projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2/dlm: add DEREF_DONE message
[linux-block.git] / fs / ocfs2 / dlm / dlmcommon.h
CommitLineData
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
0467ae95 40#define DLM_HASH_SIZE_DEFAULT (1 << 17)
c8f33b6e
JB		 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
751155a9 52enum dlm_mle_type {
079ffb74
SM		 53 DLM_MLE_BLOCK = 0,
54 DLM_MLE_MASTER = 1,
55 DLM_MLE_MIGRATION = 2,
56 DLM_MLE_NUM_TYPES = 3,
751155a9
SM		 57};
58
751155a9 59struct dlm_master_list_entry {
2ed6c750 60 struct hlist_node master_hash_node;
751155a9
SM		 61 struct list_head hb_events;
62 struct dlm_ctxt *dlm;
63 spinlock_t spinlock;
64 wait_queue_head_t wq;
65 atomic_t woken;
66 struct kref mle_refs;
67 int inuse;
68 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
69 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
70 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
71 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
72 u8 master;
73 u8 new_master;
74 enum dlm_mle_type type;
75 struct o2hb_callback_func mle_hb_up;
76 struct o2hb_callback_func mle_hb_down;
7141514b
SM		 77 struct dlm_lock_resource *mleres;
78 unsigned char mname[DLM_LOCKID_NAME_MAX];
79 unsigned int mnamelen;
80 unsigned int mnamehash;
751155a9
SM		 81};
82
6714d8e8
KH		 83enum dlm_ast_type {
84 DLM_AST = 0,
079ffb74
SM		 85 DLM_BAST = 1,
86 DLM_ASTUNLOCK = 2,
6714d8e8
KH		 87};
88
89
90#define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
91 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
92 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)
93
94#define DLM_RECOVERY_LOCK_NAME "$RECOVERY"
95#define DLM_RECOVERY_LOCK_NAME_LEN 9
96
97static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
98{
99 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
100 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
101 return 1;
102 return 0;
103}
104
466d1a45
KH		 105#define DLM_RECO_STATE_ACTIVE 0x0001
106#define DLM_RECO_STATE_FINALIZE 0x0002
6714d8e8
KH		 107
108struct dlm_recovery_ctxt
109{
110 struct list_head resources;
6714d8e8
KH		 111 struct list_head node_data;
112 u8 new_master;
113 u8 dead_node;
114 u16 state;
115 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
116 wait_queue_head_t event;
117};
118
119enum dlm_ctxt_state {
120 DLM_CTXT_NEW = 0,
079ffb74
SM		 121 DLM_CTXT_JOINED = 1,
122 DLM_CTXT_IN_SHUTDOWN = 2,
123 DLM_CTXT_LEAVING = 3,
6714d8e8
KH		 124};
125
126struct dlm_ctxt
127{
128 struct list_head list;
03d864c0 129 struct hlist_head **lockres_hash;
6714d8e8
KH		 130 struct list_head dirty_list;
131 struct list_head purge_list;
132 struct list_head pending_asts;
133 struct list_head pending_basts;
29576f8b 134 struct list_head tracking_list;
6714d8e8
KH		 135 unsigned int purge_count;
136 spinlock_t spinlock;
137 spinlock_t ast_lock;
b0d4f817 138 spinlock_t track_lock;
6714d8e8
KH		 139 char *name;
140 u8 node_num;
141 u32 key;
142 u8 joining_node;
143 wait_queue_head_t dlm_join_events;
144 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
145 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
bddefdee 146 unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
6714d8e8
KH		 147 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
148 struct dlm_recovery_ctxt reco;
149 spinlock_t master_lock;
e2b66ddc 150 struct hlist_head **master_hash;
6714d8e8
KH		 151 struct list_head mle_hb_events;
152
153 /* these give a really vague idea of the system load */
2041d8fd
SM		 154 atomic_t mle_tot_count[DLM_MLE_NUM_TYPES];
155 atomic_t mle_cur_count[DLM_MLE_NUM_TYPES];
6800791a
SM		 156 atomic_t res_tot_count;
157 atomic_t res_cur_count;
6714d8e8 158
007dce53 159 struct dlm_debug_ctxt *dlm_debug_ctxt;
6325b4a2
SM		 160 struct dentry *dlm_debugfs_subroot;
161
6714d8e8
KH		 162 /* NOTE: Next three are protected by dlm_domain_lock */
163 struct kref dlm_refs;
164 enum dlm_ctxt_state dlm_state;
165 unsigned int num_joins;
166
167 struct o2hb_callback_func dlm_hb_up;
168 struct o2hb_callback_func dlm_hb_down;
169 struct task_struct *dlm_thread_task;
170 struct task_struct *dlm_reco_thread_task;
3156d267 171 struct workqueue_struct *dlm_worker;
6714d8e8
KH		 172 wait_queue_head_t dlm_thread_wq;
173 wait_queue_head_t dlm_reco_thread_wq;
174 wait_queue_head_t ast_wq;
175 wait_queue_head_t migration_wq;
176
177 struct work_struct dispatched_work;
178 struct list_head work_list;
179 spinlock_t work_lock;
180 struct list_head dlm_domain_handlers;
181 struct list_head dlm_eviction_callbacks;
d24fbcda
JB		 182
183 /* The filesystem specifies this at domain registration. We
184 * cache it here to know what to tell other nodes. */
185 struct dlm_protocol_version fs_locking_proto;
186 /* This is the inter-dlm communication version */
187 struct dlm_protocol_version dlm_locking_proto;
6714d8e8
KH		 188};
189
03d864c0
DP		 190static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
191{
192 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
193}
194
e2b66ddc
SM		 195static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm,
196 unsigned i)
197{
198 return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] +
199 (i % DLM_BUCKETS_PER_PAGE);
200}
201
6714d8e8
KH		 202/* these keventd work queue items are for less-frequently
203 * called functions that cannot be directly called from the
204 * net message handlers for some reason, usually because
205 * they need to send net messages of their own. */
c4028958 206void dlm_dispatch_work(struct work_struct *work);
6714d8e8
KH		 207
208struct dlm_lock_resource;
209struct dlm_work_item;
210
211typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);
212
213struct dlm_request_all_locks_priv
214{
215 u8 reco_master;
216 u8 dead_node;
217};
218
219struct dlm_mig_lockres_priv
220{
221 struct dlm_lock_resource *lockres;
222 u8 real_master;
52987e2a 223 u8 extra_ref;
6714d8e8
KH		 224};
225
226struct dlm_assert_master_priv
227{
228 struct dlm_lock_resource *lockres;
229 u8 request_from;
230 u32 flags;
231 unsigned ignore_higher:1;
232};
233
f3f85464
SM		 234struct dlm_deref_lockres_priv
235{
236 struct dlm_lock_resource *deref_res;
237 u8 deref_node;
238};
6714d8e8
KH		 239
240struct dlm_work_item
241{
242 struct list_head list;
243 dlm_workfunc_t *func;
244 struct dlm_ctxt *dlm;
245 void *data;
246 union {
247 struct dlm_request_all_locks_priv ral;
248 struct dlm_mig_lockres_priv ml;
249 struct dlm_assert_master_priv am;
f3f85464 250 struct dlm_deref_lockres_priv dl;
6714d8e8
KH		 251 } u;
252};
253
254static inline void dlm_init_work_item(struct dlm_ctxt *dlm,
255 struct dlm_work_item *i,
256 dlm_workfunc_t *f, void *data)
257{
258 memset(i, 0, sizeof(*i));
259 i->func = f;
260 INIT_LIST_HEAD(&i->list);
261 i->data = data;
262 i->dlm = dlm; /* must have already done a dlm_grab on this! */
263}
264
265
266
267static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm,
268 u8 node)
269{
270 assert_spin_locked(&dlm->spinlock);
271
272 dlm->joining_node = node;
273 wake_up(&dlm->dlm_join_events);
274}
275
276#define DLM_LOCK_RES_UNINITED 0x00000001
277#define DLM_LOCK_RES_RECOVERING 0x00000002
278#define DLM_LOCK_RES_READY 0x00000004
279#define DLM_LOCK_RES_DIRTY 0x00000008
280#define DLM_LOCK_RES_IN_PROGRESS 0x00000010
281#define DLM_LOCK_RES_MIGRATING 0x00000020
ba2bf218 282#define DLM_LOCK_RES_DROPPING_REF 0x00000040
ddc09c8d 283#define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000
3b8118cf 284#define DLM_LOCK_RES_SETREF_INPROG 0x00002000
6714d8e8 285
44465a7d
KH		 286/* max milliseconds to wait to sync up a network failure with a node death */
287#define DLM_NODE_DEATH_WAIT_MAX (5 * 1000)
288
6714d8e8
KH		 289#define DLM_PURGE_INTERVAL_MS (8 * 1000)
290
291struct dlm_lock_resource
292{
293 /* WARNING: Please see the comment in dlm_init_lockres before
294 * adding fields here. */
81f2094a 295 struct hlist_node hash_node;
65c491d8 296 struct qstr lockname;
6714d8e8
KH		 297 struct kref refs;
298
6ff06a93
KH		 299 /*
300 * Please keep granted, converting, and blocked in this order,
301 * as some funcs want to iterate over all lists.
302 *
303 * All four lists are protected by the hash's reference.
304 */
6714d8e8
KH		 305 struct list_head granted;
306 struct list_head converting;
307 struct list_head blocked;
6ff06a93 308 struct list_head purge;
6714d8e8 309
6ff06a93
KH		 310 /*
311 * These two lists require you to hold an additional reference
312 * while they are on the list.
313 */
6714d8e8
KH		 314 struct list_head dirty;
315 struct list_head recovering; // dlm_recovery_ctxt.resources list
316
29576f8b
SM		 317 /* Added during init and removed during release */
318 struct list_head tracking; /* dlm->tracking_list */
319
6714d8e8
KH		 320 /* unused lock resources have their last_used stamped and are
321 * put on a list for the dlm thread to run. */
6714d8e8
KH		 322 unsigned long last_used;
323
b0d4f817
SM		 324 struct dlm_ctxt *dlm;
325
6714d8e8
KH		 326 unsigned migration_pending:1;
327 atomic_t asts_reserved;
328 spinlock_t spinlock;
329 wait_queue_head_t wq;
330 u8 owner; //node which owns the lock resource, or unknown
331 u16 state;
6714d8e8 332 char lvb[DLM_LVB_LEN];
ba2bf218 333 unsigned int inflight_locks;
ac4fef4d 334 unsigned int inflight_assert_workers;
ba2bf218 335 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
6714d8e8
KH		 336};
337
338struct dlm_migratable_lock
339{
340 __be64 cookie;
341
342 /* these 3 are just padding for the in-memory structure, but
343 * list and flags are actually used when sent over the wire */
344 __be16 pad1;
345 u8 list; // 0=granted, 1=converting, 2=blocked
346 u8 flags;
347
348 s8 type;
349 s8 convert_type;
350 s8 highest_blocked;
351 u8 node;
352}; // 16 bytes
353
354struct dlm_lock
355{
356 struct dlm_migratable_lock ml;
357
358 struct list_head list;
359 struct list_head ast_list;
360 struct list_head bast_list;
361 struct dlm_lock_resource *lockres;
362 spinlock_t spinlock;
363 struct kref lock_refs;
364
365 // ast and bast must be callable while holding a spinlock!
366 dlm_astlockfunc_t *ast;
367 dlm_bastlockfunc_t *bast;
368 void *astdata;
369 struct dlm_lockstatus *lksb;
370 unsigned ast_pending:1,
371 bast_pending:1,
372 convert_pending:1,
373 lock_pending:1,
374 cancel_pending:1,
375 unlock_pending:1,
376 lksb_kernel_allocated:1;
377};
378
6714d8e8
KH		 379enum dlm_lockres_list {
380 DLM_GRANTED_LIST = 0,
079ffb74
SM		 381 DLM_CONVERTING_LIST = 1,
382 DLM_BLOCKED_LIST = 2,
6714d8e8
KH		 383};
384
8bc674cb
KH		 385static inline int dlm_lvb_is_empty(char *lvb)
386{
387 int i;
388 for (i=0; i<DLM_LVB_LEN; i++)
389 if (lvb[i])
390 return 0;
391 return 1;
392}
393
9f62e960
SM		 394static inline char *dlm_list_in_text(enum dlm_lockres_list idx)
395{
396 if (idx == DLM_GRANTED_LIST)
397 return "granted";
398 else if (idx == DLM_CONVERTING_LIST)
399 return "converting";
400 else if (idx == DLM_BLOCKED_LIST)
401 return "blocked";
402 else
403 return "unknown";
404}
405
6714d8e8
KH		 406static inline struct list_head *
407dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
408{
409 struct list_head *ret = NULL;
410 if (idx == DLM_GRANTED_LIST)
411 ret = &res->granted;
412 else if (idx == DLM_CONVERTING_LIST)
413 ret = &res->converting;
414 else if (idx == DLM_BLOCKED_LIST)
415 ret = &res->blocked;
416 else
417 BUG();
418 return ret;
419}
420
421
422
423
424struct dlm_node_iter
425{
426 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
427 int curnode;
428};
429
430
431enum {
079ffb74
SM		 432 DLM_MASTER_REQUEST_MSG = 500,
433 DLM_UNUSED_MSG1 = 501,
434 DLM_ASSERT_MASTER_MSG = 502,
435 DLM_CREATE_LOCK_MSG = 503,
436 DLM_CONVERT_LOCK_MSG = 504,
437 DLM_PROXY_AST_MSG = 505,
438 DLM_UNLOCK_LOCK_MSG = 506,
439 DLM_DEREF_LOCKRES_MSG = 507,
440 DLM_MIGRATE_REQUEST_MSG = 508,
441 DLM_MIG_LOCKRES_MSG = 509,
442 DLM_QUERY_JOIN_MSG = 510,
443 DLM_ASSERT_JOINED_MSG = 511,
444 DLM_CANCEL_JOIN_MSG = 512,
445 DLM_EXIT_DOMAIN_MSG = 513,
446 DLM_MASTER_REQUERY_MSG = 514,
447 DLM_LOCK_REQUEST_MSG = 515,
448 DLM_RECO_DATA_DONE_MSG = 516,
449 DLM_BEGIN_RECO_MSG = 517,
450 DLM_FINALIZE_RECO_MSG = 518,
451 DLM_QUERY_REGION = 519,
452 DLM_QUERY_NODEINFO = 520,
bddefdee 453 DLM_BEGIN_EXIT_DOMAIN_MSG = 521,
60d663cb 454 DLM_DEREF_LOCKRES_DONE = 522,
6714d8e8
KH		 455};
456
457struct dlm_reco_node_data
458{
459 int state;
460 u8 node_num;
461 struct list_head list;
462};
463
464enum {
465 DLM_RECO_NODE_DATA_DEAD = -1,
466 DLM_RECO_NODE_DATA_INIT = 0,
079ffb74
SM		 467 DLM_RECO_NODE_DATA_REQUESTING = 1,
468 DLM_RECO_NODE_DATA_REQUESTED = 2,
469 DLM_RECO_NODE_DATA_RECEIVING = 3,
470 DLM_RECO_NODE_DATA_DONE = 4,
471 DLM_RECO_NODE_DATA_FINALIZE_SENT = 5,
6714d8e8
KH		 472};
473
474
475enum {
476 DLM_MASTER_RESP_NO = 0,
079ffb74
SM		 477 DLM_MASTER_RESP_YES = 1,
478 DLM_MASTER_RESP_MAYBE = 2,
479 DLM_MASTER_RESP_ERROR = 3,
6714d8e8
KH		 480};
481
482
483struct dlm_master_request
484{
485 u8 node_idx;
486 u8 namelen;
487 __be16 pad1;
488 __be32 flags;
489
490 u8 name[O2NM_MAX_NAME_LEN];
491};
492
ba2bf218
KH		 493#define DLM_ASSERT_RESPONSE_REASSERT 0x00000001
494#define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002
495
6714d8e8
KH		 496#define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001
497#define DLM_ASSERT_MASTER_REQUERY 0x00000002
498#define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
499struct dlm_assert_master
500{
501 u8 node_idx;
502 u8 namelen;
503 __be16 pad1;
504 __be32 flags;
505
506 u8 name[O2NM_MAX_NAME_LEN];
507};
508
ba2bf218
KH		 509#define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001
510
6714d8e8
KH		 511struct dlm_migrate_request
512{
513 u8 master;
514 u8 new_master;
515 u8 namelen;
516 u8 pad1;
517 __be32 pad2;
518 u8 name[O2NM_MAX_NAME_LEN];
519};
520
521struct dlm_master_requery
522{
523 u8 pad1;
524 u8 pad2;
525 u8 node_idx;
526 u8 namelen;
527 __be32 pad3;
528 u8 name[O2NM_MAX_NAME_LEN];
529};
530
531#define DLM_MRES_RECOVERY 0x01
532#define DLM_MRES_MIGRATION 0x02
533#define DLM_MRES_ALL_DONE 0x04
534
535/*
536 * We would like to get one whole lockres into a single network
537 * message whenever possible. Generally speaking, there will be
538 * at most one dlm_lock on a lockres for each node in the cluster,
539 * plus (infrequently) any additional locks coming in from userdlm.
540 *
541 * struct _dlm_lockres_page
542 * {
543 * dlm_migratable_lockres mres;
544 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
545 * u8 pad[DLM_MIG_LOCKRES_RESERVED];
546 * };
547 *
548 * from ../cluster/tcp.h
39b29af0 549 * O2NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg))
6714d8e8
KH		 550 * (roughly 4080 bytes)
551 * and sizeof(dlm_migratable_lockres) = 112 bytes
552 * and sizeof(dlm_migratable_lock) = 16 bytes
553 *
554 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
555 * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
556 *
557 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) +
558 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED =
559 * NET_MAX_PAYLOAD_BYTES
560 * (240 * 16) + 112 + 128 = 4080
561 *
562 * So a lockres would need more than 240 locks before it would
563 * use more than one network packet to recover. Not too bad.
564 */
565#define DLM_MAX_MIGRATABLE_LOCKS 240
566
567struct dlm_migratable_lockres
568{
569 u8 master;
570 u8 lockname_len;
571 u8 num_locks; // locks sent in this structure
572 u8 flags;
573 __be32 total_locks; // locks to be sent for this migration cookie
574 __be64 mig_cookie; // cookie for this lockres migration
575 // or zero if not needed
576 // 16 bytes
577 u8 lockname[DLM_LOCKID_NAME_MAX];
578 // 48 bytes
579 u8 lvb[DLM_LVB_LEN];
580 // 112 bytes
581 struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112
582};
583#define DLM_MIG_LOCKRES_MAX_LEN \
584 (sizeof(struct dlm_migratable_lockres) + \
585 (sizeof(struct dlm_migratable_lock) * \
586 DLM_MAX_MIGRATABLE_LOCKS) )
587
588/* from above, 128 bytes
589 * for some undetermined future use */
39b29af0 590#define DLM_MIG_LOCKRES_RESERVED (O2NET_MAX_PAYLOAD_BYTES - \
6714d8e8
KH		 591 DLM_MIG_LOCKRES_MAX_LEN)
592
593struct dlm_create_lock
594{
595 __be64 cookie;
596
597 __be32 flags;
598 u8 pad1;
599 u8 node_idx;
600 s8 requested_type;
601 u8 namelen;
602
603 u8 name[O2NM_MAX_NAME_LEN];
604};
605
606struct dlm_convert_lock
607{
608 __be64 cookie;
609
610 __be32 flags;
611 u8 pad1;
612 u8 node_idx;
613 s8 requested_type;
614 u8 namelen;
615
616 u8 name[O2NM_MAX_NAME_LEN];
617
618 s8 lvb[0];
619};
620#define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN)
621
622struct dlm_unlock_lock
623{
624 __be64 cookie;
625
626 __be32 flags;
627 __be16 pad1;
628 u8 node_idx;
629 u8 namelen;
630
631 u8 name[O2NM_MAX_NAME_LEN];
632
633 s8 lvb[0];
634};
635#define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN)
636
637struct dlm_proxy_ast
638{
639 __be64 cookie;
640
641 __be32 flags;
642 u8 node_idx;
643 u8 type;
644 u8 blocked_type;
645 u8 namelen;
646
647 u8 name[O2NM_MAX_NAME_LEN];
648
649 s8 lvb[0];
650};
651#define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN)
652
653#define DLM_MOD_KEY (0x666c6172)
d24fbcda 654enum dlm_query_join_response_code {
6714d8e8 655 JOIN_DISALLOW = 0,
079ffb74
SM		 656 JOIN_OK = 1,
657 JOIN_OK_NO_MAP = 2,
658 JOIN_PROTOCOL_MISMATCH = 3,
d24fbcda
JB		 659};
660
0f71b7b4
JB		 661struct dlm_query_join_packet {
662 u8 code; /* Response code. dlm_minor and fs_minor
663 are only valid if this is JOIN_OK */
664 u8 dlm_minor; /* The minor version of the protocol the
665 dlm is speaking. */
666 u8 fs_minor; /* The minor version of the protocol the
667 filesystem is speaking. */
668 u8 reserved;
669};
670
d24fbcda 671union dlm_query_join_response {
85158410 672 __be32 intval;
0f71b7b4 673 struct dlm_query_join_packet packet;
6714d8e8
KH		 674};
675
676struct dlm_lock_request
677{
678 u8 node_idx;
679 u8 dead_node;
680 __be16 pad1;
681 __be32 pad2;
682};
683
684struct dlm_reco_data_done
685{
686 u8 node_idx;
687 u8 dead_node;
688 __be16 pad1;
689 __be32 pad2;
690
691 /* unused for now */
692 /* eventually we can use this to attempt
693 * lvb recovery based on each node's info */
694 u8 reco_lvb[DLM_LVB_LEN];
695};
696
697struct dlm_begin_reco
698{
699 u8 node_idx;
700 u8 dead_node;
701 __be16 pad1;
702 __be32 pad2;
703};
704
705
1faf2894
SE		 706#define BITS_PER_BYTE 8
707#define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE)
708
6714d8e8
KH		 709struct dlm_query_join_request
710{
711 u8 node_idx;
712 u8 pad1[2];
713 u8 name_len;
d24fbcda
JB		 714 struct dlm_protocol_version dlm_proto;
715 struct dlm_protocol_version fs_proto;
6714d8e8 716 u8 domain[O2NM_MAX_NAME_LEN];
1faf2894 717 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)];
6714d8e8
KH		 718};
719
720struct dlm_assert_joined
721{
722 u8 node_idx;
723 u8 pad1[2];
724 u8 name_len;
725 u8 domain[O2NM_MAX_NAME_LEN];
726};
727
728struct dlm_cancel_join
729{
730 u8 node_idx;
731 u8 pad1[2];
732 u8 name_len;
733 u8 domain[O2NM_MAX_NAME_LEN];
734};
735
ea203441
SM		 736struct dlm_query_region {
737 u8 qr_node;
738 u8 qr_numregions;
739 u8 qr_namelen;
740 u8 pad1;
741 u8 qr_domain[O2NM_MAX_NAME_LEN];
742 u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS];
743};
744
18cfdf1b
SM		 745struct dlm_node_info {
746 u8 ni_nodenum;
747 u8 pad1;
85158410
AV		 748 __be16 ni_ipv4_port;
749 __be32 ni_ipv4_address;
18cfdf1b
SM		 750};
751
752struct dlm_query_nodeinfo {
753 u8 qn_nodenum;
754 u8 qn_numnodes;
755 u8 qn_namelen;
756 u8 pad1;
757 u8 qn_domain[O2NM_MAX_NAME_LEN];
758 struct dlm_node_info qn_nodes[O2NM_MAX_NODES];
759};
760
6714d8e8
KH		 761struct dlm_exit_domain
762{
763 u8 node_idx;
764 u8 pad1[3];
765};
766
767struct dlm_finalize_reco
768{
769 u8 node_idx;
770 u8 dead_node;
466d1a45
KH		 771 u8 flags;
772 u8 pad1;
6714d8e8
KH		 773 __be32 pad2;
774};
775
ba2bf218
KH		 776struct dlm_deref_lockres
777{
778 u32 pad1;
779 u16 pad2;
780 u8 node_idx;
781 u8 namelen;
782
783 u8 name[O2NM_MAX_NAME_LEN];
784};
785
60d663cb 786struct dlm_deref_lockres_done {
787 u32 pad1;
788 u16 pad2;
789 u8 node_idx;
790 u8 namelen;
791
792 u8 name[O2NM_MAX_NAME_LEN];
793};
794
6714d8e8
KH		 795static inline enum dlm_status
796__dlm_lockres_state_to_status(struct dlm_lock_resource *res)
797{
798 enum dlm_status status = DLM_NORMAL;
799
800 assert_spin_locked(&res->spinlock);
801
802 if (res->state & DLM_LOCK_RES_RECOVERING)
803 status = DLM_RECOVERING;
804 else if (res->state & DLM_LOCK_RES_MIGRATING)
805 status = DLM_MIGRATING;
806 else if (res->state & DLM_LOCK_RES_IN_PROGRESS)
807 status = DLM_FORWARD;
808
809 return status;
810}
811
29004858
KH		 812static inline u8 dlm_get_lock_cookie_node(u64 cookie)
813{
814 u8 ret;
815 cookie >>= 56;
816 ret = (u8)(cookie & 0xffULL);
817 return ret;
818}
819
820static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
821{
822 unsigned long long ret;
823 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
824 return ret;
825}
826
6714d8e8
KH		 827struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
828 struct dlm_lockstatus *lksb);
829void dlm_lock_get(struct dlm_lock *lock);
830void dlm_lock_put(struct dlm_lock *lock);
831
832void dlm_lock_attach_lockres(struct dlm_lock *lock,
833 struct dlm_lock_resource *res);
834
d74c9803
KH		 835int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
836 void **ret_data);
837int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
838 void **ret_data);
839int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
840 void **ret_data);
6714d8e8
KH		 841
842void dlm_revert_pending_convert(struct dlm_lock_resource *res,
843 struct dlm_lock *lock);
844void dlm_revert_pending_lock(struct dlm_lock_resource *res,
845 struct dlm_lock *lock);
846
d74c9803
KH		 847int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
848 void **ret_data);
6714d8e8
KH		 849void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
850 struct dlm_lock *lock);
851void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
852 struct dlm_lock *lock);
853
854int dlm_launch_thread(struct dlm_ctxt *dlm);
855void dlm_complete_thread(struct dlm_ctxt *dlm);
856int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
857void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
858void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
c03872f5 859void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
e2faea4c 860int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
ed8625c6
SM		 861void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
862void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
6714d8e8
KH		 863
864void dlm_put(struct dlm_ctxt *dlm);
865struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
866int dlm_domain_fully_joined(struct dlm_ctxt *dlm);
867
868void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
869 struct dlm_lock_resource *res);
870void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
871 struct dlm_lock_resource *res);
95c4f581
MF		 872static inline void dlm_lockres_get(struct dlm_lock_resource *res)
873{
874 /* This is called on every lookup, so it might be worth
875 * inlining. */
876 kref_get(&res->refs);
877}
6714d8e8 878void dlm_lockres_put(struct dlm_lock_resource *res);
e9f0b6a6
SM		 879void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
880void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
ba2bf218
KH		 881struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
882 const char *name,
883 unsigned int len,
884 unsigned int hash);
6714d8e8
KH		 885struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
886 const char *name,
a3d33291
MF		 887 unsigned int len,
888 unsigned int hash);
6714d8e8
KH		 889struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
890 const char *name,
891 unsigned int len);
892
893int dlm_is_host_down(int errno);
7d62a978 894
6714d8e8
KH		 895struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
896 const char *lockid,
3384f3df 897 int namelen,
6714d8e8
KH		 898 int flags);
899struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
900 const char *name,
901 unsigned int namelen);
902
8d400b81
SM		 903void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
904 struct dlm_lock_resource *res, int bit);
905void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
906 struct dlm_lock_resource *res, int bit);
907
908void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
909 struct dlm_lock_resource *res);
910void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
911 struct dlm_lock_resource *res);
ba2bf218 912
ac4fef4d
X		 913void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm,
914 struct dlm_lock_resource *res);
915
6714d8e8
KH		 916void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
917void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
d9ef7522
WW		 918void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
919void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
6714d8e8
KH		 920void dlm_do_local_ast(struct dlm_ctxt *dlm,
921 struct dlm_lock_resource *res,
922 struct dlm_lock *lock);
923int dlm_do_remote_ast(struct dlm_ctxt *dlm,
924 struct dlm_lock_resource *res,
925 struct dlm_lock *lock);
926void dlm_do_local_bast(struct dlm_ctxt *dlm,
927 struct dlm_lock_resource *res,
928 struct dlm_lock *lock,
929 int blocked_type);
930int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm,
931 struct dlm_lock_resource *res,
932 struct dlm_lock *lock,
933 int msg_type,
934 int blocked_type, int flags);
935static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm,
936 struct dlm_lock_resource *res,
937 struct dlm_lock *lock,
938 int blocked_type)
939{
940 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST,
941 blocked_type, 0);
942}
943
944static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm,
945 struct dlm_lock_resource *res,
946 struct dlm_lock *lock,
947 int flags)
948{
949 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST,
950 0, flags);
951}
952
953void dlm_print_one_lock_resource(struct dlm_lock_resource *res);
954void __dlm_print_one_lock_resource(struct dlm_lock_resource *res);
955
956u8 dlm_nm_this_node(struct dlm_ctxt *dlm);
957void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
958void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
959
960
961int dlm_nm_init(struct dlm_ctxt *dlm);
962int dlm_heartbeat_init(struct dlm_ctxt *dlm);
963void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data);
964void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data);
965
ba2bf218 966int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
6714d8e8
KH		 967int dlm_finish_migration(struct dlm_ctxt *dlm,
968 struct dlm_lock_resource *res,
969 u8 old_master);
970void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
971 struct dlm_lock_resource *res);
972void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res);
973
d74c9803
KH		 974int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
975 void **ret_data);
976int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
977 void **ret_data);
3b8118cf 978void dlm_assert_master_post_handler(int status, void *data, void *ret_data);
d74c9803
KH		 979int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
980 void **ret_data);
60d663cb 981int dlm_deref_lockres_done_handler(struct o2net_msg *msg, u32 len, void *data,
982 void **ret_data);
d74c9803
KH		 983int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
984 void **ret_data);
985int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
986 void **ret_data);
987int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
988 void **ret_data);
989int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
990 void **ret_data);
991int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
992 void **ret_data);
993int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
994 void **ret_data);
995int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
996 void **ret_data);
c03872f5
KH		 997int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
998 u8 nodenum, u8 *real_master);
c03872f5 999
6714d8e8
KH		 1000
1001int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
1002 struct dlm_lock_resource *res,
1003 int ignore_higher,
1004 u8 request_from,
1005 u32 flags);
1006
1007
1008int dlm_send_one_lockres(struct dlm_ctxt *dlm,
1009 struct dlm_lock_resource *res,
1010 struct dlm_migratable_lockres *mres,
1011 u8 send_to,
1012 u8 flags);
1013void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
1014 struct dlm_lock_resource *res);
1015
1016/* will exit holding res->spinlock, but may drop in function */
1017void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags);
6714d8e8
KH		 1018
1019/* will exit holding res->spinlock, but may drop in function */
1020static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res)
1021{
1022 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS|
1023 DLM_LOCK_RES_RECOVERING|
1024 DLM_LOCK_RES_MIGRATING));
1025}
1026
1c084577
SM		 1027void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1028void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1029
724bdca9
SM		 1030/* create/destroy slab caches */
1031int dlm_init_master_caches(void);
1032void dlm_destroy_master_caches(void);
1033
1034int dlm_init_lock_cache(void);
1035void dlm_destroy_lock_cache(void);
6714d8e8
KH		 1036
1037int dlm_init_mle_cache(void);
1038void dlm_destroy_mle_cache(void);
724bdca9 1039
6714d8e8 1040void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up);
ba2bf218
KH		 1041int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
1042 struct dlm_lock_resource *res);
6714d8e8
KH		 1043void dlm_clean_master_list(struct dlm_ctxt *dlm,
1044 u8 dead_node);
5dad6c39 1045void dlm_force_free_mles(struct dlm_ctxt *dlm);
6714d8e8 1046int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
ba2bf218 1047int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
69d72b06 1048int __dlm_lockres_unused(struct dlm_lock_resource *res);
6714d8e8
KH		 1049
1050static inline const char * dlm_lock_mode_name(int mode)
1051{
1052 switch (mode) {
1053 case LKM_EXMODE:
1054 return "EX";
1055 case LKM_PRMODE:
1056 return "PR";
1057 case LKM_NLMODE:
1058 return "NL";
1059 }
1060 return "UNKNOWN";
1061}
1062
1063
1064static inline int dlm_lock_compatible(int existing, int request)
1065{
1066 /* NO_LOCK compatible with all */
1067 if (request == LKM_NLMODE ||
1068 existing == LKM_NLMODE)
1069 return 1;
1070
1071 /* EX incompatible with all non-NO_LOCK */
1072 if (request == LKM_EXMODE)
1073 return 0;
1074
1075 /* request must be PR, which is compatible with PR */
1076 if (existing == LKM_PRMODE)
1077 return 1;
1078
1079 return 0;
1080}
1081
1082static inline int dlm_lock_on_list(struct list_head *head,
1083 struct dlm_lock *lock)
1084{
6714d8e8
KH		 1085 struct dlm_lock *tmplock;
1086
df53cd3b 1087 list_for_each_entry(tmplock, head, list) {
6714d8e8
KH		 1088 if (tmplock == lock)
1089 return 1;
1090 }
1091 return 0;
1092}
1093
1094
1095static inline enum dlm_status dlm_err_to_dlm_status(int err)
1096{
1097 enum dlm_status ret;
1098 if (err == -ENOMEM)
1099 ret = DLM_SYSERR;
1100 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL))
1101 ret = DLM_NOLOCKMGR;
1102 else if (err == -EINVAL)
1103 ret = DLM_BADPARAM;
1104 else if (err == -ENAMETOOLONG)
1105 ret = DLM_IVBUFLEN;
1106 else
1107 ret = DLM_BADARGS;
1108 return ret;
1109}
1110
1111
1112static inline void dlm_node_iter_init(unsigned long *map,
1113 struct dlm_node_iter *iter)
1114{
1115 memcpy(iter->node_map, map, sizeof(iter->node_map));
1116 iter->curnode = -1;
1117}
1118
1119static inline int dlm_node_iter_next(struct dlm_node_iter *iter)
1120{
1121 int bit;
1122 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1);
1123 if (bit >= O2NM_MAX_NODES) {
1124 iter->curnode = O2NM_MAX_NODES;
1125 return -ENOENT;
1126 }
1127 iter->curnode = bit;
1128 return bit;
1129}
1130
7d62a978
SM		 1131static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
1132 struct dlm_lock_resource *res,
1133 u8 owner)
1134{
1135 assert_spin_locked(&res->spinlock);
1136
1137 res->owner = owner;
1138}
6714d8e8 1139
7d62a978
SM		 1140static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
1141 struct dlm_lock_resource *res,
1142 u8 owner)
1143{
1144 assert_spin_locked(&res->spinlock);
1145
1146 if (owner != res->owner)
1147 dlm_set_lockres_owner(dlm, res, owner);
1148}
6714d8e8
KH		 1149
1150#endif /* DLMCOMMON_H */
Linux 6.x block layer and io_uring tree(s)