dlm: reject normal unlock when lock is waiting for lookup
[deliverable/linux.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;
1854 int error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid();
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
1888 for (;;) {
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 }
1902
1903 if (ret_nodeid == our_nodeid) {
1904 r->res_first_lkid = 0;
1905 r->res_nodeid = 0;
1906 lkb->lkb_nodeid = 0;
1907 } else {
1908 r->res_first_lkid = lkb->lkb_id;
1909 r->res_nodeid = ret_nodeid;
1910 lkb->lkb_nodeid = ret_nodeid;
1911 }
1912 return 0;
1913}
1914
1915static void process_lookup_list(struct dlm_rsb *r)
1916{
1917 struct dlm_lkb *lkb, *safe;
1918
1919 list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
ef0c2bb0 1920 list_del_init(&lkb->lkb_rsb_lookup);
e7fd4179
DT
1921 _request_lock(r, lkb);
1922 schedule();
1923 }
1924}
1925
1926/* confirm_master -- confirm (or deny) an rsb's master nodeid */
1927
1928static void confirm_master(struct dlm_rsb *r, int error)
1929{
1930 struct dlm_lkb *lkb;
1931
1932 if (!r->res_first_lkid)
1933 return;
1934
1935 switch (error) {
1936 case 0:
1937 case -EINPROGRESS:
1938 r->res_first_lkid = 0;
1939 process_lookup_list(r);
1940 break;
1941
1942 case -EAGAIN:
aec64e1b
DT
1943 case -EBADR:
1944 case -ENOTBLK:
1945 /* the remote request failed and won't be retried (it was
1946 a NOQUEUE, or has been canceled/unlocked); make a waiting
1947 lkb the first_lkid */
e7fd4179
DT
1948
1949 r->res_first_lkid = 0;
1950
1951 if (!list_empty(&r->res_lookup)) {
1952 lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
1953 lkb_rsb_lookup);
ef0c2bb0 1954 list_del_init(&lkb->lkb_rsb_lookup);
e7fd4179
DT
1955 r->res_first_lkid = lkb->lkb_id;
1956 _request_lock(r, lkb);
1957 } else
1958 r->res_nodeid = -1;
1959 break;
1960
1961 default:
1962 log_error(r->res_ls, "confirm_master unknown error %d", error);
1963 }
1964}
1965
1966static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags,
d7db923e 1967 int namelen, unsigned long timeout_cs, void *ast,
3bcd3687 1968 void *astarg, void *bast, struct dlm_args *args)
e7fd4179
DT
1969{
1970 int rv = -EINVAL;
1971
1972 /* check for invalid arg usage */
1973
1974 if (mode < 0 || mode > DLM_LOCK_EX)
1975 goto out;
1976
1977 if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN))
1978 goto out;
1979
1980 if (flags & DLM_LKF_CANCEL)
1981 goto out;
1982
1983 if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT))
1984 goto out;
1985
1986 if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT))
1987 goto out;
1988
1989 if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE)
1990 goto out;
1991
1992 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT)
1993 goto out;
1994
1995 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT)
1996 goto out;
1997
1998 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE)
1999 goto out;
2000
2001 if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL)
2002 goto out;
2003
2004 if (!ast || !lksb)
2005 goto out;
2006
2007 if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr)
2008 goto out;
2009
e7fd4179
DT
2010 if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid)
2011 goto out;
2012
2013 /* these args will be copied to the lkb in validate_lock_args,
2014 it cannot be done now because when converting locks, fields in
2015 an active lkb cannot be modified before locking the rsb */
2016
2017 args->flags = flags;
2018 args->astaddr = ast;
2019 args->astparam = (long) astarg;
2020 args->bastaddr = bast;
d7db923e 2021 args->timeout = timeout_cs;
e7fd4179
DT
2022 args->mode = mode;
2023 args->lksb = lksb;
e7fd4179
DT
2024 rv = 0;
2025 out:
2026 return rv;
2027}
2028
2029static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
2030{
2031 if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK |
2032 DLM_LKF_FORCEUNLOCK))
2033 return -EINVAL;
2034
ef0c2bb0
DT
2035 if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
2036 return -EINVAL;
2037
e7fd4179
DT
2038 args->flags = flags;
2039 args->astparam = (long) astarg;
2040 return 0;
2041}
2042
2043static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
2044 struct dlm_args *args)
2045{
2046 int rv = -EINVAL;
2047
2048 if (args->flags & DLM_LKF_CONVERT) {
2049 if (lkb->lkb_flags & DLM_IFL_MSTCPY)
2050 goto out;
2051
2052 if (args->flags & DLM_LKF_QUECVT &&
2053 !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1])
2054 goto out;
2055
2056 rv = -EBUSY;
2057 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
2058 goto out;
2059
2060 if (lkb->lkb_wait_type)
2061 goto out;
ef0c2bb0
DT
2062
2063 if (is_overlap(lkb))
2064 goto out;
e7fd4179
DT
2065 }
2066
2067 lkb->lkb_exflags = args->flags;
2068 lkb->lkb_sbflags = 0;
2069 lkb->lkb_astaddr = args->astaddr;
2070 lkb->lkb_astparam = args->astparam;
2071 lkb->lkb_bastaddr = args->bastaddr;
2072 lkb->lkb_rqmode = args->mode;
2073 lkb->lkb_lksb = args->lksb;
2074 lkb->lkb_lvbptr = args->lksb->sb_lvbptr;
2075 lkb->lkb_ownpid = (int) current->pid;
d7db923e 2076 lkb->lkb_timeout_cs = args->timeout;
e7fd4179
DT
2077 rv = 0;
2078 out:
2079 return rv;
2080}
2081
ef0c2bb0
DT
2082/* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
2083 for success */
2084
2085/* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
2086 because there may be a lookup in progress and it's valid to do
2087 cancel/unlockf on it */
2088
e7fd4179
DT
2089static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
2090{
ef0c2bb0 2091 struct dlm_ls *ls = lkb->lkb_resource->res_ls;
e7fd4179
DT
2092 int rv = -EINVAL;
2093
ef0c2bb0
DT
2094 if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
2095 log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
2096 dlm_print_lkb(lkb);
e7fd4179 2097 goto out;
ef0c2bb0 2098 }
e7fd4179 2099
ef0c2bb0
DT
2100 /* an lkb may still exist even though the lock is EOL'ed due to a
2101 cancel, unlock or failed noqueue request; an app can't use these
2102 locks; return same error as if the lkid had not been found at all */
e7fd4179 2103
ef0c2bb0
DT
2104 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
2105 log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
2106 rv = -ENOENT;
e7fd4179 2107 goto out;
ef0c2bb0 2108 }
e7fd4179 2109
ef0c2bb0
DT
2110 /* an lkb may be waiting for an rsb lookup to complete where the
2111 lookup was initiated by another lock */
2112
42dc1601
DT
2113 if (!list_empty(&lkb->lkb_rsb_lookup)) {
2114 if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
ef0c2bb0
DT
2115 log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
2116 list_del_init(&lkb->lkb_rsb_lookup);
2117 queue_cast(lkb->lkb_resource, lkb,
2118 args->flags & DLM_LKF_CANCEL ?
2119 -DLM_ECANCEL : -DLM_EUNLOCK);
2120 unhold_lkb(lkb); /* undoes create_lkb() */
ef0c2bb0 2121 }
42dc1601
DT
2122 /* caller changes -EBUSY to 0 for CANCEL and FORCEUNLOCK */
2123 rv = -EBUSY;
2124 goto out;
ef0c2bb0
DT
2125 }
2126
2127 /* cancel not allowed with another cancel/unlock in progress */
2128
2129 if (args->flags & DLM_LKF_CANCEL) {
2130 if (lkb->lkb_exflags & DLM_LKF_CANCEL)
2131 goto out;
2132
2133 if (is_overlap(lkb))
2134 goto out;
2135
3ae1acf9
DT
2136 /* don't let scand try to do a cancel */
2137 del_timeout(lkb);
2138
ef0c2bb0
DT
2139 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2140 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2141 rv = -EBUSY;
2142 goto out;
2143 }
2144
2145 switch (lkb->lkb_wait_type) {
2146 case DLM_MSG_LOOKUP:
2147 case DLM_MSG_REQUEST:
2148 lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
2149 rv = -EBUSY;
2150 goto out;
2151 case DLM_MSG_UNLOCK:
2152 case DLM_MSG_CANCEL:
2153 goto out;
2154 }
2155 /* add_to_waiters() will set OVERLAP_CANCEL */
2156 goto out_ok;
2157 }
2158
2159 /* do we need to allow a force-unlock if there's a normal unlock
2160 already in progress? in what conditions could the normal unlock
2161 fail such that we'd want to send a force-unlock to be sure? */
2162
2163 if (args->flags & DLM_LKF_FORCEUNLOCK) {
2164 if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
2165 goto out;
2166
2167 if (is_overlap_unlock(lkb))
2168 goto out;
e7fd4179 2169
3ae1acf9
DT
2170 /* don't let scand try to do a cancel */
2171 del_timeout(lkb);
2172
ef0c2bb0
DT
2173 if (lkb->lkb_flags & DLM_IFL_RESEND) {
2174 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2175 rv = -EBUSY;
2176 goto out;
2177 }
2178
2179 switch (lkb->lkb_wait_type) {
2180 case DLM_MSG_LOOKUP:
2181 case DLM_MSG_REQUEST:
2182 lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
2183 rv = -EBUSY;
2184 goto out;
2185 case DLM_MSG_UNLOCK:
2186 goto out;
2187 }
2188 /* add_to_waiters() will set OVERLAP_UNLOCK */
2189 goto out_ok;
2190 }
2191
2192 /* normal unlock not allowed if there's any op in progress */
e7fd4179 2193 rv = -EBUSY;
ef0c2bb0 2194 if (lkb->lkb_wait_type || lkb->lkb_wait_count)
e7fd4179
DT
2195 goto out;
2196
2197 out_ok:
ef0c2bb0
DT
2198 /* an overlapping op shouldn't blow away exflags from other op */
2199 lkb->lkb_exflags |= args->flags;
e7fd4179
DT
2200 lkb->lkb_sbflags = 0;
2201 lkb->lkb_astparam = args->astparam;
e7fd4179
DT
2202 rv = 0;
2203 out:
ef0c2bb0
DT
2204 if (rv)
2205 log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
2206 lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
2207 args->flags, lkb->lkb_wait_type,
2208 lkb->lkb_resource->res_name);
e7fd4179
DT
2209 return rv;
2210}
2211
2212/*
2213 * Four stage 4 varieties:
2214 * do_request(), do_convert(), do_unlock(), do_cancel()
2215 * These are called on the master node for the given lock and
2216 * from the central locking logic.
2217 */
2218
2219static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2220{
2221 int error = 0;
2222
c85d65e9 2223 if (can_be_granted(r, lkb, 1, NULL)) {
e7fd4179
DT
2224 grant_lock(r, lkb);
2225 queue_cast(r, lkb, 0);
2226 goto out;
2227 }
2228
2229 if (can_be_queued(lkb)) {
2230 error = -EINPROGRESS;
2231 add_lkb(r, lkb, DLM_LKSTS_WAITING);
2232 send_blocking_asts(r, lkb);
3ae1acf9 2233 add_timeout(lkb);
e7fd4179
DT
2234 goto out;
2235 }
2236
2237 error = -EAGAIN;
2238 if (force_blocking_asts(lkb))
2239 send_blocking_asts_all(r, lkb);
2240 queue_cast(r, lkb, -EAGAIN);
2241
2242 out:
2243 return error;
2244}
2245
2246static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2247{
2248 int error = 0;
c85d65e9 2249 int deadlk = 0;
e7fd4179
DT
2250
2251 /* changing an existing lock may allow others to be granted */
2252
c85d65e9 2253 if (can_be_granted(r, lkb, 1, &deadlk)) {
e7fd4179
DT
2254 grant_lock(r, lkb);
2255 queue_cast(r, lkb, 0);
2256 grant_pending_locks(r);
2257 goto out;
2258 }
2259
c85d65e9
DT
2260 /* can_be_granted() detected that this lock would block in a conversion
2261 deadlock, so we leave it on the granted queue and return EDEADLK in
2262 the ast for the convert. */
2263
2264 if (deadlk) {
2265 /* it's left on the granted queue */
2266 log_debug(r->res_ls, "deadlock %x node %d sts%d g%d r%d %s",
2267 lkb->lkb_id, lkb->lkb_nodeid, lkb->lkb_status,
2268 lkb->lkb_grmode, lkb->lkb_rqmode, r->res_name);
2269 revert_lock(r, lkb);
2270 queue_cast(r, lkb, -EDEADLK);
2271 error = -EDEADLK;
2272 goto out;
2273 }
2274
7d3c1feb
DT
2275 /* is_demoted() means the can_be_granted() above set the grmode
2276 to NL, and left us on the granted queue. This auto-demotion
2277 (due to CONVDEADLK) might mean other locks, and/or this lock, are
2278 now grantable. We have to try to grant other converting locks
2279 before we try again to grant this one. */
2280
2281 if (is_demoted(lkb)) {
36509258 2282 grant_pending_convert(r, DLM_LOCK_IV, NULL);
7d3c1feb
DT
2283 if (_can_be_granted(r, lkb, 1)) {
2284 grant_lock(r, lkb);
2285 queue_cast(r, lkb, 0);
e7fd4179 2286 grant_pending_locks(r);
7d3c1feb
DT
2287 goto out;
2288 }
2289 /* else fall through and move to convert queue */
2290 }
2291
2292 if (can_be_queued(lkb)) {
e7fd4179
DT
2293 error = -EINPROGRESS;
2294 del_lkb(r, lkb);
2295 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
2296 send_blocking_asts(r, lkb);
3ae1acf9 2297 add_timeout(lkb);
e7fd4179
DT
2298 goto out;
2299 }
2300
2301 error = -EAGAIN;
2302 if (force_blocking_asts(lkb))
2303 send_blocking_asts_all(r, lkb);
2304 queue_cast(r, lkb, -EAGAIN);
2305
2306 out:
2307 return error;
2308}
2309
2310static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2311{
2312 remove_lock(r, lkb);
2313 queue_cast(r, lkb, -DLM_EUNLOCK);
2314 grant_pending_locks(r);
2315 return -DLM_EUNLOCK;
2316}
2317
ef0c2bb0 2318/* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
907b9bce 2319
e7fd4179
DT
2320static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2321{
ef0c2bb0
DT
2322 int error;
2323
2324 error = revert_lock(r, lkb);
2325 if (error) {
2326 queue_cast(r, lkb, -DLM_ECANCEL);
2327 grant_pending_locks(r);
2328 return -DLM_ECANCEL;
2329 }
2330 return 0;
e7fd4179
DT
2331}
2332
2333/*
2334 * Four stage 3 varieties:
2335 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock()
2336 */
2337
2338/* add a new lkb to a possibly new rsb, called by requesting process */
2339
2340static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2341{
2342 int error;
2343
2344 /* set_master: sets lkb nodeid from r */
2345
2346 error = set_master(r, lkb);
2347 if (error < 0)
2348 goto out;
2349 if (error) {
2350 error = 0;
2351 goto out;
2352 }
2353
2354 if (is_remote(r))
2355 /* receive_request() calls do_request() on remote node */
2356 error = send_request(r, lkb);
2357 else
2358 error = do_request(r, lkb);
2359 out:
2360 return error;
2361}
2362
3bcd3687 2363/* change some property of an existing lkb, e.g. mode */
e7fd4179
DT
2364
2365static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2366{
2367 int error;
2368
2369 if (is_remote(r))
2370 /* receive_convert() calls do_convert() on remote node */
2371 error = send_convert(r, lkb);
2372 else
2373 error = do_convert(r, lkb);
2374
2375 return error;
2376}
2377
2378/* remove an existing lkb from the granted queue */
2379
2380static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2381{
2382 int error;
2383
2384 if (is_remote(r))
2385 /* receive_unlock() calls do_unlock() on remote node */
2386 error = send_unlock(r, lkb);
2387 else
2388 error = do_unlock(r, lkb);
2389
2390 return error;
2391}
2392
2393/* remove an existing lkb from the convert or wait queue */
2394
2395static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2396{
2397 int error;
2398
2399 if (is_remote(r))
2400 /* receive_cancel() calls do_cancel() on remote node */
2401 error = send_cancel(r, lkb);
2402 else
2403 error = do_cancel(r, lkb);
2404
2405 return error;
2406}
2407
2408/*
2409 * Four stage 2 varieties:
2410 * request_lock(), convert_lock(), unlock_lock(), cancel_lock()
2411 */
2412
2413static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name,
2414 int len, struct dlm_args *args)
2415{
2416 struct dlm_rsb *r;
2417 int error;
2418
2419 error = validate_lock_args(ls, lkb, args);
2420 if (error)
2421 goto out;
2422
2423 error = find_rsb(ls, name, len, R_CREATE, &r);
2424 if (error)
2425 goto out;
2426
2427 lock_rsb(r);
2428
2429 attach_lkb(r, lkb);
2430 lkb->lkb_lksb->sb_lkid = lkb->lkb_id;
2431
2432 error = _request_lock(r, lkb);
2433
2434 unlock_rsb(r);
2435 put_rsb(r);
2436
2437 out:
2438 return error;
2439}
2440
2441static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2442 struct dlm_args *args)
2443{
2444 struct dlm_rsb *r;
2445 int error;
2446
2447 r = lkb->lkb_resource;
2448
2449 hold_rsb(r);
2450 lock_rsb(r);
2451
2452 error = validate_lock_args(ls, lkb, args);
2453 if (error)
2454 goto out;
2455
2456 error = _convert_lock(r, lkb);
2457 out:
2458 unlock_rsb(r);
2459 put_rsb(r);
2460 return error;
2461}
2462
2463static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2464 struct dlm_args *args)
2465{
2466 struct dlm_rsb *r;
2467 int error;
2468
2469 r = lkb->lkb_resource;
2470
2471 hold_rsb(r);
2472 lock_rsb(r);
2473
2474 error = validate_unlock_args(lkb, args);
2475 if (error)
2476 goto out;
2477
2478 error = _unlock_lock(r, lkb);
2479 out:
2480 unlock_rsb(r);
2481 put_rsb(r);
2482 return error;
2483}
2484
2485static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb,
2486 struct dlm_args *args)
2487{
2488 struct dlm_rsb *r;
2489 int error;
2490
2491 r = lkb->lkb_resource;
2492
2493 hold_rsb(r);
2494 lock_rsb(r);
2495
2496 error = validate_unlock_args(lkb, args);
2497 if (error)
2498 goto out;
2499
2500 error = _cancel_lock(r, lkb);
2501 out:
2502 unlock_rsb(r);
2503 put_rsb(r);
2504 return error;
2505}
2506
2507/*
2508 * Two stage 1 varieties: dlm_lock() and dlm_unlock()
2509 */
2510
2511int dlm_lock(dlm_lockspace_t *lockspace,
2512 int mode,
2513 struct dlm_lksb *lksb,
2514 uint32_t flags,
2515 void *name,
2516 unsigned int namelen,
2517 uint32_t parent_lkid,
2518 void (*ast) (void *astarg),
2519 void *astarg,
3bcd3687 2520 void (*bast) (void *astarg, int mode))
e7fd4179
DT
2521{
2522 struct dlm_ls *ls;
2523 struct dlm_lkb *lkb;
2524 struct dlm_args args;
2525 int error, convert = flags & DLM_LKF_CONVERT;
2526
2527 ls = dlm_find_lockspace_local(lockspace);
2528 if (!ls)
2529 return -EINVAL;
2530
85e86edf 2531 dlm_lock_recovery(ls);
e7fd4179
DT
2532
2533 if (convert)
2534 error = find_lkb(ls, lksb->sb_lkid, &lkb);
2535 else
2536 error = create_lkb(ls, &lkb);
2537
2538 if (error)
2539 goto out;
2540
d7db923e 2541 error = set_lock_args(mode, lksb, flags, namelen, 0, ast,
3bcd3687 2542 astarg, bast, &args);
e7fd4179
DT
2543 if (error)
2544 goto out_put;
2545
2546 if (convert)
2547 error = convert_lock(ls, lkb, &args);
2548 else
2549 error = request_lock(ls, lkb, name, namelen, &args);
2550
2551 if (error == -EINPROGRESS)
2552 error = 0;
2553 out_put:
2554 if (convert || error)
b3f58d8f 2555 __put_lkb(ls, lkb);
c85d65e9 2556 if (error == -EAGAIN || error == -EDEADLK)
e7fd4179
DT
2557 error = 0;
2558 out:
85e86edf 2559 dlm_unlock_recovery(ls);
e7fd4179
DT
2560 dlm_put_lockspace(ls);
2561 return error;
2562}
2563
2564int dlm_unlock(dlm_lockspace_t *lockspace,
2565 uint32_t lkid,
2566 uint32_t flags,
2567 struct dlm_lksb *lksb,
2568 void *astarg)
2569{
2570 struct dlm_ls *ls;
2571 struct dlm_lkb *lkb;
2572 struct dlm_args args;
2573 int error;
2574
2575 ls = dlm_find_lockspace_local(lockspace);
2576 if (!ls)
2577 return -EINVAL;
2578
85e86edf 2579 dlm_lock_recovery(ls);
e7fd4179
DT
2580
2581 error = find_lkb(ls, lkid, &lkb);
2582 if (error)
2583 goto out;
2584
2585 error = set_unlock_args(flags, astarg, &args);
2586 if (error)
2587 goto out_put;
2588
2589 if (flags & DLM_LKF_CANCEL)
2590 error = cancel_lock(ls, lkb, &args);
2591 else
2592 error = unlock_lock(ls, lkb, &args);
2593
2594 if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
2595 error = 0;
ef0c2bb0
DT
2596 if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
2597 error = 0;
e7fd4179 2598 out_put:
b3f58d8f 2599 dlm_put_lkb(lkb);
e7fd4179 2600 out:
85e86edf 2601 dlm_unlock_recovery(ls);
e7fd4179
DT
2602 dlm_put_lockspace(ls);
2603 return error;
2604}
2605
2606/*
2607 * send/receive routines for remote operations and replies
2608 *
2609 * send_args
2610 * send_common
2611 * send_request receive_request
2612 * send_convert receive_convert
2613 * send_unlock receive_unlock
2614 * send_cancel receive_cancel
2615 * send_grant receive_grant
2616 * send_bast receive_bast
2617 * send_lookup receive_lookup
2618 * send_remove receive_remove
2619 *
2620 * send_common_reply
2621 * receive_request_reply send_request_reply
2622 * receive_convert_reply send_convert_reply
2623 * receive_unlock_reply send_unlock_reply
2624 * receive_cancel_reply send_cancel_reply
2625 * receive_lookup_reply send_lookup_reply
2626 */
2627
7e4dac33
DT
2628static int _create_message(struct dlm_ls *ls, int mb_len,
2629 int to_nodeid, int mstype,
2630 struct dlm_message **ms_ret,
2631 struct dlm_mhandle **mh_ret)
e7fd4179
DT
2632{
2633 struct dlm_message *ms;
2634 struct dlm_mhandle *mh;
2635 char *mb;
e7fd4179
DT
2636
2637 /* get_buffer gives us a message handle (mh) that we need to
2638 pass into lowcomms_commit and a message buffer (mb) that we
2639 write our data into */
2640
44f487a5 2641 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
e7fd4179
DT
2642 if (!mh)
2643 return -ENOBUFS;
2644
2645 memset(mb, 0, mb_len);
2646
2647 ms = (struct dlm_message *) mb;
2648
2649 ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
7e4dac33 2650 ms->m_header.h_lockspace = ls->ls_global_id;
e7fd4179
DT
2651 ms->m_header.h_nodeid = dlm_our_nodeid();
2652 ms->m_header.h_length = mb_len;
2653 ms->m_header.h_cmd = DLM_MSG;
2654
2655 ms->m_type = mstype;
2656
2657 *mh_ret = mh;
2658 *ms_ret = ms;
2659 return 0;
2660}
2661
7e4dac33
DT
2662static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
2663 int to_nodeid, int mstype,
2664 struct dlm_message **ms_ret,
2665 struct dlm_mhandle **mh_ret)
2666{
2667 int mb_len = sizeof(struct dlm_message);
2668
2669 switch (mstype) {
2670 case DLM_MSG_REQUEST:
2671 case DLM_MSG_LOOKUP:
2672 case DLM_MSG_REMOVE:
2673 mb_len += r->res_length;
2674 break;
2675 case DLM_MSG_CONVERT:
2676 case DLM_MSG_UNLOCK:
2677 case DLM_MSG_REQUEST_REPLY:
2678 case DLM_MSG_CONVERT_REPLY:
2679 case DLM_MSG_GRANT:
2680 if (lkb && lkb->lkb_lvbptr)
2681 mb_len += r->res_ls->ls_lvblen;
2682 break;
2683 }
2684
2685 return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
2686 ms_ret, mh_ret);
2687}
2688
e7fd4179
DT
2689/* further lowcomms enhancements or alternate implementations may make
2690 the return value from this function useful at some point */
2691
2692static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms)
2693{
2694 dlm_message_out(ms);
2695 dlm_lowcomms_commit_buffer(mh);
2696 return 0;
2697}
2698
2699static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb,
2700 struct dlm_message *ms)
2701{
2702 ms->m_nodeid = lkb->lkb_nodeid;
2703 ms->m_pid = lkb->lkb_ownpid;
2704 ms->m_lkid = lkb->lkb_id;
2705 ms->m_remid = lkb->lkb_remid;
2706 ms->m_exflags = lkb->lkb_exflags;
2707 ms->m_sbflags = lkb->lkb_sbflags;
2708 ms->m_flags = lkb->lkb_flags;
2709 ms->m_lvbseq = lkb->lkb_lvbseq;
2710 ms->m_status = lkb->lkb_status;
2711 ms->m_grmode = lkb->lkb_grmode;
2712 ms->m_rqmode = lkb->lkb_rqmode;
2713 ms->m_hash = r->res_hash;
2714
2715 /* m_result and m_bastmode are set from function args,
2716 not from lkb fields */
2717
2718 if (lkb->lkb_bastaddr)
2719 ms->m_asts |= AST_BAST;
2720 if (lkb->lkb_astaddr)
2721 ms->m_asts |= AST_COMP;
2722
da49f36f
DT
2723 /* compare with switch in create_message; send_remove() doesn't
2724 use send_args() */
e7fd4179 2725
da49f36f
DT
2726 switch (ms->m_type) {
2727 case DLM_MSG_REQUEST:
2728 case DLM_MSG_LOOKUP:
2729 memcpy(ms->m_extra, r->res_name, r->res_length);
2730 break;
2731 case DLM_MSG_CONVERT:
2732 case DLM_MSG_UNLOCK:
2733 case DLM_MSG_REQUEST_REPLY:
2734 case DLM_MSG_CONVERT_REPLY:
2735 case DLM_MSG_GRANT:
2736 if (!lkb->lkb_lvbptr)
2737 break;
e7fd4179 2738 memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
da49f36f
DT
2739 break;
2740 }
e7fd4179
DT
2741}
2742
2743static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
2744{
2745 struct dlm_message *ms;
2746 struct dlm_mhandle *mh;
2747 int to_nodeid, error;
2748
ef0c2bb0
DT
2749 error = add_to_waiters(lkb, mstype);
2750 if (error)
2751 return error;
e7fd4179
DT
2752
2753 to_nodeid = r->res_nodeid;
2754
2755 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2756 if (error)
2757 goto fail;
2758
2759 send_args(r, lkb, ms);
2760
2761 error = send_message(mh, ms);
2762 if (error)
2763 goto fail;
2764 return 0;
2765
2766 fail:
ef0c2bb0 2767 remove_from_waiters(lkb, msg_reply_type(mstype));
e7fd4179
DT
2768 return error;
2769}
2770
2771static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb)
2772{
2773 return send_common(r, lkb, DLM_MSG_REQUEST);
2774}
2775
2776static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
2777{
2778 int error;
2779
2780 error = send_common(r, lkb, DLM_MSG_CONVERT);
2781
2782 /* down conversions go without a reply from the master */
2783 if (!error && down_conversion(lkb)) {
ef0c2bb0
DT
2784 remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
2785 r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
e7fd4179 2786 r->res_ls->ls_stub_ms.m_result = 0;
32f105a1 2787 r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
e7fd4179
DT
2788 __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
2789 }
2790
2791 return error;
2792}
2793
2794/* FIXME: if this lkb is the only lock we hold on the rsb, then set
2795 MASTER_UNCERTAIN to force the next request on the rsb to confirm
2796 that the master is still correct. */
2797
2798static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
2799{
2800 return send_common(r, lkb, DLM_MSG_UNLOCK);
2801}
2802
2803static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
2804{
2805 return send_common(r, lkb, DLM_MSG_CANCEL);
2806}
2807
2808static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
2809{
2810 struct dlm_message *ms;
2811 struct dlm_mhandle *mh;
2812 int to_nodeid, error;
2813
2814 to_nodeid = lkb->lkb_nodeid;
2815
2816 error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
2817 if (error)
2818 goto out;
2819
2820 send_args(r, lkb, ms);
2821
2822 ms->m_result = 0;
2823
2824 error = send_message(mh, ms);
2825 out:
2826 return error;
2827}
2828
2829static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
2830{
2831 struct dlm_message *ms;
2832 struct dlm_mhandle *mh;
2833 int to_nodeid, error;
2834
2835 to_nodeid = lkb->lkb_nodeid;
2836
2837 error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
2838 if (error)
2839 goto out;
2840
2841 send_args(r, lkb, ms);
2842
2843 ms->m_bastmode = mode;
2844
2845 error = send_message(mh, ms);
2846 out:
2847 return error;
2848}
2849
2850static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
2851{
2852 struct dlm_message *ms;
2853 struct dlm_mhandle *mh;
2854 int to_nodeid, error;
2855
ef0c2bb0
DT
2856 error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
2857 if (error)
2858 return error;
e7fd4179
DT
2859
2860 to_nodeid = dlm_dir_nodeid(r);
2861
2862 error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
2863 if (error)
2864 goto fail;
2865
2866 send_args(r, lkb, ms);
2867
2868 error = send_message(mh, ms);
2869 if (error)
2870 goto fail;
2871 return 0;
2872
2873 fail:
ef0c2bb0 2874 remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
e7fd4179
DT
2875 return error;
2876}
2877
2878static int send_remove(struct dlm_rsb *r)
2879{
2880 struct dlm_message *ms;
2881 struct dlm_mhandle *mh;
2882 int to_nodeid, error;
2883
2884 to_nodeid = dlm_dir_nodeid(r);
2885
2886 error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
2887 if (error)
2888 goto out;
2889
2890 memcpy(ms->m_extra, r->res_name, r->res_length);
2891 ms->m_hash = r->res_hash;
2892
2893 error = send_message(mh, ms);
2894 out:
2895 return error;
2896}
2897
2898static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
2899 int mstype, int rv)
2900{
2901 struct dlm_message *ms;
2902 struct dlm_mhandle *mh;
2903 int to_nodeid, error;
2904
2905 to_nodeid = lkb->lkb_nodeid;
2906
2907 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
2908 if (error)
2909 goto out;
2910
2911 send_args(r, lkb, ms);
2912
2913 ms->m_result = rv;
2914
2915 error = send_message(mh, ms);
2916 out:
2917 return error;
2918}
2919
2920static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2921{
2922 return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv);
2923}
2924
2925static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2926{
2927 return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv);
2928}
2929
2930static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2931{
2932 return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv);
2933}
2934
2935static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
2936{
2937 return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv);
2938}
2939
2940static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in,
2941 int ret_nodeid, int rv)
2942{
2943 struct dlm_rsb *r = &ls->ls_stub_rsb;
2944 struct dlm_message *ms;
2945 struct dlm_mhandle *mh;
2946 int error, nodeid = ms_in->m_header.h_nodeid;
2947
2948 error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
2949 if (error)
2950 goto out;
2951
2952 ms->m_lkid = ms_in->m_lkid;
2953 ms->m_result = rv;
2954 ms->m_nodeid = ret_nodeid;
2955
2956 error = send_message(mh, ms);
2957 out:
2958 return error;
2959}
2960
2961/* which args we save from a received message depends heavily on the type
2962 of message, unlike the send side where we can safely send everything about
2963 the lkb for any type of message */
2964
2965static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms)
2966{
2967 lkb->lkb_exflags = ms->m_exflags;
6f90a8b1 2968 lkb->lkb_sbflags = ms->m_sbflags;
e7fd4179
DT
2969 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2970 (ms->m_flags & 0x0000FFFF);
2971}
2972
2973static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
2974{
2975 lkb->lkb_sbflags = ms->m_sbflags;
2976 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
2977 (ms->m_flags & 0x0000FFFF);
2978}
2979
2980static int receive_extralen(struct dlm_message *ms)
2981{
2982 return (ms->m_header.h_length - sizeof(struct dlm_message));
2983}
2984
e7fd4179
DT
2985static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb,
2986 struct dlm_message *ms)
2987{
2988 int len;
2989
2990 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
2991 if (!lkb->lkb_lvbptr)
52bda2b5 2992 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
e7fd4179
DT
2993 if (!lkb->lkb_lvbptr)
2994 return -ENOMEM;
2995 len = receive_extralen(ms);
2996 memcpy(lkb->lkb_lvbptr, ms->m_extra, len);
2997 }
2998 return 0;
2999}
3000
3001static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3002 struct dlm_message *ms)
3003{
3004 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3005 lkb->lkb_ownpid = ms->m_pid;
3006 lkb->lkb_remid = ms->m_lkid;
3007 lkb->lkb_grmode = DLM_LOCK_IV;
3008 lkb->lkb_rqmode = ms->m_rqmode;
3009 lkb->lkb_bastaddr = (void *) (long) (ms->m_asts & AST_BAST);
3010 lkb->lkb_astaddr = (void *) (long) (ms->m_asts & AST_COMP);
3011
8d07fd50
DT
3012 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
3013 /* lkb was just created so there won't be an lvb yet */
52bda2b5 3014 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
8d07fd50
DT
3015 if (!lkb->lkb_lvbptr)
3016 return -ENOMEM;
3017 }
e7fd4179
DT
3018
3019 return 0;
3020}
3021
3022static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3023 struct dlm_message *ms)
3024{
e7fd4179
DT
3025 if (lkb->lkb_status != DLM_LKSTS_GRANTED)
3026 return -EBUSY;
3027
e7fd4179
DT
3028 if (receive_lvb(ls, lkb, ms))
3029 return -ENOMEM;
3030
3031 lkb->lkb_rqmode = ms->m_rqmode;
3032 lkb->lkb_lvbseq = ms->m_lvbseq;
3033
3034 return 0;
3035}
3036
3037static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
3038 struct dlm_message *ms)
3039{
e7fd4179
DT
3040 if (receive_lvb(ls, lkb, ms))
3041 return -ENOMEM;
3042 return 0;
3043}
3044
3045/* We fill in the stub-lkb fields with the info that send_xxxx_reply()
3046 uses to send a reply and that the remote end uses to process the reply. */
3047
3048static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms)
3049{
3050 struct dlm_lkb *lkb = &ls->ls_stub_lkb;
3051 lkb->lkb_nodeid = ms->m_header.h_nodeid;
3052 lkb->lkb_remid = ms->m_lkid;
3053}
3054
c54e04b0
DT
3055/* This is called after the rsb is locked so that we can safely inspect
3056 fields in the lkb. */
3057
3058static int validate_message(struct dlm_lkb *lkb, struct dlm_message *ms)
3059{
3060 int from = ms->m_header.h_nodeid;
3061 int error = 0;
3062
3063 switch (ms->m_type) {
3064 case DLM_MSG_CONVERT:
3065 case DLM_MSG_UNLOCK:
3066 case DLM_MSG_CANCEL:
3067 if (!is_master_copy(lkb) || lkb->lkb_nodeid != from)
3068 error = -EINVAL;
3069 break;
3070
3071 case DLM_MSG_CONVERT_REPLY:
3072 case DLM_MSG_UNLOCK_REPLY:
3073 case DLM_MSG_CANCEL_REPLY:
3074 case DLM_MSG_GRANT:
3075 case DLM_MSG_BAST:
3076 if (!is_process_copy(lkb) || lkb->lkb_nodeid != from)
3077 error = -EINVAL;
3078 break;
3079
3080 case DLM_MSG_REQUEST_REPLY:
3081 if (!is_process_copy(lkb))
3082 error = -EINVAL;
3083 else if (lkb->lkb_nodeid != -1 && lkb->lkb_nodeid != from)
3084 error = -EINVAL;
3085 break;
3086
3087 default:
3088 error = -EINVAL;
3089 }
3090
3091 if (error)
3092 log_error(lkb->lkb_resource->res_ls,
3093 "ignore invalid message %d from %d %x %x %x %d",
3094 ms->m_type, from, lkb->lkb_id, lkb->lkb_remid,
3095 lkb->lkb_flags, lkb->lkb_nodeid);
3096 return error;
3097}
3098
e7fd4179
DT
3099static void receive_request(struct dlm_ls *ls, struct dlm_message *ms)
3100{
3101 struct dlm_lkb *lkb;
3102 struct dlm_rsb *r;
3103 int error, namelen;
3104
3105 error = create_lkb(ls, &lkb);
3106 if (error)
3107 goto fail;
3108
3109 receive_flags(lkb, ms);
3110 lkb->lkb_flags |= DLM_IFL_MSTCPY;
3111 error = receive_request_args(ls, lkb, ms);
3112 if (error) {
b3f58d8f 3113 __put_lkb(ls, lkb);
e7fd4179
DT
3114 goto fail;
3115 }
3116
3117 namelen = receive_extralen(ms);
3118
3119 error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r);
3120 if (error) {
b3f58d8f 3121 __put_lkb(ls, lkb);
e7fd4179
DT
3122 goto fail;
3123 }
3124
3125 lock_rsb(r);
3126
3127 attach_lkb(r, lkb);
3128 error = do_request(r, lkb);
3129 send_request_reply(r, lkb, error);
3130
3131 unlock_rsb(r);
3132 put_rsb(r);
3133
3134 if (error == -EINPROGRESS)
3135 error = 0;
3136 if (error)
b3f58d8f 3137 dlm_put_lkb(lkb);
e7fd4179
DT
3138 return;
3139
3140 fail:
3141 setup_stub_lkb(ls, ms);
3142 send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3143}
3144
3145static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms)
3146{
3147 struct dlm_lkb *lkb;
3148 struct dlm_rsb *r;
90135925 3149 int error, reply = 1;
e7fd4179
DT
3150
3151 error = find_lkb(ls, ms->m_remid, &lkb);
3152 if (error)
3153 goto fail;
3154
3155 r = lkb->lkb_resource;
3156
3157 hold_rsb(r);
3158 lock_rsb(r);
3159
c54e04b0
DT
3160 error = validate_message(lkb, ms);
3161 if (error)
3162 goto out;
3163
e7fd4179
DT
3164 receive_flags(lkb, ms);
3165 error = receive_convert_args(ls, lkb, ms);
3166 if (error)
c54e04b0 3167 goto out_reply;
e7fd4179
DT
3168 reply = !down_conversion(lkb);
3169
3170 error = do_convert(r, lkb);
c54e04b0 3171 out_reply:
e7fd4179
DT
3172 if (reply)
3173 send_convert_reply(r, lkb, error);
c54e04b0 3174 out:
e7fd4179
DT
3175 unlock_rsb(r);
3176 put_rsb(r);
b3f58d8f 3177 dlm_put_lkb(lkb);
e7fd4179
DT
3178 return;
3179
3180 fail:
3181 setup_stub_lkb(ls, ms);
3182 send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3183}
3184
3185static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms)
3186{
3187 struct dlm_lkb *lkb;
3188 struct dlm_rsb *r;
3189 int error;
3190
3191 error = find_lkb(ls, ms->m_remid, &lkb);
3192 if (error)
3193 goto fail;
3194
3195 r = lkb->lkb_resource;
3196
3197 hold_rsb(r);
3198 lock_rsb(r);
3199
c54e04b0
DT
3200 error = validate_message(lkb, ms);
3201 if (error)
3202 goto out;
3203
e7fd4179
DT
3204 receive_flags(lkb, ms);
3205 error = receive_unlock_args(ls, lkb, ms);
3206 if (error)
c54e04b0 3207 goto out_reply;
e7fd4179
DT
3208
3209 error = do_unlock(r, lkb);
c54e04b0 3210 out_reply:
e7fd4179 3211 send_unlock_reply(r, lkb, error);
c54e04b0 3212 out:
e7fd4179
DT
3213 unlock_rsb(r);
3214 put_rsb(r);
b3f58d8f 3215 dlm_put_lkb(lkb);
e7fd4179
DT
3216 return;
3217
3218 fail:
3219 setup_stub_lkb(ls, ms);
3220 send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3221}
3222
3223static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms)
3224{
3225 struct dlm_lkb *lkb;
3226 struct dlm_rsb *r;
3227 int error;
3228
3229 error = find_lkb(ls, ms->m_remid, &lkb);
3230 if (error)
3231 goto fail;
3232
3233 receive_flags(lkb, ms);
3234
3235 r = lkb->lkb_resource;
3236
3237 hold_rsb(r);
3238 lock_rsb(r);
3239
c54e04b0
DT
3240 error = validate_message(lkb, ms);
3241 if (error)
3242 goto out;
3243
e7fd4179
DT
3244 error = do_cancel(r, lkb);
3245 send_cancel_reply(r, lkb, error);
c54e04b0 3246 out:
e7fd4179
DT
3247 unlock_rsb(r);
3248 put_rsb(r);
b3f58d8f 3249 dlm_put_lkb(lkb);
e7fd4179
DT
3250 return;
3251
3252 fail:
3253 setup_stub_lkb(ls, ms);
3254 send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error);
3255}
3256
3257static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
3258{
3259 struct dlm_lkb *lkb;
3260 struct dlm_rsb *r;
3261 int error;
3262
3263 error = find_lkb(ls, ms->m_remid, &lkb);
3264 if (error) {
c54e04b0
DT
3265 log_debug(ls, "receive_grant from %d no lkb %x",
3266 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT
3267 return;
3268 }
e7fd4179
DT
3269
3270 r = lkb->lkb_resource;
3271
3272 hold_rsb(r);
3273 lock_rsb(r);
3274
c54e04b0
DT
3275 error = validate_message(lkb, ms);
3276 if (error)
3277 goto out;
3278
e7fd4179 3279 receive_flags_reply(lkb, ms);
7d3c1feb
DT
3280 if (is_altmode(lkb))
3281 munge_altmode(lkb, ms);
e7fd4179
DT
3282 grant_lock_pc(r, lkb, ms);
3283 queue_cast(r, lkb, 0);
c54e04b0 3284 out:
e7fd4179
DT
3285 unlock_rsb(r);
3286 put_rsb(r);
b3f58d8f 3287 dlm_put_lkb(lkb);
e7fd4179
DT
3288}
3289
3290static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms)
3291{
3292 struct dlm_lkb *lkb;
3293 struct dlm_rsb *r;
3294 int error;
3295
3296 error = find_lkb(ls, ms->m_remid, &lkb);
3297 if (error) {
c54e04b0
DT
3298 log_debug(ls, "receive_bast from %d no lkb %x",
3299 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT
3300 return;
3301 }
e7fd4179
DT
3302
3303 r = lkb->lkb_resource;
3304
3305 hold_rsb(r);
3306 lock_rsb(r);
3307
c54e04b0
DT
3308 error = validate_message(lkb, ms);
3309 if (error)
3310 goto out;
e7fd4179 3311
c54e04b0
DT
3312 queue_bast(r, lkb, ms->m_bastmode);
3313 out:
e7fd4179
DT
3314 unlock_rsb(r);
3315 put_rsb(r);
b3f58d8f 3316 dlm_put_lkb(lkb);
e7fd4179
DT
3317}
3318
3319static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms)
3320{
3321 int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid;
3322
3323 from_nodeid = ms->m_header.h_nodeid;
3324 our_nodeid = dlm_our_nodeid();
3325
3326 len = receive_extralen(ms);
3327
3328 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3329 if (dir_nodeid != our_nodeid) {
3330 log_error(ls, "lookup dir_nodeid %d from %d",
3331 dir_nodeid, from_nodeid);
3332 error = -EINVAL;
3333 ret_nodeid = -1;
3334 goto out;
3335 }
3336
3337 error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid);
3338
3339 /* Optimization: we're master so treat lookup as a request */
3340 if (!error && ret_nodeid == our_nodeid) {
3341 receive_request(ls, ms);
3342 return;
3343 }
3344 out:
3345 send_lookup_reply(ls, ms, ret_nodeid, error);
3346}
3347
3348static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
3349{
3350 int len, dir_nodeid, from_nodeid;
3351
3352 from_nodeid = ms->m_header.h_nodeid;
3353
3354 len = receive_extralen(ms);
3355
3356 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash);
3357 if (dir_nodeid != dlm_our_nodeid()) {
3358 log_error(ls, "remove dir entry dir_nodeid %d from %d",
3359 dir_nodeid, from_nodeid);
3360 return;
3361 }
3362
3363 dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
3364}
3365
8499137d
DT
3366static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
3367{
3368 do_purge(ls, ms->m_nodeid, ms->m_pid);
3369}
3370
e7fd4179
DT
3371static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
3372{
3373 struct dlm_lkb *lkb;
3374 struct dlm_rsb *r;
ef0c2bb0 3375 int error, mstype, result;
e7fd4179
DT
3376
3377 error = find_lkb(ls, ms->m_remid, &lkb);
3378 if (error) {
c54e04b0
DT
3379 log_debug(ls, "receive_request_reply from %d no lkb %x",
3380 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT
3381 return;
3382 }
e7fd4179 3383
e7fd4179
DT
3384 r = lkb->lkb_resource;
3385 hold_rsb(r);
3386 lock_rsb(r);
3387
c54e04b0
DT
3388 error = validate_message(lkb, ms);
3389 if (error)
3390 goto out;
3391
ef0c2bb0
DT
3392 mstype = lkb->lkb_wait_type;
3393 error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
3394 if (error)
3395 goto out;
3396
e7fd4179
DT
3397 /* Optimization: the dir node was also the master, so it took our
3398 lookup as a request and sent request reply instead of lookup reply */
3399 if (mstype == DLM_MSG_LOOKUP) {
3400 r->res_nodeid = ms->m_header.h_nodeid;
3401 lkb->lkb_nodeid = r->res_nodeid;
3402 }
3403
ef0c2bb0
DT
3404 /* this is the value returned from do_request() on the master */
3405 result = ms->m_result;
3406
3407 switch (result) {
e7fd4179 3408 case -EAGAIN:
ef0c2bb0 3409 /* request would block (be queued) on remote master */
e7fd4179
DT
3410 queue_cast(r, lkb, -EAGAIN);
3411 confirm_master(r, -EAGAIN);
ef0c2bb0 3412 unhold_lkb(lkb); /* undoes create_lkb() */
e7fd4179
DT
3413 break;
3414
3415 case -EINPROGRESS:
3416 case 0:
3417 /* request was queued or granted on remote master */
3418 receive_flags_reply(lkb, ms);
3419 lkb->lkb_remid = ms->m_lkid;
7d3c1feb
DT
3420 if (is_altmode(lkb))
3421 munge_altmode(lkb, ms);
3ae1acf9 3422 if (result) {
e7fd4179 3423 add_lkb(r, lkb, DLM_LKSTS_WAITING);
3ae1acf9
DT
3424 add_timeout(lkb);
3425 } else {
e7fd4179
DT
3426 grant_lock_pc(r, lkb, ms);
3427 queue_cast(r, lkb, 0);
3428 }
ef0c2bb0 3429 confirm_master(r, result);
e7fd4179
DT
3430 break;
3431
597d0cae 3432 case -EBADR:
e7fd4179
DT
3433 case -ENOTBLK:
3434 /* find_rsb failed to find rsb or rsb wasn't master */
ef0c2bb0
DT
3435 log_debug(ls, "receive_request_reply %x %x master diff %d %d",
3436 lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
e7fd4179
DT
3437 r->res_nodeid = -1;
3438 lkb->lkb_nodeid = -1;
ef0c2bb0
DT
3439
3440 if (is_overlap(lkb)) {
3441 /* we'll ignore error in cancel/unlock reply */
3442 queue_cast_overlap(r, lkb);
aec64e1b 3443 confirm_master(r, result);
ef0c2bb0
DT
3444 unhold_lkb(lkb); /* undoes create_lkb() */
3445 } else
3446 _request_lock(r, lkb);
e7fd4179
DT
3447 break;
3448
3449 default:
ef0c2bb0
DT
3450 log_error(ls, "receive_request_reply %x error %d",
3451 lkb->lkb_id, result);
e7fd4179
DT
3452 }
3453
ef0c2bb0
DT
3454 if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
3455 log_debug(ls, "receive_request_reply %x result %d unlock",
3456 lkb->lkb_id, result);
3457 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3458 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3459 send_unlock(r, lkb);
3460 } else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
3461 log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
3462 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3463 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3464 send_cancel(r, lkb);
3465 } else {
3466 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
3467 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
3468 }
3469 out:
e7fd4179
DT
3470 unlock_rsb(r);
3471 put_rsb(r);
b3f58d8f 3472 dlm_put_lkb(lkb);
e7fd4179
DT
3473}
3474
3475static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
3476 struct dlm_message *ms)
3477{
e7fd4179 3478 /* this is the value returned from do_convert() on the master */
ef0c2bb0 3479 switch (ms->m_result) {
e7fd4179
DT
3480 case -EAGAIN:
3481 /* convert would block (be queued) on remote master */
3482 queue_cast(r, lkb, -EAGAIN);
3483 break;
3484
c85d65e9
DT
3485 case -EDEADLK:
3486 receive_flags_reply(lkb, ms);
3487 revert_lock_pc(r, lkb);
3488 queue_cast(r, lkb, -EDEADLK);
3489 break;
3490
e7fd4179
DT
3491 case -EINPROGRESS:
3492 /* convert was queued on remote master */
7d3c1feb
DT
3493 receive_flags_reply(lkb, ms);
3494 if (is_demoted(lkb))
3495 munge_demoted(lkb, ms);
e7fd4179
DT
3496 del_lkb(r, lkb);
3497 add_lkb(r, lkb, DLM_LKSTS_CONVERT);
3ae1acf9 3498 add_timeout(lkb);
e7fd4179
DT
3499 break;
3500
3501 case 0:
3502 /* convert was granted on remote master */
3503 receive_flags_reply(lkb, ms);
7d3c1feb
DT
3504 if (is_demoted(lkb))
3505 munge_demoted(lkb, ms);
e7fd4179
DT
3506 grant_lock_pc(r, lkb, ms);
3507 queue_cast(r, lkb, 0);
3508 break;
3509
3510 default:
ef0c2bb0
DT
3511 log_error(r->res_ls, "receive_convert_reply %x error %d",
3512 lkb->lkb_id, ms->m_result);
e7fd4179
DT
3513 }
3514}
3515
3516static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3517{
3518 struct dlm_rsb *r = lkb->lkb_resource;
ef0c2bb0 3519 int error;
e7fd4179
DT
3520
3521 hold_rsb(r);
3522 lock_rsb(r);
3523
c54e04b0
DT
3524 error = validate_message(lkb, ms);
3525 if (error)
3526 goto out;
3527
ef0c2bb0
DT
3528 /* stub reply can happen with waiters_mutex held */
3529 error = remove_from_waiters_ms(lkb, ms);
3530 if (error)
3531 goto out;
e7fd4179 3532
ef0c2bb0
DT
3533 __receive_convert_reply(r, lkb, ms);
3534 out:
e7fd4179
DT
3535 unlock_rsb(r);
3536 put_rsb(r);
3537}
3538
3539static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
3540{
3541 struct dlm_lkb *lkb;
3542 int error;
3543
3544 error = find_lkb(ls, ms->m_remid, &lkb);
3545 if (error) {
c54e04b0
DT
3546 log_debug(ls, "receive_convert_reply from %d no lkb %x",
3547 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT
3548 return;
3549 }
e7fd4179 3550
e7fd4179 3551 _receive_convert_reply(lkb, ms);
b3f58d8f 3552 dlm_put_lkb(lkb);
e7fd4179
DT
3553}
3554
3555static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3556{
3557 struct dlm_rsb *r = lkb->lkb_resource;
ef0c2bb0 3558 int error;
e7fd4179
DT
3559
3560 hold_rsb(r);
3561 lock_rsb(r);
3562
c54e04b0
DT
3563 error = validate_message(lkb, ms);
3564 if (error)
3565 goto out;
3566
ef0c2bb0
DT
3567 /* stub reply can happen with waiters_mutex held */
3568 error = remove_from_waiters_ms(lkb, ms);
3569 if (error)
3570 goto out;
3571
e7fd4179
DT
3572 /* this is the value returned from do_unlock() on the master */
3573
ef0c2bb0 3574 switch (ms->m_result) {
e7fd4179
DT
3575 case -DLM_EUNLOCK:
3576 receive_flags_reply(lkb, ms);
3577 remove_lock_pc(r, lkb);
3578 queue_cast(r, lkb, -DLM_EUNLOCK);
3579 break;
ef0c2bb0
DT
3580 case -ENOENT:
3581 break;
e7fd4179 3582 default:
ef0c2bb0
DT
3583 log_error(r->res_ls, "receive_unlock_reply %x error %d",
3584 lkb->lkb_id, ms->m_result);
e7fd4179 3585 }
ef0c2bb0 3586 out:
e7fd4179
DT
3587 unlock_rsb(r);
3588 put_rsb(r);
3589}
3590
3591static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
3592{
3593 struct dlm_lkb *lkb;
3594 int error;
3595
3596 error = find_lkb(ls, ms->m_remid, &lkb);
3597 if (error) {
c54e04b0
DT
3598 log_debug(ls, "receive_unlock_reply from %d no lkb %x",
3599 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT
3600 return;
3601 }
e7fd4179 3602
e7fd4179 3603 _receive_unlock_reply(lkb, ms);
b3f58d8f 3604 dlm_put_lkb(lkb);
e7fd4179
DT
3605}
3606
3607static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
3608{
3609 struct dlm_rsb *r = lkb->lkb_resource;
ef0c2bb0 3610 int error;
e7fd4179
DT
3611
3612 hold_rsb(r);
3613 lock_rsb(r);
3614
c54e04b0
DT
3615 error = validate_message(lkb, ms);
3616 if (error)
3617 goto out;
3618
ef0c2bb0
DT
3619 /* stub reply can happen with waiters_mutex held */
3620 error = remove_from_waiters_ms(lkb, ms);
3621 if (error)
3622 goto out;
3623
e7fd4179
DT
3624 /* this is the value returned from do_cancel() on the master */
3625
ef0c2bb0 3626 switch (ms->m_result) {
e7fd4179
DT
3627 case -DLM_ECANCEL:
3628 receive_flags_reply(lkb, ms);
3629 revert_lock_pc(r, lkb);
84d8cd69 3630 queue_cast(r, lkb, -DLM_ECANCEL);
ef0c2bb0
DT
3631 break;
3632 case 0:
e7fd4179
DT
3633 break;
3634 default:
ef0c2bb0
DT
3635 log_error(r->res_ls, "receive_cancel_reply %x error %d",
3636 lkb->lkb_id, ms->m_result);
e7fd4179 3637 }
ef0c2bb0 3638 out:
e7fd4179
DT
3639 unlock_rsb(r);
3640 put_rsb(r);
3641}
3642
3643static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
3644{
3645 struct dlm_lkb *lkb;
3646 int error;
3647
3648 error = find_lkb(ls, ms->m_remid, &lkb);
3649 if (error) {
c54e04b0
DT
3650 log_debug(ls, "receive_cancel_reply from %d no lkb %x",
3651 ms->m_header.h_nodeid, ms->m_remid);
e7fd4179
DT
3652 return;
3653 }
e7fd4179 3654
e7fd4179 3655 _receive_cancel_reply(lkb, ms);
b3f58d8f 3656 dlm_put_lkb(lkb);
e7fd4179
DT
3657}
3658
3659static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
3660{
3661 struct dlm_lkb *lkb;
3662 struct dlm_rsb *r;
3663 int error, ret_nodeid;
3664
3665 error = find_lkb(ls, ms->m_lkid, &lkb);
3666 if (error) {
3667 log_error(ls, "receive_lookup_reply no lkb");
3668 return;
3669 }
3670
ef0c2bb0 3671 /* ms->m_result is the value returned by dlm_dir_lookup on dir node
e7fd4179 3672 FIXME: will a non-zero error ever be returned? */
e7fd4179
DT
3673
3674 r = lkb->lkb_resource;
3675 hold_rsb(r);
3676 lock_rsb(r);
3677
ef0c2bb0
DT
3678 error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
3679 if (error)
3680 goto out;
3681
e7fd4179
DT
3682 ret_nodeid = ms->m_nodeid;
3683 if (ret_nodeid == dlm_our_nodeid()) {
3684 r->res_nodeid = 0;
3685 ret_nodeid = 0;
3686 r->res_first_lkid = 0;
3687 } else {
3688 /* set_master() will copy res_nodeid to lkb_nodeid */
3689 r->res_nodeid = ret_nodeid;
3690 }
3691
ef0c2bb0
DT
3692 if (is_overlap(lkb)) {
3693 log_debug(ls, "receive_lookup_reply %x unlock %x",
3694 lkb->lkb_id, lkb->lkb_flags);
3695 queue_cast_overlap(r, lkb);
3696 unhold_lkb(lkb); /* undoes create_lkb() */
3697 goto out_list;
3698 }
3699
e7fd4179
DT
3700 _request_lock(r, lkb);
3701
ef0c2bb0 3702 out_list:
e7fd4179
DT
3703 if (!ret_nodeid)
3704 process_lookup_list(r);
ef0c2bb0 3705 out:
e7fd4179
DT
3706 unlock_rsb(r);
3707 put_rsb(r);
b3f58d8f 3708 dlm_put_lkb(lkb);
e7fd4179
DT
3709}
3710
c36258b5 3711static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
e7fd4179 3712{
46b43eed
DT
3713 if (!dlm_is_member(ls, ms->m_header.h_nodeid)) {
3714 log_debug(ls, "ignore non-member message %d from %d %x %x %d",
3715 ms->m_type, ms->m_header.h_nodeid, ms->m_lkid,
3716 ms->m_remid, ms->m_result);
3717 return;
3718 }
3719
e7fd4179
DT
3720 switch (ms->m_type) {
3721
3722 /* messages sent to a master node */
3723
3724 case DLM_MSG_REQUEST:
3725 receive_request(ls, ms);
3726 break;
3727
3728 case DLM_MSG_CONVERT:
3729 receive_convert(ls, ms);
3730 break;
3731
3732 case DLM_MSG_UNLOCK:
3733 receive_unlock(ls, ms);
3734 break;
3735
3736 case DLM_MSG_CANCEL:
3737 receive_cancel(ls, ms);
3738 break;
3739
3740 /* messages sent from a master node (replies to above) */
3741
3742 case DLM_MSG_REQUEST_REPLY:
3743 receive_request_reply(ls, ms);
3744 break;
3745
3746 case DLM_MSG_CONVERT_REPLY:
3747 receive_convert_reply(ls, ms);
3748 break;
3749
3750 case DLM_MSG_UNLOCK_REPLY:
3751 receive_unlock_reply(ls, ms);
3752 break;
3753
3754 case DLM_MSG_CANCEL_REPLY:
3755 receive_cancel_reply(ls, ms);
3756 break;
3757
3758 /* messages sent from a master node (only two types of async msg) */
3759
3760 case DLM_MSG_GRANT:
3761 receive_grant(ls, ms);
3762 break;
3763
3764 case DLM_MSG_BAST:
3765 receive_bast(ls, ms);
3766 break;
3767
3768 /* messages sent to a dir node */
3769
3770 case DLM_MSG_LOOKUP:
3771 receive_lookup(ls, ms);
3772 break;
3773
3774 case DLM_MSG_REMOVE:
3775 receive_remove(ls, ms);
3776 break;
3777
3778 /* messages sent from a dir node (remove has no reply) */
3779
3780 case DLM_MSG_LOOKUP_REPLY:
3781 receive_lookup_reply(ls, ms);
3782 break;
3783
8499137d
DT
3784 /* other messages */
3785
3786 case DLM_MSG_PURGE:
3787 receive_purge(ls, ms);
3788 break;
3789
e7fd4179
DT
3790 default:
3791 log_error(ls, "unknown message type %d", ms->m_type);
3792 }
3793
e7fd4179 3794 dlm_astd_wake();
e7fd4179
DT
3795}
3796
c36258b5
DT
3797/* If the lockspace is in recovery mode (locking stopped), then normal
3798 messages are saved on the requestqueue for processing after recovery is
3799 done. When not in recovery mode, we wait for dlm_recoverd to drain saved
3800 messages off the requestqueue before we process new ones. This occurs right
3801 after recovery completes when we transition from saving all messages on
3802 requestqueue, to processing all the saved messages, to processing new
3803 messages as they arrive. */
e7fd4179 3804
c36258b5
DT
3805static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
3806 int nodeid)
3807{
3808 if (dlm_locking_stopped(ls)) {
3809 dlm_add_requestqueue(ls, nodeid, (struct dlm_header *) ms);
3810 } else {
3811 dlm_wait_requestqueue(ls);
3812 _receive_message(ls, ms);
3813 }
3814}
3815
3816/* This is called by dlm_recoverd to process messages that were saved on
3817 the requestqueue. */
3818
3819void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
3820{
3821 _receive_message(ls, ms);
3822}
3823
3824/* This is called by the midcomms layer when something is received for
3825 the lockspace. It could be either a MSG (normal message sent as part of
3826 standard locking activity) or an RCOM (recovery message sent as part of
3827 lockspace recovery). */
3828
3829void dlm_receive_buffer(struct dlm_header *hd, int nodeid)
3830{
3831 struct dlm_message *ms = (struct dlm_message *) hd;
3832 struct dlm_rcom *rc = (struct dlm_rcom *) hd;
3833 struct dlm_ls *ls;
3834 int type = 0;
3835
3836 switch (hd->h_cmd) {
3837 case DLM_MSG:
3838 dlm_message_in(ms);
3839 type = ms->m_type;
3840 break;
3841 case DLM_RCOM:
3842 dlm_rcom_in(rc);
3843 type = rc->rc_type;
3844 break;
3845 default:
3846 log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
3847 return;
3848 }
3849
3850 if (hd->h_nodeid != nodeid) {
3851 log_print("invalid h_nodeid %d from %d lockspace %x",
3852 hd->h_nodeid, nodeid, hd->h_lockspace);
3853 return;
3854 }
3855
3856 ls = dlm_find_lockspace_global(hd->h_lockspace);
3857 if (!ls) {
3858 log_print("invalid h_lockspace %x from %d cmd %d type %d",
3859 hd->h_lockspace, nodeid, hd->h_cmd, type);
3860
3861 if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
3862 dlm_send_ls_not_ready(nodeid, rc);
3863 return;
3864 }
3865
3866 /* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
3867 be inactive (in this ls) before transitioning to recovery mode */
3868
3869 down_read(&ls->ls_recv_active);
3870 if (hd->h_cmd == DLM_MSG)
3871 dlm_receive_message(ls, ms, nodeid);
3872 else
3873 dlm_receive_rcom(ls, rc, nodeid);
3874 up_read(&ls->ls_recv_active);
3875
3876 dlm_put_lockspace(ls);
3877}
e7fd4179
DT
3878
3879static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
3880{
3881 if (middle_conversion(lkb)) {
3882 hold_lkb(lkb);
ef0c2bb0 3883 ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
e7fd4179 3884 ls->ls_stub_ms.m_result = -EINPROGRESS;
075529b5 3885 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
c54e04b0 3886 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
e7fd4179
DT
3887 _receive_convert_reply(lkb, &ls->ls_stub_ms);
3888
3889 /* Same special case as in receive_rcom_lock_args() */
3890 lkb->lkb_grmode = DLM_LOCK_IV;
3891 rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT);
3892 unhold_lkb(lkb);
3893
3894 } else if (lkb->lkb_rqmode >= lkb->lkb_grmode) {
3895 lkb->lkb_flags |= DLM_IFL_RESEND;
3896 }
3897
3898 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down
3899 conversions are async; there's no reply from the remote master */
3900}
3901
3902/* A waiting lkb needs recovery if the master node has failed, or
3903 the master node is changing (only when no directory is used) */
3904
3905static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb)
3906{
3907 if (dlm_is_removed(ls, lkb->lkb_nodeid))
3908 return 1;
3909
3910 if (!dlm_no_directory(ls))
3911 return 0;
3912
3913 if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid)
3914 return 1;
3915
3916 return 0;
3917}
3918
3919/* Recovery for locks that are waiting for replies from nodes that are now
3920 gone. We can just complete unlocks and cancels by faking a reply from the
3921 dead node. Requests and up-conversions we flag to be resent after
3922 recovery. Down-conversions can just be completed with a fake reply like
3923 unlocks. Conversions between PR and CW need special attention. */
3924
3925void dlm_recover_waiters_pre(struct dlm_ls *ls)
3926{
3927 struct dlm_lkb *lkb, *safe;
601342ce 3928 int wait_type, stub_unlock_result, stub_cancel_result;
e7fd4179 3929
90135925 3930 mutex_lock(&ls->ls_waiters_mutex);
e7fd4179
DT
3931
3932 list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
3933 log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
3934 lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
3935
3936 /* all outstanding lookups, regardless of destination will be
3937 resent after recovery is done */
3938
3939 if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) {
3940 lkb->lkb_flags |= DLM_IFL_RESEND;
3941 continue;
3942 }
3943
3944 if (!waiter_needs_recovery(ls, lkb))
3945 continue;
3946
601342ce
DT
3947 wait_type = lkb->lkb_wait_type;
3948 stub_unlock_result = -DLM_EUNLOCK;
3949 stub_cancel_result = -DLM_ECANCEL;
3950
3951 /* Main reply may have been received leaving a zero wait_type,
3952 but a reply for the overlapping op may not have been
3953 received. In that case we need to fake the appropriate
3954 reply for the overlap op. */
3955
3956 if (!wait_type) {
3957 if (is_overlap_cancel(lkb)) {
3958 wait_type = DLM_MSG_CANCEL;
3959 if (lkb->lkb_grmode == DLM_LOCK_IV)
3960 stub_cancel_result = 0;
3961 }
3962 if (is_overlap_unlock(lkb)) {
3963 wait_type = DLM_MSG_UNLOCK;
3964 if (lkb->lkb_grmode == DLM_LOCK_IV)
3965 stub_unlock_result = -ENOENT;
3966 }
3967
3968 log_debug(ls, "rwpre overlap %x %x %d %d %d",
3969 lkb->lkb_id, lkb->lkb_flags, wait_type,
3970 stub_cancel_result, stub_unlock_result);
3971 }
3972
3973 switch (wait_type) {
e7fd4179
DT
3974
3975 case DLM_MSG_REQUEST:
3976 lkb->lkb_flags |= DLM_IFL_RESEND;
3977 break;
3978
3979 case DLM_MSG_CONVERT:
3980 recover_convert_waiter(ls, lkb);
3981 break;
3982
3983 case DLM_MSG_UNLOCK:
3984 hold_lkb(lkb);
ef0c2bb0 3985 ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
601342ce 3986 ls->ls_stub_ms.m_result = stub_unlock_result;
075529b5 3987 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
c54e04b0 3988 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
e7fd4179 3989 _receive_unlock_reply(lkb, &ls->ls_stub_ms);
b3f58d8f 3990 dlm_put_lkb(lkb);
e7fd4179
DT
3991 break;
3992
3993 case DLM_MSG_CANCEL:
3994 hold_lkb(lkb);
ef0c2bb0 3995 ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
601342ce 3996 ls->ls_stub_ms.m_result = stub_cancel_result;
075529b5 3997 ls->ls_stub_ms.m_flags = lkb->lkb_flags;
c54e04b0 3998 ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
e7fd4179 3999 _receive_cancel_reply(lkb, &ls->ls_stub_ms);
b3f58d8f 4000 dlm_put_lkb(lkb);
e7fd4179
DT
4001 break;
4002
4003 default:
601342ce
DT
4004 log_error(ls, "invalid lkb wait_type %d %d",
4005 lkb->lkb_wait_type, wait_type);
e7fd4179 4006 }
81456807 4007 schedule();
e7fd4179 4008 }
90135925 4009 mutex_unlock(&ls->ls_waiters_mutex);
e7fd4179
DT
4010}
4011
ef0c2bb0 4012static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
e7fd4179
DT
4013{
4014 struct dlm_lkb *lkb;
ef0c2bb0 4015 int found = 0;
e7fd4179 4016
90135925 4017 mutex_lock(&ls->ls_waiters_mutex);
e7fd4179
DT
4018 list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
4019 if (lkb->lkb_flags & DLM_IFL_RESEND) {
ef0c2bb0
DT
4020 hold_lkb(lkb);
4021 found = 1;
e7fd4179
DT
4022 break;
4023 }
4024 }
90135925 4025 mutex_unlock(&ls->ls_waiters_mutex);
e7fd4179 4026
ef0c2bb0 4027 if (!found)
e7fd4179 4028 lkb = NULL;
ef0c2bb0 4029 return lkb;
e7fd4179
DT
4030}
4031
4032/* Deal with lookups and lkb's marked RESEND from _pre. We may now be the
4033 master or dir-node for r. Processing the lkb may result in it being placed
4034 back on waiters. */
4035
ef0c2bb0
DT
4036/* We do this after normal locking has been enabled and any saved messages
4037 (in requestqueue) have been processed. We should be confident that at
4038 this point we won't get or process a reply to any of these waiting
4039 operations. But, new ops may be coming in on the rsbs/locks here from
4040 userspace or remotely. */
4041
4042/* there may have been an overlap unlock/cancel prior to recovery or after
4043 recovery. if before, the lkb may still have a pos wait_count; if after, the
4044 overlap flag would just have been set and nothing new sent. we can be
4045 confident here than any replies to either the initial op or overlap ops
4046 prior to recovery have been received. */
4047
e7fd4179
DT
4048int dlm_recover_waiters_post(struct dlm_ls *ls)
4049{
4050 struct dlm_lkb *lkb;
4051 struct dlm_rsb *r;
ef0c2bb0 4052 int error = 0, mstype, err, oc, ou;
e7fd4179
DT
4053
4054 while (1) {
4055 if (dlm_locking_stopped(ls)) {
4056 log_debug(ls, "recover_waiters_post aborted");
4057 error = -EINTR;
4058 break;
4059 }
4060
ef0c2bb0
DT
4061 lkb = find_resend_waiter(ls);
4062 if (!lkb)
e7fd4179
DT
4063 break;
4064
4065 r = lkb->lkb_resource;
ef0c2bb0
DT
4066 hold_rsb(r);
4067 lock_rsb(r);
4068
4069 mstype = lkb->lkb_wait_type;
4070 oc = is_overlap_cancel(lkb);
4071 ou = is_overlap_unlock(lkb);
4072 err = 0;
e7fd4179
DT
4073
4074 log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
4075 lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
4076
ef0c2bb0
DT
4077 /* At this point we assume that we won't get a reply to any
4078 previous op or overlap op on this lock. First, do a big
4079 remove_from_waiters() for all previous ops. */
4080
4081 lkb->lkb_flags &= ~DLM_IFL_RESEND;
4082 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
4083 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
4084 lkb->lkb_wait_type = 0;
4085 lkb->lkb_wait_count = 0;
4086 mutex_lock(&ls->ls_waiters_mutex);
4087 list_del_init(&lkb->lkb_wait_reply);
4088 mutex_unlock(&ls->ls_waiters_mutex);
4089 unhold_lkb(lkb); /* for waiters list */
4090
4091 if (oc || ou) {
4092 /* do an unlock or cancel instead of resending */
4093 switch (mstype) {
4094 case DLM_MSG_LOOKUP:
4095 case DLM_MSG_REQUEST:
4096 queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
4097 -DLM_ECANCEL);
4098 unhold_lkb(lkb); /* undoes create_lkb() */
4099 break;
4100 case DLM_MSG_CONVERT:
4101 if (oc) {
4102 queue_cast(r, lkb, -DLM_ECANCEL);
4103 } else {
4104 lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
4105 _unlock_lock(r, lkb);
4106 }
4107 break;
4108 default:
4109 err = 1;
4110 }
4111 } else {
4112 switch (mstype) {
4113 case DLM_MSG_LOOKUP:
4114 case DLM_MSG_REQUEST:
4115 _request_lock(r, lkb);
4116 if (is_master(r))
4117 confirm_master(r, 0);
4118 break;
4119 case DLM_MSG_CONVERT:
4120 _convert_lock(r, lkb);
4121 break;
4122 default:
4123 err = 1;
4124 }
e7fd4179 4125 }
ef0c2bb0
DT
4126
4127 if (err)
4128 log_error(ls, "recover_waiters_post %x %d %x %d %d",
4129 lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
4130 unlock_rsb(r);
4131 put_rsb(r);
4132 dlm_put_lkb(lkb);
e7fd4179
DT
4133 }
4134
4135 return error;
4136}
4137
4138static void purge_queue(struct dlm_rsb *r, struct list_head *queue,
4139 int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb))
4140{
4141 struct dlm_ls *ls = r->res_ls;
4142 struct dlm_lkb *lkb, *safe;
4143
4144 list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) {
4145 if (test(ls, lkb)) {
97a35d1e 4146 rsb_set_flag(r, RSB_LOCKS_PURGED);
e7fd4179
DT
4147 del_lkb(r, lkb);
4148 /* this put should free the lkb */
b3f58d8f 4149 if (!dlm_put_lkb(lkb))
e7fd4179
DT
4150 log_error(ls, "purged lkb not released");
4151 }
4152 }
4153}
4154
4155static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4156{
4157 return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid));
4158}
4159
4160static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb)
4161{
4162 return is_master_copy(lkb);
4163}
4164
4165static void purge_dead_locks(struct dlm_rsb *r)
4166{
4167 purge_queue(r, &r->res_grantqueue, &purge_dead_test);
4168 purge_queue(r, &r->res_convertqueue, &purge_dead_test);
4169 purge_queue(r, &r->res_waitqueue, &purge_dead_test);
4170}
4171
4172void dlm_purge_mstcpy_locks(struct dlm_rsb *r)
4173{
4174 purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test);
4175 purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test);
4176 purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test);
4177}
4178
4179/* Get rid of locks held by nodes that are gone. */
4180
4181int dlm_purge_locks(struct dlm_ls *ls)
4182{
4183 struct dlm_rsb *r;
4184
4185 log_debug(ls, "dlm_purge_locks");
4186
4187 down_write(&ls->ls_root_sem);
4188 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
4189 hold_rsb(r);
4190 lock_rsb(r);
4191 if (is_master(r))
4192 purge_dead_locks(r);
4193 unlock_rsb(r);
4194 unhold_rsb(r);
4195
4196 schedule();
4197 }
4198 up_write(&ls->ls_root_sem);
4199
4200 return 0;
4201}
4202
97a35d1e
DT
4203static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket)
4204{
4205 struct dlm_rsb *r, *r_ret = NULL;
4206
4207 read_lock(&ls->ls_rsbtbl[bucket].lock);
4208 list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) {
4209 if (!rsb_flag(r, RSB_LOCKS_PURGED))
4210 continue;
4211 hold_rsb(r);
4212 rsb_clear_flag(r, RSB_LOCKS_PURGED);
4213 r_ret = r;
4214 break;
4215 }
4216 read_unlock(&ls->ls_rsbtbl[bucket].lock);
4217 return r_ret;
4218}
4219
4220void dlm_grant_after_purge(struct dlm_ls *ls)
e7fd4179
DT
4221{
4222 struct dlm_rsb *r;
2b4e926a 4223 int bucket = 0;
e7fd4179 4224
2b4e926a
DT
4225 while (1) {
4226 r = find_purged_rsb(ls, bucket);
4227 if (!r) {
4228 if (bucket == ls->ls_rsbtbl_size - 1)
4229 break;
4230 bucket++;
97a35d1e 4231 continue;
2b4e926a 4232 }
97a35d1e
DT
4233 lock_rsb(r);
4234 if (is_master(r)) {
4235 grant_pending_locks(r);
4236 confirm_master(r, 0);
e7fd4179 4237 }
97a35d1e
DT
4238 unlock_rsb(r);
4239 put_rsb(r);
2b4e926a 4240 schedule();
e7fd4179 4241 }
e7fd4179
DT
4242}
4243
4244static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid,
4245 uint32_t remid)
4246{
4247 struct dlm_lkb *lkb;
4248
4249 list_for_each_entry(lkb, head, lkb_statequeue) {
4250 if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid)
4251 return lkb;
4252 }
4253 return NULL;
4254}
4255
4256static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid,
4257 uint32_t remid)
4258{
4259 struct dlm_lkb *lkb;
4260
4261 lkb = search_remid_list(&r->res_grantqueue, nodeid, remid);
4262 if (lkb)
4263 return lkb;
4264 lkb = search_remid_list(&r->res_convertqueue, nodeid, remid);
4265 if (lkb)
4266 return lkb;
4267 lkb = search_remid_list(&r->res_waitqueue, nodeid, remid);
4268 if (lkb)
4269 return lkb;
4270 return NULL;
4271}
4272
4273static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
4274 struct dlm_rsb *r, struct dlm_rcom *rc)
4275{
4276 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4277 int lvblen;
4278
4279 lkb->lkb_nodeid = rc->rc_header.h_nodeid;
4280 lkb->lkb_ownpid = rl->rl_ownpid;
4281 lkb->lkb_remid = rl->rl_lkid;
4282 lkb->lkb_exflags = rl->rl_exflags;
4283 lkb->lkb_flags = rl->rl_flags & 0x0000FFFF;
4284 lkb->lkb_flags |= DLM_IFL_MSTCPY;
4285 lkb->lkb_lvbseq = rl->rl_lvbseq;
4286 lkb->lkb_rqmode = rl->rl_rqmode;
4287 lkb->lkb_grmode = rl->rl_grmode;
4288 /* don't set lkb_status because add_lkb wants to itself */
4289
4290 lkb->lkb_bastaddr = (void *) (long) (rl->rl_asts & AST_BAST);
4291 lkb->lkb_astaddr = (void *) (long) (rl->rl_asts & AST_COMP);
4292
e7fd4179 4293 if (lkb->lkb_exflags & DLM_LKF_VALBLK) {
52bda2b5 4294 lkb->lkb_lvbptr = dlm_allocate_lvb(ls);
e7fd4179
DT
4295 if (!lkb->lkb_lvbptr)
4296 return -ENOMEM;
4297 lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) -
4298 sizeof(struct rcom_lock);
4299 memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen);
4300 }
4301
4302 /* Conversions between PR and CW (middle modes) need special handling.
4303 The real granted mode of these converting locks cannot be determined
4304 until all locks have been rebuilt on the rsb (recover_conversion) */
4305
4306 if (rl->rl_wait_type == DLM_MSG_CONVERT && middle_conversion(lkb)) {
4307 rl->rl_status = DLM_LKSTS_CONVERT;
4308 lkb->lkb_grmode = DLM_LOCK_IV;
4309 rsb_set_flag(r, RSB_RECOVER_CONVERT);
4310 }
4311
4312 return 0;
4313}
4314
4315/* This lkb may have been recovered in a previous aborted recovery so we need
4316 to check if the rsb already has an lkb with the given remote nodeid/lkid.
4317 If so we just send back a standard reply. If not, we create a new lkb with
4318 the given values and send back our lkid. We send back our lkid by sending
4319 back the rcom_lock struct we got but with the remid field filled in. */
4320
4321int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4322{
4323 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4324 struct dlm_rsb *r;
4325 struct dlm_lkb *lkb;
4326 int error;
4327
4328 if (rl->rl_parent_lkid) {
4329 error = -EOPNOTSUPP;
4330 goto out;
4331 }
4332
4333 error = find_rsb(ls, rl->rl_name, rl->rl_namelen, R_MASTER, &r);
4334 if (error)
4335 goto out;
4336
4337 lock_rsb(r);
4338
4339 lkb = search_remid(r, rc->rc_header.h_nodeid, rl->rl_lkid);
4340 if (lkb) {
4341 error = -EEXIST;
4342 goto out_remid;
4343 }
4344
4345 error = create_lkb(ls, &lkb);
4346 if (error)
4347 goto out_unlock;
4348
4349 error = receive_rcom_lock_args(ls, lkb, r, rc);
4350 if (error) {
b3f58d8f 4351 __put_lkb(ls, lkb);
e7fd4179
DT
4352 goto out_unlock;
4353 }
4354
4355 attach_lkb(r, lkb);
4356 add_lkb(r, lkb, rl->rl_status);
4357 error = 0;
4358
4359 out_remid:
4360 /* this is the new value returned to the lock holder for
4361 saving in its process-copy lkb */
4362 rl->rl_remid = lkb->lkb_id;
4363
4364 out_unlock:
4365 unlock_rsb(r);
4366 put_rsb(r);
4367 out:
4368 if (error)
11b2498b 4369 log_debug(ls, "recover_master_copy %d %x", error, rl->rl_lkid);
e7fd4179
DT
4370 rl->rl_result = error;
4371 return error;
4372}
4373
4374int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc)
4375{
4376 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf;
4377 struct dlm_rsb *r;
4378 struct dlm_lkb *lkb;
4379 int error;
4380
4381 error = find_lkb(ls, rl->rl_lkid, &lkb);
4382 if (error) {
4383 log_error(ls, "recover_process_copy no lkid %x", rl->rl_lkid);
4384 return error;
4385 }
4386
4387 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
4388
4389 error = rl->rl_result;
4390
4391 r = lkb->lkb_resource;
4392 hold_rsb(r);
4393 lock_rsb(r);
4394
4395 switch (error) {
dc200a88
DT
4396 case -EBADR:
4397 /* There's a chance the new master received our lock before
4398 dlm_recover_master_reply(), this wouldn't happen if we did
4399 a barrier between recover_masters and recover_locks. */
4400 log_debug(ls, "master copy not ready %x r %lx %s", lkb->lkb_id,
4401 (unsigned long)r, r->res_name);
4402 dlm_send_rcom_lock(r, lkb);
4403 goto out;
e7fd4179
DT
4404 case -EEXIST:
4405 log_debug(ls, "master copy exists %x", lkb->lkb_id);
4406 /* fall through */
4407 case 0:
4408 lkb->lkb_remid = rl->rl_remid;
4409 break;
4410 default:
4411 log_error(ls, "dlm_recover_process_copy unknown error %d %x",
4412 error, lkb->lkb_id);
4413 }
4414
4415 /* an ack for dlm_recover_locks() which waits for replies from
4416 all the locks it sends to new masters */
4417 dlm_recovered_lock(r);
dc200a88 4418 out:
e7fd4179
DT
4419 unlock_rsb(r);
4420 put_rsb(r);
b3f58d8f 4421 dlm_put_lkb(lkb);
e7fd4179
DT
4422
4423 return 0;
4424}
4425
597d0cae
DT
4426int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
4427 int mode, uint32_t flags, void *name, unsigned int namelen,
d7db923e 4428 unsigned long timeout_cs)
597d0cae
DT
4429{
4430 struct dlm_lkb *lkb;
4431 struct dlm_args args;
4432 int error;
4433
85e86edf 4434 dlm_lock_recovery(ls);
597d0cae
DT
4435
4436 error = create_lkb(ls, &lkb);
4437 if (error) {
4438 kfree(ua);
4439 goto out;
4440 }
4441
4442 if (flags & DLM_LKF_VALBLK) {
62a0f623 4443 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
597d0cae
DT
4444 if (!ua->lksb.sb_lvbptr) {
4445 kfree(ua);
4446 __put_lkb(ls, lkb);
4447 error = -ENOMEM;
4448 goto out;
4449 }
4450 }
4451
52bda2b5 4452 /* After ua is attached to lkb it will be freed by dlm_free_lkb().
597d0cae
DT
4453 When DLM_IFL_USER is set, the dlm knows that this is a userspace
4454 lock and that lkb_astparam is the dlm_user_args structure. */
4455
d7db923e 4456 error = set_lock_args(mode, &ua->lksb, flags, namelen, timeout_cs,
32f105a1 4457 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
597d0cae
DT
4458 lkb->lkb_flags |= DLM_IFL_USER;
4459 ua->old_mode = DLM_LOCK_IV;
4460
4461 if (error) {
4462 __put_lkb(ls, lkb);
4463 goto out;
4464 }
4465
4466 error = request_lock(ls, lkb, name, namelen, &args);
4467
4468 switch (error) {
4469 case 0:
4470 break;
4471 case -EINPROGRESS:
4472 error = 0;
4473 break;
4474 case -EAGAIN:
4475 error = 0;
4476 /* fall through */
4477 default:
4478 __put_lkb(ls, lkb);
4479 goto out;
4480 }
4481
4482 /* add this new lkb to the per-process list of locks */
4483 spin_lock(&ua->proc->locks_spin);
ef0c2bb0 4484 hold_lkb(lkb);
597d0cae
DT
4485 list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
4486 spin_unlock(&ua->proc->locks_spin);
4487 out:
85e86edf 4488 dlm_unlock_recovery(ls);
597d0cae
DT
4489 return error;
4490}
4491
4492int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
d7db923e
DT
4493 int mode, uint32_t flags, uint32_t lkid, char *lvb_in,
4494 unsigned long timeout_cs)
597d0cae
DT
4495{
4496 struct dlm_lkb *lkb;
4497 struct dlm_args args;
4498 struct dlm_user_args *ua;
4499 int error;
4500
85e86edf 4501 dlm_lock_recovery(ls);
597d0cae
DT
4502
4503 error = find_lkb(ls, lkid, &lkb);
4504 if (error)
4505 goto out;
4506
4507 /* user can change the params on its lock when it converts it, or
4508 add an lvb that didn't exist before */
4509
4510 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4511
4512 if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) {
62a0f623 4513 ua->lksb.sb_lvbptr = kzalloc(DLM_USER_LVB_LEN, GFP_KERNEL);
597d0cae
DT
4514 if (!ua->lksb.sb_lvbptr) {
4515 error = -ENOMEM;
4516 goto out_put;
4517 }
4518 }
4519 if (lvb_in && ua->lksb.sb_lvbptr)
4520 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
4521
d7db923e 4522 ua->xid = ua_tmp->xid;
597d0cae
DT
4523 ua->castparam = ua_tmp->castparam;
4524 ua->castaddr = ua_tmp->castaddr;
4525 ua->bastparam = ua_tmp->bastparam;
4526 ua->bastaddr = ua_tmp->bastaddr;
10948eb4 4527 ua->user_lksb = ua_tmp->user_lksb;
597d0cae
DT
4528 ua->old_mode = lkb->lkb_grmode;
4529
d7db923e
DT
4530 error = set_lock_args(mode, &ua->lksb, flags, 0, timeout_cs,
4531 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args);
597d0cae
DT
4532 if (error)
4533 goto out_put;
4534
4535 error = convert_lock(ls, lkb, &args);
4536
c85d65e9 4537 if (error == -EINPROGRESS || error == -EAGAIN || error == -EDEADLK)
597d0cae
DT
4538 error = 0;
4539 out_put:
4540 dlm_put_lkb(lkb);
4541 out:
85e86edf 4542 dlm_unlock_recovery(ls);
597d0cae
DT
4543 kfree(ua_tmp);
4544 return error;
4545}
4546
4547int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4548 uint32_t flags, uint32_t lkid, char *lvb_in)
4549{
4550 struct dlm_lkb *lkb;
4551 struct dlm_args args;
4552 struct dlm_user_args *ua;
4553 int error;
4554
85e86edf 4555 dlm_lock_recovery(ls);
597d0cae
DT
4556
4557 error = find_lkb(ls, lkid, &lkb);
4558 if (error)
4559 goto out;
4560
4561 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4562
4563 if (lvb_in && ua->lksb.sb_lvbptr)
4564 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
b434eda6
PC
4565 if (ua_tmp->castparam)
4566 ua->castparam = ua_tmp->castparam;
cc346d55 4567 ua->user_lksb = ua_tmp->user_lksb;
597d0cae
DT
4568
4569 error = set_unlock_args(flags, ua, &args);
4570 if (error)
4571 goto out_put;
4572
4573 error = unlock_lock(ls, lkb, &args);
4574
4575 if (error == -DLM_EUNLOCK)
4576 error = 0;
ef0c2bb0
DT
4577 /* from validate_unlock_args() */
4578 if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
4579 error = 0;
597d0cae
DT
4580 if (error)
4581 goto out_put;
4582
4583 spin_lock(&ua->proc->locks_spin);
a1bc86e6
DT
4584 /* dlm_user_add_ast() may have already taken lkb off the proc list */
4585 if (!list_empty(&lkb->lkb_ownqueue))
4586 list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
597d0cae 4587 spin_unlock(&ua->proc->locks_spin);
597d0cae
DT
4588 out_put:
4589 dlm_put_lkb(lkb);
4590 out:
85e86edf 4591 dlm_unlock_recovery(ls);
ef0c2bb0 4592 kfree(ua_tmp);
597d0cae
DT
4593 return error;
4594}
4595
4596int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
4597 uint32_t flags, uint32_t lkid)
4598{
4599 struct dlm_lkb *lkb;
4600 struct dlm_args args;
4601 struct dlm_user_args *ua;
4602 int error;
4603
85e86edf 4604 dlm_lock_recovery(ls);
597d0cae
DT
4605
4606 error = find_lkb(ls, lkid, &lkb);
4607 if (error)
4608 goto out;
4609
4610 ua = (struct dlm_user_args *)lkb->lkb_astparam;
b434eda6
PC
4611 if (ua_tmp->castparam)
4612 ua->castparam = ua_tmp->castparam;
c059f70e 4613 ua->user_lksb = ua_tmp->user_lksb;
597d0cae
DT
4614
4615 error = set_unlock_args(flags, ua, &args);
4616 if (error)
4617 goto out_put;
4618
4619 error = cancel_lock(ls, lkb, &args);
4620
4621 if (error == -DLM_ECANCEL)
4622 error = 0;
ef0c2bb0
DT
4623 /* from validate_unlock_args() */
4624 if (error == -EBUSY)
4625 error = 0;
597d0cae
DT
4626 out_put:
4627 dlm_put_lkb(lkb);
4628 out:
85e86edf 4629 dlm_unlock_recovery(ls);
ef0c2bb0 4630 kfree(ua_tmp);
597d0cae
DT
4631 return error;
4632}
4633
8b4021fa
DT
4634int dlm_user_deadlock(struct dlm_ls *ls, uint32_t flags, uint32_t lkid)
4635{
4636 struct dlm_lkb *lkb;
4637 struct dlm_args args;
4638 struct dlm_user_args *ua;
4639 struct dlm_rsb *r;
4640 int error;
4641
4642 dlm_lock_recovery(ls);
4643
4644 error = find_lkb(ls, lkid, &lkb);
4645 if (error)
4646 goto out;
4647
4648 ua = (struct dlm_user_args *)lkb->lkb_astparam;
4649
4650 error = set_unlock_args(flags, ua, &args);
4651 if (error)
4652 goto out_put;
4653
4654 /* same as cancel_lock(), but set DEADLOCK_CANCEL after lock_rsb */
4655
4656 r = lkb->lkb_resource;
4657 hold_rsb(r);
4658 lock_rsb(r);
4659
4660 error = validate_unlock_args(lkb, &args);
4661 if (error)
4662 goto out_r;
4663 lkb->lkb_flags |= DLM_IFL_DEADLOCK_CANCEL;
4664
4665 error = _cancel_lock(r, lkb);
4666 out_r:
4667 unlock_rsb(r);
4668 put_rsb(r);
4669
4670 if (error == -DLM_ECANCEL)
4671 error = 0;
4672 /* from validate_unlock_args() */
4673 if (error == -EBUSY)
4674 error = 0;
4675 out_put:
4676 dlm_put_lkb(lkb);
4677 out:
4678 dlm_unlock_recovery(ls);
4679 return error;
4680}
4681
ef0c2bb0
DT
4682/* lkb's that are removed from the waiters list by revert are just left on the
4683 orphans list with the granted orphan locks, to be freed by purge */
4684
597d0cae
DT
4685static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4686{
4687 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
ef0c2bb0
DT
4688 struct dlm_args args;
4689 int error;
597d0cae 4690
ef0c2bb0
DT
4691 hold_lkb(lkb);
4692 mutex_lock(&ls->ls_orphans_mutex);
4693 list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
4694 mutex_unlock(&ls->ls_orphans_mutex);
597d0cae 4695
ef0c2bb0
DT
4696 set_unlock_args(0, ua, &args);
4697
4698 error = cancel_lock(ls, lkb, &args);
4699 if (error == -DLM_ECANCEL)
4700 error = 0;
4701 return error;
597d0cae
DT
4702}
4703
4704/* The force flag allows the unlock to go ahead even if the lkb isn't granted.
4705 Regardless of what rsb queue the lock is on, it's removed and freed. */
4706
4707static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
4708{
4709 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
4710 struct dlm_args args;
4711 int error;
4712
597d0cae
DT
4713 set_unlock_args(DLM_LKF_FORCEUNLOCK, ua, &args);
4714
4715 error = unlock_lock(ls, lkb, &args);
4716 if (error == -DLM_EUNLOCK)
4717 error = 0;
4718 return error;
4719}
4720
ef0c2bb0
DT
4721/* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
4722 (which does lock_rsb) due to deadlock with receiving a message that does
4723 lock_rsb followed by dlm_user_add_ast() */
4724
4725static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
4726 struct dlm_user_proc *proc)
4727{
4728 struct dlm_lkb *lkb = NULL;
4729
4730 mutex_lock(&ls->ls_clear_proc_locks);
4731 if (list_empty(&proc->locks))
4732 goto out;
4733
4734 lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
4735 list_del_init(&lkb->lkb_ownqueue);
4736
4737 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
4738 lkb->lkb_flags |= DLM_IFL_ORPHAN;
4739 else
4740 lkb->lkb_flags |= DLM_IFL_DEAD;
4741 out:
4742 mutex_unlock(&ls->ls_clear_proc_locks);
4743 return lkb;
4744}
4745
597d0cae
DT
4746/* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
4747 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
4748 which we clear here. */
4749
4750/* proc CLOSING flag is set so no more device_reads should look at proc->asts
4751 list, and no more device_writes should add lkb's to proc->locks list; so we
4752 shouldn't need to take asts_spin or locks_spin here. this assumes that
4753 device reads/writes/closes are serialized -- FIXME: we may need to serialize
4754 them ourself. */
4755
4756void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4757{
4758 struct dlm_lkb *lkb, *safe;
4759
85e86edf 4760 dlm_lock_recovery(ls);
597d0cae 4761
ef0c2bb0
DT
4762 while (1) {
4763 lkb = del_proc_lock(ls, proc);
4764 if (!lkb)
4765 break;
84d8cd69 4766 del_timeout(lkb);
ef0c2bb0 4767 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
597d0cae 4768 orphan_proc_lock(ls, lkb);
ef0c2bb0 4769 else
597d0cae 4770 unlock_proc_lock(ls, lkb);
597d0cae
DT
4771
4772 /* this removes the reference for the proc->locks list
4773 added by dlm_user_request, it may result in the lkb
4774 being freed */
4775
4776 dlm_put_lkb(lkb);
4777 }
a1bc86e6 4778
ef0c2bb0
DT
4779 mutex_lock(&ls->ls_clear_proc_locks);
4780
a1bc86e6
DT
4781 /* in-progress unlocks */
4782 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4783 list_del_init(&lkb->lkb_ownqueue);
4784 lkb->lkb_flags |= DLM_IFL_DEAD;
4785 dlm_put_lkb(lkb);
4786 }
4787
4788 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
8a358ca8 4789 lkb->lkb_ast_type = 0;
a1bc86e6
DT
4790 list_del(&lkb->lkb_astqueue);
4791 dlm_put_lkb(lkb);
4792 }
4793
597d0cae 4794 mutex_unlock(&ls->ls_clear_proc_locks);
85e86edf 4795 dlm_unlock_recovery(ls);
597d0cae 4796}
a1bc86e6 4797
8499137d
DT
4798static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
4799{
4800 struct dlm_lkb *lkb, *safe;
4801
4802 while (1) {
4803 lkb = NULL;
4804 spin_lock(&proc->locks_spin);
4805 if (!list_empty(&proc->locks)) {
4806 lkb = list_entry(proc->locks.next, struct dlm_lkb,
4807 lkb_ownqueue);
4808 list_del_init(&lkb->lkb_ownqueue);
4809 }
4810 spin_unlock(&proc->locks_spin);
4811
4812 if (!lkb)
4813 break;
4814
4815 lkb->lkb_flags |= DLM_IFL_DEAD;
4816 unlock_proc_lock(ls, lkb);
4817 dlm_put_lkb(lkb); /* ref from proc->locks list */
4818 }
4819
4820 spin_lock(&proc->locks_spin);
4821 list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
4822 list_del_init(&lkb->lkb_ownqueue);
4823 lkb->lkb_flags |= DLM_IFL_DEAD;
4824 dlm_put_lkb(lkb);
4825 }
4826 spin_unlock(&proc->locks_spin);
4827
4828 spin_lock(&proc->asts_spin);
4829 list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
4830 list_del(&lkb->lkb_astqueue);
4831 dlm_put_lkb(lkb);
4832 }
4833 spin_unlock(&proc->asts_spin);
4834}
4835
4836/* pid of 0 means purge all orphans */
4837
4838static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
4839{
4840 struct dlm_lkb *lkb, *safe;
4841
4842 mutex_lock(&ls->ls_orphans_mutex);
4843 list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
4844 if (pid && lkb->lkb_ownpid != pid)
4845 continue;
4846 unlock_proc_lock(ls, lkb);
4847 list_del_init(&lkb->lkb_ownqueue);
4848 dlm_put_lkb(lkb);
4849 }
4850 mutex_unlock(&ls->ls_orphans_mutex);
4851}
4852
4853static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
4854{
4855 struct dlm_message *ms;
4856 struct dlm_mhandle *mh;
4857 int error;
4858
4859 error = _create_message(ls, sizeof(struct dlm_message), nodeid,
4860 DLM_MSG_PURGE, &ms, &mh);
4861 if (error)
4862 return error;
4863 ms->m_nodeid = nodeid;
4864 ms->m_pid = pid;
4865
4866 return send_message(mh, ms);
4867}
4868
4869int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
4870 int nodeid, int pid)
4871{
4872 int error = 0;
4873
4874 if (nodeid != dlm_our_nodeid()) {
4875 error = send_purge(ls, nodeid, pid);
4876 } else {
85e86edf 4877 dlm_lock_recovery(ls);
8499137d
DT
4878 if (pid == current->pid)
4879 purge_proc_locks(ls, proc);
4880 else
4881 do_purge(ls, nodeid, pid);
85e86edf 4882 dlm_unlock_recovery(ls);
8499137d
DT
4883 }
4884 return error;
4885}
4886
This page took 0.385665 seconds and 5 git commands to generate.