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>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
44 #include "extent_map.h"
54 #include "buffer_head_io.h"
56 static int ocfs2_sync_inode(struct inode
*inode
)
58 filemap_fdatawrite(inode
->i_mapping
);
59 return sync_mapping_buffers(inode
->i_mapping
);
62 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
65 int mode
= file
->f_flags
;
66 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
68 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
69 file
->f_dentry
->d_name
.len
, file
->f_dentry
->d_name
.name
);
71 spin_lock(&oi
->ip_lock
);
73 /* Check that the inode hasn't been wiped from disk by another
74 * node. If it hasn't then we're safe as long as we hold the
75 * spin lock until our increment of open count. */
76 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
77 spin_unlock(&oi
->ip_lock
);
84 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
87 spin_unlock(&oi
->ip_lock
);
94 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
96 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
98 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
99 file
->f_dentry
->d_name
.len
,
100 file
->f_dentry
->d_name
.name
);
102 spin_lock(&oi
->ip_lock
);
103 if (!--oi
->ip_open_count
)
104 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
105 spin_unlock(&oi
->ip_lock
);
112 static int ocfs2_sync_file(struct file
*file
,
113 struct dentry
*dentry
,
118 struct inode
*inode
= dentry
->d_inode
;
119 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
121 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
122 dentry
->d_name
.len
, dentry
->d_name
.name
);
124 err
= ocfs2_sync_inode(dentry
->d_inode
);
128 journal
= osb
->journal
->j_journal
;
129 err
= journal_force_commit(journal
);
134 return (err
< 0) ? -EIO
: 0;
137 int ocfs2_set_inode_size(struct ocfs2_journal_handle
*handle
,
139 struct buffer_head
*fe_bh
,
145 i_size_write(inode
, new_i_size
);
146 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(new_i_size
);
147 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
149 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
160 static int ocfs2_simple_size_update(struct inode
*inode
,
161 struct buffer_head
*di_bh
,
165 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
166 struct ocfs2_journal_handle
*handle
= NULL
;
168 handle
= ocfs2_start_trans(osb
, NULL
,
169 OCFS2_INODE_UPDATE_CREDITS
);
170 if (handle
== NULL
) {
176 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
181 ocfs2_commit_trans(handle
);
186 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
188 struct buffer_head
*fe_bh
,
192 struct ocfs2_journal_handle
*handle
;
196 /* TODO: This needs to actually orphan the inode in this
199 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
200 if (IS_ERR(handle
)) {
201 status
= PTR_ERR(handle
);
206 status
= ocfs2_set_inode_size(handle
, inode
, fe_bh
, new_i_size
);
210 ocfs2_commit_trans(handle
);
216 static int ocfs2_truncate_file(struct inode
*inode
,
217 struct buffer_head
*di_bh
,
221 struct ocfs2_dinode
*fe
= NULL
;
222 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
223 struct ocfs2_truncate_context
*tc
= NULL
;
225 mlog_entry("(inode = %llu, new_i_size = %llu\n",
226 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
227 (unsigned long long)new_i_size
);
229 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
231 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
232 if (!OCFS2_IS_VALID_DINODE(fe
)) {
233 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
238 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
239 "Inode %llu, inode i_size = %lld != di "
240 "i_size = %llu, i_flags = 0x%x\n",
241 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
243 (unsigned long long)le64_to_cpu(fe
->i_size
),
244 le32_to_cpu(fe
->i_flags
));
246 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
247 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
248 (unsigned long long)le64_to_cpu(fe
->i_size
),
249 (unsigned long long)new_i_size
);
255 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
256 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
257 (unsigned long long)le64_to_cpu(fe
->i_size
),
258 (unsigned long long)new_i_size
);
260 /* lets handle the simple truncate cases before doing any more
261 * cluster locking. */
262 if (new_i_size
== le64_to_cpu(fe
->i_size
))
265 /* This forces other nodes to sync and drop their pages. Do
266 * this even if we have a truncate without allocation change -
267 * ocfs2 cluster sizes can be much greater than page size, so
268 * we have to truncate them anyway. */
269 status
= ocfs2_data_lock(inode
, 1);
274 ocfs2_data_unlock(inode
, 1);
276 if (le32_to_cpu(fe
->i_clusters
) ==
277 ocfs2_clusters_for_bytes(osb
->sb
, new_i_size
)) {
278 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
280 /* No allocation change is required, so lets fast path
282 status
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
288 /* alright, we're going to need to do a full blown alloc size
289 * change. Orphan the inode so that recovery can complete the
290 * truncate if necessary. This does the task of marking
292 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
298 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
304 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
310 /* TODO: orphan dir cleanup here. */
318 * extend allocation only here.
319 * we'll update all the disk stuff, and oip->alloc_size
321 * expect stuff to be locked, a transaction started and enough data /
322 * metadata reservations in the contexts.
324 * Will return -EAGAIN, and a reason if a restart is needed.
325 * If passed in, *reason will always be set, even in error.
327 int ocfs2_do_extend_allocation(struct ocfs2_super
*osb
,
330 struct buffer_head
*fe_bh
,
331 struct ocfs2_journal_handle
*handle
,
332 struct ocfs2_alloc_context
*data_ac
,
333 struct ocfs2_alloc_context
*meta_ac
,
334 enum ocfs2_alloc_restarted
*reason_ret
)
338 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
339 enum ocfs2_alloc_restarted reason
= RESTART_NONE
;
340 u32 bit_off
, num_bits
;
343 BUG_ON(!clusters_to_add
);
345 free_extents
= ocfs2_num_free_extents(osb
, inode
, fe
);
346 if (free_extents
< 0) {
347 status
= free_extents
;
352 /* there are two cases which could cause us to EAGAIN in the
353 * we-need-more-metadata case:
354 * 1) we haven't reserved *any*
355 * 2) we are so fragmented, we've needed to add metadata too
357 if (!free_extents
&& !meta_ac
) {
358 mlog(0, "we haven't reserved any metadata!\n");
360 reason
= RESTART_META
;
362 } else if ((!free_extents
)
363 && (ocfs2_alloc_context_bits_left(meta_ac
)
364 < ocfs2_extend_meta_needed(fe
))) {
365 mlog(0, "filesystem is really fragmented...\n");
367 reason
= RESTART_META
;
371 status
= ocfs2_claim_clusters(osb
, handle
, data_ac
, 1,
372 &bit_off
, &num_bits
);
374 if (status
!= -ENOSPC
)
379 BUG_ON(num_bits
> clusters_to_add
);
381 /* reserve our write early -- insert_extent may update the inode */
382 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
383 OCFS2_JOURNAL_ACCESS_WRITE
);
389 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
390 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
391 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
392 status
= ocfs2_insert_extent(osb
, handle
, inode
, fe_bh
, block
,
399 le32_add_cpu(&fe
->i_clusters
, num_bits
);
400 spin_lock(&OCFS2_I(inode
)->ip_lock
);
401 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
402 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
404 status
= ocfs2_journal_dirty(handle
, fe_bh
);
410 clusters_to_add
-= num_bits
;
412 if (clusters_to_add
) {
413 mlog(0, "need to alloc once more, clusters = %u, wanted = "
414 "%u\n", fe
->i_clusters
, clusters_to_add
);
416 reason
= RESTART_TRANS
;
422 *reason_ret
= reason
;
426 static int ocfs2_extend_allocation(struct inode
*inode
,
430 int restart_func
= 0;
431 int drop_alloc_sem
= 0;
432 int credits
, num_free_extents
;
434 struct buffer_head
*bh
= NULL
;
435 struct ocfs2_dinode
*fe
= NULL
;
436 struct ocfs2_journal_handle
*handle
= NULL
;
437 struct ocfs2_alloc_context
*data_ac
= NULL
;
438 struct ocfs2_alloc_context
*meta_ac
= NULL
;
439 enum ocfs2_alloc_restarted why
;
440 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
442 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
444 status
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &bh
,
445 OCFS2_BH_CACHED
, inode
);
451 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
452 if (!OCFS2_IS_VALID_DINODE(fe
)) {
453 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
459 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
461 mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
462 "clusters_to_add = %u\n",
463 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, i_size_read(inode
),
464 fe
->i_clusters
, clusters_to_add
);
466 handle
= ocfs2_alloc_handle(osb
);
467 if (handle
== NULL
) {
473 num_free_extents
= ocfs2_num_free_extents(osb
,
476 if (num_free_extents
< 0) {
477 status
= num_free_extents
;
482 if (!num_free_extents
) {
483 status
= ocfs2_reserve_new_metadata(osb
,
488 if (status
!= -ENOSPC
)
494 status
= ocfs2_reserve_clusters(osb
,
499 if (status
!= -ENOSPC
)
504 /* blocks peope in read/write from reading our allocation
505 * until we're done changing it. We depend on i_mutex to block
506 * other extend/truncate calls while we're here. Ordering wrt
507 * start_trans is important here -- always do it before! */
508 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
511 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
512 handle
= ocfs2_start_trans(osb
, handle
, credits
);
513 if (IS_ERR(handle
)) {
514 status
= PTR_ERR(handle
);
520 restarted_transaction
:
521 /* reserve a write to the file entry early on - that we if we
522 * run out of credits in the allocation path, we can still
524 status
= ocfs2_journal_access(handle
, inode
, bh
,
525 OCFS2_JOURNAL_ACCESS_WRITE
);
531 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
533 status
= ocfs2_do_extend_allocation(osb
,
541 if ((status
< 0) && (status
!= -EAGAIN
)) {
542 if (status
!= -ENOSPC
)
547 status
= ocfs2_journal_dirty(handle
, bh
);
553 spin_lock(&OCFS2_I(inode
)->ip_lock
);
554 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
555 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
557 if (why
!= RESTART_NONE
&& clusters_to_add
) {
558 if (why
== RESTART_META
) {
559 mlog(0, "restarting function.\n");
562 BUG_ON(why
!= RESTART_TRANS
);
564 mlog(0, "restarting transaction.\n");
565 /* TODO: This can be more intelligent. */
566 credits
= ocfs2_calc_extend_credits(osb
->sb
,
569 status
= ocfs2_extend_trans(handle
, credits
);
571 /* handle still has to be committed at
577 goto restarted_transaction
;
581 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
582 fe
->i_clusters
, (unsigned long long)fe
->i_size
);
583 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
584 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
587 if (drop_alloc_sem
) {
588 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
592 ocfs2_commit_trans(handle
);
596 ocfs2_free_alloc_context(data_ac
);
600 ocfs2_free_alloc_context(meta_ac
);
603 if ((!status
) && restart_func
) {
616 /* Some parts of this taken from generic_cont_expand, which turned out
617 * to be too fragile to do exactly what we need without us having to
618 * worry about recursive locking in ->prepare_write() and
619 * ->commit_write(). */
620 static int ocfs2_write_zero_page(struct inode
*inode
,
623 struct address_space
*mapping
= inode
->i_mapping
;
627 struct ocfs2_journal_handle
*handle
= NULL
;
630 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
631 /* ugh. in prepare/commit_write, if from==to==start of block, we
632 ** skip the prepare. make sure we never send an offset for the start
635 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
638 index
= size
>> PAGE_CACHE_SHIFT
;
640 page
= grab_cache_page(mapping
, index
);
647 ret
= ocfs2_prepare_write_nolock(inode
, page
, offset
, offset
);
653 if (ocfs2_should_order_data(inode
)) {
654 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
656 if (IS_ERR(handle
)) {
657 ret
= PTR_ERR(handle
);
663 /* must not update i_size! */
664 ret
= block_commit_write(page
, offset
, offset
);
671 ocfs2_commit_trans(handle
);
674 page_cache_release(page
);
679 static int ocfs2_zero_extend(struct inode
*inode
,
684 struct super_block
*sb
= inode
->i_sb
;
686 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
687 while (start_off
< zero_to_size
) {
688 ret
= ocfs2_write_zero_page(inode
, start_off
);
694 start_off
+= sb
->s_blocksize
;
697 * Very large extends have the potential to lock up
698 * the cpu for extended periods of time.
708 * A tail_to_skip value > 0 indicates that we're being called from
709 * ocfs2_file_aio_write(). This has the following implications:
711 * - we don't want to update i_size
712 * - di_bh will be NULL, which is fine because it's only used in the
713 * case where we want to update i_size.
714 * - ocfs2_zero_extend() will then only be filling the hole created
715 * between i_size and the start of the write.
717 static int ocfs2_extend_file(struct inode
*inode
,
718 struct buffer_head
*di_bh
,
725 BUG_ON(!tail_to_skip
&& !di_bh
);
727 /* setattr sometimes calls us like this. */
731 if (i_size_read(inode
) == new_i_size
)
733 BUG_ON(new_i_size
< i_size_read(inode
));
735 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
736 OCFS2_I(inode
)->ip_clusters
;
739 * protect the pages that ocfs2_zero_extend is going to be
740 * pulling into the page cache.. we do this before the
741 * metadata extend so that we don't get into the situation
742 * where we've extended the metadata but can't get the data
745 ret
= ocfs2_data_lock(inode
, 1);
751 if (clusters_to_add
) {
752 ret
= ocfs2_extend_allocation(inode
, clusters_to_add
);
760 * Call this even if we don't add any clusters to the tree. We
761 * still need to zero the area between the old i_size and the
764 ret
= ocfs2_zero_extend(inode
, (u64
)new_i_size
- tail_to_skip
);
771 /* We're being called from ocfs2_setattr() which wants
772 * us to update i_size */
773 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
779 ocfs2_data_unlock(inode
, 1);
785 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
787 int status
= 0, size_change
;
788 struct inode
*inode
= dentry
->d_inode
;
789 struct super_block
*sb
= inode
->i_sb
;
790 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
791 struct buffer_head
*bh
= NULL
;
792 struct ocfs2_journal_handle
*handle
= NULL
;
794 mlog_entry("(0x%p, '%.*s')\n", dentry
,
795 dentry
->d_name
.len
, dentry
->d_name
.name
);
797 if (attr
->ia_valid
& ATTR_MODE
)
798 mlog(0, "mode change: %d\n", attr
->ia_mode
);
799 if (attr
->ia_valid
& ATTR_UID
)
800 mlog(0, "uid change: %d\n", attr
->ia_uid
);
801 if (attr
->ia_valid
& ATTR_GID
)
802 mlog(0, "gid change: %d\n", attr
->ia_gid
);
803 if (attr
->ia_valid
& ATTR_SIZE
)
804 mlog(0, "size change...\n");
805 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
806 mlog(0, "time change...\n");
808 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
809 | ATTR_GID | ATTR_UID | ATTR_MODE)
810 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
811 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
815 status
= inode_change_ok(inode
, attr
);
819 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
821 status
= ocfs2_rw_lock(inode
, 1);
828 status
= ocfs2_meta_lock(inode
, NULL
, &bh
, 1);
830 if (status
!= -ENOENT
)
835 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
836 if (i_size_read(inode
) > attr
->ia_size
)
837 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
839 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
, 0);
841 if (status
!= -ENOSPC
)
848 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
849 if (IS_ERR(handle
)) {
850 status
= PTR_ERR(handle
);
855 status
= inode_setattr(inode
, attr
);
861 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
866 ocfs2_commit_trans(handle
);
868 ocfs2_meta_unlock(inode
, 1);
871 ocfs2_rw_unlock(inode
, 1);
880 int ocfs2_getattr(struct vfsmount
*mnt
,
881 struct dentry
*dentry
,
884 struct inode
*inode
= dentry
->d_inode
;
885 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
886 struct ocfs2_super
*osb
= sb
->s_fs_info
;
891 err
= ocfs2_inode_revalidate(dentry
);
898 generic_fillattr(inode
, stat
);
900 /* We set the blksize from the cluster size for performance */
901 stat
->blksize
= osb
->s_clustersize
;
909 static int ocfs2_write_remove_suid(struct inode
*inode
)
912 struct buffer_head
*bh
= NULL
;
913 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
914 struct ocfs2_journal_handle
*handle
;
915 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
916 struct ocfs2_dinode
*di
;
918 mlog_entry("(Inode %llu, mode 0%o)\n",
919 (unsigned long long)oi
->ip_blkno
, inode
->i_mode
);
921 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
922 if (handle
== NULL
) {
928 ret
= ocfs2_read_block(osb
, oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
934 ret
= ocfs2_journal_access(handle
, inode
, bh
,
935 OCFS2_JOURNAL_ACCESS_WRITE
);
941 inode
->i_mode
&= ~S_ISUID
;
942 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
943 inode
->i_mode
&= ~S_ISGID
;
945 di
= (struct ocfs2_dinode
*) bh
->b_data
;
946 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
948 ret
= ocfs2_journal_dirty(handle
, bh
);
954 ocfs2_commit_trans(handle
);
960 static inline int ocfs2_write_should_remove_suid(struct inode
*inode
)
962 mode_t mode
= inode
->i_mode
;
964 if (!capable(CAP_FSETID
)) {
965 if (unlikely(mode
& S_ISUID
))
968 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
974 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
975 const struct iovec
*iov
,
976 unsigned long nr_segs
,
979 int ret
, rw_level
= -1, meta_level
= -1, have_alloc_sem
= 0;
981 struct file
*filp
= iocb
->ki_filp
;
982 struct inode
*inode
= filp
->f_dentry
->d_inode
;
983 loff_t newsize
, saved_pos
;
985 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
986 (unsigned int)nr_segs
,
987 filp
->f_dentry
->d_name
.len
,
988 filp
->f_dentry
->d_name
.name
);
990 /* happy write of zero bytes */
991 if (iocb
->ki_left
== 0)
995 mlog(0, "bad inode\n");
999 mutex_lock(&inode
->i_mutex
);
1000 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1001 if (filp
->f_flags
& O_DIRECT
) {
1003 down_read(&inode
->i_alloc_sem
);
1006 /* concurrent O_DIRECT writes are allowed */
1007 rw_level
= (filp
->f_flags
& O_DIRECT
) ? 0 : 1;
1008 ret
= ocfs2_rw_lock(inode
, rw_level
);
1016 * We sample i_size under a read level meta lock to see if our write
1017 * is extending the file, if it is we back off and get a write level
1020 meta_level
= (filp
->f_flags
& O_APPEND
) ? 1 : 0;
1022 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, meta_level
);
1029 /* Clear suid / sgid if necessary. We do this here
1030 * instead of later in the write path because
1031 * remove_suid() calls ->setattr without any hint that
1032 * we may have already done our cluster locking. Since
1033 * ocfs2_setattr() *must* take cluster locks to
1034 * proceeed, this will lead us to recursively lock the
1035 * inode. There's also the dinode i_size state which
1036 * can be lost via setattr during extending writes (we
1037 * set inode->i_size at the end of a write. */
1038 if (ocfs2_write_should_remove_suid(inode
)) {
1039 if (meta_level
== 0) {
1040 ocfs2_meta_unlock(inode
, meta_level
);
1045 ret
= ocfs2_write_remove_suid(inode
);
1052 /* work on a copy of ppos until we're sure that we won't have
1053 * to recalculate it due to relocking. */
1054 if (filp
->f_flags
& O_APPEND
) {
1055 saved_pos
= i_size_read(inode
);
1056 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1058 saved_pos
= iocb
->ki_pos
;
1060 newsize
= iocb
->ki_left
+ saved_pos
;
1062 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1063 (long long) saved_pos
, (long long) newsize
,
1064 (long long) i_size_read(inode
));
1066 /* No need for a higher level metadata lock if we're
1067 * never going past i_size. */
1068 if (newsize
<= i_size_read(inode
))
1071 if (meta_level
== 0) {
1072 ocfs2_meta_unlock(inode
, meta_level
);
1077 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1078 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1079 OCFS2_I(inode
)->ip_clusters
;
1080 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1082 mlog(0, "Writing at EOF, may need more allocation: "
1083 "i_size = %lld, newsize = %lld, need %u clusters\n",
1084 (long long) i_size_read(inode
), (long long) newsize
,
1087 /* We only want to continue the rest of this loop if
1088 * our extend will actually require more
1093 ret
= ocfs2_extend_file(inode
, NULL
, newsize
, iocb
->ki_left
);
1102 /* ok, we're done with i_size and alloc work */
1103 iocb
->ki_pos
= saved_pos
;
1104 ocfs2_meta_unlock(inode
, meta_level
);
1107 /* communicate with ocfs2_dio_end_io */
1108 ocfs2_iocb_set_rw_locked(iocb
);
1110 ret
= generic_file_aio_write_nolock(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
1112 /* buffered aio wouldn't have proper lock coverage today */
1113 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1116 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1117 * function pointer which is called when o_direct io completes so that
1118 * it can unlock our rw lock. (it's the clustered equivalent of
1119 * i_alloc_sem; protects truncate from racing with pending ios).
1120 * Unfortunately there are error cases which call end_io and others
1121 * that don't. so we don't have to unlock the rw_lock if either an
1122 * async dio is going to do it in the future or an end_io after an
1123 * error has already done it.
1125 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1131 if (meta_level
!= -1)
1132 ocfs2_meta_unlock(inode
, meta_level
);
1134 up_read(&inode
->i_alloc_sem
);
1136 ocfs2_rw_unlock(inode
, rw_level
);
1137 mutex_unlock(&inode
->i_mutex
);
1143 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
1144 const struct iovec
*iov
,
1145 unsigned long nr_segs
,
1148 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0;
1149 struct file
*filp
= iocb
->ki_filp
;
1150 struct inode
*inode
= filp
->f_dentry
->d_inode
;
1152 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
1153 (unsigned int)nr_segs
,
1154 filp
->f_dentry
->d_name
.len
,
1155 filp
->f_dentry
->d_name
.name
);
1164 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1165 * need locks to protect pending reads from racing with truncate.
1167 if (filp
->f_flags
& O_DIRECT
) {
1168 down_read(&inode
->i_alloc_sem
);
1171 ret
= ocfs2_rw_lock(inode
, 0);
1177 /* communicate with ocfs2_dio_end_io */
1178 ocfs2_iocb_set_rw_locked(iocb
);
1182 * We're fine letting folks race truncates and extending
1183 * writes with read across the cluster, just like they can
1184 * locally. Hence no rw_lock during read.
1186 * Take and drop the meta data lock to update inode fields
1187 * like i_size. This allows the checks down below
1188 * generic_file_aio_read() a chance of actually working.
1190 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, 0);
1195 ocfs2_meta_unlock(inode
, 0);
1197 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
1199 mlog(ML_ERROR
, "generic_file_aio_read returned -EINVAL\n");
1201 /* buffered aio wouldn't have proper lock coverage today */
1202 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1204 /* see ocfs2_file_aio_write */
1205 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1212 up_read(&inode
->i_alloc_sem
);
1214 ocfs2_rw_unlock(inode
, rw_level
);
1220 struct inode_operations ocfs2_file_iops
= {
1221 .setattr
= ocfs2_setattr
,
1222 .getattr
= ocfs2_getattr
,
1225 struct inode_operations ocfs2_special_file_iops
= {
1226 .setattr
= ocfs2_setattr
,
1227 .getattr
= ocfs2_getattr
,
1230 const struct file_operations ocfs2_fops
= {
1231 .read
= do_sync_read
,
1232 .write
= do_sync_write
,
1233 .sendfile
= generic_file_sendfile
,
1235 .fsync
= ocfs2_sync_file
,
1236 .release
= ocfs2_file_release
,
1237 .open
= ocfs2_file_open
,
1238 .aio_read
= ocfs2_file_aio_read
,
1239 .aio_write
= ocfs2_file_aio_write
,
1240 .ioctl
= ocfs2_ioctl
,
1243 const struct file_operations ocfs2_dops
= {
1244 .read
= generic_read_dir
,
1245 .readdir
= ocfs2_readdir
,
1246 .fsync
= ocfs2_sync_file
,
1247 .ioctl
= ocfs2_ioctl
,