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 = %llu, new_i_size = %llu\n",
224 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
225 (unsigned long long)new_i_size
);
227 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
229 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
230 if (!OCFS2_IS_VALID_DINODE(fe
)) {
231 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
236 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
237 "Inode %llu, inode i_size = %lld != di "
238 "i_size = %llu, i_flags = 0x%x\n",
239 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
241 (unsigned long long)le64_to_cpu(fe
->i_size
),
242 le32_to_cpu(fe
->i_flags
));
244 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
245 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
246 (unsigned long long)le64_to_cpu(fe
->i_size
),
247 (unsigned long long)new_i_size
);
253 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
254 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
255 (unsigned long long)le64_to_cpu(fe
->i_size
),
256 (unsigned long long)new_i_size
);
258 /* lets handle the simple truncate cases before doing any more
259 * cluster locking. */
260 if (new_i_size
== le64_to_cpu(fe
->i_size
))
263 /* This forces other nodes to sync and drop their pages. Do
264 * this even if we have a truncate without allocation change -
265 * ocfs2 cluster sizes can be much greater than page size, so
266 * we have to truncate them anyway. */
267 status
= ocfs2_data_lock(inode
, 1);
272 ocfs2_data_unlock(inode
, 1);
274 if (le32_to_cpu(fe
->i_clusters
) ==
275 ocfs2_clusters_for_bytes(osb
->sb
, new_i_size
)) {
276 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
278 /* No allocation change is required, so lets fast path
280 status
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
286 /* alright, we're going to need to do a full blown alloc size
287 * change. Orphan the inode so that recovery can complete the
288 * truncate if necessary. This does the task of marking
290 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
296 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
302 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
308 /* TODO: orphan dir cleanup here. */
316 * extend allocation only here.
317 * we'll update all the disk stuff, and oip->alloc_size
319 * expect stuff to be locked, a transaction started and enough data /
320 * metadata reservations in the contexts.
322 * Will return -EAGAIN, and a reason if a restart is needed.
323 * If passed in, *reason will always be set, even in error.
325 int ocfs2_do_extend_allocation(struct ocfs2_super
*osb
,
328 struct buffer_head
*fe_bh
,
329 struct ocfs2_journal_handle
*handle
,
330 struct ocfs2_alloc_context
*data_ac
,
331 struct ocfs2_alloc_context
*meta_ac
,
332 enum ocfs2_alloc_restarted
*reason_ret
)
336 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
337 enum ocfs2_alloc_restarted reason
= RESTART_NONE
;
338 u32 bit_off
, num_bits
;
341 BUG_ON(!clusters_to_add
);
343 free_extents
= ocfs2_num_free_extents(osb
, inode
, fe
);
344 if (free_extents
< 0) {
345 status
= free_extents
;
350 /* there are two cases which could cause us to EAGAIN in the
351 * we-need-more-metadata case:
352 * 1) we haven't reserved *any*
353 * 2) we are so fragmented, we've needed to add metadata too
355 if (!free_extents
&& !meta_ac
) {
356 mlog(0, "we haven't reserved any metadata!\n");
358 reason
= RESTART_META
;
360 } else if ((!free_extents
)
361 && (ocfs2_alloc_context_bits_left(meta_ac
)
362 < ocfs2_extend_meta_needed(fe
))) {
363 mlog(0, "filesystem is really fragmented...\n");
365 reason
= RESTART_META
;
369 status
= ocfs2_claim_clusters(osb
, handle
, data_ac
, 1,
370 &bit_off
, &num_bits
);
372 if (status
!= -ENOSPC
)
377 BUG_ON(num_bits
> clusters_to_add
);
379 /* reserve our write early -- insert_extent may update the inode */
380 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
381 OCFS2_JOURNAL_ACCESS_WRITE
);
387 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
388 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
389 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
390 status
= ocfs2_insert_extent(osb
, handle
, inode
, fe_bh
, block
,
397 le32_add_cpu(&fe
->i_clusters
, num_bits
);
398 spin_lock(&OCFS2_I(inode
)->ip_lock
);
399 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
400 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
402 status
= ocfs2_journal_dirty(handle
, fe_bh
);
408 clusters_to_add
-= num_bits
;
410 if (clusters_to_add
) {
411 mlog(0, "need to alloc once more, clusters = %u, wanted = "
412 "%u\n", fe
->i_clusters
, clusters_to_add
);
414 reason
= RESTART_TRANS
;
420 *reason_ret
= reason
;
424 static int ocfs2_extend_allocation(struct inode
*inode
,
428 int restart_func
= 0;
429 int drop_alloc_sem
= 0;
430 int credits
, num_free_extents
;
432 struct buffer_head
*bh
= NULL
;
433 struct ocfs2_dinode
*fe
= NULL
;
434 struct ocfs2_journal_handle
*handle
= NULL
;
435 struct ocfs2_alloc_context
*data_ac
= NULL
;
436 struct ocfs2_alloc_context
*meta_ac
= NULL
;
437 enum ocfs2_alloc_restarted why
;
438 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
440 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
442 status
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &bh
,
443 OCFS2_BH_CACHED
, inode
);
449 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
450 if (!OCFS2_IS_VALID_DINODE(fe
)) {
451 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
457 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
459 mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
460 "clusters_to_add = %u\n",
461 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, i_size_read(inode
),
462 fe
->i_clusters
, clusters_to_add
);
464 handle
= ocfs2_alloc_handle(osb
);
465 if (handle
== NULL
) {
471 num_free_extents
= ocfs2_num_free_extents(osb
,
474 if (num_free_extents
< 0) {
475 status
= num_free_extents
;
480 if (!num_free_extents
) {
481 status
= ocfs2_reserve_new_metadata(osb
,
486 if (status
!= -ENOSPC
)
492 status
= ocfs2_reserve_clusters(osb
,
497 if (status
!= -ENOSPC
)
502 /* blocks peope in read/write from reading our allocation
503 * until we're done changing it. We depend on i_mutex to block
504 * other extend/truncate calls while we're here. Ordering wrt
505 * start_trans is important here -- always do it before! */
506 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
509 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
510 handle
= ocfs2_start_trans(osb
, handle
, credits
);
511 if (IS_ERR(handle
)) {
512 status
= PTR_ERR(handle
);
518 restarted_transaction
:
519 /* reserve a write to the file entry early on - that we if we
520 * run out of credits in the allocation path, we can still
522 status
= ocfs2_journal_access(handle
, inode
, bh
,
523 OCFS2_JOURNAL_ACCESS_WRITE
);
529 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
531 status
= ocfs2_do_extend_allocation(osb
,
539 if ((status
< 0) && (status
!= -EAGAIN
)) {
540 if (status
!= -ENOSPC
)
545 status
= ocfs2_journal_dirty(handle
, bh
);
551 spin_lock(&OCFS2_I(inode
)->ip_lock
);
552 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
553 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
555 if (why
!= RESTART_NONE
&& clusters_to_add
) {
556 if (why
== RESTART_META
) {
557 mlog(0, "restarting function.\n");
560 BUG_ON(why
!= RESTART_TRANS
);
562 mlog(0, "restarting transaction.\n");
563 /* TODO: This can be more intelligent. */
564 credits
= ocfs2_calc_extend_credits(osb
->sb
,
567 status
= ocfs2_extend_trans(handle
, credits
);
569 /* handle still has to be committed at
575 goto restarted_transaction
;
579 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
580 fe
->i_clusters
, (unsigned long long)fe
->i_size
);
581 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
582 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
585 if (drop_alloc_sem
) {
586 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
590 ocfs2_commit_trans(handle
);
594 ocfs2_free_alloc_context(data_ac
);
598 ocfs2_free_alloc_context(meta_ac
);
601 if ((!status
) && restart_func
) {
614 /* Some parts of this taken from generic_cont_expand, which turned out
615 * to be too fragile to do exactly what we need without us having to
616 * worry about recursive locking in ->commit_write(). */
617 static int ocfs2_write_zero_page(struct inode
*inode
,
620 struct address_space
*mapping
= inode
->i_mapping
;
624 struct ocfs2_journal_handle
*handle
= NULL
;
627 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
628 /* ugh. in prepare/commit_write, if from==to==start of block, we
629 ** skip the prepare. make sure we never send an offset for the start
632 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
635 index
= size
>> PAGE_CACHE_SHIFT
;
637 page
= grab_cache_page(mapping
, index
);
644 ret
= ocfs2_prepare_write(NULL
, page
, offset
, offset
);
650 if (ocfs2_should_order_data(inode
)) {
651 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
653 if (IS_ERR(handle
)) {
654 ret
= PTR_ERR(handle
);
660 /* must not update i_size! */
661 ret
= block_commit_write(page
, offset
, offset
);
668 ocfs2_commit_trans(handle
);
671 page_cache_release(page
);
676 static int ocfs2_zero_extend(struct inode
*inode
,
681 struct super_block
*sb
= inode
->i_sb
;
683 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
684 while (start_off
< zero_to_size
) {
685 ret
= ocfs2_write_zero_page(inode
, start_off
);
691 start_off
+= sb
->s_blocksize
;
698 static int ocfs2_extend_file(struct inode
*inode
,
699 struct buffer_head
*di_bh
,
705 /* setattr sometimes calls us like this. */
709 if (i_size_read(inode
) == new_i_size
)
711 BUG_ON(new_i_size
< i_size_read(inode
));
713 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
714 OCFS2_I(inode
)->ip_clusters
;
716 if (clusters_to_add
) {
717 ret
= ocfs2_extend_allocation(inode
, clusters_to_add
);
723 ret
= ocfs2_zero_extend(inode
, new_i_size
);
730 /* No allocation required, we just use this helper to
731 * do a trivial update of i_size. */
732 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
742 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
744 int status
= 0, size_change
;
745 struct inode
*inode
= dentry
->d_inode
;
746 struct super_block
*sb
= inode
->i_sb
;
747 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
748 struct buffer_head
*bh
= NULL
;
749 struct ocfs2_journal_handle
*handle
= NULL
;
751 mlog_entry("(0x%p, '%.*s')\n", dentry
,
752 dentry
->d_name
.len
, dentry
->d_name
.name
);
754 if (attr
->ia_valid
& ATTR_MODE
)
755 mlog(0, "mode change: %d\n", attr
->ia_mode
);
756 if (attr
->ia_valid
& ATTR_UID
)
757 mlog(0, "uid change: %d\n", attr
->ia_uid
);
758 if (attr
->ia_valid
& ATTR_GID
)
759 mlog(0, "gid change: %d\n", attr
->ia_gid
);
760 if (attr
->ia_valid
& ATTR_SIZE
)
761 mlog(0, "size change...\n");
762 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
763 mlog(0, "time change...\n");
765 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
766 | ATTR_GID | ATTR_UID | ATTR_MODE)
767 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
768 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
772 status
= inode_change_ok(inode
, attr
);
776 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
778 status
= ocfs2_rw_lock(inode
, 1);
785 status
= ocfs2_meta_lock(inode
, NULL
, &bh
, 1);
787 if (status
!= -ENOENT
)
792 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
793 if (i_size_read(inode
) > attr
->ia_size
)
794 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
796 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
);
798 if (status
!= -ENOSPC
)
805 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
806 if (IS_ERR(handle
)) {
807 status
= PTR_ERR(handle
);
812 status
= inode_setattr(inode
, attr
);
818 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
823 ocfs2_commit_trans(handle
);
825 ocfs2_meta_unlock(inode
, 1);
828 ocfs2_rw_unlock(inode
, 1);
837 int ocfs2_getattr(struct vfsmount
*mnt
,
838 struct dentry
*dentry
,
841 struct inode
*inode
= dentry
->d_inode
;
842 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
843 struct ocfs2_super
*osb
= sb
->s_fs_info
;
848 err
= ocfs2_inode_revalidate(dentry
);
855 generic_fillattr(inode
, stat
);
857 /* We set the blksize from the cluster size for performance */
858 stat
->blksize
= osb
->s_clustersize
;
866 static int ocfs2_write_remove_suid(struct inode
*inode
)
869 struct buffer_head
*bh
= NULL
;
870 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
871 struct ocfs2_journal_handle
*handle
;
872 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
873 struct ocfs2_dinode
*di
;
875 mlog_entry("(Inode %llu, mode 0%o)\n",
876 (unsigned long long)oi
->ip_blkno
, inode
->i_mode
);
878 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
879 if (handle
== NULL
) {
885 ret
= ocfs2_read_block(osb
, oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
891 ret
= ocfs2_journal_access(handle
, inode
, bh
,
892 OCFS2_JOURNAL_ACCESS_WRITE
);
898 inode
->i_mode
&= ~S_ISUID
;
899 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
900 inode
->i_mode
&= ~S_ISGID
;
902 di
= (struct ocfs2_dinode
*) bh
->b_data
;
903 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
905 ret
= ocfs2_journal_dirty(handle
, bh
);
911 ocfs2_commit_trans(handle
);
917 static inline int ocfs2_write_should_remove_suid(struct inode
*inode
)
919 mode_t mode
= inode
->i_mode
;
921 if (!capable(CAP_FSETID
)) {
922 if (unlikely(mode
& S_ISUID
))
925 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
931 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
932 const char __user
*buf
,
936 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
938 int ret
, rw_level
= -1, meta_level
= -1, have_alloc_sem
= 0;
940 struct file
*filp
= iocb
->ki_filp
;
941 struct inode
*inode
= filp
->f_dentry
->d_inode
;
942 loff_t newsize
, saved_pos
;
944 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
946 filp
->f_dentry
->d_name
.len
,
947 filp
->f_dentry
->d_name
.name
);
949 /* happy write of zero bytes */
954 mlog(0, "bad inode\n");
958 mutex_lock(&inode
->i_mutex
);
959 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
960 if (filp
->f_flags
& O_DIRECT
) {
962 down_read(&inode
->i_alloc_sem
);
965 /* concurrent O_DIRECT writes are allowed */
966 rw_level
= (filp
->f_flags
& O_DIRECT
) ? 0 : 1;
967 ret
= ocfs2_rw_lock(inode
, rw_level
);
975 * We sample i_size under a read level meta lock to see if our write
976 * is extending the file, if it is we back off and get a write level
979 meta_level
= (filp
->f_flags
& O_APPEND
) ? 1 : 0;
981 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, meta_level
);
988 /* Clear suid / sgid if necessary. We do this here
989 * instead of later in the write path because
990 * remove_suid() calls ->setattr without any hint that
991 * we may have already done our cluster locking. Since
992 * ocfs2_setattr() *must* take cluster locks to
993 * proceeed, this will lead us to recursively lock the
994 * inode. There's also the dinode i_size state which
995 * can be lost via setattr during extending writes (we
996 * set inode->i_size at the end of a write. */
997 if (ocfs2_write_should_remove_suid(inode
)) {
998 if (meta_level
== 0) {
999 ocfs2_meta_unlock(inode
, meta_level
);
1004 ret
= ocfs2_write_remove_suid(inode
);
1011 /* work on a copy of ppos until we're sure that we won't have
1012 * to recalculate it due to relocking. */
1013 if (filp
->f_flags
& O_APPEND
) {
1014 saved_pos
= i_size_read(inode
);
1015 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1017 saved_pos
= iocb
->ki_pos
;
1019 newsize
= count
+ saved_pos
;
1021 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1022 (long long) saved_pos
, (long long) newsize
,
1023 (long long) i_size_read(inode
));
1025 /* No need for a higher level metadata lock if we're
1026 * never going past i_size. */
1027 if (newsize
<= i_size_read(inode
))
1030 if (meta_level
== 0) {
1031 ocfs2_meta_unlock(inode
, meta_level
);
1036 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1037 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1038 OCFS2_I(inode
)->ip_clusters
;
1039 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1041 mlog(0, "Writing at EOF, may need more allocation: "
1042 "i_size = %lld, newsize = %lld, need %u clusters\n",
1043 (long long) i_size_read(inode
), (long long) newsize
,
1046 /* We only want to continue the rest of this loop if
1047 * our extend will actually require more
1052 ret
= ocfs2_extend_allocation(inode
, clusters
);
1059 /* Fill any holes which would've been created by this
1060 * write. If we're O_APPEND, this will wind up
1061 * (correctly) being a noop. */
1062 ret
= ocfs2_zero_extend(inode
, (u64
) newsize
- count
);
1070 /* ok, we're done with i_size and alloc work */
1071 iocb
->ki_pos
= saved_pos
;
1072 ocfs2_meta_unlock(inode
, meta_level
);
1075 /* communicate with ocfs2_dio_end_io */
1076 ocfs2_iocb_set_rw_locked(iocb
);
1078 ret
= generic_file_aio_write_nolock(iocb
, &local_iov
, 1, &iocb
->ki_pos
);
1080 /* buffered aio wouldn't have proper lock coverage today */
1081 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1084 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1085 * function pointer which is called when o_direct io completes so that
1086 * it can unlock our rw lock. (it's the clustered equivalent of
1087 * i_alloc_sem; protects truncate from racing with pending ios).
1088 * Unfortunately there are error cases which call end_io and others
1089 * that don't. so we don't have to unlock the rw_lock if either an
1090 * async dio is going to do it in the future or an end_io after an
1091 * error has already done it.
1093 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1099 if (meta_level
!= -1)
1100 ocfs2_meta_unlock(inode
, meta_level
);
1102 up_read(&inode
->i_alloc_sem
);
1104 ocfs2_rw_unlock(inode
, rw_level
);
1105 mutex_unlock(&inode
->i_mutex
);
1111 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
1116 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0;
1117 struct file
*filp
= iocb
->ki_filp
;
1118 struct inode
*inode
= filp
->f_dentry
->d_inode
;
1120 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
1121 (unsigned int)count
,
1122 filp
->f_dentry
->d_name
.len
,
1123 filp
->f_dentry
->d_name
.name
);
1132 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1133 * need locks to protect pending reads from racing with truncate.
1135 if (filp
->f_flags
& O_DIRECT
) {
1136 down_read(&inode
->i_alloc_sem
);
1139 ret
= ocfs2_rw_lock(inode
, 0);
1145 /* communicate with ocfs2_dio_end_io */
1146 ocfs2_iocb_set_rw_locked(iocb
);
1149 ret
= generic_file_aio_read(iocb
, buf
, count
, iocb
->ki_pos
);
1151 mlog(ML_ERROR
, "generic_file_aio_read returned -EINVAL\n");
1153 /* buffered aio wouldn't have proper lock coverage today */
1154 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1156 /* see ocfs2_file_aio_write */
1157 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1164 up_read(&inode
->i_alloc_sem
);
1166 ocfs2_rw_unlock(inode
, rw_level
);
1172 struct inode_operations ocfs2_file_iops
= {
1173 .setattr
= ocfs2_setattr
,
1174 .getattr
= ocfs2_getattr
,
1177 struct inode_operations ocfs2_special_file_iops
= {
1178 .setattr
= ocfs2_setattr
,
1179 .getattr
= ocfs2_getattr
,
1182 const struct file_operations ocfs2_fops
= {
1183 .read
= do_sync_read
,
1184 .write
= do_sync_write
,
1185 .sendfile
= generic_file_sendfile
,
1187 .fsync
= ocfs2_sync_file
,
1188 .release
= ocfs2_file_release
,
1189 .open
= ocfs2_file_open
,
1190 .aio_read
= ocfs2_file_aio_read
,
1191 .aio_write
= ocfs2_file_aio_write
,
1194 const struct file_operations ocfs2_dops
= {
1195 .read
= generic_read_dir
,
1196 .readdir
= ocfs2_readdir
,
1197 .fsync
= ocfs2_sync_file
,