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_iget(struct ocfs2_super
*osb
, u64 blkno
, int flags
)
94 struct inode
*inode
= NULL
;
95 struct super_block
*sb
= osb
->sb
;
96 struct ocfs2_find_inode_args args
;
98 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
100 /* Ok. By now we've either got the offsets passed to us by the
101 * caller, or we just pulled them off the bh. Lets do some
102 * sanity checks to make sure they're OK. */
104 inode
= ERR_PTR(-EINVAL
);
105 mlog_errno(PTR_ERR(inode
));
109 args
.fi_blkno
= blkno
;
110 args
.fi_flags
= flags
;
111 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
113 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
114 ocfs2_init_locked_inode
, &args
);
115 /* inode was *not* in the inode cache. 2.6.x requires
116 * us to do our own read_inode call and unlock it
118 if (inode
&& inode
->i_state
& I_NEW
) {
119 mlog(0, "Inode was not in inode cache, reading it.\n");
120 ocfs2_read_locked_inode(inode
, &args
);
121 unlock_new_inode(inode
);
124 inode
= ERR_PTR(-ENOMEM
);
125 mlog_errno(PTR_ERR(inode
));
128 if (is_bad_inode(inode
)) {
130 inode
= ERR_PTR(-ESTALE
);
135 if (!IS_ERR(inode
)) {
136 mlog(0, "returning inode with number %llu\n",
137 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
138 mlog_exit_ptr(inode
);
146 * here's how inodes get read from disk:
147 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
148 * found? : return the in-memory inode
149 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
152 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
154 struct ocfs2_find_inode_args
*args
= NULL
;
155 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
158 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
162 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
164 if (oi
->ip_blkno
!= args
->fi_blkno
)
174 * initialize the new inode, but don't do anything that would cause
176 * return 0 on success, 1 on failure
178 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
180 struct ocfs2_find_inode_args
*args
= opaque
;
182 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
184 inode
->i_ino
= args
->fi_ino
;
185 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
191 int ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
194 struct super_block
*sb
;
195 struct ocfs2_super
*osb
;
196 int status
= -EINVAL
;
198 mlog_entry("(0x%p, size:%llu)\n", inode
,
199 (unsigned long long)fe
->i_size
);
204 /* this means that read_inode cannot create a superblock inode
205 * today. change if needed. */
206 if (!OCFS2_IS_VALID_DINODE(fe
) ||
207 !(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
208 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
209 "signature = %.*s, flags = 0x%x\n",
211 (unsigned long long)le64_to_cpu(fe
->i_blkno
), 7,
212 fe
->i_signature
, le32_to_cpu(fe
->i_flags
));
216 if (le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
) {
217 mlog(ML_ERROR
, "file entry generation does not match "
218 "superblock! osb->fs_generation=%x, "
219 "fe->i_fs_generation=%x\n",
220 osb
->fs_generation
, le32_to_cpu(fe
->i_fs_generation
));
224 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
225 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
227 inode
->i_version
= 1;
228 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
229 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
230 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
231 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
232 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
234 /* Fast symlinks will have i_size but no allocated clusters. */
235 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
238 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
239 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
240 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
241 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
242 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
243 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
244 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
245 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
247 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
249 "ip_blkno %llu != i_blkno %llu!\n",
250 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
251 (unsigned long long)fe
->i_blkno
);
253 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
255 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
256 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
258 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
259 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
260 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
261 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
262 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
263 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
264 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
265 /* we can't actually hit this as read_inode can't
266 * handle superblocks today ;-) */
270 switch (inode
->i_mode
& S_IFMT
) {
272 inode
->i_fop
= &ocfs2_fops
;
273 inode
->i_op
= &ocfs2_file_iops
;
274 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
277 inode
->i_op
= &ocfs2_dir_iops
;
278 inode
->i_fop
= &ocfs2_dops
;
279 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
282 if (ocfs2_inode_is_fast_symlink(inode
))
283 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
285 inode
->i_op
= &ocfs2_symlink_inode_operations
;
286 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
289 inode
->i_op
= &ocfs2_special_file_iops
;
290 init_special_inode(inode
, inode
->i_mode
,
296 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
297 le64_to_cpu(fe
->i_blkno
));
300 * If we ever want to create system files from kernel,
301 * the generation argument to
302 * ocfs2_inode_lock_res_init() will have to change.
304 BUG_ON(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
));
306 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
307 OCFS2_LOCK_TYPE_META
, 0, inode
);
309 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
310 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
313 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
314 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
317 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_data_lockres
,
318 OCFS2_LOCK_TYPE_DATA
, inode
->i_generation
,
321 ocfs2_set_inode_flags(inode
);
329 static int ocfs2_read_locked_inode(struct inode
*inode
,
330 struct ocfs2_find_inode_args
*args
)
332 struct super_block
*sb
;
333 struct ocfs2_super
*osb
;
334 struct ocfs2_dinode
*fe
;
335 struct buffer_head
*bh
= NULL
;
336 int status
, can_lock
;
339 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
342 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
343 mlog(ML_ERROR
, "bad inode\n");
350 mlog(ML_ERROR
, "bad inode args\n");
351 make_bad_inode(inode
);
356 * To improve performance of cold-cache inode stats, we take
357 * the cluster lock here if possible.
359 * Generally, OCFS2 never trusts the contents of an inode
360 * unless it's holding a cluster lock, so taking it here isn't
361 * a correctness issue as much as it is a performance
364 * There are three times when taking the lock is not a good idea:
366 * 1) During startup, before we have initialized the DLM.
368 * 2) If we are reading certain system files which never get
369 * cluster locks (local alloc, truncate log).
371 * 3) If the process doing the iget() is responsible for
372 * orphan dir recovery. We're holding the orphan dir lock and
373 * can get into a deadlock with another process on another
374 * node in ->delete_inode().
376 * #1 and #2 can be simply solved by never taking the lock
377 * here for system files (which are the only type we read
378 * during mount). It's a heavier approach, but our main
379 * concern is user-accesible files anyway.
381 * #3 works itself out because we'll eventually take the
382 * cluster lock before trusting anything anyway.
384 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
385 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
386 && !ocfs2_mount_local(osb
);
389 * To maintain backwards compatibility with older versions of
390 * ocfs2-tools, we still store the generation value for system
391 * files. The only ones that actually matter to userspace are
392 * the journals, but it's easier and inexpensive to just flag
393 * all system files similarly.
395 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
396 generation
= osb
->fs_generation
;
398 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
399 OCFS2_LOCK_TYPE_META
,
402 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
403 OCFS2_LOCK_TYPE_OPEN
,
407 status
= ocfs2_open_lock(inode
);
409 make_bad_inode(inode
);
413 status
= ocfs2_meta_lock(inode
, NULL
, 0);
415 make_bad_inode(inode
);
421 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
422 status
= ocfs2_try_open_lock(inode
, 0);
424 make_bad_inode(inode
);
429 status
= ocfs2_read_block(osb
, args
->fi_blkno
, &bh
, 0,
430 can_lock
? inode
: NULL
);
437 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
438 if (!OCFS2_IS_VALID_DINODE(fe
)) {
439 mlog(ML_ERROR
, "Invalid dinode #%llu: signature = %.*s\n",
440 (unsigned long long)fe
->i_blkno
, 7, fe
->i_signature
);
445 * This is a code bug. Right now the caller needs to
446 * understand whether it is asking for a system file inode or
447 * not so the proper lock names can be built.
449 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
450 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
451 "Inode %llu: system file state is ambigous\n",
452 (unsigned long long)args
->fi_blkno
);
454 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
455 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
456 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
458 if (ocfs2_populate_inode(inode
, fe
, 0) < 0)
461 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
467 ocfs2_meta_unlock(inode
, 0);
470 make_bad_inode(inode
);
479 void ocfs2_sync_blockdev(struct super_block
*sb
)
481 sync_blockdev(sb
->s_bdev
);
484 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
486 struct buffer_head
*fe_bh
)
489 struct ocfs2_truncate_context
*tc
= NULL
;
490 struct ocfs2_dinode
*fe
;
491 handle_t
*handle
= NULL
;
495 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
497 if (fe
->i_clusters
) {
498 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
499 if (IS_ERR(handle
)) {
500 status
= PTR_ERR(handle
);
505 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
506 OCFS2_JOURNAL_ACCESS_WRITE
);
512 i_size_write(inode
, 0);
514 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
520 ocfs2_commit_trans(osb
, handle
);
523 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
529 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
538 ocfs2_commit_trans(osb
, handle
);
543 static int ocfs2_remove_inode(struct inode
*inode
,
544 struct buffer_head
*di_bh
,
545 struct inode
*orphan_dir_inode
,
546 struct buffer_head
*orphan_dir_bh
)
549 struct inode
*inode_alloc_inode
= NULL
;
550 struct buffer_head
*inode_alloc_bh
= NULL
;
552 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
553 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
556 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
557 le16_to_cpu(di
->i_suballoc_slot
));
558 if (!inode_alloc_inode
) {
564 mutex_lock(&inode_alloc_inode
->i_mutex
);
565 status
= ocfs2_meta_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
567 mutex_unlock(&inode_alloc_inode
->i_mutex
);
573 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
574 if (IS_ERR(handle
)) {
575 status
= PTR_ERR(handle
);
580 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
587 /* set the inodes dtime */
588 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
589 OCFS2_JOURNAL_ACCESS_WRITE
);
595 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
596 le32_and_cpu(&di
->i_flags
, ~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
598 status
= ocfs2_journal_dirty(handle
, di_bh
);
604 ocfs2_remove_from_cache(inode
, di_bh
);
606 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
612 ocfs2_commit_trans(osb
, handle
);
614 ocfs2_meta_unlock(inode_alloc_inode
, 1);
615 mutex_unlock(&inode_alloc_inode
->i_mutex
);
616 brelse(inode_alloc_bh
);
618 iput(inode_alloc_inode
);
624 * Serialize with orphan dir recovery. If the process doing
625 * recovery on this orphan dir does an iget() with the dir
626 * i_mutex held, we'll deadlock here. Instead we detect this
627 * and exit early - recovery will wipe this inode for us.
629 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
634 spin_lock(&osb
->osb_lock
);
635 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
636 mlog(0, "Recovery is happening on orphan dir %d, will skip "
637 "this inode\n", slot
);
641 /* This signals to the orphan recovery process that it should
642 * wait for us to handle the wipe. */
643 osb
->osb_orphan_wipes
[slot
]++;
645 spin_unlock(&osb
->osb_lock
);
649 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
652 spin_lock(&osb
->osb_lock
);
653 osb
->osb_orphan_wipes
[slot
]--;
654 spin_unlock(&osb
->osb_lock
);
656 wake_up(&osb
->osb_wipe_event
);
659 static int ocfs2_wipe_inode(struct inode
*inode
,
660 struct buffer_head
*di_bh
)
662 int status
, orphaned_slot
;
663 struct inode
*orphan_dir_inode
= NULL
;
664 struct buffer_head
*orphan_dir_bh
= NULL
;
665 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
666 struct ocfs2_dinode
*di
;
668 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
669 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
671 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
675 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
676 ORPHAN_DIR_SYSTEM_INODE
,
678 if (!orphan_dir_inode
) {
684 /* Lock the orphan dir. The lock will be held for the entire
685 * delete_inode operation. We do this now to avoid races with
686 * recovery completion on other nodes. */
687 mutex_lock(&orphan_dir_inode
->i_mutex
);
688 status
= ocfs2_meta_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
690 mutex_unlock(&orphan_dir_inode
->i_mutex
);
696 /* we do this while holding the orphan dir lock because we
697 * don't want recovery being run from another node to vote for
698 * an inode delete on us -- this will result in two nodes
699 * truncating the same file! */
700 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
703 goto bail_unlock_dir
;
706 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
712 ocfs2_meta_unlock(orphan_dir_inode
, 1);
713 mutex_unlock(&orphan_dir_inode
->i_mutex
);
714 brelse(orphan_dir_bh
);
716 iput(orphan_dir_inode
);
717 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
722 /* There is a series of simple checks that should be done before a
723 * vote is even considered. Encapsulate those in this function. */
724 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
727 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
728 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
730 /* We shouldn't be getting here for the root directory
732 if (inode
== osb
->root_inode
) {
733 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
737 /* If we're coming from process_vote we can't go into our own
738 * voting [hello, deadlock city!], so unforuntately we just
739 * have to skip deleting this guy. That's OK though because
740 * the node who's doing the actual deleting should handle it
742 if (current
== osb
->vote_task
) {
743 mlog(0, "Skipping delete of %lu because we're currently "
744 "in process_vote\n", inode
->i_ino
);
748 spin_lock(&oi
->ip_lock
);
749 /* OCFS2 *never* deletes system files. This should technically
750 * never get here as system file inodes should always have a
751 * positive link count. */
752 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
753 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
754 (unsigned long long)oi
->ip_blkno
);
758 /* If we have voted "yes" on the wipe of this inode for
759 * another node, it will be marked here so we can safely skip
760 * it. Recovery will cleanup any inodes we might inadvertantly
762 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
763 mlog(0, "Skipping delete of %lu because another node "
764 "has done this for us.\n", inode
->i_ino
);
770 spin_unlock(&oi
->ip_lock
);
775 /* Query the cluster to determine whether we should wipe an inode from
778 * Requires the inode to have the cluster lock. */
779 static int ocfs2_query_inode_wipe(struct inode
*inode
,
780 struct buffer_head
*di_bh
,
784 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
785 struct ocfs2_dinode
*di
;
789 /* While we were waiting for the cluster lock in
790 * ocfs2_delete_inode, another node might have asked to delete
791 * the inode. Recheck our flags to catch this. */
792 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
793 mlog(0, "Skipping delete of %llu because flags changed\n",
794 (unsigned long long)oi
->ip_blkno
);
798 /* Now that we have an up to date inode, we can double check
800 if (inode
->i_nlink
) {
801 mlog(0, "Skipping delete of %llu because nlink = %u\n",
802 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
806 /* Do some basic inode verification... */
807 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
808 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
809 /* for lack of a better error? */
812 "Inode %llu (on-disk %llu) not orphaned! "
813 "Disk flags 0x%x, inode flags 0x%x\n",
814 (unsigned long long)oi
->ip_blkno
,
815 (unsigned long long)di
->i_blkno
, di
->i_flags
,
820 /* has someone already deleted us?! baaad... */
828 * This is how ocfs2 determines whether an inode is still live
829 * within the cluster. Every node takes a shared read lock on
830 * the inode open lock in ocfs2_read_locked_inode(). When we
831 * get to ->delete_inode(), each node tries to convert it's
832 * lock to an exclusive. Trylocks are serialized by the inode
833 * meta data lock. If the upconvert suceeds, we know the inode
834 * is no longer live and can be deleted.
836 * Though we call this with the meta data lock held, the
837 * trylock keeps us from ABBA deadlock.
839 status
= ocfs2_try_open_lock(inode
, 1);
840 if (status
== -EAGAIN
) {
842 mlog(0, "Skipping delete of %llu because it is in use on"
843 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
852 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
853 (unsigned long long)oi
->ip_blkno
,
854 le16_to_cpu(di
->i_orphaned_slot
));
860 /* Support function for ocfs2_delete_inode. Will help us keep the
861 * inode data in a consistent state for clear_inode. Always truncates
862 * pages, optionally sync's them first. */
863 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
866 mlog(0, "Cleanup inode %llu, sync = %d\n",
867 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
869 write_inode_now(inode
, 1);
870 truncate_inode_pages(&inode
->i_data
, 0);
873 void ocfs2_delete_inode(struct inode
*inode
)
876 sigset_t blocked
, oldset
;
877 struct buffer_head
*di_bh
= NULL
;
879 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
881 if (is_bad_inode(inode
)) {
882 mlog(0, "Skipping delete of bad inode\n");
886 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
887 /* It's probably not necessary to truncate_inode_pages
888 * here but we do it for safety anyway (it will most
889 * likely be a no-op anyway) */
890 ocfs2_cleanup_delete_inode(inode
, 0);
894 /* We want to block signals in delete_inode as the lock and
895 * messaging paths may return us -ERESTARTSYS. Which would
896 * cause us to exit early, resulting in inodes being orphaned
898 sigfillset(&blocked
);
899 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
902 ocfs2_cleanup_delete_inode(inode
, 1);
906 /* Lock down the inode. This gives us an up to date view of
907 * it's metadata (for verification), and allows us to
908 * serialize delete_inode votes.
910 * Even though we might be doing a truncate, we don't take the
911 * allocation lock here as it won't be needed - nobody will
912 * have the file open.
914 status
= ocfs2_meta_lock(inode
, &di_bh
, 1);
916 if (status
!= -ENOENT
)
918 ocfs2_cleanup_delete_inode(inode
, 0);
922 /* Query the cluster. This will be the final decision made
923 * before we go ahead and wipe the inode. */
924 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
925 if (!wipe
|| status
< 0) {
926 /* Error and inode busy vote both mean we won't be
927 * removing the inode, so they take almost the same
932 /* Someone in the cluster has voted to not wipe this
933 * inode, or it was never completely orphaned. Write
934 * out the pages and exit now. */
935 ocfs2_cleanup_delete_inode(inode
, 1);
936 goto bail_unlock_inode
;
939 ocfs2_cleanup_delete_inode(inode
, 0);
941 status
= ocfs2_wipe_inode(inode
, di_bh
);
943 if (status
!= -EDEADLK
)
945 goto bail_unlock_inode
;
949 * Mark the inode as successfully deleted.
951 * This is important for ocfs2_clear_inode() as it will check
952 * this flag and skip any checkpointing work
954 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
955 * the LVB for other nodes.
957 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
960 ocfs2_meta_unlock(inode
, 1);
963 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
971 void ocfs2_clear_inode(struct inode
*inode
)
974 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
981 mlog(0, "Clearing inode: %llu, nlink = %u\n",
982 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
984 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
985 "Inode=%lu\n", inode
->i_ino
);
987 /* For remove delete_inode vote, we hold open lock before,
988 * now it is time to unlock PR and EX open locks. */
989 ocfs2_open_unlock(inode
);
991 /* Do these before all the other work so that we don't bounce
992 * the vote thread while waiting to destroy the locks. */
993 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
994 ocfs2_mark_lockres_freeing(&oi
->ip_meta_lockres
);
995 ocfs2_mark_lockres_freeing(&oi
->ip_data_lockres
);
996 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
998 /* We very well may get a clear_inode before all an inodes
999 * metadata has hit disk. Of course, we can't drop any cluster
1000 * locks until the journal has finished with it. The only
1001 * exception here are successfully wiped inodes - their
1002 * metadata can now be considered to be part of the system
1003 * inodes from which it came. */
1004 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1005 ocfs2_checkpoint_inode(inode
);
1007 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1008 "Clear inode of %llu, inode has io markers\n",
1009 (unsigned long long)oi
->ip_blkno
);
1011 ocfs2_extent_map_trunc(inode
, 0);
1013 status
= ocfs2_drop_inode_locks(inode
);
1017 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1018 ocfs2_lock_res_free(&oi
->ip_meta_lockres
);
1019 ocfs2_lock_res_free(&oi
->ip_data_lockres
);
1020 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1022 ocfs2_metadata_cache_purge(inode
);
1024 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1025 "Clear inode of %llu, inode has %u cache items\n",
1026 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1028 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1029 "Clear inode of %llu, inode has a bad flag\n",
1030 (unsigned long long)oi
->ip_blkno
);
1032 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1033 "Clear inode of %llu, inode is locked\n",
1034 (unsigned long long)oi
->ip_blkno
);
1036 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1037 "Clear inode of %llu, io_mutex is locked\n",
1038 (unsigned long long)oi
->ip_blkno
);
1039 mutex_unlock(&oi
->ip_io_mutex
);
1042 * down_trylock() returns 0, down_write_trylock() returns 1
1045 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1046 "Clear inode of %llu, alloc_sem is locked\n",
1047 (unsigned long long)oi
->ip_blkno
);
1048 up_write(&oi
->ip_alloc_sem
);
1050 mlog_bug_on_msg(oi
->ip_open_count
,
1051 "Clear inode of %llu has open count %d\n",
1052 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1054 /* Clear all other flags. */
1055 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1056 oi
->ip_created_trans
= 0;
1057 oi
->ip_last_trans
= 0;
1058 oi
->ip_dir_start_lookup
= 0;
1059 oi
->ip_blkno
= 0ULL;
1065 /* Called under inode_lock, with no more references on the
1066 * struct inode, so it's safe here to check the flags field
1067 * and to manipulate i_nlink without any other locks. */
1068 void ocfs2_drop_inode(struct inode
*inode
)
1070 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1074 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1075 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1077 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1078 generic_delete_inode(inode
);
1080 generic_drop_inode(inode
);
1086 * TODO: this should probably be merged into ocfs2_get_block
1088 * However, you now need to pay attention to the cont_prepare_write()
1089 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1090 * expects never to extend).
1092 struct buffer_head
*ocfs2_bread(struct inode
*inode
,
1093 int block
, int *err
, int reada
)
1095 struct buffer_head
*bh
= NULL
;
1098 int readflags
= OCFS2_BH_CACHED
;
1101 readflags
|= OCFS2_BH_READAHEAD
;
1103 if (((u64
)block
<< inode
->i_sb
->s_blocksize_bits
) >=
1104 i_size_read(inode
)) {
1109 tmperr
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
1116 tmperr
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), p_blkno
, &bh
,
1136 * This is called from our getattr.
1138 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1140 struct inode
*inode
= dentry
->d_inode
;
1143 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1144 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1147 mlog(0, "eep, no inode!\n");
1152 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1153 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1154 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1155 mlog(0, "inode deleted!\n");
1159 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1161 /* Let ocfs2_meta_lock do the work of updating our struct
1163 status
= ocfs2_meta_lock(inode
, NULL
, 0);
1165 if (status
!= -ENOENT
)
1169 ocfs2_meta_unlock(inode
, 0);
1177 * Updates a disk inode from a
1179 * Only takes ip_lock.
1181 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1182 struct inode
*inode
,
1183 struct buffer_head
*bh
)
1186 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1188 mlog_entry("(inode %llu)\n",
1189 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1191 status
= ocfs2_journal_access(handle
, inode
, bh
,
1192 OCFS2_JOURNAL_ACCESS_WRITE
);
1198 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1199 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1200 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1201 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1203 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1204 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1205 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1206 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1207 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1208 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1209 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1210 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1211 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1212 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1213 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1215 status
= ocfs2_journal_dirty(handle
, bh
);
1228 * Updates a struct inode from a disk inode.
1229 * does no i/o, only takes ip_lock.
1231 void ocfs2_refresh_inode(struct inode
*inode
,
1232 struct ocfs2_dinode
*fe
)
1234 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1236 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1237 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1238 ocfs2_set_inode_flags(inode
);
1239 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1240 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1241 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1242 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1243 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1244 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1245 inode
->i_blocks
= 0;
1247 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1248 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1249 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1250 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1251 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1252 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1253 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1255 spin_unlock(&OCFS2_I(inode
)->ip_lock
);