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
);
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
);
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
);
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
);
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) {
294 * Unlock the page and cycle ip_alloc_sem so that we don't
295 * busyloop waiting for ip_alloc_sem to unlock
297 ret
= AOP_TRUNCATED_PAGE
;
300 down_read(&oi
->ip_alloc_sem
);
301 up_read(&oi
->ip_alloc_sem
);
302 goto out_inode_unlock
;
306 * i_size might have just been updated as we grabed the meta lock. We
307 * might now be discovering a truncate that hit on another node.
308 * block_read_full_page->get_block freaks out if it is asked to read
309 * beyond the end of a file, so we check here. Callers
310 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
311 * and notice that the page they just read isn't needed.
313 * XXX sys_readahead() seems to get that wrong?
315 if (start
>= i_size_read(inode
)) {
316 zero_user(page
, 0, PAGE_SIZE
);
317 SetPageUptodate(page
);
322 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
323 ret
= ocfs2_readpage_inline(inode
, page
);
325 ret
= block_read_full_page(page
, ocfs2_get_block
);
329 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
331 ocfs2_inode_unlock(inode
, 0);
339 * This is used only for read-ahead. Failures or difficult to handle
340 * situations are safe to ignore.
342 * Right now, we don't bother with BH_Boundary - in-inode extent lists
343 * are quite large (243 extents on 4k blocks), so most inodes don't
344 * grow out to a tree. If need be, detecting boundary extents could
345 * trivially be added in a future version of ocfs2_get_block().
347 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
348 struct list_head
*pages
, unsigned nr_pages
)
351 struct inode
*inode
= mapping
->host
;
352 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
357 * Use the nonblocking flag for the dlm code to avoid page
358 * lock inversion, but don't bother with retrying.
360 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
364 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
365 ocfs2_inode_unlock(inode
, 0);
370 * Don't bother with inline-data. There isn't anything
371 * to read-ahead in that case anyway...
373 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
377 * Check whether a remote node truncated this file - we just
378 * drop out in that case as it's not worth handling here.
380 last
= list_entry(pages
->prev
, struct page
, lru
);
381 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
382 if (start
>= i_size_read(inode
))
385 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
388 up_read(&oi
->ip_alloc_sem
);
389 ocfs2_inode_unlock(inode
, 0);
394 /* Note: Because we don't support holes, our allocation has
395 * already happened (allocation writes zeros to the file data)
396 * so we don't have to worry about ordered writes in
399 * ->writepage is called during the process of invalidating the page cache
400 * during blocked lock processing. It can't block on any cluster locks
401 * to during block mapping. It's relying on the fact that the block
402 * mapping can't have disappeared under the dirty pages that it is
403 * being asked to write back.
405 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
407 trace_ocfs2_writepage(
408 (unsigned long long)OCFS2_I(page
->mapping
->host
)->ip_blkno
,
411 return block_write_full_page(page
, ocfs2_get_block
, wbc
);
414 /* Taken from ext3. We don't necessarily need the full blown
415 * functionality yet, but IMHO it's better to cut and paste the whole
416 * thing so we can avoid introducing our own bugs (and easily pick up
417 * their fixes when they happen) --Mark */
418 int walk_page_buffers( handle_t
*handle
,
419 struct buffer_head
*head
,
423 int (*fn
)( handle_t
*handle
,
424 struct buffer_head
*bh
))
426 struct buffer_head
*bh
;
427 unsigned block_start
, block_end
;
428 unsigned blocksize
= head
->b_size
;
430 struct buffer_head
*next
;
432 for ( bh
= head
, block_start
= 0;
433 ret
== 0 && (bh
!= head
|| !block_start
);
434 block_start
= block_end
, bh
= next
)
436 next
= bh
->b_this_page
;
437 block_end
= block_start
+ blocksize
;
438 if (block_end
<= from
|| block_start
>= to
) {
439 if (partial
&& !buffer_uptodate(bh
))
443 err
= (*fn
)(handle
, bh
);
450 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
455 struct inode
*inode
= mapping
->host
;
457 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
458 (unsigned long long)block
);
460 /* We don't need to lock journal system files, since they aren't
461 * accessed concurrently from multiple nodes.
463 if (!INODE_JOURNAL(inode
)) {
464 err
= ocfs2_inode_lock(inode
, NULL
, 0);
470 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
473 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
474 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
477 if (!INODE_JOURNAL(inode
)) {
478 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
479 ocfs2_inode_unlock(inode
, 0);
483 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
484 (unsigned long long)block
);
490 status
= err
? 0 : p_blkno
;
496 * TODO: Make this into a generic get_blocks function.
498 * From do_direct_io in direct-io.c:
499 * "So what we do is to permit the ->get_blocks function to populate
500 * bh.b_size with the size of IO which is permitted at this offset and
503 * This function is called directly from get_more_blocks in direct-io.c.
505 * called like this: dio->get_blocks(dio->inode, fs_startblk,
506 * fs_count, map_bh, dio->rw == WRITE);
508 * Note that we never bother to allocate blocks here, and thus ignore the
511 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
512 struct buffer_head
*bh_result
, int create
)
515 u64 p_blkno
, inode_blocks
, contig_blocks
;
516 unsigned int ext_flags
;
517 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
518 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
520 /* This function won't even be called if the request isn't all
521 * nicely aligned and of the right size, so there's no need
522 * for us to check any of that. */
524 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
526 /* This figures out the size of the next contiguous block, and
527 * our logical offset */
528 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
529 &contig_blocks
, &ext_flags
);
531 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
532 (unsigned long long)iblock
);
537 /* We should already CoW the refcounted extent in case of create. */
538 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
541 * get_more_blocks() expects us to describe a hole by clearing
542 * the mapped bit on bh_result().
544 * Consider an unwritten extent as a hole.
546 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
547 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
549 clear_buffer_mapped(bh_result
);
551 /* make sure we don't map more than max_blocks blocks here as
552 that's all the kernel will handle at this point. */
553 if (max_blocks
< contig_blocks
)
554 contig_blocks
= max_blocks
;
555 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
561 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
562 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
563 * to protect io on one node from truncation on another.
565 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
572 struct inode
*inode
= file_inode(iocb
->ki_filp
);
574 wait_queue_head_t
*wq
= ocfs2_ioend_wq(inode
);
576 /* this io's submitter should not have unlocked this before we could */
577 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
579 if (ocfs2_iocb_is_sem_locked(iocb
))
580 ocfs2_iocb_clear_sem_locked(iocb
);
582 if (ocfs2_iocb_is_unaligned_aio(iocb
)) {
583 ocfs2_iocb_clear_unaligned_aio(iocb
);
585 if (atomic_dec_and_test(&OCFS2_I(inode
)->ip_unaligned_aio
) &&
586 waitqueue_active(wq
)) {
591 ocfs2_iocb_clear_rw_locked(iocb
);
593 level
= ocfs2_iocb_rw_locked_level(iocb
);
594 ocfs2_rw_unlock(inode
, level
);
596 inode_dio_done(inode
);
598 aio_complete(iocb
, ret
, 0);
602 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
603 * from ext3. PageChecked() bits have been removed as OCFS2 does not
604 * do journalled data.
606 static void ocfs2_invalidatepage(struct page
*page
, unsigned int offset
,
609 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
611 jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
614 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
616 journal_t
*journal
= OCFS2_SB(page
->mapping
->host
->i_sb
)->journal
->j_journal
;
618 if (!page_has_buffers(page
))
620 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
623 static ssize_t
ocfs2_direct_IO(int rw
,
625 const struct iovec
*iov
,
627 unsigned long nr_segs
)
629 struct file
*file
= iocb
->ki_filp
;
630 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
633 * Fallback to buffered I/O if we see an inode without
636 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
639 /* Fallback to buffered I/O if we are appending. */
640 if (i_size_read(inode
) <= offset
)
643 return __blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
644 iov
, offset
, nr_segs
,
645 ocfs2_direct_IO_get_blocks
,
646 ocfs2_dio_end_io
, NULL
, 0);
649 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
654 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
656 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
659 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
661 cluster_start
= cpos
% cpp
;
662 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
664 cluster_end
= cluster_start
+ osb
->s_clustersize
;
667 BUG_ON(cluster_start
> PAGE_SIZE
);
668 BUG_ON(cluster_end
> PAGE_SIZE
);
671 *start
= cluster_start
;
677 * 'from' and 'to' are the region in the page to avoid zeroing.
679 * If pagesize > clustersize, this function will avoid zeroing outside
680 * of the cluster boundary.
682 * from == to == 0 is code for "zero the entire cluster region"
684 static void ocfs2_clear_page_regions(struct page
*page
,
685 struct ocfs2_super
*osb
, u32 cpos
,
686 unsigned from
, unsigned to
)
689 unsigned int cluster_start
, cluster_end
;
691 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
693 kaddr
= kmap_atomic(page
);
696 if (from
> cluster_start
)
697 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
698 if (to
< cluster_end
)
699 memset(kaddr
+ to
, 0, cluster_end
- to
);
701 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
704 kunmap_atomic(kaddr
);
708 * Nonsparse file systems fully allocate before we get to the write
709 * code. This prevents ocfs2_write() from tagging the write as an
710 * allocating one, which means ocfs2_map_page_blocks() might try to
711 * read-in the blocks at the tail of our file. Avoid reading them by
712 * testing i_size against each block offset.
714 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
715 unsigned int block_start
)
717 u64 offset
= page_offset(page
) + block_start
;
719 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
722 if (i_size_read(inode
) > offset
)
729 * Some of this taken from __block_write_begin(). We already have our
730 * mapping by now though, and the entire write will be allocating or
731 * it won't, so not much need to use BH_New.
733 * This will also skip zeroing, which is handled externally.
735 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
736 struct inode
*inode
, unsigned int from
,
737 unsigned int to
, int new)
740 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
741 unsigned int block_end
, block_start
;
742 unsigned int bsize
= 1 << inode
->i_blkbits
;
744 if (!page_has_buffers(page
))
745 create_empty_buffers(page
, bsize
, 0);
747 head
= page_buffers(page
);
748 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
749 bh
= bh
->b_this_page
, block_start
+= bsize
) {
750 block_end
= block_start
+ bsize
;
752 clear_buffer_new(bh
);
755 * Ignore blocks outside of our i/o range -
756 * they may belong to unallocated clusters.
758 if (block_start
>= to
|| block_end
<= from
) {
759 if (PageUptodate(page
))
760 set_buffer_uptodate(bh
);
765 * For an allocating write with cluster size >= page
766 * size, we always write the entire page.
771 if (!buffer_mapped(bh
)) {
772 map_bh(bh
, inode
->i_sb
, *p_blkno
);
773 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
776 if (PageUptodate(page
)) {
777 if (!buffer_uptodate(bh
))
778 set_buffer_uptodate(bh
);
779 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
781 ocfs2_should_read_blk(inode
, page
, block_start
) &&
782 (block_start
< from
|| block_end
> to
)) {
783 ll_rw_block(READ
, 1, &bh
);
787 *p_blkno
= *p_blkno
+ 1;
791 * If we issued read requests - let them complete.
793 while(wait_bh
> wait
) {
794 wait_on_buffer(*--wait_bh
);
795 if (!buffer_uptodate(*wait_bh
))
799 if (ret
== 0 || !new)
803 * If we get -EIO above, zero out any newly allocated blocks
804 * to avoid exposing stale data.
809 block_end
= block_start
+ bsize
;
810 if (block_end
<= from
)
812 if (block_start
>= to
)
815 zero_user(page
, block_start
, bh
->b_size
);
816 set_buffer_uptodate(bh
);
817 mark_buffer_dirty(bh
);
820 block_start
= block_end
;
821 bh
= bh
->b_this_page
;
822 } while (bh
!= head
);
827 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
828 #define OCFS2_MAX_CTXT_PAGES 1
830 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
833 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
836 * Describe the state of a single cluster to be written to.
838 struct ocfs2_write_cluster_desc
{
842 * Give this a unique field because c_phys eventually gets
846 unsigned c_unwritten
;
847 unsigned c_needs_zero
;
850 struct ocfs2_write_ctxt
{
851 /* Logical cluster position / len of write */
855 /* First cluster allocated in a nonsparse extend */
856 u32 w_first_new_cpos
;
858 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
861 * This is true if page_size > cluster_size.
863 * It triggers a set of special cases during write which might
864 * have to deal with allocating writes to partial pages.
866 unsigned int w_large_pages
;
869 * Pages involved in this write.
871 * w_target_page is the page being written to by the user.
873 * w_pages is an array of pages which always contains
874 * w_target_page, and in the case of an allocating write with
875 * page_size < cluster size, it will contain zero'd and mapped
876 * pages adjacent to w_target_page which need to be written
877 * out in so that future reads from that region will get
880 unsigned int w_num_pages
;
881 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
882 struct page
*w_target_page
;
885 * w_target_locked is used for page_mkwrite path indicating no unlocking
886 * against w_target_page in ocfs2_write_end_nolock.
888 unsigned int w_target_locked
:1;
891 * ocfs2_write_end() uses this to know what the real range to
892 * write in the target should be.
894 unsigned int w_target_from
;
895 unsigned int w_target_to
;
898 * We could use journal_current_handle() but this is cleaner,
903 struct buffer_head
*w_di_bh
;
905 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
908 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
912 for(i
= 0; i
< num_pages
; i
++) {
914 unlock_page(pages
[i
]);
915 mark_page_accessed(pages
[i
]);
916 page_cache_release(pages
[i
]);
921 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
926 * w_target_locked is only set to true in the page_mkwrite() case.
927 * The intent is to allow us to lock the target page from write_begin()
928 * to write_end(). The caller must hold a ref on w_target_page.
930 if (wc
->w_target_locked
) {
931 BUG_ON(!wc
->w_target_page
);
932 for (i
= 0; i
< wc
->w_num_pages
; i
++) {
933 if (wc
->w_target_page
== wc
->w_pages
[i
]) {
934 wc
->w_pages
[i
] = NULL
;
938 mark_page_accessed(wc
->w_target_page
);
939 page_cache_release(wc
->w_target_page
);
941 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
947 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
948 struct ocfs2_super
*osb
, loff_t pos
,
949 unsigned len
, struct buffer_head
*di_bh
)
952 struct ocfs2_write_ctxt
*wc
;
954 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
958 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
959 wc
->w_first_new_cpos
= UINT_MAX
;
960 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
961 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
965 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
966 wc
->w_large_pages
= 1;
968 wc
->w_large_pages
= 0;
970 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
978 * If a page has any new buffers, zero them out here, and mark them uptodate
979 * and dirty so they'll be written out (in order to prevent uninitialised
980 * block data from leaking). And clear the new bit.
982 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
984 unsigned int block_start
, block_end
;
985 struct buffer_head
*head
, *bh
;
987 BUG_ON(!PageLocked(page
));
988 if (!page_has_buffers(page
))
991 bh
= head
= page_buffers(page
);
994 block_end
= block_start
+ bh
->b_size
;
996 if (buffer_new(bh
)) {
997 if (block_end
> from
&& block_start
< to
) {
998 if (!PageUptodate(page
)) {
1001 start
= max(from
, block_start
);
1002 end
= min(to
, block_end
);
1004 zero_user_segment(page
, start
, end
);
1005 set_buffer_uptodate(bh
);
1008 clear_buffer_new(bh
);
1009 mark_buffer_dirty(bh
);
1013 block_start
= block_end
;
1014 bh
= bh
->b_this_page
;
1015 } while (bh
!= head
);
1019 * Only called when we have a failure during allocating write to write
1020 * zero's to the newly allocated region.
1022 static void ocfs2_write_failure(struct inode
*inode
,
1023 struct ocfs2_write_ctxt
*wc
,
1024 loff_t user_pos
, unsigned user_len
)
1027 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1028 to
= user_pos
+ user_len
;
1029 struct page
*tmppage
;
1031 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1033 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1034 tmppage
= wc
->w_pages
[i
];
1036 if (page_has_buffers(tmppage
)) {
1037 if (ocfs2_should_order_data(inode
))
1038 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1040 block_commit_write(tmppage
, from
, to
);
1045 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1046 struct ocfs2_write_ctxt
*wc
,
1047 struct page
*page
, u32 cpos
,
1048 loff_t user_pos
, unsigned user_len
,
1052 unsigned int map_from
= 0, map_to
= 0;
1053 unsigned int cluster_start
, cluster_end
;
1054 unsigned int user_data_from
= 0, user_data_to
= 0;
1056 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1057 &cluster_start
, &cluster_end
);
1059 /* treat the write as new if the a hole/lseek spanned across
1060 * the page boundary.
1062 new = new | ((i_size_read(inode
) <= page_offset(page
)) &&
1063 (page_offset(page
) <= user_pos
));
1065 if (page
== wc
->w_target_page
) {
1066 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1067 map_to
= map_from
+ user_len
;
1070 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1071 cluster_start
, cluster_end
,
1074 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1075 map_from
, map_to
, new);
1081 user_data_from
= map_from
;
1082 user_data_to
= map_to
;
1084 map_from
= cluster_start
;
1085 map_to
= cluster_end
;
1089 * If we haven't allocated the new page yet, we
1090 * shouldn't be writing it out without copying user
1091 * data. This is likely a math error from the caller.
1095 map_from
= cluster_start
;
1096 map_to
= cluster_end
;
1098 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1099 cluster_start
, cluster_end
, new);
1107 * Parts of newly allocated pages need to be zero'd.
1109 * Above, we have also rewritten 'to' and 'from' - as far as
1110 * the rest of the function is concerned, the entire cluster
1111 * range inside of a page needs to be written.
1113 * We can skip this if the page is up to date - it's already
1114 * been zero'd from being read in as a hole.
1116 if (new && !PageUptodate(page
))
1117 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1118 cpos
, user_data_from
, user_data_to
);
1120 flush_dcache_page(page
);
1127 * This function will only grab one clusters worth of pages.
1129 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1130 struct ocfs2_write_ctxt
*wc
,
1131 u32 cpos
, loff_t user_pos
,
1132 unsigned user_len
, int new,
1133 struct page
*mmap_page
)
1136 unsigned long start
, target_index
, end_index
, index
;
1137 struct inode
*inode
= mapping
->host
;
1140 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1143 * Figure out how many pages we'll be manipulating here. For
1144 * non allocating write, we just change the one
1145 * page. Otherwise, we'll need a whole clusters worth. If we're
1146 * writing past i_size, we only need enough pages to cover the
1147 * last page of the write.
1150 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1151 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1153 * We need the index *past* the last page we could possibly
1154 * touch. This is the page past the end of the write or
1155 * i_size, whichever is greater.
1157 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1158 BUG_ON(last_byte
< 1);
1159 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1160 if ((start
+ wc
->w_num_pages
) > end_index
)
1161 wc
->w_num_pages
= end_index
- start
;
1163 wc
->w_num_pages
= 1;
1164 start
= target_index
;
1167 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1170 if (index
== target_index
&& mmap_page
) {
1172 * ocfs2_pagemkwrite() is a little different
1173 * and wants us to directly use the page
1176 lock_page(mmap_page
);
1178 /* Exit and let the caller retry */
1179 if (mmap_page
->mapping
!= mapping
) {
1180 WARN_ON(mmap_page
->mapping
);
1181 unlock_page(mmap_page
);
1186 page_cache_get(mmap_page
);
1187 wc
->w_pages
[i
] = mmap_page
;
1188 wc
->w_target_locked
= true;
1190 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1192 if (!wc
->w_pages
[i
]) {
1198 wait_for_stable_page(wc
->w_pages
[i
]);
1200 if (index
== target_index
)
1201 wc
->w_target_page
= wc
->w_pages
[i
];
1205 wc
->w_target_locked
= false;
1210 * Prepare a single cluster for write one cluster into the file.
1212 static int ocfs2_write_cluster(struct address_space
*mapping
,
1213 u32 phys
, unsigned int unwritten
,
1214 unsigned int should_zero
,
1215 struct ocfs2_alloc_context
*data_ac
,
1216 struct ocfs2_alloc_context
*meta_ac
,
1217 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1218 loff_t user_pos
, unsigned user_len
)
1221 u64 v_blkno
, p_blkno
;
1222 struct inode
*inode
= mapping
->host
;
1223 struct ocfs2_extent_tree et
;
1225 new = phys
== 0 ? 1 : 0;
1230 * This is safe to call with the page locks - it won't take
1231 * any additional semaphores or cluster locks.
1234 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1235 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1236 wc
->w_handle
, data_ac
,
1239 * This shouldn't happen because we must have already
1240 * calculated the correct meta data allocation required. The
1241 * internal tree allocation code should know how to increase
1242 * transaction credits itself.
1244 * If need be, we could handle -EAGAIN for a
1245 * RESTART_TRANS here.
1247 mlog_bug_on_msg(ret
== -EAGAIN
,
1248 "Inode %llu: EAGAIN return during allocation.\n",
1249 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1254 } else if (unwritten
) {
1255 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1257 ret
= ocfs2_mark_extent_written(inode
, &et
,
1258 wc
->w_handle
, cpos
, 1, phys
,
1259 meta_ac
, &wc
->w_dealloc
);
1267 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1269 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1272 * The only reason this should fail is due to an inability to
1273 * find the extent added.
1275 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1278 ocfs2_error(inode
->i_sb
, "Corrupting extend for inode %llu, "
1279 "at logical block %llu",
1280 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1281 (unsigned long long)v_blkno
);
1285 BUG_ON(p_blkno
== 0);
1287 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1290 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1291 wc
->w_pages
[i
], cpos
,
1302 * We only have cleanup to do in case of allocating write.
1305 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1312 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1313 struct ocfs2_alloc_context
*data_ac
,
1314 struct ocfs2_alloc_context
*meta_ac
,
1315 struct ocfs2_write_ctxt
*wc
,
1316 loff_t pos
, unsigned len
)
1320 unsigned int local_len
= len
;
1321 struct ocfs2_write_cluster_desc
*desc
;
1322 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1324 for (i
= 0; i
< wc
->w_clen
; i
++) {
1325 desc
= &wc
->w_desc
[i
];
1328 * We have to make sure that the total write passed in
1329 * doesn't extend past a single cluster.
1332 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1333 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1334 local_len
= osb
->s_clustersize
- cluster_off
;
1336 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1340 wc
, desc
->c_cpos
, pos
, local_len
);
1356 * ocfs2_write_end() wants to know which parts of the target page it
1357 * should complete the write on. It's easiest to compute them ahead of
1358 * time when a more complete view of the write is available.
1360 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1361 struct ocfs2_write_ctxt
*wc
,
1362 loff_t pos
, unsigned len
, int alloc
)
1364 struct ocfs2_write_cluster_desc
*desc
;
1366 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1367 wc
->w_target_to
= wc
->w_target_from
+ len
;
1373 * Allocating write - we may have different boundaries based
1374 * on page size and cluster size.
1376 * NOTE: We can no longer compute one value from the other as
1377 * the actual write length and user provided length may be
1381 if (wc
->w_large_pages
) {
1383 * We only care about the 1st and last cluster within
1384 * our range and whether they should be zero'd or not. Either
1385 * value may be extended out to the start/end of a
1386 * newly allocated cluster.
1388 desc
= &wc
->w_desc
[0];
1389 if (desc
->c_needs_zero
)
1390 ocfs2_figure_cluster_boundaries(osb
,
1395 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1396 if (desc
->c_needs_zero
)
1397 ocfs2_figure_cluster_boundaries(osb
,
1402 wc
->w_target_from
= 0;
1403 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1408 * Populate each single-cluster write descriptor in the write context
1409 * with information about the i/o to be done.
1411 * Returns the number of clusters that will have to be allocated, as
1412 * well as a worst case estimate of the number of extent records that
1413 * would have to be created during a write to an unwritten region.
1415 static int ocfs2_populate_write_desc(struct inode
*inode
,
1416 struct ocfs2_write_ctxt
*wc
,
1417 unsigned int *clusters_to_alloc
,
1418 unsigned int *extents_to_split
)
1421 struct ocfs2_write_cluster_desc
*desc
;
1422 unsigned int num_clusters
= 0;
1423 unsigned int ext_flags
= 0;
1427 *clusters_to_alloc
= 0;
1428 *extents_to_split
= 0;
1430 for (i
= 0; i
< wc
->w_clen
; i
++) {
1431 desc
= &wc
->w_desc
[i
];
1432 desc
->c_cpos
= wc
->w_cpos
+ i
;
1434 if (num_clusters
== 0) {
1436 * Need to look up the next extent record.
1438 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1439 &num_clusters
, &ext_flags
);
1445 /* We should already CoW the refcountd extent. */
1446 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1449 * Assume worst case - that we're writing in
1450 * the middle of the extent.
1452 * We can assume that the write proceeds from
1453 * left to right, in which case the extent
1454 * insert code is smart enough to coalesce the
1455 * next splits into the previous records created.
1457 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1458 *extents_to_split
= *extents_to_split
+ 2;
1461 * Only increment phys if it doesn't describe
1468 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1469 * file that got extended. w_first_new_cpos tells us
1470 * where the newly allocated clusters are so we can
1473 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1475 desc
->c_needs_zero
= 1;
1478 desc
->c_phys
= phys
;
1481 desc
->c_needs_zero
= 1;
1482 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1485 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1486 desc
->c_unwritten
= 1;
1487 desc
->c_needs_zero
= 1;
1498 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1499 struct inode
*inode
,
1500 struct ocfs2_write_ctxt
*wc
)
1503 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1506 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1508 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1515 * If we don't set w_num_pages then this page won't get unlocked
1516 * and freed on cleanup of the write context.
1518 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1519 wc
->w_num_pages
= 1;
1521 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1522 if (IS_ERR(handle
)) {
1523 ret
= PTR_ERR(handle
);
1528 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1529 OCFS2_JOURNAL_ACCESS_WRITE
);
1531 ocfs2_commit_trans(osb
, handle
);
1537 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1538 ocfs2_set_inode_data_inline(inode
, di
);
1540 if (!PageUptodate(page
)) {
1541 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1543 ocfs2_commit_trans(osb
, handle
);
1549 wc
->w_handle
= handle
;
1554 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1556 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1558 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1563 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1564 struct inode
*inode
, loff_t pos
,
1565 unsigned len
, struct page
*mmap_page
,
1566 struct ocfs2_write_ctxt
*wc
)
1568 int ret
, written
= 0;
1569 loff_t end
= pos
+ len
;
1570 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1571 struct ocfs2_dinode
*di
= NULL
;
1573 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi
->ip_blkno
,
1574 len
, (unsigned long long)pos
,
1575 oi
->ip_dyn_features
);
1578 * Handle inodes which already have inline data 1st.
1580 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1581 if (mmap_page
== NULL
&&
1582 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1583 goto do_inline_write
;
1586 * The write won't fit - we have to give this inode an
1587 * inline extent list now.
1589 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1596 * Check whether the inode can accept inline data.
1598 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1602 * Check whether the write can fit.
1604 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1606 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1610 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1617 * This signals to the caller that the data can be written
1622 return written
? written
: ret
;
1626 * This function only does anything for file systems which can't
1627 * handle sparse files.
1629 * What we want to do here is fill in any hole between the current end
1630 * of allocation and the end of our write. That way the rest of the
1631 * write path can treat it as an non-allocating write, which has no
1632 * special case code for sparse/nonsparse files.
1634 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1635 struct buffer_head
*di_bh
,
1636 loff_t pos
, unsigned len
,
1637 struct ocfs2_write_ctxt
*wc
)
1640 loff_t newsize
= pos
+ len
;
1642 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1644 if (newsize
<= i_size_read(inode
))
1647 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1651 wc
->w_first_new_cpos
=
1652 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1657 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
1662 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1663 if (pos
> i_size_read(inode
))
1664 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
1670 * Try to flush truncate logs if we can free enough clusters from it.
1671 * As for return value, "< 0" means error, "0" no space and "1" means
1672 * we have freed enough spaces and let the caller try to allocate again.
1674 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
1675 unsigned int needed
)
1679 unsigned int truncated_clusters
;
1681 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
1682 truncated_clusters
= osb
->truncated_clusters
;
1683 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
1686 * Check whether we can succeed in allocating if we free
1689 if (truncated_clusters
< needed
)
1692 ret
= ocfs2_flush_truncate_log(osb
);
1698 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
1699 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
1706 int ocfs2_write_begin_nolock(struct file
*filp
,
1707 struct address_space
*mapping
,
1708 loff_t pos
, unsigned len
, unsigned flags
,
1709 struct page
**pagep
, void **fsdata
,
1710 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1712 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1713 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
1714 struct ocfs2_write_ctxt
*wc
;
1715 struct inode
*inode
= mapping
->host
;
1716 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1717 struct ocfs2_dinode
*di
;
1718 struct ocfs2_alloc_context
*data_ac
= NULL
;
1719 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1721 struct ocfs2_extent_tree et
;
1722 int try_free
= 1, ret1
;
1725 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1731 if (ocfs2_supports_inline_data(osb
)) {
1732 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1744 if (ocfs2_sparse_alloc(osb
))
1745 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
1747 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
1754 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1758 } else if (ret
== 1) {
1759 clusters_need
= wc
->w_clen
;
1760 ret
= ocfs2_refcount_cow(inode
, filp
, di_bh
,
1761 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1768 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1774 clusters_need
+= clusters_to_alloc
;
1776 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1778 trace_ocfs2_write_begin_nolock(
1779 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1780 (long long)i_size_read(inode
),
1781 le32_to_cpu(di
->i_clusters
),
1782 pos
, len
, flags
, mmap_page
,
1783 clusters_to_alloc
, extents_to_split
);
1786 * We set w_target_from, w_target_to here so that
1787 * ocfs2_write_end() knows which range in the target page to
1788 * write out. An allocation requires that we write the entire
1791 if (clusters_to_alloc
|| extents_to_split
) {
1793 * XXX: We are stretching the limits of
1794 * ocfs2_lock_allocators(). It greatly over-estimates
1795 * the work to be done.
1797 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1799 ret
= ocfs2_lock_allocators(inode
, &et
,
1800 clusters_to_alloc
, extents_to_split
,
1801 &data_ac
, &meta_ac
);
1808 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
1810 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
1817 * We have to zero sparse allocated clusters, unwritten extent clusters,
1818 * and non-sparse clusters we just extended. For non-sparse writes,
1819 * we know zeros will only be needed in the first and/or last cluster.
1821 if (clusters_to_alloc
|| extents_to_split
||
1822 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
1823 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
1824 cluster_of_pages
= 1;
1826 cluster_of_pages
= 0;
1828 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
1830 handle
= ocfs2_start_trans(osb
, credits
);
1831 if (IS_ERR(handle
)) {
1832 ret
= PTR_ERR(handle
);
1837 wc
->w_handle
= handle
;
1839 if (clusters_to_alloc
) {
1840 ret
= dquot_alloc_space_nodirty(inode
,
1841 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1846 * We don't want this to fail in ocfs2_write_end(), so do it
1849 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1850 OCFS2_JOURNAL_ACCESS_WRITE
);
1857 * Fill our page array first. That way we've grabbed enough so
1858 * that we can zero and flush if we error after adding the
1861 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
1862 cluster_of_pages
, mmap_page
);
1863 if (ret
&& ret
!= -EAGAIN
) {
1869 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
1870 * the target page. In this case, we exit with no error and no target
1871 * page. This will trigger the caller, page_mkwrite(), to re-try
1874 if (ret
== -EAGAIN
) {
1875 BUG_ON(wc
->w_target_page
);
1880 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
1888 ocfs2_free_alloc_context(data_ac
);
1890 ocfs2_free_alloc_context(meta_ac
);
1893 *pagep
= wc
->w_target_page
;
1897 if (clusters_to_alloc
)
1898 dquot_free_space(inode
,
1899 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
1901 ocfs2_commit_trans(osb
, handle
);
1904 ocfs2_free_write_ctxt(wc
);
1907 ocfs2_free_alloc_context(data_ac
);
1909 ocfs2_free_alloc_context(meta_ac
);
1911 if (ret
== -ENOSPC
&& try_free
) {
1913 * Try to free some truncate log so that we can have enough
1914 * clusters to allocate.
1918 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
1929 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
1930 loff_t pos
, unsigned len
, unsigned flags
,
1931 struct page
**pagep
, void **fsdata
)
1934 struct buffer_head
*di_bh
= NULL
;
1935 struct inode
*inode
= mapping
->host
;
1937 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1944 * Take alloc sem here to prevent concurrent lookups. That way
1945 * the mapping, zeroing and tree manipulation within
1946 * ocfs2_write() will be safe against ->readpage(). This
1947 * should also serve to lock out allocation from a shared
1950 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1952 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
1953 fsdata
, di_bh
, NULL
);
1964 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1967 ocfs2_inode_unlock(inode
, 1);
1972 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
1973 unsigned len
, unsigned *copied
,
1974 struct ocfs2_dinode
*di
,
1975 struct ocfs2_write_ctxt
*wc
)
1979 if (unlikely(*copied
< len
)) {
1980 if (!PageUptodate(wc
->w_target_page
)) {
1986 kaddr
= kmap_atomic(wc
->w_target_page
);
1987 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
1988 kunmap_atomic(kaddr
);
1990 trace_ocfs2_write_end_inline(
1991 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1992 (unsigned long long)pos
, *copied
,
1993 le16_to_cpu(di
->id2
.i_data
.id_count
),
1994 le16_to_cpu(di
->i_dyn_features
));
1997 int ocfs2_write_end_nolock(struct address_space
*mapping
,
1998 loff_t pos
, unsigned len
, unsigned copied
,
1999 struct page
*page
, void *fsdata
)
2002 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
2003 struct inode
*inode
= mapping
->host
;
2004 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2005 struct ocfs2_write_ctxt
*wc
= fsdata
;
2006 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
2007 handle_t
*handle
= wc
->w_handle
;
2008 struct page
*tmppage
;
2010 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2011 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
2012 goto out_write_size
;
2015 if (unlikely(copied
< len
)) {
2016 if (!PageUptodate(wc
->w_target_page
))
2019 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
2022 flush_dcache_page(wc
->w_target_page
);
2024 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
2025 tmppage
= wc
->w_pages
[i
];
2027 if (tmppage
== wc
->w_target_page
) {
2028 from
= wc
->w_target_from
;
2029 to
= wc
->w_target_to
;
2031 BUG_ON(from
> PAGE_CACHE_SIZE
||
2032 to
> PAGE_CACHE_SIZE
||
2036 * Pages adjacent to the target (if any) imply
2037 * a hole-filling write in which case we want
2038 * to flush their entire range.
2041 to
= PAGE_CACHE_SIZE
;
2044 if (page_has_buffers(tmppage
)) {
2045 if (ocfs2_should_order_data(inode
))
2046 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
2047 block_commit_write(tmppage
, from
, to
);
2053 if (pos
> inode
->i_size
) {
2054 i_size_write(inode
, pos
);
2055 mark_inode_dirty(inode
);
2057 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
2058 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
2059 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2060 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
2061 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2062 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2064 ocfs2_commit_trans(osb
, handle
);
2066 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2068 ocfs2_free_write_ctxt(wc
);
2073 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2074 loff_t pos
, unsigned len
, unsigned copied
,
2075 struct page
*page
, void *fsdata
)
2078 struct inode
*inode
= mapping
->host
;
2080 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2082 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2083 ocfs2_inode_unlock(inode
, 1);
2088 const struct address_space_operations ocfs2_aops
= {
2089 .readpage
= ocfs2_readpage
,
2090 .readpages
= ocfs2_readpages
,
2091 .writepage
= ocfs2_writepage
,
2092 .write_begin
= ocfs2_write_begin
,
2093 .write_end
= ocfs2_write_end
,
2095 .direct_IO
= ocfs2_direct_IO
,
2096 .invalidatepage
= ocfs2_invalidatepage
,
2097 .releasepage
= ocfs2_releasepage
,
2098 .migratepage
= buffer_migrate_page
,
2099 .is_partially_uptodate
= block_is_partially_uptodate
,
2100 .error_remove_page
= generic_error_remove_page
,