xref: /openbmc/linux/fs/dlm/lock.c (revision f2d8e15b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3 *******************************************************************************
4 **
5 **  Copyright (C) 2005-2010 Red Hat, Inc.  All rights reserved.
6 **
7 **
8 *******************************************************************************
9 ******************************************************************************/
10 
11 /* Central locking logic has four stages:
12 
13    dlm_lock()
14    dlm_unlock()
15 
16    request_lock(ls, lkb)
17    convert_lock(ls, lkb)
18    unlock_lock(ls, lkb)
19    cancel_lock(ls, lkb)
20 
21    _request_lock(r, lkb)
22    _convert_lock(r, lkb)
23    _unlock_lock(r, lkb)
24    _cancel_lock(r, lkb)
25 
26    do_request(r, lkb)
27    do_convert(r, lkb)
28    do_unlock(r, lkb)
29    do_cancel(r, lkb)
30 
31    Stage 1 (lock, unlock) is mainly about checking input args and
32    splitting into one of the four main operations:
33 
34        dlm_lock          = request_lock
35        dlm_lock+CONVERT  = convert_lock
36        dlm_unlock        = unlock_lock
37        dlm_unlock+CANCEL = cancel_lock
38 
39    Stage 2, xxxx_lock(), just finds and locks the relevant rsb which is
40    provided to the next stage.
41 
42    Stage 3, _xxxx_lock(), determines if the operation is local or remote.
43    When remote, it calls send_xxxx(), when local it calls do_xxxx().
44 
45    Stage 4, do_xxxx(), is the guts of the operation.  It manipulates the
46    given rsb and lkb and queues callbacks.
47 
48    For remote operations, send_xxxx() results in the corresponding do_xxxx()
49    function being executed on the remote node.  The connecting send/receive
50    calls on local (L) and remote (R) nodes:
51 
52    L: send_xxxx()              ->  R: receive_xxxx()
53                                    R: do_xxxx()
54    L: receive_xxxx_reply()     <-  R: send_xxxx_reply()
55 */
56 #include <trace/events/dlm.h>
57 
58 #include <linux/types.h>
59 #include <linux/rbtree.h>
60 #include <linux/slab.h>
61 #include "dlm_internal.h"
62 #include <linux/dlm_device.h>
63 #include "memory.h"
64 #include "midcomms.h"
65 #include "requestqueue.h"
66 #include "util.h"
67 #include "dir.h"
68 #include "member.h"
69 #include "lockspace.h"
70 #include "ast.h"
71 #include "lock.h"
72 #include "rcom.h"
73 #include "recover.h"
74 #include "lvb_table.h"
75 #include "user.h"
76 #include "config.h"
77 
78 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb);
79 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb);
80 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb);
81 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb);
82 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb);
83 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode);
84 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb);
85 static int send_remove(struct dlm_rsb *r);
86 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
87 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
88 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
89 				    struct dlm_message *ms);
90 static int receive_extralen(struct dlm_message *ms);
91 static void do_purge(struct dlm_ls *ls, int nodeid, int pid);
92 static void del_timeout(struct dlm_lkb *lkb);
93 static void toss_rsb(struct kref *kref);
94 
95 /*
96  * Lock compatibilty matrix - thanks Steve
97  * UN = Unlocked state. Not really a state, used as a flag
98  * PD = Padding. Used to make the matrix a nice power of two in size
99  * Other states are the same as the VMS DLM.
100  * Usage: matrix[grmode+1][rqmode+1]  (although m[rq+1][gr+1] is the same)
101  */
102 
103 static const int __dlm_compat_matrix[8][8] = {
104       /* UN NL CR CW PR PW EX PD */
105         {1, 1, 1, 1, 1, 1, 1, 0},       /* UN */
106         {1, 1, 1, 1, 1, 1, 1, 0},       /* NL */
107         {1, 1, 1, 1, 1, 1, 0, 0},       /* CR */
108         {1, 1, 1, 1, 0, 0, 0, 0},       /* CW */
109         {1, 1, 1, 0, 1, 0, 0, 0},       /* PR */
110         {1, 1, 1, 0, 0, 0, 0, 0},       /* PW */
111         {1, 1, 0, 0, 0, 0, 0, 0},       /* EX */
112         {0, 0, 0, 0, 0, 0, 0, 0}        /* PD */
113 };
114 
115 /*
116  * This defines the direction of transfer of LVB data.
117  * Granted mode is the row; requested mode is the column.
118  * Usage: matrix[grmode+1][rqmode+1]
119  * 1 = LVB is returned to the caller
120  * 0 = LVB is written to the resource
121  * -1 = nothing happens to the LVB
122  */
123 
124 const int dlm_lvb_operations[8][8] = {
125         /* UN   NL  CR  CW  PR  PW  EX  PD*/
126         {  -1,  1,  1,  1,  1,  1,  1, -1 }, /* UN */
127         {  -1,  1,  1,  1,  1,  1,  1,  0 }, /* NL */
128         {  -1, -1,  1,  1,  1,  1,  1,  0 }, /* CR */
129         {  -1, -1, -1,  1,  1,  1,  1,  0 }, /* CW */
130         {  -1, -1, -1, -1,  1,  1,  1,  0 }, /* PR */
131         {  -1,  0,  0,  0,  0,  0,  1,  0 }, /* PW */
132         {  -1,  0,  0,  0,  0,  0,  0,  0 }, /* EX */
133         {  -1,  0,  0,  0,  0,  0,  0,  0 }  /* PD */
134 };
135 
136 #define modes_compat(gr, rq) \
137 	__dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
138 
139 int dlm_modes_compat(int mode1, int mode2)
140 {
141 	return __dlm_compat_matrix[mode1 + 1][mode2 + 1];
142 }
143 
144 /*
145  * Compatibility matrix for conversions with QUECVT set.
146  * Granted mode is the row; requested mode is the column.
147  * Usage: matrix[grmode+1][rqmode+1]
148  */
149 
150 static const int __quecvt_compat_matrix[8][8] = {
151       /* UN NL CR CW PR PW EX PD */
152         {0, 0, 0, 0, 0, 0, 0, 0},       /* UN */
153         {0, 0, 1, 1, 1, 1, 1, 0},       /* NL */
154         {0, 0, 0, 1, 1, 1, 1, 0},       /* CR */
155         {0, 0, 0, 0, 1, 1, 1, 0},       /* CW */
156         {0, 0, 0, 1, 0, 1, 1, 0},       /* PR */
157         {0, 0, 0, 0, 0, 0, 1, 0},       /* PW */
158         {0, 0, 0, 0, 0, 0, 0, 0},       /* EX */
159         {0, 0, 0, 0, 0, 0, 0, 0}        /* PD */
160 };
161 
162 void dlm_print_lkb(struct dlm_lkb *lkb)
163 {
164 	printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x "
165 	       "sts %d rq %d gr %d wait_type %d wait_nodeid %d seq %llu\n",
166 	       lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags,
167 	       lkb->lkb_flags, lkb->lkb_status, lkb->lkb_rqmode,
168 	       lkb->lkb_grmode, lkb->lkb_wait_type, lkb->lkb_wait_nodeid,
169 	       (unsigned long long)lkb->lkb_recover_seq);
170 }
171 
172 static void dlm_print_rsb(struct dlm_rsb *r)
173 {
174 	printk(KERN_ERR "rsb: nodeid %d master %d dir %d flags %lx first %x "
175 	       "rlc %d name %s\n",
176 	       r->res_nodeid, r->res_master_nodeid, r->res_dir_nodeid,
177 	       r->res_flags, r->res_first_lkid, r->res_recover_locks_count,
178 	       r->res_name);
179 }
180 
181 void dlm_dump_rsb(struct dlm_rsb *r)
182 {
183 	struct dlm_lkb *lkb;
184 
185 	dlm_print_rsb(r);
186 
187 	printk(KERN_ERR "rsb: root_list empty %d recover_list empty %d\n",
188 	       list_empty(&r->res_root_list), list_empty(&r->res_recover_list));
189 	printk(KERN_ERR "rsb lookup list\n");
190 	list_for_each_entry(lkb, &r->res_lookup, lkb_rsb_lookup)
191 		dlm_print_lkb(lkb);
192 	printk(KERN_ERR "rsb grant queue:\n");
193 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue)
194 		dlm_print_lkb(lkb);
195 	printk(KERN_ERR "rsb convert queue:\n");
196 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue)
197 		dlm_print_lkb(lkb);
198 	printk(KERN_ERR "rsb wait queue:\n");
199 	list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue)
200 		dlm_print_lkb(lkb);
201 }
202 
203 /* Threads cannot use the lockspace while it's being recovered */
204 
205 static inline void dlm_lock_recovery(struct dlm_ls *ls)
206 {
207 	down_read(&ls->ls_in_recovery);
208 }
209 
210 void dlm_unlock_recovery(struct dlm_ls *ls)
211 {
212 	up_read(&ls->ls_in_recovery);
213 }
214 
215 int dlm_lock_recovery_try(struct dlm_ls *ls)
216 {
217 	return down_read_trylock(&ls->ls_in_recovery);
218 }
219 
220 static inline int can_be_queued(struct dlm_lkb *lkb)
221 {
222 	return !(lkb->lkb_exflags & DLM_LKF_NOQUEUE);
223 }
224 
225 static inline int force_blocking_asts(struct dlm_lkb *lkb)
226 {
227 	return (lkb->lkb_exflags & DLM_LKF_NOQUEUEBAST);
228 }
229 
230 static inline int is_demoted(struct dlm_lkb *lkb)
231 {
232 	return (lkb->lkb_sbflags & DLM_SBF_DEMOTED);
233 }
234 
235 static inline int is_altmode(struct dlm_lkb *lkb)
236 {
237 	return (lkb->lkb_sbflags & DLM_SBF_ALTMODE);
238 }
239 
240 static inline int is_granted(struct dlm_lkb *lkb)
241 {
242 	return (lkb->lkb_status == DLM_LKSTS_GRANTED);
243 }
244 
245 static inline int is_remote(struct dlm_rsb *r)
246 {
247 	DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r););
248 	return !!r->res_nodeid;
249 }
250 
251 static inline int is_process_copy(struct dlm_lkb *lkb)
252 {
253 	return (lkb->lkb_nodeid && !(lkb->lkb_flags & DLM_IFL_MSTCPY));
254 }
255 
256 static inline int is_master_copy(struct dlm_lkb *lkb)
257 {
258 	return (lkb->lkb_flags & DLM_IFL_MSTCPY) ? 1 : 0;
259 }
260 
261 static inline int middle_conversion(struct dlm_lkb *lkb)
262 {
263 	if ((lkb->lkb_grmode==DLM_LOCK_PR && lkb->lkb_rqmode==DLM_LOCK_CW) ||
264 	    (lkb->lkb_rqmode==DLM_LOCK_PR && lkb->lkb_grmode==DLM_LOCK_CW))
265 		return 1;
266 	return 0;
267 }
268 
269 static inline int down_conversion(struct dlm_lkb *lkb)
270 {
271 	return (!middle_conversion(lkb) && lkb->lkb_rqmode < lkb->lkb_grmode);
272 }
273 
274 static inline int is_overlap_unlock(struct dlm_lkb *lkb)
275 {
276 	return lkb->lkb_flags & DLM_IFL_OVERLAP_UNLOCK;
277 }
278 
279 static inline int is_overlap_cancel(struct dlm_lkb *lkb)
280 {
281 	return lkb->lkb_flags & DLM_IFL_OVERLAP_CANCEL;
282 }
283 
284 static inline int is_overlap(struct dlm_lkb *lkb)
285 {
286 	return (lkb->lkb_flags & (DLM_IFL_OVERLAP_UNLOCK |
287 				  DLM_IFL_OVERLAP_CANCEL));
288 }
289 
290 static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
291 {
292 	if (is_master_copy(lkb))
293 		return;
294 
295 	del_timeout(lkb);
296 
297 	DLM_ASSERT(lkb->lkb_lksb, dlm_print_lkb(lkb););
298 
299 #ifdef CONFIG_DLM_DEPRECATED_API
300 	/* if the operation was a cancel, then return -DLM_ECANCEL, if a
301 	   timeout caused the cancel then return -ETIMEDOUT */
302 	if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_TIMEOUT_CANCEL)) {
303 		lkb->lkb_flags &= ~DLM_IFL_TIMEOUT_CANCEL;
304 		rv = -ETIMEDOUT;
305 	}
306 #endif
307 
308 	if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_DEADLOCK_CANCEL)) {
309 		lkb->lkb_flags &= ~DLM_IFL_DEADLOCK_CANCEL;
310 		rv = -EDEADLK;
311 	}
312 
313 	dlm_add_cb(lkb, DLM_CB_CAST, lkb->lkb_grmode, rv, lkb->lkb_sbflags);
314 }
315 
316 static inline void queue_cast_overlap(struct dlm_rsb *r, struct dlm_lkb *lkb)
317 {
318 	queue_cast(r, lkb,
319 		   is_overlap_unlock(lkb) ? -DLM_EUNLOCK : -DLM_ECANCEL);
320 }
321 
322 static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
323 {
324 	if (is_master_copy(lkb)) {
325 		send_bast(r, lkb, rqmode);
326 	} else {
327 		dlm_add_cb(lkb, DLM_CB_BAST, rqmode, 0, 0);
328 	}
329 }
330 
331 /*
332  * Basic operations on rsb's and lkb's
333  */
334 
335 /* This is only called to add a reference when the code already holds
336    a valid reference to the rsb, so there's no need for locking. */
337 
338 static inline void hold_rsb(struct dlm_rsb *r)
339 {
340 	kref_get(&r->res_ref);
341 }
342 
343 void dlm_hold_rsb(struct dlm_rsb *r)
344 {
345 	hold_rsb(r);
346 }
347 
348 /* When all references to the rsb are gone it's transferred to
349    the tossed list for later disposal. */
350 
351 static void put_rsb(struct dlm_rsb *r)
352 {
353 	struct dlm_ls *ls = r->res_ls;
354 	uint32_t bucket = r->res_bucket;
355 	int rv;
356 
357 	rv = kref_put_lock(&r->res_ref, toss_rsb,
358 			   &ls->ls_rsbtbl[bucket].lock);
359 	if (rv)
360 		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
361 }
362 
363 void dlm_put_rsb(struct dlm_rsb *r)
364 {
365 	put_rsb(r);
366 }
367 
368 static int pre_rsb_struct(struct dlm_ls *ls)
369 {
370 	struct dlm_rsb *r1, *r2;
371 	int count = 0;
372 
373 	spin_lock(&ls->ls_new_rsb_spin);
374 	if (ls->ls_new_rsb_count > dlm_config.ci_new_rsb_count / 2) {
375 		spin_unlock(&ls->ls_new_rsb_spin);
376 		return 0;
377 	}
378 	spin_unlock(&ls->ls_new_rsb_spin);
379 
380 	r1 = dlm_allocate_rsb(ls);
381 	r2 = dlm_allocate_rsb(ls);
382 
383 	spin_lock(&ls->ls_new_rsb_spin);
384 	if (r1) {
385 		list_add(&r1->res_hashchain, &ls->ls_new_rsb);
386 		ls->ls_new_rsb_count++;
387 	}
388 	if (r2) {
389 		list_add(&r2->res_hashchain, &ls->ls_new_rsb);
390 		ls->ls_new_rsb_count++;
391 	}
392 	count = ls->ls_new_rsb_count;
393 	spin_unlock(&ls->ls_new_rsb_spin);
394 
395 	if (!count)
396 		return -ENOMEM;
397 	return 0;
398 }
399 
400 /* If ls->ls_new_rsb is empty, return -EAGAIN, so the caller can
401    unlock any spinlocks, go back and call pre_rsb_struct again.
402    Otherwise, take an rsb off the list and return it. */
403 
404 static int get_rsb_struct(struct dlm_ls *ls, char *name, int len,
405 			  struct dlm_rsb **r_ret)
406 {
407 	struct dlm_rsb *r;
408 	int count;
409 
410 	spin_lock(&ls->ls_new_rsb_spin);
411 	if (list_empty(&ls->ls_new_rsb)) {
412 		count = ls->ls_new_rsb_count;
413 		spin_unlock(&ls->ls_new_rsb_spin);
414 		log_debug(ls, "find_rsb retry %d %d %s",
415 			  count, dlm_config.ci_new_rsb_count, name);
416 		return -EAGAIN;
417 	}
418 
419 	r = list_first_entry(&ls->ls_new_rsb, struct dlm_rsb, res_hashchain);
420 	list_del(&r->res_hashchain);
421 	/* Convert the empty list_head to a NULL rb_node for tree usage: */
422 	memset(&r->res_hashnode, 0, sizeof(struct rb_node));
423 	ls->ls_new_rsb_count--;
424 	spin_unlock(&ls->ls_new_rsb_spin);
425 
426 	r->res_ls = ls;
427 	r->res_length = len;
428 	memcpy(r->res_name, name, len);
429 	mutex_init(&r->res_mutex);
430 
431 	INIT_LIST_HEAD(&r->res_lookup);
432 	INIT_LIST_HEAD(&r->res_grantqueue);
433 	INIT_LIST_HEAD(&r->res_convertqueue);
434 	INIT_LIST_HEAD(&r->res_waitqueue);
435 	INIT_LIST_HEAD(&r->res_root_list);
436 	INIT_LIST_HEAD(&r->res_recover_list);
437 
438 	*r_ret = r;
439 	return 0;
440 }
441 
442 static int rsb_cmp(struct dlm_rsb *r, const char *name, int nlen)
443 {
444 	char maxname[DLM_RESNAME_MAXLEN];
445 
446 	memset(maxname, 0, DLM_RESNAME_MAXLEN);
447 	memcpy(maxname, name, nlen);
448 	return memcmp(r->res_name, maxname, DLM_RESNAME_MAXLEN);
449 }
450 
451 int dlm_search_rsb_tree(struct rb_root *tree, char *name, int len,
452 			struct dlm_rsb **r_ret)
453 {
454 	struct rb_node *node = tree->rb_node;
455 	struct dlm_rsb *r;
456 	int rc;
457 
458 	while (node) {
459 		r = rb_entry(node, struct dlm_rsb, res_hashnode);
460 		rc = rsb_cmp(r, name, len);
461 		if (rc < 0)
462 			node = node->rb_left;
463 		else if (rc > 0)
464 			node = node->rb_right;
465 		else
466 			goto found;
467 	}
468 	*r_ret = NULL;
469 	return -EBADR;
470 
471  found:
472 	*r_ret = r;
473 	return 0;
474 }
475 
476 static int rsb_insert(struct dlm_rsb *rsb, struct rb_root *tree)
477 {
478 	struct rb_node **newn = &tree->rb_node;
479 	struct rb_node *parent = NULL;
480 	int rc;
481 
482 	while (*newn) {
483 		struct dlm_rsb *cur = rb_entry(*newn, struct dlm_rsb,
484 					       res_hashnode);
485 
486 		parent = *newn;
487 		rc = rsb_cmp(cur, rsb->res_name, rsb->res_length);
488 		if (rc < 0)
489 			newn = &parent->rb_left;
490 		else if (rc > 0)
491 			newn = &parent->rb_right;
492 		else {
493 			log_print("rsb_insert match");
494 			dlm_dump_rsb(rsb);
495 			dlm_dump_rsb(cur);
496 			return -EEXIST;
497 		}
498 	}
499 
500 	rb_link_node(&rsb->res_hashnode, parent, newn);
501 	rb_insert_color(&rsb->res_hashnode, tree);
502 	return 0;
503 }
504 
505 /*
506  * Find rsb in rsbtbl and potentially create/add one
507  *
508  * Delaying the release of rsb's has a similar benefit to applications keeping
509  * NL locks on an rsb, but without the guarantee that the cached master value
510  * will still be valid when the rsb is reused.  Apps aren't always smart enough
511  * to keep NL locks on an rsb that they may lock again shortly; this can lead
512  * to excessive master lookups and removals if we don't delay the release.
513  *
514  * Searching for an rsb means looking through both the normal list and toss
515  * list.  When found on the toss list the rsb is moved to the normal list with
516  * ref count of 1; when found on normal list the ref count is incremented.
517  *
518  * rsb's on the keep list are being used locally and refcounted.
519  * rsb's on the toss list are not being used locally, and are not refcounted.
520  *
521  * The toss list rsb's were either
522  * - previously used locally but not any more (were on keep list, then
523  *   moved to toss list when last refcount dropped)
524  * - created and put on toss list as a directory record for a lookup
525  *   (we are the dir node for the res, but are not using the res right now,
526  *   but some other node is)
527  *
528  * The purpose of find_rsb() is to return a refcounted rsb for local use.
529  * So, if the given rsb is on the toss list, it is moved to the keep list
530  * before being returned.
531  *
532  * toss_rsb() happens when all local usage of the rsb is done, i.e. no
533  * more refcounts exist, so the rsb is moved from the keep list to the
534  * toss list.
535  *
536  * rsb's on both keep and toss lists are used for doing a name to master
537  * lookups.  rsb's that are in use locally (and being refcounted) are on
538  * the keep list, rsb's that are not in use locally (not refcounted) and
539  * only exist for name/master lookups are on the toss list.
540  *
541  * rsb's on the toss list who's dir_nodeid is not local can have stale
542  * name/master mappings.  So, remote requests on such rsb's can potentially
543  * return with an error, which means the mapping is stale and needs to
544  * be updated with a new lookup.  (The idea behind MASTER UNCERTAIN and
545  * first_lkid is to keep only a single outstanding request on an rsb
546  * while that rsb has a potentially stale master.)
547  */
548 
549 static int find_rsb_dir(struct dlm_ls *ls, char *name, int len,
550 			uint32_t hash, uint32_t b,
551 			int dir_nodeid, int from_nodeid,
552 			unsigned int flags, struct dlm_rsb **r_ret)
553 {
554 	struct dlm_rsb *r = NULL;
555 	int our_nodeid = dlm_our_nodeid();
556 	int from_local = 0;
557 	int from_other = 0;
558 	int from_dir = 0;
559 	int create = 0;
560 	int error;
561 
562 	if (flags & R_RECEIVE_REQUEST) {
563 		if (from_nodeid == dir_nodeid)
564 			from_dir = 1;
565 		else
566 			from_other = 1;
567 	} else if (flags & R_REQUEST) {
568 		from_local = 1;
569 	}
570 
571 	/*
572 	 * flags & R_RECEIVE_RECOVER is from dlm_recover_master_copy, so
573 	 * from_nodeid has sent us a lock in dlm_recover_locks, believing
574 	 * we're the new master.  Our local recovery may not have set
575 	 * res_master_nodeid to our_nodeid yet, so allow either.  Don't
576 	 * create the rsb; dlm_recover_process_copy() will handle EBADR
577 	 * by resending.
578 	 *
579 	 * If someone sends us a request, we are the dir node, and we do
580 	 * not find the rsb anywhere, then recreate it.  This happens if
581 	 * someone sends us a request after we have removed/freed an rsb
582 	 * from our toss list.  (They sent a request instead of lookup
583 	 * because they are using an rsb from their toss list.)
584 	 */
585 
586 	if (from_local || from_dir ||
587 	    (from_other && (dir_nodeid == our_nodeid))) {
588 		create = 1;
589 	}
590 
591  retry:
592 	if (create) {
593 		error = pre_rsb_struct(ls);
594 		if (error < 0)
595 			goto out;
596 	}
597 
598 	spin_lock(&ls->ls_rsbtbl[b].lock);
599 
600 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
601 	if (error)
602 		goto do_toss;
603 
604 	/*
605 	 * rsb is active, so we can't check master_nodeid without lock_rsb.
606 	 */
607 
608 	kref_get(&r->res_ref);
609 	goto out_unlock;
610 
611 
612  do_toss:
613 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
614 	if (error)
615 		goto do_new;
616 
617 	/*
618 	 * rsb found inactive (master_nodeid may be out of date unless
619 	 * we are the dir_nodeid or were the master)  No other thread
620 	 * is using this rsb because it's on the toss list, so we can
621 	 * look at or update res_master_nodeid without lock_rsb.
622 	 */
623 
624 	if ((r->res_master_nodeid != our_nodeid) && from_other) {
625 		/* our rsb was not master, and another node (not the dir node)
626 		   has sent us a request */
627 		log_debug(ls, "find_rsb toss from_other %d master %d dir %d %s",
628 			  from_nodeid, r->res_master_nodeid, dir_nodeid,
629 			  r->res_name);
630 		error = -ENOTBLK;
631 		goto out_unlock;
632 	}
633 
634 	if ((r->res_master_nodeid != our_nodeid) && from_dir) {
635 		/* don't think this should ever happen */
636 		log_error(ls, "find_rsb toss from_dir %d master %d",
637 			  from_nodeid, r->res_master_nodeid);
638 		dlm_print_rsb(r);
639 		/* fix it and go on */
640 		r->res_master_nodeid = our_nodeid;
641 		r->res_nodeid = 0;
642 		rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
643 		r->res_first_lkid = 0;
644 	}
645 
646 	if (from_local && (r->res_master_nodeid != our_nodeid)) {
647 		/* Because we have held no locks on this rsb,
648 		   res_master_nodeid could have become stale. */
649 		rsb_set_flag(r, RSB_MASTER_UNCERTAIN);
650 		r->res_first_lkid = 0;
651 	}
652 
653 	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
654 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
655 	goto out_unlock;
656 
657 
658  do_new:
659 	/*
660 	 * rsb not found
661 	 */
662 
663 	if (error == -EBADR && !create)
664 		goto out_unlock;
665 
666 	error = get_rsb_struct(ls, name, len, &r);
667 	if (error == -EAGAIN) {
668 		spin_unlock(&ls->ls_rsbtbl[b].lock);
669 		goto retry;
670 	}
671 	if (error)
672 		goto out_unlock;
673 
674 	r->res_hash = hash;
675 	r->res_bucket = b;
676 	r->res_dir_nodeid = dir_nodeid;
677 	kref_init(&r->res_ref);
678 
679 	if (from_dir) {
680 		/* want to see how often this happens */
681 		log_debug(ls, "find_rsb new from_dir %d recreate %s",
682 			  from_nodeid, r->res_name);
683 		r->res_master_nodeid = our_nodeid;
684 		r->res_nodeid = 0;
685 		goto out_add;
686 	}
687 
688 	if (from_other && (dir_nodeid != our_nodeid)) {
689 		/* should never happen */
690 		log_error(ls, "find_rsb new from_other %d dir %d our %d %s",
691 			  from_nodeid, dir_nodeid, our_nodeid, r->res_name);
692 		dlm_free_rsb(r);
693 		r = NULL;
694 		error = -ENOTBLK;
695 		goto out_unlock;
696 	}
697 
698 	if (from_other) {
699 		log_debug(ls, "find_rsb new from_other %d dir %d %s",
700 			  from_nodeid, dir_nodeid, r->res_name);
701 	}
702 
703 	if (dir_nodeid == our_nodeid) {
704 		/* When we are the dir nodeid, we can set the master
705 		   node immediately */
706 		r->res_master_nodeid = our_nodeid;
707 		r->res_nodeid = 0;
708 	} else {
709 		/* set_master will send_lookup to dir_nodeid */
710 		r->res_master_nodeid = 0;
711 		r->res_nodeid = -1;
712 	}
713 
714  out_add:
715 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
716  out_unlock:
717 	spin_unlock(&ls->ls_rsbtbl[b].lock);
718  out:
719 	*r_ret = r;
720 	return error;
721 }
722 
723 /* During recovery, other nodes can send us new MSTCPY locks (from
724    dlm_recover_locks) before we've made ourself master (in
725    dlm_recover_masters). */
726 
727 static int find_rsb_nodir(struct dlm_ls *ls, char *name, int len,
728 			  uint32_t hash, uint32_t b,
729 			  int dir_nodeid, int from_nodeid,
730 			  unsigned int flags, struct dlm_rsb **r_ret)
731 {
732 	struct dlm_rsb *r = NULL;
733 	int our_nodeid = dlm_our_nodeid();
734 	int recover = (flags & R_RECEIVE_RECOVER);
735 	int error;
736 
737  retry:
738 	error = pre_rsb_struct(ls);
739 	if (error < 0)
740 		goto out;
741 
742 	spin_lock(&ls->ls_rsbtbl[b].lock);
743 
744 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
745 	if (error)
746 		goto do_toss;
747 
748 	/*
749 	 * rsb is active, so we can't check master_nodeid without lock_rsb.
750 	 */
751 
752 	kref_get(&r->res_ref);
753 	goto out_unlock;
754 
755 
756  do_toss:
757 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
758 	if (error)
759 		goto do_new;
760 
761 	/*
762 	 * rsb found inactive. No other thread is using this rsb because
763 	 * it's on the toss list, so we can look at or update
764 	 * res_master_nodeid without lock_rsb.
765 	 */
766 
767 	if (!recover && (r->res_master_nodeid != our_nodeid) && from_nodeid) {
768 		/* our rsb is not master, and another node has sent us a
769 		   request; this should never happen */
770 		log_error(ls, "find_rsb toss from_nodeid %d master %d dir %d",
771 			  from_nodeid, r->res_master_nodeid, dir_nodeid);
772 		dlm_print_rsb(r);
773 		error = -ENOTBLK;
774 		goto out_unlock;
775 	}
776 
777 	if (!recover && (r->res_master_nodeid != our_nodeid) &&
778 	    (dir_nodeid == our_nodeid)) {
779 		/* our rsb is not master, and we are dir; may as well fix it;
780 		   this should never happen */
781 		log_error(ls, "find_rsb toss our %d master %d dir %d",
782 			  our_nodeid, r->res_master_nodeid, dir_nodeid);
783 		dlm_print_rsb(r);
784 		r->res_master_nodeid = our_nodeid;
785 		r->res_nodeid = 0;
786 	}
787 
788 	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
789 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
790 	goto out_unlock;
791 
792 
793  do_new:
794 	/*
795 	 * rsb not found
796 	 */
797 
798 	error = get_rsb_struct(ls, name, len, &r);
799 	if (error == -EAGAIN) {
800 		spin_unlock(&ls->ls_rsbtbl[b].lock);
801 		goto retry;
802 	}
803 	if (error)
804 		goto out_unlock;
805 
806 	r->res_hash = hash;
807 	r->res_bucket = b;
808 	r->res_dir_nodeid = dir_nodeid;
809 	r->res_master_nodeid = dir_nodeid;
810 	r->res_nodeid = (dir_nodeid == our_nodeid) ? 0 : dir_nodeid;
811 	kref_init(&r->res_ref);
812 
813 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
814  out_unlock:
815 	spin_unlock(&ls->ls_rsbtbl[b].lock);
816  out:
817 	*r_ret = r;
818 	return error;
819 }
820 
821 static int find_rsb(struct dlm_ls *ls, char *name, int len, int from_nodeid,
822 		    unsigned int flags, struct dlm_rsb **r_ret)
823 {
824 	uint32_t hash, b;
825 	int dir_nodeid;
826 
827 	if (len > DLM_RESNAME_MAXLEN)
828 		return -EINVAL;
829 
830 	hash = jhash(name, len, 0);
831 	b = hash & (ls->ls_rsbtbl_size - 1);
832 
833 	dir_nodeid = dlm_hash2nodeid(ls, hash);
834 
835 	if (dlm_no_directory(ls))
836 		return find_rsb_nodir(ls, name, len, hash, b, dir_nodeid,
837 				      from_nodeid, flags, r_ret);
838 	else
839 		return find_rsb_dir(ls, name, len, hash, b, dir_nodeid,
840 				      from_nodeid, flags, r_ret);
841 }
842 
843 /* we have received a request and found that res_master_nodeid != our_nodeid,
844    so we need to return an error or make ourself the master */
845 
846 static int validate_master_nodeid(struct dlm_ls *ls, struct dlm_rsb *r,
847 				  int from_nodeid)
848 {
849 	if (dlm_no_directory(ls)) {
850 		log_error(ls, "find_rsb keep from_nodeid %d master %d dir %d",
851 			  from_nodeid, r->res_master_nodeid,
852 			  r->res_dir_nodeid);
853 		dlm_print_rsb(r);
854 		return -ENOTBLK;
855 	}
856 
857 	if (from_nodeid != r->res_dir_nodeid) {
858 		/* our rsb is not master, and another node (not the dir node)
859 	   	   has sent us a request.  this is much more common when our
860 	   	   master_nodeid is zero, so limit debug to non-zero.  */
861 
862 		if (r->res_master_nodeid) {
863 			log_debug(ls, "validate master from_other %d master %d "
864 				  "dir %d first %x %s", from_nodeid,
865 				  r->res_master_nodeid, r->res_dir_nodeid,
866 				  r->res_first_lkid, r->res_name);
867 		}
868 		return -ENOTBLK;
869 	} else {
870 		/* our rsb is not master, but the dir nodeid has sent us a
871 	   	   request; this could happen with master 0 / res_nodeid -1 */
872 
873 		if (r->res_master_nodeid) {
874 			log_error(ls, "validate master from_dir %d master %d "
875 				  "first %x %s",
876 				  from_nodeid, r->res_master_nodeid,
877 				  r->res_first_lkid, r->res_name);
878 		}
879 
880 		r->res_master_nodeid = dlm_our_nodeid();
881 		r->res_nodeid = 0;
882 		return 0;
883 	}
884 }
885 
886 static void __dlm_master_lookup(struct dlm_ls *ls, struct dlm_rsb *r, int our_nodeid,
887 				int from_nodeid, bool toss_list, unsigned int flags,
888 				int *r_nodeid, int *result)
889 {
890 	int fix_master = (flags & DLM_LU_RECOVER_MASTER);
891 	int from_master = (flags & DLM_LU_RECOVER_DIR);
892 
893 	if (r->res_dir_nodeid != our_nodeid) {
894 		/* should not happen, but may as well fix it and carry on */
895 		log_error(ls, "%s res_dir %d our %d %s", __func__,
896 			  r->res_dir_nodeid, our_nodeid, r->res_name);
897 		r->res_dir_nodeid = our_nodeid;
898 	}
899 
900 	if (fix_master && dlm_is_removed(ls, r->res_master_nodeid)) {
901 		/* Recovery uses this function to set a new master when
902 		 * the previous master failed.  Setting NEW_MASTER will
903 		 * force dlm_recover_masters to call recover_master on this
904 		 * rsb even though the res_nodeid is no longer removed.
905 		 */
906 
907 		r->res_master_nodeid = from_nodeid;
908 		r->res_nodeid = from_nodeid;
909 		rsb_set_flag(r, RSB_NEW_MASTER);
910 
911 		if (toss_list) {
912 			/* I don't think we should ever find it on toss list. */
913 			log_error(ls, "%s fix_master on toss", __func__);
914 			dlm_dump_rsb(r);
915 		}
916 	}
917 
918 	if (from_master && (r->res_master_nodeid != from_nodeid)) {
919 		/* this will happen if from_nodeid became master during
920 		 * a previous recovery cycle, and we aborted the previous
921 		 * cycle before recovering this master value
922 		 */
923 
924 		log_limit(ls, "%s from_master %d master_nodeid %d res_nodeid %d first %x %s",
925 			  __func__, from_nodeid, r->res_master_nodeid,
926 			  r->res_nodeid, r->res_first_lkid, r->res_name);
927 
928 		if (r->res_master_nodeid == our_nodeid) {
929 			log_error(ls, "from_master %d our_master", from_nodeid);
930 			dlm_dump_rsb(r);
931 			goto ret_assign;
932 		}
933 
934 		r->res_master_nodeid = from_nodeid;
935 		r->res_nodeid = from_nodeid;
936 		rsb_set_flag(r, RSB_NEW_MASTER);
937 	}
938 
939 	if (!r->res_master_nodeid) {
940 		/* this will happen if recovery happens while we're looking
941 		 * up the master for this rsb
942 		 */
943 
944 		log_debug(ls, "%s master 0 to %d first %x %s", __func__,
945 			  from_nodeid, r->res_first_lkid, r->res_name);
946 		r->res_master_nodeid = from_nodeid;
947 		r->res_nodeid = from_nodeid;
948 	}
949 
950 	if (!from_master && !fix_master &&
951 	    (r->res_master_nodeid == from_nodeid)) {
952 		/* this can happen when the master sends remove, the dir node
953 		 * finds the rsb on the keep list and ignores the remove,
954 		 * and the former master sends a lookup
955 		 */
956 
957 		log_limit(ls, "%s from master %d flags %x first %x %s",
958 			  __func__, from_nodeid, flags, r->res_first_lkid,
959 			  r->res_name);
960 	}
961 
962  ret_assign:
963 	*r_nodeid = r->res_master_nodeid;
964 	if (result)
965 		*result = DLM_LU_MATCH;
966 }
967 
968 /*
969  * We're the dir node for this res and another node wants to know the
970  * master nodeid.  During normal operation (non recovery) this is only
971  * called from receive_lookup(); master lookups when the local node is
972  * the dir node are done by find_rsb().
973  *
974  * normal operation, we are the dir node for a resource
975  * . _request_lock
976  * . set_master
977  * . send_lookup
978  * . receive_lookup
979  * . dlm_master_lookup flags 0
980  *
981  * recover directory, we are rebuilding dir for all resources
982  * . dlm_recover_directory
983  * . dlm_rcom_names
984  *   remote node sends back the rsb names it is master of and we are dir of
985  * . dlm_master_lookup RECOVER_DIR (fix_master 0, from_master 1)
986  *   we either create new rsb setting remote node as master, or find existing
987  *   rsb and set master to be the remote node.
988  *
989  * recover masters, we are finding the new master for resources
990  * . dlm_recover_masters
991  * . recover_master
992  * . dlm_send_rcom_lookup
993  * . receive_rcom_lookup
994  * . dlm_master_lookup RECOVER_MASTER (fix_master 1, from_master 0)
995  */
996 
997 int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, char *name, int len,
998 		      unsigned int flags, int *r_nodeid, int *result)
999 {
1000 	struct dlm_rsb *r = NULL;
1001 	uint32_t hash, b;
1002 	int our_nodeid = dlm_our_nodeid();
1003 	int dir_nodeid, error;
1004 
1005 	if (len > DLM_RESNAME_MAXLEN)
1006 		return -EINVAL;
1007 
1008 	if (from_nodeid == our_nodeid) {
1009 		log_error(ls, "dlm_master_lookup from our_nodeid %d flags %x",
1010 			  our_nodeid, flags);
1011 		return -EINVAL;
1012 	}
1013 
1014 	hash = jhash(name, len, 0);
1015 	b = hash & (ls->ls_rsbtbl_size - 1);
1016 
1017 	dir_nodeid = dlm_hash2nodeid(ls, hash);
1018 	if (dir_nodeid != our_nodeid) {
1019 		log_error(ls, "dlm_master_lookup from %d dir %d our %d h %x %d",
1020 			  from_nodeid, dir_nodeid, our_nodeid, hash,
1021 			  ls->ls_num_nodes);
1022 		*r_nodeid = -1;
1023 		return -EINVAL;
1024 	}
1025 
1026  retry:
1027 	error = pre_rsb_struct(ls);
1028 	if (error < 0)
1029 		return error;
1030 
1031 	spin_lock(&ls->ls_rsbtbl[b].lock);
1032 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
1033 	if (!error) {
1034 		/* because the rsb is active, we need to lock_rsb before
1035 		 * checking/changing re_master_nodeid
1036 		 */
1037 
1038 		hold_rsb(r);
1039 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1040 		lock_rsb(r);
1041 
1042 		__dlm_master_lookup(ls, r, our_nodeid, from_nodeid, false,
1043 				    flags, r_nodeid, result);
1044 
1045 		/* the rsb was active */
1046 		unlock_rsb(r);
1047 		put_rsb(r);
1048 
1049 		return 0;
1050 	}
1051 
1052 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
1053 	if (error)
1054 		goto not_found;
1055 
1056 	/* because the rsb is inactive (on toss list), it's not refcounted
1057 	 * and lock_rsb is not used, but is protected by the rsbtbl lock
1058 	 */
1059 
1060 	__dlm_master_lookup(ls, r, our_nodeid, from_nodeid, true, flags,
1061 			    r_nodeid, result);
1062 
1063 	r->res_toss_time = jiffies;
1064 	/* the rsb was inactive (on toss list) */
1065 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1066 
1067 	return 0;
1068 
1069  not_found:
1070 	error = get_rsb_struct(ls, name, len, &r);
1071 	if (error == -EAGAIN) {
1072 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1073 		goto retry;
1074 	}
1075 	if (error)
1076 		goto out_unlock;
1077 
1078 	r->res_hash = hash;
1079 	r->res_bucket = b;
1080 	r->res_dir_nodeid = our_nodeid;
1081 	r->res_master_nodeid = from_nodeid;
1082 	r->res_nodeid = from_nodeid;
1083 	kref_init(&r->res_ref);
1084 	r->res_toss_time = jiffies;
1085 
1086 	error = rsb_insert(r, &ls->ls_rsbtbl[b].toss);
1087 	if (error) {
1088 		/* should never happen */
1089 		dlm_free_rsb(r);
1090 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1091 		goto retry;
1092 	}
1093 
1094 	if (result)
1095 		*result = DLM_LU_ADD;
1096 	*r_nodeid = from_nodeid;
1097  out_unlock:
1098 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1099 	return error;
1100 }
1101 
1102 static void dlm_dump_rsb_hash(struct dlm_ls *ls, uint32_t hash)
1103 {
1104 	struct rb_node *n;
1105 	struct dlm_rsb *r;
1106 	int i;
1107 
1108 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1109 		spin_lock(&ls->ls_rsbtbl[i].lock);
1110 		for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
1111 			r = rb_entry(n, struct dlm_rsb, res_hashnode);
1112 			if (r->res_hash == hash)
1113 				dlm_dump_rsb(r);
1114 		}
1115 		spin_unlock(&ls->ls_rsbtbl[i].lock);
1116 	}
1117 }
1118 
1119 void dlm_dump_rsb_name(struct dlm_ls *ls, char *name, int len)
1120 {
1121 	struct dlm_rsb *r = NULL;
1122 	uint32_t hash, b;
1123 	int error;
1124 
1125 	hash = jhash(name, len, 0);
1126 	b = hash & (ls->ls_rsbtbl_size - 1);
1127 
1128 	spin_lock(&ls->ls_rsbtbl[b].lock);
1129 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
1130 	if (!error)
1131 		goto out_dump;
1132 
1133 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
1134 	if (error)
1135 		goto out;
1136  out_dump:
1137 	dlm_dump_rsb(r);
1138  out:
1139 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1140 }
1141 
1142 static void toss_rsb(struct kref *kref)
1143 {
1144 	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
1145 	struct dlm_ls *ls = r->res_ls;
1146 
1147 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
1148 	kref_init(&r->res_ref);
1149 	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[r->res_bucket].keep);
1150 	rsb_insert(r, &ls->ls_rsbtbl[r->res_bucket].toss);
1151 	r->res_toss_time = jiffies;
1152 	ls->ls_rsbtbl[r->res_bucket].flags |= DLM_RTF_SHRINK;
1153 	if (r->res_lvbptr) {
1154 		dlm_free_lvb(r->res_lvbptr);
1155 		r->res_lvbptr = NULL;
1156 	}
1157 }
1158 
1159 /* See comment for unhold_lkb */
1160 
1161 static void unhold_rsb(struct dlm_rsb *r)
1162 {
1163 	int rv;
1164 	rv = kref_put(&r->res_ref, toss_rsb);
1165 	DLM_ASSERT(!rv, dlm_dump_rsb(r););
1166 }
1167 
1168 static void kill_rsb(struct kref *kref)
1169 {
1170 	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
1171 
1172 	/* All work is done after the return from kref_put() so we
1173 	   can release the write_lock before the remove and free. */
1174 
1175 	DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
1176 	DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
1177 	DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
1178 	DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
1179 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
1180 	DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
1181 }
1182 
1183 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
1184    The rsb must exist as long as any lkb's for it do. */
1185 
1186 static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
1187 {
1188 	hold_rsb(r);
1189 	lkb->lkb_resource = r;
1190 }
1191 
1192 static void detach_lkb(struct dlm_lkb *lkb)
1193 {
1194 	if (lkb->lkb_resource) {
1195 		put_rsb(lkb->lkb_resource);
1196 		lkb->lkb_resource = NULL;
1197 	}
1198 }
1199 
1200 static int _create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret,
1201 		       int start, int end)
1202 {
1203 	struct dlm_lkb *lkb;
1204 	int rv;
1205 
1206 	lkb = dlm_allocate_lkb(ls);
1207 	if (!lkb)
1208 		return -ENOMEM;
1209 
1210 	lkb->lkb_nodeid = -1;
1211 	lkb->lkb_grmode = DLM_LOCK_IV;
1212 	kref_init(&lkb->lkb_ref);
1213 	INIT_LIST_HEAD(&lkb->lkb_ownqueue);
1214 	INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
1215 #ifdef CONFIG_DLM_DEPRECATED_API
1216 	INIT_LIST_HEAD(&lkb->lkb_time_list);
1217 #endif
1218 	INIT_LIST_HEAD(&lkb->lkb_cb_list);
1219 	mutex_init(&lkb->lkb_cb_mutex);
1220 	INIT_WORK(&lkb->lkb_cb_work, dlm_callback_work);
1221 
1222 	idr_preload(GFP_NOFS);
1223 	spin_lock(&ls->ls_lkbidr_spin);
1224 	rv = idr_alloc(&ls->ls_lkbidr, lkb, start, end, GFP_NOWAIT);
1225 	if (rv >= 0)
1226 		lkb->lkb_id = rv;
1227 	spin_unlock(&ls->ls_lkbidr_spin);
1228 	idr_preload_end();
1229 
1230 	if (rv < 0) {
1231 		log_error(ls, "create_lkb idr error %d", rv);
1232 		dlm_free_lkb(lkb);
1233 		return rv;
1234 	}
1235 
1236 	*lkb_ret = lkb;
1237 	return 0;
1238 }
1239 
1240 static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
1241 {
1242 	return _create_lkb(ls, lkb_ret, 1, 0);
1243 }
1244 
1245 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
1246 {
1247 	struct dlm_lkb *lkb;
1248 
1249 	spin_lock(&ls->ls_lkbidr_spin);
1250 	lkb = idr_find(&ls->ls_lkbidr, lkid);
1251 	if (lkb)
1252 		kref_get(&lkb->lkb_ref);
1253 	spin_unlock(&ls->ls_lkbidr_spin);
1254 
1255 	*lkb_ret = lkb;
1256 	return lkb ? 0 : -ENOENT;
1257 }
1258 
1259 static void kill_lkb(struct kref *kref)
1260 {
1261 	struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
1262 
1263 	/* All work is done after the return from kref_put() so we
1264 	   can release the write_lock before the detach_lkb */
1265 
1266 	DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
1267 }
1268 
1269 /* __put_lkb() is used when an lkb may not have an rsb attached to
1270    it so we need to provide the lockspace explicitly */
1271 
1272 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
1273 {
1274 	uint32_t lkid = lkb->lkb_id;
1275 	int rv;
1276 
1277 	rv = kref_put_lock(&lkb->lkb_ref, kill_lkb,
1278 			   &ls->ls_lkbidr_spin);
1279 	if (rv) {
1280 		idr_remove(&ls->ls_lkbidr, lkid);
1281 		spin_unlock(&ls->ls_lkbidr_spin);
1282 
1283 		detach_lkb(lkb);
1284 
1285 		/* for local/process lkbs, lvbptr points to caller's lksb */
1286 		if (lkb->lkb_lvbptr && is_master_copy(lkb))
1287 			dlm_free_lvb(lkb->lkb_lvbptr);
1288 		dlm_free_lkb(lkb);
1289 	}
1290 
1291 	return rv;
1292 }
1293 
1294 int dlm_put_lkb(struct dlm_lkb *lkb)
1295 {
1296 	struct dlm_ls *ls;
1297 
1298 	DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb););
1299 	DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb););
1300 
1301 	ls = lkb->lkb_resource->res_ls;
1302 	return __put_lkb(ls, lkb);
1303 }
1304 
1305 /* This is only called to add a reference when the code already holds
1306    a valid reference to the lkb, so there's no need for locking. */
1307 
1308 static inline void hold_lkb(struct dlm_lkb *lkb)
1309 {
1310 	kref_get(&lkb->lkb_ref);
1311 }
1312 
1313 static void unhold_lkb_assert(struct kref *kref)
1314 {
1315 	struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
1316 
1317 	DLM_ASSERT(false, dlm_print_lkb(lkb););
1318 }
1319 
1320 /* This is called when we need to remove a reference and are certain
1321    it's not the last ref.  e.g. del_lkb is always called between a
1322    find_lkb/put_lkb and is always the inverse of a previous add_lkb.
1323    put_lkb would work fine, but would involve unnecessary locking */
1324 
1325 static inline void unhold_lkb(struct dlm_lkb *lkb)
1326 {
1327 	kref_put(&lkb->lkb_ref, unhold_lkb_assert);
1328 }
1329 
1330 static void lkb_add_ordered(struct list_head *new, struct list_head *head,
1331 			    int mode)
1332 {
1333 	struct dlm_lkb *lkb = NULL, *iter;
1334 
1335 	list_for_each_entry(iter, head, lkb_statequeue)
1336 		if (iter->lkb_rqmode < mode) {
1337 			lkb = iter;
1338 			list_add_tail(new, &iter->lkb_statequeue);
1339 			break;
1340 		}
1341 
1342 	if (!lkb)
1343 		list_add_tail(new, head);
1344 }
1345 
1346 /* add/remove lkb to rsb's grant/convert/wait queue */
1347 
1348 static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status)
1349 {
1350 	kref_get(&lkb->lkb_ref);
1351 
1352 	DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
1353 
1354 	lkb->lkb_timestamp = ktime_get();
1355 
1356 	lkb->lkb_status = status;
1357 
1358 	switch (status) {
1359 	case DLM_LKSTS_WAITING:
1360 		if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
1361 			list_add(&lkb->lkb_statequeue, &r->res_waitqueue);
1362 		else
1363 			list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue);
1364 		break;
1365 	case DLM_LKSTS_GRANTED:
1366 		/* convention says granted locks kept in order of grmode */
1367 		lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue,
1368 				lkb->lkb_grmode);
1369 		break;
1370 	case DLM_LKSTS_CONVERT:
1371 		if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
1372 			list_add(&lkb->lkb_statequeue, &r->res_convertqueue);
1373 		else
1374 			list_add_tail(&lkb->lkb_statequeue,
1375 				      &r->res_convertqueue);
1376 		break;
1377 	default:
1378 		DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status););
1379 	}
1380 }
1381 
1382 static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
1383 {
1384 	lkb->lkb_status = 0;
1385 	list_del(&lkb->lkb_statequeue);
1386 	unhold_lkb(lkb);
1387 }
1388 
1389 static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
1390 {
1391 	hold_lkb(lkb);
1392 	del_lkb(r, lkb);
1393 	add_lkb(r, lkb, sts);
1394 	unhold_lkb(lkb);
1395 }
1396 
1397 static int msg_reply_type(int mstype)
1398 {
1399 	switch (mstype) {
1400 	case DLM_MSG_REQUEST:
1401 		return DLM_MSG_REQUEST_REPLY;
1402 	case DLM_MSG_CONVERT:
1403 		return DLM_MSG_CONVERT_REPLY;
1404 	case DLM_MSG_UNLOCK:
1405 		return DLM_MSG_UNLOCK_REPLY;
1406 	case DLM_MSG_CANCEL:
1407 		return DLM_MSG_CANCEL_REPLY;
1408 	case DLM_MSG_LOOKUP:
1409 		return DLM_MSG_LOOKUP_REPLY;
1410 	}
1411 	return -1;
1412 }
1413 
1414 /* add/remove lkb from global waiters list of lkb's waiting for
1415    a reply from a remote node */
1416 
1417 static int add_to_waiters(struct dlm_lkb *lkb, int mstype, int to_nodeid)
1418 {
1419 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1420 	int error = 0;
1421 
1422 	mutex_lock(&ls->ls_waiters_mutex);
1423 
1424 	if (is_overlap_unlock(lkb) ||
1425 	    (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
1426 		error = -EINVAL;
1427 		goto out;
1428 	}
1429 
1430 	if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
1431 		switch (mstype) {
1432 		case DLM_MSG_UNLOCK:
1433 			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
1434 			break;
1435 		case DLM_MSG_CANCEL:
1436 			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
1437 			break;
1438 		default:
1439 			error = -EBUSY;
1440 			goto out;
1441 		}
1442 		lkb->lkb_wait_count++;
1443 		hold_lkb(lkb);
1444 
1445 		log_debug(ls, "addwait %x cur %d overlap %d count %d f %x",
1446 			  lkb->lkb_id, lkb->lkb_wait_type, mstype,
1447 			  lkb->lkb_wait_count, lkb->lkb_flags);
1448 		goto out;
1449 	}
1450 
1451 	DLM_ASSERT(!lkb->lkb_wait_count,
1452 		   dlm_print_lkb(lkb);
1453 		   printk("wait_count %d\n", lkb->lkb_wait_count););
1454 
1455 	lkb->lkb_wait_count++;
1456 	lkb->lkb_wait_type = mstype;
1457 	lkb->lkb_wait_nodeid = to_nodeid; /* for debugging */
1458 	hold_lkb(lkb);
1459 	list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
1460  out:
1461 	if (error)
1462 		log_error(ls, "addwait error %x %d flags %x %d %d %s",
1463 			  lkb->lkb_id, error, lkb->lkb_flags, mstype,
1464 			  lkb->lkb_wait_type, lkb->lkb_resource->res_name);
1465 	mutex_unlock(&ls->ls_waiters_mutex);
1466 	return error;
1467 }
1468 
1469 /* We clear the RESEND flag because we might be taking an lkb off the waiters
1470    list as part of process_requestqueue (e.g. a lookup that has an optimized
1471    request reply on the requestqueue) between dlm_recover_waiters_pre() which
1472    set RESEND and dlm_recover_waiters_post() */
1473 
1474 static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype,
1475 				struct dlm_message *ms)
1476 {
1477 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1478 	int overlap_done = 0;
1479 
1480 	if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
1481 		log_debug(ls, "remwait %x unlock_reply overlap", lkb->lkb_id);
1482 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
1483 		overlap_done = 1;
1484 		goto out_del;
1485 	}
1486 
1487 	if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
1488 		log_debug(ls, "remwait %x cancel_reply overlap", lkb->lkb_id);
1489 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
1490 		overlap_done = 1;
1491 		goto out_del;
1492 	}
1493 
1494 	/* Cancel state was preemptively cleared by a successful convert,
1495 	   see next comment, nothing to do. */
1496 
1497 	if ((mstype == DLM_MSG_CANCEL_REPLY) &&
1498 	    (lkb->lkb_wait_type != DLM_MSG_CANCEL)) {
1499 		log_debug(ls, "remwait %x cancel_reply wait_type %d",
1500 			  lkb->lkb_id, lkb->lkb_wait_type);
1501 		return -1;
1502 	}
1503 
1504 	/* Remove for the convert reply, and premptively remove for the
1505 	   cancel reply.  A convert has been granted while there's still
1506 	   an outstanding cancel on it (the cancel is moot and the result
1507 	   in the cancel reply should be 0).  We preempt the cancel reply
1508 	   because the app gets the convert result and then can follow up
1509 	   with another op, like convert.  This subsequent op would see the
1510 	   lingering state of the cancel and fail with -EBUSY. */
1511 
1512 	if ((mstype == DLM_MSG_CONVERT_REPLY) &&
1513 	    (lkb->lkb_wait_type == DLM_MSG_CONVERT) &&
1514 	    is_overlap_cancel(lkb) && ms && !ms->m_result) {
1515 		log_debug(ls, "remwait %x convert_reply zap overlap_cancel",
1516 			  lkb->lkb_id);
1517 		lkb->lkb_wait_type = 0;
1518 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
1519 		lkb->lkb_wait_count--;
1520 		unhold_lkb(lkb);
1521 		goto out_del;
1522 	}
1523 
1524 	/* N.B. type of reply may not always correspond to type of original
1525 	   msg due to lookup->request optimization, verify others? */
1526 
1527 	if (lkb->lkb_wait_type) {
1528 		lkb->lkb_wait_type = 0;
1529 		goto out_del;
1530 	}
1531 
1532 	log_error(ls, "remwait error %x remote %d %x msg %d flags %x no wait",
1533 		  lkb->lkb_id, ms ? le32_to_cpu(ms->m_header.h_nodeid) : 0,
1534 		  lkb->lkb_remid, mstype, lkb->lkb_flags);
1535 	return -1;
1536 
1537  out_del:
1538 	/* the force-unlock/cancel has completed and we haven't recvd a reply
1539 	   to the op that was in progress prior to the unlock/cancel; we
1540 	   give up on any reply to the earlier op.  FIXME: not sure when/how
1541 	   this would happen */
1542 
1543 	if (overlap_done && lkb->lkb_wait_type) {
1544 		log_error(ls, "remwait error %x reply %d wait_type %d overlap",
1545 			  lkb->lkb_id, mstype, lkb->lkb_wait_type);
1546 		lkb->lkb_wait_count--;
1547 		unhold_lkb(lkb);
1548 		lkb->lkb_wait_type = 0;
1549 	}
1550 
1551 	DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
1552 
1553 	lkb->lkb_flags &= ~DLM_IFL_RESEND;
1554 	lkb->lkb_wait_count--;
1555 	if (!lkb->lkb_wait_count)
1556 		list_del_init(&lkb->lkb_wait_reply);
1557 	unhold_lkb(lkb);
1558 	return 0;
1559 }
1560 
1561 static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
1562 {
1563 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1564 	int error;
1565 
1566 	mutex_lock(&ls->ls_waiters_mutex);
1567 	error = _remove_from_waiters(lkb, mstype, NULL);
1568 	mutex_unlock(&ls->ls_waiters_mutex);
1569 	return error;
1570 }
1571 
1572 /* Handles situations where we might be processing a "fake" or "stub" reply in
1573    which we can't try to take waiters_mutex again. */
1574 
1575 static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
1576 {
1577 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1578 	int error;
1579 
1580 	if (ms->m_flags != cpu_to_le32(DLM_IFL_STUB_MS))
1581 		mutex_lock(&ls->ls_waiters_mutex);
1582 	error = _remove_from_waiters(lkb, le32_to_cpu(ms->m_type), ms);
1583 	if (ms->m_flags != cpu_to_le32(DLM_IFL_STUB_MS))
1584 		mutex_unlock(&ls->ls_waiters_mutex);
1585 	return error;
1586 }
1587 
1588 /* If there's an rsb for the same resource being removed, ensure
1589  * that the remove message is sent before the new lookup message.
1590  */
1591 
1592 #define DLM_WAIT_PENDING_COND(ls, r)		\
1593 	(ls->ls_remove_len &&			\
1594 	 !rsb_cmp(r, ls->ls_remove_name,	\
1595 		  ls->ls_remove_len))
1596 
1597 static void wait_pending_remove(struct dlm_rsb *r)
1598 {
1599 	struct dlm_ls *ls = r->res_ls;
1600  restart:
1601 	spin_lock(&ls->ls_remove_spin);
1602 	if (DLM_WAIT_PENDING_COND(ls, r)) {
1603 		log_debug(ls, "delay lookup for remove dir %d %s",
1604 			  r->res_dir_nodeid, r->res_name);
1605 		spin_unlock(&ls->ls_remove_spin);
1606 		wait_event(ls->ls_remove_wait, !DLM_WAIT_PENDING_COND(ls, r));
1607 		goto restart;
1608 	}
1609 	spin_unlock(&ls->ls_remove_spin);
1610 }
1611 
1612 /*
1613  * ls_remove_spin protects ls_remove_name and ls_remove_len which are
1614  * read by other threads in wait_pending_remove.  ls_remove_names
1615  * and ls_remove_lens are only used by the scan thread, so they do
1616  * not need protection.
1617  */
1618 
1619 static void shrink_bucket(struct dlm_ls *ls, int b)
1620 {
1621 	struct rb_node *n, *next;
1622 	struct dlm_rsb *r;
1623 	char *name;
1624 	int our_nodeid = dlm_our_nodeid();
1625 	int remote_count = 0;
1626 	int need_shrink = 0;
1627 	int i, len, rv;
1628 
1629 	memset(&ls->ls_remove_lens, 0, sizeof(int) * DLM_REMOVE_NAMES_MAX);
1630 
1631 	spin_lock(&ls->ls_rsbtbl[b].lock);
1632 
1633 	if (!(ls->ls_rsbtbl[b].flags & DLM_RTF_SHRINK)) {
1634 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1635 		return;
1636 	}
1637 
1638 	for (n = rb_first(&ls->ls_rsbtbl[b].toss); n; n = next) {
1639 		next = rb_next(n);
1640 		r = rb_entry(n, struct dlm_rsb, res_hashnode);
1641 
1642 		/* If we're the directory record for this rsb, and
1643 		   we're not the master of it, then we need to wait
1644 		   for the master node to send us a dir remove for
1645 		   before removing the dir record. */
1646 
1647 		if (!dlm_no_directory(ls) &&
1648 		    (r->res_master_nodeid != our_nodeid) &&
1649 		    (dlm_dir_nodeid(r) == our_nodeid)) {
1650 			continue;
1651 		}
1652 
1653 		need_shrink = 1;
1654 
1655 		if (!time_after_eq(jiffies, r->res_toss_time +
1656 				   dlm_config.ci_toss_secs * HZ)) {
1657 			continue;
1658 		}
1659 
1660 		if (!dlm_no_directory(ls) &&
1661 		    (r->res_master_nodeid == our_nodeid) &&
1662 		    (dlm_dir_nodeid(r) != our_nodeid)) {
1663 
1664 			/* We're the master of this rsb but we're not
1665 			   the directory record, so we need to tell the
1666 			   dir node to remove the dir record. */
1667 
1668 			ls->ls_remove_lens[remote_count] = r->res_length;
1669 			memcpy(ls->ls_remove_names[remote_count], r->res_name,
1670 			       DLM_RESNAME_MAXLEN);
1671 			remote_count++;
1672 
1673 			if (remote_count >= DLM_REMOVE_NAMES_MAX)
1674 				break;
1675 			continue;
1676 		}
1677 
1678 		if (!kref_put(&r->res_ref, kill_rsb)) {
1679 			log_error(ls, "tossed rsb in use %s", r->res_name);
1680 			continue;
1681 		}
1682 
1683 		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
1684 		dlm_free_rsb(r);
1685 	}
1686 
1687 	if (need_shrink)
1688 		ls->ls_rsbtbl[b].flags |= DLM_RTF_SHRINK;
1689 	else
1690 		ls->ls_rsbtbl[b].flags &= ~DLM_RTF_SHRINK;
1691 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1692 
1693 	/*
1694 	 * While searching for rsb's to free, we found some that require
1695 	 * remote removal.  We leave them in place and find them again here
1696 	 * so there is a very small gap between removing them from the toss
1697 	 * list and sending the removal.  Keeping this gap small is
1698 	 * important to keep us (the master node) from being out of sync
1699 	 * with the remote dir node for very long.
1700 	 *
1701 	 * From the time the rsb is removed from toss until just after
1702 	 * send_remove, the rsb name is saved in ls_remove_name.  A new
1703 	 * lookup checks this to ensure that a new lookup message for the
1704 	 * same resource name is not sent just before the remove message.
1705 	 */
1706 
1707 	for (i = 0; i < remote_count; i++) {
1708 		name = ls->ls_remove_names[i];
1709 		len = ls->ls_remove_lens[i];
1710 
1711 		spin_lock(&ls->ls_rsbtbl[b].lock);
1712 		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
1713 		if (rv) {
1714 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1715 			log_debug(ls, "remove_name not toss %s", name);
1716 			continue;
1717 		}
1718 
1719 		if (r->res_master_nodeid != our_nodeid) {
1720 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1721 			log_debug(ls, "remove_name master %d dir %d our %d %s",
1722 				  r->res_master_nodeid, r->res_dir_nodeid,
1723 				  our_nodeid, name);
1724 			continue;
1725 		}
1726 
1727 		if (r->res_dir_nodeid == our_nodeid) {
1728 			/* should never happen */
1729 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1730 			log_error(ls, "remove_name dir %d master %d our %d %s",
1731 				  r->res_dir_nodeid, r->res_master_nodeid,
1732 				  our_nodeid, name);
1733 			continue;
1734 		}
1735 
1736 		if (!time_after_eq(jiffies, r->res_toss_time +
1737 				   dlm_config.ci_toss_secs * HZ)) {
1738 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1739 			log_debug(ls, "remove_name toss_time %lu now %lu %s",
1740 				  r->res_toss_time, jiffies, name);
1741 			continue;
1742 		}
1743 
1744 		if (!kref_put(&r->res_ref, kill_rsb)) {
1745 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1746 			log_error(ls, "remove_name in use %s", name);
1747 			continue;
1748 		}
1749 
1750 		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
1751 
1752 		/* block lookup of same name until we've sent remove */
1753 		spin_lock(&ls->ls_remove_spin);
1754 		ls->ls_remove_len = len;
1755 		memcpy(ls->ls_remove_name, name, DLM_RESNAME_MAXLEN);
1756 		spin_unlock(&ls->ls_remove_spin);
1757 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1758 
1759 		send_remove(r);
1760 
1761 		/* allow lookup of name again */
1762 		spin_lock(&ls->ls_remove_spin);
1763 		ls->ls_remove_len = 0;
1764 		memset(ls->ls_remove_name, 0, DLM_RESNAME_MAXLEN);
1765 		spin_unlock(&ls->ls_remove_spin);
1766 		wake_up(&ls->ls_remove_wait);
1767 
1768 		dlm_free_rsb(r);
1769 	}
1770 }
1771 
1772 void dlm_scan_rsbs(struct dlm_ls *ls)
1773 {
1774 	int i;
1775 
1776 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1777 		shrink_bucket(ls, i);
1778 		if (dlm_locking_stopped(ls))
1779 			break;
1780 		cond_resched();
1781 	}
1782 }
1783 
1784 #ifdef CONFIG_DLM_DEPRECATED_API
1785 static void add_timeout(struct dlm_lkb *lkb)
1786 {
1787 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1788 
1789 	if (is_master_copy(lkb))
1790 		return;
1791 
1792 	if (test_bit(LSFL_TIMEWARN, &ls->ls_flags) &&
1793 	    !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1794 		lkb->lkb_flags |= DLM_IFL_WATCH_TIMEWARN;
1795 		goto add_it;
1796 	}
1797 	if (lkb->lkb_exflags & DLM_LKF_TIMEOUT)
1798 		goto add_it;
1799 	return;
1800 
1801  add_it:
1802 	DLM_ASSERT(list_empty(&lkb->lkb_time_list), dlm_print_lkb(lkb););
1803 	mutex_lock(&ls->ls_timeout_mutex);
1804 	hold_lkb(lkb);
1805 	list_add_tail(&lkb->lkb_time_list, &ls->ls_timeout);
1806 	mutex_unlock(&ls->ls_timeout_mutex);
1807 }
1808 
1809 static void del_timeout(struct dlm_lkb *lkb)
1810 {
1811 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1812 
1813 	mutex_lock(&ls->ls_timeout_mutex);
1814 	if (!list_empty(&lkb->lkb_time_list)) {
1815 		list_del_init(&lkb->lkb_time_list);
1816 		unhold_lkb(lkb);
1817 	}
1818 	mutex_unlock(&ls->ls_timeout_mutex);
1819 }
1820 
1821 /* FIXME: is it safe to look at lkb_exflags, lkb_flags, lkb_timestamp, and
1822    lkb_lksb_timeout without lock_rsb?  Note: we can't lock timeout_mutex
1823    and then lock rsb because of lock ordering in add_timeout.  We may need
1824    to specify some special timeout-related bits in the lkb that are just to
1825    be accessed under the timeout_mutex. */
1826 
1827 void dlm_scan_timeout(struct dlm_ls *ls)
1828 {
1829 	struct dlm_rsb *r;
1830 	struct dlm_lkb *lkb = NULL, *iter;
1831 	int do_cancel, do_warn;
1832 	s64 wait_us;
1833 
1834 	for (;;) {
1835 		if (dlm_locking_stopped(ls))
1836 			break;
1837 
1838 		do_cancel = 0;
1839 		do_warn = 0;
1840 		mutex_lock(&ls->ls_timeout_mutex);
1841 		list_for_each_entry(iter, &ls->ls_timeout, lkb_time_list) {
1842 
1843 			wait_us = ktime_to_us(ktime_sub(ktime_get(),
1844 							iter->lkb_timestamp));
1845 
1846 			if ((iter->lkb_exflags & DLM_LKF_TIMEOUT) &&
1847 			    wait_us >= (iter->lkb_timeout_cs * 10000))
1848 				do_cancel = 1;
1849 
1850 			if ((iter->lkb_flags & DLM_IFL_WATCH_TIMEWARN) &&
1851 			    wait_us >= dlm_config.ci_timewarn_cs * 10000)
1852 				do_warn = 1;
1853 
1854 			if (!do_cancel && !do_warn)
1855 				continue;
1856 			hold_lkb(iter);
1857 			lkb = iter;
1858 			break;
1859 		}
1860 		mutex_unlock(&ls->ls_timeout_mutex);
1861 
1862 		if (!lkb)
1863 			break;
1864 
1865 		r = lkb->lkb_resource;
1866 		hold_rsb(r);
1867 		lock_rsb(r);
1868 
1869 		if (do_warn) {
1870 			/* clear flag so we only warn once */
1871 			lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1872 			if (!(lkb->lkb_exflags & DLM_LKF_TIMEOUT))
1873 				del_timeout(lkb);
1874 			dlm_timeout_warn(lkb);
1875 		}
1876 
1877 		if (do_cancel) {
1878 			log_debug(ls, "timeout cancel %x node %d %s",
1879 				  lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1880 			lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1881 			lkb->lkb_flags |= DLM_IFL_TIMEOUT_CANCEL;
1882 			del_timeout(lkb);
1883 			_cancel_lock(r, lkb);
1884 		}
1885 
1886 		unlock_rsb(r);
1887 		unhold_rsb(r);
1888 		dlm_put_lkb(lkb);
1889 	}
1890 }
1891 
1892 /* This is only called by dlm_recoverd, and we rely on dlm_ls_stop() stopping
1893    dlm_recoverd before checking/setting ls_recover_begin. */
1894 
1895 void dlm_adjust_timeouts(struct dlm_ls *ls)
1896 {
1897 	struct dlm_lkb *lkb;
1898 	u64 adj_us = jiffies_to_usecs(jiffies - ls->ls_recover_begin);
1899 
1900 	ls->ls_recover_begin = 0;
1901 	mutex_lock(&ls->ls_timeout_mutex);
1902 	list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list)
1903 		lkb->lkb_timestamp = ktime_add_us(lkb->lkb_timestamp, adj_us);
1904 	mutex_unlock(&ls->ls_timeout_mutex);
1905 }
1906 #else
1907 static void add_timeout(struct dlm_lkb *lkb) { }
1908 static void del_timeout(struct dlm_lkb *lkb) { }
1909 #endif
1910 
1911 /* lkb is master or local copy */
1912 
1913 static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1914 {
1915 	int b, len = r->res_ls->ls_lvblen;
1916 
1917 	/* b=1 lvb returned to caller
1918 	   b=0 lvb written to rsb or invalidated
1919 	   b=-1 do nothing */
1920 
1921 	b =  dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1922 
1923 	if (b == 1) {
1924 		if (!lkb->lkb_lvbptr)
1925 			return;
1926 
1927 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1928 			return;
1929 
1930 		if (!r->res_lvbptr)
1931 			return;
1932 
1933 		memcpy(lkb->lkb_lvbptr, r->res_lvbptr, len);
1934 		lkb->lkb_lvbseq = r->res_lvbseq;
1935 
1936 	} else if (b == 0) {
1937 		if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1938 			rsb_set_flag(r, RSB_VALNOTVALID);
1939 			return;
1940 		}
1941 
1942 		if (!lkb->lkb_lvbptr)
1943 			return;
1944 
1945 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1946 			return;
1947 
1948 		if (!r->res_lvbptr)
1949 			r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1950 
1951 		if (!r->res_lvbptr)
1952 			return;
1953 
1954 		memcpy(r->res_lvbptr, lkb->lkb_lvbptr, len);
1955 		r->res_lvbseq++;
1956 		lkb->lkb_lvbseq = r->res_lvbseq;
1957 		rsb_clear_flag(r, RSB_VALNOTVALID);
1958 	}
1959 
1960 	if (rsb_flag(r, RSB_VALNOTVALID))
1961 		lkb->lkb_sbflags |= DLM_SBF_VALNOTVALID;
1962 }
1963 
1964 static void set_lvb_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1965 {
1966 	if (lkb->lkb_grmode < DLM_LOCK_PW)
1967 		return;
1968 
1969 	if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1970 		rsb_set_flag(r, RSB_VALNOTVALID);
1971 		return;
1972 	}
1973 
1974 	if (!lkb->lkb_lvbptr)
1975 		return;
1976 
1977 	if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1978 		return;
1979 
1980 	if (!r->res_lvbptr)
1981 		r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1982 
1983 	if (!r->res_lvbptr)
1984 		return;
1985 
1986 	memcpy(r->res_lvbptr, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
1987 	r->res_lvbseq++;
1988 	rsb_clear_flag(r, RSB_VALNOTVALID);
1989 }
1990 
1991 /* lkb is process copy (pc) */
1992 
1993 static void set_lvb_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1994 			    struct dlm_message *ms)
1995 {
1996 	int b;
1997 
1998 	if (!lkb->lkb_lvbptr)
1999 		return;
2000 
2001 	if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
2002 		return;
2003 
2004 	b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
2005 	if (b == 1) {
2006 		int len = receive_extralen(ms);
2007 		if (len > r->res_ls->ls_lvblen)
2008 			len = r->res_ls->ls_lvblen;
2009 		memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
2010 		lkb->lkb_lvbseq = le32_to_cpu(ms->m_lvbseq);
2011 	}
2012 }
2013 
2014 /* Manipulate lkb's on rsb's convert/granted/waiting queues
2015    remove_lock -- used for unlock, removes lkb from granted
2016    revert_lock -- used for cancel, moves lkb from convert to granted
2017    grant_lock  -- used for request and convert, adds lkb to granted or
2018                   moves lkb from convert or waiting to granted
2019 
2020    Each of these is used for master or local copy lkb's.  There is
2021    also a _pc() variation used to make the corresponding change on
2022    a process copy (pc) lkb. */
2023 
2024 static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2025 {
2026 	del_lkb(r, lkb);
2027 	lkb->lkb_grmode = DLM_LOCK_IV;
2028 	/* this unhold undoes the original ref from create_lkb()
2029 	   so this leads to the lkb being freed */
2030 	unhold_lkb(lkb);
2031 }
2032 
2033 static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2034 {
2035 	set_lvb_unlock(r, lkb);
2036 	_remove_lock(r, lkb);
2037 }
2038 
2039 static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
2040 {
2041 	_remove_lock(r, lkb);
2042 }
2043 
2044 /* returns: 0 did nothing
2045 	    1 moved lock to granted
2046 	   -1 removed lock */
2047 
2048 static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2049 {
2050 	int rv = 0;
2051 
2052 	lkb->lkb_rqmode = DLM_LOCK_IV;
2053 
2054 	switch (lkb->lkb_status) {
2055 	case DLM_LKSTS_GRANTED:
2056 		break;
2057 	case DLM_LKSTS_CONVERT:
2058 		move_lkb(r, lkb, DLM_LKSTS_GRANTED);
2059 		rv = 1;
2060 		break;
2061 	case DLM_LKSTS_WAITING:
2062 		del_lkb(r, lkb);
2063 		lkb->lkb_grmode = DLM_LOCK_IV;
2064 		/* this unhold undoes the original ref from create_lkb()
2065 		   so this leads to the lkb being freed */
2066 		unhold_lkb(lkb);
2067 		rv = -1;
2068 		break;
2069 	default:
2070 		log_print("invalid status for revert %d", lkb->lkb_status);
2071 	}
2072 	return rv;
2073 }
2074 
2075 static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
2076 {
2077 	return revert_lock(r, lkb);
2078 }
2079 
2080 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2081 {
2082 	if (lkb->lkb_grmode != lkb->lkb_rqmode) {
2083 		lkb->lkb_grmode = lkb->lkb_rqmode;
2084 		if (lkb->lkb_status)
2085 			move_lkb(r, lkb, DLM_LKSTS_GRANTED);
2086 		else
2087 			add_lkb(r, lkb, DLM_LKSTS_GRANTED);
2088 	}
2089 
2090 	lkb->lkb_rqmode = DLM_LOCK_IV;
2091 	lkb->lkb_highbast = 0;
2092 }
2093 
2094 static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2095 {
2096 	set_lvb_lock(r, lkb);
2097 	_grant_lock(r, lkb);
2098 }
2099 
2100 static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
2101 			  struct dlm_message *ms)
2102 {
2103 	set_lvb_lock_pc(r, lkb, ms);
2104 	_grant_lock(r, lkb);
2105 }
2106 
2107 /* called by grant_pending_locks() which means an async grant message must
2108    be sent to the requesting node in addition to granting the lock if the
2109    lkb belongs to a remote node. */
2110 
2111 static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
2112 {
2113 	grant_lock(r, lkb);
2114 	if (is_master_copy(lkb))
2115 		send_grant(r, lkb);
2116 	else
2117 		queue_cast(r, lkb, 0);
2118 }
2119 
2120 /* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
2121    change the granted/requested modes.  We're munging things accordingly in
2122    the process copy.
2123    CONVDEADLK: our grmode may have been forced down to NL to resolve a
2124    conversion deadlock
2125    ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
2126    compatible with other granted locks */
2127 
2128 static void munge_demoted(struct dlm_lkb *lkb)
2129 {
2130 	if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
2131 		log_print("munge_demoted %x invalid modes gr %d rq %d",
2132 			  lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
2133 		return;
2134 	}
2135 
2136 	lkb->lkb_grmode = DLM_LOCK_NL;
2137 }
2138 
2139 static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
2140 {
2141 	if (ms->m_type != cpu_to_le32(DLM_MSG_REQUEST_REPLY) &&
2142 	    ms->m_type != cpu_to_le32(DLM_MSG_GRANT)) {
2143 		log_print("munge_altmode %x invalid reply type %d",
2144 			  lkb->lkb_id, le32_to_cpu(ms->m_type));
2145 		return;
2146 	}
2147 
2148 	if (lkb->lkb_exflags & DLM_LKF_ALTPR)
2149 		lkb->lkb_rqmode = DLM_LOCK_PR;
2150 	else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
2151 		lkb->lkb_rqmode = DLM_LOCK_CW;
2152 	else {
2153 		log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
2154 		dlm_print_lkb(lkb);
2155 	}
2156 }
2157 
2158 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
2159 {
2160 	struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
2161 					   lkb_statequeue);
2162 	if (lkb->lkb_id == first->lkb_id)
2163 		return 1;
2164 
2165 	return 0;
2166 }
2167 
2168 /* Check if the given lkb conflicts with another lkb on the queue. */
2169 
2170 static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb)
2171 {
2172 	struct dlm_lkb *this;
2173 
2174 	list_for_each_entry(this, head, lkb_statequeue) {
2175 		if (this == lkb)
2176 			continue;
2177 		if (!modes_compat(this, lkb))
2178 			return 1;
2179 	}
2180 	return 0;
2181 }
2182 
2183 /*
2184  * "A conversion deadlock arises with a pair of lock requests in the converting
2185  * queue for one resource.  The granted mode of each lock blocks the requested
2186  * mode of the other lock."
2187  *
2188  * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
2189  * convert queue from being granted, then deadlk/demote lkb.
2190  *
2191  * Example:
2192  * Granted Queue: empty
2193  * Convert Queue: NL->EX (first lock)
2194  *                PR->EX (second lock)
2195  *
2196  * The first lock can't be granted because of the granted mode of the second
2197  * lock and the second lock can't be granted because it's not first in the
2198  * list.  We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
2199  * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
2200  * flag set and return DEMOTED in the lksb flags.
2201  *
2202  * Originally, this function detected conv-deadlk in a more limited scope:
2203  * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
2204  * - if lkb1 was the first entry in the queue (not just earlier), and was
2205  *   blocked by the granted mode of lkb2, and there was nothing on the
2206  *   granted queue preventing lkb1 from being granted immediately, i.e.
2207  *   lkb2 was the only thing preventing lkb1 from being granted.
2208  *
2209  * That second condition meant we'd only say there was conv-deadlk if
2210  * resolving it (by demotion) would lead to the first lock on the convert
2211  * queue being granted right away.  It allowed conversion deadlocks to exist
2212  * between locks on the convert queue while they couldn't be granted anyway.
2213  *
2214  * Now, we detect and take action on conversion deadlocks immediately when
2215  * they're created, even if they may not be immediately consequential.  If
2216  * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
2217  * mode that would prevent lkb1's conversion from being granted, we do a
2218  * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
2219  * I think this means that the lkb_is_ahead condition below should always
2220  * be zero, i.e. there will never be conv-deadlk between two locks that are
2221  * both already on the convert queue.
2222  */
2223 
2224 static int conversion_deadlock_detect(struct dlm_rsb *r, struct dlm_lkb *lkb2)
2225 {
2226 	struct dlm_lkb *lkb1;
2227 	int lkb_is_ahead = 0;
2228 
2229 	list_for_each_entry(lkb1, &r->res_convertqueue, lkb_statequeue) {
2230 		if (lkb1 == lkb2) {
2231 			lkb_is_ahead = 1;
2232 			continue;
2233 		}
2234 
2235 		if (!lkb_is_ahead) {
2236 			if (!modes_compat(lkb2, lkb1))
2237 				return 1;
2238 		} else {
2239 			if (!modes_compat(lkb2, lkb1) &&
2240 			    !modes_compat(lkb1, lkb2))
2241 				return 1;
2242 		}
2243 	}
2244 	return 0;
2245 }
2246 
2247 /*
2248  * Return 1 if the lock can be granted, 0 otherwise.
2249  * Also detect and resolve conversion deadlocks.
2250  *
2251  * lkb is the lock to be granted
2252  *
2253  * now is 1 if the function is being called in the context of the
2254  * immediate request, it is 0 if called later, after the lock has been
2255  * queued.
2256  *
2257  * recover is 1 if dlm_recover_grant() is trying to grant conversions
2258  * after recovery.
2259  *
2260  * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
2261  */
2262 
2263 static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
2264 			   int recover)
2265 {
2266 	int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV);
2267 
2268 	/*
2269 	 * 6-10: Version 5.4 introduced an option to address the phenomenon of
2270 	 * a new request for a NL mode lock being blocked.
2271 	 *
2272 	 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
2273 	 * request, then it would be granted.  In essence, the use of this flag
2274 	 * tells the Lock Manager to expedite theis request by not considering
2275 	 * what may be in the CONVERTING or WAITING queues...  As of this
2276 	 * writing, the EXPEDITE flag can be used only with new requests for NL
2277 	 * mode locks.  This flag is not valid for conversion requests.
2278 	 *
2279 	 * A shortcut.  Earlier checks return an error if EXPEDITE is used in a
2280 	 * conversion or used with a non-NL requested mode.  We also know an
2281 	 * EXPEDITE request is always granted immediately, so now must always
2282 	 * be 1.  The full condition to grant an expedite request: (now &&
2283 	 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
2284 	 * therefore be shortened to just checking the flag.
2285 	 */
2286 
2287 	if (lkb->lkb_exflags & DLM_LKF_EXPEDITE)
2288 		return 1;
2289 
2290 	/*
2291 	 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
2292 	 * added to the remaining conditions.
2293 	 */
2294 
2295 	if (queue_conflict(&r->res_grantqueue, lkb))
2296 		return 0;
2297 
2298 	/*
2299 	 * 6-3: By default, a conversion request is immediately granted if the
2300 	 * requested mode is compatible with the modes of all other granted
2301 	 * locks
2302 	 */
2303 
2304 	if (queue_conflict(&r->res_convertqueue, lkb))
2305 		return 0;
2306 
2307 	/*
2308 	 * The RECOVER_GRANT flag means dlm_recover_grant() is granting
2309 	 * locks for a recovered rsb, on which lkb's have been rebuilt.
2310 	 * The lkb's may have been rebuilt on the queues in a different
2311 	 * order than they were in on the previous master.  So, granting
2312 	 * queued conversions in order after recovery doesn't make sense
2313 	 * since the order hasn't been preserved anyway.  The new order
2314 	 * could also have created a new "in place" conversion deadlock.
2315 	 * (e.g. old, failed master held granted EX, with PR->EX, NL->EX.
2316 	 * After recovery, there would be no granted locks, and possibly
2317 	 * NL->EX, PR->EX, an in-place conversion deadlock.)  So, after
2318 	 * recovery, grant conversions without considering order.
2319 	 */
2320 
2321 	if (conv && recover)
2322 		return 1;
2323 
2324 	/*
2325 	 * 6-5: But the default algorithm for deciding whether to grant or
2326 	 * queue conversion requests does not by itself guarantee that such
2327 	 * requests are serviced on a "first come first serve" basis.  This, in
2328 	 * turn, can lead to a phenomenon known as "indefinate postponement".
2329 	 *
2330 	 * 6-7: This issue is dealt with by using the optional QUECVT flag with
2331 	 * the system service employed to request a lock conversion.  This flag
2332 	 * forces certain conversion requests to be queued, even if they are
2333 	 * compatible with the granted modes of other locks on the same
2334 	 * resource.  Thus, the use of this flag results in conversion requests
2335 	 * being ordered on a "first come first servce" basis.
2336 	 *
2337 	 * DCT: This condition is all about new conversions being able to occur
2338 	 * "in place" while the lock remains on the granted queue (assuming
2339 	 * nothing else conflicts.)  IOW if QUECVT isn't set, a conversion
2340 	 * doesn't _have_ to go onto the convert queue where it's processed in
2341 	 * order.  The "now" variable is necessary to distinguish converts
2342 	 * being received and processed for the first time now, because once a
2343 	 * convert is moved to the conversion queue the condition below applies
2344 	 * requiring fifo granting.
2345 	 */
2346 
2347 	if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT))
2348 		return 1;
2349 
2350 	/*
2351 	 * Even if the convert is compat with all granted locks,
2352 	 * QUECVT forces it behind other locks on the convert queue.
2353 	 */
2354 
2355 	if (now && conv && (lkb->lkb_exflags & DLM_LKF_QUECVT)) {
2356 		if (list_empty(&r->res_convertqueue))
2357 			return 1;
2358 		else
2359 			return 0;
2360 	}
2361 
2362 	/*
2363 	 * The NOORDER flag is set to avoid the standard vms rules on grant
2364 	 * order.
2365 	 */
2366 
2367 	if (lkb->lkb_exflags & DLM_LKF_NOORDER)
2368 		return 1;
2369 
2370 	/*
2371 	 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
2372 	 * granted until all other conversion requests ahead of it are granted
2373 	 * and/or canceled.
2374 	 */
2375 
2376 	if (!now && conv && first_in_list(lkb, &r->res_convertqueue))
2377 		return 1;
2378 
2379 	/*
2380 	 * 6-4: By default, a new request is immediately granted only if all
2381 	 * three of the following conditions are satisfied when the request is
2382 	 * issued:
2383 	 * - The queue of ungranted conversion requests for the resource is
2384 	 *   empty.
2385 	 * - The queue of ungranted new requests for the resource is empty.
2386 	 * - The mode of the new request is compatible with the most
2387 	 *   restrictive mode of all granted locks on the resource.
2388 	 */
2389 
2390 	if (now && !conv && list_empty(&r->res_convertqueue) &&
2391 	    list_empty(&r->res_waitqueue))
2392 		return 1;
2393 
2394 	/*
2395 	 * 6-4: Once a lock request is in the queue of ungranted new requests,
2396 	 * it cannot be granted until the queue of ungranted conversion
2397 	 * requests is empty, all ungranted new requests ahead of it are
2398 	 * granted and/or canceled, and it is compatible with the granted mode
2399 	 * of the most restrictive lock granted on the resource.
2400 	 */
2401 
2402 	if (!now && !conv && list_empty(&r->res_convertqueue) &&
2403 	    first_in_list(lkb, &r->res_waitqueue))
2404 		return 1;
2405 
2406 	return 0;
2407 }
2408 
2409 static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
2410 			  int recover, int *err)
2411 {
2412 	int rv;
2413 	int8_t alt = 0, rqmode = lkb->lkb_rqmode;
2414 	int8_t is_convert = (lkb->lkb_grmode != DLM_LOCK_IV);
2415 
2416 	if (err)
2417 		*err = 0;
2418 
2419 	rv = _can_be_granted(r, lkb, now, recover);
2420 	if (rv)
2421 		goto out;
2422 
2423 	/*
2424 	 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
2425 	 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
2426 	 * cancels one of the locks.
2427 	 */
2428 
2429 	if (is_convert && can_be_queued(lkb) &&
2430 	    conversion_deadlock_detect(r, lkb)) {
2431 		if (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) {
2432 			lkb->lkb_grmode = DLM_LOCK_NL;
2433 			lkb->lkb_sbflags |= DLM_SBF_DEMOTED;
2434 		} else if (err) {
2435 			*err = -EDEADLK;
2436 		} else {
2437 			log_print("can_be_granted deadlock %x now %d",
2438 				  lkb->lkb_id, now);
2439 			dlm_dump_rsb(r);
2440 		}
2441 		goto out;
2442 	}
2443 
2444 	/*
2445 	 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
2446 	 * to grant a request in a mode other than the normal rqmode.  It's a
2447 	 * simple way to provide a big optimization to applications that can
2448 	 * use them.
2449 	 */
2450 
2451 	if (rqmode != DLM_LOCK_PR && (lkb->lkb_exflags & DLM_LKF_ALTPR))
2452 		alt = DLM_LOCK_PR;
2453 	else if (rqmode != DLM_LOCK_CW && (lkb->lkb_exflags & DLM_LKF_ALTCW))
2454 		alt = DLM_LOCK_CW;
2455 
2456 	if (alt) {
2457 		lkb->lkb_rqmode = alt;
2458 		rv = _can_be_granted(r, lkb, now, 0);
2459 		if (rv)
2460 			lkb->lkb_sbflags |= DLM_SBF_ALTMODE;
2461 		else
2462 			lkb->lkb_rqmode = rqmode;
2463 	}
2464  out:
2465 	return rv;
2466 }
2467 
2468 /* Returns the highest requested mode of all blocked conversions; sets
2469    cw if there's a blocked conversion to DLM_LOCK_CW. */
2470 
2471 static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw,
2472 				 unsigned int *count)
2473 {
2474 	struct dlm_lkb *lkb, *s;
2475 	int recover = rsb_flag(r, RSB_RECOVER_GRANT);
2476 	int hi, demoted, quit, grant_restart, demote_restart;
2477 	int deadlk;
2478 
2479 	quit = 0;
2480  restart:
2481 	grant_restart = 0;
2482 	demote_restart = 0;
2483 	hi = DLM_LOCK_IV;
2484 
2485 	list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) {
2486 		demoted = is_demoted(lkb);
2487 		deadlk = 0;
2488 
2489 		if (can_be_granted(r, lkb, 0, recover, &deadlk)) {
2490 			grant_lock_pending(r, lkb);
2491 			grant_restart = 1;
2492 			if (count)
2493 				(*count)++;
2494 			continue;
2495 		}
2496 
2497 		if (!demoted && is_demoted(lkb)) {
2498 			log_print("WARN: pending demoted %x node %d %s",
2499 				  lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
2500 			demote_restart = 1;
2501 			continue;
2502 		}
2503 
2504 		if (deadlk) {
2505 			/*
2506 			 * If DLM_LKB_NODLKWT flag is set and conversion
2507 			 * deadlock is detected, we request blocking AST and
2508 			 * down (or cancel) conversion.
2509 			 */
2510 			if (lkb->lkb_exflags & DLM_LKF_NODLCKWT) {
2511 				if (lkb->lkb_highbast < lkb->lkb_rqmode) {
2512 					queue_bast(r, lkb, lkb->lkb_rqmode);
2513 					lkb->lkb_highbast = lkb->lkb_rqmode;
2514 				}
2515 			} else {
2516 				log_print("WARN: pending deadlock %x node %d %s",
2517 					  lkb->lkb_id, lkb->lkb_nodeid,
2518 					  r->res_name);
2519 				dlm_dump_rsb(r);
2520 			}
2521 			continue;
2522 		}
2523 
2524 		hi = max_t(int, lkb->lkb_rqmode, hi);
2525 
2526 		if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
2527 			*cw = 1;
2528 	}
2529 
2530 	if (grant_restart)
2531 		goto restart;
2532 	if (demote_restart && !quit) {
2533 		quit = 1;
2534 		goto restart;
2535 	}
2536 
2537 	return max_t(int, high, hi);
2538 }
2539 
2540 static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw,
2541 			      unsigned int *count)
2542 {
2543 	struct dlm_lkb *lkb, *s;
2544 
2545 	list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
2546 		if (can_be_granted(r, lkb, 0, 0, NULL)) {
2547 			grant_lock_pending(r, lkb);
2548 			if (count)
2549 				(*count)++;
2550 		} else {
2551 			high = max_t(int, lkb->lkb_rqmode, high);
2552 			if (lkb->lkb_rqmode == DLM_LOCK_CW)
2553 				*cw = 1;
2554 		}
2555 	}
2556 
2557 	return high;
2558 }
2559 
2560 /* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
2561    on either the convert or waiting queue.
2562    high is the largest rqmode of all locks blocked on the convert or
2563    waiting queue. */
2564 
2565 static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
2566 {
2567 	if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
2568 		if (gr->lkb_highbast < DLM_LOCK_EX)
2569 			return 1;
2570 		return 0;
2571 	}
2572 
2573 	if (gr->lkb_highbast < high &&
2574 	    !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
2575 		return 1;
2576 	return 0;
2577 }
2578 
2579 static void grant_pending_locks(struct dlm_rsb *r, unsigned int *count)
2580 {
2581 	struct dlm_lkb *lkb, *s;
2582 	int high = DLM_LOCK_IV;
2583 	int cw = 0;
2584 
2585 	if (!is_master(r)) {
2586 		log_print("grant_pending_locks r nodeid %d", r->res_nodeid);
2587 		dlm_dump_rsb(r);
2588 		return;
2589 	}
2590 
2591 	high = grant_pending_convert(r, high, &cw, count);
2592 	high = grant_pending_wait(r, high, &cw, count);
2593 
2594 	if (high == DLM_LOCK_IV)
2595 		return;
2596 
2597 	/*
2598 	 * If there are locks left on the wait/convert queue then send blocking
2599 	 * ASTs to granted locks based on the largest requested mode (high)
2600 	 * found above.
2601 	 */
2602 
2603 	list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
2604 		if (lkb->lkb_bastfn && lock_requires_bast(lkb, high, cw)) {
2605 			if (cw && high == DLM_LOCK_PR &&
2606 			    lkb->lkb_grmode == DLM_LOCK_PR)
2607 				queue_bast(r, lkb, DLM_LOCK_CW);
2608 			else
2609 				queue_bast(r, lkb, high);
2610 			lkb->lkb_highbast = high;
2611 		}
2612 	}
2613 }
2614 
2615 static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
2616 {
2617 	if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
2618 	    (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
2619 		if (gr->lkb_highbast < DLM_LOCK_EX)
2620 			return 1;
2621 		return 0;
2622 	}
2623 
2624 	if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
2625 		return 1;
2626 	return 0;
2627 }
2628 
2629 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
2630 			    struct dlm_lkb *lkb)
2631 {
2632 	struct dlm_lkb *gr;
2633 
2634 	list_for_each_entry(gr, head, lkb_statequeue) {
2635 		/* skip self when sending basts to convertqueue */
2636 		if (gr == lkb)
2637 			continue;
2638 		if (gr->lkb_bastfn && modes_require_bast(gr, lkb)) {
2639 			queue_bast(r, gr, lkb->lkb_rqmode);
2640 			gr->lkb_highbast = lkb->lkb_rqmode;
2641 		}
2642 	}
2643 }
2644 
2645 static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb)
2646 {
2647 	send_bast_queue(r, &r->res_grantqueue, lkb);
2648 }
2649 
2650 static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb)
2651 {
2652 	send_bast_queue(r, &r->res_grantqueue, lkb);
2653 	send_bast_queue(r, &r->res_convertqueue, lkb);
2654 }
2655 
2656 /* set_master(r, lkb) -- set the master nodeid of a resource
2657 
2658    The purpose of this function is to set the nodeid field in the given
2659    lkb using the nodeid field in the given rsb.  If the rsb's nodeid is
2660    known, it can just be copied to the lkb and the function will return
2661    0.  If the rsb's nodeid is _not_ known, it needs to be looked up
2662    before it can be copied to the lkb.
2663 
2664    When the rsb nodeid is being looked up remotely, the initial lkb
2665    causing the lookup is kept on the ls_waiters list waiting for the
2666    lookup reply.  Other lkb's waiting for the same rsb lookup are kept
2667    on the rsb's res_lookup list until the master is verified.
2668 
2669    Return values:
2670    0: nodeid is set in rsb/lkb and the caller should go ahead and use it
2671    1: the rsb master is not available and the lkb has been placed on
2672       a wait queue
2673 */
2674 
2675 static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb)
2676 {
2677 	int our_nodeid = dlm_our_nodeid();
2678 
2679 	if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) {
2680 		rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
2681 		r->res_first_lkid = lkb->lkb_id;
2682 		lkb->lkb_nodeid = r->res_nodeid;
2683 		return 0;
2684 	}
2685 
2686 	if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) {
2687 		list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup);
2688 		return 1;
2689 	}
2690 
2691 	if (r->res_master_nodeid == our_nodeid) {
2692 		lkb->lkb_nodeid = 0;
2693 		return 0;
2694 	}
2695 
2696 	if (r->res_master_nodeid) {
2697 		lkb->lkb_nodeid = r->res_master_nodeid;
2698 		return 0;
2699 	}
2700 
2701 	if (dlm_dir_nodeid(r) == our_nodeid) {
2702 		/* This is a somewhat unusual case; find_rsb will usually
2703 		   have set res_master_nodeid when dir nodeid is local, but
2704 		   there are cases where we become the dir node after we've
2705 		   past find_rsb and go through _request_lock again.
2706 		   confirm_master() or process_lookup_list() needs to be
2707 		   called after this. */
2708 		log_debug(r->res_ls, "set_master %x self master %d dir %d %s",
2709 			  lkb->lkb_id, r->res_master_nodeid, r->res_dir_nodeid,
2710 			  r->res_name);
2711 		r->res_master_nodeid = our_nodeid;
2712 		r->res_nodeid = 0;
2713 		lkb->lkb_nodeid = 0;
2714 		return 0;
2715 	}
2716 
2717 	wait_pending_remove(r);
2718 
2719 	r->res_first_lkid = lkb->lkb_id;
2720 	send_lookup(r, lkb);
2721 	return 1;
2722 }
2723 
2724 static void process_lookup_list(struct dlm_rsb *r)
2725 {
2726 	struct dlm_lkb *lkb, *safe;
2727 
2728 	list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
2729 		list_del_init(&lkb->lkb_rsb_lookup);
2730 		_request_lock(r, lkb);
2731 		schedule();
2732 	}
2733 }
2734 
2735 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
2736 
2737 static void confirm_master(struct dlm_rsb *r, int error)
2738 {
2739 	struct dlm_lkb *lkb;
2740 
2741 	if (!r->res_first_lkid)
2742 		return;
2743 
2744 	switch (error) {
2745 	case 0:
2746 	case -EINPROGRESS:
2747 		r->res_first_lkid = 0;
2748 		process_lookup_list(r);
2749 		break;
2750 
2751 	case -EAGAIN:
2752 	case -EBADR:
2753 	case -ENOTBLK:
2754 		/* the remote request failed and won't be retried (it was
2755 		   a NOQUEUE, or has been canceled/unlocked); make a waiting
2756 		   lkb the first_lkid */
2757 
2758 		r->res_first_lkid = 0;
2759 
2760 		if (!list_empty(&r->res_lookup)) {
2761 			lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
2762 					 lkb_rsb_lookup);
2763 			list_del_init(&lkb->lkb_rsb_lookup);
2764 			r->res_first_lkid = lkb->lkb_id;
2765 			_request_lock(r, lkb);
2766 		}
2767 		break;
2768 
2769 	default:
2770 		log_error(r->res_ls, "confirm_master unknown error %d", error);
2771 	}
2772 }
2773 
2774 #ifdef CONFIG_DLM_DEPRECATED_API
2775 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
2776 			 int namelen, unsigned long timeout_cs,
2777 			 void (*ast) (void *astparam),
2778 			 void *astparam,
2779 			 void (*bast) (void *astparam, int mode),
2780 			 struct dlm_args *args)
2781 #else
2782 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
2783 			 int namelen, void (*ast)(void *astparam),
2784 			 void *astparam,
2785 			 void (*bast)(void *astparam, int mode),
2786 			 struct dlm_args *args)
2787 #endif
2788 {
2789 	int rv = -EINVAL;
2790 
2791 	/* check for invalid arg usage */
2792 
2793 	if (mode < 0 || mode > DLM_LOCK_EX)
2794 		goto out;
2795 
2796 	if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
2797 		goto out;
2798 
2799 	if (flags & DLM_LKF_CANCEL)
2800 		goto out;
2801 
2802 	if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
2803 		goto out;
2804 
2805 	if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
2806 		goto out;
2807 
2808 	if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
2809 		goto out;
2810 
2811 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
2812 		goto out;
2813 
2814 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
2815 		goto out;
2816 
2817 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2818 		goto out;
2819 
2820 	if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2821 		goto out;
2822 
2823 	if (!ast || !lksb)
2824 		goto out;
2825 
2826 	if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2827 		goto out;
2828 
2829 	if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2830 		goto out;
2831 
2832 	/* these args will be copied to the lkb in validate_lock_args,
2833 	   it cannot be done now because when converting locks, fields in
2834 	   an active lkb cannot be modified before locking the rsb */
2835 
2836 	args->flags = flags;
2837 	args->astfn = ast;
2838 	args->astparam = astparam;
2839 	args->bastfn = bast;
2840 #ifdef CONFIG_DLM_DEPRECATED_API
2841 	args->timeout = timeout_cs;
2842 #endif
2843 	args->mode = mode;
2844 	args->lksb = lksb;
2845 	rv = 0;
2846  out:
2847 	return rv;
2848 }
2849 
2850 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2851 {
2852 	if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2853  		      DLM_LKF_FORCEUNLOCK))
2854 		return -EINVAL;
2855 
2856 	if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2857 		return -EINVAL;
2858 
2859 	args->flags = flags;
2860 	args->astparam = astarg;
2861 	return 0;
2862 }
2863 
2864 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2865 			      struct dlm_args *args)
2866 {
2867 	int rv = -EINVAL;
2868 
2869 	if (args->flags & DLM_LKF_CONVERT) {
2870 		if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2871 			goto out;
2872 
2873 		if (args->flags & DLM_LKF_QUECVT &&
2874 		    !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2875 			goto out;
2876 
2877 		rv = -EBUSY;
2878 		if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2879 			goto out;
2880 
2881 		/* lock not allowed if there's any op in progress */
2882 		if (lkb->lkb_wait_type || lkb->lkb_wait_count)
2883 			goto out;
2884 
2885 		if (is_overlap(lkb))
2886 			goto out;
2887 	}
2888 
2889 	lkb->lkb_exflags = args->flags;
2890 	lkb->lkb_sbflags = 0;
2891 	lkb->lkb_astfn = args->astfn;
2892 	lkb->lkb_astparam = args->astparam;
2893 	lkb->lkb_bastfn = args->bastfn;
2894 	lkb->lkb_rqmode = args->mode;
2895 	lkb->lkb_lksb = args->lksb;
2896 	lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2897 	lkb->lkb_ownpid = (int) current->pid;
2898 #ifdef CONFIG_DLM_DEPRECATED_API
2899 	lkb->lkb_timeout_cs = args->timeout;
2900 #endif
2901 	rv = 0;
2902  out:
2903 	if (rv)
2904 		log_debug(ls, "validate_lock_args %d %x %x %x %d %d %s",
2905 			  rv, lkb->lkb_id, lkb->lkb_flags, args->flags,
2906 			  lkb->lkb_status, lkb->lkb_wait_type,
2907 			  lkb->lkb_resource->res_name);
2908 	return rv;
2909 }
2910 
2911 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2912    for success */
2913 
2914 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2915    because there may be a lookup in progress and it's valid to do
2916    cancel/unlockf on it */
2917 
2918 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2919 {
2920 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
2921 	int rv = -EINVAL;
2922 
2923 	if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2924 		log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2925 		dlm_print_lkb(lkb);
2926 		goto out;
2927 	}
2928 
2929 	/* an lkb may still exist even though the lock is EOL'ed due to a
2930 	   cancel, unlock or failed noqueue request; an app can't use these
2931 	   locks; return same error as if the lkid had not been found at all */
2932 
2933 	if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2934 		log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2935 		rv = -ENOENT;
2936 		goto out;
2937 	}
2938 
2939 	/* an lkb may be waiting for an rsb lookup to complete where the
2940 	   lookup was initiated by another lock */
2941 
2942 	if (!list_empty(&lkb->lkb_rsb_lookup)) {
2943 		if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
2944 			log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2945 			list_del_init(&lkb->lkb_rsb_lookup);
2946 			queue_cast(lkb->lkb_resource, lkb,
2947 				   args->flags & DLM_LKF_CANCEL ?
2948 				   -DLM_ECANCEL : -DLM_EUNLOCK);
2949 			unhold_lkb(lkb); /* undoes create_lkb() */
2950 		}
2951 		/* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2952 		rv = -EBUSY;
2953 		goto out;
2954 	}
2955 
2956 	/* cancel not allowed with another cancel/unlock in progress */
2957 
2958 	if (args->flags & DLM_LKF_CANCEL) {
2959 		if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2960 			goto out;
2961 
2962 		if (is_overlap(lkb))
2963 			goto out;
2964 
2965 		/* don't let scand try to do a cancel */
2966 		del_timeout(lkb);
2967 
2968 		if (lkb->lkb_flags & DLM_IFL_RESEND) {
2969 			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2970 			rv = -EBUSY;
2971 			goto out;
2972 		}
2973 
2974 		/* there's nothing to cancel */
2975 		if (lkb->lkb_status == DLM_LKSTS_GRANTED &&
2976 		    !lkb->lkb_wait_type) {
2977 			rv = -EBUSY;
2978 			goto out;
2979 		}
2980 
2981 		switch (lkb->lkb_wait_type) {
2982 		case DLM_MSG_LOOKUP:
2983 		case DLM_MSG_REQUEST:
2984 			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2985 			rv = -EBUSY;
2986 			goto out;
2987 		case DLM_MSG_UNLOCK:
2988 		case DLM_MSG_CANCEL:
2989 			goto out;
2990 		}
2991 		/* add_to_waiters() will set OVERLAP_CANCEL */
2992 		goto out_ok;
2993 	}
2994 
2995 	/* do we need to allow a force-unlock if there's a normal unlock
2996 	   already in progress?  in what conditions could the normal unlock
2997 	   fail such that we'd want to send a force-unlock to be sure? */
2998 
2999 	if (args->flags & DLM_LKF_FORCEUNLOCK) {
3000 		if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
3001 			goto out;
3002 
3003 		if (is_overlap_unlock(lkb))
3004 			goto out;
3005 
3006 		/* don't let scand try to do a cancel */
3007 		del_timeout(lkb);
3008 
3009 		if (lkb->lkb_flags & DLM_IFL_RESEND) {
3010 			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
3011 			rv = -EBUSY;
3012 			goto out;
3013 		}
3014 
3015 		switch (lkb->lkb_wait_type) {
3016 		case DLM_MSG_LOOKUP:
3017 		case DLM_MSG_REQUEST:
3018 			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
3019 			rv = -EBUSY;
3020 			goto out;
3021 		case DLM_MSG_UNLOCK:
3022 			goto out;
3023 		}
3024 		/* add_to_waiters() will set OVERLAP_UNLOCK */
3025 		goto out_ok;
3026 	}
3027 
3028 	/* normal unlock not allowed if there's any op in progress */
3029 	rv = -EBUSY;
3030 	if (lkb->lkb_wait_type || lkb->lkb_wait_count)
3031 		goto out;
3032 
3033  out_ok:
3034 	/* an overlapping op shouldn't blow away exflags from other op */
3035 	lkb->lkb_exflags |= args->flags;
3036 	lkb->lkb_sbflags = 0;
3037 	lkb->lkb_astparam = args->astparam;
3038 	rv = 0;
3039  out:
3040 	if (rv)
3041 		log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
3042 			  lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
3043 			  args->flags, lkb->lkb_wait_type,
3044 			  lkb->lkb_resource->res_name);
3045 	return rv;
3046 }
3047 
3048 /*
3049  * Four stage 4 varieties:
3050  * do_request(), do_convert(), do_unlock(), do_cancel()
3051  * These are called on the master node for the given lock and
3052  * from the central locking logic.
3053  */
3054 
3055 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
3056 {
3057 	int error = 0;
3058 
3059 	if (can_be_granted(r, lkb, 1, 0, NULL)) {
3060 		grant_lock(r, lkb);
3061 		queue_cast(r, lkb, 0);
3062 		goto out;
3063 	}
3064 
3065 	if (can_be_queued(lkb)) {
3066 		error = -EINPROGRESS;
3067 		add_lkb(r, lkb, DLM_LKSTS_WAITING);
3068 		add_timeout(lkb);
3069 		goto out;
3070 	}
3071 
3072 	error = -EAGAIN;
3073 	queue_cast(r, lkb, -EAGAIN);
3074  out:
3075 	return error;
3076 }
3077 
3078 static void do_request_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
3079 			       int error)
3080 {
3081 	switch (error) {
3082 	case -EAGAIN:
3083 		if (force_blocking_asts(lkb))
3084 			send_blocking_asts_all(r, lkb);
3085 		break;
3086 	case -EINPROGRESS:
3087 		send_blocking_asts(r, lkb);
3088 		break;
3089 	}
3090 }
3091 
3092 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
3093 {
3094 	int error = 0;
3095 	int deadlk = 0;
3096 
3097 	/* changing an existing lock may allow others to be granted */
3098 
3099 	if (can_be_granted(r, lkb, 1, 0, &deadlk)) {
3100 		grant_lock(r, lkb);
3101 		queue_cast(r, lkb, 0);
3102 		goto out;
3103 	}
3104 
3105 	/* can_be_granted() detected that this lock would block in a conversion
3106 	   deadlock, so we leave it on the granted queue and return EDEADLK in
3107 	   the ast for the convert. */
3108 
3109 	if (deadlk && !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
3110 		/* it's left on the granted queue */
3111 		revert_lock(r, lkb);
3112 		queue_cast(r, lkb, -EDEADLK);
3113 		error = -EDEADLK;
3114 		goto out;
3115 	}
3116 
3117 	/* is_demoted() means the can_be_granted() above set the grmode
3118 	   to NL, and left us on the granted queue.  This auto-demotion
3119 	   (due to CONVDEADLK) might mean other locks, and/or this lock, are
3120 	   now grantable.  We have to try to grant other converting locks
3121 	   before we try again to grant this one. */
3122 
3123 	if (is_demoted(lkb)) {
3124 		grant_pending_convert(r, DLM_LOCK_IV, NULL, NULL);
3125 		if (_can_be_granted(r, lkb, 1, 0)) {
3126 			grant_lock(r, lkb);
3127 			queue_cast(r, lkb, 0);
3128 			goto out;
3129 		}
3130 		/* else fall through and move to convert queue */
3131 	}
3132 
3133 	if (can_be_queued(lkb)) {
3134 		error = -EINPROGRESS;
3135 		del_lkb(r, lkb);
3136 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3137 		add_timeout(lkb);
3138 		goto out;
3139 	}
3140 
3141 	error = -EAGAIN;
3142 	queue_cast(r, lkb, -EAGAIN);
3143  out:
3144 	return error;
3145 }
3146 
3147 static void do_convert_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
3148 			       int error)
3149 {
3150 	switch (error) {
3151 	case 0:
3152 		grant_pending_locks(r, NULL);
3153 		/* grant_pending_locks also sends basts */
3154 		break;
3155 	case -EAGAIN:
3156 		if (force_blocking_asts(lkb))
3157 			send_blocking_asts_all(r, lkb);
3158 		break;
3159 	case -EINPROGRESS:
3160 		send_blocking_asts(r, lkb);
3161 		break;
3162 	}
3163 }
3164 
3165 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3166 {
3167 	remove_lock(r, lkb);
3168 	queue_cast(r, lkb, -DLM_EUNLOCK);
3169 	return -DLM_EUNLOCK;
3170 }
3171 
3172 static void do_unlock_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
3173 			      int error)
3174 {
3175 	grant_pending_locks(r, NULL);
3176 }
3177 
3178 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
3179 
3180 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
3181 {
3182 	int error;
3183 
3184 	error = revert_lock(r, lkb);
3185 	if (error) {
3186 		queue_cast(r, lkb, -DLM_ECANCEL);
3187 		return -DLM_ECANCEL;
3188 	}
3189 	return 0;
3190 }
3191 
3192 static void do_cancel_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
3193 			      int error)
3194 {
3195 	if (error)
3196 		grant_pending_locks(r, NULL);
3197 }
3198 
3199 /*
3200  * Four stage 3 varieties:
3201  * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
3202  */
3203 
3204 /* add a new lkb to a possibly new rsb, called by requesting process */
3205 
3206 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3207 {
3208 	int error;
3209 
3210 	/* set_master: sets lkb nodeid from r */
3211 
3212 	error = set_master(r, lkb);
3213 	if (error < 0)
3214 		goto out;
3215 	if (error) {
3216 		error = 0;
3217 		goto out;
3218 	}
3219 
3220 	if (is_remote(r)) {
3221 		/* receive_request() calls do_request() on remote node */
3222 		error = send_request(r, lkb);
3223 	} else {
3224 		error = do_request(r, lkb);
3225 		/* for remote locks the request_reply is sent
3226 		   between do_request and do_request_effects */
3227 		do_request_effects(r, lkb, error);
3228 	}
3229  out:
3230 	return error;
3231 }
3232 
3233 /* change some property of an existing lkb, e.g. mode */
3234 
3235 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3236 {
3237 	int error;
3238 
3239 	if (is_remote(r)) {
3240 		/* receive_convert() calls do_convert() on remote node */
3241 		error = send_convert(r, lkb);
3242 	} else {
3243 		error = do_convert(r, lkb);
3244 		/* for remote locks the convert_reply is sent
3245 		   between do_convert and do_convert_effects */
3246 		do_convert_effects(r, lkb, error);
3247 	}
3248 
3249 	return error;
3250 }
3251 
3252 /* remove an existing lkb from the granted queue */
3253 
3254 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3255 {
3256 	int error;
3257 
3258 	if (is_remote(r)) {
3259 		/* receive_unlock() calls do_unlock() on remote node */
3260 		error = send_unlock(r, lkb);
3261 	} else {
3262 		error = do_unlock(r, lkb);
3263 		/* for remote locks the unlock_reply is sent
3264 		   between do_unlock and do_unlock_effects */
3265 		do_unlock_effects(r, lkb, error);
3266 	}
3267 
3268 	return error;
3269 }
3270 
3271 /* remove an existing lkb from the convert or wait queue */
3272 
3273 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3274 {
3275 	int error;
3276 
3277 	if (is_remote(r)) {
3278 		/* receive_cancel() calls do_cancel() on remote node */
3279 		error = send_cancel(r, lkb);
3280 	} else {
3281 		error = do_cancel(r, lkb);
3282 		/* for remote locks the cancel_reply is sent
3283 		   between do_cancel and do_cancel_effects */
3284 		do_cancel_effects(r, lkb, error);
3285 	}
3286 
3287 	return error;
3288 }
3289 
3290 /*
3291  * Four stage 2 varieties:
3292  * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
3293  */
3294 
3295 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
3296 			int len, struct dlm_args *args)
3297 {
3298 	struct dlm_rsb *r;
3299 	int error;
3300 
3301 	error = validate_lock_args(ls, lkb, args);
3302 	if (error)
3303 		return error;
3304 
3305 	error = find_rsb(ls, name, len, 0, R_REQUEST, &r);
3306 	if (error)
3307 		return error;
3308 
3309 	lock_rsb(r);
3310 
3311 	attach_lkb(r, lkb);
3312 	lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
3313 
3314 	error = _request_lock(r, lkb);
3315 
3316 	unlock_rsb(r);
3317 	put_rsb(r);
3318 	return error;
3319 }
3320 
3321 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3322 			struct dlm_args *args)
3323 {
3324 	struct dlm_rsb *r;
3325 	int error;
3326 
3327 	r = lkb->lkb_resource;
3328 
3329 	hold_rsb(r);
3330 	lock_rsb(r);
3331 
3332 	error = validate_lock_args(ls, lkb, args);
3333 	if (error)
3334 		goto out;
3335 
3336 	error = _convert_lock(r, lkb);
3337  out:
3338 	unlock_rsb(r);
3339 	put_rsb(r);
3340 	return error;
3341 }
3342 
3343 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3344 		       struct dlm_args *args)
3345 {
3346 	struct dlm_rsb *r;
3347 	int error;
3348 
3349 	r = lkb->lkb_resource;
3350 
3351 	hold_rsb(r);
3352 	lock_rsb(r);
3353 
3354 	error = validate_unlock_args(lkb, args);
3355 	if (error)
3356 		goto out;
3357 
3358 	error = _unlock_lock(r, lkb);
3359  out:
3360 	unlock_rsb(r);
3361 	put_rsb(r);
3362 	return error;
3363 }
3364 
3365 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3366 		       struct dlm_args *args)
3367 {
3368 	struct dlm_rsb *r;
3369 	int error;
3370 
3371 	r = lkb->lkb_resource;
3372 
3373 	hold_rsb(r);
3374 	lock_rsb(r);
3375 
3376 	error = validate_unlock_args(lkb, args);
3377 	if (error)
3378 		goto out;
3379 
3380 	error = _cancel_lock(r, lkb);
3381  out:
3382 	unlock_rsb(r);
3383 	put_rsb(r);
3384 	return error;
3385 }
3386 
3387 /*
3388  * Two stage 1 varieties:  dlm_lock() and dlm_unlock()
3389  */
3390 
3391 int dlm_lock(dlm_lockspace_t *lockspace,
3392 	     int mode,
3393 	     struct dlm_lksb *lksb,
3394 	     uint32_t flags,
3395 	     void *name,
3396 	     unsigned int namelen,
3397 	     uint32_t parent_lkid,
3398 	     void (*ast) (void *astarg),
3399 	     void *astarg,
3400 	     void (*bast) (void *astarg, int mode))
3401 {
3402 	struct dlm_ls *ls;
3403 	struct dlm_lkb *lkb;
3404 	struct dlm_args args;
3405 	int error, convert = flags & DLM_LKF_CONVERT;
3406 
3407 	ls = dlm_find_lockspace_local(lockspace);
3408 	if (!ls)
3409 		return -EINVAL;
3410 
3411 	dlm_lock_recovery(ls);
3412 
3413 	if (convert)
3414 		error = find_lkb(ls, lksb->sb_lkid, &lkb);
3415 	else
3416 		error = create_lkb(ls, &lkb);
3417 
3418 	if (error)
3419 		goto out;
3420 
3421 	trace_dlm_lock_start(ls, lkb, name, namelen, mode, flags);
3422 
3423 #ifdef CONFIG_DLM_DEPRECATED_API
3424 	error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
3425 			      astarg, bast, &args);
3426 #else
3427 	error = set_lock_args(mode, lksb, flags, namelen, ast, astarg, bast,
3428 			      &args);
3429 #endif
3430 	if (error)
3431 		goto out_put;
3432 
3433 	if (convert)
3434 		error = convert_lock(ls, lkb, &args);
3435 	else
3436 		error = request_lock(ls, lkb, name, namelen, &args);
3437 
3438 	if (error == -EINPROGRESS)
3439 		error = 0;
3440  out_put:
3441 	trace_dlm_lock_end(ls, lkb, name, namelen, mode, flags, error);
3442 
3443 	if (convert || error)
3444 		__put_lkb(ls, lkb);
3445 	if (error == -EAGAIN || error == -EDEADLK)
3446 		error = 0;
3447  out:
3448 	dlm_unlock_recovery(ls);
3449 	dlm_put_lockspace(ls);
3450 	return error;
3451 }
3452 
3453 int dlm_unlock(dlm_lockspace_t *lockspace,
3454 	       uint32_t lkid,
3455 	       uint32_t flags,
3456 	       struct dlm_lksb *lksb,
3457 	       void *astarg)
3458 {
3459 	struct dlm_ls *ls;
3460 	struct dlm_lkb *lkb;
3461 	struct dlm_args args;
3462 	int error;
3463 
3464 	ls = dlm_find_lockspace_local(lockspace);
3465 	if (!ls)
3466 		return -EINVAL;
3467 
3468 	dlm_lock_recovery(ls);
3469 
3470 	error = find_lkb(ls, lkid, &lkb);
3471 	if (error)
3472 		goto out;
3473 
3474 	trace_dlm_unlock_start(ls, lkb, flags);
3475 
3476 	error = set_unlock_args(flags, astarg, &args);
3477 	if (error)
3478 		goto out_put;
3479 
3480 	if (flags & DLM_LKF_CANCEL)
3481 		error = cancel_lock(ls, lkb, &args);
3482 	else
3483 		error = unlock_lock(ls, lkb, &args);
3484 
3485 	if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
3486 		error = 0;
3487 	if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
3488 		error = 0;
3489  out_put:
3490 	trace_dlm_unlock_end(ls, lkb, flags, error);
3491 
3492 	dlm_put_lkb(lkb);
3493  out:
3494 	dlm_unlock_recovery(ls);
3495 	dlm_put_lockspace(ls);
3496 	return error;
3497 }
3498 
3499 /*
3500  * send/receive routines for remote operations and replies
3501  *
3502  * send_args
3503  * send_common
3504  * send_request			receive_request
3505  * send_convert			receive_convert
3506  * send_unlock			receive_unlock
3507  * send_cancel			receive_cancel
3508  * send_grant			receive_grant
3509  * send_bast			receive_bast
3510  * send_lookup			receive_lookup
3511  * send_remove			receive_remove
3512  *
3513  * 				send_common_reply
3514  * receive_request_reply	send_request_reply
3515  * receive_convert_reply	send_convert_reply
3516  * receive_unlock_reply		send_unlock_reply
3517  * receive_cancel_reply		send_cancel_reply
3518  * receive_lookup_reply		send_lookup_reply
3519  */
3520 
3521 static int _create_message(struct dlm_ls *ls, int mb_len,
3522 			   int to_nodeid, int mstype,
3523 			   struct dlm_message **ms_ret,
3524 			   struct dlm_mhandle **mh_ret)
3525 {
3526 	struct dlm_message *ms;
3527 	struct dlm_mhandle *mh;
3528 	char *mb;
3529 
3530 	/* get_buffer gives us a message handle (mh) that we need to
3531 	   pass into midcomms_commit and a message buffer (mb) that we
3532 	   write our data into */
3533 
3534 	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, GFP_NOFS, &mb);
3535 	if (!mh)
3536 		return -ENOBUFS;
3537 
3538 	ms = (struct dlm_message *) mb;
3539 
3540 	ms->m_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
3541 	ms->m_header.u.h_lockspace = cpu_to_le32(ls->ls_global_id);
3542 	ms->m_header.h_nodeid = cpu_to_le32(dlm_our_nodeid());
3543 	ms->m_header.h_length = cpu_to_le16(mb_len);
3544 	ms->m_header.h_cmd = DLM_MSG;
3545 
3546 	ms->m_type = cpu_to_le32(mstype);
3547 
3548 	*mh_ret = mh;
3549 	*ms_ret = ms;
3550 	return 0;
3551 }
3552 
3553 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
3554 			  int to_nodeid, int mstype,
3555 			  struct dlm_message **ms_ret,
3556 			  struct dlm_mhandle **mh_ret)
3557 {
3558 	int mb_len = sizeof(struct dlm_message);
3559 
3560 	switch (mstype) {
3561 	case DLM_MSG_REQUEST:
3562 	case DLM_MSG_LOOKUP:
3563 	case DLM_MSG_REMOVE:
3564 		mb_len += r->res_length;
3565 		break;
3566 	case DLM_MSG_CONVERT:
3567 	case DLM_MSG_UNLOCK:
3568 	case DLM_MSG_REQUEST_REPLY:
3569 	case DLM_MSG_CONVERT_REPLY:
3570 	case DLM_MSG_GRANT:
3571 		if (lkb && lkb->lkb_lvbptr)
3572 			mb_len += r->res_ls->ls_lvblen;
3573 		break;
3574 	}
3575 
3576 	return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
3577 			       ms_ret, mh_ret);
3578 }
3579 
3580 /* further lowcomms enhancements or alternate implementations may make
3581    the return value from this function useful at some point */
3582 
3583 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
3584 {
3585 	dlm_midcomms_commit_mhandle(mh);
3586 	return 0;
3587 }
3588 
3589 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
3590 		      struct dlm_message *ms)
3591 {
3592 	ms->m_nodeid   = cpu_to_le32(lkb->lkb_nodeid);
3593 	ms->m_pid      = cpu_to_le32(lkb->lkb_ownpid);
3594 	ms->m_lkid     = cpu_to_le32(lkb->lkb_id);
3595 	ms->m_remid    = cpu_to_le32(lkb->lkb_remid);
3596 	ms->m_exflags  = cpu_to_le32(lkb->lkb_exflags);
3597 	ms->m_sbflags  = cpu_to_le32(lkb->lkb_sbflags);
3598 	ms->m_flags    = cpu_to_le32(lkb->lkb_flags);
3599 	ms->m_lvbseq   = cpu_to_le32(lkb->lkb_lvbseq);
3600 	ms->m_status   = cpu_to_le32(lkb->lkb_status);
3601 	ms->m_grmode   = cpu_to_le32(lkb->lkb_grmode);
3602 	ms->m_rqmode   = cpu_to_le32(lkb->lkb_rqmode);
3603 	ms->m_hash     = cpu_to_le32(r->res_hash);
3604 
3605 	/* m_result and m_bastmode are set from function args,
3606 	   not from lkb fields */
3607 
3608 	if (lkb->lkb_bastfn)
3609 		ms->m_asts |= cpu_to_le32(DLM_CB_BAST);
3610 	if (lkb->lkb_astfn)
3611 		ms->m_asts |= cpu_to_le32(DLM_CB_CAST);
3612 
3613 	/* compare with switch in create_message; send_remove() doesn't
3614 	   use send_args() */
3615 
3616 	switch (ms->m_type) {
3617 	case cpu_to_le32(DLM_MSG_REQUEST):
3618 	case cpu_to_le32(DLM_MSG_LOOKUP):
3619 		memcpy(ms->m_extra, r->res_name, r->res_length);
3620 		break;
3621 	case cpu_to_le32(DLM_MSG_CONVERT):
3622 	case cpu_to_le32(DLM_MSG_UNLOCK):
3623 	case cpu_to_le32(DLM_MSG_REQUEST_REPLY):
3624 	case cpu_to_le32(DLM_MSG_CONVERT_REPLY):
3625 	case cpu_to_le32(DLM_MSG_GRANT):
3626 		if (!lkb->lkb_lvbptr)
3627 			break;
3628 		memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
3629 		break;
3630 	}
3631 }
3632 
3633 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
3634 {
3635 	struct dlm_message *ms;
3636 	struct dlm_mhandle *mh;
3637 	int to_nodeid, error;
3638 
3639 	to_nodeid = r->res_nodeid;
3640 
3641 	error = add_to_waiters(lkb, mstype, to_nodeid);
3642 	if (error)
3643 		return error;
3644 
3645 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
3646 	if (error)
3647 		goto fail;
3648 
3649 	send_args(r, lkb, ms);
3650 
3651 	error = send_message(mh, ms);
3652 	if (error)
3653 		goto fail;
3654 	return 0;
3655 
3656  fail:
3657 	remove_from_waiters(lkb, msg_reply_type(mstype));
3658 	return error;
3659 }
3660 
3661 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
3662 {
3663 	return send_common(r, lkb, DLM_MSG_REQUEST);
3664 }
3665 
3666 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
3667 {
3668 	int error;
3669 
3670 	error = send_common(r, lkb, DLM_MSG_CONVERT);
3671 
3672 	/* down conversions go without a reply from the master */
3673 	if (!error && down_conversion(lkb)) {
3674 		remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
3675 		r->res_ls->ls_stub_ms.m_flags = cpu_to_le32(DLM_IFL_STUB_MS);
3676 		r->res_ls->ls_stub_ms.m_type = cpu_to_le32(DLM_MSG_CONVERT_REPLY);
3677 		r->res_ls->ls_stub_ms.m_result = 0;
3678 		__receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
3679 	}
3680 
3681 	return error;
3682 }
3683 
3684 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
3685    MASTER_UNCERTAIN to force the next request on the rsb to confirm
3686    that the master is still correct. */
3687 
3688 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3689 {
3690 	return send_common(r, lkb, DLM_MSG_UNLOCK);
3691 }
3692 
3693 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
3694 {
3695 	return send_common(r, lkb, DLM_MSG_CANCEL);
3696 }
3697 
3698 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
3699 {
3700 	struct dlm_message *ms;
3701 	struct dlm_mhandle *mh;
3702 	int to_nodeid, error;
3703 
3704 	to_nodeid = lkb->lkb_nodeid;
3705 
3706 	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
3707 	if (error)
3708 		goto out;
3709 
3710 	send_args(r, lkb, ms);
3711 
3712 	ms->m_result = 0;
3713 
3714 	error = send_message(mh, ms);
3715  out:
3716 	return error;
3717 }
3718 
3719 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
3720 {
3721 	struct dlm_message *ms;
3722 	struct dlm_mhandle *mh;
3723 	int to_nodeid, error;
3724 
3725 	to_nodeid = lkb->lkb_nodeid;
3726 
3727 	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
3728 	if (error)
3729 		goto out;
3730 
3731 	send_args(r, lkb, ms);
3732 
3733 	ms->m_bastmode = cpu_to_le32(mode);
3734 
3735 	error = send_message(mh, ms);
3736  out:
3737 	return error;
3738 }
3739 
3740 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
3741 {
3742 	struct dlm_message *ms;
3743 	struct dlm_mhandle *mh;
3744 	int to_nodeid, error;
3745 
3746 	to_nodeid = dlm_dir_nodeid(r);
3747 
3748 	error = add_to_waiters(lkb, DLM_MSG_LOOKUP, to_nodeid);
3749 	if (error)
3750 		return error;
3751 
3752 	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
3753 	if (error)
3754 		goto fail;
3755 
3756 	send_args(r, lkb, ms);
3757 
3758 	error = send_message(mh, ms);
3759 	if (error)
3760 		goto fail;
3761 	return 0;
3762 
3763  fail:
3764 	remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3765 	return error;
3766 }
3767 
3768 static int send_remove(struct dlm_rsb *r)
3769 {
3770 	struct dlm_message *ms;
3771 	struct dlm_mhandle *mh;
3772 	int to_nodeid, error;
3773 
3774 	to_nodeid = dlm_dir_nodeid(r);
3775 
3776 	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
3777 	if (error)
3778 		goto out;
3779 
3780 	memcpy(ms->m_extra, r->res_name, r->res_length);
3781 	ms->m_hash = cpu_to_le32(r->res_hash);
3782 
3783 	error = send_message(mh, ms);
3784  out:
3785 	return error;
3786 }
3787 
3788 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3789 			     int mstype, int rv)
3790 {
3791 	struct dlm_message *ms;
3792 	struct dlm_mhandle *mh;
3793 	int to_nodeid, error;
3794 
3795 	to_nodeid = lkb->lkb_nodeid;
3796 
3797 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
3798 	if (error)
3799 		goto out;
3800 
3801 	send_args(r, lkb, ms);
3802 
3803 	ms->m_result = cpu_to_le32(to_dlm_errno(rv));
3804 
3805 	error = send_message(mh, ms);
3806  out:
3807 	return error;
3808 }
3809 
3810 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3811 {
3812 	return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
3813 }
3814 
3815 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3816 {
3817 	return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
3818 }
3819 
3820 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3821 {
3822 	return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
3823 }
3824 
3825 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3826 {
3827 	return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
3828 }
3829 
3830 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
3831 			     int ret_nodeid, int rv)
3832 {
3833 	struct dlm_rsb *r = &ls->ls_stub_rsb;
3834 	struct dlm_message *ms;
3835 	struct dlm_mhandle *mh;
3836 	int error, nodeid = le32_to_cpu(ms_in->m_header.h_nodeid);
3837 
3838 	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
3839 	if (error)
3840 		goto out;
3841 
3842 	ms->m_lkid = ms_in->m_lkid;
3843 	ms->m_result = cpu_to_le32(to_dlm_errno(rv));
3844 	ms->m_nodeid = cpu_to_le32(ret_nodeid);
3845 
3846 	error = send_message(mh, ms);
3847  out:
3848 	return error;
3849 }
3850 
3851 /* which args we save from a received message depends heavily on the type
3852    of message, unlike the send side where we can safely send everything about
3853    the lkb for any type of message */
3854 
3855 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
3856 {
3857 	lkb->lkb_exflags = le32_to_cpu(ms->m_exflags);
3858 	lkb->lkb_sbflags = le32_to_cpu(ms->m_sbflags);
3859 	lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
3860 			  (le32_to_cpu(ms->m_flags) & 0x0000FFFF);
3861 }
3862 
3863 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3864 {
3865 	if (ms->m_flags == cpu_to_le32(DLM_IFL_STUB_MS))
3866 		return;
3867 
3868 	lkb->lkb_sbflags = le32_to_cpu(ms->m_sbflags);
3869 	lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
3870 			 (le32_to_cpu(ms->m_flags) & 0x0000FFFF);
3871 }
3872 
3873 static int receive_extralen(struct dlm_message *ms)
3874 {
3875 	return (le16_to_cpu(ms->m_header.h_length) -
3876 		sizeof(struct dlm_message));
3877 }
3878 
3879 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
3880 		       struct dlm_message *ms)
3881 {
3882 	int len;
3883 
3884 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3885 		if (!lkb->lkb_lvbptr)
3886 			lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3887 		if (!lkb->lkb_lvbptr)
3888 			return -ENOMEM;
3889 		len = receive_extralen(ms);
3890 		if (len > ls->ls_lvblen)
3891 			len = ls->ls_lvblen;
3892 		memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
3893 	}
3894 	return 0;
3895 }
3896 
3897 static void fake_bastfn(void *astparam, int mode)
3898 {
3899 	log_print("fake_bastfn should not be called");
3900 }
3901 
3902 static void fake_astfn(void *astparam)
3903 {
3904 	log_print("fake_astfn should not be called");
3905 }
3906 
3907 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3908 				struct dlm_message *ms)
3909 {
3910 	lkb->lkb_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
3911 	lkb->lkb_ownpid = le32_to_cpu(ms->m_pid);
3912 	lkb->lkb_remid = le32_to_cpu(ms->m_lkid);
3913 	lkb->lkb_grmode = DLM_LOCK_IV;
3914 	lkb->lkb_rqmode = le32_to_cpu(ms->m_rqmode);
3915 
3916 	lkb->lkb_bastfn = (ms->m_asts & cpu_to_le32(DLM_CB_BAST)) ? &fake_bastfn : NULL;
3917 	lkb->lkb_astfn = (ms->m_asts & cpu_to_le32(DLM_CB_CAST)) ? &fake_astfn : NULL;
3918 
3919 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3920 		/* lkb was just created so there won't be an lvb yet */
3921 		lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3922 		if (!lkb->lkb_lvbptr)
3923 			return -ENOMEM;
3924 	}
3925 
3926 	return 0;
3927 }
3928 
3929 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3930 				struct dlm_message *ms)
3931 {
3932 	if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3933 		return -EBUSY;
3934 
3935 	if (receive_lvb(ls, lkb, ms))
3936 		return -ENOMEM;
3937 
3938 	lkb->lkb_rqmode = le32_to_cpu(ms->m_rqmode);
3939 	lkb->lkb_lvbseq = le32_to_cpu(ms->m_lvbseq);
3940 
3941 	return 0;
3942 }
3943 
3944 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3945 			       struct dlm_message *ms)
3946 {
3947 	if (receive_lvb(ls, lkb, ms))
3948 		return -ENOMEM;
3949 	return 0;
3950 }
3951 
3952 /* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3953    uses to send a reply and that the remote end uses to process the reply. */
3954 
3955 static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3956 {
3957 	struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3958 	lkb->lkb_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
3959 	lkb->lkb_remid = le32_to_cpu(ms->m_lkid);
3960 }
3961 
3962 /* This is called after the rsb is locked so that we can safely inspect
3963    fields in the lkb. */
3964 
3965 static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3966 {
3967 	int from = le32_to_cpu(ms->m_header.h_nodeid);
3968 	int error = 0;
3969 
3970 	/* currently mixing of user/kernel locks are not supported */
3971 	if (ms->m_flags & cpu_to_le32(DLM_IFL_USER) &&
3972 	    ~lkb->lkb_flags & DLM_IFL_USER) {
3973 		log_error(lkb->lkb_resource->res_ls,
3974 			  "got user dlm message for a kernel lock");
3975 		error = -EINVAL;
3976 		goto out;
3977 	}
3978 
3979 	switch (ms->m_type) {
3980 	case cpu_to_le32(DLM_MSG_CONVERT):
3981 	case cpu_to_le32(DLM_MSG_UNLOCK):
3982 	case cpu_to_le32(DLM_MSG_CANCEL):
3983 		if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3984 			error = -EINVAL;
3985 		break;
3986 
3987 	case cpu_to_le32(DLM_MSG_CONVERT_REPLY):
3988 	case cpu_to_le32(DLM_MSG_UNLOCK_REPLY):
3989 	case cpu_to_le32(DLM_MSG_CANCEL_REPLY):
3990 	case cpu_to_le32(DLM_MSG_GRANT):
3991 	case cpu_to_le32(DLM_MSG_BAST):
3992 		if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3993 			error = -EINVAL;
3994 		break;
3995 
3996 	case cpu_to_le32(DLM_MSG_REQUEST_REPLY):
3997 		if (!is_process_copy(lkb))
3998 			error = -EINVAL;
3999 		else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
4000 			error = -EINVAL;
4001 		break;
4002 
4003 	default:
4004 		error = -EINVAL;
4005 	}
4006 
4007 out:
4008 	if (error)
4009 		log_error(lkb->lkb_resource->res_ls,
4010 			  "ignore invalid message %d from %d %x %x %x %d",
4011 			  le32_to_cpu(ms->m_type), from, lkb->lkb_id,
4012 			  lkb->lkb_remid, lkb->lkb_flags, lkb->lkb_nodeid);
4013 	return error;
4014 }
4015 
4016 static void send_repeat_remove(struct dlm_ls *ls, char *ms_name, int len)
4017 {
4018 	char name[DLM_RESNAME_MAXLEN + 1];
4019 	struct dlm_message *ms;
4020 	struct dlm_mhandle *mh;
4021 	struct dlm_rsb *r;
4022 	uint32_t hash, b;
4023 	int rv, dir_nodeid;
4024 
4025 	memset(name, 0, sizeof(name));
4026 	memcpy(name, ms_name, len);
4027 
4028 	hash = jhash(name, len, 0);
4029 	b = hash & (ls->ls_rsbtbl_size - 1);
4030 
4031 	dir_nodeid = dlm_hash2nodeid(ls, hash);
4032 
4033 	log_error(ls, "send_repeat_remove dir %d %s", dir_nodeid, name);
4034 
4035 	spin_lock(&ls->ls_rsbtbl[b].lock);
4036 	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
4037 	if (!rv) {
4038 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4039 		log_error(ls, "repeat_remove on keep %s", name);
4040 		return;
4041 	}
4042 
4043 	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
4044 	if (!rv) {
4045 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4046 		log_error(ls, "repeat_remove on toss %s", name);
4047 		return;
4048 	}
4049 
4050 	/* use ls->remove_name2 to avoid conflict with shrink? */
4051 
4052 	spin_lock(&ls->ls_remove_spin);
4053 	ls->ls_remove_len = len;
4054 	memcpy(ls->ls_remove_name, name, DLM_RESNAME_MAXLEN);
4055 	spin_unlock(&ls->ls_remove_spin);
4056 	spin_unlock(&ls->ls_rsbtbl[b].lock);
4057 
4058 	rv = _create_message(ls, sizeof(struct dlm_message) + len,
4059 			     dir_nodeid, DLM_MSG_REMOVE, &ms, &mh);
4060 	if (rv)
4061 		goto out;
4062 
4063 	memcpy(ms->m_extra, name, len);
4064 	ms->m_hash = cpu_to_le32(hash);
4065 
4066 	send_message(mh, ms);
4067 
4068 out:
4069 	spin_lock(&ls->ls_remove_spin);
4070 	ls->ls_remove_len = 0;
4071 	memset(ls->ls_remove_name, 0, DLM_RESNAME_MAXLEN);
4072 	spin_unlock(&ls->ls_remove_spin);
4073 	wake_up(&ls->ls_remove_wait);
4074 }
4075 
4076 static int receive_request(struct dlm_ls *ls, struct dlm_message *ms)
4077 {
4078 	struct dlm_lkb *lkb;
4079 	struct dlm_rsb *r;
4080 	int from_nodeid;
4081 	int error, namelen = 0;
4082 
4083 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4084 
4085 	error = create_lkb(ls, &lkb);
4086 	if (error)
4087 		goto fail;
4088 
4089 	receive_flags(lkb, ms);
4090 	lkb->lkb_flags |= DLM_IFL_MSTCPY;
4091 	error = receive_request_args(ls, lkb, ms);
4092 	if (error) {
4093 		__put_lkb(ls, lkb);
4094 		goto fail;
4095 	}
4096 
4097 	/* The dir node is the authority on whether we are the master
4098 	   for this rsb or not, so if the master sends us a request, we should
4099 	   recreate the rsb if we've destroyed it.   This race happens when we
4100 	   send a remove message to the dir node at the same time that the dir
4101 	   node sends us a request for the rsb. */
4102 
4103 	namelen = receive_extralen(ms);
4104 
4105 	error = find_rsb(ls, ms->m_extra, namelen, from_nodeid,
4106 			 R_RECEIVE_REQUEST, &r);
4107 	if (error) {
4108 		__put_lkb(ls, lkb);
4109 		goto fail;
4110 	}
4111 
4112 	lock_rsb(r);
4113 
4114 	if (r->res_master_nodeid != dlm_our_nodeid()) {
4115 		error = validate_master_nodeid(ls, r, from_nodeid);
4116 		if (error) {
4117 			unlock_rsb(r);
4118 			put_rsb(r);
4119 			__put_lkb(ls, lkb);
4120 			goto fail;
4121 		}
4122 	}
4123 
4124 	attach_lkb(r, lkb);
4125 	error = do_request(r, lkb);
4126 	send_request_reply(r, lkb, error);
4127 	do_request_effects(r, lkb, error);
4128 
4129 	unlock_rsb(r);
4130 	put_rsb(r);
4131 
4132 	if (error == -EINPROGRESS)
4133 		error = 0;
4134 	if (error)
4135 		dlm_put_lkb(lkb);
4136 	return 0;
4137 
4138  fail:
4139 	/* TODO: instead of returning ENOTBLK, add the lkb to res_lookup
4140 	   and do this receive_request again from process_lookup_list once
4141 	   we get the lookup reply.  This would avoid a many repeated
4142 	   ENOTBLK request failures when the lookup reply designating us
4143 	   as master is delayed. */
4144 
4145 	/* We could repeatedly return -EBADR here if our send_remove() is
4146 	   delayed in being sent/arriving/being processed on the dir node.
4147 	   Another node would repeatedly lookup up the master, and the dir
4148 	   node would continue returning our nodeid until our send_remove
4149 	   took effect.
4150 
4151 	   We send another remove message in case our previous send_remove
4152 	   was lost/ignored/missed somehow. */
4153 
4154 	if (error != -ENOTBLK) {
4155 		log_limit(ls, "receive_request %x from %d %d",
4156 			  le32_to_cpu(ms->m_lkid), from_nodeid, error);
4157 	}
4158 
4159 	if (namelen && error == -EBADR) {
4160 		send_repeat_remove(ls, ms->m_extra, namelen);
4161 		msleep(1000);
4162 	}
4163 
4164 	setup_stub_lkb(ls, ms);
4165 	send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
4166 	return error;
4167 }
4168 
4169 static int receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
4170 {
4171 	struct dlm_lkb *lkb;
4172 	struct dlm_rsb *r;
4173 	int error, reply = 1;
4174 
4175 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4176 	if (error)
4177 		goto fail;
4178 
4179 	if (lkb->lkb_remid != le32_to_cpu(ms->m_lkid)) {
4180 		log_error(ls, "receive_convert %x remid %x recover_seq %llu "
4181 			  "remote %d %x", lkb->lkb_id, lkb->lkb_remid,
4182 			  (unsigned long long)lkb->lkb_recover_seq,
4183 			  le32_to_cpu(ms->m_header.h_nodeid),
4184 			  le32_to_cpu(ms->m_lkid));
4185 		error = -ENOENT;
4186 		dlm_put_lkb(lkb);
4187 		goto fail;
4188 	}
4189 
4190 	r = lkb->lkb_resource;
4191 
4192 	hold_rsb(r);
4193 	lock_rsb(r);
4194 
4195 	error = validate_message(lkb, ms);
4196 	if (error)
4197 		goto out;
4198 
4199 	receive_flags(lkb, ms);
4200 
4201 	error = receive_convert_args(ls, lkb, ms);
4202 	if (error) {
4203 		send_convert_reply(r, lkb, error);
4204 		goto out;
4205 	}
4206 
4207 	reply = !down_conversion(lkb);
4208 
4209 	error = do_convert(r, lkb);
4210 	if (reply)
4211 		send_convert_reply(r, lkb, error);
4212 	do_convert_effects(r, lkb, error);
4213  out:
4214 	unlock_rsb(r);
4215 	put_rsb(r);
4216 	dlm_put_lkb(lkb);
4217 	return 0;
4218 
4219  fail:
4220 	setup_stub_lkb(ls, ms);
4221 	send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
4222 	return error;
4223 }
4224 
4225 static int receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
4226 {
4227 	struct dlm_lkb *lkb;
4228 	struct dlm_rsb *r;
4229 	int error;
4230 
4231 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4232 	if (error)
4233 		goto fail;
4234 
4235 	if (lkb->lkb_remid != le32_to_cpu(ms->m_lkid)) {
4236 		log_error(ls, "receive_unlock %x remid %x remote %d %x",
4237 			  lkb->lkb_id, lkb->lkb_remid,
4238 			  le32_to_cpu(ms->m_header.h_nodeid),
4239 			  le32_to_cpu(ms->m_lkid));
4240 		error = -ENOENT;
4241 		dlm_put_lkb(lkb);
4242 		goto fail;
4243 	}
4244 
4245 	r = lkb->lkb_resource;
4246 
4247 	hold_rsb(r);
4248 	lock_rsb(r);
4249 
4250 	error = validate_message(lkb, ms);
4251 	if (error)
4252 		goto out;
4253 
4254 	receive_flags(lkb, ms);
4255 
4256 	error = receive_unlock_args(ls, lkb, ms);
4257 	if (error) {
4258 		send_unlock_reply(r, lkb, error);
4259 		goto out;
4260 	}
4261 
4262 	error = do_unlock(r, lkb);
4263 	send_unlock_reply(r, lkb, error);
4264 	do_unlock_effects(r, lkb, error);
4265  out:
4266 	unlock_rsb(r);
4267 	put_rsb(r);
4268 	dlm_put_lkb(lkb);
4269 	return 0;
4270 
4271  fail:
4272 	setup_stub_lkb(ls, ms);
4273 	send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
4274 	return error;
4275 }
4276 
4277 static int receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
4278 {
4279 	struct dlm_lkb *lkb;
4280 	struct dlm_rsb *r;
4281 	int error;
4282 
4283 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4284 	if (error)
4285 		goto fail;
4286 
4287 	receive_flags(lkb, ms);
4288 
4289 	r = lkb->lkb_resource;
4290 
4291 	hold_rsb(r);
4292 	lock_rsb(r);
4293 
4294 	error = validate_message(lkb, ms);
4295 	if (error)
4296 		goto out;
4297 
4298 	error = do_cancel(r, lkb);
4299 	send_cancel_reply(r, lkb, error);
4300 	do_cancel_effects(r, lkb, error);
4301  out:
4302 	unlock_rsb(r);
4303 	put_rsb(r);
4304 	dlm_put_lkb(lkb);
4305 	return 0;
4306 
4307  fail:
4308 	setup_stub_lkb(ls, ms);
4309 	send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
4310 	return error;
4311 }
4312 
4313 static int receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
4314 {
4315 	struct dlm_lkb *lkb;
4316 	struct dlm_rsb *r;
4317 	int error;
4318 
4319 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4320 	if (error)
4321 		return error;
4322 
4323 	r = lkb->lkb_resource;
4324 
4325 	hold_rsb(r);
4326 	lock_rsb(r);
4327 
4328 	error = validate_message(lkb, ms);
4329 	if (error)
4330 		goto out;
4331 
4332 	receive_flags_reply(lkb, ms);
4333 	if (is_altmode(lkb))
4334 		munge_altmode(lkb, ms);
4335 	grant_lock_pc(r, lkb, ms);
4336 	queue_cast(r, lkb, 0);
4337  out:
4338 	unlock_rsb(r);
4339 	put_rsb(r);
4340 	dlm_put_lkb(lkb);
4341 	return 0;
4342 }
4343 
4344 static int receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
4345 {
4346 	struct dlm_lkb *lkb;
4347 	struct dlm_rsb *r;
4348 	int error;
4349 
4350 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4351 	if (error)
4352 		return error;
4353 
4354 	r = lkb->lkb_resource;
4355 
4356 	hold_rsb(r);
4357 	lock_rsb(r);
4358 
4359 	error = validate_message(lkb, ms);
4360 	if (error)
4361 		goto out;
4362 
4363 	queue_bast(r, lkb, le32_to_cpu(ms->m_bastmode));
4364 	lkb->lkb_highbast = le32_to_cpu(ms->m_bastmode);
4365  out:
4366 	unlock_rsb(r);
4367 	put_rsb(r);
4368 	dlm_put_lkb(lkb);
4369 	return 0;
4370 }
4371 
4372 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
4373 {
4374 	int len, error, ret_nodeid, from_nodeid, our_nodeid;
4375 
4376 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4377 	our_nodeid = dlm_our_nodeid();
4378 
4379 	len = receive_extralen(ms);
4380 
4381 	error = dlm_master_lookup(ls, from_nodeid, ms->m_extra, len, 0,
4382 				  &ret_nodeid, NULL);
4383 
4384 	/* Optimization: we're master so treat lookup as a request */
4385 	if (!error && ret_nodeid == our_nodeid) {
4386 		receive_request(ls, ms);
4387 		return;
4388 	}
4389 	send_lookup_reply(ls, ms, ret_nodeid, error);
4390 }
4391 
4392 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
4393 {
4394 	char name[DLM_RESNAME_MAXLEN+1];
4395 	struct dlm_rsb *r;
4396 	uint32_t hash, b;
4397 	int rv, len, dir_nodeid, from_nodeid;
4398 
4399 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4400 
4401 	len = receive_extralen(ms);
4402 
4403 	if (len > DLM_RESNAME_MAXLEN) {
4404 		log_error(ls, "receive_remove from %d bad len %d",
4405 			  from_nodeid, len);
4406 		return;
4407 	}
4408 
4409 	dir_nodeid = dlm_hash2nodeid(ls, le32_to_cpu(ms->m_hash));
4410 	if (dir_nodeid != dlm_our_nodeid()) {
4411 		log_error(ls, "receive_remove from %d bad nodeid %d",
4412 			  from_nodeid, dir_nodeid);
4413 		return;
4414 	}
4415 
4416 	/* Look for name on rsbtbl.toss, if it's there, kill it.
4417 	   If it's on rsbtbl.keep, it's being used, and we should ignore this
4418 	   message.  This is an expected race between the dir node sending a
4419 	   request to the master node at the same time as the master node sends
4420 	   a remove to the dir node.  The resolution to that race is for the
4421 	   dir node to ignore the remove message, and the master node to
4422 	   recreate the master rsb when it gets a request from the dir node for
4423 	   an rsb it doesn't have. */
4424 
4425 	memset(name, 0, sizeof(name));
4426 	memcpy(name, ms->m_extra, len);
4427 
4428 	hash = jhash(name, len, 0);
4429 	b = hash & (ls->ls_rsbtbl_size - 1);
4430 
4431 	spin_lock(&ls->ls_rsbtbl[b].lock);
4432 
4433 	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
4434 	if (rv) {
4435 		/* verify the rsb is on keep list per comment above */
4436 		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
4437 		if (rv) {
4438 			/* should not happen */
4439 			log_error(ls, "receive_remove from %d not found %s",
4440 				  from_nodeid, name);
4441 			spin_unlock(&ls->ls_rsbtbl[b].lock);
4442 			return;
4443 		}
4444 		if (r->res_master_nodeid != from_nodeid) {
4445 			/* should not happen */
4446 			log_error(ls, "receive_remove keep from %d master %d",
4447 				  from_nodeid, r->res_master_nodeid);
4448 			dlm_print_rsb(r);
4449 			spin_unlock(&ls->ls_rsbtbl[b].lock);
4450 			return;
4451 		}
4452 
4453 		log_debug(ls, "receive_remove from %d master %d first %x %s",
4454 			  from_nodeid, r->res_master_nodeid, r->res_first_lkid,
4455 			  name);
4456 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4457 		return;
4458 	}
4459 
4460 	if (r->res_master_nodeid != from_nodeid) {
4461 		log_error(ls, "receive_remove toss from %d master %d",
4462 			  from_nodeid, r->res_master_nodeid);
4463 		dlm_print_rsb(r);
4464 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4465 		return;
4466 	}
4467 
4468 	if (kref_put(&r->res_ref, kill_rsb)) {
4469 		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
4470 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4471 		dlm_free_rsb(r);
4472 	} else {
4473 		log_error(ls, "receive_remove from %d rsb ref error",
4474 			  from_nodeid);
4475 		dlm_print_rsb(r);
4476 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4477 	}
4478 }
4479 
4480 static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
4481 {
4482 	do_purge(ls, le32_to_cpu(ms->m_nodeid), le32_to_cpu(ms->m_pid));
4483 }
4484 
4485 static int receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
4486 {
4487 	struct dlm_lkb *lkb;
4488 	struct dlm_rsb *r;
4489 	int error, mstype, result;
4490 	int from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4491 
4492 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4493 	if (error)
4494 		return error;
4495 
4496 	r = lkb->lkb_resource;
4497 	hold_rsb(r);
4498 	lock_rsb(r);
4499 
4500 	error = validate_message(lkb, ms);
4501 	if (error)
4502 		goto out;
4503 
4504 	mstype = lkb->lkb_wait_type;
4505 	error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
4506 	if (error) {
4507 		log_error(ls, "receive_request_reply %x remote %d %x result %d",
4508 			  lkb->lkb_id, from_nodeid, le32_to_cpu(ms->m_lkid),
4509 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4510 		dlm_dump_rsb(r);
4511 		goto out;
4512 	}
4513 
4514 	/* Optimization: the dir node was also the master, so it took our
4515 	   lookup as a request and sent request reply instead of lookup reply */
4516 	if (mstype == DLM_MSG_LOOKUP) {
4517 		r->res_master_nodeid = from_nodeid;
4518 		r->res_nodeid = from_nodeid;
4519 		lkb->lkb_nodeid = from_nodeid;
4520 	}
4521 
4522 	/* this is the value returned from do_request() on the master */
4523 	result = from_dlm_errno(le32_to_cpu(ms->m_result));
4524 
4525 	switch (result) {
4526 	case -EAGAIN:
4527 		/* request would block (be queued) on remote master */
4528 		queue_cast(r, lkb, -EAGAIN);
4529 		confirm_master(r, -EAGAIN);
4530 		unhold_lkb(lkb); /* undoes create_lkb() */
4531 		break;
4532 
4533 	case -EINPROGRESS:
4534 	case 0:
4535 		/* request was queued or granted on remote master */
4536 		receive_flags_reply(lkb, ms);
4537 		lkb->lkb_remid = le32_to_cpu(ms->m_lkid);
4538 		if (is_altmode(lkb))
4539 			munge_altmode(lkb, ms);
4540 		if (result) {
4541 			add_lkb(r, lkb, DLM_LKSTS_WAITING);
4542 			add_timeout(lkb);
4543 		} else {
4544 			grant_lock_pc(r, lkb, ms);
4545 			queue_cast(r, lkb, 0);
4546 		}
4547 		confirm_master(r, result);
4548 		break;
4549 
4550 	case -EBADR:
4551 	case -ENOTBLK:
4552 		/* find_rsb failed to find rsb or rsb wasn't master */
4553 		log_limit(ls, "receive_request_reply %x from %d %d "
4554 			  "master %d dir %d first %x %s", lkb->lkb_id,
4555 			  from_nodeid, result, r->res_master_nodeid,
4556 			  r->res_dir_nodeid, r->res_first_lkid, r->res_name);
4557 
4558 		if (r->res_dir_nodeid != dlm_our_nodeid() &&
4559 		    r->res_master_nodeid != dlm_our_nodeid()) {
4560 			/* cause _request_lock->set_master->send_lookup */
4561 			r->res_master_nodeid = 0;
4562 			r->res_nodeid = -1;
4563 			lkb->lkb_nodeid = -1;
4564 		}
4565 
4566 		if (is_overlap(lkb)) {
4567 			/* we'll ignore error in cancel/unlock reply */
4568 			queue_cast_overlap(r, lkb);
4569 			confirm_master(r, result);
4570 			unhold_lkb(lkb); /* undoes create_lkb() */
4571 		} else {
4572 			_request_lock(r, lkb);
4573 
4574 			if (r->res_master_nodeid == dlm_our_nodeid())
4575 				confirm_master(r, 0);
4576 		}
4577 		break;
4578 
4579 	default:
4580 		log_error(ls, "receive_request_reply %x error %d",
4581 			  lkb->lkb_id, result);
4582 	}
4583 
4584 	if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
4585 		log_debug(ls, "receive_request_reply %x result %d unlock",
4586 			  lkb->lkb_id, result);
4587 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4588 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4589 		send_unlock(r, lkb);
4590 	} else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
4591 		log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
4592 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4593 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4594 		send_cancel(r, lkb);
4595 	} else {
4596 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4597 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4598 	}
4599  out:
4600 	unlock_rsb(r);
4601 	put_rsb(r);
4602 	dlm_put_lkb(lkb);
4603 	return 0;
4604 }
4605 
4606 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
4607 				    struct dlm_message *ms)
4608 {
4609 	/* this is the value returned from do_convert() on the master */
4610 	switch (from_dlm_errno(le32_to_cpu(ms->m_result))) {
4611 	case -EAGAIN:
4612 		/* convert would block (be queued) on remote master */
4613 		queue_cast(r, lkb, -EAGAIN);
4614 		break;
4615 
4616 	case -EDEADLK:
4617 		receive_flags_reply(lkb, ms);
4618 		revert_lock_pc(r, lkb);
4619 		queue_cast(r, lkb, -EDEADLK);
4620 		break;
4621 
4622 	case -EINPROGRESS:
4623 		/* convert was queued on remote master */
4624 		receive_flags_reply(lkb, ms);
4625 		if (is_demoted(lkb))
4626 			munge_demoted(lkb);
4627 		del_lkb(r, lkb);
4628 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
4629 		add_timeout(lkb);
4630 		break;
4631 
4632 	case 0:
4633 		/* convert was granted on remote master */
4634 		receive_flags_reply(lkb, ms);
4635 		if (is_demoted(lkb))
4636 			munge_demoted(lkb);
4637 		grant_lock_pc(r, lkb, ms);
4638 		queue_cast(r, lkb, 0);
4639 		break;
4640 
4641 	default:
4642 		log_error(r->res_ls, "receive_convert_reply %x remote %d %x %d",
4643 			  lkb->lkb_id, le32_to_cpu(ms->m_header.h_nodeid),
4644 			  le32_to_cpu(ms->m_lkid),
4645 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4646 		dlm_print_rsb(r);
4647 		dlm_print_lkb(lkb);
4648 	}
4649 }
4650 
4651 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
4652 {
4653 	struct dlm_rsb *r = lkb->lkb_resource;
4654 	int error;
4655 
4656 	hold_rsb(r);
4657 	lock_rsb(r);
4658 
4659 	error = validate_message(lkb, ms);
4660 	if (error)
4661 		goto out;
4662 
4663 	/* stub reply can happen with waiters_mutex held */
4664 	error = remove_from_waiters_ms(lkb, ms);
4665 	if (error)
4666 		goto out;
4667 
4668 	__receive_convert_reply(r, lkb, ms);
4669  out:
4670 	unlock_rsb(r);
4671 	put_rsb(r);
4672 }
4673 
4674 static int receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
4675 {
4676 	struct dlm_lkb *lkb;
4677 	int error;
4678 
4679 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4680 	if (error)
4681 		return error;
4682 
4683 	_receive_convert_reply(lkb, ms);
4684 	dlm_put_lkb(lkb);
4685 	return 0;
4686 }
4687 
4688 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
4689 {
4690 	struct dlm_rsb *r = lkb->lkb_resource;
4691 	int error;
4692 
4693 	hold_rsb(r);
4694 	lock_rsb(r);
4695 
4696 	error = validate_message(lkb, ms);
4697 	if (error)
4698 		goto out;
4699 
4700 	/* stub reply can happen with waiters_mutex held */
4701 	error = remove_from_waiters_ms(lkb, ms);
4702 	if (error)
4703 		goto out;
4704 
4705 	/* this is the value returned from do_unlock() on the master */
4706 
4707 	switch (from_dlm_errno(le32_to_cpu(ms->m_result))) {
4708 	case -DLM_EUNLOCK:
4709 		receive_flags_reply(lkb, ms);
4710 		remove_lock_pc(r, lkb);
4711 		queue_cast(r, lkb, -DLM_EUNLOCK);
4712 		break;
4713 	case -ENOENT:
4714 		break;
4715 	default:
4716 		log_error(r->res_ls, "receive_unlock_reply %x error %d",
4717 			  lkb->lkb_id, from_dlm_errno(le32_to_cpu(ms->m_result)));
4718 	}
4719  out:
4720 	unlock_rsb(r);
4721 	put_rsb(r);
4722 }
4723 
4724 static int receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
4725 {
4726 	struct dlm_lkb *lkb;
4727 	int error;
4728 
4729 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4730 	if (error)
4731 		return error;
4732 
4733 	_receive_unlock_reply(lkb, ms);
4734 	dlm_put_lkb(lkb);
4735 	return 0;
4736 }
4737 
4738 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
4739 {
4740 	struct dlm_rsb *r = lkb->lkb_resource;
4741 	int error;
4742 
4743 	hold_rsb(r);
4744 	lock_rsb(r);
4745 
4746 	error = validate_message(lkb, ms);
4747 	if (error)
4748 		goto out;
4749 
4750 	/* stub reply can happen with waiters_mutex held */
4751 	error = remove_from_waiters_ms(lkb, ms);
4752 	if (error)
4753 		goto out;
4754 
4755 	/* this is the value returned from do_cancel() on the master */
4756 
4757 	switch (from_dlm_errno(le32_to_cpu(ms->m_result))) {
4758 	case -DLM_ECANCEL:
4759 		receive_flags_reply(lkb, ms);
4760 		revert_lock_pc(r, lkb);
4761 		queue_cast(r, lkb, -DLM_ECANCEL);
4762 		break;
4763 	case 0:
4764 		break;
4765 	default:
4766 		log_error(r->res_ls, "receive_cancel_reply %x error %d",
4767 			  lkb->lkb_id,
4768 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4769 	}
4770  out:
4771 	unlock_rsb(r);
4772 	put_rsb(r);
4773 }
4774 
4775 static int receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
4776 {
4777 	struct dlm_lkb *lkb;
4778 	int error;
4779 
4780 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4781 	if (error)
4782 		return error;
4783 
4784 	_receive_cancel_reply(lkb, ms);
4785 	dlm_put_lkb(lkb);
4786 	return 0;
4787 }
4788 
4789 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
4790 {
4791 	struct dlm_lkb *lkb;
4792 	struct dlm_rsb *r;
4793 	int error, ret_nodeid;
4794 	int do_lookup_list = 0;
4795 
4796 	error = find_lkb(ls, le32_to_cpu(ms->m_lkid), &lkb);
4797 	if (error) {
4798 		log_error(ls, "%s no lkid %x", __func__,
4799 			  le32_to_cpu(ms->m_lkid));
4800 		return;
4801 	}
4802 
4803 	/* ms->m_result is the value returned by dlm_master_lookup on dir node
4804 	   FIXME: will a non-zero error ever be returned? */
4805 
4806 	r = lkb->lkb_resource;
4807 	hold_rsb(r);
4808 	lock_rsb(r);
4809 
4810 	error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
4811 	if (error)
4812 		goto out;
4813 
4814 	ret_nodeid = le32_to_cpu(ms->m_nodeid);
4815 
4816 	/* We sometimes receive a request from the dir node for this
4817 	   rsb before we've received the dir node's loookup_reply for it.
4818 	   The request from the dir node implies we're the master, so we set
4819 	   ourself as master in receive_request_reply, and verify here that
4820 	   we are indeed the master. */
4821 
4822 	if (r->res_master_nodeid && (r->res_master_nodeid != ret_nodeid)) {
4823 		/* This should never happen */
4824 		log_error(ls, "receive_lookup_reply %x from %d ret %d "
4825 			  "master %d dir %d our %d first %x %s",
4826 			  lkb->lkb_id, le32_to_cpu(ms->m_header.h_nodeid),
4827 			  ret_nodeid, r->res_master_nodeid, r->res_dir_nodeid,
4828 			  dlm_our_nodeid(), r->res_first_lkid, r->res_name);
4829 	}
4830 
4831 	if (ret_nodeid == dlm_our_nodeid()) {
4832 		r->res_master_nodeid = ret_nodeid;
4833 		r->res_nodeid = 0;
4834 		do_lookup_list = 1;
4835 		r->res_first_lkid = 0;
4836 	} else if (ret_nodeid == -1) {
4837 		/* the remote node doesn't believe it's the dir node */
4838 		log_error(ls, "receive_lookup_reply %x from %d bad ret_nodeid",
4839 			  lkb->lkb_id, le32_to_cpu(ms->m_header.h_nodeid));
4840 		r->res_master_nodeid = 0;
4841 		r->res_nodeid = -1;
4842 		lkb->lkb_nodeid = -1;
4843 	} else {
4844 		/* set_master() will set lkb_nodeid from r */
4845 		r->res_master_nodeid = ret_nodeid;
4846 		r->res_nodeid = ret_nodeid;
4847 	}
4848 
4849 	if (is_overlap(lkb)) {
4850 		log_debug(ls, "receive_lookup_reply %x unlock %x",
4851 			  lkb->lkb_id, lkb->lkb_flags);
4852 		queue_cast_overlap(r, lkb);
4853 		unhold_lkb(lkb); /* undoes create_lkb() */
4854 		goto out_list;
4855 	}
4856 
4857 	_request_lock(r, lkb);
4858 
4859  out_list:
4860 	if (do_lookup_list)
4861 		process_lookup_list(r);
4862  out:
4863 	unlock_rsb(r);
4864 	put_rsb(r);
4865 	dlm_put_lkb(lkb);
4866 }
4867 
4868 static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms,
4869 			     uint32_t saved_seq)
4870 {
4871 	int error = 0, noent = 0;
4872 
4873 	if (!dlm_is_member(ls, le32_to_cpu(ms->m_header.h_nodeid))) {
4874 		log_limit(ls, "receive %d from non-member %d %x %x %d",
4875 			  le32_to_cpu(ms->m_type),
4876 			  le32_to_cpu(ms->m_header.h_nodeid),
4877 			  le32_to_cpu(ms->m_lkid), le32_to_cpu(ms->m_remid),
4878 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4879 		return;
4880 	}
4881 
4882 	switch (ms->m_type) {
4883 
4884 	/* messages sent to a master node */
4885 
4886 	case cpu_to_le32(DLM_MSG_REQUEST):
4887 		error = receive_request(ls, ms);
4888 		break;
4889 
4890 	case cpu_to_le32(DLM_MSG_CONVERT):
4891 		error = receive_convert(ls, ms);
4892 		break;
4893 
4894 	case cpu_to_le32(DLM_MSG_UNLOCK):
4895 		error = receive_unlock(ls, ms);
4896 		break;
4897 
4898 	case cpu_to_le32(DLM_MSG_CANCEL):
4899 		noent = 1;
4900 		error = receive_cancel(ls, ms);
4901 		break;
4902 
4903 	/* messages sent from a master node (replies to above) */
4904 
4905 	case cpu_to_le32(DLM_MSG_REQUEST_REPLY):
4906 		error = receive_request_reply(ls, ms);
4907 		break;
4908 
4909 	case cpu_to_le32(DLM_MSG_CONVERT_REPLY):
4910 		error = receive_convert_reply(ls, ms);
4911 		break;
4912 
4913 	case cpu_to_le32(DLM_MSG_UNLOCK_REPLY):
4914 		error = receive_unlock_reply(ls, ms);
4915 		break;
4916 
4917 	case cpu_to_le32(DLM_MSG_CANCEL_REPLY):
4918 		error = receive_cancel_reply(ls, ms);
4919 		break;
4920 
4921 	/* messages sent from a master node (only two types of async msg) */
4922 
4923 	case cpu_to_le32(DLM_MSG_GRANT):
4924 		noent = 1;
4925 		error = receive_grant(ls, ms);
4926 		break;
4927 
4928 	case cpu_to_le32(DLM_MSG_BAST):
4929 		noent = 1;
4930 		error = receive_bast(ls, ms);
4931 		break;
4932 
4933 	/* messages sent to a dir node */
4934 
4935 	case cpu_to_le32(DLM_MSG_LOOKUP):
4936 		receive_lookup(ls, ms);
4937 		break;
4938 
4939 	case cpu_to_le32(DLM_MSG_REMOVE):
4940 		receive_remove(ls, ms);
4941 		break;
4942 
4943 	/* messages sent from a dir node (remove has no reply) */
4944 
4945 	case cpu_to_le32(DLM_MSG_LOOKUP_REPLY):
4946 		receive_lookup_reply(ls, ms);
4947 		break;
4948 
4949 	/* other messages */
4950 
4951 	case cpu_to_le32(DLM_MSG_PURGE):
4952 		receive_purge(ls, ms);
4953 		break;
4954 
4955 	default:
4956 		log_error(ls, "unknown message type %d",
4957 			  le32_to_cpu(ms->m_type));
4958 	}
4959 
4960 	/*
4961 	 * When checking for ENOENT, we're checking the result of
4962 	 * find_lkb(m_remid):
4963 	 *
4964 	 * The lock id referenced in the message wasn't found.  This may
4965 	 * happen in normal usage for the async messages and cancel, so
4966 	 * only use log_debug for them.
4967 	 *
4968 	 * Some errors are expected and normal.
4969 	 */
4970 
4971 	if (error == -ENOENT && noent) {
4972 		log_debug(ls, "receive %d no %x remote %d %x saved_seq %u",
4973 			  le32_to_cpu(ms->m_type), le32_to_cpu(ms->m_remid),
4974 			  le32_to_cpu(ms->m_header.h_nodeid),
4975 			  le32_to_cpu(ms->m_lkid), saved_seq);
4976 	} else if (error == -ENOENT) {
4977 		log_error(ls, "receive %d no %x remote %d %x saved_seq %u",
4978 			  le32_to_cpu(ms->m_type), le32_to_cpu(ms->m_remid),
4979 			  le32_to_cpu(ms->m_header.h_nodeid),
4980 			  le32_to_cpu(ms->m_lkid), saved_seq);
4981 
4982 		if (ms->m_type == cpu_to_le32(DLM_MSG_CONVERT))
4983 			dlm_dump_rsb_hash(ls, le32_to_cpu(ms->m_hash));
4984 	}
4985 
4986 	if (error == -EINVAL) {
4987 		log_error(ls, "receive %d inval from %d lkid %x remid %x "
4988 			  "saved_seq %u",
4989 			  le32_to_cpu(ms->m_type),
4990 			  le32_to_cpu(ms->m_header.h_nodeid),
4991 			  le32_to_cpu(ms->m_lkid), le32_to_cpu(ms->m_remid),
4992 			  saved_seq);
4993 	}
4994 }
4995 
4996 /* If the lockspace is in recovery mode (locking stopped), then normal
4997    messages are saved on the requestqueue for processing after recovery is
4998    done.  When not in recovery mode, we wait for dlm_recoverd to drain saved
4999    messages off the requestqueue before we process new ones. This occurs right
5000    after recovery completes when we transition from saving all messages on
5001    requestqueue, to processing all the saved messages, to processing new
5002    messages as they arrive. */
5003 
5004 static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
5005 				int nodeid)
5006 {
5007 	if (dlm_locking_stopped(ls)) {
5008 		/* If we were a member of this lockspace, left, and rejoined,
5009 		   other nodes may still be sending us messages from the
5010 		   lockspace generation before we left. */
5011 		if (!ls->ls_generation) {
5012 			log_limit(ls, "receive %d from %d ignore old gen",
5013 				  le32_to_cpu(ms->m_type), nodeid);
5014 			return;
5015 		}
5016 
5017 		dlm_add_requestqueue(ls, nodeid, ms);
5018 	} else {
5019 		dlm_wait_requestqueue(ls);
5020 		_receive_message(ls, ms, 0);
5021 	}
5022 }
5023 
5024 /* This is called by dlm_recoverd to process messages that were saved on
5025    the requestqueue. */
5026 
5027 void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms,
5028 			       uint32_t saved_seq)
5029 {
5030 	_receive_message(ls, ms, saved_seq);
5031 }
5032 
5033 /* This is called by the midcomms layer when something is received for
5034    the lockspace.  It could be either a MSG (normal message sent as part of
5035    standard locking activity) or an RCOM (recovery message sent as part of
5036    lockspace recovery). */
5037 
5038 void dlm_receive_buffer(union dlm_packet *p, int nodeid)
5039 {
5040 	struct dlm_header *hd = &p->header;
5041 	struct dlm_ls *ls;
5042 	int type = 0;
5043 
5044 	switch (hd->h_cmd) {
5045 	case DLM_MSG:
5046 		type = le32_to_cpu(p->message.m_type);
5047 		break;
5048 	case DLM_RCOM:
5049 		type = le32_to_cpu(p->rcom.rc_type);
5050 		break;
5051 	default:
5052 		log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
5053 		return;
5054 	}
5055 
5056 	if (le32_to_cpu(hd->h_nodeid) != nodeid) {
5057 		log_print("invalid h_nodeid %d from %d lockspace %x",
5058 			  le32_to_cpu(hd->h_nodeid), nodeid,
5059 			  le32_to_cpu(hd->u.h_lockspace));
5060 		return;
5061 	}
5062 
5063 	ls = dlm_find_lockspace_global(le32_to_cpu(hd->u.h_lockspace));
5064 	if (!ls) {
5065 		if (dlm_config.ci_log_debug) {
5066 			printk_ratelimited(KERN_DEBUG "dlm: invalid lockspace "
5067 				"%u from %d cmd %d type %d\n",
5068 				le32_to_cpu(hd->u.h_lockspace), nodeid,
5069 				hd->h_cmd, type);
5070 		}
5071 
5072 		if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
5073 			dlm_send_ls_not_ready(nodeid, &p->rcom);
5074 		return;
5075 	}
5076 
5077 	/* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
5078 	   be inactive (in this ls) before transitioning to recovery mode */
5079 
5080 	down_read(&ls->ls_recv_active);
5081 	if (hd->h_cmd == DLM_MSG)
5082 		dlm_receive_message(ls, &p->message, nodeid);
5083 	else
5084 		dlm_receive_rcom(ls, &p->rcom, nodeid);
5085 	up_read(&ls->ls_recv_active);
5086 
5087 	dlm_put_lockspace(ls);
5088 }
5089 
5090 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb,
5091 				   struct dlm_message *ms_stub)
5092 {
5093 	if (middle_conversion(lkb)) {
5094 		hold_lkb(lkb);
5095 		memset(ms_stub, 0, sizeof(struct dlm_message));
5096 		ms_stub->m_flags = cpu_to_le32(DLM_IFL_STUB_MS);
5097 		ms_stub->m_type = cpu_to_le32(DLM_MSG_CONVERT_REPLY);
5098 		ms_stub->m_result = cpu_to_le32(to_dlm_errno(-EINPROGRESS));
5099 		ms_stub->m_header.h_nodeid = cpu_to_le32(lkb->lkb_nodeid);
5100 		_receive_convert_reply(lkb, ms_stub);
5101 
5102 		/* Same special case as in receive_rcom_lock_args() */
5103 		lkb->lkb_grmode = DLM_LOCK_IV;
5104 		rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
5105 		unhold_lkb(lkb);
5106 
5107 	} else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
5108 		lkb->lkb_flags |= DLM_IFL_RESEND;
5109 	}
5110 
5111 	/* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
5112 	   conversions are async; there's no reply from the remote master */
5113 }
5114 
5115 /* A waiting lkb needs recovery if the master node has failed, or
5116    the master node is changing (only when no directory is used) */
5117 
5118 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb,
5119 				 int dir_nodeid)
5120 {
5121 	if (dlm_no_directory(ls))
5122 		return 1;
5123 
5124 	if (dlm_is_removed(ls, lkb->lkb_wait_nodeid))
5125 		return 1;
5126 
5127 	return 0;
5128 }
5129 
5130 /* Recovery for locks that are waiting for replies from nodes that are now
5131    gone.  We can just complete unlocks and cancels by faking a reply from the
5132    dead node.  Requests and up-conversions we flag to be resent after
5133    recovery.  Down-conversions can just be completed with a fake reply like
5134    unlocks.  Conversions between PR and CW need special attention. */
5135 
5136 void dlm_recover_waiters_pre(struct dlm_ls *ls)
5137 {
5138 	struct dlm_lkb *lkb, *safe;
5139 	struct dlm_message *ms_stub;
5140 	int wait_type, stub_unlock_result, stub_cancel_result;
5141 	int dir_nodeid;
5142 
5143 	ms_stub = kmalloc(sizeof(*ms_stub), GFP_KERNEL);
5144 	if (!ms_stub)
5145 		return;
5146 
5147 	mutex_lock(&ls->ls_waiters_mutex);
5148 
5149 	list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
5150 
5151 		dir_nodeid = dlm_dir_nodeid(lkb->lkb_resource);
5152 
5153 		/* exclude debug messages about unlocks because there can be so
5154 		   many and they aren't very interesting */
5155 
5156 		if (lkb->lkb_wait_type != DLM_MSG_UNLOCK) {
5157 			log_debug(ls, "waiter %x remote %x msg %d r_nodeid %d "
5158 				  "lkb_nodeid %d wait_nodeid %d dir_nodeid %d",
5159 				  lkb->lkb_id,
5160 				  lkb->lkb_remid,
5161 				  lkb->lkb_wait_type,
5162 				  lkb->lkb_resource->res_nodeid,
5163 				  lkb->lkb_nodeid,
5164 				  lkb->lkb_wait_nodeid,
5165 				  dir_nodeid);
5166 		}
5167 
5168 		/* all outstanding lookups, regardless of destination  will be
5169 		   resent after recovery is done */
5170 
5171 		if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
5172 			lkb->lkb_flags |= DLM_IFL_RESEND;
5173 			continue;
5174 		}
5175 
5176 		if (!waiter_needs_recovery(ls, lkb, dir_nodeid))
5177 			continue;
5178 
5179 		wait_type = lkb->lkb_wait_type;
5180 		stub_unlock_result = -DLM_EUNLOCK;
5181 		stub_cancel_result = -DLM_ECANCEL;
5182 
5183 		/* Main reply may have been received leaving a zero wait_type,
5184 		   but a reply for the overlapping op may not have been
5185 		   received.  In that case we need to fake the appropriate
5186 		   reply for the overlap op. */
5187 
5188 		if (!wait_type) {
5189 			if (is_overlap_cancel(lkb)) {
5190 				wait_type = DLM_MSG_CANCEL;
5191 				if (lkb->lkb_grmode == DLM_LOCK_IV)
5192 					stub_cancel_result = 0;
5193 			}
5194 			if (is_overlap_unlock(lkb)) {
5195 				wait_type = DLM_MSG_UNLOCK;
5196 				if (lkb->lkb_grmode == DLM_LOCK_IV)
5197 					stub_unlock_result = -ENOENT;
5198 			}
5199 
5200 			log_debug(ls, "rwpre overlap %x %x %d %d %d",
5201 				  lkb->lkb_id, lkb->lkb_flags, wait_type,
5202 				  stub_cancel_result, stub_unlock_result);
5203 		}
5204 
5205 		switch (wait_type) {
5206 
5207 		case DLM_MSG_REQUEST:
5208 			lkb->lkb_flags |= DLM_IFL_RESEND;
5209 			break;
5210 
5211 		case DLM_MSG_CONVERT:
5212 			recover_convert_waiter(ls, lkb, ms_stub);
5213 			break;
5214 
5215 		case DLM_MSG_UNLOCK:
5216 			hold_lkb(lkb);
5217 			memset(ms_stub, 0, sizeof(struct dlm_message));
5218 			ms_stub->m_flags = cpu_to_le32(DLM_IFL_STUB_MS);
5219 			ms_stub->m_type = cpu_to_le32(DLM_MSG_UNLOCK_REPLY);
5220 			ms_stub->m_result = cpu_to_le32(to_dlm_errno(stub_unlock_result));
5221 			ms_stub->m_header.h_nodeid = cpu_to_le32(lkb->lkb_nodeid);
5222 			_receive_unlock_reply(lkb, ms_stub);
5223 			dlm_put_lkb(lkb);
5224 			break;
5225 
5226 		case DLM_MSG_CANCEL:
5227 			hold_lkb(lkb);
5228 			memset(ms_stub, 0, sizeof(struct dlm_message));
5229 			ms_stub->m_flags = cpu_to_le32(DLM_IFL_STUB_MS);
5230 			ms_stub->m_type = cpu_to_le32(DLM_MSG_CANCEL_REPLY);
5231 			ms_stub->m_result = cpu_to_le32(to_dlm_errno(stub_cancel_result));
5232 			ms_stub->m_header.h_nodeid = cpu_to_le32(lkb->lkb_nodeid);
5233 			_receive_cancel_reply(lkb, ms_stub);
5234 			dlm_put_lkb(lkb);
5235 			break;
5236 
5237 		default:
5238 			log_error(ls, "invalid lkb wait_type %d %d",
5239 				  lkb->lkb_wait_type, wait_type);
5240 		}
5241 		schedule();
5242 	}
5243 	mutex_unlock(&ls->ls_waiters_mutex);
5244 	kfree(ms_stub);
5245 }
5246 
5247 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
5248 {
5249 	struct dlm_lkb *lkb = NULL, *iter;
5250 
5251 	mutex_lock(&ls->ls_waiters_mutex);
5252 	list_for_each_entry(iter, &ls->ls_waiters, lkb_wait_reply) {
5253 		if (iter->lkb_flags & DLM_IFL_RESEND) {
5254 			hold_lkb(iter);
5255 			lkb = iter;
5256 			break;
5257 		}
5258 	}
5259 	mutex_unlock(&ls->ls_waiters_mutex);
5260 
5261 	return lkb;
5262 }
5263 
5264 /* Deal with lookups and lkb's marked RESEND from _pre.  We may now be the
5265    master or dir-node for r.  Processing the lkb may result in it being placed
5266    back on waiters. */
5267 
5268 /* We do this after normal locking has been enabled and any saved messages
5269    (in requestqueue) have been processed.  We should be confident that at
5270    this point we won't get or process a reply to any of these waiting
5271    operations.  But, new ops may be coming in on the rsbs/locks here from
5272    userspace or remotely. */
5273 
5274 /* there may have been an overlap unlock/cancel prior to recovery or after
5275    recovery.  if before, the lkb may still have a pos wait_count; if after, the
5276    overlap flag would just have been set and nothing new sent.  we can be
5277    confident here than any replies to either the initial op or overlap ops
5278    prior to recovery have been received. */
5279 
5280 int dlm_recover_waiters_post(struct dlm_ls *ls)
5281 {
5282 	struct dlm_lkb *lkb;
5283 	struct dlm_rsb *r;
5284 	int error = 0, mstype, err, oc, ou;
5285 
5286 	while (1) {
5287 		if (dlm_locking_stopped(ls)) {
5288 			log_debug(ls, "recover_waiters_post aborted");
5289 			error = -EINTR;
5290 			break;
5291 		}
5292 
5293 		lkb = find_resend_waiter(ls);
5294 		if (!lkb)
5295 			break;
5296 
5297 		r = lkb->lkb_resource;
5298 		hold_rsb(r);
5299 		lock_rsb(r);
5300 
5301 		mstype = lkb->lkb_wait_type;
5302 		oc = is_overlap_cancel(lkb);
5303 		ou = is_overlap_unlock(lkb);
5304 		err = 0;
5305 
5306 		log_debug(ls, "waiter %x remote %x msg %d r_nodeid %d "
5307 			  "lkb_nodeid %d wait_nodeid %d dir_nodeid %d "
5308 			  "overlap %d %d", lkb->lkb_id, lkb->lkb_remid, mstype,
5309 			  r->res_nodeid, lkb->lkb_nodeid, lkb->lkb_wait_nodeid,
5310 			  dlm_dir_nodeid(r), oc, ou);
5311 
5312 		/* At this point we assume that we won't get a reply to any
5313 		   previous op or overlap op on this lock.  First, do a big
5314 		   remove_from_waiters() for all previous ops. */
5315 
5316 		lkb->lkb_flags &= ~DLM_IFL_RESEND;
5317 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
5318 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
5319 		lkb->lkb_wait_type = 0;
5320 		/* drop all wait_count references we still
5321 		 * hold a reference for this iteration.
5322 		 */
5323 		while (lkb->lkb_wait_count) {
5324 			lkb->lkb_wait_count--;
5325 			unhold_lkb(lkb);
5326 		}
5327 		mutex_lock(&ls->ls_waiters_mutex);
5328 		list_del_init(&lkb->lkb_wait_reply);
5329 		mutex_unlock(&ls->ls_waiters_mutex);
5330 
5331 		if (oc || ou) {
5332 			/* do an unlock or cancel instead of resending */
5333 			switch (mstype) {
5334 			case DLM_MSG_LOOKUP:
5335 			case DLM_MSG_REQUEST:
5336 				queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
5337 							-DLM_ECANCEL);
5338 				unhold_lkb(lkb); /* undoes create_lkb() */
5339 				break;
5340 			case DLM_MSG_CONVERT:
5341 				if (oc) {
5342 					queue_cast(r, lkb, -DLM_ECANCEL);
5343 				} else {
5344 					lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
5345 					_unlock_lock(r, lkb);
5346 				}
5347 				break;
5348 			default:
5349 				err = 1;
5350 			}
5351 		} else {
5352 			switch (mstype) {
5353 			case DLM_MSG_LOOKUP:
5354 			case DLM_MSG_REQUEST:
5355 				_request_lock(r, lkb);
5356 				if (is_master(r))
5357 					confirm_master(r, 0);
5358 				break;
5359 			case DLM_MSG_CONVERT:
5360 				_convert_lock(r, lkb);
5361 				break;
5362 			default:
5363 				err = 1;
5364 			}
5365 		}
5366 
5367 		if (err) {
5368 			log_error(ls, "waiter %x msg %d r_nodeid %d "
5369 				  "dir_nodeid %d overlap %d %d",
5370 				  lkb->lkb_id, mstype, r->res_nodeid,
5371 				  dlm_dir_nodeid(r), oc, ou);
5372 		}
5373 		unlock_rsb(r);
5374 		put_rsb(r);
5375 		dlm_put_lkb(lkb);
5376 	}
5377 
5378 	return error;
5379 }
5380 
5381 static void purge_mstcpy_list(struct dlm_ls *ls, struct dlm_rsb *r,
5382 			      struct list_head *list)
5383 {
5384 	struct dlm_lkb *lkb, *safe;
5385 
5386 	list_for_each_entry_safe(lkb, safe, list, lkb_statequeue) {
5387 		if (!is_master_copy(lkb))
5388 			continue;
5389 
5390 		/* don't purge lkbs we've added in recover_master_copy for
5391 		   the current recovery seq */
5392 
5393 		if (lkb->lkb_recover_seq == ls->ls_recover_seq)
5394 			continue;
5395 
5396 		del_lkb(r, lkb);
5397 
5398 		/* this put should free the lkb */
5399 		if (!dlm_put_lkb(lkb))
5400 			log_error(ls, "purged mstcpy lkb not released");
5401 	}
5402 }
5403 
5404 void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
5405 {
5406 	struct dlm_ls *ls = r->res_ls;
5407 
5408 	purge_mstcpy_list(ls, r, &r->res_grantqueue);
5409 	purge_mstcpy_list(ls, r, &r->res_convertqueue);
5410 	purge_mstcpy_list(ls, r, &r->res_waitqueue);
5411 }
5412 
5413 static void purge_dead_list(struct dlm_ls *ls, struct dlm_rsb *r,
5414 			    struct list_head *list,
5415 			    int nodeid_gone, unsigned int *count)
5416 {
5417 	struct dlm_lkb *lkb, *safe;
5418 
5419 	list_for_each_entry_safe(lkb, safe, list, lkb_statequeue) {
5420 		if (!is_master_copy(lkb))
5421 			continue;
5422 
5423 		if ((lkb->lkb_nodeid == nodeid_gone) ||
5424 		    dlm_is_removed(ls, lkb->lkb_nodeid)) {
5425 
5426 			/* tell recover_lvb to invalidate the lvb
5427 			   because a node holding EX/PW failed */
5428 			if ((lkb->lkb_exflags & DLM_LKF_VALBLK) &&
5429 			    (lkb->lkb_grmode >= DLM_LOCK_PW)) {
5430 				rsb_set_flag(r, RSB_RECOVER_LVB_INVAL);
5431 			}
5432 
5433 			del_lkb(r, lkb);
5434 
5435 			/* this put should free the lkb */
5436 			if (!dlm_put_lkb(lkb))
5437 				log_error(ls, "purged dead lkb not released");
5438 
5439 			rsb_set_flag(r, RSB_RECOVER_GRANT);
5440 
5441 			(*count)++;
5442 		}
5443 	}
5444 }
5445 
5446 /* Get rid of locks held by nodes that are gone. */
5447 
5448 void dlm_recover_purge(struct dlm_ls *ls)
5449 {
5450 	struct dlm_rsb *r;
5451 	struct dlm_member *memb;
5452 	int nodes_count = 0;
5453 	int nodeid_gone = 0;
5454 	unsigned int lkb_count = 0;
5455 
5456 	/* cache one removed nodeid to optimize the common
5457 	   case of a single node removed */
5458 
5459 	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
5460 		nodes_count++;
5461 		nodeid_gone = memb->nodeid;
5462 	}
5463 
5464 	if (!nodes_count)
5465 		return;
5466 
5467 	down_write(&ls->ls_root_sem);
5468 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
5469 		hold_rsb(r);
5470 		lock_rsb(r);
5471 		if (is_master(r)) {
5472 			purge_dead_list(ls, r, &r->res_grantqueue,
5473 					nodeid_gone, &lkb_count);
5474 			purge_dead_list(ls, r, &r->res_convertqueue,
5475 					nodeid_gone, &lkb_count);
5476 			purge_dead_list(ls, r, &r->res_waitqueue,
5477 					nodeid_gone, &lkb_count);
5478 		}
5479 		unlock_rsb(r);
5480 		unhold_rsb(r);
5481 		cond_resched();
5482 	}
5483 	up_write(&ls->ls_root_sem);
5484 
5485 	if (lkb_count)
5486 		log_rinfo(ls, "dlm_recover_purge %u locks for %u nodes",
5487 			  lkb_count, nodes_count);
5488 }
5489 
5490 static struct dlm_rsb *find_grant_rsb(struct dlm_ls *ls, int bucket)
5491 {
5492 	struct rb_node *n;
5493 	struct dlm_rsb *r;
5494 
5495 	spin_lock(&ls->ls_rsbtbl[bucket].lock);
5496 	for (n = rb_first(&ls->ls_rsbtbl[bucket].keep); n; n = rb_next(n)) {
5497 		r = rb_entry(n, struct dlm_rsb, res_hashnode);
5498 
5499 		if (!rsb_flag(r, RSB_RECOVER_GRANT))
5500 			continue;
5501 		if (!is_master(r)) {
5502 			rsb_clear_flag(r, RSB_RECOVER_GRANT);
5503 			continue;
5504 		}
5505 		hold_rsb(r);
5506 		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
5507 		return r;
5508 	}
5509 	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
5510 	return NULL;
5511 }
5512 
5513 /*
5514  * Attempt to grant locks on resources that we are the master of.
5515  * Locks may have become grantable during recovery because locks
5516  * from departed nodes have been purged (or not rebuilt), allowing
5517  * previously blocked locks to now be granted.  The subset of rsb's
5518  * we are interested in are those with lkb's on either the convert or
5519  * waiting queues.
5520  *
5521  * Simplest would be to go through each master rsb and check for non-empty
5522  * convert or waiting queues, and attempt to grant on those rsbs.
5523  * Checking the queues requires lock_rsb, though, for which we'd need
5524  * to release the rsbtbl lock.  This would make iterating through all
5525  * rsb's very inefficient.  So, we rely on earlier recovery routines
5526  * to set RECOVER_GRANT on any rsb's that we should attempt to grant
5527  * locks for.
5528  */
5529 
5530 void dlm_recover_grant(struct dlm_ls *ls)
5531 {
5532 	struct dlm_rsb *r;
5533 	int bucket = 0;
5534 	unsigned int count = 0;
5535 	unsigned int rsb_count = 0;
5536 	unsigned int lkb_count = 0;
5537 
5538 	while (1) {
5539 		r = find_grant_rsb(ls, bucket);
5540 		if (!r) {
5541 			if (bucket == ls->ls_rsbtbl_size - 1)
5542 				break;
5543 			bucket++;
5544 			continue;
5545 		}
5546 		rsb_count++;
5547 		count = 0;
5548 		lock_rsb(r);
5549 		/* the RECOVER_GRANT flag is checked in the grant path */
5550 		grant_pending_locks(r, &count);
5551 		rsb_clear_flag(r, RSB_RECOVER_GRANT);
5552 		lkb_count += count;
5553 		confirm_master(r, 0);
5554 		unlock_rsb(r);
5555 		put_rsb(r);
5556 		cond_resched();
5557 	}
5558 
5559 	if (lkb_count)
5560 		log_rinfo(ls, "dlm_recover_grant %u locks on %u resources",
5561 			  lkb_count, rsb_count);
5562 }
5563 
5564 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
5565 					 uint32_t remid)
5566 {
5567 	struct dlm_lkb *lkb;
5568 
5569 	list_for_each_entry(lkb, head, lkb_statequeue) {
5570 		if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
5571 			return lkb;
5572 	}
5573 	return NULL;
5574 }
5575 
5576 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
5577 				    uint32_t remid)
5578 {
5579 	struct dlm_lkb *lkb;
5580 
5581 	lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
5582 	if (lkb)
5583 		return lkb;
5584 	lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
5585 	if (lkb)
5586 		return lkb;
5587 	lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
5588 	if (lkb)
5589 		return lkb;
5590 	return NULL;
5591 }
5592 
5593 /* needs at least dlm_rcom + rcom_lock */
5594 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
5595 				  struct dlm_rsb *r, struct dlm_rcom *rc)
5596 {
5597 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
5598 
5599 	lkb->lkb_nodeid = le32_to_cpu(rc->rc_header.h_nodeid);
5600 	lkb->lkb_ownpid = le32_to_cpu(rl->rl_ownpid);
5601 	lkb->lkb_remid = le32_to_cpu(rl->rl_lkid);
5602 	lkb->lkb_exflags = le32_to_cpu(rl->rl_exflags);
5603 	lkb->lkb_flags = le32_to_cpu(rl->rl_flags) & 0x0000FFFF;
5604 	lkb->lkb_flags |= DLM_IFL_MSTCPY;
5605 	lkb->lkb_lvbseq = le32_to_cpu(rl->rl_lvbseq);
5606 	lkb->lkb_rqmode = rl->rl_rqmode;
5607 	lkb->lkb_grmode = rl->rl_grmode;
5608 	/* don't set lkb_status because add_lkb wants to itself */
5609 
5610 	lkb->lkb_bastfn = (rl->rl_asts & DLM_CB_BAST) ? &fake_bastfn : NULL;
5611 	lkb->lkb_astfn = (rl->rl_asts & DLM_CB_CAST) ? &fake_astfn : NULL;
5612 
5613 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
5614 		int lvblen = le16_to_cpu(rc->rc_header.h_length) -
5615 			sizeof(struct dlm_rcom) - sizeof(struct rcom_lock);
5616 		if (lvblen > ls->ls_lvblen)
5617 			return -EINVAL;
5618 		lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
5619 		if (!lkb->lkb_lvbptr)
5620 			return -ENOMEM;
5621 		memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
5622 	}
5623 
5624 	/* Conversions between PR and CW (middle modes) need special handling.
5625 	   The real granted mode of these converting locks cannot be determined
5626 	   until all locks have been rebuilt on the rsb (recover_conversion) */
5627 
5628 	if (rl->rl_wait_type == cpu_to_le16(DLM_MSG_CONVERT) &&
5629 	    middle_conversion(lkb)) {
5630 		rl->rl_status = DLM_LKSTS_CONVERT;
5631 		lkb->lkb_grmode = DLM_LOCK_IV;
5632 		rsb_set_flag(r, RSB_RECOVER_CONVERT);
5633 	}
5634 
5635 	return 0;
5636 }
5637 
5638 /* This lkb may have been recovered in a previous aborted recovery so we need
5639    to check if the rsb already has an lkb with the given remote nodeid/lkid.
5640    If so we just send back a standard reply.  If not, we create a new lkb with
5641    the given values and send back our lkid.  We send back our lkid by sending
5642    back the rcom_lock struct we got but with the remid field filled in. */
5643 
5644 /* needs at least dlm_rcom + rcom_lock */
5645 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
5646 {
5647 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
5648 	struct dlm_rsb *r;
5649 	struct dlm_lkb *lkb;
5650 	uint32_t remid = 0;
5651 	int from_nodeid = le32_to_cpu(rc->rc_header.h_nodeid);
5652 	int error;
5653 
5654 	if (rl->rl_parent_lkid) {
5655 		error = -EOPNOTSUPP;
5656 		goto out;
5657 	}
5658 
5659 	remid = le32_to_cpu(rl->rl_lkid);
5660 
5661 	/* In general we expect the rsb returned to be R_MASTER, but we don't
5662 	   have to require it.  Recovery of masters on one node can overlap
5663 	   recovery of locks on another node, so one node can send us MSTCPY
5664 	   locks before we've made ourselves master of this rsb.  We can still
5665 	   add new MSTCPY locks that we receive here without any harm; when
5666 	   we make ourselves master, dlm_recover_masters() won't touch the
5667 	   MSTCPY locks we've received early. */
5668 
5669 	error = find_rsb(ls, rl->rl_name, le16_to_cpu(rl->rl_namelen),
5670 			 from_nodeid, R_RECEIVE_RECOVER, &r);
5671 	if (error)
5672 		goto out;
5673 
5674 	lock_rsb(r);
5675 
5676 	if (dlm_no_directory(ls) && (dlm_dir_nodeid(r) != dlm_our_nodeid())) {
5677 		log_error(ls, "dlm_recover_master_copy remote %d %x not dir",
5678 			  from_nodeid, remid);
5679 		error = -EBADR;
5680 		goto out_unlock;
5681 	}
5682 
5683 	lkb = search_remid(r, from_nodeid, remid);
5684 	if (lkb) {
5685 		error = -EEXIST;
5686 		goto out_remid;
5687 	}
5688 
5689 	error = create_lkb(ls, &lkb);
5690 	if (error)
5691 		goto out_unlock;
5692 
5693 	error = receive_rcom_lock_args(ls, lkb, r, rc);
5694 	if (error) {
5695 		__put_lkb(ls, lkb);
5696 		goto out_unlock;
5697 	}
5698 
5699 	attach_lkb(r, lkb);
5700 	add_lkb(r, lkb, rl->rl_status);
5701 	ls->ls_recover_locks_in++;
5702 
5703 	if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
5704 		rsb_set_flag(r, RSB_RECOVER_GRANT);
5705 
5706  out_remid:
5707 	/* this is the new value returned to the lock holder for
5708 	   saving in its process-copy lkb */
5709 	rl->rl_remid = cpu_to_le32(lkb->lkb_id);
5710 
5711 	lkb->lkb_recover_seq = ls->ls_recover_seq;
5712 
5713  out_unlock:
5714 	unlock_rsb(r);
5715 	put_rsb(r);
5716  out:
5717 	if (error && error != -EEXIST)
5718 		log_rinfo(ls, "dlm_recover_master_copy remote %d %x error %d",
5719 			  from_nodeid, remid, error);
5720 	rl->rl_result = cpu_to_le32(error);
5721 	return error;
5722 }
5723 
5724 /* needs at least dlm_rcom + rcom_lock */
5725 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
5726 {
5727 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
5728 	struct dlm_rsb *r;
5729 	struct dlm_lkb *lkb;
5730 	uint32_t lkid, remid;
5731 	int error, result;
5732 
5733 	lkid = le32_to_cpu(rl->rl_lkid);
5734 	remid = le32_to_cpu(rl->rl_remid);
5735 	result = le32_to_cpu(rl->rl_result);
5736 
5737 	error = find_lkb(ls, lkid, &lkb);
5738 	if (error) {
5739 		log_error(ls, "dlm_recover_process_copy no %x remote %d %x %d",
5740 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5741 			  result);
5742 		return error;
5743 	}
5744 
5745 	r = lkb->lkb_resource;
5746 	hold_rsb(r);
5747 	lock_rsb(r);
5748 
5749 	if (!is_process_copy(lkb)) {
5750 		log_error(ls, "dlm_recover_process_copy bad %x remote %d %x %d",
5751 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5752 			  result);
5753 		dlm_dump_rsb(r);
5754 		unlock_rsb(r);
5755 		put_rsb(r);
5756 		dlm_put_lkb(lkb);
5757 		return -EINVAL;
5758 	}
5759 
5760 	switch (result) {
5761 	case -EBADR:
5762 		/* There's a chance the new master received our lock before
5763 		   dlm_recover_master_reply(), this wouldn't happen if we did
5764 		   a barrier between recover_masters and recover_locks. */
5765 
5766 		log_debug(ls, "dlm_recover_process_copy %x remote %d %x %d",
5767 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5768 			  result);
5769 
5770 		dlm_send_rcom_lock(r, lkb);
5771 		goto out;
5772 	case -EEXIST:
5773 	case 0:
5774 		lkb->lkb_remid = remid;
5775 		break;
5776 	default:
5777 		log_error(ls, "dlm_recover_process_copy %x remote %d %x %d unk",
5778 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5779 			  result);
5780 	}
5781 
5782 	/* an ack for dlm_recover_locks() which waits for replies from
5783 	   all the locks it sends to new masters */
5784 	dlm_recovered_lock(r);
5785  out:
5786 	unlock_rsb(r);
5787 	put_rsb(r);
5788 	dlm_put_lkb(lkb);
5789 
5790 	return 0;
5791 }
5792 
5793 #ifdef CONFIG_DLM_DEPRECATED_API
5794 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
5795 		     int mode, uint32_t flags, void *name, unsigned int namelen,
5796 		     unsigned long timeout_cs)
5797 #else
5798 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
5799 		     int mode, uint32_t flags, void *name, unsigned int namelen)
5800 #endif
5801 {
5802 	struct dlm_lkb *lkb;
5803 	struct dlm_args args;
5804 	int error;
5805 
5806 	dlm_lock_recovery(ls);
5807 
5808 	error = create_lkb(ls, &lkb);
5809 	if (error) {
5810 		kfree(ua);
5811 		goto out;
5812 	}
5813 
5814 	if (flags & DLM_LKF_VALBLK) {
5815 		ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_NOFS);
5816 		if (!ua->lksb.sb_lvbptr) {
5817 			kfree(ua);
5818 			__put_lkb(ls, lkb);
5819 			error = -ENOMEM;
5820 			goto out;
5821 		}
5822 	}
5823 #ifdef CONFIG_DLM_DEPRECATED_API
5824 	error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
5825 			      fake_astfn, ua, fake_bastfn, &args);
5826 #else
5827 	error = set_lock_args(mode, &ua->lksb, flags, namelen, fake_astfn, ua,
5828 			      fake_bastfn, &args);
5829 #endif
5830 	if (error) {
5831 		kfree(ua->lksb.sb_lvbptr);
5832 		ua->lksb.sb_lvbptr = NULL;
5833 		kfree(ua);
5834 		__put_lkb(ls, lkb);
5835 		goto out;
5836 	}
5837 
5838 	/* After ua is attached to lkb it will be freed by dlm_free_lkb().
5839 	   When DLM_IFL_USER is set, the dlm knows that this is a userspace
5840 	   lock and that lkb_astparam is the dlm_user_args structure. */
5841 	lkb->lkb_flags |= DLM_IFL_USER;
5842 	error = request_lock(ls, lkb, name, namelen, &args);
5843 
5844 	switch (error) {
5845 	case 0:
5846 		break;
5847 	case -EINPROGRESS:
5848 		error = 0;
5849 		break;
5850 	case -EAGAIN:
5851 		error = 0;
5852 		fallthrough;
5853 	default:
5854 		__put_lkb(ls, lkb);
5855 		goto out;
5856 	}
5857 
5858 	/* add this new lkb to the per-process list of locks */
5859 	spin_lock(&ua->proc->locks_spin);
5860 	hold_lkb(lkb);
5861 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
5862 	spin_unlock(&ua->proc->locks_spin);
5863  out:
5864 	dlm_unlock_recovery(ls);
5865 	return error;
5866 }
5867 
5868 #ifdef CONFIG_DLM_DEPRECATED_API
5869 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5870 		     int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
5871 		     unsigned long timeout_cs)
5872 #else
5873 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5874 		     int mode, uint32_t flags, uint32_t lkid, char *lvb_in)
5875 #endif
5876 {
5877 	struct dlm_lkb *lkb;
5878 	struct dlm_args args;
5879 	struct dlm_user_args *ua;
5880 	int error;
5881 
5882 	dlm_lock_recovery(ls);
5883 
5884 	error = find_lkb(ls, lkid, &lkb);
5885 	if (error)
5886 		goto out;
5887 
5888 	/* user can change the params on its lock when it converts it, or
5889 	   add an lvb that didn't exist before */
5890 
5891 	ua = lkb->lkb_ua;
5892 
5893 	if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
5894 		ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_NOFS);
5895 		if (!ua->lksb.sb_lvbptr) {
5896 			error = -ENOMEM;
5897 			goto out_put;
5898 		}
5899 	}
5900 	if (lvb_in && ua->lksb.sb_lvbptr)
5901 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
5902 
5903 	ua->xid = ua_tmp->xid;
5904 	ua->castparam = ua_tmp->castparam;
5905 	ua->castaddr = ua_tmp->castaddr;
5906 	ua->bastparam = ua_tmp->bastparam;
5907 	ua->bastaddr = ua_tmp->bastaddr;
5908 	ua->user_lksb = ua_tmp->user_lksb;
5909 
5910 #ifdef CONFIG_DLM_DEPRECATED_API
5911 	error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
5912 			      fake_astfn, ua, fake_bastfn, &args);
5913 #else
5914 	error = set_lock_args(mode, &ua->lksb, flags, 0, fake_astfn, ua,
5915 			      fake_bastfn, &args);
5916 #endif
5917 	if (error)
5918 		goto out_put;
5919 
5920 	error = convert_lock(ls, lkb, &args);
5921 
5922 	if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
5923 		error = 0;
5924  out_put:
5925 	dlm_put_lkb(lkb);
5926  out:
5927 	dlm_unlock_recovery(ls);
5928 	kfree(ua_tmp);
5929 	return error;
5930 }
5931 
5932 /*
5933  * The caller asks for an orphan lock on a given resource with a given mode.
5934  * If a matching lock exists, it's moved to the owner's list of locks and
5935  * the lkid is returned.
5936  */
5937 
5938 int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5939 		     int mode, uint32_t flags, void *name, unsigned int namelen,
5940 		     uint32_t *lkid)
5941 {
5942 	struct dlm_lkb *lkb = NULL, *iter;
5943 	struct dlm_user_args *ua;
5944 	int found_other_mode = 0;
5945 	int rv = 0;
5946 
5947 	mutex_lock(&ls->ls_orphans_mutex);
5948 	list_for_each_entry(iter, &ls->ls_orphans, lkb_ownqueue) {
5949 		if (iter->lkb_resource->res_length != namelen)
5950 			continue;
5951 		if (memcmp(iter->lkb_resource->res_name, name, namelen))
5952 			continue;
5953 		if (iter->lkb_grmode != mode) {
5954 			found_other_mode = 1;
5955 			continue;
5956 		}
5957 
5958 		lkb = iter;
5959 		list_del_init(&iter->lkb_ownqueue);
5960 		iter->lkb_flags &= ~DLM_IFL_ORPHAN;
5961 		*lkid = iter->lkb_id;
5962 		break;
5963 	}
5964 	mutex_unlock(&ls->ls_orphans_mutex);
5965 
5966 	if (!lkb && found_other_mode) {
5967 		rv = -EAGAIN;
5968 		goto out;
5969 	}
5970 
5971 	if (!lkb) {
5972 		rv = -ENOENT;
5973 		goto out;
5974 	}
5975 
5976 	lkb->lkb_exflags = flags;
5977 	lkb->lkb_ownpid = (int) current->pid;
5978 
5979 	ua = lkb->lkb_ua;
5980 
5981 	ua->proc = ua_tmp->proc;
5982 	ua->xid = ua_tmp->xid;
5983 	ua->castparam = ua_tmp->castparam;
5984 	ua->castaddr = ua_tmp->castaddr;
5985 	ua->bastparam = ua_tmp->bastparam;
5986 	ua->bastaddr = ua_tmp->bastaddr;
5987 	ua->user_lksb = ua_tmp->user_lksb;
5988 
5989 	/*
5990 	 * The lkb reference from the ls_orphans list was not
5991 	 * removed above, and is now considered the reference
5992 	 * for the proc locks list.
5993 	 */
5994 
5995 	spin_lock(&ua->proc->locks_spin);
5996 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
5997 	spin_unlock(&ua->proc->locks_spin);
5998  out:
5999 	kfree(ua_tmp);
6000 	return rv;
6001 }
6002 
6003 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
6004 		    uint32_t flags, uint32_t lkid, char *lvb_in)
6005 {
6006 	struct dlm_lkb *lkb;
6007 	struct dlm_args args;
6008 	struct dlm_user_args *ua;
6009 	int error;
6010 
6011 	dlm_lock_recovery(ls);
6012 
6013 	error = find_lkb(ls, lkid, &lkb);
6014 	if (error)
6015 		goto out;
6016 
6017 	ua = lkb->lkb_ua;
6018 
6019 	if (lvb_in && ua->lksb.sb_lvbptr)
6020 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
6021 	if (ua_tmp->castparam)
6022 		ua->castparam = ua_tmp->castparam;
6023 	ua->user_lksb = ua_tmp->user_lksb;
6024 
6025 	error = set_unlock_args(flags, ua, &args);
6026 	if (error)
6027 		goto out_put;
6028 
6029 	error = unlock_lock(ls, lkb, &args);
6030 
6031 	if (error == -DLM_EUNLOCK)
6032 		error = 0;
6033 	/* from validate_unlock_args() */
6034 	if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
6035 		error = 0;
6036 	if (error)
6037 		goto out_put;
6038 
6039 	spin_lock(&ua->proc->locks_spin);
6040 	/* dlm_user_add_cb() may have already taken lkb off the proc list */
6041 	if (!list_empty(&lkb->lkb_ownqueue))
6042 		list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
6043 	spin_unlock(&ua->proc->locks_spin);
6044  out_put:
6045 	dlm_put_lkb(lkb);
6046  out:
6047 	dlm_unlock_recovery(ls);
6048 	kfree(ua_tmp);
6049 	return error;
6050 }
6051 
6052 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
6053 		    uint32_t flags, uint32_t lkid)
6054 {
6055 	struct dlm_lkb *lkb;
6056 	struct dlm_args args;
6057 	struct dlm_user_args *ua;
6058 	int error;
6059 
6060 	dlm_lock_recovery(ls);
6061 
6062 	error = find_lkb(ls, lkid, &lkb);
6063 	if (error)
6064 		goto out;
6065 
6066 	ua = lkb->lkb_ua;
6067 	if (ua_tmp->castparam)
6068 		ua->castparam = ua_tmp->castparam;
6069 	ua->user_lksb = ua_tmp->user_lksb;
6070 
6071 	error = set_unlock_args(flags, ua, &args);
6072 	if (error)
6073 		goto out_put;
6074 
6075 	error = cancel_lock(ls, lkb, &args);
6076 
6077 	if (error == -DLM_ECANCEL)
6078 		error = 0;
6079 	/* from validate_unlock_args() */
6080 	if (error == -EBUSY)
6081 		error = 0;
6082  out_put:
6083 	dlm_put_lkb(lkb);
6084  out:
6085 	dlm_unlock_recovery(ls);
6086 	kfree(ua_tmp);
6087 	return error;
6088 }
6089 
6090 int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
6091 {
6092 	struct dlm_lkb *lkb;
6093 	struct dlm_args args;
6094 	struct dlm_user_args *ua;
6095 	struct dlm_rsb *r;
6096 	int error;
6097 
6098 	dlm_lock_recovery(ls);
6099 
6100 	error = find_lkb(ls, lkid, &lkb);
6101 	if (error)
6102 		goto out;
6103 
6104 	ua = lkb->lkb_ua;
6105 
6106 	error = set_unlock_args(flags, ua, &args);
6107 	if (error)
6108 		goto out_put;
6109 
6110 	/* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
6111 
6112 	r = lkb->lkb_resource;
6113 	hold_rsb(r);
6114 	lock_rsb(r);
6115 
6116 	error = validate_unlock_args(lkb, &args);
6117 	if (error)
6118 		goto out_r;
6119 	lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
6120 
6121 	error = _cancel_lock(r, lkb);
6122  out_r:
6123 	unlock_rsb(r);
6124 	put_rsb(r);
6125 
6126 	if (error == -DLM_ECANCEL)
6127 		error = 0;
6128 	/* from validate_unlock_args() */
6129 	if (error == -EBUSY)
6130 		error = 0;
6131  out_put:
6132 	dlm_put_lkb(lkb);
6133  out:
6134 	dlm_unlock_recovery(ls);
6135 	return error;
6136 }
6137 
6138 /* lkb's that are removed from the waiters list by revert are just left on the
6139    orphans list with the granted orphan locks, to be freed by purge */
6140 
6141 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
6142 {
6143 	struct dlm_args args;
6144 	int error;
6145 
6146 	hold_lkb(lkb); /* reference for the ls_orphans list */
6147 	mutex_lock(&ls->ls_orphans_mutex);
6148 	list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
6149 	mutex_unlock(&ls->ls_orphans_mutex);
6150 
6151 	set_unlock_args(0, lkb->lkb_ua, &args);
6152 
6153 	error = cancel_lock(ls, lkb, &args);
6154 	if (error == -DLM_ECANCEL)
6155 		error = 0;
6156 	return error;
6157 }
6158 
6159 /* The FORCEUNLOCK flag allows the unlock to go ahead even if the lkb isn't
6160    granted.  Regardless of what rsb queue the lock is on, it's removed and
6161    freed.  The IVVALBLK flag causes the lvb on the resource to be invalidated
6162    if our lock is PW/EX (it's ignored if our granted mode is smaller.) */
6163 
6164 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
6165 {
6166 	struct dlm_args args;
6167 	int error;
6168 
6169 	set_unlock_args(DLM_LKF_FORCEUNLOCK | DLM_LKF_IVVALBLK,
6170 			lkb->lkb_ua, &args);
6171 
6172 	error = unlock_lock(ls, lkb, &args);
6173 	if (error == -DLM_EUNLOCK)
6174 		error = 0;
6175 	return error;
6176 }
6177 
6178 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
6179    (which does lock_rsb) due to deadlock with receiving a message that does
6180    lock_rsb followed by dlm_user_add_cb() */
6181 
6182 static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
6183 				     struct dlm_user_proc *proc)
6184 {
6185 	struct dlm_lkb *lkb = NULL;
6186 
6187 	mutex_lock(&ls->ls_clear_proc_locks);
6188 	if (list_empty(&proc->locks))
6189 		goto out;
6190 
6191 	lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
6192 	list_del_init(&lkb->lkb_ownqueue);
6193 
6194 	if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
6195 		lkb->lkb_flags |= DLM_IFL_ORPHAN;
6196 	else
6197 		lkb->lkb_flags |= DLM_IFL_DEAD;
6198  out:
6199 	mutex_unlock(&ls->ls_clear_proc_locks);
6200 	return lkb;
6201 }
6202 
6203 /* The ls_clear_proc_locks mutex protects against dlm_user_add_cb() which
6204    1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
6205    which we clear here. */
6206 
6207 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
6208    list, and no more device_writes should add lkb's to proc->locks list; so we
6209    shouldn't need to take asts_spin or locks_spin here.  this assumes that
6210    device reads/writes/closes are serialized -- FIXME: we may need to serialize
6211    them ourself. */
6212 
6213 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
6214 {
6215 	struct dlm_lkb *lkb, *safe;
6216 
6217 	dlm_lock_recovery(ls);
6218 
6219 	while (1) {
6220 		lkb = del_proc_lock(ls, proc);
6221 		if (!lkb)
6222 			break;
6223 		del_timeout(lkb);
6224 		if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
6225 			orphan_proc_lock(ls, lkb);
6226 		else
6227 			unlock_proc_lock(ls, lkb);
6228 
6229 		/* this removes the reference for the proc->locks list
6230 		   added by dlm_user_request, it may result in the lkb
6231 		   being freed */
6232 
6233 		dlm_put_lkb(lkb);
6234 	}
6235 
6236 	mutex_lock(&ls->ls_clear_proc_locks);
6237 
6238 	/* in-progress unlocks */
6239 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
6240 		list_del_init(&lkb->lkb_ownqueue);
6241 		lkb->lkb_flags |= DLM_IFL_DEAD;
6242 		dlm_put_lkb(lkb);
6243 	}
6244 
6245 	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_cb_list) {
6246 		memset(&lkb->lkb_callbacks, 0,
6247 		       sizeof(struct dlm_callback) * DLM_CALLBACKS_SIZE);
6248 		list_del_init(&lkb->lkb_cb_list);
6249 		dlm_put_lkb(lkb);
6250 	}
6251 
6252 	mutex_unlock(&ls->ls_clear_proc_locks);
6253 	dlm_unlock_recovery(ls);
6254 }
6255 
6256 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
6257 {
6258 	struct dlm_lkb *lkb, *safe;
6259 
6260 	while (1) {
6261 		lkb = NULL;
6262 		spin_lock(&proc->locks_spin);
6263 		if (!list_empty(&proc->locks)) {
6264 			lkb = list_entry(proc->locks.next, struct dlm_lkb,
6265 					 lkb_ownqueue);
6266 			list_del_init(&lkb->lkb_ownqueue);
6267 		}
6268 		spin_unlock(&proc->locks_spin);
6269 
6270 		if (!lkb)
6271 			break;
6272 
6273 		lkb->lkb_flags |= DLM_IFL_DEAD;
6274 		unlock_proc_lock(ls, lkb);
6275 		dlm_put_lkb(lkb); /* ref from proc->locks list */
6276 	}
6277 
6278 	spin_lock(&proc->locks_spin);
6279 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
6280 		list_del_init(&lkb->lkb_ownqueue);
6281 		lkb->lkb_flags |= DLM_IFL_DEAD;
6282 		dlm_put_lkb(lkb);
6283 	}
6284 	spin_unlock(&proc->locks_spin);
6285 
6286 	spin_lock(&proc->asts_spin);
6287 	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_cb_list) {
6288 		memset(&lkb->lkb_callbacks, 0,
6289 		       sizeof(struct dlm_callback) * DLM_CALLBACKS_SIZE);
6290 		list_del_init(&lkb->lkb_cb_list);
6291 		dlm_put_lkb(lkb);
6292 	}
6293 	spin_unlock(&proc->asts_spin);
6294 }
6295 
6296 /* pid of 0 means purge all orphans */
6297 
6298 static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
6299 {
6300 	struct dlm_lkb *lkb, *safe;
6301 
6302 	mutex_lock(&ls->ls_orphans_mutex);
6303 	list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
6304 		if (pid && lkb->lkb_ownpid != pid)
6305 			continue;
6306 		unlock_proc_lock(ls, lkb);
6307 		list_del_init(&lkb->lkb_ownqueue);
6308 		dlm_put_lkb(lkb);
6309 	}
6310 	mutex_unlock(&ls->ls_orphans_mutex);
6311 }
6312 
6313 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
6314 {
6315 	struct dlm_message *ms;
6316 	struct dlm_mhandle *mh;
6317 	int error;
6318 
6319 	error = _create_message(ls, sizeof(struct dlm_message), nodeid,
6320 				DLM_MSG_PURGE, &ms, &mh);
6321 	if (error)
6322 		return error;
6323 	ms->m_nodeid = cpu_to_le32(nodeid);
6324 	ms->m_pid = cpu_to_le32(pid);
6325 
6326 	return send_message(mh, ms);
6327 }
6328 
6329 int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
6330 		   int nodeid, int pid)
6331 {
6332 	int error = 0;
6333 
6334 	if (nodeid && (nodeid != dlm_our_nodeid())) {
6335 		error = send_purge(ls, nodeid, pid);
6336 	} else {
6337 		dlm_lock_recovery(ls);
6338 		if (pid == current->pid)
6339 			purge_proc_locks(ls, proc);
6340 		else
6341 			do_purge(ls, nodeid, pid);
6342 		dlm_unlock_recovery(ls);
6343 	}
6344 	return error;
6345 }
6346 
6347 /* debug functionality */
6348 int dlm_debug_add_lkb(struct dlm_ls *ls, uint32_t lkb_id, char *name, int len,
6349 		      int lkb_nodeid, unsigned int lkb_flags, int lkb_status)
6350 {
6351 	struct dlm_lksb *lksb;
6352 	struct dlm_lkb *lkb;
6353 	struct dlm_rsb *r;
6354 	int error;
6355 
6356 	/* we currently can't set a valid user lock */
6357 	if (lkb_flags & DLM_IFL_USER)
6358 		return -EOPNOTSUPP;
6359 
6360 	lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
6361 	if (!lksb)
6362 		return -ENOMEM;
6363 
6364 	error = _create_lkb(ls, &lkb, lkb_id, lkb_id + 1);
6365 	if (error) {
6366 		kfree(lksb);
6367 		return error;
6368 	}
6369 
6370 	lkb->lkb_flags = lkb_flags;
6371 	lkb->lkb_nodeid = lkb_nodeid;
6372 	lkb->lkb_lksb = lksb;
6373 	/* user specific pointer, just don't have it NULL for kernel locks */
6374 	if (~lkb_flags & DLM_IFL_USER)
6375 		lkb->lkb_astparam = (void *)0xDEADBEEF;
6376 
6377 	error = find_rsb(ls, name, len, 0, R_REQUEST, &r);
6378 	if (error) {
6379 		kfree(lksb);
6380 		__put_lkb(ls, lkb);
6381 		return error;
6382 	}
6383 
6384 	lock_rsb(r);
6385 	attach_lkb(r, lkb);
6386 	add_lkb(r, lkb, lkb_status);
6387 	unlock_rsb(r);
6388 	put_rsb(r);
6389 
6390 	return 0;
6391 }
6392 
6393 int dlm_debug_add_lkb_to_waiters(struct dlm_ls *ls, uint32_t lkb_id,
6394 				 int mstype, int to_nodeid)
6395 {
6396 	struct dlm_lkb *lkb;
6397 	int error;
6398 
6399 	error = find_lkb(ls, lkb_id, &lkb);
6400 	if (error)
6401 		return error;
6402 
6403 	error = add_to_waiters(lkb, mstype, to_nodeid);
6404 	dlm_put_lkb(lkb);
6405 	return error;
6406 }
6407 
6408