4 * This is the central lockd service.
6 * FIXME: Separate the lockd NFS server functionality from the lockd NFS
7 * client functionality. Oh why didn't Sun create two separate
8 * services in the first place?
10 * Authors: Olaf Kirch (okir@monad.swb.de)
12 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
15 #include <linux/module.h>
16 #include <linux/init.h>
17 #include <linux/sysctl.h>
18 #include <linux/moduleparam.h>
20 #include <linux/sched/signal.h>
21 #include <linux/errno.h>
23 #include <linux/uio.h>
24 #include <linux/smp.h>
25 #include <linux/mutex.h>
26 #include <linux/kthread.h>
27 #include <linux/freezer.h>
28 #include <linux/inetdevice.h>
30 #include <linux/sunrpc/types.h>
31 #include <linux/sunrpc/stats.h>
32 #include <linux/sunrpc/clnt.h>
33 #include <linux/sunrpc/svc.h>
34 #include <linux/sunrpc/svcsock.h>
35 #include <linux/sunrpc/svc_xprt.h>
37 #include <net/addrconf.h>
39 #include <linux/lockd/lockd.h>
40 #include <linux/nfs.h>
45 #define NLMDBG_FACILITY NLMDBG_SVC
46 #define LOCKD_BUFSIZE (1024 + NLMSVC_XDRSIZE)
47 #define ALLOWED_SIGS (sigmask(SIGKILL))
49 static struct svc_program nlmsvc_program
;
51 const struct nlmsvc_binding
*nlmsvc_ops
;
52 EXPORT_SYMBOL_GPL(nlmsvc_ops
);
54 static DEFINE_MUTEX(nlmsvc_mutex
);
55 static unsigned int nlmsvc_users
;
56 static struct task_struct
*nlmsvc_task
;
57 static struct svc_rqst
*nlmsvc_rqst
;
58 unsigned long nlmsvc_timeout
;
60 static atomic_t nlm_ntf_refcnt
= ATOMIC_INIT(0);
61 static DECLARE_WAIT_QUEUE_HEAD(nlm_ntf_wq
);
63 unsigned int lockd_net_id
;
66 * These can be set at insmod time (useful for NFS as root filesystem),
67 * and also changed through the sysctl interface. -- Jamie Lokier, Aug 2003
69 static unsigned long nlm_grace_period
;
70 static unsigned long nlm_timeout
= LOCKD_DFLT_TIMEO
;
71 static int nlm_udpport
, nlm_tcpport
;
73 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
74 static unsigned int nlm_max_connections
= 1024;
77 * Constants needed for the sysctl interface.
79 static const unsigned long nlm_grace_period_min
= 0;
80 static const unsigned long nlm_grace_period_max
= 240;
81 static const unsigned long nlm_timeout_min
= 3;
82 static const unsigned long nlm_timeout_max
= 20;
83 static const int nlm_port_min
= 0, nlm_port_max
= 65535;
86 static struct ctl_table_header
* nlm_sysctl_table
;
89 static unsigned long get_lockd_grace_period(void)
91 /* Note: nlm_timeout should always be nonzero */
93 return roundup(nlm_grace_period
, nlm_timeout
) * HZ
;
95 return nlm_timeout
* 5 * HZ
;
98 static void grace_ender(struct work_struct
*grace
)
100 struct delayed_work
*dwork
= to_delayed_work(grace
);
101 struct lockd_net
*ln
= container_of(dwork
, struct lockd_net
,
104 locks_end_grace(&ln
->lockd_manager
);
107 static void set_grace_period(struct net
*net
)
109 unsigned long grace_period
= get_lockd_grace_period();
110 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
112 locks_start_grace(net
, &ln
->lockd_manager
);
113 cancel_delayed_work_sync(&ln
->grace_period_end
);
114 schedule_delayed_work(&ln
->grace_period_end
, grace_period
);
117 static void restart_grace(void)
120 struct net
*net
= &init_net
;
121 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
123 cancel_delayed_work_sync(&ln
->grace_period_end
);
124 locks_end_grace(&ln
->lockd_manager
);
125 nlmsvc_invalidate_all();
126 set_grace_period(net
);
131 * This is the lockd kernel thread
137 struct svc_rqst
*rqstp
= vrqstp
;
138 struct net
*net
= &init_net
;
139 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
141 /* try_to_freeze() is called from svc_recv() */
144 /* Allow SIGKILL to tell lockd to drop all of its locks */
145 allow_signal(SIGKILL
);
147 dprintk("NFS locking service started (ver " LOCKD_VERSION
").\n");
150 * The main request loop. We don't terminate until the last
151 * NFS mount or NFS daemon has gone away.
153 while (!kthread_should_stop()) {
154 long timeout
= MAX_SCHEDULE_TIMEOUT
;
155 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
157 /* update sv_maxconn if it has changed */
158 rqstp
->rq_server
->sv_maxconn
= nlm_max_connections
;
161 flush_signals(current
);
166 timeout
= nlmsvc_retry_blocked();
169 * Find a socket with data available and call its
172 err
= svc_recv(rqstp
, timeout
);
173 if (err
== -EAGAIN
|| err
== -EINTR
)
175 dprintk("lockd: request from %s\n",
176 svc_print_addr(rqstp
, buf
, sizeof(buf
)));
180 flush_signals(current
);
182 nlmsvc_invalidate_all();
183 nlm_shutdown_hosts();
184 cancel_delayed_work_sync(&ln
->grace_period_end
);
185 locks_end_grace(&ln
->lockd_manager
);
189 static int create_lockd_listener(struct svc_serv
*serv
, const char *name
,
190 struct net
*net
, const int family
,
191 const unsigned short port
)
193 struct svc_xprt
*xprt
;
195 xprt
= svc_find_xprt(serv
, name
, net
, family
, 0);
197 return svc_create_xprt(serv
, name
, net
, family
, port
,
203 static int create_lockd_family(struct svc_serv
*serv
, struct net
*net
,
208 err
= create_lockd_listener(serv
, "udp", net
, family
, nlm_udpport
);
212 return create_lockd_listener(serv
, "tcp", net
, family
, nlm_tcpport
);
216 * Ensure there are active UDP and TCP listeners for lockd.
218 * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
219 * local services (such as rpc.statd) still require UDP, and
220 * some NFS servers do not yet support NLM over TCP.
222 * Returns zero if all listeners are available; otherwise a
223 * negative errno value is returned.
225 static int make_socks(struct svc_serv
*serv
, struct net
*net
)
230 err
= create_lockd_family(serv
, net
, PF_INET
);
234 err
= create_lockd_family(serv
, net
, PF_INET6
);
235 if (err
< 0 && err
!= -EAFNOSUPPORT
)
244 "lockd_up: makesock failed, error=%d\n", err
);
245 svc_shutdown_net(serv
, net
);
249 static int lockd_up_net(struct svc_serv
*serv
, struct net
*net
)
251 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
254 if (ln
->nlmsvc_users
++)
257 error
= svc_bind(serv
, net
);
261 error
= make_socks(serv
, net
);
264 set_grace_period(net
);
265 dprintk("%s: per-net data created; net=%x\n", __func__
, net
->ns
.inum
);
273 static void lockd_down_net(struct svc_serv
*serv
, struct net
*net
)
275 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
277 if (ln
->nlmsvc_users
) {
278 if (--ln
->nlmsvc_users
== 0) {
279 nlm_shutdown_hosts_net(net
);
280 cancel_delayed_work_sync(&ln
->grace_period_end
);
281 locks_end_grace(&ln
->lockd_manager
);
282 svc_shutdown_net(serv
, net
);
283 dprintk("%s: per-net data destroyed; net=%x\n",
284 __func__
, net
->ns
.inum
);
287 pr_err("%s: no users! task=%p, net=%x\n",
288 __func__
, nlmsvc_task
, net
->ns
.inum
);
293 static int lockd_inetaddr_event(struct notifier_block
*this,
294 unsigned long event
, void *ptr
)
296 struct in_ifaddr
*ifa
= (struct in_ifaddr
*)ptr
;
297 struct sockaddr_in sin
;
299 if ((event
!= NETDEV_DOWN
) ||
300 !atomic_inc_not_zero(&nlm_ntf_refcnt
))
304 dprintk("lockd_inetaddr_event: removed %pI4\n",
306 sin
.sin_family
= AF_INET
;
307 sin
.sin_addr
.s_addr
= ifa
->ifa_local
;
308 svc_age_temp_xprts_now(nlmsvc_rqst
->rq_server
,
309 (struct sockaddr
*)&sin
);
311 atomic_dec(&nlm_ntf_refcnt
);
312 wake_up(&nlm_ntf_wq
);
318 static struct notifier_block lockd_inetaddr_notifier
= {
319 .notifier_call
= lockd_inetaddr_event
,
322 #if IS_ENABLED(CONFIG_IPV6)
323 static int lockd_inet6addr_event(struct notifier_block
*this,
324 unsigned long event
, void *ptr
)
326 struct inet6_ifaddr
*ifa
= (struct inet6_ifaddr
*)ptr
;
327 struct sockaddr_in6 sin6
;
329 if ((event
!= NETDEV_DOWN
) ||
330 !atomic_inc_not_zero(&nlm_ntf_refcnt
))
334 dprintk("lockd_inet6addr_event: removed %pI6\n", &ifa
->addr
);
335 sin6
.sin6_family
= AF_INET6
;
336 sin6
.sin6_addr
= ifa
->addr
;
337 if (ipv6_addr_type(&sin6
.sin6_addr
) & IPV6_ADDR_LINKLOCAL
)
338 sin6
.sin6_scope_id
= ifa
->idev
->dev
->ifindex
;
339 svc_age_temp_xprts_now(nlmsvc_rqst
->rq_server
,
340 (struct sockaddr
*)&sin6
);
342 atomic_dec(&nlm_ntf_refcnt
);
343 wake_up(&nlm_ntf_wq
);
349 static struct notifier_block lockd_inet6addr_notifier
= {
350 .notifier_call
= lockd_inet6addr_event
,
354 static void lockd_unregister_notifiers(void)
356 unregister_inetaddr_notifier(&lockd_inetaddr_notifier
);
357 #if IS_ENABLED(CONFIG_IPV6)
358 unregister_inet6addr_notifier(&lockd_inet6addr_notifier
);
360 wait_event(nlm_ntf_wq
, atomic_read(&nlm_ntf_refcnt
) == 0);
363 static void lockd_svc_exit_thread(void)
365 atomic_dec(&nlm_ntf_refcnt
);
366 lockd_unregister_notifiers();
367 svc_exit_thread(nlmsvc_rqst
);
370 static int lockd_start_svc(struct svc_serv
*serv
)
378 * Create the kernel thread and wait for it to start.
380 nlmsvc_rqst
= svc_prepare_thread(serv
, &serv
->sv_pools
[0], NUMA_NO_NODE
);
381 if (IS_ERR(nlmsvc_rqst
)) {
382 error
= PTR_ERR(nlmsvc_rqst
);
384 "lockd_up: svc_rqst allocation failed, error=%d\n",
386 lockd_unregister_notifiers();
390 atomic_inc(&nlm_ntf_refcnt
);
391 svc_sock_update_bufs(serv
);
392 serv
->sv_maxconn
= nlm_max_connections
;
394 nlmsvc_task
= kthread_create(lockd
, nlmsvc_rqst
, "%s", serv
->sv_name
);
395 if (IS_ERR(nlmsvc_task
)) {
396 error
= PTR_ERR(nlmsvc_task
);
398 "lockd_up: kthread_run failed, error=%d\n", error
);
401 nlmsvc_rqst
->rq_task
= nlmsvc_task
;
402 wake_up_process(nlmsvc_task
);
404 dprintk("lockd_up: service started\n");
408 lockd_svc_exit_thread();
415 static const struct svc_serv_ops lockd_sv_ops
= {
416 .svo_shutdown
= svc_rpcb_cleanup
,
417 .svo_enqueue_xprt
= svc_xprt_do_enqueue
,
420 static struct svc_serv
*lockd_create_svc(void)
422 struct svc_serv
*serv
;
425 * Check whether we're already up and running.
429 * Note: increase service usage, because later in case of error
430 * svc_destroy() will be called.
432 svc_get(nlmsvc_rqst
->rq_server
);
433 return nlmsvc_rqst
->rq_server
;
437 * Sanity check: if there's no pid,
438 * we should be the first user ...
442 "lockd_up: no pid, %d users??\n", nlmsvc_users
);
445 nlm_timeout
= LOCKD_DFLT_TIMEO
;
446 nlmsvc_timeout
= nlm_timeout
* HZ
;
448 serv
= svc_create(&nlmsvc_program
, LOCKD_BUFSIZE
, &lockd_sv_ops
);
450 printk(KERN_WARNING
"lockd_up: create service failed\n");
451 return ERR_PTR(-ENOMEM
);
453 register_inetaddr_notifier(&lockd_inetaddr_notifier
);
454 #if IS_ENABLED(CONFIG_IPV6)
455 register_inet6addr_notifier(&lockd_inet6addr_notifier
);
457 dprintk("lockd_up: service created\n");
462 * Bring up the lockd process if it's not already up.
464 int lockd_up(struct net
*net
)
466 struct svc_serv
*serv
;
469 mutex_lock(&nlmsvc_mutex
);
471 serv
= lockd_create_svc();
473 error
= PTR_ERR(serv
);
477 error
= lockd_up_net(serv
, net
);
479 lockd_unregister_notifiers();
483 error
= lockd_start_svc(serv
);
485 lockd_down_net(serv
, net
);
490 * Note: svc_serv structures have an initial use count of 1,
491 * so we exit through here on both success and failure.
496 mutex_unlock(&nlmsvc_mutex
);
499 EXPORT_SYMBOL_GPL(lockd_up
);
502 * Decrement the user count and bring down lockd if we're the last.
505 lockd_down(struct net
*net
)
507 mutex_lock(&nlmsvc_mutex
);
508 lockd_down_net(nlmsvc_rqst
->rq_server
, net
);
513 printk(KERN_ERR
"lockd_down: no users! task=%p\n",
519 printk(KERN_ERR
"lockd_down: no lockd running.\n");
522 kthread_stop(nlmsvc_task
);
523 dprintk("lockd_down: service stopped\n");
524 lockd_svc_exit_thread();
525 dprintk("lockd_down: service destroyed\n");
529 mutex_unlock(&nlmsvc_mutex
);
531 EXPORT_SYMBOL_GPL(lockd_down
);
536 * Sysctl parameters (same as module parameters, different interface).
539 static struct ctl_table nlm_sysctls
[] = {
541 .procname
= "nlm_grace_period",
542 .data
= &nlm_grace_period
,
543 .maxlen
= sizeof(unsigned long),
545 .proc_handler
= proc_doulongvec_minmax
,
546 .extra1
= (unsigned long *) &nlm_grace_period_min
,
547 .extra2
= (unsigned long *) &nlm_grace_period_max
,
550 .procname
= "nlm_timeout",
551 .data
= &nlm_timeout
,
552 .maxlen
= sizeof(unsigned long),
554 .proc_handler
= proc_doulongvec_minmax
,
555 .extra1
= (unsigned long *) &nlm_timeout_min
,
556 .extra2
= (unsigned long *) &nlm_timeout_max
,
559 .procname
= "nlm_udpport",
560 .data
= &nlm_udpport
,
561 .maxlen
= sizeof(int),
563 .proc_handler
= proc_dointvec_minmax
,
564 .extra1
= (int *) &nlm_port_min
,
565 .extra2
= (int *) &nlm_port_max
,
568 .procname
= "nlm_tcpport",
569 .data
= &nlm_tcpport
,
570 .maxlen
= sizeof(int),
572 .proc_handler
= proc_dointvec_minmax
,
573 .extra1
= (int *) &nlm_port_min
,
574 .extra2
= (int *) &nlm_port_max
,
577 .procname
= "nsm_use_hostnames",
578 .data
= &nsm_use_hostnames
,
579 .maxlen
= sizeof(int),
581 .proc_handler
= proc_dointvec
,
584 .procname
= "nsm_local_state",
585 .data
= &nsm_local_state
,
586 .maxlen
= sizeof(int),
588 .proc_handler
= proc_dointvec
,
593 static struct ctl_table nlm_sysctl_dir
[] = {
597 .child
= nlm_sysctls
,
602 static struct ctl_table nlm_sysctl_root
[] = {
606 .child
= nlm_sysctl_dir
,
611 #endif /* CONFIG_SYSCTL */
614 * Module (and sysfs) parameters.
617 #define param_set_min_max(name, type, which_strtol, min, max) \
618 static int param_set_##name(const char *val, const struct kernel_param *kp) \
621 __typeof__(type) num = which_strtol(val, &endp, 0); \
622 if (endp == val || *endp || num < (min) || num > (max)) \
624 *((type *) kp->arg) = num; \
628 static inline int is_callback(u32 proc
)
630 return proc
== NLMPROC_GRANTED
631 || proc
== NLMPROC_GRANTED_MSG
632 || proc
== NLMPROC_TEST_RES
633 || proc
== NLMPROC_LOCK_RES
634 || proc
== NLMPROC_CANCEL_RES
635 || proc
== NLMPROC_UNLOCK_RES
636 || proc
== NLMPROC_NSM_NOTIFY
;
640 static int lockd_authenticate(struct svc_rqst
*rqstp
)
642 rqstp
->rq_client
= NULL
;
643 switch (rqstp
->rq_authop
->flavour
) {
646 if (rqstp
->rq_proc
== 0)
648 if (is_callback(rqstp
->rq_proc
)) {
649 /* Leave it to individual procedures to
650 * call nlmsvc_lookup_host(rqstp)
654 return svc_set_client(rqstp
);
660 param_set_min_max(port
, int, simple_strtol
, 0, 65535)
661 param_set_min_max(grace_period
, unsigned long, simple_strtoul
,
662 nlm_grace_period_min
, nlm_grace_period_max
)
663 param_set_min_max(timeout
, unsigned long, simple_strtoul
,
664 nlm_timeout_min
, nlm_timeout_max
)
666 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
667 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION
".");
668 MODULE_LICENSE("GPL");
670 module_param_call(nlm_grace_period
, param_set_grace_period
, param_get_ulong
,
671 &nlm_grace_period
, 0644);
672 module_param_call(nlm_timeout
, param_set_timeout
, param_get_ulong
,
674 module_param_call(nlm_udpport
, param_set_port
, param_get_int
,
676 module_param_call(nlm_tcpport
, param_set_port
, param_get_int
,
678 module_param(nsm_use_hostnames
, bool, 0644);
679 module_param(nlm_max_connections
, uint
, 0644);
681 static int lockd_init_net(struct net
*net
)
683 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
685 INIT_DELAYED_WORK(&ln
->grace_period_end
, grace_ender
);
686 INIT_LIST_HEAD(&ln
->lockd_manager
.list
);
687 ln
->lockd_manager
.block_opens
= false;
688 INIT_LIST_HEAD(&ln
->nsm_handles
);
692 static void lockd_exit_net(struct net
*net
)
694 struct lockd_net
*ln
= net_generic(net
, lockd_net_id
);
696 WARN_ONCE(!list_empty(&ln
->lockd_manager
.list
),
697 "net %x %s: lockd_manager.list is not empty\n",
698 net
->ns
.inum
, __func__
);
699 WARN_ONCE(!list_empty(&ln
->nsm_handles
),
700 "net %x %s: nsm_handles list is not empty\n",
701 net
->ns
.inum
, __func__
);
702 WARN_ONCE(delayed_work_pending(&ln
->grace_period_end
),
703 "net %x %s: grace_period_end was not cancelled\n",
704 net
->ns
.inum
, __func__
);
707 static struct pernet_operations lockd_net_ops
= {
708 .init
= lockd_init_net
,
709 .exit
= lockd_exit_net
,
711 .size
= sizeof(struct lockd_net
),
716 * Initialising and terminating the module.
719 static int __init
init_nlm(void)
725 nlm_sysctl_table
= register_sysctl_table(nlm_sysctl_root
);
726 if (nlm_sysctl_table
== NULL
)
729 err
= register_pernet_subsys(&lockd_net_ops
);
733 err
= lockd_create_procfs();
740 unregister_pernet_subsys(&lockd_net_ops
);
743 unregister_sysctl_table(nlm_sysctl_table
);
749 static void __exit
exit_nlm(void)
751 /* FIXME: delete all NLM clients */
752 nlm_shutdown_hosts();
753 lockd_remove_procfs();
754 unregister_pernet_subsys(&lockd_net_ops
);
756 unregister_sysctl_table(nlm_sysctl_table
);
760 module_init(init_nlm
);
761 module_exit(exit_nlm
);
764 * Define NLM program and procedures
766 static unsigned int nlmsvc_version1_count
[17];
767 static const struct svc_version nlmsvc_version1
= {
770 .vs_proc
= nlmsvc_procedures
,
771 .vs_count
= nlmsvc_version1_count
,
772 .vs_xdrsize
= NLMSVC_XDRSIZE
,
774 static unsigned int nlmsvc_version3_count
[24];
775 static const struct svc_version nlmsvc_version3
= {
778 .vs_proc
= nlmsvc_procedures
,
779 .vs_count
= nlmsvc_version3_count
,
780 .vs_xdrsize
= NLMSVC_XDRSIZE
,
782 #ifdef CONFIG_LOCKD_V4
783 static unsigned int nlmsvc_version4_count
[24];
784 static const struct svc_version nlmsvc_version4
= {
787 .vs_proc
= nlmsvc_procedures4
,
788 .vs_count
= nlmsvc_version4_count
,
789 .vs_xdrsize
= NLMSVC_XDRSIZE
,
792 static const struct svc_version
*nlmsvc_version
[] = {
793 [1] = &nlmsvc_version1
,
794 [3] = &nlmsvc_version3
,
795 #ifdef CONFIG_LOCKD_V4
796 [4] = &nlmsvc_version4
,
800 static struct svc_stat nlmsvc_stats
;
802 #define NLM_NRVERS ARRAY_SIZE(nlmsvc_version)
803 static struct svc_program nlmsvc_program
= {
804 .pg_prog
= NLM_PROGRAM
, /* program number */
805 .pg_nvers
= NLM_NRVERS
, /* number of entries in nlmsvc_version */
806 .pg_vers
= nlmsvc_version
, /* version table */
807 .pg_name
= "lockd", /* service name */
808 .pg_class
= "nfsd", /* share authentication with nfsd */
809 .pg_stats
= &nlmsvc_stats
, /* stats table */
810 .pg_authenticate
= &lockd_authenticate
/* export authentication */