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
);
281 ocfs2_journal_dirty(handle
, bh
);
284 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
290 static int ocfs2_set_inode_size(handle_t
*handle
,
292 struct buffer_head
*fe_bh
,
298 i_size_write(inode
, new_i_size
);
299 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
300 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
302 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
313 int ocfs2_simple_size_update(struct inode
*inode
,
314 struct buffer_head
*di_bh
,
318 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
319 handle_t
*handle
= NULL
;
321 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
322 if (IS_ERR(handle
)) {
323 ret
= PTR_ERR(handle
);
328 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
333 ocfs2_commit_trans(osb
, handle
);
338 static int ocfs2_cow_file_pos(struct inode
*inode
,
339 struct buffer_head
*fe_bh
,
343 u32 phys
, cpos
= offset
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
344 unsigned int num_clusters
= 0;
345 unsigned int ext_flags
= 0;
348 * If the new offset is aligned to the range of the cluster, there is
349 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
352 if ((offset
& (OCFS2_SB(inode
->i_sb
)->s_clustersize
- 1)) == 0)
355 status
= ocfs2_get_clusters(inode
, cpos
, &phys
,
356 &num_clusters
, &ext_flags
);
362 if (!(ext_flags
& OCFS2_EXT_REFCOUNTED
))
365 return ocfs2_refcount_cow(inode
, fe_bh
, cpos
, 1, cpos
+1);
371 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
373 struct buffer_head
*fe_bh
,
378 struct ocfs2_dinode
*di
;
384 * We need to CoW the cluster contains the offset if it is reflinked
385 * since we will call ocfs2_zero_range_for_truncate later which will
386 * write "0" from offset to the end of the cluster.
388 status
= ocfs2_cow_file_pos(inode
, fe_bh
, new_i_size
);
394 /* TODO: This needs to actually orphan the inode in this
397 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
398 if (IS_ERR(handle
)) {
399 status
= PTR_ERR(handle
);
404 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
405 OCFS2_JOURNAL_ACCESS_WRITE
);
412 * Do this before setting i_size.
414 cluster_bytes
= ocfs2_align_bytes_to_clusters(inode
->i_sb
, new_i_size
);
415 status
= ocfs2_zero_range_for_truncate(inode
, handle
, new_i_size
,
422 i_size_write(inode
, new_i_size
);
423 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
425 di
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
426 di
->i_size
= cpu_to_le64(new_i_size
);
427 di
->i_ctime
= di
->i_mtime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
428 di
->i_ctime_nsec
= di
->i_mtime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
430 ocfs2_journal_dirty(handle
, fe_bh
);
433 ocfs2_commit_trans(osb
, handle
);
440 static int ocfs2_truncate_file(struct inode
*inode
,
441 struct buffer_head
*di_bh
,
445 struct ocfs2_dinode
*fe
= NULL
;
446 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
448 mlog_entry("(inode = %llu, new_i_size = %llu\n",
449 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
450 (unsigned long long)new_i_size
);
452 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
453 * already validated it */
454 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
456 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
457 "Inode %llu, inode i_size = %lld != di "
458 "i_size = %llu, i_flags = 0x%x\n",
459 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
461 (unsigned long long)le64_to_cpu(fe
->i_size
),
462 le32_to_cpu(fe
->i_flags
));
464 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
465 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
466 (unsigned long long)le64_to_cpu(fe
->i_size
),
467 (unsigned long long)new_i_size
);
473 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
474 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
475 (unsigned long long)le64_to_cpu(fe
->i_size
),
476 (unsigned long long)new_i_size
);
478 /* lets handle the simple truncate cases before doing any more
479 * cluster locking. */
480 if (new_i_size
== le64_to_cpu(fe
->i_size
))
483 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
485 ocfs2_resv_discard(&osb
->osb_la_resmap
,
486 &OCFS2_I(inode
)->ip_la_data_resv
);
489 * The inode lock forced other nodes to sync and drop their
490 * pages, which (correctly) happens even if we have a truncate
491 * without allocation change - ocfs2 cluster sizes can be much
492 * greater than page size, so we have to truncate them
495 unmap_mapping_range(inode
->i_mapping
, new_i_size
+ PAGE_SIZE
- 1, 0, 1);
496 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
498 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
499 status
= ocfs2_truncate_inline(inode
, di_bh
, new_i_size
,
500 i_size_read(inode
), 1);
504 goto bail_unlock_sem
;
507 /* alright, we're going to need to do a full blown alloc size
508 * change. Orphan the inode so that recovery can complete the
509 * truncate if necessary. This does the task of marking
511 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
514 goto bail_unlock_sem
;
517 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
);
520 goto bail_unlock_sem
;
523 /* TODO: orphan dir cleanup here. */
525 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
528 if (!status
&& OCFS2_I(inode
)->ip_clusters
== 0)
529 status
= ocfs2_try_remove_refcount_tree(inode
, di_bh
);
536 * extend file allocation only here.
537 * we'll update all the disk stuff, and oip->alloc_size
539 * expect stuff to be locked, a transaction started and enough data /
540 * metadata reservations in the contexts.
542 * Will return -EAGAIN, and a reason if a restart is needed.
543 * If passed in, *reason will always be set, even in error.
545 int ocfs2_add_inode_data(struct ocfs2_super
*osb
,
550 struct buffer_head
*fe_bh
,
552 struct ocfs2_alloc_context
*data_ac
,
553 struct ocfs2_alloc_context
*meta_ac
,
554 enum ocfs2_alloc_restarted
*reason_ret
)
557 struct ocfs2_extent_tree et
;
559 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), fe_bh
);
560 ret
= ocfs2_add_clusters_in_btree(handle
, &et
, logical_offset
,
561 clusters_to_add
, mark_unwritten
,
562 data_ac
, meta_ac
, reason_ret
);
567 static int __ocfs2_extend_allocation(struct inode
*inode
, u32 logical_start
,
568 u32 clusters_to_add
, int mark_unwritten
)
571 int restart_func
= 0;
574 struct buffer_head
*bh
= NULL
;
575 struct ocfs2_dinode
*fe
= NULL
;
576 handle_t
*handle
= NULL
;
577 struct ocfs2_alloc_context
*data_ac
= NULL
;
578 struct ocfs2_alloc_context
*meta_ac
= NULL
;
579 enum ocfs2_alloc_restarted why
;
580 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
581 struct ocfs2_extent_tree et
;
584 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
587 * This function only exists for file systems which don't
590 BUG_ON(mark_unwritten
&& !ocfs2_sparse_alloc(osb
));
592 status
= ocfs2_read_inode_block(inode
, &bh
);
597 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
600 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
602 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
603 "clusters_to_add = %u\n",
604 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
605 (long long)i_size_read(inode
), le32_to_cpu(fe
->i_clusters
),
607 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), bh
);
608 status
= ocfs2_lock_allocators(inode
, &et
, clusters_to_add
, 0,
615 credits
= ocfs2_calc_extend_credits(osb
->sb
, &fe
->id2
.i_list
,
617 handle
= ocfs2_start_trans(osb
, credits
);
618 if (IS_ERR(handle
)) {
619 status
= PTR_ERR(handle
);
625 restarted_transaction
:
626 status
= dquot_alloc_space_nodirty(inode
,
627 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
632 /* reserve a write to the file entry early on - that we if we
633 * run out of credits in the allocation path, we can still
635 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
636 OCFS2_JOURNAL_ACCESS_WRITE
);
642 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
644 status
= ocfs2_add_inode_data(osb
,
654 if ((status
< 0) && (status
!= -EAGAIN
)) {
655 if (status
!= -ENOSPC
)
660 ocfs2_journal_dirty(handle
, bh
);
662 spin_lock(&OCFS2_I(inode
)->ip_lock
);
663 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
664 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
665 /* Release unused quota reservation */
666 dquot_free_space(inode
,
667 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
670 if (why
!= RESTART_NONE
&& clusters_to_add
) {
671 if (why
== RESTART_META
) {
672 mlog(0, "restarting function.\n");
676 BUG_ON(why
!= RESTART_TRANS
);
678 mlog(0, "restarting transaction.\n");
679 /* TODO: This can be more intelligent. */
680 credits
= ocfs2_calc_extend_credits(osb
->sb
,
683 status
= ocfs2_extend_trans(handle
, credits
);
685 /* handle still has to be committed at
691 goto restarted_transaction
;
695 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
696 le32_to_cpu(fe
->i_clusters
),
697 (unsigned long long)le64_to_cpu(fe
->i_size
));
698 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
699 OCFS2_I(inode
)->ip_clusters
, (long long)i_size_read(inode
));
702 if (status
< 0 && did_quota
)
703 dquot_free_space(inode
,
704 ocfs2_clusters_to_bytes(osb
->sb
, clusters_to_add
));
706 ocfs2_commit_trans(osb
, handle
);
710 ocfs2_free_alloc_context(data_ac
);
714 ocfs2_free_alloc_context(meta_ac
);
717 if ((!status
) && restart_func
) {
728 /* Some parts of this taken from generic_cont_expand, which turned out
729 * to be too fragile to do exactly what we need without us having to
730 * worry about recursive locking in ->write_begin() and ->write_end(). */
731 static int ocfs2_write_zero_page(struct inode
*inode
,
734 struct address_space
*mapping
= inode
->i_mapping
;
738 handle_t
*handle
= NULL
;
741 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
742 /* ugh. in prepare/commit_write, if from==to==start of block, we
743 ** skip the prepare. make sure we never send an offset for the start
746 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
749 index
= size
>> PAGE_CACHE_SHIFT
;
751 page
= grab_cache_page(mapping
, index
);
758 ret
= ocfs2_prepare_write_nolock(inode
, page
, offset
, offset
);
764 if (ocfs2_should_order_data(inode
)) {
765 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
767 if (IS_ERR(handle
)) {
768 ret
= PTR_ERR(handle
);
774 /* must not update i_size! */
775 ret
= block_commit_write(page
, offset
, offset
);
782 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
785 page_cache_release(page
);
790 static int ocfs2_zero_extend(struct inode
*inode
,
795 struct super_block
*sb
= inode
->i_sb
;
797 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
798 while (start_off
< zero_to_size
) {
799 ret
= ocfs2_write_zero_page(inode
, start_off
);
805 start_off
+= sb
->s_blocksize
;
808 * Very large extends have the potential to lock up
809 * the cpu for extended periods of time.
818 int ocfs2_extend_no_holes(struct inode
*inode
, u64 new_i_size
, u64 zero_to
)
822 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
824 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
);
825 if (clusters_to_add
< oi
->ip_clusters
)
828 clusters_to_add
-= oi
->ip_clusters
;
830 if (clusters_to_add
) {
831 ret
= __ocfs2_extend_allocation(inode
, oi
->ip_clusters
,
840 * Call this even if we don't add any clusters to the tree. We
841 * still need to zero the area between the old i_size and the
844 ret
= ocfs2_zero_extend(inode
, zero_to
);
852 static int ocfs2_extend_file(struct inode
*inode
,
853 struct buffer_head
*di_bh
,
857 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
861 /* setattr sometimes calls us like this. */
865 if (i_size_read(inode
) == new_i_size
)
867 BUG_ON(new_i_size
< i_size_read(inode
));
870 * Fall through for converting inline data, even if the fs
871 * supports sparse files.
873 * The check for inline data here is legal - nobody can add
874 * the feature since we have i_mutex. We must check it again
875 * after acquiring ip_alloc_sem though, as paths like mmap
876 * might have raced us to converting the inode to extents.
878 if (!(oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
879 && ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
880 goto out_update_size
;
883 * The alloc sem blocks people in read/write from reading our
884 * allocation until we're done changing it. We depend on
885 * i_mutex to block other extend/truncate calls while we're
888 down_write(&oi
->ip_alloc_sem
);
890 if (oi
->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
892 * We can optimize small extends by keeping the inodes
895 if (ocfs2_size_fits_inline_data(di_bh
, new_i_size
)) {
896 up_write(&oi
->ip_alloc_sem
);
897 goto out_update_size
;
900 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
902 up_write(&oi
->ip_alloc_sem
);
909 if (!ocfs2_sparse_alloc(OCFS2_SB(inode
->i_sb
)))
910 ret
= ocfs2_extend_no_holes(inode
, new_i_size
, new_i_size
);
912 up_write(&oi
->ip_alloc_sem
);
920 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
928 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
930 int status
= 0, size_change
;
931 struct inode
*inode
= dentry
->d_inode
;
932 struct super_block
*sb
= inode
->i_sb
;
933 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
934 struct buffer_head
*bh
= NULL
;
935 handle_t
*handle
= NULL
;
936 struct dquot
*transfer_to
[MAXQUOTAS
] = { };
939 mlog_entry("(0x%p, '%.*s')\n", dentry
,
940 dentry
->d_name
.len
, dentry
->d_name
.name
);
942 /* ensuring we don't even attempt to truncate a symlink */
943 if (S_ISLNK(inode
->i_mode
))
944 attr
->ia_valid
&= ~ATTR_SIZE
;
946 if (attr
->ia_valid
& ATTR_MODE
)
947 mlog(0, "mode change: %d\n", attr
->ia_mode
);
948 if (attr
->ia_valid
& ATTR_UID
)
949 mlog(0, "uid change: %d\n", attr
->ia_uid
);
950 if (attr
->ia_valid
& ATTR_GID
)
951 mlog(0, "gid change: %d\n", attr
->ia_gid
);
952 if (attr
->ia_valid
& ATTR_SIZE
)
953 mlog(0, "size change...\n");
954 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
955 mlog(0, "time change...\n");
957 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
958 | ATTR_GID | ATTR_UID | ATTR_MODE)
959 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
960 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
964 status
= inode_change_ok(inode
, attr
);
968 if (is_quota_modification(inode
, attr
))
969 dquot_initialize(inode
);
970 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
972 status
= ocfs2_rw_lock(inode
, 1);
979 status
= ocfs2_inode_lock(inode
, &bh
, 1);
981 if (status
!= -ENOENT
)
986 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
987 status
= inode_newsize_ok(inode
, attr
->ia_size
);
991 if (i_size_read(inode
) > attr
->ia_size
) {
992 if (ocfs2_should_order_data(inode
)) {
993 status
= ocfs2_begin_ordered_truncate(inode
,
998 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
1000 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
1002 if (status
!= -ENOSPC
)
1009 if ((attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
1010 (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
1012 * Gather pointers to quota structures so that allocation /
1013 * freeing of quota structures happens here and not inside
1014 * dquot_transfer() where we have problems with lock ordering
1016 if (attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
1017 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
1018 OCFS2_FEATURE_RO_COMPAT_USRQUOTA
)) {
1019 transfer_to
[USRQUOTA
] = dqget(sb
, attr
->ia_uid
,
1021 if (!transfer_to
[USRQUOTA
]) {
1026 if (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
1027 && OCFS2_HAS_RO_COMPAT_FEATURE(sb
,
1028 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA
)) {
1029 transfer_to
[GRPQUOTA
] = dqget(sb
, attr
->ia_gid
,
1031 if (!transfer_to
[GRPQUOTA
]) {
1036 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
+
1037 2 * ocfs2_quota_trans_credits(sb
));
1038 if (IS_ERR(handle
)) {
1039 status
= PTR_ERR(handle
);
1043 status
= __dquot_transfer(inode
, transfer_to
);
1047 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1048 if (IS_ERR(handle
)) {
1049 status
= PTR_ERR(handle
);
1056 * This will intentionally not wind up calling vmtruncate(),
1057 * since all the work for a size change has been done above.
1058 * Otherwise, we could get into problems with truncate as
1059 * ip_alloc_sem is used there to protect against i_size
1062 status
= inode_setattr(inode
, attr
);
1068 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
1073 ocfs2_commit_trans(osb
, handle
);
1075 ocfs2_inode_unlock(inode
, 1);
1078 ocfs2_rw_unlock(inode
, 1);
1082 /* Release quota pointers in case we acquired them */
1083 for (qtype
= 0; qtype
< MAXQUOTAS
; qtype
++)
1084 dqput(transfer_to
[qtype
]);
1086 if (!status
&& attr
->ia_valid
& ATTR_MODE
) {
1087 status
= ocfs2_acl_chmod(inode
);
1096 int ocfs2_getattr(struct vfsmount
*mnt
,
1097 struct dentry
*dentry
,
1100 struct inode
*inode
= dentry
->d_inode
;
1101 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
1102 struct ocfs2_super
*osb
= sb
->s_fs_info
;
1107 err
= ocfs2_inode_revalidate(dentry
);
1114 generic_fillattr(inode
, stat
);
1116 /* We set the blksize from the cluster size for performance */
1117 stat
->blksize
= osb
->s_clustersize
;
1125 int ocfs2_permission(struct inode
*inode
, int mask
)
1131 ret
= ocfs2_inode_lock(inode
, NULL
, 0);
1138 ret
= generic_permission(inode
, mask
, ocfs2_check_acl
);
1140 ocfs2_inode_unlock(inode
, 0);
1146 static int __ocfs2_write_remove_suid(struct inode
*inode
,
1147 struct buffer_head
*bh
)
1151 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1152 struct ocfs2_dinode
*di
;
1154 mlog_entry("(Inode %llu, mode 0%o)\n",
1155 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_mode
);
1157 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1158 if (IS_ERR(handle
)) {
1159 ret
= PTR_ERR(handle
);
1164 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
1165 OCFS2_JOURNAL_ACCESS_WRITE
);
1171 inode
->i_mode
&= ~S_ISUID
;
1172 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
1173 inode
->i_mode
&= ~S_ISGID
;
1175 di
= (struct ocfs2_dinode
*) bh
->b_data
;
1176 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
1178 ocfs2_journal_dirty(handle
, bh
);
1181 ocfs2_commit_trans(osb
, handle
);
1188 * Will look for holes and unwritten extents in the range starting at
1189 * pos for count bytes (inclusive).
1191 static int ocfs2_check_range_for_holes(struct inode
*inode
, loff_t pos
,
1195 unsigned int extent_flags
;
1196 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1197 struct super_block
*sb
= inode
->i_sb
;
1199 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1200 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1203 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1210 if (phys_cpos
== 0 || (extent_flags
& OCFS2_EXT_UNWRITTEN
)) {
1215 if (extent_len
> clusters
)
1216 extent_len
= clusters
;
1218 clusters
-= extent_len
;
1225 static int ocfs2_write_remove_suid(struct inode
*inode
)
1228 struct buffer_head
*bh
= NULL
;
1230 ret
= ocfs2_read_inode_block(inode
, &bh
);
1236 ret
= __ocfs2_write_remove_suid(inode
, bh
);
1243 * Allocate enough extents to cover the region starting at byte offset
1244 * start for len bytes. Existing extents are skipped, any extents
1245 * added are marked as "unwritten".
1247 static int ocfs2_allocate_unwritten_extents(struct inode
*inode
,
1251 u32 cpos
, phys_cpos
, clusters
, alloc_size
;
1252 u64 end
= start
+ len
;
1253 struct buffer_head
*di_bh
= NULL
;
1255 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1256 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
1263 * Nothing to do if the requested reservation range
1264 * fits within the inode.
1266 if (ocfs2_size_fits_inline_data(di_bh
, end
))
1269 ret
= ocfs2_convert_inline_data_to_extents(inode
, di_bh
);
1277 * We consider both start and len to be inclusive.
1279 cpos
= start
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1280 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, start
+ len
);
1284 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
,
1292 * Hole or existing extent len can be arbitrary, so
1293 * cap it to our own allocation request.
1295 if (alloc_size
> clusters
)
1296 alloc_size
= clusters
;
1300 * We already have an allocation at this
1301 * region so we can safely skip it.
1306 ret
= __ocfs2_extend_allocation(inode
, cpos
, alloc_size
, 1);
1315 clusters
-= alloc_size
;
1326 * Truncate a byte range, avoiding pages within partial clusters. This
1327 * preserves those pages for the zeroing code to write to.
1329 static void ocfs2_truncate_cluster_pages(struct inode
*inode
, u64 byte_start
,
1332 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1334 struct address_space
*mapping
= inode
->i_mapping
;
1336 start
= (loff_t
)ocfs2_align_bytes_to_clusters(inode
->i_sb
, byte_start
);
1337 end
= byte_start
+ byte_len
;
1338 end
= end
& ~(osb
->s_clustersize
- 1);
1341 unmap_mapping_range(mapping
, start
, end
- start
, 0);
1342 truncate_inode_pages_range(mapping
, start
, end
- 1);
1346 static int ocfs2_zero_partial_clusters(struct inode
*inode
,
1350 u64 tmpend
, end
= start
+ len
;
1351 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1352 unsigned int csize
= osb
->s_clustersize
;
1356 * The "start" and "end" values are NOT necessarily part of
1357 * the range whose allocation is being deleted. Rather, this
1358 * is what the user passed in with the request. We must zero
1359 * partial clusters here. There's no need to worry about
1360 * physical allocation - the zeroing code knows to skip holes.
1362 mlog(0, "byte start: %llu, end: %llu\n",
1363 (unsigned long long)start
, (unsigned long long)end
);
1366 * If both edges are on a cluster boundary then there's no
1367 * zeroing required as the region is part of the allocation to
1370 if ((start
& (csize
- 1)) == 0 && (end
& (csize
- 1)) == 0)
1373 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1374 if (IS_ERR(handle
)) {
1375 ret
= PTR_ERR(handle
);
1381 * We want to get the byte offset of the end of the 1st cluster.
1383 tmpend
= (u64
)osb
->s_clustersize
+ (start
& ~(osb
->s_clustersize
- 1));
1387 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1388 (unsigned long long)start
, (unsigned long long)tmpend
);
1390 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, tmpend
);
1396 * This may make start and end equal, but the zeroing
1397 * code will skip any work in that case so there's no
1398 * need to catch it up here.
1400 start
= end
& ~(osb
->s_clustersize
- 1);
1402 mlog(0, "2nd range: start: %llu, end: %llu\n",
1403 (unsigned long long)start
, (unsigned long long)end
);
1405 ret
= ocfs2_zero_range_for_truncate(inode
, handle
, start
, end
);
1410 ocfs2_commit_trans(osb
, handle
);
1415 static int ocfs2_find_rec(struct ocfs2_extent_list
*el
, u32 pos
)
1418 struct ocfs2_extent_rec
*rec
= NULL
;
1420 for (i
= le16_to_cpu(el
->l_next_free_rec
) - 1; i
>= 0; i
--) {
1422 rec
= &el
->l_recs
[i
];
1424 if (le32_to_cpu(rec
->e_cpos
) < pos
)
1432 * Helper to calculate the punching pos and length in one run, we handle the
1433 * following three cases in order:
1435 * - remove the entire record
1436 * - remove a partial record
1437 * - no record needs to be removed (hole-punching completed)
1439 static void ocfs2_calc_trunc_pos(struct inode
*inode
,
1440 struct ocfs2_extent_list
*el
,
1441 struct ocfs2_extent_rec
*rec
,
1442 u32 trunc_start
, u32
*trunc_cpos
,
1443 u32
*trunc_len
, u32
*trunc_end
,
1444 u64
*blkno
, int *done
)
1449 range
= le32_to_cpu(rec
->e_cpos
) + ocfs2_rec_clusters(el
, rec
);
1451 if (le32_to_cpu(rec
->e_cpos
) >= trunc_start
) {
1452 *trunc_cpos
= le32_to_cpu(rec
->e_cpos
);
1454 * Skip holes if any.
1456 if (range
< *trunc_end
)
1458 *trunc_len
= *trunc_end
- le32_to_cpu(rec
->e_cpos
);
1459 *blkno
= le64_to_cpu(rec
->e_blkno
);
1460 *trunc_end
= le32_to_cpu(rec
->e_cpos
);
1461 } else if (range
> trunc_start
) {
1462 *trunc_cpos
= trunc_start
;
1463 *trunc_len
= *trunc_end
- trunc_start
;
1464 coff
= trunc_start
- le32_to_cpu(rec
->e_cpos
);
1465 *blkno
= le64_to_cpu(rec
->e_blkno
) +
1466 ocfs2_clusters_to_blocks(inode
->i_sb
, coff
);
1467 *trunc_end
= trunc_start
;
1470 * It may have two following possibilities:
1472 * - last record has been removed
1473 * - trunc_start was within a hole
1475 * both two cases mean the completion of hole punching.
1483 static int ocfs2_remove_inode_range(struct inode
*inode
,
1484 struct buffer_head
*di_bh
, u64 byte_start
,
1487 int ret
= 0, flags
= 0, done
= 0, i
;
1488 u32 trunc_start
, trunc_len
, trunc_end
, trunc_cpos
, phys_cpos
;
1490 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1491 struct ocfs2_cached_dealloc_ctxt dealloc
;
1492 struct address_space
*mapping
= inode
->i_mapping
;
1493 struct ocfs2_extent_tree et
;
1494 struct ocfs2_path
*path
= NULL
;
1495 struct ocfs2_extent_list
*el
= NULL
;
1496 struct ocfs2_extent_rec
*rec
= NULL
;
1497 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1498 u64 blkno
, refcount_loc
= le64_to_cpu(di
->i_refcount_loc
);
1500 ocfs2_init_dinode_extent_tree(&et
, INODE_CACHE(inode
), di_bh
);
1501 ocfs2_init_dealloc_ctxt(&dealloc
);
1506 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1507 ret
= ocfs2_truncate_inline(inode
, di_bh
, byte_start
,
1508 byte_start
+ byte_len
, 0);
1514 * There's no need to get fancy with the page cache
1515 * truncate of an inline-data inode. We're talking
1516 * about less than a page here, which will be cached
1517 * in the dinode buffer anyway.
1519 unmap_mapping_range(mapping
, 0, 0, 0);
1520 truncate_inode_pages(mapping
, 0);
1525 * For reflinks, we may need to CoW 2 clusters which might be
1526 * partially zero'd later, if hole's start and end offset were
1527 * within one cluster(means is not exactly aligned to clustersize).
1530 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
) {
1532 ret
= ocfs2_cow_file_pos(inode
, di_bh
, byte_start
);
1538 ret
= ocfs2_cow_file_pos(inode
, di_bh
, byte_start
+ byte_len
);
1545 trunc_start
= ocfs2_clusters_for_bytes(osb
->sb
, byte_start
);
1546 trunc_end
= (byte_start
+ byte_len
) >> osb
->s_clustersize_bits
;
1547 cluster_in_el
= trunc_end
;
1549 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n",
1550 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1551 (unsigned long long)byte_start
,
1552 (unsigned long long)byte_len
, trunc_start
, trunc_end
);
1554 ret
= ocfs2_zero_partial_clusters(inode
, byte_start
, byte_len
);
1560 path
= ocfs2_new_path_from_et(&et
);
1567 while (trunc_end
> trunc_start
) {
1569 ret
= ocfs2_find_path(INODE_CACHE(inode
), path
,
1576 el
= path_leaf_el(path
);
1578 i
= ocfs2_find_rec(el
, trunc_end
);
1580 * Need to go to previous extent block.
1583 if (path
->p_tree_depth
== 0)
1586 ret
= ocfs2_find_cpos_for_left_leaf(inode
->i_sb
,
1595 * We've reached the leftmost extent block,
1596 * it's safe to leave.
1598 if (cluster_in_el
== 0)
1602 * The 'pos' searched for previous extent block is
1603 * always one cluster less than actual trunc_end.
1605 trunc_end
= cluster_in_el
+ 1;
1607 ocfs2_reinit_path(path
, 1);
1612 rec
= &el
->l_recs
[i
];
1614 ocfs2_calc_trunc_pos(inode
, el
, rec
, trunc_start
, &trunc_cpos
,
1615 &trunc_len
, &trunc_end
, &blkno
, &done
);
1619 flags
= rec
->e_flags
;
1620 phys_cpos
= ocfs2_blocks_to_clusters(inode
->i_sb
, blkno
);
1622 ret
= ocfs2_remove_btree_range(inode
, &et
, trunc_cpos
,
1623 phys_cpos
, trunc_len
, flags
,
1624 &dealloc
, refcount_loc
);
1630 cluster_in_el
= trunc_end
;
1632 ocfs2_reinit_path(path
, 1);
1635 ocfs2_truncate_cluster_pages(inode
, byte_start
, byte_len
);
1638 ocfs2_schedule_truncate_log_flush(osb
, 1);
1639 ocfs2_run_deallocs(osb
, &dealloc
);
1645 * Parts of this function taken from xfs_change_file_space()
1647 static int __ocfs2_change_file_space(struct file
*file
, struct inode
*inode
,
1648 loff_t f_pos
, unsigned int cmd
,
1649 struct ocfs2_space_resv
*sr
,
1655 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1656 struct buffer_head
*di_bh
= NULL
;
1658 unsigned long long max_off
= inode
->i_sb
->s_maxbytes
;
1660 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
))
1663 mutex_lock(&inode
->i_mutex
);
1666 * This prevents concurrent writes on other nodes
1668 ret
= ocfs2_rw_lock(inode
, 1);
1674 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1680 if (inode
->i_flags
& (S_IMMUTABLE
|S_APPEND
)) {
1682 goto out_inode_unlock
;
1685 switch (sr
->l_whence
) {
1686 case 0: /*SEEK_SET*/
1688 case 1: /*SEEK_CUR*/
1689 sr
->l_start
+= f_pos
;
1691 case 2: /*SEEK_END*/
1692 sr
->l_start
+= i_size_read(inode
);
1696 goto out_inode_unlock
;
1700 llen
= sr
->l_len
> 0 ? sr
->l_len
- 1 : sr
->l_len
;
1703 || sr
->l_start
> max_off
1704 || (sr
->l_start
+ llen
) < 0
1705 || (sr
->l_start
+ llen
) > max_off
) {
1707 goto out_inode_unlock
;
1709 size
= sr
->l_start
+ sr
->l_len
;
1711 if (cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) {
1712 if (sr
->l_len
<= 0) {
1714 goto out_inode_unlock
;
1718 if (file
&& should_remove_suid(file
->f_path
.dentry
)) {
1719 ret
= __ocfs2_write_remove_suid(inode
, di_bh
);
1722 goto out_inode_unlock
;
1726 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1728 case OCFS2_IOC_RESVSP
:
1729 case OCFS2_IOC_RESVSP64
:
1731 * This takes unsigned offsets, but the signed ones we
1732 * pass have been checked against overflow above.
1734 ret
= ocfs2_allocate_unwritten_extents(inode
, sr
->l_start
,
1737 case OCFS2_IOC_UNRESVSP
:
1738 case OCFS2_IOC_UNRESVSP64
:
1739 ret
= ocfs2_remove_inode_range(inode
, di_bh
, sr
->l_start
,
1745 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
1748 goto out_inode_unlock
;
1752 * We update c/mtime for these changes
1754 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
1755 if (IS_ERR(handle
)) {
1756 ret
= PTR_ERR(handle
);
1758 goto out_inode_unlock
;
1761 if (change_size
&& i_size_read(inode
) < size
)
1762 i_size_write(inode
, size
);
1764 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1765 ret
= ocfs2_mark_inode_dirty(handle
, inode
, di_bh
);
1769 ocfs2_commit_trans(osb
, handle
);
1773 ocfs2_inode_unlock(inode
, 1);
1775 ocfs2_rw_unlock(inode
, 1);
1778 mutex_unlock(&inode
->i_mutex
);
1782 int ocfs2_change_file_space(struct file
*file
, unsigned int cmd
,
1783 struct ocfs2_space_resv
*sr
)
1785 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1786 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1788 if ((cmd
== OCFS2_IOC_RESVSP
|| cmd
== OCFS2_IOC_RESVSP64
) &&
1789 !ocfs2_writes_unwritten_extents(osb
))
1791 else if ((cmd
== OCFS2_IOC_UNRESVSP
|| cmd
== OCFS2_IOC_UNRESVSP64
) &&
1792 !ocfs2_sparse_alloc(osb
))
1795 if (!S_ISREG(inode
->i_mode
))
1798 if (!(file
->f_mode
& FMODE_WRITE
))
1801 return __ocfs2_change_file_space(file
, inode
, file
->f_pos
, cmd
, sr
, 0);
1804 static long ocfs2_fallocate(struct inode
*inode
, int mode
, loff_t offset
,
1807 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1808 struct ocfs2_space_resv sr
;
1809 int change_size
= 1;
1811 if (!ocfs2_writes_unwritten_extents(osb
))
1814 if (S_ISDIR(inode
->i_mode
))
1817 if (mode
& FALLOC_FL_KEEP_SIZE
)
1821 sr
.l_start
= (s64
)offset
;
1822 sr
.l_len
= (s64
)len
;
1824 return __ocfs2_change_file_space(NULL
, inode
, offset
,
1825 OCFS2_IOC_RESVSP64
, &sr
, change_size
);
1828 int ocfs2_check_range_for_refcount(struct inode
*inode
, loff_t pos
,
1832 unsigned int extent_flags
;
1833 u32 cpos
, clusters
, extent_len
, phys_cpos
;
1834 struct super_block
*sb
= inode
->i_sb
;
1836 if (!ocfs2_refcount_tree(OCFS2_SB(inode
->i_sb
)) ||
1837 !(OCFS2_I(inode
)->ip_dyn_features
& OCFS2_HAS_REFCOUNT_FL
) ||
1838 OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
)
1841 cpos
= pos
>> OCFS2_SB(sb
)->s_clustersize_bits
;
1842 clusters
= ocfs2_clusters_for_bytes(sb
, pos
+ count
) - cpos
;
1845 ret
= ocfs2_get_clusters(inode
, cpos
, &phys_cpos
, &extent_len
,
1852 if (phys_cpos
&& (extent_flags
& OCFS2_EXT_REFCOUNTED
)) {
1857 if (extent_len
> clusters
)
1858 extent_len
= clusters
;
1860 clusters
-= extent_len
;
1867 static int ocfs2_prepare_inode_for_refcount(struct inode
*inode
,
1868 loff_t pos
, size_t count
,
1872 struct buffer_head
*di_bh
= NULL
;
1873 u32 cpos
= pos
>> OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
1875 ocfs2_clusters_for_bytes(inode
->i_sb
, pos
+ count
) - cpos
;
1877 ret
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1885 ret
= ocfs2_refcount_cow(inode
, di_bh
, cpos
, clusters
, UINT_MAX
);
1893 static int ocfs2_prepare_inode_for_write(struct dentry
*dentry
,
1900 int ret
= 0, meta_level
= 0;
1901 struct inode
*inode
= dentry
->d_inode
;
1902 loff_t saved_pos
, end
;
1905 * We start with a read level meta lock and only jump to an ex
1906 * if we need to make modifications here.
1909 ret
= ocfs2_inode_lock(inode
, NULL
, meta_level
);
1916 /* Clear suid / sgid if necessary. We do this here
1917 * instead of later in the write path because
1918 * remove_suid() calls ->setattr without any hint that
1919 * we may have already done our cluster locking. Since
1920 * ocfs2_setattr() *must* take cluster locks to
1921 * proceeed, this will lead us to recursively lock the
1922 * inode. There's also the dinode i_size state which
1923 * can be lost via setattr during extending writes (we
1924 * set inode->i_size at the end of a write. */
1925 if (should_remove_suid(dentry
)) {
1926 if (meta_level
== 0) {
1927 ocfs2_inode_unlock(inode
, meta_level
);
1932 ret
= ocfs2_write_remove_suid(inode
);
1939 /* work on a copy of ppos until we're sure that we won't have
1940 * to recalculate it due to relocking. */
1942 saved_pos
= i_size_read(inode
);
1943 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1948 end
= saved_pos
+ count
;
1950 ret
= ocfs2_check_range_for_refcount(inode
, saved_pos
, count
);
1952 ocfs2_inode_unlock(inode
, meta_level
);
1955 ret
= ocfs2_prepare_inode_for_refcount(inode
,
1971 * Skip the O_DIRECT checks if we don't need
1974 if (!direct_io
|| !(*direct_io
))
1978 * There's no sane way to do direct writes to an inode
1981 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
1987 * Allowing concurrent direct writes means
1988 * i_size changes wouldn't be synchronized, so
1989 * one node could wind up truncating another
1992 if (end
> i_size_read(inode
)) {
1998 * We don't fill holes during direct io, so
1999 * check for them here. If any are found, the
2000 * caller will have to retake some cluster
2001 * locks and initiate the io as buffered.
2003 ret
= ocfs2_check_range_for_holes(inode
, saved_pos
, count
);
2016 if (meta_level
>= 0)
2017 ocfs2_inode_unlock(inode
, meta_level
);
2023 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
2024 const struct iovec
*iov
,
2025 unsigned long nr_segs
,
2028 int ret
, direct_io
, appending
, rw_level
, have_alloc_sem
= 0;
2029 int can_do_direct
, has_refcount
= 0;
2030 ssize_t written
= 0;
2031 size_t ocount
; /* original count */
2032 size_t count
; /* after file limit checks */
2033 loff_t old_size
, *ppos
= &iocb
->ki_pos
;
2035 struct file
*file
= iocb
->ki_filp
;
2036 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2037 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2039 mlog_entry("(0x%p, %u, '%.*s')\n", file
,
2040 (unsigned int)nr_segs
,
2041 file
->f_path
.dentry
->d_name
.len
,
2042 file
->f_path
.dentry
->d_name
.name
);
2044 if (iocb
->ki_left
== 0)
2047 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
2049 appending
= file
->f_flags
& O_APPEND
? 1 : 0;
2050 direct_io
= file
->f_flags
& O_DIRECT
? 1 : 0;
2052 mutex_lock(&inode
->i_mutex
);
2055 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
2057 down_read(&inode
->i_alloc_sem
);
2061 /* concurrent O_DIRECT writes are allowed */
2062 rw_level
= !direct_io
;
2063 ret
= ocfs2_rw_lock(inode
, rw_level
);
2069 can_do_direct
= direct_io
;
2070 ret
= ocfs2_prepare_inode_for_write(file
->f_path
.dentry
, ppos
,
2071 iocb
->ki_left
, appending
,
2072 &can_do_direct
, &has_refcount
);
2079 * We can't complete the direct I/O as requested, fall back to
2082 if (direct_io
&& !can_do_direct
) {
2083 ocfs2_rw_unlock(inode
, rw_level
);
2084 up_read(&inode
->i_alloc_sem
);
2094 * To later detect whether a journal commit for sync writes is
2095 * necessary, we sample i_size, and cluster count here.
2097 old_size
= i_size_read(inode
);
2098 old_clusters
= OCFS2_I(inode
)->ip_clusters
;
2100 /* communicate with ocfs2_dio_end_io */
2101 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2103 ret
= generic_segment_checks(iov
, &nr_segs
, &ocount
,
2109 ret
= generic_write_checks(file
, ppos
, &count
,
2110 S_ISBLK(inode
->i_mode
));
2115 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, *ppos
,
2116 ppos
, count
, ocount
);
2119 * direct write may have instantiated a few
2120 * blocks outside i_size. Trim these off again.
2121 * Don't need i_size_read because we hold i_mutex.
2123 if (*ppos
+ count
> inode
->i_size
)
2124 vmtruncate(inode
, inode
->i_size
);
2129 current
->backing_dev_info
= file
->f_mapping
->backing_dev_info
;
2130 written
= generic_file_buffered_write(iocb
, iov
, nr_segs
, *ppos
,
2132 current
->backing_dev_info
= NULL
;
2136 /* buffered aio wouldn't have proper lock coverage today */
2137 BUG_ON(ret
== -EIOCBQUEUED
&& !(file
->f_flags
& O_DIRECT
));
2139 if (((file
->f_flags
& O_DSYNC
) && !direct_io
) || IS_SYNC(inode
) ||
2140 ((file
->f_flags
& O_DIRECT
) && has_refcount
)) {
2141 ret
= filemap_fdatawrite_range(file
->f_mapping
, pos
,
2146 if (!ret
&& ((old_size
!= i_size_read(inode
)) ||
2147 (old_clusters
!= OCFS2_I(inode
)->ip_clusters
) ||
2149 ret
= jbd2_journal_force_commit(osb
->journal
->j_journal
);
2155 ret
= filemap_fdatawait_range(file
->f_mapping
, pos
,
2160 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2161 * function pointer which is called when o_direct io completes so that
2162 * it can unlock our rw lock. (it's the clustered equivalent of
2163 * i_alloc_sem; protects truncate from racing with pending ios).
2164 * Unfortunately there are error cases which call end_io and others
2165 * that don't. so we don't have to unlock the rw_lock if either an
2166 * async dio is going to do it in the future or an end_io after an
2167 * error has already done it.
2169 if ((ret
== -EIOCBQUEUED
) || (!ocfs2_iocb_is_rw_locked(iocb
))) {
2176 ocfs2_rw_unlock(inode
, rw_level
);
2180 up_read(&inode
->i_alloc_sem
);
2182 mutex_unlock(&inode
->i_mutex
);
2190 static int ocfs2_splice_to_file(struct pipe_inode_info
*pipe
,
2192 struct splice_desc
*sd
)
2196 ret
= ocfs2_prepare_inode_for_write(out
->f_path
.dentry
, &sd
->pos
,
2197 sd
->total_len
, 0, NULL
, NULL
);
2203 return splice_from_pipe_feed(pipe
, sd
, pipe_to_file
);
2206 static ssize_t
ocfs2_file_splice_write(struct pipe_inode_info
*pipe
,
2213 struct address_space
*mapping
= out
->f_mapping
;
2214 struct inode
*inode
= mapping
->host
;
2215 struct splice_desc sd
= {
2222 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out
, pipe
,
2224 out
->f_path
.dentry
->d_name
.len
,
2225 out
->f_path
.dentry
->d_name
.name
);
2228 mutex_lock_nested(&pipe
->inode
->i_mutex
, I_MUTEX_PARENT
);
2230 splice_from_pipe_begin(&sd
);
2232 ret
= splice_from_pipe_next(pipe
, &sd
);
2236 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2237 ret
= ocfs2_rw_lock(inode
, 1);
2241 ret
= ocfs2_splice_to_file(pipe
, out
, &sd
);
2242 ocfs2_rw_unlock(inode
, 1);
2244 mutex_unlock(&inode
->i_mutex
);
2246 splice_from_pipe_end(pipe
, &sd
);
2249 mutex_unlock(&pipe
->inode
->i_mutex
);
2252 ret
= sd
.num_spliced
;
2255 unsigned long nr_pages
;
2258 nr_pages
= (ret
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2260 err
= generic_write_sync(out
, *ppos
, ret
);
2266 balance_dirty_pages_ratelimited_nr(mapping
, nr_pages
);
2273 static ssize_t
ocfs2_file_splice_read(struct file
*in
,
2275 struct pipe_inode_info
*pipe
,
2279 int ret
= 0, lock_level
= 0;
2280 struct inode
*inode
= in
->f_path
.dentry
->d_inode
;
2282 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in
, pipe
,
2284 in
->f_path
.dentry
->d_name
.len
,
2285 in
->f_path
.dentry
->d_name
.name
);
2288 * See the comment in ocfs2_file_aio_read()
2290 ret
= ocfs2_inode_lock_atime(inode
, in
->f_vfsmnt
, &lock_level
);
2295 ocfs2_inode_unlock(inode
, lock_level
);
2297 ret
= generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
2304 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
2305 const struct iovec
*iov
,
2306 unsigned long nr_segs
,
2309 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0, lock_level
= 0;
2310 struct file
*filp
= iocb
->ki_filp
;
2311 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
2313 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
2314 (unsigned int)nr_segs
,
2315 filp
->f_path
.dentry
->d_name
.len
,
2316 filp
->f_path
.dentry
->d_name
.name
);
2325 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2326 * need locks to protect pending reads from racing with truncate.
2328 if (filp
->f_flags
& O_DIRECT
) {
2329 down_read(&inode
->i_alloc_sem
);
2332 ret
= ocfs2_rw_lock(inode
, 0);
2338 /* communicate with ocfs2_dio_end_io */
2339 ocfs2_iocb_set_rw_locked(iocb
, rw_level
);
2343 * We're fine letting folks race truncates and extending
2344 * writes with read across the cluster, just like they can
2345 * locally. Hence no rw_lock during read.
2347 * Take and drop the meta data lock to update inode fields
2348 * like i_size. This allows the checks down below
2349 * generic_file_aio_read() a chance of actually working.
2351 ret
= ocfs2_inode_lock_atime(inode
, filp
->f_vfsmnt
, &lock_level
);
2356 ocfs2_inode_unlock(inode
, lock_level
);
2358 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
2360 mlog(0, "generic_file_aio_read returned -EINVAL\n");
2362 /* buffered aio wouldn't have proper lock coverage today */
2363 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
2365 /* see ocfs2_file_aio_write */
2366 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
2373 up_read(&inode
->i_alloc_sem
);
2375 ocfs2_rw_unlock(inode
, rw_level
);
2381 const struct inode_operations ocfs2_file_iops
= {
2382 .setattr
= ocfs2_setattr
,
2383 .getattr
= ocfs2_getattr
,
2384 .permission
= ocfs2_permission
,
2385 .setxattr
= generic_setxattr
,
2386 .getxattr
= generic_getxattr
,
2387 .listxattr
= ocfs2_listxattr
,
2388 .removexattr
= generic_removexattr
,
2389 .fallocate
= ocfs2_fallocate
,
2390 .fiemap
= ocfs2_fiemap
,
2393 const struct inode_operations ocfs2_special_file_iops
= {
2394 .setattr
= ocfs2_setattr
,
2395 .getattr
= ocfs2_getattr
,
2396 .permission
= ocfs2_permission
,
2400 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2401 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2403 const struct file_operations ocfs2_fops
= {
2404 .llseek
= generic_file_llseek
,
2405 .read
= do_sync_read
,
2406 .write
= do_sync_write
,
2408 .fsync
= ocfs2_sync_file
,
2409 .release
= ocfs2_file_release
,
2410 .open
= ocfs2_file_open
,
2411 .aio_read
= ocfs2_file_aio_read
,
2412 .aio_write
= ocfs2_file_aio_write
,
2413 .unlocked_ioctl
= ocfs2_ioctl
,
2414 #ifdef CONFIG_COMPAT
2415 .compat_ioctl
= ocfs2_compat_ioctl
,
2418 .flock
= ocfs2_flock
,
2419 .splice_read
= ocfs2_file_splice_read
,
2420 .splice_write
= ocfs2_file_splice_write
,
2423 const struct file_operations ocfs2_dops
= {
2424 .llseek
= generic_file_llseek
,
2425 .read
= generic_read_dir
,
2426 .readdir
= ocfs2_readdir
,
2427 .fsync
= ocfs2_sync_file
,
2428 .release
= ocfs2_dir_release
,
2429 .open
= ocfs2_dir_open
,
2430 .unlocked_ioctl
= ocfs2_ioctl
,
2431 #ifdef CONFIG_COMPAT
2432 .compat_ioctl
= ocfs2_compat_ioctl
,
2435 .flock
= ocfs2_flock
,
2439 * POSIX-lockless variants of our file_operations.
2441 * These will be used if the underlying cluster stack does not support
2442 * posix file locking, if the user passes the "localflocks" mount
2443 * option, or if we have a local-only fs.
2445 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2446 * so we still want it in the case of no stack support for
2447 * plocks. Internally, it will do the right thing when asked to ignore
2450 const struct file_operations ocfs2_fops_no_plocks
= {
2451 .llseek
= generic_file_llseek
,
2452 .read
= do_sync_read
,
2453 .write
= do_sync_write
,
2455 .fsync
= ocfs2_sync_file
,
2456 .release
= ocfs2_file_release
,
2457 .open
= ocfs2_file_open
,
2458 .aio_read
= ocfs2_file_aio_read
,
2459 .aio_write
= ocfs2_file_aio_write
,
2460 .unlocked_ioctl
= ocfs2_ioctl
,
2461 #ifdef CONFIG_COMPAT
2462 .compat_ioctl
= ocfs2_compat_ioctl
,
2464 .flock
= ocfs2_flock
,
2465 .splice_read
= ocfs2_file_splice_read
,
2466 .splice_write
= ocfs2_file_splice_write
,
2469 const struct file_operations ocfs2_dops_no_plocks
= {
2470 .llseek
= generic_file_llseek
,
2471 .read
= generic_read_dir
,
2472 .readdir
= ocfs2_readdir
,
2473 .fsync
= ocfs2_sync_file
,
2474 .release
= ocfs2_dir_release
,
2475 .open
= ocfs2_dir_open
,
2476 .unlocked_ioctl
= ocfs2_ioctl
,
2477 #ifdef CONFIG_COMPAT
2478 .compat_ioctl
= ocfs2_compat_ioctl
,
2480 .flock
= ocfs2_flock
,