1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * vfs' aops, fops, dops and iops
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
32 #include <asm/byteorder.h>
34 #define MLOG_MASK_PREFIX ML_INODE
35 #include <cluster/masklog.h>
41 #include "extent_map.h"
43 #include "heartbeat.h"
54 #include "buffer_head_io.h"
56 struct ocfs2_find_inode_args
60 unsigned int fi_flags
;
61 unsigned int fi_sysfile_type
;
64 static struct lock_class_key ocfs2_sysfile_lock_key
[NUM_SYSTEM_INODES
];
66 static int ocfs2_read_locked_inode(struct inode
*inode
,
67 struct ocfs2_find_inode_args
*args
);
68 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
69 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
70 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
72 struct buffer_head
*fe_bh
);
74 void ocfs2_set_inode_flags(struct inode
*inode
)
76 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
78 inode
->i_flags
&= ~(S_IMMUTABLE
|
79 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
81 if (flags
& OCFS2_IMMUTABLE_FL
)
82 inode
->i_flags
|= S_IMMUTABLE
;
84 if (flags
& OCFS2_SYNC_FL
)
85 inode
->i_flags
|= S_SYNC
;
86 if (flags
& OCFS2_APPEND_FL
)
87 inode
->i_flags
|= S_APPEND
;
88 if (flags
& OCFS2_NOATIME_FL
)
89 inode
->i_flags
|= S_NOATIME
;
90 if (flags
& OCFS2_DIRSYNC_FL
)
91 inode
->i_flags
|= S_DIRSYNC
;
94 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
95 void ocfs2_get_inode_flags(struct ocfs2_inode_info
*oi
)
97 unsigned int flags
= oi
->vfs_inode
.i_flags
;
99 oi
->ip_attr
&= ~(OCFS2_SYNC_FL
|OCFS2_APPEND_FL
|
100 OCFS2_IMMUTABLE_FL
|OCFS2_NOATIME_FL
|OCFS2_DIRSYNC_FL
);
102 oi
->ip_attr
|= OCFS2_SYNC_FL
;
103 if (flags
& S_APPEND
)
104 oi
->ip_attr
|= OCFS2_APPEND_FL
;
105 if (flags
& S_IMMUTABLE
)
106 oi
->ip_attr
|= OCFS2_IMMUTABLE_FL
;
107 if (flags
& S_NOATIME
)
108 oi
->ip_attr
|= OCFS2_NOATIME_FL
;
109 if (flags
& S_DIRSYNC
)
110 oi
->ip_attr
|= OCFS2_DIRSYNC_FL
;
113 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, unsigned flags
,
116 struct inode
*inode
= NULL
;
117 struct super_block
*sb
= osb
->sb
;
118 struct ocfs2_find_inode_args args
;
120 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
122 /* Ok. By now we've either got the offsets passed to us by the
123 * caller, or we just pulled them off the bh. Lets do some
124 * sanity checks to make sure they're OK. */
126 inode
= ERR_PTR(-EINVAL
);
127 mlog_errno(PTR_ERR(inode
));
131 args
.fi_blkno
= blkno
;
132 args
.fi_flags
= flags
;
133 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
134 args
.fi_sysfile_type
= sysfile_type
;
136 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
137 ocfs2_init_locked_inode
, &args
);
138 /* inode was *not* in the inode cache. 2.6.x requires
139 * us to do our own read_inode call and unlock it
141 if (inode
&& inode
->i_state
& I_NEW
) {
142 mlog(0, "Inode was not in inode cache, reading it.\n");
143 ocfs2_read_locked_inode(inode
, &args
);
144 unlock_new_inode(inode
);
147 inode
= ERR_PTR(-ENOMEM
);
148 mlog_errno(PTR_ERR(inode
));
151 if (is_bad_inode(inode
)) {
153 inode
= ERR_PTR(-ESTALE
);
158 if (!IS_ERR(inode
)) {
159 mlog(0, "returning inode with number %llu\n",
160 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
161 mlog_exit_ptr(inode
);
169 * here's how inodes get read from disk:
170 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
171 * found? : return the in-memory inode
172 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
175 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
177 struct ocfs2_find_inode_args
*args
= NULL
;
178 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
181 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
185 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
187 if (oi
->ip_blkno
!= args
->fi_blkno
)
197 * initialize the new inode, but don't do anything that would cause
199 * return 0 on success, 1 on failure
201 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
203 struct ocfs2_find_inode_args
*args
= opaque
;
205 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
207 inode
->i_ino
= args
->fi_ino
;
208 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
209 if (args
->fi_sysfile_type
!= 0)
210 lockdep_set_class(&inode
->i_mutex
,
211 &ocfs2_sysfile_lock_key
[args
->fi_sysfile_type
]);
217 int ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
220 struct super_block
*sb
;
221 struct ocfs2_super
*osb
;
222 int status
= -EINVAL
;
225 mlog_entry("(0x%p, size:%llu)\n", inode
,
226 (unsigned long long)le64_to_cpu(fe
->i_size
));
231 if ((osb
->s_mount_opt
& OCFS2_MOUNT_LOCALFLOCKS
) ||
232 ocfs2_mount_local(osb
) || !ocfs2_stack_supports_plocks())
235 /* this means that read_inode cannot create a superblock inode
236 * today. change if needed. */
237 if (!OCFS2_IS_VALID_DINODE(fe
) ||
238 !(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
239 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
240 "signature = %.*s, flags = 0x%x\n",
242 (unsigned long long)le64_to_cpu(fe
->i_blkno
), 7,
243 fe
->i_signature
, le32_to_cpu(fe
->i_flags
));
247 if (le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
) {
248 mlog(ML_ERROR
, "file entry generation does not match "
249 "superblock! osb->fs_generation=%x, "
250 "fe->i_fs_generation=%x\n",
251 osb
->fs_generation
, le32_to_cpu(fe
->i_fs_generation
));
255 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
256 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
257 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
259 inode
->i_version
= 1;
260 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
261 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
262 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
263 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
264 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
266 /* Fast symlinks will have i_size but no allocated clusters. */
267 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
270 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
271 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
272 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
273 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
274 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
275 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
276 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
277 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
279 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
281 "ip_blkno %llu != i_blkno %llu!\n",
282 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
283 (unsigned long long)le64_to_cpu(fe
->i_blkno
));
285 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
287 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
288 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
290 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
291 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
292 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
293 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
294 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
295 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
296 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
297 /* we can't actually hit this as read_inode can't
298 * handle superblocks today ;-) */
302 switch (inode
->i_mode
& S_IFMT
) {
305 inode
->i_fop
= &ocfs2_fops
;
307 inode
->i_fop
= &ocfs2_fops_no_plocks
;
308 inode
->i_op
= &ocfs2_file_iops
;
309 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
312 inode
->i_op
= &ocfs2_dir_iops
;
314 inode
->i_fop
= &ocfs2_dops
;
316 inode
->i_fop
= &ocfs2_dops_no_plocks
;
317 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
320 if (ocfs2_inode_is_fast_symlink(inode
))
321 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
323 inode
->i_op
= &ocfs2_symlink_inode_operations
;
324 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
327 inode
->i_op
= &ocfs2_special_file_iops
;
328 init_special_inode(inode
, inode
->i_mode
,
334 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
335 le64_to_cpu(fe
->i_blkno
));
338 * If we ever want to create system files from kernel,
339 * the generation argument to
340 * ocfs2_inode_lock_res_init() will have to change.
342 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
344 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
345 OCFS2_LOCK_TYPE_META
, 0, inode
);
347 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
348 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
351 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
352 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
355 ocfs2_set_inode_flags(inode
);
363 static int ocfs2_read_locked_inode(struct inode
*inode
,
364 struct ocfs2_find_inode_args
*args
)
366 struct super_block
*sb
;
367 struct ocfs2_super
*osb
;
368 struct ocfs2_dinode
*fe
;
369 struct buffer_head
*bh
= NULL
;
370 int status
, can_lock
;
373 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
376 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
377 mlog(ML_ERROR
, "bad inode\n");
384 mlog(ML_ERROR
, "bad inode args\n");
385 make_bad_inode(inode
);
390 * To improve performance of cold-cache inode stats, we take
391 * the cluster lock here if possible.
393 * Generally, OCFS2 never trusts the contents of an inode
394 * unless it's holding a cluster lock, so taking it here isn't
395 * a correctness issue as much as it is a performance
398 * There are three times when taking the lock is not a good idea:
400 * 1) During startup, before we have initialized the DLM.
402 * 2) If we are reading certain system files which never get
403 * cluster locks (local alloc, truncate log).
405 * 3) If the process doing the iget() is responsible for
406 * orphan dir recovery. We're holding the orphan dir lock and
407 * can get into a deadlock with another process on another
408 * node in ->delete_inode().
410 * #1 and #2 can be simply solved by never taking the lock
411 * here for system files (which are the only type we read
412 * during mount). It's a heavier approach, but our main
413 * concern is user-accesible files anyway.
415 * #3 works itself out because we'll eventually take the
416 * cluster lock before trusting anything anyway.
418 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
419 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
420 && !ocfs2_mount_local(osb
);
423 * To maintain backwards compatibility with older versions of
424 * ocfs2-tools, we still store the generation value for system
425 * files. The only ones that actually matter to userspace are
426 * the journals, but it's easier and inexpensive to just flag
427 * all system files similarly.
429 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
430 generation
= osb
->fs_generation
;
432 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_inode_lockres
,
433 OCFS2_LOCK_TYPE_META
,
436 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
437 OCFS2_LOCK_TYPE_OPEN
,
441 status
= ocfs2_open_lock(inode
);
443 make_bad_inode(inode
);
447 status
= ocfs2_inode_lock(inode
, NULL
, 0);
449 make_bad_inode(inode
);
455 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
456 status
= ocfs2_try_open_lock(inode
, 0);
458 make_bad_inode(inode
);
463 status
= ocfs2_read_block(osb
, args
->fi_blkno
, &bh
, 0,
464 can_lock
? inode
: NULL
);
471 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
472 if (!OCFS2_IS_VALID_DINODE(fe
)) {
473 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
474 (unsigned long long)args
->fi_blkno
, 7,
480 * This is a code bug. Right now the caller needs to
481 * understand whether it is asking for a system file inode or
482 * not so the proper lock names can be built.
484 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
485 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
486 "Inode %llu: system file state is ambigous\n",
487 (unsigned long long)args
->fi_blkno
);
489 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
490 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
491 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
493 if (ocfs2_populate_inode(inode
, fe
, 0) < 0)
496 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
502 ocfs2_inode_unlock(inode
, 0);
505 make_bad_inode(inode
);
514 void ocfs2_sync_blockdev(struct super_block
*sb
)
516 sync_blockdev(sb
->s_bdev
);
519 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
521 struct buffer_head
*fe_bh
)
524 struct ocfs2_truncate_context
*tc
= NULL
;
525 struct ocfs2_dinode
*fe
;
526 handle_t
*handle
= NULL
;
530 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
533 * This check will also skip truncate of inodes with inline
534 * data and fast symlinks.
536 if (fe
->i_clusters
) {
537 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
538 if (IS_ERR(handle
)) {
539 status
= PTR_ERR(handle
);
544 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
545 OCFS2_JOURNAL_ACCESS_WRITE
);
551 i_size_write(inode
, 0);
553 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
559 ocfs2_commit_trans(osb
, handle
);
562 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
568 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
577 ocfs2_commit_trans(osb
, handle
);
582 static int ocfs2_remove_inode(struct inode
*inode
,
583 struct buffer_head
*di_bh
,
584 struct inode
*orphan_dir_inode
,
585 struct buffer_head
*orphan_dir_bh
)
588 struct inode
*inode_alloc_inode
= NULL
;
589 struct buffer_head
*inode_alloc_bh
= NULL
;
591 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
592 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
595 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
596 le16_to_cpu(di
->i_suballoc_slot
));
597 if (!inode_alloc_inode
) {
603 mutex_lock(&inode_alloc_inode
->i_mutex
);
604 status
= ocfs2_inode_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
606 mutex_unlock(&inode_alloc_inode
->i_mutex
);
612 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
613 if (IS_ERR(handle
)) {
614 status
= PTR_ERR(handle
);
619 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
626 /* set the inodes dtime */
627 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
628 OCFS2_JOURNAL_ACCESS_WRITE
);
634 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
635 di
->i_flags
&= cpu_to_le32(~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
637 status
= ocfs2_journal_dirty(handle
, di_bh
);
643 ocfs2_remove_from_cache(inode
, di_bh
);
645 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
651 ocfs2_commit_trans(osb
, handle
);
653 ocfs2_inode_unlock(inode_alloc_inode
, 1);
654 mutex_unlock(&inode_alloc_inode
->i_mutex
);
655 brelse(inode_alloc_bh
);
657 iput(inode_alloc_inode
);
663 * Serialize with orphan dir recovery. If the process doing
664 * recovery on this orphan dir does an iget() with the dir
665 * i_mutex held, we'll deadlock here. Instead we detect this
666 * and exit early - recovery will wipe this inode for us.
668 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
673 spin_lock(&osb
->osb_lock
);
674 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
675 mlog(0, "Recovery is happening on orphan dir %d, will skip "
676 "this inode\n", slot
);
680 /* This signals to the orphan recovery process that it should
681 * wait for us to handle the wipe. */
682 osb
->osb_orphan_wipes
[slot
]++;
684 spin_unlock(&osb
->osb_lock
);
688 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
691 spin_lock(&osb
->osb_lock
);
692 osb
->osb_orphan_wipes
[slot
]--;
693 spin_unlock(&osb
->osb_lock
);
695 wake_up(&osb
->osb_wipe_event
);
698 static int ocfs2_wipe_inode(struct inode
*inode
,
699 struct buffer_head
*di_bh
)
701 int status
, orphaned_slot
;
702 struct inode
*orphan_dir_inode
= NULL
;
703 struct buffer_head
*orphan_dir_bh
= NULL
;
704 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
705 struct ocfs2_dinode
*di
;
707 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
708 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
710 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
714 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
715 ORPHAN_DIR_SYSTEM_INODE
,
717 if (!orphan_dir_inode
) {
723 /* Lock the orphan dir. The lock will be held for the entire
724 * delete_inode operation. We do this now to avoid races with
725 * recovery completion on other nodes. */
726 mutex_lock(&orphan_dir_inode
->i_mutex
);
727 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
729 mutex_unlock(&orphan_dir_inode
->i_mutex
);
735 /* we do this while holding the orphan dir lock because we
736 * don't want recovery being run from another node to try an
737 * inode delete underneath us -- this will result in two nodes
738 * truncating the same file! */
739 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
742 goto bail_unlock_dir
;
745 /*Free extended attribute resources associated with this inode.*/
746 status
= ocfs2_xattr_remove(inode
, di_bh
);
749 goto bail_unlock_dir
;
752 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
758 ocfs2_inode_unlock(orphan_dir_inode
, 1);
759 mutex_unlock(&orphan_dir_inode
->i_mutex
);
760 brelse(orphan_dir_bh
);
762 iput(orphan_dir_inode
);
763 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
768 /* There is a series of simple checks that should be done before a
769 * trylock is even considered. Encapsulate those in this function. */
770 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
773 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
774 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
776 /* We shouldn't be getting here for the root directory
778 if (inode
== osb
->root_inode
) {
779 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
783 /* If we're coming from downconvert_thread we can't go into our own
784 * voting [hello, deadlock city!], so unforuntately we just
785 * have to skip deleting this guy. That's OK though because
786 * the node who's doing the actual deleting should handle it
788 if (current
== osb
->dc_task
) {
789 mlog(0, "Skipping delete of %lu because we're currently "
790 "in downconvert\n", inode
->i_ino
);
794 spin_lock(&oi
->ip_lock
);
795 /* OCFS2 *never* deletes system files. This should technically
796 * never get here as system file inodes should always have a
797 * positive link count. */
798 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
799 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
800 (unsigned long long)oi
->ip_blkno
);
804 /* If we have allowd wipe of this inode for another node, it
805 * will be marked here so we can safely skip it. Recovery will
806 * cleanup any inodes we might inadvertantly skip here. */
807 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
808 mlog(0, "Skipping delete of %lu because another node "
809 "has done this for us.\n", inode
->i_ino
);
815 spin_unlock(&oi
->ip_lock
);
820 /* Query the cluster to determine whether we should wipe an inode from
823 * Requires the inode to have the cluster lock. */
824 static int ocfs2_query_inode_wipe(struct inode
*inode
,
825 struct buffer_head
*di_bh
,
829 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
830 struct ocfs2_dinode
*di
;
834 /* While we were waiting for the cluster lock in
835 * ocfs2_delete_inode, another node might have asked to delete
836 * the inode. Recheck our flags to catch this. */
837 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
838 mlog(0, "Skipping delete of %llu because flags changed\n",
839 (unsigned long long)oi
->ip_blkno
);
843 /* Now that we have an up to date inode, we can double check
845 if (inode
->i_nlink
) {
846 mlog(0, "Skipping delete of %llu because nlink = %u\n",
847 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
851 /* Do some basic inode verification... */
852 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
853 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
854 /* for lack of a better error? */
857 "Inode %llu (on-disk %llu) not orphaned! "
858 "Disk flags 0x%x, inode flags 0x%x\n",
859 (unsigned long long)oi
->ip_blkno
,
860 (unsigned long long)le64_to_cpu(di
->i_blkno
),
861 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
865 /* has someone already deleted us?! baaad... */
873 * This is how ocfs2 determines whether an inode is still live
874 * within the cluster. Every node takes a shared read lock on
875 * the inode open lock in ocfs2_read_locked_inode(). When we
876 * get to ->delete_inode(), each node tries to convert it's
877 * lock to an exclusive. Trylocks are serialized by the inode
878 * meta data lock. If the upconvert suceeds, we know the inode
879 * is no longer live and can be deleted.
881 * Though we call this with the meta data lock held, the
882 * trylock keeps us from ABBA deadlock.
884 status
= ocfs2_try_open_lock(inode
, 1);
885 if (status
== -EAGAIN
) {
887 mlog(0, "Skipping delete of %llu because it is in use on "
888 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
897 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
898 (unsigned long long)oi
->ip_blkno
,
899 le16_to_cpu(di
->i_orphaned_slot
));
905 /* Support function for ocfs2_delete_inode. Will help us keep the
906 * inode data in a consistent state for clear_inode. Always truncates
907 * pages, optionally sync's them first. */
908 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
911 mlog(0, "Cleanup inode %llu, sync = %d\n",
912 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
914 write_inode_now(inode
, 1);
915 truncate_inode_pages(&inode
->i_data
, 0);
918 void ocfs2_delete_inode(struct inode
*inode
)
921 sigset_t blocked
, oldset
;
922 struct buffer_head
*di_bh
= NULL
;
924 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
926 if (is_bad_inode(inode
)) {
927 mlog(0, "Skipping delete of bad inode\n");
931 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
932 /* It's probably not necessary to truncate_inode_pages
933 * here but we do it for safety anyway (it will most
934 * likely be a no-op anyway) */
935 ocfs2_cleanup_delete_inode(inode
, 0);
939 /* We want to block signals in delete_inode as the lock and
940 * messaging paths may return us -ERESTARTSYS. Which would
941 * cause us to exit early, resulting in inodes being orphaned
943 sigfillset(&blocked
);
944 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
947 ocfs2_cleanup_delete_inode(inode
, 1);
951 /* Lock down the inode. This gives us an up to date view of
952 * it's metadata (for verification), and allows us to
953 * serialize delete_inode on multiple nodes.
955 * Even though we might be doing a truncate, we don't take the
956 * allocation lock here as it won't be needed - nobody will
957 * have the file open.
959 status
= ocfs2_inode_lock(inode
, &di_bh
, 1);
961 if (status
!= -ENOENT
)
963 ocfs2_cleanup_delete_inode(inode
, 0);
967 /* Query the cluster. This will be the final decision made
968 * before we go ahead and wipe the inode. */
969 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
970 if (!wipe
|| status
< 0) {
971 /* Error and remote inode busy both mean we won't be
972 * removing the inode, so they take almost the same
977 /* Someone in the cluster has disallowed a wipe of
978 * this inode, or it was never completely
979 * orphaned. Write out the pages and exit now. */
980 ocfs2_cleanup_delete_inode(inode
, 1);
981 goto bail_unlock_inode
;
984 ocfs2_cleanup_delete_inode(inode
, 0);
986 status
= ocfs2_wipe_inode(inode
, di_bh
);
988 if (status
!= -EDEADLK
)
990 goto bail_unlock_inode
;
994 * Mark the inode as successfully deleted.
996 * This is important for ocfs2_clear_inode() as it will check
997 * this flag and skip any checkpointing work
999 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1000 * the LVB for other nodes.
1002 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
1005 ocfs2_inode_unlock(inode
, 1);
1008 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
1016 void ocfs2_clear_inode(struct inode
*inode
)
1019 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1026 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1027 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1029 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1030 "Inode=%lu\n", inode
->i_ino
);
1032 /* To preven remote deletes we hold open lock before, now it
1033 * is time to unlock PR and EX open locks. */
1034 ocfs2_open_unlock(inode
);
1036 /* Do these before all the other work so that we don't bounce
1037 * the downconvert thread while waiting to destroy the locks. */
1038 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1039 ocfs2_mark_lockres_freeing(&oi
->ip_inode_lockres
);
1040 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
1042 /* We very well may get a clear_inode before all an inodes
1043 * metadata has hit disk. Of course, we can't drop any cluster
1044 * locks until the journal has finished with it. The only
1045 * exception here are successfully wiped inodes - their
1046 * metadata can now be considered to be part of the system
1047 * inodes from which it came. */
1048 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1049 ocfs2_checkpoint_inode(inode
);
1051 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1052 "Clear inode of %llu, inode has io markers\n",
1053 (unsigned long long)oi
->ip_blkno
);
1055 ocfs2_extent_map_trunc(inode
, 0);
1057 status
= ocfs2_drop_inode_locks(inode
);
1061 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1062 ocfs2_lock_res_free(&oi
->ip_inode_lockres
);
1063 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1065 ocfs2_metadata_cache_purge(inode
);
1067 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1068 "Clear inode of %llu, inode has %u cache items\n",
1069 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1071 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1072 "Clear inode of %llu, inode has a bad flag\n",
1073 (unsigned long long)oi
->ip_blkno
);
1075 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1076 "Clear inode of %llu, inode is locked\n",
1077 (unsigned long long)oi
->ip_blkno
);
1079 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1080 "Clear inode of %llu, io_mutex is locked\n",
1081 (unsigned long long)oi
->ip_blkno
);
1082 mutex_unlock(&oi
->ip_io_mutex
);
1085 * down_trylock() returns 0, down_write_trylock() returns 1
1088 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1089 "Clear inode of %llu, alloc_sem is locked\n",
1090 (unsigned long long)oi
->ip_blkno
);
1091 up_write(&oi
->ip_alloc_sem
);
1093 mlog_bug_on_msg(oi
->ip_open_count
,
1094 "Clear inode of %llu has open count %d\n",
1095 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1097 /* Clear all other flags. */
1098 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1099 oi
->ip_created_trans
= 0;
1100 oi
->ip_last_trans
= 0;
1101 oi
->ip_dir_start_lookup
= 0;
1102 oi
->ip_blkno
= 0ULL;
1108 /* Called under inode_lock, with no more references on the
1109 * struct inode, so it's safe here to check the flags field
1110 * and to manipulate i_nlink without any other locks. */
1111 void ocfs2_drop_inode(struct inode
*inode
)
1113 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1117 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1118 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1120 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1121 generic_delete_inode(inode
);
1123 generic_drop_inode(inode
);
1129 * TODO: this should probably be merged into ocfs2_get_block
1131 * However, you now need to pay attention to the cont_prepare_write()
1132 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1133 * expects never to extend).
1135 struct buffer_head
*ocfs2_bread(struct inode
*inode
,
1136 int block
, int *err
, int reada
)
1138 struct buffer_head
*bh
= NULL
;
1141 int readflags
= OCFS2_BH_CACHED
;
1144 readflags
|= OCFS2_BH_READAHEAD
;
1146 if (((u64
)block
<< inode
->i_sb
->s_blocksize_bits
) >=
1147 i_size_read(inode
)) {
1152 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
1153 tmperr
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
1155 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
1161 tmperr
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), p_blkno
, &bh
,
1181 * This is called from our getattr.
1183 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1185 struct inode
*inode
= dentry
->d_inode
;
1188 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1189 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1192 mlog(0, "eep, no inode!\n");
1197 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1198 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1199 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1200 mlog(0, "inode deleted!\n");
1204 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1206 /* Let ocfs2_inode_lock do the work of updating our struct
1208 status
= ocfs2_inode_lock(inode
, NULL
, 0);
1210 if (status
!= -ENOENT
)
1214 ocfs2_inode_unlock(inode
, 0);
1222 * Updates a disk inode from a
1224 * Only takes ip_lock.
1226 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1227 struct inode
*inode
,
1228 struct buffer_head
*bh
)
1231 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1233 mlog_entry("(inode %llu)\n",
1234 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1236 status
= ocfs2_journal_access(handle
, inode
, bh
,
1237 OCFS2_JOURNAL_ACCESS_WRITE
);
1243 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1244 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1245 ocfs2_get_inode_flags(OCFS2_I(inode
));
1246 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1247 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1248 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1250 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1251 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1252 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1253 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1254 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1255 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1256 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1257 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1258 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1259 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1260 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1262 status
= ocfs2_journal_dirty(handle
, bh
);
1275 * Updates a struct inode from a disk inode.
1276 * does no i/o, only takes ip_lock.
1278 void ocfs2_refresh_inode(struct inode
*inode
,
1279 struct ocfs2_dinode
*fe
)
1281 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1283 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1284 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1285 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1286 ocfs2_set_inode_flags(inode
);
1287 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1288 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1289 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1290 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1291 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1292 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1293 inode
->i_blocks
= 0;
1295 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1296 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1297 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1298 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1299 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1300 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1301 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1303 spin_unlock(&OCFS2_I(inode
)->ip_lock
);