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 << 14)
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(_n, _l)
51
52 enum dlm_mle_type {
53 DLM_MLE_BLOCK,
54 DLM_MLE_MASTER,
55 DLM_MLE_MIGRATION
56 };
57
58 struct dlm_lock_name {
59 u8 len;
60 u8 name[DLM_LOCKID_NAME_MAX];
61 };
62
63 struct dlm_master_list_entry {
64 struct list_head list;
65 struct list_head hb_events;
66 struct dlm_ctxt *dlm;
67 spinlock_t spinlock;
68 wait_queue_head_t wq;
69 atomic_t woken;
70 struct kref mle_refs;
71 int inuse;
72 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
73 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
74 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
75 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
76 u8 master;
77 u8 new_master;
78 enum dlm_mle_type type;
79 struct o2hb_callback_func mle_hb_up;
80 struct o2hb_callback_func mle_hb_down;
81 union {
82 struct dlm_lock_resource *res;
83 struct dlm_lock_name name;
84 } u;
85 };
86
87 enum dlm_ast_type {
88 DLM_AST = 0,
89 DLM_BAST,
90 DLM_ASTUNLOCK
91 };
92
93
94 #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
95 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
96 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)
97
98 #define DLM_RECOVERY_LOCK_NAME "$RECOVERY"
99 #define DLM_RECOVERY_LOCK_NAME_LEN 9
100
dlm_is_recovery_lock(const char * lock_name,int name_len)101 static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
102 {
103 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
104 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
105 return 1;
106 return 0;
107 }
108
109 #define DLM_RECO_STATE_ACTIVE 0x0001
110 #define DLM_RECO_STATE_FINALIZE 0x0002
111
112 struct dlm_recovery_ctxt
113 {
114 struct list_head resources;
115 struct list_head received;
116 struct list_head node_data;
117 u8 new_master;
118 u8 dead_node;
119 u16 state;
120 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
121 wait_queue_head_t event;
122 };
123
124 enum dlm_ctxt_state {
125 DLM_CTXT_NEW = 0,
126 DLM_CTXT_JOINED,
127 DLM_CTXT_IN_SHUTDOWN,
128 DLM_CTXT_LEAVING,
129 };
130
131 struct dlm_ctxt
132 {
133 struct list_head list;
134 struct hlist_head **lockres_hash;
135 struct list_head dirty_list;
136 struct list_head purge_list;
137 struct list_head pending_asts;
138 struct list_head pending_basts;
139 struct list_head tracking_list;
140 unsigned int purge_count;
141 spinlock_t spinlock;
142 spinlock_t ast_lock;
143 spinlock_t track_lock;
144 char *name;
145 u8 node_num;
146 u32 key;
147 u8 joining_node;
148 wait_queue_head_t dlm_join_events;
149 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
150 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
151 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
152 struct dlm_recovery_ctxt reco;
153 spinlock_t master_lock;
154 struct list_head master_list;
155 struct list_head mle_hb_events;
156
157 /* these give a really vague idea of the system load */
158 atomic_t local_resources;
159 atomic_t remote_resources;
160 atomic_t unknown_resources;
161
162 struct dlm_debug_ctxt *dlm_debug_ctxt;
163 struct dentry *dlm_debugfs_subroot;
164
165 /* NOTE: Next three are protected by dlm_domain_lock */
166 struct kref dlm_refs;
167 enum dlm_ctxt_state dlm_state;
168 unsigned int num_joins;
169
170 struct o2hb_callback_func dlm_hb_up;
171 struct o2hb_callback_func dlm_hb_down;
172 struct task_struct *dlm_thread_task;
173 struct task_struct *dlm_reco_thread_task;
174 struct workqueue_struct *dlm_worker;
175 wait_queue_head_t dlm_thread_wq;
176 wait_queue_head_t dlm_reco_thread_wq;
177 wait_queue_head_t ast_wq;
178 wait_queue_head_t migration_wq;
179
180 struct work_struct dispatched_work;
181 struct list_head work_list;
182 spinlock_t work_lock;
183 struct list_head dlm_domain_handlers;
184 struct list_head dlm_eviction_callbacks;
185
186 /* The filesystem specifies this at domain registration. We
187 * cache it here to know what to tell other nodes. */
188 struct dlm_protocol_version fs_locking_proto;
189 /* This is the inter-dlm communication version */
190 struct dlm_protocol_version dlm_locking_proto;
191 };
192
dlm_lockres_hash(struct dlm_ctxt * dlm,unsigned i)193 static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
194 {
195 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
196 }
197
198 /* these keventd work queue items are for less-frequently
199 * called functions that cannot be directly called from the
200 * net message handlers for some reason, usually because
201 * they need to send net messages of their own. */
202 void dlm_dispatch_work(struct work_struct *work);
203
204 struct dlm_lock_resource;
205 struct dlm_work_item;
206
207 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);
208
209 struct dlm_request_all_locks_priv
210 {
211 u8 reco_master;
212 u8 dead_node;
213 };
214
215 struct dlm_mig_lockres_priv
216 {
217 struct dlm_lock_resource *lockres;
218 u8 real_master;
219 u8 extra_ref;
220 };
221
222 struct dlm_assert_master_priv
223 {
224 struct dlm_lock_resource *lockres;
225 u8 request_from;
226 u32 flags;
227 unsigned ignore_higher:1;
228 };
229
230 struct dlm_deref_lockres_priv
231 {
232 struct dlm_lock_resource *deref_res;
233 u8 deref_node;
234 };
235
236 struct dlm_work_item
237 {
238 struct list_head list;
239 dlm_workfunc_t *func;
240 struct dlm_ctxt *dlm;
241 void *data;
242 union {
243 struct dlm_request_all_locks_priv ral;
244 struct dlm_mig_lockres_priv ml;
245 struct dlm_assert_master_priv am;
246 struct dlm_deref_lockres_priv dl;
247 } u;
248 };
249
dlm_init_work_item(struct dlm_ctxt * dlm,struct dlm_work_item * i,dlm_workfunc_t * f,void * data)250 static inline void dlm_init_work_item(struct dlm_ctxt *dlm,
251 struct dlm_work_item *i,
252 dlm_workfunc_t *f, void *data)
253 {
254 memset(i, 0, sizeof(*i));
255 i->func = f;
256 INIT_LIST_HEAD(&i->list);
257 i->data = data;
258 i->dlm = dlm; /* must have already done a dlm_grab on this! */
259 }
260
261
262
__dlm_set_joining_node(struct dlm_ctxt * dlm,u8 node)263 static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm,
264 u8 node)
265 {
266 assert_spin_locked(&dlm->spinlock);
267
268 dlm->joining_node = node;
269 wake_up(&dlm->dlm_join_events);
270 }
271
272 #define DLM_LOCK_RES_UNINITED 0x00000001
273 #define DLM_LOCK_RES_RECOVERING 0x00000002
274 #define DLM_LOCK_RES_READY 0x00000004
275 #define DLM_LOCK_RES_DIRTY 0x00000008
276 #define DLM_LOCK_RES_IN_PROGRESS 0x00000010
277 #define DLM_LOCK_RES_MIGRATING 0x00000020
278 #define DLM_LOCK_RES_DROPPING_REF 0x00000040
279 #define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000
280 #define DLM_LOCK_RES_SETREF_INPROG 0x00002000
281
282 /* max milliseconds to wait to sync up a network failure with a node death */
283 #define DLM_NODE_DEATH_WAIT_MAX (5 * 1000)
284
285 #define DLM_PURGE_INTERVAL_MS (8 * 1000)
286
287 struct dlm_lock_resource
288 {
289 /* WARNING: Please see the comment in dlm_init_lockres before
290 * adding fields here. */
291 struct hlist_node hash_node;
292 struct qstr lockname;
293 struct kref refs;
294
295 /*
296 * Please keep granted, converting, and blocked in this order,
297 * as some funcs want to iterate over all lists.
298 *
299 * All four lists are protected by the hash's reference.
300 */
301 struct list_head granted;
302 struct list_head converting;
303 struct list_head blocked;
304 struct list_head purge;
305
306 /*
307 * These two lists require you to hold an additional reference
308 * while they are on the list.
309 */
310 struct list_head dirty;
311 struct list_head recovering; // dlm_recovery_ctxt.resources list
312
313 /* Added during init and removed during release */
314 struct list_head tracking; /* dlm->tracking_list */
315
316 /* unused lock resources have their last_used stamped and are
317 * put on a list for the dlm thread to run. */
318 unsigned long last_used;
319
320 struct dlm_ctxt *dlm;
321
322 unsigned migration_pending:1;
323 atomic_t asts_reserved;
324 spinlock_t spinlock;
325 wait_queue_head_t wq;
326 u8 owner; //node which owns the lock resource, or unknown
327 u16 state;
328 char lvb[DLM_LVB_LEN];
329 unsigned int inflight_locks;
330 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
331 };
332
333 struct dlm_migratable_lock
334 {
335 __be64 cookie;
336
337 /* these 3 are just padding for the in-memory structure, but
338 * list and flags are actually used when sent over the wire */
339 __be16 pad1;
340 u8 list; // 0=granted, 1=converting, 2=blocked
341 u8 flags;
342
343 s8 type;
344 s8 convert_type;
345 s8 highest_blocked;
346 u8 node;
347 }; // 16 bytes
348
349 struct dlm_lock
350 {
351 struct dlm_migratable_lock ml;
352
353 struct list_head list;
354 struct list_head ast_list;
355 struct list_head bast_list;
356 struct dlm_lock_resource *lockres;
357 spinlock_t spinlock;
358 struct kref lock_refs;
359
360 // ast and bast must be callable while holding a spinlock!
361 dlm_astlockfunc_t *ast;
362 dlm_bastlockfunc_t *bast;
363 void *astdata;
364 struct dlm_lockstatus *lksb;
365 unsigned ast_pending:1,
366 bast_pending:1,
367 convert_pending:1,
368 lock_pending:1,
369 cancel_pending:1,
370 unlock_pending:1,
371 lksb_kernel_allocated:1;
372 };
373
374
375 #define DLM_LKSB_UNUSED1 0x01
376 #define DLM_LKSB_PUT_LVB 0x02
377 #define DLM_LKSB_GET_LVB 0x04
378 #define DLM_LKSB_UNUSED2 0x08
379 #define DLM_LKSB_UNUSED3 0x10
380 #define DLM_LKSB_UNUSED4 0x20
381 #define DLM_LKSB_UNUSED5 0x40
382 #define DLM_LKSB_UNUSED6 0x80
383
384
385 enum dlm_lockres_list {
386 DLM_GRANTED_LIST = 0,
387 DLM_CONVERTING_LIST,
388 DLM_BLOCKED_LIST
389 };
390
dlm_lvb_is_empty(char * lvb)391 static inline int dlm_lvb_is_empty(char *lvb)
392 {
393 int i;
394 for (i=0; i<DLM_LVB_LEN; i++)
395 if (lvb[i])
396 return 0;
397 return 1;
398 }
399
400 static inline struct list_head *
dlm_list_idx_to_ptr(struct dlm_lock_resource * res,enum dlm_lockres_list idx)401 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
402 {
403 struct list_head *ret = NULL;
404 if (idx == DLM_GRANTED_LIST)
405 ret = &res->granted;
406 else if (idx == DLM_CONVERTING_LIST)
407 ret = &res->converting;
408 else if (idx == DLM_BLOCKED_LIST)
409 ret = &res->blocked;
410 else
411 BUG();
412 return ret;
413 }
414
415
416
417
418 struct dlm_node_iter
419 {
420 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
421 int curnode;
422 };
423
424
425 enum {
426 DLM_MASTER_REQUEST_MSG = 500,
427 DLM_UNUSED_MSG1, /* 501 */
428 DLM_ASSERT_MASTER_MSG, /* 502 */
429 DLM_CREATE_LOCK_MSG, /* 503 */
430 DLM_CONVERT_LOCK_MSG, /* 504 */
431 DLM_PROXY_AST_MSG, /* 505 */
432 DLM_UNLOCK_LOCK_MSG, /* 506 */
433 DLM_DEREF_LOCKRES_MSG, /* 507 */
434 DLM_MIGRATE_REQUEST_MSG, /* 508 */
435 DLM_MIG_LOCKRES_MSG, /* 509 */
436 DLM_QUERY_JOIN_MSG, /* 510 */
437 DLM_ASSERT_JOINED_MSG, /* 511 */
438 DLM_CANCEL_JOIN_MSG, /* 512 */
439 DLM_EXIT_DOMAIN_MSG, /* 513 */
440 DLM_MASTER_REQUERY_MSG, /* 514 */
441 DLM_LOCK_REQUEST_MSG, /* 515 */
442 DLM_RECO_DATA_DONE_MSG, /* 516 */
443 DLM_BEGIN_RECO_MSG, /* 517 */
444 DLM_FINALIZE_RECO_MSG /* 518 */
445 };
446
447 struct dlm_reco_node_data
448 {
449 int state;
450 u8 node_num;
451 struct list_head list;
452 };
453
454 enum {
455 DLM_RECO_NODE_DATA_DEAD = -1,
456 DLM_RECO_NODE_DATA_INIT = 0,
457 DLM_RECO_NODE_DATA_REQUESTING,
458 DLM_RECO_NODE_DATA_REQUESTED,
459 DLM_RECO_NODE_DATA_RECEIVING,
460 DLM_RECO_NODE_DATA_DONE,
461 DLM_RECO_NODE_DATA_FINALIZE_SENT,
462 };
463
464
465 enum {
466 DLM_MASTER_RESP_NO = 0,
467 DLM_MASTER_RESP_YES,
468 DLM_MASTER_RESP_MAYBE,
469 DLM_MASTER_RESP_ERROR
470 };
471
472
473 struct dlm_master_request
474 {
475 u8 node_idx;
476 u8 namelen;
477 __be16 pad1;
478 __be32 flags;
479
480 u8 name[O2NM_MAX_NAME_LEN];
481 };
482
483 #define DLM_ASSERT_RESPONSE_REASSERT 0x00000001
484 #define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002
485
486 #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001
487 #define DLM_ASSERT_MASTER_REQUERY 0x00000002
488 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
489 struct dlm_assert_master
490 {
491 u8 node_idx;
492 u8 namelen;
493 __be16 pad1;
494 __be32 flags;
495
496 u8 name[O2NM_MAX_NAME_LEN];
497 };
498
499 #define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001
500
501 struct dlm_migrate_request
502 {
503 u8 master;
504 u8 new_master;
505 u8 namelen;
506 u8 pad1;
507 __be32 pad2;
508 u8 name[O2NM_MAX_NAME_LEN];
509 };
510
511 struct dlm_master_requery
512 {
513 u8 pad1;
514 u8 pad2;
515 u8 node_idx;
516 u8 namelen;
517 __be32 pad3;
518 u8 name[O2NM_MAX_NAME_LEN];
519 };
520
521 #define DLM_MRES_RECOVERY 0x01
522 #define DLM_MRES_MIGRATION 0x02
523 #define DLM_MRES_ALL_DONE 0x04
524
525 /*
526 * We would like to get one whole lockres into a single network
527 * message whenever possible. Generally speaking, there will be
528 * at most one dlm_lock on a lockres for each node in the cluster,
529 * plus (infrequently) any additional locks coming in from userdlm.
530 *
531 * struct _dlm_lockres_page
532 * {
533 * dlm_migratable_lockres mres;
534 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
535 * u8 pad[DLM_MIG_LOCKRES_RESERVED];
536 * };
537 *
538 * from ../cluster/tcp.h
539 * NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg))
540 * (roughly 4080 bytes)
541 * and sizeof(dlm_migratable_lockres) = 112 bytes
542 * and sizeof(dlm_migratable_lock) = 16 bytes
543 *
544 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
545 * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
546 *
547 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) +
548 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED =
549 * NET_MAX_PAYLOAD_BYTES
550 * (240 * 16) + 112 + 128 = 4080
551 *
552 * So a lockres would need more than 240 locks before it would
553 * use more than one network packet to recover. Not too bad.
554 */
555 #define DLM_MAX_MIGRATABLE_LOCKS 240
556
557 struct dlm_migratable_lockres
558 {
559 u8 master;
560 u8 lockname_len;
561 u8 num_locks; // locks sent in this structure
562 u8 flags;
563 __be32 total_locks; // locks to be sent for this migration cookie
564 __be64 mig_cookie; // cookie for this lockres migration
565 // or zero if not needed
566 // 16 bytes
567 u8 lockname[DLM_LOCKID_NAME_MAX];
568 // 48 bytes
569 u8 lvb[DLM_LVB_LEN];
570 // 112 bytes
571 struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112
572 };
573 #define DLM_MIG_LOCKRES_MAX_LEN \
574 (sizeof(struct dlm_migratable_lockres) + \
575 (sizeof(struct dlm_migratable_lock) * \
576 DLM_MAX_MIGRATABLE_LOCKS) )
577
578 /* from above, 128 bytes
579 * for some undetermined future use */
580 #define DLM_MIG_LOCKRES_RESERVED (NET_MAX_PAYLOAD_BYTES - \
581 DLM_MIG_LOCKRES_MAX_LEN)
582
583 struct dlm_create_lock
584 {
585 __be64 cookie;
586
587 __be32 flags;
588 u8 pad1;
589 u8 node_idx;
590 s8 requested_type;
591 u8 namelen;
592
593 u8 name[O2NM_MAX_NAME_LEN];
594 };
595
596 struct dlm_convert_lock
597 {
598 __be64 cookie;
599
600 __be32 flags;
601 u8 pad1;
602 u8 node_idx;
603 s8 requested_type;
604 u8 namelen;
605
606 u8 name[O2NM_MAX_NAME_LEN];
607
608 s8 lvb[0];
609 };
610 #define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN)
611
612 struct dlm_unlock_lock
613 {
614 __be64 cookie;
615
616 __be32 flags;
617 __be16 pad1;
618 u8 node_idx;
619 u8 namelen;
620
621 u8 name[O2NM_MAX_NAME_LEN];
622
623 s8 lvb[0];
624 };
625 #define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN)
626
627 struct dlm_proxy_ast
628 {
629 __be64 cookie;
630
631 __be32 flags;
632 u8 node_idx;
633 u8 type;
634 u8 blocked_type;
635 u8 namelen;
636
637 u8 name[O2NM_MAX_NAME_LEN];
638
639 s8 lvb[0];
640 };
641 #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN)
642
643 #define DLM_MOD_KEY (0x666c6172)
644 enum dlm_query_join_response_code {
645 JOIN_DISALLOW = 0,
646 JOIN_OK,
647 JOIN_OK_NO_MAP,
648 JOIN_PROTOCOL_MISMATCH,
649 };
650
651 struct dlm_query_join_packet {
652 u8 code; /* Response code. dlm_minor and fs_minor
653 are only valid if this is JOIN_OK */
654 u8 dlm_minor; /* The minor version of the protocol the
655 dlm is speaking. */
656 u8 fs_minor; /* The minor version of the protocol the
657 filesystem is speaking. */
658 u8 reserved;
659 };
660
661 union dlm_query_join_response {
662 u32 intval;
663 struct dlm_query_join_packet packet;
664 };
665
666 struct dlm_lock_request
667 {
668 u8 node_idx;
669 u8 dead_node;
670 __be16 pad1;
671 __be32 pad2;
672 };
673
674 struct dlm_reco_data_done
675 {
676 u8 node_idx;
677 u8 dead_node;
678 __be16 pad1;
679 __be32 pad2;
680
681 /* unused for now */
682 /* eventually we can use this to attempt
683 * lvb recovery based on each node's info */
684 u8 reco_lvb[DLM_LVB_LEN];
685 };
686
687 struct dlm_begin_reco
688 {
689 u8 node_idx;
690 u8 dead_node;
691 __be16 pad1;
692 __be32 pad2;
693 };
694
695
696 #define BITS_PER_BYTE 8
697 #define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE)
698
699 struct dlm_query_join_request
700 {
701 u8 node_idx;
702 u8 pad1[2];
703 u8 name_len;
704 struct dlm_protocol_version dlm_proto;
705 struct dlm_protocol_version fs_proto;
706 u8 domain[O2NM_MAX_NAME_LEN];
707 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)];
708 };
709
710 struct dlm_assert_joined
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_cancel_join
719 {
720 u8 node_idx;
721 u8 pad1[2];
722 u8 name_len;
723 u8 domain[O2NM_MAX_NAME_LEN];
724 };
725
726 struct dlm_exit_domain
727 {
728 u8 node_idx;
729 u8 pad1[3];
730 };
731
732 struct dlm_finalize_reco
733 {
734 u8 node_idx;
735 u8 dead_node;
736 u8 flags;
737 u8 pad1;
738 __be32 pad2;
739 };
740
741 struct dlm_deref_lockres
742 {
743 u32 pad1;
744 u16 pad2;
745 u8 node_idx;
746 u8 namelen;
747
748 u8 name[O2NM_MAX_NAME_LEN];
749 };
750
751 static inline enum dlm_status
__dlm_lockres_state_to_status(struct dlm_lock_resource * res)752 __dlm_lockres_state_to_status(struct dlm_lock_resource *res)
753 {
754 enum dlm_status status = DLM_NORMAL;
755
756 assert_spin_locked(&res->spinlock);
757
758 if (res->state & DLM_LOCK_RES_RECOVERING)
759 status = DLM_RECOVERING;
760 else if (res->state & DLM_LOCK_RES_MIGRATING)
761 status = DLM_MIGRATING;
762 else if (res->state & DLM_LOCK_RES_IN_PROGRESS)
763 status = DLM_FORWARD;
764
765 return status;
766 }
767
dlm_get_lock_cookie_node(u64 cookie)768 static inline u8 dlm_get_lock_cookie_node(u64 cookie)
769 {
770 u8 ret;
771 cookie >>= 56;
772 ret = (u8)(cookie & 0xffULL);
773 return ret;
774 }
775
dlm_get_lock_cookie_seq(u64 cookie)776 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
777 {
778 unsigned long long ret;
779 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
780 return ret;
781 }
782
783 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
784 struct dlm_lockstatus *lksb);
785 void dlm_lock_get(struct dlm_lock *lock);
786 void dlm_lock_put(struct dlm_lock *lock);
787
788 void dlm_lock_attach_lockres(struct dlm_lock *lock,
789 struct dlm_lock_resource *res);
790
791 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
792 void **ret_data);
793 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
794 void **ret_data);
795 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
796 void **ret_data);
797
798 void dlm_revert_pending_convert(struct dlm_lock_resource *res,
799 struct dlm_lock *lock);
800 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
801 struct dlm_lock *lock);
802
803 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
804 void **ret_data);
805 void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
806 struct dlm_lock *lock);
807 void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
808 struct dlm_lock *lock);
809
810 int dlm_launch_thread(struct dlm_ctxt *dlm);
811 void dlm_complete_thread(struct dlm_ctxt *dlm);
812 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
813 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
814 void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
815 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
816 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
817 int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
818 int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
819
820 void dlm_put(struct dlm_ctxt *dlm);
821 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
822 int dlm_domain_fully_joined(struct dlm_ctxt *dlm);
823
824 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
825 struct dlm_lock_resource *res);
826 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
827 struct dlm_lock_resource *res);
dlm_lockres_get(struct dlm_lock_resource * res)828 static inline void dlm_lockres_get(struct dlm_lock_resource *res)
829 {
830 /* This is called on every lookup, so it might be worth
831 * inlining. */
832 kref_get(&res->refs);
833 }
834 void dlm_lockres_put(struct dlm_lock_resource *res);
835 void __dlm_unhash_lockres(struct dlm_lock_resource *res);
836 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
837 struct dlm_lock_resource *res);
838 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
839 const char *name,
840 unsigned int len,
841 unsigned int hash);
842 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
843 const char *name,
844 unsigned int len,
845 unsigned int hash);
846 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
847 const char *name,
848 unsigned int len);
849
850 int dlm_is_host_down(int errno);
851 void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
852 struct dlm_lock_resource *res,
853 u8 owner);
854 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
855 const char *lockid,
856 int namelen,
857 int flags);
858 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
859 const char *name,
860 unsigned int namelen);
861
862 #define dlm_lockres_set_refmap_bit(bit,res) \
863 __dlm_lockres_set_refmap_bit(bit,res,__FILE__,__LINE__)
864 #define dlm_lockres_clear_refmap_bit(bit,res) \
865 __dlm_lockres_clear_refmap_bit(bit,res,__FILE__,__LINE__)
866
__dlm_lockres_set_refmap_bit(int bit,struct dlm_lock_resource * res,const char * file,int line)867 static inline void __dlm_lockres_set_refmap_bit(int bit,
868 struct dlm_lock_resource *res,
869 const char *file,
870 int line)
871 {
872 //printk("%s:%d:%.*s: setting bit %d\n", file, line,
873 // res->lockname.len, res->lockname.name, bit);
874 set_bit(bit, res->refmap);
875 }
876
__dlm_lockres_clear_refmap_bit(int bit,struct dlm_lock_resource * res,const char * file,int line)877 static inline void __dlm_lockres_clear_refmap_bit(int bit,
878 struct dlm_lock_resource *res,
879 const char *file,
880 int line)
881 {
882 //printk("%s:%d:%.*s: clearing bit %d\n", file, line,
883 // res->lockname.len, res->lockname.name, bit);
884 clear_bit(bit, res->refmap);
885 }
886
887 void __dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
888 struct dlm_lock_resource *res,
889 const char *file,
890 int line);
891 void __dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
892 struct dlm_lock_resource *res,
893 int new_lockres,
894 const char *file,
895 int line);
896 #define dlm_lockres_drop_inflight_ref(d,r) \
897 __dlm_lockres_drop_inflight_ref(d,r,__FILE__,__LINE__)
898 #define dlm_lockres_grab_inflight_ref(d,r) \
899 __dlm_lockres_grab_inflight_ref(d,r,0,__FILE__,__LINE__)
900 #define dlm_lockres_grab_inflight_ref_new(d,r) \
901 __dlm_lockres_grab_inflight_ref(d,r,1,__FILE__,__LINE__)
902
903 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
904 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
905 void dlm_do_local_ast(struct dlm_ctxt *dlm,
906 struct dlm_lock_resource *res,
907 struct dlm_lock *lock);
908 int dlm_do_remote_ast(struct dlm_ctxt *dlm,
909 struct dlm_lock_resource *res,
910 struct dlm_lock *lock);
911 void dlm_do_local_bast(struct dlm_ctxt *dlm,
912 struct dlm_lock_resource *res,
913 struct dlm_lock *lock,
914 int blocked_type);
915 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm,
916 struct dlm_lock_resource *res,
917 struct dlm_lock *lock,
918 int msg_type,
919 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)920 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm,
921 struct dlm_lock_resource *res,
922 struct dlm_lock *lock,
923 int blocked_type)
924 {
925 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST,
926 blocked_type, 0);
927 }
928
dlm_send_proxy_ast(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags)929 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm,
930 struct dlm_lock_resource *res,
931 struct dlm_lock *lock,
932 int flags)
933 {
934 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST,
935 0, flags);
936 }
937
938 void dlm_print_one_lock_resource(struct dlm_lock_resource *res);
939 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res);
940
941 u8 dlm_nm_this_node(struct dlm_ctxt *dlm);
942 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
943 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
944
945
946 int dlm_nm_init(struct dlm_ctxt *dlm);
947 int dlm_heartbeat_init(struct dlm_ctxt *dlm);
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_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
967 void **ret_data);
968 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
969 void **ret_data);
970 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
971 void **ret_data);
972 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
973 void **ret_data);
974 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
975 void **ret_data);
976 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
977 void **ret_data);
978 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
979 void **ret_data);
980 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
981 u8 nodenum, u8 *real_master);
982
983
984 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
985 struct dlm_lock_resource *res,
986 int ignore_higher,
987 u8 request_from,
988 u32 flags);
989
990
991 int dlm_send_one_lockres(struct dlm_ctxt *dlm,
992 struct dlm_lock_resource *res,
993 struct dlm_migratable_lockres *mres,
994 u8 send_to,
995 u8 flags);
996 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
997 struct dlm_lock_resource *res);
998
999 /* will exit holding res->spinlock, but may drop in function */
1000 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags);
1001 void __dlm_wait_on_lockres_flags_set(struct dlm_lock_resource *res, int flags);
1002
1003 /* will exit holding res->spinlock, but may drop in function */
__dlm_wait_on_lockres(struct dlm_lock_resource * res)1004 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res)
1005 {
1006 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS|
1007 DLM_LOCK_RES_RECOVERING|
1008 DLM_LOCK_RES_MIGRATING));
1009 }
1010
1011 /* create/destroy slab caches */
1012 int dlm_init_master_caches(void);
1013 void dlm_destroy_master_caches(void);
1014
1015 int dlm_init_lock_cache(void);
1016 void dlm_destroy_lock_cache(void);
1017
1018 int dlm_init_mle_cache(void);
1019 void dlm_destroy_mle_cache(void);
1020
1021 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up);
1022 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
1023 struct dlm_lock_resource *res);
1024 void dlm_clean_master_list(struct dlm_ctxt *dlm,
1025 u8 dead_node);
1026 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
1027 int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
1028 int __dlm_lockres_unused(struct dlm_lock_resource *res);
1029
dlm_lock_mode_name(int mode)1030 static inline const char * dlm_lock_mode_name(int mode)
1031 {
1032 switch (mode) {
1033 case LKM_EXMODE:
1034 return "EX";
1035 case LKM_PRMODE:
1036 return "PR";
1037 case LKM_NLMODE:
1038 return "NL";
1039 }
1040 return "UNKNOWN";
1041 }
1042
1043
dlm_lock_compatible(int existing,int request)1044 static inline int dlm_lock_compatible(int existing, int request)
1045 {
1046 /* NO_LOCK compatible with all */
1047 if (request == LKM_NLMODE ||
1048 existing == LKM_NLMODE)
1049 return 1;
1050
1051 /* EX incompatible with all non-NO_LOCK */
1052 if (request == LKM_EXMODE)
1053 return 0;
1054
1055 /* request must be PR, which is compatible with PR */
1056 if (existing == LKM_PRMODE)
1057 return 1;
1058
1059 return 0;
1060 }
1061
dlm_lock_on_list(struct list_head * head,struct dlm_lock * lock)1062 static inline int dlm_lock_on_list(struct list_head *head,
1063 struct dlm_lock *lock)
1064 {
1065 struct list_head *iter;
1066 struct dlm_lock *tmplock;
1067
1068 list_for_each(iter, head) {
1069 tmplock = list_entry(iter, struct dlm_lock, list);
1070 if (tmplock == lock)
1071 return 1;
1072 }
1073 return 0;
1074 }
1075
1076
dlm_err_to_dlm_status(int err)1077 static inline enum dlm_status dlm_err_to_dlm_status(int err)
1078 {
1079 enum dlm_status ret;
1080 if (err == -ENOMEM)
1081 ret = DLM_SYSERR;
1082 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL))
1083 ret = DLM_NOLOCKMGR;
1084 else if (err == -EINVAL)
1085 ret = DLM_BADPARAM;
1086 else if (err == -ENAMETOOLONG)
1087 ret = DLM_IVBUFLEN;
1088 else
1089 ret = DLM_BADARGS;
1090 return ret;
1091 }
1092
1093
dlm_node_iter_init(unsigned long * map,struct dlm_node_iter * iter)1094 static inline void dlm_node_iter_init(unsigned long *map,
1095 struct dlm_node_iter *iter)
1096 {
1097 memcpy(iter->node_map, map, sizeof(iter->node_map));
1098 iter->curnode = -1;
1099 }
1100
dlm_node_iter_next(struct dlm_node_iter * iter)1101 static inline int dlm_node_iter_next(struct dlm_node_iter *iter)
1102 {
1103 int bit;
1104 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1);
1105 if (bit >= O2NM_MAX_NODES) {
1106 iter->curnode = O2NM_MAX_NODES;
1107 return -ENOENT;
1108 }
1109 iter->curnode = bit;
1110 return bit;
1111 }
1112
1113
1114
1115 #endif /* DLMCOMMON_H */
1116