]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/dlm/lock.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
dlm: change error message to debug
[mirror_ubuntu-artful-kernel.git] / fs / dlm / lock.c
CommitLineData
e7fd4179
DT		 1/******************************************************************************
2*******************************************************************************
3**
46b43eed 4** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
e7fd4179
DT		 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*/
597d0cae 58#include <linux/types.h>
e7fd4179 59#include "dlm_internal.h"
597d0cae 60#include <linux/dlm_device.h>
e7fd4179
DT		 61#include "memory.h"
62#include "lowcomms.h"
63#include "requestqueue.h"
64#include "util.h"
65#include "dir.h"
66#include "member.h"
67#include "lockspace.h"
68#include "ast.h"
69#include "lock.h"
70#include "rcom.h"
71#include "recover.h"
72#include "lvb_table.h"
597d0cae 73#include "user.h"
e7fd4179
DT		 74#include "config.h"
75
76static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb);
77static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb);
78static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb);
79static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb);
80static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb);
81static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode);
82static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb);
83static int send_remove(struct dlm_rsb *r);
84static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
3ae1acf9 85static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
e7fd4179
DT		 86static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
87 struct dlm_message *ms);
88static int receive_extralen(struct dlm_message *ms);
8499137d 89static void do_purge(struct dlm_ls *ls, int nodeid, int pid);
3ae1acf9 90static void del_timeout(struct dlm_lkb *lkb);
e7fd4179
DT		 91
92/*
93 * Lock compatibilty matrix - thanks Steve
94 * UN = Unlocked state. Not really a state, used as a flag
95 * PD = Padding. Used to make the matrix a nice power of two in size
96 * Other states are the same as the VMS DLM.
97 * Usage: matrix[grmode+1][rqmode+1] (although m[rq+1][gr+1] is the same)
98 */
99
100static const int __dlm_compat_matrix[8][8] = {
101 /* UN NL CR CW PR PW EX PD */
102 {1, 1, 1, 1, 1, 1, 1, 0}, /* UN */
103 {1, 1, 1, 1, 1, 1, 1, 0}, /* NL */
104 {1, 1, 1, 1, 1, 1, 0, 0}, /* CR */
105 {1, 1, 1, 1, 0, 0, 0, 0}, /* CW */
106 {1, 1, 1, 0, 1, 0, 0, 0}, /* PR */
107 {1, 1, 1, 0, 0, 0, 0, 0}, /* PW */
108 {1, 1, 0, 0, 0, 0, 0, 0}, /* EX */
109 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
110};
111
112/*
113 * This defines the direction of transfer of LVB data.
114 * Granted mode is the row; requested mode is the column.
115 * Usage: matrix[grmode+1][rqmode+1]
116 * 1 = LVB is returned to the caller
117 * 0 = LVB is written to the resource
118 * -1 = nothing happens to the LVB
119 */
120
121const int dlm_lvb_operations[8][8] = {
122 /* UN NL CR CW PR PW EX PD*/
123 { -1, 1, 1, 1, 1, 1, 1, -1 }, /* UN */
124 { -1, 1, 1, 1, 1, 1, 1, 0 }, /* NL */
125 { -1, -1, 1, 1, 1, 1, 1, 0 }, /* CR */
126 { -1, -1, -1, 1, 1, 1, 1, 0 }, /* CW */
127 { -1, -1, -1, -1, 1, 1, 1, 0 }, /* PR */
128 { -1, 0, 0, 0, 0, 0, 1, 0 }, /* PW */
129 { -1, 0, 0, 0, 0, 0, 0, 0 }, /* EX */
130 { -1, 0, 0, 0, 0, 0, 0, 0 } /* PD */
131};
e7fd4179
DT		 132
133#define modes_compat(gr, rq) \
134 __dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1]
135
136int dlm_modes_compat(int mode1, int mode2)
137{
138 return __dlm_compat_matrix[mode1 + 1][mode2 + 1];
139}
140
141/*
142 * Compatibility matrix for conversions with QUECVT set.
143 * Granted mode is the row; requested mode is the column.
144 * Usage: matrix[grmode+1][rqmode+1]
145 */
146
147static const int __quecvt_compat_matrix[8][8] = {
148 /* UN NL CR CW PR PW EX PD */
149 {0, 0, 0, 0, 0, 0, 0, 0}, /* UN */
150 {0, 0, 1, 1, 1, 1, 1, 0}, /* NL */
151 {0, 0, 0, 1, 1, 1, 1, 0}, /* CR */
152 {0, 0, 0, 0, 1, 1, 1, 0}, /* CW */
153 {0, 0, 0, 1, 0, 1, 1, 0}, /* PR */
154 {0, 0, 0, 0, 0, 0, 1, 0}, /* PW */
155 {0, 0, 0, 0, 0, 0, 0, 0}, /* EX */
156 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */
157};
158
597d0cae 159void dlm_print_lkb(struct dlm_lkb *lkb)
e7fd4179
DT		 160{
161 printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x\n"
162 " status %d rqmode %d grmode %d wait_type %d ast_type %d\n",
163 lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags,
164 lkb->lkb_flags, lkb->lkb_status, lkb->lkb_rqmode,
165 lkb->lkb_grmode, lkb->lkb_wait_type, lkb->lkb_ast_type);
166}
167
168void dlm_print_rsb(struct dlm_rsb *r)
169{
170 printk(KERN_ERR "rsb: nodeid %d flags %lx first %x rlc %d name %s\n",
171 r->res_nodeid, r->res_flags, r->res_first_lkid,
172 r->res_recover_locks_count, r->res_name);
173}
174
a345da3e
DT		 175void dlm_dump_rsb(struct dlm_rsb *r)
176{
177 struct dlm_lkb *lkb;
178
179 dlm_print_rsb(r);
180
181 printk(KERN_ERR "rsb: root_list empty %d recover_list empty %d\n",
182 list_empty(&r->res_root_list), list_empty(&r->res_recover_list));
183 printk(KERN_ERR "rsb lookup list\n");
184 list_for_each_entry(lkb, &r->res_lookup, lkb_rsb_lookup)
185 dlm_print_lkb(lkb);
186 printk(KERN_ERR "rsb grant queue:\n");
187 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue)
188 dlm_print_lkb(lkb);
189 printk(KERN_ERR "rsb convert queue:\n");
190 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue)
191 dlm_print_lkb(lkb);
192 printk(KERN_ERR "rsb wait queue:\n");
193 list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue)
194 dlm_print_lkb(lkb);
195}
196
e7fd4179
DT		 197/* Threads cannot use the lockspace while it's being recovered */
198
85e86edf 199static inline void dlm_lock_recovery(struct dlm_ls *ls)
e7fd4179
DT		 200{
201 down_read(&ls->ls_in_recovery);
202}
203
85e86edf 204void dlm_unlock_recovery(struct dlm_ls *ls)
e7fd4179
DT		 205{
206 up_read(&ls->ls_in_recovery);
207}
208
85e86edf 209int dlm_lock_recovery_try(struct dlm_ls *ls)
e7fd4179
DT		 210{
211 return down_read_trylock(&ls->ls_in_recovery);
212}
213
214static inline int can_be_queued(struct dlm_lkb *lkb)
215{
216 return !(lkb->lkb_exflags & DLM_LKF_NOQUEUE);
217}
218
219static inline int force_blocking_asts(struct dlm_lkb *lkb)
220{
221 return (lkb->lkb_exflags & DLM_LKF_NOQUEUEBAST);
222}
223
224static inline int is_demoted(struct dlm_lkb *lkb)
225{
226 return (lkb->lkb_sbflags & DLM_SBF_DEMOTED);
227}
228
7d3c1feb
DT		 229static inline int is_altmode(struct dlm_lkb *lkb)
230{
231 return (lkb->lkb_sbflags & DLM_SBF_ALTMODE);
232}
233
234static inline int is_granted(struct dlm_lkb *lkb)
235{
236 return (lkb->lkb_status == DLM_LKSTS_GRANTED);
237}
238
e7fd4179
DT		 239static inline int is_remote(struct dlm_rsb *r)
240{
241 DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r););
242 return !!r->res_nodeid;
243}
244
245static inline int is_process_copy(struct dlm_lkb *lkb)
246{
247 return (lkb->lkb_nodeid && !(lkb->lkb_flags & DLM_IFL_MSTCPY));
248}
249
250static inline int is_master_copy(struct dlm_lkb *lkb)
251{
252 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
253 DLM_ASSERT(lkb->lkb_nodeid, dlm_print_lkb(lkb););
90135925 254 return (lkb->lkb_flags & DLM_IFL_MSTCPY) ? 1 : 0;
e7fd4179
DT		 255}
256
257static inline int middle_conversion(struct dlm_lkb *lkb)
258{
259 if ((lkb->lkb_grmode==DLM_LOCK_PR && lkb->lkb_rqmode==DLM_LOCK_CW) ||
260 (lkb->lkb_rqmode==DLM_LOCK_PR && lkb->lkb_grmode==DLM_LOCK_CW))
90135925
DT		 261 return 1;
262 return 0;
e7fd4179
DT		 263}
264
265static inline int down_conversion(struct dlm_lkb *lkb)
266{
267 return (!middle_conversion(lkb) && lkb->lkb_rqmode < lkb->lkb_grmode);
268}
269
ef0c2bb0
DT		 270static inline int is_overlap_unlock(struct dlm_lkb *lkb)
271{
272 return lkb->lkb_flags & DLM_IFL_OVERLAP_UNLOCK;
273}
274
275static inline int is_overlap_cancel(struct dlm_lkb *lkb)
276{
277 return lkb->lkb_flags & DLM_IFL_OVERLAP_CANCEL;
278}
279
280static inline int is_overlap(struct dlm_lkb *lkb)
281{
282 return (lkb->lkb_flags & (DLM_IFL_OVERLAP_UNLOCK |
283 DLM_IFL_OVERLAP_CANCEL));
284}
285
e7fd4179
DT		 286static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
287{
288 if (is_master_copy(lkb))
289 return;
290
3ae1acf9
DT		 291 del_timeout(lkb);
292
e7fd4179
DT		 293 DLM_ASSERT(lkb->lkb_lksb, dlm_print_lkb(lkb););
294
3ae1acf9
DT		 295 /* if the operation was a cancel, then return -DLM_ECANCEL, if a
296 timeout caused the cancel then return -ETIMEDOUT */
297 if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_TIMEOUT_CANCEL)) {
298 lkb->lkb_flags &= ~DLM_IFL_TIMEOUT_CANCEL;
299 rv = -ETIMEDOUT;
300 }
301
8b4021fa
DT		 302 if (rv == -DLM_ECANCEL && (lkb->lkb_flags & DLM_IFL_DEADLOCK_CANCEL)) {
303 lkb->lkb_flags &= ~DLM_IFL_DEADLOCK_CANCEL;
304 rv = -EDEADLK;
305 }
306
e7fd4179
DT		 307 lkb->lkb_lksb->sb_status = rv;
308 lkb->lkb_lksb->sb_flags = lkb->lkb_sbflags;
309
310 dlm_add_ast(lkb, AST_COMP);
311}
312
ef0c2bb0
DT		 313static inline void queue_cast_overlap(struct dlm_rsb *r, struct dlm_lkb *lkb)
314{
315 queue_cast(r, lkb,
316 is_overlap_unlock(lkb) ? -DLM_EUNLOCK : -DLM_ECANCEL);
317}
318
e7fd4179
DT		 319static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
320{
321 if (is_master_copy(lkb))
322 send_bast(r, lkb, rqmode);
323 else {
324 lkb->lkb_bastmode = rqmode;
325 dlm_add_ast(lkb, AST_BAST);
326 }
327}
328
329/*
330 * Basic operations on rsb's and lkb's
331 */
332
333static struct dlm_rsb *create_rsb(struct dlm_ls *ls, char *name, int len)
334{
335 struct dlm_rsb *r;
336
52bda2b5 337 r = dlm_allocate_rsb(ls, len);
e7fd4179
DT		 338 if (!r)
339 return NULL;
340
341 r->res_ls = ls;
342 r->res_length = len;
343 memcpy(r->res_name, name, len);
90135925 344 mutex_init(&r->res_mutex);
e7fd4179
DT		 345
346 INIT_LIST_HEAD(&r->res_lookup);
347 INIT_LIST_HEAD(&r->res_grantqueue);
348 INIT_LIST_HEAD(&r->res_convertqueue);
349 INIT_LIST_HEAD(&r->res_waitqueue);
350 INIT_LIST_HEAD(&r->res_root_list);
351 INIT_LIST_HEAD(&r->res_recover_list);
352
353 return r;
354}
355
356static int search_rsb_list(struct list_head *head, char *name, int len,
357 unsigned int flags, struct dlm_rsb **r_ret)
358{
359 struct dlm_rsb *r;
360 int error = 0;
361
362 list_for_each_entry(r, head, res_hashchain) {
363 if (len == r->res_length && !memcmp(name, r->res_name, len))
364 goto found;
365 }
597d0cae 366 return -EBADR;
e7fd4179
DT		 367
368 found:
369 if (r->res_nodeid && (flags & R_MASTER))
370 error = -ENOTBLK;
371 *r_ret = r;
372 return error;
373}
374
375static int _search_rsb(struct dlm_ls *ls, char *name, int len, int b,
376 unsigned int flags, struct dlm_rsb **r_ret)
377{
378 struct dlm_rsb *r;
379 int error;
380
381 error = search_rsb_list(&ls->ls_rsbtbl[b].list, name, len, flags, &r);
382 if (!error) {
383 kref_get(&r->res_ref);
384 goto out;
385 }
386 error = search_rsb_list(&ls->ls_rsbtbl[b].toss, name, len, flags, &r);
387 if (error)
388 goto out;
389
390 list_move(&r->res_hashchain, &ls->ls_rsbtbl[b].list);
391
392 if (dlm_no_directory(ls))
393 goto out;
394
395 if (r->res_nodeid == -1) {
396 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
397 r->res_first_lkid = 0;
398 } else if (r->res_nodeid > 0) {
399 rsb_set_flag(r, RSB_MASTER_UNCERTAIN);
400 r->res_first_lkid = 0;
401 } else {
402 DLM_ASSERT(r->res_nodeid == 0, dlm_print_rsb(r););
403 DLM_ASSERT(!rsb_flag(r, RSB_MASTER_UNCERTAIN),);
404 }
405 out:
406 *r_ret = r;
407 return error;
408}
409
410static int search_rsb(struct dlm_ls *ls, char *name, int len, int b,
411 unsigned int flags, struct dlm_rsb **r_ret)
412{
413 int error;
414 write_lock(&ls->ls_rsbtbl[b].lock);
415 error = _search_rsb(ls, name, len, b, flags, r_ret);
416 write_unlock(&ls->ls_rsbtbl[b].lock);
417 return error;
418}
419
420/*
421 * Find rsb in rsbtbl and potentially create/add one
422 *
423 * Delaying the release of rsb's has a similar benefit to applications keeping
424 * NL locks on an rsb, but without the guarantee that the cached master value
425 * will still be valid when the rsb is reused. Apps aren't always smart enough
426 * to keep NL locks on an rsb that they may lock again shortly; this can lead
427 * to excessive master lookups and removals if we don't delay the release.
428 *
429 * Searching for an rsb means looking through both the normal list and toss
430 * list. When found on the toss list the rsb is moved to the normal list with
431 * ref count of 1; when found on normal list the ref count is incremented.
432 */
433
434static int find_rsb(struct dlm_ls *ls, char *name, int namelen,
435 unsigned int flags, struct dlm_rsb **r_ret)
436{
437 struct dlm_rsb *r, *tmp;
438 uint32_t hash, bucket;
439 int error = 0;
440
441 if (dlm_no_directory(ls))
442 flags |= R_CREATE;
443
444 hash = jhash(name, namelen, 0);
445 bucket = hash & (ls->ls_rsbtbl_size - 1);
446
447 error = search_rsb(ls, name, namelen, bucket, flags, &r);
448 if (!error)
449 goto out;
450
597d0cae 451 if (error == -EBADR && !(flags & R_CREATE))
e7fd4179
DT		 452 goto out;
453
454 /* the rsb was found but wasn't a master copy */
455 if (error == -ENOTBLK)
456 goto out;
457
458 error = -ENOMEM;
459 r = create_rsb(ls, name, namelen);
460 if (!r)
461 goto out;
462
463 r->res_hash = hash;
464 r->res_bucket = bucket;
465 r->res_nodeid = -1;
466 kref_init(&r->res_ref);
467
468 /* With no directory, the master can be set immediately */
469 if (dlm_no_directory(ls)) {
470 int nodeid = dlm_dir_nodeid(r);
471 if (nodeid == dlm_our_nodeid())
472 nodeid = 0;
473 r->res_nodeid = nodeid;
474 }
475
476 write_lock(&ls->ls_rsbtbl[bucket].lock);
477 error = _search_rsb(ls, name, namelen, bucket, 0, &tmp);
478 if (!error) {
479 write_unlock(&ls->ls_rsbtbl[bucket].lock);
52bda2b5 480 dlm_free_rsb(r);
e7fd4179
DT		 481 r = tmp;
482 goto out;
483 }
484 list_add(&r->res_hashchain, &ls->ls_rsbtbl[bucket].list);
485 write_unlock(&ls->ls_rsbtbl[bucket].lock);
486 error = 0;
487 out:
488 *r_ret = r;
489 return error;
490}
491
492int dlm_find_rsb(struct dlm_ls *ls, char *name, int namelen,
493 unsigned int flags, struct dlm_rsb **r_ret)
494{
495 return find_rsb(ls, name, namelen, flags, r_ret);
496}
497
498/* This is only called to add a reference when the code already holds
499 a valid reference to the rsb, so there's no need for locking. */
500
501static inline void hold_rsb(struct dlm_rsb *r)
502{
503 kref_get(&r->res_ref);
504}
505
506void dlm_hold_rsb(struct dlm_rsb *r)
507{
508 hold_rsb(r);
509}
510
511static void toss_rsb(struct kref *kref)
512{
513 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
514 struct dlm_ls *ls = r->res_ls;
515
516 DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
517 kref_init(&r->res_ref);
518 list_move(&r->res_hashchain, &ls->ls_rsbtbl[r->res_bucket].toss);
519 r->res_toss_time = jiffies;
520 if (r->res_lvbptr) {
52bda2b5 521 dlm_free_lvb(r->res_lvbptr);
e7fd4179
DT		 522 r->res_lvbptr = NULL;
523 }
524}
525
526/* When all references to the rsb are gone it's transfered to
527 the tossed list for later disposal. */
528
529static void put_rsb(struct dlm_rsb *r)
530{
531 struct dlm_ls *ls = r->res_ls;
532 uint32_t bucket = r->res_bucket;
533
534 write_lock(&ls->ls_rsbtbl[bucket].lock);
535 kref_put(&r->res_ref, toss_rsb);
536 write_unlock(&ls->ls_rsbtbl[bucket].lock);
537}
538
539void dlm_put_rsb(struct dlm_rsb *r)
540{
541 put_rsb(r);
542}
543
544/* See comment for unhold_lkb */
545
546static void unhold_rsb(struct dlm_rsb *r)
547{
548 int rv;
549 rv = kref_put(&r->res_ref, toss_rsb);
a345da3e 550 DLM_ASSERT(!rv, dlm_dump_rsb(r););
e7fd4179
DT		 551}
552
553static void kill_rsb(struct kref *kref)
554{
555 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
556
557 /* All work is done after the return from kref_put() so we
558 can release the write_lock before the remove and free. */
559
a345da3e
DT		 560 DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
561 DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
562 DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
563 DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
564 DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
565 DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
e7fd4179
DT		 566}
567
568/* Attaching/detaching lkb's from rsb's is for rsb reference counting.
569 The rsb must exist as long as any lkb's for it do. */
570
571static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
572{
573 hold_rsb(r);
574 lkb->lkb_resource = r;
575}
576
577static void detach_lkb(struct dlm_lkb *lkb)
578{
579 if (lkb->lkb_resource) {
580 put_rsb(lkb->lkb_resource);
581 lkb->lkb_resource = NULL;
582 }
583}
584
585static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
586{
587 struct dlm_lkb *lkb, *tmp;
588 uint32_t lkid = 0;
589 uint16_t bucket;
590
52bda2b5 591 lkb = dlm_allocate_lkb(ls);
e7fd4179
DT		 592 if (!lkb)
593 return -ENOMEM;
594
595 lkb->lkb_nodeid = -1;
596 lkb->lkb_grmode = DLM_LOCK_IV;
597 kref_init(&lkb->lkb_ref);
34e22bed 598 INIT_LIST_HEAD(&lkb->lkb_ownqueue);
ef0c2bb0 599 INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
3ae1acf9 600 INIT_LIST_HEAD(&lkb->lkb_time_list);
e7fd4179
DT		 601
602 get_random_bytes(&bucket, sizeof(bucket));
603 bucket &= (ls->ls_lkbtbl_size - 1);
604
605 write_lock(&ls->ls_lkbtbl[bucket].lock);
606
607 /* counter can roll over so we must verify lkid is not in use */
608
609 while (lkid == 0) {
ce03f12b 610 lkid = (bucket << 16) | ls->ls_lkbtbl[bucket].counter++;
e7fd4179
DT		 611
612 list_for_each_entry(tmp, &ls->ls_lkbtbl[bucket].list,
613 lkb_idtbl_list) {
614 if (tmp->lkb_id != lkid)
615 continue;
616 lkid = 0;
617 break;
618 }
619 }
620
621 lkb->lkb_id = lkid;
622 list_add(&lkb->lkb_idtbl_list, &ls->ls_lkbtbl[bucket].list);
623 write_unlock(&ls->ls_lkbtbl[bucket].lock);
624
625 *lkb_ret = lkb;
626 return 0;
627}
628
629static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid)
630{
e7fd4179 631 struct dlm_lkb *lkb;
ce03f12b 632 uint16_t bucket = (lkid >> 16);
e7fd4179
DT		 633
634 list_for_each_entry(lkb, &ls->ls_lkbtbl[bucket].list, lkb_idtbl_list) {
635 if (lkb->lkb_id == lkid)
636 return lkb;
637 }
638 return NULL;
639}
640
641static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
642{
643 struct dlm_lkb *lkb;
ce03f12b 644 uint16_t bucket = (lkid >> 16);
e7fd4179
DT		 645
646 if (bucket >= ls->ls_lkbtbl_size)
647 return -EBADSLT;
648
649 read_lock(&ls->ls_lkbtbl[bucket].lock);
650 lkb = __find_lkb(ls, lkid);
651 if (lkb)
652 kref_get(&lkb->lkb_ref);
653 read_unlock(&ls->ls_lkbtbl[bucket].lock);
654
655 *lkb_ret = lkb;
656 return lkb ? 0 : -ENOENT;
657}
658
659static void kill_lkb(struct kref *kref)
660{
661 struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref);
662
663 /* All work is done after the return from kref_put() so we
664 can release the write_lock before the detach_lkb */
665
666 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
667}
668
b3f58d8f
DT		 669/* __put_lkb() is used when an lkb may not have an rsb attached to
670 it so we need to provide the lockspace explicitly */
671
672static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
e7fd4179 673{
ce03f12b 674 uint16_t bucket = (lkb->lkb_id >> 16);
e7fd4179
DT		 675
676 write_lock(&ls->ls_lkbtbl[bucket].lock);
677 if (kref_put(&lkb->lkb_ref, kill_lkb)) {
678 list_del(&lkb->lkb_idtbl_list);
679 write_unlock(&ls->ls_lkbtbl[bucket].lock);
680
681 detach_lkb(lkb);
682
683 /* for local/process lkbs, lvbptr points to caller's lksb */
684 if (lkb->lkb_lvbptr && is_master_copy(lkb))
52bda2b5
DT		 685 dlm_free_lvb(lkb->lkb_lvbptr);
686 dlm_free_lkb(lkb);
e7fd4179
DT		 687 return 1;
688 } else {
689 write_unlock(&ls->ls_lkbtbl[bucket].lock);
690 return 0;
691 }
692}
693
694int dlm_put_lkb(struct dlm_lkb *lkb)
695{
b3f58d8f
DT		 696 struct dlm_ls *ls;
697
698 DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb););
699 DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb););
700
701 ls = lkb->lkb_resource->res_ls;
702 return __put_lkb(ls, lkb);
e7fd4179
DT		 703}
704
705/* This is only called to add a reference when the code already holds
706 a valid reference to the lkb, so there's no need for locking. */
707
708static inline void hold_lkb(struct dlm_lkb *lkb)
709{
710 kref_get(&lkb->lkb_ref);
711}
712
713/* This is called when we need to remove a reference and are certain
714 it's not the last ref. e.g. del_lkb is always called between a
715 find_lkb/put_lkb and is always the inverse of a previous add_lkb.
716 put_lkb would work fine, but would involve unnecessary locking */
717
718static inline void unhold_lkb(struct dlm_lkb *lkb)
719{
720 int rv;
721 rv = kref_put(&lkb->lkb_ref, kill_lkb);
722 DLM_ASSERT(!rv, dlm_print_lkb(lkb););
723}
724
725static void lkb_add_ordered(struct list_head *new, struct list_head *head,
726 int mode)
727{
728 struct dlm_lkb *lkb = NULL;
729
730 list_for_each_entry(lkb, head, lkb_statequeue)
731 if (lkb->lkb_rqmode < mode)
732 break;
733
734 if (!lkb)
735 list_add_tail(new, head);
736 else
737 __list_add(new, lkb->lkb_statequeue.prev, &lkb->lkb_statequeue);
738}
739
740/* add/remove lkb to rsb's grant/convert/wait queue */
741
742static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status)
743{
744 kref_get(&lkb->lkb_ref);
745
746 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb););
747
748 lkb->lkb_status = status;
749
750 switch (status) {
751 case DLM_LKSTS_WAITING:
752 if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
753 list_add(&lkb->lkb_statequeue, &r->res_waitqueue);
754 else
755 list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue);
756 break;
757 case DLM_LKSTS_GRANTED:
758 /* convention says granted locks kept in order of grmode */
759 lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue,
760 lkb->lkb_grmode);
761 break;
762 case DLM_LKSTS_CONVERT:
763 if (lkb->lkb_exflags & DLM_LKF_HEADQUE)
764 list_add(&lkb->lkb_statequeue, &r->res_convertqueue);
765 else
766 list_add_tail(&lkb->lkb_statequeue,
767 &r->res_convertqueue);
768 break;
769 default:
770 DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status););
771 }
772}
773
774static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb)
775{
776 lkb->lkb_status = 0;
777 list_del(&lkb->lkb_statequeue);
778 unhold_lkb(lkb);
779}
780
781static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
782{
783 hold_lkb(lkb);
784 del_lkb(r, lkb);
785 add_lkb(r, lkb, sts);
786 unhold_lkb(lkb);
787}
788
ef0c2bb0
DT		 789static int msg_reply_type(int mstype)
790{
791 switch (mstype) {
792 case DLM_MSG_REQUEST:
793 return DLM_MSG_REQUEST_REPLY;
794 case DLM_MSG_CONVERT:
795 return DLM_MSG_CONVERT_REPLY;
796 case DLM_MSG_UNLOCK:
797 return DLM_MSG_UNLOCK_REPLY;
798 case DLM_MSG_CANCEL:
799 return DLM_MSG_CANCEL_REPLY;
800 case DLM_MSG_LOOKUP:
801 return DLM_MSG_LOOKUP_REPLY;
802 }
803 return -1;
804}
805
e7fd4179
DT		 806/* add/remove lkb from global waiters list of lkb's waiting for
807 a reply from a remote node */
808
ef0c2bb0 809static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
e7fd4179
DT		 810{
811 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
ef0c2bb0 812 int error = 0;
e7fd4179 813
90135925 814 mutex_lock(&ls->ls_waiters_mutex);
ef0c2bb0
DT		 815
816 if (is_overlap_unlock(lkb) ||
817 (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
818 error = -EINVAL;
819 goto out;
820 }
821
822 if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
823 switch (mstype) {
824 case DLM_MSG_UNLOCK:
825 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
826 break;
827 case DLM_MSG_CANCEL:
828 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
829 break;
830 default:
831 error = -EBUSY;
832 goto out;
833 }
834 lkb->lkb_wait_count++;
835 hold_lkb(lkb);
836
837 log_debug(ls, "add overlap %x cur %d new %d count %d flags %x",
838 lkb->lkb_id, lkb->lkb_wait_type, mstype,
839 lkb->lkb_wait_count, lkb->lkb_flags);
e7fd4179
DT		 840 goto out;
841 }
ef0c2bb0
DT		 842
843 DLM_ASSERT(!lkb->lkb_wait_count,
844 dlm_print_lkb(lkb);
845 printk("wait_count %d\n", lkb->lkb_wait_count););
846
847 lkb->lkb_wait_count++;
e7fd4179 848 lkb->lkb_wait_type = mstype;
ef0c2bb0 849 hold_lkb(lkb);
e7fd4179
DT		 850 list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
851 out:
ef0c2bb0
DT		 852 if (error)
853 log_error(ls, "add_to_waiters %x error %d flags %x %d %d %s",
854 lkb->lkb_id, error, lkb->lkb_flags, mstype,
855 lkb->lkb_wait_type, lkb->lkb_resource->res_name);
90135925 856 mutex_unlock(&ls->ls_waiters_mutex);
ef0c2bb0 857 return error;
e7fd4179
DT		 858}
859
b790c3b7
DT		 860/* We clear the RESEND flag because we might be taking an lkb off the waiters
861 list as part of process_requestqueue (e.g. a lookup that has an optimized
862 request reply on the requestqueue) between dlm_recover_waiters_pre() which
863 set RESEND and dlm_recover_waiters_post() */
864
ef0c2bb0 865static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
e7fd4179 866{
ef0c2bb0
DT		 867 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
868 int overlap_done = 0;
e7fd4179 869
ef0c2bb0
DT		 870 if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
871 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
872 overlap_done = 1;
873 goto out_del;
e7fd4179 874 }
ef0c2bb0
DT		 875
876 if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
877 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
878 overlap_done = 1;
879 goto out_del;
880 }
881
882 /* N.B. type of reply may not always correspond to type of original
883 msg due to lookup->request optimization, verify others? */
884
885 if (lkb->lkb_wait_type) {
886 lkb->lkb_wait_type = 0;
887 goto out_del;
888 }
889
890 log_error(ls, "remove_from_waiters lkid %x flags %x types %d %d",
891 lkb->lkb_id, lkb->lkb_flags, mstype, lkb->lkb_wait_type);
892 return -1;
893
894 out_del:
895 /* the force-unlock/cancel has completed and we haven't recvd a reply
896 to the op that was in progress prior to the unlock/cancel; we
897 give up on any reply to the earlier op. FIXME: not sure when/how
898 this would happen */
899
900 if (overlap_done && lkb->lkb_wait_type) {
901 log_error(ls, "remove_from_waiters %x reply %d give up on %d",
902 lkb->lkb_id, mstype, lkb->lkb_wait_type);
903 lkb->lkb_wait_count--;
904 lkb->lkb_wait_type = 0;
905 }
906
907 DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
908
b790c3b7 909 lkb->lkb_flags &= ~DLM_IFL_RESEND;
ef0c2bb0
DT		 910 lkb->lkb_wait_count--;
911 if (!lkb->lkb_wait_count)
912 list_del_init(&lkb->lkb_wait_reply);
e7fd4179 913 unhold_lkb(lkb);
ef0c2bb0 914 return 0;
e7fd4179
DT		 915}
916
ef0c2bb0 917static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
e7fd4179
DT		 918{
919 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
920 int error;
921
90135925 922 mutex_lock(&ls->ls_waiters_mutex);
ef0c2bb0 923 error = _remove_from_waiters(lkb, mstype);
90135925 924 mutex_unlock(&ls->ls_waiters_mutex);
e7fd4179
DT		 925 return error;
926}
927
ef0c2bb0
DT		 928/* Handles situations where we might be processing a "fake" or "stub" reply in
929 which we can't try to take waiters_mutex again. */
930
931static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
932{
933 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
934 int error;
935
936 if (ms != &ls->ls_stub_ms)
937 mutex_lock(&ls->ls_waiters_mutex);
938 error = _remove_from_waiters(lkb, ms->m_type);
939 if (ms != &ls->ls_stub_ms)
940 mutex_unlock(&ls->ls_waiters_mutex);
941 return error;
942}
943
e7fd4179
DT		 944static void dir_remove(struct dlm_rsb *r)
945{
946 int to_nodeid;
947
948 if (dlm_no_directory(r->res_ls))
949 return;
950
951 to_nodeid = dlm_dir_nodeid(r);
952 if (to_nodeid != dlm_our_nodeid())
953 send_remove(r);
954 else
955 dlm_dir_remove_entry(r->res_ls, to_nodeid,
956 r->res_name, r->res_length);
957}
958
959/* FIXME: shouldn't this be able to exit as soon as one non-due rsb is
960 found since they are in order of newest to oldest? */
961
962static int shrink_bucket(struct dlm_ls *ls, int b)
963{
964 struct dlm_rsb *r;
965 int count = 0, found;
966
967 for (;;) {
90135925 968 found = 0;
e7fd4179
DT		 969 write_lock(&ls->ls_rsbtbl[b].lock);
970 list_for_each_entry_reverse(r, &ls->ls_rsbtbl[b].toss,
971 res_hashchain) {
972 if (!time_after_eq(jiffies, r->res_toss_time +
68c817a1 973 dlm_config.ci_toss_secs * HZ))
e7fd4179 974 continue;
90135925 975 found = 1;
e7fd4179
DT		 976 break;
977 }
978
979 if (!found) {
980 write_unlock(&ls->ls_rsbtbl[b].lock);
981 break;
982 }
983
984 if (kref_put(&r->res_ref, kill_rsb)) {
985 list_del(&r->res_hashchain);
986 write_unlock(&ls->ls_rsbtbl[b].lock);
987
988 if (is_master(r))
989 dir_remove(r);
52bda2b5 990 dlm_free_rsb(r);
e7fd4179
DT		 991 count++;
992 } else {
993 write_unlock(&ls->ls_rsbtbl[b].lock);
994 log_error(ls, "tossed rsb in use %s", r->res_name);
995 }
996 }
997
998 return count;
999}
1000
1001void dlm_scan_rsbs(struct dlm_ls *ls)
1002{
1003 int i;
1004
e7fd4179
DT		 1005 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
1006 shrink_bucket(ls, i);
85e86edf
DT		 1007 if (dlm_locking_stopped(ls))
1008 break;
e7fd4179
DT		 1009 cond_resched();
1010 }
1011}
1012
3ae1acf9
DT		 1013static void add_timeout(struct dlm_lkb *lkb)
1014{
1015 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1016
84d8cd69
DT		 1017 if (is_master_copy(lkb)) {
1018 lkb->lkb_timestamp = jiffies;
3ae1acf9 1019 return;
84d8cd69 1020 }
3ae1acf9
DT		 1021
1022 if (test_bit(LSFL_TIMEWARN, &ls->ls_flags) &&
1023 !(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1024 lkb->lkb_flags |= DLM_IFL_WATCH_TIMEWARN;
1025 goto add_it;
1026 }
84d8cd69
DT		 1027 if (lkb->lkb_exflags & DLM_LKF_TIMEOUT)
1028 goto add_it;
3ae1acf9
DT		 1029 return;
1030
1031 add_it:
1032 DLM_ASSERT(list_empty(&lkb->lkb_time_list), dlm_print_lkb(lkb););
1033 mutex_lock(&ls->ls_timeout_mutex);
1034 hold_lkb(lkb);
1035 lkb->lkb_timestamp = jiffies;
1036 list_add_tail(&lkb->lkb_time_list, &ls->ls_timeout);
1037 mutex_unlock(&ls->ls_timeout_mutex);
1038}
1039
1040static void del_timeout(struct dlm_lkb *lkb)
1041{
1042 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
1043
1044 mutex_lock(&ls->ls_timeout_mutex);
1045 if (!list_empty(&lkb->lkb_time_list)) {
1046 list_del_init(&lkb->lkb_time_list);
1047 unhold_lkb(lkb);
1048 }
1049 mutex_unlock(&ls->ls_timeout_mutex);
1050}
1051
1052/* FIXME: is it safe to look at lkb_exflags, lkb_flags, lkb_timestamp, and
1053 lkb_lksb_timeout without lock_rsb? Note: we can't lock timeout_mutex
1054 and then lock rsb because of lock ordering in add_timeout. We may need
1055 to specify some special timeout-related bits in the lkb that are just to
1056 be accessed under the timeout_mutex. */
1057
1058void dlm_scan_timeout(struct dlm_ls *ls)
1059{
1060 struct dlm_rsb *r;
1061 struct dlm_lkb *lkb;
1062 int do_cancel, do_warn;
1063
1064 for (;;) {
1065 if (dlm_locking_stopped(ls))
1066 break;
1067
1068 do_cancel = 0;
1069 do_warn = 0;
1070 mutex_lock(&ls->ls_timeout_mutex);
1071 list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list) {
1072
1073 if ((lkb->lkb_exflags & DLM_LKF_TIMEOUT) &&
1074 time_after_eq(jiffies, lkb->lkb_timestamp +
1075 lkb->lkb_timeout_cs * HZ/100))
1076 do_cancel = 1;
1077
1078 if ((lkb->lkb_flags & DLM_IFL_WATCH_TIMEWARN) &&
1079 time_after_eq(jiffies, lkb->lkb_timestamp +
1080 dlm_config.ci_timewarn_cs * HZ/100))
1081 do_warn = 1;
1082
1083 if (!do_cancel && !do_warn)
1084 continue;
1085 hold_lkb(lkb);
1086 break;
1087 }
1088 mutex_unlock(&ls->ls_timeout_mutex);
1089
1090 if (!do_cancel && !do_warn)
1091 break;
1092
1093 r = lkb->lkb_resource;
1094 hold_rsb(r);
1095 lock_rsb(r);
1096
1097 if (do_warn) {
1098 /* clear flag so we only warn once */
1099 lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1100 if (!(lkb->lkb_exflags & DLM_LKF_TIMEOUT))
1101 del_timeout(lkb);
1102 dlm_timeout_warn(lkb);
1103 }
1104
1105 if (do_cancel) {
b3cab7b9 1106 log_debug(ls, "timeout cancel %x node %d %s",
639aca41 1107 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
3ae1acf9
DT		 1108 lkb->lkb_flags &= ~DLM_IFL_WATCH_TIMEWARN;
1109 lkb->lkb_flags |= DLM_IFL_TIMEOUT_CANCEL;
1110 del_timeout(lkb);
1111 _cancel_lock(r, lkb);
1112 }
1113
1114 unlock_rsb(r);
1115 unhold_rsb(r);
1116 dlm_put_lkb(lkb);
1117 }
1118}
1119
1120/* This is only called by dlm_recoverd, and we rely on dlm_ls_stop() stopping
1121 dlm_recoverd before checking/setting ls_recover_begin. */
1122
1123void dlm_adjust_timeouts(struct dlm_ls *ls)
1124{
1125 struct dlm_lkb *lkb;
1126 long adj = jiffies - ls->ls_recover_begin;
1127
1128 ls->ls_recover_begin = 0;
1129 mutex_lock(&ls->ls_timeout_mutex);
1130 list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list)
1131 lkb->lkb_timestamp += adj;
1132 mutex_unlock(&ls->ls_timeout_mutex);
1133}
1134
e7fd4179
DT		 1135/* lkb is master or local copy */
1136
1137static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1138{
1139 int b, len = r->res_ls->ls_lvblen;
1140
1141 /* b=1 lvb returned to caller
1142 b=0 lvb written to rsb or invalidated
1143 b=-1 do nothing */
1144
1145 b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
1146
1147 if (b == 1) {
1148 if (!lkb->lkb_lvbptr)
1149 return;
1150
1151 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1152 return;
1153
1154 if (!r->res_lvbptr)
1155 return;
1156
1157 memcpy(lkb->lkb_lvbptr, r->res_lvbptr, len);
1158 lkb->lkb_lvbseq = r->res_lvbseq;
1159
1160 } else if (b == 0) {
1161 if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1162 rsb_set_flag(r, RSB_VALNOTVALID);
1163 return;
1164 }
1165
1166 if (!lkb->lkb_lvbptr)
1167 return;
1168
1169 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1170 return;
1171
1172 if (!r->res_lvbptr)
52bda2b5 1173 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
e7fd4179
DT		 1174
1175 if (!r->res_lvbptr)
1176 return;
1177
1178 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, len);
1179 r->res_lvbseq++;
1180 lkb->lkb_lvbseq = r->res_lvbseq;
1181 rsb_clear_flag(r, RSB_VALNOTVALID);
1182 }
1183
1184 if (rsb_flag(r, RSB_VALNOTVALID))
1185 lkb->lkb_sbflags |= DLM_SBF_VALNOTVALID;
1186}
1187
1188static void set_lvb_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1189{
1190 if (lkb->lkb_grmode < DLM_LOCK_PW)
1191 return;
1192
1193 if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) {
1194 rsb_set_flag(r, RSB_VALNOTVALID);
1195 return;
1196 }
1197
1198 if (!lkb->lkb_lvbptr)
1199 return;
1200
1201 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1202 return;
1203
1204 if (!r->res_lvbptr)
52bda2b5 1205 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
e7fd4179
DT		 1206
1207 if (!r->res_lvbptr)
1208 return;
1209
1210 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
1211 r->res_lvbseq++;
1212 rsb_clear_flag(r, RSB_VALNOTVALID);
1213}
1214
1215/* lkb is process copy (pc) */
1216
1217static void set_lvb_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1218 struct dlm_message *ms)
1219{
1220 int b;
1221
1222 if (!lkb->lkb_lvbptr)
1223 return;
1224
1225 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
1226 return;
1227
597d0cae 1228 b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1];
e7fd4179
DT		 1229 if (b == 1) {
1230 int len = receive_extralen(ms);
1231 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
1232 lkb->lkb_lvbseq = ms->m_lvbseq;
1233 }
1234}
1235
1236/* Manipulate lkb's on rsb's convert/granted/waiting queues
1237 remove_lock -- used for unlock, removes lkb from granted
1238 revert_lock -- used for cancel, moves lkb from convert to granted
1239 grant_lock -- used for request and convert, adds lkb to granted or
1240 moves lkb from convert or waiting to granted
1241
1242 Each of these is used for master or local copy lkb's. There is
1243 also a _pc() variation used to make the corresponding change on
1244 a process copy (pc) lkb. */
1245
1246static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1247{
1248 del_lkb(r, lkb);
1249 lkb->lkb_grmode = DLM_LOCK_IV;
1250 /* this unhold undoes the original ref from create_lkb()
1251 so this leads to the lkb being freed */
1252 unhold_lkb(lkb);
1253}
1254
1255static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1256{
1257 set_lvb_unlock(r, lkb);
1258 _remove_lock(r, lkb);
1259}
1260
1261static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
1262{
1263 _remove_lock(r, lkb);
1264}
1265
ef0c2bb0
DT		 1266/* returns: 0 did nothing
1267 1 moved lock to granted
1268 -1 removed lock */
1269
1270static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
e7fd4179 1271{
ef0c2bb0
DT		 1272 int rv = 0;
1273
e7fd4179
DT		 1274 lkb->lkb_rqmode = DLM_LOCK_IV;
1275
1276 switch (lkb->lkb_status) {
597d0cae
DT		 1277 case DLM_LKSTS_GRANTED:
1278 break;
e7fd4179
DT		 1279 case DLM_LKSTS_CONVERT:
1280 move_lkb(r, lkb, DLM_LKSTS_GRANTED);
ef0c2bb0 1281 rv = 1;
e7fd4179
DT		 1282 break;
1283 case DLM_LKSTS_WAITING:
1284 del_lkb(r, lkb);
1285 lkb->lkb_grmode = DLM_LOCK_IV;
1286 /* this unhold undoes the original ref from create_lkb()
1287 so this leads to the lkb being freed */
1288 unhold_lkb(lkb);
ef0c2bb0 1289 rv = -1;
e7fd4179
DT		 1290 break;
1291 default:
1292 log_print("invalid status for revert %d", lkb->lkb_status);
1293 }
ef0c2bb0 1294 return rv;
e7fd4179
DT		 1295}
1296
ef0c2bb0 1297static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
e7fd4179 1298{
ef0c2bb0 1299 return revert_lock(r, lkb);
e7fd4179
DT		 1300}
1301
1302static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1303{
1304 if (lkb->lkb_grmode != lkb->lkb_rqmode) {
1305 lkb->lkb_grmode = lkb->lkb_rqmode;
1306 if (lkb->lkb_status)
1307 move_lkb(r, lkb, DLM_LKSTS_GRANTED);
1308 else
1309 add_lkb(r, lkb, DLM_LKSTS_GRANTED);
1310 }
1311
1312 lkb->lkb_rqmode = DLM_LOCK_IV;
e7fd4179
DT		 1313}
1314
1315static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
1316{
1317 set_lvb_lock(r, lkb);
1318 _grant_lock(r, lkb);
1319 lkb->lkb_highbast = 0;
1320}
1321
1322static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb,
1323 struct dlm_message *ms)
1324{
1325 set_lvb_lock_pc(r, lkb, ms);
1326 _grant_lock(r, lkb);
1327}
1328
1329/* called by grant_pending_locks() which means an async grant message must
1330 be sent to the requesting node in addition to granting the lock if the
1331 lkb belongs to a remote node. */
1332
1333static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
1334{
1335 grant_lock(r, lkb);
1336 if (is_master_copy(lkb))
1337 send_grant(r, lkb);
1338 else
1339 queue_cast(r, lkb, 0);
1340}
1341
7d3c1feb
DT		 1342/* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
1343 change the granted/requested modes. We're munging things accordingly in
1344 the process copy.
1345 CONVDEADLK: our grmode may have been forced down to NL to resolve a
1346 conversion deadlock
1347 ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
1348 compatible with other granted locks */
1349
1350static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
1351{
1352 if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
1353 log_print("munge_demoted %x invalid reply type %d",
1354 lkb->lkb_id, ms->m_type);
1355 return;
1356 }
1357
1358 if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
1359 log_print("munge_demoted %x invalid modes gr %d rq %d",
1360 lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
1361 return;
1362 }
1363
1364 lkb->lkb_grmode = DLM_LOCK_NL;
1365}
1366
1367static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
1368{
1369 if (ms->m_type != DLM_MSG_REQUEST_REPLY &&
1370 ms->m_type != DLM_MSG_GRANT) {
1371 log_print("munge_altmode %x invalid reply type %d",
1372 lkb->lkb_id, ms->m_type);
1373 return;
1374 }
1375
1376 if (lkb->lkb_exflags & DLM_LKF_ALTPR)
1377 lkb->lkb_rqmode = DLM_LOCK_PR;
1378 else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
1379 lkb->lkb_rqmode = DLM_LOCK_CW;
1380 else {
1381 log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
1382 dlm_print_lkb(lkb);
1383 }
1384}
1385
e7fd4179
DT		 1386static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
1387{
1388 struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
1389 lkb_statequeue);
1390 if (lkb->lkb_id == first->lkb_id)
90135925 1391 return 1;
e7fd4179 1392
90135925 1393 return 0;
e7fd4179
DT		 1394}
1395
e7fd4179
DT		 1396/* Check if the given lkb conflicts with another lkb on the queue. */
1397
1398static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb)
1399{
1400 struct dlm_lkb *this;
1401
1402 list_for_each_entry(this, head, lkb_statequeue) {
1403 if (this == lkb)
1404 continue;
3bcd3687 1405 if (!modes_compat(this, lkb))
90135925 1406 return 1;
e7fd4179 1407 }
90135925 1408 return 0;
e7fd4179
DT		 1409}
1410
1411/*
1412 * "A conversion deadlock arises with a pair of lock requests in the converting
1413 * queue for one resource. The granted mode of each lock blocks the requested
1414 * mode of the other lock."
1415 *
c85d65e9
DT		 1416 * Part 2: if the granted mode of lkb is preventing an earlier lkb in the
1417 * convert queue from being granted, then deadlk/demote lkb.
e7fd4179
DT		 1418 *
1419 * Example:
1420 * Granted Queue: empty
1421 * Convert Queue: NL->EX (first lock)
1422 * PR->EX (second lock)
1423 *
1424 * The first lock can't be granted because of the granted mode of the second
1425 * lock and the second lock can't be granted because it's not first in the
c85d65e9
DT		 1426 * list. We either cancel lkb's conversion (PR->EX) and return EDEADLK, or we
1427 * demote the granted mode of lkb (from PR to NL) if it has the CONVDEADLK
1428 * flag set and return DEMOTED in the lksb flags.
e7fd4179 1429 *
c85d65e9
DT		 1430 * Originally, this function detected conv-deadlk in a more limited scope:
1431 * - if !modes_compat(lkb1, lkb2) && !modes_compat(lkb2, lkb1), or
1432 * - if lkb1 was the first entry in the queue (not just earlier), and was
1433 * blocked by the granted mode of lkb2, and there was nothing on the
1434 * granted queue preventing lkb1 from being granted immediately, i.e.
1435 * lkb2 was the only thing preventing lkb1 from being granted.
1436 *
1437 * That second condition meant we'd only say there was conv-deadlk if
1438 * resolving it (by demotion) would lead to the first lock on the convert
1439 * queue being granted right away. It allowed conversion deadlocks to exist
1440 * between locks on the convert queue while they couldn't be granted anyway.
1441 *
1442 * Now, we detect and take action on conversion deadlocks immediately when
1443 * they're created, even if they may not be immediately consequential. If
1444 * lkb1 exists anywhere in the convert queue and lkb2 comes in with a granted
1445 * mode that would prevent lkb1's conversion from being granted, we do a
1446 * deadlk/demote on lkb2 right away and don't let it onto the convert queue.
1447 * I think this means that the lkb_is_ahead condition below should always
1448 * be zero, i.e. there will never be conv-deadlk between two locks that are
1449 * both already on the convert queue.
e7fd4179
DT		 1450 */
1451
c85d65e9 1452static int conversion_deadlock_detect(struct dlm_rsb *r, struct dlm_lkb *lkb2)
e7fd4179 1453{
c85d65e9
DT		 1454 struct dlm_lkb *lkb1;
1455 int lkb_is_ahead = 0;
e7fd4179 1456
c85d65e9
DT		 1457 list_for_each_entry(lkb1, &r->res_convertqueue, lkb_statequeue) {
1458 if (lkb1 == lkb2) {
1459 lkb_is_ahead = 1;
e7fd4179
DT		 1460 continue;
1461 }
1462
c85d65e9
DT		 1463 if (!lkb_is_ahead) {
1464 if (!modes_compat(lkb2, lkb1))
1465 return 1;
1466 } else {
1467 if (!modes_compat(lkb2, lkb1) &&
1468 !modes_compat(lkb1, lkb2))
1469 return 1;
1470 }
e7fd4179 1471 }
90135925 1472 return 0;
e7fd4179
DT		 1473}
1474
1475/*
1476 * Return 1 if the lock can be granted, 0 otherwise.
1477 * Also detect and resolve conversion deadlocks.
1478 *
1479 * lkb is the lock to be granted
1480 *
1481 * now is 1 if the function is being called in the context of the
1482 * immediate request, it is 0 if called later, after the lock has been
1483 * queued.
1484 *
1485 * References are from chapter 6 of "VAXcluster Principles" by Roy Davis
1486 */
1487
1488static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now)
1489{
1490 int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV);
1491
1492 /*
1493 * 6-10: Version 5.4 introduced an option to address the phenomenon of
1494 * a new request for a NL mode lock being blocked.
1495 *
1496 * 6-11: If the optional EXPEDITE flag is used with the new NL mode
1497 * request, then it would be granted. In essence, the use of this flag
1498 * tells the Lock Manager to expedite theis request by not considering
1499 * what may be in the CONVERTING or WAITING queues... As of this
1500 * writing, the EXPEDITE flag can be used only with new requests for NL
1501 * mode locks. This flag is not valid for conversion requests.
1502 *
1503 * A shortcut. Earlier checks return an error if EXPEDITE is used in a
1504 * conversion or used with a non-NL requested mode. We also know an
1505 * EXPEDITE request is always granted immediately, so now must always
1506 * be 1. The full condition to grant an expedite request: (now &&
1507 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can
1508 * therefore be shortened to just checking the flag.
1509 */
1510
1511 if (lkb->lkb_exflags & DLM_LKF_EXPEDITE)
90135925 1512 return 1;
e7fd4179
DT		 1513
1514 /*
1515 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be
1516 * added to the remaining conditions.
1517 */
1518
1519 if (queue_conflict(&r->res_grantqueue, lkb))
1520 goto out;
1521
1522 /*
1523 * 6-3: By default, a conversion request is immediately granted if the
1524 * requested mode is compatible with the modes of all other granted
1525 * locks
1526 */
1527
1528 if (queue_conflict(&r->res_convertqueue, lkb))
1529 goto out;
1530
1531 /*
1532 * 6-5: But the default algorithm for deciding whether to grant or
1533 * queue conversion requests does not by itself guarantee that such
1534 * requests are serviced on a "first come first serve" basis. This, in
1535 * turn, can lead to a phenomenon known as "indefinate postponement".
1536 *
1537 * 6-7: This issue is dealt with by using the optional QUECVT flag with
1538 * the system service employed to request a lock conversion. This flag
1539 * forces certain conversion requests to be queued, even if they are
1540 * compatible with the granted modes of other locks on the same
1541 * resource. Thus, the use of this flag results in conversion requests
1542 * being ordered on a "first come first servce" basis.
1543 *
1544 * DCT: This condition is all about new conversions being able to occur
1545 * "in place" while the lock remains on the granted queue (assuming
1546 * nothing else conflicts.) IOW if QUECVT isn't set, a conversion
1547 * doesn't _have_ to go onto the convert queue where it's processed in
1548 * order. The "now" variable is necessary to distinguish converts
1549 * being received and processed for the first time now, because once a
1550 * convert is moved to the conversion queue the condition below applies
1551 * requiring fifo granting.
1552 */
1553
1554 if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT))
90135925 1555 return 1;
e7fd4179
DT		 1556
1557 /*
3bcd3687
DT		 1558 * The NOORDER flag is set to avoid the standard vms rules on grant
1559 * order.
e7fd4179
DT		 1560 */
1561
1562 if (lkb->lkb_exflags & DLM_LKF_NOORDER)
90135925 1563 return 1;
e7fd4179
DT		 1564
1565 /*
1566 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be
1567 * granted until all other conversion requests ahead of it are granted
1568 * and/or canceled.
1569 */
1570
1571 if (!now && conv && first_in_list(lkb, &r->res_convertqueue))
90135925 1572 return 1;
e7fd4179
DT		 1573
1574 /*
1575 * 6-4: By default, a new request is immediately granted only if all
1576 * three of the following conditions are satisfied when the request is
1577 * issued:
1578 * - The queue of ungranted conversion requests for the resource is
1579 * empty.
1580 * - The queue of ungranted new requests for the resource is empty.
1581 * - The mode of the new request is compatible with the most
1582 * restrictive mode of all granted locks on the resource.
1583 */
1584
1585 if (now && !conv && list_empty(&r->res_convertqueue) &&
1586 list_empty(&r->res_waitqueue))
90135925 1587 return 1;
e7fd4179
DT		 1588
1589 /*
1590 * 6-4: Once a lock request is in the queue of ungranted new requests,
1591 * it cannot be granted until the queue of ungranted conversion
1592 * requests is empty, all ungranted new requests ahead of it are
1593 * granted and/or canceled, and it is compatible with the granted mode
1594 * of the most restrictive lock granted on the resource.
1595 */
1596
1597 if (!now && !conv && list_empty(&r->res_convertqueue) &&
1598 first_in_list(lkb, &r->res_waitqueue))
90135925 1599 return 1;
e7fd4179 1600 out:
90135925 1601 return 0;
e7fd4179
DT		 1602}
1603
c85d65e9
DT		 1604static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
1605 int *err)
e7fd4179 1606{
e7fd4179
DT		 1607 int rv;
1608 int8_t alt = 0, rqmode = lkb->lkb_rqmode;
c85d65e9
DT		 1609 int8_t is_convert = (lkb->lkb_grmode != DLM_LOCK_IV);
1610
1611 if (err)
1612 *err = 0;
e7fd4179
DT		 1613
1614 rv = _can_be_granted(r, lkb, now);
1615 if (rv)
1616 goto out;
1617
c85d65e9
DT		 1618 /*
1619 * The CONVDEADLK flag is non-standard and tells the dlm to resolve
1620 * conversion deadlocks by demoting grmode to NL, otherwise the dlm
1621 * cancels one of the locks.
1622 */
1623
1624 if (is_convert && can_be_queued(lkb) &&
1625 conversion_deadlock_detect(r, lkb)) {
1626 if (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) {
1627 lkb->lkb_grmode = DLM_LOCK_NL;
1628 lkb->lkb_sbflags |= DLM_SBF_DEMOTED;
1629 } else if (!(lkb->lkb_exflags & DLM_LKF_NODLCKWT)) {
1630 if (err)
1631 *err = -EDEADLK;
1632 else {
1633 log_print("can_be_granted deadlock %x now %d",
1634 lkb->lkb_id, now);
1635 dlm_dump_rsb(r);
1636 }
1637 }
e7fd4179 1638 goto out;
c85d65e9 1639 }
e7fd4179 1640
c85d65e9
DT		 1641 /*
1642 * The ALTPR and ALTCW flags are non-standard and tell the dlm to try
1643 * to grant a request in a mode other than the normal rqmode. It's a
1644 * simple way to provide a big optimization to applications that can
1645 * use them.
1646 */
1647
1648 if (rqmode != DLM_LOCK_PR && (lkb->lkb_exflags & DLM_LKF_ALTPR))
e7fd4179 1649 alt = DLM_LOCK_PR;
c85d65e9 1650 else if (rqmode != DLM_LOCK_CW && (lkb->lkb_exflags & DLM_LKF_ALTCW))
e7fd4179
DT		 1651 alt = DLM_LOCK_CW;
1652
1653 if (alt) {
1654 lkb->lkb_rqmode = alt;
1655 rv = _can_be_granted(r, lkb, now);
1656 if (rv)
1657 lkb->lkb_sbflags |= DLM_SBF_ALTMODE;
1658 else
1659 lkb->lkb_rqmode = rqmode;
1660 }
1661 out:
1662 return rv;
1663}
1664
c85d65e9
DT		 1665/* FIXME: I don't think that can_be_granted() can/will demote or find deadlock
1666 for locks pending on the convert list. Once verified (watch for these
1667 log_prints), we should be able to just call _can_be_granted() and not
1668 bother with the demote/deadlk cases here (and there's no easy way to deal
1669 with a deadlk here, we'd have to generate something like grant_lock with
1670 the deadlk error.) */
1671
36509258
DT		 1672/* Returns the highest requested mode of all blocked conversions; sets
1673 cw if there's a blocked conversion to DLM_LOCK_CW. */
c85d65e9 1674
36509258 1675static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw)
e7fd4179
DT		 1676{
1677 struct dlm_lkb *lkb, *s;
1678 int hi, demoted, quit, grant_restart, demote_restart;
c85d65e9 1679 int deadlk;
e7fd4179
DT		 1680
1681 quit = 0;
1682 restart:
1683 grant_restart = 0;
1684 demote_restart = 0;
1685 hi = DLM_LOCK_IV;
1686
1687 list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) {
1688 demoted = is_demoted(lkb);
c85d65e9
DT		 1689 deadlk = 0;
1690
1691 if (can_be_granted(r, lkb, 0, &deadlk)) {
e7fd4179
DT		 1692 grant_lock_pending(r, lkb);
1693 grant_restart = 1;
c85d65e9 1694 continue;
e7fd4179 1695 }
c85d65e9
DT		 1696
1697 if (!demoted && is_demoted(lkb)) {
1698 log_print("WARN: pending demoted %x node %d %s",
1699 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1700 demote_restart = 1;
1701 continue;
1702 }
1703
1704 if (deadlk) {
1705 log_print("WARN: pending deadlock %x node %d %s",
1706 lkb->lkb_id, lkb->lkb_nodeid, r->res_name);
1707 dlm_dump_rsb(r);
1708 continue;
1709 }
1710
1711 hi = max_t(int, lkb->lkb_rqmode, hi);
36509258
DT		 1712
1713 if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
1714 *cw = 1;
e7fd4179
DT		 1715 }
1716
1717 if (grant_restart)
1718 goto restart;
1719 if (demote_restart && !quit) {
1720 quit = 1;
1721 goto restart;
1722 }
1723
1724 return max_t(int, high, hi);
1725}
1726
36509258 1727static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw)
e7fd4179
DT		 1728{
1729 struct dlm_lkb *lkb, *s;
1730
1731 list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
c85d65e9 1732 if (can_be_granted(r, lkb, 0, NULL))
e7fd4179 1733 grant_lock_pending(r, lkb);
36509258 1734 else {
e7fd4179 1735 high = max_t(int, lkb->lkb_rqmode, high);
36509258
DT		 1736 if (lkb->lkb_rqmode == DLM_LOCK_CW)
1737 *cw = 1;
1738 }
e7fd4179
DT		 1739 }
1740
1741 return high;
1742}
1743
36509258
DT		 1744/* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
1745 on either the convert or waiting queue.
1746 high is the largest rqmode of all locks blocked on the convert or
1747 waiting queue. */
1748
1749static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
1750{
1751 if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
1752 if (gr->lkb_highbast < DLM_LOCK_EX)
1753 return 1;
1754 return 0;
1755 }
1756
1757 if (gr->lkb_highbast < high &&
1758 !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
1759 return 1;
1760 return 0;
1761}
1762
e7fd4179
DT		 1763static void grant_pending_locks(struct dlm_rsb *r)
1764{
1765 struct dlm_lkb *lkb, *s;
1766 int high = DLM_LOCK_IV;
36509258 1767 int cw = 0;
e7fd4179 1768
a345da3e 1769 DLM_ASSERT(is_master(r), dlm_dump_rsb(r););
e7fd4179 1770
36509258
DT		 1771 high = grant_pending_convert(r, high, &cw);
1772 high = grant_pending_wait(r, high, &cw);
e7fd4179
DT		 1773
1774 if (high == DLM_LOCK_IV)
1775 return;
1776
1777 /*
1778 * If there are locks left on the wait/convert queue then send blocking
1779 * ASTs to granted locks based on the largest requested mode (high)
36509258 1780 * found above.
e7fd4179
DT		 1781 */
1782
1783 list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
36509258
DT		 1784 if (lkb->lkb_bastaddr && lock_requires_bast(lkb, high, cw)) {
1785 if (cw && high == DLM_LOCK_PR)
1786 queue_bast(r, lkb, DLM_LOCK_CW);
1787 else
1788 queue_bast(r, lkb, high);
e7fd4179
DT		 1789 lkb->lkb_highbast = high;
1790 }
1791 }
1792}
1793
36509258
DT		 1794static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
1795{
1796 if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
1797 (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
1798 if (gr->lkb_highbast < DLM_LOCK_EX)
1799 return 1;
1800 return 0;
1801 }
1802
1803 if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
1804 return 1;
1805 return 0;
1806}
1807
e7fd4179
DT		 1808static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
1809 struct dlm_lkb *lkb)
1810{
1811 struct dlm_lkb *gr;
1812
1813 list_for_each_entry(gr, head, lkb_statequeue) {
36509258 1814 if (gr->lkb_bastaddr && modes_require_bast(gr, lkb)) {
e7fd4179
DT		 1815 queue_bast(r, gr, lkb->lkb_rqmode);
1816 gr->lkb_highbast = lkb->lkb_rqmode;
1817 }
1818 }
1819}
1820
1821static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb)
1822{
1823 send_bast_queue(r, &r->res_grantqueue, lkb);
1824}
1825
1826static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb)
1827{
1828 send_bast_queue(r, &r->res_grantqueue, lkb);
1829 send_bast_queue(r, &r->res_convertqueue, lkb);
1830}
1831
1832/* set_master(r, lkb) -- set the master nodeid of a resource
1833
1834 The purpose of this function is to set the nodeid field in the given
1835 lkb using the nodeid field in the given rsb. If the rsb's nodeid is
1836 known, it can just be copied to the lkb and the function will return
1837 0. If the rsb's nodeid is _not_ known, it needs to be looked up
1838 before it can be copied to the lkb.
1839
1840 When the rsb nodeid is being looked up remotely, the initial lkb
1841 causing the lookup is kept on the ls_waiters list waiting for the
1842 lookup reply. Other lkb's waiting for the same rsb lookup are kept
1843 on the rsb's res_lookup list until the master is verified.
1844
1845 Return values:
1846 0: nodeid is set in rsb/lkb and the caller should go ahead and use it
1847 1: the rsb master is not available and the lkb has been placed on
1848 a wait queue
1849*/
1850
1851static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb)
1852{
1853 struct dlm_ls *ls = r->res_ls;
755b5eb8 1854 int i, error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid();
e7fd4179
DT		 1855
1856 if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) {
1857 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN);
1858 r->res_first_lkid = lkb->lkb_id;
1859 lkb->lkb_nodeid = r->res_nodeid;
1860 return 0;
1861 }
1862
1863 if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) {
1864 list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup);
1865 return 1;
1866 }
1867
1868 if (r->res_nodeid == 0) {
1869 lkb->lkb_nodeid = 0;
1870 return 0;
1871 }
1872
1873 if (r->res_nodeid > 0) {
1874 lkb->lkb_nodeid = r->res_nodeid;
1875 return 0;
1876 }
1877
a345da3e 1878 DLM_ASSERT(r->res_nodeid == -1, dlm_dump_rsb(r););
e7fd4179
DT		 1879
1880 dir_nodeid = dlm_dir_nodeid(r);
1881
1882 if (dir_nodeid != our_nodeid) {
1883 r->res_first_lkid = lkb->lkb_id;
1884 send_lookup(r, lkb);
1885 return 1;
1886 }
1887
755b5eb8 1888 for (i = 0; i < 2; i++) {
e7fd4179
DT		 1889 /* It's possible for dlm_scand to remove an old rsb for
1890 this same resource from the toss list, us to create
1891 a new one, look up the master locally, and find it
1892 already exists just before dlm_scand does the
1893 dir_remove() on the previous rsb. */
1894
1895 error = dlm_dir_lookup(ls, our_nodeid, r->res_name,
1896 r->res_length, &ret_nodeid);
1897 if (!error)
1898 break;
1899 log_debug(ls, "dir_lookup error %d %s", error, r->res_name);
1900 schedule();
1901 }
755b5eb8
DT		 1902 if (error && error != -EEXIST)
1903 return error;
e7fd4179
DT		 1904
1905 if (ret_nodeid == our_nodeid) {
1906 r->res_first_lkid = 0;
1907 r->res_nodeid = 0;
1908 lkb->lkb_nodeid = 0;
1909 } else {
1910 r->res_first_lkid = lkb->lkb_id;
1911 r->res_nodeid = ret_nodeid;
1912 lkb->lkb_nodeid = ret_nodeid;
1913 }
1914 return 0;
1915}
1916
1917static void process_lookup_list(struct dlm_rsb *r)
1918{
1919 struct dlm_lkb *lkb, *safe;
1920
1921 list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
ef0c2bb0 1922 list_del_init(&lkb->lkb_rsb_lookup);
e7fd4179
DT		 1923 _request_lock(r, lkb);
1924 schedule();
1925 }
1926}
1927
1928/* confirm_master -- confirm (or deny) an rsb's master nodeid */
1929
1930static void confirm_master(struct dlm_rsb *r, int error)
1931{
1932 struct dlm_lkb *lkb;
1933
1934 if (!r->res_first_lkid)
1935 return;
1936
1937 switch (error) {
1938 case 0:
1939 case -EINPROGRESS:
1940 r->res_first_lkid = 0;
1941 process_lookup_list(r);
1942 break;
1943
1944 case -EAGAIN:
aec64e1b
DT		 1945 case -EBADR:
1946 case -ENOTBLK:
1947 /* the remote request failed and won't be retried (it was
1948 a NOQUEUE, or has been canceled/unlocked); make a waiting
1949 lkb the first_lkid */
e7fd4179
DT		 1950
1951 r->res_first_lkid = 0;
1952
1953 if (!list_empty(&r->res_lookup)) {
1954 lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
1955 lkb_rsb_lookup);
ef0c2bb0 1956 list_del_init(&lkb->lkb_rsb_lookup);
e7fd4179
DT		 1957 r->res_first_lkid = lkb->lkb_id;
1958 _request_lock(r, lkb);
1959 } else
1960 r->res_nodeid = -1;
1961 break;
1962
1963 default:
1964 log_error(r->res_ls, "confirm_master unknown error %d", error);
1965 }
1966}
1967
1968static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
d7db923e 1969 int namelen, unsigned long timeout_cs, void *ast,
3bcd3687 1970 void *astarg, void *bast, struct dlm_args *args)
e7fd4179
DT		 1971{
1972 int rv = -EINVAL;
1973
1974 /* check for invalid arg usage */
1975
1976 if (mode < 0 || mode > DLM_LOCK_EX)
1977 goto out;
1978
1979 if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
1980 goto out;
1981
1982 if (flags & DLM_LKF_CANCEL)
1983 goto out;
1984
1985 if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
1986 goto out;
1987
1988 if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
1989 goto out;
1990
1991 if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
1992 goto out;
1993
1994 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
1995 goto out;
1996
1997 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
1998 goto out;
1999
2000 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
2001 goto out;
2002
2003 if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2004 goto out;
2005
2006 if (!ast || !lksb)
2007 goto out;
2008
2009 if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2010 goto out;
2011
e7fd4179
DT		 2012 if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2013 goto out;
2014
2015 /* these args will be copied to the lkb in validate_lock_args,
2016 it cannot be done now because when converting locks, fields in
2017 an active lkb cannot be modified before locking the rsb */
2018
2019 args->flags = flags;
2020 args->astaddr = ast;
2021 args->astparam = (long) astarg;
2022 args->bastaddr = bast;
d7db923e 2023 args->timeout = timeout_cs;
e7fd4179
DT		 2024 args->mode = mode;
2025 args->lksb = lksb;
e7fd4179
DT		 2026 rv = 0;
2027 out:
2028 return rv;
2029}
2030
2031static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2032{
2033 if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2034 DLM_LKF_FORCEUNLOCK))
2035 return -EINVAL;
2036
ef0c2bb0
DT		 2037 if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2038 return -EINVAL;
2039
e7fd4179
DT		 2040 args->flags = flags;
2041 args->astparam = (long) astarg;
2042 return 0;
2043}
2044
2045static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2046 struct dlm_args *args)
2047{
2048 int rv = -EINVAL;
2049
2050 if (args->flags & DLM_LKF_CONVERT) {
2051 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2052 goto out;
2053
2054 if (args->flags & DLM_LKF_QUECVT &&
2055 !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2056 goto out;
2057
2058 rv = -EBUSY;
2059 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2060 goto out;
2061
2062 if (lkb->lkb_wait_type)
2063 goto out;
ef0c2bb0
DT		 2064
2065 if (is_overlap(lkb))
2066 goto out;
e7fd4179
DT		 2067 }
2068
2069 lkb->lkb_exflags = args->flags;
2070 lkb->lkb_sbflags = 0;
2071 lkb->lkb_astaddr = args->astaddr;
2072 lkb->lkb_astparam = args->astparam;
2073 lkb->lkb_bastaddr = args->bastaddr;
2074 lkb->lkb_rqmode = args->mode;
2075 lkb->lkb_lksb = args->lksb;
2076 lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2077 lkb->lkb_ownpid = (int) current->pid;
d7db923e 2078 lkb->lkb_timeout_cs = args->timeout;
e7fd4179
DT		 2079 rv = 0;
2080 out:
2081 return rv;
2082}
2083
ef0c2bb0
DT		 2084/* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2085 for success */
2086
2087/* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2088 because there may be a lookup in progress and it's valid to do
2089 cancel/unlockf on it */
2090
e7fd4179
DT		 2091static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2092{
ef0c2bb0 2093 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
e7fd4179
DT		 2094 int rv = -EINVAL;
2095
ef0c2bb0
DT		 2096 if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2097 log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2098 dlm_print_lkb(lkb);
e7fd4179 2099 goto out;
ef0c2bb0 2100 }
e7fd4179 2101
ef0c2bb0
DT		 2102 /* an lkb may still exist even though the lock is EOL'ed due to a
2103 cancel, unlock or failed noqueue request; an app can't use these
2104 locks; return same error as if the lkid had not been found at all */
e7fd4179 2105
ef0c2bb0
DT		 2106 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2107 log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2108 rv = -ENOENT;
e7fd4179 2109 goto out;
ef0c2bb0 2110 }
e7fd4179 2111
ef0c2bb0
DT		 2112 /* an lkb may be waiting for an rsb lookup to complete where the
2113 lookup was initiated by another lock */
2114
42dc1601
DT		 2115 if (!list_empty(&lkb->lkb_rsb_lookup)) {
2116 if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
ef0c2bb0
DT		 2117 log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2118 list_del_init(&lkb->lkb_rsb_lookup);
2119 queue_cast(lkb->lkb_resource, lkb,
2120 args->flags & DLM_LKF_CANCEL ?
2121 -DLM_ECANCEL : -DLM_EUNLOCK);
2122 unhold_lkb(lkb); /* undoes create_lkb() */
ef0c2bb0 2123 }
42dc1601
DT		 2124 /* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2125 rv = -EBUSY;
2126 goto out;
ef0c2bb0
DT		 2127 }
2128
2129 /* cancel not allowed with another cancel/unlock in progress */
2130
2131 if (args->flags & DLM_LKF_CANCEL) {
2132 if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2133 goto out;
2134
2135 if (is_overlap(lkb))
2136 goto out;
2137
3ae1acf9
DT		 2138 /* don't let scand try to do a cancel */
2139 del_timeout(lkb);
2140
ef0c2bb0
DT		 2141 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2142 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2143 rv = -EBUSY;
2144 goto out;
2145 }
2146
2147 switch (lkb->lkb_wait_type) {
2148 case DLM_MSG_LOOKUP:
2149 case DLM_MSG_REQUEST:
2150 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2151 rv = -EBUSY;
2152 goto out;
2153 case DLM_MSG_UNLOCK:
2154 case DLM_MSG_CANCEL:
2155 goto out;
2156 }
2157 /* add_to_waiters() will set OVERLAP_CANCEL */
2158 goto out_ok;
2159 }
2160
2161 /* do we need to allow a force-unlock if there's a normal unlock
2162 already in progress? in what conditions could the normal unlock
2163 fail such that we'd want to send a force-unlock to be sure? */
2164
2165 if (args->flags & DLM_LKF_FORCEUNLOCK) {
2166 if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2167 goto out;
2168
2169 if (is_overlap_unlock(lkb))
2170 goto out;
e7fd4179 2171
3ae1acf9
DT		 2172 /* don't let scand try to do a cancel */
2173 del_timeout(lkb);
2174
ef0c2bb0
DT		 2175 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2176 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2177 rv = -EBUSY;
2178 goto out;
2179 }
2180
2181 switch (lkb->lkb_wait_type) {
2182 case DLM_MSG_LOOKUP:
2183 case DLM_MSG_REQUEST:
2184 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2185 rv = -EBUSY;
2186 goto out;
2187 case DLM_MSG_UNLOCK:
2188 goto out;
2189 }
2190 /* add_to_waiters() will set OVERLAP_UNLOCK */
2191 goto out_ok;
2192 }
2193
2194 /* normal unlock not allowed if there's any op in progress */
e7fd4179 2195 rv = -EBUSY;
ef0c2bb0 2196 if (lkb->lkb_wait_type || lkb->lkb_wait_count)
e7fd4179
DT		 2197 goto out;
2198
2199 out_ok:
ef0c2bb0
DT		 2200 /* an overlapping op shouldn't blow away exflags from other op */
2201 lkb->lkb_exflags |= args->flags;
e7fd4179
DT		 2202 lkb->lkb_sbflags = 0;
2203 lkb->lkb_astparam = args->astparam;
e7fd4179
DT		 2204 rv = 0;
2205 out:
ef0c2bb0
DT		 2206 if (rv)
2207 log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
2208 lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
2209 args->flags, lkb->lkb_wait_type,
2210 lkb->lkb_resource->res_name);
e7fd4179
DT		 2211 return rv;
2212}
2213
2214/*
2215 * Four stage 4 varieties:
2216 * do_request(), do_convert(), do_unlock(), do_cancel()
2217 * These are called on the master node for the given lock and
2218 * from the central locking logic.
2219 */
2220
2221static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2222{
2223 int error = 0;
2224
c85d65e9 2225 if (can_be_granted(r, lkb, 1, NULL)) {
e7fd4179
DT		 2226 grant_lock(r, lkb);
2227 queue_cast(r, lkb, 0);
2228 goto out;
2229 }
2230
2231 if (can_be_queued(lkb)) {
2232 error = -EINPROGRESS;
2233 add_lkb(r, lkb, DLM_LKSTS_WAITING);
2234 send_blocking_asts(r, lkb);
3ae1acf9 2235 add_timeout(lkb);
e7fd4179
DT		 2236 goto out;
2237 }
2238
2239 error = -EAGAIN;
2240 if (force_blocking_asts(lkb))
2241 send_blocking_asts_all(r, lkb);
2242 queue_cast(r, lkb, -EAGAIN);
2243
2244 out:
2245 return error;
2246}
2247
2248static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2249{
2250 int error = 0;
c85d65e9 2251 int deadlk = 0;
e7fd4179
DT		 2252
2253 /* changing an existing lock may allow others to be granted */
2254
c85d65e9 2255 if (can_be_granted(r, lkb, 1, &deadlk)) {
e7fd4179
DT		 2256 grant_lock(r, lkb);
2257 queue_cast(r, lkb, 0);
2258 grant_pending_locks(r);
2259 goto out;
2260 }
2261
c85d65e9
DT		 2262 /* can_be_granted() detected that this lock would block in a conversion
2263 deadlock, so we leave it on the granted queue and return EDEADLK in
2264 the ast for the convert. */
2265
2266 if (deadlk) {
2267 /* it's left on the granted queue */
2268 log_debug(r->res_ls, "deadlock %x node %d sts%d g%d r%d %s",
2269 lkb->lkb_id, lkb->lkb_nodeid, lkb->lkb_status,
2270 lkb->lkb_grmode, lkb->lkb_rqmode, r->res_name);
2271 revert_lock(r, lkb);
2272 queue_cast(r, lkb, -EDEADLK);
2273 error = -EDEADLK;
2274 goto out;
2275 }
2276
7d3c1feb
DT		 2277 /* is_demoted() means the can_be_granted() above set the grmode
2278 to NL, and left us on the granted queue. This auto-demotion
2279 (due to CONVDEADLK) might mean other locks, and/or this lock, are
2280 now grantable. We have to try to grant other converting locks
2281 before we try again to grant this one. */
2282
2283 if (is_demoted(lkb)) {
36509258 2284 grant_pending_convert(r, DLM_LOCK_IV, NULL);
7d3c1feb
DT		 2285 if (_can_be_granted(r, lkb, 1)) {
2286 grant_lock(r, lkb);
2287 queue_cast(r, lkb, 0);
e7fd4179 2288 grant_pending_locks(r);
7d3c1feb
DT		 2289 goto out;
2290 }
2291 /* else fall through and move to convert queue */
2292 }
2293
2294 if (can_be_queued(lkb)) {
e7fd4179
DT		 2295 error = -EINPROGRESS;
2296 del_lkb(r, lkb);
2297 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2298 send_blocking_asts(r, lkb);
3ae1acf9 2299 add_timeout(lkb);
e7fd4179
DT		 2300 goto out;
2301 }
2302
2303 error = -EAGAIN;
2304 if (force_blocking_asts(lkb))
2305 send_blocking_asts_all(r, lkb);
2306 queue_cast(r, lkb, -EAGAIN);
2307
2308 out:
2309 return error;
2310}
2311
2312static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2313{
2314 remove_lock(r, lkb);
2315 queue_cast(r, lkb, -DLM_EUNLOCK);
2316 grant_pending_locks(r);
2317 return -DLM_EUNLOCK;
2318}
2319
ef0c2bb0 2320/* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
907b9bce 2321
e7fd4179
DT		 2322static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2323{
ef0c2bb0
DT		 2324 int error;
2325
2326 error = revert_lock(r, lkb);
2327 if (error) {
2328 queue_cast(r, lkb, -DLM_ECANCEL);
2329 grant_pending_locks(r);
2330 return -DLM_ECANCEL;
2331 }
2332 return 0;
e7fd4179
DT		 2333}
2334
2335/*
2336 * Four stage 3 varieties:
2337 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2338 */
2339
2340/* add a new lkb to a possibly new rsb, called by requesting process */
2341
2342static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2343{
2344 int error;
2345
2346 /* set_master: sets lkb nodeid from r */
2347
2348 error = set_master(r, lkb);
2349 if (error < 0)
2350 goto out;
2351 if (error) {
2352 error = 0;
2353 goto out;
2354 }
2355
2356 if (is_remote(r))
2357 /* receive_request() calls do_request() on remote node */
2358 error = send_request(r, lkb);
2359 else
2360 error = do_request(r, lkb);
2361 out:
2362 return error;
2363}
2364
3bcd3687 2365/* change some property of an existing lkb, e.g. mode */
e7fd4179
DT		 2366
2367static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2368{
2369 int error;
2370
2371 if (is_remote(r))
2372 /* receive_convert() calls do_convert() on remote node */
2373 error = send_convert(r, lkb);
2374 else
2375 error = do_convert(r, lkb);
2376
2377 return error;
2378}
2379
2380/* remove an existing lkb from the granted queue */
2381
2382static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2383{
2384 int error;
2385
2386 if (is_remote(r))
2387 /* receive_unlock() calls do_unlock() on remote node */
2388 error = send_unlock(r, lkb);
2389 else
2390 error = do_unlock(r, lkb);
2391
2392 return error;
2393}
2394
2395/* remove an existing lkb from the convert or wait queue */
2396
2397static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2398{
2399 int error;
2400
2401 if (is_remote(r))
2402 /* receive_cancel() calls do_cancel() on remote node */
2403 error = send_cancel(r, lkb);
2404 else
2405 error = do_cancel(r, lkb);
2406
2407 return error;
2408}
2409
2410/*
2411 * Four stage 2 varieties:
2412 * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2413 */
2414
2415static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
2416 int len, struct dlm_args *args)
2417{
2418 struct dlm_rsb *r;
2419 int error;
2420
2421 error = validate_lock_args(ls, lkb, args);
2422 if (error)
2423 goto out;
2424
2425 error = find_rsb(ls, name, len, R_CREATE, &r);
2426 if (error)
2427 goto out;
2428
2429 lock_rsb(r);
2430
2431 attach_lkb(r, lkb);
2432 lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
2433
2434 error = _request_lock(r, lkb);
2435
2436 unlock_rsb(r);
2437 put_rsb(r);
2438
2439 out:
2440 return error;
2441}
2442
2443static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2444 struct dlm_args *args)
2445{
2446 struct dlm_rsb *r;
2447 int error;
2448
2449 r = lkb->lkb_resource;
2450
2451 hold_rsb(r);
2452 lock_rsb(r);
2453
2454 error = validate_lock_args(ls, lkb, args);
2455 if (error)
2456 goto out;
2457
2458 error = _convert_lock(r, lkb);
2459 out:
2460 unlock_rsb(r);
2461 put_rsb(r);
2462 return error;
2463}
2464
2465static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2466 struct dlm_args *args)
2467{
2468 struct dlm_rsb *r;
2469 int error;
2470
2471 r = lkb->lkb_resource;
2472
2473 hold_rsb(r);
2474 lock_rsb(r);
2475
2476 error = validate_unlock_args(lkb, args);
2477 if (error)
2478 goto out;
2479
2480 error = _unlock_lock(r, lkb);
2481 out:
2482 unlock_rsb(r);
2483 put_rsb(r);
2484 return error;
2485}
2486
2487static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2488 struct dlm_args *args)
2489{
2490 struct dlm_rsb *r;
2491 int error;
2492
2493 r = lkb->lkb_resource;
2494
2495 hold_rsb(r);
2496 lock_rsb(r);
2497
2498 error = validate_unlock_args(lkb, args);
2499 if (error)
2500 goto out;
2501
2502 error = _cancel_lock(r, lkb);
2503 out:
2504 unlock_rsb(r);
2505 put_rsb(r);
2506 return error;
2507}
2508
2509/*
2510 * Two stage 1 varieties: dlm_lock() and dlm_unlock()
2511 */
2512
2513int dlm_lock(dlm_lockspace_t *lockspace,
2514 int mode,
2515 struct dlm_lksb *lksb,
2516 uint32_t flags,
2517 void *name,
2518 unsigned int namelen,
2519 uint32_t parent_lkid,
2520 void (*ast) (void *astarg),
2521 void *astarg,
3bcd3687 2522 void (*bast) (void *astarg, int mode))
e7fd4179
DT		 2523{
2524 struct dlm_ls *ls;
2525 struct dlm_lkb *lkb;
2526 struct dlm_args args;
2527 int error, convert = flags & DLM_LKF_CONVERT;
2528
2529 ls = dlm_find_lockspace_local(lockspace);
2530 if (!ls)
2531 return -EINVAL;
2532
85e86edf 2533 dlm_lock_recovery(ls);
e7fd4179
DT		 2534
2535 if (convert)
2536 error = find_lkb(ls, lksb->sb_lkid, &lkb);
2537 else
2538 error = create_lkb(ls, &lkb);
2539
2540 if (error)
2541 goto out;
2542
d7db923e 2543 error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
3bcd3687 2544 astarg, bast, &args);
e7fd4179
DT		 2545 if (error)
2546 goto out_put;
2547
2548 if (convert)
2549 error = convert_lock(ls, lkb, &args);
2550 else
2551 error = request_lock(ls, lkb, name, namelen, &args);
2552
2553 if (error == -EINPROGRESS)
2554 error = 0;
2555 out_put:
2556 if (convert || error)
b3f58d8f 2557 __put_lkb(ls, lkb);
c85d65e9 2558 if (error == -EAGAIN || error == -EDEADLK)
e7fd4179
DT		 2559 error = 0;
2560 out:
85e86edf 2561 dlm_unlock_recovery(ls);
e7fd4179
DT		 2562 dlm_put_lockspace(ls);
2563 return error;
2564}
2565
2566int dlm_unlock(dlm_lockspace_t *lockspace,
2567 uint32_t lkid,
2568 uint32_t flags,
2569 struct dlm_lksb *lksb,
2570 void *astarg)
2571{
2572 struct dlm_ls *ls;
2573 struct dlm_lkb *lkb;
2574 struct dlm_args args;
2575 int error;
2576
2577 ls = dlm_find_lockspace_local(lockspace);
2578 if (!ls)
2579 return -EINVAL;
2580
85e86edf 2581 dlm_lock_recovery(ls);
e7fd4179
DT		 2582
2583 error = find_lkb(ls, lkid, &lkb);
2584 if (error)
2585 goto out;
2586
2587 error = set_unlock_args(flags, astarg, &args);
2588 if (error)
2589 goto out_put;
2590
2591 if (flags & DLM_LKF_CANCEL)
2592 error = cancel_lock(ls, lkb, &args);
2593 else
2594 error = unlock_lock(ls, lkb, &args);
2595
2596 if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
2597 error = 0;
ef0c2bb0
DT		 2598 if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
2599 error = 0;
e7fd4179 2600 out_put:
b3f58d8f 2601 dlm_put_lkb(lkb);
e7fd4179 2602 out:
85e86edf 2603 dlm_unlock_recovery(ls);
e7fd4179
DT		 2604 dlm_put_lockspace(ls);
2605 return error;
2606}
2607
2608/*
2609 * send/receive routines for remote operations and replies
2610 *
2611 * send_args
2612 * send_common
2613 * send_request receive_request
2614 * send_convert receive_convert
2615 * send_unlock receive_unlock
2616 * send_cancel receive_cancel
2617 * send_grant receive_grant
2618 * send_bast receive_bast
2619 * send_lookup receive_lookup
2620 * send_remove receive_remove
2621 *
2622 * send_common_reply
2623 * receive_request_reply send_request_reply
2624 * receive_convert_reply send_convert_reply
2625 * receive_unlock_reply send_unlock_reply
2626 * receive_cancel_reply send_cancel_reply
2627 * receive_lookup_reply send_lookup_reply
2628 */
2629
7e4dac33
DT		 2630static int _create_message(struct dlm_ls *ls, int mb_len,
2631 int to_nodeid, int mstype,
2632 struct dlm_message **ms_ret,
2633 struct dlm_mhandle **mh_ret)
e7fd4179
DT		 2634{
2635 struct dlm_message *ms;
2636 struct dlm_mhandle *mh;
2637 char *mb;
e7fd4179
DT		 2638
2639 /* get_buffer gives us a message handle (mh) that we need to
2640 pass into lowcomms_commit and a message buffer (mb) that we
2641 write our data into */
2642
44f487a5 2643 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
e7fd4179
DT		 2644 if (!mh)
2645 return -ENOBUFS;
2646
2647 memset(mb, 0, mb_len);
2648
2649 ms = (struct dlm_message *) mb;
2650
2651 ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
7e4dac33 2652 ms->m_header.h_lockspace = ls->ls_global_id;
e7fd4179
DT		 2653 ms->m_header.h_nodeid = dlm_our_nodeid();
2654 ms->m_header.h_length = mb_len;
2655 ms->m_header.h_cmd = DLM_MSG;
2656
2657 ms->m_type = mstype;
2658
2659 *mh_ret = mh;
2660 *ms_ret = ms;
2661 return 0;
2662}
2663
7e4dac33
DT		 2664static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
2665 int to_nodeid, int mstype,
2666 struct dlm_message **ms_ret,
2667 struct dlm_mhandle **mh_ret)
2668{
2669 int mb_len = sizeof(struct dlm_message);
2670
2671 switch (mstype) {
2672 case DLM_MSG_REQUEST:
2673 case DLM_MSG_LOOKUP:
2674 case DLM_MSG_REMOVE:
2675 mb_len += r->res_length;
2676 break;
2677 case DLM_MSG_CONVERT:
2678 case DLM_MSG_UNLOCK:
2679 case DLM_MSG_REQUEST_REPLY:
2680 case DLM_MSG_CONVERT_REPLY:
2681 case DLM_MSG_GRANT:
2682 if (lkb && lkb->lkb_lvbptr)
2683 mb_len += r->res_ls->ls_lvblen;
2684 break;
2685 }
2686
2687 return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
2688 ms_ret, mh_ret);
2689}
2690
e7fd4179
DT		 2691/* further lowcomms enhancements or alternate implementations may make
2692 the return value from this function useful at some point */
2693
2694static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
2695{
2696 dlm_message_out(ms);
2697 dlm_lowcomms_commit_buffer(mh);
2698 return 0;
2699}
2700
2701static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
2702 struct dlm_message *ms)
2703{
2704 ms->m_nodeid = lkb->lkb_nodeid;
2705 ms->m_pid = lkb->lkb_ownpid;
2706 ms->m_lkid = lkb->lkb_id;
2707 ms->m_remid = lkb->lkb_remid;
2708 ms->m_exflags = lkb->lkb_exflags;
2709 ms->m_sbflags = lkb->lkb_sbflags;
2710 ms->m_flags = lkb->lkb_flags;
2711 ms->m_lvbseq = lkb->lkb_lvbseq;
2712 ms->m_status = lkb->lkb_status;
2713 ms->m_grmode = lkb->lkb_grmode;
2714 ms->m_rqmode = lkb->lkb_rqmode;
2715 ms->m_hash = r->res_hash;
2716
2717 /* m_result and m_bastmode are set from function args,
2718 not from lkb fields */
2719
2720 if (lkb->lkb_bastaddr)
2721 ms->m_asts |= AST_BAST;
2722 if (lkb->lkb_astaddr)
2723 ms->m_asts |= AST_COMP;
2724
da49f36f
DT		 2725 /* compare with switch in create_message; send_remove() doesn't
2726 use send_args() */
e7fd4179 2727
da49f36f
DT		 2728 switch (ms->m_type) {
2729 case DLM_MSG_REQUEST:
2730 case DLM_MSG_LOOKUP:
2731 memcpy(ms->m_extra, r->res_name, r->res_length);
2732 break;
2733 case DLM_MSG_CONVERT:
2734 case DLM_MSG_UNLOCK:
2735 case DLM_MSG_REQUEST_REPLY:
2736 case DLM_MSG_CONVERT_REPLY:
2737 case DLM_MSG_GRANT:
2738 if (!lkb->lkb_lvbptr)
2739 break;
e7fd4179 2740 memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
da49f36f
DT		 2741 break;
2742 }
e7fd4179
DT		 2743}
2744
2745static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
2746{
2747 struct dlm_message *ms;
2748 struct dlm_mhandle *mh;
2749 int to_nodeid, error;
2750
ef0c2bb0
DT		 2751 error = add_to_waiters(lkb, mstype);
2752 if (error)
2753 return error;
e7fd4179
DT		 2754
2755 to_nodeid = r->res_nodeid;
2756
2757 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2758 if (error)
2759 goto fail;
2760
2761 send_args(r, lkb, ms);
2762
2763 error = send_message(mh, ms);
2764 if (error)
2765 goto fail;
2766 return 0;
2767
2768 fail:
ef0c2bb0 2769 remove_from_waiters(lkb, msg_reply_type(mstype));
e7fd4179
DT		 2770 return error;
2771}
2772
2773static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2774{
2775 return send_common(r, lkb, DLM_MSG_REQUEST);
2776}
2777
2778static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2779{
2780 int error;
2781
2782 error = send_common(r, lkb, DLM_MSG_CONVERT);
2783
2784 /* down conversions go without a reply from the master */
2785 if (!error && down_conversion(lkb)) {
ef0c2bb0
DT		 2786 remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
2787 r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
e7fd4179 2788 r->res_ls->ls_stub_ms.m_result = 0;
32f105a1 2789 r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
e7fd4179
DT		 2790 __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
2791 }
2792
2793 return error;
2794}
2795
2796/* FIXME: if this lkb is the only lock we hold on the rsb, then set
2797 MASTER_UNCERTAIN to force the next request on the rsb to confirm
2798 that the master is still correct. */
2799
2800static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2801{
2802 return send_common(r, lkb, DLM_MSG_UNLOCK);
2803}
2804
2805static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2806{
2807 return send_common(r, lkb, DLM_MSG_CANCEL);
2808}
2809
2810static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
2811{
2812 struct dlm_message *ms;
2813 struct dlm_mhandle *mh;
2814 int to_nodeid, error;
2815
2816 to_nodeid = lkb->lkb_nodeid;
2817
2818 error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
2819 if (error)
2820 goto out;
2821
2822 send_args(r, lkb, ms);
2823
2824 ms->m_result = 0;
2825
2826 error = send_message(mh, ms);
2827 out:
2828 return error;
2829}
2830
2831static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
2832{
2833 struct dlm_message *ms;
2834 struct dlm_mhandle *mh;
2835 int to_nodeid, error;
2836
2837 to_nodeid = lkb->lkb_nodeid;
2838
2839 error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
2840 if (error)
2841 goto out;
2842
2843 send_args(r, lkb, ms);
2844
2845 ms->m_bastmode = mode;
2846
2847 error = send_message(mh, ms);
2848 out:
2849 return error;
2850}
2851
2852static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
2853{
2854 struct dlm_message *ms;
2855 struct dlm_mhandle *mh;
2856 int to_nodeid, error;
2857
ef0c2bb0
DT		 2858 error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
2859 if (error)
2860 return error;
e7fd4179
DT		 2861
2862 to_nodeid = dlm_dir_nodeid(r);
2863
2864 error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
2865 if (error)
2866 goto fail;
2867
2868 send_args(r, lkb, ms);
2869
2870 error = send_message(mh, ms);
2871 if (error)
2872 goto fail;
2873 return 0;
2874
2875 fail:
ef0c2bb0 2876 remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
e7fd4179
DT		 2877 return error;
2878}
2879
2880static int send_remove(struct dlm_rsb *r)
2881{
2882 struct dlm_message *ms;
2883 struct dlm_mhandle *mh;
2884 int to_nodeid, error;
2885
2886 to_nodeid = dlm_dir_nodeid(r);
2887
2888 error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
2889 if (error)
2890 goto out;
2891
2892 memcpy(ms->m_extra, r->res_name, r->res_length);
2893 ms->m_hash = r->res_hash;
2894
2895 error = send_message(mh, ms);
2896 out:
2897 return error;
2898}
2899
2900static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
2901 int mstype, int rv)
2902{
2903 struct dlm_message *ms;
2904 struct dlm_mhandle *mh;
2905 int to_nodeid, error;
2906
2907 to_nodeid = lkb->lkb_nodeid;
2908
2909 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2910 if (error)
2911 goto out;
2912
2913 send_args(r, lkb, ms);
2914
2915 ms->m_result = rv;
2916
2917 error = send_message(mh, ms);
2918 out:
2919 return error;
2920}
2921
2922static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2923{
2924 return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
2925}
2926
2927static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2928{
2929 return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
2930}
2931
2932static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2933{
2934 return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
2935}
2936
2937static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2938{
2939 return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
2940}
2941
2942static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
2943 int ret_nodeid, int rv)
2944{
2945 struct dlm_rsb *r = &ls->ls_stub_rsb;
2946 struct dlm_message *ms;
2947 struct dlm_mhandle *mh;
2948 int error, nodeid = ms_in->m_header.h_nodeid;
2949
2950 error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
2951 if (error)
2952 goto out;
2953
2954 ms->m_lkid = ms_in->m_lkid;
2955 ms->m_result = rv;
2956 ms->m_nodeid = ret_nodeid;
2957
2958 error = send_message(mh, ms);
2959 out:
2960 return error;
2961}
2962
2963/* which args we save from a received message depends heavily on the type
2964 of message, unlike the send side where we can safely send everything about
2965 the lkb for any type of message */
2966
2967static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
2968{
2969 lkb->lkb_exflags = ms->m_exflags;
6f90a8b1 2970 lkb->lkb_sbflags = ms->m_sbflags;
e7fd4179
DT		 2971 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2972 (ms->m_flags & 0x0000FFFF);
2973}
2974
2975static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
2976{
2977 lkb->lkb_sbflags = ms->m_sbflags;
2978 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2979 (ms->m_flags & 0x0000FFFF);
2980}
2981
2982static int receive_extralen(struct dlm_message *ms)
2983{
2984 return (ms->m_header.h_length - sizeof(struct dlm_message));
2985}
2986
e7fd4179
DT		 2987static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
2988 struct dlm_message *ms)
2989{
2990 int len;
2991
2992 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
2993 if (!lkb->lkb_lvbptr)
52bda2b5 2994 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
e7fd4179
DT		 2995 if (!lkb->lkb_lvbptr)
2996 return -ENOMEM;
2997 len = receive_extralen(ms);
2998 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
2999 }
3000 return 0;
3001}
3002
3003static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3004 struct dlm_message *ms)
3005{
3006 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3007 lkb->lkb_ownpid = ms->m_pid;
3008 lkb->lkb_remid = ms->m_lkid;
3009 lkb->lkb_grmode = DLM_LOCK_IV;
3010 lkb->lkb_rqmode = ms->m_rqmode;
3011 lkb->lkb_bastaddr = (void *) (long) (ms->m_asts & AST_BAST);
3012 lkb->lkb_astaddr = (void *) (long) (ms->m_asts & AST_COMP);
3013
8d07fd50
DT		 3014 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3015 /* lkb was just created so there won't be an lvb yet */
52bda2b5 3016 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
8d07fd50
DT		 3017 if (!lkb->lkb_lvbptr)
3018 return -ENOMEM;
3019 }
e7fd4179
DT		 3020
3021 return 0;
3022}
3023
3024static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3025 struct dlm_message *ms)
3026{
e7fd4179
DT		 3027 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3028 return -EBUSY;
3029
e7fd4179
DT		 3030 if (receive_lvb(ls, lkb, ms))
3031 return -ENOMEM;
3032
3033 lkb->lkb_rqmode = ms->m_rqmode;
3034 lkb->lkb_lvbseq = ms->m_lvbseq;
3035
3036 return 0;
3037}
3038
3039static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3040 struct dlm_message *ms)
3041{
e7fd4179
DT		 3042 if (receive_lvb(ls, lkb, ms))
3043 return -ENOMEM;
3044 return 0;
3045}
3046
3047/* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3048 uses to send a reply and that the remote end uses to process the reply. */
3049
3050static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3051{
3052 struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3053 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3054 lkb->lkb_remid = ms->m_lkid;
3055}
3056
c54e04b0
DT		 3057/* This is called after the rsb is locked so that we can safely inspect
3058 fields in the lkb. */
3059
3060static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3061{
3062 int from = ms->m_header.h_nodeid;
3063 int error = 0;
3064
3065 switch (ms->m_type) {
3066 case DLM_MSG_CONVERT:
3067 case DLM_MSG_UNLOCK:
3068 case DLM_MSG_CANCEL:
3069 if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3070 error = -EINVAL;
3071 break;
3072
3073 case DLM_MSG_CONVERT_REPLY:
3074 case DLM_MSG_UNLOCK_REPLY:
3075 case DLM_MSG_CANCEL_REPLY:
3076 case DLM_MSG_GRANT:
3077 case DLM_MSG_BAST:
3078 if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3079 error = -EINVAL;
3080 break;
3081
3082 case DLM_MSG_REQUEST_REPLY:
3083 if (!is_process_copy(lkb))
3084 error = -EINVAL;
3085 else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
3086 error = -EINVAL;
3087 break;
3088
3089 default:
3090 error = -EINVAL;
3091 }
3092
3093 if (error)
3094 log_error(lkb->lkb_resource->res_ls,
3095 "ignore invalid message %d from %d %x %x %x %d",
3096 ms->m_type, from, lkb->lkb_id, lkb->lkb_remid,
3097 lkb->lkb_flags, lkb->lkb_nodeid);
3098 return error;
3099}
3100
e7fd4179
DT		 3101static void receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3102{
3103 struct dlm_lkb *lkb;
3104 struct dlm_rsb *r;
3105 int error, namelen;
3106
3107 error = create_lkb(ls, &lkb);
3108 if (error)
3109 goto fail;
3110
3111 receive_flags(lkb, ms);
3112 lkb->lkb_flags |= DLM_IFL_MSTCPY;
3113 error = receive_request_args(ls, lkb, ms);
3114 if (error) {
b3f58d8f 3115 __put_lkb(ls, lkb);
e7fd4179
DT		 3116 goto fail;
3117 }
3118
3119 namelen = receive_extralen(ms);
3120
3121 error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r);
3122 if (error) {
b3f58d8f 3123 __put_lkb(ls, lkb);
e7fd4179
DT		 3124 goto fail;
3125 }
3126
3127 lock_rsb(r);
3128
3129 attach_lkb(r, lkb);
3130 error = do_request(r, lkb);
3131 send_request_reply(r, lkb, error);
3132
3133 unlock_rsb(r);
3134 put_rsb(r);
3135
3136 if (error == -EINPROGRESS)
3137 error = 0;
3138 if (error)
b3f58d8f 3139 dlm_put_lkb(lkb);
e7fd4179
DT		 3140 return;
3141
3142 fail:
3143 setup_stub_lkb(ls, ms);
3144 send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3145}
3146
3147static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3148{
3149 struct dlm_lkb *lkb;
3150 struct dlm_rsb *r;
90135925 3151 int error, reply = 1;
e7fd4179
DT		 3152
3153 error = find_lkb(ls, ms->m_remid, &lkb);
3154 if (error)
3155 goto fail;
3156
3157 r = lkb->lkb_resource;
3158
3159 hold_rsb(r);
3160 lock_rsb(r);
3161
c54e04b0
DT		 3162 error = validate_message(lkb, ms);
3163 if (error)
3164 goto out;
3165
e7fd4179
DT		 3166 receive_flags(lkb, ms);
3167 error = receive_convert_args(ls, lkb, ms);
3168 if (error)
c54e04b0 3169 goto out_reply;
e7fd4179
DT		 3170 reply = !down_conversion(lkb);
3171
3172 error = do_convert(r, lkb);
c54e04b0 3173 out_reply:
e7fd4179
DT		 3174 if (reply)
3175 send_convert_reply(r, lkb, error);
c54e04b0 3176 out:
e7fd4179
DT		 3177 unlock_rsb(r);
3178 put_rsb(r);
b3f58d8f 3179 dlm_put_lkb(lkb);
e7fd4179
DT		 3180 return;
3181
3182 fail:
3183 setup_stub_lkb(ls, ms);
3184 send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3185}
3186
3187static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3188{
3189 struct dlm_lkb *lkb;
3190 struct dlm_rsb *r;
3191 int error;
3192
3193 error = find_lkb(ls, ms->m_remid, &lkb);
3194 if (error)
3195 goto fail;
3196
3197 r = lkb->lkb_resource;
3198
3199 hold_rsb(r);
3200 lock_rsb(r);
3201
c54e04b0
DT		 3202 error = validate_message(lkb, ms);
3203 if (error)
3204 goto out;
3205
e7fd4179
DT		 3206 receive_flags(lkb, ms);
3207 error = receive_unlock_args(ls, lkb, ms);
3208 if (error)
c54e04b0 3209 goto out_reply;
e7fd4179
DT		 3210
3211 error = do_unlock(r, lkb);
c54e04b0 3212 out_reply:
e7fd4179 3213 send_unlock_reply(r, lkb, error);
c54e04b0 3214 out:
e7fd4179
DT		 3215 unlock_rsb(r);
3216 put_rsb(r);
b3f58d8f 3217 dlm_put_lkb(lkb);
e7fd4179
DT		 3218 return;
3219
3220 fail:
3221 setup_stub_lkb(ls, ms);
3222 send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3223}
3224
3225static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
3226{
3227 struct dlm_lkb *lkb;
3228 struct dlm_rsb *r;
3229 int error;
3230
3231 error = find_lkb(ls, ms->m_remid, &lkb);
3232 if (error)
3233 goto fail;
3234
3235 receive_flags(lkb, ms);
3236
3237 r = lkb->lkb_resource;
3238
3239 hold_rsb(r);
3240 lock_rsb(r);
3241
c54e04b0
DT		 3242 error = validate_message(lkb, ms);
3243 if (error)
3244 goto out;
3245
e7fd4179
DT		 3246 error = do_cancel(r, lkb);
3247 send_cancel_reply(r, lkb, error);
c54e04b0 3248 out:
e7fd4179
DT		 3249 unlock_rsb(r);
3250 put_rsb(r);
b3f58d8f 3251 dlm_put_lkb(lkb);
e7fd4179
DT		 3252 return;
3253
3254 fail:
3255 setup_stub_lkb(ls, ms);
3256 send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3257}
3258
3259static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
3260{
3261 struct dlm_lkb *lkb;
3262 struct dlm_rsb *r;
3263 int error;
3264
3265 error = find_lkb(ls, ms->m_remid, &lkb);
3266 if (error) {
c54e04b0
DT		 3267 log_debug(ls, "receive_grant from %d no lkb %x",
3268 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT		 3269 return;
3270 }
e7fd4179
DT		 3271
3272 r = lkb->lkb_resource;
3273
3274 hold_rsb(r);
3275 lock_rsb(r);
3276
c54e04b0
DT		 3277 error = validate_message(lkb, ms);
3278 if (error)
3279 goto out;
3280
e7fd4179 3281 receive_flags_reply(lkb, ms);
7d3c1feb
DT		 3282 if (is_altmode(lkb))
3283 munge_altmode(lkb, ms);
e7fd4179
DT		 3284 grant_lock_pc(r, lkb, ms);
3285 queue_cast(r, lkb, 0);
c54e04b0 3286 out:
e7fd4179
DT		 3287 unlock_rsb(r);
3288 put_rsb(r);
b3f58d8f 3289 dlm_put_lkb(lkb);
e7fd4179
DT		 3290}
3291
3292static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
3293{
3294 struct dlm_lkb *lkb;
3295 struct dlm_rsb *r;
3296 int error;
3297
3298 error = find_lkb(ls, ms->m_remid, &lkb);
3299 if (error) {
c54e04b0
DT		 3300 log_debug(ls, "receive_bast from %d no lkb %x",
3301 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT		 3302 return;
3303 }
e7fd4179
DT		 3304
3305 r = lkb->lkb_resource;
3306
3307 hold_rsb(r);
3308 lock_rsb(r);
3309
c54e04b0
DT		 3310 error = validate_message(lkb, ms);
3311 if (error)
3312 goto out;
e7fd4179 3313
c54e04b0
DT		 3314 queue_bast(r, lkb, ms->m_bastmode);
3315 out:
e7fd4179
DT		 3316 unlock_rsb(r);
3317 put_rsb(r);
b3f58d8f 3318 dlm_put_lkb(lkb);
e7fd4179
DT		 3319}
3320
3321static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
3322{
3323 int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid;
3324
3325 from_nodeid = ms->m_header.h_nodeid;
3326 our_nodeid = dlm_our_nodeid();
3327
3328 len = receive_extralen(ms);
3329
3330 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3331 if (dir_nodeid != our_nodeid) {
3332 log_error(ls, "lookup dir_nodeid %d from %d",
3333 dir_nodeid, from_nodeid);
3334 error = -EINVAL;
3335 ret_nodeid = -1;
3336 goto out;
3337 }
3338
3339 error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid);
3340
3341 /* Optimization: we're master so treat lookup as a request */
3342 if (!error && ret_nodeid == our_nodeid) {
3343 receive_request(ls, ms);
3344 return;
3345 }
3346 out:
3347 send_lookup_reply(ls, ms, ret_nodeid, error);
3348}
3349
3350static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
3351{
3352 int len, dir_nodeid, from_nodeid;
3353
3354 from_nodeid = ms->m_header.h_nodeid;
3355
3356 len = receive_extralen(ms);
3357
3358 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3359 if (dir_nodeid != dlm_our_nodeid()) {
3360 log_error(ls, "remove dir entry dir_nodeid %d from %d",
3361 dir_nodeid, from_nodeid);
3362 return;
3363 }
3364
3365 dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
3366}
3367
8499137d
DT		 3368static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
3369{
3370 do_purge(ls, ms->m_nodeid, ms->m_pid);
3371}
3372
e7fd4179
DT		 3373static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
3374{
3375 struct dlm_lkb *lkb;
3376 struct dlm_rsb *r;
ef0c2bb0 3377 int error, mstype, result;
e7fd4179
DT		 3378
3379 error = find_lkb(ls, ms->m_remid, &lkb);
3380 if (error) {
c54e04b0
DT		 3381 log_debug(ls, "receive_request_reply from %d no lkb %x",
3382 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT		 3383 return;
3384 }
e7fd4179 3385
e7fd4179
DT		 3386 r = lkb->lkb_resource;
3387 hold_rsb(r);
3388 lock_rsb(r);
3389
c54e04b0
DT		 3390 error = validate_message(lkb, ms);
3391 if (error)
3392 goto out;
3393
ef0c2bb0
DT		 3394 mstype = lkb->lkb_wait_type;
3395 error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
3396 if (error)
3397 goto out;
3398
e7fd4179
DT		 3399 /* Optimization: the dir node was also the master, so it took our
3400 lookup as a request and sent request reply instead of lookup reply */
3401 if (mstype == DLM_MSG_LOOKUP) {
3402 r->res_nodeid = ms->m_header.h_nodeid;
3403 lkb->lkb_nodeid = r->res_nodeid;
3404 }
3405
ef0c2bb0
DT		 3406 /* this is the value returned from do_request() on the master */
3407 result = ms->m_result;
3408
3409 switch (result) {
e7fd4179 3410 case -EAGAIN:
ef0c2bb0 3411 /* request would block (be queued) on remote master */
e7fd4179
DT		 3412 queue_cast(r, lkb, -EAGAIN);
3413 confirm_master(r, -EAGAIN);
ef0c2bb0 3414 unhold_lkb(lkb); /* undoes create_lkb() */
e7fd4179
DT		 3415 break;
3416
3417 case -EINPROGRESS:
3418 case 0:
3419 /* request was queued or granted on remote master */
3420 receive_flags_reply(lkb, ms);
3421 lkb->lkb_remid = ms->m_lkid;
7d3c1feb
DT		 3422 if (is_altmode(lkb))
3423 munge_altmode(lkb, ms);
3ae1acf9 3424 if (result) {
e7fd4179 3425 add_lkb(r, lkb, DLM_LKSTS_WAITING);
3ae1acf9
DT		 3426 add_timeout(lkb);
3427 } else {
e7fd4179
DT		 3428 grant_lock_pc(r, lkb, ms);
3429 queue_cast(r, lkb, 0);
3430 }
ef0c2bb0 3431 confirm_master(r, result);
e7fd4179
DT		 3432 break;
3433
597d0cae 3434 case -EBADR:
e7fd4179
DT		 3435 case -ENOTBLK:
3436 /* find_rsb failed to find rsb or rsb wasn't master */
ef0c2bb0
DT		 3437 log_debug(ls, "receive_request_reply %x %x master diff %d %d",
3438 lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
e7fd4179
DT		 3439 r->res_nodeid = -1;
3440 lkb->lkb_nodeid = -1;
ef0c2bb0
DT		 3441
3442 if (is_overlap(lkb)) {
3443 /* we'll ignore error in cancel/unlock reply */
3444 queue_cast_overlap(r, lkb);
aec64e1b 3445 confirm_master(r, result);
ef0c2bb0
DT		 3446 unhold_lkb(lkb); /* undoes create_lkb() */
3447 } else
3448 _request_lock(r, lkb);
e7fd4179
DT		 3449 break;
3450
3451 default:
ef0c2bb0
DT		 3452 log_error(ls, "receive_request_reply %x error %d",
3453 lkb->lkb_id, result);
e7fd4179
DT		 3454 }
3455
ef0c2bb0
DT		 3456 if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
3457 log_debug(ls, "receive_request_reply %x result %d unlock",
3458 lkb->lkb_id, result);
3459 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3460 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3461 send_unlock(r, lkb);
3462 } else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
3463 log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
3464 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3465 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3466 send_cancel(r, lkb);
3467 } else {
3468 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3469 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3470 }
3471 out:
e7fd4179
DT		 3472 unlock_rsb(r);
3473 put_rsb(r);
b3f58d8f 3474 dlm_put_lkb(lkb);
e7fd4179
DT		 3475}
3476
3477static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3478 struct dlm_message *ms)
3479{
e7fd4179 3480 /* this is the value returned from do_convert() on the master */
ef0c2bb0 3481 switch (ms->m_result) {
e7fd4179
DT		 3482 case -EAGAIN:
3483 /* convert would block (be queued) on remote master */
3484 queue_cast(r, lkb, -EAGAIN);
3485 break;
3486
c85d65e9
DT		 3487 case -EDEADLK:
3488 receive_flags_reply(lkb, ms);
3489 revert_lock_pc(r, lkb);
3490 queue_cast(r, lkb, -EDEADLK);
3491 break;
3492
e7fd4179
DT		 3493 case -EINPROGRESS:
3494 /* convert was queued on remote master */
7d3c1feb
DT		 3495 receive_flags_reply(lkb, ms);
3496 if (is_demoted(lkb))
3497 munge_demoted(lkb, ms);
e7fd4179
DT		 3498 del_lkb(r, lkb);
3499 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3ae1acf9 3500 add_timeout(lkb);
e7fd4179
DT		 3501 break;
3502
3503 case 0:
3504 /* convert was granted on remote master */
3505 receive_flags_reply(lkb, ms);
7d3c1feb
DT		 3506 if (is_demoted(lkb))
3507 munge_demoted(lkb, ms);
e7fd4179
DT		 3508 grant_lock_pc(r, lkb, ms);
3509 queue_cast(r, lkb, 0);
3510 break;
3511
3512 default:
ef0c2bb0
DT		 3513 log_error(r->res_ls, "receive_convert_reply %x error %d",
3514 lkb->lkb_id, ms->m_result);
e7fd4179
DT		 3515 }
3516}
3517
3518static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3519{
3520 struct dlm_rsb *r = lkb->lkb_resource;
ef0c2bb0 3521 int error;
e7fd4179
DT		 3522
3523 hold_rsb(r);
3524 lock_rsb(r);
3525
c54e04b0
DT		 3526 error = validate_message(lkb, ms);
3527 if (error)
3528 goto out;
3529
ef0c2bb0
DT		 3530 /* stub reply can happen with waiters_mutex held */
3531 error = remove_from_waiters_ms(lkb, ms);
3532 if (error)
3533 goto out;
e7fd4179 3534
ef0c2bb0
DT		 3535 __receive_convert_reply(r, lkb, ms);
3536 out:
e7fd4179
DT		 3537 unlock_rsb(r);
3538 put_rsb(r);
3539}
3540
3541static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
3542{
3543 struct dlm_lkb *lkb;
3544 int error;
3545
3546 error = find_lkb(ls, ms->m_remid, &lkb);
3547 if (error) {
c54e04b0
DT		 3548 log_debug(ls, "receive_convert_reply from %d no lkb %x",
3549 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT		 3550 return;
3551 }
e7fd4179 3552
e7fd4179 3553 _receive_convert_reply(lkb, ms);
b3f58d8f 3554 dlm_put_lkb(lkb);
e7fd4179
DT		 3555}
3556
3557static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3558{
3559 struct dlm_rsb *r = lkb->lkb_resource;
ef0c2bb0 3560 int error;
e7fd4179
DT		 3561
3562 hold_rsb(r);
3563 lock_rsb(r);
3564
c54e04b0
DT		 3565 error = validate_message(lkb, ms);
3566 if (error)
3567 goto out;
3568
ef0c2bb0
DT		 3569 /* stub reply can happen with waiters_mutex held */
3570 error = remove_from_waiters_ms(lkb, ms);
3571 if (error)
3572 goto out;
3573
e7fd4179
DT		 3574 /* this is the value returned from do_unlock() on the master */
3575
ef0c2bb0 3576 switch (ms->m_result) {
e7fd4179
DT		 3577 case -DLM_EUNLOCK:
3578 receive_flags_reply(lkb, ms);
3579 remove_lock_pc(r, lkb);
3580 queue_cast(r, lkb, -DLM_EUNLOCK);
3581 break;
ef0c2bb0
DT		 3582 case -ENOENT:
3583 break;
e7fd4179 3584 default:
ef0c2bb0
DT		 3585 log_error(r->res_ls, "receive_unlock_reply %x error %d",
3586 lkb->lkb_id, ms->m_result);
e7fd4179 3587 }
ef0c2bb0 3588 out:
e7fd4179
DT		 3589 unlock_rsb(r);
3590 put_rsb(r);
3591}
3592
3593static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
3594{
3595 struct dlm_lkb *lkb;
3596 int error;
3597
3598 error = find_lkb(ls, ms->m_remid, &lkb);
3599 if (error) {
c54e04b0
DT		 3600 log_debug(ls, "receive_unlock_reply from %d no lkb %x",
3601 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT		 3602 return;
3603 }
e7fd4179 3604
e7fd4179 3605 _receive_unlock_reply(lkb, ms);
b3f58d8f 3606 dlm_put_lkb(lkb);
e7fd4179
DT		 3607}
3608
3609static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3610{
3611 struct dlm_rsb *r = lkb->lkb_resource;
ef0c2bb0 3612 int error;
e7fd4179
DT		 3613
3614 hold_rsb(r);
3615 lock_rsb(r);
3616
c54e04b0
DT		 3617 error = validate_message(lkb, ms);
3618 if (error)
3619 goto out;
3620
ef0c2bb0
DT		 3621 /* stub reply can happen with waiters_mutex held */
3622 error = remove_from_waiters_ms(lkb, ms);
3623 if (error)
3624 goto out;
3625
e7fd4179
DT		 3626 /* this is the value returned from do_cancel() on the master */
3627
ef0c2bb0 3628 switch (ms->m_result) {
e7fd4179
DT		 3629 case -DLM_ECANCEL:
3630 receive_flags_reply(lkb, ms);
3631 revert_lock_pc(r, lkb);
84d8cd69 3632 queue_cast(r, lkb, -DLM_ECANCEL);
ef0c2bb0
DT		 3633 break;
3634 case 0:
e7fd4179
DT		 3635 break;
3636 default:
ef0c2bb0
DT		 3637 log_error(r->res_ls, "receive_cancel_reply %x error %d",
3638 lkb->lkb_id, ms->m_result);
e7fd4179 3639 }
ef0c2bb0 3640 out:
e7fd4179
DT		 3641 unlock_rsb(r);
3642 put_rsb(r);
3643}
3644
3645static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
3646{
3647 struct dlm_lkb *lkb;
3648 int error;
3649
3650 error = find_lkb(ls, ms->m_remid, &lkb);
3651 if (error) {
c54e04b0
DT		 3652 log_debug(ls, "receive_cancel_reply from %d no lkb %x",
3653 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT		 3654 return;
3655 }
e7fd4179 3656
e7fd4179 3657 _receive_cancel_reply(lkb, ms);
b3f58d8f 3658 dlm_put_lkb(lkb);
e7fd4179
DT		 3659}
3660
3661static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
3662{
3663 struct dlm_lkb *lkb;
3664 struct dlm_rsb *r;
3665 int error, ret_nodeid;
3666
3667 error = find_lkb(ls, ms->m_lkid, &lkb);
3668 if (error) {
3669 log_error(ls, "receive_lookup_reply no lkb");
3670 return;
3671 }
3672
ef0c2bb0 3673 /* ms->m_result is the value returned by dlm_dir_lookup on dir node
e7fd4179 3674 FIXME: will a non-zero error ever be returned? */
e7fd4179
DT		 3675
3676 r = lkb->lkb_resource;
3677 hold_rsb(r);
3678 lock_rsb(r);
3679
ef0c2bb0
DT		 3680 error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3681 if (error)
3682 goto out;
3683
e7fd4179
DT		 3684 ret_nodeid = ms->m_nodeid;
3685 if (ret_nodeid == dlm_our_nodeid()) {
3686 r->res_nodeid = 0;
3687 ret_nodeid = 0;
3688 r->res_first_lkid = 0;
3689 } else {
3690 /* set_master() will copy res_nodeid to lkb_nodeid */
3691 r->res_nodeid = ret_nodeid;
3692 }
3693
ef0c2bb0
DT		 3694 if (is_overlap(lkb)) {
3695 log_debug(ls, "receive_lookup_reply %x unlock %x",
3696 lkb->lkb_id, lkb->lkb_flags);
3697 queue_cast_overlap(r, lkb);
3698 unhold_lkb(lkb); /* undoes create_lkb() */
3699 goto out_list;
3700 }
3701
e7fd4179
DT		 3702 _request_lock(r, lkb);
3703
ef0c2bb0 3704 out_list:
e7fd4179
DT		 3705 if (!ret_nodeid)
3706 process_lookup_list(r);
ef0c2bb0 3707 out:
e7fd4179
DT		 3708 unlock_rsb(r);
3709 put_rsb(r);
b3f58d8f 3710 dlm_put_lkb(lkb);
e7fd4179
DT		 3711}
3712
c36258b5 3713static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
e7fd4179 3714{
46b43eed
DT		 3715 if (!dlm_is_member(ls, ms->m_header.h_nodeid)) {
3716 log_debug(ls, "ignore non-member message %d from %d %x %x %d",
3717 ms->m_type, ms->m_header.h_nodeid, ms->m_lkid,
3718 ms->m_remid, ms->m_result);
3719 return;
3720 }
3721
e7fd4179
DT		 3722 switch (ms->m_type) {
3723
3724 /* messages sent to a master node */
3725
3726 case DLM_MSG_REQUEST:
3727 receive_request(ls, ms);
3728 break;
3729
3730 case DLM_MSG_CONVERT:
3731 receive_convert(ls, ms);
3732 break;
3733
3734 case DLM_MSG_UNLOCK:
3735 receive_unlock(ls, ms);
3736 break;
3737
3738 case DLM_MSG_CANCEL:
3739 receive_cancel(ls, ms);
3740 break;
3741
3742 /* messages sent from a master node (replies to above) */
3743
3744 case DLM_MSG_REQUEST_REPLY:
3745 receive_request_reply(ls, ms);
3746 break;
3747
3748 case DLM_MSG_CONVERT_REPLY:
3749 receive_convert_reply(ls, ms);
3750 break;
3751
3752 case DLM_MSG_UNLOCK_REPLY:
3753 receive_unlock_reply(ls, ms);
3754 break;
3755
3756 case DLM_MSG_CANCEL_REPLY:
3757 receive_cancel_reply(ls, ms);
3758 break;
3759
3760 /* messages sent from a master node (only two types of async msg) */
3761
3762 case DLM_MSG_GRANT:
3763 receive_grant(ls, ms);
3764 break;
3765
3766 case DLM_MSG_BAST:
3767 receive_bast(ls, ms);
3768 break;
3769
3770 /* messages sent to a dir node */
3771
3772 case DLM_MSG_LOOKUP:
3773 receive_lookup(ls, ms);
3774 break;
3775
3776 case DLM_MSG_REMOVE:
3777 receive_remove(ls, ms);
3778 break;
3779
3780 /* messages sent from a dir node (remove has no reply) */
3781
3782 case DLM_MSG_LOOKUP_REPLY:
3783 receive_lookup_reply(ls, ms);
3784 break;
3785
8499137d
DT		 3786 /* other messages */
3787
3788 case DLM_MSG_PURGE:
3789 receive_purge(ls, ms);
3790 break;
3791
e7fd4179
DT		 3792 default:
3793 log_error(ls, "unknown message type %d", ms->m_type);
3794 }
3795
e7fd4179 3796 dlm_astd_wake();
e7fd4179
DT		 3797}
3798
c36258b5
DT		 3799/* If the lockspace is in recovery mode (locking stopped), then normal
3800 messages are saved on the requestqueue for processing after recovery is
3801 done. When not in recovery mode, we wait for dlm_recoverd to drain saved
3802 messages off the requestqueue before we process new ones. This occurs right
3803 after recovery completes when we transition from saving all messages on
3804 requestqueue, to processing all the saved messages, to processing new
3805 messages as they arrive. */
e7fd4179 3806
c36258b5
DT		 3807static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
3808 int nodeid)
3809{
3810 if (dlm_locking_stopped(ls)) {
3811 dlm_add_requestqueue(ls, nodeid, (struct dlm_header *) ms);
3812 } else {
3813 dlm_wait_requestqueue(ls);
3814 _receive_message(ls, ms);
3815 }
3816}
3817
3818/* This is called by dlm_recoverd to process messages that were saved on
3819 the requestqueue. */
3820
3821void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
3822{
3823 _receive_message(ls, ms);
3824}
3825
3826/* This is called by the midcomms layer when something is received for
3827 the lockspace. It could be either a MSG (normal message sent as part of
3828 standard locking activity) or an RCOM (recovery message sent as part of
3829 lockspace recovery). */
3830
3831void dlm_receive_buffer(struct dlm_header *hd, int nodeid)
3832{
3833 struct dlm_message *ms = (struct dlm_message *) hd;
3834 struct dlm_rcom *rc = (struct dlm_rcom *) hd;
3835 struct dlm_ls *ls;
3836 int type = 0;
3837
3838 switch (hd->h_cmd) {
3839 case DLM_MSG:
3840 dlm_message_in(ms);
3841 type = ms->m_type;
3842 break;
3843 case DLM_RCOM:
3844 dlm_rcom_in(rc);
3845 type = rc->rc_type;
3846 break;
3847 default:
3848 log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
3849 return;
3850 }
3851
3852 if (hd->h_nodeid != nodeid) {
3853 log_print("invalid h_nodeid %d from %d lockspace %x",
3854 hd->h_nodeid, nodeid, hd->h_lockspace);
3855 return;
3856 }
3857
3858 ls = dlm_find_lockspace_global(hd->h_lockspace);
3859 if (!ls) {
594199eb
DT		 3860 if (dlm_config.ci_log_debug)
3861 log_print("invalid lockspace %x from %d cmd %d type %d",
3862 hd->h_lockspace, nodeid, hd->h_cmd, type);
c36258b5
DT		 3863
3864 if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
3865 dlm_send_ls_not_ready(nodeid, rc);
3866 return;
3867 }
3868
3869 /* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3870 be inactive (in this ls) before transitioning to recovery mode */
3871
3872 down_read(&ls->ls_recv_active);
3873 if (hd->h_cmd == DLM_MSG)
3874 dlm_receive_message(ls, ms, nodeid);
3875 else
3876 dlm_receive_rcom(ls, rc, nodeid);
3877 up_read(&ls->ls_recv_active);
3878
3879 dlm_put_lockspace(ls);
3880}
e7fd4179
DT		 3881
3882static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
3883{
3884 if (middle_conversion(lkb)) {
3885 hold_lkb(lkb);
ef0c2bb0 3886 ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
e7fd4179 3887 ls->ls_stub_ms.m_result = -EINPROGRESS;
075529b5 3888 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
c54e04b0 3889 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
e7fd4179
DT		 3890 _receive_convert_reply(lkb, &ls->ls_stub_ms);
3891
3892 /* Same special case as in receive_rcom_lock_args() */
3893 lkb->lkb_grmode = DLM_LOCK_IV;
3894 rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
3895 unhold_lkb(lkb);
3896
3897 } else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
3898 lkb->lkb_flags |= DLM_IFL_RESEND;
3899 }
3900
3901 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
3902 conversions are async; there's no reply from the remote master */
3903}
3904
3905/* A waiting lkb needs recovery if the master node has failed, or
3906 the master node is changing (only when no directory is used) */
3907
3908static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
3909{
3910 if (dlm_is_removed(ls, lkb->lkb_nodeid))
3911 return 1;
3912
3913 if (!dlm_no_directory(ls))
3914 return 0;
3915
3916 if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid)
3917 return 1;
3918
3919 return 0;
3920}
3921
3922/* Recovery for locks that are waiting for replies from nodes that are now
3923 gone. We can just complete unlocks and cancels by faking a reply from the
3924 dead node. Requests and up-conversions we flag to be resent after
3925 recovery. Down-conversions can just be completed with a fake reply like
3926 unlocks. Conversions between PR and CW need special attention. */
3927
3928void dlm_recover_waiters_pre(struct dlm_ls *ls)
3929{
3930 struct dlm_lkb *lkb, *safe;
601342ce 3931 int wait_type, stub_unlock_result, stub_cancel_result;
e7fd4179 3932
90135925 3933 mutex_lock(&ls->ls_waiters_mutex);
e7fd4179
DT		 3934
3935 list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
3936 log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
3937 lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
3938
3939 /* all outstanding lookups, regardless of destination will be
3940 resent after recovery is done */
3941
3942 if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
3943 lkb->lkb_flags |= DLM_IFL_RESEND;
3944 continue;
3945 }
3946
3947 if (!waiter_needs_recovery(ls, lkb))
3948 continue;
3949
601342ce
DT		 3950 wait_type = lkb->lkb_wait_type;
3951 stub_unlock_result = -DLM_EUNLOCK;
3952 stub_cancel_result = -DLM_ECANCEL;
3953
3954 /* Main reply may have been received leaving a zero wait_type,
3955 but a reply for the overlapping op may not have been
3956 received. In that case we need to fake the appropriate
3957 reply for the overlap op. */
3958
3959 if (!wait_type) {
3960 if (is_overlap_cancel(lkb)) {
3961 wait_type = DLM_MSG_CANCEL;
3962 if (lkb->lkb_grmode == DLM_LOCK_IV)
3963 stub_cancel_result = 0;
3964 }
3965 if (is_overlap_unlock(lkb)) {
3966 wait_type = DLM_MSG_UNLOCK;
3967 if (lkb->lkb_grmode == DLM_LOCK_IV)
3968 stub_unlock_result = -ENOENT;
3969 }
3970
3971 log_debug(ls, "rwpre overlap %x %x %d %d %d",
3972 lkb->lkb_id, lkb->lkb_flags, wait_type,
3973 stub_cancel_result, stub_unlock_result);
3974 }
3975
3976 switch (wait_type) {
e7fd4179
DT		 3977
3978 case DLM_MSG_REQUEST:
3979 lkb->lkb_flags |= DLM_IFL_RESEND;
3980 break;
3981
3982 case DLM_MSG_CONVERT:
3983 recover_convert_waiter(ls, lkb);
3984 break;
3985
3986 case DLM_MSG_UNLOCK:
3987 hold_lkb(lkb);
ef0c2bb0 3988 ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
601342ce 3989 ls->ls_stub_ms.m_result = stub_unlock_result;
075529b5 3990 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
c54e04b0 3991 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
e7fd4179 3992 _receive_unlock_reply(lkb, &ls->ls_stub_ms);
b3f58d8f 3993 dlm_put_lkb(lkb);
e7fd4179
DT		 3994 break;
3995
3996 case DLM_MSG_CANCEL:
3997 hold_lkb(lkb);
ef0c2bb0 3998 ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
601342ce 3999 ls->ls_stub_ms.m_result = stub_cancel_result;
075529b5 4000 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
c54e04b0 4001 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
e7fd4179 4002 _receive_cancel_reply(lkb, &ls->ls_stub_ms);
b3f58d8f 4003 dlm_put_lkb(lkb);
e7fd4179
DT		 4004 break;
4005
4006 default:
601342ce
DT		 4007 log_error(ls, "invalid lkb wait_type %d %d",
4008 lkb->lkb_wait_type, wait_type);
e7fd4179 4009 }
81456807 4010 schedule();
e7fd4179 4011 }
90135925 4012 mutex_unlock(&ls->ls_waiters_mutex);
e7fd4179
DT		 4013}
4014
ef0c2bb0 4015static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
e7fd4179
DT		 4016{
4017 struct dlm_lkb *lkb;
ef0c2bb0 4018 int found = 0;
e7fd4179 4019
90135925 4020 mutex_lock(&ls->ls_waiters_mutex);
e7fd4179
DT		 4021 list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
4022 if (lkb->lkb_flags & DLM_IFL_RESEND) {
ef0c2bb0
DT		 4023 hold_lkb(lkb);
4024 found = 1;
e7fd4179
DT		 4025 break;
4026 }
4027 }
90135925 4028 mutex_unlock(&ls->ls_waiters_mutex);
e7fd4179 4029
ef0c2bb0 4030 if (!found)
e7fd4179 4031 lkb = NULL;
ef0c2bb0 4032 return lkb;
e7fd4179
DT		 4033}
4034
4035/* Deal with lookups and lkb's marked RESEND from _pre. We may now be the
4036 master or dir-node for r. Processing the lkb may result in it being placed
4037 back on waiters. */
4038
ef0c2bb0
DT		 4039/* We do this after normal locking has been enabled and any saved messages
4040 (in requestqueue) have been processed. We should be confident that at
4041 this point we won't get or process a reply to any of these waiting
4042 operations. But, new ops may be coming in on the rsbs/locks here from
4043 userspace or remotely. */
4044
4045/* there may have been an overlap unlock/cancel prior to recovery or after
4046 recovery. if before, the lkb may still have a pos wait_count; if after, the
4047 overlap flag would just have been set and nothing new sent. we can be
4048 confident here than any replies to either the initial op or overlap ops
4049 prior to recovery have been received. */
4050
e7fd4179
DT		 4051int dlm_recover_waiters_post(struct dlm_ls *ls)
4052{
4053 struct dlm_lkb *lkb;
4054 struct dlm_rsb *r;
ef0c2bb0 4055 int error = 0, mstype, err, oc, ou;
e7fd4179
DT		 4056
4057 while (1) {
4058 if (dlm_locking_stopped(ls)) {
4059 log_debug(ls, "recover_waiters_post aborted");
4060 error = -EINTR;
4061 break;
4062 }
4063
ef0c2bb0
DT		 4064 lkb = find_resend_waiter(ls);
4065 if (!lkb)
e7fd4179
DT		 4066 break;
4067
4068 r = lkb->lkb_resource;
ef0c2bb0
DT		 4069 hold_rsb(r);
4070 lock_rsb(r);
4071
4072 mstype = lkb->lkb_wait_type;
4073 oc = is_overlap_cancel(lkb);
4074 ou = is_overlap_unlock(lkb);
4075 err = 0;
e7fd4179
DT		 4076
4077 log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
4078 lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
4079
ef0c2bb0
DT		 4080 /* At this point we assume that we won't get a reply to any
4081 previous op or overlap op on this lock. First, do a big
4082 remove_from_waiters() for all previous ops. */
4083
4084 lkb->lkb_flags &= ~DLM_IFL_RESEND;
4085 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4086 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4087 lkb->lkb_wait_type = 0;
4088 lkb->lkb_wait_count = 0;
4089 mutex_lock(&ls->ls_waiters_mutex);
4090 list_del_init(&lkb->lkb_wait_reply);
4091 mutex_unlock(&ls->ls_waiters_mutex);
4092 unhold_lkb(lkb); /* for waiters list */
4093
4094 if (oc || ou) {
4095 /* do an unlock or cancel instead of resending */
4096 switch (mstype) {
4097 case DLM_MSG_LOOKUP:
4098 case DLM_MSG_REQUEST:
4099 queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
4100 -DLM_ECANCEL);
4101 unhold_lkb(lkb); /* undoes create_lkb() */
4102 break;
4103 case DLM_MSG_CONVERT:
4104 if (oc) {
4105 queue_cast(r, lkb, -DLM_ECANCEL);
4106 } else {
4107 lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
4108 _unlock_lock(r, lkb);
4109 }
4110 break;
4111 default:
4112 err = 1;
4113 }
4114 } else {
4115 switch (mstype) {
4116 case DLM_MSG_LOOKUP:
4117 case DLM_MSG_REQUEST:
4118 _request_lock(r, lkb);
4119 if (is_master(r))
4120 confirm_master(r, 0);
4121 break;
4122 case DLM_MSG_CONVERT:
4123 _convert_lock(r, lkb);
4124 break;
4125 default:
4126 err = 1;
4127 }
e7fd4179 4128 }
ef0c2bb0
DT		 4129
4130 if (err)
4131 log_error(ls, "recover_waiters_post %x %d %x %d %d",
4132 lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
4133 unlock_rsb(r);
4134 put_rsb(r);
4135 dlm_put_lkb(lkb);
e7fd4179
DT		 4136 }
4137
4138 return error;
4139}
4140
4141static void purge_queue(struct dlm_rsb *r, struct list_head *queue,
4142 int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb))
4143{
4144 struct dlm_ls *ls = r->res_ls;
4145 struct dlm_lkb *lkb, *safe;
4146
4147 list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) {
4148 if (test(ls, lkb)) {
97a35d1e 4149 rsb_set_flag(r, RSB_LOCKS_PURGED);
e7fd4179
DT		 4150 del_lkb(r, lkb);
4151 /* this put should free the lkb */
b3f58d8f 4152 if (!dlm_put_lkb(lkb))
e7fd4179
DT		 4153 log_error(ls, "purged lkb not released");
4154 }
4155 }
4156}
4157
4158static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4159{
4160 return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid));
4161}
4162
4163static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4164{
4165 return is_master_copy(lkb);
4166}
4167
4168static void purge_dead_locks(struct dlm_rsb *r)
4169{
4170 purge_queue(r, &r->res_grantqueue, &purge_dead_test);
4171 purge_queue(r, &r->res_convertqueue, &purge_dead_test);
4172 purge_queue(r, &r->res_waitqueue, &purge_dead_test);
4173}
4174
4175void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
4176{
4177 purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test);
4178 purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test);
4179 purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test);
4180}
4181
4182/* Get rid of locks held by nodes that are gone. */
4183
4184int dlm_purge_locks(struct dlm_ls *ls)
4185{
4186 struct dlm_rsb *r;
4187
4188 log_debug(ls, "dlm_purge_locks");
4189
4190 down_write(&ls->ls_root_sem);
4191 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
4192 hold_rsb(r);
4193 lock_rsb(r);
4194 if (is_master(r))
4195 purge_dead_locks(r);
4196 unlock_rsb(r);
4197 unhold_rsb(r);
4198
4199 schedule();
4200 }
4201 up_write(&ls->ls_root_sem);
4202
4203 return 0;
4204}
4205
97a35d1e
DT		 4206static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket)
4207{
4208 struct dlm_rsb *r, *r_ret = NULL;
4209
4210 read_lock(&ls->ls_rsbtbl[bucket].lock);
4211 list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) {
4212 if (!rsb_flag(r, RSB_LOCKS_PURGED))
4213 continue;
4214 hold_rsb(r);
4215 rsb_clear_flag(r, RSB_LOCKS_PURGED);
4216 r_ret = r;
4217 break;
4218 }
4219 read_unlock(&ls->ls_rsbtbl[bucket].lock);
4220 return r_ret;
4221}
4222
4223void dlm_grant_after_purge(struct dlm_ls *ls)
e7fd4179
DT		 4224{
4225 struct dlm_rsb *r;
2b4e926a 4226 int bucket = 0;
e7fd4179 4227
2b4e926a
DT		 4228 while (1) {
4229 r = find_purged_rsb(ls, bucket);
4230 if (!r) {
4231 if (bucket == ls->ls_rsbtbl_size - 1)
4232 break;
4233 bucket++;
97a35d1e 4234 continue;
2b4e926a 4235 }
97a35d1e
DT		 4236 lock_rsb(r);
4237 if (is_master(r)) {
4238 grant_pending_locks(r);
4239 confirm_master(r, 0);
e7fd4179 4240 }
97a35d1e
DT		 4241 unlock_rsb(r);
4242 put_rsb(r);
2b4e926a 4243 schedule();
e7fd4179 4244 }
e7fd4179
DT		 4245}
4246
4247static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
4248 uint32_t remid)
4249{
4250 struct dlm_lkb *lkb;
4251
4252 list_for_each_entry(lkb, head, lkb_statequeue) {
4253 if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
4254 return lkb;
4255 }
4256 return NULL;
4257}
4258
4259static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
4260 uint32_t remid)
4261{
4262 struct dlm_lkb *lkb;
4263
4264 lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
4265 if (lkb)
4266 return lkb;
4267 lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
4268 if (lkb)
4269 return lkb;
4270 lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
4271 if (lkb)
4272 return lkb;
4273 return NULL;
4274}
4275
4276static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
4277 struct dlm_rsb *r, struct dlm_rcom *rc)
4278{
4279 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4280 int lvblen;
4281
4282 lkb->lkb_nodeid = rc->rc_header.h_nodeid;
4283 lkb->lkb_ownpid = rl->rl_ownpid;
4284 lkb->lkb_remid = rl->rl_lkid;
4285 lkb->lkb_exflags = rl->rl_exflags;
4286 lkb->lkb_flags = rl->rl_flags & 0x0000FFFF;
4287 lkb->lkb_flags |= DLM_IFL_MSTCPY;
4288 lkb->lkb_lvbseq = rl->rl_lvbseq;
4289 lkb->lkb_rqmode = rl->rl_rqmode;
4290 lkb->lkb_grmode = rl->rl_grmode;
4291 /* don't set lkb_status because add_lkb wants to itself */
4292
4293 lkb->lkb_bastaddr = (void *) (long) (rl->rl_asts & AST_BAST);
4294 lkb->lkb_astaddr = (void *) (long) (rl->rl_asts & AST_COMP);
4295
e7fd4179 4296 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
52bda2b5 4297 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
e7fd4179
DT		 4298 if (!lkb->lkb_lvbptr)
4299 return -ENOMEM;
4300 lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) -
4301 sizeof(struct rcom_lock);
4302 memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
4303 }
4304
4305 /* Conversions between PR and CW (middle modes) need special handling.
4306 The real granted mode of these converting locks cannot be determined
4307 until all locks have been rebuilt on the rsb (recover_conversion) */
4308
4309 if (rl->rl_wait_type == DLM_MSG_CONVERT && middle_conversion(lkb)) {
4310 rl->rl_status = DLM_LKSTS_CONVERT;
4311 lkb->lkb_grmode = DLM_LOCK_IV;
4312 rsb_set_flag(r, RSB_RECOVER_CONVERT);
4313 }
4314
4315 return 0;
4316}
4317
4318/* This lkb may have been recovered in a previous aborted recovery so we need
4319 to check if the rsb already has an lkb with the given remote nodeid/lkid.
4320 If so we just send back a standard reply. If not, we create a new lkb with
4321 the given values and send back our lkid. We send back our lkid by sending
4322 back the rcom_lock struct we got but with the remid field filled in. */
4323
4324int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4325{
4326 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4327 struct dlm_rsb *r;
4328 struct dlm_lkb *lkb;
4329 int error;
4330
4331 if (rl->rl_parent_lkid) {
4332 error = -EOPNOTSUPP;
4333 goto out;
4334 }
4335
4336 error = find_rsb(ls, rl->rl_name, rl->rl_namelen, R_MASTER, &r);
4337 if (error)
4338 goto out;
4339
4340 lock_rsb(r);
4341
4342 lkb = search_remid(r, rc->rc_header.h_nodeid, rl->rl_lkid);
4343 if (lkb) {
4344 error = -EEXIST;
4345 goto out_remid;
4346 }
4347
4348 error = create_lkb(ls, &lkb);
4349 if (error)
4350 goto out_unlock;
4351
4352 error = receive_rcom_lock_args(ls, lkb, r, rc);
4353 if (error) {
b3f58d8f 4354 __put_lkb(ls, lkb);
e7fd4179
DT		 4355 goto out_unlock;
4356 }
4357
4358 attach_lkb(r, lkb);
4359 add_lkb(r, lkb, rl->rl_status);
4360 error = 0;
4361
4362 out_remid:
4363 /* this is the new value returned to the lock holder for
4364 saving in its process-copy lkb */
4365 rl->rl_remid = lkb->lkb_id;
4366
4367 out_unlock:
4368 unlock_rsb(r);
4369 put_rsb(r);
4370 out:
4371 if (error)
11b2498b 4372 log_debug(ls, "recover_master_copy %d %x", error, rl->rl_lkid);
e7fd4179
DT		 4373 rl->rl_result = error;
4374 return error;
4375}
4376
4377int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4378{
4379 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4380 struct dlm_rsb *r;
4381 struct dlm_lkb *lkb;
4382 int error;
4383
4384 error = find_lkb(ls, rl->rl_lkid, &lkb);
4385 if (error) {
4386 log_error(ls, "recover_process_copy no lkid %x", rl->rl_lkid);
4387 return error;
4388 }
4389
4390 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
4391
4392 error = rl->rl_result;
4393
4394 r = lkb->lkb_resource;
4395 hold_rsb(r);
4396 lock_rsb(r);
4397
4398 switch (error) {
dc200a88
DT		 4399 case -EBADR:
4400 /* There's a chance the new master received our lock before
4401 dlm_recover_master_reply(), this wouldn't happen if we did
4402 a barrier between recover_masters and recover_locks. */
4403 log_debug(ls, "master copy not ready %x r %lx %s", lkb->lkb_id,
4404 (unsigned long)r, r->res_name);
4405 dlm_send_rcom_lock(r, lkb);
4406 goto out;
e7fd4179
DT		 4407 case -EEXIST:
4408 log_debug(ls, "master copy exists %x", lkb->lkb_id);
4409 /* fall through */
4410 case 0:
4411 lkb->lkb_remid = rl->rl_remid;
4412 break;
4413 default:
4414 log_error(ls, "dlm_recover_process_copy unknown error %d %x",
4415 error, lkb->lkb_id);
4416 }
4417
4418 /* an ack for dlm_recover_locks() which waits for replies from
4419 all the locks it sends to new masters */
4420 dlm_recovered_lock(r);
dc200a88 4421 out:
e7fd4179
DT		 4422 unlock_rsb(r);
4423 put_rsb(r);
b3f58d8f 4424 dlm_put_lkb(lkb);
e7fd4179
DT		 4425
4426 return 0;
4427}
4428
597d0cae
DT		 4429int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
4430 int mode, uint32_t flags, void *name, unsigned int namelen,
d7db923e 4431 unsigned long timeout_cs)
597d0cae
DT		 4432{
4433 struct dlm_lkb *lkb;
4434 struct dlm_args args;
4435 int error;
4436
85e86edf 4437 dlm_lock_recovery(ls);
597d0cae
DT		 4438
4439 error = create_lkb(ls, &lkb);
4440 if (error) {
4441 kfree(ua);
4442 goto out;
4443 }
4444
4445 if (flags & DLM_LKF_VALBLK) {
62a0f623 4446 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
597d0cae
DT		 4447 if (!ua->lksb.sb_lvbptr) {
4448 kfree(ua);
4449 __put_lkb(ls, lkb);
4450 error = -ENOMEM;
4451 goto out;
4452 }
4453 }
4454
52bda2b5 4455 /* After ua is attached to lkb it will be freed by dlm_free_lkb().
597d0cae
DT		 4456 When DLM_IFL_USER is set, the dlm knows that this is a userspace
4457 lock and that lkb_astparam is the dlm_user_args structure. */
4458
d7db923e 4459 error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
32f105a1 4460 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
597d0cae
DT		 4461 lkb->lkb_flags |= DLM_IFL_USER;
4462 ua->old_mode = DLM_LOCK_IV;
4463
4464 if (error) {
4465 __put_lkb(ls, lkb);
4466 goto out;
4467 }
4468
4469 error = request_lock(ls, lkb, name, namelen, &args);
4470
4471 switch (error) {
4472 case 0:
4473 break;
4474 case -EINPROGRESS:
4475 error = 0;
4476 break;
4477 case -EAGAIN:
4478 error = 0;
4479 /* fall through */
4480 default:
4481 __put_lkb(ls, lkb);
4482 goto out;
4483 }
4484
4485 /* add this new lkb to the per-process list of locks */
4486 spin_lock(&ua->proc->locks_spin);
ef0c2bb0 4487 hold_lkb(lkb);
597d0cae
DT		 4488 list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
4489 spin_unlock(&ua->proc->locks_spin);
4490 out:
85e86edf 4491 dlm_unlock_recovery(ls);
597d0cae
DT		 4492 return error;
4493}
4494
4495int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
d7db923e
DT		 4496 int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
4497 unsigned long timeout_cs)
597d0cae
DT		 4498{
4499 struct dlm_lkb *lkb;
4500 struct dlm_args args;
4501 struct dlm_user_args *ua;
4502 int error;
4503
85e86edf 4504 dlm_lock_recovery(ls);
597d0cae
DT		 4505
4506 error = find_lkb(ls, lkid, &lkb);
4507 if (error)
4508 goto out;
4509
4510 /* user can change the params on its lock when it converts it, or
4511 add an lvb that didn't exist before */
4512
4513 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4514
4515 if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
62a0f623 4516 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
597d0cae
DT		 4517 if (!ua->lksb.sb_lvbptr) {
4518 error = -ENOMEM;
4519 goto out_put;
4520 }
4521 }
4522 if (lvb_in && ua->lksb.sb_lvbptr)
4523 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4524
d7db923e 4525 ua->xid = ua_tmp->xid;
597d0cae
DT		 4526 ua->castparam = ua_tmp->castparam;
4527 ua->castaddr = ua_tmp->castaddr;
4528 ua->bastparam = ua_tmp->bastparam;
4529 ua->bastaddr = ua_tmp->bastaddr;
10948eb4 4530 ua->user_lksb = ua_tmp->user_lksb;
597d0cae
DT		 4531 ua->old_mode = lkb->lkb_grmode;
4532
d7db923e
DT		 4533 error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
4534 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
597d0cae
DT		 4535 if (error)
4536 goto out_put;
4537
4538 error = convert_lock(ls, lkb, &args);
4539
c85d65e9 4540 if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
597d0cae
DT		 4541 error = 0;
4542 out_put:
4543 dlm_put_lkb(lkb);
4544 out:
85e86edf 4545 dlm_unlock_recovery(ls);
597d0cae
DT		 4546 kfree(ua_tmp);
4547 return error;
4548}
4549
4550int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4551 uint32_t flags, uint32_t lkid, char *lvb_in)
4552{
4553 struct dlm_lkb *lkb;
4554 struct dlm_args args;
4555 struct dlm_user_args *ua;
4556 int error;
4557
85e86edf 4558 dlm_lock_recovery(ls);
597d0cae
DT		 4559
4560 error = find_lkb(ls, lkid, &lkb);
4561 if (error)
4562 goto out;
4563
4564 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4565
4566 if (lvb_in && ua->lksb.sb_lvbptr)
4567 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
b434eda6
PC		 4568 if (ua_tmp->castparam)
4569 ua->castparam = ua_tmp->castparam;
cc346d55 4570 ua->user_lksb = ua_tmp->user_lksb;
597d0cae
DT		 4571
4572 error = set_unlock_args(flags, ua, &args);
4573 if (error)
4574 goto out_put;
4575
4576 error = unlock_lock(ls, lkb, &args);
4577
4578 if (error == -DLM_EUNLOCK)
4579 error = 0;
ef0c2bb0
DT		 4580 /* from validate_unlock_args() */
4581 if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
4582 error = 0;
597d0cae
DT		 4583 if (error)
4584 goto out_put;
4585
4586 spin_lock(&ua->proc->locks_spin);
a1bc86e6
DT		 4587 /* dlm_user_add_ast() may have already taken lkb off the proc list */
4588 if (!list_empty(&lkb->lkb_ownqueue))
4589 list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
597d0cae 4590 spin_unlock(&ua->proc->locks_spin);
597d0cae
DT		 4591 out_put:
4592 dlm_put_lkb(lkb);
4593 out:
85e86edf 4594 dlm_unlock_recovery(ls);
ef0c2bb0 4595 kfree(ua_tmp);
597d0cae
DT		 4596 return error;
4597}
4598
4599int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4600 uint32_t flags, uint32_t lkid)
4601{
4602 struct dlm_lkb *lkb;
4603 struct dlm_args args;
4604 struct dlm_user_args *ua;
4605 int error;
4606
85e86edf 4607 dlm_lock_recovery(ls);
597d0cae
DT		 4608
4609 error = find_lkb(ls, lkid, &lkb);
4610 if (error)
4611 goto out;
4612
4613 ua = (struct dlm_user_args *)lkb->lkb_astparam;
b434eda6
PC		 4614 if (ua_tmp->castparam)
4615 ua->castparam = ua_tmp->castparam;
c059f70e 4616 ua->user_lksb = ua_tmp->user_lksb;
597d0cae
DT		 4617
4618 error = set_unlock_args(flags, ua, &args);
4619 if (error)
4620 goto out_put;
4621
4622 error = cancel_lock(ls, lkb, &args);
4623
4624 if (error == -DLM_ECANCEL)
4625 error = 0;
ef0c2bb0
DT		 4626 /* from validate_unlock_args() */
4627 if (error == -EBUSY)
4628 error = 0;
597d0cae
DT		 4629 out_put:
4630 dlm_put_lkb(lkb);
4631 out:
85e86edf 4632 dlm_unlock_recovery(ls);
ef0c2bb0 4633 kfree(ua_tmp);
597d0cae
DT		 4634 return error;
4635}
4636
8b4021fa
DT		 4637int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
4638{
4639 struct dlm_lkb *lkb;
4640 struct dlm_args args;
4641 struct dlm_user_args *ua;
4642 struct dlm_rsb *r;
4643 int error;
4644
4645 dlm_lock_recovery(ls);
4646
4647 error = find_lkb(ls, lkid, &lkb);
4648 if (error)
4649 goto out;
4650
4651 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4652
4653 error = set_unlock_args(flags, ua, &args);
4654 if (error)
4655 goto out_put;
4656
4657 /* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4658
4659 r = lkb->lkb_resource;
4660 hold_rsb(r);
4661 lock_rsb(r);
4662
4663 error = validate_unlock_args(lkb, &args);
4664 if (error)
4665 goto out_r;
4666 lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
4667
4668 error = _cancel_lock(r, lkb);
4669 out_r:
4670 unlock_rsb(r);
4671 put_rsb(r);
4672
4673 if (error == -DLM_ECANCEL)
4674 error = 0;
4675 /* from validate_unlock_args() */
4676 if (error == -EBUSY)
4677 error = 0;
4678 out_put:
4679 dlm_put_lkb(lkb);
4680 out:
4681 dlm_unlock_recovery(ls);
4682 return error;
4683}
4684
ef0c2bb0
DT		 4685/* lkb's that are removed from the waiters list by revert are just left on the
4686 orphans list with the granted orphan locks, to be freed by purge */
4687
597d0cae
DT		 4688static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4689{
4690 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
ef0c2bb0
DT		 4691 struct dlm_args args;
4692 int error;
597d0cae 4693
ef0c2bb0
DT		 4694 hold_lkb(lkb);
4695 mutex_lock(&ls->ls_orphans_mutex);
4696 list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
4697 mutex_unlock(&ls->ls_orphans_mutex);
597d0cae 4698
ef0c2bb0
DT		 4699 set_unlock_args(0, ua, &args);
4700
4701 error = cancel_lock(ls, lkb, &args);
4702 if (error == -DLM_ECANCEL)
4703 error = 0;
4704 return error;
597d0cae
DT		 4705}
4706
4707/* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4708 Regardless of what rsb queue the lock is on, it's removed and freed. */
4709
4710static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4711{
4712 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
4713 struct dlm_args args;
4714 int error;
4715
597d0cae
DT		 4716 set_unlock_args(DLM_LKF_FORCEUNLOCK, ua, &args);
4717
4718 error = unlock_lock(ls, lkb, &args);
4719 if (error == -DLM_EUNLOCK)
4720 error = 0;
4721 return error;
4722}
4723
ef0c2bb0
DT		 4724/* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4725 (which does lock_rsb) due to deadlock with receiving a message that does
4726 lock_rsb followed by dlm_user_add_ast() */
4727
4728static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
4729 struct dlm_user_proc *proc)
4730{
4731 struct dlm_lkb *lkb = NULL;
4732
4733 mutex_lock(&ls->ls_clear_proc_locks);
4734 if (list_empty(&proc->locks))
4735 goto out;
4736
4737 lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
4738 list_del_init(&lkb->lkb_ownqueue);
4739
4740 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4741 lkb->lkb_flags |= DLM_IFL_ORPHAN;
4742 else
4743 lkb->lkb_flags |= DLM_IFL_DEAD;
4744 out:
4745 mutex_unlock(&ls->ls_clear_proc_locks);
4746 return lkb;
4747}
4748
597d0cae
DT		 4749/* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4750 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4751 which we clear here. */
4752
4753/* proc CLOSING flag is set so no more device_reads should look at proc->asts
4754 list, and no more device_writes should add lkb's to proc->locks list; so we
4755 shouldn't need to take asts_spin or locks_spin here. this assumes that
4756 device reads/writes/closes are serialized -- FIXME: we may need to serialize
4757 them ourself. */
4758
4759void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4760{
4761 struct dlm_lkb *lkb, *safe;
4762
85e86edf 4763 dlm_lock_recovery(ls);
597d0cae 4764
ef0c2bb0
DT		 4765 while (1) {
4766 lkb = del_proc_lock(ls, proc);
4767 if (!lkb)
4768 break;
84d8cd69 4769 del_timeout(lkb);
ef0c2bb0 4770 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
597d0cae 4771 orphan_proc_lock(ls, lkb);
ef0c2bb0 4772 else
597d0cae 4773 unlock_proc_lock(ls, lkb);
597d0cae
DT		 4774
4775 /* this removes the reference for the proc->locks list
4776 added by dlm_user_request, it may result in the lkb
4777 being freed */
4778
4779 dlm_put_lkb(lkb);
4780 }
a1bc86e6 4781
ef0c2bb0
DT		 4782 mutex_lock(&ls->ls_clear_proc_locks);
4783
a1bc86e6
DT		 4784 /* in-progress unlocks */
4785 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4786 list_del_init(&lkb->lkb_ownqueue);
4787 lkb->lkb_flags |= DLM_IFL_DEAD;
4788 dlm_put_lkb(lkb);
4789 }
4790
4791 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
8a358ca8 4792 lkb->lkb_ast_type = 0;
a1bc86e6
DT		 4793 list_del(&lkb->lkb_astqueue);
4794 dlm_put_lkb(lkb);
4795 }
4796
597d0cae 4797 mutex_unlock(&ls->ls_clear_proc_locks);
85e86edf 4798 dlm_unlock_recovery(ls);
597d0cae 4799}
a1bc86e6 4800
8499137d
DT		 4801static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4802{
4803 struct dlm_lkb *lkb, *safe;
4804
4805 while (1) {
4806 lkb = NULL;
4807 spin_lock(&proc->locks_spin);
4808 if (!list_empty(&proc->locks)) {
4809 lkb = list_entry(proc->locks.next, struct dlm_lkb,
4810 lkb_ownqueue);
4811 list_del_init(&lkb->lkb_ownqueue);
4812 }
4813 spin_unlock(&proc->locks_spin);
4814
4815 if (!lkb)
4816 break;
4817
4818 lkb->lkb_flags |= DLM_IFL_DEAD;
4819 unlock_proc_lock(ls, lkb);
4820 dlm_put_lkb(lkb); /* ref from proc->locks list */
4821 }
4822
4823 spin_lock(&proc->locks_spin);
4824 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4825 list_del_init(&lkb->lkb_ownqueue);
4826 lkb->lkb_flags |= DLM_IFL_DEAD;
4827 dlm_put_lkb(lkb);
4828 }
4829 spin_unlock(&proc->locks_spin);
4830
4831 spin_lock(&proc->asts_spin);
4832 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4833 list_del(&lkb->lkb_astqueue);
4834 dlm_put_lkb(lkb);
4835 }
4836 spin_unlock(&proc->asts_spin);
4837}
4838
4839/* pid of 0 means purge all orphans */
4840
4841static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
4842{
4843 struct dlm_lkb *lkb, *safe;
4844
4845 mutex_lock(&ls->ls_orphans_mutex);
4846 list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
4847 if (pid && lkb->lkb_ownpid != pid)
4848 continue;
4849 unlock_proc_lock(ls, lkb);
4850 list_del_init(&lkb->lkb_ownqueue);
4851 dlm_put_lkb(lkb);
4852 }
4853 mutex_unlock(&ls->ls_orphans_mutex);
4854}
4855
4856static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
4857{
4858 struct dlm_message *ms;
4859 struct dlm_mhandle *mh;
4860 int error;
4861
4862 error = _create_message(ls, sizeof(struct dlm_message), nodeid,
4863 DLM_MSG_PURGE, &ms, &mh);
4864 if (error)
4865 return error;
4866 ms->m_nodeid = nodeid;
4867 ms->m_pid = pid;
4868
4869 return send_message(mh, ms);
4870}
4871
4872int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
4873 int nodeid, int pid)
4874{
4875 int error = 0;
4876
4877 if (nodeid != dlm_our_nodeid()) {
4878 error = send_purge(ls, nodeid, pid);
4879 } else {
85e86edf 4880 dlm_lock_recovery(ls);
8499137d
DT		 4881 if (pid == current->pid)
4882 purge_proc_locks(ls, proc);
4883 else
4884 do_purge(ls, nodeid, pid);
85e86edf 4885 dlm_unlock_recovery(ls);
8499137d
DT		 4886 }
4887 return error;
4888}
4889
Ubuntu Artful kernel mirror