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 #ifdef CONFIG_DLM_DEPRECATED_API
149 unsigned long timeout;
157 * A lock can be one of three types:
159 * local copy lock is mastered locally
160 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
161 * process copy lock is mastered on a remote node
162 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
163 * master copy master node's copy of a lock owned by remote node
164 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
166 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
167 * dlm_unlock. The dlm does not modify these or use any private flags in
168 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags
169 * are sent as-is to the remote master when the lock is remote.
171 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
172 * Some internal flags are shared between the master and process nodes;
173 * these shared flags are kept in the lower two bytes. One of these
174 * flags set on the master copy will be propagated to the process copy
175 * and v.v. Other internal flags are private to the master or process
176 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes.
178 * lkb_sbflags: status block flags. These flags are copied directly into
179 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
180 * ast. All defined in dlm.h with DLM_SBF_ prefix.
182 * lkb_status: the lock status indicates which rsb queue the lock is
183 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT
185 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
186 * reply is needed. Only set when the lkb is on the lockspace waiters
187 * list awaiting a reply from a remote node.
189 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
190 * is a master copy, nodeid specifies the remote lock holder, when the
191 * lkb is a process copy, the nodeid specifies the lock master.
196 #define DLM_LKSTS_WAITING 1
197 #define DLM_LKSTS_GRANTED 2
198 #define DLM_LKSTS_CONVERT 3
202 #define DLM_IFL_MSTCPY 0x00010000
203 #define DLM_IFL_RESEND 0x00020000
204 #define DLM_IFL_DEAD 0x00040000
205 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000
206 #define DLM_IFL_OVERLAP_CANCEL 0x00100000
207 #define DLM_IFL_ENDOFLIFE 0x00200000
208 #ifdef CONFIG_DLM_DEPRECATED_API
209 #define DLM_IFL_WATCH_TIMEWARN 0x00400000
210 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000
212 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000
213 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */
214 /* least significant 2 bytes are message changed, they are full transmitted
215 * but at receive side only the 2 bytes LSB will be set.
217 * Even wireshark dlm dissector does only evaluate the lower bytes and note
218 * that they may not be used on transceiver side, we assume the higher bytes
219 * are for internal use or reserved so long they are not parsed on receiver
222 #define DLM_IFL_USER 0x00000001
223 #define DLM_IFL_ORPHAN 0x00000002
225 #define DLM_CALLBACKS_SIZE 6
227 #define DLM_CB_CAST 0x00000001
228 #define DLM_CB_BAST 0x00000002
229 #define DLM_CB_SKIP 0x00000004
231 struct dlm_callback {
233 uint32_t flags; /* DLM_CBF_ */
234 int sb_status; /* copy to lksb status */
235 uint8_t sb_flags; /* copy to lksb flags */
236 int8_t mode; /* rq mode of bast, gr mode of cast */
240 struct dlm_rsb *lkb_resource; /* the rsb */
242 int lkb_nodeid; /* copied from rsb */
243 int lkb_ownpid; /* pid of lock owner */
244 uint32_t lkb_id; /* our lock ID */
245 uint32_t lkb_remid; /* lock ID on remote partner */
246 uint32_t lkb_exflags; /* external flags from caller */
247 uint32_t lkb_sbflags; /* lksb flags */
248 uint32_t lkb_flags; /* internal flags */
249 uint32_t lkb_lvbseq; /* lvb sequence number */
251 int8_t lkb_status; /* granted, waiting, convert */
252 int8_t lkb_rqmode; /* requested lock mode */
253 int8_t lkb_grmode; /* granted lock mode */
254 int8_t lkb_highbast; /* highest mode bast sent for */
256 int8_t lkb_wait_type; /* type of reply waiting for */
257 int8_t lkb_wait_count;
258 int lkb_wait_nodeid; /* for debugging */
260 struct list_head lkb_statequeue; /* rsb g/c/w list */
261 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */
262 struct list_head lkb_wait_reply; /* waiting for remote reply */
263 struct list_head lkb_ownqueue; /* list of locks for a process */
264 ktime_t lkb_timestamp;
266 #ifdef CONFIG_DLM_DEPRECATED_API
267 struct list_head lkb_time_list;
268 unsigned long lkb_timeout_cs;
271 struct mutex lkb_cb_mutex;
272 struct work_struct lkb_cb_work;
273 struct list_head lkb_cb_list; /* for ls_cb_delay or proc->asts */
274 struct dlm_callback lkb_callbacks[DLM_CALLBACKS_SIZE];
275 struct dlm_callback lkb_last_cast;
276 struct dlm_callback lkb_last_bast;
277 ktime_t lkb_last_cast_time; /* for debugging */
278 ktime_t lkb_last_bast_time; /* for debugging */
280 uint64_t lkb_recover_seq; /* from ls_recover_seq */
283 struct dlm_lksb *lkb_lksb; /* caller's status block */
284 void (*lkb_astfn) (void *astparam);
285 void (*lkb_bastfn) (void *astparam, int mode);
287 void *lkb_astparam; /* caller's ast arg */
288 struct dlm_user_args *lkb_ua;
293 * res_master_nodeid is "normal": 0 is unset/invalid, non-zero is the real
294 * nodeid, even when nodeid is our_nodeid.
296 * res_nodeid is "odd": -1 is unset/invalid, zero means our_nodeid,
297 * greater than zero when another nodeid.
299 * (TODO: remove res_nodeid and only use res_master_nodeid)
303 struct dlm_ls *res_ls; /* the lockspace */
305 struct mutex res_mutex;
306 unsigned long res_flags;
307 int res_length; /* length of rsb name */
309 int res_master_nodeid;
311 int res_id; /* for ls_recover_idr */
314 uint32_t res_bucket; /* rsbtbl */
315 unsigned long res_toss_time;
316 uint32_t res_first_lkid;
317 struct list_head res_lookup; /* lkbs waiting on first */
319 struct list_head res_hashchain;
320 struct rb_node res_hashnode; /* rsbtbl */
322 struct list_head res_grantqueue;
323 struct list_head res_convertqueue;
324 struct list_head res_waitqueue;
326 struct list_head res_root_list; /* used for recovery */
327 struct list_head res_recover_list; /* used for recovery */
328 int res_recover_locks_count;
331 char res_name[DLM_RESNAME_MAXLEN+1];
334 /* dlm_master_lookup() flags */
336 #define DLM_LU_RECOVER_DIR 1
337 #define DLM_LU_RECOVER_MASTER 2
339 /* dlm_master_lookup() results */
341 #define DLM_LU_MATCH 1
344 /* find_rsb() flags */
346 #define R_REQUEST 0x00000001
347 #define R_RECEIVE_REQUEST 0x00000002
348 #define R_RECEIVE_RECOVER 0x00000004
353 RSB_MASTER_UNCERTAIN,
355 RSB_VALNOTVALID_PREV,
360 RSB_RECOVER_LVB_INVAL,
363 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
365 __set_bit(flag, &r->res_flags);
368 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
370 __clear_bit(flag, &r->res_flags);
373 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
375 return test_bit(flag, &r->res_flags);
379 /* dlm_header is first element of all structs sent between nodes */
381 #define DLM_HEADER_MAJOR 0x00030000
382 #define DLM_HEADER_MINOR 0x00000002
384 #define DLM_VERSION_3_1 0x00030001
385 #define DLM_VERSION_3_2 0x00030002
387 #define DLM_HEADER_SLOTS 0x00000001
398 /* for DLM_MSG and DLM_RCOM */
400 /* for DLM_ACK and DLM_OPTS */
403 __le32 h_nodeid; /* nodeid of sender */
405 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */
409 #define DLM_MSG_REQUEST 1
410 #define DLM_MSG_CONVERT 2
411 #define DLM_MSG_UNLOCK 3
412 #define DLM_MSG_CANCEL 4
413 #define DLM_MSG_REQUEST_REPLY 5
414 #define DLM_MSG_CONVERT_REPLY 6
415 #define DLM_MSG_UNLOCK_REPLY 7
416 #define DLM_MSG_CANCEL_REPLY 8
417 #define DLM_MSG_GRANT 9
418 #define DLM_MSG_BAST 10
419 #define DLM_MSG_LOOKUP 11
420 #define DLM_MSG_REMOVE 12
421 #define DLM_MSG_LOOKUP_REPLY 13
422 #define DLM_MSG_PURGE 14
425 struct dlm_header m_header;
426 __le32 m_type; /* DLM_MSG_ */
429 __le32 m_lkid; /* lkid on sender */
430 __le32 m_remid; /* lkid on receiver */
431 __le32 m_parent_lkid;
432 __le32 m_parent_remid;
443 __le32 m_result; /* 0 or -EXXX */
444 char m_extra[]; /* name or lvb */
448 #define DLM_RS_NODES 0x00000001
449 #define DLM_RS_NODES_ALL 0x00000002
450 #define DLM_RS_DIR 0x00000004
451 #define DLM_RS_DIR_ALL 0x00000008
452 #define DLM_RS_LOCKS 0x00000010
453 #define DLM_RS_LOCKS_ALL 0x00000020
454 #define DLM_RS_DONE 0x00000040
455 #define DLM_RS_DONE_ALL 0x00000080
457 #define DLM_RCOM_STATUS 1
458 #define DLM_RCOM_NAMES 2
459 #define DLM_RCOM_LOOKUP 3
460 #define DLM_RCOM_LOCK 4
461 #define DLM_RCOM_STATUS_REPLY 5
462 #define DLM_RCOM_NAMES_REPLY 6
463 #define DLM_RCOM_LOOKUP_REPLY 7
464 #define DLM_RCOM_LOCK_REPLY 8
467 struct dlm_header rc_header;
468 __le32 rc_type; /* DLM_RCOM_ */
469 __le32 rc_result; /* multi-purpose */
470 __le64 rc_id; /* match reply with request */
471 __le64 rc_seq; /* sender's ls_recover_seq */
472 __le64 rc_seq_reply; /* remote ls_recover_seq */
476 struct dlm_opt_header {
480 /* need to be 8 byte aligned */
484 /* encapsulation header */
486 struct dlm_header o_header;
495 struct dlm_header header; /* common to other two */
496 struct dlm_message message;
497 struct dlm_rcom rcom;
498 struct dlm_opts opts;
501 #define DLM_RSF_NEED_SLOTS 0x00000001
503 /* RCOM_STATUS data */
510 /* RCOM_STATUS_REPLY data */
515 /* DLM_HEADER_SLOTS adds: */
519 __le32 rf_generation;
535 __le32 rl_parent_lkid;
536 __le32 rl_parent_remid;
547 char rl_name[DLM_RESNAME_MAXLEN];
552 * The max number of resources per rsbtbl bucket that shrink will attempt
553 * to remove in each iteration.
556 #define DLM_REMOVE_NAMES_MAX 8
559 struct list_head ls_list; /* list of lockspaces */
560 dlm_lockspace_t *ls_local_handle;
561 uint32_t ls_global_id; /* global unique lockspace ID */
562 uint32_t ls_generation;
565 atomic_t ls_count; /* refcount of processes in
566 the dlm using this ls */
567 wait_queue_head_t ls_count_wait;
568 int ls_create_count; /* create/release refcount */
569 unsigned long ls_flags; /* LSFL_ */
570 unsigned long ls_scan_time;
571 struct kobject ls_kobj;
573 struct idr ls_lkbidr;
574 spinlock_t ls_lkbidr_spin;
576 struct dlm_rsbtable *ls_rsbtbl;
577 uint32_t ls_rsbtbl_size;
579 struct mutex ls_waiters_mutex;
580 struct list_head ls_waiters; /* lkbs needing a reply */
582 struct mutex ls_orphans_mutex;
583 struct list_head ls_orphans;
585 #ifdef CONFIG_DLM_DEPRECATED_API
586 struct mutex ls_timeout_mutex;
587 struct list_head ls_timeout;
590 spinlock_t ls_new_rsb_spin;
591 int ls_new_rsb_count;
592 struct list_head ls_new_rsb; /* new rsb structs */
594 spinlock_t ls_remove_spin;
595 wait_queue_head_t ls_remove_wait;
596 char ls_remove_name[DLM_RESNAME_MAXLEN+1];
597 char *ls_remove_names[DLM_REMOVE_NAMES_MAX];
599 int ls_remove_lens[DLM_REMOVE_NAMES_MAX];
601 struct list_head ls_nodes; /* current nodes in ls */
602 struct list_head ls_nodes_gone; /* dead node list, recovery */
603 int ls_num_nodes; /* number of nodes in ls */
611 struct dlm_slot *ls_slots;
613 struct dlm_rsb ls_stub_rsb; /* for returning errors */
614 struct dlm_lkb ls_stub_lkb; /* for returning errors */
615 struct dlm_message ls_stub_ms; /* for faking a reply */
617 struct dentry *ls_debug_rsb_dentry; /* debugfs */
618 struct dentry *ls_debug_waiters_dentry; /* debugfs */
619 struct dentry *ls_debug_locks_dentry; /* debugfs */
620 struct dentry *ls_debug_all_dentry; /* debugfs */
621 struct dentry *ls_debug_toss_dentry; /* debugfs */
623 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */
624 int ls_uevent_result;
625 struct completion ls_recovery_done;
626 int ls_recovery_result;
628 struct miscdevice ls_device;
630 struct workqueue_struct *ls_callback_wq;
632 /* recovery related */
634 struct mutex ls_cb_mutex;
635 struct list_head ls_cb_delay; /* save for queue_work later */
636 struct timer_list ls_timer;
637 struct task_struct *ls_recoverd_task;
638 struct mutex ls_recoverd_active;
639 spinlock_t ls_recover_lock;
640 unsigned long ls_recover_begin; /* jiffies timestamp */
641 uint32_t ls_recover_status; /* DLM_RS_ */
642 uint64_t ls_recover_seq;
643 struct dlm_recover *ls_recover_args;
644 struct rw_semaphore ls_in_recovery; /* block local requests */
645 struct rw_semaphore ls_recv_active; /* block dlm_recv */
646 struct list_head ls_requestqueue;/* queue remote requests */
647 atomic_t ls_requestqueue_cnt;
648 wait_queue_head_t ls_requestqueue_wait;
649 struct mutex ls_requestqueue_mutex;
650 struct dlm_rcom *ls_recover_buf;
651 int ls_recover_nodeid; /* for debugging */
652 unsigned int ls_recover_dir_sent_res; /* for log info */
653 unsigned int ls_recover_dir_sent_msg; /* for log info */
654 unsigned int ls_recover_locks_in; /* for log info */
655 uint64_t ls_rcom_seq;
656 spinlock_t ls_rcom_spin;
657 struct list_head ls_recover_list;
658 spinlock_t ls_recover_list_lock;
659 int ls_recover_list_count;
660 struct idr ls_recover_idr;
661 spinlock_t ls_recover_idr_lock;
662 wait_queue_head_t ls_wait_general;
663 wait_queue_head_t ls_recover_lock_wait;
664 spinlock_t ls_clear_proc_locks;
666 struct list_head ls_root_list; /* root resources */
667 struct rw_semaphore ls_root_sem; /* protect root_list */
669 const struct dlm_lockspace_ops *ls_ops;
677 * LSFL_RECOVER_STOP - dlm_ls_stop() sets this to tell dlm recovery routines
678 * that they should abort what they're doing so new recovery can be started.
680 * LSFL_RECOVER_DOWN - dlm_ls_stop() sets this to tell dlm_recoverd that it
681 * should do down_write() on the in_recovery rw_semaphore. (doing down_write
682 * within dlm_ls_stop causes complaints about the lock acquired/released
683 * in different contexts.)
685 * LSFL_RECOVER_LOCK - dlm_recoverd holds the in_recovery rw_semaphore.
686 * It sets this after it is done with down_write() on the in_recovery
687 * rw_semaphore and clears it after it has released the rw_semaphore.
689 * LSFL_RECOVER_WORK - dlm_ls_start() sets this to tell dlm_recoverd that it
690 * should begin recovery of the lockspace.
692 * LSFL_RUNNING - set when normal locking activity is enabled.
693 * dlm_ls_stop() clears this to tell dlm locking routines that they should
694 * quit what they are doing so recovery can run. dlm_recoverd sets
695 * this after recovery is finished.
698 #define LSFL_RECOVER_STOP 0
699 #define LSFL_RECOVER_DOWN 1
700 #define LSFL_RECOVER_LOCK 2
701 #define LSFL_RECOVER_WORK 3
702 #define LSFL_RUNNING 4
704 #define LSFL_RCOM_READY 5
705 #define LSFL_RCOM_WAIT 6
706 #define LSFL_UEVENT_WAIT 7
707 #ifdef CONFIG_DLM_DEPRECATED_API
708 #define LSFL_TIMEWARN 8
710 #define LSFL_CB_DELAY 9
711 #define LSFL_NODIR 10
713 /* much of this is just saving user space pointers associated with the
714 lock that we pass back to the user lib with an ast */
716 struct dlm_user_args {
717 struct dlm_user_proc *proc; /* each process that opens the lockspace
718 device has private data
719 (dlm_user_proc) on the struct file,
720 the process's locks point back to it*/
721 struct dlm_lksb lksb;
722 struct dlm_lksb __user *user_lksb;
723 void __user *castparam;
724 void __user *castaddr;
725 void __user *bastparam;
726 void __user *bastaddr;
730 #define DLM_PROC_FLAGS_CLOSING 1
731 #define DLM_PROC_FLAGS_COMPAT 2
733 /* locks list is kept so we can remove all a process's locks when it
734 exits (or orphan those that are persistent) */
736 struct dlm_user_proc {
737 dlm_lockspace_t *lockspace;
738 unsigned long flags; /* DLM_PROC_FLAGS */
739 struct list_head asts;
740 spinlock_t asts_spin;
741 struct list_head locks;
742 spinlock_t locks_spin;
743 struct list_head unlocking;
744 wait_queue_head_t wait;
747 static inline int dlm_locking_stopped(struct dlm_ls *ls)
749 return !test_bit(LSFL_RUNNING, &ls->ls_flags);
752 static inline int dlm_recovery_stopped(struct dlm_ls *ls)
754 return test_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
757 static inline int dlm_no_directory(struct dlm_ls *ls)
759 return test_bit(LSFL_NODIR, &ls->ls_flags);
762 #ifdef CONFIG_DLM_DEPRECATED_API
763 int dlm_netlink_init(void);
764 void dlm_netlink_exit(void);
765 void dlm_timeout_warn(struct dlm_lkb *lkb);
767 static inline int dlm_netlink_init(void) { return 0; }
768 static inline void dlm_netlink_exit(void) { };
769 static inline void dlm_timeout_warn(struct dlm_lkb *lkb) { };
771 int dlm_plock_init(void);
772 void dlm_plock_exit(void);
774 #ifdef CONFIG_DLM_DEBUG
775 void dlm_register_debugfs(void);
776 void dlm_unregister_debugfs(void);
777 void dlm_create_debug_file(struct dlm_ls *ls);
778 void dlm_delete_debug_file(struct dlm_ls *ls);
779 void *dlm_create_debug_comms_file(int nodeid, void *data);
780 void dlm_delete_debug_comms_file(void *ctx);
782 static inline void dlm_register_debugfs(void) { }
783 static inline void dlm_unregister_debugfs(void) { }
784 static inline void dlm_create_debug_file(struct dlm_ls *ls) { }
785 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { }
786 static inline void *dlm_create_debug_comms_file(int nodeid, void *data) { return NULL; }
787 static inline void dlm_delete_debug_comms_file(void *ctx) { }
790 #endif /* __DLM_INTERNAL_DOT_H__ */