Commit | Line | Data |
---|---|---|
6714d8e8 KH |
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 | ||
c8f33b6e JB |
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 | |
03d864c0 DP |
46 | #define DLM_BUCKETS_PER_PAGE (PAGE_SIZE / sizeof(struct hlist_head)) |
47 | #define DLM_HASH_BUCKETS (DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE) | |
6714d8e8 | 48 | |
a3d33291 MF |
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 | ||
6714d8e8 KH |
52 | enum dlm_ast_type { |
53 | DLM_AST = 0, | |
54 | DLM_BAST, | |
55 | DLM_ASTUNLOCK | |
56 | }; | |
57 | ||
58 | ||
59 | #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \ | |
60 | LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \ | |
61 | LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE) | |
62 | ||
63 | #define DLM_RECOVERY_LOCK_NAME "$RECOVERY" | |
64 | #define DLM_RECOVERY_LOCK_NAME_LEN 9 | |
65 | ||
66 | static inline int dlm_is_recovery_lock(const char *lock_name, int name_len) | |
67 | { | |
68 | if (name_len == DLM_RECOVERY_LOCK_NAME_LEN && | |
69 | memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0) | |
70 | return 1; | |
71 | return 0; | |
72 | } | |
73 | ||
466d1a45 KH |
74 | #define DLM_RECO_STATE_ACTIVE 0x0001 |
75 | #define DLM_RECO_STATE_FINALIZE 0x0002 | |
6714d8e8 KH |
76 | |
77 | struct dlm_recovery_ctxt | |
78 | { | |
79 | struct list_head resources; | |
80 | struct list_head received; | |
81 | struct list_head node_data; | |
82 | u8 new_master; | |
83 | u8 dead_node; | |
84 | u16 state; | |
85 | unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
86 | wait_queue_head_t event; | |
87 | }; | |
88 | ||
89 | enum dlm_ctxt_state { | |
90 | DLM_CTXT_NEW = 0, | |
91 | DLM_CTXT_JOINED, | |
92 | DLM_CTXT_IN_SHUTDOWN, | |
93 | DLM_CTXT_LEAVING, | |
94 | }; | |
95 | ||
96 | struct dlm_ctxt | |
97 | { | |
98 | struct list_head list; | |
03d864c0 | 99 | struct hlist_head **lockres_hash; |
6714d8e8 KH |
100 | struct list_head dirty_list; |
101 | struct list_head purge_list; | |
102 | struct list_head pending_asts; | |
103 | struct list_head pending_basts; | |
104 | unsigned int purge_count; | |
105 | spinlock_t spinlock; | |
106 | spinlock_t ast_lock; | |
107 | char *name; | |
108 | u8 node_num; | |
109 | u32 key; | |
110 | u8 joining_node; | |
111 | wait_queue_head_t dlm_join_events; | |
112 | unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
113 | unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
114 | unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
115 | struct dlm_recovery_ctxt reco; | |
116 | spinlock_t master_lock; | |
117 | struct list_head master_list; | |
118 | struct list_head mle_hb_events; | |
119 | ||
120 | /* these give a really vague idea of the system load */ | |
121 | atomic_t local_resources; | |
122 | atomic_t remote_resources; | |
123 | atomic_t unknown_resources; | |
124 | ||
125 | /* NOTE: Next three are protected by dlm_domain_lock */ | |
126 | struct kref dlm_refs; | |
127 | enum dlm_ctxt_state dlm_state; | |
128 | unsigned int num_joins; | |
129 | ||
130 | struct o2hb_callback_func dlm_hb_up; | |
131 | struct o2hb_callback_func dlm_hb_down; | |
132 | struct task_struct *dlm_thread_task; | |
133 | struct task_struct *dlm_reco_thread_task; | |
3156d267 | 134 | struct workqueue_struct *dlm_worker; |
6714d8e8 KH |
135 | wait_queue_head_t dlm_thread_wq; |
136 | wait_queue_head_t dlm_reco_thread_wq; | |
137 | wait_queue_head_t ast_wq; | |
138 | wait_queue_head_t migration_wq; | |
139 | ||
140 | struct work_struct dispatched_work; | |
141 | struct list_head work_list; | |
142 | spinlock_t work_lock; | |
143 | struct list_head dlm_domain_handlers; | |
144 | struct list_head dlm_eviction_callbacks; | |
145 | }; | |
146 | ||
03d864c0 DP |
147 | static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i) |
148 | { | |
149 | return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE); | |
150 | } | |
151 | ||
6714d8e8 KH |
152 | /* these keventd work queue items are for less-frequently |
153 | * called functions that cannot be directly called from the | |
154 | * net message handlers for some reason, usually because | |
155 | * they need to send net messages of their own. */ | |
c4028958 | 156 | void dlm_dispatch_work(struct work_struct *work); |
6714d8e8 KH |
157 | |
158 | struct dlm_lock_resource; | |
159 | struct dlm_work_item; | |
160 | ||
161 | typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *); | |
162 | ||
163 | struct dlm_request_all_locks_priv | |
164 | { | |
165 | u8 reco_master; | |
166 | u8 dead_node; | |
167 | }; | |
168 | ||
169 | struct dlm_mig_lockres_priv | |
170 | { | |
171 | struct dlm_lock_resource *lockres; | |
172 | u8 real_master; | |
173 | }; | |
174 | ||
175 | struct dlm_assert_master_priv | |
176 | { | |
177 | struct dlm_lock_resource *lockres; | |
178 | u8 request_from; | |
179 | u32 flags; | |
180 | unsigned ignore_higher:1; | |
181 | }; | |
182 | ||
183 | ||
184 | struct dlm_work_item | |
185 | { | |
186 | struct list_head list; | |
187 | dlm_workfunc_t *func; | |
188 | struct dlm_ctxt *dlm; | |
189 | void *data; | |
190 | union { | |
191 | struct dlm_request_all_locks_priv ral; | |
192 | struct dlm_mig_lockres_priv ml; | |
193 | struct dlm_assert_master_priv am; | |
194 | } u; | |
195 | }; | |
196 | ||
197 | static inline void dlm_init_work_item(struct dlm_ctxt *dlm, | |
198 | struct dlm_work_item *i, | |
199 | dlm_workfunc_t *f, void *data) | |
200 | { | |
201 | memset(i, 0, sizeof(*i)); | |
202 | i->func = f; | |
203 | INIT_LIST_HEAD(&i->list); | |
204 | i->data = data; | |
205 | i->dlm = dlm; /* must have already done a dlm_grab on this! */ | |
206 | } | |
207 | ||
208 | ||
209 | ||
210 | static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm, | |
211 | u8 node) | |
212 | { | |
213 | assert_spin_locked(&dlm->spinlock); | |
214 | ||
215 | dlm->joining_node = node; | |
216 | wake_up(&dlm->dlm_join_events); | |
217 | } | |
218 | ||
219 | #define DLM_LOCK_RES_UNINITED 0x00000001 | |
220 | #define DLM_LOCK_RES_RECOVERING 0x00000002 | |
221 | #define DLM_LOCK_RES_READY 0x00000004 | |
222 | #define DLM_LOCK_RES_DIRTY 0x00000008 | |
223 | #define DLM_LOCK_RES_IN_PROGRESS 0x00000010 | |
224 | #define DLM_LOCK_RES_MIGRATING 0x00000020 | |
ba2bf218 | 225 | #define DLM_LOCK_RES_DROPPING_REF 0x00000040 |
6714d8e8 | 226 | |
44465a7d KH |
227 | /* max milliseconds to wait to sync up a network failure with a node death */ |
228 | #define DLM_NODE_DEATH_WAIT_MAX (5 * 1000) | |
229 | ||
6714d8e8 KH |
230 | #define DLM_PURGE_INTERVAL_MS (8 * 1000) |
231 | ||
232 | struct dlm_lock_resource | |
233 | { | |
234 | /* WARNING: Please see the comment in dlm_init_lockres before | |
235 | * adding fields here. */ | |
81f2094a | 236 | struct hlist_node hash_node; |
65c491d8 | 237 | struct qstr lockname; |
6714d8e8 KH |
238 | struct kref refs; |
239 | ||
6ff06a93 KH |
240 | /* |
241 | * Please keep granted, converting, and blocked in this order, | |
242 | * as some funcs want to iterate over all lists. | |
243 | * | |
244 | * All four lists are protected by the hash's reference. | |
245 | */ | |
6714d8e8 KH |
246 | struct list_head granted; |
247 | struct list_head converting; | |
248 | struct list_head blocked; | |
6ff06a93 | 249 | struct list_head purge; |
6714d8e8 | 250 | |
6ff06a93 KH |
251 | /* |
252 | * These two lists require you to hold an additional reference | |
253 | * while they are on the list. | |
254 | */ | |
6714d8e8 KH |
255 | struct list_head dirty; |
256 | struct list_head recovering; // dlm_recovery_ctxt.resources list | |
257 | ||
258 | /* unused lock resources have their last_used stamped and are | |
259 | * put on a list for the dlm thread to run. */ | |
6714d8e8 KH |
260 | unsigned long last_used; |
261 | ||
262 | unsigned migration_pending:1; | |
263 | atomic_t asts_reserved; | |
264 | spinlock_t spinlock; | |
265 | wait_queue_head_t wq; | |
266 | u8 owner; //node which owns the lock resource, or unknown | |
267 | u16 state; | |
6714d8e8 | 268 | char lvb[DLM_LVB_LEN]; |
ba2bf218 KH |
269 | unsigned int inflight_locks; |
270 | unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
6714d8e8 KH |
271 | }; |
272 | ||
273 | struct dlm_migratable_lock | |
274 | { | |
275 | __be64 cookie; | |
276 | ||
277 | /* these 3 are just padding for the in-memory structure, but | |
278 | * list and flags are actually used when sent over the wire */ | |
279 | __be16 pad1; | |
280 | u8 list; // 0=granted, 1=converting, 2=blocked | |
281 | u8 flags; | |
282 | ||
283 | s8 type; | |
284 | s8 convert_type; | |
285 | s8 highest_blocked; | |
286 | u8 node; | |
287 | }; // 16 bytes | |
288 | ||
289 | struct dlm_lock | |
290 | { | |
291 | struct dlm_migratable_lock ml; | |
292 | ||
293 | struct list_head list; | |
294 | struct list_head ast_list; | |
295 | struct list_head bast_list; | |
296 | struct dlm_lock_resource *lockres; | |
297 | spinlock_t spinlock; | |
298 | struct kref lock_refs; | |
299 | ||
300 | // ast and bast must be callable while holding a spinlock! | |
301 | dlm_astlockfunc_t *ast; | |
302 | dlm_bastlockfunc_t *bast; | |
303 | void *astdata; | |
304 | struct dlm_lockstatus *lksb; | |
305 | unsigned ast_pending:1, | |
306 | bast_pending:1, | |
307 | convert_pending:1, | |
308 | lock_pending:1, | |
309 | cancel_pending:1, | |
310 | unlock_pending:1, | |
311 | lksb_kernel_allocated:1; | |
312 | }; | |
313 | ||
314 | ||
315 | #define DLM_LKSB_UNUSED1 0x01 | |
316 | #define DLM_LKSB_PUT_LVB 0x02 | |
317 | #define DLM_LKSB_GET_LVB 0x04 | |
318 | #define DLM_LKSB_UNUSED2 0x08 | |
319 | #define DLM_LKSB_UNUSED3 0x10 | |
320 | #define DLM_LKSB_UNUSED4 0x20 | |
321 | #define DLM_LKSB_UNUSED5 0x40 | |
322 | #define DLM_LKSB_UNUSED6 0x80 | |
323 | ||
324 | ||
325 | enum dlm_lockres_list { | |
326 | DLM_GRANTED_LIST = 0, | |
327 | DLM_CONVERTING_LIST, | |
328 | DLM_BLOCKED_LIST | |
329 | }; | |
330 | ||
8bc674cb KH |
331 | static inline int dlm_lvb_is_empty(char *lvb) |
332 | { | |
333 | int i; | |
334 | for (i=0; i<DLM_LVB_LEN; i++) | |
335 | if (lvb[i]) | |
336 | return 0; | |
337 | return 1; | |
338 | } | |
339 | ||
6714d8e8 KH |
340 | static inline struct list_head * |
341 | dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx) | |
342 | { | |
343 | struct list_head *ret = NULL; | |
344 | if (idx == DLM_GRANTED_LIST) | |
345 | ret = &res->granted; | |
346 | else if (idx == DLM_CONVERTING_LIST) | |
347 | ret = &res->converting; | |
348 | else if (idx == DLM_BLOCKED_LIST) | |
349 | ret = &res->blocked; | |
350 | else | |
351 | BUG(); | |
352 | return ret; | |
353 | } | |
354 | ||
355 | ||
356 | ||
357 | ||
358 | struct dlm_node_iter | |
359 | { | |
360 | unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
361 | int curnode; | |
362 | }; | |
363 | ||
364 | ||
365 | enum { | |
366 | DLM_MASTER_REQUEST_MSG = 500, | |
367 | DLM_UNUSED_MSG1, /* 501 */ | |
368 | DLM_ASSERT_MASTER_MSG, /* 502 */ | |
369 | DLM_CREATE_LOCK_MSG, /* 503 */ | |
370 | DLM_CONVERT_LOCK_MSG, /* 504 */ | |
371 | DLM_PROXY_AST_MSG, /* 505 */ | |
372 | DLM_UNLOCK_LOCK_MSG, /* 506 */ | |
ba2bf218 | 373 | DLM_DEREF_LOCKRES_MSG, /* 507 */ |
6714d8e8 KH |
374 | DLM_MIGRATE_REQUEST_MSG, /* 508 */ |
375 | DLM_MIG_LOCKRES_MSG, /* 509 */ | |
376 | DLM_QUERY_JOIN_MSG, /* 510 */ | |
377 | DLM_ASSERT_JOINED_MSG, /* 511 */ | |
378 | DLM_CANCEL_JOIN_MSG, /* 512 */ | |
379 | DLM_EXIT_DOMAIN_MSG, /* 513 */ | |
380 | DLM_MASTER_REQUERY_MSG, /* 514 */ | |
381 | DLM_LOCK_REQUEST_MSG, /* 515 */ | |
382 | DLM_RECO_DATA_DONE_MSG, /* 516 */ | |
383 | DLM_BEGIN_RECO_MSG, /* 517 */ | |
384 | DLM_FINALIZE_RECO_MSG /* 518 */ | |
385 | }; | |
386 | ||
387 | struct dlm_reco_node_data | |
388 | { | |
389 | int state; | |
390 | u8 node_num; | |
391 | struct list_head list; | |
392 | }; | |
393 | ||
394 | enum { | |
395 | DLM_RECO_NODE_DATA_DEAD = -1, | |
396 | DLM_RECO_NODE_DATA_INIT = 0, | |
397 | DLM_RECO_NODE_DATA_REQUESTING, | |
398 | DLM_RECO_NODE_DATA_REQUESTED, | |
399 | DLM_RECO_NODE_DATA_RECEIVING, | |
400 | DLM_RECO_NODE_DATA_DONE, | |
401 | DLM_RECO_NODE_DATA_FINALIZE_SENT, | |
402 | }; | |
403 | ||
404 | ||
405 | enum { | |
406 | DLM_MASTER_RESP_NO = 0, | |
407 | DLM_MASTER_RESP_YES, | |
408 | DLM_MASTER_RESP_MAYBE, | |
409 | DLM_MASTER_RESP_ERROR | |
410 | }; | |
411 | ||
412 | ||
413 | struct dlm_master_request | |
414 | { | |
415 | u8 node_idx; | |
416 | u8 namelen; | |
417 | __be16 pad1; | |
418 | __be32 flags; | |
419 | ||
420 | u8 name[O2NM_MAX_NAME_LEN]; | |
421 | }; | |
422 | ||
ba2bf218 KH |
423 | #define DLM_ASSERT_RESPONSE_REASSERT 0x00000001 |
424 | #define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002 | |
425 | ||
6714d8e8 KH |
426 | #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001 |
427 | #define DLM_ASSERT_MASTER_REQUERY 0x00000002 | |
428 | #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004 | |
429 | struct dlm_assert_master | |
430 | { | |
431 | u8 node_idx; | |
432 | u8 namelen; | |
433 | __be16 pad1; | |
434 | __be32 flags; | |
435 | ||
436 | u8 name[O2NM_MAX_NAME_LEN]; | |
437 | }; | |
438 | ||
ba2bf218 KH |
439 | #define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001 |
440 | ||
6714d8e8 KH |
441 | struct dlm_migrate_request |
442 | { | |
443 | u8 master; | |
444 | u8 new_master; | |
445 | u8 namelen; | |
446 | u8 pad1; | |
447 | __be32 pad2; | |
448 | u8 name[O2NM_MAX_NAME_LEN]; | |
449 | }; | |
450 | ||
451 | struct dlm_master_requery | |
452 | { | |
453 | u8 pad1; | |
454 | u8 pad2; | |
455 | u8 node_idx; | |
456 | u8 namelen; | |
457 | __be32 pad3; | |
458 | u8 name[O2NM_MAX_NAME_LEN]; | |
459 | }; | |
460 | ||
461 | #define DLM_MRES_RECOVERY 0x01 | |
462 | #define DLM_MRES_MIGRATION 0x02 | |
463 | #define DLM_MRES_ALL_DONE 0x04 | |
464 | ||
465 | /* | |
466 | * We would like to get one whole lockres into a single network | |
467 | * message whenever possible. Generally speaking, there will be | |
468 | * at most one dlm_lock on a lockres for each node in the cluster, | |
469 | * plus (infrequently) any additional locks coming in from userdlm. | |
470 | * | |
471 | * struct _dlm_lockres_page | |
472 | * { | |
473 | * dlm_migratable_lockres mres; | |
474 | * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS]; | |
475 | * u8 pad[DLM_MIG_LOCKRES_RESERVED]; | |
476 | * }; | |
477 | * | |
478 | * from ../cluster/tcp.h | |
479 | * NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg)) | |
480 | * (roughly 4080 bytes) | |
481 | * and sizeof(dlm_migratable_lockres) = 112 bytes | |
482 | * and sizeof(dlm_migratable_lock) = 16 bytes | |
483 | * | |
484 | * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and | |
485 | * DLM_MIG_LOCKRES_RESERVED=128 means we have this: | |
486 | * | |
487 | * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) + | |
488 | * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED = | |
489 | * NET_MAX_PAYLOAD_BYTES | |
490 | * (240 * 16) + 112 + 128 = 4080 | |
491 | * | |
492 | * So a lockres would need more than 240 locks before it would | |
493 | * use more than one network packet to recover. Not too bad. | |
494 | */ | |
495 | #define DLM_MAX_MIGRATABLE_LOCKS 240 | |
496 | ||
497 | struct dlm_migratable_lockres | |
498 | { | |
499 | u8 master; | |
500 | u8 lockname_len; | |
501 | u8 num_locks; // locks sent in this structure | |
502 | u8 flags; | |
503 | __be32 total_locks; // locks to be sent for this migration cookie | |
504 | __be64 mig_cookie; // cookie for this lockres migration | |
505 | // or zero if not needed | |
506 | // 16 bytes | |
507 | u8 lockname[DLM_LOCKID_NAME_MAX]; | |
508 | // 48 bytes | |
509 | u8 lvb[DLM_LVB_LEN]; | |
510 | // 112 bytes | |
511 | struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112 | |
512 | }; | |
513 | #define DLM_MIG_LOCKRES_MAX_LEN \ | |
514 | (sizeof(struct dlm_migratable_lockres) + \ | |
515 | (sizeof(struct dlm_migratable_lock) * \ | |
516 | DLM_MAX_MIGRATABLE_LOCKS) ) | |
517 | ||
518 | /* from above, 128 bytes | |
519 | * for some undetermined future use */ | |
520 | #define DLM_MIG_LOCKRES_RESERVED (NET_MAX_PAYLOAD_BYTES - \ | |
521 | DLM_MIG_LOCKRES_MAX_LEN) | |
522 | ||
523 | struct dlm_create_lock | |
524 | { | |
525 | __be64 cookie; | |
526 | ||
527 | __be32 flags; | |
528 | u8 pad1; | |
529 | u8 node_idx; | |
530 | s8 requested_type; | |
531 | u8 namelen; | |
532 | ||
533 | u8 name[O2NM_MAX_NAME_LEN]; | |
534 | }; | |
535 | ||
536 | struct dlm_convert_lock | |
537 | { | |
538 | __be64 cookie; | |
539 | ||
540 | __be32 flags; | |
541 | u8 pad1; | |
542 | u8 node_idx; | |
543 | s8 requested_type; | |
544 | u8 namelen; | |
545 | ||
546 | u8 name[O2NM_MAX_NAME_LEN]; | |
547 | ||
548 | s8 lvb[0]; | |
549 | }; | |
550 | #define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN) | |
551 | ||
552 | struct dlm_unlock_lock | |
553 | { | |
554 | __be64 cookie; | |
555 | ||
556 | __be32 flags; | |
557 | __be16 pad1; | |
558 | u8 node_idx; | |
559 | u8 namelen; | |
560 | ||
561 | u8 name[O2NM_MAX_NAME_LEN]; | |
562 | ||
563 | s8 lvb[0]; | |
564 | }; | |
565 | #define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN) | |
566 | ||
567 | struct dlm_proxy_ast | |
568 | { | |
569 | __be64 cookie; | |
570 | ||
571 | __be32 flags; | |
572 | u8 node_idx; | |
573 | u8 type; | |
574 | u8 blocked_type; | |
575 | u8 namelen; | |
576 | ||
577 | u8 name[O2NM_MAX_NAME_LEN]; | |
578 | ||
579 | s8 lvb[0]; | |
580 | }; | |
581 | #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN) | |
582 | ||
583 | #define DLM_MOD_KEY (0x666c6172) | |
584 | enum dlm_query_join_response { | |
585 | JOIN_DISALLOW = 0, | |
586 | JOIN_OK, | |
587 | JOIN_OK_NO_MAP, | |
588 | }; | |
589 | ||
590 | struct dlm_lock_request | |
591 | { | |
592 | u8 node_idx; | |
593 | u8 dead_node; | |
594 | __be16 pad1; | |
595 | __be32 pad2; | |
596 | }; | |
597 | ||
598 | struct dlm_reco_data_done | |
599 | { | |
600 | u8 node_idx; | |
601 | u8 dead_node; | |
602 | __be16 pad1; | |
603 | __be32 pad2; | |
604 | ||
605 | /* unused for now */ | |
606 | /* eventually we can use this to attempt | |
607 | * lvb recovery based on each node's info */ | |
608 | u8 reco_lvb[DLM_LVB_LEN]; | |
609 | }; | |
610 | ||
611 | struct dlm_begin_reco | |
612 | { | |
613 | u8 node_idx; | |
614 | u8 dead_node; | |
615 | __be16 pad1; | |
616 | __be32 pad2; | |
617 | }; | |
618 | ||
619 | ||
620 | struct dlm_query_join_request | |
621 | { | |
622 | u8 node_idx; | |
623 | u8 pad1[2]; | |
624 | u8 name_len; | |
625 | u8 domain[O2NM_MAX_NAME_LEN]; | |
626 | }; | |
627 | ||
628 | struct dlm_assert_joined | |
629 | { | |
630 | u8 node_idx; | |
631 | u8 pad1[2]; | |
632 | u8 name_len; | |
633 | u8 domain[O2NM_MAX_NAME_LEN]; | |
634 | }; | |
635 | ||
636 | struct dlm_cancel_join | |
637 | { | |
638 | u8 node_idx; | |
639 | u8 pad1[2]; | |
640 | u8 name_len; | |
641 | u8 domain[O2NM_MAX_NAME_LEN]; | |
642 | }; | |
643 | ||
644 | struct dlm_exit_domain | |
645 | { | |
646 | u8 node_idx; | |
647 | u8 pad1[3]; | |
648 | }; | |
649 | ||
650 | struct dlm_finalize_reco | |
651 | { | |
652 | u8 node_idx; | |
653 | u8 dead_node; | |
466d1a45 KH |
654 | u8 flags; |
655 | u8 pad1; | |
6714d8e8 KH |
656 | __be32 pad2; |
657 | }; | |
658 | ||
ba2bf218 KH |
659 | struct dlm_deref_lockres |
660 | { | |
661 | u32 pad1; | |
662 | u16 pad2; | |
663 | u8 node_idx; | |
664 | u8 namelen; | |
665 | ||
666 | u8 name[O2NM_MAX_NAME_LEN]; | |
667 | }; | |
668 | ||
6714d8e8 KH |
669 | static inline enum dlm_status |
670 | __dlm_lockres_state_to_status(struct dlm_lock_resource *res) | |
671 | { | |
672 | enum dlm_status status = DLM_NORMAL; | |
673 | ||
674 | assert_spin_locked(&res->spinlock); | |
675 | ||
676 | if (res->state & DLM_LOCK_RES_RECOVERING) | |
677 | status = DLM_RECOVERING; | |
678 | else if (res->state & DLM_LOCK_RES_MIGRATING) | |
679 | status = DLM_MIGRATING; | |
680 | else if (res->state & DLM_LOCK_RES_IN_PROGRESS) | |
681 | status = DLM_FORWARD; | |
682 | ||
683 | return status; | |
684 | } | |
685 | ||
29004858 KH |
686 | static inline u8 dlm_get_lock_cookie_node(u64 cookie) |
687 | { | |
688 | u8 ret; | |
689 | cookie >>= 56; | |
690 | ret = (u8)(cookie & 0xffULL); | |
691 | return ret; | |
692 | } | |
693 | ||
694 | static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie) | |
695 | { | |
696 | unsigned long long ret; | |
697 | ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL; | |
698 | return ret; | |
699 | } | |
700 | ||
6714d8e8 KH |
701 | struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, |
702 | struct dlm_lockstatus *lksb); | |
703 | void dlm_lock_get(struct dlm_lock *lock); | |
704 | void dlm_lock_put(struct dlm_lock *lock); | |
705 | ||
706 | void dlm_lock_attach_lockres(struct dlm_lock *lock, | |
707 | struct dlm_lock_resource *res); | |
708 | ||
709 | int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data); | |
710 | int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data); | |
711 | int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data); | |
712 | ||
713 | void dlm_revert_pending_convert(struct dlm_lock_resource *res, | |
714 | struct dlm_lock *lock); | |
715 | void dlm_revert_pending_lock(struct dlm_lock_resource *res, | |
716 | struct dlm_lock *lock); | |
717 | ||
718 | int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data); | |
719 | void dlm_commit_pending_cancel(struct dlm_lock_resource *res, | |
720 | struct dlm_lock *lock); | |
721 | void dlm_commit_pending_unlock(struct dlm_lock_resource *res, | |
722 | struct dlm_lock *lock); | |
723 | ||
724 | int dlm_launch_thread(struct dlm_ctxt *dlm); | |
725 | void dlm_complete_thread(struct dlm_ctxt *dlm); | |
726 | int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); | |
727 | void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); | |
728 | void dlm_wait_for_recovery(struct dlm_ctxt *dlm); | |
c03872f5 | 729 | void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); |
e2faea4c | 730 | int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node); |
44465a7d | 731 | int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout); |
b7084ab5 | 732 | int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout); |
6714d8e8 KH |
733 | |
734 | void dlm_put(struct dlm_ctxt *dlm); | |
735 | struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm); | |
736 | int dlm_domain_fully_joined(struct dlm_ctxt *dlm); | |
737 | ||
738 | void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, | |
739 | struct dlm_lock_resource *res); | |
740 | void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, | |
741 | struct dlm_lock_resource *res); | |
ba2bf218 KH |
742 | int dlm_purge_lockres(struct dlm_ctxt *dlm, |
743 | struct dlm_lock_resource *lockres); | |
95c4f581 MF |
744 | static inline void dlm_lockres_get(struct dlm_lock_resource *res) |
745 | { | |
746 | /* This is called on every lookup, so it might be worth | |
747 | * inlining. */ | |
748 | kref_get(&res->refs); | |
749 | } | |
6714d8e8 KH |
750 | void dlm_lockres_put(struct dlm_lock_resource *res); |
751 | void __dlm_unhash_lockres(struct dlm_lock_resource *res); | |
752 | void __dlm_insert_lockres(struct dlm_ctxt *dlm, | |
753 | struct dlm_lock_resource *res); | |
ba2bf218 KH |
754 | struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, |
755 | const char *name, | |
756 | unsigned int len, | |
757 | unsigned int hash); | |
6714d8e8 KH |
758 | struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, |
759 | const char *name, | |
a3d33291 MF |
760 | unsigned int len, |
761 | unsigned int hash); | |
6714d8e8 KH |
762 | struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, |
763 | const char *name, | |
764 | unsigned int len); | |
765 | ||
766 | int dlm_is_host_down(int errno); | |
767 | void dlm_change_lockres_owner(struct dlm_ctxt *dlm, | |
768 | struct dlm_lock_resource *res, | |
769 | u8 owner); | |
770 | struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, | |
771 | const char *lockid, | |
3384f3df | 772 | int namelen, |
6714d8e8 KH |
773 | int flags); |
774 | struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, | |
775 | const char *name, | |
776 | unsigned int namelen); | |
777 | ||
ba2bf218 KH |
778 | #define dlm_lockres_set_refmap_bit(bit,res) \ |
779 | __dlm_lockres_set_refmap_bit(bit,res,__FILE__,__LINE__) | |
780 | #define dlm_lockres_clear_refmap_bit(bit,res) \ | |
781 | __dlm_lockres_clear_refmap_bit(bit,res,__FILE__,__LINE__) | |
782 | ||
783 | static inline void __dlm_lockres_set_refmap_bit(int bit, | |
784 | struct dlm_lock_resource *res, | |
785 | const char *file, | |
786 | int line) | |
787 | { | |
788 | //printk("%s:%d:%.*s: setting bit %d\n", file, line, | |
789 | // res->lockname.len, res->lockname.name, bit); | |
790 | set_bit(bit, res->refmap); | |
791 | } | |
792 | ||
793 | static inline void __dlm_lockres_clear_refmap_bit(int bit, | |
794 | struct dlm_lock_resource *res, | |
795 | const char *file, | |
796 | int line) | |
797 | { | |
798 | //printk("%s:%d:%.*s: clearing bit %d\n", file, line, | |
799 | // res->lockname.len, res->lockname.name, bit); | |
800 | clear_bit(bit, res->refmap); | |
801 | } | |
802 | ||
803 | void __dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, | |
804 | struct dlm_lock_resource *res, | |
805 | const char *file, | |
806 | int line); | |
807 | void __dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, | |
808 | struct dlm_lock_resource *res, | |
809 | int new_lockres, | |
810 | const char *file, | |
811 | int line); | |
812 | #define dlm_lockres_drop_inflight_ref(d,r) \ | |
813 | __dlm_lockres_drop_inflight_ref(d,r,__FILE__,__LINE__) | |
814 | #define dlm_lockres_grab_inflight_ref(d,r) \ | |
815 | __dlm_lockres_grab_inflight_ref(d,r,0,__FILE__,__LINE__) | |
816 | #define dlm_lockres_grab_inflight_ref_new(d,r) \ | |
817 | __dlm_lockres_grab_inflight_ref(d,r,1,__FILE__,__LINE__) | |
818 | ||
6714d8e8 KH |
819 | void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); |
820 | void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); | |
821 | void dlm_do_local_ast(struct dlm_ctxt *dlm, | |
822 | struct dlm_lock_resource *res, | |
823 | struct dlm_lock *lock); | |
824 | int dlm_do_remote_ast(struct dlm_ctxt *dlm, | |
825 | struct dlm_lock_resource *res, | |
826 | struct dlm_lock *lock); | |
827 | void dlm_do_local_bast(struct dlm_ctxt *dlm, | |
828 | struct dlm_lock_resource *res, | |
829 | struct dlm_lock *lock, | |
830 | int blocked_type); | |
831 | int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, | |
832 | struct dlm_lock_resource *res, | |
833 | struct dlm_lock *lock, | |
834 | int msg_type, | |
835 | int blocked_type, int flags); | |
836 | static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm, | |
837 | struct dlm_lock_resource *res, | |
838 | struct dlm_lock *lock, | |
839 | int blocked_type) | |
840 | { | |
841 | return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST, | |
842 | blocked_type, 0); | |
843 | } | |
844 | ||
845 | static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm, | |
846 | struct dlm_lock_resource *res, | |
847 | struct dlm_lock *lock, | |
848 | int flags) | |
849 | { | |
850 | return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST, | |
851 | 0, flags); | |
852 | } | |
853 | ||
854 | void dlm_print_one_lock_resource(struct dlm_lock_resource *res); | |
855 | void __dlm_print_one_lock_resource(struct dlm_lock_resource *res); | |
856 | ||
857 | u8 dlm_nm_this_node(struct dlm_ctxt *dlm); | |
858 | void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); | |
859 | void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); | |
860 | ||
861 | ||
862 | int dlm_nm_init(struct dlm_ctxt *dlm); | |
863 | int dlm_heartbeat_init(struct dlm_ctxt *dlm); | |
864 | void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data); | |
865 | void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data); | |
866 | ||
867 | int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); | |
868 | int dlm_migrate_lockres(struct dlm_ctxt *dlm, | |
869 | struct dlm_lock_resource *res, | |
870 | u8 target); | |
ba2bf218 | 871 | int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); |
6714d8e8 KH |
872 | int dlm_finish_migration(struct dlm_ctxt *dlm, |
873 | struct dlm_lock_resource *res, | |
874 | u8 old_master); | |
875 | void dlm_lockres_release_ast(struct dlm_ctxt *dlm, | |
876 | struct dlm_lock_resource *res); | |
877 | void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res); | |
878 | ||
879 | int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data); | |
880 | int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data); | |
ba2bf218 | 881 | int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data); |
6714d8e8 KH |
882 | int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data); |
883 | int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data); | |
884 | int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data); | |
885 | int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data); | |
886 | int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data); | |
887 | int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data); | |
888 | int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data); | |
c03872f5 KH |
889 | int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, |
890 | u8 nodenum, u8 *real_master); | |
c03872f5 | 891 | |
6714d8e8 KH |
892 | |
893 | int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, | |
894 | struct dlm_lock_resource *res, | |
895 | int ignore_higher, | |
896 | u8 request_from, | |
897 | u32 flags); | |
898 | ||
899 | ||
900 | int dlm_send_one_lockres(struct dlm_ctxt *dlm, | |
901 | struct dlm_lock_resource *res, | |
902 | struct dlm_migratable_lockres *mres, | |
903 | u8 send_to, | |
904 | u8 flags); | |
905 | void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm, | |
906 | struct dlm_lock_resource *res); | |
907 | ||
908 | /* will exit holding res->spinlock, but may drop in function */ | |
909 | void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags); | |
910 | void __dlm_wait_on_lockres_flags_set(struct dlm_lock_resource *res, int flags); | |
911 | ||
912 | /* will exit holding res->spinlock, but may drop in function */ | |
913 | static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res) | |
914 | { | |
915 | __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS| | |
916 | DLM_LOCK_RES_RECOVERING| | |
917 | DLM_LOCK_RES_MIGRATING)); | |
918 | } | |
919 | ||
920 | ||
921 | int dlm_init_mle_cache(void); | |
922 | void dlm_destroy_mle_cache(void); | |
923 | void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up); | |
ba2bf218 KH |
924 | int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, |
925 | struct dlm_lock_resource *res); | |
6714d8e8 KH |
926 | void dlm_clean_master_list(struct dlm_ctxt *dlm, |
927 | u8 dead_node); | |
928 | int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); | |
ba2bf218 | 929 | int __dlm_lockres_has_locks(struct dlm_lock_resource *res); |
69d72b06 | 930 | int __dlm_lockres_unused(struct dlm_lock_resource *res); |
6714d8e8 KH |
931 | |
932 | static inline const char * dlm_lock_mode_name(int mode) | |
933 | { | |
934 | switch (mode) { | |
935 | case LKM_EXMODE: | |
936 | return "EX"; | |
937 | case LKM_PRMODE: | |
938 | return "PR"; | |
939 | case LKM_NLMODE: | |
940 | return "NL"; | |
941 | } | |
942 | return "UNKNOWN"; | |
943 | } | |
944 | ||
945 | ||
946 | static inline int dlm_lock_compatible(int existing, int request) | |
947 | { | |
948 | /* NO_LOCK compatible with all */ | |
949 | if (request == LKM_NLMODE || | |
950 | existing == LKM_NLMODE) | |
951 | return 1; | |
952 | ||
953 | /* EX incompatible with all non-NO_LOCK */ | |
954 | if (request == LKM_EXMODE) | |
955 | return 0; | |
956 | ||
957 | /* request must be PR, which is compatible with PR */ | |
958 | if (existing == LKM_PRMODE) | |
959 | return 1; | |
960 | ||
961 | return 0; | |
962 | } | |
963 | ||
964 | static inline int dlm_lock_on_list(struct list_head *head, | |
965 | struct dlm_lock *lock) | |
966 | { | |
967 | struct list_head *iter; | |
968 | struct dlm_lock *tmplock; | |
969 | ||
970 | list_for_each(iter, head) { | |
971 | tmplock = list_entry(iter, struct dlm_lock, list); | |
972 | if (tmplock == lock) | |
973 | return 1; | |
974 | } | |
975 | return 0; | |
976 | } | |
977 | ||
978 | ||
979 | static inline enum dlm_status dlm_err_to_dlm_status(int err) | |
980 | { | |
981 | enum dlm_status ret; | |
982 | if (err == -ENOMEM) | |
983 | ret = DLM_SYSERR; | |
984 | else if (err == -ETIMEDOUT || o2net_link_down(err, NULL)) | |
985 | ret = DLM_NOLOCKMGR; | |
986 | else if (err == -EINVAL) | |
987 | ret = DLM_BADPARAM; | |
988 | else if (err == -ENAMETOOLONG) | |
989 | ret = DLM_IVBUFLEN; | |
990 | else | |
991 | ret = DLM_BADARGS; | |
992 | return ret; | |
993 | } | |
994 | ||
995 | ||
996 | static inline void dlm_node_iter_init(unsigned long *map, | |
997 | struct dlm_node_iter *iter) | |
998 | { | |
999 | memcpy(iter->node_map, map, sizeof(iter->node_map)); | |
1000 | iter->curnode = -1; | |
1001 | } | |
1002 | ||
1003 | static inline int dlm_node_iter_next(struct dlm_node_iter *iter) | |
1004 | { | |
1005 | int bit; | |
1006 | bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1); | |
1007 | if (bit >= O2NM_MAX_NODES) { | |
1008 | iter->curnode = O2NM_MAX_NODES; | |
1009 | return -ENOENT; | |
1010 | } | |
1011 | iter->curnode = bit; | |
1012 | return bit; | |
1013 | } | |
1014 | ||
1015 | ||
1016 | ||
1017 | #endif /* DLMCOMMON_H */ |