1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) 2005-2007 Red Hat, Inc. All rights reserved.
6 ** This copyrighted material is made available to anyone wishing to use,
7 ** modify, copy, or redistribute it subject to the terms and conditions
8 ** of the GNU General Public License v.2.
10 *******************************************************************************
11 ******************************************************************************/
13 /* Central locking logic has four stages:
33 Stage 1 (lock, unlock) is mainly about checking input args and
34 splitting into one of the four main operations:
36 dlm_lock = request_lock
37 dlm_lock+CONVERT = convert_lock
38 dlm_unlock = unlock_lock
39 dlm_unlock+CANCEL = cancel_lock
41 Stage 2, xxxx_lock(), just finds and locks the relevant rsb which is
42 provided to the next stage.
44 Stage 3, _xxxx_lock(), determines if the operation is local or remote.
45 When remote, it calls send_xxxx(), when local it calls do_xxxx().
47 Stage 4, do_xxxx(), is the guts of the operation. It manipulates the
48 given rsb and lkb and queues callbacks.
50 For remote operations, send_xxxx() results in the corresponding do_xxxx()
51 function being executed on the remote node. The connecting send/receive
52 calls on local (L) and remote (R) nodes:
54 L: send_xxxx() -> R: receive_xxxx()
56 L: receive_xxxx_reply() <- R: send_xxxx_reply()
58 #include <linux/types.h>
59 #include "dlm_internal.h"
60 #include <linux/dlm_device.h>
63 #include "requestqueue.h"
67 #include "lockspace.h"
72 #include "lvb_table.h"
76 static int send_request(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
77 static int send_convert(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
78 static int send_unlock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
79 static int send_cancel(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
80 static int send_grant(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
81 static int send_bast(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int mode
);
82 static int send_lookup(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
83 static int send_remove(struct dlm_rsb
*r
);
84 static int _request_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
85 static int _cancel_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
);
86 static void __receive_convert_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
87 struct dlm_message
*ms
);
88 static int receive_extralen(struct dlm_message
*ms
);
89 static void do_purge(struct dlm_ls
*ls
, int nodeid
, int pid
);
90 static void del_timeout(struct dlm_lkb
*lkb
);
93 * Lock compatibilty matrix - thanks Steve
94 * UN = Unlocked state. Not really a state, used as a flag
95 * PD = Padding. Used to make the matrix a nice power of two in size
96 * Other states are the same as the VMS DLM.
97 * Usage: matrix[grmode+1][rqmode+1] (although m[rq+1][gr+1] is the same)
100 static const int __dlm_compat_matrix
[8][8] = {
101 /* UN NL CR CW PR PW EX PD */
102 {1, 1, 1, 1, 1, 1, 1, 0}, /* UN */
103 {1, 1, 1, 1, 1, 1, 1, 0}, /* NL */
104 {1, 1, 1, 1, 1, 1, 0, 0}, /* CR */
105 {1, 1, 1, 1, 0, 0, 0, 0}, /* CW */
106 {1, 1, 1, 0, 1, 0, 0, 0}, /* PR */
107 {1, 1, 1, 0, 0, 0, 0, 0}, /* PW */
108 {1, 1, 0, 0, 0, 0, 0, 0}, /* EX */
109 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
113 * This defines the direction of transfer of LVB data.
114 * Granted mode is the row; requested mode is the column.
115 * Usage: matrix[grmode+1][rqmode+1]
116 * 1 = LVB is returned to the caller
117 * 0 = LVB is written to the resource
118 * -1 = nothing happens to the LVB
121 const int dlm_lvb_operations
[8][8] = {
122 /* UN NL CR CW PR PW EX PD*/
123 { -1, 1, 1, 1, 1, 1, 1, -1 }, /* UN */
124 { -1, 1, 1, 1, 1, 1, 1, 0 }, /* NL */
125 { -1, -1, 1, 1, 1, 1, 1, 0 }, /* CR */
126 { -1, -1, -1, 1, 1, 1, 1, 0 }, /* CW */
127 { -1, -1, -1, -1, 1, 1, 1, 0 }, /* PR */
128 { -1, 0, 0, 0, 0, 0, 1, 0 }, /* PW */
129 { -1, 0, 0, 0, 0, 0, 0, 0 }, /* EX */
130 { -1, 0, 0, 0, 0, 0, 0, 0 } /* PD */
133 #define modes_compat(gr, rq) \
134 __dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
136 int dlm_modes_compat(int mode1
, int mode2
)
138 return __dlm_compat_matrix
[mode1
+ 1][mode2
+ 1];
142 * Compatibility matrix for conversions with QUECVT set.
143 * Granted mode is the row; requested mode is the column.
144 * Usage: matrix[grmode+1][rqmode+1]
147 static const int __quecvt_compat_matrix
[8][8] = {
148 /* UN NL CR CW PR PW EX PD */
149 {0, 0, 0, 0, 0, 0, 0, 0}, /* UN */
150 {0, 0, 1, 1, 1, 1, 1, 0}, /* NL */
151 {0, 0, 0, 1, 1, 1, 1, 0}, /* CR */
152 {0, 0, 0, 0, 1, 1, 1, 0}, /* CW */
153 {0, 0, 0, 1, 0, 1, 1, 0}, /* PR */
154 {0, 0, 0, 0, 0, 0, 1, 0}, /* PW */
155 {0, 0, 0, 0, 0, 0, 0, 0}, /* EX */
156 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
159 void dlm_print_lkb(struct dlm_lkb
*lkb
)
161 printk(KERN_ERR
"lkb: nodeid %d id %x remid %x exflags %x flags %x\n"
162 " status %d rqmode %d grmode %d wait_type %d ast_type %d\n",
163 lkb
->lkb_nodeid
, lkb
->lkb_id
, lkb
->lkb_remid
, lkb
->lkb_exflags
,
164 lkb
->lkb_flags
, lkb
->lkb_status
, lkb
->lkb_rqmode
,
165 lkb
->lkb_grmode
, lkb
->lkb_wait_type
, lkb
->lkb_ast_type
);
168 void dlm_print_rsb(struct dlm_rsb
*r
)
170 printk(KERN_ERR
"rsb: nodeid %d flags %lx first %x rlc %d name %s\n",
171 r
->res_nodeid
, r
->res_flags
, r
->res_first_lkid
,
172 r
->res_recover_locks_count
, r
->res_name
);
175 void dlm_dump_rsb(struct dlm_rsb
*r
)
181 printk(KERN_ERR
"rsb: root_list empty %d recover_list empty %d\n",
182 list_empty(&r
->res_root_list
), list_empty(&r
->res_recover_list
));
183 printk(KERN_ERR
"rsb lookup list\n");
184 list_for_each_entry(lkb
, &r
->res_lookup
, lkb_rsb_lookup
)
186 printk(KERN_ERR
"rsb grant queue:\n");
187 list_for_each_entry(lkb
, &r
->res_grantqueue
, lkb_statequeue
)
189 printk(KERN_ERR
"rsb convert queue:\n");
190 list_for_each_entry(lkb
, &r
->res_convertqueue
, lkb_statequeue
)
192 printk(KERN_ERR
"rsb wait queue:\n");
193 list_for_each_entry(lkb
, &r
->res_waitqueue
, lkb_statequeue
)
197 /* Threads cannot use the lockspace while it's being recovered */
199 static inline void dlm_lock_recovery(struct dlm_ls
*ls
)
201 down_read(&ls
->ls_in_recovery
);
204 void dlm_unlock_recovery(struct dlm_ls
*ls
)
206 up_read(&ls
->ls_in_recovery
);
209 int dlm_lock_recovery_try(struct dlm_ls
*ls
)
211 return down_read_trylock(&ls
->ls_in_recovery
);
214 static inline int can_be_queued(struct dlm_lkb
*lkb
)
216 return !(lkb
->lkb_exflags
& DLM_LKF_NOQUEUE
);
219 static inline int force_blocking_asts(struct dlm_lkb
*lkb
)
221 return (lkb
->lkb_exflags
& DLM_LKF_NOQUEUEBAST
);
224 static inline int is_demoted(struct dlm_lkb
*lkb
)
226 return (lkb
->lkb_sbflags
& DLM_SBF_DEMOTED
);
229 static inline int is_altmode(struct dlm_lkb
*lkb
)
231 return (lkb
->lkb_sbflags
& DLM_SBF_ALTMODE
);
234 static inline int is_granted(struct dlm_lkb
*lkb
)
236 return (lkb
->lkb_status
== DLM_LKSTS_GRANTED
);
239 static inline int is_remote(struct dlm_rsb
*r
)
241 DLM_ASSERT(r
->res_nodeid
>= 0, dlm_print_rsb(r
););
242 return !!r
->res_nodeid
;
245 static inline int is_process_copy(struct dlm_lkb
*lkb
)
247 return (lkb
->lkb_nodeid
&& !(lkb
->lkb_flags
& DLM_IFL_MSTCPY
));
250 static inline int is_master_copy(struct dlm_lkb
*lkb
)
252 if (lkb
->lkb_flags
& DLM_IFL_MSTCPY
)
253 DLM_ASSERT(lkb
->lkb_nodeid
, dlm_print_lkb(lkb
););
254 return (lkb
->lkb_flags
& DLM_IFL_MSTCPY
) ? 1 : 0;
257 static inline int middle_conversion(struct dlm_lkb
*lkb
)
259 if ((lkb
->lkb_grmode
==DLM_LOCK_PR
&& lkb
->lkb_rqmode
==DLM_LOCK_CW
) ||
260 (lkb
->lkb_rqmode
==DLM_LOCK_PR
&& lkb
->lkb_grmode
==DLM_LOCK_CW
))
265 static inline int down_conversion(struct dlm_lkb
*lkb
)
267 return (!middle_conversion(lkb
) && lkb
->lkb_rqmode
< lkb
->lkb_grmode
);
270 static inline int is_overlap_unlock(struct dlm_lkb
*lkb
)
272 return lkb
->lkb_flags
& DLM_IFL_OVERLAP_UNLOCK
;
275 static inline int is_overlap_cancel(struct dlm_lkb
*lkb
)
277 return lkb
->lkb_flags
& DLM_IFL_OVERLAP_CANCEL
;
280 static inline int is_overlap(struct dlm_lkb
*lkb
)
282 return (lkb
->lkb_flags
& (DLM_IFL_OVERLAP_UNLOCK
|
283 DLM_IFL_OVERLAP_CANCEL
));
286 static void queue_cast(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int rv
)
288 if (is_master_copy(lkb
))
293 DLM_ASSERT(lkb
->lkb_lksb
, dlm_print_lkb(lkb
););
295 /* if the operation was a cancel, then return -DLM_ECANCEL, if a
296 timeout caused the cancel then return -ETIMEDOUT */
297 if (rv
== -DLM_ECANCEL
&& (lkb
->lkb_flags
& DLM_IFL_TIMEOUT_CANCEL
)) {
298 lkb
->lkb_flags
&= ~DLM_IFL_TIMEOUT_CANCEL
;
302 if (rv
== -DLM_ECANCEL
&& (lkb
->lkb_flags
& DLM_IFL_DEADLOCK_CANCEL
)) {
303 lkb
->lkb_flags
&= ~DLM_IFL_DEADLOCK_CANCEL
;
307 lkb
->lkb_lksb
->sb_status
= rv
;
308 lkb
->lkb_lksb
->sb_flags
= lkb
->lkb_sbflags
;
310 dlm_add_ast(lkb
, AST_COMP
);
313 static inline void queue_cast_overlap(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
316 is_overlap_unlock(lkb
) ? -DLM_EUNLOCK
: -DLM_ECANCEL
);
319 static void queue_bast(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int rqmode
)
321 if (is_master_copy(lkb
))
322 send_bast(r
, lkb
, rqmode
);
324 lkb
->lkb_bastmode
= rqmode
;
325 dlm_add_ast(lkb
, AST_BAST
);
330 * Basic operations on rsb's and lkb's
333 static struct dlm_rsb
*create_rsb(struct dlm_ls
*ls
, char *name
, int len
)
337 r
= dlm_allocate_rsb(ls
, len
);
343 memcpy(r
->res_name
, name
, len
);
344 mutex_init(&r
->res_mutex
);
346 INIT_LIST_HEAD(&r
->res_lookup
);
347 INIT_LIST_HEAD(&r
->res_grantqueue
);
348 INIT_LIST_HEAD(&r
->res_convertqueue
);
349 INIT_LIST_HEAD(&r
->res_waitqueue
);
350 INIT_LIST_HEAD(&r
->res_root_list
);
351 INIT_LIST_HEAD(&r
->res_recover_list
);
356 static int search_rsb_list(struct list_head
*head
, char *name
, int len
,
357 unsigned int flags
, struct dlm_rsb
**r_ret
)
362 list_for_each_entry(r
, head
, res_hashchain
) {
363 if (len
== r
->res_length
&& !memcmp(name
, r
->res_name
, len
))
369 if (r
->res_nodeid
&& (flags
& R_MASTER
))
375 static int _search_rsb(struct dlm_ls
*ls
, char *name
, int len
, int b
,
376 unsigned int flags
, struct dlm_rsb
**r_ret
)
381 error
= search_rsb_list(&ls
->ls_rsbtbl
[b
].list
, name
, len
, flags
, &r
);
383 kref_get(&r
->res_ref
);
386 error
= search_rsb_list(&ls
->ls_rsbtbl
[b
].toss
, name
, len
, flags
, &r
);
390 list_move(&r
->res_hashchain
, &ls
->ls_rsbtbl
[b
].list
);
392 if (dlm_no_directory(ls
))
395 if (r
->res_nodeid
== -1) {
396 rsb_clear_flag(r
, RSB_MASTER_UNCERTAIN
);
397 r
->res_first_lkid
= 0;
398 } else if (r
->res_nodeid
> 0) {
399 rsb_set_flag(r
, RSB_MASTER_UNCERTAIN
);
400 r
->res_first_lkid
= 0;
402 DLM_ASSERT(r
->res_nodeid
== 0, dlm_print_rsb(r
););
403 DLM_ASSERT(!rsb_flag(r
, RSB_MASTER_UNCERTAIN
),);
410 static int search_rsb(struct dlm_ls
*ls
, char *name
, int len
, int b
,
411 unsigned int flags
, struct dlm_rsb
**r_ret
)
414 write_lock(&ls
->ls_rsbtbl
[b
].lock
);
415 error
= _search_rsb(ls
, name
, len
, b
, flags
, r_ret
);
416 write_unlock(&ls
->ls_rsbtbl
[b
].lock
);
421 * Find rsb in rsbtbl and potentially create/add one
423 * Delaying the release of rsb's has a similar benefit to applications keeping
424 * NL locks on an rsb, but without the guarantee that the cached master value
425 * will still be valid when the rsb is reused. Apps aren't always smart enough
426 * to keep NL locks on an rsb that they may lock again shortly; this can lead
427 * to excessive master lookups and removals if we don't delay the release.
429 * Searching for an rsb means looking through both the normal list and toss
430 * list. When found on the toss list the rsb is moved to the normal list with
431 * ref count of 1; when found on normal list the ref count is incremented.
434 static int find_rsb(struct dlm_ls
*ls
, char *name
, int namelen
,
435 unsigned int flags
, struct dlm_rsb
**r_ret
)
437 struct dlm_rsb
*r
, *tmp
;
438 uint32_t hash
, bucket
;
441 if (dlm_no_directory(ls
))
444 hash
= jhash(name
, namelen
, 0);
445 bucket
= hash
& (ls
->ls_rsbtbl_size
- 1);
447 error
= search_rsb(ls
, name
, namelen
, bucket
, flags
, &r
);
451 if (error
== -EBADR
&& !(flags
& R_CREATE
))
454 /* the rsb was found but wasn't a master copy */
455 if (error
== -ENOTBLK
)
459 r
= create_rsb(ls
, name
, namelen
);
464 r
->res_bucket
= bucket
;
466 kref_init(&r
->res_ref
);
468 /* With no directory, the master can be set immediately */
469 if (dlm_no_directory(ls
)) {
470 int nodeid
= dlm_dir_nodeid(r
);
471 if (nodeid
== dlm_our_nodeid())
473 r
->res_nodeid
= nodeid
;
476 write_lock(&ls
->ls_rsbtbl
[bucket
].lock
);
477 error
= _search_rsb(ls
, name
, namelen
, bucket
, 0, &tmp
);
479 write_unlock(&ls
->ls_rsbtbl
[bucket
].lock
);
484 list_add(&r
->res_hashchain
, &ls
->ls_rsbtbl
[bucket
].list
);
485 write_unlock(&ls
->ls_rsbtbl
[bucket
].lock
);
492 int dlm_find_rsb(struct dlm_ls
*ls
, char *name
, int namelen
,
493 unsigned int flags
, struct dlm_rsb
**r_ret
)
495 return find_rsb(ls
, name
, namelen
, flags
, r_ret
);
498 /* This is only called to add a reference when the code already holds
499 a valid reference to the rsb, so there's no need for locking. */
501 static inline void hold_rsb(struct dlm_rsb
*r
)
503 kref_get(&r
->res_ref
);
506 void dlm_hold_rsb(struct dlm_rsb
*r
)
511 static void toss_rsb(struct kref
*kref
)
513 struct dlm_rsb
*r
= container_of(kref
, struct dlm_rsb
, res_ref
);
514 struct dlm_ls
*ls
= r
->res_ls
;
516 DLM_ASSERT(list_empty(&r
->res_root_list
), dlm_print_rsb(r
););
517 kref_init(&r
->res_ref
);
518 list_move(&r
->res_hashchain
, &ls
->ls_rsbtbl
[r
->res_bucket
].toss
);
519 r
->res_toss_time
= jiffies
;
521 dlm_free_lvb(r
->res_lvbptr
);
522 r
->res_lvbptr
= NULL
;
526 /* When all references to the rsb are gone it's transfered to
527 the tossed list for later disposal. */
529 static void put_rsb(struct dlm_rsb
*r
)
531 struct dlm_ls
*ls
= r
->res_ls
;
532 uint32_t bucket
= r
->res_bucket
;
534 write_lock(&ls
->ls_rsbtbl
[bucket
].lock
);
535 kref_put(&r
->res_ref
, toss_rsb
);
536 write_unlock(&ls
->ls_rsbtbl
[bucket
].lock
);
539 void dlm_put_rsb(struct dlm_rsb
*r
)
544 /* See comment for unhold_lkb */
546 static void unhold_rsb(struct dlm_rsb
*r
)
549 rv
= kref_put(&r
->res_ref
, toss_rsb
);
550 DLM_ASSERT(!rv
, dlm_dump_rsb(r
););
553 static void kill_rsb(struct kref
*kref
)
555 struct dlm_rsb
*r
= container_of(kref
, struct dlm_rsb
, res_ref
);
557 /* All work is done after the return from kref_put() so we
558 can release the write_lock before the remove and free. */
560 DLM_ASSERT(list_empty(&r
->res_lookup
), dlm_dump_rsb(r
););
561 DLM_ASSERT(list_empty(&r
->res_grantqueue
), dlm_dump_rsb(r
););
562 DLM_ASSERT(list_empty(&r
->res_convertqueue
), dlm_dump_rsb(r
););
563 DLM_ASSERT(list_empty(&r
->res_waitqueue
), dlm_dump_rsb(r
););
564 DLM_ASSERT(list_empty(&r
->res_root_list
), dlm_dump_rsb(r
););
565 DLM_ASSERT(list_empty(&r
->res_recover_list
), dlm_dump_rsb(r
););
568 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
569 The rsb must exist as long as any lkb's for it do. */
571 static void attach_lkb(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
574 lkb
->lkb_resource
= r
;
577 static void detach_lkb(struct dlm_lkb
*lkb
)
579 if (lkb
->lkb_resource
) {
580 put_rsb(lkb
->lkb_resource
);
581 lkb
->lkb_resource
= NULL
;
585 static int create_lkb(struct dlm_ls
*ls
, struct dlm_lkb
**lkb_ret
)
587 struct dlm_lkb
*lkb
, *tmp
;
591 lkb
= dlm_allocate_lkb(ls
);
595 lkb
->lkb_nodeid
= -1;
596 lkb
->lkb_grmode
= DLM_LOCK_IV
;
597 kref_init(&lkb
->lkb_ref
);
598 INIT_LIST_HEAD(&lkb
->lkb_ownqueue
);
599 INIT_LIST_HEAD(&lkb
->lkb_rsb_lookup
);
600 INIT_LIST_HEAD(&lkb
->lkb_time_list
);
602 get_random_bytes(&bucket
, sizeof(bucket
));
603 bucket
&= (ls
->ls_lkbtbl_size
- 1);
605 write_lock(&ls
->ls_lkbtbl
[bucket
].lock
);
607 /* counter can roll over so we must verify lkid is not in use */
610 lkid
= (bucket
<< 16) | ls
->ls_lkbtbl
[bucket
].counter
++;
612 list_for_each_entry(tmp
, &ls
->ls_lkbtbl
[bucket
].list
,
614 if (tmp
->lkb_id
!= lkid
)
622 list_add(&lkb
->lkb_idtbl_list
, &ls
->ls_lkbtbl
[bucket
].list
);
623 write_unlock(&ls
->ls_lkbtbl
[bucket
].lock
);
629 static struct dlm_lkb
*__find_lkb(struct dlm_ls
*ls
, uint32_t lkid
)
632 uint16_t bucket
= (lkid
>> 16);
634 list_for_each_entry(lkb
, &ls
->ls_lkbtbl
[bucket
].list
, lkb_idtbl_list
) {
635 if (lkb
->lkb_id
== lkid
)
641 static int find_lkb(struct dlm_ls
*ls
, uint32_t lkid
, struct dlm_lkb
**lkb_ret
)
644 uint16_t bucket
= (lkid
>> 16);
646 if (bucket
>= ls
->ls_lkbtbl_size
)
649 read_lock(&ls
->ls_lkbtbl
[bucket
].lock
);
650 lkb
= __find_lkb(ls
, lkid
);
652 kref_get(&lkb
->lkb_ref
);
653 read_unlock(&ls
->ls_lkbtbl
[bucket
].lock
);
656 return lkb
? 0 : -ENOENT
;
659 static void kill_lkb(struct kref
*kref
)
661 struct dlm_lkb
*lkb
= container_of(kref
, struct dlm_lkb
, lkb_ref
);
663 /* All work is done after the return from kref_put() so we
664 can release the write_lock before the detach_lkb */
666 DLM_ASSERT(!lkb
->lkb_status
, dlm_print_lkb(lkb
););
669 /* __put_lkb() is used when an lkb may not have an rsb attached to
670 it so we need to provide the lockspace explicitly */
672 static int __put_lkb(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
674 uint16_t bucket
= (lkb
->lkb_id
>> 16);
676 write_lock(&ls
->ls_lkbtbl
[bucket
].lock
);
677 if (kref_put(&lkb
->lkb_ref
, kill_lkb
)) {
678 list_del(&lkb
->lkb_idtbl_list
);
679 write_unlock(&ls
->ls_lkbtbl
[bucket
].lock
);
683 /* for local/process lkbs, lvbptr points to caller's lksb */
684 if (lkb
->lkb_lvbptr
&& is_master_copy(lkb
))
685 dlm_free_lvb(lkb
->lkb_lvbptr
);
689 write_unlock(&ls
->ls_lkbtbl
[bucket
].lock
);
694 int dlm_put_lkb(struct dlm_lkb
*lkb
)
698 DLM_ASSERT(lkb
->lkb_resource
, dlm_print_lkb(lkb
););
699 DLM_ASSERT(lkb
->lkb_resource
->res_ls
, dlm_print_lkb(lkb
););
701 ls
= lkb
->lkb_resource
->res_ls
;
702 return __put_lkb(ls
, lkb
);
705 /* This is only called to add a reference when the code already holds
706 a valid reference to the lkb, so there's no need for locking. */
708 static inline void hold_lkb(struct dlm_lkb
*lkb
)
710 kref_get(&lkb
->lkb_ref
);
713 /* This is called when we need to remove a reference and are certain
714 it's not the last ref. e.g. del_lkb is always called between a
715 find_lkb/put_lkb and is always the inverse of a previous add_lkb.
716 put_lkb would work fine, but would involve unnecessary locking */
718 static inline void unhold_lkb(struct dlm_lkb
*lkb
)
721 rv
= kref_put(&lkb
->lkb_ref
, kill_lkb
);
722 DLM_ASSERT(!rv
, dlm_print_lkb(lkb
););
725 static void lkb_add_ordered(struct list_head
*new, struct list_head
*head
,
728 struct dlm_lkb
*lkb
= NULL
;
730 list_for_each_entry(lkb
, head
, lkb_statequeue
)
731 if (lkb
->lkb_rqmode
< mode
)
735 list_add_tail(new, head
);
737 __list_add(new, lkb
->lkb_statequeue
.prev
, &lkb
->lkb_statequeue
);
740 /* add/remove lkb to rsb's grant/convert/wait queue */
742 static void add_lkb(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int status
)
744 kref_get(&lkb
->lkb_ref
);
746 DLM_ASSERT(!lkb
->lkb_status
, dlm_print_lkb(lkb
););
748 lkb
->lkb_status
= status
;
751 case DLM_LKSTS_WAITING
:
752 if (lkb
->lkb_exflags
& DLM_LKF_HEADQUE
)
753 list_add(&lkb
->lkb_statequeue
, &r
->res_waitqueue
);
755 list_add_tail(&lkb
->lkb_statequeue
, &r
->res_waitqueue
);
757 case DLM_LKSTS_GRANTED
:
758 /* convention says granted locks kept in order of grmode */
759 lkb_add_ordered(&lkb
->lkb_statequeue
, &r
->res_grantqueue
,
762 case DLM_LKSTS_CONVERT
:
763 if (lkb
->lkb_exflags
& DLM_LKF_HEADQUE
)
764 list_add(&lkb
->lkb_statequeue
, &r
->res_convertqueue
);
766 list_add_tail(&lkb
->lkb_statequeue
,
767 &r
->res_convertqueue
);
770 DLM_ASSERT(0, dlm_print_lkb(lkb
); printk("sts=%d\n", status
););
774 static void del_lkb(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
777 list_del(&lkb
->lkb_statequeue
);
781 static void move_lkb(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int sts
)
785 add_lkb(r
, lkb
, sts
);
789 static int msg_reply_type(int mstype
)
792 case DLM_MSG_REQUEST
:
793 return DLM_MSG_REQUEST_REPLY
;
794 case DLM_MSG_CONVERT
:
795 return DLM_MSG_CONVERT_REPLY
;
797 return DLM_MSG_UNLOCK_REPLY
;
799 return DLM_MSG_CANCEL_REPLY
;
801 return DLM_MSG_LOOKUP_REPLY
;
806 /* add/remove lkb from global waiters list of lkb's waiting for
807 a reply from a remote node */
809 static int add_to_waiters(struct dlm_lkb
*lkb
, int mstype
)
811 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
814 mutex_lock(&ls
->ls_waiters_mutex
);
816 if (is_overlap_unlock(lkb
) ||
817 (is_overlap_cancel(lkb
) && (mstype
== DLM_MSG_CANCEL
))) {
822 if (lkb
->lkb_wait_type
|| is_overlap_cancel(lkb
)) {
825 lkb
->lkb_flags
|= DLM_IFL_OVERLAP_UNLOCK
;
828 lkb
->lkb_flags
|= DLM_IFL_OVERLAP_CANCEL
;
834 lkb
->lkb_wait_count
++;
837 log_debug(ls
, "add overlap %x cur %d new %d count %d flags %x",
838 lkb
->lkb_id
, lkb
->lkb_wait_type
, mstype
,
839 lkb
->lkb_wait_count
, lkb
->lkb_flags
);
843 DLM_ASSERT(!lkb
->lkb_wait_count
,
845 printk("wait_count %d\n", lkb
->lkb_wait_count
););
847 lkb
->lkb_wait_count
++;
848 lkb
->lkb_wait_type
= mstype
;
850 list_add(&lkb
->lkb_wait_reply
, &ls
->ls_waiters
);
853 log_error(ls
, "add_to_waiters %x error %d flags %x %d %d %s",
854 lkb
->lkb_id
, error
, lkb
->lkb_flags
, mstype
,
855 lkb
->lkb_wait_type
, lkb
->lkb_resource
->res_name
);
856 mutex_unlock(&ls
->ls_waiters_mutex
);
860 /* We clear the RESEND flag because we might be taking an lkb off the waiters
861 list as part of process_requestqueue (e.g. a lookup that has an optimized
862 request reply on the requestqueue) between dlm_recover_waiters_pre() which
863 set RESEND and dlm_recover_waiters_post() */
865 static int _remove_from_waiters(struct dlm_lkb
*lkb
, int mstype
)
867 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
868 int overlap_done
= 0;
870 if (is_overlap_unlock(lkb
) && (mstype
== DLM_MSG_UNLOCK_REPLY
)) {
871 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_UNLOCK
;
876 if (is_overlap_cancel(lkb
) && (mstype
== DLM_MSG_CANCEL_REPLY
)) {
877 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_CANCEL
;
882 /* N.B. type of reply may not always correspond to type of original
883 msg due to lookup->request optimization, verify others? */
885 if (lkb
->lkb_wait_type
) {
886 lkb
->lkb_wait_type
= 0;
890 log_error(ls
, "remove_from_waiters lkid %x flags %x types %d %d",
891 lkb
->lkb_id
, lkb
->lkb_flags
, mstype
, lkb
->lkb_wait_type
);
895 /* the force-unlock/cancel has completed and we haven't recvd a reply
896 to the op that was in progress prior to the unlock/cancel; we
897 give up on any reply to the earlier op. FIXME: not sure when/how
900 if (overlap_done
&& lkb
->lkb_wait_type
) {
901 log_error(ls
, "remove_from_waiters %x reply %d give up on %d",
902 lkb
->lkb_id
, mstype
, lkb
->lkb_wait_type
);
903 lkb
->lkb_wait_count
--;
904 lkb
->lkb_wait_type
= 0;
907 DLM_ASSERT(lkb
->lkb_wait_count
, dlm_print_lkb(lkb
););
909 lkb
->lkb_flags
&= ~DLM_IFL_RESEND
;
910 lkb
->lkb_wait_count
--;
911 if (!lkb
->lkb_wait_count
)
912 list_del_init(&lkb
->lkb_wait_reply
);
917 static int remove_from_waiters(struct dlm_lkb
*lkb
, int mstype
)
919 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
922 mutex_lock(&ls
->ls_waiters_mutex
);
923 error
= _remove_from_waiters(lkb
, mstype
);
924 mutex_unlock(&ls
->ls_waiters_mutex
);
928 /* Handles situations where we might be processing a "fake" or "stub" reply in
929 which we can't try to take waiters_mutex again. */
931 static int remove_from_waiters_ms(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
933 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
936 if (ms
!= &ls
->ls_stub_ms
)
937 mutex_lock(&ls
->ls_waiters_mutex
);
938 error
= _remove_from_waiters(lkb
, ms
->m_type
);
939 if (ms
!= &ls
->ls_stub_ms
)
940 mutex_unlock(&ls
->ls_waiters_mutex
);
944 static void dir_remove(struct dlm_rsb
*r
)
948 if (dlm_no_directory(r
->res_ls
))
951 to_nodeid
= dlm_dir_nodeid(r
);
952 if (to_nodeid
!= dlm_our_nodeid())
955 dlm_dir_remove_entry(r
->res_ls
, to_nodeid
,
956 r
->res_name
, r
->res_length
);
959 /* FIXME: shouldn't this be able to exit as soon as one non-due rsb is
960 found since they are in order of newest to oldest? */
962 static int shrink_bucket(struct dlm_ls
*ls
, int b
)
965 int count
= 0, found
;
969 write_lock(&ls
->ls_rsbtbl
[b
].lock
);
970 list_for_each_entry_reverse(r
, &ls
->ls_rsbtbl
[b
].toss
,
972 if (!time_after_eq(jiffies
, r
->res_toss_time
+
973 dlm_config
.ci_toss_secs
* HZ
))
980 write_unlock(&ls
->ls_rsbtbl
[b
].lock
);
984 if (kref_put(&r
->res_ref
, kill_rsb
)) {
985 list_del(&r
->res_hashchain
);
986 write_unlock(&ls
->ls_rsbtbl
[b
].lock
);
993 write_unlock(&ls
->ls_rsbtbl
[b
].lock
);
994 log_error(ls
, "tossed rsb in use %s", r
->res_name
);
1001 void dlm_scan_rsbs(struct dlm_ls
*ls
)
1005 for (i
= 0; i
< ls
->ls_rsbtbl_size
; i
++) {
1006 shrink_bucket(ls
, i
);
1007 if (dlm_locking_stopped(ls
))
1013 static void add_timeout(struct dlm_lkb
*lkb
)
1015 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
1017 if (is_master_copy(lkb
)) {
1018 lkb
->lkb_timestamp
= jiffies
;
1022 if (test_bit(LSFL_TIMEWARN
, &ls
->ls_flags
) &&
1023 !(lkb
->lkb_exflags
& DLM_LKF_NODLCKWT
)) {
1024 lkb
->lkb_flags
|= DLM_IFL_WATCH_TIMEWARN
;
1027 if (lkb
->lkb_exflags
& DLM_LKF_TIMEOUT
)
1032 DLM_ASSERT(list_empty(&lkb
->lkb_time_list
), dlm_print_lkb(lkb
););
1033 mutex_lock(&ls
->ls_timeout_mutex
);
1035 lkb
->lkb_timestamp
= jiffies
;
1036 list_add_tail(&lkb
->lkb_time_list
, &ls
->ls_timeout
);
1037 mutex_unlock(&ls
->ls_timeout_mutex
);
1040 static void del_timeout(struct dlm_lkb
*lkb
)
1042 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
1044 mutex_lock(&ls
->ls_timeout_mutex
);
1045 if (!list_empty(&lkb
->lkb_time_list
)) {
1046 list_del_init(&lkb
->lkb_time_list
);
1049 mutex_unlock(&ls
->ls_timeout_mutex
);
1052 /* FIXME: is it safe to look at lkb_exflags, lkb_flags, lkb_timestamp, and
1053 lkb_lksb_timeout without lock_rsb? Note: we can't lock timeout_mutex
1054 and then lock rsb because of lock ordering in add_timeout. We may need
1055 to specify some special timeout-related bits in the lkb that are just to
1056 be accessed under the timeout_mutex. */
1058 void dlm_scan_timeout(struct dlm_ls
*ls
)
1061 struct dlm_lkb
*lkb
;
1062 int do_cancel
, do_warn
;
1065 if (dlm_locking_stopped(ls
))
1070 mutex_lock(&ls
->ls_timeout_mutex
);
1071 list_for_each_entry(lkb
, &ls
->ls_timeout
, lkb_time_list
) {
1073 if ((lkb
->lkb_exflags
& DLM_LKF_TIMEOUT
) &&
1074 time_after_eq(jiffies
, lkb
->lkb_timestamp
+
1075 lkb
->lkb_timeout_cs
* HZ
/100))
1078 if ((lkb
->lkb_flags
& DLM_IFL_WATCH_TIMEWARN
) &&
1079 time_after_eq(jiffies
, lkb
->lkb_timestamp
+
1080 dlm_config
.ci_timewarn_cs
* HZ
/100))
1083 if (!do_cancel
&& !do_warn
)
1088 mutex_unlock(&ls
->ls_timeout_mutex
);
1090 if (!do_cancel
&& !do_warn
)
1093 r
= lkb
->lkb_resource
;
1098 /* clear flag so we only warn once */
1099 lkb
->lkb_flags
&= ~DLM_IFL_WATCH_TIMEWARN
;
1100 if (!(lkb
->lkb_exflags
& DLM_LKF_TIMEOUT
))
1102 dlm_timeout_warn(lkb
);
1106 log_debug(ls
, "timeout cancel %x node %d %s",
1107 lkb
->lkb_id
, lkb
->lkb_nodeid
, r
->res_name
);
1108 lkb
->lkb_flags
&= ~DLM_IFL_WATCH_TIMEWARN
;
1109 lkb
->lkb_flags
|= DLM_IFL_TIMEOUT_CANCEL
;
1111 _cancel_lock(r
, lkb
);
1120 /* This is only called by dlm_recoverd, and we rely on dlm_ls_stop() stopping
1121 dlm_recoverd before checking/setting ls_recover_begin. */
1123 void dlm_adjust_timeouts(struct dlm_ls
*ls
)
1125 struct dlm_lkb
*lkb
;
1126 long adj
= jiffies
- ls
->ls_recover_begin
;
1128 ls
->ls_recover_begin
= 0;
1129 mutex_lock(&ls
->ls_timeout_mutex
);
1130 list_for_each_entry(lkb
, &ls
->ls_timeout
, lkb_time_list
)
1131 lkb
->lkb_timestamp
+= adj
;
1132 mutex_unlock(&ls
->ls_timeout_mutex
);
1135 /* lkb is master or local copy */
1137 static void set_lvb_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1139 int b
, len
= r
->res_ls
->ls_lvblen
;
1141 /* b=1 lvb returned to caller
1142 b=0 lvb written to rsb or invalidated
1145 b
= dlm_lvb_operations
[lkb
->lkb_grmode
+ 1][lkb
->lkb_rqmode
+ 1];
1148 if (!lkb
->lkb_lvbptr
)
1151 if (!(lkb
->lkb_exflags
& DLM_LKF_VALBLK
))
1157 memcpy(lkb
->lkb_lvbptr
, r
->res_lvbptr
, len
);
1158 lkb
->lkb_lvbseq
= r
->res_lvbseq
;
1160 } else if (b
== 0) {
1161 if (lkb
->lkb_exflags
& DLM_LKF_IVVALBLK
) {
1162 rsb_set_flag(r
, RSB_VALNOTVALID
);
1166 if (!lkb
->lkb_lvbptr
)
1169 if (!(lkb
->lkb_exflags
& DLM_LKF_VALBLK
))
1173 r
->res_lvbptr
= dlm_allocate_lvb(r
->res_ls
);
1178 memcpy(r
->res_lvbptr
, lkb
->lkb_lvbptr
, len
);
1180 lkb
->lkb_lvbseq
= r
->res_lvbseq
;
1181 rsb_clear_flag(r
, RSB_VALNOTVALID
);
1184 if (rsb_flag(r
, RSB_VALNOTVALID
))
1185 lkb
->lkb_sbflags
|= DLM_SBF_VALNOTVALID
;
1188 static void set_lvb_unlock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1190 if (lkb
->lkb_grmode
< DLM_LOCK_PW
)
1193 if (lkb
->lkb_exflags
& DLM_LKF_IVVALBLK
) {
1194 rsb_set_flag(r
, RSB_VALNOTVALID
);
1198 if (!lkb
->lkb_lvbptr
)
1201 if (!(lkb
->lkb_exflags
& DLM_LKF_VALBLK
))
1205 r
->res_lvbptr
= dlm_allocate_lvb(r
->res_ls
);
1210 memcpy(r
->res_lvbptr
, lkb
->lkb_lvbptr
, r
->res_ls
->ls_lvblen
);
1212 rsb_clear_flag(r
, RSB_VALNOTVALID
);
1215 /* lkb is process copy (pc) */
1217 static void set_lvb_lock_pc(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
1218 struct dlm_message
*ms
)
1222 if (!lkb
->lkb_lvbptr
)
1225 if (!(lkb
->lkb_exflags
& DLM_LKF_VALBLK
))
1228 b
= dlm_lvb_operations
[lkb
->lkb_grmode
+ 1][lkb
->lkb_rqmode
+ 1];
1230 int len
= receive_extralen(ms
);
1231 memcpy(lkb
->lkb_lvbptr
, ms
->m_extra
, len
);
1232 lkb
->lkb_lvbseq
= ms
->m_lvbseq
;
1236 /* Manipulate lkb's on rsb's convert/granted/waiting queues
1237 remove_lock -- used for unlock, removes lkb from granted
1238 revert_lock -- used for cancel, moves lkb from convert to granted
1239 grant_lock -- used for request and convert, adds lkb to granted or
1240 moves lkb from convert or waiting to granted
1242 Each of these is used for master or local copy lkb's. There is
1243 also a _pc() variation used to make the corresponding change on
1244 a process copy (pc) lkb. */
1246 static void _remove_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1249 lkb
->lkb_grmode
= DLM_LOCK_IV
;
1250 /* this unhold undoes the original ref from create_lkb()
1251 so this leads to the lkb being freed */
1255 static void remove_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1257 set_lvb_unlock(r
, lkb
);
1258 _remove_lock(r
, lkb
);
1261 static void remove_lock_pc(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1263 _remove_lock(r
, lkb
);
1266 /* returns: 0 did nothing
1267 1 moved lock to granted
1270 static int revert_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1274 lkb
->lkb_rqmode
= DLM_LOCK_IV
;
1276 switch (lkb
->lkb_status
) {
1277 case DLM_LKSTS_GRANTED
:
1279 case DLM_LKSTS_CONVERT
:
1280 move_lkb(r
, lkb
, DLM_LKSTS_GRANTED
);
1283 case DLM_LKSTS_WAITING
:
1285 lkb
->lkb_grmode
= DLM_LOCK_IV
;
1286 /* this unhold undoes the original ref from create_lkb()
1287 so this leads to the lkb being freed */
1292 log_print("invalid status for revert %d", lkb
->lkb_status
);
1297 static int revert_lock_pc(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1299 return revert_lock(r
, lkb
);
1302 static void _grant_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1304 if (lkb
->lkb_grmode
!= lkb
->lkb_rqmode
) {
1305 lkb
->lkb_grmode
= lkb
->lkb_rqmode
;
1306 if (lkb
->lkb_status
)
1307 move_lkb(r
, lkb
, DLM_LKSTS_GRANTED
);
1309 add_lkb(r
, lkb
, DLM_LKSTS_GRANTED
);
1312 lkb
->lkb_rqmode
= DLM_LOCK_IV
;
1315 static void grant_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1317 set_lvb_lock(r
, lkb
);
1318 _grant_lock(r
, lkb
);
1319 lkb
->lkb_highbast
= 0;
1322 static void grant_lock_pc(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
1323 struct dlm_message
*ms
)
1325 set_lvb_lock_pc(r
, lkb
, ms
);
1326 _grant_lock(r
, lkb
);
1329 /* called by grant_pending_locks() which means an async grant message must
1330 be sent to the requesting node in addition to granting the lock if the
1331 lkb belongs to a remote node. */
1333 static void grant_lock_pending(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1336 if (is_master_copy(lkb
))
1339 queue_cast(r
, lkb
, 0);
1342 /* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
1343 change the granted/requested modes. We're munging things accordingly in
1345 CONVDEADLK: our grmode may have been forced down to NL to resolve a
1347 ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
1348 compatible with other granted locks */
1350 static void munge_demoted(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
1352 if (ms
->m_type
!= DLM_MSG_CONVERT_REPLY
) {
1353 log_print("munge_demoted %x invalid reply type %d",
1354 lkb
->lkb_id
, ms
->m_type
);
1358 if (lkb
->lkb_rqmode
== DLM_LOCK_IV
|| lkb
->lkb_grmode
== DLM_LOCK_IV
) {
1359 log_print("munge_demoted %x invalid modes gr %d rq %d",
1360 lkb
->lkb_id
, lkb
->lkb_grmode
, lkb
->lkb_rqmode
);
1364 lkb
->lkb_grmode
= DLM_LOCK_NL
;
1367 static void munge_altmode(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
1369 if (ms
->m_type
!= DLM_MSG_REQUEST_REPLY
&&
1370 ms
->m_type
!= DLM_MSG_GRANT
) {
1371 log_print("munge_altmode %x invalid reply type %d",
1372 lkb
->lkb_id
, ms
->m_type
);
1376 if (lkb
->lkb_exflags
& DLM_LKF_ALTPR
)
1377 lkb
->lkb_rqmode
= DLM_LOCK_PR
;
1378 else if (lkb
->lkb_exflags
& DLM_LKF_ALTCW
)
1379 lkb
->lkb_rqmode
= DLM_LOCK_CW
;
1381 log_print("munge_altmode invalid exflags %x", lkb
->lkb_exflags
);
1386 static inline int first_in_list(struct dlm_lkb
*lkb
, struct list_head
*head
)
1388 struct dlm_lkb
*first
= list_entry(head
->next
, struct dlm_lkb
,
1390 if (lkb
->lkb_id
== first
->lkb_id
)
1396 /* Check if the given lkb conflicts with another lkb on the queue. */
1398 static int queue_conflict(struct list_head
*head
, struct dlm_lkb
*lkb
)
1400 struct dlm_lkb
*this;
1402 list_for_each_entry(this, head
, lkb_statequeue
) {
1405 if (!modes_compat(this, lkb
))
1412 * "A conversion deadlock arises with a pair of lock requests in the converting
1413 * queue for one resource. The granted mode of each lock blocks the requested
1414 * mode of the other lock."
1416 * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
1417 * convert queue from being granted, then deadlk/demote lkb.
1420 * Granted Queue: empty
1421 * Convert Queue: NL->EX (first lock)
1422 * PR->EX (second lock)
1424 * The first lock can't be granted because of the granted mode of the second
1425 * lock and the second lock can't be granted because it's not first in the
1426 * list. We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
1427 * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
1428 * flag set and return DEMOTED in the lksb flags.
1430 * Originally, this function detected conv-deadlk in a more limited scope:
1431 * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
1432 * - if lkb1 was the first entry in the queue (not just earlier), and was
1433 * blocked by the granted mode of lkb2, and there was nothing on the
1434 * granted queue preventing lkb1 from being granted immediately, i.e.
1435 * lkb2 was the only thing preventing lkb1 from being granted.
1437 * That second condition meant we'd only say there was conv-deadlk if
1438 * resolving it (by demotion) would lead to the first lock on the convert
1439 * queue being granted right away. It allowed conversion deadlocks to exist
1440 * between locks on the convert queue while they couldn't be granted anyway.
1442 * Now, we detect and take action on conversion deadlocks immediately when
1443 * they're created, even if they may not be immediately consequential. If
1444 * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
1445 * mode that would prevent lkb1's conversion from being granted, we do a
1446 * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
1447 * I think this means that the lkb_is_ahead condition below should always
1448 * be zero, i.e. there will never be conv-deadlk between two locks that are
1449 * both already on the convert queue.
1452 static int conversion_deadlock_detect(struct dlm_rsb
*r
, struct dlm_lkb
*lkb2
)
1454 struct dlm_lkb
*lkb1
;
1455 int lkb_is_ahead
= 0;
1457 list_for_each_entry(lkb1
, &r
->res_convertqueue
, lkb_statequeue
) {
1463 if (!lkb_is_ahead
) {
1464 if (!modes_compat(lkb2
, lkb1
))
1467 if (!modes_compat(lkb2
, lkb1
) &&
1468 !modes_compat(lkb1
, lkb2
))
1476 * Return 1 if the lock can be granted, 0 otherwise.
1477 * Also detect and resolve conversion deadlocks.
1479 * lkb is the lock to be granted
1481 * now is 1 if the function is being called in the context of the
1482 * immediate request, it is 0 if called later, after the lock has been
1485 * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
1488 static int _can_be_granted(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int now
)
1490 int8_t conv
= (lkb
->lkb_grmode
!= DLM_LOCK_IV
);
1493 * 6-10: Version 5.4 introduced an option to address the phenomenon of
1494 * a new request for a NL mode lock being blocked.
1496 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
1497 * request, then it would be granted. In essence, the use of this flag
1498 * tells the Lock Manager to expedite theis request by not considering
1499 * what may be in the CONVERTING or WAITING queues... As of this
1500 * writing, the EXPEDITE flag can be used only with new requests for NL
1501 * mode locks. This flag is not valid for conversion requests.
1503 * A shortcut. Earlier checks return an error if EXPEDITE is used in a
1504 * conversion or used with a non-NL requested mode. We also know an
1505 * EXPEDITE request is always granted immediately, so now must always
1506 * be 1. The full condition to grant an expedite request: (now &&
1507 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
1508 * therefore be shortened to just checking the flag.
1511 if (lkb
->lkb_exflags
& DLM_LKF_EXPEDITE
)
1515 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
1516 * added to the remaining conditions.
1519 if (queue_conflict(&r
->res_grantqueue
, lkb
))
1523 * 6-3: By default, a conversion request is immediately granted if the
1524 * requested mode is compatible with the modes of all other granted
1528 if (queue_conflict(&r
->res_convertqueue
, lkb
))
1532 * 6-5: But the default algorithm for deciding whether to grant or
1533 * queue conversion requests does not by itself guarantee that such
1534 * requests are serviced on a "first come first serve" basis. This, in
1535 * turn, can lead to a phenomenon known as "indefinate postponement".
1537 * 6-7: This issue is dealt with by using the optional QUECVT flag with
1538 * the system service employed to request a lock conversion. This flag
1539 * forces certain conversion requests to be queued, even if they are
1540 * compatible with the granted modes of other locks on the same
1541 * resource. Thus, the use of this flag results in conversion requests
1542 * being ordered on a "first come first servce" basis.
1544 * DCT: This condition is all about new conversions being able to occur
1545 * "in place" while the lock remains on the granted queue (assuming
1546 * nothing else conflicts.) IOW if QUECVT isn't set, a conversion
1547 * doesn't _have_ to go onto the convert queue where it's processed in
1548 * order. The "now" variable is necessary to distinguish converts
1549 * being received and processed for the first time now, because once a
1550 * convert is moved to the conversion queue the condition below applies
1551 * requiring fifo granting.
1554 if (now
&& conv
&& !(lkb
->lkb_exflags
& DLM_LKF_QUECVT
))
1558 * The NOORDER flag is set to avoid the standard vms rules on grant
1562 if (lkb
->lkb_exflags
& DLM_LKF_NOORDER
)
1566 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
1567 * granted until all other conversion requests ahead of it are granted
1571 if (!now
&& conv
&& first_in_list(lkb
, &r
->res_convertqueue
))
1575 * 6-4: By default, a new request is immediately granted only if all
1576 * three of the following conditions are satisfied when the request is
1578 * - The queue of ungranted conversion requests for the resource is
1580 * - The queue of ungranted new requests for the resource is empty.
1581 * - The mode of the new request is compatible with the most
1582 * restrictive mode of all granted locks on the resource.
1585 if (now
&& !conv
&& list_empty(&r
->res_convertqueue
) &&
1586 list_empty(&r
->res_waitqueue
))
1590 * 6-4: Once a lock request is in the queue of ungranted new requests,
1591 * it cannot be granted until the queue of ungranted conversion
1592 * requests is empty, all ungranted new requests ahead of it are
1593 * granted and/or canceled, and it is compatible with the granted mode
1594 * of the most restrictive lock granted on the resource.
1597 if (!now
&& !conv
&& list_empty(&r
->res_convertqueue
) &&
1598 first_in_list(lkb
, &r
->res_waitqueue
))
1604 static int can_be_granted(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int now
,
1608 int8_t alt
= 0, rqmode
= lkb
->lkb_rqmode
;
1609 int8_t is_convert
= (lkb
->lkb_grmode
!= DLM_LOCK_IV
);
1614 rv
= _can_be_granted(r
, lkb
, now
);
1619 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
1620 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
1621 * cancels one of the locks.
1624 if (is_convert
&& can_be_queued(lkb
) &&
1625 conversion_deadlock_detect(r
, lkb
)) {
1626 if (lkb
->lkb_exflags
& DLM_LKF_CONVDEADLK
) {
1627 lkb
->lkb_grmode
= DLM_LOCK_NL
;
1628 lkb
->lkb_sbflags
|= DLM_SBF_DEMOTED
;
1629 } else if (!(lkb
->lkb_exflags
& DLM_LKF_NODLCKWT
)) {
1633 log_print("can_be_granted deadlock %x now %d",
1642 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
1643 * to grant a request in a mode other than the normal rqmode. It's a
1644 * simple way to provide a big optimization to applications that can
1648 if (rqmode
!= DLM_LOCK_PR
&& (lkb
->lkb_exflags
& DLM_LKF_ALTPR
))
1650 else if (rqmode
!= DLM_LOCK_CW
&& (lkb
->lkb_exflags
& DLM_LKF_ALTCW
))
1654 lkb
->lkb_rqmode
= alt
;
1655 rv
= _can_be_granted(r
, lkb
, now
);
1657 lkb
->lkb_sbflags
|= DLM_SBF_ALTMODE
;
1659 lkb
->lkb_rqmode
= rqmode
;
1665 /* FIXME: I don't think that can_be_granted() can/will demote or find deadlock
1666 for locks pending on the convert list. Once verified (watch for these
1667 log_prints), we should be able to just call _can_be_granted() and not
1668 bother with the demote/deadlk cases here (and there's no easy way to deal
1669 with a deadlk here, we'd have to generate something like grant_lock with
1670 the deadlk error.) */
1672 /* Returns the highest requested mode of all blocked conversions; sets
1673 cw if there's a blocked conversion to DLM_LOCK_CW. */
1675 static int grant_pending_convert(struct dlm_rsb
*r
, int high
, int *cw
)
1677 struct dlm_lkb
*lkb
, *s
;
1678 int hi
, demoted
, quit
, grant_restart
, demote_restart
;
1687 list_for_each_entry_safe(lkb
, s
, &r
->res_convertqueue
, lkb_statequeue
) {
1688 demoted
= is_demoted(lkb
);
1691 if (can_be_granted(r
, lkb
, 0, &deadlk
)) {
1692 grant_lock_pending(r
, lkb
);
1697 if (!demoted
&& is_demoted(lkb
)) {
1698 log_print("WARN: pending demoted %x node %d %s",
1699 lkb
->lkb_id
, lkb
->lkb_nodeid
, r
->res_name
);
1705 log_print("WARN: pending deadlock %x node %d %s",
1706 lkb
->lkb_id
, lkb
->lkb_nodeid
, r
->res_name
);
1711 hi
= max_t(int, lkb
->lkb_rqmode
, hi
);
1713 if (cw
&& lkb
->lkb_rqmode
== DLM_LOCK_CW
)
1719 if (demote_restart
&& !quit
) {
1724 return max_t(int, high
, hi
);
1727 static int grant_pending_wait(struct dlm_rsb
*r
, int high
, int *cw
)
1729 struct dlm_lkb
*lkb
, *s
;
1731 list_for_each_entry_safe(lkb
, s
, &r
->res_waitqueue
, lkb_statequeue
) {
1732 if (can_be_granted(r
, lkb
, 0, NULL
))
1733 grant_lock_pending(r
, lkb
);
1735 high
= max_t(int, lkb
->lkb_rqmode
, high
);
1736 if (lkb
->lkb_rqmode
== DLM_LOCK_CW
)
1744 /* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
1745 on either the convert or waiting queue.
1746 high is the largest rqmode of all locks blocked on the convert or
1749 static int lock_requires_bast(struct dlm_lkb
*gr
, int high
, int cw
)
1751 if (gr
->lkb_grmode
== DLM_LOCK_PR
&& cw
) {
1752 if (gr
->lkb_highbast
< DLM_LOCK_EX
)
1757 if (gr
->lkb_highbast
< high
&&
1758 !__dlm_compat_matrix
[gr
->lkb_grmode
+1][high
+1])
1763 static void grant_pending_locks(struct dlm_rsb
*r
)
1765 struct dlm_lkb
*lkb
, *s
;
1766 int high
= DLM_LOCK_IV
;
1769 DLM_ASSERT(is_master(r
), dlm_dump_rsb(r
););
1771 high
= grant_pending_convert(r
, high
, &cw
);
1772 high
= grant_pending_wait(r
, high
, &cw
);
1774 if (high
== DLM_LOCK_IV
)
1778 * If there are locks left on the wait/convert queue then send blocking
1779 * ASTs to granted locks based on the largest requested mode (high)
1783 list_for_each_entry_safe(lkb
, s
, &r
->res_grantqueue
, lkb_statequeue
) {
1784 if (lkb
->lkb_bastaddr
&& lock_requires_bast(lkb
, high
, cw
)) {
1785 if (cw
&& high
== DLM_LOCK_PR
)
1786 queue_bast(r
, lkb
, DLM_LOCK_CW
);
1788 queue_bast(r
, lkb
, high
);
1789 lkb
->lkb_highbast
= high
;
1794 static int modes_require_bast(struct dlm_lkb
*gr
, struct dlm_lkb
*rq
)
1796 if ((gr
->lkb_grmode
== DLM_LOCK_PR
&& rq
->lkb_rqmode
== DLM_LOCK_CW
) ||
1797 (gr
->lkb_grmode
== DLM_LOCK_CW
&& rq
->lkb_rqmode
== DLM_LOCK_PR
)) {
1798 if (gr
->lkb_highbast
< DLM_LOCK_EX
)
1803 if (gr
->lkb_highbast
< rq
->lkb_rqmode
&& !modes_compat(gr
, rq
))
1808 static void send_bast_queue(struct dlm_rsb
*r
, struct list_head
*head
,
1809 struct dlm_lkb
*lkb
)
1813 list_for_each_entry(gr
, head
, lkb_statequeue
) {
1814 if (gr
->lkb_bastaddr
&& modes_require_bast(gr
, lkb
)) {
1815 queue_bast(r
, gr
, lkb
->lkb_rqmode
);
1816 gr
->lkb_highbast
= lkb
->lkb_rqmode
;
1821 static void send_blocking_asts(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1823 send_bast_queue(r
, &r
->res_grantqueue
, lkb
);
1826 static void send_blocking_asts_all(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1828 send_bast_queue(r
, &r
->res_grantqueue
, lkb
);
1829 send_bast_queue(r
, &r
->res_convertqueue
, lkb
);
1832 /* set_master(r, lkb) -- set the master nodeid of a resource
1834 The purpose of this function is to set the nodeid field in the given
1835 lkb using the nodeid field in the given rsb. If the rsb's nodeid is
1836 known, it can just be copied to the lkb and the function will return
1837 0. If the rsb's nodeid is _not_ known, it needs to be looked up
1838 before it can be copied to the lkb.
1840 When the rsb nodeid is being looked up remotely, the initial lkb
1841 causing the lookup is kept on the ls_waiters list waiting for the
1842 lookup reply. Other lkb's waiting for the same rsb lookup are kept
1843 on the rsb's res_lookup list until the master is verified.
1846 0: nodeid is set in rsb/lkb and the caller should go ahead and use it
1847 1: the rsb master is not available and the lkb has been placed on
1851 static int set_master(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
1853 struct dlm_ls
*ls
= r
->res_ls
;
1854 int error
, dir_nodeid
, ret_nodeid
, our_nodeid
= dlm_our_nodeid();
1856 if (rsb_flag(r
, RSB_MASTER_UNCERTAIN
)) {
1857 rsb_clear_flag(r
, RSB_MASTER_UNCERTAIN
);
1858 r
->res_first_lkid
= lkb
->lkb_id
;
1859 lkb
->lkb_nodeid
= r
->res_nodeid
;
1863 if (r
->res_first_lkid
&& r
->res_first_lkid
!= lkb
->lkb_id
) {
1864 list_add_tail(&lkb
->lkb_rsb_lookup
, &r
->res_lookup
);
1868 if (r
->res_nodeid
== 0) {
1869 lkb
->lkb_nodeid
= 0;
1873 if (r
->res_nodeid
> 0) {
1874 lkb
->lkb_nodeid
= r
->res_nodeid
;
1878 DLM_ASSERT(r
->res_nodeid
== -1, dlm_dump_rsb(r
););
1880 dir_nodeid
= dlm_dir_nodeid(r
);
1882 if (dir_nodeid
!= our_nodeid
) {
1883 r
->res_first_lkid
= lkb
->lkb_id
;
1884 send_lookup(r
, lkb
);
1889 /* It's possible for dlm_scand to remove an old rsb for
1890 this same resource from the toss list, us to create
1891 a new one, look up the master locally, and find it
1892 already exists just before dlm_scand does the
1893 dir_remove() on the previous rsb. */
1895 error
= dlm_dir_lookup(ls
, our_nodeid
, r
->res_name
,
1896 r
->res_length
, &ret_nodeid
);
1899 log_debug(ls
, "dir_lookup error %d %s", error
, r
->res_name
);
1903 if (ret_nodeid
== our_nodeid
) {
1904 r
->res_first_lkid
= 0;
1906 lkb
->lkb_nodeid
= 0;
1908 r
->res_first_lkid
= lkb
->lkb_id
;
1909 r
->res_nodeid
= ret_nodeid
;
1910 lkb
->lkb_nodeid
= ret_nodeid
;
1915 static void process_lookup_list(struct dlm_rsb
*r
)
1917 struct dlm_lkb
*lkb
, *safe
;
1919 list_for_each_entry_safe(lkb
, safe
, &r
->res_lookup
, lkb_rsb_lookup
) {
1920 list_del_init(&lkb
->lkb_rsb_lookup
);
1921 _request_lock(r
, lkb
);
1926 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
1928 static void confirm_master(struct dlm_rsb
*r
, int error
)
1930 struct dlm_lkb
*lkb
;
1932 if (!r
->res_first_lkid
)
1938 r
->res_first_lkid
= 0;
1939 process_lookup_list(r
);
1943 /* the remote master didn't queue our NOQUEUE request;
1944 make a waiting lkb the first_lkid */
1946 r
->res_first_lkid
= 0;
1948 if (!list_empty(&r
->res_lookup
)) {
1949 lkb
= list_entry(r
->res_lookup
.next
, struct dlm_lkb
,
1951 list_del_init(&lkb
->lkb_rsb_lookup
);
1952 r
->res_first_lkid
= lkb
->lkb_id
;
1953 _request_lock(r
, lkb
);
1959 log_error(r
->res_ls
, "confirm_master unknown error %d", error
);
1963 static int set_lock_args(int mode
, struct dlm_lksb
*lksb
, uint32_t flags
,
1964 int namelen
, unsigned long timeout_cs
, void *ast
,
1965 void *astarg
, void *bast
, struct dlm_args
*args
)
1969 /* check for invalid arg usage */
1971 if (mode
< 0 || mode
> DLM_LOCK_EX
)
1974 if (!(flags
& DLM_LKF_CONVERT
) && (namelen
> DLM_RESNAME_MAXLEN
))
1977 if (flags
& DLM_LKF_CANCEL
)
1980 if (flags
& DLM_LKF_QUECVT
&& !(flags
& DLM_LKF_CONVERT
))
1983 if (flags
& DLM_LKF_CONVDEADLK
&& !(flags
& DLM_LKF_CONVERT
))
1986 if (flags
& DLM_LKF_CONVDEADLK
&& flags
& DLM_LKF_NOQUEUE
)
1989 if (flags
& DLM_LKF_EXPEDITE
&& flags
& DLM_LKF_CONVERT
)
1992 if (flags
& DLM_LKF_EXPEDITE
&& flags
& DLM_LKF_QUECVT
)
1995 if (flags
& DLM_LKF_EXPEDITE
&& flags
& DLM_LKF_NOQUEUE
)
1998 if (flags
& DLM_LKF_EXPEDITE
&& mode
!= DLM_LOCK_NL
)
2004 if (flags
& DLM_LKF_VALBLK
&& !lksb
->sb_lvbptr
)
2007 if (flags
& DLM_LKF_CONVERT
&& !lksb
->sb_lkid
)
2010 /* these args will be copied to the lkb in validate_lock_args,
2011 it cannot be done now because when converting locks, fields in
2012 an active lkb cannot be modified before locking the rsb */
2014 args
->flags
= flags
;
2015 args
->astaddr
= ast
;
2016 args
->astparam
= (long) astarg
;
2017 args
->bastaddr
= bast
;
2018 args
->timeout
= timeout_cs
;
2026 static int set_unlock_args(uint32_t flags
, void *astarg
, struct dlm_args
*args
)
2028 if (flags
& ~(DLM_LKF_CANCEL
| DLM_LKF_VALBLK
| DLM_LKF_IVVALBLK
|
2029 DLM_LKF_FORCEUNLOCK
))
2032 if (flags
& DLM_LKF_CANCEL
&& flags
& DLM_LKF_FORCEUNLOCK
)
2035 args
->flags
= flags
;
2036 args
->astparam
= (long) astarg
;
2040 static int validate_lock_args(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
2041 struct dlm_args
*args
)
2045 if (args
->flags
& DLM_LKF_CONVERT
) {
2046 if (lkb
->lkb_flags
& DLM_IFL_MSTCPY
)
2049 if (args
->flags
& DLM_LKF_QUECVT
&&
2050 !__quecvt_compat_matrix
[lkb
->lkb_grmode
+1][args
->mode
+1])
2054 if (lkb
->lkb_status
!= DLM_LKSTS_GRANTED
)
2057 if (lkb
->lkb_wait_type
)
2060 if (is_overlap(lkb
))
2064 lkb
->lkb_exflags
= args
->flags
;
2065 lkb
->lkb_sbflags
= 0;
2066 lkb
->lkb_astaddr
= args
->astaddr
;
2067 lkb
->lkb_astparam
= args
->astparam
;
2068 lkb
->lkb_bastaddr
= args
->bastaddr
;
2069 lkb
->lkb_rqmode
= args
->mode
;
2070 lkb
->lkb_lksb
= args
->lksb
;
2071 lkb
->lkb_lvbptr
= args
->lksb
->sb_lvbptr
;
2072 lkb
->lkb_ownpid
= (int) current
->pid
;
2073 lkb
->lkb_timeout_cs
= args
->timeout
;
2079 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2082 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2083 because there may be a lookup in progress and it's valid to do
2084 cancel/unlockf on it */
2086 static int validate_unlock_args(struct dlm_lkb
*lkb
, struct dlm_args
*args
)
2088 struct dlm_ls
*ls
= lkb
->lkb_resource
->res_ls
;
2091 if (lkb
->lkb_flags
& DLM_IFL_MSTCPY
) {
2092 log_error(ls
, "unlock on MSTCPY %x", lkb
->lkb_id
);
2097 /* an lkb may still exist even though the lock is EOL'ed due to a
2098 cancel, unlock or failed noqueue request; an app can't use these
2099 locks; return same error as if the lkid had not been found at all */
2101 if (lkb
->lkb_flags
& DLM_IFL_ENDOFLIFE
) {
2102 log_debug(ls
, "unlock on ENDOFLIFE %x", lkb
->lkb_id
);
2107 /* an lkb may be waiting for an rsb lookup to complete where the
2108 lookup was initiated by another lock */
2110 if (args
->flags
& (DLM_LKF_CANCEL
| DLM_LKF_FORCEUNLOCK
)) {
2111 if (!list_empty(&lkb
->lkb_rsb_lookup
)) {
2112 log_debug(ls
, "unlock on rsb_lookup %x", lkb
->lkb_id
);
2113 list_del_init(&lkb
->lkb_rsb_lookup
);
2114 queue_cast(lkb
->lkb_resource
, lkb
,
2115 args
->flags
& DLM_LKF_CANCEL
?
2116 -DLM_ECANCEL
: -DLM_EUNLOCK
);
2117 unhold_lkb(lkb
); /* undoes create_lkb() */
2123 /* cancel not allowed with another cancel/unlock in progress */
2125 if (args
->flags
& DLM_LKF_CANCEL
) {
2126 if (lkb
->lkb_exflags
& DLM_LKF_CANCEL
)
2129 if (is_overlap(lkb
))
2132 /* don't let scand try to do a cancel */
2135 if (lkb
->lkb_flags
& DLM_IFL_RESEND
) {
2136 lkb
->lkb_flags
|= DLM_IFL_OVERLAP_CANCEL
;
2141 switch (lkb
->lkb_wait_type
) {
2142 case DLM_MSG_LOOKUP
:
2143 case DLM_MSG_REQUEST
:
2144 lkb
->lkb_flags
|= DLM_IFL_OVERLAP_CANCEL
;
2147 case DLM_MSG_UNLOCK
:
2148 case DLM_MSG_CANCEL
:
2151 /* add_to_waiters() will set OVERLAP_CANCEL */
2155 /* do we need to allow a force-unlock if there's a normal unlock
2156 already in progress? in what conditions could the normal unlock
2157 fail such that we'd want to send a force-unlock to be sure? */
2159 if (args
->flags
& DLM_LKF_FORCEUNLOCK
) {
2160 if (lkb
->lkb_exflags
& DLM_LKF_FORCEUNLOCK
)
2163 if (is_overlap_unlock(lkb
))
2166 /* don't let scand try to do a cancel */
2169 if (lkb
->lkb_flags
& DLM_IFL_RESEND
) {
2170 lkb
->lkb_flags
|= DLM_IFL_OVERLAP_UNLOCK
;
2175 switch (lkb
->lkb_wait_type
) {
2176 case DLM_MSG_LOOKUP
:
2177 case DLM_MSG_REQUEST
:
2178 lkb
->lkb_flags
|= DLM_IFL_OVERLAP_UNLOCK
;
2181 case DLM_MSG_UNLOCK
:
2184 /* add_to_waiters() will set OVERLAP_UNLOCK */
2188 /* normal unlock not allowed if there's any op in progress */
2190 if (lkb
->lkb_wait_type
|| lkb
->lkb_wait_count
)
2194 /* an overlapping op shouldn't blow away exflags from other op */
2195 lkb
->lkb_exflags
|= args
->flags
;
2196 lkb
->lkb_sbflags
= 0;
2197 lkb
->lkb_astparam
= args
->astparam
;
2201 log_debug(ls
, "validate_unlock_args %d %x %x %x %x %d %s", rv
,
2202 lkb
->lkb_id
, lkb
->lkb_flags
, lkb
->lkb_exflags
,
2203 args
->flags
, lkb
->lkb_wait_type
,
2204 lkb
->lkb_resource
->res_name
);
2209 * Four stage 4 varieties:
2210 * do_request(), do_convert(), do_unlock(), do_cancel()
2211 * These are called on the master node for the given lock and
2212 * from the central locking logic.
2215 static int do_request(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2219 if (can_be_granted(r
, lkb
, 1, NULL
)) {
2221 queue_cast(r
, lkb
, 0);
2225 if (can_be_queued(lkb
)) {
2226 error
= -EINPROGRESS
;
2227 add_lkb(r
, lkb
, DLM_LKSTS_WAITING
);
2228 send_blocking_asts(r
, lkb
);
2234 if (force_blocking_asts(lkb
))
2235 send_blocking_asts_all(r
, lkb
);
2236 queue_cast(r
, lkb
, -EAGAIN
);
2242 static int do_convert(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2247 /* changing an existing lock may allow others to be granted */
2249 if (can_be_granted(r
, lkb
, 1, &deadlk
)) {
2251 queue_cast(r
, lkb
, 0);
2252 grant_pending_locks(r
);
2256 /* can_be_granted() detected that this lock would block in a conversion
2257 deadlock, so we leave it on the granted queue and return EDEADLK in
2258 the ast for the convert. */
2261 /* it's left on the granted queue */
2262 log_debug(r
->res_ls
, "deadlock %x node %d sts%d g%d r%d %s",
2263 lkb
->lkb_id
, lkb
->lkb_nodeid
, lkb
->lkb_status
,
2264 lkb
->lkb_grmode
, lkb
->lkb_rqmode
, r
->res_name
);
2265 revert_lock(r
, lkb
);
2266 queue_cast(r
, lkb
, -EDEADLK
);
2271 /* is_demoted() means the can_be_granted() above set the grmode
2272 to NL, and left us on the granted queue. This auto-demotion
2273 (due to CONVDEADLK) might mean other locks, and/or this lock, are
2274 now grantable. We have to try to grant other converting locks
2275 before we try again to grant this one. */
2277 if (is_demoted(lkb
)) {
2278 grant_pending_convert(r
, DLM_LOCK_IV
, NULL
);
2279 if (_can_be_granted(r
, lkb
, 1)) {
2281 queue_cast(r
, lkb
, 0);
2282 grant_pending_locks(r
);
2285 /* else fall through and move to convert queue */
2288 if (can_be_queued(lkb
)) {
2289 error
= -EINPROGRESS
;
2291 add_lkb(r
, lkb
, DLM_LKSTS_CONVERT
);
2292 send_blocking_asts(r
, lkb
);
2298 if (force_blocking_asts(lkb
))
2299 send_blocking_asts_all(r
, lkb
);
2300 queue_cast(r
, lkb
, -EAGAIN
);
2306 static int do_unlock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2308 remove_lock(r
, lkb
);
2309 queue_cast(r
, lkb
, -DLM_EUNLOCK
);
2310 grant_pending_locks(r
);
2311 return -DLM_EUNLOCK
;
2314 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
2316 static int do_cancel(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2320 error
= revert_lock(r
, lkb
);
2322 queue_cast(r
, lkb
, -DLM_ECANCEL
);
2323 grant_pending_locks(r
);
2324 return -DLM_ECANCEL
;
2330 * Four stage 3 varieties:
2331 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2334 /* add a new lkb to a possibly new rsb, called by requesting process */
2336 static int _request_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2340 /* set_master: sets lkb nodeid from r */
2342 error
= set_master(r
, lkb
);
2351 /* receive_request() calls do_request() on remote node */
2352 error
= send_request(r
, lkb
);
2354 error
= do_request(r
, lkb
);
2359 /* change some property of an existing lkb, e.g. mode */
2361 static int _convert_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2366 /* receive_convert() calls do_convert() on remote node */
2367 error
= send_convert(r
, lkb
);
2369 error
= do_convert(r
, lkb
);
2374 /* remove an existing lkb from the granted queue */
2376 static int _unlock_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2381 /* receive_unlock() calls do_unlock() on remote node */
2382 error
= send_unlock(r
, lkb
);
2384 error
= do_unlock(r
, lkb
);
2389 /* remove an existing lkb from the convert or wait queue */
2391 static int _cancel_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2396 /* receive_cancel() calls do_cancel() on remote node */
2397 error
= send_cancel(r
, lkb
);
2399 error
= do_cancel(r
, lkb
);
2405 * Four stage 2 varieties:
2406 * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2409 static int request_lock(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
, char *name
,
2410 int len
, struct dlm_args
*args
)
2415 error
= validate_lock_args(ls
, lkb
, args
);
2419 error
= find_rsb(ls
, name
, len
, R_CREATE
, &r
);
2426 lkb
->lkb_lksb
->sb_lkid
= lkb
->lkb_id
;
2428 error
= _request_lock(r
, lkb
);
2437 static int convert_lock(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
2438 struct dlm_args
*args
)
2443 r
= lkb
->lkb_resource
;
2448 error
= validate_lock_args(ls
, lkb
, args
);
2452 error
= _convert_lock(r
, lkb
);
2459 static int unlock_lock(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
2460 struct dlm_args
*args
)
2465 r
= lkb
->lkb_resource
;
2470 error
= validate_unlock_args(lkb
, args
);
2474 error
= _unlock_lock(r
, lkb
);
2481 static int cancel_lock(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
2482 struct dlm_args
*args
)
2487 r
= lkb
->lkb_resource
;
2492 error
= validate_unlock_args(lkb
, args
);
2496 error
= _cancel_lock(r
, lkb
);
2504 * Two stage 1 varieties: dlm_lock() and dlm_unlock()
2507 int dlm_lock(dlm_lockspace_t
*lockspace
,
2509 struct dlm_lksb
*lksb
,
2512 unsigned int namelen
,
2513 uint32_t parent_lkid
,
2514 void (*ast
) (void *astarg
),
2516 void (*bast
) (void *astarg
, int mode
))
2519 struct dlm_lkb
*lkb
;
2520 struct dlm_args args
;
2521 int error
, convert
= flags
& DLM_LKF_CONVERT
;
2523 ls
= dlm_find_lockspace_local(lockspace
);
2527 dlm_lock_recovery(ls
);
2530 error
= find_lkb(ls
, lksb
->sb_lkid
, &lkb
);
2532 error
= create_lkb(ls
, &lkb
);
2537 error
= set_lock_args(mode
, lksb
, flags
, namelen
, 0, ast
,
2538 astarg
, bast
, &args
);
2543 error
= convert_lock(ls
, lkb
, &args
);
2545 error
= request_lock(ls
, lkb
, name
, namelen
, &args
);
2547 if (error
== -EINPROGRESS
)
2550 if (convert
|| error
)
2552 if (error
== -EAGAIN
|| error
== -EDEADLK
)
2555 dlm_unlock_recovery(ls
);
2556 dlm_put_lockspace(ls
);
2560 int dlm_unlock(dlm_lockspace_t
*lockspace
,
2563 struct dlm_lksb
*lksb
,
2567 struct dlm_lkb
*lkb
;
2568 struct dlm_args args
;
2571 ls
= dlm_find_lockspace_local(lockspace
);
2575 dlm_lock_recovery(ls
);
2577 error
= find_lkb(ls
, lkid
, &lkb
);
2581 error
= set_unlock_args(flags
, astarg
, &args
);
2585 if (flags
& DLM_LKF_CANCEL
)
2586 error
= cancel_lock(ls
, lkb
, &args
);
2588 error
= unlock_lock(ls
, lkb
, &args
);
2590 if (error
== -DLM_EUNLOCK
|| error
== -DLM_ECANCEL
)
2592 if (error
== -EBUSY
&& (flags
& (DLM_LKF_CANCEL
| DLM_LKF_FORCEUNLOCK
)))
2597 dlm_unlock_recovery(ls
);
2598 dlm_put_lockspace(ls
);
2603 * send/receive routines for remote operations and replies
2607 * send_request receive_request
2608 * send_convert receive_convert
2609 * send_unlock receive_unlock
2610 * send_cancel receive_cancel
2611 * send_grant receive_grant
2612 * send_bast receive_bast
2613 * send_lookup receive_lookup
2614 * send_remove receive_remove
2617 * receive_request_reply send_request_reply
2618 * receive_convert_reply send_convert_reply
2619 * receive_unlock_reply send_unlock_reply
2620 * receive_cancel_reply send_cancel_reply
2621 * receive_lookup_reply send_lookup_reply
2624 static int _create_message(struct dlm_ls
*ls
, int mb_len
,
2625 int to_nodeid
, int mstype
,
2626 struct dlm_message
**ms_ret
,
2627 struct dlm_mhandle
**mh_ret
)
2629 struct dlm_message
*ms
;
2630 struct dlm_mhandle
*mh
;
2633 /* get_buffer gives us a message handle (mh) that we need to
2634 pass into lowcomms_commit and a message buffer (mb) that we
2635 write our data into */
2637 mh
= dlm_lowcomms_get_buffer(to_nodeid
, mb_len
, ls
->ls_allocation
, &mb
);
2641 memset(mb
, 0, mb_len
);
2643 ms
= (struct dlm_message
*) mb
;
2645 ms
->m_header
.h_version
= (DLM_HEADER_MAJOR
| DLM_HEADER_MINOR
);
2646 ms
->m_header
.h_lockspace
= ls
->ls_global_id
;
2647 ms
->m_header
.h_nodeid
= dlm_our_nodeid();
2648 ms
->m_header
.h_length
= mb_len
;
2649 ms
->m_header
.h_cmd
= DLM_MSG
;
2651 ms
->m_type
= mstype
;
2658 static int create_message(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
2659 int to_nodeid
, int mstype
,
2660 struct dlm_message
**ms_ret
,
2661 struct dlm_mhandle
**mh_ret
)
2663 int mb_len
= sizeof(struct dlm_message
);
2666 case DLM_MSG_REQUEST
:
2667 case DLM_MSG_LOOKUP
:
2668 case DLM_MSG_REMOVE
:
2669 mb_len
+= r
->res_length
;
2671 case DLM_MSG_CONVERT
:
2672 case DLM_MSG_UNLOCK
:
2673 case DLM_MSG_REQUEST_REPLY
:
2674 case DLM_MSG_CONVERT_REPLY
:
2676 if (lkb
&& lkb
->lkb_lvbptr
)
2677 mb_len
+= r
->res_ls
->ls_lvblen
;
2681 return _create_message(r
->res_ls
, mb_len
, to_nodeid
, mstype
,
2685 /* further lowcomms enhancements or alternate implementations may make
2686 the return value from this function useful at some point */
2688 static int send_message(struct dlm_mhandle
*mh
, struct dlm_message
*ms
)
2690 dlm_message_out(ms
);
2691 dlm_lowcomms_commit_buffer(mh
);
2695 static void send_args(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
2696 struct dlm_message
*ms
)
2698 ms
->m_nodeid
= lkb
->lkb_nodeid
;
2699 ms
->m_pid
= lkb
->lkb_ownpid
;
2700 ms
->m_lkid
= lkb
->lkb_id
;
2701 ms
->m_remid
= lkb
->lkb_remid
;
2702 ms
->m_exflags
= lkb
->lkb_exflags
;
2703 ms
->m_sbflags
= lkb
->lkb_sbflags
;
2704 ms
->m_flags
= lkb
->lkb_flags
;
2705 ms
->m_lvbseq
= lkb
->lkb_lvbseq
;
2706 ms
->m_status
= lkb
->lkb_status
;
2707 ms
->m_grmode
= lkb
->lkb_grmode
;
2708 ms
->m_rqmode
= lkb
->lkb_rqmode
;
2709 ms
->m_hash
= r
->res_hash
;
2711 /* m_result and m_bastmode are set from function args,
2712 not from lkb fields */
2714 if (lkb
->lkb_bastaddr
)
2715 ms
->m_asts
|= AST_BAST
;
2716 if (lkb
->lkb_astaddr
)
2717 ms
->m_asts
|= AST_COMP
;
2719 /* compare with switch in create_message; send_remove() doesn't
2722 switch (ms
->m_type
) {
2723 case DLM_MSG_REQUEST
:
2724 case DLM_MSG_LOOKUP
:
2725 memcpy(ms
->m_extra
, r
->res_name
, r
->res_length
);
2727 case DLM_MSG_CONVERT
:
2728 case DLM_MSG_UNLOCK
:
2729 case DLM_MSG_REQUEST_REPLY
:
2730 case DLM_MSG_CONVERT_REPLY
:
2732 if (!lkb
->lkb_lvbptr
)
2734 memcpy(ms
->m_extra
, lkb
->lkb_lvbptr
, r
->res_ls
->ls_lvblen
);
2739 static int send_common(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int mstype
)
2741 struct dlm_message
*ms
;
2742 struct dlm_mhandle
*mh
;
2743 int to_nodeid
, error
;
2745 error
= add_to_waiters(lkb
, mstype
);
2749 to_nodeid
= r
->res_nodeid
;
2751 error
= create_message(r
, lkb
, to_nodeid
, mstype
, &ms
, &mh
);
2755 send_args(r
, lkb
, ms
);
2757 error
= send_message(mh
, ms
);
2763 remove_from_waiters(lkb
, msg_reply_type(mstype
));
2767 static int send_request(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2769 return send_common(r
, lkb
, DLM_MSG_REQUEST
);
2772 static int send_convert(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2776 error
= send_common(r
, lkb
, DLM_MSG_CONVERT
);
2778 /* down conversions go without a reply from the master */
2779 if (!error
&& down_conversion(lkb
)) {
2780 remove_from_waiters(lkb
, DLM_MSG_CONVERT_REPLY
);
2781 r
->res_ls
->ls_stub_ms
.m_type
= DLM_MSG_CONVERT_REPLY
;
2782 r
->res_ls
->ls_stub_ms
.m_result
= 0;
2783 r
->res_ls
->ls_stub_ms
.m_flags
= lkb
->lkb_flags
;
2784 __receive_convert_reply(r
, lkb
, &r
->res_ls
->ls_stub_ms
);
2790 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
2791 MASTER_UNCERTAIN to force the next request on the rsb to confirm
2792 that the master is still correct. */
2794 static int send_unlock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2796 return send_common(r
, lkb
, DLM_MSG_UNLOCK
);
2799 static int send_cancel(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2801 return send_common(r
, lkb
, DLM_MSG_CANCEL
);
2804 static int send_grant(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2806 struct dlm_message
*ms
;
2807 struct dlm_mhandle
*mh
;
2808 int to_nodeid
, error
;
2810 to_nodeid
= lkb
->lkb_nodeid
;
2812 error
= create_message(r
, lkb
, to_nodeid
, DLM_MSG_GRANT
, &ms
, &mh
);
2816 send_args(r
, lkb
, ms
);
2820 error
= send_message(mh
, ms
);
2825 static int send_bast(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int mode
)
2827 struct dlm_message
*ms
;
2828 struct dlm_mhandle
*mh
;
2829 int to_nodeid
, error
;
2831 to_nodeid
= lkb
->lkb_nodeid
;
2833 error
= create_message(r
, NULL
, to_nodeid
, DLM_MSG_BAST
, &ms
, &mh
);
2837 send_args(r
, lkb
, ms
);
2839 ms
->m_bastmode
= mode
;
2841 error
= send_message(mh
, ms
);
2846 static int send_lookup(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
2848 struct dlm_message
*ms
;
2849 struct dlm_mhandle
*mh
;
2850 int to_nodeid
, error
;
2852 error
= add_to_waiters(lkb
, DLM_MSG_LOOKUP
);
2856 to_nodeid
= dlm_dir_nodeid(r
);
2858 error
= create_message(r
, NULL
, to_nodeid
, DLM_MSG_LOOKUP
, &ms
, &mh
);
2862 send_args(r
, lkb
, ms
);
2864 error
= send_message(mh
, ms
);
2870 remove_from_waiters(lkb
, DLM_MSG_LOOKUP_REPLY
);
2874 static int send_remove(struct dlm_rsb
*r
)
2876 struct dlm_message
*ms
;
2877 struct dlm_mhandle
*mh
;
2878 int to_nodeid
, error
;
2880 to_nodeid
= dlm_dir_nodeid(r
);
2882 error
= create_message(r
, NULL
, to_nodeid
, DLM_MSG_REMOVE
, &ms
, &mh
);
2886 memcpy(ms
->m_extra
, r
->res_name
, r
->res_length
);
2887 ms
->m_hash
= r
->res_hash
;
2889 error
= send_message(mh
, ms
);
2894 static int send_common_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
2897 struct dlm_message
*ms
;
2898 struct dlm_mhandle
*mh
;
2899 int to_nodeid
, error
;
2901 to_nodeid
= lkb
->lkb_nodeid
;
2903 error
= create_message(r
, lkb
, to_nodeid
, mstype
, &ms
, &mh
);
2907 send_args(r
, lkb
, ms
);
2911 error
= send_message(mh
, ms
);
2916 static int send_request_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int rv
)
2918 return send_common_reply(r
, lkb
, DLM_MSG_REQUEST_REPLY
, rv
);
2921 static int send_convert_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int rv
)
2923 return send_common_reply(r
, lkb
, DLM_MSG_CONVERT_REPLY
, rv
);
2926 static int send_unlock_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int rv
)
2928 return send_common_reply(r
, lkb
, DLM_MSG_UNLOCK_REPLY
, rv
);
2931 static int send_cancel_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
, int rv
)
2933 return send_common_reply(r
, lkb
, DLM_MSG_CANCEL_REPLY
, rv
);
2936 static int send_lookup_reply(struct dlm_ls
*ls
, struct dlm_message
*ms_in
,
2937 int ret_nodeid
, int rv
)
2939 struct dlm_rsb
*r
= &ls
->ls_stub_rsb
;
2940 struct dlm_message
*ms
;
2941 struct dlm_mhandle
*mh
;
2942 int error
, nodeid
= ms_in
->m_header
.h_nodeid
;
2944 error
= create_message(r
, NULL
, nodeid
, DLM_MSG_LOOKUP_REPLY
, &ms
, &mh
);
2948 ms
->m_lkid
= ms_in
->m_lkid
;
2950 ms
->m_nodeid
= ret_nodeid
;
2952 error
= send_message(mh
, ms
);
2957 /* which args we save from a received message depends heavily on the type
2958 of message, unlike the send side where we can safely send everything about
2959 the lkb for any type of message */
2961 static void receive_flags(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
2963 lkb
->lkb_exflags
= ms
->m_exflags
;
2964 lkb
->lkb_sbflags
= ms
->m_sbflags
;
2965 lkb
->lkb_flags
= (lkb
->lkb_flags
& 0xFFFF0000) |
2966 (ms
->m_flags
& 0x0000FFFF);
2969 static void receive_flags_reply(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
2971 lkb
->lkb_sbflags
= ms
->m_sbflags
;
2972 lkb
->lkb_flags
= (lkb
->lkb_flags
& 0xFFFF0000) |
2973 (ms
->m_flags
& 0x0000FFFF);
2976 static int receive_extralen(struct dlm_message
*ms
)
2978 return (ms
->m_header
.h_length
- sizeof(struct dlm_message
));
2981 static int receive_lvb(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
2982 struct dlm_message
*ms
)
2986 if (lkb
->lkb_exflags
& DLM_LKF_VALBLK
) {
2987 if (!lkb
->lkb_lvbptr
)
2988 lkb
->lkb_lvbptr
= dlm_allocate_lvb(ls
);
2989 if (!lkb
->lkb_lvbptr
)
2991 len
= receive_extralen(ms
);
2992 memcpy(lkb
->lkb_lvbptr
, ms
->m_extra
, len
);
2997 static int receive_request_args(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
2998 struct dlm_message
*ms
)
3000 lkb
->lkb_nodeid
= ms
->m_header
.h_nodeid
;
3001 lkb
->lkb_ownpid
= ms
->m_pid
;
3002 lkb
->lkb_remid
= ms
->m_lkid
;
3003 lkb
->lkb_grmode
= DLM_LOCK_IV
;
3004 lkb
->lkb_rqmode
= ms
->m_rqmode
;
3005 lkb
->lkb_bastaddr
= (void *) (long) (ms
->m_asts
& AST_BAST
);
3006 lkb
->lkb_astaddr
= (void *) (long) (ms
->m_asts
& AST_COMP
);
3008 DLM_ASSERT(is_master_copy(lkb
), dlm_print_lkb(lkb
););
3010 if (lkb
->lkb_exflags
& DLM_LKF_VALBLK
) {
3011 /* lkb was just created so there won't be an lvb yet */
3012 lkb
->lkb_lvbptr
= dlm_allocate_lvb(ls
);
3013 if (!lkb
->lkb_lvbptr
)
3020 static int receive_convert_args(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
3021 struct dlm_message
*ms
)
3023 if (lkb
->lkb_nodeid
!= ms
->m_header
.h_nodeid
) {
3024 log_error(ls
, "convert_args nodeid %d %d lkid %x %x",
3025 lkb
->lkb_nodeid
, ms
->m_header
.h_nodeid
,
3026 lkb
->lkb_id
, lkb
->lkb_remid
);
3030 if (!is_master_copy(lkb
))
3033 if (lkb
->lkb_status
!= DLM_LKSTS_GRANTED
)
3036 if (receive_lvb(ls
, lkb
, ms
))
3039 lkb
->lkb_rqmode
= ms
->m_rqmode
;
3040 lkb
->lkb_lvbseq
= ms
->m_lvbseq
;
3045 static int receive_unlock_args(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
3046 struct dlm_message
*ms
)
3048 if (!is_master_copy(lkb
))
3050 if (receive_lvb(ls
, lkb
, ms
))
3055 /* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3056 uses to send a reply and that the remote end uses to process the reply. */
3058 static void setup_stub_lkb(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3060 struct dlm_lkb
*lkb
= &ls
->ls_stub_lkb
;
3061 lkb
->lkb_nodeid
= ms
->m_header
.h_nodeid
;
3062 lkb
->lkb_remid
= ms
->m_lkid
;
3065 static void receive_request(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3067 struct dlm_lkb
*lkb
;
3071 error
= create_lkb(ls
, &lkb
);
3075 receive_flags(lkb
, ms
);
3076 lkb
->lkb_flags
|= DLM_IFL_MSTCPY
;
3077 error
= receive_request_args(ls
, lkb
, ms
);
3083 namelen
= receive_extralen(ms
);
3085 error
= find_rsb(ls
, ms
->m_extra
, namelen
, R_MASTER
, &r
);
3094 error
= do_request(r
, lkb
);
3095 send_request_reply(r
, lkb
, error
);
3100 if (error
== -EINPROGRESS
)
3107 setup_stub_lkb(ls
, ms
);
3108 send_request_reply(&ls
->ls_stub_rsb
, &ls
->ls_stub_lkb
, error
);
3111 static void receive_convert(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3113 struct dlm_lkb
*lkb
;
3115 int error
, reply
= 1;
3117 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3121 r
= lkb
->lkb_resource
;
3126 receive_flags(lkb
, ms
);
3127 error
= receive_convert_args(ls
, lkb
, ms
);
3130 reply
= !down_conversion(lkb
);
3132 error
= do_convert(r
, lkb
);
3135 send_convert_reply(r
, lkb
, error
);
3143 setup_stub_lkb(ls
, ms
);
3144 send_convert_reply(&ls
->ls_stub_rsb
, &ls
->ls_stub_lkb
, error
);
3147 static void receive_unlock(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3149 struct dlm_lkb
*lkb
;
3153 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3157 r
= lkb
->lkb_resource
;
3162 receive_flags(lkb
, ms
);
3163 error
= receive_unlock_args(ls
, lkb
, ms
);
3167 error
= do_unlock(r
, lkb
);
3169 send_unlock_reply(r
, lkb
, error
);
3177 setup_stub_lkb(ls
, ms
);
3178 send_unlock_reply(&ls
->ls_stub_rsb
, &ls
->ls_stub_lkb
, error
);
3181 static void receive_cancel(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3183 struct dlm_lkb
*lkb
;
3187 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3191 receive_flags(lkb
, ms
);
3193 r
= lkb
->lkb_resource
;
3198 error
= do_cancel(r
, lkb
);
3199 send_cancel_reply(r
, lkb
, error
);
3207 setup_stub_lkb(ls
, ms
);
3208 send_cancel_reply(&ls
->ls_stub_rsb
, &ls
->ls_stub_lkb
, error
);
3211 static void receive_grant(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3213 struct dlm_lkb
*lkb
;
3217 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3219 log_error(ls
, "receive_grant no lkb");
3222 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
3224 r
= lkb
->lkb_resource
;
3229 receive_flags_reply(lkb
, ms
);
3230 if (is_altmode(lkb
))
3231 munge_altmode(lkb
, ms
);
3232 grant_lock_pc(r
, lkb
, ms
);
3233 queue_cast(r
, lkb
, 0);
3240 static void receive_bast(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3242 struct dlm_lkb
*lkb
;
3246 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3248 log_error(ls
, "receive_bast no lkb");
3251 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
3253 r
= lkb
->lkb_resource
;
3258 queue_bast(r
, lkb
, ms
->m_bastmode
);
3265 static void receive_lookup(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3267 int len
, error
, ret_nodeid
, dir_nodeid
, from_nodeid
, our_nodeid
;
3269 from_nodeid
= ms
->m_header
.h_nodeid
;
3270 our_nodeid
= dlm_our_nodeid();
3272 len
= receive_extralen(ms
);
3274 dir_nodeid
= dlm_hash2nodeid(ls
, ms
->m_hash
);
3275 if (dir_nodeid
!= our_nodeid
) {
3276 log_error(ls
, "lookup dir_nodeid %d from %d",
3277 dir_nodeid
, from_nodeid
);
3283 error
= dlm_dir_lookup(ls
, from_nodeid
, ms
->m_extra
, len
, &ret_nodeid
);
3285 /* Optimization: we're master so treat lookup as a request */
3286 if (!error
&& ret_nodeid
== our_nodeid
) {
3287 receive_request(ls
, ms
);
3291 send_lookup_reply(ls
, ms
, ret_nodeid
, error
);
3294 static void receive_remove(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3296 int len
, dir_nodeid
, from_nodeid
;
3298 from_nodeid
= ms
->m_header
.h_nodeid
;
3300 len
= receive_extralen(ms
);
3302 dir_nodeid
= dlm_hash2nodeid(ls
, ms
->m_hash
);
3303 if (dir_nodeid
!= dlm_our_nodeid()) {
3304 log_error(ls
, "remove dir entry dir_nodeid %d from %d",
3305 dir_nodeid
, from_nodeid
);
3309 dlm_dir_remove_entry(ls
, from_nodeid
, ms
->m_extra
, len
);
3312 static void receive_purge(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3314 do_purge(ls
, ms
->m_nodeid
, ms
->m_pid
);
3317 static void receive_request_reply(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3319 struct dlm_lkb
*lkb
;
3321 int error
, mstype
, result
;
3323 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3325 log_error(ls
, "receive_request_reply no lkb");
3328 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
3330 r
= lkb
->lkb_resource
;
3334 mstype
= lkb
->lkb_wait_type
;
3335 error
= remove_from_waiters(lkb
, DLM_MSG_REQUEST_REPLY
);
3339 /* Optimization: the dir node was also the master, so it took our
3340 lookup as a request and sent request reply instead of lookup reply */
3341 if (mstype
== DLM_MSG_LOOKUP
) {
3342 r
->res_nodeid
= ms
->m_header
.h_nodeid
;
3343 lkb
->lkb_nodeid
= r
->res_nodeid
;
3346 /* this is the value returned from do_request() on the master */
3347 result
= ms
->m_result
;
3351 /* request would block (be queued) on remote master */
3352 queue_cast(r
, lkb
, -EAGAIN
);
3353 confirm_master(r
, -EAGAIN
);
3354 unhold_lkb(lkb
); /* undoes create_lkb() */
3359 /* request was queued or granted on remote master */
3360 receive_flags_reply(lkb
, ms
);
3361 lkb
->lkb_remid
= ms
->m_lkid
;
3362 if (is_altmode(lkb
))
3363 munge_altmode(lkb
, ms
);
3365 add_lkb(r
, lkb
, DLM_LKSTS_WAITING
);
3368 grant_lock_pc(r
, lkb
, ms
);
3369 queue_cast(r
, lkb
, 0);
3371 confirm_master(r
, result
);
3376 /* find_rsb failed to find rsb or rsb wasn't master */
3377 log_debug(ls
, "receive_request_reply %x %x master diff %d %d",
3378 lkb
->lkb_id
, lkb
->lkb_flags
, r
->res_nodeid
, result
);
3380 lkb
->lkb_nodeid
= -1;
3382 if (is_overlap(lkb
)) {
3383 /* we'll ignore error in cancel/unlock reply */
3384 queue_cast_overlap(r
, lkb
);
3385 unhold_lkb(lkb
); /* undoes create_lkb() */
3387 _request_lock(r
, lkb
);
3391 log_error(ls
, "receive_request_reply %x error %d",
3392 lkb
->lkb_id
, result
);
3395 if (is_overlap_unlock(lkb
) && (result
== 0 || result
== -EINPROGRESS
)) {
3396 log_debug(ls
, "receive_request_reply %x result %d unlock",
3397 lkb
->lkb_id
, result
);
3398 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_UNLOCK
;
3399 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_CANCEL
;
3400 send_unlock(r
, lkb
);
3401 } else if (is_overlap_cancel(lkb
) && (result
== -EINPROGRESS
)) {
3402 log_debug(ls
, "receive_request_reply %x cancel", lkb
->lkb_id
);
3403 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_UNLOCK
;
3404 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_CANCEL
;
3405 send_cancel(r
, lkb
);
3407 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_CANCEL
;
3408 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_UNLOCK
;
3416 static void __receive_convert_reply(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
3417 struct dlm_message
*ms
)
3419 /* this is the value returned from do_convert() on the master */
3420 switch (ms
->m_result
) {
3422 /* convert would block (be queued) on remote master */
3423 queue_cast(r
, lkb
, -EAGAIN
);
3427 receive_flags_reply(lkb
, ms
);
3428 revert_lock_pc(r
, lkb
);
3429 queue_cast(r
, lkb
, -EDEADLK
);
3433 /* convert was queued on remote master */
3434 receive_flags_reply(lkb
, ms
);
3435 if (is_demoted(lkb
))
3436 munge_demoted(lkb
, ms
);
3438 add_lkb(r
, lkb
, DLM_LKSTS_CONVERT
);
3443 /* convert was granted on remote master */
3444 receive_flags_reply(lkb
, ms
);
3445 if (is_demoted(lkb
))
3446 munge_demoted(lkb
, ms
);
3447 grant_lock_pc(r
, lkb
, ms
);
3448 queue_cast(r
, lkb
, 0);
3452 log_error(r
->res_ls
, "receive_convert_reply %x error %d",
3453 lkb
->lkb_id
, ms
->m_result
);
3457 static void _receive_convert_reply(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
3459 struct dlm_rsb
*r
= lkb
->lkb_resource
;
3465 /* stub reply can happen with waiters_mutex held */
3466 error
= remove_from_waiters_ms(lkb
, ms
);
3470 __receive_convert_reply(r
, lkb
, ms
);
3476 static void receive_convert_reply(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3478 struct dlm_lkb
*lkb
;
3481 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3483 log_error(ls
, "receive_convert_reply no lkb");
3486 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
3488 _receive_convert_reply(lkb
, ms
);
3492 static void _receive_unlock_reply(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
3494 struct dlm_rsb
*r
= lkb
->lkb_resource
;
3500 /* stub reply can happen with waiters_mutex held */
3501 error
= remove_from_waiters_ms(lkb
, ms
);
3505 /* this is the value returned from do_unlock() on the master */
3507 switch (ms
->m_result
) {
3509 receive_flags_reply(lkb
, ms
);
3510 remove_lock_pc(r
, lkb
);
3511 queue_cast(r
, lkb
, -DLM_EUNLOCK
);
3516 log_error(r
->res_ls
, "receive_unlock_reply %x error %d",
3517 lkb
->lkb_id
, ms
->m_result
);
3524 static void receive_unlock_reply(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3526 struct dlm_lkb
*lkb
;
3529 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3531 log_error(ls
, "receive_unlock_reply no lkb");
3534 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
3536 _receive_unlock_reply(lkb
, ms
);
3540 static void _receive_cancel_reply(struct dlm_lkb
*lkb
, struct dlm_message
*ms
)
3542 struct dlm_rsb
*r
= lkb
->lkb_resource
;
3548 /* stub reply can happen with waiters_mutex held */
3549 error
= remove_from_waiters_ms(lkb
, ms
);
3553 /* this is the value returned from do_cancel() on the master */
3555 switch (ms
->m_result
) {
3557 receive_flags_reply(lkb
, ms
);
3558 revert_lock_pc(r
, lkb
);
3559 queue_cast(r
, lkb
, -DLM_ECANCEL
);
3564 log_error(r
->res_ls
, "receive_cancel_reply %x error %d",
3565 lkb
->lkb_id
, ms
->m_result
);
3572 static void receive_cancel_reply(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3574 struct dlm_lkb
*lkb
;
3577 error
= find_lkb(ls
, ms
->m_remid
, &lkb
);
3579 log_error(ls
, "receive_cancel_reply no lkb");
3582 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
3584 _receive_cancel_reply(lkb
, ms
);
3588 static void receive_lookup_reply(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3590 struct dlm_lkb
*lkb
;
3592 int error
, ret_nodeid
;
3594 error
= find_lkb(ls
, ms
->m_lkid
, &lkb
);
3596 log_error(ls
, "receive_lookup_reply no lkb");
3600 /* ms->m_result is the value returned by dlm_dir_lookup on dir node
3601 FIXME: will a non-zero error ever be returned? */
3603 r
= lkb
->lkb_resource
;
3607 error
= remove_from_waiters(lkb
, DLM_MSG_LOOKUP_REPLY
);
3611 ret_nodeid
= ms
->m_nodeid
;
3612 if (ret_nodeid
== dlm_our_nodeid()) {
3615 r
->res_first_lkid
= 0;
3617 /* set_master() will copy res_nodeid to lkb_nodeid */
3618 r
->res_nodeid
= ret_nodeid
;
3621 if (is_overlap(lkb
)) {
3622 log_debug(ls
, "receive_lookup_reply %x unlock %x",
3623 lkb
->lkb_id
, lkb
->lkb_flags
);
3624 queue_cast_overlap(r
, lkb
);
3625 unhold_lkb(lkb
); /* undoes create_lkb() */
3629 _request_lock(r
, lkb
);
3633 process_lookup_list(r
);
3640 static void _receive_message(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3642 switch (ms
->m_type
) {
3644 /* messages sent to a master node */
3646 case DLM_MSG_REQUEST
:
3647 receive_request(ls
, ms
);
3650 case DLM_MSG_CONVERT
:
3651 receive_convert(ls
, ms
);
3654 case DLM_MSG_UNLOCK
:
3655 receive_unlock(ls
, ms
);
3658 case DLM_MSG_CANCEL
:
3659 receive_cancel(ls
, ms
);
3662 /* messages sent from a master node (replies to above) */
3664 case DLM_MSG_REQUEST_REPLY
:
3665 receive_request_reply(ls
, ms
);
3668 case DLM_MSG_CONVERT_REPLY
:
3669 receive_convert_reply(ls
, ms
);
3672 case DLM_MSG_UNLOCK_REPLY
:
3673 receive_unlock_reply(ls
, ms
);
3676 case DLM_MSG_CANCEL_REPLY
:
3677 receive_cancel_reply(ls
, ms
);
3680 /* messages sent from a master node (only two types of async msg) */
3683 receive_grant(ls
, ms
);
3687 receive_bast(ls
, ms
);
3690 /* messages sent to a dir node */
3692 case DLM_MSG_LOOKUP
:
3693 receive_lookup(ls
, ms
);
3696 case DLM_MSG_REMOVE
:
3697 receive_remove(ls
, ms
);
3700 /* messages sent from a dir node (remove has no reply) */
3702 case DLM_MSG_LOOKUP_REPLY
:
3703 receive_lookup_reply(ls
, ms
);
3706 /* other messages */
3709 receive_purge(ls
, ms
);
3713 log_error(ls
, "unknown message type %d", ms
->m_type
);
3719 /* If the lockspace is in recovery mode (locking stopped), then normal
3720 messages are saved on the requestqueue for processing after recovery is
3721 done. When not in recovery mode, we wait for dlm_recoverd to drain saved
3722 messages off the requestqueue before we process new ones. This occurs right
3723 after recovery completes when we transition from saving all messages on
3724 requestqueue, to processing all the saved messages, to processing new
3725 messages as they arrive. */
3727 static void dlm_receive_message(struct dlm_ls
*ls
, struct dlm_message
*ms
,
3730 if (dlm_locking_stopped(ls
)) {
3731 dlm_add_requestqueue(ls
, nodeid
, (struct dlm_header
*) ms
);
3733 dlm_wait_requestqueue(ls
);
3734 _receive_message(ls
, ms
);
3738 /* This is called by dlm_recoverd to process messages that were saved on
3739 the requestqueue. */
3741 void dlm_receive_message_saved(struct dlm_ls
*ls
, struct dlm_message
*ms
)
3743 _receive_message(ls
, ms
);
3746 /* This is called by the midcomms layer when something is received for
3747 the lockspace. It could be either a MSG (normal message sent as part of
3748 standard locking activity) or an RCOM (recovery message sent as part of
3749 lockspace recovery). */
3751 void dlm_receive_buffer(struct dlm_header
*hd
, int nodeid
)
3753 struct dlm_message
*ms
= (struct dlm_message
*) hd
;
3754 struct dlm_rcom
*rc
= (struct dlm_rcom
*) hd
;
3758 switch (hd
->h_cmd
) {
3768 log_print("invalid h_cmd %d from %u", hd
->h_cmd
, nodeid
);
3772 if (hd
->h_nodeid
!= nodeid
) {
3773 log_print("invalid h_nodeid %d from %d lockspace %x",
3774 hd
->h_nodeid
, nodeid
, hd
->h_lockspace
);
3778 ls
= dlm_find_lockspace_global(hd
->h_lockspace
);
3780 log_print("invalid h_lockspace %x from %d cmd %d type %d",
3781 hd
->h_lockspace
, nodeid
, hd
->h_cmd
, type
);
3783 if (hd
->h_cmd
== DLM_RCOM
&& type
== DLM_RCOM_STATUS
)
3784 dlm_send_ls_not_ready(nodeid
, rc
);
3788 /* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3789 be inactive (in this ls) before transitioning to recovery mode */
3791 down_read(&ls
->ls_recv_active
);
3792 if (hd
->h_cmd
== DLM_MSG
)
3793 dlm_receive_message(ls
, ms
, nodeid
);
3795 dlm_receive_rcom(ls
, rc
, nodeid
);
3796 up_read(&ls
->ls_recv_active
);
3798 dlm_put_lockspace(ls
);
3801 static void recover_convert_waiter(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
3803 if (middle_conversion(lkb
)) {
3805 ls
->ls_stub_ms
.m_type
= DLM_MSG_CONVERT_REPLY
;
3806 ls
->ls_stub_ms
.m_result
= -EINPROGRESS
;
3807 ls
->ls_stub_ms
.m_flags
= lkb
->lkb_flags
;
3808 _receive_convert_reply(lkb
, &ls
->ls_stub_ms
);
3810 /* Same special case as in receive_rcom_lock_args() */
3811 lkb
->lkb_grmode
= DLM_LOCK_IV
;
3812 rsb_set_flag(lkb
->lkb_resource
, RSB_RECOVER_CONVERT
);
3815 } else if (lkb
->lkb_rqmode
>= lkb
->lkb_grmode
) {
3816 lkb
->lkb_flags
|= DLM_IFL_RESEND
;
3819 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
3820 conversions are async; there's no reply from the remote master */
3823 /* A waiting lkb needs recovery if the master node has failed, or
3824 the master node is changing (only when no directory is used) */
3826 static int waiter_needs_recovery(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
3828 if (dlm_is_removed(ls
, lkb
->lkb_nodeid
))
3831 if (!dlm_no_directory(ls
))
3834 if (dlm_dir_nodeid(lkb
->lkb_resource
) != lkb
->lkb_nodeid
)
3840 /* Recovery for locks that are waiting for replies from nodes that are now
3841 gone. We can just complete unlocks and cancels by faking a reply from the
3842 dead node. Requests and up-conversions we flag to be resent after
3843 recovery. Down-conversions can just be completed with a fake reply like
3844 unlocks. Conversions between PR and CW need special attention. */
3846 void dlm_recover_waiters_pre(struct dlm_ls
*ls
)
3848 struct dlm_lkb
*lkb
, *safe
;
3849 int wait_type
, stub_unlock_result
, stub_cancel_result
;
3851 mutex_lock(&ls
->ls_waiters_mutex
);
3853 list_for_each_entry_safe(lkb
, safe
, &ls
->ls_waiters
, lkb_wait_reply
) {
3854 log_debug(ls
, "pre recover waiter lkid %x type %d flags %x",
3855 lkb
->lkb_id
, lkb
->lkb_wait_type
, lkb
->lkb_flags
);
3857 /* all outstanding lookups, regardless of destination will be
3858 resent after recovery is done */
3860 if (lkb
->lkb_wait_type
== DLM_MSG_LOOKUP
) {
3861 lkb
->lkb_flags
|= DLM_IFL_RESEND
;
3865 if (!waiter_needs_recovery(ls
, lkb
))
3868 wait_type
= lkb
->lkb_wait_type
;
3869 stub_unlock_result
= -DLM_EUNLOCK
;
3870 stub_cancel_result
= -DLM_ECANCEL
;
3872 /* Main reply may have been received leaving a zero wait_type,
3873 but a reply for the overlapping op may not have been
3874 received. In that case we need to fake the appropriate
3875 reply for the overlap op. */
3878 if (is_overlap_cancel(lkb
)) {
3879 wait_type
= DLM_MSG_CANCEL
;
3880 if (lkb
->lkb_grmode
== DLM_LOCK_IV
)
3881 stub_cancel_result
= 0;
3883 if (is_overlap_unlock(lkb
)) {
3884 wait_type
= DLM_MSG_UNLOCK
;
3885 if (lkb
->lkb_grmode
== DLM_LOCK_IV
)
3886 stub_unlock_result
= -ENOENT
;
3889 log_debug(ls
, "rwpre overlap %x %x %d %d %d",
3890 lkb
->lkb_id
, lkb
->lkb_flags
, wait_type
,
3891 stub_cancel_result
, stub_unlock_result
);
3894 switch (wait_type
) {
3896 case DLM_MSG_REQUEST
:
3897 lkb
->lkb_flags
|= DLM_IFL_RESEND
;
3900 case DLM_MSG_CONVERT
:
3901 recover_convert_waiter(ls
, lkb
);
3904 case DLM_MSG_UNLOCK
:
3906 ls
->ls_stub_ms
.m_type
= DLM_MSG_UNLOCK_REPLY
;
3907 ls
->ls_stub_ms
.m_result
= stub_unlock_result
;
3908 ls
->ls_stub_ms
.m_flags
= lkb
->lkb_flags
;
3909 _receive_unlock_reply(lkb
, &ls
->ls_stub_ms
);
3913 case DLM_MSG_CANCEL
:
3915 ls
->ls_stub_ms
.m_type
= DLM_MSG_CANCEL_REPLY
;
3916 ls
->ls_stub_ms
.m_result
= stub_cancel_result
;
3917 ls
->ls_stub_ms
.m_flags
= lkb
->lkb_flags
;
3918 _receive_cancel_reply(lkb
, &ls
->ls_stub_ms
);
3923 log_error(ls
, "invalid lkb wait_type %d %d",
3924 lkb
->lkb_wait_type
, wait_type
);
3928 mutex_unlock(&ls
->ls_waiters_mutex
);
3931 static struct dlm_lkb
*find_resend_waiter(struct dlm_ls
*ls
)
3933 struct dlm_lkb
*lkb
;
3936 mutex_lock(&ls
->ls_waiters_mutex
);
3937 list_for_each_entry(lkb
, &ls
->ls_waiters
, lkb_wait_reply
) {
3938 if (lkb
->lkb_flags
& DLM_IFL_RESEND
) {
3944 mutex_unlock(&ls
->ls_waiters_mutex
);
3951 /* Deal with lookups and lkb's marked RESEND from _pre. We may now be the
3952 master or dir-node for r. Processing the lkb may result in it being placed
3955 /* We do this after normal locking has been enabled and any saved messages
3956 (in requestqueue) have been processed. We should be confident that at
3957 this point we won't get or process a reply to any of these waiting
3958 operations. But, new ops may be coming in on the rsbs/locks here from
3959 userspace or remotely. */
3961 /* there may have been an overlap unlock/cancel prior to recovery or after
3962 recovery. if before, the lkb may still have a pos wait_count; if after, the
3963 overlap flag would just have been set and nothing new sent. we can be
3964 confident here than any replies to either the initial op or overlap ops
3965 prior to recovery have been received. */
3967 int dlm_recover_waiters_post(struct dlm_ls
*ls
)
3969 struct dlm_lkb
*lkb
;
3971 int error
= 0, mstype
, err
, oc
, ou
;
3974 if (dlm_locking_stopped(ls
)) {
3975 log_debug(ls
, "recover_waiters_post aborted");
3980 lkb
= find_resend_waiter(ls
);
3984 r
= lkb
->lkb_resource
;
3988 mstype
= lkb
->lkb_wait_type
;
3989 oc
= is_overlap_cancel(lkb
);
3990 ou
= is_overlap_unlock(lkb
);
3993 log_debug(ls
, "recover_waiters_post %x type %d flags %x %s",
3994 lkb
->lkb_id
, mstype
, lkb
->lkb_flags
, r
->res_name
);
3996 /* At this point we assume that we won't get a reply to any
3997 previous op or overlap op on this lock. First, do a big
3998 remove_from_waiters() for all previous ops. */
4000 lkb
->lkb_flags
&= ~DLM_IFL_RESEND
;
4001 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_UNLOCK
;
4002 lkb
->lkb_flags
&= ~DLM_IFL_OVERLAP_CANCEL
;
4003 lkb
->lkb_wait_type
= 0;
4004 lkb
->lkb_wait_count
= 0;
4005 mutex_lock(&ls
->ls_waiters_mutex
);
4006 list_del_init(&lkb
->lkb_wait_reply
);
4007 mutex_unlock(&ls
->ls_waiters_mutex
);
4008 unhold_lkb(lkb
); /* for waiters list */
4011 /* do an unlock or cancel instead of resending */
4013 case DLM_MSG_LOOKUP
:
4014 case DLM_MSG_REQUEST
:
4015 queue_cast(r
, lkb
, ou
? -DLM_EUNLOCK
:
4017 unhold_lkb(lkb
); /* undoes create_lkb() */
4019 case DLM_MSG_CONVERT
:
4021 queue_cast(r
, lkb
, -DLM_ECANCEL
);
4023 lkb
->lkb_exflags
|= DLM_LKF_FORCEUNLOCK
;
4024 _unlock_lock(r
, lkb
);
4032 case DLM_MSG_LOOKUP
:
4033 case DLM_MSG_REQUEST
:
4034 _request_lock(r
, lkb
);
4036 confirm_master(r
, 0);
4038 case DLM_MSG_CONVERT
:
4039 _convert_lock(r
, lkb
);
4047 log_error(ls
, "recover_waiters_post %x %d %x %d %d",
4048 lkb
->lkb_id
, mstype
, lkb
->lkb_flags
, oc
, ou
);
4057 static void purge_queue(struct dlm_rsb
*r
, struct list_head
*queue
,
4058 int (*test
)(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
))
4060 struct dlm_ls
*ls
= r
->res_ls
;
4061 struct dlm_lkb
*lkb
, *safe
;
4063 list_for_each_entry_safe(lkb
, safe
, queue
, lkb_statequeue
) {
4064 if (test(ls
, lkb
)) {
4065 rsb_set_flag(r
, RSB_LOCKS_PURGED
);
4067 /* this put should free the lkb */
4068 if (!dlm_put_lkb(lkb
))
4069 log_error(ls
, "purged lkb not released");
4074 static int purge_dead_test(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
4076 return (is_master_copy(lkb
) && dlm_is_removed(ls
, lkb
->lkb_nodeid
));
4079 static int purge_mstcpy_test(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
4081 return is_master_copy(lkb
);
4084 static void purge_dead_locks(struct dlm_rsb
*r
)
4086 purge_queue(r
, &r
->res_grantqueue
, &purge_dead_test
);
4087 purge_queue(r
, &r
->res_convertqueue
, &purge_dead_test
);
4088 purge_queue(r
, &r
->res_waitqueue
, &purge_dead_test
);
4091 void dlm_purge_mstcpy_locks(struct dlm_rsb
*r
)
4093 purge_queue(r
, &r
->res_grantqueue
, &purge_mstcpy_test
);
4094 purge_queue(r
, &r
->res_convertqueue
, &purge_mstcpy_test
);
4095 purge_queue(r
, &r
->res_waitqueue
, &purge_mstcpy_test
);
4098 /* Get rid of locks held by nodes that are gone. */
4100 int dlm_purge_locks(struct dlm_ls
*ls
)
4104 log_debug(ls
, "dlm_purge_locks");
4106 down_write(&ls
->ls_root_sem
);
4107 list_for_each_entry(r
, &ls
->ls_root_list
, res_root_list
) {
4111 purge_dead_locks(r
);
4117 up_write(&ls
->ls_root_sem
);
4122 static struct dlm_rsb
*find_purged_rsb(struct dlm_ls
*ls
, int bucket
)
4124 struct dlm_rsb
*r
, *r_ret
= NULL
;
4126 read_lock(&ls
->ls_rsbtbl
[bucket
].lock
);
4127 list_for_each_entry(r
, &ls
->ls_rsbtbl
[bucket
].list
, res_hashchain
) {
4128 if (!rsb_flag(r
, RSB_LOCKS_PURGED
))
4131 rsb_clear_flag(r
, RSB_LOCKS_PURGED
);
4135 read_unlock(&ls
->ls_rsbtbl
[bucket
].lock
);
4139 void dlm_grant_after_purge(struct dlm_ls
*ls
)
4145 r
= find_purged_rsb(ls
, bucket
);
4147 if (bucket
== ls
->ls_rsbtbl_size
- 1)
4154 grant_pending_locks(r
);
4155 confirm_master(r
, 0);
4163 static struct dlm_lkb
*search_remid_list(struct list_head
*head
, int nodeid
,
4166 struct dlm_lkb
*lkb
;
4168 list_for_each_entry(lkb
, head
, lkb_statequeue
) {
4169 if (lkb
->lkb_nodeid
== nodeid
&& lkb
->lkb_remid
== remid
)
4175 static struct dlm_lkb
*search_remid(struct dlm_rsb
*r
, int nodeid
,
4178 struct dlm_lkb
*lkb
;
4180 lkb
= search_remid_list(&r
->res_grantqueue
, nodeid
, remid
);
4183 lkb
= search_remid_list(&r
->res_convertqueue
, nodeid
, remid
);
4186 lkb
= search_remid_list(&r
->res_waitqueue
, nodeid
, remid
);
4192 static int receive_rcom_lock_args(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
,
4193 struct dlm_rsb
*r
, struct dlm_rcom
*rc
)
4195 struct rcom_lock
*rl
= (struct rcom_lock
*) rc
->rc_buf
;
4198 lkb
->lkb_nodeid
= rc
->rc_header
.h_nodeid
;
4199 lkb
->lkb_ownpid
= rl
->rl_ownpid
;
4200 lkb
->lkb_remid
= rl
->rl_lkid
;
4201 lkb
->lkb_exflags
= rl
->rl_exflags
;
4202 lkb
->lkb_flags
= rl
->rl_flags
& 0x0000FFFF;
4203 lkb
->lkb_flags
|= DLM_IFL_MSTCPY
;
4204 lkb
->lkb_lvbseq
= rl
->rl_lvbseq
;
4205 lkb
->lkb_rqmode
= rl
->rl_rqmode
;
4206 lkb
->lkb_grmode
= rl
->rl_grmode
;
4207 /* don't set lkb_status because add_lkb wants to itself */
4209 lkb
->lkb_bastaddr
= (void *) (long) (rl
->rl_asts
& AST_BAST
);
4210 lkb
->lkb_astaddr
= (void *) (long) (rl
->rl_asts
& AST_COMP
);
4212 if (lkb
->lkb_exflags
& DLM_LKF_VALBLK
) {
4213 lkb
->lkb_lvbptr
= dlm_allocate_lvb(ls
);
4214 if (!lkb
->lkb_lvbptr
)
4216 lvblen
= rc
->rc_header
.h_length
- sizeof(struct dlm_rcom
) -
4217 sizeof(struct rcom_lock
);
4218 memcpy(lkb
->lkb_lvbptr
, rl
->rl_lvb
, lvblen
);
4221 /* Conversions between PR and CW (middle modes) need special handling.
4222 The real granted mode of these converting locks cannot be determined
4223 until all locks have been rebuilt on the rsb (recover_conversion) */
4225 if (rl
->rl_wait_type
== DLM_MSG_CONVERT
&& middle_conversion(lkb
)) {
4226 rl
->rl_status
= DLM_LKSTS_CONVERT
;
4227 lkb
->lkb_grmode
= DLM_LOCK_IV
;
4228 rsb_set_flag(r
, RSB_RECOVER_CONVERT
);
4234 /* This lkb may have been recovered in a previous aborted recovery so we need
4235 to check if the rsb already has an lkb with the given remote nodeid/lkid.
4236 If so we just send back a standard reply. If not, we create a new lkb with
4237 the given values and send back our lkid. We send back our lkid by sending
4238 back the rcom_lock struct we got but with the remid field filled in. */
4240 int dlm_recover_master_copy(struct dlm_ls
*ls
, struct dlm_rcom
*rc
)
4242 struct rcom_lock
*rl
= (struct rcom_lock
*) rc
->rc_buf
;
4244 struct dlm_lkb
*lkb
;
4247 if (rl
->rl_parent_lkid
) {
4248 error
= -EOPNOTSUPP
;
4252 error
= find_rsb(ls
, rl
->rl_name
, rl
->rl_namelen
, R_MASTER
, &r
);
4258 lkb
= search_remid(r
, rc
->rc_header
.h_nodeid
, rl
->rl_lkid
);
4264 error
= create_lkb(ls
, &lkb
);
4268 error
= receive_rcom_lock_args(ls
, lkb
, r
, rc
);
4275 add_lkb(r
, lkb
, rl
->rl_status
);
4279 /* this is the new value returned to the lock holder for
4280 saving in its process-copy lkb */
4281 rl
->rl_remid
= lkb
->lkb_id
;
4288 log_debug(ls
, "recover_master_copy %d %x", error
, rl
->rl_lkid
);
4289 rl
->rl_result
= error
;
4293 int dlm_recover_process_copy(struct dlm_ls
*ls
, struct dlm_rcom
*rc
)
4295 struct rcom_lock
*rl
= (struct rcom_lock
*) rc
->rc_buf
;
4297 struct dlm_lkb
*lkb
;
4300 error
= find_lkb(ls
, rl
->rl_lkid
, &lkb
);
4302 log_error(ls
, "recover_process_copy no lkid %x", rl
->rl_lkid
);
4306 DLM_ASSERT(is_process_copy(lkb
), dlm_print_lkb(lkb
););
4308 error
= rl
->rl_result
;
4310 r
= lkb
->lkb_resource
;
4316 /* There's a chance the new master received our lock before
4317 dlm_recover_master_reply(), this wouldn't happen if we did
4318 a barrier between recover_masters and recover_locks. */
4319 log_debug(ls
, "master copy not ready %x r %lx %s", lkb
->lkb_id
,
4320 (unsigned long)r
, r
->res_name
);
4321 dlm_send_rcom_lock(r
, lkb
);
4324 log_debug(ls
, "master copy exists %x", lkb
->lkb_id
);
4327 lkb
->lkb_remid
= rl
->rl_remid
;
4330 log_error(ls
, "dlm_recover_process_copy unknown error %d %x",
4331 error
, lkb
->lkb_id
);
4334 /* an ack for dlm_recover_locks() which waits for replies from
4335 all the locks it sends to new masters */
4336 dlm_recovered_lock(r
);
4345 int dlm_user_request(struct dlm_ls
*ls
, struct dlm_user_args
*ua
,
4346 int mode
, uint32_t flags
, void *name
, unsigned int namelen
,
4347 unsigned long timeout_cs
)
4349 struct dlm_lkb
*lkb
;
4350 struct dlm_args args
;
4353 dlm_lock_recovery(ls
);
4355 error
= create_lkb(ls
, &lkb
);
4361 if (flags
& DLM_LKF_VALBLK
) {
4362 ua
->lksb
.sb_lvbptr
= kzalloc(DLM_USER_LVB_LEN
, GFP_KERNEL
);
4363 if (!ua
->lksb
.sb_lvbptr
) {
4371 /* After ua is attached to lkb it will be freed by dlm_free_lkb().
4372 When DLM_IFL_USER is set, the dlm knows that this is a userspace
4373 lock and that lkb_astparam is the dlm_user_args structure. */
4375 error
= set_lock_args(mode
, &ua
->lksb
, flags
, namelen
, timeout_cs
,
4376 DLM_FAKE_USER_AST
, ua
, DLM_FAKE_USER_AST
, &args
);
4377 lkb
->lkb_flags
|= DLM_IFL_USER
;
4378 ua
->old_mode
= DLM_LOCK_IV
;
4385 error
= request_lock(ls
, lkb
, name
, namelen
, &args
);
4401 /* add this new lkb to the per-process list of locks */
4402 spin_lock(&ua
->proc
->locks_spin
);
4404 list_add_tail(&lkb
->lkb_ownqueue
, &ua
->proc
->locks
);
4405 spin_unlock(&ua
->proc
->locks_spin
);
4407 dlm_unlock_recovery(ls
);
4411 int dlm_user_convert(struct dlm_ls
*ls
, struct dlm_user_args
*ua_tmp
,
4412 int mode
, uint32_t flags
, uint32_t lkid
, char *lvb_in
,
4413 unsigned long timeout_cs
)
4415 struct dlm_lkb
*lkb
;
4416 struct dlm_args args
;
4417 struct dlm_user_args
*ua
;
4420 dlm_lock_recovery(ls
);
4422 error
= find_lkb(ls
, lkid
, &lkb
);
4426 /* user can change the params on its lock when it converts it, or
4427 add an lvb that didn't exist before */
4429 ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
4431 if (flags
& DLM_LKF_VALBLK
&& !ua
->lksb
.sb_lvbptr
) {
4432 ua
->lksb
.sb_lvbptr
= kzalloc(DLM_USER_LVB_LEN
, GFP_KERNEL
);
4433 if (!ua
->lksb
.sb_lvbptr
) {
4438 if (lvb_in
&& ua
->lksb
.sb_lvbptr
)
4439 memcpy(ua
->lksb
.sb_lvbptr
, lvb_in
, DLM_USER_LVB_LEN
);
4441 ua
->xid
= ua_tmp
->xid
;
4442 ua
->castparam
= ua_tmp
->castparam
;
4443 ua
->castaddr
= ua_tmp
->castaddr
;
4444 ua
->bastparam
= ua_tmp
->bastparam
;
4445 ua
->bastaddr
= ua_tmp
->bastaddr
;
4446 ua
->user_lksb
= ua_tmp
->user_lksb
;
4447 ua
->old_mode
= lkb
->lkb_grmode
;
4449 error
= set_lock_args(mode
, &ua
->lksb
, flags
, 0, timeout_cs
,
4450 DLM_FAKE_USER_AST
, ua
, DLM_FAKE_USER_AST
, &args
);
4454 error
= convert_lock(ls
, lkb
, &args
);
4456 if (error
== -EINPROGRESS
|| error
== -EAGAIN
|| error
== -EDEADLK
)
4461 dlm_unlock_recovery(ls
);
4466 int dlm_user_unlock(struct dlm_ls
*ls
, struct dlm_user_args
*ua_tmp
,
4467 uint32_t flags
, uint32_t lkid
, char *lvb_in
)
4469 struct dlm_lkb
*lkb
;
4470 struct dlm_args args
;
4471 struct dlm_user_args
*ua
;
4474 dlm_lock_recovery(ls
);
4476 error
= find_lkb(ls
, lkid
, &lkb
);
4480 ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
4482 if (lvb_in
&& ua
->lksb
.sb_lvbptr
)
4483 memcpy(ua
->lksb
.sb_lvbptr
, lvb_in
, DLM_USER_LVB_LEN
);
4484 if (ua_tmp
->castparam
)
4485 ua
->castparam
= ua_tmp
->castparam
;
4486 ua
->user_lksb
= ua_tmp
->user_lksb
;
4488 error
= set_unlock_args(flags
, ua
, &args
);
4492 error
= unlock_lock(ls
, lkb
, &args
);
4494 if (error
== -DLM_EUNLOCK
)
4496 /* from validate_unlock_args() */
4497 if (error
== -EBUSY
&& (flags
& DLM_LKF_FORCEUNLOCK
))
4502 spin_lock(&ua
->proc
->locks_spin
);
4503 /* dlm_user_add_ast() may have already taken lkb off the proc list */
4504 if (!list_empty(&lkb
->lkb_ownqueue
))
4505 list_move(&lkb
->lkb_ownqueue
, &ua
->proc
->unlocking
);
4506 spin_unlock(&ua
->proc
->locks_spin
);
4510 dlm_unlock_recovery(ls
);
4515 int dlm_user_cancel(struct dlm_ls
*ls
, struct dlm_user_args
*ua_tmp
,
4516 uint32_t flags
, uint32_t lkid
)
4518 struct dlm_lkb
*lkb
;
4519 struct dlm_args args
;
4520 struct dlm_user_args
*ua
;
4523 dlm_lock_recovery(ls
);
4525 error
= find_lkb(ls
, lkid
, &lkb
);
4529 ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
4530 if (ua_tmp
->castparam
)
4531 ua
->castparam
= ua_tmp
->castparam
;
4532 ua
->user_lksb
= ua_tmp
->user_lksb
;
4534 error
= set_unlock_args(flags
, ua
, &args
);
4538 error
= cancel_lock(ls
, lkb
, &args
);
4540 if (error
== -DLM_ECANCEL
)
4542 /* from validate_unlock_args() */
4543 if (error
== -EBUSY
)
4548 dlm_unlock_recovery(ls
);
4553 int dlm_user_deadlock(struct dlm_ls
*ls
, uint32_t flags
, uint32_t lkid
)
4555 struct dlm_lkb
*lkb
;
4556 struct dlm_args args
;
4557 struct dlm_user_args
*ua
;
4561 dlm_lock_recovery(ls
);
4563 error
= find_lkb(ls
, lkid
, &lkb
);
4567 ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
4569 error
= set_unlock_args(flags
, ua
, &args
);
4573 /* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4575 r
= lkb
->lkb_resource
;
4579 error
= validate_unlock_args(lkb
, &args
);
4582 lkb
->lkb_flags
|= DLM_IFL_DEADLOCK_CANCEL
;
4584 error
= _cancel_lock(r
, lkb
);
4589 if (error
== -DLM_ECANCEL
)
4591 /* from validate_unlock_args() */
4592 if (error
== -EBUSY
)
4597 dlm_unlock_recovery(ls
);
4601 /* lkb's that are removed from the waiters list by revert are just left on the
4602 orphans list with the granted orphan locks, to be freed by purge */
4604 static int orphan_proc_lock(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
4606 struct dlm_user_args
*ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
4607 struct dlm_args args
;
4611 mutex_lock(&ls
->ls_orphans_mutex
);
4612 list_add_tail(&lkb
->lkb_ownqueue
, &ls
->ls_orphans
);
4613 mutex_unlock(&ls
->ls_orphans_mutex
);
4615 set_unlock_args(0, ua
, &args
);
4617 error
= cancel_lock(ls
, lkb
, &args
);
4618 if (error
== -DLM_ECANCEL
)
4623 /* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4624 Regardless of what rsb queue the lock is on, it's removed and freed. */
4626 static int unlock_proc_lock(struct dlm_ls
*ls
, struct dlm_lkb
*lkb
)
4628 struct dlm_user_args
*ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
4629 struct dlm_args args
;
4632 set_unlock_args(DLM_LKF_FORCEUNLOCK
, ua
, &args
);
4634 error
= unlock_lock(ls
, lkb
, &args
);
4635 if (error
== -DLM_EUNLOCK
)
4640 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4641 (which does lock_rsb) due to deadlock with receiving a message that does
4642 lock_rsb followed by dlm_user_add_ast() */
4644 static struct dlm_lkb
*del_proc_lock(struct dlm_ls
*ls
,
4645 struct dlm_user_proc
*proc
)
4647 struct dlm_lkb
*lkb
= NULL
;
4649 mutex_lock(&ls
->ls_clear_proc_locks
);
4650 if (list_empty(&proc
->locks
))
4653 lkb
= list_entry(proc
->locks
.next
, struct dlm_lkb
, lkb_ownqueue
);
4654 list_del_init(&lkb
->lkb_ownqueue
);
4656 if (lkb
->lkb_exflags
& DLM_LKF_PERSISTENT
)
4657 lkb
->lkb_flags
|= DLM_IFL_ORPHAN
;
4659 lkb
->lkb_flags
|= DLM_IFL_DEAD
;
4661 mutex_unlock(&ls
->ls_clear_proc_locks
);
4665 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4666 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4667 which we clear here. */
4669 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
4670 list, and no more device_writes should add lkb's to proc->locks list; so we
4671 shouldn't need to take asts_spin or locks_spin here. this assumes that
4672 device reads/writes/closes are serialized -- FIXME: we may need to serialize
4675 void dlm_clear_proc_locks(struct dlm_ls
*ls
, struct dlm_user_proc
*proc
)
4677 struct dlm_lkb
*lkb
, *safe
;
4679 dlm_lock_recovery(ls
);
4682 lkb
= del_proc_lock(ls
, proc
);
4686 if (lkb
->lkb_exflags
& DLM_LKF_PERSISTENT
)
4687 orphan_proc_lock(ls
, lkb
);
4689 unlock_proc_lock(ls
, lkb
);
4691 /* this removes the reference for the proc->locks list
4692 added by dlm_user_request, it may result in the lkb
4698 mutex_lock(&ls
->ls_clear_proc_locks
);
4700 /* in-progress unlocks */
4701 list_for_each_entry_safe(lkb
, safe
, &proc
->unlocking
, lkb_ownqueue
) {
4702 list_del_init(&lkb
->lkb_ownqueue
);
4703 lkb
->lkb_flags
|= DLM_IFL_DEAD
;
4707 list_for_each_entry_safe(lkb
, safe
, &proc
->asts
, lkb_astqueue
) {
4708 lkb
->lkb_ast_type
= 0;
4709 list_del(&lkb
->lkb_astqueue
);
4713 mutex_unlock(&ls
->ls_clear_proc_locks
);
4714 dlm_unlock_recovery(ls
);
4717 static void purge_proc_locks(struct dlm_ls
*ls
, struct dlm_user_proc
*proc
)
4719 struct dlm_lkb
*lkb
, *safe
;
4723 spin_lock(&proc
->locks_spin
);
4724 if (!list_empty(&proc
->locks
)) {
4725 lkb
= list_entry(proc
->locks
.next
, struct dlm_lkb
,
4727 list_del_init(&lkb
->lkb_ownqueue
);
4729 spin_unlock(&proc
->locks_spin
);
4734 lkb
->lkb_flags
|= DLM_IFL_DEAD
;
4735 unlock_proc_lock(ls
, lkb
);
4736 dlm_put_lkb(lkb
); /* ref from proc->locks list */
4739 spin_lock(&proc
->locks_spin
);
4740 list_for_each_entry_safe(lkb
, safe
, &proc
->unlocking
, lkb_ownqueue
) {
4741 list_del_init(&lkb
->lkb_ownqueue
);
4742 lkb
->lkb_flags
|= DLM_IFL_DEAD
;
4745 spin_unlock(&proc
->locks_spin
);
4747 spin_lock(&proc
->asts_spin
);
4748 list_for_each_entry_safe(lkb
, safe
, &proc
->asts
, lkb_astqueue
) {
4749 list_del(&lkb
->lkb_astqueue
);
4752 spin_unlock(&proc
->asts_spin
);
4755 /* pid of 0 means purge all orphans */
4757 static void do_purge(struct dlm_ls
*ls
, int nodeid
, int pid
)
4759 struct dlm_lkb
*lkb
, *safe
;
4761 mutex_lock(&ls
->ls_orphans_mutex
);
4762 list_for_each_entry_safe(lkb
, safe
, &ls
->ls_orphans
, lkb_ownqueue
) {
4763 if (pid
&& lkb
->lkb_ownpid
!= pid
)
4765 unlock_proc_lock(ls
, lkb
);
4766 list_del_init(&lkb
->lkb_ownqueue
);
4769 mutex_unlock(&ls
->ls_orphans_mutex
);
4772 static int send_purge(struct dlm_ls
*ls
, int nodeid
, int pid
)
4774 struct dlm_message
*ms
;
4775 struct dlm_mhandle
*mh
;
4778 error
= _create_message(ls
, sizeof(struct dlm_message
), nodeid
,
4779 DLM_MSG_PURGE
, &ms
, &mh
);
4782 ms
->m_nodeid
= nodeid
;
4785 return send_message(mh
, ms
);
4788 int dlm_user_purge(struct dlm_ls
*ls
, struct dlm_user_proc
*proc
,
4789 int nodeid
, int pid
)
4793 if (nodeid
!= dlm_our_nodeid()) {
4794 error
= send_purge(ls
, nodeid
, pid
);
4796 dlm_lock_recovery(ls
);
4797 if (pid
== current
->pid
)
4798 purge_proc_locks(ls
, proc
);
4800 do_purge(ls
, nodeid
, pid
);
4801 dlm_unlock_recovery(ls
);