1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * File open, close, extend, truncate
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
26 #include <linux/capability.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/pagemap.h>
32 #include <linux/uio.h>
33 #include <linux/sched.h>
34 #include <linux/splice.h>
35 #include <linux/mount.h>
36 #include <linux/writeback.h>
38 #define MLOG_MASK_PREFIX ML_INODE
39 #include <cluster/masklog.h>
47 #include "extent_map.h"
57 #include "buffer_head_io.h"
59 static int ocfs2_sync_inode(struct inode
*inode
)
61 filemap_fdatawrite(inode
->i_mapping
);
62 return sync_mapping_buffers(inode
->i_mapping
);
65 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
68 int mode
= file
->f_flags
;
69 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
71 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
72 file
->f_path
.dentry
->d_name
.len
, file
->f_path
.dentry
->d_name
.name
);
74 spin_lock(&oi
->ip_lock
);
76 /* Check that the inode hasn't been wiped from disk by another
77 * node. If it hasn't then we're safe as long as we hold the
78 * spin lock until our increment of open count. */
79 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
80 spin_unlock(&oi
->ip_lock
);
87 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
90 spin_unlock(&oi
->ip_lock
);
97 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
99 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
101 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
102 file
->f_path
.dentry
->d_name
.len
,
103 file
->f_path
.dentry
->d_name
.name
);
105 spin_lock(&oi
->ip_lock
);
106 if (!--oi
->ip_open_count
)
107 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
108 spin_unlock(&oi
->ip_lock
);
115 static int ocfs2_sync_file(struct file
*file
,
116 struct dentry
*dentry
,
121 struct inode
*inode
= dentry
->d_inode
;
122 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
124 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
125 dentry
->d_name
.len
, dentry
->d_name
.name
);
127 err
= ocfs2_sync_inode(dentry
->d_inode
);
131 journal
= osb
->journal
->j_journal
;
132 err
= journal_force_commit(journal
);
137 return (err
< 0) ? -EIO
: 0;
140 int ocfs2_should_update_atime(struct inode
*inode
,
141 struct vfsmount
*vfsmnt
)
144 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
146 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
149 if ((inode
->i_flags
& S_NOATIME
) ||
150 ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
154 * We can be called with no vfsmnt structure - NFSD will
157 * Note that our action here is different than touch_atime() -
158 * if we can't tell whether this is a noatime mount, then we
159 * don't know whether to trust the value of s_atime_quantum.
164 if ((vfsmnt
->mnt_flags
& MNT_NOATIME
) ||
165 ((vfsmnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
168 if (vfsmnt
->mnt_flags
& MNT_RELATIME
) {
169 if ((timespec_compare(&inode
->i_atime
, &inode
->i_mtime
) <= 0) ||
170 (timespec_compare(&inode
->i_atime
, &inode
->i_ctime
) <= 0))
177 if ((now
.tv_sec
- inode
->i_atime
.tv_sec
<= osb
->s_atime_quantum
))
183 int ocfs2_update_inode_atime(struct inode
*inode
,
184 struct buffer_head
*bh
)
187 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
192 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
193 if (handle
== NULL
) {
199 inode
->i_atime
= CURRENT_TIME
;
200 ret
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
204 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
210 static int ocfs2_set_inode_size(handle_t
*handle
,
212 struct buffer_head
*fe_bh
,
218 i_size_write(inode
, new_i_size
);
219 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
220 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
222 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
233 static int ocfs2_simple_size_update(struct inode
*inode
,
234 struct buffer_head
*di_bh
,
238 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
239 handle_t
*handle
= NULL
;
241 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
242 if (handle
== NULL
) {
248 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
253 ocfs2_commit_trans(osb
, handle
);
258 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
260 struct buffer_head
*fe_bh
,
265 struct ocfs2_dinode
*di
;
270 /* TODO: This needs to actually orphan the inode in this
273 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
274 if (IS_ERR(handle
)) {
275 status
= PTR_ERR(handle
);
280 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
281 OCFS2_JOURNAL_ACCESS_WRITE
);
288 * Do this before setting i_size.
290 cluster_bytes
= ocfs2_align_bytes_to_clusters(inode
->i_sb
, new_i_size
);
291 status
= ocfs2_zero_range_for_truncate(inode
, handle
, new_i_size
,
298 i_size_write(inode
, new_i_size
);
299 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(new_i_size
);
300 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
302 di
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
303 di
->i_size
= cpu_to_le64(new_i_size
);
304 di
->i_ctime
= di
->i_mtime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
305 di
->i_ctime_nsec
= di
->i_mtime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
307 status
= ocfs2_journal_dirty(handle
, fe_bh
);
312 ocfs2_commit_trans(osb
, handle
);
319 static int ocfs2_truncate_file(struct inode
*inode
,
320 struct buffer_head
*di_bh
,
324 struct ocfs2_dinode
*fe
= NULL
;
325 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
326 struct ocfs2_truncate_context
*tc
= NULL
;
328 mlog_entry("(inode = %llu, new_i_size = %llu\n",
329 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
330 (unsigned long long)new_i_size
);
332 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
333 if (!OCFS2_IS_VALID_DINODE(fe
)) {
334 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
339 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
340 "Inode %llu, inode i_size = %lld != di "
341 "i_size = %llu, i_flags = 0x%x\n",
342 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
344 (unsigned long long)le64_to_cpu(fe
->i_size
),
345 le32_to_cpu(fe
->i_flags
));
347 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
348 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
349 (unsigned long long)le64_to_cpu(fe
->i_size
),
350 (unsigned long long)new_i_size
);
356 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
357 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
358 (unsigned long long)le64_to_cpu(fe
->i_size
),
359 (unsigned long long)new_i_size
);
361 /* lets handle the simple truncate cases before doing any more
362 * cluster locking. */
363 if (new_i_size
== le64_to_cpu(fe
->i_size
))
366 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
368 /* This forces other nodes to sync and drop their pages. Do
369 * this even if we have a truncate without allocation change -
370 * ocfs2 cluster sizes can be much greater than page size, so
371 * we have to truncate them anyway. */
372 status
= ocfs2_data_lock(inode
, 1);
374 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
380 unmap_mapping_range(inode
->i_mapping
, new_i_size
+ PAGE_SIZE
- 1, 0, 1);
381 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
383 /* alright, we're going to need to do a full blown alloc size
384 * change. Orphan the inode so that recovery can complete the
385 * truncate if necessary. This does the task of marking
387 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
390 goto bail_unlock_data
;
393 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
396 goto bail_unlock_data
;
399 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
402 goto bail_unlock_data
;
405 /* TODO: orphan dir cleanup here. */
407 ocfs2_data_unlock(inode
, 1);
409 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
418 * extend allocation only here.
419 * we'll update all the disk stuff, and oip->alloc_size
421 * expect stuff to be locked, a transaction started and enough data /
422 * metadata reservations in the contexts.
424 * Will return -EAGAIN, and a reason if a restart is needed.
425 * If passed in, *reason will always be set, even in error.
427 int ocfs2_do_extend_allocation(struct ocfs2_super
*osb
,
432 struct buffer_head
*fe_bh
,
434 struct ocfs2_alloc_context
*data_ac
,
435 struct ocfs2_alloc_context
*meta_ac
,
436 enum ocfs2_alloc_restarted
*reason_ret
)
440 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
441 enum ocfs2_alloc_restarted reason
= RESTART_NONE
;
442 u32 bit_off
, num_bits
;
446 BUG_ON(!clusters_to_add
);
449 flags
= OCFS2_EXT_UNWRITTEN
;
451 free_extents
= ocfs2_num_free_extents(osb
, inode
, fe
);
452 if (free_extents
< 0) {
453 status
= free_extents
;
458 /* there are two cases which could cause us to EAGAIN in the
459 * we-need-more-metadata case:
460 * 1) we haven't reserved *any*
461 * 2) we are so fragmented, we've needed to add metadata too
463 if (!free_extents
&& !meta_ac
) {
464 mlog(0, "we haven't reserved any metadata!\n");
466 reason
= RESTART_META
;
468 } else if ((!free_extents
)
469 && (ocfs2_alloc_context_bits_left(meta_ac
)
470 < ocfs2_extend_meta_needed(fe
))) {
471 mlog(0, "filesystem is really fragmented...\n");
473 reason
= RESTART_META
;
477 status
= ocfs2_claim_clusters(osb
, handle
, data_ac
, 1,
478 &bit_off
, &num_bits
);
480 if (status
!= -ENOSPC
)
485 BUG_ON(num_bits
> clusters_to_add
);
487 /* reserve our write early -- insert_extent may update the inode */
488 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
489 OCFS2_JOURNAL_ACCESS_WRITE
);
495 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
496 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
497 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
498 status
= ocfs2_insert_extent(osb
, handle
, inode
, fe_bh
,
499 *logical_offset
, block
, num_bits
,
506 status
= ocfs2_journal_dirty(handle
, fe_bh
);
512 clusters_to_add
-= num_bits
;
513 *logical_offset
+= num_bits
;
515 if (clusters_to_add
) {
516 mlog(0, "need to alloc once more, clusters = %u, wanted = "
517 "%u\n", fe
->i_clusters
, clusters_to_add
);
519 reason
= RESTART_TRANS
;
525 *reason_ret
= reason
;
530 * For a given allocation, determine which allocators will need to be
531 * accessed, and lock them, reserving the appropriate number of bits.
533 * Sparse file systems call this from ocfs2_write_begin_nolock()
534 * and ocfs2_allocate_unwritten_extents().
536 * File systems which don't support holes call this from
537 * ocfs2_extend_allocation().
539 int ocfs2_lock_allocators(struct inode
*inode
, struct ocfs2_dinode
*di
,
540 u32 clusters_to_add
, u32 extents_to_split
,
541 struct ocfs2_alloc_context
**data_ac
,
542 struct ocfs2_alloc_context
**meta_ac
)
544 int ret
= 0, num_free_extents
;
545 unsigned int max_recs_needed
= clusters_to_add
+ 2 * extents_to_split
;
546 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
552 BUG_ON(clusters_to_add
!= 0 && data_ac
== NULL
);
554 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
555 "clusters_to_add = %u, extents_to_split = %u\n",
556 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, i_size_read(inode
),
557 le32_to_cpu(di
->i_clusters
), clusters_to_add
, extents_to_split
);
559 num_free_extents
= ocfs2_num_free_extents(osb
, inode
, di
);
560 if (num_free_extents
< 0) {
561 ret
= num_free_extents
;
567 * Sparse allocation file systems need to be more conservative
568 * with reserving room for expansion - the actual allocation
569 * happens while we've got a journal handle open so re-taking
570 * a cluster lock (because we ran out of room for another
571 * extent) will violate ordering rules.
573 * Most of the time we'll only be seeing this 1 cluster at a time
576 * Always lock for any unwritten extents - we might want to
577 * add blocks during a split.
579 if (!num_free_extents
||
580 (ocfs2_sparse_alloc(osb
) && num_free_extents
< max_recs_needed
)) {
581 ret
= ocfs2_reserve_new_metadata(osb
, di
, meta_ac
);
589 if (clusters_to_add
== 0)
592 ret
= ocfs2_reserve_clusters(osb
, clusters_to_add
, data_ac
);
602 ocfs2_free_alloc_context(*meta_ac
);
607 * We cannot have an error and a non null *data_ac.
614 static int __ocfs2_extend_allocation(struct inode
*inode
, u32 logical_start
,
615 u32 clusters_to_add
, int mark_unwritten
)
618 int restart_func
= 0;
621 struct buffer_head
*bh
= NULL
;
622 struct ocfs2_dinode
*fe
= NULL
;
623 handle_t
*handle
= NULL
;
624 struct ocfs2_alloc_context
*data_ac
= NULL
;
625 struct ocfs2_alloc_context
*meta_ac
= NULL
;
626 enum ocfs2_alloc_restarted why
;
627 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
629 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
632 * This function only exists for file systems which don't
635 BUG_ON(mark_unwritten
&& !ocfs2_sparse_alloc(osb
));
637 status
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &bh
,
638 OCFS2_BH_CACHED
, inode
);
644 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
645 if (!OCFS2_IS_VALID_DINODE(fe
)) {
646 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
652 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
654 status
= ocfs2_lock_allocators(inode
, fe
, clusters_to_add
, 0, &data_ac
,
661 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
662 handle
= ocfs2_start_trans(osb
, credits
);
663 if (IS_ERR(handle
)) {
664 status
= PTR_ERR(handle
);
670 restarted_transaction
:
671 /* reserve a write to the file entry early on - that we if we
672 * run out of credits in the allocation path, we can still
674 status
= ocfs2_journal_access(handle
, inode
, bh
,
675 OCFS2_JOURNAL_ACCESS_WRITE
);
681 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
683 status
= ocfs2_do_extend_allocation(osb
,
693 if ((status
< 0) && (status
!= -EAGAIN
)) {
694 if (status
!= -ENOSPC
)
699 status
= ocfs2_journal_dirty(handle
, bh
);
705 spin_lock(&OCFS2_I(inode
)->ip_lock
);
706 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
707 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
709 if (why
!= RESTART_NONE
&& clusters_to_add
) {
710 if (why
== RESTART_META
) {
711 mlog(0, "restarting function.\n");
714 BUG_ON(why
!= RESTART_TRANS
);
716 mlog(0, "restarting transaction.\n");
717 /* TODO: This can be more intelligent. */
718 credits
= ocfs2_calc_extend_credits(osb
->sb
,
721 status
= ocfs2_extend_trans(handle
, credits
);
723 /* handle still has to be committed at
729 goto restarted_transaction
;
733 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
734 le32_to_cpu(fe
->i_clusters
),
735 (unsigned long long)le64_to_cpu(fe
->i_size
));
736 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
737 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
741 ocfs2_commit_trans(osb
, handle
);
745 ocfs2_free_alloc_context(data_ac
);
749 ocfs2_free_alloc_context(meta_ac
);
752 if ((!status
) && restart_func
) {
765 static int ocfs2_extend_allocation(struct inode
*inode
, u32 logical_start
,
766 u32 clusters_to_add
, int mark_unwritten
)
771 * The alloc sem blocks peope in read/write from reading our
772 * allocation until we're done changing it. We depend on
773 * i_mutex to block other extend/truncate calls while we're
776 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
777 ret
= __ocfs2_extend_allocation(inode
, logical_start
, clusters_to_add
,
779 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
784 /* Some parts of this taken from generic_cont_expand, which turned out
785 * to be too fragile to do exactly what we need without us having to
786 * worry about recursive locking in ->prepare_write() and
787 * ->commit_write(). */
788 static int ocfs2_write_zero_page(struct inode
*inode
,
791 struct address_space
*mapping
= inode
->i_mapping
;
795 handle_t
*handle
= NULL
;
798 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
799 /* ugh. in prepare/commit_write, if from==to==start of block, we
800 ** skip the prepare. make sure we never send an offset for the start
803 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
806 index
= size
>> PAGE_CACHE_SHIFT
;
808 page
= grab_cache_page(mapping
, index
);
815 ret
= ocfs2_prepare_write_nolock(inode
, page
, offset
, offset
);
821 if (ocfs2_should_order_data(inode
)) {
822 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
824 if (IS_ERR(handle
)) {
825 ret
= PTR_ERR(handle
);
831 /* must not update i_size! */
832 ret
= block_commit_write(page
, offset
, offset
);
839 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
842 page_cache_release(page
);
847 static int ocfs2_zero_extend(struct inode
*inode
,
852 struct super_block
*sb
= inode
->i_sb
;
854 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
855 while (start_off
< zero_to_size
) {
856 ret
= ocfs2_write_zero_page(inode
, start_off
);
862 start_off
+= sb
->s_blocksize
;
865 * Very large extends have the potential to lock up
866 * the cpu for extended periods of time.
876 * A tail_to_skip value > 0 indicates that we're being called from
877 * ocfs2_file_aio_write(). This has the following implications:
879 * - we don't want to update i_size
880 * - di_bh will be NULL, which is fine because it's only used in the
881 * case where we want to update i_size.
882 * - ocfs2_zero_extend() will then only be filling the hole created
883 * between i_size and the start of the write.
885 static int ocfs2_extend_file(struct inode
*inode
,
886 struct buffer_head
*di_bh
,
891 u32 clusters_to_add
= 0;
893 BUG_ON(!tail_to_skip
&& !di_bh
);
895 /* setattr sometimes calls us like this. */
899 if (i_size_read(inode
) == new_i_size
)
901 BUG_ON(new_i_size
< i_size_read(inode
));
903 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
))) {
904 BUG_ON(tail_to_skip
!= 0);
905 goto out_update_size
;
908 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
909 OCFS2_I(inode
)->ip_clusters
;
912 * protect the pages that ocfs2_zero_extend is going to be
913 * pulling into the page cache.. we do this before the
914 * metadata extend so that we don't get into the situation
915 * where we've extended the metadata but can't get the data
918 ret
= ocfs2_data_lock(inode
, 1);
924 if (clusters_to_add
) {
925 ret
= ocfs2_extend_allocation(inode
,
926 OCFS2_I(inode
)->ip_clusters
,
935 * Call this even if we don't add any clusters to the tree. We
936 * still need to zero the area between the old i_size and the
939 ret
= ocfs2_zero_extend(inode
, (u64
)new_i_size
- tail_to_skip
);
947 /* We're being called from ocfs2_setattr() which wants
948 * us to update i_size */
949 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
955 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
956 ocfs2_data_unlock(inode
, 1);
962 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
964 int status
= 0, size_change
;
965 struct inode
*inode
= dentry
->d_inode
;
966 struct super_block
*sb
= inode
->i_sb
;
967 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
968 struct buffer_head
*bh
= NULL
;
969 handle_t
*handle
= NULL
;
971 mlog_entry("(0x%p, '%.*s')\n", dentry
,
972 dentry
->d_name
.len
, dentry
->d_name
.name
);
974 if (attr
->ia_valid
& ATTR_MODE
)
975 mlog(0, "mode change: %d\n", attr
->ia_mode
);
976 if (attr
->ia_valid
& ATTR_UID
)
977 mlog(0, "uid change: %d\n", attr
->ia_uid
);
978 if (attr
->ia_valid
& ATTR_GID
)
979 mlog(0, "gid change: %d\n", attr
->ia_gid
);
980 if (attr
->ia_valid
& ATTR_SIZE
)
981 mlog(0, "size change...\n");
982 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
983 mlog(0, "time change...\n");
985 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
986 | ATTR_GID | ATTR_UID | ATTR_MODE)
987 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
988 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
992 status
= inode_change_ok(inode
, attr
);
996 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
998 status
= ocfs2_rw_lock(inode
, 1);
1005 status
= ocfs2_meta_lock(inode
, &bh
, 1);
1007 if (status
!= -ENOENT
)
1009 goto bail_unlock_rw
;
1012 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
1013 if (i_size_read(inode
) > attr
->ia_size
)
1014 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
1016 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
, 0);
1018 if (status
!= -ENOSPC
)
1025 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1026 if (IS_ERR(handle
)) {
1027 status
= PTR_ERR(handle
);
1033 * This will intentionally not wind up calling vmtruncate(),
1034 * since all the work for a size change has been done above.
1035 * Otherwise, we could get into problems with truncate as
1036 * ip_alloc_sem is used there to protect against i_size
1039 status
= inode_setattr(inode
, attr
);
1045 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
1050 ocfs2_commit_trans(osb
, handle
);
1052 ocfs2_meta_unlock(inode
, 1);
1055 ocfs2_rw_unlock(inode
, 1);
1064 int ocfs2_getattr(struct vfsmount
*mnt
,
1065 struct dentry
*dentry
,
1068 struct inode
*inode
= dentry
->d_inode
;
1069 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1070 struct ocfs2_super
*osb
= sb
->s_fs_info
;
1075 err
= ocfs2_inode_revalidate(dentry
);
1082 generic_fillattr(inode
, stat
);
1084 /* We set the blksize from the cluster size for performance */
1085 stat
->blksize
= osb
->s_clustersize
;
1093 int ocfs2_permission(struct inode
*inode
, int mask
, struct nameidata
*nd
)
1099 ret
= ocfs2_meta_lock(inode
, NULL
, 0);
1106 ret
= generic_permission(inode
, mask
, NULL
);
1108 ocfs2_meta_unlock(inode
, 0);
1114 static int __ocfs2_write_remove_suid(struct inode
*inode
,
1115 struct buffer_head
*bh
)
1119 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1120 struct ocfs2_dinode
*di
;
1122 mlog_entry("(Inode %llu, mode 0%o)\n",
1123 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_mode
);
1125 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1126 if (handle
== NULL
) {
1132 ret
= ocfs2_journal_access(handle
, inode
, bh
,
1133 OCFS2_JOURNAL_ACCESS_WRITE
);
1139 inode
->i_mode
&= ~S_ISUID
;
1140 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
1141 inode
->i_mode
&= ~S_ISGID
;
1143 di
= (struct ocfs2_dinode
*) bh
->b_data
;
1144 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
1146 ret
= ocfs2_journal_dirty(handle
, bh
);
1151 ocfs2_commit_trans(osb
, handle
);
1158 * Will look for holes and unwritten extents in the range starting at
1159 * pos for count bytes (inclusive).
1161 static int ocfs2_check_range_for_holes(struct inode
*inode
, loff_t pos
,
1165 unsigned int extent_flags
;
1166 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1167 struct super_block
*sb
= inode
->i_sb
;
1169 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1170 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1173 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1180 if (phys_cpos
== 0 || (extent_flags
& OCFS2_EXT_UNWRITTEN
)) {
1185 if (extent_len
> clusters
)
1186 extent_len
= clusters
;
1188 clusters
-= extent_len
;
1195 static int ocfs2_write_remove_suid(struct inode
*inode
)
1198 struct buffer_head
*bh
= NULL
;
1199 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1201 ret
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
),
1202 oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
1208 ret
= __ocfs2_write_remove_suid(inode
, bh
);
1215 * Allocate enough extents to cover the region starting at byte offset
1216 * start for len bytes. Existing extents are skipped, any extents
1217 * added are marked as "unwritten".
1219 static int ocfs2_allocate_unwritten_extents(struct inode
*inode
,
1223 u32 cpos
, phys_cpos
, clusters
, alloc_size
;
1226 * We consider both start and len to be inclusive.
1228 cpos
= start
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1229 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, start
+ len
);
1233 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1241 * Hole or existing extent len can be arbitrary, so
1242 * cap it to our own allocation request.
1244 if (alloc_size
> clusters
)
1245 alloc_size
= clusters
;
1249 * We already have an allocation at this
1250 * region so we can safely skip it.
1255 ret
= __ocfs2_extend_allocation(inode
, cpos
, alloc_size
, 1);
1264 clusters
-= alloc_size
;
1272 static int __ocfs2_remove_inode_range(struct inode
*inode
,
1273 struct buffer_head
*di_bh
,
1274 u32 cpos
, u32 phys_cpos
, u32 len
,
1275 struct ocfs2_cached_dealloc_ctxt
*dealloc
)
1278 u64 phys_blkno
= ocfs2_clusters_to_blocks(inode
->i_sb
, phys_cpos
);
1279 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1280 struct inode
*tl_inode
= osb
->osb_tl_inode
;
1282 struct ocfs2_alloc_context
*meta_ac
= NULL
;
1283 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1285 ret
= ocfs2_lock_allocators(inode
, di
, 0, 1, NULL
, &meta_ac
);
1291 mutex_lock(&tl_inode
->i_mutex
);
1293 if (ocfs2_truncate_log_needs_flush(osb
)) {
1294 ret
= __ocfs2_flush_truncate_log(osb
);
1301 handle
= ocfs2_start_trans(osb
, OCFS2_REMOVE_EXTENT_CREDITS
);
1302 if (handle
== NULL
) {
1308 ret
= ocfs2_journal_access(handle
, inode
, di_bh
,
1309 OCFS2_JOURNAL_ACCESS_WRITE
);
1315 ret
= ocfs2_remove_extent(inode
, di_bh
, cpos
, len
, handle
, meta_ac
,
1322 OCFS2_I(inode
)->ip_clusters
-= len
;
1323 di
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1325 ret
= ocfs2_journal_dirty(handle
, di_bh
);
1331 ret
= ocfs2_truncate_log_append(osb
, handle
, phys_blkno
, len
);
1336 ocfs2_commit_trans(osb
, handle
);
1338 mutex_unlock(&tl_inode
->i_mutex
);
1341 ocfs2_free_alloc_context(meta_ac
);
1347 * Truncate a byte range, avoiding pages within partial clusters. This
1348 * preserves those pages for the zeroing code to write to.
1350 static void ocfs2_truncate_cluster_pages(struct inode
*inode
, u64 byte_start
,
1353 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1355 struct address_space
*mapping
= inode
->i_mapping
;
1357 start
= (loff_t
)ocfs2_align_bytes_to_clusters(inode
->i_sb
, byte_start
);
1358 end
= byte_start
+ byte_len
;
1359 end
= end
& ~(osb
->s_clustersize
- 1);
1362 unmap_mapping_range(mapping
, start
, end
- start
, 0);
1363 truncate_inode_pages_range(mapping
, start
, end
- 1);
1367 static int ocfs2_zero_partial_clusters(struct inode
*inode
,
1371 u64 tmpend
, end
= start
+ len
;
1372 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1373 unsigned int csize
= osb
->s_clustersize
;
1377 * The "start" and "end" values are NOT necessarily part of
1378 * the range whose allocation is being deleted. Rather, this
1379 * is what the user passed in with the request. We must zero
1380 * partial clusters here. There's no need to worry about
1381 * physical allocation - the zeroing code knows to skip holes.
1383 mlog(0, "byte start: %llu, end: %llu\n",
1384 (unsigned long long)start
, (unsigned long long)end
);
1387 * If both edges are on a cluster boundary then there's no
1388 * zeroing required as the region is part of the allocation to
1391 if ((start
& (csize
- 1)) == 0 && (end
& (csize
- 1)) == 0)
1394 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1395 if (handle
== NULL
) {
1402 * We want to get the byte offset of the end of the 1st cluster.
1404 tmpend
= (u64
)osb
->s_clustersize
+ (start
& ~(osb
->s_clustersize
- 1));
1408 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1409 (unsigned long long)start
, (unsigned long long)tmpend
);
1411 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, tmpend
);
1417 * This may make start and end equal, but the zeroing
1418 * code will skip any work in that case so there's no
1419 * need to catch it up here.
1421 start
= end
& ~(osb
->s_clustersize
- 1);
1423 mlog(0, "2nd range: start: %llu, end: %llu\n",
1424 (unsigned long long)start
, (unsigned long long)end
);
1426 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, end
);
1431 ocfs2_commit_trans(osb
, handle
);
1436 static int ocfs2_remove_inode_range(struct inode
*inode
,
1437 struct buffer_head
*di_bh
, u64 byte_start
,
1441 u32 trunc_start
, trunc_len
, cpos
, phys_cpos
, alloc_size
;
1442 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1443 struct ocfs2_cached_dealloc_ctxt dealloc
;
1445 ocfs2_init_dealloc_ctxt(&dealloc
);
1450 trunc_start
= ocfs2_clusters_for_bytes(osb
->sb
, byte_start
);
1451 trunc_len
= (byte_start
+ byte_len
) >> osb
->s_clustersize_bits
;
1452 if (trunc_len
>= trunc_start
)
1453 trunc_len
-= trunc_start
;
1457 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1458 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1459 (unsigned long long)byte_start
,
1460 (unsigned long long)byte_len
, trunc_start
, trunc_len
);
1462 ret
= ocfs2_zero_partial_clusters(inode
, byte_start
, byte_len
);
1470 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1477 if (alloc_size
> trunc_len
)
1478 alloc_size
= trunc_len
;
1480 /* Only do work for non-holes */
1481 if (phys_cpos
!= 0) {
1482 ret
= __ocfs2_remove_inode_range(inode
, di_bh
, cpos
,
1483 phys_cpos
, alloc_size
,
1492 trunc_len
-= alloc_size
;
1495 ocfs2_truncate_cluster_pages(inode
, byte_start
, byte_len
);
1498 ocfs2_schedule_truncate_log_flush(osb
, 1);
1499 ocfs2_run_deallocs(osb
, &dealloc
);
1505 * Parts of this function taken from xfs_change_file_space()
1507 int ocfs2_change_file_space(struct file
*file
, unsigned int cmd
,
1508 struct ocfs2_space_resv
*sr
)
1512 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1513 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1514 struct buffer_head
*di_bh
= NULL
;
1516 unsigned long long max_off
= ocfs2_max_file_offset(inode
->i_sb
->s_blocksize_bits
);
1518 if ((cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) &&
1519 !ocfs2_writes_unwritten_extents(osb
))
1521 else if ((cmd
== OCFS2_IOC_UNRESVSP
|| cmd
== OCFS2_IOC_UNRESVSP64
) &&
1522 !ocfs2_sparse_alloc(osb
))
1525 if (!S_ISREG(inode
->i_mode
))
1528 if (!(file
->f_mode
& FMODE_WRITE
))
1531 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
1534 mutex_lock(&inode
->i_mutex
);
1537 * This prevents concurrent writes on other nodes
1539 ret
= ocfs2_rw_lock(inode
, 1);
1545 ret
= ocfs2_meta_lock(inode
, &di_bh
, 1);
1551 if (inode
->i_flags
& (S_IMMUTABLE
|S_APPEND
)) {
1553 goto out_meta_unlock
;
1556 switch (sr
->l_whence
) {
1557 case 0: /*SEEK_SET*/
1559 case 1: /*SEEK_CUR*/
1560 sr
->l_start
+= file
->f_pos
;
1562 case 2: /*SEEK_END*/
1563 sr
->l_start
+= i_size_read(inode
);
1567 goto out_meta_unlock
;
1571 llen
= sr
->l_len
> 0 ? sr
->l_len
- 1 : sr
->l_len
;
1574 || sr
->l_start
> max_off
1575 || (sr
->l_start
+ llen
) < 0
1576 || (sr
->l_start
+ llen
) > max_off
) {
1578 goto out_meta_unlock
;
1581 if (cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) {
1582 if (sr
->l_len
<= 0) {
1584 goto out_meta_unlock
;
1588 if (should_remove_suid(file
->f_path
.dentry
)) {
1589 ret
= __ocfs2_write_remove_suid(inode
, di_bh
);
1592 goto out_meta_unlock
;
1596 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1598 case OCFS2_IOC_RESVSP
:
1599 case OCFS2_IOC_RESVSP64
:
1601 * This takes unsigned offsets, but the signed ones we
1602 * pass have been checked against overflow above.
1604 ret
= ocfs2_allocate_unwritten_extents(inode
, sr
->l_start
,
1607 case OCFS2_IOC_UNRESVSP
:
1608 case OCFS2_IOC_UNRESVSP64
:
1609 ret
= ocfs2_remove_inode_range(inode
, di_bh
, sr
->l_start
,
1615 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1618 goto out_meta_unlock
;
1622 * We update c/mtime for these changes
1624 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1625 if (IS_ERR(handle
)) {
1626 ret
= PTR_ERR(handle
);
1628 goto out_meta_unlock
;
1631 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1632 ret
= ocfs2_mark_inode_dirty(handle
, inode
, di_bh
);
1636 ocfs2_commit_trans(osb
, handle
);
1640 ocfs2_meta_unlock(inode
, 1);
1642 ocfs2_rw_unlock(inode
, 1);
1644 mutex_unlock(&inode
->i_mutex
);
1649 static int ocfs2_prepare_inode_for_write(struct dentry
*dentry
,
1655 int ret
= 0, meta_level
= appending
;
1656 struct inode
*inode
= dentry
->d_inode
;
1658 loff_t newsize
, saved_pos
;
1661 * We sample i_size under a read level meta lock to see if our write
1662 * is extending the file, if it is we back off and get a write level
1666 ret
= ocfs2_meta_lock(inode
, NULL
, meta_level
);
1673 /* Clear suid / sgid if necessary. We do this here
1674 * instead of later in the write path because
1675 * remove_suid() calls ->setattr without any hint that
1676 * we may have already done our cluster locking. Since
1677 * ocfs2_setattr() *must* take cluster locks to
1678 * proceeed, this will lead us to recursively lock the
1679 * inode. There's also the dinode i_size state which
1680 * can be lost via setattr during extending writes (we
1681 * set inode->i_size at the end of a write. */
1682 if (should_remove_suid(dentry
)) {
1683 if (meta_level
== 0) {
1684 ocfs2_meta_unlock(inode
, meta_level
);
1689 ret
= ocfs2_write_remove_suid(inode
);
1696 /* work on a copy of ppos until we're sure that we won't have
1697 * to recalculate it due to relocking. */
1699 saved_pos
= i_size_read(inode
);
1700 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1705 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
))) {
1706 loff_t end
= saved_pos
+ count
;
1709 * Skip the O_DIRECT checks if we don't need
1712 if (!direct_io
|| !(*direct_io
))
1716 * Allowing concurrent direct writes means
1717 * i_size changes wouldn't be synchronized, so
1718 * one node could wind up truncating another
1721 if (end
> i_size_read(inode
)) {
1727 * We don't fill holes during direct io, so
1728 * check for them here. If any are found, the
1729 * caller will have to retake some cluster
1730 * locks and initiate the io as buffered.
1732 ret
= ocfs2_check_range_for_holes(inode
, saved_pos
,
1743 * The rest of this loop is concerned with legacy file
1744 * systems which don't support sparse files.
1747 newsize
= count
+ saved_pos
;
1749 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1750 (long long) saved_pos
, (long long) newsize
,
1751 (long long) i_size_read(inode
));
1753 /* No need for a higher level metadata lock if we're
1754 * never going past i_size. */
1755 if (newsize
<= i_size_read(inode
))
1758 if (meta_level
== 0) {
1759 ocfs2_meta_unlock(inode
, meta_level
);
1764 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1765 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1766 OCFS2_I(inode
)->ip_clusters
;
1767 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1769 mlog(0, "Writing at EOF, may need more allocation: "
1770 "i_size = %lld, newsize = %lld, need %u clusters\n",
1771 (long long) i_size_read(inode
), (long long) newsize
,
1774 /* We only want to continue the rest of this loop if
1775 * our extend will actually require more
1780 ret
= ocfs2_extend_file(inode
, NULL
, newsize
, count
);
1793 ocfs2_meta_unlock(inode
, meta_level
);
1800 ocfs2_set_next_iovec(const struct iovec
**iovp
, size_t *basep
, size_t bytes
)
1802 const struct iovec
*iov
= *iovp
;
1803 size_t base
= *basep
;
1806 int copy
= min(bytes
, iov
->iov_len
- base
);
1810 if (iov
->iov_len
== base
) {
1819 static struct page
* ocfs2_get_write_source(char **ret_src_buf
,
1820 const struct iovec
*cur_iov
,
1824 char *buf
= cur_iov
->iov_base
+ iov_offset
;
1825 struct page
*src_page
= NULL
;
1828 off
= (unsigned long)(buf
) & ~PAGE_CACHE_MASK
;
1830 if (!segment_eq(get_fs(), KERNEL_DS
)) {
1832 * Pull in the user page. We want to do this outside
1833 * of the meta data locks in order to preserve locking
1834 * order in case of page fault.
1836 ret
= get_user_pages(current
, current
->mm
,
1837 (unsigned long)buf
& PAGE_CACHE_MASK
, 1,
1838 0, 0, &src_page
, NULL
);
1840 *ret_src_buf
= kmap(src_page
) + off
;
1842 src_page
= ERR_PTR(-EFAULT
);
1850 static void ocfs2_put_write_source(struct page
*page
)
1854 page_cache_release(page
);
1858 static ssize_t
ocfs2_file_buffered_write(struct file
*file
, loff_t
*ppos
,
1859 const struct iovec
*iov
,
1860 unsigned long nr_segs
,
1862 ssize_t o_direct_written
)
1865 ssize_t copied
, total
= 0;
1866 size_t iov_offset
= 0, bytes
;
1868 const struct iovec
*cur_iov
= iov
;
1869 struct page
*user_page
, *page
;
1874 * handle partial DIO write. Adjust cur_iov if needed.
1876 ocfs2_set_next_iovec(&cur_iov
, &iov_offset
, o_direct_written
);
1881 user_page
= ocfs2_get_write_source(&buf
, cur_iov
, iov_offset
);
1882 if (IS_ERR(user_page
)) {
1883 ret
= PTR_ERR(user_page
);
1887 /* Stay within our page boundaries */
1888 bytes
= min((PAGE_CACHE_SIZE
- ((unsigned long)pos
& ~PAGE_CACHE_MASK
)),
1889 (PAGE_CACHE_SIZE
- ((unsigned long)buf
& ~PAGE_CACHE_MASK
)));
1890 /* Stay within the vector boundary */
1891 bytes
= min_t(size_t, bytes
, cur_iov
->iov_len
- iov_offset
);
1892 /* Stay within count */
1893 bytes
= min(bytes
, count
);
1896 ret
= ocfs2_write_begin(file
, file
->f_mapping
, pos
, bytes
, 0,
1903 dst
= kmap_atomic(page
, KM_USER0
);
1904 memcpy(dst
+ (pos
& (PAGE_CACHE_SIZE
- 1)), buf
, bytes
);
1905 kunmap_atomic(dst
, KM_USER0
);
1906 flush_dcache_page(page
);
1907 ocfs2_put_write_source(user_page
);
1909 copied
= ocfs2_write_end(file
, file
->f_mapping
, pos
, bytes
,
1910 bytes
, page
, fsdata
);
1918 *ppos
= pos
+ copied
;
1921 ocfs2_set_next_iovec(&cur_iov
, &iov_offset
, copied
);
1925 return total
? total
: ret
;
1928 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
1929 const struct iovec
*iov
,
1930 unsigned long nr_segs
,
1933 int ret
, direct_io
, appending
, rw_level
, have_alloc_sem
= 0;
1934 int can_do_direct
, sync
= 0;
1935 ssize_t written
= 0;
1936 size_t ocount
; /* original count */
1937 size_t count
; /* after file limit checks */
1938 loff_t
*ppos
= &iocb
->ki_pos
;
1939 struct file
*file
= iocb
->ki_filp
;
1940 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1942 mlog_entry("(0x%p, %u, '%.*s')\n", file
,
1943 (unsigned int)nr_segs
,
1944 file
->f_path
.dentry
->d_name
.len
,
1945 file
->f_path
.dentry
->d_name
.name
);
1947 if (iocb
->ki_left
== 0)
1950 ret
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
1956 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1958 appending
= file
->f_flags
& O_APPEND
? 1 : 0;
1959 direct_io
= file
->f_flags
& O_DIRECT
? 1 : 0;
1961 mutex_lock(&inode
->i_mutex
);
1964 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1966 down_read(&inode
->i_alloc_sem
);
1970 /* concurrent O_DIRECT writes are allowed */
1971 rw_level
= !direct_io
;
1972 ret
= ocfs2_rw_lock(inode
, rw_level
);
1978 can_do_direct
= direct_io
;
1979 ret
= ocfs2_prepare_inode_for_write(file
->f_path
.dentry
, ppos
,
1980 iocb
->ki_left
, appending
,
1988 * We can't complete the direct I/O as requested, fall back to
1991 if (direct_io
&& !can_do_direct
) {
1992 ocfs2_rw_unlock(inode
, rw_level
);
1993 up_read(&inode
->i_alloc_sem
);
2003 if (!sync
&& ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
)))
2007 * XXX: Is it ok to execute these checks a second time?
2009 ret
= generic_write_checks(file
, ppos
, &count
, S_ISBLK(inode
->i_mode
));
2014 * Set pos so that sync_page_range_nolock() below understands
2015 * where to start from. We might've moved it around via the
2016 * calls above. The range we want to actually sync starts from
2022 /* communicate with ocfs2_dio_end_io */
2023 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2026 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, *ppos
,
2027 ppos
, count
, ocount
);
2033 written
= ocfs2_file_buffered_write(file
, ppos
, iov
, nr_segs
,
2037 if (ret
!= -EFAULT
|| ret
!= -ENOSPC
)
2044 /* buffered aio wouldn't have proper lock coverage today */
2045 BUG_ON(ret
== -EIOCBQUEUED
&& !(file
->f_flags
& O_DIRECT
));
2048 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2049 * function pointer which is called when o_direct io completes so that
2050 * it can unlock our rw lock. (it's the clustered equivalent of
2051 * i_alloc_sem; protects truncate from racing with pending ios).
2052 * Unfortunately there are error cases which call end_io and others
2053 * that don't. so we don't have to unlock the rw_lock if either an
2054 * async dio is going to do it in the future or an end_io after an
2055 * error has already done it.
2057 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
2064 ocfs2_rw_unlock(inode
, rw_level
);
2068 up_read(&inode
->i_alloc_sem
);
2070 if (written
> 0 && sync
) {
2073 err
= sync_page_range_nolock(inode
, file
->f_mapping
, pos
, count
);
2078 mutex_unlock(&inode
->i_mutex
);
2081 return written
? written
: ret
;
2084 static int ocfs2_splice_write_actor(struct pipe_inode_info
*pipe
,
2085 struct pipe_buffer
*buf
,
2086 struct splice_desc
*sd
)
2090 struct file
*file
= sd
->u
.file
;
2091 unsigned int offset
;
2092 struct page
*page
= NULL
;
2096 ret
= buf
->ops
->confirm(pipe
, buf
);
2100 offset
= sd
->pos
& ~PAGE_CACHE_MASK
;
2102 if (count
+ offset
> PAGE_CACHE_SIZE
)
2103 count
= PAGE_CACHE_SIZE
- offset
;
2105 ret
= ocfs2_write_begin(file
, file
->f_mapping
, sd
->pos
, count
, 0,
2112 src
= buf
->ops
->map(pipe
, buf
, 1);
2113 dst
= kmap_atomic(page
, KM_USER1
);
2114 memcpy(dst
+ offset
, src
+ buf
->offset
, count
);
2115 kunmap_atomic(page
, KM_USER1
);
2116 buf
->ops
->unmap(pipe
, buf
, src
);
2118 copied
= ocfs2_write_end(file
, file
->f_mapping
, sd
->pos
, count
, count
,
2127 return copied
? copied
: ret
;
2130 static ssize_t
__ocfs2_file_splice_write(struct pipe_inode_info
*pipe
,
2137 struct address_space
*mapping
= out
->f_mapping
;
2138 struct inode
*inode
= mapping
->host
;
2139 struct splice_desc sd
= {
2146 ret
= __splice_from_pipe(pipe
, &sd
, ocfs2_splice_write_actor
);
2150 if (unlikely((out
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2151 err
= generic_osync_inode(inode
, mapping
,
2152 OSYNC_METADATA
|OSYNC_DATA
);
2161 static ssize_t
ocfs2_file_splice_write(struct pipe_inode_info
*pipe
,
2168 struct inode
*inode
= out
->f_path
.dentry
->d_inode
;
2170 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out
, pipe
,
2172 out
->f_path
.dentry
->d_name
.len
,
2173 out
->f_path
.dentry
->d_name
.name
);
2175 inode_double_lock(inode
, pipe
->inode
);
2177 ret
= ocfs2_rw_lock(inode
, 1);
2183 ret
= ocfs2_prepare_inode_for_write(out
->f_path
.dentry
, ppos
, len
, 0,
2190 /* ok, we're done with i_size and alloc work */
2191 ret
= __ocfs2_file_splice_write(pipe
, out
, ppos
, len
, flags
);
2194 ocfs2_rw_unlock(inode
, 1);
2196 inode_double_unlock(inode
, pipe
->inode
);
2202 static ssize_t
ocfs2_file_splice_read(struct file
*in
,
2204 struct pipe_inode_info
*pipe
,
2209 struct inode
*inode
= in
->f_path
.dentry
->d_inode
;
2211 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in
, pipe
,
2213 in
->f_path
.dentry
->d_name
.len
,
2214 in
->f_path
.dentry
->d_name
.name
);
2217 * See the comment in ocfs2_file_aio_read()
2219 ret
= ocfs2_meta_lock(inode
, NULL
, 0);
2224 ocfs2_meta_unlock(inode
, 0);
2226 ret
= generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
2233 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
2234 const struct iovec
*iov
,
2235 unsigned long nr_segs
,
2238 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0, lock_level
= 0;
2239 struct file
*filp
= iocb
->ki_filp
;
2240 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
2242 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
2243 (unsigned int)nr_segs
,
2244 filp
->f_path
.dentry
->d_name
.len
,
2245 filp
->f_path
.dentry
->d_name
.name
);
2254 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2255 * need locks to protect pending reads from racing with truncate.
2257 if (filp
->f_flags
& O_DIRECT
) {
2258 down_read(&inode
->i_alloc_sem
);
2261 ret
= ocfs2_rw_lock(inode
, 0);
2267 /* communicate with ocfs2_dio_end_io */
2268 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2272 * We're fine letting folks race truncates and extending
2273 * writes with read across the cluster, just like they can
2274 * locally. Hence no rw_lock during read.
2276 * Take and drop the meta data lock to update inode fields
2277 * like i_size. This allows the checks down below
2278 * generic_file_aio_read() a chance of actually working.
2280 ret
= ocfs2_meta_lock_atime(inode
, filp
->f_vfsmnt
, &lock_level
);
2285 ocfs2_meta_unlock(inode
, lock_level
);
2287 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
2289 mlog(ML_ERROR
, "generic_file_aio_read returned -EINVAL\n");
2291 /* buffered aio wouldn't have proper lock coverage today */
2292 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
2294 /* see ocfs2_file_aio_write */
2295 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
2302 up_read(&inode
->i_alloc_sem
);
2304 ocfs2_rw_unlock(inode
, rw_level
);
2310 const struct inode_operations ocfs2_file_iops
= {
2311 .setattr
= ocfs2_setattr
,
2312 .getattr
= ocfs2_getattr
,
2313 .permission
= ocfs2_permission
,
2316 const struct inode_operations ocfs2_special_file_iops
= {
2317 .setattr
= ocfs2_setattr
,
2318 .getattr
= ocfs2_getattr
,
2319 .permission
= ocfs2_permission
,
2322 const struct file_operations ocfs2_fops
= {
2323 .read
= do_sync_read
,
2324 .write
= do_sync_write
,
2326 .fsync
= ocfs2_sync_file
,
2327 .release
= ocfs2_file_release
,
2328 .open
= ocfs2_file_open
,
2329 .aio_read
= ocfs2_file_aio_read
,
2330 .aio_write
= ocfs2_file_aio_write
,
2331 .ioctl
= ocfs2_ioctl
,
2332 #ifdef CONFIG_COMPAT
2333 .compat_ioctl
= ocfs2_compat_ioctl
,
2335 .splice_read
= ocfs2_file_splice_read
,
2336 .splice_write
= ocfs2_file_splice_write
,
2339 const struct file_operations ocfs2_dops
= {
2340 .read
= generic_read_dir
,
2341 .readdir
= ocfs2_readdir
,
2342 .fsync
= ocfs2_sync_file
,
2343 .ioctl
= ocfs2_ioctl
,
2344 #ifdef CONFIG_COMPAT
2345 .compat_ioctl
= ocfs2_compat_ioctl
,