xref: /netbsd-src/sys/nfs/nfs_syscalls.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: nfs_syscalls.c,v 1.127 2007/12/04 17:42:31 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_syscalls.c	8.5 (Berkeley) 3/30/95
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: nfs_syscalls.c,v 1.127 2007/12/04 17:42:31 yamt Exp $");
39 
40 #include "fs_nfs.h"
41 #include "opt_nfs.h"
42 #include "opt_nfsserver.h"
43 #include "opt_iso.h"
44 #include "opt_inet.h"
45 #include "opt_compat_netbsd.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/file.h>
51 #include <sys/stat.h>
52 #include <sys/vnode.h>
53 #include <sys/mount.h>
54 #include <sys/proc.h>
55 #include <sys/uio.h>
56 #include <sys/malloc.h>
57 #include <sys/kmem.h>
58 #include <sys/buf.h>
59 #include <sys/mbuf.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/signalvar.h>
63 #include <sys/domain.h>
64 #include <sys/protosw.h>
65 #include <sys/namei.h>
66 #include <sys/syslog.h>
67 #include <sys/filedesc.h>
68 #include <sys/kthread.h>
69 #include <sys/kauth.h>
70 #include <sys/syscallargs.h>
71 
72 #include <netinet/in.h>
73 #include <netinet/tcp.h>
74 #ifdef ISO
75 #include <netiso/iso.h>
76 #endif
77 #include <nfs/xdr_subs.h>
78 #include <nfs/rpcv2.h>
79 #include <nfs/nfsproto.h>
80 #include <nfs/nfs.h>
81 #include <nfs/nfsm_subs.h>
82 #include <nfs/nfsrvcache.h>
83 #include <nfs/nfsmount.h>
84 #include <nfs/nfsnode.h>
85 #include <nfs/nfsrtt.h>
86 #include <nfs/nfs_var.h>
87 
88 /* Global defs. */
89 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
90 						struct nfssvc_sock *,
91 						struct lwp *, struct mbuf **));
92 extern int nfsrvw_procrastinate;
93 
94 struct nfssvc_sock *nfs_udpsock;
95 #ifdef ISO
96 struct nfssvc_sock *nfs_cltpsock;
97 #endif
98 #ifdef INET6
99 struct nfssvc_sock *nfs_udp6sock;
100 #endif
101 int nuidhash_max = NFS_MAXUIDHASH;
102 #ifdef NFSSERVER
103 static int nfs_numnfsd = 0;
104 static struct nfsdrt nfsdrt;
105 #endif
106 
107 #ifdef NFSSERVER
108 kmutex_t nfsd_lock;
109 struct nfssvc_sockhead nfssvc_sockhead;
110 kcondvar_t nfsd_initcv;
111 struct nfssvc_sockhead nfssvc_sockpending;
112 struct nfsdhead nfsd_head;
113 struct nfsdidlehead nfsd_idle_head;
114 
115 int nfssvc_sockhead_flag;
116 int nfsd_head_flag;
117 #endif
118 
119 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
120 
121 #ifdef NFS
122 /*
123  * locking order:
124  *	nfs_iodlist_lock -> nid_lock -> nm_lock
125  */
126 kmutex_t nfs_iodlist_lock;
127 struct nfs_iodlist nfs_iodlist_idle;
128 struct nfs_iodlist nfs_iodlist_all;
129 int nfs_niothreads = -1; /* == "0, and has never been set" */
130 #endif
131 
132 #ifdef NFSSERVER
133 static struct nfssvc_sock *nfsrv_sockalloc __P((void));
134 static void nfsrv_sockfree __P((struct nfssvc_sock *));
135 static void nfsd_rt __P((int, struct nfsrv_descript *, int));
136 #endif
137 
138 /*
139  * NFS server system calls
140  */
141 
142 
143 /*
144  * Nfs server pseudo system call for the nfsd's
145  * Based on the flag value it either:
146  * - adds a socket to the selection list
147  * - remains in the kernel as an nfsd
148  * - remains in the kernel as an nfsiod
149  */
150 int
151 sys_nfssvc(struct lwp *l, void *v, register_t *retval)
152 {
153 	struct sys_nfssvc_args /* {
154 		syscallarg(int) flag;
155 		syscallarg(void *) argp;
156 	} */ *uap = v;
157 	int error;
158 #ifdef NFSSERVER
159 	struct file *fp;
160 	struct mbuf *nam;
161 	struct nfsd_args nfsdarg;
162 	struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
163 	struct nfsd *nfsd;
164 	struct nfssvc_sock *slp;
165 	struct nfsuid *nuidp;
166 #endif
167 
168 	/*
169 	 * Must be super user
170 	 */
171 	error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
172 	    NULL);
173 	if (error)
174 		return (error);
175 
176 	/* Initialize NFS server / client shared data. */
177 	nfs_init();
178 
179 #ifdef NFSSERVER
180 	mutex_enter(&nfsd_lock);
181 	while (nfssvc_sockhead_flag & SLP_INIT) {
182 		cv_wait(&nfsd_initcv, &nfsd_lock);
183 	}
184 	mutex_exit(&nfsd_lock);
185 #endif
186 	if (SCARG(uap, flag) & NFSSVC_BIOD) {
187 #if defined(NFS) && defined(COMPAT_14)
188 		error = kpause("nfsbiod", true, 0, NULL); /* dummy impl */
189 #else
190 		error = ENOSYS;
191 #endif
192 	} else if (SCARG(uap, flag) & NFSSVC_MNTD) {
193 		error = ENOSYS;
194 	} else if (SCARG(uap, flag) & NFSSVC_ADDSOCK) {
195 #ifndef NFSSERVER
196 		error = ENOSYS;
197 #else
198 		error = copyin(SCARG(uap, argp), (void *)&nfsdarg,
199 		    sizeof(nfsdarg));
200 		if (error)
201 			return (error);
202 		/* getsock() will use the descriptor for us */
203 		error = getsock(l->l_proc->p_fd, nfsdarg.sock, &fp);
204 		if (error)
205 			return (error);
206 		/*
207 		 * Get the client address for connected sockets.
208 		 */
209 		if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
210 			nam = (struct mbuf *)0;
211 		else {
212 			error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen,
213 				MT_SONAME);
214 			if (error) {
215 				FILE_UNUSE(fp, NULL);
216 				return (error);
217 			}
218 		}
219 		error = nfssvc_addsock(fp, nam);
220 		FILE_UNUSE(fp, NULL);
221 #endif /* !NFSSERVER */
222 	} else if (SCARG(uap, flag) & NFSSVC_SETEXPORTSLIST) {
223 #ifndef NFSSERVER
224 		error = ENOSYS;
225 #else
226 		struct export_args *args;
227 		struct mountd_exports_list mel;
228 
229 		error = copyin(SCARG(uap, argp), &mel, sizeof(mel));
230 		if (error != 0)
231 			return error;
232 
233 		args = (struct export_args *)malloc(mel.mel_nexports *
234 		    sizeof(struct export_args), M_TEMP, M_WAITOK);
235 		error = copyin(mel.mel_exports, args, mel.mel_nexports *
236 		    sizeof(struct export_args));
237 		if (error != 0) {
238 			free(args, M_TEMP);
239 			return error;
240 		}
241 		mel.mel_exports = args;
242 
243 		error = mountd_set_exports_list(&mel, l);
244 
245 		free(args, M_TEMP);
246 #endif /* !NFSSERVER */
247 	} else {
248 #ifndef NFSSERVER
249 		error = ENOSYS;
250 #else
251 		error = copyin(SCARG(uap, argp), (void *)nsd, sizeof (*nsd));
252 		if (error)
253 			return (error);
254 		if ((SCARG(uap, flag) & NFSSVC_AUTHIN) &&
255 		    ((nfsd = nsd->nsd_nfsd)) != NULL &&
256 		    (nfsd->nfsd_slp->ns_flags & SLP_VALID)) {
257 			slp = nfsd->nfsd_slp;
258 
259 			/*
260 			 * First check to see if another nfsd has already
261 			 * added this credential.
262 			 */
263 			LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid),
264 			    nu_hash) {
265 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
266 				    nsd->nsd_cr.cr_uid &&
267 				    (!nfsd->nfsd_nd->nd_nam2 ||
268 				     netaddr_match(NU_NETFAM(nuidp),
269 				     &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
270 					break;
271 			}
272 			if (nuidp) {
273 			    kauth_cred_hold(nuidp->nu_cr);
274 			    nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
275 			    nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
276 			} else {
277 			    /*
278 			     * Nope, so we will.
279 			     */
280 			    if (slp->ns_numuids < nuidhash_max) {
281 				slp->ns_numuids++;
282 				nuidp = (struct nfsuid *)
283 				   malloc(sizeof (struct nfsuid), M_NFSUID,
284 					M_WAITOK);
285 			    } else
286 				nuidp = (struct nfsuid *)0;
287 			    if ((slp->ns_flags & SLP_VALID) == 0) {
288 				if (nuidp)
289 				    free((void *)nuidp, M_NFSUID);
290 			    } else {
291 				if (nuidp == (struct nfsuid *)0) {
292 				    nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
293 				    LIST_REMOVE(nuidp, nu_hash);
294 				    TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
295 					nu_lru);
296 				    if (nuidp->nu_flag & NU_NAM)
297 					m_freem(nuidp->nu_nam);
298 			        }
299 				nuidp->nu_flag = 0;
300 				kauth_uucred_to_cred(nuidp->nu_cr,
301 				    &nsd->nsd_cr);
302 				nuidp->nu_timestamp = nsd->nsd_timestamp;
303 				nuidp->nu_expire = time_second + nsd->nsd_ttl;
304 				/*
305 				 * and save the session key in nu_key.
306 				 */
307 				memcpy(nuidp->nu_key, nsd->nsd_key,
308 				    sizeof(nsd->nsd_key));
309 				if (nfsd->nfsd_nd->nd_nam2) {
310 				    struct sockaddr_in *saddr;
311 
312 				    saddr = mtod(nfsd->nfsd_nd->nd_nam2,
313 					 struct sockaddr_in *);
314 				    switch (saddr->sin_family) {
315 				    case AF_INET:
316 					nuidp->nu_flag |= NU_INETADDR;
317 					nuidp->nu_inetaddr =
318 					     saddr->sin_addr.s_addr;
319 					break;
320 				    case AF_ISO:
321 				    default:
322 					nuidp->nu_flag |= NU_NAM;
323 					nuidp->nu_nam = m_copym(
324 					    nfsd->nfsd_nd->nd_nam2, 0,
325 					     M_COPYALL, M_WAIT);
326 					break;
327 				    };
328 				}
329 				TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
330 					nu_lru);
331 				LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
332 					nuidp, nu_hash);
333 				kauth_cred_hold(nuidp->nu_cr);
334 				nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
335 				nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
336 			    }
337 			}
338 		}
339 		if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) &&
340 		    (nfsd = nsd->nsd_nfsd))
341 			nfsd->nfsd_flag |= NFSD_AUTHFAIL;
342 		error = nfssvc_nfsd(nsd, SCARG(uap, argp), l);
343 #endif /* !NFSSERVER */
344 	}
345 	if (error == EINTR || error == ERESTART)
346 		error = 0;
347 	return (error);
348 }
349 
350 #ifdef NFSSERVER
351 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
352 MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
353 
354 static struct nfssvc_sock *
355 nfsrv_sockalloc()
356 {
357 	struct nfssvc_sock *slp;
358 
359 	slp = (struct nfssvc_sock *)
360 	    malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK);
361 	memset(slp, 0, sizeof (struct nfssvc_sock));
362 	/* XXX could be IPL_SOFTNET */
363 	mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_VM);
364 	mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_VM);
365 	cv_init(&slp->ns_cv, "nfsdsock");
366 	TAILQ_INIT(&slp->ns_uidlruhead);
367 	LIST_INIT(&slp->ns_tq);
368 	SIMPLEQ_INIT(&slp->ns_sendq);
369 	mutex_enter(&nfsd_lock);
370 	TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
371 	mutex_exit(&nfsd_lock);
372 
373 	return slp;
374 }
375 
376 static void
377 nfsrv_sockfree(struct nfssvc_sock *slp)
378 {
379 
380 	KASSERT(slp->ns_so == NULL);
381 	KASSERT(slp->ns_fp == NULL);
382 	KASSERT((slp->ns_flags & SLP_VALID) == 0);
383 	mutex_destroy(&slp->ns_lock);
384 	mutex_destroy(&slp->ns_alock);
385 	cv_destroy(&slp->ns_cv);
386 	free(slp, M_NFSSVC);
387 }
388 
389 /*
390  * Adds a socket to the list for servicing by nfsds.
391  */
392 int
393 nfssvc_addsock(fp, mynam)
394 	struct file *fp;
395 	struct mbuf *mynam;
396 {
397 	struct mbuf *m;
398 	int siz;
399 	struct nfssvc_sock *slp;
400 	struct socket *so;
401 	struct nfssvc_sock *tslp;
402 	int error, s;
403 
404 	so = (struct socket *)fp->f_data;
405 	tslp = (struct nfssvc_sock *)0;
406 	/*
407 	 * Add it to the list, as required.
408 	 */
409 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
410 #ifdef INET6
411 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
412 			tslp = nfs_udp6sock;
413 		else
414 #endif
415 		tslp = nfs_udpsock;
416 		if (tslp->ns_flags & SLP_VALID) {
417 			m_freem(mynam);
418 			return (EPERM);
419 		}
420 #ifdef ISO
421 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
422 		tslp = nfs_cltpsock;
423 		if (tslp->ns_flags & SLP_VALID) {
424 			m_freem(mynam);
425 			return (EPERM);
426 		}
427 #endif /* ISO */
428 	}
429 	if (so->so_type == SOCK_STREAM)
430 		siz = NFS_MAXPACKET + sizeof (u_long);
431 	else
432 		siz = NFS_MAXPACKET;
433 	error = soreserve(so, siz, siz);
434 	if (error) {
435 		m_freem(mynam);
436 		return (error);
437 	}
438 
439 	/*
440 	 * Set protocol specific options { for now TCP only } and
441 	 * reserve some space. For datagram sockets, this can get called
442 	 * repeatedly for the same socket, but that isn't harmful.
443 	 */
444 	if (so->so_type == SOCK_STREAM) {
445 		m = m_get(M_WAIT, MT_SOOPTS);
446 		MCLAIM(m, &nfs_mowner);
447 		*mtod(m, int32_t *) = 1;
448 		m->m_len = sizeof(int32_t);
449 		sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
450 	}
451 	if ((so->so_proto->pr_domain->dom_family == AF_INET
452 #ifdef INET6
453 	    || so->so_proto->pr_domain->dom_family == AF_INET6
454 #endif
455 	    ) &&
456 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
457 		m = m_get(M_WAIT, MT_SOOPTS);
458 		MCLAIM(m, &nfs_mowner);
459 		*mtod(m, int32_t *) = 1;
460 		m->m_len = sizeof(int32_t);
461 		sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
462 	}
463 	so->so_rcv.sb_flags &= ~SB_NOINTR;
464 	so->so_rcv.sb_timeo = 0;
465 	so->so_snd.sb_flags &= ~SB_NOINTR;
466 	so->so_snd.sb_timeo = 0;
467 	if (tslp) {
468 		slp = tslp;
469 	} else {
470 		slp = nfsrv_sockalloc();
471 	}
472 	slp->ns_so = so;
473 	slp->ns_nam = mynam;
474 	mutex_enter(&fp->f_lock);
475 	fp->f_count++;
476 	mutex_exit(&fp->f_lock);
477 	slp->ns_fp = fp;
478 	slp->ns_flags = SLP_VALID;
479 	slp->ns_aflags = SLP_A_NEEDQ;
480 	slp->ns_gflags = 0;
481 	slp->ns_sflags = 0;
482 	s = splsoftnet();
483 	so->so_upcallarg = (void *)slp;
484 	so->so_upcall = nfsrv_soupcall;
485 	so->so_rcv.sb_flags |= SB_UPCALL;
486 	splx(s);
487 	nfsrv_wakenfsd(slp);
488 	return (0);
489 }
490 
491 /*
492  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
493  * until it is killed by a signal.
494  */
495 int
496 nfssvc_nfsd(nsd, argp, l)
497 	struct nfsd_srvargs *nsd;
498 	void *argp;
499 	struct lwp *l;
500 {
501 	struct timeval tv;
502 	struct mbuf *m;
503 	struct nfssvc_sock *slp;
504 	struct nfsd *nfsd = nsd->nsd_nfsd;
505 	struct nfsrv_descript *nd = NULL;
506 	struct mbuf *mreq;
507 	u_quad_t cur_usec;
508 	int error = 0, cacherep, siz, sotype, writes_todo;
509 	struct proc *p = l->l_proc;
510 	int s;
511 	bool doreinit;
512 
513 #ifndef nolint
514 	cacherep = RC_DOIT;
515 	writes_todo = 0;
516 #endif
517 	uvm_lwp_hold(l);
518 	if (nfsd == NULL) {
519 		nsd->nsd_nfsd = nfsd =
520 			malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK);
521 		memset(nfsd, 0, sizeof (struct nfsd));
522 		cv_init(&nfsd->nfsd_cv, "nfsd");
523 		nfsd->nfsd_procp = p;
524 		mutex_enter(&nfsd_lock);
525 		while ((nfssvc_sockhead_flag & SLP_INIT) != 0) {
526 			KASSERT(nfs_numnfsd == 0);
527 			cv_wait(&nfsd_initcv, &nfsd_lock);
528 		}
529 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
530 		nfs_numnfsd++;
531 		mutex_exit(&nfsd_lock);
532 	}
533 	/*
534 	 * Loop getting rpc requests until SIGKILL.
535 	 */
536 	for (;;) {
537 		bool dummy;
538 
539 		if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
540 		    != 0) {
541 			preempt();
542 		}
543 		if (nfsd->nfsd_slp == NULL) {
544 			mutex_enter(&nfsd_lock);
545 			while (nfsd->nfsd_slp == NULL &&
546 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
547 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
548 				    nfsd_idle);
549 				error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock);
550 				if (error) {
551 					slp = nfsd->nfsd_slp;
552 					nfsd->nfsd_slp = NULL;
553 					if (!slp)
554 						SLIST_REMOVE(&nfsd_idle_head,
555 						    nfsd, nfsd, nfsd_idle);
556 					mutex_exit(&nfsd_lock);
557 					if (slp) {
558 						nfsrv_wakenfsd(slp);
559 						nfsrv_slpderef(slp);
560 					}
561 					goto done;
562 				}
563 			}
564 			if (nfsd->nfsd_slp == NULL &&
565 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
566 				slp = TAILQ_FIRST(&nfssvc_sockpending);
567 				if (slp) {
568 					KASSERT((slp->ns_gflags & SLP_G_DOREC)
569 					    != 0);
570 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
571 					    ns_pending);
572 					slp->ns_gflags &= ~SLP_G_DOREC;
573 					slp->ns_sref++;
574 					nfsd->nfsd_slp = slp;
575 				} else
576 					nfsd_head_flag &= ~NFSD_CHECKSLP;
577 			}
578 			KASSERT(nfsd->nfsd_slp == NULL ||
579 			    nfsd->nfsd_slp->ns_sref > 0);
580 			mutex_exit(&nfsd_lock);
581 			if ((slp = nfsd->nfsd_slp) == NULL)
582 				continue;
583 			if (slp->ns_flags & SLP_VALID) {
584 				bool more;
585 
586 				if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) {
587 					nfsrv_rcv(slp);
588 				}
589 				if (nfsdsock_testbits(slp, SLP_A_DISCONN)) {
590 					nfsrv_zapsock(slp);
591 				}
592 				error = nfsrv_dorec(slp, nfsd, &nd, &more);
593 				getmicrotime(&tv);
594 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
595 					(u_quad_t)tv.tv_usec;
596 				writes_todo = 0;
597 				if (error) {
598 					struct nfsrv_descript *nd2;
599 
600 					mutex_enter(&nfsd_lock);
601 					nd2 = LIST_FIRST(&slp->ns_tq);
602 					if (nd2 != NULL &&
603 					    nd2->nd_time <= cur_usec) {
604 						error = 0;
605 						cacherep = RC_DOIT;
606 						writes_todo = 1;
607 					}
608 					mutex_exit(&nfsd_lock);
609 				}
610 				if (error == 0 && more) {
611 					nfsrv_wakenfsd(slp);
612 				}
613 			}
614 		} else {
615 			error = 0;
616 			slp = nfsd->nfsd_slp;
617 		}
618 		KASSERT(slp != NULL);
619 		KASSERT(nfsd->nfsd_slp == slp);
620 		if (error || (slp->ns_flags & SLP_VALID) == 0) {
621 			if (nd) {
622 				nfsdreq_free(nd);
623 				nd = NULL;
624 			}
625 			nfsd->nfsd_slp = NULL;
626 			nfsrv_slpderef(slp);
627 			continue;
628 		}
629 		sotype = slp->ns_so->so_type;
630 		if (nd) {
631 			getmicrotime(&nd->nd_starttime);
632 			if (nd->nd_nam2)
633 				nd->nd_nam = nd->nd_nam2;
634 			else
635 				nd->nd_nam = slp->ns_nam;
636 
637 			/*
638 			 * Check to see if authorization is needed.
639 			 */
640 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
641 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
642 				nsd->nsd_haddr = mtod(nd->nd_nam,
643 				    struct sockaddr_in *)->sin_addr.s_addr;
644 				nsd->nsd_authlen = nfsd->nfsd_authlen;
645 				nsd->nsd_verflen = nfsd->nfsd_verflen;
646 				if (!copyout(nfsd->nfsd_authstr,
647 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
648 				    !copyout(nfsd->nfsd_verfstr,
649 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
650 				    !copyout(nsd, argp, sizeof (*nsd))) {
651 					uvm_lwp_rele(l);
652 					return (ENEEDAUTH);
653 				}
654 				cacherep = RC_DROPIT;
655 			} else
656 				cacherep = nfsrv_getcache(nd, slp, &mreq);
657 
658 			if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
659 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
660 				nd->nd_procnum = NFSPROC_NOOP;
661 				nd->nd_repstat =
662 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
663 				cacherep = RC_DOIT;
664 			}
665 		}
666 
667 		/*
668 		 * Loop to get all the write rpc relies that have been
669 		 * gathered together.
670 		 */
671 		do {
672 #ifdef DIAGNOSTIC
673 			int lockcount;
674 #endif
675 			switch (cacherep) {
676 			case RC_DOIT:
677 #ifdef DIAGNOSTIC
678 				/*
679 				 * NFS server procs should neither release
680 				 * locks already held, nor leave things
681 				 * locked.  Catch this sooner, rather than
682 				 * later (when we try to relock something we
683 				 * already have locked).  Careful inspection
684 				 * of the failing routine usually turns up the
685 				 * lock leak.. once we know what it is..
686 				 */
687 				lockcount = l->l_locks;
688 #endif
689 				mreq = NULL;
690 				netexport_rdlock();
691 				if (writes_todo || nd == NULL ||
692 				     (!(nd->nd_flag & ND_NFSV3) &&
693 				     nd->nd_procnum == NFSPROC_WRITE &&
694 				     nfsrvw_procrastinate > 0))
695 					error = nfsrv_writegather(&nd, slp,
696 					    l, &mreq);
697 				else
698 					error =
699 					    (*(nfsrv3_procs[nd->nd_procnum]))
700 					    (nd, slp, l, &mreq);
701 				netexport_rdunlock();
702 #ifdef DIAGNOSTIC
703 				if (l->l_locks != lockcount) {
704 					/*
705 					 * If you see this panic, audit
706 					 * nfsrv3_procs[nd->nd_procnum] for
707 					 * vnode locking errors (usually, it's
708 					 * due to forgetting to vput()
709 					 * something).
710 					 */
711 #ifdef DEBUG
712 					extern void printlockedvnodes(void);
713 					printlockedvnodes();
714 #endif
715 					printf("nfsd: locking botch in op %d"
716 					    " (before %d, after %d)\n",
717 					    nd ? nd->nd_procnum : -1,
718 					    lockcount, l->l_locks);
719 				}
720 #endif
721 				if (mreq == NULL) {
722 					if (nd != NULL) {
723 						if (nd->nd_nam2)
724 							m_free(nd->nd_nam2);
725 						if (nd->nd_mrep)
726 							m_freem(nd->nd_mrep);
727 					}
728 					break;
729 				}
730 				if (error) {
731 					nfsstats.srv_errs++;
732 					nfsrv_updatecache(nd, false, mreq);
733 					if (nd->nd_nam2)
734 						m_freem(nd->nd_nam2);
735 					break;
736 				}
737 				nfsstats.srvrpccnt[nd->nd_procnum]++;
738 				nfsrv_updatecache(nd, true, mreq);
739 				nd->nd_mrep = (struct mbuf *)0;
740 			case RC_REPLY:
741 				m = mreq;
742 				siz = 0;
743 				while (m) {
744 					siz += m->m_len;
745 					m = m->m_next;
746 				}
747 				if (siz <= 0 || siz > NFS_MAXPACKET) {
748 					printf("mbuf siz=%d\n",siz);
749 					panic("Bad nfs svc reply");
750 				}
751 				m = mreq;
752 				m->m_pkthdr.len = siz;
753 				m->m_pkthdr.rcvif = (struct ifnet *)0;
754 				/*
755 				 * For stream protocols, prepend a Sun RPC
756 				 * Record Mark.
757 				 */
758 				if (sotype == SOCK_STREAM) {
759 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
760 					*mtod(m, u_int32_t *) =
761 					    htonl(0x80000000 | siz);
762 				}
763 				nd->nd_mreq = m;
764 				if (nfsrtton) {
765 					nfsd_rt(slp->ns_so->so_type, nd,
766 					    cacherep);
767 				}
768 				error = nfsdsock_sendreply(slp, nd);
769 				nd = NULL;
770 				if (error == EPIPE)
771 					nfsrv_zapsock(slp);
772 				if (error == EINTR || error == ERESTART) {
773 					nfsd->nfsd_slp = NULL;
774 					nfsrv_slpderef(slp);
775 					goto done;
776 				}
777 				break;
778 			case RC_DROPIT:
779 				if (nfsrtton)
780 					nfsd_rt(sotype, nd, cacherep);
781 				m_freem(nd->nd_mrep);
782 				m_freem(nd->nd_nam2);
783 				break;
784 			}
785 			if (nd) {
786 				nfsdreq_free(nd);
787 				nd = NULL;
788 			}
789 
790 			/*
791 			 * Check to see if there are outstanding writes that
792 			 * need to be serviced.
793 			 */
794 			getmicrotime(&tv);
795 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
796 			    (u_quad_t)tv.tv_usec;
797 			s = splsoftclock();
798 			if (LIST_FIRST(&slp->ns_tq) &&
799 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
800 				cacherep = RC_DOIT;
801 				writes_todo = 1;
802 			} else
803 				writes_todo = 0;
804 			splx(s);
805 		} while (writes_todo);
806 		if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) {
807 			nfsd->nfsd_slp = NULL;
808 			nfsrv_slpderef(slp);
809 		}
810 	}
811 done:
812 	mutex_enter(&nfsd_lock);
813 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
814 	doreinit = --nfs_numnfsd == 0;
815 	if (doreinit)
816 		nfssvc_sockhead_flag |= SLP_INIT;
817 	mutex_exit(&nfsd_lock);
818 	cv_destroy(&nfsd->nfsd_cv);
819 	free(nfsd, M_NFSD);
820 	nsd->nsd_nfsd = NULL;
821 	if (doreinit)
822 		nfsrv_init(true);	/* Reinitialize everything */
823 	uvm_lwp_rele(l);
824 	return (error);
825 }
826 
827 /*
828  * Shut down a socket associated with an nfssvc_sock structure.
829  * Should be called with the send lock set, if required.
830  * The trick here is to increment the sref at the start, so that the nfsds
831  * will stop using it and clear ns_flag at the end so that it will not be
832  * reassigned during cleanup.
833  *
834  * called at splsoftnet.
835  */
836 void
837 nfsrv_zapsock(slp)
838 	struct nfssvc_sock *slp;
839 {
840 	struct nfsuid *nuidp, *nnuidp;
841 	struct nfsrv_descript *nwp;
842 	struct socket *so;
843 	struct mbuf *m;
844 	int s;
845 
846 	if (nfsdsock_drain(slp)) {
847 		return;
848 	}
849 	mutex_enter(&nfsd_lock);
850 	if (slp->ns_gflags & SLP_G_DOREC) {
851 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
852 		slp->ns_gflags &= ~SLP_G_DOREC;
853 	}
854 	mutex_exit(&nfsd_lock);
855 
856 	so = slp->ns_so;
857 	KASSERT(so != NULL);
858 	s = splsoftnet();
859 	so->so_upcall = NULL;
860 	so->so_upcallarg = NULL;
861 	so->so_rcv.sb_flags &= ~SB_UPCALL;
862 	splx(s);
863 	soshutdown(so, SHUT_RDWR);
864 
865 	if (slp->ns_nam)
866 		m_free(slp->ns_nam);
867 	m_freem(slp->ns_raw);
868 	m = slp->ns_rec;
869 	while (m != NULL) {
870 		struct mbuf *n;
871 
872 		n = m->m_nextpkt;
873 		m_freem(m);
874 		m = n;
875 	}
876 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
877 	    nuidp = nnuidp) {
878 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
879 		LIST_REMOVE(nuidp, nu_hash);
880 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
881 		if (nuidp->nu_flag & NU_NAM)
882 			m_freem(nuidp->nu_nam);
883 		free((void *)nuidp, M_NFSUID);
884 	}
885 	mutex_enter(&nfsd_lock);
886 	while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) {
887 		LIST_REMOVE(nwp, nd_tq);
888 		mutex_exit(&nfsd_lock);
889 		nfsdreq_free(nwp);
890 		mutex_enter(&nfsd_lock);
891 	}
892 	mutex_exit(&nfsd_lock);
893 }
894 
895 /*
896  * Derefence a server socket structure. If it has no more references and
897  * is no longer valid, you can throw it away.
898  */
899 void
900 nfsrv_slpderef(slp)
901 	struct nfssvc_sock *slp;
902 {
903 	uint32_t ref;
904 
905 	mutex_enter(&nfsd_lock);
906 	KASSERT(slp->ns_sref > 0);
907 	ref = --slp->ns_sref;
908 	mutex_exit(&nfsd_lock);
909 	if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) {
910 		struct file *fp;
911 
912 		mutex_enter(&nfsd_lock);
913 		KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0);
914 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
915 		mutex_exit(&nfsd_lock);
916 
917 		fp = slp->ns_fp;
918 		if (fp != NULL) {
919 			slp->ns_fp = NULL;
920 			KASSERT(fp != NULL);
921 			KASSERT(fp->f_data == slp->ns_so);
922 			mutex_enter(&fp->f_lock);
923 			FILE_USE(fp);
924 			closef(fp, (struct lwp *)0);
925 			slp->ns_so = NULL;
926 		}
927 
928 		nfsrv_sockfree(slp);
929 	}
930 }
931 
932 /*
933  * Initialize the data structures for the server.
934  * Handshake with any new nfsds starting up to avoid any chance of
935  * corruption.
936  */
937 void
938 nfsrv_init(terminating)
939 	int terminating;
940 {
941 	struct nfssvc_sock *slp;
942 
943 	if (!terminating) {
944 		/* XXX could be IPL_SOFTNET */
945 		mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_VM);
946 		cv_init(&nfsd_initcv, "nfsdinit");
947 	}
948 
949 	mutex_enter(&nfsd_lock);
950 	if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0)
951 		panic("nfsd init");
952 	nfssvc_sockhead_flag |= SLP_INIT;
953 
954 	if (terminating) {
955 		KASSERT(SLIST_EMPTY(&nfsd_idle_head));
956 		KASSERT(TAILQ_EMPTY(&nfsd_head));
957 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
958 			mutex_exit(&nfsd_lock);
959 			KASSERT(slp->ns_sref == 0);
960 			slp->ns_sref++;
961 			nfsrv_zapsock(slp);
962 			nfsrv_slpderef(slp);
963 			mutex_enter(&nfsd_lock);
964 		}
965 		KASSERT(TAILQ_EMPTY(&nfssvc_sockpending));
966 		mutex_exit(&nfsd_lock);
967 		nfsrv_cleancache();	/* And clear out server cache */
968 	} else {
969 		mutex_exit(&nfsd_lock);
970 		nfs_pub.np_valid = 0;
971 	}
972 
973 	TAILQ_INIT(&nfssvc_sockhead);
974 	TAILQ_INIT(&nfssvc_sockpending);
975 
976 	TAILQ_INIT(&nfsd_head);
977 	SLIST_INIT(&nfsd_idle_head);
978 	nfsd_head_flag &= ~NFSD_CHECKSLP;
979 
980 	nfs_udpsock = nfsrv_sockalloc();
981 
982 #ifdef INET6
983 	nfs_udp6sock = nfsrv_sockalloc();
984 #endif
985 
986 #ifdef ISO
987 	nfs_cltpsock = nfsrv_sockalloc();
988 #endif
989 
990 	mutex_enter(&nfsd_lock);
991 	nfssvc_sockhead_flag &= ~SLP_INIT;
992 	cv_broadcast(&nfsd_initcv);
993 	mutex_exit(&nfsd_lock);
994 }
995 
996 /*
997  * Add entries to the server monitor log.
998  */
999 static void
1000 nfsd_rt(sotype, nd, cacherep)
1001 	int sotype;
1002 	struct nfsrv_descript *nd;
1003 	int cacherep;
1004 {
1005 	struct timeval tv;
1006 	struct drt *rt;
1007 
1008 	rt = &nfsdrt.drt[nfsdrt.pos];
1009 	if (cacherep == RC_DOIT)
1010 		rt->flag = 0;
1011 	else if (cacherep == RC_REPLY)
1012 		rt->flag = DRT_CACHEREPLY;
1013 	else
1014 		rt->flag = DRT_CACHEDROP;
1015 	if (sotype == SOCK_STREAM)
1016 		rt->flag |= DRT_TCP;
1017 	if (nd->nd_flag & ND_NFSV3)
1018 		rt->flag |= DRT_NFSV3;
1019 	rt->proc = nd->nd_procnum;
1020 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
1021 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
1022 	else
1023 	    rt->ipadr = INADDR_ANY;
1024 	getmicrotime(&tv);
1025 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
1026 		(tv.tv_usec - nd->nd_starttime.tv_usec);
1027 	rt->tstamp = tv;
1028 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
1029 }
1030 #endif /* NFSSERVER */
1031 
1032 #ifdef NFS
1033 
1034 int nfs_defect = 0;
1035 /*
1036  * Asynchronous I/O threads for client nfs.
1037  * They do read-ahead and write-behind operations on the block I/O cache.
1038  * Never returns unless it fails or gets killed.
1039  */
1040 
1041 static void
1042 nfssvc_iod(void *arg)
1043 {
1044 	struct buf *bp;
1045 	struct nfs_iod *myiod;
1046 	struct nfsmount *nmp;
1047 
1048 	KERNEL_LOCK(1, curlwp);
1049 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
1050 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
1051 	KERNEL_UNLOCK_LAST(curlwp);
1052 	cv_init(&myiod->nid_cv, "nfsiod");
1053 	myiod->nid_exiting = false;
1054 	myiod->nid_mount = NULL;
1055 	mutex_enter(&nfs_iodlist_lock);
1056 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
1057 	mutex_exit(&nfs_iodlist_lock);
1058 
1059 	for (;;) {
1060 		mutex_enter(&nfs_iodlist_lock);
1061 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
1062 		mutex_exit(&nfs_iodlist_lock);
1063 
1064 		mutex_enter(&myiod->nid_lock);
1065 		while (/*CONSTCOND*/ true) {
1066 			nmp = myiod->nid_mount;
1067 			if (nmp) {
1068 				myiod->nid_mount = NULL;
1069 				break;
1070 			}
1071 			if (__predict_false(myiod->nid_exiting)) {
1072 				/*
1073 				 * drop nid_lock to preserve locking order.
1074 				 */
1075 				mutex_exit(&myiod->nid_lock);
1076 				mutex_enter(&nfs_iodlist_lock);
1077 				mutex_enter(&myiod->nid_lock);
1078 				/*
1079 				 * recheck nid_mount because nfs_asyncio can
1080 				 * pick us in the meantime as we are still on
1081 				 * nfs_iodlist_lock.
1082 				 */
1083 				if (myiod->nid_mount != NULL) {
1084 					mutex_exit(&nfs_iodlist_lock);
1085 					continue;
1086 				}
1087 				LIST_REMOVE(myiod, nid_idle);
1088 				mutex_exit(&nfs_iodlist_lock);
1089 				goto quit;
1090 			}
1091 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
1092 		}
1093 		mutex_exit(&myiod->nid_lock);
1094 
1095 		mutex_enter(&nmp->nm_lock);
1096 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
1097 			/* Take one off the front of the list */
1098 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
1099 			nmp->nm_bufqlen--;
1100 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
1101 				cv_broadcast(&nmp->nm_aiocv);
1102 			}
1103 			mutex_exit(&nmp->nm_lock);
1104 			KERNEL_LOCK(1, curlwp);
1105 			(void)nfs_doio(bp);
1106 			KERNEL_UNLOCK_LAST(curlwp);
1107 			mutex_enter(&nmp->nm_lock);
1108 			/*
1109 			 * If there are more than one iod on this mount,
1110 			 * then defect so that the iods can be shared out
1111 			 * fairly between the mounts
1112 			 */
1113 			if (nfs_defect && nmp->nm_bufqiods > 1) {
1114 				break;
1115 			}
1116 		}
1117 		KASSERT(nmp->nm_bufqiods > 0);
1118 		nmp->nm_bufqiods--;
1119 		mutex_exit(&nmp->nm_lock);
1120 	}
1121 quit:
1122 	KASSERT(myiod->nid_mount == NULL);
1123 	mutex_exit(&myiod->nid_lock);
1124 
1125 	cv_destroy(&myiod->nid_cv);
1126 	KERNEL_LOCK(1, curlwp);
1127 	mutex_destroy(&myiod->nid_lock);
1128 	kmem_free(myiod, sizeof(*myiod));
1129 	KERNEL_UNLOCK_LAST(curlwp);
1130 
1131 	kthread_exit(0);
1132 }
1133 
1134 void
1135 nfs_iodinit()
1136 {
1137 
1138 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
1139 	LIST_INIT(&nfs_iodlist_all);
1140 	LIST_INIT(&nfs_iodlist_idle);
1141 }
1142 
1143 int
1144 nfs_set_niothreads(int newval)
1145 {
1146 	struct nfs_iod *nid;
1147 	int error = 0;
1148 
1149 #if defined(MULTIPROCESSOR)
1150         int hold_count;
1151 #endif /* defined(MULTIPROCESSOR) */
1152 
1153 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
1154 
1155 	mutex_enter(&nfs_iodlist_lock);
1156 	/* clamp to sane range */
1157 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
1158 
1159 	while (nfs_numasync != nfs_niothreads && error == 0) {
1160 		while (nfs_numasync < nfs_niothreads) {
1161 
1162 			/*
1163 			 * kthread_create can wait for pagedaemon and
1164 			 * pagedaemon can wait for nfsiod which needs to acquire
1165 			 * nfs_iodlist_lock.
1166 			 */
1167 
1168 			mutex_exit(&nfs_iodlist_lock);
1169 			KERNEL_LOCK(1, curlwp);
1170 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
1171 			    nfssvc_iod, NULL, NULL, "nfsio");
1172 			KERNEL_UNLOCK_LAST(curlwp);
1173 			mutex_enter(&nfs_iodlist_lock);
1174 			if (error) {
1175 				/* give up */
1176 				nfs_niothreads = nfs_numasync;
1177 				break;
1178 			}
1179 			nfs_numasync++;
1180 		}
1181 		while (nfs_numasync > nfs_niothreads) {
1182 			nid = LIST_FIRST(&nfs_iodlist_all);
1183 			if (nid == NULL) {
1184 				/* iod has not started yet. */
1185 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
1186 				continue;
1187 			}
1188 			LIST_REMOVE(nid, nid_all);
1189 			mutex_enter(&nid->nid_lock);
1190 			KASSERT(!nid->nid_exiting);
1191 			nid->nid_exiting = true;
1192 			cv_signal(&nid->nid_cv);
1193 			mutex_exit(&nid->nid_lock);
1194 			nfs_numasync--;
1195 		}
1196 	}
1197 	mutex_exit(&nfs_iodlist_lock);
1198 
1199 	KERNEL_LOCK(hold_count, curlwp);
1200 	return error;
1201 }
1202 
1203 /*
1204  * Get an authorization string for the uid by having the mount_nfs sitting
1205  * on this mount point porpous out of the kernel and do it.
1206  */
1207 int
1208 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
1209 	struct nfsmount *nmp;
1210 	struct nfsreq *rep;
1211 	kauth_cred_t cred;
1212 	char **auth_str;
1213 	int *auth_len;
1214 	char *verf_str;
1215 	int *verf_len;
1216 	NFSKERBKEY_T key;		/* return session key */
1217 {
1218 	int error = 0;
1219 
1220 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
1221 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
1222 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
1223 			"nfsauth1", 2 * hz);
1224 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
1225 		if (error) {
1226 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
1227 			return (error);
1228 		}
1229 	}
1230 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
1231 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
1232 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
1233 	nmp->nm_verfstr = verf_str;
1234 	nmp->nm_verflen = *verf_len;
1235 	nmp->nm_authuid = kauth_cred_geteuid(cred);
1236 	wakeup((void *)&nmp->nm_authstr);
1237 
1238 	/*
1239 	 * And wait for mount_nfs to do its stuff.
1240 	 */
1241 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
1242 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
1243 			"nfsauth2", 2 * hz);
1244 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
1245 	}
1246 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
1247 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
1248 		error = EAUTH;
1249 	}
1250 	if (error)
1251 		free((void *)*auth_str, M_TEMP);
1252 	else {
1253 		*auth_len = nmp->nm_authlen;
1254 		*verf_len = nmp->nm_verflen;
1255 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
1256 	}
1257 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
1258 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
1259 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
1260 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
1261 		wakeup((void *)&nmp->nm_authtype);
1262 	}
1263 	return (error);
1264 }
1265 
1266 /*
1267  * Get a nickname authenticator and verifier.
1268  */
1269 int
1270 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
1271     int *auth_len, char *verf_str, int verf_len)
1272 {
1273 	struct timeval ktvin, ktvout, tv;
1274 	struct nfsuid *nuidp;
1275 	u_int32_t *nickp, *verfp;
1276 
1277 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
1278 
1279 #ifdef DIAGNOSTIC
1280 	if (verf_len < (4 * NFSX_UNSIGNED))
1281 		panic("nfs_getnickauth verf too small");
1282 #endif
1283 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
1284 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
1285 			break;
1286 	}
1287 	if (!nuidp || nuidp->nu_expire < time_second)
1288 		return (EACCES);
1289 
1290 	/*
1291 	 * Move to the end of the lru list (end of lru == most recently used).
1292 	 */
1293 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
1294 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
1295 
1296 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
1297 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
1298 	*nickp = txdr_unsigned(nuidp->nu_nickname);
1299 	*auth_str = (char *)nickp;
1300 	*auth_len = 2 * NFSX_UNSIGNED;
1301 
1302 	/*
1303 	 * Now we must encrypt the verifier and package it up.
1304 	 */
1305 	verfp = (u_int32_t *)verf_str;
1306 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
1307 	getmicrotime(&tv);
1308 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
1309 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
1310 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
1311 		nuidp->nu_timestamp = tv;
1312 	else
1313 		nuidp->nu_timestamp.tv_usec++;
1314 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
1315 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1316 
1317 	/*
1318 	 * Now encrypt the timestamp verifier in ecb mode using the session
1319 	 * key.
1320 	 */
1321 #ifdef NFSKERB
1322 	XXX
1323 #endif
1324 
1325 	*verfp++ = ktvout.tv_sec;
1326 	*verfp++ = ktvout.tv_usec;
1327 	*verfp = 0;
1328 	return (0);
1329 }
1330 
1331 /*
1332  * Save the current nickname in a hash list entry on the mount point.
1333  */
1334 int
1335 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
1336 	struct nfsmount *nmp;
1337 	kauth_cred_t cred;
1338 	int len;
1339 	NFSKERBKEY_T key;
1340 	struct mbuf **mdp;
1341 	char **dposp;
1342 	struct mbuf *mrep;
1343 {
1344 	struct nfsuid *nuidp;
1345 	u_int32_t *tl;
1346 	int32_t t1;
1347 	struct mbuf *md = *mdp;
1348 	struct timeval ktvin, ktvout;
1349 	u_int32_t nick;
1350 	char *dpos = *dposp, *cp2;
1351 	int deltasec, error = 0;
1352 
1353 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
1354 
1355 	if (len == (3 * NFSX_UNSIGNED)) {
1356 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1357 		ktvin.tv_sec = *tl++;
1358 		ktvin.tv_usec = *tl++;
1359 		nick = fxdr_unsigned(u_int32_t, *tl);
1360 
1361 		/*
1362 		 * Decrypt the timestamp in ecb mode.
1363 		 */
1364 #ifdef NFSKERB
1365 		XXX
1366 #endif
1367 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1368 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1369 		deltasec = time_second - ktvout.tv_sec;
1370 		if (deltasec < 0)
1371 			deltasec = -deltasec;
1372 		/*
1373 		 * If ok, add it to the hash list for the mount point.
1374 		 */
1375 		if (deltasec <= NFS_KERBCLOCKSKEW) {
1376 			if (nmp->nm_numuids < nuidhash_max) {
1377 				nmp->nm_numuids++;
1378 				nuidp = (struct nfsuid *)
1379 				   malloc(sizeof (struct nfsuid), M_NFSUID,
1380 					M_WAITOK);
1381 			} else {
1382 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1383 				LIST_REMOVE(nuidp, nu_hash);
1384 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1385 					nu_lru);
1386 			}
1387 			nuidp->nu_flag = 0;
1388 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
1389 			nuidp->nu_expire = time_second + NFS_KERBTTL;
1390 			nuidp->nu_timestamp = ktvout;
1391 			nuidp->nu_nickname = nick;
1392 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
1393 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1394 				nu_lru);
1395 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
1396 				nuidp, nu_hash);
1397 		}
1398 	} else
1399 		nfsm_adv(nfsm_rndup(len));
1400 nfsmout:
1401 	*mdp = md;
1402 	*dposp = dpos;
1403 	return (error);
1404 }
1405 #endif /* NFS */
1406