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 int ocfs2_direct_IO_zero_extend(struct ocfs2_super
*osb
,
668 struct inode
*inode
, loff_t offset
,
669 u64 zero_len
, int cluster_align
)
672 u32 v_cpos
= ocfs2_bytes_to_clusters(osb
->sb
, i_size_read(inode
));
673 unsigned int num_clusters
= 0;
674 unsigned int ext_flags
= 0;
677 if (offset
<= i_size_read(inode
) || cluster_align
)
680 ret
= ocfs2_get_clusters(inode
, v_cpos
, &p_cpos
, &num_clusters
,
687 if (p_cpos
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
)) {
688 u64 s
= i_size_read(inode
);
689 sector_t sector
= (p_cpos
<< (osb
->s_clustersize_bits
- 9)) +
690 (do_div(s
, osb
->s_clustersize
) >> 9);
692 ret
= blkdev_issue_zeroout(osb
->sb
->s_bdev
, sector
,
693 zero_len
>> 9, GFP_NOFS
, false);
701 static int ocfs2_direct_IO_extend_no_holes(struct ocfs2_super
*osb
,
702 struct inode
*inode
, loff_t offset
)
704 u64 zero_start
, zero_len
, total_zero_len
;
705 u32 p_cpos
= 0, clusters_to_add
;
706 u32 v_cpos
= ocfs2_bytes_to_clusters(osb
->sb
, i_size_read(inode
));
707 unsigned int num_clusters
= 0;
708 unsigned int ext_flags
= 0;
709 u32 size_div
, offset_div
;
714 u64 s
= i_size_read(inode
);
716 offset_div
= do_div(o
, osb
->s_clustersize
);
717 size_div
= do_div(s
, osb
->s_clustersize
);
720 if (offset
<= i_size_read(inode
))
723 clusters_to_add
= ocfs2_bytes_to_clusters(inode
->i_sb
, offset
) -
724 ocfs2_bytes_to_clusters(inode
->i_sb
, i_size_read(inode
));
725 total_zero_len
= offset
- i_size_read(inode
);
727 total_zero_len
-= offset_div
;
729 /* Allocate clusters to fill out holes, and this is only needed
730 * when we add more than one clusters. Otherwise the cluster will
731 * be allocated during direct IO */
732 if (clusters_to_add
> 1) {
733 ret
= ocfs2_extend_allocation(inode
,
734 OCFS2_I(inode
)->ip_clusters
,
735 clusters_to_add
- 1, 0);
742 while (total_zero_len
) {
743 ret
= ocfs2_get_clusters(inode
, v_cpos
, &p_cpos
, &num_clusters
,
750 zero_start
= ocfs2_clusters_to_bytes(osb
->sb
, p_cpos
) +
752 zero_len
= ocfs2_clusters_to_bytes(osb
->sb
, num_clusters
) -
754 zero_len
= min(total_zero_len
, zero_len
);
756 if (p_cpos
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
)) {
757 ret
= blkdev_issue_zeroout(osb
->sb
->s_bdev
,
758 zero_start
>> 9, zero_len
>> 9,
766 total_zero_len
-= zero_len
;
767 v_cpos
+= ocfs2_bytes_to_clusters(osb
->sb
, zero_len
+ size_div
);
769 /* Only at first iteration can be cluster not aligned.
770 * So set size_div to 0 for the rest */
778 static ssize_t
ocfs2_direct_IO_write(struct kiocb
*iocb
,
779 struct iov_iter
*iter
,
784 bool orphaned
= false;
785 int is_overwrite
= 0;
786 struct file
*file
= iocb
->ki_filp
;
787 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
788 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
789 struct buffer_head
*di_bh
= NULL
;
790 size_t count
= iter
->count
;
791 journal_t
*journal
= osb
->journal
->j_journal
;
792 u64 zero_len_head
, zero_len_tail
;
793 int cluster_align_head
, cluster_align_tail
;
794 loff_t final_size
= offset
+ count
;
795 int append_write
= offset
>= i_size_read(inode
) ? 1 : 0;
796 unsigned int num_clusters
= 0;
797 unsigned int ext_flags
= 0;
801 u64 s
= i_size_read(inode
);
803 zero_len_head
= do_div(o
, 1 << osb
->s_clustersize_bits
);
804 cluster_align_head
= !zero_len_head
;
806 zero_len_tail
= osb
->s_clustersize
-
807 do_div(s
, osb
->s_clustersize
);
808 if ((offset
- i_size_read(inode
)) < zero_len_tail
)
809 zero_len_tail
= offset
- i_size_read(inode
);
810 cluster_align_tail
= !zero_len_tail
;
814 * when final_size > inode->i_size, inode->i_size will be
815 * updated after direct write, so add the inode to orphan
818 if (final_size
> i_size_read(inode
)) {
819 ret
= ocfs2_add_inode_to_orphan(osb
, inode
);
828 ret
= ocfs2_inode_lock(inode
, NULL
, 1);
834 /* zeroing out the previously allocated cluster tail
835 * that but not zeroed */
836 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
837 ret
= ocfs2_direct_IO_zero_extend(osb
, inode
, offset
,
838 zero_len_tail
, cluster_align_tail
);
840 ret
= ocfs2_direct_IO_extend_no_holes(osb
, inode
,
844 ocfs2_inode_unlock(inode
, 1);
848 is_overwrite
= ocfs2_is_overwrite(osb
, inode
, offset
);
849 if (is_overwrite
< 0) {
850 mlog_errno(is_overwrite
);
851 ocfs2_inode_unlock(inode
, 1);
855 ocfs2_inode_unlock(inode
, 1);
858 written
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
859 offset
, ocfs2_direct_IO_get_blocks
,
860 ocfs2_dio_end_io
, NULL
, 0);
861 if (unlikely(written
< 0)) {
862 loff_t i_size
= i_size_read(inode
);
864 if (offset
+ count
> i_size
) {
865 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
871 if (i_size
== i_size_read(inode
)) {
872 ret
= ocfs2_truncate_file(inode
, di_bh
,
878 ocfs2_inode_unlock(inode
, 1);
884 ocfs2_inode_unlock(inode
, 1);
887 ret
= jbd2_journal_force_commit(journal
);
891 } else if (written
> 0 && append_write
&& !is_overwrite
&&
892 !cluster_align_head
) {
893 /* zeroing out the allocated cluster head */
895 u32 v_cpos
= ocfs2_bytes_to_clusters(osb
->sb
, offset
);
897 ret
= ocfs2_inode_lock(inode
, NULL
, 0);
903 ret
= ocfs2_get_clusters(inode
, v_cpos
, &p_cpos
,
904 &num_clusters
, &ext_flags
);
907 ocfs2_inode_unlock(inode
, 0);
911 BUG_ON(!p_cpos
|| (ext_flags
& OCFS2_EXT_UNWRITTEN
));
913 ret
= blkdev_issue_zeroout(osb
->sb
->s_bdev
,
914 p_cpos
<< (osb
->s_clustersize_bits
- 9),
915 zero_len_head
>> 9, GFP_NOFS
, false);
919 ocfs2_inode_unlock(inode
, 0);
925 int update_isize
= written
> 0 ? 1 : 0;
926 loff_t end
= update_isize
? offset
+ written
: 0;
928 tmp_ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
935 tmp_ret
= ocfs2_del_inode_from_orphan(osb
, inode
, di_bh
,
943 ocfs2_inode_unlock(inode
, 1);
945 tmp_ret
= jbd2_journal_force_commit(journal
);
958 static ssize_t
ocfs2_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
961 struct file
*file
= iocb
->ki_filp
;
962 struct inode
*inode
= file_inode(file
)->i_mapping
->host
;
963 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
964 int full_coherency
= !(osb
->s_mount_opt
&
965 OCFS2_MOUNT_COHERENCY_BUFFERED
);
968 * Fallback to buffered I/O if we see an inode without
971 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
974 /* Fallback to buffered I/O if we are appending and
975 * concurrent O_DIRECT writes are allowed.
977 if (i_size_read(inode
) <= offset
&& !full_coherency
)
980 if (iov_iter_rw(iter
) == READ
)
981 return __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
983 ocfs2_direct_IO_get_blocks
,
984 ocfs2_dio_end_io
, NULL
, 0);
986 return ocfs2_direct_IO_write(iocb
, iter
, offset
);
989 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super
*osb
,
994 unsigned int cluster_start
= 0, cluster_end
= PAGE_CACHE_SIZE
;
996 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
)) {
999 cpp
= 1 << (PAGE_CACHE_SHIFT
- osb
->s_clustersize_bits
);
1001 cluster_start
= cpos
% cpp
;
1002 cluster_start
= cluster_start
<< osb
->s_clustersize_bits
;
1004 cluster_end
= cluster_start
+ osb
->s_clustersize
;
1007 BUG_ON(cluster_start
> PAGE_SIZE
);
1008 BUG_ON(cluster_end
> PAGE_SIZE
);
1011 *start
= cluster_start
;
1017 * 'from' and 'to' are the region in the page to avoid zeroing.
1019 * If pagesize > clustersize, this function will avoid zeroing outside
1020 * of the cluster boundary.
1022 * from == to == 0 is code for "zero the entire cluster region"
1024 static void ocfs2_clear_page_regions(struct page
*page
,
1025 struct ocfs2_super
*osb
, u32 cpos
,
1026 unsigned from
, unsigned to
)
1029 unsigned int cluster_start
, cluster_end
;
1031 ocfs2_figure_cluster_boundaries(osb
, cpos
, &cluster_start
, &cluster_end
);
1033 kaddr
= kmap_atomic(page
);
1036 if (from
> cluster_start
)
1037 memset(kaddr
+ cluster_start
, 0, from
- cluster_start
);
1038 if (to
< cluster_end
)
1039 memset(kaddr
+ to
, 0, cluster_end
- to
);
1041 memset(kaddr
+ cluster_start
, 0, cluster_end
- cluster_start
);
1044 kunmap_atomic(kaddr
);
1048 * Nonsparse file systems fully allocate before we get to the write
1049 * code. This prevents ocfs2_write() from tagging the write as an
1050 * allocating one, which means ocfs2_map_page_blocks() might try to
1051 * read-in the blocks at the tail of our file. Avoid reading them by
1052 * testing i_size against each block offset.
1054 static int ocfs2_should_read_blk(struct inode
*inode
, struct page
*page
,
1055 unsigned int block_start
)
1057 u64 offset
= page_offset(page
) + block_start
;
1059 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
1062 if (i_size_read(inode
) > offset
)
1069 * Some of this taken from __block_write_begin(). We already have our
1070 * mapping by now though, and the entire write will be allocating or
1071 * it won't, so not much need to use BH_New.
1073 * This will also skip zeroing, which is handled externally.
1075 int ocfs2_map_page_blocks(struct page
*page
, u64
*p_blkno
,
1076 struct inode
*inode
, unsigned int from
,
1077 unsigned int to
, int new)
1080 struct buffer_head
*head
, *bh
, *wait
[2], **wait_bh
= wait
;
1081 unsigned int block_end
, block_start
;
1082 unsigned int bsize
= 1 << inode
->i_blkbits
;
1084 if (!page_has_buffers(page
))
1085 create_empty_buffers(page
, bsize
, 0);
1087 head
= page_buffers(page
);
1088 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1089 bh
= bh
->b_this_page
, block_start
+= bsize
) {
1090 block_end
= block_start
+ bsize
;
1092 clear_buffer_new(bh
);
1095 * Ignore blocks outside of our i/o range -
1096 * they may belong to unallocated clusters.
1098 if (block_start
>= to
|| block_end
<= from
) {
1099 if (PageUptodate(page
))
1100 set_buffer_uptodate(bh
);
1105 * For an allocating write with cluster size >= page
1106 * size, we always write the entire page.
1111 if (!buffer_mapped(bh
)) {
1112 map_bh(bh
, inode
->i_sb
, *p_blkno
);
1113 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
1116 if (PageUptodate(page
)) {
1117 if (!buffer_uptodate(bh
))
1118 set_buffer_uptodate(bh
);
1119 } else if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1121 ocfs2_should_read_blk(inode
, page
, block_start
) &&
1122 (block_start
< from
|| block_end
> to
)) {
1123 ll_rw_block(READ
, 1, &bh
);
1127 *p_blkno
= *p_blkno
+ 1;
1131 * If we issued read requests - let them complete.
1133 while(wait_bh
> wait
) {
1134 wait_on_buffer(*--wait_bh
);
1135 if (!buffer_uptodate(*wait_bh
))
1139 if (ret
== 0 || !new)
1143 * If we get -EIO above, zero out any newly allocated blocks
1144 * to avoid exposing stale data.
1149 block_end
= block_start
+ bsize
;
1150 if (block_end
<= from
)
1152 if (block_start
>= to
)
1155 zero_user(page
, block_start
, bh
->b_size
);
1156 set_buffer_uptodate(bh
);
1157 mark_buffer_dirty(bh
);
1160 block_start
= block_end
;
1161 bh
= bh
->b_this_page
;
1162 } while (bh
!= head
);
1167 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
1168 #define OCFS2_MAX_CTXT_PAGES 1
1170 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
1173 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
1176 * Describe the state of a single cluster to be written to.
1178 struct ocfs2_write_cluster_desc
{
1182 * Give this a unique field because c_phys eventually gets
1186 unsigned c_unwritten
;
1187 unsigned c_needs_zero
;
1190 struct ocfs2_write_ctxt
{
1191 /* Logical cluster position / len of write */
1195 /* First cluster allocated in a nonsparse extend */
1196 u32 w_first_new_cpos
;
1198 struct ocfs2_write_cluster_desc w_desc
[OCFS2_MAX_CLUSTERS_PER_PAGE
];
1201 * This is true if page_size > cluster_size.
1203 * It triggers a set of special cases during write which might
1204 * have to deal with allocating writes to partial pages.
1206 unsigned int w_large_pages
;
1209 * Pages involved in this write.
1211 * w_target_page is the page being written to by the user.
1213 * w_pages is an array of pages which always contains
1214 * w_target_page, and in the case of an allocating write with
1215 * page_size < cluster size, it will contain zero'd and mapped
1216 * pages adjacent to w_target_page which need to be written
1217 * out in so that future reads from that region will get
1220 unsigned int w_num_pages
;
1221 struct page
*w_pages
[OCFS2_MAX_CTXT_PAGES
];
1222 struct page
*w_target_page
;
1225 * w_target_locked is used for page_mkwrite path indicating no unlocking
1226 * against w_target_page in ocfs2_write_end_nolock.
1228 unsigned int w_target_locked
:1;
1231 * ocfs2_write_end() uses this to know what the real range to
1232 * write in the target should be.
1234 unsigned int w_target_from
;
1235 unsigned int w_target_to
;
1238 * We could use journal_current_handle() but this is cleaner,
1243 struct buffer_head
*w_di_bh
;
1245 struct ocfs2_cached_dealloc_ctxt w_dealloc
;
1248 void ocfs2_unlock_and_free_pages(struct page
**pages
, int num_pages
)
1252 for(i
= 0; i
< num_pages
; i
++) {
1254 unlock_page(pages
[i
]);
1255 mark_page_accessed(pages
[i
]);
1256 page_cache_release(pages
[i
]);
1261 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt
*wc
)
1266 * w_target_locked is only set to true in the page_mkwrite() case.
1267 * The intent is to allow us to lock the target page from write_begin()
1268 * to write_end(). The caller must hold a ref on w_target_page.
1270 if (wc
->w_target_locked
) {
1271 BUG_ON(!wc
->w_target_page
);
1272 for (i
= 0; i
< wc
->w_num_pages
; i
++) {
1273 if (wc
->w_target_page
== wc
->w_pages
[i
]) {
1274 wc
->w_pages
[i
] = NULL
;
1278 mark_page_accessed(wc
->w_target_page
);
1279 page_cache_release(wc
->w_target_page
);
1281 ocfs2_unlock_and_free_pages(wc
->w_pages
, wc
->w_num_pages
);
1284 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt
*wc
)
1286 ocfs2_unlock_pages(wc
);
1287 brelse(wc
->w_di_bh
);
1291 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt
**wcp
,
1292 struct ocfs2_super
*osb
, loff_t pos
,
1293 unsigned len
, struct buffer_head
*di_bh
)
1296 struct ocfs2_write_ctxt
*wc
;
1298 wc
= kzalloc(sizeof(struct ocfs2_write_ctxt
), GFP_NOFS
);
1302 wc
->w_cpos
= pos
>> osb
->s_clustersize_bits
;
1303 wc
->w_first_new_cpos
= UINT_MAX
;
1304 cend
= (pos
+ len
- 1) >> osb
->s_clustersize_bits
;
1305 wc
->w_clen
= cend
- wc
->w_cpos
+ 1;
1307 wc
->w_di_bh
= di_bh
;
1309 if (unlikely(PAGE_CACHE_SHIFT
> osb
->s_clustersize_bits
))
1310 wc
->w_large_pages
= 1;
1312 wc
->w_large_pages
= 0;
1314 ocfs2_init_dealloc_ctxt(&wc
->w_dealloc
);
1322 * If a page has any new buffers, zero them out here, and mark them uptodate
1323 * and dirty so they'll be written out (in order to prevent uninitialised
1324 * block data from leaking). And clear the new bit.
1326 static void ocfs2_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1328 unsigned int block_start
, block_end
;
1329 struct buffer_head
*head
, *bh
;
1331 BUG_ON(!PageLocked(page
));
1332 if (!page_has_buffers(page
))
1335 bh
= head
= page_buffers(page
);
1338 block_end
= block_start
+ bh
->b_size
;
1340 if (buffer_new(bh
)) {
1341 if (block_end
> from
&& block_start
< to
) {
1342 if (!PageUptodate(page
)) {
1343 unsigned start
, end
;
1345 start
= max(from
, block_start
);
1346 end
= min(to
, block_end
);
1348 zero_user_segment(page
, start
, end
);
1349 set_buffer_uptodate(bh
);
1352 clear_buffer_new(bh
);
1353 mark_buffer_dirty(bh
);
1357 block_start
= block_end
;
1358 bh
= bh
->b_this_page
;
1359 } while (bh
!= head
);
1363 * Only called when we have a failure during allocating write to write
1364 * zero's to the newly allocated region.
1366 static void ocfs2_write_failure(struct inode
*inode
,
1367 struct ocfs2_write_ctxt
*wc
,
1368 loff_t user_pos
, unsigned user_len
)
1371 unsigned from
= user_pos
& (PAGE_CACHE_SIZE
- 1),
1372 to
= user_pos
+ user_len
;
1373 struct page
*tmppage
;
1375 ocfs2_zero_new_buffers(wc
->w_target_page
, from
, to
);
1377 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1378 tmppage
= wc
->w_pages
[i
];
1380 if (page_has_buffers(tmppage
)) {
1381 if (ocfs2_should_order_data(inode
))
1382 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
1384 block_commit_write(tmppage
, from
, to
);
1389 static int ocfs2_prepare_page_for_write(struct inode
*inode
, u64
*p_blkno
,
1390 struct ocfs2_write_ctxt
*wc
,
1391 struct page
*page
, u32 cpos
,
1392 loff_t user_pos
, unsigned user_len
,
1396 unsigned int map_from
= 0, map_to
= 0;
1397 unsigned int cluster_start
, cluster_end
;
1398 unsigned int user_data_from
= 0, user_data_to
= 0;
1400 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode
->i_sb
), cpos
,
1401 &cluster_start
, &cluster_end
);
1403 /* treat the write as new if the a hole/lseek spanned across
1404 * the page boundary.
1406 new = new | ((i_size_read(inode
) <= page_offset(page
)) &&
1407 (page_offset(page
) <= user_pos
));
1409 if (page
== wc
->w_target_page
) {
1410 map_from
= user_pos
& (PAGE_CACHE_SIZE
- 1);
1411 map_to
= map_from
+ user_len
;
1414 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1415 cluster_start
, cluster_end
,
1418 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1419 map_from
, map_to
, new);
1425 user_data_from
= map_from
;
1426 user_data_to
= map_to
;
1428 map_from
= cluster_start
;
1429 map_to
= cluster_end
;
1433 * If we haven't allocated the new page yet, we
1434 * shouldn't be writing it out without copying user
1435 * data. This is likely a math error from the caller.
1439 map_from
= cluster_start
;
1440 map_to
= cluster_end
;
1442 ret
= ocfs2_map_page_blocks(page
, p_blkno
, inode
,
1443 cluster_start
, cluster_end
, new);
1451 * Parts of newly allocated pages need to be zero'd.
1453 * Above, we have also rewritten 'to' and 'from' - as far as
1454 * the rest of the function is concerned, the entire cluster
1455 * range inside of a page needs to be written.
1457 * We can skip this if the page is up to date - it's already
1458 * been zero'd from being read in as a hole.
1460 if (new && !PageUptodate(page
))
1461 ocfs2_clear_page_regions(page
, OCFS2_SB(inode
->i_sb
),
1462 cpos
, user_data_from
, user_data_to
);
1464 flush_dcache_page(page
);
1471 * This function will only grab one clusters worth of pages.
1473 static int ocfs2_grab_pages_for_write(struct address_space
*mapping
,
1474 struct ocfs2_write_ctxt
*wc
,
1475 u32 cpos
, loff_t user_pos
,
1476 unsigned user_len
, int new,
1477 struct page
*mmap_page
)
1480 unsigned long start
, target_index
, end_index
, index
;
1481 struct inode
*inode
= mapping
->host
;
1484 target_index
= user_pos
>> PAGE_CACHE_SHIFT
;
1487 * Figure out how many pages we'll be manipulating here. For
1488 * non allocating write, we just change the one
1489 * page. Otherwise, we'll need a whole clusters worth. If we're
1490 * writing past i_size, we only need enough pages to cover the
1491 * last page of the write.
1494 wc
->w_num_pages
= ocfs2_pages_per_cluster(inode
->i_sb
);
1495 start
= ocfs2_align_clusters_to_page_index(inode
->i_sb
, cpos
);
1497 * We need the index *past* the last page we could possibly
1498 * touch. This is the page past the end of the write or
1499 * i_size, whichever is greater.
1501 last_byte
= max(user_pos
+ user_len
, i_size_read(inode
));
1502 BUG_ON(last_byte
< 1);
1503 end_index
= ((last_byte
- 1) >> PAGE_CACHE_SHIFT
) + 1;
1504 if ((start
+ wc
->w_num_pages
) > end_index
)
1505 wc
->w_num_pages
= end_index
- start
;
1507 wc
->w_num_pages
= 1;
1508 start
= target_index
;
1511 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1514 if (index
== target_index
&& mmap_page
) {
1516 * ocfs2_pagemkwrite() is a little different
1517 * and wants us to directly use the page
1520 lock_page(mmap_page
);
1522 /* Exit and let the caller retry */
1523 if (mmap_page
->mapping
!= mapping
) {
1524 WARN_ON(mmap_page
->mapping
);
1525 unlock_page(mmap_page
);
1530 page_cache_get(mmap_page
);
1531 wc
->w_pages
[i
] = mmap_page
;
1532 wc
->w_target_locked
= true;
1534 wc
->w_pages
[i
] = find_or_create_page(mapping
, index
,
1536 if (!wc
->w_pages
[i
]) {
1542 wait_for_stable_page(wc
->w_pages
[i
]);
1544 if (index
== target_index
)
1545 wc
->w_target_page
= wc
->w_pages
[i
];
1549 wc
->w_target_locked
= false;
1554 * Prepare a single cluster for write one cluster into the file.
1556 static int ocfs2_write_cluster(struct address_space
*mapping
,
1557 u32 phys
, unsigned int unwritten
,
1558 unsigned int should_zero
,
1559 struct ocfs2_alloc_context
*data_ac
,
1560 struct ocfs2_alloc_context
*meta_ac
,
1561 struct ocfs2_write_ctxt
*wc
, u32 cpos
,
1562 loff_t user_pos
, unsigned user_len
)
1565 u64 v_blkno
, p_blkno
;
1566 struct inode
*inode
= mapping
->host
;
1567 struct ocfs2_extent_tree et
;
1569 new = phys
== 0 ? 1 : 0;
1574 * This is safe to call with the page locks - it won't take
1575 * any additional semaphores or cluster locks.
1578 ret
= ocfs2_add_inode_data(OCFS2_SB(inode
->i_sb
), inode
,
1579 &tmp_pos
, 1, 0, wc
->w_di_bh
,
1580 wc
->w_handle
, data_ac
,
1583 * This shouldn't happen because we must have already
1584 * calculated the correct meta data allocation required. The
1585 * internal tree allocation code should know how to increase
1586 * transaction credits itself.
1588 * If need be, we could handle -EAGAIN for a
1589 * RESTART_TRANS here.
1591 mlog_bug_on_msg(ret
== -EAGAIN
,
1592 "Inode %llu: EAGAIN return during allocation.\n",
1593 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1598 } else if (unwritten
) {
1599 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
1601 ret
= ocfs2_mark_extent_written(inode
, &et
,
1602 wc
->w_handle
, cpos
, 1, phys
,
1603 meta_ac
, &wc
->w_dealloc
);
1611 v_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, cpos
);
1613 v_blkno
= user_pos
>> inode
->i_sb
->s_blocksize_bits
;
1616 * The only reason this should fail is due to an inability to
1617 * find the extent added.
1619 ret
= ocfs2_extent_map_get_blocks(inode
, v_blkno
, &p_blkno
, NULL
,
1622 mlog(ML_ERROR
, "Get physical blkno failed for inode %llu, "
1623 "at logical block %llu",
1624 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1625 (unsigned long long)v_blkno
);
1629 BUG_ON(p_blkno
== 0);
1631 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
1634 tmpret
= ocfs2_prepare_page_for_write(inode
, &p_blkno
, wc
,
1635 wc
->w_pages
[i
], cpos
,
1646 * We only have cleanup to do in case of allocating write.
1649 ocfs2_write_failure(inode
, wc
, user_pos
, user_len
);
1656 static int ocfs2_write_cluster_by_desc(struct address_space
*mapping
,
1657 struct ocfs2_alloc_context
*data_ac
,
1658 struct ocfs2_alloc_context
*meta_ac
,
1659 struct ocfs2_write_ctxt
*wc
,
1660 loff_t pos
, unsigned len
)
1664 unsigned int local_len
= len
;
1665 struct ocfs2_write_cluster_desc
*desc
;
1666 struct ocfs2_super
*osb
= OCFS2_SB(mapping
->host
->i_sb
);
1668 for (i
= 0; i
< wc
->w_clen
; i
++) {
1669 desc
= &wc
->w_desc
[i
];
1672 * We have to make sure that the total write passed in
1673 * doesn't extend past a single cluster.
1676 cluster_off
= pos
& (osb
->s_clustersize
- 1);
1677 if ((cluster_off
+ local_len
) > osb
->s_clustersize
)
1678 local_len
= osb
->s_clustersize
- cluster_off
;
1680 ret
= ocfs2_write_cluster(mapping
, desc
->c_phys
,
1684 wc
, desc
->c_cpos
, pos
, local_len
);
1700 * ocfs2_write_end() wants to know which parts of the target page it
1701 * should complete the write on. It's easiest to compute them ahead of
1702 * time when a more complete view of the write is available.
1704 static void ocfs2_set_target_boundaries(struct ocfs2_super
*osb
,
1705 struct ocfs2_write_ctxt
*wc
,
1706 loff_t pos
, unsigned len
, int alloc
)
1708 struct ocfs2_write_cluster_desc
*desc
;
1710 wc
->w_target_from
= pos
& (PAGE_CACHE_SIZE
- 1);
1711 wc
->w_target_to
= wc
->w_target_from
+ len
;
1717 * Allocating write - we may have different boundaries based
1718 * on page size and cluster size.
1720 * NOTE: We can no longer compute one value from the other as
1721 * the actual write length and user provided length may be
1725 if (wc
->w_large_pages
) {
1727 * We only care about the 1st and last cluster within
1728 * our range and whether they should be zero'd or not. Either
1729 * value may be extended out to the start/end of a
1730 * newly allocated cluster.
1732 desc
= &wc
->w_desc
[0];
1733 if (desc
->c_needs_zero
)
1734 ocfs2_figure_cluster_boundaries(osb
,
1739 desc
= &wc
->w_desc
[wc
->w_clen
- 1];
1740 if (desc
->c_needs_zero
)
1741 ocfs2_figure_cluster_boundaries(osb
,
1746 wc
->w_target_from
= 0;
1747 wc
->w_target_to
= PAGE_CACHE_SIZE
;
1752 * Populate each single-cluster write descriptor in the write context
1753 * with information about the i/o to be done.
1755 * Returns the number of clusters that will have to be allocated, as
1756 * well as a worst case estimate of the number of extent records that
1757 * would have to be created during a write to an unwritten region.
1759 static int ocfs2_populate_write_desc(struct inode
*inode
,
1760 struct ocfs2_write_ctxt
*wc
,
1761 unsigned int *clusters_to_alloc
,
1762 unsigned int *extents_to_split
)
1765 struct ocfs2_write_cluster_desc
*desc
;
1766 unsigned int num_clusters
= 0;
1767 unsigned int ext_flags
= 0;
1771 *clusters_to_alloc
= 0;
1772 *extents_to_split
= 0;
1774 for (i
= 0; i
< wc
->w_clen
; i
++) {
1775 desc
= &wc
->w_desc
[i
];
1776 desc
->c_cpos
= wc
->w_cpos
+ i
;
1778 if (num_clusters
== 0) {
1780 * Need to look up the next extent record.
1782 ret
= ocfs2_get_clusters(inode
, desc
->c_cpos
, &phys
,
1783 &num_clusters
, &ext_flags
);
1789 /* We should already CoW the refcountd extent. */
1790 BUG_ON(ext_flags
& OCFS2_EXT_REFCOUNTED
);
1793 * Assume worst case - that we're writing in
1794 * the middle of the extent.
1796 * We can assume that the write proceeds from
1797 * left to right, in which case the extent
1798 * insert code is smart enough to coalesce the
1799 * next splits into the previous records created.
1801 if (ext_flags
& OCFS2_EXT_UNWRITTEN
)
1802 *extents_to_split
= *extents_to_split
+ 2;
1805 * Only increment phys if it doesn't describe
1812 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1813 * file that got extended. w_first_new_cpos tells us
1814 * where the newly allocated clusters are so we can
1817 if (desc
->c_cpos
>= wc
->w_first_new_cpos
) {
1819 desc
->c_needs_zero
= 1;
1822 desc
->c_phys
= phys
;
1825 desc
->c_needs_zero
= 1;
1826 *clusters_to_alloc
= *clusters_to_alloc
+ 1;
1829 if (ext_flags
& OCFS2_EXT_UNWRITTEN
) {
1830 desc
->c_unwritten
= 1;
1831 desc
->c_needs_zero
= 1;
1842 static int ocfs2_write_begin_inline(struct address_space
*mapping
,
1843 struct inode
*inode
,
1844 struct ocfs2_write_ctxt
*wc
)
1847 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1850 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1852 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1853 if (IS_ERR(handle
)) {
1854 ret
= PTR_ERR(handle
);
1859 page
= find_or_create_page(mapping
, 0, GFP_NOFS
);
1861 ocfs2_commit_trans(osb
, handle
);
1867 * If we don't set w_num_pages then this page won't get unlocked
1868 * and freed on cleanup of the write context.
1870 wc
->w_pages
[0] = wc
->w_target_page
= page
;
1871 wc
->w_num_pages
= 1;
1873 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
1874 OCFS2_JOURNAL_ACCESS_WRITE
);
1876 ocfs2_commit_trans(osb
, handle
);
1882 if (!(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
))
1883 ocfs2_set_inode_data_inline(inode
, di
);
1885 if (!PageUptodate(page
)) {
1886 ret
= ocfs2_read_inline_data(inode
, page
, wc
->w_di_bh
);
1888 ocfs2_commit_trans(osb
, handle
);
1894 wc
->w_handle
= handle
;
1899 int ocfs2_size_fits_inline_data(struct buffer_head
*di_bh
, u64 new_size
)
1901 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1903 if (new_size
<= le16_to_cpu(di
->id2
.i_data
.id_count
))
1908 static int ocfs2_try_to_write_inline_data(struct address_space
*mapping
,
1909 struct inode
*inode
, loff_t pos
,
1910 unsigned len
, struct page
*mmap_page
,
1911 struct ocfs2_write_ctxt
*wc
)
1913 int ret
, written
= 0;
1914 loff_t end
= pos
+ len
;
1915 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1916 struct ocfs2_dinode
*di
= NULL
;
1918 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi
->ip_blkno
,
1919 len
, (unsigned long long)pos
,
1920 oi
->ip_dyn_features
);
1923 * Handle inodes which already have inline data 1st.
1925 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1926 if (mmap_page
== NULL
&&
1927 ocfs2_size_fits_inline_data(wc
->w_di_bh
, end
))
1928 goto do_inline_write
;
1931 * The write won't fit - we have to give this inode an
1932 * inline extent list now.
1934 ret
= ocfs2_convert_inline_data_to_extents(inode
, wc
->w_di_bh
);
1941 * Check whether the inode can accept inline data.
1943 if (oi
->ip_clusters
!= 0 || i_size_read(inode
) != 0)
1947 * Check whether the write can fit.
1949 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
1951 end
> ocfs2_max_inline_data_with_xattr(inode
->i_sb
, di
))
1955 ret
= ocfs2_write_begin_inline(mapping
, inode
, wc
);
1962 * This signals to the caller that the data can be written
1967 return written
? written
: ret
;
1971 * This function only does anything for file systems which can't
1972 * handle sparse files.
1974 * What we want to do here is fill in any hole between the current end
1975 * of allocation and the end of our write. That way the rest of the
1976 * write path can treat it as an non-allocating write, which has no
1977 * special case code for sparse/nonsparse files.
1979 static int ocfs2_expand_nonsparse_inode(struct inode
*inode
,
1980 struct buffer_head
*di_bh
,
1981 loff_t pos
, unsigned len
,
1982 struct ocfs2_write_ctxt
*wc
)
1985 loff_t newsize
= pos
+ len
;
1987 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
1989 if (newsize
<= i_size_read(inode
))
1992 ret
= ocfs2_extend_no_holes(inode
, di_bh
, newsize
, pos
);
1996 wc
->w_first_new_cpos
=
1997 ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
2002 static int ocfs2_zero_tail(struct inode
*inode
, struct buffer_head
*di_bh
,
2007 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)));
2008 if (pos
> i_size_read(inode
))
2009 ret
= ocfs2_zero_extend(inode
, di_bh
, pos
);
2015 * Try to flush truncate logs if we can free enough clusters from it.
2016 * As for return value, "< 0" means error, "0" no space and "1" means
2017 * we have freed enough spaces and let the caller try to allocate again.
2019 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super
*osb
,
2020 unsigned int needed
)
2024 unsigned int truncated_clusters
;
2026 mutex_lock(&osb
->osb_tl_inode
->i_mutex
);
2027 truncated_clusters
= osb
->truncated_clusters
;
2028 mutex_unlock(&osb
->osb_tl_inode
->i_mutex
);
2031 * Check whether we can succeed in allocating if we free
2034 if (truncated_clusters
< needed
)
2037 ret
= ocfs2_flush_truncate_log(osb
);
2043 if (jbd2_journal_start_commit(osb
->journal
->j_journal
, &target
)) {
2044 jbd2_log_wait_commit(osb
->journal
->j_journal
, target
);
2051 int ocfs2_write_begin_nolock(struct file
*filp
,
2052 struct address_space
*mapping
,
2053 loff_t pos
, unsigned len
, unsigned flags
,
2054 struct page
**pagep
, void **fsdata
,
2055 struct buffer_head
*di_bh
, struct page
*mmap_page
)
2057 int ret
, cluster_of_pages
, credits
= OCFS2_INODE_UPDATE_CREDITS
;
2058 unsigned int clusters_to_alloc
, extents_to_split
, clusters_need
= 0;
2059 struct ocfs2_write_ctxt
*wc
;
2060 struct inode
*inode
= mapping
->host
;
2061 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2062 struct ocfs2_dinode
*di
;
2063 struct ocfs2_alloc_context
*data_ac
= NULL
;
2064 struct ocfs2_alloc_context
*meta_ac
= NULL
;
2066 struct ocfs2_extent_tree et
;
2067 int try_free
= 1, ret1
;
2070 ret
= ocfs2_alloc_write_ctxt(&wc
, osb
, pos
, len
, di_bh
);
2076 if (ocfs2_supports_inline_data(osb
)) {
2077 ret
= ocfs2_try_to_write_inline_data(mapping
, inode
, pos
, len
,
2089 if (ocfs2_sparse_alloc(osb
))
2090 ret
= ocfs2_zero_tail(inode
, di_bh
, pos
);
2092 ret
= ocfs2_expand_nonsparse_inode(inode
, di_bh
, pos
, len
,
2099 ret
= ocfs2_check_range_for_refcount(inode
, pos
, len
);
2103 } else if (ret
== 1) {
2104 clusters_need
= wc
->w_clen
;
2105 ret
= ocfs2_refcount_cow(inode
, di_bh
,
2106 wc
->w_cpos
, wc
->w_clen
, UINT_MAX
);
2113 ret
= ocfs2_populate_write_desc(inode
, wc
, &clusters_to_alloc
,
2119 clusters_need
+= clusters_to_alloc
;
2121 di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
2123 trace_ocfs2_write_begin_nolock(
2124 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
2125 (long long)i_size_read(inode
),
2126 le32_to_cpu(di
->i_clusters
),
2127 pos
, len
, flags
, mmap_page
,
2128 clusters_to_alloc
, extents_to_split
);
2131 * We set w_target_from, w_target_to here so that
2132 * ocfs2_write_end() knows which range in the target page to
2133 * write out. An allocation requires that we write the entire
2136 if (clusters_to_alloc
|| extents_to_split
) {
2138 * XXX: We are stretching the limits of
2139 * ocfs2_lock_allocators(). It greatly over-estimates
2140 * the work to be done.
2142 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
),
2144 ret
= ocfs2_lock_allocators(inode
, &et
,
2145 clusters_to_alloc
, extents_to_split
,
2146 &data_ac
, &meta_ac
);
2153 data_ac
->ac_resv
= &OCFS2_I(inode
)->ip_la_data_resv
;
2155 credits
= ocfs2_calc_extend_credits(inode
->i_sb
,
2161 * We have to zero sparse allocated clusters, unwritten extent clusters,
2162 * and non-sparse clusters we just extended. For non-sparse writes,
2163 * we know zeros will only be needed in the first and/or last cluster.
2165 if (clusters_to_alloc
|| extents_to_split
||
2166 (wc
->w_clen
&& (wc
->w_desc
[0].c_needs_zero
||
2167 wc
->w_desc
[wc
->w_clen
- 1].c_needs_zero
)))
2168 cluster_of_pages
= 1;
2170 cluster_of_pages
= 0;
2172 ocfs2_set_target_boundaries(osb
, wc
, pos
, len
, cluster_of_pages
);
2174 handle
= ocfs2_start_trans(osb
, credits
);
2175 if (IS_ERR(handle
)) {
2176 ret
= PTR_ERR(handle
);
2181 wc
->w_handle
= handle
;
2183 if (clusters_to_alloc
) {
2184 ret
= dquot_alloc_space_nodirty(inode
,
2185 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
2190 * We don't want this to fail in ocfs2_write_end(), so do it
2193 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), wc
->w_di_bh
,
2194 OCFS2_JOURNAL_ACCESS_WRITE
);
2201 * Fill our page array first. That way we've grabbed enough so
2202 * that we can zero and flush if we error after adding the
2205 ret
= ocfs2_grab_pages_for_write(mapping
, wc
, wc
->w_cpos
, pos
, len
,
2206 cluster_of_pages
, mmap_page
);
2207 if (ret
&& ret
!= -EAGAIN
) {
2213 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
2214 * the target page. In this case, we exit with no error and no target
2215 * page. This will trigger the caller, page_mkwrite(), to re-try
2218 if (ret
== -EAGAIN
) {
2219 BUG_ON(wc
->w_target_page
);
2224 ret
= ocfs2_write_cluster_by_desc(mapping
, data_ac
, meta_ac
, wc
, pos
,
2232 ocfs2_free_alloc_context(data_ac
);
2234 ocfs2_free_alloc_context(meta_ac
);
2237 *pagep
= wc
->w_target_page
;
2241 if (clusters_to_alloc
)
2242 dquot_free_space(inode
,
2243 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_alloc
));
2245 ocfs2_commit_trans(osb
, handle
);
2248 ocfs2_free_write_ctxt(wc
);
2251 ocfs2_free_alloc_context(data_ac
);
2255 ocfs2_free_alloc_context(meta_ac
);
2259 if (ret
== -ENOSPC
&& try_free
) {
2261 * Try to free some truncate log so that we can have enough
2262 * clusters to allocate.
2266 ret1
= ocfs2_try_to_free_truncate_log(osb
, clusters_need
);
2277 static int ocfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
2278 loff_t pos
, unsigned len
, unsigned flags
,
2279 struct page
**pagep
, void **fsdata
)
2282 struct buffer_head
*di_bh
= NULL
;
2283 struct inode
*inode
= mapping
->host
;
2285 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2292 * Take alloc sem here to prevent concurrent lookups. That way
2293 * the mapping, zeroing and tree manipulation within
2294 * ocfs2_write() will be safe against ->readpage(). This
2295 * should also serve to lock out allocation from a shared
2298 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2300 ret
= ocfs2_write_begin_nolock(file
, mapping
, pos
, len
, flags
, pagep
,
2301 fsdata
, di_bh
, NULL
);
2312 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2315 ocfs2_inode_unlock(inode
, 1);
2320 static void ocfs2_write_end_inline(struct inode
*inode
, loff_t pos
,
2321 unsigned len
, unsigned *copied
,
2322 struct ocfs2_dinode
*di
,
2323 struct ocfs2_write_ctxt
*wc
)
2327 if (unlikely(*copied
< len
)) {
2328 if (!PageUptodate(wc
->w_target_page
)) {
2334 kaddr
= kmap_atomic(wc
->w_target_page
);
2335 memcpy(di
->id2
.i_data
.id_data
+ pos
, kaddr
+ pos
, *copied
);
2336 kunmap_atomic(kaddr
);
2338 trace_ocfs2_write_end_inline(
2339 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
2340 (unsigned long long)pos
, *copied
,
2341 le16_to_cpu(di
->id2
.i_data
.id_count
),
2342 le16_to_cpu(di
->i_dyn_features
));
2345 int ocfs2_write_end_nolock(struct address_space
*mapping
,
2346 loff_t pos
, unsigned len
, unsigned copied
,
2347 struct page
*page
, void *fsdata
)
2350 unsigned from
, to
, start
= pos
& (PAGE_CACHE_SIZE
- 1);
2351 struct inode
*inode
= mapping
->host
;
2352 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2353 struct ocfs2_write_ctxt
*wc
= fsdata
;
2354 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)wc
->w_di_bh
->b_data
;
2355 handle_t
*handle
= wc
->w_handle
;
2356 struct page
*tmppage
;
2358 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2359 ocfs2_write_end_inline(inode
, pos
, len
, &copied
, di
, wc
);
2360 goto out_write_size
;
2363 if (unlikely(copied
< len
)) {
2364 if (!PageUptodate(wc
->w_target_page
))
2367 ocfs2_zero_new_buffers(wc
->w_target_page
, start
+copied
,
2370 flush_dcache_page(wc
->w_target_page
);
2372 for(i
= 0; i
< wc
->w_num_pages
; i
++) {
2373 tmppage
= wc
->w_pages
[i
];
2375 if (tmppage
== wc
->w_target_page
) {
2376 from
= wc
->w_target_from
;
2377 to
= wc
->w_target_to
;
2379 BUG_ON(from
> PAGE_CACHE_SIZE
||
2380 to
> PAGE_CACHE_SIZE
||
2384 * Pages adjacent to the target (if any) imply
2385 * a hole-filling write in which case we want
2386 * to flush their entire range.
2389 to
= PAGE_CACHE_SIZE
;
2392 if (page_has_buffers(tmppage
)) {
2393 if (ocfs2_should_order_data(inode
))
2394 ocfs2_jbd2_file_inode(wc
->w_handle
, inode
);
2395 block_commit_write(tmppage
, from
, to
);
2401 if (pos
> i_size_read(inode
)) {
2402 i_size_write(inode
, pos
);
2403 mark_inode_dirty(inode
);
2405 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
2406 di
->i_size
= cpu_to_le64((u64
)i_size_read(inode
));
2407 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2408 di
->i_mtime
= di
->i_ctime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
2409 di
->i_mtime_nsec
= di
->i_ctime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
2410 ocfs2_update_inode_fsync_trans(handle
, inode
, 1);
2411 ocfs2_journal_dirty(handle
, wc
->w_di_bh
);
2413 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2414 * lock, or it will cause a deadlock since journal commit threads holds
2415 * this lock and will ask for the page lock when flushing the data.
2416 * put it here to preserve the unlock order.
2418 ocfs2_unlock_pages(wc
);
2420 ocfs2_commit_trans(osb
, handle
);
2422 ocfs2_run_deallocs(osb
, &wc
->w_dealloc
);
2424 brelse(wc
->w_di_bh
);
2430 static int ocfs2_write_end(struct file
*file
, struct address_space
*mapping
,
2431 loff_t pos
, unsigned len
, unsigned copied
,
2432 struct page
*page
, void *fsdata
)
2435 struct inode
*inode
= mapping
->host
;
2437 ret
= ocfs2_write_end_nolock(mapping
, pos
, len
, copied
, page
, fsdata
);
2439 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
2440 ocfs2_inode_unlock(inode
, 1);
2445 const struct address_space_operations ocfs2_aops
= {
2446 .readpage
= ocfs2_readpage
,
2447 .readpages
= ocfs2_readpages
,
2448 .writepage
= ocfs2_writepage
,
2449 .write_begin
= ocfs2_write_begin
,
2450 .write_end
= ocfs2_write_end
,
2452 .direct_IO
= ocfs2_direct_IO
,
2453 .invalidatepage
= block_invalidatepage
,
2454 .releasepage
= ocfs2_releasepage
,
2455 .migratepage
= buffer_migrate_page
,
2456 .is_partially_uptodate
= block_is_partially_uptodate
,
2457 .error_remove_page
= generic_error_remove_page
,