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>
32 #include <linux/uio.h>
34 #include <cluster/masklog.h>
41 #include "extent_map.h"
48 #include "refcounttree.h"
49 #include "ocfs2_trace.h"
51 #include "buffer_head_io.h"
56 static int ocfs2_symlink_get_block(struct inode
*inode
, sector_t iblock
,
57 struct buffer_head
*bh_result
, int create
)
61 struct ocfs2_dinode
*fe
= NULL
;
62 struct buffer_head
*bh
= NULL
;
63 struct buffer_head
*buffer_cache_bh
= NULL
;
64 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
67 trace_ocfs2_symlink_get_block(
68 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
69 (unsigned long long)iblock
, bh_result
, create
);
71 BUG_ON(ocfs2_inode_is_fast_symlink(inode
));
73 if ((iblock
<< inode
->i_sb
->s_blocksize_bits
) > PATH_MAX
+ 1) {
74 mlog(ML_ERROR
, "block offset > PATH_MAX: %llu",
75 (unsigned long long)iblock
);
79 status
= ocfs2_read_inode_block(inode
, &bh
);
84 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
86 if ((u64
)iblock
>= ocfs2_clusters_to_blocks(inode
->i_sb
,
87 le32_to_cpu(fe
->i_clusters
))) {
89 mlog(ML_ERROR
, "block offset is outside the allocated size: "
90 "%llu\n", (unsigned long long)iblock
);
94 /* We don't use the page cache to create symlink data, so if
95 * need be, copy it over from the buffer cache. */
96 if (!buffer_uptodate(bh_result
) && ocfs2_inode_is_new(inode
)) {
97 u64 blkno
= le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) +
99 buffer_cache_bh
= sb_getblk(osb
->sb
, blkno
);
100 if (!buffer_cache_bh
) {
102 mlog(ML_ERROR
, "couldn't getblock for symlink!\n");
106 /* we haven't locked out transactions, so a commit
107 * could've happened. Since we've got a reference on
108 * the bh, even if it commits while we're doing the
109 * copy, the data is still good. */
110 if (buffer_jbd(buffer_cache_bh
)
111 && ocfs2_inode_is_new(inode
)) {
112 kaddr
= kmap_atomic(bh_result
->b_page
);
114 mlog(ML_ERROR
, "couldn't kmap!\n");
117 memcpy(kaddr
+ (bh_result
->b_size
* iblock
),
118 buffer_cache_bh
->b_data
,
120 kunmap_atomic(kaddr
);
121 set_buffer_uptodate(bh_result
);
123 brelse(buffer_cache_bh
);
126 map_bh(bh_result
, inode
->i_sb
,
127 le64_to_cpu(fe
->id2
.i_list
.l_recs
[0].e_blkno
) + iblock
);
137 int ocfs2_get_block(struct inode
*inode
, sector_t iblock
,
138 struct buffer_head
*bh_result
, int create
)
141 unsigned int ext_flags
;
142 u64 max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
143 u64 p_blkno
, count
, past_eof
;
144 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
146 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
147 (unsigned long long)iblock
, bh_result
, create
);
149 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SYSTEM_FILE
)
150 mlog(ML_NOTICE
, "get_block on system inode 0x%p (%lu)\n",
151 inode
, inode
->i_ino
);
153 if (S_ISLNK(inode
->i_mode
)) {
154 /* this always does I/O for some reason. */
155 err
= ocfs2_symlink_get_block(inode
, iblock
, bh_result
, create
);
159 err
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
, &count
,
162 mlog(ML_ERROR
, "Error %d from get_blocks(0x%p, %llu, 1, "
163 "%llu, NULL)\n", err
, inode
, (unsigned long long)iblock
,
164 (unsigned long long)p_blkno
);
168 if (max_blocks
< count
)
172 * ocfs2 never allocates in this function - the only time we
173 * need to use BH_New is when we're extending i_size on a file
174 * system which doesn't support holes, in which case BH_New
175 * allows __block_write_begin() to zero.
177 * If we see this on a sparse file system, then a truncate has
178 * raced us and removed the cluster. In this case, we clear
179 * the buffers dirty and uptodate bits and let the buffer code
180 * ignore it as a hole.
182 if (create
&& p_blkno
== 0 && ocfs2_sparse_alloc(osb
)) {
183 clear_buffer_dirty(bh_result
);
184 clear_buffer_uptodate(bh_result
);
188 /* Treat the unwritten extent as a hole for zeroing purposes. */
189 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
190 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
192 bh_result
->b_size
= count
<< inode
->i_blkbits
;
194 if (!ocfs2_sparse_alloc(osb
)) {
198 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
199 (unsigned long long)iblock
,
200 (unsigned long long)p_blkno
,
201 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
202 mlog(ML_ERROR
, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode
), OCFS2_I(inode
)->ip_clusters
);
208 past_eof
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
210 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
211 (unsigned long long)past_eof
);
212 if (create
&& (iblock
>= past_eof
))
213 set_buffer_new(bh_result
);
222 int ocfs2_read_inline_data(struct inode
*inode
, struct page
*page
,
223 struct buffer_head
*di_bh
)
227 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
229 if (!(le16_to_cpu(di
->i_dyn_features
) & OCFS2_INLINE_DATA_FL
)) {
230 ocfs2_error(inode
->i_sb
, "Inode %llu lost inline data flag",
231 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
235 size
= i_size_read(inode
);
237 if (size
> PAGE_CACHE_SIZE
||
238 size
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
)) {
239 ocfs2_error(inode
->i_sb
,
240 "Inode %llu has with inline data has bad size: %Lu",
241 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
242 (unsigned long long)size
);
246 kaddr
= kmap_atomic(page
);
248 memcpy(kaddr
, di
->id2
.i_data
.id_data
, size
);
249 /* Clear the remaining part of the page */
250 memset(kaddr
+ size
, 0, PAGE_CACHE_SIZE
- size
);
251 flush_dcache_page(page
);
252 kunmap_atomic(kaddr
);
254 SetPageUptodate(page
);
259 static int ocfs2_readpage_inline(struct inode
*inode
, struct page
*page
)
262 struct buffer_head
*di_bh
= NULL
;
264 BUG_ON(!PageLocked(page
));
265 BUG_ON(!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
));
267 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
273 ret
= ocfs2_read_inline_data(inode
, page
, di_bh
);
281 static int ocfs2_readpage(struct file
*file
, struct page
*page
)
283 struct inode
*inode
= page
->mapping
->host
;
284 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
285 loff_t start
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
288 trace_ocfs2_readpage((unsigned long long)oi
->ip_blkno
,
289 (page
? page
->index
: 0));
291 ret
= ocfs2_inode_lock_with_page(inode
, NULL
, 0, page
);
293 if (ret
== AOP_TRUNCATED_PAGE
)
299 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
301 * Unlock the page and cycle ip_alloc_sem so that we don't
302 * busyloop waiting for ip_alloc_sem to unlock
304 ret
= AOP_TRUNCATED_PAGE
;
307 down_read(&oi
->ip_alloc_sem
);
308 up_read(&oi
->ip_alloc_sem
);
309 goto out_inode_unlock
;
313 * i_size might have just been updated as we grabed the meta lock. We
314 * might now be discovering a truncate that hit on another node.
315 * block_read_full_page->get_block freaks out if it is asked to read
316 * beyond the end of a file, so we check here. Callers
317 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
318 * and notice that the page they just read isn't needed.
320 * XXX sys_readahead() seems to get that wrong?
322 if (start
>= i_size_read(inode
)) {
323 zero_user(page
, 0, PAGE_SIZE
);
324 SetPageUptodate(page
);
329 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
330 ret
= ocfs2_readpage_inline(inode
, page
);
332 ret
= block_read_full_page(page
, ocfs2_get_block
);
336 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
338 ocfs2_inode_unlock(inode
, 0);
346 * This is used only for read-ahead. Failures or difficult to handle
347 * situations are safe to ignore.
349 * Right now, we don't bother with BH_Boundary - in-inode extent lists
350 * are quite large (243 extents on 4k blocks), so most inodes don't
351 * grow out to a tree. If need be, detecting boundary extents could
352 * trivially be added in a future version of ocfs2_get_block().
354 static int ocfs2_readpages(struct file
*filp
, struct address_space
*mapping
,
355 struct list_head
*pages
, unsigned nr_pages
)
358 struct inode
*inode
= mapping
->host
;
359 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
364 * Use the nonblocking flag for the dlm code to avoid page
365 * lock inversion, but don't bother with retrying.
367 ret
= ocfs2_inode_lock_full(inode
, NULL
, 0, OCFS2_LOCK_NONBLOCK
);
371 if (down_read_trylock(&oi
->ip_alloc_sem
) == 0) {
372 ocfs2_inode_unlock(inode
, 0);
377 * Don't bother with inline-data. There isn't anything
378 * to read-ahead in that case anyway...
380 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
384 * Check whether a remote node truncated this file - we just
385 * drop out in that case as it's not worth handling here.
387 last
= list_entry(pages
->prev
, struct page
, lru
);
388 start
= (loff_t
)last
->index
<< PAGE_CACHE_SHIFT
;
389 if (start
>= i_size_read(inode
))
392 err
= mpage_readpages(mapping
, pages
, nr_pages
, ocfs2_get_block
);
395 up_read(&oi
->ip_alloc_sem
);
396 ocfs2_inode_unlock(inode
, 0);
401 /* Note: Because we don't support holes, our allocation has
402 * already happened (allocation writes zeros to the file data)
403 * so we don't have to worry about ordered writes in
406 * ->writepage is called during the process of invalidating the page cache
407 * during blocked lock processing. It can't block on any cluster locks
408 * to during block mapping. It's relying on the fact that the block
409 * mapping can't have disappeared under the dirty pages that it is
410 * being asked to write back.
412 static int ocfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
414 trace_ocfs2_writepage(
415 (unsigned long long)OCFS2_I(page
->mapping
->host
)->ip_blkno
,
418 return block_write_full_page(page
, ocfs2_get_block
, wbc
);
421 /* Taken from ext3. We don't necessarily need the full blown
422 * functionality yet, but IMHO it's better to cut and paste the whole
423 * thing so we can avoid introducing our own bugs (and easily pick up
424 * their fixes when they happen) --Mark */
425 int walk_page_buffers( handle_t
*handle
,
426 struct buffer_head
*head
,
430 int (*fn
)( handle_t
*handle
,
431 struct buffer_head
*bh
))
433 struct buffer_head
*bh
;
434 unsigned block_start
, block_end
;
435 unsigned blocksize
= head
->b_size
;
437 struct buffer_head
*next
;
439 for ( bh
= head
, block_start
= 0;
440 ret
== 0 && (bh
!= head
|| !block_start
);
441 block_start
= block_end
, bh
= next
)
443 next
= bh
->b_this_page
;
444 block_end
= block_start
+ blocksize
;
445 if (block_end
<= from
|| block_start
>= to
) {
446 if (partial
&& !buffer_uptodate(bh
))
450 err
= (*fn
)(handle
, bh
);
457 static sector_t
ocfs2_bmap(struct address_space
*mapping
, sector_t block
)
462 struct inode
*inode
= mapping
->host
;
464 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
465 (unsigned long long)block
);
467 /* We don't need to lock journal system files, since they aren't
468 * accessed concurrently from multiple nodes.
470 if (!INODE_JOURNAL(inode
)) {
471 err
= ocfs2_inode_lock(inode
, NULL
, 0);
477 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
480 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
481 err
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
484 if (!INODE_JOURNAL(inode
)) {
485 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
486 ocfs2_inode_unlock(inode
, 0);
490 mlog(ML_ERROR
, "get_blocks() failed, block = %llu\n",
491 (unsigned long long)block
);
497 status
= err
? 0 : p_blkno
;
503 * TODO: Make this into a generic get_blocks function.
505 * From do_direct_io in direct-io.c:
506 * "So what we do is to permit the ->get_blocks function to populate
507 * bh.b_size with the size of IO which is permitted at this offset and
510 * This function is called directly from get_more_blocks in direct-io.c.
512 * called like this: dio->get_blocks(dio->inode, fs_startblk,
513 * fs_count, map_bh, dio->rw == WRITE);
515 static int ocfs2_direct_IO_get_blocks(struct inode
*inode
, sector_t iblock
,
516 struct buffer_head
*bh_result
, int create
)
520 int alloc_locked
= 0;
521 u64 p_blkno
, inode_blocks
, contig_blocks
;
522 unsigned int ext_flags
;
523 unsigned char blocksize_bits
= inode
->i_sb
->s_blocksize_bits
;
524 unsigned long max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
525 unsigned long len
= bh_result
->b_size
;
526 unsigned int clusters_to_alloc
= 0;
528 cpos
= ocfs2_blocks_to_clusters(inode
->i_sb
, iblock
);
530 /* This function won't even be called if the request isn't all
531 * nicely aligned and of the right size, so there's no need
532 * for us to check any of that. */
534 inode_blocks
= ocfs2_blocks_for_bytes(inode
->i_sb
, i_size_read(inode
));
536 /* This figures out the size of the next contiguous block, and
537 * our logical offset */
538 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
539 &contig_blocks
, &ext_flags
);
541 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
542 (unsigned long long)iblock
);
547 /* We should already CoW the refcounted extent in case of create. */
548 BUG_ON(create
&& (ext_flags
& OCFS2_EXT_REFCOUNTED
));
550 /* allocate blocks if no p_blkno is found, and create == 1 */
551 if (!p_blkno
&& create
) {
552 ret
= ocfs2_inode_lock(inode
, NULL
, 1);
560 /* fill hole, allocate blocks can't be larger than the size
562 clusters_to_alloc
= ocfs2_clusters_for_bytes(inode
->i_sb
, len
);
563 if (clusters_to_alloc
> contig_blocks
)
564 clusters_to_alloc
= contig_blocks
;
566 /* allocate extent and insert them into the extent tree */
567 ret
= ocfs2_extend_allocation(inode
, cpos
,
568 clusters_to_alloc
, 0);
574 ret
= ocfs2_extent_map_get_blocks(inode
, iblock
, &p_blkno
,
575 &contig_blocks
, &ext_flags
);
577 mlog(ML_ERROR
, "get_blocks() failed iblock=%llu\n",
578 (unsigned long long)iblock
);
585 * get_more_blocks() expects us to describe a hole by clearing
586 * the mapped bit on bh_result().
588 * Consider an unwritten extent as a hole.
590 if (p_blkno
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
591 map_bh(bh_result
, inode
->i_sb
, p_blkno
);
593 clear_buffer_mapped(bh_result
);
595 /* make sure we don't map more than max_blocks blocks here as
596 that's all the kernel will handle at this point. */
597 if (max_blocks
< contig_blocks
)
598 contig_blocks
= max_blocks
;
599 bh_result
->b_size
= contig_blocks
<< blocksize_bits
;
602 ocfs2_inode_unlock(inode
, 1);
607 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
608 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
609 * to protect io on one node from truncation on another.
611 static void ocfs2_dio_end_io(struct kiocb
*iocb
,
616 struct inode
*inode
= file_inode(iocb
->ki_filp
);
619 /* this io's submitter should not have unlocked this before we could */
620 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb
));
622 if (ocfs2_iocb_is_sem_locked(iocb
))
623 ocfs2_iocb_clear_sem_locked(iocb
);
625 if (ocfs2_iocb_is_unaligned_aio(iocb
)) {
626 ocfs2_iocb_clear_unaligned_aio(iocb
);
628 mutex_unlock(&OCFS2_I(inode
)->ip_unaligned_aio
);
631 ocfs2_iocb_clear_rw_locked(iocb
);
633 level
= ocfs2_iocb_rw_locked_level(iocb
);
634 ocfs2_rw_unlock(inode
, level
);
637 static int ocfs2_releasepage(struct page
*page
, gfp_t wait
)
639 if (!page_has_buffers(page
))
641 return try_to_free_buffers(page
);
644 static int ocfs2_is_overwrite(struct ocfs2_super
*osb
,
645 struct inode
*inode
, loff_t offset
)
650 unsigned int num_clusters
= 0;
651 unsigned int ext_flags
= 0;
653 v_cpos
= ocfs2_bytes_to_clusters(osb
->sb
, offset
);
654 ret
= ocfs2_get_clusters(inode
, v_cpos
, &p_cpos
,
655 &num_clusters
, &ext_flags
);
661 if (p_cpos
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
))
667 static ssize_t
ocfs2_direct_IO_write(struct kiocb
*iocb
,
668 struct iov_iter
*iter
,
673 bool orphaned
= false;
674 int is_overwrite
= 0;
675 struct file
*file
= iocb
->ki_filp
;
676 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
677 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
678 struct buffer_head
*di_bh
= NULL
;
679 size_t count
= iter
->count
;
680 journal_t
*journal
= osb
->journal
->j_journal
;
683 loff_t final_size
= offset
+ count
;
684 int append_write
= offset
>= i_size_read(inode
) ? 1 : 0;
685 unsigned int num_clusters
= 0;
686 unsigned int ext_flags
= 0;
691 zero_len
= do_div(o
, 1 << osb
->s_clustersize_bits
);
692 cluster_align
= !zero_len
;
696 * when final_size > inode->i_size, inode->i_size will be
697 * updated after direct write, so add the inode to orphan
700 if (final_size
> i_size_read(inode
)) {
701 ret
= ocfs2_add_inode_to_orphan(osb
, inode
);
710 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
716 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
717 ret
= ocfs2_zero_extend(inode
, di_bh
, offset
);
719 ret
= ocfs2_extend_no_holes(inode
, di_bh
, offset
,
723 ocfs2_inode_unlock(inode
, 1);
728 is_overwrite
= ocfs2_is_overwrite(osb
, inode
, offset
);
729 if (is_overwrite
< 0) {
730 mlog_errno(is_overwrite
);
731 ocfs2_inode_unlock(inode
, 1);
736 ocfs2_inode_unlock(inode
, 1);
741 written
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
742 offset
, 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
)
845 if (iov_iter_rw(iter
) == READ
)
846 return __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
848 ocfs2_direct_IO_get_blocks
,
849 ocfs2_dio_end_io
, NULL
, 0);
851 return ocfs2_direct_IO_write(iocb
, iter
, offset
);
854 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
859 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
861 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
864 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
866 cluster_start
= cpos
% cpp
;
867 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
869 cluster_end
= cluster_start
+ osb
->s_clustersize
;
872 BUG_ON(cluster_start
> PAGE_SIZE
);
873 BUG_ON(cluster_end
> PAGE_SIZE
);
876 *start
= cluster_start
;
882 * 'from' and 'to' are the region in the page to avoid zeroing.
884 * If pagesize > clustersize, this function will avoid zeroing outside
885 * of the cluster boundary.
887 * from == to == 0 is code for "zero the entire cluster region"
889 static void ocfs2_clear_page_regions(struct page
*page
,
890 struct ocfs2_super
*osb
, u32 cpos
,
891 unsigned from
, unsigned to
)
894 unsigned int cluster_start
, cluster_end
;
896 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
898 kaddr
= kmap_atomic(page
);
901 if (from
> cluster_start
)
902 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
903 if (to
< cluster_end
)
904 memset(kaddr
+ to
, 0, cluster_end
- to
);
906 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
909 kunmap_atomic(kaddr
);
913 * Nonsparse file systems fully allocate before we get to the write
914 * code. This prevents ocfs2_write() from tagging the write as an
915 * allocating one, which means ocfs2_map_page_blocks() might try to
916 * read-in the blocks at the tail of our file. Avoid reading them by
917 * testing i_size against each block offset.
919 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
920 unsigned int block_start
)
922 u64 offset
= page_offset(page
) + block_start
;
924 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
927 if (i_size_read(inode
) > offset
)
934 * Some of this taken from __block_write_begin(). We already have our
935 * mapping by now though, and the entire write will be allocating or
936 * it won't, so not much need to use BH_New.
938 * This will also skip zeroing, which is handled externally.
940 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
941 struct inode
*inode
, unsigned int from
,
942 unsigned int to
, int new)
945 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
946 unsigned int block_end
, block_start
;
947 unsigned int bsize
= 1 << inode
->i_blkbits
;
949 if (!page_has_buffers(page
))
950 create_empty_buffers(page
, bsize
, 0);
952 head
= page_buffers(page
);
953 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
954 bh
= bh
->b_this_page
, block_start
+= bsize
) {
955 block_end
= block_start
+ bsize
;
957 clear_buffer_new(bh
);
960 * Ignore blocks outside of our i/o range -
961 * they may belong to unallocated clusters.
963 if (block_start
>= to
|| block_end
<= from
) {
964 if (PageUptodate(page
))
965 set_buffer_uptodate(bh
);
970 * For an allocating write with cluster size >= page
971 * size, we always write the entire page.
976 if (!buffer_mapped(bh
)) {
977 map_bh(bh
, inode
->i_sb
, *p_blkno
);
978 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
981 if (PageUptodate(page
)) {
982 if (!buffer_uptodate(bh
))
983 set_buffer_uptodate(bh
);
984 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
986 ocfs2_should_read_blk(inode
, page
, block_start
) &&
987 (block_start
< from
|| block_end
> to
)) {
988 ll_rw_block(READ
, 1, &bh
);
992 *p_blkno
= *p_blkno
+ 1;
996 * If we issued read requests - let them complete.
998 while(wait_bh
> wait
) {
999 wait_on_buffer(*--wait_bh
);
1000 if (!buffer_uptodate(*wait_bh
))
1004 if (ret
== 0 || !new)
1008 * If we get -EIO above, zero out any newly allocated blocks
1009 * to avoid exposing stale data.
1014 block_end
= block_start
+ bsize
;
1015 if (block_end
<= from
)
1017 if (block_start
>= to
)
1020 zero_user(page
, block_start
, bh
->b_size
);
1021 set_buffer_uptodate(bh
);
1022 mark_buffer_dirty(bh
);
1025 block_start
= block_end
;
1026 bh
= bh
->b_this_page
;
1027 } while (bh
!= head
);
1032 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
1033 #define OCFS2_MAX_CTXT_PAGES 1
1035 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
1038 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
1041 * Describe the state of a single cluster to be written to.
1043 struct ocfs2_write_cluster_desc
{
1047 * Give this a unique field because c_phys eventually gets
1051 unsigned c_unwritten
;
1052 unsigned c_needs_zero
;
1055 struct ocfs2_write_ctxt
{
1056 /* Logical cluster position / len of write */
1060 /* First cluster allocated in a nonsparse extend */
1061 u32 w_first_new_cpos
;
1063 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
1066 * This is true if page_size > cluster_size.
1068 * It triggers a set of special cases during write which might
1069 * have to deal with allocating writes to partial pages.
1071 unsigned int w_large_pages
;
1074 * Pages involved in this write.
1076 * w_target_page is the page being written to by the user.
1078 * w_pages is an array of pages which always contains
1079 * w_target_page, and in the case of an allocating write with
1080 * page_size < cluster size, it will contain zero'd and mapped
1081 * pages adjacent to w_target_page which need to be written
1082 * out in so that future reads from that region will get
1085 unsigned int w_num_pages
;
1086 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
1087 struct page
*w_target_page
;
1090 * w_target_locked is used for page_mkwrite path indicating no unlocking
1091 * against w_target_page in ocfs2_write_end_nolock.
1093 unsigned int w_target_locked
:1;
1096 * ocfs2_write_end() uses this to know what the real range to
1097 * write in the target should be.
1099 unsigned int w_target_from
;
1100 unsigned int w_target_to
;
1103 * We could use journal_current_handle() but this is cleaner,
1108 struct buffer_head
*w_di_bh
;
1110 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
1113 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
1117 for(i
= 0; i
< num_pages
; i
++) {
1119 unlock_page(pages
[i
]);
1120 mark_page_accessed(pages
[i
]);
1121 page_cache_release(pages
[i
]);
1126 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt
*wc
)
1131 * w_target_locked is only set to true in the page_mkwrite() case.
1132 * The intent is to allow us to lock the target page from write_begin()
1133 * to write_end(). The caller must hold a ref on w_target_page.
1135 if (wc
->w_target_locked
) {
1136 BUG_ON(!wc
->w_target_page
);
1137 for (i
= 0; i
< wc
->w_num_pages
; i
++) {
1138 if (wc
->w_target_page
== wc
->w_pages
[i
]) {
1139 wc
->w_pages
[i
] = NULL
;
1143 mark_page_accessed(wc
->w_target_page
);
1144 page_cache_release(wc
->w_target_page
);
1146 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
1149 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
1151 ocfs2_unlock_pages(wc
);
1152 brelse(wc
->w_di_bh
);
1156 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
1157 struct ocfs2_super
*osb
, loff_t pos
,
1158 unsigned len
, struct buffer_head
*di_bh
)
1161 struct ocfs2_write_ctxt
*wc
;
1163 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
1167 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
1168 wc
->w_first_new_cpos
= UINT_MAX
;
1169 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
1170 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
1172 wc
->w_di_bh
= di_bh
;
1174 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
1175 wc
->w_large_pages
= 1;
1177 wc
->w_large_pages
= 0;
1179 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
1187 * If a page has any new buffers, zero them out here, and mark them uptodate
1188 * and dirty so they'll be written out (in order to prevent uninitialised
1189 * block data from leaking). And clear the new bit.
1191 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1193 unsigned int block_start
, block_end
;
1194 struct buffer_head
*head
, *bh
;
1196 BUG_ON(!PageLocked(page
));
1197 if (!page_has_buffers(page
))
1200 bh
= head
= page_buffers(page
);
1203 block_end
= block_start
+ bh
->b_size
;
1205 if (buffer_new(bh
)) {
1206 if (block_end
> from
&& block_start
< to
) {
1207 if (!PageUptodate(page
)) {
1208 unsigned start
, end
;
1210 start
= max(from
, block_start
);
1211 end
= min(to
, block_end
);
1213 zero_user_segment(page
, start
, end
);
1214 set_buffer_uptodate(bh
);
1217 clear_buffer_new(bh
);
1218 mark_buffer_dirty(bh
);
1222 block_start
= block_end
;
1223 bh
= bh
->b_this_page
;
1224 } while (bh
!= head
);
1228 * Only called when we have a failure during allocating write to write
1229 * zero's to the newly allocated region.
1231 static void ocfs2_write_failure(struct inode
*inode
,
1232 struct ocfs2_write_ctxt
*wc
,
1233 loff_t user_pos
, unsigned user_len
)
1236 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1237 to
= user_pos
+ user_len
;
1238 struct page
*tmppage
;
1240 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1242 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1243 tmppage
= wc
->w_pages
[i
];
1245 if (page_has_buffers(tmppage
)) {
1246 if (ocfs2_should_order_data(inode
))
1247 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1249 block_commit_write(tmppage
, from
, to
);
1254 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1255 struct ocfs2_write_ctxt
*wc
,
1256 struct page
*page
, u32 cpos
,
1257 loff_t user_pos
, unsigned user_len
,
1261 unsigned int map_from
= 0, map_to
= 0;
1262 unsigned int cluster_start
, cluster_end
;
1263 unsigned int user_data_from
= 0, user_data_to
= 0;
1265 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1266 &cluster_start
, &cluster_end
);
1268 /* treat the write as new if the a hole/lseek spanned across
1269 * the page boundary.
1271 new = new | ((i_size_read(inode
) <= page_offset(page
)) &&
1272 (page_offset(page
) <= user_pos
));
1274 if (page
== wc
->w_target_page
) {
1275 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1276 map_to
= map_from
+ user_len
;
1279 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1280 cluster_start
, cluster_end
,
1283 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1284 map_from
, map_to
, new);
1290 user_data_from
= map_from
;
1291 user_data_to
= map_to
;
1293 map_from
= cluster_start
;
1294 map_to
= cluster_end
;
1298 * If we haven't allocated the new page yet, we
1299 * shouldn't be writing it out without copying user
1300 * data. This is likely a math error from the caller.
1304 map_from
= cluster_start
;
1305 map_to
= cluster_end
;
1307 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1308 cluster_start
, cluster_end
, new);
1316 * Parts of newly allocated pages need to be zero'd.
1318 * Above, we have also rewritten 'to' and 'from' - as far as
1319 * the rest of the function is concerned, the entire cluster
1320 * range inside of a page needs to be written.
1322 * We can skip this if the page is up to date - it's already
1323 * been zero'd from being read in as a hole.
1325 if (new && !PageUptodate(page
))
1326 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1327 cpos
, user_data_from
, user_data_to
);
1329 flush_dcache_page(page
);
1336 * This function will only grab one clusters worth of pages.
1338 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1339 struct ocfs2_write_ctxt
*wc
,
1340 u32 cpos
, loff_t user_pos
,
1341 unsigned user_len
, int new,
1342 struct page
*mmap_page
)
1345 unsigned long start
, target_index
, end_index
, index
;
1346 struct inode
*inode
= mapping
->host
;
1349 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1352 * Figure out how many pages we'll be manipulating here. For
1353 * non allocating write, we just change the one
1354 * page. Otherwise, we'll need a whole clusters worth. If we're
1355 * writing past i_size, we only need enough pages to cover the
1356 * last page of the write.
1359 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1360 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1362 * We need the index *past* the last page we could possibly
1363 * touch. This is the page past the end of the write or
1364 * i_size, whichever is greater.
1366 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1367 BUG_ON(last_byte
< 1);
1368 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1369 if ((start
+ wc
->w_num_pages
) > end_index
)
1370 wc
->w_num_pages
= end_index
- start
;
1372 wc
->w_num_pages
= 1;
1373 start
= target_index
;
1376 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1379 if (index
== target_index
&& mmap_page
) {
1381 * ocfs2_pagemkwrite() is a little different
1382 * and wants us to directly use the page
1385 lock_page(mmap_page
);
1387 /* Exit and let the caller retry */
1388 if (mmap_page
->mapping
!= mapping
) {
1389 WARN_ON(mmap_page
->mapping
);
1390 unlock_page(mmap_page
);
1395 page_cache_get(mmap_page
);
1396 wc
->w_pages
[i
] = mmap_page
;
1397 wc
->w_target_locked
= true;
1399 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1401 if (!wc
->w_pages
[i
]) {
1407 wait_for_stable_page(wc
->w_pages
[i
]);
1409 if (index
== target_index
)
1410 wc
->w_target_page
= wc
->w_pages
[i
];
1414 wc
->w_target_locked
= false;
1419 * Prepare a single cluster for write one cluster into the file.
1421 static int ocfs2_write_cluster(struct address_space
*mapping
,
1422 u32 phys
, unsigned int unwritten
,
1423 unsigned int should_zero
,
1424 struct ocfs2_alloc_context
*data_ac
,
1425 struct ocfs2_alloc_context
*meta_ac
,
1426 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1427 loff_t user_pos
, unsigned user_len
)
1430 u64 v_blkno
, p_blkno
;
1431 struct inode
*inode
= mapping
->host
;
1432 struct ocfs2_extent_tree et
;
1434 new = phys
== 0 ? 1 : 0;
1439 * This is safe to call with the page locks - it won't take
1440 * any additional semaphores or cluster locks.
1443 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1444 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1445 wc
->w_handle
, data_ac
,
1448 * This shouldn't happen because we must have already
1449 * calculated the correct meta data allocation required. The
1450 * internal tree allocation code should know how to increase
1451 * transaction credits itself.
1453 * If need be, we could handle -EAGAIN for a
1454 * RESTART_TRANS here.
1456 mlog_bug_on_msg(ret
== -EAGAIN
,
1457 "Inode %llu: EAGAIN return during allocation.\n",
1458 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1463 } else if (unwritten
) {
1464 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1466 ret
= ocfs2_mark_extent_written(inode
, &et
,
1467 wc
->w_handle
, cpos
, 1, phys
,
1468 meta_ac
, &wc
->w_dealloc
);
1476 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1478 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1481 * The only reason this should fail is due to an inability to
1482 * find the extent added.
1484 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1487 mlog(ML_ERROR
, "Get physical blkno failed for inode %llu, "
1488 "at logical block %llu",
1489 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1490 (unsigned long long)v_blkno
);
1494 BUG_ON(p_blkno
== 0);
1496 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1499 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1500 wc
->w_pages
[i
], cpos
,
1511 * We only have cleanup to do in case of allocating write.
1514 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1521 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1522 struct ocfs2_alloc_context
*data_ac
,
1523 struct ocfs2_alloc_context
*meta_ac
,
1524 struct ocfs2_write_ctxt
*wc
,
1525 loff_t pos
, unsigned len
)
1529 unsigned int local_len
= len
;
1530 struct ocfs2_write_cluster_desc
*desc
;
1531 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1533 for (i
= 0; i
< wc
->w_clen
; i
++) {
1534 desc
= &wc
->w_desc
[i
];
1537 * We have to make sure that the total write passed in
1538 * doesn't extend past a single cluster.
1541 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1542 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1543 local_len
= osb
->s_clustersize
- cluster_off
;
1545 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1549 wc
, desc
->c_cpos
, pos
, local_len
);
1565 * ocfs2_write_end() wants to know which parts of the target page it
1566 * should complete the write on. It's easiest to compute them ahead of
1567 * time when a more complete view of the write is available.
1569 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1570 struct ocfs2_write_ctxt
*wc
,
1571 loff_t pos
, unsigned len
, int alloc
)
1573 struct ocfs2_write_cluster_desc
*desc
;
1575 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1576 wc
->w_target_to
= wc
->w_target_from
+ len
;
1582 * Allocating write - we may have different boundaries based
1583 * on page size and cluster size.
1585 * NOTE: We can no longer compute one value from the other as
1586 * the actual write length and user provided length may be
1590 if (wc
->w_large_pages
) {
1592 * We only care about the 1st and last cluster within
1593 * our range and whether they should be zero'd or not. Either
1594 * value may be extended out to the start/end of a
1595 * newly allocated cluster.
1597 desc
= &wc
->w_desc
[0];
1598 if (desc
->c_needs_zero
)
1599 ocfs2_figure_cluster_boundaries(osb
,
1604 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1605 if (desc
->c_needs_zero
)
1606 ocfs2_figure_cluster_boundaries(osb
,
1611 wc
->w_target_from
= 0;
1612 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1617 * Populate each single-cluster write descriptor in the write context
1618 * with information about the i/o to be done.
1620 * Returns the number of clusters that will have to be allocated, as
1621 * well as a worst case estimate of the number of extent records that
1622 * would have to be created during a write to an unwritten region.
1624 static int ocfs2_populate_write_desc(struct inode
*inode
,
1625 struct ocfs2_write_ctxt
*wc
,
1626 unsigned int *clusters_to_alloc
,
1627 unsigned int *extents_to_split
)
1630 struct ocfs2_write_cluster_desc
*desc
;
1631 unsigned int num_clusters
= 0;
1632 unsigned int ext_flags
= 0;
1636 *clusters_to_alloc
= 0;
1637 *extents_to_split
= 0;
1639 for (i
= 0; i
< wc
->w_clen
; i
++) {
1640 desc
= &wc
->w_desc
[i
];
1641 desc
->c_cpos
= wc
->w_cpos
+ i
;
1643 if (num_clusters
== 0) {
1645 * Need to look up the next extent record.
1647 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1648 &num_clusters
, &ext_flags
);
1654 /* We should already CoW the refcountd extent. */
1655 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1658 * Assume worst case - that we're writing in
1659 * the middle of the extent.
1661 * We can assume that the write proceeds from
1662 * left to right, in which case the extent
1663 * insert code is smart enough to coalesce the
1664 * next splits into the previous records created.
1666 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1667 *extents_to_split
= *extents_to_split
+ 2;
1670 * Only increment phys if it doesn't describe
1677 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1678 * file that got extended. w_first_new_cpos tells us
1679 * where the newly allocated clusters are so we can
1682 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1684 desc
->c_needs_zero
= 1;
1687 desc
->c_phys
= phys
;
1690 desc
->c_needs_zero
= 1;
1691 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1694 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1695 desc
->c_unwritten
= 1;
1696 desc
->c_needs_zero
= 1;
1707 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1708 struct inode
*inode
,
1709 struct ocfs2_write_ctxt
*wc
)
1712 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1715 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1717 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1718 if (IS_ERR(handle
)) {
1719 ret
= PTR_ERR(handle
);
1724 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1726 ocfs2_commit_trans(osb
, handle
);
1732 * If we don't set w_num_pages then this page won't get unlocked
1733 * and freed on cleanup of the write context.
1735 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1736 wc
->w_num_pages
= 1;
1738 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1739 OCFS2_JOURNAL_ACCESS_WRITE
);
1741 ocfs2_commit_trans(osb
, handle
);
1747 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1748 ocfs2_set_inode_data_inline(inode
, di
);
1750 if (!PageUptodate(page
)) {
1751 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1753 ocfs2_commit_trans(osb
, handle
);
1759 wc
->w_handle
= handle
;
1764 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1766 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1768 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1773 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1774 struct inode
*inode
, loff_t pos
,
1775 unsigned len
, struct page
*mmap_page
,
1776 struct ocfs2_write_ctxt
*wc
)
1778 int ret
, written
= 0;
1779 loff_t end
= pos
+ len
;
1780 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1781 struct ocfs2_dinode
*di
= NULL
;
1783 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi
->ip_blkno
,
1784 len
, (unsigned long long)pos
,
1785 oi
->ip_dyn_features
);
1788 * Handle inodes which already have inline data 1st.
1790 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1791 if (mmap_page
== NULL
&&
1792 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1793 goto do_inline_write
;
1796 * The write won't fit - we have to give this inode an
1797 * inline extent list now.
1799 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1806 * Check whether the inode can accept inline data.
1808 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1812 * Check whether the write can fit.
1814 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1816 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1820 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1827 * This signals to the caller that the data can be written
1832 return written
? written
: ret
;
1836 * This function only does anything for file systems which can't
1837 * handle sparse files.
1839 * What we want to do here is fill in any hole between the current end
1840 * of allocation and the end of our write. That way the rest of the
1841 * write path can treat it as an non-allocating write, which has no
1842 * special case code for sparse/nonsparse files.
1844 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1845 struct buffer_head
*di_bh
,
1846 loff_t pos
, unsigned len
,
1847 struct ocfs2_write_ctxt
*wc
)
1850 loff_t newsize
= pos
+ len
;
1852 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1854 if (newsize
<= i_size_read(inode
))
1857 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1861 wc
->w_first_new_cpos
=
1862 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
1867 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
1872 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1873 if (pos
> i_size_read(inode
))
1874 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
1880 * Try to flush truncate logs if we can free enough clusters from it.
1881 * As for return value, "< 0" means error, "0" no space and "1" means
1882 * we have freed enough spaces and let the caller try to allocate again.
1884 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
1885 unsigned int needed
)
1889 unsigned int truncated_clusters
;
1891 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
1892 truncated_clusters
= osb
->truncated_clusters
;
1893 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
1896 * Check whether we can succeed in allocating if we free
1899 if (truncated_clusters
< needed
)
1902 ret
= ocfs2_flush_truncate_log(osb
);
1908 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
1909 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
1916 int ocfs2_write_begin_nolock(struct file
*filp
,
1917 struct address_space
*mapping
,
1918 loff_t pos
, unsigned len
, unsigned flags
,
1919 struct page
**pagep
, void **fsdata
,
1920 struct buffer_head
*di_bh
, struct page
*mmap_page
)
1922 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
1923 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
1924 struct ocfs2_write_ctxt
*wc
;
1925 struct inode
*inode
= mapping
->host
;
1926 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1927 struct ocfs2_dinode
*di
;
1928 struct ocfs2_alloc_context
*data_ac
= NULL
;
1929 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1931 struct ocfs2_extent_tree et
;
1932 int try_free
= 1, ret1
;
1935 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
1941 if (ocfs2_supports_inline_data(osb
)) {
1942 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1954 if (ocfs2_sparse_alloc(osb
))
1955 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
1957 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
1964 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
1968 } else if (ret
== 1) {
1969 clusters_need
= wc
->w_clen
;
1970 ret
= ocfs2_refcount_cow(inode
, di_bh
,
1971 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
1978 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
1984 clusters_need
+= clusters_to_alloc
;
1986 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1988 trace_ocfs2_write_begin_nolock(
1989 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1990 (long long)i_size_read(inode
),
1991 le32_to_cpu(di
->i_clusters
),
1992 pos
, len
, flags
, mmap_page
,
1993 clusters_to_alloc
, extents_to_split
);
1996 * We set w_target_from, w_target_to here so that
1997 * ocfs2_write_end() knows which range in the target page to
1998 * write out. An allocation requires that we write the entire
2001 if (clusters_to_alloc
|| extents_to_split
) {
2003 * XXX: We are stretching the limits of
2004 * ocfs2_lock_allocators(). It greatly over-estimates
2005 * the work to be done.
2007 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
2009 ret
= ocfs2_lock_allocators(inode
, &et
,
2010 clusters_to_alloc
, extents_to_split
,
2011 &data_ac
, &meta_ac
);
2018 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
2020 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
2026 * We have to zero sparse allocated clusters, unwritten extent clusters,
2027 * and non-sparse clusters we just extended. For non-sparse writes,
2028 * we know zeros will only be needed in the first and/or last cluster.
2030 if (clusters_to_alloc
|| extents_to_split
||
2031 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
2032 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
2033 cluster_of_pages
= 1;
2035 cluster_of_pages
= 0;
2037 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
2039 handle
= ocfs2_start_trans(osb
, credits
);
2040 if (IS_ERR(handle
)) {
2041 ret
= PTR_ERR(handle
);
2046 wc
->w_handle
= handle
;
2048 if (clusters_to_alloc
) {
2049 ret
= dquot_alloc_space_nodirty(inode
,
2050 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
2055 * We don't want this to fail in ocfs2_write_end(), so do it
2058 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
2059 OCFS2_JOURNAL_ACCESS_WRITE
);
2066 * Fill our page array first. That way we've grabbed enough so
2067 * that we can zero and flush if we error after adding the
2070 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
2071 cluster_of_pages
, mmap_page
);
2072 if (ret
&& ret
!= -EAGAIN
) {
2078 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
2079 * the target page. In this case, we exit with no error and no target
2080 * page. This will trigger the caller, page_mkwrite(), to re-try
2083 if (ret
== -EAGAIN
) {
2084 BUG_ON(wc
->w_target_page
);
2089 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
2097 ocfs2_free_alloc_context(data_ac
);
2099 ocfs2_free_alloc_context(meta_ac
);
2102 *pagep
= wc
->w_target_page
;
2106 if (clusters_to_alloc
)
2107 dquot_free_space(inode
,
2108 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
2110 ocfs2_commit_trans(osb
, handle
);
2113 ocfs2_free_write_ctxt(wc
);
2116 ocfs2_free_alloc_context(data_ac
);
2120 ocfs2_free_alloc_context(meta_ac
);
2124 if (ret
== -ENOSPC
&& try_free
) {
2126 * Try to free some truncate log so that we can have enough
2127 * clusters to allocate.
2131 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
2142 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
2143 loff_t pos
, unsigned len
, unsigned flags
,
2144 struct page
**pagep
, void **fsdata
)
2147 struct buffer_head
*di_bh
= NULL
;
2148 struct inode
*inode
= mapping
->host
;
2150 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2157 * Take alloc sem here to prevent concurrent lookups. That way
2158 * the mapping, zeroing and tree manipulation within
2159 * ocfs2_write() will be safe against ->readpage(). This
2160 * should also serve to lock out allocation from a shared
2163 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2165 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
2166 fsdata
, di_bh
, NULL
);
2177 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2180 ocfs2_inode_unlock(inode
, 1);
2185 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
2186 unsigned len
, unsigned *copied
,
2187 struct ocfs2_dinode
*di
,
2188 struct ocfs2_write_ctxt
*wc
)
2192 if (unlikely(*copied
< len
)) {
2193 if (!PageUptodate(wc
->w_target_page
)) {
2199 kaddr
= kmap_atomic(wc
->w_target_page
);
2200 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
2201 kunmap_atomic(kaddr
);
2203 trace_ocfs2_write_end_inline(
2204 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
2205 (unsigned long long)pos
, *copied
,
2206 le16_to_cpu(di
->id2
.i_data
.id_count
),
2207 le16_to_cpu(di
->i_dyn_features
));
2210 int ocfs2_write_end_nolock(struct address_space
*mapping
,
2211 loff_t pos
, unsigned len
, unsigned copied
,
2212 struct page
*page
, void *fsdata
)
2215 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
2216 struct inode
*inode
= mapping
->host
;
2217 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2218 struct ocfs2_write_ctxt
*wc
= fsdata
;
2219 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
2220 handle_t
*handle
= wc
->w_handle
;
2221 struct page
*tmppage
;
2223 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2224 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
2225 goto out_write_size
;
2228 if (unlikely(copied
< len
)) {
2229 if (!PageUptodate(wc
->w_target_page
))
2232 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
2235 flush_dcache_page(wc
->w_target_page
);
2237 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
2238 tmppage
= wc
->w_pages
[i
];
2240 if (tmppage
== wc
->w_target_page
) {
2241 from
= wc
->w_target_from
;
2242 to
= wc
->w_target_to
;
2244 BUG_ON(from
> PAGE_CACHE_SIZE
||
2245 to
> PAGE_CACHE_SIZE
||
2249 * Pages adjacent to the target (if any) imply
2250 * a hole-filling write in which case we want
2251 * to flush their entire range.
2254 to
= PAGE_CACHE_SIZE
;
2257 if (page_has_buffers(tmppage
)) {
2258 if (ocfs2_should_order_data(inode
))
2259 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
2260 block_commit_write(tmppage
, from
, to
);
2266 if (pos
> i_size_read(inode
)) {
2267 i_size_write(inode
, pos
);
2268 mark_inode_dirty(inode
);
2270 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
2271 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
2272 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2273 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
2274 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2275 ocfs2_update_inode_fsync_trans(handle
, inode
, 1);
2276 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2278 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2279 * lock, or it will cause a deadlock since journal commit threads holds
2280 * this lock and will ask for the page lock when flushing the data.
2281 * put it here to preserve the unlock order.
2283 ocfs2_unlock_pages(wc
);
2285 ocfs2_commit_trans(osb
, handle
);
2287 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2289 brelse(wc
->w_di_bh
);
2295 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2296 loff_t pos
, unsigned len
, unsigned copied
,
2297 struct page
*page
, void *fsdata
)
2300 struct inode
*inode
= mapping
->host
;
2302 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2304 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2305 ocfs2_inode_unlock(inode
, 1);
2310 const struct address_space_operations ocfs2_aops
= {
2311 .readpage
= ocfs2_readpage
,
2312 .readpages
= ocfs2_readpages
,
2313 .writepage
= ocfs2_writepage
,
2314 .write_begin
= ocfs2_write_begin
,
2315 .write_end
= ocfs2_write_end
,
2317 .direct_IO
= ocfs2_direct_IO
,
2318 .invalidatepage
= block_invalidatepage
,
2319 .releasepage
= ocfs2_releasepage
,
2320 .migratepage
= buffer_migrate_page
,
2321 .is_partially_uptodate
= block_is_partially_uptodate
,
2322 .error_remove_page
= generic_error_remove_page
,