1 /******************************************************************************
2 *******************************************************************************
3 **
4 **  Copyright (C) 2005-2010 Red Hat, Inc.  All rights reserved.
5 **
6 **  This copyrighted material is made available to anyone wishing to use,
7 **  modify, copy, or redistribute it subject to the terms and conditions
8 **  of the GNU General Public License v.2.
9 **
10 *******************************************************************************
11 ******************************************************************************/
12 
13 /* Central locking logic has four stages:
14 
15    dlm_lock()
16    dlm_unlock()
17 
18    request_lock(ls, lkb)
19    convert_lock(ls, lkb)
20    unlock_lock(ls, lkb)
21    cancel_lock(ls, lkb)
22 
23    _request_lock(r, lkb)
24    _convert_lock(r, lkb)
25    _unlock_lock(r, lkb)
26    _cancel_lock(r, lkb)
27 
28    do_request(r, lkb)
29    do_convert(r, lkb)
30    do_unlock(r, lkb)
31    do_cancel(r, lkb)
32 
33    Stage 1 (lock, unlock) is mainly about checking input args and
34    splitting into one of the four main operations:
35 
36        dlm_lock          = request_lock
37        dlm_lock+CONVERT  = convert_lock
38        dlm_unlock        = unlock_lock
39        dlm_unlock+CANCEL = cancel_lock
40 
41    Stage 2, xxxx_lock(), just finds and locks the relevant rsb which is
42    provided to the next stage.
43 
44    Stage 3, _xxxx_lock(), determines if the operation is local or remote.
45    When remote, it calls send_xxxx(), when local it calls do_xxxx().
46 
47    Stage 4, do_xxxx(), is the guts of the operation.  It manipulates the
48    given rsb and lkb and queues callbacks.
49 
50    For remote operations, send_xxxx() results in the corresponding do_xxxx()
51    function being executed on the remote node.  The connecting send/receive
52    calls on local (L) and remote (R) nodes:
53 
54    L: send_xxxx()              ->  R: receive_xxxx()
55                                    R: do_xxxx()
56    L: receive_xxxx_reply()     <-  R: send_xxxx_reply()
57 */
58 #include <linux/types.h>
59 #include <linux/slab.h>
60 #include "dlm_internal.h"
61 #include <linux/dlm_device.h>
62 #include "memory.h"
63 #include "lowcomms.h"
64 #include "requestqueue.h"
65 #include "util.h"
66 #include "dir.h"
67 #include "member.h"
68 #include "lockspace.h"
69 #include "ast.h"
70 #include "lock.h"
71 #include "rcom.h"
72 #include "recover.h"
73 #include "lvb_table.h"
74 #include "user.h"
75 #include "config.h"
76 
77 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb);
78 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb);
79 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb);
80 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb);
81 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb);
82 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode);
83 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb);
84 static int send_remove(struct dlm_rsb *r);
85 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
86 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
87 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
88 				    struct dlm_message *ms);
89 static int receive_extralen(struct dlm_message *ms);
90 static void do_purge(struct dlm_ls *ls, int nodeid, int pid);
91 static void del_timeout(struct dlm_lkb *lkb);
92 
93 /*
94  * Lock compatibilty matrix - thanks Steve
95  * UN = Unlocked state. Not really a state, used as a flag
96  * PD = Padding. Used to make the matrix a nice power of two in size
97  * Other states are the same as the VMS DLM.
98  * Usage: matrix[grmode+1][rqmode+1]  (although m[rq+1][gr+1] is the same)
99  */
100 
101 static const int __dlm_compat_matrix[8][8] = {
102       /* UN NL CR CW PR PW EX PD */
103         {1, 1, 1, 1, 1, 1, 1, 0},       /* UN */
104         {1, 1, 1, 1, 1, 1, 1, 0},       /* NL */
105         {1, 1, 1, 1, 1, 1, 0, 0},       /* CR */
106         {1, 1, 1, 1, 0, 0, 0, 0},       /* CW */
107         {1, 1, 1, 0, 1, 0, 0, 0},       /* PR */
108         {1, 1, 1, 0, 0, 0, 0, 0},       /* PW */
109         {1, 1, 0, 0, 0, 0, 0, 0},       /* EX */
110         {0, 0, 0, 0, 0, 0, 0, 0}        /* PD */
111 };
112 
113 /*
114  * This defines the direction of transfer of LVB data.
115  * Granted mode is the row; requested mode is the column.
116  * Usage: matrix[grmode+1][rqmode+1]
117  * 1 = LVB is returned to the caller
118  * 0 = LVB is written to the resource
119  * -1 = nothing happens to the LVB
120  */
121 
122 const int dlm_lvb_operations[8][8] = {
123         /* UN   NL  CR  CW  PR  PW  EX  PD*/
124         {  -1,  1,  1,  1,  1,  1,  1, -1 }, /* UN */
125         {  -1,  1,  1,  1,  1,  1,  1,  0 }, /* NL */
126         {  -1, -1,  1,  1,  1,  1,  1,  0 }, /* CR */
127         {  -1, -1, -1,  1,  1,  1,  1,  0 }, /* CW */
128         {  -1, -1, -1, -1,  1,  1,  1,  0 }, /* PR */
129         {  -1,  0,  0,  0,  0,  0,  1,  0 }, /* PW */
130         {  -1,  0,  0,  0,  0,  0,  0,  0 }, /* EX */
131         {  -1,  0,  0,  0,  0,  0,  0,  0 }  /* PD */
132 };
133 
134 #define modes_compat(gr, rq) \
135 	__dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
136 
dlm_modes_compat(int mode1,int mode2)137 int dlm_modes_compat(int mode1, int mode2)
138 {
139 	return __dlm_compat_matrix[mode1 + 1][mode2 + 1];
140 }
141 
142 /*
143  * Compatibility matrix for conversions with QUECVT set.
144  * Granted mode is the row; requested mode is the column.
145  * Usage: matrix[grmode+1][rqmode+1]
146  */
147 
148 static const int __quecvt_compat_matrix[8][8] = {
149       /* UN NL CR CW PR PW EX PD */
150         {0, 0, 0, 0, 0, 0, 0, 0},       /* UN */
151         {0, 0, 1, 1, 1, 1, 1, 0},       /* NL */
152         {0, 0, 0, 1, 1, 1, 1, 0},       /* CR */
153         {0, 0, 0, 0, 1, 1, 1, 0},       /* CW */
154         {0, 0, 0, 1, 0, 1, 1, 0},       /* PR */
155         {0, 0, 0, 0, 0, 0, 1, 0},       /* PW */
156         {0, 0, 0, 0, 0, 0, 0, 0},       /* EX */
157         {0, 0, 0, 0, 0, 0, 0, 0}        /* PD */
158 };
159 
dlm_print_lkb(struct dlm_lkb * lkb)160 void dlm_print_lkb(struct dlm_lkb *lkb)
161 {
162 	printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x\n"
163 	       "     status %d rqmode %d grmode %d wait_type %d\n",
164 	       lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags,
165 	       lkb->lkb_flags, lkb->lkb_status, lkb->lkb_rqmode,
166 	       lkb->lkb_grmode, lkb->lkb_wait_type);
167 }
168 
dlm_print_rsb(struct dlm_rsb * r)169 static void dlm_print_rsb(struct dlm_rsb *r)
170 {
171 	printk(KERN_ERR "rsb: nodeid %d flags %lx first %x rlc %d name %s\n",
172 	       r->res_nodeid, r->res_flags, r->res_first_lkid,
173 	       r->res_recover_locks_count, r->res_name);
174 }
175 
dlm_dump_rsb(struct dlm_rsb * r)176 void dlm_dump_rsb(struct dlm_rsb *r)
177 {
178 	struct dlm_lkb *lkb;
179 
180 	dlm_print_rsb(r);
181 
182 	printk(KERN_ERR "rsb: root_list empty %d recover_list empty %d\n",
183 	       list_empty(&r->res_root_list), list_empty(&r->res_recover_list));
184 	printk(KERN_ERR "rsb lookup list\n");
185 	list_for_each_entry(lkb, &r->res_lookup, lkb_rsb_lookup)
186 		dlm_print_lkb(lkb);
187 	printk(KERN_ERR "rsb grant queue:\n");
188 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue)
189 		dlm_print_lkb(lkb);
190 	printk(KERN_ERR "rsb convert queue:\n");
191 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue)
192 		dlm_print_lkb(lkb);
193 	printk(KERN_ERR "rsb wait queue:\n");
194 	list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue)
195 		dlm_print_lkb(lkb);
196 }
197 
198 /* Threads cannot use the lockspace while it's being recovered */
199 
dlm_lock_recovery(struct dlm_ls * ls)200 static inline void dlm_lock_recovery(struct dlm_ls *ls)
201 {
202 	down_read(&ls->ls_in_recovery);
203 }
204 
dlm_unlock_recovery(struct dlm_ls * ls)205 void dlm_unlock_recovery(struct dlm_ls *ls)
206 {
207 	up_read(&ls->ls_in_recovery);
208 }
209 
dlm_lock_recovery_try(struct dlm_ls * ls)210 int dlm_lock_recovery_try(struct dlm_ls *ls)
211 {
212 	return down_read_trylock(&ls->ls_in_recovery);
213 }
214 
can_be_queued(struct dlm_lkb * lkb)215 static inline int can_be_queued(struct dlm_lkb *lkb)
216 {
217 	return !(lkb->lkb_exflags & DLM_LKF_NOQUEUE);
218 }
219 
force_blocking_asts(struct dlm_lkb * lkb)220 static inline int force_blocking_asts(struct dlm_lkb *lkb)
221 {
222 	return (lkb->lkb_exflags & DLM_LKF_NOQUEUEBAST);
223 }
224 
is_demoted(struct dlm_lkb * lkb)225 static inline int is_demoted(struct dlm_lkb *lkb)
226 {
227 	return (lkb->lkb_sbflags & DLM_SBF_DEMOTED);
228 }
229 
is_altmode(struct dlm_lkb * lkb)230 static inline int is_altmode(struct dlm_lkb *lkb)
231 {
232 	return (lkb->lkb_sbflags & DLM_SBF_ALTMODE);
233 }
234 
is_granted(struct dlm_lkb * lkb)235 static inline int is_granted(struct dlm_lkb *lkb)
236 {
237 	return (lkb->lkb_status == DLM_LKSTS_GRANTED);
238 }
239 
is_remote(struct dlm_rsb * r)240 static inline int is_remote(struct dlm_rsb *r)
241 {
242 	DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r););
243 	return !!r->res_nodeid;
244 }
245 
is_process_copy(struct dlm_lkb * lkb)246 static inline int is_process_copy(struct dlm_lkb *lkb)
247 {
248 	return (lkb->lkb_nodeid && !(lkb->lkb_flags & DLM_IFL_MSTCPY));
249 }
250 
is_master_copy(struct dlm_lkb * lkb)251 static inline int is_master_copy(struct dlm_lkb *lkb)
252 {
253 	if (lkb->lkb_flags & DLM_IFL_MSTCPY)
254 		DLM_ASSERT(lkb->lkb_nodeid, dlm_print_lkb(lkb););
255 	return (lkb->lkb_flags & DLM_IFL_MSTCPY) ? 1 : 0;
256 }
257 
middle_conversion(struct dlm_lkb * lkb)258 static inline int middle_conversion(struct dlm_lkb *lkb)
259 {
260 	if ((lkb->lkb_grmode==DLM_LOCK_PR && lkb->lkb_rqmode==DLM_LOCK_CW) ||
261 	    (lkb->lkb_rqmode==DLM_LOCK_PR && lkb->lkb_grmode==DLM_LOCK_CW))
262 		return 1;
263 	return 0;
264 }
265 
down_conversion(struct dlm_lkb * lkb)266 static inline int down_conversion(struct dlm_lkb *lkb)
267 {
268 	return (!middle_conversion(lkb) && lkb->lkb_rqmode < lkb->lkb_grmode);
269 }
270 
is_overlap_unlock(struct dlm_lkb * lkb)271 static inline int is_overlap_unlock(struct dlm_lkb *lkb)
272 {
273 	return lkb->lkb_flags & DLM_IFL_OVERLAP_UNLOCK;
274 }
275 
is_overlap_cancel(struct dlm_lkb * lkb)276 static inline int is_overlap_cancel(struct dlm_lkb *lkb)
277 {
278 	return lkb->lkb_flags & DLM_IFL_OVERLAP_CANCEL;
279 }
280 
is_overlap(struct dlm_lkb * lkb)281 static inline int is_overlap(struct dlm_lkb *lkb)
282 {
283 	return (lkb->lkb_flags & (DLM_IFL_OVERLAP_UNLOCK |
284 				  DLM_IFL_OVERLAP_CANCEL));
285 }
286 
queue_cast(struct dlm_rsb * r,struct dlm_lkb * lkb,int rv)287 static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
288 {
289 	if (is_master_copy(lkb))
290 		return;
291 
292 	del_timeout(lkb);
293 
294 	DLM_ASSERT(lkb->lkb_lksb, dlm_print_lkb(lkb););
295 
296 	/* if the operation was a cancel, then return -DLM_ECANCEL, if a
297 	   timeout caused the cancel then return -ETIMEDOUT */
298 	if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_TIMEOUT_CANCEL)) {
299 		lkb->lkb_flags &= ~DLM_IFL_TIMEOUT_CANCEL;
300 		rv = -ETIMEDOUT;
301 	}
302 
303 	if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_DEADLOCK_CANCEL)) {
304 		lkb->lkb_flags &= ~DLM_IFL_DEADLOCK_CANCEL;
305 		rv = -EDEADLK;
306 	}
307 
308 	dlm_add_ast(lkb, DLM_CB_CAST, lkb->lkb_grmode, rv, lkb->lkb_sbflags);
309 }
310 
queue_cast_overlap(struct dlm_rsb * r,struct dlm_lkb * lkb)311 static inline void queue_cast_overlap(struct dlm_rsb *r, struct dlm_lkb *lkb)
312 {
313 	queue_cast(r, lkb,
314 		   is_overlap_unlock(lkb) ? -DLM_EUNLOCK : -DLM_ECANCEL);
315 }
316 
queue_bast(struct dlm_rsb * r,struct dlm_lkb * lkb,int rqmode)317 static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
318 {
319 	if (is_master_copy(lkb)) {
320 		send_bast(r, lkb, rqmode);
321 	} else {
322 		dlm_add_ast(lkb, DLM_CB_BAST, rqmode, 0, 0);
323 	}
324 }
325 
326 /*
327  * Basic operations on rsb's and lkb's
328  */
329 
create_rsb(struct dlm_ls * ls,char * name,int len)330 static struct dlm_rsb *create_rsb(struct dlm_ls *ls, char *name, int len)
331 {
332 	struct dlm_rsb *r;
333 
334 	r = dlm_allocate_rsb(ls, len);
335 	if (!r)
336 		return NULL;
337 
338 	r->res_ls = ls;
339 	r->res_length = len;
340 	memcpy(r->res_name, name, len);
341 	mutex_init(&r->res_mutex);
342 
343 	INIT_LIST_HEAD(&r->res_lookup);
344 	INIT_LIST_HEAD(&r->res_grantqueue);
345 	INIT_LIST_HEAD(&r->res_convertqueue);
346 	INIT_LIST_HEAD(&r->res_waitqueue);
347 	INIT_LIST_HEAD(&r->res_root_list);
348 	INIT_LIST_HEAD(&r->res_recover_list);
349 
350 	return r;
351 }
352 
search_rsb_list(struct list_head * head,char * name,int len,unsigned int flags,struct dlm_rsb ** r_ret)353 static int search_rsb_list(struct list_head *head, char *name, int len,
354 			   unsigned int flags, struct dlm_rsb **r_ret)
355 {
356 	struct dlm_rsb *r;
357 	int error = 0;
358 
359 	list_for_each_entry(r, head, res_hashchain) {
360 		if (len == r->res_length && !memcmp(name, r->res_name, len))
361 			goto found;
362 	}
363 	*r_ret = NULL;
364 	return -EBADR;
365 
366  found:
367 	if (r->res_nodeid && (flags & R_MASTER))
368 		error = -ENOTBLK;
369 	*r_ret = r;
370 	return error;
371 }
372 
_search_rsb(struct dlm_ls * ls,char * name,int len,int b,unsigned int flags,struct dlm_rsb ** r_ret)373 static int _search_rsb(struct dlm_ls *ls, char *name, int len, int b,
374 		       unsigned int flags, struct dlm_rsb **r_ret)
375 {
376 	struct dlm_rsb *r;
377 	int error;
378 
379 	error = search_rsb_list(&ls->ls_rsbtbl[b].list, name, len, flags, &r);
380 	if (!error) {
381 		kref_get(&r->res_ref);
382 		goto out;
383 	}
384 	error = search_rsb_list(&ls->ls_rsbtbl[b].toss, name, len, flags, &r);
385 	if (error)
386 		goto out;
387 
388 	list_move(&r->res_hashchain, &ls->ls_rsbtbl[b].list);
389 
390 	if (dlm_no_directory(ls))
391 		goto out;
392 
393 	if (r->res_nodeid == -1) {
394 		rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
395 		r->res_first_lkid = 0;
396 	} else if (r->res_nodeid > 0) {
397 		rsb_set_flag(r, RSB_MASTER_UNCERTAIN);
398 		r->res_first_lkid = 0;
399 	} else {
400 		DLM_ASSERT(r->res_nodeid == 0, dlm_print_rsb(r););
401 		DLM_ASSERT(!rsb_flag(r, RSB_MASTER_UNCERTAIN),);
402 	}
403  out:
404 	*r_ret = r;
405 	return error;
406 }
407 
search_rsb(struct dlm_ls * ls,char * name,int len,int b,unsigned int flags,struct dlm_rsb ** r_ret)408 static int search_rsb(struct dlm_ls *ls, char *name, int len, int b,
409 		      unsigned int flags, struct dlm_rsb **r_ret)
410 {
411 	int error;
412 	spin_lock(&ls->ls_rsbtbl[b].lock);
413 	error = _search_rsb(ls, name, len, b, flags, r_ret);
414 	spin_unlock(&ls->ls_rsbtbl[b].lock);
415 	return error;
416 }
417 
418 /*
419  * Find rsb in rsbtbl and potentially create/add one
420  *
421  * Delaying the release of rsb's has a similar benefit to applications keeping
422  * NL locks on an rsb, but without the guarantee that the cached master value
423  * will still be valid when the rsb is reused.  Apps aren't always smart enough
424  * to keep NL locks on an rsb that they may lock again shortly; this can lead
425  * to excessive master lookups and removals if we don't delay the release.
426  *
427  * Searching for an rsb means looking through both the normal list and toss
428  * list.  When found on the toss list the rsb is moved to the normal list with
429  * ref count of 1; when found on normal list the ref count is incremented.
430  */
431 
find_rsb(struct dlm_ls * ls,char * name,int namelen,unsigned int flags,struct dlm_rsb ** r_ret)432 static int find_rsb(struct dlm_ls *ls, char *name, int namelen,
433 		    unsigned int flags, struct dlm_rsb **r_ret)
434 {
435 	struct dlm_rsb *r = NULL, *tmp;
436 	uint32_t hash, bucket;
437 	int error = -EINVAL;
438 
439 	if (namelen > DLM_RESNAME_MAXLEN)
440 		goto out;
441 
442 	if (dlm_no_directory(ls))
443 		flags |= R_CREATE;
444 
445 	error = 0;
446 	hash = jhash(name, namelen, 0);
447 	bucket = hash & (ls->ls_rsbtbl_size - 1);
448 
449 	error = search_rsb(ls, name, namelen, bucket, flags, &r);
450 	if (!error)
451 		goto out;
452 
453 	if (error == -EBADR && !(flags & R_CREATE))
454 		goto out;
455 
456 	/* the rsb was found but wasn't a master copy */
457 	if (error == -ENOTBLK)
458 		goto out;
459 
460 	error = -ENOMEM;
461 	r = create_rsb(ls, name, namelen);
462 	if (!r)
463 		goto out;
464 
465 	r->res_hash = hash;
466 	r->res_bucket = bucket;
467 	r->res_nodeid = -1;
468 	kref_init(&r->res_ref);
469 
470 	/* With no directory, the master can be set immediately */
471 	if (dlm_no_directory(ls)) {
472 		int nodeid = dlm_dir_nodeid(r);
473 		if (nodeid == dlm_our_nodeid())
474 			nodeid = 0;
475 		r->res_nodeid = nodeid;
476 	}
477 
478 	spin_lock(&ls->ls_rsbtbl[bucket].lock);
479 	error = _search_rsb(ls, name, namelen, bucket, 0, &tmp);
480 	if (!error) {
481 		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
482 		dlm_free_rsb(r);
483 		r = tmp;
484 		goto out;
485 	}
486 	list_add(&r->res_hashchain, &ls->ls_rsbtbl[bucket].list);
487 	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
488 	error = 0;
489  out:
490 	*r_ret = r;
491 	return error;
492 }
493 
494 /* This is only called to add a reference when the code already holds
495    a valid reference to the rsb, so there's no need for locking. */
496 
hold_rsb(struct dlm_rsb * r)497 static inline void hold_rsb(struct dlm_rsb *r)
498 {
499 	kref_get(&r->res_ref);
500 }
501 
dlm_hold_rsb(struct dlm_rsb * r)502 void dlm_hold_rsb(struct dlm_rsb *r)
503 {
504 	hold_rsb(r);
505 }
506 
toss_rsb(struct kref * kref)507 static void toss_rsb(struct kref *kref)
508 {
509 	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
510 	struct dlm_ls *ls = r->res_ls;
511 
512 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
513 	kref_init(&r->res_ref);
514 	list_move(&r->res_hashchain, &ls->ls_rsbtbl[r->res_bucket].toss);
515 	r->res_toss_time = jiffies;
516 	if (r->res_lvbptr) {
517 		dlm_free_lvb(r->res_lvbptr);
518 		r->res_lvbptr = NULL;
519 	}
520 }
521 
522 /* When all references to the rsb are gone it's transferred to
523    the tossed list for later disposal. */
524 
put_rsb(struct dlm_rsb * r)525 static void put_rsb(struct dlm_rsb *r)
526 {
527 	struct dlm_ls *ls = r->res_ls;
528 	uint32_t bucket = r->res_bucket;
529 
530 	spin_lock(&ls->ls_rsbtbl[bucket].lock);
531 	kref_put(&r->res_ref, toss_rsb);
532 	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
533 }
534 
dlm_put_rsb(struct dlm_rsb * r)535 void dlm_put_rsb(struct dlm_rsb *r)
536 {
537 	put_rsb(r);
538 }
539 
540 /* See comment for unhold_lkb */
541 
unhold_rsb(struct dlm_rsb * r)542 static void unhold_rsb(struct dlm_rsb *r)
543 {
544 	int rv;
545 	rv = kref_put(&r->res_ref, toss_rsb);
546 	DLM_ASSERT(!rv, dlm_dump_rsb(r););
547 }
548 
kill_rsb(struct kref * kref)549 static void kill_rsb(struct kref *kref)
550 {
551 	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
552 
553 	/* All work is done after the return from kref_put() so we
554 	   can release the write_lock before the remove and free. */
555 
556 	DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
557 	DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
558 	DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
559 	DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
560 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
561 	DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
562 }
563 
564 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
565    The rsb must exist as long as any lkb's for it do. */
566 
attach_lkb(struct dlm_rsb * r,struct dlm_lkb * lkb)567 static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
568 {
569 	hold_rsb(r);
570 	lkb->lkb_resource = r;
571 }
572 
detach_lkb(struct dlm_lkb * lkb)573 static void detach_lkb(struct dlm_lkb *lkb)
574 {
575 	if (lkb->lkb_resource) {
576 		put_rsb(lkb->lkb_resource);
577 		lkb->lkb_resource = NULL;
578 	}
579 }
580 
create_lkb(struct dlm_ls * ls,struct dlm_lkb ** lkb_ret)581 static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
582 {
583 	struct dlm_lkb *lkb, *tmp;
584 	uint32_t lkid = 0;
585 	uint16_t bucket;
586 
587 	lkb = dlm_allocate_lkb(ls);
588 	if (!lkb)
589 		return -ENOMEM;
590 
591 	lkb->lkb_nodeid = -1;
592 	lkb->lkb_grmode = DLM_LOCK_IV;
593 	kref_init(&lkb->lkb_ref);
594 	INIT_LIST_HEAD(&lkb->lkb_ownqueue);
595 	INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
596 	INIT_LIST_HEAD(&lkb->lkb_time_list);
597 	INIT_LIST_HEAD(&lkb->lkb_astqueue);
598 
599 	get_random_bytes(&bucket, sizeof(bucket));
600 	bucket &= (ls->ls_lkbtbl_size - 1);
601 
602 	write_lock(&ls->ls_lkbtbl[bucket].lock);
603 
604 	/* counter can roll over so we must verify lkid is not in use */
605 
606 	while (lkid == 0) {
607 		lkid = (bucket << 16) | ls->ls_lkbtbl[bucket].counter++;
608 
609 		list_for_each_entry(tmp, &ls->ls_lkbtbl[bucket].list,
610 				    lkb_idtbl_list) {
611 			if (tmp->lkb_id != lkid)
612 				continue;
613 			lkid = 0;
614 			break;
615 		}
616 	}
617 
618 	lkb->lkb_id = lkid;
619 	list_add(&lkb->lkb_idtbl_list, &ls->ls_lkbtbl[bucket].list);
620 	write_unlock(&ls->ls_lkbtbl[bucket].lock);
621 
622 	*lkb_ret = lkb;
623 	return 0;
624 }
625 
__find_lkb(struct dlm_ls * ls,uint32_t lkid)626 static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid)
627 {
628 	struct dlm_lkb *lkb;
629 	uint16_t bucket = (lkid >> 16);
630 
631 	list_for_each_entry(lkb, &ls->ls_lkbtbl[bucket].list, lkb_idtbl_list) {
632 		if (lkb->lkb_id == lkid)
633 			return lkb;
634 	}
635 	return NULL;
636 }
637 
find_lkb(struct dlm_ls * ls,uint32_t lkid,struct dlm_lkb ** lkb_ret)638 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
639 {
640 	struct dlm_lkb *lkb;
641 	uint16_t bucket = (lkid >> 16);
642 
643 	if (bucket >= ls->ls_lkbtbl_size)
644 		return -EBADSLT;
645 
646 	read_lock(&ls->ls_lkbtbl[bucket].lock);
647 	lkb = __find_lkb(ls, lkid);
648 	if (lkb)
649 		kref_get(&lkb->lkb_ref);
650 	read_unlock(&ls->ls_lkbtbl[bucket].lock);
651 
652 	*lkb_ret = lkb;
653 	return lkb ? 0 : -ENOENT;
654 }
655 
kill_lkb(struct kref * kref)656 static void kill_lkb(struct kref *kref)
657 {
658 	struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
659 
660 	/* All work is done after the return from kref_put() so we
661 	   can release the write_lock before the detach_lkb */
662 
663 	DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
664 }
665 
666 /* __put_lkb() is used when an lkb may not have an rsb attached to
667    it so we need to provide the lockspace explicitly */
668 
__put_lkb(struct dlm_ls * ls,struct dlm_lkb * lkb)669 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
670 {
671 	uint16_t bucket = (lkb->lkb_id >> 16);
672 
673 	write_lock(&ls->ls_lkbtbl[bucket].lock);
674 	if (kref_put(&lkb->lkb_ref, kill_lkb)) {
675 		list_del(&lkb->lkb_idtbl_list);
676 		write_unlock(&ls->ls_lkbtbl[bucket].lock);
677 
678 		detach_lkb(lkb);
679 
680 		/* for local/process lkbs, lvbptr points to caller's lksb */
681 		if (lkb->lkb_lvbptr && is_master_copy(lkb))
682 			dlm_free_lvb(lkb->lkb_lvbptr);
683 		dlm_free_lkb(lkb);
684 		return 1;
685 	} else {
686 		write_unlock(&ls->ls_lkbtbl[bucket].lock);
687 		return 0;
688 	}
689 }
690 
dlm_put_lkb(struct dlm_lkb * lkb)691 int dlm_put_lkb(struct dlm_lkb *lkb)
692 {
693 	struct dlm_ls *ls;
694 
695 	DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb););
696 	DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb););
697 
698 	ls = lkb->lkb_resource->res_ls;
699 	return __put_lkb(ls, lkb);
700 }
701 
702 /* This is only called to add a reference when the code already holds
703    a valid reference to the lkb, so there's no need for locking. */
704 
hold_lkb(struct dlm_lkb * lkb)705 static inline void hold_lkb(struct dlm_lkb *lkb)
706 {
707 	kref_get(&lkb->lkb_ref);
708 }
709 
710 /* This is called when we need to remove a reference and are certain
711    it's not the last ref.  e.g. del_lkb is always called between a
712    find_lkb/put_lkb and is always the inverse of a previous add_lkb.
713    put_lkb would work fine, but would involve unnecessary locking */
714 
unhold_lkb(struct dlm_lkb * lkb)715 static inline void unhold_lkb(struct dlm_lkb *lkb)
716 {
717 	int rv;
718 	rv = kref_put(&lkb->lkb_ref, kill_lkb);
719 	DLM_ASSERT(!rv, dlm_print_lkb(lkb););
720 }
721 
lkb_add_ordered(struct list_head * new,struct list_head * head,int mode)722 static void lkb_add_ordered(struct list_head *new, struct list_head *head,
723 			    int mode)
724 {
725 	struct dlm_lkb *lkb = NULL;
726 
727 	list_for_each_entry(lkb, head, lkb_statequeue)
728 		if (lkb->lkb_rqmode < mode)
729 			break;
730 
731 	__list_add(new, lkb->lkb_statequeue.prev, &lkb->lkb_statequeue);
732 }
733 
734 /* add/remove lkb to rsb's grant/convert/wait queue */
735 
add_lkb(struct dlm_rsb * r,struct dlm_lkb * lkb,int status)736 static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status)
737 {
738 	kref_get(&lkb->lkb_ref);
739 
740 	DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
741 
742 	lkb->lkb_timestamp = ktime_get();
743 
744 	lkb->lkb_status = status;
745 
746 	switch (status) {
747 	case DLM_LKSTS_WAITING:
748 		if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
749 			list_add(&lkb->lkb_statequeue, &r->res_waitqueue);
750 		else
751 			list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue);
752 		break;
753 	case DLM_LKSTS_GRANTED:
754 		/* convention says granted locks kept in order of grmode */
755 		lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue,
756 				lkb->lkb_grmode);
757 		break;
758 	case DLM_LKSTS_CONVERT:
759 		if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
760 			list_add(&lkb->lkb_statequeue, &r->res_convertqueue);
761 		else
762 			list_add_tail(&lkb->lkb_statequeue,
763 				      &r->res_convertqueue);
764 		break;
765 	default:
766 		DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status););
767 	}
768 }
769 
del_lkb(struct dlm_rsb * r,struct dlm_lkb * lkb)770 static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
771 {
772 	lkb->lkb_status = 0;
773 	list_del(&lkb->lkb_statequeue);
774 	unhold_lkb(lkb);
775 }
776 
move_lkb(struct dlm_rsb * r,struct dlm_lkb * lkb,int sts)777 static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
778 {
779 	hold_lkb(lkb);
780 	del_lkb(r, lkb);
781 	add_lkb(r, lkb, sts);
782 	unhold_lkb(lkb);
783 }
784 
msg_reply_type(int mstype)785 static int msg_reply_type(int mstype)
786 {
787 	switch (mstype) {
788 	case DLM_MSG_REQUEST:
789 		return DLM_MSG_REQUEST_REPLY;
790 	case DLM_MSG_CONVERT:
791 		return DLM_MSG_CONVERT_REPLY;
792 	case DLM_MSG_UNLOCK:
793 		return DLM_MSG_UNLOCK_REPLY;
794 	case DLM_MSG_CANCEL:
795 		return DLM_MSG_CANCEL_REPLY;
796 	case DLM_MSG_LOOKUP:
797 		return DLM_MSG_LOOKUP_REPLY;
798 	}
799 	return -1;
800 }
801 
802 /* add/remove lkb from global waiters list of lkb's waiting for
803    a reply from a remote node */
804 
add_to_waiters(struct dlm_lkb * lkb,int mstype)805 static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
806 {
807 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
808 	int error = 0;
809 
810 	mutex_lock(&ls->ls_waiters_mutex);
811 
812 	if (is_overlap_unlock(lkb) ||
813 	    (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
814 		error = -EINVAL;
815 		goto out;
816 	}
817 
818 	if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
819 		switch (mstype) {
820 		case DLM_MSG_UNLOCK:
821 			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
822 			break;
823 		case DLM_MSG_CANCEL:
824 			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
825 			break;
826 		default:
827 			error = -EBUSY;
828 			goto out;
829 		}
830 		lkb->lkb_wait_count++;
831 		hold_lkb(lkb);
832 
833 		log_debug(ls, "addwait %x cur %d overlap %d count %d f %x",
834 			  lkb->lkb_id, lkb->lkb_wait_type, mstype,
835 			  lkb->lkb_wait_count, lkb->lkb_flags);
836 		goto out;
837 	}
838 
839 	DLM_ASSERT(!lkb->lkb_wait_count,
840 		   dlm_print_lkb(lkb);
841 		   printk("wait_count %d\n", lkb->lkb_wait_count););
842 
843 	lkb->lkb_wait_count++;
844 	lkb->lkb_wait_type = mstype;
845 	hold_lkb(lkb);
846 	list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
847  out:
848 	if (error)
849 		log_error(ls, "addwait error %x %d flags %x %d %d %s",
850 			  lkb->lkb_id, error, lkb->lkb_flags, mstype,
851 			  lkb->lkb_wait_type, lkb->lkb_resource->res_name);
852 	mutex_unlock(&ls->ls_waiters_mutex);
853 	return error;
854 }
855 
856 /* We clear the RESEND flag because we might be taking an lkb off the waiters
857    list as part of process_requestqueue (e.g. a lookup that has an optimized
858    request reply on the requestqueue) between dlm_recover_waiters_pre() which
859    set RESEND and dlm_recover_waiters_post() */
860 
_remove_from_waiters(struct dlm_lkb * lkb,int mstype,struct dlm_message * ms)861 static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype,
862 				struct dlm_message *ms)
863 {
864 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
865 	int overlap_done = 0;
866 
867 	if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
868 		log_debug(ls, "remwait %x unlock_reply overlap", lkb->lkb_id);
869 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
870 		overlap_done = 1;
871 		goto out_del;
872 	}
873 
874 	if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
875 		log_debug(ls, "remwait %x cancel_reply overlap", lkb->lkb_id);
876 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
877 		overlap_done = 1;
878 		goto out_del;
879 	}
880 
881 	/* Cancel state was preemptively cleared by a successful convert,
882 	   see next comment, nothing to do. */
883 
884 	if ((mstype == DLM_MSG_CANCEL_REPLY) &&
885 	    (lkb->lkb_wait_type != DLM_MSG_CANCEL)) {
886 		log_debug(ls, "remwait %x cancel_reply wait_type %d",
887 			  lkb->lkb_id, lkb->lkb_wait_type);
888 		return -1;
889 	}
890 
891 	/* Remove for the convert reply, and premptively remove for the
892 	   cancel reply.  A convert has been granted while there's still
893 	   an outstanding cancel on it (the cancel is moot and the result
894 	   in the cancel reply should be 0).  We preempt the cancel reply
895 	   because the app gets the convert result and then can follow up
896 	   with another op, like convert.  This subsequent op would see the
897 	   lingering state of the cancel and fail with -EBUSY. */
898 
899 	if ((mstype == DLM_MSG_CONVERT_REPLY) &&
900 	    (lkb->lkb_wait_type == DLM_MSG_CONVERT) &&
901 	    is_overlap_cancel(lkb) && ms && !ms->m_result) {
902 		log_debug(ls, "remwait %x convert_reply zap overlap_cancel",
903 			  lkb->lkb_id);
904 		lkb->lkb_wait_type = 0;
905 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
906 		lkb->lkb_wait_count--;
907 		goto out_del;
908 	}
909 
910 	/* N.B. type of reply may not always correspond to type of original
911 	   msg due to lookup->request optimization, verify others? */
912 
913 	if (lkb->lkb_wait_type) {
914 		lkb->lkb_wait_type = 0;
915 		goto out_del;
916 	}
917 
918 	log_error(ls, "remwait error %x reply %d flags %x no wait_type",
919 		  lkb->lkb_id, mstype, lkb->lkb_flags);
920 	return -1;
921 
922  out_del:
923 	/* the force-unlock/cancel has completed and we haven't recvd a reply
924 	   to the op that was in progress prior to the unlock/cancel; we
925 	   give up on any reply to the earlier op.  FIXME: not sure when/how
926 	   this would happen */
927 
928 	if (overlap_done && lkb->lkb_wait_type) {
929 		log_error(ls, "remwait error %x reply %d wait_type %d overlap",
930 			  lkb->lkb_id, mstype, lkb->lkb_wait_type);
931 		lkb->lkb_wait_count--;
932 		lkb->lkb_wait_type = 0;
933 	}
934 
935 	DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
936 
937 	lkb->lkb_flags &= ~DLM_IFL_RESEND;
938 	lkb->lkb_wait_count--;
939 	if (!lkb->lkb_wait_count)
940 		list_del_init(&lkb->lkb_wait_reply);
941 	unhold_lkb(lkb);
942 	return 0;
943 }
944 
remove_from_waiters(struct dlm_lkb * lkb,int mstype)945 static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
946 {
947 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
948 	int error;
949 
950 	mutex_lock(&ls->ls_waiters_mutex);
951 	error = _remove_from_waiters(lkb, mstype, NULL);
952 	mutex_unlock(&ls->ls_waiters_mutex);
953 	return error;
954 }
955 
956 /* Handles situations where we might be processing a "fake" or "stub" reply in
957    which we can't try to take waiters_mutex again. */
958 
remove_from_waiters_ms(struct dlm_lkb * lkb,struct dlm_message * ms)959 static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
960 {
961 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
962 	int error;
963 
964 	if (ms != &ls->ls_stub_ms)
965 		mutex_lock(&ls->ls_waiters_mutex);
966 	error = _remove_from_waiters(lkb, ms->m_type, ms);
967 	if (ms != &ls->ls_stub_ms)
968 		mutex_unlock(&ls->ls_waiters_mutex);
969 	return error;
970 }
971 
dir_remove(struct dlm_rsb * r)972 static void dir_remove(struct dlm_rsb *r)
973 {
974 	int to_nodeid;
975 
976 	if (dlm_no_directory(r->res_ls))
977 		return;
978 
979 	to_nodeid = dlm_dir_nodeid(r);
980 	if (to_nodeid != dlm_our_nodeid())
981 		send_remove(r);
982 	else
983 		dlm_dir_remove_entry(r->res_ls, to_nodeid,
984 				     r->res_name, r->res_length);
985 }
986 
987 /* FIXME: shouldn't this be able to exit as soon as one non-due rsb is
988    found since they are in order of newest to oldest? */
989 
shrink_bucket(struct dlm_ls * ls,int b)990 static int shrink_bucket(struct dlm_ls *ls, int b)
991 {
992 	struct dlm_rsb *r;
993 	int count = 0, found;
994 
995 	for (;;) {
996 		found = 0;
997 		spin_lock(&ls->ls_rsbtbl[b].lock);
998 		list_for_each_entry_reverse(r, &ls->ls_rsbtbl[b].toss,
999 					    res_hashchain) {
1000 			if (!time_after_eq(jiffies, r->res_toss_time +
1001 					   dlm_config.ci_toss_secs * HZ))
1002 				continue;
1003 			found = 1;
1004 			break;
1005 		}
1006 
1007 		if (!found) {
1008 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1009 			break;
1010 		}
1011 
1012 		if (kref_put(&r->res_ref, kill_rsb)) {
1013 			list_del(&r->res_hashchain);
1014 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1015 
1016 			if (is_master(r))
1017 				dir_remove(r);
1018 			dlm_free_rsb(r);
1019 			count++;
1020 		} else {
1021 			spin_unlock(&ls->ls_rsbtbl[b].lock);
1022 			log_error(ls, "tossed rsb in use %s", r->res_name);
1023 		}
1024 	}
1025 
1026 	return count;
1027 }
1028 
dlm_scan_rsbs(struct dlm_ls * ls)1029 void dlm_scan_rsbs(struct dlm_ls *ls)
1030 {
1031 	int i;
1032 
1033 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1034 		shrink_bucket(ls, i);
1035 		if (dlm_locking_stopped(ls))
1036 			break;
1037 		cond_resched();
1038 	}
1039 }
1040 
add_timeout(struct dlm_lkb * lkb)1041 static void add_timeout(struct dlm_lkb *lkb)
1042 {
1043 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1044 
1045 	if (is_master_copy(lkb))
1046 		return;
1047 
1048 	if (test_bit(LSFL_TIMEWARN, &ls->ls_flags) &&
1049 	    !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1050 		lkb->lkb_flags |= DLM_IFL_WATCH_TIMEWARN;
1051 		goto add_it;
1052 	}
1053 	if (lkb->lkb_exflags & DLM_LKF_TIMEOUT)
1054 		goto add_it;
1055 	return;
1056 
1057  add_it:
1058 	DLM_ASSERT(list_empty(&lkb->lkb_time_list), dlm_print_lkb(lkb););
1059 	mutex_lock(&ls->ls_timeout_mutex);
1060 	hold_lkb(lkb);
1061 	list_add_tail(&lkb->lkb_time_list, &ls->ls_timeout);
1062 	mutex_unlock(&ls->ls_timeout_mutex);
1063 }
1064 
del_timeout(struct dlm_lkb * lkb)1065 static void del_timeout(struct dlm_lkb *lkb)
1066 {
1067 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1068 
1069 	mutex_lock(&ls->ls_timeout_mutex);
1070 	if (!list_empty(&lkb->lkb_time_list)) {
1071 		list_del_init(&lkb->lkb_time_list);
1072 		unhold_lkb(lkb);
1073 	}
1074 	mutex_unlock(&ls->ls_timeout_mutex);
1075 }
1076 
1077 /* FIXME: is it safe to look at lkb_exflags, lkb_flags, lkb_timestamp, and
1078    lkb_lksb_timeout without lock_rsb?  Note: we can't lock timeout_mutex
1079    and then lock rsb because of lock ordering in add_timeout.  We may need
1080    to specify some special timeout-related bits in the lkb that are just to
1081    be accessed under the timeout_mutex. */
1082 
dlm_scan_timeout(struct dlm_ls * ls)1083 void dlm_scan_timeout(struct dlm_ls *ls)
1084 {
1085 	struct dlm_rsb *r;
1086 	struct dlm_lkb *lkb;
1087 	int do_cancel, do_warn;
1088 	s64 wait_us;
1089 
1090 	for (;;) {
1091 		if (dlm_locking_stopped(ls))
1092 			break;
1093 
1094 		do_cancel = 0;
1095 		do_warn = 0;
1096 		mutex_lock(&ls->ls_timeout_mutex);
1097 		list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list) {
1098 
1099 			wait_us = ktime_to_us(ktime_sub(ktime_get(),
1100 					      		lkb->lkb_timestamp));
1101 
1102 			if ((lkb->lkb_exflags & DLM_LKF_TIMEOUT) &&
1103 			    wait_us >= (lkb->lkb_timeout_cs * 10000))
1104 				do_cancel = 1;
1105 
1106 			if ((lkb->lkb_flags & DLM_IFL_WATCH_TIMEWARN) &&
1107 			    wait_us >= dlm_config.ci_timewarn_cs * 10000)
1108 				do_warn = 1;
1109 
1110 			if (!do_cancel && !do_warn)
1111 				continue;
1112 			hold_lkb(lkb);
1113 			break;
1114 		}
1115 		mutex_unlock(&ls->ls_timeout_mutex);
1116 
1117 		if (!do_cancel && !do_warn)
1118 			break;
1119 
1120 		r = lkb->lkb_resource;
1121 		hold_rsb(r);
1122 		lock_rsb(r);
1123 
1124 		if (do_warn) {
1125 			/* clear flag so we only warn once */
1126 			lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1127 			if (!(lkb->lkb_exflags & DLM_LKF_TIMEOUT))
1128 				del_timeout(lkb);
1129 			dlm_timeout_warn(lkb);
1130 		}
1131 
1132 		if (do_cancel) {
1133 			log_debug(ls, "timeout cancel %x node %d %s",
1134 				  lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1135 			lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1136 			lkb->lkb_flags |= DLM_IFL_TIMEOUT_CANCEL;
1137 			del_timeout(lkb);
1138 			_cancel_lock(r, lkb);
1139 		}
1140 
1141 		unlock_rsb(r);
1142 		unhold_rsb(r);
1143 		dlm_put_lkb(lkb);
1144 	}
1145 }
1146 
1147 /* This is only called by dlm_recoverd, and we rely on dlm_ls_stop() stopping
1148    dlm_recoverd before checking/setting ls_recover_begin. */
1149 
dlm_adjust_timeouts(struct dlm_ls * ls)1150 void dlm_adjust_timeouts(struct dlm_ls *ls)
1151 {
1152 	struct dlm_lkb *lkb;
1153 	u64 adj_us = jiffies_to_usecs(jiffies - ls->ls_recover_begin);
1154 
1155 	ls->ls_recover_begin = 0;
1156 	mutex_lock(&ls->ls_timeout_mutex);
1157 	list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list)
1158 		lkb->lkb_timestamp = ktime_add_us(lkb->lkb_timestamp, adj_us);
1159 	mutex_unlock(&ls->ls_timeout_mutex);
1160 }
1161 
1162 /* lkb is master or local copy */
1163 
set_lvb_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)1164 static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1165 {
1166 	int b, len = r->res_ls->ls_lvblen;
1167 
1168 	/* b=1 lvb returned to caller
1169 	   b=0 lvb written to rsb or invalidated
1170 	   b=-1 do nothing */
1171 
1172 	b =  dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1173 
1174 	if (b == 1) {
1175 		if (!lkb->lkb_lvbptr)
1176 			return;
1177 
1178 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1179 			return;
1180 
1181 		if (!r->res_lvbptr)
1182 			return;
1183 
1184 		memcpy(lkb->lkb_lvbptr, r->res_lvbptr, len);
1185 		lkb->lkb_lvbseq = r->res_lvbseq;
1186 
1187 	} else if (b == 0) {
1188 		if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1189 			rsb_set_flag(r, RSB_VALNOTVALID);
1190 			return;
1191 		}
1192 
1193 		if (!lkb->lkb_lvbptr)
1194 			return;
1195 
1196 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1197 			return;
1198 
1199 		if (!r->res_lvbptr)
1200 			r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1201 
1202 		if (!r->res_lvbptr)
1203 			return;
1204 
1205 		memcpy(r->res_lvbptr, lkb->lkb_lvbptr, len);
1206 		r->res_lvbseq++;
1207 		lkb->lkb_lvbseq = r->res_lvbseq;
1208 		rsb_clear_flag(r, RSB_VALNOTVALID);
1209 	}
1210 
1211 	if (rsb_flag(r, RSB_VALNOTVALID))
1212 		lkb->lkb_sbflags |= DLM_SBF_VALNOTVALID;
1213 }
1214 
set_lvb_unlock(struct dlm_rsb * r,struct dlm_lkb * lkb)1215 static void set_lvb_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1216 {
1217 	if (lkb->lkb_grmode < DLM_LOCK_PW)
1218 		return;
1219 
1220 	if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1221 		rsb_set_flag(r, RSB_VALNOTVALID);
1222 		return;
1223 	}
1224 
1225 	if (!lkb->lkb_lvbptr)
1226 		return;
1227 
1228 	if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1229 		return;
1230 
1231 	if (!r->res_lvbptr)
1232 		r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
1233 
1234 	if (!r->res_lvbptr)
1235 		return;
1236 
1237 	memcpy(r->res_lvbptr, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
1238 	r->res_lvbseq++;
1239 	rsb_clear_flag(r, RSB_VALNOTVALID);
1240 }
1241 
1242 /* lkb is process copy (pc) */
1243 
set_lvb_lock_pc(struct dlm_rsb * r,struct dlm_lkb * lkb,struct dlm_message * ms)1244 static void set_lvb_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1245 			    struct dlm_message *ms)
1246 {
1247 	int b;
1248 
1249 	if (!lkb->lkb_lvbptr)
1250 		return;
1251 
1252 	if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1253 		return;
1254 
1255 	b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1256 	if (b == 1) {
1257 		int len = receive_extralen(ms);
1258 		if (len > DLM_RESNAME_MAXLEN)
1259 			len = DLM_RESNAME_MAXLEN;
1260 		memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
1261 		lkb->lkb_lvbseq = ms->m_lvbseq;
1262 	}
1263 }
1264 
1265 /* Manipulate lkb's on rsb's convert/granted/waiting queues
1266    remove_lock -- used for unlock, removes lkb from granted
1267    revert_lock -- used for cancel, moves lkb from convert to granted
1268    grant_lock  -- used for request and convert, adds lkb to granted or
1269                   moves lkb from convert or waiting to granted
1270 
1271    Each of these is used for master or local copy lkb's.  There is
1272    also a _pc() variation used to make the corresponding change on
1273    a process copy (pc) lkb. */
1274 
_remove_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)1275 static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1276 {
1277 	del_lkb(r, lkb);
1278 	lkb->lkb_grmode = DLM_LOCK_IV;
1279 	/* this unhold undoes the original ref from create_lkb()
1280 	   so this leads to the lkb being freed */
1281 	unhold_lkb(lkb);
1282 }
1283 
remove_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)1284 static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1285 {
1286 	set_lvb_unlock(r, lkb);
1287 	_remove_lock(r, lkb);
1288 }
1289 
remove_lock_pc(struct dlm_rsb * r,struct dlm_lkb * lkb)1290 static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1291 {
1292 	_remove_lock(r, lkb);
1293 }
1294 
1295 /* returns: 0 did nothing
1296 	    1 moved lock to granted
1297 	   -1 removed lock */
1298 
revert_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)1299 static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1300 {
1301 	int rv = 0;
1302 
1303 	lkb->lkb_rqmode = DLM_LOCK_IV;
1304 
1305 	switch (lkb->lkb_status) {
1306 	case DLM_LKSTS_GRANTED:
1307 		break;
1308 	case DLM_LKSTS_CONVERT:
1309 		move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1310 		rv = 1;
1311 		break;
1312 	case DLM_LKSTS_WAITING:
1313 		del_lkb(r, lkb);
1314 		lkb->lkb_grmode = DLM_LOCK_IV;
1315 		/* this unhold undoes the original ref from create_lkb()
1316 		   so this leads to the lkb being freed */
1317 		unhold_lkb(lkb);
1318 		rv = -1;
1319 		break;
1320 	default:
1321 		log_print("invalid status for revert %d", lkb->lkb_status);
1322 	}
1323 	return rv;
1324 }
1325 
revert_lock_pc(struct dlm_rsb * r,struct dlm_lkb * lkb)1326 static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1327 {
1328 	return revert_lock(r, lkb);
1329 }
1330 
_grant_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)1331 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1332 {
1333 	if (lkb->lkb_grmode != lkb->lkb_rqmode) {
1334 		lkb->lkb_grmode = lkb->lkb_rqmode;
1335 		if (lkb->lkb_status)
1336 			move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1337 		else
1338 			add_lkb(r, lkb, DLM_LKSTS_GRANTED);
1339 	}
1340 
1341 	lkb->lkb_rqmode = DLM_LOCK_IV;
1342 }
1343 
grant_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)1344 static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1345 {
1346 	set_lvb_lock(r, lkb);
1347 	_grant_lock(r, lkb);
1348 	lkb->lkb_highbast = 0;
1349 }
1350 
grant_lock_pc(struct dlm_rsb * r,struct dlm_lkb * lkb,struct dlm_message * ms)1351 static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1352 			  struct dlm_message *ms)
1353 {
1354 	set_lvb_lock_pc(r, lkb, ms);
1355 	_grant_lock(r, lkb);
1356 }
1357 
1358 /* called by grant_pending_locks() which means an async grant message must
1359    be sent to the requesting node in addition to granting the lock if the
1360    lkb belongs to a remote node. */
1361 
grant_lock_pending(struct dlm_rsb * r,struct dlm_lkb * lkb)1362 static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
1363 {
1364 	grant_lock(r, lkb);
1365 	if (is_master_copy(lkb))
1366 		send_grant(r, lkb);
1367 	else
1368 		queue_cast(r, lkb, 0);
1369 }
1370 
1371 /* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
1372    change the granted/requested modes.  We're munging things accordingly in
1373    the process copy.
1374    CONVDEADLK: our grmode may have been forced down to NL to resolve a
1375    conversion deadlock
1376    ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
1377    compatible with other granted locks */
1378 
munge_demoted(struct dlm_lkb * lkb,struct dlm_message * ms)1379 static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
1380 {
1381 	if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
1382 		log_print("munge_demoted %x invalid reply type %d",
1383 			  lkb->lkb_id, ms->m_type);
1384 		return;
1385 	}
1386 
1387 	if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
1388 		log_print("munge_demoted %x invalid modes gr %d rq %d",
1389 			  lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
1390 		return;
1391 	}
1392 
1393 	lkb->lkb_grmode = DLM_LOCK_NL;
1394 }
1395 
munge_altmode(struct dlm_lkb * lkb,struct dlm_message * ms)1396 static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
1397 {
1398 	if (ms->m_type != DLM_MSG_REQUEST_REPLY &&
1399 	    ms->m_type != DLM_MSG_GRANT) {
1400 		log_print("munge_altmode %x invalid reply type %d",
1401 			  lkb->lkb_id, ms->m_type);
1402 		return;
1403 	}
1404 
1405 	if (lkb->lkb_exflags & DLM_LKF_ALTPR)
1406 		lkb->lkb_rqmode = DLM_LOCK_PR;
1407 	else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
1408 		lkb->lkb_rqmode = DLM_LOCK_CW;
1409 	else {
1410 		log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
1411 		dlm_print_lkb(lkb);
1412 	}
1413 }
1414 
first_in_list(struct dlm_lkb * lkb,struct list_head * head)1415 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
1416 {
1417 	struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
1418 					   lkb_statequeue);
1419 	if (lkb->lkb_id == first->lkb_id)
1420 		return 1;
1421 
1422 	return 0;
1423 }
1424 
1425 /* Check if the given lkb conflicts with another lkb on the queue. */
1426 
queue_conflict(struct list_head * head,struct dlm_lkb * lkb)1427 static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb)
1428 {
1429 	struct dlm_lkb *this;
1430 
1431 	list_for_each_entry(this, head, lkb_statequeue) {
1432 		if (this == lkb)
1433 			continue;
1434 		if (!modes_compat(this, lkb))
1435 			return 1;
1436 	}
1437 	return 0;
1438 }
1439 
1440 /*
1441  * "A conversion deadlock arises with a pair of lock requests in the converting
1442  * queue for one resource.  The granted mode of each lock blocks the requested
1443  * mode of the other lock."
1444  *
1445  * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
1446  * convert queue from being granted, then deadlk/demote lkb.
1447  *
1448  * Example:
1449  * Granted Queue: empty
1450  * Convert Queue: NL->EX (first lock)
1451  *                PR->EX (second lock)
1452  *
1453  * The first lock can't be granted because of the granted mode of the second
1454  * lock and the second lock can't be granted because it's not first in the
1455  * list.  We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
1456  * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
1457  * flag set and return DEMOTED in the lksb flags.
1458  *
1459  * Originally, this function detected conv-deadlk in a more limited scope:
1460  * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
1461  * - if lkb1 was the first entry in the queue (not just earlier), and was
1462  *   blocked by the granted mode of lkb2, and there was nothing on the
1463  *   granted queue preventing lkb1 from being granted immediately, i.e.
1464  *   lkb2 was the only thing preventing lkb1 from being granted.
1465  *
1466  * That second condition meant we'd only say there was conv-deadlk if
1467  * resolving it (by demotion) would lead to the first lock on the convert
1468  * queue being granted right away.  It allowed conversion deadlocks to exist
1469  * between locks on the convert queue while they couldn't be granted anyway.
1470  *
1471  * Now, we detect and take action on conversion deadlocks immediately when
1472  * they're created, even if they may not be immediately consequential.  If
1473  * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
1474  * mode that would prevent lkb1's conversion from being granted, we do a
1475  * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
1476  * I think this means that the lkb_is_ahead condition below should always
1477  * be zero, i.e. there will never be conv-deadlk between two locks that are
1478  * both already on the convert queue.
1479  */
1480 
conversion_deadlock_detect(struct dlm_rsb * r,struct dlm_lkb * lkb2)1481 static int conversion_deadlock_detect(struct dlm_rsb *r, struct dlm_lkb *lkb2)
1482 {
1483 	struct dlm_lkb *lkb1;
1484 	int lkb_is_ahead = 0;
1485 
1486 	list_for_each_entry(lkb1, &r->res_convertqueue, lkb_statequeue) {
1487 		if (lkb1 == lkb2) {
1488 			lkb_is_ahead = 1;
1489 			continue;
1490 		}
1491 
1492 		if (!lkb_is_ahead) {
1493 			if (!modes_compat(lkb2, lkb1))
1494 				return 1;
1495 		} else {
1496 			if (!modes_compat(lkb2, lkb1) &&
1497 			    !modes_compat(lkb1, lkb2))
1498 				return 1;
1499 		}
1500 	}
1501 	return 0;
1502 }
1503 
1504 /*
1505  * Return 1 if the lock can be granted, 0 otherwise.
1506  * Also detect and resolve conversion deadlocks.
1507  *
1508  * lkb is the lock to be granted
1509  *
1510  * now is 1 if the function is being called in the context of the
1511  * immediate request, it is 0 if called later, after the lock has been
1512  * queued.
1513  *
1514  * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
1515  */
1516 
_can_be_granted(struct dlm_rsb * r,struct dlm_lkb * lkb,int now)1517 static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now)
1518 {
1519 	int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV);
1520 
1521 	/*
1522 	 * 6-10: Version 5.4 introduced an option to address the phenomenon of
1523 	 * a new request for a NL mode lock being blocked.
1524 	 *
1525 	 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
1526 	 * request, then it would be granted.  In essence, the use of this flag
1527 	 * tells the Lock Manager to expedite theis request by not considering
1528 	 * what may be in the CONVERTING or WAITING queues...  As of this
1529 	 * writing, the EXPEDITE flag can be used only with new requests for NL
1530 	 * mode locks.  This flag is not valid for conversion requests.
1531 	 *
1532 	 * A shortcut.  Earlier checks return an error if EXPEDITE is used in a
1533 	 * conversion or used with a non-NL requested mode.  We also know an
1534 	 * EXPEDITE request is always granted immediately, so now must always
1535 	 * be 1.  The full condition to grant an expedite request: (now &&
1536 	 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
1537 	 * therefore be shortened to just checking the flag.
1538 	 */
1539 
1540 	if (lkb->lkb_exflags & DLM_LKF_EXPEDITE)
1541 		return 1;
1542 
1543 	/*
1544 	 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
1545 	 * added to the remaining conditions.
1546 	 */
1547 
1548 	if (queue_conflict(&r->res_grantqueue, lkb))
1549 		goto out;
1550 
1551 	/*
1552 	 * 6-3: By default, a conversion request is immediately granted if the
1553 	 * requested mode is compatible with the modes of all other granted
1554 	 * locks
1555 	 */
1556 
1557 	if (queue_conflict(&r->res_convertqueue, lkb))
1558 		goto out;
1559 
1560 	/*
1561 	 * 6-5: But the default algorithm for deciding whether to grant or
1562 	 * queue conversion requests does not by itself guarantee that such
1563 	 * requests are serviced on a "first come first serve" basis.  This, in
1564 	 * turn, can lead to a phenomenon known as "indefinate postponement".
1565 	 *
1566 	 * 6-7: This issue is dealt with by using the optional QUECVT flag with
1567 	 * the system service employed to request a lock conversion.  This flag
1568 	 * forces certain conversion requests to be queued, even if they are
1569 	 * compatible with the granted modes of other locks on the same
1570 	 * resource.  Thus, the use of this flag results in conversion requests
1571 	 * being ordered on a "first come first servce" basis.
1572 	 *
1573 	 * DCT: This condition is all about new conversions being able to occur
1574 	 * "in place" while the lock remains on the granted queue (assuming
1575 	 * nothing else conflicts.)  IOW if QUECVT isn't set, a conversion
1576 	 * doesn't _have_ to go onto the convert queue where it's processed in
1577 	 * order.  The "now" variable is necessary to distinguish converts
1578 	 * being received and processed for the first time now, because once a
1579 	 * convert is moved to the conversion queue the condition below applies
1580 	 * requiring fifo granting.
1581 	 */
1582 
1583 	if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT))
1584 		return 1;
1585 
1586 	/*
1587 	 * The NOORDER flag is set to avoid the standard vms rules on grant
1588 	 * order.
1589 	 */
1590 
1591 	if (lkb->lkb_exflags & DLM_LKF_NOORDER)
1592 		return 1;
1593 
1594 	/*
1595 	 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
1596 	 * granted until all other conversion requests ahead of it are granted
1597 	 * and/or canceled.
1598 	 */
1599 
1600 	if (!now && conv && first_in_list(lkb, &r->res_convertqueue))
1601 		return 1;
1602 
1603 	/*
1604 	 * 6-4: By default, a new request is immediately granted only if all
1605 	 * three of the following conditions are satisfied when the request is
1606 	 * issued:
1607 	 * - The queue of ungranted conversion requests for the resource is
1608 	 *   empty.
1609 	 * - The queue of ungranted new requests for the resource is empty.
1610 	 * - The mode of the new request is compatible with the most
1611 	 *   restrictive mode of all granted locks on the resource.
1612 	 */
1613 
1614 	if (now && !conv && list_empty(&r->res_convertqueue) &&
1615 	    list_empty(&r->res_waitqueue))
1616 		return 1;
1617 
1618 	/*
1619 	 * 6-4: Once a lock request is in the queue of ungranted new requests,
1620 	 * it cannot be granted until the queue of ungranted conversion
1621 	 * requests is empty, all ungranted new requests ahead of it are
1622 	 * granted and/or canceled, and it is compatible with the granted mode
1623 	 * of the most restrictive lock granted on the resource.
1624 	 */
1625 
1626 	if (!now && !conv && list_empty(&r->res_convertqueue) &&
1627 	    first_in_list(lkb, &r->res_waitqueue))
1628 		return 1;
1629  out:
1630 	return 0;
1631 }
1632 
can_be_granted(struct dlm_rsb * r,struct dlm_lkb * lkb,int now,int * err)1633 static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
1634 			  int *err)
1635 {
1636 	int rv;
1637 	int8_t alt = 0, rqmode = lkb->lkb_rqmode;
1638 	int8_t is_convert = (lkb->lkb_grmode != DLM_LOCK_IV);
1639 
1640 	if (err)
1641 		*err = 0;
1642 
1643 	rv = _can_be_granted(r, lkb, now);
1644 	if (rv)
1645 		goto out;
1646 
1647 	/*
1648 	 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
1649 	 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
1650 	 * cancels one of the locks.
1651 	 */
1652 
1653 	if (is_convert && can_be_queued(lkb) &&
1654 	    conversion_deadlock_detect(r, lkb)) {
1655 		if (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) {
1656 			lkb->lkb_grmode = DLM_LOCK_NL;
1657 			lkb->lkb_sbflags |= DLM_SBF_DEMOTED;
1658 		} else if (!(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1659 			if (err)
1660 				*err = -EDEADLK;
1661 			else {
1662 				log_print("can_be_granted deadlock %x now %d",
1663 					  lkb->lkb_id, now);
1664 				dlm_dump_rsb(r);
1665 			}
1666 		}
1667 		goto out;
1668 	}
1669 
1670 	/*
1671 	 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
1672 	 * to grant a request in a mode other than the normal rqmode.  It's a
1673 	 * simple way to provide a big optimization to applications that can
1674 	 * use them.
1675 	 */
1676 
1677 	if (rqmode != DLM_LOCK_PR && (lkb->lkb_exflags & DLM_LKF_ALTPR))
1678 		alt = DLM_LOCK_PR;
1679 	else if (rqmode != DLM_LOCK_CW && (lkb->lkb_exflags & DLM_LKF_ALTCW))
1680 		alt = DLM_LOCK_CW;
1681 
1682 	if (alt) {
1683 		lkb->lkb_rqmode = alt;
1684 		rv = _can_be_granted(r, lkb, now);
1685 		if (rv)
1686 			lkb->lkb_sbflags |= DLM_SBF_ALTMODE;
1687 		else
1688 			lkb->lkb_rqmode = rqmode;
1689 	}
1690  out:
1691 	return rv;
1692 }
1693 
1694 /* FIXME: I don't think that can_be_granted() can/will demote or find deadlock
1695    for locks pending on the convert list.  Once verified (watch for these
1696    log_prints), we should be able to just call _can_be_granted() and not
1697    bother with the demote/deadlk cases here (and there's no easy way to deal
1698    with a deadlk here, we'd have to generate something like grant_lock with
1699    the deadlk error.) */
1700 
1701 /* Returns the highest requested mode of all blocked conversions; sets
1702    cw if there's a blocked conversion to DLM_LOCK_CW. */
1703 
grant_pending_convert(struct dlm_rsb * r,int high,int * cw)1704 static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw)
1705 {
1706 	struct dlm_lkb *lkb, *s;
1707 	int hi, demoted, quit, grant_restart, demote_restart;
1708 	int deadlk;
1709 
1710 	quit = 0;
1711  restart:
1712 	grant_restart = 0;
1713 	demote_restart = 0;
1714 	hi = DLM_LOCK_IV;
1715 
1716 	list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) {
1717 		demoted = is_demoted(lkb);
1718 		deadlk = 0;
1719 
1720 		if (can_be_granted(r, lkb, 0, &deadlk)) {
1721 			grant_lock_pending(r, lkb);
1722 			grant_restart = 1;
1723 			continue;
1724 		}
1725 
1726 		if (!demoted && is_demoted(lkb)) {
1727 			log_print("WARN: pending demoted %x node %d %s",
1728 				  lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1729 			demote_restart = 1;
1730 			continue;
1731 		}
1732 
1733 		if (deadlk) {
1734 			log_print("WARN: pending deadlock %x node %d %s",
1735 				  lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1736 			dlm_dump_rsb(r);
1737 			continue;
1738 		}
1739 
1740 		hi = max_t(int, lkb->lkb_rqmode, hi);
1741 
1742 		if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
1743 			*cw = 1;
1744 	}
1745 
1746 	if (grant_restart)
1747 		goto restart;
1748 	if (demote_restart && !quit) {
1749 		quit = 1;
1750 		goto restart;
1751 	}
1752 
1753 	return max_t(int, high, hi);
1754 }
1755 
grant_pending_wait(struct dlm_rsb * r,int high,int * cw)1756 static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw)
1757 {
1758 	struct dlm_lkb *lkb, *s;
1759 
1760 	list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
1761 		if (can_be_granted(r, lkb, 0, NULL))
1762 			grant_lock_pending(r, lkb);
1763                 else {
1764 			high = max_t(int, lkb->lkb_rqmode, high);
1765 			if (lkb->lkb_rqmode == DLM_LOCK_CW)
1766 				*cw = 1;
1767 		}
1768 	}
1769 
1770 	return high;
1771 }
1772 
1773 /* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
1774    on either the convert or waiting queue.
1775    high is the largest rqmode of all locks blocked on the convert or
1776    waiting queue. */
1777 
lock_requires_bast(struct dlm_lkb * gr,int high,int cw)1778 static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
1779 {
1780 	if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
1781 		if (gr->lkb_highbast < DLM_LOCK_EX)
1782 			return 1;
1783 		return 0;
1784 	}
1785 
1786 	if (gr->lkb_highbast < high &&
1787 	    !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
1788 		return 1;
1789 	return 0;
1790 }
1791 
grant_pending_locks(struct dlm_rsb * r)1792 static void grant_pending_locks(struct dlm_rsb *r)
1793 {
1794 	struct dlm_lkb *lkb, *s;
1795 	int high = DLM_LOCK_IV;
1796 	int cw = 0;
1797 
1798 	DLM_ASSERT(is_master(r), dlm_dump_rsb(r););
1799 
1800 	high = grant_pending_convert(r, high, &cw);
1801 	high = grant_pending_wait(r, high, &cw);
1802 
1803 	if (high == DLM_LOCK_IV)
1804 		return;
1805 
1806 	/*
1807 	 * If there are locks left on the wait/convert queue then send blocking
1808 	 * ASTs to granted locks based on the largest requested mode (high)
1809 	 * found above.
1810 	 */
1811 
1812 	list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
1813 		if (lkb->lkb_bastfn && lock_requires_bast(lkb, high, cw)) {
1814 			if (cw && high == DLM_LOCK_PR &&
1815 			    lkb->lkb_grmode == DLM_LOCK_PR)
1816 				queue_bast(r, lkb, DLM_LOCK_CW);
1817 			else
1818 				queue_bast(r, lkb, high);
1819 			lkb->lkb_highbast = high;
1820 		}
1821 	}
1822 }
1823 
modes_require_bast(struct dlm_lkb * gr,struct dlm_lkb * rq)1824 static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
1825 {
1826 	if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
1827 	    (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
1828 		if (gr->lkb_highbast < DLM_LOCK_EX)
1829 			return 1;
1830 		return 0;
1831 	}
1832 
1833 	if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
1834 		return 1;
1835 	return 0;
1836 }
1837 
send_bast_queue(struct dlm_rsb * r,struct list_head * head,struct dlm_lkb * lkb)1838 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
1839 			    struct dlm_lkb *lkb)
1840 {
1841 	struct dlm_lkb *gr;
1842 
1843 	list_for_each_entry(gr, head, lkb_statequeue) {
1844 		/* skip self when sending basts to convertqueue */
1845 		if (gr == lkb)
1846 			continue;
1847 		if (gr->lkb_bastfn && modes_require_bast(gr, lkb)) {
1848 			queue_bast(r, gr, lkb->lkb_rqmode);
1849 			gr->lkb_highbast = lkb->lkb_rqmode;
1850 		}
1851 	}
1852 }
1853 
send_blocking_asts(struct dlm_rsb * r,struct dlm_lkb * lkb)1854 static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb)
1855 {
1856 	send_bast_queue(r, &r->res_grantqueue, lkb);
1857 }
1858 
send_blocking_asts_all(struct dlm_rsb * r,struct dlm_lkb * lkb)1859 static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb)
1860 {
1861 	send_bast_queue(r, &r->res_grantqueue, lkb);
1862 	send_bast_queue(r, &r->res_convertqueue, lkb);
1863 }
1864 
1865 /* set_master(r, lkb) -- set the master nodeid of a resource
1866 
1867    The purpose of this function is to set the nodeid field in the given
1868    lkb using the nodeid field in the given rsb.  If the rsb's nodeid is
1869    known, it can just be copied to the lkb and the function will return
1870    0.  If the rsb's nodeid is _not_ known, it needs to be looked up
1871    before it can be copied to the lkb.
1872 
1873    When the rsb nodeid is being looked up remotely, the initial lkb
1874    causing the lookup is kept on the ls_waiters list waiting for the
1875    lookup reply.  Other lkb's waiting for the same rsb lookup are kept
1876    on the rsb's res_lookup list until the master is verified.
1877 
1878    Return values:
1879    0: nodeid is set in rsb/lkb and the caller should go ahead and use it
1880    1: the rsb master is not available and the lkb has been placed on
1881       a wait queue
1882 */
1883 
set_master(struct dlm_rsb * r,struct dlm_lkb * lkb)1884 static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb)
1885 {
1886 	struct dlm_ls *ls = r->res_ls;
1887 	int i, error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid();
1888 
1889 	if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) {
1890 		rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
1891 		r->res_first_lkid = lkb->lkb_id;
1892 		lkb->lkb_nodeid = r->res_nodeid;
1893 		return 0;
1894 	}
1895 
1896 	if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) {
1897 		list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup);
1898 		return 1;
1899 	}
1900 
1901 	if (r->res_nodeid == 0) {
1902 		lkb->lkb_nodeid = 0;
1903 		return 0;
1904 	}
1905 
1906 	if (r->res_nodeid > 0) {
1907 		lkb->lkb_nodeid = r->res_nodeid;
1908 		return 0;
1909 	}
1910 
1911 	DLM_ASSERT(r->res_nodeid == -1, dlm_dump_rsb(r););
1912 
1913 	dir_nodeid = dlm_dir_nodeid(r);
1914 
1915 	if (dir_nodeid != our_nodeid) {
1916 		r->res_first_lkid = lkb->lkb_id;
1917 		send_lookup(r, lkb);
1918 		return 1;
1919 	}
1920 
1921 	for (i = 0; i < 2; i++) {
1922 		/* It's possible for dlm_scand to remove an old rsb for
1923 		   this same resource from the toss list, us to create
1924 		   a new one, look up the master locally, and find it
1925 		   already exists just before dlm_scand does the
1926 		   dir_remove() on the previous rsb. */
1927 
1928 		error = dlm_dir_lookup(ls, our_nodeid, r->res_name,
1929 				       r->res_length, &ret_nodeid);
1930 		if (!error)
1931 			break;
1932 		log_debug(ls, "dir_lookup error %d %s", error, r->res_name);
1933 		schedule();
1934 	}
1935 	if (error && error != -EEXIST)
1936 		return error;
1937 
1938 	if (ret_nodeid == our_nodeid) {
1939 		r->res_first_lkid = 0;
1940 		r->res_nodeid = 0;
1941 		lkb->lkb_nodeid = 0;
1942 	} else {
1943 		r->res_first_lkid = lkb->lkb_id;
1944 		r->res_nodeid = ret_nodeid;
1945 		lkb->lkb_nodeid = ret_nodeid;
1946 	}
1947 	return 0;
1948 }
1949 
process_lookup_list(struct dlm_rsb * r)1950 static void process_lookup_list(struct dlm_rsb *r)
1951 {
1952 	struct dlm_lkb *lkb, *safe;
1953 
1954 	list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
1955 		list_del_init(&lkb->lkb_rsb_lookup);
1956 		_request_lock(r, lkb);
1957 		schedule();
1958 	}
1959 }
1960 
1961 /* confirm_master -- confirm (or deny) an rsb's master nodeid */
1962 
confirm_master(struct dlm_rsb * r,int error)1963 static void confirm_master(struct dlm_rsb *r, int error)
1964 {
1965 	struct dlm_lkb *lkb;
1966 
1967 	if (!r->res_first_lkid)
1968 		return;
1969 
1970 	switch (error) {
1971 	case 0:
1972 	case -EINPROGRESS:
1973 		r->res_first_lkid = 0;
1974 		process_lookup_list(r);
1975 		break;
1976 
1977 	case -EAGAIN:
1978 	case -EBADR:
1979 	case -ENOTBLK:
1980 		/* the remote request failed and won't be retried (it was
1981 		   a NOQUEUE, or has been canceled/unlocked); make a waiting
1982 		   lkb the first_lkid */
1983 
1984 		r->res_first_lkid = 0;
1985 
1986 		if (!list_empty(&r->res_lookup)) {
1987 			lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
1988 					 lkb_rsb_lookup);
1989 			list_del_init(&lkb->lkb_rsb_lookup);
1990 			r->res_first_lkid = lkb->lkb_id;
1991 			_request_lock(r, lkb);
1992 		}
1993 		break;
1994 
1995 	default:
1996 		log_error(r->res_ls, "confirm_master unknown error %d", error);
1997 	}
1998 }
1999 
set_lock_args(int mode,struct dlm_lksb * lksb,uint32_t flags,int namelen,unsigned long timeout_cs,void (* ast)(void * astparam),void * astparam,void (* bast)(void * astparam,int mode),struct dlm_args * args)2000 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
2001 			 int namelen, unsigned long timeout_cs,
2002 			 void (*ast) (void *astparam),
2003 			 void *astparam,
2004 			 void (*bast) (void *astparam, int mode),
2005 			 struct dlm_args *args)
2006 {
2007 	int rv = -EINVAL;
2008 
2009 	/* check for invalid arg usage */
2010 
2011 	if (mode < 0 || mode > DLM_LOCK_EX)
2012 		goto out;
2013 
2014 	if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
2015 		goto out;
2016 
2017 	if (flags & DLM_LKF_CANCEL)
2018 		goto out;
2019 
2020 	if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
2021 		goto out;
2022 
2023 	if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
2024 		goto out;
2025 
2026 	if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
2027 		goto out;
2028 
2029 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
2030 		goto out;
2031 
2032 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
2033 		goto out;
2034 
2035 	if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2036 		goto out;
2037 
2038 	if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2039 		goto out;
2040 
2041 	if (!ast || !lksb)
2042 		goto out;
2043 
2044 	if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2045 		goto out;
2046 
2047 	if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2048 		goto out;
2049 
2050 	/* these args will be copied to the lkb in validate_lock_args,
2051 	   it cannot be done now because when converting locks, fields in
2052 	   an active lkb cannot be modified before locking the rsb */
2053 
2054 	args->flags = flags;
2055 	args->astfn = ast;
2056 	args->astparam = astparam;
2057 	args->bastfn = bast;
2058 	args->timeout = timeout_cs;
2059 	args->mode = mode;
2060 	args->lksb = lksb;
2061 	rv = 0;
2062  out:
2063 	return rv;
2064 }
2065 
set_unlock_args(uint32_t flags,void * astarg,struct dlm_args * args)2066 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2067 {
2068 	if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2069  		      DLM_LKF_FORCEUNLOCK))
2070 		return -EINVAL;
2071 
2072 	if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2073 		return -EINVAL;
2074 
2075 	args->flags = flags;
2076 	args->astparam = astarg;
2077 	return 0;
2078 }
2079 
validate_lock_args(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_args * args)2080 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2081 			      struct dlm_args *args)
2082 {
2083 	int rv = -EINVAL;
2084 
2085 	if (args->flags & DLM_LKF_CONVERT) {
2086 		if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2087 			goto out;
2088 
2089 		if (args->flags & DLM_LKF_QUECVT &&
2090 		    !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2091 			goto out;
2092 
2093 		rv = -EBUSY;
2094 		if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2095 			goto out;
2096 
2097 		if (lkb->lkb_wait_type)
2098 			goto out;
2099 
2100 		if (is_overlap(lkb))
2101 			goto out;
2102 	}
2103 
2104 	lkb->lkb_exflags = args->flags;
2105 	lkb->lkb_sbflags = 0;
2106 	lkb->lkb_astfn = args->astfn;
2107 	lkb->lkb_astparam = args->astparam;
2108 	lkb->lkb_bastfn = args->bastfn;
2109 	lkb->lkb_rqmode = args->mode;
2110 	lkb->lkb_lksb = args->lksb;
2111 	lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2112 	lkb->lkb_ownpid = (int) current->pid;
2113 	lkb->lkb_timeout_cs = args->timeout;
2114 	rv = 0;
2115  out:
2116 	if (rv)
2117 		log_debug(ls, "validate_lock_args %d %x %x %x %d %d %s",
2118 			  rv, lkb->lkb_id, lkb->lkb_flags, args->flags,
2119 			  lkb->lkb_status, lkb->lkb_wait_type,
2120 			  lkb->lkb_resource->res_name);
2121 	return rv;
2122 }
2123 
2124 /* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2125    for success */
2126 
2127 /* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2128    because there may be a lookup in progress and it's valid to do
2129    cancel/unlockf on it */
2130 
validate_unlock_args(struct dlm_lkb * lkb,struct dlm_args * args)2131 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2132 {
2133 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
2134 	int rv = -EINVAL;
2135 
2136 	if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2137 		log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2138 		dlm_print_lkb(lkb);
2139 		goto out;
2140 	}
2141 
2142 	/* an lkb may still exist even though the lock is EOL'ed due to a
2143 	   cancel, unlock or failed noqueue request; an app can't use these
2144 	   locks; return same error as if the lkid had not been found at all */
2145 
2146 	if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2147 		log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2148 		rv = -ENOENT;
2149 		goto out;
2150 	}
2151 
2152 	/* an lkb may be waiting for an rsb lookup to complete where the
2153 	   lookup was initiated by another lock */
2154 
2155 	if (!list_empty(&lkb->lkb_rsb_lookup)) {
2156 		if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
2157 			log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2158 			list_del_init(&lkb->lkb_rsb_lookup);
2159 			queue_cast(lkb->lkb_resource, lkb,
2160 				   args->flags & DLM_LKF_CANCEL ?
2161 				   -DLM_ECANCEL : -DLM_EUNLOCK);
2162 			unhold_lkb(lkb); /* undoes create_lkb() */
2163 		}
2164 		/* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2165 		rv = -EBUSY;
2166 		goto out;
2167 	}
2168 
2169 	/* cancel not allowed with another cancel/unlock in progress */
2170 
2171 	if (args->flags & DLM_LKF_CANCEL) {
2172 		if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2173 			goto out;
2174 
2175 		if (is_overlap(lkb))
2176 			goto out;
2177 
2178 		/* don't let scand try to do a cancel */
2179 		del_timeout(lkb);
2180 
2181 		if (lkb->lkb_flags & DLM_IFL_RESEND) {
2182 			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2183 			rv = -EBUSY;
2184 			goto out;
2185 		}
2186 
2187 		/* there's nothing to cancel */
2188 		if (lkb->lkb_status == DLM_LKSTS_GRANTED &&
2189 		    !lkb->lkb_wait_type) {
2190 			rv = -EBUSY;
2191 			goto out;
2192 		}
2193 
2194 		switch (lkb->lkb_wait_type) {
2195 		case DLM_MSG_LOOKUP:
2196 		case DLM_MSG_REQUEST:
2197 			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2198 			rv = -EBUSY;
2199 			goto out;
2200 		case DLM_MSG_UNLOCK:
2201 		case DLM_MSG_CANCEL:
2202 			goto out;
2203 		}
2204 		/* add_to_waiters() will set OVERLAP_CANCEL */
2205 		goto out_ok;
2206 	}
2207 
2208 	/* do we need to allow a force-unlock if there's a normal unlock
2209 	   already in progress?  in what conditions could the normal unlock
2210 	   fail such that we'd want to send a force-unlock to be sure? */
2211 
2212 	if (args->flags & DLM_LKF_FORCEUNLOCK) {
2213 		if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2214 			goto out;
2215 
2216 		if (is_overlap_unlock(lkb))
2217 			goto out;
2218 
2219 		/* don't let scand try to do a cancel */
2220 		del_timeout(lkb);
2221 
2222 		if (lkb->lkb_flags & DLM_IFL_RESEND) {
2223 			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2224 			rv = -EBUSY;
2225 			goto out;
2226 		}
2227 
2228 		switch (lkb->lkb_wait_type) {
2229 		case DLM_MSG_LOOKUP:
2230 		case DLM_MSG_REQUEST:
2231 			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2232 			rv = -EBUSY;
2233 			goto out;
2234 		case DLM_MSG_UNLOCK:
2235 			goto out;
2236 		}
2237 		/* add_to_waiters() will set OVERLAP_UNLOCK */
2238 		goto out_ok;
2239 	}
2240 
2241 	/* normal unlock not allowed if there's any op in progress */
2242 	rv = -EBUSY;
2243 	if (lkb->lkb_wait_type || lkb->lkb_wait_count)
2244 		goto out;
2245 
2246  out_ok:
2247 	/* an overlapping op shouldn't blow away exflags from other op */
2248 	lkb->lkb_exflags |= args->flags;
2249 	lkb->lkb_sbflags = 0;
2250 	lkb->lkb_astparam = args->astparam;
2251 	rv = 0;
2252  out:
2253 	if (rv)
2254 		log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
2255 			  lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
2256 			  args->flags, lkb->lkb_wait_type,
2257 			  lkb->lkb_resource->res_name);
2258 	return rv;
2259 }
2260 
2261 /*
2262  * Four stage 4 varieties:
2263  * do_request(), do_convert(), do_unlock(), do_cancel()
2264  * These are called on the master node for the given lock and
2265  * from the central locking logic.
2266  */
2267 
do_request(struct dlm_rsb * r,struct dlm_lkb * lkb)2268 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2269 {
2270 	int error = 0;
2271 
2272 	if (can_be_granted(r, lkb, 1, NULL)) {
2273 		grant_lock(r, lkb);
2274 		queue_cast(r, lkb, 0);
2275 		goto out;
2276 	}
2277 
2278 	if (can_be_queued(lkb)) {
2279 		error = -EINPROGRESS;
2280 		add_lkb(r, lkb, DLM_LKSTS_WAITING);
2281 		add_timeout(lkb);
2282 		goto out;
2283 	}
2284 
2285 	error = -EAGAIN;
2286 	queue_cast(r, lkb, -EAGAIN);
2287  out:
2288 	return error;
2289 }
2290 
do_request_effects(struct dlm_rsb * r,struct dlm_lkb * lkb,int error)2291 static void do_request_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2292 			       int error)
2293 {
2294 	switch (error) {
2295 	case -EAGAIN:
2296 		if (force_blocking_asts(lkb))
2297 			send_blocking_asts_all(r, lkb);
2298 		break;
2299 	case -EINPROGRESS:
2300 		send_blocking_asts(r, lkb);
2301 		break;
2302 	}
2303 }
2304 
do_convert(struct dlm_rsb * r,struct dlm_lkb * lkb)2305 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2306 {
2307 	int error = 0;
2308 	int deadlk = 0;
2309 
2310 	/* changing an existing lock may allow others to be granted */
2311 
2312 	if (can_be_granted(r, lkb, 1, &deadlk)) {
2313 		grant_lock(r, lkb);
2314 		queue_cast(r, lkb, 0);
2315 		goto out;
2316 	}
2317 
2318 	/* can_be_granted() detected that this lock would block in a conversion
2319 	   deadlock, so we leave it on the granted queue and return EDEADLK in
2320 	   the ast for the convert. */
2321 
2322 	if (deadlk) {
2323 		/* it's left on the granted queue */
2324 		log_debug(r->res_ls, "deadlock %x node %d sts%d g%d r%d %s",
2325 			  lkb->lkb_id, lkb->lkb_nodeid, lkb->lkb_status,
2326 			  lkb->lkb_grmode, lkb->lkb_rqmode, r->res_name);
2327 		revert_lock(r, lkb);
2328 		queue_cast(r, lkb, -EDEADLK);
2329 		error = -EDEADLK;
2330 		goto out;
2331 	}
2332 
2333 	/* is_demoted() means the can_be_granted() above set the grmode
2334 	   to NL, and left us on the granted queue.  This auto-demotion
2335 	   (due to CONVDEADLK) might mean other locks, and/or this lock, are
2336 	   now grantable.  We have to try to grant other converting locks
2337 	   before we try again to grant this one. */
2338 
2339 	if (is_demoted(lkb)) {
2340 		grant_pending_convert(r, DLM_LOCK_IV, NULL);
2341 		if (_can_be_granted(r, lkb, 1)) {
2342 			grant_lock(r, lkb);
2343 			queue_cast(r, lkb, 0);
2344 			goto out;
2345 		}
2346 		/* else fall through and move to convert queue */
2347 	}
2348 
2349 	if (can_be_queued(lkb)) {
2350 		error = -EINPROGRESS;
2351 		del_lkb(r, lkb);
2352 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2353 		add_timeout(lkb);
2354 		goto out;
2355 	}
2356 
2357 	error = -EAGAIN;
2358 	queue_cast(r, lkb, -EAGAIN);
2359  out:
2360 	return error;
2361 }
2362 
do_convert_effects(struct dlm_rsb * r,struct dlm_lkb * lkb,int error)2363 static void do_convert_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2364 			       int error)
2365 {
2366 	switch (error) {
2367 	case 0:
2368 		grant_pending_locks(r);
2369 		/* grant_pending_locks also sends basts */
2370 		break;
2371 	case -EAGAIN:
2372 		if (force_blocking_asts(lkb))
2373 			send_blocking_asts_all(r, lkb);
2374 		break;
2375 	case -EINPROGRESS:
2376 		send_blocking_asts(r, lkb);
2377 		break;
2378 	}
2379 }
2380 
do_unlock(struct dlm_rsb * r,struct dlm_lkb * lkb)2381 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2382 {
2383 	remove_lock(r, lkb);
2384 	queue_cast(r, lkb, -DLM_EUNLOCK);
2385 	return -DLM_EUNLOCK;
2386 }
2387 
do_unlock_effects(struct dlm_rsb * r,struct dlm_lkb * lkb,int error)2388 static void do_unlock_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2389 			      int error)
2390 {
2391 	grant_pending_locks(r);
2392 }
2393 
2394 /* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
2395 
do_cancel(struct dlm_rsb * r,struct dlm_lkb * lkb)2396 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2397 {
2398 	int error;
2399 
2400 	error = revert_lock(r, lkb);
2401 	if (error) {
2402 		queue_cast(r, lkb, -DLM_ECANCEL);
2403 		return -DLM_ECANCEL;
2404 	}
2405 	return 0;
2406 }
2407 
do_cancel_effects(struct dlm_rsb * r,struct dlm_lkb * lkb,int error)2408 static void do_cancel_effects(struct dlm_rsb *r, struct dlm_lkb *lkb,
2409 			      int error)
2410 {
2411 	if (error)
2412 		grant_pending_locks(r);
2413 }
2414 
2415 /*
2416  * Four stage 3 varieties:
2417  * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2418  */
2419 
2420 /* add a new lkb to a possibly new rsb, called by requesting process */
2421 
_request_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)2422 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2423 {
2424 	int error;
2425 
2426 	/* set_master: sets lkb nodeid from r */
2427 
2428 	error = set_master(r, lkb);
2429 	if (error < 0)
2430 		goto out;
2431 	if (error) {
2432 		error = 0;
2433 		goto out;
2434 	}
2435 
2436 	if (is_remote(r)) {
2437 		/* receive_request() calls do_request() on remote node */
2438 		error = send_request(r, lkb);
2439 	} else {
2440 		error = do_request(r, lkb);
2441 		/* for remote locks the request_reply is sent
2442 		   between do_request and do_request_effects */
2443 		do_request_effects(r, lkb, error);
2444 	}
2445  out:
2446 	return error;
2447 }
2448 
2449 /* change some property of an existing lkb, e.g. mode */
2450 
_convert_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)2451 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2452 {
2453 	int error;
2454 
2455 	if (is_remote(r)) {
2456 		/* receive_convert() calls do_convert() on remote node */
2457 		error = send_convert(r, lkb);
2458 	} else {
2459 		error = do_convert(r, lkb);
2460 		/* for remote locks the convert_reply is sent
2461 		   between do_convert and do_convert_effects */
2462 		do_convert_effects(r, lkb, error);
2463 	}
2464 
2465 	return error;
2466 }
2467 
2468 /* remove an existing lkb from the granted queue */
2469 
_unlock_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)2470 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2471 {
2472 	int error;
2473 
2474 	if (is_remote(r)) {
2475 		/* receive_unlock() calls do_unlock() on remote node */
2476 		error = send_unlock(r, lkb);
2477 	} else {
2478 		error = do_unlock(r, lkb);
2479 		/* for remote locks the unlock_reply is sent
2480 		   between do_unlock and do_unlock_effects */
2481 		do_unlock_effects(r, lkb, error);
2482 	}
2483 
2484 	return error;
2485 }
2486 
2487 /* remove an existing lkb from the convert or wait queue */
2488 
_cancel_lock(struct dlm_rsb * r,struct dlm_lkb * lkb)2489 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2490 {
2491 	int error;
2492 
2493 	if (is_remote(r)) {
2494 		/* receive_cancel() calls do_cancel() on remote node */
2495 		error = send_cancel(r, lkb);
2496 	} else {
2497 		error = do_cancel(r, lkb);
2498 		/* for remote locks the cancel_reply is sent
2499 		   between do_cancel and do_cancel_effects */
2500 		do_cancel_effects(r, lkb, error);
2501 	}
2502 
2503 	return error;
2504 }
2505 
2506 /*
2507  * Four stage 2 varieties:
2508  * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2509  */
2510 
request_lock(struct dlm_ls * ls,struct dlm_lkb * lkb,char * name,int len,struct dlm_args * args)2511 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
2512 			int len, struct dlm_args *args)
2513 {
2514 	struct dlm_rsb *r;
2515 	int error;
2516 
2517 	error = validate_lock_args(ls, lkb, args);
2518 	if (error)
2519 		goto out;
2520 
2521 	error = find_rsb(ls, name, len, R_CREATE, &r);
2522 	if (error)
2523 		goto out;
2524 
2525 	lock_rsb(r);
2526 
2527 	attach_lkb(r, lkb);
2528 	lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
2529 
2530 	error = _request_lock(r, lkb);
2531 
2532 	unlock_rsb(r);
2533 	put_rsb(r);
2534 
2535  out:
2536 	return error;
2537 }
2538 
convert_lock(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_args * args)2539 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2540 			struct dlm_args *args)
2541 {
2542 	struct dlm_rsb *r;
2543 	int error;
2544 
2545 	r = lkb->lkb_resource;
2546 
2547 	hold_rsb(r);
2548 	lock_rsb(r);
2549 
2550 	error = validate_lock_args(ls, lkb, args);
2551 	if (error)
2552 		goto out;
2553 
2554 	error = _convert_lock(r, lkb);
2555  out:
2556 	unlock_rsb(r);
2557 	put_rsb(r);
2558 	return error;
2559 }
2560 
unlock_lock(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_args * args)2561 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2562 		       struct dlm_args *args)
2563 {
2564 	struct dlm_rsb *r;
2565 	int error;
2566 
2567 	r = lkb->lkb_resource;
2568 
2569 	hold_rsb(r);
2570 	lock_rsb(r);
2571 
2572 	error = validate_unlock_args(lkb, args);
2573 	if (error)
2574 		goto out;
2575 
2576 	error = _unlock_lock(r, lkb);
2577  out:
2578 	unlock_rsb(r);
2579 	put_rsb(r);
2580 	return error;
2581 }
2582 
cancel_lock(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_args * args)2583 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2584 		       struct dlm_args *args)
2585 {
2586 	struct dlm_rsb *r;
2587 	int error;
2588 
2589 	r = lkb->lkb_resource;
2590 
2591 	hold_rsb(r);
2592 	lock_rsb(r);
2593 
2594 	error = validate_unlock_args(lkb, args);
2595 	if (error)
2596 		goto out;
2597 
2598 	error = _cancel_lock(r, lkb);
2599  out:
2600 	unlock_rsb(r);
2601 	put_rsb(r);
2602 	return error;
2603 }
2604 
2605 /*
2606  * Two stage 1 varieties:  dlm_lock() and dlm_unlock()
2607  */
2608 
dlm_lock(dlm_lockspace_t * lockspace,int mode,struct dlm_lksb * lksb,uint32_t flags,void * name,unsigned int namelen,uint32_t parent_lkid,void (* ast)(void * astarg),void * astarg,void (* bast)(void * astarg,int mode))2609 int dlm_lock(dlm_lockspace_t *lockspace,
2610 	     int mode,
2611 	     struct dlm_lksb *lksb,
2612 	     uint32_t flags,
2613 	     void *name,
2614 	     unsigned int namelen,
2615 	     uint32_t parent_lkid,
2616 	     void (*ast) (void *astarg),
2617 	     void *astarg,
2618 	     void (*bast) (void *astarg, int mode))
2619 {
2620 	struct dlm_ls *ls;
2621 	struct dlm_lkb *lkb;
2622 	struct dlm_args args;
2623 	int error, convert = flags & DLM_LKF_CONVERT;
2624 
2625 	ls = dlm_find_lockspace_local(lockspace);
2626 	if (!ls)
2627 		return -EINVAL;
2628 
2629 	dlm_lock_recovery(ls);
2630 
2631 	if (convert)
2632 		error = find_lkb(ls, lksb->sb_lkid, &lkb);
2633 	else
2634 		error = create_lkb(ls, &lkb);
2635 
2636 	if (error)
2637 		goto out;
2638 
2639 	error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
2640 			      astarg, bast, &args);
2641 	if (error)
2642 		goto out_put;
2643 
2644 	if (convert)
2645 		error = convert_lock(ls, lkb, &args);
2646 	else
2647 		error = request_lock(ls, lkb, name, namelen, &args);
2648 
2649 	if (error == -EINPROGRESS)
2650 		error = 0;
2651  out_put:
2652 	if (convert || error)
2653 		__put_lkb(ls, lkb);
2654 	if (error == -EAGAIN || error == -EDEADLK)
2655 		error = 0;
2656  out:
2657 	dlm_unlock_recovery(ls);
2658 	dlm_put_lockspace(ls);
2659 	return error;
2660 }
2661 
dlm_unlock(dlm_lockspace_t * lockspace,uint32_t lkid,uint32_t flags,struct dlm_lksb * lksb,void * astarg)2662 int dlm_unlock(dlm_lockspace_t *lockspace,
2663 	       uint32_t lkid,
2664 	       uint32_t flags,
2665 	       struct dlm_lksb *lksb,
2666 	       void *astarg)
2667 {
2668 	struct dlm_ls *ls;
2669 	struct dlm_lkb *lkb;
2670 	struct dlm_args args;
2671 	int error;
2672 
2673 	ls = dlm_find_lockspace_local(lockspace);
2674 	if (!ls)
2675 		return -EINVAL;
2676 
2677 	dlm_lock_recovery(ls);
2678 
2679 	error = find_lkb(ls, lkid, &lkb);
2680 	if (error)
2681 		goto out;
2682 
2683 	error = set_unlock_args(flags, astarg, &args);
2684 	if (error)
2685 		goto out_put;
2686 
2687 	if (flags & DLM_LKF_CANCEL)
2688 		error = cancel_lock(ls, lkb, &args);
2689 	else
2690 		error = unlock_lock(ls, lkb, &args);
2691 
2692 	if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
2693 		error = 0;
2694 	if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
2695 		error = 0;
2696  out_put:
2697 	dlm_put_lkb(lkb);
2698  out:
2699 	dlm_unlock_recovery(ls);
2700 	dlm_put_lockspace(ls);
2701 	return error;
2702 }
2703 
2704 /*
2705  * send/receive routines for remote operations and replies
2706  *
2707  * send_args
2708  * send_common
2709  * send_request			receive_request
2710  * send_convert			receive_convert
2711  * send_unlock			receive_unlock
2712  * send_cancel			receive_cancel
2713  * send_grant			receive_grant
2714  * send_bast			receive_bast
2715  * send_lookup			receive_lookup
2716  * send_remove			receive_remove
2717  *
2718  * 				send_common_reply
2719  * receive_request_reply	send_request_reply
2720  * receive_convert_reply	send_convert_reply
2721  * receive_unlock_reply		send_unlock_reply
2722  * receive_cancel_reply		send_cancel_reply
2723  * receive_lookup_reply		send_lookup_reply
2724  */
2725 
_create_message(struct dlm_ls * ls,int mb_len,int to_nodeid,int mstype,struct dlm_message ** ms_ret,struct dlm_mhandle ** mh_ret)2726 static int _create_message(struct dlm_ls *ls, int mb_len,
2727 			   int to_nodeid, int mstype,
2728 			   struct dlm_message **ms_ret,
2729 			   struct dlm_mhandle **mh_ret)
2730 {
2731 	struct dlm_message *ms;
2732 	struct dlm_mhandle *mh;
2733 	char *mb;
2734 
2735 	/* get_buffer gives us a message handle (mh) that we need to
2736 	   pass into lowcomms_commit and a message buffer (mb) that we
2737 	   write our data into */
2738 
2739 	mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_NOFS, &mb);
2740 	if (!mh)
2741 		return -ENOBUFS;
2742 
2743 	memset(mb, 0, mb_len);
2744 
2745 	ms = (struct dlm_message *) mb;
2746 
2747 	ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
2748 	ms->m_header.h_lockspace = ls->ls_global_id;
2749 	ms->m_header.h_nodeid = dlm_our_nodeid();
2750 	ms->m_header.h_length = mb_len;
2751 	ms->m_header.h_cmd = DLM_MSG;
2752 
2753 	ms->m_type = mstype;
2754 
2755 	*mh_ret = mh;
2756 	*ms_ret = ms;
2757 	return 0;
2758 }
2759 
create_message(struct dlm_rsb * r,struct dlm_lkb * lkb,int to_nodeid,int mstype,struct dlm_message ** ms_ret,struct dlm_mhandle ** mh_ret)2760 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
2761 			  int to_nodeid, int mstype,
2762 			  struct dlm_message **ms_ret,
2763 			  struct dlm_mhandle **mh_ret)
2764 {
2765 	int mb_len = sizeof(struct dlm_message);
2766 
2767 	switch (mstype) {
2768 	case DLM_MSG_REQUEST:
2769 	case DLM_MSG_LOOKUP:
2770 	case DLM_MSG_REMOVE:
2771 		mb_len += r->res_length;
2772 		break;
2773 	case DLM_MSG_CONVERT:
2774 	case DLM_MSG_UNLOCK:
2775 	case DLM_MSG_REQUEST_REPLY:
2776 	case DLM_MSG_CONVERT_REPLY:
2777 	case DLM_MSG_GRANT:
2778 		if (lkb && lkb->lkb_lvbptr)
2779 			mb_len += r->res_ls->ls_lvblen;
2780 		break;
2781 	}
2782 
2783 	return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
2784 			       ms_ret, mh_ret);
2785 }
2786 
2787 /* further lowcomms enhancements or alternate implementations may make
2788    the return value from this function useful at some point */
2789 
send_message(struct dlm_mhandle * mh,struct dlm_message * ms)2790 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
2791 {
2792 	dlm_message_out(ms);
2793 	dlm_lowcomms_commit_buffer(mh);
2794 	return 0;
2795 }
2796 
send_args(struct dlm_rsb * r,struct dlm_lkb * lkb,struct dlm_message * ms)2797 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
2798 		      struct dlm_message *ms)
2799 {
2800 	ms->m_nodeid   = lkb->lkb_nodeid;
2801 	ms->m_pid      = lkb->lkb_ownpid;
2802 	ms->m_lkid     = lkb->lkb_id;
2803 	ms->m_remid    = lkb->lkb_remid;
2804 	ms->m_exflags  = lkb->lkb_exflags;
2805 	ms->m_sbflags  = lkb->lkb_sbflags;
2806 	ms->m_flags    = lkb->lkb_flags;
2807 	ms->m_lvbseq   = lkb->lkb_lvbseq;
2808 	ms->m_status   = lkb->lkb_status;
2809 	ms->m_grmode   = lkb->lkb_grmode;
2810 	ms->m_rqmode   = lkb->lkb_rqmode;
2811 	ms->m_hash     = r->res_hash;
2812 
2813 	/* m_result and m_bastmode are set from function args,
2814 	   not from lkb fields */
2815 
2816 	if (lkb->lkb_bastfn)
2817 		ms->m_asts |= DLM_CB_BAST;
2818 	if (lkb->lkb_astfn)
2819 		ms->m_asts |= DLM_CB_CAST;
2820 
2821 	/* compare with switch in create_message; send_remove() doesn't
2822 	   use send_args() */
2823 
2824 	switch (ms->m_type) {
2825 	case DLM_MSG_REQUEST:
2826 	case DLM_MSG_LOOKUP:
2827 		memcpy(ms->m_extra, r->res_name, r->res_length);
2828 		break;
2829 	case DLM_MSG_CONVERT:
2830 	case DLM_MSG_UNLOCK:
2831 	case DLM_MSG_REQUEST_REPLY:
2832 	case DLM_MSG_CONVERT_REPLY:
2833 	case DLM_MSG_GRANT:
2834 		if (!lkb->lkb_lvbptr)
2835 			break;
2836 		memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
2837 		break;
2838 	}
2839 }
2840 
send_common(struct dlm_rsb * r,struct dlm_lkb * lkb,int mstype)2841 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
2842 {
2843 	struct dlm_message *ms;
2844 	struct dlm_mhandle *mh;
2845 	int to_nodeid, error;
2846 
2847 	error = add_to_waiters(lkb, mstype);
2848 	if (error)
2849 		return error;
2850 
2851 	to_nodeid = r->res_nodeid;
2852 
2853 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2854 	if (error)
2855 		goto fail;
2856 
2857 	send_args(r, lkb, ms);
2858 
2859 	error = send_message(mh, ms);
2860 	if (error)
2861 		goto fail;
2862 	return 0;
2863 
2864  fail:
2865 	remove_from_waiters(lkb, msg_reply_type(mstype));
2866 	return error;
2867 }
2868 
send_request(struct dlm_rsb * r,struct dlm_lkb * lkb)2869 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2870 {
2871 	return send_common(r, lkb, DLM_MSG_REQUEST);
2872 }
2873 
send_convert(struct dlm_rsb * r,struct dlm_lkb * lkb)2874 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2875 {
2876 	int error;
2877 
2878 	error = send_common(r, lkb, DLM_MSG_CONVERT);
2879 
2880 	/* down conversions go without a reply from the master */
2881 	if (!error && down_conversion(lkb)) {
2882 		remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
2883 		r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
2884 		r->res_ls->ls_stub_ms.m_result = 0;
2885 		r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
2886 		__receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
2887 	}
2888 
2889 	return error;
2890 }
2891 
2892 /* FIXME: if this lkb is the only lock we hold on the rsb, then set
2893    MASTER_UNCERTAIN to force the next request on the rsb to confirm
2894    that the master is still correct. */
2895 
send_unlock(struct dlm_rsb * r,struct dlm_lkb * lkb)2896 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2897 {
2898 	return send_common(r, lkb, DLM_MSG_UNLOCK);
2899 }
2900 
send_cancel(struct dlm_rsb * r,struct dlm_lkb * lkb)2901 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2902 {
2903 	return send_common(r, lkb, DLM_MSG_CANCEL);
2904 }
2905 
send_grant(struct dlm_rsb * r,struct dlm_lkb * lkb)2906 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
2907 {
2908 	struct dlm_message *ms;
2909 	struct dlm_mhandle *mh;
2910 	int to_nodeid, error;
2911 
2912 	to_nodeid = lkb->lkb_nodeid;
2913 
2914 	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
2915 	if (error)
2916 		goto out;
2917 
2918 	send_args(r, lkb, ms);
2919 
2920 	ms->m_result = 0;
2921 
2922 	error = send_message(mh, ms);
2923  out:
2924 	return error;
2925 }
2926 
send_bast(struct dlm_rsb * r,struct dlm_lkb * lkb,int mode)2927 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
2928 {
2929 	struct dlm_message *ms;
2930 	struct dlm_mhandle *mh;
2931 	int to_nodeid, error;
2932 
2933 	to_nodeid = lkb->lkb_nodeid;
2934 
2935 	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
2936 	if (error)
2937 		goto out;
2938 
2939 	send_args(r, lkb, ms);
2940 
2941 	ms->m_bastmode = mode;
2942 
2943 	error = send_message(mh, ms);
2944  out:
2945 	return error;
2946 }
2947 
send_lookup(struct dlm_rsb * r,struct dlm_lkb * lkb)2948 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
2949 {
2950 	struct dlm_message *ms;
2951 	struct dlm_mhandle *mh;
2952 	int to_nodeid, error;
2953 
2954 	error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
2955 	if (error)
2956 		return error;
2957 
2958 	to_nodeid = dlm_dir_nodeid(r);
2959 
2960 	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
2961 	if (error)
2962 		goto fail;
2963 
2964 	send_args(r, lkb, ms);
2965 
2966 	error = send_message(mh, ms);
2967 	if (error)
2968 		goto fail;
2969 	return 0;
2970 
2971  fail:
2972 	remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
2973 	return error;
2974 }
2975 
send_remove(struct dlm_rsb * r)2976 static int send_remove(struct dlm_rsb *r)
2977 {
2978 	struct dlm_message *ms;
2979 	struct dlm_mhandle *mh;
2980 	int to_nodeid, error;
2981 
2982 	to_nodeid = dlm_dir_nodeid(r);
2983 
2984 	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
2985 	if (error)
2986 		goto out;
2987 
2988 	memcpy(ms->m_extra, r->res_name, r->res_length);
2989 	ms->m_hash = r->res_hash;
2990 
2991 	error = send_message(mh, ms);
2992  out:
2993 	return error;
2994 }
2995 
send_common_reply(struct dlm_rsb * r,struct dlm_lkb * lkb,int mstype,int rv)2996 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
2997 			     int mstype, int rv)
2998 {
2999 	struct dlm_message *ms;
3000 	struct dlm_mhandle *mh;
3001 	int to_nodeid, error;
3002 
3003 	to_nodeid = lkb->lkb_nodeid;
3004 
3005 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
3006 	if (error)
3007 		goto out;
3008 
3009 	send_args(r, lkb, ms);
3010 
3011 	ms->m_result = rv;
3012 
3013 	error = send_message(mh, ms);
3014  out:
3015 	return error;
3016 }
3017 
send_request_reply(struct dlm_rsb * r,struct dlm_lkb * lkb,int rv)3018 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3019 {
3020 	return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
3021 }
3022 
send_convert_reply(struct dlm_rsb * r,struct dlm_lkb * lkb,int rv)3023 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3024 {
3025 	return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
3026 }
3027 
send_unlock_reply(struct dlm_rsb * r,struct dlm_lkb * lkb,int rv)3028 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3029 {
3030 	return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
3031 }
3032 
send_cancel_reply(struct dlm_rsb * r,struct dlm_lkb * lkb,int rv)3033 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
3034 {
3035 	return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
3036 }
3037 
send_lookup_reply(struct dlm_ls * ls,struct dlm_message * ms_in,int ret_nodeid,int rv)3038 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
3039 			     int ret_nodeid, int rv)
3040 {
3041 	struct dlm_rsb *r = &ls->ls_stub_rsb;
3042 	struct dlm_message *ms;
3043 	struct dlm_mhandle *mh;
3044 	int error, nodeid = ms_in->m_header.h_nodeid;
3045 
3046 	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
3047 	if (error)
3048 		goto out;
3049 
3050 	ms->m_lkid = ms_in->m_lkid;
3051 	ms->m_result = rv;
3052 	ms->m_nodeid = ret_nodeid;
3053 
3054 	error = send_message(mh, ms);
3055  out:
3056 	return error;
3057 }
3058 
3059 /* which args we save from a received message depends heavily on the type
3060    of message, unlike the send side where we can safely send everything about
3061    the lkb for any type of message */
3062 
receive_flags(struct dlm_lkb * lkb,struct dlm_message * ms)3063 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
3064 {
3065 	lkb->lkb_exflags = ms->m_exflags;
3066 	lkb->lkb_sbflags = ms->m_sbflags;
3067 	lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
3068 		         (ms->m_flags & 0x0000FFFF);
3069 }
3070 
receive_flags_reply(struct dlm_lkb * lkb,struct dlm_message * ms)3071 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3072 {
3073 	lkb->lkb_sbflags = ms->m_sbflags;
3074 	lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
3075 		         (ms->m_flags & 0x0000FFFF);
3076 }
3077 
receive_extralen(struct dlm_message * ms)3078 static int receive_extralen(struct dlm_message *ms)
3079 {
3080 	return (ms->m_header.h_length - sizeof(struct dlm_message));
3081 }
3082 
receive_lvb(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_message * ms)3083 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
3084 		       struct dlm_message *ms)
3085 {
3086 	int len;
3087 
3088 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3089 		if (!lkb->lkb_lvbptr)
3090 			lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3091 		if (!lkb->lkb_lvbptr)
3092 			return -ENOMEM;
3093 		len = receive_extralen(ms);
3094 		if (len > DLM_RESNAME_MAXLEN)
3095 			len = DLM_RESNAME_MAXLEN;
3096 		memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
3097 	}
3098 	return 0;
3099 }
3100 
fake_bastfn(void * astparam,int mode)3101 static void fake_bastfn(void *astparam, int mode)
3102 {
3103 	log_print("fake_bastfn should not be called");
3104 }
3105 
fake_astfn(void * astparam)3106 static void fake_astfn(void *astparam)
3107 {
3108 	log_print("fake_astfn should not be called");
3109 }
3110 
receive_request_args(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_message * ms)3111 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3112 				struct dlm_message *ms)
3113 {
3114 	lkb->lkb_nodeid = ms->m_header.h_nodeid;
3115 	lkb->lkb_ownpid = ms->m_pid;
3116 	lkb->lkb_remid = ms->m_lkid;
3117 	lkb->lkb_grmode = DLM_LOCK_IV;
3118 	lkb->lkb_rqmode = ms->m_rqmode;
3119 
3120 	lkb->lkb_bastfn = (ms->m_asts & DLM_CB_BAST) ? &fake_bastfn : NULL;
3121 	lkb->lkb_astfn = (ms->m_asts & DLM_CB_CAST) ? &fake_astfn : NULL;
3122 
3123 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3124 		/* lkb was just created so there won't be an lvb yet */
3125 		lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
3126 		if (!lkb->lkb_lvbptr)
3127 			return -ENOMEM;
3128 	}
3129 
3130 	return 0;
3131 }
3132 
receive_convert_args(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_message * ms)3133 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3134 				struct dlm_message *ms)
3135 {
3136 	if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3137 		return -EBUSY;
3138 
3139 	if (receive_lvb(ls, lkb, ms))
3140 		return -ENOMEM;
3141 
3142 	lkb->lkb_rqmode = ms->m_rqmode;
3143 	lkb->lkb_lvbseq = ms->m_lvbseq;
3144 
3145 	return 0;
3146 }
3147 
receive_unlock_args(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_message * ms)3148 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3149 			       struct dlm_message *ms)
3150 {
3151 	if (receive_lvb(ls, lkb, ms))
3152 		return -ENOMEM;
3153 	return 0;
3154 }
3155 
3156 /* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3157    uses to send a reply and that the remote end uses to process the reply. */
3158 
setup_stub_lkb(struct dlm_ls * ls,struct dlm_message * ms)3159 static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3160 {
3161 	struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3162 	lkb->lkb_nodeid = ms->m_header.h_nodeid;
3163 	lkb->lkb_remid = ms->m_lkid;
3164 }
3165 
3166 /* This is called after the rsb is locked so that we can safely inspect
3167    fields in the lkb. */
3168 
validate_message(struct dlm_lkb * lkb,struct dlm_message * ms)3169 static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3170 {
3171 	int from = ms->m_header.h_nodeid;
3172 	int error = 0;
3173 
3174 	switch (ms->m_type) {
3175 	case DLM_MSG_CONVERT:
3176 	case DLM_MSG_UNLOCK:
3177 	case DLM_MSG_CANCEL:
3178 		if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3179 			error = -EINVAL;
3180 		break;
3181 
3182 	case DLM_MSG_CONVERT_REPLY:
3183 	case DLM_MSG_UNLOCK_REPLY:
3184 	case DLM_MSG_CANCEL_REPLY:
3185 	case DLM_MSG_GRANT:
3186 	case DLM_MSG_BAST:
3187 		if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3188 			error = -EINVAL;
3189 		break;
3190 
3191 	case DLM_MSG_REQUEST_REPLY:
3192 		if (!is_process_copy(lkb))
3193 			error = -EINVAL;
3194 		else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
3195 			error = -EINVAL;
3196 		break;
3197 
3198 	default:
3199 		error = -EINVAL;
3200 	}
3201 
3202 	if (error)
3203 		log_error(lkb->lkb_resource->res_ls,
3204 			  "ignore invalid message %d from %d %x %x %x %d",
3205 			  ms->m_type, from, lkb->lkb_id, lkb->lkb_remid,
3206 			  lkb->lkb_flags, lkb->lkb_nodeid);
3207 	return error;
3208 }
3209 
receive_request(struct dlm_ls * ls,struct dlm_message * ms)3210 static void receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3211 {
3212 	struct dlm_lkb *lkb;
3213 	struct dlm_rsb *r;
3214 	int error, namelen;
3215 
3216 	error = create_lkb(ls, &lkb);
3217 	if (error)
3218 		goto fail;
3219 
3220 	receive_flags(lkb, ms);
3221 	lkb->lkb_flags |= DLM_IFL_MSTCPY;
3222 	error = receive_request_args(ls, lkb, ms);
3223 	if (error) {
3224 		__put_lkb(ls, lkb);
3225 		goto fail;
3226 	}
3227 
3228 	namelen = receive_extralen(ms);
3229 
3230 	error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r);
3231 	if (error) {
3232 		__put_lkb(ls, lkb);
3233 		goto fail;
3234 	}
3235 
3236 	lock_rsb(r);
3237 
3238 	attach_lkb(r, lkb);
3239 	error = do_request(r, lkb);
3240 	send_request_reply(r, lkb, error);
3241 	do_request_effects(r, lkb, error);
3242 
3243 	unlock_rsb(r);
3244 	put_rsb(r);
3245 
3246 	if (error == -EINPROGRESS)
3247 		error = 0;
3248 	if (error)
3249 		dlm_put_lkb(lkb);
3250 	return;
3251 
3252  fail:
3253 	setup_stub_lkb(ls, ms);
3254 	send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3255 }
3256 
receive_convert(struct dlm_ls * ls,struct dlm_message * ms)3257 static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3258 {
3259 	struct dlm_lkb *lkb;
3260 	struct dlm_rsb *r;
3261 	int error, reply = 1;
3262 
3263 	error = find_lkb(ls, ms->m_remid, &lkb);
3264 	if (error)
3265 		goto fail;
3266 
3267 	r = lkb->lkb_resource;
3268 
3269 	hold_rsb(r);
3270 	lock_rsb(r);
3271 
3272 	error = validate_message(lkb, ms);
3273 	if (error)
3274 		goto out;
3275 
3276 	receive_flags(lkb, ms);
3277 
3278 	error = receive_convert_args(ls, lkb, ms);
3279 	if (error) {
3280 		send_convert_reply(r, lkb, error);
3281 		goto out;
3282 	}
3283 
3284 	reply = !down_conversion(lkb);
3285 
3286 	error = do_convert(r, lkb);
3287 	if (reply)
3288 		send_convert_reply(r, lkb, error);
3289 	do_convert_effects(r, lkb, error);
3290  out:
3291 	unlock_rsb(r);
3292 	put_rsb(r);
3293 	dlm_put_lkb(lkb);
3294 	return;
3295 
3296  fail:
3297 	setup_stub_lkb(ls, ms);
3298 	send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3299 }
3300 
receive_unlock(struct dlm_ls * ls,struct dlm_message * ms)3301 static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3302 {
3303 	struct dlm_lkb *lkb;
3304 	struct dlm_rsb *r;
3305 	int error;
3306 
3307 	error = find_lkb(ls, ms->m_remid, &lkb);
3308 	if (error)
3309 		goto fail;
3310 
3311 	r = lkb->lkb_resource;
3312 
3313 	hold_rsb(r);
3314 	lock_rsb(r);
3315 
3316 	error = validate_message(lkb, ms);
3317 	if (error)
3318 		goto out;
3319 
3320 	receive_flags(lkb, ms);
3321 
3322 	error = receive_unlock_args(ls, lkb, ms);
3323 	if (error) {
3324 		send_unlock_reply(r, lkb, error);
3325 		goto out;
3326 	}
3327 
3328 	error = do_unlock(r, lkb);
3329 	send_unlock_reply(r, lkb, error);
3330 	do_unlock_effects(r, lkb, error);
3331  out:
3332 	unlock_rsb(r);
3333 	put_rsb(r);
3334 	dlm_put_lkb(lkb);
3335 	return;
3336 
3337  fail:
3338 	setup_stub_lkb(ls, ms);
3339 	send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3340 }
3341 
receive_cancel(struct dlm_ls * ls,struct dlm_message * ms)3342 static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
3343 {
3344 	struct dlm_lkb *lkb;
3345 	struct dlm_rsb *r;
3346 	int error;
3347 
3348 	error = find_lkb(ls, ms->m_remid, &lkb);
3349 	if (error)
3350 		goto fail;
3351 
3352 	receive_flags(lkb, ms);
3353 
3354 	r = lkb->lkb_resource;
3355 
3356 	hold_rsb(r);
3357 	lock_rsb(r);
3358 
3359 	error = validate_message(lkb, ms);
3360 	if (error)
3361 		goto out;
3362 
3363 	error = do_cancel(r, lkb);
3364 	send_cancel_reply(r, lkb, error);
3365 	do_cancel_effects(r, lkb, error);
3366  out:
3367 	unlock_rsb(r);
3368 	put_rsb(r);
3369 	dlm_put_lkb(lkb);
3370 	return;
3371 
3372  fail:
3373 	setup_stub_lkb(ls, ms);
3374 	send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3375 }
3376 
receive_grant(struct dlm_ls * ls,struct dlm_message * ms)3377 static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
3378 {
3379 	struct dlm_lkb *lkb;
3380 	struct dlm_rsb *r;
3381 	int error;
3382 
3383 	error = find_lkb(ls, ms->m_remid, &lkb);
3384 	if (error) {
3385 		log_debug(ls, "receive_grant from %d no lkb %x",
3386 			  ms->m_header.h_nodeid, ms->m_remid);
3387 		return;
3388 	}
3389 
3390 	r = lkb->lkb_resource;
3391 
3392 	hold_rsb(r);
3393 	lock_rsb(r);
3394 
3395 	error = validate_message(lkb, ms);
3396 	if (error)
3397 		goto out;
3398 
3399 	receive_flags_reply(lkb, ms);
3400 	if (is_altmode(lkb))
3401 		munge_altmode(lkb, ms);
3402 	grant_lock_pc(r, lkb, ms);
3403 	queue_cast(r, lkb, 0);
3404  out:
3405 	unlock_rsb(r);
3406 	put_rsb(r);
3407 	dlm_put_lkb(lkb);
3408 }
3409 
receive_bast(struct dlm_ls * ls,struct dlm_message * ms)3410 static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
3411 {
3412 	struct dlm_lkb *lkb;
3413 	struct dlm_rsb *r;
3414 	int error;
3415 
3416 	error = find_lkb(ls, ms->m_remid, &lkb);
3417 	if (error) {
3418 		log_debug(ls, "receive_bast from %d no lkb %x",
3419 			  ms->m_header.h_nodeid, ms->m_remid);
3420 		return;
3421 	}
3422 
3423 	r = lkb->lkb_resource;
3424 
3425 	hold_rsb(r);
3426 	lock_rsb(r);
3427 
3428 	error = validate_message(lkb, ms);
3429 	if (error)
3430 		goto out;
3431 
3432 	queue_bast(r, lkb, ms->m_bastmode);
3433  out:
3434 	unlock_rsb(r);
3435 	put_rsb(r);
3436 	dlm_put_lkb(lkb);
3437 }
3438 
receive_lookup(struct dlm_ls * ls,struct dlm_message * ms)3439 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
3440 {
3441 	int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid;
3442 
3443 	from_nodeid = ms->m_header.h_nodeid;
3444 	our_nodeid = dlm_our_nodeid();
3445 
3446 	len = receive_extralen(ms);
3447 
3448 	dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3449 	if (dir_nodeid != our_nodeid) {
3450 		log_error(ls, "lookup dir_nodeid %d from %d",
3451 			  dir_nodeid, from_nodeid);
3452 		error = -EINVAL;
3453 		ret_nodeid = -1;
3454 		goto out;
3455 	}
3456 
3457 	error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid);
3458 
3459 	/* Optimization: we're master so treat lookup as a request */
3460 	if (!error && ret_nodeid == our_nodeid) {
3461 		receive_request(ls, ms);
3462 		return;
3463 	}
3464  out:
3465 	send_lookup_reply(ls, ms, ret_nodeid, error);
3466 }
3467 
receive_remove(struct dlm_ls * ls,struct dlm_message * ms)3468 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
3469 {
3470 	int len, dir_nodeid, from_nodeid;
3471 
3472 	from_nodeid = ms->m_header.h_nodeid;
3473 
3474 	len = receive_extralen(ms);
3475 
3476 	dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3477 	if (dir_nodeid != dlm_our_nodeid()) {
3478 		log_error(ls, "remove dir entry dir_nodeid %d from %d",
3479 			  dir_nodeid, from_nodeid);
3480 		return;
3481 	}
3482 
3483 	dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
3484 }
3485 
receive_purge(struct dlm_ls * ls,struct dlm_message * ms)3486 static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
3487 {
3488 	do_purge(ls, ms->m_nodeid, ms->m_pid);
3489 }
3490 
receive_request_reply(struct dlm_ls * ls,struct dlm_message * ms)3491 static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
3492 {
3493 	struct dlm_lkb *lkb;
3494 	struct dlm_rsb *r;
3495 	int error, mstype, result;
3496 
3497 	error = find_lkb(ls, ms->m_remid, &lkb);
3498 	if (error) {
3499 		log_debug(ls, "receive_request_reply from %d no lkb %x",
3500 			  ms->m_header.h_nodeid, ms->m_remid);
3501 		return;
3502 	}
3503 
3504 	r = lkb->lkb_resource;
3505 	hold_rsb(r);
3506 	lock_rsb(r);
3507 
3508 	error = validate_message(lkb, ms);
3509 	if (error)
3510 		goto out;
3511 
3512 	mstype = lkb->lkb_wait_type;
3513 	error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
3514 	if (error)
3515 		goto out;
3516 
3517 	/* Optimization: the dir node was also the master, so it took our
3518 	   lookup as a request and sent request reply instead of lookup reply */
3519 	if (mstype == DLM_MSG_LOOKUP) {
3520 		r->res_nodeid = ms->m_header.h_nodeid;
3521 		lkb->lkb_nodeid = r->res_nodeid;
3522 	}
3523 
3524 	/* this is the value returned from do_request() on the master */
3525 	result = ms->m_result;
3526 
3527 	switch (result) {
3528 	case -EAGAIN:
3529 		/* request would block (be queued) on remote master */
3530 		queue_cast(r, lkb, -EAGAIN);
3531 		confirm_master(r, -EAGAIN);
3532 		unhold_lkb(lkb); /* undoes create_lkb() */
3533 		break;
3534 
3535 	case -EINPROGRESS:
3536 	case 0:
3537 		/* request was queued or granted on remote master */
3538 		receive_flags_reply(lkb, ms);
3539 		lkb->lkb_remid = ms->m_lkid;
3540 		if (is_altmode(lkb))
3541 			munge_altmode(lkb, ms);
3542 		if (result) {
3543 			add_lkb(r, lkb, DLM_LKSTS_WAITING);
3544 			add_timeout(lkb);
3545 		} else {
3546 			grant_lock_pc(r, lkb, ms);
3547 			queue_cast(r, lkb, 0);
3548 		}
3549 		confirm_master(r, result);
3550 		break;
3551 
3552 	case -EBADR:
3553 	case -ENOTBLK:
3554 		/* find_rsb failed to find rsb or rsb wasn't master */
3555 		log_debug(ls, "receive_request_reply %x %x master diff %d %d",
3556 			  lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
3557 		r->res_nodeid = -1;
3558 		lkb->lkb_nodeid = -1;
3559 
3560 		if (is_overlap(lkb)) {
3561 			/* we'll ignore error in cancel/unlock reply */
3562 			queue_cast_overlap(r, lkb);
3563 			confirm_master(r, result);
3564 			unhold_lkb(lkb); /* undoes create_lkb() */
3565 		} else
3566 			_request_lock(r, lkb);
3567 		break;
3568 
3569 	default:
3570 		log_error(ls, "receive_request_reply %x error %d",
3571 			  lkb->lkb_id, result);
3572 	}
3573 
3574 	if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
3575 		log_debug(ls, "receive_request_reply %x result %d unlock",
3576 			  lkb->lkb_id, result);
3577 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3578 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3579 		send_unlock(r, lkb);
3580 	} else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
3581 		log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
3582 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3583 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3584 		send_cancel(r, lkb);
3585 	} else {
3586 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3587 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3588 	}
3589  out:
3590 	unlock_rsb(r);
3591 	put_rsb(r);
3592 	dlm_put_lkb(lkb);
3593 }
3594 
__receive_convert_reply(struct dlm_rsb * r,struct dlm_lkb * lkb,struct dlm_message * ms)3595 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3596 				    struct dlm_message *ms)
3597 {
3598 	/* this is the value returned from do_convert() on the master */
3599 	switch (ms->m_result) {
3600 	case -EAGAIN:
3601 		/* convert would block (be queued) on remote master */
3602 		queue_cast(r, lkb, -EAGAIN);
3603 		break;
3604 
3605 	case -EDEADLK:
3606 		receive_flags_reply(lkb, ms);
3607 		revert_lock_pc(r, lkb);
3608 		queue_cast(r, lkb, -EDEADLK);
3609 		break;
3610 
3611 	case -EINPROGRESS:
3612 		/* convert was queued on remote master */
3613 		receive_flags_reply(lkb, ms);
3614 		if (is_demoted(lkb))
3615 			munge_demoted(lkb, ms);
3616 		del_lkb(r, lkb);
3617 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3618 		add_timeout(lkb);
3619 		break;
3620 
3621 	case 0:
3622 		/* convert was granted on remote master */
3623 		receive_flags_reply(lkb, ms);
3624 		if (is_demoted(lkb))
3625 			munge_demoted(lkb, ms);
3626 		grant_lock_pc(r, lkb, ms);
3627 		queue_cast(r, lkb, 0);
3628 		break;
3629 
3630 	default:
3631 		log_error(r->res_ls, "receive_convert_reply %x error %d",
3632 			  lkb->lkb_id, ms->m_result);
3633 	}
3634 }
3635 
_receive_convert_reply(struct dlm_lkb * lkb,struct dlm_message * ms)3636 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3637 {
3638 	struct dlm_rsb *r = lkb->lkb_resource;
3639 	int error;
3640 
3641 	hold_rsb(r);
3642 	lock_rsb(r);
3643 
3644 	error = validate_message(lkb, ms);
3645 	if (error)
3646 		goto out;
3647 
3648 	/* stub reply can happen with waiters_mutex held */
3649 	error = remove_from_waiters_ms(lkb, ms);
3650 	if (error)
3651 		goto out;
3652 
3653 	__receive_convert_reply(r, lkb, ms);
3654  out:
3655 	unlock_rsb(r);
3656 	put_rsb(r);
3657 }
3658 
receive_convert_reply(struct dlm_ls * ls,struct dlm_message * ms)3659 static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
3660 {
3661 	struct dlm_lkb *lkb;
3662 	int error;
3663 
3664 	error = find_lkb(ls, ms->m_remid, &lkb);
3665 	if (error) {
3666 		log_debug(ls, "receive_convert_reply from %d no lkb %x",
3667 			  ms->m_header.h_nodeid, ms->m_remid);
3668 		return;
3669 	}
3670 
3671 	_receive_convert_reply(lkb, ms);
3672 	dlm_put_lkb(lkb);
3673 }
3674 
_receive_unlock_reply(struct dlm_lkb * lkb,struct dlm_message * ms)3675 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3676 {
3677 	struct dlm_rsb *r = lkb->lkb_resource;
3678 	int error;
3679 
3680 	hold_rsb(r);
3681 	lock_rsb(r);
3682 
3683 	error = validate_message(lkb, ms);
3684 	if (error)
3685 		goto out;
3686 
3687 	/* stub reply can happen with waiters_mutex held */
3688 	error = remove_from_waiters_ms(lkb, ms);
3689 	if (error)
3690 		goto out;
3691 
3692 	/* this is the value returned from do_unlock() on the master */
3693 
3694 	switch (ms->m_result) {
3695 	case -DLM_EUNLOCK:
3696 		receive_flags_reply(lkb, ms);
3697 		remove_lock_pc(r, lkb);
3698 		queue_cast(r, lkb, -DLM_EUNLOCK);
3699 		break;
3700 	case -ENOENT:
3701 		break;
3702 	default:
3703 		log_error(r->res_ls, "receive_unlock_reply %x error %d",
3704 			  lkb->lkb_id, ms->m_result);
3705 	}
3706  out:
3707 	unlock_rsb(r);
3708 	put_rsb(r);
3709 }
3710 
receive_unlock_reply(struct dlm_ls * ls,struct dlm_message * ms)3711 static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
3712 {
3713 	struct dlm_lkb *lkb;
3714 	int error;
3715 
3716 	error = find_lkb(ls, ms->m_remid, &lkb);
3717 	if (error) {
3718 		log_debug(ls, "receive_unlock_reply from %d no lkb %x",
3719 			  ms->m_header.h_nodeid, ms->m_remid);
3720 		return;
3721 	}
3722 
3723 	_receive_unlock_reply(lkb, ms);
3724 	dlm_put_lkb(lkb);
3725 }
3726 
_receive_cancel_reply(struct dlm_lkb * lkb,struct dlm_message * ms)3727 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3728 {
3729 	struct dlm_rsb *r = lkb->lkb_resource;
3730 	int error;
3731 
3732 	hold_rsb(r);
3733 	lock_rsb(r);
3734 
3735 	error = validate_message(lkb, ms);
3736 	if (error)
3737 		goto out;
3738 
3739 	/* stub reply can happen with waiters_mutex held */
3740 	error = remove_from_waiters_ms(lkb, ms);
3741 	if (error)
3742 		goto out;
3743 
3744 	/* this is the value returned from do_cancel() on the master */
3745 
3746 	switch (ms->m_result) {
3747 	case -DLM_ECANCEL:
3748 		receive_flags_reply(lkb, ms);
3749 		revert_lock_pc(r, lkb);
3750 		queue_cast(r, lkb, -DLM_ECANCEL);
3751 		break;
3752 	case 0:
3753 		break;
3754 	default:
3755 		log_error(r->res_ls, "receive_cancel_reply %x error %d",
3756 			  lkb->lkb_id, ms->m_result);
3757 	}
3758  out:
3759 	unlock_rsb(r);
3760 	put_rsb(r);
3761 }
3762 
receive_cancel_reply(struct dlm_ls * ls,struct dlm_message * ms)3763 static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
3764 {
3765 	struct dlm_lkb *lkb;
3766 	int error;
3767 
3768 	error = find_lkb(ls, ms->m_remid, &lkb);
3769 	if (error) {
3770 		log_debug(ls, "receive_cancel_reply from %d no lkb %x",
3771 			  ms->m_header.h_nodeid, ms->m_remid);
3772 		return;
3773 	}
3774 
3775 	_receive_cancel_reply(lkb, ms);
3776 	dlm_put_lkb(lkb);
3777 }
3778 
receive_lookup_reply(struct dlm_ls * ls,struct dlm_message * ms)3779 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
3780 {
3781 	struct dlm_lkb *lkb;
3782 	struct dlm_rsb *r;
3783 	int error, ret_nodeid;
3784 
3785 	error = find_lkb(ls, ms->m_lkid, &lkb);
3786 	if (error) {
3787 		log_error(ls, "receive_lookup_reply no lkb");
3788 		return;
3789 	}
3790 
3791 	/* ms->m_result is the value returned by dlm_dir_lookup on dir node
3792 	   FIXME: will a non-zero error ever be returned? */
3793 
3794 	r = lkb->lkb_resource;
3795 	hold_rsb(r);
3796 	lock_rsb(r);
3797 
3798 	error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3799 	if (error)
3800 		goto out;
3801 
3802 	ret_nodeid = ms->m_nodeid;
3803 	if (ret_nodeid == dlm_our_nodeid()) {
3804 		r->res_nodeid = 0;
3805 		ret_nodeid = 0;
3806 		r->res_first_lkid = 0;
3807 	} else {
3808 		/* set_master() will copy res_nodeid to lkb_nodeid */
3809 		r->res_nodeid = ret_nodeid;
3810 	}
3811 
3812 	if (is_overlap(lkb)) {
3813 		log_debug(ls, "receive_lookup_reply %x unlock %x",
3814 			  lkb->lkb_id, lkb->lkb_flags);
3815 		queue_cast_overlap(r, lkb);
3816 		unhold_lkb(lkb); /* undoes create_lkb() */
3817 		goto out_list;
3818 	}
3819 
3820 	_request_lock(r, lkb);
3821 
3822  out_list:
3823 	if (!ret_nodeid)
3824 		process_lookup_list(r);
3825  out:
3826 	unlock_rsb(r);
3827 	put_rsb(r);
3828 	dlm_put_lkb(lkb);
3829 }
3830 
_receive_message(struct dlm_ls * ls,struct dlm_message * ms)3831 static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
3832 {
3833 	if (!dlm_is_member(ls, ms->m_header.h_nodeid)) {
3834 		log_debug(ls, "ignore non-member message %d from %d %x %x %d",
3835 			  ms->m_type, ms->m_header.h_nodeid, ms->m_lkid,
3836 			  ms->m_remid, ms->m_result);
3837 		return;
3838 	}
3839 
3840 	switch (ms->m_type) {
3841 
3842 	/* messages sent to a master node */
3843 
3844 	case DLM_MSG_REQUEST:
3845 		receive_request(ls, ms);
3846 		break;
3847 
3848 	case DLM_MSG_CONVERT:
3849 		receive_convert(ls, ms);
3850 		break;
3851 
3852 	case DLM_MSG_UNLOCK:
3853 		receive_unlock(ls, ms);
3854 		break;
3855 
3856 	case DLM_MSG_CANCEL:
3857 		receive_cancel(ls, ms);
3858 		break;
3859 
3860 	/* messages sent from a master node (replies to above) */
3861 
3862 	case DLM_MSG_REQUEST_REPLY:
3863 		receive_request_reply(ls, ms);
3864 		break;
3865 
3866 	case DLM_MSG_CONVERT_REPLY:
3867 		receive_convert_reply(ls, ms);
3868 		break;
3869 
3870 	case DLM_MSG_UNLOCK_REPLY:
3871 		receive_unlock_reply(ls, ms);
3872 		break;
3873 
3874 	case DLM_MSG_CANCEL_REPLY:
3875 		receive_cancel_reply(ls, ms);
3876 		break;
3877 
3878 	/* messages sent from a master node (only two types of async msg) */
3879 
3880 	case DLM_MSG_GRANT:
3881 		receive_grant(ls, ms);
3882 		break;
3883 
3884 	case DLM_MSG_BAST:
3885 		receive_bast(ls, ms);
3886 		break;
3887 
3888 	/* messages sent to a dir node */
3889 
3890 	case DLM_MSG_LOOKUP:
3891 		receive_lookup(ls, ms);
3892 		break;
3893 
3894 	case DLM_MSG_REMOVE:
3895 		receive_remove(ls, ms);
3896 		break;
3897 
3898 	/* messages sent from a dir node (remove has no reply) */
3899 
3900 	case DLM_MSG_LOOKUP_REPLY:
3901 		receive_lookup_reply(ls, ms);
3902 		break;
3903 
3904 	/* other messages */
3905 
3906 	case DLM_MSG_PURGE:
3907 		receive_purge(ls, ms);
3908 		break;
3909 
3910 	default:
3911 		log_error(ls, "unknown message type %d", ms->m_type);
3912 	}
3913 
3914 	dlm_astd_wake();
3915 }
3916 
3917 /* If the lockspace is in recovery mode (locking stopped), then normal
3918    messages are saved on the requestqueue for processing after recovery is
3919    done.  When not in recovery mode, we wait for dlm_recoverd to drain saved
3920    messages off the requestqueue before we process new ones. This occurs right
3921    after recovery completes when we transition from saving all messages on
3922    requestqueue, to processing all the saved messages, to processing new
3923    messages as they arrive. */
3924 
dlm_receive_message(struct dlm_ls * ls,struct dlm_message * ms,int nodeid)3925 static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
3926 				int nodeid)
3927 {
3928 	if (dlm_locking_stopped(ls)) {
3929 		dlm_add_requestqueue(ls, nodeid, ms);
3930 	} else {
3931 		dlm_wait_requestqueue(ls);
3932 		_receive_message(ls, ms);
3933 	}
3934 }
3935 
3936 /* This is called by dlm_recoverd to process messages that were saved on
3937    the requestqueue. */
3938 
dlm_receive_message_saved(struct dlm_ls * ls,struct dlm_message * ms)3939 void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
3940 {
3941 	_receive_message(ls, ms);
3942 }
3943 
3944 /* This is called by the midcomms layer when something is received for
3945    the lockspace.  It could be either a MSG (normal message sent as part of
3946    standard locking activity) or an RCOM (recovery message sent as part of
3947    lockspace recovery). */
3948 
dlm_receive_buffer(union dlm_packet * p,int nodeid)3949 void dlm_receive_buffer(union dlm_packet *p, int nodeid)
3950 {
3951 	struct dlm_header *hd = &p->header;
3952 	struct dlm_ls *ls;
3953 	int type = 0;
3954 
3955 	switch (hd->h_cmd) {
3956 	case DLM_MSG:
3957 		dlm_message_in(&p->message);
3958 		type = p->message.m_type;
3959 		break;
3960 	case DLM_RCOM:
3961 		dlm_rcom_in(&p->rcom);
3962 		type = p->rcom.rc_type;
3963 		break;
3964 	default:
3965 		log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
3966 		return;
3967 	}
3968 
3969 	if (hd->h_nodeid != nodeid) {
3970 		log_print("invalid h_nodeid %d from %d lockspace %x",
3971 			  hd->h_nodeid, nodeid, hd->h_lockspace);
3972 		return;
3973 	}
3974 
3975 	ls = dlm_find_lockspace_global(hd->h_lockspace);
3976 	if (!ls) {
3977 		if (dlm_config.ci_log_debug)
3978 			log_print("invalid lockspace %x from %d cmd %d type %d",
3979 				  hd->h_lockspace, nodeid, hd->h_cmd, type);
3980 
3981 		if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
3982 			dlm_send_ls_not_ready(nodeid, &p->rcom);
3983 		return;
3984 	}
3985 
3986 	/* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3987 	   be inactive (in this ls) before transitioning to recovery mode */
3988 
3989 	down_read(&ls->ls_recv_active);
3990 	if (hd->h_cmd == DLM_MSG)
3991 		dlm_receive_message(ls, &p->message, nodeid);
3992 	else
3993 		dlm_receive_rcom(ls, &p->rcom, nodeid);
3994 	up_read(&ls->ls_recv_active);
3995 
3996 	dlm_put_lockspace(ls);
3997 }
3998 
recover_convert_waiter(struct dlm_ls * ls,struct dlm_lkb * lkb)3999 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
4000 {
4001 	if (middle_conversion(lkb)) {
4002 		hold_lkb(lkb);
4003 		ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
4004 		ls->ls_stub_ms.m_result = -EINPROGRESS;
4005 		ls->ls_stub_ms.m_flags = lkb->lkb_flags;
4006 		ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
4007 		_receive_convert_reply(lkb, &ls->ls_stub_ms);
4008 
4009 		/* Same special case as in receive_rcom_lock_args() */
4010 		lkb->lkb_grmode = DLM_LOCK_IV;
4011 		rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
4012 		unhold_lkb(lkb);
4013 
4014 	} else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
4015 		lkb->lkb_flags |= DLM_IFL_RESEND;
4016 	}
4017 
4018 	/* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
4019 	   conversions are async; there's no reply from the remote master */
4020 }
4021 
4022 /* A waiting lkb needs recovery if the master node has failed, or
4023    the master node is changing (only when no directory is used) */
4024 
waiter_needs_recovery(struct dlm_ls * ls,struct dlm_lkb * lkb)4025 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
4026 {
4027 	if (dlm_is_removed(ls, lkb->lkb_nodeid))
4028 		return 1;
4029 
4030 	if (!dlm_no_directory(ls))
4031 		return 0;
4032 
4033 	if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid)
4034 		return 1;
4035 
4036 	return 0;
4037 }
4038 
4039 /* Recovery for locks that are waiting for replies from nodes that are now
4040    gone.  We can just complete unlocks and cancels by faking a reply from the
4041    dead node.  Requests and up-conversions we flag to be resent after
4042    recovery.  Down-conversions can just be completed with a fake reply like
4043    unlocks.  Conversions between PR and CW need special attention. */
4044 
dlm_recover_waiters_pre(struct dlm_ls * ls)4045 void dlm_recover_waiters_pre(struct dlm_ls *ls)
4046 {
4047 	struct dlm_lkb *lkb, *safe;
4048 	int wait_type, stub_unlock_result, stub_cancel_result;
4049 
4050 	mutex_lock(&ls->ls_waiters_mutex);
4051 
4052 	list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
4053 		log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
4054 			  lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
4055 
4056 		/* all outstanding lookups, regardless of destination  will be
4057 		   resent after recovery is done */
4058 
4059 		if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
4060 			lkb->lkb_flags |= DLM_IFL_RESEND;
4061 			continue;
4062 		}
4063 
4064 		if (!waiter_needs_recovery(ls, lkb))
4065 			continue;
4066 
4067 		wait_type = lkb->lkb_wait_type;
4068 		stub_unlock_result = -DLM_EUNLOCK;
4069 		stub_cancel_result = -DLM_ECANCEL;
4070 
4071 		/* Main reply may have been received leaving a zero wait_type,
4072 		   but a reply for the overlapping op may not have been
4073 		   received.  In that case we need to fake the appropriate
4074 		   reply for the overlap op. */
4075 
4076 		if (!wait_type) {
4077 			if (is_overlap_cancel(lkb)) {
4078 				wait_type = DLM_MSG_CANCEL;
4079 				if (lkb->lkb_grmode == DLM_LOCK_IV)
4080 					stub_cancel_result = 0;
4081 			}
4082 			if (is_overlap_unlock(lkb)) {
4083 				wait_type = DLM_MSG_UNLOCK;
4084 				if (lkb->lkb_grmode == DLM_LOCK_IV)
4085 					stub_unlock_result = -ENOENT;
4086 			}
4087 
4088 			log_debug(ls, "rwpre overlap %x %x %d %d %d",
4089 				  lkb->lkb_id, lkb->lkb_flags, wait_type,
4090 				  stub_cancel_result, stub_unlock_result);
4091 		}
4092 
4093 		switch (wait_type) {
4094 
4095 		case DLM_MSG_REQUEST:
4096 			lkb->lkb_flags |= DLM_IFL_RESEND;
4097 			break;
4098 
4099 		case DLM_MSG_CONVERT:
4100 			recover_convert_waiter(ls, lkb);
4101 			break;
4102 
4103 		case DLM_MSG_UNLOCK:
4104 			hold_lkb(lkb);
4105 			ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
4106 			ls->ls_stub_ms.m_result = stub_unlock_result;
4107 			ls->ls_stub_ms.m_flags = lkb->lkb_flags;
4108 			ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
4109 			_receive_unlock_reply(lkb, &ls->ls_stub_ms);
4110 			dlm_put_lkb(lkb);
4111 			break;
4112 
4113 		case DLM_MSG_CANCEL:
4114 			hold_lkb(lkb);
4115 			ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
4116 			ls->ls_stub_ms.m_result = stub_cancel_result;
4117 			ls->ls_stub_ms.m_flags = lkb->lkb_flags;
4118 			ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
4119 			_receive_cancel_reply(lkb, &ls->ls_stub_ms);
4120 			dlm_put_lkb(lkb);
4121 			break;
4122 
4123 		default:
4124 			log_error(ls, "invalid lkb wait_type %d %d",
4125 				  lkb->lkb_wait_type, wait_type);
4126 		}
4127 		schedule();
4128 	}
4129 	mutex_unlock(&ls->ls_waiters_mutex);
4130 }
4131 
find_resend_waiter(struct dlm_ls * ls)4132 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
4133 {
4134 	struct dlm_lkb *lkb;
4135 	int found = 0;
4136 
4137 	mutex_lock(&ls->ls_waiters_mutex);
4138 	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
4139 		if (lkb->lkb_flags & DLM_IFL_RESEND) {
4140 			hold_lkb(lkb);
4141 			found = 1;
4142 			break;
4143 		}
4144 	}
4145 	mutex_unlock(&ls->ls_waiters_mutex);
4146 
4147 	if (!found)
4148 		lkb = NULL;
4149 	return lkb;
4150 }
4151 
4152 /* Deal with lookups and lkb's marked RESEND from _pre.  We may now be the
4153    master or dir-node for r.  Processing the lkb may result in it being placed
4154    back on waiters. */
4155 
4156 /* We do this after normal locking has been enabled and any saved messages
4157    (in requestqueue) have been processed.  We should be confident that at
4158    this point we won't get or process a reply to any of these waiting
4159    operations.  But, new ops may be coming in on the rsbs/locks here from
4160    userspace or remotely. */
4161 
4162 /* there may have been an overlap unlock/cancel prior to recovery or after
4163    recovery.  if before, the lkb may still have a pos wait_count; if after, the
4164    overlap flag would just have been set and nothing new sent.  we can be
4165    confident here than any replies to either the initial op or overlap ops
4166    prior to recovery have been received. */
4167 
dlm_recover_waiters_post(struct dlm_ls * ls)4168 int dlm_recover_waiters_post(struct dlm_ls *ls)
4169 {
4170 	struct dlm_lkb *lkb;
4171 	struct dlm_rsb *r;
4172 	int error = 0, mstype, err, oc, ou;
4173 
4174 	while (1) {
4175 		if (dlm_locking_stopped(ls)) {
4176 			log_debug(ls, "recover_waiters_post aborted");
4177 			error = -EINTR;
4178 			break;
4179 		}
4180 
4181 		lkb = find_resend_waiter(ls);
4182 		if (!lkb)
4183 			break;
4184 
4185 		r = lkb->lkb_resource;
4186 		hold_rsb(r);
4187 		lock_rsb(r);
4188 
4189 		mstype = lkb->lkb_wait_type;
4190 		oc = is_overlap_cancel(lkb);
4191 		ou = is_overlap_unlock(lkb);
4192 		err = 0;
4193 
4194 		log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
4195 			  lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
4196 
4197 		/* At this point we assume that we won't get a reply to any
4198 		   previous op or overlap op on this lock.  First, do a big
4199 		   remove_from_waiters() for all previous ops. */
4200 
4201 		lkb->lkb_flags &= ~DLM_IFL_RESEND;
4202 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4203 		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4204 		lkb->lkb_wait_type = 0;
4205 		lkb->lkb_wait_count = 0;
4206 		mutex_lock(&ls->ls_waiters_mutex);
4207 		list_del_init(&lkb->lkb_wait_reply);
4208 		mutex_unlock(&ls->ls_waiters_mutex);
4209 		unhold_lkb(lkb); /* for waiters list */
4210 
4211 		if (oc || ou) {
4212 			/* do an unlock or cancel instead of resending */
4213 			switch (mstype) {
4214 			case DLM_MSG_LOOKUP:
4215 			case DLM_MSG_REQUEST:
4216 				queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
4217 							-DLM_ECANCEL);
4218 				unhold_lkb(lkb); /* undoes create_lkb() */
4219 				break;
4220 			case DLM_MSG_CONVERT:
4221 				if (oc) {
4222 					queue_cast(r, lkb, -DLM_ECANCEL);
4223 				} else {
4224 					lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
4225 					_unlock_lock(r, lkb);
4226 				}
4227 				break;
4228 			default:
4229 				err = 1;
4230 			}
4231 		} else {
4232 			switch (mstype) {
4233 			case DLM_MSG_LOOKUP:
4234 			case DLM_MSG_REQUEST:
4235 				_request_lock(r, lkb);
4236 				if (is_master(r))
4237 					confirm_master(r, 0);
4238 				break;
4239 			case DLM_MSG_CONVERT:
4240 				_convert_lock(r, lkb);
4241 				break;
4242 			default:
4243 				err = 1;
4244 			}
4245 		}
4246 
4247 		if (err)
4248 			log_error(ls, "recover_waiters_post %x %d %x %d %d",
4249 			  	  lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
4250 		unlock_rsb(r);
4251 		put_rsb(r);
4252 		dlm_put_lkb(lkb);
4253 	}
4254 
4255 	return error;
4256 }
4257 
purge_queue(struct dlm_rsb * r,struct list_head * queue,int (* test)(struct dlm_ls * ls,struct dlm_lkb * lkb))4258 static void purge_queue(struct dlm_rsb *r, struct list_head *queue,
4259 			int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb))
4260 {
4261 	struct dlm_ls *ls = r->res_ls;
4262 	struct dlm_lkb *lkb, *safe;
4263 
4264 	list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) {
4265 		if (test(ls, lkb)) {
4266 			rsb_set_flag(r, RSB_LOCKS_PURGED);
4267 			del_lkb(r, lkb);
4268 			/* this put should free the lkb */
4269 			if (!dlm_put_lkb(lkb))
4270 				log_error(ls, "purged lkb not released");
4271 		}
4272 	}
4273 }
4274 
purge_dead_test(struct dlm_ls * ls,struct dlm_lkb * lkb)4275 static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4276 {
4277 	return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid));
4278 }
4279 
purge_mstcpy_test(struct dlm_ls * ls,struct dlm_lkb * lkb)4280 static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4281 {
4282 	return is_master_copy(lkb);
4283 }
4284 
purge_dead_locks(struct dlm_rsb * r)4285 static void purge_dead_locks(struct dlm_rsb *r)
4286 {
4287 	purge_queue(r, &r->res_grantqueue, &purge_dead_test);
4288 	purge_queue(r, &r->res_convertqueue, &purge_dead_test);
4289 	purge_queue(r, &r->res_waitqueue, &purge_dead_test);
4290 }
4291 
dlm_purge_mstcpy_locks(struct dlm_rsb * r)4292 void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
4293 {
4294 	purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test);
4295 	purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test);
4296 	purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test);
4297 }
4298 
4299 /* Get rid of locks held by nodes that are gone. */
4300 
dlm_purge_locks(struct dlm_ls * ls)4301 int dlm_purge_locks(struct dlm_ls *ls)
4302 {
4303 	struct dlm_rsb *r;
4304 
4305 	log_debug(ls, "dlm_purge_locks");
4306 
4307 	down_write(&ls->ls_root_sem);
4308 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
4309 		hold_rsb(r);
4310 		lock_rsb(r);
4311 		if (is_master(r))
4312 			purge_dead_locks(r);
4313 		unlock_rsb(r);
4314 		unhold_rsb(r);
4315 
4316 		schedule();
4317 	}
4318 	up_write(&ls->ls_root_sem);
4319 
4320 	return 0;
4321 }
4322 
find_purged_rsb(struct dlm_ls * ls,int bucket)4323 static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket)
4324 {
4325 	struct dlm_rsb *r, *r_ret = NULL;
4326 
4327 	spin_lock(&ls->ls_rsbtbl[bucket].lock);
4328 	list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) {
4329 		if (!rsb_flag(r, RSB_LOCKS_PURGED))
4330 			continue;
4331 		hold_rsb(r);
4332 		rsb_clear_flag(r, RSB_LOCKS_PURGED);
4333 		r_ret = r;
4334 		break;
4335 	}
4336 	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
4337 	return r_ret;
4338 }
4339 
dlm_grant_after_purge(struct dlm_ls * ls)4340 void dlm_grant_after_purge(struct dlm_ls *ls)
4341 {
4342 	struct dlm_rsb *r;
4343 	int bucket = 0;
4344 
4345 	while (1) {
4346 		r = find_purged_rsb(ls, bucket);
4347 		if (!r) {
4348 			if (bucket == ls->ls_rsbtbl_size - 1)
4349 				break;
4350 			bucket++;
4351 			continue;
4352 		}
4353 		lock_rsb(r);
4354 		if (is_master(r)) {
4355 			grant_pending_locks(r);
4356 			confirm_master(r, 0);
4357 		}
4358 		unlock_rsb(r);
4359 		put_rsb(r);
4360 		schedule();
4361 	}
4362 }
4363 
search_remid_list(struct list_head * head,int nodeid,uint32_t remid)4364 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
4365 					 uint32_t remid)
4366 {
4367 	struct dlm_lkb *lkb;
4368 
4369 	list_for_each_entry(lkb, head, lkb_statequeue) {
4370 		if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
4371 			return lkb;
4372 	}
4373 	return NULL;
4374 }
4375 
search_remid(struct dlm_rsb * r,int nodeid,uint32_t remid)4376 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
4377 				    uint32_t remid)
4378 {
4379 	struct dlm_lkb *lkb;
4380 
4381 	lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
4382 	if (lkb)
4383 		return lkb;
4384 	lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
4385 	if (lkb)
4386 		return lkb;
4387 	lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
4388 	if (lkb)
4389 		return lkb;
4390 	return NULL;
4391 }
4392 
4393 /* needs at least dlm_rcom + rcom_lock */
receive_rcom_lock_args(struct dlm_ls * ls,struct dlm_lkb * lkb,struct dlm_rsb * r,struct dlm_rcom * rc)4394 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
4395 				  struct dlm_rsb *r, struct dlm_rcom *rc)
4396 {
4397 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4398 
4399 	lkb->lkb_nodeid = rc->rc_header.h_nodeid;
4400 	lkb->lkb_ownpid = le32_to_cpu(rl->rl_ownpid);
4401 	lkb->lkb_remid = le32_to_cpu(rl->rl_lkid);
4402 	lkb->lkb_exflags = le32_to_cpu(rl->rl_exflags);
4403 	lkb->lkb_flags = le32_to_cpu(rl->rl_flags) & 0x0000FFFF;
4404 	lkb->lkb_flags |= DLM_IFL_MSTCPY;
4405 	lkb->lkb_lvbseq = le32_to_cpu(rl->rl_lvbseq);
4406 	lkb->lkb_rqmode = rl->rl_rqmode;
4407 	lkb->lkb_grmode = rl->rl_grmode;
4408 	/* don't set lkb_status because add_lkb wants to itself */
4409 
4410 	lkb->lkb_bastfn = (rl->rl_asts & DLM_CB_BAST) ? &fake_bastfn : NULL;
4411 	lkb->lkb_astfn = (rl->rl_asts & DLM_CB_CAST) ? &fake_astfn : NULL;
4412 
4413 	if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
4414 		int lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) -
4415 			 sizeof(struct rcom_lock);
4416 		if (lvblen > ls->ls_lvblen)
4417 			return -EINVAL;
4418 		lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
4419 		if (!lkb->lkb_lvbptr)
4420 			return -ENOMEM;
4421 		memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
4422 	}
4423 
4424 	/* Conversions between PR and CW (middle modes) need special handling.
4425 	   The real granted mode of these converting locks cannot be determined
4426 	   until all locks have been rebuilt on the rsb (recover_conversion) */
4427 
4428 	if (rl->rl_wait_type == cpu_to_le16(DLM_MSG_CONVERT) &&
4429 	    middle_conversion(lkb)) {
4430 		rl->rl_status = DLM_LKSTS_CONVERT;
4431 		lkb->lkb_grmode = DLM_LOCK_IV;
4432 		rsb_set_flag(r, RSB_RECOVER_CONVERT);
4433 	}
4434 
4435 	return 0;
4436 }
4437 
4438 /* This lkb may have been recovered in a previous aborted recovery so we need
4439    to check if the rsb already has an lkb with the given remote nodeid/lkid.
4440    If so we just send back a standard reply.  If not, we create a new lkb with
4441    the given values and send back our lkid.  We send back our lkid by sending
4442    back the rcom_lock struct we got but with the remid field filled in. */
4443 
4444 /* needs at least dlm_rcom + rcom_lock */
dlm_recover_master_copy(struct dlm_ls * ls,struct dlm_rcom * rc)4445 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4446 {
4447 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4448 	struct dlm_rsb *r;
4449 	struct dlm_lkb *lkb;
4450 	int error;
4451 
4452 	if (rl->rl_parent_lkid) {
4453 		error = -EOPNOTSUPP;
4454 		goto out;
4455 	}
4456 
4457 	error = find_rsb(ls, rl->rl_name, le16_to_cpu(rl->rl_namelen),
4458 			 R_MASTER, &r);
4459 	if (error)
4460 		goto out;
4461 
4462 	lock_rsb(r);
4463 
4464 	lkb = search_remid(r, rc->rc_header.h_nodeid, le32_to_cpu(rl->rl_lkid));
4465 	if (lkb) {
4466 		error = -EEXIST;
4467 		goto out_remid;
4468 	}
4469 
4470 	error = create_lkb(ls, &lkb);
4471 	if (error)
4472 		goto out_unlock;
4473 
4474 	error = receive_rcom_lock_args(ls, lkb, r, rc);
4475 	if (error) {
4476 		__put_lkb(ls, lkb);
4477 		goto out_unlock;
4478 	}
4479 
4480 	attach_lkb(r, lkb);
4481 	add_lkb(r, lkb, rl->rl_status);
4482 	error = 0;
4483 
4484  out_remid:
4485 	/* this is the new value returned to the lock holder for
4486 	   saving in its process-copy lkb */
4487 	rl->rl_remid = cpu_to_le32(lkb->lkb_id);
4488 
4489  out_unlock:
4490 	unlock_rsb(r);
4491 	put_rsb(r);
4492  out:
4493 	if (error)
4494 		log_debug(ls, "recover_master_copy %d %x", error,
4495 			  le32_to_cpu(rl->rl_lkid));
4496 	rl->rl_result = cpu_to_le32(error);
4497 	return error;
4498 }
4499 
4500 /* needs at least dlm_rcom + rcom_lock */
dlm_recover_process_copy(struct dlm_ls * ls,struct dlm_rcom * rc)4501 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4502 {
4503 	struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4504 	struct dlm_rsb *r;
4505 	struct dlm_lkb *lkb;
4506 	int error;
4507 
4508 	error = find_lkb(ls, le32_to_cpu(rl->rl_lkid), &lkb);
4509 	if (error) {
4510 		log_error(ls, "recover_process_copy no lkid %x",
4511 				le32_to_cpu(rl->rl_lkid));
4512 		return error;
4513 	}
4514 
4515 	DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
4516 
4517 	error = le32_to_cpu(rl->rl_result);
4518 
4519 	r = lkb->lkb_resource;
4520 	hold_rsb(r);
4521 	lock_rsb(r);
4522 
4523 	switch (error) {
4524 	case -EBADR:
4525 		/* There's a chance the new master received our lock before
4526 		   dlm_recover_master_reply(), this wouldn't happen if we did
4527 		   a barrier between recover_masters and recover_locks. */
4528 		log_debug(ls, "master copy not ready %x r %lx %s", lkb->lkb_id,
4529 			  (unsigned long)r, r->res_name);
4530 		dlm_send_rcom_lock(r, lkb);
4531 		goto out;
4532 	case -EEXIST:
4533 		log_debug(ls, "master copy exists %x", lkb->lkb_id);
4534 		/* fall through */
4535 	case 0:
4536 		lkb->lkb_remid = le32_to_cpu(rl->rl_remid);
4537 		break;
4538 	default:
4539 		log_error(ls, "dlm_recover_process_copy unknown error %d %x",
4540 			  error, lkb->lkb_id);
4541 	}
4542 
4543 	/* an ack for dlm_recover_locks() which waits for replies from
4544 	   all the locks it sends to new masters */
4545 	dlm_recovered_lock(r);
4546  out:
4547 	unlock_rsb(r);
4548 	put_rsb(r);
4549 	dlm_put_lkb(lkb);
4550 
4551 	return 0;
4552 }
4553 
dlm_user_request(struct dlm_ls * ls,struct dlm_user_args * ua,int mode,uint32_t flags,void * name,unsigned int namelen,unsigned long timeout_cs)4554 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
4555 		     int mode, uint32_t flags, void *name, unsigned int namelen,
4556 		     unsigned long timeout_cs)
4557 {
4558 	struct dlm_lkb *lkb;
4559 	struct dlm_args args;
4560 	int error;
4561 
4562 	dlm_lock_recovery(ls);
4563 
4564 	error = create_lkb(ls, &lkb);
4565 	if (error) {
4566 		kfree(ua);
4567 		goto out;
4568 	}
4569 
4570 	if (flags & DLM_LKF_VALBLK) {
4571 		ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_NOFS);
4572 		if (!ua->lksb.sb_lvbptr) {
4573 			kfree(ua);
4574 			__put_lkb(ls, lkb);
4575 			error = -ENOMEM;
4576 			goto out;
4577 		}
4578 	}
4579 
4580 	/* After ua is attached to lkb it will be freed by dlm_free_lkb().
4581 	   When DLM_IFL_USER is set, the dlm knows that this is a userspace
4582 	   lock and that lkb_astparam is the dlm_user_args structure. */
4583 
4584 	error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
4585 			      fake_astfn, ua, fake_bastfn, &args);
4586 	lkb->lkb_flags |= DLM_IFL_USER;
4587 
4588 	if (error) {
4589 		__put_lkb(ls, lkb);
4590 		goto out;
4591 	}
4592 
4593 	error = request_lock(ls, lkb, name, namelen, &args);
4594 
4595 	switch (error) {
4596 	case 0:
4597 		break;
4598 	case -EINPROGRESS:
4599 		error = 0;
4600 		break;
4601 	case -EAGAIN:
4602 		error = 0;
4603 		/* fall through */
4604 	default:
4605 		__put_lkb(ls, lkb);
4606 		goto out;
4607 	}
4608 
4609 	/* add this new lkb to the per-process list of locks */
4610 	spin_lock(&ua->proc->locks_spin);
4611 	hold_lkb(lkb);
4612 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
4613 	spin_unlock(&ua->proc->locks_spin);
4614  out:
4615 	dlm_unlock_recovery(ls);
4616 	return error;
4617 }
4618 
dlm_user_convert(struct dlm_ls * ls,struct dlm_user_args * ua_tmp,int mode,uint32_t flags,uint32_t lkid,char * lvb_in,unsigned long timeout_cs)4619 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4620 		     int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
4621 		     unsigned long timeout_cs)
4622 {
4623 	struct dlm_lkb *lkb;
4624 	struct dlm_args args;
4625 	struct dlm_user_args *ua;
4626 	int error;
4627 
4628 	dlm_lock_recovery(ls);
4629 
4630 	error = find_lkb(ls, lkid, &lkb);
4631 	if (error)
4632 		goto out;
4633 
4634 	/* user can change the params on its lock when it converts it, or
4635 	   add an lvb that didn't exist before */
4636 
4637 	ua = lkb->lkb_ua;
4638 
4639 	if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
4640 		ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_NOFS);
4641 		if (!ua->lksb.sb_lvbptr) {
4642 			error = -ENOMEM;
4643 			goto out_put;
4644 		}
4645 	}
4646 	if (lvb_in && ua->lksb.sb_lvbptr)
4647 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4648 
4649 	ua->xid = ua_tmp->xid;
4650 	ua->castparam = ua_tmp->castparam;
4651 	ua->castaddr = ua_tmp->castaddr;
4652 	ua->bastparam = ua_tmp->bastparam;
4653 	ua->bastaddr = ua_tmp->bastaddr;
4654 	ua->user_lksb = ua_tmp->user_lksb;
4655 
4656 	error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
4657 			      fake_astfn, ua, fake_bastfn, &args);
4658 	if (error)
4659 		goto out_put;
4660 
4661 	error = convert_lock(ls, lkb, &args);
4662 
4663 	if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
4664 		error = 0;
4665  out_put:
4666 	dlm_put_lkb(lkb);
4667  out:
4668 	dlm_unlock_recovery(ls);
4669 	kfree(ua_tmp);
4670 	return error;
4671 }
4672 
dlm_user_unlock(struct dlm_ls * ls,struct dlm_user_args * ua_tmp,uint32_t flags,uint32_t lkid,char * lvb_in)4673 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4674 		    uint32_t flags, uint32_t lkid, char *lvb_in)
4675 {
4676 	struct dlm_lkb *lkb;
4677 	struct dlm_args args;
4678 	struct dlm_user_args *ua;
4679 	int error;
4680 
4681 	dlm_lock_recovery(ls);
4682 
4683 	error = find_lkb(ls, lkid, &lkb);
4684 	if (error)
4685 		goto out;
4686 
4687 	ua = lkb->lkb_ua;
4688 
4689 	if (lvb_in && ua->lksb.sb_lvbptr)
4690 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4691 	if (ua_tmp->castparam)
4692 		ua->castparam = ua_tmp->castparam;
4693 	ua->user_lksb = ua_tmp->user_lksb;
4694 
4695 	error = set_unlock_args(flags, ua, &args);
4696 	if (error)
4697 		goto out_put;
4698 
4699 	error = unlock_lock(ls, lkb, &args);
4700 
4701 	if (error == -DLM_EUNLOCK)
4702 		error = 0;
4703 	/* from validate_unlock_args() */
4704 	if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
4705 		error = 0;
4706 	if (error)
4707 		goto out_put;
4708 
4709 	spin_lock(&ua->proc->locks_spin);
4710 	/* dlm_user_add_ast() may have already taken lkb off the proc list */
4711 	if (!list_empty(&lkb->lkb_ownqueue))
4712 		list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
4713 	spin_unlock(&ua->proc->locks_spin);
4714  out_put:
4715 	dlm_put_lkb(lkb);
4716  out:
4717 	dlm_unlock_recovery(ls);
4718 	kfree(ua_tmp);
4719 	return error;
4720 }
4721 
dlm_user_cancel(struct dlm_ls * ls,struct dlm_user_args * ua_tmp,uint32_t flags,uint32_t lkid)4722 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4723 		    uint32_t flags, uint32_t lkid)
4724 {
4725 	struct dlm_lkb *lkb;
4726 	struct dlm_args args;
4727 	struct dlm_user_args *ua;
4728 	int error;
4729 
4730 	dlm_lock_recovery(ls);
4731 
4732 	error = find_lkb(ls, lkid, &lkb);
4733 	if (error)
4734 		goto out;
4735 
4736 	ua = lkb->lkb_ua;
4737 	if (ua_tmp->castparam)
4738 		ua->castparam = ua_tmp->castparam;
4739 	ua->user_lksb = ua_tmp->user_lksb;
4740 
4741 	error = set_unlock_args(flags, ua, &args);
4742 	if (error)
4743 		goto out_put;
4744 
4745 	error = cancel_lock(ls, lkb, &args);
4746 
4747 	if (error == -DLM_ECANCEL)
4748 		error = 0;
4749 	/* from validate_unlock_args() */
4750 	if (error == -EBUSY)
4751 		error = 0;
4752  out_put:
4753 	dlm_put_lkb(lkb);
4754  out:
4755 	dlm_unlock_recovery(ls);
4756 	kfree(ua_tmp);
4757 	return error;
4758 }
4759 
dlm_user_deadlock(struct dlm_ls * ls,uint32_t flags,uint32_t lkid)4760 int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
4761 {
4762 	struct dlm_lkb *lkb;
4763 	struct dlm_args args;
4764 	struct dlm_user_args *ua;
4765 	struct dlm_rsb *r;
4766 	int error;
4767 
4768 	dlm_lock_recovery(ls);
4769 
4770 	error = find_lkb(ls, lkid, &lkb);
4771 	if (error)
4772 		goto out;
4773 
4774 	ua = lkb->lkb_ua;
4775 
4776 	error = set_unlock_args(flags, ua, &args);
4777 	if (error)
4778 		goto out_put;
4779 
4780 	/* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4781 
4782 	r = lkb->lkb_resource;
4783 	hold_rsb(r);
4784 	lock_rsb(r);
4785 
4786 	error = validate_unlock_args(lkb, &args);
4787 	if (error)
4788 		goto out_r;
4789 	lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
4790 
4791 	error = _cancel_lock(r, lkb);
4792  out_r:
4793 	unlock_rsb(r);
4794 	put_rsb(r);
4795 
4796 	if (error == -DLM_ECANCEL)
4797 		error = 0;
4798 	/* from validate_unlock_args() */
4799 	if (error == -EBUSY)
4800 		error = 0;
4801  out_put:
4802 	dlm_put_lkb(lkb);
4803  out:
4804 	dlm_unlock_recovery(ls);
4805 	return error;
4806 }
4807 
4808 /* lkb's that are removed from the waiters list by revert are just left on the
4809    orphans list with the granted orphan locks, to be freed by purge */
4810 
orphan_proc_lock(struct dlm_ls * ls,struct dlm_lkb * lkb)4811 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4812 {
4813 	struct dlm_args args;
4814 	int error;
4815 
4816 	hold_lkb(lkb);
4817 	mutex_lock(&ls->ls_orphans_mutex);
4818 	list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
4819 	mutex_unlock(&ls->ls_orphans_mutex);
4820 
4821 	set_unlock_args(0, lkb->lkb_ua, &args);
4822 
4823 	error = cancel_lock(ls, lkb, &args);
4824 	if (error == -DLM_ECANCEL)
4825 		error = 0;
4826 	return error;
4827 }
4828 
4829 /* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4830    Regardless of what rsb queue the lock is on, it's removed and freed. */
4831 
unlock_proc_lock(struct dlm_ls * ls,struct dlm_lkb * lkb)4832 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4833 {
4834 	struct dlm_args args;
4835 	int error;
4836 
4837 	set_unlock_args(DLM_LKF_FORCEUNLOCK, lkb->lkb_ua, &args);
4838 
4839 	error = unlock_lock(ls, lkb, &args);
4840 	if (error == -DLM_EUNLOCK)
4841 		error = 0;
4842 	return error;
4843 }
4844 
4845 /* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4846    (which does lock_rsb) due to deadlock with receiving a message that does
4847    lock_rsb followed by dlm_user_add_ast() */
4848 
del_proc_lock(struct dlm_ls * ls,struct dlm_user_proc * proc)4849 static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
4850 				     struct dlm_user_proc *proc)
4851 {
4852 	struct dlm_lkb *lkb = NULL;
4853 
4854 	mutex_lock(&ls->ls_clear_proc_locks);
4855 	if (list_empty(&proc->locks))
4856 		goto out;
4857 
4858 	lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
4859 	list_del_init(&lkb->lkb_ownqueue);
4860 
4861 	if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4862 		lkb->lkb_flags |= DLM_IFL_ORPHAN;
4863 	else
4864 		lkb->lkb_flags |= DLM_IFL_DEAD;
4865  out:
4866 	mutex_unlock(&ls->ls_clear_proc_locks);
4867 	return lkb;
4868 }
4869 
4870 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4871    1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4872    which we clear here. */
4873 
4874 /* proc CLOSING flag is set so no more device_reads should look at proc->asts
4875    list, and no more device_writes should add lkb's to proc->locks list; so we
4876    shouldn't need to take asts_spin or locks_spin here.  this assumes that
4877    device reads/writes/closes are serialized -- FIXME: we may need to serialize
4878    them ourself. */
4879 
dlm_clear_proc_locks(struct dlm_ls * ls,struct dlm_user_proc * proc)4880 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4881 {
4882 	struct dlm_lkb *lkb, *safe;
4883 
4884 	dlm_lock_recovery(ls);
4885 
4886 	while (1) {
4887 		lkb = del_proc_lock(ls, proc);
4888 		if (!lkb)
4889 			break;
4890 		del_timeout(lkb);
4891 		if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4892 			orphan_proc_lock(ls, lkb);
4893 		else
4894 			unlock_proc_lock(ls, lkb);
4895 
4896 		/* this removes the reference for the proc->locks list
4897 		   added by dlm_user_request, it may result in the lkb
4898 		   being freed */
4899 
4900 		dlm_put_lkb(lkb);
4901 	}
4902 
4903 	mutex_lock(&ls->ls_clear_proc_locks);
4904 
4905 	/* in-progress unlocks */
4906 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4907 		list_del_init(&lkb->lkb_ownqueue);
4908 		lkb->lkb_flags |= DLM_IFL_DEAD;
4909 		dlm_put_lkb(lkb);
4910 	}
4911 
4912 	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4913 		memset(&lkb->lkb_callbacks, 0,
4914 		       sizeof(struct dlm_callback) * DLM_CALLBACKS_SIZE);
4915 		list_del_init(&lkb->lkb_astqueue);
4916 		dlm_put_lkb(lkb);
4917 	}
4918 
4919 	mutex_unlock(&ls->ls_clear_proc_locks);
4920 	dlm_unlock_recovery(ls);
4921 }
4922 
purge_proc_locks(struct dlm_ls * ls,struct dlm_user_proc * proc)4923 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4924 {
4925 	struct dlm_lkb *lkb, *safe;
4926 
4927 	while (1) {
4928 		lkb = NULL;
4929 		spin_lock(&proc->locks_spin);
4930 		if (!list_empty(&proc->locks)) {
4931 			lkb = list_entry(proc->locks.next, struct dlm_lkb,
4932 					 lkb_ownqueue);
4933 			list_del_init(&lkb->lkb_ownqueue);
4934 		}
4935 		spin_unlock(&proc->locks_spin);
4936 
4937 		if (!lkb)
4938 			break;
4939 
4940 		lkb->lkb_flags |= DLM_IFL_DEAD;
4941 		unlock_proc_lock(ls, lkb);
4942 		dlm_put_lkb(lkb); /* ref from proc->locks list */
4943 	}
4944 
4945 	spin_lock(&proc->locks_spin);
4946 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4947 		list_del_init(&lkb->lkb_ownqueue);
4948 		lkb->lkb_flags |= DLM_IFL_DEAD;
4949 		dlm_put_lkb(lkb);
4950 	}
4951 	spin_unlock(&proc->locks_spin);
4952 
4953 	spin_lock(&proc->asts_spin);
4954 	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4955 		memset(&lkb->lkb_callbacks, 0,
4956 		       sizeof(struct dlm_callback) * DLM_CALLBACKS_SIZE);
4957 		list_del_init(&lkb->lkb_astqueue);
4958 		dlm_put_lkb(lkb);
4959 	}
4960 	spin_unlock(&proc->asts_spin);
4961 }
4962 
4963 /* pid of 0 means purge all orphans */
4964 
do_purge(struct dlm_ls * ls,int nodeid,int pid)4965 static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
4966 {
4967 	struct dlm_lkb *lkb, *safe;
4968 
4969 	mutex_lock(&ls->ls_orphans_mutex);
4970 	list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
4971 		if (pid && lkb->lkb_ownpid != pid)
4972 			continue;
4973 		unlock_proc_lock(ls, lkb);
4974 		list_del_init(&lkb->lkb_ownqueue);
4975 		dlm_put_lkb(lkb);
4976 	}
4977 	mutex_unlock(&ls->ls_orphans_mutex);
4978 }
4979 
send_purge(struct dlm_ls * ls,int nodeid,int pid)4980 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
4981 {
4982 	struct dlm_message *ms;
4983 	struct dlm_mhandle *mh;
4984 	int error;
4985 
4986 	error = _create_message(ls, sizeof(struct dlm_message), nodeid,
4987 				DLM_MSG_PURGE, &ms, &mh);
4988 	if (error)
4989 		return error;
4990 	ms->m_nodeid = nodeid;
4991 	ms->m_pid = pid;
4992 
4993 	return send_message(mh, ms);
4994 }
4995 
dlm_user_purge(struct dlm_ls * ls,struct dlm_user_proc * proc,int nodeid,int pid)4996 int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
4997 		   int nodeid, int pid)
4998 {
4999 	int error = 0;
5000 
5001 	if (nodeid != dlm_our_nodeid()) {
5002 		error = send_purge(ls, nodeid, pid);
5003 	} else {
5004 		dlm_lock_recovery(ls);
5005 		if (pid == current->pid)
5006 			purge_proc_locks(ls, proc);
5007 		else
5008 			do_purge(ls, nodeid, pid);
5009 		dlm_unlock_recovery(ls);
5010 	}
5011 	return error;
5012 }
5013 
5014