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?
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,
192 const struct cred *cred)
194 struct svc_xprt *xprt;
196 xprt = svc_find_xprt(serv, name, net, family, 0);
198 return svc_create_xprt(serv, name, net, family, port,
199 SVC_SOCK_DEFAULTS, cred);
204 static int create_lockd_family(struct svc_serv *serv, struct net *net,
205 const int family, const struct cred *cred)
209 err = create_lockd_listener(serv, "udp", net, family, nlm_udpport,
214 return create_lockd_listener(serv, "tcp", net, family, nlm_tcpport,
219 * Ensure there are active UDP and TCP listeners for lockd.
221 * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
222 * local services (such as rpc.statd) still require UDP, and
223 * some NFS servers do not yet support NLM over TCP.
225 * Returns zero if all listeners are available; otherwise a
226 * negative errno value is returned.
228 static int make_socks(struct svc_serv *serv, struct net *net,
229 const struct cred *cred)
234 err = create_lockd_family(serv, net, PF_INET, cred);
238 err = create_lockd_family(serv, net, PF_INET6, cred);
239 if (err < 0 && err != -EAFNOSUPPORT)
248 "lockd_up: makesock failed, error=%d\n", err);
249 svc_shutdown_net(serv, net);
253 static int lockd_up_net(struct svc_serv *serv, struct net *net,
254 const struct cred *cred)
256 struct lockd_net *ln = net_generic(net, lockd_net_id);
259 if (ln->nlmsvc_users++)
262 error = svc_bind(serv, net);
266 error = make_socks(serv, net, cred);
269 set_grace_period(net);
270 dprintk("%s: per-net data created; net=%x\n", __func__, net->ns.inum);
278 static void lockd_down_net(struct svc_serv *serv, struct net *net)
280 struct lockd_net *ln = net_generic(net, lockd_net_id);
282 if (ln->nlmsvc_users) {
283 if (--ln->nlmsvc_users == 0) {
284 nlm_shutdown_hosts_net(net);
285 cancel_delayed_work_sync(&ln->grace_period_end);
286 locks_end_grace(&ln->lockd_manager);
287 svc_shutdown_net(serv, net);
288 dprintk("%s: per-net data destroyed; net=%x\n",
289 __func__, net->ns.inum);
292 pr_err("%s: no users! task=%p, net=%x\n",
293 __func__, nlmsvc_task, net->ns.inum);
298 static int lockd_inetaddr_event(struct notifier_block *this,
299 unsigned long event, void *ptr)
301 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
302 struct sockaddr_in sin;
304 if ((event != NETDEV_DOWN) ||
305 !atomic_inc_not_zero(&nlm_ntf_refcnt))
309 dprintk("lockd_inetaddr_event: removed %pI4\n",
311 sin.sin_family = AF_INET;
312 sin.sin_addr.s_addr = ifa->ifa_local;
313 svc_age_temp_xprts_now(nlmsvc_rqst->rq_server,
314 (struct sockaddr *)&sin);
316 atomic_dec(&nlm_ntf_refcnt);
317 wake_up(&nlm_ntf_wq);
323 static struct notifier_block lockd_inetaddr_notifier = {
324 .notifier_call = lockd_inetaddr_event,
327 #if IS_ENABLED(CONFIG_IPV6)
328 static int lockd_inet6addr_event(struct notifier_block *this,
329 unsigned long event, void *ptr)
331 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
332 struct sockaddr_in6 sin6;
334 if ((event != NETDEV_DOWN) ||
335 !atomic_inc_not_zero(&nlm_ntf_refcnt))
339 dprintk("lockd_inet6addr_event: removed %pI6\n", &ifa->addr);
340 sin6.sin6_family = AF_INET6;
341 sin6.sin6_addr = ifa->addr;
342 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
343 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
344 svc_age_temp_xprts_now(nlmsvc_rqst->rq_server,
345 (struct sockaddr *)&sin6);
347 atomic_dec(&nlm_ntf_refcnt);
348 wake_up(&nlm_ntf_wq);
354 static struct notifier_block lockd_inet6addr_notifier = {
355 .notifier_call = lockd_inet6addr_event,
359 static void lockd_unregister_notifiers(void)
361 unregister_inetaddr_notifier(&lockd_inetaddr_notifier);
362 #if IS_ENABLED(CONFIG_IPV6)
363 unregister_inet6addr_notifier(&lockd_inet6addr_notifier);
365 wait_event(nlm_ntf_wq, atomic_read(&nlm_ntf_refcnt) == 0);
368 static void lockd_svc_exit_thread(void)
370 atomic_dec(&nlm_ntf_refcnt);
371 lockd_unregister_notifiers();
372 svc_exit_thread(nlmsvc_rqst);
375 static int lockd_start_svc(struct svc_serv *serv)
383 * Create the kernel thread and wait for it to start.
385 nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
386 if (IS_ERR(nlmsvc_rqst)) {
387 error = PTR_ERR(nlmsvc_rqst);
389 "lockd_up: svc_rqst allocation failed, error=%d\n",
391 lockd_unregister_notifiers();
395 atomic_inc(&nlm_ntf_refcnt);
396 svc_sock_update_bufs(serv);
397 serv->sv_maxconn = nlm_max_connections;
399 nlmsvc_task = kthread_create(lockd, nlmsvc_rqst, "%s", serv->sv_name);
400 if (IS_ERR(nlmsvc_task)) {
401 error = PTR_ERR(nlmsvc_task);
403 "lockd_up: kthread_run failed, error=%d\n", error);
406 nlmsvc_rqst->rq_task = nlmsvc_task;
407 wake_up_process(nlmsvc_task);
409 dprintk("lockd_up: service started\n");
413 lockd_svc_exit_thread();
420 static const struct svc_serv_ops lockd_sv_ops = {
421 .svo_shutdown = svc_rpcb_cleanup,
422 .svo_enqueue_xprt = svc_xprt_do_enqueue,
425 static struct svc_serv *lockd_create_svc(void)
427 struct svc_serv *serv;
430 * Check whether we're already up and running.
434 * Note: increase service usage, because later in case of error
435 * svc_destroy() will be called.
437 svc_get(nlmsvc_rqst->rq_server);
438 return nlmsvc_rqst->rq_server;
442 * Sanity check: if there's no pid,
443 * we should be the first user ...
447 "lockd_up: no pid, %d users??\n", nlmsvc_users);
450 nlm_timeout = LOCKD_DFLT_TIMEO;
451 nlmsvc_timeout = nlm_timeout * HZ;
453 serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, &lockd_sv_ops);
455 printk(KERN_WARNING "lockd_up: create service failed\n");
456 return ERR_PTR(-ENOMEM);
458 register_inetaddr_notifier(&lockd_inetaddr_notifier);
459 #if IS_ENABLED(CONFIG_IPV6)
460 register_inet6addr_notifier(&lockd_inet6addr_notifier);
462 dprintk("lockd_up: service created\n");
467 * Bring up the lockd process if it's not already up.
469 int lockd_up(struct net *net, const struct cred *cred)
471 struct svc_serv *serv;
474 mutex_lock(&nlmsvc_mutex);
476 serv = lockd_create_svc();
478 error = PTR_ERR(serv);
482 error = lockd_up_net(serv, net, cred);
484 lockd_unregister_notifiers();
488 error = lockd_start_svc(serv);
490 lockd_down_net(serv, net);
495 * Note: svc_serv structures have an initial use count of 1,
496 * so we exit through here on both success and failure.
501 mutex_unlock(&nlmsvc_mutex);
504 EXPORT_SYMBOL_GPL(lockd_up);
507 * Decrement the user count and bring down lockd if we're the last.
510 lockd_down(struct net *net)
512 mutex_lock(&nlmsvc_mutex);
513 lockd_down_net(nlmsvc_rqst->rq_server, net);
518 printk(KERN_ERR "lockd_down: no users! task=%p\n",
524 printk(KERN_ERR "lockd_down: no lockd running.\n");
527 kthread_stop(nlmsvc_task);
528 dprintk("lockd_down: service stopped\n");
529 lockd_svc_exit_thread();
530 dprintk("lockd_down: service destroyed\n");
534 mutex_unlock(&nlmsvc_mutex);
536 EXPORT_SYMBOL_GPL(lockd_down);
541 * Sysctl parameters (same as module parameters, different interface).
544 static struct ctl_table nlm_sysctls[] = {
546 .procname = "nlm_grace_period",
547 .data = &nlm_grace_period,
548 .maxlen = sizeof(unsigned long),
550 .proc_handler = proc_doulongvec_minmax,
551 .extra1 = (unsigned long *) &nlm_grace_period_min,
552 .extra2 = (unsigned long *) &nlm_grace_period_max,
555 .procname = "nlm_timeout",
556 .data = &nlm_timeout,
557 .maxlen = sizeof(unsigned long),
559 .proc_handler = proc_doulongvec_minmax,
560 .extra1 = (unsigned long *) &nlm_timeout_min,
561 .extra2 = (unsigned long *) &nlm_timeout_max,
564 .procname = "nlm_udpport",
565 .data = &nlm_udpport,
566 .maxlen = sizeof(int),
568 .proc_handler = proc_dointvec_minmax,
569 .extra1 = (int *) &nlm_port_min,
570 .extra2 = (int *) &nlm_port_max,
573 .procname = "nlm_tcpport",
574 .data = &nlm_tcpport,
575 .maxlen = sizeof(int),
577 .proc_handler = proc_dointvec_minmax,
578 .extra1 = (int *) &nlm_port_min,
579 .extra2 = (int *) &nlm_port_max,
582 .procname = "nsm_use_hostnames",
583 .data = &nsm_use_hostnames,
584 .maxlen = sizeof(int),
586 .proc_handler = proc_dointvec,
589 .procname = "nsm_local_state",
590 .data = &nsm_local_state,
591 .maxlen = sizeof(int),
593 .proc_handler = proc_dointvec,
598 static struct ctl_table nlm_sysctl_dir[] = {
602 .child = nlm_sysctls,
607 static struct ctl_table nlm_sysctl_root[] = {
611 .child = nlm_sysctl_dir,
616 #endif /* CONFIG_SYSCTL */
619 * Module (and sysfs) parameters.
622 #define param_set_min_max(name, type, which_strtol, min, max) \
623 static int param_set_##name(const char *val, const struct kernel_param *kp) \
626 __typeof__(type) num = which_strtol(val, &endp, 0); \
627 if (endp == val || *endp || num < (min) || num > (max)) \
629 *((type *) kp->arg) = num; \
633 static inline int is_callback(u32 proc)
635 return proc == NLMPROC_GRANTED
636 || proc == NLMPROC_GRANTED_MSG
637 || proc == NLMPROC_TEST_RES
638 || proc == NLMPROC_LOCK_RES
639 || proc == NLMPROC_CANCEL_RES
640 || proc == NLMPROC_UNLOCK_RES
641 || proc == NLMPROC_NSM_NOTIFY;
645 static int lockd_authenticate(struct svc_rqst *rqstp)
647 rqstp->rq_client = NULL;
648 switch (rqstp->rq_authop->flavour) {
651 if (rqstp->rq_proc == 0)
653 if (is_callback(rqstp->rq_proc)) {
654 /* Leave it to individual procedures to
655 * call nlmsvc_lookup_host(rqstp)
659 return svc_set_client(rqstp);
665 param_set_min_max(port, int, simple_strtol, 0, 65535)
666 param_set_min_max(grace_period, unsigned long, simple_strtoul,
667 nlm_grace_period_min, nlm_grace_period_max)
668 param_set_min_max(timeout, unsigned long, simple_strtoul,
669 nlm_timeout_min, nlm_timeout_max)
672 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
673 MODULE_LICENSE("GPL");
675 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
676 &nlm_grace_period, 0644);
677 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
679 module_param_call(nlm_udpport, param_set_port, param_get_int,
681 module_param_call(nlm_tcpport, param_set_port, param_get_int,
683 module_param(nsm_use_hostnames, bool, 0644);
684 module_param(nlm_max_connections, uint, 0644);
686 static int lockd_init_net(struct net *net)
688 struct lockd_net *ln = net_generic(net, lockd_net_id);
690 INIT_DELAYED_WORK(&ln->grace_period_end, grace_ender);
691 INIT_LIST_HEAD(&ln->lockd_manager.list);
692 ln->lockd_manager.block_opens = false;
693 INIT_LIST_HEAD(&ln->nsm_handles);
697 static void lockd_exit_net(struct net *net)
699 struct lockd_net *ln = net_generic(net, lockd_net_id);
701 WARN_ONCE(!list_empty(&ln->lockd_manager.list),
702 "net %x %s: lockd_manager.list is not empty\n",
703 net->ns.inum, __func__);
704 WARN_ONCE(!list_empty(&ln->nsm_handles),
705 "net %x %s: nsm_handles list is not empty\n",
706 net->ns.inum, __func__);
707 WARN_ONCE(delayed_work_pending(&ln->grace_period_end),
708 "net %x %s: grace_period_end was not cancelled\n",
709 net->ns.inum, __func__);
712 static struct pernet_operations lockd_net_ops = {
713 .init = lockd_init_net,
714 .exit = lockd_exit_net,
716 .size = sizeof(struct lockd_net),
721 * Initialising and terminating the module.
724 static int __init init_nlm(void)
730 nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
731 if (nlm_sysctl_table == NULL)
734 err = register_pernet_subsys(&lockd_net_ops);
738 err = lockd_create_procfs();
745 unregister_pernet_subsys(&lockd_net_ops);
748 unregister_sysctl_table(nlm_sysctl_table);
754 static void __exit exit_nlm(void)
756 /* FIXME: delete all NLM clients */
757 nlm_shutdown_hosts();
758 lockd_remove_procfs();
759 unregister_pernet_subsys(&lockd_net_ops);
761 unregister_sysctl_table(nlm_sysctl_table);
765 module_init(init_nlm);
766 module_exit(exit_nlm);
769 * Define NLM program and procedures
771 static unsigned int nlmsvc_version1_count[17];
772 static const struct svc_version nlmsvc_version1 = {
775 .vs_proc = nlmsvc_procedures,
776 .vs_count = nlmsvc_version1_count,
777 .vs_xdrsize = NLMSVC_XDRSIZE,
779 static unsigned int nlmsvc_version3_count[24];
780 static const struct svc_version nlmsvc_version3 = {
783 .vs_proc = nlmsvc_procedures,
784 .vs_count = nlmsvc_version3_count,
785 .vs_xdrsize = NLMSVC_XDRSIZE,
787 #ifdef CONFIG_LOCKD_V4
788 static unsigned int nlmsvc_version4_count[24];
789 static const struct svc_version nlmsvc_version4 = {
792 .vs_proc = nlmsvc_procedures4,
793 .vs_count = nlmsvc_version4_count,
794 .vs_xdrsize = NLMSVC_XDRSIZE,
797 static const struct svc_version *nlmsvc_version[] = {
798 [1] = &nlmsvc_version1,
799 [3] = &nlmsvc_version3,
800 #ifdef CONFIG_LOCKD_V4
801 [4] = &nlmsvc_version4,
805 static struct svc_stat nlmsvc_stats;
807 #define NLM_NRVERS ARRAY_SIZE(nlmsvc_version)
808 static struct svc_program nlmsvc_program = {
809 .pg_prog = NLM_PROGRAM, /* program number */
810 .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */
811 .pg_vers = nlmsvc_version, /* version table */
812 .pg_name = "lockd", /* service name */
813 .pg_class = "nfsd", /* share authentication with nfsd */
814 .pg_stats = &nlmsvc_stats, /* stats table */
815 .pg_authenticate = &lockd_authenticate, /* export authentication */
816 .pg_init_request = svc_generic_init_request,
817 .pg_rpcbind_set = svc_generic_rpcbind_set,