Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
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> | |
353ab6e9 | 27 | #include <linux/mutex.h> |
1da177e4 LT |
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> | |
24e36663 | 34 | #include <net/ip.h> |
1da177e4 LT |
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 | ||
42 | static struct svc_program nlmsvc_program; | |
43 | ||
44 | struct nlmsvc_binding * nlmsvc_ops; | |
45 | EXPORT_SYMBOL(nlmsvc_ops); | |
46 | ||
353ab6e9 | 47 | static DEFINE_MUTEX(nlmsvc_mutex); |
1da177e4 LT |
48 | static unsigned int nlmsvc_users; |
49 | static pid_t nlmsvc_pid; | |
24e36663 | 50 | static struct svc_serv *nlmsvc_serv; |
1da177e4 LT |
51 | int nlmsvc_grace_period; |
52 | unsigned long nlmsvc_timeout; | |
53 | ||
353ab6e9 | 54 | static DECLARE_COMPLETION(lockd_start_done); |
1da177e4 LT |
55 | static DECLARE_WAIT_QUEUE_HEAD(lockd_exit); |
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 | ||
65 | /* | |
66 | * Constants needed for the sysctl interface. | |
67 | */ | |
68 | static const unsigned long nlm_grace_period_min = 0; | |
69 | static const unsigned long nlm_grace_period_max = 240; | |
70 | static const unsigned long nlm_timeout_min = 3; | |
71 | static const unsigned long nlm_timeout_max = 20; | |
72 | static const int nlm_port_min = 0, nlm_port_max = 65535; | |
73 | ||
74 | static struct ctl_table_header * nlm_sysctl_table; | |
75 | ||
76 | static unsigned long set_grace_period(void) | |
77 | { | |
78 | unsigned long grace_period; | |
79 | ||
80 | /* Note: nlm_timeout should always be nonzero */ | |
81 | if (nlm_grace_period) | |
82 | grace_period = ((nlm_grace_period + nlm_timeout - 1) | |
83 | / nlm_timeout) * nlm_timeout * HZ; | |
84 | else | |
85 | grace_period = nlm_timeout * 5 * HZ; | |
86 | nlmsvc_grace_period = 1; | |
87 | return grace_period + jiffies; | |
88 | } | |
89 | ||
90 | static inline void clear_grace_period(void) | |
91 | { | |
92 | nlmsvc_grace_period = 0; | |
93 | } | |
94 | ||
95 | /* | |
96 | * This is the lockd kernel thread | |
97 | */ | |
98 | static void | |
99 | lockd(struct svc_rqst *rqstp) | |
100 | { | |
101 | struct svc_serv *serv = rqstp->rq_server; | |
102 | int err = 0; | |
103 | unsigned long grace_period_expire; | |
104 | ||
105 | /* Lock module and set up kernel thread */ | |
106 | /* lockd_up is waiting for us to startup, so will | |
107 | * be holding a reference to this module, so it | |
108 | * is safe to just claim another reference | |
109 | */ | |
110 | __module_get(THIS_MODULE); | |
111 | lock_kernel(); | |
112 | ||
113 | /* | |
114 | * Let our maker know we're running. | |
115 | */ | |
116 | nlmsvc_pid = current->pid; | |
24e36663 | 117 | nlmsvc_serv = serv; |
353ab6e9 | 118 | complete(&lockd_start_done); |
1da177e4 LT |
119 | |
120 | daemonize("lockd"); | |
121 | ||
122 | /* Process request with signals blocked, but allow SIGKILL. */ | |
123 | allow_signal(SIGKILL); | |
124 | ||
125 | /* kick rpciod */ | |
126 | rpciod_up(); | |
127 | ||
128 | dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n"); | |
129 | ||
130 | if (!nlm_timeout) | |
131 | nlm_timeout = LOCKD_DFLT_TIMEO; | |
132 | nlmsvc_timeout = nlm_timeout * HZ; | |
133 | ||
134 | grace_period_expire = set_grace_period(); | |
135 | ||
136 | /* | |
137 | * The main request loop. We don't terminate until the last | |
138 | * NFS mount or NFS daemon has gone away, and we've been sent a | |
139 | * signal, or else another process has taken over our job. | |
140 | */ | |
141 | while ((nlmsvc_users || !signalled()) && nlmsvc_pid == current->pid) { | |
142 | long timeout = MAX_SCHEDULE_TIMEOUT; | |
143 | ||
144 | if (signalled()) { | |
145 | flush_signals(current); | |
146 | if (nlmsvc_ops) { | |
147 | nlmsvc_invalidate_all(); | |
148 | grace_period_expire = set_grace_period(); | |
149 | } | |
150 | } | |
151 | ||
152 | /* | |
153 | * Retry any blocked locks that have been notified by | |
154 | * the VFS. Don't do this during grace period. | |
155 | * (Theoretically, there shouldn't even be blocked locks | |
156 | * during grace period). | |
157 | */ | |
158 | if (!nlmsvc_grace_period) { | |
159 | timeout = nlmsvc_retry_blocked(); | |
160 | } else if (time_before(grace_period_expire, jiffies)) | |
161 | clear_grace_period(); | |
162 | ||
163 | /* | |
164 | * Find a socket with data available and call its | |
165 | * recvfrom routine. | |
166 | */ | |
167 | err = svc_recv(serv, rqstp, timeout); | |
168 | if (err == -EAGAIN || err == -EINTR) | |
169 | continue; | |
170 | if (err < 0) { | |
171 | printk(KERN_WARNING | |
172 | "lockd: terminating on error %d\n", | |
173 | -err); | |
174 | break; | |
175 | } | |
176 | ||
177 | dprintk("lockd: request from %08x\n", | |
178 | (unsigned)ntohl(rqstp->rq_addr.sin_addr.s_addr)); | |
179 | ||
180 | svc_process(serv, rqstp); | |
181 | ||
182 | } | |
183 | ||
2bd61579 TM |
184 | flush_signals(current); |
185 | ||
1da177e4 LT |
186 | /* |
187 | * Check whether there's a new lockd process before | |
188 | * shutting down the hosts and clearing the slot. | |
189 | */ | |
190 | if (!nlmsvc_pid || current->pid == nlmsvc_pid) { | |
191 | if (nlmsvc_ops) | |
192 | nlmsvc_invalidate_all(); | |
193 | nlm_shutdown_hosts(); | |
194 | nlmsvc_pid = 0; | |
24e36663 | 195 | nlmsvc_serv = NULL; |
1da177e4 LT |
196 | } else |
197 | printk(KERN_DEBUG | |
198 | "lockd: new process, skipping host shutdown\n"); | |
199 | wake_up(&lockd_exit); | |
46be925f | 200 | |
1da177e4 LT |
201 | /* Exit the RPC thread */ |
202 | svc_exit_thread(rqstp); | |
203 | ||
204 | /* release rpciod */ | |
205 | rpciod_down(); | |
206 | ||
207 | /* Release module */ | |
208 | unlock_kernel(); | |
209 | module_put_and_exit(0); | |
210 | } | |
211 | ||
24e36663 N |
212 | |
213 | static int find_socket(struct svc_serv *serv, int proto) | |
214 | { | |
215 | struct svc_sock *svsk; | |
216 | int found = 0; | |
217 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) | |
218 | if (svsk->sk_sk->sk_protocol == proto) { | |
219 | found = 1; | |
220 | break; | |
221 | } | |
222 | return found; | |
223 | } | |
224 | ||
225 | static int make_socks(struct svc_serv *serv, int proto) | |
226 | { | |
227 | /* Make any sockets that are needed but not present. | |
228 | * If nlm_udpport or nlm_tcpport were set as module | |
229 | * options, make those sockets unconditionally | |
230 | */ | |
231 | int err = 0; | |
232 | if (proto == IPPROTO_UDP || nlm_udpport) | |
233 | if (!find_socket(serv, IPPROTO_UDP)) | |
234 | err = svc_makesock(serv, IPPROTO_UDP, nlm_udpport); | |
235 | if (err) | |
236 | return err; | |
237 | if (proto == IPPROTO_TCP || nlm_tcpport) | |
238 | if (!find_socket(serv, IPPROTO_TCP)) | |
239 | err= svc_makesock(serv, IPPROTO_TCP, nlm_tcpport); | |
240 | return err; | |
241 | } | |
242 | ||
1da177e4 LT |
243 | /* |
244 | * Bring up the lockd process if it's not already up. | |
245 | */ | |
246 | int | |
24e36663 | 247 | lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */ |
1da177e4 LT |
248 | { |
249 | static int warned; | |
250 | struct svc_serv * serv; | |
251 | int error = 0; | |
252 | ||
353ab6e9 | 253 | mutex_lock(&nlmsvc_mutex); |
1da177e4 LT |
254 | /* |
255 | * Unconditionally increment the user count ... this is | |
256 | * the number of clients who _want_ a lockd process. | |
257 | */ | |
258 | nlmsvc_users++; | |
259 | /* | |
260 | * Check whether we're already up and running. | |
261 | */ | |
24e36663 N |
262 | if (nlmsvc_pid) { |
263 | error = make_socks(nlmsvc_serv, proto); | |
1da177e4 | 264 | goto out; |
24e36663 | 265 | } |
1da177e4 LT |
266 | |
267 | /* | |
268 | * Sanity check: if there's no pid, | |
269 | * we should be the first user ... | |
270 | */ | |
271 | if (nlmsvc_users > 1) | |
272 | printk(KERN_WARNING | |
273 | "lockd_up: no pid, %d users??\n", nlmsvc_users); | |
274 | ||
275 | error = -ENOMEM; | |
bc591ccf | 276 | serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL); |
1da177e4 LT |
277 | if (!serv) { |
278 | printk(KERN_WARNING "lockd_up: create service failed\n"); | |
279 | goto out; | |
280 | } | |
281 | ||
24e36663 | 282 | if ((error = make_socks(serv, proto)) < 0) { |
1da177e4 LT |
283 | if (warned++ == 0) |
284 | printk(KERN_WARNING | |
285 | "lockd_up: makesock failed, error=%d\n", error); | |
286 | goto destroy_and_out; | |
287 | } | |
288 | warned = 0; | |
289 | ||
290 | /* | |
291 | * Create the kernel thread and wait for it to start. | |
292 | */ | |
293 | error = svc_create_thread(lockd, serv); | |
294 | if (error) { | |
295 | printk(KERN_WARNING | |
296 | "lockd_up: create thread failed, error=%d\n", error); | |
297 | goto destroy_and_out; | |
298 | } | |
353ab6e9 | 299 | wait_for_completion(&lockd_start_done); |
1da177e4 LT |
300 | |
301 | /* | |
302 | * Note: svc_serv structures have an initial use count of 1, | |
303 | * so we exit through here on both success and failure. | |
304 | */ | |
305 | destroy_and_out: | |
306 | svc_destroy(serv); | |
307 | out: | |
353ab6e9 | 308 | mutex_unlock(&nlmsvc_mutex); |
1da177e4 LT |
309 | return error; |
310 | } | |
311 | EXPORT_SYMBOL(lockd_up); | |
312 | ||
313 | /* | |
314 | * Decrement the user count and bring down lockd if we're the last. | |
315 | */ | |
316 | void | |
317 | lockd_down(void) | |
318 | { | |
319 | static int warned; | |
320 | ||
353ab6e9 | 321 | mutex_lock(&nlmsvc_mutex); |
1da177e4 LT |
322 | if (nlmsvc_users) { |
323 | if (--nlmsvc_users) | |
324 | goto out; | |
325 | } else | |
326 | printk(KERN_WARNING "lockd_down: no users! pid=%d\n", nlmsvc_pid); | |
327 | ||
328 | if (!nlmsvc_pid) { | |
329 | if (warned++ == 0) | |
330 | printk(KERN_WARNING "lockd_down: no lockd running.\n"); | |
331 | goto out; | |
332 | } | |
333 | warned = 0; | |
334 | ||
335 | kill_proc(nlmsvc_pid, SIGKILL, 1); | |
336 | /* | |
337 | * Wait for the lockd process to exit, but since we're holding | |
338 | * the lockd semaphore, we can't wait around forever ... | |
339 | */ | |
340 | clear_thread_flag(TIF_SIGPENDING); | |
341 | interruptible_sleep_on_timeout(&lockd_exit, HZ); | |
342 | if (nlmsvc_pid) { | |
343 | printk(KERN_WARNING | |
344 | "lockd_down: lockd failed to exit, clearing pid\n"); | |
345 | nlmsvc_pid = 0; | |
346 | } | |
347 | spin_lock_irq(¤t->sighand->siglock); | |
348 | recalc_sigpending(); | |
349 | spin_unlock_irq(¤t->sighand->siglock); | |
350 | out: | |
353ab6e9 | 351 | mutex_unlock(&nlmsvc_mutex); |
1da177e4 LT |
352 | } |
353 | EXPORT_SYMBOL(lockd_down); | |
354 | ||
355 | /* | |
356 | * Sysctl parameters (same as module parameters, different interface). | |
357 | */ | |
358 | ||
359 | /* Something that isn't CTL_ANY, CTL_NONE or a value that may clash. */ | |
360 | #define CTL_UNNUMBERED -2 | |
361 | ||
362 | static ctl_table nlm_sysctls[] = { | |
363 | { | |
364 | .ctl_name = CTL_UNNUMBERED, | |
365 | .procname = "nlm_grace_period", | |
366 | .data = &nlm_grace_period, | |
7ee91ec1 | 367 | .maxlen = sizeof(unsigned long), |
1da177e4 LT |
368 | .mode = 0644, |
369 | .proc_handler = &proc_doulongvec_minmax, | |
370 | .extra1 = (unsigned long *) &nlm_grace_period_min, | |
371 | .extra2 = (unsigned long *) &nlm_grace_period_max, | |
372 | }, | |
373 | { | |
374 | .ctl_name = CTL_UNNUMBERED, | |
375 | .procname = "nlm_timeout", | |
376 | .data = &nlm_timeout, | |
7ee91ec1 | 377 | .maxlen = sizeof(unsigned long), |
1da177e4 LT |
378 | .mode = 0644, |
379 | .proc_handler = &proc_doulongvec_minmax, | |
380 | .extra1 = (unsigned long *) &nlm_timeout_min, | |
381 | .extra2 = (unsigned long *) &nlm_timeout_max, | |
382 | }, | |
383 | { | |
384 | .ctl_name = CTL_UNNUMBERED, | |
385 | .procname = "nlm_udpport", | |
386 | .data = &nlm_udpport, | |
387 | .maxlen = sizeof(int), | |
388 | .mode = 0644, | |
389 | .proc_handler = &proc_dointvec_minmax, | |
390 | .extra1 = (int *) &nlm_port_min, | |
391 | .extra2 = (int *) &nlm_port_max, | |
392 | }, | |
393 | { | |
394 | .ctl_name = CTL_UNNUMBERED, | |
395 | .procname = "nlm_tcpport", | |
396 | .data = &nlm_tcpport, | |
397 | .maxlen = sizeof(int), | |
398 | .mode = 0644, | |
399 | .proc_handler = &proc_dointvec_minmax, | |
400 | .extra1 = (int *) &nlm_port_min, | |
401 | .extra2 = (int *) &nlm_port_max, | |
402 | }, | |
403 | { .ctl_name = 0 } | |
404 | }; | |
405 | ||
406 | static ctl_table nlm_sysctl_dir[] = { | |
407 | { | |
408 | .ctl_name = CTL_UNNUMBERED, | |
409 | .procname = "nfs", | |
410 | .mode = 0555, | |
411 | .child = nlm_sysctls, | |
412 | }, | |
413 | { .ctl_name = 0 } | |
414 | }; | |
415 | ||
416 | static ctl_table nlm_sysctl_root[] = { | |
417 | { | |
418 | .ctl_name = CTL_FS, | |
419 | .procname = "fs", | |
420 | .mode = 0555, | |
421 | .child = nlm_sysctl_dir, | |
422 | }, | |
423 | { .ctl_name = 0 } | |
424 | }; | |
425 | ||
426 | /* | |
427 | * Module (and driverfs) parameters. | |
428 | */ | |
429 | ||
430 | #define param_set_min_max(name, type, which_strtol, min, max) \ | |
431 | static int param_set_##name(const char *val, struct kernel_param *kp) \ | |
432 | { \ | |
433 | char *endp; \ | |
434 | __typeof__(type) num = which_strtol(val, &endp, 0); \ | |
435 | if (endp == val || *endp || num < (min) || num > (max)) \ | |
436 | return -EINVAL; \ | |
437 | *((int *) kp->arg) = num; \ | |
438 | return 0; \ | |
439 | } | |
440 | ||
441 | static inline int is_callback(u32 proc) | |
442 | { | |
443 | return proc == NLMPROC_GRANTED | |
444 | || proc == NLMPROC_GRANTED_MSG | |
445 | || proc == NLMPROC_TEST_RES | |
446 | || proc == NLMPROC_LOCK_RES | |
447 | || proc == NLMPROC_CANCEL_RES | |
448 | || proc == NLMPROC_UNLOCK_RES | |
449 | || proc == NLMPROC_NSM_NOTIFY; | |
450 | } | |
451 | ||
452 | ||
453 | static int lockd_authenticate(struct svc_rqst *rqstp) | |
454 | { | |
455 | rqstp->rq_client = NULL; | |
456 | switch (rqstp->rq_authop->flavour) { | |
457 | case RPC_AUTH_NULL: | |
458 | case RPC_AUTH_UNIX: | |
459 | if (rqstp->rq_proc == 0) | |
460 | return SVC_OK; | |
461 | if (is_callback(rqstp->rq_proc)) { | |
462 | /* Leave it to individual procedures to | |
463 | * call nlmsvc_lookup_host(rqstp) | |
464 | */ | |
465 | return SVC_OK; | |
466 | } | |
467 | return svc_set_client(rqstp); | |
468 | } | |
469 | return SVC_DENIED; | |
470 | } | |
471 | ||
472 | ||
473 | param_set_min_max(port, int, simple_strtol, 0, 65535) | |
474 | param_set_min_max(grace_period, unsigned long, simple_strtoul, | |
475 | nlm_grace_period_min, nlm_grace_period_max) | |
476 | param_set_min_max(timeout, unsigned long, simple_strtoul, | |
477 | nlm_timeout_min, nlm_timeout_max) | |
478 | ||
479 | MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); | |
480 | MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION "."); | |
481 | MODULE_LICENSE("GPL"); | |
482 | ||
483 | module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong, | |
484 | &nlm_grace_period, 0644); | |
485 | module_param_call(nlm_timeout, param_set_timeout, param_get_ulong, | |
486 | &nlm_timeout, 0644); | |
487 | module_param_call(nlm_udpport, param_set_port, param_get_int, | |
488 | &nlm_udpport, 0644); | |
489 | module_param_call(nlm_tcpport, param_set_port, param_get_int, | |
490 | &nlm_tcpport, 0644); | |
491 | ||
492 | /* | |
493 | * Initialising and terminating the module. | |
494 | */ | |
495 | ||
496 | static int __init init_nlm(void) | |
497 | { | |
498 | nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root, 0); | |
499 | return nlm_sysctl_table ? 0 : -ENOMEM; | |
500 | } | |
501 | ||
502 | static void __exit exit_nlm(void) | |
503 | { | |
504 | /* FIXME: delete all NLM clients */ | |
505 | nlm_shutdown_hosts(); | |
506 | unregister_sysctl_table(nlm_sysctl_table); | |
507 | } | |
508 | ||
509 | module_init(init_nlm); | |
510 | module_exit(exit_nlm); | |
511 | ||
512 | /* | |
513 | * Define NLM program and procedures | |
514 | */ | |
515 | static struct svc_version nlmsvc_version1 = { | |
516 | .vs_vers = 1, | |
517 | .vs_nproc = 17, | |
518 | .vs_proc = nlmsvc_procedures, | |
519 | .vs_xdrsize = NLMSVC_XDRSIZE, | |
520 | }; | |
521 | static struct svc_version nlmsvc_version3 = { | |
522 | .vs_vers = 3, | |
523 | .vs_nproc = 24, | |
524 | .vs_proc = nlmsvc_procedures, | |
525 | .vs_xdrsize = NLMSVC_XDRSIZE, | |
526 | }; | |
527 | #ifdef CONFIG_LOCKD_V4 | |
528 | static struct svc_version nlmsvc_version4 = { | |
529 | .vs_vers = 4, | |
530 | .vs_nproc = 24, | |
531 | .vs_proc = nlmsvc_procedures4, | |
532 | .vs_xdrsize = NLMSVC_XDRSIZE, | |
533 | }; | |
534 | #endif | |
535 | static struct svc_version * nlmsvc_version[] = { | |
536 | [1] = &nlmsvc_version1, | |
537 | [3] = &nlmsvc_version3, | |
538 | #ifdef CONFIG_LOCKD_V4 | |
539 | [4] = &nlmsvc_version4, | |
540 | #endif | |
541 | }; | |
542 | ||
543 | static struct svc_stat nlmsvc_stats; | |
544 | ||
e8c96f8c | 545 | #define NLM_NRVERS ARRAY_SIZE(nlmsvc_version) |
1da177e4 LT |
546 | static struct svc_program nlmsvc_program = { |
547 | .pg_prog = NLM_PROGRAM, /* program number */ | |
548 | .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */ | |
549 | .pg_vers = nlmsvc_version, /* version table */ | |
550 | .pg_name = "lockd", /* service name */ | |
551 | .pg_class = "nfsd", /* share authentication with nfsd */ | |
552 | .pg_stats = &nlmsvc_stats, /* stats table */ | |
553 | .pg_authenticate = &lockd_authenticate /* export authentication */ | |
554 | }; |