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>
31 #include <linux/blkdev.h>
33 #include <cluster/masklog.h>
40 #include "extent_map.h"
47 #include "refcounttree.h"
48 #include "ocfs2_trace.h"
50 #include "buffer_head_io.h"
55 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
56 struct buffer_head
*bh_result
, int create
)
60 struct ocfs2_dinode
*fe
= NULL
;
61 struct buffer_head
*bh
= NULL
;
62 struct buffer_head
*buffer_cache_bh
= NULL
;
63 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
66 trace_ocfs2_symlink_get_block(
67 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
68 (unsigned long long)iblock
, bh_result
, create
);
70 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
72 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
73 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
74 (unsigned long long)iblock
);
78 status
= ocfs2_read_inode_block(inode
, &bh
);
83 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
85 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
86 le32_to_cpu(fe
->i_clusters
))) {
88 mlog(ML_ERROR
, "block offset is outside the allocated size: "
89 "%llu\n", (unsigned long long)iblock
);
93 /* We don't use the page cache to create symlink data, so if
94 * need be, copy it over from the buffer cache. */
95 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
96 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
98 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
99 if (!buffer_cache_bh
) {
101 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
105 /* we haven't locked out transactions, so a commit
106 * could've happened. Since we've got a reference on
107 * the bh, even if it commits while we're doing the
108 * copy, the data is still good. */
109 if (buffer_jbd(buffer_cache_bh
)
110 && ocfs2_inode_is_new(inode
)) {
111 kaddr
= kmap_atomic(bh_result
->b_page
);
113 mlog(ML_ERROR
, "couldn't kmap!\n");
116 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
117 buffer_cache_bh
->b_data
,
119 kunmap_atomic(kaddr
);
120 set_buffer_uptodate(bh_result
);
122 brelse(buffer_cache_bh
);
125 map_bh(bh_result
, inode
->i_sb
,
126 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
136 int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
137 struct buffer_head
*bh_result
, int create
)
140 unsigned int ext_flags
;
141 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
142 u64 p_blkno
, count
, past_eof
;
143 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
145 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
146 (unsigned long long)iblock
, bh_result
, create
);
148 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
149 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
150 inode
, inode
->i_ino
);
152 if (S_ISLNK(inode
->i_mode
)) {
153 /* this always does I/O for some reason. */
154 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
158 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
161 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
162 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
163 (unsigned long long)p_blkno
);
167 if (max_blocks
< count
)
171 * ocfs2 never allocates in this function - the only time we
172 * need to use BH_New is when we're extending i_size on a file
173 * system which doesn't support holes, in which case BH_New
174 * allows __block_write_begin() to zero.
176 * If we see this on a sparse file system, then a truncate has
177 * raced us and removed the cluster. In this case, we clear
178 * the buffers dirty and uptodate bits and let the buffer code
179 * ignore it as a hole.
181 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
182 clear_buffer_dirty(bh_result
);
183 clear_buffer_uptodate(bh_result
);
187 /* Treat the unwritten extent as a hole for zeroing purposes. */
188 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
189 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
191 bh_result
->b_size
= count
<< inode
->i_blkbits
;
193 if (!ocfs2_sparse_alloc(osb
)) {
197 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
198 (unsigned long long)iblock
,
199 (unsigned long long)p_blkno
,
200 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
201 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
207 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
209 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
210 (unsigned long long)past_eof
);
211 if (create
&& (iblock
>= past_eof
))
212 set_buffer_new(bh_result
);
221 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
222 struct buffer_head
*di_bh
)
226 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
228 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
229 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
230 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
234 size
= i_size_read(inode
);
236 if (size
> PAGE_CACHE_SIZE
||
237 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
238 ocfs2_error(inode
->i_sb
,
239 "Inode %llu has with inline data has bad size: %Lu",
240 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
241 (unsigned long long)size
);
245 kaddr
= kmap_atomic(page
);
247 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
248 /* Clear the remaining part of the page */
249 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
250 flush_dcache_page(page
);
251 kunmap_atomic(kaddr
);
253 SetPageUptodate(page
);
258 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
261 struct buffer_head
*di_bh
= NULL
;
263 BUG_ON(!PageLocked(page
));
264 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
266 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
272 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
280 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
282 struct inode
*inode
= page
->mapping
->host
;
283 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
284 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
287 trace_ocfs2_readpage((unsigned long long)oi
->ip_blkno
,
288 (page
? page
->index
: 0));
290 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
292 if (ret
== AOP_TRUNCATED_PAGE
)
298 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
300 * Unlock the page and cycle ip_alloc_sem so that we don't
301 * busyloop waiting for ip_alloc_sem to unlock
303 ret
= AOP_TRUNCATED_PAGE
;
306 down_read(&oi
->ip_alloc_sem
);
307 up_read(&oi
->ip_alloc_sem
);
308 goto out_inode_unlock
;
312 * i_size might have just been updated as we grabed the meta lock. We
313 * might now be discovering a truncate that hit on another node.
314 * block_read_full_page->get_block freaks out if it is asked to read
315 * beyond the end of a file, so we check here. Callers
316 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
317 * and notice that the page they just read isn't needed.
319 * XXX sys_readahead() seems to get that wrong?
321 if (start
>= i_size_read(inode
)) {
322 zero_user(page
, 0, PAGE_SIZE
);
323 SetPageUptodate(page
);
328 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
329 ret
= ocfs2_readpage_inline(inode
, page
);
331 ret
= block_read_full_page(page
, ocfs2_get_block
);
335 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
337 ocfs2_inode_unlock(inode
, 0);
345 * This is used only for read-ahead. Failures or difficult to handle
346 * situations are safe to ignore.
348 * Right now, we don't bother with BH_Boundary - in-inode extent lists
349 * are quite large (243 extents on 4k blocks), so most inodes don't
350 * grow out to a tree. If need be, detecting boundary extents could
351 * trivially be added in a future version of ocfs2_get_block().
353 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
354 struct list_head
*pages
, unsigned nr_pages
)
357 struct inode
*inode
= mapping
->host
;
358 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
363 * Use the nonblocking flag for the dlm code to avoid page
364 * lock inversion, but don't bother with retrying.
366 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
370 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
371 ocfs2_inode_unlock(inode
, 0);
376 * Don't bother with inline-data. There isn't anything
377 * to read-ahead in that case anyway...
379 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
383 * Check whether a remote node truncated this file - we just
384 * drop out in that case as it's not worth handling here.
386 last
= list_entry(pages
->prev
, struct page
, lru
);
387 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
388 if (start
>= i_size_read(inode
))
391 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
394 up_read(&oi
->ip_alloc_sem
);
395 ocfs2_inode_unlock(inode
, 0);
400 /* Note: Because we don't support holes, our allocation has
401 * already happened (allocation writes zeros to the file data)
402 * so we don't have to worry about ordered writes in
405 * ->writepage is called during the process of invalidating the page cache
406 * during blocked lock processing. It can't block on any cluster locks
407 * to during block mapping. It's relying on the fact that the block
408 * mapping can't have disappeared under the dirty pages that it is
409 * being asked to write back.
411 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
413 trace_ocfs2_writepage(
414 (unsigned long long)OCFS2_I(page
->mapping
->host
)->ip_blkno
,
417 return block_write_full_page(page
, ocfs2_get_block
, wbc
);
420 /* Taken from ext3. We don't necessarily need the full blown
421 * functionality yet, but IMHO it's better to cut and paste the whole
422 * thing so we can avoid introducing our own bugs (and easily pick up
423 * their fixes when they happen) --Mark */
424 int walk_page_buffers( handle_t
*handle
,
425 struct buffer_head
*head
,
429 int (*fn
)( handle_t
*handle
,
430 struct buffer_head
*bh
))
432 struct buffer_head
*bh
;
433 unsigned block_start
, block_end
;
434 unsigned blocksize
= head
->b_size
;
436 struct buffer_head
*next
;
438 for ( bh
= head
, block_start
= 0;
439 ret
== 0 && (bh
!= head
|| !block_start
);
440 block_start
= block_end
, bh
= next
)
442 next
= bh
->b_this_page
;
443 block_end
= block_start
+ blocksize
;
444 if (block_end
<= from
|| block_start
>= to
) {
445 if (partial
&& !buffer_uptodate(bh
))
449 err
= (*fn
)(handle
, bh
);
456 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
461 struct inode
*inode
= mapping
->host
;
463 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
464 (unsigned long long)block
);
466 /* We don't need to lock journal system files, since they aren't
467 * accessed concurrently from multiple nodes.
469 if (!INODE_JOURNAL(inode
)) {
470 err
= ocfs2_inode_lock(inode
, NULL
, 0);
476 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
479 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
480 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
483 if (!INODE_JOURNAL(inode
)) {
484 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
485 ocfs2_inode_unlock(inode
, 0);
489 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
490 (unsigned long long)block
);
496 status
= err
? 0 : p_blkno
;
502 * TODO: Make this into a generic get_blocks function.
504 * From do_direct_io in direct-io.c:
505 * "So what we do is to permit the ->get_blocks function to populate
506 * bh.b_size with the size of IO which is permitted at this offset and
509 * This function is called directly from get_more_blocks in direct-io.c.
511 * called like this: dio->get_blocks(dio->inode, fs_startblk,
512 * fs_count, map_bh, dio->rw == WRITE);
514 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
515 struct buffer_head
*bh_result
, int create
)
519 int alloc_locked
= 0;
520 u64 p_blkno
, inode_blocks
, contig_blocks
;
521 unsigned int ext_flags
;
522 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
523 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
524 unsigned long len
= bh_result
->b_size
;
525 unsigned int clusters_to_alloc
= 0;
527 cpos
= ocfs2_blocks_to_clusters(inode
->i_sb
, iblock
);
529 /* This function won't even be called if the request isn't all
530 * nicely aligned and of the right size, so there's no need
531 * for us to check any of that. */
533 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
535 /* This figures out the size of the next contiguous block, and
536 * our logical offset */
537 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
538 &contig_blocks
, &ext_flags
);
540 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
541 (unsigned long long)iblock
);
546 /* We should already CoW the refcounted extent in case of create. */
547 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
549 /* allocate blocks if no p_blkno is found, and create == 1 */
550 if (!p_blkno
&& create
) {
551 ret
= ocfs2_inode_lock(inode
, NULL
, 1);
559 /* fill hole, allocate blocks can't be larger than the size
561 clusters_to_alloc
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
562 if (clusters_to_alloc
> contig_blocks
)
563 clusters_to_alloc
= contig_blocks
;
565 /* allocate extent and insert them into the extent tree */
566 ret
= ocfs2_extend_allocation(inode
, cpos
,
567 clusters_to_alloc
, 0);
573 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
574 &contig_blocks
, &ext_flags
);
576 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
577 (unsigned long long)iblock
);
584 * get_more_blocks() expects us to describe a hole by clearing
585 * the mapped bit on bh_result().
587 * Consider an unwritten extent as a hole.
589 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
590 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
592 clear_buffer_mapped(bh_result
);
594 /* make sure we don't map more than max_blocks blocks here as
595 that's all the kernel will handle at this point. */
596 if (max_blocks
< contig_blocks
)
597 contig_blocks
= max_blocks
;
598 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
601 ocfs2_inode_unlock(inode
, 1);
606 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
607 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
608 * to protect io on one node from truncation on another.
610 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
615 struct inode
*inode
= file_inode(iocb
->ki_filp
);
618 /* this io's submitter should not have unlocked this before we could */
619 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
621 if (ocfs2_iocb_is_sem_locked(iocb
))
622 ocfs2_iocb_clear_sem_locked(iocb
);
624 if (ocfs2_iocb_is_unaligned_aio(iocb
)) {
625 ocfs2_iocb_clear_unaligned_aio(iocb
);
627 mutex_unlock(&OCFS2_I(inode
)->ip_unaligned_aio
);
630 ocfs2_iocb_clear_rw_locked(iocb
);
632 level
= ocfs2_iocb_rw_locked_level(iocb
);
633 ocfs2_rw_unlock(inode
, level
);
636 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
638 if (!page_has_buffers(page
))
640 return try_to_free_buffers(page
);
643 static int ocfs2_is_overwrite(struct ocfs2_super
*osb
,
644 struct inode
*inode
, loff_t offset
)
649 unsigned int num_clusters
= 0;
650 unsigned int ext_flags
= 0;
652 v_cpos
= ocfs2_bytes_to_clusters(osb
->sb
, offset
);
653 ret
= ocfs2_get_clusters(inode
, v_cpos
, &p_cpos
,
654 &num_clusters
, &ext_flags
);
660 if (p_cpos
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
666 static ssize_t
ocfs2_direct_IO_write(struct kiocb
*iocb
,
667 struct iov_iter
*iter
,
672 bool orphaned
= false;
673 int is_overwrite
= 0;
674 struct file
*file
= iocb
->ki_filp
;
675 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
676 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
677 struct buffer_head
*di_bh
= NULL
;
678 size_t count
= iter
->count
;
679 journal_t
*journal
= osb
->journal
->j_journal
;
682 loff_t final_size
= offset
+ count
;
683 int append_write
= offset
>= i_size_read(inode
) ? 1 : 0;
684 unsigned int num_clusters
= 0;
685 unsigned int ext_flags
= 0;
690 zero_len
= do_div(o
, 1 << osb
->s_clustersize_bits
);
691 cluster_align
= !zero_len
;
695 * when final_size > inode->i_size, inode->i_size will be
696 * updated after direct write, so add the inode to orphan
699 if (final_size
> i_size_read(inode
)) {
700 ret
= ocfs2_add_inode_to_orphan(osb
, inode
);
709 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
715 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
716 ret
= ocfs2_zero_extend(inode
, di_bh
, offset
);
718 ret
= ocfs2_extend_no_holes(inode
, di_bh
, offset
,
722 ocfs2_inode_unlock(inode
, 1);
727 is_overwrite
= ocfs2_is_overwrite(osb
, inode
, offset
);
728 if (is_overwrite
< 0) {
729 mlog_errno(is_overwrite
);
730 ocfs2_inode_unlock(inode
, 1);
735 ocfs2_inode_unlock(inode
, 1);
740 written
= __blockdev_direct_IO(WRITE
, iocb
, inode
, inode
->i_sb
->s_bdev
,
742 ocfs2_direct_IO_get_blocks
,
743 ocfs2_dio_end_io
, NULL
, 0);
744 if (unlikely(written
< 0)) {
745 loff_t i_size
= i_size_read(inode
);
747 if (offset
+ count
> i_size
) {
748 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
754 if (i_size
== i_size_read(inode
)) {
755 ret
= ocfs2_truncate_file(inode
, di_bh
,
761 ocfs2_inode_unlock(inode
, 1);
767 ocfs2_inode_unlock(inode
, 1);
770 ret
= jbd2_journal_force_commit(journal
);
774 } else if (written
< 0 && append_write
&& !is_overwrite
&&
777 u32 v_cpos
= ocfs2_bytes_to_clusters(osb
->sb
, offset
);
779 ret
= ocfs2_get_clusters(inode
, v_cpos
, &p_cpos
,
780 &num_clusters
, &ext_flags
);
786 BUG_ON(!p_cpos
|| (ext_flags
& OCFS2_EXT_UNWRITTEN
));
788 ret
= blkdev_issue_zeroout(osb
->sb
->s_bdev
,
789 p_cpos
<< (osb
->s_clustersize_bits
- 9),
790 zero_len
>> 9, GFP_KERNEL
, false);
798 int update_isize
= written
> 0 ? 1 : 0;
799 loff_t end
= update_isize
? offset
+ written
: 0;
801 tmp_ret
= ocfs2_del_inode_from_orphan(osb
, inode
,
808 tmp_ret
= jbd2_journal_force_commit(journal
);
821 static ssize_t
ocfs2_direct_IO(int rw
,
823 struct iov_iter
*iter
,
826 struct file
*file
= iocb
->ki_filp
;
827 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
828 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
829 int full_coherency
= !(osb
->s_mount_opt
&
830 OCFS2_MOUNT_COHERENCY_BUFFERED
);
833 * Fallback to buffered I/O if we see an inode without
836 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
839 /* Fallback to buffered I/O if we are appending and
840 * concurrent O_DIRECT writes are allowed.
842 if (i_size_read(inode
) <= offset
&& !full_coherency
)
846 return __blockdev_direct_IO(rw
, iocb
, inode
,
849 ocfs2_direct_IO_get_blocks
,
850 ocfs2_dio_end_io
, NULL
, 0);
852 return ocfs2_direct_IO_write(iocb
, iter
, offset
);
855 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
860 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
862 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
865 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
867 cluster_start
= cpos
% cpp
;
868 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
870 cluster_end
= cluster_start
+ osb
->s_clustersize
;
873 BUG_ON(cluster_start
> PAGE_SIZE
);
874 BUG_ON(cluster_end
> PAGE_SIZE
);
877 *start
= cluster_start
;
883 * 'from' and 'to' are the region in the page to avoid zeroing.
885 * If pagesize > clustersize, this function will avoid zeroing outside
886 * of the cluster boundary.
888 * from == to == 0 is code for "zero the entire cluster region"
890 static void ocfs2_clear_page_regions(struct page
*page
,
891 struct ocfs2_super
*osb
, u32 cpos
,
892 unsigned from
, unsigned to
)
895 unsigned int cluster_start
, cluster_end
;
897 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
899 kaddr
= kmap_atomic(page
);
902 if (from
> cluster_start
)
903 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
904 if (to
< cluster_end
)
905 memset(kaddr
+ to
, 0, cluster_end
- to
);
907 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
910 kunmap_atomic(kaddr
);
914 * Nonsparse file systems fully allocate before we get to the write
915 * code. This prevents ocfs2_write() from tagging the write as an
916 * allocating one, which means ocfs2_map_page_blocks() might try to
917 * read-in the blocks at the tail of our file. Avoid reading them by
918 * testing i_size against each block offset.
920 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
921 unsigned int block_start
)
923 u64 offset
= page_offset(page
) + block_start
;
925 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
928 if (i_size_read(inode
) > offset
)
935 * Some of this taken from __block_write_begin(). We already have our
936 * mapping by now though, and the entire write will be allocating or
937 * it won't, so not much need to use BH_New.
939 * This will also skip zeroing, which is handled externally.
941 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
942 struct inode
*inode
, unsigned int from
,
943 unsigned int to
, int new)
946 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
947 unsigned int block_end
, block_start
;
948 unsigned int bsize
= 1 << inode
->i_blkbits
;
950 if (!page_has_buffers(page
))
951 create_empty_buffers(page
, bsize
, 0);
953 head
= page_buffers(page
);
954 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
955 bh
= bh
->b_this_page
, block_start
+= bsize
) {
956 block_end
= block_start
+ bsize
;
958 clear_buffer_new(bh
);
961 * Ignore blocks outside of our i/o range -
962 * they may belong to unallocated clusters.
964 if (block_start
>= to
|| block_end
<= from
) {
965 if (PageUptodate(page
))
966 set_buffer_uptodate(bh
);
971 * For an allocating write with cluster size >= page
972 * size, we always write the entire page.
977 if (!buffer_mapped(bh
)) {
978 map_bh(bh
, inode
->i_sb
, *p_blkno
);
979 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
982 if (PageUptodate(page
)) {
983 if (!buffer_uptodate(bh
))
984 set_buffer_uptodate(bh
);
985 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
987 ocfs2_should_read_blk(inode
, page
, block_start
) &&
988 (block_start
< from
|| block_end
> to
)) {
989 ll_rw_block(READ
, 1, &bh
);
993 *p_blkno
= *p_blkno
+ 1;
997 * If we issued read requests - let them complete.
999 while(wait_bh
> wait
) {
1000 wait_on_buffer(*--wait_bh
);
1001 if (!buffer_uptodate(*wait_bh
))
1005 if (ret
== 0 || !new)
1009 * If we get -EIO above, zero out any newly allocated blocks
1010 * to avoid exposing stale data.
1015 block_end
= block_start
+ bsize
;
1016 if (block_end
<= from
)
1018 if (block_start
>= to
)
1021 zero_user(page
, block_start
, bh
->b_size
);
1022 set_buffer_uptodate(bh
);
1023 mark_buffer_dirty(bh
);
1026 block_start
= block_end
;
1027 bh
= bh
->b_this_page
;
1028 } while (bh
!= head
);
1033 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
1034 #define OCFS2_MAX_CTXT_PAGES 1
1036 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
1039 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
1042 * Describe the state of a single cluster to be written to.
1044 struct ocfs2_write_cluster_desc
{
1048 * Give this a unique field because c_phys eventually gets
1052 unsigned c_unwritten
;
1053 unsigned c_needs_zero
;
1056 struct ocfs2_write_ctxt
{
1057 /* Logical cluster position / len of write */
1061 /* First cluster allocated in a nonsparse extend */
1062 u32 w_first_new_cpos
;
1064 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
1067 * This is true if page_size > cluster_size.
1069 * It triggers a set of special cases during write which might
1070 * have to deal with allocating writes to partial pages.
1072 unsigned int w_large_pages
;
1075 * Pages involved in this write.
1077 * w_target_page is the page being written to by the user.
1079 * w_pages is an array of pages which always contains
1080 * w_target_page, and in the case of an allocating write with
1081 * page_size < cluster size, it will contain zero'd and mapped
1082 * pages adjacent to w_target_page which need to be written
1083 * out in so that future reads from that region will get
1086 unsigned int w_num_pages
;
1087 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
1088 struct page
*w_target_page
;
1091 * w_target_locked is used for page_mkwrite path indicating no unlocking
1092 * against w_target_page in ocfs2_write_end_nolock.
1094 unsigned int w_target_locked
:1;
1097 * ocfs2_write_end() uses this to know what the real range to
1098 * write in the target should be.
1100 unsigned int w_target_from
;
1101 unsigned int w_target_to
;
1104 * We could use journal_current_handle() but this is cleaner,
1109 struct buffer_head
*w_di_bh
;
1111 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
1114 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
1118 for(i
= 0; i
< num_pages
; i
++) {
1120 unlock_page(pages
[i
]);
1121 mark_page_accessed(pages
[i
]);
1122 page_cache_release(pages
[i
]);
1127 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt
*wc
)
1132 * w_target_locked is only set to true in the page_mkwrite() case.
1133 * The intent is to allow us to lock the target page from write_begin()
1134 * to write_end(). The caller must hold a ref on w_target_page.
1136 if (wc
->w_target_locked
) {
1137 BUG_ON(!wc
->w_target_page
);
1138 for (i
= 0; i
< wc
->w_num_pages
; i
++) {
1139 if (wc
->w_target_page
== wc
->w_pages
[i
]) {
1140 wc
->w_pages
[i
] = NULL
;
1144 mark_page_accessed(wc
->w_target_page
);
1145 page_cache_release(wc
->w_target_page
);
1147 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
1150 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
1152 ocfs2_unlock_pages(wc
);
1153 brelse(wc
->w_di_bh
);
1157 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
1158 struct ocfs2_super
*osb
, loff_t pos
,
1159 unsigned len
, struct buffer_head
*di_bh
)
1162 struct ocfs2_write_ctxt
*wc
;
1164 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
1168 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
1169 wc
->w_first_new_cpos
= UINT_MAX
;
1170 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
1171 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
1173 wc
->w_di_bh
= di_bh
;
1175 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
1176 wc
->w_large_pages
= 1;
1178 wc
->w_large_pages
= 0;
1180 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
1188 * If a page has any new buffers, zero them out here, and mark them uptodate
1189 * and dirty so they'll be written out (in order to prevent uninitialised
1190 * block data from leaking). And clear the new bit.
1192 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1194 unsigned int block_start
, block_end
;
1195 struct buffer_head
*head
, *bh
;
1197 BUG_ON(!PageLocked(page
));
1198 if (!page_has_buffers(page
))
1201 bh
= head
= page_buffers(page
);
1204 block_end
= block_start
+ bh
->b_size
;
1206 if (buffer_new(bh
)) {
1207 if (block_end
> from
&& block_start
< to
) {
1208 if (!PageUptodate(page
)) {
1209 unsigned start
, end
;
1211 start
= max(from
, block_start
);
1212 end
= min(to
, block_end
);
1214 zero_user_segment(page
, start
, end
);
1215 set_buffer_uptodate(bh
);
1218 clear_buffer_new(bh
);
1219 mark_buffer_dirty(bh
);
1223 block_start
= block_end
;
1224 bh
= bh
->b_this_page
;
1225 } while (bh
!= head
);
1229 * Only called when we have a failure during allocating write to write
1230 * zero's to the newly allocated region.
1232 static void ocfs2_write_failure(struct inode
*inode
,
1233 struct ocfs2_write_ctxt
*wc
,
1234 loff_t user_pos
, unsigned user_len
)
1237 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1238 to
= user_pos
+ user_len
;
1239 struct page
*tmppage
;
1241 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1243 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1244 tmppage
= wc
->w_pages
[i
];
1246 if (page_has_buffers(tmppage
)) {
1247 if (ocfs2_should_order_data(inode
))
1248 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1250 block_commit_write(tmppage
, from
, to
);
1255 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1256 struct ocfs2_write_ctxt
*wc
,
1257 struct page
*page
, u32 cpos
,
1258 loff_t user_pos
, unsigned user_len
,
1262 unsigned int map_from
= 0, map_to
= 0;
1263 unsigned int cluster_start
, cluster_end
;
1264 unsigned int user_data_from
= 0, user_data_to
= 0;
1266 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1267 &cluster_start
, &cluster_end
);
1269 /* treat the write as new if the a hole/lseek spanned across
1270 * the page boundary.
1272 new = new | ((i_size_read(inode
) <= page_offset(page
)) &&
1273 (page_offset(page
) <= user_pos
));
1275 if (page
== wc
->w_target_page
) {
1276 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1277 map_to
= map_from
+ user_len
;
1280 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1281 cluster_start
, cluster_end
,
1284 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1285 map_from
, map_to
, new);
1291 user_data_from
= map_from
;
1292 user_data_to
= map_to
;
1294 map_from
= cluster_start
;
1295 map_to
= cluster_end
;
1299 * If we haven't allocated the new page yet, we
1300 * shouldn't be writing it out without copying user
1301 * data. This is likely a math error from the caller.
1305 map_from
= cluster_start
;
1306 map_to
= cluster_end
;
1308 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1309 cluster_start
, cluster_end
, new);
1317 * Parts of newly allocated pages need to be zero'd.
1319 * Above, we have also rewritten 'to' and 'from' - as far as
1320 * the rest of the function is concerned, the entire cluster
1321 * range inside of a page needs to be written.
1323 * We can skip this if the page is up to date - it's already
1324 * been zero'd from being read in as a hole.
1326 if (new && !PageUptodate(page
))
1327 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1328 cpos
, user_data_from
, user_data_to
);
1330 flush_dcache_page(page
);
1337 * This function will only grab one clusters worth of pages.
1339 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1340 struct ocfs2_write_ctxt
*wc
,
1341 u32 cpos
, loff_t user_pos
,
1342 unsigned user_len
, int new,
1343 struct page
*mmap_page
)
1346 unsigned long start
, target_index
, end_index
, index
;
1347 struct inode
*inode
= mapping
->host
;
1350 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1353 * Figure out how many pages we'll be manipulating here. For
1354 * non allocating write, we just change the one
1355 * page. Otherwise, we'll need a whole clusters worth. If we're
1356 * writing past i_size, we only need enough pages to cover the
1357 * last page of the write.
1360 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1361 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1363 * We need the index *past* the last page we could possibly
1364 * touch. This is the page past the end of the write or
1365 * i_size, whichever is greater.
1367 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1368 BUG_ON(last_byte
< 1);
1369 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1370 if ((start
+ wc
->w_num_pages
) > end_index
)
1371 wc
->w_num_pages
= end_index
- start
;
1373 wc
->w_num_pages
= 1;
1374 start
= target_index
;
1377 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1380 if (index
== target_index
&& mmap_page
) {
1382 * ocfs2_pagemkwrite() is a little different
1383 * and wants us to directly use the page
1386 lock_page(mmap_page
);
1388 /* Exit and let the caller retry */
1389 if (mmap_page
->mapping
!= mapping
) {
1390 WARN_ON(mmap_page
->mapping
);
1391 unlock_page(mmap_page
);
1396 page_cache_get(mmap_page
);
1397 wc
->w_pages
[i
] = mmap_page
;
1398 wc
->w_target_locked
= true;
1400 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1402 if (!wc
->w_pages
[i
]) {
1408 wait_for_stable_page(wc
->w_pages
[i
]);
1410 if (index
== target_index
)
1411 wc
->w_target_page
= wc
->w_pages
[i
];
1415 wc
->w_target_locked
= false;
1420 * Prepare a single cluster for write one cluster into the file.
1422 static int ocfs2_write_cluster(struct address_space
*mapping
,
1423 u32 phys
, unsigned int unwritten
,
1424 unsigned int should_zero
,
1425 struct ocfs2_alloc_context
*data_ac
,
1426 struct ocfs2_alloc_context
*meta_ac
,
1427 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1428 loff_t user_pos
, unsigned user_len
)
1431 u64 v_blkno
, p_blkno
;
1432 struct inode
*inode
= mapping
->host
;
1433 struct ocfs2_extent_tree et
;
1435 new = phys
== 0 ? 1 : 0;
1440 * This is safe to call with the page locks - it won't take
1441 * any additional semaphores or cluster locks.
1444 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1445 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1446 wc
->w_handle
, data_ac
,
1449 * This shouldn't happen because we must have already
1450 * calculated the correct meta data allocation required. The
1451 * internal tree allocation code should know how to increase
1452 * transaction credits itself.
1454 * If need be, we could handle -EAGAIN for a
1455 * RESTART_TRANS here.
1457 mlog_bug_on_msg(ret
== -EAGAIN
,
1458 "Inode %llu: EAGAIN return during allocation.\n",
1459 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1464 } else if (unwritten
) {
1465 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1467 ret
= ocfs2_mark_extent_written(inode
, &et
,
1468 wc
->w_handle
, cpos
, 1, phys
,
1469 meta_ac
, &wc
->w_dealloc
);
1477 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1479 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1482 * The only reason this should fail is due to an inability to
1483 * find the extent added.
1485 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1488 mlog(ML_ERROR
, "Get physical blkno failed for inode %llu, "
1489 "at logical block %llu",
1490 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1491 (unsigned long long)v_blkno
);
1495 BUG_ON(p_blkno
== 0);
1497 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1500 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1501 wc
->w_pages
[i
], cpos
,
1512 * We only have cleanup to do in case of allocating write.
1515 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1522 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1523 struct ocfs2_alloc_context
*data_ac
,
1524 struct ocfs2_alloc_context
*meta_ac
,
1525 struct ocfs2_write_ctxt
*wc
,
1526 loff_t pos
, unsigned len
)
1530 unsigned int local_len
= len
;
1531 struct ocfs2_write_cluster_desc
*desc
;
1532 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1534 for (i
= 0; i
< wc
->w_clen
; i
++) {
1535 desc
= &wc
->w_desc
[i
];
1538 * We have to make sure that the total write passed in
1539 * doesn't extend past a single cluster.
1542 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1543 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1544 local_len
= osb
->s_clustersize
- cluster_off
;
1546 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1550 wc
, desc
->c_cpos
, pos
, local_len
);
1566 * ocfs2_write_end() wants to know which parts of the target page it
1567 * should complete the write on. It's easiest to compute them ahead of
1568 * time when a more complete view of the write is available.
1570 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1571 struct ocfs2_write_ctxt
*wc
,
1572 loff_t pos
, unsigned len
, int alloc
)
1574 struct ocfs2_write_cluster_desc
*desc
;
1576 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1577 wc
->w_target_to
= wc
->w_target_from
+ len
;
1583 * Allocating write - we may have different boundaries based
1584 * on page size and cluster size.
1586 * NOTE: We can no longer compute one value from the other as
1587 * the actual write length and user provided length may be
1591 if (wc
->w_large_pages
) {
1593 * We only care about the 1st and last cluster within
1594 * our range and whether they should be zero'd or not. Either
1595 * value may be extended out to the start/end of a
1596 * newly allocated cluster.
1598 desc
= &wc
->w_desc
[0];
1599 if (desc
->c_needs_zero
)
1600 ocfs2_figure_cluster_boundaries(osb
,
1605 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1606 if (desc
->c_needs_zero
)
1607 ocfs2_figure_cluster_boundaries(osb
,
1612 wc
->w_target_from
= 0;
1613 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1618 * Populate each single-cluster write descriptor in the write context
1619 * with information about the i/o to be done.
1621 * Returns the number of clusters that will have to be allocated, as
1622 * well as a worst case estimate of the number of extent records that
1623 * would have to be created during a write to an unwritten region.
1625 static int ocfs2_populate_write_desc(struct inode
*inode
,
1626 struct ocfs2_write_ctxt
*wc
,
1627 unsigned int *clusters_to_alloc
,
1628 unsigned int *extents_to_split
)
1631 struct ocfs2_write_cluster_desc
*desc
;
1632 unsigned int num_clusters
= 0;
1633 unsigned int ext_flags
= 0;
1637 *clusters_to_alloc
= 0;
1638 *extents_to_split
= 0;
1640 for (i
= 0; i
< wc
->w_clen
; i
++) {
1641 desc
= &wc
->w_desc
[i
];
1642 desc
->c_cpos
= wc
->w_cpos
+ i
;
1644 if (num_clusters
== 0) {
1646 * Need to look up the next extent record.
1648 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1649 &num_clusters
, &ext_flags
);
1655 /* We should already CoW the refcountd extent. */
1656 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1659 * Assume worst case - that we're writing in
1660 * the middle of the extent.
1662 * We can assume that the write proceeds from
1663 * left to right, in which case the extent
1664 * insert code is smart enough to coalesce the
1665 * next splits into the previous records created.
1667 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1668 *extents_to_split
= *extents_to_split
+ 2;
1671 * Only increment phys if it doesn't describe
1678 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1679 * file that got extended. w_first_new_cpos tells us
1680 * where the newly allocated clusters are so we can
1683 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1685 desc
->c_needs_zero
= 1;
1688 desc
->c_phys
= phys
;
1691 desc
->c_needs_zero
= 1;
1692 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1695 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1696 desc
->c_unwritten
= 1;
1697 desc
->c_needs_zero
= 1;
1708 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1709 struct inode
*inode
,
1710 struct ocfs2_write_ctxt
*wc
)
1713 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1716 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1718 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1719 if (IS_ERR(handle
)) {
1720 ret
= PTR_ERR(handle
);
1725 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1727 ocfs2_commit_trans(osb
, handle
);
1733 * If we don't set w_num_pages then this page won't get unlocked
1734 * and freed on cleanup of the write context.
1736 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1737 wc
->w_num_pages
= 1;
1739 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1740 OCFS2_JOURNAL_ACCESS_WRITE
);
1742 ocfs2_commit_trans(osb
, handle
);
1748 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1749 ocfs2_set_inode_data_inline(inode
, di
);
1751 if (!PageUptodate(page
)) {
1752 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1754 ocfs2_commit_trans(osb
, handle
);
1760 wc
->w_handle
= handle
;
1765 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1767 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1769 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1774 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1775 struct inode
*inode
, loff_t pos
,
1776 unsigned len
, struct page
*mmap_page
,
1777 struct ocfs2_write_ctxt
*wc
)
1779 int ret
, written
= 0;
1780 loff_t end
= pos
+ len
;
1781 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1782 struct ocfs2_dinode
*di
= NULL
;
1784 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi
->ip_blkno
,
1785 len
, (unsigned long long)pos
,
1786 oi
->ip_dyn_features
);
1789 * Handle inodes which already have inline data 1st.
1791 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1792 if (mmap_page
== NULL
&&
1793 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1794 goto do_inline_write
;
1797 * The write won't fit - we have to give this inode an
1798 * inline extent list now.
1800 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1807 * Check whether the inode can accept inline data.
1809 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1813 * Check whether the write can fit.
1815 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1817 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1821 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1828 * This signals to the caller that the data can be written
1833 return written
? written
: ret
;
1837 * This function only does anything for file systems which can't
1838 * handle sparse files.
1840 * What we want to do here is fill in any hole between the current end
1841 * of allocation and the end of our write. That way the rest of the
1842 * write path can treat it as an non-allocating write, which has no
1843 * special case code for sparse/nonsparse files.
1845 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1846 struct buffer_head
*di_bh
,
1847 loff_t pos
, unsigned len
,
1848 struct ocfs2_write_ctxt
*wc
)
1851 loff_t newsize
= pos
+ len
;
1853 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1855 if (newsize
<= i_size_read(inode
))
1858 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1862 wc
->w_first_new_cpos
=
1863 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1868 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
1873 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1874 if (pos
> i_size_read(inode
))
1875 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
1881 * Try to flush truncate logs if we can free enough clusters from it.
1882 * As for return value, "< 0" means error, "0" no space and "1" means
1883 * we have freed enough spaces and let the caller try to allocate again.
1885 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
1886 unsigned int needed
)
1890 unsigned int truncated_clusters
;
1892 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
1893 truncated_clusters
= osb
->truncated_clusters
;
1894 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
1897 * Check whether we can succeed in allocating if we free
1900 if (truncated_clusters
< needed
)
1903 ret
= ocfs2_flush_truncate_log(osb
);
1909 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
1910 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
1917 int ocfs2_write_begin_nolock(struct file
*filp
,
1918 struct address_space
*mapping
,
1919 loff_t pos
, unsigned len
, unsigned flags
,
1920 struct page
**pagep
, void **fsdata
,
1921 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1923 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1924 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
1925 struct ocfs2_write_ctxt
*wc
;
1926 struct inode
*inode
= mapping
->host
;
1927 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1928 struct ocfs2_dinode
*di
;
1929 struct ocfs2_alloc_context
*data_ac
= NULL
;
1930 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1932 struct ocfs2_extent_tree et
;
1933 int try_free
= 1, ret1
;
1936 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1942 if (ocfs2_supports_inline_data(osb
)) {
1943 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1955 if (ocfs2_sparse_alloc(osb
))
1956 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
1958 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
1965 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1969 } else if (ret
== 1) {
1970 clusters_need
= wc
->w_clen
;
1971 ret
= ocfs2_refcount_cow(inode
, di_bh
,
1972 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1979 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1985 clusters_need
+= clusters_to_alloc
;
1987 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1989 trace_ocfs2_write_begin_nolock(
1990 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1991 (long long)i_size_read(inode
),
1992 le32_to_cpu(di
->i_clusters
),
1993 pos
, len
, flags
, mmap_page
,
1994 clusters_to_alloc
, extents_to_split
);
1997 * We set w_target_from, w_target_to here so that
1998 * ocfs2_write_end() knows which range in the target page to
1999 * write out. An allocation requires that we write the entire
2002 if (clusters_to_alloc
|| extents_to_split
) {
2004 * XXX: We are stretching the limits of
2005 * ocfs2_lock_allocators(). It greatly over-estimates
2006 * the work to be done.
2008 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
2010 ret
= ocfs2_lock_allocators(inode
, &et
,
2011 clusters_to_alloc
, extents_to_split
,
2012 &data_ac
, &meta_ac
);
2019 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
2021 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
2027 * We have to zero sparse allocated clusters, unwritten extent clusters,
2028 * and non-sparse clusters we just extended. For non-sparse writes,
2029 * we know zeros will only be needed in the first and/or last cluster.
2031 if (clusters_to_alloc
|| extents_to_split
||
2032 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
2033 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
2034 cluster_of_pages
= 1;
2036 cluster_of_pages
= 0;
2038 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
2040 handle
= ocfs2_start_trans(osb
, credits
);
2041 if (IS_ERR(handle
)) {
2042 ret
= PTR_ERR(handle
);
2047 wc
->w_handle
= handle
;
2049 if (clusters_to_alloc
) {
2050 ret
= dquot_alloc_space_nodirty(inode
,
2051 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
2056 * We don't want this to fail in ocfs2_write_end(), so do it
2059 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
2060 OCFS2_JOURNAL_ACCESS_WRITE
);
2067 * Fill our page array first. That way we've grabbed enough so
2068 * that we can zero and flush if we error after adding the
2071 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
2072 cluster_of_pages
, mmap_page
);
2073 if (ret
&& ret
!= -EAGAIN
) {
2079 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
2080 * the target page. In this case, we exit with no error and no target
2081 * page. This will trigger the caller, page_mkwrite(), to re-try
2084 if (ret
== -EAGAIN
) {
2085 BUG_ON(wc
->w_target_page
);
2090 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
2098 ocfs2_free_alloc_context(data_ac
);
2100 ocfs2_free_alloc_context(meta_ac
);
2103 *pagep
= wc
->w_target_page
;
2107 if (clusters_to_alloc
)
2108 dquot_free_space(inode
,
2109 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
2111 ocfs2_commit_trans(osb
, handle
);
2114 ocfs2_free_write_ctxt(wc
);
2117 ocfs2_free_alloc_context(data_ac
);
2121 ocfs2_free_alloc_context(meta_ac
);
2125 if (ret
== -ENOSPC
&& try_free
) {
2127 * Try to free some truncate log so that we can have enough
2128 * clusters to allocate.
2132 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
2143 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
2144 loff_t pos
, unsigned len
, unsigned flags
,
2145 struct page
**pagep
, void **fsdata
)
2148 struct buffer_head
*di_bh
= NULL
;
2149 struct inode
*inode
= mapping
->host
;
2151 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2158 * Take alloc sem here to prevent concurrent lookups. That way
2159 * the mapping, zeroing and tree manipulation within
2160 * ocfs2_write() will be safe against ->readpage(). This
2161 * should also serve to lock out allocation from a shared
2164 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2166 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
2167 fsdata
, di_bh
, NULL
);
2178 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2181 ocfs2_inode_unlock(inode
, 1);
2186 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
2187 unsigned len
, unsigned *copied
,
2188 struct ocfs2_dinode
*di
,
2189 struct ocfs2_write_ctxt
*wc
)
2193 if (unlikely(*copied
< len
)) {
2194 if (!PageUptodate(wc
->w_target_page
)) {
2200 kaddr
= kmap_atomic(wc
->w_target_page
);
2201 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
2202 kunmap_atomic(kaddr
);
2204 trace_ocfs2_write_end_inline(
2205 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
2206 (unsigned long long)pos
, *copied
,
2207 le16_to_cpu(di
->id2
.i_data
.id_count
),
2208 le16_to_cpu(di
->i_dyn_features
));
2211 int ocfs2_write_end_nolock(struct address_space
*mapping
,
2212 loff_t pos
, unsigned len
, unsigned copied
,
2213 struct page
*page
, void *fsdata
)
2216 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
2217 struct inode
*inode
= mapping
->host
;
2218 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2219 struct ocfs2_write_ctxt
*wc
= fsdata
;
2220 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
2221 handle_t
*handle
= wc
->w_handle
;
2222 struct page
*tmppage
;
2224 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2225 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
2226 goto out_write_size
;
2229 if (unlikely(copied
< len
)) {
2230 if (!PageUptodate(wc
->w_target_page
))
2233 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
2236 flush_dcache_page(wc
->w_target_page
);
2238 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
2239 tmppage
= wc
->w_pages
[i
];
2241 if (tmppage
== wc
->w_target_page
) {
2242 from
= wc
->w_target_from
;
2243 to
= wc
->w_target_to
;
2245 BUG_ON(from
> PAGE_CACHE_SIZE
||
2246 to
> PAGE_CACHE_SIZE
||
2250 * Pages adjacent to the target (if any) imply
2251 * a hole-filling write in which case we want
2252 * to flush their entire range.
2255 to
= PAGE_CACHE_SIZE
;
2258 if (page_has_buffers(tmppage
)) {
2259 if (ocfs2_should_order_data(inode
))
2260 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
2261 block_commit_write(tmppage
, from
, to
);
2267 if (pos
> i_size_read(inode
)) {
2268 i_size_write(inode
, pos
);
2269 mark_inode_dirty(inode
);
2271 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
2272 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
2273 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2274 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
2275 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2276 ocfs2_update_inode_fsync_trans(handle
, inode
, 1);
2277 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2279 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2280 * lock, or it will cause a deadlock since journal commit threads holds
2281 * this lock and will ask for the page lock when flushing the data.
2282 * put it here to preserve the unlock order.
2284 ocfs2_unlock_pages(wc
);
2286 ocfs2_commit_trans(osb
, handle
);
2288 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2290 brelse(wc
->w_di_bh
);
2296 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2297 loff_t pos
, unsigned len
, unsigned copied
,
2298 struct page
*page
, void *fsdata
)
2301 struct inode
*inode
= mapping
->host
;
2303 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2305 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2306 ocfs2_inode_unlock(inode
, 1);
2311 const struct address_space_operations ocfs2_aops
= {
2312 .readpage
= ocfs2_readpage
,
2313 .readpages
= ocfs2_readpages
,
2314 .writepage
= ocfs2_writepage
,
2315 .write_begin
= ocfs2_write_begin
,
2316 .write_end
= ocfs2_write_end
,
2318 .direct_IO
= ocfs2_direct_IO
,
2319 .invalidatepage
= block_invalidatepage
,
2320 .releasepage
= ocfs2_releasepage
,
2321 .migratepage
= buffer_migrate_page
,
2322 .is_partially_uptodate
= block_is_partially_uptodate
,
2323 .error_remove_page
= generic_error_remove_page
,