1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5 ** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved.
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 <linux/idr.h>
41 #include <linux/ratelimit.h>
42 #include <asm/uaccess.h>
44 #include <linux/dlm.h>
47 /* Size of the temp buffer midcomms allocates on the stack.
48 We try to make this large enough so most messages fit.
49 FIXME: should sctp make this unnecessary? */
51 #define DLM_INBUF_LEN 148
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)
71 #define log_debug(ls, fmt, args...) \
73 if (dlm_config.ci_log_debug) \
74 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
75 (ls)->ls_name , ##args); \
78 #define log_limit(ls, fmt, args...) \
80 if (dlm_config.ci_log_debug) \
81 printk_ratelimited(KERN_DEBUG "dlm: %s: " fmt "\n", \
82 (ls)->ls_name , ##args); \
85 #define DLM_ASSERT(x, do) \
89 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \
90 "DLM: assertion: \"%s\"\n" \
91 "DLM: time = %lu\n", \
92 __LINE__, __FILE__, #x, jiffies); \
96 panic("DLM: Record message above and reboot.\n"); \
101 struct dlm_direntry {
102 struct list_head list;
103 uint32_t master_nodeid;
108 struct dlm_dirtable {
109 struct list_head list;
113 struct dlm_rsbtable {
121 * Lockspace member (per node in a ls)
125 struct list_head list;
135 * Save and manage recovery state for a lockspace.
139 struct list_head list;
140 struct dlm_config_node *nodes;
146 * Pass input args to second stage locking function.
151 void (*astfn) (void *astparam);
153 void (*bastfn) (void *astparam, int mode);
155 struct dlm_lksb *lksb;
156 unsigned long timeout;
163 * A lock can be one of three types:
165 * local copy lock is mastered locally
166 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
167 * process copy lock is mastered on a remote node
168 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
169 * master copy master node's copy of a lock owned by remote node
170 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
172 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
173 * dlm_unlock. The dlm does not modify these or use any private flags in
174 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags
175 * are sent as-is to the remote master when the lock is remote.
177 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
178 * Some internal flags are shared between the master and process nodes;
179 * these shared flags are kept in the lower two bytes. One of these
180 * flags set on the master copy will be propagated to the process copy
181 * and v.v. Other internal flags are private to the master or process
182 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes.
184 * lkb_sbflags: status block flags. These flags are copied directly into
185 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
186 * ast. All defined in dlm.h with DLM_SBF_ prefix.
188 * lkb_status: the lock status indicates which rsb queue the lock is
189 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT
191 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
192 * reply is needed. Only set when the lkb is on the lockspace waiters
193 * list awaiting a reply from a remote node.
195 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
196 * is a master copy, nodeid specifies the remote lock holder, when the
197 * lkb is a process copy, the nodeid specifies the lock master.
202 #define DLM_LKSTS_WAITING 1
203 #define DLM_LKSTS_GRANTED 2
204 #define DLM_LKSTS_CONVERT 3
208 #define DLM_IFL_MSTCPY 0x00010000
209 #define DLM_IFL_RESEND 0x00020000
210 #define DLM_IFL_DEAD 0x00040000
211 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000
212 #define DLM_IFL_OVERLAP_CANCEL 0x00100000
213 #define DLM_IFL_ENDOFLIFE 0x00200000
214 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
215 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
216 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000
217 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */
218 #define DLM_IFL_USER 0x00000001
219 #define DLM_IFL_ORPHAN 0x00000002
221 #define DLM_CALLBACKS_SIZE 6
223 #define DLM_CB_CAST 0x00000001
224 #define DLM_CB_BAST 0x00000002
225 #define DLM_CB_SKIP 0x00000004
227 struct dlm_callback {
229 uint32_t flags; /* DLM_CBF_ */
230 int sb_status; /* copy to lksb status */
231 uint8_t sb_flags; /* copy to lksb flags */
232 int8_t mode; /* rq mode of bast, gr mode of cast */
236 struct dlm_rsb *lkb_resource; /* the rsb */
238 int lkb_nodeid; /* copied from rsb */
239 int lkb_ownpid; /* pid of lock owner */
240 uint32_t lkb_id; /* our lock ID */
241 uint32_t lkb_remid; /* lock ID on remote partner */
242 uint32_t lkb_exflags; /* external flags from caller */
243 uint32_t lkb_sbflags; /* lksb flags */
244 uint32_t lkb_flags; /* internal flags */
245 uint32_t lkb_lvbseq; /* lvb sequence number */
247 int8_t lkb_status; /* granted, waiting, convert */
248 int8_t lkb_rqmode; /* requested lock mode */
249 int8_t lkb_grmode; /* granted lock mode */
250 int8_t lkb_highbast; /* highest mode bast sent for */
252 int8_t lkb_wait_type; /* type of reply waiting for */
253 int8_t lkb_wait_count;
254 int lkb_wait_nodeid; /* for debugging */
256 struct list_head lkb_statequeue; /* rsb g/c/w list */
257 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */
258 struct list_head lkb_wait_reply; /* waiting for remote reply */
259 struct list_head lkb_ownqueue; /* list of locks for a process */
260 struct list_head lkb_time_list;
261 ktime_t lkb_timestamp;
262 ktime_t lkb_wait_time;
263 unsigned long lkb_timeout_cs;
265 struct mutex lkb_cb_mutex;
266 struct work_struct lkb_cb_work;
267 struct list_head lkb_cb_list; /* for ls_cb_delay or proc->asts */
268 struct dlm_callback lkb_callbacks[DLM_CALLBACKS_SIZE];
269 struct dlm_callback lkb_last_cast;
270 struct dlm_callback lkb_last_bast;
271 ktime_t lkb_last_cast_time; /* for debugging */
272 ktime_t lkb_last_bast_time; /* for debugging */
274 uint64_t lkb_recover_seq; /* from ls_recover_seq */
277 struct dlm_lksb *lkb_lksb; /* caller's status block */
278 void (*lkb_astfn) (void *astparam);
279 void (*lkb_bastfn) (void *astparam, int mode);
281 void *lkb_astparam; /* caller's ast arg */
282 struct dlm_user_args *lkb_ua;
288 struct dlm_ls *res_ls; /* the lockspace */
290 struct mutex res_mutex;
291 unsigned long res_flags;
292 int res_length; /* length of rsb name */
296 uint32_t res_bucket; /* rsbtbl */
297 unsigned long res_toss_time;
298 uint32_t res_first_lkid;
299 struct list_head res_lookup; /* lkbs waiting on first */
301 struct list_head res_hashchain;
302 struct rb_node res_hashnode; /* rsbtbl */
304 struct list_head res_grantqueue;
305 struct list_head res_convertqueue;
306 struct list_head res_waitqueue;
308 struct list_head res_root_list; /* used for recovery */
309 struct list_head res_recover_list; /* used for recovery */
310 int res_recover_locks_count;
313 char res_name[DLM_RESNAME_MAXLEN+1];
316 /* find_rsb() flags */
318 #define R_MASTER 1 /* only return rsb if it's a master */
319 #define R_CREATE 2 /* create/add rsb if not found */
324 RSB_MASTER_UNCERTAIN,
326 RSB_VALNOTVALID_PREV,
333 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
335 __set_bit(flag, &r->res_flags);
338 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
340 __clear_bit(flag, &r->res_flags);
343 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
345 return test_bit(flag, &r->res_flags);
349 /* dlm_header is first element of all structs sent between nodes */
351 #define DLM_HEADER_MAJOR 0x00030000
352 #define DLM_HEADER_MINOR 0x00000001
354 #define DLM_HEADER_SLOTS 0x00000001
361 uint32_t h_lockspace;
362 uint32_t h_nodeid; /* nodeid of sender */
364 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */
369 #define DLM_MSG_REQUEST 1
370 #define DLM_MSG_CONVERT 2
371 #define DLM_MSG_UNLOCK 3
372 #define DLM_MSG_CANCEL 4
373 #define DLM_MSG_REQUEST_REPLY 5
374 #define DLM_MSG_CONVERT_REPLY 6
375 #define DLM_MSG_UNLOCK_REPLY 7
376 #define DLM_MSG_CANCEL_REPLY 8
377 #define DLM_MSG_GRANT 9
378 #define DLM_MSG_BAST 10
379 #define DLM_MSG_LOOKUP 11
380 #define DLM_MSG_REMOVE 12
381 #define DLM_MSG_LOOKUP_REPLY 13
382 #define DLM_MSG_PURGE 14
385 struct dlm_header m_header;
386 uint32_t m_type; /* DLM_MSG_ */
389 uint32_t m_lkid; /* lkid on sender */
390 uint32_t m_remid; /* lkid on receiver */
391 uint32_t m_parent_lkid;
392 uint32_t m_parent_remid;
403 int m_result; /* 0 or -EXXX */
404 char m_extra[0]; /* name or lvb */
408 #define DLM_RS_NODES 0x00000001
409 #define DLM_RS_NODES_ALL 0x00000002
410 #define DLM_RS_DIR 0x00000004
411 #define DLM_RS_DIR_ALL 0x00000008
412 #define DLM_RS_LOCKS 0x00000010
413 #define DLM_RS_LOCKS_ALL 0x00000020
414 #define DLM_RS_DONE 0x00000040
415 #define DLM_RS_DONE_ALL 0x00000080
417 #define DLM_RCOM_STATUS 1
418 #define DLM_RCOM_NAMES 2
419 #define DLM_RCOM_LOOKUP 3
420 #define DLM_RCOM_LOCK 4
421 #define DLM_RCOM_STATUS_REPLY 5
422 #define DLM_RCOM_NAMES_REPLY 6
423 #define DLM_RCOM_LOOKUP_REPLY 7
424 #define DLM_RCOM_LOCK_REPLY 8
427 struct dlm_header rc_header;
428 uint32_t rc_type; /* DLM_RCOM_ */
429 int rc_result; /* multi-purpose */
430 uint64_t rc_id; /* match reply with request */
431 uint64_t rc_seq; /* sender's ls_recover_seq */
432 uint64_t rc_seq_reply; /* remote ls_recover_seq */
437 struct dlm_header header; /* common to other two */
438 struct dlm_message message;
439 struct dlm_rcom rcom;
442 #define DLM_RSF_NEED_SLOTS 0x00000001
444 /* RCOM_STATUS data */
451 /* RCOM_STATUS_REPLY data */
456 /* DLM_HEADER_SLOTS adds: */
460 __le32 rf_generation;
476 __le32 rl_parent_lkid;
477 __le32 rl_parent_remid;
488 char rl_name[DLM_RESNAME_MAXLEN];
493 struct list_head ls_list; /* list of lockspaces */
494 dlm_lockspace_t *ls_local_handle;
495 uint32_t ls_global_id; /* global unique lockspace ID */
496 uint32_t ls_generation;
499 int ls_count; /* refcount of processes in
500 the dlm using this ls */
501 int ls_create_count; /* create/release refcount */
502 unsigned long ls_flags; /* LSFL_ */
503 unsigned long ls_scan_time;
504 struct kobject ls_kobj;
506 struct idr ls_lkbidr;
507 spinlock_t ls_lkbidr_spin;
509 struct dlm_rsbtable *ls_rsbtbl;
510 uint32_t ls_rsbtbl_size;
512 struct dlm_dirtable *ls_dirtbl;
513 uint32_t ls_dirtbl_size;
515 struct mutex ls_waiters_mutex;
516 struct list_head ls_waiters; /* lkbs needing a reply */
518 struct mutex ls_orphans_mutex;
519 struct list_head ls_orphans;
521 struct mutex ls_timeout_mutex;
522 struct list_head ls_timeout;
524 spinlock_t ls_new_rsb_spin;
525 int ls_new_rsb_count;
526 struct list_head ls_new_rsb; /* new rsb structs */
528 struct list_head ls_nodes; /* current nodes in ls */
529 struct list_head ls_nodes_gone; /* dead node list, recovery */
530 int ls_num_nodes; /* number of nodes in ls */
538 struct dlm_slot *ls_slots;
540 struct dlm_rsb ls_stub_rsb; /* for returning errors */
541 struct dlm_lkb ls_stub_lkb; /* for returning errors */
542 struct dlm_message ls_stub_ms; /* for faking a reply */
544 struct dentry *ls_debug_rsb_dentry; /* debugfs */
545 struct dentry *ls_debug_waiters_dentry; /* debugfs */
546 struct dentry *ls_debug_locks_dentry; /* debugfs */
547 struct dentry *ls_debug_all_dentry; /* debugfs */
549 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */
550 int ls_uevent_result;
551 struct completion ls_members_done;
552 int ls_members_result;
554 struct miscdevice ls_device;
556 struct workqueue_struct *ls_callback_wq;
558 /* recovery related */
560 struct mutex ls_cb_mutex;
561 struct list_head ls_cb_delay; /* save for queue_work later */
562 struct timer_list ls_timer;
563 struct task_struct *ls_recoverd_task;
564 struct mutex ls_recoverd_active;
565 spinlock_t ls_recover_lock;
566 unsigned long ls_recover_begin; /* jiffies timestamp */
567 uint32_t ls_recover_status; /* DLM_RS_ */
568 uint64_t ls_recover_seq;
569 struct dlm_recover *ls_recover_args;
570 struct rw_semaphore ls_in_recovery; /* block local requests */
571 struct rw_semaphore ls_recv_active; /* block dlm_recv */
572 struct list_head ls_requestqueue;/* queue remote requests */
573 struct mutex ls_requestqueue_mutex;
574 struct dlm_rcom *ls_recover_buf;
575 int ls_recover_nodeid; /* for debugging */
576 unsigned int ls_recover_locks_in; /* for log info */
577 uint64_t ls_rcom_seq;
578 spinlock_t ls_rcom_spin;
579 struct list_head ls_recover_list;
580 spinlock_t ls_recover_list_lock;
581 int ls_recover_list_count;
582 wait_queue_head_t ls_wait_general;
583 struct mutex ls_clear_proc_locks;
585 struct list_head ls_root_list; /* root resources */
586 struct rw_semaphore ls_root_sem; /* protect root_list */
588 const struct dlm_lockspace_ops *ls_ops;
596 #define LSFL_RUNNING 1
597 #define LSFL_RECOVERY_STOP 2
598 #define LSFL_RCOM_READY 3
599 #define LSFL_RCOM_WAIT 4
600 #define LSFL_UEVENT_WAIT 5
601 #define LSFL_TIMEWARN 6
602 #define LSFL_CB_DELAY 7
605 /* much of this is just saving user space pointers associated with the
606 lock that we pass back to the user lib with an ast */
608 struct dlm_user_args {
609 struct dlm_user_proc *proc; /* each process that opens the lockspace
610 device has private data
611 (dlm_user_proc) on the struct file,
612 the process's locks point back to it*/
613 struct dlm_lksb lksb;
614 struct dlm_lksb __user *user_lksb;
615 void __user *castparam;
616 void __user *castaddr;
617 void __user *bastparam;
618 void __user *bastaddr;
622 #define DLM_PROC_FLAGS_CLOSING 1
623 #define DLM_PROC_FLAGS_COMPAT 2
625 /* locks list is kept so we can remove all a process's locks when it
626 exits (or orphan those that are persistent) */
628 struct dlm_user_proc {
629 dlm_lockspace_t *lockspace;
630 unsigned long flags; /* DLM_PROC_FLAGS */
631 struct list_head asts;
632 spinlock_t asts_spin;
633 struct list_head locks;
634 spinlock_t locks_spin;
635 struct list_head unlocking;
636 wait_queue_head_t wait;
639 static inline int dlm_locking_stopped(struct dlm_ls *ls)
641 return !test_bit(LSFL_RUNNING, &ls->ls_flags);
644 static inline int dlm_recovery_stopped(struct dlm_ls *ls)
646 return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
649 static inline int dlm_no_directory(struct dlm_ls *ls)
651 return test_bit(LSFL_NODIR, &ls->ls_flags);
654 int dlm_netlink_init(void);
655 void dlm_netlink_exit(void);
656 void dlm_timeout_warn(struct dlm_lkb *lkb);
657 int dlm_plock_init(void);
658 void dlm_plock_exit(void);
660 #ifdef CONFIG_DLM_DEBUG
661 int dlm_register_debugfs(void);
662 void dlm_unregister_debugfs(void);
663 int dlm_create_debug_file(struct dlm_ls *ls);
664 void dlm_delete_debug_file(struct dlm_ls *ls);
666 static inline int dlm_register_debugfs(void) { return 0; }
667 static inline void dlm_unregister_debugfs(void) { }
668 static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; }
669 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { }
672 #endif /* __DLM_INTERNAL_DOT_H__ */