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_QUOTA_FL
)) {
287 inode
->i_flags
|= S_NOQUOTA
;
288 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
289 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
290 /* we can't actually hit this as read_inode can't
291 * handle superblocks today ;-) */
295 switch (inode
->i_mode
& S_IFMT
) {
298 inode
->i_fop
= &ocfs2_fops
;
300 inode
->i_fop
= &ocfs2_fops_no_plocks
;
301 inode
->i_op
= &ocfs2_file_iops
;
302 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
305 inode
->i_op
= &ocfs2_dir_iops
;
307 inode
->i_fop
= &ocfs2_dops
;
309 inode
->i_fop
= &ocfs2_dops_no_plocks
;
310 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
313 if (ocfs2_inode_is_fast_symlink(inode
))
314 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
316 inode
->i_op
= &ocfs2_symlink_inode_operations
;
317 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
320 inode
->i_op
= &ocfs2_special_file_iops
;
321 init_special_inode(inode
, inode
->i_mode
,
327 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
328 le64_to_cpu(fe
->i_blkno
));
331 * If we ever want to create system files from kernel,
332 * the generation argument to
333 * ocfs2_inode_lock_res_init() will have to change.
335 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
337 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
338 OCFS2_LOCK_TYPE_META
, 0, inode
);
340 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
341 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
344 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
345 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
348 ocfs2_set_inode_flags(inode
);
353 static int ocfs2_read_locked_inode(struct inode
*inode
,
354 struct ocfs2_find_inode_args
*args
)
356 struct super_block
*sb
;
357 struct ocfs2_super
*osb
;
358 struct ocfs2_dinode
*fe
;
359 struct buffer_head
*bh
= NULL
;
360 int status
, can_lock
;
363 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
366 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
367 mlog(ML_ERROR
, "bad inode\n");
374 mlog(ML_ERROR
, "bad inode args\n");
375 make_bad_inode(inode
);
380 * To improve performance of cold-cache inode stats, we take
381 * the cluster lock here if possible.
383 * Generally, OCFS2 never trusts the contents of an inode
384 * unless it's holding a cluster lock, so taking it here isn't
385 * a correctness issue as much as it is a performance
388 * There are three times when taking the lock is not a good idea:
390 * 1) During startup, before we have initialized the DLM.
392 * 2) If we are reading certain system files which never get
393 * cluster locks (local alloc, truncate log).
395 * 3) If the process doing the iget() is responsible for
396 * orphan dir recovery. We're holding the orphan dir lock and
397 * can get into a deadlock with another process on another
398 * node in ->delete_inode().
400 * #1 and #2 can be simply solved by never taking the lock
401 * here for system files (which are the only type we read
402 * during mount). It's a heavier approach, but our main
403 * concern is user-accesible files anyway.
405 * #3 works itself out because we'll eventually take the
406 * cluster lock before trusting anything anyway.
408 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
409 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
410 && !ocfs2_mount_local(osb
);
413 * To maintain backwards compatibility with older versions of
414 * ocfs2-tools, we still store the generation value for system
415 * files. The only ones that actually matter to userspace are
416 * the journals, but it's easier and inexpensive to just flag
417 * all system files similarly.
419 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
420 generation
= osb
->fs_generation
;
422 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
423 OCFS2_LOCK_TYPE_META
,
426 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
427 OCFS2_LOCK_TYPE_OPEN
,
431 status
= ocfs2_open_lock(inode
);
433 make_bad_inode(inode
);
437 status
= ocfs2_inode_lock(inode
, NULL
, 0);
439 make_bad_inode(inode
);
445 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
446 status
= ocfs2_try_open_lock(inode
, 0);
448 make_bad_inode(inode
);
454 status
= ocfs2_read_inode_block_full(inode
, &bh
,
455 OCFS2_BH_IGNORE_CACHE
);
457 status
= ocfs2_read_blocks_sync(osb
, args
->fi_blkno
, 1, &bh
);
459 status
= ocfs2_validate_inode_block(osb
->sb
, bh
);
467 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
470 * This is a code bug. Right now the caller needs to
471 * understand whether it is asking for a system file inode or
472 * not so the proper lock names can be built.
474 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
475 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
476 "Inode %llu: system file state is ambigous\n",
477 (unsigned long long)args
->fi_blkno
);
479 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
480 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
481 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
483 ocfs2_populate_inode(inode
, fe
, 0);
485 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
491 ocfs2_inode_unlock(inode
, 0);
494 make_bad_inode(inode
);
503 void ocfs2_sync_blockdev(struct super_block
*sb
)
505 sync_blockdev(sb
->s_bdev
);
508 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
510 struct buffer_head
*fe_bh
)
513 struct ocfs2_truncate_context
*tc
= NULL
;
514 struct ocfs2_dinode
*fe
;
515 handle_t
*handle
= NULL
;
519 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
522 * This check will also skip truncate of inodes with inline
523 * data and fast symlinks.
525 if (fe
->i_clusters
) {
526 if (ocfs2_should_order_data(inode
))
527 ocfs2_begin_ordered_truncate(inode
, 0);
529 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
530 if (IS_ERR(handle
)) {
531 status
= PTR_ERR(handle
);
536 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
537 OCFS2_JOURNAL_ACCESS_WRITE
);
543 i_size_write(inode
, 0);
545 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
551 ocfs2_commit_trans(osb
, handle
);
554 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
560 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
569 ocfs2_commit_trans(osb
, handle
);
574 static int ocfs2_remove_inode(struct inode
*inode
,
575 struct buffer_head
*di_bh
,
576 struct inode
*orphan_dir_inode
,
577 struct buffer_head
*orphan_dir_bh
)
580 struct inode
*inode_alloc_inode
= NULL
;
581 struct buffer_head
*inode_alloc_bh
= NULL
;
583 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
584 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
587 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
588 le16_to_cpu(di
->i_suballoc_slot
));
589 if (!inode_alloc_inode
) {
595 mutex_lock(&inode_alloc_inode
->i_mutex
);
596 status
= ocfs2_inode_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
598 mutex_unlock(&inode_alloc_inode
->i_mutex
);
604 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
605 if (IS_ERR(handle
)) {
606 status
= PTR_ERR(handle
);
611 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
618 /* set the inodes dtime */
619 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
620 OCFS2_JOURNAL_ACCESS_WRITE
);
626 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
627 di
->i_flags
&= cpu_to_le32(~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
629 status
= ocfs2_journal_dirty(handle
, di_bh
);
635 ocfs2_remove_from_cache(inode
, di_bh
);
637 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
643 ocfs2_commit_trans(osb
, handle
);
645 ocfs2_inode_unlock(inode_alloc_inode
, 1);
646 mutex_unlock(&inode_alloc_inode
->i_mutex
);
647 brelse(inode_alloc_bh
);
649 iput(inode_alloc_inode
);
655 * Serialize with orphan dir recovery. If the process doing
656 * recovery on this orphan dir does an iget() with the dir
657 * i_mutex held, we'll deadlock here. Instead we detect this
658 * and exit early - recovery will wipe this inode for us.
660 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
665 spin_lock(&osb
->osb_lock
);
666 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
667 mlog(0, "Recovery is happening on orphan dir %d, will skip "
668 "this inode\n", slot
);
672 /* This signals to the orphan recovery process that it should
673 * wait for us to handle the wipe. */
674 osb
->osb_orphan_wipes
[slot
]++;
676 spin_unlock(&osb
->osb_lock
);
680 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
683 spin_lock(&osb
->osb_lock
);
684 osb
->osb_orphan_wipes
[slot
]--;
685 spin_unlock(&osb
->osb_lock
);
687 wake_up(&osb
->osb_wipe_event
);
690 static int ocfs2_wipe_inode(struct inode
*inode
,
691 struct buffer_head
*di_bh
)
693 int status
, orphaned_slot
;
694 struct inode
*orphan_dir_inode
= NULL
;
695 struct buffer_head
*orphan_dir_bh
= NULL
;
696 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
697 struct ocfs2_dinode
*di
;
699 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
700 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
702 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
706 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
707 ORPHAN_DIR_SYSTEM_INODE
,
709 if (!orphan_dir_inode
) {
715 /* Lock the orphan dir. The lock will be held for the entire
716 * delete_inode operation. We do this now to avoid races with
717 * recovery completion on other nodes. */
718 mutex_lock(&orphan_dir_inode
->i_mutex
);
719 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
721 mutex_unlock(&orphan_dir_inode
->i_mutex
);
727 /* we do this while holding the orphan dir lock because we
728 * don't want recovery being run from another node to try an
729 * inode delete underneath us -- this will result in two nodes
730 * truncating the same file! */
731 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
734 goto bail_unlock_dir
;
737 /*Free extended attribute resources associated with this inode.*/
738 status
= ocfs2_xattr_remove(inode
, di_bh
);
741 goto bail_unlock_dir
;
744 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
750 ocfs2_inode_unlock(orphan_dir_inode
, 1);
751 mutex_unlock(&orphan_dir_inode
->i_mutex
);
752 brelse(orphan_dir_bh
);
754 iput(orphan_dir_inode
);
755 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
760 /* There is a series of simple checks that should be done before a
761 * trylock is even considered. Encapsulate those in this function. */
762 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
765 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
766 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
768 /* We shouldn't be getting here for the root directory
770 if (inode
== osb
->root_inode
) {
771 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
775 /* If we're coming from downconvert_thread we can't go into our own
776 * voting [hello, deadlock city!], so unforuntately we just
777 * have to skip deleting this guy. That's OK though because
778 * the node who's doing the actual deleting should handle it
780 if (current
== osb
->dc_task
) {
781 mlog(0, "Skipping delete of %lu because we're currently "
782 "in downconvert\n", inode
->i_ino
);
786 spin_lock(&oi
->ip_lock
);
787 /* OCFS2 *never* deletes system files. This should technically
788 * never get here as system file inodes should always have a
789 * positive link count. */
790 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
791 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
792 (unsigned long long)oi
->ip_blkno
);
796 /* If we have allowd wipe of this inode for another node, it
797 * will be marked here so we can safely skip it. Recovery will
798 * cleanup any inodes we might inadvertantly skip here. */
799 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
800 mlog(0, "Skipping delete of %lu because another node "
801 "has done this for us.\n", inode
->i_ino
);
807 spin_unlock(&oi
->ip_lock
);
812 /* Query the cluster to determine whether we should wipe an inode from
815 * Requires the inode to have the cluster lock. */
816 static int ocfs2_query_inode_wipe(struct inode
*inode
,
817 struct buffer_head
*di_bh
,
821 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
822 struct ocfs2_dinode
*di
;
826 /* While we were waiting for the cluster lock in
827 * ocfs2_delete_inode, another node might have asked to delete
828 * the inode. Recheck our flags to catch this. */
829 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
830 mlog(0, "Skipping delete of %llu because flags changed\n",
831 (unsigned long long)oi
->ip_blkno
);
835 /* Now that we have an up to date inode, we can double check
837 if (inode
->i_nlink
) {
838 mlog(0, "Skipping delete of %llu because nlink = %u\n",
839 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
843 /* Do some basic inode verification... */
844 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
845 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
846 /* for lack of a better error? */
849 "Inode %llu (on-disk %llu) not orphaned! "
850 "Disk flags 0x%x, inode flags 0x%x\n",
851 (unsigned long long)oi
->ip_blkno
,
852 (unsigned long long)le64_to_cpu(di
->i_blkno
),
853 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
857 /* has someone already deleted us?! baaad... */
865 * This is how ocfs2 determines whether an inode is still live
866 * within the cluster. Every node takes a shared read lock on
867 * the inode open lock in ocfs2_read_locked_inode(). When we
868 * get to ->delete_inode(), each node tries to convert it's
869 * lock to an exclusive. Trylocks are serialized by the inode
870 * meta data lock. If the upconvert suceeds, we know the inode
871 * is no longer live and can be deleted.
873 * Though we call this with the meta data lock held, the
874 * trylock keeps us from ABBA deadlock.
876 status
= ocfs2_try_open_lock(inode
, 1);
877 if (status
== -EAGAIN
) {
879 mlog(0, "Skipping delete of %llu because it is in use on "
880 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
889 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
890 (unsigned long long)oi
->ip_blkno
,
891 le16_to_cpu(di
->i_orphaned_slot
));
897 /* Support function for ocfs2_delete_inode. Will help us keep the
898 * inode data in a consistent state for clear_inode. Always truncates
899 * pages, optionally sync's them first. */
900 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
903 mlog(0, "Cleanup inode %llu, sync = %d\n",
904 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
906 write_inode_now(inode
, 1);
907 truncate_inode_pages(&inode
->i_data
, 0);
910 void ocfs2_delete_inode(struct inode
*inode
)
913 sigset_t blocked
, oldset
;
914 struct buffer_head
*di_bh
= NULL
;
916 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
918 if (is_bad_inode(inode
)) {
919 mlog(0, "Skipping delete of bad inode\n");
923 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
924 /* It's probably not necessary to truncate_inode_pages
925 * here but we do it for safety anyway (it will most
926 * likely be a no-op anyway) */
927 ocfs2_cleanup_delete_inode(inode
, 0);
931 /* We want to block signals in delete_inode as the lock and
932 * messaging paths may return us -ERESTARTSYS. Which would
933 * cause us to exit early, resulting in inodes being orphaned
935 sigfillset(&blocked
);
936 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
939 ocfs2_cleanup_delete_inode(inode
, 1);
943 /* Lock down the inode. This gives us an up to date view of
944 * it's metadata (for verification), and allows us to
945 * serialize delete_inode on multiple nodes.
947 * Even though we might be doing a truncate, we don't take the
948 * allocation lock here as it won't be needed - nobody will
949 * have the file open.
951 status
= ocfs2_inode_lock(inode
, &di_bh
, 1);
953 if (status
!= -ENOENT
)
955 ocfs2_cleanup_delete_inode(inode
, 0);
959 /* Query the cluster. This will be the final decision made
960 * before we go ahead and wipe the inode. */
961 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
962 if (!wipe
|| status
< 0) {
963 /* Error and remote inode busy both mean we won't be
964 * removing the inode, so they take almost the same
969 /* Someone in the cluster has disallowed a wipe of
970 * this inode, or it was never completely
971 * orphaned. Write out the pages and exit now. */
972 ocfs2_cleanup_delete_inode(inode
, 1);
973 goto bail_unlock_inode
;
976 ocfs2_cleanup_delete_inode(inode
, 0);
978 status
= ocfs2_wipe_inode(inode
, di_bh
);
980 if (status
!= -EDEADLK
)
982 goto bail_unlock_inode
;
986 * Mark the inode as successfully deleted.
988 * This is important for ocfs2_clear_inode() as it will check
989 * this flag and skip any checkpointing work
991 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
992 * the LVB for other nodes.
994 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
997 ocfs2_inode_unlock(inode
, 1);
1000 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
1008 void ocfs2_clear_inode(struct inode
*inode
)
1011 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1018 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1019 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1021 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1022 "Inode=%lu\n", inode
->i_ino
);
1024 /* To preven remote deletes we hold open lock before, now it
1025 * is time to unlock PR and EX open locks. */
1026 ocfs2_open_unlock(inode
);
1028 /* Do these before all the other work so that we don't bounce
1029 * the downconvert thread while waiting to destroy the locks. */
1030 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1031 ocfs2_mark_lockres_freeing(&oi
->ip_inode_lockres
);
1032 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
1034 /* We very well may get a clear_inode before all an inodes
1035 * metadata has hit disk. Of course, we can't drop any cluster
1036 * locks until the journal has finished with it. The only
1037 * exception here are successfully wiped inodes - their
1038 * metadata can now be considered to be part of the system
1039 * inodes from which it came. */
1040 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1041 ocfs2_checkpoint_inode(inode
);
1043 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1044 "Clear inode of %llu, inode has io markers\n",
1045 (unsigned long long)oi
->ip_blkno
);
1047 ocfs2_extent_map_trunc(inode
, 0);
1049 status
= ocfs2_drop_inode_locks(inode
);
1053 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1054 ocfs2_lock_res_free(&oi
->ip_inode_lockres
);
1055 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1057 ocfs2_metadata_cache_purge(inode
);
1059 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1060 "Clear inode of %llu, inode has %u cache items\n",
1061 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1063 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1064 "Clear inode of %llu, inode has a bad flag\n",
1065 (unsigned long long)oi
->ip_blkno
);
1067 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1068 "Clear inode of %llu, inode is locked\n",
1069 (unsigned long long)oi
->ip_blkno
);
1071 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1072 "Clear inode of %llu, io_mutex is locked\n",
1073 (unsigned long long)oi
->ip_blkno
);
1074 mutex_unlock(&oi
->ip_io_mutex
);
1077 * down_trylock() returns 0, down_write_trylock() returns 1
1080 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1081 "Clear inode of %llu, alloc_sem is locked\n",
1082 (unsigned long long)oi
->ip_blkno
);
1083 up_write(&oi
->ip_alloc_sem
);
1085 mlog_bug_on_msg(oi
->ip_open_count
,
1086 "Clear inode of %llu has open count %d\n",
1087 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1089 /* Clear all other flags. */
1090 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1091 oi
->ip_created_trans
= 0;
1092 oi
->ip_last_trans
= 0;
1093 oi
->ip_dir_start_lookup
= 0;
1094 oi
->ip_blkno
= 0ULL;
1097 * ip_jinode is used to track txns against this inode. We ensure that
1098 * the journal is flushed before journal shutdown. Thus it is safe to
1099 * have inodes get cleaned up after journal shutdown.
1101 jbd2_journal_release_jbd_inode(OCFS2_SB(inode
->i_sb
)->journal
->j_journal
,
1108 /* Called under inode_lock, with no more references on the
1109 * struct inode, so it's safe here to check the flags field
1110 * and to manipulate i_nlink without any other locks. */
1111 void ocfs2_drop_inode(struct inode
*inode
)
1113 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1117 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1118 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1120 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1121 generic_delete_inode(inode
);
1123 generic_drop_inode(inode
);
1129 * This is called from our getattr.
1131 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1133 struct inode
*inode
= dentry
->d_inode
;
1136 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1137 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1140 mlog(0, "eep, no inode!\n");
1145 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1146 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1147 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1148 mlog(0, "inode deleted!\n");
1152 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1154 /* Let ocfs2_inode_lock do the work of updating our struct
1156 status
= ocfs2_inode_lock(inode
, NULL
, 0);
1158 if (status
!= -ENOENT
)
1162 ocfs2_inode_unlock(inode
, 0);
1170 * Updates a disk inode from a
1172 * Only takes ip_lock.
1174 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1175 struct inode
*inode
,
1176 struct buffer_head
*bh
)
1179 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1181 mlog_entry("(inode %llu)\n",
1182 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1184 status
= ocfs2_journal_access(handle
, inode
, bh
,
1185 OCFS2_JOURNAL_ACCESS_WRITE
);
1191 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1192 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1193 ocfs2_get_inode_flags(OCFS2_I(inode
));
1194 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1195 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1196 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1198 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1199 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1200 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1201 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1202 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1203 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1204 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1205 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1206 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1207 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1208 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1210 status
= ocfs2_journal_dirty(handle
, bh
);
1223 * Updates a struct inode from a disk inode.
1224 * does no i/o, only takes ip_lock.
1226 void ocfs2_refresh_inode(struct inode
*inode
,
1227 struct ocfs2_dinode
*fe
)
1229 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1231 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1232 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1233 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1234 ocfs2_set_inode_flags(inode
);
1235 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1236 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1237 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1238 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1239 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1240 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1241 inode
->i_blocks
= 0;
1243 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1244 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1245 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1246 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1247 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1248 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1249 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1251 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1254 int ocfs2_validate_inode_block(struct super_block
*sb
,
1255 struct buffer_head
*bh
)
1258 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)bh
->b_data
;
1260 mlog(0, "Validating dinode %llu\n",
1261 (unsigned long long)bh
->b_blocknr
);
1263 BUG_ON(!buffer_uptodate(bh
));
1265 if (!OCFS2_IS_VALID_DINODE(di
)) {
1266 ocfs2_error(sb
, "Invalid dinode #%llu: signature = %.*s\n",
1267 (unsigned long long)bh
->b_blocknr
, 7,
1272 if (le64_to_cpu(di
->i_blkno
) != bh
->b_blocknr
) {
1273 ocfs2_error(sb
, "Invalid dinode #%llu: i_blkno is %llu\n",
1274 (unsigned long long)bh
->b_blocknr
,
1275 (unsigned long long)le64_to_cpu(di
->i_blkno
));
1279 if (!(di
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
1281 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1282 (unsigned long long)bh
->b_blocknr
);
1286 if (le32_to_cpu(di
->i_fs_generation
) !=
1287 OCFS2_SB(sb
)->fs_generation
) {
1289 "Invalid dinode #%llu: fs_generation is %u\n",
1290 (unsigned long long)bh
->b_blocknr
,
1291 le32_to_cpu(di
->i_fs_generation
));
1301 int ocfs2_read_inode_block_full(struct inode
*inode
, struct buffer_head
**bh
,
1305 struct buffer_head
*tmp
= *bh
;
1307 rc
= ocfs2_read_blocks(inode
, OCFS2_I(inode
)->ip_blkno
, 1, &tmp
,
1308 flags
, ocfs2_validate_inode_block
);
1310 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1317 int ocfs2_read_inode_block(struct inode
*inode
, struct buffer_head
**bh
)
1319 return ocfs2_read_inode_block_full(inode
, bh
, 0);