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_rinfo(ls, fmt, args...) \
71 if (dlm_config.ci_log_info) \
72 printk(KERN_INFO "dlm: %s: " fmt "\n", \
73 (ls)->ls_name, ##args); \
74 else if (dlm_config.ci_log_debug) \
75 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
76 (ls)->ls_name , ##args); \
79 #define log_debug(ls, fmt, args...) \
81 if (dlm_config.ci_log_debug) \
82 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
83 (ls)->ls_name , ##args); \
86 #define log_limit(ls, fmt, args...) \
88 if (dlm_config.ci_log_debug) \
89 printk_ratelimited(KERN_DEBUG "dlm: %s: " fmt "\n", \
90 (ls)->ls_name , ##args); \
93 #define DLM_ASSERT(x, do) \
97 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \
98 "DLM: assertion: \"%s\"\n" \
99 "DLM: time = %lu\n", \
100 __LINE__, __FILE__, #x, jiffies); \
104 panic("DLM: Record message above and reboot.\n"); \
109 #define DLM_RTF_SHRINK 0x00000001
111 struct dlm_rsbtable
{
120 * Lockspace member (per node in a ls)
124 struct list_head list
;
134 * Save and manage recovery state for a lockspace.
138 struct list_head list
;
139 struct dlm_config_node
*nodes
;
145 * Pass input args to second stage locking function.
150 void (*astfn
) (void *astparam
);
152 void (*bastfn
) (void *astparam
, int mode
);
154 struct dlm_lksb
*lksb
;
155 unsigned long timeout
;
162 * A lock can be one of three types:
164 * local copy lock is mastered locally
165 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
166 * process copy lock is mastered on a remote node
167 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
168 * master copy master node's copy of a lock owned by remote node
169 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
171 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
172 * dlm_unlock. The dlm does not modify these or use any private flags in
173 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags
174 * are sent as-is to the remote master when the lock is remote.
176 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
177 * Some internal flags are shared between the master and process nodes;
178 * these shared flags are kept in the lower two bytes. One of these
179 * flags set on the master copy will be propagated to the process copy
180 * and v.v. Other internal flags are private to the master or process
181 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes.
183 * lkb_sbflags: status block flags. These flags are copied directly into
184 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
185 * ast. All defined in dlm.h with DLM_SBF_ prefix.
187 * lkb_status: the lock status indicates which rsb queue the lock is
188 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT
190 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
191 * reply is needed. Only set when the lkb is on the lockspace waiters
192 * list awaiting a reply from a remote node.
194 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
195 * is a master copy, nodeid specifies the remote lock holder, when the
196 * lkb is a process copy, the nodeid specifies the lock master.
201 #define DLM_LKSTS_WAITING 1
202 #define DLM_LKSTS_GRANTED 2
203 #define DLM_LKSTS_CONVERT 3
207 #define DLM_IFL_MSTCPY 0x00010000
208 #define DLM_IFL_RESEND 0x00020000
209 #define DLM_IFL_DEAD 0x00040000
210 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000
211 #define DLM_IFL_OVERLAP_CANCEL 0x00100000
212 #define DLM_IFL_ENDOFLIFE 0x00200000
213 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
214 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
215 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000
216 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */
217 #define DLM_IFL_USER 0x00000001
218 #define DLM_IFL_ORPHAN 0x00000002
220 #define DLM_CALLBACKS_SIZE 6
222 #define DLM_CB_CAST 0x00000001
223 #define DLM_CB_BAST 0x00000002
224 #define DLM_CB_SKIP 0x00000004
226 struct dlm_callback
{
228 uint32_t flags
; /* DLM_CBF_ */
229 int sb_status
; /* copy to lksb status */
230 uint8_t sb_flags
; /* copy to lksb flags */
231 int8_t mode
; /* rq mode of bast, gr mode of cast */
235 struct dlm_rsb
*lkb_resource
; /* the rsb */
237 int lkb_nodeid
; /* copied from rsb */
238 int lkb_ownpid
; /* pid of lock owner */
239 uint32_t lkb_id
; /* our lock ID */
240 uint32_t lkb_remid
; /* lock ID on remote partner */
241 uint32_t lkb_exflags
; /* external flags from caller */
242 uint32_t lkb_sbflags
; /* lksb flags */
243 uint32_t lkb_flags
; /* internal flags */
244 uint32_t lkb_lvbseq
; /* lvb sequence number */
246 int8_t lkb_status
; /* granted, waiting, convert */
247 int8_t lkb_rqmode
; /* requested lock mode */
248 int8_t lkb_grmode
; /* granted lock mode */
249 int8_t lkb_highbast
; /* highest mode bast sent for */
251 int8_t lkb_wait_type
; /* type of reply waiting for */
252 int8_t lkb_wait_count
;
253 int lkb_wait_nodeid
; /* for debugging */
255 struct list_head lkb_statequeue
; /* rsb g/c/w list */
256 struct list_head lkb_rsb_lookup
; /* waiting for rsb lookup */
257 struct list_head lkb_wait_reply
; /* waiting for remote reply */
258 struct list_head lkb_ownqueue
; /* list of locks for a process */
259 struct list_head lkb_time_list
;
260 ktime_t lkb_timestamp
;
261 ktime_t lkb_wait_time
;
262 unsigned long lkb_timeout_cs
;
264 struct mutex lkb_cb_mutex
;
265 struct work_struct lkb_cb_work
;
266 struct list_head lkb_cb_list
; /* for ls_cb_delay or proc->asts */
267 struct dlm_callback lkb_callbacks
[DLM_CALLBACKS_SIZE
];
268 struct dlm_callback lkb_last_cast
;
269 struct dlm_callback lkb_last_bast
;
270 ktime_t lkb_last_cast_time
; /* for debugging */
271 ktime_t lkb_last_bast_time
; /* for debugging */
273 uint64_t lkb_recover_seq
; /* from ls_recover_seq */
276 struct dlm_lksb
*lkb_lksb
; /* caller's status block */
277 void (*lkb_astfn
) (void *astparam
);
278 void (*lkb_bastfn
) (void *astparam
, int mode
);
280 void *lkb_astparam
; /* caller's ast arg */
281 struct dlm_user_args
*lkb_ua
;
286 * res_master_nodeid is "normal": 0 is unset/invalid, non-zero is the real
287 * nodeid, even when nodeid is our_nodeid.
289 * res_nodeid is "odd": -1 is unset/invalid, zero means our_nodeid,
290 * greater than zero when another nodeid.
292 * (TODO: remove res_nodeid and only use res_master_nodeid)
296 struct dlm_ls
*res_ls
; /* the lockspace */
298 struct mutex res_mutex
;
299 unsigned long res_flags
;
300 int res_length
; /* length of rsb name */
302 int res_master_nodeid
;
304 int res_id
; /* for ls_recover_idr */
307 uint32_t res_bucket
; /* rsbtbl */
308 unsigned long res_toss_time
;
309 uint32_t res_first_lkid
;
310 struct list_head res_lookup
; /* lkbs waiting on first */
312 struct list_head res_hashchain
;
313 struct rb_node res_hashnode
; /* rsbtbl */
315 struct list_head res_grantqueue
;
316 struct list_head res_convertqueue
;
317 struct list_head res_waitqueue
;
319 struct list_head res_root_list
; /* used for recovery */
320 struct list_head res_recover_list
; /* used for recovery */
321 int res_recover_locks_count
;
324 char res_name
[DLM_RESNAME_MAXLEN
+1];
327 /* dlm_master_lookup() flags */
329 #define DLM_LU_RECOVER_DIR 1
330 #define DLM_LU_RECOVER_MASTER 2
332 /* dlm_master_lookup() results */
334 #define DLM_LU_MATCH 1
337 /* find_rsb() flags */
339 #define R_REQUEST 0x00000001
340 #define R_RECEIVE_REQUEST 0x00000002
341 #define R_RECEIVE_RECOVER 0x00000004
346 RSB_MASTER_UNCERTAIN
,
348 RSB_VALNOTVALID_PREV
,
353 RSB_RECOVER_LVB_INVAL
,
356 static inline void rsb_set_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
358 __set_bit(flag
, &r
->res_flags
);
361 static inline void rsb_clear_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
363 __clear_bit(flag
, &r
->res_flags
);
366 static inline int rsb_flag(struct dlm_rsb
*r
, enum rsb_flags flag
)
368 return test_bit(flag
, &r
->res_flags
);
372 /* dlm_header is first element of all structs sent between nodes */
374 #define DLM_HEADER_MAJOR 0x00030000
375 #define DLM_HEADER_MINOR 0x00000001
377 #define DLM_HEADER_SLOTS 0x00000001
384 uint32_t h_lockspace
;
385 uint32_t h_nodeid
; /* nodeid of sender */
387 uint8_t h_cmd
; /* DLM_MSG, DLM_RCOM */
392 #define DLM_MSG_REQUEST 1
393 #define DLM_MSG_CONVERT 2
394 #define DLM_MSG_UNLOCK 3
395 #define DLM_MSG_CANCEL 4
396 #define DLM_MSG_REQUEST_REPLY 5
397 #define DLM_MSG_CONVERT_REPLY 6
398 #define DLM_MSG_UNLOCK_REPLY 7
399 #define DLM_MSG_CANCEL_REPLY 8
400 #define DLM_MSG_GRANT 9
401 #define DLM_MSG_BAST 10
402 #define DLM_MSG_LOOKUP 11
403 #define DLM_MSG_REMOVE 12
404 #define DLM_MSG_LOOKUP_REPLY 13
405 #define DLM_MSG_PURGE 14
408 struct dlm_header m_header
;
409 uint32_t m_type
; /* DLM_MSG_ */
412 uint32_t m_lkid
; /* lkid on sender */
413 uint32_t m_remid
; /* lkid on receiver */
414 uint32_t m_parent_lkid
;
415 uint32_t m_parent_remid
;
426 int m_result
; /* 0 or -EXXX */
427 char m_extra
[0]; /* name or lvb */
431 #define DLM_RS_NODES 0x00000001
432 #define DLM_RS_NODES_ALL 0x00000002
433 #define DLM_RS_DIR 0x00000004
434 #define DLM_RS_DIR_ALL 0x00000008
435 #define DLM_RS_LOCKS 0x00000010
436 #define DLM_RS_LOCKS_ALL 0x00000020
437 #define DLM_RS_DONE 0x00000040
438 #define DLM_RS_DONE_ALL 0x00000080
440 #define DLM_RCOM_STATUS 1
441 #define DLM_RCOM_NAMES 2
442 #define DLM_RCOM_LOOKUP 3
443 #define DLM_RCOM_LOCK 4
444 #define DLM_RCOM_STATUS_REPLY 5
445 #define DLM_RCOM_NAMES_REPLY 6
446 #define DLM_RCOM_LOOKUP_REPLY 7
447 #define DLM_RCOM_LOCK_REPLY 8
450 struct dlm_header rc_header
;
451 uint32_t rc_type
; /* DLM_RCOM_ */
452 int rc_result
; /* multi-purpose */
453 uint64_t rc_id
; /* match reply with request */
454 uint64_t rc_seq
; /* sender's ls_recover_seq */
455 uint64_t rc_seq_reply
; /* remote ls_recover_seq */
460 struct dlm_header header
; /* common to other two */
461 struct dlm_message message
;
462 struct dlm_rcom rcom
;
465 #define DLM_RSF_NEED_SLOTS 0x00000001
467 /* RCOM_STATUS data */
474 /* RCOM_STATUS_REPLY data */
479 /* DLM_HEADER_SLOTS adds: */
483 __le32 rf_generation
;
499 __le32 rl_parent_lkid
;
500 __le32 rl_parent_remid
;
511 char rl_name
[DLM_RESNAME_MAXLEN
];
516 * The max number of resources per rsbtbl bucket that shrink will attempt
517 * to remove in each iteration.
520 #define DLM_REMOVE_NAMES_MAX 8
523 struct list_head ls_list
; /* list of lockspaces */
524 dlm_lockspace_t
*ls_local_handle
;
525 uint32_t ls_global_id
; /* global unique lockspace ID */
526 uint32_t ls_generation
;
529 int ls_count
; /* refcount of processes in
530 the dlm using this ls */
531 int ls_create_count
; /* create/release refcount */
532 unsigned long ls_flags
; /* LSFL_ */
533 unsigned long ls_scan_time
;
534 struct kobject ls_kobj
;
536 struct idr ls_lkbidr
;
537 spinlock_t ls_lkbidr_spin
;
539 struct dlm_rsbtable
*ls_rsbtbl
;
540 uint32_t ls_rsbtbl_size
;
542 struct mutex ls_waiters_mutex
;
543 struct list_head ls_waiters
; /* lkbs needing a reply */
545 struct mutex ls_orphans_mutex
;
546 struct list_head ls_orphans
;
548 struct mutex ls_timeout_mutex
;
549 struct list_head ls_timeout
;
551 spinlock_t ls_new_rsb_spin
;
552 int ls_new_rsb_count
;
553 struct list_head ls_new_rsb
; /* new rsb structs */
555 spinlock_t ls_remove_spin
;
556 char ls_remove_name
[DLM_RESNAME_MAXLEN
+1];
557 char *ls_remove_names
[DLM_REMOVE_NAMES_MAX
];
559 int ls_remove_lens
[DLM_REMOVE_NAMES_MAX
];
561 struct list_head ls_nodes
; /* current nodes in ls */
562 struct list_head ls_nodes_gone
; /* dead node list, recovery */
563 int ls_num_nodes
; /* number of nodes in ls */
571 struct dlm_slot
*ls_slots
;
573 struct dlm_rsb ls_stub_rsb
; /* for returning errors */
574 struct dlm_lkb ls_stub_lkb
; /* for returning errors */
575 struct dlm_message ls_stub_ms
; /* for faking a reply */
577 struct dentry
*ls_debug_rsb_dentry
; /* debugfs */
578 struct dentry
*ls_debug_waiters_dentry
; /* debugfs */
579 struct dentry
*ls_debug_locks_dentry
; /* debugfs */
580 struct dentry
*ls_debug_all_dentry
; /* debugfs */
581 struct dentry
*ls_debug_toss_dentry
; /* debugfs */
583 wait_queue_head_t ls_uevent_wait
; /* user part of join/leave */
584 int ls_uevent_result
;
585 struct completion ls_members_done
;
586 int ls_members_result
;
588 struct miscdevice ls_device
;
590 struct workqueue_struct
*ls_callback_wq
;
592 /* recovery related */
594 struct mutex ls_cb_mutex
;
595 struct list_head ls_cb_delay
; /* save for queue_work later */
596 struct timer_list ls_timer
;
597 struct task_struct
*ls_recoverd_task
;
598 struct mutex ls_recoverd_active
;
599 spinlock_t ls_recover_lock
;
600 unsigned long ls_recover_begin
; /* jiffies timestamp */
601 uint32_t ls_recover_status
; /* DLM_RS_ */
602 uint64_t ls_recover_seq
;
603 struct dlm_recover
*ls_recover_args
;
604 struct rw_semaphore ls_in_recovery
; /* block local requests */
605 struct rw_semaphore ls_recv_active
; /* block dlm_recv */
606 struct list_head ls_requestqueue
;/* queue remote requests */
607 struct mutex ls_requestqueue_mutex
;
608 struct dlm_rcom
*ls_recover_buf
;
609 int ls_recover_nodeid
; /* for debugging */
610 unsigned int ls_recover_dir_sent_res
; /* for log info */
611 unsigned int ls_recover_dir_sent_msg
; /* for log info */
612 unsigned int ls_recover_locks_in
; /* for log info */
613 uint64_t ls_rcom_seq
;
614 spinlock_t ls_rcom_spin
;
615 struct list_head ls_recover_list
;
616 spinlock_t ls_recover_list_lock
;
617 int ls_recover_list_count
;
618 struct idr ls_recover_idr
;
619 spinlock_t ls_recover_idr_lock
;
620 wait_queue_head_t ls_wait_general
;
621 wait_queue_head_t ls_recover_lock_wait
;
622 struct mutex ls_clear_proc_locks
;
624 struct list_head ls_root_list
; /* root resources */
625 struct rw_semaphore ls_root_sem
; /* protect root_list */
627 const struct dlm_lockspace_ops
*ls_ops
;
635 * LSFL_RECOVER_STOP - dlm_ls_stop() sets this to tell dlm recovery routines
636 * that they should abort what they're doing so new recovery can be started.
638 * LSFL_RECOVER_DOWN - dlm_ls_stop() sets this to tell dlm_recoverd that it
639 * should do down_write() on the in_recovery rw_semaphore. (doing down_write
640 * within dlm_ls_stop causes complaints about the lock acquired/released
641 * in different contexts.)
643 * LSFL_RECOVER_LOCK - dlm_recoverd holds the in_recovery rw_semaphore.
644 * It sets this after it is done with down_write() on the in_recovery
645 * rw_semaphore and clears it after it has released the rw_semaphore.
647 * LSFL_RECOVER_WORK - dlm_ls_start() sets this to tell dlm_recoverd that it
648 * should begin recovery of the lockspace.
650 * LSFL_RUNNING - set when normal locking activity is enabled.
651 * dlm_ls_stop() clears this to tell dlm locking routines that they should
652 * quit what they are doing so recovery can run. dlm_recoverd sets
653 * this after recovery is finished.
656 #define LSFL_RECOVER_STOP 0
657 #define LSFL_RECOVER_DOWN 1
658 #define LSFL_RECOVER_LOCK 2
659 #define LSFL_RECOVER_WORK 3
660 #define LSFL_RUNNING 4
662 #define LSFL_RCOM_READY 5
663 #define LSFL_RCOM_WAIT 6
664 #define LSFL_UEVENT_WAIT 7
665 #define LSFL_TIMEWARN 8
666 #define LSFL_CB_DELAY 9
667 #define LSFL_NODIR 10
669 /* much of this is just saving user space pointers associated with the
670 lock that we pass back to the user lib with an ast */
672 struct dlm_user_args
{
673 struct dlm_user_proc
*proc
; /* each process that opens the lockspace
674 device has private data
675 (dlm_user_proc) on the struct file,
676 the process's locks point back to it*/
677 struct dlm_lksb lksb
;
678 struct dlm_lksb __user
*user_lksb
;
679 void __user
*castparam
;
680 void __user
*castaddr
;
681 void __user
*bastparam
;
682 void __user
*bastaddr
;
686 #define DLM_PROC_FLAGS_CLOSING 1
687 #define DLM_PROC_FLAGS_COMPAT 2
689 /* locks list is kept so we can remove all a process's locks when it
690 exits (or orphan those that are persistent) */
692 struct dlm_user_proc
{
693 dlm_lockspace_t
*lockspace
;
694 unsigned long flags
; /* DLM_PROC_FLAGS */
695 struct list_head asts
;
696 spinlock_t asts_spin
;
697 struct list_head locks
;
698 spinlock_t locks_spin
;
699 struct list_head unlocking
;
700 wait_queue_head_t wait
;
703 static inline int dlm_locking_stopped(struct dlm_ls
*ls
)
705 return !test_bit(LSFL_RUNNING
, &ls
->ls_flags
);
708 static inline int dlm_recovery_stopped(struct dlm_ls
*ls
)
710 return test_bit(LSFL_RECOVER_STOP
, &ls
->ls_flags
);
713 static inline int dlm_no_directory(struct dlm_ls
*ls
)
715 return test_bit(LSFL_NODIR
, &ls
->ls_flags
);
718 int dlm_netlink_init(void);
719 void dlm_netlink_exit(void);
720 void dlm_timeout_warn(struct dlm_lkb
*lkb
);
721 int dlm_plock_init(void);
722 void dlm_plock_exit(void);
724 #ifdef CONFIG_DLM_DEBUG
725 int dlm_register_debugfs(void);
726 void dlm_unregister_debugfs(void);
727 int dlm_create_debug_file(struct dlm_ls
*ls
);
728 void dlm_delete_debug_file(struct dlm_ls
*ls
);
730 static inline int dlm_register_debugfs(void) { return 0; }
731 static inline void dlm_unregister_debugfs(void) { }
732 static inline int dlm_create_debug_file(struct dlm_ls
*ls
) { return 0; }
733 static inline void dlm_delete_debug_file(struct dlm_ls
*ls
) { }
736 #endif /* __DLM_INTERNAL_DOT_H__ */