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 | { | |
1da177e4 LT |
101 | int err = 0; |
102 | unsigned long grace_period_expire; | |
103 | ||
104 | /* Lock module and set up kernel thread */ | |
105 | /* lockd_up is waiting for us to startup, so will | |
106 | * be holding a reference to this module, so it | |
107 | * is safe to just claim another reference | |
108 | */ | |
109 | __module_get(THIS_MODULE); | |
110 | lock_kernel(); | |
111 | ||
112 | /* | |
113 | * Let our maker know we're running. | |
114 | */ | |
115 | nlmsvc_pid = current->pid; | |
6fb2b47f | 116 | nlmsvc_serv = rqstp->rq_server; |
353ab6e9 | 117 | complete(&lockd_start_done); |
1da177e4 LT |
118 | |
119 | daemonize("lockd"); | |
120 | ||
121 | /* Process request with signals blocked, but allow SIGKILL. */ | |
122 | allow_signal(SIGKILL); | |
123 | ||
124 | /* kick rpciod */ | |
125 | rpciod_up(); | |
126 | ||
127 | dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n"); | |
128 | ||
129 | if (!nlm_timeout) | |
130 | nlm_timeout = LOCKD_DFLT_TIMEO; | |
131 | nlmsvc_timeout = nlm_timeout * HZ; | |
132 | ||
133 | grace_period_expire = set_grace_period(); | |
134 | ||
135 | /* | |
136 | * The main request loop. We don't terminate until the last | |
137 | * NFS mount or NFS daemon has gone away, and we've been sent a | |
138 | * signal, or else another process has taken over our job. | |
139 | */ | |
140 | while ((nlmsvc_users || !signalled()) && nlmsvc_pid == current->pid) { | |
141 | long timeout = MAX_SCHEDULE_TIMEOUT; | |
142 | ||
143 | if (signalled()) { | |
144 | flush_signals(current); | |
145 | if (nlmsvc_ops) { | |
146 | nlmsvc_invalidate_all(); | |
147 | grace_period_expire = set_grace_period(); | |
148 | } | |
149 | } | |
150 | ||
151 | /* | |
152 | * Retry any blocked locks that have been notified by | |
153 | * the VFS. Don't do this during grace period. | |
154 | * (Theoretically, there shouldn't even be blocked locks | |
155 | * during grace period). | |
156 | */ | |
157 | if (!nlmsvc_grace_period) { | |
158 | timeout = nlmsvc_retry_blocked(); | |
159 | } else if (time_before(grace_period_expire, jiffies)) | |
160 | clear_grace_period(); | |
161 | ||
162 | /* | |
163 | * Find a socket with data available and call its | |
164 | * recvfrom routine. | |
165 | */ | |
6fb2b47f | 166 | err = svc_recv(rqstp, timeout); |
1da177e4 LT |
167 | if (err == -EAGAIN || err == -EINTR) |
168 | continue; | |
169 | if (err < 0) { | |
170 | printk(KERN_WARNING | |
171 | "lockd: terminating on error %d\n", | |
172 | -err); | |
173 | break; | |
174 | } | |
175 | ||
176 | dprintk("lockd: request from %08x\n", | |
177 | (unsigned)ntohl(rqstp->rq_addr.sin_addr.s_addr)); | |
178 | ||
6fb2b47f | 179 | svc_process(rqstp); |
1da177e4 LT |
180 | |
181 | } | |
182 | ||
2bd61579 TM |
183 | flush_signals(current); |
184 | ||
1da177e4 LT |
185 | /* |
186 | * Check whether there's a new lockd process before | |
187 | * shutting down the hosts and clearing the slot. | |
188 | */ | |
189 | if (!nlmsvc_pid || current->pid == nlmsvc_pid) { | |
190 | if (nlmsvc_ops) | |
191 | nlmsvc_invalidate_all(); | |
192 | nlm_shutdown_hosts(); | |
193 | nlmsvc_pid = 0; | |
24e36663 | 194 | nlmsvc_serv = NULL; |
1da177e4 LT |
195 | } else |
196 | printk(KERN_DEBUG | |
197 | "lockd: new process, skipping host shutdown\n"); | |
198 | wake_up(&lockd_exit); | |
46be925f | 199 | |
1da177e4 LT |
200 | /* Exit the RPC thread */ |
201 | svc_exit_thread(rqstp); | |
202 | ||
203 | /* release rpciod */ | |
204 | rpciod_down(); | |
205 | ||
206 | /* Release module */ | |
207 | unlock_kernel(); | |
208 | module_put_and_exit(0); | |
209 | } | |
210 | ||
24e36663 N |
211 | |
212 | static int find_socket(struct svc_serv *serv, int proto) | |
213 | { | |
214 | struct svc_sock *svsk; | |
215 | int found = 0; | |
216 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) | |
217 | if (svsk->sk_sk->sk_protocol == proto) { | |
218 | found = 1; | |
219 | break; | |
220 | } | |
221 | return found; | |
222 | } | |
223 | ||
224 | static int make_socks(struct svc_serv *serv, int proto) | |
225 | { | |
226 | /* Make any sockets that are needed but not present. | |
227 | * If nlm_udpport or nlm_tcpport were set as module | |
228 | * options, make those sockets unconditionally | |
229 | */ | |
230 | int err = 0; | |
231 | if (proto == IPPROTO_UDP || nlm_udpport) | |
232 | if (!find_socket(serv, IPPROTO_UDP)) | |
233 | err = svc_makesock(serv, IPPROTO_UDP, nlm_udpport); | |
234 | if (err) | |
235 | return err; | |
236 | if (proto == IPPROTO_TCP || nlm_tcpport) | |
237 | if (!find_socket(serv, IPPROTO_TCP)) | |
238 | err= svc_makesock(serv, IPPROTO_TCP, nlm_tcpport); | |
239 | return err; | |
240 | } | |
241 | ||
1da177e4 LT |
242 | /* |
243 | * Bring up the lockd process if it's not already up. | |
244 | */ | |
245 | int | |
24e36663 | 246 | lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */ |
1da177e4 LT |
247 | { |
248 | static int warned; | |
249 | struct svc_serv * serv; | |
250 | int error = 0; | |
251 | ||
353ab6e9 | 252 | mutex_lock(&nlmsvc_mutex); |
1da177e4 LT |
253 | /* |
254 | * Unconditionally increment the user count ... this is | |
255 | * the number of clients who _want_ a lockd process. | |
256 | */ | |
257 | nlmsvc_users++; | |
258 | /* | |
259 | * Check whether we're already up and running. | |
260 | */ | |
24e36663 N |
261 | if (nlmsvc_pid) { |
262 | error = make_socks(nlmsvc_serv, proto); | |
1da177e4 | 263 | goto out; |
24e36663 | 264 | } |
1da177e4 LT |
265 | |
266 | /* | |
267 | * Sanity check: if there's no pid, | |
268 | * we should be the first user ... | |
269 | */ | |
270 | if (nlmsvc_users > 1) | |
271 | printk(KERN_WARNING | |
272 | "lockd_up: no pid, %d users??\n", nlmsvc_users); | |
273 | ||
274 | error = -ENOMEM; | |
bc591ccf | 275 | serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL); |
1da177e4 LT |
276 | if (!serv) { |
277 | printk(KERN_WARNING "lockd_up: create service failed\n"); | |
278 | goto out; | |
279 | } | |
280 | ||
24e36663 | 281 | if ((error = make_socks(serv, proto)) < 0) { |
1da177e4 LT |
282 | if (warned++ == 0) |
283 | printk(KERN_WARNING | |
284 | "lockd_up: makesock failed, error=%d\n", error); | |
285 | goto destroy_and_out; | |
286 | } | |
287 | warned = 0; | |
288 | ||
289 | /* | |
290 | * Create the kernel thread and wait for it to start. | |
291 | */ | |
292 | error = svc_create_thread(lockd, serv); | |
293 | if (error) { | |
294 | printk(KERN_WARNING | |
295 | "lockd_up: create thread failed, error=%d\n", error); | |
296 | goto destroy_and_out; | |
297 | } | |
353ab6e9 | 298 | wait_for_completion(&lockd_start_done); |
1da177e4 LT |
299 | |
300 | /* | |
301 | * Note: svc_serv structures have an initial use count of 1, | |
302 | * so we exit through here on both success and failure. | |
303 | */ | |
304 | destroy_and_out: | |
305 | svc_destroy(serv); | |
306 | out: | |
353ab6e9 | 307 | mutex_unlock(&nlmsvc_mutex); |
1da177e4 LT |
308 | return error; |
309 | } | |
310 | EXPORT_SYMBOL(lockd_up); | |
311 | ||
312 | /* | |
313 | * Decrement the user count and bring down lockd if we're the last. | |
314 | */ | |
315 | void | |
316 | lockd_down(void) | |
317 | { | |
318 | static int warned; | |
319 | ||
353ab6e9 | 320 | mutex_lock(&nlmsvc_mutex); |
1da177e4 LT |
321 | if (nlmsvc_users) { |
322 | if (--nlmsvc_users) | |
323 | goto out; | |
324 | } else | |
325 | printk(KERN_WARNING "lockd_down: no users! pid=%d\n", nlmsvc_pid); | |
326 | ||
327 | if (!nlmsvc_pid) { | |
328 | if (warned++ == 0) | |
329 | printk(KERN_WARNING "lockd_down: no lockd running.\n"); | |
330 | goto out; | |
331 | } | |
332 | warned = 0; | |
333 | ||
334 | kill_proc(nlmsvc_pid, SIGKILL, 1); | |
335 | /* | |
336 | * Wait for the lockd process to exit, but since we're holding | |
337 | * the lockd semaphore, we can't wait around forever ... | |
338 | */ | |
339 | clear_thread_flag(TIF_SIGPENDING); | |
340 | interruptible_sleep_on_timeout(&lockd_exit, HZ); | |
341 | if (nlmsvc_pid) { | |
342 | printk(KERN_WARNING | |
343 | "lockd_down: lockd failed to exit, clearing pid\n"); | |
344 | nlmsvc_pid = 0; | |
345 | } | |
346 | spin_lock_irq(¤t->sighand->siglock); | |
347 | recalc_sigpending(); | |
348 | spin_unlock_irq(¤t->sighand->siglock); | |
349 | out: | |
353ab6e9 | 350 | mutex_unlock(&nlmsvc_mutex); |
1da177e4 LT |
351 | } |
352 | EXPORT_SYMBOL(lockd_down); | |
353 | ||
354 | /* | |
355 | * Sysctl parameters (same as module parameters, different interface). | |
356 | */ | |
357 | ||
358 | /* Something that isn't CTL_ANY, CTL_NONE or a value that may clash. */ | |
359 | #define CTL_UNNUMBERED -2 | |
360 | ||
361 | static ctl_table nlm_sysctls[] = { | |
362 | { | |
363 | .ctl_name = CTL_UNNUMBERED, | |
364 | .procname = "nlm_grace_period", | |
365 | .data = &nlm_grace_period, | |
7ee91ec1 | 366 | .maxlen = sizeof(unsigned long), |
1da177e4 LT |
367 | .mode = 0644, |
368 | .proc_handler = &proc_doulongvec_minmax, | |
369 | .extra1 = (unsigned long *) &nlm_grace_period_min, | |
370 | .extra2 = (unsigned long *) &nlm_grace_period_max, | |
371 | }, | |
372 | { | |
373 | .ctl_name = CTL_UNNUMBERED, | |
374 | .procname = "nlm_timeout", | |
375 | .data = &nlm_timeout, | |
7ee91ec1 | 376 | .maxlen = sizeof(unsigned long), |
1da177e4 LT |
377 | .mode = 0644, |
378 | .proc_handler = &proc_doulongvec_minmax, | |
379 | .extra1 = (unsigned long *) &nlm_timeout_min, | |
380 | .extra2 = (unsigned long *) &nlm_timeout_max, | |
381 | }, | |
382 | { | |
383 | .ctl_name = CTL_UNNUMBERED, | |
384 | .procname = "nlm_udpport", | |
385 | .data = &nlm_udpport, | |
386 | .maxlen = sizeof(int), | |
387 | .mode = 0644, | |
388 | .proc_handler = &proc_dointvec_minmax, | |
389 | .extra1 = (int *) &nlm_port_min, | |
390 | .extra2 = (int *) &nlm_port_max, | |
391 | }, | |
392 | { | |
393 | .ctl_name = CTL_UNNUMBERED, | |
394 | .procname = "nlm_tcpport", | |
395 | .data = &nlm_tcpport, | |
396 | .maxlen = sizeof(int), | |
397 | .mode = 0644, | |
398 | .proc_handler = &proc_dointvec_minmax, | |
399 | .extra1 = (int *) &nlm_port_min, | |
400 | .extra2 = (int *) &nlm_port_max, | |
401 | }, | |
402 | { .ctl_name = 0 } | |
403 | }; | |
404 | ||
405 | static ctl_table nlm_sysctl_dir[] = { | |
406 | { | |
407 | .ctl_name = CTL_UNNUMBERED, | |
408 | .procname = "nfs", | |
409 | .mode = 0555, | |
410 | .child = nlm_sysctls, | |
411 | }, | |
412 | { .ctl_name = 0 } | |
413 | }; | |
414 | ||
415 | static ctl_table nlm_sysctl_root[] = { | |
416 | { | |
417 | .ctl_name = CTL_FS, | |
418 | .procname = "fs", | |
419 | .mode = 0555, | |
420 | .child = nlm_sysctl_dir, | |
421 | }, | |
422 | { .ctl_name = 0 } | |
423 | }; | |
424 | ||
425 | /* | |
426 | * Module (and driverfs) parameters. | |
427 | */ | |
428 | ||
429 | #define param_set_min_max(name, type, which_strtol, min, max) \ | |
430 | static int param_set_##name(const char *val, struct kernel_param *kp) \ | |
431 | { \ | |
432 | char *endp; \ | |
433 | __typeof__(type) num = which_strtol(val, &endp, 0); \ | |
434 | if (endp == val || *endp || num < (min) || num > (max)) \ | |
435 | return -EINVAL; \ | |
436 | *((int *) kp->arg) = num; \ | |
437 | return 0; \ | |
438 | } | |
439 | ||
440 | static inline int is_callback(u32 proc) | |
441 | { | |
442 | return proc == NLMPROC_GRANTED | |
443 | || proc == NLMPROC_GRANTED_MSG | |
444 | || proc == NLMPROC_TEST_RES | |
445 | || proc == NLMPROC_LOCK_RES | |
446 | || proc == NLMPROC_CANCEL_RES | |
447 | || proc == NLMPROC_UNLOCK_RES | |
448 | || proc == NLMPROC_NSM_NOTIFY; | |
449 | } | |
450 | ||
451 | ||
452 | static int lockd_authenticate(struct svc_rqst *rqstp) | |
453 | { | |
454 | rqstp->rq_client = NULL; | |
455 | switch (rqstp->rq_authop->flavour) { | |
456 | case RPC_AUTH_NULL: | |
457 | case RPC_AUTH_UNIX: | |
458 | if (rqstp->rq_proc == 0) | |
459 | return SVC_OK; | |
460 | if (is_callback(rqstp->rq_proc)) { | |
461 | /* Leave it to individual procedures to | |
462 | * call nlmsvc_lookup_host(rqstp) | |
463 | */ | |
464 | return SVC_OK; | |
465 | } | |
466 | return svc_set_client(rqstp); | |
467 | } | |
468 | return SVC_DENIED; | |
469 | } | |
470 | ||
471 | ||
472 | param_set_min_max(port, int, simple_strtol, 0, 65535) | |
473 | param_set_min_max(grace_period, unsigned long, simple_strtoul, | |
474 | nlm_grace_period_min, nlm_grace_period_max) | |
475 | param_set_min_max(timeout, unsigned long, simple_strtoul, | |
476 | nlm_timeout_min, nlm_timeout_max) | |
477 | ||
478 | MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); | |
479 | MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION "."); | |
480 | MODULE_LICENSE("GPL"); | |
481 | ||
482 | module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong, | |
483 | &nlm_grace_period, 0644); | |
484 | module_param_call(nlm_timeout, param_set_timeout, param_get_ulong, | |
485 | &nlm_timeout, 0644); | |
486 | module_param_call(nlm_udpport, param_set_port, param_get_int, | |
487 | &nlm_udpport, 0644); | |
488 | module_param_call(nlm_tcpport, param_set_port, param_get_int, | |
489 | &nlm_tcpport, 0644); | |
490 | ||
491 | /* | |
492 | * Initialising and terminating the module. | |
493 | */ | |
494 | ||
495 | static int __init init_nlm(void) | |
496 | { | |
497 | nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root, 0); | |
498 | return nlm_sysctl_table ? 0 : -ENOMEM; | |
499 | } | |
500 | ||
501 | static void __exit exit_nlm(void) | |
502 | { | |
503 | /* FIXME: delete all NLM clients */ | |
504 | nlm_shutdown_hosts(); | |
505 | unregister_sysctl_table(nlm_sysctl_table); | |
506 | } | |
507 | ||
508 | module_init(init_nlm); | |
509 | module_exit(exit_nlm); | |
510 | ||
511 | /* | |
512 | * Define NLM program and procedures | |
513 | */ | |
514 | static struct svc_version nlmsvc_version1 = { | |
515 | .vs_vers = 1, | |
516 | .vs_nproc = 17, | |
517 | .vs_proc = nlmsvc_procedures, | |
518 | .vs_xdrsize = NLMSVC_XDRSIZE, | |
519 | }; | |
520 | static struct svc_version nlmsvc_version3 = { | |
521 | .vs_vers = 3, | |
522 | .vs_nproc = 24, | |
523 | .vs_proc = nlmsvc_procedures, | |
524 | .vs_xdrsize = NLMSVC_XDRSIZE, | |
525 | }; | |
526 | #ifdef CONFIG_LOCKD_V4 | |
527 | static struct svc_version nlmsvc_version4 = { | |
528 | .vs_vers = 4, | |
529 | .vs_nproc = 24, | |
530 | .vs_proc = nlmsvc_procedures4, | |
531 | .vs_xdrsize = NLMSVC_XDRSIZE, | |
532 | }; | |
533 | #endif | |
534 | static struct svc_version * nlmsvc_version[] = { | |
535 | [1] = &nlmsvc_version1, | |
536 | [3] = &nlmsvc_version3, | |
537 | #ifdef CONFIG_LOCKD_V4 | |
538 | [4] = &nlmsvc_version4, | |
539 | #endif | |
540 | }; | |
541 | ||
542 | static struct svc_stat nlmsvc_stats; | |
543 | ||
e8c96f8c | 544 | #define NLM_NRVERS ARRAY_SIZE(nlmsvc_version) |
1da177e4 LT |
545 | static struct svc_program nlmsvc_program = { |
546 | .pg_prog = NLM_PROGRAM, /* program number */ | |
547 | .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */ | |
548 | .pg_vers = nlmsvc_version, /* version table */ | |
549 | .pg_name = "lockd", /* service name */ | |
550 | .pg_class = "nfsd", /* share authentication with nfsd */ | |
551 | .pg_stats = &nlmsvc_stats, /* stats table */ | |
552 | .pg_authenticate = &lockd_authenticate /* export authentication */ | |
553 | }; |