1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * vfs' aops, fops, dops and iops
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/quotaops.h>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
42 #include "blockcheck.h"
44 #include "extent_map.h"
46 #include "heartbeat.h"
57 #include "buffer_head_io.h"
59 struct ocfs2_find_inode_args
63 unsigned int fi_flags
;
64 unsigned int fi_sysfile_type
;
67 static struct lock_class_key ocfs2_sysfile_lock_key
[NUM_SYSTEM_INODES
];
69 static int ocfs2_read_locked_inode(struct inode
*inode
,
70 struct ocfs2_find_inode_args
*args
);
71 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
72 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
73 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
75 struct buffer_head
*fe_bh
);
77 void ocfs2_set_inode_flags(struct inode
*inode
)
79 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
81 inode
->i_flags
&= ~(S_IMMUTABLE
|
82 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
84 if (flags
& OCFS2_IMMUTABLE_FL
)
85 inode
->i_flags
|= S_IMMUTABLE
;
87 if (flags
& OCFS2_SYNC_FL
)
88 inode
->i_flags
|= S_SYNC
;
89 if (flags
& OCFS2_APPEND_FL
)
90 inode
->i_flags
|= S_APPEND
;
91 if (flags
& OCFS2_NOATIME_FL
)
92 inode
->i_flags
|= S_NOATIME
;
93 if (flags
& OCFS2_DIRSYNC_FL
)
94 inode
->i_flags
|= S_DIRSYNC
;
97 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
98 void ocfs2_get_inode_flags(struct ocfs2_inode_info
*oi
)
100 unsigned int flags
= oi
->vfs_inode
.i_flags
;
102 oi
->ip_attr
&= ~(OCFS2_SYNC_FL
|OCFS2_APPEND_FL
|
103 OCFS2_IMMUTABLE_FL
|OCFS2_NOATIME_FL
|OCFS2_DIRSYNC_FL
);
105 oi
->ip_attr
|= OCFS2_SYNC_FL
;
106 if (flags
& S_APPEND
)
107 oi
->ip_attr
|= OCFS2_APPEND_FL
;
108 if (flags
& S_IMMUTABLE
)
109 oi
->ip_attr
|= OCFS2_IMMUTABLE_FL
;
110 if (flags
& S_NOATIME
)
111 oi
->ip_attr
|= OCFS2_NOATIME_FL
;
112 if (flags
& S_DIRSYNC
)
113 oi
->ip_attr
|= OCFS2_DIRSYNC_FL
;
116 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, unsigned flags
,
119 struct inode
*inode
= NULL
;
120 struct super_block
*sb
= osb
->sb
;
121 struct ocfs2_find_inode_args args
;
123 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
125 /* Ok. By now we've either got the offsets passed to us by the
126 * caller, or we just pulled them off the bh. Lets do some
127 * sanity checks to make sure they're OK. */
129 inode
= ERR_PTR(-EINVAL
);
130 mlog_errno(PTR_ERR(inode
));
134 args
.fi_blkno
= blkno
;
135 args
.fi_flags
= flags
;
136 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
137 args
.fi_sysfile_type
= sysfile_type
;
139 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
140 ocfs2_init_locked_inode
, &args
);
141 /* inode was *not* in the inode cache. 2.6.x requires
142 * us to do our own read_inode call and unlock it
144 if (inode
&& inode
->i_state
& I_NEW
) {
145 mlog(0, "Inode was not in inode cache, reading it.\n");
146 ocfs2_read_locked_inode(inode
, &args
);
147 unlock_new_inode(inode
);
150 inode
= ERR_PTR(-ENOMEM
);
151 mlog_errno(PTR_ERR(inode
));
154 if (is_bad_inode(inode
)) {
156 inode
= ERR_PTR(-ESTALE
);
161 if (!IS_ERR(inode
)) {
162 mlog(0, "returning inode with number %llu\n",
163 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
164 mlog_exit_ptr(inode
);
172 * here's how inodes get read from disk:
173 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
174 * found? : return the in-memory inode
175 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
178 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
180 struct ocfs2_find_inode_args
*args
= NULL
;
181 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
184 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
188 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
190 if (oi
->ip_blkno
!= args
->fi_blkno
)
200 * initialize the new inode, but don't do anything that would cause
202 * return 0 on success, 1 on failure
204 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
206 struct ocfs2_find_inode_args
*args
= opaque
;
208 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
210 inode
->i_ino
= args
->fi_ino
;
211 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
212 if (args
->fi_sysfile_type
!= 0)
213 lockdep_set_class(&inode
->i_mutex
,
214 &ocfs2_sysfile_lock_key
[args
->fi_sysfile_type
]);
220 void ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
223 struct super_block
*sb
;
224 struct ocfs2_super
*osb
;
227 mlog_entry("(0x%p, size:%llu)\n", inode
,
228 (unsigned long long)le64_to_cpu(fe
->i_size
));
233 if ((osb
->s_mount_opt
& OCFS2_MOUNT_LOCALFLOCKS
) ||
234 ocfs2_mount_local(osb
) || !ocfs2_stack_supports_plocks())
238 * These have all been checked by ocfs2_read_inode_block() or set
239 * by ocfs2_mknod_locked(), so a failure is a code bug.
241 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
)); /* This means that read_inode
242 cannot create a superblock
243 inode today. change if
245 BUG_ON(!(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
)));
246 BUG_ON(le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
);
249 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
250 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
251 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
253 inode
->i_version
= 1;
254 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
255 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
256 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
257 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
258 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
260 /* Fast symlinks will have i_size but no allocated clusters. */
261 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
264 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
265 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
266 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
267 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
268 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
269 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
270 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
271 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
273 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
275 "ip_blkno %llu != i_blkno %llu!\n",
276 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
277 (unsigned long long)le64_to_cpu(fe
->i_blkno
));
279 inode
->i_nlink
= ocfs2_read_links_count(fe
);
281 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) {
282 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
283 inode
->i_flags
|= S_NOQUOTA
;
286 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
287 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
288 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
289 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
290 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
291 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_QUOTA_FL
)) {
292 inode
->i_flags
|= S_NOQUOTA
;
293 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
294 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
295 /* we can't actually hit this as read_inode can't
296 * handle superblocks today ;-) */
300 switch (inode
->i_mode
& S_IFMT
) {
303 inode
->i_fop
= &ocfs2_fops
;
305 inode
->i_fop
= &ocfs2_fops_no_plocks
;
306 inode
->i_op
= &ocfs2_file_iops
;
307 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
310 inode
->i_op
= &ocfs2_dir_iops
;
312 inode
->i_fop
= &ocfs2_dops
;
314 inode
->i_fop
= &ocfs2_dops_no_plocks
;
315 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
318 if (ocfs2_inode_is_fast_symlink(inode
))
319 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
321 inode
->i_op
= &ocfs2_symlink_inode_operations
;
322 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
325 inode
->i_op
= &ocfs2_special_file_iops
;
326 init_special_inode(inode
, inode
->i_mode
,
332 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
333 le64_to_cpu(fe
->i_blkno
));
336 * If we ever want to create system files from kernel,
337 * the generation argument to
338 * ocfs2_inode_lock_res_init() will have to change.
340 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
342 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
343 OCFS2_LOCK_TYPE_META
, 0, inode
);
345 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
346 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
349 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
350 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
353 ocfs2_set_inode_flags(inode
);
355 OCFS2_I(inode
)->ip_last_used_slot
= 0;
356 OCFS2_I(inode
)->ip_last_used_group
= 0;
360 static int ocfs2_read_locked_inode(struct inode
*inode
,
361 struct ocfs2_find_inode_args
*args
)
363 struct super_block
*sb
;
364 struct ocfs2_super
*osb
;
365 struct ocfs2_dinode
*fe
;
366 struct buffer_head
*bh
= NULL
;
367 int status
, can_lock
;
370 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
373 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
374 mlog(ML_ERROR
, "bad inode\n");
381 mlog(ML_ERROR
, "bad inode args\n");
382 make_bad_inode(inode
);
387 * To improve performance of cold-cache inode stats, we take
388 * the cluster lock here if possible.
390 * Generally, OCFS2 never trusts the contents of an inode
391 * unless it's holding a cluster lock, so taking it here isn't
392 * a correctness issue as much as it is a performance
395 * There are three times when taking the lock is not a good idea:
397 * 1) During startup, before we have initialized the DLM.
399 * 2) If we are reading certain system files which never get
400 * cluster locks (local alloc, truncate log).
402 * 3) If the process doing the iget() is responsible for
403 * orphan dir recovery. We're holding the orphan dir lock and
404 * can get into a deadlock with another process on another
405 * node in ->delete_inode().
407 * #1 and #2 can be simply solved by never taking the lock
408 * here for system files (which are the only type we read
409 * during mount). It's a heavier approach, but our main
410 * concern is user-accesible files anyway.
412 * #3 works itself out because we'll eventually take the
413 * cluster lock before trusting anything anyway.
415 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
416 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
417 && !ocfs2_mount_local(osb
);
420 * To maintain backwards compatibility with older versions of
421 * ocfs2-tools, we still store the generation value for system
422 * files. The only ones that actually matter to userspace are
423 * the journals, but it's easier and inexpensive to just flag
424 * all system files similarly.
426 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
427 generation
= osb
->fs_generation
;
429 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
430 OCFS2_LOCK_TYPE_META
,
433 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
434 OCFS2_LOCK_TYPE_OPEN
,
438 status
= ocfs2_open_lock(inode
);
440 make_bad_inode(inode
);
444 status
= ocfs2_inode_lock(inode
, NULL
, 0);
446 make_bad_inode(inode
);
452 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
453 status
= ocfs2_try_open_lock(inode
, 0);
455 make_bad_inode(inode
);
461 status
= ocfs2_read_inode_block_full(inode
, &bh
,
462 OCFS2_BH_IGNORE_CACHE
);
464 status
= ocfs2_read_blocks_sync(osb
, args
->fi_blkno
, 1, &bh
);
466 status
= ocfs2_validate_inode_block(osb
->sb
, bh
);
474 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
477 * This is a code bug. Right now the caller needs to
478 * understand whether it is asking for a system file inode or
479 * not so the proper lock names can be built.
481 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
482 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
483 "Inode %llu: system file state is ambigous\n",
484 (unsigned long long)args
->fi_blkno
);
486 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
487 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
488 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
490 ocfs2_populate_inode(inode
, fe
, 0);
492 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
498 ocfs2_inode_unlock(inode
, 0);
501 make_bad_inode(inode
);
510 void ocfs2_sync_blockdev(struct super_block
*sb
)
512 sync_blockdev(sb
->s_bdev
);
515 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
517 struct buffer_head
*fe_bh
)
520 struct ocfs2_truncate_context
*tc
= NULL
;
521 struct ocfs2_dinode
*fe
;
522 handle_t
*handle
= NULL
;
526 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
529 * This check will also skip truncate of inodes with inline
530 * data and fast symlinks.
532 if (fe
->i_clusters
) {
533 if (ocfs2_should_order_data(inode
))
534 ocfs2_begin_ordered_truncate(inode
, 0);
536 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
537 if (IS_ERR(handle
)) {
538 status
= PTR_ERR(handle
);
543 status
= ocfs2_journal_access_di(handle
, inode
, fe_bh
,
544 OCFS2_JOURNAL_ACCESS_WRITE
);
550 i_size_write(inode
, 0);
552 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
558 ocfs2_commit_trans(osb
, handle
);
561 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
567 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
576 ocfs2_commit_trans(osb
, handle
);
581 static int ocfs2_remove_inode(struct inode
*inode
,
582 struct buffer_head
*di_bh
,
583 struct inode
*orphan_dir_inode
,
584 struct buffer_head
*orphan_dir_bh
)
587 struct inode
*inode_alloc_inode
= NULL
;
588 struct buffer_head
*inode_alloc_bh
= NULL
;
590 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
591 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
594 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
595 le16_to_cpu(di
->i_suballoc_slot
));
596 if (!inode_alloc_inode
) {
602 mutex_lock(&inode_alloc_inode
->i_mutex
);
603 status
= ocfs2_inode_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
605 mutex_unlock(&inode_alloc_inode
->i_mutex
);
611 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
+
612 ocfs2_quota_trans_credits(inode
->i_sb
));
613 if (IS_ERR(handle
)) {
614 status
= PTR_ERR(handle
);
619 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
626 /* set the inodes dtime */
627 status
= ocfs2_journal_access_di(handle
, inode
, di_bh
,
628 OCFS2_JOURNAL_ACCESS_WRITE
);
634 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
635 di
->i_flags
&= cpu_to_le32(~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
637 status
= ocfs2_journal_dirty(handle
, di_bh
);
643 ocfs2_remove_from_cache(inode
, di_bh
);
644 vfs_dq_free_inode(inode
);
646 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
652 ocfs2_commit_trans(osb
, handle
);
654 ocfs2_inode_unlock(inode_alloc_inode
, 1);
655 mutex_unlock(&inode_alloc_inode
->i_mutex
);
656 brelse(inode_alloc_bh
);
658 iput(inode_alloc_inode
);
664 * Serialize with orphan dir recovery. If the process doing
665 * recovery on this orphan dir does an iget() with the dir
666 * i_mutex held, we'll deadlock here. Instead we detect this
667 * and exit early - recovery will wipe this inode for us.
669 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
674 spin_lock(&osb
->osb_lock
);
675 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
676 mlog(0, "Recovery is happening on orphan dir %d, will skip "
677 "this inode\n", slot
);
681 /* This signals to the orphan recovery process that it should
682 * wait for us to handle the wipe. */
683 osb
->osb_orphan_wipes
[slot
]++;
685 spin_unlock(&osb
->osb_lock
);
689 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
692 spin_lock(&osb
->osb_lock
);
693 osb
->osb_orphan_wipes
[slot
]--;
694 spin_unlock(&osb
->osb_lock
);
696 wake_up(&osb
->osb_wipe_event
);
699 static int ocfs2_wipe_inode(struct inode
*inode
,
700 struct buffer_head
*di_bh
)
702 int status
, orphaned_slot
;
703 struct inode
*orphan_dir_inode
= NULL
;
704 struct buffer_head
*orphan_dir_bh
= NULL
;
705 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
706 struct ocfs2_dinode
*di
;
708 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
709 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
711 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
715 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
716 ORPHAN_DIR_SYSTEM_INODE
,
718 if (!orphan_dir_inode
) {
724 /* Lock the orphan dir. The lock will be held for the entire
725 * delete_inode operation. We do this now to avoid races with
726 * recovery completion on other nodes. */
727 mutex_lock(&orphan_dir_inode
->i_mutex
);
728 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
730 mutex_unlock(&orphan_dir_inode
->i_mutex
);
736 /* we do this while holding the orphan dir lock because we
737 * don't want recovery being run from another node to try an
738 * inode delete underneath us -- this will result in two nodes
739 * truncating the same file! */
740 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
743 goto bail_unlock_dir
;
746 /* Remove any dir index tree */
747 if (S_ISDIR(inode
->i_mode
)) {
748 status
= ocfs2_dx_dir_truncate(inode
, di_bh
);
751 goto bail_unlock_dir
;
755 /*Free extended attribute resources associated with this inode.*/
756 status
= ocfs2_xattr_remove(inode
, di_bh
);
759 goto bail_unlock_dir
;
762 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
768 ocfs2_inode_unlock(orphan_dir_inode
, 1);
769 mutex_unlock(&orphan_dir_inode
->i_mutex
);
770 brelse(orphan_dir_bh
);
772 iput(orphan_dir_inode
);
773 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
778 /* There is a series of simple checks that should be done before a
779 * trylock is even considered. Encapsulate those in this function. */
780 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
783 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
784 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
786 /* We shouldn't be getting here for the root directory
788 if (inode
== osb
->root_inode
) {
789 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
793 /* If we're coming from downconvert_thread we can't go into our own
794 * voting [hello, deadlock city!], so unforuntately we just
795 * have to skip deleting this guy. That's OK though because
796 * the node who's doing the actual deleting should handle it
798 if (current
== osb
->dc_task
) {
799 mlog(0, "Skipping delete of %lu because we're currently "
800 "in downconvert\n", inode
->i_ino
);
804 spin_lock(&oi
->ip_lock
);
805 /* OCFS2 *never* deletes system files. This should technically
806 * never get here as system file inodes should always have a
807 * positive link count. */
808 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
809 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
810 (unsigned long long)oi
->ip_blkno
);
814 /* If we have allowd wipe of this inode for another node, it
815 * will be marked here so we can safely skip it. Recovery will
816 * cleanup any inodes we might inadvertantly skip here. */
817 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
818 mlog(0, "Skipping delete of %lu because another node "
819 "has done this for us.\n", inode
->i_ino
);
825 spin_unlock(&oi
->ip_lock
);
830 /* Query the cluster to determine whether we should wipe an inode from
833 * Requires the inode to have the cluster lock. */
834 static int ocfs2_query_inode_wipe(struct inode
*inode
,
835 struct buffer_head
*di_bh
,
839 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
840 struct ocfs2_dinode
*di
;
844 /* While we were waiting for the cluster lock in
845 * ocfs2_delete_inode, another node might have asked to delete
846 * the inode. Recheck our flags to catch this. */
847 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
848 mlog(0, "Skipping delete of %llu because flags changed\n",
849 (unsigned long long)oi
->ip_blkno
);
853 /* Now that we have an up to date inode, we can double check
855 if (inode
->i_nlink
) {
856 mlog(0, "Skipping delete of %llu because nlink = %u\n",
857 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
861 /* Do some basic inode verification... */
862 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
863 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
864 /* for lack of a better error? */
867 "Inode %llu (on-disk %llu) not orphaned! "
868 "Disk flags 0x%x, inode flags 0x%x\n",
869 (unsigned long long)oi
->ip_blkno
,
870 (unsigned long long)le64_to_cpu(di
->i_blkno
),
871 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
875 /* has someone already deleted us?! baaad... */
883 * This is how ocfs2 determines whether an inode is still live
884 * within the cluster. Every node takes a shared read lock on
885 * the inode open lock in ocfs2_read_locked_inode(). When we
886 * get to ->delete_inode(), each node tries to convert it's
887 * lock to an exclusive. Trylocks are serialized by the inode
888 * meta data lock. If the upconvert suceeds, we know the inode
889 * is no longer live and can be deleted.
891 * Though we call this with the meta data lock held, the
892 * trylock keeps us from ABBA deadlock.
894 status
= ocfs2_try_open_lock(inode
, 1);
895 if (status
== -EAGAIN
) {
897 mlog(0, "Skipping delete of %llu because it is in use on "
898 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
907 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
908 (unsigned long long)oi
->ip_blkno
,
909 le16_to_cpu(di
->i_orphaned_slot
));
915 /* Support function for ocfs2_delete_inode. Will help us keep the
916 * inode data in a consistent state for clear_inode. Always truncates
917 * pages, optionally sync's them first. */
918 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
921 mlog(0, "Cleanup inode %llu, sync = %d\n",
922 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
924 write_inode_now(inode
, 1);
925 truncate_inode_pages(&inode
->i_data
, 0);
928 void ocfs2_delete_inode(struct inode
*inode
)
931 sigset_t blocked
, oldset
;
932 struct buffer_head
*di_bh
= NULL
;
934 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
936 /* When we fail in read_inode() we mark inode as bad. The second test
937 * catches the case when inode allocation fails before allocating
938 * a block for inode. */
939 if (is_bad_inode(inode
) || !OCFS2_I(inode
)->ip_blkno
) {
940 mlog(0, "Skipping delete of bad inode\n");
944 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
945 /* It's probably not necessary to truncate_inode_pages
946 * here but we do it for safety anyway (it will most
947 * likely be a no-op anyway) */
948 ocfs2_cleanup_delete_inode(inode
, 0);
952 /* We want to block signals in delete_inode as the lock and
953 * messaging paths may return us -ERESTARTSYS. Which would
954 * cause us to exit early, resulting in inodes being orphaned
956 sigfillset(&blocked
);
957 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
960 ocfs2_cleanup_delete_inode(inode
, 1);
964 /* Lock down the inode. This gives us an up to date view of
965 * it's metadata (for verification), and allows us to
966 * serialize delete_inode on multiple nodes.
968 * Even though we might be doing a truncate, we don't take the
969 * allocation lock here as it won't be needed - nobody will
970 * have the file open.
972 status
= ocfs2_inode_lock(inode
, &di_bh
, 1);
974 if (status
!= -ENOENT
)
976 ocfs2_cleanup_delete_inode(inode
, 0);
980 /* Query the cluster. This will be the final decision made
981 * before we go ahead and wipe the inode. */
982 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
983 if (!wipe
|| status
< 0) {
984 /* Error and remote inode busy both mean we won't be
985 * removing the inode, so they take almost the same
990 /* Someone in the cluster has disallowed a wipe of
991 * this inode, or it was never completely
992 * orphaned. Write out the pages and exit now. */
993 ocfs2_cleanup_delete_inode(inode
, 1);
994 goto bail_unlock_inode
;
997 ocfs2_cleanup_delete_inode(inode
, 0);
999 status
= ocfs2_wipe_inode(inode
, di_bh
);
1001 if (status
!= -EDEADLK
)
1003 goto bail_unlock_inode
;
1007 * Mark the inode as successfully deleted.
1009 * This is important for ocfs2_clear_inode() as it will check
1010 * this flag and skip any checkpointing work
1012 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1013 * the LVB for other nodes.
1015 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
1018 ocfs2_inode_unlock(inode
, 1);
1021 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
1029 void ocfs2_clear_inode(struct inode
*inode
)
1032 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1039 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1040 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1042 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1043 "Inode=%lu\n", inode
->i_ino
);
1045 /* To preven remote deletes we hold open lock before, now it
1046 * is time to unlock PR and EX open locks. */
1047 ocfs2_open_unlock(inode
);
1049 /* Do these before all the other work so that we don't bounce
1050 * the downconvert thread while waiting to destroy the locks. */
1051 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1052 ocfs2_mark_lockres_freeing(&oi
->ip_inode_lockres
);
1053 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
1055 /* We very well may get a clear_inode before all an inodes
1056 * metadata has hit disk. Of course, we can't drop any cluster
1057 * locks until the journal has finished with it. The only
1058 * exception here are successfully wiped inodes - their
1059 * metadata can now be considered to be part of the system
1060 * inodes from which it came. */
1061 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1062 ocfs2_checkpoint_inode(inode
);
1064 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1065 "Clear inode of %llu, inode has io markers\n",
1066 (unsigned long long)oi
->ip_blkno
);
1068 ocfs2_extent_map_trunc(inode
, 0);
1070 status
= ocfs2_drop_inode_locks(inode
);
1074 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1075 ocfs2_lock_res_free(&oi
->ip_inode_lockres
);
1076 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1078 ocfs2_metadata_cache_purge(inode
);
1080 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1081 "Clear inode of %llu, inode has %u cache items\n",
1082 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1084 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1085 "Clear inode of %llu, inode has a bad flag\n",
1086 (unsigned long long)oi
->ip_blkno
);
1088 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1089 "Clear inode of %llu, inode is locked\n",
1090 (unsigned long long)oi
->ip_blkno
);
1092 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1093 "Clear inode of %llu, io_mutex is locked\n",
1094 (unsigned long long)oi
->ip_blkno
);
1095 mutex_unlock(&oi
->ip_io_mutex
);
1098 * down_trylock() returns 0, down_write_trylock() returns 1
1101 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1102 "Clear inode of %llu, alloc_sem is locked\n",
1103 (unsigned long long)oi
->ip_blkno
);
1104 up_write(&oi
->ip_alloc_sem
);
1106 mlog_bug_on_msg(oi
->ip_open_count
,
1107 "Clear inode of %llu has open count %d\n",
1108 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1110 /* Clear all other flags. */
1111 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1112 oi
->ip_created_trans
= 0;
1113 oi
->ip_last_trans
= 0;
1114 oi
->ip_dir_start_lookup
= 0;
1115 oi
->ip_blkno
= 0ULL;
1118 * ip_jinode is used to track txns against this inode. We ensure that
1119 * the journal is flushed before journal shutdown. Thus it is safe to
1120 * have inodes get cleaned up after journal shutdown.
1122 jbd2_journal_release_jbd_inode(OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
1129 /* Called under inode_lock, with no more references on the
1130 * struct inode, so it's safe here to check the flags field
1131 * and to manipulate i_nlink without any other locks. */
1132 void ocfs2_drop_inode(struct inode
*inode
)
1134 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1138 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1139 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1141 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1142 generic_delete_inode(inode
);
1144 generic_drop_inode(inode
);
1150 * This is called from our getattr.
1152 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1154 struct inode
*inode
= dentry
->d_inode
;
1157 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1158 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1161 mlog(0, "eep, no inode!\n");
1166 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1167 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1168 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1169 mlog(0, "inode deleted!\n");
1173 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1175 /* Let ocfs2_inode_lock do the work of updating our struct
1177 status
= ocfs2_inode_lock(inode
, NULL
, 0);
1179 if (status
!= -ENOENT
)
1183 ocfs2_inode_unlock(inode
, 0);
1191 * Updates a disk inode from a
1193 * Only takes ip_lock.
1195 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1196 struct inode
*inode
,
1197 struct buffer_head
*bh
)
1200 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1202 mlog_entry("(inode %llu)\n",
1203 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1205 status
= ocfs2_journal_access_di(handle
, inode
, bh
,
1206 OCFS2_JOURNAL_ACCESS_WRITE
);
1212 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1213 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1214 ocfs2_get_inode_flags(OCFS2_I(inode
));
1215 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1216 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1217 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1219 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1220 ocfs2_set_links_count(fe
, inode
->i_nlink
);
1221 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1222 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1223 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1224 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1225 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1226 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1227 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1228 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1229 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1231 status
= ocfs2_journal_dirty(handle
, bh
);
1244 * Updates a struct inode from a disk inode.
1245 * does no i/o, only takes ip_lock.
1247 void ocfs2_refresh_inode(struct inode
*inode
,
1248 struct ocfs2_dinode
*fe
)
1250 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1252 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1253 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1254 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1255 ocfs2_set_inode_flags(inode
);
1256 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1257 inode
->i_nlink
= ocfs2_read_links_count(fe
);
1258 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1259 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1260 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1261 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1262 inode
->i_blocks
= 0;
1264 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1265 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1266 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1267 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1268 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1269 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1270 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1272 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1275 int ocfs2_validate_inode_block(struct super_block
*sb
,
1276 struct buffer_head
*bh
)
1279 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1281 mlog(0, "Validating dinode %llu\n",
1282 (unsigned long long)bh
->b_blocknr
);
1284 BUG_ON(!buffer_uptodate(bh
));
1287 * If the ecc fails, we return the error but otherwise
1288 * leave the filesystem running. We know any error is
1289 * local to this block.
1291 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &di
->i_check
);
1293 mlog(ML_ERROR
, "Checksum failed for dinode %llu\n",
1294 (unsigned long long)bh
->b_blocknr
);
1299 * Errors after here are fatal.
1304 if (!OCFS2_IS_VALID_DINODE(di
)) {
1305 ocfs2_error(sb
, "Invalid dinode #%llu: signature = %.*s\n",
1306 (unsigned long long)bh
->b_blocknr
, 7,
1311 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1312 ocfs2_error(sb
, "Invalid dinode #%llu: i_blkno is %llu\n",
1313 (unsigned long long)bh
->b_blocknr
,
1314 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1318 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1320 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1321 (unsigned long long)bh
->b_blocknr
);
1325 if (le32_to_cpu(di
->i_fs_generation
) !=
1326 OCFS2_SB(sb
)->fs_generation
) {
1328 "Invalid dinode #%llu: fs_generation is %u\n",
1329 (unsigned long long)bh
->b_blocknr
,
1330 le32_to_cpu(di
->i_fs_generation
));
1340 int ocfs2_read_inode_block_full(struct inode
*inode
, struct buffer_head
**bh
,
1344 struct buffer_head
*tmp
= *bh
;
1346 rc
= ocfs2_read_blocks(inode
, OCFS2_I(inode
)->ip_blkno
, 1, &tmp
,
1347 flags
, ocfs2_validate_inode_block
);
1349 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1356 int ocfs2_read_inode_block(struct inode
*inode
, struct buffer_head
**bh
)
1358 return ocfs2_read_inode_block_full(inode
, bh
, 0);