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
,
179 struct dentry
*dentry
,
184 struct inode
*inode
= dentry
->d_inode
;
185 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
187 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
188 dentry
->d_name
.len
, dentry
->d_name
.name
);
190 err
= ocfs2_sync_inode(dentry
->d_inode
);
194 if (datasync
&& !(inode
->i_state
& I_DIRTY_DATASYNC
))
197 journal
= osb
->journal
->j_journal
;
198 err
= jbd2_journal_force_commit(journal
);
203 return (err
< 0) ? -EIO
: 0;
206 int ocfs2_should_update_atime(struct inode
*inode
,
207 struct vfsmount
*vfsmnt
)
210 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
212 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
215 if ((inode
->i_flags
& S_NOATIME
) ||
216 ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
220 * We can be called with no vfsmnt structure - NFSD will
223 * Note that our action here is different than touch_atime() -
224 * if we can't tell whether this is a noatime mount, then we
225 * don't know whether to trust the value of s_atime_quantum.
230 if ((vfsmnt
->mnt_flags
& MNT_NOATIME
) ||
231 ((vfsmnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
234 if (vfsmnt
->mnt_flags
& MNT_RELATIME
) {
235 if ((timespec_compare(&inode
->i_atime
, &inode
->i_mtime
) <= 0) ||
236 (timespec_compare(&inode
->i_atime
, &inode
->i_ctime
) <= 0))
243 if ((now
.tv_sec
- inode
->i_atime
.tv_sec
<= osb
->s_atime_quantum
))
249 int ocfs2_update_inode_atime(struct inode
*inode
,
250 struct buffer_head
*bh
)
253 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
255 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) bh
->b_data
;
259 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
260 if (IS_ERR(handle
)) {
261 ret
= PTR_ERR(handle
);
266 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
267 OCFS2_JOURNAL_ACCESS_WRITE
);
274 * Don't use ocfs2_mark_inode_dirty() here as we don't always
275 * have i_mutex to guard against concurrent changes to other
278 inode
->i_atime
= CURRENT_TIME
;
279 di
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
280 di
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
282 ret
= ocfs2_journal_dirty(handle
, bh
);
287 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
293 static int ocfs2_set_inode_size(handle_t
*handle
,
295 struct buffer_head
*fe_bh
,
301 i_size_write(inode
, new_i_size
);
302 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
303 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
305 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
316 int ocfs2_simple_size_update(struct inode
*inode
,
317 struct buffer_head
*di_bh
,
321 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
322 handle_t
*handle
= NULL
;
324 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
325 if (IS_ERR(handle
)) {
326 ret
= PTR_ERR(handle
);
331 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
336 ocfs2_commit_trans(osb
, handle
);
341 static int ocfs2_cow_file_pos(struct inode
*inode
,
342 struct buffer_head
*fe_bh
,
346 u32 phys
, cpos
= offset
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
347 unsigned int num_clusters
= 0;
348 unsigned int ext_flags
= 0;
351 * If the new offset is aligned to the range of the cluster, there is
352 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
355 if ((offset
& (OCFS2_SB(inode
->i_sb
)->s_clustersize
- 1)) == 0)
358 status
= ocfs2_get_clusters(inode
, cpos
, &phys
,
359 &num_clusters
, &ext_flags
);
365 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
368 return ocfs2_refcount_cow(inode
, fe_bh
, cpos
, 1, cpos
+1);
374 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
376 struct buffer_head
*fe_bh
,
381 struct ocfs2_dinode
*di
;
387 * We need to CoW the cluster contains the offset if it is reflinked
388 * since we will call ocfs2_zero_range_for_truncate later which will
389 * write "0" from offset to the end of the cluster.
391 status
= ocfs2_cow_file_pos(inode
, fe_bh
, new_i_size
);
397 /* TODO: This needs to actually orphan the inode in this
400 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
401 if (IS_ERR(handle
)) {
402 status
= PTR_ERR(handle
);
407 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
408 OCFS2_JOURNAL_ACCESS_WRITE
);
415 * Do this before setting i_size.
417 cluster_bytes
= ocfs2_align_bytes_to_clusters(inode
->i_sb
, new_i_size
);
418 status
= ocfs2_zero_range_for_truncate(inode
, handle
, new_i_size
,
425 i_size_write(inode
, new_i_size
);
426 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
428 di
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
429 di
->i_size
= cpu_to_le64(new_i_size
);
430 di
->i_ctime
= di
->i_mtime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
431 di
->i_ctime_nsec
= di
->i_mtime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
433 status
= ocfs2_journal_dirty(handle
, fe_bh
);
438 ocfs2_commit_trans(osb
, handle
);
445 static int ocfs2_truncate_file(struct inode
*inode
,
446 struct buffer_head
*di_bh
,
450 struct ocfs2_dinode
*fe
= NULL
;
451 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
452 struct ocfs2_truncate_context
*tc
= NULL
;
454 mlog_entry("(inode = %llu, new_i_size = %llu\n",
455 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
456 (unsigned long long)new_i_size
);
458 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
459 * already validated it */
460 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
462 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
463 "Inode %llu, inode i_size = %lld != di "
464 "i_size = %llu, i_flags = 0x%x\n",
465 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
467 (unsigned long long)le64_to_cpu(fe
->i_size
),
468 le32_to_cpu(fe
->i_flags
));
470 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
471 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
472 (unsigned long long)le64_to_cpu(fe
->i_size
),
473 (unsigned long long)new_i_size
);
479 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
480 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
481 (unsigned long long)le64_to_cpu(fe
->i_size
),
482 (unsigned long long)new_i_size
);
484 /* lets handle the simple truncate cases before doing any more
485 * cluster locking. */
486 if (new_i_size
== le64_to_cpu(fe
->i_size
))
489 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
492 * The inode lock forced other nodes to sync and drop their
493 * pages, which (correctly) happens even if we have a truncate
494 * without allocation change - ocfs2 cluster sizes can be much
495 * greater than page size, so we have to truncate them
498 unmap_mapping_range(inode
->i_mapping
, new_i_size
+ PAGE_SIZE
- 1, 0, 1);
499 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
501 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
502 status
= ocfs2_truncate_inline(inode
, di_bh
, new_i_size
,
503 i_size_read(inode
), 1);
507 goto bail_unlock_sem
;
510 /* alright, we're going to need to do a full blown alloc size
511 * change. Orphan the inode so that recovery can complete the
512 * truncate if necessary. This does the task of marking
514 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
517 goto bail_unlock_sem
;
520 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
523 goto bail_unlock_sem
;
526 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
529 goto bail_unlock_sem
;
532 /* TODO: orphan dir cleanup here. */
534 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
537 if (!status
&& OCFS2_I(inode
)->ip_clusters
== 0)
538 status
= ocfs2_try_remove_refcount_tree(inode
, di_bh
);
545 * extend file allocation only here.
546 * we'll update all the disk stuff, and oip->alloc_size
548 * expect stuff to be locked, a transaction started and enough data /
549 * metadata reservations in the contexts.
551 * Will return -EAGAIN, and a reason if a restart is needed.
552 * If passed in, *reason will always be set, even in error.
554 int ocfs2_add_inode_data(struct ocfs2_super
*osb
,
559 struct buffer_head
*fe_bh
,
561 struct ocfs2_alloc_context
*data_ac
,
562 struct ocfs2_alloc_context
*meta_ac
,
563 enum ocfs2_alloc_restarted
*reason_ret
)
566 struct ocfs2_extent_tree et
;
568 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), fe_bh
);
569 ret
= ocfs2_add_clusters_in_btree(handle
, &et
, logical_offset
,
570 clusters_to_add
, mark_unwritten
,
571 data_ac
, meta_ac
, reason_ret
);
576 static int __ocfs2_extend_allocation(struct inode
*inode
, u32 logical_start
,
577 u32 clusters_to_add
, int mark_unwritten
)
580 int restart_func
= 0;
583 struct buffer_head
*bh
= NULL
;
584 struct ocfs2_dinode
*fe
= NULL
;
585 handle_t
*handle
= NULL
;
586 struct ocfs2_alloc_context
*data_ac
= NULL
;
587 struct ocfs2_alloc_context
*meta_ac
= NULL
;
588 enum ocfs2_alloc_restarted why
;
589 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
590 struct ocfs2_extent_tree et
;
593 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
596 * This function only exists for file systems which don't
599 BUG_ON(mark_unwritten
&& !ocfs2_sparse_alloc(osb
));
601 status
= ocfs2_read_inode_block(inode
, &bh
);
606 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
609 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
611 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
612 "clusters_to_add = %u\n",
613 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
614 (long long)i_size_read(inode
), le32_to_cpu(fe
->i_clusters
),
616 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), bh
);
617 status
= ocfs2_lock_allocators(inode
, &et
, clusters_to_add
, 0,
624 credits
= ocfs2_calc_extend_credits(osb
->sb
, &fe
->id2
.i_list
,
626 handle
= ocfs2_start_trans(osb
, credits
);
627 if (IS_ERR(handle
)) {
628 status
= PTR_ERR(handle
);
634 restarted_transaction
:
635 status
= dquot_alloc_space_nodirty(inode
,
636 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
641 /* reserve a write to the file entry early on - that we if we
642 * run out of credits in the allocation path, we can still
644 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
645 OCFS2_JOURNAL_ACCESS_WRITE
);
651 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
653 status
= ocfs2_add_inode_data(osb
,
663 if ((status
< 0) && (status
!= -EAGAIN
)) {
664 if (status
!= -ENOSPC
)
669 status
= ocfs2_journal_dirty(handle
, bh
);
675 spin_lock(&OCFS2_I(inode
)->ip_lock
);
676 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
677 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
678 /* Release unused quota reservation */
679 dquot_free_space(inode
,
680 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
683 if (why
!= RESTART_NONE
&& clusters_to_add
) {
684 if (why
== RESTART_META
) {
685 mlog(0, "restarting function.\n");
689 BUG_ON(why
!= RESTART_TRANS
);
691 mlog(0, "restarting transaction.\n");
692 /* TODO: This can be more intelligent. */
693 credits
= ocfs2_calc_extend_credits(osb
->sb
,
696 status
= ocfs2_extend_trans(handle
, credits
);
698 /* handle still has to be committed at
704 goto restarted_transaction
;
708 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
709 le32_to_cpu(fe
->i_clusters
),
710 (unsigned long long)le64_to_cpu(fe
->i_size
));
711 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
712 OCFS2_I(inode
)->ip_clusters
, (long long)i_size_read(inode
));
715 if (status
< 0 && did_quota
)
716 dquot_free_space(inode
,
717 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
719 ocfs2_commit_trans(osb
, handle
);
723 ocfs2_free_alloc_context(data_ac
);
727 ocfs2_free_alloc_context(meta_ac
);
730 if ((!status
) && restart_func
) {
742 * While a write will already be ordering the data, a truncate will not.
743 * Thus, we need to explicitly order the zeroed pages.
745 static handle_t
*ocfs2_zero_start_ordered_transaction(struct inode
*inode
)
747 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
748 handle_t
*handle
= NULL
;
751 if (!ocfs2_should_order_data(inode
))
754 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
755 if (IS_ERR(handle
)) {
761 ret
= ocfs2_jbd2_file_inode(handle
, inode
);
768 ocfs2_commit_trans(osb
, handle
);
769 handle
= ERR_PTR(ret
);
774 /* Some parts of this taken from generic_cont_expand, which turned out
775 * to be too fragile to do exactly what we need without us having to
776 * worry about recursive locking in ->write_begin() and ->write_end(). */
777 static int ocfs2_write_zero_page(struct inode
*inode
, u64 abs_from
,
780 struct address_space
*mapping
= inode
->i_mapping
;
782 unsigned long index
= abs_from
>> PAGE_CACHE_SHIFT
;
783 handle_t
*handle
= NULL
;
785 unsigned zero_from
, zero_to
, block_start
, block_end
;
787 BUG_ON(abs_from
>= abs_to
);
788 BUG_ON(abs_to
> (((u64
)index
+ 1) << PAGE_CACHE_SHIFT
));
789 BUG_ON(abs_from
& (inode
->i_blkbits
- 1));
791 page
= grab_cache_page(mapping
, index
);
798 /* Get the offsets within the page that we want to zero */
799 zero_from
= abs_from
& (PAGE_CACHE_SIZE
- 1);
800 zero_to
= abs_to
& (PAGE_CACHE_SIZE
- 1);
802 zero_to
= PAGE_CACHE_SIZE
;
804 /* We know that zero_from is block aligned */
805 for (block_start
= zero_from
; block_start
< zero_to
;
806 block_start
= block_end
) {
807 block_end
= block_start
+ (1 << inode
->i_blkbits
);
810 * block_start is block-aligned. Bump it by one to
811 * force ocfs2_{prepare,commit}_write() to zero the
814 ret
= ocfs2_prepare_write_nolock(inode
, page
,
823 handle
= ocfs2_zero_start_ordered_transaction(inode
);
824 if (IS_ERR(handle
)) {
825 ret
= PTR_ERR(handle
);
831 /* must not update i_size! */
832 ret
= block_commit_write(page
, block_start
+ 1,
841 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
845 page_cache_release(page
);
850 static int ocfs2_zero_extend(struct inode
*inode
,
854 u64 start_off
, next_off
;
855 struct super_block
*sb
= inode
->i_sb
;
857 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
858 while (start_off
< zero_to_size
) {
859 next_off
= (start_off
& PAGE_CACHE_MASK
) + PAGE_CACHE_SIZE
;
860 if (next_off
> zero_to_size
)
861 next_off
= zero_to_size
;
862 ret
= ocfs2_write_zero_page(inode
, start_off
, next_off
);
868 start_off
= next_off
;
871 * Very large extends have the potential to lock up
872 * the cpu for extended periods of time.
881 int ocfs2_extend_no_holes(struct inode
*inode
, u64 new_i_size
, u64 zero_to
)
885 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
887 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
);
888 if (clusters_to_add
< oi
->ip_clusters
)
891 clusters_to_add
-= oi
->ip_clusters
;
893 if (clusters_to_add
) {
894 ret
= __ocfs2_extend_allocation(inode
, oi
->ip_clusters
,
903 * Call this even if we don't add any clusters to the tree. We
904 * still need to zero the area between the old i_size and the
907 ret
= ocfs2_zero_extend(inode
, zero_to
);
915 static int ocfs2_extend_file(struct inode
*inode
,
916 struct buffer_head
*di_bh
,
920 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
924 /* setattr sometimes calls us like this. */
928 if (i_size_read(inode
) == new_i_size
)
930 BUG_ON(new_i_size
< i_size_read(inode
));
933 * Fall through for converting inline data, even if the fs
934 * supports sparse files.
936 * The check for inline data here is legal - nobody can add
937 * the feature since we have i_mutex. We must check it again
938 * after acquiring ip_alloc_sem though, as paths like mmap
939 * might have raced us to converting the inode to extents.
941 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
942 && ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
943 goto out_update_size
;
946 * The alloc sem blocks people in read/write from reading our
947 * allocation until we're done changing it. We depend on
948 * i_mutex to block other extend/truncate calls while we're
951 down_write(&oi
->ip_alloc_sem
);
953 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
955 * We can optimize small extends by keeping the inodes
958 if (ocfs2_size_fits_inline_data(di_bh
, new_i_size
)) {
959 up_write(&oi
->ip_alloc_sem
);
960 goto out_update_size
;
963 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
965 up_write(&oi
->ip_alloc_sem
);
972 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
973 ret
= ocfs2_extend_no_holes(inode
, new_i_size
, new_i_size
);
975 up_write(&oi
->ip_alloc_sem
);
983 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
991 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
993 int status
= 0, size_change
;
994 struct inode
*inode
= dentry
->d_inode
;
995 struct super_block
*sb
= inode
->i_sb
;
996 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
997 struct buffer_head
*bh
= NULL
;
998 handle_t
*handle
= NULL
;
1000 struct dquot
*transfer_from
[MAXQUOTAS
] = { };
1001 struct dquot
*transfer_to
[MAXQUOTAS
] = { };
1003 mlog_entry("(0x%p, '%.*s')\n", dentry
,
1004 dentry
->d_name
.len
, dentry
->d_name
.name
);
1006 /* ensuring we don't even attempt to truncate a symlink */
1007 if (S_ISLNK(inode
->i_mode
))
1008 attr
->ia_valid
&= ~ATTR_SIZE
;
1010 if (attr
->ia_valid
& ATTR_MODE
)
1011 mlog(0, "mode change: %d\n", attr
->ia_mode
);
1012 if (attr
->ia_valid
& ATTR_UID
)
1013 mlog(0, "uid change: %d\n", attr
->ia_uid
);
1014 if (attr
->ia_valid
& ATTR_GID
)
1015 mlog(0, "gid change: %d\n", attr
->ia_gid
);
1016 if (attr
->ia_valid
& ATTR_SIZE
)
1017 mlog(0, "size change...\n");
1018 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
1019 mlog(0, "time change...\n");
1021 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1022 | ATTR_GID | ATTR_UID | ATTR_MODE)
1023 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
1024 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
1028 status
= inode_change_ok(inode
, attr
);
1032 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
1034 dquot_initialize(inode
);
1036 status
= ocfs2_rw_lock(inode
, 1);
1043 status
= ocfs2_inode_lock(inode
, &bh
, 1);
1045 if (status
!= -ENOENT
)
1047 goto bail_unlock_rw
;
1050 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
1051 status
= inode_newsize_ok(inode
, attr
->ia_size
);
1055 if (i_size_read(inode
) > attr
->ia_size
) {
1056 if (ocfs2_should_order_data(inode
)) {
1057 status
= ocfs2_begin_ordered_truncate(inode
,
1062 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
1064 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
1066 if (status
!= -ENOSPC
)
1073 if ((attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
1074 (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
1076 * Gather pointers to quota structures so that allocation /
1077 * freeing of quota structures happens here and not inside
1078 * dquot_transfer() where we have problems with lock ordering
1080 if (attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
1081 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
1082 OCFS2_FEATURE_RO_COMPAT_USRQUOTA
)) {
1083 transfer_to
[USRQUOTA
] = dqget(sb
, attr
->ia_uid
,
1085 transfer_from
[USRQUOTA
] = dqget(sb
, inode
->i_uid
,
1087 if (!transfer_to
[USRQUOTA
] || !transfer_from
[USRQUOTA
]) {
1092 if (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
1093 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
1094 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
)) {
1095 transfer_to
[GRPQUOTA
] = dqget(sb
, attr
->ia_gid
,
1097 transfer_from
[GRPQUOTA
] = dqget(sb
, inode
->i_gid
,
1099 if (!transfer_to
[GRPQUOTA
] || !transfer_from
[GRPQUOTA
]) {
1104 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
+
1105 2 * ocfs2_quota_trans_credits(sb
));
1106 if (IS_ERR(handle
)) {
1107 status
= PTR_ERR(handle
);
1111 status
= dquot_transfer(inode
, attr
);
1115 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1116 if (IS_ERR(handle
)) {
1117 status
= PTR_ERR(handle
);
1124 * This will intentionally not wind up calling vmtruncate(),
1125 * since all the work for a size change has been done above.
1126 * Otherwise, we could get into problems with truncate as
1127 * ip_alloc_sem is used there to protect against i_size
1130 status
= inode_setattr(inode
, attr
);
1136 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
1141 ocfs2_commit_trans(osb
, handle
);
1143 ocfs2_inode_unlock(inode
, 1);
1146 ocfs2_rw_unlock(inode
, 1);
1150 /* Release quota pointers in case we acquired them */
1151 for (qtype
= 0; qtype
< MAXQUOTAS
; qtype
++) {
1152 dqput(transfer_to
[qtype
]);
1153 dqput(transfer_from
[qtype
]);
1156 if (!status
&& attr
->ia_valid
& ATTR_MODE
) {
1157 status
= ocfs2_acl_chmod(inode
);
1166 int ocfs2_getattr(struct vfsmount
*mnt
,
1167 struct dentry
*dentry
,
1170 struct inode
*inode
= dentry
->d_inode
;
1171 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1172 struct ocfs2_super
*osb
= sb
->s_fs_info
;
1177 err
= ocfs2_inode_revalidate(dentry
);
1184 generic_fillattr(inode
, stat
);
1186 /* We set the blksize from the cluster size for performance */
1187 stat
->blksize
= osb
->s_clustersize
;
1195 int ocfs2_permission(struct inode
*inode
, int mask
)
1201 ret
= ocfs2_inode_lock(inode
, NULL
, 0);
1208 ret
= generic_permission(inode
, mask
, ocfs2_check_acl
);
1210 ocfs2_inode_unlock(inode
, 0);
1216 static int __ocfs2_write_remove_suid(struct inode
*inode
,
1217 struct buffer_head
*bh
)
1221 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1222 struct ocfs2_dinode
*di
;
1224 mlog_entry("(Inode %llu, mode 0%o)\n",
1225 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_mode
);
1227 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1228 if (IS_ERR(handle
)) {
1229 ret
= PTR_ERR(handle
);
1234 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
1235 OCFS2_JOURNAL_ACCESS_WRITE
);
1241 inode
->i_mode
&= ~S_ISUID
;
1242 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
1243 inode
->i_mode
&= ~S_ISGID
;
1245 di
= (struct ocfs2_dinode
*) bh
->b_data
;
1246 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
1248 ret
= ocfs2_journal_dirty(handle
, bh
);
1253 ocfs2_commit_trans(osb
, handle
);
1260 * Will look for holes and unwritten extents in the range starting at
1261 * pos for count bytes (inclusive).
1263 static int ocfs2_check_range_for_holes(struct inode
*inode
, loff_t pos
,
1267 unsigned int extent_flags
;
1268 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1269 struct super_block
*sb
= inode
->i_sb
;
1271 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1272 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1275 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1282 if (phys_cpos
== 0 || (extent_flags
& OCFS2_EXT_UNWRITTEN
)) {
1287 if (extent_len
> clusters
)
1288 extent_len
= clusters
;
1290 clusters
-= extent_len
;
1297 static int ocfs2_write_remove_suid(struct inode
*inode
)
1300 struct buffer_head
*bh
= NULL
;
1302 ret
= ocfs2_read_inode_block(inode
, &bh
);
1308 ret
= __ocfs2_write_remove_suid(inode
, bh
);
1315 * Allocate enough extents to cover the region starting at byte offset
1316 * start for len bytes. Existing extents are skipped, any extents
1317 * added are marked as "unwritten".
1319 static int ocfs2_allocate_unwritten_extents(struct inode
*inode
,
1323 u32 cpos
, phys_cpos
, clusters
, alloc_size
;
1324 u64 end
= start
+ len
;
1325 struct buffer_head
*di_bh
= NULL
;
1327 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1328 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
1335 * Nothing to do if the requested reservation range
1336 * fits within the inode.
1338 if (ocfs2_size_fits_inline_data(di_bh
, end
))
1341 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
1349 * We consider both start and len to be inclusive.
1351 cpos
= start
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1352 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, start
+ len
);
1356 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1364 * Hole or existing extent len can be arbitrary, so
1365 * cap it to our own allocation request.
1367 if (alloc_size
> clusters
)
1368 alloc_size
= clusters
;
1372 * We already have an allocation at this
1373 * region so we can safely skip it.
1378 ret
= __ocfs2_extend_allocation(inode
, cpos
, alloc_size
, 1);
1387 clusters
-= alloc_size
;
1398 * Truncate a byte range, avoiding pages within partial clusters. This
1399 * preserves those pages for the zeroing code to write to.
1401 static void ocfs2_truncate_cluster_pages(struct inode
*inode
, u64 byte_start
,
1404 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1406 struct address_space
*mapping
= inode
->i_mapping
;
1408 start
= (loff_t
)ocfs2_align_bytes_to_clusters(inode
->i_sb
, byte_start
);
1409 end
= byte_start
+ byte_len
;
1410 end
= end
& ~(osb
->s_clustersize
- 1);
1413 unmap_mapping_range(mapping
, start
, end
- start
, 0);
1414 truncate_inode_pages_range(mapping
, start
, end
- 1);
1418 static int ocfs2_zero_partial_clusters(struct inode
*inode
,
1422 u64 tmpend
, end
= start
+ len
;
1423 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1424 unsigned int csize
= osb
->s_clustersize
;
1428 * The "start" and "end" values are NOT necessarily part of
1429 * the range whose allocation is being deleted. Rather, this
1430 * is what the user passed in with the request. We must zero
1431 * partial clusters here. There's no need to worry about
1432 * physical allocation - the zeroing code knows to skip holes.
1434 mlog(0, "byte start: %llu, end: %llu\n",
1435 (unsigned long long)start
, (unsigned long long)end
);
1438 * If both edges are on a cluster boundary then there's no
1439 * zeroing required as the region is part of the allocation to
1442 if ((start
& (csize
- 1)) == 0 && (end
& (csize
- 1)) == 0)
1445 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1446 if (IS_ERR(handle
)) {
1447 ret
= PTR_ERR(handle
);
1453 * We want to get the byte offset of the end of the 1st cluster.
1455 tmpend
= (u64
)osb
->s_clustersize
+ (start
& ~(osb
->s_clustersize
- 1));
1459 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1460 (unsigned long long)start
, (unsigned long long)tmpend
);
1462 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, tmpend
);
1468 * This may make start and end equal, but the zeroing
1469 * code will skip any work in that case so there's no
1470 * need to catch it up here.
1472 start
= end
& ~(osb
->s_clustersize
- 1);
1474 mlog(0, "2nd range: start: %llu, end: %llu\n",
1475 (unsigned long long)start
, (unsigned long long)end
);
1477 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, end
);
1482 ocfs2_commit_trans(osb
, handle
);
1487 static int ocfs2_remove_inode_range(struct inode
*inode
,
1488 struct buffer_head
*di_bh
, u64 byte_start
,
1492 u32 trunc_start
, trunc_len
, cpos
, phys_cpos
, alloc_size
;
1493 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1494 struct ocfs2_cached_dealloc_ctxt dealloc
;
1495 struct address_space
*mapping
= inode
->i_mapping
;
1496 struct ocfs2_extent_tree et
;
1498 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), di_bh
);
1499 ocfs2_init_dealloc_ctxt(&dealloc
);
1504 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1505 ret
= ocfs2_truncate_inline(inode
, di_bh
, byte_start
,
1506 byte_start
+ byte_len
, 0);
1512 * There's no need to get fancy with the page cache
1513 * truncate of an inline-data inode. We're talking
1514 * about less than a page here, which will be cached
1515 * in the dinode buffer anyway.
1517 unmap_mapping_range(mapping
, 0, 0, 0);
1518 truncate_inode_pages(mapping
, 0);
1522 trunc_start
= ocfs2_clusters_for_bytes(osb
->sb
, byte_start
);
1523 trunc_len
= (byte_start
+ byte_len
) >> osb
->s_clustersize_bits
;
1524 if (trunc_len
>= trunc_start
)
1525 trunc_len
-= trunc_start
;
1529 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1530 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1531 (unsigned long long)byte_start
,
1532 (unsigned long long)byte_len
, trunc_start
, trunc_len
);
1534 ret
= ocfs2_zero_partial_clusters(inode
, byte_start
, byte_len
);
1542 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1549 if (alloc_size
> trunc_len
)
1550 alloc_size
= trunc_len
;
1552 /* Only do work for non-holes */
1553 if (phys_cpos
!= 0) {
1554 ret
= ocfs2_remove_btree_range(inode
, &et
, cpos
,
1555 phys_cpos
, alloc_size
,
1564 trunc_len
-= alloc_size
;
1567 ocfs2_truncate_cluster_pages(inode
, byte_start
, byte_len
);
1570 ocfs2_schedule_truncate_log_flush(osb
, 1);
1571 ocfs2_run_deallocs(osb
, &dealloc
);
1577 * Parts of this function taken from xfs_change_file_space()
1579 static int __ocfs2_change_file_space(struct file
*file
, struct inode
*inode
,
1580 loff_t f_pos
, unsigned int cmd
,
1581 struct ocfs2_space_resv
*sr
,
1587 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1588 struct buffer_head
*di_bh
= NULL
;
1590 unsigned long long max_off
= inode
->i_sb
->s_maxbytes
;
1592 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
1595 mutex_lock(&inode
->i_mutex
);
1598 * This prevents concurrent writes on other nodes
1600 ret
= ocfs2_rw_lock(inode
, 1);
1606 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1612 if (inode
->i_flags
& (S_IMMUTABLE
|S_APPEND
)) {
1614 goto out_inode_unlock
;
1617 switch (sr
->l_whence
) {
1618 case 0: /*SEEK_SET*/
1620 case 1: /*SEEK_CUR*/
1621 sr
->l_start
+= f_pos
;
1623 case 2: /*SEEK_END*/
1624 sr
->l_start
+= i_size_read(inode
);
1628 goto out_inode_unlock
;
1632 llen
= sr
->l_len
> 0 ? sr
->l_len
- 1 : sr
->l_len
;
1635 || sr
->l_start
> max_off
1636 || (sr
->l_start
+ llen
) < 0
1637 || (sr
->l_start
+ llen
) > max_off
) {
1639 goto out_inode_unlock
;
1641 size
= sr
->l_start
+ sr
->l_len
;
1643 if (cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) {
1644 if (sr
->l_len
<= 0) {
1646 goto out_inode_unlock
;
1650 if (file
&& should_remove_suid(file
->f_path
.dentry
)) {
1651 ret
= __ocfs2_write_remove_suid(inode
, di_bh
);
1654 goto out_inode_unlock
;
1658 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1660 case OCFS2_IOC_RESVSP
:
1661 case OCFS2_IOC_RESVSP64
:
1663 * This takes unsigned offsets, but the signed ones we
1664 * pass have been checked against overflow above.
1666 ret
= ocfs2_allocate_unwritten_extents(inode
, sr
->l_start
,
1669 case OCFS2_IOC_UNRESVSP
:
1670 case OCFS2_IOC_UNRESVSP64
:
1671 ret
= ocfs2_remove_inode_range(inode
, di_bh
, sr
->l_start
,
1677 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1680 goto out_inode_unlock
;
1684 * We update c/mtime for these changes
1686 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1687 if (IS_ERR(handle
)) {
1688 ret
= PTR_ERR(handle
);
1690 goto out_inode_unlock
;
1693 if (change_size
&& i_size_read(inode
) < size
)
1694 i_size_write(inode
, size
);
1696 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1697 ret
= ocfs2_mark_inode_dirty(handle
, inode
, di_bh
);
1701 ocfs2_commit_trans(osb
, handle
);
1705 ocfs2_inode_unlock(inode
, 1);
1707 ocfs2_rw_unlock(inode
, 1);
1710 mutex_unlock(&inode
->i_mutex
);
1714 int ocfs2_change_file_space(struct file
*file
, unsigned int cmd
,
1715 struct ocfs2_space_resv
*sr
)
1717 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1718 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1720 if ((cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) &&
1721 !ocfs2_writes_unwritten_extents(osb
))
1723 else if ((cmd
== OCFS2_IOC_UNRESVSP
|| cmd
== OCFS2_IOC_UNRESVSP64
) &&
1724 !ocfs2_sparse_alloc(osb
))
1727 if (!S_ISREG(inode
->i_mode
))
1730 if (!(file
->f_mode
& FMODE_WRITE
))
1733 return __ocfs2_change_file_space(file
, inode
, file
->f_pos
, cmd
, sr
, 0);
1736 static long ocfs2_fallocate(struct inode
*inode
, int mode
, loff_t offset
,
1739 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1740 struct ocfs2_space_resv sr
;
1741 int change_size
= 1;
1743 if (!ocfs2_writes_unwritten_extents(osb
))
1746 if (S_ISDIR(inode
->i_mode
))
1749 if (mode
& FALLOC_FL_KEEP_SIZE
)
1753 sr
.l_start
= (s64
)offset
;
1754 sr
.l_len
= (s64
)len
;
1756 return __ocfs2_change_file_space(NULL
, inode
, offset
,
1757 OCFS2_IOC_RESVSP64
, &sr
, change_size
);
1760 int ocfs2_check_range_for_refcount(struct inode
*inode
, loff_t pos
,
1764 unsigned int extent_flags
;
1765 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1766 struct super_block
*sb
= inode
->i_sb
;
1768 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
)) ||
1769 !(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
) ||
1770 OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
1773 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1774 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1777 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1784 if (phys_cpos
&& (extent_flags
& OCFS2_EXT_REFCOUNTED
)) {
1789 if (extent_len
> clusters
)
1790 extent_len
= clusters
;
1792 clusters
-= extent_len
;
1799 static int ocfs2_prepare_inode_for_refcount(struct inode
*inode
,
1800 loff_t pos
, size_t count
,
1804 struct buffer_head
*di_bh
= NULL
;
1805 u32 cpos
= pos
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1807 ocfs2_clusters_for_bytes(inode
->i_sb
, pos
+ count
) - cpos
;
1809 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1817 ret
= ocfs2_refcount_cow(inode
, di_bh
, cpos
, clusters
, UINT_MAX
);
1825 static int ocfs2_prepare_inode_for_write(struct dentry
*dentry
,
1832 int ret
= 0, meta_level
= 0;
1833 struct inode
*inode
= dentry
->d_inode
;
1834 loff_t saved_pos
, end
;
1837 * We start with a read level meta lock and only jump to an ex
1838 * if we need to make modifications here.
1841 ret
= ocfs2_inode_lock(inode
, NULL
, meta_level
);
1848 /* Clear suid / sgid if necessary. We do this here
1849 * instead of later in the write path because
1850 * remove_suid() calls ->setattr without any hint that
1851 * we may have already done our cluster locking. Since
1852 * ocfs2_setattr() *must* take cluster locks to
1853 * proceeed, this will lead us to recursively lock the
1854 * inode. There's also the dinode i_size state which
1855 * can be lost via setattr during extending writes (we
1856 * set inode->i_size at the end of a write. */
1857 if (should_remove_suid(dentry
)) {
1858 if (meta_level
== 0) {
1859 ocfs2_inode_unlock(inode
, meta_level
);
1864 ret
= ocfs2_write_remove_suid(inode
);
1871 /* work on a copy of ppos until we're sure that we won't have
1872 * to recalculate it due to relocking. */
1874 saved_pos
= i_size_read(inode
);
1875 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1880 end
= saved_pos
+ count
;
1882 ret
= ocfs2_check_range_for_refcount(inode
, saved_pos
, count
);
1884 ocfs2_inode_unlock(inode
, meta_level
);
1887 ret
= ocfs2_prepare_inode_for_refcount(inode
,
1903 * Skip the O_DIRECT checks if we don't need
1906 if (!direct_io
|| !(*direct_io
))
1910 * There's no sane way to do direct writes to an inode
1913 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1919 * Allowing concurrent direct writes means
1920 * i_size changes wouldn't be synchronized, so
1921 * one node could wind up truncating another
1924 if (end
> i_size_read(inode
)) {
1930 * We don't fill holes during direct io, so
1931 * check for them here. If any are found, the
1932 * caller will have to retake some cluster
1933 * locks and initiate the io as buffered.
1935 ret
= ocfs2_check_range_for_holes(inode
, saved_pos
, count
);
1948 if (meta_level
>= 0)
1949 ocfs2_inode_unlock(inode
, meta_level
);
1955 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
1956 const struct iovec
*iov
,
1957 unsigned long nr_segs
,
1960 int ret
, direct_io
, appending
, rw_level
, have_alloc_sem
= 0;
1961 int can_do_direct
, has_refcount
= 0;
1962 ssize_t written
= 0;
1963 size_t ocount
; /* original count */
1964 size_t count
; /* after file limit checks */
1965 loff_t old_size
, *ppos
= &iocb
->ki_pos
;
1967 struct file
*file
= iocb
->ki_filp
;
1968 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1969 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1971 mlog_entry("(0x%p, %u, '%.*s')\n", file
,
1972 (unsigned int)nr_segs
,
1973 file
->f_path
.dentry
->d_name
.len
,
1974 file
->f_path
.dentry
->d_name
.name
);
1976 if (iocb
->ki_left
== 0)
1979 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1981 appending
= file
->f_flags
& O_APPEND
? 1 : 0;
1982 direct_io
= file
->f_flags
& O_DIRECT
? 1 : 0;
1984 mutex_lock(&inode
->i_mutex
);
1987 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1989 down_read(&inode
->i_alloc_sem
);
1993 /* concurrent O_DIRECT writes are allowed */
1994 rw_level
= !direct_io
;
1995 ret
= ocfs2_rw_lock(inode
, rw_level
);
2001 can_do_direct
= direct_io
;
2002 ret
= ocfs2_prepare_inode_for_write(file
->f_path
.dentry
, ppos
,
2003 iocb
->ki_left
, appending
,
2004 &can_do_direct
, &has_refcount
);
2011 * We can't complete the direct I/O as requested, fall back to
2014 if (direct_io
&& !can_do_direct
) {
2015 ocfs2_rw_unlock(inode
, rw_level
);
2016 up_read(&inode
->i_alloc_sem
);
2026 * To later detect whether a journal commit for sync writes is
2027 * necessary, we sample i_size, and cluster count here.
2029 old_size
= i_size_read(inode
);
2030 old_clusters
= OCFS2_I(inode
)->ip_clusters
;
2032 /* communicate with ocfs2_dio_end_io */
2033 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2035 ret
= generic_segment_checks(iov
, &nr_segs
, &ocount
,
2041 ret
= generic_write_checks(file
, ppos
, &count
,
2042 S_ISBLK(inode
->i_mode
));
2047 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, *ppos
,
2048 ppos
, count
, ocount
);
2051 * direct write may have instantiated a few
2052 * blocks outside i_size. Trim these off again.
2053 * Don't need i_size_read because we hold i_mutex.
2055 if (*ppos
+ count
> inode
->i_size
)
2056 vmtruncate(inode
, inode
->i_size
);
2061 current
->backing_dev_info
= file
->f_mapping
->backing_dev_info
;
2062 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
, *ppos
,
2064 current
->backing_dev_info
= NULL
;
2068 /* buffered aio wouldn't have proper lock coverage today */
2069 BUG_ON(ret
== -EIOCBQUEUED
&& !(file
->f_flags
& O_DIRECT
));
2071 if (((file
->f_flags
& O_DSYNC
) && !direct_io
) || IS_SYNC(inode
) ||
2072 ((file
->f_flags
& O_DIRECT
) && has_refcount
)) {
2073 ret
= filemap_fdatawrite_range(file
->f_mapping
, pos
,
2078 if (!ret
&& ((old_size
!= i_size_read(inode
)) ||
2079 (old_clusters
!= OCFS2_I(inode
)->ip_clusters
) ||
2081 ret
= jbd2_journal_force_commit(osb
->journal
->j_journal
);
2087 ret
= filemap_fdatawait_range(file
->f_mapping
, pos
,
2092 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2093 * function pointer which is called when o_direct io completes so that
2094 * it can unlock our rw lock. (it's the clustered equivalent of
2095 * i_alloc_sem; protects truncate from racing with pending ios).
2096 * Unfortunately there are error cases which call end_io and others
2097 * that don't. so we don't have to unlock the rw_lock if either an
2098 * async dio is going to do it in the future or an end_io after an
2099 * error has already done it.
2101 if ((ret
== -EIOCBQUEUED
) || (!ocfs2_iocb_is_rw_locked(iocb
))) {
2108 ocfs2_rw_unlock(inode
, rw_level
);
2112 up_read(&inode
->i_alloc_sem
);
2114 mutex_unlock(&inode
->i_mutex
);
2122 static int ocfs2_splice_to_file(struct pipe_inode_info
*pipe
,
2124 struct splice_desc
*sd
)
2128 ret
= ocfs2_prepare_inode_for_write(out
->f_path
.dentry
, &sd
->pos
,
2129 sd
->total_len
, 0, NULL
, NULL
);
2135 return splice_from_pipe_feed(pipe
, sd
, pipe_to_file
);
2138 static ssize_t
ocfs2_file_splice_write(struct pipe_inode_info
*pipe
,
2145 struct address_space
*mapping
= out
->f_mapping
;
2146 struct inode
*inode
= mapping
->host
;
2147 struct splice_desc sd
= {
2154 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out
, pipe
,
2156 out
->f_path
.dentry
->d_name
.len
,
2157 out
->f_path
.dentry
->d_name
.name
);
2160 mutex_lock_nested(&pipe
->inode
->i_mutex
, I_MUTEX_PARENT
);
2162 splice_from_pipe_begin(&sd
);
2164 ret
= splice_from_pipe_next(pipe
, &sd
);
2168 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2169 ret
= ocfs2_rw_lock(inode
, 1);
2173 ret
= ocfs2_splice_to_file(pipe
, out
, &sd
);
2174 ocfs2_rw_unlock(inode
, 1);
2176 mutex_unlock(&inode
->i_mutex
);
2178 splice_from_pipe_end(pipe
, &sd
);
2181 mutex_unlock(&pipe
->inode
->i_mutex
);
2184 ret
= sd
.num_spliced
;
2187 unsigned long nr_pages
;
2190 nr_pages
= (ret
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2192 err
= generic_write_sync(out
, *ppos
, ret
);
2198 balance_dirty_pages_ratelimited_nr(mapping
, nr_pages
);
2205 static ssize_t
ocfs2_file_splice_read(struct file
*in
,
2207 struct pipe_inode_info
*pipe
,
2211 int ret
= 0, lock_level
= 0;
2212 struct inode
*inode
= in
->f_path
.dentry
->d_inode
;
2214 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in
, pipe
,
2216 in
->f_path
.dentry
->d_name
.len
,
2217 in
->f_path
.dentry
->d_name
.name
);
2220 * See the comment in ocfs2_file_aio_read()
2222 ret
= ocfs2_inode_lock_atime(inode
, in
->f_vfsmnt
, &lock_level
);
2227 ocfs2_inode_unlock(inode
, lock_level
);
2229 ret
= generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
2236 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
2237 const struct iovec
*iov
,
2238 unsigned long nr_segs
,
2241 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0, lock_level
= 0;
2242 struct file
*filp
= iocb
->ki_filp
;
2243 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
2245 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
2246 (unsigned int)nr_segs
,
2247 filp
->f_path
.dentry
->d_name
.len
,
2248 filp
->f_path
.dentry
->d_name
.name
);
2257 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2258 * need locks to protect pending reads from racing with truncate.
2260 if (filp
->f_flags
& O_DIRECT
) {
2261 down_read(&inode
->i_alloc_sem
);
2264 ret
= ocfs2_rw_lock(inode
, 0);
2270 /* communicate with ocfs2_dio_end_io */
2271 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2275 * We're fine letting folks race truncates and extending
2276 * writes with read across the cluster, just like they can
2277 * locally. Hence no rw_lock during read.
2279 * Take and drop the meta data lock to update inode fields
2280 * like i_size. This allows the checks down below
2281 * generic_file_aio_read() a chance of actually working.
2283 ret
= ocfs2_inode_lock_atime(inode
, filp
->f_vfsmnt
, &lock_level
);
2288 ocfs2_inode_unlock(inode
, lock_level
);
2290 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
2292 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2294 /* buffered aio wouldn't have proper lock coverage today */
2295 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
2297 /* see ocfs2_file_aio_write */
2298 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
2305 up_read(&inode
->i_alloc_sem
);
2307 ocfs2_rw_unlock(inode
, rw_level
);
2313 const struct inode_operations ocfs2_file_iops
= {
2314 .setattr
= ocfs2_setattr
,
2315 .getattr
= ocfs2_getattr
,
2316 .permission
= ocfs2_permission
,
2317 .setxattr
= generic_setxattr
,
2318 .getxattr
= generic_getxattr
,
2319 .listxattr
= ocfs2_listxattr
,
2320 .removexattr
= generic_removexattr
,
2321 .fallocate
= ocfs2_fallocate
,
2322 .fiemap
= ocfs2_fiemap
,
2325 const struct inode_operations ocfs2_special_file_iops
= {
2326 .setattr
= ocfs2_setattr
,
2327 .getattr
= ocfs2_getattr
,
2328 .permission
= ocfs2_permission
,
2332 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2333 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2335 const struct file_operations ocfs2_fops
= {
2336 .llseek
= generic_file_llseek
,
2337 .read
= do_sync_read
,
2338 .write
= do_sync_write
,
2340 .fsync
= ocfs2_sync_file
,
2341 .release
= ocfs2_file_release
,
2342 .open
= ocfs2_file_open
,
2343 .aio_read
= ocfs2_file_aio_read
,
2344 .aio_write
= ocfs2_file_aio_write
,
2345 .unlocked_ioctl
= ocfs2_ioctl
,
2346 #ifdef CONFIG_COMPAT
2347 .compat_ioctl
= ocfs2_compat_ioctl
,
2350 .flock
= ocfs2_flock
,
2351 .splice_read
= ocfs2_file_splice_read
,
2352 .splice_write
= ocfs2_file_splice_write
,
2355 const struct file_operations ocfs2_dops
= {
2356 .llseek
= generic_file_llseek
,
2357 .read
= generic_read_dir
,
2358 .readdir
= ocfs2_readdir
,
2359 .fsync
= ocfs2_sync_file
,
2360 .release
= ocfs2_dir_release
,
2361 .open
= ocfs2_dir_open
,
2362 .unlocked_ioctl
= ocfs2_ioctl
,
2363 #ifdef CONFIG_COMPAT
2364 .compat_ioctl
= ocfs2_compat_ioctl
,
2367 .flock
= ocfs2_flock
,
2371 * POSIX-lockless variants of our file_operations.
2373 * These will be used if the underlying cluster stack does not support
2374 * posix file locking, if the user passes the "localflocks" mount
2375 * option, or if we have a local-only fs.
2377 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2378 * so we still want it in the case of no stack support for
2379 * plocks. Internally, it will do the right thing when asked to ignore
2382 const struct file_operations ocfs2_fops_no_plocks
= {
2383 .llseek
= generic_file_llseek
,
2384 .read
= do_sync_read
,
2385 .write
= do_sync_write
,
2387 .fsync
= ocfs2_sync_file
,
2388 .release
= ocfs2_file_release
,
2389 .open
= ocfs2_file_open
,
2390 .aio_read
= ocfs2_file_aio_read
,
2391 .aio_write
= ocfs2_file_aio_write
,
2392 .unlocked_ioctl
= ocfs2_ioctl
,
2393 #ifdef CONFIG_COMPAT
2394 .compat_ioctl
= ocfs2_compat_ioctl
,
2396 .flock
= ocfs2_flock
,
2397 .splice_read
= ocfs2_file_splice_read
,
2398 .splice_write
= ocfs2_file_splice_write
,
2401 const struct file_operations ocfs2_dops_no_plocks
= {
2402 .llseek
= generic_file_llseek
,
2403 .read
= generic_read_dir
,
2404 .readdir
= ocfs2_readdir
,
2405 .fsync
= ocfs2_sync_file
,
2406 .release
= ocfs2_dir_release
,
2407 .open
= ocfs2_dir_open
,
2408 .unlocked_ioctl
= ocfs2_ioctl
,
2409 #ifdef CONFIG_COMPAT
2410 .compat_ioctl
= ocfs2_compat_ioctl
,
2412 .flock
= ocfs2_flock
,