1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
23 #include <linux/slab.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <asm/byteorder.h>
27 #include <linux/swap.h>
28 #include <linux/pipe_fs_i.h>
29 #include <linux/mpage.h>
30 #include <linux/quotaops.h>
32 #include <cluster/masklog.h>
39 #include "extent_map.h"
46 #include "refcounttree.h"
47 #include "ocfs2_trace.h"
49 #include "buffer_head_io.h"
51 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
52 struct buffer_head
*bh_result
, int create
)
56 struct ocfs2_dinode
*fe
= NULL
;
57 struct buffer_head
*bh
= NULL
;
58 struct buffer_head
*buffer_cache_bh
= NULL
;
59 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
62 trace_ocfs2_symlink_get_block(
63 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
64 (unsigned long long)iblock
, bh_result
, create
);
66 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
68 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
69 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
70 (unsigned long long)iblock
);
74 status
= ocfs2_read_inode_block(inode
, &bh
);
79 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
81 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
82 le32_to_cpu(fe
->i_clusters
))) {
83 mlog(ML_ERROR
, "block offset is outside the allocated size: "
84 "%llu\n", (unsigned long long)iblock
);
88 /* We don't use the page cache to create symlink data, so if
89 * need be, copy it over from the buffer cache. */
90 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
91 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
93 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
94 if (!buffer_cache_bh
) {
95 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
99 /* we haven't locked out transactions, so a commit
100 * could've happened. Since we've got a reference on
101 * the bh, even if it commits while we're doing the
102 * copy, the data is still good. */
103 if (buffer_jbd(buffer_cache_bh
)
104 && ocfs2_inode_is_new(inode
)) {
105 kaddr
= kmap_atomic(bh_result
->b_page
, KM_USER0
);
107 mlog(ML_ERROR
, "couldn't kmap!\n");
110 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
111 buffer_cache_bh
->b_data
,
113 kunmap_atomic(kaddr
, KM_USER0
);
114 set_buffer_uptodate(bh_result
);
116 brelse(buffer_cache_bh
);
119 map_bh(bh_result
, inode
->i_sb
,
120 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 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
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_write_begin() 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
);
201 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
203 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
204 (unsigned long long)past_eof
);
205 if (create
&& (iblock
>= past_eof
))
206 set_buffer_new(bh_result
);
215 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
216 struct buffer_head
*di_bh
)
220 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
222 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
223 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
224 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
228 size
= i_size_read(inode
);
230 if (size
> PAGE_CACHE_SIZE
||
231 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
232 ocfs2_error(inode
->i_sb
,
233 "Inode %llu has with inline data has bad size: %Lu",
234 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
235 (unsigned long long)size
);
239 kaddr
= kmap_atomic(page
, KM_USER0
);
241 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
242 /* Clear the remaining part of the page */
243 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
244 flush_dcache_page(page
);
245 kunmap_atomic(kaddr
, KM_USER0
);
247 SetPageUptodate(page
);
252 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
255 struct buffer_head
*di_bh
= NULL
;
257 BUG_ON(!PageLocked(page
));
258 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
260 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
266 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
274 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
276 struct inode
*inode
= page
->mapping
->host
;
277 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
278 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
281 trace_ocfs2_readpage((unsigned long long)oi
->ip_blkno
,
282 (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);
331 * This is used only for read-ahead. Failures or difficult to handle
332 * situations are safe to ignore.
334 * Right now, we don't bother with BH_Boundary - in-inode extent lists
335 * are quite large (243 extents on 4k blocks), so most inodes don't
336 * grow out to a tree. If need be, detecting boundary extents could
337 * trivially be added in a future version of ocfs2_get_block().
339 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
340 struct list_head
*pages
, unsigned nr_pages
)
343 struct inode
*inode
= mapping
->host
;
344 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
349 * Use the nonblocking flag for the dlm code to avoid page
350 * lock inversion, but don't bother with retrying.
352 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
356 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
357 ocfs2_inode_unlock(inode
, 0);
362 * Don't bother with inline-data. There isn't anything
363 * to read-ahead in that case anyway...
365 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
369 * Check whether a remote node truncated this file - we just
370 * drop out in that case as it's not worth handling here.
372 last
= list_entry(pages
->prev
, struct page
, lru
);
373 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
374 if (start
>= i_size_read(inode
))
377 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
380 up_read(&oi
->ip_alloc_sem
);
381 ocfs2_inode_unlock(inode
, 0);
386 /* Note: Because we don't support holes, our allocation has
387 * already happened (allocation writes zeros to the file data)
388 * so we don't have to worry about ordered writes in
391 * ->writepage is called during the process of invalidating the page cache
392 * during blocked lock processing. It can't block on any cluster locks
393 * to during block mapping. It's relying on the fact that the block
394 * mapping can't have disappeared under the dirty pages that it is
395 * being asked to write back.
397 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
399 trace_ocfs2_writepage(
400 (unsigned long long)OCFS2_I(page
->mapping
->host
)->ip_blkno
,
403 return block_write_full_page(page
, ocfs2_get_block
, wbc
);
406 /* Taken from ext3. We don't necessarily need the full blown
407 * functionality yet, but IMHO it's better to cut and paste the whole
408 * thing so we can avoid introducing our own bugs (and easily pick up
409 * their fixes when they happen) --Mark */
410 int walk_page_buffers( handle_t
*handle
,
411 struct buffer_head
*head
,
415 int (*fn
)( handle_t
*handle
,
416 struct buffer_head
*bh
))
418 struct buffer_head
*bh
;
419 unsigned block_start
, block_end
;
420 unsigned blocksize
= head
->b_size
;
422 struct buffer_head
*next
;
424 for ( bh
= head
, block_start
= 0;
425 ret
== 0 && (bh
!= head
|| !block_start
);
426 block_start
= block_end
, bh
= next
)
428 next
= bh
->b_this_page
;
429 block_end
= block_start
+ blocksize
;
430 if (block_end
<= from
|| block_start
>= to
) {
431 if (partial
&& !buffer_uptodate(bh
))
435 err
= (*fn
)(handle
, bh
);
442 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
447 struct inode
*inode
= mapping
->host
;
449 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
450 (unsigned long long)block
);
452 /* We don't need to lock journal system files, since they aren't
453 * accessed concurrently from multiple nodes.
455 if (!INODE_JOURNAL(inode
)) {
456 err
= ocfs2_inode_lock(inode
, NULL
, 0);
462 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
465 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
466 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
469 if (!INODE_JOURNAL(inode
)) {
470 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
471 ocfs2_inode_unlock(inode
, 0);
475 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
476 (unsigned long long)block
);
482 status
= err
? 0 : p_blkno
;
488 * TODO: Make this into a generic get_blocks function.
490 * From do_direct_io in direct-io.c:
491 * "So what we do is to permit the ->get_blocks function to populate
492 * bh.b_size with the size of IO which is permitted at this offset and
495 * This function is called directly from get_more_blocks in direct-io.c.
497 * called like this: dio->get_blocks(dio->inode, fs_startblk,
498 * fs_count, map_bh, dio->rw == WRITE);
500 * Note that we never bother to allocate blocks here, and thus ignore the
503 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
504 struct buffer_head
*bh_result
, int create
)
507 u64 p_blkno
, inode_blocks
, contig_blocks
;
508 unsigned int ext_flags
;
509 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
510 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
512 /* This function won't even be called if the request isn't all
513 * nicely aligned and of the right size, so there's no need
514 * for us to check any of that. */
516 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
518 /* This figures out the size of the next contiguous block, and
519 * our logical offset */
520 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
521 &contig_blocks
, &ext_flags
);
523 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
524 (unsigned long long)iblock
);
529 /* We should already CoW the refcounted extent in case of create. */
530 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
533 * get_more_blocks() expects us to describe a hole by clearing
534 * the mapped bit on bh_result().
536 * Consider an unwritten extent as a hole.
538 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
539 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
541 clear_buffer_mapped(bh_result
);
543 /* make sure we don't map more than max_blocks blocks here as
544 that's all the kernel will handle at this point. */
545 if (max_blocks
< contig_blocks
)
546 contig_blocks
= max_blocks
;
547 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
553 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
554 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
555 * to protect io on one node from truncation on another.
557 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
564 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
567 /* this io's submitter should not have unlocked this before we could */
568 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
570 if (ocfs2_iocb_is_sem_locked(iocb
))
571 ocfs2_iocb_clear_sem_locked(iocb
);
573 ocfs2_iocb_clear_rw_locked(iocb
);
575 level
= ocfs2_iocb_rw_locked_level(iocb
);
576 ocfs2_rw_unlock(inode
, level
);
579 aio_complete(iocb
, ret
, 0);
583 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
584 * from ext3. PageChecked() bits have been removed as OCFS2 does not
585 * do journalled data.
587 static void ocfs2_invalidatepage(struct page
*page
, unsigned long offset
)
589 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
591 jbd2_journal_invalidatepage(journal
, page
, offset
);
594 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
596 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
598 if (!page_has_buffers(page
))
600 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
603 static ssize_t
ocfs2_direct_IO(int rw
,
605 const struct iovec
*iov
,
607 unsigned long nr_segs
)
609 struct file
*file
= iocb
->ki_filp
;
610 struct inode
*inode
= file
->f_path
.dentry
->d_inode
->i_mapping
->host
;
613 * Fallback to buffered I/O if we see an inode without
616 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
619 /* Fallback to buffered I/O if we are appending. */
620 if (i_size_read(inode
) <= offset
)
623 return __blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
624 iov
, offset
, nr_segs
,
625 ocfs2_direct_IO_get_blocks
,
626 ocfs2_dio_end_io
, NULL
, 0);
629 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
634 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
636 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
639 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
641 cluster_start
= cpos
% cpp
;
642 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
644 cluster_end
= cluster_start
+ osb
->s_clustersize
;
647 BUG_ON(cluster_start
> PAGE_SIZE
);
648 BUG_ON(cluster_end
> PAGE_SIZE
);
651 *start
= cluster_start
;
657 * 'from' and 'to' are the region in the page to avoid zeroing.
659 * If pagesize > clustersize, this function will avoid zeroing outside
660 * of the cluster boundary.
662 * from == to == 0 is code for "zero the entire cluster region"
664 static void ocfs2_clear_page_regions(struct page
*page
,
665 struct ocfs2_super
*osb
, u32 cpos
,
666 unsigned from
, unsigned to
)
669 unsigned int cluster_start
, cluster_end
;
671 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
673 kaddr
= kmap_atomic(page
, KM_USER0
);
676 if (from
> cluster_start
)
677 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
678 if (to
< cluster_end
)
679 memset(kaddr
+ to
, 0, cluster_end
- to
);
681 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
684 kunmap_atomic(kaddr
, KM_USER0
);
688 * Nonsparse file systems fully allocate before we get to the write
689 * code. This prevents ocfs2_write() from tagging the write as an
690 * allocating one, which means ocfs2_map_page_blocks() might try to
691 * read-in the blocks at the tail of our file. Avoid reading them by
692 * testing i_size against each block offset.
694 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
695 unsigned int block_start
)
697 u64 offset
= page_offset(page
) + block_start
;
699 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
702 if (i_size_read(inode
) > offset
)
709 * Some of this taken from __block_write_begin(). We already have our
710 * mapping by now though, and the entire write will be allocating or
711 * it won't, so not much need to use BH_New.
713 * This will also skip zeroing, which is handled externally.
715 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
716 struct inode
*inode
, unsigned int from
,
717 unsigned int to
, int new)
720 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
721 unsigned int block_end
, block_start
;
722 unsigned int bsize
= 1 << inode
->i_blkbits
;
724 if (!page_has_buffers(page
))
725 create_empty_buffers(page
, bsize
, 0);
727 head
= page_buffers(page
);
728 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
729 bh
= bh
->b_this_page
, block_start
+= bsize
) {
730 block_end
= block_start
+ bsize
;
732 clear_buffer_new(bh
);
735 * Ignore blocks outside of our i/o range -
736 * they may belong to unallocated clusters.
738 if (block_start
>= to
|| block_end
<= from
) {
739 if (PageUptodate(page
))
740 set_buffer_uptodate(bh
);
745 * For an allocating write with cluster size >= page
746 * size, we always write the entire page.
751 if (!buffer_mapped(bh
)) {
752 map_bh(bh
, inode
->i_sb
, *p_blkno
);
753 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
756 if (PageUptodate(page
)) {
757 if (!buffer_uptodate(bh
))
758 set_buffer_uptodate(bh
);
759 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
761 ocfs2_should_read_blk(inode
, page
, block_start
) &&
762 (block_start
< from
|| block_end
> to
)) {
763 ll_rw_block(READ
, 1, &bh
);
767 *p_blkno
= *p_blkno
+ 1;
771 * If we issued read requests - let them complete.
773 while(wait_bh
> wait
) {
774 wait_on_buffer(*--wait_bh
);
775 if (!buffer_uptodate(*wait_bh
))
779 if (ret
== 0 || !new)
783 * If we get -EIO above, zero out any newly allocated blocks
784 * to avoid exposing stale data.
789 block_end
= block_start
+ bsize
;
790 if (block_end
<= from
)
792 if (block_start
>= to
)
795 zero_user(page
, block_start
, bh
->b_size
);
796 set_buffer_uptodate(bh
);
797 mark_buffer_dirty(bh
);
800 block_start
= block_end
;
801 bh
= bh
->b_this_page
;
802 } while (bh
!= head
);
807 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
808 #define OCFS2_MAX_CTXT_PAGES 1
810 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
813 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
816 * Describe the state of a single cluster to be written to.
818 struct ocfs2_write_cluster_desc
{
822 * Give this a unique field because c_phys eventually gets
826 unsigned c_unwritten
;
827 unsigned c_needs_zero
;
830 struct ocfs2_write_ctxt
{
831 /* Logical cluster position / len of write */
835 /* First cluster allocated in a nonsparse extend */
836 u32 w_first_new_cpos
;
838 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
841 * This is true if page_size > cluster_size.
843 * It triggers a set of special cases during write which might
844 * have to deal with allocating writes to partial pages.
846 unsigned int w_large_pages
;
849 * Pages involved in this write.
851 * w_target_page is the page being written to by the user.
853 * w_pages is an array of pages which always contains
854 * w_target_page, and in the case of an allocating write with
855 * page_size < cluster size, it will contain zero'd and mapped
856 * pages adjacent to w_target_page which need to be written
857 * out in so that future reads from that region will get
860 unsigned int w_num_pages
;
861 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
862 struct page
*w_target_page
;
865 * ocfs2_write_end() uses this to know what the real range to
866 * write in the target should be.
868 unsigned int w_target_from
;
869 unsigned int w_target_to
;
872 * We could use journal_current_handle() but this is cleaner,
877 struct buffer_head
*w_di_bh
;
879 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
882 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
886 for(i
= 0; i
< num_pages
; i
++) {
888 unlock_page(pages
[i
]);
889 mark_page_accessed(pages
[i
]);
890 page_cache_release(pages
[i
]);
895 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
897 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
903 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
904 struct ocfs2_super
*osb
, loff_t pos
,
905 unsigned len
, struct buffer_head
*di_bh
)
908 struct ocfs2_write_ctxt
*wc
;
910 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
914 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
915 wc
->w_first_new_cpos
= UINT_MAX
;
916 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
917 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
921 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
922 wc
->w_large_pages
= 1;
924 wc
->w_large_pages
= 0;
926 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
934 * If a page has any new buffers, zero them out here, and mark them uptodate
935 * and dirty so they'll be written out (in order to prevent uninitialised
936 * block data from leaking). And clear the new bit.
938 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
940 unsigned int block_start
, block_end
;
941 struct buffer_head
*head
, *bh
;
943 BUG_ON(!PageLocked(page
));
944 if (!page_has_buffers(page
))
947 bh
= head
= page_buffers(page
);
950 block_end
= block_start
+ bh
->b_size
;
952 if (buffer_new(bh
)) {
953 if (block_end
> from
&& block_start
< to
) {
954 if (!PageUptodate(page
)) {
957 start
= max(from
, block_start
);
958 end
= min(to
, block_end
);
960 zero_user_segment(page
, start
, end
);
961 set_buffer_uptodate(bh
);
964 clear_buffer_new(bh
);
965 mark_buffer_dirty(bh
);
969 block_start
= block_end
;
970 bh
= bh
->b_this_page
;
971 } while (bh
!= head
);
975 * Only called when we have a failure during allocating write to write
976 * zero's to the newly allocated region.
978 static void ocfs2_write_failure(struct inode
*inode
,
979 struct ocfs2_write_ctxt
*wc
,
980 loff_t user_pos
, unsigned user_len
)
983 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
984 to
= user_pos
+ user_len
;
985 struct page
*tmppage
;
987 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
989 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
990 tmppage
= wc
->w_pages
[i
];
992 if (page_has_buffers(tmppage
)) {
993 if (ocfs2_should_order_data(inode
))
994 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
996 block_commit_write(tmppage
, from
, to
);
1001 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1002 struct ocfs2_write_ctxt
*wc
,
1003 struct page
*page
, u32 cpos
,
1004 loff_t user_pos
, unsigned user_len
,
1008 unsigned int map_from
= 0, map_to
= 0;
1009 unsigned int cluster_start
, cluster_end
;
1010 unsigned int user_data_from
= 0, user_data_to
= 0;
1012 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1013 &cluster_start
, &cluster_end
);
1015 /* treat the write as new if the a hole/lseek spanned across
1016 * the page boundary.
1018 new = new | ((i_size_read(inode
) <= page_offset(page
)) &&
1019 (page_offset(page
) <= user_pos
));
1021 if (page
== wc
->w_target_page
) {
1022 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1023 map_to
= map_from
+ user_len
;
1026 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1027 cluster_start
, cluster_end
,
1030 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1031 map_from
, map_to
, new);
1037 user_data_from
= map_from
;
1038 user_data_to
= map_to
;
1040 map_from
= cluster_start
;
1041 map_to
= cluster_end
;
1045 * If we haven't allocated the new page yet, we
1046 * shouldn't be writing it out without copying user
1047 * data. This is likely a math error from the caller.
1051 map_from
= cluster_start
;
1052 map_to
= cluster_end
;
1054 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1055 cluster_start
, cluster_end
, new);
1063 * Parts of newly allocated pages need to be zero'd.
1065 * Above, we have also rewritten 'to' and 'from' - as far as
1066 * the rest of the function is concerned, the entire cluster
1067 * range inside of a page needs to be written.
1069 * We can skip this if the page is up to date - it's already
1070 * been zero'd from being read in as a hole.
1072 if (new && !PageUptodate(page
))
1073 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1074 cpos
, user_data_from
, user_data_to
);
1076 flush_dcache_page(page
);
1083 * This function will only grab one clusters worth of pages.
1085 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1086 struct ocfs2_write_ctxt
*wc
,
1087 u32 cpos
, loff_t user_pos
,
1088 unsigned user_len
, int new,
1089 struct page
*mmap_page
)
1092 unsigned long start
, target_index
, end_index
, index
;
1093 struct inode
*inode
= mapping
->host
;
1096 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1099 * Figure out how many pages we'll be manipulating here. For
1100 * non allocating write, we just change the one
1101 * page. Otherwise, we'll need a whole clusters worth. If we're
1102 * writing past i_size, we only need enough pages to cover the
1103 * last page of the write.
1106 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1107 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1109 * We need the index *past* the last page we could possibly
1110 * touch. This is the page past the end of the write or
1111 * i_size, whichever is greater.
1113 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1114 BUG_ON(last_byte
< 1);
1115 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1116 if ((start
+ wc
->w_num_pages
) > end_index
)
1117 wc
->w_num_pages
= end_index
- start
;
1119 wc
->w_num_pages
= 1;
1120 start
= target_index
;
1123 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1126 if (index
== target_index
&& mmap_page
) {
1128 * ocfs2_pagemkwrite() is a little different
1129 * and wants us to directly use the page
1132 lock_page(mmap_page
);
1134 if (mmap_page
->mapping
!= mapping
) {
1135 unlock_page(mmap_page
);
1137 * Sanity check - the locking in
1138 * ocfs2_pagemkwrite() should ensure
1139 * that this code doesn't trigger.
1146 page_cache_get(mmap_page
);
1147 wc
->w_pages
[i
] = mmap_page
;
1149 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1151 if (!wc
->w_pages
[i
]) {
1158 if (index
== target_index
)
1159 wc
->w_target_page
= wc
->w_pages
[i
];
1166 * Prepare a single cluster for write one cluster into the file.
1168 static int ocfs2_write_cluster(struct address_space
*mapping
,
1169 u32 phys
, unsigned int unwritten
,
1170 unsigned int should_zero
,
1171 struct ocfs2_alloc_context
*data_ac
,
1172 struct ocfs2_alloc_context
*meta_ac
,
1173 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1174 loff_t user_pos
, unsigned user_len
)
1177 u64 v_blkno
, p_blkno
;
1178 struct inode
*inode
= mapping
->host
;
1179 struct ocfs2_extent_tree et
;
1181 new = phys
== 0 ? 1 : 0;
1186 * This is safe to call with the page locks - it won't take
1187 * any additional semaphores or cluster locks.
1190 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1191 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1192 wc
->w_handle
, data_ac
,
1195 * This shouldn't happen because we must have already
1196 * calculated the correct meta data allocation required. The
1197 * internal tree allocation code should know how to increase
1198 * transaction credits itself.
1200 * If need be, we could handle -EAGAIN for a
1201 * RESTART_TRANS here.
1203 mlog_bug_on_msg(ret
== -EAGAIN
,
1204 "Inode %llu: EAGAIN return during allocation.\n",
1205 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1210 } else if (unwritten
) {
1211 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1213 ret
= ocfs2_mark_extent_written(inode
, &et
,
1214 wc
->w_handle
, cpos
, 1, phys
,
1215 meta_ac
, &wc
->w_dealloc
);
1223 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1225 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1228 * The only reason this should fail is due to an inability to
1229 * find the extent added.
1231 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1234 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1235 "at logical block %llu",
1236 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1237 (unsigned long long)v_blkno
);
1241 BUG_ON(p_blkno
== 0);
1243 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1246 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1247 wc
->w_pages
[i
], cpos
,
1258 * We only have cleanup to do in case of allocating write.
1261 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1268 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1269 struct ocfs2_alloc_context
*data_ac
,
1270 struct ocfs2_alloc_context
*meta_ac
,
1271 struct ocfs2_write_ctxt
*wc
,
1272 loff_t pos
, unsigned len
)
1276 unsigned int local_len
= len
;
1277 struct ocfs2_write_cluster_desc
*desc
;
1278 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1280 for (i
= 0; i
< wc
->w_clen
; i
++) {
1281 desc
= &wc
->w_desc
[i
];
1284 * We have to make sure that the total write passed in
1285 * doesn't extend past a single cluster.
1288 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1289 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1290 local_len
= osb
->s_clustersize
- cluster_off
;
1292 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1296 wc
, desc
->c_cpos
, pos
, local_len
);
1312 * ocfs2_write_end() wants to know which parts of the target page it
1313 * should complete the write on. It's easiest to compute them ahead of
1314 * time when a more complete view of the write is available.
1316 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1317 struct ocfs2_write_ctxt
*wc
,
1318 loff_t pos
, unsigned len
, int alloc
)
1320 struct ocfs2_write_cluster_desc
*desc
;
1322 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1323 wc
->w_target_to
= wc
->w_target_from
+ len
;
1329 * Allocating write - we may have different boundaries based
1330 * on page size and cluster size.
1332 * NOTE: We can no longer compute one value from the other as
1333 * the actual write length and user provided length may be
1337 if (wc
->w_large_pages
) {
1339 * We only care about the 1st and last cluster within
1340 * our range and whether they should be zero'd or not. Either
1341 * value may be extended out to the start/end of a
1342 * newly allocated cluster.
1344 desc
= &wc
->w_desc
[0];
1345 if (desc
->c_needs_zero
)
1346 ocfs2_figure_cluster_boundaries(osb
,
1351 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1352 if (desc
->c_needs_zero
)
1353 ocfs2_figure_cluster_boundaries(osb
,
1358 wc
->w_target_from
= 0;
1359 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1364 * Populate each single-cluster write descriptor in the write context
1365 * with information about the i/o to be done.
1367 * Returns the number of clusters that will have to be allocated, as
1368 * well as a worst case estimate of the number of extent records that
1369 * would have to be created during a write to an unwritten region.
1371 static int ocfs2_populate_write_desc(struct inode
*inode
,
1372 struct ocfs2_write_ctxt
*wc
,
1373 unsigned int *clusters_to_alloc
,
1374 unsigned int *extents_to_split
)
1377 struct ocfs2_write_cluster_desc
*desc
;
1378 unsigned int num_clusters
= 0;
1379 unsigned int ext_flags
= 0;
1383 *clusters_to_alloc
= 0;
1384 *extents_to_split
= 0;
1386 for (i
= 0; i
< wc
->w_clen
; i
++) {
1387 desc
= &wc
->w_desc
[i
];
1388 desc
->c_cpos
= wc
->w_cpos
+ i
;
1390 if (num_clusters
== 0) {
1392 * Need to look up the next extent record.
1394 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1395 &num_clusters
, &ext_flags
);
1401 /* We should already CoW the refcountd extent. */
1402 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1405 * Assume worst case - that we're writing in
1406 * the middle of the extent.
1408 * We can assume that the write proceeds from
1409 * left to right, in which case the extent
1410 * insert code is smart enough to coalesce the
1411 * next splits into the previous records created.
1413 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1414 *extents_to_split
= *extents_to_split
+ 2;
1417 * Only increment phys if it doesn't describe
1424 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1425 * file that got extended. w_first_new_cpos tells us
1426 * where the newly allocated clusters are so we can
1429 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1431 desc
->c_needs_zero
= 1;
1434 desc
->c_phys
= phys
;
1437 desc
->c_needs_zero
= 1;
1438 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1441 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1442 desc
->c_unwritten
= 1;
1443 desc
->c_needs_zero
= 1;
1454 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1455 struct inode
*inode
,
1456 struct ocfs2_write_ctxt
*wc
)
1459 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1462 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1464 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1471 * If we don't set w_num_pages then this page won't get unlocked
1472 * and freed on cleanup of the write context.
1474 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1475 wc
->w_num_pages
= 1;
1477 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1478 if (IS_ERR(handle
)) {
1479 ret
= PTR_ERR(handle
);
1484 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1485 OCFS2_JOURNAL_ACCESS_WRITE
);
1487 ocfs2_commit_trans(osb
, handle
);
1493 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1494 ocfs2_set_inode_data_inline(inode
, di
);
1496 if (!PageUptodate(page
)) {
1497 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1499 ocfs2_commit_trans(osb
, handle
);
1505 wc
->w_handle
= handle
;
1510 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1512 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1514 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1519 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1520 struct inode
*inode
, loff_t pos
,
1521 unsigned len
, struct page
*mmap_page
,
1522 struct ocfs2_write_ctxt
*wc
)
1524 int ret
, written
= 0;
1525 loff_t end
= pos
+ len
;
1526 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1527 struct ocfs2_dinode
*di
= NULL
;
1529 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi
->ip_blkno
,
1530 len
, (unsigned long long)pos
,
1531 oi
->ip_dyn_features
);
1534 * Handle inodes which already have inline data 1st.
1536 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1537 if (mmap_page
== NULL
&&
1538 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1539 goto do_inline_write
;
1542 * The write won't fit - we have to give this inode an
1543 * inline extent list now.
1545 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1552 * Check whether the inode can accept inline data.
1554 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1558 * Check whether the write can fit.
1560 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1562 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1566 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1573 * This signals to the caller that the data can be written
1578 return written
? written
: ret
;
1582 * This function only does anything for file systems which can't
1583 * handle sparse files.
1585 * What we want to do here is fill in any hole between the current end
1586 * of allocation and the end of our write. That way the rest of the
1587 * write path can treat it as an non-allocating write, which has no
1588 * special case code for sparse/nonsparse files.
1590 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1591 struct buffer_head
*di_bh
,
1592 loff_t pos
, unsigned len
,
1593 struct ocfs2_write_ctxt
*wc
)
1596 loff_t newsize
= pos
+ len
;
1598 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1600 if (newsize
<= i_size_read(inode
))
1603 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1607 wc
->w_first_new_cpos
=
1608 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1613 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
1618 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1619 if (pos
> i_size_read(inode
))
1620 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
1626 * Try to flush truncate logs if we can free enough clusters from it.
1627 * As for return value, "< 0" means error, "0" no space and "1" means
1628 * we have freed enough spaces and let the caller try to allocate again.
1630 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
1631 unsigned int needed
)
1635 unsigned int truncated_clusters
;
1637 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
1638 truncated_clusters
= osb
->truncated_clusters
;
1639 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
1642 * Check whether we can succeed in allocating if we free
1645 if (truncated_clusters
< needed
)
1648 ret
= ocfs2_flush_truncate_log(osb
);
1654 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
1655 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
1662 int ocfs2_write_begin_nolock(struct file
*filp
,
1663 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
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1669 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
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
;
1678 int try_free
= 1, ret1
;
1681 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1687 if (ocfs2_supports_inline_data(osb
)) {
1688 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1700 if (ocfs2_sparse_alloc(osb
))
1701 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
1703 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
1710 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1714 } else if (ret
== 1) {
1715 clusters_need
= wc
->w_clen
;
1716 ret
= ocfs2_refcount_cow(inode
, filp
, di_bh
,
1717 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1724 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1730 clusters_need
+= clusters_to_alloc
;
1732 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1734 trace_ocfs2_write_begin_nolock(
1735 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1736 (long long)i_size_read(inode
),
1737 le32_to_cpu(di
->i_clusters
),
1738 pos
, len
, flags
, mmap_page
,
1739 clusters_to_alloc
, extents_to_split
);
1742 * We set w_target_from, w_target_to here so that
1743 * ocfs2_write_end() knows which range in the target page to
1744 * write out. An allocation requires that we write the entire
1747 if (clusters_to_alloc
|| extents_to_split
) {
1749 * XXX: We are stretching the limits of
1750 * ocfs2_lock_allocators(). It greatly over-estimates
1751 * the work to be done.
1753 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1755 ret
= ocfs2_lock_allocators(inode
, &et
,
1756 clusters_to_alloc
, extents_to_split
,
1757 &data_ac
, &meta_ac
);
1764 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
1766 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1773 * We have to zero sparse allocated clusters, unwritten extent clusters,
1774 * and non-sparse clusters we just extended. For non-sparse writes,
1775 * we know zeros will only be needed in the first and/or last cluster.
1777 if (clusters_to_alloc
|| extents_to_split
||
1778 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
1779 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
1780 cluster_of_pages
= 1;
1782 cluster_of_pages
= 0;
1784 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1786 handle
= ocfs2_start_trans(osb
, credits
);
1787 if (IS_ERR(handle
)) {
1788 ret
= PTR_ERR(handle
);
1793 wc
->w_handle
= handle
;
1795 if (clusters_to_alloc
) {
1796 ret
= dquot_alloc_space_nodirty(inode
,
1797 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1802 * We don't want this to fail in ocfs2_write_end(), so do it
1805 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1806 OCFS2_JOURNAL_ACCESS_WRITE
);
1813 * Fill our page array first. That way we've grabbed enough so
1814 * that we can zero and flush if we error after adding the
1817 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
1818 cluster_of_pages
, mmap_page
);
1824 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1832 ocfs2_free_alloc_context(data_ac
);
1834 ocfs2_free_alloc_context(meta_ac
);
1837 *pagep
= wc
->w_target_page
;
1841 if (clusters_to_alloc
)
1842 dquot_free_space(inode
,
1843 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1845 ocfs2_commit_trans(osb
, handle
);
1848 ocfs2_free_write_ctxt(wc
);
1851 ocfs2_free_alloc_context(data_ac
);
1853 ocfs2_free_alloc_context(meta_ac
);
1855 if (ret
== -ENOSPC
&& try_free
) {
1857 * Try to free some truncate log so that we can have enough
1858 * clusters to allocate.
1862 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
1873 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1874 loff_t pos
, unsigned len
, unsigned flags
,
1875 struct page
**pagep
, void **fsdata
)
1878 struct buffer_head
*di_bh
= NULL
;
1879 struct inode
*inode
= mapping
->host
;
1881 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1888 * Take alloc sem here to prevent concurrent lookups. That way
1889 * the mapping, zeroing and tree manipulation within
1890 * ocfs2_write() will be safe against ->readpage(). This
1891 * should also serve to lock out allocation from a shared
1894 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1896 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
1897 fsdata
, di_bh
, NULL
);
1908 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1911 ocfs2_inode_unlock(inode
, 1);
1916 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1917 unsigned len
, unsigned *copied
,
1918 struct ocfs2_dinode
*di
,
1919 struct ocfs2_write_ctxt
*wc
)
1923 if (unlikely(*copied
< len
)) {
1924 if (!PageUptodate(wc
->w_target_page
)) {
1930 kaddr
= kmap_atomic(wc
->w_target_page
, KM_USER0
);
1931 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1932 kunmap_atomic(kaddr
, KM_USER0
);
1934 trace_ocfs2_write_end_inline(
1935 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1936 (unsigned long long)pos
, *copied
,
1937 le16_to_cpu(di
->id2
.i_data
.id_count
),
1938 le16_to_cpu(di
->i_dyn_features
));
1941 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1942 loff_t pos
, unsigned len
, unsigned copied
,
1943 struct page
*page
, void *fsdata
)
1946 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
1947 struct inode
*inode
= mapping
->host
;
1948 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1949 struct ocfs2_write_ctxt
*wc
= fsdata
;
1950 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1951 handle_t
*handle
= wc
->w_handle
;
1952 struct page
*tmppage
;
1954 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1955 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
1956 goto out_write_size
;
1959 if (unlikely(copied
< len
)) {
1960 if (!PageUptodate(wc
->w_target_page
))
1963 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
1966 flush_dcache_page(wc
->w_target_page
);
1968 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1969 tmppage
= wc
->w_pages
[i
];
1971 if (tmppage
== wc
->w_target_page
) {
1972 from
= wc
->w_target_from
;
1973 to
= wc
->w_target_to
;
1975 BUG_ON(from
> PAGE_CACHE_SIZE
||
1976 to
> PAGE_CACHE_SIZE
||
1980 * Pages adjacent to the target (if any) imply
1981 * a hole-filling write in which case we want
1982 * to flush their entire range.
1985 to
= PAGE_CACHE_SIZE
;
1988 if (page_has_buffers(tmppage
)) {
1989 if (ocfs2_should_order_data(inode
))
1990 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1991 block_commit_write(tmppage
, from
, to
);
1997 if (pos
> inode
->i_size
) {
1998 i_size_write(inode
, pos
);
1999 mark_inode_dirty(inode
);
2001 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
2002 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
2003 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2004 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
2005 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2006 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2008 ocfs2_commit_trans(osb
, handle
);
2010 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2012 ocfs2_free_write_ctxt(wc
);
2017 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2018 loff_t pos
, unsigned len
, unsigned copied
,
2019 struct page
*page
, void *fsdata
)
2022 struct inode
*inode
= mapping
->host
;
2024 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2026 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2027 ocfs2_inode_unlock(inode
, 1);
2032 const struct address_space_operations ocfs2_aops
= {
2033 .readpage
= ocfs2_readpage
,
2034 .readpages
= ocfs2_readpages
,
2035 .writepage
= ocfs2_writepage
,
2036 .write_begin
= ocfs2_write_begin
,
2037 .write_end
= ocfs2_write_end
,
2039 .direct_IO
= ocfs2_direct_IO
,
2040 .invalidatepage
= ocfs2_invalidatepage
,
2041 .releasepage
= ocfs2_releasepage
,
2042 .migratepage
= buffer_migrate_page
,
2043 .is_partially_uptodate
= block_is_partially_uptodate
,
2044 .error_remove_page
= generic_error_remove_page
,