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>
32 #include <asm/byteorder.h>
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
41 #include "extent_map.h"
43 #include "heartbeat.h"
54 #include "buffer_head_io.h"
56 struct ocfs2_find_inode_args
60 unsigned int fi_flags
;
61 unsigned int fi_sysfile_type
;
64 static struct lock_class_key ocfs2_sysfile_lock_key
[NUM_SYSTEM_INODES
];
66 static int ocfs2_read_locked_inode(struct inode
*inode
,
67 struct ocfs2_find_inode_args
*args
);
68 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
69 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
70 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
72 struct buffer_head
*fe_bh
);
74 void ocfs2_set_inode_flags(struct inode
*inode
)
76 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
78 inode
->i_flags
&= ~(S_IMMUTABLE
|
79 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
81 if (flags
& OCFS2_IMMUTABLE_FL
)
82 inode
->i_flags
|= S_IMMUTABLE
;
84 if (flags
& OCFS2_SYNC_FL
)
85 inode
->i_flags
|= S_SYNC
;
86 if (flags
& OCFS2_APPEND_FL
)
87 inode
->i_flags
|= S_APPEND
;
88 if (flags
& OCFS2_NOATIME_FL
)
89 inode
->i_flags
|= S_NOATIME
;
90 if (flags
& OCFS2_DIRSYNC_FL
)
91 inode
->i_flags
|= S_DIRSYNC
;
94 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
95 void ocfs2_get_inode_flags(struct ocfs2_inode_info
*oi
)
97 unsigned int flags
= oi
->vfs_inode
.i_flags
;
99 oi
->ip_attr
&= ~(OCFS2_SYNC_FL
|OCFS2_APPEND_FL
|
100 OCFS2_IMMUTABLE_FL
|OCFS2_NOATIME_FL
|OCFS2_DIRSYNC_FL
);
102 oi
->ip_attr
|= OCFS2_SYNC_FL
;
103 if (flags
& S_APPEND
)
104 oi
->ip_attr
|= OCFS2_APPEND_FL
;
105 if (flags
& S_IMMUTABLE
)
106 oi
->ip_attr
|= OCFS2_IMMUTABLE_FL
;
107 if (flags
& S_NOATIME
)
108 oi
->ip_attr
|= OCFS2_NOATIME_FL
;
109 if (flags
& S_DIRSYNC
)
110 oi
->ip_attr
|= OCFS2_DIRSYNC_FL
;
113 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, unsigned flags
,
116 struct inode
*inode
= NULL
;
117 struct super_block
*sb
= osb
->sb
;
118 struct ocfs2_find_inode_args args
;
120 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
122 /* Ok. By now we've either got the offsets passed to us by the
123 * caller, or we just pulled them off the bh. Lets do some
124 * sanity checks to make sure they're OK. */
126 inode
= ERR_PTR(-EINVAL
);
127 mlog_errno(PTR_ERR(inode
));
131 args
.fi_blkno
= blkno
;
132 args
.fi_flags
= flags
;
133 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
134 args
.fi_sysfile_type
= sysfile_type
;
136 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
137 ocfs2_init_locked_inode
, &args
);
138 /* inode was *not* in the inode cache. 2.6.x requires
139 * us to do our own read_inode call and unlock it
141 if (inode
&& inode
->i_state
& I_NEW
) {
142 mlog(0, "Inode was not in inode cache, reading it.\n");
143 ocfs2_read_locked_inode(inode
, &args
);
144 unlock_new_inode(inode
);
147 inode
= ERR_PTR(-ENOMEM
);
148 mlog_errno(PTR_ERR(inode
));
151 if (is_bad_inode(inode
)) {
153 inode
= ERR_PTR(-ESTALE
);
158 if (!IS_ERR(inode
)) {
159 mlog(0, "returning inode with number %llu\n",
160 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
161 mlog_exit_ptr(inode
);
169 * here's how inodes get read from disk:
170 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
171 * found? : return the in-memory inode
172 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
175 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
177 struct ocfs2_find_inode_args
*args
= NULL
;
178 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
181 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
185 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
187 if (oi
->ip_blkno
!= args
->fi_blkno
)
197 * initialize the new inode, but don't do anything that would cause
199 * return 0 on success, 1 on failure
201 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
203 struct ocfs2_find_inode_args
*args
= opaque
;
205 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
207 inode
->i_ino
= args
->fi_ino
;
208 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
209 if (args
->fi_sysfile_type
!= 0)
210 lockdep_set_class(&inode
->i_mutex
,
211 &ocfs2_sysfile_lock_key
[args
->fi_sysfile_type
]);
217 void ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
220 struct super_block
*sb
;
221 struct ocfs2_super
*osb
;
224 mlog_entry("(0x%p, size:%llu)\n", inode
,
225 (unsigned long long)le64_to_cpu(fe
->i_size
));
230 if ((osb
->s_mount_opt
& OCFS2_MOUNT_LOCALFLOCKS
) ||
231 ocfs2_mount_local(osb
) || !ocfs2_stack_supports_plocks())
235 * These have all been checked by ocfs2_read_inode_block() or set
236 * by ocfs2_mknod_locked(), so a failure is a code bug.
238 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
)); /* This means that read_inode
239 cannot create a superblock
240 inode today. change if
242 BUG_ON(!(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
)));
243 BUG_ON(le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
);
246 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
247 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
248 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
250 inode
->i_version
= 1;
251 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
252 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
253 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
254 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
255 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
257 /* Fast symlinks will have i_size but no allocated clusters. */
258 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
261 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
262 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
263 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
264 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
265 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
266 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
267 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
268 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
270 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
272 "ip_blkno %llu != i_blkno %llu!\n",
273 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
274 (unsigned long long)le64_to_cpu(fe
->i_blkno
));
276 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
278 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
279 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
281 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
282 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
283 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
284 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
285 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
286 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
287 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
288 /* we can't actually hit this as read_inode can't
289 * handle superblocks today ;-) */
293 switch (inode
->i_mode
& S_IFMT
) {
296 inode
->i_fop
= &ocfs2_fops
;
298 inode
->i_fop
= &ocfs2_fops_no_plocks
;
299 inode
->i_op
= &ocfs2_file_iops
;
300 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
303 inode
->i_op
= &ocfs2_dir_iops
;
305 inode
->i_fop
= &ocfs2_dops
;
307 inode
->i_fop
= &ocfs2_dops_no_plocks
;
308 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
311 if (ocfs2_inode_is_fast_symlink(inode
))
312 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
314 inode
->i_op
= &ocfs2_symlink_inode_operations
;
315 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
318 inode
->i_op
= &ocfs2_special_file_iops
;
319 init_special_inode(inode
, inode
->i_mode
,
325 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
326 le64_to_cpu(fe
->i_blkno
));
329 * If we ever want to create system files from kernel,
330 * the generation argument to
331 * ocfs2_inode_lock_res_init() will have to change.
333 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
335 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
336 OCFS2_LOCK_TYPE_META
, 0, inode
);
338 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
339 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
342 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
343 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
346 ocfs2_set_inode_flags(inode
);
351 static int ocfs2_read_locked_inode(struct inode
*inode
,
352 struct ocfs2_find_inode_args
*args
)
354 struct super_block
*sb
;
355 struct ocfs2_super
*osb
;
356 struct ocfs2_dinode
*fe
;
357 struct buffer_head
*bh
= NULL
;
358 int status
, can_lock
;
361 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
364 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
365 mlog(ML_ERROR
, "bad inode\n");
372 mlog(ML_ERROR
, "bad inode args\n");
373 make_bad_inode(inode
);
378 * To improve performance of cold-cache inode stats, we take
379 * the cluster lock here if possible.
381 * Generally, OCFS2 never trusts the contents of an inode
382 * unless it's holding a cluster lock, so taking it here isn't
383 * a correctness issue as much as it is a performance
386 * There are three times when taking the lock is not a good idea:
388 * 1) During startup, before we have initialized the DLM.
390 * 2) If we are reading certain system files which never get
391 * cluster locks (local alloc, truncate log).
393 * 3) If the process doing the iget() is responsible for
394 * orphan dir recovery. We're holding the orphan dir lock and
395 * can get into a deadlock with another process on another
396 * node in ->delete_inode().
398 * #1 and #2 can be simply solved by never taking the lock
399 * here for system files (which are the only type we read
400 * during mount). It's a heavier approach, but our main
401 * concern is user-accesible files anyway.
403 * #3 works itself out because we'll eventually take the
404 * cluster lock before trusting anything anyway.
406 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
407 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
408 && !ocfs2_mount_local(osb
);
411 * To maintain backwards compatibility with older versions of
412 * ocfs2-tools, we still store the generation value for system
413 * files. The only ones that actually matter to userspace are
414 * the journals, but it's easier and inexpensive to just flag
415 * all system files similarly.
417 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
418 generation
= osb
->fs_generation
;
420 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
421 OCFS2_LOCK_TYPE_META
,
424 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
425 OCFS2_LOCK_TYPE_OPEN
,
429 status
= ocfs2_open_lock(inode
);
431 make_bad_inode(inode
);
435 status
= ocfs2_inode_lock(inode
, NULL
, 0);
437 make_bad_inode(inode
);
443 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
444 status
= ocfs2_try_open_lock(inode
, 0);
446 make_bad_inode(inode
);
452 status
= ocfs2_read_inode_block_full(inode
, &bh
,
453 OCFS2_BH_IGNORE_CACHE
);
455 status
= ocfs2_read_blocks_sync(osb
, args
->fi_blkno
, 1, &bh
);
457 status
= ocfs2_validate_inode_block(osb
->sb
, bh
);
465 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
468 * This is a code bug. Right now the caller needs to
469 * understand whether it is asking for a system file inode or
470 * not so the proper lock names can be built.
472 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
473 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
474 "Inode %llu: system file state is ambigous\n",
475 (unsigned long long)args
->fi_blkno
);
477 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
478 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
479 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
481 ocfs2_populate_inode(inode
, fe
, 0);
483 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
489 ocfs2_inode_unlock(inode
, 0);
492 make_bad_inode(inode
);
501 void ocfs2_sync_blockdev(struct super_block
*sb
)
503 sync_blockdev(sb
->s_bdev
);
506 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
508 struct buffer_head
*fe_bh
)
511 struct ocfs2_truncate_context
*tc
= NULL
;
512 struct ocfs2_dinode
*fe
;
513 handle_t
*handle
= NULL
;
517 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
520 * This check will also skip truncate of inodes with inline
521 * data and fast symlinks.
523 if (fe
->i_clusters
) {
524 if (ocfs2_should_order_data(inode
))
525 ocfs2_begin_ordered_truncate(inode
, 0);
527 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
528 if (IS_ERR(handle
)) {
529 status
= PTR_ERR(handle
);
534 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
535 OCFS2_JOURNAL_ACCESS_WRITE
);
541 i_size_write(inode
, 0);
543 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
549 ocfs2_commit_trans(osb
, handle
);
552 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
558 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
567 ocfs2_commit_trans(osb
, handle
);
572 static int ocfs2_remove_inode(struct inode
*inode
,
573 struct buffer_head
*di_bh
,
574 struct inode
*orphan_dir_inode
,
575 struct buffer_head
*orphan_dir_bh
)
578 struct inode
*inode_alloc_inode
= NULL
;
579 struct buffer_head
*inode_alloc_bh
= NULL
;
581 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
582 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
585 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
586 le16_to_cpu(di
->i_suballoc_slot
));
587 if (!inode_alloc_inode
) {
593 mutex_lock(&inode_alloc_inode
->i_mutex
);
594 status
= ocfs2_inode_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
596 mutex_unlock(&inode_alloc_inode
->i_mutex
);
602 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
603 if (IS_ERR(handle
)) {
604 status
= PTR_ERR(handle
);
609 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
616 /* set the inodes dtime */
617 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
618 OCFS2_JOURNAL_ACCESS_WRITE
);
624 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
625 di
->i_flags
&= cpu_to_le32(~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
627 status
= ocfs2_journal_dirty(handle
, di_bh
);
633 ocfs2_remove_from_cache(inode
, di_bh
);
635 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
641 ocfs2_commit_trans(osb
, handle
);
643 ocfs2_inode_unlock(inode_alloc_inode
, 1);
644 mutex_unlock(&inode_alloc_inode
->i_mutex
);
645 brelse(inode_alloc_bh
);
647 iput(inode_alloc_inode
);
653 * Serialize with orphan dir recovery. If the process doing
654 * recovery on this orphan dir does an iget() with the dir
655 * i_mutex held, we'll deadlock here. Instead we detect this
656 * and exit early - recovery will wipe this inode for us.
658 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
663 spin_lock(&osb
->osb_lock
);
664 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
665 mlog(0, "Recovery is happening on orphan dir %d, will skip "
666 "this inode\n", slot
);
670 /* This signals to the orphan recovery process that it should
671 * wait for us to handle the wipe. */
672 osb
->osb_orphan_wipes
[slot
]++;
674 spin_unlock(&osb
->osb_lock
);
678 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
681 spin_lock(&osb
->osb_lock
);
682 osb
->osb_orphan_wipes
[slot
]--;
683 spin_unlock(&osb
->osb_lock
);
685 wake_up(&osb
->osb_wipe_event
);
688 static int ocfs2_wipe_inode(struct inode
*inode
,
689 struct buffer_head
*di_bh
)
691 int status
, orphaned_slot
;
692 struct inode
*orphan_dir_inode
= NULL
;
693 struct buffer_head
*orphan_dir_bh
= NULL
;
694 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
695 struct ocfs2_dinode
*di
;
697 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
698 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
700 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
704 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
705 ORPHAN_DIR_SYSTEM_INODE
,
707 if (!orphan_dir_inode
) {
713 /* Lock the orphan dir. The lock will be held for the entire
714 * delete_inode operation. We do this now to avoid races with
715 * recovery completion on other nodes. */
716 mutex_lock(&orphan_dir_inode
->i_mutex
);
717 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
719 mutex_unlock(&orphan_dir_inode
->i_mutex
);
725 /* we do this while holding the orphan dir lock because we
726 * don't want recovery being run from another node to try an
727 * inode delete underneath us -- this will result in two nodes
728 * truncating the same file! */
729 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
732 goto bail_unlock_dir
;
735 /*Free extended attribute resources associated with this inode.*/
736 status
= ocfs2_xattr_remove(inode
, di_bh
);
739 goto bail_unlock_dir
;
742 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
748 ocfs2_inode_unlock(orphan_dir_inode
, 1);
749 mutex_unlock(&orphan_dir_inode
->i_mutex
);
750 brelse(orphan_dir_bh
);
752 iput(orphan_dir_inode
);
753 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
758 /* There is a series of simple checks that should be done before a
759 * trylock is even considered. Encapsulate those in this function. */
760 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
763 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
764 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
766 /* We shouldn't be getting here for the root directory
768 if (inode
== osb
->root_inode
) {
769 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
773 /* If we're coming from downconvert_thread we can't go into our own
774 * voting [hello, deadlock city!], so unforuntately we just
775 * have to skip deleting this guy. That's OK though because
776 * the node who's doing the actual deleting should handle it
778 if (current
== osb
->dc_task
) {
779 mlog(0, "Skipping delete of %lu because we're currently "
780 "in downconvert\n", inode
->i_ino
);
784 spin_lock(&oi
->ip_lock
);
785 /* OCFS2 *never* deletes system files. This should technically
786 * never get here as system file inodes should always have a
787 * positive link count. */
788 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
789 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
790 (unsigned long long)oi
->ip_blkno
);
794 /* If we have allowd wipe of this inode for another node, it
795 * will be marked here so we can safely skip it. Recovery will
796 * cleanup any inodes we might inadvertantly skip here. */
797 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
798 mlog(0, "Skipping delete of %lu because another node "
799 "has done this for us.\n", inode
->i_ino
);
805 spin_unlock(&oi
->ip_lock
);
810 /* Query the cluster to determine whether we should wipe an inode from
813 * Requires the inode to have the cluster lock. */
814 static int ocfs2_query_inode_wipe(struct inode
*inode
,
815 struct buffer_head
*di_bh
,
819 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
820 struct ocfs2_dinode
*di
;
824 /* While we were waiting for the cluster lock in
825 * ocfs2_delete_inode, another node might have asked to delete
826 * the inode. Recheck our flags to catch this. */
827 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
828 mlog(0, "Skipping delete of %llu because flags changed\n",
829 (unsigned long long)oi
->ip_blkno
);
833 /* Now that we have an up to date inode, we can double check
835 if (inode
->i_nlink
) {
836 mlog(0, "Skipping delete of %llu because nlink = %u\n",
837 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
841 /* Do some basic inode verification... */
842 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
843 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
844 /* for lack of a better error? */
847 "Inode %llu (on-disk %llu) not orphaned! "
848 "Disk flags 0x%x, inode flags 0x%x\n",
849 (unsigned long long)oi
->ip_blkno
,
850 (unsigned long long)le64_to_cpu(di
->i_blkno
),
851 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
855 /* has someone already deleted us?! baaad... */
863 * This is how ocfs2 determines whether an inode is still live
864 * within the cluster. Every node takes a shared read lock on
865 * the inode open lock in ocfs2_read_locked_inode(). When we
866 * get to ->delete_inode(), each node tries to convert it's
867 * lock to an exclusive. Trylocks are serialized by the inode
868 * meta data lock. If the upconvert suceeds, we know the inode
869 * is no longer live and can be deleted.
871 * Though we call this with the meta data lock held, the
872 * trylock keeps us from ABBA deadlock.
874 status
= ocfs2_try_open_lock(inode
, 1);
875 if (status
== -EAGAIN
) {
877 mlog(0, "Skipping delete of %llu because it is in use on "
878 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
887 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
888 (unsigned long long)oi
->ip_blkno
,
889 le16_to_cpu(di
->i_orphaned_slot
));
895 /* Support function for ocfs2_delete_inode. Will help us keep the
896 * inode data in a consistent state for clear_inode. Always truncates
897 * pages, optionally sync's them first. */
898 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
901 mlog(0, "Cleanup inode %llu, sync = %d\n",
902 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
904 write_inode_now(inode
, 1);
905 truncate_inode_pages(&inode
->i_data
, 0);
908 void ocfs2_delete_inode(struct inode
*inode
)
911 sigset_t blocked
, oldset
;
912 struct buffer_head
*di_bh
= NULL
;
914 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
916 if (is_bad_inode(inode
)) {
917 mlog(0, "Skipping delete of bad inode\n");
921 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
922 /* It's probably not necessary to truncate_inode_pages
923 * here but we do it for safety anyway (it will most
924 * likely be a no-op anyway) */
925 ocfs2_cleanup_delete_inode(inode
, 0);
929 /* We want to block signals in delete_inode as the lock and
930 * messaging paths may return us -ERESTARTSYS. Which would
931 * cause us to exit early, resulting in inodes being orphaned
933 sigfillset(&blocked
);
934 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
937 ocfs2_cleanup_delete_inode(inode
, 1);
941 /* Lock down the inode. This gives us an up to date view of
942 * it's metadata (for verification), and allows us to
943 * serialize delete_inode on multiple nodes.
945 * Even though we might be doing a truncate, we don't take the
946 * allocation lock here as it won't be needed - nobody will
947 * have the file open.
949 status
= ocfs2_inode_lock(inode
, &di_bh
, 1);
951 if (status
!= -ENOENT
)
953 ocfs2_cleanup_delete_inode(inode
, 0);
957 /* Query the cluster. This will be the final decision made
958 * before we go ahead and wipe the inode. */
959 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
960 if (!wipe
|| status
< 0) {
961 /* Error and remote inode busy both mean we won't be
962 * removing the inode, so they take almost the same
967 /* Someone in the cluster has disallowed a wipe of
968 * this inode, or it was never completely
969 * orphaned. Write out the pages and exit now. */
970 ocfs2_cleanup_delete_inode(inode
, 1);
971 goto bail_unlock_inode
;
974 ocfs2_cleanup_delete_inode(inode
, 0);
976 status
= ocfs2_wipe_inode(inode
, di_bh
);
978 if (status
!= -EDEADLK
)
980 goto bail_unlock_inode
;
984 * Mark the inode as successfully deleted.
986 * This is important for ocfs2_clear_inode() as it will check
987 * this flag and skip any checkpointing work
989 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
990 * the LVB for other nodes.
992 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
995 ocfs2_inode_unlock(inode
, 1);
998 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
1006 void ocfs2_clear_inode(struct inode
*inode
)
1009 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1016 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1017 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1019 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1020 "Inode=%lu\n", inode
->i_ino
);
1022 /* To preven remote deletes we hold open lock before, now it
1023 * is time to unlock PR and EX open locks. */
1024 ocfs2_open_unlock(inode
);
1026 /* Do these before all the other work so that we don't bounce
1027 * the downconvert thread while waiting to destroy the locks. */
1028 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1029 ocfs2_mark_lockres_freeing(&oi
->ip_inode_lockres
);
1030 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
1032 /* We very well may get a clear_inode before all an inodes
1033 * metadata has hit disk. Of course, we can't drop any cluster
1034 * locks until the journal has finished with it. The only
1035 * exception here are successfully wiped inodes - their
1036 * metadata can now be considered to be part of the system
1037 * inodes from which it came. */
1038 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1039 ocfs2_checkpoint_inode(inode
);
1041 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1042 "Clear inode of %llu, inode has io markers\n",
1043 (unsigned long long)oi
->ip_blkno
);
1045 ocfs2_extent_map_trunc(inode
, 0);
1047 status
= ocfs2_drop_inode_locks(inode
);
1051 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1052 ocfs2_lock_res_free(&oi
->ip_inode_lockres
);
1053 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1055 ocfs2_metadata_cache_purge(inode
);
1057 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1058 "Clear inode of %llu, inode has %u cache items\n",
1059 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1061 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1062 "Clear inode of %llu, inode has a bad flag\n",
1063 (unsigned long long)oi
->ip_blkno
);
1065 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1066 "Clear inode of %llu, inode is locked\n",
1067 (unsigned long long)oi
->ip_blkno
);
1069 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1070 "Clear inode of %llu, io_mutex is locked\n",
1071 (unsigned long long)oi
->ip_blkno
);
1072 mutex_unlock(&oi
->ip_io_mutex
);
1075 * down_trylock() returns 0, down_write_trylock() returns 1
1078 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1079 "Clear inode of %llu, alloc_sem is locked\n",
1080 (unsigned long long)oi
->ip_blkno
);
1081 up_write(&oi
->ip_alloc_sem
);
1083 mlog_bug_on_msg(oi
->ip_open_count
,
1084 "Clear inode of %llu has open count %d\n",
1085 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1087 /* Clear all other flags. */
1088 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1089 oi
->ip_created_trans
= 0;
1090 oi
->ip_last_trans
= 0;
1091 oi
->ip_dir_start_lookup
= 0;
1092 oi
->ip_blkno
= 0ULL;
1095 * ip_jinode is used to track txns against this inode. We ensure that
1096 * the journal is flushed before journal shutdown. Thus it is safe to
1097 * have inodes get cleaned up after journal shutdown.
1099 jbd2_journal_release_jbd_inode(OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
1106 /* Called under inode_lock, with no more references on the
1107 * struct inode, so it's safe here to check the flags field
1108 * and to manipulate i_nlink without any other locks. */
1109 void ocfs2_drop_inode(struct inode
*inode
)
1111 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1115 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1116 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1118 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1119 generic_delete_inode(inode
);
1121 generic_drop_inode(inode
);
1127 * This is called from our getattr.
1129 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1131 struct inode
*inode
= dentry
->d_inode
;
1134 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1135 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1138 mlog(0, "eep, no inode!\n");
1143 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1144 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1145 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1146 mlog(0, "inode deleted!\n");
1150 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1152 /* Let ocfs2_inode_lock do the work of updating our struct
1154 status
= ocfs2_inode_lock(inode
, NULL
, 0);
1156 if (status
!= -ENOENT
)
1160 ocfs2_inode_unlock(inode
, 0);
1168 * Updates a disk inode from a
1170 * Only takes ip_lock.
1172 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1173 struct inode
*inode
,
1174 struct buffer_head
*bh
)
1177 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1179 mlog_entry("(inode %llu)\n",
1180 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1182 status
= ocfs2_journal_access(handle
, inode
, bh
,
1183 OCFS2_JOURNAL_ACCESS_WRITE
);
1189 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1190 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1191 ocfs2_get_inode_flags(OCFS2_I(inode
));
1192 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1193 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1194 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1196 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1197 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1198 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1199 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1200 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1201 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1202 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1203 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1204 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1205 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1206 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1208 status
= ocfs2_journal_dirty(handle
, bh
);
1221 * Updates a struct inode from a disk inode.
1222 * does no i/o, only takes ip_lock.
1224 void ocfs2_refresh_inode(struct inode
*inode
,
1225 struct ocfs2_dinode
*fe
)
1227 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1229 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1230 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1231 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1232 ocfs2_set_inode_flags(inode
);
1233 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1234 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1235 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1236 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1237 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1238 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1239 inode
->i_blocks
= 0;
1241 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1242 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1243 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1244 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1245 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1246 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1247 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1249 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1252 int ocfs2_validate_inode_block(struct super_block
*sb
,
1253 struct buffer_head
*bh
)
1256 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1258 mlog(0, "Validating dinode %llu\n",
1259 (unsigned long long)bh
->b_blocknr
);
1261 BUG_ON(!buffer_uptodate(bh
));
1263 if (!OCFS2_IS_VALID_DINODE(di
)) {
1264 ocfs2_error(sb
, "Invalid dinode #%llu: signature = %.*s\n",
1265 (unsigned long long)bh
->b_blocknr
, 7,
1270 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1271 ocfs2_error(sb
, "Invalid dinode #%llu: i_blkno is %llu\n",
1272 (unsigned long long)bh
->b_blocknr
,
1273 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1277 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1279 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1280 (unsigned long long)bh
->b_blocknr
);
1284 if (le32_to_cpu(di
->i_fs_generation
) !=
1285 OCFS2_SB(sb
)->fs_generation
) {
1287 "Invalid dinode #%llu: fs_generation is %u\n",
1288 (unsigned long long)bh
->b_blocknr
,
1289 le32_to_cpu(di
->i_fs_generation
));
1299 int ocfs2_read_inode_block_full(struct inode
*inode
, struct buffer_head
**bh
,
1303 struct buffer_head
*tmp
= *bh
;
1305 rc
= ocfs2_read_blocks(inode
, OCFS2_I(inode
)->ip_blkno
, 1, &tmp
,
1306 flags
, ocfs2_validate_inode_block
);
1308 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1315 int ocfs2_read_inode_block(struct inode
*inode
, struct buffer_head
**bh
)
1317 return ocfs2_read_inode_block_full(inode
, bh
, 0);