Commit | Line | Data |
---|---|---|
e7fd4179 DT |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5 | ** Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
6 | ** | |
7 | ** This copyrighted material is made available to anyone wishing to use, | |
8 | ** modify, copy, or redistribute it subject to the terms and conditions | |
9 | ** of the GNU General Public License v.2. | |
10 | ** | |
11 | ******************************************************************************* | |
12 | ******************************************************************************/ | |
13 | ||
14 | #ifndef __DLM_INTERNAL_DOT_H__ | |
15 | #define __DLM_INTERNAL_DOT_H__ | |
16 | ||
17 | /* | |
18 | * This is the main header file to be included in each DLM source file. | |
19 | */ | |
20 | ||
21 | #include <linux/module.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/sched.h> | |
24 | #include <linux/types.h> | |
25 | #include <linux/ctype.h> | |
26 | #include <linux/spinlock.h> | |
27 | #include <linux/vmalloc.h> | |
28 | #include <linux/list.h> | |
29 | #include <linux/errno.h> | |
30 | #include <linux/random.h> | |
31 | #include <linux/delay.h> | |
32 | #include <linux/socket.h> | |
33 | #include <linux/kthread.h> | |
34 | #include <linux/kobject.h> | |
35 | #include <linux/kref.h> | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/jhash.h> | |
90135925 | 38 | #include <linux/mutex.h> |
e7fd4179 DT |
39 | #include <asm/semaphore.h> |
40 | #include <asm/uaccess.h> | |
41 | ||
42 | #include <linux/dlm.h> | |
43 | ||
44 | #define DLM_LOCKSPACE_LEN 64 | |
45 | ||
e7fd4179 DT |
46 | /* Size of the temp buffer midcomms allocates on the stack. |
47 | We try to make this large enough so most messages fit. | |
48 | FIXME: should sctp make this unnecessary? */ | |
49 | ||
50 | #define DLM_INBUF_LEN 148 | |
51 | ||
52 | struct dlm_ls; | |
53 | struct dlm_lkb; | |
54 | struct dlm_rsb; | |
55 | struct dlm_member; | |
56 | struct dlm_lkbtable; | |
57 | struct dlm_rsbtable; | |
58 | struct dlm_dirtable; | |
59 | struct dlm_direntry; | |
60 | struct dlm_recover; | |
61 | struct dlm_header; | |
62 | struct dlm_message; | |
63 | struct dlm_rcom; | |
64 | struct dlm_mhandle; | |
65 | ||
66 | #define log_print(fmt, args...) \ | |
67 | printk(KERN_ERR "dlm: "fmt"\n" , ##args) | |
68 | #define log_error(ls, fmt, args...) \ | |
69 | printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args) | |
70 | ||
71 | #ifdef DLM_LOG_DEBUG | |
72 | #define log_debug(ls, fmt, args...) log_error(ls, fmt, ##args) | |
73 | #else | |
74 | #define log_debug(ls, fmt, args...) | |
75 | #endif | |
76 | ||
77 | #define DLM_ASSERT(x, do) \ | |
78 | { \ | |
79 | if (!(x)) \ | |
80 | { \ | |
81 | printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \ | |
82 | "DLM: assertion: \"%s\"\n" \ | |
83 | "DLM: time = %lu\n", \ | |
84 | __LINE__, __FILE__, #x, jiffies); \ | |
85 | {do} \ | |
86 | printk("\n"); \ | |
87 | BUG(); \ | |
88 | panic("DLM: Record message above and reboot.\n"); \ | |
89 | } \ | |
90 | } | |
91 | ||
92 | ||
93 | struct dlm_direntry { | |
94 | struct list_head list; | |
95 | uint32_t master_nodeid; | |
96 | uint16_t length; | |
97 | char name[1]; | |
98 | }; | |
99 | ||
100 | struct dlm_dirtable { | |
101 | struct list_head list; | |
102 | rwlock_t lock; | |
103 | }; | |
104 | ||
105 | struct dlm_rsbtable { | |
106 | struct list_head list; | |
107 | struct list_head toss; | |
108 | rwlock_t lock; | |
109 | }; | |
110 | ||
111 | struct dlm_lkbtable { | |
112 | struct list_head list; | |
113 | rwlock_t lock; | |
114 | uint16_t counter; | |
115 | }; | |
116 | ||
117 | /* | |
118 | * Lockspace member (per node in a ls) | |
119 | */ | |
120 | ||
121 | struct dlm_member { | |
122 | struct list_head list; | |
123 | int nodeid; | |
124 | int weight; | |
125 | }; | |
126 | ||
127 | /* | |
128 | * Save and manage recovery state for a lockspace. | |
129 | */ | |
130 | ||
131 | struct dlm_recover { | |
132 | struct list_head list; | |
133 | int *nodeids; | |
134 | int node_count; | |
135 | uint64_t seq; | |
136 | }; | |
137 | ||
138 | /* | |
139 | * Pass input args to second stage locking function. | |
140 | */ | |
141 | ||
142 | struct dlm_args { | |
143 | uint32_t flags; | |
144 | void *astaddr; | |
145 | long astparam; | |
146 | void *bastaddr; | |
147 | int mode; | |
148 | struct dlm_lksb *lksb; | |
e7fd4179 DT |
149 | }; |
150 | ||
151 | ||
152 | /* | |
153 | * Lock block | |
154 | * | |
155 | * A lock can be one of three types: | |
156 | * | |
157 | * local copy lock is mastered locally | |
158 | * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set) | |
159 | * process copy lock is mastered on a remote node | |
160 | * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set) | |
161 | * master copy master node's copy of a lock owned by remote node | |
162 | * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set) | |
163 | * | |
164 | * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or | |
165 | * dlm_unlock. The dlm does not modify these or use any private flags in | |
166 | * this field; it only contains DLM_LKF_ flags from dlm.h. These flags | |
167 | * are sent as-is to the remote master when the lock is remote. | |
168 | * | |
169 | * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h. | |
170 | * Some internal flags are shared between the master and process nodes; | |
171 | * these shared flags are kept in the lower two bytes. One of these | |
172 | * flags set on the master copy will be propagated to the process copy | |
173 | * and v.v. Other internal flags are private to the master or process | |
174 | * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes. | |
175 | * | |
176 | * lkb_sbflags: status block flags. These flags are copied directly into | |
177 | * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion | |
178 | * ast. All defined in dlm.h with DLM_SBF_ prefix. | |
179 | * | |
180 | * lkb_status: the lock status indicates which rsb queue the lock is | |
181 | * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT | |
182 | * | |
183 | * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a | |
184 | * reply is needed. Only set when the lkb is on the lockspace waiters | |
185 | * list awaiting a reply from a remote node. | |
186 | * | |
187 | * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb | |
188 | * is a master copy, nodeid specifies the remote lock holder, when the | |
189 | * lkb is a process copy, the nodeid specifies the lock master. | |
190 | */ | |
191 | ||
192 | /* lkb_ast_type */ | |
193 | ||
194 | #define AST_COMP 1 | |
195 | #define AST_BAST 2 | |
196 | ||
e7fd4179 DT |
197 | /* lkb_status */ |
198 | ||
199 | #define DLM_LKSTS_WAITING 1 | |
200 | #define DLM_LKSTS_GRANTED 2 | |
201 | #define DLM_LKSTS_CONVERT 3 | |
202 | ||
203 | /* lkb_flags */ | |
204 | ||
205 | #define DLM_IFL_MSTCPY 0x00010000 | |
206 | #define DLM_IFL_RESEND 0x00020000 | |
e7fd4179 DT |
207 | |
208 | struct dlm_lkb { | |
209 | struct dlm_rsb *lkb_resource; /* the rsb */ | |
210 | struct kref lkb_ref; | |
211 | int lkb_nodeid; /* copied from rsb */ | |
212 | int lkb_ownpid; /* pid of lock owner */ | |
213 | uint32_t lkb_id; /* our lock ID */ | |
214 | uint32_t lkb_remid; /* lock ID on remote partner */ | |
215 | uint32_t lkb_exflags; /* external flags from caller */ | |
216 | uint32_t lkb_sbflags; /* lksb flags */ | |
217 | uint32_t lkb_flags; /* internal flags */ | |
218 | uint32_t lkb_lvbseq; /* lvb sequence number */ | |
219 | ||
220 | int8_t lkb_status; /* granted, waiting, convert */ | |
221 | int8_t lkb_rqmode; /* requested lock mode */ | |
222 | int8_t lkb_grmode; /* granted lock mode */ | |
223 | int8_t lkb_bastmode; /* requested mode */ | |
224 | int8_t lkb_highbast; /* highest mode bast sent for */ | |
225 | ||
226 | int8_t lkb_wait_type; /* type of reply waiting for */ | |
227 | int8_t lkb_ast_type; /* type of ast queued for */ | |
228 | ||
229 | struct list_head lkb_idtbl_list; /* lockspace lkbtbl */ | |
230 | struct list_head lkb_statequeue; /* rsb g/c/w list */ | |
231 | struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */ | |
232 | struct list_head lkb_wait_reply; /* waiting for remote reply */ | |
233 | struct list_head lkb_astqueue; /* need ast to be sent */ | |
234 | ||
e7fd4179 DT |
235 | char *lkb_lvbptr; |
236 | struct dlm_lksb *lkb_lksb; /* caller's status block */ | |
237 | void *lkb_astaddr; /* caller's ast function */ | |
238 | void *lkb_bastaddr; /* caller's bast function */ | |
239 | long lkb_astparam; /* caller's ast arg */ | |
240 | }; | |
241 | ||
242 | ||
243 | struct dlm_rsb { | |
244 | struct dlm_ls *res_ls; /* the lockspace */ | |
245 | struct kref res_ref; | |
90135925 | 246 | struct mutex res_mutex; |
e7fd4179 DT |
247 | unsigned long res_flags; |
248 | int res_length; /* length of rsb name */ | |
249 | int res_nodeid; | |
250 | uint32_t res_lvbseq; | |
251 | uint32_t res_hash; | |
252 | uint32_t res_bucket; /* rsbtbl */ | |
253 | unsigned long res_toss_time; | |
254 | uint32_t res_first_lkid; | |
255 | struct list_head res_lookup; /* lkbs waiting on first */ | |
256 | struct list_head res_hashchain; /* rsbtbl */ | |
257 | struct list_head res_grantqueue; | |
258 | struct list_head res_convertqueue; | |
259 | struct list_head res_waitqueue; | |
260 | ||
261 | struct list_head res_root_list; /* used for recovery */ | |
262 | struct list_head res_recover_list; /* used for recovery */ | |
263 | int res_recover_locks_count; | |
264 | ||
265 | char *res_lvbptr; | |
266 | char res_name[1]; | |
267 | }; | |
268 | ||
269 | /* find_rsb() flags */ | |
270 | ||
271 | #define R_MASTER 1 /* only return rsb if it's a master */ | |
272 | #define R_CREATE 2 /* create/add rsb if not found */ | |
273 | ||
274 | /* rsb_flags */ | |
275 | ||
276 | enum rsb_flags { | |
277 | RSB_MASTER_UNCERTAIN, | |
278 | RSB_VALNOTVALID, | |
279 | RSB_VALNOTVALID_PREV, | |
280 | RSB_NEW_MASTER, | |
281 | RSB_NEW_MASTER2, | |
282 | RSB_RECOVER_CONVERT, | |
283 | }; | |
284 | ||
285 | static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag) | |
286 | { | |
287 | __set_bit(flag, &r->res_flags); | |
288 | } | |
289 | ||
290 | static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag) | |
291 | { | |
292 | __clear_bit(flag, &r->res_flags); | |
293 | } | |
294 | ||
295 | static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag) | |
296 | { | |
297 | return test_bit(flag, &r->res_flags); | |
298 | } | |
299 | ||
300 | ||
301 | /* dlm_header is first element of all structs sent between nodes */ | |
302 | ||
303 | #define DLM_HEADER_MAJOR 0x00020000 | |
304 | #define DLM_HEADER_MINOR 0x00000001 | |
305 | ||
306 | #define DLM_MSG 1 | |
307 | #define DLM_RCOM 2 | |
308 | ||
309 | struct dlm_header { | |
310 | uint32_t h_version; | |
311 | uint32_t h_lockspace; | |
312 | uint32_t h_nodeid; /* nodeid of sender */ | |
313 | uint16_t h_length; | |
314 | uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */ | |
315 | uint8_t h_pad; | |
316 | }; | |
317 | ||
318 | ||
319 | #define DLM_MSG_REQUEST 1 | |
320 | #define DLM_MSG_CONVERT 2 | |
321 | #define DLM_MSG_UNLOCK 3 | |
322 | #define DLM_MSG_CANCEL 4 | |
323 | #define DLM_MSG_REQUEST_REPLY 5 | |
324 | #define DLM_MSG_CONVERT_REPLY 6 | |
325 | #define DLM_MSG_UNLOCK_REPLY 7 | |
326 | #define DLM_MSG_CANCEL_REPLY 8 | |
327 | #define DLM_MSG_GRANT 9 | |
328 | #define DLM_MSG_BAST 10 | |
329 | #define DLM_MSG_LOOKUP 11 | |
330 | #define DLM_MSG_REMOVE 12 | |
331 | #define DLM_MSG_LOOKUP_REPLY 13 | |
332 | ||
333 | struct dlm_message { | |
334 | struct dlm_header m_header; | |
335 | uint32_t m_type; /* DLM_MSG_ */ | |
336 | uint32_t m_nodeid; | |
337 | uint32_t m_pid; | |
338 | uint32_t m_lkid; /* lkid on sender */ | |
339 | uint32_t m_remid; /* lkid on receiver */ | |
340 | uint32_t m_parent_lkid; | |
341 | uint32_t m_parent_remid; | |
342 | uint32_t m_exflags; | |
343 | uint32_t m_sbflags; | |
344 | uint32_t m_flags; | |
345 | uint32_t m_lvbseq; | |
346 | uint32_t m_hash; | |
347 | int m_status; | |
348 | int m_grmode; | |
349 | int m_rqmode; | |
350 | int m_bastmode; | |
351 | int m_asts; | |
352 | int m_result; /* 0 or -EXXX */ | |
e7fd4179 DT |
353 | char m_extra[0]; /* name or lvb */ |
354 | }; | |
355 | ||
356 | ||
357 | #define DLM_RS_NODES 0x00000001 | |
358 | #define DLM_RS_NODES_ALL 0x00000002 | |
359 | #define DLM_RS_DIR 0x00000004 | |
360 | #define DLM_RS_DIR_ALL 0x00000008 | |
361 | #define DLM_RS_LOCKS 0x00000010 | |
362 | #define DLM_RS_LOCKS_ALL 0x00000020 | |
363 | #define DLM_RS_DONE 0x00000040 | |
364 | #define DLM_RS_DONE_ALL 0x00000080 | |
365 | ||
366 | #define DLM_RCOM_STATUS 1 | |
367 | #define DLM_RCOM_NAMES 2 | |
368 | #define DLM_RCOM_LOOKUP 3 | |
369 | #define DLM_RCOM_LOCK 4 | |
370 | #define DLM_RCOM_STATUS_REPLY 5 | |
371 | #define DLM_RCOM_NAMES_REPLY 6 | |
372 | #define DLM_RCOM_LOOKUP_REPLY 7 | |
373 | #define DLM_RCOM_LOCK_REPLY 8 | |
374 | ||
375 | struct dlm_rcom { | |
376 | struct dlm_header rc_header; | |
377 | uint32_t rc_type; /* DLM_RCOM_ */ | |
378 | int rc_result; /* multi-purpose */ | |
379 | uint64_t rc_id; /* match reply with request */ | |
380 | char rc_buf[0]; | |
381 | }; | |
382 | ||
383 | struct rcom_config { | |
384 | uint32_t rf_lvblen; | |
385 | uint32_t rf_lsflags; | |
386 | uint64_t rf_unused; | |
387 | }; | |
388 | ||
389 | struct rcom_lock { | |
390 | uint32_t rl_ownpid; | |
391 | uint32_t rl_lkid; | |
392 | uint32_t rl_remid; | |
393 | uint32_t rl_parent_lkid; | |
394 | uint32_t rl_parent_remid; | |
395 | uint32_t rl_exflags; | |
396 | uint32_t rl_flags; | |
397 | uint32_t rl_lvbseq; | |
398 | int rl_result; | |
399 | int8_t rl_rqmode; | |
400 | int8_t rl_grmode; | |
401 | int8_t rl_status; | |
402 | int8_t rl_asts; | |
403 | uint16_t rl_wait_type; | |
404 | uint16_t rl_namelen; | |
e7fd4179 DT |
405 | char rl_name[DLM_RESNAME_MAXLEN]; |
406 | char rl_lvb[0]; | |
407 | }; | |
408 | ||
409 | struct dlm_ls { | |
410 | struct list_head ls_list; /* list of lockspaces */ | |
411 | uint32_t ls_global_id; /* global unique lockspace ID */ | |
412 | uint32_t ls_exflags; | |
413 | int ls_lvblen; | |
414 | int ls_count; /* reference count */ | |
415 | unsigned long ls_flags; /* LSFL_ */ | |
416 | struct kobject ls_kobj; | |
417 | ||
418 | struct dlm_rsbtable *ls_rsbtbl; | |
419 | uint32_t ls_rsbtbl_size; | |
420 | ||
421 | struct dlm_lkbtable *ls_lkbtbl; | |
422 | uint32_t ls_lkbtbl_size; | |
423 | ||
424 | struct dlm_dirtable *ls_dirtbl; | |
425 | uint32_t ls_dirtbl_size; | |
426 | ||
90135925 | 427 | struct mutex ls_waiters_mutex; |
e7fd4179 DT |
428 | struct list_head ls_waiters; /* lkbs needing a reply */ |
429 | ||
430 | struct list_head ls_nodes; /* current nodes in ls */ | |
431 | struct list_head ls_nodes_gone; /* dead node list, recovery */ | |
432 | int ls_num_nodes; /* number of nodes in ls */ | |
433 | int ls_low_nodeid; | |
434 | int ls_total_weight; | |
435 | int *ls_node_array; | |
436 | ||
437 | struct dlm_rsb ls_stub_rsb; /* for returning errors */ | |
438 | struct dlm_lkb ls_stub_lkb; /* for returning errors */ | |
439 | struct dlm_message ls_stub_ms; /* for faking a reply */ | |
440 | ||
441 | struct dentry *ls_debug_dentry; /* debugfs */ | |
442 | ||
443 | wait_queue_head_t ls_uevent_wait; /* user part of join/leave */ | |
444 | int ls_uevent_result; | |
445 | ||
446 | /* recovery related */ | |
447 | ||
448 | struct timer_list ls_timer; | |
449 | struct task_struct *ls_recoverd_task; | |
90135925 | 450 | struct mutex ls_recoverd_active; |
e7fd4179 DT |
451 | spinlock_t ls_recover_lock; |
452 | uint32_t ls_recover_status; /* DLM_RS_ */ | |
453 | uint64_t ls_recover_seq; | |
454 | struct dlm_recover *ls_recover_args; | |
455 | struct rw_semaphore ls_in_recovery; /* block local requests */ | |
456 | struct list_head ls_requestqueue;/* queue remote requests */ | |
90135925 | 457 | struct mutex ls_requestqueue_mutex; |
e7fd4179 DT |
458 | char *ls_recover_buf; |
459 | struct list_head ls_recover_list; | |
460 | spinlock_t ls_recover_list_lock; | |
461 | int ls_recover_list_count; | |
462 | wait_queue_head_t ls_wait_general; | |
463 | ||
464 | struct list_head ls_root_list; /* root resources */ | |
465 | struct rw_semaphore ls_root_sem; /* protect root_list */ | |
466 | ||
467 | int ls_namelen; | |
468 | char ls_name[1]; | |
469 | }; | |
470 | ||
471 | #define LSFL_WORK 0 | |
472 | #define LSFL_RUNNING 1 | |
473 | #define LSFL_RECOVERY_STOP 2 | |
474 | #define LSFL_RCOM_READY 3 | |
475 | #define LSFL_UEVENT_WAIT 4 | |
476 | ||
477 | static inline int dlm_locking_stopped(struct dlm_ls *ls) | |
478 | { | |
479 | return !test_bit(LSFL_RUNNING, &ls->ls_flags); | |
480 | } | |
481 | ||
482 | static inline int dlm_recovery_stopped(struct dlm_ls *ls) | |
483 | { | |
484 | return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); | |
485 | } | |
486 | ||
487 | static inline int dlm_no_directory(struct dlm_ls *ls) | |
488 | { | |
489 | return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0; | |
490 | } | |
491 | ||
492 | #endif /* __DLM_INTERNAL_DOT_H__ */ | |
493 |