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
64 #define log_print(fmt, args...) \
65 printk(KERN_ERR "dlm: "fmt"\n" , ##args)
66 #define log_error(ls, fmt, args...) \
67 printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args)
69 #define log_debug(ls, fmt, args...) \
71 if (dlm_config.ci_log_debug) \
72 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
73 (ls)->ls_name , ##args); \
76 #define log_limit(ls, fmt, args...) \
78 if (dlm_config.ci_log_debug) \
79 printk_ratelimited(KERN_DEBUG "dlm: %s: " fmt "\n", \
80 (ls)->ls_name , ##args); \
83 #define DLM_ASSERT(x, do) \
87 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \
88 "DLM: assertion: \"%s\"\n" \
89 "DLM: time = %lu\n", \
90 __LINE__, __FILE__, #x, jiffies); \
94 panic("DLM: Record message above and reboot.\n"); \
99 #define DLM_RTF_SHRINK 0x00000001
101 struct dlm_rsbtable
{
110 * Lockspace member (per node in a ls)
114 struct list_head list
;
124 * Save and manage recovery state for a lockspace.
128 struct list_head list
;
129 struct dlm_config_node
*nodes
;
135 * Pass input args to second stage locking function.
140 void (*astfn
) (void *astparam
);
142 void (*bastfn
) (void *astparam
, int mode
);
144 struct dlm_lksb
*lksb
;
145 unsigned long timeout
;
152 * A lock can be one of three types:
154 * local copy lock is mastered locally
155 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
156 * process copy lock is mastered on a remote node
157 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
158 * master copy master node's copy of a lock owned by remote node
159 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
161 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
162 * dlm_unlock. The dlm does not modify these or use any private flags in
163 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags
164 * are sent as-is to the remote master when the lock is remote.
166 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
167 * Some internal flags are shared between the master and process nodes;
168 * these shared flags are kept in the lower two bytes. One of these
169 * flags set on the master copy will be propagated to the process copy
170 * and v.v. Other internal flags are private to the master or process
171 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes.
173 * lkb_sbflags: status block flags. These flags are copied directly into
174 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
175 * ast. All defined in dlm.h with DLM_SBF_ prefix.
177 * lkb_status: the lock status indicates which rsb queue the lock is
178 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT
180 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
181 * reply is needed. Only set when the lkb is on the lockspace waiters
182 * list awaiting a reply from a remote node.
184 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
185 * is a master copy, nodeid specifies the remote lock holder, when the
186 * lkb is a process copy, the nodeid specifies the lock master.
191 #define DLM_LKSTS_WAITING 1
192 #define DLM_LKSTS_GRANTED 2
193 #define DLM_LKSTS_CONVERT 3
197 #define DLM_IFL_MSTCPY 0x00010000
198 #define DLM_IFL_RESEND 0x00020000
199 #define DLM_IFL_DEAD 0x00040000
200 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000
201 #define DLM_IFL_OVERLAP_CANCEL 0x00100000
202 #define DLM_IFL_ENDOFLIFE 0x00200000
203 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
204 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
205 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000
206 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */
207 #define DLM_IFL_USER 0x00000001
208 #define DLM_IFL_ORPHAN 0x00000002
210 #define DLM_CALLBACKS_SIZE 6
212 #define DLM_CB_CAST 0x00000001
213 #define DLM_CB_BAST 0x00000002
214 #define DLM_CB_SKIP 0x00000004
216 struct dlm_callback
{
218 uint32_t flags
; /* DLM_CBF_ */
219 int sb_status
; /* copy to lksb status */
220 uint8_t sb_flags
; /* copy to lksb flags */
221 int8_t mode
; /* rq mode of bast, gr mode of cast */
225 struct dlm_rsb
*lkb_resource
; /* the rsb */
227 int lkb_nodeid
; /* copied from rsb */
228 int lkb_ownpid
; /* pid of lock owner */
229 uint32_t lkb_id
; /* our lock ID */
230 uint32_t lkb_remid
; /* lock ID on remote partner */
231 uint32_t lkb_exflags
; /* external flags from caller */
232 uint32_t lkb_sbflags
; /* lksb flags */
233 uint32_t lkb_flags
; /* internal flags */
234 uint32_t lkb_lvbseq
; /* lvb sequence number */
236 int8_t lkb_status
; /* granted, waiting, convert */
237 int8_t lkb_rqmode
; /* requested lock mode */
238 int8_t lkb_grmode
; /* granted lock mode */
239 int8_t lkb_highbast
; /* highest mode bast sent for */
241 int8_t lkb_wait_type
; /* type of reply waiting for */
242 int8_t lkb_wait_count
;
243 int lkb_wait_nodeid
; /* for debugging */
245 struct list_head lkb_statequeue
; /* rsb g/c/w list */
246 struct list_head lkb_rsb_lookup
; /* waiting for rsb lookup */
247 struct list_head lkb_wait_reply
; /* waiting for remote reply */
248 struct list_head lkb_ownqueue
; /* list of locks for a process */
249 struct list_head lkb_time_list
;
250 ktime_t lkb_timestamp
;
251 ktime_t lkb_wait_time
;
252 unsigned long lkb_timeout_cs
;
254 struct mutex lkb_cb_mutex
;
255 struct work_struct lkb_cb_work
;
256 struct list_head lkb_cb_list
; /* for ls_cb_delay or proc->asts */
257 struct dlm_callback lkb_callbacks
[DLM_CALLBACKS_SIZE
];
258 struct dlm_callback lkb_last_cast
;
259 struct dlm_callback lkb_last_bast
;
260 ktime_t lkb_last_cast_time
; /* for debugging */
261 ktime_t lkb_last_bast_time
; /* for debugging */
263 uint64_t lkb_recover_seq
; /* from ls_recover_seq */
266 struct dlm_lksb
*lkb_lksb
; /* caller's status block */
267 void (*lkb_astfn
) (void *astparam
);
268 void (*lkb_bastfn
) (void *astparam
, int mode
);
270 void *lkb_astparam
; /* caller's ast arg */
271 struct dlm_user_args
*lkb_ua
;
276 * res_master_nodeid is "normal": 0 is unset/invalid, non-zero is the real
277 * nodeid, even when nodeid is our_nodeid.
279 * res_nodeid is "odd": -1 is unset/invalid, zero means our_nodeid,
280 * greater than zero when another nodeid.
282 * (TODO: remove res_nodeid and only use res_master_nodeid)
286 struct dlm_ls
*res_ls
; /* the lockspace */
288 struct mutex res_mutex
;
289 unsigned long res_flags
;
290 int res_length
; /* length of rsb name */
292 int res_master_nodeid
;
294 int res_id
; /* for ls_recover_idr */
297 uint32_t res_bucket
; /* rsbtbl */
298 unsigned long res_toss_time
;
299 uint32_t res_first_lkid
;
300 struct list_head res_lookup
; /* lkbs waiting on first */
302 struct list_head res_hashchain
;
303 struct rb_node res_hashnode
; /* rsbtbl */
305 struct list_head res_grantqueue
;
306 struct list_head res_convertqueue
;
307 struct list_head res_waitqueue
;
309 struct list_head res_root_list
; /* used for recovery */
310 struct list_head res_recover_list
; /* used for recovery */
311 int res_recover_locks_count
;
314 char res_name
[DLM_RESNAME_MAXLEN
+1];
317 /* dlm_master_lookup() flags */
319 #define DLM_LU_RECOVER_DIR 1
320 #define DLM_LU_RECOVER_MASTER 2
322 /* dlm_master_lookup() results */
324 #define DLM_LU_MATCH 1
327 /* find_rsb() flags */
329 #define R_REQUEST 0x00000001
330 #define R_RECEIVE_REQUEST 0x00000002
331 #define R_RECEIVE_RECOVER 0x00000004
336 RSB_MASTER_UNCERTAIN
,
338 RSB_VALNOTVALID_PREV
,
343 RSB_RECOVER_LVB_INVAL
,
346 static inline void rsb_set_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
348 __set_bit(flag
, &r
->res_flags
);
351 static inline void rsb_clear_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
353 __clear_bit(flag
, &r
->res_flags
);
356 static inline int rsb_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
358 return test_bit(flag
, &r
->res_flags
);
362 /* dlm_header is first element of all structs sent between nodes */
364 #define DLM_HEADER_MAJOR 0x00030000
365 #define DLM_HEADER_MINOR 0x00000001
367 #define DLM_HEADER_SLOTS 0x00000001
374 uint32_t h_lockspace
;
375 uint32_t h_nodeid
; /* nodeid of sender */
377 uint8_t h_cmd
; /* DLM_MSG, DLM_RCOM */
382 #define DLM_MSG_REQUEST 1
383 #define DLM_MSG_CONVERT 2
384 #define DLM_MSG_UNLOCK 3
385 #define DLM_MSG_CANCEL 4
386 #define DLM_MSG_REQUEST_REPLY 5
387 #define DLM_MSG_CONVERT_REPLY 6
388 #define DLM_MSG_UNLOCK_REPLY 7
389 #define DLM_MSG_CANCEL_REPLY 8
390 #define DLM_MSG_GRANT 9
391 #define DLM_MSG_BAST 10
392 #define DLM_MSG_LOOKUP 11
393 #define DLM_MSG_REMOVE 12
394 #define DLM_MSG_LOOKUP_REPLY 13
395 #define DLM_MSG_PURGE 14
398 struct dlm_header m_header
;
399 uint32_t m_type
; /* DLM_MSG_ */
402 uint32_t m_lkid
; /* lkid on sender */
403 uint32_t m_remid
; /* lkid on receiver */
404 uint32_t m_parent_lkid
;
405 uint32_t m_parent_remid
;
416 int m_result
; /* 0 or -EXXX */
417 char m_extra
[0]; /* name or lvb */
421 #define DLM_RS_NODES 0x00000001
422 #define DLM_RS_NODES_ALL 0x00000002
423 #define DLM_RS_DIR 0x00000004
424 #define DLM_RS_DIR_ALL 0x00000008
425 #define DLM_RS_LOCKS 0x00000010
426 #define DLM_RS_LOCKS_ALL 0x00000020
427 #define DLM_RS_DONE 0x00000040
428 #define DLM_RS_DONE_ALL 0x00000080
430 #define DLM_RCOM_STATUS 1
431 #define DLM_RCOM_NAMES 2
432 #define DLM_RCOM_LOOKUP 3
433 #define DLM_RCOM_LOCK 4
434 #define DLM_RCOM_STATUS_REPLY 5
435 #define DLM_RCOM_NAMES_REPLY 6
436 #define DLM_RCOM_LOOKUP_REPLY 7
437 #define DLM_RCOM_LOCK_REPLY 8
440 struct dlm_header rc_header
;
441 uint32_t rc_type
; /* DLM_RCOM_ */
442 int rc_result
; /* multi-purpose */
443 uint64_t rc_id
; /* match reply with request */
444 uint64_t rc_seq
; /* sender's ls_recover_seq */
445 uint64_t rc_seq_reply
; /* remote ls_recover_seq */
450 struct dlm_header header
; /* common to other two */
451 struct dlm_message message
;
452 struct dlm_rcom rcom
;
455 #define DLM_RSF_NEED_SLOTS 0x00000001
457 /* RCOM_STATUS data */
464 /* RCOM_STATUS_REPLY data */
469 /* DLM_HEADER_SLOTS adds: */
473 __le32 rf_generation
;
489 __le32 rl_parent_lkid
;
490 __le32 rl_parent_remid
;
501 char rl_name
[DLM_RESNAME_MAXLEN
];
506 * The max number of resources per rsbtbl bucket that shrink will attempt
507 * to remove in each iteration.
510 #define DLM_REMOVE_NAMES_MAX 8
513 struct list_head ls_list
; /* list of lockspaces */
514 dlm_lockspace_t
*ls_local_handle
;
515 uint32_t ls_global_id
; /* global unique lockspace ID */
516 uint32_t ls_generation
;
519 int ls_count
; /* refcount of processes in
520 the dlm using this ls */
521 int ls_create_count
; /* create/release refcount */
522 unsigned long ls_flags
; /* LSFL_ */
523 unsigned long ls_scan_time
;
524 struct kobject ls_kobj
;
526 struct idr ls_lkbidr
;
527 spinlock_t ls_lkbidr_spin
;
529 struct dlm_rsbtable
*ls_rsbtbl
;
530 uint32_t ls_rsbtbl_size
;
532 struct mutex ls_waiters_mutex
;
533 struct list_head ls_waiters
; /* lkbs needing a reply */
535 struct mutex ls_orphans_mutex
;
536 struct list_head ls_orphans
;
538 struct mutex ls_timeout_mutex
;
539 struct list_head ls_timeout
;
541 spinlock_t ls_new_rsb_spin
;
542 int ls_new_rsb_count
;
543 struct list_head ls_new_rsb
; /* new rsb structs */
545 spinlock_t ls_remove_spin
;
546 char ls_remove_name
[DLM_RESNAME_MAXLEN
+1];
547 char *ls_remove_names
[DLM_REMOVE_NAMES_MAX
];
549 int ls_remove_lens
[DLM_REMOVE_NAMES_MAX
];
551 struct list_head ls_nodes
; /* current nodes in ls */
552 struct list_head ls_nodes_gone
; /* dead node list, recovery */
553 int ls_num_nodes
; /* number of nodes in ls */
561 struct dlm_slot
*ls_slots
;
563 struct dlm_rsb ls_stub_rsb
; /* for returning errors */
564 struct dlm_lkb ls_stub_lkb
; /* for returning errors */
565 struct dlm_message ls_stub_ms
; /* for faking a reply */
567 struct dentry
*ls_debug_rsb_dentry
; /* debugfs */
568 struct dentry
*ls_debug_waiters_dentry
; /* debugfs */
569 struct dentry
*ls_debug_locks_dentry
; /* debugfs */
570 struct dentry
*ls_debug_all_dentry
; /* debugfs */
571 struct dentry
*ls_debug_toss_dentry
; /* debugfs */
573 wait_queue_head_t ls_uevent_wait
; /* user part of join/leave */
574 int ls_uevent_result
;
575 struct completion ls_members_done
;
576 int ls_members_result
;
578 struct miscdevice ls_device
;
580 struct workqueue_struct
*ls_callback_wq
;
582 /* recovery related */
584 struct mutex ls_cb_mutex
;
585 struct list_head ls_cb_delay
; /* save for queue_work later */
586 struct timer_list ls_timer
;
587 struct task_struct
*ls_recoverd_task
;
588 struct mutex ls_recoverd_active
;
589 spinlock_t ls_recover_lock
;
590 unsigned long ls_recover_begin
; /* jiffies timestamp */
591 uint32_t ls_recover_status
; /* DLM_RS_ */
592 uint64_t ls_recover_seq
;
593 struct dlm_recover
*ls_recover_args
;
594 struct rw_semaphore ls_in_recovery
; /* block local requests */
595 struct rw_semaphore ls_recv_active
; /* block dlm_recv */
596 struct list_head ls_requestqueue
;/* queue remote requests */
597 struct mutex ls_requestqueue_mutex
;
598 struct dlm_rcom
*ls_recover_buf
;
599 int ls_recover_nodeid
; /* for debugging */
600 unsigned int ls_recover_dir_sent_res
; /* for log info */
601 unsigned int ls_recover_dir_sent_msg
; /* for log info */
602 unsigned int ls_recover_locks_in
; /* for log info */
603 uint64_t ls_rcom_seq
;
604 spinlock_t ls_rcom_spin
;
605 struct list_head ls_recover_list
;
606 spinlock_t ls_recover_list_lock
;
607 int ls_recover_list_count
;
608 struct idr ls_recover_idr
;
609 spinlock_t ls_recover_idr_lock
;
610 wait_queue_head_t ls_wait_general
;
611 wait_queue_head_t ls_recover_lock_wait
;
612 struct mutex ls_clear_proc_locks
;
614 struct list_head ls_root_list
; /* root resources */
615 struct rw_semaphore ls_root_sem
; /* protect root_list */
617 const struct dlm_lockspace_ops
*ls_ops
;
625 * LSFL_RECOVER_STOP - dlm_ls_stop() sets this to tell dlm recovery routines
626 * that they should abort what they're doing so new recovery can be started.
628 * LSFL_RECOVER_DOWN - dlm_ls_stop() sets this to tell dlm_recoverd that it
629 * should do down_write() on the in_recovery rw_semaphore. (doing down_write
630 * within dlm_ls_stop causes complaints about the lock acquired/released
631 * in different contexts.)
633 * LSFL_RECOVER_LOCK - dlm_recoverd holds the in_recovery rw_semaphore.
634 * It sets this after it is done with down_write() on the in_recovery
635 * rw_semaphore and clears it after it has released the rw_semaphore.
637 * LSFL_RECOVER_WORK - dlm_ls_start() sets this to tell dlm_recoverd that it
638 * should begin recovery of the lockspace.
640 * LSFL_RUNNING - set when normal locking activity is enabled.
641 * dlm_ls_stop() clears this to tell dlm locking routines that they should
642 * quit what they are doing so recovery can run. dlm_recoverd sets
643 * this after recovery is finished.
646 #define LSFL_RECOVER_STOP 0
647 #define LSFL_RECOVER_DOWN 1
648 #define LSFL_RECOVER_LOCK 2
649 #define LSFL_RECOVER_WORK 3
650 #define LSFL_RUNNING 4
652 #define LSFL_RCOM_READY 5
653 #define LSFL_RCOM_WAIT 6
654 #define LSFL_UEVENT_WAIT 7
655 #define LSFL_TIMEWARN 8
656 #define LSFL_CB_DELAY 9
657 #define LSFL_NODIR 10
659 /* much of this is just saving user space pointers associated with the
660 lock that we pass back to the user lib with an ast */
662 struct dlm_user_args
{
663 struct dlm_user_proc
*proc
; /* each process that opens the lockspace
664 device has private data
665 (dlm_user_proc) on the struct file,
666 the process's locks point back to it*/
667 struct dlm_lksb lksb
;
668 struct dlm_lksb __user
*user_lksb
;
669 void __user
*castparam
;
670 void __user
*castaddr
;
671 void __user
*bastparam
;
672 void __user
*bastaddr
;
676 #define DLM_PROC_FLAGS_CLOSING 1
677 #define DLM_PROC_FLAGS_COMPAT 2
679 /* locks list is kept so we can remove all a process's locks when it
680 exits (or orphan those that are persistent) */
682 struct dlm_user_proc
{
683 dlm_lockspace_t
*lockspace
;
684 unsigned long flags
; /* DLM_PROC_FLAGS */
685 struct list_head asts
;
686 spinlock_t asts_spin
;
687 struct list_head locks
;
688 spinlock_t locks_spin
;
689 struct list_head unlocking
;
690 wait_queue_head_t wait
;
693 static inline int dlm_locking_stopped(struct dlm_ls
*ls
)
695 return !test_bit(LSFL_RUNNING
, &ls
->ls_flags
);
698 static inline int dlm_recovery_stopped(struct dlm_ls
*ls
)
700 return test_bit(LSFL_RECOVER_STOP
, &ls
->ls_flags
);
703 static inline int dlm_no_directory(struct dlm_ls
*ls
)
705 return test_bit(LSFL_NODIR
, &ls
->ls_flags
);
708 int dlm_netlink_init(void);
709 void dlm_netlink_exit(void);
710 void dlm_timeout_warn(struct dlm_lkb
*lkb
);
711 int dlm_plock_init(void);
712 void dlm_plock_exit(void);
714 #ifdef CONFIG_DLM_DEBUG
715 int dlm_register_debugfs(void);
716 void dlm_unregister_debugfs(void);
717 int dlm_create_debug_file(struct dlm_ls
*ls
);
718 void dlm_delete_debug_file(struct dlm_ls
*ls
);
720 static inline int dlm_register_debugfs(void) { return 0; }
721 static inline void dlm_unregister_debugfs(void) { }
722 static inline int dlm_create_debug_file(struct dlm_ls
*ls
) { return 0; }
723 static inline void dlm_delete_debug_file(struct dlm_ls
*ls
) { }
726 #endif /* __DLM_INTERNAL_DOT_H__ */