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"
53 #include "buffer_head_io.h"
55 static int ocfs2_sync_inode(struct inode
*inode
)
57 filemap_fdatawrite(inode
->i_mapping
);
58 return sync_mapping_buffers(inode
->i_mapping
);
61 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
64 int mode
= file
->f_flags
;
65 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
67 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
68 file
->f_dentry
->d_name
.len
, file
->f_dentry
->d_name
.name
);
70 spin_lock(&oi
->ip_lock
);
72 /* Check that the inode hasn't been wiped from disk by another
73 * node. If it hasn't then we're safe as long as we hold the
74 * spin lock until our increment of open count. */
75 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
76 spin_unlock(&oi
->ip_lock
);
83 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
86 spin_unlock(&oi
->ip_lock
);
93 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
95 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
97 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
98 file
->f_dentry
->d_name
.len
,
99 file
->f_dentry
->d_name
.name
);
101 spin_lock(&oi
->ip_lock
);
102 if (!--oi
->ip_open_count
)
103 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
104 spin_unlock(&oi
->ip_lock
);
111 static int ocfs2_sync_file(struct file
*file
,
112 struct dentry
*dentry
,
117 struct inode
*inode
= dentry
->d_inode
;
118 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
120 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
121 dentry
->d_name
.len
, dentry
->d_name
.name
);
123 err
= ocfs2_sync_inode(dentry
->d_inode
);
127 journal
= osb
->journal
->j_journal
;
128 err
= journal_force_commit(journal
);
133 return (err
< 0) ? -EIO
: 0;
136 int ocfs2_set_inode_size(struct ocfs2_journal_handle
*handle
,
138 struct buffer_head
*fe_bh
,
144 i_size_write(inode
, new_i_size
);
145 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(new_i_size
);
146 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
148 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
159 static int ocfs2_simple_size_update(struct inode
*inode
,
160 struct buffer_head
*di_bh
,
164 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
165 struct ocfs2_journal_handle
*handle
= NULL
;
167 handle
= ocfs2_start_trans(osb
, NULL
,
168 OCFS2_INODE_UPDATE_CREDITS
);
169 if (handle
== NULL
) {
175 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
180 ocfs2_commit_trans(handle
);
185 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
187 struct buffer_head
*fe_bh
,
191 struct ocfs2_journal_handle
*handle
;
195 /* TODO: This needs to actually orphan the inode in this
198 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
199 if (IS_ERR(handle
)) {
200 status
= PTR_ERR(handle
);
205 status
= ocfs2_set_inode_size(handle
, inode
, fe_bh
, new_i_size
);
209 ocfs2_commit_trans(handle
);
215 static int ocfs2_truncate_file(struct inode
*inode
,
216 struct buffer_head
*di_bh
,
220 struct ocfs2_dinode
*fe
= NULL
;
221 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
222 struct ocfs2_truncate_context
*tc
= NULL
;
224 mlog_entry("(inode = %llu, new_i_size = %llu\n",
225 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
226 (unsigned long long)new_i_size
);
228 truncate_inode_pages(inode
->i_mapping
, new_i_size
);
230 fe
= (struct ocfs2_dinode
*) di_bh
->b_data
;
231 if (!OCFS2_IS_VALID_DINODE(fe
)) {
232 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
237 mlog_bug_on_msg(le64_to_cpu(fe
->i_size
) != i_size_read(inode
),
238 "Inode %llu, inode i_size = %lld != di "
239 "i_size = %llu, i_flags = 0x%x\n",
240 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
242 (unsigned long long)le64_to_cpu(fe
->i_size
),
243 le32_to_cpu(fe
->i_flags
));
245 if (new_i_size
> le64_to_cpu(fe
->i_size
)) {
246 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
247 (unsigned long long)le64_to_cpu(fe
->i_size
),
248 (unsigned long long)new_i_size
);
254 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
255 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
256 (unsigned long long)le64_to_cpu(fe
->i_size
),
257 (unsigned long long)new_i_size
);
259 /* lets handle the simple truncate cases before doing any more
260 * cluster locking. */
261 if (new_i_size
== le64_to_cpu(fe
->i_size
))
264 /* This forces other nodes to sync and drop their pages. Do
265 * this even if we have a truncate without allocation change -
266 * ocfs2 cluster sizes can be much greater than page size, so
267 * we have to truncate them anyway. */
268 status
= ocfs2_data_lock(inode
, 1);
273 ocfs2_data_unlock(inode
, 1);
275 if (le32_to_cpu(fe
->i_clusters
) ==
276 ocfs2_clusters_for_bytes(osb
->sb
, new_i_size
)) {
277 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
279 /* No allocation change is required, so lets fast path
281 status
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
287 /* alright, we're going to need to do a full blown alloc size
288 * change. Orphan the inode so that recovery can complete the
289 * truncate if necessary. This does the task of marking
291 status
= ocfs2_orphan_for_truncate(osb
, inode
, di_bh
, new_i_size
);
297 status
= ocfs2_prepare_truncate(osb
, inode
, di_bh
, &tc
);
303 status
= ocfs2_commit_truncate(osb
, inode
, di_bh
, tc
);
309 /* TODO: orphan dir cleanup here. */
317 * extend allocation only here.
318 * we'll update all the disk stuff, and oip->alloc_size
320 * expect stuff to be locked, a transaction started and enough data /
321 * metadata reservations in the contexts.
323 * Will return -EAGAIN, and a reason if a restart is needed.
324 * If passed in, *reason will always be set, even in error.
326 int ocfs2_do_extend_allocation(struct ocfs2_super
*osb
,
329 struct buffer_head
*fe_bh
,
330 struct ocfs2_journal_handle
*handle
,
331 struct ocfs2_alloc_context
*data_ac
,
332 struct ocfs2_alloc_context
*meta_ac
,
333 enum ocfs2_alloc_restarted
*reason_ret
)
337 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
338 enum ocfs2_alloc_restarted reason
= RESTART_NONE
;
339 u32 bit_off
, num_bits
;
342 BUG_ON(!clusters_to_add
);
344 free_extents
= ocfs2_num_free_extents(osb
, inode
, fe
);
345 if (free_extents
< 0) {
346 status
= free_extents
;
351 /* there are two cases which could cause us to EAGAIN in the
352 * we-need-more-metadata case:
353 * 1) we haven't reserved *any*
354 * 2) we are so fragmented, we've needed to add metadata too
356 if (!free_extents
&& !meta_ac
) {
357 mlog(0, "we haven't reserved any metadata!\n");
359 reason
= RESTART_META
;
361 } else if ((!free_extents
)
362 && (ocfs2_alloc_context_bits_left(meta_ac
)
363 < ocfs2_extend_meta_needed(fe
))) {
364 mlog(0, "filesystem is really fragmented...\n");
366 reason
= RESTART_META
;
370 status
= ocfs2_claim_clusters(osb
, handle
, data_ac
, 1,
371 &bit_off
, &num_bits
);
373 if (status
!= -ENOSPC
)
378 BUG_ON(num_bits
> clusters_to_add
);
380 /* reserve our write early -- insert_extent may update the inode */
381 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
382 OCFS2_JOURNAL_ACCESS_WRITE
);
388 block
= ocfs2_clusters_to_blocks(osb
->sb
, bit_off
);
389 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
390 num_bits
, bit_off
, (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
391 status
= ocfs2_insert_extent(osb
, handle
, inode
, fe_bh
, block
,
398 le32_add_cpu(&fe
->i_clusters
, num_bits
);
399 spin_lock(&OCFS2_I(inode
)->ip_lock
);
400 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
401 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
403 status
= ocfs2_journal_dirty(handle
, fe_bh
);
409 clusters_to_add
-= num_bits
;
411 if (clusters_to_add
) {
412 mlog(0, "need to alloc once more, clusters = %u, wanted = "
413 "%u\n", fe
->i_clusters
, clusters_to_add
);
415 reason
= RESTART_TRANS
;
421 *reason_ret
= reason
;
425 static int ocfs2_extend_allocation(struct inode
*inode
,
429 int restart_func
= 0;
430 int drop_alloc_sem
= 0;
431 int credits
, num_free_extents
;
433 struct buffer_head
*bh
= NULL
;
434 struct ocfs2_dinode
*fe
= NULL
;
435 struct ocfs2_journal_handle
*handle
= NULL
;
436 struct ocfs2_alloc_context
*data_ac
= NULL
;
437 struct ocfs2_alloc_context
*meta_ac
= NULL
;
438 enum ocfs2_alloc_restarted why
;
439 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
441 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add
);
443 status
= ocfs2_read_block(osb
, OCFS2_I(inode
)->ip_blkno
, &bh
,
444 OCFS2_BH_CACHED
, inode
);
450 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
451 if (!OCFS2_IS_VALID_DINODE(fe
)) {
452 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
458 BUG_ON(le32_to_cpu(fe
->i_clusters
) != OCFS2_I(inode
)->ip_clusters
);
460 mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
461 "clusters_to_add = %u\n",
462 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, i_size_read(inode
),
463 fe
->i_clusters
, clusters_to_add
);
465 handle
= ocfs2_alloc_handle(osb
);
466 if (handle
== NULL
) {
472 num_free_extents
= ocfs2_num_free_extents(osb
,
475 if (num_free_extents
< 0) {
476 status
= num_free_extents
;
481 if (!num_free_extents
) {
482 status
= ocfs2_reserve_new_metadata(osb
,
487 if (status
!= -ENOSPC
)
493 status
= ocfs2_reserve_clusters(osb
,
498 if (status
!= -ENOSPC
)
503 /* blocks peope in read/write from reading our allocation
504 * until we're done changing it. We depend on i_mutex to block
505 * other extend/truncate calls while we're here. Ordering wrt
506 * start_trans is important here -- always do it before! */
507 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
510 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
511 handle
= ocfs2_start_trans(osb
, handle
, credits
);
512 if (IS_ERR(handle
)) {
513 status
= PTR_ERR(handle
);
519 restarted_transaction
:
520 /* reserve a write to the file entry early on - that we if we
521 * run out of credits in the allocation path, we can still
523 status
= ocfs2_journal_access(handle
, inode
, bh
,
524 OCFS2_JOURNAL_ACCESS_WRITE
);
530 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
532 status
= ocfs2_do_extend_allocation(osb
,
540 if ((status
< 0) && (status
!= -EAGAIN
)) {
541 if (status
!= -ENOSPC
)
546 status
= ocfs2_journal_dirty(handle
, bh
);
552 spin_lock(&OCFS2_I(inode
)->ip_lock
);
553 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
554 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
556 if (why
!= RESTART_NONE
&& clusters_to_add
) {
557 if (why
== RESTART_META
) {
558 mlog(0, "restarting function.\n");
561 BUG_ON(why
!= RESTART_TRANS
);
563 mlog(0, "restarting transaction.\n");
564 /* TODO: This can be more intelligent. */
565 credits
= ocfs2_calc_extend_credits(osb
->sb
,
568 status
= ocfs2_extend_trans(handle
, credits
);
570 /* handle still has to be committed at
576 goto restarted_transaction
;
580 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
581 fe
->i_clusters
, (unsigned long long)fe
->i_size
);
582 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
583 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
586 if (drop_alloc_sem
) {
587 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
591 ocfs2_commit_trans(handle
);
595 ocfs2_free_alloc_context(data_ac
);
599 ocfs2_free_alloc_context(meta_ac
);
602 if ((!status
) && restart_func
) {
615 /* Some parts of this taken from generic_cont_expand, which turned out
616 * to be too fragile to do exactly what we need without us having to
617 * worry about recursive locking in ->prepare_write() and
618 * ->commit_write(). */
619 static int ocfs2_write_zero_page(struct inode
*inode
,
622 struct address_space
*mapping
= inode
->i_mapping
;
626 struct ocfs2_journal_handle
*handle
= NULL
;
629 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
630 /* ugh. in prepare/commit_write, if from==to==start of block, we
631 ** skip the prepare. make sure we never send an offset for the start
634 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
637 index
= size
>> PAGE_CACHE_SHIFT
;
639 page
= grab_cache_page(mapping
, index
);
646 ret
= ocfs2_prepare_write_nolock(inode
, page
, offset
, offset
);
652 if (ocfs2_should_order_data(inode
)) {
653 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
655 if (IS_ERR(handle
)) {
656 ret
= PTR_ERR(handle
);
662 /* must not update i_size! */
663 ret
= block_commit_write(page
, offset
, offset
);
670 ocfs2_commit_trans(handle
);
673 page_cache_release(page
);
678 static int ocfs2_zero_extend(struct inode
*inode
,
683 struct super_block
*sb
= inode
->i_sb
;
685 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
686 while (start_off
< zero_to_size
) {
687 ret
= ocfs2_write_zero_page(inode
, start_off
);
693 start_off
+= sb
->s_blocksize
;
701 * A tail_to_skip value > 0 indicates that we're being called from
702 * ocfs2_file_aio_write(). This has the following implications:
704 * - we don't want to update i_size
705 * - di_bh will be NULL, which is fine because it's only used in the
706 * case where we want to update i_size.
707 * - ocfs2_zero_extend() will then only be filling the hole created
708 * between i_size and the start of the write.
710 static int ocfs2_extend_file(struct inode
*inode
,
711 struct buffer_head
*di_bh
,
718 BUG_ON(!tail_to_skip
&& !di_bh
);
720 /* setattr sometimes calls us like this. */
724 if (i_size_read(inode
) == new_i_size
)
726 BUG_ON(new_i_size
< i_size_read(inode
));
728 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
729 OCFS2_I(inode
)->ip_clusters
;
731 if (clusters_to_add
) {
733 * protect the pages that ocfs2_zero_extend is going to
734 * be pulling into the page cache.. we do this before the
735 * metadata extend so that we don't get into the situation
736 * where we've extended the metadata but can't get the data
739 ret
= ocfs2_data_lock(inode
, 1);
745 ret
= ocfs2_extend_allocation(inode
, clusters_to_add
);
751 ret
= ocfs2_zero_extend(inode
, (u64
)new_i_size
- tail_to_skip
);
759 /* We're being called from ocfs2_setattr() which wants
760 * us to update i_size */
761 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
767 if (clusters_to_add
) /* this is the only case in which we lock */
768 ocfs2_data_unlock(inode
, 1);
774 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
776 int status
= 0, size_change
;
777 struct inode
*inode
= dentry
->d_inode
;
778 struct super_block
*sb
= inode
->i_sb
;
779 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
780 struct buffer_head
*bh
= NULL
;
781 struct ocfs2_journal_handle
*handle
= NULL
;
783 mlog_entry("(0x%p, '%.*s')\n", dentry
,
784 dentry
->d_name
.len
, dentry
->d_name
.name
);
786 if (attr
->ia_valid
& ATTR_MODE
)
787 mlog(0, "mode change: %d\n", attr
->ia_mode
);
788 if (attr
->ia_valid
& ATTR_UID
)
789 mlog(0, "uid change: %d\n", attr
->ia_uid
);
790 if (attr
->ia_valid
& ATTR_GID
)
791 mlog(0, "gid change: %d\n", attr
->ia_gid
);
792 if (attr
->ia_valid
& ATTR_SIZE
)
793 mlog(0, "size change...\n");
794 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
795 mlog(0, "time change...\n");
797 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
798 | ATTR_GID | ATTR_UID | ATTR_MODE)
799 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
800 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
804 status
= inode_change_ok(inode
, attr
);
808 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
810 status
= ocfs2_rw_lock(inode
, 1);
817 status
= ocfs2_meta_lock(inode
, NULL
, &bh
, 1);
819 if (status
!= -ENOENT
)
824 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
825 if (i_size_read(inode
) > attr
->ia_size
)
826 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
828 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
, 0);
830 if (status
!= -ENOSPC
)
837 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
838 if (IS_ERR(handle
)) {
839 status
= PTR_ERR(handle
);
844 status
= inode_setattr(inode
, attr
);
850 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
855 ocfs2_commit_trans(handle
);
857 ocfs2_meta_unlock(inode
, 1);
860 ocfs2_rw_unlock(inode
, 1);
869 int ocfs2_getattr(struct vfsmount
*mnt
,
870 struct dentry
*dentry
,
873 struct inode
*inode
= dentry
->d_inode
;
874 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
875 struct ocfs2_super
*osb
= sb
->s_fs_info
;
880 err
= ocfs2_inode_revalidate(dentry
);
887 generic_fillattr(inode
, stat
);
889 /* We set the blksize from the cluster size for performance */
890 stat
->blksize
= osb
->s_clustersize
;
898 static int ocfs2_write_remove_suid(struct inode
*inode
)
901 struct buffer_head
*bh
= NULL
;
902 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
903 struct ocfs2_journal_handle
*handle
;
904 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
905 struct ocfs2_dinode
*di
;
907 mlog_entry("(Inode %llu, mode 0%o)\n",
908 (unsigned long long)oi
->ip_blkno
, inode
->i_mode
);
910 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_INODE_UPDATE_CREDITS
);
911 if (handle
== NULL
) {
917 ret
= ocfs2_read_block(osb
, oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
923 ret
= ocfs2_journal_access(handle
, inode
, bh
,
924 OCFS2_JOURNAL_ACCESS_WRITE
);
930 inode
->i_mode
&= ~S_ISUID
;
931 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
932 inode
->i_mode
&= ~S_ISGID
;
934 di
= (struct ocfs2_dinode
*) bh
->b_data
;
935 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
937 ret
= ocfs2_journal_dirty(handle
, bh
);
943 ocfs2_commit_trans(handle
);
949 static inline int ocfs2_write_should_remove_suid(struct inode
*inode
)
951 mode_t mode
= inode
->i_mode
;
953 if (!capable(CAP_FSETID
)) {
954 if (unlikely(mode
& S_ISUID
))
957 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
963 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
964 const char __user
*buf
,
968 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
,
970 int ret
, rw_level
= -1, meta_level
= -1, have_alloc_sem
= 0;
972 struct file
*filp
= iocb
->ki_filp
;
973 struct inode
*inode
= filp
->f_dentry
->d_inode
;
974 loff_t newsize
, saved_pos
;
976 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
978 filp
->f_dentry
->d_name
.len
,
979 filp
->f_dentry
->d_name
.name
);
981 /* happy write of zero bytes */
986 mlog(0, "bad inode\n");
990 mutex_lock(&inode
->i_mutex
);
991 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
992 if (filp
->f_flags
& O_DIRECT
) {
994 down_read(&inode
->i_alloc_sem
);
997 /* concurrent O_DIRECT writes are allowed */
998 rw_level
= (filp
->f_flags
& O_DIRECT
) ? 0 : 1;
999 ret
= ocfs2_rw_lock(inode
, rw_level
);
1007 * We sample i_size under a read level meta lock to see if our write
1008 * is extending the file, if it is we back off and get a write level
1011 meta_level
= (filp
->f_flags
& O_APPEND
) ? 1 : 0;
1013 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, meta_level
);
1020 /* Clear suid / sgid if necessary. We do this here
1021 * instead of later in the write path because
1022 * remove_suid() calls ->setattr without any hint that
1023 * we may have already done our cluster locking. Since
1024 * ocfs2_setattr() *must* take cluster locks to
1025 * proceeed, this will lead us to recursively lock the
1026 * inode. There's also the dinode i_size state which
1027 * can be lost via setattr during extending writes (we
1028 * set inode->i_size at the end of a write. */
1029 if (ocfs2_write_should_remove_suid(inode
)) {
1030 if (meta_level
== 0) {
1031 ocfs2_meta_unlock(inode
, meta_level
);
1036 ret
= ocfs2_write_remove_suid(inode
);
1043 /* work on a copy of ppos until we're sure that we won't have
1044 * to recalculate it due to relocking. */
1045 if (filp
->f_flags
& O_APPEND
) {
1046 saved_pos
= i_size_read(inode
);
1047 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1049 saved_pos
= iocb
->ki_pos
;
1051 newsize
= count
+ saved_pos
;
1053 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1054 (long long) saved_pos
, (long long) newsize
,
1055 (long long) i_size_read(inode
));
1057 /* No need for a higher level metadata lock if we're
1058 * never going past i_size. */
1059 if (newsize
<= i_size_read(inode
))
1062 if (meta_level
== 0) {
1063 ocfs2_meta_unlock(inode
, meta_level
);
1068 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1069 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1070 OCFS2_I(inode
)->ip_clusters
;
1071 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1073 mlog(0, "Writing at EOF, may need more allocation: "
1074 "i_size = %lld, newsize = %lld, need %u clusters\n",
1075 (long long) i_size_read(inode
), (long long) newsize
,
1078 /* We only want to continue the rest of this loop if
1079 * our extend will actually require more
1084 ret
= ocfs2_extend_file(inode
, NULL
, newsize
, count
);
1093 /* ok, we're done with i_size and alloc work */
1094 iocb
->ki_pos
= saved_pos
;
1095 ocfs2_meta_unlock(inode
, meta_level
);
1098 /* communicate with ocfs2_dio_end_io */
1099 ocfs2_iocb_set_rw_locked(iocb
);
1101 ret
= generic_file_aio_write_nolock(iocb
, &local_iov
, 1, &iocb
->ki_pos
);
1103 /* buffered aio wouldn't have proper lock coverage today */
1104 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1107 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1108 * function pointer which is called when o_direct io completes so that
1109 * it can unlock our rw lock. (it's the clustered equivalent of
1110 * i_alloc_sem; protects truncate from racing with pending ios).
1111 * Unfortunately there are error cases which call end_io and others
1112 * that don't. so we don't have to unlock the rw_lock if either an
1113 * async dio is going to do it in the future or an end_io after an
1114 * error has already done it.
1116 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1122 if (meta_level
!= -1)
1123 ocfs2_meta_unlock(inode
, meta_level
);
1125 up_read(&inode
->i_alloc_sem
);
1127 ocfs2_rw_unlock(inode
, rw_level
);
1128 mutex_unlock(&inode
->i_mutex
);
1134 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
1139 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0;
1140 struct file
*filp
= iocb
->ki_filp
;
1141 struct inode
*inode
= filp
->f_dentry
->d_inode
;
1143 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp
, buf
,
1144 (unsigned int)count
,
1145 filp
->f_dentry
->d_name
.len
,
1146 filp
->f_dentry
->d_name
.name
);
1155 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1156 * need locks to protect pending reads from racing with truncate.
1158 if (filp
->f_flags
& O_DIRECT
) {
1159 down_read(&inode
->i_alloc_sem
);
1162 ret
= ocfs2_rw_lock(inode
, 0);
1168 /* communicate with ocfs2_dio_end_io */
1169 ocfs2_iocb_set_rw_locked(iocb
);
1173 * We're fine letting folks race truncates and extending
1174 * writes with read across the cluster, just like they can
1175 * locally. Hence no rw_lock during read.
1177 * Take and drop the meta data lock to update inode fields
1178 * like i_size. This allows the checks down below
1179 * generic_file_aio_read() a chance of actually working.
1181 ret
= ocfs2_meta_lock(inode
, NULL
, NULL
, 0);
1186 ocfs2_meta_unlock(inode
, 0);
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 const 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
,
1231 .ioctl
= ocfs2_ioctl
,
1234 const struct file_operations ocfs2_dops
= {
1235 .read
= generic_read_dir
,
1236 .readdir
= ocfs2_readdir
,
1237 .fsync
= ocfs2_sync_file
,
1238 .ioctl
= ocfs2_ioctl
,