NLM: don't let lockd exit on unexpected svc_recv errors (try #2)
[safe/jmp/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/kthread.h>
29 #include <linux/freezer.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 <net/ip.h>
37 #include <linux/lockd/lockd.h>
38 #include <linux/lockd/sm_inter.h>
39 #include <linux/nfs.h>
40
41 #define NLMDBG_FACILITY         NLMDBG_SVC
42 #define LOCKD_BUFSIZE           (1024 + NLMSVC_XDRSIZE)
43 #define ALLOWED_SIGS            (sigmask(SIGKILL))
44
45 static struct svc_program       nlmsvc_program;
46
47 struct nlmsvc_binding *         nlmsvc_ops;
48 EXPORT_SYMBOL(nlmsvc_ops);
49
50 static DEFINE_MUTEX(nlmsvc_mutex);
51 static unsigned int             nlmsvc_users;
52 static struct task_struct       *nlmsvc_task;
53 static struct svc_serv          *nlmsvc_serv;
54 int                             nlmsvc_grace_period;
55 unsigned long                   nlmsvc_timeout;
56
57 /*
58  * These can be set at insmod time (useful for NFS as root filesystem),
59  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
60  */
61 static unsigned long            nlm_grace_period;
62 static unsigned long            nlm_timeout = LOCKD_DFLT_TIMEO;
63 static int                      nlm_udpport, nlm_tcpport;
64 int                             nsm_use_hostnames = 0;
65
66 /*
67  * Constants needed for the sysctl interface.
68  */
69 static const unsigned long      nlm_grace_period_min = 0;
70 static const unsigned long      nlm_grace_period_max = 240;
71 static const unsigned long      nlm_timeout_min = 3;
72 static const unsigned long      nlm_timeout_max = 20;
73 static const int                nlm_port_min = 0, nlm_port_max = 65535;
74
75 static struct ctl_table_header * nlm_sysctl_table;
76
77 static 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
86 unsigned long get_nfs_grace_period(void)
87 {
88         unsigned long lockdgrace = get_lockd_grace_period();
89         unsigned long nfsdgrace = 0;
90
91         if (nlmsvc_ops)
92                 nfsdgrace = nlmsvc_ops->get_grace_period();
93
94         return max(lockdgrace, nfsdgrace);
95 }
96 EXPORT_SYMBOL(get_nfs_grace_period);
97
98 static unsigned long set_grace_period(void)
99 {
100         nlmsvc_grace_period = 1;
101         return get_nfs_grace_period() + jiffies;
102 }
103
104 static inline void clear_grace_period(void)
105 {
106         nlmsvc_grace_period = 0;
107 }
108
109 /*
110  * This is the lockd kernel thread
111  */
112 static int
113 lockd(void *vrqstp)
114 {
115         int             err = 0, preverr = 0;
116         struct svc_rqst *rqstp = vrqstp;
117         unsigned long grace_period_expire;
118
119         /* try_to_freeze() is called from svc_recv() */
120         set_freezable();
121
122         /* Allow SIGKILL to tell lockd to drop all of its locks */
123         allow_signal(SIGKILL);
124
125         dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
126
127         /*
128          * FIXME: it would be nice if lockd didn't spend its entire life
129          * running under the BKL. At the very least, it would be good to
130          * have someone clarify what it's intended to protect here. I've
131          * seen some handwavy posts about posix locking needing to be
132          * done under the BKL, but it's far from clear.
133          */
134         lock_kernel();
135
136         if (!nlm_timeout)
137                 nlm_timeout = LOCKD_DFLT_TIMEO;
138         nlmsvc_timeout = nlm_timeout * HZ;
139
140         grace_period_expire = set_grace_period();
141
142         /*
143          * The main request loop. We don't terminate until the last
144          * NFS mount or NFS daemon has gone away.
145          */
146         while (!kthread_should_stop()) {
147                 long timeout = MAX_SCHEDULE_TIMEOUT;
148                 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
149
150                 if (signalled()) {
151                         flush_signals(current);
152                         if (nlmsvc_ops) {
153                                 nlmsvc_invalidate_all();
154                                 grace_period_expire = set_grace_period();
155                         }
156                         continue;
157                 }
158
159                 /*
160                  * Retry any blocked locks that have been notified by
161                  * the VFS. Don't do this during grace period.
162                  * (Theoretically, there shouldn't even be blocked locks
163                  * during grace period).
164                  */
165                 if (!nlmsvc_grace_period) {
166                         timeout = nlmsvc_retry_blocked();
167                 } else if (time_before(grace_period_expire, jiffies))
168                         clear_grace_period();
169
170                 /*
171                  * Find a socket with data available and call its
172                  * recvfrom routine.
173                  */
174                 err = svc_recv(rqstp, timeout);
175                 if (err == -EAGAIN || err == -EINTR) {
176                         preverr = err;
177                         continue;
178                 }
179                 if (err < 0) {
180                         if (err != preverr) {
181                                 printk(KERN_WARNING "%s: unexpected error "
182                                         "from svc_recv (%d)\n", __func__, err);
183                                 preverr = err;
184                         }
185                         schedule_timeout_interruptible(HZ);
186                         continue;
187                 }
188                 preverr = err;
189
190                 dprintk("lockd: request from %s\n",
191                                 svc_print_addr(rqstp, buf, sizeof(buf)));
192
193                 svc_process(rqstp);
194         }
195
196         flush_signals(current);
197         if (nlmsvc_ops)
198                 nlmsvc_invalidate_all();
199         nlm_shutdown_hosts();
200
201         unlock_kernel();
202
203         nlmsvc_task = NULL;
204         nlmsvc_serv = NULL;
205
206         /* Exit the RPC thread */
207         svc_exit_thread(rqstp);
208
209         return 0;
210 }
211
212 /*
213  * Make any sockets that are needed but not present.
214  * If nlm_udpport or nlm_tcpport were set as module
215  * options, make those sockets unconditionally
216  */
217 static int make_socks(struct svc_serv *serv, int proto)
218 {
219         static int warned;
220         struct svc_xprt *xprt;
221         int err = 0;
222
223         if (proto == IPPROTO_UDP || nlm_udpport) {
224                 xprt = svc_find_xprt(serv, "udp", 0, 0);
225                 if (!xprt)
226                         err = svc_create_xprt(serv, "udp", nlm_udpport,
227                                               SVC_SOCK_DEFAULTS);
228                 else
229                         svc_xprt_put(xprt);
230         }
231         if (err >= 0 && (proto == IPPROTO_TCP || nlm_tcpport)) {
232                 xprt = svc_find_xprt(serv, "tcp", 0, 0);
233                 if (!xprt)
234                         err = svc_create_xprt(serv, "tcp", nlm_tcpport,
235                                               SVC_SOCK_DEFAULTS);
236                 else
237                         svc_xprt_put(xprt);
238         }
239         if (err >= 0) {
240                 warned = 0;
241                 err = 0;
242         } else 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  */
251 int
252 lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
253 {
254         struct svc_serv *serv;
255         struct svc_rqst *rqstp;
256         int             error = 0;
257
258         mutex_lock(&nlmsvc_mutex);
259         /*
260          * Check whether we're already up and running.
261          */
262         if (nlmsvc_serv) {
263                 if (proto)
264                         error = make_socks(nlmsvc_serv, proto);
265                 goto out;
266         }
267
268         /*
269          * Sanity check: if there's no pid,
270          * we should be the first user ...
271          */
272         if (nlmsvc_users)
273                 printk(KERN_WARNING
274                         "lockd_up: no pid, %d users??\n", nlmsvc_users);
275
276         error = -ENOMEM;
277         serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
278         if (!serv) {
279                 printk(KERN_WARNING "lockd_up: create service failed\n");
280                 goto out;
281         }
282
283         if ((error = make_socks(serv, proto)) < 0)
284                 goto destroy_and_out;
285
286         /*
287          * Create the kernel thread and wait for it to start.
288          */
289         rqstp = svc_prepare_thread(serv, &serv->sv_pools[0]);
290         if (IS_ERR(rqstp)) {
291                 error = PTR_ERR(rqstp);
292                 printk(KERN_WARNING
293                         "lockd_up: svc_rqst allocation failed, error=%d\n",
294                         error);
295                 goto destroy_and_out;
296         }
297
298         svc_sock_update_bufs(serv);
299         nlmsvc_serv = rqstp->rq_server;
300
301         nlmsvc_task = kthread_run(lockd, rqstp, serv->sv_name);
302         if (IS_ERR(nlmsvc_task)) {
303                 error = PTR_ERR(nlmsvc_task);
304                 nlmsvc_task = NULL;
305                 nlmsvc_serv = NULL;
306                 printk(KERN_WARNING
307                         "lockd_up: kthread_run failed, error=%d\n", error);
308                 svc_exit_thread(rqstp);
309                 goto destroy_and_out;
310         }
311
312         /*
313          * Note: svc_serv structures have an initial use count of 1,
314          * so we exit through here on both success and failure.
315          */
316 destroy_and_out:
317         svc_destroy(serv);
318 out:
319         if (!error)
320                 nlmsvc_users++;
321         mutex_unlock(&nlmsvc_mutex);
322         return error;
323 }
324 EXPORT_SYMBOL(lockd_up);
325
326 /*
327  * Decrement the user count and bring down lockd if we're the last.
328  */
329 void
330 lockd_down(void)
331 {
332         mutex_lock(&nlmsvc_mutex);
333         if (nlmsvc_users) {
334                 if (--nlmsvc_users)
335                         goto out;
336         } else {
337                 printk(KERN_ERR "lockd_down: no users! task=%p\n",
338                         nlmsvc_task);
339                 BUG();
340         }
341
342         if (!nlmsvc_task) {
343                 printk(KERN_ERR "lockd_down: no lockd running.\n");
344                 BUG();
345         }
346         kthread_stop(nlmsvc_task);
347 out:
348         mutex_unlock(&nlmsvc_mutex);
349 }
350 EXPORT_SYMBOL(lockd_down);
351
352 /*
353  * Sysctl parameters (same as module parameters, different interface).
354  */
355
356 static ctl_table nlm_sysctls[] = {
357         {
358                 .ctl_name       = CTL_UNNUMBERED,
359                 .procname       = "nlm_grace_period",
360                 .data           = &nlm_grace_period,
361                 .maxlen         = sizeof(unsigned long),
362                 .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,
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                 .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         },
397         {
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         },
405         {
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         },
413         { .ctl_name = 0 }
414 };
415
416 static 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
426 static 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 sysfs) parameters.
438  */
439
440 #define param_set_min_max(name, type, which_strtol, min, max)           \
441 static 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
451 static 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
463 static 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
483 param_set_min_max(port, int, simple_strtol, 0, 65535)
484 param_set_min_max(grace_period, unsigned long, simple_strtoul,
485                   nlm_grace_period_min, nlm_grace_period_max)
486 param_set_min_max(timeout, unsigned long, simple_strtoul,
487                   nlm_timeout_min, nlm_timeout_max)
488
489 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
490 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
491 MODULE_LICENSE("GPL");
492
493 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
494                   &nlm_grace_period, 0644);
495 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
496                   &nlm_timeout, 0644);
497 module_param_call(nlm_udpport, param_set_port, param_get_int,
498                   &nlm_udpport, 0644);
499 module_param_call(nlm_tcpport, param_set_port, param_get_int,
500                   &nlm_tcpport, 0644);
501 module_param(nsm_use_hostnames, bool, 0644);
502
503 /*
504  * Initialising and terminating the module.
505  */
506
507 static int __init init_nlm(void)
508 {
509         nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
510         return nlm_sysctl_table ? 0 : -ENOMEM;
511 }
512
513 static 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
520 module_init(init_nlm);
521 module_exit(exit_nlm);
522
523 /*
524  * Define NLM program and procedures
525  */
526 static struct svc_version       nlmsvc_version1 = {
527                 .vs_vers        = 1,
528                 .vs_nproc       = 17,
529                 .vs_proc        = nlmsvc_procedures,
530                 .vs_xdrsize     = NLMSVC_XDRSIZE,
531 };
532 static 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
539 static 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
546 static 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
554 static struct svc_stat          nlmsvc_stats;
555
556 #define NLM_NRVERS      ARRAY_SIZE(nlmsvc_version)
557 static 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 };