1 // SPDX-License-Identifier: GPL-2.0-only
5 * This is the central lockd service.
7 * FIXME: Separate the lockd NFS server functionality from the lockd NFS
8 * client functionality. Oh why didn't Sun create two separate
9 * services in the first place?
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sysctl.h>
19 #include <linux/moduleparam.h>
21 #include <linux/sched/signal.h>
22 #include <linux/errno.h>
24 #include <linux/uio.h>
25 #include <linux/smp.h>
26 #include <linux/mutex.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/inetdevice.h>
31 #include <linux/sunrpc/types.h>
32 #include <linux/sunrpc/stats.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/svc.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <linux/sunrpc/svc_xprt.h>
38 #include <net/addrconf.h>
40 #include <linux/lockd/lockd.h>
41 #include <linux/nfs.h>
46 #define NLMDBG_FACILITY NLMDBG_SVC
47 #define LOCKD_BUFSIZE (1024 + NLMSVC_XDRSIZE)
48 #define ALLOWED_SIGS (sigmask(SIGKILL))
50 static struct svc_program nlmsvc_program;
52 const struct nlmsvc_binding *nlmsvc_ops;
53 EXPORT_SYMBOL_GPL(nlmsvc_ops);
55 static DEFINE_MUTEX(nlmsvc_mutex);
56 static unsigned int nlmsvc_users;
57 static struct task_struct *nlmsvc_task;
58 static struct svc_rqst *nlmsvc_rqst;
59 unsigned long nlmsvc_timeout;
61 static atomic_t nlm_ntf_refcnt = ATOMIC_INIT(0);
62 static DECLARE_WAIT_QUEUE_HEAD(nlm_ntf_wq);
64 unsigned int lockd_net_id;
67 * These can be set at insmod time (useful for NFS as root filesystem),
68 * and also changed through the sysctl interface. -- Jamie Lokier, Aug 2003
70 static unsigned long nlm_grace_period;
71 static unsigned long nlm_timeout = LOCKD_DFLT_TIMEO;
72 static int nlm_udpport, nlm_tcpport;
74 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
75 static unsigned int nlm_max_connections = 1024;
78 * Constants needed for the sysctl interface.
80 static const unsigned long nlm_grace_period_min = 0;
81 static const unsigned long nlm_grace_period_max = 240;
82 static const unsigned long nlm_timeout_min = 3;
83 static const unsigned long nlm_timeout_max = 20;
84 static const int nlm_port_min = 0, nlm_port_max = 65535;
87 static struct ctl_table_header * nlm_sysctl_table;
90 static unsigned long get_lockd_grace_period(void)
92 /* Note: nlm_timeout should always be nonzero */
94 return roundup(nlm_grace_period, nlm_timeout) * HZ;
96 return nlm_timeout * 5 * HZ;
99 static void grace_ender(struct work_struct *grace)
101 struct delayed_work *dwork = to_delayed_work(grace);
102 struct lockd_net *ln = container_of(dwork, struct lockd_net,
105 locks_end_grace(&ln->lockd_manager);
108 static void set_grace_period(struct net *net)
110 unsigned long grace_period = get_lockd_grace_period();
111 struct lockd_net *ln = net_generic(net, lockd_net_id);
113 locks_start_grace(net, &ln->lockd_manager);
114 cancel_delayed_work_sync(&ln->grace_period_end);
115 schedule_delayed_work(&ln->grace_period_end, grace_period);
118 static void restart_grace(void)
121 struct net *net = &init_net;
122 struct lockd_net *ln = net_generic(net, lockd_net_id);
124 cancel_delayed_work_sync(&ln->grace_period_end);
125 locks_end_grace(&ln->lockd_manager);
126 nlmsvc_invalidate_all();
127 set_grace_period(net);
132 * This is the lockd kernel thread
138 struct svc_rqst *rqstp = vrqstp;
139 struct net *net = &init_net;
140 struct lockd_net *ln = net_generic(net, lockd_net_id);
142 /* try_to_freeze() is called from svc_recv() */
145 /* Allow SIGKILL to tell lockd to drop all of its locks */
146 allow_signal(SIGKILL);
148 dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
151 * The main request loop. We don't terminate until the last
152 * NFS mount or NFS daemon has gone away.
154 while (!kthread_should_stop()) {
155 long timeout = MAX_SCHEDULE_TIMEOUT;
156 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
158 /* update sv_maxconn if it has changed */
159 rqstp->rq_server->sv_maxconn = nlm_max_connections;
162 flush_signals(current);
167 timeout = nlmsvc_retry_blocked();
170 * Find a socket with data available and call its
173 err = svc_recv(rqstp, timeout);
174 if (err == -EAGAIN || err == -EINTR)
176 dprintk("lockd: request from %s\n",
177 svc_print_addr(rqstp, buf, sizeof(buf)));
181 flush_signals(current);
183 nlmsvc_invalidate_all();
184 nlm_shutdown_hosts();
185 cancel_delayed_work_sync(&ln->grace_period_end);
186 locks_end_grace(&ln->lockd_manager);
190 static int create_lockd_listener(struct svc_serv *serv, const char *name,
191 struct net *net, const int family,
192 const unsigned short port,
193 const struct cred *cred)
195 struct svc_xprt *xprt;
197 xprt = svc_find_xprt(serv, name, net, family, 0);
199 return svc_create_xprt(serv, name, net, family, port,
200 SVC_SOCK_DEFAULTS, cred);
205 static int create_lockd_family(struct svc_serv *serv, struct net *net,
206 const int family, const struct cred *cred)
210 err = create_lockd_listener(serv, "udp", net, family, nlm_udpport,
215 return create_lockd_listener(serv, "tcp", net, family, nlm_tcpport,
220 * Ensure there are active UDP and TCP listeners for lockd.
222 * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
223 * local services (such as rpc.statd) still require UDP, and
224 * some NFS servers do not yet support NLM over TCP.
226 * Returns zero if all listeners are available; otherwise a
227 * negative errno value is returned.
229 static int make_socks(struct svc_serv *serv, struct net *net,
230 const struct cred *cred)
235 err = create_lockd_family(serv, net, PF_INET, cred);
239 err = create_lockd_family(serv, net, PF_INET6, cred);
240 if (err < 0 && err != -EAFNOSUPPORT)
249 "lockd_up: makesock failed, error=%d\n", err);
250 svc_shutdown_net(serv, net);
254 static int lockd_up_net(struct svc_serv *serv, struct net *net,
255 const struct cred *cred)
257 struct lockd_net *ln = net_generic(net, lockd_net_id);
260 if (ln->nlmsvc_users++)
263 error = svc_bind(serv, net);
267 error = make_socks(serv, net, cred);
270 set_grace_period(net);
271 dprintk("%s: per-net data created; net=%x\n", __func__, net->ns.inum);
279 static void lockd_down_net(struct svc_serv *serv, struct net *net)
281 struct lockd_net *ln = net_generic(net, lockd_net_id);
283 if (ln->nlmsvc_users) {
284 if (--ln->nlmsvc_users == 0) {
285 nlm_shutdown_hosts_net(net);
286 cancel_delayed_work_sync(&ln->grace_period_end);
287 locks_end_grace(&ln->lockd_manager);
288 svc_shutdown_net(serv, net);
289 dprintk("%s: per-net data destroyed; net=%x\n",
290 __func__, net->ns.inum);
293 pr_err("%s: no users! task=%p, net=%x\n",
294 __func__, nlmsvc_task, net->ns.inum);
299 static int lockd_inetaddr_event(struct notifier_block *this,
300 unsigned long event, void *ptr)
302 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
303 struct sockaddr_in sin;
305 if ((event != NETDEV_DOWN) ||
306 !atomic_inc_not_zero(&nlm_ntf_refcnt))
310 dprintk("lockd_inetaddr_event: removed %pI4\n",
312 sin.sin_family = AF_INET;
313 sin.sin_addr.s_addr = ifa->ifa_local;
314 svc_age_temp_xprts_now(nlmsvc_rqst->rq_server,
315 (struct sockaddr *)&sin);
317 atomic_dec(&nlm_ntf_refcnt);
318 wake_up(&nlm_ntf_wq);
324 static struct notifier_block lockd_inetaddr_notifier = {
325 .notifier_call = lockd_inetaddr_event,
328 #if IS_ENABLED(CONFIG_IPV6)
329 static int lockd_inet6addr_event(struct notifier_block *this,
330 unsigned long event, void *ptr)
332 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
333 struct sockaddr_in6 sin6;
335 if ((event != NETDEV_DOWN) ||
336 !atomic_inc_not_zero(&nlm_ntf_refcnt))
340 dprintk("lockd_inet6addr_event: removed %pI6\n", &ifa->addr);
341 sin6.sin6_family = AF_INET6;
342 sin6.sin6_addr = ifa->addr;
343 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
344 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
345 svc_age_temp_xprts_now(nlmsvc_rqst->rq_server,
346 (struct sockaddr *)&sin6);
348 atomic_dec(&nlm_ntf_refcnt);
349 wake_up(&nlm_ntf_wq);
355 static struct notifier_block lockd_inet6addr_notifier = {
356 .notifier_call = lockd_inet6addr_event,
360 static void lockd_unregister_notifiers(void)
362 unregister_inetaddr_notifier(&lockd_inetaddr_notifier);
363 #if IS_ENABLED(CONFIG_IPV6)
364 unregister_inet6addr_notifier(&lockd_inet6addr_notifier);
366 wait_event(nlm_ntf_wq, atomic_read(&nlm_ntf_refcnt) == 0);
369 static void lockd_svc_exit_thread(void)
371 atomic_dec(&nlm_ntf_refcnt);
372 lockd_unregister_notifiers();
373 svc_exit_thread(nlmsvc_rqst);
376 static int lockd_start_svc(struct svc_serv *serv)
384 * Create the kernel thread and wait for it to start.
386 nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
387 if (IS_ERR(nlmsvc_rqst)) {
388 error = PTR_ERR(nlmsvc_rqst);
390 "lockd_up: svc_rqst allocation failed, error=%d\n",
392 lockd_unregister_notifiers();
396 atomic_inc(&nlm_ntf_refcnt);
397 svc_sock_update_bufs(serv);
398 serv->sv_maxconn = nlm_max_connections;
400 nlmsvc_task = kthread_create(lockd, nlmsvc_rqst, "%s", serv->sv_name);
401 if (IS_ERR(nlmsvc_task)) {
402 error = PTR_ERR(nlmsvc_task);
404 "lockd_up: kthread_run failed, error=%d\n", error);
407 nlmsvc_rqst->rq_task = nlmsvc_task;
408 wake_up_process(nlmsvc_task);
410 dprintk("lockd_up: service started\n");
414 lockd_svc_exit_thread();
421 static const struct svc_serv_ops lockd_sv_ops = {
422 .svo_shutdown = svc_rpcb_cleanup,
423 .svo_enqueue_xprt = svc_xprt_do_enqueue,
426 static struct svc_serv *lockd_create_svc(void)
428 struct svc_serv *serv;
431 * Check whether we're already up and running.
435 * Note: increase service usage, because later in case of error
436 * svc_destroy() will be called.
438 svc_get(nlmsvc_rqst->rq_server);
439 return nlmsvc_rqst->rq_server;
443 * Sanity check: if there's no pid,
444 * we should be the first user ...
448 "lockd_up: no pid, %d users??\n", nlmsvc_users);
451 nlm_timeout = LOCKD_DFLT_TIMEO;
452 nlmsvc_timeout = nlm_timeout * HZ;
454 serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, &lockd_sv_ops);
456 printk(KERN_WARNING "lockd_up: create service failed\n");
457 return ERR_PTR(-ENOMEM);
459 register_inetaddr_notifier(&lockd_inetaddr_notifier);
460 #if IS_ENABLED(CONFIG_IPV6)
461 register_inet6addr_notifier(&lockd_inet6addr_notifier);
463 dprintk("lockd_up: service created\n");
468 * Bring up the lockd process if it's not already up.
470 int lockd_up(struct net *net, const struct cred *cred)
472 struct svc_serv *serv;
475 mutex_lock(&nlmsvc_mutex);
477 serv = lockd_create_svc();
479 error = PTR_ERR(serv);
483 error = lockd_up_net(serv, net, cred);
485 lockd_unregister_notifiers();
489 error = lockd_start_svc(serv);
491 lockd_down_net(serv, net);
496 * Note: svc_serv structures have an initial use count of 1,
497 * so we exit through here on both success and failure.
502 mutex_unlock(&nlmsvc_mutex);
505 EXPORT_SYMBOL_GPL(lockd_up);
508 * Decrement the user count and bring down lockd if we're the last.
511 lockd_down(struct net *net)
513 mutex_lock(&nlmsvc_mutex);
514 lockd_down_net(nlmsvc_rqst->rq_server, net);
519 printk(KERN_ERR "lockd_down: no users! task=%p\n",
525 printk(KERN_ERR "lockd_down: no lockd running.\n");
528 kthread_stop(nlmsvc_task);
529 dprintk("lockd_down: service stopped\n");
530 lockd_svc_exit_thread();
531 dprintk("lockd_down: service destroyed\n");
535 mutex_unlock(&nlmsvc_mutex);
537 EXPORT_SYMBOL_GPL(lockd_down);
542 * Sysctl parameters (same as module parameters, different interface).
545 static struct ctl_table nlm_sysctls[] = {
547 .procname = "nlm_grace_period",
548 .data = &nlm_grace_period,
549 .maxlen = sizeof(unsigned long),
551 .proc_handler = proc_doulongvec_minmax,
552 .extra1 = (unsigned long *) &nlm_grace_period_min,
553 .extra2 = (unsigned long *) &nlm_grace_period_max,
556 .procname = "nlm_timeout",
557 .data = &nlm_timeout,
558 .maxlen = sizeof(unsigned long),
560 .proc_handler = proc_doulongvec_minmax,
561 .extra1 = (unsigned long *) &nlm_timeout_min,
562 .extra2 = (unsigned long *) &nlm_timeout_max,
565 .procname = "nlm_udpport",
566 .data = &nlm_udpport,
567 .maxlen = sizeof(int),
569 .proc_handler = proc_dointvec_minmax,
570 .extra1 = (int *) &nlm_port_min,
571 .extra2 = (int *) &nlm_port_max,
574 .procname = "nlm_tcpport",
575 .data = &nlm_tcpport,
576 .maxlen = sizeof(int),
578 .proc_handler = proc_dointvec_minmax,
579 .extra1 = (int *) &nlm_port_min,
580 .extra2 = (int *) &nlm_port_max,
583 .procname = "nsm_use_hostnames",
584 .data = &nsm_use_hostnames,
585 .maxlen = sizeof(int),
587 .proc_handler = proc_dointvec,
590 .procname = "nsm_local_state",
591 .data = &nsm_local_state,
592 .maxlen = sizeof(int),
594 .proc_handler = proc_dointvec,
599 static struct ctl_table nlm_sysctl_dir[] = {
603 .child = nlm_sysctls,
608 static struct ctl_table nlm_sysctl_root[] = {
612 .child = nlm_sysctl_dir,
617 #endif /* CONFIG_SYSCTL */
620 * Module (and sysfs) parameters.
623 #define param_set_min_max(name, type, which_strtol, min, max) \
624 static int param_set_##name(const char *val, const struct kernel_param *kp) \
627 __typeof__(type) num = which_strtol(val, &endp, 0); \
628 if (endp == val || *endp || num < (min) || num > (max)) \
630 *((type *) kp->arg) = num; \
634 static inline int is_callback(u32 proc)
636 return proc == NLMPROC_GRANTED
637 || proc == NLMPROC_GRANTED_MSG
638 || proc == NLMPROC_TEST_RES
639 || proc == NLMPROC_LOCK_RES
640 || proc == NLMPROC_CANCEL_RES
641 || proc == NLMPROC_UNLOCK_RES
642 || proc == NLMPROC_NSM_NOTIFY;
646 static int lockd_authenticate(struct svc_rqst *rqstp)
648 rqstp->rq_client = NULL;
649 switch (rqstp->rq_authop->flavour) {
652 if (rqstp->rq_proc == 0)
654 if (is_callback(rqstp->rq_proc)) {
655 /* Leave it to individual procedures to
656 * call nlmsvc_lookup_host(rqstp)
660 return svc_set_client(rqstp);
666 param_set_min_max(port, int, simple_strtol, 0, 65535)
667 param_set_min_max(grace_period, unsigned long, simple_strtoul,
668 nlm_grace_period_min, nlm_grace_period_max)
669 param_set_min_max(timeout, unsigned long, simple_strtoul,
670 nlm_timeout_min, nlm_timeout_max)
673 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
674 MODULE_LICENSE("GPL");
676 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
677 &nlm_grace_period, 0644);
678 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
680 module_param_call(nlm_udpport, param_set_port, param_get_int,
682 module_param_call(nlm_tcpport, param_set_port, param_get_int,
684 module_param(nsm_use_hostnames, bool, 0644);
685 module_param(nlm_max_connections, uint, 0644);
687 static int lockd_init_net(struct net *net)
689 struct lockd_net *ln = net_generic(net, lockd_net_id);
691 INIT_DELAYED_WORK(&ln->grace_period_end, grace_ender);
692 INIT_LIST_HEAD(&ln->lockd_manager.list);
693 ln->lockd_manager.block_opens = false;
694 INIT_LIST_HEAD(&ln->nsm_handles);
698 static void lockd_exit_net(struct net *net)
700 struct lockd_net *ln = net_generic(net, lockd_net_id);
702 WARN_ONCE(!list_empty(&ln->lockd_manager.list),
703 "net %x %s: lockd_manager.list is not empty\n",
704 net->ns.inum, __func__);
705 WARN_ONCE(!list_empty(&ln->nsm_handles),
706 "net %x %s: nsm_handles list is not empty\n",
707 net->ns.inum, __func__);
708 WARN_ONCE(delayed_work_pending(&ln->grace_period_end),
709 "net %x %s: grace_period_end was not cancelled\n",
710 net->ns.inum, __func__);
713 static struct pernet_operations lockd_net_ops = {
714 .init = lockd_init_net,
715 .exit = lockd_exit_net,
717 .size = sizeof(struct lockd_net),
722 * Initialising and terminating the module.
725 static int __init init_nlm(void)
731 nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
732 if (nlm_sysctl_table == NULL)
735 err = register_pernet_subsys(&lockd_net_ops);
739 err = lockd_create_procfs();
746 unregister_pernet_subsys(&lockd_net_ops);
749 unregister_sysctl_table(nlm_sysctl_table);
755 static void __exit exit_nlm(void)
757 /* FIXME: delete all NLM clients */
758 nlm_shutdown_hosts();
759 lockd_remove_procfs();
760 unregister_pernet_subsys(&lockd_net_ops);
762 unregister_sysctl_table(nlm_sysctl_table);
766 module_init(init_nlm);
767 module_exit(exit_nlm);
770 * Define NLM program and procedures
772 static unsigned int nlmsvc_version1_count[17];
773 static const struct svc_version nlmsvc_version1 = {
776 .vs_proc = nlmsvc_procedures,
777 .vs_count = nlmsvc_version1_count,
778 .vs_xdrsize = NLMSVC_XDRSIZE,
780 static unsigned int nlmsvc_version3_count[24];
781 static const struct svc_version nlmsvc_version3 = {
784 .vs_proc = nlmsvc_procedures,
785 .vs_count = nlmsvc_version3_count,
786 .vs_xdrsize = NLMSVC_XDRSIZE,
788 #ifdef CONFIG_LOCKD_V4
789 static unsigned int nlmsvc_version4_count[24];
790 static const struct svc_version nlmsvc_version4 = {
793 .vs_proc = nlmsvc_procedures4,
794 .vs_count = nlmsvc_version4_count,
795 .vs_xdrsize = NLMSVC_XDRSIZE,
798 static const struct svc_version *nlmsvc_version[] = {
799 [1] = &nlmsvc_version1,
800 [3] = &nlmsvc_version3,
801 #ifdef CONFIG_LOCKD_V4
802 [4] = &nlmsvc_version4,
806 static struct svc_stat nlmsvc_stats;
808 #define NLM_NRVERS ARRAY_SIZE(nlmsvc_version)
809 static struct svc_program nlmsvc_program = {
810 .pg_prog = NLM_PROGRAM, /* program number */
811 .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */
812 .pg_vers = nlmsvc_version, /* version table */
813 .pg_name = "lockd", /* service name */
814 .pg_class = "nfsd", /* share authentication with nfsd */
815 .pg_stats = &nlmsvc_stats, /* stats table */
816 .pg_authenticate = &lockd_authenticate, /* export authentication */
817 .pg_init_request = svc_generic_init_request,
818 .pg_rpcbind_set = svc_generic_rpcbind_set,