xref: /openbmc/linux/fs/dlm/lock.c (revision 745b7908)
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, bool local);
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 toss_rsb(struct kref *kref);
93 
94 /*
95  * Lock compatibilty matrix - thanks Steve
96  * UN = Unlocked state. Not really a state, used as a flag
97  * PD = Padding. Used to make the matrix a nice power of two in size
98  * Other states are the same as the VMS DLM.
99  * Usage: matrix[grmode+1][rqmode+1]  (although m[rq+1][gr+1] is the same)
100  */
101 
102 static const int __dlm_compat_matrix[8][8] = {
103       /* UN NL CR CW PR PW EX PD */
104         {1, 1, 1, 1, 1, 1, 1, 0},       /* UN */
105         {1, 1, 1, 1, 1, 1, 1, 0},       /* NL */
106         {1, 1, 1, 1, 1, 1, 0, 0},       /* CR */
107         {1, 1, 1, 1, 0, 0, 0, 0},       /* CW */
108         {1, 1, 1, 0, 1, 0, 0, 0},       /* PR */
109         {1, 1, 1, 0, 0, 0, 0, 0},       /* PW */
110         {1, 1, 0, 0, 0, 0, 0, 0},       /* EX */
111         {0, 0, 0, 0, 0, 0, 0, 0}        /* PD */
112 };
113 
114 /*
115  * This defines the direction of transfer of LVB data.
116  * Granted mode is the row; requested mode is the column.
117  * Usage: matrix[grmode+1][rqmode+1]
118  * 1 = LVB is returned to the caller
119  * 0 = LVB is written to the resource
120  * -1 = nothing happens to the LVB
121  */
122 
123 const int dlm_lvb_operations[8][8] = {
124         /* UN   NL  CR  CW  PR  PW  EX  PD*/
125         {  -1,  1,  1,  1,  1,  1,  1, -1 }, /* UN */
126         {  -1,  1,  1,  1,  1,  1,  1,  0 }, /* NL */
127         {  -1, -1,  1,  1,  1,  1,  1,  0 }, /* CR */
128         {  -1, -1, -1,  1,  1,  1,  1,  0 }, /* CW */
129         {  -1, -1, -1, -1,  1,  1,  1,  0 }, /* PR */
130         {  -1,  0,  0,  0,  0,  0,  1,  0 }, /* PW */
131         {  -1,  0,  0,  0,  0,  0,  0,  0 }, /* EX */
132         {  -1,  0,  0,  0,  0,  0,  0,  0 }  /* PD */
133 };
134 
135 #define modes_compat(gr, rq) \
136 	__dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
137 
138 int dlm_modes_compat(int mode1, int mode2)
139 {
140 	return __dlm_compat_matrix[mode1 + 1][mode2 + 1];
141 }
142 
143 /*
144  * Compatibility matrix for conversions with QUECVT set.
145  * Granted mode is the row; requested mode is the column.
146  * Usage: matrix[grmode+1][rqmode+1]
147  */
148 
149 static const int __quecvt_compat_matrix[8][8] = {
150       /* UN NL CR CW PR PW EX PD */
151         {0, 0, 0, 0, 0, 0, 0, 0},       /* UN */
152         {0, 0, 1, 1, 1, 1, 1, 0},       /* NL */
153         {0, 0, 0, 1, 1, 1, 1, 0},       /* CR */
154         {0, 0, 0, 0, 1, 1, 1, 0},       /* CW */
155         {0, 0, 0, 1, 0, 1, 1, 0},       /* PR */
156         {0, 0, 0, 0, 0, 0, 1, 0},       /* PW */
157         {0, 0, 0, 0, 0, 0, 0, 0},       /* EX */
158         {0, 0, 0, 0, 0, 0, 0, 0}        /* PD */
159 };
160 
161 void dlm_print_lkb(struct dlm_lkb *lkb)
162 {
163 	printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x "
164 	       "sts %d rq %d gr %d wait_type %d wait_nodeid %d seq %llu\n",
165 	       lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags,
166 	       dlm_iflags_val(lkb), lkb->lkb_status, lkb->lkb_rqmode,
167 	       lkb->lkb_grmode, lkb->lkb_wait_type, lkb->lkb_wait_nodeid,
168 	       (unsigned long long)lkb->lkb_recover_seq);
169 }
170 
171 static void dlm_print_rsb(struct dlm_rsb *r)
172 {
173 	printk(KERN_ERR "rsb: nodeid %d master %d dir %d flags %lx first %x "
174 	       "rlc %d name %s\n",
175 	       r->res_nodeid, r->res_master_nodeid, r->res_dir_nodeid,
176 	       r->res_flags, r->res_first_lkid, r->res_recover_locks_count,
177 	       r->res_name);
178 }
179 
180 void dlm_dump_rsb(struct dlm_rsb *r)
181 {
182 	struct dlm_lkb *lkb;
183 
184 	dlm_print_rsb(r);
185 
186 	printk(KERN_ERR "rsb: root_list empty %d recover_list empty %d\n",
187 	       list_empty(&r->res_root_list), list_empty(&r->res_recover_list));
188 	printk(KERN_ERR "rsb lookup list\n");
189 	list_for_each_entry(lkb, &r->res_lookup, lkb_rsb_lookup)
190 		dlm_print_lkb(lkb);
191 	printk(KERN_ERR "rsb grant queue:\n");
192 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue)
193 		dlm_print_lkb(lkb);
194 	printk(KERN_ERR "rsb convert queue:\n");
195 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue)
196 		dlm_print_lkb(lkb);
197 	printk(KERN_ERR "rsb wait queue:\n");
198 	list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue)
199 		dlm_print_lkb(lkb);
200 }
201 
202 /* Threads cannot use the lockspace while it's being recovered */
203 
204 static inline void dlm_lock_recovery(struct dlm_ls *ls)
205 {
206 	down_read(&ls->ls_in_recovery);
207 }
208 
209 void dlm_unlock_recovery(struct dlm_ls *ls)
210 {
211 	up_read(&ls->ls_in_recovery);
212 }
213 
214 int dlm_lock_recovery_try(struct dlm_ls *ls)
215 {
216 	return down_read_trylock(&ls->ls_in_recovery);
217 }
218 
219 static inline int can_be_queued(struct dlm_lkb *lkb)
220 {
221 	return !(lkb->lkb_exflags & DLM_LKF_NOQUEUE);
222 }
223 
224 static inline int force_blocking_asts(struct dlm_lkb *lkb)
225 {
226 	return (lkb->lkb_exflags & DLM_LKF_NOQUEUEBAST);
227 }
228 
229 static inline int is_demoted(struct dlm_lkb *lkb)
230 {
231 	return test_bit(DLM_SBF_DEMOTED_BIT, &lkb->lkb_sbflags);
232 }
233 
234 static inline int is_altmode(struct dlm_lkb *lkb)
235 {
236 	return test_bit(DLM_SBF_ALTMODE_BIT, &lkb->lkb_sbflags);
237 }
238 
239 static inline int is_granted(struct dlm_lkb *lkb)
240 {
241 	return (lkb->lkb_status == DLM_LKSTS_GRANTED);
242 }
243 
244 static inline int is_remote(struct dlm_rsb *r)
245 {
246 	DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r););
247 	return !!r->res_nodeid;
248 }
249 
250 static inline int is_process_copy(struct dlm_lkb *lkb)
251 {
252 	return lkb->lkb_nodeid &&
253 	       !test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags);
254 }
255 
256 static inline int is_master_copy(struct dlm_lkb *lkb)
257 {
258 	return test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags);
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 test_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags);
277 }
278 
279 static inline int is_overlap_cancel(struct dlm_lkb *lkb)
280 {
281 	return test_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
282 }
283 
284 static inline int is_overlap(struct dlm_lkb *lkb)
285 {
286 	return test_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags) ||
287 	       test_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
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 	DLM_ASSERT(lkb->lkb_lksb, dlm_print_lkb(lkb););
296 
297 	if (rv == -DLM_ECANCEL &&
298 	    test_and_clear_bit(DLM_IFL_DEADLOCK_CANCEL_BIT, &lkb->lkb_iflags))
299 		rv = -EDEADLK;
300 
301 	dlm_add_cb(lkb, DLM_CB_CAST, lkb->lkb_grmode, rv, dlm_sbflags_val(lkb));
302 }
303 
304 static inline void queue_cast_overlap(struct dlm_rsb *r, struct dlm_lkb *lkb)
305 {
306 	queue_cast(r, lkb,
307 		   is_overlap_unlock(lkb) ? -DLM_EUNLOCK : -DLM_ECANCEL);
308 }
309 
310 static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
311 {
312 	if (is_master_copy(lkb)) {
313 		send_bast(r, lkb, rqmode);
314 	} else {
315 		dlm_add_cb(lkb, DLM_CB_BAST, rqmode, 0, 0);
316 	}
317 }
318 
319 /*
320  * Basic operations on rsb's and lkb's
321  */
322 
323 /* This is only called to add a reference when the code already holds
324    a valid reference to the rsb, so there's no need for locking. */
325 
326 static inline void hold_rsb(struct dlm_rsb *r)
327 {
328 	kref_get(&r->res_ref);
329 }
330 
331 void dlm_hold_rsb(struct dlm_rsb *r)
332 {
333 	hold_rsb(r);
334 }
335 
336 /* When all references to the rsb are gone it's transferred to
337    the tossed list for later disposal. */
338 
339 static void put_rsb(struct dlm_rsb *r)
340 {
341 	struct dlm_ls *ls = r->res_ls;
342 	uint32_t bucket = r->res_bucket;
343 	int rv;
344 
345 	rv = kref_put_lock(&r->res_ref, toss_rsb,
346 			   &ls->ls_rsbtbl[bucket].lock);
347 	if (rv)
348 		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
349 }
350 
351 void dlm_put_rsb(struct dlm_rsb *r)
352 {
353 	put_rsb(r);
354 }
355 
356 static int pre_rsb_struct(struct dlm_ls *ls)
357 {
358 	struct dlm_rsb *r1, *r2;
359 	int count = 0;
360 
361 	spin_lock(&ls->ls_new_rsb_spin);
362 	if (ls->ls_new_rsb_count > dlm_config.ci_new_rsb_count / 2) {
363 		spin_unlock(&ls->ls_new_rsb_spin);
364 		return 0;
365 	}
366 	spin_unlock(&ls->ls_new_rsb_spin);
367 
368 	r1 = dlm_allocate_rsb(ls);
369 	r2 = dlm_allocate_rsb(ls);
370 
371 	spin_lock(&ls->ls_new_rsb_spin);
372 	if (r1) {
373 		list_add(&r1->res_hashchain, &ls->ls_new_rsb);
374 		ls->ls_new_rsb_count++;
375 	}
376 	if (r2) {
377 		list_add(&r2->res_hashchain, &ls->ls_new_rsb);
378 		ls->ls_new_rsb_count++;
379 	}
380 	count = ls->ls_new_rsb_count;
381 	spin_unlock(&ls->ls_new_rsb_spin);
382 
383 	if (!count)
384 		return -ENOMEM;
385 	return 0;
386 }
387 
388 /* If ls->ls_new_rsb is empty, return -EAGAIN, so the caller can
389    unlock any spinlocks, go back and call pre_rsb_struct again.
390    Otherwise, take an rsb off the list and return it. */
391 
392 static int get_rsb_struct(struct dlm_ls *ls, const void *name, int len,
393 			  struct dlm_rsb **r_ret)
394 {
395 	struct dlm_rsb *r;
396 	int count;
397 
398 	spin_lock(&ls->ls_new_rsb_spin);
399 	if (list_empty(&ls->ls_new_rsb)) {
400 		count = ls->ls_new_rsb_count;
401 		spin_unlock(&ls->ls_new_rsb_spin);
402 		log_debug(ls, "find_rsb retry %d %d %s",
403 			  count, dlm_config.ci_new_rsb_count,
404 			  (const char *)name);
405 		return -EAGAIN;
406 	}
407 
408 	r = list_first_entry(&ls->ls_new_rsb, struct dlm_rsb, res_hashchain);
409 	list_del(&r->res_hashchain);
410 	/* Convert the empty list_head to a NULL rb_node for tree usage: */
411 	memset(&r->res_hashnode, 0, sizeof(struct rb_node));
412 	ls->ls_new_rsb_count--;
413 	spin_unlock(&ls->ls_new_rsb_spin);
414 
415 	r->res_ls = ls;
416 	r->res_length = len;
417 	memcpy(r->res_name, name, len);
418 	mutex_init(&r->res_mutex);
419 
420 	INIT_LIST_HEAD(&r->res_lookup);
421 	INIT_LIST_HEAD(&r->res_grantqueue);
422 	INIT_LIST_HEAD(&r->res_convertqueue);
423 	INIT_LIST_HEAD(&r->res_waitqueue);
424 	INIT_LIST_HEAD(&r->res_root_list);
425 	INIT_LIST_HEAD(&r->res_recover_list);
426 
427 	*r_ret = r;
428 	return 0;
429 }
430 
431 static int rsb_cmp(struct dlm_rsb *r, const char *name, int nlen)
432 {
433 	char maxname[DLM_RESNAME_MAXLEN];
434 
435 	memset(maxname, 0, DLM_RESNAME_MAXLEN);
436 	memcpy(maxname, name, nlen);
437 	return memcmp(r->res_name, maxname, DLM_RESNAME_MAXLEN);
438 }
439 
440 int dlm_search_rsb_tree(struct rb_root *tree, const void *name, int len,
441 			struct dlm_rsb **r_ret)
442 {
443 	struct rb_node *node = tree->rb_node;
444 	struct dlm_rsb *r;
445 	int rc;
446 
447 	while (node) {
448 		r = rb_entry(node, struct dlm_rsb, res_hashnode);
449 		rc = rsb_cmp(r, name, len);
450 		if (rc < 0)
451 			node = node->rb_left;
452 		else if (rc > 0)
453 			node = node->rb_right;
454 		else
455 			goto found;
456 	}
457 	*r_ret = NULL;
458 	return -EBADR;
459 
460  found:
461 	*r_ret = r;
462 	return 0;
463 }
464 
465 static int rsb_insert(struct dlm_rsb *rsb, struct rb_root *tree)
466 {
467 	struct rb_node **newn = &tree->rb_node;
468 	struct rb_node *parent = NULL;
469 	int rc;
470 
471 	while (*newn) {
472 		struct dlm_rsb *cur = rb_entry(*newn, struct dlm_rsb,
473 					       res_hashnode);
474 
475 		parent = *newn;
476 		rc = rsb_cmp(cur, rsb->res_name, rsb->res_length);
477 		if (rc < 0)
478 			newn = &parent->rb_left;
479 		else if (rc > 0)
480 			newn = &parent->rb_right;
481 		else {
482 			log_print("rsb_insert match");
483 			dlm_dump_rsb(rsb);
484 			dlm_dump_rsb(cur);
485 			return -EEXIST;
486 		}
487 	}
488 
489 	rb_link_node(&rsb->res_hashnode, parent, newn);
490 	rb_insert_color(&rsb->res_hashnode, tree);
491 	return 0;
492 }
493 
494 /*
495  * Find rsb in rsbtbl and potentially create/add one
496  *
497  * Delaying the release of rsb's has a similar benefit to applications keeping
498  * NL locks on an rsb, but without the guarantee that the cached master value
499  * will still be valid when the rsb is reused.  Apps aren't always smart enough
500  * to keep NL locks on an rsb that they may lock again shortly; this can lead
501  * to excessive master lookups and removals if we don't delay the release.
502  *
503  * Searching for an rsb means looking through both the normal list and toss
504  * list.  When found on the toss list the rsb is moved to the normal list with
505  * ref count of 1; when found on normal list the ref count is incremented.
506  *
507  * rsb's on the keep list are being used locally and refcounted.
508  * rsb's on the toss list are not being used locally, and are not refcounted.
509  *
510  * The toss list rsb's were either
511  * - previously used locally but not any more (were on keep list, then
512  *   moved to toss list when last refcount dropped)
513  * - created and put on toss list as a directory record for a lookup
514  *   (we are the dir node for the res, but are not using the res right now,
515  *   but some other node is)
516  *
517  * The purpose of find_rsb() is to return a refcounted rsb for local use.
518  * So, if the given rsb is on the toss list, it is moved to the keep list
519  * before being returned.
520  *
521  * toss_rsb() happens when all local usage of the rsb is done, i.e. no
522  * more refcounts exist, so the rsb is moved from the keep list to the
523  * toss list.
524  *
525  * rsb's on both keep and toss lists are used for doing a name to master
526  * lookups.  rsb's that are in use locally (and being refcounted) are on
527  * the keep list, rsb's that are not in use locally (not refcounted) and
528  * only exist for name/master lookups are on the toss list.
529  *
530  * rsb's on the toss list who's dir_nodeid is not local can have stale
531  * name/master mappings.  So, remote requests on such rsb's can potentially
532  * return with an error, which means the mapping is stale and needs to
533  * be updated with a new lookup.  (The idea behind MASTER UNCERTAIN and
534  * first_lkid is to keep only a single outstanding request on an rsb
535  * while that rsb has a potentially stale master.)
536  */
537 
538 static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
539 			uint32_t hash, uint32_t b,
540 			int dir_nodeid, int from_nodeid,
541 			unsigned int flags, struct dlm_rsb **r_ret)
542 {
543 	struct dlm_rsb *r = NULL;
544 	int our_nodeid = dlm_our_nodeid();
545 	int from_local = 0;
546 	int from_other = 0;
547 	int from_dir = 0;
548 	int create = 0;
549 	int error;
550 
551 	if (flags & R_RECEIVE_REQUEST) {
552 		if (from_nodeid == dir_nodeid)
553 			from_dir = 1;
554 		else
555 			from_other = 1;
556 	} else if (flags & R_REQUEST) {
557 		from_local = 1;
558 	}
559 
560 	/*
561 	 * flags & R_RECEIVE_RECOVER is from dlm_recover_master_copy, so
562 	 * from_nodeid has sent us a lock in dlm_recover_locks, believing
563 	 * we're the new master.  Our local recovery may not have set
564 	 * res_master_nodeid to our_nodeid yet, so allow either.  Don't
565 	 * create the rsb; dlm_recover_process_copy() will handle EBADR
566 	 * by resending.
567 	 *
568 	 * If someone sends us a request, we are the dir node, and we do
569 	 * not find the rsb anywhere, then recreate it.  This happens if
570 	 * someone sends us a request after we have removed/freed an rsb
571 	 * from our toss list.  (They sent a request instead of lookup
572 	 * because they are using an rsb from their toss list.)
573 	 */
574 
575 	if (from_local || from_dir ||
576 	    (from_other && (dir_nodeid == our_nodeid))) {
577 		create = 1;
578 	}
579 
580  retry:
581 	if (create) {
582 		error = pre_rsb_struct(ls);
583 		if (error < 0)
584 			goto out;
585 	}
586 
587 	spin_lock(&ls->ls_rsbtbl[b].lock);
588 
589 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
590 	if (error)
591 		goto do_toss;
592 
593 	/*
594 	 * rsb is active, so we can't check master_nodeid without lock_rsb.
595 	 */
596 
597 	kref_get(&r->res_ref);
598 	goto out_unlock;
599 
600 
601  do_toss:
602 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
603 	if (error)
604 		goto do_new;
605 
606 	/*
607 	 * rsb found inactive (master_nodeid may be out of date unless
608 	 * we are the dir_nodeid or were the master)  No other thread
609 	 * is using this rsb because it's on the toss list, so we can
610 	 * look at or update res_master_nodeid without lock_rsb.
611 	 */
612 
613 	if ((r->res_master_nodeid != our_nodeid) && from_other) {
614 		/* our rsb was not master, and another node (not the dir node)
615 		   has sent us a request */
616 		log_debug(ls, "find_rsb toss from_other %d master %d dir %d %s",
617 			  from_nodeid, r->res_master_nodeid, dir_nodeid,
618 			  r->res_name);
619 		error = -ENOTBLK;
620 		goto out_unlock;
621 	}
622 
623 	if ((r->res_master_nodeid != our_nodeid) && from_dir) {
624 		/* don't think this should ever happen */
625 		log_error(ls, "find_rsb toss from_dir %d master %d",
626 			  from_nodeid, r->res_master_nodeid);
627 		dlm_print_rsb(r);
628 		/* fix it and go on */
629 		r->res_master_nodeid = our_nodeid;
630 		r->res_nodeid = 0;
631 		rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
632 		r->res_first_lkid = 0;
633 	}
634 
635 	if (from_local && (r->res_master_nodeid != our_nodeid)) {
636 		/* Because we have held no locks on this rsb,
637 		   res_master_nodeid could have become stale. */
638 		rsb_set_flag(r, RSB_MASTER_UNCERTAIN);
639 		r->res_first_lkid = 0;
640 	}
641 
642 	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
643 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
644 	goto out_unlock;
645 
646 
647  do_new:
648 	/*
649 	 * rsb not found
650 	 */
651 
652 	if (error == -EBADR && !create)
653 		goto out_unlock;
654 
655 	error = get_rsb_struct(ls, name, len, &r);
656 	if (error == -EAGAIN) {
657 		spin_unlock(&ls->ls_rsbtbl[b].lock);
658 		goto retry;
659 	}
660 	if (error)
661 		goto out_unlock;
662 
663 	r->res_hash = hash;
664 	r->res_bucket = b;
665 	r->res_dir_nodeid = dir_nodeid;
666 	kref_init(&r->res_ref);
667 
668 	if (from_dir) {
669 		/* want to see how often this happens */
670 		log_debug(ls, "find_rsb new from_dir %d recreate %s",
671 			  from_nodeid, r->res_name);
672 		r->res_master_nodeid = our_nodeid;
673 		r->res_nodeid = 0;
674 		goto out_add;
675 	}
676 
677 	if (from_other && (dir_nodeid != our_nodeid)) {
678 		/* should never happen */
679 		log_error(ls, "find_rsb new from_other %d dir %d our %d %s",
680 			  from_nodeid, dir_nodeid, our_nodeid, r->res_name);
681 		dlm_free_rsb(r);
682 		r = NULL;
683 		error = -ENOTBLK;
684 		goto out_unlock;
685 	}
686 
687 	if (from_other) {
688 		log_debug(ls, "find_rsb new from_other %d dir %d %s",
689 			  from_nodeid, dir_nodeid, r->res_name);
690 	}
691 
692 	if (dir_nodeid == our_nodeid) {
693 		/* When we are the dir nodeid, we can set the master
694 		   node immediately */
695 		r->res_master_nodeid = our_nodeid;
696 		r->res_nodeid = 0;
697 	} else {
698 		/* set_master will send_lookup to dir_nodeid */
699 		r->res_master_nodeid = 0;
700 		r->res_nodeid = -1;
701 	}
702 
703  out_add:
704 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
705  out_unlock:
706 	spin_unlock(&ls->ls_rsbtbl[b].lock);
707  out:
708 	*r_ret = r;
709 	return error;
710 }
711 
712 /* During recovery, other nodes can send us new MSTCPY locks (from
713    dlm_recover_locks) before we've made ourself master (in
714    dlm_recover_masters). */
715 
716 static int find_rsb_nodir(struct dlm_ls *ls, const void *name, int len,
717 			  uint32_t hash, uint32_t b,
718 			  int dir_nodeid, int from_nodeid,
719 			  unsigned int flags, struct dlm_rsb **r_ret)
720 {
721 	struct dlm_rsb *r = NULL;
722 	int our_nodeid = dlm_our_nodeid();
723 	int recover = (flags & R_RECEIVE_RECOVER);
724 	int error;
725 
726  retry:
727 	error = pre_rsb_struct(ls);
728 	if (error < 0)
729 		goto out;
730 
731 	spin_lock(&ls->ls_rsbtbl[b].lock);
732 
733 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
734 	if (error)
735 		goto do_toss;
736 
737 	/*
738 	 * rsb is active, so we can't check master_nodeid without lock_rsb.
739 	 */
740 
741 	kref_get(&r->res_ref);
742 	goto out_unlock;
743 
744 
745  do_toss:
746 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
747 	if (error)
748 		goto do_new;
749 
750 	/*
751 	 * rsb found inactive. No other thread is using this rsb because
752 	 * it's on the toss list, so we can look at or update
753 	 * res_master_nodeid without lock_rsb.
754 	 */
755 
756 	if (!recover && (r->res_master_nodeid != our_nodeid) && from_nodeid) {
757 		/* our rsb is not master, and another node has sent us a
758 		   request; this should never happen */
759 		log_error(ls, "find_rsb toss from_nodeid %d master %d dir %d",
760 			  from_nodeid, r->res_master_nodeid, dir_nodeid);
761 		dlm_print_rsb(r);
762 		error = -ENOTBLK;
763 		goto out_unlock;
764 	}
765 
766 	if (!recover && (r->res_master_nodeid != our_nodeid) &&
767 	    (dir_nodeid == our_nodeid)) {
768 		/* our rsb is not master, and we are dir; may as well fix it;
769 		   this should never happen */
770 		log_error(ls, "find_rsb toss our %d master %d dir %d",
771 			  our_nodeid, r->res_master_nodeid, dir_nodeid);
772 		dlm_print_rsb(r);
773 		r->res_master_nodeid = our_nodeid;
774 		r->res_nodeid = 0;
775 	}
776 
777 	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
778 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
779 	goto out_unlock;
780 
781 
782  do_new:
783 	/*
784 	 * rsb not found
785 	 */
786 
787 	error = get_rsb_struct(ls, name, len, &r);
788 	if (error == -EAGAIN) {
789 		spin_unlock(&ls->ls_rsbtbl[b].lock);
790 		goto retry;
791 	}
792 	if (error)
793 		goto out_unlock;
794 
795 	r->res_hash = hash;
796 	r->res_bucket = b;
797 	r->res_dir_nodeid = dir_nodeid;
798 	r->res_master_nodeid = dir_nodeid;
799 	r->res_nodeid = (dir_nodeid == our_nodeid) ? 0 : dir_nodeid;
800 	kref_init(&r->res_ref);
801 
802 	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
803  out_unlock:
804 	spin_unlock(&ls->ls_rsbtbl[b].lock);
805  out:
806 	*r_ret = r;
807 	return error;
808 }
809 
810 static int find_rsb(struct dlm_ls *ls, const void *name, int len,
811 		    int from_nodeid, unsigned int flags,
812 		    struct dlm_rsb **r_ret)
813 {
814 	uint32_t hash, b;
815 	int dir_nodeid;
816 
817 	if (len > DLM_RESNAME_MAXLEN)
818 		return -EINVAL;
819 
820 	hash = jhash(name, len, 0);
821 	b = hash & (ls->ls_rsbtbl_size - 1);
822 
823 	dir_nodeid = dlm_hash2nodeid(ls, hash);
824 
825 	if (dlm_no_directory(ls))
826 		return find_rsb_nodir(ls, name, len, hash, b, dir_nodeid,
827 				      from_nodeid, flags, r_ret);
828 	else
829 		return find_rsb_dir(ls, name, len, hash, b, dir_nodeid,
830 				      from_nodeid, flags, r_ret);
831 }
832 
833 /* we have received a request and found that res_master_nodeid != our_nodeid,
834    so we need to return an error or make ourself the master */
835 
836 static int validate_master_nodeid(struct dlm_ls *ls, struct dlm_rsb *r,
837 				  int from_nodeid)
838 {
839 	if (dlm_no_directory(ls)) {
840 		log_error(ls, "find_rsb keep from_nodeid %d master %d dir %d",
841 			  from_nodeid, r->res_master_nodeid,
842 			  r->res_dir_nodeid);
843 		dlm_print_rsb(r);
844 		return -ENOTBLK;
845 	}
846 
847 	if (from_nodeid != r->res_dir_nodeid) {
848 		/* our rsb is not master, and another node (not the dir node)
849 	   	   has sent us a request.  this is much more common when our
850 	   	   master_nodeid is zero, so limit debug to non-zero.  */
851 
852 		if (r->res_master_nodeid) {
853 			log_debug(ls, "validate master from_other %d master %d "
854 				  "dir %d first %x %s", from_nodeid,
855 				  r->res_master_nodeid, r->res_dir_nodeid,
856 				  r->res_first_lkid, r->res_name);
857 		}
858 		return -ENOTBLK;
859 	} else {
860 		/* our rsb is not master, but the dir nodeid has sent us a
861 	   	   request; this could happen with master 0 / res_nodeid -1 */
862 
863 		if (r->res_master_nodeid) {
864 			log_error(ls, "validate master from_dir %d master %d "
865 				  "first %x %s",
866 				  from_nodeid, r->res_master_nodeid,
867 				  r->res_first_lkid, r->res_name);
868 		}
869 
870 		r->res_master_nodeid = dlm_our_nodeid();
871 		r->res_nodeid = 0;
872 		return 0;
873 	}
874 }
875 
876 static void __dlm_master_lookup(struct dlm_ls *ls, struct dlm_rsb *r, int our_nodeid,
877 				int from_nodeid, bool toss_list, unsigned int flags,
878 				int *r_nodeid, int *result)
879 {
880 	int fix_master = (flags & DLM_LU_RECOVER_MASTER);
881 	int from_master = (flags & DLM_LU_RECOVER_DIR);
882 
883 	if (r->res_dir_nodeid != our_nodeid) {
884 		/* should not happen, but may as well fix it and carry on */
885 		log_error(ls, "%s res_dir %d our %d %s", __func__,
886 			  r->res_dir_nodeid, our_nodeid, r->res_name);
887 		r->res_dir_nodeid = our_nodeid;
888 	}
889 
890 	if (fix_master && dlm_is_removed(ls, r->res_master_nodeid)) {
891 		/* Recovery uses this function to set a new master when
892 		 * the previous master failed.  Setting NEW_MASTER will
893 		 * force dlm_recover_masters to call recover_master on this
894 		 * rsb even though the res_nodeid is no longer removed.
895 		 */
896 
897 		r->res_master_nodeid = from_nodeid;
898 		r->res_nodeid = from_nodeid;
899 		rsb_set_flag(r, RSB_NEW_MASTER);
900 
901 		if (toss_list) {
902 			/* I don't think we should ever find it on toss list. */
903 			log_error(ls, "%s fix_master on toss", __func__);
904 			dlm_dump_rsb(r);
905 		}
906 	}
907 
908 	if (from_master && (r->res_master_nodeid != from_nodeid)) {
909 		/* this will happen if from_nodeid became master during
910 		 * a previous recovery cycle, and we aborted the previous
911 		 * cycle before recovering this master value
912 		 */
913 
914 		log_limit(ls, "%s from_master %d master_nodeid %d res_nodeid %d first %x %s",
915 			  __func__, from_nodeid, r->res_master_nodeid,
916 			  r->res_nodeid, r->res_first_lkid, r->res_name);
917 
918 		if (r->res_master_nodeid == our_nodeid) {
919 			log_error(ls, "from_master %d our_master", from_nodeid);
920 			dlm_dump_rsb(r);
921 			goto ret_assign;
922 		}
923 
924 		r->res_master_nodeid = from_nodeid;
925 		r->res_nodeid = from_nodeid;
926 		rsb_set_flag(r, RSB_NEW_MASTER);
927 	}
928 
929 	if (!r->res_master_nodeid) {
930 		/* this will happen if recovery happens while we're looking
931 		 * up the master for this rsb
932 		 */
933 
934 		log_debug(ls, "%s master 0 to %d first %x %s", __func__,
935 			  from_nodeid, r->res_first_lkid, r->res_name);
936 		r->res_master_nodeid = from_nodeid;
937 		r->res_nodeid = from_nodeid;
938 	}
939 
940 	if (!from_master && !fix_master &&
941 	    (r->res_master_nodeid == from_nodeid)) {
942 		/* this can happen when the master sends remove, the dir node
943 		 * finds the rsb on the keep list and ignores the remove,
944 		 * and the former master sends a lookup
945 		 */
946 
947 		log_limit(ls, "%s from master %d flags %x first %x %s",
948 			  __func__, from_nodeid, flags, r->res_first_lkid,
949 			  r->res_name);
950 	}
951 
952  ret_assign:
953 	*r_nodeid = r->res_master_nodeid;
954 	if (result)
955 		*result = DLM_LU_MATCH;
956 }
957 
958 /*
959  * We're the dir node for this res and another node wants to know the
960  * master nodeid.  During normal operation (non recovery) this is only
961  * called from receive_lookup(); master lookups when the local node is
962  * the dir node are done by find_rsb().
963  *
964  * normal operation, we are the dir node for a resource
965  * . _request_lock
966  * . set_master
967  * . send_lookup
968  * . receive_lookup
969  * . dlm_master_lookup flags 0
970  *
971  * recover directory, we are rebuilding dir for all resources
972  * . dlm_recover_directory
973  * . dlm_rcom_names
974  *   remote node sends back the rsb names it is master of and we are dir of
975  * . dlm_master_lookup RECOVER_DIR (fix_master 0, from_master 1)
976  *   we either create new rsb setting remote node as master, or find existing
977  *   rsb and set master to be the remote node.
978  *
979  * recover masters, we are finding the new master for resources
980  * . dlm_recover_masters
981  * . recover_master
982  * . dlm_send_rcom_lookup
983  * . receive_rcom_lookup
984  * . dlm_master_lookup RECOVER_MASTER (fix_master 1, from_master 0)
985  */
986 
987 int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, char *name, int len,
988 		      unsigned int flags, int *r_nodeid, int *result)
989 {
990 	struct dlm_rsb *r = NULL;
991 	uint32_t hash, b;
992 	int our_nodeid = dlm_our_nodeid();
993 	int dir_nodeid, error;
994 
995 	if (len > DLM_RESNAME_MAXLEN)
996 		return -EINVAL;
997 
998 	if (from_nodeid == our_nodeid) {
999 		log_error(ls, "dlm_master_lookup from our_nodeid %d flags %x",
1000 			  our_nodeid, flags);
1001 		return -EINVAL;
1002 	}
1003 
1004 	hash = jhash(name, len, 0);
1005 	b = hash & (ls->ls_rsbtbl_size - 1);
1006 
1007 	dir_nodeid = dlm_hash2nodeid(ls, hash);
1008 	if (dir_nodeid != our_nodeid) {
1009 		log_error(ls, "dlm_master_lookup from %d dir %d our %d h %x %d",
1010 			  from_nodeid, dir_nodeid, our_nodeid, hash,
1011 			  ls->ls_num_nodes);
1012 		*r_nodeid = -1;
1013 		return -EINVAL;
1014 	}
1015 
1016  retry:
1017 	error = pre_rsb_struct(ls);
1018 	if (error < 0)
1019 		return error;
1020 
1021 	spin_lock(&ls->ls_rsbtbl[b].lock);
1022 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
1023 	if (!error) {
1024 		/* because the rsb is active, we need to lock_rsb before
1025 		 * checking/changing re_master_nodeid
1026 		 */
1027 
1028 		hold_rsb(r);
1029 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1030 		lock_rsb(r);
1031 
1032 		__dlm_master_lookup(ls, r, our_nodeid, from_nodeid, false,
1033 				    flags, r_nodeid, result);
1034 
1035 		/* the rsb was active */
1036 		unlock_rsb(r);
1037 		put_rsb(r);
1038 
1039 		return 0;
1040 	}
1041 
1042 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
1043 	if (error)
1044 		goto not_found;
1045 
1046 	/* because the rsb is inactive (on toss list), it's not refcounted
1047 	 * and lock_rsb is not used, but is protected by the rsbtbl lock
1048 	 */
1049 
1050 	__dlm_master_lookup(ls, r, our_nodeid, from_nodeid, true, flags,
1051 			    r_nodeid, result);
1052 
1053 	r->res_toss_time = jiffies;
1054 	/* the rsb was inactive (on toss list) */
1055 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1056 
1057 	return 0;
1058 
1059  not_found:
1060 	error = get_rsb_struct(ls, name, len, &r);
1061 	if (error == -EAGAIN) {
1062 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1063 		goto retry;
1064 	}
1065 	if (error)
1066 		goto out_unlock;
1067 
1068 	r->res_hash = hash;
1069 	r->res_bucket = b;
1070 	r->res_dir_nodeid = our_nodeid;
1071 	r->res_master_nodeid = from_nodeid;
1072 	r->res_nodeid = from_nodeid;
1073 	kref_init(&r->res_ref);
1074 	r->res_toss_time = jiffies;
1075 
1076 	error = rsb_insert(r, &ls->ls_rsbtbl[b].toss);
1077 	if (error) {
1078 		/* should never happen */
1079 		dlm_free_rsb(r);
1080 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1081 		goto retry;
1082 	}
1083 
1084 	if (result)
1085 		*result = DLM_LU_ADD;
1086 	*r_nodeid = from_nodeid;
1087  out_unlock:
1088 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1089 	return error;
1090 }
1091 
1092 static void dlm_dump_rsb_hash(struct dlm_ls *ls, uint32_t hash)
1093 {
1094 	struct rb_node *n;
1095 	struct dlm_rsb *r;
1096 	int i;
1097 
1098 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1099 		spin_lock(&ls->ls_rsbtbl[i].lock);
1100 		for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
1101 			r = rb_entry(n, struct dlm_rsb, res_hashnode);
1102 			if (r->res_hash == hash)
1103 				dlm_dump_rsb(r);
1104 		}
1105 		spin_unlock(&ls->ls_rsbtbl[i].lock);
1106 	}
1107 }
1108 
1109 void dlm_dump_rsb_name(struct dlm_ls *ls, char *name, int len)
1110 {
1111 	struct dlm_rsb *r = NULL;
1112 	uint32_t hash, b;
1113 	int error;
1114 
1115 	hash = jhash(name, len, 0);
1116 	b = hash & (ls->ls_rsbtbl_size - 1);
1117 
1118 	spin_lock(&ls->ls_rsbtbl[b].lock);
1119 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
1120 	if (!error)
1121 		goto out_dump;
1122 
1123 	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
1124 	if (error)
1125 		goto out;
1126  out_dump:
1127 	dlm_dump_rsb(r);
1128  out:
1129 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1130 }
1131 
1132 static void toss_rsb(struct kref *kref)
1133 {
1134 	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
1135 	struct dlm_ls *ls = r->res_ls;
1136 
1137 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
1138 	kref_init(&r->res_ref);
1139 	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[r->res_bucket].keep);
1140 	rsb_insert(r, &ls->ls_rsbtbl[r->res_bucket].toss);
1141 	r->res_toss_time = jiffies;
1142 	set_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[r->res_bucket].flags);
1143 	if (r->res_lvbptr) {
1144 		dlm_free_lvb(r->res_lvbptr);
1145 		r->res_lvbptr = NULL;
1146 	}
1147 }
1148 
1149 /* See comment for unhold_lkb */
1150 
1151 static void unhold_rsb(struct dlm_rsb *r)
1152 {
1153 	int rv;
1154 	rv = kref_put(&r->res_ref, toss_rsb);
1155 	DLM_ASSERT(!rv, dlm_dump_rsb(r););
1156 }
1157 
1158 static void kill_rsb(struct kref *kref)
1159 {
1160 	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
1161 
1162 	/* All work is done after the return from kref_put() so we
1163 	   can release the write_lock before the remove and free. */
1164 
1165 	DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
1166 	DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
1167 	DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
1168 	DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
1169 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
1170 	DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
1171 }
1172 
1173 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
1174    The rsb must exist as long as any lkb's for it do. */
1175 
1176 static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
1177 {
1178 	hold_rsb(r);
1179 	lkb->lkb_resource = r;
1180 }
1181 
1182 static void detach_lkb(struct dlm_lkb *lkb)
1183 {
1184 	if (lkb->lkb_resource) {
1185 		put_rsb(lkb->lkb_resource);
1186 		lkb->lkb_resource = NULL;
1187 	}
1188 }
1189 
1190 static int _create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret,
1191 		       int start, int end)
1192 {
1193 	struct dlm_lkb *lkb;
1194 	int rv;
1195 
1196 	lkb = dlm_allocate_lkb(ls);
1197 	if (!lkb)
1198 		return -ENOMEM;
1199 
1200 	lkb->lkb_last_bast_mode = -1;
1201 	lkb->lkb_nodeid = -1;
1202 	lkb->lkb_grmode = DLM_LOCK_IV;
1203 	kref_init(&lkb->lkb_ref);
1204 	INIT_LIST_HEAD(&lkb->lkb_ownqueue);
1205 	INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
1206 	INIT_LIST_HEAD(&lkb->lkb_cb_list);
1207 	INIT_LIST_HEAD(&lkb->lkb_callbacks);
1208 	spin_lock_init(&lkb->lkb_cb_lock);
1209 	INIT_WORK(&lkb->lkb_cb_work, dlm_callback_work);
1210 
1211 	idr_preload(GFP_NOFS);
1212 	spin_lock(&ls->ls_lkbidr_spin);
1213 	rv = idr_alloc(&ls->ls_lkbidr, lkb, start, end, GFP_NOWAIT);
1214 	if (rv >= 0)
1215 		lkb->lkb_id = rv;
1216 	spin_unlock(&ls->ls_lkbidr_spin);
1217 	idr_preload_end();
1218 
1219 	if (rv < 0) {
1220 		log_error(ls, "create_lkb idr error %d", rv);
1221 		dlm_free_lkb(lkb);
1222 		return rv;
1223 	}
1224 
1225 	*lkb_ret = lkb;
1226 	return 0;
1227 }
1228 
1229 static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
1230 {
1231 	return _create_lkb(ls, lkb_ret, 1, 0);
1232 }
1233 
1234 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
1235 {
1236 	struct dlm_lkb *lkb;
1237 
1238 	spin_lock(&ls->ls_lkbidr_spin);
1239 	lkb = idr_find(&ls->ls_lkbidr, lkid);
1240 	if (lkb)
1241 		kref_get(&lkb->lkb_ref);
1242 	spin_unlock(&ls->ls_lkbidr_spin);
1243 
1244 	*lkb_ret = lkb;
1245 	return lkb ? 0 : -ENOENT;
1246 }
1247 
1248 static void kill_lkb(struct kref *kref)
1249 {
1250 	struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
1251 
1252 	/* All work is done after the return from kref_put() so we
1253 	   can release the write_lock before the detach_lkb */
1254 
1255 	DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
1256 }
1257 
1258 /* __put_lkb() is used when an lkb may not have an rsb attached to
1259    it so we need to provide the lockspace explicitly */
1260 
1261 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
1262 {
1263 	uint32_t lkid = lkb->lkb_id;
1264 	int rv;
1265 
1266 	rv = kref_put_lock(&lkb->lkb_ref, kill_lkb,
1267 			   &ls->ls_lkbidr_spin);
1268 	if (rv) {
1269 		idr_remove(&ls->ls_lkbidr, lkid);
1270 		spin_unlock(&ls->ls_lkbidr_spin);
1271 
1272 		detach_lkb(lkb);
1273 
1274 		/* for local/process lkbs, lvbptr points to caller's lksb */
1275 		if (lkb->lkb_lvbptr && is_master_copy(lkb))
1276 			dlm_free_lvb(lkb->lkb_lvbptr);
1277 		dlm_free_lkb(lkb);
1278 	}
1279 
1280 	return rv;
1281 }
1282 
1283 int dlm_put_lkb(struct dlm_lkb *lkb)
1284 {
1285 	struct dlm_ls *ls;
1286 
1287 	DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb););
1288 	DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb););
1289 
1290 	ls = lkb->lkb_resource->res_ls;
1291 	return __put_lkb(ls, lkb);
1292 }
1293 
1294 /* This is only called to add a reference when the code already holds
1295    a valid reference to the lkb, so there's no need for locking. */
1296 
1297 static inline void hold_lkb(struct dlm_lkb *lkb)
1298 {
1299 	kref_get(&lkb->lkb_ref);
1300 }
1301 
1302 static void unhold_lkb_assert(struct kref *kref)
1303 {
1304 	struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
1305 
1306 	DLM_ASSERT(false, dlm_print_lkb(lkb););
1307 }
1308 
1309 /* This is called when we need to remove a reference and are certain
1310    it's not the last ref.  e.g. del_lkb is always called between a
1311    find_lkb/put_lkb and is always the inverse of a previous add_lkb.
1312    put_lkb would work fine, but would involve unnecessary locking */
1313 
1314 static inline void unhold_lkb(struct dlm_lkb *lkb)
1315 {
1316 	kref_put(&lkb->lkb_ref, unhold_lkb_assert);
1317 }
1318 
1319 static void lkb_add_ordered(struct list_head *new, struct list_head *head,
1320 			    int mode)
1321 {
1322 	struct dlm_lkb *lkb = NULL, *iter;
1323 
1324 	list_for_each_entry(iter, head, lkb_statequeue)
1325 		if (iter->lkb_rqmode < mode) {
1326 			lkb = iter;
1327 			list_add_tail(new, &iter->lkb_statequeue);
1328 			break;
1329 		}
1330 
1331 	if (!lkb)
1332 		list_add_tail(new, head);
1333 }
1334 
1335 /* add/remove lkb to rsb's grant/convert/wait queue */
1336 
1337 static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status)
1338 {
1339 	kref_get(&lkb->lkb_ref);
1340 
1341 	DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
1342 
1343 	lkb->lkb_timestamp = ktime_get();
1344 
1345 	lkb->lkb_status = status;
1346 
1347 	switch (status) {
1348 	case DLM_LKSTS_WAITING:
1349 		if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
1350 			list_add(&lkb->lkb_statequeue, &r->res_waitqueue);
1351 		else
1352 			list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue);
1353 		break;
1354 	case DLM_LKSTS_GRANTED:
1355 		/* convention says granted locks kept in order of grmode */
1356 		lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue,
1357 				lkb->lkb_grmode);
1358 		break;
1359 	case DLM_LKSTS_CONVERT:
1360 		if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
1361 			list_add(&lkb->lkb_statequeue, &r->res_convertqueue);
1362 		else
1363 			list_add_tail(&lkb->lkb_statequeue,
1364 				      &r->res_convertqueue);
1365 		break;
1366 	default:
1367 		DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status););
1368 	}
1369 }
1370 
1371 static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
1372 {
1373 	lkb->lkb_status = 0;
1374 	list_del(&lkb->lkb_statequeue);
1375 	unhold_lkb(lkb);
1376 }
1377 
1378 static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
1379 {
1380 	hold_lkb(lkb);
1381 	del_lkb(r, lkb);
1382 	add_lkb(r, lkb, sts);
1383 	unhold_lkb(lkb);
1384 }
1385 
1386 static int msg_reply_type(int mstype)
1387 {
1388 	switch (mstype) {
1389 	case DLM_MSG_REQUEST:
1390 		return DLM_MSG_REQUEST_REPLY;
1391 	case DLM_MSG_CONVERT:
1392 		return DLM_MSG_CONVERT_REPLY;
1393 	case DLM_MSG_UNLOCK:
1394 		return DLM_MSG_UNLOCK_REPLY;
1395 	case DLM_MSG_CANCEL:
1396 		return DLM_MSG_CANCEL_REPLY;
1397 	case DLM_MSG_LOOKUP:
1398 		return DLM_MSG_LOOKUP_REPLY;
1399 	}
1400 	return -1;
1401 }
1402 
1403 /* add/remove lkb from global waiters list of lkb's waiting for
1404    a reply from a remote node */
1405 
1406 static int add_to_waiters(struct dlm_lkb *lkb, int mstype, int to_nodeid)
1407 {
1408 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1409 	int error = 0;
1410 
1411 	mutex_lock(&ls->ls_waiters_mutex);
1412 
1413 	if (is_overlap_unlock(lkb) ||
1414 	    (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
1415 		error = -EINVAL;
1416 		goto out;
1417 	}
1418 
1419 	if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
1420 		switch (mstype) {
1421 		case DLM_MSG_UNLOCK:
1422 			set_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags);
1423 			break;
1424 		case DLM_MSG_CANCEL:
1425 			set_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
1426 			break;
1427 		default:
1428 			error = -EBUSY;
1429 			goto out;
1430 		}
1431 		lkb->lkb_wait_count++;
1432 		hold_lkb(lkb);
1433 
1434 		log_debug(ls, "addwait %x cur %d overlap %d count %d f %x",
1435 			  lkb->lkb_id, lkb->lkb_wait_type, mstype,
1436 			  lkb->lkb_wait_count, dlm_iflags_val(lkb));
1437 		goto out;
1438 	}
1439 
1440 	DLM_ASSERT(!lkb->lkb_wait_count,
1441 		   dlm_print_lkb(lkb);
1442 		   printk("wait_count %d\n", lkb->lkb_wait_count););
1443 
1444 	lkb->lkb_wait_count++;
1445 	lkb->lkb_wait_type = mstype;
1446 	lkb->lkb_wait_nodeid = to_nodeid; /* for debugging */
1447 	hold_lkb(lkb);
1448 	list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
1449  out:
1450 	if (error)
1451 		log_error(ls, "addwait error %x %d flags %x %d %d %s",
1452 			  lkb->lkb_id, error, dlm_iflags_val(lkb), mstype,
1453 			  lkb->lkb_wait_type, lkb->lkb_resource->res_name);
1454 	mutex_unlock(&ls->ls_waiters_mutex);
1455 	return error;
1456 }
1457 
1458 /* We clear the RESEND flag because we might be taking an lkb off the waiters
1459    list as part of process_requestqueue (e.g. a lookup that has an optimized
1460    request reply on the requestqueue) between dlm_recover_waiters_pre() which
1461    set RESEND and dlm_recover_waiters_post() */
1462 
1463 static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype,
1464 				struct dlm_message *ms)
1465 {
1466 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1467 	int overlap_done = 0;
1468 
1469 	if (mstype == DLM_MSG_UNLOCK_REPLY &&
1470 	    test_and_clear_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags)) {
1471 		log_debug(ls, "remwait %x unlock_reply overlap", lkb->lkb_id);
1472 		overlap_done = 1;
1473 		goto out_del;
1474 	}
1475 
1476 	if (mstype == DLM_MSG_CANCEL_REPLY &&
1477 	    test_and_clear_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags)) {
1478 		log_debug(ls, "remwait %x cancel_reply overlap", lkb->lkb_id);
1479 		overlap_done = 1;
1480 		goto out_del;
1481 	}
1482 
1483 	/* Cancel state was preemptively cleared by a successful convert,
1484 	   see next comment, nothing to do. */
1485 
1486 	if ((mstype == DLM_MSG_CANCEL_REPLY) &&
1487 	    (lkb->lkb_wait_type != DLM_MSG_CANCEL)) {
1488 		log_debug(ls, "remwait %x cancel_reply wait_type %d",
1489 			  lkb->lkb_id, lkb->lkb_wait_type);
1490 		return -1;
1491 	}
1492 
1493 	/* Remove for the convert reply, and premptively remove for the
1494 	   cancel reply.  A convert has been granted while there's still
1495 	   an outstanding cancel on it (the cancel is moot and the result
1496 	   in the cancel reply should be 0).  We preempt the cancel reply
1497 	   because the app gets the convert result and then can follow up
1498 	   with another op, like convert.  This subsequent op would see the
1499 	   lingering state of the cancel and fail with -EBUSY. */
1500 
1501 	if ((mstype == DLM_MSG_CONVERT_REPLY) &&
1502 	    (lkb->lkb_wait_type == DLM_MSG_CONVERT) && ms && !ms->m_result &&
1503 	    test_and_clear_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags)) {
1504 		log_debug(ls, "remwait %x convert_reply zap overlap_cancel",
1505 			  lkb->lkb_id);
1506 		lkb->lkb_wait_type = 0;
1507 		lkb->lkb_wait_count--;
1508 		unhold_lkb(lkb);
1509 		goto out_del;
1510 	}
1511 
1512 	/* N.B. type of reply may not always correspond to type of original
1513 	   msg due to lookup->request optimization, verify others? */
1514 
1515 	if (lkb->lkb_wait_type) {
1516 		lkb->lkb_wait_type = 0;
1517 		goto out_del;
1518 	}
1519 
1520 	log_error(ls, "remwait error %x remote %d %x msg %d flags %x no wait",
1521 		  lkb->lkb_id, ms ? le32_to_cpu(ms->m_header.h_nodeid) : 0,
1522 		  lkb->lkb_remid, mstype, dlm_iflags_val(lkb));
1523 	return -1;
1524 
1525  out_del:
1526 	/* the force-unlock/cancel has completed and we haven't recvd a reply
1527 	   to the op that was in progress prior to the unlock/cancel; we
1528 	   give up on any reply to the earlier op.  FIXME: not sure when/how
1529 	   this would happen */
1530 
1531 	if (overlap_done && lkb->lkb_wait_type) {
1532 		log_error(ls, "remwait error %x reply %d wait_type %d overlap",
1533 			  lkb->lkb_id, mstype, lkb->lkb_wait_type);
1534 		lkb->lkb_wait_count--;
1535 		unhold_lkb(lkb);
1536 		lkb->lkb_wait_type = 0;
1537 	}
1538 
1539 	DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
1540 
1541 	clear_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags);
1542 	lkb->lkb_wait_count--;
1543 	if (!lkb->lkb_wait_count)
1544 		list_del_init(&lkb->lkb_wait_reply);
1545 	unhold_lkb(lkb);
1546 	return 0;
1547 }
1548 
1549 static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
1550 {
1551 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1552 	int error;
1553 
1554 	mutex_lock(&ls->ls_waiters_mutex);
1555 	error = _remove_from_waiters(lkb, mstype, NULL);
1556 	mutex_unlock(&ls->ls_waiters_mutex);
1557 	return error;
1558 }
1559 
1560 /* Handles situations where we might be processing a "fake" or "local" reply in
1561    which we can't try to take waiters_mutex again. */
1562 
1563 static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms,
1564 				  bool local)
1565 {
1566 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1567 	int error;
1568 
1569 	if (!local)
1570 		mutex_lock(&ls->ls_waiters_mutex);
1571 	error = _remove_from_waiters(lkb, le32_to_cpu(ms->m_type), ms);
1572 	if (!local)
1573 		mutex_unlock(&ls->ls_waiters_mutex);
1574 	return error;
1575 }
1576 
1577 static void shrink_bucket(struct dlm_ls *ls, int b)
1578 {
1579 	struct rb_node *n, *next;
1580 	struct dlm_rsb *r;
1581 	char *name;
1582 	int our_nodeid = dlm_our_nodeid();
1583 	int remote_count = 0;
1584 	int need_shrink = 0;
1585 	int i, len, rv;
1586 
1587 	memset(&ls->ls_remove_lens, 0, sizeof(int) * DLM_REMOVE_NAMES_MAX);
1588 
1589 	spin_lock(&ls->ls_rsbtbl[b].lock);
1590 
1591 	if (!test_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[b].flags)) {
1592 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1593 		return;
1594 	}
1595 
1596 	for (n = rb_first(&ls->ls_rsbtbl[b].toss); n; n = next) {
1597 		next = rb_next(n);
1598 		r = rb_entry(n, struct dlm_rsb, res_hashnode);
1599 
1600 		/* If we're the directory record for this rsb, and
1601 		   we're not the master of it, then we need to wait
1602 		   for the master node to send us a dir remove for
1603 		   before removing the dir record. */
1604 
1605 		if (!dlm_no_directory(ls) &&
1606 		    (r->res_master_nodeid != our_nodeid) &&
1607 		    (dlm_dir_nodeid(r) == our_nodeid)) {
1608 			continue;
1609 		}
1610 
1611 		need_shrink = 1;
1612 
1613 		if (!time_after_eq(jiffies, r->res_toss_time +
1614 				   dlm_config.ci_toss_secs * HZ)) {
1615 			continue;
1616 		}
1617 
1618 		if (!dlm_no_directory(ls) &&
1619 		    (r->res_master_nodeid == our_nodeid) &&
1620 		    (dlm_dir_nodeid(r) != our_nodeid)) {
1621 
1622 			/* We're the master of this rsb but we're not
1623 			   the directory record, so we need to tell the
1624 			   dir node to remove the dir record. */
1625 
1626 			ls->ls_remove_lens[remote_count] = r->res_length;
1627 			memcpy(ls->ls_remove_names[remote_count], r->res_name,
1628 			       DLM_RESNAME_MAXLEN);
1629 			remote_count++;
1630 
1631 			if (remote_count >= DLM_REMOVE_NAMES_MAX)
1632 				break;
1633 			continue;
1634 		}
1635 
1636 		if (!kref_put(&r->res_ref, kill_rsb)) {
1637 			log_error(ls, "tossed rsb in use %s", r->res_name);
1638 			continue;
1639 		}
1640 
1641 		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
1642 		dlm_free_rsb(r);
1643 	}
1644 
1645 	if (need_shrink)
1646 		set_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[b].flags);
1647 	else
1648 		clear_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[b].flags);
1649 	spin_unlock(&ls->ls_rsbtbl[b].lock);
1650 
1651 	/*
1652 	 * While searching for rsb's to free, we found some that require
1653 	 * remote removal.  We leave them in place and find them again here
1654 	 * so there is a very small gap between removing them from the toss
1655 	 * list and sending the removal.  Keeping this gap small is
1656 	 * important to keep us (the master node) from being out of sync
1657 	 * with the remote dir node for very long.
1658 	 */
1659 
1660 	for (i = 0; i < remote_count; i++) {
1661 		name = ls->ls_remove_names[i];
1662 		len = ls->ls_remove_lens[i];
1663 
1664 		spin_lock(&ls->ls_rsbtbl[b].lock);
1665 		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
1666 		if (rv) {
1667 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1668 			log_debug(ls, "remove_name not toss %s", name);
1669 			continue;
1670 		}
1671 
1672 		if (r->res_master_nodeid != our_nodeid) {
1673 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1674 			log_debug(ls, "remove_name master %d dir %d our %d %s",
1675 				  r->res_master_nodeid, r->res_dir_nodeid,
1676 				  our_nodeid, name);
1677 			continue;
1678 		}
1679 
1680 		if (r->res_dir_nodeid == our_nodeid) {
1681 			/* should never happen */
1682 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1683 			log_error(ls, "remove_name dir %d master %d our %d %s",
1684 				  r->res_dir_nodeid, r->res_master_nodeid,
1685 				  our_nodeid, name);
1686 			continue;
1687 		}
1688 
1689 		if (!time_after_eq(jiffies, r->res_toss_time +
1690 				   dlm_config.ci_toss_secs * HZ)) {
1691 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1692 			log_debug(ls, "remove_name toss_time %lu now %lu %s",
1693 				  r->res_toss_time, jiffies, name);
1694 			continue;
1695 		}
1696 
1697 		if (!kref_put(&r->res_ref, kill_rsb)) {
1698 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1699 			log_error(ls, "remove_name in use %s", name);
1700 			continue;
1701 		}
1702 
1703 		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
1704 		send_remove(r);
1705 		spin_unlock(&ls->ls_rsbtbl[b].lock);
1706 
1707 		dlm_free_rsb(r);
1708 	}
1709 }
1710 
1711 void dlm_scan_rsbs(struct dlm_ls *ls)
1712 {
1713 	int i;
1714 
1715 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1716 		shrink_bucket(ls, i);
1717 		if (dlm_locking_stopped(ls))
1718 			break;
1719 		cond_resched();
1720 	}
1721 }
1722 
1723 /* lkb is master or local copy */
1724 
1725 static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1726 {
1727 	int b, len = r->res_ls->ls_lvblen;
1728 
1729 	/* b=1 lvb returned to caller
1730 	   b=0 lvb written to rsb or invalidated
1731 	   b=-1 do nothing */
1732 
1733 	b =  dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1734 
1735 	if (b == 1) {
1736 		if (!lkb->lkb_lvbptr)
1737 			return;
1738 
1739 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1740 			return;
1741 
1742 		if (!r->res_lvbptr)
1743 			return;
1744 
1745 		memcpy(lkb->lkb_lvbptr, r->res_lvbptr, len);
1746 		lkb->lkb_lvbseq = r->res_lvbseq;
1747 
1748 	} else if (b == 0) {
1749 		if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1750 			rsb_set_flag(r, RSB_VALNOTVALID);
1751 			return;
1752 		}
1753 
1754 		if (!lkb->lkb_lvbptr)
1755 			return;
1756 
1757 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1758 			return;
1759 
1760 		if (!r->res_lvbptr)
1761 			r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1762 
1763 		if (!r->res_lvbptr)
1764 			return;
1765 
1766 		memcpy(r->res_lvbptr, lkb->lkb_lvbptr, len);
1767 		r->res_lvbseq++;
1768 		lkb->lkb_lvbseq = r->res_lvbseq;
1769 		rsb_clear_flag(r, RSB_VALNOTVALID);
1770 	}
1771 
1772 	if (rsb_flag(r, RSB_VALNOTVALID))
1773 		set_bit(DLM_SBF_VALNOTVALID_BIT, &lkb->lkb_sbflags);
1774 }
1775 
1776 static void set_lvb_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1777 {
1778 	if (lkb->lkb_grmode < DLM_LOCK_PW)
1779 		return;
1780 
1781 	if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1782 		rsb_set_flag(r, RSB_VALNOTVALID);
1783 		return;
1784 	}
1785 
1786 	if (!lkb->lkb_lvbptr)
1787 		return;
1788 
1789 	if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1790 		return;
1791 
1792 	if (!r->res_lvbptr)
1793 		r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1794 
1795 	if (!r->res_lvbptr)
1796 		return;
1797 
1798 	memcpy(r->res_lvbptr, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
1799 	r->res_lvbseq++;
1800 	rsb_clear_flag(r, RSB_VALNOTVALID);
1801 }
1802 
1803 /* lkb is process copy (pc) */
1804 
1805 static void set_lvb_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1806 			    struct dlm_message *ms)
1807 {
1808 	int b;
1809 
1810 	if (!lkb->lkb_lvbptr)
1811 		return;
1812 
1813 	if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1814 		return;
1815 
1816 	b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1817 	if (b == 1) {
1818 		int len = receive_extralen(ms);
1819 		if (len > r->res_ls->ls_lvblen)
1820 			len = r->res_ls->ls_lvblen;
1821 		memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
1822 		lkb->lkb_lvbseq = le32_to_cpu(ms->m_lvbseq);
1823 	}
1824 }
1825 
1826 /* Manipulate lkb's on rsb's convert/granted/waiting queues
1827    remove_lock -- used for unlock, removes lkb from granted
1828    revert_lock -- used for cancel, moves lkb from convert to granted
1829    grant_lock  -- used for request and convert, adds lkb to granted or
1830                   moves lkb from convert or waiting to granted
1831 
1832    Each of these is used for master or local copy lkb's.  There is
1833    also a _pc() variation used to make the corresponding change on
1834    a process copy (pc) lkb. */
1835 
1836 static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1837 {
1838 	del_lkb(r, lkb);
1839 	lkb->lkb_grmode = DLM_LOCK_IV;
1840 	/* this unhold undoes the original ref from create_lkb()
1841 	   so this leads to the lkb being freed */
1842 	unhold_lkb(lkb);
1843 }
1844 
1845 static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1846 {
1847 	set_lvb_unlock(r, lkb);
1848 	_remove_lock(r, lkb);
1849 }
1850 
1851 static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1852 {
1853 	_remove_lock(r, lkb);
1854 }
1855 
1856 /* returns: 0 did nothing
1857 	    1 moved lock to granted
1858 	   -1 removed lock */
1859 
1860 static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1861 {
1862 	int rv = 0;
1863 
1864 	lkb->lkb_rqmode = DLM_LOCK_IV;
1865 
1866 	switch (lkb->lkb_status) {
1867 	case DLM_LKSTS_GRANTED:
1868 		break;
1869 	case DLM_LKSTS_CONVERT:
1870 		move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1871 		rv = 1;
1872 		break;
1873 	case DLM_LKSTS_WAITING:
1874 		del_lkb(r, lkb);
1875 		lkb->lkb_grmode = DLM_LOCK_IV;
1876 		/* this unhold undoes the original ref from create_lkb()
1877 		   so this leads to the lkb being freed */
1878 		unhold_lkb(lkb);
1879 		rv = -1;
1880 		break;
1881 	default:
1882 		log_print("invalid status for revert %d", lkb->lkb_status);
1883 	}
1884 	return rv;
1885 }
1886 
1887 static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1888 {
1889 	return revert_lock(r, lkb);
1890 }
1891 
1892 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1893 {
1894 	if (lkb->lkb_grmode != lkb->lkb_rqmode) {
1895 		lkb->lkb_grmode = lkb->lkb_rqmode;
1896 		if (lkb->lkb_status)
1897 			move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1898 		else
1899 			add_lkb(r, lkb, DLM_LKSTS_GRANTED);
1900 	}
1901 
1902 	lkb->lkb_rqmode = DLM_LOCK_IV;
1903 	lkb->lkb_highbast = 0;
1904 }
1905 
1906 static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1907 {
1908 	set_lvb_lock(r, lkb);
1909 	_grant_lock(r, lkb);
1910 }
1911 
1912 static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1913 			  struct dlm_message *ms)
1914 {
1915 	set_lvb_lock_pc(r, lkb, ms);
1916 	_grant_lock(r, lkb);
1917 }
1918 
1919 /* called by grant_pending_locks() which means an async grant message must
1920    be sent to the requesting node in addition to granting the lock if the
1921    lkb belongs to a remote node. */
1922 
1923 static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
1924 {
1925 	grant_lock(r, lkb);
1926 	if (is_master_copy(lkb))
1927 		send_grant(r, lkb);
1928 	else
1929 		queue_cast(r, lkb, 0);
1930 }
1931 
1932 /* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
1933    change the granted/requested modes.  We're munging things accordingly in
1934    the process copy.
1935    CONVDEADLK: our grmode may have been forced down to NL to resolve a
1936    conversion deadlock
1937    ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
1938    compatible with other granted locks */
1939 
1940 static void munge_demoted(struct dlm_lkb *lkb)
1941 {
1942 	if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
1943 		log_print("munge_demoted %x invalid modes gr %d rq %d",
1944 			  lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
1945 		return;
1946 	}
1947 
1948 	lkb->lkb_grmode = DLM_LOCK_NL;
1949 }
1950 
1951 static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
1952 {
1953 	if (ms->m_type != cpu_to_le32(DLM_MSG_REQUEST_REPLY) &&
1954 	    ms->m_type != cpu_to_le32(DLM_MSG_GRANT)) {
1955 		log_print("munge_altmode %x invalid reply type %d",
1956 			  lkb->lkb_id, le32_to_cpu(ms->m_type));
1957 		return;
1958 	}
1959 
1960 	if (lkb->lkb_exflags & DLM_LKF_ALTPR)
1961 		lkb->lkb_rqmode = DLM_LOCK_PR;
1962 	else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
1963 		lkb->lkb_rqmode = DLM_LOCK_CW;
1964 	else {
1965 		log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
1966 		dlm_print_lkb(lkb);
1967 	}
1968 }
1969 
1970 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
1971 {
1972 	struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
1973 					   lkb_statequeue);
1974 	if (lkb->lkb_id == first->lkb_id)
1975 		return 1;
1976 
1977 	return 0;
1978 }
1979 
1980 /* Check if the given lkb conflicts with another lkb on the queue. */
1981 
1982 static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb)
1983 {
1984 	struct dlm_lkb *this;
1985 
1986 	list_for_each_entry(this, head, lkb_statequeue) {
1987 		if (this == lkb)
1988 			continue;
1989 		if (!modes_compat(this, lkb))
1990 			return 1;
1991 	}
1992 	return 0;
1993 }
1994 
1995 /*
1996  * "A conversion deadlock arises with a pair of lock requests in the converting
1997  * queue for one resource.  The granted mode of each lock blocks the requested
1998  * mode of the other lock."
1999  *
2000  * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
2001  * convert queue from being granted, then deadlk/demote lkb.
2002  *
2003  * Example:
2004  * Granted Queue: empty
2005  * Convert Queue: NL->EX (first lock)
2006  *                PR->EX (second lock)
2007  *
2008  * The first lock can't be granted because of the granted mode of the second
2009  * lock and the second lock can't be granted because it's not first in the
2010  * list.  We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
2011  * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
2012  * flag set and return DEMOTED in the lksb flags.
2013  *
2014  * Originally, this function detected conv-deadlk in a more limited scope:
2015  * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
2016  * - if lkb1 was the first entry in the queue (not just earlier), and was
2017  *   blocked by the granted mode of lkb2, and there was nothing on the
2018  *   granted queue preventing lkb1 from being granted immediately, i.e.
2019  *   lkb2 was the only thing preventing lkb1 from being granted.
2020  *
2021  * That second condition meant we'd only say there was conv-deadlk if
2022  * resolving it (by demotion) would lead to the first lock on the convert
2023  * queue being granted right away.  It allowed conversion deadlocks to exist
2024  * between locks on the convert queue while they couldn't be granted anyway.
2025  *
2026  * Now, we detect and take action on conversion deadlocks immediately when
2027  * they're created, even if they may not be immediately consequential.  If
2028  * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
2029  * mode that would prevent lkb1's conversion from being granted, we do a
2030  * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
2031  * I think this means that the lkb_is_ahead condition below should always
2032  * be zero, i.e. there will never be conv-deadlk between two locks that are
2033  * both already on the convert queue.
2034  */
2035 
2036 static int conversion_deadlock_detect(struct dlm_rsb *r, struct dlm_lkb *lkb2)
2037 {
2038 	struct dlm_lkb *lkb1;
2039 	int lkb_is_ahead = 0;
2040 
2041 	list_for_each_entry(lkb1, &r->res_convertqueue, lkb_statequeue) {
2042 		if (lkb1 == lkb2) {
2043 			lkb_is_ahead = 1;
2044 			continue;
2045 		}
2046 
2047 		if (!lkb_is_ahead) {
2048 			if (!modes_compat(lkb2, lkb1))
2049 				return 1;
2050 		} else {
2051 			if (!modes_compat(lkb2, lkb1) &&
2052 			    !modes_compat(lkb1, lkb2))
2053 				return 1;
2054 		}
2055 	}
2056 	return 0;
2057 }
2058 
2059 /*
2060  * Return 1 if the lock can be granted, 0 otherwise.
2061  * Also detect and resolve conversion deadlocks.
2062  *
2063  * lkb is the lock to be granted
2064  *
2065  * now is 1 if the function is being called in the context of the
2066  * immediate request, it is 0 if called later, after the lock has been
2067  * queued.
2068  *
2069  * recover is 1 if dlm_recover_grant() is trying to grant conversions
2070  * after recovery.
2071  *
2072  * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
2073  */
2074 
2075 static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
2076 			   int recover)
2077 {
2078 	int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV);
2079 
2080 	/*
2081 	 * 6-10: Version 5.4 introduced an option to address the phenomenon of
2082 	 * a new request for a NL mode lock being blocked.
2083 	 *
2084 	 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
2085 	 * request, then it would be granted.  In essence, the use of this flag
2086 	 * tells the Lock Manager to expedite theis request by not considering
2087 	 * what may be in the CONVERTING or WAITING queues...  As of this
2088 	 * writing, the EXPEDITE flag can be used only with new requests for NL
2089 	 * mode locks.  This flag is not valid for conversion requests.
2090 	 *
2091 	 * A shortcut.  Earlier checks return an error if EXPEDITE is used in a
2092 	 * conversion or used with a non-NL requested mode.  We also know an
2093 	 * EXPEDITE request is always granted immediately, so now must always
2094 	 * be 1.  The full condition to grant an expedite request: (now &&
2095 	 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
2096 	 * therefore be shortened to just checking the flag.
2097 	 */
2098 
2099 	if (lkb->lkb_exflags & DLM_LKF_EXPEDITE)
2100 		return 1;
2101 
2102 	/*
2103 	 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
2104 	 * added to the remaining conditions.
2105 	 */
2106 
2107 	if (queue_conflict(&r->res_grantqueue, lkb))
2108 		return 0;
2109 
2110 	/*
2111 	 * 6-3: By default, a conversion request is immediately granted if the
2112 	 * requested mode is compatible with the modes of all other granted
2113 	 * locks
2114 	 */
2115 
2116 	if (queue_conflict(&r->res_convertqueue, lkb))
2117 		return 0;
2118 
2119 	/*
2120 	 * The RECOVER_GRANT flag means dlm_recover_grant() is granting
2121 	 * locks for a recovered rsb, on which lkb's have been rebuilt.
2122 	 * The lkb's may have been rebuilt on the queues in a different
2123 	 * order than they were in on the previous master.  So, granting
2124 	 * queued conversions in order after recovery doesn't make sense
2125 	 * since the order hasn't been preserved anyway.  The new order
2126 	 * could also have created a new "in place" conversion deadlock.
2127 	 * (e.g. old, failed master held granted EX, with PR->EX, NL->EX.
2128 	 * After recovery, there would be no granted locks, and possibly
2129 	 * NL->EX, PR->EX, an in-place conversion deadlock.)  So, after
2130 	 * recovery, grant conversions without considering order.
2131 	 */
2132 
2133 	if (conv && recover)
2134 		return 1;
2135 
2136 	/*
2137 	 * 6-5: But the default algorithm for deciding whether to grant or
2138 	 * queue conversion requests does not by itself guarantee that such
2139 	 * requests are serviced on a "first come first serve" basis.  This, in
2140 	 * turn, can lead to a phenomenon known as "indefinate postponement".
2141 	 *
2142 	 * 6-7: This issue is dealt with by using the optional QUECVT flag with
2143 	 * the system service employed to request a lock conversion.  This flag
2144 	 * forces certain conversion requests to be queued, even if they are
2145 	 * compatible with the granted modes of other locks on the same
2146 	 * resource.  Thus, the use of this flag results in conversion requests
2147 	 * being ordered on a "first come first servce" basis.
2148 	 *
2149 	 * DCT: This condition is all about new conversions being able to occur
2150 	 * "in place" while the lock remains on the granted queue (assuming
2151 	 * nothing else conflicts.)  IOW if QUECVT isn't set, a conversion
2152 	 * doesn't _have_ to go onto the convert queue where it's processed in
2153 	 * order.  The "now" variable is necessary to distinguish converts
2154 	 * being received and processed for the first time now, because once a
2155 	 * convert is moved to the conversion queue the condition below applies
2156 	 * requiring fifo granting.
2157 	 */
2158 
2159 	if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT))
2160 		return 1;
2161 
2162 	/*
2163 	 * Even if the convert is compat with all granted locks,
2164 	 * QUECVT forces it behind other locks on the convert queue.
2165 	 */
2166 
2167 	if (now && conv && (lkb->lkb_exflags & DLM_LKF_QUECVT)) {
2168 		if (list_empty(&r->res_convertqueue))
2169 			return 1;
2170 		else
2171 			return 0;
2172 	}
2173 
2174 	/*
2175 	 * The NOORDER flag is set to avoid the standard vms rules on grant
2176 	 * order.
2177 	 */
2178 
2179 	if (lkb->lkb_exflags & DLM_LKF_NOORDER)
2180 		return 1;
2181 
2182 	/*
2183 	 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
2184 	 * granted until all other conversion requests ahead of it are granted
2185 	 * and/or canceled.
2186 	 */
2187 
2188 	if (!now && conv && first_in_list(lkb, &r->res_convertqueue))
2189 		return 1;
2190 
2191 	/*
2192 	 * 6-4: By default, a new request is immediately granted only if all
2193 	 * three of the following conditions are satisfied when the request is
2194 	 * issued:
2195 	 * - The queue of ungranted conversion requests for the resource is
2196 	 *   empty.
2197 	 * - The queue of ungranted new requests for the resource is empty.
2198 	 * - The mode of the new request is compatible with the most
2199 	 *   restrictive mode of all granted locks on the resource.
2200 	 */
2201 
2202 	if (now && !conv && list_empty(&r->res_convertqueue) &&
2203 	    list_empty(&r->res_waitqueue))
2204 		return 1;
2205 
2206 	/*
2207 	 * 6-4: Once a lock request is in the queue of ungranted new requests,
2208 	 * it cannot be granted until the queue of ungranted conversion
2209 	 * requests is empty, all ungranted new requests ahead of it are
2210 	 * granted and/or canceled, and it is compatible with the granted mode
2211 	 * of the most restrictive lock granted on the resource.
2212 	 */
2213 
2214 	if (!now && !conv && list_empty(&r->res_convertqueue) &&
2215 	    first_in_list(lkb, &r->res_waitqueue))
2216 		return 1;
2217 
2218 	return 0;
2219 }
2220 
2221 static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
2222 			  int recover, int *err)
2223 {
2224 	int rv;
2225 	int8_t alt = 0, rqmode = lkb->lkb_rqmode;
2226 	int8_t is_convert = (lkb->lkb_grmode != DLM_LOCK_IV);
2227 
2228 	if (err)
2229 		*err = 0;
2230 
2231 	rv = _can_be_granted(r, lkb, now, recover);
2232 	if (rv)
2233 		goto out;
2234 
2235 	/*
2236 	 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
2237 	 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
2238 	 * cancels one of the locks.
2239 	 */
2240 
2241 	if (is_convert && can_be_queued(lkb) &&
2242 	    conversion_deadlock_detect(r, lkb)) {
2243 		if (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) {
2244 			lkb->lkb_grmode = DLM_LOCK_NL;
2245 			set_bit(DLM_SBF_DEMOTED_BIT, &lkb->lkb_sbflags);
2246 		} else if (err) {
2247 			*err = -EDEADLK;
2248 		} else {
2249 			log_print("can_be_granted deadlock %x now %d",
2250 				  lkb->lkb_id, now);
2251 			dlm_dump_rsb(r);
2252 		}
2253 		goto out;
2254 	}
2255 
2256 	/*
2257 	 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
2258 	 * to grant a request in a mode other than the normal rqmode.  It's a
2259 	 * simple way to provide a big optimization to applications that can
2260 	 * use them.
2261 	 */
2262 
2263 	if (rqmode != DLM_LOCK_PR && (lkb->lkb_exflags & DLM_LKF_ALTPR))
2264 		alt = DLM_LOCK_PR;
2265 	else if (rqmode != DLM_LOCK_CW && (lkb->lkb_exflags & DLM_LKF_ALTCW))
2266 		alt = DLM_LOCK_CW;
2267 
2268 	if (alt) {
2269 		lkb->lkb_rqmode = alt;
2270 		rv = _can_be_granted(r, lkb, now, 0);
2271 		if (rv)
2272 			set_bit(DLM_SBF_ALTMODE_BIT, &lkb->lkb_sbflags);
2273 		else
2274 			lkb->lkb_rqmode = rqmode;
2275 	}
2276  out:
2277 	return rv;
2278 }
2279 
2280 /* Returns the highest requested mode of all blocked conversions; sets
2281    cw if there's a blocked conversion to DLM_LOCK_CW. */
2282 
2283 static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw,
2284 				 unsigned int *count)
2285 {
2286 	struct dlm_lkb *lkb, *s;
2287 	int recover = rsb_flag(r, RSB_RECOVER_GRANT);
2288 	int hi, demoted, quit, grant_restart, demote_restart;
2289 	int deadlk;
2290 
2291 	quit = 0;
2292  restart:
2293 	grant_restart = 0;
2294 	demote_restart = 0;
2295 	hi = DLM_LOCK_IV;
2296 
2297 	list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) {
2298 		demoted = is_demoted(lkb);
2299 		deadlk = 0;
2300 
2301 		if (can_be_granted(r, lkb, 0, recover, &deadlk)) {
2302 			grant_lock_pending(r, lkb);
2303 			grant_restart = 1;
2304 			if (count)
2305 				(*count)++;
2306 			continue;
2307 		}
2308 
2309 		if (!demoted && is_demoted(lkb)) {
2310 			log_print("WARN: pending demoted %x node %d %s",
2311 				  lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
2312 			demote_restart = 1;
2313 			continue;
2314 		}
2315 
2316 		if (deadlk) {
2317 			/*
2318 			 * If DLM_LKB_NODLKWT flag is set and conversion
2319 			 * deadlock is detected, we request blocking AST and
2320 			 * down (or cancel) conversion.
2321 			 */
2322 			if (lkb->lkb_exflags & DLM_LKF_NODLCKWT) {
2323 				if (lkb->lkb_highbast < lkb->lkb_rqmode) {
2324 					queue_bast(r, lkb, lkb->lkb_rqmode);
2325 					lkb->lkb_highbast = lkb->lkb_rqmode;
2326 				}
2327 			} else {
2328 				log_print("WARN: pending deadlock %x node %d %s",
2329 					  lkb->lkb_id, lkb->lkb_nodeid,
2330 					  r->res_name);
2331 				dlm_dump_rsb(r);
2332 			}
2333 			continue;
2334 		}
2335 
2336 		hi = max_t(int, lkb->lkb_rqmode, hi);
2337 
2338 		if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
2339 			*cw = 1;
2340 	}
2341 
2342 	if (grant_restart)
2343 		goto restart;
2344 	if (demote_restart && !quit) {
2345 		quit = 1;
2346 		goto restart;
2347 	}
2348 
2349 	return max_t(int, high, hi);
2350 }
2351 
2352 static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw,
2353 			      unsigned int *count)
2354 {
2355 	struct dlm_lkb *lkb, *s;
2356 
2357 	list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
2358 		if (can_be_granted(r, lkb, 0, 0, NULL)) {
2359 			grant_lock_pending(r, lkb);
2360 			if (count)
2361 				(*count)++;
2362 		} else {
2363 			high = max_t(int, lkb->lkb_rqmode, high);
2364 			if (lkb->lkb_rqmode == DLM_LOCK_CW)
2365 				*cw = 1;
2366 		}
2367 	}
2368 
2369 	return high;
2370 }
2371 
2372 /* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
2373    on either the convert or waiting queue.
2374    high is the largest rqmode of all locks blocked on the convert or
2375    waiting queue. */
2376 
2377 static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
2378 {
2379 	if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
2380 		if (gr->lkb_highbast < DLM_LOCK_EX)
2381 			return 1;
2382 		return 0;
2383 	}
2384 
2385 	if (gr->lkb_highbast < high &&
2386 	    !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
2387 		return 1;
2388 	return 0;
2389 }
2390 
2391 static void grant_pending_locks(struct dlm_rsb *r, unsigned int *count)
2392 {
2393 	struct dlm_lkb *lkb, *s;
2394 	int high = DLM_LOCK_IV;
2395 	int cw = 0;
2396 
2397 	if (!is_master(r)) {
2398 		log_print("grant_pending_locks r nodeid %d", r->res_nodeid);
2399 		dlm_dump_rsb(r);
2400 		return;
2401 	}
2402 
2403 	high = grant_pending_convert(r, high, &cw, count);
2404 	high = grant_pending_wait(r, high, &cw, count);
2405 
2406 	if (high == DLM_LOCK_IV)
2407 		return;
2408 
2409 	/*
2410 	 * If there are locks left on the wait/convert queue then send blocking
2411 	 * ASTs to granted locks based on the largest requested mode (high)
2412 	 * found above.
2413 	 */
2414 
2415 	list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
2416 		if (lkb->lkb_bastfn && lock_requires_bast(lkb, high, cw)) {
2417 			if (cw && high == DLM_LOCK_PR &&
2418 			    lkb->lkb_grmode == DLM_LOCK_PR)
2419 				queue_bast(r, lkb, DLM_LOCK_CW);
2420 			else
2421 				queue_bast(r, lkb, high);
2422 			lkb->lkb_highbast = high;
2423 		}
2424 	}
2425 }
2426 
2427 static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
2428 {
2429 	if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
2430 	    (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
2431 		if (gr->lkb_highbast < DLM_LOCK_EX)
2432 			return 1;
2433 		return 0;
2434 	}
2435 
2436 	if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
2437 		return 1;
2438 	return 0;
2439 }
2440 
2441 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
2442 			    struct dlm_lkb *lkb)
2443 {
2444 	struct dlm_lkb *gr;
2445 
2446 	list_for_each_entry(gr, head, lkb_statequeue) {
2447 		/* skip self when sending basts to convertqueue */
2448 		if (gr == lkb)
2449 			continue;
2450 		if (gr->lkb_bastfn && modes_require_bast(gr, lkb)) {
2451 			queue_bast(r, gr, lkb->lkb_rqmode);
2452 			gr->lkb_highbast = lkb->lkb_rqmode;
2453 		}
2454 	}
2455 }
2456 
2457 static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb)
2458 {
2459 	send_bast_queue(r, &r->res_grantqueue, lkb);
2460 }
2461 
2462 static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb)
2463 {
2464 	send_bast_queue(r, &r->res_grantqueue, lkb);
2465 	send_bast_queue(r, &r->res_convertqueue, lkb);
2466 }
2467 
2468 /* set_master(r, lkb) -- set the master nodeid of a resource
2469 
2470    The purpose of this function is to set the nodeid field in the given
2471    lkb using the nodeid field in the given rsb.  If the rsb's nodeid is
2472    known, it can just be copied to the lkb and the function will return
2473    0.  If the rsb's nodeid is _not_ known, it needs to be looked up
2474    before it can be copied to the lkb.
2475 
2476    When the rsb nodeid is being looked up remotely, the initial lkb
2477    causing the lookup is kept on the ls_waiters list waiting for the
2478    lookup reply.  Other lkb's waiting for the same rsb lookup are kept
2479    on the rsb's res_lookup list until the master is verified.
2480 
2481    Return values:
2482    0: nodeid is set in rsb/lkb and the caller should go ahead and use it
2483    1: the rsb master is not available and the lkb has been placed on
2484       a wait queue
2485 */
2486 
2487 static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb)
2488 {
2489 	int our_nodeid = dlm_our_nodeid();
2490 
2491 	if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) {
2492 		rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
2493 		r->res_first_lkid = lkb->lkb_id;
2494 		lkb->lkb_nodeid = r->res_nodeid;
2495 		return 0;
2496 	}
2497 
2498 	if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) {
2499 		list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup);
2500 		return 1;
2501 	}
2502 
2503 	if (r->res_master_nodeid == our_nodeid) {
2504 		lkb->lkb_nodeid = 0;
2505 		return 0;
2506 	}
2507 
2508 	if (r->res_master_nodeid) {
2509 		lkb->lkb_nodeid = r->res_master_nodeid;
2510 		return 0;
2511 	}
2512 
2513 	if (dlm_dir_nodeid(r) == our_nodeid) {
2514 		/* This is a somewhat unusual case; find_rsb will usually
2515 		   have set res_master_nodeid when dir nodeid is local, but
2516 		   there are cases where we become the dir node after we've
2517 		   past find_rsb and go through _request_lock again.
2518 		   confirm_master() or process_lookup_list() needs to be
2519 		   called after this. */
2520 		log_debug(r->res_ls, "set_master %x self master %d dir %d %s",
2521 			  lkb->lkb_id, r->res_master_nodeid, r->res_dir_nodeid,
2522 			  r->res_name);
2523 		r->res_master_nodeid = our_nodeid;
2524 		r->res_nodeid = 0;
2525 		lkb->lkb_nodeid = 0;
2526 		return 0;
2527 	}
2528 
2529 	r->res_first_lkid = lkb->lkb_id;
2530 	send_lookup(r, lkb);
2531 	return 1;
2532 }
2533 
2534 static void process_lookup_list(struct dlm_rsb *r)
2535 {
2536 	struct dlm_lkb *lkb, *safe;
2537 
2538 	list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
2539 		list_del_init(&lkb->lkb_rsb_lookup);
2540 		_request_lock(r, lkb);
2541 		schedule();
2542 	}
2543 }
2544 
2545 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
2546 
2547 static void confirm_master(struct dlm_rsb *r, int error)
2548 {
2549 	struct dlm_lkb *lkb;
2550 
2551 	if (!r->res_first_lkid)
2552 		return;
2553 
2554 	switch (error) {
2555 	case 0:
2556 	case -EINPROGRESS:
2557 		r->res_first_lkid = 0;
2558 		process_lookup_list(r);
2559 		break;
2560 
2561 	case -EAGAIN:
2562 	case -EBADR:
2563 	case -ENOTBLK:
2564 		/* the remote request failed and won't be retried (it was
2565 		   a NOQUEUE, or has been canceled/unlocked); make a waiting
2566 		   lkb the first_lkid */
2567 
2568 		r->res_first_lkid = 0;
2569 
2570 		if (!list_empty(&r->res_lookup)) {
2571 			lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
2572 					 lkb_rsb_lookup);
2573 			list_del_init(&lkb->lkb_rsb_lookup);
2574 			r->res_first_lkid = lkb->lkb_id;
2575 			_request_lock(r, lkb);
2576 		}
2577 		break;
2578 
2579 	default:
2580 		log_error(r->res_ls, "confirm_master unknown error %d", error);
2581 	}
2582 }
2583 
2584 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
2585 			 int namelen, void (*ast)(void *astparam),
2586 			 void *astparam,
2587 			 void (*bast)(void *astparam, int mode),
2588 			 struct dlm_args *args)
2589 {
2590 	int rv = -EINVAL;
2591 
2592 	/* check for invalid arg usage */
2593 
2594 	if (mode < 0 || mode > DLM_LOCK_EX)
2595 		goto out;
2596 
2597 	if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
2598 		goto out;
2599 
2600 	if (flags & DLM_LKF_CANCEL)
2601 		goto out;
2602 
2603 	if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
2604 		goto out;
2605 
2606 	if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
2607 		goto out;
2608 
2609 	if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
2610 		goto out;
2611 
2612 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
2613 		goto out;
2614 
2615 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
2616 		goto out;
2617 
2618 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2619 		goto out;
2620 
2621 	if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2622 		goto out;
2623 
2624 	if (!ast || !lksb)
2625 		goto out;
2626 
2627 	if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2628 		goto out;
2629 
2630 	if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2631 		goto out;
2632 
2633 	/* these args will be copied to the lkb in validate_lock_args,
2634 	   it cannot be done now because when converting locks, fields in
2635 	   an active lkb cannot be modified before locking the rsb */
2636 
2637 	args->flags = flags;
2638 	args->astfn = ast;
2639 	args->astparam = astparam;
2640 	args->bastfn = bast;
2641 	args->mode = mode;
2642 	args->lksb = lksb;
2643 	rv = 0;
2644  out:
2645 	return rv;
2646 }
2647 
2648 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2649 {
2650 	if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2651  		      DLM_LKF_FORCEUNLOCK))
2652 		return -EINVAL;
2653 
2654 	if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2655 		return -EINVAL;
2656 
2657 	args->flags = flags;
2658 	args->astparam = astarg;
2659 	return 0;
2660 }
2661 
2662 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2663 			      struct dlm_args *args)
2664 {
2665 	int rv = -EBUSY;
2666 
2667 	if (args->flags & DLM_LKF_CONVERT) {
2668 		if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2669 			goto out;
2670 
2671 		/* lock not allowed if there's any op in progress */
2672 		if (lkb->lkb_wait_type || lkb->lkb_wait_count)
2673 			goto out;
2674 
2675 		if (is_overlap(lkb))
2676 			goto out;
2677 
2678 		rv = -EINVAL;
2679 		if (test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags))
2680 			goto out;
2681 
2682 		if (args->flags & DLM_LKF_QUECVT &&
2683 		    !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2684 			goto out;
2685 	}
2686 
2687 	lkb->lkb_exflags = args->flags;
2688 	dlm_set_sbflags_val(lkb, 0);
2689 	lkb->lkb_astfn = args->astfn;
2690 	lkb->lkb_astparam = args->astparam;
2691 	lkb->lkb_bastfn = args->bastfn;
2692 	lkb->lkb_rqmode = args->mode;
2693 	lkb->lkb_lksb = args->lksb;
2694 	lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2695 	lkb->lkb_ownpid = (int) current->pid;
2696 	rv = 0;
2697  out:
2698 	switch (rv) {
2699 	case 0:
2700 		break;
2701 	case -EINVAL:
2702 		/* annoy the user because dlm usage is wrong */
2703 		WARN_ON(1);
2704 		log_error(ls, "%s %d %x %x %x %d %d %s", __func__,
2705 			  rv, lkb->lkb_id, dlm_iflags_val(lkb), args->flags,
2706 			  lkb->lkb_status, lkb->lkb_wait_type,
2707 			  lkb->lkb_resource->res_name);
2708 		break;
2709 	default:
2710 		log_debug(ls, "%s %d %x %x %x %d %d %s", __func__,
2711 			  rv, lkb->lkb_id, dlm_iflags_val(lkb), args->flags,
2712 			  lkb->lkb_status, lkb->lkb_wait_type,
2713 			  lkb->lkb_resource->res_name);
2714 		break;
2715 	}
2716 
2717 	return rv;
2718 }
2719 
2720 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2721    for success */
2722 
2723 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2724    because there may be a lookup in progress and it's valid to do
2725    cancel/unlockf on it */
2726 
2727 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2728 {
2729 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
2730 	int rv = -EBUSY;
2731 
2732 	/* normal unlock not allowed if there's any op in progress */
2733 	if (!(args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) &&
2734 	    (lkb->lkb_wait_type || lkb->lkb_wait_count))
2735 		goto out;
2736 
2737 	/* an lkb may be waiting for an rsb lookup to complete where the
2738 	   lookup was initiated by another lock */
2739 
2740 	if (!list_empty(&lkb->lkb_rsb_lookup)) {
2741 		if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
2742 			log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2743 			list_del_init(&lkb->lkb_rsb_lookup);
2744 			queue_cast(lkb->lkb_resource, lkb,
2745 				   args->flags & DLM_LKF_CANCEL ?
2746 				   -DLM_ECANCEL : -DLM_EUNLOCK);
2747 			unhold_lkb(lkb); /* undoes create_lkb() */
2748 		}
2749 		/* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2750 		goto out;
2751 	}
2752 
2753 	rv = -EINVAL;
2754 	if (test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags)) {
2755 		log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2756 		dlm_print_lkb(lkb);
2757 		goto out;
2758 	}
2759 
2760 	/* an lkb may still exist even though the lock is EOL'ed due to a
2761 	 * cancel, unlock or failed noqueue request; an app can't use these
2762 	 * locks; return same error as if the lkid had not been found at all
2763 	 */
2764 
2765 	if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) {
2766 		log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2767 		rv = -ENOENT;
2768 		goto out;
2769 	}
2770 
2771 	/* cancel not allowed with another cancel/unlock in progress */
2772 
2773 	if (args->flags & DLM_LKF_CANCEL) {
2774 		if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2775 			goto out;
2776 
2777 		if (is_overlap(lkb))
2778 			goto out;
2779 
2780 		if (test_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags)) {
2781 			set_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
2782 			rv = -EBUSY;
2783 			goto out;
2784 		}
2785 
2786 		/* there's nothing to cancel */
2787 		if (lkb->lkb_status == DLM_LKSTS_GRANTED &&
2788 		    !lkb->lkb_wait_type) {
2789 			rv = -EBUSY;
2790 			goto out;
2791 		}
2792 
2793 		switch (lkb->lkb_wait_type) {
2794 		case DLM_MSG_LOOKUP:
2795 		case DLM_MSG_REQUEST:
2796 			set_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
2797 			rv = -EBUSY;
2798 			goto out;
2799 		case DLM_MSG_UNLOCK:
2800 		case DLM_MSG_CANCEL:
2801 			goto out;
2802 		}
2803 		/* add_to_waiters() will set OVERLAP_CANCEL */
2804 		goto out_ok;
2805 	}
2806 
2807 	/* do we need to allow a force-unlock if there's a normal unlock
2808 	   already in progress?  in what conditions could the normal unlock
2809 	   fail such that we'd want to send a force-unlock to be sure? */
2810 
2811 	if (args->flags & DLM_LKF_FORCEUNLOCK) {
2812 		if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2813 			goto out;
2814 
2815 		if (is_overlap_unlock(lkb))
2816 			goto out;
2817 
2818 		if (test_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags)) {
2819 			set_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags);
2820 			rv = -EBUSY;
2821 			goto out;
2822 		}
2823 
2824 		switch (lkb->lkb_wait_type) {
2825 		case DLM_MSG_LOOKUP:
2826 		case DLM_MSG_REQUEST:
2827 			set_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags);
2828 			rv = -EBUSY;
2829 			goto out;
2830 		case DLM_MSG_UNLOCK:
2831 			goto out;
2832 		}
2833 		/* add_to_waiters() will set OVERLAP_UNLOCK */
2834 	}
2835 
2836  out_ok:
2837 	/* an overlapping op shouldn't blow away exflags from other op */
2838 	lkb->lkb_exflags |= args->flags;
2839 	dlm_set_sbflags_val(lkb, 0);
2840 	lkb->lkb_astparam = args->astparam;
2841 	rv = 0;
2842  out:
2843 	switch (rv) {
2844 	case 0:
2845 		break;
2846 	case -EINVAL:
2847 		/* annoy the user because dlm usage is wrong */
2848 		WARN_ON(1);
2849 		log_error(ls, "%s %d %x %x %x %x %d %s", __func__, rv,
2850 			  lkb->lkb_id, dlm_iflags_val(lkb), lkb->lkb_exflags,
2851 			  args->flags, lkb->lkb_wait_type,
2852 			  lkb->lkb_resource->res_name);
2853 		break;
2854 	default:
2855 		log_debug(ls, "%s %d %x %x %x %x %d %s", __func__, rv,
2856 			  lkb->lkb_id, dlm_iflags_val(lkb), lkb->lkb_exflags,
2857 			  args->flags, lkb->lkb_wait_type,
2858 			  lkb->lkb_resource->res_name);
2859 		break;
2860 	}
2861 
2862 	return rv;
2863 }
2864 
2865 /*
2866  * Four stage 4 varieties:
2867  * do_request(), do_convert(), do_unlock(), do_cancel()
2868  * These are called on the master node for the given lock and
2869  * from the central locking logic.
2870  */
2871 
2872 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2873 {
2874 	int error = 0;
2875 
2876 	if (can_be_granted(r, lkb, 1, 0, NULL)) {
2877 		grant_lock(r, lkb);
2878 		queue_cast(r, lkb, 0);
2879 		goto out;
2880 	}
2881 
2882 	if (can_be_queued(lkb)) {
2883 		error = -EINPROGRESS;
2884 		add_lkb(r, lkb, DLM_LKSTS_WAITING);
2885 		goto out;
2886 	}
2887 
2888 	error = -EAGAIN;
2889 	queue_cast(r, lkb, -EAGAIN);
2890  out:
2891 	return error;
2892 }
2893 
2894 static void do_request_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2895 			       int error)
2896 {
2897 	switch (error) {
2898 	case -EAGAIN:
2899 		if (force_blocking_asts(lkb))
2900 			send_blocking_asts_all(r, lkb);
2901 		break;
2902 	case -EINPROGRESS:
2903 		send_blocking_asts(r, lkb);
2904 		break;
2905 	}
2906 }
2907 
2908 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2909 {
2910 	int error = 0;
2911 	int deadlk = 0;
2912 
2913 	/* changing an existing lock may allow others to be granted */
2914 
2915 	if (can_be_granted(r, lkb, 1, 0, &deadlk)) {
2916 		grant_lock(r, lkb);
2917 		queue_cast(r, lkb, 0);
2918 		goto out;
2919 	}
2920 
2921 	/* can_be_granted() detected that this lock would block in a conversion
2922 	   deadlock, so we leave it on the granted queue and return EDEADLK in
2923 	   the ast for the convert. */
2924 
2925 	if (deadlk && !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
2926 		/* it's left on the granted queue */
2927 		revert_lock(r, lkb);
2928 		queue_cast(r, lkb, -EDEADLK);
2929 		error = -EDEADLK;
2930 		goto out;
2931 	}
2932 
2933 	/* is_demoted() means the can_be_granted() above set the grmode
2934 	   to NL, and left us on the granted queue.  This auto-demotion
2935 	   (due to CONVDEADLK) might mean other locks, and/or this lock, are
2936 	   now grantable.  We have to try to grant other converting locks
2937 	   before we try again to grant this one. */
2938 
2939 	if (is_demoted(lkb)) {
2940 		grant_pending_convert(r, DLM_LOCK_IV, NULL, NULL);
2941 		if (_can_be_granted(r, lkb, 1, 0)) {
2942 			grant_lock(r, lkb);
2943 			queue_cast(r, lkb, 0);
2944 			goto out;
2945 		}
2946 		/* else fall through and move to convert queue */
2947 	}
2948 
2949 	if (can_be_queued(lkb)) {
2950 		error = -EINPROGRESS;
2951 		del_lkb(r, lkb);
2952 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2953 		goto out;
2954 	}
2955 
2956 	error = -EAGAIN;
2957 	queue_cast(r, lkb, -EAGAIN);
2958  out:
2959 	return error;
2960 }
2961 
2962 static void do_convert_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2963 			       int error)
2964 {
2965 	switch (error) {
2966 	case 0:
2967 		grant_pending_locks(r, NULL);
2968 		/* grant_pending_locks also sends basts */
2969 		break;
2970 	case -EAGAIN:
2971 		if (force_blocking_asts(lkb))
2972 			send_blocking_asts_all(r, lkb);
2973 		break;
2974 	case -EINPROGRESS:
2975 		send_blocking_asts(r, lkb);
2976 		break;
2977 	}
2978 }
2979 
2980 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2981 {
2982 	remove_lock(r, lkb);
2983 	queue_cast(r, lkb, -DLM_EUNLOCK);
2984 	return -DLM_EUNLOCK;
2985 }
2986 
2987 static void do_unlock_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2988 			      int error)
2989 {
2990 	grant_pending_locks(r, NULL);
2991 }
2992 
2993 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
2994 
2995 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2996 {
2997 	int error;
2998 
2999 	error = revert_lock(r, lkb);
3000 	if (error) {
3001 		queue_cast(r, lkb, -DLM_ECANCEL);
3002 		return -DLM_ECANCEL;
3003 	}
3004 	return 0;
3005 }
3006 
3007 static void do_cancel_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
3008 			      int error)
3009 {
3010 	if (error)
3011 		grant_pending_locks(r, NULL);
3012 }
3013 
3014 /*
3015  * Four stage 3 varieties:
3016  * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
3017  */
3018 
3019 /* add a new lkb to a possibly new rsb, called by requesting process */
3020 
3021 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3022 {
3023 	int error;
3024 
3025 	/* set_master: sets lkb nodeid from r */
3026 
3027 	error = set_master(r, lkb);
3028 	if (error < 0)
3029 		goto out;
3030 	if (error) {
3031 		error = 0;
3032 		goto out;
3033 	}
3034 
3035 	if (is_remote(r)) {
3036 		/* receive_request() calls do_request() on remote node */
3037 		error = send_request(r, lkb);
3038 	} else {
3039 		error = do_request(r, lkb);
3040 		/* for remote locks the request_reply is sent
3041 		   between do_request and do_request_effects */
3042 		do_request_effects(r, lkb, error);
3043 	}
3044  out:
3045 	return error;
3046 }
3047 
3048 /* change some property of an existing lkb, e.g. mode */
3049 
3050 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3051 {
3052 	int error;
3053 
3054 	if (is_remote(r)) {
3055 		/* receive_convert() calls do_convert() on remote node */
3056 		error = send_convert(r, lkb);
3057 	} else {
3058 		error = do_convert(r, lkb);
3059 		/* for remote locks the convert_reply is sent
3060 		   between do_convert and do_convert_effects */
3061 		do_convert_effects(r, lkb, error);
3062 	}
3063 
3064 	return error;
3065 }
3066 
3067 /* remove an existing lkb from the granted queue */
3068 
3069 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3070 {
3071 	int error;
3072 
3073 	if (is_remote(r)) {
3074 		/* receive_unlock() calls do_unlock() on remote node */
3075 		error = send_unlock(r, lkb);
3076 	} else {
3077 		error = do_unlock(r, lkb);
3078 		/* for remote locks the unlock_reply is sent
3079 		   between do_unlock and do_unlock_effects */
3080 		do_unlock_effects(r, lkb, error);
3081 	}
3082 
3083 	return error;
3084 }
3085 
3086 /* remove an existing lkb from the convert or wait queue */
3087 
3088 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3089 {
3090 	int error;
3091 
3092 	if (is_remote(r)) {
3093 		/* receive_cancel() calls do_cancel() on remote node */
3094 		error = send_cancel(r, lkb);
3095 	} else {
3096 		error = do_cancel(r, lkb);
3097 		/* for remote locks the cancel_reply is sent
3098 		   between do_cancel and do_cancel_effects */
3099 		do_cancel_effects(r, lkb, error);
3100 	}
3101 
3102 	return error;
3103 }
3104 
3105 /*
3106  * Four stage 2 varieties:
3107  * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
3108  */
3109 
3110 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3111 			const void *name, int len,
3112 			struct dlm_args *args)
3113 {
3114 	struct dlm_rsb *r;
3115 	int error;
3116 
3117 	error = validate_lock_args(ls, lkb, args);
3118 	if (error)
3119 		return error;
3120 
3121 	error = find_rsb(ls, name, len, 0, R_REQUEST, &r);
3122 	if (error)
3123 		return error;
3124 
3125 	lock_rsb(r);
3126 
3127 	attach_lkb(r, lkb);
3128 	lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
3129 
3130 	error = _request_lock(r, lkb);
3131 
3132 	unlock_rsb(r);
3133 	put_rsb(r);
3134 	return error;
3135 }
3136 
3137 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3138 			struct dlm_args *args)
3139 {
3140 	struct dlm_rsb *r;
3141 	int error;
3142 
3143 	r = lkb->lkb_resource;
3144 
3145 	hold_rsb(r);
3146 	lock_rsb(r);
3147 
3148 	error = validate_lock_args(ls, lkb, args);
3149 	if (error)
3150 		goto out;
3151 
3152 	error = _convert_lock(r, lkb);
3153  out:
3154 	unlock_rsb(r);
3155 	put_rsb(r);
3156 	return error;
3157 }
3158 
3159 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3160 		       struct dlm_args *args)
3161 {
3162 	struct dlm_rsb *r;
3163 	int error;
3164 
3165 	r = lkb->lkb_resource;
3166 
3167 	hold_rsb(r);
3168 	lock_rsb(r);
3169 
3170 	error = validate_unlock_args(lkb, args);
3171 	if (error)
3172 		goto out;
3173 
3174 	error = _unlock_lock(r, lkb);
3175  out:
3176 	unlock_rsb(r);
3177 	put_rsb(r);
3178 	return error;
3179 }
3180 
3181 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
3182 		       struct dlm_args *args)
3183 {
3184 	struct dlm_rsb *r;
3185 	int error;
3186 
3187 	r = lkb->lkb_resource;
3188 
3189 	hold_rsb(r);
3190 	lock_rsb(r);
3191 
3192 	error = validate_unlock_args(lkb, args);
3193 	if (error)
3194 		goto out;
3195 
3196 	error = _cancel_lock(r, lkb);
3197  out:
3198 	unlock_rsb(r);
3199 	put_rsb(r);
3200 	return error;
3201 }
3202 
3203 /*
3204  * Two stage 1 varieties:  dlm_lock() and dlm_unlock()
3205  */
3206 
3207 int dlm_lock(dlm_lockspace_t *lockspace,
3208 	     int mode,
3209 	     struct dlm_lksb *lksb,
3210 	     uint32_t flags,
3211 	     const void *name,
3212 	     unsigned int namelen,
3213 	     uint32_t parent_lkid,
3214 	     void (*ast) (void *astarg),
3215 	     void *astarg,
3216 	     void (*bast) (void *astarg, int mode))
3217 {
3218 	struct dlm_ls *ls;
3219 	struct dlm_lkb *lkb;
3220 	struct dlm_args args;
3221 	int error, convert = flags & DLM_LKF_CONVERT;
3222 
3223 	ls = dlm_find_lockspace_local(lockspace);
3224 	if (!ls)
3225 		return -EINVAL;
3226 
3227 	dlm_lock_recovery(ls);
3228 
3229 	if (convert)
3230 		error = find_lkb(ls, lksb->sb_lkid, &lkb);
3231 	else
3232 		error = create_lkb(ls, &lkb);
3233 
3234 	if (error)
3235 		goto out;
3236 
3237 	trace_dlm_lock_start(ls, lkb, name, namelen, mode, flags);
3238 
3239 	error = set_lock_args(mode, lksb, flags, namelen, ast, astarg, bast,
3240 			      &args);
3241 	if (error)
3242 		goto out_put;
3243 
3244 	if (convert)
3245 		error = convert_lock(ls, lkb, &args);
3246 	else
3247 		error = request_lock(ls, lkb, name, namelen, &args);
3248 
3249 	if (error == -EINPROGRESS)
3250 		error = 0;
3251  out_put:
3252 	trace_dlm_lock_end(ls, lkb, name, namelen, mode, flags, error, true);
3253 
3254 	if (convert || error)
3255 		__put_lkb(ls, lkb);
3256 	if (error == -EAGAIN || error == -EDEADLK)
3257 		error = 0;
3258  out:
3259 	dlm_unlock_recovery(ls);
3260 	dlm_put_lockspace(ls);
3261 	return error;
3262 }
3263 
3264 int dlm_unlock(dlm_lockspace_t *lockspace,
3265 	       uint32_t lkid,
3266 	       uint32_t flags,
3267 	       struct dlm_lksb *lksb,
3268 	       void *astarg)
3269 {
3270 	struct dlm_ls *ls;
3271 	struct dlm_lkb *lkb;
3272 	struct dlm_args args;
3273 	int error;
3274 
3275 	ls = dlm_find_lockspace_local(lockspace);
3276 	if (!ls)
3277 		return -EINVAL;
3278 
3279 	dlm_lock_recovery(ls);
3280 
3281 	error = find_lkb(ls, lkid, &lkb);
3282 	if (error)
3283 		goto out;
3284 
3285 	trace_dlm_unlock_start(ls, lkb, flags);
3286 
3287 	error = set_unlock_args(flags, astarg, &args);
3288 	if (error)
3289 		goto out_put;
3290 
3291 	if (flags & DLM_LKF_CANCEL)
3292 		error = cancel_lock(ls, lkb, &args);
3293 	else
3294 		error = unlock_lock(ls, lkb, &args);
3295 
3296 	if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
3297 		error = 0;
3298 	if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
3299 		error = 0;
3300  out_put:
3301 	trace_dlm_unlock_end(ls, lkb, flags, error);
3302 
3303 	dlm_put_lkb(lkb);
3304  out:
3305 	dlm_unlock_recovery(ls);
3306 	dlm_put_lockspace(ls);
3307 	return error;
3308 }
3309 
3310 /*
3311  * send/receive routines for remote operations and replies
3312  *
3313  * send_args
3314  * send_common
3315  * send_request			receive_request
3316  * send_convert			receive_convert
3317  * send_unlock			receive_unlock
3318  * send_cancel			receive_cancel
3319  * send_grant			receive_grant
3320  * send_bast			receive_bast
3321  * send_lookup			receive_lookup
3322  * send_remove			receive_remove
3323  *
3324  * 				send_common_reply
3325  * receive_request_reply	send_request_reply
3326  * receive_convert_reply	send_convert_reply
3327  * receive_unlock_reply		send_unlock_reply
3328  * receive_cancel_reply		send_cancel_reply
3329  * receive_lookup_reply		send_lookup_reply
3330  */
3331 
3332 static int _create_message(struct dlm_ls *ls, int mb_len,
3333 			   int to_nodeid, int mstype,
3334 			   struct dlm_message **ms_ret,
3335 			   struct dlm_mhandle **mh_ret,
3336 			   gfp_t allocation)
3337 {
3338 	struct dlm_message *ms;
3339 	struct dlm_mhandle *mh;
3340 	char *mb;
3341 
3342 	/* get_buffer gives us a message handle (mh) that we need to
3343 	   pass into midcomms_commit and a message buffer (mb) that we
3344 	   write our data into */
3345 
3346 	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, allocation, &mb);
3347 	if (!mh)
3348 		return -ENOBUFS;
3349 
3350 	ms = (struct dlm_message *) mb;
3351 
3352 	ms->m_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
3353 	ms->m_header.u.h_lockspace = cpu_to_le32(ls->ls_global_id);
3354 	ms->m_header.h_nodeid = cpu_to_le32(dlm_our_nodeid());
3355 	ms->m_header.h_length = cpu_to_le16(mb_len);
3356 	ms->m_header.h_cmd = DLM_MSG;
3357 
3358 	ms->m_type = cpu_to_le32(mstype);
3359 
3360 	*mh_ret = mh;
3361 	*ms_ret = ms;
3362 	return 0;
3363 }
3364 
3365 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
3366 			  int to_nodeid, int mstype,
3367 			  struct dlm_message **ms_ret,
3368 			  struct dlm_mhandle **mh_ret,
3369 			  gfp_t allocation)
3370 {
3371 	int mb_len = sizeof(struct dlm_message);
3372 
3373 	switch (mstype) {
3374 	case DLM_MSG_REQUEST:
3375 	case DLM_MSG_LOOKUP:
3376 	case DLM_MSG_REMOVE:
3377 		mb_len += r->res_length;
3378 		break;
3379 	case DLM_MSG_CONVERT:
3380 	case DLM_MSG_UNLOCK:
3381 	case DLM_MSG_REQUEST_REPLY:
3382 	case DLM_MSG_CONVERT_REPLY:
3383 	case DLM_MSG_GRANT:
3384 		if (lkb && lkb->lkb_lvbptr && (lkb->lkb_exflags & DLM_LKF_VALBLK))
3385 			mb_len += r->res_ls->ls_lvblen;
3386 		break;
3387 	}
3388 
3389 	return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
3390 			       ms_ret, mh_ret, allocation);
3391 }
3392 
3393 /* further lowcomms enhancements or alternate implementations may make
3394    the return value from this function useful at some point */
3395 
3396 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms,
3397 			const void *name, int namelen)
3398 {
3399 	dlm_midcomms_commit_mhandle(mh, name, namelen);
3400 	return 0;
3401 }
3402 
3403 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
3404 		      struct dlm_message *ms)
3405 {
3406 	ms->m_nodeid   = cpu_to_le32(lkb->lkb_nodeid);
3407 	ms->m_pid      = cpu_to_le32(lkb->lkb_ownpid);
3408 	ms->m_lkid     = cpu_to_le32(lkb->lkb_id);
3409 	ms->m_remid    = cpu_to_le32(lkb->lkb_remid);
3410 	ms->m_exflags  = cpu_to_le32(lkb->lkb_exflags);
3411 	ms->m_sbflags  = cpu_to_le32(dlm_sbflags_val(lkb));
3412 	ms->m_flags    = cpu_to_le32(dlm_dflags_val(lkb));
3413 	ms->m_lvbseq   = cpu_to_le32(lkb->lkb_lvbseq);
3414 	ms->m_status   = cpu_to_le32(lkb->lkb_status);
3415 	ms->m_grmode   = cpu_to_le32(lkb->lkb_grmode);
3416 	ms->m_rqmode   = cpu_to_le32(lkb->lkb_rqmode);
3417 	ms->m_hash     = cpu_to_le32(r->res_hash);
3418 
3419 	/* m_result and m_bastmode are set from function args,
3420 	   not from lkb fields */
3421 
3422 	if (lkb->lkb_bastfn)
3423 		ms->m_asts |= cpu_to_le32(DLM_CB_BAST);
3424 	if (lkb->lkb_astfn)
3425 		ms->m_asts |= cpu_to_le32(DLM_CB_CAST);
3426 
3427 	/* compare with switch in create_message; send_remove() doesn't
3428 	   use send_args() */
3429 
3430 	switch (ms->m_type) {
3431 	case cpu_to_le32(DLM_MSG_REQUEST):
3432 	case cpu_to_le32(DLM_MSG_LOOKUP):
3433 		memcpy(ms->m_extra, r->res_name, r->res_length);
3434 		break;
3435 	case cpu_to_le32(DLM_MSG_CONVERT):
3436 	case cpu_to_le32(DLM_MSG_UNLOCK):
3437 	case cpu_to_le32(DLM_MSG_REQUEST_REPLY):
3438 	case cpu_to_le32(DLM_MSG_CONVERT_REPLY):
3439 	case cpu_to_le32(DLM_MSG_GRANT):
3440 		if (!lkb->lkb_lvbptr || !(lkb->lkb_exflags & DLM_LKF_VALBLK))
3441 			break;
3442 		memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
3443 		break;
3444 	}
3445 }
3446 
3447 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
3448 {
3449 	struct dlm_message *ms;
3450 	struct dlm_mhandle *mh;
3451 	int to_nodeid, error;
3452 
3453 	to_nodeid = r->res_nodeid;
3454 
3455 	error = add_to_waiters(lkb, mstype, to_nodeid);
3456 	if (error)
3457 		return error;
3458 
3459 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh, GFP_NOFS);
3460 	if (error)
3461 		goto fail;
3462 
3463 	send_args(r, lkb, ms);
3464 
3465 	error = send_message(mh, ms, r->res_name, r->res_length);
3466 	if (error)
3467 		goto fail;
3468 	return 0;
3469 
3470  fail:
3471 	remove_from_waiters(lkb, msg_reply_type(mstype));
3472 	return error;
3473 }
3474 
3475 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
3476 {
3477 	return send_common(r, lkb, DLM_MSG_REQUEST);
3478 }
3479 
3480 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
3481 {
3482 	int error;
3483 
3484 	error = send_common(r, lkb, DLM_MSG_CONVERT);
3485 
3486 	/* down conversions go without a reply from the master */
3487 	if (!error && down_conversion(lkb)) {
3488 		remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
3489 		r->res_ls->ls_local_ms.m_type = cpu_to_le32(DLM_MSG_CONVERT_REPLY);
3490 		r->res_ls->ls_local_ms.m_result = 0;
3491 		__receive_convert_reply(r, lkb, &r->res_ls->ls_local_ms, true);
3492 	}
3493 
3494 	return error;
3495 }
3496 
3497 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
3498    MASTER_UNCERTAIN to force the next request on the rsb to confirm
3499    that the master is still correct. */
3500 
3501 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
3502 {
3503 	return send_common(r, lkb, DLM_MSG_UNLOCK);
3504 }
3505 
3506 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
3507 {
3508 	return send_common(r, lkb, DLM_MSG_CANCEL);
3509 }
3510 
3511 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
3512 {
3513 	struct dlm_message *ms;
3514 	struct dlm_mhandle *mh;
3515 	int to_nodeid, error;
3516 
3517 	to_nodeid = lkb->lkb_nodeid;
3518 
3519 	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh,
3520 			       GFP_NOFS);
3521 	if (error)
3522 		goto out;
3523 
3524 	send_args(r, lkb, ms);
3525 
3526 	ms->m_result = 0;
3527 
3528 	error = send_message(mh, ms, r->res_name, r->res_length);
3529  out:
3530 	return error;
3531 }
3532 
3533 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
3534 {
3535 	struct dlm_message *ms;
3536 	struct dlm_mhandle *mh;
3537 	int to_nodeid, error;
3538 
3539 	to_nodeid = lkb->lkb_nodeid;
3540 
3541 	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh,
3542 			       GFP_NOFS);
3543 	if (error)
3544 		goto out;
3545 
3546 	send_args(r, lkb, ms);
3547 
3548 	ms->m_bastmode = cpu_to_le32(mode);
3549 
3550 	error = send_message(mh, ms, r->res_name, r->res_length);
3551  out:
3552 	return error;
3553 }
3554 
3555 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
3556 {
3557 	struct dlm_message *ms;
3558 	struct dlm_mhandle *mh;
3559 	int to_nodeid, error;
3560 
3561 	to_nodeid = dlm_dir_nodeid(r);
3562 
3563 	error = add_to_waiters(lkb, DLM_MSG_LOOKUP, to_nodeid);
3564 	if (error)
3565 		return error;
3566 
3567 	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh,
3568 			       GFP_NOFS);
3569 	if (error)
3570 		goto fail;
3571 
3572 	send_args(r, lkb, ms);
3573 
3574 	error = send_message(mh, ms, r->res_name, r->res_length);
3575 	if (error)
3576 		goto fail;
3577 	return 0;
3578 
3579  fail:
3580 	remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3581 	return error;
3582 }
3583 
3584 static int send_remove(struct dlm_rsb *r)
3585 {
3586 	struct dlm_message *ms;
3587 	struct dlm_mhandle *mh;
3588 	int to_nodeid, error;
3589 
3590 	to_nodeid = dlm_dir_nodeid(r);
3591 
3592 	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh,
3593 			       GFP_ATOMIC);
3594 	if (error)
3595 		goto out;
3596 
3597 	memcpy(ms->m_extra, r->res_name, r->res_length);
3598 	ms->m_hash = cpu_to_le32(r->res_hash);
3599 
3600 	error = send_message(mh, ms, r->res_name, r->res_length);
3601  out:
3602 	return error;
3603 }
3604 
3605 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3606 			     int mstype, int rv)
3607 {
3608 	struct dlm_message *ms;
3609 	struct dlm_mhandle *mh;
3610 	int to_nodeid, error;
3611 
3612 	to_nodeid = lkb->lkb_nodeid;
3613 
3614 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh, GFP_NOFS);
3615 	if (error)
3616 		goto out;
3617 
3618 	send_args(r, lkb, ms);
3619 
3620 	ms->m_result = cpu_to_le32(to_dlm_errno(rv));
3621 
3622 	error = send_message(mh, ms, r->res_name, r->res_length);
3623  out:
3624 	return error;
3625 }
3626 
3627 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3628 {
3629 	return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
3630 }
3631 
3632 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3633 {
3634 	return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
3635 }
3636 
3637 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3638 {
3639 	return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
3640 }
3641 
3642 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3643 {
3644 	return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
3645 }
3646 
3647 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
3648 			     int ret_nodeid, int rv)
3649 {
3650 	struct dlm_rsb *r = &ls->ls_local_rsb;
3651 	struct dlm_message *ms;
3652 	struct dlm_mhandle *mh;
3653 	int error, nodeid = le32_to_cpu(ms_in->m_header.h_nodeid);
3654 
3655 	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh,
3656 			       GFP_NOFS);
3657 	if (error)
3658 		goto out;
3659 
3660 	ms->m_lkid = ms_in->m_lkid;
3661 	ms->m_result = cpu_to_le32(to_dlm_errno(rv));
3662 	ms->m_nodeid = cpu_to_le32(ret_nodeid);
3663 
3664 	error = send_message(mh, ms, ms_in->m_extra, receive_extralen(ms_in));
3665  out:
3666 	return error;
3667 }
3668 
3669 /* which args we save from a received message depends heavily on the type
3670    of message, unlike the send side where we can safely send everything about
3671    the lkb for any type of message */
3672 
3673 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
3674 {
3675 	lkb->lkb_exflags = le32_to_cpu(ms->m_exflags);
3676 	dlm_set_sbflags_val(lkb, le32_to_cpu(ms->m_sbflags));
3677 	dlm_set_dflags_val(lkb, le32_to_cpu(ms->m_flags));
3678 }
3679 
3680 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms,
3681 				bool local)
3682 {
3683 	if (local)
3684 		return;
3685 
3686 	dlm_set_sbflags_val(lkb, le32_to_cpu(ms->m_sbflags));
3687 	dlm_set_dflags_val(lkb, le32_to_cpu(ms->m_flags));
3688 }
3689 
3690 static int receive_extralen(struct dlm_message *ms)
3691 {
3692 	return (le16_to_cpu(ms->m_header.h_length) -
3693 		sizeof(struct dlm_message));
3694 }
3695 
3696 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
3697 		       struct dlm_message *ms)
3698 {
3699 	int len;
3700 
3701 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3702 		if (!lkb->lkb_lvbptr)
3703 			lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3704 		if (!lkb->lkb_lvbptr)
3705 			return -ENOMEM;
3706 		len = receive_extralen(ms);
3707 		if (len > ls->ls_lvblen)
3708 			len = ls->ls_lvblen;
3709 		memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
3710 	}
3711 	return 0;
3712 }
3713 
3714 static void fake_bastfn(void *astparam, int mode)
3715 {
3716 	log_print("fake_bastfn should not be called");
3717 }
3718 
3719 static void fake_astfn(void *astparam)
3720 {
3721 	log_print("fake_astfn should not be called");
3722 }
3723 
3724 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3725 				struct dlm_message *ms)
3726 {
3727 	lkb->lkb_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
3728 	lkb->lkb_ownpid = le32_to_cpu(ms->m_pid);
3729 	lkb->lkb_remid = le32_to_cpu(ms->m_lkid);
3730 	lkb->lkb_grmode = DLM_LOCK_IV;
3731 	lkb->lkb_rqmode = le32_to_cpu(ms->m_rqmode);
3732 
3733 	lkb->lkb_bastfn = (ms->m_asts & cpu_to_le32(DLM_CB_BAST)) ? &fake_bastfn : NULL;
3734 	lkb->lkb_astfn = (ms->m_asts & cpu_to_le32(DLM_CB_CAST)) ? &fake_astfn : NULL;
3735 
3736 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3737 		/* lkb was just created so there won't be an lvb yet */
3738 		lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3739 		if (!lkb->lkb_lvbptr)
3740 			return -ENOMEM;
3741 	}
3742 
3743 	return 0;
3744 }
3745 
3746 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3747 				struct dlm_message *ms)
3748 {
3749 	if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3750 		return -EBUSY;
3751 
3752 	if (receive_lvb(ls, lkb, ms))
3753 		return -ENOMEM;
3754 
3755 	lkb->lkb_rqmode = le32_to_cpu(ms->m_rqmode);
3756 	lkb->lkb_lvbseq = le32_to_cpu(ms->m_lvbseq);
3757 
3758 	return 0;
3759 }
3760 
3761 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3762 			       struct dlm_message *ms)
3763 {
3764 	if (receive_lvb(ls, lkb, ms))
3765 		return -ENOMEM;
3766 	return 0;
3767 }
3768 
3769 /* We fill in the local-lkb fields with the info that send_xxxx_reply()
3770    uses to send a reply and that the remote end uses to process the reply. */
3771 
3772 static void setup_local_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3773 {
3774 	struct dlm_lkb *lkb = &ls->ls_local_lkb;
3775 	lkb->lkb_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
3776 	lkb->lkb_remid = le32_to_cpu(ms->m_lkid);
3777 }
3778 
3779 /* This is called after the rsb is locked so that we can safely inspect
3780    fields in the lkb. */
3781 
3782 static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3783 {
3784 	int from = le32_to_cpu(ms->m_header.h_nodeid);
3785 	int error = 0;
3786 
3787 	/* currently mixing of user/kernel locks are not supported */
3788 	if (ms->m_flags & cpu_to_le32(BIT(DLM_DFL_USER_BIT)) &&
3789 	    !test_bit(DLM_DFL_USER_BIT, &lkb->lkb_dflags)) {
3790 		log_error(lkb->lkb_resource->res_ls,
3791 			  "got user dlm message for a kernel lock");
3792 		error = -EINVAL;
3793 		goto out;
3794 	}
3795 
3796 	switch (ms->m_type) {
3797 	case cpu_to_le32(DLM_MSG_CONVERT):
3798 	case cpu_to_le32(DLM_MSG_UNLOCK):
3799 	case cpu_to_le32(DLM_MSG_CANCEL):
3800 		if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3801 			error = -EINVAL;
3802 		break;
3803 
3804 	case cpu_to_le32(DLM_MSG_CONVERT_REPLY):
3805 	case cpu_to_le32(DLM_MSG_UNLOCK_REPLY):
3806 	case cpu_to_le32(DLM_MSG_CANCEL_REPLY):
3807 	case cpu_to_le32(DLM_MSG_GRANT):
3808 	case cpu_to_le32(DLM_MSG_BAST):
3809 		if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3810 			error = -EINVAL;
3811 		break;
3812 
3813 	case cpu_to_le32(DLM_MSG_REQUEST_REPLY):
3814 		if (!is_process_copy(lkb))
3815 			error = -EINVAL;
3816 		else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
3817 			error = -EINVAL;
3818 		break;
3819 
3820 	default:
3821 		error = -EINVAL;
3822 	}
3823 
3824 out:
3825 	if (error)
3826 		log_error(lkb->lkb_resource->res_ls,
3827 			  "ignore invalid message %d from %d %x %x %x %d",
3828 			  le32_to_cpu(ms->m_type), from, lkb->lkb_id,
3829 			  lkb->lkb_remid, dlm_iflags_val(lkb),
3830 			  lkb->lkb_nodeid);
3831 	return error;
3832 }
3833 
3834 static int receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3835 {
3836 	struct dlm_lkb *lkb;
3837 	struct dlm_rsb *r;
3838 	int from_nodeid;
3839 	int error, namelen = 0;
3840 
3841 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
3842 
3843 	error = create_lkb(ls, &lkb);
3844 	if (error)
3845 		goto fail;
3846 
3847 	receive_flags(lkb, ms);
3848 	set_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags);
3849 	error = receive_request_args(ls, lkb, ms);
3850 	if (error) {
3851 		__put_lkb(ls, lkb);
3852 		goto fail;
3853 	}
3854 
3855 	/* The dir node is the authority on whether we are the master
3856 	   for this rsb or not, so if the master sends us a request, we should
3857 	   recreate the rsb if we've destroyed it.   This race happens when we
3858 	   send a remove message to the dir node at the same time that the dir
3859 	   node sends us a request for the rsb. */
3860 
3861 	namelen = receive_extralen(ms);
3862 
3863 	error = find_rsb(ls, ms->m_extra, namelen, from_nodeid,
3864 			 R_RECEIVE_REQUEST, &r);
3865 	if (error) {
3866 		__put_lkb(ls, lkb);
3867 		goto fail;
3868 	}
3869 
3870 	lock_rsb(r);
3871 
3872 	if (r->res_master_nodeid != dlm_our_nodeid()) {
3873 		error = validate_master_nodeid(ls, r, from_nodeid);
3874 		if (error) {
3875 			unlock_rsb(r);
3876 			put_rsb(r);
3877 			__put_lkb(ls, lkb);
3878 			goto fail;
3879 		}
3880 	}
3881 
3882 	attach_lkb(r, lkb);
3883 	error = do_request(r, lkb);
3884 	send_request_reply(r, lkb, error);
3885 	do_request_effects(r, lkb, error);
3886 
3887 	unlock_rsb(r);
3888 	put_rsb(r);
3889 
3890 	if (error == -EINPROGRESS)
3891 		error = 0;
3892 	if (error)
3893 		dlm_put_lkb(lkb);
3894 	return 0;
3895 
3896  fail:
3897 	/* TODO: instead of returning ENOTBLK, add the lkb to res_lookup
3898 	   and do this receive_request again from process_lookup_list once
3899 	   we get the lookup reply.  This would avoid a many repeated
3900 	   ENOTBLK request failures when the lookup reply designating us
3901 	   as master is delayed. */
3902 
3903 	if (error != -ENOTBLK) {
3904 		log_limit(ls, "receive_request %x from %d %d",
3905 			  le32_to_cpu(ms->m_lkid), from_nodeid, error);
3906 	}
3907 
3908 	setup_local_lkb(ls, ms);
3909 	send_request_reply(&ls->ls_local_rsb, &ls->ls_local_lkb, error);
3910 	return error;
3911 }
3912 
3913 static int receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3914 {
3915 	struct dlm_lkb *lkb;
3916 	struct dlm_rsb *r;
3917 	int error, reply = 1;
3918 
3919 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
3920 	if (error)
3921 		goto fail;
3922 
3923 	if (lkb->lkb_remid != le32_to_cpu(ms->m_lkid)) {
3924 		log_error(ls, "receive_convert %x remid %x recover_seq %llu "
3925 			  "remote %d %x", lkb->lkb_id, lkb->lkb_remid,
3926 			  (unsigned long long)lkb->lkb_recover_seq,
3927 			  le32_to_cpu(ms->m_header.h_nodeid),
3928 			  le32_to_cpu(ms->m_lkid));
3929 		error = -ENOENT;
3930 		dlm_put_lkb(lkb);
3931 		goto fail;
3932 	}
3933 
3934 	r = lkb->lkb_resource;
3935 
3936 	hold_rsb(r);
3937 	lock_rsb(r);
3938 
3939 	error = validate_message(lkb, ms);
3940 	if (error)
3941 		goto out;
3942 
3943 	receive_flags(lkb, ms);
3944 
3945 	error = receive_convert_args(ls, lkb, ms);
3946 	if (error) {
3947 		send_convert_reply(r, lkb, error);
3948 		goto out;
3949 	}
3950 
3951 	reply = !down_conversion(lkb);
3952 
3953 	error = do_convert(r, lkb);
3954 	if (reply)
3955 		send_convert_reply(r, lkb, error);
3956 	do_convert_effects(r, lkb, error);
3957  out:
3958 	unlock_rsb(r);
3959 	put_rsb(r);
3960 	dlm_put_lkb(lkb);
3961 	return 0;
3962 
3963  fail:
3964 	setup_local_lkb(ls, ms);
3965 	send_convert_reply(&ls->ls_local_rsb, &ls->ls_local_lkb, error);
3966 	return error;
3967 }
3968 
3969 static int receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3970 {
3971 	struct dlm_lkb *lkb;
3972 	struct dlm_rsb *r;
3973 	int error;
3974 
3975 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
3976 	if (error)
3977 		goto fail;
3978 
3979 	if (lkb->lkb_remid != le32_to_cpu(ms->m_lkid)) {
3980 		log_error(ls, "receive_unlock %x remid %x remote %d %x",
3981 			  lkb->lkb_id, lkb->lkb_remid,
3982 			  le32_to_cpu(ms->m_header.h_nodeid),
3983 			  le32_to_cpu(ms->m_lkid));
3984 		error = -ENOENT;
3985 		dlm_put_lkb(lkb);
3986 		goto fail;
3987 	}
3988 
3989 	r = lkb->lkb_resource;
3990 
3991 	hold_rsb(r);
3992 	lock_rsb(r);
3993 
3994 	error = validate_message(lkb, ms);
3995 	if (error)
3996 		goto out;
3997 
3998 	receive_flags(lkb, ms);
3999 
4000 	error = receive_unlock_args(ls, lkb, ms);
4001 	if (error) {
4002 		send_unlock_reply(r, lkb, error);
4003 		goto out;
4004 	}
4005 
4006 	error = do_unlock(r, lkb);
4007 	send_unlock_reply(r, lkb, error);
4008 	do_unlock_effects(r, lkb, error);
4009  out:
4010 	unlock_rsb(r);
4011 	put_rsb(r);
4012 	dlm_put_lkb(lkb);
4013 	return 0;
4014 
4015  fail:
4016 	setup_local_lkb(ls, ms);
4017 	send_unlock_reply(&ls->ls_local_rsb, &ls->ls_local_lkb, error);
4018 	return error;
4019 }
4020 
4021 static int receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
4022 {
4023 	struct dlm_lkb *lkb;
4024 	struct dlm_rsb *r;
4025 	int error;
4026 
4027 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4028 	if (error)
4029 		goto fail;
4030 
4031 	receive_flags(lkb, ms);
4032 
4033 	r = lkb->lkb_resource;
4034 
4035 	hold_rsb(r);
4036 	lock_rsb(r);
4037 
4038 	error = validate_message(lkb, ms);
4039 	if (error)
4040 		goto out;
4041 
4042 	error = do_cancel(r, lkb);
4043 	send_cancel_reply(r, lkb, error);
4044 	do_cancel_effects(r, lkb, error);
4045  out:
4046 	unlock_rsb(r);
4047 	put_rsb(r);
4048 	dlm_put_lkb(lkb);
4049 	return 0;
4050 
4051  fail:
4052 	setup_local_lkb(ls, ms);
4053 	send_cancel_reply(&ls->ls_local_rsb, &ls->ls_local_lkb, error);
4054 	return error;
4055 }
4056 
4057 static int receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
4058 {
4059 	struct dlm_lkb *lkb;
4060 	struct dlm_rsb *r;
4061 	int error;
4062 
4063 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4064 	if (error)
4065 		return error;
4066 
4067 	r = lkb->lkb_resource;
4068 
4069 	hold_rsb(r);
4070 	lock_rsb(r);
4071 
4072 	error = validate_message(lkb, ms);
4073 	if (error)
4074 		goto out;
4075 
4076 	receive_flags_reply(lkb, ms, false);
4077 	if (is_altmode(lkb))
4078 		munge_altmode(lkb, ms);
4079 	grant_lock_pc(r, lkb, ms);
4080 	queue_cast(r, lkb, 0);
4081  out:
4082 	unlock_rsb(r);
4083 	put_rsb(r);
4084 	dlm_put_lkb(lkb);
4085 	return 0;
4086 }
4087 
4088 static int receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
4089 {
4090 	struct dlm_lkb *lkb;
4091 	struct dlm_rsb *r;
4092 	int error;
4093 
4094 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4095 	if (error)
4096 		return error;
4097 
4098 	r = lkb->lkb_resource;
4099 
4100 	hold_rsb(r);
4101 	lock_rsb(r);
4102 
4103 	error = validate_message(lkb, ms);
4104 	if (error)
4105 		goto out;
4106 
4107 	queue_bast(r, lkb, le32_to_cpu(ms->m_bastmode));
4108 	lkb->lkb_highbast = le32_to_cpu(ms->m_bastmode);
4109  out:
4110 	unlock_rsb(r);
4111 	put_rsb(r);
4112 	dlm_put_lkb(lkb);
4113 	return 0;
4114 }
4115 
4116 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
4117 {
4118 	int len, error, ret_nodeid, from_nodeid, our_nodeid;
4119 
4120 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4121 	our_nodeid = dlm_our_nodeid();
4122 
4123 	len = receive_extralen(ms);
4124 
4125 	error = dlm_master_lookup(ls, from_nodeid, ms->m_extra, len, 0,
4126 				  &ret_nodeid, NULL);
4127 
4128 	/* Optimization: we're master so treat lookup as a request */
4129 	if (!error && ret_nodeid == our_nodeid) {
4130 		receive_request(ls, ms);
4131 		return;
4132 	}
4133 	send_lookup_reply(ls, ms, ret_nodeid, error);
4134 }
4135 
4136 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
4137 {
4138 	char name[DLM_RESNAME_MAXLEN+1];
4139 	struct dlm_rsb *r;
4140 	uint32_t hash, b;
4141 	int rv, len, dir_nodeid, from_nodeid;
4142 
4143 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4144 
4145 	len = receive_extralen(ms);
4146 
4147 	if (len > DLM_RESNAME_MAXLEN) {
4148 		log_error(ls, "receive_remove from %d bad len %d",
4149 			  from_nodeid, len);
4150 		return;
4151 	}
4152 
4153 	dir_nodeid = dlm_hash2nodeid(ls, le32_to_cpu(ms->m_hash));
4154 	if (dir_nodeid != dlm_our_nodeid()) {
4155 		log_error(ls, "receive_remove from %d bad nodeid %d",
4156 			  from_nodeid, dir_nodeid);
4157 		return;
4158 	}
4159 
4160 	/* Look for name on rsbtbl.toss, if it's there, kill it.
4161 	   If it's on rsbtbl.keep, it's being used, and we should ignore this
4162 	   message.  This is an expected race between the dir node sending a
4163 	   request to the master node at the same time as the master node sends
4164 	   a remove to the dir node.  The resolution to that race is for the
4165 	   dir node to ignore the remove message, and the master node to
4166 	   recreate the master rsb when it gets a request from the dir node for
4167 	   an rsb it doesn't have. */
4168 
4169 	memset(name, 0, sizeof(name));
4170 	memcpy(name, ms->m_extra, len);
4171 
4172 	hash = jhash(name, len, 0);
4173 	b = hash & (ls->ls_rsbtbl_size - 1);
4174 
4175 	spin_lock(&ls->ls_rsbtbl[b].lock);
4176 
4177 	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
4178 	if (rv) {
4179 		/* verify the rsb is on keep list per comment above */
4180 		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
4181 		if (rv) {
4182 			/* should not happen */
4183 			log_error(ls, "receive_remove from %d not found %s",
4184 				  from_nodeid, name);
4185 			spin_unlock(&ls->ls_rsbtbl[b].lock);
4186 			return;
4187 		}
4188 		if (r->res_master_nodeid != from_nodeid) {
4189 			/* should not happen */
4190 			log_error(ls, "receive_remove keep from %d master %d",
4191 				  from_nodeid, r->res_master_nodeid);
4192 			dlm_print_rsb(r);
4193 			spin_unlock(&ls->ls_rsbtbl[b].lock);
4194 			return;
4195 		}
4196 
4197 		log_debug(ls, "receive_remove from %d master %d first %x %s",
4198 			  from_nodeid, r->res_master_nodeid, r->res_first_lkid,
4199 			  name);
4200 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4201 		return;
4202 	}
4203 
4204 	if (r->res_master_nodeid != from_nodeid) {
4205 		log_error(ls, "receive_remove toss from %d master %d",
4206 			  from_nodeid, r->res_master_nodeid);
4207 		dlm_print_rsb(r);
4208 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4209 		return;
4210 	}
4211 
4212 	if (kref_put(&r->res_ref, kill_rsb)) {
4213 		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
4214 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4215 		dlm_free_rsb(r);
4216 	} else {
4217 		log_error(ls, "receive_remove from %d rsb ref error",
4218 			  from_nodeid);
4219 		dlm_print_rsb(r);
4220 		spin_unlock(&ls->ls_rsbtbl[b].lock);
4221 	}
4222 }
4223 
4224 static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
4225 {
4226 	do_purge(ls, le32_to_cpu(ms->m_nodeid), le32_to_cpu(ms->m_pid));
4227 }
4228 
4229 static int receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
4230 {
4231 	struct dlm_lkb *lkb;
4232 	struct dlm_rsb *r;
4233 	int error, mstype, result;
4234 	int from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
4235 
4236 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4237 	if (error)
4238 		return error;
4239 
4240 	r = lkb->lkb_resource;
4241 	hold_rsb(r);
4242 	lock_rsb(r);
4243 
4244 	error = validate_message(lkb, ms);
4245 	if (error)
4246 		goto out;
4247 
4248 	mstype = lkb->lkb_wait_type;
4249 	error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
4250 	if (error) {
4251 		log_error(ls, "receive_request_reply %x remote %d %x result %d",
4252 			  lkb->lkb_id, from_nodeid, le32_to_cpu(ms->m_lkid),
4253 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4254 		dlm_dump_rsb(r);
4255 		goto out;
4256 	}
4257 
4258 	/* Optimization: the dir node was also the master, so it took our
4259 	   lookup as a request and sent request reply instead of lookup reply */
4260 	if (mstype == DLM_MSG_LOOKUP) {
4261 		r->res_master_nodeid = from_nodeid;
4262 		r->res_nodeid = from_nodeid;
4263 		lkb->lkb_nodeid = from_nodeid;
4264 	}
4265 
4266 	/* this is the value returned from do_request() on the master */
4267 	result = from_dlm_errno(le32_to_cpu(ms->m_result));
4268 
4269 	switch (result) {
4270 	case -EAGAIN:
4271 		/* request would block (be queued) on remote master */
4272 		queue_cast(r, lkb, -EAGAIN);
4273 		confirm_master(r, -EAGAIN);
4274 		unhold_lkb(lkb); /* undoes create_lkb() */
4275 		break;
4276 
4277 	case -EINPROGRESS:
4278 	case 0:
4279 		/* request was queued or granted on remote master */
4280 		receive_flags_reply(lkb, ms, false);
4281 		lkb->lkb_remid = le32_to_cpu(ms->m_lkid);
4282 		if (is_altmode(lkb))
4283 			munge_altmode(lkb, ms);
4284 		if (result) {
4285 			add_lkb(r, lkb, DLM_LKSTS_WAITING);
4286 		} else {
4287 			grant_lock_pc(r, lkb, ms);
4288 			queue_cast(r, lkb, 0);
4289 		}
4290 		confirm_master(r, result);
4291 		break;
4292 
4293 	case -EBADR:
4294 	case -ENOTBLK:
4295 		/* find_rsb failed to find rsb or rsb wasn't master */
4296 		log_limit(ls, "receive_request_reply %x from %d %d "
4297 			  "master %d dir %d first %x %s", lkb->lkb_id,
4298 			  from_nodeid, result, r->res_master_nodeid,
4299 			  r->res_dir_nodeid, r->res_first_lkid, r->res_name);
4300 
4301 		if (r->res_dir_nodeid != dlm_our_nodeid() &&
4302 		    r->res_master_nodeid != dlm_our_nodeid()) {
4303 			/* cause _request_lock->set_master->send_lookup */
4304 			r->res_master_nodeid = 0;
4305 			r->res_nodeid = -1;
4306 			lkb->lkb_nodeid = -1;
4307 		}
4308 
4309 		if (is_overlap(lkb)) {
4310 			/* we'll ignore error in cancel/unlock reply */
4311 			queue_cast_overlap(r, lkb);
4312 			confirm_master(r, result);
4313 			unhold_lkb(lkb); /* undoes create_lkb() */
4314 		} else {
4315 			_request_lock(r, lkb);
4316 
4317 			if (r->res_master_nodeid == dlm_our_nodeid())
4318 				confirm_master(r, 0);
4319 		}
4320 		break;
4321 
4322 	default:
4323 		log_error(ls, "receive_request_reply %x error %d",
4324 			  lkb->lkb_id, result);
4325 	}
4326 
4327 	if ((result == 0 || result == -EINPROGRESS) &&
4328 	    test_and_clear_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags)) {
4329 		log_debug(ls, "receive_request_reply %x result %d unlock",
4330 			  lkb->lkb_id, result);
4331 		clear_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
4332 		send_unlock(r, lkb);
4333 	} else if ((result == -EINPROGRESS) &&
4334 		   test_and_clear_bit(DLM_IFL_OVERLAP_CANCEL_BIT,
4335 				      &lkb->lkb_iflags)) {
4336 		log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
4337 		clear_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags);
4338 		send_cancel(r, lkb);
4339 	} else {
4340 		clear_bit(DLM_IFL_OVERLAP_CANCEL_BIT, &lkb->lkb_iflags);
4341 		clear_bit(DLM_IFL_OVERLAP_UNLOCK_BIT, &lkb->lkb_iflags);
4342 	}
4343  out:
4344 	unlock_rsb(r);
4345 	put_rsb(r);
4346 	dlm_put_lkb(lkb);
4347 	return 0;
4348 }
4349 
4350 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
4351 				    struct dlm_message *ms, bool local)
4352 {
4353 	/* this is the value returned from do_convert() on the master */
4354 	switch (from_dlm_errno(le32_to_cpu(ms->m_result))) {
4355 	case -EAGAIN:
4356 		/* convert would block (be queued) on remote master */
4357 		queue_cast(r, lkb, -EAGAIN);
4358 		break;
4359 
4360 	case -EDEADLK:
4361 		receive_flags_reply(lkb, ms, local);
4362 		revert_lock_pc(r, lkb);
4363 		queue_cast(r, lkb, -EDEADLK);
4364 		break;
4365 
4366 	case -EINPROGRESS:
4367 		/* convert was queued on remote master */
4368 		receive_flags_reply(lkb, ms, local);
4369 		if (is_demoted(lkb))
4370 			munge_demoted(lkb);
4371 		del_lkb(r, lkb);
4372 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
4373 		break;
4374 
4375 	case 0:
4376 		/* convert was granted on remote master */
4377 		receive_flags_reply(lkb, ms, local);
4378 		if (is_demoted(lkb))
4379 			munge_demoted(lkb);
4380 		grant_lock_pc(r, lkb, ms);
4381 		queue_cast(r, lkb, 0);
4382 		break;
4383 
4384 	default:
4385 		log_error(r->res_ls, "receive_convert_reply %x remote %d %x %d",
4386 			  lkb->lkb_id, le32_to_cpu(ms->m_header.h_nodeid),
4387 			  le32_to_cpu(ms->m_lkid),
4388 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4389 		dlm_print_rsb(r);
4390 		dlm_print_lkb(lkb);
4391 	}
4392 }
4393 
4394 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms,
4395 				   bool local)
4396 {
4397 	struct dlm_rsb *r = lkb->lkb_resource;
4398 	int error;
4399 
4400 	hold_rsb(r);
4401 	lock_rsb(r);
4402 
4403 	error = validate_message(lkb, ms);
4404 	if (error)
4405 		goto out;
4406 
4407 	/* local reply can happen with waiters_mutex held */
4408 	error = remove_from_waiters_ms(lkb, ms, local);
4409 	if (error)
4410 		goto out;
4411 
4412 	__receive_convert_reply(r, lkb, ms, local);
4413  out:
4414 	unlock_rsb(r);
4415 	put_rsb(r);
4416 }
4417 
4418 static int receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
4419 {
4420 	struct dlm_lkb *lkb;
4421 	int error;
4422 
4423 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4424 	if (error)
4425 		return error;
4426 
4427 	_receive_convert_reply(lkb, ms, false);
4428 	dlm_put_lkb(lkb);
4429 	return 0;
4430 }
4431 
4432 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms,
4433 				  bool local)
4434 {
4435 	struct dlm_rsb *r = lkb->lkb_resource;
4436 	int error;
4437 
4438 	hold_rsb(r);
4439 	lock_rsb(r);
4440 
4441 	error = validate_message(lkb, ms);
4442 	if (error)
4443 		goto out;
4444 
4445 	/* local reply can happen with waiters_mutex held */
4446 	error = remove_from_waiters_ms(lkb, ms, local);
4447 	if (error)
4448 		goto out;
4449 
4450 	/* this is the value returned from do_unlock() on the master */
4451 
4452 	switch (from_dlm_errno(le32_to_cpu(ms->m_result))) {
4453 	case -DLM_EUNLOCK:
4454 		receive_flags_reply(lkb, ms, local);
4455 		remove_lock_pc(r, lkb);
4456 		queue_cast(r, lkb, -DLM_EUNLOCK);
4457 		break;
4458 	case -ENOENT:
4459 		break;
4460 	default:
4461 		log_error(r->res_ls, "receive_unlock_reply %x error %d",
4462 			  lkb->lkb_id, from_dlm_errno(le32_to_cpu(ms->m_result)));
4463 	}
4464  out:
4465 	unlock_rsb(r);
4466 	put_rsb(r);
4467 }
4468 
4469 static int receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
4470 {
4471 	struct dlm_lkb *lkb;
4472 	int error;
4473 
4474 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4475 	if (error)
4476 		return error;
4477 
4478 	_receive_unlock_reply(lkb, ms, false);
4479 	dlm_put_lkb(lkb);
4480 	return 0;
4481 }
4482 
4483 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms,
4484 				  bool local)
4485 {
4486 	struct dlm_rsb *r = lkb->lkb_resource;
4487 	int error;
4488 
4489 	hold_rsb(r);
4490 	lock_rsb(r);
4491 
4492 	error = validate_message(lkb, ms);
4493 	if (error)
4494 		goto out;
4495 
4496 	/* local reply can happen with waiters_mutex held */
4497 	error = remove_from_waiters_ms(lkb, ms, local);
4498 	if (error)
4499 		goto out;
4500 
4501 	/* this is the value returned from do_cancel() on the master */
4502 
4503 	switch (from_dlm_errno(le32_to_cpu(ms->m_result))) {
4504 	case -DLM_ECANCEL:
4505 		receive_flags_reply(lkb, ms, local);
4506 		revert_lock_pc(r, lkb);
4507 		queue_cast(r, lkb, -DLM_ECANCEL);
4508 		break;
4509 	case 0:
4510 		break;
4511 	default:
4512 		log_error(r->res_ls, "receive_cancel_reply %x error %d",
4513 			  lkb->lkb_id,
4514 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4515 	}
4516  out:
4517 	unlock_rsb(r);
4518 	put_rsb(r);
4519 }
4520 
4521 static int receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
4522 {
4523 	struct dlm_lkb *lkb;
4524 	int error;
4525 
4526 	error = find_lkb(ls, le32_to_cpu(ms->m_remid), &lkb);
4527 	if (error)
4528 		return error;
4529 
4530 	_receive_cancel_reply(lkb, ms, false);
4531 	dlm_put_lkb(lkb);
4532 	return 0;
4533 }
4534 
4535 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
4536 {
4537 	struct dlm_lkb *lkb;
4538 	struct dlm_rsb *r;
4539 	int error, ret_nodeid;
4540 	int do_lookup_list = 0;
4541 
4542 	error = find_lkb(ls, le32_to_cpu(ms->m_lkid), &lkb);
4543 	if (error) {
4544 		log_error(ls, "%s no lkid %x", __func__,
4545 			  le32_to_cpu(ms->m_lkid));
4546 		return;
4547 	}
4548 
4549 	/* ms->m_result is the value returned by dlm_master_lookup on dir node
4550 	   FIXME: will a non-zero error ever be returned? */
4551 
4552 	r = lkb->lkb_resource;
4553 	hold_rsb(r);
4554 	lock_rsb(r);
4555 
4556 	error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
4557 	if (error)
4558 		goto out;
4559 
4560 	ret_nodeid = le32_to_cpu(ms->m_nodeid);
4561 
4562 	/* We sometimes receive a request from the dir node for this
4563 	   rsb before we've received the dir node's loookup_reply for it.
4564 	   The request from the dir node implies we're the master, so we set
4565 	   ourself as master in receive_request_reply, and verify here that
4566 	   we are indeed the master. */
4567 
4568 	if (r->res_master_nodeid && (r->res_master_nodeid != ret_nodeid)) {
4569 		/* This should never happen */
4570 		log_error(ls, "receive_lookup_reply %x from %d ret %d "
4571 			  "master %d dir %d our %d first %x %s",
4572 			  lkb->lkb_id, le32_to_cpu(ms->m_header.h_nodeid),
4573 			  ret_nodeid, r->res_master_nodeid, r->res_dir_nodeid,
4574 			  dlm_our_nodeid(), r->res_first_lkid, r->res_name);
4575 	}
4576 
4577 	if (ret_nodeid == dlm_our_nodeid()) {
4578 		r->res_master_nodeid = ret_nodeid;
4579 		r->res_nodeid = 0;
4580 		do_lookup_list = 1;
4581 		r->res_first_lkid = 0;
4582 	} else if (ret_nodeid == -1) {
4583 		/* the remote node doesn't believe it's the dir node */
4584 		log_error(ls, "receive_lookup_reply %x from %d bad ret_nodeid",
4585 			  lkb->lkb_id, le32_to_cpu(ms->m_header.h_nodeid));
4586 		r->res_master_nodeid = 0;
4587 		r->res_nodeid = -1;
4588 		lkb->lkb_nodeid = -1;
4589 	} else {
4590 		/* set_master() will set lkb_nodeid from r */
4591 		r->res_master_nodeid = ret_nodeid;
4592 		r->res_nodeid = ret_nodeid;
4593 	}
4594 
4595 	if (is_overlap(lkb)) {
4596 		log_debug(ls, "receive_lookup_reply %x unlock %x",
4597 			  lkb->lkb_id, dlm_iflags_val(lkb));
4598 		queue_cast_overlap(r, lkb);
4599 		unhold_lkb(lkb); /* undoes create_lkb() */
4600 		goto out_list;
4601 	}
4602 
4603 	_request_lock(r, lkb);
4604 
4605  out_list:
4606 	if (do_lookup_list)
4607 		process_lookup_list(r);
4608  out:
4609 	unlock_rsb(r);
4610 	put_rsb(r);
4611 	dlm_put_lkb(lkb);
4612 }
4613 
4614 static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms,
4615 			     uint32_t saved_seq)
4616 {
4617 	int error = 0, noent = 0;
4618 
4619 	if (!dlm_is_member(ls, le32_to_cpu(ms->m_header.h_nodeid))) {
4620 		log_limit(ls, "receive %d from non-member %d %x %x %d",
4621 			  le32_to_cpu(ms->m_type),
4622 			  le32_to_cpu(ms->m_header.h_nodeid),
4623 			  le32_to_cpu(ms->m_lkid), le32_to_cpu(ms->m_remid),
4624 			  from_dlm_errno(le32_to_cpu(ms->m_result)));
4625 		return;
4626 	}
4627 
4628 	switch (ms->m_type) {
4629 
4630 	/* messages sent to a master node */
4631 
4632 	case cpu_to_le32(DLM_MSG_REQUEST):
4633 		error = receive_request(ls, ms);
4634 		break;
4635 
4636 	case cpu_to_le32(DLM_MSG_CONVERT):
4637 		error = receive_convert(ls, ms);
4638 		break;
4639 
4640 	case cpu_to_le32(DLM_MSG_UNLOCK):
4641 		error = receive_unlock(ls, ms);
4642 		break;
4643 
4644 	case cpu_to_le32(DLM_MSG_CANCEL):
4645 		noent = 1;
4646 		error = receive_cancel(ls, ms);
4647 		break;
4648 
4649 	/* messages sent from a master node (replies to above) */
4650 
4651 	case cpu_to_le32(DLM_MSG_REQUEST_REPLY):
4652 		error = receive_request_reply(ls, ms);
4653 		break;
4654 
4655 	case cpu_to_le32(DLM_MSG_CONVERT_REPLY):
4656 		error = receive_convert_reply(ls, ms);
4657 		break;
4658 
4659 	case cpu_to_le32(DLM_MSG_UNLOCK_REPLY):
4660 		error = receive_unlock_reply(ls, ms);
4661 		break;
4662 
4663 	case cpu_to_le32(DLM_MSG_CANCEL_REPLY):
4664 		error = receive_cancel_reply(ls, ms);
4665 		break;
4666 
4667 	/* messages sent from a master node (only two types of async msg) */
4668 
4669 	case cpu_to_le32(DLM_MSG_GRANT):
4670 		noent = 1;
4671 		error = receive_grant(ls, ms);
4672 		break;
4673 
4674 	case cpu_to_le32(DLM_MSG_BAST):
4675 		noent = 1;
4676 		error = receive_bast(ls, ms);
4677 		break;
4678 
4679 	/* messages sent to a dir node */
4680 
4681 	case cpu_to_le32(DLM_MSG_LOOKUP):
4682 		receive_lookup(ls, ms);
4683 		break;
4684 
4685 	case cpu_to_le32(DLM_MSG_REMOVE):
4686 		receive_remove(ls, ms);
4687 		break;
4688 
4689 	/* messages sent from a dir node (remove has no reply) */
4690 
4691 	case cpu_to_le32(DLM_MSG_LOOKUP_REPLY):
4692 		receive_lookup_reply(ls, ms);
4693 		break;
4694 
4695 	/* other messages */
4696 
4697 	case cpu_to_le32(DLM_MSG_PURGE):
4698 		receive_purge(ls, ms);
4699 		break;
4700 
4701 	default:
4702 		log_error(ls, "unknown message type %d",
4703 			  le32_to_cpu(ms->m_type));
4704 	}
4705 
4706 	/*
4707 	 * When checking for ENOENT, we're checking the result of
4708 	 * find_lkb(m_remid):
4709 	 *
4710 	 * The lock id referenced in the message wasn't found.  This may
4711 	 * happen in normal usage for the async messages and cancel, so
4712 	 * only use log_debug for them.
4713 	 *
4714 	 * Some errors are expected and normal.
4715 	 */
4716 
4717 	if (error == -ENOENT && noent) {
4718 		log_debug(ls, "receive %d no %x remote %d %x saved_seq %u",
4719 			  le32_to_cpu(ms->m_type), le32_to_cpu(ms->m_remid),
4720 			  le32_to_cpu(ms->m_header.h_nodeid),
4721 			  le32_to_cpu(ms->m_lkid), saved_seq);
4722 	} else if (error == -ENOENT) {
4723 		log_error(ls, "receive %d no %x remote %d %x saved_seq %u",
4724 			  le32_to_cpu(ms->m_type), le32_to_cpu(ms->m_remid),
4725 			  le32_to_cpu(ms->m_header.h_nodeid),
4726 			  le32_to_cpu(ms->m_lkid), saved_seq);
4727 
4728 		if (ms->m_type == cpu_to_le32(DLM_MSG_CONVERT))
4729 			dlm_dump_rsb_hash(ls, le32_to_cpu(ms->m_hash));
4730 	}
4731 
4732 	if (error == -EINVAL) {
4733 		log_error(ls, "receive %d inval from %d lkid %x remid %x "
4734 			  "saved_seq %u",
4735 			  le32_to_cpu(ms->m_type),
4736 			  le32_to_cpu(ms->m_header.h_nodeid),
4737 			  le32_to_cpu(ms->m_lkid), le32_to_cpu(ms->m_remid),
4738 			  saved_seq);
4739 	}
4740 }
4741 
4742 /* If the lockspace is in recovery mode (locking stopped), then normal
4743    messages are saved on the requestqueue for processing after recovery is
4744    done.  When not in recovery mode, we wait for dlm_recoverd to drain saved
4745    messages off the requestqueue before we process new ones. This occurs right
4746    after recovery completes when we transition from saving all messages on
4747    requestqueue, to processing all the saved messages, to processing new
4748    messages as they arrive. */
4749 
4750 static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
4751 				int nodeid)
4752 {
4753 	if (dlm_locking_stopped(ls)) {
4754 		/* If we were a member of this lockspace, left, and rejoined,
4755 		   other nodes may still be sending us messages from the
4756 		   lockspace generation before we left. */
4757 		if (!ls->ls_generation) {
4758 			log_limit(ls, "receive %d from %d ignore old gen",
4759 				  le32_to_cpu(ms->m_type), nodeid);
4760 			return;
4761 		}
4762 
4763 		dlm_add_requestqueue(ls, nodeid, ms);
4764 	} else {
4765 		dlm_wait_requestqueue(ls);
4766 		_receive_message(ls, ms, 0);
4767 	}
4768 }
4769 
4770 /* This is called by dlm_recoverd to process messages that were saved on
4771    the requestqueue. */
4772 
4773 void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms,
4774 			       uint32_t saved_seq)
4775 {
4776 	_receive_message(ls, ms, saved_seq);
4777 }
4778 
4779 /* This is called by the midcomms layer when something is received for
4780    the lockspace.  It could be either a MSG (normal message sent as part of
4781    standard locking activity) or an RCOM (recovery message sent as part of
4782    lockspace recovery). */
4783 
4784 void dlm_receive_buffer(union dlm_packet *p, int nodeid)
4785 {
4786 	struct dlm_header *hd = &p->header;
4787 	struct dlm_ls *ls;
4788 	int type = 0;
4789 
4790 	switch (hd->h_cmd) {
4791 	case DLM_MSG:
4792 		type = le32_to_cpu(p->message.m_type);
4793 		break;
4794 	case DLM_RCOM:
4795 		type = le32_to_cpu(p->rcom.rc_type);
4796 		break;
4797 	default:
4798 		log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
4799 		return;
4800 	}
4801 
4802 	if (le32_to_cpu(hd->h_nodeid) != nodeid) {
4803 		log_print("invalid h_nodeid %d from %d lockspace %x",
4804 			  le32_to_cpu(hd->h_nodeid), nodeid,
4805 			  le32_to_cpu(hd->u.h_lockspace));
4806 		return;
4807 	}
4808 
4809 	ls = dlm_find_lockspace_global(le32_to_cpu(hd->u.h_lockspace));
4810 	if (!ls) {
4811 		if (dlm_config.ci_log_debug) {
4812 			printk_ratelimited(KERN_DEBUG "dlm: invalid lockspace "
4813 				"%u from %d cmd %d type %d\n",
4814 				le32_to_cpu(hd->u.h_lockspace), nodeid,
4815 				hd->h_cmd, type);
4816 		}
4817 
4818 		if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
4819 			dlm_send_ls_not_ready(nodeid, &p->rcom);
4820 		return;
4821 	}
4822 
4823 	/* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
4824 	   be inactive (in this ls) before transitioning to recovery mode */
4825 
4826 	down_read(&ls->ls_recv_active);
4827 	if (hd->h_cmd == DLM_MSG)
4828 		dlm_receive_message(ls, &p->message, nodeid);
4829 	else if (hd->h_cmd == DLM_RCOM)
4830 		dlm_receive_rcom(ls, &p->rcom, nodeid);
4831 	else
4832 		log_error(ls, "invalid h_cmd %d from %d lockspace %x",
4833 			  hd->h_cmd, nodeid, le32_to_cpu(hd->u.h_lockspace));
4834 	up_read(&ls->ls_recv_active);
4835 
4836 	dlm_put_lockspace(ls);
4837 }
4838 
4839 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb,
4840 				   struct dlm_message *ms_local)
4841 {
4842 	if (middle_conversion(lkb)) {
4843 		hold_lkb(lkb);
4844 		memset(ms_local, 0, sizeof(struct dlm_message));
4845 		ms_local->m_type = cpu_to_le32(DLM_MSG_CONVERT_REPLY);
4846 		ms_local->m_result = cpu_to_le32(to_dlm_errno(-EINPROGRESS));
4847 		ms_local->m_header.h_nodeid = cpu_to_le32(lkb->lkb_nodeid);
4848 		_receive_convert_reply(lkb, ms_local, true);
4849 
4850 		/* Same special case as in receive_rcom_lock_args() */
4851 		lkb->lkb_grmode = DLM_LOCK_IV;
4852 		rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
4853 		unhold_lkb(lkb);
4854 
4855 	} else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
4856 		set_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags);
4857 	}
4858 
4859 	/* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
4860 	   conversions are async; there's no reply from the remote master */
4861 }
4862 
4863 /* A waiting lkb needs recovery if the master node has failed, or
4864    the master node is changing (only when no directory is used) */
4865 
4866 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb,
4867 				 int dir_nodeid)
4868 {
4869 	if (dlm_no_directory(ls))
4870 		return 1;
4871 
4872 	if (dlm_is_removed(ls, lkb->lkb_wait_nodeid))
4873 		return 1;
4874 
4875 	return 0;
4876 }
4877 
4878 /* Recovery for locks that are waiting for replies from nodes that are now
4879    gone.  We can just complete unlocks and cancels by faking a reply from the
4880    dead node.  Requests and up-conversions we flag to be resent after
4881    recovery.  Down-conversions can just be completed with a fake reply like
4882    unlocks.  Conversions between PR and CW need special attention. */
4883 
4884 void dlm_recover_waiters_pre(struct dlm_ls *ls)
4885 {
4886 	struct dlm_lkb *lkb, *safe;
4887 	struct dlm_message *ms_local;
4888 	int wait_type, local_unlock_result, local_cancel_result;
4889 	int dir_nodeid;
4890 
4891 	ms_local = kmalloc(sizeof(*ms_local), GFP_KERNEL);
4892 	if (!ms_local)
4893 		return;
4894 
4895 	mutex_lock(&ls->ls_waiters_mutex);
4896 
4897 	list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
4898 
4899 		dir_nodeid = dlm_dir_nodeid(lkb->lkb_resource);
4900 
4901 		/* exclude debug messages about unlocks because there can be so
4902 		   many and they aren't very interesting */
4903 
4904 		if (lkb->lkb_wait_type != DLM_MSG_UNLOCK) {
4905 			log_debug(ls, "waiter %x remote %x msg %d r_nodeid %d "
4906 				  "lkb_nodeid %d wait_nodeid %d dir_nodeid %d",
4907 				  lkb->lkb_id,
4908 				  lkb->lkb_remid,
4909 				  lkb->lkb_wait_type,
4910 				  lkb->lkb_resource->res_nodeid,
4911 				  lkb->lkb_nodeid,
4912 				  lkb->lkb_wait_nodeid,
4913 				  dir_nodeid);
4914 		}
4915 
4916 		/* all outstanding lookups, regardless of destination  will be
4917 		   resent after recovery is done */
4918 
4919 		if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
4920 			set_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags);
4921 			continue;
4922 		}
4923 
4924 		if (!waiter_needs_recovery(ls, lkb, dir_nodeid))
4925 			continue;
4926 
4927 		wait_type = lkb->lkb_wait_type;
4928 		local_unlock_result = -DLM_EUNLOCK;
4929 		local_cancel_result = -DLM_ECANCEL;
4930 
4931 		/* Main reply may have been received leaving a zero wait_type,
4932 		   but a reply for the overlapping op may not have been
4933 		   received.  In that case we need to fake the appropriate
4934 		   reply for the overlap op. */
4935 
4936 		if (!wait_type) {
4937 			if (is_overlap_cancel(lkb)) {
4938 				wait_type = DLM_MSG_CANCEL;
4939 				if (lkb->lkb_grmode == DLM_LOCK_IV)
4940 					local_cancel_result = 0;
4941 			}
4942 			if (is_overlap_unlock(lkb)) {
4943 				wait_type = DLM_MSG_UNLOCK;
4944 				if (lkb->lkb_grmode == DLM_LOCK_IV)
4945 					local_unlock_result = -ENOENT;
4946 			}
4947 
4948 			log_debug(ls, "rwpre overlap %x %x %d %d %d",
4949 				  lkb->lkb_id, dlm_iflags_val(lkb), wait_type,
4950 				  local_cancel_result, local_unlock_result);
4951 		}
4952 
4953 		switch (wait_type) {
4954 
4955 		case DLM_MSG_REQUEST:
4956 			set_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags);
4957 			break;
4958 
4959 		case DLM_MSG_CONVERT:
4960 			recover_convert_waiter(ls, lkb, ms_local);
4961 			break;
4962 
4963 		case DLM_MSG_UNLOCK:
4964 			hold_lkb(lkb);
4965 			memset(ms_local, 0, sizeof(struct dlm_message));
4966 			ms_local->m_type = cpu_to_le32(DLM_MSG_UNLOCK_REPLY);
4967 			ms_local->m_result = cpu_to_le32(to_dlm_errno(local_unlock_result));
4968 			ms_local->m_header.h_nodeid = cpu_to_le32(lkb->lkb_nodeid);
4969 			_receive_unlock_reply(lkb, ms_local, true);
4970 			dlm_put_lkb(lkb);
4971 			break;
4972 
4973 		case DLM_MSG_CANCEL:
4974 			hold_lkb(lkb);
4975 			memset(ms_local, 0, sizeof(struct dlm_message));
4976 			ms_local->m_type = cpu_to_le32(DLM_MSG_CANCEL_REPLY);
4977 			ms_local->m_result = cpu_to_le32(to_dlm_errno(local_cancel_result));
4978 			ms_local->m_header.h_nodeid = cpu_to_le32(lkb->lkb_nodeid);
4979 			_receive_cancel_reply(lkb, ms_local, true);
4980 			dlm_put_lkb(lkb);
4981 			break;
4982 
4983 		default:
4984 			log_error(ls, "invalid lkb wait_type %d %d",
4985 				  lkb->lkb_wait_type, wait_type);
4986 		}
4987 		schedule();
4988 	}
4989 	mutex_unlock(&ls->ls_waiters_mutex);
4990 	kfree(ms_local);
4991 }
4992 
4993 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
4994 {
4995 	struct dlm_lkb *lkb = NULL, *iter;
4996 
4997 	mutex_lock(&ls->ls_waiters_mutex);
4998 	list_for_each_entry(iter, &ls->ls_waiters, lkb_wait_reply) {
4999 		if (test_bit(DLM_IFL_RESEND_BIT, &iter->lkb_iflags)) {
5000 			hold_lkb(iter);
5001 			lkb = iter;
5002 			break;
5003 		}
5004 	}
5005 	mutex_unlock(&ls->ls_waiters_mutex);
5006 
5007 	return lkb;
5008 }
5009 
5010 /* Deal with lookups and lkb's marked RESEND from _pre.  We may now be the
5011    master or dir-node for r.  Processing the lkb may result in it being placed
5012    back on waiters. */
5013 
5014 /* We do this after normal locking has been enabled and any saved messages
5015    (in requestqueue) have been processed.  We should be confident that at
5016    this point we won't get or process a reply to any of these waiting
5017    operations.  But, new ops may be coming in on the rsbs/locks here from
5018    userspace or remotely. */
5019 
5020 /* there may have been an overlap unlock/cancel prior to recovery or after
5021    recovery.  if before, the lkb may still have a pos wait_count; if after, the
5022    overlap flag would just have been set and nothing new sent.  we can be
5023    confident here than any replies to either the initial op or overlap ops
5024    prior to recovery have been received. */
5025 
5026 int dlm_recover_waiters_post(struct dlm_ls *ls)
5027 {
5028 	struct dlm_lkb *lkb;
5029 	struct dlm_rsb *r;
5030 	int error = 0, mstype, err, oc, ou;
5031 
5032 	while (1) {
5033 		if (dlm_locking_stopped(ls)) {
5034 			log_debug(ls, "recover_waiters_post aborted");
5035 			error = -EINTR;
5036 			break;
5037 		}
5038 
5039 		lkb = find_resend_waiter(ls);
5040 		if (!lkb)
5041 			break;
5042 
5043 		r = lkb->lkb_resource;
5044 		hold_rsb(r);
5045 		lock_rsb(r);
5046 
5047 		mstype = lkb->lkb_wait_type;
5048 		oc = test_and_clear_bit(DLM_IFL_OVERLAP_CANCEL_BIT,
5049 					&lkb->lkb_iflags);
5050 		ou = test_and_clear_bit(DLM_IFL_OVERLAP_UNLOCK_BIT,
5051 					&lkb->lkb_iflags);
5052 		err = 0;
5053 
5054 		log_debug(ls, "waiter %x remote %x msg %d r_nodeid %d "
5055 			  "lkb_nodeid %d wait_nodeid %d dir_nodeid %d "
5056 			  "overlap %d %d", lkb->lkb_id, lkb->lkb_remid, mstype,
5057 			  r->res_nodeid, lkb->lkb_nodeid, lkb->lkb_wait_nodeid,
5058 			  dlm_dir_nodeid(r), oc, ou);
5059 
5060 		/* At this point we assume that we won't get a reply to any
5061 		   previous op or overlap op on this lock.  First, do a big
5062 		   remove_from_waiters() for all previous ops. */
5063 
5064 		clear_bit(DLM_IFL_RESEND_BIT, &lkb->lkb_iflags);
5065 		lkb->lkb_wait_type = 0;
5066 		/* drop all wait_count references we still
5067 		 * hold a reference for this iteration.
5068 		 */
5069 		while (lkb->lkb_wait_count) {
5070 			lkb->lkb_wait_count--;
5071 			unhold_lkb(lkb);
5072 		}
5073 		mutex_lock(&ls->ls_waiters_mutex);
5074 		list_del_init(&lkb->lkb_wait_reply);
5075 		mutex_unlock(&ls->ls_waiters_mutex);
5076 
5077 		if (oc || ou) {
5078 			/* do an unlock or cancel instead of resending */
5079 			switch (mstype) {
5080 			case DLM_MSG_LOOKUP:
5081 			case DLM_MSG_REQUEST:
5082 				queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
5083 							-DLM_ECANCEL);
5084 				unhold_lkb(lkb); /* undoes create_lkb() */
5085 				break;
5086 			case DLM_MSG_CONVERT:
5087 				if (oc) {
5088 					queue_cast(r, lkb, -DLM_ECANCEL);
5089 				} else {
5090 					lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
5091 					_unlock_lock(r, lkb);
5092 				}
5093 				break;
5094 			default:
5095 				err = 1;
5096 			}
5097 		} else {
5098 			switch (mstype) {
5099 			case DLM_MSG_LOOKUP:
5100 			case DLM_MSG_REQUEST:
5101 				_request_lock(r, lkb);
5102 				if (is_master(r))
5103 					confirm_master(r, 0);
5104 				break;
5105 			case DLM_MSG_CONVERT:
5106 				_convert_lock(r, lkb);
5107 				break;
5108 			default:
5109 				err = 1;
5110 			}
5111 		}
5112 
5113 		if (err) {
5114 			log_error(ls, "waiter %x msg %d r_nodeid %d "
5115 				  "dir_nodeid %d overlap %d %d",
5116 				  lkb->lkb_id, mstype, r->res_nodeid,
5117 				  dlm_dir_nodeid(r), oc, ou);
5118 		}
5119 		unlock_rsb(r);
5120 		put_rsb(r);
5121 		dlm_put_lkb(lkb);
5122 	}
5123 
5124 	return error;
5125 }
5126 
5127 static void purge_mstcpy_list(struct dlm_ls *ls, struct dlm_rsb *r,
5128 			      struct list_head *list)
5129 {
5130 	struct dlm_lkb *lkb, *safe;
5131 
5132 	list_for_each_entry_safe(lkb, safe, list, lkb_statequeue) {
5133 		if (!is_master_copy(lkb))
5134 			continue;
5135 
5136 		/* don't purge lkbs we've added in recover_master_copy for
5137 		   the current recovery seq */
5138 
5139 		if (lkb->lkb_recover_seq == ls->ls_recover_seq)
5140 			continue;
5141 
5142 		del_lkb(r, lkb);
5143 
5144 		/* this put should free the lkb */
5145 		if (!dlm_put_lkb(lkb))
5146 			log_error(ls, "purged mstcpy lkb not released");
5147 	}
5148 }
5149 
5150 void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
5151 {
5152 	struct dlm_ls *ls = r->res_ls;
5153 
5154 	purge_mstcpy_list(ls, r, &r->res_grantqueue);
5155 	purge_mstcpy_list(ls, r, &r->res_convertqueue);
5156 	purge_mstcpy_list(ls, r, &r->res_waitqueue);
5157 }
5158 
5159 static void purge_dead_list(struct dlm_ls *ls, struct dlm_rsb *r,
5160 			    struct list_head *list,
5161 			    int nodeid_gone, unsigned int *count)
5162 {
5163 	struct dlm_lkb *lkb, *safe;
5164 
5165 	list_for_each_entry_safe(lkb, safe, list, lkb_statequeue) {
5166 		if (!is_master_copy(lkb))
5167 			continue;
5168 
5169 		if ((lkb->lkb_nodeid == nodeid_gone) ||
5170 		    dlm_is_removed(ls, lkb->lkb_nodeid)) {
5171 
5172 			/* tell recover_lvb to invalidate the lvb
5173 			   because a node holding EX/PW failed */
5174 			if ((lkb->lkb_exflags & DLM_LKF_VALBLK) &&
5175 			    (lkb->lkb_grmode >= DLM_LOCK_PW)) {
5176 				rsb_set_flag(r, RSB_RECOVER_LVB_INVAL);
5177 			}
5178 
5179 			del_lkb(r, lkb);
5180 
5181 			/* this put should free the lkb */
5182 			if (!dlm_put_lkb(lkb))
5183 				log_error(ls, "purged dead lkb not released");
5184 
5185 			rsb_set_flag(r, RSB_RECOVER_GRANT);
5186 
5187 			(*count)++;
5188 		}
5189 	}
5190 }
5191 
5192 /* Get rid of locks held by nodes that are gone. */
5193 
5194 void dlm_recover_purge(struct dlm_ls *ls)
5195 {
5196 	struct dlm_rsb *r;
5197 	struct dlm_member *memb;
5198 	int nodes_count = 0;
5199 	int nodeid_gone = 0;
5200 	unsigned int lkb_count = 0;
5201 
5202 	/* cache one removed nodeid to optimize the common
5203 	   case of a single node removed */
5204 
5205 	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
5206 		nodes_count++;
5207 		nodeid_gone = memb->nodeid;
5208 	}
5209 
5210 	if (!nodes_count)
5211 		return;
5212 
5213 	down_write(&ls->ls_root_sem);
5214 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
5215 		hold_rsb(r);
5216 		lock_rsb(r);
5217 		if (is_master(r)) {
5218 			purge_dead_list(ls, r, &r->res_grantqueue,
5219 					nodeid_gone, &lkb_count);
5220 			purge_dead_list(ls, r, &r->res_convertqueue,
5221 					nodeid_gone, &lkb_count);
5222 			purge_dead_list(ls, r, &r->res_waitqueue,
5223 					nodeid_gone, &lkb_count);
5224 		}
5225 		unlock_rsb(r);
5226 		unhold_rsb(r);
5227 		cond_resched();
5228 	}
5229 	up_write(&ls->ls_root_sem);
5230 
5231 	if (lkb_count)
5232 		log_rinfo(ls, "dlm_recover_purge %u locks for %u nodes",
5233 			  lkb_count, nodes_count);
5234 }
5235 
5236 static struct dlm_rsb *find_grant_rsb(struct dlm_ls *ls, int bucket)
5237 {
5238 	struct rb_node *n;
5239 	struct dlm_rsb *r;
5240 
5241 	spin_lock(&ls->ls_rsbtbl[bucket].lock);
5242 	for (n = rb_first(&ls->ls_rsbtbl[bucket].keep); n; n = rb_next(n)) {
5243 		r = rb_entry(n, struct dlm_rsb, res_hashnode);
5244 
5245 		if (!rsb_flag(r, RSB_RECOVER_GRANT))
5246 			continue;
5247 		if (!is_master(r)) {
5248 			rsb_clear_flag(r, RSB_RECOVER_GRANT);
5249 			continue;
5250 		}
5251 		hold_rsb(r);
5252 		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
5253 		return r;
5254 	}
5255 	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
5256 	return NULL;
5257 }
5258 
5259 /*
5260  * Attempt to grant locks on resources that we are the master of.
5261  * Locks may have become grantable during recovery because locks
5262  * from departed nodes have been purged (or not rebuilt), allowing
5263  * previously blocked locks to now be granted.  The subset of rsb's
5264  * we are interested in are those with lkb's on either the convert or
5265  * waiting queues.
5266  *
5267  * Simplest would be to go through each master rsb and check for non-empty
5268  * convert or waiting queues, and attempt to grant on those rsbs.
5269  * Checking the queues requires lock_rsb, though, for which we'd need
5270  * to release the rsbtbl lock.  This would make iterating through all
5271  * rsb's very inefficient.  So, we rely on earlier recovery routines
5272  * to set RECOVER_GRANT on any rsb's that we should attempt to grant
5273  * locks for.
5274  */
5275 
5276 void dlm_recover_grant(struct dlm_ls *ls)
5277 {
5278 	struct dlm_rsb *r;
5279 	int bucket = 0;
5280 	unsigned int count = 0;
5281 	unsigned int rsb_count = 0;
5282 	unsigned int lkb_count = 0;
5283 
5284 	while (1) {
5285 		r = find_grant_rsb(ls, bucket);
5286 		if (!r) {
5287 			if (bucket == ls->ls_rsbtbl_size - 1)
5288 				break;
5289 			bucket++;
5290 			continue;
5291 		}
5292 		rsb_count++;
5293 		count = 0;
5294 		lock_rsb(r);
5295 		/* the RECOVER_GRANT flag is checked in the grant path */
5296 		grant_pending_locks(r, &count);
5297 		rsb_clear_flag(r, RSB_RECOVER_GRANT);
5298 		lkb_count += count;
5299 		confirm_master(r, 0);
5300 		unlock_rsb(r);
5301 		put_rsb(r);
5302 		cond_resched();
5303 	}
5304 
5305 	if (lkb_count)
5306 		log_rinfo(ls, "dlm_recover_grant %u locks on %u resources",
5307 			  lkb_count, rsb_count);
5308 }
5309 
5310 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
5311 					 uint32_t remid)
5312 {
5313 	struct dlm_lkb *lkb;
5314 
5315 	list_for_each_entry(lkb, head, lkb_statequeue) {
5316 		if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
5317 			return lkb;
5318 	}
5319 	return NULL;
5320 }
5321 
5322 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
5323 				    uint32_t remid)
5324 {
5325 	struct dlm_lkb *lkb;
5326 
5327 	lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
5328 	if (lkb)
5329 		return lkb;
5330 	lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
5331 	if (lkb)
5332 		return lkb;
5333 	lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
5334 	if (lkb)
5335 		return lkb;
5336 	return NULL;
5337 }
5338 
5339 /* needs at least dlm_rcom + rcom_lock */
5340 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
5341 				  struct dlm_rsb *r, struct dlm_rcom *rc)
5342 {
5343 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
5344 
5345 	lkb->lkb_nodeid = le32_to_cpu(rc->rc_header.h_nodeid);
5346 	lkb->lkb_ownpid = le32_to_cpu(rl->rl_ownpid);
5347 	lkb->lkb_remid = le32_to_cpu(rl->rl_lkid);
5348 	lkb->lkb_exflags = le32_to_cpu(rl->rl_exflags);
5349 	dlm_set_dflags_val(lkb, le32_to_cpu(rl->rl_flags));
5350 	set_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags);
5351 	lkb->lkb_lvbseq = le32_to_cpu(rl->rl_lvbseq);
5352 	lkb->lkb_rqmode = rl->rl_rqmode;
5353 	lkb->lkb_grmode = rl->rl_grmode;
5354 	/* don't set lkb_status because add_lkb wants to itself */
5355 
5356 	lkb->lkb_bastfn = (rl->rl_asts & DLM_CB_BAST) ? &fake_bastfn : NULL;
5357 	lkb->lkb_astfn = (rl->rl_asts & DLM_CB_CAST) ? &fake_astfn : NULL;
5358 
5359 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
5360 		int lvblen = le16_to_cpu(rc->rc_header.h_length) -
5361 			sizeof(struct dlm_rcom) - sizeof(struct rcom_lock);
5362 		if (lvblen > ls->ls_lvblen)
5363 			return -EINVAL;
5364 		lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
5365 		if (!lkb->lkb_lvbptr)
5366 			return -ENOMEM;
5367 		memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
5368 	}
5369 
5370 	/* Conversions between PR and CW (middle modes) need special handling.
5371 	   The real granted mode of these converting locks cannot be determined
5372 	   until all locks have been rebuilt on the rsb (recover_conversion) */
5373 
5374 	if (rl->rl_wait_type == cpu_to_le16(DLM_MSG_CONVERT) &&
5375 	    middle_conversion(lkb)) {
5376 		rl->rl_status = DLM_LKSTS_CONVERT;
5377 		lkb->lkb_grmode = DLM_LOCK_IV;
5378 		rsb_set_flag(r, RSB_RECOVER_CONVERT);
5379 	}
5380 
5381 	return 0;
5382 }
5383 
5384 /* This lkb may have been recovered in a previous aborted recovery so we need
5385    to check if the rsb already has an lkb with the given remote nodeid/lkid.
5386    If so we just send back a standard reply.  If not, we create a new lkb with
5387    the given values and send back our lkid.  We send back our lkid by sending
5388    back the rcom_lock struct we got but with the remid field filled in. */
5389 
5390 /* needs at least dlm_rcom + rcom_lock */
5391 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
5392 {
5393 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
5394 	struct dlm_rsb *r;
5395 	struct dlm_lkb *lkb;
5396 	uint32_t remid = 0;
5397 	int from_nodeid = le32_to_cpu(rc->rc_header.h_nodeid);
5398 	int error;
5399 
5400 	if (rl->rl_parent_lkid) {
5401 		error = -EOPNOTSUPP;
5402 		goto out;
5403 	}
5404 
5405 	remid = le32_to_cpu(rl->rl_lkid);
5406 
5407 	/* In general we expect the rsb returned to be R_MASTER, but we don't
5408 	   have to require it.  Recovery of masters on one node can overlap
5409 	   recovery of locks on another node, so one node can send us MSTCPY
5410 	   locks before we've made ourselves master of this rsb.  We can still
5411 	   add new MSTCPY locks that we receive here without any harm; when
5412 	   we make ourselves master, dlm_recover_masters() won't touch the
5413 	   MSTCPY locks we've received early. */
5414 
5415 	error = find_rsb(ls, rl->rl_name, le16_to_cpu(rl->rl_namelen),
5416 			 from_nodeid, R_RECEIVE_RECOVER, &r);
5417 	if (error)
5418 		goto out;
5419 
5420 	lock_rsb(r);
5421 
5422 	if (dlm_no_directory(ls) && (dlm_dir_nodeid(r) != dlm_our_nodeid())) {
5423 		log_error(ls, "dlm_recover_master_copy remote %d %x not dir",
5424 			  from_nodeid, remid);
5425 		error = -EBADR;
5426 		goto out_unlock;
5427 	}
5428 
5429 	lkb = search_remid(r, from_nodeid, remid);
5430 	if (lkb) {
5431 		error = -EEXIST;
5432 		goto out_remid;
5433 	}
5434 
5435 	error = create_lkb(ls, &lkb);
5436 	if (error)
5437 		goto out_unlock;
5438 
5439 	error = receive_rcom_lock_args(ls, lkb, r, rc);
5440 	if (error) {
5441 		__put_lkb(ls, lkb);
5442 		goto out_unlock;
5443 	}
5444 
5445 	attach_lkb(r, lkb);
5446 	add_lkb(r, lkb, rl->rl_status);
5447 	ls->ls_recover_locks_in++;
5448 
5449 	if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
5450 		rsb_set_flag(r, RSB_RECOVER_GRANT);
5451 
5452  out_remid:
5453 	/* this is the new value returned to the lock holder for
5454 	   saving in its process-copy lkb */
5455 	rl->rl_remid = cpu_to_le32(lkb->lkb_id);
5456 
5457 	lkb->lkb_recover_seq = ls->ls_recover_seq;
5458 
5459  out_unlock:
5460 	unlock_rsb(r);
5461 	put_rsb(r);
5462  out:
5463 	if (error && error != -EEXIST)
5464 		log_rinfo(ls, "dlm_recover_master_copy remote %d %x error %d",
5465 			  from_nodeid, remid, error);
5466 	rl->rl_result = cpu_to_le32(error);
5467 	return error;
5468 }
5469 
5470 /* needs at least dlm_rcom + rcom_lock */
5471 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
5472 {
5473 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
5474 	struct dlm_rsb *r;
5475 	struct dlm_lkb *lkb;
5476 	uint32_t lkid, remid;
5477 	int error, result;
5478 
5479 	lkid = le32_to_cpu(rl->rl_lkid);
5480 	remid = le32_to_cpu(rl->rl_remid);
5481 	result = le32_to_cpu(rl->rl_result);
5482 
5483 	error = find_lkb(ls, lkid, &lkb);
5484 	if (error) {
5485 		log_error(ls, "dlm_recover_process_copy no %x remote %d %x %d",
5486 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5487 			  result);
5488 		return error;
5489 	}
5490 
5491 	r = lkb->lkb_resource;
5492 	hold_rsb(r);
5493 	lock_rsb(r);
5494 
5495 	if (!is_process_copy(lkb)) {
5496 		log_error(ls, "dlm_recover_process_copy bad %x remote %d %x %d",
5497 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5498 			  result);
5499 		dlm_dump_rsb(r);
5500 		unlock_rsb(r);
5501 		put_rsb(r);
5502 		dlm_put_lkb(lkb);
5503 		return -EINVAL;
5504 	}
5505 
5506 	switch (result) {
5507 	case -EBADR:
5508 		/* There's a chance the new master received our lock before
5509 		   dlm_recover_master_reply(), this wouldn't happen if we did
5510 		   a barrier between recover_masters and recover_locks. */
5511 
5512 		log_debug(ls, "dlm_recover_process_copy %x remote %d %x %d",
5513 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5514 			  result);
5515 
5516 		dlm_send_rcom_lock(r, lkb);
5517 		goto out;
5518 	case -EEXIST:
5519 	case 0:
5520 		lkb->lkb_remid = remid;
5521 		break;
5522 	default:
5523 		log_error(ls, "dlm_recover_process_copy %x remote %d %x %d unk",
5524 			  lkid, le32_to_cpu(rc->rc_header.h_nodeid), remid,
5525 			  result);
5526 	}
5527 
5528 	/* an ack for dlm_recover_locks() which waits for replies from
5529 	   all the locks it sends to new masters */
5530 	dlm_recovered_lock(r);
5531  out:
5532 	unlock_rsb(r);
5533 	put_rsb(r);
5534 	dlm_put_lkb(lkb);
5535 
5536 	return 0;
5537 }
5538 
5539 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
5540 		     int mode, uint32_t flags, void *name, unsigned int namelen)
5541 {
5542 	struct dlm_lkb *lkb;
5543 	struct dlm_args args;
5544 	bool do_put = true;
5545 	int error;
5546 
5547 	dlm_lock_recovery(ls);
5548 
5549 	error = create_lkb(ls, &lkb);
5550 	if (error) {
5551 		kfree(ua);
5552 		goto out;
5553 	}
5554 
5555 	trace_dlm_lock_start(ls, lkb, name, namelen, mode, flags);
5556 
5557 	if (flags & DLM_LKF_VALBLK) {
5558 		ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_NOFS);
5559 		if (!ua->lksb.sb_lvbptr) {
5560 			kfree(ua);
5561 			error = -ENOMEM;
5562 			goto out_put;
5563 		}
5564 	}
5565 	error = set_lock_args(mode, &ua->lksb, flags, namelen, fake_astfn, ua,
5566 			      fake_bastfn, &args);
5567 	if (error) {
5568 		kfree(ua->lksb.sb_lvbptr);
5569 		ua->lksb.sb_lvbptr = NULL;
5570 		kfree(ua);
5571 		goto out_put;
5572 	}
5573 
5574 	/* After ua is attached to lkb it will be freed by dlm_free_lkb().
5575 	   When DLM_DFL_USER_BIT is set, the dlm knows that this is a userspace
5576 	   lock and that lkb_astparam is the dlm_user_args structure. */
5577 	set_bit(DLM_DFL_USER_BIT, &lkb->lkb_dflags);
5578 	error = request_lock(ls, lkb, name, namelen, &args);
5579 
5580 	switch (error) {
5581 	case 0:
5582 		break;
5583 	case -EINPROGRESS:
5584 		error = 0;
5585 		break;
5586 	case -EAGAIN:
5587 		error = 0;
5588 		fallthrough;
5589 	default:
5590 		goto out_put;
5591 	}
5592 
5593 	/* add this new lkb to the per-process list of locks */
5594 	spin_lock(&ua->proc->locks_spin);
5595 	hold_lkb(lkb);
5596 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
5597 	spin_unlock(&ua->proc->locks_spin);
5598 	do_put = false;
5599  out_put:
5600 	trace_dlm_lock_end(ls, lkb, name, namelen, mode, flags, error, false);
5601 	if (do_put)
5602 		__put_lkb(ls, lkb);
5603  out:
5604 	dlm_unlock_recovery(ls);
5605 	return error;
5606 }
5607 
5608 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5609 		     int mode, uint32_t flags, uint32_t lkid, char *lvb_in)
5610 {
5611 	struct dlm_lkb *lkb;
5612 	struct dlm_args args;
5613 	struct dlm_user_args *ua;
5614 	int error;
5615 
5616 	dlm_lock_recovery(ls);
5617 
5618 	error = find_lkb(ls, lkid, &lkb);
5619 	if (error)
5620 		goto out;
5621 
5622 	trace_dlm_lock_start(ls, lkb, NULL, 0, mode, flags);
5623 
5624 	/* user can change the params on its lock when it converts it, or
5625 	   add an lvb that didn't exist before */
5626 
5627 	ua = lkb->lkb_ua;
5628 
5629 	if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
5630 		ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_NOFS);
5631 		if (!ua->lksb.sb_lvbptr) {
5632 			error = -ENOMEM;
5633 			goto out_put;
5634 		}
5635 	}
5636 	if (lvb_in && ua->lksb.sb_lvbptr)
5637 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
5638 
5639 	ua->xid = ua_tmp->xid;
5640 	ua->castparam = ua_tmp->castparam;
5641 	ua->castaddr = ua_tmp->castaddr;
5642 	ua->bastparam = ua_tmp->bastparam;
5643 	ua->bastaddr = ua_tmp->bastaddr;
5644 	ua->user_lksb = ua_tmp->user_lksb;
5645 
5646 	error = set_lock_args(mode, &ua->lksb, flags, 0, fake_astfn, ua,
5647 			      fake_bastfn, &args);
5648 	if (error)
5649 		goto out_put;
5650 
5651 	error = convert_lock(ls, lkb, &args);
5652 
5653 	if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
5654 		error = 0;
5655  out_put:
5656 	trace_dlm_lock_end(ls, lkb, NULL, 0, mode, flags, error, false);
5657 	dlm_put_lkb(lkb);
5658  out:
5659 	dlm_unlock_recovery(ls);
5660 	kfree(ua_tmp);
5661 	return error;
5662 }
5663 
5664 /*
5665  * The caller asks for an orphan lock on a given resource with a given mode.
5666  * If a matching lock exists, it's moved to the owner's list of locks and
5667  * the lkid is returned.
5668  */
5669 
5670 int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5671 		     int mode, uint32_t flags, void *name, unsigned int namelen,
5672 		     uint32_t *lkid)
5673 {
5674 	struct dlm_lkb *lkb = NULL, *iter;
5675 	struct dlm_user_args *ua;
5676 	int found_other_mode = 0;
5677 	int rv = 0;
5678 
5679 	mutex_lock(&ls->ls_orphans_mutex);
5680 	list_for_each_entry(iter, &ls->ls_orphans, lkb_ownqueue) {
5681 		if (iter->lkb_resource->res_length != namelen)
5682 			continue;
5683 		if (memcmp(iter->lkb_resource->res_name, name, namelen))
5684 			continue;
5685 		if (iter->lkb_grmode != mode) {
5686 			found_other_mode = 1;
5687 			continue;
5688 		}
5689 
5690 		lkb = iter;
5691 		list_del_init(&iter->lkb_ownqueue);
5692 		clear_bit(DLM_DFL_ORPHAN_BIT, &iter->lkb_dflags);
5693 		*lkid = iter->lkb_id;
5694 		break;
5695 	}
5696 	mutex_unlock(&ls->ls_orphans_mutex);
5697 
5698 	if (!lkb && found_other_mode) {
5699 		rv = -EAGAIN;
5700 		goto out;
5701 	}
5702 
5703 	if (!lkb) {
5704 		rv = -ENOENT;
5705 		goto out;
5706 	}
5707 
5708 	lkb->lkb_exflags = flags;
5709 	lkb->lkb_ownpid = (int) current->pid;
5710 
5711 	ua = lkb->lkb_ua;
5712 
5713 	ua->proc = ua_tmp->proc;
5714 	ua->xid = ua_tmp->xid;
5715 	ua->castparam = ua_tmp->castparam;
5716 	ua->castaddr = ua_tmp->castaddr;
5717 	ua->bastparam = ua_tmp->bastparam;
5718 	ua->bastaddr = ua_tmp->bastaddr;
5719 	ua->user_lksb = ua_tmp->user_lksb;
5720 
5721 	/*
5722 	 * The lkb reference from the ls_orphans list was not
5723 	 * removed above, and is now considered the reference
5724 	 * for the proc locks list.
5725 	 */
5726 
5727 	spin_lock(&ua->proc->locks_spin);
5728 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
5729 	spin_unlock(&ua->proc->locks_spin);
5730  out:
5731 	kfree(ua_tmp);
5732 	return rv;
5733 }
5734 
5735 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5736 		    uint32_t flags, uint32_t lkid, char *lvb_in)
5737 {
5738 	struct dlm_lkb *lkb;
5739 	struct dlm_args args;
5740 	struct dlm_user_args *ua;
5741 	int error;
5742 
5743 	dlm_lock_recovery(ls);
5744 
5745 	error = find_lkb(ls, lkid, &lkb);
5746 	if (error)
5747 		goto out;
5748 
5749 	trace_dlm_unlock_start(ls, lkb, flags);
5750 
5751 	ua = lkb->lkb_ua;
5752 
5753 	if (lvb_in && ua->lksb.sb_lvbptr)
5754 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
5755 	if (ua_tmp->castparam)
5756 		ua->castparam = ua_tmp->castparam;
5757 	ua->user_lksb = ua_tmp->user_lksb;
5758 
5759 	error = set_unlock_args(flags, ua, &args);
5760 	if (error)
5761 		goto out_put;
5762 
5763 	error = unlock_lock(ls, lkb, &args);
5764 
5765 	if (error == -DLM_EUNLOCK)
5766 		error = 0;
5767 	/* from validate_unlock_args() */
5768 	if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
5769 		error = 0;
5770 	if (error)
5771 		goto out_put;
5772 
5773 	spin_lock(&ua->proc->locks_spin);
5774 	/* dlm_user_add_cb() may have already taken lkb off the proc list */
5775 	if (!list_empty(&lkb->lkb_ownqueue))
5776 		list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
5777 	spin_unlock(&ua->proc->locks_spin);
5778  out_put:
5779 	trace_dlm_unlock_end(ls, lkb, flags, error);
5780 	dlm_put_lkb(lkb);
5781  out:
5782 	dlm_unlock_recovery(ls);
5783 	kfree(ua_tmp);
5784 	return error;
5785 }
5786 
5787 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
5788 		    uint32_t flags, uint32_t lkid)
5789 {
5790 	struct dlm_lkb *lkb;
5791 	struct dlm_args args;
5792 	struct dlm_user_args *ua;
5793 	int error;
5794 
5795 	dlm_lock_recovery(ls);
5796 
5797 	error = find_lkb(ls, lkid, &lkb);
5798 	if (error)
5799 		goto out;
5800 
5801 	trace_dlm_unlock_start(ls, lkb, flags);
5802 
5803 	ua = lkb->lkb_ua;
5804 	if (ua_tmp->castparam)
5805 		ua->castparam = ua_tmp->castparam;
5806 	ua->user_lksb = ua_tmp->user_lksb;
5807 
5808 	error = set_unlock_args(flags, ua, &args);
5809 	if (error)
5810 		goto out_put;
5811 
5812 	error = cancel_lock(ls, lkb, &args);
5813 
5814 	if (error == -DLM_ECANCEL)
5815 		error = 0;
5816 	/* from validate_unlock_args() */
5817 	if (error == -EBUSY)
5818 		error = 0;
5819  out_put:
5820 	trace_dlm_unlock_end(ls, lkb, flags, error);
5821 	dlm_put_lkb(lkb);
5822  out:
5823 	dlm_unlock_recovery(ls);
5824 	kfree(ua_tmp);
5825 	return error;
5826 }
5827 
5828 int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
5829 {
5830 	struct dlm_lkb *lkb;
5831 	struct dlm_args args;
5832 	struct dlm_user_args *ua;
5833 	struct dlm_rsb *r;
5834 	int error;
5835 
5836 	dlm_lock_recovery(ls);
5837 
5838 	error = find_lkb(ls, lkid, &lkb);
5839 	if (error)
5840 		goto out;
5841 
5842 	trace_dlm_unlock_start(ls, lkb, flags);
5843 
5844 	ua = lkb->lkb_ua;
5845 
5846 	error = set_unlock_args(flags, ua, &args);
5847 	if (error)
5848 		goto out_put;
5849 
5850 	/* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
5851 
5852 	r = lkb->lkb_resource;
5853 	hold_rsb(r);
5854 	lock_rsb(r);
5855 
5856 	error = validate_unlock_args(lkb, &args);
5857 	if (error)
5858 		goto out_r;
5859 	set_bit(DLM_IFL_DEADLOCK_CANCEL_BIT, &lkb->lkb_iflags);
5860 
5861 	error = _cancel_lock(r, lkb);
5862  out_r:
5863 	unlock_rsb(r);
5864 	put_rsb(r);
5865 
5866 	if (error == -DLM_ECANCEL)
5867 		error = 0;
5868 	/* from validate_unlock_args() */
5869 	if (error == -EBUSY)
5870 		error = 0;
5871  out_put:
5872 	trace_dlm_unlock_end(ls, lkb, flags, error);
5873 	dlm_put_lkb(lkb);
5874  out:
5875 	dlm_unlock_recovery(ls);
5876 	return error;
5877 }
5878 
5879 /* lkb's that are removed from the waiters list by revert are just left on the
5880    orphans list with the granted orphan locks, to be freed by purge */
5881 
5882 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
5883 {
5884 	struct dlm_args args;
5885 	int error;
5886 
5887 	hold_lkb(lkb); /* reference for the ls_orphans list */
5888 	mutex_lock(&ls->ls_orphans_mutex);
5889 	list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
5890 	mutex_unlock(&ls->ls_orphans_mutex);
5891 
5892 	set_unlock_args(0, lkb->lkb_ua, &args);
5893 
5894 	error = cancel_lock(ls, lkb, &args);
5895 	if (error == -DLM_ECANCEL)
5896 		error = 0;
5897 	return error;
5898 }
5899 
5900 /* The FORCEUNLOCK flag allows the unlock to go ahead even if the lkb isn't
5901    granted.  Regardless of what rsb queue the lock is on, it's removed and
5902    freed.  The IVVALBLK flag causes the lvb on the resource to be invalidated
5903    if our lock is PW/EX (it's ignored if our granted mode is smaller.) */
5904 
5905 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
5906 {
5907 	struct dlm_args args;
5908 	int error;
5909 
5910 	set_unlock_args(DLM_LKF_FORCEUNLOCK | DLM_LKF_IVVALBLK,
5911 			lkb->lkb_ua, &args);
5912 
5913 	error = unlock_lock(ls, lkb, &args);
5914 	if (error == -DLM_EUNLOCK)
5915 		error = 0;
5916 	return error;
5917 }
5918 
5919 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
5920    (which does lock_rsb) due to deadlock with receiving a message that does
5921    lock_rsb followed by dlm_user_add_cb() */
5922 
5923 static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
5924 				     struct dlm_user_proc *proc)
5925 {
5926 	struct dlm_lkb *lkb = NULL;
5927 
5928 	spin_lock(&ls->ls_clear_proc_locks);
5929 	if (list_empty(&proc->locks))
5930 		goto out;
5931 
5932 	lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
5933 	list_del_init(&lkb->lkb_ownqueue);
5934 
5935 	if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
5936 		set_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags);
5937 	else
5938 		set_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags);
5939  out:
5940 	spin_unlock(&ls->ls_clear_proc_locks);
5941 	return lkb;
5942 }
5943 
5944 /* The ls_clear_proc_locks mutex protects against dlm_user_add_cb() which
5945    1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
5946    which we clear here. */
5947 
5948 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
5949    list, and no more device_writes should add lkb's to proc->locks list; so we
5950    shouldn't need to take asts_spin or locks_spin here.  this assumes that
5951    device reads/writes/closes are serialized -- FIXME: we may need to serialize
5952    them ourself. */
5953 
5954 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
5955 {
5956 	struct dlm_lkb *lkb, *safe;
5957 
5958 	dlm_lock_recovery(ls);
5959 
5960 	while (1) {
5961 		lkb = del_proc_lock(ls, proc);
5962 		if (!lkb)
5963 			break;
5964 		if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
5965 			orphan_proc_lock(ls, lkb);
5966 		else
5967 			unlock_proc_lock(ls, lkb);
5968 
5969 		/* this removes the reference for the proc->locks list
5970 		   added by dlm_user_request, it may result in the lkb
5971 		   being freed */
5972 
5973 		dlm_put_lkb(lkb);
5974 	}
5975 
5976 	spin_lock(&ls->ls_clear_proc_locks);
5977 
5978 	/* in-progress unlocks */
5979 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
5980 		list_del_init(&lkb->lkb_ownqueue);
5981 		set_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags);
5982 		dlm_put_lkb(lkb);
5983 	}
5984 
5985 	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_cb_list) {
5986 		dlm_purge_lkb_callbacks(lkb);
5987 		list_del_init(&lkb->lkb_cb_list);
5988 		dlm_put_lkb(lkb);
5989 	}
5990 
5991 	spin_unlock(&ls->ls_clear_proc_locks);
5992 	dlm_unlock_recovery(ls);
5993 }
5994 
5995 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
5996 {
5997 	struct dlm_lkb *lkb, *safe;
5998 
5999 	while (1) {
6000 		lkb = NULL;
6001 		spin_lock(&proc->locks_spin);
6002 		if (!list_empty(&proc->locks)) {
6003 			lkb = list_entry(proc->locks.next, struct dlm_lkb,
6004 					 lkb_ownqueue);
6005 			list_del_init(&lkb->lkb_ownqueue);
6006 		}
6007 		spin_unlock(&proc->locks_spin);
6008 
6009 		if (!lkb)
6010 			break;
6011 
6012 		set_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags);
6013 		unlock_proc_lock(ls, lkb);
6014 		dlm_put_lkb(lkb); /* ref from proc->locks list */
6015 	}
6016 
6017 	spin_lock(&proc->locks_spin);
6018 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
6019 		list_del_init(&lkb->lkb_ownqueue);
6020 		set_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags);
6021 		dlm_put_lkb(lkb);
6022 	}
6023 	spin_unlock(&proc->locks_spin);
6024 
6025 	spin_lock(&proc->asts_spin);
6026 	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_cb_list) {
6027 		dlm_purge_lkb_callbacks(lkb);
6028 		list_del_init(&lkb->lkb_cb_list);
6029 		dlm_put_lkb(lkb);
6030 	}
6031 	spin_unlock(&proc->asts_spin);
6032 }
6033 
6034 /* pid of 0 means purge all orphans */
6035 
6036 static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
6037 {
6038 	struct dlm_lkb *lkb, *safe;
6039 
6040 	mutex_lock(&ls->ls_orphans_mutex);
6041 	list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
6042 		if (pid && lkb->lkb_ownpid != pid)
6043 			continue;
6044 		unlock_proc_lock(ls, lkb);
6045 		list_del_init(&lkb->lkb_ownqueue);
6046 		dlm_put_lkb(lkb);
6047 	}
6048 	mutex_unlock(&ls->ls_orphans_mutex);
6049 }
6050 
6051 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
6052 {
6053 	struct dlm_message *ms;
6054 	struct dlm_mhandle *mh;
6055 	int error;
6056 
6057 	error = _create_message(ls, sizeof(struct dlm_message), nodeid,
6058 				DLM_MSG_PURGE, &ms, &mh, GFP_NOFS);
6059 	if (error)
6060 		return error;
6061 	ms->m_nodeid = cpu_to_le32(nodeid);
6062 	ms->m_pid = cpu_to_le32(pid);
6063 
6064 	return send_message(mh, ms, NULL, 0);
6065 }
6066 
6067 int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
6068 		   int nodeid, int pid)
6069 {
6070 	int error = 0;
6071 
6072 	if (nodeid && (nodeid != dlm_our_nodeid())) {
6073 		error = send_purge(ls, nodeid, pid);
6074 	} else {
6075 		dlm_lock_recovery(ls);
6076 		if (pid == current->pid)
6077 			purge_proc_locks(ls, proc);
6078 		else
6079 			do_purge(ls, nodeid, pid);
6080 		dlm_unlock_recovery(ls);
6081 	}
6082 	return error;
6083 }
6084 
6085 /* debug functionality */
6086 int dlm_debug_add_lkb(struct dlm_ls *ls, uint32_t lkb_id, char *name, int len,
6087 		      int lkb_nodeid, unsigned int lkb_dflags, int lkb_status)
6088 {
6089 	struct dlm_lksb *lksb;
6090 	struct dlm_lkb *lkb;
6091 	struct dlm_rsb *r;
6092 	int error;
6093 
6094 	/* we currently can't set a valid user lock */
6095 	if (lkb_dflags & BIT(DLM_DFL_USER_BIT))
6096 		return -EOPNOTSUPP;
6097 
6098 	lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
6099 	if (!lksb)
6100 		return -ENOMEM;
6101 
6102 	error = _create_lkb(ls, &lkb, lkb_id, lkb_id + 1);
6103 	if (error) {
6104 		kfree(lksb);
6105 		return error;
6106 	}
6107 
6108 	dlm_set_dflags_val(lkb, lkb_dflags);
6109 	lkb->lkb_nodeid = lkb_nodeid;
6110 	lkb->lkb_lksb = lksb;
6111 	/* user specific pointer, just don't have it NULL for kernel locks */
6112 	if (~lkb_dflags & BIT(DLM_DFL_USER_BIT))
6113 		lkb->lkb_astparam = (void *)0xDEADBEEF;
6114 
6115 	error = find_rsb(ls, name, len, 0, R_REQUEST, &r);
6116 	if (error) {
6117 		kfree(lksb);
6118 		__put_lkb(ls, lkb);
6119 		return error;
6120 	}
6121 
6122 	lock_rsb(r);
6123 	attach_lkb(r, lkb);
6124 	add_lkb(r, lkb, lkb_status);
6125 	unlock_rsb(r);
6126 	put_rsb(r);
6127 
6128 	return 0;
6129 }
6130 
6131 int dlm_debug_add_lkb_to_waiters(struct dlm_ls *ls, uint32_t lkb_id,
6132 				 int mstype, int to_nodeid)
6133 {
6134 	struct dlm_lkb *lkb;
6135 	int error;
6136 
6137 	error = find_lkb(ls, lkb_id, &lkb);
6138 	if (error)
6139 		return error;
6140 
6141 	error = add_to_waiters(lkb, mstype, to_nodeid);
6142 	dlm_put_lkb(lkb);
6143 	return error;
6144 }
6145 
6146