blob: 8ca18085e68d0e796d0089b3b932aabddf835d3f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#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/slab.h>
25#include <linux/smp.h>
26#include <linux/smp_lock.h>
Ingo Molnar353ab6e2006-03-26 01:37:12 -080027#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
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>
NeilBrown24e36662006-10-02 02:17:45 -070034#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/lockd/lockd.h>
Olaf Kirch460f5ca2006-10-04 02:16:03 -070036#include <linux/lockd/sm_inter.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/nfs.h>
38
39#define NLMDBG_FACILITY NLMDBG_SVC
40#define LOCKD_BUFSIZE (1024 + NLMSVC_XDRSIZE)
41#define ALLOWED_SIGS (sigmask(SIGKILL))
42
43static struct svc_program nlmsvc_program;
44
45struct nlmsvc_binding * nlmsvc_ops;
46EXPORT_SYMBOL(nlmsvc_ops);
47
Ingo Molnar353ab6e2006-03-26 01:37:12 -080048static DEFINE_MUTEX(nlmsvc_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049static unsigned int nlmsvc_users;
50static pid_t nlmsvc_pid;
NeilBrown24e36662006-10-02 02:17:45 -070051static struct svc_serv *nlmsvc_serv;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052int nlmsvc_grace_period;
53unsigned long nlmsvc_timeout;
54
Ingo Molnar353ab6e2006-03-26 01:37:12 -080055static DECLARE_COMPLETION(lockd_start_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static DECLARE_WAIT_QUEUE_HEAD(lockd_exit);
57
58/*
59 * These can be set at insmod time (useful for NFS as root filesystem),
60 * and also changed through the sysctl interface. -- Jamie Lokier, Aug 2003
61 */
62static unsigned long nlm_grace_period;
63static unsigned long nlm_timeout = LOCKD_DFLT_TIMEO;
64static int nlm_udpport, nlm_tcpport;
Olaf Kirchabd1f502006-10-04 02:16:01 -070065int nsm_use_hostnames = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67/*
68 * Constants needed for the sysctl interface.
69 */
70static const unsigned long nlm_grace_period_min = 0;
71static const unsigned long nlm_grace_period_max = 240;
72static const unsigned long nlm_timeout_min = 3;
73static const unsigned long nlm_timeout_max = 20;
74static const int nlm_port_min = 0, nlm_port_max = 65535;
75
76static struct ctl_table_header * nlm_sysctl_table;
77
78static unsigned long set_grace_period(void)
79{
80 unsigned long grace_period;
81
82 /* Note: nlm_timeout should always be nonzero */
83 if (nlm_grace_period)
84 grace_period = ((nlm_grace_period + nlm_timeout - 1)
85 / nlm_timeout) * nlm_timeout * HZ;
86 else
87 grace_period = nlm_timeout * 5 * HZ;
88 nlmsvc_grace_period = 1;
89 return grace_period + jiffies;
90}
91
92static inline void clear_grace_period(void)
93{
94 nlmsvc_grace_period = 0;
95}
96
97/*
98 * This is the lockd kernel thread
99 */
100static void
101lockd(struct svc_rqst *rqstp)
102{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 int err = 0;
104 unsigned long grace_period_expire;
105
106 /* Lock module and set up kernel thread */
107 /* lockd_up is waiting for us to startup, so will
108 * be holding a reference to this module, so it
109 * is safe to just claim another reference
110 */
111 __module_get(THIS_MODULE);
112 lock_kernel();
113
114 /*
115 * Let our maker know we're running.
116 */
117 nlmsvc_pid = current->pid;
NeilBrown6fb2b472006-10-02 02:17:50 -0700118 nlmsvc_serv = rqstp->rq_server;
Ingo Molnar353ab6e2006-03-26 01:37:12 -0800119 complete(&lockd_start_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 daemonize("lockd");
122
123 /* Process request with signals blocked, but allow SIGKILL. */
124 allow_signal(SIGKILL);
125
126 /* kick rpciod */
127 rpciod_up();
128
129 dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
130
131 if (!nlm_timeout)
132 nlm_timeout = LOCKD_DFLT_TIMEO;
133 nlmsvc_timeout = nlm_timeout * HZ;
134
135 grace_period_expire = set_grace_period();
136
137 /*
138 * The main request loop. We don't terminate until the last
139 * NFS mount or NFS daemon has gone away, and we've been sent a
140 * signal, or else another process has taken over our job.
141 */
142 while ((nlmsvc_users || !signalled()) && nlmsvc_pid == current->pid) {
143 long timeout = MAX_SCHEDULE_TIMEOUT;
144
145 if (signalled()) {
146 flush_signals(current);
147 if (nlmsvc_ops) {
148 nlmsvc_invalidate_all();
149 grace_period_expire = set_grace_period();
150 }
151 }
152
153 /*
154 * Retry any blocked locks that have been notified by
155 * the VFS. Don't do this during grace period.
156 * (Theoretically, there shouldn't even be blocked locks
157 * during grace period).
158 */
159 if (!nlmsvc_grace_period) {
160 timeout = nlmsvc_retry_blocked();
161 } else if (time_before(grace_period_expire, jiffies))
162 clear_grace_period();
163
164 /*
165 * Find a socket with data available and call its
166 * recvfrom routine.
167 */
NeilBrown6fb2b472006-10-02 02:17:50 -0700168 err = svc_recv(rqstp, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 if (err == -EAGAIN || err == -EINTR)
170 continue;
171 if (err < 0) {
172 printk(KERN_WARNING
173 "lockd: terminating on error %d\n",
174 -err);
175 break;
176 }
177
178 dprintk("lockd: request from %08x\n",
179 (unsigned)ntohl(rqstp->rq_addr.sin_addr.s_addr));
180
NeilBrown6fb2b472006-10-02 02:17:50 -0700181 svc_process(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 }
184
Trond Myklebust2bd61572006-01-03 09:55:19 +0100185 flush_signals(current);
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 /*
188 * Check whether there's a new lockd process before
189 * shutting down the hosts and clearing the slot.
190 */
191 if (!nlmsvc_pid || current->pid == nlmsvc_pid) {
192 if (nlmsvc_ops)
193 nlmsvc_invalidate_all();
194 nlm_shutdown_hosts();
195 nlmsvc_pid = 0;
NeilBrown24e36662006-10-02 02:17:45 -0700196 nlmsvc_serv = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 } else
198 printk(KERN_DEBUG
199 "lockd: new process, skipping host shutdown\n");
200 wake_up(&lockd_exit);
NeilBrown46be9252005-06-23 22:03:11 -0700201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 /* Exit the RPC thread */
203 svc_exit_thread(rqstp);
204
205 /* release rpciod */
206 rpciod_down();
207
208 /* Release module */
209 unlock_kernel();
210 module_put_and_exit(0);
211}
212
NeilBrown24e36662006-10-02 02:17:45 -0700213
214static int find_socket(struct svc_serv *serv, int proto)
215{
216 struct svc_sock *svsk;
217 int found = 0;
218 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list)
219 if (svsk->sk_sk->sk_protocol == proto) {
220 found = 1;
221 break;
222 }
223 return found;
224}
225
226static int make_socks(struct svc_serv *serv, int proto)
227{
228 /* Make any sockets that are needed but not present.
229 * If nlm_udpport or nlm_tcpport were set as module
230 * options, make those sockets unconditionally
231 */
NeilBrown7dcf91e2006-10-02 02:17:52 -0700232 static int warned;
NeilBrown24e36662006-10-02 02:17:45 -0700233 int err = 0;
234 if (proto == IPPROTO_UDP || nlm_udpport)
235 if (!find_socket(serv, IPPROTO_UDP))
236 err = svc_makesock(serv, IPPROTO_UDP, nlm_udpport);
NeilBrown7dcf91e2006-10-02 02:17:52 -0700237 if (err == 0 && (proto == IPPROTO_TCP || nlm_tcpport))
NeilBrown24e36662006-10-02 02:17:45 -0700238 if (!find_socket(serv, IPPROTO_TCP))
239 err= svc_makesock(serv, IPPROTO_TCP, nlm_tcpport);
NeilBrown7dcf91e2006-10-02 02:17:52 -0700240 if (!err)
241 warned = 0;
242 else if (warned++ == 0)
243 printk(KERN_WARNING
244 "lockd_up: makesock failed, error=%d\n", err);
NeilBrown24e36662006-10-02 02:17:45 -0700245 return err;
246}
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248/*
249 * Bring up the lockd process if it's not already up.
250 */
251int
NeilBrown24e36662006-10-02 02:17:45 -0700252lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 struct svc_serv * serv;
255 int error = 0;
256
Ingo Molnar353ab6e2006-03-26 01:37:12 -0800257 mutex_lock(&nlmsvc_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * Check whether we're already up and running.
260 */
NeilBrown24e36662006-10-02 02:17:45 -0700261 if (nlmsvc_pid) {
NeilBrown4a3ae422006-10-02 02:17:53 -0700262 if (proto)
263 error = make_socks(nlmsvc_serv, proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 goto out;
NeilBrown24e36662006-10-02 02:17:45 -0700265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
267 /*
268 * Sanity check: if there's no pid,
269 * we should be the first user ...
270 */
NeilBrown4a3ae422006-10-02 02:17:53 -0700271 if (nlmsvc_users)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 printk(KERN_WARNING
273 "lockd_up: no pid, %d users??\n", nlmsvc_users);
274
275 error = -ENOMEM;
NeilBrownbc591cc2006-10-02 02:17:44 -0700276 serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 if (!serv) {
278 printk(KERN_WARNING "lockd_up: create service failed\n");
279 goto out;
280 }
281
NeilBrown7dcf91e2006-10-02 02:17:52 -0700282 if ((error = make_socks(serv, proto)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 goto destroy_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 /*
286 * Create the kernel thread and wait for it to start.
287 */
288 error = svc_create_thread(lockd, serv);
289 if (error) {
290 printk(KERN_WARNING
291 "lockd_up: create thread failed, error=%d\n", error);
292 goto destroy_and_out;
293 }
Ingo Molnar353ab6e2006-03-26 01:37:12 -0800294 wait_for_completion(&lockd_start_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
296 /*
297 * Note: svc_serv structures have an initial use count of 1,
298 * so we exit through here on both success and failure.
299 */
300destroy_and_out:
301 svc_destroy(serv);
302out:
NeilBrown4a3ae422006-10-02 02:17:53 -0700303 if (!error)
304 nlmsvc_users++;
Ingo Molnar353ab6e2006-03-26 01:37:12 -0800305 mutex_unlock(&nlmsvc_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return error;
307}
308EXPORT_SYMBOL(lockd_up);
309
310/*
311 * Decrement the user count and bring down lockd if we're the last.
312 */
313void
314lockd_down(void)
315{
316 static int warned;
317
Ingo Molnar353ab6e2006-03-26 01:37:12 -0800318 mutex_lock(&nlmsvc_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 if (nlmsvc_users) {
320 if (--nlmsvc_users)
321 goto out;
322 } else
323 printk(KERN_WARNING "lockd_down: no users! pid=%d\n", nlmsvc_pid);
324
325 if (!nlmsvc_pid) {
326 if (warned++ == 0)
327 printk(KERN_WARNING "lockd_down: no lockd running.\n");
328 goto out;
329 }
330 warned = 0;
331
332 kill_proc(nlmsvc_pid, SIGKILL, 1);
333 /*
334 * Wait for the lockd process to exit, but since we're holding
335 * the lockd semaphore, we can't wait around forever ...
336 */
337 clear_thread_flag(TIF_SIGPENDING);
338 interruptible_sleep_on_timeout(&lockd_exit, HZ);
339 if (nlmsvc_pid) {
340 printk(KERN_WARNING
341 "lockd_down: lockd failed to exit, clearing pid\n");
342 nlmsvc_pid = 0;
343 }
344 spin_lock_irq(&current->sighand->siglock);
345 recalc_sigpending();
346 spin_unlock_irq(&current->sighand->siglock);
347out:
Ingo Molnar353ab6e2006-03-26 01:37:12 -0800348 mutex_unlock(&nlmsvc_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349}
350EXPORT_SYMBOL(lockd_down);
351
352/*
353 * Sysctl parameters (same as module parameters, different interface).
354 */
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356static ctl_table nlm_sysctls[] = {
357 {
358 .ctl_name = CTL_UNNUMBERED,
359 .procname = "nlm_grace_period",
360 .data = &nlm_grace_period,
Steve Dickson7ee91ec2005-07-13 01:10:47 -0700361 .maxlen = sizeof(unsigned long),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 .mode = 0644,
363 .proc_handler = &proc_doulongvec_minmax,
364 .extra1 = (unsigned long *) &nlm_grace_period_min,
365 .extra2 = (unsigned long *) &nlm_grace_period_max,
366 },
367 {
368 .ctl_name = CTL_UNNUMBERED,
369 .procname = "nlm_timeout",
370 .data = &nlm_timeout,
Steve Dickson7ee91ec2005-07-13 01:10:47 -0700371 .maxlen = sizeof(unsigned long),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 .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 .ctl_name = CTL_UNNUMBERED,
379 .procname = "nlm_udpport",
380 .data = &nlm_udpport,
381 .maxlen = sizeof(int),
382 .mode = 0644,
383 .proc_handler = &proc_dointvec_minmax,
384 .extra1 = (int *) &nlm_port_min,
385 .extra2 = (int *) &nlm_port_max,
386 },
387 {
388 .ctl_name = CTL_UNNUMBERED,
389 .procname = "nlm_tcpport",
390 .data = &nlm_tcpport,
391 .maxlen = sizeof(int),
392 .mode = 0644,
393 .proc_handler = &proc_dointvec_minmax,
394 .extra1 = (int *) &nlm_port_min,
395 .extra2 = (int *) &nlm_port_max,
396 },
Olaf Kirchabd1f502006-10-04 02:16:01 -0700397 {
398 .ctl_name = CTL_UNNUMBERED,
399 .procname = "nsm_use_hostnames",
400 .data = &nsm_use_hostnames,
401 .maxlen = sizeof(int),
402 .mode = 0644,
403 .proc_handler = &proc_dointvec,
404 },
Olaf Kirch460f5ca2006-10-04 02:16:03 -0700405 {
406 .ctl_name = CTL_UNNUMBERED,
407 .procname = "nsm_local_state",
408 .data = &nsm_local_state,
409 .maxlen = sizeof(int),
410 .mode = 0644,
411 .proc_handler = &proc_dointvec,
412 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 { .ctl_name = 0 }
414};
415
416static ctl_table nlm_sysctl_dir[] = {
417 {
418 .ctl_name = CTL_UNNUMBERED,
419 .procname = "nfs",
420 .mode = 0555,
421 .child = nlm_sysctls,
422 },
423 { .ctl_name = 0 }
424};
425
426static ctl_table nlm_sysctl_root[] = {
427 {
428 .ctl_name = CTL_FS,
429 .procname = "fs",
430 .mode = 0555,
431 .child = nlm_sysctl_dir,
432 },
433 { .ctl_name = 0 }
434};
435
436/*
437 * Module (and driverfs) parameters.
438 */
439
440#define param_set_min_max(name, type, which_strtol, min, max) \
441static int param_set_##name(const char *val, struct kernel_param *kp) \
442{ \
443 char *endp; \
444 __typeof__(type) num = which_strtol(val, &endp, 0); \
445 if (endp == val || *endp || num < (min) || num > (max)) \
446 return -EINVAL; \
447 *((int *) kp->arg) = num; \
448 return 0; \
449}
450
451static inline int is_callback(u32 proc)
452{
453 return proc == NLMPROC_GRANTED
454 || proc == NLMPROC_GRANTED_MSG
455 || proc == NLMPROC_TEST_RES
456 || proc == NLMPROC_LOCK_RES
457 || proc == NLMPROC_CANCEL_RES
458 || proc == NLMPROC_UNLOCK_RES
459 || proc == NLMPROC_NSM_NOTIFY;
460}
461
462
463static int lockd_authenticate(struct svc_rqst *rqstp)
464{
465 rqstp->rq_client = NULL;
466 switch (rqstp->rq_authop->flavour) {
467 case RPC_AUTH_NULL:
468 case RPC_AUTH_UNIX:
469 if (rqstp->rq_proc == 0)
470 return SVC_OK;
471 if (is_callback(rqstp->rq_proc)) {
472 /* Leave it to individual procedures to
473 * call nlmsvc_lookup_host(rqstp)
474 */
475 return SVC_OK;
476 }
477 return svc_set_client(rqstp);
478 }
479 return SVC_DENIED;
480}
481
482
483param_set_min_max(port, int, simple_strtol, 0, 65535)
484param_set_min_max(grace_period, unsigned long, simple_strtoul,
485 nlm_grace_period_min, nlm_grace_period_max)
486param_set_min_max(timeout, unsigned long, simple_strtoul,
487 nlm_timeout_min, nlm_timeout_max)
488
489MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
490MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
491MODULE_LICENSE("GPL");
492
493module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
494 &nlm_grace_period, 0644);
495module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
496 &nlm_timeout, 0644);
497module_param_call(nlm_udpport, param_set_port, param_get_int,
498 &nlm_udpport, 0644);
499module_param_call(nlm_tcpport, param_set_port, param_get_int,
500 &nlm_tcpport, 0644);
Olaf Kirchabd1f502006-10-04 02:16:01 -0700501module_param(nsm_use_hostnames, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
503/*
504 * Initialising and terminating the module.
505 */
506
507static int __init init_nlm(void)
508{
509 nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root, 0);
510 return nlm_sysctl_table ? 0 : -ENOMEM;
511}
512
513static void __exit exit_nlm(void)
514{
515 /* FIXME: delete all NLM clients */
516 nlm_shutdown_hosts();
517 unregister_sysctl_table(nlm_sysctl_table);
518}
519
520module_init(init_nlm);
521module_exit(exit_nlm);
522
523/*
524 * Define NLM program and procedures
525 */
526static struct svc_version nlmsvc_version1 = {
527 .vs_vers = 1,
528 .vs_nproc = 17,
529 .vs_proc = nlmsvc_procedures,
530 .vs_xdrsize = NLMSVC_XDRSIZE,
531};
532static struct svc_version nlmsvc_version3 = {
533 .vs_vers = 3,
534 .vs_nproc = 24,
535 .vs_proc = nlmsvc_procedures,
536 .vs_xdrsize = NLMSVC_XDRSIZE,
537};
538#ifdef CONFIG_LOCKD_V4
539static struct svc_version nlmsvc_version4 = {
540 .vs_vers = 4,
541 .vs_nproc = 24,
542 .vs_proc = nlmsvc_procedures4,
543 .vs_xdrsize = NLMSVC_XDRSIZE,
544};
545#endif
546static struct svc_version * nlmsvc_version[] = {
547 [1] = &nlmsvc_version1,
548 [3] = &nlmsvc_version3,
549#ifdef CONFIG_LOCKD_V4
550 [4] = &nlmsvc_version4,
551#endif
552};
553
554static struct svc_stat nlmsvc_stats;
555
Tobias Klausere8c96f82006-03-24 03:15:34 -0800556#define NLM_NRVERS ARRAY_SIZE(nlmsvc_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557static struct svc_program nlmsvc_program = {
558 .pg_prog = NLM_PROGRAM, /* program number */
559 .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */
560 .pg_vers = nlmsvc_version, /* version table */
561 .pg_name = "lockd", /* service name */
562 .pg_class = "nfsd", /* share authentication with nfsd */
563 .pg_stats = &nlmsvc_stats, /* stats table */
564 .pg_authenticate = &lockd_authenticate /* export authentication */
565};