xref: /original-bsd/sys/nfs/nfs_nqlease.c (revision deff14a8)
1 /*
2  * Copyright (c) 1992, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * %sccs.include.redist.c%
9  *
10  *	@(#)nfs_nqlease.c	8.5 (Berkeley) 08/18/94
11  */
12 
13 /*
14  * References:
15  *	Cary G. Gray and David R. Cheriton, "Leases: An Efficient Fault-Tolerant
16  *		Mechanism for Distributed File Cache Consistency",
17  *		In Proc. of the Twelfth ACM Symposium on Operating Systems
18  *		Principals, pg. 202-210, Litchfield Park, AZ, Dec. 1989.
19  *	Michael N. Nelson, Brent B. Welch and John K. Ousterhout, "Caching
20  *		in the Sprite Network File System", ACM TOCS 6(1),
21  *		pages 134-154, February 1988.
22  *	V. Srinivasan and Jeffrey C. Mogul, "Spritely NFS: Implementation and
23  *		Performance of Cache-Consistency Protocols", Digital
24  *		Equipment Corporation WRL Research Report 89/5, May 1989.
25  */
26 #include <sys/param.h>
27 #include <sys/vnode.h>
28 #include <sys/mount.h>
29 #include <sys/kernel.h>
30 #include <sys/proc.h>
31 #include <sys/systm.h>
32 #include <sys/mbuf.h>
33 #include <sys/socket.h>
34 #include <sys/socketvar.h>
35 #include <sys/file.h>
36 #include <sys/buf.h>
37 #include <sys/stat.h>
38 #include <sys/protosw.h>
39 
40 #include <netinet/in.h>
41 #include <nfs/rpcv2.h>
42 #include <nfs/nfsv2.h>
43 #include <nfs/nfs.h>
44 #include <nfs/nfsm_subs.h>
45 #include <nfs/xdr_subs.h>
46 #include <nfs/nqnfs.h>
47 #include <nfs/nfsnode.h>
48 #include <nfs/nfsmount.h>
49 
50 time_t nqnfsstarttime = (time_t)0;
51 u_long nqnfs_prog, nqnfs_vers;
52 int nqsrv_clockskew = NQ_CLOCKSKEW;
53 int nqsrv_writeslack = NQ_WRITESLACK;
54 int nqsrv_maxlease = NQ_MAXLEASE;
55 int nqsrv_maxnumlease = NQ_MAXNUMLEASE;
56 void nqsrv_instimeq(), nqsrv_send_eviction(), nfs_sndunlock();
57 void nqsrv_unlocklease(), nqsrv_waitfor_expiry(), nfsrv_slpderef();
58 void nqsrv_addhost(), nqsrv_locklease(), nqnfs_serverd();
59 void nqnfs_clientlease();
60 struct mbuf *nfsm_rpchead();
61 
62 /*
63  * Signifies which rpcs can have piggybacked lease requests
64  */
65 int nqnfs_piggy[NFS_NPROCS] = {
66 	0,
67 	NQL_READ,
68 	NQL_WRITE,
69 	0,
70 	NQL_READ,
71 	NQL_READ,
72 	NQL_READ,
73 	0,
74 	NQL_WRITE,
75 	0,
76 	0,
77 	0,
78 	0,
79 	0,
80 	0,
81 	0,
82 	NQL_READ,
83 	0,
84 	NQL_READ,
85 	0,
86 	0,
87 	0,
88 	0,
89 };
90 
91 extern nfstype nfs_type[9];
92 extern struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock;
93 extern int nfsd_waiting;
94 
95 #define TRUE	1
96 #define	FALSE	0
97 
98 /*
99  * Get or check for a lease for "vp", based on NQL_CHECK flag.
100  * The rules are as follows:
101  * - if a current non-caching lease, reply non-caching
102  * - if a current lease for same host only, extend lease
103  * - if a read cachable lease and a read lease request
104  *	add host to list any reply cachable
105  * - else { set non-cachable for read-write sharing }
106  *	send eviction notice messages to all other hosts that have lease
107  *	wait for lease termination { either by receiving vacated messages
108  *					from all the other hosts or expiry
109  *					via. timeout }
110  *	modify lease to non-cachable
111  * - else if no current lease, issue new one
112  * - reply
113  * - return boolean TRUE iff nam should be m_freem()'d
114  * NB: Since nqnfs_serverd() is called from a timer, any potential tsleep()
115  *     in here must be framed by nqsrv_locklease() and nqsrv_unlocklease().
116  *     nqsrv_locklease() is coded such that at least one of LC_LOCKED and
117  *     LC_WANTED is set whenever a process is tsleeping in it. The exception
118  *     is when a new lease is being allocated, since it is not in the timer
119  *     queue yet. (Ditto for the splsoftclock() and splx(s) calls)
120  */
121 nqsrv_getlease(vp, duration, flags, nd, nam, cachablep, frev, cred)
122 	struct vnode *vp;
123 	u_long *duration;
124 	int flags;
125 	struct nfsd *nd;
126 	struct mbuf *nam;
127 	int *cachablep;
128 	u_quad_t *frev;
129 	struct ucred *cred;
130 {
131 	register struct nqlease *lp;
132 	register struct nqfhhashhead *lpp;
133 	register struct nqhost *lph;
134 	struct nqlease *tlp;
135 	struct nqm **lphp;
136 	struct vattr vattr;
137 	fhandle_t fh;
138 	int i, ok, error, s;
139 
140 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
141 		return (0);
142 	if (*duration > nqsrv_maxlease)
143 		*duration = nqsrv_maxlease;
144 	if (error = VOP_GETATTR(vp, &vattr, cred, nd->nd_procp))
145 		return (error);
146 	*frev = vattr.va_filerev;
147 	s = splsoftclock();
148 	tlp = vp->v_lease;
149 	if ((flags & NQL_CHECK) == 0)
150 		nfsstats.srvnqnfs_getleases++;
151 	if (tlp == 0) {
152 		/*
153 		 * Find the lease by searching the hash list.
154 		 */
155 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
156 		if (error = VFS_VPTOFH(vp, &fh.fh_fid)) {
157 			splx(s);
158 			return (error);
159 		}
160 		lpp = NQFHHASH(fh.fh_fid.fid_data);
161 		for (lp = lpp->lh_first; lp != 0; lp = lp->lc_hash.le_next)
162 			if (fh.fh_fsid.val[0] == lp->lc_fsid.val[0] &&
163 			    fh.fh_fsid.val[1] == lp->lc_fsid.val[1] &&
164 			    !bcmp(fh.fh_fid.fid_data, lp->lc_fiddata,
165 				  fh.fh_fid.fid_len - sizeof (long))) {
166 				/* Found it */
167 				lp->lc_vp = vp;
168 				vp->v_lease = lp;
169 				tlp = lp;
170 				break;
171 			}
172 	} else
173 		lp = tlp;
174 	if (lp != 0) {
175 		if ((lp->lc_flag & LC_NONCACHABLE) ||
176 		    (lp->lc_morehosts == (struct nqm *)0 &&
177 		     nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host)))
178 			goto doreply;
179 		if ((flags & NQL_READ) && (lp->lc_flag & LC_WRITE) == 0) {
180 			if (flags & NQL_CHECK)
181 				goto doreply;
182 			if (nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host))
183 				goto doreply;
184 			i = 0;
185 			if (lp->lc_morehosts) {
186 				lph = lp->lc_morehosts->lpm_hosts;
187 				lphp = &lp->lc_morehosts->lpm_next;
188 				ok = 1;
189 			} else {
190 				lphp = &lp->lc_morehosts;
191 				ok = 0;
192 			}
193 			while (ok && (lph->lph_flag & LC_VALID)) {
194 				if (nqsrv_cmpnam(nd->nd_slp, nam, lph))
195 					goto doreply;
196 				if (++i == LC_MOREHOSTSIZ) {
197 					i = 0;
198 					if (*lphp) {
199 						lph = (*lphp)->lpm_hosts;
200 						lphp = &((*lphp)->lpm_next);
201 					} else
202 						ok = 0;
203 				} else
204 					lph++;
205 			}
206 			nqsrv_locklease(lp);
207 			if (!ok) {
208 				*lphp = (struct nqm *)
209 					malloc(sizeof (struct nqm),
210 						M_NQMHOST, M_WAITOK);
211 				bzero((caddr_t)*lphp, sizeof (struct nqm));
212 				lph = (*lphp)->lpm_hosts;
213 			}
214 			nqsrv_addhost(lph, nd->nd_slp, nam);
215 			nqsrv_unlocklease(lp);
216 		} else {
217 			lp->lc_flag |= LC_NONCACHABLE;
218 			nqsrv_locklease(lp);
219 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
220 			nqsrv_waitfor_expiry(lp);
221 			nqsrv_unlocklease(lp);
222 		}
223 doreply:
224 		/*
225 		 * Update the lease and return
226 		 */
227 		if ((flags & NQL_CHECK) == 0)
228 			nqsrv_instimeq(lp, *duration);
229 		if (lp->lc_flag & LC_NONCACHABLE)
230 			*cachablep = 0;
231 		else {
232 			*cachablep = 1;
233 			if (flags & NQL_WRITE)
234 				lp->lc_flag |= LC_WRITTEN;
235 		}
236 		splx(s);
237 		return (0);
238 	}
239 	splx(s);
240 	if (flags & NQL_CHECK)
241 		return (0);
242 
243 	/*
244 	 * Allocate new lease
245 	 * The value of nqsrv_maxnumlease should be set generously, so that
246 	 * the following "printf" happens infrequently.
247 	 */
248 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
249 		printf("Nqnfs server, too many leases\n");
250 		do {
251 			(void) tsleep((caddr_t)&lbolt, PSOCK,
252 					"nqsrvnuml", 0);
253 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
254 	}
255 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
256 	bzero((caddr_t)lp, sizeof (struct nqlease));
257 	if (flags & NQL_WRITE)
258 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
259 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nam);
260 	lp->lc_vp = vp;
261 	lp->lc_fsid = fh.fh_fsid;
262 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
263 	LIST_INSERT_HEAD(lpp, lp, lc_hash);
264 	vp->v_lease = lp;
265 	s = splsoftclock();
266 	nqsrv_instimeq(lp, *duration);
267 	splx(s);
268 	*cachablep = 1;
269 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
270 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
271 	return (0);
272 }
273 
274 /*
275  * Local lease check for server syscalls.
276  * Just set up args and let nqsrv_getlease() do the rest.
277  */
278 lease_check(ap)
279 	struct vop_lease_args /* {
280 		struct vnode *a_vp;
281 		struct proc *a_p;
282 		struct ucred *a_cred;
283 		int a_flag;
284 	} */ *ap;
285 {
286 	int duration = 0, cache;
287 	struct nfsd nfsd;
288 	u_quad_t frev;
289 
290 	nfsd.nd_slp = NQLOCALSLP;
291 	nfsd.nd_procp = ap->a_p;
292 	(void) nqsrv_getlease(ap->a_vp, &duration, NQL_CHECK | ap->a_flag,
293 		&nfsd, (struct mbuf *)0, &cache, &frev, ap->a_cred);
294 	return (0);
295 }
296 
297 /*
298  * Add a host to an nqhost structure for a lease.
299  */
300 void
301 nqsrv_addhost(lph, slp, nam)
302 	register struct nqhost *lph;
303 	struct nfssvc_sock *slp;
304 	struct mbuf *nam;
305 {
306 	register struct sockaddr_in *saddr;
307 
308 	if (slp == NQLOCALSLP)
309 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
310 	else if (slp == nfs_udpsock) {
311 		saddr = mtod(nam, struct sockaddr_in *);
312 		lph->lph_flag |= (LC_VALID | LC_UDP);
313 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
314 		lph->lph_port = saddr->sin_port;
315 	} else if (slp == nfs_cltpsock) {
316 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
317 		lph->lph_flag |= (LC_VALID | LC_CLTP);
318 	} else {
319 		lph->lph_flag |= (LC_VALID | LC_SREF);
320 		lph->lph_slp = slp;
321 		slp->ns_sref++;
322 	}
323 }
324 
325 /*
326  * Update the lease expiry time and position it in the timer queue correctly.
327  */
328 void
329 nqsrv_instimeq(lp, duration)
330 	register struct nqlease *lp;
331 	u_long duration;
332 {
333 	register struct nqlease *tlp;
334 	time_t newexpiry;
335 
336 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
337 	if (lp->lc_expiry == newexpiry)
338 		return;
339 	if (lp->lc_timer.cqe_next != 0)
340 		CIRCLEQ_REMOVE(&nqtimerhead, lp, lc_timer);
341 	lp->lc_expiry = newexpiry;
342 
343 	/*
344 	 * Find where in the queue it should be.
345 	 */
346 	tlp = nqtimerhead.cqh_last;
347 	while (tlp != (void *)&nqtimerhead && tlp->lc_expiry > newexpiry)
348 		tlp = tlp->lc_timer.cqe_prev;
349 	if (tlp == nqtimerhead.cqh_last)
350 		NQSTORENOVRAM(newexpiry);
351 	if (tlp == (void *)&nqtimerhead) {
352 		CIRCLEQ_INSERT_HEAD(&nqtimerhead, lp, lc_timer);
353 	} else {
354 		CIRCLEQ_INSERT_AFTER(&nqtimerhead, tlp, lp, lc_timer);
355 	}
356 }
357 
358 /*
359  * Compare the requesting host address with the lph entry in the lease.
360  * Return true iff it is the same.
361  * This is somewhat messy due to the union in the nqhost structure.
362  * The local host is indicated by the special value of NQLOCALSLP for slp.
363  */
364 nqsrv_cmpnam(slp, nam, lph)
365 	register struct nfssvc_sock *slp;
366 	struct mbuf *nam;
367 	register struct nqhost *lph;
368 {
369 	register struct sockaddr_in *saddr;
370 	struct mbuf *addr;
371 	union nethostaddr lhaddr;
372 	int ret;
373 
374 	if (slp == NQLOCALSLP) {
375 		if (lph->lph_flag & LC_LOCAL)
376 			return (1);
377 		else
378 			return (0);
379 	}
380 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
381 		addr = nam;
382 	else
383 		addr = slp->ns_nam;
384 	if (lph->lph_flag & LC_UDP)
385 		ret = netaddr_match(AF_INET, &lph->lph_haddr, addr);
386 	else if (lph->lph_flag & LC_CLTP)
387 		ret = netaddr_match(AF_ISO, &lph->lph_claddr, addr);
388 	else {
389 		if ((lph->lph_slp->ns_flag & SLP_VALID) == 0)
390 			return (0);
391 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
392 		if (saddr->sin_family == AF_INET)
393 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
394 		else
395 			lhaddr.had_nam = lph->lph_slp->ns_nam;
396 		ret = netaddr_match(saddr->sin_family, &lhaddr, addr);
397 	}
398 	return (ret);
399 }
400 
401 /*
402  * Send out eviction notice messages to all other hosts for the lease.
403  */
404 void
405 nqsrv_send_eviction(vp, lp, slp, nam, cred)
406 	struct vnode *vp;
407 	register struct nqlease *lp;
408 	struct nfssvc_sock *slp;
409 	struct mbuf *nam;
410 	struct ucred *cred;
411 {
412 	register struct nqhost *lph = &lp->lc_host;
413 	register struct mbuf *m;
414 	register int siz;
415 	struct nqm *lphnext = lp->lc_morehosts;
416 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
417 	struct socket *so;
418 	struct sockaddr_in *saddr;
419 	fhandle_t *fhp;
420 	caddr_t bpos, cp;
421 	u_long xid;
422 	int len = 1, ok = 1, i = 0;
423 	int sotype, *solockp;
424 
425 	while (ok && (lph->lph_flag & LC_VALID)) {
426 		if (nqsrv_cmpnam(slp, nam, lph))
427 			lph->lph_flag |= LC_VACATED;
428 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
429 			if (lph->lph_flag & LC_UDP) {
430 				MGET(nam2, M_WAIT, MT_SONAME);
431 				saddr = mtod(nam2, struct sockaddr_in *);
432 				nam2->m_len = saddr->sin_len =
433 					sizeof (struct sockaddr_in);
434 				saddr->sin_family = AF_INET;
435 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
436 				saddr->sin_port = lph->lph_port;
437 				so = nfs_udpsock->ns_so;
438 			} else if (lph->lph_flag & LC_CLTP) {
439 				nam2 = lph->lph_nam;
440 				so = nfs_cltpsock->ns_so;
441 			} else if (lph->lph_slp->ns_flag & SLP_VALID) {
442 				nam2 = (struct mbuf *)0;
443 				so = lph->lph_slp->ns_so;
444 			} else
445 				goto nextone;
446 			sotype = so->so_type;
447 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
448 				solockp = &lph->lph_slp->ns_solock;
449 			else
450 				solockp = (int *)0;
451 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
452 				NFSX_FH);
453 			nfsm_build(cp, caddr_t, NFSX_FH);
454 			bzero(cp, NFSX_FH);
455 			fhp = (fhandle_t *)cp;
456 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
457 			VFS_VPTOFH(vp, &fhp->fh_fid);
458 			m = mreq;
459 			siz = 0;
460 			while (m) {
461 				siz += m->m_len;
462 				m = m->m_next;
463 			}
464 			if (siz <= 0 || siz > NFS_MAXPACKET) {
465 				printf("mbuf siz=%d\n",siz);
466 				panic("Bad nfs svc reply");
467 			}
468 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
469 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
470 				mreq, siz, &mheadend, &xid);
471 			/*
472 			 * For stream protocols, prepend a Sun RPC
473 			 * Record Mark.
474 			 */
475 			if (sotype == SOCK_STREAM) {
476 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
477 				*mtod(m, u_long *) = htonl(0x80000000 |
478 					(m->m_pkthdr.len - NFSX_UNSIGNED));
479 			}
480 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
481 			    (lph->lph_slp->ns_flag & SLP_VALID) == 0) ||
482 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
483 				m_freem(m);
484 			else {
485 				if (solockp)
486 					*solockp |= NFSMNT_SNDLOCK;
487 				(void) nfs_send(so, nam2, m,
488 						(struct nfsreq *)0);
489 				if (solockp)
490 					nfs_sndunlock(solockp);
491 			}
492 			if (lph->lph_flag & LC_UDP)
493 				MFREE(nam2, m);
494 		}
495 nextone:
496 		if (++i == len) {
497 			if (lphnext) {
498 				i = 0;
499 				len = LC_MOREHOSTSIZ;
500 				lph = lphnext->lpm_hosts;
501 				lphnext = lphnext->lpm_next;
502 			} else
503 				ok = 0;
504 		} else
505 			lph++;
506 	}
507 }
508 
509 /*
510  * Wait for the lease to expire.
511  * This will occur when all clients have sent "vacated" messages to
512  * this server OR when it expires do to timeout.
513  */
514 void
515 nqsrv_waitfor_expiry(lp)
516 	register struct nqlease *lp;
517 {
518 	register struct nqhost *lph;
519 	register int i;
520 	struct nqm *lphnext;
521 	int len, ok;
522 
523 tryagain:
524 	if (time.tv_sec > lp->lc_expiry)
525 		return;
526 	lph = &lp->lc_host;
527 	lphnext = lp->lc_morehosts;
528 	len = 1;
529 	i = 0;
530 	ok = 1;
531 	while (ok && (lph->lph_flag & LC_VALID)) {
532 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
533 			lp->lc_flag |= LC_EXPIREDWANTED;
534 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
535 					"nqexp", 0);
536 			goto tryagain;
537 		}
538 		if (++i == len) {
539 			if (lphnext) {
540 				i = 0;
541 				len = LC_MOREHOSTSIZ;
542 				lph = lphnext->lpm_hosts;
543 				lphnext = lphnext->lpm_next;
544 			} else
545 				ok = 0;
546 		} else
547 			lph++;
548 	}
549 }
550 
551 /*
552  * Nqnfs server timer that maintains the server lease queue.
553  * Scan the lease queue for expired entries:
554  * - when one is found, wakeup anyone waiting for it
555  *   else dequeue and free
556  */
557 void
558 nqnfs_serverd()
559 {
560 	register struct nqlease *lp, *lq;
561 	register struct nqhost *lph;
562 	struct nqlease *nextlp;
563 	struct nqm *lphnext, *olphnext;
564 	struct mbuf *n;
565 	int i, len, ok;
566 
567 	for (lp = nqtimerhead.cqh_first; lp != (void *)&nqtimerhead;
568 	    lp = nextlp) {
569 		if (lp->lc_expiry >= time.tv_sec)
570 			break;
571 		nextlp = lp->lc_timer.cqe_next;
572 		if (lp->lc_flag & LC_EXPIREDWANTED) {
573 			lp->lc_flag &= ~LC_EXPIREDWANTED;
574 			wakeup((caddr_t)&lp->lc_flag);
575 		} else if ((lp->lc_flag & (LC_LOCKED | LC_WANTED)) == 0) {
576 		    /*
577 		     * Make a best effort at keeping a write caching lease long
578 		     * enough by not deleting it until it has been explicitly
579 		     * vacated or there have been no writes in the previous
580 		     * write_slack seconds since expiry and the nfsds are not
581 		     * all busy. The assumption is that if the nfsds are not
582 		     * all busy now (no queue of nfs requests), then the client
583 		     * would have been able to do at least one write to the
584 		     * file during the last write_slack seconds if it was still
585 		     * trying to push writes to the server.
586 		     */
587 		    if ((lp->lc_flag & (LC_WRITE | LC_VACATED)) == LC_WRITE &&
588 			((lp->lc_flag & LC_WRITTEN) || nfsd_waiting == 0)) {
589 			lp->lc_flag &= ~LC_WRITTEN;
590 			nqsrv_instimeq(lp, nqsrv_writeslack);
591 		    } else {
592 			CIRCLEQ_REMOVE(&nqtimerhead, lp, lc_timer);
593 			LIST_REMOVE(lp, lc_hash);
594 			/*
595 			 * This soft reference may no longer be valid, but
596 			 * no harm done. The worst case is if the vnode was
597 			 * recycled and has another valid lease reference,
598 			 * which is dereferenced prematurely.
599 			 */
600 			lp->lc_vp->v_lease = (struct nqlease *)0;
601 			lph = &lp->lc_host;
602 			lphnext = lp->lc_morehosts;
603 			olphnext = (struct nqm *)0;
604 			len = 1;
605 			i = 0;
606 			ok = 1;
607 			while (ok && (lph->lph_flag & LC_VALID)) {
608 				if (lph->lph_flag & LC_CLTP)
609 					MFREE(lph->lph_nam, n);
610 				if (lph->lph_flag & LC_SREF)
611 					nfsrv_slpderef(lph->lph_slp);
612 				if (++i == len) {
613 					if (olphnext) {
614 						free((caddr_t)olphnext, M_NQMHOST);
615 						olphnext = (struct nqm *)0;
616 					}
617 					if (lphnext) {
618 						olphnext = lphnext;
619 						i = 0;
620 						len = LC_MOREHOSTSIZ;
621 						lph = lphnext->lpm_hosts;
622 						lphnext = lphnext->lpm_next;
623 					} else
624 						ok = 0;
625 				} else
626 					lph++;
627 			}
628 			FREE((caddr_t)lp, M_NQLEASE);
629 			if (olphnext)
630 				free((caddr_t)olphnext, M_NQMHOST);
631 			nfsstats.srvnqnfs_leases--;
632 		    }
633 		}
634 	}
635 }
636 
637 /*
638  * Called from nfssvc_nfsd() for a getlease rpc request.
639  * Do the from/to xdr translation and call nqsrv_getlease() to
640  * do the real work.
641  */
642 nqnfsrv_getlease(nfsd, mrep, md, dpos, cred, nam, mrq)
643 	struct nfsd *nfsd;
644 	struct mbuf *mrep, *md;
645 	caddr_t dpos;
646 	struct ucred *cred;
647 	struct mbuf *nam, **mrq;
648 {
649 	register struct nfsv2_fattr *fp;
650 	struct vattr va;
651 	register struct vattr *vap = &va;
652 	struct vnode *vp;
653 	nfsv2fh_t nfh;
654 	fhandle_t *fhp;
655 	register u_long *tl;
656 	register long t1;
657 	u_quad_t frev;
658 	caddr_t bpos;
659 	int error = 0;
660 	char *cp2;
661 	struct mbuf *mb, *mb2, *mreq;
662 	int flags, rdonly, cache;
663 
664 	fhp = &nfh.fh_generic;
665 	nfsm_srvmtofh(fhp);
666 	nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
667 	flags = fxdr_unsigned(int, *tl++);
668 	nfsd->nd_duration = fxdr_unsigned(int, *tl);
669 	if (error = nfsrv_fhtovp(fhp,
670 	    TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
671 		nfsm_reply(0);
672 	if (rdonly && flags == NQL_WRITE) {
673 		error = EROFS;
674 		nfsm_reply(0);
675 	}
676 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
677 		nam, &cache, &frev, cred);
678 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
679 	vput(vp);
680 	nfsm_reply(NFSX_NQFATTR + 4*NFSX_UNSIGNED);
681 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
682 	*tl++ = txdr_unsigned(cache);
683 	*tl++ = txdr_unsigned(nfsd->nd_duration);
684 	txdr_hyper(&frev, tl);
685 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_NQFATTR);
686 	nfsm_srvfillattr;
687 	nfsm_srvdone;
688 }
689 
690 /*
691  * Called from nfssvc_nfsd() when a "vacated" message is received from a
692  * client. Find the entry and expire it.
693  */
694 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
695 	struct nfsd *nfsd;
696 	struct mbuf *mrep, *md;
697 	caddr_t dpos;
698 	struct ucred *cred;
699 	struct mbuf *nam, **mrq;
700 {
701 	register struct nqlease *lp;
702 	register struct nqhost *lph;
703 	struct nqlease *tlp = (struct nqlease *)0;
704 	nfsv2fh_t nfh;
705 	fhandle_t *fhp;
706 	register u_long *tl;
707 	register long t1;
708 	struct nqm *lphnext;
709 	int error = 0, i, len, ok, gotit = 0;
710 	char *cp2;
711 
712 	fhp = &nfh.fh_generic;
713 	nfsm_srvmtofh(fhp);
714 	m_freem(mrep);
715 	/*
716 	 * Find the lease by searching the hash list.
717 	 */
718 	for (lp = NQFHHASH(fhp->fh_fid.fid_data)->lh_first; lp != 0;
719 	    lp = lp->lc_hash.le_next)
720 		if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
721 		    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
722 		    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
723 			  MAXFIDSZ)) {
724 			/* Found it */
725 			tlp = lp;
726 			break;
727 		}
728 	if (tlp != 0) {
729 		lp = tlp;
730 		len = 1;
731 		i = 0;
732 		lph = &lp->lc_host;
733 		lphnext = lp->lc_morehosts;
734 		ok = 1;
735 		while (ok && (lph->lph_flag & LC_VALID)) {
736 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
737 				lph->lph_flag |= LC_VACATED;
738 				gotit++;
739 				break;
740 			}
741 			if (++i == len) {
742 				if (lphnext) {
743 					len = LC_MOREHOSTSIZ;
744 					i = 0;
745 					lph = lphnext->lpm_hosts;
746 					lphnext = lphnext->lpm_next;
747 				} else
748 					ok = 0;
749 			} else
750 				lph++;
751 		}
752 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
753 			lp->lc_flag &= ~LC_EXPIREDWANTED;
754 			wakeup((caddr_t)&lp->lc_flag);
755 		}
756 nfsmout:
757 		return (EPERM);
758 	}
759 	return (EPERM);
760 }
761 
762 /*
763  * Client get lease rpc function.
764  */
765 nqnfs_getlease(vp, rwflag, cred, p)
766 	register struct vnode *vp;
767 	int rwflag;
768 	struct ucred *cred;
769 	struct proc *p;
770 {
771 	register u_long *tl;
772 	register caddr_t cp;
773 	register long t1;
774 	register struct nfsnode *np;
775 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
776 	caddr_t bpos, dpos, cp2;
777 	time_t reqtime;
778 	int error = 0;
779 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
780 	int cachable;
781 	u_quad_t frev;
782 
783 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
784 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
785 		 &bpos);
786 	nfsm_fhtom(vp);
787 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
788 	*tl++ = txdr_unsigned(rwflag);
789 	*tl = txdr_unsigned(nmp->nm_leaseterm);
790 	reqtime = time.tv_sec;
791 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
792 	np = VTONFS(vp);
793 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
794 	cachable = fxdr_unsigned(int, *tl++);
795 	reqtime += fxdr_unsigned(int, *tl++);
796 	if (reqtime > time.tv_sec) {
797 		fxdr_hyper(tl, &frev);
798 		nqnfs_clientlease(nmp, np, rwflag, cachable, reqtime, frev);
799 		nfsm_loadattr(vp, (struct vattr *)0);
800 	} else
801 		error = NQNFS_EXPIRED;
802 	nfsm_reqdone;
803 	return (error);
804 }
805 
806 /*
807  * Client vacated message function.
808  */
809 nqnfs_vacated(vp, cred)
810 	register struct vnode *vp;
811 	struct ucred *cred;
812 {
813 	register caddr_t cp;
814 	register struct mbuf *m;
815 	register int i;
816 	caddr_t bpos;
817 	u_long xid;
818 	int error = 0;
819 	struct mbuf *mreq, *mb, *mb2, *mheadend;
820 	struct nfsmount *nmp;
821 	struct nfsreq myrep;
822 
823 	nmp = VFSTONFS(vp->v_mount);
824 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
825 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
826 	nfsm_fhtom(vp);
827 	m = mreq;
828 	i = 0;
829 	while (m) {
830 		i += m->m_len;
831 		m = m->m_next;
832 	}
833 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
834 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
835 		mreq, i, &mheadend, &xid);
836 	if (nmp->nm_sotype == SOCK_STREAM) {
837 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
838 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
839 			NFSX_UNSIGNED));
840 	}
841 	myrep.r_flags = 0;
842 	myrep.r_nmp = nmp;
843 	if (nmp->nm_soflags & PR_CONNREQUIRED)
844 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
845 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
846 	if (nmp->nm_soflags & PR_CONNREQUIRED)
847 		nfs_sndunlock(&nmp->nm_flag);
848 	return (error);
849 }
850 
851 /*
852  * Called for client side callbacks
853  */
854 nqnfs_callback(nmp, mrep, md, dpos)
855 	struct nfsmount *nmp;
856 	struct mbuf *mrep, *md;
857 	caddr_t dpos;
858 {
859 	register struct vnode *vp;
860 	register u_long *tl;
861 	register long t1;
862 	nfsv2fh_t nfh;
863 	fhandle_t *fhp;
864 	struct nfsnode *np;
865 	struct nfsd nd;
866 	int error;
867 	char *cp2;
868 
869 	nd.nd_mrep = mrep;
870 	nd.nd_md = md;
871 	nd.nd_dpos = dpos;
872 	if (error = nfs_getreq(&nd, FALSE))
873 		return (error);
874 	md = nd.nd_md;
875 	dpos = nd.nd_dpos;
876 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
877 		m_freem(mrep);
878 		return (EPERM);
879 	}
880 	fhp = &nfh.fh_generic;
881 	nfsm_srvmtofh(fhp);
882 	m_freem(mrep);
883 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
884 		return (error);
885 	vp = NFSTOV(np);
886 	if (np->n_timer.cqe_next != 0) {
887 		np->n_expiry = 0;
888 		np->n_flag |= NQNFSEVICTED;
889 		if (nmp->nm_timerhead.cqh_first != np) {
890 			CIRCLEQ_REMOVE(&nmp->nm_timerhead, np, n_timer);
891 			CIRCLEQ_INSERT_HEAD(&nmp->nm_timerhead, np, n_timer);
892 		}
893 	}
894 	vrele(vp);
895 	nfsm_srvdone;
896 }
897 
898 /*
899  * Nqnfs client helper daemon. Runs once a second to expire leases.
900  * It also get authorization strings for "kerb" mounts.
901  * It must start at the beginning of the list again after any potential
902  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
903  * the list asynchronously.
904  */
905 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
906 	register struct nfsmount *nmp;
907 	struct ucred *cred;
908 	struct nfsd_cargs *ncd;
909 	int flag;
910 	caddr_t argp;
911 	struct proc *p;
912 {
913 	register struct nfsnode *np;
914 	struct vnode *vp;
915 	struct nfsreq myrep;
916 	int error, vpid;
917 
918 	/*
919 	 * First initialize some variables
920 	 */
921 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
922 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
923 
924 	/*
925 	 * If an authorization string is being passed in, get it.
926 	 */
927 	if ((flag & NFSSVC_GOTAUTH) &&
928 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
929 		if (nmp->nm_flag & NFSMNT_HASAUTH)
930 			panic("cld kerb");
931 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
932 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
933 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
934 				ncd->ncd_authlen) == 0) {
935 				nmp->nm_authtype = ncd->ncd_authtype;
936 				nmp->nm_authlen = ncd->ncd_authlen;
937 			} else
938 				nmp->nm_flag |= NFSMNT_AUTHERR;
939 		} else
940 			nmp->nm_flag |= NFSMNT_AUTHERR;
941 		nmp->nm_flag |= NFSMNT_HASAUTH;
942 		wakeup((caddr_t)&nmp->nm_authlen);
943 	} else
944 		nmp->nm_flag |= NFSMNT_WAITAUTH;
945 
946 	/*
947 	 * Loop every second updating queue until there is a termination sig.
948 	 */
949 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
950 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
951 		/*
952 		 * If there are no outstanding requests (and therefore no
953 		 * processes in nfs_reply) and there is data in the receive
954 		 * queue, poke for callbacks.
955 		 */
956 		if (nfs_reqq.tqh_first == 0 && nmp->nm_so &&
957 		    nmp->nm_so->so_rcv.sb_cc > 0) {
958 		    myrep.r_flags = R_GETONEREP;
959 		    myrep.r_nmp = nmp;
960 		    myrep.r_mrep = (struct mbuf *)0;
961 		    myrep.r_procp = (struct proc *)0;
962 		    (void) nfs_reply(&myrep);
963 		}
964 
965 		/*
966 		 * Loop through the leases, updating as required.
967 		 */
968 		np = nmp->nm_timerhead.cqh_first;
969 		while (np != (void *)&nmp->nm_timerhead &&
970 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
971 			vp = NFSTOV(np);
972 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
973 			vpid = vp->v_id;
974 			if (np->n_expiry < time.tv_sec) {
975 			   if (vget(vp, 1) == 0) {
976 			     nmp->nm_inprog = vp;
977 			     if (vpid == vp->v_id) {
978 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
979 				CIRCLEQ_REMOVE(&nmp->nm_timerhead, np, n_timer);
980 				np->n_timer.cqe_next = 0;
981 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
982 				    && vp->v_type == VREG) {
983 					if (np->n_flag & NQNFSEVICTED) {
984 						(void) nfs_vinvalbuf(vp,
985 						       V_SAVE, cred, p, 0);
986 						np->n_flag &= ~NQNFSEVICTED;
987 						(void) nqnfs_vacated(vp, cred);
988 					} else {
989 						(void) VOP_FSYNC(vp, cred,
990 						    MNT_WAIT, p);
991 						np->n_flag &= ~NMODIFIED;
992 					}
993 				}
994 			      }
995 			      vrele(vp);
996 			      nmp->nm_inprog = NULLVP;
997 			    }
998 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
999 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1000 				 == NQNFSWRITE && vp->v_dirtyblkhd.lh_first &&
1001 				 vget(vp, 1) == 0) {
1002 				 nmp->nm_inprog = vp;
1003 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1004 				 if (vpid == vp->v_id &&
1005 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1006 					np->n_brev = np->n_lrev;
1007 				 vrele(vp);
1008 				 nmp->nm_inprog = NULLVP;
1009 			    }
1010 			} else
1011 				break;
1012 			if (np == nmp->nm_timerhead.cqh_first)
1013 				break;
1014 			np = nmp->nm_timerhead.cqh_first;
1015 		}
1016 	    }
1017 
1018 	    /*
1019 	     * Get an authorization string, if required.
1020 	     */
1021 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1022 		ncd->ncd_authuid = nmp->nm_authuid;
1023 		if (copyout((caddr_t)ncd, argp, sizeof (struct nfsd_cargs)))
1024 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1025 		else
1026 			return (ENEEDAUTH);
1027 	    }
1028 
1029 	    /*
1030 	     * Wait a bit (no pun) and do it again.
1031 	     */
1032 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1033 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1034 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1035 			"nqnfstimr", hz / 3);
1036 		    if (error == EINTR || error == ERESTART)
1037 			(void) dounmount(nmp->nm_mountp, 0, p);
1038 	    }
1039 	}
1040 	free((caddr_t)nmp, M_NFSMNT);
1041 	if (error == EWOULDBLOCK)
1042 		error = 0;
1043 	return (error);
1044 }
1045 
1046 /*
1047  * Adjust all timer queue expiry times when the time of day clock is changed.
1048  * Called from the settimeofday() syscall.
1049  */
1050 void
1051 lease_updatetime(deltat)
1052 	register int deltat;
1053 {
1054 	register struct nqlease *lp;
1055 	register struct nfsnode *np;
1056 	struct mount *mp;
1057 	struct nfsmount *nmp;
1058 	int s;
1059 
1060 	if (nqnfsstarttime != 0)
1061 		nqnfsstarttime += deltat;
1062 	s = splsoftclock();
1063 	for (lp = nqtimerhead.cqh_first; lp != (void *)&nqtimerhead;
1064 	    lp = lp->lc_timer.cqe_next)
1065 		lp->lc_expiry += deltat;
1066 	splx(s);
1067 
1068 	/*
1069 	 * Search the mount list for all nqnfs mounts and do their timer
1070 	 * queues.
1071 	 */
1072 	for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) {
1073 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1074 			nmp = VFSTONFS(mp);
1075 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1076 				for (np = nmp->nm_timerhead.cqh_first;
1077 				    np != (void *)&nmp->nm_timerhead;
1078 				    np = np->n_timer.cqe_next) {
1079 					np->n_expiry += deltat;
1080 				}
1081 			}
1082 		}
1083 	}
1084 }
1085 
1086 /*
1087  * Lock a server lease.
1088  */
1089 void
1090 nqsrv_locklease(lp)
1091 	struct nqlease *lp;
1092 {
1093 
1094 	while (lp->lc_flag & LC_LOCKED) {
1095 		lp->lc_flag |= LC_WANTED;
1096 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1097 	}
1098 	lp->lc_flag |= LC_LOCKED;
1099 	lp->lc_flag &= ~LC_WANTED;
1100 }
1101 
1102 /*
1103  * Unlock a server lease.
1104  */
1105 void
1106 nqsrv_unlocklease(lp)
1107 	struct nqlease *lp;
1108 {
1109 
1110 	lp->lc_flag &= ~LC_LOCKED;
1111 	if (lp->lc_flag & LC_WANTED)
1112 		wakeup((caddr_t)lp);
1113 }
1114 
1115 /*
1116  * Update a client lease.
1117  */
1118 void
1119 nqnfs_clientlease(nmp, np, rwflag, cachable, expiry, frev)
1120 	register struct nfsmount *nmp;
1121 	register struct nfsnode *np;
1122 	int rwflag, cachable;
1123 	time_t expiry;
1124 	u_quad_t frev;
1125 {
1126 	register struct nfsnode *tp;
1127 
1128 	if (np->n_timer.cqe_next != 0) {
1129 		CIRCLEQ_REMOVE(&nmp->nm_timerhead, np, n_timer);
1130 		if (rwflag == NQL_WRITE)
1131 			np->n_flag |= NQNFSWRITE;
1132 	} else if (rwflag == NQL_READ)
1133 		np->n_flag &= ~NQNFSWRITE;
1134 	else
1135 		np->n_flag |= NQNFSWRITE;
1136 	if (cachable)
1137 		np->n_flag &= ~NQNFSNONCACHE;
1138 	else
1139 		np->n_flag |= NQNFSNONCACHE;
1140 	np->n_expiry = expiry;
1141 	np->n_lrev = frev;
1142 	tp = nmp->nm_timerhead.cqh_last;
1143 	while (tp != (void *)&nmp->nm_timerhead && tp->n_expiry > np->n_expiry)
1144 		tp = tp->n_timer.cqe_prev;
1145 	if (tp == (void *)&nmp->nm_timerhead) {
1146 		CIRCLEQ_INSERT_HEAD(&nmp->nm_timerhead, np, n_timer);
1147 	} else {
1148 		CIRCLEQ_INSERT_AFTER(&nmp->nm_timerhead, tp, np, n_timer);
1149 	}
1150 }
1151