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 "extent_map.h"
44 #include "heartbeat.h"
55 #include "buffer_head_io.h"
57 struct ocfs2_find_inode_args
61 unsigned int fi_flags
;
62 unsigned int fi_sysfile_type
;
65 static struct lock_class_key ocfs2_sysfile_lock_key
[NUM_SYSTEM_INODES
];
67 static int ocfs2_read_locked_inode(struct inode
*inode
,
68 struct ocfs2_find_inode_args
*args
);
69 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
70 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
71 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
73 struct buffer_head
*fe_bh
);
75 void ocfs2_set_inode_flags(struct inode
*inode
)
77 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
79 inode
->i_flags
&= ~(S_IMMUTABLE
|
80 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
82 if (flags
& OCFS2_IMMUTABLE_FL
)
83 inode
->i_flags
|= S_IMMUTABLE
;
85 if (flags
& OCFS2_SYNC_FL
)
86 inode
->i_flags
|= S_SYNC
;
87 if (flags
& OCFS2_APPEND_FL
)
88 inode
->i_flags
|= S_APPEND
;
89 if (flags
& OCFS2_NOATIME_FL
)
90 inode
->i_flags
|= S_NOATIME
;
91 if (flags
& OCFS2_DIRSYNC_FL
)
92 inode
->i_flags
|= S_DIRSYNC
;
95 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
96 void ocfs2_get_inode_flags(struct ocfs2_inode_info
*oi
)
98 unsigned int flags
= oi
->vfs_inode
.i_flags
;
100 oi
->ip_attr
&= ~(OCFS2_SYNC_FL
|OCFS2_APPEND_FL
|
101 OCFS2_IMMUTABLE_FL
|OCFS2_NOATIME_FL
|OCFS2_DIRSYNC_FL
);
103 oi
->ip_attr
|= OCFS2_SYNC_FL
;
104 if (flags
& S_APPEND
)
105 oi
->ip_attr
|= OCFS2_APPEND_FL
;
106 if (flags
& S_IMMUTABLE
)
107 oi
->ip_attr
|= OCFS2_IMMUTABLE_FL
;
108 if (flags
& S_NOATIME
)
109 oi
->ip_attr
|= OCFS2_NOATIME_FL
;
110 if (flags
& S_DIRSYNC
)
111 oi
->ip_attr
|= OCFS2_DIRSYNC_FL
;
114 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, unsigned flags
,
117 struct inode
*inode
= NULL
;
118 struct super_block
*sb
= osb
->sb
;
119 struct ocfs2_find_inode_args args
;
121 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
123 /* Ok. By now we've either got the offsets passed to us by the
124 * caller, or we just pulled them off the bh. Lets do some
125 * sanity checks to make sure they're OK. */
127 inode
= ERR_PTR(-EINVAL
);
128 mlog_errno(PTR_ERR(inode
));
132 args
.fi_blkno
= blkno
;
133 args
.fi_flags
= flags
;
134 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
135 args
.fi_sysfile_type
= sysfile_type
;
137 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
138 ocfs2_init_locked_inode
, &args
);
139 /* inode was *not* in the inode cache. 2.6.x requires
140 * us to do our own read_inode call and unlock it
142 if (inode
&& inode
->i_state
& I_NEW
) {
143 mlog(0, "Inode was not in inode cache, reading it.\n");
144 ocfs2_read_locked_inode(inode
, &args
);
145 unlock_new_inode(inode
);
148 inode
= ERR_PTR(-ENOMEM
);
149 mlog_errno(PTR_ERR(inode
));
152 if (is_bad_inode(inode
)) {
154 inode
= ERR_PTR(-ESTALE
);
159 if (!IS_ERR(inode
)) {
160 mlog(0, "returning inode with number %llu\n",
161 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
162 mlog_exit_ptr(inode
);
170 * here's how inodes get read from disk:
171 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
172 * found? : return the in-memory inode
173 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
176 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
178 struct ocfs2_find_inode_args
*args
= NULL
;
179 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
182 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
186 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
188 if (oi
->ip_blkno
!= args
->fi_blkno
)
198 * initialize the new inode, but don't do anything that would cause
200 * return 0 on success, 1 on failure
202 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
204 struct ocfs2_find_inode_args
*args
= opaque
;
206 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
208 inode
->i_ino
= args
->fi_ino
;
209 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
210 if (args
->fi_sysfile_type
!= 0)
211 lockdep_set_class(&inode
->i_mutex
,
212 &ocfs2_sysfile_lock_key
[args
->fi_sysfile_type
]);
218 void ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
221 struct super_block
*sb
;
222 struct ocfs2_super
*osb
;
225 mlog_entry("(0x%p, size:%llu)\n", inode
,
226 (unsigned long long)le64_to_cpu(fe
->i_size
));
231 if ((osb
->s_mount_opt
& OCFS2_MOUNT_LOCALFLOCKS
) ||
232 ocfs2_mount_local(osb
) || !ocfs2_stack_supports_plocks())
236 * These have all been checked by ocfs2_read_inode_block() or set
237 * by ocfs2_mknod_locked(), so a failure is a code bug.
239 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
)); /* This means that read_inode
240 cannot create a superblock
241 inode today. change if
243 BUG_ON(!(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
)));
244 BUG_ON(le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
);
247 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
248 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
249 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
251 inode
->i_version
= 1;
252 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
253 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
254 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
255 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
256 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
258 /* Fast symlinks will have i_size but no allocated clusters. */
259 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
262 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
263 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
264 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
265 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
266 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
267 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
268 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
269 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
271 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
273 "ip_blkno %llu != i_blkno %llu!\n",
274 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
275 (unsigned long long)le64_to_cpu(fe
->i_blkno
));
277 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
279 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) {
280 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
281 inode
->i_flags
|= S_NOQUOTA
;
284 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
285 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
286 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
287 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
288 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
289 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_QUOTA_FL
)) {
290 inode
->i_flags
|= S_NOQUOTA
;
291 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
292 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
293 /* we can't actually hit this as read_inode can't
294 * handle superblocks today ;-) */
298 switch (inode
->i_mode
& S_IFMT
) {
301 inode
->i_fop
= &ocfs2_fops
;
303 inode
->i_fop
= &ocfs2_fops_no_plocks
;
304 inode
->i_op
= &ocfs2_file_iops
;
305 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
308 inode
->i_op
= &ocfs2_dir_iops
;
310 inode
->i_fop
= &ocfs2_dops
;
312 inode
->i_fop
= &ocfs2_dops_no_plocks
;
313 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
316 if (ocfs2_inode_is_fast_symlink(inode
))
317 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
319 inode
->i_op
= &ocfs2_symlink_inode_operations
;
320 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
323 inode
->i_op
= &ocfs2_special_file_iops
;
324 init_special_inode(inode
, inode
->i_mode
,
330 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
331 le64_to_cpu(fe
->i_blkno
));
334 * If we ever want to create system files from kernel,
335 * the generation argument to
336 * ocfs2_inode_lock_res_init() will have to change.
338 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
340 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
341 OCFS2_LOCK_TYPE_META
, 0, inode
);
343 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
344 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
347 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
348 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
351 ocfs2_set_inode_flags(inode
);
356 static int ocfs2_read_locked_inode(struct inode
*inode
,
357 struct ocfs2_find_inode_args
*args
)
359 struct super_block
*sb
;
360 struct ocfs2_super
*osb
;
361 struct ocfs2_dinode
*fe
;
362 struct buffer_head
*bh
= NULL
;
363 int status
, can_lock
;
366 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
369 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
370 mlog(ML_ERROR
, "bad inode\n");
377 mlog(ML_ERROR
, "bad inode args\n");
378 make_bad_inode(inode
);
383 * To improve performance of cold-cache inode stats, we take
384 * the cluster lock here if possible.
386 * Generally, OCFS2 never trusts the contents of an inode
387 * unless it's holding a cluster lock, so taking it here isn't
388 * a correctness issue as much as it is a performance
391 * There are three times when taking the lock is not a good idea:
393 * 1) During startup, before we have initialized the DLM.
395 * 2) If we are reading certain system files which never get
396 * cluster locks (local alloc, truncate log).
398 * 3) If the process doing the iget() is responsible for
399 * orphan dir recovery. We're holding the orphan dir lock and
400 * can get into a deadlock with another process on another
401 * node in ->delete_inode().
403 * #1 and #2 can be simply solved by never taking the lock
404 * here for system files (which are the only type we read
405 * during mount). It's a heavier approach, but our main
406 * concern is user-accesible files anyway.
408 * #3 works itself out because we'll eventually take the
409 * cluster lock before trusting anything anyway.
411 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
412 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
413 && !ocfs2_mount_local(osb
);
416 * To maintain backwards compatibility with older versions of
417 * ocfs2-tools, we still store the generation value for system
418 * files. The only ones that actually matter to userspace are
419 * the journals, but it's easier and inexpensive to just flag
420 * all system files similarly.
422 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
423 generation
= osb
->fs_generation
;
425 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
426 OCFS2_LOCK_TYPE_META
,
429 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
430 OCFS2_LOCK_TYPE_OPEN
,
434 status
= ocfs2_open_lock(inode
);
436 make_bad_inode(inode
);
440 status
= ocfs2_inode_lock(inode
, NULL
, 0);
442 make_bad_inode(inode
);
448 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
449 status
= ocfs2_try_open_lock(inode
, 0);
451 make_bad_inode(inode
);
457 status
= ocfs2_read_inode_block_full(inode
, &bh
,
458 OCFS2_BH_IGNORE_CACHE
);
460 status
= ocfs2_read_blocks_sync(osb
, args
->fi_blkno
, 1, &bh
);
462 status
= ocfs2_validate_inode_block(osb
->sb
, bh
);
470 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
473 * This is a code bug. Right now the caller needs to
474 * understand whether it is asking for a system file inode or
475 * not so the proper lock names can be built.
477 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
478 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
479 "Inode %llu: system file state is ambigous\n",
480 (unsigned long long)args
->fi_blkno
);
482 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
483 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
484 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
486 ocfs2_populate_inode(inode
, fe
, 0);
488 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
494 ocfs2_inode_unlock(inode
, 0);
497 make_bad_inode(inode
);
506 void ocfs2_sync_blockdev(struct super_block
*sb
)
508 sync_blockdev(sb
->s_bdev
);
511 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
513 struct buffer_head
*fe_bh
)
516 struct ocfs2_truncate_context
*tc
= NULL
;
517 struct ocfs2_dinode
*fe
;
518 handle_t
*handle
= NULL
;
522 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
525 * This check will also skip truncate of inodes with inline
526 * data and fast symlinks.
528 if (fe
->i_clusters
) {
529 if (ocfs2_should_order_data(inode
))
530 ocfs2_begin_ordered_truncate(inode
, 0);
532 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
533 if (IS_ERR(handle
)) {
534 status
= PTR_ERR(handle
);
539 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
540 OCFS2_JOURNAL_ACCESS_WRITE
);
546 i_size_write(inode
, 0);
548 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
554 ocfs2_commit_trans(osb
, handle
);
557 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
563 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
572 ocfs2_commit_trans(osb
, handle
);
577 static int ocfs2_remove_inode(struct inode
*inode
,
578 struct buffer_head
*di_bh
,
579 struct inode
*orphan_dir_inode
,
580 struct buffer_head
*orphan_dir_bh
)
583 struct inode
*inode_alloc_inode
= NULL
;
584 struct buffer_head
*inode_alloc_bh
= NULL
;
586 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
587 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
590 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
591 le16_to_cpu(di
->i_suballoc_slot
));
592 if (!inode_alloc_inode
) {
598 mutex_lock(&inode_alloc_inode
->i_mutex
);
599 status
= ocfs2_inode_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
601 mutex_unlock(&inode_alloc_inode
->i_mutex
);
607 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
+
608 ocfs2_quota_trans_credits(inode
->i_sb
));
609 if (IS_ERR(handle
)) {
610 status
= PTR_ERR(handle
);
615 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
622 /* set the inodes dtime */
623 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
624 OCFS2_JOURNAL_ACCESS_WRITE
);
630 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
631 di
->i_flags
&= cpu_to_le32(~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
633 status
= ocfs2_journal_dirty(handle
, di_bh
);
639 ocfs2_remove_from_cache(inode
, di_bh
);
640 vfs_dq_free_inode(inode
);
642 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
648 ocfs2_commit_trans(osb
, handle
);
650 ocfs2_inode_unlock(inode_alloc_inode
, 1);
651 mutex_unlock(&inode_alloc_inode
->i_mutex
);
652 brelse(inode_alloc_bh
);
654 iput(inode_alloc_inode
);
660 * Serialize with orphan dir recovery. If the process doing
661 * recovery on this orphan dir does an iget() with the dir
662 * i_mutex held, we'll deadlock here. Instead we detect this
663 * and exit early - recovery will wipe this inode for us.
665 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
670 spin_lock(&osb
->osb_lock
);
671 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
672 mlog(0, "Recovery is happening on orphan dir %d, will skip "
673 "this inode\n", slot
);
677 /* This signals to the orphan recovery process that it should
678 * wait for us to handle the wipe. */
679 osb
->osb_orphan_wipes
[slot
]++;
681 spin_unlock(&osb
->osb_lock
);
685 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
688 spin_lock(&osb
->osb_lock
);
689 osb
->osb_orphan_wipes
[slot
]--;
690 spin_unlock(&osb
->osb_lock
);
692 wake_up(&osb
->osb_wipe_event
);
695 static int ocfs2_wipe_inode(struct inode
*inode
,
696 struct buffer_head
*di_bh
)
698 int status
, orphaned_slot
;
699 struct inode
*orphan_dir_inode
= NULL
;
700 struct buffer_head
*orphan_dir_bh
= NULL
;
701 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
702 struct ocfs2_dinode
*di
;
704 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
705 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
707 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
711 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
712 ORPHAN_DIR_SYSTEM_INODE
,
714 if (!orphan_dir_inode
) {
720 /* Lock the orphan dir. The lock will be held for the entire
721 * delete_inode operation. We do this now to avoid races with
722 * recovery completion on other nodes. */
723 mutex_lock(&orphan_dir_inode
->i_mutex
);
724 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
726 mutex_unlock(&orphan_dir_inode
->i_mutex
);
732 /* we do this while holding the orphan dir lock because we
733 * don't want recovery being run from another node to try an
734 * inode delete underneath us -- this will result in two nodes
735 * truncating the same file! */
736 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
739 goto bail_unlock_dir
;
742 /*Free extended attribute resources associated with this inode.*/
743 status
= ocfs2_xattr_remove(inode
, di_bh
);
746 goto bail_unlock_dir
;
749 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
755 ocfs2_inode_unlock(orphan_dir_inode
, 1);
756 mutex_unlock(&orphan_dir_inode
->i_mutex
);
757 brelse(orphan_dir_bh
);
759 iput(orphan_dir_inode
);
760 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
765 /* There is a series of simple checks that should be done before a
766 * trylock is even considered. Encapsulate those in this function. */
767 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
770 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
771 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
773 /* We shouldn't be getting here for the root directory
775 if (inode
== osb
->root_inode
) {
776 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
780 /* If we're coming from downconvert_thread we can't go into our own
781 * voting [hello, deadlock city!], so unforuntately we just
782 * have to skip deleting this guy. That's OK though because
783 * the node who's doing the actual deleting should handle it
785 if (current
== osb
->dc_task
) {
786 mlog(0, "Skipping delete of %lu because we're currently "
787 "in downconvert\n", inode
->i_ino
);
791 spin_lock(&oi
->ip_lock
);
792 /* OCFS2 *never* deletes system files. This should technically
793 * never get here as system file inodes should always have a
794 * positive link count. */
795 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
796 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
797 (unsigned long long)oi
->ip_blkno
);
801 /* If we have allowd wipe of this inode for another node, it
802 * will be marked here so we can safely skip it. Recovery will
803 * cleanup any inodes we might inadvertantly skip here. */
804 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
805 mlog(0, "Skipping delete of %lu because another node "
806 "has done this for us.\n", inode
->i_ino
);
812 spin_unlock(&oi
->ip_lock
);
817 /* Query the cluster to determine whether we should wipe an inode from
820 * Requires the inode to have the cluster lock. */
821 static int ocfs2_query_inode_wipe(struct inode
*inode
,
822 struct buffer_head
*di_bh
,
826 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
827 struct ocfs2_dinode
*di
;
831 /* While we were waiting for the cluster lock in
832 * ocfs2_delete_inode, another node might have asked to delete
833 * the inode. Recheck our flags to catch this. */
834 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
835 mlog(0, "Skipping delete of %llu because flags changed\n",
836 (unsigned long long)oi
->ip_blkno
);
840 /* Now that we have an up to date inode, we can double check
842 if (inode
->i_nlink
) {
843 mlog(0, "Skipping delete of %llu because nlink = %u\n",
844 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
848 /* Do some basic inode verification... */
849 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
850 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
851 /* for lack of a better error? */
854 "Inode %llu (on-disk %llu) not orphaned! "
855 "Disk flags 0x%x, inode flags 0x%x\n",
856 (unsigned long long)oi
->ip_blkno
,
857 (unsigned long long)le64_to_cpu(di
->i_blkno
),
858 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
862 /* has someone already deleted us?! baaad... */
870 * This is how ocfs2 determines whether an inode is still live
871 * within the cluster. Every node takes a shared read lock on
872 * the inode open lock in ocfs2_read_locked_inode(). When we
873 * get to ->delete_inode(), each node tries to convert it's
874 * lock to an exclusive. Trylocks are serialized by the inode
875 * meta data lock. If the upconvert suceeds, we know the inode
876 * is no longer live and can be deleted.
878 * Though we call this with the meta data lock held, the
879 * trylock keeps us from ABBA deadlock.
881 status
= ocfs2_try_open_lock(inode
, 1);
882 if (status
== -EAGAIN
) {
884 mlog(0, "Skipping delete of %llu because it is in use on "
885 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
894 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
895 (unsigned long long)oi
->ip_blkno
,
896 le16_to_cpu(di
->i_orphaned_slot
));
902 /* Support function for ocfs2_delete_inode. Will help us keep the
903 * inode data in a consistent state for clear_inode. Always truncates
904 * pages, optionally sync's them first. */
905 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
908 mlog(0, "Cleanup inode %llu, sync = %d\n",
909 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
911 write_inode_now(inode
, 1);
912 truncate_inode_pages(&inode
->i_data
, 0);
915 void ocfs2_delete_inode(struct inode
*inode
)
918 sigset_t blocked
, oldset
;
919 struct buffer_head
*di_bh
= NULL
;
921 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
923 /* When we fail in read_inode() we mark inode as bad. The second test
924 * catches the case when inode allocation fails before allocating
925 * a block for inode. */
926 if (is_bad_inode(inode
) || !OCFS2_I(inode
)->ip_blkno
) {
927 mlog(0, "Skipping delete of bad inode\n");
931 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
932 /* It's probably not necessary to truncate_inode_pages
933 * here but we do it for safety anyway (it will most
934 * likely be a no-op anyway) */
935 ocfs2_cleanup_delete_inode(inode
, 0);
939 /* We want to block signals in delete_inode as the lock and
940 * messaging paths may return us -ERESTARTSYS. Which would
941 * cause us to exit early, resulting in inodes being orphaned
943 sigfillset(&blocked
);
944 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
947 ocfs2_cleanup_delete_inode(inode
, 1);
951 /* Lock down the inode. This gives us an up to date view of
952 * it's metadata (for verification), and allows us to
953 * serialize delete_inode on multiple nodes.
955 * Even though we might be doing a truncate, we don't take the
956 * allocation lock here as it won't be needed - nobody will
957 * have the file open.
959 status
= ocfs2_inode_lock(inode
, &di_bh
, 1);
961 if (status
!= -ENOENT
)
963 ocfs2_cleanup_delete_inode(inode
, 0);
967 /* Query the cluster. This will be the final decision made
968 * before we go ahead and wipe the inode. */
969 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
970 if (!wipe
|| status
< 0) {
971 /* Error and remote inode busy both mean we won't be
972 * removing the inode, so they take almost the same
977 /* Someone in the cluster has disallowed a wipe of
978 * this inode, or it was never completely
979 * orphaned. Write out the pages and exit now. */
980 ocfs2_cleanup_delete_inode(inode
, 1);
981 goto bail_unlock_inode
;
984 ocfs2_cleanup_delete_inode(inode
, 0);
986 status
= ocfs2_wipe_inode(inode
, di_bh
);
988 if (status
!= -EDEADLK
)
990 goto bail_unlock_inode
;
994 * Mark the inode as successfully deleted.
996 * This is important for ocfs2_clear_inode() as it will check
997 * this flag and skip any checkpointing work
999 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1000 * the LVB for other nodes.
1002 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
1005 ocfs2_inode_unlock(inode
, 1);
1008 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
1016 void ocfs2_clear_inode(struct inode
*inode
)
1019 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1026 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1027 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1029 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1030 "Inode=%lu\n", inode
->i_ino
);
1032 /* To preven remote deletes we hold open lock before, now it
1033 * is time to unlock PR and EX open locks. */
1034 ocfs2_open_unlock(inode
);
1036 /* Do these before all the other work so that we don't bounce
1037 * the downconvert thread while waiting to destroy the locks. */
1038 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1039 ocfs2_mark_lockres_freeing(&oi
->ip_inode_lockres
);
1040 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
1042 /* We very well may get a clear_inode before all an inodes
1043 * metadata has hit disk. Of course, we can't drop any cluster
1044 * locks until the journal has finished with it. The only
1045 * exception here are successfully wiped inodes - their
1046 * metadata can now be considered to be part of the system
1047 * inodes from which it came. */
1048 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1049 ocfs2_checkpoint_inode(inode
);
1051 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1052 "Clear inode of %llu, inode has io markers\n",
1053 (unsigned long long)oi
->ip_blkno
);
1055 ocfs2_extent_map_trunc(inode
, 0);
1057 status
= ocfs2_drop_inode_locks(inode
);
1061 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1062 ocfs2_lock_res_free(&oi
->ip_inode_lockres
);
1063 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1065 ocfs2_metadata_cache_purge(inode
);
1067 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1068 "Clear inode of %llu, inode has %u cache items\n",
1069 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1071 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1072 "Clear inode of %llu, inode has a bad flag\n",
1073 (unsigned long long)oi
->ip_blkno
);
1075 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1076 "Clear inode of %llu, inode is locked\n",
1077 (unsigned long long)oi
->ip_blkno
);
1079 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1080 "Clear inode of %llu, io_mutex is locked\n",
1081 (unsigned long long)oi
->ip_blkno
);
1082 mutex_unlock(&oi
->ip_io_mutex
);
1085 * down_trylock() returns 0, down_write_trylock() returns 1
1088 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1089 "Clear inode of %llu, alloc_sem is locked\n",
1090 (unsigned long long)oi
->ip_blkno
);
1091 up_write(&oi
->ip_alloc_sem
);
1093 mlog_bug_on_msg(oi
->ip_open_count
,
1094 "Clear inode of %llu has open count %d\n",
1095 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1097 /* Clear all other flags. */
1098 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1099 oi
->ip_created_trans
= 0;
1100 oi
->ip_last_trans
= 0;
1101 oi
->ip_dir_start_lookup
= 0;
1102 oi
->ip_blkno
= 0ULL;
1105 * ip_jinode is used to track txns against this inode. We ensure that
1106 * the journal is flushed before journal shutdown. Thus it is safe to
1107 * have inodes get cleaned up after journal shutdown.
1109 jbd2_journal_release_jbd_inode(OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
1116 /* Called under inode_lock, with no more references on the
1117 * struct inode, so it's safe here to check the flags field
1118 * and to manipulate i_nlink without any other locks. */
1119 void ocfs2_drop_inode(struct inode
*inode
)
1121 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1125 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1126 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1128 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1129 generic_delete_inode(inode
);
1131 generic_drop_inode(inode
);
1137 * This is called from our getattr.
1139 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1141 struct inode
*inode
= dentry
->d_inode
;
1144 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1145 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1148 mlog(0, "eep, no inode!\n");
1153 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1154 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1155 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1156 mlog(0, "inode deleted!\n");
1160 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1162 /* Let ocfs2_inode_lock do the work of updating our struct
1164 status
= ocfs2_inode_lock(inode
, NULL
, 0);
1166 if (status
!= -ENOENT
)
1170 ocfs2_inode_unlock(inode
, 0);
1178 * Updates a disk inode from a
1180 * Only takes ip_lock.
1182 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1183 struct inode
*inode
,
1184 struct buffer_head
*bh
)
1187 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1189 mlog_entry("(inode %llu)\n",
1190 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1192 status
= ocfs2_journal_access(handle
, inode
, bh
,
1193 OCFS2_JOURNAL_ACCESS_WRITE
);
1199 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1200 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1201 ocfs2_get_inode_flags(OCFS2_I(inode
));
1202 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1203 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1204 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1206 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1207 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1208 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1209 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1210 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1211 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1212 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1213 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1214 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1215 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1216 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1218 status
= ocfs2_journal_dirty(handle
, bh
);
1231 * Updates a struct inode from a disk inode.
1232 * does no i/o, only takes ip_lock.
1234 void ocfs2_refresh_inode(struct inode
*inode
,
1235 struct ocfs2_dinode
*fe
)
1237 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1239 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1240 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1241 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1242 ocfs2_set_inode_flags(inode
);
1243 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1244 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1245 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1246 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1247 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1248 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1249 inode
->i_blocks
= 0;
1251 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1252 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1253 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1254 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1255 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1256 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1257 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1259 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1262 int ocfs2_validate_inode_block(struct super_block
*sb
,
1263 struct buffer_head
*bh
)
1266 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1268 mlog(0, "Validating dinode %llu\n",
1269 (unsigned long long)bh
->b_blocknr
);
1271 BUG_ON(!buffer_uptodate(bh
));
1273 if (!OCFS2_IS_VALID_DINODE(di
)) {
1274 ocfs2_error(sb
, "Invalid dinode #%llu: signature = %.*s\n",
1275 (unsigned long long)bh
->b_blocknr
, 7,
1280 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1281 ocfs2_error(sb
, "Invalid dinode #%llu: i_blkno is %llu\n",
1282 (unsigned long long)bh
->b_blocknr
,
1283 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1287 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1289 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1290 (unsigned long long)bh
->b_blocknr
);
1294 if (le32_to_cpu(di
->i_fs_generation
) !=
1295 OCFS2_SB(sb
)->fs_generation
) {
1297 "Invalid dinode #%llu: fs_generation is %u\n",
1298 (unsigned long long)bh
->b_blocknr
,
1299 le32_to_cpu(di
->i_fs_generation
));
1309 int ocfs2_read_inode_block_full(struct inode
*inode
, struct buffer_head
**bh
,
1313 struct buffer_head
*tmp
= *bh
;
1315 rc
= ocfs2_read_blocks(inode
, OCFS2_I(inode
)->ip_blkno
, 1, &tmp
,
1316 flags
, ocfs2_validate_inode_block
);
1318 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1325 int ocfs2_read_inode_block(struct inode
*inode
, struct buffer_head
**bh
)
1327 return ocfs2_read_inode_block_full(inode
, bh
, 0);