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/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/quotaops.h>
32 #include <asm/byteorder.h>
34 #include <cluster/masklog.h>
40 #include "blockcheck.h"
42 #include "extent_map.h"
44 #include "heartbeat.h"
54 #include "refcounttree.h"
55 #include "ocfs2_trace.h"
56 #include "filecheck.h"
58 #include "buffer_head_io.h"
60 struct ocfs2_find_inode_args
64 unsigned int fi_flags
;
65 unsigned int fi_sysfile_type
;
68 static struct lock_class_key ocfs2_sysfile_lock_key
[NUM_SYSTEM_INODES
];
70 static int ocfs2_read_locked_inode(struct inode
*inode
,
71 struct ocfs2_find_inode_args
*args
);
72 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
73 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
74 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
76 struct buffer_head
*fe_bh
);
78 static int ocfs2_filecheck_read_inode_block_full(struct inode
*inode
,
79 struct buffer_head
**bh
,
81 static int ocfs2_filecheck_validate_inode_block(struct super_block
*sb
,
82 struct buffer_head
*bh
);
83 static int ocfs2_filecheck_repair_inode_block(struct super_block
*sb
,
84 struct buffer_head
*bh
);
86 void ocfs2_set_inode_flags(struct inode
*inode
)
88 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
90 inode
->i_flags
&= ~(S_IMMUTABLE
|
91 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
93 if (flags
& OCFS2_IMMUTABLE_FL
)
94 inode
->i_flags
|= S_IMMUTABLE
;
96 if (flags
& OCFS2_SYNC_FL
)
97 inode
->i_flags
|= S_SYNC
;
98 if (flags
& OCFS2_APPEND_FL
)
99 inode
->i_flags
|= S_APPEND
;
100 if (flags
& OCFS2_NOATIME_FL
)
101 inode
->i_flags
|= S_NOATIME
;
102 if (flags
& OCFS2_DIRSYNC_FL
)
103 inode
->i_flags
|= S_DIRSYNC
;
106 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
107 void ocfs2_get_inode_flags(struct ocfs2_inode_info
*oi
)
109 unsigned int flags
= oi
->vfs_inode
.i_flags
;
111 oi
->ip_attr
&= ~(OCFS2_SYNC_FL
|OCFS2_APPEND_FL
|
112 OCFS2_IMMUTABLE_FL
|OCFS2_NOATIME_FL
|OCFS2_DIRSYNC_FL
);
114 oi
->ip_attr
|= OCFS2_SYNC_FL
;
115 if (flags
& S_APPEND
)
116 oi
->ip_attr
|= OCFS2_APPEND_FL
;
117 if (flags
& S_IMMUTABLE
)
118 oi
->ip_attr
|= OCFS2_IMMUTABLE_FL
;
119 if (flags
& S_NOATIME
)
120 oi
->ip_attr
|= OCFS2_NOATIME_FL
;
121 if (flags
& S_DIRSYNC
)
122 oi
->ip_attr
|= OCFS2_DIRSYNC_FL
;
125 struct inode
*ocfs2_ilookup(struct super_block
*sb
, u64 blkno
)
127 struct ocfs2_find_inode_args args
;
129 args
.fi_blkno
= blkno
;
131 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
132 args
.fi_sysfile_type
= 0;
134 return ilookup5(sb
, blkno
, ocfs2_find_actor
, &args
);
136 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, unsigned flags
,
140 struct inode
*inode
= NULL
;
141 struct super_block
*sb
= osb
->sb
;
142 struct ocfs2_find_inode_args args
;
143 journal_t
*journal
= OCFS2_SB(sb
)->journal
->j_journal
;
145 trace_ocfs2_iget_begin((unsigned long long)blkno
, flags
,
148 /* Ok. By now we've either got the offsets passed to us by the
149 * caller, or we just pulled them off the bh. Lets do some
150 * sanity checks to make sure they're OK. */
152 inode
= ERR_PTR(-EINVAL
);
153 mlog_errno(PTR_ERR(inode
));
157 args
.fi_blkno
= blkno
;
158 args
.fi_flags
= flags
;
159 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
160 args
.fi_sysfile_type
= sysfile_type
;
162 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
163 ocfs2_init_locked_inode
, &args
);
164 /* inode was *not* in the inode cache. 2.6.x requires
165 * us to do our own read_inode call and unlock it
168 inode
= ERR_PTR(-ENOMEM
);
169 mlog_errno(PTR_ERR(inode
));
172 trace_ocfs2_iget5_locked(inode
->i_state
);
173 if (inode
->i_state
& I_NEW
) {
174 rc
= ocfs2_read_locked_inode(inode
, &args
);
175 unlock_new_inode(inode
);
177 if (is_bad_inode(inode
)) {
179 if ((flags
& OCFS2_FI_FLAG_FILECHECK_CHK
) ||
180 (flags
& OCFS2_FI_FLAG_FILECHECK_FIX
))
181 /* Return OCFS2_FILECHECK_ERR_XXX related errno */
184 inode
= ERR_PTR(-ESTALE
);
189 * Set transaction id's of transactions that have to be committed
190 * to finish f[data]sync. We set them to currently running transaction
191 * as we cannot be sure that the inode or some of its metadata isn't
192 * part of the transaction - the inode could have been reclaimed and
193 * now it is reread from disk.
196 transaction_t
*transaction
;
198 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
200 read_lock(&journal
->j_state_lock
);
201 if (journal
->j_running_transaction
)
202 transaction
= journal
->j_running_transaction
;
204 transaction
= journal
->j_committing_transaction
;
206 tid
= transaction
->t_tid
;
208 tid
= journal
->j_commit_sequence
;
209 read_unlock(&journal
->j_state_lock
);
210 oi
->i_sync_tid
= tid
;
211 oi
->i_datasync_tid
= tid
;
215 if (!IS_ERR(inode
)) {
216 trace_ocfs2_iget_end(inode
,
217 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
225 * here's how inodes get read from disk:
226 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
227 * found? : return the in-memory inode
228 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
231 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
233 struct ocfs2_find_inode_args
*args
= NULL
;
234 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
239 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
241 trace_ocfs2_find_actor(inode
, inode
->i_ino
, opaque
, args
->fi_blkno
);
243 if (oi
->ip_blkno
!= args
->fi_blkno
)
252 * initialize the new inode, but don't do anything that would cause
254 * return 0 on success, 1 on failure
256 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
258 struct ocfs2_find_inode_args
*args
= opaque
;
259 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key
,
260 ocfs2_file_ip_alloc_sem_key
;
262 inode
->i_ino
= args
->fi_ino
;
263 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
264 if (args
->fi_sysfile_type
!= 0)
265 lockdep_set_class(&inode
->i_rwsem
,
266 &ocfs2_sysfile_lock_key
[args
->fi_sysfile_type
]);
267 if (args
->fi_sysfile_type
== USER_QUOTA_SYSTEM_INODE
||
268 args
->fi_sysfile_type
== GROUP_QUOTA_SYSTEM_INODE
||
269 args
->fi_sysfile_type
== LOCAL_USER_QUOTA_SYSTEM_INODE
||
270 args
->fi_sysfile_type
== LOCAL_GROUP_QUOTA_SYSTEM_INODE
)
271 lockdep_set_class(&OCFS2_I(inode
)->ip_alloc_sem
,
272 &ocfs2_quota_ip_alloc_sem_key
);
274 lockdep_set_class(&OCFS2_I(inode
)->ip_alloc_sem
,
275 &ocfs2_file_ip_alloc_sem_key
);
280 void ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
283 struct super_block
*sb
;
284 struct ocfs2_super
*osb
;
290 if ((osb
->s_mount_opt
& OCFS2_MOUNT_LOCALFLOCKS
) ||
291 ocfs2_mount_local(osb
) || !ocfs2_stack_supports_plocks())
295 * These have all been checked by ocfs2_read_inode_block() or set
296 * by ocfs2_mknod_locked(), so a failure is a code bug.
298 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
)); /* This means that read_inode
299 cannot create a superblock
300 inode today. change if
302 BUG_ON(!(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
)));
303 BUG_ON(le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
);
306 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
307 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
308 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
310 inode
->i_version
= 1;
311 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
312 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
313 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
314 i_uid_write(inode
, le32_to_cpu(fe
->i_uid
));
315 i_gid_write(inode
, le32_to_cpu(fe
->i_gid
));
317 /* Fast symlinks will have i_size but no allocated clusters. */
318 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
) {
320 inode
->i_mapping
->a_ops
= &ocfs2_fast_symlink_aops
;
322 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
323 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
325 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
326 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
327 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
328 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
329 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
330 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
332 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
334 "ip_blkno %llu != i_blkno %llu!\n",
335 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
336 (unsigned long long)le64_to_cpu(fe
->i_blkno
));
338 set_nlink(inode
, ocfs2_read_links_count(fe
));
340 trace_ocfs2_populate_inode(OCFS2_I(inode
)->ip_blkno
,
341 le32_to_cpu(fe
->i_flags
));
342 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) {
343 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
344 inode
->i_flags
|= S_NOQUOTA
;
347 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
348 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
349 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
350 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
351 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_QUOTA_FL
)) {
352 inode
->i_flags
|= S_NOQUOTA
;
353 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
354 /* we can't actually hit this as read_inode can't
355 * handle superblocks today ;-) */
359 switch (inode
->i_mode
& S_IFMT
) {
362 inode
->i_fop
= &ocfs2_fops
;
364 inode
->i_fop
= &ocfs2_fops_no_plocks
;
365 inode
->i_op
= &ocfs2_file_iops
;
366 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
369 inode
->i_op
= &ocfs2_dir_iops
;
371 inode
->i_fop
= &ocfs2_dops
;
373 inode
->i_fop
= &ocfs2_dops_no_plocks
;
374 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
375 OCFS2_I(inode
)->ip_dir_lock_gen
= 1;
378 inode
->i_op
= &ocfs2_symlink_inode_operations
;
379 inode_nohighmem(inode
);
380 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
383 inode
->i_op
= &ocfs2_special_file_iops
;
384 init_special_inode(inode
, inode
->i_mode
,
390 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
391 le64_to_cpu(fe
->i_blkno
));
394 * If we ever want to create system files from kernel,
395 * the generation argument to
396 * ocfs2_inode_lock_res_init() will have to change.
398 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
400 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
401 OCFS2_LOCK_TYPE_META
, 0, inode
);
403 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
404 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
407 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
408 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
411 ocfs2_set_inode_flags(inode
);
413 OCFS2_I(inode
)->ip_last_used_slot
= 0;
414 OCFS2_I(inode
)->ip_last_used_group
= 0;
416 if (S_ISDIR(inode
->i_mode
))
417 ocfs2_resv_set_type(&OCFS2_I(inode
)->ip_la_data_resv
,
418 OCFS2_RESV_FLAG_DIR
);
421 static int ocfs2_read_locked_inode(struct inode
*inode
,
422 struct ocfs2_find_inode_args
*args
)
424 struct super_block
*sb
;
425 struct ocfs2_super
*osb
;
426 struct ocfs2_dinode
*fe
;
427 struct buffer_head
*bh
= NULL
;
428 int status
, can_lock
, lock_level
= 0;
436 * To improve performance of cold-cache inode stats, we take
437 * the cluster lock here if possible.
439 * Generally, OCFS2 never trusts the contents of an inode
440 * unless it's holding a cluster lock, so taking it here isn't
441 * a correctness issue as much as it is a performance
444 * There are three times when taking the lock is not a good idea:
446 * 1) During startup, before we have initialized the DLM.
448 * 2) If we are reading certain system files which never get
449 * cluster locks (local alloc, truncate log).
451 * 3) If the process doing the iget() is responsible for
452 * orphan dir recovery. We're holding the orphan dir lock and
453 * can get into a deadlock with another process on another
454 * node in ->delete_inode().
456 * #1 and #2 can be simply solved by never taking the lock
457 * here for system files (which are the only type we read
458 * during mount). It's a heavier approach, but our main
459 * concern is user-accessible files anyway.
461 * #3 works itself out because we'll eventually take the
462 * cluster lock before trusting anything anyway.
464 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
465 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
466 && !ocfs2_mount_local(osb
);
468 trace_ocfs2_read_locked_inode(
469 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, can_lock
);
472 * To maintain backwards compatibility with older versions of
473 * ocfs2-tools, we still store the generation value for system
474 * files. The only ones that actually matter to userspace are
475 * the journals, but it's easier and inexpensive to just flag
476 * all system files similarly.
478 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
479 generation
= osb
->fs_generation
;
481 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
482 OCFS2_LOCK_TYPE_META
,
485 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
486 OCFS2_LOCK_TYPE_OPEN
,
490 status
= ocfs2_open_lock(inode
);
492 make_bad_inode(inode
);
496 status
= ocfs2_inode_lock(inode
, NULL
, lock_level
);
498 make_bad_inode(inode
);
504 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
505 status
= ocfs2_try_open_lock(inode
, 0);
507 make_bad_inode(inode
);
513 if (args
->fi_flags
& OCFS2_FI_FLAG_FILECHECK_CHK
)
514 status
= ocfs2_filecheck_read_inode_block_full(inode
,
515 &bh
, OCFS2_BH_IGNORE_CACHE
, 0);
516 else if (args
->fi_flags
& OCFS2_FI_FLAG_FILECHECK_FIX
)
517 status
= ocfs2_filecheck_read_inode_block_full(inode
,
518 &bh
, OCFS2_BH_IGNORE_CACHE
, 1);
520 status
= ocfs2_read_inode_block_full(inode
,
521 &bh
, OCFS2_BH_IGNORE_CACHE
);
523 status
= ocfs2_read_blocks_sync(osb
, args
->fi_blkno
, 1, &bh
);
525 * If buffer is in jbd, then its checksum may not have been
528 if (!status
&& !buffer_jbd(bh
)) {
529 if (args
->fi_flags
& OCFS2_FI_FLAG_FILECHECK_CHK
)
530 status
= ocfs2_filecheck_validate_inode_block(
532 else if (args
->fi_flags
& OCFS2_FI_FLAG_FILECHECK_FIX
)
533 status
= ocfs2_filecheck_repair_inode_block(
536 status
= ocfs2_validate_inode_block(
546 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
549 * This is a code bug. Right now the caller needs to
550 * understand whether it is asking for a system file inode or
551 * not so the proper lock names can be built.
553 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
554 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
555 "Inode %llu: system file state is ambigous\n",
556 (unsigned long long)args
->fi_blkno
);
558 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
559 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
560 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
562 ocfs2_populate_inode(inode
, fe
, 0);
564 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
566 if (buffer_dirty(bh
) && !buffer_jbd(bh
)) {
568 ocfs2_inode_unlock(inode
, lock_level
);
570 ocfs2_inode_lock(inode
, NULL
, lock_level
);
572 status
= ocfs2_write_block(osb
, bh
, INODE_CACHE(inode
));
583 ocfs2_inode_unlock(inode
, lock_level
);
586 make_bad_inode(inode
);
593 void ocfs2_sync_blockdev(struct super_block
*sb
)
595 sync_blockdev(sb
->s_bdev
);
598 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
600 struct buffer_head
*fe_bh
)
603 struct ocfs2_dinode
*fe
;
604 handle_t
*handle
= NULL
;
606 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
609 * This check will also skip truncate of inodes with inline
610 * data and fast symlinks.
612 if (fe
->i_clusters
) {
613 if (ocfs2_should_order_data(inode
))
614 ocfs2_begin_ordered_truncate(inode
, 0);
616 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
617 if (IS_ERR(handle
)) {
618 status
= PTR_ERR(handle
);
624 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
626 OCFS2_JOURNAL_ACCESS_WRITE
);
632 i_size_write(inode
, 0);
634 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
640 ocfs2_commit_trans(osb
, handle
);
643 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
);
652 ocfs2_commit_trans(osb
, handle
);
656 static int ocfs2_remove_inode(struct inode
*inode
,
657 struct buffer_head
*di_bh
,
658 struct inode
*orphan_dir_inode
,
659 struct buffer_head
*orphan_dir_bh
)
662 struct inode
*inode_alloc_inode
= NULL
;
663 struct buffer_head
*inode_alloc_bh
= NULL
;
665 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
666 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
669 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
670 le16_to_cpu(di
->i_suballoc_slot
));
671 if (!inode_alloc_inode
) {
677 inode_lock(inode_alloc_inode
);
678 status
= ocfs2_inode_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
680 inode_unlock(inode_alloc_inode
);
686 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
+
687 ocfs2_quota_trans_credits(inode
->i_sb
));
688 if (IS_ERR(handle
)) {
689 status
= PTR_ERR(handle
);
694 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SKIP_ORPHAN_DIR
)) {
695 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
696 orphan_dir_bh
, false);
703 /* set the inodes dtime */
704 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
705 OCFS2_JOURNAL_ACCESS_WRITE
);
711 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
712 di
->i_flags
&= cpu_to_le32(~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
713 ocfs2_journal_dirty(handle
, di_bh
);
715 ocfs2_remove_from_cache(INODE_CACHE(inode
), di_bh
);
716 dquot_free_inode(inode
);
718 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
724 ocfs2_commit_trans(osb
, handle
);
726 ocfs2_inode_unlock(inode_alloc_inode
, 1);
727 inode_unlock(inode_alloc_inode
);
728 brelse(inode_alloc_bh
);
730 iput(inode_alloc_inode
);
736 * Serialize with orphan dir recovery. If the process doing
737 * recovery on this orphan dir does an iget() with the dir
738 * i_mutex held, we'll deadlock here. Instead we detect this
739 * and exit early - recovery will wipe this inode for us.
741 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
746 spin_lock(&osb
->osb_lock
);
747 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
751 /* This signals to the orphan recovery process that it should
752 * wait for us to handle the wipe. */
753 osb
->osb_orphan_wipes
[slot
]++;
755 spin_unlock(&osb
->osb_lock
);
756 trace_ocfs2_check_orphan_recovery_state(slot
, ret
);
760 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
763 spin_lock(&osb
->osb_lock
);
764 osb
->osb_orphan_wipes
[slot
]--;
765 spin_unlock(&osb
->osb_lock
);
767 wake_up(&osb
->osb_wipe_event
);
770 static int ocfs2_wipe_inode(struct inode
*inode
,
771 struct buffer_head
*di_bh
)
773 int status
, orphaned_slot
= -1;
774 struct inode
*orphan_dir_inode
= NULL
;
775 struct buffer_head
*orphan_dir_bh
= NULL
;
776 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
777 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
779 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SKIP_ORPHAN_DIR
)) {
780 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
782 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
786 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
787 ORPHAN_DIR_SYSTEM_INODE
,
789 if (!orphan_dir_inode
) {
795 /* Lock the orphan dir. The lock will be held for the entire
796 * delete_inode operation. We do this now to avoid races with
797 * recovery completion on other nodes. */
798 inode_lock(orphan_dir_inode
);
799 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
801 inode_unlock(orphan_dir_inode
);
808 /* we do this while holding the orphan dir lock because we
809 * don't want recovery being run from another node to try an
810 * inode delete underneath us -- this will result in two nodes
811 * truncating the same file! */
812 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
815 goto bail_unlock_dir
;
818 /* Remove any dir index tree */
819 if (S_ISDIR(inode
->i_mode
)) {
820 status
= ocfs2_dx_dir_truncate(inode
, di_bh
);
823 goto bail_unlock_dir
;
827 /*Free extended attribute resources associated with this inode.*/
828 status
= ocfs2_xattr_remove(inode
, di_bh
);
831 goto bail_unlock_dir
;
834 status
= ocfs2_remove_refcount_tree(inode
, di_bh
);
837 goto bail_unlock_dir
;
840 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
846 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_SKIP_ORPHAN_DIR
)
849 ocfs2_inode_unlock(orphan_dir_inode
, 1);
850 inode_unlock(orphan_dir_inode
);
851 brelse(orphan_dir_bh
);
853 iput(orphan_dir_inode
);
854 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
859 /* There is a series of simple checks that should be done before a
860 * trylock is even considered. Encapsulate those in this function. */
861 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
864 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
865 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
867 trace_ocfs2_inode_is_valid_to_delete(current
, osb
->dc_task
,
868 (unsigned long long)oi
->ip_blkno
,
871 /* We shouldn't be getting here for the root directory
873 if (inode
== osb
->root_inode
) {
874 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
879 * If we're coming from downconvert_thread we can't go into our own
880 * voting [hello, deadlock city!] so we cannot delete the inode. But
881 * since we dropped last inode ref when downconverting dentry lock,
882 * we cannot have the file open and thus the node doing unlink will
883 * take care of deleting the inode.
885 if (current
== osb
->dc_task
)
888 spin_lock(&oi
->ip_lock
);
889 /* OCFS2 *never* deletes system files. This should technically
890 * never get here as system file inodes should always have a
891 * positive link count. */
892 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
893 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
894 (unsigned long long)oi
->ip_blkno
);
900 spin_unlock(&oi
->ip_lock
);
905 /* Query the cluster to determine whether we should wipe an inode from
908 * Requires the inode to have the cluster lock. */
909 static int ocfs2_query_inode_wipe(struct inode
*inode
,
910 struct buffer_head
*di_bh
,
913 int status
= 0, reason
= 0;
914 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
915 struct ocfs2_dinode
*di
;
919 trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi
->ip_blkno
,
922 /* While we were waiting for the cluster lock in
923 * ocfs2_delete_inode, another node might have asked to delete
924 * the inode. Recheck our flags to catch this. */
925 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
930 /* Now that we have an up to date inode, we can double check
935 /* Do some basic inode verification... */
936 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
937 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
)) &&
938 !(oi
->ip_flags
& OCFS2_INODE_SKIP_ORPHAN_DIR
)) {
940 * Inodes in the orphan dir must have ORPHANED_FL. The only
941 * inodes that come back out of the orphan dir are reflink
942 * targets. A reflink target may be moved out of the orphan
943 * dir between the time we scan the directory and the time we
944 * process it. This would lead to HAS_REFCOUNT_FL being set but
947 if (di
->i_dyn_features
& cpu_to_le16(OCFS2_HAS_REFCOUNT_FL
)) {
952 /* for lack of a better error? */
955 "Inode %llu (on-disk %llu) not orphaned! "
956 "Disk flags 0x%x, inode flags 0x%x\n",
957 (unsigned long long)oi
->ip_blkno
,
958 (unsigned long long)le64_to_cpu(di
->i_blkno
),
959 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
963 /* has someone already deleted us?! baaad... */
971 * This is how ocfs2 determines whether an inode is still live
972 * within the cluster. Every node takes a shared read lock on
973 * the inode open lock in ocfs2_read_locked_inode(). When we
974 * get to ->delete_inode(), each node tries to convert it's
975 * lock to an exclusive. Trylocks are serialized by the inode
976 * meta data lock. If the upconvert succeeds, we know the inode
977 * is no longer live and can be deleted.
979 * Though we call this with the meta data lock held, the
980 * trylock keeps us from ABBA deadlock.
982 status
= ocfs2_try_open_lock(inode
, 1);
983 if (status
== -EAGAIN
) {
994 trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di
->i_orphaned_slot
));
997 trace_ocfs2_query_inode_wipe_end(status
, reason
);
1001 /* Support function for ocfs2_delete_inode. Will help us keep the
1002 * inode data in a consistent state for clear_inode. Always truncates
1003 * pages, optionally sync's them first. */
1004 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
1007 trace_ocfs2_cleanup_delete_inode(
1008 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
1010 filemap_write_and_wait(inode
->i_mapping
);
1011 truncate_inode_pages_final(&inode
->i_data
);
1014 static void ocfs2_delete_inode(struct inode
*inode
)
1018 struct buffer_head
*di_bh
= NULL
;
1019 struct ocfs2_dinode
*di
= NULL
;
1021 trace_ocfs2_delete_inode(inode
->i_ino
,
1022 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1023 is_bad_inode(inode
));
1025 /* When we fail in read_inode() we mark inode as bad. The second test
1026 * catches the case when inode allocation fails before allocating
1027 * a block for inode. */
1028 if (is_bad_inode(inode
) || !OCFS2_I(inode
)->ip_blkno
)
1031 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
1032 /* It's probably not necessary to truncate_inode_pages
1033 * here but we do it for safety anyway (it will most
1034 * likely be a no-op anyway) */
1035 ocfs2_cleanup_delete_inode(inode
, 0);
1039 dquot_initialize(inode
);
1041 /* We want to block signals in delete_inode as the lock and
1042 * messaging paths may return us -ERESTARTSYS. Which would
1043 * cause us to exit early, resulting in inodes being orphaned
1045 ocfs2_block_signals(&oldset
);
1048 * Synchronize us against ocfs2_get_dentry. We take this in
1049 * shared mode so that all nodes can still concurrently
1052 status
= ocfs2_nfs_sync_lock(OCFS2_SB(inode
->i_sb
), 0);
1054 mlog(ML_ERROR
, "getting nfs sync lock(PR) failed %d\n", status
);
1055 ocfs2_cleanup_delete_inode(inode
, 0);
1058 /* Lock down the inode. This gives us an up to date view of
1059 * it's metadata (for verification), and allows us to
1060 * serialize delete_inode on multiple nodes.
1062 * Even though we might be doing a truncate, we don't take the
1063 * allocation lock here as it won't be needed - nobody will
1064 * have the file open.
1066 status
= ocfs2_inode_lock(inode
, &di_bh
, 1);
1068 if (status
!= -ENOENT
)
1070 ocfs2_cleanup_delete_inode(inode
, 0);
1071 goto bail_unlock_nfs_sync
;
1074 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
1075 /* Skip inode deletion and wait for dio orphan entry recovered
1077 if (unlikely(di
->i_flags
& cpu_to_le32(OCFS2_DIO_ORPHANED_FL
))) {
1078 ocfs2_cleanup_delete_inode(inode
, 0);
1079 goto bail_unlock_inode
;
1082 /* Query the cluster. This will be the final decision made
1083 * before we go ahead and wipe the inode. */
1084 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
1085 if (!wipe
|| status
< 0) {
1086 /* Error and remote inode busy both mean we won't be
1087 * removing the inode, so they take almost the same
1092 /* Someone in the cluster has disallowed a wipe of
1093 * this inode, or it was never completely
1094 * orphaned. Write out the pages and exit now. */
1095 ocfs2_cleanup_delete_inode(inode
, 1);
1096 goto bail_unlock_inode
;
1099 ocfs2_cleanup_delete_inode(inode
, 0);
1101 status
= ocfs2_wipe_inode(inode
, di_bh
);
1103 if (status
!= -EDEADLK
)
1105 goto bail_unlock_inode
;
1109 * Mark the inode as successfully deleted.
1111 * This is important for ocfs2_clear_inode() as it will check
1112 * this flag and skip any checkpointing work
1114 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1115 * the LVB for other nodes.
1117 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
1120 ocfs2_inode_unlock(inode
, 1);
1123 bail_unlock_nfs_sync
:
1124 ocfs2_nfs_sync_unlock(OCFS2_SB(inode
->i_sb
), 0);
1127 ocfs2_unblock_signals(&oldset
);
1132 static void ocfs2_clear_inode(struct inode
*inode
)
1135 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1136 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1139 trace_ocfs2_clear_inode((unsigned long long)oi
->ip_blkno
,
1142 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1143 "Inode=%lu\n", inode
->i_ino
);
1147 /* To preven remote deletes we hold open lock before, now it
1148 * is time to unlock PR and EX open locks. */
1149 ocfs2_open_unlock(inode
);
1151 /* Do these before all the other work so that we don't bounce
1152 * the downconvert thread while waiting to destroy the locks. */
1153 ocfs2_mark_lockres_freeing(osb
, &oi
->ip_rw_lockres
);
1154 ocfs2_mark_lockres_freeing(osb
, &oi
->ip_inode_lockres
);
1155 ocfs2_mark_lockres_freeing(osb
, &oi
->ip_open_lockres
);
1157 ocfs2_resv_discard(&OCFS2_SB(inode
->i_sb
)->osb_la_resmap
,
1158 &oi
->ip_la_data_resv
);
1159 ocfs2_resv_init_once(&oi
->ip_la_data_resv
);
1161 /* We very well may get a clear_inode before all an inodes
1162 * metadata has hit disk. Of course, we can't drop any cluster
1163 * locks until the journal has finished with it. The only
1164 * exception here are successfully wiped inodes - their
1165 * metadata can now be considered to be part of the system
1166 * inodes from which it came. */
1167 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1168 ocfs2_checkpoint_inode(inode
);
1170 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1171 "Clear inode of %llu, inode has io markers\n",
1172 (unsigned long long)oi
->ip_blkno
);
1173 mlog_bug_on_msg(!list_empty(&oi
->ip_unwritten_list
),
1174 "Clear inode of %llu, inode has unwritten extents\n",
1175 (unsigned long long)oi
->ip_blkno
);
1177 ocfs2_extent_map_trunc(inode
, 0);
1179 status
= ocfs2_drop_inode_locks(inode
);
1183 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1184 ocfs2_lock_res_free(&oi
->ip_inode_lockres
);
1185 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1187 ocfs2_metadata_cache_exit(INODE_CACHE(inode
));
1189 mlog_bug_on_msg(INODE_CACHE(inode
)->ci_num_cached
,
1190 "Clear inode of %llu, inode has %u cache items\n",
1191 (unsigned long long)oi
->ip_blkno
,
1192 INODE_CACHE(inode
)->ci_num_cached
);
1194 mlog_bug_on_msg(!(INODE_CACHE(inode
)->ci_flags
& OCFS2_CACHE_FL_INLINE
),
1195 "Clear inode of %llu, inode has a bad flag\n",
1196 (unsigned long long)oi
->ip_blkno
);
1198 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1199 "Clear inode of %llu, inode is locked\n",
1200 (unsigned long long)oi
->ip_blkno
);
1202 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1203 "Clear inode of %llu, io_mutex is locked\n",
1204 (unsigned long long)oi
->ip_blkno
);
1205 mutex_unlock(&oi
->ip_io_mutex
);
1208 * down_trylock() returns 0, down_write_trylock() returns 1
1211 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1212 "Clear inode of %llu, alloc_sem is locked\n",
1213 (unsigned long long)oi
->ip_blkno
);
1214 up_write(&oi
->ip_alloc_sem
);
1216 mlog_bug_on_msg(oi
->ip_open_count
,
1217 "Clear inode of %llu has open count %d\n",
1218 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1220 /* Clear all other flags. */
1222 oi
->ip_dir_start_lookup
= 0;
1223 oi
->ip_blkno
= 0ULL;
1226 * ip_jinode is used to track txns against this inode. We ensure that
1227 * the journal is flushed before journal shutdown. Thus it is safe to
1228 * have inodes get cleaned up after journal shutdown.
1230 jbd2_journal_release_jbd_inode(OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
1234 void ocfs2_evict_inode(struct inode
*inode
)
1236 if (!inode
->i_nlink
||
1237 (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)) {
1238 ocfs2_delete_inode(inode
);
1240 truncate_inode_pages_final(&inode
->i_data
);
1242 ocfs2_clear_inode(inode
);
1245 /* Called under inode_lock, with no more references on the
1246 * struct inode, so it's safe here to check the flags field
1247 * and to manipulate i_nlink without any other locks. */
1248 int ocfs2_drop_inode(struct inode
*inode
)
1250 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1252 trace_ocfs2_drop_inode((unsigned long long)oi
->ip_blkno
,
1253 inode
->i_nlink
, oi
->ip_flags
);
1255 assert_spin_locked(&inode
->i_lock
);
1256 inode
->i_state
|= I_WILL_FREE
;
1257 spin_unlock(&inode
->i_lock
);
1258 write_inode_now(inode
, 1);
1259 spin_lock(&inode
->i_lock
);
1260 WARN_ON(inode
->i_state
& I_NEW
);
1261 inode
->i_state
&= ~I_WILL_FREE
;
1267 * This is called from our getattr.
1269 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1271 struct inode
*inode
= d_inode(dentry
);
1274 trace_ocfs2_inode_revalidate(inode
,
1275 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL,
1276 inode
? (unsigned long long)OCFS2_I(inode
)->ip_flags
: 0);
1283 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1284 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1285 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1289 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1291 /* Let ocfs2_inode_lock do the work of updating our struct
1293 status
= ocfs2_inode_lock(inode
, NULL
, 0);
1295 if (status
!= -ENOENT
)
1299 ocfs2_inode_unlock(inode
, 0);
1305 * Updates a disk inode from a
1307 * Only takes ip_lock.
1309 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1310 struct inode
*inode
,
1311 struct buffer_head
*bh
)
1314 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1316 trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1318 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), bh
,
1319 OCFS2_JOURNAL_ACCESS_WRITE
);
1325 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1326 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1327 ocfs2_get_inode_flags(OCFS2_I(inode
));
1328 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1329 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1330 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1332 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1333 ocfs2_set_links_count(fe
, inode
->i_nlink
);
1334 fe
->i_uid
= cpu_to_le32(i_uid_read(inode
));
1335 fe
->i_gid
= cpu_to_le32(i_gid_read(inode
));
1336 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1337 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1338 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1339 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1340 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1341 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1342 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1344 ocfs2_journal_dirty(handle
, bh
);
1345 ocfs2_update_inode_fsync_trans(handle
, inode
, 1);
1352 * Updates a struct inode from a disk inode.
1353 * does no i/o, only takes ip_lock.
1355 void ocfs2_refresh_inode(struct inode
*inode
,
1356 struct ocfs2_dinode
*fe
)
1358 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1360 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1361 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1362 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1363 ocfs2_set_inode_flags(inode
);
1364 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1365 set_nlink(inode
, ocfs2_read_links_count(fe
));
1366 i_uid_write(inode
, le32_to_cpu(fe
->i_uid
));
1367 i_gid_write(inode
, le32_to_cpu(fe
->i_gid
));
1368 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1369 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1370 inode
->i_blocks
= 0;
1372 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1373 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1374 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1375 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1376 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1377 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1378 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1380 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1383 int ocfs2_validate_inode_block(struct super_block
*sb
,
1384 struct buffer_head
*bh
)
1387 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1389 trace_ocfs2_validate_inode_block((unsigned long long)bh
->b_blocknr
);
1391 BUG_ON(!buffer_uptodate(bh
));
1394 * If the ecc fails, we return the error but otherwise
1395 * leave the filesystem running. We know any error is
1396 * local to this block.
1398 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &di
->i_check
);
1400 mlog(ML_ERROR
, "Checksum failed for dinode %llu\n",
1401 (unsigned long long)bh
->b_blocknr
);
1406 * Errors after here are fatal.
1411 if (!OCFS2_IS_VALID_DINODE(di
)) {
1412 rc
= ocfs2_error(sb
, "Invalid dinode #%llu: signature = %.*s\n",
1413 (unsigned long long)bh
->b_blocknr
, 7,
1418 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1419 rc
= ocfs2_error(sb
, "Invalid dinode #%llu: i_blkno is %llu\n",
1420 (unsigned long long)bh
->b_blocknr
,
1421 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1425 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1426 rc
= ocfs2_error(sb
,
1427 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1428 (unsigned long long)bh
->b_blocknr
);
1432 if (le32_to_cpu(di
->i_fs_generation
) !=
1433 OCFS2_SB(sb
)->fs_generation
) {
1434 rc
= ocfs2_error(sb
,
1435 "Invalid dinode #%llu: fs_generation is %u\n",
1436 (unsigned long long)bh
->b_blocknr
,
1437 le32_to_cpu(di
->i_fs_generation
));
1447 static int ocfs2_filecheck_validate_inode_block(struct super_block
*sb
,
1448 struct buffer_head
*bh
)
1451 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1453 trace_ocfs2_filecheck_validate_inode_block(
1454 (unsigned long long)bh
->b_blocknr
);
1456 BUG_ON(!buffer_uptodate(bh
));
1459 * Call ocfs2_validate_meta_ecc() first since it has ecc repair
1460 * function, but we should not return error immediately when ecc
1461 * validation fails, because the reason is quite likely the invalid
1462 * inode number inputed.
1464 rc
= ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &di
->i_check
);
1467 "Filecheck: checksum failed for dinode %llu\n",
1468 (unsigned long long)bh
->b_blocknr
);
1469 rc
= -OCFS2_FILECHECK_ERR_BLOCKECC
;
1472 if (!OCFS2_IS_VALID_DINODE(di
)) {
1474 "Filecheck: invalid dinode #%llu: signature = %.*s\n",
1475 (unsigned long long)bh
->b_blocknr
, 7, di
->i_signature
);
1476 rc
= -OCFS2_FILECHECK_ERR_INVALIDINO
;
1481 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1483 "Filecheck: invalid dinode #%llu: i_blkno is %llu\n",
1484 (unsigned long long)bh
->b_blocknr
,
1485 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1486 rc
= -OCFS2_FILECHECK_ERR_BLOCKNO
;
1490 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1492 "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL "
1494 (unsigned long long)bh
->b_blocknr
);
1495 rc
= -OCFS2_FILECHECK_ERR_VALIDFLAG
;
1499 if (le32_to_cpu(di
->i_fs_generation
) !=
1500 OCFS2_SB(sb
)->fs_generation
) {
1502 "Filecheck: invalid dinode #%llu: fs_generation is %u\n",
1503 (unsigned long long)bh
->b_blocknr
,
1504 le32_to_cpu(di
->i_fs_generation
));
1505 rc
= -OCFS2_FILECHECK_ERR_GENERATION
;
1513 static int ocfs2_filecheck_repair_inode_block(struct super_block
*sb
,
1514 struct buffer_head
*bh
)
1517 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1519 if (!ocfs2_filecheck_validate_inode_block(sb
, bh
))
1522 trace_ocfs2_filecheck_repair_inode_block(
1523 (unsigned long long)bh
->b_blocknr
);
1525 if (ocfs2_is_hard_readonly(OCFS2_SB(sb
)) ||
1526 ocfs2_is_soft_readonly(OCFS2_SB(sb
))) {
1528 "Filecheck: cannot repair dinode #%llu "
1529 "on readonly filesystem\n",
1530 (unsigned long long)bh
->b_blocknr
);
1531 return -OCFS2_FILECHECK_ERR_READONLY
;
1534 if (buffer_jbd(bh
)) {
1536 "Filecheck: cannot repair dinode #%llu, "
1537 "its buffer is in jbd\n",
1538 (unsigned long long)bh
->b_blocknr
);
1539 return -OCFS2_FILECHECK_ERR_INJBD
;
1542 if (!OCFS2_IS_VALID_DINODE(di
)) {
1543 /* Cannot fix invalid inode block */
1544 return -OCFS2_FILECHECK_ERR_INVALIDINO
;
1547 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1548 /* Cannot just add VALID_FL flag back as a fix,
1549 * need more things to check here.
1551 return -OCFS2_FILECHECK_ERR_VALIDFLAG
;
1554 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1555 di
->i_blkno
= cpu_to_le64(bh
->b_blocknr
);
1558 "Filecheck: reset dinode #%llu: i_blkno to %llu\n",
1559 (unsigned long long)bh
->b_blocknr
,
1560 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1563 if (le32_to_cpu(di
->i_fs_generation
) !=
1564 OCFS2_SB(sb
)->fs_generation
) {
1565 di
->i_fs_generation
= cpu_to_le32(OCFS2_SB(sb
)->fs_generation
);
1568 "Filecheck: reset dinode #%llu: fs_generation to %u\n",
1569 (unsigned long long)bh
->b_blocknr
,
1570 le32_to_cpu(di
->i_fs_generation
));
1573 if (changed
|| ocfs2_validate_meta_ecc(sb
, bh
->b_data
, &di
->i_check
)) {
1574 ocfs2_compute_meta_ecc(sb
, bh
->b_data
, &di
->i_check
);
1575 mark_buffer_dirty(bh
);
1577 "Filecheck: reset dinode #%llu: compute meta ecc\n",
1578 (unsigned long long)bh
->b_blocknr
);
1585 ocfs2_filecheck_read_inode_block_full(struct inode
*inode
,
1586 struct buffer_head
**bh
,
1587 int flags
, int type
)
1590 struct buffer_head
*tmp
= *bh
;
1592 if (!type
) /* Check inode block */
1593 rc
= ocfs2_read_blocks(INODE_CACHE(inode
),
1594 OCFS2_I(inode
)->ip_blkno
,
1596 ocfs2_filecheck_validate_inode_block
);
1597 else /* Repair inode block */
1598 rc
= ocfs2_read_blocks(INODE_CACHE(inode
),
1599 OCFS2_I(inode
)->ip_blkno
,
1601 ocfs2_filecheck_repair_inode_block
);
1603 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1610 int ocfs2_read_inode_block_full(struct inode
*inode
, struct buffer_head
**bh
,
1614 struct buffer_head
*tmp
= *bh
;
1616 rc
= ocfs2_read_blocks(INODE_CACHE(inode
), OCFS2_I(inode
)->ip_blkno
,
1617 1, &tmp
, flags
, ocfs2_validate_inode_block
);
1619 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1626 int ocfs2_read_inode_block(struct inode
*inode
, struct buffer_head
**bh
)
1628 return ocfs2_read_inode_block_full(inode
, bh
, 0);
1632 static u64
ocfs2_inode_cache_owner(struct ocfs2_caching_info
*ci
)
1634 struct ocfs2_inode_info
*oi
= cache_info_to_inode(ci
);
1636 return oi
->ip_blkno
;
1639 static struct super_block
*ocfs2_inode_cache_get_super(struct ocfs2_caching_info
*ci
)
1641 struct ocfs2_inode_info
*oi
= cache_info_to_inode(ci
);
1643 return oi
->vfs_inode
.i_sb
;
1646 static void ocfs2_inode_cache_lock(struct ocfs2_caching_info
*ci
)
1648 struct ocfs2_inode_info
*oi
= cache_info_to_inode(ci
);
1650 spin_lock(&oi
->ip_lock
);
1653 static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info
*ci
)
1655 struct ocfs2_inode_info
*oi
= cache_info_to_inode(ci
);
1657 spin_unlock(&oi
->ip_lock
);
1660 static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info
*ci
)
1662 struct ocfs2_inode_info
*oi
= cache_info_to_inode(ci
);
1664 mutex_lock(&oi
->ip_io_mutex
);
1667 static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info
*ci
)
1669 struct ocfs2_inode_info
*oi
= cache_info_to_inode(ci
);
1671 mutex_unlock(&oi
->ip_io_mutex
);
1674 const struct ocfs2_caching_operations ocfs2_inode_caching_ops
= {
1675 .co_owner
= ocfs2_inode_cache_owner
,
1676 .co_get_super
= ocfs2_inode_cache_get_super
,
1677 .co_cache_lock
= ocfs2_inode_cache_lock
,
1678 .co_cache_unlock
= ocfs2_inode_cache_unlock
,
1679 .co_io_lock
= ocfs2_inode_cache_io_lock
,
1680 .co_io_unlock
= ocfs2_inode_cache_io_unlock
,