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/smp_lock.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 #define OCFS2_FI_FLAG_NOWAIT 0x1
58 #define OCFS2_FI_FLAG_DELETE 0x2
59 struct ocfs2_find_inode_args
63 unsigned int fi_flags
;
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 struct inode
*ocfs2_ilookup_for_vote(struct ocfs2_super
*osb
,
98 struct ocfs2_find_inode_args args
;
100 /* ocfs2_ilookup_for_vote should *only* be called from the
102 BUG_ON(current
!= osb
->vote_task
);
104 args
.fi_blkno
= blkno
;
105 args
.fi_flags
= OCFS2_FI_FLAG_NOWAIT
;
107 args
.fi_flags
|= OCFS2_FI_FLAG_DELETE
;
108 args
.fi_ino
= ino_from_blkno(osb
->sb
, blkno
);
109 return ilookup5(osb
->sb
, args
.fi_ino
, ocfs2_find_actor
, &args
);
112 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
)
114 struct inode
*inode
= NULL
;
115 struct super_block
*sb
= osb
->sb
;
116 struct ocfs2_find_inode_args args
;
118 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
120 /* Ok. By now we've either got the offsets passed to us by the
121 * caller, or we just pulled them off the bh. Lets do some
122 * sanity checks to make sure they're OK. */
124 inode
= ERR_PTR(-EINVAL
);
125 mlog_errno(PTR_ERR(inode
));
129 args
.fi_blkno
= blkno
;
131 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
133 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
134 ocfs2_init_locked_inode
, &args
);
135 /* inode was *not* in the inode cache. 2.6.x requires
136 * us to do our own read_inode call and unlock it
138 if (inode
&& inode
->i_state
& I_NEW
) {
139 mlog(0, "Inode was not in inode cache, reading it.\n");
140 ocfs2_read_locked_inode(inode
, &args
);
141 unlock_new_inode(inode
);
144 inode
= ERR_PTR(-ENOMEM
);
145 mlog_errno(PTR_ERR(inode
));
148 if (is_bad_inode(inode
)) {
150 inode
= ERR_PTR(-ESTALE
);
151 mlog_errno(PTR_ERR(inode
));
156 if (!IS_ERR(inode
)) {
157 mlog(0, "returning inode with number %llu\n",
158 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
159 mlog_exit_ptr(inode
);
161 mlog_errno(PTR_ERR(inode
));
168 * here's how inodes get read from disk:
169 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
170 * found? : return the in-memory inode
171 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
174 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
176 struct ocfs2_find_inode_args
*args
= NULL
;
177 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
180 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
184 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
186 if (oi
->ip_blkno
!= args
->fi_blkno
)
189 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
190 * ocfs2_ilookup_for_vote which won't create an inode for one
191 * that isn't found. The vote thread which doesn't want to get
192 * an inode which is in the process of going away - otherwise
193 * the call to __wait_on_freeing_inode in find_inode_fast will
194 * cause it to deadlock on an inode which may be waiting on a
195 * vote (or lock release) in delete_inode */
196 if ((args
->fi_flags
& OCFS2_FI_FLAG_NOWAIT
) &&
197 (inode
->i_state
& (I_FREEING
|I_CLEAR
))) {
198 /* As stated above, we're not going to return an
199 * inode. In the case of a delete vote, the voting
200 * code is going to signal the other node to go
201 * ahead. Mark that state here, so this freeing inode
202 * has the state when it gets to delete_inode. */
203 if (args
->fi_flags
& OCFS2_FI_FLAG_DELETE
) {
204 spin_lock(&oi
->ip_lock
);
205 ocfs2_mark_inode_remotely_deleted(inode
);
206 spin_unlock(&oi
->ip_lock
);
218 * initialize the new inode, but don't do anything that would cause
220 * return 0 on success, 1 on failure
222 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
224 struct ocfs2_find_inode_args
*args
= opaque
;
226 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
228 inode
->i_ino
= args
->fi_ino
;
229 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
235 int ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
238 struct super_block
*sb
;
239 struct ocfs2_super
*osb
;
240 int status
= -EINVAL
;
242 mlog_entry("(0x%p, size:%llu)\n", inode
,
243 (unsigned long long)fe
->i_size
);
248 /* this means that read_inode cannot create a superblock inode
249 * today. change if needed. */
250 if (!OCFS2_IS_VALID_DINODE(fe
) ||
251 !(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
252 mlog(ML_ERROR
, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
253 "signature = %.*s, flags = 0x%x\n",
255 (unsigned long long)le64_to_cpu(fe
->i_blkno
), 7,
256 fe
->i_signature
, le32_to_cpu(fe
->i_flags
));
260 if (le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
) {
261 mlog(ML_ERROR
, "file entry generation does not match "
262 "superblock! osb->fs_generation=%x, "
263 "fe->i_fs_generation=%x\n",
264 osb
->fs_generation
, le32_to_cpu(fe
->i_fs_generation
));
268 inode
->i_version
= 1;
269 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
270 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
271 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
272 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
273 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
274 inode
->i_blksize
= (u32
)osb
->s_clustersize
;
276 /* Fast symlinks will have i_size but no allocated clusters. */
277 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
281 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe
->i_size
));
282 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
283 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
284 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
285 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
286 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
287 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
288 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
290 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
292 "ip_blkno %llu != i_blkno %llu!\n",
293 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
294 (unsigned long long)fe
->i_blkno
);
296 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
297 OCFS2_I(inode
)->ip_orphaned_slot
= OCFS2_INVALID_SLOT
;
298 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
301 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
302 le64_to_cpu(fe
->i_blkno
));
304 mlog(0, "blkno = %llu, ino = %lu, create_ino = %s\n",
305 (unsigned long long)fe
->i_blkno
, inode
->i_ino
, create_ino
? "true" : "false");
307 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
309 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
310 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
311 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
312 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
313 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
314 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
315 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
316 /* we can't actually hit this as read_inode can't
317 * handle superblocks today ;-) */
321 switch (inode
->i_mode
& S_IFMT
) {
323 inode
->i_fop
= &ocfs2_fops
;
324 inode
->i_op
= &ocfs2_file_iops
;
325 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
328 inode
->i_op
= &ocfs2_dir_iops
;
329 inode
->i_fop
= &ocfs2_dops
;
330 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
333 if (ocfs2_inode_is_fast_symlink(inode
))
334 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
336 inode
->i_op
= &ocfs2_symlink_inode_operations
;
337 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
340 inode
->i_op
= &ocfs2_special_file_iops
;
341 init_special_inode(inode
, inode
->i_mode
,
346 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
347 OCFS2_LOCK_TYPE_RW
, inode
);
348 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
349 OCFS2_LOCK_TYPE_META
, inode
);
350 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_data_lockres
,
351 OCFS2_LOCK_TYPE_DATA
, inode
);
353 ocfs2_set_inode_flags(inode
);
354 inode
->i_flags
|= S_NOATIME
;
362 static int ocfs2_read_locked_inode(struct inode
*inode
,
363 struct ocfs2_find_inode_args
*args
)
365 struct super_block
*sb
;
366 struct ocfs2_super
*osb
;
367 struct ocfs2_dinode
*fe
;
368 struct buffer_head
*bh
= NULL
;
372 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
375 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
376 mlog(ML_ERROR
, "bad inode\n");
383 mlog(ML_ERROR
, "bad inode args\n");
384 make_bad_inode(inode
);
388 /* Read the FE off disk. This is safe because the kernel only
389 * does one read_inode2 for a new inode, and if it doesn't
390 * exist yet then nobody can be working on it! */
391 status
= ocfs2_read_block(osb
, args
->fi_blkno
, &bh
, 0, NULL
);
394 make_bad_inode(inode
);
398 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
399 if (!OCFS2_IS_VALID_DINODE(fe
)) {
400 mlog(ML_ERROR
, "Invalid dinode #%llu: signature = %.*s\n",
401 (unsigned long long)fe
->i_blkno
, 7, fe
->i_signature
);
402 make_bad_inode(inode
);
406 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
409 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
410 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
411 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
414 if (ocfs2_populate_inode(inode
, fe
, 0) < 0) {
415 mlog(ML_ERROR
, "populate failed! i_blkno=%llu, i_ino=%lu\n",
416 (unsigned long long)fe
->i_blkno
, inode
->i_ino
);
417 make_bad_inode(inode
);
421 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
424 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
436 void ocfs2_sync_blockdev(struct super_block
*sb
)
438 sync_blockdev(sb
->s_bdev
);
441 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
443 struct buffer_head
*fe_bh
)
446 struct ocfs2_journal_handle
*handle
= NULL
;
447 struct ocfs2_truncate_context
*tc
= NULL
;
448 struct ocfs2_dinode
*fe
;
452 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
454 /* zero allocation, zero truncate :) */
458 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_INODE_UPDATE_CREDITS
);
459 if (IS_ERR(handle
)) {
460 status
= PTR_ERR(handle
);
466 status
= ocfs2_set_inode_size(handle
, inode
, fe_bh
, 0ULL);
472 ocfs2_commit_trans(handle
);
475 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
481 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
488 ocfs2_commit_trans(handle
);
494 static int ocfs2_remove_inode(struct inode
*inode
,
495 struct buffer_head
*di_bh
,
496 struct inode
*orphan_dir_inode
,
497 struct buffer_head
*orphan_dir_bh
)
500 struct inode
*inode_alloc_inode
= NULL
;
501 struct buffer_head
*inode_alloc_bh
= NULL
;
502 struct ocfs2_journal_handle
*handle
;
503 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
504 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
507 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
508 le16_to_cpu(di
->i_suballoc_slot
));
509 if (!inode_alloc_inode
) {
515 mutex_lock(&inode_alloc_inode
->i_mutex
);
516 status
= ocfs2_meta_lock(inode_alloc_inode
, NULL
, &inode_alloc_bh
, 1);
518 mutex_unlock(&inode_alloc_inode
->i_mutex
);
524 handle
= ocfs2_start_trans(osb
, NULL
, OCFS2_DELETE_INODE_CREDITS
);
525 if (IS_ERR(handle
)) {
526 status
= PTR_ERR(handle
);
531 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
538 /* set the inodes dtime */
539 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
540 OCFS2_JOURNAL_ACCESS_WRITE
);
546 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
547 le32_and_cpu(&di
->i_flags
, ~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
549 status
= ocfs2_journal_dirty(handle
, di_bh
);
555 ocfs2_remove_from_cache(inode
, di_bh
);
557 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
563 ocfs2_commit_trans(handle
);
565 ocfs2_meta_unlock(inode_alloc_inode
, 1);
566 mutex_unlock(&inode_alloc_inode
->i_mutex
);
567 brelse(inode_alloc_bh
);
569 iput(inode_alloc_inode
);
575 * Serialize with orphan dir recovery. If the process doing
576 * recovery on this orphan dir does an iget() with the dir
577 * i_mutex held, we'll deadlock here. Instead we detect this
578 * and exit early - recovery will wipe this inode for us.
580 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
585 spin_lock(&osb
->osb_lock
);
586 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
587 mlog(0, "Recovery is happening on orphan dir %d, will skip "
588 "this inode\n", slot
);
592 /* This signals to the orphan recovery process that it should
593 * wait for us to handle the wipe. */
594 osb
->osb_orphan_wipes
[slot
]++;
596 spin_unlock(&osb
->osb_lock
);
600 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
603 spin_lock(&osb
->osb_lock
);
604 osb
->osb_orphan_wipes
[slot
]--;
605 spin_unlock(&osb
->osb_lock
);
607 wake_up(&osb
->osb_wipe_event
);
610 static int ocfs2_wipe_inode(struct inode
*inode
,
611 struct buffer_head
*di_bh
)
613 int status
, orphaned_slot
;
614 struct inode
*orphan_dir_inode
= NULL
;
615 struct buffer_head
*orphan_dir_bh
= NULL
;
616 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
618 /* We've already voted on this so it should be readonly - no
619 * spinlock needed. */
620 orphaned_slot
= OCFS2_I(inode
)->ip_orphaned_slot
;
622 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
626 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
627 ORPHAN_DIR_SYSTEM_INODE
,
629 if (!orphan_dir_inode
) {
635 /* Lock the orphan dir. The lock will be held for the entire
636 * delete_inode operation. We do this now to avoid races with
637 * recovery completion on other nodes. */
638 mutex_lock(&orphan_dir_inode
->i_mutex
);
639 status
= ocfs2_meta_lock(orphan_dir_inode
, NULL
, &orphan_dir_bh
, 1);
641 mutex_unlock(&orphan_dir_inode
->i_mutex
);
647 /* we do this while holding the orphan dir lock because we
648 * don't want recovery being run from another node to vote for
649 * an inode delete on us -- this will result in two nodes
650 * truncating the same file! */
651 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
654 goto bail_unlock_dir
;
657 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
663 ocfs2_meta_unlock(orphan_dir_inode
, 1);
664 mutex_unlock(&orphan_dir_inode
->i_mutex
);
665 brelse(orphan_dir_bh
);
667 iput(orphan_dir_inode
);
668 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
673 /* There is a series of simple checks that should be done before a
674 * vote is even considered. Encapsulate those in this function. */
675 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
678 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
679 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
681 /* We shouldn't be getting here for the root directory
683 if (inode
== osb
->root_inode
) {
684 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
688 /* If we're coming from process_vote we can't go into our own
689 * voting [hello, deadlock city!], so unforuntately we just
690 * have to skip deleting this guy. That's OK though because
691 * the node who's doing the actual deleting should handle it
693 if (current
== osb
->vote_task
) {
694 mlog(0, "Skipping delete of %lu because we're currently "
695 "in process_vote\n", inode
->i_ino
);
699 spin_lock(&oi
->ip_lock
);
700 /* OCFS2 *never* deletes system files. This should technically
701 * never get here as system file inodes should always have a
702 * positive link count. */
703 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
704 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
705 (unsigned long long)oi
->ip_blkno
);
709 /* If we have voted "yes" on the wipe of this inode for
710 * another node, it will be marked here so we can safely skip
711 * it. Recovery will cleanup any inodes we might inadvertantly
713 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
714 mlog(0, "Skipping delete of %lu because another node "
715 "has done this for us.\n", inode
->i_ino
);
721 spin_unlock(&oi
->ip_lock
);
726 /* Query the cluster to determine whether we should wipe an inode from
729 * Requires the inode to have the cluster lock. */
730 static int ocfs2_query_inode_wipe(struct inode
*inode
,
731 struct buffer_head
*di_bh
,
735 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
736 struct ocfs2_dinode
*di
;
740 /* While we were waiting for the cluster lock in
741 * ocfs2_delete_inode, another node might have asked to delete
742 * the inode. Recheck our flags to catch this. */
743 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
744 mlog(0, "Skipping delete of %llu because flags changed\n",
745 (unsigned long long)oi
->ip_blkno
);
749 /* Now that we have an up to date inode, we can double check
751 if (inode
->i_nlink
) {
752 mlog(0, "Skipping delete of %llu because nlink = %u\n",
753 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
757 /* Do some basic inode verification... */
758 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
759 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
760 /* for lack of a better error? */
763 "Inode %llu (on-disk %llu) not orphaned! "
764 "Disk flags 0x%x, inode flags 0x%x\n",
765 (unsigned long long)oi
->ip_blkno
,
766 (unsigned long long)di
->i_blkno
, di
->i_flags
,
771 /* has someone already deleted us?! baaad... */
778 status
= ocfs2_request_delete_vote(inode
);
779 /* -EBUSY means that other nodes are still using the
780 * inode. We're done here though, so avoid doing anything on
781 * disk and let them worry about deleting it. */
782 if (status
== -EBUSY
) {
784 mlog(0, "Skipping delete of %llu because it is in use on"
785 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
793 spin_lock(&oi
->ip_lock
);
794 if (oi
->ip_orphaned_slot
== OCFS2_INVALID_SLOT
) {
795 /* Nobody knew which slot this inode was orphaned
796 * into. This may happen during node death and
797 * recovery knows how to clean it up so we can safely
798 * ignore this inode for now on. */
799 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
800 (unsigned long long)oi
->ip_blkno
);
804 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
805 (unsigned long long)oi
->ip_blkno
, oi
->ip_orphaned_slot
);
807 spin_unlock(&oi
->ip_lock
);
813 /* Support function for ocfs2_delete_inode. Will help us keep the
814 * inode data in a consistent state for clear_inode. Always truncates
815 * pages, optionally sync's them first. */
816 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
819 mlog(0, "Cleanup inode %llu, sync = %d\n",
820 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
822 write_inode_now(inode
, 1);
823 truncate_inode_pages(&inode
->i_data
, 0);
826 void ocfs2_delete_inode(struct inode
*inode
)
829 sigset_t blocked
, oldset
;
830 struct buffer_head
*di_bh
= NULL
;
832 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
834 if (is_bad_inode(inode
)) {
835 mlog(0, "Skipping delete of bad inode\n");
839 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
840 /* It's probably not necessary to truncate_inode_pages
841 * here but we do it for safety anyway (it will most
842 * likely be a no-op anyway) */
843 ocfs2_cleanup_delete_inode(inode
, 0);
847 /* We want to block signals in delete_inode as the lock and
848 * messaging paths may return us -ERESTARTSYS. Which would
849 * cause us to exit early, resulting in inodes being orphaned
851 sigfillset(&blocked
);
852 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
855 ocfs2_cleanup_delete_inode(inode
, 1);
859 /* Lock down the inode. This gives us an up to date view of
860 * it's metadata (for verification), and allows us to
861 * serialize delete_inode votes.
863 * Even though we might be doing a truncate, we don't take the
864 * allocation lock here as it won't be needed - nobody will
865 * have the file open.
867 status
= ocfs2_meta_lock(inode
, NULL
, &di_bh
, 1);
869 if (status
!= -ENOENT
)
871 ocfs2_cleanup_delete_inode(inode
, 0);
875 /* Query the cluster. This will be the final decision made
876 * before we go ahead and wipe the inode. */
877 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
878 if (!wipe
|| status
< 0) {
879 /* Error and inode busy vote both mean we won't be
880 * removing the inode, so they take almost the same
885 /* Someone in the cluster has voted to not wipe this
886 * inode, or it was never completely orphaned. Write
887 * out the pages and exit now. */
888 ocfs2_cleanup_delete_inode(inode
, 1);
889 goto bail_unlock_inode
;
892 ocfs2_cleanup_delete_inode(inode
, 0);
894 status
= ocfs2_wipe_inode(inode
, di_bh
);
896 if (status
!= -EDEADLK
)
898 goto bail_unlock_inode
;
901 /* Mark the inode as successfully deleted. This is important
902 * for ocfs2_clear_inode as it will check this flag and skip
903 * any checkpointing work */
904 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
907 ocfs2_meta_unlock(inode
, 1);
910 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
918 void ocfs2_clear_inode(struct inode
*inode
)
921 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
928 mlog(0, "Clearing inode: %llu, nlink = %u\n",
929 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
931 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
932 "Inode=%lu\n", inode
->i_ino
);
934 /* Do these before all the other work so that we don't bounce
935 * the vote thread while waiting to destroy the locks. */
936 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
937 ocfs2_mark_lockres_freeing(&oi
->ip_meta_lockres
);
938 ocfs2_mark_lockres_freeing(&oi
->ip_data_lockres
);
940 /* We very well may get a clear_inode before all an inodes
941 * metadata has hit disk. Of course, we can't drop any cluster
942 * locks until the journal has finished with it. The only
943 * exception here are successfully wiped inodes - their
944 * metadata can now be considered to be part of the system
945 * inodes from which it came. */
946 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
947 ocfs2_checkpoint_inode(inode
);
949 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
950 "Clear inode of %llu, inode has io markers\n",
951 (unsigned long long)oi
->ip_blkno
);
953 ocfs2_extent_map_drop(inode
, 0);
954 ocfs2_extent_map_init(inode
);
956 status
= ocfs2_drop_inode_locks(inode
);
960 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
961 ocfs2_lock_res_free(&oi
->ip_meta_lockres
);
962 ocfs2_lock_res_free(&oi
->ip_data_lockres
);
964 ocfs2_metadata_cache_purge(inode
);
966 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
967 "Clear inode of %llu, inode has %u cache items\n",
968 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
970 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
971 "Clear inode of %llu, inode has a bad flag\n",
972 (unsigned long long)oi
->ip_blkno
);
974 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
975 "Clear inode of %llu, inode is locked\n",
976 (unsigned long long)oi
->ip_blkno
);
978 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
979 "Clear inode of %llu, io_mutex is locked\n",
980 (unsigned long long)oi
->ip_blkno
);
981 mutex_unlock(&oi
->ip_io_mutex
);
984 * down_trylock() returns 0, down_write_trylock() returns 1
987 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
988 "Clear inode of %llu, alloc_sem is locked\n",
989 (unsigned long long)oi
->ip_blkno
);
990 up_write(&oi
->ip_alloc_sem
);
992 mlog_bug_on_msg(oi
->ip_open_count
,
993 "Clear inode of %llu has open count %d\n",
994 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
995 mlog_bug_on_msg(!list_empty(&oi
->ip_handle_list
),
996 "Clear inode of %llu has non empty handle list\n",
997 (unsigned long long)oi
->ip_blkno
);
998 mlog_bug_on_msg(oi
->ip_handle
,
999 "Clear inode of %llu has non empty handle pointer\n",
1000 (unsigned long long)oi
->ip_blkno
);
1002 /* Clear all other flags. */
1003 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1004 oi
->ip_created_trans
= 0;
1005 oi
->ip_last_trans
= 0;
1006 oi
->ip_dir_start_lookup
= 0;
1007 oi
->ip_blkno
= 0ULL;
1013 /* Called under inode_lock, with no more references on the
1014 * struct inode, so it's safe here to check the flags field
1015 * and to manipulate i_nlink without any other locks. */
1016 void ocfs2_drop_inode(struct inode
*inode
)
1018 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1022 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1023 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1025 /* Testing ip_orphaned_slot here wouldn't work because we may
1026 * not have gotten a delete_inode vote from any other nodes
1028 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
) {
1029 mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
1033 generic_drop_inode(inode
);
1039 * TODO: this should probably be merged into ocfs2_get_block
1041 * However, you now need to pay attention to the cont_prepare_write()
1042 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1043 * expects never to extend).
1045 struct buffer_head
*ocfs2_bread(struct inode
*inode
,
1046 int block
, int *err
, int reada
)
1048 struct buffer_head
*bh
= NULL
;
1051 int readflags
= OCFS2_BH_CACHED
;
1054 readflags
|= OCFS2_BH_READAHEAD
;
1056 if (((u64
)block
<< inode
->i_sb
->s_blocksize_bits
) >=
1057 i_size_read(inode
)) {
1062 tmperr
= ocfs2_extent_map_get_blocks(inode
, block
, 1,
1069 tmperr
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), p_blkno
, &bh
,
1089 * This is called from our getattr.
1091 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1093 struct inode
*inode
= dentry
->d_inode
;
1096 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1097 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1100 mlog(0, "eep, no inode!\n");
1105 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1106 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1107 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1108 mlog(0, "inode deleted!\n");
1112 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1114 /* Let ocfs2_meta_lock do the work of updating our struct
1116 status
= ocfs2_meta_lock(inode
, NULL
, NULL
, 0);
1118 if (status
!= -ENOENT
)
1122 ocfs2_meta_unlock(inode
, 0);
1130 * Updates a disk inode from a
1132 * Only takes ip_lock.
1134 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle
*handle
,
1135 struct inode
*inode
,
1136 struct buffer_head
*bh
)
1139 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1141 mlog_entry("(inode %llu)\n",
1142 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1144 status
= ocfs2_journal_access(handle
, inode
, bh
,
1145 OCFS2_JOURNAL_ACCESS_WRITE
);
1151 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1152 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1153 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1154 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1156 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1157 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1158 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1159 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1160 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1161 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1162 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1163 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1164 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1165 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1166 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1168 status
= ocfs2_journal_dirty(handle
, bh
);
1181 * Updates a struct inode from a disk inode.
1182 * does no i/o, only takes ip_lock.
1184 void ocfs2_refresh_inode(struct inode
*inode
,
1185 struct ocfs2_dinode
*fe
)
1187 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1189 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1191 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1192 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1193 ocfs2_set_inode_flags(inode
);
1194 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1195 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1196 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1197 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1198 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1199 inode
->i_blksize
= (u32
) osb
->s_clustersize
;
1200 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1201 inode
->i_blocks
= 0;
1203 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(i_size_read(inode
));
1204 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1205 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1206 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1207 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1208 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1209 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1211 spin_unlock(&OCFS2_I(inode
)->ip_lock
);