1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * dlmcommon.h
6 *
7 * Copyright (C) 2004 Oracle. All rights reserved.
8 */
9
10 #ifndef DLMCOMMON_H
11 #define DLMCOMMON_H
12
13 #include <linux/kref.h>
14
15 #define DLM_HB_NODE_DOWN_PRI (0xf000000)
16 #define DLM_HB_NODE_UP_PRI (0x8000000)
17
18 #define DLM_LOCKID_NAME_MAX 32
19
20 #define DLM_DOMAIN_NAME_MAX_LEN 255
21 #define DLM_LOCK_RES_OWNER_UNKNOWN O2NM_MAX_NODES
22 #define DLM_THREAD_SHUFFLE_INTERVAL 5 // flush everything every 5 passes
23 #define DLM_THREAD_MS 200 // flush at least every 200 ms
24
25 #define DLM_HASH_SIZE_DEFAULT (1 << 17)
26 #if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE
27 # define DLM_HASH_PAGES 1
28 #else
29 # define DLM_HASH_PAGES (DLM_HASH_SIZE_DEFAULT / PAGE_SIZE)
30 #endif
31 #define DLM_BUCKETS_PER_PAGE (PAGE_SIZE / sizeof(struct hlist_head))
32 #define DLM_HASH_BUCKETS (DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
33
34 /* Intended to make it easier for us to switch out hash functions */
35 #define dlm_lockid_hash(_n, _l) full_name_hash(NULL, _n, _l)
36
37 enum dlm_mle_type {
38 DLM_MLE_BLOCK = 0,
39 DLM_MLE_MASTER = 1,
40 DLM_MLE_MIGRATION = 2,
41 DLM_MLE_NUM_TYPES = 3,
42 };
43
44 struct dlm_master_list_entry {
45 struct hlist_node master_hash_node;
46 struct list_head hb_events;
47 struct dlm_ctxt *dlm;
48 spinlock_t spinlock;
49 wait_queue_head_t wq;
50 atomic_t woken;
51 struct kref mle_refs;
52 int inuse;
53 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
54 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
55 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
56 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
57 u8 master;
58 u8 new_master;
59 enum dlm_mle_type type;
60 struct o2hb_callback_func mle_hb_up;
61 struct o2hb_callback_func mle_hb_down;
62 struct dlm_lock_resource *mleres;
63 unsigned char mname[DLM_LOCKID_NAME_MAX];
64 unsigned int mnamelen;
65 unsigned int mnamehash;
66 };
67
68 enum dlm_ast_type {
69 DLM_AST = 0,
70 DLM_BAST = 1,
71 DLM_ASTUNLOCK = 2,
72 };
73
74
75 #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
76 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
77 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)
78
79 #define DLM_RECOVERY_LOCK_NAME "$RECOVERY"
80 #define DLM_RECOVERY_LOCK_NAME_LEN 9
81
dlm_is_recovery_lock(const char * lock_name,int name_len)82 static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
83 {
84 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
85 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
86 return 1;
87 return 0;
88 }
89
90 #define DLM_RECO_STATE_ACTIVE 0x0001
91 #define DLM_RECO_STATE_FINALIZE 0x0002
92
93 struct dlm_recovery_ctxt
94 {
95 struct list_head resources;
96 struct list_head node_data;
97 u8 new_master;
98 u8 dead_node;
99 u16 state;
100 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
101 wait_queue_head_t event;
102 };
103
104 enum dlm_ctxt_state {
105 DLM_CTXT_NEW = 0,
106 DLM_CTXT_JOINED = 1,
107 DLM_CTXT_IN_SHUTDOWN = 2,
108 DLM_CTXT_LEAVING = 3,
109 };
110
111 struct dlm_ctxt
112 {
113 struct list_head list;
114 struct hlist_head **lockres_hash;
115 struct list_head dirty_list;
116 struct list_head purge_list;
117 struct list_head pending_asts;
118 struct list_head pending_basts;
119 struct list_head tracking_list;
120 unsigned int purge_count;
121 spinlock_t spinlock;
122 spinlock_t ast_lock;
123 spinlock_t track_lock;
124 char *name;
125 u8 node_num;
126 u32 key;
127 u8 joining_node;
128 u8 migrate_done; /* set to 1 means node has migrated all lock resources */
129 wait_queue_head_t dlm_join_events;
130 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
131 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
132 unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
133 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
134 struct dlm_recovery_ctxt reco;
135 spinlock_t master_lock;
136 struct hlist_head **master_hash;
137 struct list_head mle_hb_events;
138
139 /* these give a really vague idea of the system load */
140 atomic_t mle_tot_count[DLM_MLE_NUM_TYPES];
141 atomic_t mle_cur_count[DLM_MLE_NUM_TYPES];
142 atomic_t res_tot_count;
143 atomic_t res_cur_count;
144
145 struct dentry *dlm_debugfs_subroot;
146
147 /* NOTE: Next three are protected by dlm_domain_lock */
148 struct kref dlm_refs;
149 enum dlm_ctxt_state dlm_state;
150 unsigned int num_joins;
151
152 struct o2hb_callback_func dlm_hb_up;
153 struct o2hb_callback_func dlm_hb_down;
154 struct task_struct *dlm_thread_task;
155 struct task_struct *dlm_reco_thread_task;
156 struct workqueue_struct *dlm_worker;
157 wait_queue_head_t dlm_thread_wq;
158 wait_queue_head_t dlm_reco_thread_wq;
159 wait_queue_head_t ast_wq;
160 wait_queue_head_t migration_wq;
161
162 struct work_struct dispatched_work;
163 struct list_head work_list;
164 spinlock_t work_lock;
165 struct list_head dlm_domain_handlers;
166 struct list_head dlm_eviction_callbacks;
167
168 /* The filesystem specifies this at domain registration. We
169 * cache it here to know what to tell other nodes. */
170 struct dlm_protocol_version fs_locking_proto;
171 /* This is the inter-dlm communication version */
172 struct dlm_protocol_version dlm_locking_proto;
173 };
174
dlm_lockres_hash(struct dlm_ctxt * dlm,unsigned i)175 static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
176 {
177 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
178 }
179
dlm_master_hash(struct dlm_ctxt * dlm,unsigned i)180 static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm,
181 unsigned i)
182 {
183 return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] +
184 (i % DLM_BUCKETS_PER_PAGE);
185 }
186
187 /* these keventd work queue items are for less-frequently
188 * called functions that cannot be directly called from the
189 * net message handlers for some reason, usually because
190 * they need to send net messages of their own. */
191 void dlm_dispatch_work(struct work_struct *work);
192
193 struct dlm_lock_resource;
194 struct dlm_work_item;
195
196 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);
197
198 struct dlm_request_all_locks_priv
199 {
200 u8 reco_master;
201 u8 dead_node;
202 };
203
204 struct dlm_mig_lockres_priv
205 {
206 struct dlm_lock_resource *lockres;
207 u8 real_master;
208 u8 extra_ref;
209 };
210
211 struct dlm_assert_master_priv
212 {
213 struct dlm_lock_resource *lockres;
214 u8 request_from;
215 u32 flags;
216 unsigned ignore_higher:1;
217 };
218
219 struct dlm_deref_lockres_priv
220 {
221 struct dlm_lock_resource *deref_res;
222 u8 deref_node;
223 };
224
225 struct dlm_work_item
226 {
227 struct list_head list;
228 dlm_workfunc_t *func;
229 struct dlm_ctxt *dlm;
230 void *data;
231 union {
232 struct dlm_request_all_locks_priv ral;
233 struct dlm_mig_lockres_priv ml;
234 struct dlm_assert_master_priv am;
235 struct dlm_deref_lockres_priv dl;
236 } u;
237 };
238
dlm_init_work_item(struct dlm_ctxt * dlm,struct dlm_work_item * i,dlm_workfunc_t * f,void * data)239 static inline void dlm_init_work_item(struct dlm_ctxt *dlm,
240 struct dlm_work_item *i,
241 dlm_workfunc_t *f, void *data)
242 {
243 memset(i, 0, sizeof(*i));
244 i->func = f;
245 INIT_LIST_HEAD(&i->list);
246 i->data = data;
247 i->dlm = dlm; /* must have already done a dlm_grab on this! */
248 }
249
250
251
__dlm_set_joining_node(struct dlm_ctxt * dlm,u8 node)252 static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm,
253 u8 node)
254 {
255 assert_spin_locked(&dlm->spinlock);
256
257 dlm->joining_node = node;
258 wake_up(&dlm->dlm_join_events);
259 }
260
261 #define DLM_LOCK_RES_UNINITED 0x00000001
262 #define DLM_LOCK_RES_RECOVERING 0x00000002
263 #define DLM_LOCK_RES_READY 0x00000004
264 #define DLM_LOCK_RES_DIRTY 0x00000008
265 #define DLM_LOCK_RES_IN_PROGRESS 0x00000010
266 #define DLM_LOCK_RES_MIGRATING 0x00000020
267 #define DLM_LOCK_RES_DROPPING_REF 0x00000040
268 #define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000
269 #define DLM_LOCK_RES_SETREF_INPROG 0x00002000
270 #define DLM_LOCK_RES_RECOVERY_WAITING 0x00004000
271
272 /* max milliseconds to wait to sync up a network failure with a node death */
273 #define DLM_NODE_DEATH_WAIT_MAX (5 * 1000)
274
275 #define DLM_PURGE_INTERVAL_MS (8 * 1000)
276
277 struct dlm_lock_resource
278 {
279 /* WARNING: Please see the comment in dlm_init_lockres before
280 * adding fields here. */
281 struct hlist_node hash_node;
282 struct qstr lockname;
283 struct kref refs;
284
285 /*
286 * Please keep granted, converting, and blocked in this order,
287 * as some funcs want to iterate over all lists.
288 *
289 * All four lists are protected by the hash's reference.
290 */
291 struct list_head granted;
292 struct list_head converting;
293 struct list_head blocked;
294 struct list_head purge;
295
296 /*
297 * These two lists require you to hold an additional reference
298 * while they are on the list.
299 */
300 struct list_head dirty;
301 struct list_head recovering; // dlm_recovery_ctxt.resources list
302
303 /* Added during init and removed during release */
304 struct list_head tracking; /* dlm->tracking_list */
305
306 /* unused lock resources have their last_used stamped and are
307 * put on a list for the dlm thread to run. */
308 unsigned long last_used;
309
310 struct dlm_ctxt *dlm;
311
312 unsigned migration_pending:1;
313 atomic_t asts_reserved;
314 spinlock_t spinlock;
315 wait_queue_head_t wq;
316 u8 owner; //node which owns the lock resource, or unknown
317 u16 state;
318 char lvb[DLM_LVB_LEN];
319 unsigned int inflight_locks;
320 unsigned int inflight_assert_workers;
321 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
322 };
323
324 struct dlm_migratable_lock
325 {
326 __be64 cookie;
327
328 /* these 3 are just padding for the in-memory structure, but
329 * list and flags are actually used when sent over the wire */
330 __be16 pad1;
331 u8 list; // 0=granted, 1=converting, 2=blocked
332 u8 flags;
333
334 s8 type;
335 s8 convert_type;
336 s8 highest_blocked;
337 u8 node;
338 }; // 16 bytes
339
340 struct dlm_lock
341 {
342 struct dlm_migratable_lock ml;
343
344 struct list_head list;
345 struct list_head ast_list;
346 struct list_head bast_list;
347 struct dlm_lock_resource *lockres;
348 spinlock_t spinlock;
349 struct kref lock_refs;
350
351 // ast and bast must be callable while holding a spinlock!
352 dlm_astlockfunc_t *ast;
353 dlm_bastlockfunc_t *bast;
354 void *astdata;
355 struct dlm_lockstatus *lksb;
356 unsigned ast_pending:1,
357 bast_pending:1,
358 convert_pending:1,
359 lock_pending:1,
360 cancel_pending:1,
361 unlock_pending:1,
362 lksb_kernel_allocated:1;
363 };
364
365 enum dlm_lockres_list {
366 DLM_GRANTED_LIST = 0,
367 DLM_CONVERTING_LIST = 1,
368 DLM_BLOCKED_LIST = 2,
369 };
370
dlm_lvb_is_empty(char * lvb)371 static inline int dlm_lvb_is_empty(char *lvb)
372 {
373 int i;
374 for (i=0; i<DLM_LVB_LEN; i++)
375 if (lvb[i])
376 return 0;
377 return 1;
378 }
379
dlm_list_in_text(enum dlm_lockres_list idx)380 static inline char *dlm_list_in_text(enum dlm_lockres_list idx)
381 {
382 if (idx == DLM_GRANTED_LIST)
383 return "granted";
384 else if (idx == DLM_CONVERTING_LIST)
385 return "converting";
386 else if (idx == DLM_BLOCKED_LIST)
387 return "blocked";
388 else
389 return "unknown";
390 }
391
392 static inline struct list_head *
dlm_list_idx_to_ptr(struct dlm_lock_resource * res,enum dlm_lockres_list idx)393 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
394 {
395 struct list_head *ret = NULL;
396 if (idx == DLM_GRANTED_LIST)
397 ret = &res->granted;
398 else if (idx == DLM_CONVERTING_LIST)
399 ret = &res->converting;
400 else if (idx == DLM_BLOCKED_LIST)
401 ret = &res->blocked;
402 else
403 BUG();
404 return ret;
405 }
406
407
408
409
410 struct dlm_node_iter
411 {
412 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
413 int curnode;
414 };
415
416
417 enum {
418 DLM_MASTER_REQUEST_MSG = 500,
419 DLM_UNUSED_MSG1 = 501,
420 DLM_ASSERT_MASTER_MSG = 502,
421 DLM_CREATE_LOCK_MSG = 503,
422 DLM_CONVERT_LOCK_MSG = 504,
423 DLM_PROXY_AST_MSG = 505,
424 DLM_UNLOCK_LOCK_MSG = 506,
425 DLM_DEREF_LOCKRES_MSG = 507,
426 DLM_MIGRATE_REQUEST_MSG = 508,
427 DLM_MIG_LOCKRES_MSG = 509,
428 DLM_QUERY_JOIN_MSG = 510,
429 DLM_ASSERT_JOINED_MSG = 511,
430 DLM_CANCEL_JOIN_MSG = 512,
431 DLM_EXIT_DOMAIN_MSG = 513,
432 DLM_MASTER_REQUERY_MSG = 514,
433 DLM_LOCK_REQUEST_MSG = 515,
434 DLM_RECO_DATA_DONE_MSG = 516,
435 DLM_BEGIN_RECO_MSG = 517,
436 DLM_FINALIZE_RECO_MSG = 518,
437 DLM_QUERY_REGION = 519,
438 DLM_QUERY_NODEINFO = 520,
439 DLM_BEGIN_EXIT_DOMAIN_MSG = 521,
440 DLM_DEREF_LOCKRES_DONE = 522,
441 };
442
443 struct dlm_reco_node_data
444 {
445 int state;
446 u8 node_num;
447 struct list_head list;
448 };
449
450 enum {
451 DLM_RECO_NODE_DATA_DEAD = -1,
452 DLM_RECO_NODE_DATA_INIT = 0,
453 DLM_RECO_NODE_DATA_REQUESTING = 1,
454 DLM_RECO_NODE_DATA_REQUESTED = 2,
455 DLM_RECO_NODE_DATA_RECEIVING = 3,
456 DLM_RECO_NODE_DATA_DONE = 4,
457 DLM_RECO_NODE_DATA_FINALIZE_SENT = 5,
458 };
459
460
461 enum {
462 DLM_MASTER_RESP_NO = 0,
463 DLM_MASTER_RESP_YES = 1,
464 DLM_MASTER_RESP_MAYBE = 2,
465 DLM_MASTER_RESP_ERROR = 3,
466 };
467
468
469 struct dlm_master_request
470 {
471 u8 node_idx;
472 u8 namelen;
473 __be16 pad1;
474 __be32 flags;
475
476 u8 name[O2NM_MAX_NAME_LEN];
477 };
478
479 #define DLM_ASSERT_RESPONSE_REASSERT 0x00000001
480 #define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002
481
482 #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001
483 #define DLM_ASSERT_MASTER_REQUERY 0x00000002
484 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
485 struct dlm_assert_master
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_MIGRATE_RESPONSE_MASTERY_REF 0x00000001
496
497 struct dlm_migrate_request
498 {
499 u8 master;
500 u8 new_master;
501 u8 namelen;
502 u8 pad1;
503 __be32 pad2;
504 u8 name[O2NM_MAX_NAME_LEN];
505 };
506
507 struct dlm_master_requery
508 {
509 u8 pad1;
510 u8 pad2;
511 u8 node_idx;
512 u8 namelen;
513 __be32 pad3;
514 u8 name[O2NM_MAX_NAME_LEN];
515 };
516
517 #define DLM_MRES_RECOVERY 0x01
518 #define DLM_MRES_MIGRATION 0x02
519 #define DLM_MRES_ALL_DONE 0x04
520
521 /*
522 * We would like to get one whole lockres into a single network
523 * message whenever possible. Generally speaking, there will be
524 * at most one dlm_lock on a lockres for each node in the cluster,
525 * plus (infrequently) any additional locks coming in from userdlm.
526 *
527 * struct _dlm_lockres_page
528 * {
529 * dlm_migratable_lockres mres;
530 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
531 * u8 pad[DLM_MIG_LOCKRES_RESERVED];
532 * };
533 *
534 * from ../cluster/tcp.h
535 * O2NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg))
536 * (roughly 4080 bytes)
537 * and sizeof(dlm_migratable_lockres) = 112 bytes
538 * and sizeof(dlm_migratable_lock) = 16 bytes
539 *
540 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
541 * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
542 *
543 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) +
544 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED =
545 * NET_MAX_PAYLOAD_BYTES
546 * (240 * 16) + 112 + 128 = 4080
547 *
548 * So a lockres would need more than 240 locks before it would
549 * use more than one network packet to recover. Not too bad.
550 */
551 #define DLM_MAX_MIGRATABLE_LOCKS 240
552
553 struct dlm_migratable_lockres
554 {
555 u8 master;
556 u8 lockname_len;
557 u8 num_locks; // locks sent in this structure
558 u8 flags;
559 __be32 total_locks; // locks to be sent for this migration cookie
560 __be64 mig_cookie; // cookie for this lockres migration
561 // or zero if not needed
562 // 16 bytes
563 u8 lockname[DLM_LOCKID_NAME_MAX];
564 // 48 bytes
565 u8 lvb[DLM_LVB_LEN];
566 // 112 bytes
567 struct dlm_migratable_lock ml[]; // 16 bytes each, begins at byte 112
568 };
569 #define DLM_MIG_LOCKRES_MAX_LEN \
570 (sizeof(struct dlm_migratable_lockres) + \
571 (sizeof(struct dlm_migratable_lock) * \
572 DLM_MAX_MIGRATABLE_LOCKS) )
573
574 /* from above, 128 bytes
575 * for some undetermined future use */
576 #define DLM_MIG_LOCKRES_RESERVED (O2NET_MAX_PAYLOAD_BYTES - \
577 DLM_MIG_LOCKRES_MAX_LEN)
578
579 struct dlm_create_lock
580 {
581 __be64 cookie;
582
583 __be32 flags;
584 u8 pad1;
585 u8 node_idx;
586 s8 requested_type;
587 u8 namelen;
588
589 u8 name[O2NM_MAX_NAME_LEN];
590 };
591
592 struct dlm_convert_lock
593 {
594 __be64 cookie;
595
596 __be32 flags;
597 u8 pad1;
598 u8 node_idx;
599 s8 requested_type;
600 u8 namelen;
601
602 u8 name[O2NM_MAX_NAME_LEN];
603
604 s8 lvb[];
605 };
606 #define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN)
607
608 struct dlm_unlock_lock
609 {
610 __be64 cookie;
611
612 __be32 flags;
613 __be16 pad1;
614 u8 node_idx;
615 u8 namelen;
616
617 u8 name[O2NM_MAX_NAME_LEN];
618
619 s8 lvb[];
620 };
621 #define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN)
622
623 struct dlm_proxy_ast
624 {
625 __be64 cookie;
626
627 __be32 flags;
628 u8 node_idx;
629 u8 type;
630 u8 blocked_type;
631 u8 namelen;
632
633 u8 name[O2NM_MAX_NAME_LEN];
634
635 s8 lvb[];
636 };
637 #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN)
638
639 #define DLM_MOD_KEY (0x666c6172)
640 enum dlm_query_join_response_code {
641 JOIN_DISALLOW = 0,
642 JOIN_OK = 1,
643 JOIN_OK_NO_MAP = 2,
644 JOIN_PROTOCOL_MISMATCH = 3,
645 };
646
647 struct dlm_query_join_packet {
648 u8 code; /* Response code. dlm_minor and fs_minor
649 are only valid if this is JOIN_OK */
650 u8 dlm_minor; /* The minor version of the protocol the
651 dlm is speaking. */
652 u8 fs_minor; /* The minor version of the protocol the
653 filesystem is speaking. */
654 u8 reserved;
655 };
656
657 union dlm_query_join_response {
658 __be32 intval;
659 struct dlm_query_join_packet packet;
660 };
661
662 struct dlm_lock_request
663 {
664 u8 node_idx;
665 u8 dead_node;
666 __be16 pad1;
667 __be32 pad2;
668 };
669
670 struct dlm_reco_data_done
671 {
672 u8 node_idx;
673 u8 dead_node;
674 __be16 pad1;
675 __be32 pad2;
676
677 /* unused for now */
678 /* eventually we can use this to attempt
679 * lvb recovery based on each node's info */
680 u8 reco_lvb[DLM_LVB_LEN];
681 };
682
683 struct dlm_begin_reco
684 {
685 u8 node_idx;
686 u8 dead_node;
687 __be16 pad1;
688 __be32 pad2;
689 };
690
691 struct dlm_query_join_request
692 {
693 u8 node_idx;
694 u8 pad1[2];
695 u8 name_len;
696 struct dlm_protocol_version dlm_proto;
697 struct dlm_protocol_version fs_proto;
698 u8 domain[O2NM_MAX_NAME_LEN];
699 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)];
700 };
701
702 struct dlm_assert_joined
703 {
704 u8 node_idx;
705 u8 pad1[2];
706 u8 name_len;
707 u8 domain[O2NM_MAX_NAME_LEN];
708 };
709
710 struct dlm_cancel_join
711 {
712 u8 node_idx;
713 u8 pad1[2];
714 u8 name_len;
715 u8 domain[O2NM_MAX_NAME_LEN];
716 };
717
718 struct dlm_query_region {
719 u8 qr_node;
720 u8 qr_numregions;
721 u8 qr_namelen;
722 u8 pad1;
723 u8 qr_domain[O2NM_MAX_NAME_LEN];
724 u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS];
725 };
726
727 struct dlm_node_info {
728 u8 ni_nodenum;
729 u8 pad1;
730 __be16 ni_ipv4_port;
731 __be32 ni_ipv4_address;
732 };
733
734 struct dlm_query_nodeinfo {
735 u8 qn_nodenum;
736 u8 qn_numnodes;
737 u8 qn_namelen;
738 u8 pad1;
739 u8 qn_domain[O2NM_MAX_NAME_LEN];
740 struct dlm_node_info qn_nodes[O2NM_MAX_NODES];
741 };
742
743 struct dlm_exit_domain
744 {
745 u8 node_idx;
746 u8 pad1[3];
747 };
748
749 struct dlm_finalize_reco
750 {
751 u8 node_idx;
752 u8 dead_node;
753 u8 flags;
754 u8 pad1;
755 __be32 pad2;
756 };
757
758 struct dlm_deref_lockres
759 {
760 u32 pad1;
761 u16 pad2;
762 u8 node_idx;
763 u8 namelen;
764
765 u8 name[O2NM_MAX_NAME_LEN];
766 };
767
768 enum {
769 DLM_DEREF_RESPONSE_DONE = 0,
770 DLM_DEREF_RESPONSE_INPROG = 1,
771 };
772
773 struct dlm_deref_lockres_done {
774 u32 pad1;
775 u16 pad2;
776 u8 node_idx;
777 u8 namelen;
778
779 u8 name[O2NM_MAX_NAME_LEN];
780 };
781
782 static inline enum dlm_status
__dlm_lockres_state_to_status(struct dlm_lock_resource * res)783 __dlm_lockres_state_to_status(struct dlm_lock_resource *res)
784 {
785 enum dlm_status status = DLM_NORMAL;
786
787 assert_spin_locked(&res->spinlock);
788
789 if (res->state & (DLM_LOCK_RES_RECOVERING|
790 DLM_LOCK_RES_RECOVERY_WAITING))
791 status = DLM_RECOVERING;
792 else if (res->state & DLM_LOCK_RES_MIGRATING)
793 status = DLM_MIGRATING;
794 else if (res->state & DLM_LOCK_RES_IN_PROGRESS)
795 status = DLM_FORWARD;
796
797 return status;
798 }
799
dlm_get_lock_cookie_node(u64 cookie)800 static inline u8 dlm_get_lock_cookie_node(u64 cookie)
801 {
802 u8 ret;
803 cookie >>= 56;
804 ret = (u8)(cookie & 0xffULL);
805 return ret;
806 }
807
dlm_get_lock_cookie_seq(u64 cookie)808 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
809 {
810 unsigned long long ret;
811 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
812 return ret;
813 }
814
815 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
816 struct dlm_lockstatus *lksb);
817 void dlm_lock_get(struct dlm_lock *lock);
818 void dlm_lock_put(struct dlm_lock *lock);
819
820 void dlm_lock_attach_lockres(struct dlm_lock *lock,
821 struct dlm_lock_resource *res);
822
823 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
824 void **ret_data);
825 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
826 void **ret_data);
827 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
828 void **ret_data);
829
830 void dlm_revert_pending_convert(struct dlm_lock_resource *res,
831 struct dlm_lock *lock);
832 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
833 struct dlm_lock *lock);
834
835 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
836 void **ret_data);
837 void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
838 struct dlm_lock *lock);
839 void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
840 struct dlm_lock *lock);
841
842 int dlm_launch_thread(struct dlm_ctxt *dlm);
843 void dlm_complete_thread(struct dlm_ctxt *dlm);
844 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
845 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
846 void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
847 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
848 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
849 void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
850 void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
851
852 void dlm_put(struct dlm_ctxt *dlm);
853 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
854 int dlm_domain_fully_joined(struct dlm_ctxt *dlm);
855
856 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
857 struct dlm_lock_resource *res);
858 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
859 struct dlm_lock_resource *res);
dlm_lockres_get(struct dlm_lock_resource * res)860 static inline void dlm_lockres_get(struct dlm_lock_resource *res)
861 {
862 /* This is called on every lookup, so it might be worth
863 * inlining. */
864 kref_get(&res->refs);
865 }
866 void dlm_lockres_put(struct dlm_lock_resource *res);
867 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
868 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
869 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
870 const char *name,
871 unsigned int len,
872 unsigned int hash);
873 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
874 const char *name,
875 unsigned int len,
876 unsigned int hash);
877 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
878 const char *name,
879 unsigned int len);
880
881 int dlm_is_host_down(int errno);
882
883 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
884 const char *lockid,
885 int namelen,
886 int flags);
887 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
888 const char *name,
889 unsigned int namelen);
890
891 void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
892 struct dlm_lock_resource *res, int bit);
893 void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
894 struct dlm_lock_resource *res, int bit);
895
896 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
897 struct dlm_lock_resource *res);
898 void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
899 struct dlm_lock_resource *res);
900
901 void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm,
902 struct dlm_lock_resource *res);
903
904 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
905 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
906 void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
907 void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
908 void dlm_do_local_ast(struct dlm_ctxt *dlm,
909 struct dlm_lock_resource *res,
910 struct dlm_lock *lock);
911 int dlm_do_remote_ast(struct dlm_ctxt *dlm,
912 struct dlm_lock_resource *res,
913 struct dlm_lock *lock);
914 void dlm_do_local_bast(struct dlm_ctxt *dlm,
915 struct dlm_lock_resource *res,
916 struct dlm_lock *lock,
917 int blocked_type);
918 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm,
919 struct dlm_lock_resource *res,
920 struct dlm_lock *lock,
921 int msg_type,
922 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)923 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm,
924 struct dlm_lock_resource *res,
925 struct dlm_lock *lock,
926 int blocked_type)
927 {
928 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST,
929 blocked_type, 0);
930 }
931
dlm_send_proxy_ast(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags)932 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm,
933 struct dlm_lock_resource *res,
934 struct dlm_lock *lock,
935 int flags)
936 {
937 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST,
938 0, flags);
939 }
940
941 void dlm_print_one_lock_resource(struct dlm_lock_resource *res);
942 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res);
943
944 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
945 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
946
947
948 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data);
949 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data);
950
951 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
952 int dlm_finish_migration(struct dlm_ctxt *dlm,
953 struct dlm_lock_resource *res,
954 u8 old_master);
955 void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
956 struct dlm_lock_resource *res);
957 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res);
958
959 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
960 void **ret_data);
961 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
962 void **ret_data);
963 void dlm_assert_master_post_handler(int status, void *data, void *ret_data);
964 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
965 void **ret_data);
966 int dlm_deref_lockres_done_handler(struct o2net_msg *msg, u32 len, void *data,
967 void **ret_data);
968 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
969 void **ret_data);
970 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
971 void **ret_data);
972 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
973 void **ret_data);
974 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
975 void **ret_data);
976 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
977 void **ret_data);
978 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
979 void **ret_data);
980 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
981 void **ret_data);
982 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
983 u8 nodenum, u8 *real_master);
984
985 void __dlm_do_purge_lockres(struct dlm_ctxt *dlm,
986 struct dlm_lock_resource *res);
987
988 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
989 struct dlm_lock_resource *res,
990 int ignore_higher,
991 u8 request_from,
992 u32 flags);
993
994
995 int dlm_send_one_lockres(struct dlm_ctxt *dlm,
996 struct dlm_lock_resource *res,
997 struct dlm_migratable_lockres *mres,
998 u8 send_to,
999 u8 flags);
1000 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
1001 struct dlm_lock_resource *res);
1002
1003 /* will exit holding res->spinlock, but may drop in function */
1004 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags);
1005
1006 /* will exit holding res->spinlock, but may drop in function */
__dlm_wait_on_lockres(struct dlm_lock_resource * res)1007 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res)
1008 {
1009 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS|
1010 DLM_LOCK_RES_RECOVERING|
1011 DLM_LOCK_RES_RECOVERY_WAITING|
1012 DLM_LOCK_RES_MIGRATING));
1013 }
1014
1015 void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1016 void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1017
1018 /* create/destroy slab caches */
1019 int dlm_init_master_caches(void);
1020 void dlm_destroy_master_caches(void);
1021
1022 int dlm_init_lock_cache(void);
1023 void dlm_destroy_lock_cache(void);
1024
1025 int dlm_init_mle_cache(void);
1026 void dlm_destroy_mle_cache(void);
1027
1028 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up);
1029 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
1030 struct dlm_lock_resource *res);
1031 void dlm_clean_master_list(struct dlm_ctxt *dlm,
1032 u8 dead_node);
1033 void dlm_force_free_mles(struct dlm_ctxt *dlm);
1034 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
1035 int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
1036 int __dlm_lockres_unused(struct dlm_lock_resource *res);
1037
dlm_lock_mode_name(int mode)1038 static inline const char * dlm_lock_mode_name(int mode)
1039 {
1040 switch (mode) {
1041 case LKM_EXMODE:
1042 return "EX";
1043 case LKM_PRMODE:
1044 return "PR";
1045 case LKM_NLMODE:
1046 return "NL";
1047 }
1048 return "UNKNOWN";
1049 }
1050
1051
dlm_lock_compatible(int existing,int request)1052 static inline int dlm_lock_compatible(int existing, int request)
1053 {
1054 /* NO_LOCK compatible with all */
1055 if (request == LKM_NLMODE ||
1056 existing == LKM_NLMODE)
1057 return 1;
1058
1059 /* EX incompatible with all non-NO_LOCK */
1060 if (request == LKM_EXMODE)
1061 return 0;
1062
1063 /* request must be PR, which is compatible with PR */
1064 if (existing == LKM_PRMODE)
1065 return 1;
1066
1067 return 0;
1068 }
1069
dlm_lock_on_list(struct list_head * head,struct dlm_lock * lock)1070 static inline int dlm_lock_on_list(struct list_head *head,
1071 struct dlm_lock *lock)
1072 {
1073 struct dlm_lock *tmplock;
1074
1075 list_for_each_entry(tmplock, head, list) {
1076 if (tmplock == lock)
1077 return 1;
1078 }
1079 return 0;
1080 }
1081
1082
dlm_err_to_dlm_status(int err)1083 static inline enum dlm_status dlm_err_to_dlm_status(int err)
1084 {
1085 enum dlm_status ret;
1086 if (err == -ENOMEM)
1087 ret = DLM_SYSERR;
1088 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL))
1089 ret = DLM_NOLOCKMGR;
1090 else if (err == -EINVAL)
1091 ret = DLM_BADPARAM;
1092 else if (err == -ENAMETOOLONG)
1093 ret = DLM_IVBUFLEN;
1094 else
1095 ret = DLM_BADARGS;
1096 return ret;
1097 }
1098
1099
dlm_node_iter_init(unsigned long * map,struct dlm_node_iter * iter)1100 static inline void dlm_node_iter_init(unsigned long *map,
1101 struct dlm_node_iter *iter)
1102 {
1103 memcpy(iter->node_map, map, sizeof(iter->node_map));
1104 iter->curnode = -1;
1105 }
1106
dlm_node_iter_next(struct dlm_node_iter * iter)1107 static inline int dlm_node_iter_next(struct dlm_node_iter *iter)
1108 {
1109 int bit;
1110 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1);
1111 if (bit >= O2NM_MAX_NODES) {
1112 iter->curnode = O2NM_MAX_NODES;
1113 return -ENOENT;
1114 }
1115 iter->curnode = bit;
1116 return bit;
1117 }
1118
dlm_set_lockres_owner(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 owner)1119 static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
1120 struct dlm_lock_resource *res,
1121 u8 owner)
1122 {
1123 assert_spin_locked(&res->spinlock);
1124
1125 res->owner = owner;
1126 }
1127
dlm_change_lockres_owner(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 owner)1128 static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
1129 struct dlm_lock_resource *res,
1130 u8 owner)
1131 {
1132 assert_spin_locked(&res->spinlock);
1133
1134 if (owner != res->owner)
1135 dlm_set_lockres_owner(dlm, res, owner);
1136 }
1137
1138 #endif /* DLMCOMMON_H */
1139