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>
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
43 #include "extent_map.h"
52 #include "buffer_head_io.h"
54 static int ocfs2_sync_inode(struct inode
*inode
)
56 filemap_fdatawrite(inode
->i_mapping
);
57 return sync_mapping_buffers(inode
->i_mapping
);
60 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
63 int mode
= file
->f_flags
;
64 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
66 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
67 file
->f_dentry
->d_name
.len
, file
->f_dentry
->d_name
.name
);
69 spin_lock(&oi
->ip_lock
);
71 /* Check that the inode hasn't been wiped from disk by another
72 * node. If it hasn't then we're safe as long as we hold the
73 * spin lock until our increment of open count. */
74 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
75 spin_unlock(&oi
->ip_lock
);
82 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
85 spin_unlock(&oi
->ip_lock
);
92 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
94 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
96 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
97 file
->f_dentry
->d_name
.len
,
98 file
->f_dentry
->d_name
.name
);
100 spin_lock(&oi
->ip_lock
);
101 if (!--oi
->ip_open_count
)
102 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
103 spin_unlock(&oi
->ip_lock
);
110 static int ocfs2_sync_file(struct file
*file
,
111 struct dentry
*dentry
,
116 struct inode
*inode
= dentry
->d_inode
;
117 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
119 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
120 dentry
->d_name
.len
, dentry
->d_name
.name
);
122 err
= ocfs2_sync_inode(dentry
->d_inode
);
126 journal
= osb
->journal
->j_journal
;
127 err
= journal_force_commit(journal
);
132 return (err
< 0) ? -EIO
: 0;
135 int ocfs2_set_inode_size(struct ocfs2_journal_handle
*handle
,
137 struct buffer_head
*fe_bh
,
143 i_size_write(inode
, new_i_size
);
144 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(new_i_size
);
145 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
147 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
158 static int ocfs2_simple_size_update(struct inode
*inode
,
159 struct buffer_head
*di_bh
,
163 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
164 struct ocfs2_journal_handle
*handle
= NULL
;
166 handle
= ocfs2_start_trans(osb
, NULL
,
167 OCFS2_INODE_UPDATE_CREDITS
);
168 if (handle
== NULL
) {
174 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
179 ocfs2_commit_trans(handle
);
184 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
186 struct buffer_head
*fe_bh
,
190 struct ocfs2_journal_handle
*handle
;
194 /* TODO: This needs to actually orphan the inode in this
197 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
198 if (IS_ERR(handle
)) {
199 status
= PTR_ERR(handle
);
204 status
= ocfs2_set_inode_size(handle
, inode
, fe_bh
, new_i_size
);
208 ocfs2_commit_trans(handle
);
214 static int ocfs2_truncate_file(struct inode
*inode
,
215 struct buffer_head
*di_bh
,
219 struct ocfs2_dinode
*fe
= NULL
;
220 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
221 struct ocfs2_truncate_context
*tc
= NULL
;
223 mlog_entry("(inode = %"MLFu64
", new_i_size = %"MLFu64
"\n",
224 OCFS2_I(inode
)->ip_blkno
, new_i_size
);
226 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
228 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
229 if (!OCFS2_IS_VALID_DINODE(fe
)) {
230 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
235 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
236 "Inode %"MLFu64
", inode i_size = %lld != di "
237 "i_size = %"MLFu64
", i_flags = 0x%x\n",
238 OCFS2_I(inode
)->ip_blkno
,
240 le64_to_cpu(fe
->i_size
), le32_to_cpu(fe
->i_flags
));
242 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
243 mlog(0, "asked to truncate file with size (%"MLFu64
") "
244 "to size (%"MLFu64
")!\n",
245 le64_to_cpu(fe
->i_size
), new_i_size
);
251 mlog(0, "inode %"MLFu64
", i_size = %"MLFu64
", new_i_size = %"MLFu64
"\n",
252 le64_to_cpu(fe
->i_blkno
), le64_to_cpu(fe
->i_size
), new_i_size
);
254 /* lets handle the simple truncate cases before doing any more
255 * cluster locking. */
256 if (new_i_size
== le64_to_cpu(fe
->i_size
))
259 if (le32_to_cpu(fe
->i_clusters
) ==
260 ocfs2_clusters_for_bytes(osb
->sb
, new_i_size
)) {
261 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
263 /* No allocation change is required, so lets fast path
265 status
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
271 /* This forces other nodes to sync and drop their pages */
272 status
= ocfs2_data_lock(inode
, 1);
277 ocfs2_data_unlock(inode
, 1);
279 /* alright, we're going to need to do a full blown alloc size
280 * change. Orphan the inode so that recovery can complete the
281 * truncate if necessary. This does the task of marking
283 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
289 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
295 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
301 /* TODO: orphan dir cleanup here. */
309 * extend allocation only here.
310 * we'll update all the disk stuff, and oip->alloc_size
312 * expect stuff to be locked, a transaction started and enough data /
313 * metadata reservations in the contexts.
315 * Will return -EAGAIN, and a reason if a restart is needed.
316 * If passed in, *reason will always be set, even in error.
318 int ocfs2_do_extend_allocation(struct ocfs2_super
*osb
,
321 struct buffer_head
*fe_bh
,
322 struct ocfs2_journal_handle
*handle
,
323 struct ocfs2_alloc_context
*data_ac
,
324 struct ocfs2_alloc_context
*meta_ac
,
325 enum ocfs2_alloc_restarted
*reason_ret
)
329 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
330 enum ocfs2_alloc_restarted reason
= RESTART_NONE
;
331 u32 bit_off
, num_bits
;
334 BUG_ON(!clusters_to_add
);
336 free_extents
= ocfs2_num_free_extents(osb
, inode
, fe
);
337 if (free_extents
< 0) {
338 status
= free_extents
;
343 /* there are two cases which could cause us to EAGAIN in the
344 * we-need-more-metadata case:
345 * 1) we haven't reserved *any*
346 * 2) we are so fragmented, we've needed to add metadata too
348 if (!free_extents
&& !meta_ac
) {
349 mlog(0, "we haven't reserved any metadata!\n");
351 reason
= RESTART_META
;
353 } else if ((!free_extents
)
354 && (ocfs2_alloc_context_bits_left(meta_ac
)
355 < ocfs2_extend_meta_needed(fe
))) {
356 mlog(0, "filesystem is really fragmented...\n");
358 reason
= RESTART_META
;
362 status
= ocfs2_claim_clusters(osb
, handle
, data_ac
, 1,
363 &bit_off
, &num_bits
);
365 if (status
!= -ENOSPC
)
370 BUG_ON(num_bits
> clusters_to_add
);
372 /* reserve our write early -- insert_extent may update the inode */
373 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
374 OCFS2_JOURNAL_ACCESS_WRITE
);
380 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
381 mlog(0, "Allocating %u clusters at block %u for inode %"MLFu64
"\n",
382 num_bits
, bit_off
, OCFS2_I(inode
)->ip_blkno
);
383 status
= ocfs2_insert_extent(osb
, handle
, inode
, fe_bh
, block
,
390 le32_add_cpu(&fe
->i_clusters
, num_bits
);
391 spin_lock(&OCFS2_I(inode
)->ip_lock
);
392 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
393 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
395 status
= ocfs2_journal_dirty(handle
, fe_bh
);
401 clusters_to_add
-= num_bits
;
403 if (clusters_to_add
) {
404 mlog(0, "need to alloc once more, clusters = %u, wanted = "
405 "%u\n", fe
->i_clusters
, clusters_to_add
);
407 reason
= RESTART_TRANS
;
413 *reason_ret
= reason
;
417 static int ocfs2_extend_allocation(struct inode
*inode
,
421 int restart_func
= 0;
422 int drop_alloc_sem
= 0;
423 int credits
, num_free_extents
;
425 struct buffer_head
*bh
= NULL
;
426 struct ocfs2_dinode
*fe
= NULL
;
427 struct ocfs2_journal_handle
*handle
= NULL
;
428 struct ocfs2_alloc_context
*data_ac
= NULL
;
429 struct ocfs2_alloc_context
*meta_ac
= NULL
;
430 enum ocfs2_alloc_restarted why
;
431 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
433 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
435 status
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &bh
,
436 OCFS2_BH_CACHED
, inode
);
442 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
443 if (!OCFS2_IS_VALID_DINODE(fe
)) {
444 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
450 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
452 mlog(0, "extend inode %"MLFu64
", i_size = %lld, fe->i_clusters = %u, "
453 "clusters_to_add = %u\n",
454 OCFS2_I(inode
)->ip_blkno
, i_size_read(inode
),
455 fe
->i_clusters
, clusters_to_add
);
457 handle
= ocfs2_alloc_handle(osb
);
458 if (handle
== NULL
) {
464 num_free_extents
= ocfs2_num_free_extents(osb
,
467 if (num_free_extents
< 0) {
468 status
= num_free_extents
;
473 if (!num_free_extents
) {
474 status
= ocfs2_reserve_new_metadata(osb
,
479 if (status
!= -ENOSPC
)
485 status
= ocfs2_reserve_clusters(osb
,
490 if (status
!= -ENOSPC
)
495 /* blocks peope in read/write from reading our allocation
496 * until we're done changing it. We depend on i_mutex to block
497 * other extend/truncate calls while we're here. Ordering wrt
498 * start_trans is important here -- always do it before! */
499 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
502 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
503 handle
= ocfs2_start_trans(osb
, handle
, credits
);
504 if (IS_ERR(handle
)) {
505 status
= PTR_ERR(handle
);
511 restarted_transaction
:
512 /* reserve a write to the file entry early on - that we if we
513 * run out of credits in the allocation path, we can still
515 status
= ocfs2_journal_access(handle
, inode
, bh
,
516 OCFS2_JOURNAL_ACCESS_WRITE
);
522 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
524 status
= ocfs2_do_extend_allocation(osb
,
532 if ((status
< 0) && (status
!= -EAGAIN
)) {
533 if (status
!= -ENOSPC
)
538 status
= ocfs2_journal_dirty(handle
, bh
);
544 spin_lock(&OCFS2_I(inode
)->ip_lock
);
545 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
546 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
548 if (why
!= RESTART_NONE
&& clusters_to_add
) {
549 if (why
== RESTART_META
) {
550 mlog(0, "restarting function.\n");
553 BUG_ON(why
!= RESTART_TRANS
);
555 mlog(0, "restarting transaction.\n");
556 /* TODO: This can be more intelligent. */
557 credits
= ocfs2_calc_extend_credits(osb
->sb
,
560 status
= ocfs2_extend_trans(handle
, credits
);
562 /* handle still has to be committed at
568 goto restarted_transaction
;
572 mlog(0, "fe: i_clusters = %u, i_size=%"MLFu64
"\n",
573 fe
->i_clusters
, fe
->i_size
);
574 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
575 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
578 if (drop_alloc_sem
) {
579 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
583 ocfs2_commit_trans(handle
);
587 ocfs2_free_alloc_context(data_ac
);
591 ocfs2_free_alloc_context(meta_ac
);
594 if ((!status
) && restart_func
) {
607 /* Some parts of this taken from generic_cont_expand, which turned out
608 * to be too fragile to do exactly what we need without us having to
609 * worry about recursive locking in ->commit_write(). */
610 static int ocfs2_write_zero_page(struct inode
*inode
,
613 struct address_space
*mapping
= inode
->i_mapping
;
617 struct ocfs2_journal_handle
*handle
= NULL
;
620 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
621 /* ugh. in prepare/commit_write, if from==to==start of block, we
622 ** skip the prepare. make sure we never send an offset for the start
625 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
628 index
= size
>> PAGE_CACHE_SHIFT
;
630 page
= grab_cache_page(mapping
, index
);
637 ret
= ocfs2_prepare_write(NULL
, page
, offset
, offset
);
643 if (ocfs2_should_order_data(inode
)) {
644 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
646 if (IS_ERR(handle
)) {
647 ret
= PTR_ERR(handle
);
653 /* must not update i_size! */
654 ret
= block_commit_write(page
, offset
, offset
);
661 ocfs2_commit_trans(handle
);
664 page_cache_release(page
);
669 static int ocfs2_zero_extend(struct inode
*inode
,
674 struct super_block
*sb
= inode
->i_sb
;
676 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
677 while (start_off
< zero_to_size
) {
678 ret
= ocfs2_write_zero_page(inode
, start_off
);
684 start_off
+= sb
->s_blocksize
;
691 static int ocfs2_extend_file(struct inode
*inode
,
692 struct buffer_head
*di_bh
,
698 /* setattr sometimes calls us like this. */
702 if (i_size_read(inode
) == new_i_size
)
704 BUG_ON(new_i_size
< i_size_read(inode
));
706 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
707 OCFS2_I(inode
)->ip_clusters
;
709 if (clusters_to_add
) {
710 ret
= ocfs2_extend_allocation(inode
, clusters_to_add
);
716 ret
= ocfs2_zero_extend(inode
, new_i_size
);
723 /* No allocation required, we just use this helper to
724 * do a trivial update of i_size. */
725 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
735 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
737 int status
= 0, size_change
;
738 struct inode
*inode
= dentry
->d_inode
;
739 struct super_block
*sb
= inode
->i_sb
;
740 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
741 struct buffer_head
*bh
= NULL
;
742 struct ocfs2_journal_handle
*handle
= NULL
;
744 mlog_entry("(0x%p, '%.*s')\n", dentry
,
745 dentry
->d_name
.len
, dentry
->d_name
.name
);
747 if (attr
->ia_valid
& ATTR_MODE
)
748 mlog(0, "mode change: %d\n", attr
->ia_mode
);
749 if (attr
->ia_valid
& ATTR_UID
)
750 mlog(0, "uid change: %d\n", attr
->ia_uid
);
751 if (attr
->ia_valid
& ATTR_GID
)
752 mlog(0, "gid change: %d\n", attr
->ia_gid
);
753 if (attr
->ia_valid
& ATTR_SIZE
)
754 mlog(0, "size change...\n");
755 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
756 mlog(0, "time change...\n");
758 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
759 | ATTR_GID | ATTR_UID | ATTR_MODE)
760 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
761 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
765 status
= inode_change_ok(inode
, attr
);
769 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
771 status
= ocfs2_rw_lock(inode
, 1);
778 status
= ocfs2_meta_lock(inode
, NULL
, &bh
, 1);
780 if (status
!= -ENOENT
)
785 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
786 if (i_size_read(inode
) > attr
->ia_size
)
787 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
789 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
791 if (status
!= -ENOSPC
)
798 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
799 if (IS_ERR(handle
)) {
800 status
= PTR_ERR(handle
);
805 status
= inode_setattr(inode
, attr
);
811 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
816 ocfs2_commit_trans(handle
);
818 ocfs2_meta_unlock(inode
, 1);
821 ocfs2_rw_unlock(inode
, 1);
830 int ocfs2_getattr(struct vfsmount
*mnt
,
831 struct dentry
*dentry
,
834 struct inode
*inode
= dentry
->d_inode
;
835 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
836 struct ocfs2_super
*osb
= sb
->s_fs_info
;
841 err
= ocfs2_inode_revalidate(dentry
);
848 generic_fillattr(inode
, stat
);
850 /* We set the blksize from the cluster size for performance */
851 stat
->blksize
= osb
->s_clustersize
;
859 static int ocfs2_write_remove_suid(struct inode
*inode
)
862 struct buffer_head
*bh
= NULL
;
863 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
864 struct ocfs2_journal_handle
*handle
;
865 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
866 struct ocfs2_dinode
*di
;
868 mlog_entry("(Inode %"MLFu64
", mode 0%o)\n", oi
->ip_blkno
,
871 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
872 if (handle
== NULL
) {
878 ret
= ocfs2_read_block(osb
, oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
884 ret
= ocfs2_journal_access(handle
, inode
, bh
,
885 OCFS2_JOURNAL_ACCESS_WRITE
);
891 inode
->i_mode
&= ~S_ISUID
;
892 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
893 inode
->i_mode
&= ~S_ISGID
;
895 di
= (struct ocfs2_dinode
*) bh
->b_data
;
896 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
898 ret
= ocfs2_journal_dirty(handle
, bh
);
904 ocfs2_commit_trans(handle
);
910 static inline int ocfs2_write_should_remove_suid(struct inode
*inode
)
912 mode_t mode
= inode
->i_mode
;
914 if (!capable(CAP_FSETID
)) {
915 if (unlikely(mode
& S_ISUID
))
918 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
924 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
925 const char __user
*buf
,
929 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
931 int ret
, rw_level
= -1, meta_level
= -1, have_alloc_sem
= 0;
933 struct file
*filp
= iocb
->ki_filp
;
934 struct inode
*inode
= filp
->f_dentry
->d_inode
;
935 loff_t newsize
, saved_pos
;
936 #ifdef OCFS2_ORACORE_WORKAROUNDS
937 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
940 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
942 filp
->f_dentry
->d_name
.len
,
943 filp
->f_dentry
->d_name
.name
);
945 /* happy write of zero bytes */
950 mlog(0, "bad inode\n");
954 #ifdef OCFS2_ORACORE_WORKAROUNDS
955 /* ugh, work around some applications which open everything O_DIRECT +
956 * O_APPEND and really don't mean to use O_DIRECT. */
957 if (osb
->s_mount_opt
& OCFS2_MOUNT_COMPAT_OCFS
&&
958 (filp
->f_flags
& O_APPEND
) && (filp
->f_flags
& O_DIRECT
))
959 filp
->f_flags
&= ~O_DIRECT
;
962 mutex_lock(&inode
->i_mutex
);
963 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
964 if (filp
->f_flags
& O_DIRECT
) {
966 down_read(&inode
->i_alloc_sem
);
969 /* concurrent O_DIRECT writes are allowed */
970 rw_level
= (filp
->f_flags
& O_DIRECT
) ? 0 : 1;
971 ret
= ocfs2_rw_lock(inode
, rw_level
);
979 * We sample i_size under a read level meta lock to see if our write
980 * is extending the file, if it is we back off and get a write level
983 meta_level
= (filp
->f_flags
& O_APPEND
) ? 1 : 0;
985 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, meta_level
);
992 /* Clear suid / sgid if necessary. We do this here
993 * instead of later in the write path because
994 * remove_suid() calls ->setattr without any hint that
995 * we may have already done our cluster locking. Since
996 * ocfs2_setattr() *must* take cluster locks to
997 * proceeed, this will lead us to recursively lock the
998 * inode. There's also the dinode i_size state which
999 * can be lost via setattr during extending writes (we
1000 * set inode->i_size at the end of a write. */
1001 if (ocfs2_write_should_remove_suid(inode
)) {
1002 if (meta_level
== 0) {
1003 ocfs2_meta_unlock(inode
, meta_level
);
1008 ret
= ocfs2_write_remove_suid(inode
);
1015 /* work on a copy of ppos until we're sure that we won't have
1016 * to recalculate it due to relocking. */
1017 if (filp
->f_flags
& O_APPEND
) {
1018 saved_pos
= i_size_read(inode
);
1019 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1021 saved_pos
= iocb
->ki_pos
;
1023 newsize
= count
+ saved_pos
;
1025 mlog(0, "pos=%lld newsize=%"MLFu64
" cursize=%lld\n",
1026 saved_pos
, newsize
, i_size_read(inode
));
1028 /* No need for a higher level metadata lock if we're
1029 * never going past i_size. */
1030 if (newsize
<= i_size_read(inode
))
1033 if (meta_level
== 0) {
1034 ocfs2_meta_unlock(inode
, meta_level
);
1039 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1040 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1041 OCFS2_I(inode
)->ip_clusters
;
1042 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1044 mlog(0, "Writing at EOF, may need more allocation: "
1045 "i_size = %lld, newsize = %"MLFu64
", need %u clusters\n",
1046 i_size_read(inode
), newsize
, clusters
);
1048 /* We only want to continue the rest of this loop if
1049 * our extend will actually require more
1054 ret
= ocfs2_extend_allocation(inode
, clusters
);
1061 /* Fill any holes which would've been created by this
1062 * write. If we're O_APPEND, this will wind up
1063 * (correctly) being a noop. */
1064 ret
= ocfs2_zero_extend(inode
, (u64
) newsize
- count
);
1072 /* ok, we're done with i_size and alloc work */
1073 iocb
->ki_pos
= saved_pos
;
1074 ocfs2_meta_unlock(inode
, meta_level
);
1077 /* communicate with ocfs2_dio_end_io */
1078 ocfs2_iocb_set_rw_locked(iocb
);
1080 #ifdef OCFS2_ORACORE_WORKAROUNDS
1081 if (osb
->s_mount_opt
& OCFS2_MOUNT_COMPAT_OCFS
&&
1082 filp
->f_flags
& O_DIRECT
) {
1083 unsigned int saved_flags
= filp
->f_flags
;
1084 int sector_size
= 1 << osb
->s_sectsize_bits
;
1086 if ((saved_pos
& (sector_size
- 1)) ||
1087 (count
& (sector_size
- 1)) ||
1088 ((unsigned long)buf
& (sector_size
- 1))) {
1089 filp
->f_flags
|= O_SYNC
;
1090 filp
->f_flags
&= ~O_DIRECT
;
1093 ret
= generic_file_aio_write_nolock(iocb
, &local_iov
, 1,
1096 filp
->f_flags
= saved_flags
;
1099 ret
= generic_file_aio_write_nolock(iocb
, &local_iov
, 1,
1102 /* buffered aio wouldn't have proper lock coverage today */
1103 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1106 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1107 * function pointer which is called when o_direct io completes so that
1108 * it can unlock our rw lock. (it's the clustered equivalent of
1109 * i_alloc_sem; protects truncate from racing with pending ios).
1110 * Unfortunately there are error cases which call end_io and others
1111 * that don't. so we don't have to unlock the rw_lock if either an
1112 * async dio is going to do it in the future or an end_io after an
1113 * error has already done it.
1115 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1121 if (meta_level
!= -1)
1122 ocfs2_meta_unlock(inode
, meta_level
);
1124 up_read(&inode
->i_alloc_sem
);
1126 ocfs2_rw_unlock(inode
, rw_level
);
1127 mutex_unlock(&inode
->i_mutex
);
1133 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
1138 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0;
1139 struct file
*filp
= iocb
->ki_filp
;
1140 struct inode
*inode
= filp
->f_dentry
->d_inode
;
1141 #ifdef OCFS2_ORACORE_WORKAROUNDS
1142 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1145 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
1146 (unsigned int)count
,
1147 filp
->f_dentry
->d_name
.len
,
1148 filp
->f_dentry
->d_name
.name
);
1156 #ifdef OCFS2_ORACORE_WORKAROUNDS
1157 if (osb
->s_mount_opt
& OCFS2_MOUNT_COMPAT_OCFS
) {
1158 if (filp
->f_flags
& O_DIRECT
) {
1159 int sector_size
= 1 << osb
->s_sectsize_bits
;
1161 if ((pos
& (sector_size
- 1)) ||
1162 (count
& (sector_size
- 1)) ||
1163 ((unsigned long)buf
& (sector_size
- 1)) ||
1164 (i_size_read(inode
) & (sector_size
-1))) {
1165 filp
->f_flags
&= ~O_DIRECT
;
1172 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1173 * need locks to protect pending reads from racing with truncate.
1175 if (filp
->f_flags
& O_DIRECT
) {
1176 down_read(&inode
->i_alloc_sem
);
1179 ret
= ocfs2_rw_lock(inode
, 0);
1185 /* communicate with ocfs2_dio_end_io */
1186 ocfs2_iocb_set_rw_locked(iocb
);
1189 ret
= generic_file_aio_read(iocb
, buf
, count
, iocb
->ki_pos
);
1191 mlog(ML_ERROR
, "generic_file_aio_read returned -EINVAL\n");
1193 /* buffered aio wouldn't have proper lock coverage today */
1194 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1196 /* see ocfs2_file_aio_write */
1197 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1204 up_read(&inode
->i_alloc_sem
);
1206 ocfs2_rw_unlock(inode
, rw_level
);
1212 struct inode_operations ocfs2_file_iops
= {
1213 .setattr
= ocfs2_setattr
,
1214 .getattr
= ocfs2_getattr
,
1217 struct inode_operations ocfs2_special_file_iops
= {
1218 .setattr
= ocfs2_setattr
,
1219 .getattr
= ocfs2_getattr
,
1222 struct file_operations ocfs2_fops
= {
1223 .read
= do_sync_read
,
1224 .write
= do_sync_write
,
1225 .sendfile
= generic_file_sendfile
,
1227 .fsync
= ocfs2_sync_file
,
1228 .release
= ocfs2_file_release
,
1229 .open
= ocfs2_file_open
,
1230 .aio_read
= ocfs2_file_aio_read
,
1231 .aio_write
= ocfs2_file_aio_write
,
1234 struct file_operations ocfs2_dops
= {
1235 .read
= generic_read_dir
,
1236 .readdir
= ocfs2_readdir
,
1237 .fsync
= ocfs2_sync_file
,