1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * File open, close, extend, truncate
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
26 #include <linux/capability.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/pagemap.h>
32 #include <linux/uio.h>
33 #include <linux/sched.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
44 #include "extent_map.h"
54 #include "buffer_head_io.h"
56 static int ocfs2_sync_inode(struct inode
*inode
)
58 filemap_fdatawrite(inode
->i_mapping
);
59 return sync_mapping_buffers(inode
->i_mapping
);
62 static int ocfs2_file_open(struct inode
*inode
, struct file
*file
)
65 int mode
= file
->f_flags
;
66 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
68 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
69 file
->f_dentry
->d_name
.len
, file
->f_dentry
->d_name
.name
);
71 spin_lock(&oi
->ip_lock
);
73 /* Check that the inode hasn't been wiped from disk by another
74 * node. If it hasn't then we're safe as long as we hold the
75 * spin lock until our increment of open count. */
76 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
77 spin_unlock(&oi
->ip_lock
);
84 oi
->ip_flags
|= OCFS2_INODE_OPEN_DIRECT
;
87 spin_unlock(&oi
->ip_lock
);
94 static int ocfs2_file_release(struct inode
*inode
, struct file
*file
)
96 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
98 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode
, file
,
99 file
->f_dentry
->d_name
.len
,
100 file
->f_dentry
->d_name
.name
);
102 spin_lock(&oi
->ip_lock
);
103 if (!--oi
->ip_open_count
)
104 oi
->ip_flags
&= ~OCFS2_INODE_OPEN_DIRECT
;
105 spin_unlock(&oi
->ip_lock
);
112 static int ocfs2_sync_file(struct file
*file
,
113 struct dentry
*dentry
,
118 struct inode
*inode
= dentry
->d_inode
;
119 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
121 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file
, dentry
, datasync
,
122 dentry
->d_name
.len
, dentry
->d_name
.name
);
124 err
= ocfs2_sync_inode(dentry
->d_inode
);
128 journal
= osb
->journal
->j_journal
;
129 err
= journal_force_commit(journal
);
134 return (err
< 0) ? -EIO
: 0;
137 int ocfs2_set_inode_size(handle_t
*handle
,
139 struct buffer_head
*fe_bh
,
145 i_size_write(inode
, new_i_size
);
146 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(new_i_size
);
147 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
149 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
160 static int ocfs2_simple_size_update(struct inode
*inode
,
161 struct buffer_head
*di_bh
,
165 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
166 handle_t
*handle
= NULL
;
168 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
169 if (handle
== NULL
) {
175 ret
= ocfs2_set_inode_size(handle
, inode
, di_bh
,
180 ocfs2_commit_trans(osb
, handle
);
185 static int ocfs2_orphan_for_truncate(struct ocfs2_super
*osb
,
187 struct buffer_head
*fe_bh
,
195 /* TODO: This needs to actually orphan the inode in this
198 handle
= ocfs2_start_trans(osb
, 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(osb
, 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
,
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 handle_t
*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 num_free_extents
= ocfs2_num_free_extents(osb
,
468 if (num_free_extents
< 0) {
469 status
= num_free_extents
;
474 if (!num_free_extents
) {
475 status
= ocfs2_reserve_new_metadata(osb
, fe
, &meta_ac
);
477 if (status
!= -ENOSPC
)
483 status
= ocfs2_reserve_clusters(osb
, clusters_to_add
, &data_ac
);
485 if (status
!= -ENOSPC
)
490 /* blocks peope in read/write from reading our allocation
491 * until we're done changing it. We depend on i_mutex to block
492 * other extend/truncate calls while we're here. Ordering wrt
493 * start_trans is important here -- always do it before! */
494 down_write(&OCFS2_I(inode
)->ip_alloc_sem
);
497 credits
= ocfs2_calc_extend_credits(osb
->sb
, fe
, clusters_to_add
);
498 handle
= ocfs2_start_trans(osb
, credits
);
499 if (IS_ERR(handle
)) {
500 status
= PTR_ERR(handle
);
506 restarted_transaction
:
507 /* reserve a write to the file entry early on - that we if we
508 * run out of credits in the allocation path, we can still
510 status
= ocfs2_journal_access(handle
, inode
, bh
,
511 OCFS2_JOURNAL_ACCESS_WRITE
);
517 prev_clusters
= OCFS2_I(inode
)->ip_clusters
;
519 status
= ocfs2_do_extend_allocation(osb
,
527 if ((status
< 0) && (status
!= -EAGAIN
)) {
528 if (status
!= -ENOSPC
)
533 status
= ocfs2_journal_dirty(handle
, bh
);
539 spin_lock(&OCFS2_I(inode
)->ip_lock
);
540 clusters_to_add
-= (OCFS2_I(inode
)->ip_clusters
- prev_clusters
);
541 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
543 if (why
!= RESTART_NONE
&& clusters_to_add
) {
544 if (why
== RESTART_META
) {
545 mlog(0, "restarting function.\n");
548 BUG_ON(why
!= RESTART_TRANS
);
550 mlog(0, "restarting transaction.\n");
551 /* TODO: This can be more intelligent. */
552 credits
= ocfs2_calc_extend_credits(osb
->sb
,
555 status
= ocfs2_extend_trans(handle
, credits
);
557 /* handle still has to be committed at
563 goto restarted_transaction
;
567 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
568 fe
->i_clusters
, (unsigned long long)fe
->i_size
);
569 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
570 OCFS2_I(inode
)->ip_clusters
, i_size_read(inode
));
573 if (drop_alloc_sem
) {
574 up_write(&OCFS2_I(inode
)->ip_alloc_sem
);
578 ocfs2_commit_trans(osb
, handle
);
582 ocfs2_free_alloc_context(data_ac
);
586 ocfs2_free_alloc_context(meta_ac
);
589 if ((!status
) && restart_func
) {
602 /* Some parts of this taken from generic_cont_expand, which turned out
603 * to be too fragile to do exactly what we need without us having to
604 * worry about recursive locking in ->prepare_write() and
605 * ->commit_write(). */
606 static int ocfs2_write_zero_page(struct inode
*inode
,
609 struct address_space
*mapping
= inode
->i_mapping
;
613 handle_t
*handle
= NULL
;
616 offset
= (size
& (PAGE_CACHE_SIZE
-1)); /* Within page */
617 /* ugh. in prepare/commit_write, if from==to==start of block, we
618 ** skip the prepare. make sure we never send an offset for the start
621 if ((offset
& (inode
->i_sb
->s_blocksize
- 1)) == 0) {
624 index
= size
>> PAGE_CACHE_SHIFT
;
626 page
= grab_cache_page(mapping
, index
);
633 ret
= ocfs2_prepare_write_nolock(inode
, page
, offset
, offset
);
639 if (ocfs2_should_order_data(inode
)) {
640 handle
= ocfs2_start_walk_page_trans(inode
, page
, offset
,
642 if (IS_ERR(handle
)) {
643 ret
= PTR_ERR(handle
);
649 /* must not update i_size! */
650 ret
= block_commit_write(page
, offset
, offset
);
657 ocfs2_commit_trans(OCFS2_SB(inode
->i_sb
), handle
);
660 page_cache_release(page
);
665 static int ocfs2_zero_extend(struct inode
*inode
,
670 struct super_block
*sb
= inode
->i_sb
;
672 start_off
= ocfs2_align_bytes_to_blocks(sb
, i_size_read(inode
));
673 while (start_off
< zero_to_size
) {
674 ret
= ocfs2_write_zero_page(inode
, start_off
);
680 start_off
+= sb
->s_blocksize
;
683 * Very large extends have the potential to lock up
684 * the cpu for extended periods of time.
694 * A tail_to_skip value > 0 indicates that we're being called from
695 * ocfs2_file_aio_write(). This has the following implications:
697 * - we don't want to update i_size
698 * - di_bh will be NULL, which is fine because it's only used in the
699 * case where we want to update i_size.
700 * - ocfs2_zero_extend() will then only be filling the hole created
701 * between i_size and the start of the write.
703 static int ocfs2_extend_file(struct inode
*inode
,
704 struct buffer_head
*di_bh
,
711 BUG_ON(!tail_to_skip
&& !di_bh
);
713 /* setattr sometimes calls us like this. */
717 if (i_size_read(inode
) == new_i_size
)
719 BUG_ON(new_i_size
< i_size_read(inode
));
721 clusters_to_add
= ocfs2_clusters_for_bytes(inode
->i_sb
, new_i_size
) -
722 OCFS2_I(inode
)->ip_clusters
;
725 * protect the pages that ocfs2_zero_extend is going to be
726 * pulling into the page cache.. we do this before the
727 * metadata extend so that we don't get into the situation
728 * where we've extended the metadata but can't get the data
731 ret
= ocfs2_data_lock(inode
, 1);
737 if (clusters_to_add
) {
738 ret
= ocfs2_extend_allocation(inode
, clusters_to_add
);
746 * Call this even if we don't add any clusters to the tree. We
747 * still need to zero the area between the old i_size and the
750 ret
= ocfs2_zero_extend(inode
, (u64
)new_i_size
- tail_to_skip
);
757 /* We're being called from ocfs2_setattr() which wants
758 * us to update i_size */
759 ret
= ocfs2_simple_size_update(inode
, di_bh
, new_i_size
);
765 ocfs2_data_unlock(inode
, 1);
771 int ocfs2_setattr(struct dentry
*dentry
, struct iattr
*attr
)
773 int status
= 0, size_change
;
774 struct inode
*inode
= dentry
->d_inode
;
775 struct super_block
*sb
= inode
->i_sb
;
776 struct ocfs2_super
*osb
= OCFS2_SB(sb
);
777 struct buffer_head
*bh
= NULL
;
778 handle_t
*handle
= NULL
;
780 mlog_entry("(0x%p, '%.*s')\n", dentry
,
781 dentry
->d_name
.len
, dentry
->d_name
.name
);
783 if (attr
->ia_valid
& ATTR_MODE
)
784 mlog(0, "mode change: %d\n", attr
->ia_mode
);
785 if (attr
->ia_valid
& ATTR_UID
)
786 mlog(0, "uid change: %d\n", attr
->ia_uid
);
787 if (attr
->ia_valid
& ATTR_GID
)
788 mlog(0, "gid change: %d\n", attr
->ia_gid
);
789 if (attr
->ia_valid
& ATTR_SIZE
)
790 mlog(0, "size change...\n");
791 if (attr
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_CTIME
))
792 mlog(0, "time change...\n");
794 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
795 | ATTR_GID | ATTR_UID | ATTR_MODE)
796 if (!(attr
->ia_valid
& OCFS2_VALID_ATTRS
)) {
797 mlog(0, "can't handle attrs: 0x%x\n", attr
->ia_valid
);
801 status
= inode_change_ok(inode
, attr
);
805 size_change
= S_ISREG(inode
->i_mode
) && attr
->ia_valid
& ATTR_SIZE
;
807 status
= ocfs2_rw_lock(inode
, 1);
814 status
= ocfs2_meta_lock(inode
, &bh
, 1);
816 if (status
!= -ENOENT
)
821 if (size_change
&& attr
->ia_size
!= i_size_read(inode
)) {
822 if (i_size_read(inode
) > attr
->ia_size
)
823 status
= ocfs2_truncate_file(inode
, bh
, attr
->ia_size
);
825 status
= ocfs2_extend_file(inode
, bh
, attr
->ia_size
, 0);
827 if (status
!= -ENOSPC
)
834 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
835 if (IS_ERR(handle
)) {
836 status
= PTR_ERR(handle
);
841 status
= inode_setattr(inode
, attr
);
847 status
= ocfs2_mark_inode_dirty(handle
, inode
, bh
);
852 ocfs2_commit_trans(osb
, handle
);
854 ocfs2_meta_unlock(inode
, 1);
857 ocfs2_rw_unlock(inode
, 1);
866 int ocfs2_getattr(struct vfsmount
*mnt
,
867 struct dentry
*dentry
,
870 struct inode
*inode
= dentry
->d_inode
;
871 struct super_block
*sb
= dentry
->d_inode
->i_sb
;
872 struct ocfs2_super
*osb
= sb
->s_fs_info
;
877 err
= ocfs2_inode_revalidate(dentry
);
884 generic_fillattr(inode
, stat
);
886 /* We set the blksize from the cluster size for performance */
887 stat
->blksize
= osb
->s_clustersize
;
895 static int ocfs2_write_remove_suid(struct inode
*inode
)
898 struct buffer_head
*bh
= NULL
;
899 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
901 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
902 struct ocfs2_dinode
*di
;
904 mlog_entry("(Inode %llu, mode 0%o)\n",
905 (unsigned long long)oi
->ip_blkno
, inode
->i_mode
);
907 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
908 if (handle
== NULL
) {
914 ret
= ocfs2_read_block(osb
, oi
->ip_blkno
, &bh
, OCFS2_BH_CACHED
, inode
);
920 ret
= ocfs2_journal_access(handle
, inode
, bh
,
921 OCFS2_JOURNAL_ACCESS_WRITE
);
927 inode
->i_mode
&= ~S_ISUID
;
928 if ((inode
->i_mode
& S_ISGID
) && (inode
->i_mode
& S_IXGRP
))
929 inode
->i_mode
&= ~S_ISGID
;
931 di
= (struct ocfs2_dinode
*) bh
->b_data
;
932 di
->i_mode
= cpu_to_le16(inode
->i_mode
);
934 ret
= ocfs2_journal_dirty(handle
, bh
);
940 ocfs2_commit_trans(osb
, handle
);
946 static inline int ocfs2_write_should_remove_suid(struct inode
*inode
)
948 mode_t mode
= inode
->i_mode
;
950 if (!capable(CAP_FSETID
)) {
951 if (unlikely(mode
& S_ISUID
))
954 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
960 static ssize_t
ocfs2_file_aio_write(struct kiocb
*iocb
,
961 const struct iovec
*iov
,
962 unsigned long nr_segs
,
965 int ret
, rw_level
= -1, meta_level
= -1, have_alloc_sem
= 0;
967 struct file
*filp
= iocb
->ki_filp
;
968 struct inode
*inode
= filp
->f_dentry
->d_inode
;
969 loff_t newsize
, saved_pos
;
971 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
972 (unsigned int)nr_segs
,
973 filp
->f_dentry
->d_name
.len
,
974 filp
->f_dentry
->d_name
.name
);
976 /* happy write of zero bytes */
977 if (iocb
->ki_left
== 0)
981 mlog(0, "bad inode\n");
985 mutex_lock(&inode
->i_mutex
);
986 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
987 if (filp
->f_flags
& O_DIRECT
) {
989 down_read(&inode
->i_alloc_sem
);
992 /* concurrent O_DIRECT writes are allowed */
993 rw_level
= (filp
->f_flags
& O_DIRECT
) ? 0 : 1;
994 ret
= ocfs2_rw_lock(inode
, rw_level
);
1002 * We sample i_size under a read level meta lock to see if our write
1003 * is extending the file, if it is we back off and get a write level
1006 meta_level
= (filp
->f_flags
& O_APPEND
) ? 1 : 0;
1008 ret
= ocfs2_meta_lock(inode
, NULL
, meta_level
);
1015 /* Clear suid / sgid if necessary. We do this here
1016 * instead of later in the write path because
1017 * remove_suid() calls ->setattr without any hint that
1018 * we may have already done our cluster locking. Since
1019 * ocfs2_setattr() *must* take cluster locks to
1020 * proceeed, this will lead us to recursively lock the
1021 * inode. There's also the dinode i_size state which
1022 * can be lost via setattr during extending writes (we
1023 * set inode->i_size at the end of a write. */
1024 if (ocfs2_write_should_remove_suid(inode
)) {
1025 if (meta_level
== 0) {
1026 ocfs2_meta_unlock(inode
, meta_level
);
1031 ret
= ocfs2_write_remove_suid(inode
);
1038 /* work on a copy of ppos until we're sure that we won't have
1039 * to recalculate it due to relocking. */
1040 if (filp
->f_flags
& O_APPEND
) {
1041 saved_pos
= i_size_read(inode
);
1042 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos
);
1044 saved_pos
= iocb
->ki_pos
;
1046 newsize
= iocb
->ki_left
+ saved_pos
;
1048 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1049 (long long) saved_pos
, (long long) newsize
,
1050 (long long) i_size_read(inode
));
1052 /* No need for a higher level metadata lock if we're
1053 * never going past i_size. */
1054 if (newsize
<= i_size_read(inode
))
1057 if (meta_level
== 0) {
1058 ocfs2_meta_unlock(inode
, meta_level
);
1063 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1064 clusters
= ocfs2_clusters_for_bytes(inode
->i_sb
, newsize
) -
1065 OCFS2_I(inode
)->ip_clusters
;
1066 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1068 mlog(0, "Writing at EOF, may need more allocation: "
1069 "i_size = %lld, newsize = %lld, need %u clusters\n",
1070 (long long) i_size_read(inode
), (long long) newsize
,
1073 /* We only want to continue the rest of this loop if
1074 * our extend will actually require more
1079 ret
= ocfs2_extend_file(inode
, NULL
, newsize
, iocb
->ki_left
);
1088 /* ok, we're done with i_size and alloc work */
1089 iocb
->ki_pos
= saved_pos
;
1090 ocfs2_meta_unlock(inode
, meta_level
);
1093 /* communicate with ocfs2_dio_end_io */
1094 ocfs2_iocb_set_rw_locked(iocb
);
1096 ret
= generic_file_aio_write_nolock(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
1098 /* buffered aio wouldn't have proper lock coverage today */
1099 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1102 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1103 * function pointer which is called when o_direct io completes so that
1104 * it can unlock our rw lock. (it's the clustered equivalent of
1105 * i_alloc_sem; protects truncate from racing with pending ios).
1106 * Unfortunately there are error cases which call end_io and others
1107 * that don't. so we don't have to unlock the rw_lock if either an
1108 * async dio is going to do it in the future or an end_io after an
1109 * error has already done it.
1111 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1117 if (meta_level
!= -1)
1118 ocfs2_meta_unlock(inode
, meta_level
);
1120 up_read(&inode
->i_alloc_sem
);
1122 ocfs2_rw_unlock(inode
, rw_level
);
1123 mutex_unlock(&inode
->i_mutex
);
1129 static ssize_t
ocfs2_file_aio_read(struct kiocb
*iocb
,
1130 const struct iovec
*iov
,
1131 unsigned long nr_segs
,
1134 int ret
= 0, rw_level
= -1, have_alloc_sem
= 0;
1135 struct file
*filp
= iocb
->ki_filp
;
1136 struct inode
*inode
= filp
->f_dentry
->d_inode
;
1138 mlog_entry("(0x%p, %u, '%.*s')\n", filp
,
1139 (unsigned int)nr_segs
,
1140 filp
->f_dentry
->d_name
.len
,
1141 filp
->f_dentry
->d_name
.name
);
1150 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1151 * need locks to protect pending reads from racing with truncate.
1153 if (filp
->f_flags
& O_DIRECT
) {
1154 down_read(&inode
->i_alloc_sem
);
1157 ret
= ocfs2_rw_lock(inode
, 0);
1163 /* communicate with ocfs2_dio_end_io */
1164 ocfs2_iocb_set_rw_locked(iocb
);
1168 * We're fine letting folks race truncates and extending
1169 * writes with read across the cluster, just like they can
1170 * locally. Hence no rw_lock during read.
1172 * Take and drop the meta data lock to update inode fields
1173 * like i_size. This allows the checks down below
1174 * generic_file_aio_read() a chance of actually working.
1176 ret
= ocfs2_meta_lock(inode
, NULL
, 0);
1181 ocfs2_meta_unlock(inode
, 0);
1183 ret
= generic_file_aio_read(iocb
, iov
, nr_segs
, iocb
->ki_pos
);
1185 mlog(ML_ERROR
, "generic_file_aio_read returned -EINVAL\n");
1187 /* buffered aio wouldn't have proper lock coverage today */
1188 BUG_ON(ret
== -EIOCBQUEUED
&& !(filp
->f_flags
& O_DIRECT
));
1190 /* see ocfs2_file_aio_write */
1191 if (ret
== -EIOCBQUEUED
|| !ocfs2_iocb_is_rw_locked(iocb
)) {
1198 up_read(&inode
->i_alloc_sem
);
1200 ocfs2_rw_unlock(inode
, rw_level
);
1206 struct inode_operations ocfs2_file_iops
= {
1207 .setattr
= ocfs2_setattr
,
1208 .getattr
= ocfs2_getattr
,
1211 struct inode_operations ocfs2_special_file_iops
= {
1212 .setattr
= ocfs2_setattr
,
1213 .getattr
= ocfs2_getattr
,
1216 const struct file_operations ocfs2_fops
= {
1217 .read
= do_sync_read
,
1218 .write
= do_sync_write
,
1219 .sendfile
= generic_file_sendfile
,
1221 .fsync
= ocfs2_sync_file
,
1222 .release
= ocfs2_file_release
,
1223 .open
= ocfs2_file_open
,
1224 .aio_read
= ocfs2_file_aio_read
,
1225 .aio_write
= ocfs2_file_aio_write
,
1226 .ioctl
= ocfs2_ioctl
,
1229 const struct file_operations ocfs2_dops
= {
1230 .read
= generic_read_dir
,
1231 .readdir
= ocfs2_readdir
,
1232 .fsync
= ocfs2_sync_file
,
1233 .ioctl
= ocfs2_ioctl
,