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>
37 #include <linux/falloc.h>
38 #include <linux/quotaops.h>
40 #define MLOG_MASK_PREFIX ML_INODE
41 #include <cluster/masklog.h>
49 #include "extent_map.h"
63 #include "buffer_head_io.h"
65 static int ocfs2_sync_inode(struct inode
*inode
)
67 filemap_fdatawrite(inode
->i_mapping
);
68 return sync_mapping_buffers(inode
->i_mapping
);
71 static int ocfs2_init_file_private(struct inode
*inode
, struct file
*file
)
73 struct ocfs2_file_private
*fp
;
75 fp
= kzalloc(sizeof(struct ocfs2_file_private
), GFP_KERNEL
);
80 mutex_init(&fp
->fp_mutex
);
81 ocfs2_file_lock_res_init(&fp
->fp_flock
, fp
);
82 file
->private_data
= fp
;
87 static void ocfs2_free_file_private(struct inode
*inode
, struct file
*file
)
89 struct ocfs2_file_private
*fp
= file
->private_data
;
90 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
93 ocfs2_simple_drop_lockres(osb
, &fp
->fp_flock
);
94 ocfs2_lock_res_free(&fp
->fp_flock
);
96 file
->private_data
= NULL
;
100 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
103 int mode
= file
->f_flags
;
104 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
106 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
107 file
->f_path
.dentry
->d_name
.len
, file
->f_path
.dentry
->d_name
.name
);
109 spin_lock(&oi
->ip_lock
);
111 /* Check that the inode hasn't been wiped from disk by another
112 * node. If it hasn't then we're safe as long as we hold the
113 * spin lock until our increment of open count. */
114 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
115 spin_unlock(&oi
->ip_lock
);
122 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
125 spin_unlock(&oi
->ip_lock
);
127 status
= ocfs2_init_file_private(inode
, file
);
130 * We want to set open count back if we're failing the
133 spin_lock(&oi
->ip_lock
);
135 spin_unlock(&oi
->ip_lock
);
143 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
145 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
147 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
148 file
->f_path
.dentry
->d_name
.len
,
149 file
->f_path
.dentry
->d_name
.name
);
151 spin_lock(&oi
->ip_lock
);
152 if (!--oi
->ip_open_count
)
153 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
154 spin_unlock(&oi
->ip_lock
);
156 ocfs2_free_file_private(inode
, file
);
163 static int ocfs2_dir_open(struct inode
*inode
, struct file
*file
)
165 return ocfs2_init_file_private(inode
, file
);
168 static int ocfs2_dir_release(struct inode
*inode
, struct file
*file
)
170 ocfs2_free_file_private(inode
, file
);
174 static int ocfs2_sync_file(struct file
*file
,
175 struct dentry
*dentry
,
180 struct inode
*inode
= dentry
->d_inode
;
181 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
183 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
184 dentry
->d_name
.len
, dentry
->d_name
.name
);
186 err
= ocfs2_sync_inode(dentry
->d_inode
);
190 if (datasync
&& !(inode
->i_state
& I_DIRTY_DATASYNC
))
193 journal
= osb
->journal
->j_journal
;
194 err
= jbd2_journal_force_commit(journal
);
199 return (err
< 0) ? -EIO
: 0;
202 int ocfs2_should_update_atime(struct inode
*inode
,
203 struct vfsmount
*vfsmnt
)
206 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
208 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
211 if ((inode
->i_flags
& S_NOATIME
) ||
212 ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
216 * We can be called with no vfsmnt structure - NFSD will
219 * Note that our action here is different than touch_atime() -
220 * if we can't tell whether this is a noatime mount, then we
221 * don't know whether to trust the value of s_atime_quantum.
226 if ((vfsmnt
->mnt_flags
& MNT_NOATIME
) ||
227 ((vfsmnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
230 if (vfsmnt
->mnt_flags
& MNT_RELATIME
) {
231 if ((timespec_compare(&inode
->i_atime
, &inode
->i_mtime
) <= 0) ||
232 (timespec_compare(&inode
->i_atime
, &inode
->i_ctime
) <= 0))
239 if ((now
.tv_sec
- inode
->i_atime
.tv_sec
<= osb
->s_atime_quantum
))
245 int ocfs2_update_inode_atime(struct inode
*inode
,
246 struct buffer_head
*bh
)
249 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
251 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) bh
->b_data
;
255 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
256 if (IS_ERR(handle
)) {
257 ret
= PTR_ERR(handle
);
262 ret
= ocfs2_journal_access_di(handle
, inode
, bh
,
263 OCFS2_JOURNAL_ACCESS_WRITE
);
270 * Don't use ocfs2_mark_inode_dirty() here as we don't always
271 * have i_mutex to guard against concurrent changes to other
274 inode
->i_atime
= CURRENT_TIME
;
275 di
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
276 di
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
278 ret
= ocfs2_journal_dirty(handle
, bh
);
283 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
289 static int ocfs2_set_inode_size(handle_t
*handle
,
291 struct buffer_head
*fe_bh
,
297 i_size_write(inode
, new_i_size
);
298 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
299 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
301 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
312 int ocfs2_simple_size_update(struct inode
*inode
,
313 struct buffer_head
*di_bh
,
317 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
318 handle_t
*handle
= NULL
;
320 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
321 if (IS_ERR(handle
)) {
322 ret
= PTR_ERR(handle
);
327 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
332 ocfs2_commit_trans(osb
, handle
);
337 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
339 struct buffer_head
*fe_bh
,
344 struct ocfs2_dinode
*di
;
349 /* TODO: This needs to actually orphan the inode in this
352 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
353 if (IS_ERR(handle
)) {
354 status
= PTR_ERR(handle
);
359 status
= ocfs2_journal_access_di(handle
, inode
, fe_bh
,
360 OCFS2_JOURNAL_ACCESS_WRITE
);
367 * Do this before setting i_size.
369 cluster_bytes
= ocfs2_align_bytes_to_clusters(inode
->i_sb
, new_i_size
);
370 status
= ocfs2_zero_range_for_truncate(inode
, handle
, new_i_size
,
377 i_size_write(inode
, new_i_size
);
378 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
380 di
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
381 di
->i_size
= cpu_to_le64(new_i_size
);
382 di
->i_ctime
= di
->i_mtime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
383 di
->i_ctime_nsec
= di
->i_mtime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
385 status
= ocfs2_journal_dirty(handle
, fe_bh
);
390 ocfs2_commit_trans(osb
, handle
);
397 static int ocfs2_truncate_file(struct inode
*inode
,
398 struct buffer_head
*di_bh
,
402 struct ocfs2_dinode
*fe
= NULL
;
403 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
404 struct ocfs2_truncate_context
*tc
= NULL
;
406 mlog_entry("(inode = %llu, new_i_size = %llu\n",
407 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
408 (unsigned long long)new_i_size
);
410 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
411 * already validated it */
412 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
414 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
415 "Inode %llu, inode i_size = %lld != di "
416 "i_size = %llu, i_flags = 0x%x\n",
417 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
419 (unsigned long long)le64_to_cpu(fe
->i_size
),
420 le32_to_cpu(fe
->i_flags
));
422 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
423 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
424 (unsigned long long)le64_to_cpu(fe
->i_size
),
425 (unsigned long long)new_i_size
);
431 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
432 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
433 (unsigned long long)le64_to_cpu(fe
->i_size
),
434 (unsigned long long)new_i_size
);
436 /* lets handle the simple truncate cases before doing any more
437 * cluster locking. */
438 if (new_i_size
== le64_to_cpu(fe
->i_size
))
441 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
444 * The inode lock forced other nodes to sync and drop their
445 * pages, which (correctly) happens even if we have a truncate
446 * without allocation change - ocfs2 cluster sizes can be much
447 * greater than page size, so we have to truncate them
450 unmap_mapping_range(inode
->i_mapping
, new_i_size
+ PAGE_SIZE
- 1, 0, 1);
451 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
453 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
454 status
= ocfs2_truncate_inline(inode
, di_bh
, new_i_size
,
455 i_size_read(inode
), 1);
459 goto bail_unlock_sem
;
462 /* alright, we're going to need to do a full blown alloc size
463 * change. Orphan the inode so that recovery can complete the
464 * truncate if necessary. This does the task of marking
466 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
469 goto bail_unlock_sem
;
472 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
475 goto bail_unlock_sem
;
478 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
481 goto bail_unlock_sem
;
484 /* TODO: orphan dir cleanup here. */
486 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
495 * extend file allocation only here.
496 * we'll update all the disk stuff, and oip->alloc_size
498 * expect stuff to be locked, a transaction started and enough data /
499 * metadata reservations in the contexts.
501 * Will return -EAGAIN, and a reason if a restart is needed.
502 * If passed in, *reason will always be set, even in error.
504 int ocfs2_add_inode_data(struct ocfs2_super
*osb
,
509 struct buffer_head
*fe_bh
,
511 struct ocfs2_alloc_context
*data_ac
,
512 struct ocfs2_alloc_context
*meta_ac
,
513 enum ocfs2_alloc_restarted
*reason_ret
)
516 struct ocfs2_extent_tree et
;
518 ocfs2_init_dinode_extent_tree(&et
, inode
, fe_bh
);
519 ret
= ocfs2_add_clusters_in_btree(osb
, inode
, logical_offset
,
520 clusters_to_add
, mark_unwritten
,
522 data_ac
, meta_ac
, reason_ret
);
527 static int __ocfs2_extend_allocation(struct inode
*inode
, u32 logical_start
,
528 u32 clusters_to_add
, int mark_unwritten
)
531 int restart_func
= 0;
534 struct buffer_head
*bh
= NULL
;
535 struct ocfs2_dinode
*fe
= NULL
;
536 handle_t
*handle
= NULL
;
537 struct ocfs2_alloc_context
*data_ac
= NULL
;
538 struct ocfs2_alloc_context
*meta_ac
= NULL
;
539 enum ocfs2_alloc_restarted why
;
540 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
541 struct ocfs2_extent_tree et
;
544 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
547 * This function only exists for file systems which don't
550 BUG_ON(mark_unwritten
&& !ocfs2_sparse_alloc(osb
));
552 status
= ocfs2_read_inode_block(inode
, &bh
);
557 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
560 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
562 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
563 "clusters_to_add = %u\n",
564 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
565 (long long)i_size_read(inode
), le32_to_cpu(fe
->i_clusters
),
567 ocfs2_init_dinode_extent_tree(&et
, inode
, bh
);
568 status
= ocfs2_lock_allocators(inode
, &et
, clusters_to_add
, 0,
575 credits
= ocfs2_calc_extend_credits(osb
->sb
, &fe
->id2
.i_list
,
577 handle
= ocfs2_start_trans(osb
, credits
);
578 if (IS_ERR(handle
)) {
579 status
= PTR_ERR(handle
);
585 restarted_transaction
:
586 if (vfs_dq_alloc_space_nodirty(inode
, ocfs2_clusters_to_bytes(osb
->sb
,
593 /* reserve a write to the file entry early on - that we if we
594 * run out of credits in the allocation path, we can still
596 status
= ocfs2_journal_access_di(handle
, inode
, bh
,
597 OCFS2_JOURNAL_ACCESS_WRITE
);
603 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
605 status
= ocfs2_add_inode_data(osb
,
615 if ((status
< 0) && (status
!= -EAGAIN
)) {
616 if (status
!= -ENOSPC
)
621 status
= ocfs2_journal_dirty(handle
, bh
);
627 spin_lock(&OCFS2_I(inode
)->ip_lock
);
628 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
629 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
630 /* Release unused quota reservation */
631 vfs_dq_free_space(inode
,
632 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
635 if (why
!= RESTART_NONE
&& clusters_to_add
) {
636 if (why
== RESTART_META
) {
637 mlog(0, "restarting function.\n");
640 BUG_ON(why
!= RESTART_TRANS
);
642 mlog(0, "restarting transaction.\n");
643 /* TODO: This can be more intelligent. */
644 credits
= ocfs2_calc_extend_credits(osb
->sb
,
647 status
= ocfs2_extend_trans(handle
, credits
);
649 /* handle still has to be committed at
655 goto restarted_transaction
;
659 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
660 le32_to_cpu(fe
->i_clusters
),
661 (unsigned long long)le64_to_cpu(fe
->i_size
));
662 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
663 OCFS2_I(inode
)->ip_clusters
, (long long)i_size_read(inode
));
666 if (status
< 0 && did_quota
)
667 vfs_dq_free_space(inode
,
668 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
670 ocfs2_commit_trans(osb
, handle
);
674 ocfs2_free_alloc_context(data_ac
);
678 ocfs2_free_alloc_context(meta_ac
);
681 if ((!status
) && restart_func
) {
692 /* Some parts of this taken from generic_cont_expand, which turned out
693 * to be too fragile to do exactly what we need without us having to
694 * worry about recursive locking in ->write_begin() and ->write_end(). */
695 static int ocfs2_write_zero_page(struct inode
*inode
,
698 struct address_space
*mapping
= inode
->i_mapping
;
702 handle_t
*handle
= NULL
;
705 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
706 /* ugh. in prepare/commit_write, if from==to==start of block, we
707 ** skip the prepare. make sure we never send an offset for the start
710 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
713 index
= size
>> PAGE_CACHE_SHIFT
;
715 page
= grab_cache_page(mapping
, index
);
722 ret
= ocfs2_prepare_write_nolock(inode
, page
, offset
, offset
);
728 if (ocfs2_should_order_data(inode
)) {
729 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
731 if (IS_ERR(handle
)) {
732 ret
= PTR_ERR(handle
);
738 /* must not update i_size! */
739 ret
= block_commit_write(page
, offset
, offset
);
746 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
749 page_cache_release(page
);
754 static int ocfs2_zero_extend(struct inode
*inode
,
759 struct super_block
*sb
= inode
->i_sb
;
761 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
762 while (start_off
< zero_to_size
) {
763 ret
= ocfs2_write_zero_page(inode
, start_off
);
769 start_off
+= sb
->s_blocksize
;
772 * Very large extends have the potential to lock up
773 * the cpu for extended periods of time.
782 int ocfs2_extend_no_holes(struct inode
*inode
, u64 new_i_size
, u64 zero_to
)
786 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
788 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
);
789 if (clusters_to_add
< oi
->ip_clusters
)
792 clusters_to_add
-= oi
->ip_clusters
;
794 if (clusters_to_add
) {
795 ret
= __ocfs2_extend_allocation(inode
, oi
->ip_clusters
,
804 * Call this even if we don't add any clusters to the tree. We
805 * still need to zero the area between the old i_size and the
808 ret
= ocfs2_zero_extend(inode
, zero_to
);
816 static int ocfs2_extend_file(struct inode
*inode
,
817 struct buffer_head
*di_bh
,
821 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
825 /* setattr sometimes calls us like this. */
829 if (i_size_read(inode
) == new_i_size
)
831 BUG_ON(new_i_size
< i_size_read(inode
));
834 * Fall through for converting inline data, even if the fs
835 * supports sparse files.
837 * The check for inline data here is legal - nobody can add
838 * the feature since we have i_mutex. We must check it again
839 * after acquiring ip_alloc_sem though, as paths like mmap
840 * might have raced us to converting the inode to extents.
842 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
843 && ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
844 goto out_update_size
;
847 * The alloc sem blocks people in read/write from reading our
848 * allocation until we're done changing it. We depend on
849 * i_mutex to block other extend/truncate calls while we're
852 down_write(&oi
->ip_alloc_sem
);
854 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
856 * We can optimize small extends by keeping the inodes
859 if (ocfs2_size_fits_inline_data(di_bh
, new_i_size
)) {
860 up_write(&oi
->ip_alloc_sem
);
861 goto out_update_size
;
864 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
866 up_write(&oi
->ip_alloc_sem
);
873 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
874 ret
= ocfs2_extend_no_holes(inode
, new_i_size
, new_i_size
);
876 up_write(&oi
->ip_alloc_sem
);
884 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
892 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
894 int status
= 0, size_change
;
895 struct inode
*inode
= dentry
->d_inode
;
896 struct super_block
*sb
= inode
->i_sb
;
897 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
898 struct buffer_head
*bh
= NULL
;
899 handle_t
*handle
= NULL
;
901 struct dquot
*transfer_from
[MAXQUOTAS
] = { };
902 struct dquot
*transfer_to
[MAXQUOTAS
] = { };
904 mlog_entry("(0x%p, '%.*s')\n", dentry
,
905 dentry
->d_name
.len
, dentry
->d_name
.name
);
907 /* ensuring we don't even attempt to truncate a symlink */
908 if (S_ISLNK(inode
->i_mode
))
909 attr
->ia_valid
&= ~ATTR_SIZE
;
911 if (attr
->ia_valid
& ATTR_MODE
)
912 mlog(0, "mode change: %d\n", attr
->ia_mode
);
913 if (attr
->ia_valid
& ATTR_UID
)
914 mlog(0, "uid change: %d\n", attr
->ia_uid
);
915 if (attr
->ia_valid
& ATTR_GID
)
916 mlog(0, "gid change: %d\n", attr
->ia_gid
);
917 if (attr
->ia_valid
& ATTR_SIZE
)
918 mlog(0, "size change...\n");
919 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
920 mlog(0, "time change...\n");
922 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
923 | ATTR_GID | ATTR_UID | ATTR_MODE)
924 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
925 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
929 status
= inode_change_ok(inode
, attr
);
933 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
935 status
= ocfs2_rw_lock(inode
, 1);
942 status
= ocfs2_inode_lock(inode
, &bh
, 1);
944 if (status
!= -ENOENT
)
949 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
950 if (attr
->ia_size
> sb
->s_maxbytes
) {
955 if (i_size_read(inode
) > attr
->ia_size
) {
956 if (ocfs2_should_order_data(inode
)) {
957 status
= ocfs2_begin_ordered_truncate(inode
,
962 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
964 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
966 if (status
!= -ENOSPC
)
973 if ((attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
974 (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
976 * Gather pointers to quota structures so that allocation /
977 * freeing of quota structures happens here and not inside
978 * vfs_dq_transfer() where we have problems with lock ordering
980 if (attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
981 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
982 OCFS2_FEATURE_RO_COMPAT_USRQUOTA
)) {
983 transfer_to
[USRQUOTA
] = dqget(sb
, attr
->ia_uid
,
985 transfer_from
[USRQUOTA
] = dqget(sb
, inode
->i_uid
,
987 if (!transfer_to
[USRQUOTA
] || !transfer_from
[USRQUOTA
]) {
992 if (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
993 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
994 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
)) {
995 transfer_to
[GRPQUOTA
] = dqget(sb
, attr
->ia_gid
,
997 transfer_from
[GRPQUOTA
] = dqget(sb
, inode
->i_gid
,
999 if (!transfer_to
[GRPQUOTA
] || !transfer_from
[GRPQUOTA
]) {
1004 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
+
1005 2 * ocfs2_quota_trans_credits(sb
));
1006 if (IS_ERR(handle
)) {
1007 status
= PTR_ERR(handle
);
1011 status
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
1015 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1016 if (IS_ERR(handle
)) {
1017 status
= PTR_ERR(handle
);
1024 * This will intentionally not wind up calling vmtruncate(),
1025 * since all the work for a size change has been done above.
1026 * Otherwise, we could get into problems with truncate as
1027 * ip_alloc_sem is used there to protect against i_size
1030 status
= inode_setattr(inode
, attr
);
1036 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
1041 ocfs2_commit_trans(osb
, handle
);
1043 ocfs2_inode_unlock(inode
, 1);
1046 ocfs2_rw_unlock(inode
, 1);
1050 /* Release quota pointers in case we acquired them */
1051 for (qtype
= 0; qtype
< MAXQUOTAS
; qtype
++) {
1052 dqput(transfer_to
[qtype
]);
1053 dqput(transfer_from
[qtype
]);
1056 if (!status
&& attr
->ia_valid
& ATTR_MODE
) {
1057 status
= ocfs2_acl_chmod(inode
);
1066 int ocfs2_getattr(struct vfsmount
*mnt
,
1067 struct dentry
*dentry
,
1070 struct inode
*inode
= dentry
->d_inode
;
1071 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1072 struct ocfs2_super
*osb
= sb
->s_fs_info
;
1077 err
= ocfs2_inode_revalidate(dentry
);
1084 generic_fillattr(inode
, stat
);
1086 /* We set the blksize from the cluster size for performance */
1087 stat
->blksize
= osb
->s_clustersize
;
1095 int ocfs2_permission(struct inode
*inode
, int mask
)
1101 ret
= ocfs2_inode_lock(inode
, NULL
, 0);
1108 ret
= generic_permission(inode
, mask
, ocfs2_check_acl
);
1110 ocfs2_inode_unlock(inode
, 0);
1116 static int __ocfs2_write_remove_suid(struct inode
*inode
,
1117 struct buffer_head
*bh
)
1121 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1122 struct ocfs2_dinode
*di
;
1124 mlog_entry("(Inode %llu, mode 0%o)\n",
1125 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_mode
);
1127 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1128 if (IS_ERR(handle
)) {
1129 ret
= PTR_ERR(handle
);
1134 ret
= ocfs2_journal_access_di(handle
, inode
, bh
,
1135 OCFS2_JOURNAL_ACCESS_WRITE
);
1141 inode
->i_mode
&= ~S_ISUID
;
1142 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
1143 inode
->i_mode
&= ~S_ISGID
;
1145 di
= (struct ocfs2_dinode
*) bh
->b_data
;
1146 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
1148 ret
= ocfs2_journal_dirty(handle
, bh
);
1153 ocfs2_commit_trans(osb
, handle
);
1160 * Will look for holes and unwritten extents in the range starting at
1161 * pos for count bytes (inclusive).
1163 static int ocfs2_check_range_for_holes(struct inode
*inode
, loff_t pos
,
1167 unsigned int extent_flags
;
1168 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1169 struct super_block
*sb
= inode
->i_sb
;
1171 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1172 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1175 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1182 if (phys_cpos
== 0 || (extent_flags
& OCFS2_EXT_UNWRITTEN
)) {
1187 if (extent_len
> clusters
)
1188 extent_len
= clusters
;
1190 clusters
-= extent_len
;
1197 static int ocfs2_write_remove_suid(struct inode
*inode
)
1200 struct buffer_head
*bh
= NULL
;
1202 ret
= ocfs2_read_inode_block(inode
, &bh
);
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
;
1224 u64 end
= start
+ len
;
1225 struct buffer_head
*di_bh
= NULL
;
1227 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1228 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
1235 * Nothing to do if the requested reservation range
1236 * fits within the inode.
1238 if (ocfs2_size_fits_inline_data(di_bh
, end
))
1241 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
1249 * We consider both start and len to be inclusive.
1251 cpos
= start
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1252 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, start
+ len
);
1256 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1264 * Hole or existing extent len can be arbitrary, so
1265 * cap it to our own allocation request.
1267 if (alloc_size
> clusters
)
1268 alloc_size
= clusters
;
1272 * We already have an allocation at this
1273 * region so we can safely skip it.
1278 ret
= __ocfs2_extend_allocation(inode
, cpos
, alloc_size
, 1);
1287 clusters
-= alloc_size
;
1298 * Truncate a byte range, avoiding pages within partial clusters. This
1299 * preserves those pages for the zeroing code to write to.
1301 static void ocfs2_truncate_cluster_pages(struct inode
*inode
, u64 byte_start
,
1304 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1306 struct address_space
*mapping
= inode
->i_mapping
;
1308 start
= (loff_t
)ocfs2_align_bytes_to_clusters(inode
->i_sb
, byte_start
);
1309 end
= byte_start
+ byte_len
;
1310 end
= end
& ~(osb
->s_clustersize
- 1);
1313 unmap_mapping_range(mapping
, start
, end
- start
, 0);
1314 truncate_inode_pages_range(mapping
, start
, end
- 1);
1318 static int ocfs2_zero_partial_clusters(struct inode
*inode
,
1322 u64 tmpend
, end
= start
+ len
;
1323 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1324 unsigned int csize
= osb
->s_clustersize
;
1328 * The "start" and "end" values are NOT necessarily part of
1329 * the range whose allocation is being deleted. Rather, this
1330 * is what the user passed in with the request. We must zero
1331 * partial clusters here. There's no need to worry about
1332 * physical allocation - the zeroing code knows to skip holes.
1334 mlog(0, "byte start: %llu, end: %llu\n",
1335 (unsigned long long)start
, (unsigned long long)end
);
1338 * If both edges are on a cluster boundary then there's no
1339 * zeroing required as the region is part of the allocation to
1342 if ((start
& (csize
- 1)) == 0 && (end
& (csize
- 1)) == 0)
1345 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1346 if (IS_ERR(handle
)) {
1347 ret
= PTR_ERR(handle
);
1353 * We want to get the byte offset of the end of the 1st cluster.
1355 tmpend
= (u64
)osb
->s_clustersize
+ (start
& ~(osb
->s_clustersize
- 1));
1359 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1360 (unsigned long long)start
, (unsigned long long)tmpend
);
1362 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, tmpend
);
1368 * This may make start and end equal, but the zeroing
1369 * code will skip any work in that case so there's no
1370 * need to catch it up here.
1372 start
= end
& ~(osb
->s_clustersize
- 1);
1374 mlog(0, "2nd range: start: %llu, end: %llu\n",
1375 (unsigned long long)start
, (unsigned long long)end
);
1377 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, end
);
1382 ocfs2_commit_trans(osb
, handle
);
1387 static int ocfs2_remove_inode_range(struct inode
*inode
,
1388 struct buffer_head
*di_bh
, u64 byte_start
,
1392 u32 trunc_start
, trunc_len
, cpos
, phys_cpos
, alloc_size
;
1393 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1394 struct ocfs2_cached_dealloc_ctxt dealloc
;
1395 struct address_space
*mapping
= inode
->i_mapping
;
1396 struct ocfs2_extent_tree et
;
1398 ocfs2_init_dinode_extent_tree(&et
, inode
, di_bh
);
1399 ocfs2_init_dealloc_ctxt(&dealloc
);
1404 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1405 ret
= ocfs2_truncate_inline(inode
, di_bh
, byte_start
,
1406 byte_start
+ byte_len
, 0);
1412 * There's no need to get fancy with the page cache
1413 * truncate of an inline-data inode. We're talking
1414 * about less than a page here, which will be cached
1415 * in the dinode buffer anyway.
1417 unmap_mapping_range(mapping
, 0, 0, 0);
1418 truncate_inode_pages(mapping
, 0);
1422 trunc_start
= ocfs2_clusters_for_bytes(osb
->sb
, byte_start
);
1423 trunc_len
= (byte_start
+ byte_len
) >> osb
->s_clustersize_bits
;
1424 if (trunc_len
>= trunc_start
)
1425 trunc_len
-= trunc_start
;
1429 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1430 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1431 (unsigned long long)byte_start
,
1432 (unsigned long long)byte_len
, trunc_start
, trunc_len
);
1434 ret
= ocfs2_zero_partial_clusters(inode
, byte_start
, byte_len
);
1442 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1449 if (alloc_size
> trunc_len
)
1450 alloc_size
= trunc_len
;
1452 /* Only do work for non-holes */
1453 if (phys_cpos
!= 0) {
1454 ret
= ocfs2_remove_btree_range(inode
, &et
, cpos
,
1455 phys_cpos
, alloc_size
,
1464 trunc_len
-= alloc_size
;
1467 ocfs2_truncate_cluster_pages(inode
, byte_start
, byte_len
);
1470 ocfs2_schedule_truncate_log_flush(osb
, 1);
1471 ocfs2_run_deallocs(osb
, &dealloc
);
1477 * Parts of this function taken from xfs_change_file_space()
1479 static int __ocfs2_change_file_space(struct file
*file
, struct inode
*inode
,
1480 loff_t f_pos
, unsigned int cmd
,
1481 struct ocfs2_space_resv
*sr
,
1487 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1488 struct buffer_head
*di_bh
= NULL
;
1490 unsigned long long max_off
= inode
->i_sb
->s_maxbytes
;
1492 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
1495 mutex_lock(&inode
->i_mutex
);
1498 * This prevents concurrent writes on other nodes
1500 ret
= ocfs2_rw_lock(inode
, 1);
1506 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1512 if (inode
->i_flags
& (S_IMMUTABLE
|S_APPEND
)) {
1514 goto out_inode_unlock
;
1517 switch (sr
->l_whence
) {
1518 case 0: /*SEEK_SET*/
1520 case 1: /*SEEK_CUR*/
1521 sr
->l_start
+= f_pos
;
1523 case 2: /*SEEK_END*/
1524 sr
->l_start
+= i_size_read(inode
);
1528 goto out_inode_unlock
;
1532 llen
= sr
->l_len
> 0 ? sr
->l_len
- 1 : sr
->l_len
;
1535 || sr
->l_start
> max_off
1536 || (sr
->l_start
+ llen
) < 0
1537 || (sr
->l_start
+ llen
) > max_off
) {
1539 goto out_inode_unlock
;
1541 size
= sr
->l_start
+ sr
->l_len
;
1543 if (cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) {
1544 if (sr
->l_len
<= 0) {
1546 goto out_inode_unlock
;
1550 if (file
&& should_remove_suid(file
->f_path
.dentry
)) {
1551 ret
= __ocfs2_write_remove_suid(inode
, di_bh
);
1554 goto out_inode_unlock
;
1558 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1560 case OCFS2_IOC_RESVSP
:
1561 case OCFS2_IOC_RESVSP64
:
1563 * This takes unsigned offsets, but the signed ones we
1564 * pass have been checked against overflow above.
1566 ret
= ocfs2_allocate_unwritten_extents(inode
, sr
->l_start
,
1569 case OCFS2_IOC_UNRESVSP
:
1570 case OCFS2_IOC_UNRESVSP64
:
1571 ret
= ocfs2_remove_inode_range(inode
, di_bh
, sr
->l_start
,
1577 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1580 goto out_inode_unlock
;
1584 * We update c/mtime for these changes
1586 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1587 if (IS_ERR(handle
)) {
1588 ret
= PTR_ERR(handle
);
1590 goto out_inode_unlock
;
1593 if (change_size
&& i_size_read(inode
) < size
)
1594 i_size_write(inode
, size
);
1596 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1597 ret
= ocfs2_mark_inode_dirty(handle
, inode
, di_bh
);
1601 ocfs2_commit_trans(osb
, handle
);
1605 ocfs2_inode_unlock(inode
, 1);
1607 ocfs2_rw_unlock(inode
, 1);
1610 mutex_unlock(&inode
->i_mutex
);
1614 int ocfs2_change_file_space(struct file
*file
, unsigned int cmd
,
1615 struct ocfs2_space_resv
*sr
)
1617 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1618 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1620 if ((cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) &&
1621 !ocfs2_writes_unwritten_extents(osb
))
1623 else if ((cmd
== OCFS2_IOC_UNRESVSP
|| cmd
== OCFS2_IOC_UNRESVSP64
) &&
1624 !ocfs2_sparse_alloc(osb
))
1627 if (!S_ISREG(inode
->i_mode
))
1630 if (!(file
->f_mode
& FMODE_WRITE
))
1633 return __ocfs2_change_file_space(file
, inode
, file
->f_pos
, cmd
, sr
, 0);
1636 static long ocfs2_fallocate(struct inode
*inode
, int mode
, loff_t offset
,
1639 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1640 struct ocfs2_space_resv sr
;
1641 int change_size
= 1;
1643 if (!ocfs2_writes_unwritten_extents(osb
))
1646 if (S_ISDIR(inode
->i_mode
))
1649 if (mode
& FALLOC_FL_KEEP_SIZE
)
1653 sr
.l_start
= (s64
)offset
;
1654 sr
.l_len
= (s64
)len
;
1656 return __ocfs2_change_file_space(NULL
, inode
, offset
,
1657 OCFS2_IOC_RESVSP64
, &sr
, change_size
);
1660 static int ocfs2_prepare_inode_for_write(struct dentry
*dentry
,
1666 int ret
= 0, meta_level
= 0;
1667 struct inode
*inode
= dentry
->d_inode
;
1668 loff_t saved_pos
, end
;
1671 * We start with a read level meta lock and only jump to an ex
1672 * if we need to make modifications here.
1675 ret
= ocfs2_inode_lock(inode
, NULL
, meta_level
);
1682 /* Clear suid / sgid if necessary. We do this here
1683 * instead of later in the write path because
1684 * remove_suid() calls ->setattr without any hint that
1685 * we may have already done our cluster locking. Since
1686 * ocfs2_setattr() *must* take cluster locks to
1687 * proceeed, this will lead us to recursively lock the
1688 * inode. There's also the dinode i_size state which
1689 * can be lost via setattr during extending writes (we
1690 * set inode->i_size at the end of a write. */
1691 if (should_remove_suid(dentry
)) {
1692 if (meta_level
== 0) {
1693 ocfs2_inode_unlock(inode
, meta_level
);
1698 ret
= ocfs2_write_remove_suid(inode
);
1705 /* work on a copy of ppos until we're sure that we won't have
1706 * to recalculate it due to relocking. */
1708 saved_pos
= i_size_read(inode
);
1709 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1714 end
= saved_pos
+ count
;
1717 * Skip the O_DIRECT checks if we don't need
1720 if (!direct_io
|| !(*direct_io
))
1724 * There's no sane way to do direct writes to an inode
1727 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1733 * Allowing concurrent direct writes means
1734 * i_size changes wouldn't be synchronized, so
1735 * one node could wind up truncating another
1738 if (end
> i_size_read(inode
)) {
1744 * We don't fill holes during direct io, so
1745 * check for them here. If any are found, the
1746 * caller will have to retake some cluster
1747 * locks and initiate the io as buffered.
1749 ret
= ocfs2_check_range_for_holes(inode
, saved_pos
, count
);
1762 ocfs2_inode_unlock(inode
, meta_level
);
1768 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
1769 const struct iovec
*iov
,
1770 unsigned long nr_segs
,
1773 int ret
, direct_io
, appending
, rw_level
, have_alloc_sem
= 0;
1775 ssize_t written
= 0;
1776 size_t ocount
; /* original count */
1777 size_t count
; /* after file limit checks */
1778 loff_t old_size
, *ppos
= &iocb
->ki_pos
;
1780 struct file
*file
= iocb
->ki_filp
;
1781 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1782 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1784 mlog_entry("(0x%p, %u, '%.*s')\n", file
,
1785 (unsigned int)nr_segs
,
1786 file
->f_path
.dentry
->d_name
.len
,
1787 file
->f_path
.dentry
->d_name
.name
);
1789 if (iocb
->ki_left
== 0)
1792 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1794 appending
= file
->f_flags
& O_APPEND
? 1 : 0;
1795 direct_io
= file
->f_flags
& O_DIRECT
? 1 : 0;
1797 mutex_lock(&inode
->i_mutex
);
1800 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1802 down_read(&inode
->i_alloc_sem
);
1806 /* concurrent O_DIRECT writes are allowed */
1807 rw_level
= !direct_io
;
1808 ret
= ocfs2_rw_lock(inode
, rw_level
);
1814 can_do_direct
= direct_io
;
1815 ret
= ocfs2_prepare_inode_for_write(file
->f_path
.dentry
, ppos
,
1816 iocb
->ki_left
, appending
,
1824 * We can't complete the direct I/O as requested, fall back to
1827 if (direct_io
&& !can_do_direct
) {
1828 ocfs2_rw_unlock(inode
, rw_level
);
1829 up_read(&inode
->i_alloc_sem
);
1839 * To later detect whether a journal commit for sync writes is
1840 * necessary, we sample i_size, and cluster count here.
1842 old_size
= i_size_read(inode
);
1843 old_clusters
= OCFS2_I(inode
)->ip_clusters
;
1845 /* communicate with ocfs2_dio_end_io */
1846 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
1849 ret
= generic_segment_checks(iov
, &nr_segs
, &ocount
,
1854 ret
= generic_write_checks(file
, ppos
, &count
,
1855 S_ISBLK(inode
->i_mode
));
1859 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, *ppos
,
1860 ppos
, count
, ocount
);
1863 * direct write may have instantiated a few
1864 * blocks outside i_size. Trim these off again.
1865 * Don't need i_size_read because we hold i_mutex.
1867 if (*ppos
+ count
> inode
->i_size
)
1868 vmtruncate(inode
, inode
->i_size
);
1873 written
= generic_file_aio_write_nolock(iocb
, iov
, nr_segs
,
1878 /* buffered aio wouldn't have proper lock coverage today */
1879 BUG_ON(ret
== -EIOCBQUEUED
&& !(file
->f_flags
& O_DIRECT
));
1881 if ((file
->f_flags
& O_SYNC
&& !direct_io
) || IS_SYNC(inode
)) {
1883 * The generic write paths have handled getting data
1884 * to disk, but since we don't make use of the dirty
1885 * inode list, a manual journal commit is necessary
1888 if (old_size
!= i_size_read(inode
) ||
1889 old_clusters
!= OCFS2_I(inode
)->ip_clusters
) {
1890 ret
= jbd2_journal_force_commit(osb
->journal
->j_journal
);
1897 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1898 * function pointer which is called when o_direct io completes so that
1899 * it can unlock our rw lock. (it's the clustered equivalent of
1900 * i_alloc_sem; protects truncate from racing with pending ios).
1901 * Unfortunately there are error cases which call end_io and others
1902 * that don't. so we don't have to unlock the rw_lock if either an
1903 * async dio is going to do it in the future or an end_io after an
1904 * error has already done it.
1906 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1913 ocfs2_rw_unlock(inode
, rw_level
);
1917 up_read(&inode
->i_alloc_sem
);
1919 mutex_unlock(&inode
->i_mutex
);
1927 static int ocfs2_splice_to_file(struct pipe_inode_info
*pipe
,
1929 struct splice_desc
*sd
)
1933 ret
= ocfs2_prepare_inode_for_write(out
->f_path
.dentry
, &sd
->pos
,
1934 sd
->total_len
, 0, NULL
);
1940 return splice_from_pipe_feed(pipe
, sd
, pipe_to_file
);
1943 static ssize_t
ocfs2_file_splice_write(struct pipe_inode_info
*pipe
,
1950 struct address_space
*mapping
= out
->f_mapping
;
1951 struct inode
*inode
= mapping
->host
;
1952 struct splice_desc sd
= {
1959 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out
, pipe
,
1961 out
->f_path
.dentry
->d_name
.len
,
1962 out
->f_path
.dentry
->d_name
.name
);
1965 mutex_lock_nested(&pipe
->inode
->i_mutex
, I_MUTEX_PARENT
);
1967 splice_from_pipe_begin(&sd
);
1969 ret
= splice_from_pipe_next(pipe
, &sd
);
1973 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
1974 ret
= ocfs2_rw_lock(inode
, 1);
1978 ret
= ocfs2_splice_to_file(pipe
, out
, &sd
);
1979 ocfs2_rw_unlock(inode
, 1);
1981 mutex_unlock(&inode
->i_mutex
);
1983 splice_from_pipe_end(pipe
, &sd
);
1986 mutex_unlock(&pipe
->inode
->i_mutex
);
1989 ret
= sd
.num_spliced
;
1992 unsigned long nr_pages
;
1995 nr_pages
= (ret
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1998 * If file or inode is SYNC and we actually wrote some data,
2001 if (unlikely((out
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
2004 mutex_lock(&inode
->i_mutex
);
2005 err
= ocfs2_rw_lock(inode
, 1);
2009 err
= generic_osync_inode(inode
, mapping
,
2010 OSYNC_METADATA
|OSYNC_DATA
);
2011 ocfs2_rw_unlock(inode
, 1);
2013 mutex_unlock(&inode
->i_mutex
);
2018 balance_dirty_pages_ratelimited_nr(mapping
, nr_pages
);
2025 static ssize_t
ocfs2_file_splice_read(struct file
*in
,
2027 struct pipe_inode_info
*pipe
,
2031 int ret
= 0, lock_level
= 0;
2032 struct inode
*inode
= in
->f_path
.dentry
->d_inode
;
2034 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in
, pipe
,
2036 in
->f_path
.dentry
->d_name
.len
,
2037 in
->f_path
.dentry
->d_name
.name
);
2040 * See the comment in ocfs2_file_aio_read()
2042 ret
= ocfs2_inode_lock_atime(inode
, in
->f_vfsmnt
, &lock_level
);
2047 ocfs2_inode_unlock(inode
, lock_level
);
2049 ret
= generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
2056 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
2057 const struct iovec
*iov
,
2058 unsigned long nr_segs
,
2061 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0, lock_level
= 0;
2062 struct file
*filp
= iocb
->ki_filp
;
2063 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
2065 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
2066 (unsigned int)nr_segs
,
2067 filp
->f_path
.dentry
->d_name
.len
,
2068 filp
->f_path
.dentry
->d_name
.name
);
2077 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2078 * need locks to protect pending reads from racing with truncate.
2080 if (filp
->f_flags
& O_DIRECT
) {
2081 down_read(&inode
->i_alloc_sem
);
2084 ret
= ocfs2_rw_lock(inode
, 0);
2090 /* communicate with ocfs2_dio_end_io */
2091 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2095 * We're fine letting folks race truncates and extending
2096 * writes with read across the cluster, just like they can
2097 * locally. Hence no rw_lock during read.
2099 * Take and drop the meta data lock to update inode fields
2100 * like i_size. This allows the checks down below
2101 * generic_file_aio_read() a chance of actually working.
2103 ret
= ocfs2_inode_lock_atime(inode
, filp
->f_vfsmnt
, &lock_level
);
2108 ocfs2_inode_unlock(inode
, lock_level
);
2110 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
2112 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2114 /* buffered aio wouldn't have proper lock coverage today */
2115 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
2117 /* see ocfs2_file_aio_write */
2118 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
2125 up_read(&inode
->i_alloc_sem
);
2127 ocfs2_rw_unlock(inode
, rw_level
);
2133 const struct inode_operations ocfs2_file_iops
= {
2134 .setattr
= ocfs2_setattr
,
2135 .getattr
= ocfs2_getattr
,
2136 .permission
= ocfs2_permission
,
2137 .setxattr
= generic_setxattr
,
2138 .getxattr
= generic_getxattr
,
2139 .listxattr
= ocfs2_listxattr
,
2140 .removexattr
= generic_removexattr
,
2141 .fallocate
= ocfs2_fallocate
,
2142 .fiemap
= ocfs2_fiemap
,
2145 const struct inode_operations ocfs2_special_file_iops
= {
2146 .setattr
= ocfs2_setattr
,
2147 .getattr
= ocfs2_getattr
,
2148 .permission
= ocfs2_permission
,
2152 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2153 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2155 const struct file_operations ocfs2_fops
= {
2156 .llseek
= generic_file_llseek
,
2157 .read
= do_sync_read
,
2158 .write
= do_sync_write
,
2160 .fsync
= ocfs2_sync_file
,
2161 .release
= ocfs2_file_release
,
2162 .open
= ocfs2_file_open
,
2163 .aio_read
= ocfs2_file_aio_read
,
2164 .aio_write
= ocfs2_file_aio_write
,
2165 .unlocked_ioctl
= ocfs2_ioctl
,
2166 #ifdef CONFIG_COMPAT
2167 .compat_ioctl
= ocfs2_compat_ioctl
,
2170 .flock
= ocfs2_flock
,
2171 .splice_read
= ocfs2_file_splice_read
,
2172 .splice_write
= ocfs2_file_splice_write
,
2175 const struct file_operations ocfs2_dops
= {
2176 .llseek
= generic_file_llseek
,
2177 .read
= generic_read_dir
,
2178 .readdir
= ocfs2_readdir
,
2179 .fsync
= ocfs2_sync_file
,
2180 .release
= ocfs2_dir_release
,
2181 .open
= ocfs2_dir_open
,
2182 .unlocked_ioctl
= ocfs2_ioctl
,
2183 #ifdef CONFIG_COMPAT
2184 .compat_ioctl
= ocfs2_compat_ioctl
,
2187 .flock
= ocfs2_flock
,
2191 * POSIX-lockless variants of our file_operations.
2193 * These will be used if the underlying cluster stack does not support
2194 * posix file locking, if the user passes the "localflocks" mount
2195 * option, or if we have a local-only fs.
2197 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2198 * so we still want it in the case of no stack support for
2199 * plocks. Internally, it will do the right thing when asked to ignore
2202 const struct file_operations ocfs2_fops_no_plocks
= {
2203 .llseek
= generic_file_llseek
,
2204 .read
= do_sync_read
,
2205 .write
= do_sync_write
,
2207 .fsync
= ocfs2_sync_file
,
2208 .release
= ocfs2_file_release
,
2209 .open
= ocfs2_file_open
,
2210 .aio_read
= ocfs2_file_aio_read
,
2211 .aio_write
= ocfs2_file_aio_write
,
2212 .unlocked_ioctl
= ocfs2_ioctl
,
2213 #ifdef CONFIG_COMPAT
2214 .compat_ioctl
= ocfs2_compat_ioctl
,
2216 .flock
= ocfs2_flock
,
2217 .splice_read
= ocfs2_file_splice_read
,
2218 .splice_write
= ocfs2_file_splice_write
,
2221 const struct file_operations ocfs2_dops_no_plocks
= {
2222 .llseek
= generic_file_llseek
,
2223 .read
= generic_read_dir
,
2224 .readdir
= ocfs2_readdir
,
2225 .fsync
= ocfs2_sync_file
,
2226 .release
= ocfs2_dir_release
,
2227 .open
= ocfs2_dir_open
,
2228 .unlocked_ioctl
= ocfs2_ioctl
,
2229 #ifdef CONFIG_COMPAT
2230 .compat_ioctl
= ocfs2_compat_ioctl
,
2232 .flock
= ocfs2_flock
,