kvm tools, setup: Create private directory
[linux-2.6/next.git] / fs / dlm / dlm_internal.h
blob0262451eb9c6e90cf77d71fb955fff7f60c44668
1 /******************************************************************************
2 *******************************************************************************
3 **
4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5 ** Copyright (C) 2004-2010 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.
11 *******************************************************************************
12 ******************************************************************************/
14 #ifndef __DLM_INTERNAL_DOT_H__
15 #define __DLM_INTERNAL_DOT_H__
18 * This is the main header file to be included in each DLM source file.
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>
38 #include <linux/miscdevice.h>
39 #include <linux/mutex.h>
40 #include <asm/uaccess.h>
42 #include <linux/dlm.h>
43 #include "config.h"
45 /* Size of the temp buffer midcomms allocates on the stack.
46 We try to make this large enough so most messages fit.
47 FIXME: should sctp make this unnecessary? */
49 #define DLM_INBUF_LEN 148
51 struct dlm_ls;
52 struct dlm_lkb;
53 struct dlm_rsb;
54 struct dlm_member;
55 struct dlm_lkbtable;
56 struct dlm_rsbtable;
57 struct dlm_dirtable;
58 struct dlm_direntry;
59 struct dlm_recover;
60 struct dlm_header;
61 struct dlm_message;
62 struct dlm_rcom;
63 struct dlm_mhandle;
65 #define log_print(fmt, args...) \
66 printk(KERN_ERR "dlm: "fmt"\n" , ##args)
67 #define log_error(ls, fmt, args...) \
68 printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args)
70 #define log_debug(ls, fmt, args...) \
71 do { \
72 if (dlm_config.ci_log_debug) \
73 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
74 (ls)->ls_name , ##args); \
75 } while (0)
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 } \
93 struct dlm_direntry {
94 struct list_head list;
95 uint32_t master_nodeid;
96 uint16_t length;
97 char name[1];
100 struct dlm_dirtable {
101 struct list_head list;
102 spinlock_t lock;
105 struct dlm_rsbtable {
106 struct list_head list;
107 struct list_head toss;
108 spinlock_t lock;
111 struct dlm_lkbtable {
112 struct list_head list;
113 rwlock_t lock;
114 uint16_t counter;
118 * Lockspace member (per node in a ls)
121 struct dlm_member {
122 struct list_head list;
123 int nodeid;
124 int weight;
128 * Save and manage recovery state for a lockspace.
131 struct dlm_recover {
132 struct list_head list;
133 int *nodeids; /* nodeids of all members */
134 int node_count;
135 int *new; /* nodeids of new members */
136 int new_count;
137 uint64_t seq;
141 * Pass input args to second stage locking function.
144 struct dlm_args {
145 uint32_t flags;
146 void (*astfn) (void *astparam);
147 void *astparam;
148 void (*bastfn) (void *astparam, int mode);
149 int mode;
150 struct dlm_lksb *lksb;
151 unsigned long timeout;
156 * Lock block
158 * A lock can be one of three types:
160 * local copy lock is mastered locally
161 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
162 * process copy lock is mastered on a remote node
163 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
164 * master copy master node's copy of a lock owned by remote node
165 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
167 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
168 * dlm_unlock. The dlm does not modify these or use any private flags in
169 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags
170 * are sent as-is to the remote master when the lock is remote.
172 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
173 * Some internal flags are shared between the master and process nodes;
174 * these shared flags are kept in the lower two bytes. One of these
175 * flags set on the master copy will be propagated to the process copy
176 * and v.v. Other internal flags are private to the master or process
177 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes.
179 * lkb_sbflags: status block flags. These flags are copied directly into
180 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
181 * ast. All defined in dlm.h with DLM_SBF_ prefix.
183 * lkb_status: the lock status indicates which rsb queue the lock is
184 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT
186 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
187 * reply is needed. Only set when the lkb is on the lockspace waiters
188 * list awaiting a reply from a remote node.
190 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
191 * is a master copy, nodeid specifies the remote lock holder, when the
192 * lkb is a process copy, the nodeid specifies the lock master.
195 /* lkb_status */
197 #define DLM_LKSTS_WAITING 1
198 #define DLM_LKSTS_GRANTED 2
199 #define DLM_LKSTS_CONVERT 3
201 /* lkb_flags */
203 #define DLM_IFL_MSTCPY 0x00010000
204 #define DLM_IFL_RESEND 0x00020000
205 #define DLM_IFL_DEAD 0x00040000
206 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000
207 #define DLM_IFL_OVERLAP_CANCEL 0x00100000
208 #define DLM_IFL_ENDOFLIFE 0x00200000
209 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
210 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
211 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000
212 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */
213 #define DLM_IFL_USER 0x00000001
214 #define DLM_IFL_ORPHAN 0x00000002
216 #define DLM_CALLBACKS_SIZE 6
218 #define DLM_CB_CAST 0x00000001
219 #define DLM_CB_BAST 0x00000002
220 #define DLM_CB_SKIP 0x00000004
222 struct dlm_callback {
223 uint64_t seq;
224 uint32_t flags; /* DLM_CBF_ */
225 int sb_status; /* copy to lksb status */
226 uint8_t sb_flags; /* copy to lksb flags */
227 int8_t mode; /* rq mode of bast, gr mode of cast */
230 struct dlm_lkb {
231 struct dlm_rsb *lkb_resource; /* the rsb */
232 struct kref lkb_ref;
233 int lkb_nodeid; /* copied from rsb */
234 int lkb_ownpid; /* pid of lock owner */
235 uint32_t lkb_id; /* our lock ID */
236 uint32_t lkb_remid; /* lock ID on remote partner */
237 uint32_t lkb_exflags; /* external flags from caller */
238 uint32_t lkb_sbflags; /* lksb flags */
239 uint32_t lkb_flags; /* internal flags */
240 uint32_t lkb_lvbseq; /* lvb sequence number */
242 int8_t lkb_status; /* granted, waiting, convert */
243 int8_t lkb_rqmode; /* requested lock mode */
244 int8_t lkb_grmode; /* granted lock mode */
245 int8_t lkb_highbast; /* highest mode bast sent for */
247 int8_t lkb_wait_type; /* type of reply waiting for */
248 int8_t lkb_wait_count;
249 int lkb_wait_nodeid; /* for debugging */
251 struct list_head lkb_idtbl_list; /* lockspace lkbtbl */
252 struct list_head lkb_statequeue; /* rsb g/c/w list */
253 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */
254 struct list_head lkb_wait_reply; /* waiting for remote reply */
255 struct list_head lkb_astqueue; /* need ast to be sent */
256 struct list_head lkb_ownqueue; /* list of locks for a process */
257 struct list_head lkb_time_list;
258 ktime_t lkb_timestamp;
259 ktime_t lkb_wait_time;
260 unsigned long lkb_timeout_cs;
262 struct dlm_callback lkb_callbacks[DLM_CALLBACKS_SIZE];
263 struct dlm_callback lkb_last_cast;
264 struct dlm_callback lkb_last_bast;
265 ktime_t lkb_last_cast_time; /* for debugging */
266 ktime_t lkb_last_bast_time; /* for debugging */
268 char *lkb_lvbptr;
269 struct dlm_lksb *lkb_lksb; /* caller's status block */
270 void (*lkb_astfn) (void *astparam);
271 void (*lkb_bastfn) (void *astparam, int mode);
272 union {
273 void *lkb_astparam; /* caller's ast arg */
274 struct dlm_user_args *lkb_ua;
279 struct dlm_rsb {
280 struct dlm_ls *res_ls; /* the lockspace */
281 struct kref res_ref;
282 struct mutex res_mutex;
283 unsigned long res_flags;
284 int res_length; /* length of rsb name */
285 int res_nodeid;
286 uint32_t res_lvbseq;
287 uint32_t res_hash;
288 uint32_t res_bucket; /* rsbtbl */
289 unsigned long res_toss_time;
290 uint32_t res_first_lkid;
291 struct list_head res_lookup; /* lkbs waiting on first */
292 struct list_head res_hashchain; /* rsbtbl */
293 struct list_head res_grantqueue;
294 struct list_head res_convertqueue;
295 struct list_head res_waitqueue;
297 struct list_head res_root_list; /* used for recovery */
298 struct list_head res_recover_list; /* used for recovery */
299 int res_recover_locks_count;
301 char *res_lvbptr;
302 char res_name[1];
305 /* find_rsb() flags */
307 #define R_MASTER 1 /* only return rsb if it's a master */
308 #define R_CREATE 2 /* create/add rsb if not found */
310 /* rsb_flags */
312 enum rsb_flags {
313 RSB_MASTER_UNCERTAIN,
314 RSB_VALNOTVALID,
315 RSB_VALNOTVALID_PREV,
316 RSB_NEW_MASTER,
317 RSB_NEW_MASTER2,
318 RSB_RECOVER_CONVERT,
319 RSB_LOCKS_PURGED,
322 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
324 __set_bit(flag, &r->res_flags);
327 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
329 __clear_bit(flag, &r->res_flags);
332 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
334 return test_bit(flag, &r->res_flags);
338 /* dlm_header is first element of all structs sent between nodes */
340 #define DLM_HEADER_MAJOR 0x00030000
341 #define DLM_HEADER_MINOR 0x00000000
343 #define DLM_MSG 1
344 #define DLM_RCOM 2
346 struct dlm_header {
347 uint32_t h_version;
348 uint32_t h_lockspace;
349 uint32_t h_nodeid; /* nodeid of sender */
350 uint16_t h_length;
351 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */
352 uint8_t h_pad;
356 #define DLM_MSG_REQUEST 1
357 #define DLM_MSG_CONVERT 2
358 #define DLM_MSG_UNLOCK 3
359 #define DLM_MSG_CANCEL 4
360 #define DLM_MSG_REQUEST_REPLY 5
361 #define DLM_MSG_CONVERT_REPLY 6
362 #define DLM_MSG_UNLOCK_REPLY 7
363 #define DLM_MSG_CANCEL_REPLY 8
364 #define DLM_MSG_GRANT 9
365 #define DLM_MSG_BAST 10
366 #define DLM_MSG_LOOKUP 11
367 #define DLM_MSG_REMOVE 12
368 #define DLM_MSG_LOOKUP_REPLY 13
369 #define DLM_MSG_PURGE 14
371 struct dlm_message {
372 struct dlm_header m_header;
373 uint32_t m_type; /* DLM_MSG_ */
374 uint32_t m_nodeid;
375 uint32_t m_pid;
376 uint32_t m_lkid; /* lkid on sender */
377 uint32_t m_remid; /* lkid on receiver */
378 uint32_t m_parent_lkid;
379 uint32_t m_parent_remid;
380 uint32_t m_exflags;
381 uint32_t m_sbflags;
382 uint32_t m_flags;
383 uint32_t m_lvbseq;
384 uint32_t m_hash;
385 int m_status;
386 int m_grmode;
387 int m_rqmode;
388 int m_bastmode;
389 int m_asts;
390 int m_result; /* 0 or -EXXX */
391 char m_extra[0]; /* name or lvb */
395 #define DLM_RS_NODES 0x00000001
396 #define DLM_RS_NODES_ALL 0x00000002
397 #define DLM_RS_DIR 0x00000004
398 #define DLM_RS_DIR_ALL 0x00000008
399 #define DLM_RS_LOCKS 0x00000010
400 #define DLM_RS_LOCKS_ALL 0x00000020
401 #define DLM_RS_DONE 0x00000040
402 #define DLM_RS_DONE_ALL 0x00000080
404 #define DLM_RCOM_STATUS 1
405 #define DLM_RCOM_NAMES 2
406 #define DLM_RCOM_LOOKUP 3
407 #define DLM_RCOM_LOCK 4
408 #define DLM_RCOM_STATUS_REPLY 5
409 #define DLM_RCOM_NAMES_REPLY 6
410 #define DLM_RCOM_LOOKUP_REPLY 7
411 #define DLM_RCOM_LOCK_REPLY 8
413 struct dlm_rcom {
414 struct dlm_header rc_header;
415 uint32_t rc_type; /* DLM_RCOM_ */
416 int rc_result; /* multi-purpose */
417 uint64_t rc_id; /* match reply with request */
418 uint64_t rc_seq; /* sender's ls_recover_seq */
419 uint64_t rc_seq_reply; /* remote ls_recover_seq */
420 char rc_buf[0];
423 union dlm_packet {
424 struct dlm_header header; /* common to other two */
425 struct dlm_message message;
426 struct dlm_rcom rcom;
429 struct rcom_config {
430 __le32 rf_lvblen;
431 __le32 rf_lsflags;
432 __le64 rf_unused;
435 struct rcom_lock {
436 __le32 rl_ownpid;
437 __le32 rl_lkid;
438 __le32 rl_remid;
439 __le32 rl_parent_lkid;
440 __le32 rl_parent_remid;
441 __le32 rl_exflags;
442 __le32 rl_flags;
443 __le32 rl_lvbseq;
444 __le32 rl_result;
445 int8_t rl_rqmode;
446 int8_t rl_grmode;
447 int8_t rl_status;
448 int8_t rl_asts;
449 __le16 rl_wait_type;
450 __le16 rl_namelen;
451 char rl_name[DLM_RESNAME_MAXLEN];
452 char rl_lvb[0];
455 struct dlm_ls {
456 struct list_head ls_list; /* list of lockspaces */
457 dlm_lockspace_t *ls_local_handle;
458 uint32_t ls_global_id; /* global unique lockspace ID */
459 uint32_t ls_exflags;
460 int ls_lvblen;
461 int ls_count; /* refcount of processes in
462 the dlm using this ls */
463 int ls_create_count; /* create/release refcount */
464 unsigned long ls_flags; /* LSFL_ */
465 unsigned long ls_scan_time;
466 struct kobject ls_kobj;
468 struct dlm_rsbtable *ls_rsbtbl;
469 uint32_t ls_rsbtbl_size;
471 struct dlm_lkbtable *ls_lkbtbl;
472 uint32_t ls_lkbtbl_size;
474 struct dlm_dirtable *ls_dirtbl;
475 uint32_t ls_dirtbl_size;
477 struct mutex ls_waiters_mutex;
478 struct list_head ls_waiters; /* lkbs needing a reply */
480 struct mutex ls_orphans_mutex;
481 struct list_head ls_orphans;
483 struct mutex ls_timeout_mutex;
484 struct list_head ls_timeout;
486 struct list_head ls_nodes; /* current nodes in ls */
487 struct list_head ls_nodes_gone; /* dead node list, recovery */
488 int ls_num_nodes; /* number of nodes in ls */
489 int ls_low_nodeid;
490 int ls_total_weight;
491 int *ls_node_array;
493 struct dlm_rsb ls_stub_rsb; /* for returning errors */
494 struct dlm_lkb ls_stub_lkb; /* for returning errors */
495 struct dlm_message ls_stub_ms; /* for faking a reply */
497 struct dentry *ls_debug_rsb_dentry; /* debugfs */
498 struct dentry *ls_debug_waiters_dentry; /* debugfs */
499 struct dentry *ls_debug_locks_dentry; /* debugfs */
500 struct dentry *ls_debug_all_dentry; /* debugfs */
502 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */
503 int ls_uevent_result;
504 struct completion ls_members_done;
505 int ls_members_result;
507 struct miscdevice ls_device;
509 /* recovery related */
511 struct timer_list ls_timer;
512 struct task_struct *ls_recoverd_task;
513 struct mutex ls_recoverd_active;
514 spinlock_t ls_recover_lock;
515 unsigned long ls_recover_begin; /* jiffies timestamp */
516 uint32_t ls_recover_status; /* DLM_RS_ */
517 uint64_t ls_recover_seq;
518 struct dlm_recover *ls_recover_args;
519 struct rw_semaphore ls_in_recovery; /* block local requests */
520 struct rw_semaphore ls_recv_active; /* block dlm_recv */
521 struct list_head ls_requestqueue;/* queue remote requests */
522 struct mutex ls_requestqueue_mutex;
523 struct dlm_rcom *ls_recover_buf;
524 int ls_recover_nodeid; /* for debugging */
525 uint64_t ls_rcom_seq;
526 spinlock_t ls_rcom_spin;
527 struct list_head ls_recover_list;
528 spinlock_t ls_recover_list_lock;
529 int ls_recover_list_count;
530 wait_queue_head_t ls_wait_general;
531 struct mutex ls_clear_proc_locks;
533 struct list_head ls_root_list; /* root resources */
534 struct rw_semaphore ls_root_sem; /* protect root_list */
536 int ls_namelen;
537 char ls_name[1];
540 #define LSFL_WORK 0
541 #define LSFL_RUNNING 1
542 #define LSFL_RECOVERY_STOP 2
543 #define LSFL_RCOM_READY 3
544 #define LSFL_RCOM_WAIT 4
545 #define LSFL_UEVENT_WAIT 5
546 #define LSFL_TIMEWARN 6
548 /* much of this is just saving user space pointers associated with the
549 lock that we pass back to the user lib with an ast */
551 struct dlm_user_args {
552 struct dlm_user_proc *proc; /* each process that opens the lockspace
553 device has private data
554 (dlm_user_proc) on the struct file,
555 the process's locks point back to it*/
556 struct dlm_lksb lksb;
557 struct dlm_lksb __user *user_lksb;
558 void __user *castparam;
559 void __user *castaddr;
560 void __user *bastparam;
561 void __user *bastaddr;
562 uint64_t xid;
565 #define DLM_PROC_FLAGS_CLOSING 1
566 #define DLM_PROC_FLAGS_COMPAT 2
568 /* locks list is kept so we can remove all a process's locks when it
569 exits (or orphan those that are persistent) */
571 struct dlm_user_proc {
572 dlm_lockspace_t *lockspace;
573 unsigned long flags; /* DLM_PROC_FLAGS */
574 struct list_head asts;
575 spinlock_t asts_spin;
576 struct list_head locks;
577 spinlock_t locks_spin;
578 struct list_head unlocking;
579 wait_queue_head_t wait;
582 static inline int dlm_locking_stopped(struct dlm_ls *ls)
584 return !test_bit(LSFL_RUNNING, &ls->ls_flags);
587 static inline int dlm_recovery_stopped(struct dlm_ls *ls)
589 return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
592 static inline int dlm_no_directory(struct dlm_ls *ls)
594 return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0;
597 int dlm_netlink_init(void);
598 void dlm_netlink_exit(void);
599 void dlm_timeout_warn(struct dlm_lkb *lkb);
600 int dlm_plock_init(void);
601 void dlm_plock_exit(void);
603 #ifdef CONFIG_DLM_DEBUG
604 int dlm_register_debugfs(void);
605 void dlm_unregister_debugfs(void);
606 int dlm_create_debug_file(struct dlm_ls *ls);
607 void dlm_delete_debug_file(struct dlm_ls *ls);
608 #else
609 static inline int dlm_register_debugfs(void) { return 0; }
610 static inline void dlm_unregister_debugfs(void) { }
611 static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; }
612 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { }
613 #endif
615 #endif /* __DLM_INTERNAL_DOT_H__ */