1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) 2005-2007 Red Hat, Inc. All rights reserved.
6 ** This copyrighted material is made available to anyone wishing to use,
7 ** modify, copy, or redistribute it subject to the terms and conditions
8 ** of the GNU General Public License v.2.
10 *******************************************************************************
11 ******************************************************************************/
13 /* Central locking logic has four stages:
33 Stage 1 (lock, unlock) is mainly about checking input args and
34 splitting into one of the four main operations:
36 dlm_lock = request_lock
37 dlm_lock+CONVERT = convert_lock
38 dlm_unlock = unlock_lock
39 dlm_unlock+CANCEL = cancel_lock
41 Stage 2, xxxx_lock(), just finds and locks the relevant rsb which is
42 provided to the next stage.
44 Stage 3, _xxxx_lock(), determines if the operation is local or remote.
45 When remote, it calls send_xxxx(), when local it calls do_xxxx().
47 Stage 4, do_xxxx(), is the guts of the operation. It manipulates the
48 given rsb and lkb and queues callbacks.
50 For remote operations, send_xxxx() results in the corresponding do_xxxx()
51 function being executed on the remote node. The connecting send/receive
52 calls on local (L) and remote (R) nodes:
54 L: send_xxxx() -> R: receive_xxxx()
56 L: receive_xxxx_reply() <- R: send_xxxx_reply()
58 #include <linux/types.h>
59 #include "dlm_internal.h"
60 #include <linux/dlm_device.h>
63 #include "requestqueue.h"
67 #include "lockspace.h"
72 #include "lvb_table.h"
76 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb);
77 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb);
78 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb);
79 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb);
80 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb);
81 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode);
82 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb);
83 static int send_remove(struct dlm_rsb *r);
84 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
85 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
86 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
87 struct dlm_message *ms);
88 static int receive_extralen(struct dlm_message *ms);
89 static void do_purge(struct dlm_ls *ls, int nodeid, int pid);
90 static void del_timeout(struct dlm_lkb *lkb);
93 * Lock compatibilty matrix - thanks Steve
94 * UN = Unlocked state. Not really a state, used as a flag
95 * PD = Padding. Used to make the matrix a nice power of two in size
96 * Other states are the same as the VMS DLM.
97 * Usage: matrix[grmode+1][rqmode+1] (although m[rq+1][gr+1] is the same)
100 static const int __dlm_compat_matrix[8][8] = {
101 /* UN NL CR CW PR PW EX PD */
102 {1, 1, 1, 1, 1, 1, 1, 0}, /* UN */
103 {1, 1, 1, 1, 1, 1, 1, 0}, /* NL */
104 {1, 1, 1, 1, 1, 1, 0, 0}, /* CR */
105 {1, 1, 1, 1, 0, 0, 0, 0}, /* CW */
106 {1, 1, 1, 0, 1, 0, 0, 0}, /* PR */
107 {1, 1, 1, 0, 0, 0, 0, 0}, /* PW */
108 {1, 1, 0, 0, 0, 0, 0, 0}, /* EX */
109 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
113 * This defines the direction of transfer of LVB data.
114 * Granted mode is the row; requested mode is the column.
115 * Usage: matrix[grmode+1][rqmode+1]
116 * 1 = LVB is returned to the caller
117 * 0 = LVB is written to the resource
118 * -1 = nothing happens to the LVB
121 const int dlm_lvb_operations[8][8] = {
122 /* UN NL CR CW PR PW EX PD*/
123 { -1, 1, 1, 1, 1, 1, 1, -1 }, /* UN */
124 { -1, 1, 1, 1, 1, 1, 1, 0 }, /* NL */
125 { -1, -1, 1, 1, 1, 1, 1, 0 }, /* CR */
126 { -1, -1, -1, 1, 1, 1, 1, 0 }, /* CW */
127 { -1, -1, -1, -1, 1, 1, 1, 0 }, /* PR */
128 { -1, 0, 0, 0, 0, 0, 1, 0 }, /* PW */
129 { -1, 0, 0, 0, 0, 0, 0, 0 }, /* EX */
130 { -1, 0, 0, 0, 0, 0, 0, 0 } /* PD */
133 #define modes_compat(gr, rq) \
134 __dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
136 int dlm_modes_compat(int mode1, int mode2)
138 return __dlm_compat_matrix[mode1 + 1][mode2 + 1];
142 * Compatibility matrix for conversions with QUECVT set.
143 * Granted mode is the row; requested mode is the column.
144 * Usage: matrix[grmode+1][rqmode+1]
147 static const int __quecvt_compat_matrix[8][8] = {
148 /* UN NL CR CW PR PW EX PD */
149 {0, 0, 0, 0, 0, 0, 0, 0}, /* UN */
150 {0, 0, 1, 1, 1, 1, 1, 0}, /* NL */
151 {0, 0, 0, 1, 1, 1, 1, 0}, /* CR */
152 {0, 0, 0, 0, 1, 1, 1, 0}, /* CW */
153 {0, 0, 0, 1, 0, 1, 1, 0}, /* PR */
154 {0, 0, 0, 0, 0, 0, 1, 0}, /* PW */
155 {0, 0, 0, 0, 0, 0, 0, 0}, /* EX */
156 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
159 void dlm_print_lkb(struct dlm_lkb *lkb)
161 printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x\n"
162 " status %d rqmode %d grmode %d wait_type %d ast_type %d\n",
163 lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags,
164 lkb->lkb_flags, lkb->lkb_status, lkb->lkb_rqmode,
165 lkb->lkb_grmode, lkb->lkb_wait_type, lkb->lkb_ast_type);
168 void dlm_print_rsb(struct dlm_rsb *r)
170 printk(KERN_ERR "rsb: nodeid %d flags %lx first %x rlc %d name %s\n",
171 r->res_nodeid, r->res_flags, r->res_first_lkid,
172 r->res_recover_locks_count, r->res_name);
175 void dlm_dump_rsb(struct dlm_rsb *r)
181 printk(KERN_ERR "rsb: root_list empty %d recover_list empty %d\n",
182 list_empty(&r->res_root_list), list_empty(&r->res_recover_list));
183 printk(KERN_ERR "rsb lookup list\n");
184 list_for_each_entry(lkb, &r->res_lookup, lkb_rsb_lookup)
186 printk(KERN_ERR "rsb grant queue:\n");
187 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue)
189 printk(KERN_ERR "rsb convert queue:\n");
190 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue)
192 printk(KERN_ERR "rsb wait queue:\n");
193 list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue)
197 /* Threads cannot use the lockspace while it's being recovered */
199 static inline void dlm_lock_recovery(struct dlm_ls *ls)
201 down_read(&ls->ls_in_recovery);
204 void dlm_unlock_recovery(struct dlm_ls *ls)
206 up_read(&ls->ls_in_recovery);
209 int dlm_lock_recovery_try(struct dlm_ls *ls)
211 return down_read_trylock(&ls->ls_in_recovery);
214 static inline int can_be_queued(struct dlm_lkb *lkb)
216 return !(lkb->lkb_exflags & DLM_LKF_NOQUEUE);
219 static inline int force_blocking_asts(struct dlm_lkb *lkb)
221 return (lkb->lkb_exflags & DLM_LKF_NOQUEUEBAST);
224 static inline int is_demoted(struct dlm_lkb *lkb)
226 return (lkb->lkb_sbflags & DLM_SBF_DEMOTED);
229 static inline int is_altmode(struct dlm_lkb *lkb)
231 return (lkb->lkb_sbflags & DLM_SBF_ALTMODE);
234 static inline int is_granted(struct dlm_lkb *lkb)
236 return (lkb->lkb_status == DLM_LKSTS_GRANTED);
239 static inline int is_remote(struct dlm_rsb *r)
241 DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r););
242 return !!r->res_nodeid;
245 static inline int is_process_copy(struct dlm_lkb *lkb)
247 return (lkb->lkb_nodeid && !(lkb->lkb_flags & DLM_IFL_MSTCPY));
250 static inline int is_master_copy(struct dlm_lkb *lkb)
252 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
253 DLM_ASSERT(lkb->lkb_nodeid, dlm_print_lkb(lkb););
254 return (lkb->lkb_flags & DLM_IFL_MSTCPY) ? 1 : 0;
257 static inline int middle_conversion(struct dlm_lkb *lkb)
259 if ((lkb->lkb_grmode==DLM_LOCK_PR && lkb->lkb_rqmode==DLM_LOCK_CW) ||
260 (lkb->lkb_rqmode==DLM_LOCK_PR && lkb->lkb_grmode==DLM_LOCK_CW))
265 static inline int down_conversion(struct dlm_lkb *lkb)
267 return (!middle_conversion(lkb) && lkb->lkb_rqmode < lkb->lkb_grmode);
270 static inline int is_overlap_unlock(struct dlm_lkb *lkb)
272 return lkb->lkb_flags & DLM_IFL_OVERLAP_UNLOCK;
275 static inline int is_overlap_cancel(struct dlm_lkb *lkb)
277 return lkb->lkb_flags & DLM_IFL_OVERLAP_CANCEL;
280 static inline int is_overlap(struct dlm_lkb *lkb)
282 return (lkb->lkb_flags & (DLM_IFL_OVERLAP_UNLOCK |
283 DLM_IFL_OVERLAP_CANCEL));
286 static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
288 if (is_master_copy(lkb))
293 DLM_ASSERT(lkb->lkb_lksb, dlm_print_lkb(lkb););
295 /* if the operation was a cancel, then return -DLM_ECANCEL, if a
296 timeout caused the cancel then return -ETIMEDOUT */
297 if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_TIMEOUT_CANCEL)) {
298 lkb->lkb_flags &= ~DLM_IFL_TIMEOUT_CANCEL;
302 if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_DEADLOCK_CANCEL)) {
303 lkb->lkb_flags &= ~DLM_IFL_DEADLOCK_CANCEL;
307 lkb->lkb_lksb->sb_status = rv;
308 lkb->lkb_lksb->sb_flags = lkb->lkb_sbflags;
310 dlm_add_ast(lkb, AST_COMP);
313 static inline void queue_cast_overlap(struct dlm_rsb *r, struct dlm_lkb *lkb)
316 is_overlap_unlock(lkb) ? -DLM_EUNLOCK : -DLM_ECANCEL);
319 static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
321 if (is_master_copy(lkb))
322 send_bast(r, lkb, rqmode);
324 lkb->lkb_bastmode = rqmode;
325 dlm_add_ast(lkb, AST_BAST);
330 * Basic operations on rsb's and lkb's
333 static struct dlm_rsb *create_rsb(struct dlm_ls *ls, char *name, int len)
337 r = dlm_allocate_rsb(ls, len);
343 memcpy(r->res_name, name, len);
344 mutex_init(&r->res_mutex);
346 INIT_LIST_HEAD(&r->res_lookup);
347 INIT_LIST_HEAD(&r->res_grantqueue);
348 INIT_LIST_HEAD(&r->res_convertqueue);
349 INIT_LIST_HEAD(&r->res_waitqueue);
350 INIT_LIST_HEAD(&r->res_root_list);
351 INIT_LIST_HEAD(&r->res_recover_list);
356 static int search_rsb_list(struct list_head *head, char *name, int len,
357 unsigned int flags, struct dlm_rsb **r_ret)
362 list_for_each_entry(r, head, res_hashchain) {
363 if (len == r->res_length && !memcmp(name, r->res_name, len))
369 if (r->res_nodeid && (flags & R_MASTER))
375 static int _search_rsb(struct dlm_ls *ls, char *name, int len, int b,
376 unsigned int flags, struct dlm_rsb **r_ret)
381 error = search_rsb_list(&ls->ls_rsbtbl[b].list, name, len, flags, &r);
383 kref_get(&r->res_ref);
386 error = search_rsb_list(&ls->ls_rsbtbl[b].toss, name, len, flags, &r);
390 list_move(&r->res_hashchain, &ls->ls_rsbtbl[b].list);
392 if (dlm_no_directory(ls))
395 if (r->res_nodeid == -1) {
396 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
397 r->res_first_lkid = 0;
398 } else if (r->res_nodeid > 0) {
399 rsb_set_flag(r, RSB_MASTER_UNCERTAIN);
400 r->res_first_lkid = 0;
402 DLM_ASSERT(r->res_nodeid == 0, dlm_print_rsb(r););
403 DLM_ASSERT(!rsb_flag(r, RSB_MASTER_UNCERTAIN),);
410 static int search_rsb(struct dlm_ls *ls, char *name, int len, int b,
411 unsigned int flags, struct dlm_rsb **r_ret)
414 write_lock(&ls->ls_rsbtbl[b].lock);
415 error = _search_rsb(ls, name, len, b, flags, r_ret);
416 write_unlock(&ls->ls_rsbtbl[b].lock);
421 * Find rsb in rsbtbl and potentially create/add one
423 * Delaying the release of rsb's has a similar benefit to applications keeping
424 * NL locks on an rsb, but without the guarantee that the cached master value
425 * will still be valid when the rsb is reused. Apps aren't always smart enough
426 * to keep NL locks on an rsb that they may lock again shortly; this can lead
427 * to excessive master lookups and removals if we don't delay the release.
429 * Searching for an rsb means looking through both the normal list and toss
430 * list. When found on the toss list the rsb is moved to the normal list with
431 * ref count of 1; when found on normal list the ref count is incremented.
434 static int find_rsb(struct dlm_ls *ls, char *name, int namelen,
435 unsigned int flags, struct dlm_rsb **r_ret)
437 struct dlm_rsb *r, *tmp;
438 uint32_t hash, bucket;
441 if (dlm_no_directory(ls))
444 hash = jhash(name, namelen, 0);
445 bucket = hash & (ls->ls_rsbtbl_size - 1);
447 error = search_rsb(ls, name, namelen, bucket, flags, &r);
451 if (error == -EBADR && !(flags & R_CREATE))
454 /* the rsb was found but wasn't a master copy */
455 if (error == -ENOTBLK)
459 r = create_rsb(ls, name, namelen);
464 r->res_bucket = bucket;
466 kref_init(&r->res_ref);
468 /* With no directory, the master can be set immediately */
469 if (dlm_no_directory(ls)) {
470 int nodeid = dlm_dir_nodeid(r);
471 if (nodeid == dlm_our_nodeid())
473 r->res_nodeid = nodeid;
476 write_lock(&ls->ls_rsbtbl[bucket].lock);
477 error = _search_rsb(ls, name, namelen, bucket, 0, &tmp);
479 write_unlock(&ls->ls_rsbtbl[bucket].lock);
484 list_add(&r->res_hashchain, &ls->ls_rsbtbl[bucket].list);
485 write_unlock(&ls->ls_rsbtbl[bucket].lock);
492 int dlm_find_rsb(struct dlm_ls *ls, char *name, int namelen,
493 unsigned int flags, struct dlm_rsb **r_ret)
495 return find_rsb(ls, name, namelen, flags, r_ret);
498 /* This is only called to add a reference when the code already holds
499 a valid reference to the rsb, so there's no need for locking. */
501 static inline void hold_rsb(struct dlm_rsb *r)
503 kref_get(&r->res_ref);
506 void dlm_hold_rsb(struct dlm_rsb *r)
511 static void toss_rsb(struct kref *kref)
513 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
514 struct dlm_ls *ls = r->res_ls;
516 DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
517 kref_init(&r->res_ref);
518 list_move(&r->res_hashchain, &ls->ls_rsbtbl[r->res_bucket].toss);
519 r->res_toss_time = jiffies;
521 dlm_free_lvb(r->res_lvbptr);
522 r->res_lvbptr = NULL;
526 /* When all references to the rsb are gone it's transfered to
527 the tossed list for later disposal. */
529 static void put_rsb(struct dlm_rsb *r)
531 struct dlm_ls *ls = r->res_ls;
532 uint32_t bucket = r->res_bucket;
534 write_lock(&ls->ls_rsbtbl[bucket].lock);
535 kref_put(&r->res_ref, toss_rsb);
536 write_unlock(&ls->ls_rsbtbl[bucket].lock);
539 void dlm_put_rsb(struct dlm_rsb *r)
544 /* See comment for unhold_lkb */
546 static void unhold_rsb(struct dlm_rsb *r)
549 rv = kref_put(&r->res_ref, toss_rsb);
550 DLM_ASSERT(!rv, dlm_dump_rsb(r););
553 static void kill_rsb(struct kref *kref)
555 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
557 /* All work is done after the return from kref_put() so we
558 can release the write_lock before the remove and free. */
560 DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
561 DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
562 DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
563 DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
564 DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
565 DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
568 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
569 The rsb must exist as long as any lkb's for it do. */
571 static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
574 lkb->lkb_resource = r;
577 static void detach_lkb(struct dlm_lkb *lkb)
579 if (lkb->lkb_resource) {
580 put_rsb(lkb->lkb_resource);
581 lkb->lkb_resource = NULL;
585 static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
587 struct dlm_lkb *lkb, *tmp;
591 lkb = dlm_allocate_lkb(ls);
595 lkb->lkb_nodeid = -1;
596 lkb->lkb_grmode = DLM_LOCK_IV;
597 kref_init(&lkb->lkb_ref);
598 INIT_LIST_HEAD(&lkb->lkb_ownqueue);
599 INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
600 INIT_LIST_HEAD(&lkb->lkb_time_list);
602 get_random_bytes(&bucket, sizeof(bucket));
603 bucket &= (ls->ls_lkbtbl_size - 1);
605 write_lock(&ls->ls_lkbtbl[bucket].lock);
607 /* counter can roll over so we must verify lkid is not in use */
610 lkid = (bucket << 16) | ls->ls_lkbtbl[bucket].counter++;
612 list_for_each_entry(tmp, &ls->ls_lkbtbl[bucket].list,
614 if (tmp->lkb_id != lkid)
622 list_add(&lkb->lkb_idtbl_list, &ls->ls_lkbtbl[bucket].list);
623 write_unlock(&ls->ls_lkbtbl[bucket].lock);
629 static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid)
632 uint16_t bucket = (lkid >> 16);
634 list_for_each_entry(lkb, &ls->ls_lkbtbl[bucket].list, lkb_idtbl_list) {
635 if (lkb->lkb_id == lkid)
641 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
644 uint16_t bucket = (lkid >> 16);
646 if (bucket >= ls->ls_lkbtbl_size)
649 read_lock(&ls->ls_lkbtbl[bucket].lock);
650 lkb = __find_lkb(ls, lkid);
652 kref_get(&lkb->lkb_ref);
653 read_unlock(&ls->ls_lkbtbl[bucket].lock);
656 return lkb ? 0 : -ENOENT;
659 static void kill_lkb(struct kref *kref)
661 struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
663 /* All work is done after the return from kref_put() so we
664 can release the write_lock before the detach_lkb */
666 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
669 /* __put_lkb() is used when an lkb may not have an rsb attached to
670 it so we need to provide the lockspace explicitly */
672 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
674 uint16_t bucket = (lkb->lkb_id >> 16);
676 write_lock(&ls->ls_lkbtbl[bucket].lock);
677 if (kref_put(&lkb->lkb_ref, kill_lkb)) {
678 list_del(&lkb->lkb_idtbl_list);
679 write_unlock(&ls->ls_lkbtbl[bucket].lock);
683 /* for local/process lkbs, lvbptr points to caller's lksb */
684 if (lkb->lkb_lvbptr && is_master_copy(lkb))
685 dlm_free_lvb(lkb->lkb_lvbptr);
689 write_unlock(&ls->ls_lkbtbl[bucket].lock);
694 int dlm_put_lkb(struct dlm_lkb *lkb)
698 DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb););
699 DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb););
701 ls = lkb->lkb_resource->res_ls;
702 return __put_lkb(ls, lkb);
705 /* This is only called to add a reference when the code already holds
706 a valid reference to the lkb, so there's no need for locking. */
708 static inline void hold_lkb(struct dlm_lkb *lkb)
710 kref_get(&lkb->lkb_ref);
713 /* This is called when we need to remove a reference and are certain
714 it's not the last ref. e.g. del_lkb is always called between a
715 find_lkb/put_lkb and is always the inverse of a previous add_lkb.
716 put_lkb would work fine, but would involve unnecessary locking */
718 static inline void unhold_lkb(struct dlm_lkb *lkb)
721 rv = kref_put(&lkb->lkb_ref, kill_lkb);
722 DLM_ASSERT(!rv, dlm_print_lkb(lkb););
725 static void lkb_add_ordered(struct list_head *new, struct list_head *head,
728 struct dlm_lkb *lkb = NULL;
730 list_for_each_entry(lkb, head, lkb_statequeue)
731 if (lkb->lkb_rqmode < mode)
735 list_add_tail(new, head);
737 __list_add(new, lkb->lkb_statequeue.prev, &lkb->lkb_statequeue);
740 /* add/remove lkb to rsb's grant/convert/wait queue */
742 static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status)
744 kref_get(&lkb->lkb_ref);
746 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
748 lkb->lkb_status = status;
751 case DLM_LKSTS_WAITING:
752 if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
753 list_add(&lkb->lkb_statequeue, &r->res_waitqueue);
755 list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue);
757 case DLM_LKSTS_GRANTED:
758 /* convention says granted locks kept in order of grmode */
759 lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue,
762 case DLM_LKSTS_CONVERT:
763 if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
764 list_add(&lkb->lkb_statequeue, &r->res_convertqueue);
766 list_add_tail(&lkb->lkb_statequeue,
767 &r->res_convertqueue);
770 DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status););
774 static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
777 list_del(&lkb->lkb_statequeue);
781 static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
785 add_lkb(r, lkb, sts);
789 static int msg_reply_type(int mstype)
792 case DLM_MSG_REQUEST:
793 return DLM_MSG_REQUEST_REPLY;
794 case DLM_MSG_CONVERT:
795 return DLM_MSG_CONVERT_REPLY;
797 return DLM_MSG_UNLOCK_REPLY;
799 return DLM_MSG_CANCEL_REPLY;
801 return DLM_MSG_LOOKUP_REPLY;
806 /* add/remove lkb from global waiters list of lkb's waiting for
807 a reply from a remote node */
809 static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
811 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
814 mutex_lock(&ls->ls_waiters_mutex);
816 if (is_overlap_unlock(lkb) ||
817 (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
822 if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
825 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
828 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
834 lkb->lkb_wait_count++;
837 log_debug(ls, "add overlap %x cur %d new %d count %d flags %x",
838 lkb->lkb_id, lkb->lkb_wait_type, mstype,
839 lkb->lkb_wait_count, lkb->lkb_flags);
843 DLM_ASSERT(!lkb->lkb_wait_count,
845 printk("wait_count %d\n", lkb->lkb_wait_count););
847 lkb->lkb_wait_count++;
848 lkb->lkb_wait_type = mstype;
850 list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
853 log_error(ls, "add_to_waiters %x error %d flags %x %d %d %s",
854 lkb->lkb_id, error, lkb->lkb_flags, mstype,
855 lkb->lkb_wait_type, lkb->lkb_resource->res_name);
856 mutex_unlock(&ls->ls_waiters_mutex);
860 /* We clear the RESEND flag because we might be taking an lkb off the waiters
861 list as part of process_requestqueue (e.g. a lookup that has an optimized
862 request reply on the requestqueue) between dlm_recover_waiters_pre() which
863 set RESEND and dlm_recover_waiters_post() */
865 static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
867 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
868 int overlap_done = 0;
870 if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
871 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
876 if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
877 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
882 /* N.B. type of reply may not always correspond to type of original
883 msg due to lookup->request optimization, verify others? */
885 if (lkb->lkb_wait_type) {
886 lkb->lkb_wait_type = 0;
890 log_error(ls, "remove_from_waiters lkid %x flags %x types %d %d",
891 lkb->lkb_id, lkb->lkb_flags, mstype, lkb->lkb_wait_type);
895 /* the force-unlock/cancel has completed and we haven't recvd a reply
896 to the op that was in progress prior to the unlock/cancel; we
897 give up on any reply to the earlier op. FIXME: not sure when/how
900 if (overlap_done && lkb->lkb_wait_type) {
901 log_error(ls, "remove_from_waiters %x reply %d give up on %d",
902 lkb->lkb_id, mstype, lkb->lkb_wait_type);
903 lkb->lkb_wait_count--;
904 lkb->lkb_wait_type = 0;
907 DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
909 lkb->lkb_flags &= ~DLM_IFL_RESEND;
910 lkb->lkb_wait_count--;
911 if (!lkb->lkb_wait_count)
912 list_del_init(&lkb->lkb_wait_reply);
917 static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
919 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
922 mutex_lock(&ls->ls_waiters_mutex);
923 error = _remove_from_waiters(lkb, mstype);
924 mutex_unlock(&ls->ls_waiters_mutex);
928 /* Handles situations where we might be processing a "fake" or "stub" reply in
929 which we can't try to take waiters_mutex again. */
931 static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
933 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
936 if (ms != &ls->ls_stub_ms)
937 mutex_lock(&ls->ls_waiters_mutex);
938 error = _remove_from_waiters(lkb, ms->m_type);
939 if (ms != &ls->ls_stub_ms)
940 mutex_unlock(&ls->ls_waiters_mutex);
944 static void dir_remove(struct dlm_rsb *r)
948 if (dlm_no_directory(r->res_ls))
951 to_nodeid = dlm_dir_nodeid(r);
952 if (to_nodeid != dlm_our_nodeid())
955 dlm_dir_remove_entry(r->res_ls, to_nodeid,
956 r->res_name, r->res_length);
959 /* FIXME: shouldn't this be able to exit as soon as one non-due rsb is
960 found since they are in order of newest to oldest? */
962 static int shrink_bucket(struct dlm_ls *ls, int b)
965 int count = 0, found;
969 write_lock(&ls->ls_rsbtbl[b].lock);
970 list_for_each_entry_reverse(r, &ls->ls_rsbtbl[b].toss,
972 if (!time_after_eq(jiffies, r->res_toss_time +
973 dlm_config.ci_toss_secs * HZ))
980 write_unlock(&ls->ls_rsbtbl[b].lock);
984 if (kref_put(&r->res_ref, kill_rsb)) {
985 list_del(&r->res_hashchain);
986 write_unlock(&ls->ls_rsbtbl[b].lock);
993 write_unlock(&ls->ls_rsbtbl[b].lock);
994 log_error(ls, "tossed rsb in use %s", r->res_name);
1001 void dlm_scan_rsbs(struct dlm_ls *ls)
1005 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1006 shrink_bucket(ls, i);
1007 if (dlm_locking_stopped(ls))
1013 static void add_timeout(struct dlm_lkb *lkb)
1015 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1017 if (is_master_copy(lkb)) {
1018 lkb->lkb_timestamp = jiffies;
1022 if (test_bit(LSFL_TIMEWARN, &ls->ls_flags) &&
1023 !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1024 lkb->lkb_flags |= DLM_IFL_WATCH_TIMEWARN;
1027 if (lkb->lkb_exflags & DLM_LKF_TIMEOUT)
1032 DLM_ASSERT(list_empty(&lkb->lkb_time_list), dlm_print_lkb(lkb););
1033 mutex_lock(&ls->ls_timeout_mutex);
1035 lkb->lkb_timestamp = jiffies;
1036 list_add_tail(&lkb->lkb_time_list, &ls->ls_timeout);
1037 mutex_unlock(&ls->ls_timeout_mutex);
1040 static void del_timeout(struct dlm_lkb *lkb)
1042 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1044 mutex_lock(&ls->ls_timeout_mutex);
1045 if (!list_empty(&lkb->lkb_time_list)) {
1046 list_del_init(&lkb->lkb_time_list);
1049 mutex_unlock(&ls->ls_timeout_mutex);
1052 /* FIXME: is it safe to look at lkb_exflags, lkb_flags, lkb_timestamp, and
1053 lkb_lksb_timeout without lock_rsb? Note: we can't lock timeout_mutex
1054 and then lock rsb because of lock ordering in add_timeout. We may need
1055 to specify some special timeout-related bits in the lkb that are just to
1056 be accessed under the timeout_mutex. */
1058 void dlm_scan_timeout(struct dlm_ls *ls)
1061 struct dlm_lkb *lkb;
1062 int do_cancel, do_warn;
1065 if (dlm_locking_stopped(ls))
1070 mutex_lock(&ls->ls_timeout_mutex);
1071 list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list) {
1073 if ((lkb->lkb_exflags & DLM_LKF_TIMEOUT) &&
1074 time_after_eq(jiffies, lkb->lkb_timestamp +
1075 lkb->lkb_timeout_cs * HZ/100))
1078 if ((lkb->lkb_flags & DLM_IFL_WATCH_TIMEWARN) &&
1079 time_after_eq(jiffies, lkb->lkb_timestamp +
1080 dlm_config.ci_timewarn_cs * HZ/100))
1083 if (!do_cancel && !do_warn)
1088 mutex_unlock(&ls->ls_timeout_mutex);
1090 if (!do_cancel && !do_warn)
1093 r = lkb->lkb_resource;
1098 /* clear flag so we only warn once */
1099 lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1100 if (!(lkb->lkb_exflags & DLM_LKF_TIMEOUT))
1102 dlm_timeout_warn(lkb);
1106 log_debug(ls, "timeout cancel %x node %d %s",
1107 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1108 lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1109 lkb->lkb_flags |= DLM_IFL_TIMEOUT_CANCEL;
1111 _cancel_lock(r, lkb);
1120 /* This is only called by dlm_recoverd, and we rely on dlm_ls_stop() stopping
1121 dlm_recoverd before checking/setting ls_recover_begin. */
1123 void dlm_adjust_timeouts(struct dlm_ls *ls)
1125 struct dlm_lkb *lkb;
1126 long adj = jiffies - ls->ls_recover_begin;
1128 ls->ls_recover_begin = 0;
1129 mutex_lock(&ls->ls_timeout_mutex);
1130 list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list)
1131 lkb->lkb_timestamp += adj;
1132 mutex_unlock(&ls->ls_timeout_mutex);
1135 /* lkb is master or local copy */
1137 static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1139 int b, len = r->res_ls->ls_lvblen;
1141 /* b=1 lvb returned to caller
1142 b=0 lvb written to rsb or invalidated
1145 b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1148 if (!lkb->lkb_lvbptr)
1151 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1157 memcpy(lkb->lkb_lvbptr, r->res_lvbptr, len);
1158 lkb->lkb_lvbseq = r->res_lvbseq;
1160 } else if (b == 0) {
1161 if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1162 rsb_set_flag(r, RSB_VALNOTVALID);
1166 if (!lkb->lkb_lvbptr)
1169 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1173 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1178 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, len);
1180 lkb->lkb_lvbseq = r->res_lvbseq;
1181 rsb_clear_flag(r, RSB_VALNOTVALID);
1184 if (rsb_flag(r, RSB_VALNOTVALID))
1185 lkb->lkb_sbflags |= DLM_SBF_VALNOTVALID;
1188 static void set_lvb_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1190 if (lkb->lkb_grmode < DLM_LOCK_PW)
1193 if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1194 rsb_set_flag(r, RSB_VALNOTVALID);
1198 if (!lkb->lkb_lvbptr)
1201 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1205 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1210 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
1212 rsb_clear_flag(r, RSB_VALNOTVALID);
1215 /* lkb is process copy (pc) */
1217 static void set_lvb_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1218 struct dlm_message *ms)
1222 if (!lkb->lkb_lvbptr)
1225 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1228 b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1230 int len = receive_extralen(ms);
1231 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
1232 lkb->lkb_lvbseq = ms->m_lvbseq;
1236 /* Manipulate lkb's on rsb's convert/granted/waiting queues
1237 remove_lock -- used for unlock, removes lkb from granted
1238 revert_lock -- used for cancel, moves lkb from convert to granted
1239 grant_lock -- used for request and convert, adds lkb to granted or
1240 moves lkb from convert or waiting to granted
1242 Each of these is used for master or local copy lkb's. There is
1243 also a _pc() variation used to make the corresponding change on
1244 a process copy (pc) lkb. */
1246 static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1249 lkb->lkb_grmode = DLM_LOCK_IV;
1250 /* this unhold undoes the original ref from create_lkb()
1251 so this leads to the lkb being freed */
1255 static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1257 set_lvb_unlock(r, lkb);
1258 _remove_lock(r, lkb);
1261 static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1263 _remove_lock(r, lkb);
1266 /* returns: 0 did nothing
1267 1 moved lock to granted
1270 static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1274 lkb->lkb_rqmode = DLM_LOCK_IV;
1276 switch (lkb->lkb_status) {
1277 case DLM_LKSTS_GRANTED:
1279 case DLM_LKSTS_CONVERT:
1280 move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1283 case DLM_LKSTS_WAITING:
1285 lkb->lkb_grmode = DLM_LOCK_IV;
1286 /* this unhold undoes the original ref from create_lkb()
1287 so this leads to the lkb being freed */
1292 log_print("invalid status for revert %d", lkb->lkb_status);
1297 static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1299 return revert_lock(r, lkb);
1302 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1304 if (lkb->lkb_grmode != lkb->lkb_rqmode) {
1305 lkb->lkb_grmode = lkb->lkb_rqmode;
1306 if (lkb->lkb_status)
1307 move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1309 add_lkb(r, lkb, DLM_LKSTS_GRANTED);
1312 lkb->lkb_rqmode = DLM_LOCK_IV;
1315 static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1317 set_lvb_lock(r, lkb);
1318 _grant_lock(r, lkb);
1319 lkb->lkb_highbast = 0;
1322 static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1323 struct dlm_message *ms)
1325 set_lvb_lock_pc(r, lkb, ms);
1326 _grant_lock(r, lkb);
1329 /* called by grant_pending_locks() which means an async grant message must
1330 be sent to the requesting node in addition to granting the lock if the
1331 lkb belongs to a remote node. */
1333 static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
1336 if (is_master_copy(lkb))
1339 queue_cast(r, lkb, 0);
1342 /* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
1343 change the granted/requested modes. We're munging things accordingly in
1345 CONVDEADLK: our grmode may have been forced down to NL to resolve a
1347 ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
1348 compatible with other granted locks */
1350 static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
1352 if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
1353 log_print("munge_demoted %x invalid reply type %d",
1354 lkb->lkb_id, ms->m_type);
1358 if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
1359 log_print("munge_demoted %x invalid modes gr %d rq %d",
1360 lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
1364 lkb->lkb_grmode = DLM_LOCK_NL;
1367 static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
1369 if (ms->m_type != DLM_MSG_REQUEST_REPLY &&
1370 ms->m_type != DLM_MSG_GRANT) {
1371 log_print("munge_altmode %x invalid reply type %d",
1372 lkb->lkb_id, ms->m_type);
1376 if (lkb->lkb_exflags & DLM_LKF_ALTPR)
1377 lkb->lkb_rqmode = DLM_LOCK_PR;
1378 else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
1379 lkb->lkb_rqmode = DLM_LOCK_CW;
1381 log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
1386 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
1388 struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
1390 if (lkb->lkb_id == first->lkb_id)
1396 /* Check if the given lkb conflicts with another lkb on the queue. */
1398 static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb)
1400 struct dlm_lkb *this;
1402 list_for_each_entry(this, head, lkb_statequeue) {
1405 if (!modes_compat(this, lkb))
1412 * "A conversion deadlock arises with a pair of lock requests in the converting
1413 * queue for one resource. The granted mode of each lock blocks the requested
1414 * mode of the other lock."
1416 * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
1417 * convert queue from being granted, then deadlk/demote lkb.
1420 * Granted Queue: empty
1421 * Convert Queue: NL->EX (first lock)
1422 * PR->EX (second lock)
1424 * The first lock can't be granted because of the granted mode of the second
1425 * lock and the second lock can't be granted because it's not first in the
1426 * list. We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
1427 * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
1428 * flag set and return DEMOTED in the lksb flags.
1430 * Originally, this function detected conv-deadlk in a more limited scope:
1431 * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
1432 * - if lkb1 was the first entry in the queue (not just earlier), and was
1433 * blocked by the granted mode of lkb2, and there was nothing on the
1434 * granted queue preventing lkb1 from being granted immediately, i.e.
1435 * lkb2 was the only thing preventing lkb1 from being granted.
1437 * That second condition meant we'd only say there was conv-deadlk if
1438 * resolving it (by demotion) would lead to the first lock on the convert
1439 * queue being granted right away. It allowed conversion deadlocks to exist
1440 * between locks on the convert queue while they couldn't be granted anyway.
1442 * Now, we detect and take action on conversion deadlocks immediately when
1443 * they're created, even if they may not be immediately consequential. If
1444 * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
1445 * mode that would prevent lkb1's conversion from being granted, we do a
1446 * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
1447 * I think this means that the lkb_is_ahead condition below should always
1448 * be zero, i.e. there will never be conv-deadlk between two locks that are
1449 * both already on the convert queue.
1452 static int conversion_deadlock_detect(struct dlm_rsb *r, struct dlm_lkb *lkb2)
1454 struct dlm_lkb *lkb1;
1455 int lkb_is_ahead = 0;
1457 list_for_each_entry(lkb1, &r->res_convertqueue, lkb_statequeue) {
1463 if (!lkb_is_ahead) {
1464 if (!modes_compat(lkb2, lkb1))
1467 if (!modes_compat(lkb2, lkb1) &&
1468 !modes_compat(lkb1, lkb2))
1476 * Return 1 if the lock can be granted, 0 otherwise.
1477 * Also detect and resolve conversion deadlocks.
1479 * lkb is the lock to be granted
1481 * now is 1 if the function is being called in the context of the
1482 * immediate request, it is 0 if called later, after the lock has been
1485 * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
1488 static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now)
1490 int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV);
1493 * 6-10: Version 5.4 introduced an option to address the phenomenon of
1494 * a new request for a NL mode lock being blocked.
1496 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
1497 * request, then it would be granted. In essence, the use of this flag
1498 * tells the Lock Manager to expedite theis request by not considering
1499 * what may be in the CONVERTING or WAITING queues... As of this
1500 * writing, the EXPEDITE flag can be used only with new requests for NL
1501 * mode locks. This flag is not valid for conversion requests.
1503 * A shortcut. Earlier checks return an error if EXPEDITE is used in a
1504 * conversion or used with a non-NL requested mode. We also know an
1505 * EXPEDITE request is always granted immediately, so now must always
1506 * be 1. The full condition to grant an expedite request: (now &&
1507 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
1508 * therefore be shortened to just checking the flag.
1511 if (lkb->lkb_exflags & DLM_LKF_EXPEDITE)
1515 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
1516 * added to the remaining conditions.
1519 if (queue_conflict(&r->res_grantqueue, lkb))
1523 * 6-3: By default, a conversion request is immediately granted if the
1524 * requested mode is compatible with the modes of all other granted
1528 if (queue_conflict(&r->res_convertqueue, lkb))
1532 * 6-5: But the default algorithm for deciding whether to grant or
1533 * queue conversion requests does not by itself guarantee that such
1534 * requests are serviced on a "first come first serve" basis. This, in
1535 * turn, can lead to a phenomenon known as "indefinate postponement".
1537 * 6-7: This issue is dealt with by using the optional QUECVT flag with
1538 * the system service employed to request a lock conversion. This flag
1539 * forces certain conversion requests to be queued, even if they are
1540 * compatible with the granted modes of other locks on the same
1541 * resource. Thus, the use of this flag results in conversion requests
1542 * being ordered on a "first come first servce" basis.
1544 * DCT: This condition is all about new conversions being able to occur
1545 * "in place" while the lock remains on the granted queue (assuming
1546 * nothing else conflicts.) IOW if QUECVT isn't set, a conversion
1547 * doesn't _have_ to go onto the convert queue where it's processed in
1548 * order. The "now" variable is necessary to distinguish converts
1549 * being received and processed for the first time now, because once a
1550 * convert is moved to the conversion queue the condition below applies
1551 * requiring fifo granting.
1554 if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT))
1558 * The NOORDER flag is set to avoid the standard vms rules on grant
1562 if (lkb->lkb_exflags & DLM_LKF_NOORDER)
1566 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
1567 * granted until all other conversion requests ahead of it are granted
1571 if (!now && conv && first_in_list(lkb, &r->res_convertqueue))
1575 * 6-4: By default, a new request is immediately granted only if all
1576 * three of the following conditions are satisfied when the request is
1578 * - The queue of ungranted conversion requests for the resource is
1580 * - The queue of ungranted new requests for the resource is empty.
1581 * - The mode of the new request is compatible with the most
1582 * restrictive mode of all granted locks on the resource.
1585 if (now && !conv && list_empty(&r->res_convertqueue) &&
1586 list_empty(&r->res_waitqueue))
1590 * 6-4: Once a lock request is in the queue of ungranted new requests,
1591 * it cannot be granted until the queue of ungranted conversion
1592 * requests is empty, all ungranted new requests ahead of it are
1593 * granted and/or canceled, and it is compatible with the granted mode
1594 * of the most restrictive lock granted on the resource.
1597 if (!now && !conv && list_empty(&r->res_convertqueue) &&
1598 first_in_list(lkb, &r->res_waitqueue))
1604 static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
1608 int8_t alt = 0, rqmode = lkb->lkb_rqmode;
1609 int8_t is_convert = (lkb->lkb_grmode != DLM_LOCK_IV);
1614 rv = _can_be_granted(r, lkb, now);
1619 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
1620 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
1621 * cancels one of the locks.
1624 if (is_convert && can_be_queued(lkb) &&
1625 conversion_deadlock_detect(r, lkb)) {
1626 if (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) {
1627 lkb->lkb_grmode = DLM_LOCK_NL;
1628 lkb->lkb_sbflags |= DLM_SBF_DEMOTED;
1629 } else if (!(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1633 log_print("can_be_granted deadlock %x now %d",
1642 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
1643 * to grant a request in a mode other than the normal rqmode. It's a
1644 * simple way to provide a big optimization to applications that can
1648 if (rqmode != DLM_LOCK_PR && (lkb->lkb_exflags & DLM_LKF_ALTPR))
1650 else if (rqmode != DLM_LOCK_CW && (lkb->lkb_exflags & DLM_LKF_ALTCW))
1654 lkb->lkb_rqmode = alt;
1655 rv = _can_be_granted(r, lkb, now);
1657 lkb->lkb_sbflags |= DLM_SBF_ALTMODE;
1659 lkb->lkb_rqmode = rqmode;
1665 /* FIXME: I don't think that can_be_granted() can/will demote or find deadlock
1666 for locks pending on the convert list. Once verified (watch for these
1667 log_prints), we should be able to just call _can_be_granted() and not
1668 bother with the demote/deadlk cases here (and there's no easy way to deal
1669 with a deadlk here, we'd have to generate something like grant_lock with
1670 the deadlk error.) */
1672 /* Returns the highest requested mode of all blocked conversions; sets
1673 cw if there's a blocked conversion to DLM_LOCK_CW. */
1675 static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw)
1677 struct dlm_lkb *lkb, *s;
1678 int hi, demoted, quit, grant_restart, demote_restart;
1687 list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) {
1688 demoted = is_demoted(lkb);
1691 if (can_be_granted(r, lkb, 0, &deadlk)) {
1692 grant_lock_pending(r, lkb);
1697 if (!demoted && is_demoted(lkb)) {
1698 log_print("WARN: pending demoted %x node %d %s",
1699 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1705 log_print("WARN: pending deadlock %x node %d %s",
1706 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1711 hi = max_t(int, lkb->lkb_rqmode, hi);
1713 if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
1719 if (demote_restart && !quit) {
1724 return max_t(int, high, hi);
1727 static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw)
1729 struct dlm_lkb *lkb, *s;
1731 list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
1732 if (can_be_granted(r, lkb, 0, NULL))
1733 grant_lock_pending(r, lkb);
1735 high = max_t(int, lkb->lkb_rqmode, high);
1736 if (lkb->lkb_rqmode == DLM_LOCK_CW)
1744 /* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
1745 on either the convert or waiting queue.
1746 high is the largest rqmode of all locks blocked on the convert or
1749 static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
1751 if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
1752 if (gr->lkb_highbast < DLM_LOCK_EX)
1757 if (gr->lkb_highbast < high &&
1758 !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
1763 static void grant_pending_locks(struct dlm_rsb *r)
1765 struct dlm_lkb *lkb, *s;
1766 int high = DLM_LOCK_IV;
1769 DLM_ASSERT(is_master(r), dlm_dump_rsb(r););
1771 high = grant_pending_convert(r, high, &cw);
1772 high = grant_pending_wait(r, high, &cw);
1774 if (high == DLM_LOCK_IV)
1778 * If there are locks left on the wait/convert queue then send blocking
1779 * ASTs to granted locks based on the largest requested mode (high)
1783 list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
1784 if (lkb->lkb_bastaddr && lock_requires_bast(lkb, high, cw)) {
1785 if (cw && high == DLM_LOCK_PR)
1786 queue_bast(r, lkb, DLM_LOCK_CW);
1788 queue_bast(r, lkb, high);
1789 lkb->lkb_highbast = high;
1794 static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
1796 if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
1797 (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
1798 if (gr->lkb_highbast < DLM_LOCK_EX)
1803 if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
1808 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
1809 struct dlm_lkb *lkb)
1813 list_for_each_entry(gr, head, lkb_statequeue) {
1814 if (gr->lkb_bastaddr && modes_require_bast(gr, lkb)) {
1815 queue_bast(r, gr, lkb->lkb_rqmode);
1816 gr->lkb_highbast = lkb->lkb_rqmode;
1821 static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb)
1823 send_bast_queue(r, &r->res_grantqueue, lkb);
1826 static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb)
1828 send_bast_queue(r, &r->res_grantqueue, lkb);
1829 send_bast_queue(r, &r->res_convertqueue, lkb);
1832 /* set_master(r, lkb) -- set the master nodeid of a resource
1834 The purpose of this function is to set the nodeid field in the given
1835 lkb using the nodeid field in the given rsb. If the rsb's nodeid is
1836 known, it can just be copied to the lkb and the function will return
1837 0. If the rsb's nodeid is _not_ known, it needs to be looked up
1838 before it can be copied to the lkb.
1840 When the rsb nodeid is being looked up remotely, the initial lkb
1841 causing the lookup is kept on the ls_waiters list waiting for the
1842 lookup reply. Other lkb's waiting for the same rsb lookup are kept
1843 on the rsb's res_lookup list until the master is verified.
1846 0: nodeid is set in rsb/lkb and the caller should go ahead and use it
1847 1: the rsb master is not available and the lkb has been placed on
1851 static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb)
1853 struct dlm_ls *ls = r->res_ls;
1854 int error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid();
1856 if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) {
1857 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
1858 r->res_first_lkid = lkb->lkb_id;
1859 lkb->lkb_nodeid = r->res_nodeid;
1863 if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) {
1864 list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup);
1868 if (r->res_nodeid == 0) {
1869 lkb->lkb_nodeid = 0;
1873 if (r->res_nodeid > 0) {
1874 lkb->lkb_nodeid = r->res_nodeid;
1878 DLM_ASSERT(r->res_nodeid == -1, dlm_dump_rsb(r););
1880 dir_nodeid = dlm_dir_nodeid(r);
1882 if (dir_nodeid != our_nodeid) {
1883 r->res_first_lkid = lkb->lkb_id;
1884 send_lookup(r, lkb);
1889 /* It's possible for dlm_scand to remove an old rsb for
1890 this same resource from the toss list, us to create
1891 a new one, look up the master locally, and find it
1892 already exists just before dlm_scand does the
1893 dir_remove() on the previous rsb. */
1895 error = dlm_dir_lookup(ls, our_nodeid, r->res_name,
1896 r->res_length, &ret_nodeid);
1899 log_debug(ls, "dir_lookup error %d %s", error, r->res_name);
1903 if (ret_nodeid == our_nodeid) {
1904 r->res_first_lkid = 0;
1906 lkb->lkb_nodeid = 0;
1908 r->res_first_lkid = lkb->lkb_id;
1909 r->res_nodeid = ret_nodeid;
1910 lkb->lkb_nodeid = ret_nodeid;
1915 static void process_lookup_list(struct dlm_rsb *r)
1917 struct dlm_lkb *lkb, *safe;
1919 list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
1920 list_del_init(&lkb->lkb_rsb_lookup);
1921 _request_lock(r, lkb);
1926 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
1928 static void confirm_master(struct dlm_rsb *r, int error)
1930 struct dlm_lkb *lkb;
1932 if (!r->res_first_lkid)
1938 r->res_first_lkid = 0;
1939 process_lookup_list(r);
1945 /* the remote request failed and won't be retried (it was
1946 a NOQUEUE, or has been canceled/unlocked); make a waiting
1947 lkb the first_lkid */
1949 r->res_first_lkid = 0;
1951 if (!list_empty(&r->res_lookup)) {
1952 lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
1954 list_del_init(&lkb->lkb_rsb_lookup);
1955 r->res_first_lkid = lkb->lkb_id;
1956 _request_lock(r, lkb);
1962 log_error(r->res_ls, "confirm_master unknown error %d", error);
1966 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
1967 int namelen, unsigned long timeout_cs, void *ast,
1968 void *astarg, void *bast, struct dlm_args *args)
1972 /* check for invalid arg usage */
1974 if (mode < 0 || mode > DLM_LOCK_EX)
1977 if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
1980 if (flags & DLM_LKF_CANCEL)
1983 if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
1986 if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
1989 if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
1992 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
1995 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
1998 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2001 if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2007 if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2010 if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2013 /* these args will be copied to the lkb in validate_lock_args,
2014 it cannot be done now because when converting locks, fields in
2015 an active lkb cannot be modified before locking the rsb */
2017 args->flags = flags;
2018 args->astaddr = ast;
2019 args->astparam = (long) astarg;
2020 args->bastaddr = bast;
2021 args->timeout = timeout_cs;
2029 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2031 if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2032 DLM_LKF_FORCEUNLOCK))
2035 if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2038 args->flags = flags;
2039 args->astparam = (long) astarg;
2043 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2044 struct dlm_args *args)
2048 if (args->flags & DLM_LKF_CONVERT) {
2049 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2052 if (args->flags & DLM_LKF_QUECVT &&
2053 !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2057 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2060 if (lkb->lkb_wait_type)
2063 if (is_overlap(lkb))
2067 lkb->lkb_exflags = args->flags;
2068 lkb->lkb_sbflags = 0;
2069 lkb->lkb_astaddr = args->astaddr;
2070 lkb->lkb_astparam = args->astparam;
2071 lkb->lkb_bastaddr = args->bastaddr;
2072 lkb->lkb_rqmode = args->mode;
2073 lkb->lkb_lksb = args->lksb;
2074 lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2075 lkb->lkb_ownpid = (int) current->pid;
2076 lkb->lkb_timeout_cs = args->timeout;
2082 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2085 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2086 because there may be a lookup in progress and it's valid to do
2087 cancel/unlockf on it */
2089 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2091 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
2094 if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2095 log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2100 /* an lkb may still exist even though the lock is EOL'ed due to a
2101 cancel, unlock or failed noqueue request; an app can't use these
2102 locks; return same error as if the lkid had not been found at all */
2104 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2105 log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2110 /* an lkb may be waiting for an rsb lookup to complete where the
2111 lookup was initiated by another lock */
2113 if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
2114 if (!list_empty(&lkb->lkb_rsb_lookup)) {
2115 log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2116 list_del_init(&lkb->lkb_rsb_lookup);
2117 queue_cast(lkb->lkb_resource, lkb,
2118 args->flags & DLM_LKF_CANCEL ?
2119 -DLM_ECANCEL : -DLM_EUNLOCK);
2120 unhold_lkb(lkb); /* undoes create_lkb() */
2126 /* cancel not allowed with another cancel/unlock in progress */
2128 if (args->flags & DLM_LKF_CANCEL) {
2129 if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2132 if (is_overlap(lkb))
2135 /* don't let scand try to do a cancel */
2138 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2139 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2144 switch (lkb->lkb_wait_type) {
2145 case DLM_MSG_LOOKUP:
2146 case DLM_MSG_REQUEST:
2147 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2150 case DLM_MSG_UNLOCK:
2151 case DLM_MSG_CANCEL:
2154 /* add_to_waiters() will set OVERLAP_CANCEL */
2158 /* do we need to allow a force-unlock if there's a normal unlock
2159 already in progress? in what conditions could the normal unlock
2160 fail such that we'd want to send a force-unlock to be sure? */
2162 if (args->flags & DLM_LKF_FORCEUNLOCK) {
2163 if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2166 if (is_overlap_unlock(lkb))
2169 /* don't let scand try to do a cancel */
2172 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2173 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2178 switch (lkb->lkb_wait_type) {
2179 case DLM_MSG_LOOKUP:
2180 case DLM_MSG_REQUEST:
2181 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2184 case DLM_MSG_UNLOCK:
2187 /* add_to_waiters() will set OVERLAP_UNLOCK */
2191 /* normal unlock not allowed if there's any op in progress */
2193 if (lkb->lkb_wait_type || lkb->lkb_wait_count)
2197 /* an overlapping op shouldn't blow away exflags from other op */
2198 lkb->lkb_exflags |= args->flags;
2199 lkb->lkb_sbflags = 0;
2200 lkb->lkb_astparam = args->astparam;
2204 log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
2205 lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
2206 args->flags, lkb->lkb_wait_type,
2207 lkb->lkb_resource->res_name);
2212 * Four stage 4 varieties:
2213 * do_request(), do_convert(), do_unlock(), do_cancel()
2214 * These are called on the master node for the given lock and
2215 * from the central locking logic.
2218 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2222 if (can_be_granted(r, lkb, 1, NULL)) {
2224 queue_cast(r, lkb, 0);
2228 if (can_be_queued(lkb)) {
2229 error = -EINPROGRESS;
2230 add_lkb(r, lkb, DLM_LKSTS_WAITING);
2231 send_blocking_asts(r, lkb);
2237 if (force_blocking_asts(lkb))
2238 send_blocking_asts_all(r, lkb);
2239 queue_cast(r, lkb, -EAGAIN);
2245 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2250 /* changing an existing lock may allow others to be granted */
2252 if (can_be_granted(r, lkb, 1, &deadlk)) {
2254 queue_cast(r, lkb, 0);
2255 grant_pending_locks(r);
2259 /* can_be_granted() detected that this lock would block in a conversion
2260 deadlock, so we leave it on the granted queue and return EDEADLK in
2261 the ast for the convert. */
2264 /* it's left on the granted queue */
2265 log_debug(r->res_ls, "deadlock %x node %d sts%d g%d r%d %s",
2266 lkb->lkb_id, lkb->lkb_nodeid, lkb->lkb_status,
2267 lkb->lkb_grmode, lkb->lkb_rqmode, r->res_name);
2268 revert_lock(r, lkb);
2269 queue_cast(r, lkb, -EDEADLK);
2274 /* is_demoted() means the can_be_granted() above set the grmode
2275 to NL, and left us on the granted queue. This auto-demotion
2276 (due to CONVDEADLK) might mean other locks, and/or this lock, are
2277 now grantable. We have to try to grant other converting locks
2278 before we try again to grant this one. */
2280 if (is_demoted(lkb)) {
2281 grant_pending_convert(r, DLM_LOCK_IV, NULL);
2282 if (_can_be_granted(r, lkb, 1)) {
2284 queue_cast(r, lkb, 0);
2285 grant_pending_locks(r);
2288 /* else fall through and move to convert queue */
2291 if (can_be_queued(lkb)) {
2292 error = -EINPROGRESS;
2294 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2295 send_blocking_asts(r, lkb);
2301 if (force_blocking_asts(lkb))
2302 send_blocking_asts_all(r, lkb);
2303 queue_cast(r, lkb, -EAGAIN);
2309 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2311 remove_lock(r, lkb);
2312 queue_cast(r, lkb, -DLM_EUNLOCK);
2313 grant_pending_locks(r);
2314 return -DLM_EUNLOCK;
2317 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
2319 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2323 error = revert_lock(r, lkb);
2325 queue_cast(r, lkb, -DLM_ECANCEL);
2326 grant_pending_locks(r);
2327 return -DLM_ECANCEL;
2333 * Four stage 3 varieties:
2334 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2337 /* add a new lkb to a possibly new rsb, called by requesting process */
2339 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2343 /* set_master: sets lkb nodeid from r */
2345 error = set_master(r, lkb);
2354 /* receive_request() calls do_request() on remote node */
2355 error = send_request(r, lkb);
2357 error = do_request(r, lkb);
2362 /* change some property of an existing lkb, e.g. mode */
2364 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2369 /* receive_convert() calls do_convert() on remote node */
2370 error = send_convert(r, lkb);
2372 error = do_convert(r, lkb);
2377 /* remove an existing lkb from the granted queue */
2379 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2384 /* receive_unlock() calls do_unlock() on remote node */
2385 error = send_unlock(r, lkb);
2387 error = do_unlock(r, lkb);
2392 /* remove an existing lkb from the convert or wait queue */
2394 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2399 /* receive_cancel() calls do_cancel() on remote node */
2400 error = send_cancel(r, lkb);
2402 error = do_cancel(r, lkb);
2408 * Four stage 2 varieties:
2409 * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2412 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
2413 int len, struct dlm_args *args)
2418 error = validate_lock_args(ls, lkb, args);
2422 error = find_rsb(ls, name, len, R_CREATE, &r);
2429 lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
2431 error = _request_lock(r, lkb);
2440 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2441 struct dlm_args *args)
2446 r = lkb->lkb_resource;
2451 error = validate_lock_args(ls, lkb, args);
2455 error = _convert_lock(r, lkb);
2462 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2463 struct dlm_args *args)
2468 r = lkb->lkb_resource;
2473 error = validate_unlock_args(lkb, args);
2477 error = _unlock_lock(r, lkb);
2484 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2485 struct dlm_args *args)
2490 r = lkb->lkb_resource;
2495 error = validate_unlock_args(lkb, args);
2499 error = _cancel_lock(r, lkb);
2507 * Two stage 1 varieties: dlm_lock() and dlm_unlock()
2510 int dlm_lock(dlm_lockspace_t *lockspace,
2512 struct dlm_lksb *lksb,
2515 unsigned int namelen,
2516 uint32_t parent_lkid,
2517 void (*ast) (void *astarg),
2519 void (*bast) (void *astarg, int mode))
2522 struct dlm_lkb *lkb;
2523 struct dlm_args args;
2524 int error, convert = flags & DLM_LKF_CONVERT;
2526 ls = dlm_find_lockspace_local(lockspace);
2530 dlm_lock_recovery(ls);
2533 error = find_lkb(ls, lksb->sb_lkid, &lkb);
2535 error = create_lkb(ls, &lkb);
2540 error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
2541 astarg, bast, &args);
2546 error = convert_lock(ls, lkb, &args);
2548 error = request_lock(ls, lkb, name, namelen, &args);
2550 if (error == -EINPROGRESS)
2553 if (convert || error)
2555 if (error == -EAGAIN || error == -EDEADLK)
2558 dlm_unlock_recovery(ls);
2559 dlm_put_lockspace(ls);
2563 int dlm_unlock(dlm_lockspace_t *lockspace,
2566 struct dlm_lksb *lksb,
2570 struct dlm_lkb *lkb;
2571 struct dlm_args args;
2574 ls = dlm_find_lockspace_local(lockspace);
2578 dlm_lock_recovery(ls);
2580 error = find_lkb(ls, lkid, &lkb);
2584 error = set_unlock_args(flags, astarg, &args);
2588 if (flags & DLM_LKF_CANCEL)
2589 error = cancel_lock(ls, lkb, &args);
2591 error = unlock_lock(ls, lkb, &args);
2593 if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
2595 if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
2600 dlm_unlock_recovery(ls);
2601 dlm_put_lockspace(ls);
2606 * send/receive routines for remote operations and replies
2610 * send_request receive_request
2611 * send_convert receive_convert
2612 * send_unlock receive_unlock
2613 * send_cancel receive_cancel
2614 * send_grant receive_grant
2615 * send_bast receive_bast
2616 * send_lookup receive_lookup
2617 * send_remove receive_remove
2620 * receive_request_reply send_request_reply
2621 * receive_convert_reply send_convert_reply
2622 * receive_unlock_reply send_unlock_reply
2623 * receive_cancel_reply send_cancel_reply
2624 * receive_lookup_reply send_lookup_reply
2627 static int _create_message(struct dlm_ls *ls, int mb_len,
2628 int to_nodeid, int mstype,
2629 struct dlm_message **ms_ret,
2630 struct dlm_mhandle **mh_ret)
2632 struct dlm_message *ms;
2633 struct dlm_mhandle *mh;
2636 /* get_buffer gives us a message handle (mh) that we need to
2637 pass into lowcomms_commit and a message buffer (mb) that we
2638 write our data into */
2640 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
2644 memset(mb, 0, mb_len);
2646 ms = (struct dlm_message *) mb;
2648 ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
2649 ms->m_header.h_lockspace = ls->ls_global_id;
2650 ms->m_header.h_nodeid = dlm_our_nodeid();
2651 ms->m_header.h_length = mb_len;
2652 ms->m_header.h_cmd = DLM_MSG;
2654 ms->m_type = mstype;
2661 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
2662 int to_nodeid, int mstype,
2663 struct dlm_message **ms_ret,
2664 struct dlm_mhandle **mh_ret)
2666 int mb_len = sizeof(struct dlm_message);
2669 case DLM_MSG_REQUEST:
2670 case DLM_MSG_LOOKUP:
2671 case DLM_MSG_REMOVE:
2672 mb_len += r->res_length;
2674 case DLM_MSG_CONVERT:
2675 case DLM_MSG_UNLOCK:
2676 case DLM_MSG_REQUEST_REPLY:
2677 case DLM_MSG_CONVERT_REPLY:
2679 if (lkb && lkb->lkb_lvbptr)
2680 mb_len += r->res_ls->ls_lvblen;
2684 return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
2688 /* further lowcomms enhancements or alternate implementations may make
2689 the return value from this function useful at some point */
2691 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
2693 dlm_message_out(ms);
2694 dlm_lowcomms_commit_buffer(mh);
2698 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
2699 struct dlm_message *ms)
2701 ms->m_nodeid = lkb->lkb_nodeid;
2702 ms->m_pid = lkb->lkb_ownpid;
2703 ms->m_lkid = lkb->lkb_id;
2704 ms->m_remid = lkb->lkb_remid;
2705 ms->m_exflags = lkb->lkb_exflags;
2706 ms->m_sbflags = lkb->lkb_sbflags;
2707 ms->m_flags = lkb->lkb_flags;
2708 ms->m_lvbseq = lkb->lkb_lvbseq;
2709 ms->m_status = lkb->lkb_status;
2710 ms->m_grmode = lkb->lkb_grmode;
2711 ms->m_rqmode = lkb->lkb_rqmode;
2712 ms->m_hash = r->res_hash;
2714 /* m_result and m_bastmode are set from function args,
2715 not from lkb fields */
2717 if (lkb->lkb_bastaddr)
2718 ms->m_asts |= AST_BAST;
2719 if (lkb->lkb_astaddr)
2720 ms->m_asts |= AST_COMP;
2722 /* compare with switch in create_message; send_remove() doesn't
2725 switch (ms->m_type) {
2726 case DLM_MSG_REQUEST:
2727 case DLM_MSG_LOOKUP:
2728 memcpy(ms->m_extra, r->res_name, r->res_length);
2730 case DLM_MSG_CONVERT:
2731 case DLM_MSG_UNLOCK:
2732 case DLM_MSG_REQUEST_REPLY:
2733 case DLM_MSG_CONVERT_REPLY:
2735 if (!lkb->lkb_lvbptr)
2737 memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
2742 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
2744 struct dlm_message *ms;
2745 struct dlm_mhandle *mh;
2746 int to_nodeid, error;
2748 error = add_to_waiters(lkb, mstype);
2752 to_nodeid = r->res_nodeid;
2754 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2758 send_args(r, lkb, ms);
2760 error = send_message(mh, ms);
2766 remove_from_waiters(lkb, msg_reply_type(mstype));
2770 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2772 return send_common(r, lkb, DLM_MSG_REQUEST);
2775 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2779 error = send_common(r, lkb, DLM_MSG_CONVERT);
2781 /* down conversions go without a reply from the master */
2782 if (!error && down_conversion(lkb)) {
2783 remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
2784 r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
2785 r->res_ls->ls_stub_ms.m_result = 0;
2786 r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
2787 __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
2793 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
2794 MASTER_UNCERTAIN to force the next request on the rsb to confirm
2795 that the master is still correct. */
2797 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2799 return send_common(r, lkb, DLM_MSG_UNLOCK);
2802 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2804 return send_common(r, lkb, DLM_MSG_CANCEL);
2807 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
2809 struct dlm_message *ms;
2810 struct dlm_mhandle *mh;
2811 int to_nodeid, error;
2813 to_nodeid = lkb->lkb_nodeid;
2815 error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
2819 send_args(r, lkb, ms);
2823 error = send_message(mh, ms);
2828 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
2830 struct dlm_message *ms;
2831 struct dlm_mhandle *mh;
2832 int to_nodeid, error;
2834 to_nodeid = lkb->lkb_nodeid;
2836 error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
2840 send_args(r, lkb, ms);
2842 ms->m_bastmode = mode;
2844 error = send_message(mh, ms);
2849 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
2851 struct dlm_message *ms;
2852 struct dlm_mhandle *mh;
2853 int to_nodeid, error;
2855 error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
2859 to_nodeid = dlm_dir_nodeid(r);
2861 error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
2865 send_args(r, lkb, ms);
2867 error = send_message(mh, ms);
2873 remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
2877 static int send_remove(struct dlm_rsb *r)
2879 struct dlm_message *ms;
2880 struct dlm_mhandle *mh;
2881 int to_nodeid, error;
2883 to_nodeid = dlm_dir_nodeid(r);
2885 error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
2889 memcpy(ms->m_extra, r->res_name, r->res_length);
2890 ms->m_hash = r->res_hash;
2892 error = send_message(mh, ms);
2897 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
2900 struct dlm_message *ms;
2901 struct dlm_mhandle *mh;
2902 int to_nodeid, error;
2904 to_nodeid = lkb->lkb_nodeid;
2906 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2910 send_args(r, lkb, ms);
2914 error = send_message(mh, ms);
2919 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2921 return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
2924 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2926 return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
2929 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2931 return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
2934 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2936 return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
2939 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
2940 int ret_nodeid, int rv)
2942 struct dlm_rsb *r = &ls->ls_stub_rsb;
2943 struct dlm_message *ms;
2944 struct dlm_mhandle *mh;
2945 int error, nodeid = ms_in->m_header.h_nodeid;
2947 error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
2951 ms->m_lkid = ms_in->m_lkid;
2953 ms->m_nodeid = ret_nodeid;
2955 error = send_message(mh, ms);
2960 /* which args we save from a received message depends heavily on the type
2961 of message, unlike the send side where we can safely send everything about
2962 the lkb for any type of message */
2964 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
2966 lkb->lkb_exflags = ms->m_exflags;
2967 lkb->lkb_sbflags = ms->m_sbflags;
2968 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2969 (ms->m_flags & 0x0000FFFF);
2972 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
2974 lkb->lkb_sbflags = ms->m_sbflags;
2975 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2976 (ms->m_flags & 0x0000FFFF);
2979 static int receive_extralen(struct dlm_message *ms)
2981 return (ms->m_header.h_length - sizeof(struct dlm_message));
2984 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
2985 struct dlm_message *ms)
2989 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
2990 if (!lkb->lkb_lvbptr)
2991 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
2992 if (!lkb->lkb_lvbptr)
2994 len = receive_extralen(ms);
2995 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
3000 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3001 struct dlm_message *ms)
3003 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3004 lkb->lkb_ownpid = ms->m_pid;
3005 lkb->lkb_remid = ms->m_lkid;
3006 lkb->lkb_grmode = DLM_LOCK_IV;
3007 lkb->lkb_rqmode = ms->m_rqmode;
3008 lkb->lkb_bastaddr = (void *) (long) (ms->m_asts & AST_BAST);
3009 lkb->lkb_astaddr = (void *) (long) (ms->m_asts & AST_COMP);
3011 DLM_ASSERT(is_master_copy(lkb), dlm_print_lkb(lkb););
3013 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3014 /* lkb was just created so there won't be an lvb yet */
3015 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3016 if (!lkb->lkb_lvbptr)
3023 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3024 struct dlm_message *ms)
3026 if (lkb->lkb_nodeid != ms->m_header.h_nodeid) {
3027 log_error(ls, "convert_args nodeid %d %d lkid %x %x",
3028 lkb->lkb_nodeid, ms->m_header.h_nodeid,
3029 lkb->lkb_id, lkb->lkb_remid);
3033 if (!is_master_copy(lkb))
3036 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3039 if (receive_lvb(ls, lkb, ms))
3042 lkb->lkb_rqmode = ms->m_rqmode;
3043 lkb->lkb_lvbseq = ms->m_lvbseq;
3048 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3049 struct dlm_message *ms)
3051 if (!is_master_copy(lkb))
3053 if (receive_lvb(ls, lkb, ms))
3058 /* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3059 uses to send a reply and that the remote end uses to process the reply. */
3061 static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3063 struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3064 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3065 lkb->lkb_remid = ms->m_lkid;
3068 static void receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3070 struct dlm_lkb *lkb;
3074 error = create_lkb(ls, &lkb);
3078 receive_flags(lkb, ms);
3079 lkb->lkb_flags |= DLM_IFL_MSTCPY;
3080 error = receive_request_args(ls, lkb, ms);
3086 namelen = receive_extralen(ms);
3088 error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r);
3097 error = do_request(r, lkb);
3098 send_request_reply(r, lkb, error);
3103 if (error == -EINPROGRESS)
3110 setup_stub_lkb(ls, ms);
3111 send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3114 static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3116 struct dlm_lkb *lkb;
3118 int error, reply = 1;
3120 error = find_lkb(ls, ms->m_remid, &lkb);
3124 r = lkb->lkb_resource;
3129 receive_flags(lkb, ms);
3130 error = receive_convert_args(ls, lkb, ms);
3133 reply = !down_conversion(lkb);
3135 error = do_convert(r, lkb);
3138 send_convert_reply(r, lkb, error);
3146 setup_stub_lkb(ls, ms);
3147 send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3150 static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3152 struct dlm_lkb *lkb;
3156 error = find_lkb(ls, ms->m_remid, &lkb);
3160 r = lkb->lkb_resource;
3165 receive_flags(lkb, ms);
3166 error = receive_unlock_args(ls, lkb, ms);
3170 error = do_unlock(r, lkb);
3172 send_unlock_reply(r, lkb, error);
3180 setup_stub_lkb(ls, ms);
3181 send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3184 static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
3186 struct dlm_lkb *lkb;
3190 error = find_lkb(ls, ms->m_remid, &lkb);
3194 receive_flags(lkb, ms);
3196 r = lkb->lkb_resource;
3201 error = do_cancel(r, lkb);
3202 send_cancel_reply(r, lkb, error);
3210 setup_stub_lkb(ls, ms);
3211 send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3214 static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
3216 struct dlm_lkb *lkb;
3220 error = find_lkb(ls, ms->m_remid, &lkb);
3222 log_error(ls, "receive_grant no lkb");
3225 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
3227 r = lkb->lkb_resource;
3232 receive_flags_reply(lkb, ms);
3233 if (is_altmode(lkb))
3234 munge_altmode(lkb, ms);
3235 grant_lock_pc(r, lkb, ms);
3236 queue_cast(r, lkb, 0);
3243 static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
3245 struct dlm_lkb *lkb;
3249 error = find_lkb(ls, ms->m_remid, &lkb);
3251 log_error(ls, "receive_bast no lkb");
3254 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
3256 r = lkb->lkb_resource;
3261 queue_bast(r, lkb, ms->m_bastmode);
3268 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
3270 int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid;
3272 from_nodeid = ms->m_header.h_nodeid;
3273 our_nodeid = dlm_our_nodeid();
3275 len = receive_extralen(ms);
3277 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3278 if (dir_nodeid != our_nodeid) {
3279 log_error(ls, "lookup dir_nodeid %d from %d",
3280 dir_nodeid, from_nodeid);
3286 error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid);
3288 /* Optimization: we're master so treat lookup as a request */
3289 if (!error && ret_nodeid == our_nodeid) {
3290 receive_request(ls, ms);
3294 send_lookup_reply(ls, ms, ret_nodeid, error);
3297 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
3299 int len, dir_nodeid, from_nodeid;
3301 from_nodeid = ms->m_header.h_nodeid;
3303 len = receive_extralen(ms);
3305 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3306 if (dir_nodeid != dlm_our_nodeid()) {
3307 log_error(ls, "remove dir entry dir_nodeid %d from %d",
3308 dir_nodeid, from_nodeid);
3312 dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
3315 static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
3317 do_purge(ls, ms->m_nodeid, ms->m_pid);
3320 static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
3322 struct dlm_lkb *lkb;
3324 int error, mstype, result;
3326 error = find_lkb(ls, ms->m_remid, &lkb);
3328 log_error(ls, "receive_request_reply no lkb");
3331 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
3333 r = lkb->lkb_resource;
3337 mstype = lkb->lkb_wait_type;
3338 error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
3342 /* Optimization: the dir node was also the master, so it took our
3343 lookup as a request and sent request reply instead of lookup reply */
3344 if (mstype == DLM_MSG_LOOKUP) {
3345 r->res_nodeid = ms->m_header.h_nodeid;
3346 lkb->lkb_nodeid = r->res_nodeid;
3349 /* this is the value returned from do_request() on the master */
3350 result = ms->m_result;
3354 /* request would block (be queued) on remote master */
3355 queue_cast(r, lkb, -EAGAIN);
3356 confirm_master(r, -EAGAIN);
3357 unhold_lkb(lkb); /* undoes create_lkb() */
3362 /* request was queued or granted on remote master */
3363 receive_flags_reply(lkb, ms);
3364 lkb->lkb_remid = ms->m_lkid;
3365 if (is_altmode(lkb))
3366 munge_altmode(lkb, ms);
3368 add_lkb(r, lkb, DLM_LKSTS_WAITING);
3371 grant_lock_pc(r, lkb, ms);
3372 queue_cast(r, lkb, 0);
3374 confirm_master(r, result);
3379 /* find_rsb failed to find rsb or rsb wasn't master */
3380 log_debug(ls, "receive_request_reply %x %x master diff %d %d",
3381 lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
3383 lkb->lkb_nodeid = -1;
3385 if (is_overlap(lkb)) {
3386 /* we'll ignore error in cancel/unlock reply */
3387 queue_cast_overlap(r, lkb);
3388 confirm_master(r, result);
3389 unhold_lkb(lkb); /* undoes create_lkb() */
3391 _request_lock(r, lkb);
3395 log_error(ls, "receive_request_reply %x error %d",
3396 lkb->lkb_id, result);
3399 if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
3400 log_debug(ls, "receive_request_reply %x result %d unlock",
3401 lkb->lkb_id, result);
3402 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3403 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3404 send_unlock(r, lkb);
3405 } else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
3406 log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
3407 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3408 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3409 send_cancel(r, lkb);
3411 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3412 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3420 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3421 struct dlm_message *ms)
3423 /* this is the value returned from do_convert() on the master */
3424 switch (ms->m_result) {
3426 /* convert would block (be queued) on remote master */
3427 queue_cast(r, lkb, -EAGAIN);
3431 receive_flags_reply(lkb, ms);
3432 revert_lock_pc(r, lkb);
3433 queue_cast(r, lkb, -EDEADLK);
3437 /* convert was queued on remote master */
3438 receive_flags_reply(lkb, ms);
3439 if (is_demoted(lkb))
3440 munge_demoted(lkb, ms);
3442 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3447 /* convert was granted on remote master */
3448 receive_flags_reply(lkb, ms);
3449 if (is_demoted(lkb))
3450 munge_demoted(lkb, ms);
3451 grant_lock_pc(r, lkb, ms);
3452 queue_cast(r, lkb, 0);
3456 log_error(r->res_ls, "receive_convert_reply %x error %d",
3457 lkb->lkb_id, ms->m_result);
3461 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3463 struct dlm_rsb *r = lkb->lkb_resource;
3469 /* stub reply can happen with waiters_mutex held */
3470 error = remove_from_waiters_ms(lkb, ms);
3474 __receive_convert_reply(r, lkb, ms);
3480 static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
3482 struct dlm_lkb *lkb;
3485 error = find_lkb(ls, ms->m_remid, &lkb);
3487 log_error(ls, "receive_convert_reply no lkb");
3490 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
3492 _receive_convert_reply(lkb, ms);
3496 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3498 struct dlm_rsb *r = lkb->lkb_resource;
3504 /* stub reply can happen with waiters_mutex held */
3505 error = remove_from_waiters_ms(lkb, ms);
3509 /* this is the value returned from do_unlock() on the master */
3511 switch (ms->m_result) {
3513 receive_flags_reply(lkb, ms);
3514 remove_lock_pc(r, lkb);
3515 queue_cast(r, lkb, -DLM_EUNLOCK);
3520 log_error(r->res_ls, "receive_unlock_reply %x error %d",
3521 lkb->lkb_id, ms->m_result);
3528 static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
3530 struct dlm_lkb *lkb;
3533 error = find_lkb(ls, ms->m_remid, &lkb);
3535 log_error(ls, "receive_unlock_reply no lkb");
3538 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
3540 _receive_unlock_reply(lkb, ms);
3544 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3546 struct dlm_rsb *r = lkb->lkb_resource;
3552 /* stub reply can happen with waiters_mutex held */
3553 error = remove_from_waiters_ms(lkb, ms);
3557 /* this is the value returned from do_cancel() on the master */
3559 switch (ms->m_result) {
3561 receive_flags_reply(lkb, ms);
3562 revert_lock_pc(r, lkb);
3563 queue_cast(r, lkb, -DLM_ECANCEL);
3568 log_error(r->res_ls, "receive_cancel_reply %x error %d",
3569 lkb->lkb_id, ms->m_result);
3576 static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
3578 struct dlm_lkb *lkb;
3581 error = find_lkb(ls, ms->m_remid, &lkb);
3583 log_error(ls, "receive_cancel_reply no lkb");
3586 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
3588 _receive_cancel_reply(lkb, ms);
3592 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
3594 struct dlm_lkb *lkb;
3596 int error, ret_nodeid;
3598 error = find_lkb(ls, ms->m_lkid, &lkb);
3600 log_error(ls, "receive_lookup_reply no lkb");
3604 /* ms->m_result is the value returned by dlm_dir_lookup on dir node
3605 FIXME: will a non-zero error ever be returned? */
3607 r = lkb->lkb_resource;
3611 error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3615 ret_nodeid = ms->m_nodeid;
3616 if (ret_nodeid == dlm_our_nodeid()) {
3619 r->res_first_lkid = 0;
3621 /* set_master() will copy res_nodeid to lkb_nodeid */
3622 r->res_nodeid = ret_nodeid;
3625 if (is_overlap(lkb)) {
3626 log_debug(ls, "receive_lookup_reply %x unlock %x",
3627 lkb->lkb_id, lkb->lkb_flags);
3628 queue_cast_overlap(r, lkb);
3629 unhold_lkb(lkb); /* undoes create_lkb() */
3633 _request_lock(r, lkb);
3637 process_lookup_list(r);
3644 static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
3646 switch (ms->m_type) {
3648 /* messages sent to a master node */
3650 case DLM_MSG_REQUEST:
3651 receive_request(ls, ms);
3654 case DLM_MSG_CONVERT:
3655 receive_convert(ls, ms);
3658 case DLM_MSG_UNLOCK:
3659 receive_unlock(ls, ms);
3662 case DLM_MSG_CANCEL:
3663 receive_cancel(ls, ms);
3666 /* messages sent from a master node (replies to above) */
3668 case DLM_MSG_REQUEST_REPLY:
3669 receive_request_reply(ls, ms);
3672 case DLM_MSG_CONVERT_REPLY:
3673 receive_convert_reply(ls, ms);
3676 case DLM_MSG_UNLOCK_REPLY:
3677 receive_unlock_reply(ls, ms);
3680 case DLM_MSG_CANCEL_REPLY:
3681 receive_cancel_reply(ls, ms);
3684 /* messages sent from a master node (only two types of async msg) */
3687 receive_grant(ls, ms);
3691 receive_bast(ls, ms);
3694 /* messages sent to a dir node */
3696 case DLM_MSG_LOOKUP:
3697 receive_lookup(ls, ms);
3700 case DLM_MSG_REMOVE:
3701 receive_remove(ls, ms);
3704 /* messages sent from a dir node (remove has no reply) */
3706 case DLM_MSG_LOOKUP_REPLY:
3707 receive_lookup_reply(ls, ms);
3710 /* other messages */
3713 receive_purge(ls, ms);
3717 log_error(ls, "unknown message type %d", ms->m_type);
3723 /* If the lockspace is in recovery mode (locking stopped), then normal
3724 messages are saved on the requestqueue for processing after recovery is
3725 done. When not in recovery mode, we wait for dlm_recoverd to drain saved
3726 messages off the requestqueue before we process new ones. This occurs right
3727 after recovery completes when we transition from saving all messages on
3728 requestqueue, to processing all the saved messages, to processing new
3729 messages as they arrive. */
3731 static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
3734 if (dlm_locking_stopped(ls)) {
3735 dlm_add_requestqueue(ls, nodeid, (struct dlm_header *) ms);
3737 dlm_wait_requestqueue(ls);
3738 _receive_message(ls, ms);
3742 /* This is called by dlm_recoverd to process messages that were saved on
3743 the requestqueue. */
3745 void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
3747 _receive_message(ls, ms);
3750 /* This is called by the midcomms layer when something is received for
3751 the lockspace. It could be either a MSG (normal message sent as part of
3752 standard locking activity) or an RCOM (recovery message sent as part of
3753 lockspace recovery). */
3755 void dlm_receive_buffer(struct dlm_header *hd, int nodeid)
3757 struct dlm_message *ms = (struct dlm_message *) hd;
3758 struct dlm_rcom *rc = (struct dlm_rcom *) hd;
3762 switch (hd->h_cmd) {
3772 log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
3776 if (hd->h_nodeid != nodeid) {
3777 log_print("invalid h_nodeid %d from %d lockspace %x",
3778 hd->h_nodeid, nodeid, hd->h_lockspace);
3782 ls = dlm_find_lockspace_global(hd->h_lockspace);
3784 log_print("invalid h_lockspace %x from %d cmd %d type %d",
3785 hd->h_lockspace, nodeid, hd->h_cmd, type);
3787 if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
3788 dlm_send_ls_not_ready(nodeid, rc);
3792 /* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3793 be inactive (in this ls) before transitioning to recovery mode */
3795 down_read(&ls->ls_recv_active);
3796 if (hd->h_cmd == DLM_MSG)
3797 dlm_receive_message(ls, ms, nodeid);
3799 dlm_receive_rcom(ls, rc, nodeid);
3800 up_read(&ls->ls_recv_active);
3802 dlm_put_lockspace(ls);
3805 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
3807 if (middle_conversion(lkb)) {
3809 ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
3810 ls->ls_stub_ms.m_result = -EINPROGRESS;
3811 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
3812 _receive_convert_reply(lkb, &ls->ls_stub_ms);
3814 /* Same special case as in receive_rcom_lock_args() */
3815 lkb->lkb_grmode = DLM_LOCK_IV;
3816 rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
3819 } else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
3820 lkb->lkb_flags |= DLM_IFL_RESEND;
3823 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
3824 conversions are async; there's no reply from the remote master */
3827 /* A waiting lkb needs recovery if the master node has failed, or
3828 the master node is changing (only when no directory is used) */
3830 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
3832 if (dlm_is_removed(ls, lkb->lkb_nodeid))
3835 if (!dlm_no_directory(ls))
3838 if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid)
3844 /* Recovery for locks that are waiting for replies from nodes that are now
3845 gone. We can just complete unlocks and cancels by faking a reply from the
3846 dead node. Requests and up-conversions we flag to be resent after
3847 recovery. Down-conversions can just be completed with a fake reply like
3848 unlocks. Conversions between PR and CW need special attention. */
3850 void dlm_recover_waiters_pre(struct dlm_ls *ls)
3852 struct dlm_lkb *lkb, *safe;
3853 int wait_type, stub_unlock_result, stub_cancel_result;
3855 mutex_lock(&ls->ls_waiters_mutex);
3857 list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
3858 log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
3859 lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
3861 /* all outstanding lookups, regardless of destination will be
3862 resent after recovery is done */
3864 if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
3865 lkb->lkb_flags |= DLM_IFL_RESEND;
3869 if (!waiter_needs_recovery(ls, lkb))
3872 wait_type = lkb->lkb_wait_type;
3873 stub_unlock_result = -DLM_EUNLOCK;
3874 stub_cancel_result = -DLM_ECANCEL;
3876 /* Main reply may have been received leaving a zero wait_type,
3877 but a reply for the overlapping op may not have been
3878 received. In that case we need to fake the appropriate
3879 reply for the overlap op. */
3882 if (is_overlap_cancel(lkb)) {
3883 wait_type = DLM_MSG_CANCEL;
3884 if (lkb->lkb_grmode == DLM_LOCK_IV)
3885 stub_cancel_result = 0;
3887 if (is_overlap_unlock(lkb)) {
3888 wait_type = DLM_MSG_UNLOCK;
3889 if (lkb->lkb_grmode == DLM_LOCK_IV)
3890 stub_unlock_result = -ENOENT;
3893 log_debug(ls, "rwpre overlap %x %x %d %d %d",
3894 lkb->lkb_id, lkb->lkb_flags, wait_type,
3895 stub_cancel_result, stub_unlock_result);
3898 switch (wait_type) {
3900 case DLM_MSG_REQUEST:
3901 lkb->lkb_flags |= DLM_IFL_RESEND;
3904 case DLM_MSG_CONVERT:
3905 recover_convert_waiter(ls, lkb);
3908 case DLM_MSG_UNLOCK:
3910 ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
3911 ls->ls_stub_ms.m_result = stub_unlock_result;
3912 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
3913 _receive_unlock_reply(lkb, &ls->ls_stub_ms);
3917 case DLM_MSG_CANCEL:
3919 ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
3920 ls->ls_stub_ms.m_result = stub_cancel_result;
3921 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
3922 _receive_cancel_reply(lkb, &ls->ls_stub_ms);
3927 log_error(ls, "invalid lkb wait_type %d %d",
3928 lkb->lkb_wait_type, wait_type);
3932 mutex_unlock(&ls->ls_waiters_mutex);
3935 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
3937 struct dlm_lkb *lkb;
3940 mutex_lock(&ls->ls_waiters_mutex);
3941 list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
3942 if (lkb->lkb_flags & DLM_IFL_RESEND) {
3948 mutex_unlock(&ls->ls_waiters_mutex);
3955 /* Deal with lookups and lkb's marked RESEND from _pre. We may now be the
3956 master or dir-node for r. Processing the lkb may result in it being placed
3959 /* We do this after normal locking has been enabled and any saved messages
3960 (in requestqueue) have been processed. We should be confident that at
3961 this point we won't get or process a reply to any of these waiting
3962 operations. But, new ops may be coming in on the rsbs/locks here from
3963 userspace or remotely. */
3965 /* there may have been an overlap unlock/cancel prior to recovery or after
3966 recovery. if before, the lkb may still have a pos wait_count; if after, the
3967 overlap flag would just have been set and nothing new sent. we can be
3968 confident here than any replies to either the initial op or overlap ops
3969 prior to recovery have been received. */
3971 int dlm_recover_waiters_post(struct dlm_ls *ls)
3973 struct dlm_lkb *lkb;
3975 int error = 0, mstype, err, oc, ou;
3978 if (dlm_locking_stopped(ls)) {
3979 log_debug(ls, "recover_waiters_post aborted");
3984 lkb = find_resend_waiter(ls);
3988 r = lkb->lkb_resource;
3992 mstype = lkb->lkb_wait_type;
3993 oc = is_overlap_cancel(lkb);
3994 ou = is_overlap_unlock(lkb);
3997 log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
3998 lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
4000 /* At this point we assume that we won't get a reply to any
4001 previous op or overlap op on this lock. First, do a big
4002 remove_from_waiters() for all previous ops. */
4004 lkb->lkb_flags &= ~DLM_IFL_RESEND;
4005 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4006 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4007 lkb->lkb_wait_type = 0;
4008 lkb->lkb_wait_count = 0;
4009 mutex_lock(&ls->ls_waiters_mutex);
4010 list_del_init(&lkb->lkb_wait_reply);
4011 mutex_unlock(&ls->ls_waiters_mutex);
4012 unhold_lkb(lkb); /* for waiters list */
4015 /* do an unlock or cancel instead of resending */
4017 case DLM_MSG_LOOKUP:
4018 case DLM_MSG_REQUEST:
4019 queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
4021 unhold_lkb(lkb); /* undoes create_lkb() */
4023 case DLM_MSG_CONVERT:
4025 queue_cast(r, lkb, -DLM_ECANCEL);
4027 lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
4028 _unlock_lock(r, lkb);
4036 case DLM_MSG_LOOKUP:
4037 case DLM_MSG_REQUEST:
4038 _request_lock(r, lkb);
4040 confirm_master(r, 0);
4042 case DLM_MSG_CONVERT:
4043 _convert_lock(r, lkb);
4051 log_error(ls, "recover_waiters_post %x %d %x %d %d",
4052 lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
4061 static void purge_queue(struct dlm_rsb *r, struct list_head *queue,
4062 int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb))
4064 struct dlm_ls *ls = r->res_ls;
4065 struct dlm_lkb *lkb, *safe;
4067 list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) {
4068 if (test(ls, lkb)) {
4069 rsb_set_flag(r, RSB_LOCKS_PURGED);
4071 /* this put should free the lkb */
4072 if (!dlm_put_lkb(lkb))
4073 log_error(ls, "purged lkb not released");
4078 static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4080 return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid));
4083 static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4085 return is_master_copy(lkb);
4088 static void purge_dead_locks(struct dlm_rsb *r)
4090 purge_queue(r, &r->res_grantqueue, &purge_dead_test);
4091 purge_queue(r, &r->res_convertqueue, &purge_dead_test);
4092 purge_queue(r, &r->res_waitqueue, &purge_dead_test);
4095 void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
4097 purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test);
4098 purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test);
4099 purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test);
4102 /* Get rid of locks held by nodes that are gone. */
4104 int dlm_purge_locks(struct dlm_ls *ls)
4108 log_debug(ls, "dlm_purge_locks");
4110 down_write(&ls->ls_root_sem);
4111 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
4115 purge_dead_locks(r);
4121 up_write(&ls->ls_root_sem);
4126 static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket)
4128 struct dlm_rsb *r, *r_ret = NULL;
4130 read_lock(&ls->ls_rsbtbl[bucket].lock);
4131 list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) {
4132 if (!rsb_flag(r, RSB_LOCKS_PURGED))
4135 rsb_clear_flag(r, RSB_LOCKS_PURGED);
4139 read_unlock(&ls->ls_rsbtbl[bucket].lock);
4143 void dlm_grant_after_purge(struct dlm_ls *ls)
4149 r = find_purged_rsb(ls, bucket);
4151 if (bucket == ls->ls_rsbtbl_size - 1)
4158 grant_pending_locks(r);
4159 confirm_master(r, 0);
4167 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
4170 struct dlm_lkb *lkb;
4172 list_for_each_entry(lkb, head, lkb_statequeue) {
4173 if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
4179 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
4182 struct dlm_lkb *lkb;
4184 lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
4187 lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
4190 lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
4196 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
4197 struct dlm_rsb *r, struct dlm_rcom *rc)
4199 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4202 lkb->lkb_nodeid = rc->rc_header.h_nodeid;
4203 lkb->lkb_ownpid = rl->rl_ownpid;
4204 lkb->lkb_remid = rl->rl_lkid;
4205 lkb->lkb_exflags = rl->rl_exflags;
4206 lkb->lkb_flags = rl->rl_flags & 0x0000FFFF;
4207 lkb->lkb_flags |= DLM_IFL_MSTCPY;
4208 lkb->lkb_lvbseq = rl->rl_lvbseq;
4209 lkb->lkb_rqmode = rl->rl_rqmode;
4210 lkb->lkb_grmode = rl->rl_grmode;
4211 /* don't set lkb_status because add_lkb wants to itself */
4213 lkb->lkb_bastaddr = (void *) (long) (rl->rl_asts & AST_BAST);
4214 lkb->lkb_astaddr = (void *) (long) (rl->rl_asts & AST_COMP);
4216 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
4217 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
4218 if (!lkb->lkb_lvbptr)
4220 lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) -
4221 sizeof(struct rcom_lock);
4222 memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
4225 /* Conversions between PR and CW (middle modes) need special handling.
4226 The real granted mode of these converting locks cannot be determined
4227 until all locks have been rebuilt on the rsb (recover_conversion) */
4229 if (rl->rl_wait_type == DLM_MSG_CONVERT && middle_conversion(lkb)) {
4230 rl->rl_status = DLM_LKSTS_CONVERT;
4231 lkb->lkb_grmode = DLM_LOCK_IV;
4232 rsb_set_flag(r, RSB_RECOVER_CONVERT);
4238 /* This lkb may have been recovered in a previous aborted recovery so we need
4239 to check if the rsb already has an lkb with the given remote nodeid/lkid.
4240 If so we just send back a standard reply. If not, we create a new lkb with
4241 the given values and send back our lkid. We send back our lkid by sending
4242 back the rcom_lock struct we got but with the remid field filled in. */
4244 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4246 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4248 struct dlm_lkb *lkb;
4251 if (rl->rl_parent_lkid) {
4252 error = -EOPNOTSUPP;
4256 error = find_rsb(ls, rl->rl_name, rl->rl_namelen, R_MASTER, &r);
4262 lkb = search_remid(r, rc->rc_header.h_nodeid, rl->rl_lkid);
4268 error = create_lkb(ls, &lkb);
4272 error = receive_rcom_lock_args(ls, lkb, r, rc);
4279 add_lkb(r, lkb, rl->rl_status);
4283 /* this is the new value returned to the lock holder for
4284 saving in its process-copy lkb */
4285 rl->rl_remid = lkb->lkb_id;
4292 log_debug(ls, "recover_master_copy %d %x", error, rl->rl_lkid);
4293 rl->rl_result = error;
4297 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4299 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4301 struct dlm_lkb *lkb;
4304 error = find_lkb(ls, rl->rl_lkid, &lkb);
4306 log_error(ls, "recover_process_copy no lkid %x", rl->rl_lkid);
4310 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
4312 error = rl->rl_result;
4314 r = lkb->lkb_resource;
4320 /* There's a chance the new master received our lock before
4321 dlm_recover_master_reply(), this wouldn't happen if we did
4322 a barrier between recover_masters and recover_locks. */
4323 log_debug(ls, "master copy not ready %x r %lx %s", lkb->lkb_id,
4324 (unsigned long)r, r->res_name);
4325 dlm_send_rcom_lock(r, lkb);
4328 log_debug(ls, "master copy exists %x", lkb->lkb_id);
4331 lkb->lkb_remid = rl->rl_remid;
4334 log_error(ls, "dlm_recover_process_copy unknown error %d %x",
4335 error, lkb->lkb_id);
4338 /* an ack for dlm_recover_locks() which waits for replies from
4339 all the locks it sends to new masters */
4340 dlm_recovered_lock(r);
4349 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
4350 int mode, uint32_t flags, void *name, unsigned int namelen,
4351 unsigned long timeout_cs)
4353 struct dlm_lkb *lkb;
4354 struct dlm_args args;
4357 dlm_lock_recovery(ls);
4359 error = create_lkb(ls, &lkb);
4365 if (flags & DLM_LKF_VALBLK) {
4366 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
4367 if (!ua->lksb.sb_lvbptr) {
4375 /* After ua is attached to lkb it will be freed by dlm_free_lkb().
4376 When DLM_IFL_USER is set, the dlm knows that this is a userspace
4377 lock and that lkb_astparam is the dlm_user_args structure. */
4379 error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
4380 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
4381 lkb->lkb_flags |= DLM_IFL_USER;
4382 ua->old_mode = DLM_LOCK_IV;
4389 error = request_lock(ls, lkb, name, namelen, &args);
4405 /* add this new lkb to the per-process list of locks */
4406 spin_lock(&ua->proc->locks_spin);
4408 list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
4409 spin_unlock(&ua->proc->locks_spin);
4411 dlm_unlock_recovery(ls);
4415 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4416 int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
4417 unsigned long timeout_cs)
4419 struct dlm_lkb *lkb;
4420 struct dlm_args args;
4421 struct dlm_user_args *ua;
4424 dlm_lock_recovery(ls);
4426 error = find_lkb(ls, lkid, &lkb);
4430 /* user can change the params on its lock when it converts it, or
4431 add an lvb that didn't exist before */
4433 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4435 if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
4436 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
4437 if (!ua->lksb.sb_lvbptr) {
4442 if (lvb_in && ua->lksb.sb_lvbptr)
4443 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4445 ua->xid = ua_tmp->xid;
4446 ua->castparam = ua_tmp->castparam;
4447 ua->castaddr = ua_tmp->castaddr;
4448 ua->bastparam = ua_tmp->bastparam;
4449 ua->bastaddr = ua_tmp->bastaddr;
4450 ua->user_lksb = ua_tmp->user_lksb;
4451 ua->old_mode = lkb->lkb_grmode;
4453 error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
4454 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
4458 error = convert_lock(ls, lkb, &args);
4460 if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
4465 dlm_unlock_recovery(ls);
4470 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4471 uint32_t flags, uint32_t lkid, char *lvb_in)
4473 struct dlm_lkb *lkb;
4474 struct dlm_args args;
4475 struct dlm_user_args *ua;
4478 dlm_lock_recovery(ls);
4480 error = find_lkb(ls, lkid, &lkb);
4484 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4486 if (lvb_in && ua->lksb.sb_lvbptr)
4487 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4488 if (ua_tmp->castparam)
4489 ua->castparam = ua_tmp->castparam;
4490 ua->user_lksb = ua_tmp->user_lksb;
4492 error = set_unlock_args(flags, ua, &args);
4496 error = unlock_lock(ls, lkb, &args);
4498 if (error == -DLM_EUNLOCK)
4500 /* from validate_unlock_args() */
4501 if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
4506 spin_lock(&ua->proc->locks_spin);
4507 /* dlm_user_add_ast() may have already taken lkb off the proc list */
4508 if (!list_empty(&lkb->lkb_ownqueue))
4509 list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
4510 spin_unlock(&ua->proc->locks_spin);
4514 dlm_unlock_recovery(ls);
4519 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4520 uint32_t flags, uint32_t lkid)
4522 struct dlm_lkb *lkb;
4523 struct dlm_args args;
4524 struct dlm_user_args *ua;
4527 dlm_lock_recovery(ls);
4529 error = find_lkb(ls, lkid, &lkb);
4533 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4534 if (ua_tmp->castparam)
4535 ua->castparam = ua_tmp->castparam;
4536 ua->user_lksb = ua_tmp->user_lksb;
4538 error = set_unlock_args(flags, ua, &args);
4542 error = cancel_lock(ls, lkb, &args);
4544 if (error == -DLM_ECANCEL)
4546 /* from validate_unlock_args() */
4547 if (error == -EBUSY)
4552 dlm_unlock_recovery(ls);
4557 int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
4559 struct dlm_lkb *lkb;
4560 struct dlm_args args;
4561 struct dlm_user_args *ua;
4565 dlm_lock_recovery(ls);
4567 error = find_lkb(ls, lkid, &lkb);
4571 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4573 error = set_unlock_args(flags, ua, &args);
4577 /* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4579 r = lkb->lkb_resource;
4583 error = validate_unlock_args(lkb, &args);
4586 lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
4588 error = _cancel_lock(r, lkb);
4593 if (error == -DLM_ECANCEL)
4595 /* from validate_unlock_args() */
4596 if (error == -EBUSY)
4601 dlm_unlock_recovery(ls);
4605 /* lkb's that are removed from the waiters list by revert are just left on the
4606 orphans list with the granted orphan locks, to be freed by purge */
4608 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4610 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
4611 struct dlm_args args;
4615 mutex_lock(&ls->ls_orphans_mutex);
4616 list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
4617 mutex_unlock(&ls->ls_orphans_mutex);
4619 set_unlock_args(0, ua, &args);
4621 error = cancel_lock(ls, lkb, &args);
4622 if (error == -DLM_ECANCEL)
4627 /* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4628 Regardless of what rsb queue the lock is on, it's removed and freed. */
4630 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4632 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
4633 struct dlm_args args;
4636 set_unlock_args(DLM_LKF_FORCEUNLOCK, ua, &args);
4638 error = unlock_lock(ls, lkb, &args);
4639 if (error == -DLM_EUNLOCK)
4644 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4645 (which does lock_rsb) due to deadlock with receiving a message that does
4646 lock_rsb followed by dlm_user_add_ast() */
4648 static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
4649 struct dlm_user_proc *proc)
4651 struct dlm_lkb *lkb = NULL;
4653 mutex_lock(&ls->ls_clear_proc_locks);
4654 if (list_empty(&proc->locks))
4657 lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
4658 list_del_init(&lkb->lkb_ownqueue);
4660 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4661 lkb->lkb_flags |= DLM_IFL_ORPHAN;
4663 lkb->lkb_flags |= DLM_IFL_DEAD;
4665 mutex_unlock(&ls->ls_clear_proc_locks);
4669 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4670 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4671 which we clear here. */
4673 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
4674 list, and no more device_writes should add lkb's to proc->locks list; so we
4675 shouldn't need to take asts_spin or locks_spin here. this assumes that
4676 device reads/writes/closes are serialized -- FIXME: we may need to serialize
4679 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4681 struct dlm_lkb *lkb, *safe;
4683 dlm_lock_recovery(ls);
4686 lkb = del_proc_lock(ls, proc);
4690 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4691 orphan_proc_lock(ls, lkb);
4693 unlock_proc_lock(ls, lkb);
4695 /* this removes the reference for the proc->locks list
4696 added by dlm_user_request, it may result in the lkb
4702 mutex_lock(&ls->ls_clear_proc_locks);
4704 /* in-progress unlocks */
4705 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4706 list_del_init(&lkb->lkb_ownqueue);
4707 lkb->lkb_flags |= DLM_IFL_DEAD;
4711 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4712 lkb->lkb_ast_type = 0;
4713 list_del(&lkb->lkb_astqueue);
4717 mutex_unlock(&ls->ls_clear_proc_locks);
4718 dlm_unlock_recovery(ls);
4721 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4723 struct dlm_lkb *lkb, *safe;
4727 spin_lock(&proc->locks_spin);
4728 if (!list_empty(&proc->locks)) {
4729 lkb = list_entry(proc->locks.next, struct dlm_lkb,
4731 list_del_init(&lkb->lkb_ownqueue);
4733 spin_unlock(&proc->locks_spin);
4738 lkb->lkb_flags |= DLM_IFL_DEAD;
4739 unlock_proc_lock(ls, lkb);
4740 dlm_put_lkb(lkb); /* ref from proc->locks list */
4743 spin_lock(&proc->locks_spin);
4744 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4745 list_del_init(&lkb->lkb_ownqueue);
4746 lkb->lkb_flags |= DLM_IFL_DEAD;
4749 spin_unlock(&proc->locks_spin);
4751 spin_lock(&proc->asts_spin);
4752 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4753 list_del(&lkb->lkb_astqueue);
4756 spin_unlock(&proc->asts_spin);
4759 /* pid of 0 means purge all orphans */
4761 static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
4763 struct dlm_lkb *lkb, *safe;
4765 mutex_lock(&ls->ls_orphans_mutex);
4766 list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
4767 if (pid && lkb->lkb_ownpid != pid)
4769 unlock_proc_lock(ls, lkb);
4770 list_del_init(&lkb->lkb_ownqueue);
4773 mutex_unlock(&ls->ls_orphans_mutex);
4776 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
4778 struct dlm_message *ms;
4779 struct dlm_mhandle *mh;
4782 error = _create_message(ls, sizeof(struct dlm_message), nodeid,
4783 DLM_MSG_PURGE, &ms, &mh);
4786 ms->m_nodeid = nodeid;
4789 return send_message(mh, ms);
4792 int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
4793 int nodeid, int pid)
4797 if (nodeid != dlm_our_nodeid()) {
4798 error = send_purge(ls, nodeid, pid);
4800 dlm_lock_recovery(ls);
4801 if (pid == current->pid)
4802 purge_proc_locks(ls, proc);
4804 do_purge(ls, nodeid, pid);
4805 dlm_unlock_recovery(ls);