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 struct ocfs2_find_inode_args
61 unsigned int fi_flags
;
64 static int ocfs2_read_locked_inode(struct inode
*inode
,
65 struct ocfs2_find_inode_args
*args
);
66 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
67 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
68 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
70 struct buffer_head
*fe_bh
);
72 void ocfs2_set_inode_flags(struct inode
*inode
)
74 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
76 inode
->i_flags
&= ~(S_IMMUTABLE
|
77 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
79 if (flags
& OCFS2_IMMUTABLE_FL
)
80 inode
->i_flags
|= S_IMMUTABLE
;
82 if (flags
& OCFS2_SYNC_FL
)
83 inode
->i_flags
|= S_SYNC
;
84 if (flags
& OCFS2_APPEND_FL
)
85 inode
->i_flags
|= S_APPEND
;
86 if (flags
& OCFS2_NOATIME_FL
)
87 inode
->i_flags
|= S_NOATIME
;
88 if (flags
& OCFS2_DIRSYNC_FL
)
89 inode
->i_flags
|= S_DIRSYNC
;
92 struct inode
*ocfs2_ilookup_for_vote(struct ocfs2_super
*osb
,
96 struct ocfs2_find_inode_args args
;
98 /* ocfs2_ilookup_for_vote should *only* be called from the
100 BUG_ON(current
!= osb
->vote_task
);
102 args
.fi_blkno
= blkno
;
103 args
.fi_flags
= OCFS2_FI_FLAG_NOWAIT
;
105 args
.fi_flags
|= OCFS2_FI_FLAG_DELETE
;
106 args
.fi_ino
= ino_from_blkno(osb
->sb
, blkno
);
107 return ilookup5(osb
->sb
, args
.fi_ino
, ocfs2_find_actor
, &args
);
110 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, int flags
)
112 struct inode
*inode
= NULL
;
113 struct super_block
*sb
= osb
->sb
;
114 struct ocfs2_find_inode_args args
;
116 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
118 /* Ok. By now we've either got the offsets passed to us by the
119 * caller, or we just pulled them off the bh. Lets do some
120 * sanity checks to make sure they're OK. */
122 inode
= ERR_PTR(-EINVAL
);
123 mlog_errno(PTR_ERR(inode
));
127 args
.fi_blkno
= blkno
;
128 args
.fi_flags
= flags
;
129 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
131 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
132 ocfs2_init_locked_inode
, &args
);
133 /* inode was *not* in the inode cache. 2.6.x requires
134 * us to do our own read_inode call and unlock it
136 if (inode
&& inode
->i_state
& I_NEW
) {
137 mlog(0, "Inode was not in inode cache, reading it.\n");
138 ocfs2_read_locked_inode(inode
, &args
);
139 unlock_new_inode(inode
);
142 inode
= ERR_PTR(-ENOMEM
);
143 mlog_errno(PTR_ERR(inode
));
146 if (is_bad_inode(inode
)) {
148 inode
= ERR_PTR(-ESTALE
);
149 mlog_errno(PTR_ERR(inode
));
154 if (!IS_ERR(inode
)) {
155 mlog(0, "returning inode with number %llu\n",
156 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
157 mlog_exit_ptr(inode
);
159 mlog_errno(PTR_ERR(inode
));
166 * here's how inodes get read from disk:
167 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
168 * found? : return the in-memory inode
169 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
172 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
174 struct ocfs2_find_inode_args
*args
= NULL
;
175 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
178 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
182 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
184 if (oi
->ip_blkno
!= args
->fi_blkno
)
187 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
188 * ocfs2_ilookup_for_vote which won't create an inode for one
189 * that isn't found. The vote thread which doesn't want to get
190 * an inode which is in the process of going away - otherwise
191 * the call to __wait_on_freeing_inode in find_inode_fast will
192 * cause it to deadlock on an inode which may be waiting on a
193 * vote (or lock release) in delete_inode */
194 if ((args
->fi_flags
& OCFS2_FI_FLAG_NOWAIT
) &&
195 (inode
->i_state
& (I_FREEING
|I_CLEAR
))) {
196 /* As stated above, we're not going to return an
197 * inode. In the case of a delete vote, the voting
198 * code is going to signal the other node to go
199 * ahead. Mark that state here, so this freeing inode
200 * has the state when it gets to delete_inode. */
201 if (args
->fi_flags
& OCFS2_FI_FLAG_DELETE
) {
202 spin_lock(&oi
->ip_lock
);
203 ocfs2_mark_inode_remotely_deleted(inode
);
204 spin_unlock(&oi
->ip_lock
);
216 * initialize the new inode, but don't do anything that would cause
218 * return 0 on success, 1 on failure
220 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
222 struct ocfs2_find_inode_args
*args
= opaque
;
224 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
226 inode
->i_ino
= args
->fi_ino
;
227 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
233 int ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
236 struct super_block
*sb
;
237 struct ocfs2_super
*osb
;
238 int status
= -EINVAL
;
240 mlog_entry("(0x%p, size:%llu)\n", inode
,
241 (unsigned long long)fe
->i_size
);
246 /* this means that read_inode cannot create a superblock inode
247 * today. change if needed. */
248 if (!OCFS2_IS_VALID_DINODE(fe
) ||
249 !(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
250 mlog(ML_ERROR
, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
251 "signature = %.*s, flags = 0x%x\n",
253 (unsigned long long)le64_to_cpu(fe
->i_blkno
), 7,
254 fe
->i_signature
, le32_to_cpu(fe
->i_flags
));
258 if (le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
) {
259 mlog(ML_ERROR
, "file entry generation does not match "
260 "superblock! osb->fs_generation=%x, "
261 "fe->i_fs_generation=%x\n",
262 osb
->fs_generation
, le32_to_cpu(fe
->i_fs_generation
));
266 inode
->i_version
= 1;
267 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
268 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
269 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
270 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
271 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
273 /* Fast symlinks will have i_size but no allocated clusters. */
274 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
278 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe
->i_size
));
279 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
280 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
281 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
282 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
283 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
284 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
285 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
287 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
289 "ip_blkno %llu != i_blkno %llu!\n",
290 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
291 (unsigned long long)fe
->i_blkno
);
293 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
294 OCFS2_I(inode
)->ip_orphaned_slot
= OCFS2_INVALID_SLOT
;
295 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
297 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
299 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
300 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
302 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
303 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
304 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
305 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
306 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
307 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
308 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
309 /* we can't actually hit this as read_inode can't
310 * handle superblocks today ;-) */
314 switch (inode
->i_mode
& S_IFMT
) {
316 inode
->i_fop
= &ocfs2_fops
;
317 inode
->i_op
= &ocfs2_file_iops
;
318 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
321 inode
->i_op
= &ocfs2_dir_iops
;
322 inode
->i_fop
= &ocfs2_dops
;
323 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
326 if (ocfs2_inode_is_fast_symlink(inode
))
327 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
329 inode
->i_op
= &ocfs2_symlink_inode_operations
;
330 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
333 inode
->i_op
= &ocfs2_special_file_iops
;
334 init_special_inode(inode
, inode
->i_mode
,
340 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
341 le64_to_cpu(fe
->i_blkno
));
344 * If we ever want to create system files from kernel,
345 * the generation argument to
346 * ocfs2_inode_lock_res_init() will have to change.
348 BUG_ON(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
));
350 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
351 OCFS2_LOCK_TYPE_META
, 0, inode
);
354 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
355 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
358 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_data_lockres
,
359 OCFS2_LOCK_TYPE_DATA
, inode
->i_generation
,
362 ocfs2_set_inode_flags(inode
);
370 static int ocfs2_read_locked_inode(struct inode
*inode
,
371 struct ocfs2_find_inode_args
*args
)
373 struct super_block
*sb
;
374 struct ocfs2_super
*osb
;
375 struct ocfs2_dinode
*fe
;
376 struct buffer_head
*bh
= NULL
;
377 int status
, can_lock
;
380 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
383 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
384 mlog(ML_ERROR
, "bad inode\n");
391 mlog(ML_ERROR
, "bad inode args\n");
392 make_bad_inode(inode
);
397 * To improve performance of cold-cache inode stats, we take
398 * the cluster lock here if possible.
400 * Generally, OCFS2 never trusts the contents of an inode
401 * unless it's holding a cluster lock, so taking it here isn't
402 * a correctness issue as much as it is a performance
405 * There are three times when taking the lock is not a good idea:
407 * 1) During startup, before we have initialized the DLM.
409 * 2) If we are reading certain system files which never get
410 * cluster locks (local alloc, truncate log).
412 * 3) If the process doing the iget() is responsible for
413 * orphan dir recovery. We're holding the orphan dir lock and
414 * can get into a deadlock with another process on another
415 * node in ->delete_inode().
417 * #1 and #2 can be simply solved by never taking the lock
418 * here for system files (which are the only type we read
419 * during mount). It's a heavier approach, but our main
420 * concern is user-accesible files anyway.
422 * #3 works itself out because we'll eventually take the
423 * cluster lock before trusting anything anyway.
425 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
426 && !(args
->fi_flags
& OCFS2_FI_FLAG_NOLOCK
);
429 * To maintain backwards compatibility with older versions of
430 * ocfs2-tools, we still store the generation value for system
431 * files. The only ones that actually matter to userspace are
432 * the journals, but it's easier and inexpensive to just flag
433 * all system files similarly.
435 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
436 generation
= osb
->fs_generation
;
438 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
439 OCFS2_LOCK_TYPE_META
,
443 status
= ocfs2_meta_lock(inode
, NULL
, 0);
445 make_bad_inode(inode
);
451 status
= ocfs2_read_block(osb
, args
->fi_blkno
, &bh
, 0,
452 can_lock
? inode
: NULL
);
459 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
460 if (!OCFS2_IS_VALID_DINODE(fe
)) {
461 mlog(ML_ERROR
, "Invalid dinode #%llu: signature = %.*s\n",
462 (unsigned long long)fe
->i_blkno
, 7, fe
->i_signature
);
467 * This is a code bug. Right now the caller needs to
468 * understand whether it is asking for a system file inode or
469 * not so the proper lock names can be built.
471 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
472 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
473 "Inode %llu: system file state is ambigous\n",
474 (unsigned long long)args
->fi_blkno
);
476 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
477 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
478 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
480 if (ocfs2_populate_inode(inode
, fe
, 0) < 0) {
481 mlog(ML_ERROR
, "populate failed! i_blkno=%llu, i_ino=%lu\n",
482 (unsigned long long)fe
->i_blkno
, inode
->i_ino
);
486 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
492 ocfs2_meta_unlock(inode
, 0);
495 make_bad_inode(inode
);
504 void ocfs2_sync_blockdev(struct super_block
*sb
)
506 sync_blockdev(sb
->s_bdev
);
509 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
511 struct buffer_head
*fe_bh
)
514 handle_t
*handle
= NULL
;
515 struct ocfs2_truncate_context
*tc
= NULL
;
516 struct ocfs2_dinode
*fe
;
520 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
522 /* zero allocation, zero truncate :) */
526 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
527 if (IS_ERR(handle
)) {
528 status
= PTR_ERR(handle
);
534 status
= ocfs2_set_inode_size(handle
, inode
, fe_bh
, 0ULL);
540 ocfs2_commit_trans(osb
, handle
);
543 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
549 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
556 ocfs2_commit_trans(osb
, handle
);
562 static int ocfs2_remove_inode(struct inode
*inode
,
563 struct buffer_head
*di_bh
,
564 struct inode
*orphan_dir_inode
,
565 struct buffer_head
*orphan_dir_bh
)
568 struct inode
*inode_alloc_inode
= NULL
;
569 struct buffer_head
*inode_alloc_bh
= NULL
;
571 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
572 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
575 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
576 le16_to_cpu(di
->i_suballoc_slot
));
577 if (!inode_alloc_inode
) {
583 mutex_lock(&inode_alloc_inode
->i_mutex
);
584 status
= ocfs2_meta_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
586 mutex_unlock(&inode_alloc_inode
->i_mutex
);
592 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
593 if (IS_ERR(handle
)) {
594 status
= PTR_ERR(handle
);
599 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
606 /* set the inodes dtime */
607 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
608 OCFS2_JOURNAL_ACCESS_WRITE
);
614 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
615 le32_and_cpu(&di
->i_flags
, ~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
617 status
= ocfs2_journal_dirty(handle
, di_bh
);
623 ocfs2_remove_from_cache(inode
, di_bh
);
625 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
631 ocfs2_commit_trans(osb
, handle
);
633 ocfs2_meta_unlock(inode_alloc_inode
, 1);
634 mutex_unlock(&inode_alloc_inode
->i_mutex
);
635 brelse(inode_alloc_bh
);
637 iput(inode_alloc_inode
);
643 * Serialize with orphan dir recovery. If the process doing
644 * recovery on this orphan dir does an iget() with the dir
645 * i_mutex held, we'll deadlock here. Instead we detect this
646 * and exit early - recovery will wipe this inode for us.
648 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
653 spin_lock(&osb
->osb_lock
);
654 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
655 mlog(0, "Recovery is happening on orphan dir %d, will skip "
656 "this inode\n", slot
);
660 /* This signals to the orphan recovery process that it should
661 * wait for us to handle the wipe. */
662 osb
->osb_orphan_wipes
[slot
]++;
664 spin_unlock(&osb
->osb_lock
);
668 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
671 spin_lock(&osb
->osb_lock
);
672 osb
->osb_orphan_wipes
[slot
]--;
673 spin_unlock(&osb
->osb_lock
);
675 wake_up(&osb
->osb_wipe_event
);
678 static int ocfs2_wipe_inode(struct inode
*inode
,
679 struct buffer_head
*di_bh
)
681 int status
, orphaned_slot
;
682 struct inode
*orphan_dir_inode
= NULL
;
683 struct buffer_head
*orphan_dir_bh
= NULL
;
684 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
686 /* We've already voted on this so it should be readonly - no
687 * spinlock needed. */
688 orphaned_slot
= OCFS2_I(inode
)->ip_orphaned_slot
;
690 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
694 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
695 ORPHAN_DIR_SYSTEM_INODE
,
697 if (!orphan_dir_inode
) {
703 /* Lock the orphan dir. The lock will be held for the entire
704 * delete_inode operation. We do this now to avoid races with
705 * recovery completion on other nodes. */
706 mutex_lock(&orphan_dir_inode
->i_mutex
);
707 status
= ocfs2_meta_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
709 mutex_unlock(&orphan_dir_inode
->i_mutex
);
715 /* we do this while holding the orphan dir lock because we
716 * don't want recovery being run from another node to vote for
717 * an inode delete on us -- this will result in two nodes
718 * truncating the same file! */
719 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
722 goto bail_unlock_dir
;
725 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
731 ocfs2_meta_unlock(orphan_dir_inode
, 1);
732 mutex_unlock(&orphan_dir_inode
->i_mutex
);
733 brelse(orphan_dir_bh
);
735 iput(orphan_dir_inode
);
736 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
741 /* There is a series of simple checks that should be done before a
742 * vote is even considered. Encapsulate those in this function. */
743 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
746 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
747 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
749 /* We shouldn't be getting here for the root directory
751 if (inode
== osb
->root_inode
) {
752 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
756 /* If we're coming from process_vote we can't go into our own
757 * voting [hello, deadlock city!], so unforuntately we just
758 * have to skip deleting this guy. That's OK though because
759 * the node who's doing the actual deleting should handle it
761 if (current
== osb
->vote_task
) {
762 mlog(0, "Skipping delete of %lu because we're currently "
763 "in process_vote\n", inode
->i_ino
);
767 spin_lock(&oi
->ip_lock
);
768 /* OCFS2 *never* deletes system files. This should technically
769 * never get here as system file inodes should always have a
770 * positive link count. */
771 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
772 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
773 (unsigned long long)oi
->ip_blkno
);
777 /* If we have voted "yes" on the wipe of this inode for
778 * another node, it will be marked here so we can safely skip
779 * it. Recovery will cleanup any inodes we might inadvertantly
781 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
782 mlog(0, "Skipping delete of %lu because another node "
783 "has done this for us.\n", inode
->i_ino
);
789 spin_unlock(&oi
->ip_lock
);
794 /* Query the cluster to determine whether we should wipe an inode from
797 * Requires the inode to have the cluster lock. */
798 static int ocfs2_query_inode_wipe(struct inode
*inode
,
799 struct buffer_head
*di_bh
,
803 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
804 struct ocfs2_dinode
*di
;
808 /* While we were waiting for the cluster lock in
809 * ocfs2_delete_inode, another node might have asked to delete
810 * the inode. Recheck our flags to catch this. */
811 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
812 mlog(0, "Skipping delete of %llu because flags changed\n",
813 (unsigned long long)oi
->ip_blkno
);
817 /* Now that we have an up to date inode, we can double check
819 if (inode
->i_nlink
) {
820 mlog(0, "Skipping delete of %llu because nlink = %u\n",
821 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
825 /* Do some basic inode verification... */
826 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
827 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
828 /* for lack of a better error? */
831 "Inode %llu (on-disk %llu) not orphaned! "
832 "Disk flags 0x%x, inode flags 0x%x\n",
833 (unsigned long long)oi
->ip_blkno
,
834 (unsigned long long)di
->i_blkno
, di
->i_flags
,
839 /* has someone already deleted us?! baaad... */
846 status
= ocfs2_request_delete_vote(inode
);
847 /* -EBUSY means that other nodes are still using the
848 * inode. We're done here though, so avoid doing anything on
849 * disk and let them worry about deleting it. */
850 if (status
== -EBUSY
) {
852 mlog(0, "Skipping delete of %llu because it is in use on"
853 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
861 spin_lock(&oi
->ip_lock
);
862 if (oi
->ip_orphaned_slot
== OCFS2_INVALID_SLOT
) {
863 /* Nobody knew which slot this inode was orphaned
864 * into. This may happen during node death and
865 * recovery knows how to clean it up so we can safely
866 * ignore this inode for now on. */
867 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
868 (unsigned long long)oi
->ip_blkno
);
872 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
873 (unsigned long long)oi
->ip_blkno
, oi
->ip_orphaned_slot
);
875 spin_unlock(&oi
->ip_lock
);
881 /* Support function for ocfs2_delete_inode. Will help us keep the
882 * inode data in a consistent state for clear_inode. Always truncates
883 * pages, optionally sync's them first. */
884 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
887 mlog(0, "Cleanup inode %llu, sync = %d\n",
888 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
890 write_inode_now(inode
, 1);
891 truncate_inode_pages(&inode
->i_data
, 0);
894 void ocfs2_delete_inode(struct inode
*inode
)
897 sigset_t blocked
, oldset
;
898 struct buffer_head
*di_bh
= NULL
;
900 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
902 if (is_bad_inode(inode
)) {
903 mlog(0, "Skipping delete of bad inode\n");
907 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
908 /* It's probably not necessary to truncate_inode_pages
909 * here but we do it for safety anyway (it will most
910 * likely be a no-op anyway) */
911 ocfs2_cleanup_delete_inode(inode
, 0);
915 /* We want to block signals in delete_inode as the lock and
916 * messaging paths may return us -ERESTARTSYS. Which would
917 * cause us to exit early, resulting in inodes being orphaned
919 sigfillset(&blocked
);
920 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
923 ocfs2_cleanup_delete_inode(inode
, 1);
927 /* Lock down the inode. This gives us an up to date view of
928 * it's metadata (for verification), and allows us to
929 * serialize delete_inode votes.
931 * Even though we might be doing a truncate, we don't take the
932 * allocation lock here as it won't be needed - nobody will
933 * have the file open.
935 status
= ocfs2_meta_lock(inode
, &di_bh
, 1);
937 if (status
!= -ENOENT
)
939 ocfs2_cleanup_delete_inode(inode
, 0);
943 /* Query the cluster. This will be the final decision made
944 * before we go ahead and wipe the inode. */
945 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
946 if (!wipe
|| status
< 0) {
947 /* Error and inode busy vote both mean we won't be
948 * removing the inode, so they take almost the same
953 /* Someone in the cluster has voted to not wipe this
954 * inode, or it was never completely orphaned. Write
955 * out the pages and exit now. */
956 ocfs2_cleanup_delete_inode(inode
, 1);
957 goto bail_unlock_inode
;
960 ocfs2_cleanup_delete_inode(inode
, 0);
962 status
= ocfs2_wipe_inode(inode
, di_bh
);
964 if (status
!= -EDEADLK
)
966 goto bail_unlock_inode
;
970 * Mark the inode as successfully deleted.
972 * This is important for ocfs2_clear_inode() as it will check
973 * this flag and skip any checkpointing work
975 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
976 * the LVB for other nodes.
978 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
981 ocfs2_meta_unlock(inode
, 1);
984 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
992 void ocfs2_clear_inode(struct inode
*inode
)
995 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1002 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1003 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1005 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1006 "Inode=%lu\n", inode
->i_ino
);
1008 /* Do these before all the other work so that we don't bounce
1009 * the vote thread while waiting to destroy the locks. */
1010 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1011 ocfs2_mark_lockres_freeing(&oi
->ip_meta_lockres
);
1012 ocfs2_mark_lockres_freeing(&oi
->ip_data_lockres
);
1014 /* We very well may get a clear_inode before all an inodes
1015 * metadata has hit disk. Of course, we can't drop any cluster
1016 * locks until the journal has finished with it. The only
1017 * exception here are successfully wiped inodes - their
1018 * metadata can now be considered to be part of the system
1019 * inodes from which it came. */
1020 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1021 ocfs2_checkpoint_inode(inode
);
1023 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1024 "Clear inode of %llu, inode has io markers\n",
1025 (unsigned long long)oi
->ip_blkno
);
1027 ocfs2_extent_map_drop(inode
, 0);
1028 ocfs2_extent_map_init(inode
);
1030 status
= ocfs2_drop_inode_locks(inode
);
1034 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1035 ocfs2_lock_res_free(&oi
->ip_meta_lockres
);
1036 ocfs2_lock_res_free(&oi
->ip_data_lockres
);
1038 ocfs2_metadata_cache_purge(inode
);
1040 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1041 "Clear inode of %llu, inode has %u cache items\n",
1042 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1044 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1045 "Clear inode of %llu, inode has a bad flag\n",
1046 (unsigned long long)oi
->ip_blkno
);
1048 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1049 "Clear inode of %llu, inode is locked\n",
1050 (unsigned long long)oi
->ip_blkno
);
1052 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1053 "Clear inode of %llu, io_mutex is locked\n",
1054 (unsigned long long)oi
->ip_blkno
);
1055 mutex_unlock(&oi
->ip_io_mutex
);
1058 * down_trylock() returns 0, down_write_trylock() returns 1
1061 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1062 "Clear inode of %llu, alloc_sem is locked\n",
1063 (unsigned long long)oi
->ip_blkno
);
1064 up_write(&oi
->ip_alloc_sem
);
1066 mlog_bug_on_msg(oi
->ip_open_count
,
1067 "Clear inode of %llu has open count %d\n",
1068 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1070 /* Clear all other flags. */
1071 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1072 oi
->ip_created_trans
= 0;
1073 oi
->ip_last_trans
= 0;
1074 oi
->ip_dir_start_lookup
= 0;
1075 oi
->ip_blkno
= 0ULL;
1081 /* Called under inode_lock, with no more references on the
1082 * struct inode, so it's safe here to check the flags field
1083 * and to manipulate i_nlink without any other locks. */
1084 void ocfs2_drop_inode(struct inode
*inode
)
1086 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1090 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1091 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1093 /* Testing ip_orphaned_slot here wouldn't work because we may
1094 * not have gotten a delete_inode vote from any other nodes
1096 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1097 generic_delete_inode(inode
);
1099 generic_drop_inode(inode
);
1105 * TODO: this should probably be merged into ocfs2_get_block
1107 * However, you now need to pay attention to the cont_prepare_write()
1108 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1109 * expects never to extend).
1111 struct buffer_head
*ocfs2_bread(struct inode
*inode
,
1112 int block
, int *err
, int reada
)
1114 struct buffer_head
*bh
= NULL
;
1117 int readflags
= OCFS2_BH_CACHED
;
1120 readflags
|= OCFS2_BH_READAHEAD
;
1122 if (((u64
)block
<< inode
->i_sb
->s_blocksize_bits
) >=
1123 i_size_read(inode
)) {
1128 tmperr
= ocfs2_extent_map_get_blocks(inode
, block
, 1,
1135 tmperr
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), p_blkno
, &bh
,
1155 * This is called from our getattr.
1157 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1159 struct inode
*inode
= dentry
->d_inode
;
1162 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1163 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1166 mlog(0, "eep, no inode!\n");
1171 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1172 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1173 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1174 mlog(0, "inode deleted!\n");
1178 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1180 /* Let ocfs2_meta_lock do the work of updating our struct
1182 status
= ocfs2_meta_lock(inode
, NULL
, 0);
1184 if (status
!= -ENOENT
)
1188 ocfs2_meta_unlock(inode
, 0);
1196 * Updates a disk inode from a
1198 * Only takes ip_lock.
1200 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1201 struct inode
*inode
,
1202 struct buffer_head
*bh
)
1205 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1207 mlog_entry("(inode %llu)\n",
1208 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1210 status
= ocfs2_journal_access(handle
, inode
, bh
,
1211 OCFS2_JOURNAL_ACCESS_WRITE
);
1217 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1218 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1219 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1220 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1222 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1223 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1224 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1225 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1226 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1227 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1228 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1229 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1230 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1231 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1232 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1234 status
= ocfs2_journal_dirty(handle
, bh
);
1247 * Updates a struct inode from a disk inode.
1248 * does no i/o, only takes ip_lock.
1250 void ocfs2_refresh_inode(struct inode
*inode
,
1251 struct ocfs2_dinode
*fe
)
1253 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1255 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1256 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1257 ocfs2_set_inode_flags(inode
);
1258 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1259 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1260 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1261 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1262 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1263 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1264 inode
->i_blocks
= 0;
1266 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(i_size_read(inode
));
1267 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1268 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1269 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1270 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1271 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1272 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1274 spin_unlock(&OCFS2_I(inode
)->ip_lock
);