Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/lockd/svc.c
4 *
5 * This is the central lockd service.
6 *
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?
10 *
11 * Authors: Olaf Kirch (okir@monad.swb.de)
12 *
13 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
14 */
15
16#include <linux/module.h>
17#include <linux/init.h>
18#include <linux/sysctl.h>
19#include <linux/moduleparam.h>
20
21#include <linux/sched/signal.h>
22#include <linux/errno.h>
23#include <linux/in.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>
30
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>
37#include <net/ip.h>
38#include <net/addrconf.h>
39#include <net/ipv6.h>
40#include <linux/lockd/lockd.h>
41#include <linux/nfs.h>
42
43#include "netns.h"
44#include "procfs.h"
45
46#define NLMDBG_FACILITY NLMDBG_SVC
47#define LOCKD_BUFSIZE (1024 + NLMSVC_XDRSIZE)
48#define ALLOWED_SIGS (sigmask(SIGKILL))
49
50static struct svc_program nlmsvc_program;
51
52const struct nlmsvc_binding *nlmsvc_ops;
53EXPORT_SYMBOL_GPL(nlmsvc_ops);
54
55static DEFINE_MUTEX(nlmsvc_mutex);
56static unsigned int nlmsvc_users;
57static struct task_struct *nlmsvc_task;
58static struct svc_rqst *nlmsvc_rqst;
59unsigned long nlmsvc_timeout;
60
61static atomic_t nlm_ntf_refcnt = ATOMIC_INIT(0);
62static DECLARE_WAIT_QUEUE_HEAD(nlm_ntf_wq);
63
64unsigned int lockd_net_id;
65
66/*
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
69 */
70static unsigned long nlm_grace_period;
71static unsigned long nlm_timeout = LOCKD_DFLT_TIMEO;
72static int nlm_udpport, nlm_tcpport;
73
74/* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
75static unsigned int nlm_max_connections = 1024;
76
77/*
78 * Constants needed for the sysctl interface.
79 */
80static const unsigned long nlm_grace_period_min = 0;
81static const unsigned long nlm_grace_period_max = 240;
82static const unsigned long nlm_timeout_min = 3;
83static const unsigned long nlm_timeout_max = 20;
84static const int nlm_port_min = 0, nlm_port_max = 65535;
85
86#ifdef CONFIG_SYSCTL
87static struct ctl_table_header * nlm_sysctl_table;
88#endif
89
90static unsigned long get_lockd_grace_period(void)
91{
92 /* Note: nlm_timeout should always be nonzero */
93 if (nlm_grace_period)
94 return roundup(nlm_grace_period, nlm_timeout) * HZ;
95 else
96 return nlm_timeout * 5 * HZ;
97}
98
99static void grace_ender(struct work_struct *grace)
100{
101 struct delayed_work *dwork = to_delayed_work(grace);
102 struct lockd_net *ln = container_of(dwork, struct lockd_net,
103 grace_period_end);
104
105 locks_end_grace(&ln->lockd_manager);
106}
107
108static void set_grace_period(struct net *net)
109{
110 unsigned long grace_period = get_lockd_grace_period();
111 struct lockd_net *ln = net_generic(net, lockd_net_id);
112
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);
116}
117
118static void restart_grace(void)
119{
120 if (nlmsvc_ops) {
121 struct net *net = &init_net;
122 struct lockd_net *ln = net_generic(net, lockd_net_id);
123
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);
128 }
129}
130
131/*
132 * This is the lockd kernel thread
133 */
134static int
135lockd(void *vrqstp)
136{
137 int err = 0;
138 struct svc_rqst *rqstp = vrqstp;
139 struct net *net = &init_net;
140 struct lockd_net *ln = net_generic(net, lockd_net_id);
141
142 /* try_to_freeze() is called from svc_recv() */
143 set_freezable();
144
145 /* Allow SIGKILL to tell lockd to drop all of its locks */
146 allow_signal(SIGKILL);
147
148 dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
149
150 /*
151 * The main request loop. We don't terminate until the last
152 * NFS mount or NFS daemon has gone away.
153 */
154 while (!kthread_should_stop()) {
155 long timeout = MAX_SCHEDULE_TIMEOUT;
156 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
157
158 /* update sv_maxconn if it has changed */
159 rqstp->rq_server->sv_maxconn = nlm_max_connections;
160
161 if (signalled()) {
162 flush_signals(current);
163 restart_grace();
164 continue;
165 }
166
167 timeout = nlmsvc_retry_blocked();
168
169 /*
170 * Find a socket with data available and call its
171 * recvfrom routine.
172 */
173 err = svc_recv(rqstp, timeout);
174 if (err == -EAGAIN || err == -EINTR)
175 continue;
176 dprintk("lockd: request from %s\n",
177 svc_print_addr(rqstp, buf, sizeof(buf)));
178
179 svc_process(rqstp);
180 }
181 flush_signals(current);
182 if (nlmsvc_ops)
183 nlmsvc_invalidate_all();
184 nlm_shutdown_hosts();
185 cancel_delayed_work_sync(&ln->grace_period_end);
186 locks_end_grace(&ln->lockd_manager);
187 return 0;
188}
189
190static 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)
194{
195 struct svc_xprt *xprt;
196
197 xprt = svc_find_xprt(serv, name, net, family, 0);
198 if (xprt == NULL)
199 return svc_create_xprt(serv, name, net, family, port,
200 SVC_SOCK_DEFAULTS, cred);
201 svc_xprt_put(xprt);
202 return 0;
203}
204
205static int create_lockd_family(struct svc_serv *serv, struct net *net,
206 const int family, const struct cred *cred)
207{
208 int err;
209
210 err = create_lockd_listener(serv, "udp", net, family, nlm_udpport,
211 cred);
212 if (err < 0)
213 return err;
214
215 return create_lockd_listener(serv, "tcp", net, family, nlm_tcpport,
216 cred);
217}
218
219/*
220 * Ensure there are active UDP and TCP listeners for lockd.
221 *
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.
225 *
226 * Returns zero if all listeners are available; otherwise a
227 * negative errno value is returned.
228 */
229static int make_socks(struct svc_serv *serv, struct net *net,
230 const struct cred *cred)
231{
232 static int warned;
233 int err;
234
235 err = create_lockd_family(serv, net, PF_INET, cred);
236 if (err < 0)
237 goto out_err;
238
239 err = create_lockd_family(serv, net, PF_INET6, cred);
240 if (err < 0 && err != -EAFNOSUPPORT)
241 goto out_err;
242
243 warned = 0;
244 return 0;
245
246out_err:
247 if (warned++ == 0)
248 printk(KERN_WARNING
249 "lockd_up: makesock failed, error=%d\n", err);
250 svc_shutdown_net(serv, net);
251 return err;
252}
253
254static int lockd_up_net(struct svc_serv *serv, struct net *net,
255 const struct cred *cred)
256{
257 struct lockd_net *ln = net_generic(net, lockd_net_id);
258 int error;
259
260 if (ln->nlmsvc_users++)
261 return 0;
262
263 error = svc_bind(serv, net);
264 if (error)
265 goto err_bind;
266
267 error = make_socks(serv, net, cred);
268 if (error < 0)
269 goto err_bind;
270 set_grace_period(net);
271 dprintk("%s: per-net data created; net=%x\n", __func__, net->ns.inum);
272 return 0;
273
274err_bind:
275 ln->nlmsvc_users--;
276 return error;
277}
278
279static void lockd_down_net(struct svc_serv *serv, struct net *net)
280{
281 struct lockd_net *ln = net_generic(net, lockd_net_id);
282
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);
291 }
292 } else {
293 pr_err("%s: no users! task=%p, net=%x\n",
294 __func__, nlmsvc_task, net->ns.inum);
295 BUG();
296 }
297}
298
299static int lockd_inetaddr_event(struct notifier_block *this,
300 unsigned long event, void *ptr)
301{
302 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
303 struct sockaddr_in sin;
304
305 if ((event != NETDEV_DOWN) ||
306 !atomic_inc_not_zero(&nlm_ntf_refcnt))
307 goto out;
308
309 if (nlmsvc_rqst) {
310 dprintk("lockd_inetaddr_event: removed %pI4\n",
311 &ifa->ifa_local);
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);
316 }
317 atomic_dec(&nlm_ntf_refcnt);
318 wake_up(&nlm_ntf_wq);
319
320out:
321 return NOTIFY_DONE;
322}
323
324static struct notifier_block lockd_inetaddr_notifier = {
325 .notifier_call = lockd_inetaddr_event,
326};
327
328#if IS_ENABLED(CONFIG_IPV6)
329static int lockd_inet6addr_event(struct notifier_block *this,
330 unsigned long event, void *ptr)
331{
332 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
333 struct sockaddr_in6 sin6;
334
335 if ((event != NETDEV_DOWN) ||
336 !atomic_inc_not_zero(&nlm_ntf_refcnt))
337 goto out;
338
339 if (nlmsvc_rqst) {
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);
347 }
348 atomic_dec(&nlm_ntf_refcnt);
349 wake_up(&nlm_ntf_wq);
350
351out:
352 return NOTIFY_DONE;
353}
354
355static struct notifier_block lockd_inet6addr_notifier = {
356 .notifier_call = lockd_inet6addr_event,
357};
358#endif
359
360static void lockd_unregister_notifiers(void)
361{
362 unregister_inetaddr_notifier(&lockd_inetaddr_notifier);
363#if IS_ENABLED(CONFIG_IPV6)
364 unregister_inet6addr_notifier(&lockd_inet6addr_notifier);
365#endif
366 wait_event(nlm_ntf_wq, atomic_read(&nlm_ntf_refcnt) == 0);
367}
368
369static void lockd_svc_exit_thread(void)
370{
371 atomic_dec(&nlm_ntf_refcnt);
372 lockd_unregister_notifiers();
373 svc_exit_thread(nlmsvc_rqst);
374}
375
376static int lockd_start_svc(struct svc_serv *serv)
377{
378 int error;
379
380 if (nlmsvc_rqst)
381 return 0;
382
383 /*
384 * Create the kernel thread and wait for it to start.
385 */
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);
389 printk(KERN_WARNING
390 "lockd_up: svc_rqst allocation failed, error=%d\n",
391 error);
392 lockd_unregister_notifiers();
393 goto out_rqst;
394 }
395
396 atomic_inc(&nlm_ntf_refcnt);
397 svc_sock_update_bufs(serv);
398 serv->sv_maxconn = nlm_max_connections;
399
400 nlmsvc_task = kthread_create(lockd, nlmsvc_rqst, "%s", serv->sv_name);
401 if (IS_ERR(nlmsvc_task)) {
402 error = PTR_ERR(nlmsvc_task);
403 printk(KERN_WARNING
404 "lockd_up: kthread_run failed, error=%d\n", error);
405 goto out_task;
406 }
407 nlmsvc_rqst->rq_task = nlmsvc_task;
408 wake_up_process(nlmsvc_task);
409
410 dprintk("lockd_up: service started\n");
411 return 0;
412
413out_task:
414 lockd_svc_exit_thread();
415 nlmsvc_task = NULL;
416out_rqst:
417 nlmsvc_rqst = NULL;
418 return error;
419}
420
421static const struct svc_serv_ops lockd_sv_ops = {
422 .svo_shutdown = svc_rpcb_cleanup,
423 .svo_enqueue_xprt = svc_xprt_do_enqueue,
424};
425
426static struct svc_serv *lockd_create_svc(void)
427{
428 struct svc_serv *serv;
429
430 /*
431 * Check whether we're already up and running.
432 */
433 if (nlmsvc_rqst) {
434 /*
435 * Note: increase service usage, because later in case of error
436 * svc_destroy() will be called.
437 */
438 svc_get(nlmsvc_rqst->rq_server);
439 return nlmsvc_rqst->rq_server;
440 }
441
442 /*
443 * Sanity check: if there's no pid,
444 * we should be the first user ...
445 */
446 if (nlmsvc_users)
447 printk(KERN_WARNING
448 "lockd_up: no pid, %d users??\n", nlmsvc_users);
449
450 if (!nlm_timeout)
451 nlm_timeout = LOCKD_DFLT_TIMEO;
452 nlmsvc_timeout = nlm_timeout * HZ;
453
454 serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, &lockd_sv_ops);
455 if (!serv) {
456 printk(KERN_WARNING "lockd_up: create service failed\n");
457 return ERR_PTR(-ENOMEM);
458 }
459 register_inetaddr_notifier(&lockd_inetaddr_notifier);
460#if IS_ENABLED(CONFIG_IPV6)
461 register_inet6addr_notifier(&lockd_inet6addr_notifier);
462#endif
463 dprintk("lockd_up: service created\n");
464 return serv;
465}
466
467/*
468 * Bring up the lockd process if it's not already up.
469 */
470int lockd_up(struct net *net, const struct cred *cred)
471{
472 struct svc_serv *serv;
473 int error;
474
475 mutex_lock(&nlmsvc_mutex);
476
477 serv = lockd_create_svc();
478 if (IS_ERR(serv)) {
479 error = PTR_ERR(serv);
480 goto err_create;
481 }
482
483 error = lockd_up_net(serv, net, cred);
484 if (error < 0) {
485 lockd_unregister_notifiers();
486 goto err_put;
487 }
488
489 error = lockd_start_svc(serv);
490 if (error < 0) {
491 lockd_down_net(serv, net);
492 goto err_put;
493 }
494 nlmsvc_users++;
495 /*
496 * Note: svc_serv structures have an initial use count of 1,
497 * so we exit through here on both success and failure.
498 */
499err_put:
500 svc_destroy(serv);
501err_create:
502 mutex_unlock(&nlmsvc_mutex);
503 return error;
504}
505EXPORT_SYMBOL_GPL(lockd_up);
506
507/*
508 * Decrement the user count and bring down lockd if we're the last.
509 */
510void
511lockd_down(struct net *net)
512{
513 mutex_lock(&nlmsvc_mutex);
514 lockd_down_net(nlmsvc_rqst->rq_server, net);
515 if (nlmsvc_users) {
516 if (--nlmsvc_users)
517 goto out;
518 } else {
519 printk(KERN_ERR "lockd_down: no users! task=%p\n",
520 nlmsvc_task);
521 BUG();
522 }
523
524 if (!nlmsvc_task) {
525 printk(KERN_ERR "lockd_down: no lockd running.\n");
526 BUG();
527 }
528 kthread_stop(nlmsvc_task);
529 dprintk("lockd_down: service stopped\n");
530 lockd_svc_exit_thread();
531 dprintk("lockd_down: service destroyed\n");
532 nlmsvc_task = NULL;
533 nlmsvc_rqst = NULL;
534out:
535 mutex_unlock(&nlmsvc_mutex);
536}
537EXPORT_SYMBOL_GPL(lockd_down);
538
539#ifdef CONFIG_SYSCTL
540
541/*
542 * Sysctl parameters (same as module parameters, different interface).
543 */
544
545static struct ctl_table nlm_sysctls[] = {
546 {
547 .procname = "nlm_grace_period",
548 .data = &nlm_grace_period,
549 .maxlen = sizeof(unsigned long),
550 .mode = 0644,
551 .proc_handler = proc_doulongvec_minmax,
552 .extra1 = (unsigned long *) &nlm_grace_period_min,
553 .extra2 = (unsigned long *) &nlm_grace_period_max,
554 },
555 {
556 .procname = "nlm_timeout",
557 .data = &nlm_timeout,
558 .maxlen = sizeof(unsigned long),
559 .mode = 0644,
560 .proc_handler = proc_doulongvec_minmax,
561 .extra1 = (unsigned long *) &nlm_timeout_min,
562 .extra2 = (unsigned long *) &nlm_timeout_max,
563 },
564 {
565 .procname = "nlm_udpport",
566 .data = &nlm_udpport,
567 .maxlen = sizeof(int),
568 .mode = 0644,
569 .proc_handler = proc_dointvec_minmax,
570 .extra1 = (int *) &nlm_port_min,
571 .extra2 = (int *) &nlm_port_max,
572 },
573 {
574 .procname = "nlm_tcpport",
575 .data = &nlm_tcpport,
576 .maxlen = sizeof(int),
577 .mode = 0644,
578 .proc_handler = proc_dointvec_minmax,
579 .extra1 = (int *) &nlm_port_min,
580 .extra2 = (int *) &nlm_port_max,
581 },
582 {
583 .procname = "nsm_use_hostnames",
584 .data = &nsm_use_hostnames,
585 .maxlen = sizeof(int),
586 .mode = 0644,
587 .proc_handler = proc_dointvec,
588 },
589 {
590 .procname = "nsm_local_state",
591 .data = &nsm_local_state,
592 .maxlen = sizeof(int),
593 .mode = 0644,
594 .proc_handler = proc_dointvec,
595 },
596 { }
597};
598
599static struct ctl_table nlm_sysctl_dir[] = {
600 {
601 .procname = "nfs",
602 .mode = 0555,
603 .child = nlm_sysctls,
604 },
605 { }
606};
607
608static struct ctl_table nlm_sysctl_root[] = {
609 {
610 .procname = "fs",
611 .mode = 0555,
612 .child = nlm_sysctl_dir,
613 },
614 { }
615};
616
617#endif /* CONFIG_SYSCTL */
618
619/*
620 * Module (and sysfs) parameters.
621 */
622
623#define param_set_min_max(name, type, which_strtol, min, max) \
624static int param_set_##name(const char *val, const struct kernel_param *kp) \
625{ \
626 char *endp; \
627 __typeof__(type) num = which_strtol(val, &endp, 0); \
628 if (endp == val || *endp || num < (min) || num > (max)) \
629 return -EINVAL; \
630 *((type *) kp->arg) = num; \
631 return 0; \
632}
633
634static inline int is_callback(u32 proc)
635{
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;
643}
644
645
646static int lockd_authenticate(struct svc_rqst *rqstp)
647{
648 rqstp->rq_client = NULL;
649 switch (rqstp->rq_authop->flavour) {
650 case RPC_AUTH_NULL:
651 case RPC_AUTH_UNIX:
652 if (rqstp->rq_proc == 0)
653 return SVC_OK;
654 if (is_callback(rqstp->rq_proc)) {
655 /* Leave it to individual procedures to
656 * call nlmsvc_lookup_host(rqstp)
657 */
658 return SVC_OK;
659 }
660 return svc_set_client(rqstp);
661 }
662 return SVC_DENIED;
663}
664
665
666param_set_min_max(port, int, simple_strtol, 0, 65535)
667param_set_min_max(grace_period, unsigned long, simple_strtoul,
668 nlm_grace_period_min, nlm_grace_period_max)
669param_set_min_max(timeout, unsigned long, simple_strtoul,
670 nlm_timeout_min, nlm_timeout_max)
671
672MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
673MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
674MODULE_LICENSE("GPL");
675
676module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
677 &nlm_grace_period, 0644);
678module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
679 &nlm_timeout, 0644);
680module_param_call(nlm_udpport, param_set_port, param_get_int,
681 &nlm_udpport, 0644);
682module_param_call(nlm_tcpport, param_set_port, param_get_int,
683 &nlm_tcpport, 0644);
684module_param(nsm_use_hostnames, bool, 0644);
685module_param(nlm_max_connections, uint, 0644);
686
687static int lockd_init_net(struct net *net)
688{
689 struct lockd_net *ln = net_generic(net, lockd_net_id);
690
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);
695 return 0;
696}
697
698static void lockd_exit_net(struct net *net)
699{
700 struct lockd_net *ln = net_generic(net, lockd_net_id);
701
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__);
711}
712
713static struct pernet_operations lockd_net_ops = {
714 .init = lockd_init_net,
715 .exit = lockd_exit_net,
716 .id = &lockd_net_id,
717 .size = sizeof(struct lockd_net),
718};
719
720
721/*
722 * Initialising and terminating the module.
723 */
724
725static int __init init_nlm(void)
726{
727 int err;
728
729#ifdef CONFIG_SYSCTL
730 err = -ENOMEM;
731 nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
732 if (nlm_sysctl_table == NULL)
733 goto err_sysctl;
734#endif
735 err = register_pernet_subsys(&lockd_net_ops);
736 if (err)
737 goto err_pernet;
738
739 err = lockd_create_procfs();
740 if (err)
741 goto err_procfs;
742
743 return 0;
744
745err_procfs:
746 unregister_pernet_subsys(&lockd_net_ops);
747err_pernet:
748#ifdef CONFIG_SYSCTL
749 unregister_sysctl_table(nlm_sysctl_table);
750err_sysctl:
751#endif
752 return err;
753}
754
755static void __exit exit_nlm(void)
756{
757 /* FIXME: delete all NLM clients */
758 nlm_shutdown_hosts();
759 lockd_remove_procfs();
760 unregister_pernet_subsys(&lockd_net_ops);
761#ifdef CONFIG_SYSCTL
762 unregister_sysctl_table(nlm_sysctl_table);
763#endif
764}
765
766module_init(init_nlm);
767module_exit(exit_nlm);
768
769/*
770 * Define NLM program and procedures
771 */
772static unsigned int nlmsvc_version1_count[17];
773static const struct svc_version nlmsvc_version1 = {
774 .vs_vers = 1,
775 .vs_nproc = 17,
776 .vs_proc = nlmsvc_procedures,
777 .vs_count = nlmsvc_version1_count,
778 .vs_xdrsize = NLMSVC_XDRSIZE,
779};
780static unsigned int nlmsvc_version3_count[24];
781static const struct svc_version nlmsvc_version3 = {
782 .vs_vers = 3,
783 .vs_nproc = 24,
784 .vs_proc = nlmsvc_procedures,
785 .vs_count = nlmsvc_version3_count,
786 .vs_xdrsize = NLMSVC_XDRSIZE,
787};
788#ifdef CONFIG_LOCKD_V4
789static unsigned int nlmsvc_version4_count[24];
790static const struct svc_version nlmsvc_version4 = {
791 .vs_vers = 4,
792 .vs_nproc = 24,
793 .vs_proc = nlmsvc_procedures4,
794 .vs_count = nlmsvc_version4_count,
795 .vs_xdrsize = NLMSVC_XDRSIZE,
796};
797#endif
798static const struct svc_version *nlmsvc_version[] = {
799 [1] = &nlmsvc_version1,
800 [3] = &nlmsvc_version3,
801#ifdef CONFIG_LOCKD_V4
802 [4] = &nlmsvc_version4,
803#endif
804};
805
806static struct svc_stat nlmsvc_stats;
807
808#define NLM_NRVERS ARRAY_SIZE(nlmsvc_version)
809static 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,
819};
1/*
2 * linux/fs/lockd/svc.c
3 *
4 * This is the central lockd service.
5 *
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?
9 *
10 * Authors: Olaf Kirch (okir@monad.swb.de)
11 *
12 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
13 */
14
15#include <linux/module.h>
16#include <linux/init.h>
17#include <linux/sysctl.h>
18#include <linux/moduleparam.h>
19
20#include <linux/sched.h>
21#include <linux/errno.h>
22#include <linux/in.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
29#include <linux/sunrpc/types.h>
30#include <linux/sunrpc/stats.h>
31#include <linux/sunrpc/clnt.h>
32#include <linux/sunrpc/svc.h>
33#include <linux/sunrpc/svcsock.h>
34#include <net/ip.h>
35#include <linux/lockd/lockd.h>
36#include <linux/nfs.h>
37
38#define NLMDBG_FACILITY NLMDBG_SVC
39#define LOCKD_BUFSIZE (1024 + NLMSVC_XDRSIZE)
40#define ALLOWED_SIGS (sigmask(SIGKILL))
41
42static struct svc_program nlmsvc_program;
43
44struct nlmsvc_binding * nlmsvc_ops;
45EXPORT_SYMBOL_GPL(nlmsvc_ops);
46
47static DEFINE_MUTEX(nlmsvc_mutex);
48static unsigned int nlmsvc_users;
49static struct task_struct *nlmsvc_task;
50static struct svc_rqst *nlmsvc_rqst;
51unsigned long nlmsvc_timeout;
52
53/*
54 * These can be set at insmod time (useful for NFS as root filesystem),
55 * and also changed through the sysctl interface. -- Jamie Lokier, Aug 2003
56 */
57static unsigned long nlm_grace_period;
58static unsigned long nlm_timeout = LOCKD_DFLT_TIMEO;
59static int nlm_udpport, nlm_tcpport;
60
61/* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
62static unsigned int nlm_max_connections = 1024;
63
64/*
65 * Constants needed for the sysctl interface.
66 */
67static const unsigned long nlm_grace_period_min = 0;
68static const unsigned long nlm_grace_period_max = 240;
69static const unsigned long nlm_timeout_min = 3;
70static const unsigned long nlm_timeout_max = 20;
71static const int nlm_port_min = 0, nlm_port_max = 65535;
72
73#ifdef CONFIG_SYSCTL
74static struct ctl_table_header * nlm_sysctl_table;
75#endif
76
77static unsigned long get_lockd_grace_period(void)
78{
79 /* Note: nlm_timeout should always be nonzero */
80 if (nlm_grace_period)
81 return roundup(nlm_grace_period, nlm_timeout) * HZ;
82 else
83 return nlm_timeout * 5 * HZ;
84}
85
86static struct lock_manager lockd_manager = {
87};
88
89static void grace_ender(struct work_struct *not_used)
90{
91 locks_end_grace(&lockd_manager);
92}
93
94static DECLARE_DELAYED_WORK(grace_period_end, grace_ender);
95
96static void set_grace_period(void)
97{
98 unsigned long grace_period = get_lockd_grace_period();
99
100 locks_start_grace(&lockd_manager);
101 cancel_delayed_work_sync(&grace_period_end);
102 schedule_delayed_work(&grace_period_end, grace_period);
103}
104
105static void restart_grace(void)
106{
107 if (nlmsvc_ops) {
108 cancel_delayed_work_sync(&grace_period_end);
109 locks_end_grace(&lockd_manager);
110 nlmsvc_invalidate_all();
111 set_grace_period();
112 }
113}
114
115/*
116 * This is the lockd kernel thread
117 */
118static int
119lockd(void *vrqstp)
120{
121 int err = 0, preverr = 0;
122 struct svc_rqst *rqstp = vrqstp;
123
124 /* try_to_freeze() is called from svc_recv() */
125 set_freezable();
126
127 /* Allow SIGKILL to tell lockd to drop all of its locks */
128 allow_signal(SIGKILL);
129
130 dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
131
132 if (!nlm_timeout)
133 nlm_timeout = LOCKD_DFLT_TIMEO;
134 nlmsvc_timeout = nlm_timeout * HZ;
135
136 set_grace_period();
137
138 /*
139 * The main request loop. We don't terminate until the last
140 * NFS mount or NFS daemon has gone away.
141 */
142 while (!kthread_should_stop()) {
143 long timeout = MAX_SCHEDULE_TIMEOUT;
144 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
145
146 /* update sv_maxconn if it has changed */
147 rqstp->rq_server->sv_maxconn = nlm_max_connections;
148
149 if (signalled()) {
150 flush_signals(current);
151 restart_grace();
152 continue;
153 }
154
155 timeout = nlmsvc_retry_blocked();
156
157 /*
158 * Find a socket with data available and call its
159 * recvfrom routine.
160 */
161 err = svc_recv(rqstp, timeout);
162 if (err == -EAGAIN || err == -EINTR) {
163 preverr = err;
164 continue;
165 }
166 if (err < 0) {
167 if (err != preverr) {
168 printk(KERN_WARNING "%s: unexpected error "
169 "from svc_recv (%d)\n", __func__, err);
170 preverr = err;
171 }
172 schedule_timeout_interruptible(HZ);
173 continue;
174 }
175 preverr = err;
176
177 dprintk("lockd: request from %s\n",
178 svc_print_addr(rqstp, buf, sizeof(buf)));
179
180 svc_process(rqstp);
181 }
182 flush_signals(current);
183 cancel_delayed_work_sync(&grace_period_end);
184 locks_end_grace(&lockd_manager);
185 if (nlmsvc_ops)
186 nlmsvc_invalidate_all();
187 nlm_shutdown_hosts();
188 return 0;
189}
190
191static int create_lockd_listener(struct svc_serv *serv, const char *name,
192 const int family, const unsigned short port)
193{
194 struct svc_xprt *xprt;
195
196 xprt = svc_find_xprt(serv, name, family, 0);
197 if (xprt == NULL)
198 return svc_create_xprt(serv, name, &init_net, family, port,
199 SVC_SOCK_DEFAULTS);
200 svc_xprt_put(xprt);
201 return 0;
202}
203
204static int create_lockd_family(struct svc_serv *serv, const int family)
205{
206 int err;
207
208 err = create_lockd_listener(serv, "udp", family, nlm_udpport);
209 if (err < 0)
210 return err;
211
212 return create_lockd_listener(serv, "tcp", family, nlm_tcpport);
213}
214
215/*
216 * Ensure there are active UDP and TCP listeners for lockd.
217 *
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.
221 *
222 * Returns zero if all listeners are available; otherwise a
223 * negative errno value is returned.
224 */
225static int make_socks(struct svc_serv *serv)
226{
227 static int warned;
228 int err;
229
230 err = create_lockd_family(serv, PF_INET);
231 if (err < 0)
232 goto out_err;
233
234 err = create_lockd_family(serv, PF_INET6);
235 if (err < 0 && err != -EAFNOSUPPORT)
236 goto out_err;
237
238 warned = 0;
239 return 0;
240
241out_err:
242 if (warned++ == 0)
243 printk(KERN_WARNING
244 "lockd_up: makesock failed, error=%d\n", err);
245 return err;
246}
247
248/*
249 * Bring up the lockd process if it's not already up.
250 */
251int lockd_up(void)
252{
253 struct svc_serv *serv;
254 int error = 0;
255
256 mutex_lock(&nlmsvc_mutex);
257 /*
258 * Check whether we're already up and running.
259 */
260 if (nlmsvc_rqst)
261 goto out;
262
263 /*
264 * Sanity check: if there's no pid,
265 * we should be the first user ...
266 */
267 if (nlmsvc_users)
268 printk(KERN_WARNING
269 "lockd_up: no pid, %d users??\n", nlmsvc_users);
270
271 error = -ENOMEM;
272 serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
273 if (!serv) {
274 printk(KERN_WARNING "lockd_up: create service failed\n");
275 goto out;
276 }
277
278 error = make_socks(serv);
279 if (error < 0)
280 goto destroy_and_out;
281
282 /*
283 * Create the kernel thread and wait for it to start.
284 */
285 nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
286 if (IS_ERR(nlmsvc_rqst)) {
287 error = PTR_ERR(nlmsvc_rqst);
288 nlmsvc_rqst = NULL;
289 printk(KERN_WARNING
290 "lockd_up: svc_rqst allocation failed, error=%d\n",
291 error);
292 goto destroy_and_out;
293 }
294
295 svc_sock_update_bufs(serv);
296 serv->sv_maxconn = nlm_max_connections;
297
298 nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
299 if (IS_ERR(nlmsvc_task)) {
300 error = PTR_ERR(nlmsvc_task);
301 svc_exit_thread(nlmsvc_rqst);
302 nlmsvc_task = NULL;
303 nlmsvc_rqst = NULL;
304 printk(KERN_WARNING
305 "lockd_up: kthread_run failed, error=%d\n", error);
306 goto destroy_and_out;
307 }
308
309 /*
310 * Note: svc_serv structures have an initial use count of 1,
311 * so we exit through here on both success and failure.
312 */
313destroy_and_out:
314 svc_destroy(serv);
315out:
316 if (!error)
317 nlmsvc_users++;
318 mutex_unlock(&nlmsvc_mutex);
319 return error;
320}
321EXPORT_SYMBOL_GPL(lockd_up);
322
323/*
324 * Decrement the user count and bring down lockd if we're the last.
325 */
326void
327lockd_down(void)
328{
329 mutex_lock(&nlmsvc_mutex);
330 if (nlmsvc_users) {
331 if (--nlmsvc_users)
332 goto out;
333 } else {
334 printk(KERN_ERR "lockd_down: no users! task=%p\n",
335 nlmsvc_task);
336 BUG();
337 }
338
339 if (!nlmsvc_task) {
340 printk(KERN_ERR "lockd_down: no lockd running.\n");
341 BUG();
342 }
343 kthread_stop(nlmsvc_task);
344 svc_exit_thread(nlmsvc_rqst);
345 nlmsvc_task = NULL;
346 nlmsvc_rqst = NULL;
347out:
348 mutex_unlock(&nlmsvc_mutex);
349}
350EXPORT_SYMBOL_GPL(lockd_down);
351
352#ifdef CONFIG_SYSCTL
353
354/*
355 * Sysctl parameters (same as module parameters, different interface).
356 */
357
358static ctl_table nlm_sysctls[] = {
359 {
360 .procname = "nlm_grace_period",
361 .data = &nlm_grace_period,
362 .maxlen = sizeof(unsigned long),
363 .mode = 0644,
364 .proc_handler = proc_doulongvec_minmax,
365 .extra1 = (unsigned long *) &nlm_grace_period_min,
366 .extra2 = (unsigned long *) &nlm_grace_period_max,
367 },
368 {
369 .procname = "nlm_timeout",
370 .data = &nlm_timeout,
371 .maxlen = sizeof(unsigned long),
372 .mode = 0644,
373 .proc_handler = proc_doulongvec_minmax,
374 .extra1 = (unsigned long *) &nlm_timeout_min,
375 .extra2 = (unsigned long *) &nlm_timeout_max,
376 },
377 {
378 .procname = "nlm_udpport",
379 .data = &nlm_udpport,
380 .maxlen = sizeof(int),
381 .mode = 0644,
382 .proc_handler = proc_dointvec_minmax,
383 .extra1 = (int *) &nlm_port_min,
384 .extra2 = (int *) &nlm_port_max,
385 },
386 {
387 .procname = "nlm_tcpport",
388 .data = &nlm_tcpport,
389 .maxlen = sizeof(int),
390 .mode = 0644,
391 .proc_handler = proc_dointvec_minmax,
392 .extra1 = (int *) &nlm_port_min,
393 .extra2 = (int *) &nlm_port_max,
394 },
395 {
396 .procname = "nsm_use_hostnames",
397 .data = &nsm_use_hostnames,
398 .maxlen = sizeof(int),
399 .mode = 0644,
400 .proc_handler = proc_dointvec,
401 },
402 {
403 .procname = "nsm_local_state",
404 .data = &nsm_local_state,
405 .maxlen = sizeof(int),
406 .mode = 0644,
407 .proc_handler = proc_dointvec,
408 },
409 { }
410};
411
412static ctl_table nlm_sysctl_dir[] = {
413 {
414 .procname = "nfs",
415 .mode = 0555,
416 .child = nlm_sysctls,
417 },
418 { }
419};
420
421static ctl_table nlm_sysctl_root[] = {
422 {
423 .procname = "fs",
424 .mode = 0555,
425 .child = nlm_sysctl_dir,
426 },
427 { }
428};
429
430#endif /* CONFIG_SYSCTL */
431
432/*
433 * Module (and sysfs) parameters.
434 */
435
436#define param_set_min_max(name, type, which_strtol, min, max) \
437static int param_set_##name(const char *val, struct kernel_param *kp) \
438{ \
439 char *endp; \
440 __typeof__(type) num = which_strtol(val, &endp, 0); \
441 if (endp == val || *endp || num < (min) || num > (max)) \
442 return -EINVAL; \
443 *((int *) kp->arg) = num; \
444 return 0; \
445}
446
447static inline int is_callback(u32 proc)
448{
449 return proc == NLMPROC_GRANTED
450 || proc == NLMPROC_GRANTED_MSG
451 || proc == NLMPROC_TEST_RES
452 || proc == NLMPROC_LOCK_RES
453 || proc == NLMPROC_CANCEL_RES
454 || proc == NLMPROC_UNLOCK_RES
455 || proc == NLMPROC_NSM_NOTIFY;
456}
457
458
459static int lockd_authenticate(struct svc_rqst *rqstp)
460{
461 rqstp->rq_client = NULL;
462 switch (rqstp->rq_authop->flavour) {
463 case RPC_AUTH_NULL:
464 case RPC_AUTH_UNIX:
465 if (rqstp->rq_proc == 0)
466 return SVC_OK;
467 if (is_callback(rqstp->rq_proc)) {
468 /* Leave it to individual procedures to
469 * call nlmsvc_lookup_host(rqstp)
470 */
471 return SVC_OK;
472 }
473 return svc_set_client(rqstp);
474 }
475 return SVC_DENIED;
476}
477
478
479param_set_min_max(port, int, simple_strtol, 0, 65535)
480param_set_min_max(grace_period, unsigned long, simple_strtoul,
481 nlm_grace_period_min, nlm_grace_period_max)
482param_set_min_max(timeout, unsigned long, simple_strtoul,
483 nlm_timeout_min, nlm_timeout_max)
484
485MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
486MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
487MODULE_LICENSE("GPL");
488
489module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
490 &nlm_grace_period, 0644);
491module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
492 &nlm_timeout, 0644);
493module_param_call(nlm_udpport, param_set_port, param_get_int,
494 &nlm_udpport, 0644);
495module_param_call(nlm_tcpport, param_set_port, param_get_int,
496 &nlm_tcpport, 0644);
497module_param(nsm_use_hostnames, bool, 0644);
498module_param(nlm_max_connections, uint, 0644);
499
500/*
501 * Initialising and terminating the module.
502 */
503
504static int __init init_nlm(void)
505{
506#ifdef CONFIG_SYSCTL
507 nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
508 return nlm_sysctl_table ? 0 : -ENOMEM;
509#else
510 return 0;
511#endif
512}
513
514static void __exit exit_nlm(void)
515{
516 /* FIXME: delete all NLM clients */
517 nlm_shutdown_hosts();
518#ifdef CONFIG_SYSCTL
519 unregister_sysctl_table(nlm_sysctl_table);
520#endif
521}
522
523module_init(init_nlm);
524module_exit(exit_nlm);
525
526/*
527 * Define NLM program and procedures
528 */
529static struct svc_version nlmsvc_version1 = {
530 .vs_vers = 1,
531 .vs_nproc = 17,
532 .vs_proc = nlmsvc_procedures,
533 .vs_xdrsize = NLMSVC_XDRSIZE,
534};
535static struct svc_version nlmsvc_version3 = {
536 .vs_vers = 3,
537 .vs_nproc = 24,
538 .vs_proc = nlmsvc_procedures,
539 .vs_xdrsize = NLMSVC_XDRSIZE,
540};
541#ifdef CONFIG_LOCKD_V4
542static struct svc_version nlmsvc_version4 = {
543 .vs_vers = 4,
544 .vs_nproc = 24,
545 .vs_proc = nlmsvc_procedures4,
546 .vs_xdrsize = NLMSVC_XDRSIZE,
547};
548#endif
549static struct svc_version * nlmsvc_version[] = {
550 [1] = &nlmsvc_version1,
551 [3] = &nlmsvc_version3,
552#ifdef CONFIG_LOCKD_V4
553 [4] = &nlmsvc_version4,
554#endif
555};
556
557static struct svc_stat nlmsvc_stats;
558
559#define NLM_NRVERS ARRAY_SIZE(nlmsvc_version)
560static struct svc_program nlmsvc_program = {
561 .pg_prog = NLM_PROGRAM, /* program number */
562 .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */
563 .pg_vers = nlmsvc_version, /* version table */
564 .pg_name = "lockd", /* service name */
565 .pg_class = "nfsd", /* share authentication with nfsd */
566 .pg_stats = &nlmsvc_stats, /* stats table */
567 .pg_authenticate = &lockd_authenticate /* export authentication */
568};