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 inode
->i_version
= 1;
225 inode
->i_generation
= le32_to_cpu(fe
->i_generation
);
226 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
227 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
228 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
229 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
231 /* Fast symlinks will have i_size but no allocated clusters. */
232 if (S_ISLNK(inode
->i_mode
) && !fe
->i_clusters
)
236 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe
->i_size
));
237 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
238 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
239 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
240 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
241 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
242 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
243 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
245 if (OCFS2_I(inode
)->ip_blkno
!= le64_to_cpu(fe
->i_blkno
))
247 "ip_blkno %llu != i_blkno %llu!\n",
248 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
249 (unsigned long long)fe
->i_blkno
);
251 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
252 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
254 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
256 if (fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
))
257 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SYSTEM_FILE
;
259 if (fe
->i_flags
& cpu_to_le32(OCFS2_LOCAL_ALLOC_FL
)) {
260 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
261 mlog(0, "local alloc inode: i_ino=%lu\n", inode
->i_ino
);
262 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_BITMAP_FL
)) {
263 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_BITMAP
;
264 } else if (fe
->i_flags
& cpu_to_le32(OCFS2_SUPER_BLOCK_FL
)) {
265 mlog(0, "superblock inode: i_ino=%lu\n", inode
->i_ino
);
266 /* we can't actually hit this as read_inode can't
267 * handle superblocks today ;-) */
271 switch (inode
->i_mode
& S_IFMT
) {
273 inode
->i_fop
= &ocfs2_fops
;
274 inode
->i_op
= &ocfs2_file_iops
;
275 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
278 inode
->i_op
= &ocfs2_dir_iops
;
279 inode
->i_fop
= &ocfs2_dops
;
280 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
283 if (ocfs2_inode_is_fast_symlink(inode
))
284 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
286 inode
->i_op
= &ocfs2_symlink_inode_operations
;
287 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
290 inode
->i_op
= &ocfs2_special_file_iops
;
291 init_special_inode(inode
, inode
->i_mode
,
297 inode
->i_ino
= ino_from_blkno(inode
->i_sb
,
298 le64_to_cpu(fe
->i_blkno
));
301 * If we ever want to create system files from kernel,
302 * the generation argument to
303 * ocfs2_inode_lock_res_init() will have to change.
305 BUG_ON(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
));
307 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
308 OCFS2_LOCK_TYPE_META
, 0, inode
);
310 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
311 OCFS2_LOCK_TYPE_OPEN
, 0, inode
);
314 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_rw_lockres
,
315 OCFS2_LOCK_TYPE_RW
, inode
->i_generation
,
318 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_data_lockres
,
319 OCFS2_LOCK_TYPE_DATA
, inode
->i_generation
,
322 ocfs2_set_inode_flags(inode
);
330 static int ocfs2_read_locked_inode(struct inode
*inode
,
331 struct ocfs2_find_inode_args
*args
)
333 struct super_block
*sb
;
334 struct ocfs2_super
*osb
;
335 struct ocfs2_dinode
*fe
;
336 struct buffer_head
*bh
= NULL
;
337 int status
, can_lock
;
340 mlog_entry("(0x%p, 0x%p)\n", inode
, args
);
343 if (inode
== NULL
|| inode
->i_sb
== NULL
) {
344 mlog(ML_ERROR
, "bad inode\n");
351 mlog(ML_ERROR
, "bad inode args\n");
352 make_bad_inode(inode
);
357 * To improve performance of cold-cache inode stats, we take
358 * the cluster lock here if possible.
360 * Generally, OCFS2 never trusts the contents of an inode
361 * unless it's holding a cluster lock, so taking it here isn't
362 * a correctness issue as much as it is a performance
365 * There are three times when taking the lock is not a good idea:
367 * 1) During startup, before we have initialized the DLM.
369 * 2) If we are reading certain system files which never get
370 * cluster locks (local alloc, truncate log).
372 * 3) If the process doing the iget() is responsible for
373 * orphan dir recovery. We're holding the orphan dir lock and
374 * can get into a deadlock with another process on another
375 * node in ->delete_inode().
377 * #1 and #2 can be simply solved by never taking the lock
378 * here for system files (which are the only type we read
379 * during mount). It's a heavier approach, but our main
380 * concern is user-accesible files anyway.
382 * #3 works itself out because we'll eventually take the
383 * cluster lock before trusting anything anyway.
385 can_lock
= !(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
386 && !(args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
)
387 && !ocfs2_mount_local(osb
);
390 * To maintain backwards compatibility with older versions of
391 * ocfs2-tools, we still store the generation value for system
392 * files. The only ones that actually matter to userspace are
393 * the journals, but it's easier and inexpensive to just flag
394 * all system files similarly.
396 if (args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
)
397 generation
= osb
->fs_generation
;
399 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_meta_lockres
,
400 OCFS2_LOCK_TYPE_META
,
403 ocfs2_inode_lock_res_init(&OCFS2_I(inode
)->ip_open_lockres
,
404 OCFS2_LOCK_TYPE_OPEN
,
408 status
= ocfs2_open_lock(inode
);
410 make_bad_inode(inode
);
414 status
= ocfs2_meta_lock(inode
, NULL
, 0);
416 make_bad_inode(inode
);
422 if (args
->fi_flags
& OCFS2_FI_FLAG_ORPHAN_RECOVERY
) {
423 status
= ocfs2_try_open_lock(inode
, 0);
425 make_bad_inode(inode
);
430 status
= ocfs2_read_block(osb
, args
->fi_blkno
, &bh
, 0,
431 can_lock
? inode
: NULL
);
438 fe
= (struct ocfs2_dinode
*) bh
->b_data
;
439 if (!OCFS2_IS_VALID_DINODE(fe
)) {
440 mlog(ML_ERROR
, "Invalid dinode #%llu: signature = %.*s\n",
441 (unsigned long long)fe
->i_blkno
, 7, fe
->i_signature
);
446 * This is a code bug. Right now the caller needs to
447 * understand whether it is asking for a system file inode or
448 * not so the proper lock names can be built.
450 mlog_bug_on_msg(!!(fe
->i_flags
& cpu_to_le32(OCFS2_SYSTEM_FL
)) !=
451 !!(args
->fi_flags
& OCFS2_FI_FLAG_SYSFILE
),
452 "Inode %llu: system file state is ambigous\n",
453 (unsigned long long)args
->fi_blkno
);
455 if (S_ISCHR(le16_to_cpu(fe
->i_mode
)) ||
456 S_ISBLK(le16_to_cpu(fe
->i_mode
)))
457 inode
->i_rdev
= huge_decode_dev(le64_to_cpu(fe
->id1
.dev1
.i_rdev
));
459 if (ocfs2_populate_inode(inode
, fe
, 0) < 0)
462 BUG_ON(args
->fi_blkno
!= le64_to_cpu(fe
->i_blkno
));
468 ocfs2_meta_unlock(inode
, 0);
471 make_bad_inode(inode
);
480 void ocfs2_sync_blockdev(struct super_block
*sb
)
482 sync_blockdev(sb
->s_bdev
);
485 static int ocfs2_truncate_for_delete(struct ocfs2_super
*osb
,
487 struct buffer_head
*fe_bh
)
490 struct ocfs2_truncate_context
*tc
= NULL
;
491 struct ocfs2_dinode
*fe
;
492 handle_t
*handle
= NULL
;
496 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
498 if (fe
->i_clusters
) {
499 handle
= ocfs2_start_trans(osb
, OCFS2_INODE_UPDATE_CREDITS
);
500 if (IS_ERR(handle
)) {
501 status
= PTR_ERR(handle
);
506 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
507 OCFS2_JOURNAL_ACCESS_WRITE
);
513 i_size_write(inode
, 0);
515 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
521 ocfs2_commit_trans(osb
, handle
);
524 status
= ocfs2_prepare_truncate(osb
, inode
, fe_bh
, &tc
);
530 status
= ocfs2_commit_truncate(osb
, inode
, fe_bh
, tc
);
539 ocfs2_commit_trans(osb
, handle
);
544 static int ocfs2_remove_inode(struct inode
*inode
,
545 struct buffer_head
*di_bh
,
546 struct inode
*orphan_dir_inode
,
547 struct buffer_head
*orphan_dir_bh
)
550 struct inode
*inode_alloc_inode
= NULL
;
551 struct buffer_head
*inode_alloc_bh
= NULL
;
553 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
554 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
557 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
558 le16_to_cpu(di
->i_suballoc_slot
));
559 if (!inode_alloc_inode
) {
565 mutex_lock(&inode_alloc_inode
->i_mutex
);
566 status
= ocfs2_meta_lock(inode_alloc_inode
, &inode_alloc_bh
, 1);
568 mutex_unlock(&inode_alloc_inode
->i_mutex
);
574 handle
= ocfs2_start_trans(osb
, OCFS2_DELETE_INODE_CREDITS
);
575 if (IS_ERR(handle
)) {
576 status
= PTR_ERR(handle
);
581 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
588 /* set the inodes dtime */
589 status
= ocfs2_journal_access(handle
, inode
, di_bh
,
590 OCFS2_JOURNAL_ACCESS_WRITE
);
596 di
->i_dtime
= cpu_to_le64(CURRENT_TIME
.tv_sec
);
597 le32_and_cpu(&di
->i_flags
, ~(OCFS2_VALID_FL
| OCFS2_ORPHANED_FL
));
599 status
= ocfs2_journal_dirty(handle
, di_bh
);
605 ocfs2_remove_from_cache(inode
, di_bh
);
607 status
= ocfs2_free_dinode(handle
, inode_alloc_inode
,
613 ocfs2_commit_trans(osb
, handle
);
615 ocfs2_meta_unlock(inode_alloc_inode
, 1);
616 mutex_unlock(&inode_alloc_inode
->i_mutex
);
617 brelse(inode_alloc_bh
);
619 iput(inode_alloc_inode
);
625 * Serialize with orphan dir recovery. If the process doing
626 * recovery on this orphan dir does an iget() with the dir
627 * i_mutex held, we'll deadlock here. Instead we detect this
628 * and exit early - recovery will wipe this inode for us.
630 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super
*osb
,
635 spin_lock(&osb
->osb_lock
);
636 if (ocfs2_node_map_test_bit(osb
, &osb
->osb_recovering_orphan_dirs
, slot
)) {
637 mlog(0, "Recovery is happening on orphan dir %d, will skip "
638 "this inode\n", slot
);
642 /* This signals to the orphan recovery process that it should
643 * wait for us to handle the wipe. */
644 osb
->osb_orphan_wipes
[slot
]++;
646 spin_unlock(&osb
->osb_lock
);
650 static void ocfs2_signal_wipe_completion(struct ocfs2_super
*osb
,
653 spin_lock(&osb
->osb_lock
);
654 osb
->osb_orphan_wipes
[slot
]--;
655 spin_unlock(&osb
->osb_lock
);
657 wake_up(&osb
->osb_wipe_event
);
660 static int ocfs2_wipe_inode(struct inode
*inode
,
661 struct buffer_head
*di_bh
)
663 int status
, orphaned_slot
;
664 struct inode
*orphan_dir_inode
= NULL
;
665 struct buffer_head
*orphan_dir_bh
= NULL
;
666 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
667 struct ocfs2_dinode
*di
;
669 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
670 orphaned_slot
= le16_to_cpu(di
->i_orphaned_slot
);
672 status
= ocfs2_check_orphan_recovery_state(osb
, orphaned_slot
);
676 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
677 ORPHAN_DIR_SYSTEM_INODE
,
679 if (!orphan_dir_inode
) {
685 /* Lock the orphan dir. The lock will be held for the entire
686 * delete_inode operation. We do this now to avoid races with
687 * recovery completion on other nodes. */
688 mutex_lock(&orphan_dir_inode
->i_mutex
);
689 status
= ocfs2_meta_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
691 mutex_unlock(&orphan_dir_inode
->i_mutex
);
697 /* we do this while holding the orphan dir lock because we
698 * don't want recovery being run from another node to vote for
699 * an inode delete on us -- this will result in two nodes
700 * truncating the same file! */
701 status
= ocfs2_truncate_for_delete(osb
, inode
, di_bh
);
704 goto bail_unlock_dir
;
707 status
= ocfs2_remove_inode(inode
, di_bh
, orphan_dir_inode
,
713 ocfs2_meta_unlock(orphan_dir_inode
, 1);
714 mutex_unlock(&orphan_dir_inode
->i_mutex
);
715 brelse(orphan_dir_bh
);
717 iput(orphan_dir_inode
);
718 ocfs2_signal_wipe_completion(osb
, orphaned_slot
);
723 /* There is a series of simple checks that should be done before a
724 * vote is even considered. Encapsulate those in this function. */
725 static int ocfs2_inode_is_valid_to_delete(struct inode
*inode
)
728 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
729 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
731 /* We shouldn't be getting here for the root directory
733 if (inode
== osb
->root_inode
) {
734 mlog(ML_ERROR
, "Skipping delete of root inode.\n");
738 /* If we're coming from process_vote we can't go into our own
739 * voting [hello, deadlock city!], so unforuntately we just
740 * have to skip deleting this guy. That's OK though because
741 * the node who's doing the actual deleting should handle it
743 if (current
== osb
->vote_task
) {
744 mlog(0, "Skipping delete of %lu because we're currently "
745 "in process_vote\n", inode
->i_ino
);
749 spin_lock(&oi
->ip_lock
);
750 /* OCFS2 *never* deletes system files. This should technically
751 * never get here as system file inodes should always have a
752 * positive link count. */
753 if (oi
->ip_flags
& OCFS2_INODE_SYSTEM_FILE
) {
754 mlog(ML_ERROR
, "Skipping delete of system file %llu\n",
755 (unsigned long long)oi
->ip_blkno
);
759 /* If we have voted "yes" on the wipe of this inode for
760 * another node, it will be marked here so we can safely skip
761 * it. Recovery will cleanup any inodes we might inadvertantly
763 if (oi
->ip_flags
& OCFS2_INODE_SKIP_DELETE
) {
764 mlog(0, "Skipping delete of %lu because another node "
765 "has done this for us.\n", inode
->i_ino
);
771 spin_unlock(&oi
->ip_lock
);
776 /* Query the cluster to determine whether we should wipe an inode from
779 * Requires the inode to have the cluster lock. */
780 static int ocfs2_query_inode_wipe(struct inode
*inode
,
781 struct buffer_head
*di_bh
,
785 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
786 struct ocfs2_dinode
*di
;
790 /* While we were waiting for the cluster lock in
791 * ocfs2_delete_inode, another node might have asked to delete
792 * the inode. Recheck our flags to catch this. */
793 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
794 mlog(0, "Skipping delete of %llu because flags changed\n",
795 (unsigned long long)oi
->ip_blkno
);
799 /* Now that we have an up to date inode, we can double check
801 if (inode
->i_nlink
) {
802 mlog(0, "Skipping delete of %llu because nlink = %u\n",
803 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
);
807 /* Do some basic inode verification... */
808 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
809 if (!(di
->i_flags
& cpu_to_le32(OCFS2_ORPHANED_FL
))) {
810 /* for lack of a better error? */
813 "Inode %llu (on-disk %llu) not orphaned! "
814 "Disk flags 0x%x, inode flags 0x%x\n",
815 (unsigned long long)oi
->ip_blkno
,
816 (unsigned long long)di
->i_blkno
, di
->i_flags
,
821 /* has someone already deleted us?! baaad... */
829 * This is how ocfs2 determines whether an inode is still live
830 * within the cluster. Every node takes a shared read lock on
831 * the inode open lock in ocfs2_read_locked_inode(). When we
832 * get to ->delete_inode(), each node tries to convert it's
833 * lock to an exclusive. Trylocks are serialized by the inode
834 * meta data lock. If the upconvert suceeds, we know the inode
835 * is no longer live and can be deleted.
837 * Though we call this with the meta data lock held, the
838 * trylock keeps us from ABBA deadlock.
840 status
= ocfs2_try_open_lock(inode
, 1);
841 if (status
== -EAGAIN
) {
843 mlog(0, "Skipping delete of %llu because it is in use on"
844 "other nodes\n", (unsigned long long)oi
->ip_blkno
);
853 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
854 (unsigned long long)oi
->ip_blkno
,
855 le16_to_cpu(di
->i_orphaned_slot
));
861 /* Support function for ocfs2_delete_inode. Will help us keep the
862 * inode data in a consistent state for clear_inode. Always truncates
863 * pages, optionally sync's them first. */
864 static void ocfs2_cleanup_delete_inode(struct inode
*inode
,
867 mlog(0, "Cleanup inode %llu, sync = %d\n",
868 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, sync_data
);
870 write_inode_now(inode
, 1);
871 truncate_inode_pages(&inode
->i_data
, 0);
874 void ocfs2_delete_inode(struct inode
*inode
)
877 sigset_t blocked
, oldset
;
878 struct buffer_head
*di_bh
= NULL
;
880 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
882 if (is_bad_inode(inode
)) {
883 mlog(0, "Skipping delete of bad inode\n");
887 if (!ocfs2_inode_is_valid_to_delete(inode
)) {
888 /* It's probably not necessary to truncate_inode_pages
889 * here but we do it for safety anyway (it will most
890 * likely be a no-op anyway) */
891 ocfs2_cleanup_delete_inode(inode
, 0);
895 /* We want to block signals in delete_inode as the lock and
896 * messaging paths may return us -ERESTARTSYS. Which would
897 * cause us to exit early, resulting in inodes being orphaned
899 sigfillset(&blocked
);
900 status
= sigprocmask(SIG_BLOCK
, &blocked
, &oldset
);
903 ocfs2_cleanup_delete_inode(inode
, 1);
907 /* Lock down the inode. This gives us an up to date view of
908 * it's metadata (for verification), and allows us to
909 * serialize delete_inode votes.
911 * Even though we might be doing a truncate, we don't take the
912 * allocation lock here as it won't be needed - nobody will
913 * have the file open.
915 status
= ocfs2_meta_lock(inode
, &di_bh
, 1);
917 if (status
!= -ENOENT
)
919 ocfs2_cleanup_delete_inode(inode
, 0);
923 /* Query the cluster. This will be the final decision made
924 * before we go ahead and wipe the inode. */
925 status
= ocfs2_query_inode_wipe(inode
, di_bh
, &wipe
);
926 if (!wipe
|| status
< 0) {
927 /* Error and inode busy vote both mean we won't be
928 * removing the inode, so they take almost the same
933 /* Someone in the cluster has voted to not wipe this
934 * inode, or it was never completely orphaned. Write
935 * out the pages and exit now. */
936 ocfs2_cleanup_delete_inode(inode
, 1);
937 goto bail_unlock_inode
;
940 ocfs2_cleanup_delete_inode(inode
, 0);
942 status
= ocfs2_wipe_inode(inode
, di_bh
);
944 if (status
!= -EDEADLK
)
946 goto bail_unlock_inode
;
950 * Mark the inode as successfully deleted.
952 * This is important for ocfs2_clear_inode() as it will check
953 * this flag and skip any checkpointing work
955 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
956 * the LVB for other nodes.
958 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_DELETED
;
961 ocfs2_meta_unlock(inode
, 1);
964 status
= sigprocmask(SIG_SETMASK
, &oldset
, NULL
);
972 void ocfs2_clear_inode(struct inode
*inode
)
975 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
982 mlog(0, "Clearing inode: %llu, nlink = %u\n",
983 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, inode
->i_nlink
);
985 mlog_bug_on_msg(OCFS2_SB(inode
->i_sb
) == NULL
,
986 "Inode=%lu\n", inode
->i_ino
);
988 /* For remove delete_inode vote, we hold open lock before,
989 * now it is time to unlock PR and EX open locks. */
990 ocfs2_open_unlock(inode
);
992 /* Do these before all the other work so that we don't bounce
993 * the vote thread while waiting to destroy the locks. */
994 ocfs2_mark_lockres_freeing(&oi
->ip_rw_lockres
);
995 ocfs2_mark_lockres_freeing(&oi
->ip_meta_lockres
);
996 ocfs2_mark_lockres_freeing(&oi
->ip_data_lockres
);
997 ocfs2_mark_lockres_freeing(&oi
->ip_open_lockres
);
999 /* We very well may get a clear_inode before all an inodes
1000 * metadata has hit disk. Of course, we can't drop any cluster
1001 * locks until the journal has finished with it. The only
1002 * exception here are successfully wiped inodes - their
1003 * metadata can now be considered to be part of the system
1004 * inodes from which it came. */
1005 if (!(OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
))
1006 ocfs2_checkpoint_inode(inode
);
1008 mlog_bug_on_msg(!list_empty(&oi
->ip_io_markers
),
1009 "Clear inode of %llu, inode has io markers\n",
1010 (unsigned long long)oi
->ip_blkno
);
1012 status
= ocfs2_drop_inode_locks(inode
);
1016 ocfs2_lock_res_free(&oi
->ip_rw_lockres
);
1017 ocfs2_lock_res_free(&oi
->ip_meta_lockres
);
1018 ocfs2_lock_res_free(&oi
->ip_data_lockres
);
1019 ocfs2_lock_res_free(&oi
->ip_open_lockres
);
1021 ocfs2_metadata_cache_purge(inode
);
1023 mlog_bug_on_msg(oi
->ip_metadata_cache
.ci_num_cached
,
1024 "Clear inode of %llu, inode has %u cache items\n",
1025 (unsigned long long)oi
->ip_blkno
, oi
->ip_metadata_cache
.ci_num_cached
);
1027 mlog_bug_on_msg(!(oi
->ip_flags
& OCFS2_INODE_CACHE_INLINE
),
1028 "Clear inode of %llu, inode has a bad flag\n",
1029 (unsigned long long)oi
->ip_blkno
);
1031 mlog_bug_on_msg(spin_is_locked(&oi
->ip_lock
),
1032 "Clear inode of %llu, inode is locked\n",
1033 (unsigned long long)oi
->ip_blkno
);
1035 mlog_bug_on_msg(!mutex_trylock(&oi
->ip_io_mutex
),
1036 "Clear inode of %llu, io_mutex is locked\n",
1037 (unsigned long long)oi
->ip_blkno
);
1038 mutex_unlock(&oi
->ip_io_mutex
);
1041 * down_trylock() returns 0, down_write_trylock() returns 1
1044 mlog_bug_on_msg(!down_write_trylock(&oi
->ip_alloc_sem
),
1045 "Clear inode of %llu, alloc_sem is locked\n",
1046 (unsigned long long)oi
->ip_blkno
);
1047 up_write(&oi
->ip_alloc_sem
);
1049 mlog_bug_on_msg(oi
->ip_open_count
,
1050 "Clear inode of %llu has open count %d\n",
1051 (unsigned long long)oi
->ip_blkno
, oi
->ip_open_count
);
1053 /* Clear all other flags. */
1054 oi
->ip_flags
= OCFS2_INODE_CACHE_INLINE
;
1055 oi
->ip_created_trans
= 0;
1056 oi
->ip_last_trans
= 0;
1057 oi
->ip_dir_start_lookup
= 0;
1058 oi
->ip_blkno
= 0ULL;
1064 /* Called under inode_lock, with no more references on the
1065 * struct inode, so it's safe here to check the flags field
1066 * and to manipulate i_nlink without any other locks. */
1067 void ocfs2_drop_inode(struct inode
*inode
)
1069 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1073 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1074 (unsigned long long)oi
->ip_blkno
, inode
->i_nlink
, oi
->ip_flags
);
1076 if (oi
->ip_flags
& OCFS2_INODE_MAYBE_ORPHANED
)
1077 generic_delete_inode(inode
);
1079 generic_drop_inode(inode
);
1085 * TODO: this should probably be merged into ocfs2_get_block
1087 * However, you now need to pay attention to the cont_prepare_write()
1088 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1089 * expects never to extend).
1091 struct buffer_head
*ocfs2_bread(struct inode
*inode
,
1092 int block
, int *err
, int reada
)
1094 struct buffer_head
*bh
= NULL
;
1097 int readflags
= OCFS2_BH_CACHED
;
1100 readflags
|= OCFS2_BH_READAHEAD
;
1102 if (((u64
)block
<< inode
->i_sb
->s_blocksize_bits
) >=
1103 i_size_read(inode
)) {
1108 tmperr
= ocfs2_extent_map_get_blocks(inode
, block
, &p_blkno
, NULL
,
1115 tmperr
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), p_blkno
, &bh
,
1135 * This is called from our getattr.
1137 int ocfs2_inode_revalidate(struct dentry
*dentry
)
1139 struct inode
*inode
= dentry
->d_inode
;
1142 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode
,
1143 inode
? (unsigned long long)OCFS2_I(inode
)->ip_blkno
: 0ULL);
1146 mlog(0, "eep, no inode!\n");
1151 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1152 if (OCFS2_I(inode
)->ip_flags
& OCFS2_INODE_DELETED
) {
1153 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1154 mlog(0, "inode deleted!\n");
1158 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1160 /* Let ocfs2_meta_lock do the work of updating our struct
1162 status
= ocfs2_meta_lock(inode
, NULL
, 0);
1164 if (status
!= -ENOENT
)
1168 ocfs2_meta_unlock(inode
, 0);
1176 * Updates a disk inode from a
1178 * Only takes ip_lock.
1180 int ocfs2_mark_inode_dirty(handle_t
*handle
,
1181 struct inode
*inode
,
1182 struct buffer_head
*bh
)
1185 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) bh
->b_data
;
1187 mlog_entry("(inode %llu)\n",
1188 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1190 status
= ocfs2_journal_access(handle
, inode
, bh
,
1191 OCFS2_JOURNAL_ACCESS_WRITE
);
1197 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1198 fe
->i_clusters
= cpu_to_le32(OCFS2_I(inode
)->ip_clusters
);
1199 fe
->i_attr
= cpu_to_le32(OCFS2_I(inode
)->ip_attr
);
1200 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
1202 fe
->i_size
= cpu_to_le64(i_size_read(inode
));
1203 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1204 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
1205 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
1206 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
1207 fe
->i_atime
= cpu_to_le64(inode
->i_atime
.tv_sec
);
1208 fe
->i_atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
1209 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
1210 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
1211 fe
->i_mtime
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
1212 fe
->i_mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
1214 status
= ocfs2_journal_dirty(handle
, bh
);
1227 * Updates a struct inode from a disk inode.
1228 * does no i/o, only takes ip_lock.
1230 void ocfs2_refresh_inode(struct inode
*inode
,
1231 struct ocfs2_dinode
*fe
)
1233 spin_lock(&OCFS2_I(inode
)->ip_lock
);
1235 OCFS2_I(inode
)->ip_clusters
= le32_to_cpu(fe
->i_clusters
);
1236 OCFS2_I(inode
)->ip_attr
= le32_to_cpu(fe
->i_attr
);
1237 ocfs2_set_inode_flags(inode
);
1238 i_size_write(inode
, le64_to_cpu(fe
->i_size
));
1239 inode
->i_nlink
= le16_to_cpu(fe
->i_links_count
);
1240 inode
->i_uid
= le32_to_cpu(fe
->i_uid
);
1241 inode
->i_gid
= le32_to_cpu(fe
->i_gid
);
1242 inode
->i_mode
= le16_to_cpu(fe
->i_mode
);
1243 if (S_ISLNK(inode
->i_mode
) && le32_to_cpu(fe
->i_clusters
) == 0)
1244 inode
->i_blocks
= 0;
1246 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(i_size_read(inode
));
1247 inode
->i_atime
.tv_sec
= le64_to_cpu(fe
->i_atime
);
1248 inode
->i_atime
.tv_nsec
= le32_to_cpu(fe
->i_atime_nsec
);
1249 inode
->i_mtime
.tv_sec
= le64_to_cpu(fe
->i_mtime
);
1250 inode
->i_mtime
.tv_nsec
= le32_to_cpu(fe
->i_mtime_nsec
);
1251 inode
->i_ctime
.tv_sec
= le64_to_cpu(fe
->i_ctime
);
1252 inode
->i_ctime
.tv_nsec
= le32_to_cpu(fe
->i_ctime_nsec
);
1254 spin_unlock(&OCFS2_I(inode
)->ip_lock
);