1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmcommon.h
5 *
6 * Copyright (C) 2004 Oracle. All rights reserved.
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public
10 * License as published by the Free Software Foundation; either
11 * version 2 of the License, or (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public
19 * License along with this program; if not, write to the
20 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
21 * Boston, MA 021110-1307, USA.
22 *
23 */
24
25 #ifndef DLMCOMMON_H
26 #define DLMCOMMON_H
27
28 #include <linux/kref.h>
29
30 #define DLM_HB_NODE_DOWN_PRI (0xf000000)
31 #define DLM_HB_NODE_UP_PRI (0x8000000)
32
33 #define DLM_LOCKID_NAME_MAX 32
34
35 #define DLM_DOMAIN_NAME_MAX_LEN 255
36 #define DLM_LOCK_RES_OWNER_UNKNOWN O2NM_MAX_NODES
37 #define DLM_THREAD_SHUFFLE_INTERVAL 5 // flush everything every 5 passes
38 #define DLM_THREAD_MS 200 // flush at least every 200 ms
39
40 #define DLM_HASH_SIZE_DEFAULT (1 << 17)
41 #if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE
42 # define DLM_HASH_PAGES 1
43 #else
44 # define DLM_HASH_PAGES (DLM_HASH_SIZE_DEFAULT / PAGE_SIZE)
45 #endif
46 #define DLM_BUCKETS_PER_PAGE (PAGE_SIZE / sizeof(struct hlist_head))
47 #define DLM_HASH_BUCKETS (DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
48
49 /* Intended to make it easier for us to switch out hash functions */
50 #define dlm_lockid_hash(_n, _l) full_name_hash(NULL, _n, _l)
51
52 enum dlm_mle_type {
53 DLM_MLE_BLOCK = 0,
54 DLM_MLE_MASTER = 1,
55 DLM_MLE_MIGRATION = 2,
56 DLM_MLE_NUM_TYPES = 3,
57 };
58
59 struct dlm_master_list_entry {
60 struct hlist_node master_hash_node;
61 struct list_head hb_events;
62 struct dlm_ctxt *dlm;
63 spinlock_t spinlock;
64 wait_queue_head_t wq;
65 atomic_t woken;
66 struct kref mle_refs;
67 int inuse;
68 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
69 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
70 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
71 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
72 u8 master;
73 u8 new_master;
74 enum dlm_mle_type type;
75 struct o2hb_callback_func mle_hb_up;
76 struct o2hb_callback_func mle_hb_down;
77 struct dlm_lock_resource *mleres;
78 unsigned char mname[DLM_LOCKID_NAME_MAX];
79 unsigned int mnamelen;
80 unsigned int mnamehash;
81 };
82
83 enum dlm_ast_type {
84 DLM_AST = 0,
85 DLM_BAST = 1,
86 DLM_ASTUNLOCK = 2,
87 };
88
89
90 #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
91 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
92 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)
93
94 #define DLM_RECOVERY_LOCK_NAME "$RECOVERY"
95 #define DLM_RECOVERY_LOCK_NAME_LEN 9
96
dlm_is_recovery_lock(const char * lock_name,int name_len)97 static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
98 {
99 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
100 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
101 return 1;
102 return 0;
103 }
104
105 #define DLM_RECO_STATE_ACTIVE 0x0001
106 #define DLM_RECO_STATE_FINALIZE 0x0002
107
108 struct dlm_recovery_ctxt
109 {
110 struct list_head resources;
111 struct list_head node_data;
112 u8 new_master;
113 u8 dead_node;
114 u16 state;
115 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
116 wait_queue_head_t event;
117 };
118
119 enum dlm_ctxt_state {
120 DLM_CTXT_NEW = 0,
121 DLM_CTXT_JOINED = 1,
122 DLM_CTXT_IN_SHUTDOWN = 2,
123 DLM_CTXT_LEAVING = 3,
124 };
125
126 struct dlm_ctxt
127 {
128 struct list_head list;
129 struct hlist_head **lockres_hash;
130 struct list_head dirty_list;
131 struct list_head purge_list;
132 struct list_head pending_asts;
133 struct list_head pending_basts;
134 struct list_head tracking_list;
135 unsigned int purge_count;
136 spinlock_t spinlock;
137 spinlock_t ast_lock;
138 spinlock_t track_lock;
139 char *name;
140 u8 node_num;
141 u32 key;
142 u8 joining_node;
143 u8 migrate_done; /* set to 1 means node has migrated all lock resources */
144 wait_queue_head_t dlm_join_events;
145 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
146 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
147 unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
148 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
149 struct dlm_recovery_ctxt reco;
150 spinlock_t master_lock;
151 struct hlist_head **master_hash;
152 struct list_head mle_hb_events;
153
154 /* these give a really vague idea of the system load */
155 atomic_t mle_tot_count[DLM_MLE_NUM_TYPES];
156 atomic_t mle_cur_count[DLM_MLE_NUM_TYPES];
157 atomic_t res_tot_count;
158 atomic_t res_cur_count;
159
160 struct dlm_debug_ctxt *dlm_debug_ctxt;
161 struct dentry *dlm_debugfs_subroot;
162
163 /* NOTE: Next three are protected by dlm_domain_lock */
164 struct kref dlm_refs;
165 enum dlm_ctxt_state dlm_state;
166 unsigned int num_joins;
167
168 struct o2hb_callback_func dlm_hb_up;
169 struct o2hb_callback_func dlm_hb_down;
170 struct task_struct *dlm_thread_task;
171 struct task_struct *dlm_reco_thread_task;
172 struct workqueue_struct *dlm_worker;
173 wait_queue_head_t dlm_thread_wq;
174 wait_queue_head_t dlm_reco_thread_wq;
175 wait_queue_head_t ast_wq;
176 wait_queue_head_t migration_wq;
177
178 struct work_struct dispatched_work;
179 struct list_head work_list;
180 spinlock_t work_lock;
181 struct list_head dlm_domain_handlers;
182 struct list_head dlm_eviction_callbacks;
183
184 /* The filesystem specifies this at domain registration. We
185 * cache it here to know what to tell other nodes. */
186 struct dlm_protocol_version fs_locking_proto;
187 /* This is the inter-dlm communication version */
188 struct dlm_protocol_version dlm_locking_proto;
189 };
190
dlm_lockres_hash(struct dlm_ctxt * dlm,unsigned i)191 static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
192 {
193 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
194 }
195
dlm_master_hash(struct dlm_ctxt * dlm,unsigned i)196 static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm,
197 unsigned i)
198 {
199 return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] +
200 (i % DLM_BUCKETS_PER_PAGE);
201 }
202
203 /* these keventd work queue items are for less-frequently
204 * called functions that cannot be directly called from the
205 * net message handlers for some reason, usually because
206 * they need to send net messages of their own. */
207 void dlm_dispatch_work(struct work_struct *work);
208
209 struct dlm_lock_resource;
210 struct dlm_work_item;
211
212 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);
213
214 struct dlm_request_all_locks_priv
215 {
216 u8 reco_master;
217 u8 dead_node;
218 };
219
220 struct dlm_mig_lockres_priv
221 {
222 struct dlm_lock_resource *lockres;
223 u8 real_master;
224 u8 extra_ref;
225 };
226
227 struct dlm_assert_master_priv
228 {
229 struct dlm_lock_resource *lockres;
230 u8 request_from;
231 u32 flags;
232 unsigned ignore_higher:1;
233 };
234
235 struct dlm_deref_lockres_priv
236 {
237 struct dlm_lock_resource *deref_res;
238 u8 deref_node;
239 };
240
241 struct dlm_work_item
242 {
243 struct list_head list;
244 dlm_workfunc_t *func;
245 struct dlm_ctxt *dlm;
246 void *data;
247 union {
248 struct dlm_request_all_locks_priv ral;
249 struct dlm_mig_lockres_priv ml;
250 struct dlm_assert_master_priv am;
251 struct dlm_deref_lockres_priv dl;
252 } u;
253 };
254
dlm_init_work_item(struct dlm_ctxt * dlm,struct dlm_work_item * i,dlm_workfunc_t * f,void * data)255 static inline void dlm_init_work_item(struct dlm_ctxt *dlm,
256 struct dlm_work_item *i,
257 dlm_workfunc_t *f, void *data)
258 {
259 memset(i, 0, sizeof(*i));
260 i->func = f;
261 INIT_LIST_HEAD(&i->list);
262 i->data = data;
263 i->dlm = dlm; /* must have already done a dlm_grab on this! */
264 }
265
266
267
__dlm_set_joining_node(struct dlm_ctxt * dlm,u8 node)268 static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm,
269 u8 node)
270 {
271 assert_spin_locked(&dlm->spinlock);
272
273 dlm->joining_node = node;
274 wake_up(&dlm->dlm_join_events);
275 }
276
277 #define DLM_LOCK_RES_UNINITED 0x00000001
278 #define DLM_LOCK_RES_RECOVERING 0x00000002
279 #define DLM_LOCK_RES_READY 0x00000004
280 #define DLM_LOCK_RES_DIRTY 0x00000008
281 #define DLM_LOCK_RES_IN_PROGRESS 0x00000010
282 #define DLM_LOCK_RES_MIGRATING 0x00000020
283 #define DLM_LOCK_RES_DROPPING_REF 0x00000040
284 #define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000
285 #define DLM_LOCK_RES_SETREF_INPROG 0x00002000
286 #define DLM_LOCK_RES_RECOVERY_WAITING 0x00004000
287
288 /* max milliseconds to wait to sync up a network failure with a node death */
289 #define DLM_NODE_DEATH_WAIT_MAX (5 * 1000)
290
291 #define DLM_PURGE_INTERVAL_MS (8 * 1000)
292
293 struct dlm_lock_resource
294 {
295 /* WARNING: Please see the comment in dlm_init_lockres before
296 * adding fields here. */
297 struct hlist_node hash_node;
298 struct qstr lockname;
299 struct kref refs;
300
301 /*
302 * Please keep granted, converting, and blocked in this order,
303 * as some funcs want to iterate over all lists.
304 *
305 * All four lists are protected by the hash's reference.
306 */
307 struct list_head granted;
308 struct list_head converting;
309 struct list_head blocked;
310 struct list_head purge;
311
312 /*
313 * These two lists require you to hold an additional reference
314 * while they are on the list.
315 */
316 struct list_head dirty;
317 struct list_head recovering; // dlm_recovery_ctxt.resources list
318
319 /* Added during init and removed during release */
320 struct list_head tracking; /* dlm->tracking_list */
321
322 /* unused lock resources have their last_used stamped and are
323 * put on a list for the dlm thread to run. */
324 unsigned long last_used;
325
326 struct dlm_ctxt *dlm;
327
328 unsigned migration_pending:1;
329 atomic_t asts_reserved;
330 spinlock_t spinlock;
331 wait_queue_head_t wq;
332 u8 owner; //node which owns the lock resource, or unknown
333 u16 state;
334 char lvb[DLM_LVB_LEN];
335 unsigned int inflight_locks;
336 unsigned int inflight_assert_workers;
337 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
338 };
339
340 struct dlm_migratable_lock
341 {
342 __be64 cookie;
343
344 /* these 3 are just padding for the in-memory structure, but
345 * list and flags are actually used when sent over the wire */
346 __be16 pad1;
347 u8 list; // 0=granted, 1=converting, 2=blocked
348 u8 flags;
349
350 s8 type;
351 s8 convert_type;
352 s8 highest_blocked;
353 u8 node;
354 }; // 16 bytes
355
356 struct dlm_lock
357 {
358 struct dlm_migratable_lock ml;
359
360 struct list_head list;
361 struct list_head ast_list;
362 struct list_head bast_list;
363 struct dlm_lock_resource *lockres;
364 spinlock_t spinlock;
365 struct kref lock_refs;
366
367 // ast and bast must be callable while holding a spinlock!
368 dlm_astlockfunc_t *ast;
369 dlm_bastlockfunc_t *bast;
370 void *astdata;
371 struct dlm_lockstatus *lksb;
372 unsigned ast_pending:1,
373 bast_pending:1,
374 convert_pending:1,
375 lock_pending:1,
376 cancel_pending:1,
377 unlock_pending:1,
378 lksb_kernel_allocated:1;
379 };
380
381 enum dlm_lockres_list {
382 DLM_GRANTED_LIST = 0,
383 DLM_CONVERTING_LIST = 1,
384 DLM_BLOCKED_LIST = 2,
385 };
386
dlm_lvb_is_empty(char * lvb)387 static inline int dlm_lvb_is_empty(char *lvb)
388 {
389 int i;
390 for (i=0; i<DLM_LVB_LEN; i++)
391 if (lvb[i])
392 return 0;
393 return 1;
394 }
395
dlm_list_in_text(enum dlm_lockres_list idx)396 static inline char *dlm_list_in_text(enum dlm_lockres_list idx)
397 {
398 if (idx == DLM_GRANTED_LIST)
399 return "granted";
400 else if (idx == DLM_CONVERTING_LIST)
401 return "converting";
402 else if (idx == DLM_BLOCKED_LIST)
403 return "blocked";
404 else
405 return "unknown";
406 }
407
408 static inline struct list_head *
dlm_list_idx_to_ptr(struct dlm_lock_resource * res,enum dlm_lockres_list idx)409 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
410 {
411 struct list_head *ret = NULL;
412 if (idx == DLM_GRANTED_LIST)
413 ret = &res->granted;
414 else if (idx == DLM_CONVERTING_LIST)
415 ret = &res->converting;
416 else if (idx == DLM_BLOCKED_LIST)
417 ret = &res->blocked;
418 else
419 BUG();
420 return ret;
421 }
422
423
424
425
426 struct dlm_node_iter
427 {
428 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
429 int curnode;
430 };
431
432
433 enum {
434 DLM_MASTER_REQUEST_MSG = 500,
435 DLM_UNUSED_MSG1 = 501,
436 DLM_ASSERT_MASTER_MSG = 502,
437 DLM_CREATE_LOCK_MSG = 503,
438 DLM_CONVERT_LOCK_MSG = 504,
439 DLM_PROXY_AST_MSG = 505,
440 DLM_UNLOCK_LOCK_MSG = 506,
441 DLM_DEREF_LOCKRES_MSG = 507,
442 DLM_MIGRATE_REQUEST_MSG = 508,
443 DLM_MIG_LOCKRES_MSG = 509,
444 DLM_QUERY_JOIN_MSG = 510,
445 DLM_ASSERT_JOINED_MSG = 511,
446 DLM_CANCEL_JOIN_MSG = 512,
447 DLM_EXIT_DOMAIN_MSG = 513,
448 DLM_MASTER_REQUERY_MSG = 514,
449 DLM_LOCK_REQUEST_MSG = 515,
450 DLM_RECO_DATA_DONE_MSG = 516,
451 DLM_BEGIN_RECO_MSG = 517,
452 DLM_FINALIZE_RECO_MSG = 518,
453 DLM_QUERY_REGION = 519,
454 DLM_QUERY_NODEINFO = 520,
455 DLM_BEGIN_EXIT_DOMAIN_MSG = 521,
456 DLM_DEREF_LOCKRES_DONE = 522,
457 };
458
459 struct dlm_reco_node_data
460 {
461 int state;
462 u8 node_num;
463 struct list_head list;
464 };
465
466 enum {
467 DLM_RECO_NODE_DATA_DEAD = -1,
468 DLM_RECO_NODE_DATA_INIT = 0,
469 DLM_RECO_NODE_DATA_REQUESTING = 1,
470 DLM_RECO_NODE_DATA_REQUESTED = 2,
471 DLM_RECO_NODE_DATA_RECEIVING = 3,
472 DLM_RECO_NODE_DATA_DONE = 4,
473 DLM_RECO_NODE_DATA_FINALIZE_SENT = 5,
474 };
475
476
477 enum {
478 DLM_MASTER_RESP_NO = 0,
479 DLM_MASTER_RESP_YES = 1,
480 DLM_MASTER_RESP_MAYBE = 2,
481 DLM_MASTER_RESP_ERROR = 3,
482 };
483
484
485 struct dlm_master_request
486 {
487 u8 node_idx;
488 u8 namelen;
489 __be16 pad1;
490 __be32 flags;
491
492 u8 name[O2NM_MAX_NAME_LEN];
493 };
494
495 #define DLM_ASSERT_RESPONSE_REASSERT 0x00000001
496 #define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002
497
498 #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001
499 #define DLM_ASSERT_MASTER_REQUERY 0x00000002
500 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
501 struct dlm_assert_master
502 {
503 u8 node_idx;
504 u8 namelen;
505 __be16 pad1;
506 __be32 flags;
507
508 u8 name[O2NM_MAX_NAME_LEN];
509 };
510
511 #define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001
512
513 struct dlm_migrate_request
514 {
515 u8 master;
516 u8 new_master;
517 u8 namelen;
518 u8 pad1;
519 __be32 pad2;
520 u8 name[O2NM_MAX_NAME_LEN];
521 };
522
523 struct dlm_master_requery
524 {
525 u8 pad1;
526 u8 pad2;
527 u8 node_idx;
528 u8 namelen;
529 __be32 pad3;
530 u8 name[O2NM_MAX_NAME_LEN];
531 };
532
533 #define DLM_MRES_RECOVERY 0x01
534 #define DLM_MRES_MIGRATION 0x02
535 #define DLM_MRES_ALL_DONE 0x04
536
537 /*
538 * We would like to get one whole lockres into a single network
539 * message whenever possible. Generally speaking, there will be
540 * at most one dlm_lock on a lockres for each node in the cluster,
541 * plus (infrequently) any additional locks coming in from userdlm.
542 *
543 * struct _dlm_lockres_page
544 * {
545 * dlm_migratable_lockres mres;
546 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
547 * u8 pad[DLM_MIG_LOCKRES_RESERVED];
548 * };
549 *
550 * from ../cluster/tcp.h
551 * O2NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg))
552 * (roughly 4080 bytes)
553 * and sizeof(dlm_migratable_lockres) = 112 bytes
554 * and sizeof(dlm_migratable_lock) = 16 bytes
555 *
556 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
557 * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
558 *
559 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) +
560 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED =
561 * NET_MAX_PAYLOAD_BYTES
562 * (240 * 16) + 112 + 128 = 4080
563 *
564 * So a lockres would need more than 240 locks before it would
565 * use more than one network packet to recover. Not too bad.
566 */
567 #define DLM_MAX_MIGRATABLE_LOCKS 240
568
569 struct dlm_migratable_lockres
570 {
571 u8 master;
572 u8 lockname_len;
573 u8 num_locks; // locks sent in this structure
574 u8 flags;
575 __be32 total_locks; // locks to be sent for this migration cookie
576 __be64 mig_cookie; // cookie for this lockres migration
577 // or zero if not needed
578 // 16 bytes
579 u8 lockname[DLM_LOCKID_NAME_MAX];
580 // 48 bytes
581 u8 lvb[DLM_LVB_LEN];
582 // 112 bytes
583 struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112
584 };
585 #define DLM_MIG_LOCKRES_MAX_LEN \
586 (sizeof(struct dlm_migratable_lockres) + \
587 (sizeof(struct dlm_migratable_lock) * \
588 DLM_MAX_MIGRATABLE_LOCKS) )
589
590 /* from above, 128 bytes
591 * for some undetermined future use */
592 #define DLM_MIG_LOCKRES_RESERVED (O2NET_MAX_PAYLOAD_BYTES - \
593 DLM_MIG_LOCKRES_MAX_LEN)
594
595 struct dlm_create_lock
596 {
597 __be64 cookie;
598
599 __be32 flags;
600 u8 pad1;
601 u8 node_idx;
602 s8 requested_type;
603 u8 namelen;
604
605 u8 name[O2NM_MAX_NAME_LEN];
606 };
607
608 struct dlm_convert_lock
609 {
610 __be64 cookie;
611
612 __be32 flags;
613 u8 pad1;
614 u8 node_idx;
615 s8 requested_type;
616 u8 namelen;
617
618 u8 name[O2NM_MAX_NAME_LEN];
619
620 s8 lvb[0];
621 };
622 #define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN)
623
624 struct dlm_unlock_lock
625 {
626 __be64 cookie;
627
628 __be32 flags;
629 __be16 pad1;
630 u8 node_idx;
631 u8 namelen;
632
633 u8 name[O2NM_MAX_NAME_LEN];
634
635 s8 lvb[0];
636 };
637 #define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN)
638
639 struct dlm_proxy_ast
640 {
641 __be64 cookie;
642
643 __be32 flags;
644 u8 node_idx;
645 u8 type;
646 u8 blocked_type;
647 u8 namelen;
648
649 u8 name[O2NM_MAX_NAME_LEN];
650
651 s8 lvb[0];
652 };
653 #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN)
654
655 #define DLM_MOD_KEY (0x666c6172)
656 enum dlm_query_join_response_code {
657 JOIN_DISALLOW = 0,
658 JOIN_OK = 1,
659 JOIN_OK_NO_MAP = 2,
660 JOIN_PROTOCOL_MISMATCH = 3,
661 };
662
663 struct dlm_query_join_packet {
664 u8 code; /* Response code. dlm_minor and fs_minor
665 are only valid if this is JOIN_OK */
666 u8 dlm_minor; /* The minor version of the protocol the
667 dlm is speaking. */
668 u8 fs_minor; /* The minor version of the protocol the
669 filesystem is speaking. */
670 u8 reserved;
671 };
672
673 union dlm_query_join_response {
674 __be32 intval;
675 struct dlm_query_join_packet packet;
676 };
677
678 struct dlm_lock_request
679 {
680 u8 node_idx;
681 u8 dead_node;
682 __be16 pad1;
683 __be32 pad2;
684 };
685
686 struct dlm_reco_data_done
687 {
688 u8 node_idx;
689 u8 dead_node;
690 __be16 pad1;
691 __be32 pad2;
692
693 /* unused for now */
694 /* eventually we can use this to attempt
695 * lvb recovery based on each node's info */
696 u8 reco_lvb[DLM_LVB_LEN];
697 };
698
699 struct dlm_begin_reco
700 {
701 u8 node_idx;
702 u8 dead_node;
703 __be16 pad1;
704 __be32 pad2;
705 };
706
707
708 #define BITS_PER_BYTE 8
709 #define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE)
710
711 struct dlm_query_join_request
712 {
713 u8 node_idx;
714 u8 pad1[2];
715 u8 name_len;
716 struct dlm_protocol_version dlm_proto;
717 struct dlm_protocol_version fs_proto;
718 u8 domain[O2NM_MAX_NAME_LEN];
719 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)];
720 };
721
722 struct dlm_assert_joined
723 {
724 u8 node_idx;
725 u8 pad1[2];
726 u8 name_len;
727 u8 domain[O2NM_MAX_NAME_LEN];
728 };
729
730 struct dlm_cancel_join
731 {
732 u8 node_idx;
733 u8 pad1[2];
734 u8 name_len;
735 u8 domain[O2NM_MAX_NAME_LEN];
736 };
737
738 struct dlm_query_region {
739 u8 qr_node;
740 u8 qr_numregions;
741 u8 qr_namelen;
742 u8 pad1;
743 u8 qr_domain[O2NM_MAX_NAME_LEN];
744 u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS];
745 };
746
747 struct dlm_node_info {
748 u8 ni_nodenum;
749 u8 pad1;
750 __be16 ni_ipv4_port;
751 __be32 ni_ipv4_address;
752 };
753
754 struct dlm_query_nodeinfo {
755 u8 qn_nodenum;
756 u8 qn_numnodes;
757 u8 qn_namelen;
758 u8 pad1;
759 u8 qn_domain[O2NM_MAX_NAME_LEN];
760 struct dlm_node_info qn_nodes[O2NM_MAX_NODES];
761 };
762
763 struct dlm_exit_domain
764 {
765 u8 node_idx;
766 u8 pad1[3];
767 };
768
769 struct dlm_finalize_reco
770 {
771 u8 node_idx;
772 u8 dead_node;
773 u8 flags;
774 u8 pad1;
775 __be32 pad2;
776 };
777
778 struct dlm_deref_lockres
779 {
780 u32 pad1;
781 u16 pad2;
782 u8 node_idx;
783 u8 namelen;
784
785 u8 name[O2NM_MAX_NAME_LEN];
786 };
787
788 enum {
789 DLM_DEREF_RESPONSE_DONE = 0,
790 DLM_DEREF_RESPONSE_INPROG = 1,
791 };
792
793 struct dlm_deref_lockres_done {
794 u32 pad1;
795 u16 pad2;
796 u8 node_idx;
797 u8 namelen;
798
799 u8 name[O2NM_MAX_NAME_LEN];
800 };
801
802 static inline enum dlm_status
__dlm_lockres_state_to_status(struct dlm_lock_resource * res)803 __dlm_lockres_state_to_status(struct dlm_lock_resource *res)
804 {
805 enum dlm_status status = DLM_NORMAL;
806
807 assert_spin_locked(&res->spinlock);
808
809 if (res->state & (DLM_LOCK_RES_RECOVERING|
810 DLM_LOCK_RES_RECOVERY_WAITING))
811 status = DLM_RECOVERING;
812 else if (res->state & DLM_LOCK_RES_MIGRATING)
813 status = DLM_MIGRATING;
814 else if (res->state & DLM_LOCK_RES_IN_PROGRESS)
815 status = DLM_FORWARD;
816
817 return status;
818 }
819
dlm_get_lock_cookie_node(u64 cookie)820 static inline u8 dlm_get_lock_cookie_node(u64 cookie)
821 {
822 u8 ret;
823 cookie >>= 56;
824 ret = (u8)(cookie & 0xffULL);
825 return ret;
826 }
827
dlm_get_lock_cookie_seq(u64 cookie)828 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
829 {
830 unsigned long long ret;
831 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
832 return ret;
833 }
834
835 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
836 struct dlm_lockstatus *lksb);
837 void dlm_lock_get(struct dlm_lock *lock);
838 void dlm_lock_put(struct dlm_lock *lock);
839
840 void dlm_lock_attach_lockres(struct dlm_lock *lock,
841 struct dlm_lock_resource *res);
842
843 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
844 void **ret_data);
845 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
846 void **ret_data);
847 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
848 void **ret_data);
849
850 void dlm_revert_pending_convert(struct dlm_lock_resource *res,
851 struct dlm_lock *lock);
852 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
853 struct dlm_lock *lock);
854
855 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
856 void **ret_data);
857 void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
858 struct dlm_lock *lock);
859 void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
860 struct dlm_lock *lock);
861
862 int dlm_launch_thread(struct dlm_ctxt *dlm);
863 void dlm_complete_thread(struct dlm_ctxt *dlm);
864 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
865 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
866 void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
867 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
868 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
869 void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
870 void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
871
872 void dlm_put(struct dlm_ctxt *dlm);
873 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
874 int dlm_domain_fully_joined(struct dlm_ctxt *dlm);
875
876 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
877 struct dlm_lock_resource *res);
878 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
879 struct dlm_lock_resource *res);
dlm_lockres_get(struct dlm_lock_resource * res)880 static inline void dlm_lockres_get(struct dlm_lock_resource *res)
881 {
882 /* This is called on every lookup, so it might be worth
883 * inlining. */
884 kref_get(&res->refs);
885 }
886 void dlm_lockres_put(struct dlm_lock_resource *res);
887 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
888 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
889 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
890 const char *name,
891 unsigned int len,
892 unsigned int hash);
893 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
894 const char *name,
895 unsigned int len,
896 unsigned int hash);
897 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
898 const char *name,
899 unsigned int len);
900
901 int dlm_is_host_down(int errno);
902
903 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
904 const char *lockid,
905 int namelen,
906 int flags);
907 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
908 const char *name,
909 unsigned int namelen);
910
911 void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
912 struct dlm_lock_resource *res, int bit);
913 void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
914 struct dlm_lock_resource *res, int bit);
915
916 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
917 struct dlm_lock_resource *res);
918 void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
919 struct dlm_lock_resource *res);
920
921 void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm,
922 struct dlm_lock_resource *res);
923
924 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
925 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
926 void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
927 void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
928 void dlm_do_local_ast(struct dlm_ctxt *dlm,
929 struct dlm_lock_resource *res,
930 struct dlm_lock *lock);
931 int dlm_do_remote_ast(struct dlm_ctxt *dlm,
932 struct dlm_lock_resource *res,
933 struct dlm_lock *lock);
934 void dlm_do_local_bast(struct dlm_ctxt *dlm,
935 struct dlm_lock_resource *res,
936 struct dlm_lock *lock,
937 int blocked_type);
938 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm,
939 struct dlm_lock_resource *res,
940 struct dlm_lock *lock,
941 int msg_type,
942 int blocked_type, int flags);
dlm_send_proxy_bast(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int blocked_type)943 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm,
944 struct dlm_lock_resource *res,
945 struct dlm_lock *lock,
946 int blocked_type)
947 {
948 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST,
949 blocked_type, 0);
950 }
951
dlm_send_proxy_ast(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags)952 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm,
953 struct dlm_lock_resource *res,
954 struct dlm_lock *lock,
955 int flags)
956 {
957 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST,
958 0, flags);
959 }
960
961 void dlm_print_one_lock_resource(struct dlm_lock_resource *res);
962 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res);
963
964 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
965 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
966
967
968 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data);
969 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data);
970
971 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
972 int dlm_finish_migration(struct dlm_ctxt *dlm,
973 struct dlm_lock_resource *res,
974 u8 old_master);
975 void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
976 struct dlm_lock_resource *res);
977 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res);
978
979 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
980 void **ret_data);
981 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
982 void **ret_data);
983 void dlm_assert_master_post_handler(int status, void *data, void *ret_data);
984 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
985 void **ret_data);
986 int dlm_deref_lockres_done_handler(struct o2net_msg *msg, u32 len, void *data,
987 void **ret_data);
988 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
989 void **ret_data);
990 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
991 void **ret_data);
992 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
993 void **ret_data);
994 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
995 void **ret_data);
996 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
997 void **ret_data);
998 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
999 void **ret_data);
1000 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
1001 void **ret_data);
1002 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
1003 u8 nodenum, u8 *real_master);
1004
1005 void __dlm_do_purge_lockres(struct dlm_ctxt *dlm,
1006 struct dlm_lock_resource *res);
1007
1008 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
1009 struct dlm_lock_resource *res,
1010 int ignore_higher,
1011 u8 request_from,
1012 u32 flags);
1013
1014
1015 int dlm_send_one_lockres(struct dlm_ctxt *dlm,
1016 struct dlm_lock_resource *res,
1017 struct dlm_migratable_lockres *mres,
1018 u8 send_to,
1019 u8 flags);
1020 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
1021 struct dlm_lock_resource *res);
1022
1023 /* will exit holding res->spinlock, but may drop in function */
1024 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags);
1025
1026 /* will exit holding res->spinlock, but may drop in function */
__dlm_wait_on_lockres(struct dlm_lock_resource * res)1027 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res)
1028 {
1029 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS|
1030 DLM_LOCK_RES_RECOVERING|
1031 DLM_LOCK_RES_RECOVERY_WAITING|
1032 DLM_LOCK_RES_MIGRATING));
1033 }
1034
1035 void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1036 void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1037
1038 /* create/destroy slab caches */
1039 int dlm_init_master_caches(void);
1040 void dlm_destroy_master_caches(void);
1041
1042 int dlm_init_lock_cache(void);
1043 void dlm_destroy_lock_cache(void);
1044
1045 int dlm_init_mle_cache(void);
1046 void dlm_destroy_mle_cache(void);
1047
1048 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up);
1049 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
1050 struct dlm_lock_resource *res);
1051 void dlm_clean_master_list(struct dlm_ctxt *dlm,
1052 u8 dead_node);
1053 void dlm_force_free_mles(struct dlm_ctxt *dlm);
1054 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
1055 int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
1056 int __dlm_lockres_unused(struct dlm_lock_resource *res);
1057
dlm_lock_mode_name(int mode)1058 static inline const char * dlm_lock_mode_name(int mode)
1059 {
1060 switch (mode) {
1061 case LKM_EXMODE:
1062 return "EX";
1063 case LKM_PRMODE:
1064 return "PR";
1065 case LKM_NLMODE:
1066 return "NL";
1067 }
1068 return "UNKNOWN";
1069 }
1070
1071
dlm_lock_compatible(int existing,int request)1072 static inline int dlm_lock_compatible(int existing, int request)
1073 {
1074 /* NO_LOCK compatible with all */
1075 if (request == LKM_NLMODE ||
1076 existing == LKM_NLMODE)
1077 return 1;
1078
1079 /* EX incompatible with all non-NO_LOCK */
1080 if (request == LKM_EXMODE)
1081 return 0;
1082
1083 /* request must be PR, which is compatible with PR */
1084 if (existing == LKM_PRMODE)
1085 return 1;
1086
1087 return 0;
1088 }
1089
dlm_lock_on_list(struct list_head * head,struct dlm_lock * lock)1090 static inline int dlm_lock_on_list(struct list_head *head,
1091 struct dlm_lock *lock)
1092 {
1093 struct dlm_lock *tmplock;
1094
1095 list_for_each_entry(tmplock, head, list) {
1096 if (tmplock == lock)
1097 return 1;
1098 }
1099 return 0;
1100 }
1101
1102
dlm_err_to_dlm_status(int err)1103 static inline enum dlm_status dlm_err_to_dlm_status(int err)
1104 {
1105 enum dlm_status ret;
1106 if (err == -ENOMEM)
1107 ret = DLM_SYSERR;
1108 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL))
1109 ret = DLM_NOLOCKMGR;
1110 else if (err == -EINVAL)
1111 ret = DLM_BADPARAM;
1112 else if (err == -ENAMETOOLONG)
1113 ret = DLM_IVBUFLEN;
1114 else
1115 ret = DLM_BADARGS;
1116 return ret;
1117 }
1118
1119
dlm_node_iter_init(unsigned long * map,struct dlm_node_iter * iter)1120 static inline void dlm_node_iter_init(unsigned long *map,
1121 struct dlm_node_iter *iter)
1122 {
1123 memcpy(iter->node_map, map, sizeof(iter->node_map));
1124 iter->curnode = -1;
1125 }
1126
dlm_node_iter_next(struct dlm_node_iter * iter)1127 static inline int dlm_node_iter_next(struct dlm_node_iter *iter)
1128 {
1129 int bit;
1130 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1);
1131 if (bit >= O2NM_MAX_NODES) {
1132 iter->curnode = O2NM_MAX_NODES;
1133 return -ENOENT;
1134 }
1135 iter->curnode = bit;
1136 return bit;
1137 }
1138
dlm_set_lockres_owner(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 owner)1139 static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
1140 struct dlm_lock_resource *res,
1141 u8 owner)
1142 {
1143 assert_spin_locked(&res->spinlock);
1144
1145 res->owner = owner;
1146 }
1147
dlm_change_lockres_owner(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 owner)1148 static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
1149 struct dlm_lock_resource *res,
1150 u8 owner)
1151 {
1152 assert_spin_locked(&res->spinlock);
1153
1154 if (owner != res->owner)
1155 dlm_set_lockres_owner(dlm, res, owner);
1156 }
1157
1158 #endif /* DLMCOMMON_H */
1159
