write
759 * code. This prevents ocfs2_write() from tagging the write as an
760 * allocating one, which means ocfs2_map_page_blocks() might try to
761 * read-in the blocks at the tail of our file. Avoid reading them by
762 * testing i_size against each block offset.
764 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
765 unsigned int block_start
)
767 u64 offset
= page_offset(page
) + block_start
;
769 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
772 if (i_size_read(inode
) > offset
)
779 * Some of this taken from block_prepare_write(). We already have our
780 * mapping by now though, and the entire write will be allocating or
781 * it won't, so not much need to use BH_New.
783 * This will also skip zeroing, which is handled externally.
785 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
786 struct inode
*inode
, unsigned int from
,
787 unsigned int to
, int new)
790 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
791 unsigned int block_end
, block_start
;
792 unsigned int bsize
= 1 << inode
->i_blkbits
;
794 if (!page_has_buffers(page
))
795 create_empty_buffers(page
, bsize
, 0);
797 head
= page_buffers(page
);
798 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
799 bh
= bh
->b_this_page
, block_start
+= bsize
) {
800 block_end
= block_start
+ bsize
;
802 clear_buffer_new(bh
);
805 * Ignore blocks outside of our i/o range -
806 * they may belong to unallocated clusters.
808 if (block_start
>= to
|| block_end
<= from
) {
809 if (PageUptodate(page
))
810 set_buffer_uptodate(bh
);
815 * For an allocating write with cluster size >= page
816 * size, we always write the entire page.
821 if (!buffer_mapped(bh
)) {
822 map_bh(bh
, inode
->i_sb
, *p_blkno
);
823 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
826 if (PageUptodate(page
)) {
827 if (!buffer_uptodate(bh
))
828 set_buffer_uptodate(bh
);
829 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
831 ocfs2_should_read_blk(inode
, page
, block_start
) &&
832 (block_start
< from
|| block_end
> to
)) {
833 ll_rw_block(READ
, 1, &bh
);
837 *p_blkno
= *p_blkno
+ 1;
841 * If we issued read requests - let them complete.
843 while(wait_bh
> wait
) {
844 wait_on_buffer(*--wait_bh
);
845 if (!buffer_uptodate(*wait_bh
))
849 if (ret
== 0 || !new)
853 * If we get -EIO above, zero out any newly allocated blocks
854 * to avoid exposing stale data.
859 block_end
= block_start
+ bsize
;
860 if (block_end
<= from
)
862 if (block_start
>= to
)
865 zero_user(page
, block_start
, bh
->b_size
);
866 set_buffer_uptodate(bh
);
867 mark_buffer_dirty(bh
);
870 block_start
= block_end
;
871 bh
= bh
->b_this_page
;
872 } while (bh
!= head
);
877 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
878 #define OCFS2_MAX_CTXT_PAGES 1
880 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
883 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
886 * Describe the state of a single cluster to be written to.
888 struct ocfs2_write_cluster_desc
{
892 * Give this a unique field because c_phys eventually gets
896 unsigned c_unwritten
;
897 unsigned c_needs_zero
;
900 struct ocfs2_write_ctxt
{
901 /* Logical cluster position / len of write */
905 /* First cluster allocated in a nonsparse extend */
906 u32 w_first_new_cpos
;
908 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
911 * This is true if page_size > cluster_size.
913 * It triggers a set of special cases during write which might
914 * have to deal with allocating writes to partial pages.
916 unsigned int w_large_pages
;
919 * Pages involved in this write.
921 * w_target_page is the page being written to by the user.
923 * w_pages is an array of pages which always contains
924 * w_target_page, and in the case of an allocating write with
925 * page_size < cluster size, it will contain zero'd and mapped
926 * pages adjacent to w_target_page which need to be written
927 * out in so that future reads from that region will get
930 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
931 unsigned int w_num_pages
;
932 struct page
*w_target_page
;
935 * ocfs2_write_end() uses this to know what the real range to
936 * write in the target should be.
938 unsigned int w_target_from
;
939 unsigned int w_target_to
;
942 * We could use journal_current_handle() but this is cleaner,
947 struct buffer_head
*w_di_bh
;
949 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
952 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
956 for(i
= 0; i
< num_pages
; i
++) {
958 unlock_page(pages
[i
]);
959 mark_page_accessed(pages
[i
]);
960 page_cache_release(pages
[i
]);
965 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
967 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
973 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
974 struct ocfs2_super
*osb
, loff_t pos
,
975 unsigned len
, struct buffer_head
*di_bh
)
978 struct ocfs2_write_ctxt
*wc
;
980 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
984 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
985 wc
->w_first_new_cpos
= UINT_MAX
;
986 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
987 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
991 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
992 wc
->w_large_pages
= 1;
994 wc
->w_large_pages
= 0;
996 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
1004 * If a page has any new buffers, zero them out here, and mark them uptodate
1005 * and dirty so they'll be written out (in order to prevent uninitialised
1006 * block data from leaking). And clear the new bit.
1008 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1010 unsigned int block_start
, block_end
;
1011 struct buffer_head
*head
, *bh
;
1013 BUG_ON(!PageLocked(page
));
1014 if (!page_has_buffers(page
))
1017 bh
= head
= page_buffers(page
);
1020 block_end
= block_start
+ bh
->b_size
;
1022 if (buffer_new(bh
)) {
1023 if (block_end
> from
&& block_start
< to
) {
1024 if (!PageUptodate(page
)) {
1025 unsigned start
, end
;
1027 start
= max(from
, block_start
);
1028 end
= min(to
, block_end
);
1030 zero_user_segment(page
, start
, end
);
1031 set_buffer_uptodate(bh
);
1034 clear_buffer_new(bh
);
1035 mark_buffer_dirty(bh
);
1039 block_start
= block_end
;
1040 bh
= bh
->b_this_page
;
1041 } while (bh
!= head
);
1045 * Only called when we have a failure during allocating write to write
1046 * zero's to the newly allocated region.
1048 static void ocfs2_write_failure(struct inode
*inode
,
1049 struct ocfs2_write_ctxt
*wc
,
1050 loff_t user_pos
, unsigned user_len
)
1053 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1054 to
= user_pos
+ user_len
;
1055 struct page
*tmppage
;
1057 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1059 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1060 tmppage
= wc
->w_pages
[i
];
1062 if (page_has_buffers(tmppage
)) {
1063 if (ocfs2_should_order_data(inode
))
1064 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1066 block_commit_write(tmppage
, from
, to
);
1071 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1072 struct ocfs2_write_ctxt
*wc
,
1073 struct page
*page
, u32 cpos
,
1074 loff_t user_pos
, unsigned user_len
,
1078 unsigned int map_from
= 0, map_to
= 0;
1079 unsigned int cluster_start
, cluster_end
;
1080 unsigned int user_data_from
= 0, user_data_to
= 0;
1082 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1083 &cluster_start
, &cluster_end
);
1085 if (page
== wc
->w_target_page
) {
1086 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1087 map_to
= map_from
+ user_len
;
1090 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1091 cluster_start
, cluster_end
,
1094 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1095 map_from
, map_to
, new);
1101 user_data_from
= map_from
;
1102 user_data_to
= map_to
;
1104 map_from
= cluster_start
;
1105 map_to
= cluster_end
;
1109 * If we haven't allocated the new page yet, we
1110 * shouldn't be writing it out without copying user
1111 * data. This is likely a math error from the caller.
1115 map_from
= cluster_start
;
1116 map_to
= cluster_end
;
1118 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1119 cluster_start
, cluster_end
, new);
1127 * Parts of newly allocated pages need to be zero'd.
1129 * Above, we have also rewritten 'to' and 'from' - as far as
1130 * the rest of the function is concerned, the entire cluster
1131 * range inside of a page needs to be written.
1133 * We can skip this if the page is up to date - it's already
1134 * been zero'd from being read in as a hole.
1136 if (new && !PageUptodate(page
))
1137 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1138 cpos
, user_data_from
, user_data_to
);
1140 flush_dcache_page(page
);
1147 * This function will only grab one clusters worth of pages.
1149 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1150 struct ocfs2_write_ctxt
*wc
,
1151 u32 cpos
, loff_t user_pos
, int new,
1152 struct page
*mmap_page
)
1155 unsigned long start
, target_index
, index
;
1156 struct inode
*inode
= mapping
->host
;
1158 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1161 * Figure out how many pages we'll be manipulating here. For
1162 * non allocating write, we just change the one
1163 * page. Otherwise, we'll need a whole clusters worth.
1166 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1167 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1169 wc
->w_num_pages
= 1;
1170 start
= target_index
;
1173 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1176 if (index
== target_index
&& mmap_page
) {
1178 * ocfs2_pagemkwrite() is a little different
1179 * and wants us to directly use the page
1182 lock_page(mmap_page
);
1184 if (mmap_page
->mapping
!= mapping
) {
1185 unlock_page(mmap_page
);
1187 * Sanity check - the locking in
1188 * ocfs2_pagemkwrite() should ensure
1189 * that this code doesn't trigger.
1196 page_cache_get(mmap_page
);
1197 wc
->w_pages
[i
] = mmap_page
;
1199 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1201 if (!wc
->w_pages
[i
]) {
1208 if (index
== target_index
)
1209 wc
->w_target_page
= wc
->w_pages
[i
];
1216 * Prepare a single cluster for write one cluster into the file.
1218 static int ocfs2_write_cluster(struct address_space
*mapping
,
1219 u32 phys
, unsigned int unwritten
,
1220 unsigned int should_zero
,
1221 struct ocfs2_alloc_context
*data_ac
,
1222 struct ocfs2_alloc_context
*meta_ac
,
1223 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1224 loff_t user_pos
, unsigned user_len
)
1227 u64 v_blkno
, p_blkno
;
1228 struct inode
*inode
= mapping
->host
;
1229 struct ocfs2_extent_tree et
;
1231 new = phys
== 0 ? 1 : 0;
1236 * This is safe to call with the page locks - it won't take
1237 * any additional semaphores or cluster locks.
1240 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1241 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1242 wc
->w_handle
, data_ac
,
1245 * This shouldn't happen because we must have already
1246 * calculated the correct meta data allocation required. The
1247 * internal tree allocation code should know how to increase
1248 * transaction credits itself.
1250 * If need be, we could handle -EAGAIN for a
1251 * RESTART_TRANS here.
1253 mlog_bug_on_msg(ret
== -EAGAIN
,
1254 "Inode %llu: EAGAIN return during allocation.\n",
1255 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1260 } else if (unwritten
) {
1261 ocfs2_init_dinode_extent_tree(&et
, inode
, wc
->w_di_bh
);
1262 ret
= ocfs2_mark_extent_written(inode
, &et
,
1263 wc
->w_handle
, cpos
, 1, phys
,
1264 meta_ac
, &wc
->w_dealloc
);
1272 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1274 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1277 * The only reason this should fail is due to an inability to
1278 * find the extent added.
1280 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1283 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1284 "at logical block %llu",
1285 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1286 (unsigned long long)v_blkno
);
1290 BUG_ON(p_blkno
== 0);
1292 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1295 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1296 wc
->w_pages
[i
], cpos
,
1307 * We only have cleanup to do in case of allocating write.
1310 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1317 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1318 struct ocfs2_alloc_context
*data_ac
,
1319 struct ocfs2_alloc_context
*meta_ac
,
1320 struct ocfs2_write_ctxt
*wc
,
1321 loff_t pos
, unsigned len
)
1325 unsigned int local_len
= len
;
1326 struct ocfs2_write_cluster_desc
*desc
;
1327 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1329 for (i
= 0; i
< wc
->w_clen
; i
++) {
1330 desc
= &wc
->w_desc
[i
];
1333 * We have to make sure that the total write passed in
1334 * doesn't extend past a single cluster.
1337 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1338 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1339 local_len
= osb
->s_clustersize
- cluster_off
;
1341 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1345 wc
, desc
->c_cpos
, pos
, local_len
);
1361 * ocfs2_write_end() wants to know which parts of the target page it
1362 * should complete the write on. It's easiest to compute them ahead of
1363 * time when a more complete view of the write is available.
1365 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1366 struct ocfs2_write_ctxt
*wc
,
1367 loff_t pos
, unsigned len
, int alloc
)
1369 struct ocfs2_write_cluster_desc
*desc
;
1371 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1372 wc
->w_target_to
= wc
->w_target_from
+ len
;
1378 * Allocating write - we may have different boundaries based
1379 * on page size and cluster size.
1381 * NOTE: We can no longer compute one value from the other as
1382 * the actual write length and user provided length may be
1386 if (wc
->w_large_pages
) {
1388 * We only care about the 1st and last cluster within
1389 * our range and whether they should be zero'd or not. Either
1390 * value may be extended out to the start/end of a
1391 * newly allocated cluster.
1393 desc
= &wc
->w_desc
[0];
1394 if (desc
->c_needs_zero
)
1395 ocfs2_figure_cluster_boundaries(osb
,
1400 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1401 if (desc
->c_needs_zero
)
1402 ocfs2_figure_cluster_boundaries(osb
,
1407 wc
->w_target_from
= 0;
1408 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1413 * Populate each single-cluster write descriptor in the write context
1414 * with information about the i/o to be done.
1416 * Returns the number of clusters that will have to be allocated, as
1417 * well as a worst case estimate of the number of extent records that
1418 * would have to be created during a write to an unwritten region.
1420 static int ocfs2_populate_write_desc(struct inode
*inode
,
1421 struct ocfs2_write_ctxt
*wc
,
1422 unsigned int *clusters_to_alloc
,
1423 unsigned int *extents_to_split
)
1426 struct ocfs2_write_cluster_desc
*desc
;
1427 unsigned int num_clusters
= 0;
1428 unsigned int ext_flags
= 0;
1432 *clusters_to_alloc
= 0;
1433 *extents_to_split
= 0;
1435 for (i
= 0; i
< wc
->w_clen
; i
++) {
1436 desc
= &wc
->w_desc
[i
];
1437 desc
->c_cpos
= wc
->w_cpos
+ i
;
1439 if (num_clusters
== 0) {
1441 * Need to look up the next extent record.
1443 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1444 &num_clusters
, &ext_flags
);
1451 * Assume worst case - that we're writing in
1452 * the middle of the extent.
1454 * We can assume that the write proceeds from
1455 * left to right, in which case the extent
1456 * insert code is smart enough to coalesce the
1457 * next splits into the previous records created.
1459 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1460 *extents_to_split
= *extents_to_split
+ 2;
1463 * Only increment phys if it doesn't describe
1470 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1471 * file that got extended. w_first_new_cpos tells us
1472 * where the newly allocated clusters are so we can
1475 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1477 desc
->c_needs_zero
= 1;
1480 desc
->c_phys
= phys
;
1483 desc
->c_needs_zero
= 1;
1484 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1487 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1488 desc
->c_unwritten
= 1;
1489 desc
->c_needs_zero
= 1;
1500 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1501 struct inode
*inode
,
1502 struct ocfs2_write_ctxt
*wc
)
1505 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1508 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1510 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1517 * If we don't set w_num_pages then this page won't get unlocked
1518 * and freed on cleanup of the write context.
1520 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1521 wc
->w_num_pages
= 1;
1523 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1524 if (IS_ERR(handle
)) {
1525 ret
= PTR_ERR(handle
);
1530 ret
= ocfs2_journal_access_di(handle
, inode
, wc
->w_di_bh
,
1531 OCFS2_JOURNAL_ACCESS_WRITE
);
1533 ocfs2_commit_trans(osb
, handle
);
1539 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1540 ocfs2_set_inode_data_inline(inode
, di
);
1542 if (!PageUptodate(page
)) {
1543 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1545 ocfs2_commit_trans(osb
, handle
);
1551 wc
->w_handle
= handle
;
1556 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1558 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1560 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1565 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1566 struct inode
*inode
, loff_t pos
,
1567 unsigned len
, struct page
*mmap_page
,
1568 struct ocfs2_write_ctxt
*wc
)
1570 int ret
, written
= 0;
1571 loff_t end
= pos
+ len
;
1572 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1573 struct ocfs2_dinode
*di
= NULL
;
1575 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1576 (unsigned long long)oi
->ip_blkno
, len
, (unsigned long long)pos
,
1577 oi
->ip_dyn_features
);
1580 * Handle inodes which already have inline data 1st.
1582 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1583 if (mmap_page
== NULL
&&
1584 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1585 goto do_inline_write
;
1588 * The write won't fit - we have to give this inode an
1589 * inline extent list now.
1591 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1598 * Check whether the inode can accept inline data.
1600 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1604 * Check whether the write can fit.
1606 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1608 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1612 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1619 * This signals to the caller that the data can be written
1624 return written
? written
: ret
;
1628 * This function only does anything for file systems which can't
1629 * handle sparse files.
1631 * What we want to do here is fill in any hole between the current end
1632 * of allocation and the end of our write. That way the rest of the
1633 * write path can treat it as an non-allocating write, which has no
1634 * special case code for sparse/nonsparse files.
1636 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
, loff_t pos
,
1638 struct ocfs2_write_ctxt
*wc
)
1641 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1642 loff_t newsize
= pos
+ len
;
1644 if (ocfs2_sparse_alloc(osb
))
1647 if (newsize
<= i_size_read(inode
))
1650 ret
= ocfs2_extend_no_holes(inode
, newsize
, pos
);
1654 wc
->w_first_new_cpos
=
1655 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1660 int ocfs2_write_begin_nolock(struct address_space
*mapping
,
1661 loff_t pos
, unsigned len
, unsigned flags
,
1662 struct page
**pagep
, void **fsdata
,
1663 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1665 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1666 unsigned int clusters_to_alloc
, extents_to_split
;
1667 struct ocfs2_write_ctxt
*wc
;
1668 struct inode
*inode
= mapping
->host
;
1669 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1670 struct ocfs2_dinode
*di
;
1671 struct ocfs2_alloc_context
*data_ac
= NULL
;
1672 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1674 struct ocfs2_extent_tree et
;
1676 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1682 if (ocfs2_supports_inline_data(osb
)) {
1683 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1695 ret
= ocfs2_expand_nonsparse_inode(inode
, pos
, len
, wc
);
1701 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1708 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1711 * We set w_target_from, w_target_to here so that
1712 * ocfs2_write_end() knows which range in the target page to
1713 * write out. An allocation requires that we write the entire
1716 if (clusters_to_alloc
|| extents_to_split
) {
1718 * XXX: We are stretching the limits of
1719 * ocfs2_lock_allocators(). It greatly over-estimates
1720 * the work to be done.
1722 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1723 " clusters_to_add = %u, extents_to_split = %u\n",
1724 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1725 (long long)i_size_read(inode
), le32_to_cpu(di
->i_clusters
),
1726 clusters_to_alloc
, extents_to_split
);
1728 ocfs2_init_dinode_extent_tree(&et
, inode
, wc
->w_di_bh
);
1729 ret
= ocfs2_lock_allocators(inode
, &et
,
1730 clusters_to_alloc
, extents_to_split
,
1731 &data_ac
, &meta_ac
);
1737 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1744 * We have to zero sparse allocated clusters, unwritten extent clusters,
1745 * and non-sparse clusters we just extended. For non-sparse writes,
1746 * we know zeros will only be needed in the first and/or last cluster.
1748 if (clusters_to_alloc
|| extents_to_split
||
1749 wc
->w_desc
[0].c_needs_zero
||
1750 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)
1751 cluster_of_pages
= 1;
1753 cluster_of_pages
= 0;
1755 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1757 handle
= ocfs2_start_trans(osb
, credits
);
1758 if (IS_ERR(handle
)) {
1759 ret
= PTR_ERR(handle
);
1764 wc
->w_handle
= handle
;
1766 if (clusters_to_alloc
&& vfs_dq_alloc_space_nodirty(inode
,
1767 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
))) {
1772 * We don't want this to fail in ocfs2_write_end(), so do it
1775 ret
= ocfs2_journal_access_di(handle
, inode
, wc
->w_di_bh
,
1776 OCFS2_JOURNAL_ACCESS_WRITE
);
1783 * Fill our page array first. That way we've grabbed enough so
1784 * that we can zero and flush if we error after adding the
1787 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
,
1788 cluster_of_pages
, mmap_page
);
1794 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1802 ocfs2_free_alloc_context(data_ac
);
1804 ocfs2_free_alloc_context(meta_ac
);
1807 *pagep
= wc
->w_target_page
;
1811 if (clusters_to_alloc
)
1812 vfs_dq_free_space(inode
,
1813 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1815 ocfs2_commit_trans(osb
, handle
);
1818 ocfs2_free_write_ctxt(wc
);
1821 ocfs2_free_alloc_context(data_ac
);
1823 ocfs2_free_alloc_context(meta_ac
);
1827 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1828 loff_t pos
, unsigned len
, unsigned flags
,
1829 struct page
**pagep
, void **fsdata
)
1832 struct buffer_head
*di_bh
= NULL
;
1833 struct inode
*inode
= mapping
->host
;
1835 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1842 * Take alloc sem here to prevent concurrent lookups. That way
1843 * the mapping, zeroing and tree manipulation within
1844 * ocfs2_write() will be safe against ->readpage(). This
1845 * should also serve to lock out allocation from a shared
1848 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1850 ret
= ocfs2_write_begin_nolock(mapping
, pos
, len
, flags
, pagep
,
1851 fsdata
, di_bh
, NULL
);
1862 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1865 ocfs2_inode_unlock(inode
, 1);
1870 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1871 unsigned len
, unsigned *copied
,
1872 struct ocfs2_dinode
*di
,
1873 struct ocfs2_write_ctxt
*wc
)
1877 if (unlikely(*copied
< len
)) {
1878 if (!PageUptodate(wc
->w_target_page
)) {
1884 kaddr
= kmap_atomic(wc
->w_target_page
, KM_USER0
);
1885 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1886 kunmap_atomic(kaddr
, KM_USER0
);
1888 mlog(0, "Data written to inode at offset %llu. "
1889 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1890 (unsigned long long)pos
, *copied
,
1891 le16_to_cpu(di
->id2
.i_data
.id_count
),
1892 le16_to_cpu(di
->i_dyn_features
));
1895 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1896 loff_t pos
, unsigned len
, unsigned copied
,
1897 struct page
*page
, void *fsdata
)
1900 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1901 struct inode
*inode
= mapping
->host
;
1902 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1903 struct ocfs2_write_ctxt
*wc
= fsdata
;
1904 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1905 handle_t
*handle
= wc
->w_handle
;
1906 struct page
*tmppage
;
1908 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1909 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1910 goto out_write_size
;
1913 if (unlikely(copied
< len
)) {
1914 if (!PageUptodate(wc
->w_target_page
))
1917 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
1920 flush_dcache_page(wc
->w_target_page
);
1922 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1923 tmppage
= wc
->w_pages
[i
];
1925 if (tmppage
== wc
->w_target_page
) {
1926 from
= wc
->w_target_from
;
1927 to
= wc
->w_target_to
;
1929 BUG_ON(from
> PAGE_CACHE_SIZE
||
1930 to
> PAGE_CACHE_SIZE
||
1934 * Pages adjacent to the target (if any) imply
1935 * a hole-filling write in which case we want
1936 * to flush their entire range.
1939 to
= PAGE_CACHE_SIZE
;
1942 if (page_has_buffers(tmppage
)) {
1943 if (ocfs2_should_order_data(inode
))
1944 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1945 block_commit_write(tmppage
, from
, to
);
1951 if (pos
> inode
->i_size
) {
1952 i_size_write(inode
, pos
);
1953 mark_inode_dirty(inode
);
1955 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1956 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
1957 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1958 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1959 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1960 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
1962 ocfs2_commit_trans(osb
, handle
);
1964 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
1966 ocfs2_free_write_ctxt(wc
);
1971 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
1972 loff_t pos
, unsigned len
, unsigned copied
,
1973 struct page
*page
, void *fsdata
)
1976 struct inode
*inode
= mapping
->host
;
1978 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
1980 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1981 ocfs2_inode_unlock(inode
, 1);
1986 const struct address_space_operations ocfs2_aops
= {
1987 .readpage
= ocfs2_readpage
,
1988 .readpages
= ocfs2_readpages
,
1989 .writepage
= ocfs2_writepage
,
1990 .write_begin
= ocfs2_write_begin
,
1991 .write_end
= ocfs2_write_end
,
1993 .sync_page
= block_sync_page
,
1994 .direct_IO
= ocfs2_direct_IO
,
1995 .invalidatepage
= ocfs2_invalidatepage
,
1996 .releasepage
= ocfs2_releasepage
,
1997 .migratepage
= buffer_migrate_page
,
1998 .is_partially_uptodate
= block_is_partially_uptodate
,