xref: /linux/fs/ocfs2/dlm/dlmcommon.h (revision c0c914eca7f251c70facc37dfebeaf176601918d)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmcommon.h
5  *
6  * Copyright (C) 2004 Oracle.  All rights reserved.
7  *
8  * This program is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU General Public
10  * License as published by the Free Software Foundation; either
11  * version 2 of the License, or (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public
19  * License along with this program; if not, write to the
20  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
21  * Boston, MA 021110-1307, USA.
22  *
23  */
24 
25 #ifndef DLMCOMMON_H
26 #define DLMCOMMON_H
27 
28 #include <linux/kref.h>
29 
30 #define DLM_HB_NODE_DOWN_PRI     (0xf000000)
31 #define DLM_HB_NODE_UP_PRI       (0x8000000)
32 
33 #define DLM_LOCKID_NAME_MAX    32
34 
35 #define DLM_DOMAIN_NAME_MAX_LEN    255
36 #define DLM_LOCK_RES_OWNER_UNKNOWN     O2NM_MAX_NODES
37 #define DLM_THREAD_SHUFFLE_INTERVAL    5     // flush everything every 5 passes
38 #define DLM_THREAD_MS                  200   // flush at least every 200 ms
39 
40 #define DLM_HASH_SIZE_DEFAULT	(1 << 17)
41 #if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE
42 # define DLM_HASH_PAGES		1
43 #else
44 # define DLM_HASH_PAGES		(DLM_HASH_SIZE_DEFAULT / PAGE_SIZE)
45 #endif
46 #define DLM_BUCKETS_PER_PAGE	(PAGE_SIZE / sizeof(struct hlist_head))
47 #define DLM_HASH_BUCKETS	(DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
48 
49 /* Intended to make it easier for us to switch out hash functions */
50 #define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l)
51 
52 enum dlm_mle_type {
53 	DLM_MLE_BLOCK = 0,
54 	DLM_MLE_MASTER = 1,
55 	DLM_MLE_MIGRATION = 2,
56 	DLM_MLE_NUM_TYPES = 3,
57 };
58 
59 struct dlm_master_list_entry {
60 	struct hlist_node master_hash_node;
61 	struct list_head hb_events;
62 	struct dlm_ctxt *dlm;
63 	spinlock_t spinlock;
64 	wait_queue_head_t wq;
65 	atomic_t woken;
66 	struct kref mle_refs;
67 	int inuse;
68 	unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
69 	unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
70 	unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
71 	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
72 	u8 master;
73 	u8 new_master;
74 	enum dlm_mle_type type;
75 	struct o2hb_callback_func mle_hb_up;
76 	struct o2hb_callback_func mle_hb_down;
77 	struct dlm_lock_resource *mleres;
78 	unsigned char mname[DLM_LOCKID_NAME_MAX];
79 	unsigned int mnamelen;
80 	unsigned int mnamehash;
81 };
82 
83 enum dlm_ast_type {
84 	DLM_AST = 0,
85 	DLM_BAST = 1,
86 	DLM_ASTUNLOCK = 2,
87 };
88 
89 
90 #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
91 			 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
92 			 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)
93 
94 #define DLM_RECOVERY_LOCK_NAME       "$RECOVERY"
95 #define DLM_RECOVERY_LOCK_NAME_LEN   9
96 
97 static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
98 {
99 	if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
100 	    memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
101 		return 1;
102 	return 0;
103 }
104 
105 #define DLM_RECO_STATE_ACTIVE    0x0001
106 #define DLM_RECO_STATE_FINALIZE  0x0002
107 
108 struct dlm_recovery_ctxt
109 {
110 	struct list_head resources;
111 	struct list_head node_data;
112 	u8  new_master;
113 	u8  dead_node;
114 	u16 state;
115 	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
116 	wait_queue_head_t event;
117 };
118 
119 enum dlm_ctxt_state {
120 	DLM_CTXT_NEW = 0,
121 	DLM_CTXT_JOINED = 1,
122 	DLM_CTXT_IN_SHUTDOWN = 2,
123 	DLM_CTXT_LEAVING = 3,
124 };
125 
126 struct dlm_ctxt
127 {
128 	struct list_head list;
129 	struct hlist_head **lockres_hash;
130 	struct list_head dirty_list;
131 	struct list_head purge_list;
132 	struct list_head pending_asts;
133 	struct list_head pending_basts;
134 	struct list_head tracking_list;
135 	unsigned int purge_count;
136 	spinlock_t spinlock;
137 	spinlock_t ast_lock;
138 	spinlock_t track_lock;
139 	char *name;
140 	u8 node_num;
141 	u32 key;
142 	u8  joining_node;
143 	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)];
146 	unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
147 	unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
148 	struct dlm_recovery_ctxt reco;
149 	spinlock_t master_lock;
150 	struct hlist_head **master_hash;
151 	struct list_head mle_hb_events;
152 
153 	/* these give a really vague idea of the system load */
154 	atomic_t mle_tot_count[DLM_MLE_NUM_TYPES];
155 	atomic_t mle_cur_count[DLM_MLE_NUM_TYPES];
156 	atomic_t res_tot_count;
157 	atomic_t res_cur_count;
158 
159 	struct dlm_debug_ctxt *dlm_debug_ctxt;
160 	struct dentry *dlm_debugfs_subroot;
161 
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;
171 	struct workqueue_struct *dlm_worker;
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;
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;
188 };
189 
190 static 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 
195 static 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 
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. */
206 void dlm_dispatch_work(struct work_struct *work);
207 
208 struct dlm_lock_resource;
209 struct dlm_work_item;
210 
211 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);
212 
213 struct dlm_request_all_locks_priv
214 {
215 	u8 reco_master;
216 	u8 dead_node;
217 };
218 
219 struct dlm_mig_lockres_priv
220 {
221 	struct dlm_lock_resource *lockres;
222 	u8 real_master;
223 	u8 extra_ref;
224 };
225 
226 struct 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 
234 struct dlm_deref_lockres_priv
235 {
236 	struct dlm_lock_resource *deref_res;
237 	u8 deref_node;
238 };
239 
240 struct 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;
250 		struct dlm_deref_lockres_priv dl;
251 	} u;
252 };
253 
254 static 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 
267 static 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
282 #define DLM_LOCK_RES_DROPPING_REF         0x00000040
283 #define DLM_LOCK_RES_BLOCK_DIRTY          0x00001000
284 #define DLM_LOCK_RES_SETREF_INPROG        0x00002000
285 
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 
289 #define DLM_PURGE_INTERVAL_MS   (8 * 1000)
290 
291 struct dlm_lock_resource
292 {
293 	/* WARNING: Please see the comment in dlm_init_lockres before
294 	 * adding fields here. */
295 	struct hlist_node hash_node;
296 	struct qstr lockname;
297 	struct kref      refs;
298 
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 	 */
305 	struct list_head granted;
306 	struct list_head converting;
307 	struct list_head blocked;
308 	struct list_head purge;
309 
310 	/*
311 	 * These two lists require you to hold an additional reference
312 	 * while they are on the list.
313 	 */
314 	struct list_head dirty;
315 	struct list_head recovering; // dlm_recovery_ctxt.resources list
316 
317 	/* Added during init and removed during release */
318 	struct list_head tracking;	/* dlm->tracking_list */
319 
320 	/* unused lock resources have their last_used stamped and are
321 	 * put on a list for the dlm thread to run. */
322 	unsigned long    last_used;
323 
324 	struct dlm_ctxt *dlm;
325 
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;
332 	char lvb[DLM_LVB_LEN];
333 	unsigned int inflight_locks;
334 	unsigned int inflight_assert_workers;
335 	unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
336 };
337 
338 struct 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 
354 struct 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 
379 enum dlm_lockres_list {
380 	DLM_GRANTED_LIST = 0,
381 	DLM_CONVERTING_LIST = 1,
382 	DLM_BLOCKED_LIST = 2,
383 };
384 
385 static 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 
394 static 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 
406 static inline struct list_head *
407 dlm_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 
424 struct dlm_node_iter
425 {
426 	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
427 	int curnode;
428 };
429 
430 
431 enum {
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,
453 	DLM_BEGIN_EXIT_DOMAIN_MSG	= 521,
454 };
455 
456 struct dlm_reco_node_data
457 {
458 	int state;
459 	u8 node_num;
460 	struct list_head list;
461 };
462 
463 enum {
464 	DLM_RECO_NODE_DATA_DEAD = -1,
465 	DLM_RECO_NODE_DATA_INIT = 0,
466 	DLM_RECO_NODE_DATA_REQUESTING = 1,
467 	DLM_RECO_NODE_DATA_REQUESTED = 2,
468 	DLM_RECO_NODE_DATA_RECEIVING = 3,
469 	DLM_RECO_NODE_DATA_DONE = 4,
470 	DLM_RECO_NODE_DATA_FINALIZE_SENT = 5,
471 };
472 
473 
474 enum {
475 	DLM_MASTER_RESP_NO = 0,
476 	DLM_MASTER_RESP_YES = 1,
477 	DLM_MASTER_RESP_MAYBE = 2,
478 	DLM_MASTER_RESP_ERROR = 3,
479 };
480 
481 
482 struct dlm_master_request
483 {
484 	u8 node_idx;
485 	u8 namelen;
486 	__be16 pad1;
487 	__be32 flags;
488 
489 	u8 name[O2NM_MAX_NAME_LEN];
490 };
491 
492 #define DLM_ASSERT_RESPONSE_REASSERT       0x00000001
493 #define DLM_ASSERT_RESPONSE_MASTERY_REF    0x00000002
494 
495 #define DLM_ASSERT_MASTER_MLE_CLEANUP      0x00000001
496 #define DLM_ASSERT_MASTER_REQUERY          0x00000002
497 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
498 struct dlm_assert_master
499 {
500 	u8 node_idx;
501 	u8 namelen;
502 	__be16 pad1;
503 	__be32 flags;
504 
505 	u8 name[O2NM_MAX_NAME_LEN];
506 };
507 
508 #define DLM_MIGRATE_RESPONSE_MASTERY_REF   0x00000001
509 
510 struct dlm_migrate_request
511 {
512 	u8 master;
513 	u8 new_master;
514 	u8 namelen;
515 	u8 pad1;
516 	__be32 pad2;
517 	u8 name[O2NM_MAX_NAME_LEN];
518 };
519 
520 struct dlm_master_requery
521 {
522 	u8 pad1;
523 	u8 pad2;
524 	u8 node_idx;
525 	u8 namelen;
526 	__be32 pad3;
527 	u8 name[O2NM_MAX_NAME_LEN];
528 };
529 
530 #define DLM_MRES_RECOVERY   0x01
531 #define DLM_MRES_MIGRATION  0x02
532 #define DLM_MRES_ALL_DONE   0x04
533 
534 /*
535  * We would like to get one whole lockres into a single network
536  * message whenever possible.  Generally speaking, there will be
537  * at most one dlm_lock on a lockres for each node in the cluster,
538  * plus (infrequently) any additional locks coming in from userdlm.
539  *
540  * struct _dlm_lockres_page
541  * {
542  * 	dlm_migratable_lockres mres;
543  * 	dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
544  * 	u8 pad[DLM_MIG_LOCKRES_RESERVED];
545  * };
546  *
547  * from ../cluster/tcp.h
548  *    NET_MAX_PAYLOAD_BYTES  (4096 - sizeof(net_msg))
549  *    (roughly 4080 bytes)
550  * and sizeof(dlm_migratable_lockres) = 112 bytes
551  * and sizeof(dlm_migratable_lock) = 16 bytes
552  *
553  * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
554  * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
555  *
556  *  (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) +
557  *     sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED =
558  *        NET_MAX_PAYLOAD_BYTES
559  *  (240 * 16) + 112 + 128 = 4080
560  *
561  * So a lockres would need more than 240 locks before it would
562  * use more than one network packet to recover.  Not too bad.
563  */
564 #define DLM_MAX_MIGRATABLE_LOCKS   240
565 
566 struct dlm_migratable_lockres
567 {
568 	u8 master;
569 	u8 lockname_len;
570 	u8 num_locks;    // locks sent in this structure
571 	u8 flags;
572 	__be32 total_locks; // locks to be sent for this migration cookie
573 	__be64 mig_cookie;  // cookie for this lockres migration
574 			 // or zero if not needed
575 	// 16 bytes
576 	u8 lockname[DLM_LOCKID_NAME_MAX];
577 	// 48 bytes
578 	u8 lvb[DLM_LVB_LEN];
579 	// 112 bytes
580 	struct dlm_migratable_lock ml[0];  // 16 bytes each, begins at byte 112
581 };
582 #define DLM_MIG_LOCKRES_MAX_LEN  \
583 	(sizeof(struct dlm_migratable_lockres) + \
584 	 (sizeof(struct dlm_migratable_lock) * \
585 	  DLM_MAX_MIGRATABLE_LOCKS) )
586 
587 /* from above, 128 bytes
588  * for some undetermined future use */
589 #define DLM_MIG_LOCKRES_RESERVED   (NET_MAX_PAYLOAD_BYTES - \
590 				    DLM_MIG_LOCKRES_MAX_LEN)
591 
592 struct dlm_create_lock
593 {
594 	__be64 cookie;
595 
596 	__be32 flags;
597 	u8 pad1;
598 	u8 node_idx;
599 	s8 requested_type;
600 	u8 namelen;
601 
602 	u8 name[O2NM_MAX_NAME_LEN];
603 };
604 
605 struct dlm_convert_lock
606 {
607 	__be64 cookie;
608 
609 	__be32 flags;
610 	u8 pad1;
611 	u8 node_idx;
612 	s8 requested_type;
613 	u8 namelen;
614 
615 	u8 name[O2NM_MAX_NAME_LEN];
616 
617 	s8 lvb[0];
618 };
619 #define DLM_CONVERT_LOCK_MAX_LEN  (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN)
620 
621 struct dlm_unlock_lock
622 {
623 	__be64 cookie;
624 
625 	__be32 flags;
626 	__be16 pad1;
627 	u8 node_idx;
628 	u8 namelen;
629 
630 	u8 name[O2NM_MAX_NAME_LEN];
631 
632 	s8 lvb[0];
633 };
634 #define DLM_UNLOCK_LOCK_MAX_LEN  (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN)
635 
636 struct dlm_proxy_ast
637 {
638 	__be64 cookie;
639 
640 	__be32 flags;
641 	u8 node_idx;
642 	u8 type;
643 	u8 blocked_type;
644 	u8 namelen;
645 
646 	u8 name[O2NM_MAX_NAME_LEN];
647 
648 	s8 lvb[0];
649 };
650 #define DLM_PROXY_AST_MAX_LEN  (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN)
651 
652 #define DLM_MOD_KEY (0x666c6172)
653 enum dlm_query_join_response_code {
654 	JOIN_DISALLOW = 0,
655 	JOIN_OK = 1,
656 	JOIN_OK_NO_MAP = 2,
657 	JOIN_PROTOCOL_MISMATCH = 3,
658 };
659 
660 struct dlm_query_join_packet {
661 	u8 code;	/* Response code.  dlm_minor and fs_minor
662 			   are only valid if this is JOIN_OK */
663 	u8 dlm_minor;	/* The minor version of the protocol the
664 			   dlm is speaking. */
665 	u8 fs_minor;	/* The minor version of the protocol the
666 			   filesystem is speaking. */
667 	u8 reserved;
668 };
669 
670 union dlm_query_join_response {
671 	__be32 intval;
672 	struct dlm_query_join_packet packet;
673 };
674 
675 struct dlm_lock_request
676 {
677 	u8 node_idx;
678 	u8 dead_node;
679 	__be16 pad1;
680 	__be32 pad2;
681 };
682 
683 struct dlm_reco_data_done
684 {
685 	u8 node_idx;
686 	u8 dead_node;
687 	__be16 pad1;
688 	__be32 pad2;
689 
690 	/* unused for now */
691 	/* eventually we can use this to attempt
692 	 * lvb recovery based on each node's info */
693 	u8 reco_lvb[DLM_LVB_LEN];
694 };
695 
696 struct dlm_begin_reco
697 {
698 	u8 node_idx;
699 	u8 dead_node;
700 	__be16 pad1;
701 	__be32 pad2;
702 };
703 
704 
705 #define BITS_PER_BYTE 8
706 #define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE)
707 
708 struct dlm_query_join_request
709 {
710 	u8 node_idx;
711 	u8 pad1[2];
712 	u8 name_len;
713 	struct dlm_protocol_version dlm_proto;
714 	struct dlm_protocol_version fs_proto;
715 	u8 domain[O2NM_MAX_NAME_LEN];
716 	u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)];
717 };
718 
719 struct dlm_assert_joined
720 {
721 	u8 node_idx;
722 	u8 pad1[2];
723 	u8 name_len;
724 	u8 domain[O2NM_MAX_NAME_LEN];
725 };
726 
727 struct dlm_cancel_join
728 {
729 	u8 node_idx;
730 	u8 pad1[2];
731 	u8 name_len;
732 	u8 domain[O2NM_MAX_NAME_LEN];
733 };
734 
735 struct dlm_query_region {
736 	u8 qr_node;
737 	u8 qr_numregions;
738 	u8 qr_namelen;
739 	u8 pad1;
740 	u8 qr_domain[O2NM_MAX_NAME_LEN];
741 	u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS];
742 };
743 
744 struct dlm_node_info {
745 	u8 ni_nodenum;
746 	u8 pad1;
747 	__be16 ni_ipv4_port;
748 	__be32 ni_ipv4_address;
749 };
750 
751 struct dlm_query_nodeinfo {
752 	u8 qn_nodenum;
753 	u8 qn_numnodes;
754 	u8 qn_namelen;
755 	u8 pad1;
756 	u8 qn_domain[O2NM_MAX_NAME_LEN];
757 	struct dlm_node_info qn_nodes[O2NM_MAX_NODES];
758 };
759 
760 struct dlm_exit_domain
761 {
762 	u8 node_idx;
763 	u8 pad1[3];
764 };
765 
766 struct dlm_finalize_reco
767 {
768 	u8 node_idx;
769 	u8 dead_node;
770 	u8 flags;
771 	u8 pad1;
772 	__be32 pad2;
773 };
774 
775 struct dlm_deref_lockres
776 {
777 	u32 pad1;
778 	u16 pad2;
779 	u8 node_idx;
780 	u8 namelen;
781 
782 	u8 name[O2NM_MAX_NAME_LEN];
783 };
784 
785 static inline enum dlm_status
786 __dlm_lockres_state_to_status(struct dlm_lock_resource *res)
787 {
788 	enum dlm_status status = DLM_NORMAL;
789 
790 	assert_spin_locked(&res->spinlock);
791 
792 	if (res->state & DLM_LOCK_RES_RECOVERING)
793 		status = DLM_RECOVERING;
794 	else if (res->state & DLM_LOCK_RES_MIGRATING)
795 		status = DLM_MIGRATING;
796 	else if (res->state & DLM_LOCK_RES_IN_PROGRESS)
797 		status = DLM_FORWARD;
798 
799 	return status;
800 }
801 
802 static inline u8 dlm_get_lock_cookie_node(u64 cookie)
803 {
804 	u8 ret;
805 	cookie >>= 56;
806 	ret = (u8)(cookie & 0xffULL);
807 	return ret;
808 }
809 
810 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
811 {
812 	unsigned long long ret;
813 	ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
814 	return ret;
815 }
816 
817 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
818 			       struct dlm_lockstatus *lksb);
819 void dlm_lock_get(struct dlm_lock *lock);
820 void dlm_lock_put(struct dlm_lock *lock);
821 
822 void dlm_lock_attach_lockres(struct dlm_lock *lock,
823 			     struct dlm_lock_resource *res);
824 
825 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
826 			    void **ret_data);
827 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
828 			     void **ret_data);
829 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
830 			  void **ret_data);
831 
832 void dlm_revert_pending_convert(struct dlm_lock_resource *res,
833 				struct dlm_lock *lock);
834 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
835 			     struct dlm_lock *lock);
836 
837 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
838 			    void **ret_data);
839 void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
840 			       struct dlm_lock *lock);
841 void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
842 			       struct dlm_lock *lock);
843 
844 int dlm_launch_thread(struct dlm_ctxt *dlm);
845 void dlm_complete_thread(struct dlm_ctxt *dlm);
846 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
847 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
848 void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
849 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
850 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
851 void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
852 void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
853 
854 void dlm_put(struct dlm_ctxt *dlm);
855 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
856 int dlm_domain_fully_joined(struct dlm_ctxt *dlm);
857 
858 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
859 			      struct dlm_lock_resource *res);
860 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
861 			    struct dlm_lock_resource *res);
862 static inline void dlm_lockres_get(struct dlm_lock_resource *res)
863 {
864 	/* This is called on every lookup, so it might be worth
865 	 * inlining. */
866 	kref_get(&res->refs);
867 }
868 void dlm_lockres_put(struct dlm_lock_resource *res);
869 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
870 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
871 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
872 						     const char *name,
873 						     unsigned int len,
874 						     unsigned int hash);
875 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
876 						const char *name,
877 						unsigned int len,
878 						unsigned int hash);
879 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
880 					      const char *name,
881 					      unsigned int len);
882 
883 int dlm_is_host_down(int errno);
884 
885 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
886 						 const char *lockid,
887 						 int namelen,
888 						 int flags);
889 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
890 					  const char *name,
891 					  unsigned int namelen);
892 
893 void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
894 				struct dlm_lock_resource *res, int bit);
895 void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
896 				  struct dlm_lock_resource *res, int bit);
897 
898 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
899 				   struct dlm_lock_resource *res);
900 void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
901 				   struct dlm_lock_resource *res);
902 
903 void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm,
904 		struct dlm_lock_resource *res);
905 
906 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
907 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
908 void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
909 void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
910 void dlm_do_local_ast(struct dlm_ctxt *dlm,
911 		      struct dlm_lock_resource *res,
912 		      struct dlm_lock *lock);
913 int dlm_do_remote_ast(struct dlm_ctxt *dlm,
914 		      struct dlm_lock_resource *res,
915 		      struct dlm_lock *lock);
916 void dlm_do_local_bast(struct dlm_ctxt *dlm,
917 		       struct dlm_lock_resource *res,
918 		       struct dlm_lock *lock,
919 		       int blocked_type);
920 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm,
921 			   struct dlm_lock_resource *res,
922 			   struct dlm_lock *lock,
923 			   int msg_type,
924 			   int blocked_type, int flags);
925 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm,
926 				      struct dlm_lock_resource *res,
927 				      struct dlm_lock *lock,
928 				      int blocked_type)
929 {
930 	return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST,
931 				      blocked_type, 0);
932 }
933 
934 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm,
935 				     struct dlm_lock_resource *res,
936 				     struct dlm_lock *lock,
937 				     int flags)
938 {
939 	return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST,
940 				      0, flags);
941 }
942 
943 void dlm_print_one_lock_resource(struct dlm_lock_resource *res);
944 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res);
945 
946 u8 dlm_nm_this_node(struct dlm_ctxt *dlm);
947 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
948 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
949 
950 
951 int dlm_nm_init(struct dlm_ctxt *dlm);
952 int dlm_heartbeat_init(struct dlm_ctxt *dlm);
953 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data);
954 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data);
955 
956 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
957 int dlm_finish_migration(struct dlm_ctxt *dlm,
958 			 struct dlm_lock_resource *res,
959 			 u8 old_master);
960 void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
961 			     struct dlm_lock_resource *res);
962 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res);
963 
964 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
965 			       void **ret_data);
966 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
967 			      void **ret_data);
968 void dlm_assert_master_post_handler(int status, void *data, void *ret_data);
969 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
970 			      void **ret_data);
971 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
972 				void **ret_data);
973 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
974 			    void **ret_data);
975 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
976 			       void **ret_data);
977 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
978 				  void **ret_data);
979 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
980 			       void **ret_data);
981 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
982 			   void **ret_data);
983 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
984 			      void **ret_data);
985 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
986 			  u8 nodenum, u8 *real_master);
987 
988 
989 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
990 			       struct dlm_lock_resource *res,
991 			       int ignore_higher,
992 			       u8 request_from,
993 			       u32 flags);
994 
995 
996 int dlm_send_one_lockres(struct dlm_ctxt *dlm,
997 			 struct dlm_lock_resource *res,
998 			 struct dlm_migratable_lockres *mres,
999 			 u8 send_to,
1000 			 u8 flags);
1001 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
1002 				       struct dlm_lock_resource *res);
1003 
1004 /* will exit holding res->spinlock, but may drop in function */
1005 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags);
1006 
1007 /* will exit holding res->spinlock, but may drop in function */
1008 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res)
1009 {
1010 	__dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS|
1011 				    	  DLM_LOCK_RES_RECOVERING|
1012 					  DLM_LOCK_RES_MIGRATING));
1013 }
1014 
1015 void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1016 void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
1017 
1018 /* create/destroy slab caches */
1019 int dlm_init_master_caches(void);
1020 void dlm_destroy_master_caches(void);
1021 
1022 int dlm_init_lock_cache(void);
1023 void dlm_destroy_lock_cache(void);
1024 
1025 int dlm_init_mle_cache(void);
1026 void dlm_destroy_mle_cache(void);
1027 
1028 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up);
1029 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
1030 			 struct dlm_lock_resource *res);
1031 void dlm_clean_master_list(struct dlm_ctxt *dlm,
1032 			   u8 dead_node);
1033 void dlm_force_free_mles(struct dlm_ctxt *dlm);
1034 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
1035 int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
1036 int __dlm_lockres_unused(struct dlm_lock_resource *res);
1037 
1038 static inline const char * dlm_lock_mode_name(int mode)
1039 {
1040 	switch (mode) {
1041 		case LKM_EXMODE:
1042 			return "EX";
1043 		case LKM_PRMODE:
1044 			return "PR";
1045 		case LKM_NLMODE:
1046 			return "NL";
1047 	}
1048 	return "UNKNOWN";
1049 }
1050 
1051 
1052 static inline int dlm_lock_compatible(int existing, int request)
1053 {
1054 	/* NO_LOCK compatible with all */
1055 	if (request == LKM_NLMODE ||
1056 	    existing == LKM_NLMODE)
1057 		return 1;
1058 
1059 	/* EX incompatible with all non-NO_LOCK */
1060 	if (request == LKM_EXMODE)
1061 		return 0;
1062 
1063 	/* request must be PR, which is compatible with PR */
1064 	if (existing == LKM_PRMODE)
1065 		return 1;
1066 
1067 	return 0;
1068 }
1069 
1070 static inline int dlm_lock_on_list(struct list_head *head,
1071 				   struct dlm_lock *lock)
1072 {
1073 	struct dlm_lock *tmplock;
1074 
1075 	list_for_each_entry(tmplock, head, list) {
1076 		if (tmplock == lock)
1077 			return 1;
1078 	}
1079 	return 0;
1080 }
1081 
1082 
1083 static inline enum dlm_status dlm_err_to_dlm_status(int err)
1084 {
1085 	enum dlm_status ret;
1086 	if (err == -ENOMEM)
1087 		ret = DLM_SYSERR;
1088 	else if (err == -ETIMEDOUT || o2net_link_down(err, NULL))
1089 		ret = DLM_NOLOCKMGR;
1090 	else if (err == -EINVAL)
1091 		ret = DLM_BADPARAM;
1092 	else if (err == -ENAMETOOLONG)
1093 		ret = DLM_IVBUFLEN;
1094 	else
1095 		ret = DLM_BADARGS;
1096 	return ret;
1097 }
1098 
1099 
1100 static inline void dlm_node_iter_init(unsigned long *map,
1101 				      struct dlm_node_iter *iter)
1102 {
1103 	memcpy(iter->node_map, map, sizeof(iter->node_map));
1104 	iter->curnode = -1;
1105 }
1106 
1107 static inline int dlm_node_iter_next(struct dlm_node_iter *iter)
1108 {
1109 	int bit;
1110 	bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1);
1111 	if (bit >= O2NM_MAX_NODES) {
1112 		iter->curnode = O2NM_MAX_NODES;
1113 		return -ENOENT;
1114 	}
1115 	iter->curnode = bit;
1116 	return bit;
1117 }
1118 
1119 static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
1120 					 struct dlm_lock_resource *res,
1121 					 u8 owner)
1122 {
1123 	assert_spin_locked(&res->spinlock);
1124 
1125 	res->owner = owner;
1126 }
1127 
1128 static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
1129 					    struct dlm_lock_resource *res,
1130 					    u8 owner)
1131 {
1132 	assert_spin_locked(&res->spinlock);
1133 
1134 	if (owner != res->owner)
1135 		dlm_set_lockres_owner(dlm, res, owner);
1136 }
1137 
1138 #endif /* DLMCOMMON_H */
1139