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 <asm/uaccess.h>
43 #include <linux/dlm.h>
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? */
50 #define DLM_INBUF_LEN 148
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...) \
72 if (dlm_config.ci_log_debug) \
73 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
74 (ls)->ls_name , ##args); \
77 #define DLM_ASSERT(x, do) \
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); \
88 panic("DLM: Record message above and reboot.\n"); \
94 struct list_head list
;
95 uint32_t master_nodeid
;
100 struct dlm_dirtable
{
101 struct list_head list
;
105 struct dlm_rsbtable
{
113 * Lockspace member (per node in a ls)
117 struct list_head list
;
127 * Save and manage recovery state for a lockspace.
131 struct list_head list
;
132 struct dlm_config_node
*nodes
;
138 * Pass input args to second stage locking function.
143 void (*astfn
) (void *astparam
);
145 void (*bastfn
) (void *astparam
, int mode
);
147 struct dlm_lksb
*lksb
;
148 unsigned long timeout
;
155 * A lock can be one of three types:
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)
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.
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.
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.
180 * lkb_status: the lock status indicates which rsb queue the lock is
181 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT
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.
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.
194 #define DLM_LKSTS_WAITING 1
195 #define DLM_LKSTS_GRANTED 2
196 #define DLM_LKSTS_CONVERT 3
200 #define DLM_IFL_MSTCPY 0x00010000
201 #define DLM_IFL_RESEND 0x00020000
202 #define DLM_IFL_DEAD 0x00040000
203 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000
204 #define DLM_IFL_OVERLAP_CANCEL 0x00100000
205 #define DLM_IFL_ENDOFLIFE 0x00200000
206 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
207 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
208 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000
209 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */
210 #define DLM_IFL_USER 0x00000001
211 #define DLM_IFL_ORPHAN 0x00000002
213 #define DLM_CALLBACKS_SIZE 6
215 #define DLM_CB_CAST 0x00000001
216 #define DLM_CB_BAST 0x00000002
217 #define DLM_CB_SKIP 0x00000004
219 struct dlm_callback
{
221 uint32_t flags
; /* DLM_CBF_ */
222 int sb_status
; /* copy to lksb status */
223 uint8_t sb_flags
; /* copy to lksb flags */
224 int8_t mode
; /* rq mode of bast, gr mode of cast */
228 struct dlm_rsb
*lkb_resource
; /* the rsb */
230 int lkb_nodeid
; /* copied from rsb */
231 int lkb_ownpid
; /* pid of lock owner */
232 uint32_t lkb_id
; /* our lock ID */
233 uint32_t lkb_remid
; /* lock ID on remote partner */
234 uint32_t lkb_exflags
; /* external flags from caller */
235 uint32_t lkb_sbflags
; /* lksb flags */
236 uint32_t lkb_flags
; /* internal flags */
237 uint32_t lkb_lvbseq
; /* lvb sequence number */
239 int8_t lkb_status
; /* granted, waiting, convert */
240 int8_t lkb_rqmode
; /* requested lock mode */
241 int8_t lkb_grmode
; /* granted lock mode */
242 int8_t lkb_highbast
; /* highest mode bast sent for */
244 int8_t lkb_wait_type
; /* type of reply waiting for */
245 int8_t lkb_wait_count
;
246 int lkb_wait_nodeid
; /* for debugging */
248 struct list_head lkb_statequeue
; /* rsb g/c/w list */
249 struct list_head lkb_rsb_lookup
; /* waiting for rsb lookup */
250 struct list_head lkb_wait_reply
; /* waiting for remote reply */
251 struct list_head lkb_ownqueue
; /* list of locks for a process */
252 struct list_head lkb_time_list
;
253 ktime_t lkb_timestamp
;
254 ktime_t lkb_wait_time
;
255 unsigned long lkb_timeout_cs
;
257 struct mutex lkb_cb_mutex
;
258 struct work_struct lkb_cb_work
;
259 struct list_head lkb_cb_list
; /* for ls_cb_delay or proc->asts */
260 struct dlm_callback lkb_callbacks
[DLM_CALLBACKS_SIZE
];
261 struct dlm_callback lkb_last_cast
;
262 struct dlm_callback lkb_last_bast
;
263 ktime_t lkb_last_cast_time
; /* for debugging */
264 ktime_t lkb_last_bast_time
; /* for debugging */
267 struct dlm_lksb
*lkb_lksb
; /* caller's status block */
268 void (*lkb_astfn
) (void *astparam
);
269 void (*lkb_bastfn
) (void *astparam
, int mode
);
271 void *lkb_astparam
; /* caller's ast arg */
272 struct dlm_user_args
*lkb_ua
;
278 struct dlm_ls
*res_ls
; /* the lockspace */
280 struct mutex res_mutex
;
281 unsigned long res_flags
;
282 int res_length
; /* length of rsb name */
286 uint32_t res_bucket
; /* rsbtbl */
287 unsigned long res_toss_time
;
288 uint32_t res_first_lkid
;
289 struct list_head res_lookup
; /* lkbs waiting on first */
291 struct list_head res_hashchain
;
292 struct rb_node res_hashnode
; /* rsbtbl */
294 struct list_head res_grantqueue
;
295 struct list_head res_convertqueue
;
296 struct list_head res_waitqueue
;
298 struct list_head res_root_list
; /* used for recovery */
299 struct list_head res_recover_list
; /* used for recovery */
300 int res_recover_locks_count
;
303 char res_name
[DLM_RESNAME_MAXLEN
+1];
306 /* find_rsb() flags */
308 #define R_MASTER 1 /* only return rsb if it's a master */
309 #define R_CREATE 2 /* create/add rsb if not found */
314 RSB_MASTER_UNCERTAIN
,
316 RSB_VALNOTVALID_PREV
,
323 static inline void rsb_set_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
325 __set_bit(flag
, &r
->res_flags
);
328 static inline void rsb_clear_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
330 __clear_bit(flag
, &r
->res_flags
);
333 static inline int rsb_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
335 return test_bit(flag
, &r
->res_flags
);
339 /* dlm_header is first element of all structs sent between nodes */
341 #define DLM_HEADER_MAJOR 0x00030000
342 #define DLM_HEADER_MINOR 0x00000001
344 #define DLM_HEADER_SLOTS 0x00000001
351 uint32_t h_lockspace
;
352 uint32_t h_nodeid
; /* nodeid of sender */
354 uint8_t h_cmd
; /* DLM_MSG, DLM_RCOM */
359 #define DLM_MSG_REQUEST 1
360 #define DLM_MSG_CONVERT 2
361 #define DLM_MSG_UNLOCK 3
362 #define DLM_MSG_CANCEL 4
363 #define DLM_MSG_REQUEST_REPLY 5
364 #define DLM_MSG_CONVERT_REPLY 6
365 #define DLM_MSG_UNLOCK_REPLY 7
366 #define DLM_MSG_CANCEL_REPLY 8
367 #define DLM_MSG_GRANT 9
368 #define DLM_MSG_BAST 10
369 #define DLM_MSG_LOOKUP 11
370 #define DLM_MSG_REMOVE 12
371 #define DLM_MSG_LOOKUP_REPLY 13
372 #define DLM_MSG_PURGE 14
375 struct dlm_header m_header
;
376 uint32_t m_type
; /* DLM_MSG_ */
379 uint32_t m_lkid
; /* lkid on sender */
380 uint32_t m_remid
; /* lkid on receiver */
381 uint32_t m_parent_lkid
;
382 uint32_t m_parent_remid
;
393 int m_result
; /* 0 or -EXXX */
394 char m_extra
[0]; /* name or lvb */
398 #define DLM_RS_NODES 0x00000001
399 #define DLM_RS_NODES_ALL 0x00000002
400 #define DLM_RS_DIR 0x00000004
401 #define DLM_RS_DIR_ALL 0x00000008
402 #define DLM_RS_LOCKS 0x00000010
403 #define DLM_RS_LOCKS_ALL 0x00000020
404 #define DLM_RS_DONE 0x00000040
405 #define DLM_RS_DONE_ALL 0x00000080
407 #define DLM_RCOM_STATUS 1
408 #define DLM_RCOM_NAMES 2
409 #define DLM_RCOM_LOOKUP 3
410 #define DLM_RCOM_LOCK 4
411 #define DLM_RCOM_STATUS_REPLY 5
412 #define DLM_RCOM_NAMES_REPLY 6
413 #define DLM_RCOM_LOOKUP_REPLY 7
414 #define DLM_RCOM_LOCK_REPLY 8
417 struct dlm_header rc_header
;
418 uint32_t rc_type
; /* DLM_RCOM_ */
419 int rc_result
; /* multi-purpose */
420 uint64_t rc_id
; /* match reply with request */
421 uint64_t rc_seq
; /* sender's ls_recover_seq */
422 uint64_t rc_seq_reply
; /* remote ls_recover_seq */
427 struct dlm_header header
; /* common to other two */
428 struct dlm_message message
;
429 struct dlm_rcom rcom
;
432 #define DLM_RSF_NEED_SLOTS 0x00000001
434 /* RCOM_STATUS data */
441 /* RCOM_STATUS_REPLY data */
446 /* DLM_HEADER_SLOTS adds: */
450 __le32 rf_generation
;
466 __le32 rl_parent_lkid
;
467 __le32 rl_parent_remid
;
478 char rl_name
[DLM_RESNAME_MAXLEN
];
483 struct list_head ls_list
; /* list of lockspaces */
484 dlm_lockspace_t
*ls_local_handle
;
485 uint32_t ls_global_id
; /* global unique lockspace ID */
486 uint32_t ls_generation
;
489 int ls_count
; /* refcount of processes in
490 the dlm using this ls */
491 int ls_create_count
; /* create/release refcount */
492 unsigned long ls_flags
; /* LSFL_ */
493 unsigned long ls_scan_time
;
494 struct kobject ls_kobj
;
496 struct idr ls_lkbidr
;
497 spinlock_t ls_lkbidr_spin
;
499 struct dlm_rsbtable
*ls_rsbtbl
;
500 uint32_t ls_rsbtbl_size
;
502 struct dlm_dirtable
*ls_dirtbl
;
503 uint32_t ls_dirtbl_size
;
505 struct mutex ls_waiters_mutex
;
506 struct list_head ls_waiters
; /* lkbs needing a reply */
508 struct mutex ls_orphans_mutex
;
509 struct list_head ls_orphans
;
511 struct mutex ls_timeout_mutex
;
512 struct list_head ls_timeout
;
514 spinlock_t ls_new_rsb_spin
;
515 int ls_new_rsb_count
;
516 struct list_head ls_new_rsb
; /* new rsb structs */
518 struct list_head ls_nodes
; /* current nodes in ls */
519 struct list_head ls_nodes_gone
; /* dead node list, recovery */
520 int ls_num_nodes
; /* number of nodes in ls */
528 struct dlm_slot
*ls_slots
;
530 struct dlm_rsb ls_stub_rsb
; /* for returning errors */
531 struct dlm_lkb ls_stub_lkb
; /* for returning errors */
532 struct dlm_message ls_stub_ms
; /* for faking a reply */
534 struct dentry
*ls_debug_rsb_dentry
; /* debugfs */
535 struct dentry
*ls_debug_waiters_dentry
; /* debugfs */
536 struct dentry
*ls_debug_locks_dentry
; /* debugfs */
537 struct dentry
*ls_debug_all_dentry
; /* debugfs */
539 wait_queue_head_t ls_uevent_wait
; /* user part of join/leave */
540 int ls_uevent_result
;
541 struct completion ls_members_done
;
542 int ls_members_result
;
544 struct miscdevice ls_device
;
546 struct workqueue_struct
*ls_callback_wq
;
548 /* recovery related */
550 struct mutex ls_cb_mutex
;
551 struct list_head ls_cb_delay
; /* save for queue_work later */
552 struct timer_list ls_timer
;
553 struct task_struct
*ls_recoverd_task
;
554 struct mutex ls_recoverd_active
;
555 spinlock_t ls_recover_lock
;
556 unsigned long ls_recover_begin
; /* jiffies timestamp */
557 uint32_t ls_recover_status
; /* DLM_RS_ */
558 uint64_t ls_recover_seq
;
559 struct dlm_recover
*ls_recover_args
;
560 struct rw_semaphore ls_in_recovery
; /* block local requests */
561 struct rw_semaphore ls_recv_active
; /* block dlm_recv */
562 struct list_head ls_requestqueue
;/* queue remote requests */
563 struct mutex ls_requestqueue_mutex
;
564 struct dlm_rcom
*ls_recover_buf
;
565 int ls_recover_nodeid
; /* for debugging */
566 uint64_t ls_rcom_seq
;
567 spinlock_t ls_rcom_spin
;
568 struct list_head ls_recover_list
;
569 spinlock_t ls_recover_list_lock
;
570 int ls_recover_list_count
;
571 wait_queue_head_t ls_wait_general
;
572 struct mutex ls_clear_proc_locks
;
574 struct list_head ls_root_list
; /* root resources */
575 struct rw_semaphore ls_root_sem
; /* protect root_list */
577 const struct dlm_lockspace_ops
*ls_ops
;
585 #define LSFL_RUNNING 1
586 #define LSFL_RECOVERY_STOP 2
587 #define LSFL_RCOM_READY 3
588 #define LSFL_RCOM_WAIT 4
589 #define LSFL_UEVENT_WAIT 5
590 #define LSFL_TIMEWARN 6
591 #define LSFL_CB_DELAY 7
593 /* much of this is just saving user space pointers associated with the
594 lock that we pass back to the user lib with an ast */
596 struct dlm_user_args
{
597 struct dlm_user_proc
*proc
; /* each process that opens the lockspace
598 device has private data
599 (dlm_user_proc) on the struct file,
600 the process's locks point back to it*/
601 struct dlm_lksb lksb
;
602 struct dlm_lksb __user
*user_lksb
;
603 void __user
*castparam
;
604 void __user
*castaddr
;
605 void __user
*bastparam
;
606 void __user
*bastaddr
;
610 #define DLM_PROC_FLAGS_CLOSING 1
611 #define DLM_PROC_FLAGS_COMPAT 2
613 /* locks list is kept so we can remove all a process's locks when it
614 exits (or orphan those that are persistent) */
616 struct dlm_user_proc
{
617 dlm_lockspace_t
*lockspace
;
618 unsigned long flags
; /* DLM_PROC_FLAGS */
619 struct list_head asts
;
620 spinlock_t asts_spin
;
621 struct list_head locks
;
622 spinlock_t locks_spin
;
623 struct list_head unlocking
;
624 wait_queue_head_t wait
;
627 static inline int dlm_locking_stopped(struct dlm_ls
*ls
)
629 return !test_bit(LSFL_RUNNING
, &ls
->ls_flags
);
632 static inline int dlm_recovery_stopped(struct dlm_ls
*ls
)
634 return test_bit(LSFL_RECOVERY_STOP
, &ls
->ls_flags
);
637 static inline int dlm_no_directory(struct dlm_ls
*ls
)
639 return (ls
->ls_exflags
& DLM_LSFL_NODIR
) ? 1 : 0;
642 int dlm_netlink_init(void);
643 void dlm_netlink_exit(void);
644 void dlm_timeout_warn(struct dlm_lkb
*lkb
);
645 int dlm_plock_init(void);
646 void dlm_plock_exit(void);
648 #ifdef CONFIG_DLM_DEBUG
649 int dlm_register_debugfs(void);
650 void dlm_unregister_debugfs(void);
651 int dlm_create_debug_file(struct dlm_ls
*ls
);
652 void dlm_delete_debug_file(struct dlm_ls
*ls
);
654 static inline int dlm_register_debugfs(void) { return 0; }
655 static inline void dlm_unregister_debugfs(void) { }
656 static inline int dlm_create_debug_file(struct dlm_ls
*ls
) { return 0; }
657 static inline void dlm_delete_debug_file(struct dlm_ls
*ls
) { }
660 #endif /* __DLM_INTERNAL_DOT_H__ */