1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /******************************************************************************
3 *******************************************************************************
5 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
6 ** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved.
9 *******************************************************************************
10 ******************************************************************************/
12 #ifndef __DLM_INTERNAL_DOT_H__
13 #define __DLM_INTERNAL_DOT_H__
16 * This is the main header file to be included in each DLM source file.
19 #include <linux/slab.h>
20 #include <linux/sched.h>
21 #include <linux/types.h>
22 #include <linux/ctype.h>
23 #include <linux/spinlock.h>
24 #include <linux/vmalloc.h>
25 #include <linux/list.h>
26 #include <linux/errno.h>
27 #include <linux/random.h>
28 #include <linux/delay.h>
29 #include <linux/socket.h>
30 #include <linux/kthread.h>
31 #include <linux/kobject.h>
32 #include <linux/kref.h>
33 #include <linux/kernel.h>
34 #include <linux/jhash.h>
35 #include <linux/miscdevice.h>
36 #include <linux/mutex.h>
37 #include <linux/idr.h>
38 #include <linux/ratelimit.h>
39 #include <linux/uaccess.h>
41 #include <linux/dlm.h>
56 #define log_print(fmt, args...) \
57 printk(KERN_ERR "dlm: "fmt"\n" , ##args)
58 #define log_print_ratelimited(fmt, args...) \
59 printk_ratelimited(KERN_ERR "dlm: "fmt"\n", ##args)
60 #define log_error(ls, fmt, args...) \
61 printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args)
63 #define log_rinfo(ls, fmt, args...) \
65 if (dlm_config.ci_log_info) \
66 printk(KERN_INFO "dlm: %s: " fmt "\n", \
67 (ls)->ls_name, ##args); \
68 else if (dlm_config.ci_log_debug) \
69 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
70 (ls)->ls_name , ##args); \
73 #define log_debug(ls, fmt, args...) \
75 if (dlm_config.ci_log_debug) \
76 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
77 (ls)->ls_name , ##args); \
80 #define log_limit(ls, fmt, args...) \
82 if (dlm_config.ci_log_debug) \
83 printk_ratelimited(KERN_DEBUG "dlm: %s: " fmt "\n", \
84 (ls)->ls_name , ##args); \
87 #define DLM_ASSERT(x, do) \
91 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \
92 "DLM: assertion: \"%s\"\n" \
93 "DLM: time = %lu\n", \
94 __LINE__, __FILE__, #x, jiffies); \
97 panic("DLM: Record message above and reboot.\n"); \
102 #define DLM_RTF_SHRINK 0x00000001
104 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 */
266 uint64_t lkb_recover_seq; /* from ls_recover_seq */
269 struct dlm_lksb *lkb_lksb; /* caller's status block */
270 void (*lkb_astfn) (void *astparam);
271 void (*lkb_bastfn) (void *astparam, int mode);
273 void *lkb_astparam; /* caller's ast arg */
274 struct dlm_user_args *lkb_ua;
279 * res_master_nodeid is "normal": 0 is unset/invalid, non-zero is the real
280 * nodeid, even when nodeid is our_nodeid.
282 * res_nodeid is "odd": -1 is unset/invalid, zero means our_nodeid,
283 * greater than zero when another nodeid.
285 * (TODO: remove res_nodeid and only use res_master_nodeid)
289 struct dlm_ls *res_ls; /* the lockspace */
291 struct mutex res_mutex;
292 unsigned long res_flags;
293 int res_length; /* length of rsb name */
295 int res_master_nodeid;
297 int res_id; /* for ls_recover_idr */
300 uint32_t res_bucket; /* rsbtbl */
301 unsigned long res_toss_time;
302 uint32_t res_first_lkid;
303 struct list_head res_lookup; /* lkbs waiting on first */
305 struct list_head res_hashchain;
306 struct rb_node res_hashnode; /* rsbtbl */
308 struct list_head res_grantqueue;
309 struct list_head res_convertqueue;
310 struct list_head res_waitqueue;
312 struct list_head res_root_list; /* used for recovery */
313 struct list_head res_recover_list; /* used for recovery */
314 int res_recover_locks_count;
317 char res_name[DLM_RESNAME_MAXLEN+1];
320 /* dlm_master_lookup() flags */
322 #define DLM_LU_RECOVER_DIR 1
323 #define DLM_LU_RECOVER_MASTER 2
325 /* dlm_master_lookup() results */
327 #define DLM_LU_MATCH 1
330 /* find_rsb() flags */
332 #define R_REQUEST 0x00000001
333 #define R_RECEIVE_REQUEST 0x00000002
334 #define R_RECEIVE_RECOVER 0x00000004
339 RSB_MASTER_UNCERTAIN,
341 RSB_VALNOTVALID_PREV,
346 RSB_RECOVER_LVB_INVAL,
349 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
351 __set_bit(flag, &r->res_flags);
354 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
356 __clear_bit(flag, &r->res_flags);
359 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
361 return test_bit(flag, &r->res_flags);
365 /* dlm_header is first element of all structs sent between nodes */
367 #define DLM_HEADER_MAJOR 0x00030000
368 #define DLM_HEADER_MINOR 0x00000002
370 #define DLM_VERSION_3_1 0x00030001
371 #define DLM_VERSION_3_2 0x00030002
373 #define DLM_HEADER_SLOTS 0x00000001
384 /* for DLM_MSG and DLM_RCOM */
385 uint32_t h_lockspace;
386 /* for DLM_ACK and DLM_OPTS */
389 uint32_t h_nodeid; /* nodeid of sender */
391 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */
395 #define DLM_MSG_REQUEST 1
396 #define DLM_MSG_CONVERT 2
397 #define DLM_MSG_UNLOCK 3
398 #define DLM_MSG_CANCEL 4
399 #define DLM_MSG_REQUEST_REPLY 5
400 #define DLM_MSG_CONVERT_REPLY 6
401 #define DLM_MSG_UNLOCK_REPLY 7
402 #define DLM_MSG_CANCEL_REPLY 8
403 #define DLM_MSG_GRANT 9
404 #define DLM_MSG_BAST 10
405 #define DLM_MSG_LOOKUP 11
406 #define DLM_MSG_REMOVE 12
407 #define DLM_MSG_LOOKUP_REPLY 13
408 #define DLM_MSG_PURGE 14
411 struct dlm_header m_header;
412 uint32_t m_type; /* DLM_MSG_ */
415 uint32_t m_lkid; /* lkid on sender */
416 uint32_t m_remid; /* lkid on receiver */
417 uint32_t m_parent_lkid;
418 uint32_t m_parent_remid;
429 int m_result; /* 0 or -EXXX */
430 char m_extra[]; /* name or lvb */
434 #define DLM_RS_NODES 0x00000001
435 #define DLM_RS_NODES_ALL 0x00000002
436 #define DLM_RS_DIR 0x00000004
437 #define DLM_RS_DIR_ALL 0x00000008
438 #define DLM_RS_LOCKS 0x00000010
439 #define DLM_RS_LOCKS_ALL 0x00000020
440 #define DLM_RS_DONE 0x00000040
441 #define DLM_RS_DONE_ALL 0x00000080
443 #define DLM_RCOM_STATUS 1
444 #define DLM_RCOM_NAMES 2
445 #define DLM_RCOM_LOOKUP 3
446 #define DLM_RCOM_LOCK 4
447 #define DLM_RCOM_STATUS_REPLY 5
448 #define DLM_RCOM_NAMES_REPLY 6
449 #define DLM_RCOM_LOOKUP_REPLY 7
450 #define DLM_RCOM_LOCK_REPLY 8
453 struct dlm_header rc_header;
454 uint32_t rc_type; /* DLM_RCOM_ */
455 int rc_result; /* multi-purpose */
456 uint64_t rc_id; /* match reply with request */
457 uint64_t rc_seq; /* sender's ls_recover_seq */
458 uint64_t rc_seq_reply; /* remote ls_recover_seq */
462 struct dlm_opt_header {
466 /* need to be 8 byte aligned */
470 /* encapsulation header */
472 struct dlm_header o_header;
481 struct dlm_header header; /* common to other two */
482 struct dlm_message message;
483 struct dlm_rcom rcom;
484 struct dlm_opts opts;
487 #define DLM_RSF_NEED_SLOTS 0x00000001
489 /* RCOM_STATUS data */
496 /* RCOM_STATUS_REPLY data */
501 /* DLM_HEADER_SLOTS adds: */
505 __le32 rf_generation;
521 __le32 rl_parent_lkid;
522 __le32 rl_parent_remid;
533 char rl_name[DLM_RESNAME_MAXLEN];
538 * The max number of resources per rsbtbl bucket that shrink will attempt
539 * to remove in each iteration.
542 #define DLM_REMOVE_NAMES_MAX 8
545 struct list_head ls_list; /* list of lockspaces */
546 dlm_lockspace_t *ls_local_handle;
547 uint32_t ls_global_id; /* global unique lockspace ID */
548 uint32_t ls_generation;
551 atomic_t ls_count; /* refcount of processes in
552 the dlm using this ls */
553 wait_queue_head_t ls_count_wait;
554 int ls_create_count; /* create/release refcount */
555 unsigned long ls_flags; /* LSFL_ */
556 unsigned long ls_scan_time;
557 struct kobject ls_kobj;
559 struct idr ls_lkbidr;
560 spinlock_t ls_lkbidr_spin;
562 struct dlm_rsbtable *ls_rsbtbl;
563 uint32_t ls_rsbtbl_size;
565 struct mutex ls_waiters_mutex;
566 struct list_head ls_waiters; /* lkbs needing a reply */
568 struct mutex ls_orphans_mutex;
569 struct list_head ls_orphans;
571 struct mutex ls_timeout_mutex;
572 struct list_head ls_timeout;
574 spinlock_t ls_new_rsb_spin;
575 int ls_new_rsb_count;
576 struct list_head ls_new_rsb; /* new rsb structs */
578 spinlock_t ls_remove_spin;
579 wait_queue_head_t ls_remove_wait;
580 char ls_remove_name[DLM_RESNAME_MAXLEN+1];
581 char *ls_remove_names[DLM_REMOVE_NAMES_MAX];
583 int ls_remove_lens[DLM_REMOVE_NAMES_MAX];
585 struct list_head ls_nodes; /* current nodes in ls */
586 struct list_head ls_nodes_gone; /* dead node list, recovery */
587 int ls_num_nodes; /* number of nodes in ls */
595 struct dlm_slot *ls_slots;
597 struct dlm_rsb ls_stub_rsb; /* for returning errors */
598 struct dlm_lkb ls_stub_lkb; /* for returning errors */
599 struct dlm_message ls_stub_ms; /* for faking a reply */
601 struct dentry *ls_debug_rsb_dentry; /* debugfs */
602 struct dentry *ls_debug_waiters_dentry; /* debugfs */
603 struct dentry *ls_debug_locks_dentry; /* debugfs */
604 struct dentry *ls_debug_all_dentry; /* debugfs */
605 struct dentry *ls_debug_toss_dentry; /* debugfs */
607 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */
608 int ls_uevent_result;
609 struct completion ls_members_done;
610 int ls_members_result;
612 struct miscdevice ls_device;
614 struct workqueue_struct *ls_callback_wq;
616 /* recovery related */
618 struct mutex ls_cb_mutex;
619 struct list_head ls_cb_delay; /* save for queue_work later */
620 struct timer_list ls_timer;
621 struct task_struct *ls_recoverd_task;
622 struct mutex ls_recoverd_active;
623 spinlock_t ls_recover_lock;
624 unsigned long ls_recover_begin; /* jiffies timestamp */
625 uint32_t ls_recover_status; /* DLM_RS_ */
626 uint64_t ls_recover_seq;
627 struct dlm_recover *ls_recover_args;
628 struct rw_semaphore ls_in_recovery; /* block local requests */
629 struct rw_semaphore ls_recv_active; /* block dlm_recv */
630 struct list_head ls_requestqueue;/* queue remote requests */
631 atomic_t ls_requestqueue_cnt;
632 wait_queue_head_t ls_requestqueue_wait;
633 struct mutex ls_requestqueue_mutex;
634 struct dlm_rcom *ls_recover_buf;
635 int ls_recover_nodeid; /* for debugging */
636 unsigned int ls_recover_dir_sent_res; /* for log info */
637 unsigned int ls_recover_dir_sent_msg; /* for log info */
638 unsigned int ls_recover_locks_in; /* for log info */
639 uint64_t ls_rcom_seq;
640 spinlock_t ls_rcom_spin;
641 struct list_head ls_recover_list;
642 spinlock_t ls_recover_list_lock;
643 int ls_recover_list_count;
644 struct idr ls_recover_idr;
645 spinlock_t ls_recover_idr_lock;
646 wait_queue_head_t ls_wait_general;
647 wait_queue_head_t ls_recover_lock_wait;
648 struct mutex ls_clear_proc_locks;
650 struct list_head ls_root_list; /* root resources */
651 struct rw_semaphore ls_root_sem; /* protect root_list */
653 const struct dlm_lockspace_ops *ls_ops;
661 * LSFL_RECOVER_STOP - dlm_ls_stop() sets this to tell dlm recovery routines
662 * that they should abort what they're doing so new recovery can be started.
664 * LSFL_RECOVER_DOWN - dlm_ls_stop() sets this to tell dlm_recoverd that it
665 * should do down_write() on the in_recovery rw_semaphore. (doing down_write
666 * within dlm_ls_stop causes complaints about the lock acquired/released
667 * in different contexts.)
669 * LSFL_RECOVER_LOCK - dlm_recoverd holds the in_recovery rw_semaphore.
670 * It sets this after it is done with down_write() on the in_recovery
671 * rw_semaphore and clears it after it has released the rw_semaphore.
673 * LSFL_RECOVER_WORK - dlm_ls_start() sets this to tell dlm_recoverd that it
674 * should begin recovery of the lockspace.
676 * LSFL_RUNNING - set when normal locking activity is enabled.
677 * dlm_ls_stop() clears this to tell dlm locking routines that they should
678 * quit what they are doing so recovery can run. dlm_recoverd sets
679 * this after recovery is finished.
682 #define LSFL_RECOVER_STOP 0
683 #define LSFL_RECOVER_DOWN 1
684 #define LSFL_RECOVER_LOCK 2
685 #define LSFL_RECOVER_WORK 3
686 #define LSFL_RUNNING 4
688 #define LSFL_RCOM_READY 5
689 #define LSFL_RCOM_WAIT 6
690 #define LSFL_UEVENT_WAIT 7
691 #define LSFL_TIMEWARN 8
692 #define LSFL_CB_DELAY 9
693 #define LSFL_NODIR 10
695 /* much of this is just saving user space pointers associated with the
696 lock that we pass back to the user lib with an ast */
698 struct dlm_user_args {
699 struct dlm_user_proc *proc; /* each process that opens the lockspace
700 device has private data
701 (dlm_user_proc) on the struct file,
702 the process's locks point back to it*/
703 struct dlm_lksb lksb;
704 struct dlm_lksb __user *user_lksb;
705 void __user *castparam;
706 void __user *castaddr;
707 void __user *bastparam;
708 void __user *bastaddr;
712 #define DLM_PROC_FLAGS_CLOSING 1
713 #define DLM_PROC_FLAGS_COMPAT 2
715 /* locks list is kept so we can remove all a process's locks when it
716 exits (or orphan those that are persistent) */
718 struct dlm_user_proc {
719 dlm_lockspace_t *lockspace;
720 unsigned long flags; /* DLM_PROC_FLAGS */
721 struct list_head asts;
722 spinlock_t asts_spin;
723 struct list_head locks;
724 spinlock_t locks_spin;
725 struct list_head unlocking;
726 wait_queue_head_t wait;
729 static inline int dlm_locking_stopped(struct dlm_ls *ls)
731 return !test_bit(LSFL_RUNNING, &ls->ls_flags);
734 static inline int dlm_recovery_stopped(struct dlm_ls *ls)
736 return test_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
739 static inline int dlm_no_directory(struct dlm_ls *ls)
741 return test_bit(LSFL_NODIR, &ls->ls_flags);
744 int dlm_netlink_init(void);
745 void dlm_netlink_exit(void);
746 void dlm_timeout_warn(struct dlm_lkb *lkb);
747 int dlm_plock_init(void);
748 void dlm_plock_exit(void);
750 #ifdef CONFIG_DLM_DEBUG
751 void dlm_register_debugfs(void);
752 void dlm_unregister_debugfs(void);
753 void dlm_create_debug_file(struct dlm_ls *ls);
754 void dlm_delete_debug_file(struct dlm_ls *ls);
755 void *dlm_create_debug_comms_file(int nodeid, void *data);
756 void dlm_delete_debug_comms_file(void *ctx);
758 static inline void dlm_register_debugfs(void) { }
759 static inline void dlm_unregister_debugfs(void) { }
760 static inline void dlm_create_debug_file(struct dlm_ls *ls) { }
761 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { }
762 static inline void *dlm_create_debug_comms_file(int nodeid, void *data) { return NULL; }
763 static inline void dlm_delete_debug_comms_file(void *ctx) { }
766 #endif /* __DLM_INTERNAL_DOT_H__ */