1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Code which implements an OCFS2 specific interface to our DLM.
8 * Copyright (C) 2003, 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.
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
30 #include <linux/smp_lock.h>
31 #include <linux/crc32.h>
32 #include <linux/kthread.h>
33 #include <linux/pagemap.h>
34 #include <linux/debugfs.h>
35 #include <linux/seq_file.h>
37 #include <cluster/heartbeat.h>
38 #include <cluster/nodemanager.h>
39 #include <cluster/tcp.h>
41 #include <dlm/dlmapi.h>
43 #define MLOG_MASK_PREFIX ML_DLM_GLUE
44 #include <cluster/masklog.h>
51 #include "extent_map.h"
52 #include "heartbeat.h"
60 #include "buffer_head_io.h"
62 struct ocfs2_mask_waiter
{
63 struct list_head mw_item
;
65 struct completion mw_complete
;
66 unsigned long mw_mask
;
67 unsigned long mw_goal
;
70 static void ocfs2_inode_ast_func(void *opaque
);
71 static void ocfs2_inode_bast_func(void *opaque
,
73 static void ocfs2_dentry_ast_func(void *opaque
);
74 static void ocfs2_dentry_bast_func(void *opaque
,
76 static void ocfs2_super_ast_func(void *opaque
);
77 static void ocfs2_super_bast_func(void *opaque
,
79 static void ocfs2_rename_ast_func(void *opaque
);
80 static void ocfs2_rename_bast_func(void *opaque
,
84 * Return value from ocfs2_convert_worker_t functions.
86 * These control the precise actions of ocfs2_generic_unblock_lock()
87 * and ocfs2_process_blocked_lock()
90 enum ocfs2_unblock_action
{
91 UNBLOCK_CONTINUE
= 0, /* Continue downconvert */
92 UNBLOCK_CONTINUE_POST
= 1, /* Continue downconvert, fire
93 * ->post_unlock callback */
94 UNBLOCK_STOP_POST
= 2, /* Do not downconvert, fire
95 * ->post_unlock() callback. */
98 struct ocfs2_unblock_ctl
{
100 enum ocfs2_unblock_action unblock_action
;
103 /* so far, all locks have gotten along with the same unlock ast */
104 static void ocfs2_unlock_ast_func(void *opaque
,
105 enum dlm_status status
);
106 static int ocfs2_unblock_meta(struct ocfs2_lock_res
*lockres
,
107 struct ocfs2_unblock_ctl
*ctl
);
108 static int ocfs2_unblock_data(struct ocfs2_lock_res
*lockres
,
109 struct ocfs2_unblock_ctl
*ctl
);
110 static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res
*lockres
,
111 struct ocfs2_unblock_ctl
*ctl
);
112 static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res
*lockres
,
113 struct ocfs2_unblock_ctl
*ctl
);
114 static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res
*lockres
,
115 struct ocfs2_unblock_ctl
*ctl
);
117 static void ocfs2_dentry_post_unlock(struct ocfs2_super
*osb
,
118 struct ocfs2_lock_res
*lockres
);
120 struct ocfs2_lock_res_ops
{
122 void (*bast
)(void *, int);
123 void (*unlock_ast
)(void *, enum dlm_status
);
124 int (*unblock
)(struct ocfs2_lock_res
*, struct ocfs2_unblock_ctl
*);
125 void (*post_unlock
)(struct ocfs2_super
*, struct ocfs2_lock_res
*);
128 typedef int (ocfs2_convert_worker_t
)(struct ocfs2_lock_res
*, int);
129 static int ocfs2_generic_unblock_lock(struct ocfs2_super
*osb
,
130 struct ocfs2_lock_res
*lockres
,
131 struct ocfs2_unblock_ctl
*ctl
,
132 ocfs2_convert_worker_t
*worker
);
134 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops
= {
135 .ast
= ocfs2_inode_ast_func
,
136 .bast
= ocfs2_inode_bast_func
,
137 .unlock_ast
= ocfs2_unlock_ast_func
,
138 .unblock
= ocfs2_unblock_inode_lock
,
141 static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops
= {
142 .ast
= ocfs2_inode_ast_func
,
143 .bast
= ocfs2_inode_bast_func
,
144 .unlock_ast
= ocfs2_unlock_ast_func
,
145 .unblock
= ocfs2_unblock_meta
,
148 static struct ocfs2_lock_res_ops ocfs2_inode_data_lops
= {
149 .ast
= ocfs2_inode_ast_func
,
150 .bast
= ocfs2_inode_bast_func
,
151 .unlock_ast
= ocfs2_unlock_ast_func
,
152 .unblock
= ocfs2_unblock_data
,
155 static struct ocfs2_lock_res_ops ocfs2_super_lops
= {
156 .ast
= ocfs2_super_ast_func
,
157 .bast
= ocfs2_super_bast_func
,
158 .unlock_ast
= ocfs2_unlock_ast_func
,
159 .unblock
= ocfs2_unblock_osb_lock
,
162 static struct ocfs2_lock_res_ops ocfs2_rename_lops
= {
163 .ast
= ocfs2_rename_ast_func
,
164 .bast
= ocfs2_rename_bast_func
,
165 .unlock_ast
= ocfs2_unlock_ast_func
,
166 .unblock
= ocfs2_unblock_osb_lock
,
169 static struct ocfs2_lock_res_ops ocfs2_dentry_lops
= {
170 .ast
= ocfs2_dentry_ast_func
,
171 .bast
= ocfs2_dentry_bast_func
,
172 .unlock_ast
= ocfs2_unlock_ast_func
,
173 .unblock
= ocfs2_unblock_dentry_lock
,
174 .post_unlock
= ocfs2_dentry_post_unlock
,
177 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res
*lockres
)
179 return lockres
->l_type
== OCFS2_LOCK_TYPE_META
||
180 lockres
->l_type
== OCFS2_LOCK_TYPE_DATA
||
181 lockres
->l_type
== OCFS2_LOCK_TYPE_RW
;
184 static inline int ocfs2_is_super_lock(struct ocfs2_lock_res
*lockres
)
186 return lockres
->l_type
== OCFS2_LOCK_TYPE_SUPER
;
189 static inline int ocfs2_is_rename_lock(struct ocfs2_lock_res
*lockres
)
191 return lockres
->l_type
== OCFS2_LOCK_TYPE_RENAME
;
194 static inline struct ocfs2_super
*ocfs2_lock_res_super(struct ocfs2_lock_res
*lockres
)
196 BUG_ON(!ocfs2_is_super_lock(lockres
)
197 && !ocfs2_is_rename_lock(lockres
));
199 return (struct ocfs2_super
*) lockres
->l_priv
;
202 static inline struct inode
*ocfs2_lock_res_inode(struct ocfs2_lock_res
*lockres
)
204 BUG_ON(!ocfs2_is_inode_lock(lockres
));
206 return (struct inode
*) lockres
->l_priv
;
209 static inline struct ocfs2_dentry_lock
*ocfs2_lock_res_dl(struct ocfs2_lock_res
*lockres
)
211 BUG_ON(lockres
->l_type
!= OCFS2_LOCK_TYPE_DENTRY
);
213 return (struct ocfs2_dentry_lock
*)lockres
->l_priv
;
216 static int ocfs2_lock_create(struct ocfs2_super
*osb
,
217 struct ocfs2_lock_res
*lockres
,
220 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res
*lockres
,
222 static void ocfs2_cluster_unlock(struct ocfs2_super
*osb
,
223 struct ocfs2_lock_res
*lockres
,
225 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res
*lockres
);
226 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res
*lockres
);
227 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res
*lockres
);
228 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res
*lockres
, int level
);
229 static void ocfs2_schedule_blocked_lock(struct ocfs2_super
*osb
,
230 struct ocfs2_lock_res
*lockres
);
231 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res
*lockres
,
233 #define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \
234 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
235 "resource %s: %s\n", dlm_errname(_stat), _func, \
236 _lockres->l_name, dlm_errmsg(_stat)); \
238 static void ocfs2_vote_on_unlock(struct ocfs2_super
*osb
,
239 struct ocfs2_lock_res
*lockres
);
240 static int ocfs2_meta_lock_update(struct inode
*inode
,
241 struct buffer_head
**bh
);
242 static void ocfs2_drop_osb_locks(struct ocfs2_super
*osb
);
243 static inline int ocfs2_highest_compat_lock_level(int level
);
244 static inline int ocfs2_can_downconvert_meta_lock(struct inode
*inode
,
245 struct ocfs2_lock_res
*lockres
,
248 static void ocfs2_build_lock_name(enum ocfs2_lock_type type
,
257 BUG_ON(type
>= OCFS2_NUM_LOCK_TYPES
);
259 len
= snprintf(name
, OCFS2_LOCK_ID_MAX_LEN
, "%c%s%016llx%08x",
260 ocfs2_lock_type_char(type
), OCFS2_LOCK_ID_PAD
,
261 (long long)blkno
, generation
);
263 BUG_ON(len
!= (OCFS2_LOCK_ID_MAX_LEN
- 1));
265 mlog(0, "built lock resource with name: %s\n", name
);
270 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock
);
272 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res
*res
,
273 struct ocfs2_dlm_debug
*dlm_debug
)
275 mlog(0, "Add tracking for lockres %s\n", res
->l_name
);
277 spin_lock(&ocfs2_dlm_tracking_lock
);
278 list_add(&res
->l_debug_list
, &dlm_debug
->d_lockres_tracking
);
279 spin_unlock(&ocfs2_dlm_tracking_lock
);
282 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res
*res
)
284 spin_lock(&ocfs2_dlm_tracking_lock
);
285 if (!list_empty(&res
->l_debug_list
))
286 list_del_init(&res
->l_debug_list
);
287 spin_unlock(&ocfs2_dlm_tracking_lock
);
290 static void ocfs2_lock_res_init_common(struct ocfs2_super
*osb
,
291 struct ocfs2_lock_res
*res
,
292 enum ocfs2_lock_type type
,
293 struct ocfs2_lock_res_ops
*ops
,
300 res
->l_level
= LKM_IVMODE
;
301 res
->l_requested
= LKM_IVMODE
;
302 res
->l_blocking
= LKM_IVMODE
;
303 res
->l_action
= OCFS2_AST_INVALID
;
304 res
->l_unlock_action
= OCFS2_UNLOCK_INVALID
;
306 res
->l_flags
= OCFS2_LOCK_INITIALIZED
;
308 ocfs2_add_lockres_tracking(res
, osb
->osb_dlm_debug
);
311 void ocfs2_lock_res_init_once(struct ocfs2_lock_res
*res
)
313 /* This also clears out the lock status block */
314 memset(res
, 0, sizeof(struct ocfs2_lock_res
));
315 spin_lock_init(&res
->l_lock
);
316 init_waitqueue_head(&res
->l_event
);
317 INIT_LIST_HEAD(&res
->l_blocked_list
);
318 INIT_LIST_HEAD(&res
->l_mask_waiters
);
321 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res
*res
,
322 enum ocfs2_lock_type type
,
323 unsigned int generation
,
326 struct ocfs2_lock_res_ops
*ops
;
329 case OCFS2_LOCK_TYPE_RW
:
330 ops
= &ocfs2_inode_rw_lops
;
332 case OCFS2_LOCK_TYPE_META
:
333 ops
= &ocfs2_inode_meta_lops
;
335 case OCFS2_LOCK_TYPE_DATA
:
336 ops
= &ocfs2_inode_data_lops
;
339 mlog_bug_on_msg(1, "type: %d\n", type
);
340 ops
= NULL
; /* thanks, gcc */
344 ocfs2_build_lock_name(type
, OCFS2_I(inode
)->ip_blkno
,
345 generation
, res
->l_name
);
346 ocfs2_lock_res_init_common(OCFS2_SB(inode
->i_sb
), res
, type
, ops
, inode
);
349 static __u64
ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res
*lockres
)
351 __be64 inode_blkno_be
;
353 memcpy(&inode_blkno_be
, &lockres
->l_name
[OCFS2_DENTRY_LOCK_INO_START
],
356 return be64_to_cpu(inode_blkno_be
);
359 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock
*dl
,
360 u64 parent
, struct inode
*inode
)
363 u64 inode_blkno
= OCFS2_I(inode
)->ip_blkno
;
364 __be64 inode_blkno_be
= cpu_to_be64(inode_blkno
);
365 struct ocfs2_lock_res
*lockres
= &dl
->dl_lockres
;
367 ocfs2_lock_res_init_once(lockres
);
370 * Unfortunately, the standard lock naming scheme won't work
371 * here because we have two 16 byte values to use. Instead,
372 * we'll stuff the inode number as a binary value. We still
373 * want error prints to show something without garbling the
374 * display, so drop a null byte in there before the inode
375 * number. A future version of OCFS2 will likely use all
376 * binary lock names. The stringified names have been a
377 * tremendous aid in debugging, but now that the debugfs
378 * interface exists, we can mangle things there if need be.
380 * NOTE: We also drop the standard "pad" value (the total lock
381 * name size stays the same though - the last part is all
382 * zeros due to the memset in ocfs2_lock_res_init_once()
384 len
= snprintf(lockres
->l_name
, OCFS2_DENTRY_LOCK_INO_START
,
386 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY
),
389 BUG_ON(len
!= (OCFS2_DENTRY_LOCK_INO_START
- 1));
391 memcpy(&lockres
->l_name
[OCFS2_DENTRY_LOCK_INO_START
], &inode_blkno_be
,
394 ocfs2_lock_res_init_common(OCFS2_SB(inode
->i_sb
), lockres
,
395 OCFS2_LOCK_TYPE_DENTRY
, &ocfs2_dentry_lops
,
399 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res
*res
,
400 struct ocfs2_super
*osb
)
402 /* Superblock lockres doesn't come from a slab so we call init
403 * once on it manually. */
404 ocfs2_lock_res_init_once(res
);
405 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER
, OCFS2_SUPER_BLOCK_BLKNO
,
407 ocfs2_lock_res_init_common(osb
, res
, OCFS2_LOCK_TYPE_SUPER
,
408 &ocfs2_super_lops
, osb
);
411 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res
*res
,
412 struct ocfs2_super
*osb
)
414 /* Rename lockres doesn't come from a slab so we call init
415 * once on it manually. */
416 ocfs2_lock_res_init_once(res
);
417 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME
, 0, 0, res
->l_name
);
418 ocfs2_lock_res_init_common(osb
, res
, OCFS2_LOCK_TYPE_RENAME
,
419 &ocfs2_rename_lops
, osb
);
422 void ocfs2_lock_res_free(struct ocfs2_lock_res
*res
)
426 if (!(res
->l_flags
& OCFS2_LOCK_INITIALIZED
))
429 ocfs2_remove_lockres_tracking(res
);
431 mlog_bug_on_msg(!list_empty(&res
->l_blocked_list
),
432 "Lockres %s is on the blocked list\n",
434 mlog_bug_on_msg(!list_empty(&res
->l_mask_waiters
),
435 "Lockres %s has mask waiters pending\n",
437 mlog_bug_on_msg(spin_is_locked(&res
->l_lock
),
438 "Lockres %s is locked\n",
440 mlog_bug_on_msg(res
->l_ro_holders
,
441 "Lockres %s has %u ro holders\n",
442 res
->l_name
, res
->l_ro_holders
);
443 mlog_bug_on_msg(res
->l_ex_holders
,
444 "Lockres %s has %u ex holders\n",
445 res
->l_name
, res
->l_ex_holders
);
447 /* Need to clear out the lock status block for the dlm */
448 memset(&res
->l_lksb
, 0, sizeof(res
->l_lksb
));
454 static inline void ocfs2_inc_holders(struct ocfs2_lock_res
*lockres
,
463 lockres
->l_ex_holders
++;
466 lockres
->l_ro_holders
++;
475 static inline void ocfs2_dec_holders(struct ocfs2_lock_res
*lockres
,
484 BUG_ON(!lockres
->l_ex_holders
);
485 lockres
->l_ex_holders
--;
488 BUG_ON(!lockres
->l_ro_holders
);
489 lockres
->l_ro_holders
--;
497 /* WARNING: This function lives in a world where the only three lock
498 * levels are EX, PR, and NL. It *will* have to be adjusted when more
499 * lock types are added. */
500 static inline int ocfs2_highest_compat_lock_level(int level
)
502 int new_level
= LKM_EXMODE
;
504 if (level
== LKM_EXMODE
)
505 new_level
= LKM_NLMODE
;
506 else if (level
== LKM_PRMODE
)
507 new_level
= LKM_PRMODE
;
511 static void lockres_set_flags(struct ocfs2_lock_res
*lockres
,
512 unsigned long newflags
)
514 struct list_head
*pos
, *tmp
;
515 struct ocfs2_mask_waiter
*mw
;
517 assert_spin_locked(&lockres
->l_lock
);
519 lockres
->l_flags
= newflags
;
521 list_for_each_safe(pos
, tmp
, &lockres
->l_mask_waiters
) {
522 mw
= list_entry(pos
, struct ocfs2_mask_waiter
, mw_item
);
523 if ((lockres
->l_flags
& mw
->mw_mask
) != mw
->mw_goal
)
526 list_del_init(&mw
->mw_item
);
528 complete(&mw
->mw_complete
);
531 static void lockres_or_flags(struct ocfs2_lock_res
*lockres
, unsigned long or)
533 lockres_set_flags(lockres
, lockres
->l_flags
| or);
535 static void lockres_clear_flags(struct ocfs2_lock_res
*lockres
,
538 lockres_set_flags(lockres
, lockres
->l_flags
& ~clear
);
541 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res
*lockres
)
545 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_BUSY
));
546 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_ATTACHED
));
547 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_BLOCKED
));
548 BUG_ON(lockres
->l_blocking
<= LKM_NLMODE
);
550 lockres
->l_level
= lockres
->l_requested
;
551 if (lockres
->l_level
<=
552 ocfs2_highest_compat_lock_level(lockres
->l_blocking
)) {
553 lockres
->l_blocking
= LKM_NLMODE
;
554 lockres_clear_flags(lockres
, OCFS2_LOCK_BLOCKED
);
556 lockres_clear_flags(lockres
, OCFS2_LOCK_BUSY
);
561 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res
*lockres
)
565 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_BUSY
));
566 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_ATTACHED
));
568 /* Convert from RO to EX doesn't really need anything as our
569 * information is already up to data. Convert from NL to
570 * *anything* however should mark ourselves as needing an
572 if (lockres
->l_level
== LKM_NLMODE
)
573 lockres_or_flags(lockres
, OCFS2_LOCK_NEEDS_REFRESH
);
575 lockres
->l_level
= lockres
->l_requested
;
576 lockres_clear_flags(lockres
, OCFS2_LOCK_BUSY
);
581 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res
*lockres
)
585 BUG_ON((!lockres
->l_flags
& OCFS2_LOCK_BUSY
));
586 BUG_ON(lockres
->l_flags
& OCFS2_LOCK_ATTACHED
);
588 if (lockres
->l_requested
> LKM_NLMODE
&&
589 !(lockres
->l_flags
& OCFS2_LOCK_LOCAL
))
590 lockres_or_flags(lockres
, OCFS2_LOCK_NEEDS_REFRESH
);
592 lockres
->l_level
= lockres
->l_requested
;
593 lockres_or_flags(lockres
, OCFS2_LOCK_ATTACHED
);
594 lockres_clear_flags(lockres
, OCFS2_LOCK_BUSY
);
599 static void ocfs2_inode_ast_func(void *opaque
)
601 struct ocfs2_lock_res
*lockres
= opaque
;
603 struct dlm_lockstatus
*lksb
;
608 inode
= ocfs2_lock_res_inode(lockres
);
610 mlog(0, "AST fired for inode %llu, l_action = %u, type = %s\n",
611 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, lockres
->l_action
,
612 ocfs2_lock_type_string(lockres
->l_type
));
614 BUG_ON(!ocfs2_is_inode_lock(lockres
));
616 spin_lock_irqsave(&lockres
->l_lock
, flags
);
618 lksb
= &(lockres
->l_lksb
);
619 if (lksb
->status
!= DLM_NORMAL
) {
620 mlog(ML_ERROR
, "ocfs2_inode_ast_func: lksb status value of %u "
621 "on inode %llu\n", lksb
->status
,
622 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
623 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
628 switch(lockres
->l_action
) {
629 case OCFS2_AST_ATTACH
:
630 ocfs2_generic_handle_attach_action(lockres
);
631 lockres_clear_flags(lockres
, OCFS2_LOCK_LOCAL
);
633 case OCFS2_AST_CONVERT
:
634 ocfs2_generic_handle_convert_action(lockres
);
636 case OCFS2_AST_DOWNCONVERT
:
637 ocfs2_generic_handle_downconvert_action(lockres
);
640 mlog(ML_ERROR
, "lockres %s: ast fired with invalid action: %u "
641 "lockres flags = 0x%lx, unlock action: %u\n",
642 lockres
->l_name
, lockres
->l_action
, lockres
->l_flags
,
643 lockres
->l_unlock_action
);
648 /* data and rw locking ignores refresh flag for now. */
649 if (lockres
->l_type
!= OCFS2_LOCK_TYPE_META
)
650 lockres_clear_flags(lockres
, OCFS2_LOCK_NEEDS_REFRESH
);
652 /* set it to something invalid so if we get called again we
654 lockres
->l_action
= OCFS2_AST_INVALID
;
655 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
656 wake_up(&lockres
->l_event
);
661 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res
*lockres
,
664 int needs_downconvert
= 0;
667 assert_spin_locked(&lockres
->l_lock
);
669 lockres_or_flags(lockres
, OCFS2_LOCK_BLOCKED
);
671 if (level
> lockres
->l_blocking
) {
672 /* only schedule a downconvert if we haven't already scheduled
673 * one that goes low enough to satisfy the level we're
674 * blocking. this also catches the case where we get
676 if (ocfs2_highest_compat_lock_level(level
) <
677 ocfs2_highest_compat_lock_level(lockres
->l_blocking
))
678 needs_downconvert
= 1;
680 lockres
->l_blocking
= level
;
683 mlog_exit(needs_downconvert
);
684 return needs_downconvert
;
687 static void ocfs2_generic_bast_func(struct ocfs2_super
*osb
,
688 struct ocfs2_lock_res
*lockres
,
691 int needs_downconvert
;
696 BUG_ON(level
<= LKM_NLMODE
);
698 spin_lock_irqsave(&lockres
->l_lock
, flags
);
699 needs_downconvert
= ocfs2_generic_handle_bast(lockres
, level
);
700 if (needs_downconvert
)
701 ocfs2_schedule_blocked_lock(osb
, lockres
);
702 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
704 wake_up(&lockres
->l_event
);
706 ocfs2_kick_vote_thread(osb
);
711 static void ocfs2_inode_bast_func(void *opaque
, int level
)
713 struct ocfs2_lock_res
*lockres
= opaque
;
715 struct ocfs2_super
*osb
;
719 BUG_ON(!ocfs2_is_inode_lock(lockres
));
721 inode
= ocfs2_lock_res_inode(lockres
);
722 osb
= OCFS2_SB(inode
->i_sb
);
724 mlog(0, "BAST fired for inode %llu, blocking %d, level %d type %s\n",
725 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, level
,
726 lockres
->l_level
, ocfs2_lock_type_string(lockres
->l_type
));
728 ocfs2_generic_bast_func(osb
, lockres
, level
);
733 static void ocfs2_generic_ast_func(struct ocfs2_lock_res
*lockres
,
736 struct dlm_lockstatus
*lksb
= &lockres
->l_lksb
;
739 spin_lock_irqsave(&lockres
->l_lock
, flags
);
741 if (lksb
->status
!= DLM_NORMAL
) {
742 mlog(ML_ERROR
, "lockres %s: lksb status value of %u!\n",
743 lockres
->l_name
, lksb
->status
);
744 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
748 switch(lockres
->l_action
) {
749 case OCFS2_AST_ATTACH
:
750 ocfs2_generic_handle_attach_action(lockres
);
752 case OCFS2_AST_CONVERT
:
753 ocfs2_generic_handle_convert_action(lockres
);
755 case OCFS2_AST_DOWNCONVERT
:
756 ocfs2_generic_handle_downconvert_action(lockres
);
763 lockres_clear_flags(lockres
, OCFS2_LOCK_NEEDS_REFRESH
);
765 /* set it to something invalid so if we get called again we
767 lockres
->l_action
= OCFS2_AST_INVALID
;
769 wake_up(&lockres
->l_event
);
770 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
773 static void ocfs2_super_ast_func(void *opaque
)
775 struct ocfs2_lock_res
*lockres
= opaque
;
778 mlog(0, "Superblock AST fired\n");
780 BUG_ON(!ocfs2_is_super_lock(lockres
));
781 ocfs2_generic_ast_func(lockres
, 0);
786 static void ocfs2_super_bast_func(void *opaque
,
789 struct ocfs2_lock_res
*lockres
= opaque
;
790 struct ocfs2_super
*osb
;
793 mlog(0, "Superblock BAST fired\n");
795 BUG_ON(!ocfs2_is_super_lock(lockres
));
796 osb
= ocfs2_lock_res_super(lockres
);
797 ocfs2_generic_bast_func(osb
, lockres
, level
);
802 static void ocfs2_rename_ast_func(void *opaque
)
804 struct ocfs2_lock_res
*lockres
= opaque
;
808 mlog(0, "Rename AST fired\n");
810 BUG_ON(!ocfs2_is_rename_lock(lockres
));
812 ocfs2_generic_ast_func(lockres
, 1);
817 static void ocfs2_rename_bast_func(void *opaque
,
820 struct ocfs2_lock_res
*lockres
= opaque
;
821 struct ocfs2_super
*osb
;
825 mlog(0, "Rename BAST fired\n");
827 BUG_ON(!ocfs2_is_rename_lock(lockres
));
829 osb
= ocfs2_lock_res_super(lockres
);
830 ocfs2_generic_bast_func(osb
, lockres
, level
);
835 static void ocfs2_dentry_ast_func(void *opaque
)
837 struct ocfs2_lock_res
*lockres
= opaque
;
841 ocfs2_generic_ast_func(lockres
, 1);
844 static void ocfs2_dentry_bast_func(void *opaque
, int level
)
846 struct ocfs2_lock_res
*lockres
= opaque
;
847 struct ocfs2_dentry_lock
*dl
= lockres
->l_priv
;
848 struct ocfs2_super
*osb
= OCFS2_SB(dl
->dl_inode
->i_sb
);
850 mlog(0, "Dentry bast: level: %d, name: %s\n", level
,
853 ocfs2_generic_bast_func(osb
, lockres
, level
);
856 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res
*lockres
,
862 spin_lock_irqsave(&lockres
->l_lock
, flags
);
863 lockres_clear_flags(lockres
, OCFS2_LOCK_BUSY
);
865 lockres
->l_action
= OCFS2_AST_INVALID
;
867 lockres
->l_unlock_action
= OCFS2_UNLOCK_INVALID
;
868 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
870 wake_up(&lockres
->l_event
);
874 /* Note: If we detect another process working on the lock (i.e.,
875 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
876 * to do the right thing in that case.
878 static int ocfs2_lock_create(struct ocfs2_super
*osb
,
879 struct ocfs2_lock_res
*lockres
,
884 enum dlm_status status
;
889 mlog(0, "lock %s, level = %d, flags = %d\n", lockres
->l_name
, level
,
892 spin_lock_irqsave(&lockres
->l_lock
, flags
);
893 if ((lockres
->l_flags
& OCFS2_LOCK_ATTACHED
) ||
894 (lockres
->l_flags
& OCFS2_LOCK_BUSY
)) {
895 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
899 lockres
->l_action
= OCFS2_AST_ATTACH
;
900 lockres
->l_requested
= level
;
901 lockres_or_flags(lockres
, OCFS2_LOCK_BUSY
);
902 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
904 status
= dlmlock(osb
->dlm
,
909 OCFS2_LOCK_ID_MAX_LEN
- 1,
912 lockres
->l_ops
->bast
);
913 if (status
!= DLM_NORMAL
) {
914 ocfs2_log_dlm_error("dlmlock", status
, lockres
);
916 ocfs2_recover_from_dlm_error(lockres
, 1);
919 mlog(0, "lock %s, successfull return from dlmlock\n", lockres
->l_name
);
926 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res
*lockres
,
932 spin_lock_irqsave(&lockres
->l_lock
, flags
);
933 ret
= lockres
->l_flags
& flag
;
934 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
939 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res
*lockres
)
942 wait_event(lockres
->l_event
,
943 !ocfs2_check_wait_flag(lockres
, OCFS2_LOCK_BUSY
));
946 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res
*lockres
)
949 wait_event(lockres
->l_event
,
950 !ocfs2_check_wait_flag(lockres
, OCFS2_LOCK_REFRESHING
));
953 /* predict what lock level we'll be dropping down to on behalf
954 * of another node, and return true if the currently wanted
955 * level will be compatible with it. */
956 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res
*lockres
,
959 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_BLOCKED
));
961 return wanted
<= ocfs2_highest_compat_lock_level(lockres
->l_blocking
);
964 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter
*mw
)
966 INIT_LIST_HEAD(&mw
->mw_item
);
967 init_completion(&mw
->mw_complete
);
970 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter
*mw
)
972 wait_for_completion(&mw
->mw_complete
);
973 /* Re-arm the completion in case we want to wait on it again */
974 INIT_COMPLETION(mw
->mw_complete
);
975 return mw
->mw_status
;
978 static void lockres_add_mask_waiter(struct ocfs2_lock_res
*lockres
,
979 struct ocfs2_mask_waiter
*mw
,
983 BUG_ON(!list_empty(&mw
->mw_item
));
985 assert_spin_locked(&lockres
->l_lock
);
987 list_add_tail(&mw
->mw_item
, &lockres
->l_mask_waiters
);
992 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
993 * if the mask still hadn't reached its goal */
994 static int lockres_remove_mask_waiter(struct ocfs2_lock_res
*lockres
,
995 struct ocfs2_mask_waiter
*mw
)
1000 spin_lock_irqsave(&lockres
->l_lock
, flags
);
1001 if (!list_empty(&mw
->mw_item
)) {
1002 if ((lockres
->l_flags
& mw
->mw_mask
) != mw
->mw_goal
)
1005 list_del_init(&mw
->mw_item
);
1006 init_completion(&mw
->mw_complete
);
1008 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1014 static int ocfs2_cluster_lock(struct ocfs2_super
*osb
,
1015 struct ocfs2_lock_res
*lockres
,
1020 struct ocfs2_mask_waiter mw
;
1021 enum dlm_status status
;
1022 int wait
, catch_signals
= !(osb
->s_mount_opt
& OCFS2_MOUNT_NOINTR
);
1023 int ret
= 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1024 unsigned long flags
;
1028 ocfs2_init_mask_waiter(&mw
);
1033 if (catch_signals
&& signal_pending(current
)) {
1038 spin_lock_irqsave(&lockres
->l_lock
, flags
);
1040 mlog_bug_on_msg(lockres
->l_flags
& OCFS2_LOCK_FREEING
,
1041 "Cluster lock called on freeing lockres %s! flags "
1042 "0x%lx\n", lockres
->l_name
, lockres
->l_flags
);
1044 /* We only compare against the currently granted level
1045 * here. If the lock is blocked waiting on a downconvert,
1046 * we'll get caught below. */
1047 if (lockres
->l_flags
& OCFS2_LOCK_BUSY
&&
1048 level
> lockres
->l_level
) {
1049 /* is someone sitting in dlm_lock? If so, wait on
1051 lockres_add_mask_waiter(lockres
, &mw
, OCFS2_LOCK_BUSY
, 0);
1056 if (!(lockres
->l_flags
& OCFS2_LOCK_ATTACHED
)) {
1057 /* lock has not been created yet. */
1058 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1060 ret
= ocfs2_lock_create(osb
, lockres
, LKM_NLMODE
, 0);
1068 if (lockres
->l_flags
& OCFS2_LOCK_BLOCKED
&&
1069 !ocfs2_may_continue_on_blocked_lock(lockres
, level
)) {
1070 /* is the lock is currently blocked on behalf of
1072 lockres_add_mask_waiter(lockres
, &mw
, OCFS2_LOCK_BLOCKED
, 0);
1077 if (level
> lockres
->l_level
) {
1078 if (lockres
->l_action
!= OCFS2_AST_INVALID
)
1079 mlog(ML_ERROR
, "lockres %s has action %u pending\n",
1080 lockres
->l_name
, lockres
->l_action
);
1082 lockres
->l_action
= OCFS2_AST_CONVERT
;
1083 lockres
->l_requested
= level
;
1084 lockres_or_flags(lockres
, OCFS2_LOCK_BUSY
);
1085 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1087 BUG_ON(level
== LKM_IVMODE
);
1088 BUG_ON(level
== LKM_NLMODE
);
1090 mlog(0, "lock %s, convert from %d to level = %d\n",
1091 lockres
->l_name
, lockres
->l_level
, level
);
1093 /* call dlm_lock to upgrade lock now */
1094 status
= dlmlock(osb
->dlm
,
1097 lkm_flags
|LKM_CONVERT
|LKM_VALBLK
,
1099 OCFS2_LOCK_ID_MAX_LEN
- 1,
1100 lockres
->l_ops
->ast
,
1102 lockres
->l_ops
->bast
);
1103 if (status
!= DLM_NORMAL
) {
1104 if ((lkm_flags
& LKM_NOQUEUE
) &&
1105 (status
== DLM_NOTQUEUED
))
1108 ocfs2_log_dlm_error("dlmlock", status
,
1112 ocfs2_recover_from_dlm_error(lockres
, 1);
1116 mlog(0, "lock %s, successfull return from dlmlock\n",
1119 /* At this point we've gone inside the dlm and need to
1120 * complete our work regardless. */
1123 /* wait for busy to clear and carry on */
1127 /* Ok, if we get here then we're good to go. */
1128 ocfs2_inc_holders(lockres
, level
);
1132 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1135 * This is helping work around a lock inversion between the page lock
1136 * and dlm locks. One path holds the page lock while calling aops
1137 * which block acquiring dlm locks. The voting thread holds dlm
1138 * locks while acquiring page locks while down converting data locks.
1139 * This block is helping an aop path notice the inversion and back
1140 * off to unlock its page lock before trying the dlm lock again.
1142 if (wait
&& arg_flags
& OCFS2_LOCK_NONBLOCK
&&
1143 mw
.mw_mask
& (OCFS2_LOCK_BUSY
|OCFS2_LOCK_BLOCKED
)) {
1145 if (lockres_remove_mask_waiter(lockres
, &mw
))
1151 ret
= ocfs2_wait_for_mask(&mw
);
1161 static void ocfs2_cluster_unlock(struct ocfs2_super
*osb
,
1162 struct ocfs2_lock_res
*lockres
,
1165 unsigned long flags
;
1168 spin_lock_irqsave(&lockres
->l_lock
, flags
);
1169 ocfs2_dec_holders(lockres
, level
);
1170 ocfs2_vote_on_unlock(osb
, lockres
);
1171 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1175 int ocfs2_create_new_lock(struct ocfs2_super
*osb
,
1176 struct ocfs2_lock_res
*lockres
,
1180 int level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1181 unsigned long flags
;
1182 int lkm_flags
= local
? LKM_LOCAL
: 0;
1184 spin_lock_irqsave(&lockres
->l_lock
, flags
);
1185 BUG_ON(lockres
->l_flags
& OCFS2_LOCK_ATTACHED
);
1186 lockres_or_flags(lockres
, OCFS2_LOCK_LOCAL
);
1187 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1189 return ocfs2_lock_create(osb
, lockres
, level
, lkm_flags
);
1192 /* Grants us an EX lock on the data and metadata resources, skipping
1193 * the normal cluster directory lookup. Use this ONLY on newly created
1194 * inodes which other nodes can't possibly see, and which haven't been
1195 * hashed in the inode hash yet. This can give us a good performance
1196 * increase as it'll skip the network broadcast normally associated
1197 * with creating a new lock resource. */
1198 int ocfs2_create_new_inode_locks(struct inode
*inode
)
1201 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1204 BUG_ON(!ocfs2_inode_is_new(inode
));
1208 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
1210 /* NOTE: That we don't increment any of the holder counts, nor
1211 * do we add anything to a journal handle. Since this is
1212 * supposed to be a new inode which the cluster doesn't know
1213 * about yet, there is no need to. As far as the LVB handling
1214 * is concerned, this is basically like acquiring an EX lock
1215 * on a resource which has an invalid one -- we'll set it
1216 * valid when we release the EX. */
1218 ret
= ocfs2_create_new_lock(osb
, &OCFS2_I(inode
)->ip_rw_lockres
, 1, 1);
1225 * We don't want to use LKM_LOCAL on a meta data lock as they
1226 * don't use a generation in their lock names.
1228 ret
= ocfs2_create_new_lock(osb
, &OCFS2_I(inode
)->ip_meta_lockres
, 1, 0);
1234 ret
= ocfs2_create_new_lock(osb
, &OCFS2_I(inode
)->ip_data_lockres
, 1, 1);
1245 int ocfs2_rw_lock(struct inode
*inode
, int write
)
1248 struct ocfs2_lock_res
*lockres
;
1254 mlog(0, "inode %llu take %s RW lock\n",
1255 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1256 write
? "EXMODE" : "PRMODE");
1258 lockres
= &OCFS2_I(inode
)->ip_rw_lockres
;
1260 level
= write
? LKM_EXMODE
: LKM_PRMODE
;
1262 status
= ocfs2_cluster_lock(OCFS2_SB(inode
->i_sb
), lockres
, level
, 0,
1271 void ocfs2_rw_unlock(struct inode
*inode
, int write
)
1273 int level
= write
? LKM_EXMODE
: LKM_PRMODE
;
1274 struct ocfs2_lock_res
*lockres
= &OCFS2_I(inode
)->ip_rw_lockres
;
1278 mlog(0, "inode %llu drop %s RW lock\n",
1279 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1280 write
? "EXMODE" : "PRMODE");
1282 ocfs2_cluster_unlock(OCFS2_SB(inode
->i_sb
), lockres
, level
);
1287 int ocfs2_data_lock_full(struct inode
*inode
,
1291 int status
= 0, level
;
1292 struct ocfs2_lock_res
*lockres
;
1298 mlog(0, "inode %llu take %s DATA lock\n",
1299 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1300 write
? "EXMODE" : "PRMODE");
1302 /* We'll allow faking a readonly data lock for
1304 if (ocfs2_is_hard_readonly(OCFS2_SB(inode
->i_sb
))) {
1312 lockres
= &OCFS2_I(inode
)->ip_data_lockres
;
1314 level
= write
? LKM_EXMODE
: LKM_PRMODE
;
1316 status
= ocfs2_cluster_lock(OCFS2_SB(inode
->i_sb
), lockres
, level
,
1318 if (status
< 0 && status
!= -EAGAIN
)
1326 /* see ocfs2_meta_lock_with_page() */
1327 int ocfs2_data_lock_with_page(struct inode
*inode
,
1333 ret
= ocfs2_data_lock_full(inode
, write
, OCFS2_LOCK_NONBLOCK
);
1334 if (ret
== -EAGAIN
) {
1336 if (ocfs2_data_lock(inode
, write
) == 0)
1337 ocfs2_data_unlock(inode
, write
);
1338 ret
= AOP_TRUNCATED_PAGE
;
1344 static void ocfs2_vote_on_unlock(struct ocfs2_super
*osb
,
1345 struct ocfs2_lock_res
*lockres
)
1351 /* If we know that another node is waiting on our lock, kick
1352 * the vote thread * pre-emptively when we reach a release
1354 if (lockres
->l_flags
& OCFS2_LOCK_BLOCKED
) {
1355 switch(lockres
->l_blocking
) {
1357 if (!lockres
->l_ex_holders
&& !lockres
->l_ro_holders
)
1361 if (!lockres
->l_ex_holders
)
1370 ocfs2_kick_vote_thread(osb
);
1375 void ocfs2_data_unlock(struct inode
*inode
,
1378 int level
= write
? LKM_EXMODE
: LKM_PRMODE
;
1379 struct ocfs2_lock_res
*lockres
= &OCFS2_I(inode
)->ip_data_lockres
;
1383 mlog(0, "inode %llu drop %s DATA lock\n",
1384 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1385 write
? "EXMODE" : "PRMODE");
1387 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode
->i_sb
)))
1388 ocfs2_cluster_unlock(OCFS2_SB(inode
->i_sb
), lockres
, level
);
1393 #define OCFS2_SEC_BITS 34
1394 #define OCFS2_SEC_SHIFT (64 - 34)
1395 #define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
1397 /* LVB only has room for 64 bits of time here so we pack it for
1399 static u64
ocfs2_pack_timespec(struct timespec
*spec
)
1402 u64 sec
= spec
->tv_sec
;
1403 u32 nsec
= spec
->tv_nsec
;
1405 res
= (sec
<< OCFS2_SEC_SHIFT
) | (nsec
& OCFS2_NSEC_MASK
);
1410 /* Call this with the lockres locked. I am reasonably sure we don't
1411 * need ip_lock in this function as anyone who would be changing those
1412 * values is supposed to be blocked in ocfs2_meta_lock right now. */
1413 static void __ocfs2_stuff_meta_lvb(struct inode
*inode
)
1415 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1416 struct ocfs2_lock_res
*lockres
= &oi
->ip_meta_lockres
;
1417 struct ocfs2_meta_lvb
*lvb
;
1421 lvb
= (struct ocfs2_meta_lvb
*) lockres
->l_lksb
.lvb
;
1424 * Invalidate the LVB of a deleted inode - this way other
1425 * nodes are forced to go to disk and discover the new inode
1428 if (oi
->ip_flags
& OCFS2_INODE_DELETED
) {
1429 lvb
->lvb_version
= 0;
1433 lvb
->lvb_version
= OCFS2_LVB_VERSION
;
1434 lvb
->lvb_isize
= cpu_to_be64(i_size_read(inode
));
1435 lvb
->lvb_iclusters
= cpu_to_be32(oi
->ip_clusters
);
1436 lvb
->lvb_iuid
= cpu_to_be32(inode
->i_uid
);
1437 lvb
->lvb_igid
= cpu_to_be32(inode
->i_gid
);
1438 lvb
->lvb_imode
= cpu_to_be16(inode
->i_mode
);
1439 lvb
->lvb_inlink
= cpu_to_be16(inode
->i_nlink
);
1440 lvb
->lvb_iatime_packed
=
1441 cpu_to_be64(ocfs2_pack_timespec(&inode
->i_atime
));
1442 lvb
->lvb_ictime_packed
=
1443 cpu_to_be64(ocfs2_pack_timespec(&inode
->i_ctime
));
1444 lvb
->lvb_imtime_packed
=
1445 cpu_to_be64(ocfs2_pack_timespec(&inode
->i_mtime
));
1446 lvb
->lvb_iattr
= cpu_to_be32(oi
->ip_attr
);
1447 lvb
->lvb_igeneration
= cpu_to_be32(inode
->i_generation
);
1450 mlog_meta_lvb(0, lockres
);
1455 static void ocfs2_unpack_timespec(struct timespec
*spec
,
1458 spec
->tv_sec
= packed_time
>> OCFS2_SEC_SHIFT
;
1459 spec
->tv_nsec
= packed_time
& OCFS2_NSEC_MASK
;
1462 static void ocfs2_refresh_inode_from_lvb(struct inode
*inode
)
1464 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1465 struct ocfs2_lock_res
*lockres
= &oi
->ip_meta_lockres
;
1466 struct ocfs2_meta_lvb
*lvb
;
1470 mlog_meta_lvb(0, lockres
);
1472 lvb
= (struct ocfs2_meta_lvb
*) lockres
->l_lksb
.lvb
;
1474 /* We're safe here without the lockres lock... */
1475 spin_lock(&oi
->ip_lock
);
1476 oi
->ip_clusters
= be32_to_cpu(lvb
->lvb_iclusters
);
1477 i_size_write(inode
, be64_to_cpu(lvb
->lvb_isize
));
1479 oi
->ip_attr
= be32_to_cpu(lvb
->lvb_iattr
);
1480 ocfs2_set_inode_flags(inode
);
1482 /* fast-symlinks are a special case */
1483 if (S_ISLNK(inode
->i_mode
) && !oi
->ip_clusters
)
1484 inode
->i_blocks
= 0;
1487 ocfs2_align_bytes_to_sectors(i_size_read(inode
));
1489 inode
->i_uid
= be32_to_cpu(lvb
->lvb_iuid
);
1490 inode
->i_gid
= be32_to_cpu(lvb
->lvb_igid
);
1491 inode
->i_mode
= be16_to_cpu(lvb
->lvb_imode
);
1492 inode
->i_nlink
= be16_to_cpu(lvb
->lvb_inlink
);
1493 ocfs2_unpack_timespec(&inode
->i_atime
,
1494 be64_to_cpu(lvb
->lvb_iatime_packed
));
1495 ocfs2_unpack_timespec(&inode
->i_mtime
,
1496 be64_to_cpu(lvb
->lvb_imtime_packed
));
1497 ocfs2_unpack_timespec(&inode
->i_ctime
,
1498 be64_to_cpu(lvb
->lvb_ictime_packed
));
1499 spin_unlock(&oi
->ip_lock
);
1504 static inline int ocfs2_meta_lvb_is_trustable(struct inode
*inode
,
1505 struct ocfs2_lock_res
*lockres
)
1507 struct ocfs2_meta_lvb
*lvb
= (struct ocfs2_meta_lvb
*) lockres
->l_lksb
.lvb
;
1509 if (lvb
->lvb_version
== OCFS2_LVB_VERSION
1510 && be32_to_cpu(lvb
->lvb_igeneration
) == inode
->i_generation
)
1515 /* Determine whether a lock resource needs to be refreshed, and
1516 * arbitrate who gets to refresh it.
1518 * 0 means no refresh needed.
1520 * > 0 means you need to refresh this and you MUST call
1521 * ocfs2_complete_lock_res_refresh afterwards. */
1522 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res
*lockres
)
1524 unsigned long flags
;
1530 spin_lock_irqsave(&lockres
->l_lock
, flags
);
1531 if (!(lockres
->l_flags
& OCFS2_LOCK_NEEDS_REFRESH
)) {
1532 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1536 if (lockres
->l_flags
& OCFS2_LOCK_REFRESHING
) {
1537 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1539 ocfs2_wait_on_refreshing_lock(lockres
);
1543 /* Ok, I'll be the one to refresh this lock. */
1544 lockres_or_flags(lockres
, OCFS2_LOCK_REFRESHING
);
1545 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1553 /* If status is non zero, I'll mark it as not being in refresh
1554 * anymroe, but i won't clear the needs refresh flag. */
1555 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res
*lockres
,
1558 unsigned long flags
;
1561 spin_lock_irqsave(&lockres
->l_lock
, flags
);
1562 lockres_clear_flags(lockres
, OCFS2_LOCK_REFRESHING
);
1564 lockres_clear_flags(lockres
, OCFS2_LOCK_NEEDS_REFRESH
);
1565 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
1567 wake_up(&lockres
->l_event
);
1572 /* may or may not return a bh if it went to disk. */
1573 static int ocfs2_meta_lock_update(struct inode
*inode
,
1574 struct buffer_head
**bh
)
1577 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
1578 struct ocfs2_lock_res
*lockres
;
1579 struct ocfs2_dinode
*fe
;
1583 spin_lock(&oi
->ip_lock
);
1584 if (oi
->ip_flags
& OCFS2_INODE_DELETED
) {
1585 mlog(0, "Orphaned inode %llu was deleted while we "
1586 "were waiting on a lock. ip_flags = 0x%x\n",
1587 (unsigned long long)oi
->ip_blkno
, oi
->ip_flags
);
1588 spin_unlock(&oi
->ip_lock
);
1592 spin_unlock(&oi
->ip_lock
);
1594 lockres
= &oi
->ip_meta_lockres
;
1596 if (!ocfs2_should_refresh_lock_res(lockres
))
1599 /* This will discard any caching information we might have had
1600 * for the inode metadata. */
1601 ocfs2_metadata_cache_purge(inode
);
1603 /* will do nothing for inode types that don't use the extent
1604 * map (directories, bitmap files, etc) */
1605 ocfs2_extent_map_trunc(inode
, 0);
1607 if (ocfs2_meta_lvb_is_trustable(inode
, lockres
)) {
1608 mlog(0, "Trusting LVB on inode %llu\n",
1609 (unsigned long long)oi
->ip_blkno
);
1610 ocfs2_refresh_inode_from_lvb(inode
);
1612 /* Boo, we have to go to disk. */
1613 /* read bh, cast, ocfs2_refresh_inode */
1614 status
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
), oi
->ip_blkno
,
1615 bh
, OCFS2_BH_CACHED
, inode
);
1620 fe
= (struct ocfs2_dinode
*) (*bh
)->b_data
;
1622 /* This is a good chance to make sure we're not
1623 * locking an invalid object.
1625 * We bug on a stale inode here because we checked
1626 * above whether it was wiped from disk. The wiping
1627 * node provides a guarantee that we receive that
1628 * message and can mark the inode before dropping any
1629 * locks associated with it. */
1630 if (!OCFS2_IS_VALID_DINODE(fe
)) {
1631 OCFS2_RO_ON_INVALID_DINODE(inode
->i_sb
, fe
);
1635 mlog_bug_on_msg(inode
->i_generation
!=
1636 le32_to_cpu(fe
->i_generation
),
1637 "Invalid dinode %llu disk generation: %u "
1638 "inode->i_generation: %u\n",
1639 (unsigned long long)oi
->ip_blkno
,
1640 le32_to_cpu(fe
->i_generation
),
1641 inode
->i_generation
);
1642 mlog_bug_on_msg(le64_to_cpu(fe
->i_dtime
) ||
1643 !(fe
->i_flags
& cpu_to_le32(OCFS2_VALID_FL
)),
1644 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
1645 (unsigned long long)oi
->ip_blkno
,
1646 (unsigned long long)le64_to_cpu(fe
->i_dtime
),
1647 le32_to_cpu(fe
->i_flags
));
1649 ocfs2_refresh_inode(inode
, fe
);
1654 ocfs2_complete_lock_res_refresh(lockres
, status
);
1660 static int ocfs2_assign_bh(struct inode
*inode
,
1661 struct buffer_head
**ret_bh
,
1662 struct buffer_head
*passed_bh
)
1667 /* Ok, the update went to disk for us, use the
1669 *ret_bh
= passed_bh
;
1675 status
= ocfs2_read_block(OCFS2_SB(inode
->i_sb
),
1676 OCFS2_I(inode
)->ip_blkno
,
1687 * returns < 0 error if the callback will never be called, otherwise
1688 * the result of the lock will be communicated via the callback.
1690 int ocfs2_meta_lock_full(struct inode
*inode
,
1691 struct ocfs2_journal_handle
*handle
,
1692 struct buffer_head
**ret_bh
,
1696 int status
, level
, dlm_flags
, acquired
;
1697 struct ocfs2_lock_res
*lockres
;
1698 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1699 struct buffer_head
*local_bh
= NULL
;
1705 mlog(0, "inode %llu, take %s META lock\n",
1706 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1707 ex
? "EXMODE" : "PRMODE");
1711 /* We'll allow faking a readonly metadata lock for
1713 if (ocfs2_is_hard_readonly(osb
)) {
1719 if (!(arg_flags
& OCFS2_META_LOCK_RECOVERY
))
1720 wait_event(osb
->recovery_event
,
1721 ocfs2_node_map_is_empty(osb
, &osb
->recovery_map
));
1724 lockres
= &OCFS2_I(inode
)->ip_meta_lockres
;
1725 level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1727 if (arg_flags
& OCFS2_META_LOCK_NOQUEUE
)
1728 dlm_flags
|= LKM_NOQUEUE
;
1730 status
= ocfs2_cluster_lock(osb
, lockres
, level
, dlm_flags
, arg_flags
);
1732 if (status
!= -EAGAIN
&& status
!= -EIOCBRETRY
)
1737 /* Notify the error cleanup path to drop the cluster lock. */
1740 /* We wait twice because a node may have died while we were in
1741 * the lower dlm layers. The second time though, we've
1742 * committed to owning this lock so we don't allow signals to
1743 * abort the operation. */
1744 if (!(arg_flags
& OCFS2_META_LOCK_RECOVERY
))
1745 wait_event(osb
->recovery_event
,
1746 ocfs2_node_map_is_empty(osb
, &osb
->recovery_map
));
1749 * We only see this flag if we're being called from
1750 * ocfs2_read_locked_inode(). It means we're locking an inode
1751 * which hasn't been populated yet, so clear the refresh flag
1752 * and let the caller handle it.
1754 if (inode
->i_state
& I_NEW
) {
1756 ocfs2_complete_lock_res_refresh(lockres
, 0);
1760 /* This is fun. The caller may want a bh back, or it may
1761 * not. ocfs2_meta_lock_update definitely wants one in, but
1762 * may or may not read one, depending on what's in the
1763 * LVB. The result of all of this is that we've *only* gone to
1764 * disk if we have to, so the complexity is worthwhile. */
1765 status
= ocfs2_meta_lock_update(inode
, &local_bh
);
1767 if (status
!= -ENOENT
)
1773 status
= ocfs2_assign_bh(inode
, ret_bh
, local_bh
);
1781 status
= ocfs2_handle_add_lock(handle
, inode
);
1788 if (ret_bh
&& (*ret_bh
)) {
1793 ocfs2_meta_unlock(inode
, ex
);
1804 * This is working around a lock inversion between tasks acquiring DLM locks
1805 * while holding a page lock and the vote thread which blocks dlm lock acquiry
1806 * while acquiring page locks.
1808 * ** These _with_page variantes are only intended to be called from aop
1809 * methods that hold page locks and return a very specific *positive* error
1810 * code that aop methods pass up to the VFS -- test for errors with != 0. **
1812 * The DLM is called such that it returns -EAGAIN if it would have blocked
1813 * waiting for the vote thread. In that case we unlock our page so the vote
1814 * thread can make progress. Once we've done this we have to return
1815 * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up
1816 * into the VFS who will then immediately retry the aop call.
1818 * We do a blocking lock and immediate unlock before returning, though, so that
1819 * the lock has a great chance of being cached on this node by the time the VFS
1820 * calls back to retry the aop. This has a potential to livelock as nodes
1821 * ping locks back and forth, but that's a risk we're willing to take to avoid
1822 * the lock inversion simply.
1824 int ocfs2_meta_lock_with_page(struct inode
*inode
,
1825 struct ocfs2_journal_handle
*handle
,
1826 struct buffer_head
**ret_bh
,
1832 ret
= ocfs2_meta_lock_full(inode
, handle
, ret_bh
, ex
,
1833 OCFS2_LOCK_NONBLOCK
);
1834 if (ret
== -EAGAIN
) {
1836 if (ocfs2_meta_lock(inode
, handle
, ret_bh
, ex
) == 0)
1837 ocfs2_meta_unlock(inode
, ex
);
1838 ret
= AOP_TRUNCATED_PAGE
;
1844 void ocfs2_meta_unlock(struct inode
*inode
,
1847 int level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1848 struct ocfs2_lock_res
*lockres
= &OCFS2_I(inode
)->ip_meta_lockres
;
1852 mlog(0, "inode %llu drop %s META lock\n",
1853 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1854 ex
? "EXMODE" : "PRMODE");
1856 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode
->i_sb
)))
1857 ocfs2_cluster_unlock(OCFS2_SB(inode
->i_sb
), lockres
, level
);
1862 int ocfs2_super_lock(struct ocfs2_super
*osb
,
1866 int level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1867 struct ocfs2_lock_res
*lockres
= &osb
->osb_super_lockres
;
1868 struct buffer_head
*bh
;
1869 struct ocfs2_slot_info
*si
= osb
->slot_info
;
1873 if (ocfs2_is_hard_readonly(osb
))
1876 status
= ocfs2_cluster_lock(osb
, lockres
, level
, 0, 0);
1882 /* The super block lock path is really in the best position to
1883 * know when resources covered by the lock need to be
1884 * refreshed, so we do it here. Of course, making sense of
1885 * everything is up to the caller :) */
1886 status
= ocfs2_should_refresh_lock_res(lockres
);
1893 status
= ocfs2_read_block(osb
, bh
->b_blocknr
, &bh
, 0,
1896 ocfs2_update_slot_info(si
);
1898 ocfs2_complete_lock_res_refresh(lockres
, status
);
1908 void ocfs2_super_unlock(struct ocfs2_super
*osb
,
1911 int level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1912 struct ocfs2_lock_res
*lockres
= &osb
->osb_super_lockres
;
1914 ocfs2_cluster_unlock(osb
, lockres
, level
);
1917 int ocfs2_rename_lock(struct ocfs2_super
*osb
)
1920 struct ocfs2_lock_res
*lockres
= &osb
->osb_rename_lockres
;
1922 if (ocfs2_is_hard_readonly(osb
))
1925 status
= ocfs2_cluster_lock(osb
, lockres
, LKM_EXMODE
, 0, 0);
1932 void ocfs2_rename_unlock(struct ocfs2_super
*osb
)
1934 struct ocfs2_lock_res
*lockres
= &osb
->osb_rename_lockres
;
1936 ocfs2_cluster_unlock(osb
, lockres
, LKM_EXMODE
);
1939 int ocfs2_dentry_lock(struct dentry
*dentry
, int ex
)
1942 int level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1943 struct ocfs2_dentry_lock
*dl
= dentry
->d_fsdata
;
1944 struct ocfs2_super
*osb
= OCFS2_SB(dentry
->d_sb
);
1948 if (ocfs2_is_hard_readonly(osb
))
1951 ret
= ocfs2_cluster_lock(osb
, &dl
->dl_lockres
, level
, 0, 0);
1958 void ocfs2_dentry_unlock(struct dentry
*dentry
, int ex
)
1960 int level
= ex
? LKM_EXMODE
: LKM_PRMODE
;
1961 struct ocfs2_dentry_lock
*dl
= dentry
->d_fsdata
;
1962 struct ocfs2_super
*osb
= OCFS2_SB(dentry
->d_sb
);
1964 ocfs2_cluster_unlock(osb
, &dl
->dl_lockres
, level
);
1967 /* Reference counting of the dlm debug structure. We want this because
1968 * open references on the debug inodes can live on after a mount, so
1969 * we can't rely on the ocfs2_super to always exist. */
1970 static void ocfs2_dlm_debug_free(struct kref
*kref
)
1972 struct ocfs2_dlm_debug
*dlm_debug
;
1974 dlm_debug
= container_of(kref
, struct ocfs2_dlm_debug
, d_refcnt
);
1979 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug
*dlm_debug
)
1982 kref_put(&dlm_debug
->d_refcnt
, ocfs2_dlm_debug_free
);
1985 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug
*debug
)
1987 kref_get(&debug
->d_refcnt
);
1990 struct ocfs2_dlm_debug
*ocfs2_new_dlm_debug(void)
1992 struct ocfs2_dlm_debug
*dlm_debug
;
1994 dlm_debug
= kmalloc(sizeof(struct ocfs2_dlm_debug
), GFP_KERNEL
);
1996 mlog_errno(-ENOMEM
);
2000 kref_init(&dlm_debug
->d_refcnt
);
2001 INIT_LIST_HEAD(&dlm_debug
->d_lockres_tracking
);
2002 dlm_debug
->d_locking_state
= NULL
;
2007 /* Access to this is arbitrated for us via seq_file->sem. */
2008 struct ocfs2_dlm_seq_priv
{
2009 struct ocfs2_dlm_debug
*p_dlm_debug
;
2010 struct ocfs2_lock_res p_iter_res
;
2011 struct ocfs2_lock_res p_tmp_res
;
2014 static struct ocfs2_lock_res
*ocfs2_dlm_next_res(struct ocfs2_lock_res
*start
,
2015 struct ocfs2_dlm_seq_priv
*priv
)
2017 struct ocfs2_lock_res
*iter
, *ret
= NULL
;
2018 struct ocfs2_dlm_debug
*dlm_debug
= priv
->p_dlm_debug
;
2020 assert_spin_locked(&ocfs2_dlm_tracking_lock
);
2022 list_for_each_entry(iter
, &start
->l_debug_list
, l_debug_list
) {
2023 /* discover the head of the list */
2024 if (&iter
->l_debug_list
== &dlm_debug
->d_lockres_tracking
) {
2025 mlog(0, "End of list found, %p\n", ret
);
2029 /* We track our "dummy" iteration lockres' by a NULL
2031 if (iter
->l_ops
!= NULL
) {
2040 static void *ocfs2_dlm_seq_start(struct seq_file
*m
, loff_t
*pos
)
2042 struct ocfs2_dlm_seq_priv
*priv
= m
->private;
2043 struct ocfs2_lock_res
*iter
;
2045 spin_lock(&ocfs2_dlm_tracking_lock
);
2046 iter
= ocfs2_dlm_next_res(&priv
->p_iter_res
, priv
);
2048 /* Since lockres' have the lifetime of their container
2049 * (which can be inodes, ocfs2_supers, etc) we want to
2050 * copy this out to a temporary lockres while still
2051 * under the spinlock. Obviously after this we can't
2052 * trust any pointers on the copy returned, but that's
2053 * ok as the information we want isn't typically held
2055 priv
->p_tmp_res
= *iter
;
2056 iter
= &priv
->p_tmp_res
;
2058 spin_unlock(&ocfs2_dlm_tracking_lock
);
2063 static void ocfs2_dlm_seq_stop(struct seq_file
*m
, void *v
)
2067 static void *ocfs2_dlm_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2069 struct ocfs2_dlm_seq_priv
*priv
= m
->private;
2070 struct ocfs2_lock_res
*iter
= v
;
2071 struct ocfs2_lock_res
*dummy
= &priv
->p_iter_res
;
2073 spin_lock(&ocfs2_dlm_tracking_lock
);
2074 iter
= ocfs2_dlm_next_res(iter
, priv
);
2075 list_del_init(&dummy
->l_debug_list
);
2077 list_add(&dummy
->l_debug_list
, &iter
->l_debug_list
);
2078 priv
->p_tmp_res
= *iter
;
2079 iter
= &priv
->p_tmp_res
;
2081 spin_unlock(&ocfs2_dlm_tracking_lock
);
2086 /* So that debugfs.ocfs2 can determine which format is being used */
2087 #define OCFS2_DLM_DEBUG_STR_VERSION 1
2088 static int ocfs2_dlm_seq_show(struct seq_file
*m
, void *v
)
2092 struct ocfs2_lock_res
*lockres
= v
;
2097 seq_printf(m
, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION
);
2099 if (lockres
->l_type
== OCFS2_LOCK_TYPE_DENTRY
)
2100 seq_printf(m
, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START
- 1,
2102 (unsigned int)ocfs2_get_dentry_lock_ino(lockres
));
2104 seq_printf(m
, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN
, lockres
->l_name
);
2106 seq_printf(m
, "%d\t"
2117 lockres
->l_unlock_action
,
2118 lockres
->l_ro_holders
,
2119 lockres
->l_ex_holders
,
2120 lockres
->l_requested
,
2121 lockres
->l_blocking
);
2123 /* Dump the raw LVB */
2124 lvb
= lockres
->l_lksb
.lvb
;
2125 for(i
= 0; i
< DLM_LVB_LEN
; i
++)
2126 seq_printf(m
, "0x%x\t", lvb
[i
]);
2129 seq_printf(m
, "\n");
2133 static struct seq_operations ocfs2_dlm_seq_ops
= {
2134 .start
= ocfs2_dlm_seq_start
,
2135 .stop
= ocfs2_dlm_seq_stop
,
2136 .next
= ocfs2_dlm_seq_next
,
2137 .show
= ocfs2_dlm_seq_show
,
2140 static int ocfs2_dlm_debug_release(struct inode
*inode
, struct file
*file
)
2142 struct seq_file
*seq
= (struct seq_file
*) file
->private_data
;
2143 struct ocfs2_dlm_seq_priv
*priv
= seq
->private;
2144 struct ocfs2_lock_res
*res
= &priv
->p_iter_res
;
2146 ocfs2_remove_lockres_tracking(res
);
2147 ocfs2_put_dlm_debug(priv
->p_dlm_debug
);
2148 return seq_release_private(inode
, file
);
2151 static int ocfs2_dlm_debug_open(struct inode
*inode
, struct file
*file
)
2154 struct ocfs2_dlm_seq_priv
*priv
;
2155 struct seq_file
*seq
;
2156 struct ocfs2_super
*osb
;
2158 priv
= kzalloc(sizeof(struct ocfs2_dlm_seq_priv
), GFP_KERNEL
);
2164 osb
= (struct ocfs2_super
*) inode
->u
.generic_ip
;
2165 ocfs2_get_dlm_debug(osb
->osb_dlm_debug
);
2166 priv
->p_dlm_debug
= osb
->osb_dlm_debug
;
2167 INIT_LIST_HEAD(&priv
->p_iter_res
.l_debug_list
);
2169 ret
= seq_open(file
, &ocfs2_dlm_seq_ops
);
2176 seq
= (struct seq_file
*) file
->private_data
;
2177 seq
->private = priv
;
2179 ocfs2_add_lockres_tracking(&priv
->p_iter_res
,
2186 static const struct file_operations ocfs2_dlm_debug_fops
= {
2187 .open
= ocfs2_dlm_debug_open
,
2188 .release
= ocfs2_dlm_debug_release
,
2190 .llseek
= seq_lseek
,
2193 static int ocfs2_dlm_init_debug(struct ocfs2_super
*osb
)
2196 struct ocfs2_dlm_debug
*dlm_debug
= osb
->osb_dlm_debug
;
2198 dlm_debug
->d_locking_state
= debugfs_create_file("locking_state",
2200 osb
->osb_debug_root
,
2202 &ocfs2_dlm_debug_fops
);
2203 if (!dlm_debug
->d_locking_state
) {
2206 "Unable to create locking state debugfs file.\n");
2210 ocfs2_get_dlm_debug(dlm_debug
);
2215 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super
*osb
)
2217 struct ocfs2_dlm_debug
*dlm_debug
= osb
->osb_dlm_debug
;
2220 debugfs_remove(dlm_debug
->d_locking_state
);
2221 ocfs2_put_dlm_debug(dlm_debug
);
2225 int ocfs2_dlm_init(struct ocfs2_super
*osb
)
2229 struct dlm_ctxt
*dlm
;
2233 status
= ocfs2_dlm_init_debug(osb
);
2239 /* launch vote thread */
2240 osb
->vote_task
= kthread_run(ocfs2_vote_thread
, osb
, "ocfs2vote");
2241 if (IS_ERR(osb
->vote_task
)) {
2242 status
= PTR_ERR(osb
->vote_task
);
2243 osb
->vote_task
= NULL
;
2248 /* used by the dlm code to make message headers unique, each
2249 * node in this domain must agree on this. */
2250 dlm_key
= crc32_le(0, osb
->uuid_str
, strlen(osb
->uuid_str
));
2252 /* for now, uuid == domain */
2253 dlm
= dlm_register_domain(osb
->uuid_str
, dlm_key
);
2255 status
= PTR_ERR(dlm
);
2260 ocfs2_super_lock_res_init(&osb
->osb_super_lockres
, osb
);
2261 ocfs2_rename_lock_res_init(&osb
->osb_rename_lockres
, osb
);
2263 dlm_register_eviction_cb(dlm
, &osb
->osb_eviction_cb
);
2270 ocfs2_dlm_shutdown_debug(osb
);
2272 kthread_stop(osb
->vote_task
);
2279 void ocfs2_dlm_shutdown(struct ocfs2_super
*osb
)
2283 dlm_unregister_eviction_cb(&osb
->osb_eviction_cb
);
2285 ocfs2_drop_osb_locks(osb
);
2287 if (osb
->vote_task
) {
2288 kthread_stop(osb
->vote_task
);
2289 osb
->vote_task
= NULL
;
2292 ocfs2_lock_res_free(&osb
->osb_super_lockres
);
2293 ocfs2_lock_res_free(&osb
->osb_rename_lockres
);
2295 dlm_unregister_domain(osb
->dlm
);
2298 ocfs2_dlm_shutdown_debug(osb
);
2303 static void ocfs2_unlock_ast_func(void *opaque
, enum dlm_status status
)
2305 struct ocfs2_lock_res
*lockres
= opaque
;
2306 unsigned long flags
;
2310 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres
->l_name
,
2311 lockres
->l_unlock_action
);
2313 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2314 /* We tried to cancel a convert request, but it was already
2315 * granted. All we want to do here is clear our unlock
2316 * state. The wake_up call done at the bottom is redundant
2317 * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
2318 * hurt anything anyway */
2319 if (status
== DLM_CANCELGRANT
&&
2320 lockres
->l_unlock_action
== OCFS2_UNLOCK_CANCEL_CONVERT
) {
2321 mlog(0, "Got cancelgrant for %s\n", lockres
->l_name
);
2323 /* We don't clear the busy flag in this case as it
2324 * should have been cleared by the ast which the dlm
2326 goto complete_unlock
;
2329 if (status
!= DLM_NORMAL
) {
2330 mlog(ML_ERROR
, "Dlm passes status %d for lock %s, "
2331 "unlock_action %d\n", status
, lockres
->l_name
,
2332 lockres
->l_unlock_action
);
2333 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2337 switch(lockres
->l_unlock_action
) {
2338 case OCFS2_UNLOCK_CANCEL_CONVERT
:
2339 mlog(0, "Cancel convert success for %s\n", lockres
->l_name
);
2340 lockres
->l_action
= OCFS2_AST_INVALID
;
2342 case OCFS2_UNLOCK_DROP_LOCK
:
2343 lockres
->l_level
= LKM_IVMODE
;
2349 lockres_clear_flags(lockres
, OCFS2_LOCK_BUSY
);
2351 lockres
->l_unlock_action
= OCFS2_UNLOCK_INVALID
;
2352 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2354 wake_up(&lockres
->l_event
);
2359 typedef void (ocfs2_pre_drop_cb_t
)(struct ocfs2_lock_res
*, void *);
2361 struct drop_lock_cb
{
2362 ocfs2_pre_drop_cb_t
*drop_func
;
2366 static int ocfs2_drop_lock(struct ocfs2_super
*osb
,
2367 struct ocfs2_lock_res
*lockres
,
2368 struct drop_lock_cb
*dcb
)
2370 enum dlm_status status
;
2371 unsigned long flags
;
2373 /* We didn't get anywhere near actually using this lockres. */
2374 if (!(lockres
->l_flags
& OCFS2_LOCK_INITIALIZED
))
2377 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2379 mlog_bug_on_msg(!(lockres
->l_flags
& OCFS2_LOCK_FREEING
),
2380 "lockres %s, flags 0x%lx\n",
2381 lockres
->l_name
, lockres
->l_flags
);
2383 while (lockres
->l_flags
& OCFS2_LOCK_BUSY
) {
2384 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2385 "%u, unlock_action = %u\n",
2386 lockres
->l_name
, lockres
->l_flags
, lockres
->l_action
,
2387 lockres
->l_unlock_action
);
2389 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2391 /* XXX: Today we just wait on any busy
2392 * locks... Perhaps we need to cancel converts in the
2394 ocfs2_wait_on_busy_lock(lockres
);
2396 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2400 dcb
->drop_func(lockres
, dcb
->drop_data
);
2402 if (lockres
->l_flags
& OCFS2_LOCK_BUSY
)
2403 mlog(ML_ERROR
, "destroying busy lock: \"%s\"\n",
2405 if (lockres
->l_flags
& OCFS2_LOCK_BLOCKED
)
2406 mlog(0, "destroying blocked lock: \"%s\"\n", lockres
->l_name
);
2408 if (!(lockres
->l_flags
& OCFS2_LOCK_ATTACHED
)) {
2409 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2413 lockres_clear_flags(lockres
, OCFS2_LOCK_ATTACHED
);
2415 /* make sure we never get here while waiting for an ast to
2417 BUG_ON(lockres
->l_action
!= OCFS2_AST_INVALID
);
2419 /* is this necessary? */
2420 lockres_or_flags(lockres
, OCFS2_LOCK_BUSY
);
2421 lockres
->l_unlock_action
= OCFS2_UNLOCK_DROP_LOCK
;
2422 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2424 mlog(0, "lock %s\n", lockres
->l_name
);
2426 status
= dlmunlock(osb
->dlm
, &lockres
->l_lksb
, LKM_VALBLK
,
2427 lockres
->l_ops
->unlock_ast
, lockres
);
2428 if (status
!= DLM_NORMAL
) {
2429 ocfs2_log_dlm_error("dlmunlock", status
, lockres
);
2430 mlog(ML_ERROR
, "lockres flags: %lu\n", lockres
->l_flags
);
2431 dlm_print_one_lock(lockres
->l_lksb
.lockid
);
2434 mlog(0, "lock %s, successfull return from dlmunlock\n",
2437 ocfs2_wait_on_busy_lock(lockres
);
2443 /* Mark the lockres as being dropped. It will no longer be
2444 * queued if blocking, but we still may have to wait on it
2445 * being dequeued from the vote thread before we can consider
2448 * You can *not* attempt to call cluster_lock on this lockres anymore. */
2449 void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res
*lockres
)
2452 struct ocfs2_mask_waiter mw
;
2453 unsigned long flags
;
2455 ocfs2_init_mask_waiter(&mw
);
2457 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2458 lockres
->l_flags
|= OCFS2_LOCK_FREEING
;
2459 while (lockres
->l_flags
& OCFS2_LOCK_QUEUED
) {
2460 lockres_add_mask_waiter(lockres
, &mw
, OCFS2_LOCK_QUEUED
, 0);
2461 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2463 mlog(0, "Waiting on lockres %s\n", lockres
->l_name
);
2465 status
= ocfs2_wait_for_mask(&mw
);
2469 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2471 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2474 void ocfs2_simple_drop_lockres(struct ocfs2_super
*osb
,
2475 struct ocfs2_lock_res
*lockres
)
2479 ocfs2_mark_lockres_freeing(lockres
);
2480 ret
= ocfs2_drop_lock(osb
, lockres
, NULL
);
2485 static void ocfs2_drop_osb_locks(struct ocfs2_super
*osb
)
2487 ocfs2_simple_drop_lockres(osb
, &osb
->osb_super_lockres
);
2488 ocfs2_simple_drop_lockres(osb
, &osb
->osb_rename_lockres
);
2491 static void ocfs2_meta_pre_drop(struct ocfs2_lock_res
*lockres
, void *data
)
2493 struct inode
*inode
= data
;
2495 /* the metadata lock requires a bit more work as we have an
2496 * LVB to worry about. */
2497 if (lockres
->l_flags
& OCFS2_LOCK_ATTACHED
&&
2498 lockres
->l_level
== LKM_EXMODE
&&
2499 !(lockres
->l_flags
& OCFS2_LOCK_NEEDS_REFRESH
))
2500 __ocfs2_stuff_meta_lvb(inode
);
2503 int ocfs2_drop_inode_locks(struct inode
*inode
)
2506 struct drop_lock_cb meta_dcb
= { ocfs2_meta_pre_drop
, inode
, };
2510 /* No need to call ocfs2_mark_lockres_freeing here -
2511 * ocfs2_clear_inode has done it for us. */
2513 err
= ocfs2_drop_lock(OCFS2_SB(inode
->i_sb
),
2514 &OCFS2_I(inode
)->ip_data_lockres
,
2521 err
= ocfs2_drop_lock(OCFS2_SB(inode
->i_sb
),
2522 &OCFS2_I(inode
)->ip_meta_lockres
,
2526 if (err
< 0 && !status
)
2529 err
= ocfs2_drop_lock(OCFS2_SB(inode
->i_sb
),
2530 &OCFS2_I(inode
)->ip_rw_lockres
,
2534 if (err
< 0 && !status
)
2541 static void ocfs2_prepare_downconvert(struct ocfs2_lock_res
*lockres
,
2544 assert_spin_locked(&lockres
->l_lock
);
2546 BUG_ON(lockres
->l_blocking
<= LKM_NLMODE
);
2548 if (lockres
->l_level
<= new_level
) {
2549 mlog(ML_ERROR
, "lockres->l_level (%u) <= new_level (%u)\n",
2550 lockres
->l_level
, new_level
);
2554 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2555 lockres
->l_name
, new_level
, lockres
->l_blocking
);
2557 lockres
->l_action
= OCFS2_AST_DOWNCONVERT
;
2558 lockres
->l_requested
= new_level
;
2559 lockres_or_flags(lockres
, OCFS2_LOCK_BUSY
);
2562 static int ocfs2_downconvert_lock(struct ocfs2_super
*osb
,
2563 struct ocfs2_lock_res
*lockres
,
2567 int ret
, dlm_flags
= LKM_CONVERT
;
2568 enum dlm_status status
;
2573 dlm_flags
|= LKM_VALBLK
;
2575 status
= dlmlock(osb
->dlm
,
2580 OCFS2_LOCK_ID_MAX_LEN
- 1,
2581 lockres
->l_ops
->ast
,
2583 lockres
->l_ops
->bast
);
2584 if (status
!= DLM_NORMAL
) {
2585 ocfs2_log_dlm_error("dlmlock", status
, lockres
);
2587 ocfs2_recover_from_dlm_error(lockres
, 1);
2597 /* returns 1 when the caller should unlock and call dlmunlock */
2598 static int ocfs2_prepare_cancel_convert(struct ocfs2_super
*osb
,
2599 struct ocfs2_lock_res
*lockres
)
2601 assert_spin_locked(&lockres
->l_lock
);
2604 mlog(0, "lock %s\n", lockres
->l_name
);
2606 if (lockres
->l_unlock_action
== OCFS2_UNLOCK_CANCEL_CONVERT
) {
2607 /* If we're already trying to cancel a lock conversion
2608 * then just drop the spinlock and allow the caller to
2609 * requeue this lock. */
2611 mlog(0, "Lockres %s, skip convert\n", lockres
->l_name
);
2615 /* were we in a convert when we got the bast fire? */
2616 BUG_ON(lockres
->l_action
!= OCFS2_AST_CONVERT
&&
2617 lockres
->l_action
!= OCFS2_AST_DOWNCONVERT
);
2618 /* set things up for the unlockast to know to just
2619 * clear out the ast_action and unset busy, etc. */
2620 lockres
->l_unlock_action
= OCFS2_UNLOCK_CANCEL_CONVERT
;
2622 mlog_bug_on_msg(!(lockres
->l_flags
& OCFS2_LOCK_BUSY
),
2623 "lock %s, invalid flags: 0x%lx\n",
2624 lockres
->l_name
, lockres
->l_flags
);
2629 static int ocfs2_cancel_convert(struct ocfs2_super
*osb
,
2630 struct ocfs2_lock_res
*lockres
)
2633 enum dlm_status status
;
2636 mlog(0, "lock %s\n", lockres
->l_name
);
2639 status
= dlmunlock(osb
->dlm
,
2642 lockres
->l_ops
->unlock_ast
,
2644 if (status
!= DLM_NORMAL
) {
2645 ocfs2_log_dlm_error("dlmunlock", status
, lockres
);
2647 ocfs2_recover_from_dlm_error(lockres
, 0);
2650 mlog(0, "lock %s return from dlmunlock\n", lockres
->l_name
);
2656 static inline int ocfs2_can_downconvert_meta_lock(struct inode
*inode
,
2657 struct ocfs2_lock_res
*lockres
,
2664 BUG_ON(new_level
!= LKM_NLMODE
&& new_level
!= LKM_PRMODE
);
2666 if (lockres
->l_flags
& OCFS2_LOCK_REFRESHING
) {
2668 mlog(0, "lockres %s currently being refreshed -- backing "
2669 "off!\n", lockres
->l_name
);
2670 } else if (new_level
== LKM_PRMODE
)
2671 ret
= !lockres
->l_ex_holders
&&
2672 ocfs2_inode_fully_checkpointed(inode
);
2673 else /* Must be NLMODE we're converting to. */
2674 ret
= !lockres
->l_ro_holders
&& !lockres
->l_ex_holders
&&
2675 ocfs2_inode_fully_checkpointed(inode
);
2681 static int ocfs2_do_unblock_meta(struct inode
*inode
,
2687 struct ocfs2_lock_res
*lockres
= &OCFS2_I(inode
)->ip_meta_lockres
;
2688 unsigned long flags
;
2690 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2694 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2696 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_BLOCKED
));
2698 mlog(0, "l_level=%d, l_blocking=%d\n", lockres
->l_level
,
2699 lockres
->l_blocking
);
2701 BUG_ON(lockres
->l_level
!= LKM_EXMODE
&&
2702 lockres
->l_level
!= LKM_PRMODE
);
2704 if (lockres
->l_flags
& OCFS2_LOCK_BUSY
) {
2706 ret
= ocfs2_prepare_cancel_convert(osb
, lockres
);
2707 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2709 ret
= ocfs2_cancel_convert(osb
, lockres
);
2716 new_level
= ocfs2_highest_compat_lock_level(lockres
->l_blocking
);
2718 mlog(0, "l_level=%d, l_blocking=%d, new_level=%d\n",
2719 lockres
->l_level
, lockres
->l_blocking
, new_level
);
2721 if (ocfs2_can_downconvert_meta_lock(inode
, lockres
, new_level
)) {
2722 if (lockres
->l_level
== LKM_EXMODE
)
2725 /* If the lock hasn't been refreshed yet (rare), then
2726 * our memory inode values are old and we skip
2727 * stuffing the lvb. There's no need to actually clear
2728 * out the lvb here as it's value is still valid. */
2729 if (!(lockres
->l_flags
& OCFS2_LOCK_NEEDS_REFRESH
)) {
2731 __ocfs2_stuff_meta_lvb(inode
);
2733 mlog(0, "lockres %s: downconverting stale lock!\n",
2736 mlog(0, "calling ocfs2_downconvert_lock with l_level=%d, "
2737 "l_blocking=%d, new_level=%d\n",
2738 lockres
->l_level
, lockres
->l_blocking
, new_level
);
2740 ocfs2_prepare_downconvert(lockres
, new_level
);
2741 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2742 ret
= ocfs2_downconvert_lock(osb
, lockres
, new_level
, set_lvb
);
2745 if (!ocfs2_inode_fully_checkpointed(inode
))
2746 ocfs2_start_checkpoint(osb
);
2749 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2756 static int ocfs2_generic_unblock_lock(struct ocfs2_super
*osb
,
2757 struct ocfs2_lock_res
*lockres
,
2758 struct ocfs2_unblock_ctl
*ctl
,
2759 ocfs2_convert_worker_t
*worker
)
2761 unsigned long flags
;
2768 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2770 BUG_ON(!(lockres
->l_flags
& OCFS2_LOCK_BLOCKED
));
2773 if (lockres
->l_flags
& OCFS2_LOCK_BUSY
) {
2775 ret
= ocfs2_prepare_cancel_convert(osb
, lockres
);
2776 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2778 ret
= ocfs2_cancel_convert(osb
, lockres
);
2785 /* if we're blocking an exclusive and we have *any* holders,
2787 if ((lockres
->l_blocking
== LKM_EXMODE
)
2788 && (lockres
->l_ex_holders
|| lockres
->l_ro_holders
)) {
2789 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2795 /* If it's a PR we're blocking, then only
2796 * requeue if we've got any EX holders */
2797 if (lockres
->l_blocking
== LKM_PRMODE
&&
2798 lockres
->l_ex_holders
) {
2799 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2805 /* If we get here, then we know that there are no more
2806 * incompatible holders (and anyone asking for an incompatible
2807 * lock is blocked). We can now downconvert the lock */
2811 /* Some lockres types want to do a bit of work before
2812 * downconverting a lock. Allow that here. The worker function
2813 * may sleep, so we save off a copy of what we're blocking as
2814 * it may change while we're not holding the spin lock. */
2815 blocking
= lockres
->l_blocking
;
2816 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2818 ctl
->unblock_action
= worker(lockres
, blocking
);
2820 if (ctl
->unblock_action
== UNBLOCK_STOP_POST
)
2823 spin_lock_irqsave(&lockres
->l_lock
, flags
);
2824 if (blocking
!= lockres
->l_blocking
) {
2825 /* If this changed underneath us, then we can't drop
2832 new_level
= ocfs2_highest_compat_lock_level(lockres
->l_blocking
);
2834 ocfs2_prepare_downconvert(lockres
, new_level
);
2835 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
2836 ret
= ocfs2_downconvert_lock(osb
, lockres
, new_level
, 0);
2842 static int ocfs2_data_convert_worker(struct ocfs2_lock_res
*lockres
,
2845 struct inode
*inode
;
2846 struct address_space
*mapping
;
2848 inode
= ocfs2_lock_res_inode(lockres
);
2849 mapping
= inode
->i_mapping
;
2851 if (filemap_fdatawrite(mapping
)) {
2852 mlog(ML_ERROR
, "Could not sync inode %llu for downconvert!",
2853 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2855 sync_mapping_buffers(mapping
);
2856 if (blocking
== LKM_EXMODE
) {
2857 truncate_inode_pages(mapping
, 0);
2858 unmap_mapping_range(mapping
, 0, 0, 0);
2860 /* We only need to wait on the I/O if we're not also
2861 * truncating pages because truncate_inode_pages waits
2862 * for us above. We don't truncate pages if we're
2863 * blocking anything < EXMODE because we want to keep
2864 * them around in that case. */
2865 filemap_fdatawait(mapping
);
2868 return UNBLOCK_CONTINUE
;
2871 int ocfs2_unblock_data(struct ocfs2_lock_res
*lockres
,
2872 struct ocfs2_unblock_ctl
*ctl
)
2875 struct inode
*inode
;
2876 struct ocfs2_super
*osb
;
2880 inode
= ocfs2_lock_res_inode(lockres
);
2881 osb
= OCFS2_SB(inode
->i_sb
);
2883 mlog(0, "unblock inode %llu\n",
2884 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2886 status
= ocfs2_generic_unblock_lock(osb
, lockres
, ctl
,
2887 ocfs2_data_convert_worker
);
2891 mlog(0, "inode %llu, requeue = %d\n",
2892 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, ctl
->requeue
);
2898 static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res
*lockres
,
2899 struct ocfs2_unblock_ctl
*ctl
)
2902 struct inode
*inode
;
2906 mlog(0, "Unblock lockres %s\n", lockres
->l_name
);
2908 inode
= ocfs2_lock_res_inode(lockres
);
2910 status
= ocfs2_generic_unblock_lock(OCFS2_SB(inode
->i_sb
),
2911 lockres
, ctl
, NULL
);
2919 static int ocfs2_unblock_meta(struct ocfs2_lock_res
*lockres
,
2920 struct ocfs2_unblock_ctl
*ctl
)
2923 struct inode
*inode
;
2927 inode
= ocfs2_lock_res_inode(lockres
);
2929 mlog(0, "unblock inode %llu\n",
2930 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2932 status
= ocfs2_do_unblock_meta(inode
, &ctl
->requeue
);
2936 mlog(0, "inode %llu, requeue = %d\n",
2937 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, ctl
->requeue
);
2944 * Does the final reference drop on our dentry lock. Right now this
2945 * happens in the vote thread, but we could choose to simplify the
2946 * dlmglue API and push these off to the ocfs2_wq in the future.
2948 static void ocfs2_dentry_post_unlock(struct ocfs2_super
*osb
,
2949 struct ocfs2_lock_res
*lockres
)
2951 struct ocfs2_dentry_lock
*dl
= ocfs2_lock_res_dl(lockres
);
2952 ocfs2_dentry_lock_put(osb
, dl
);
2956 * d_delete() matching dentries before the lock downconvert.
2958 * At this point, any process waiting to destroy the
2959 * dentry_lock due to last ref count is stopped by the
2960 * OCFS2_LOCK_QUEUED flag.
2962 * We have two potential problems
2964 * 1) If we do the last reference drop on our dentry_lock (via dput)
2965 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
2966 * the downconvert to finish. Instead we take an elevated
2967 * reference and push the drop until after we've completed our
2968 * unblock processing.
2970 * 2) There might be another process with a final reference,
2971 * waiting on us to finish processing. If this is the case, we
2972 * detect it and exit out - there's no more dentries anyway.
2974 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res
*lockres
,
2977 struct ocfs2_dentry_lock
*dl
= ocfs2_lock_res_dl(lockres
);
2978 struct ocfs2_inode_info
*oi
= OCFS2_I(dl
->dl_inode
);
2979 struct dentry
*dentry
;
2980 unsigned long flags
;
2984 * This node is blocking another node from getting a read
2985 * lock. This happens when we've renamed within a
2986 * directory. We've forced the other nodes to d_delete(), but
2987 * we never actually dropped our lock because it's still
2988 * valid. The downconvert code will retain a PR for this node,
2989 * so there's no further work to do.
2991 if (blocking
== LKM_PRMODE
)
2992 return UNBLOCK_CONTINUE
;
2995 * Mark this inode as potentially orphaned. The code in
2996 * ocfs2_delete_inode() will figure out whether it actually
2997 * needs to be freed or not.
2999 spin_lock(&oi
->ip_lock
);
3000 oi
->ip_flags
|= OCFS2_INODE_MAYBE_ORPHANED
;
3001 spin_unlock(&oi
->ip_lock
);
3004 * Yuck. We need to make sure however that the check of
3005 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3006 * respect to a reference decrement or the setting of that
3009 spin_lock_irqsave(&lockres
->l_lock
, flags
);
3010 spin_lock(&dentry_attach_lock
);
3011 if (!(lockres
->l_flags
& OCFS2_LOCK_FREEING
)
3016 spin_unlock(&dentry_attach_lock
);
3017 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
3019 mlog(0, "extra_ref = %d\n", extra_ref
);
3022 * We have a process waiting on us in ocfs2_dentry_iput(),
3023 * which means we can't have any more outstanding
3024 * aliases. There's no need to do any more work.
3027 return UNBLOCK_CONTINUE
;
3029 spin_lock(&dentry_attach_lock
);
3031 dentry
= ocfs2_find_local_alias(dl
->dl_inode
,
3032 dl
->dl_parent_blkno
, 1);
3035 spin_unlock(&dentry_attach_lock
);
3037 mlog(0, "d_delete(%.*s);\n", dentry
->d_name
.len
,
3038 dentry
->d_name
.name
);
3041 * The following dcache calls may do an
3042 * iput(). Normally we don't want that from the
3043 * downconverting thread, but in this case it's ok
3044 * because the requesting node already has an
3045 * exclusive lock on the inode, so it can't be queued
3046 * for a downconvert.
3051 spin_lock(&dentry_attach_lock
);
3053 spin_unlock(&dentry_attach_lock
);
3056 * If we are the last holder of this dentry lock, there is no
3057 * reason to downconvert so skip straight to the unlock.
3059 if (dl
->dl_count
== 1)
3060 return UNBLOCK_STOP_POST
;
3062 return UNBLOCK_CONTINUE_POST
;
3065 static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res
*lockres
,
3066 struct ocfs2_unblock_ctl
*ctl
)
3069 struct ocfs2_dentry_lock
*dl
= ocfs2_lock_res_dl(lockres
);
3070 struct ocfs2_super
*osb
= OCFS2_SB(dl
->dl_inode
->i_sb
);
3072 mlog(0, "unblock dentry lock: %llu\n",
3073 (unsigned long long)OCFS2_I(dl
->dl_inode
)->ip_blkno
);
3075 ret
= ocfs2_generic_unblock_lock(osb
,
3078 ocfs2_dentry_convert_worker
);
3082 mlog(0, "requeue = %d, post = %d\n", ctl
->requeue
, ctl
->unblock_action
);
3087 /* Generic unblock function for any lockres whose private data is an
3088 * ocfs2_super pointer. */
3089 static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res
*lockres
,
3090 struct ocfs2_unblock_ctl
*ctl
)
3093 struct ocfs2_super
*osb
;
3097 mlog(0, "Unblock lockres %s\n", lockres
->l_name
);
3099 osb
= ocfs2_lock_res_super(lockres
);
3101 status
= ocfs2_generic_unblock_lock(osb
,
3112 void ocfs2_process_blocked_lock(struct ocfs2_super
*osb
,
3113 struct ocfs2_lock_res
*lockres
)
3116 struct ocfs2_unblock_ctl ctl
= {0, 0,};
3117 unsigned long flags
;
3119 /* Our reference to the lockres in this function can be
3120 * considered valid until we remove the OCFS2_LOCK_QUEUED
3126 BUG_ON(!lockres
->l_ops
);
3127 BUG_ON(!lockres
->l_ops
->unblock
);
3129 mlog(0, "lockres %s blocked.\n", lockres
->l_name
);
3131 /* Detect whether a lock has been marked as going away while
3132 * the vote thread was processing other things. A lock can
3133 * still be marked with OCFS2_LOCK_FREEING after this check,
3134 * but short circuiting here will still save us some
3136 spin_lock_irqsave(&lockres
->l_lock
, flags
);
3137 if (lockres
->l_flags
& OCFS2_LOCK_FREEING
)
3139 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
3141 status
= lockres
->l_ops
->unblock(lockres
, &ctl
);
3145 spin_lock_irqsave(&lockres
->l_lock
, flags
);
3147 if (lockres
->l_flags
& OCFS2_LOCK_FREEING
|| !ctl
.requeue
) {
3148 lockres_clear_flags(lockres
, OCFS2_LOCK_QUEUED
);
3150 ocfs2_schedule_blocked_lock(osb
, lockres
);
3152 mlog(0, "lockres %s, requeue = %s.\n", lockres
->l_name
,
3153 ctl
.requeue
? "yes" : "no");
3154 spin_unlock_irqrestore(&lockres
->l_lock
, flags
);
3156 if (ctl
.unblock_action
!= UNBLOCK_CONTINUE
3157 && lockres
->l_ops
->post_unlock
)
3158 lockres
->l_ops
->post_unlock(osb
, lockres
);
3163 static void ocfs2_schedule_blocked_lock(struct ocfs2_super
*osb
,
3164 struct ocfs2_lock_res
*lockres
)
3168 assert_spin_locked(&lockres
->l_lock
);
3170 if (lockres
->l_flags
& OCFS2_LOCK_FREEING
) {
3171 /* Do not schedule a lock for downconvert when it's on
3172 * the way to destruction - any nodes wanting access
3173 * to the resource will get it soon. */
3174 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3175 lockres
->l_name
, lockres
->l_flags
);
3179 lockres_or_flags(lockres
, OCFS2_LOCK_QUEUED
);
3181 spin_lock(&osb
->vote_task_lock
);
3182 if (list_empty(&lockres
->l_blocked_list
)) {
3183 list_add_tail(&lockres
->l_blocked_list
,
3184 &osb
->blocked_lock_list
);
3185 osb
->blocked_lock_count
++;
3187 spin_unlock(&osb
->vote_task_lock
);
3192 /* This aids in debugging situations where a bad LVB might be involved. */
3193 void ocfs2_dump_meta_lvb_info(u64 level
,
3194 const char *function
,
3196 struct ocfs2_lock_res
*lockres
)
3198 struct ocfs2_meta_lvb
*lvb
= (struct ocfs2_meta_lvb
*) lockres
->l_lksb
.lvb
;
3200 mlog(level
, "LVB information for %s (called from %s:%u):\n",
3201 lockres
->l_name
, function
, line
);
3202 mlog(level
, "version: %u, clusters: %u, generation: 0x%x\n",
3203 lvb
->lvb_version
, be32_to_cpu(lvb
->lvb_iclusters
),
3204 be32_to_cpu(lvb
->lvb_igeneration
));
3205 mlog(level
, "size: %llu, uid %u, gid %u, mode 0x%x\n",
3206 (unsigned long long)be64_to_cpu(lvb
->lvb_isize
),
3207 be32_to_cpu(lvb
->lvb_iuid
), be32_to_cpu(lvb
->lvb_igid
),
3208 be16_to_cpu(lvb
->lvb_imode
));
3209 mlog(level
, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
3210 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb
->lvb_inlink
),
3211 (long long)be64_to_cpu(lvb
->lvb_iatime_packed
),
3212 (long long)be64_to_cpu(lvb
->lvb_ictime_packed
),
3213 (long long)be64_to_cpu(lvb
->lvb_imtime_packed
),
3214 be32_to_cpu(lvb
->lvb_iattr
));