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"
62 #include "refcounttree.h"
64 #include "buffer_head_io.h"
66 static int ocfs2_sync_inode(struct inode
*inode
)
68 filemap_fdatawrite(inode
->i_mapping
);
69 return sync_mapping_buffers(inode
->i_mapping
);
72 static int ocfs2_init_file_private(struct inode
*inode
, struct file
*file
)
74 struct ocfs2_file_private
*fp
;
76 fp
= kzalloc(sizeof(struct ocfs2_file_private
), GFP_KERNEL
);
81 mutex_init(&fp
->fp_mutex
);
82 ocfs2_file_lock_res_init(&fp
->fp_flock
, fp
);
83 file
->private_data
= fp
;
88 static void ocfs2_free_file_private(struct inode
*inode
, struct file
*file
)
90 struct ocfs2_file_private
*fp
= file
->private_data
;
91 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
94 ocfs2_simple_drop_lockres(osb
, &fp
->fp_flock
);
95 ocfs2_lock_res_free(&fp
->fp_flock
);
97 file
->private_data
= NULL
;
101 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
104 int mode
= file
->f_flags
;
105 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
107 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
108 file
->f_path
.dentry
->d_name
.len
, file
->f_path
.dentry
->d_name
.name
);
110 if (file
->f_mode
& FMODE_WRITE
)
111 dquot_initialize(inode
);
113 spin_lock(&oi
->ip_lock
);
115 /* Check that the inode hasn't been wiped from disk by another
116 * node. If it hasn't then we're safe as long as we hold the
117 * spin lock until our increment of open count. */
118 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
119 spin_unlock(&oi
->ip_lock
);
126 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
129 spin_unlock(&oi
->ip_lock
);
131 status
= ocfs2_init_file_private(inode
, file
);
134 * We want to set open count back if we're failing the
137 spin_lock(&oi
->ip_lock
);
139 spin_unlock(&oi
->ip_lock
);
147 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
149 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
151 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
152 file
->f_path
.dentry
->d_name
.len
,
153 file
->f_path
.dentry
->d_name
.name
);
155 spin_lock(&oi
->ip_lock
);
156 if (!--oi
->ip_open_count
)
157 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
158 spin_unlock(&oi
->ip_lock
);
160 ocfs2_free_file_private(inode
, file
);
167 static int ocfs2_dir_open(struct inode
*inode
, struct file
*file
)
169 return ocfs2_init_file_private(inode
, file
);
172 static int ocfs2_dir_release(struct inode
*inode
, struct file
*file
)
174 ocfs2_free_file_private(inode
, file
);
178 static int ocfs2_sync_file(struct file
*file
, int datasync
)
182 struct dentry
*dentry
= file
->f_path
.dentry
;
183 struct inode
*inode
= file
->f_mapping
->host
;
184 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
186 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
187 dentry
->d_name
.len
, dentry
->d_name
.name
);
189 err
= ocfs2_sync_inode(dentry
->d_inode
);
193 if (datasync
&& !(inode
->i_state
& I_DIRTY_DATASYNC
))
196 journal
= osb
->journal
->j_journal
;
197 err
= jbd2_journal_force_commit(journal
);
202 return (err
< 0) ? -EIO
: 0;
205 int ocfs2_should_update_atime(struct inode
*inode
,
206 struct vfsmount
*vfsmnt
)
209 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
211 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
214 if ((inode
->i_flags
& S_NOATIME
) ||
215 ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
219 * We can be called with no vfsmnt structure - NFSD will
222 * Note that our action here is different than touch_atime() -
223 * if we can't tell whether this is a noatime mount, then we
224 * don't know whether to trust the value of s_atime_quantum.
229 if ((vfsmnt
->mnt_flags
& MNT_NOATIME
) ||
230 ((vfsmnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
233 if (vfsmnt
->mnt_flags
& MNT_RELATIME
) {
234 if ((timespec_compare(&inode
->i_atime
, &inode
->i_mtime
) <= 0) ||
235 (timespec_compare(&inode
->i_atime
, &inode
->i_ctime
) <= 0))
242 if ((now
.tv_sec
- inode
->i_atime
.tv_sec
<= osb
->s_atime_quantum
))
248 int ocfs2_update_inode_atime(struct inode
*inode
,
249 struct buffer_head
*bh
)
252 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
254 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) bh
->b_data
;
258 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
259 if (IS_ERR(handle
)) {
260 ret
= PTR_ERR(handle
);
265 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
266 OCFS2_JOURNAL_ACCESS_WRITE
);
273 * Don't use ocfs2_mark_inode_dirty() here as we don't always
274 * have i_mutex to guard against concurrent changes to other
277 inode
->i_atime
= CURRENT_TIME
;
278 di
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
279 di
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
280 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_cow_file_pos(struct inode
*inode
,
338 struct buffer_head
*fe_bh
,
342 u32 phys
, cpos
= offset
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
343 unsigned int num_clusters
= 0;
344 unsigned int ext_flags
= 0;
347 * If the new offset is aligned to the range of the cluster, there is
348 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
351 if ((offset
& (OCFS2_SB(inode
->i_sb
)->s_clustersize
- 1)) == 0)
354 status
= ocfs2_get_clusters(inode
, cpos
, &phys
,
355 &num_clusters
, &ext_flags
);
361 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
364 return ocfs2_refcount_cow(inode
, fe_bh
, cpos
, 1, cpos
+1);
370 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
372 struct buffer_head
*fe_bh
,
377 struct ocfs2_dinode
*di
;
383 * We need to CoW the cluster contains the offset if it is reflinked
384 * since we will call ocfs2_zero_range_for_truncate later which will
385 * write "0" from offset to the end of the cluster.
387 status
= ocfs2_cow_file_pos(inode
, fe_bh
, new_i_size
);
393 /* TODO: This needs to actually orphan the inode in this
396 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
397 if (IS_ERR(handle
)) {
398 status
= PTR_ERR(handle
);
403 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
404 OCFS2_JOURNAL_ACCESS_WRITE
);
411 * Do this before setting i_size.
413 cluster_bytes
= ocfs2_align_bytes_to_clusters(inode
->i_sb
, new_i_size
);
414 status
= ocfs2_zero_range_for_truncate(inode
, handle
, new_i_size
,
421 i_size_write(inode
, new_i_size
);
422 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
424 di
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
425 di
->i_size
= cpu_to_le64(new_i_size
);
426 di
->i_ctime
= di
->i_mtime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
427 di
->i_ctime_nsec
= di
->i_mtime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
429 ocfs2_journal_dirty(handle
, fe_bh
);
432 ocfs2_commit_trans(osb
, handle
);
439 static int ocfs2_truncate_file(struct inode
*inode
,
440 struct buffer_head
*di_bh
,
444 struct ocfs2_dinode
*fe
= NULL
;
445 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
447 mlog_entry("(inode = %llu, new_i_size = %llu\n",
448 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
449 (unsigned long long)new_i_size
);
451 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
452 * already validated it */
453 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
455 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
456 "Inode %llu, inode i_size = %lld != di "
457 "i_size = %llu, i_flags = 0x%x\n",
458 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
460 (unsigned long long)le64_to_cpu(fe
->i_size
),
461 le32_to_cpu(fe
->i_flags
));
463 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
464 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
465 (unsigned long long)le64_to_cpu(fe
->i_size
),
466 (unsigned long long)new_i_size
);
472 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
473 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
474 (unsigned long long)le64_to_cpu(fe
->i_size
),
475 (unsigned long long)new_i_size
);
477 /* lets handle the simple truncate cases before doing any more
478 * cluster locking. */
479 if (new_i_size
== le64_to_cpu(fe
->i_size
))
482 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
484 ocfs2_resv_discard(&osb
->osb_la_resmap
,
485 &OCFS2_I(inode
)->ip_la_data_resv
);
488 * The inode lock forced other nodes to sync and drop their
489 * pages, which (correctly) happens even if we have a truncate
490 * without allocation change - ocfs2 cluster sizes can be much
491 * greater than page size, so we have to truncate them
494 unmap_mapping_range(inode
->i_mapping
, new_i_size
+ PAGE_SIZE
- 1, 0, 1);
495 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
497 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
498 status
= ocfs2_truncate_inline(inode
, di_bh
, new_i_size
,
499 i_size_read(inode
), 1);
503 goto bail_unlock_sem
;
506 /* alright, we're going to need to do a full blown alloc size
507 * change. Orphan the inode so that recovery can complete the
508 * truncate if necessary. This does the task of marking
510 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
513 goto bail_unlock_sem
;
516 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
);
519 goto bail_unlock_sem
;
522 /* TODO: orphan dir cleanup here. */
524 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
527 if (!status
&& OCFS2_I(inode
)->ip_clusters
== 0)
528 status
= ocfs2_try_remove_refcount_tree(inode
, di_bh
);
535 * extend file allocation only here.
536 * we'll update all the disk stuff, and oip->alloc_size
538 * expect stuff to be locked, a transaction started and enough data /
539 * metadata reservations in the contexts.
541 * Will return -EAGAIN, and a reason if a restart is needed.
542 * If passed in, *reason will always be set, even in error.
544 int ocfs2_add_inode_data(struct ocfs2_super
*osb
,
549 struct buffer_head
*fe_bh
,
551 struct ocfs2_alloc_context
*data_ac
,
552 struct ocfs2_alloc_context
*meta_ac
,
553 enum ocfs2_alloc_restarted
*reason_ret
)
556 struct ocfs2_extent_tree et
;
558 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), fe_bh
);
559 ret
= ocfs2_add_clusters_in_btree(handle
, &et
, logical_offset
,
560 clusters_to_add
, mark_unwritten
,
561 data_ac
, meta_ac
, reason_ret
);
566 static int __ocfs2_extend_allocation(struct inode
*inode
, u32 logical_start
,
567 u32 clusters_to_add
, int mark_unwritten
)
570 int restart_func
= 0;
573 struct buffer_head
*bh
= NULL
;
574 struct ocfs2_dinode
*fe
= NULL
;
575 handle_t
*handle
= NULL
;
576 struct ocfs2_alloc_context
*data_ac
= NULL
;
577 struct ocfs2_alloc_context
*meta_ac
= NULL
;
578 enum ocfs2_alloc_restarted why
;
579 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
580 struct ocfs2_extent_tree et
;
583 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
586 * This function only exists for file systems which don't
589 BUG_ON(mark_unwritten
&& !ocfs2_sparse_alloc(osb
));
591 status
= ocfs2_read_inode_block(inode
, &bh
);
596 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
599 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
601 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
602 "clusters_to_add = %u\n",
603 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
604 (long long)i_size_read(inode
), le32_to_cpu(fe
->i_clusters
),
606 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), bh
);
607 status
= ocfs2_lock_allocators(inode
, &et
, clusters_to_add
, 0,
614 credits
= ocfs2_calc_extend_credits(osb
->sb
, &fe
->id2
.i_list
,
616 handle
= ocfs2_start_trans(osb
, credits
);
617 if (IS_ERR(handle
)) {
618 status
= PTR_ERR(handle
);
624 restarted_transaction
:
625 status
= dquot_alloc_space_nodirty(inode
,
626 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
631 /* reserve a write to the file entry early on - that we if we
632 * run out of credits in the allocation path, we can still
634 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
635 OCFS2_JOURNAL_ACCESS_WRITE
);
641 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
643 status
= ocfs2_add_inode_data(osb
,
653 if ((status
< 0) && (status
!= -EAGAIN
)) {
654 if (status
!= -ENOSPC
)
659 ocfs2_journal_dirty(handle
, bh
);
661 spin_lock(&OCFS2_I(inode
)->ip_lock
);
662 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
663 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
664 /* Release unused quota reservation */
665 dquot_free_space(inode
,
666 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
669 if (why
!= RESTART_NONE
&& clusters_to_add
) {
670 if (why
== RESTART_META
) {
671 mlog(0, "restarting function.\n");
675 BUG_ON(why
!= RESTART_TRANS
);
677 mlog(0, "restarting transaction.\n");
678 /* TODO: This can be more intelligent. */
679 credits
= ocfs2_calc_extend_credits(osb
->sb
,
682 status
= ocfs2_extend_trans(handle
, credits
);
684 /* handle still has to be committed at
690 goto restarted_transaction
;
694 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
695 le32_to_cpu(fe
->i_clusters
),
696 (unsigned long long)le64_to_cpu(fe
->i_size
));
697 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
698 OCFS2_I(inode
)->ip_clusters
, (long long)i_size_read(inode
));
701 if (status
< 0 && did_quota
)
702 dquot_free_space(inode
,
703 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
705 ocfs2_commit_trans(osb
, handle
);
709 ocfs2_free_alloc_context(data_ac
);
713 ocfs2_free_alloc_context(meta_ac
);
716 if ((!status
) && restart_func
) {
728 * While a write will already be ordering the data, a truncate will not.
729 * Thus, we need to explicitly order the zeroed pages.
731 static handle_t
*ocfs2_zero_start_ordered_transaction(struct inode
*inode
)
733 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
734 handle_t
*handle
= NULL
;
737 if (!ocfs2_should_order_data(inode
))
740 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
741 if (IS_ERR(handle
)) {
747 ret
= ocfs2_jbd2_file_inode(handle
, inode
);
754 ocfs2_commit_trans(osb
, handle
);
755 handle
= ERR_PTR(ret
);
760 /* Some parts of this taken from generic_cont_expand, which turned out
761 * to be too fragile to do exactly what we need without us having to
762 * worry about recursive locking in ->write_begin() and ->write_end(). */
763 static int ocfs2_write_zero_page(struct inode
*inode
, u64 abs_from
,
766 struct address_space
*mapping
= inode
->i_mapping
;
768 unsigned long index
= abs_from
>> PAGE_CACHE_SHIFT
;
769 handle_t
*handle
= NULL
;
771 unsigned zero_from
, zero_to
, block_start
, block_end
;
773 BUG_ON(abs_from
>= abs_to
);
774 BUG_ON(abs_to
> (((u64
)index
+ 1) << PAGE_CACHE_SHIFT
));
775 BUG_ON(abs_from
& (inode
->i_blkbits
- 1));
777 page
= grab_cache_page(mapping
, index
);
784 /* Get the offsets within the page that we want to zero */
785 zero_from
= abs_from
& (PAGE_CACHE_SIZE
- 1);
786 zero_to
= abs_to
& (PAGE_CACHE_SIZE
- 1);
788 zero_to
= PAGE_CACHE_SIZE
;
791 "abs_from = %llu, abs_to = %llu, index = %lu, zero_from = %u, zero_to = %u\n",
792 (unsigned long long)abs_from
, (unsigned long long)abs_to
,
793 index
, zero_from
, zero_to
);
795 /* We know that zero_from is block aligned */
796 for (block_start
= zero_from
; block_start
< zero_to
;
797 block_start
= block_end
) {
798 block_end
= block_start
+ (1 << inode
->i_blkbits
);
801 * block_start is block-aligned. Bump it by one to
802 * force ocfs2_{prepare,commit}_write() to zero the
805 ret
= ocfs2_prepare_write_nolock(inode
, page
,
814 handle
= ocfs2_zero_start_ordered_transaction(inode
);
815 if (IS_ERR(handle
)) {
816 ret
= PTR_ERR(handle
);
822 /* must not update i_size! */
823 ret
= block_commit_write(page
, block_start
+ 1,
832 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
836 page_cache_release(page
);
842 * Find the next range to zero. We do this in terms of bytes because
843 * that's what ocfs2_zero_extend() wants, and it is dealing with the
844 * pagecache. We may return multiple extents.
846 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
847 * needs to be zeroed. range_start and range_end return the next zeroing
848 * range. A subsequent call should pass the previous range_end as its
849 * zero_start. If range_end is 0, there's nothing to do.
851 * Unwritten extents are skipped over. Refcounted extents are CoWd.
853 static int ocfs2_zero_extend_get_range(struct inode
*inode
,
854 struct buffer_head
*di_bh
,
855 u64 zero_start
, u64 zero_end
,
856 u64
*range_start
, u64
*range_end
)
858 int rc
= 0, needs_cow
= 0;
859 u32 p_cpos
, zero_clusters
= 0;
861 zero_start
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
862 u32 last_cpos
= ocfs2_clusters_for_bytes(inode
->i_sb
, zero_end
);
863 unsigned int num_clusters
= 0;
864 unsigned int ext_flags
= 0;
866 while (zero_cpos
< last_cpos
) {
867 rc
= ocfs2_get_clusters(inode
, zero_cpos
, &p_cpos
,
868 &num_clusters
, &ext_flags
);
874 if (p_cpos
&& !(ext_flags
& OCFS2_EXT_UNWRITTEN
)) {
875 zero_clusters
= num_clusters
;
876 if (ext_flags
& OCFS2_EXT_REFCOUNTED
)
881 zero_cpos
+= num_clusters
;
883 if (!zero_clusters
) {
888 while ((zero_cpos
+ zero_clusters
) < last_cpos
) {
889 rc
= ocfs2_get_clusters(inode
, zero_cpos
+ zero_clusters
,
890 &p_cpos
, &num_clusters
,
897 if (!p_cpos
|| (ext_flags
& OCFS2_EXT_UNWRITTEN
))
899 if (ext_flags
& OCFS2_EXT_REFCOUNTED
)
901 zero_clusters
+= num_clusters
;
903 if ((zero_cpos
+ zero_clusters
) > last_cpos
)
904 zero_clusters
= last_cpos
- zero_cpos
;
907 rc
= ocfs2_refcount_cow(inode
, di_bh
, zero_cpos
, zero_clusters
,
915 *range_start
= ocfs2_clusters_to_bytes(inode
->i_sb
, zero_cpos
);
916 *range_end
= ocfs2_clusters_to_bytes(inode
->i_sb
,
917 zero_cpos
+ zero_clusters
);
924 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
925 * has made sure that the entire range needs zeroing.
927 static int ocfs2_zero_extend_range(struct inode
*inode
, u64 range_start
,
932 u64 zero_pos
= range_start
;
934 mlog(0, "range_start = %llu, range_end = %llu\n",
935 (unsigned long long)range_start
,
936 (unsigned long long)range_end
);
937 BUG_ON(range_start
>= range_end
);
939 while (zero_pos
< range_end
) {
940 next_pos
= (zero_pos
& PAGE_CACHE_MASK
) + PAGE_CACHE_SIZE
;
941 if (next_pos
> range_end
)
942 next_pos
= range_end
;
943 rc
= ocfs2_write_zero_page(inode
, zero_pos
, next_pos
);
951 * Very large extends have the potential to lock up
952 * the cpu for extended periods of time.
960 int ocfs2_zero_extend(struct inode
*inode
, struct buffer_head
*di_bh
,
964 u64 zero_start
, range_start
= 0, range_end
= 0;
965 struct super_block
*sb
= inode
->i_sb
;
967 zero_start
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
968 mlog(0, "zero_start %llu for i_size %llu\n",
969 (unsigned long long)zero_start
,
970 (unsigned long long)i_size_read(inode
));
971 while (zero_start
< zero_to_size
) {
972 ret
= ocfs2_zero_extend_get_range(inode
, di_bh
, zero_start
,
983 if (range_start
< zero_start
)
984 range_start
= zero_start
;
985 if (range_end
> zero_to_size
)
986 range_end
= zero_to_size
;
988 ret
= ocfs2_zero_extend_range(inode
, range_start
,
994 zero_start
= range_end
;
1000 int ocfs2_extend_no_holes(struct inode
*inode
, struct buffer_head
*di_bh
,
1001 u64 new_i_size
, u64 zero_to
)
1004 u32 clusters_to_add
;
1005 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1008 * Only quota files call this without a bh, and they can't be
1011 BUG_ON(!di_bh
&& (oi
->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
));
1012 BUG_ON(!di_bh
&& !(oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
));
1014 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
);
1015 if (clusters_to_add
< oi
->ip_clusters
)
1016 clusters_to_add
= 0;
1018 clusters_to_add
-= oi
->ip_clusters
;
1020 if (clusters_to_add
) {
1021 ret
= __ocfs2_extend_allocation(inode
, oi
->ip_clusters
,
1022 clusters_to_add
, 0);
1030 * Call this even if we don't add any clusters to the tree. We
1031 * still need to zero the area between the old i_size and the
1034 ret
= ocfs2_zero_extend(inode
, di_bh
, zero_to
);
1042 static int ocfs2_extend_file(struct inode
*inode
,
1043 struct buffer_head
*di_bh
,
1047 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1051 /* setattr sometimes calls us like this. */
1052 if (new_i_size
== 0)
1055 if (i_size_read(inode
) == new_i_size
)
1057 BUG_ON(new_i_size
< i_size_read(inode
));
1060 * The alloc sem blocks people in read/write from reading our
1061 * allocation until we're done changing it. We depend on
1062 * i_mutex to block other extend/truncate calls while we're
1063 * here. We even have to hold it for sparse files because there
1064 * might be some tail zeroing.
1066 down_write(&oi
->ip_alloc_sem
);
1068 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1070 * We can optimize small extends by keeping the inodes
1073 if (ocfs2_size_fits_inline_data(di_bh
, new_i_size
)) {
1074 up_write(&oi
->ip_alloc_sem
);
1075 goto out_update_size
;
1078 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
1080 up_write(&oi
->ip_alloc_sem
);
1086 if (ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
1087 ret
= ocfs2_zero_extend(inode
, di_bh
, new_i_size
);
1089 ret
= ocfs2_extend_no_holes(inode
, di_bh
, new_i_size
,
1092 up_write(&oi
->ip_alloc_sem
);
1100 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
1108 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1110 int status
= 0, size_change
;
1111 struct inode
*inode
= dentry
->d_inode
;
1112 struct super_block
*sb
= inode
->i_sb
;
1113 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
1114 struct buffer_head
*bh
= NULL
;
1115 handle_t
*handle
= NULL
;
1116 struct dquot
*transfer_to
[MAXQUOTAS
] = { };
1119 mlog_entry("(0x%p, '%.*s')\n", dentry
,
1120 dentry
->d_name
.len
, dentry
->d_name
.name
);
1122 /* ensuring we don't even attempt to truncate a symlink */
1123 if (S_ISLNK(inode
->i_mode
))
1124 attr
->ia_valid
&= ~ATTR_SIZE
;
1126 if (attr
->ia_valid
& ATTR_MODE
)
1127 mlog(0, "mode change: %d\n", attr
->ia_mode
);
1128 if (attr
->ia_valid
& ATTR_UID
)
1129 mlog(0, "uid change: %d\n", attr
->ia_uid
);
1130 if (attr
->ia_valid
& ATTR_GID
)
1131 mlog(0, "gid change: %d\n", attr
->ia_gid
);
1132 if (attr
->ia_valid
& ATTR_SIZE
)
1133 mlog(0, "size change...\n");
1134 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
1135 mlog(0, "time change...\n");
1137 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1138 | ATTR_GID | ATTR_UID | ATTR_MODE)
1139 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
1140 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
1144 status
= inode_change_ok(inode
, attr
);
1148 if (is_quota_modification(inode
, attr
))
1149 dquot_initialize(inode
);
1150 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
1152 status
= ocfs2_rw_lock(inode
, 1);
1159 status
= ocfs2_inode_lock(inode
, &bh
, 1);
1161 if (status
!= -ENOENT
)
1163 goto bail_unlock_rw
;
1166 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
1167 status
= inode_newsize_ok(inode
, attr
->ia_size
);
1171 if (i_size_read(inode
) > attr
->ia_size
) {
1172 if (ocfs2_should_order_data(inode
)) {
1173 status
= ocfs2_begin_ordered_truncate(inode
,
1178 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
1180 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
1182 if (status
!= -ENOSPC
)
1189 if ((attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
1190 (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
1192 * Gather pointers to quota structures so that allocation /
1193 * freeing of quota structures happens here and not inside
1194 * dquot_transfer() where we have problems with lock ordering
1196 if (attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
1197 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
1198 OCFS2_FEATURE_RO_COMPAT_USRQUOTA
)) {
1199 transfer_to
[USRQUOTA
] = dqget(sb
, attr
->ia_uid
,
1201 if (!transfer_to
[USRQUOTA
]) {
1206 if (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
1207 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
1208 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
)) {
1209 transfer_to
[GRPQUOTA
] = dqget(sb
, attr
->ia_gid
,
1211 if (!transfer_to
[GRPQUOTA
]) {
1216 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
+
1217 2 * ocfs2_quota_trans_credits(sb
));
1218 if (IS_ERR(handle
)) {
1219 status
= PTR_ERR(handle
);
1223 status
= __dquot_transfer(inode
, transfer_to
);
1227 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1228 if (IS_ERR(handle
)) {
1229 status
= PTR_ERR(handle
);
1236 * This will intentionally not wind up calling truncate_setsize(),
1237 * since all the work for a size change has been done above.
1238 * Otherwise, we could get into problems with truncate as
1239 * ip_alloc_sem is used there to protect against i_size
1242 * XXX: this means the conditional below can probably be removed.
1244 if ((attr
->ia_valid
& ATTR_SIZE
) &&
1245 attr
->ia_size
!= i_size_read(inode
)) {
1246 status
= vmtruncate(inode
, attr
->ia_size
);
1253 setattr_copy(inode
, attr
);
1254 mark_inode_dirty(inode
);
1256 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
1261 ocfs2_commit_trans(osb
, handle
);
1263 ocfs2_inode_unlock(inode
, 1);
1266 ocfs2_rw_unlock(inode
, 1);
1270 /* Release quota pointers in case we acquired them */
1271 for (qtype
= 0; qtype
< MAXQUOTAS
; qtype
++)
1272 dqput(transfer_to
[qtype
]);
1274 if (!status
&& attr
->ia_valid
& ATTR_MODE
) {
1275 status
= ocfs2_acl_chmod(inode
);
1284 int ocfs2_getattr(struct vfsmount
*mnt
,
1285 struct dentry
*dentry
,
1288 struct inode
*inode
= dentry
->d_inode
;
1289 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1290 struct ocfs2_super
*osb
= sb
->s_fs_info
;
1295 err
= ocfs2_inode_revalidate(dentry
);
1302 generic_fillattr(inode
, stat
);
1304 /* We set the blksize from the cluster size for performance */
1305 stat
->blksize
= osb
->s_clustersize
;
1313 int ocfs2_permission(struct inode
*inode
, int mask
)
1319 ret
= ocfs2_inode_lock(inode
, NULL
, 0);
1326 ret
= generic_permission(inode
, mask
, ocfs2_check_acl
);
1328 ocfs2_inode_unlock(inode
, 0);
1334 static int __ocfs2_write_remove_suid(struct inode
*inode
,
1335 struct buffer_head
*bh
)
1339 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1340 struct ocfs2_dinode
*di
;
1342 mlog_entry("(Inode %llu, mode 0%o)\n",
1343 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_mode
);
1345 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1346 if (IS_ERR(handle
)) {
1347 ret
= PTR_ERR(handle
);
1352 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
1353 OCFS2_JOURNAL_ACCESS_WRITE
);
1359 inode
->i_mode
&= ~S_ISUID
;
1360 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
1361 inode
->i_mode
&= ~S_ISGID
;
1363 di
= (struct ocfs2_dinode
*) bh
->b_data
;
1364 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
1366 ocfs2_journal_dirty(handle
, bh
);
1369 ocfs2_commit_trans(osb
, handle
);
1376 * Will look for holes and unwritten extents in the range starting at
1377 * pos for count bytes (inclusive).
1379 static int ocfs2_check_range_for_holes(struct inode
*inode
, loff_t pos
,
1383 unsigned int extent_flags
;
1384 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1385 struct super_block
*sb
= inode
->i_sb
;
1387 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1388 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1391 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1398 if (phys_cpos
== 0 || (extent_flags
& OCFS2_EXT_UNWRITTEN
)) {
1403 if (extent_len
> clusters
)
1404 extent_len
= clusters
;
1406 clusters
-= extent_len
;
1413 static int ocfs2_write_remove_suid(struct inode
*inode
)
1416 struct buffer_head
*bh
= NULL
;
1418 ret
= ocfs2_read_inode_block(inode
, &bh
);
1424 ret
= __ocfs2_write_remove_suid(inode
, bh
);
1431 * Allocate enough extents to cover the region starting at byte offset
1432 * start for len bytes. Existing extents are skipped, any extents
1433 * added are marked as "unwritten".
1435 static int ocfs2_allocate_unwritten_extents(struct inode
*inode
,
1439 u32 cpos
, phys_cpos
, clusters
, alloc_size
;
1440 u64 end
= start
+ len
;
1441 struct buffer_head
*di_bh
= NULL
;
1443 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1444 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
1451 * Nothing to do if the requested reservation range
1452 * fits within the inode.
1454 if (ocfs2_size_fits_inline_data(di_bh
, end
))
1457 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
1465 * We consider both start and len to be inclusive.
1467 cpos
= start
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1468 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, start
+ len
);
1472 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1480 * Hole or existing extent len can be arbitrary, so
1481 * cap it to our own allocation request.
1483 if (alloc_size
> clusters
)
1484 alloc_size
= clusters
;
1488 * We already have an allocation at this
1489 * region so we can safely skip it.
1494 ret
= __ocfs2_extend_allocation(inode
, cpos
, alloc_size
, 1);
1503 clusters
-= alloc_size
;
1514 * Truncate a byte range, avoiding pages within partial clusters. This
1515 * preserves those pages for the zeroing code to write to.
1517 static void ocfs2_truncate_cluster_pages(struct inode
*inode
, u64 byte_start
,
1520 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1522 struct address_space
*mapping
= inode
->i_mapping
;
1524 start
= (loff_t
)ocfs2_align_bytes_to_clusters(inode
->i_sb
, byte_start
);
1525 end
= byte_start
+ byte_len
;
1526 end
= end
& ~(osb
->s_clustersize
- 1);
1529 unmap_mapping_range(mapping
, start
, end
- start
, 0);
1530 truncate_inode_pages_range(mapping
, start
, end
- 1);
1534 static int ocfs2_zero_partial_clusters(struct inode
*inode
,
1538 u64 tmpend
, end
= start
+ len
;
1539 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1540 unsigned int csize
= osb
->s_clustersize
;
1544 * The "start" and "end" values are NOT necessarily part of
1545 * the range whose allocation is being deleted. Rather, this
1546 * is what the user passed in with the request. We must zero
1547 * partial clusters here. There's no need to worry about
1548 * physical allocation - the zeroing code knows to skip holes.
1550 mlog(0, "byte start: %llu, end: %llu\n",
1551 (unsigned long long)start
, (unsigned long long)end
);
1554 * If both edges are on a cluster boundary then there's no
1555 * zeroing required as the region is part of the allocation to
1558 if ((start
& (csize
- 1)) == 0 && (end
& (csize
- 1)) == 0)
1561 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1562 if (IS_ERR(handle
)) {
1563 ret
= PTR_ERR(handle
);
1569 * We want to get the byte offset of the end of the 1st cluster.
1571 tmpend
= (u64
)osb
->s_clustersize
+ (start
& ~(osb
->s_clustersize
- 1));
1575 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1576 (unsigned long long)start
, (unsigned long long)tmpend
);
1578 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, tmpend
);
1584 * This may make start and end equal, but the zeroing
1585 * code will skip any work in that case so there's no
1586 * need to catch it up here.
1588 start
= end
& ~(osb
->s_clustersize
- 1);
1590 mlog(0, "2nd range: start: %llu, end: %llu\n",
1591 (unsigned long long)start
, (unsigned long long)end
);
1593 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, end
);
1598 ocfs2_commit_trans(osb
, handle
);
1603 static int ocfs2_find_rec(struct ocfs2_extent_list
*el
, u32 pos
)
1606 struct ocfs2_extent_rec
*rec
= NULL
;
1608 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
1610 rec
= &el
->l_recs
[i
];
1612 if (le32_to_cpu(rec
->e_cpos
) < pos
)
1620 * Helper to calculate the punching pos and length in one run, we handle the
1621 * following three cases in order:
1623 * - remove the entire record
1624 * - remove a partial record
1625 * - no record needs to be removed (hole-punching completed)
1627 static void ocfs2_calc_trunc_pos(struct inode
*inode
,
1628 struct ocfs2_extent_list
*el
,
1629 struct ocfs2_extent_rec
*rec
,
1630 u32 trunc_start
, u32
*trunc_cpos
,
1631 u32
*trunc_len
, u32
*trunc_end
,
1632 u64
*blkno
, int *done
)
1637 range
= le32_to_cpu(rec
->e_cpos
) + ocfs2_rec_clusters(el
, rec
);
1639 if (le32_to_cpu(rec
->e_cpos
) >= trunc_start
) {
1640 *trunc_cpos
= le32_to_cpu(rec
->e_cpos
);
1642 * Skip holes if any.
1644 if (range
< *trunc_end
)
1646 *trunc_len
= *trunc_end
- le32_to_cpu(rec
->e_cpos
);
1647 *blkno
= le64_to_cpu(rec
->e_blkno
);
1648 *trunc_end
= le32_to_cpu(rec
->e_cpos
);
1649 } else if (range
> trunc_start
) {
1650 *trunc_cpos
= trunc_start
;
1651 *trunc_len
= *trunc_end
- trunc_start
;
1652 coff
= trunc_start
- le32_to_cpu(rec
->e_cpos
);
1653 *blkno
= le64_to_cpu(rec
->e_blkno
) +
1654 ocfs2_clusters_to_blocks(inode
->i_sb
, coff
);
1655 *trunc_end
= trunc_start
;
1658 * It may have two following possibilities:
1660 * - last record has been removed
1661 * - trunc_start was within a hole
1663 * both two cases mean the completion of hole punching.
1671 static int ocfs2_remove_inode_range(struct inode
*inode
,
1672 struct buffer_head
*di_bh
, u64 byte_start
,
1675 int ret
= 0, flags
= 0, done
= 0, i
;
1676 u32 trunc_start
, trunc_len
, trunc_end
, trunc_cpos
, phys_cpos
;
1678 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1679 struct ocfs2_cached_dealloc_ctxt dealloc
;
1680 struct address_space
*mapping
= inode
->i_mapping
;
1681 struct ocfs2_extent_tree et
;
1682 struct ocfs2_path
*path
= NULL
;
1683 struct ocfs2_extent_list
*el
= NULL
;
1684 struct ocfs2_extent_rec
*rec
= NULL
;
1685 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1686 u64 blkno
, refcount_loc
= le64_to_cpu(di
->i_refcount_loc
);
1688 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), di_bh
);
1689 ocfs2_init_dealloc_ctxt(&dealloc
);
1694 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1695 ret
= ocfs2_truncate_inline(inode
, di_bh
, byte_start
,
1696 byte_start
+ byte_len
, 0);
1702 * There's no need to get fancy with the page cache
1703 * truncate of an inline-data inode. We're talking
1704 * about less than a page here, which will be cached
1705 * in the dinode buffer anyway.
1707 unmap_mapping_range(mapping
, 0, 0, 0);
1708 truncate_inode_pages(mapping
, 0);
1713 * For reflinks, we may need to CoW 2 clusters which might be
1714 * partially zero'd later, if hole's start and end offset were
1715 * within one cluster(means is not exactly aligned to clustersize).
1718 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
) {
1720 ret
= ocfs2_cow_file_pos(inode
, di_bh
, byte_start
);
1726 ret
= ocfs2_cow_file_pos(inode
, di_bh
, byte_start
+ byte_len
);
1733 trunc_start
= ocfs2_clusters_for_bytes(osb
->sb
, byte_start
);
1734 trunc_end
= (byte_start
+ byte_len
) >> osb
->s_clustersize_bits
;
1735 cluster_in_el
= trunc_end
;
1737 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1738 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1739 (unsigned long long)byte_start
,
1740 (unsigned long long)byte_len
, trunc_start
, trunc_end
);
1742 ret
= ocfs2_zero_partial_clusters(inode
, byte_start
, byte_len
);
1748 path
= ocfs2_new_path_from_et(&et
);
1755 while (trunc_end
> trunc_start
) {
1757 ret
= ocfs2_find_path(INODE_CACHE(inode
), path
,
1764 el
= path_leaf_el(path
);
1766 i
= ocfs2_find_rec(el
, trunc_end
);
1768 * Need to go to previous extent block.
1771 if (path
->p_tree_depth
== 0)
1774 ret
= ocfs2_find_cpos_for_left_leaf(inode
->i_sb
,
1783 * We've reached the leftmost extent block,
1784 * it's safe to leave.
1786 if (cluster_in_el
== 0)
1790 * The 'pos' searched for previous extent block is
1791 * always one cluster less than actual trunc_end.
1793 trunc_end
= cluster_in_el
+ 1;
1795 ocfs2_reinit_path(path
, 1);
1800 rec
= &el
->l_recs
[i
];
1802 ocfs2_calc_trunc_pos(inode
, el
, rec
, trunc_start
, &trunc_cpos
,
1803 &trunc_len
, &trunc_end
, &blkno
, &done
);
1807 flags
= rec
->e_flags
;
1808 phys_cpos
= ocfs2_blocks_to_clusters(inode
->i_sb
, blkno
);
1810 ret
= ocfs2_remove_btree_range(inode
, &et
, trunc_cpos
,
1811 phys_cpos
, trunc_len
, flags
,
1812 &dealloc
, refcount_loc
);
1818 cluster_in_el
= trunc_end
;
1820 ocfs2_reinit_path(path
, 1);
1823 ocfs2_truncate_cluster_pages(inode
, byte_start
, byte_len
);
1826 ocfs2_schedule_truncate_log_flush(osb
, 1);
1827 ocfs2_run_deallocs(osb
, &dealloc
);
1833 * Parts of this function taken from xfs_change_file_space()
1835 static int __ocfs2_change_file_space(struct file
*file
, struct inode
*inode
,
1836 loff_t f_pos
, unsigned int cmd
,
1837 struct ocfs2_space_resv
*sr
,
1843 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1844 struct buffer_head
*di_bh
= NULL
;
1846 unsigned long long max_off
= inode
->i_sb
->s_maxbytes
;
1848 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
1851 mutex_lock(&inode
->i_mutex
);
1854 * This prevents concurrent writes on other nodes
1856 ret
= ocfs2_rw_lock(inode
, 1);
1862 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1868 if (inode
->i_flags
& (S_IMMUTABLE
|S_APPEND
)) {
1870 goto out_inode_unlock
;
1873 switch (sr
->l_whence
) {
1874 case 0: /*SEEK_SET*/
1876 case 1: /*SEEK_CUR*/
1877 sr
->l_start
+= f_pos
;
1879 case 2: /*SEEK_END*/
1880 sr
->l_start
+= i_size_read(inode
);
1884 goto out_inode_unlock
;
1888 llen
= sr
->l_len
> 0 ? sr
->l_len
- 1 : sr
->l_len
;
1891 || sr
->l_start
> max_off
1892 || (sr
->l_start
+ llen
) < 0
1893 || (sr
->l_start
+ llen
) > max_off
) {
1895 goto out_inode_unlock
;
1897 size
= sr
->l_start
+ sr
->l_len
;
1899 if (cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) {
1900 if (sr
->l_len
<= 0) {
1902 goto out_inode_unlock
;
1906 if (file
&& should_remove_suid(file
->f_path
.dentry
)) {
1907 ret
= __ocfs2_write_remove_suid(inode
, di_bh
);
1910 goto out_inode_unlock
;
1914 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1916 case OCFS2_IOC_RESVSP
:
1917 case OCFS2_IOC_RESVSP64
:
1919 * This takes unsigned offsets, but the signed ones we
1920 * pass have been checked against overflow above.
1922 ret
= ocfs2_allocate_unwritten_extents(inode
, sr
->l_start
,
1925 case OCFS2_IOC_UNRESVSP
:
1926 case OCFS2_IOC_UNRESVSP64
:
1927 ret
= ocfs2_remove_inode_range(inode
, di_bh
, sr
->l_start
,
1933 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1936 goto out_inode_unlock
;
1940 * We update c/mtime for these changes
1942 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1943 if (IS_ERR(handle
)) {
1944 ret
= PTR_ERR(handle
);
1946 goto out_inode_unlock
;
1949 if (change_size
&& i_size_read(inode
) < size
)
1950 i_size_write(inode
, size
);
1952 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1953 ret
= ocfs2_mark_inode_dirty(handle
, inode
, di_bh
);
1957 ocfs2_commit_trans(osb
, handle
);
1961 ocfs2_inode_unlock(inode
, 1);
1963 ocfs2_rw_unlock(inode
, 1);
1966 mutex_unlock(&inode
->i_mutex
);
1970 int ocfs2_change_file_space(struct file
*file
, unsigned int cmd
,
1971 struct ocfs2_space_resv
*sr
)
1973 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1974 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1976 if ((cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) &&
1977 !ocfs2_writes_unwritten_extents(osb
))
1979 else if ((cmd
== OCFS2_IOC_UNRESVSP
|| cmd
== OCFS2_IOC_UNRESVSP64
) &&
1980 !ocfs2_sparse_alloc(osb
))
1983 if (!S_ISREG(inode
->i_mode
))
1986 if (!(file
->f_mode
& FMODE_WRITE
))
1989 return __ocfs2_change_file_space(file
, inode
, file
->f_pos
, cmd
, sr
, 0);
1992 static long ocfs2_fallocate(struct inode
*inode
, int mode
, loff_t offset
,
1995 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1996 struct ocfs2_space_resv sr
;
1997 int change_size
= 1;
1999 if (!ocfs2_writes_unwritten_extents(osb
))
2002 if (S_ISDIR(inode
->i_mode
))
2005 if (mode
& FALLOC_FL_KEEP_SIZE
)
2009 sr
.l_start
= (s64
)offset
;
2010 sr
.l_len
= (s64
)len
;
2012 return __ocfs2_change_file_space(NULL
, inode
, offset
,
2013 OCFS2_IOC_RESVSP64
, &sr
, change_size
);
2016 int ocfs2_check_range_for_refcount(struct inode
*inode
, loff_t pos
,
2020 unsigned int extent_flags
;
2021 u32 cpos
, clusters
, extent_len
, phys_cpos
;
2022 struct super_block
*sb
= inode
->i_sb
;
2024 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
)) ||
2025 !(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
) ||
2026 OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
2029 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
2030 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
2033 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
2040 if (phys_cpos
&& (extent_flags
& OCFS2_EXT_REFCOUNTED
)) {
2045 if (extent_len
> clusters
)
2046 extent_len
= clusters
;
2048 clusters
-= extent_len
;
2055 static int ocfs2_prepare_inode_for_refcount(struct inode
*inode
,
2057 loff_t pos
, size_t count
,
2061 struct buffer_head
*di_bh
= NULL
;
2062 u32 cpos
= pos
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
2064 ocfs2_clusters_for_bytes(inode
->i_sb
, pos
+ count
) - cpos
;
2066 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
2074 ret
= ocfs2_refcount_cow(inode
, di_bh
, cpos
, clusters
, UINT_MAX
);
2082 static int ocfs2_prepare_inode_for_write(struct file
*file
,
2089 int ret
= 0, meta_level
= 0;
2090 struct dentry
*dentry
= file
->f_path
.dentry
;
2091 struct inode
*inode
= dentry
->d_inode
;
2092 loff_t saved_pos
, end
;
2095 * We start with a read level meta lock and only jump to an ex
2096 * if we need to make modifications here.
2099 ret
= ocfs2_inode_lock(inode
, NULL
, meta_level
);
2106 /* Clear suid / sgid if necessary. We do this here
2107 * instead of later in the write path because
2108 * remove_suid() calls ->setattr without any hint that
2109 * we may have already done our cluster locking. Since
2110 * ocfs2_setattr() *must* take cluster locks to
2111 * proceeed, this will lead us to recursively lock the
2112 * inode. There's also the dinode i_size state which
2113 * can be lost via setattr during extending writes (we
2114 * set inode->i_size at the end of a write. */
2115 if (should_remove_suid(dentry
)) {
2116 if (meta_level
== 0) {
2117 ocfs2_inode_unlock(inode
, meta_level
);
2122 ret
= ocfs2_write_remove_suid(inode
);
2129 /* work on a copy of ppos until we're sure that we won't have
2130 * to recalculate it due to relocking. */
2132 saved_pos
= i_size_read(inode
);
2133 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
2138 end
= saved_pos
+ count
;
2140 ret
= ocfs2_check_range_for_refcount(inode
, saved_pos
, count
);
2142 ocfs2_inode_unlock(inode
, meta_level
);
2145 ret
= ocfs2_prepare_inode_for_refcount(inode
,
2162 * Skip the O_DIRECT checks if we don't need
2165 if (!direct_io
|| !(*direct_io
))
2169 * There's no sane way to do direct writes to an inode
2172 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
2178 * Allowing concurrent direct writes means
2179 * i_size changes wouldn't be synchronized, so
2180 * one node could wind up truncating another
2183 if (end
> i_size_read(inode
)) {
2189 * We don't fill holes during direct io, so
2190 * check for them here. If any are found, the
2191 * caller will have to retake some cluster
2192 * locks and initiate the io as buffered.
2194 ret
= ocfs2_check_range_for_holes(inode
, saved_pos
, count
);
2207 if (meta_level
>= 0)
2208 ocfs2_inode_unlock(inode
, meta_level
);
2214 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
2215 const struct iovec
*iov
,
2216 unsigned long nr_segs
,
2219 int ret
, direct_io
, appending
, rw_level
, have_alloc_sem
= 0;
2220 int can_do_direct
, has_refcount
= 0;
2221 ssize_t written
= 0;
2222 size_t ocount
; /* original count */
2223 size_t count
; /* after file limit checks */
2224 loff_t old_size
, *ppos
= &iocb
->ki_pos
;
2226 struct file
*file
= iocb
->ki_filp
;
2227 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2228 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2230 mlog_entry("(0x%p, %u, '%.*s')\n", file
,
2231 (unsigned int)nr_segs
,
2232 file
->f_path
.dentry
->d_name
.len
,
2233 file
->f_path
.dentry
->d_name
.name
);
2235 if (iocb
->ki_left
== 0)
2238 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2240 appending
= file
->f_flags
& O_APPEND
? 1 : 0;
2241 direct_io
= file
->f_flags
& O_DIRECT
? 1 : 0;
2243 mutex_lock(&inode
->i_mutex
);
2246 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2248 down_read(&inode
->i_alloc_sem
);
2252 /* concurrent O_DIRECT writes are allowed */
2253 rw_level
= !direct_io
;
2254 ret
= ocfs2_rw_lock(inode
, rw_level
);
2260 can_do_direct
= direct_io
;
2261 ret
= ocfs2_prepare_inode_for_write(file
, ppos
,
2262 iocb
->ki_left
, appending
,
2263 &can_do_direct
, &has_refcount
);
2270 * We can't complete the direct I/O as requested, fall back to
2273 if (direct_io
&& !can_do_direct
) {
2274 ocfs2_rw_unlock(inode
, rw_level
);
2275 up_read(&inode
->i_alloc_sem
);
2285 * To later detect whether a journal commit for sync writes is
2286 * necessary, we sample i_size, and cluster count here.
2288 old_size
= i_size_read(inode
);
2289 old_clusters
= OCFS2_I(inode
)->ip_clusters
;
2291 /* communicate with ocfs2_dio_end_io */
2292 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2294 ret
= generic_segment_checks(iov
, &nr_segs
, &ocount
,
2300 ret
= generic_write_checks(file
, ppos
, &count
,
2301 S_ISBLK(inode
->i_mode
));
2306 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, *ppos
,
2307 ppos
, count
, ocount
);
2310 * direct write may have instantiated a few
2311 * blocks outside i_size. Trim these off again.
2312 * Don't need i_size_read because we hold i_mutex.
2314 * XXX(truncate): this looks buggy because ocfs2 did not
2315 * actually implement ->truncate. Take a look at
2316 * the new truncate sequence and update this accordingly
2318 if (*ppos
+ count
> inode
->i_size
)
2319 truncate_setsize(inode
, inode
->i_size
);
2324 current
->backing_dev_info
= file
->f_mapping
->backing_dev_info
;
2325 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
, *ppos
,
2327 current
->backing_dev_info
= NULL
;
2331 /* buffered aio wouldn't have proper lock coverage today */
2332 BUG_ON(ret
== -EIOCBQUEUED
&& !(file
->f_flags
& O_DIRECT
));
2334 if (((file
->f_flags
& O_DSYNC
) && !direct_io
) || IS_SYNC(inode
) ||
2335 ((file
->f_flags
& O_DIRECT
) && has_refcount
)) {
2336 ret
= filemap_fdatawrite_range(file
->f_mapping
, pos
,
2341 if (!ret
&& ((old_size
!= i_size_read(inode
)) ||
2342 (old_clusters
!= OCFS2_I(inode
)->ip_clusters
) ||
2344 ret
= jbd2_journal_force_commit(osb
->journal
->j_journal
);
2350 ret
= filemap_fdatawait_range(file
->f_mapping
, pos
,
2355 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2356 * function pointer which is called when o_direct io completes so that
2357 * it can unlock our rw lock. (it's the clustered equivalent of
2358 * i_alloc_sem; protects truncate from racing with pending ios).
2359 * Unfortunately there are error cases which call end_io and others
2360 * that don't. so we don't have to unlock the rw_lock if either an
2361 * async dio is going to do it in the future or an end_io after an
2362 * error has already done it.
2364 if ((ret
== -EIOCBQUEUED
) || (!ocfs2_iocb_is_rw_locked(iocb
))) {
2371 ocfs2_rw_unlock(inode
, rw_level
);
2375 up_read(&inode
->i_alloc_sem
);
2377 mutex_unlock(&inode
->i_mutex
);
2385 static int ocfs2_splice_to_file(struct pipe_inode_info
*pipe
,
2387 struct splice_desc
*sd
)
2391 ret
= ocfs2_prepare_inode_for_write(out
, &sd
->pos
,
2392 sd
->total_len
, 0, NULL
, NULL
);
2398 return splice_from_pipe_feed(pipe
, sd
, pipe_to_file
);
2401 static ssize_t
ocfs2_file_splice_write(struct pipe_inode_info
*pipe
,
2408 struct address_space
*mapping
= out
->f_mapping
;
2409 struct inode
*inode
= mapping
->host
;
2410 struct splice_desc sd
= {
2417 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out
, pipe
,
2419 out
->f_path
.dentry
->d_name
.len
,
2420 out
->f_path
.dentry
->d_name
.name
);
2423 mutex_lock_nested(&pipe
->inode
->i_mutex
, I_MUTEX_PARENT
);
2425 splice_from_pipe_begin(&sd
);
2427 ret
= splice_from_pipe_next(pipe
, &sd
);
2431 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2432 ret
= ocfs2_rw_lock(inode
, 1);
2436 ret
= ocfs2_splice_to_file(pipe
, out
, &sd
);
2437 ocfs2_rw_unlock(inode
, 1);
2439 mutex_unlock(&inode
->i_mutex
);
2441 splice_from_pipe_end(pipe
, &sd
);
2444 mutex_unlock(&pipe
->inode
->i_mutex
);
2447 ret
= sd
.num_spliced
;
2450 unsigned long nr_pages
;
2453 nr_pages
= (ret
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2455 err
= generic_write_sync(out
, *ppos
, ret
);
2461 balance_dirty_pages_ratelimited_nr(mapping
, nr_pages
);
2468 static ssize_t
ocfs2_file_splice_read(struct file
*in
,
2470 struct pipe_inode_info
*pipe
,
2474 int ret
= 0, lock_level
= 0;
2475 struct inode
*inode
= in
->f_path
.dentry
->d_inode
;
2477 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in
, pipe
,
2479 in
->f_path
.dentry
->d_name
.len
,
2480 in
->f_path
.dentry
->d_name
.name
);
2483 * See the comment in ocfs2_file_aio_read()
2485 ret
= ocfs2_inode_lock_atime(inode
, in
->f_vfsmnt
, &lock_level
);
2490 ocfs2_inode_unlock(inode
, lock_level
);
2492 ret
= generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
2499 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
2500 const struct iovec
*iov
,
2501 unsigned long nr_segs
,
2504 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0, lock_level
= 0;
2505 struct file
*filp
= iocb
->ki_filp
;
2506 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
2508 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
2509 (unsigned int)nr_segs
,
2510 filp
->f_path
.dentry
->d_name
.len
,
2511 filp
->f_path
.dentry
->d_name
.name
);
2520 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2521 * need locks to protect pending reads from racing with truncate.
2523 if (filp
->f_flags
& O_DIRECT
) {
2524 down_read(&inode
->i_alloc_sem
);
2527 ret
= ocfs2_rw_lock(inode
, 0);
2533 /* communicate with ocfs2_dio_end_io */
2534 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2538 * We're fine letting folks race truncates and extending
2539 * writes with read across the cluster, just like they can
2540 * locally. Hence no rw_lock during read.
2542 * Take and drop the meta data lock to update inode fields
2543 * like i_size. This allows the checks down below
2544 * generic_file_aio_read() a chance of actually working.
2546 ret
= ocfs2_inode_lock_atime(inode
, filp
->f_vfsmnt
, &lock_level
);
2551 ocfs2_inode_unlock(inode
, lock_level
);
2553 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
2555 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2557 /* buffered aio wouldn't have proper lock coverage today */
2558 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
2560 /* see ocfs2_file_aio_write */
2561 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
2568 up_read(&inode
->i_alloc_sem
);
2570 ocfs2_rw_unlock(inode
, rw_level
);
2576 const struct inode_operations ocfs2_file_iops
= {
2577 .setattr
= ocfs2_setattr
,
2578 .getattr
= ocfs2_getattr
,
2579 .permission
= ocfs2_permission
,
2580 .setxattr
= generic_setxattr
,
2581 .getxattr
= generic_getxattr
,
2582 .listxattr
= ocfs2_listxattr
,
2583 .removexattr
= generic_removexattr
,
2584 .fallocate
= ocfs2_fallocate
,
2585 .fiemap
= ocfs2_fiemap
,
2588 const struct inode_operations ocfs2_special_file_iops
= {
2589 .setattr
= ocfs2_setattr
,
2590 .getattr
= ocfs2_getattr
,
2591 .permission
= ocfs2_permission
,
2595 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2596 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2598 const struct file_operations ocfs2_fops
= {
2599 .llseek
= generic_file_llseek
,
2600 .read
= do_sync_read
,
2601 .write
= do_sync_write
,
2603 .fsync
= ocfs2_sync_file
,
2604 .release
= ocfs2_file_release
,
2605 .open
= ocfs2_file_open
,
2606 .aio_read
= ocfs2_file_aio_read
,
2607 .aio_write
= ocfs2_file_aio_write
,
2608 .unlocked_ioctl
= ocfs2_ioctl
,
2609 #ifdef CONFIG_COMPAT
2610 .compat_ioctl
= ocfs2_compat_ioctl
,
2613 .flock
= ocfs2_flock
,
2614 .splice_read
= ocfs2_file_splice_read
,
2615 .splice_write
= ocfs2_file_splice_write
,
2618 const struct file_operations ocfs2_dops
= {
2619 .llseek
= generic_file_llseek
,
2620 .read
= generic_read_dir
,
2621 .readdir
= ocfs2_readdir
,
2622 .fsync
= ocfs2_sync_file
,
2623 .release
= ocfs2_dir_release
,
2624 .open
= ocfs2_dir_open
,
2625 .unlocked_ioctl
= ocfs2_ioctl
,
2626 #ifdef CONFIG_COMPAT
2627 .compat_ioctl
= ocfs2_compat_ioctl
,
2630 .flock
= ocfs2_flock
,
2634 * POSIX-lockless variants of our file_operations.
2636 * These will be used if the underlying cluster stack does not support
2637 * posix file locking, if the user passes the "localflocks" mount
2638 * option, or if we have a local-only fs.
2640 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2641 * so we still want it in the case of no stack support for
2642 * plocks. Internally, it will do the right thing when asked to ignore
2645 const struct file_operations ocfs2_fops_no_plocks
= {
2646 .llseek
= generic_file_llseek
,
2647 .read
= do_sync_read
,
2648 .write
= do_sync_write
,
2650 .fsync
= ocfs2_sync_file
,
2651 .release
= ocfs2_file_release
,
2652 .open
= ocfs2_file_open
,
2653 .aio_read
= ocfs2_file_aio_read
,
2654 .aio_write
= ocfs2_file_aio_write
,
2655 .unlocked_ioctl
= ocfs2_ioctl
,
2656 #ifdef CONFIG_COMPAT
2657 .compat_ioctl
= ocfs2_compat_ioctl
,
2659 .flock
= ocfs2_flock
,
2660 .splice_read
= ocfs2_file_splice_read
,
2661 .splice_write
= ocfs2_file_splice_write
,
2664 const struct file_operations ocfs2_dops_no_plocks
= {
2665 .llseek
= generic_file_llseek
,
2666 .read
= generic_read_dir
,
2667 .readdir
= ocfs2_readdir
,
2668 .fsync
= ocfs2_sync_file
,
2669 .release
= ocfs2_dir_release
,
2670 .open
= ocfs2_dir_open
,
2671 .unlocked_ioctl
= ocfs2_ioctl
,
2672 #ifdef CONFIG_COMPAT
2673 .compat_ioctl
= ocfs2_compat_ioctl
,
2675 .flock
= ocfs2_flock
,