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"
53 #include "buffer_head_io.h"
55 struct ocfs2_find_inode_args
59 unsigned int fi_flags
;
62 static int ocfs2_read_locked_inode(struct inode
*inode
,
63 struct ocfs2_find_inode_args
*args
);
64 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
);
65 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
);
66 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
68 struct buffer_head
*fe_bh
);
70 void ocfs2_set_inode_flags(struct inode
*inode
)
72 unsigned int flags
= OCFS2_I(inode
)->ip_attr
;
74 inode
->i_flags
&= ~(S_IMMUTABLE
|
75 S_SYNC
| S_APPEND
| S_NOATIME
| S_DIRSYNC
);
77 if (flags
& OCFS2_IMMUTABLE_FL
)
78 inode
->i_flags
|= S_IMMUTABLE
;
80 if (flags
& OCFS2_SYNC_FL
)
81 inode
->i_flags
|= S_SYNC
;
82 if (flags
& OCFS2_APPEND_FL
)
83 inode
->i_flags
|= S_APPEND
;
84 if (flags
& OCFS2_NOATIME_FL
)
85 inode
->i_flags
|= S_NOATIME
;
86 if (flags
& OCFS2_DIRSYNC_FL
)
87 inode
->i_flags
|= S_DIRSYNC
;
90 /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
91 void ocfs2_get_inode_flags(struct ocfs2_inode_info
*oi
)
93 unsigned int flags
= oi
->vfs_inode
.i_flags
;
95 oi
->ip_attr
&= ~(OCFS2_SYNC_FL
|OCFS2_APPEND_FL
|
96 OCFS2_IMMUTABLE_FL
|OCFS2_NOATIME_FL
|OCFS2_DIRSYNC_FL
);
98 oi
->ip_attr
|= OCFS2_SYNC_FL
;
100 oi
->ip_attr
|= OCFS2_APPEND_FL
;
101 if (flags
& S_IMMUTABLE
)
102 oi
->ip_attr
|= OCFS2_IMMUTABLE_FL
;
103 if (flags
& S_NOATIME
)
104 oi
->ip_attr
|= OCFS2_NOATIME_FL
;
105 if (flags
& S_DIRSYNC
)
106 oi
->ip_attr
|= OCFS2_DIRSYNC_FL
;
109 struct inode
*ocfs2_iget(struct ocfs2_super
*osb
, u64 blkno
, int flags
)
111 struct inode
*inode
= NULL
;
112 struct super_block
*sb
= osb
->sb
;
113 struct ocfs2_find_inode_args args
;
115 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno
);
117 /* Ok. By now we've either got the offsets passed to us by the
118 * caller, or we just pulled them off the bh. Lets do some
119 * sanity checks to make sure they're OK. */
121 inode
= ERR_PTR(-EINVAL
);
122 mlog_errno(PTR_ERR(inode
));
126 args
.fi_blkno
= blkno
;
127 args
.fi_flags
= flags
;
128 args
.fi_ino
= ino_from_blkno(sb
, blkno
);
130 inode
= iget5_locked(sb
, args
.fi_ino
, ocfs2_find_actor
,
131 ocfs2_init_locked_inode
, &args
);
132 /* inode was *not* in the inode cache. 2.6.x requires
133 * us to do our own read_inode call and unlock it
135 if (inode
&& inode
->i_state
& I_NEW
) {
136 mlog(0, "Inode was not in inode cache, reading it.\n");
137 ocfs2_read_locked_inode(inode
, &args
);
138 unlock_new_inode(inode
);
141 inode
= ERR_PTR(-ENOMEM
);
142 mlog_errno(PTR_ERR(inode
));
145 if (is_bad_inode(inode
)) {
147 inode
= ERR_PTR(-ESTALE
);
152 if (!IS_ERR(inode
)) {
153 mlog(0, "returning inode with number %llu\n",
154 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
155 mlog_exit_ptr(inode
);
163 * here's how inodes get read from disk:
164 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
165 * found? : return the in-memory inode
166 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
169 static int ocfs2_find_actor(struct inode
*inode
, void *opaque
)
171 struct ocfs2_find_inode_args
*args
= NULL
;
172 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
175 mlog_entry("(0x%p, %lu, 0x%p)\n", inode
, inode
->i_ino
, opaque
);
179 mlog_bug_on_msg(!inode
, "No inode in find actor!\n");
181 if (oi
->ip_blkno
!= args
->fi_blkno
)
191 * initialize the new inode, but don't do anything that would cause
193 * return 0 on success, 1 on failure
195 static int ocfs2_init_locked_inode(struct inode
*inode
, void *opaque
)
197 struct ocfs2_find_inode_args
*args
= opaque
;
199 mlog_entry("inode = %p, opaque = %p\n", inode
, opaque
);
201 inode
->i_ino
= args
->fi_ino
;
202 OCFS2_I(inode
)->ip_blkno
= args
->fi_blkno
;
208 int ocfs2_populate_inode(struct inode
*inode
, struct ocfs2_dinode
*fe
,
211 struct super_block
*sb
;
212 struct ocfs2_super
*osb
;
213 int status
= -EINVAL
;
215 mlog_entry("(0x%p, size:%llu)\n", inode
,
216 (unsigned long long)le64_to_cpu(fe
->i_size
));
221 /* this means that read_inode cannot create a superblock inode
222 * today. change if needed. */
223 if (!OCFS2_IS_VALID_DINODE(fe
) ||
224 !(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
))) {
225 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
226 "signature = %.*s, flags = 0x%x\n",
228 (unsigned long long)le64_to_cpu(fe
->i_blkno
), 7,
229 fe
->i_signature
, le32_to_cpu(fe
->i_flags
));
233 if (le32_to_cpu(fe
->i_fs_generation
) != osb
->fs_generation
) {
234 mlog(ML_ERROR
, "file entry generation does not match "
235 "superblock! osb->fs_generation=%x, "
236 "fe->i_fs_generation=%x\n",
237 osb
->fs_generation
, le32_to_cpu(fe
->i_fs_generation
));
241 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
242 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
243 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
245 inode
->i_version
= 1;
246 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
247 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
248 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
249 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
250 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
252 /* Fast symlinks will have i_size but no allocated clusters. */
253 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
256 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
257 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
258 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
259 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
260 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
261 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
262 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
263 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
265 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
267 "ip_blkno %llu != i_blkno %llu!\n",
268 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
269 (unsigned long long)le64_to_cpu(fe
->i_blkno
));
271 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
273 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
274 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
276 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
277 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
278 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
279 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
280 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
281 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
282 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
283 /* we can't actually hit this as read_inode can't
284 * handle superblocks today ;-) */
288 switch (inode
->i_mode
& S_IFMT
) {
290 inode
->i_fop
= &ocfs2_fops
;
291 inode
->i_op
= &ocfs2_file_iops
;
292 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
295 inode
->i_op
= &ocfs2_dir_iops
;
296 inode
->i_fop
= &ocfs2_dops
;
297 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
300 if (ocfs2_inode_is_fast_symlink(inode
))
301 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
303 inode
->i_op
= &ocfs2_symlink_inode_operations
;
304 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
307 inode
->i_op
= &ocfs2_special_file_iops
;
308 init_special_inode(inode
, inode
->i_mode
,
314 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
315 le64_to_cpu(fe
->i_blkno
));
318 * If we ever want to create system files from kernel,
319 * the generation argument to
320 * ocfs2_inode_lock_res_init() will have to change.
322 BUG_ON(le32_to_cpu(fe
->i_flags
) & OCFS2_SYSTEM_FL
);
324 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
325 OCFS2_LOCK_TYPE_META
, 0, inode
);
327 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
328 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
331 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
332 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
335 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_data_lockres
,
336 OCFS2_LOCK_TYPE_DATA
, inode
->i_generation
,
339 ocfs2_set_inode_flags(inode
);
347 static int ocfs2_read_locked_inode(struct inode
*inode
,
348 struct ocfs2_find_inode_args
*args
)
350 struct super_block
*sb
;
351 struct ocfs2_super
*osb
;
352 struct ocfs2_dinode
*fe
;
353 struct buffer_head
*bh
= NULL
;
354 int status
, can_lock
;
357 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
360 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
361 mlog(ML_ERROR
, "bad inode\n");
368 mlog(ML_ERROR
, "bad inode args\n");
369 make_bad_inode(inode
);
374 * To improve performance of cold-cache inode stats, we take
375 * the cluster lock here if possible.
377 * Generally, OCFS2 never trusts the contents of an inode
378 * unless it's holding a cluster lock, so taking it here isn't
379 * a correctness issue as much as it is a performance
382 * There are three times when taking the lock is not a good idea:
384 * 1) During startup, before we have initialized the DLM.
386 * 2) If we are reading certain system files which never get
387 * cluster locks (local alloc, truncate log).
389 * 3) If the process doing the iget() is responsible for
390 * orphan dir recovery. We're holding the orphan dir lock and
391 * can get into a deadlock with another process on another
392 * node in ->delete_inode().
394 * #1 and #2 can be simply solved by never taking the lock
395 * here for system files (which are the only type we read
396 * during mount). It's a heavier approach, but our main
397 * concern is user-accesible files anyway.
399 * #3 works itself out because we'll eventually take the
400 * cluster lock before trusting anything anyway.
402 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
403 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
404 && !ocfs2_mount_local(osb
);
407 * To maintain backwards compatibility with older versions of
408 * ocfs2-tools, we still store the generation value for system
409 * files. The only ones that actually matter to userspace are
410 * the journals, but it's easier and inexpensive to just flag
411 * all system files similarly.
413 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
414 generation
= osb
->fs_generation
;
416 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
417 OCFS2_LOCK_TYPE_META
,
420 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
421 OCFS2_LOCK_TYPE_OPEN
,
425 status
= ocfs2_open_lock(inode
);
427 make_bad_inode(inode
);
431 status
= ocfs2_meta_lock(inode
, NULL
, 0);
433 make_bad_inode(inode
);
439 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
440 status
= ocfs2_try_open_lock(inode
, 0);
442 make_bad_inode(inode
);
447 status
= ocfs2_read_block(osb
, args
->fi_blkno
, &bh
, 0,
448 can_lock
? inode
: NULL
);
455 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
456 if (!OCFS2_IS_VALID_DINODE(fe
)) {
457 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
458 (unsigned long long)args
->fi_blkno
, 7,
464 * This is a code bug. Right now the caller needs to
465 * understand whether it is asking for a system file inode or
466 * not so the proper lock names can be built.
468 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
469 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
470 "Inode %llu: system file state is ambigous\n",
471 (unsigned long long)args
->fi_blkno
);
473 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
474 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
475 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
477 if (ocfs2_populate_inode(inode
, fe
, 0) < 0)
480 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
486 ocfs2_meta_unlock(inode
, 0);
489 make_bad_inode(inode
);
498 void ocfs2_sync_blockdev(struct super_block
*sb
)
500 sync_blockdev(sb
->s_bdev
);
503 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
505 struct buffer_head
*fe_bh
)
508 struct ocfs2_truncate_context
*tc
= NULL
;
509 struct ocfs2_dinode
*fe
;
510 handle_t
*handle
= NULL
;
514 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
517 * This check will also skip truncate of inodes with inline
518 * data and fast symlinks.
520 if (fe
->i_clusters
) {
521 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
522 if (IS_ERR(handle
)) {
523 status
= PTR_ERR(handle
);
528 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
529 OCFS2_JOURNAL_ACCESS_WRITE
);
535 i_size_write(inode
, 0);
537 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
543 ocfs2_commit_trans(osb
, handle
);
546 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
552 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
561 ocfs2_commit_trans(osb
, handle
);
566 static int ocfs2_remove_inode(struct inode
*inode
,
567 struct buffer_head
*di_bh
,
568 struct inode
*orphan_dir_inode
,
569 struct buffer_head
*orphan_dir_bh
)
572 struct inode
*inode_alloc_inode
= NULL
;
573 struct buffer_head
*inode_alloc_bh
= NULL
;
575 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
576 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
579 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
580 le16_to_cpu(di
->i_suballoc_slot
));
581 if (!inode_alloc_inode
) {
587 mutex_lock(&inode_alloc_inode
->i_mutex
);
588 status
= ocfs2_meta_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
590 mutex_unlock(&inode_alloc_inode
->i_mutex
);
596 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
597 if (IS_ERR(handle
)) {
598 status
= PTR_ERR(handle
);
603 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
610 /* set the inodes dtime */
611 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
612 OCFS2_JOURNAL_ACCESS_WRITE
);
618 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
619 le32_and_cpu(&di
->i_flags
, ~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
621 status
= ocfs2_journal_dirty(handle
, di_bh
);
627 ocfs2_remove_from_cache(inode
, di_bh
);
629 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
635 ocfs2_commit_trans(osb
, handle
);
637 ocfs2_meta_unlock(inode_alloc_inode
, 1);
638 mutex_unlock(&inode_alloc_inode
->i_mutex
);
639 brelse(inode_alloc_bh
);
641 iput(inode_alloc_inode
);
647 * Serialize with orphan dir recovery. If the process doing
648 * recovery on this orphan dir does an iget() with the dir
649 * i_mutex held, we'll deadlock here. Instead we detect this
650 * and exit early - recovery will wipe this inode for us.
652 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
657 spin_lock(&osb
->osb_lock
);
658 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
659 mlog(0, "Recovery is happening on orphan dir %d, will skip "
660 "this inode\n", slot
);
664 /* This signals to the orphan recovery process that it should
665 * wait for us to handle the wipe. */
666 osb
->osb_orphan_wipes
[slot
]++;
668 spin_unlock(&osb
->osb_lock
);
672 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
675 spin_lock(&osb
->osb_lock
);
676 osb
->osb_orphan_wipes
[slot
]--;
677 spin_unlock(&osb
->osb_lock
);
679 wake_up(&osb
->osb_wipe_event
);
682 static int ocfs2_wipe_inode(struct inode
*inode
,
683 struct buffer_head
*di_bh
)
685 int status
, orphaned_slot
;
686 struct inode
*orphan_dir_inode
= NULL
;
687 struct buffer_head
*orphan_dir_bh
= NULL
;
688 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
689 struct ocfs2_dinode
*di
;
691 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
692 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
694 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
698 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
699 ORPHAN_DIR_SYSTEM_INODE
,
701 if (!orphan_dir_inode
) {
707 /* Lock the orphan dir. The lock will be held for the entire
708 * delete_inode operation. We do this now to avoid races with
709 * recovery completion on other nodes. */
710 mutex_lock(&orphan_dir_inode
->i_mutex
);
711 status
= ocfs2_meta_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
713 mutex_unlock(&orphan_dir_inode
->i_mutex
);
719 /* we do this while holding the orphan dir lock because we
720 * don't want recovery being run from another node to try an
721 * inode delete underneath us -- this will result in two nodes
722 * truncating the same file! */
723 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
726 goto bail_unlock_dir
;
729 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
735 ocfs2_meta_unlock(orphan_dir_inode
, 1);
736 mutex_unlock(&orphan_dir_inode
->i_mutex
);
737 brelse(orphan_dir_bh
);
739 iput(orphan_dir_inode
);
740 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
745 /* There is a series of simple checks that should be done before a
746 * trylock is even considered. Encapsulate those in this function. */
747 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
750 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
751 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
753 /* We shouldn't be getting here for the root directory
755 if (inode
== osb
->root_inode
) {
756 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
760 /* If we're coming from downconvert_thread we can't go into our own
761 * voting [hello, deadlock city!], so unforuntately we just
762 * have to skip deleting this guy. That's OK though because
763 * the node who's doing the actual deleting should handle it
765 if (current
== osb
->dc_task
) {
766 mlog(0, "Skipping delete of %lu because we're currently "
767 "in downconvert\n", inode
->i_ino
);
771 spin_lock(&oi
->ip_lock
);
772 /* OCFS2 *never* deletes system files. This should technically
773 * never get here as system file inodes should always have a
774 * positive link count. */
775 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
776 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
777 (unsigned long long)oi
->ip_blkno
);
781 /* If we have allowd wipe of this inode for another node, it
782 * will be marked here so we can safely skip it. Recovery will
783 * cleanup any inodes we might inadvertantly skip here. */
784 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
785 mlog(0, "Skipping delete of %lu because another node "
786 "has done this for us.\n", inode
->i_ino
);
792 spin_unlock(&oi
->ip_lock
);
797 /* Query the cluster to determine whether we should wipe an inode from
800 * Requires the inode to have the cluster lock. */
801 static int ocfs2_query_inode_wipe(struct inode
*inode
,
802 struct buffer_head
*di_bh
,
806 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
807 struct ocfs2_dinode
*di
;
811 /* While we were waiting for the cluster lock in
812 * ocfs2_delete_inode, another node might have asked to delete
813 * the inode. Recheck our flags to catch this. */
814 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
815 mlog(0, "Skipping delete of %llu because flags changed\n",
816 (unsigned long long)oi
->ip_blkno
);
820 /* Now that we have an up to date inode, we can double check
822 if (inode
->i_nlink
) {
823 mlog(0, "Skipping delete of %llu because nlink = %u\n",
824 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
828 /* Do some basic inode verification... */
829 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
830 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
831 /* for lack of a better error? */
834 "Inode %llu (on-disk %llu) not orphaned! "
835 "Disk flags 0x%x, inode flags 0x%x\n",
836 (unsigned long long)oi
->ip_blkno
,
837 (unsigned long long)le64_to_cpu(di
->i_blkno
),
838 le32_to_cpu(di
->i_flags
), oi
->ip_flags
);
842 /* has someone already deleted us?! baaad... */
850 * This is how ocfs2 determines whether an inode is still live
851 * within the cluster. Every node takes a shared read lock on
852 * the inode open lock in ocfs2_read_locked_inode(). When we
853 * get to ->delete_inode(), each node tries to convert it's
854 * lock to an exclusive. Trylocks are serialized by the inode
855 * meta data lock. If the upconvert suceeds, we know the inode
856 * is no longer live and can be deleted.
858 * Though we call this with the meta data lock held, the
859 * trylock keeps us from ABBA deadlock.
861 status
= ocfs2_try_open_lock(inode
, 1);
862 if (status
== -EAGAIN
) {
864 mlog(0, "Skipping delete of %llu because it is in use on "
865 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
874 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
875 (unsigned long long)oi
->ip_blkno
,
876 le16_to_cpu(di
->i_orphaned_slot
));
882 /* Support function for ocfs2_delete_inode. Will help us keep the
883 * inode data in a consistent state for clear_inode. Always truncates
884 * pages, optionally sync's them first. */
885 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
888 mlog(0, "Cleanup inode %llu, sync = %d\n",
889 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
891 write_inode_now(inode
, 1);
892 truncate_inode_pages(&inode
->i_data
, 0);
895 void ocfs2_delete_inode(struct inode
*inode
)
898 sigset_t blocked
, oldset
;
899 struct buffer_head
*di_bh
= NULL
;
901 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
903 if (is_bad_inode(inode
)) {
904 mlog(0, "Skipping delete of bad inode\n");
908 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
909 /* It's probably not necessary to truncate_inode_pages
910 * here but we do it for safety anyway (it will most
911 * likely be a no-op anyway) */
912 ocfs2_cleanup_delete_inode(inode
, 0);
916 /* We want to block signals in delete_inode as the lock and
917 * messaging paths may return us -ERESTARTSYS. Which would
918 * cause us to exit early, resulting in inodes being orphaned
920 sigfillset(&blocked
);
921 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
924 ocfs2_cleanup_delete_inode(inode
, 1);
928 /* Lock down the inode. This gives us an up to date view of
929 * it's metadata (for verification), and allows us to
930 * serialize delete_inode on multiple nodes.
932 * Even though we might be doing a truncate, we don't take the
933 * allocation lock here as it won't be needed - nobody will
934 * have the file open.
936 status
= ocfs2_meta_lock(inode
, &di_bh
, 1);
938 if (status
!= -ENOENT
)
940 ocfs2_cleanup_delete_inode(inode
, 0);
944 /* Query the cluster. This will be the final decision made
945 * before we go ahead and wipe the inode. */
946 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
947 if (!wipe
|| status
< 0) {
948 /* Error and remote inode busy both mean we won't be
949 * removing the inode, so they take almost the same
954 /* Someone in the cluster has disallowed a wipe of
955 * this inode, or it was never completely
956 * orphaned. Write out the pages and exit now. */
957 ocfs2_cleanup_delete_inode(inode
, 1);
958 goto bail_unlock_inode
;
961 ocfs2_cleanup_delete_inode(inode
, 0);
963 status
= ocfs2_wipe_inode(inode
, di_bh
);
965 if (status
!= -EDEADLK
)
967 goto bail_unlock_inode
;
971 * Mark the inode as successfully deleted.
973 * This is important for ocfs2_clear_inode() as it will check
974 * this flag and skip any checkpointing work
976 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
977 * the LVB for other nodes.
979 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
982 ocfs2_meta_unlock(inode
, 1);
985 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
993 void ocfs2_clear_inode(struct inode
*inode
)
996 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1003 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1004 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
1006 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
1007 "Inode=%lu\n", inode
->i_ino
);
1009 /* To preven remote deletes we hold open lock before, now it
1010 * is time to unlock PR and EX open locks. */
1011 ocfs2_open_unlock(inode
);
1013 /* Do these before all the other work so that we don't bounce
1014 * the downconvert thread while waiting to destroy the locks. */
1015 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
1016 ocfs2_mark_lockres_freeing(&oi
->ip_meta_lockres
);
1017 ocfs2_mark_lockres_freeing(&oi
->ip_data_lockres
);
1018 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
1020 /* We very well may get a clear_inode before all an inodes
1021 * metadata has hit disk. Of course, we can't drop any cluster
1022 * locks until the journal has finished with it. The only
1023 * exception here are successfully wiped inodes - their
1024 * metadata can now be considered to be part of the system
1025 * inodes from which it came. */
1026 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1027 ocfs2_checkpoint_inode(inode
);
1029 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1030 "Clear inode of %llu, inode has io markers\n",
1031 (unsigned long long)oi
->ip_blkno
);
1033 ocfs2_extent_map_trunc(inode
, 0);
1035 status
= ocfs2_drop_inode_locks(inode
);
1039 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1040 ocfs2_lock_res_free(&oi
->ip_meta_lockres
);
1041 ocfs2_lock_res_free(&oi
->ip_data_lockres
);
1042 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1044 ocfs2_metadata_cache_purge(inode
);
1046 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1047 "Clear inode of %llu, inode has %u cache items\n",
1048 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1050 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1051 "Clear inode of %llu, inode has a bad flag\n",
1052 (unsigned long long)oi
->ip_blkno
);
1054 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1055 "Clear inode of %llu, inode is locked\n",
1056 (unsigned long long)oi
->ip_blkno
);
1058 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1059 "Clear inode of %llu, io_mutex is locked\n",
1060 (unsigned long long)oi
->ip_blkno
);
1061 mutex_unlock(&oi
->ip_io_mutex
);
1064 * down_trylock() returns 0, down_write_trylock() returns 1
1067 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1068 "Clear inode of %llu, alloc_sem is locked\n",
1069 (unsigned long long)oi
->ip_blkno
);
1070 up_write(&oi
->ip_alloc_sem
);
1072 mlog_bug_on_msg(oi
->ip_open_count
,
1073 "Clear inode of %llu has open count %d\n",
1074 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1076 /* Clear all other flags. */
1077 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1078 oi
->ip_created_trans
= 0;
1079 oi
->ip_last_trans
= 0;
1080 oi
->ip_dir_start_lookup
= 0;
1081 oi
->ip_blkno
= 0ULL;
1087 /* Called under inode_lock, with no more references on the
1088 * struct inode, so it's safe here to check the flags field
1089 * and to manipulate i_nlink without any other locks. */
1090 void ocfs2_drop_inode(struct inode
*inode
)
1092 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1096 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1097 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1099 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1100 generic_delete_inode(inode
);
1102 generic_drop_inode(inode
);
1108 * TODO: this should probably be merged into ocfs2_get_block
1110 * However, you now need to pay attention to the cont_prepare_write()
1111 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1112 * expects never to extend).
1114 struct buffer_head
*ocfs2_bread(struct inode
*inode
,
1115 int block
, int *err
, int reada
)
1117 struct buffer_head
*bh
= NULL
;
1120 int readflags
= OCFS2_BH_CACHED
;
1123 readflags
|= OCFS2_BH_READAHEAD
;
1125 if (((u64
)block
<< inode
->i_sb
->s_blocksize_bits
) >=
1126 i_size_read(inode
)) {
1131 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
1132 tmperr
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
1134 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
1140 tmperr
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), p_blkno
, &bh
,
1160 * This is called from our getattr.
1162 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1164 struct inode
*inode
= dentry
->d_inode
;
1167 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1168 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1171 mlog(0, "eep, no inode!\n");
1176 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1177 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1178 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1179 mlog(0, "inode deleted!\n");
1183 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1185 /* Let ocfs2_meta_lock do the work of updating our struct
1187 status
= ocfs2_meta_lock(inode
, NULL
, 0);
1189 if (status
!= -ENOENT
)
1193 ocfs2_meta_unlock(inode
, 0);
1201 * Updates a disk inode from a
1203 * Only takes ip_lock.
1205 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1206 struct inode
*inode
,
1207 struct buffer_head
*bh
)
1210 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1212 mlog_entry("(inode %llu)\n",
1213 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1215 status
= ocfs2_journal_access(handle
, inode
, bh
,
1216 OCFS2_JOURNAL_ACCESS_WRITE
);
1222 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1223 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1224 ocfs2_get_inode_flags(OCFS2_I(inode
));
1225 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1226 fe
->i_dyn_features
= cpu_to_le16(OCFS2_I(inode
)->ip_dyn_features
);
1227 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1229 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1230 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1231 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1232 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1233 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1234 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1235 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1236 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1237 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1238 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1239 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1241 status
= ocfs2_journal_dirty(handle
, bh
);
1254 * Updates a struct inode from a disk inode.
1255 * does no i/o, only takes ip_lock.
1257 void ocfs2_refresh_inode(struct inode
*inode
,
1258 struct ocfs2_dinode
*fe
)
1260 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1262 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1263 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1264 OCFS2_I(inode
)->ip_dyn_features
= le16_to_cpu(fe
->i_dyn_features
);
1265 ocfs2_set_inode_flags(inode
);
1266 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1267 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1268 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1269 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1270 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1271 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1272 inode
->i_blocks
= 0;
1274 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1275 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1276 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1277 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1278 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1279 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1280 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1282 spin_unlock(&OCFS2_I(inode
)->ip_lock
);