1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
6 ** This copyrighted material is made available to anyone wishing to use,
7 ** modify, copy, or redistribute it subject to the terms and conditions
8 ** of the GNU General Public License v.2.
10 *******************************************************************************
11 ******************************************************************************/
13 /* Central locking logic has four stages:
33 Stage 1 (lock, unlock) is mainly about checking input args and
34 splitting into one of the four main operations:
36 dlm_lock = request_lock
37 dlm_lock+CONVERT = convert_lock
38 dlm_unlock = unlock_lock
39 dlm_unlock+CANCEL = cancel_lock
41 Stage 2, xxxx_lock(), just finds and locks the relevant rsb which is
42 provided to the next stage.
44 Stage 3, _xxxx_lock(), determines if the operation is local or remote.
45 When remote, it calls send_xxxx(), when local it calls do_xxxx().
47 Stage 4, do_xxxx(), is the guts of the operation. It manipulates the
48 given rsb and lkb and queues callbacks.
50 For remote operations, send_xxxx() results in the corresponding do_xxxx()
51 function being executed on the remote node. The connecting send/receive
52 calls on local (L) and remote (R) nodes:
54 L: send_xxxx() -> R: receive_xxxx()
56 L: receive_xxxx_reply() <- R: send_xxxx_reply()
58 #include <linux/types.h>
59 #include "dlm_internal.h"
60 #include <linux/dlm_device.h>
63 #include "requestqueue.h"
67 #include "lockspace.h"
72 #include "lvb_table.h"
76 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb);
77 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb);
78 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb);
79 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb);
80 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb);
81 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode);
82 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb);
83 static int send_remove(struct dlm_rsb *r);
84 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
85 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
86 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
87 struct dlm_message *ms);
88 static int receive_extralen(struct dlm_message *ms);
89 static void do_purge(struct dlm_ls *ls, int nodeid, int pid);
90 static void del_timeout(struct dlm_lkb *lkb);
93 * Lock compatibilty matrix - thanks Steve
94 * UN = Unlocked state. Not really a state, used as a flag
95 * PD = Padding. Used to make the matrix a nice power of two in size
96 * Other states are the same as the VMS DLM.
97 * Usage: matrix[grmode+1][rqmode+1] (although m[rq+1][gr+1] is the same)
100 static const int __dlm_compat_matrix[8][8] = {
101 /* UN NL CR CW PR PW EX PD */
102 {1, 1, 1, 1, 1, 1, 1, 0}, /* UN */
103 {1, 1, 1, 1, 1, 1, 1, 0}, /* NL */
104 {1, 1, 1, 1, 1, 1, 0, 0}, /* CR */
105 {1, 1, 1, 1, 0, 0, 0, 0}, /* CW */
106 {1, 1, 1, 0, 1, 0, 0, 0}, /* PR */
107 {1, 1, 1, 0, 0, 0, 0, 0}, /* PW */
108 {1, 1, 0, 0, 0, 0, 0, 0}, /* EX */
109 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
113 * This defines the direction of transfer of LVB data.
114 * Granted mode is the row; requested mode is the column.
115 * Usage: matrix[grmode+1][rqmode+1]
116 * 1 = LVB is returned to the caller
117 * 0 = LVB is written to the resource
118 * -1 = nothing happens to the LVB
121 const int dlm_lvb_operations[8][8] = {
122 /* UN NL CR CW PR PW EX PD*/
123 { -1, 1, 1, 1, 1, 1, 1, -1 }, /* UN */
124 { -1, 1, 1, 1, 1, 1, 1, 0 }, /* NL */
125 { -1, -1, 1, 1, 1, 1, 1, 0 }, /* CR */
126 { -1, -1, -1, 1, 1, 1, 1, 0 }, /* CW */
127 { -1, -1, -1, -1, 1, 1, 1, 0 }, /* PR */
128 { -1, 0, 0, 0, 0, 0, 1, 0 }, /* PW */
129 { -1, 0, 0, 0, 0, 0, 0, 0 }, /* EX */
130 { -1, 0, 0, 0, 0, 0, 0, 0 } /* PD */
133 #define modes_compat(gr, rq) \
134 __dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
136 int dlm_modes_compat(int mode1, int mode2)
138 return __dlm_compat_matrix[mode1 + 1][mode2 + 1];
142 * Compatibility matrix for conversions with QUECVT set.
143 * Granted mode is the row; requested mode is the column.
144 * Usage: matrix[grmode+1][rqmode+1]
147 static const int __quecvt_compat_matrix[8][8] = {
148 /* UN NL CR CW PR PW EX PD */
149 {0, 0, 0, 0, 0, 0, 0, 0}, /* UN */
150 {0, 0, 1, 1, 1, 1, 1, 0}, /* NL */
151 {0, 0, 0, 1, 1, 1, 1, 0}, /* CR */
152 {0, 0, 0, 0, 1, 1, 1, 0}, /* CW */
153 {0, 0, 0, 1, 0, 1, 1, 0}, /* PR */
154 {0, 0, 0, 0, 0, 0, 1, 0}, /* PW */
155 {0, 0, 0, 0, 0, 0, 0, 0}, /* EX */
156 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
159 void dlm_print_lkb(struct dlm_lkb *lkb)
161 printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x\n"
162 " status %d rqmode %d grmode %d wait_type %d ast_type %d\n",
163 lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags,
164 lkb->lkb_flags, lkb->lkb_status, lkb->lkb_rqmode,
165 lkb->lkb_grmode, lkb->lkb_wait_type, lkb->lkb_ast_type);
168 static 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, 0);
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 lkb->lkb_time_bast = ktime_get();
323 if (is_master_copy(lkb))
324 send_bast(r, lkb, rqmode);
326 dlm_add_ast(lkb, AST_BAST, rqmode);
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))
370 if (r->res_nodeid && (flags & R_MASTER))
376 static int _search_rsb(struct dlm_ls *ls, char *name, int len, int b,
377 unsigned int flags, struct dlm_rsb **r_ret)
382 error = search_rsb_list(&ls->ls_rsbtbl[b].list, name, len, flags, &r);
384 kref_get(&r->res_ref);
387 error = search_rsb_list(&ls->ls_rsbtbl[b].toss, name, len, flags, &r);
391 list_move(&r->res_hashchain, &ls->ls_rsbtbl[b].list);
393 if (dlm_no_directory(ls))
396 if (r->res_nodeid == -1) {
397 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
398 r->res_first_lkid = 0;
399 } else if (r->res_nodeid > 0) {
400 rsb_set_flag(r, RSB_MASTER_UNCERTAIN);
401 r->res_first_lkid = 0;
403 DLM_ASSERT(r->res_nodeid == 0, dlm_print_rsb(r););
404 DLM_ASSERT(!rsb_flag(r, RSB_MASTER_UNCERTAIN),);
411 static int search_rsb(struct dlm_ls *ls, char *name, int len, int b,
412 unsigned int flags, struct dlm_rsb **r_ret)
415 spin_lock(&ls->ls_rsbtbl[b].lock);
416 error = _search_rsb(ls, name, len, b, flags, r_ret);
417 spin_unlock(&ls->ls_rsbtbl[b].lock);
422 * Find rsb in rsbtbl and potentially create/add one
424 * Delaying the release of rsb's has a similar benefit to applications keeping
425 * NL locks on an rsb, but without the guarantee that the cached master value
426 * will still be valid when the rsb is reused. Apps aren't always smart enough
427 * to keep NL locks on an rsb that they may lock again shortly; this can lead
428 * to excessive master lookups and removals if we don't delay the release.
430 * Searching for an rsb means looking through both the normal list and toss
431 * list. When found on the toss list the rsb is moved to the normal list with
432 * ref count of 1; when found on normal list the ref count is incremented.
435 static int find_rsb(struct dlm_ls *ls, char *name, int namelen,
436 unsigned int flags, struct dlm_rsb **r_ret)
438 struct dlm_rsb *r, *tmp;
439 uint32_t hash, bucket;
442 if (namelen > DLM_RESNAME_MAXLEN)
445 if (dlm_no_directory(ls))
449 hash = jhash(name, namelen, 0);
450 bucket = hash & (ls->ls_rsbtbl_size - 1);
452 error = search_rsb(ls, name, namelen, bucket, flags, &r);
456 if (error == -EBADR && !(flags & R_CREATE))
459 /* the rsb was found but wasn't a master copy */
460 if (error == -ENOTBLK)
464 r = create_rsb(ls, name, namelen);
469 r->res_bucket = bucket;
471 kref_init(&r->res_ref);
473 /* With no directory, the master can be set immediately */
474 if (dlm_no_directory(ls)) {
475 int nodeid = dlm_dir_nodeid(r);
476 if (nodeid == dlm_our_nodeid())
478 r->res_nodeid = nodeid;
481 spin_lock(&ls->ls_rsbtbl[bucket].lock);
482 error = _search_rsb(ls, name, namelen, bucket, 0, &tmp);
484 spin_unlock(&ls->ls_rsbtbl[bucket].lock);
489 list_add(&r->res_hashchain, &ls->ls_rsbtbl[bucket].list);
490 spin_unlock(&ls->ls_rsbtbl[bucket].lock);
497 /* This is only called to add a reference when the code already holds
498 a valid reference to the rsb, so there's no need for locking. */
500 static inline void hold_rsb(struct dlm_rsb *r)
502 kref_get(&r->res_ref);
505 void dlm_hold_rsb(struct dlm_rsb *r)
510 static void toss_rsb(struct kref *kref)
512 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
513 struct dlm_ls *ls = r->res_ls;
515 DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
516 kref_init(&r->res_ref);
517 list_move(&r->res_hashchain, &ls->ls_rsbtbl[r->res_bucket].toss);
518 r->res_toss_time = jiffies;
520 dlm_free_lvb(r->res_lvbptr);
521 r->res_lvbptr = NULL;
525 /* When all references to the rsb are gone it's transfered to
526 the tossed list for later disposal. */
528 static void put_rsb(struct dlm_rsb *r)
530 struct dlm_ls *ls = r->res_ls;
531 uint32_t bucket = r->res_bucket;
533 spin_lock(&ls->ls_rsbtbl[bucket].lock);
534 kref_put(&r->res_ref, toss_rsb);
535 spin_unlock(&ls->ls_rsbtbl[bucket].lock);
538 void dlm_put_rsb(struct dlm_rsb *r)
543 /* See comment for unhold_lkb */
545 static void unhold_rsb(struct dlm_rsb *r)
548 rv = kref_put(&r->res_ref, toss_rsb);
549 DLM_ASSERT(!rv, dlm_dump_rsb(r););
552 static void kill_rsb(struct kref *kref)
554 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
556 /* All work is done after the return from kref_put() so we
557 can release the write_lock before the remove and free. */
559 DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
560 DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
561 DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
562 DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
563 DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
564 DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
567 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
568 The rsb must exist as long as any lkb's for it do. */
570 static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
573 lkb->lkb_resource = r;
576 static void detach_lkb(struct dlm_lkb *lkb)
578 if (lkb->lkb_resource) {
579 put_rsb(lkb->lkb_resource);
580 lkb->lkb_resource = NULL;
584 static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
586 struct dlm_lkb *lkb, *tmp;
590 lkb = dlm_allocate_lkb(ls);
594 lkb->lkb_nodeid = -1;
595 lkb->lkb_grmode = DLM_LOCK_IV;
596 kref_init(&lkb->lkb_ref);
597 INIT_LIST_HEAD(&lkb->lkb_ownqueue);
598 INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
599 INIT_LIST_HEAD(&lkb->lkb_time_list);
601 get_random_bytes(&bucket, sizeof(bucket));
602 bucket &= (ls->ls_lkbtbl_size - 1);
604 write_lock(&ls->ls_lkbtbl[bucket].lock);
606 /* counter can roll over so we must verify lkid is not in use */
609 lkid = (bucket << 16) | ls->ls_lkbtbl[bucket].counter++;
611 list_for_each_entry(tmp, &ls->ls_lkbtbl[bucket].list,
613 if (tmp->lkb_id != lkid)
621 list_add(&lkb->lkb_idtbl_list, &ls->ls_lkbtbl[bucket].list);
622 write_unlock(&ls->ls_lkbtbl[bucket].lock);
628 static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid)
631 uint16_t bucket = (lkid >> 16);
633 list_for_each_entry(lkb, &ls->ls_lkbtbl[bucket].list, lkb_idtbl_list) {
634 if (lkb->lkb_id == lkid)
640 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
643 uint16_t bucket = (lkid >> 16);
645 if (bucket >= ls->ls_lkbtbl_size)
648 read_lock(&ls->ls_lkbtbl[bucket].lock);
649 lkb = __find_lkb(ls, lkid);
651 kref_get(&lkb->lkb_ref);
652 read_unlock(&ls->ls_lkbtbl[bucket].lock);
655 return lkb ? 0 : -ENOENT;
658 static void kill_lkb(struct kref *kref)
660 struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
662 /* All work is done after the return from kref_put() so we
663 can release the write_lock before the detach_lkb */
665 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
668 /* __put_lkb() is used when an lkb may not have an rsb attached to
669 it so we need to provide the lockspace explicitly */
671 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
673 uint16_t bucket = (lkb->lkb_id >> 16);
675 write_lock(&ls->ls_lkbtbl[bucket].lock);
676 if (kref_put(&lkb->lkb_ref, kill_lkb)) {
677 list_del(&lkb->lkb_idtbl_list);
678 write_unlock(&ls->ls_lkbtbl[bucket].lock);
682 /* for local/process lkbs, lvbptr points to caller's lksb */
683 if (lkb->lkb_lvbptr && is_master_copy(lkb))
684 dlm_free_lvb(lkb->lkb_lvbptr);
688 write_unlock(&ls->ls_lkbtbl[bucket].lock);
693 int dlm_put_lkb(struct dlm_lkb *lkb)
697 DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb););
698 DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb););
700 ls = lkb->lkb_resource->res_ls;
701 return __put_lkb(ls, lkb);
704 /* This is only called to add a reference when the code already holds
705 a valid reference to the lkb, so there's no need for locking. */
707 static inline void hold_lkb(struct dlm_lkb *lkb)
709 kref_get(&lkb->lkb_ref);
712 /* This is called when we need to remove a reference and are certain
713 it's not the last ref. e.g. del_lkb is always called between a
714 find_lkb/put_lkb and is always the inverse of a previous add_lkb.
715 put_lkb would work fine, but would involve unnecessary locking */
717 static inline void unhold_lkb(struct dlm_lkb *lkb)
720 rv = kref_put(&lkb->lkb_ref, kill_lkb);
721 DLM_ASSERT(!rv, dlm_print_lkb(lkb););
724 static void lkb_add_ordered(struct list_head *new, struct list_head *head,
727 struct dlm_lkb *lkb = NULL;
729 list_for_each_entry(lkb, head, lkb_statequeue)
730 if (lkb->lkb_rqmode < mode)
734 list_add_tail(new, head);
736 __list_add(new, lkb->lkb_statequeue.prev, &lkb->lkb_statequeue);
739 /* add/remove lkb to rsb's grant/convert/wait queue */
741 static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status)
743 kref_get(&lkb->lkb_ref);
745 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
747 lkb->lkb_timestamp = ktime_get();
749 lkb->lkb_status = status;
752 case DLM_LKSTS_WAITING:
753 if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
754 list_add(&lkb->lkb_statequeue, &r->res_waitqueue);
756 list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue);
758 case DLM_LKSTS_GRANTED:
759 /* convention says granted locks kept in order of grmode */
760 lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue,
763 case DLM_LKSTS_CONVERT:
764 if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
765 list_add(&lkb->lkb_statequeue, &r->res_convertqueue);
767 list_add_tail(&lkb->lkb_statequeue,
768 &r->res_convertqueue);
771 DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status););
775 static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
778 list_del(&lkb->lkb_statequeue);
782 static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
786 add_lkb(r, lkb, sts);
790 static int msg_reply_type(int mstype)
793 case DLM_MSG_REQUEST:
794 return DLM_MSG_REQUEST_REPLY;
795 case DLM_MSG_CONVERT:
796 return DLM_MSG_CONVERT_REPLY;
798 return DLM_MSG_UNLOCK_REPLY;
800 return DLM_MSG_CANCEL_REPLY;
802 return DLM_MSG_LOOKUP_REPLY;
807 /* add/remove lkb from global waiters list of lkb's waiting for
808 a reply from a remote node */
810 static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
812 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
815 mutex_lock(&ls->ls_waiters_mutex);
817 if (is_overlap_unlock(lkb) ||
818 (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
823 if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
826 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
829 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
835 lkb->lkb_wait_count++;
838 log_debug(ls, "add overlap %x cur %d new %d count %d flags %x",
839 lkb->lkb_id, lkb->lkb_wait_type, mstype,
840 lkb->lkb_wait_count, lkb->lkb_flags);
844 DLM_ASSERT(!lkb->lkb_wait_count,
846 printk("wait_count %d\n", lkb->lkb_wait_count););
848 lkb->lkb_wait_count++;
849 lkb->lkb_wait_type = mstype;
851 list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
854 log_error(ls, "add_to_waiters %x error %d flags %x %d %d %s",
855 lkb->lkb_id, error, lkb->lkb_flags, mstype,
856 lkb->lkb_wait_type, lkb->lkb_resource->res_name);
857 mutex_unlock(&ls->ls_waiters_mutex);
861 /* We clear the RESEND flag because we might be taking an lkb off the waiters
862 list as part of process_requestqueue (e.g. a lookup that has an optimized
863 request reply on the requestqueue) between dlm_recover_waiters_pre() which
864 set RESEND and dlm_recover_waiters_post() */
866 static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
868 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
869 int overlap_done = 0;
871 if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
872 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
877 if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
878 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
883 /* N.B. type of reply may not always correspond to type of original
884 msg due to lookup->request optimization, verify others? */
886 if (lkb->lkb_wait_type) {
887 lkb->lkb_wait_type = 0;
891 log_error(ls, "remove_from_waiters lkid %x flags %x types %d %d",
892 lkb->lkb_id, lkb->lkb_flags, mstype, lkb->lkb_wait_type);
896 /* the force-unlock/cancel has completed and we haven't recvd a reply
897 to the op that was in progress prior to the unlock/cancel; we
898 give up on any reply to the earlier op. FIXME: not sure when/how
901 if (overlap_done && lkb->lkb_wait_type) {
902 log_error(ls, "remove_from_waiters %x reply %d give up on %d",
903 lkb->lkb_id, mstype, lkb->lkb_wait_type);
904 lkb->lkb_wait_count--;
905 lkb->lkb_wait_type = 0;
908 DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
910 lkb->lkb_flags &= ~DLM_IFL_RESEND;
911 lkb->lkb_wait_count--;
912 if (!lkb->lkb_wait_count)
913 list_del_init(&lkb->lkb_wait_reply);
918 static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
920 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
923 mutex_lock(&ls->ls_waiters_mutex);
924 error = _remove_from_waiters(lkb, mstype);
925 mutex_unlock(&ls->ls_waiters_mutex);
929 /* Handles situations where we might be processing a "fake" or "stub" reply in
930 which we can't try to take waiters_mutex again. */
932 static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
934 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
937 if (ms != &ls->ls_stub_ms)
938 mutex_lock(&ls->ls_waiters_mutex);
939 error = _remove_from_waiters(lkb, ms->m_type);
940 if (ms != &ls->ls_stub_ms)
941 mutex_unlock(&ls->ls_waiters_mutex);
945 static void dir_remove(struct dlm_rsb *r)
949 if (dlm_no_directory(r->res_ls))
952 to_nodeid = dlm_dir_nodeid(r);
953 if (to_nodeid != dlm_our_nodeid())
956 dlm_dir_remove_entry(r->res_ls, to_nodeid,
957 r->res_name, r->res_length);
960 /* FIXME: shouldn't this be able to exit as soon as one non-due rsb is
961 found since they are in order of newest to oldest? */
963 static int shrink_bucket(struct dlm_ls *ls, int b)
966 int count = 0, found;
970 spin_lock(&ls->ls_rsbtbl[b].lock);
971 list_for_each_entry_reverse(r, &ls->ls_rsbtbl[b].toss,
973 if (!time_after_eq(jiffies, r->res_toss_time +
974 dlm_config.ci_toss_secs * HZ))
981 spin_unlock(&ls->ls_rsbtbl[b].lock);
985 if (kref_put(&r->res_ref, kill_rsb)) {
986 list_del(&r->res_hashchain);
987 spin_unlock(&ls->ls_rsbtbl[b].lock);
994 spin_unlock(&ls->ls_rsbtbl[b].lock);
995 log_error(ls, "tossed rsb in use %s", r->res_name);
1002 void dlm_scan_rsbs(struct dlm_ls *ls)
1006 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1007 shrink_bucket(ls, i);
1008 if (dlm_locking_stopped(ls))
1014 static void add_timeout(struct dlm_lkb *lkb)
1016 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1018 if (is_master_copy(lkb))
1021 if (test_bit(LSFL_TIMEWARN, &ls->ls_flags) &&
1022 !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1023 lkb->lkb_flags |= DLM_IFL_WATCH_TIMEWARN;
1026 if (lkb->lkb_exflags & DLM_LKF_TIMEOUT)
1031 DLM_ASSERT(list_empty(&lkb->lkb_time_list), dlm_print_lkb(lkb););
1032 mutex_lock(&ls->ls_timeout_mutex);
1034 list_add_tail(&lkb->lkb_time_list, &ls->ls_timeout);
1035 mutex_unlock(&ls->ls_timeout_mutex);
1038 static void del_timeout(struct dlm_lkb *lkb)
1040 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1042 mutex_lock(&ls->ls_timeout_mutex);
1043 if (!list_empty(&lkb->lkb_time_list)) {
1044 list_del_init(&lkb->lkb_time_list);
1047 mutex_unlock(&ls->ls_timeout_mutex);
1050 /* FIXME: is it safe to look at lkb_exflags, lkb_flags, lkb_timestamp, and
1051 lkb_lksb_timeout without lock_rsb? Note: we can't lock timeout_mutex
1052 and then lock rsb because of lock ordering in add_timeout. We may need
1053 to specify some special timeout-related bits in the lkb that are just to
1054 be accessed under the timeout_mutex. */
1056 void dlm_scan_timeout(struct dlm_ls *ls)
1059 struct dlm_lkb *lkb;
1060 int do_cancel, do_warn;
1064 if (dlm_locking_stopped(ls))
1069 mutex_lock(&ls->ls_timeout_mutex);
1070 list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list) {
1072 wait_us = ktime_to_us(ktime_sub(ktime_get(),
1073 lkb->lkb_timestamp));
1075 if ((lkb->lkb_exflags & DLM_LKF_TIMEOUT) &&
1076 wait_us >= (lkb->lkb_timeout_cs * 10000))
1079 if ((lkb->lkb_flags & DLM_IFL_WATCH_TIMEWARN) &&
1080 wait_us >= dlm_config.ci_timewarn_cs * 10000)
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 u64 adj_us = jiffies_to_usecs(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 = ktime_add_us(lkb->lkb_timestamp, adj_us);
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 if (len > DLM_RESNAME_MAXLEN)
1232 len = DLM_RESNAME_MAXLEN;
1233 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
1234 lkb->lkb_lvbseq = ms->m_lvbseq;
1238 /* Manipulate lkb's on rsb's convert/granted/waiting queues
1239 remove_lock -- used for unlock, removes lkb from granted
1240 revert_lock -- used for cancel, moves lkb from convert to granted
1241 grant_lock -- used for request and convert, adds lkb to granted or
1242 moves lkb from convert or waiting to granted
1244 Each of these is used for master or local copy lkb's. There is
1245 also a _pc() variation used to make the corresponding change on
1246 a process copy (pc) lkb. */
1248 static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1251 lkb->lkb_grmode = DLM_LOCK_IV;
1252 /* this unhold undoes the original ref from create_lkb()
1253 so this leads to the lkb being freed */
1257 static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1259 set_lvb_unlock(r, lkb);
1260 _remove_lock(r, lkb);
1263 static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1265 _remove_lock(r, lkb);
1268 /* returns: 0 did nothing
1269 1 moved lock to granted
1272 static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1276 lkb->lkb_rqmode = DLM_LOCK_IV;
1278 switch (lkb->lkb_status) {
1279 case DLM_LKSTS_GRANTED:
1281 case DLM_LKSTS_CONVERT:
1282 move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1285 case DLM_LKSTS_WAITING:
1287 lkb->lkb_grmode = DLM_LOCK_IV;
1288 /* this unhold undoes the original ref from create_lkb()
1289 so this leads to the lkb being freed */
1294 log_print("invalid status for revert %d", lkb->lkb_status);
1299 static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1301 return revert_lock(r, lkb);
1304 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1306 if (lkb->lkb_grmode != lkb->lkb_rqmode) {
1307 lkb->lkb_grmode = lkb->lkb_rqmode;
1308 if (lkb->lkb_status)
1309 move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1311 add_lkb(r, lkb, DLM_LKSTS_GRANTED);
1314 lkb->lkb_rqmode = DLM_LOCK_IV;
1317 static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1319 set_lvb_lock(r, lkb);
1320 _grant_lock(r, lkb);
1321 lkb->lkb_highbast = 0;
1324 static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1325 struct dlm_message *ms)
1327 set_lvb_lock_pc(r, lkb, ms);
1328 _grant_lock(r, lkb);
1331 /* called by grant_pending_locks() which means an async grant message must
1332 be sent to the requesting node in addition to granting the lock if the
1333 lkb belongs to a remote node. */
1335 static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
1338 if (is_master_copy(lkb))
1341 queue_cast(r, lkb, 0);
1344 /* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
1345 change the granted/requested modes. We're munging things accordingly in
1347 CONVDEADLK: our grmode may have been forced down to NL to resolve a
1349 ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
1350 compatible with other granted locks */
1352 static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
1354 if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
1355 log_print("munge_demoted %x invalid reply type %d",
1356 lkb->lkb_id, ms->m_type);
1360 if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
1361 log_print("munge_demoted %x invalid modes gr %d rq %d",
1362 lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
1366 lkb->lkb_grmode = DLM_LOCK_NL;
1369 static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
1371 if (ms->m_type != DLM_MSG_REQUEST_REPLY &&
1372 ms->m_type != DLM_MSG_GRANT) {
1373 log_print("munge_altmode %x invalid reply type %d",
1374 lkb->lkb_id, ms->m_type);
1378 if (lkb->lkb_exflags & DLM_LKF_ALTPR)
1379 lkb->lkb_rqmode = DLM_LOCK_PR;
1380 else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
1381 lkb->lkb_rqmode = DLM_LOCK_CW;
1383 log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
1388 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
1390 struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
1392 if (lkb->lkb_id == first->lkb_id)
1398 /* Check if the given lkb conflicts with another lkb on the queue. */
1400 static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb)
1402 struct dlm_lkb *this;
1404 list_for_each_entry(this, head, lkb_statequeue) {
1407 if (!modes_compat(this, lkb))
1414 * "A conversion deadlock arises with a pair of lock requests in the converting
1415 * queue for one resource. The granted mode of each lock blocks the requested
1416 * mode of the other lock."
1418 * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
1419 * convert queue from being granted, then deadlk/demote lkb.
1422 * Granted Queue: empty
1423 * Convert Queue: NL->EX (first lock)
1424 * PR->EX (second lock)
1426 * The first lock can't be granted because of the granted mode of the second
1427 * lock and the second lock can't be granted because it's not first in the
1428 * list. We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
1429 * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
1430 * flag set and return DEMOTED in the lksb flags.
1432 * Originally, this function detected conv-deadlk in a more limited scope:
1433 * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
1434 * - if lkb1 was the first entry in the queue (not just earlier), and was
1435 * blocked by the granted mode of lkb2, and there was nothing on the
1436 * granted queue preventing lkb1 from being granted immediately, i.e.
1437 * lkb2 was the only thing preventing lkb1 from being granted.
1439 * That second condition meant we'd only say there was conv-deadlk if
1440 * resolving it (by demotion) would lead to the first lock on the convert
1441 * queue being granted right away. It allowed conversion deadlocks to exist
1442 * between locks on the convert queue while they couldn't be granted anyway.
1444 * Now, we detect and take action on conversion deadlocks immediately when
1445 * they're created, even if they may not be immediately consequential. If
1446 * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
1447 * mode that would prevent lkb1's conversion from being granted, we do a
1448 * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
1449 * I think this means that the lkb_is_ahead condition below should always
1450 * be zero, i.e. there will never be conv-deadlk between two locks that are
1451 * both already on the convert queue.
1454 static int conversion_deadlock_detect(struct dlm_rsb *r, struct dlm_lkb *lkb2)
1456 struct dlm_lkb *lkb1;
1457 int lkb_is_ahead = 0;
1459 list_for_each_entry(lkb1, &r->res_convertqueue, lkb_statequeue) {
1465 if (!lkb_is_ahead) {
1466 if (!modes_compat(lkb2, lkb1))
1469 if (!modes_compat(lkb2, lkb1) &&
1470 !modes_compat(lkb1, lkb2))
1478 * Return 1 if the lock can be granted, 0 otherwise.
1479 * Also detect and resolve conversion deadlocks.
1481 * lkb is the lock to be granted
1483 * now is 1 if the function is being called in the context of the
1484 * immediate request, it is 0 if called later, after the lock has been
1487 * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
1490 static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now)
1492 int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV);
1495 * 6-10: Version 5.4 introduced an option to address the phenomenon of
1496 * a new request for a NL mode lock being blocked.
1498 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
1499 * request, then it would be granted. In essence, the use of this flag
1500 * tells the Lock Manager to expedite theis request by not considering
1501 * what may be in the CONVERTING or WAITING queues... As of this
1502 * writing, the EXPEDITE flag can be used only with new requests for NL
1503 * mode locks. This flag is not valid for conversion requests.
1505 * A shortcut. Earlier checks return an error if EXPEDITE is used in a
1506 * conversion or used with a non-NL requested mode. We also know an
1507 * EXPEDITE request is always granted immediately, so now must always
1508 * be 1. The full condition to grant an expedite request: (now &&
1509 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
1510 * therefore be shortened to just checking the flag.
1513 if (lkb->lkb_exflags & DLM_LKF_EXPEDITE)
1517 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
1518 * added to the remaining conditions.
1521 if (queue_conflict(&r->res_grantqueue, lkb))
1525 * 6-3: By default, a conversion request is immediately granted if the
1526 * requested mode is compatible with the modes of all other granted
1530 if (queue_conflict(&r->res_convertqueue, lkb))
1534 * 6-5: But the default algorithm for deciding whether to grant or
1535 * queue conversion requests does not by itself guarantee that such
1536 * requests are serviced on a "first come first serve" basis. This, in
1537 * turn, can lead to a phenomenon known as "indefinate postponement".
1539 * 6-7: This issue is dealt with by using the optional QUECVT flag with
1540 * the system service employed to request a lock conversion. This flag
1541 * forces certain conversion requests to be queued, even if they are
1542 * compatible with the granted modes of other locks on the same
1543 * resource. Thus, the use of this flag results in conversion requests
1544 * being ordered on a "first come first servce" basis.
1546 * DCT: This condition is all about new conversions being able to occur
1547 * "in place" while the lock remains on the granted queue (assuming
1548 * nothing else conflicts.) IOW if QUECVT isn't set, a conversion
1549 * doesn't _have_ to go onto the convert queue where it's processed in
1550 * order. The "now" variable is necessary to distinguish converts
1551 * being received and processed for the first time now, because once a
1552 * convert is moved to the conversion queue the condition below applies
1553 * requiring fifo granting.
1556 if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT))
1560 * The NOORDER flag is set to avoid the standard vms rules on grant
1564 if (lkb->lkb_exflags & DLM_LKF_NOORDER)
1568 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
1569 * granted until all other conversion requests ahead of it are granted
1573 if (!now && conv && first_in_list(lkb, &r->res_convertqueue))
1577 * 6-4: By default, a new request is immediately granted only if all
1578 * three of the following conditions are satisfied when the request is
1580 * - The queue of ungranted conversion requests for the resource is
1582 * - The queue of ungranted new requests for the resource is empty.
1583 * - The mode of the new request is compatible with the most
1584 * restrictive mode of all granted locks on the resource.
1587 if (now && !conv && list_empty(&r->res_convertqueue) &&
1588 list_empty(&r->res_waitqueue))
1592 * 6-4: Once a lock request is in the queue of ungranted new requests,
1593 * it cannot be granted until the queue of ungranted conversion
1594 * requests is empty, all ungranted new requests ahead of it are
1595 * granted and/or canceled, and it is compatible with the granted mode
1596 * of the most restrictive lock granted on the resource.
1599 if (!now && !conv && list_empty(&r->res_convertqueue) &&
1600 first_in_list(lkb, &r->res_waitqueue))
1606 static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
1610 int8_t alt = 0, rqmode = lkb->lkb_rqmode;
1611 int8_t is_convert = (lkb->lkb_grmode != DLM_LOCK_IV);
1616 rv = _can_be_granted(r, lkb, now);
1621 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
1622 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
1623 * cancels one of the locks.
1626 if (is_convert && can_be_queued(lkb) &&
1627 conversion_deadlock_detect(r, lkb)) {
1628 if (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) {
1629 lkb->lkb_grmode = DLM_LOCK_NL;
1630 lkb->lkb_sbflags |= DLM_SBF_DEMOTED;
1631 } else if (!(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1635 log_print("can_be_granted deadlock %x now %d",
1644 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
1645 * to grant a request in a mode other than the normal rqmode. It's a
1646 * simple way to provide a big optimization to applications that can
1650 if (rqmode != DLM_LOCK_PR && (lkb->lkb_exflags & DLM_LKF_ALTPR))
1652 else if (rqmode != DLM_LOCK_CW && (lkb->lkb_exflags & DLM_LKF_ALTCW))
1656 lkb->lkb_rqmode = alt;
1657 rv = _can_be_granted(r, lkb, now);
1659 lkb->lkb_sbflags |= DLM_SBF_ALTMODE;
1661 lkb->lkb_rqmode = rqmode;
1667 /* FIXME: I don't think that can_be_granted() can/will demote or find deadlock
1668 for locks pending on the convert list. Once verified (watch for these
1669 log_prints), we should be able to just call _can_be_granted() and not
1670 bother with the demote/deadlk cases here (and there's no easy way to deal
1671 with a deadlk here, we'd have to generate something like grant_lock with
1672 the deadlk error.) */
1674 /* Returns the highest requested mode of all blocked conversions; sets
1675 cw if there's a blocked conversion to DLM_LOCK_CW. */
1677 static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw)
1679 struct dlm_lkb *lkb, *s;
1680 int hi, demoted, quit, grant_restart, demote_restart;
1689 list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) {
1690 demoted = is_demoted(lkb);
1693 if (can_be_granted(r, lkb, 0, &deadlk)) {
1694 grant_lock_pending(r, lkb);
1699 if (!demoted && is_demoted(lkb)) {
1700 log_print("WARN: pending demoted %x node %d %s",
1701 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1707 log_print("WARN: pending deadlock %x node %d %s",
1708 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1713 hi = max_t(int, lkb->lkb_rqmode, hi);
1715 if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
1721 if (demote_restart && !quit) {
1726 return max_t(int, high, hi);
1729 static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw)
1731 struct dlm_lkb *lkb, *s;
1733 list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
1734 if (can_be_granted(r, lkb, 0, NULL))
1735 grant_lock_pending(r, lkb);
1737 high = max_t(int, lkb->lkb_rqmode, high);
1738 if (lkb->lkb_rqmode == DLM_LOCK_CW)
1746 /* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
1747 on either the convert or waiting queue.
1748 high is the largest rqmode of all locks blocked on the convert or
1751 static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
1753 if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
1754 if (gr->lkb_highbast < DLM_LOCK_EX)
1759 if (gr->lkb_highbast < high &&
1760 !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
1765 static void grant_pending_locks(struct dlm_rsb *r)
1767 struct dlm_lkb *lkb, *s;
1768 int high = DLM_LOCK_IV;
1771 DLM_ASSERT(is_master(r), dlm_dump_rsb(r););
1773 high = grant_pending_convert(r, high, &cw);
1774 high = grant_pending_wait(r, high, &cw);
1776 if (high == DLM_LOCK_IV)
1780 * If there are locks left on the wait/convert queue then send blocking
1781 * ASTs to granted locks based on the largest requested mode (high)
1785 list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
1786 if (lkb->lkb_bastfn && lock_requires_bast(lkb, high, cw)) {
1787 if (cw && high == DLM_LOCK_PR &&
1788 lkb->lkb_grmode == DLM_LOCK_PR)
1789 queue_bast(r, lkb, DLM_LOCK_CW);
1791 queue_bast(r, lkb, high);
1792 lkb->lkb_highbast = high;
1797 static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
1799 if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
1800 (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
1801 if (gr->lkb_highbast < DLM_LOCK_EX)
1806 if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
1811 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
1812 struct dlm_lkb *lkb)
1816 list_for_each_entry(gr, head, lkb_statequeue) {
1817 if (gr->lkb_bastfn && modes_require_bast(gr, lkb)) {
1818 queue_bast(r, gr, lkb->lkb_rqmode);
1819 gr->lkb_highbast = lkb->lkb_rqmode;
1824 static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb)
1826 send_bast_queue(r, &r->res_grantqueue, lkb);
1829 static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb)
1831 send_bast_queue(r, &r->res_grantqueue, lkb);
1832 send_bast_queue(r, &r->res_convertqueue, lkb);
1835 /* set_master(r, lkb) -- set the master nodeid of a resource
1837 The purpose of this function is to set the nodeid field in the given
1838 lkb using the nodeid field in the given rsb. If the rsb's nodeid is
1839 known, it can just be copied to the lkb and the function will return
1840 0. If the rsb's nodeid is _not_ known, it needs to be looked up
1841 before it can be copied to the lkb.
1843 When the rsb nodeid is being looked up remotely, the initial lkb
1844 causing the lookup is kept on the ls_waiters list waiting for the
1845 lookup reply. Other lkb's waiting for the same rsb lookup are kept
1846 on the rsb's res_lookup list until the master is verified.
1849 0: nodeid is set in rsb/lkb and the caller should go ahead and use it
1850 1: the rsb master is not available and the lkb has been placed on
1854 static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb)
1856 struct dlm_ls *ls = r->res_ls;
1857 int i, error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid();
1859 if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) {
1860 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
1861 r->res_first_lkid = lkb->lkb_id;
1862 lkb->lkb_nodeid = r->res_nodeid;
1866 if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) {
1867 list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup);
1871 if (r->res_nodeid == 0) {
1872 lkb->lkb_nodeid = 0;
1876 if (r->res_nodeid > 0) {
1877 lkb->lkb_nodeid = r->res_nodeid;
1881 DLM_ASSERT(r->res_nodeid == -1, dlm_dump_rsb(r););
1883 dir_nodeid = dlm_dir_nodeid(r);
1885 if (dir_nodeid != our_nodeid) {
1886 r->res_first_lkid = lkb->lkb_id;
1887 send_lookup(r, lkb);
1891 for (i = 0; i < 2; i++) {
1892 /* It's possible for dlm_scand to remove an old rsb for
1893 this same resource from the toss list, us to create
1894 a new one, look up the master locally, and find it
1895 already exists just before dlm_scand does the
1896 dir_remove() on the previous rsb. */
1898 error = dlm_dir_lookup(ls, our_nodeid, r->res_name,
1899 r->res_length, &ret_nodeid);
1902 log_debug(ls, "dir_lookup error %d %s", error, r->res_name);
1905 if (error && error != -EEXIST)
1908 if (ret_nodeid == our_nodeid) {
1909 r->res_first_lkid = 0;
1911 lkb->lkb_nodeid = 0;
1913 r->res_first_lkid = lkb->lkb_id;
1914 r->res_nodeid = ret_nodeid;
1915 lkb->lkb_nodeid = ret_nodeid;
1920 static void process_lookup_list(struct dlm_rsb *r)
1922 struct dlm_lkb *lkb, *safe;
1924 list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
1925 list_del_init(&lkb->lkb_rsb_lookup);
1926 _request_lock(r, lkb);
1931 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
1933 static void confirm_master(struct dlm_rsb *r, int error)
1935 struct dlm_lkb *lkb;
1937 if (!r->res_first_lkid)
1943 r->res_first_lkid = 0;
1944 process_lookup_list(r);
1950 /* the remote request failed and won't be retried (it was
1951 a NOQUEUE, or has been canceled/unlocked); make a waiting
1952 lkb the first_lkid */
1954 r->res_first_lkid = 0;
1956 if (!list_empty(&r->res_lookup)) {
1957 lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
1959 list_del_init(&lkb->lkb_rsb_lookup);
1960 r->res_first_lkid = lkb->lkb_id;
1961 _request_lock(r, lkb);
1966 log_error(r->res_ls, "confirm_master unknown error %d", error);
1970 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
1971 int namelen, unsigned long timeout_cs,
1972 void (*ast) (void *astparam),
1974 void (*bast) (void *astparam, int mode),
1975 struct dlm_args *args)
1979 /* check for invalid arg usage */
1981 if (mode < 0 || mode > DLM_LOCK_EX)
1984 if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
1987 if (flags & DLM_LKF_CANCEL)
1990 if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
1993 if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
1996 if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
1999 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
2002 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
2005 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2008 if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2014 if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2017 if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2020 /* these args will be copied to the lkb in validate_lock_args,
2021 it cannot be done now because when converting locks, fields in
2022 an active lkb cannot be modified before locking the rsb */
2024 args->flags = flags;
2026 args->astparam = astparam;
2027 args->bastfn = bast;
2028 args->timeout = timeout_cs;
2036 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2038 if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2039 DLM_LKF_FORCEUNLOCK))
2042 if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2045 args->flags = flags;
2046 args->astparam = astarg;
2050 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2051 struct dlm_args *args)
2055 if (args->flags & DLM_LKF_CONVERT) {
2056 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2059 if (args->flags & DLM_LKF_QUECVT &&
2060 !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2064 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2067 if (lkb->lkb_wait_type)
2070 if (is_overlap(lkb))
2074 lkb->lkb_exflags = args->flags;
2075 lkb->lkb_sbflags = 0;
2076 lkb->lkb_astfn = args->astfn;
2077 lkb->lkb_astparam = args->astparam;
2078 lkb->lkb_bastfn = args->bastfn;
2079 lkb->lkb_rqmode = args->mode;
2080 lkb->lkb_lksb = args->lksb;
2081 lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2082 lkb->lkb_ownpid = (int) current->pid;
2083 lkb->lkb_timeout_cs = args->timeout;
2089 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2092 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2093 because there may be a lookup in progress and it's valid to do
2094 cancel/unlockf on it */
2096 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2098 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
2101 if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2102 log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2107 /* an lkb may still exist even though the lock is EOL'ed due to a
2108 cancel, unlock or failed noqueue request; an app can't use these
2109 locks; return same error as if the lkid had not been found at all */
2111 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2112 log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2117 /* an lkb may be waiting for an rsb lookup to complete where the
2118 lookup was initiated by another lock */
2120 if (!list_empty(&lkb->lkb_rsb_lookup)) {
2121 if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
2122 log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2123 list_del_init(&lkb->lkb_rsb_lookup);
2124 queue_cast(lkb->lkb_resource, lkb,
2125 args->flags & DLM_LKF_CANCEL ?
2126 -DLM_ECANCEL : -DLM_EUNLOCK);
2127 unhold_lkb(lkb); /* undoes create_lkb() */
2129 /* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2134 /* cancel not allowed with another cancel/unlock in progress */
2136 if (args->flags & DLM_LKF_CANCEL) {
2137 if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2140 if (is_overlap(lkb))
2143 /* don't let scand try to do a cancel */
2146 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2147 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2152 switch (lkb->lkb_wait_type) {
2153 case DLM_MSG_LOOKUP:
2154 case DLM_MSG_REQUEST:
2155 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2158 case DLM_MSG_UNLOCK:
2159 case DLM_MSG_CANCEL:
2162 /* add_to_waiters() will set OVERLAP_CANCEL */
2166 /* do we need to allow a force-unlock if there's a normal unlock
2167 already in progress? in what conditions could the normal unlock
2168 fail such that we'd want to send a force-unlock to be sure? */
2170 if (args->flags & DLM_LKF_FORCEUNLOCK) {
2171 if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2174 if (is_overlap_unlock(lkb))
2177 /* don't let scand try to do a cancel */
2180 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2181 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2186 switch (lkb->lkb_wait_type) {
2187 case DLM_MSG_LOOKUP:
2188 case DLM_MSG_REQUEST:
2189 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2192 case DLM_MSG_UNLOCK:
2195 /* add_to_waiters() will set OVERLAP_UNLOCK */
2199 /* normal unlock not allowed if there's any op in progress */
2201 if (lkb->lkb_wait_type || lkb->lkb_wait_count)
2205 /* an overlapping op shouldn't blow away exflags from other op */
2206 lkb->lkb_exflags |= args->flags;
2207 lkb->lkb_sbflags = 0;
2208 lkb->lkb_astparam = args->astparam;
2212 log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
2213 lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
2214 args->flags, lkb->lkb_wait_type,
2215 lkb->lkb_resource->res_name);
2220 * Four stage 4 varieties:
2221 * do_request(), do_convert(), do_unlock(), do_cancel()
2222 * These are called on the master node for the given lock and
2223 * from the central locking logic.
2226 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2230 if (can_be_granted(r, lkb, 1, NULL)) {
2232 queue_cast(r, lkb, 0);
2236 if (can_be_queued(lkb)) {
2237 error = -EINPROGRESS;
2238 add_lkb(r, lkb, DLM_LKSTS_WAITING);
2239 send_blocking_asts(r, lkb);
2245 if (force_blocking_asts(lkb))
2246 send_blocking_asts_all(r, lkb);
2247 queue_cast(r, lkb, -EAGAIN);
2253 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2258 /* changing an existing lock may allow others to be granted */
2260 if (can_be_granted(r, lkb, 1, &deadlk)) {
2262 queue_cast(r, lkb, 0);
2263 grant_pending_locks(r);
2267 /* can_be_granted() detected that this lock would block in a conversion
2268 deadlock, so we leave it on the granted queue and return EDEADLK in
2269 the ast for the convert. */
2272 /* it's left on the granted queue */
2273 log_debug(r->res_ls, "deadlock %x node %d sts%d g%d r%d %s",
2274 lkb->lkb_id, lkb->lkb_nodeid, lkb->lkb_status,
2275 lkb->lkb_grmode, lkb->lkb_rqmode, r->res_name);
2276 revert_lock(r, lkb);
2277 queue_cast(r, lkb, -EDEADLK);
2282 /* is_demoted() means the can_be_granted() above set the grmode
2283 to NL, and left us on the granted queue. This auto-demotion
2284 (due to CONVDEADLK) might mean other locks, and/or this lock, are
2285 now grantable. We have to try to grant other converting locks
2286 before we try again to grant this one. */
2288 if (is_demoted(lkb)) {
2289 grant_pending_convert(r, DLM_LOCK_IV, NULL);
2290 if (_can_be_granted(r, lkb, 1)) {
2292 queue_cast(r, lkb, 0);
2293 grant_pending_locks(r);
2296 /* else fall through and move to convert queue */
2299 if (can_be_queued(lkb)) {
2300 error = -EINPROGRESS;
2302 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2303 send_blocking_asts(r, lkb);
2309 if (force_blocking_asts(lkb))
2310 send_blocking_asts_all(r, lkb);
2311 queue_cast(r, lkb, -EAGAIN);
2317 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2319 remove_lock(r, lkb);
2320 queue_cast(r, lkb, -DLM_EUNLOCK);
2321 grant_pending_locks(r);
2322 return -DLM_EUNLOCK;
2325 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
2327 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2331 error = revert_lock(r, lkb);
2333 queue_cast(r, lkb, -DLM_ECANCEL);
2334 grant_pending_locks(r);
2335 return -DLM_ECANCEL;
2341 * Four stage 3 varieties:
2342 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2345 /* add a new lkb to a possibly new rsb, called by requesting process */
2347 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2351 /* set_master: sets lkb nodeid from r */
2353 error = set_master(r, lkb);
2362 /* receive_request() calls do_request() on remote node */
2363 error = send_request(r, lkb);
2365 error = do_request(r, lkb);
2370 /* change some property of an existing lkb, e.g. mode */
2372 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2377 /* receive_convert() calls do_convert() on remote node */
2378 error = send_convert(r, lkb);
2380 error = do_convert(r, lkb);
2385 /* remove an existing lkb from the granted queue */
2387 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2392 /* receive_unlock() calls do_unlock() on remote node */
2393 error = send_unlock(r, lkb);
2395 error = do_unlock(r, lkb);
2400 /* remove an existing lkb from the convert or wait queue */
2402 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2407 /* receive_cancel() calls do_cancel() on remote node */
2408 error = send_cancel(r, lkb);
2410 error = do_cancel(r, lkb);
2416 * Four stage 2 varieties:
2417 * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2420 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
2421 int len, struct dlm_args *args)
2426 error = validate_lock_args(ls, lkb, args);
2430 error = find_rsb(ls, name, len, R_CREATE, &r);
2437 lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
2439 error = _request_lock(r, lkb);
2448 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2449 struct dlm_args *args)
2454 r = lkb->lkb_resource;
2459 error = validate_lock_args(ls, lkb, args);
2463 error = _convert_lock(r, lkb);
2470 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2471 struct dlm_args *args)
2476 r = lkb->lkb_resource;
2481 error = validate_unlock_args(lkb, args);
2485 error = _unlock_lock(r, lkb);
2492 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2493 struct dlm_args *args)
2498 r = lkb->lkb_resource;
2503 error = validate_unlock_args(lkb, args);
2507 error = _cancel_lock(r, lkb);
2515 * Two stage 1 varieties: dlm_lock() and dlm_unlock()
2518 int dlm_lock(dlm_lockspace_t *lockspace,
2520 struct dlm_lksb *lksb,
2523 unsigned int namelen,
2524 uint32_t parent_lkid,
2525 void (*ast) (void *astarg),
2527 void (*bast) (void *astarg, int mode))
2530 struct dlm_lkb *lkb;
2531 struct dlm_args args;
2532 int error, convert = flags & DLM_LKF_CONVERT;
2534 ls = dlm_find_lockspace_local(lockspace);
2538 dlm_lock_recovery(ls);
2541 error = find_lkb(ls, lksb->sb_lkid, &lkb);
2543 error = create_lkb(ls, &lkb);
2548 error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
2549 astarg, bast, &args);
2554 error = convert_lock(ls, lkb, &args);
2556 error = request_lock(ls, lkb, name, namelen, &args);
2558 if (error == -EINPROGRESS)
2561 if (convert || error)
2563 if (error == -EAGAIN || error == -EDEADLK)
2566 dlm_unlock_recovery(ls);
2567 dlm_put_lockspace(ls);
2571 int dlm_unlock(dlm_lockspace_t *lockspace,
2574 struct dlm_lksb *lksb,
2578 struct dlm_lkb *lkb;
2579 struct dlm_args args;
2582 ls = dlm_find_lockspace_local(lockspace);
2586 dlm_lock_recovery(ls);
2588 error = find_lkb(ls, lkid, &lkb);
2592 error = set_unlock_args(flags, astarg, &args);
2596 if (flags & DLM_LKF_CANCEL)
2597 error = cancel_lock(ls, lkb, &args);
2599 error = unlock_lock(ls, lkb, &args);
2601 if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
2603 if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
2608 dlm_unlock_recovery(ls);
2609 dlm_put_lockspace(ls);
2614 * send/receive routines for remote operations and replies
2618 * send_request receive_request
2619 * send_convert receive_convert
2620 * send_unlock receive_unlock
2621 * send_cancel receive_cancel
2622 * send_grant receive_grant
2623 * send_bast receive_bast
2624 * send_lookup receive_lookup
2625 * send_remove receive_remove
2628 * receive_request_reply send_request_reply
2629 * receive_convert_reply send_convert_reply
2630 * receive_unlock_reply send_unlock_reply
2631 * receive_cancel_reply send_cancel_reply
2632 * receive_lookup_reply send_lookup_reply
2635 static int _create_message(struct dlm_ls *ls, int mb_len,
2636 int to_nodeid, int mstype,
2637 struct dlm_message **ms_ret,
2638 struct dlm_mhandle **mh_ret)
2640 struct dlm_message *ms;
2641 struct dlm_mhandle *mh;
2644 /* get_buffer gives us a message handle (mh) that we need to
2645 pass into lowcomms_commit and a message buffer (mb) that we
2646 write our data into */
2648 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
2652 memset(mb, 0, mb_len);
2654 ms = (struct dlm_message *) mb;
2656 ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
2657 ms->m_header.h_lockspace = ls->ls_global_id;
2658 ms->m_header.h_nodeid = dlm_our_nodeid();
2659 ms->m_header.h_length = mb_len;
2660 ms->m_header.h_cmd = DLM_MSG;
2662 ms->m_type = mstype;
2669 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
2670 int to_nodeid, int mstype,
2671 struct dlm_message **ms_ret,
2672 struct dlm_mhandle **mh_ret)
2674 int mb_len = sizeof(struct dlm_message);
2677 case DLM_MSG_REQUEST:
2678 case DLM_MSG_LOOKUP:
2679 case DLM_MSG_REMOVE:
2680 mb_len += r->res_length;
2682 case DLM_MSG_CONVERT:
2683 case DLM_MSG_UNLOCK:
2684 case DLM_MSG_REQUEST_REPLY:
2685 case DLM_MSG_CONVERT_REPLY:
2687 if (lkb && lkb->lkb_lvbptr)
2688 mb_len += r->res_ls->ls_lvblen;
2692 return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
2696 /* further lowcomms enhancements or alternate implementations may make
2697 the return value from this function useful at some point */
2699 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
2701 dlm_message_out(ms);
2702 dlm_lowcomms_commit_buffer(mh);
2706 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
2707 struct dlm_message *ms)
2709 ms->m_nodeid = lkb->lkb_nodeid;
2710 ms->m_pid = lkb->lkb_ownpid;
2711 ms->m_lkid = lkb->lkb_id;
2712 ms->m_remid = lkb->lkb_remid;
2713 ms->m_exflags = lkb->lkb_exflags;
2714 ms->m_sbflags = lkb->lkb_sbflags;
2715 ms->m_flags = lkb->lkb_flags;
2716 ms->m_lvbseq = lkb->lkb_lvbseq;
2717 ms->m_status = lkb->lkb_status;
2718 ms->m_grmode = lkb->lkb_grmode;
2719 ms->m_rqmode = lkb->lkb_rqmode;
2720 ms->m_hash = r->res_hash;
2722 /* m_result and m_bastmode are set from function args,
2723 not from lkb fields */
2725 if (lkb->lkb_bastfn)
2726 ms->m_asts |= AST_BAST;
2728 ms->m_asts |= AST_COMP;
2730 /* compare with switch in create_message; send_remove() doesn't
2733 switch (ms->m_type) {
2734 case DLM_MSG_REQUEST:
2735 case DLM_MSG_LOOKUP:
2736 memcpy(ms->m_extra, r->res_name, r->res_length);
2738 case DLM_MSG_CONVERT:
2739 case DLM_MSG_UNLOCK:
2740 case DLM_MSG_REQUEST_REPLY:
2741 case DLM_MSG_CONVERT_REPLY:
2743 if (!lkb->lkb_lvbptr)
2745 memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
2750 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
2752 struct dlm_message *ms;
2753 struct dlm_mhandle *mh;
2754 int to_nodeid, error;
2756 error = add_to_waiters(lkb, mstype);
2760 to_nodeid = r->res_nodeid;
2762 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2766 send_args(r, lkb, ms);
2768 error = send_message(mh, ms);
2774 remove_from_waiters(lkb, msg_reply_type(mstype));
2778 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2780 return send_common(r, lkb, DLM_MSG_REQUEST);
2783 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2787 error = send_common(r, lkb, DLM_MSG_CONVERT);
2789 /* down conversions go without a reply from the master */
2790 if (!error && down_conversion(lkb)) {
2791 remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
2792 r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
2793 r->res_ls->ls_stub_ms.m_result = 0;
2794 r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
2795 __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
2801 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
2802 MASTER_UNCERTAIN to force the next request on the rsb to confirm
2803 that the master is still correct. */
2805 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2807 return send_common(r, lkb, DLM_MSG_UNLOCK);
2810 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2812 return send_common(r, lkb, DLM_MSG_CANCEL);
2815 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
2817 struct dlm_message *ms;
2818 struct dlm_mhandle *mh;
2819 int to_nodeid, error;
2821 to_nodeid = lkb->lkb_nodeid;
2823 error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
2827 send_args(r, lkb, ms);
2831 error = send_message(mh, ms);
2836 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
2838 struct dlm_message *ms;
2839 struct dlm_mhandle *mh;
2840 int to_nodeid, error;
2842 to_nodeid = lkb->lkb_nodeid;
2844 error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
2848 send_args(r, lkb, ms);
2850 ms->m_bastmode = mode;
2852 error = send_message(mh, ms);
2857 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
2859 struct dlm_message *ms;
2860 struct dlm_mhandle *mh;
2861 int to_nodeid, error;
2863 error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
2867 to_nodeid = dlm_dir_nodeid(r);
2869 error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
2873 send_args(r, lkb, ms);
2875 error = send_message(mh, ms);
2881 remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
2885 static int send_remove(struct dlm_rsb *r)
2887 struct dlm_message *ms;
2888 struct dlm_mhandle *mh;
2889 int to_nodeid, error;
2891 to_nodeid = dlm_dir_nodeid(r);
2893 error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
2897 memcpy(ms->m_extra, r->res_name, r->res_length);
2898 ms->m_hash = r->res_hash;
2900 error = send_message(mh, ms);
2905 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
2908 struct dlm_message *ms;
2909 struct dlm_mhandle *mh;
2910 int to_nodeid, error;
2912 to_nodeid = lkb->lkb_nodeid;
2914 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2918 send_args(r, lkb, ms);
2922 error = send_message(mh, ms);
2927 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2929 return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
2932 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2934 return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
2937 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2939 return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
2942 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2944 return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
2947 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
2948 int ret_nodeid, int rv)
2950 struct dlm_rsb *r = &ls->ls_stub_rsb;
2951 struct dlm_message *ms;
2952 struct dlm_mhandle *mh;
2953 int error, nodeid = ms_in->m_header.h_nodeid;
2955 error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
2959 ms->m_lkid = ms_in->m_lkid;
2961 ms->m_nodeid = ret_nodeid;
2963 error = send_message(mh, ms);
2968 /* which args we save from a received message depends heavily on the type
2969 of message, unlike the send side where we can safely send everything about
2970 the lkb for any type of message */
2972 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
2974 lkb->lkb_exflags = ms->m_exflags;
2975 lkb->lkb_sbflags = ms->m_sbflags;
2976 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2977 (ms->m_flags & 0x0000FFFF);
2980 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
2982 lkb->lkb_sbflags = ms->m_sbflags;
2983 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2984 (ms->m_flags & 0x0000FFFF);
2987 static int receive_extralen(struct dlm_message *ms)
2989 return (ms->m_header.h_length - sizeof(struct dlm_message));
2992 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
2993 struct dlm_message *ms)
2997 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
2998 if (!lkb->lkb_lvbptr)
2999 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3000 if (!lkb->lkb_lvbptr)
3002 len = receive_extralen(ms);
3003 if (len > DLM_RESNAME_MAXLEN)
3004 len = DLM_RESNAME_MAXLEN;
3005 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
3010 static void fake_bastfn(void *astparam, int mode)
3012 log_print("fake_bastfn should not be called");
3015 static void fake_astfn(void *astparam)
3017 log_print("fake_astfn should not be called");
3020 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3021 struct dlm_message *ms)
3023 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3024 lkb->lkb_ownpid = ms->m_pid;
3025 lkb->lkb_remid = ms->m_lkid;
3026 lkb->lkb_grmode = DLM_LOCK_IV;
3027 lkb->lkb_rqmode = ms->m_rqmode;
3029 lkb->lkb_bastfn = (ms->m_asts & AST_BAST) ? &fake_bastfn : NULL;
3030 lkb->lkb_astfn = (ms->m_asts & AST_COMP) ? &fake_astfn : NULL;
3032 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3033 /* lkb was just created so there won't be an lvb yet */
3034 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3035 if (!lkb->lkb_lvbptr)
3042 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3043 struct dlm_message *ms)
3045 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3048 if (receive_lvb(ls, lkb, ms))
3051 lkb->lkb_rqmode = ms->m_rqmode;
3052 lkb->lkb_lvbseq = ms->m_lvbseq;
3057 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3058 struct dlm_message *ms)
3060 if (receive_lvb(ls, lkb, ms))
3065 /* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3066 uses to send a reply and that the remote end uses to process the reply. */
3068 static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3070 struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3071 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3072 lkb->lkb_remid = ms->m_lkid;
3075 /* This is called after the rsb is locked so that we can safely inspect
3076 fields in the lkb. */
3078 static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3080 int from = ms->m_header.h_nodeid;
3083 switch (ms->m_type) {
3084 case DLM_MSG_CONVERT:
3085 case DLM_MSG_UNLOCK:
3086 case DLM_MSG_CANCEL:
3087 if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3091 case DLM_MSG_CONVERT_REPLY:
3092 case DLM_MSG_UNLOCK_REPLY:
3093 case DLM_MSG_CANCEL_REPLY:
3096 if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3100 case DLM_MSG_REQUEST_REPLY:
3101 if (!is_process_copy(lkb))
3103 else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
3112 log_error(lkb->lkb_resource->res_ls,
3113 "ignore invalid message %d from %d %x %x %x %d",
3114 ms->m_type, from, lkb->lkb_id, lkb->lkb_remid,
3115 lkb->lkb_flags, lkb->lkb_nodeid);
3119 static void receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3121 struct dlm_lkb *lkb;
3125 error = create_lkb(ls, &lkb);
3129 receive_flags(lkb, ms);
3130 lkb->lkb_flags |= DLM_IFL_MSTCPY;
3131 error = receive_request_args(ls, lkb, ms);
3137 namelen = receive_extralen(ms);
3139 error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r);
3148 error = do_request(r, lkb);
3149 send_request_reply(r, lkb, error);
3154 if (error == -EINPROGRESS)
3161 setup_stub_lkb(ls, ms);
3162 send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3165 static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3167 struct dlm_lkb *lkb;
3169 int error, reply = 1;
3171 error = find_lkb(ls, ms->m_remid, &lkb);
3175 r = lkb->lkb_resource;
3180 error = validate_message(lkb, ms);
3184 receive_flags(lkb, ms);
3185 error = receive_convert_args(ls, lkb, ms);
3188 reply = !down_conversion(lkb);
3190 error = do_convert(r, lkb);
3193 send_convert_reply(r, lkb, error);
3201 setup_stub_lkb(ls, ms);
3202 send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3205 static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3207 struct dlm_lkb *lkb;
3211 error = find_lkb(ls, ms->m_remid, &lkb);
3215 r = lkb->lkb_resource;
3220 error = validate_message(lkb, ms);
3224 receive_flags(lkb, ms);
3225 error = receive_unlock_args(ls, lkb, ms);
3229 error = do_unlock(r, lkb);
3231 send_unlock_reply(r, lkb, error);
3239 setup_stub_lkb(ls, ms);
3240 send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3243 static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
3245 struct dlm_lkb *lkb;
3249 error = find_lkb(ls, ms->m_remid, &lkb);
3253 receive_flags(lkb, ms);
3255 r = lkb->lkb_resource;
3260 error = validate_message(lkb, ms);
3264 error = do_cancel(r, lkb);
3265 send_cancel_reply(r, lkb, error);
3273 setup_stub_lkb(ls, ms);
3274 send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3277 static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
3279 struct dlm_lkb *lkb;
3283 error = find_lkb(ls, ms->m_remid, &lkb);
3285 log_debug(ls, "receive_grant from %d no lkb %x",
3286 ms->m_header.h_nodeid, ms->m_remid);
3290 r = lkb->lkb_resource;
3295 error = validate_message(lkb, ms);
3299 receive_flags_reply(lkb, ms);
3300 if (is_altmode(lkb))
3301 munge_altmode(lkb, ms);
3302 grant_lock_pc(r, lkb, ms);
3303 queue_cast(r, lkb, 0);
3310 static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
3312 struct dlm_lkb *lkb;
3316 error = find_lkb(ls, ms->m_remid, &lkb);
3318 log_debug(ls, "receive_bast from %d no lkb %x",
3319 ms->m_header.h_nodeid, ms->m_remid);
3323 r = lkb->lkb_resource;
3328 error = validate_message(lkb, ms);
3332 queue_bast(r, lkb, ms->m_bastmode);
3339 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
3341 int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid;
3343 from_nodeid = ms->m_header.h_nodeid;
3344 our_nodeid = dlm_our_nodeid();
3346 len = receive_extralen(ms);
3348 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3349 if (dir_nodeid != our_nodeid) {
3350 log_error(ls, "lookup dir_nodeid %d from %d",
3351 dir_nodeid, from_nodeid);
3357 error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid);
3359 /* Optimization: we're master so treat lookup as a request */
3360 if (!error && ret_nodeid == our_nodeid) {
3361 receive_request(ls, ms);
3365 send_lookup_reply(ls, ms, ret_nodeid, error);
3368 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
3370 int len, dir_nodeid, from_nodeid;
3372 from_nodeid = ms->m_header.h_nodeid;
3374 len = receive_extralen(ms);
3376 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3377 if (dir_nodeid != dlm_our_nodeid()) {
3378 log_error(ls, "remove dir entry dir_nodeid %d from %d",
3379 dir_nodeid, from_nodeid);
3383 dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
3386 static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
3388 do_purge(ls, ms->m_nodeid, ms->m_pid);
3391 static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
3393 struct dlm_lkb *lkb;
3395 int error, mstype, result;
3397 error = find_lkb(ls, ms->m_remid, &lkb);
3399 log_debug(ls, "receive_request_reply from %d no lkb %x",
3400 ms->m_header.h_nodeid, ms->m_remid);
3404 r = lkb->lkb_resource;
3408 error = validate_message(lkb, ms);
3412 mstype = lkb->lkb_wait_type;
3413 error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
3417 /* Optimization: the dir node was also the master, so it took our
3418 lookup as a request and sent request reply instead of lookup reply */
3419 if (mstype == DLM_MSG_LOOKUP) {
3420 r->res_nodeid = ms->m_header.h_nodeid;
3421 lkb->lkb_nodeid = r->res_nodeid;
3424 /* this is the value returned from do_request() on the master */
3425 result = ms->m_result;
3429 /* request would block (be queued) on remote master */
3430 queue_cast(r, lkb, -EAGAIN);
3431 confirm_master(r, -EAGAIN);
3432 unhold_lkb(lkb); /* undoes create_lkb() */
3437 /* request was queued or granted on remote master */
3438 receive_flags_reply(lkb, ms);
3439 lkb->lkb_remid = ms->m_lkid;
3440 if (is_altmode(lkb))
3441 munge_altmode(lkb, ms);
3443 add_lkb(r, lkb, DLM_LKSTS_WAITING);
3446 grant_lock_pc(r, lkb, ms);
3447 queue_cast(r, lkb, 0);
3449 confirm_master(r, result);
3454 /* find_rsb failed to find rsb or rsb wasn't master */
3455 log_debug(ls, "receive_request_reply %x %x master diff %d %d",
3456 lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
3458 lkb->lkb_nodeid = -1;
3460 if (is_overlap(lkb)) {
3461 /* we'll ignore error in cancel/unlock reply */
3462 queue_cast_overlap(r, lkb);
3463 confirm_master(r, result);
3464 unhold_lkb(lkb); /* undoes create_lkb() */
3466 _request_lock(r, lkb);
3470 log_error(ls, "receive_request_reply %x error %d",
3471 lkb->lkb_id, result);
3474 if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
3475 log_debug(ls, "receive_request_reply %x result %d unlock",
3476 lkb->lkb_id, result);
3477 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3478 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3479 send_unlock(r, lkb);
3480 } else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
3481 log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
3482 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3483 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3484 send_cancel(r, lkb);
3486 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3487 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3495 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3496 struct dlm_message *ms)
3498 /* this is the value returned from do_convert() on the master */
3499 switch (ms->m_result) {
3501 /* convert would block (be queued) on remote master */
3502 queue_cast(r, lkb, -EAGAIN);
3506 receive_flags_reply(lkb, ms);
3507 revert_lock_pc(r, lkb);
3508 queue_cast(r, lkb, -EDEADLK);
3512 /* convert was queued on remote master */
3513 receive_flags_reply(lkb, ms);
3514 if (is_demoted(lkb))
3515 munge_demoted(lkb, ms);
3517 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3522 /* convert was granted on remote master */
3523 receive_flags_reply(lkb, ms);
3524 if (is_demoted(lkb))
3525 munge_demoted(lkb, ms);
3526 grant_lock_pc(r, lkb, ms);
3527 queue_cast(r, lkb, 0);
3531 log_error(r->res_ls, "receive_convert_reply %x error %d",
3532 lkb->lkb_id, ms->m_result);
3536 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3538 struct dlm_rsb *r = lkb->lkb_resource;
3544 error = validate_message(lkb, ms);
3548 /* stub reply can happen with waiters_mutex held */
3549 error = remove_from_waiters_ms(lkb, ms);
3553 __receive_convert_reply(r, lkb, ms);
3559 static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
3561 struct dlm_lkb *lkb;
3564 error = find_lkb(ls, ms->m_remid, &lkb);
3566 log_debug(ls, "receive_convert_reply from %d no lkb %x",
3567 ms->m_header.h_nodeid, ms->m_remid);
3571 _receive_convert_reply(lkb, ms);
3575 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3577 struct dlm_rsb *r = lkb->lkb_resource;
3583 error = validate_message(lkb, ms);
3587 /* stub reply can happen with waiters_mutex held */
3588 error = remove_from_waiters_ms(lkb, ms);
3592 /* this is the value returned from do_unlock() on the master */
3594 switch (ms->m_result) {
3596 receive_flags_reply(lkb, ms);
3597 remove_lock_pc(r, lkb);
3598 queue_cast(r, lkb, -DLM_EUNLOCK);
3603 log_error(r->res_ls, "receive_unlock_reply %x error %d",
3604 lkb->lkb_id, ms->m_result);
3611 static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
3613 struct dlm_lkb *lkb;
3616 error = find_lkb(ls, ms->m_remid, &lkb);
3618 log_debug(ls, "receive_unlock_reply from %d no lkb %x",
3619 ms->m_header.h_nodeid, ms->m_remid);
3623 _receive_unlock_reply(lkb, ms);
3627 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3629 struct dlm_rsb *r = lkb->lkb_resource;
3635 error = validate_message(lkb, ms);
3639 /* stub reply can happen with waiters_mutex held */
3640 error = remove_from_waiters_ms(lkb, ms);
3644 /* this is the value returned from do_cancel() on the master */
3646 switch (ms->m_result) {
3648 receive_flags_reply(lkb, ms);
3649 revert_lock_pc(r, lkb);
3650 queue_cast(r, lkb, -DLM_ECANCEL);
3655 log_error(r->res_ls, "receive_cancel_reply %x error %d",
3656 lkb->lkb_id, ms->m_result);
3663 static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
3665 struct dlm_lkb *lkb;
3668 error = find_lkb(ls, ms->m_remid, &lkb);
3670 log_debug(ls, "receive_cancel_reply from %d no lkb %x",
3671 ms->m_header.h_nodeid, ms->m_remid);
3675 _receive_cancel_reply(lkb, ms);
3679 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
3681 struct dlm_lkb *lkb;
3683 int error, ret_nodeid;
3685 error = find_lkb(ls, ms->m_lkid, &lkb);
3687 log_error(ls, "receive_lookup_reply no lkb");
3691 /* ms->m_result is the value returned by dlm_dir_lookup on dir node
3692 FIXME: will a non-zero error ever be returned? */
3694 r = lkb->lkb_resource;
3698 error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3702 ret_nodeid = ms->m_nodeid;
3703 if (ret_nodeid == dlm_our_nodeid()) {
3706 r->res_first_lkid = 0;
3708 /* set_master() will copy res_nodeid to lkb_nodeid */
3709 r->res_nodeid = ret_nodeid;
3712 if (is_overlap(lkb)) {
3713 log_debug(ls, "receive_lookup_reply %x unlock %x",
3714 lkb->lkb_id, lkb->lkb_flags);
3715 queue_cast_overlap(r, lkb);
3716 unhold_lkb(lkb); /* undoes create_lkb() */
3720 _request_lock(r, lkb);
3724 process_lookup_list(r);
3731 static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
3733 if (!dlm_is_member(ls, ms->m_header.h_nodeid)) {
3734 log_debug(ls, "ignore non-member message %d from %d %x %x %d",
3735 ms->m_type, ms->m_header.h_nodeid, ms->m_lkid,
3736 ms->m_remid, ms->m_result);
3740 switch (ms->m_type) {
3742 /* messages sent to a master node */
3744 case DLM_MSG_REQUEST:
3745 receive_request(ls, ms);
3748 case DLM_MSG_CONVERT:
3749 receive_convert(ls, ms);
3752 case DLM_MSG_UNLOCK:
3753 receive_unlock(ls, ms);
3756 case DLM_MSG_CANCEL:
3757 receive_cancel(ls, ms);
3760 /* messages sent from a master node (replies to above) */
3762 case DLM_MSG_REQUEST_REPLY:
3763 receive_request_reply(ls, ms);
3766 case DLM_MSG_CONVERT_REPLY:
3767 receive_convert_reply(ls, ms);
3770 case DLM_MSG_UNLOCK_REPLY:
3771 receive_unlock_reply(ls, ms);
3774 case DLM_MSG_CANCEL_REPLY:
3775 receive_cancel_reply(ls, ms);
3778 /* messages sent from a master node (only two types of async msg) */
3781 receive_grant(ls, ms);
3785 receive_bast(ls, ms);
3788 /* messages sent to a dir node */
3790 case DLM_MSG_LOOKUP:
3791 receive_lookup(ls, ms);
3794 case DLM_MSG_REMOVE:
3795 receive_remove(ls, ms);
3798 /* messages sent from a dir node (remove has no reply) */
3800 case DLM_MSG_LOOKUP_REPLY:
3801 receive_lookup_reply(ls, ms);
3804 /* other messages */
3807 receive_purge(ls, ms);
3811 log_error(ls, "unknown message type %d", ms->m_type);
3817 /* If the lockspace is in recovery mode (locking stopped), then normal
3818 messages are saved on the requestqueue for processing after recovery is
3819 done. When not in recovery mode, we wait for dlm_recoverd to drain saved
3820 messages off the requestqueue before we process new ones. This occurs right
3821 after recovery completes when we transition from saving all messages on
3822 requestqueue, to processing all the saved messages, to processing new
3823 messages as they arrive. */
3825 static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
3828 if (dlm_locking_stopped(ls)) {
3829 dlm_add_requestqueue(ls, nodeid, ms);
3831 dlm_wait_requestqueue(ls);
3832 _receive_message(ls, ms);
3836 /* This is called by dlm_recoverd to process messages that were saved on
3837 the requestqueue. */
3839 void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
3841 _receive_message(ls, ms);
3844 /* This is called by the midcomms layer when something is received for
3845 the lockspace. It could be either a MSG (normal message sent as part of
3846 standard locking activity) or an RCOM (recovery message sent as part of
3847 lockspace recovery). */
3849 void dlm_receive_buffer(union dlm_packet *p, int nodeid)
3851 struct dlm_header *hd = &p->header;
3855 switch (hd->h_cmd) {
3857 dlm_message_in(&p->message);
3858 type = p->message.m_type;
3861 dlm_rcom_in(&p->rcom);
3862 type = p->rcom.rc_type;
3865 log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
3869 if (hd->h_nodeid != nodeid) {
3870 log_print("invalid h_nodeid %d from %d lockspace %x",
3871 hd->h_nodeid, nodeid, hd->h_lockspace);
3875 ls = dlm_find_lockspace_global(hd->h_lockspace);
3877 if (dlm_config.ci_log_debug)
3878 log_print("invalid lockspace %x from %d cmd %d type %d",
3879 hd->h_lockspace, nodeid, hd->h_cmd, type);
3881 if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
3882 dlm_send_ls_not_ready(nodeid, &p->rcom);
3886 /* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3887 be inactive (in this ls) before transitioning to recovery mode */
3889 down_read(&ls->ls_recv_active);
3890 if (hd->h_cmd == DLM_MSG)
3891 dlm_receive_message(ls, &p->message, nodeid);
3893 dlm_receive_rcom(ls, &p->rcom, nodeid);
3894 up_read(&ls->ls_recv_active);
3896 dlm_put_lockspace(ls);
3899 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
3901 if (middle_conversion(lkb)) {
3903 ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
3904 ls->ls_stub_ms.m_result = -EINPROGRESS;
3905 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
3906 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
3907 _receive_convert_reply(lkb, &ls->ls_stub_ms);
3909 /* Same special case as in receive_rcom_lock_args() */
3910 lkb->lkb_grmode = DLM_LOCK_IV;
3911 rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
3914 } else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
3915 lkb->lkb_flags |= DLM_IFL_RESEND;
3918 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
3919 conversions are async; there's no reply from the remote master */
3922 /* A waiting lkb needs recovery if the master node has failed, or
3923 the master node is changing (only when no directory is used) */
3925 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
3927 if (dlm_is_removed(ls, lkb->lkb_nodeid))
3930 if (!dlm_no_directory(ls))
3933 if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid)
3939 /* Recovery for locks that are waiting for replies from nodes that are now
3940 gone. We can just complete unlocks and cancels by faking a reply from the
3941 dead node. Requests and up-conversions we flag to be resent after
3942 recovery. Down-conversions can just be completed with a fake reply like
3943 unlocks. Conversions between PR and CW need special attention. */
3945 void dlm_recover_waiters_pre(struct dlm_ls *ls)
3947 struct dlm_lkb *lkb, *safe;
3948 int wait_type, stub_unlock_result, stub_cancel_result;
3950 mutex_lock(&ls->ls_waiters_mutex);
3952 list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
3953 log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
3954 lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
3956 /* all outstanding lookups, regardless of destination will be
3957 resent after recovery is done */
3959 if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
3960 lkb->lkb_flags |= DLM_IFL_RESEND;
3964 if (!waiter_needs_recovery(ls, lkb))
3967 wait_type = lkb->lkb_wait_type;
3968 stub_unlock_result = -DLM_EUNLOCK;
3969 stub_cancel_result = -DLM_ECANCEL;
3971 /* Main reply may have been received leaving a zero wait_type,
3972 but a reply for the overlapping op may not have been
3973 received. In that case we need to fake the appropriate
3974 reply for the overlap op. */
3977 if (is_overlap_cancel(lkb)) {
3978 wait_type = DLM_MSG_CANCEL;
3979 if (lkb->lkb_grmode == DLM_LOCK_IV)
3980 stub_cancel_result = 0;
3982 if (is_overlap_unlock(lkb)) {
3983 wait_type = DLM_MSG_UNLOCK;
3984 if (lkb->lkb_grmode == DLM_LOCK_IV)
3985 stub_unlock_result = -ENOENT;
3988 log_debug(ls, "rwpre overlap %x %x %d %d %d",
3989 lkb->lkb_id, lkb->lkb_flags, wait_type,
3990 stub_cancel_result, stub_unlock_result);
3993 switch (wait_type) {
3995 case DLM_MSG_REQUEST:
3996 lkb->lkb_flags |= DLM_IFL_RESEND;
3999 case DLM_MSG_CONVERT:
4000 recover_convert_waiter(ls, lkb);
4003 case DLM_MSG_UNLOCK:
4005 ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
4006 ls->ls_stub_ms.m_result = stub_unlock_result;
4007 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
4008 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
4009 _receive_unlock_reply(lkb, &ls->ls_stub_ms);
4013 case DLM_MSG_CANCEL:
4015 ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
4016 ls->ls_stub_ms.m_result = stub_cancel_result;
4017 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
4018 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
4019 _receive_cancel_reply(lkb, &ls->ls_stub_ms);
4024 log_error(ls, "invalid lkb wait_type %d %d",
4025 lkb->lkb_wait_type, wait_type);
4029 mutex_unlock(&ls->ls_waiters_mutex);
4032 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
4034 struct dlm_lkb *lkb;
4037 mutex_lock(&ls->ls_waiters_mutex);
4038 list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
4039 if (lkb->lkb_flags & DLM_IFL_RESEND) {
4045 mutex_unlock(&ls->ls_waiters_mutex);
4052 /* Deal with lookups and lkb's marked RESEND from _pre. We may now be the
4053 master or dir-node for r. Processing the lkb may result in it being placed
4056 /* We do this after normal locking has been enabled and any saved messages
4057 (in requestqueue) have been processed. We should be confident that at
4058 this point we won't get or process a reply to any of these waiting
4059 operations. But, new ops may be coming in on the rsbs/locks here from
4060 userspace or remotely. */
4062 /* there may have been an overlap unlock/cancel prior to recovery or after
4063 recovery. if before, the lkb may still have a pos wait_count; if after, the
4064 overlap flag would just have been set and nothing new sent. we can be
4065 confident here than any replies to either the initial op or overlap ops
4066 prior to recovery have been received. */
4068 int dlm_recover_waiters_post(struct dlm_ls *ls)
4070 struct dlm_lkb *lkb;
4072 int error = 0, mstype, err, oc, ou;
4075 if (dlm_locking_stopped(ls)) {
4076 log_debug(ls, "recover_waiters_post aborted");
4081 lkb = find_resend_waiter(ls);
4085 r = lkb->lkb_resource;
4089 mstype = lkb->lkb_wait_type;
4090 oc = is_overlap_cancel(lkb);
4091 ou = is_overlap_unlock(lkb);
4094 log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
4095 lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
4097 /* At this point we assume that we won't get a reply to any
4098 previous op or overlap op on this lock. First, do a big
4099 remove_from_waiters() for all previous ops. */
4101 lkb->lkb_flags &= ~DLM_IFL_RESEND;
4102 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4103 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4104 lkb->lkb_wait_type = 0;
4105 lkb->lkb_wait_count = 0;
4106 mutex_lock(&ls->ls_waiters_mutex);
4107 list_del_init(&lkb->lkb_wait_reply);
4108 mutex_unlock(&ls->ls_waiters_mutex);
4109 unhold_lkb(lkb); /* for waiters list */
4112 /* do an unlock or cancel instead of resending */
4114 case DLM_MSG_LOOKUP:
4115 case DLM_MSG_REQUEST:
4116 queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
4118 unhold_lkb(lkb); /* undoes create_lkb() */
4120 case DLM_MSG_CONVERT:
4122 queue_cast(r, lkb, -DLM_ECANCEL);
4124 lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
4125 _unlock_lock(r, lkb);
4133 case DLM_MSG_LOOKUP:
4134 case DLM_MSG_REQUEST:
4135 _request_lock(r, lkb);
4137 confirm_master(r, 0);
4139 case DLM_MSG_CONVERT:
4140 _convert_lock(r, lkb);
4148 log_error(ls, "recover_waiters_post %x %d %x %d %d",
4149 lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
4158 static void purge_queue(struct dlm_rsb *r, struct list_head *queue,
4159 int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb))
4161 struct dlm_ls *ls = r->res_ls;
4162 struct dlm_lkb *lkb, *safe;
4164 list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) {
4165 if (test(ls, lkb)) {
4166 rsb_set_flag(r, RSB_LOCKS_PURGED);
4168 /* this put should free the lkb */
4169 if (!dlm_put_lkb(lkb))
4170 log_error(ls, "purged lkb not released");
4175 static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4177 return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid));
4180 static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4182 return is_master_copy(lkb);
4185 static void purge_dead_locks(struct dlm_rsb *r)
4187 purge_queue(r, &r->res_grantqueue, &purge_dead_test);
4188 purge_queue(r, &r->res_convertqueue, &purge_dead_test);
4189 purge_queue(r, &r->res_waitqueue, &purge_dead_test);
4192 void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
4194 purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test);
4195 purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test);
4196 purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test);
4199 /* Get rid of locks held by nodes that are gone. */
4201 int dlm_purge_locks(struct dlm_ls *ls)
4205 log_debug(ls, "dlm_purge_locks");
4207 down_write(&ls->ls_root_sem);
4208 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
4212 purge_dead_locks(r);
4218 up_write(&ls->ls_root_sem);
4223 static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket)
4225 struct dlm_rsb *r, *r_ret = NULL;
4227 spin_lock(&ls->ls_rsbtbl[bucket].lock);
4228 list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) {
4229 if (!rsb_flag(r, RSB_LOCKS_PURGED))
4232 rsb_clear_flag(r, RSB_LOCKS_PURGED);
4236 spin_unlock(&ls->ls_rsbtbl[bucket].lock);
4240 void dlm_grant_after_purge(struct dlm_ls *ls)
4246 r = find_purged_rsb(ls, bucket);
4248 if (bucket == ls->ls_rsbtbl_size - 1)
4255 grant_pending_locks(r);
4256 confirm_master(r, 0);
4264 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
4267 struct dlm_lkb *lkb;
4269 list_for_each_entry(lkb, head, lkb_statequeue) {
4270 if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
4276 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
4279 struct dlm_lkb *lkb;
4281 lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
4284 lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
4287 lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
4293 /* needs at least dlm_rcom + rcom_lock */
4294 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
4295 struct dlm_rsb *r, struct dlm_rcom *rc)
4297 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4299 lkb->lkb_nodeid = rc->rc_header.h_nodeid;
4300 lkb->lkb_ownpid = le32_to_cpu(rl->rl_ownpid);
4301 lkb->lkb_remid = le32_to_cpu(rl->rl_lkid);
4302 lkb->lkb_exflags = le32_to_cpu(rl->rl_exflags);
4303 lkb->lkb_flags = le32_to_cpu(rl->rl_flags) & 0x0000FFFF;
4304 lkb->lkb_flags |= DLM_IFL_MSTCPY;
4305 lkb->lkb_lvbseq = le32_to_cpu(rl->rl_lvbseq);
4306 lkb->lkb_rqmode = rl->rl_rqmode;
4307 lkb->lkb_grmode = rl->rl_grmode;
4308 /* don't set lkb_status because add_lkb wants to itself */
4310 lkb->lkb_bastfn = (rl->rl_asts & AST_BAST) ? &fake_bastfn : NULL;
4311 lkb->lkb_astfn = (rl->rl_asts & AST_COMP) ? &fake_astfn : NULL;
4313 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
4314 int lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) -
4315 sizeof(struct rcom_lock);
4316 if (lvblen > ls->ls_lvblen)
4318 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
4319 if (!lkb->lkb_lvbptr)
4321 memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
4324 /* Conversions between PR and CW (middle modes) need special handling.
4325 The real granted mode of these converting locks cannot be determined
4326 until all locks have been rebuilt on the rsb (recover_conversion) */
4328 if (rl->rl_wait_type == cpu_to_le16(DLM_MSG_CONVERT) &&
4329 middle_conversion(lkb)) {
4330 rl->rl_status = DLM_LKSTS_CONVERT;
4331 lkb->lkb_grmode = DLM_LOCK_IV;
4332 rsb_set_flag(r, RSB_RECOVER_CONVERT);
4338 /* This lkb may have been recovered in a previous aborted recovery so we need
4339 to check if the rsb already has an lkb with the given remote nodeid/lkid.
4340 If so we just send back a standard reply. If not, we create a new lkb with
4341 the given values and send back our lkid. We send back our lkid by sending
4342 back the rcom_lock struct we got but with the remid field filled in. */
4344 /* needs at least dlm_rcom + rcom_lock */
4345 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4347 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4349 struct dlm_lkb *lkb;
4352 if (rl->rl_parent_lkid) {
4353 error = -EOPNOTSUPP;
4357 error = find_rsb(ls, rl->rl_name, le16_to_cpu(rl->rl_namelen),
4364 lkb = search_remid(r, rc->rc_header.h_nodeid, le32_to_cpu(rl->rl_lkid));
4370 error = create_lkb(ls, &lkb);
4374 error = receive_rcom_lock_args(ls, lkb, r, rc);
4381 add_lkb(r, lkb, rl->rl_status);
4385 /* this is the new value returned to the lock holder for
4386 saving in its process-copy lkb */
4387 rl->rl_remid = cpu_to_le32(lkb->lkb_id);
4394 log_debug(ls, "recover_master_copy %d %x", error,
4395 le32_to_cpu(rl->rl_lkid));
4396 rl->rl_result = cpu_to_le32(error);
4400 /* needs at least dlm_rcom + rcom_lock */
4401 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4403 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4405 struct dlm_lkb *lkb;
4408 error = find_lkb(ls, le32_to_cpu(rl->rl_lkid), &lkb);
4410 log_error(ls, "recover_process_copy no lkid %x",
4411 le32_to_cpu(rl->rl_lkid));
4415 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
4417 error = le32_to_cpu(rl->rl_result);
4419 r = lkb->lkb_resource;
4425 /* There's a chance the new master received our lock before
4426 dlm_recover_master_reply(), this wouldn't happen if we did
4427 a barrier between recover_masters and recover_locks. */
4428 log_debug(ls, "master copy not ready %x r %lx %s", lkb->lkb_id,
4429 (unsigned long)r, r->res_name);
4430 dlm_send_rcom_lock(r, lkb);
4433 log_debug(ls, "master copy exists %x", lkb->lkb_id);
4436 lkb->lkb_remid = le32_to_cpu(rl->rl_remid);
4439 log_error(ls, "dlm_recover_process_copy unknown error %d %x",
4440 error, lkb->lkb_id);
4443 /* an ack for dlm_recover_locks() which waits for replies from
4444 all the locks it sends to new masters */
4445 dlm_recovered_lock(r);
4454 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
4455 int mode, uint32_t flags, void *name, unsigned int namelen,
4456 unsigned long timeout_cs)
4458 struct dlm_lkb *lkb;
4459 struct dlm_args args;
4462 dlm_lock_recovery(ls);
4464 error = create_lkb(ls, &lkb);
4470 if (flags & DLM_LKF_VALBLK) {
4471 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
4472 if (!ua->lksb.sb_lvbptr) {
4480 /* After ua is attached to lkb it will be freed by dlm_free_lkb().
4481 When DLM_IFL_USER is set, the dlm knows that this is a userspace
4482 lock and that lkb_astparam is the dlm_user_args structure. */
4484 error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
4485 fake_astfn, ua, fake_bastfn, &args);
4486 lkb->lkb_flags |= DLM_IFL_USER;
4487 ua->old_mode = DLM_LOCK_IV;
4494 error = request_lock(ls, lkb, name, namelen, &args);
4510 /* add this new lkb to the per-process list of locks */
4511 spin_lock(&ua->proc->locks_spin);
4513 list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
4514 spin_unlock(&ua->proc->locks_spin);
4516 dlm_unlock_recovery(ls);
4520 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4521 int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
4522 unsigned long timeout_cs)
4524 struct dlm_lkb *lkb;
4525 struct dlm_args args;
4526 struct dlm_user_args *ua;
4529 dlm_lock_recovery(ls);
4531 error = find_lkb(ls, lkid, &lkb);
4535 /* user can change the params on its lock when it converts it, or
4536 add an lvb that didn't exist before */
4540 if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
4541 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
4542 if (!ua->lksb.sb_lvbptr) {
4547 if (lvb_in && ua->lksb.sb_lvbptr)
4548 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4550 ua->xid = ua_tmp->xid;
4551 ua->castparam = ua_tmp->castparam;
4552 ua->castaddr = ua_tmp->castaddr;
4553 ua->bastparam = ua_tmp->bastparam;
4554 ua->bastaddr = ua_tmp->bastaddr;
4555 ua->user_lksb = ua_tmp->user_lksb;
4556 ua->old_mode = lkb->lkb_grmode;
4558 error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
4559 fake_astfn, ua, fake_bastfn, &args);
4563 error = convert_lock(ls, lkb, &args);
4565 if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
4570 dlm_unlock_recovery(ls);
4575 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4576 uint32_t flags, uint32_t lkid, char *lvb_in)
4578 struct dlm_lkb *lkb;
4579 struct dlm_args args;
4580 struct dlm_user_args *ua;
4583 dlm_lock_recovery(ls);
4585 error = find_lkb(ls, lkid, &lkb);
4591 if (lvb_in && ua->lksb.sb_lvbptr)
4592 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4593 if (ua_tmp->castparam)
4594 ua->castparam = ua_tmp->castparam;
4595 ua->user_lksb = ua_tmp->user_lksb;
4597 error = set_unlock_args(flags, ua, &args);
4601 error = unlock_lock(ls, lkb, &args);
4603 if (error == -DLM_EUNLOCK)
4605 /* from validate_unlock_args() */
4606 if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
4611 spin_lock(&ua->proc->locks_spin);
4612 /* dlm_user_add_ast() may have already taken lkb off the proc list */
4613 if (!list_empty(&lkb->lkb_ownqueue))
4614 list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
4615 spin_unlock(&ua->proc->locks_spin);
4619 dlm_unlock_recovery(ls);
4624 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4625 uint32_t flags, uint32_t lkid)
4627 struct dlm_lkb *lkb;
4628 struct dlm_args args;
4629 struct dlm_user_args *ua;
4632 dlm_lock_recovery(ls);
4634 error = find_lkb(ls, lkid, &lkb);
4639 if (ua_tmp->castparam)
4640 ua->castparam = ua_tmp->castparam;
4641 ua->user_lksb = ua_tmp->user_lksb;
4643 error = set_unlock_args(flags, ua, &args);
4647 error = cancel_lock(ls, lkb, &args);
4649 if (error == -DLM_ECANCEL)
4651 /* from validate_unlock_args() */
4652 if (error == -EBUSY)
4657 dlm_unlock_recovery(ls);
4662 int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
4664 struct dlm_lkb *lkb;
4665 struct dlm_args args;
4666 struct dlm_user_args *ua;
4670 dlm_lock_recovery(ls);
4672 error = find_lkb(ls, lkid, &lkb);
4678 error = set_unlock_args(flags, ua, &args);
4682 /* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4684 r = lkb->lkb_resource;
4688 error = validate_unlock_args(lkb, &args);
4691 lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
4693 error = _cancel_lock(r, lkb);
4698 if (error == -DLM_ECANCEL)
4700 /* from validate_unlock_args() */
4701 if (error == -EBUSY)
4706 dlm_unlock_recovery(ls);
4710 /* lkb's that are removed from the waiters list by revert are just left on the
4711 orphans list with the granted orphan locks, to be freed by purge */
4713 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4715 struct dlm_args args;
4719 mutex_lock(&ls->ls_orphans_mutex);
4720 list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
4721 mutex_unlock(&ls->ls_orphans_mutex);
4723 set_unlock_args(0, lkb->lkb_ua, &args);
4725 error = cancel_lock(ls, lkb, &args);
4726 if (error == -DLM_ECANCEL)
4731 /* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4732 Regardless of what rsb queue the lock is on, it's removed and freed. */
4734 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4736 struct dlm_args args;
4739 set_unlock_args(DLM_LKF_FORCEUNLOCK, lkb->lkb_ua, &args);
4741 error = unlock_lock(ls, lkb, &args);
4742 if (error == -DLM_EUNLOCK)
4747 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4748 (which does lock_rsb) due to deadlock with receiving a message that does
4749 lock_rsb followed by dlm_user_add_ast() */
4751 static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
4752 struct dlm_user_proc *proc)
4754 struct dlm_lkb *lkb = NULL;
4756 mutex_lock(&ls->ls_clear_proc_locks);
4757 if (list_empty(&proc->locks))
4760 lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
4761 list_del_init(&lkb->lkb_ownqueue);
4763 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4764 lkb->lkb_flags |= DLM_IFL_ORPHAN;
4766 lkb->lkb_flags |= DLM_IFL_DEAD;
4768 mutex_unlock(&ls->ls_clear_proc_locks);
4772 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4773 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4774 which we clear here. */
4776 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
4777 list, and no more device_writes should add lkb's to proc->locks list; so we
4778 shouldn't need to take asts_spin or locks_spin here. this assumes that
4779 device reads/writes/closes are serialized -- FIXME: we may need to serialize
4782 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4784 struct dlm_lkb *lkb, *safe;
4786 dlm_lock_recovery(ls);
4789 lkb = del_proc_lock(ls, proc);
4793 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4794 orphan_proc_lock(ls, lkb);
4796 unlock_proc_lock(ls, lkb);
4798 /* this removes the reference for the proc->locks list
4799 added by dlm_user_request, it may result in the lkb
4805 mutex_lock(&ls->ls_clear_proc_locks);
4807 /* in-progress unlocks */
4808 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4809 list_del_init(&lkb->lkb_ownqueue);
4810 lkb->lkb_flags |= DLM_IFL_DEAD;
4814 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4815 lkb->lkb_ast_type = 0;
4816 list_del(&lkb->lkb_astqueue);
4820 mutex_unlock(&ls->ls_clear_proc_locks);
4821 dlm_unlock_recovery(ls);
4824 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4826 struct dlm_lkb *lkb, *safe;
4830 spin_lock(&proc->locks_spin);
4831 if (!list_empty(&proc->locks)) {
4832 lkb = list_entry(proc->locks.next, struct dlm_lkb,
4834 list_del_init(&lkb->lkb_ownqueue);
4836 spin_unlock(&proc->locks_spin);
4841 lkb->lkb_flags |= DLM_IFL_DEAD;
4842 unlock_proc_lock(ls, lkb);
4843 dlm_put_lkb(lkb); /* ref from proc->locks list */
4846 spin_lock(&proc->locks_spin);
4847 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4848 list_del_init(&lkb->lkb_ownqueue);
4849 lkb->lkb_flags |= DLM_IFL_DEAD;
4852 spin_unlock(&proc->locks_spin);
4854 spin_lock(&proc->asts_spin);
4855 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4856 list_del(&lkb->lkb_astqueue);
4859 spin_unlock(&proc->asts_spin);
4862 /* pid of 0 means purge all orphans */
4864 static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
4866 struct dlm_lkb *lkb, *safe;
4868 mutex_lock(&ls->ls_orphans_mutex);
4869 list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
4870 if (pid && lkb->lkb_ownpid != pid)
4872 unlock_proc_lock(ls, lkb);
4873 list_del_init(&lkb->lkb_ownqueue);
4876 mutex_unlock(&ls->ls_orphans_mutex);
4879 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
4881 struct dlm_message *ms;
4882 struct dlm_mhandle *mh;
4885 error = _create_message(ls, sizeof(struct dlm_message), nodeid,
4886 DLM_MSG_PURGE, &ms, &mh);
4889 ms->m_nodeid = nodeid;
4892 return send_message(mh, ms);
4895 int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
4896 int nodeid, int pid)
4900 if (nodeid != dlm_our_nodeid()) {
4901 error = send_purge(ls, nodeid, pid);
4903 dlm_lock_recovery(ls);
4904 if (pid == current->pid)
4905 purge_proc_locks(ls, proc);
4907 do_purge(ls, nodeid, pid);
4908 dlm_unlock_recovery(ls);