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.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/uio.h>
33 #define MLOG_MASK_PREFIX ML_INODE
34 #include <cluster/masklog.h>
42 #include "extent_map.h"
51 #include "buffer_head_io.h"
53 static int ocfs2_sync_inode(struct inode
*inode
)
55 filemap_fdatawrite(inode
->i_mapping
);
56 return sync_mapping_buffers(inode
->i_mapping
);
59 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
62 int mode
= file
->f_flags
;
63 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
65 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
66 file
->f_dentry
->d_name
.len
, file
->f_dentry
->d_name
.name
);
68 spin_lock(&oi
->ip_lock
);
70 /* Check that the inode hasn't been wiped from disk by another
71 * node. If it hasn't then we're safe as long as we hold the
72 * spin lock until our increment of open count. */
73 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
74 spin_unlock(&oi
->ip_lock
);
81 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
84 spin_unlock(&oi
->ip_lock
);
91 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
93 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
95 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
96 file
->f_dentry
->d_name
.len
,
97 file
->f_dentry
->d_name
.name
);
99 spin_lock(&oi
->ip_lock
);
100 if (!--oi
->ip_open_count
)
101 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
102 spin_unlock(&oi
->ip_lock
);
109 static int ocfs2_sync_file(struct file
*file
,
110 struct dentry
*dentry
,
115 struct inode
*inode
= dentry
->d_inode
;
116 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
118 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
119 dentry
->d_name
.len
, dentry
->d_name
.name
);
121 err
= ocfs2_sync_inode(dentry
->d_inode
);
125 journal
= osb
->journal
->j_journal
;
126 err
= journal_force_commit(journal
);
131 return (err
< 0) ? -EIO
: 0;
134 int ocfs2_set_inode_size(struct ocfs2_journal_handle
*handle
,
136 struct buffer_head
*fe_bh
,
142 i_size_write(inode
, new_i_size
);
143 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(new_i_size
);
144 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
146 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
157 static int ocfs2_simple_size_update(struct inode
*inode
,
158 struct buffer_head
*di_bh
,
162 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
163 struct ocfs2_journal_handle
*handle
= NULL
;
165 handle
= ocfs2_start_trans(osb
, NULL
,
166 OCFS2_INODE_UPDATE_CREDITS
);
167 if (handle
== NULL
) {
173 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
178 ocfs2_commit_trans(handle
);
183 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
185 struct buffer_head
*fe_bh
,
189 struct ocfs2_journal_handle
*handle
;
193 /* TODO: This needs to actually orphan the inode in this
196 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
197 if (IS_ERR(handle
)) {
198 status
= PTR_ERR(handle
);
203 status
= ocfs2_set_inode_size(handle
, inode
, fe_bh
, new_i_size
);
207 ocfs2_commit_trans(handle
);
213 static int ocfs2_truncate_file(struct inode
*inode
,
214 struct buffer_head
*di_bh
,
218 struct ocfs2_dinode
*fe
= NULL
;
219 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
220 struct ocfs2_truncate_context
*tc
= NULL
;
222 mlog_entry("(inode = %"MLFu64
", new_i_size = %"MLFu64
"\n",
223 OCFS2_I(inode
)->ip_blkno
, new_i_size
);
225 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
227 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
228 if (!OCFS2_IS_VALID_DINODE(fe
)) {
229 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
234 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
235 "Inode %"MLFu64
", inode i_size = %lld != di "
236 "i_size = %"MLFu64
", i_flags = 0x%x\n",
237 OCFS2_I(inode
)->ip_blkno
,
239 le64_to_cpu(fe
->i_size
), le32_to_cpu(fe
->i_flags
));
241 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
242 mlog(0, "asked to truncate file with size (%"MLFu64
") "
243 "to size (%"MLFu64
")!\n",
244 le64_to_cpu(fe
->i_size
), new_i_size
);
250 mlog(0, "inode %"MLFu64
", i_size = %"MLFu64
", new_i_size = %"MLFu64
"\n",
251 le64_to_cpu(fe
->i_blkno
), le64_to_cpu(fe
->i_size
), new_i_size
);
253 /* lets handle the simple truncate cases before doing any more
254 * cluster locking. */
255 if (new_i_size
== le64_to_cpu(fe
->i_size
))
258 if (le32_to_cpu(fe
->i_clusters
) ==
259 ocfs2_clusters_for_bytes(osb
->sb
, new_i_size
)) {
260 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
262 /* No allocation change is required, so lets fast path
264 status
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
270 /* This forces other nodes to sync and drop their pages */
271 status
= ocfs2_data_lock(inode
, 1);
276 ocfs2_data_unlock(inode
, 1);
278 /* alright, we're going to need to do a full blown alloc size
279 * change. Orphan the inode so that recovery can complete the
280 * truncate if necessary. This does the task of marking
282 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
288 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
294 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
300 /* TODO: orphan dir cleanup here. */
308 * extend allocation only here.
309 * we'll update all the disk stuff, and oip->alloc_size
311 * expect stuff to be locked, a transaction started and enough data /
312 * metadata reservations in the contexts.
314 * Will return -EAGAIN, and a reason if a restart is needed.
315 * If passed in, *reason will always be set, even in error.
317 int ocfs2_do_extend_allocation(struct ocfs2_super
*osb
,
320 struct buffer_head
*fe_bh
,
321 struct ocfs2_journal_handle
*handle
,
322 struct ocfs2_alloc_context
*data_ac
,
323 struct ocfs2_alloc_context
*meta_ac
,
324 enum ocfs2_alloc_restarted
*reason_ret
)
328 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
329 enum ocfs2_alloc_restarted reason
= RESTART_NONE
;
330 u32 bit_off
, num_bits
;
333 BUG_ON(!clusters_to_add
);
335 free_extents
= ocfs2_num_free_extents(osb
, inode
, fe
);
336 if (free_extents
< 0) {
337 status
= free_extents
;
342 /* there are two cases which could cause us to EAGAIN in the
343 * we-need-more-metadata case:
344 * 1) we haven't reserved *any*
345 * 2) we are so fragmented, we've needed to add metadata too
347 if (!free_extents
&& !meta_ac
) {
348 mlog(0, "we haven't reserved any metadata!\n");
350 reason
= RESTART_META
;
352 } else if ((!free_extents
)
353 && (ocfs2_alloc_context_bits_left(meta_ac
)
354 < ocfs2_extend_meta_needed(fe
))) {
355 mlog(0, "filesystem is really fragmented...\n");
357 reason
= RESTART_META
;
361 status
= ocfs2_claim_clusters(osb
, handle
, data_ac
, 1,
362 &bit_off
, &num_bits
);
364 if (status
!= -ENOSPC
)
369 BUG_ON(num_bits
> clusters_to_add
);
371 /* reserve our write early -- insert_extent may update the inode */
372 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
373 OCFS2_JOURNAL_ACCESS_WRITE
);
379 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
380 mlog(0, "Allocating %u clusters at block %u for inode %"MLFu64
"\n",
381 num_bits
, bit_off
, OCFS2_I(inode
)->ip_blkno
);
382 status
= ocfs2_insert_extent(osb
, handle
, inode
, fe_bh
, block
,
389 le32_add_cpu(&fe
->i_clusters
, num_bits
);
390 spin_lock(&OCFS2_I(inode
)->ip_lock
);
391 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
392 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
394 status
= ocfs2_journal_dirty(handle
, fe_bh
);
400 clusters_to_add
-= num_bits
;
402 if (clusters_to_add
) {
403 mlog(0, "need to alloc once more, clusters = %u, wanted = "
404 "%u\n", fe
->i_clusters
, clusters_to_add
);
406 reason
= RESTART_TRANS
;
412 *reason_ret
= reason
;
416 static int ocfs2_extend_allocation(struct inode
*inode
,
420 int restart_func
= 0;
421 int drop_alloc_sem
= 0;
422 int credits
, num_free_extents
;
424 struct buffer_head
*bh
= NULL
;
425 struct ocfs2_dinode
*fe
= NULL
;
426 struct ocfs2_journal_handle
*handle
= NULL
;
427 struct ocfs2_alloc_context
*data_ac
= NULL
;
428 struct ocfs2_alloc_context
*meta_ac
= NULL
;
429 enum ocfs2_alloc_restarted why
;
430 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
432 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
434 status
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &bh
,
435 OCFS2_BH_CACHED
, inode
);
441 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
442 if (!OCFS2_IS_VALID_DINODE(fe
)) {
443 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
449 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
451 mlog(0, "extend inode %"MLFu64
", i_size = %lld, fe->i_clusters = %u, "
452 "clusters_to_add = %u\n",
453 OCFS2_I(inode
)->ip_blkno
, i_size_read(inode
),
454 fe
->i_clusters
, clusters_to_add
);
456 handle
= ocfs2_alloc_handle(osb
);
457 if (handle
== NULL
) {
463 num_free_extents
= ocfs2_num_free_extents(osb
,
466 if (num_free_extents
< 0) {
467 status
= num_free_extents
;
472 if (!num_free_extents
) {
473 status
= ocfs2_reserve_new_metadata(osb
,
478 if (status
!= -ENOSPC
)
484 status
= ocfs2_reserve_clusters(osb
,
489 if (status
!= -ENOSPC
)
494 /* blocks peope in read/write from reading our allocation
495 * until we're done changing it. We depend on i_mutex to block
496 * other extend/truncate calls while we're here. Ordering wrt
497 * start_trans is important here -- always do it before! */
498 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
501 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
502 handle
= ocfs2_start_trans(osb
, handle
, credits
);
503 if (IS_ERR(handle
)) {
504 status
= PTR_ERR(handle
);
510 restarted_transaction
:
511 /* reserve a write to the file entry early on - that we if we
512 * run out of credits in the allocation path, we can still
514 status
= ocfs2_journal_access(handle
, inode
, bh
,
515 OCFS2_JOURNAL_ACCESS_WRITE
);
521 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
523 status
= ocfs2_do_extend_allocation(osb
,
531 if ((status
< 0) && (status
!= -EAGAIN
)) {
532 if (status
!= -ENOSPC
)
537 status
= ocfs2_journal_dirty(handle
, bh
);
543 spin_lock(&OCFS2_I(inode
)->ip_lock
);
544 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
545 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
547 if (why
!= RESTART_NONE
&& clusters_to_add
) {
548 if (why
== RESTART_META
) {
549 mlog(0, "restarting function.\n");
552 BUG_ON(why
!= RESTART_TRANS
);
554 mlog(0, "restarting transaction.\n");
555 /* TODO: This can be more intelligent. */
556 credits
= ocfs2_calc_extend_credits(osb
->sb
,
559 status
= ocfs2_extend_trans(handle
, credits
);
561 /* handle still has to be committed at
567 goto restarted_transaction
;
571 mlog(0, "fe: i_clusters = %u, i_size=%"MLFu64
"\n",
572 fe
->i_clusters
, fe
->i_size
);
573 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
574 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
577 if (drop_alloc_sem
) {
578 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
582 ocfs2_commit_trans(handle
);
586 ocfs2_free_alloc_context(data_ac
);
590 ocfs2_free_alloc_context(meta_ac
);
593 if ((!status
) && restart_func
) {
606 /* Some parts of this taken from generic_cont_expand, which turned out
607 * to be too fragile to do exactly what we need without us having to
608 * worry about recursive locking in ->commit_write(). */
609 static int ocfs2_write_zero_page(struct inode
*inode
,
612 struct address_space
*mapping
= inode
->i_mapping
;
616 struct ocfs2_journal_handle
*handle
= NULL
;
619 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
620 /* ugh. in prepare/commit_write, if from==to==start of block, we
621 ** skip the prepare. make sure we never send an offset for the start
624 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
627 index
= size
>> PAGE_CACHE_SHIFT
;
629 page
= grab_cache_page(mapping
, index
);
636 ret
= ocfs2_prepare_write(NULL
, page
, offset
, offset
);
642 if (ocfs2_should_order_data(inode
)) {
643 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
645 if (IS_ERR(handle
)) {
646 ret
= PTR_ERR(handle
);
652 /* must not update i_size! */
653 ret
= block_commit_write(page
, offset
, offset
);
660 ocfs2_commit_trans(handle
);
663 page_cache_release(page
);
668 static int ocfs2_zero_extend(struct inode
*inode
,
673 struct super_block
*sb
= inode
->i_sb
;
675 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
676 while (start_off
< zero_to_size
) {
677 ret
= ocfs2_write_zero_page(inode
, start_off
);
683 start_off
+= sb
->s_blocksize
;
690 static int ocfs2_extend_file(struct inode
*inode
,
691 struct buffer_head
*di_bh
,
697 /* setattr sometimes calls us like this. */
701 if (i_size_read(inode
) == new_i_size
)
703 BUG_ON(new_i_size
< i_size_read(inode
));
705 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
706 OCFS2_I(inode
)->ip_clusters
;
708 if (clusters_to_add
) {
709 ret
= ocfs2_extend_allocation(inode
, clusters_to_add
);
715 ret
= ocfs2_zero_extend(inode
, new_i_size
);
722 /* No allocation required, we just use this helper to
723 * do a trivial update of i_size. */
724 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
734 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
736 int status
= 0, size_change
;
737 struct inode
*inode
= dentry
->d_inode
;
738 struct super_block
*sb
= inode
->i_sb
;
739 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
740 struct buffer_head
*bh
= NULL
;
741 struct ocfs2_journal_handle
*handle
= NULL
;
743 mlog_entry("(0x%p, '%.*s')\n", dentry
,
744 dentry
->d_name
.len
, dentry
->d_name
.name
);
746 if (attr
->ia_valid
& ATTR_MODE
)
747 mlog(0, "mode change: %d\n", attr
->ia_mode
);
748 if (attr
->ia_valid
& ATTR_UID
)
749 mlog(0, "uid change: %d\n", attr
->ia_uid
);
750 if (attr
->ia_valid
& ATTR_GID
)
751 mlog(0, "gid change: %d\n", attr
->ia_gid
);
752 if (attr
->ia_valid
& ATTR_SIZE
)
753 mlog(0, "size change...\n");
754 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
755 mlog(0, "time change...\n");
757 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
758 | ATTR_GID | ATTR_UID | ATTR_MODE)
759 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
760 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
764 status
= inode_change_ok(inode
, attr
);
768 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
770 status
= ocfs2_rw_lock(inode
, 1);
777 status
= ocfs2_meta_lock(inode
, NULL
, &bh
, 1);
779 if (status
!= -ENOENT
)
784 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
785 if (i_size_read(inode
) > attr
->ia_size
)
786 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
788 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
790 if (status
!= -ENOSPC
)
797 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
798 if (IS_ERR(handle
)) {
799 status
= PTR_ERR(handle
);
804 status
= inode_setattr(inode
, attr
);
810 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
815 ocfs2_commit_trans(handle
);
817 ocfs2_meta_unlock(inode
, 1);
820 ocfs2_rw_unlock(inode
, 1);
829 int ocfs2_getattr(struct vfsmount
*mnt
,
830 struct dentry
*dentry
,
833 struct inode
*inode
= dentry
->d_inode
;
834 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
835 struct ocfs2_super
*osb
= sb
->s_fs_info
;
840 err
= ocfs2_inode_revalidate(dentry
);
847 generic_fillattr(inode
, stat
);
849 /* We set the blksize from the cluster size for performance */
850 stat
->blksize
= osb
->s_clustersize
;
858 static int ocfs2_write_remove_suid(struct inode
*inode
)
861 struct buffer_head
*bh
= NULL
;
862 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
863 struct ocfs2_journal_handle
*handle
;
864 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
865 struct ocfs2_dinode
*di
;
867 mlog_entry("(Inode %"MLFu64
", mode 0%o)\n", oi
->ip_blkno
,
870 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
871 if (handle
== NULL
) {
877 ret
= ocfs2_read_block(osb
, oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
883 ret
= ocfs2_journal_access(handle
, inode
, bh
,
884 OCFS2_JOURNAL_ACCESS_WRITE
);
890 inode
->i_mode
&= ~S_ISUID
;
891 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
892 inode
->i_mode
&= ~S_ISGID
;
894 di
= (struct ocfs2_dinode
*) bh
->b_data
;
895 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
897 ret
= ocfs2_journal_dirty(handle
, bh
);
903 ocfs2_commit_trans(handle
);
909 static inline int ocfs2_write_should_remove_suid(struct inode
*inode
)
911 mode_t mode
= inode
->i_mode
;
913 if (!capable(CAP_FSETID
)) {
914 if (unlikely(mode
& S_ISUID
))
917 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
923 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
924 const char __user
*buf
,
928 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
930 int ret
, rw_level
= -1, meta_level
= -1, have_alloc_sem
= 0;
932 struct file
*filp
= iocb
->ki_filp
;
933 struct inode
*inode
= filp
->f_dentry
->d_inode
;
934 loff_t newsize
, saved_pos
;
935 #ifdef OCFS2_ORACORE_WORKAROUNDS
936 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
939 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
941 filp
->f_dentry
->d_name
.len
,
942 filp
->f_dentry
->d_name
.name
);
944 /* happy write of zero bytes */
949 mlog(0, "bad inode\n");
953 #ifdef OCFS2_ORACORE_WORKAROUNDS
954 /* ugh, work around some applications which open everything O_DIRECT +
955 * O_APPEND and really don't mean to use O_DIRECT. */
956 if (osb
->s_mount_opt
& OCFS2_MOUNT_COMPAT_OCFS
&&
957 (filp
->f_flags
& O_APPEND
) && (filp
->f_flags
& O_DIRECT
))
958 filp
->f_flags
&= ~O_DIRECT
;
961 mutex_lock(&inode
->i_mutex
);
962 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
963 if (filp
->f_flags
& O_DIRECT
) {
965 down_read(&inode
->i_alloc_sem
);
968 /* concurrent O_DIRECT writes are allowed */
969 rw_level
= (filp
->f_flags
& O_DIRECT
) ? 0 : 1;
970 ret
= ocfs2_rw_lock(inode
, rw_level
);
978 * We sample i_size under a read level meta lock to see if our write
979 * is extending the file, if it is we back off and get a write level
982 meta_level
= (filp
->f_flags
& O_APPEND
) ? 1 : 0;
984 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, meta_level
);
991 /* Clear suid / sgid if necessary. We do this here
992 * instead of later in the write path because
993 * remove_suid() calls ->setattr without any hint that
994 * we may have already done our cluster locking. Since
995 * ocfs2_setattr() *must* take cluster locks to
996 * proceeed, this will lead us to recursively lock the
997 * inode. There's also the dinode i_size state which
998 * can be lost via setattr during extending writes (we
999 * set inode->i_size at the end of a write. */
1000 if (ocfs2_write_should_remove_suid(inode
)) {
1001 if (meta_level
== 0) {
1002 ocfs2_meta_unlock(inode
, meta_level
);
1007 ret
= ocfs2_write_remove_suid(inode
);
1014 /* work on a copy of ppos until we're sure that we won't have
1015 * to recalculate it due to relocking. */
1016 if (filp
->f_flags
& O_APPEND
) {
1017 saved_pos
= i_size_read(inode
);
1018 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1020 saved_pos
= iocb
->ki_pos
;
1022 newsize
= count
+ saved_pos
;
1024 mlog(0, "pos=%lld newsize=%"MLFu64
" cursize=%lld\n",
1025 saved_pos
, newsize
, i_size_read(inode
));
1027 /* No need for a higher level metadata lock if we're
1028 * never going past i_size. */
1029 if (newsize
<= i_size_read(inode
))
1032 if (meta_level
== 0) {
1033 ocfs2_meta_unlock(inode
, meta_level
);
1038 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1039 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1040 OCFS2_I(inode
)->ip_clusters
;
1041 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1043 mlog(0, "Writing at EOF, may need more allocation: "
1044 "i_size = %lld, newsize = %"MLFu64
", need %u clusters\n",
1045 i_size_read(inode
), newsize
, clusters
);
1047 /* We only want to continue the rest of this loop if
1048 * our extend will actually require more
1053 ret
= ocfs2_extend_allocation(inode
, clusters
);
1060 /* Fill any holes which would've been created by this
1061 * write. If we're O_APPEND, this will wind up
1062 * (correctly) being a noop. */
1063 ret
= ocfs2_zero_extend(inode
, (u64
) newsize
- count
);
1071 /* ok, we're done with i_size and alloc work */
1072 iocb
->ki_pos
= saved_pos
;
1073 ocfs2_meta_unlock(inode
, meta_level
);
1076 /* communicate with ocfs2_dio_end_io */
1077 ocfs2_iocb_set_rw_locked(iocb
);
1079 #ifdef OCFS2_ORACORE_WORKAROUNDS
1080 if (osb
->s_mount_opt
& OCFS2_MOUNT_COMPAT_OCFS
&&
1081 filp
->f_flags
& O_DIRECT
) {
1082 unsigned int saved_flags
= filp
->f_flags
;
1083 int sector_size
= 1 << osb
->s_sectsize_bits
;
1085 if ((saved_pos
& (sector_size
- 1)) ||
1086 (count
& (sector_size
- 1)) ||
1087 ((unsigned long)buf
& (sector_size
- 1))) {
1088 filp
->f_flags
|= O_SYNC
;
1089 filp
->f_flags
&= ~O_DIRECT
;
1092 ret
= generic_file_aio_write_nolock(iocb
, &local_iov
, 1,
1095 filp
->f_flags
= saved_flags
;
1098 ret
= generic_file_aio_write_nolock(iocb
, &local_iov
, 1,
1101 /* buffered aio wouldn't have proper lock coverage today */
1102 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1105 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1106 * function pointer which is called when o_direct io completes so that
1107 * it can unlock our rw lock. (it's the clustered equivalent of
1108 * i_alloc_sem; protects truncate from racing with pending ios).
1109 * Unfortunately there are error cases which call end_io and others
1110 * that don't. so we don't have to unlock the rw_lock if either an
1111 * async dio is going to do it in the future or an end_io after an
1112 * error has already done it.
1114 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1120 if (meta_level
!= -1)
1121 ocfs2_meta_unlock(inode
, meta_level
);
1123 up_read(&inode
->i_alloc_sem
);
1125 ocfs2_rw_unlock(inode
, rw_level
);
1126 mutex_unlock(&inode
->i_mutex
);
1132 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
1137 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0;
1138 struct file
*filp
= iocb
->ki_filp
;
1139 struct inode
*inode
= filp
->f_dentry
->d_inode
;
1140 #ifdef OCFS2_ORACORE_WORKAROUNDS
1141 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1144 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
1145 (unsigned int)count
,
1146 filp
->f_dentry
->d_name
.len
,
1147 filp
->f_dentry
->d_name
.name
);
1155 #ifdef OCFS2_ORACORE_WORKAROUNDS
1156 if (osb
->s_mount_opt
& OCFS2_MOUNT_COMPAT_OCFS
) {
1157 if (filp
->f_flags
& O_DIRECT
) {
1158 int sector_size
= 1 << osb
->s_sectsize_bits
;
1160 if ((pos
& (sector_size
- 1)) ||
1161 (count
& (sector_size
- 1)) ||
1162 ((unsigned long)buf
& (sector_size
- 1)) ||
1163 (i_size_read(inode
) & (sector_size
-1))) {
1164 filp
->f_flags
&= ~O_DIRECT
;
1171 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1172 * need locks to protect pending reads from racing with truncate.
1174 if (filp
->f_flags
& O_DIRECT
) {
1175 down_read(&inode
->i_alloc_sem
);
1178 ret
= ocfs2_rw_lock(inode
, 0);
1184 /* communicate with ocfs2_dio_end_io */
1185 ocfs2_iocb_set_rw_locked(iocb
);
1188 ret
= generic_file_aio_read(iocb
, buf
, count
, iocb
->ki_pos
);
1190 mlog(ML_ERROR
, "generic_file_aio_read returned -EINVAL\n");
1192 /* buffered aio wouldn't have proper lock coverage today */
1193 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1195 /* see ocfs2_file_aio_write */
1196 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1203 up_read(&inode
->i_alloc_sem
);
1205 ocfs2_rw_unlock(inode
, rw_level
);
1211 struct inode_operations ocfs2_file_iops
= {
1212 .setattr
= ocfs2_setattr
,
1213 .getattr
= ocfs2_getattr
,
1216 struct inode_operations ocfs2_special_file_iops
= {
1217 .setattr
= ocfs2_setattr
,
1218 .getattr
= ocfs2_getattr
,
1221 struct file_operations ocfs2_fops
= {
1222 .read
= do_sync_read
,
1223 .write
= do_sync_write
,
1224 .sendfile
= generic_file_sendfile
,
1226 .fsync
= ocfs2_sync_file
,
1227 .release
= ocfs2_file_release
,
1228 .open
= ocfs2_file_open
,
1229 .aio_read
= ocfs2_file_aio_read
,
1230 .aio_write
= ocfs2_file_aio_write
,
1233 struct file_operations ocfs2_dops
= {
1234 .read
= generic_read_dir
,
1235 .readdir
= ocfs2_readdir
,
1236 .fsync
= ocfs2_sync_file
,