Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / fs / lockd / svc.c
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/slab.h>
25 #include <linux/smp.h>
26 #include <linux/smp_lock.h>
27 #include <linux/mutex.h>
28 #include <linux/freezer.h>
29
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 <net/ip.h>
36 #include <linux/lockd/lockd.h>
37 #include <linux/lockd/sm_inter.h>
38 #include <linux/nfs.h>
39
40 #define NLMDBG_FACILITY         NLMDBG_SVC
41 #define LOCKD_BUFSIZE           (1024 + NLMSVC_XDRSIZE)
42 #define ALLOWED_SIGS            (sigmask(SIGKILL))
43
44 static struct svc_program       nlmsvc_program;
45
46 struct nlmsvc_binding *         nlmsvc_ops;
47 EXPORT_SYMBOL(nlmsvc_ops);
48
49 static DEFINE_MUTEX(nlmsvc_mutex);
50 static unsigned int             nlmsvc_users;
51 static pid_t                    nlmsvc_pid;
52 static struct svc_serv          *nlmsvc_serv;
53 int                             nlmsvc_grace_period;
54 unsigned long                   nlmsvc_timeout;
55
56 static DECLARE_COMPLETION(lockd_start_done);
57 static DECLARE_WAIT_QUEUE_HEAD(lockd_exit);
58
59 /*
60  * These can be set at insmod time (useful for NFS as root filesystem),
61  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
62  */
63 static unsigned long            nlm_grace_period;
64 static unsigned long            nlm_timeout = LOCKD_DFLT_TIMEO;
65 static int                      nlm_udpport, nlm_tcpport;
66 int                             nsm_use_hostnames = 0;
67
68 /*
69  * Constants needed for the sysctl interface.
70  */
71 static const unsigned long      nlm_grace_period_min = 0;
72 static const unsigned long      nlm_grace_period_max = 240;
73 static const unsigned long      nlm_timeout_min = 3;
74 static const unsigned long      nlm_timeout_max = 20;
75 static const int                nlm_port_min = 0, nlm_port_max = 65535;
76
77 static struct ctl_table_header * nlm_sysctl_table;
78
79 static unsigned long get_lockd_grace_period(void)
80 {
81         /* Note: nlm_timeout should always be nonzero */
82         if (nlm_grace_period)
83                 return roundup(nlm_grace_period, nlm_timeout) * HZ;
84         else
85                 return nlm_timeout * 5 * HZ;
86 }
87
88 unsigned long get_nfs_grace_period(void)
89 {
90         unsigned long lockdgrace = get_lockd_grace_period();
91         unsigned long nfsdgrace = 0;
92
93         if (nlmsvc_ops)
94                 nfsdgrace = nlmsvc_ops->get_grace_period();
95
96         return max(lockdgrace, nfsdgrace);
97 }
98 EXPORT_SYMBOL(get_nfs_grace_period);
99
100 static unsigned long set_grace_period(void)
101 {
102         nlmsvc_grace_period = 1;
103         return get_nfs_grace_period() + jiffies;
104 }
105
106 static inline void clear_grace_period(void)
107 {
108         nlmsvc_grace_period = 0;
109 }
110
111 /*
112  * This is the lockd kernel thread
113  */
114 static void
115 lockd(struct svc_rqst *rqstp)
116 {
117         int             err = 0;
118         unsigned long grace_period_expire;
119
120         /* Lock module and set up kernel thread */
121         /* lockd_up is waiting for us to startup, so will
122          * be holding a reference to this module, so it
123          * is safe to just claim another reference
124          */
125         __module_get(THIS_MODULE);
126         lock_kernel();
127
128         /*
129          * Let our maker know we're running.
130          */
131         nlmsvc_pid = current->pid;
132         nlmsvc_serv = rqstp->rq_server;
133         complete(&lockd_start_done);
134
135         daemonize("lockd");
136         set_freezable();
137
138         /* Process request with signals blocked, but allow SIGKILL.  */
139         allow_signal(SIGKILL);
140
141         dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
142
143         if (!nlm_timeout)
144                 nlm_timeout = LOCKD_DFLT_TIMEO;
145         nlmsvc_timeout = nlm_timeout * HZ;
146
147         grace_period_expire = set_grace_period();
148
149         /*
150          * The main request loop. We don't terminate until the last
151          * NFS mount or NFS daemon has gone away, and we've been sent a
152          * signal, or else another process has taken over our job.
153          */
154         while ((nlmsvc_users || !signalled()) && nlmsvc_pid == current->pid) {
155                 long timeout = MAX_SCHEDULE_TIMEOUT;
156                 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
157
158                 if (signalled()) {
159                         flush_signals(current);
160                         if (nlmsvc_ops) {
161                                 nlmsvc_invalidate_all();
162                                 grace_period_expire = set_grace_period();
163                         }
164                 }
165
166                 /*
167                  * Retry any blocked locks that have been notified by
168                  * the VFS. Don't do this during grace period.
169                  * (Theoretically, there shouldn't even be blocked locks
170                  * during grace period).
171                  */
172                 if (!nlmsvc_grace_period) {
173                         timeout = nlmsvc_retry_blocked();
174                 } else if (time_before(grace_period_expire, jiffies))
175                         clear_grace_period();
176
177                 /*
178                  * Find a socket with data available and call its
179                  * recvfrom routine.
180                  */
181                 err = svc_recv(rqstp, timeout);
182                 if (err == -EAGAIN || err == -EINTR)
183                         continue;
184                 if (err < 0) {
185                         printk(KERN_WARNING
186                                "lockd: terminating on error %d\n",
187                                -err);
188                         break;
189                 }
190
191                 dprintk("lockd: request from %s\n",
192                                 svc_print_addr(rqstp, buf, sizeof(buf)));
193
194                 svc_process(rqstp);
195         }
196
197         flush_signals(current);
198
199         /*
200          * Check whether there's a new lockd process before
201          * shutting down the hosts and clearing the slot.
202          */
203         if (!nlmsvc_pid || current->pid == nlmsvc_pid) {
204                 if (nlmsvc_ops)
205                         nlmsvc_invalidate_all();
206                 nlm_shutdown_hosts();
207                 nlmsvc_pid = 0;
208                 nlmsvc_serv = NULL;
209         } else
210                 printk(KERN_DEBUG
211                         "lockd: new process, skipping host shutdown\n");
212         wake_up(&lockd_exit);
213
214         /* Exit the RPC thread */
215         svc_exit_thread(rqstp);
216
217         /* Release module */
218         unlock_kernel();
219         module_put_and_exit(0);
220 }
221
222 /*
223  * Make any sockets that are needed but not present.
224  * If nlm_udpport or nlm_tcpport were set as module
225  * options, make those sockets unconditionally
226  */
227 static int make_socks(struct svc_serv *serv, int proto)
228 {
229         static int warned;
230         struct svc_xprt *xprt;
231         int err = 0;
232
233         if (proto == IPPROTO_UDP || nlm_udpport) {
234                 xprt = svc_find_xprt(serv, "udp", 0, 0);
235                 if (!xprt)
236                         err = svc_create_xprt(serv, "udp", nlm_udpport,
237                                               SVC_SOCK_DEFAULTS);
238                 else
239                         svc_xprt_put(xprt);
240         }
241         if (err >= 0 && (proto == IPPROTO_TCP || nlm_tcpport)) {
242                 xprt = svc_find_xprt(serv, "tcp", 0, 0);
243                 if (!xprt)
244                         err = svc_create_xprt(serv, "tcp", nlm_tcpport,
245                                               SVC_SOCK_DEFAULTS);
246                 else
247                         svc_xprt_put(xprt);
248         }
249         if (err >= 0) {
250                 warned = 0;
251                 err = 0;
252         } else if (warned++ == 0)
253                 printk(KERN_WARNING
254                        "lockd_up: makesock failed, error=%d\n", err);
255         return err;
256 }
257
258 /*
259  * Bring up the lockd process if it's not already up.
260  */
261 int
262 lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
263 {
264         struct svc_serv *       serv;
265         int                     error = 0;
266
267         mutex_lock(&nlmsvc_mutex);
268         /*
269          * Check whether we're already up and running.
270          */
271         if (nlmsvc_pid) {
272                 if (proto)
273                         error = make_socks(nlmsvc_serv, proto);
274                 goto out;
275         }
276
277         /*
278          * Sanity check: if there's no pid,
279          * we should be the first user ...
280          */
281         if (nlmsvc_users)
282                 printk(KERN_WARNING
283                         "lockd_up: no pid, %d users??\n", nlmsvc_users);
284
285         error = -ENOMEM;
286         serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
287         if (!serv) {
288                 printk(KERN_WARNING "lockd_up: create service failed\n");
289                 goto out;
290         }
291
292         if ((error = make_socks(serv, proto)) < 0)
293                 goto destroy_and_out;
294
295         /*
296          * Create the kernel thread and wait for it to start.
297          */
298         error = svc_create_thread(lockd, serv);
299         if (error) {
300                 printk(KERN_WARNING
301                         "lockd_up: create thread failed, error=%d\n", error);
302                 goto destroy_and_out;
303         }
304         wait_for_completion(&lockd_start_done);
305
306         /*
307          * Note: svc_serv structures have an initial use count of 1,
308          * so we exit through here on both success and failure.
309          */
310 destroy_and_out:
311         svc_destroy(serv);
312 out:
313         if (!error)
314                 nlmsvc_users++;
315         mutex_unlock(&nlmsvc_mutex);
316         return error;
317 }
318 EXPORT_SYMBOL(lockd_up);
319
320 /*
321  * Decrement the user count and bring down lockd if we're the last.
322  */
323 void
324 lockd_down(void)
325 {
326         static int warned;
327
328         mutex_lock(&nlmsvc_mutex);
329         if (nlmsvc_users) {
330                 if (--nlmsvc_users)
331                         goto out;
332         } else
333                 printk(KERN_WARNING "lockd_down: no users! pid=%d\n", nlmsvc_pid);
334
335         if (!nlmsvc_pid) {
336                 if (warned++ == 0)
337                         printk(KERN_WARNING "lockd_down: no lockd running.\n"); 
338                 goto out;
339         }
340         warned = 0;
341
342         kill_proc(nlmsvc_pid, SIGKILL, 1);
343         /*
344          * Wait for the lockd process to exit, but since we're holding
345          * the lockd semaphore, we can't wait around forever ...
346          */
347         clear_thread_flag(TIF_SIGPENDING);
348         interruptible_sleep_on_timeout(&lockd_exit, HZ);
349         if (nlmsvc_pid) {
350                 printk(KERN_WARNING 
351                         "lockd_down: lockd failed to exit, clearing pid\n");
352                 nlmsvc_pid = 0;
353         }
354         spin_lock_irq(&current->sighand->siglock);
355         recalc_sigpending();
356         spin_unlock_irq(&current->sighand->siglock);
357 out:
358         mutex_unlock(&nlmsvc_mutex);
359 }
360 EXPORT_SYMBOL(lockd_down);
361
362 /*
363  * Sysctl parameters (same as module parameters, different interface).
364  */
365
366 static ctl_table nlm_sysctls[] = {
367         {
368                 .ctl_name       = CTL_UNNUMBERED,
369                 .procname       = "nlm_grace_period",
370                 .data           = &nlm_grace_period,
371                 .maxlen         = sizeof(unsigned long),
372                 .mode           = 0644,
373                 .proc_handler   = &proc_doulongvec_minmax,
374                 .extra1         = (unsigned long *) &nlm_grace_period_min,
375                 .extra2         = (unsigned long *) &nlm_grace_period_max,
376         },
377         {
378                 .ctl_name       = CTL_UNNUMBERED,
379                 .procname       = "nlm_timeout",
380                 .data           = &nlm_timeout,
381                 .maxlen         = sizeof(unsigned long),
382                 .mode           = 0644,
383                 .proc_handler   = &proc_doulongvec_minmax,
384                 .extra1         = (unsigned long *) &nlm_timeout_min,
385                 .extra2         = (unsigned long *) &nlm_timeout_max,
386         },
387         {
388                 .ctl_name       = CTL_UNNUMBERED,
389                 .procname       = "nlm_udpport",
390                 .data           = &nlm_udpport,
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         },
397         {
398                 .ctl_name       = CTL_UNNUMBERED,
399                 .procname       = "nlm_tcpport",
400                 .data           = &nlm_tcpport,
401                 .maxlen         = sizeof(int),
402                 .mode           = 0644,
403                 .proc_handler   = &proc_dointvec_minmax,
404                 .extra1         = (int *) &nlm_port_min,
405                 .extra2         = (int *) &nlm_port_max,
406         },
407         {
408                 .ctl_name       = CTL_UNNUMBERED,
409                 .procname       = "nsm_use_hostnames",
410                 .data           = &nsm_use_hostnames,
411                 .maxlen         = sizeof(int),
412                 .mode           = 0644,
413                 .proc_handler   = &proc_dointvec,
414         },
415         {
416                 .ctl_name       = CTL_UNNUMBERED,
417                 .procname       = "nsm_local_state",
418                 .data           = &nsm_local_state,
419                 .maxlen         = sizeof(int),
420                 .mode           = 0644,
421                 .proc_handler   = &proc_dointvec,
422         },
423         { .ctl_name = 0 }
424 };
425
426 static ctl_table nlm_sysctl_dir[] = {
427         {
428                 .ctl_name       = CTL_UNNUMBERED,
429                 .procname       = "nfs",
430                 .mode           = 0555,
431                 .child          = nlm_sysctls,
432         },
433         { .ctl_name = 0 }
434 };
435
436 static ctl_table nlm_sysctl_root[] = {
437         {
438                 .ctl_name       = CTL_FS,
439                 .procname       = "fs",
440                 .mode           = 0555,
441                 .child          = nlm_sysctl_dir,
442         },
443         { .ctl_name = 0 }
444 };
445
446 /*
447  * Module (and sysfs) parameters.
448  */
449
450 #define param_set_min_max(name, type, which_strtol, min, max)           \
451 static int param_set_##name(const char *val, struct kernel_param *kp)   \
452 {                                                                       \
453         char *endp;                                                     \
454         __typeof__(type) num = which_strtol(val, &endp, 0);             \
455         if (endp == val || *endp || num < (min) || num > (max))         \
456                 return -EINVAL;                                         \
457         *((int *) kp->arg) = num;                                       \
458         return 0;                                                       \
459 }
460
461 static inline int is_callback(u32 proc)
462 {
463         return proc == NLMPROC_GRANTED
464                 || proc == NLMPROC_GRANTED_MSG
465                 || proc == NLMPROC_TEST_RES
466                 || proc == NLMPROC_LOCK_RES
467                 || proc == NLMPROC_CANCEL_RES
468                 || proc == NLMPROC_UNLOCK_RES
469                 || proc == NLMPROC_NSM_NOTIFY;
470 }
471
472
473 static int lockd_authenticate(struct svc_rqst *rqstp)
474 {
475         rqstp->rq_client = NULL;
476         switch (rqstp->rq_authop->flavour) {
477                 case RPC_AUTH_NULL:
478                 case RPC_AUTH_UNIX:
479                         if (rqstp->rq_proc == 0)
480                                 return SVC_OK;
481                         if (is_callback(rqstp->rq_proc)) {
482                                 /* Leave it to individual procedures to
483                                  * call nlmsvc_lookup_host(rqstp)
484                                  */
485                                 return SVC_OK;
486                         }
487                         return svc_set_client(rqstp);
488         }
489         return SVC_DENIED;
490 }
491
492
493 param_set_min_max(port, int, simple_strtol, 0, 65535)
494 param_set_min_max(grace_period, unsigned long, simple_strtoul,
495                   nlm_grace_period_min, nlm_grace_period_max)
496 param_set_min_max(timeout, unsigned long, simple_strtoul,
497                   nlm_timeout_min, nlm_timeout_max)
498
499 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
500 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
501 MODULE_LICENSE("GPL");
502
503 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
504                   &nlm_grace_period, 0644);
505 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
506                   &nlm_timeout, 0644);
507 module_param_call(nlm_udpport, param_set_port, param_get_int,
508                   &nlm_udpport, 0644);
509 module_param_call(nlm_tcpport, param_set_port, param_get_int,
510                   &nlm_tcpport, 0644);
511 module_param(nsm_use_hostnames, bool, 0644);
512
513 /*
514  * Initialising and terminating the module.
515  */
516
517 static int __init init_nlm(void)
518 {
519         nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
520         return nlm_sysctl_table ? 0 : -ENOMEM;
521 }
522
523 static void __exit exit_nlm(void)
524 {
525         /* FIXME: delete all NLM clients */
526         nlm_shutdown_hosts();
527         unregister_sysctl_table(nlm_sysctl_table);
528 }
529
530 module_init(init_nlm);
531 module_exit(exit_nlm);
532
533 /*
534  * Define NLM program and procedures
535  */
536 static struct svc_version       nlmsvc_version1 = {
537                 .vs_vers        = 1,
538                 .vs_nproc       = 17,
539                 .vs_proc        = nlmsvc_procedures,
540                 .vs_xdrsize     = NLMSVC_XDRSIZE,
541 };
542 static struct svc_version       nlmsvc_version3 = {
543                 .vs_vers        = 3,
544                 .vs_nproc       = 24,
545                 .vs_proc        = nlmsvc_procedures,
546                 .vs_xdrsize     = NLMSVC_XDRSIZE,
547 };
548 #ifdef CONFIG_LOCKD_V4
549 static struct svc_version       nlmsvc_version4 = {
550                 .vs_vers        = 4,
551                 .vs_nproc       = 24,
552                 .vs_proc        = nlmsvc_procedures4,
553                 .vs_xdrsize     = NLMSVC_XDRSIZE,
554 };
555 #endif
556 static struct svc_version *     nlmsvc_version[] = {
557         [1] = &nlmsvc_version1,
558         [3] = &nlmsvc_version3,
559 #ifdef CONFIG_LOCKD_V4
560         [4] = &nlmsvc_version4,
561 #endif
562 };
563
564 static struct svc_stat          nlmsvc_stats;
565
566 #define NLM_NRVERS      ARRAY_SIZE(nlmsvc_version)
567 static struct svc_program       nlmsvc_program = {
568         .pg_prog                = NLM_PROGRAM,          /* program number */
569         .pg_nvers               = NLM_NRVERS,           /* number of entries in nlmsvc_version */
570         .pg_vers                = nlmsvc_version,       /* version table */
571         .pg_name                = "lockd",              /* service name */
572         .pg_class               = "nfsd",               /* share authentication with nfsd */
573         .pg_stats               = &nlmsvc_stats,        /* stats table */
574         .pg_authenticate = &lockd_authenticate  /* export authentication */
575 };