1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) 2005-2008 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 i, 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);
1888 for (i = 0; i < 2; i++) {
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);
1902 if (error && error != -EEXIST)
1905 if (ret_nodeid == our_nodeid) {
1906 r->res_first_lkid = 0;
1908 lkb->lkb_nodeid = 0;
1910 r->res_first_lkid = lkb->lkb_id;
1911 r->res_nodeid = ret_nodeid;
1912 lkb->lkb_nodeid = ret_nodeid;
1917 static void process_lookup_list(struct dlm_rsb *r)
1919 struct dlm_lkb *lkb, *safe;
1921 list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
1922 list_del_init(&lkb->lkb_rsb_lookup);
1923 _request_lock(r, lkb);
1928 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
1930 static void confirm_master(struct dlm_rsb *r, int error)
1932 struct dlm_lkb *lkb;
1934 if (!r->res_first_lkid)
1940 r->res_first_lkid = 0;
1941 process_lookup_list(r);
1947 /* the remote request failed and won't be retried (it was
1948 a NOQUEUE, or has been canceled/unlocked); make a waiting
1949 lkb the first_lkid */
1951 r->res_first_lkid = 0;
1953 if (!list_empty(&r->res_lookup)) {
1954 lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
1956 list_del_init(&lkb->lkb_rsb_lookup);
1957 r->res_first_lkid = lkb->lkb_id;
1958 _request_lock(r, lkb);
1964 log_error(r->res_ls, "confirm_master unknown error %d", error);
1968 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
1969 int namelen, unsigned long timeout_cs, void *ast,
1970 void *astarg, void *bast, struct dlm_args *args)
1974 /* check for invalid arg usage */
1976 if (mode < 0 || mode > DLM_LOCK_EX)
1979 if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
1982 if (flags & DLM_LKF_CANCEL)
1985 if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
1988 if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
1991 if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
1994 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
1997 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
2000 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2003 if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2009 if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2012 if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2015 /* these args will be copied to the lkb in validate_lock_args,
2016 it cannot be done now because when converting locks, fields in
2017 an active lkb cannot be modified before locking the rsb */
2019 args->flags = flags;
2020 args->astaddr = ast;
2021 args->astparam = (long) astarg;
2022 args->bastaddr = bast;
2023 args->timeout = timeout_cs;
2031 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2033 if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2034 DLM_LKF_FORCEUNLOCK))
2037 if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2040 args->flags = flags;
2041 args->astparam = (long) astarg;
2045 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2046 struct dlm_args *args)
2050 if (args->flags & DLM_LKF_CONVERT) {
2051 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2054 if (args->flags & DLM_LKF_QUECVT &&
2055 !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2059 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2062 if (lkb->lkb_wait_type)
2065 if (is_overlap(lkb))
2069 lkb->lkb_exflags = args->flags;
2070 lkb->lkb_sbflags = 0;
2071 lkb->lkb_astaddr = args->astaddr;
2072 lkb->lkb_astparam = args->astparam;
2073 lkb->lkb_bastaddr = args->bastaddr;
2074 lkb->lkb_rqmode = args->mode;
2075 lkb->lkb_lksb = args->lksb;
2076 lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2077 lkb->lkb_ownpid = (int) current->pid;
2078 lkb->lkb_timeout_cs = args->timeout;
2084 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2087 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2088 because there may be a lookup in progress and it's valid to do
2089 cancel/unlockf on it */
2091 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2093 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
2096 if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2097 log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2102 /* an lkb may still exist even though the lock is EOL'ed due to a
2103 cancel, unlock or failed noqueue request; an app can't use these
2104 locks; return same error as if the lkid had not been found at all */
2106 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2107 log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2112 /* an lkb may be waiting for an rsb lookup to complete where the
2113 lookup was initiated by another lock */
2115 if (!list_empty(&lkb->lkb_rsb_lookup)) {
2116 if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
2117 log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2118 list_del_init(&lkb->lkb_rsb_lookup);
2119 queue_cast(lkb->lkb_resource, lkb,
2120 args->flags & DLM_LKF_CANCEL ?
2121 -DLM_ECANCEL : -DLM_EUNLOCK);
2122 unhold_lkb(lkb); /* undoes create_lkb() */
2124 /* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2129 /* cancel not allowed with another cancel/unlock in progress */
2131 if (args->flags & DLM_LKF_CANCEL) {
2132 if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2135 if (is_overlap(lkb))
2138 /* don't let scand try to do a cancel */
2141 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2142 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2147 switch (lkb->lkb_wait_type) {
2148 case DLM_MSG_LOOKUP:
2149 case DLM_MSG_REQUEST:
2150 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2153 case DLM_MSG_UNLOCK:
2154 case DLM_MSG_CANCEL:
2157 /* add_to_waiters() will set OVERLAP_CANCEL */
2161 /* do we need to allow a force-unlock if there's a normal unlock
2162 already in progress? in what conditions could the normal unlock
2163 fail such that we'd want to send a force-unlock to be sure? */
2165 if (args->flags & DLM_LKF_FORCEUNLOCK) {
2166 if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2169 if (is_overlap_unlock(lkb))
2172 /* don't let scand try to do a cancel */
2175 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2176 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2181 switch (lkb->lkb_wait_type) {
2182 case DLM_MSG_LOOKUP:
2183 case DLM_MSG_REQUEST:
2184 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2187 case DLM_MSG_UNLOCK:
2190 /* add_to_waiters() will set OVERLAP_UNLOCK */
2194 /* normal unlock not allowed if there's any op in progress */
2196 if (lkb->lkb_wait_type || lkb->lkb_wait_count)
2200 /* an overlapping op shouldn't blow away exflags from other op */
2201 lkb->lkb_exflags |= args->flags;
2202 lkb->lkb_sbflags = 0;
2203 lkb->lkb_astparam = args->astparam;
2207 log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
2208 lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
2209 args->flags, lkb->lkb_wait_type,
2210 lkb->lkb_resource->res_name);
2215 * Four stage 4 varieties:
2216 * do_request(), do_convert(), do_unlock(), do_cancel()
2217 * These are called on the master node for the given lock and
2218 * from the central locking logic.
2221 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2225 if (can_be_granted(r, lkb, 1, NULL)) {
2227 queue_cast(r, lkb, 0);
2231 if (can_be_queued(lkb)) {
2232 error = -EINPROGRESS;
2233 add_lkb(r, lkb, DLM_LKSTS_WAITING);
2234 send_blocking_asts(r, lkb);
2240 if (force_blocking_asts(lkb))
2241 send_blocking_asts_all(r, lkb);
2242 queue_cast(r, lkb, -EAGAIN);
2248 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2253 /* changing an existing lock may allow others to be granted */
2255 if (can_be_granted(r, lkb, 1, &deadlk)) {
2257 queue_cast(r, lkb, 0);
2258 grant_pending_locks(r);
2262 /* can_be_granted() detected that this lock would block in a conversion
2263 deadlock, so we leave it on the granted queue and return EDEADLK in
2264 the ast for the convert. */
2267 /* it's left on the granted queue */
2268 log_debug(r->res_ls, "deadlock %x node %d sts%d g%d r%d %s",
2269 lkb->lkb_id, lkb->lkb_nodeid, lkb->lkb_status,
2270 lkb->lkb_grmode, lkb->lkb_rqmode, r->res_name);
2271 revert_lock(r, lkb);
2272 queue_cast(r, lkb, -EDEADLK);
2277 /* is_demoted() means the can_be_granted() above set the grmode
2278 to NL, and left us on the granted queue. This auto-demotion
2279 (due to CONVDEADLK) might mean other locks, and/or this lock, are
2280 now grantable. We have to try to grant other converting locks
2281 before we try again to grant this one. */
2283 if (is_demoted(lkb)) {
2284 grant_pending_convert(r, DLM_LOCK_IV, NULL);
2285 if (_can_be_granted(r, lkb, 1)) {
2287 queue_cast(r, lkb, 0);
2288 grant_pending_locks(r);
2291 /* else fall through and move to convert queue */
2294 if (can_be_queued(lkb)) {
2295 error = -EINPROGRESS;
2297 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2298 send_blocking_asts(r, lkb);
2304 if (force_blocking_asts(lkb))
2305 send_blocking_asts_all(r, lkb);
2306 queue_cast(r, lkb, -EAGAIN);
2312 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2314 remove_lock(r, lkb);
2315 queue_cast(r, lkb, -DLM_EUNLOCK);
2316 grant_pending_locks(r);
2317 return -DLM_EUNLOCK;
2320 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
2322 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2326 error = revert_lock(r, lkb);
2328 queue_cast(r, lkb, -DLM_ECANCEL);
2329 grant_pending_locks(r);
2330 return -DLM_ECANCEL;
2336 * Four stage 3 varieties:
2337 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2340 /* add a new lkb to a possibly new rsb, called by requesting process */
2342 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2346 /* set_master: sets lkb nodeid from r */
2348 error = set_master(r, lkb);
2357 /* receive_request() calls do_request() on remote node */
2358 error = send_request(r, lkb);
2360 error = do_request(r, lkb);
2365 /* change some property of an existing lkb, e.g. mode */
2367 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2372 /* receive_convert() calls do_convert() on remote node */
2373 error = send_convert(r, lkb);
2375 error = do_convert(r, lkb);
2380 /* remove an existing lkb from the granted queue */
2382 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2387 /* receive_unlock() calls do_unlock() on remote node */
2388 error = send_unlock(r, lkb);
2390 error = do_unlock(r, lkb);
2395 /* remove an existing lkb from the convert or wait queue */
2397 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2402 /* receive_cancel() calls do_cancel() on remote node */
2403 error = send_cancel(r, lkb);
2405 error = do_cancel(r, lkb);
2411 * Four stage 2 varieties:
2412 * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2415 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
2416 int len, struct dlm_args *args)
2421 error = validate_lock_args(ls, lkb, args);
2425 error = find_rsb(ls, name, len, R_CREATE, &r);
2432 lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
2434 error = _request_lock(r, lkb);
2443 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2444 struct dlm_args *args)
2449 r = lkb->lkb_resource;
2454 error = validate_lock_args(ls, lkb, args);
2458 error = _convert_lock(r, lkb);
2465 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2466 struct dlm_args *args)
2471 r = lkb->lkb_resource;
2476 error = validate_unlock_args(lkb, args);
2480 error = _unlock_lock(r, lkb);
2487 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2488 struct dlm_args *args)
2493 r = lkb->lkb_resource;
2498 error = validate_unlock_args(lkb, args);
2502 error = _cancel_lock(r, lkb);
2510 * Two stage 1 varieties: dlm_lock() and dlm_unlock()
2513 int dlm_lock(dlm_lockspace_t *lockspace,
2515 struct dlm_lksb *lksb,
2518 unsigned int namelen,
2519 uint32_t parent_lkid,
2520 void (*ast) (void *astarg),
2522 void (*bast) (void *astarg, int mode))
2525 struct dlm_lkb *lkb;
2526 struct dlm_args args;
2527 int error, convert = flags & DLM_LKF_CONVERT;
2529 ls = dlm_find_lockspace_local(lockspace);
2533 dlm_lock_recovery(ls);
2536 error = find_lkb(ls, lksb->sb_lkid, &lkb);
2538 error = create_lkb(ls, &lkb);
2543 error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
2544 astarg, bast, &args);
2549 error = convert_lock(ls, lkb, &args);
2551 error = request_lock(ls, lkb, name, namelen, &args);
2553 if (error == -EINPROGRESS)
2556 if (convert || error)
2558 if (error == -EAGAIN || error == -EDEADLK)
2561 dlm_unlock_recovery(ls);
2562 dlm_put_lockspace(ls);
2566 int dlm_unlock(dlm_lockspace_t *lockspace,
2569 struct dlm_lksb *lksb,
2573 struct dlm_lkb *lkb;
2574 struct dlm_args args;
2577 ls = dlm_find_lockspace_local(lockspace);
2581 dlm_lock_recovery(ls);
2583 error = find_lkb(ls, lkid, &lkb);
2587 error = set_unlock_args(flags, astarg, &args);
2591 if (flags & DLM_LKF_CANCEL)
2592 error = cancel_lock(ls, lkb, &args);
2594 error = unlock_lock(ls, lkb, &args);
2596 if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
2598 if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
2603 dlm_unlock_recovery(ls);
2604 dlm_put_lockspace(ls);
2609 * send/receive routines for remote operations and replies
2613 * send_request receive_request
2614 * send_convert receive_convert
2615 * send_unlock receive_unlock
2616 * send_cancel receive_cancel
2617 * send_grant receive_grant
2618 * send_bast receive_bast
2619 * send_lookup receive_lookup
2620 * send_remove receive_remove
2623 * receive_request_reply send_request_reply
2624 * receive_convert_reply send_convert_reply
2625 * receive_unlock_reply send_unlock_reply
2626 * receive_cancel_reply send_cancel_reply
2627 * receive_lookup_reply send_lookup_reply
2630 static int _create_message(struct dlm_ls *ls, int mb_len,
2631 int to_nodeid, int mstype,
2632 struct dlm_message **ms_ret,
2633 struct dlm_mhandle **mh_ret)
2635 struct dlm_message *ms;
2636 struct dlm_mhandle *mh;
2639 /* get_buffer gives us a message handle (mh) that we need to
2640 pass into lowcomms_commit and a message buffer (mb) that we
2641 write our data into */
2643 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
2647 memset(mb, 0, mb_len);
2649 ms = (struct dlm_message *) mb;
2651 ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
2652 ms->m_header.h_lockspace = ls->ls_global_id;
2653 ms->m_header.h_nodeid = dlm_our_nodeid();
2654 ms->m_header.h_length = mb_len;
2655 ms->m_header.h_cmd = DLM_MSG;
2657 ms->m_type = mstype;
2664 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
2665 int to_nodeid, int mstype,
2666 struct dlm_message **ms_ret,
2667 struct dlm_mhandle **mh_ret)
2669 int mb_len = sizeof(struct dlm_message);
2672 case DLM_MSG_REQUEST:
2673 case DLM_MSG_LOOKUP:
2674 case DLM_MSG_REMOVE:
2675 mb_len += r->res_length;
2677 case DLM_MSG_CONVERT:
2678 case DLM_MSG_UNLOCK:
2679 case DLM_MSG_REQUEST_REPLY:
2680 case DLM_MSG_CONVERT_REPLY:
2682 if (lkb && lkb->lkb_lvbptr)
2683 mb_len += r->res_ls->ls_lvblen;
2687 return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
2691 /* further lowcomms enhancements or alternate implementations may make
2692 the return value from this function useful at some point */
2694 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
2696 dlm_message_out(ms);
2697 dlm_lowcomms_commit_buffer(mh);
2701 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
2702 struct dlm_message *ms)
2704 ms->m_nodeid = lkb->lkb_nodeid;
2705 ms->m_pid = lkb->lkb_ownpid;
2706 ms->m_lkid = lkb->lkb_id;
2707 ms->m_remid = lkb->lkb_remid;
2708 ms->m_exflags = lkb->lkb_exflags;
2709 ms->m_sbflags = lkb->lkb_sbflags;
2710 ms->m_flags = lkb->lkb_flags;
2711 ms->m_lvbseq = lkb->lkb_lvbseq;
2712 ms->m_status = lkb->lkb_status;
2713 ms->m_grmode = lkb->lkb_grmode;
2714 ms->m_rqmode = lkb->lkb_rqmode;
2715 ms->m_hash = r->res_hash;
2717 /* m_result and m_bastmode are set from function args,
2718 not from lkb fields */
2720 if (lkb->lkb_bastaddr)
2721 ms->m_asts |= AST_BAST;
2722 if (lkb->lkb_astaddr)
2723 ms->m_asts |= AST_COMP;
2725 /* compare with switch in create_message; send_remove() doesn't
2728 switch (ms->m_type) {
2729 case DLM_MSG_REQUEST:
2730 case DLM_MSG_LOOKUP:
2731 memcpy(ms->m_extra, r->res_name, r->res_length);
2733 case DLM_MSG_CONVERT:
2734 case DLM_MSG_UNLOCK:
2735 case DLM_MSG_REQUEST_REPLY:
2736 case DLM_MSG_CONVERT_REPLY:
2738 if (!lkb->lkb_lvbptr)
2740 memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
2745 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
2747 struct dlm_message *ms;
2748 struct dlm_mhandle *mh;
2749 int to_nodeid, error;
2751 error = add_to_waiters(lkb, mstype);
2755 to_nodeid = r->res_nodeid;
2757 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2761 send_args(r, lkb, ms);
2763 error = send_message(mh, ms);
2769 remove_from_waiters(lkb, msg_reply_type(mstype));
2773 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2775 return send_common(r, lkb, DLM_MSG_REQUEST);
2778 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2782 error = send_common(r, lkb, DLM_MSG_CONVERT);
2784 /* down conversions go without a reply from the master */
2785 if (!error && down_conversion(lkb)) {
2786 remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
2787 r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
2788 r->res_ls->ls_stub_ms.m_result = 0;
2789 r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
2790 __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
2796 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
2797 MASTER_UNCERTAIN to force the next request on the rsb to confirm
2798 that the master is still correct. */
2800 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2802 return send_common(r, lkb, DLM_MSG_UNLOCK);
2805 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2807 return send_common(r, lkb, DLM_MSG_CANCEL);
2810 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
2812 struct dlm_message *ms;
2813 struct dlm_mhandle *mh;
2814 int to_nodeid, error;
2816 to_nodeid = lkb->lkb_nodeid;
2818 error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
2822 send_args(r, lkb, ms);
2826 error = send_message(mh, ms);
2831 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
2833 struct dlm_message *ms;
2834 struct dlm_mhandle *mh;
2835 int to_nodeid, error;
2837 to_nodeid = lkb->lkb_nodeid;
2839 error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
2843 send_args(r, lkb, ms);
2845 ms->m_bastmode = mode;
2847 error = send_message(mh, ms);
2852 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
2854 struct dlm_message *ms;
2855 struct dlm_mhandle *mh;
2856 int to_nodeid, error;
2858 error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
2862 to_nodeid = dlm_dir_nodeid(r);
2864 error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
2868 send_args(r, lkb, ms);
2870 error = send_message(mh, ms);
2876 remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
2880 static int send_remove(struct dlm_rsb *r)
2882 struct dlm_message *ms;
2883 struct dlm_mhandle *mh;
2884 int to_nodeid, error;
2886 to_nodeid = dlm_dir_nodeid(r);
2888 error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
2892 memcpy(ms->m_extra, r->res_name, r->res_length);
2893 ms->m_hash = r->res_hash;
2895 error = send_message(mh, ms);
2900 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
2903 struct dlm_message *ms;
2904 struct dlm_mhandle *mh;
2905 int to_nodeid, error;
2907 to_nodeid = lkb->lkb_nodeid;
2909 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2913 send_args(r, lkb, ms);
2917 error = send_message(mh, ms);
2922 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2924 return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
2927 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2929 return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
2932 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2934 return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
2937 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2939 return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
2942 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
2943 int ret_nodeid, int rv)
2945 struct dlm_rsb *r = &ls->ls_stub_rsb;
2946 struct dlm_message *ms;
2947 struct dlm_mhandle *mh;
2948 int error, nodeid = ms_in->m_header.h_nodeid;
2950 error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
2954 ms->m_lkid = ms_in->m_lkid;
2956 ms->m_nodeid = ret_nodeid;
2958 error = send_message(mh, ms);
2963 /* which args we save from a received message depends heavily on the type
2964 of message, unlike the send side where we can safely send everything about
2965 the lkb for any type of message */
2967 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
2969 lkb->lkb_exflags = ms->m_exflags;
2970 lkb->lkb_sbflags = ms->m_sbflags;
2971 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2972 (ms->m_flags & 0x0000FFFF);
2975 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
2977 lkb->lkb_sbflags = ms->m_sbflags;
2978 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2979 (ms->m_flags & 0x0000FFFF);
2982 static int receive_extralen(struct dlm_message *ms)
2984 return (ms->m_header.h_length - sizeof(struct dlm_message));
2987 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
2988 struct dlm_message *ms)
2992 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
2993 if (!lkb->lkb_lvbptr)
2994 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
2995 if (!lkb->lkb_lvbptr)
2997 len = receive_extralen(ms);
2998 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
3003 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3004 struct dlm_message *ms)
3006 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3007 lkb->lkb_ownpid = ms->m_pid;
3008 lkb->lkb_remid = ms->m_lkid;
3009 lkb->lkb_grmode = DLM_LOCK_IV;
3010 lkb->lkb_rqmode = ms->m_rqmode;
3011 lkb->lkb_bastaddr = (void *) (long) (ms->m_asts & AST_BAST);
3012 lkb->lkb_astaddr = (void *) (long) (ms->m_asts & AST_COMP);
3014 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3015 /* lkb was just created so there won't be an lvb yet */
3016 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3017 if (!lkb->lkb_lvbptr)
3024 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3025 struct dlm_message *ms)
3027 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3030 if (receive_lvb(ls, lkb, ms))
3033 lkb->lkb_rqmode = ms->m_rqmode;
3034 lkb->lkb_lvbseq = ms->m_lvbseq;
3039 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3040 struct dlm_message *ms)
3042 if (receive_lvb(ls, lkb, ms))
3047 /* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3048 uses to send a reply and that the remote end uses to process the reply. */
3050 static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3052 struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3053 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3054 lkb->lkb_remid = ms->m_lkid;
3057 /* This is called after the rsb is locked so that we can safely inspect
3058 fields in the lkb. */
3060 static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3062 int from = ms->m_header.h_nodeid;
3065 switch (ms->m_type) {
3066 case DLM_MSG_CONVERT:
3067 case DLM_MSG_UNLOCK:
3068 case DLM_MSG_CANCEL:
3069 if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3073 case DLM_MSG_CONVERT_REPLY:
3074 case DLM_MSG_UNLOCK_REPLY:
3075 case DLM_MSG_CANCEL_REPLY:
3078 if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3082 case DLM_MSG_REQUEST_REPLY:
3083 if (!is_process_copy(lkb))
3085 else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
3094 log_error(lkb->lkb_resource->res_ls,
3095 "ignore invalid message %d from %d %x %x %x %d",
3096 ms->m_type, from, lkb->lkb_id, lkb->lkb_remid,
3097 lkb->lkb_flags, lkb->lkb_nodeid);
3101 static void receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3103 struct dlm_lkb *lkb;
3107 error = create_lkb(ls, &lkb);
3111 receive_flags(lkb, ms);
3112 lkb->lkb_flags |= DLM_IFL_MSTCPY;
3113 error = receive_request_args(ls, lkb, ms);
3119 namelen = receive_extralen(ms);
3121 error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r);
3130 error = do_request(r, lkb);
3131 send_request_reply(r, lkb, error);
3136 if (error == -EINPROGRESS)
3143 setup_stub_lkb(ls, ms);
3144 send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3147 static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3149 struct dlm_lkb *lkb;
3151 int error, reply = 1;
3153 error = find_lkb(ls, ms->m_remid, &lkb);
3157 r = lkb->lkb_resource;
3162 error = validate_message(lkb, ms);
3166 receive_flags(lkb, ms);
3167 error = receive_convert_args(ls, lkb, ms);
3170 reply = !down_conversion(lkb);
3172 error = do_convert(r, lkb);
3175 send_convert_reply(r, lkb, error);
3183 setup_stub_lkb(ls, ms);
3184 send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3187 static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3189 struct dlm_lkb *lkb;
3193 error = find_lkb(ls, ms->m_remid, &lkb);
3197 r = lkb->lkb_resource;
3202 error = validate_message(lkb, ms);
3206 receive_flags(lkb, ms);
3207 error = receive_unlock_args(ls, lkb, ms);
3211 error = do_unlock(r, lkb);
3213 send_unlock_reply(r, lkb, error);
3221 setup_stub_lkb(ls, ms);
3222 send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3225 static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
3227 struct dlm_lkb *lkb;
3231 error = find_lkb(ls, ms->m_remid, &lkb);
3235 receive_flags(lkb, ms);
3237 r = lkb->lkb_resource;
3242 error = validate_message(lkb, ms);
3246 error = do_cancel(r, lkb);
3247 send_cancel_reply(r, lkb, error);
3255 setup_stub_lkb(ls, ms);
3256 send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3259 static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
3261 struct dlm_lkb *lkb;
3265 error = find_lkb(ls, ms->m_remid, &lkb);
3267 log_debug(ls, "receive_grant from %d no lkb %x",
3268 ms->m_header.h_nodeid, ms->m_remid);
3272 r = lkb->lkb_resource;
3277 error = validate_message(lkb, ms);
3281 receive_flags_reply(lkb, ms);
3282 if (is_altmode(lkb))
3283 munge_altmode(lkb, ms);
3284 grant_lock_pc(r, lkb, ms);
3285 queue_cast(r, lkb, 0);
3292 static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
3294 struct dlm_lkb *lkb;
3298 error = find_lkb(ls, ms->m_remid, &lkb);
3300 log_debug(ls, "receive_bast from %d no lkb %x",
3301 ms->m_header.h_nodeid, ms->m_remid);
3305 r = lkb->lkb_resource;
3310 error = validate_message(lkb, ms);
3314 queue_bast(r, lkb, ms->m_bastmode);
3321 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
3323 int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid;
3325 from_nodeid = ms->m_header.h_nodeid;
3326 our_nodeid = dlm_our_nodeid();
3328 len = receive_extralen(ms);
3330 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3331 if (dir_nodeid != our_nodeid) {
3332 log_error(ls, "lookup dir_nodeid %d from %d",
3333 dir_nodeid, from_nodeid);
3339 error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid);
3341 /* Optimization: we're master so treat lookup as a request */
3342 if (!error && ret_nodeid == our_nodeid) {
3343 receive_request(ls, ms);
3347 send_lookup_reply(ls, ms, ret_nodeid, error);
3350 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
3352 int len, dir_nodeid, from_nodeid;
3354 from_nodeid = ms->m_header.h_nodeid;
3356 len = receive_extralen(ms);
3358 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3359 if (dir_nodeid != dlm_our_nodeid()) {
3360 log_error(ls, "remove dir entry dir_nodeid %d from %d",
3361 dir_nodeid, from_nodeid);
3365 dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
3368 static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
3370 do_purge(ls, ms->m_nodeid, ms->m_pid);
3373 static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
3375 struct dlm_lkb *lkb;
3377 int error, mstype, result;
3379 error = find_lkb(ls, ms->m_remid, &lkb);
3381 log_debug(ls, "receive_request_reply from %d no lkb %x",
3382 ms->m_header.h_nodeid, ms->m_remid);
3386 r = lkb->lkb_resource;
3390 error = validate_message(lkb, ms);
3394 mstype = lkb->lkb_wait_type;
3395 error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
3399 /* Optimization: the dir node was also the master, so it took our
3400 lookup as a request and sent request reply instead of lookup reply */
3401 if (mstype == DLM_MSG_LOOKUP) {
3402 r->res_nodeid = ms->m_header.h_nodeid;
3403 lkb->lkb_nodeid = r->res_nodeid;
3406 /* this is the value returned from do_request() on the master */
3407 result = ms->m_result;
3411 /* request would block (be queued) on remote master */
3412 queue_cast(r, lkb, -EAGAIN);
3413 confirm_master(r, -EAGAIN);
3414 unhold_lkb(lkb); /* undoes create_lkb() */
3419 /* request was queued or granted on remote master */
3420 receive_flags_reply(lkb, ms);
3421 lkb->lkb_remid = ms->m_lkid;
3422 if (is_altmode(lkb))
3423 munge_altmode(lkb, ms);
3425 add_lkb(r, lkb, DLM_LKSTS_WAITING);
3428 grant_lock_pc(r, lkb, ms);
3429 queue_cast(r, lkb, 0);
3431 confirm_master(r, result);
3436 /* find_rsb failed to find rsb or rsb wasn't master */
3437 log_debug(ls, "receive_request_reply %x %x master diff %d %d",
3438 lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
3440 lkb->lkb_nodeid = -1;
3442 if (is_overlap(lkb)) {
3443 /* we'll ignore error in cancel/unlock reply */
3444 queue_cast_overlap(r, lkb);
3445 confirm_master(r, result);
3446 unhold_lkb(lkb); /* undoes create_lkb() */
3448 _request_lock(r, lkb);
3452 log_error(ls, "receive_request_reply %x error %d",
3453 lkb->lkb_id, result);
3456 if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
3457 log_debug(ls, "receive_request_reply %x result %d unlock",
3458 lkb->lkb_id, result);
3459 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3460 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3461 send_unlock(r, lkb);
3462 } else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
3463 log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
3464 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3465 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3466 send_cancel(r, lkb);
3468 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3469 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3477 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3478 struct dlm_message *ms)
3480 /* this is the value returned from do_convert() on the master */
3481 switch (ms->m_result) {
3483 /* convert would block (be queued) on remote master */
3484 queue_cast(r, lkb, -EAGAIN);
3488 receive_flags_reply(lkb, ms);
3489 revert_lock_pc(r, lkb);
3490 queue_cast(r, lkb, -EDEADLK);
3494 /* convert was queued on remote master */
3495 receive_flags_reply(lkb, ms);
3496 if (is_demoted(lkb))
3497 munge_demoted(lkb, ms);
3499 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3504 /* convert was granted on remote master */
3505 receive_flags_reply(lkb, ms);
3506 if (is_demoted(lkb))
3507 munge_demoted(lkb, ms);
3508 grant_lock_pc(r, lkb, ms);
3509 queue_cast(r, lkb, 0);
3513 log_error(r->res_ls, "receive_convert_reply %x error %d",
3514 lkb->lkb_id, ms->m_result);
3518 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3520 struct dlm_rsb *r = lkb->lkb_resource;
3526 error = validate_message(lkb, ms);
3530 /* stub reply can happen with waiters_mutex held */
3531 error = remove_from_waiters_ms(lkb, ms);
3535 __receive_convert_reply(r, lkb, ms);
3541 static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
3543 struct dlm_lkb *lkb;
3546 error = find_lkb(ls, ms->m_remid, &lkb);
3548 log_debug(ls, "receive_convert_reply from %d no lkb %x",
3549 ms->m_header.h_nodeid, ms->m_remid);
3553 _receive_convert_reply(lkb, ms);
3557 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3559 struct dlm_rsb *r = lkb->lkb_resource;
3565 error = validate_message(lkb, ms);
3569 /* stub reply can happen with waiters_mutex held */
3570 error = remove_from_waiters_ms(lkb, ms);
3574 /* this is the value returned from do_unlock() on the master */
3576 switch (ms->m_result) {
3578 receive_flags_reply(lkb, ms);
3579 remove_lock_pc(r, lkb);
3580 queue_cast(r, lkb, -DLM_EUNLOCK);
3585 log_error(r->res_ls, "receive_unlock_reply %x error %d",
3586 lkb->lkb_id, ms->m_result);
3593 static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
3595 struct dlm_lkb *lkb;
3598 error = find_lkb(ls, ms->m_remid, &lkb);
3600 log_debug(ls, "receive_unlock_reply from %d no lkb %x",
3601 ms->m_header.h_nodeid, ms->m_remid);
3605 _receive_unlock_reply(lkb, ms);
3609 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3611 struct dlm_rsb *r = lkb->lkb_resource;
3617 error = validate_message(lkb, ms);
3621 /* stub reply can happen with waiters_mutex held */
3622 error = remove_from_waiters_ms(lkb, ms);
3626 /* this is the value returned from do_cancel() on the master */
3628 switch (ms->m_result) {
3630 receive_flags_reply(lkb, ms);
3631 revert_lock_pc(r, lkb);
3632 queue_cast(r, lkb, -DLM_ECANCEL);
3637 log_error(r->res_ls, "receive_cancel_reply %x error %d",
3638 lkb->lkb_id, ms->m_result);
3645 static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
3647 struct dlm_lkb *lkb;
3650 error = find_lkb(ls, ms->m_remid, &lkb);
3652 log_debug(ls, "receive_cancel_reply from %d no lkb %x",
3653 ms->m_header.h_nodeid, ms->m_remid);
3657 _receive_cancel_reply(lkb, ms);
3661 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
3663 struct dlm_lkb *lkb;
3665 int error, ret_nodeid;
3667 error = find_lkb(ls, ms->m_lkid, &lkb);
3669 log_error(ls, "receive_lookup_reply no lkb");
3673 /* ms->m_result is the value returned by dlm_dir_lookup on dir node
3674 FIXME: will a non-zero error ever be returned? */
3676 r = lkb->lkb_resource;
3680 error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3684 ret_nodeid = ms->m_nodeid;
3685 if (ret_nodeid == dlm_our_nodeid()) {
3688 r->res_first_lkid = 0;
3690 /* set_master() will copy res_nodeid to lkb_nodeid */
3691 r->res_nodeid = ret_nodeid;
3694 if (is_overlap(lkb)) {
3695 log_debug(ls, "receive_lookup_reply %x unlock %x",
3696 lkb->lkb_id, lkb->lkb_flags);
3697 queue_cast_overlap(r, lkb);
3698 unhold_lkb(lkb); /* undoes create_lkb() */
3702 _request_lock(r, lkb);
3706 process_lookup_list(r);
3713 static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
3715 if (!dlm_is_member(ls, ms->m_header.h_nodeid)) {
3716 log_debug(ls, "ignore non-member message %d from %d %x %x %d",
3717 ms->m_type, ms->m_header.h_nodeid, ms->m_lkid,
3718 ms->m_remid, ms->m_result);
3722 switch (ms->m_type) {
3724 /* messages sent to a master node */
3726 case DLM_MSG_REQUEST:
3727 receive_request(ls, ms);
3730 case DLM_MSG_CONVERT:
3731 receive_convert(ls, ms);
3734 case DLM_MSG_UNLOCK:
3735 receive_unlock(ls, ms);
3738 case DLM_MSG_CANCEL:
3739 receive_cancel(ls, ms);
3742 /* messages sent from a master node (replies to above) */
3744 case DLM_MSG_REQUEST_REPLY:
3745 receive_request_reply(ls, ms);
3748 case DLM_MSG_CONVERT_REPLY:
3749 receive_convert_reply(ls, ms);
3752 case DLM_MSG_UNLOCK_REPLY:
3753 receive_unlock_reply(ls, ms);
3756 case DLM_MSG_CANCEL_REPLY:
3757 receive_cancel_reply(ls, ms);
3760 /* messages sent from a master node (only two types of async msg) */
3763 receive_grant(ls, ms);
3767 receive_bast(ls, ms);
3770 /* messages sent to a dir node */
3772 case DLM_MSG_LOOKUP:
3773 receive_lookup(ls, ms);
3776 case DLM_MSG_REMOVE:
3777 receive_remove(ls, ms);
3780 /* messages sent from a dir node (remove has no reply) */
3782 case DLM_MSG_LOOKUP_REPLY:
3783 receive_lookup_reply(ls, ms);
3786 /* other messages */
3789 receive_purge(ls, ms);
3793 log_error(ls, "unknown message type %d", ms->m_type);
3799 /* If the lockspace is in recovery mode (locking stopped), then normal
3800 messages are saved on the requestqueue for processing after recovery is
3801 done. When not in recovery mode, we wait for dlm_recoverd to drain saved
3802 messages off the requestqueue before we process new ones. This occurs right
3803 after recovery completes when we transition from saving all messages on
3804 requestqueue, to processing all the saved messages, to processing new
3805 messages as they arrive. */
3807 static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
3810 if (dlm_locking_stopped(ls)) {
3811 dlm_add_requestqueue(ls, nodeid, (struct dlm_header *) ms);
3813 dlm_wait_requestqueue(ls);
3814 _receive_message(ls, ms);
3818 /* This is called by dlm_recoverd to process messages that were saved on
3819 the requestqueue. */
3821 void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
3823 _receive_message(ls, ms);
3826 /* This is called by the midcomms layer when something is received for
3827 the lockspace. It could be either a MSG (normal message sent as part of
3828 standard locking activity) or an RCOM (recovery message sent as part of
3829 lockspace recovery). */
3831 void dlm_receive_buffer(struct dlm_header *hd, int nodeid)
3833 struct dlm_message *ms = (struct dlm_message *) hd;
3834 struct dlm_rcom *rc = (struct dlm_rcom *) hd;
3838 switch (hd->h_cmd) {
3848 log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
3852 if (hd->h_nodeid != nodeid) {
3853 log_print("invalid h_nodeid %d from %d lockspace %x",
3854 hd->h_nodeid, nodeid, hd->h_lockspace);
3858 ls = dlm_find_lockspace_global(hd->h_lockspace);
3860 log_print("invalid h_lockspace %x from %d cmd %d type %d",
3861 hd->h_lockspace, nodeid, hd->h_cmd, type);
3863 if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
3864 dlm_send_ls_not_ready(nodeid, rc);
3868 /* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3869 be inactive (in this ls) before transitioning to recovery mode */
3871 down_read(&ls->ls_recv_active);
3872 if (hd->h_cmd == DLM_MSG)
3873 dlm_receive_message(ls, ms, nodeid);
3875 dlm_receive_rcom(ls, rc, nodeid);
3876 up_read(&ls->ls_recv_active);
3878 dlm_put_lockspace(ls);
3881 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
3883 if (middle_conversion(lkb)) {
3885 ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
3886 ls->ls_stub_ms.m_result = -EINPROGRESS;
3887 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
3888 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
3889 _receive_convert_reply(lkb, &ls->ls_stub_ms);
3891 /* Same special case as in receive_rcom_lock_args() */
3892 lkb->lkb_grmode = DLM_LOCK_IV;
3893 rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
3896 } else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
3897 lkb->lkb_flags |= DLM_IFL_RESEND;
3900 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
3901 conversions are async; there's no reply from the remote master */
3904 /* A waiting lkb needs recovery if the master node has failed, or
3905 the master node is changing (only when no directory is used) */
3907 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
3909 if (dlm_is_removed(ls, lkb->lkb_nodeid))
3912 if (!dlm_no_directory(ls))
3915 if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid)
3921 /* Recovery for locks that are waiting for replies from nodes that are now
3922 gone. We can just complete unlocks and cancels by faking a reply from the
3923 dead node. Requests and up-conversions we flag to be resent after
3924 recovery. Down-conversions can just be completed with a fake reply like
3925 unlocks. Conversions between PR and CW need special attention. */
3927 void dlm_recover_waiters_pre(struct dlm_ls *ls)
3929 struct dlm_lkb *lkb, *safe;
3930 int wait_type, stub_unlock_result, stub_cancel_result;
3932 mutex_lock(&ls->ls_waiters_mutex);
3934 list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
3935 log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
3936 lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
3938 /* all outstanding lookups, regardless of destination will be
3939 resent after recovery is done */
3941 if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
3942 lkb->lkb_flags |= DLM_IFL_RESEND;
3946 if (!waiter_needs_recovery(ls, lkb))
3949 wait_type = lkb->lkb_wait_type;
3950 stub_unlock_result = -DLM_EUNLOCK;
3951 stub_cancel_result = -DLM_ECANCEL;
3953 /* Main reply may have been received leaving a zero wait_type,
3954 but a reply for the overlapping op may not have been
3955 received. In that case we need to fake the appropriate
3956 reply for the overlap op. */
3959 if (is_overlap_cancel(lkb)) {
3960 wait_type = DLM_MSG_CANCEL;
3961 if (lkb->lkb_grmode == DLM_LOCK_IV)
3962 stub_cancel_result = 0;
3964 if (is_overlap_unlock(lkb)) {
3965 wait_type = DLM_MSG_UNLOCK;
3966 if (lkb->lkb_grmode == DLM_LOCK_IV)
3967 stub_unlock_result = -ENOENT;
3970 log_debug(ls, "rwpre overlap %x %x %d %d %d",
3971 lkb->lkb_id, lkb->lkb_flags, wait_type,
3972 stub_cancel_result, stub_unlock_result);
3975 switch (wait_type) {
3977 case DLM_MSG_REQUEST:
3978 lkb->lkb_flags |= DLM_IFL_RESEND;
3981 case DLM_MSG_CONVERT:
3982 recover_convert_waiter(ls, lkb);
3985 case DLM_MSG_UNLOCK:
3987 ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
3988 ls->ls_stub_ms.m_result = stub_unlock_result;
3989 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
3990 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
3991 _receive_unlock_reply(lkb, &ls->ls_stub_ms);
3995 case DLM_MSG_CANCEL:
3997 ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
3998 ls->ls_stub_ms.m_result = stub_cancel_result;
3999 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
4000 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
4001 _receive_cancel_reply(lkb, &ls->ls_stub_ms);
4006 log_error(ls, "invalid lkb wait_type %d %d",
4007 lkb->lkb_wait_type, wait_type);
4011 mutex_unlock(&ls->ls_waiters_mutex);
4014 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
4016 struct dlm_lkb *lkb;
4019 mutex_lock(&ls->ls_waiters_mutex);
4020 list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
4021 if (lkb->lkb_flags & DLM_IFL_RESEND) {
4027 mutex_unlock(&ls->ls_waiters_mutex);
4034 /* Deal with lookups and lkb's marked RESEND from _pre. We may now be the
4035 master or dir-node for r. Processing the lkb may result in it being placed
4038 /* We do this after normal locking has been enabled and any saved messages
4039 (in requestqueue) have been processed. We should be confident that at
4040 this point we won't get or process a reply to any of these waiting
4041 operations. But, new ops may be coming in on the rsbs/locks here from
4042 userspace or remotely. */
4044 /* there may have been an overlap unlock/cancel prior to recovery or after
4045 recovery. if before, the lkb may still have a pos wait_count; if after, the
4046 overlap flag would just have been set and nothing new sent. we can be
4047 confident here than any replies to either the initial op or overlap ops
4048 prior to recovery have been received. */
4050 int dlm_recover_waiters_post(struct dlm_ls *ls)
4052 struct dlm_lkb *lkb;
4054 int error = 0, mstype, err, oc, ou;
4057 if (dlm_locking_stopped(ls)) {
4058 log_debug(ls, "recover_waiters_post aborted");
4063 lkb = find_resend_waiter(ls);
4067 r = lkb->lkb_resource;
4071 mstype = lkb->lkb_wait_type;
4072 oc = is_overlap_cancel(lkb);
4073 ou = is_overlap_unlock(lkb);
4076 log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
4077 lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
4079 /* At this point we assume that we won't get a reply to any
4080 previous op or overlap op on this lock. First, do a big
4081 remove_from_waiters() for all previous ops. */
4083 lkb->lkb_flags &= ~DLM_IFL_RESEND;
4084 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4085 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4086 lkb->lkb_wait_type = 0;
4087 lkb->lkb_wait_count = 0;
4088 mutex_lock(&ls->ls_waiters_mutex);
4089 list_del_init(&lkb->lkb_wait_reply);
4090 mutex_unlock(&ls->ls_waiters_mutex);
4091 unhold_lkb(lkb); /* for waiters list */
4094 /* do an unlock or cancel instead of resending */
4096 case DLM_MSG_LOOKUP:
4097 case DLM_MSG_REQUEST:
4098 queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
4100 unhold_lkb(lkb); /* undoes create_lkb() */
4102 case DLM_MSG_CONVERT:
4104 queue_cast(r, lkb, -DLM_ECANCEL);
4106 lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
4107 _unlock_lock(r, lkb);
4115 case DLM_MSG_LOOKUP:
4116 case DLM_MSG_REQUEST:
4117 _request_lock(r, lkb);
4119 confirm_master(r, 0);
4121 case DLM_MSG_CONVERT:
4122 _convert_lock(r, lkb);
4130 log_error(ls, "recover_waiters_post %x %d %x %d %d",
4131 lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
4140 static void purge_queue(struct dlm_rsb *r, struct list_head *queue,
4141 int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb))
4143 struct dlm_ls *ls = r->res_ls;
4144 struct dlm_lkb *lkb, *safe;
4146 list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) {
4147 if (test(ls, lkb)) {
4148 rsb_set_flag(r, RSB_LOCKS_PURGED);
4150 /* this put should free the lkb */
4151 if (!dlm_put_lkb(lkb))
4152 log_error(ls, "purged lkb not released");
4157 static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4159 return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid));
4162 static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4164 return is_master_copy(lkb);
4167 static void purge_dead_locks(struct dlm_rsb *r)
4169 purge_queue(r, &r->res_grantqueue, &purge_dead_test);
4170 purge_queue(r, &r->res_convertqueue, &purge_dead_test);
4171 purge_queue(r, &r->res_waitqueue, &purge_dead_test);
4174 void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
4176 purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test);
4177 purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test);
4178 purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test);
4181 /* Get rid of locks held by nodes that are gone. */
4183 int dlm_purge_locks(struct dlm_ls *ls)
4187 log_debug(ls, "dlm_purge_locks");
4189 down_write(&ls->ls_root_sem);
4190 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
4194 purge_dead_locks(r);
4200 up_write(&ls->ls_root_sem);
4205 static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket)
4207 struct dlm_rsb *r, *r_ret = NULL;
4209 read_lock(&ls->ls_rsbtbl[bucket].lock);
4210 list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) {
4211 if (!rsb_flag(r, RSB_LOCKS_PURGED))
4214 rsb_clear_flag(r, RSB_LOCKS_PURGED);
4218 read_unlock(&ls->ls_rsbtbl[bucket].lock);
4222 void dlm_grant_after_purge(struct dlm_ls *ls)
4228 r = find_purged_rsb(ls, bucket);
4230 if (bucket == ls->ls_rsbtbl_size - 1)
4237 grant_pending_locks(r);
4238 confirm_master(r, 0);
4246 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
4249 struct dlm_lkb *lkb;
4251 list_for_each_entry(lkb, head, lkb_statequeue) {
4252 if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
4258 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
4261 struct dlm_lkb *lkb;
4263 lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
4266 lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
4269 lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
4275 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
4276 struct dlm_rsb *r, struct dlm_rcom *rc)
4278 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4281 lkb->lkb_nodeid = rc->rc_header.h_nodeid;
4282 lkb->lkb_ownpid = rl->rl_ownpid;
4283 lkb->lkb_remid = rl->rl_lkid;
4284 lkb->lkb_exflags = rl->rl_exflags;
4285 lkb->lkb_flags = rl->rl_flags & 0x0000FFFF;
4286 lkb->lkb_flags |= DLM_IFL_MSTCPY;
4287 lkb->lkb_lvbseq = rl->rl_lvbseq;
4288 lkb->lkb_rqmode = rl->rl_rqmode;
4289 lkb->lkb_grmode = rl->rl_grmode;
4290 /* don't set lkb_status because add_lkb wants to itself */
4292 lkb->lkb_bastaddr = (void *) (long) (rl->rl_asts & AST_BAST);
4293 lkb->lkb_astaddr = (void *) (long) (rl->rl_asts & AST_COMP);
4295 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
4296 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
4297 if (!lkb->lkb_lvbptr)
4299 lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) -
4300 sizeof(struct rcom_lock);
4301 memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
4304 /* Conversions between PR and CW (middle modes) need special handling.
4305 The real granted mode of these converting locks cannot be determined
4306 until all locks have been rebuilt on the rsb (recover_conversion) */
4308 if (rl->rl_wait_type == DLM_MSG_CONVERT && middle_conversion(lkb)) {
4309 rl->rl_status = DLM_LKSTS_CONVERT;
4310 lkb->lkb_grmode = DLM_LOCK_IV;
4311 rsb_set_flag(r, RSB_RECOVER_CONVERT);
4317 /* This lkb may have been recovered in a previous aborted recovery so we need
4318 to check if the rsb already has an lkb with the given remote nodeid/lkid.
4319 If so we just send back a standard reply. If not, we create a new lkb with
4320 the given values and send back our lkid. We send back our lkid by sending
4321 back the rcom_lock struct we got but with the remid field filled in. */
4323 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4325 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4327 struct dlm_lkb *lkb;
4330 if (rl->rl_parent_lkid) {
4331 error = -EOPNOTSUPP;
4335 error = find_rsb(ls, rl->rl_name, rl->rl_namelen, R_MASTER, &r);
4341 lkb = search_remid(r, rc->rc_header.h_nodeid, rl->rl_lkid);
4347 error = create_lkb(ls, &lkb);
4351 error = receive_rcom_lock_args(ls, lkb, r, rc);
4358 add_lkb(r, lkb, rl->rl_status);
4362 /* this is the new value returned to the lock holder for
4363 saving in its process-copy lkb */
4364 rl->rl_remid = lkb->lkb_id;
4371 log_debug(ls, "recover_master_copy %d %x", error, rl->rl_lkid);
4372 rl->rl_result = error;
4376 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4378 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4380 struct dlm_lkb *lkb;
4383 error = find_lkb(ls, rl->rl_lkid, &lkb);
4385 log_error(ls, "recover_process_copy no lkid %x", rl->rl_lkid);
4389 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
4391 error = rl->rl_result;
4393 r = lkb->lkb_resource;
4399 /* There's a chance the new master received our lock before
4400 dlm_recover_master_reply(), this wouldn't happen if we did
4401 a barrier between recover_masters and recover_locks. */
4402 log_debug(ls, "master copy not ready %x r %lx %s", lkb->lkb_id,
4403 (unsigned long)r, r->res_name);
4404 dlm_send_rcom_lock(r, lkb);
4407 log_debug(ls, "master copy exists %x", lkb->lkb_id);
4410 lkb->lkb_remid = rl->rl_remid;
4413 log_error(ls, "dlm_recover_process_copy unknown error %d %x",
4414 error, lkb->lkb_id);
4417 /* an ack for dlm_recover_locks() which waits for replies from
4418 all the locks it sends to new masters */
4419 dlm_recovered_lock(r);
4428 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
4429 int mode, uint32_t flags, void *name, unsigned int namelen,
4430 unsigned long timeout_cs)
4432 struct dlm_lkb *lkb;
4433 struct dlm_args args;
4436 dlm_lock_recovery(ls);
4438 error = create_lkb(ls, &lkb);
4444 if (flags & DLM_LKF_VALBLK) {
4445 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
4446 if (!ua->lksb.sb_lvbptr) {
4454 /* After ua is attached to lkb it will be freed by dlm_free_lkb().
4455 When DLM_IFL_USER is set, the dlm knows that this is a userspace
4456 lock and that lkb_astparam is the dlm_user_args structure. */
4458 error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
4459 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
4460 lkb->lkb_flags |= DLM_IFL_USER;
4461 ua->old_mode = DLM_LOCK_IV;
4468 error = request_lock(ls, lkb, name, namelen, &args);
4484 /* add this new lkb to the per-process list of locks */
4485 spin_lock(&ua->proc->locks_spin);
4487 list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
4488 spin_unlock(&ua->proc->locks_spin);
4490 dlm_unlock_recovery(ls);
4494 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4495 int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
4496 unsigned long timeout_cs)
4498 struct dlm_lkb *lkb;
4499 struct dlm_args args;
4500 struct dlm_user_args *ua;
4503 dlm_lock_recovery(ls);
4505 error = find_lkb(ls, lkid, &lkb);
4509 /* user can change the params on its lock when it converts it, or
4510 add an lvb that didn't exist before */
4512 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4514 if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
4515 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
4516 if (!ua->lksb.sb_lvbptr) {
4521 if (lvb_in && ua->lksb.sb_lvbptr)
4522 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4524 ua->xid = ua_tmp->xid;
4525 ua->castparam = ua_tmp->castparam;
4526 ua->castaddr = ua_tmp->castaddr;
4527 ua->bastparam = ua_tmp->bastparam;
4528 ua->bastaddr = ua_tmp->bastaddr;
4529 ua->user_lksb = ua_tmp->user_lksb;
4530 ua->old_mode = lkb->lkb_grmode;
4532 error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
4533 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
4537 error = convert_lock(ls, lkb, &args);
4539 if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
4544 dlm_unlock_recovery(ls);
4549 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4550 uint32_t flags, uint32_t lkid, char *lvb_in)
4552 struct dlm_lkb *lkb;
4553 struct dlm_args args;
4554 struct dlm_user_args *ua;
4557 dlm_lock_recovery(ls);
4559 error = find_lkb(ls, lkid, &lkb);
4563 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4565 if (lvb_in && ua->lksb.sb_lvbptr)
4566 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4567 if (ua_tmp->castparam)
4568 ua->castparam = ua_tmp->castparam;
4569 ua->user_lksb = ua_tmp->user_lksb;
4571 error = set_unlock_args(flags, ua, &args);
4575 error = unlock_lock(ls, lkb, &args);
4577 if (error == -DLM_EUNLOCK)
4579 /* from validate_unlock_args() */
4580 if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
4585 spin_lock(&ua->proc->locks_spin);
4586 /* dlm_user_add_ast() may have already taken lkb off the proc list */
4587 if (!list_empty(&lkb->lkb_ownqueue))
4588 list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
4589 spin_unlock(&ua->proc->locks_spin);
4593 dlm_unlock_recovery(ls);
4598 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4599 uint32_t flags, uint32_t lkid)
4601 struct dlm_lkb *lkb;
4602 struct dlm_args args;
4603 struct dlm_user_args *ua;
4606 dlm_lock_recovery(ls);
4608 error = find_lkb(ls, lkid, &lkb);
4612 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4613 if (ua_tmp->castparam)
4614 ua->castparam = ua_tmp->castparam;
4615 ua->user_lksb = ua_tmp->user_lksb;
4617 error = set_unlock_args(flags, ua, &args);
4621 error = cancel_lock(ls, lkb, &args);
4623 if (error == -DLM_ECANCEL)
4625 /* from validate_unlock_args() */
4626 if (error == -EBUSY)
4631 dlm_unlock_recovery(ls);
4636 int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
4638 struct dlm_lkb *lkb;
4639 struct dlm_args args;
4640 struct dlm_user_args *ua;
4644 dlm_lock_recovery(ls);
4646 error = find_lkb(ls, lkid, &lkb);
4650 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4652 error = set_unlock_args(flags, ua, &args);
4656 /* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4658 r = lkb->lkb_resource;
4662 error = validate_unlock_args(lkb, &args);
4665 lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
4667 error = _cancel_lock(r, lkb);
4672 if (error == -DLM_ECANCEL)
4674 /* from validate_unlock_args() */
4675 if (error == -EBUSY)
4680 dlm_unlock_recovery(ls);
4684 /* lkb's that are removed from the waiters list by revert are just left on the
4685 orphans list with the granted orphan locks, to be freed by purge */
4687 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4689 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
4690 struct dlm_args args;
4694 mutex_lock(&ls->ls_orphans_mutex);
4695 list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
4696 mutex_unlock(&ls->ls_orphans_mutex);
4698 set_unlock_args(0, ua, &args);
4700 error = cancel_lock(ls, lkb, &args);
4701 if (error == -DLM_ECANCEL)
4706 /* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4707 Regardless of what rsb queue the lock is on, it's removed and freed. */
4709 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4711 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
4712 struct dlm_args args;
4715 set_unlock_args(DLM_LKF_FORCEUNLOCK, ua, &args);
4717 error = unlock_lock(ls, lkb, &args);
4718 if (error == -DLM_EUNLOCK)
4723 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4724 (which does lock_rsb) due to deadlock with receiving a message that does
4725 lock_rsb followed by dlm_user_add_ast() */
4727 static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
4728 struct dlm_user_proc *proc)
4730 struct dlm_lkb *lkb = NULL;
4732 mutex_lock(&ls->ls_clear_proc_locks);
4733 if (list_empty(&proc->locks))
4736 lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
4737 list_del_init(&lkb->lkb_ownqueue);
4739 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4740 lkb->lkb_flags |= DLM_IFL_ORPHAN;
4742 lkb->lkb_flags |= DLM_IFL_DEAD;
4744 mutex_unlock(&ls->ls_clear_proc_locks);
4748 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4749 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4750 which we clear here. */
4752 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
4753 list, and no more device_writes should add lkb's to proc->locks list; so we
4754 shouldn't need to take asts_spin or locks_spin here. this assumes that
4755 device reads/writes/closes are serialized -- FIXME: we may need to serialize
4758 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4760 struct dlm_lkb *lkb, *safe;
4762 dlm_lock_recovery(ls);
4765 lkb = del_proc_lock(ls, proc);
4769 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4770 orphan_proc_lock(ls, lkb);
4772 unlock_proc_lock(ls, lkb);
4774 /* this removes the reference for the proc->locks list
4775 added by dlm_user_request, it may result in the lkb
4781 mutex_lock(&ls->ls_clear_proc_locks);
4783 /* in-progress unlocks */
4784 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4785 list_del_init(&lkb->lkb_ownqueue);
4786 lkb->lkb_flags |= DLM_IFL_DEAD;
4790 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4791 lkb->lkb_ast_type = 0;
4792 list_del(&lkb->lkb_astqueue);
4796 mutex_unlock(&ls->ls_clear_proc_locks);
4797 dlm_unlock_recovery(ls);
4800 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4802 struct dlm_lkb *lkb, *safe;
4806 spin_lock(&proc->locks_spin);
4807 if (!list_empty(&proc->locks)) {
4808 lkb = list_entry(proc->locks.next, struct dlm_lkb,
4810 list_del_init(&lkb->lkb_ownqueue);
4812 spin_unlock(&proc->locks_spin);
4817 lkb->lkb_flags |= DLM_IFL_DEAD;
4818 unlock_proc_lock(ls, lkb);
4819 dlm_put_lkb(lkb); /* ref from proc->locks list */
4822 spin_lock(&proc->locks_spin);
4823 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4824 list_del_init(&lkb->lkb_ownqueue);
4825 lkb->lkb_flags |= DLM_IFL_DEAD;
4828 spin_unlock(&proc->locks_spin);
4830 spin_lock(&proc->asts_spin);
4831 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4832 list_del(&lkb->lkb_astqueue);
4835 spin_unlock(&proc->asts_spin);
4838 /* pid of 0 means purge all orphans */
4840 static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
4842 struct dlm_lkb *lkb, *safe;
4844 mutex_lock(&ls->ls_orphans_mutex);
4845 list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
4846 if (pid && lkb->lkb_ownpid != pid)
4848 unlock_proc_lock(ls, lkb);
4849 list_del_init(&lkb->lkb_ownqueue);
4852 mutex_unlock(&ls->ls_orphans_mutex);
4855 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
4857 struct dlm_message *ms;
4858 struct dlm_mhandle *mh;
4861 error = _create_message(ls, sizeof(struct dlm_message), nodeid,
4862 DLM_MSG_PURGE, &ms, &mh);
4865 ms->m_nodeid = nodeid;
4868 return send_message(mh, ms);
4871 int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
4872 int nodeid, int pid)
4876 if (nodeid != dlm_our_nodeid()) {
4877 error = send_purge(ls, nodeid, pid);
4879 dlm_lock_recovery(ls);
4880 if (pid == current->pid)
4881 purge_proc_locks(ls, proc);
4883 do_purge(ls, nodeid, pid);
4884 dlm_unlock_recovery(ls);