xref: /original-bsd/sys/nfs/nfs_nqlease.c (revision dc8f81f2)
1 /*
2  * Copyright (c) 1992 The Regents of the University of California.
3  * 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	7.17 (Berkeley) 11/15/92
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 /*
51  * List head for the lease queue and other global data.
52  * At any time a lease is linked into a list ordered by increasing expiry time.
53  */
54 #define	NQFHHASH(f)	((*((u_long *)(f)))&nqfheadhash)
55 
56 union nqsrvthead nqthead;
57 struct nqlease **nqfhead;
58 u_long nqfheadhash;
59 time_t nqnfsstarttime = (time_t)0;
60 u_long nqnfs_prog, nqnfs_vers;
61 int nqsrv_clockskew = NQ_CLOCKSKEW;
62 int nqsrv_writeslack = NQ_WRITESLACK;
63 int nqsrv_maxlease = NQ_MAXLEASE;
64 int nqsrv_maxnumlease = NQ_MAXNUMLEASE;
65 void nqsrv_instimeq(), nqsrv_send_eviction(), nfs_sndunlock();
66 void nqsrv_unlocklease(), nqsrv_waitfor_expiry(), nfsrv_slpderef();
67 void nqsrv_addhost(), nqsrv_locklease(), nqnfs_serverd();
68 void nqnfs_clientlease();
69 struct mbuf *nfsm_rpchead();
70 
71 /*
72  * Signifies which rpcs can have piggybacked lease requests
73  */
74 int nqnfs_piggy[NFS_NPROCS] = {
75 	0,
76 	NQL_READ,
77 	NQL_WRITE,
78 	0,
79 	NQL_READ,
80 	NQL_READ,
81 	NQL_READ,
82 	0,
83 	NQL_WRITE,
84 	0,
85 	0,
86 	0,
87 	0,
88 	0,
89 	0,
90 	0,
91 	NQL_READ,
92 	0,
93 	NQL_READ,
94 	0,
95 	0,
96 	0,
97 	0,
98 };
99 
100 int nnnnnn = sizeof (struct nqlease);
101 int oooooo = sizeof (struct nfsnode);
102 extern nfstype nfs_type[9];
103 extern struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock;
104 extern struct nfsd nfsd_head;
105 extern int nfsd_waiting;
106 
107 #define TRUE	1
108 #define	FALSE	0
109 
110 /*
111  * Get or check for a lease for "vp", based on NQL_CHECK flag.
112  * The rules are as follows:
113  * - if a current non-caching lease, reply non-caching
114  * - if a current lease for same host only, extend lease
115  * - if a read cachable lease and a read lease request
116  *	add host to list any reply cachable
117  * - else { set non-cachable for read-write sharing }
118  *	send eviction notice messages to all other hosts that have lease
119  *	wait for lease termination { either by receiving vacated messages
120  *					from all the other hosts or expiry
121  *					via. timeout }
122  *	modify lease to non-cachable
123  * - else if no current lease, issue new one
124  * - reply
125  * - return boolean TRUE iff nam should be m_freem()'d
126  * NB: Since nqnfs_serverd() is called from a timer, any potential tsleep()
127  *     in here must be framed by nqsrv_locklease() and nqsrv_unlocklease().
128  *     nqsrv_locklease() is coded such that at least one of LC_LOCKED and
129  *     LC_WANTED is set whenever a process is tsleeping in it. The exception
130  *     is when a new lease is being allocated, since it is not in the timer
131  *     queue yet. (Ditto for the splsoftclock() and splx(s) calls)
132  */
133 nqsrv_getlease(vp, duration, flags, nd, nam, cachablep, frev, cred)
134 	struct vnode *vp;
135 	u_long *duration;
136 	int flags;
137 	struct nfsd *nd;
138 	struct mbuf *nam;
139 	int *cachablep;
140 	u_quad_t *frev;
141 	struct ucred *cred;
142 {
143 	register struct nqlease *lp, *lq, **lpp;
144 	register struct nqhost *lph;
145 	struct nqlease *tlp;
146 	struct nqm **lphp;
147 	struct vattr vattr;
148 	fhandle_t fh;
149 	int i, ok, error, s;
150 
151 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
152 		return (0);
153 	if (*duration > nqsrv_maxlease)
154 		*duration = nqsrv_maxlease;
155 	if (error = VOP_GETATTR(vp, &vattr, cred, nd->nd_procp))
156 		return (error);
157 	*frev = vattr.va_filerev;
158 	s = splsoftclock();
159 	tlp = vp->v_lease;
160 	if ((flags & NQL_CHECK) == 0)
161 		nfsstats.srvnqnfs_getleases++;
162 	if (tlp == (struct nqlease *)0) {
163 
164 		/*
165 		 * Find the lease by searching the hash list.
166 		 */
167 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
168 		if (error = VFS_VPTOFH(vp, &fh.fh_fid)) {
169 			splx(s);
170 			return (error);
171 		}
172 		lpp = &nqfhead[NQFHHASH(fh.fh_fid.fid_data)];
173 		for (lp = *lpp; lp; lp = lp->lc_fhnext)
174 			if (fh.fh_fsid.val[0] == lp->lc_fsid.val[0] &&
175 			    fh.fh_fsid.val[1] == lp->lc_fsid.val[1] &&
176 			    !bcmp(fh.fh_fid.fid_data, lp->lc_fiddata,
177 				  fh.fh_fid.fid_len - sizeof (long))) {
178 				/* Found it */
179 				lp->lc_vp = vp;
180 				vp->v_lease = lp;
181 				tlp = lp;
182 				break;
183 			}
184 	}
185 	lp = tlp;
186 	if (lp) {
187 		if ((lp->lc_flag & LC_NONCACHABLE) ||
188 		    (lp->lc_morehosts == (struct nqm *)0 &&
189 		     nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host)))
190 			goto doreply;
191 		if ((flags & NQL_READ) && (lp->lc_flag & LC_WRITE)==0) {
192 			if (flags & NQL_CHECK)
193 				goto doreply;
194 			if (nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host))
195 				goto doreply;
196 			i = 0;
197 			if (lp->lc_morehosts) {
198 				lph = lp->lc_morehosts->lpm_hosts;
199 				lphp = &lp->lc_morehosts->lpm_next;
200 				ok = 1;
201 			} else {
202 				lphp = &lp->lc_morehosts;
203 				ok = 0;
204 			}
205 			while (ok && (lph->lph_flag & LC_VALID)) {
206 				if (nqsrv_cmpnam(nd->nd_slp, nam, lph))
207 					goto doreply;
208 				if (++i == LC_MOREHOSTSIZ) {
209 					i = 0;
210 					if (*lphp) {
211 						lph = (*lphp)->lpm_hosts;
212 						lphp = &((*lphp)->lpm_next);
213 					} else
214 						ok = 0;
215 				} else
216 					lph++;
217 			}
218 			nqsrv_locklease(lp);
219 			if (!ok) {
220 				*lphp = (struct nqm *)
221 					malloc(sizeof (struct nqm),
222 						M_NQMHOST, M_WAITOK);
223 				bzero((caddr_t)*lphp, sizeof (struct nqm));
224 				lph = (*lphp)->lpm_hosts;
225 			}
226 			nqsrv_addhost(lph, nd->nd_slp, nam);
227 			nqsrv_unlocklease(lp);
228 		} else {
229 			lp->lc_flag |= LC_NONCACHABLE;
230 			nqsrv_locklease(lp);
231 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
232 			nqsrv_waitfor_expiry(lp);
233 			nqsrv_unlocklease(lp);
234 		}
235 doreply:
236 		/*
237 		 * Update the lease and return
238 		 */
239 		if ((flags & NQL_CHECK) == 0)
240 			nqsrv_instimeq(lp, *duration);
241 		if (lp->lc_flag & LC_NONCACHABLE)
242 			*cachablep = 0;
243 		else {
244 			*cachablep = 1;
245 			if (flags & NQL_WRITE)
246 				lp->lc_flag |= LC_WRITTEN;
247 		}
248 		splx(s);
249 		return (0);
250 	}
251 	splx(s);
252 	if (flags & NQL_CHECK)
253 		return (0);
254 
255 	/*
256 	 * Allocate new lease
257 	 * The value of nqsrv_maxnumlease should be set generously, so that
258 	 * the following "printf" happens infrequently.
259 	 */
260 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
261 		printf("Nqnfs server, too many leases\n");
262 		do {
263 			(void) tsleep((caddr_t)&lbolt, PSOCK,
264 					"nqsrvnuml", 0);
265 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
266 	}
267 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
268 	bzero((caddr_t)lp, sizeof (struct nqlease));
269 	if (flags & NQL_WRITE)
270 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
271 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nam);
272 	lp->lc_vp = vp;
273 	lp->lc_fsid = fh.fh_fsid;
274 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
275 	if (lq = *lpp)
276 		lq->lc_fhprev = &lp->lc_fhnext;
277 	lp->lc_fhnext = lq;
278 	lp->lc_fhprev = lpp;
279 	*lpp = lp;
280 	vp->v_lease = lp;
281 	s = splsoftclock();
282 	nqsrv_instimeq(lp, *duration);
283 	splx(s);
284 	*cachablep = 1;
285 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
286 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
287 	return (0);
288 }
289 
290 /*
291  * Local lease check for server syscalls.
292  * Just set up args and let nqsrv_getlease() do the rest.
293  */
294 void
295 lease_check(vp, p, cred, flag)
296 	struct vnode *vp;
297 	struct proc *p;
298 	struct ucred *cred;
299 	int flag;
300 {
301 	int duration = 0, cache;
302 	struct nfsd nfsd;
303 	u_quad_t frev;
304 
305 	nfsd.nd_slp = NQLOCALSLP;
306 	nfsd.nd_procp = p;
307 	(void) nqsrv_getlease(vp, &duration, NQL_CHECK | flag, &nfsd,
308 		(struct mbuf *)0, &cache, &frev, cred);
309 }
310 
311 /*
312  * Add a host to an nqhost structure for a lease.
313  */
314 void
315 nqsrv_addhost(lph, slp, nam)
316 	register struct nqhost *lph;
317 	struct nfssvc_sock *slp;
318 	struct mbuf *nam;
319 {
320 	register struct sockaddr_in *saddr;
321 
322 	if (slp == NQLOCALSLP)
323 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
324 	else if (slp == nfs_udpsock) {
325 		saddr = mtod(nam, struct sockaddr_in *);
326 		lph->lph_flag |= (LC_VALID | LC_UDP);
327 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
328 		lph->lph_port = saddr->sin_port;
329 	} else if (slp == nfs_cltpsock) {
330 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
331 		lph->lph_flag |= (LC_VALID | LC_CLTP);
332 	} else {
333 		lph->lph_flag |= (LC_VALID | LC_SREF);
334 		lph->lph_slp = slp;
335 		slp->ns_sref++;
336 	}
337 }
338 
339 /*
340  * Update the lease expiry time and position it in the timer queue correctly.
341  */
342 void
343 nqsrv_instimeq(lp, duration)
344 	register struct nqlease *lp;
345 	u_long duration;
346 {
347 	register struct nqlease *tlp;
348 	time_t newexpiry;
349 
350 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
351 	if (lp->lc_expiry == newexpiry)
352 		return;
353 	if (lp->lc_chain1[0])
354 		remque(lp);
355 	lp->lc_expiry = newexpiry;
356 
357 	/*
358 	 * Find where in the queue it should be.
359 	 */
360 	tlp = nqthead.th_chain[1];
361 	while (tlp->lc_expiry > newexpiry && tlp != (struct nqlease *)&nqthead)
362 		tlp = tlp->lc_chain1[1];
363 	if (tlp == nqthead.th_chain[1])
364 		NQSTORENOVRAM(newexpiry);
365 	insque(lp, tlp);
366 }
367 
368 /*
369  * Compare the requesting host address with the lph entry in the lease.
370  * Return true iff it is the same.
371  * This is somewhat messy due to the union in the nqhost structure.
372  * The local host is indicated by the special value of NQLOCALSLP for slp.
373  */
374 nqsrv_cmpnam(slp, nam, lph)
375 	register struct nfssvc_sock *slp;
376 	struct mbuf *nam;
377 	register struct nqhost *lph;
378 {
379 	register struct sockaddr_in *saddr;
380 	struct mbuf *addr;
381 	union nethostaddr lhaddr;
382 	int ret;
383 
384 	if (slp == NQLOCALSLP) {
385 		if (lph->lph_flag & LC_LOCAL)
386 			return (1);
387 		else
388 			return (0);
389 	}
390 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
391 		addr = nam;
392 	else
393 		addr = slp->ns_nam;
394 	if (lph->lph_flag & LC_UDP)
395 		ret = netaddr_match(AF_INET, &lph->lph_haddr, addr);
396 	else if (lph->lph_flag & LC_CLTP)
397 		ret = netaddr_match(AF_ISO, &lph->lph_claddr, addr);
398 	else {
399 		if ((lph->lph_slp->ns_flag & SLP_VALID) == 0)
400 			return (0);
401 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
402 		if (saddr->sin_family == AF_INET)
403 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
404 		else
405 			lhaddr.had_nam = lph->lph_slp->ns_nam;
406 		ret = netaddr_match(saddr->sin_family, &lhaddr, addr);
407 	}
408 	return (ret);
409 }
410 
411 /*
412  * Send out eviction notice messages to all other hosts for the lease.
413  */
414 void
415 nqsrv_send_eviction(vp, lp, slp, nam, cred)
416 	struct vnode *vp;
417 	register struct nqlease *lp;
418 	struct nfssvc_sock *slp;
419 	struct mbuf *nam;
420 	struct ucred *cred;
421 {
422 	register struct nqhost *lph = &lp->lc_host;
423 	register struct mbuf *m;
424 	register int siz;
425 	struct nqm *lphnext = lp->lc_morehosts;
426 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
427 	struct socket *so;
428 	struct sockaddr_in *saddr;
429 	fhandle_t *fhp;
430 	caddr_t bpos, cp;
431 	u_long xid;
432 	int len = 1, ok = 1, i = 0;
433 	int sotype, *solockp;
434 
435 	while (ok && (lph->lph_flag & LC_VALID)) {
436 		if (nqsrv_cmpnam(slp, nam, lph))
437 			lph->lph_flag |= LC_VACATED;
438 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
439 			if (lph->lph_flag & LC_UDP) {
440 				MGET(nam2, M_WAIT, MT_SONAME);
441 				saddr = mtod(nam2, struct sockaddr_in *);
442 				nam2->m_len = saddr->sin_len =
443 					sizeof (struct sockaddr_in);
444 				saddr->sin_family = AF_INET;
445 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
446 				saddr->sin_port = lph->lph_port;
447 				so = nfs_udpsock->ns_so;
448 			} else if (lph->lph_flag & LC_CLTP) {
449 				nam2 = lph->lph_nam;
450 				so = nfs_cltpsock->ns_so;
451 			} else if (lph->lph_slp->ns_flag & SLP_VALID) {
452 				nam2 = (struct mbuf *)0;
453 				so = lph->lph_slp->ns_so;
454 			} else
455 				goto nextone;
456 			sotype = so->so_type;
457 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
458 				solockp = &lph->lph_slp->ns_solock;
459 			else
460 				solockp = (int *)0;
461 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
462 				NFSX_FH);
463 			nfsm_build(cp, caddr_t, NFSX_FH);
464 			bzero(cp, NFSX_FH);
465 			fhp = (fhandle_t *)cp;
466 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
467 			VFS_VPTOFH(vp, &fhp->fh_fid);
468 			m = mreq;
469 			siz = 0;
470 			while (m) {
471 				siz += m->m_len;
472 				m = m->m_next;
473 			}
474 			if (siz <= 0 || siz > NFS_MAXPACKET) {
475 				printf("mbuf siz=%d\n",siz);
476 				panic("Bad nfs svc reply");
477 			}
478 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
479 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
480 				mreq, siz, &mheadend, &xid);
481 			/*
482 			 * For stream protocols, prepend a Sun RPC
483 			 * Record Mark.
484 			 */
485 			if (sotype == SOCK_STREAM) {
486 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
487 				*mtod(m, u_long *) = htonl(0x80000000 |
488 					(m->m_pkthdr.len - NFSX_UNSIGNED));
489 			}
490 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
491 			    (lph->lph_slp->ns_flag & SLP_VALID) == 0) ||
492 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
493 				m_freem(m);
494 			else {
495 				if (solockp)
496 					*solockp |= NFSMNT_SNDLOCK;
497 				(void) nfs_send(so, nam2, m,
498 						(struct nfsreq *)0);
499 				if (solockp)
500 					nfs_sndunlock(solockp);
501 			}
502 			if (lph->lph_flag & LC_UDP)
503 				MFREE(nam2, m);
504 		}
505 nextone:
506 		if (++i == len) {
507 			if (lphnext) {
508 				i = 0;
509 				len = LC_MOREHOSTSIZ;
510 				lph = lphnext->lpm_hosts;
511 				lphnext = lphnext->lpm_next;
512 			} else
513 				ok = 0;
514 		} else
515 			lph++;
516 	}
517 }
518 
519 /*
520  * Wait for the lease to expire.
521  * This will occur when all clients have sent "vacated" messages to
522  * this server OR when it expires do to timeout.
523  */
524 void
525 nqsrv_waitfor_expiry(lp)
526 	register struct nqlease *lp;
527 {
528 	register struct nqhost *lph;
529 	register int i;
530 	struct nqm *lphnext;
531 	int len, ok;
532 
533 tryagain:
534 	if (time.tv_sec > lp->lc_expiry)
535 		return;
536 	lph = &lp->lc_host;
537 	lphnext = lp->lc_morehosts;
538 	len = 1;
539 	i = 0;
540 	ok = 1;
541 	while (ok && (lph->lph_flag & LC_VALID)) {
542 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
543 			lp->lc_flag |= LC_EXPIREDWANTED;
544 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
545 					"nqexp", 0);
546 			goto tryagain;
547 		}
548 		if (++i == len) {
549 			if (lphnext) {
550 				i = 0;
551 				len = LC_MOREHOSTSIZ;
552 				lph = lphnext->lpm_hosts;
553 				lphnext = lphnext->lpm_next;
554 			} else
555 				ok = 0;
556 		} else
557 			lph++;
558 	}
559 }
560 
561 /*
562  * Nqnfs server timer that maintains the server lease queue.
563  * Scan the lease queue for expired entries:
564  * - when one is found, wakeup anyone waiting for it
565  *   else dequeue and free
566  */
567 void
568 nqnfs_serverd()
569 {
570 	register struct nqlease *lp, *lq;
571 	register struct nqhost *lph;
572 	struct nqlease *nextlp;
573 	struct nqm *lphnext, *olphnext;
574 	struct mbuf *n;
575 	union nqsrvthead *lhp;
576 	int i, len, ok;
577 
578 	lp = nqthead.th_chain[0];
579 	while (lp != (struct nqlease *)&nqthead) {
580 		if (lp->lc_expiry >= time.tv_sec)
581 			break;
582 		nextlp = lp->lc_chain1[0];
583 		if (lp->lc_flag & LC_EXPIREDWANTED) {
584 			lp->lc_flag &= ~LC_EXPIREDWANTED;
585 			wakeup((caddr_t)&lp->lc_flag);
586 		} else if ((lp->lc_flag & (LC_LOCKED | LC_WANTED)) == 0) {
587 		    /*
588 		     * Make a best effort at keeping a write caching lease long
589 		     * enough by not deleting it until it has been explicitly
590 		     * vacated or there have been no writes in the previous
591 		     * write_slack seconds since expiry and the nfsds are not
592 		     * all busy. The assumption is that if the nfsds are not
593 		     * all busy now (no queue of nfs requests), then the client
594 		     * would have been able to do at least one write to the
595 		     * file during the last write_slack seconds if it was still
596 		     * trying to push writes to the server.
597 		     */
598 		    if ((lp->lc_flag & (LC_WRITE | LC_VACATED)) == LC_WRITE &&
599 			((lp->lc_flag & LC_WRITTEN) || nfsd_waiting == 0)) {
600 			lp->lc_flag &= ~LC_WRITTEN;
601 			nqsrv_instimeq(lp, nqsrv_writeslack);
602 		    } else {
603 			remque(lp);
604 			if (lq = lp->lc_fhnext)
605 				lq->lc_fhprev = lp->lc_fhprev;
606 			*lp->lc_fhprev = lq;
607 			/*
608 			 * This soft reference may no longer be valid, but
609 			 * no harm done. The worst case is if the vnode was
610 			 * recycled and has another valid lease reference,
611 			 * which is dereferenced prematurely.
612 			 */
613 			lp->lc_vp->v_lease = (struct nqlease *)0;
614 			lph = &lp->lc_host;
615 			lphnext = lp->lc_morehosts;
616 			olphnext = (struct nqm *)0;
617 			len = 1;
618 			i = 0;
619 			ok = 1;
620 			while (ok && (lph->lph_flag & LC_VALID)) {
621 				if (lph->lph_flag & LC_CLTP)
622 					MFREE(lph->lph_nam, n);
623 				if (lph->lph_flag & LC_SREF)
624 					nfsrv_slpderef(lph->lph_slp);
625 				if (++i == len) {
626 					if (olphnext) {
627 						free((caddr_t)olphnext, M_NQMHOST);
628 						olphnext = (struct nqm *)0;
629 					}
630 					if (lphnext) {
631 						olphnext = lphnext;
632 						i = 0;
633 						len = LC_MOREHOSTSIZ;
634 						lph = lphnext->lpm_hosts;
635 						lphnext = lphnext->lpm_next;
636 					} else
637 						ok = 0;
638 				} else
639 					lph++;
640 			}
641 			FREE((caddr_t)lp, M_NQLEASE);
642 			if (olphnext)
643 				free((caddr_t)olphnext, M_NQMHOST);
644 			nfsstats.srvnqnfs_leases--;
645 		    }
646 		}
647 		lp = nextlp;
648 	}
649 }
650 
651 /*
652  * Called from nfssvc_nfsd() for a getlease rpc request.
653  * Do the from/to xdr translation and call nqsrv_getlease() to
654  * do the real work.
655  */
656 nqnfsrv_getlease(nfsd, mrep, md, dpos, cred, nam, mrq)
657 	struct nfsd *nfsd;
658 	struct mbuf *mrep, *md;
659 	caddr_t dpos;
660 	struct ucred *cred;
661 	struct mbuf *nam, **mrq;
662 {
663 	register struct nfsv2_fattr *fp;
664 	struct vattr va;
665 	register struct vattr *vap = &va;
666 	struct vnode *vp;
667 	nfsv2fh_t nfh;
668 	fhandle_t *fhp;
669 	register u_long *tl;
670 	register long t1;
671 	u_quad_t frev;
672 	caddr_t bpos;
673 	int error = 0;
674 	char *cp2;
675 	struct mbuf *mb, *mb2, *mreq;
676 	int flags, rdonly, cache;
677 
678 	fhp = &nfh.fh_generic;
679 	nfsm_srvmtofh(fhp);
680 	nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
681 	flags = fxdr_unsigned(int, *tl++);
682 	nfsd->nd_duration = fxdr_unsigned(int, *tl);
683 	if (error = nfsrv_fhtovp(fhp, TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
684 		nfsm_reply(0);
685 	if (rdonly && flags == NQL_WRITE) {
686 		error = EROFS;
687 		nfsm_reply(0);
688 	}
689 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
690 		nam, &cache, &frev, cred);
691 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
692 	vput(vp);
693 	nfsm_reply(NFSX_NQFATTR + 4*NFSX_UNSIGNED);
694 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
695 	*tl++ = txdr_unsigned(cache);
696 	*tl++ = txdr_unsigned(nfsd->nd_duration);
697 	txdr_hyper(&frev, tl);
698 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_NQFATTR);
699 	nfsm_srvfillattr;
700 	nfsm_srvdone;
701 }
702 
703 /*
704  * Called from nfssvc_nfsd() when a "vacated" message is received from a
705  * client. Find the entry and expire it.
706  */
707 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
708 	struct nfsd *nfsd;
709 	struct mbuf *mrep, *md;
710 	caddr_t dpos;
711 	struct ucred *cred;
712 	struct mbuf *nam, **mrq;
713 {
714 	register struct nqlease *lp;
715 	register struct nqhost *lph;
716 	struct nqlease *tlp = (struct nqlease *)0;
717 	struct vnode *vp;
718 	nfsv2fh_t nfh;
719 	fhandle_t *fhp;
720 	register u_long *tl;
721 	register long t1;
722 	struct nqm *lphnext;
723 	union nqsrvthead *lhp;
724 	u_quad_t frev;
725 	int error = 0, i, len, ok, rdonly, gotit = 0;
726 	char *cp2;
727 
728 	fhp = &nfh.fh_generic;
729 	nfsm_srvmtofh(fhp);
730 	m_freem(mrep);
731 	/*
732 	 * Find the lease by searching the hash list.
733 	 */
734 	for (lp = nqfhead[NQFHHASH(fhp->fh_fid.fid_data)]; lp;
735 	     lp = lp->lc_fhnext)
736 		if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
737 		    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
738 		    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
739 			  MAXFIDSZ)) {
740 			/* Found it */
741 			tlp = lp;
742 			break;
743 		}
744 	if (tlp) {
745 		lp = tlp;
746 		len = 1;
747 		i = 0;
748 		lph = &lp->lc_host;
749 		lphnext = lp->lc_morehosts;
750 		ok = 1;
751 		while (ok && (lph->lph_flag & LC_VALID)) {
752 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
753 				lph->lph_flag |= LC_VACATED;
754 				gotit++;
755 				break;
756 			}
757 			if (++i == len) {
758 				if (lphnext) {
759 					len = LC_MOREHOSTSIZ;
760 					i = 0;
761 					lph = lphnext->lpm_hosts;
762 					lphnext = lphnext->lpm_next;
763 				} else
764 					ok = 0;
765 			} else
766 				lph++;
767 		}
768 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
769 			lp->lc_flag &= ~LC_EXPIREDWANTED;
770 			wakeup((caddr_t)&lp->lc_flag);
771 		}
772 nfsmout:
773 		return (EPERM);
774 	}
775 	return (EPERM);
776 }
777 
778 /*
779  * Client get lease rpc function.
780  */
781 nqnfs_getlease(vp, rwflag, cred, p)
782 	register struct vnode *vp;
783 	int rwflag;
784 	struct ucred *cred;
785 	struct proc *p;
786 {
787 	register u_long *tl;
788 	register caddr_t cp;
789 	register long t1;
790 	register struct nfsnode *np, *tp;
791 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
792 	caddr_t bpos, dpos, cp2;
793 	time_t reqtime;
794 	int error = 0;
795 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
796 	int cachable;
797 	u_quad_t frev;
798 
799 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
800 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
801 		 &bpos);
802 	nfsm_fhtom(vp);
803 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
804 	*tl++ = txdr_unsigned(rwflag);
805 	*tl = txdr_unsigned(nmp->nm_leaseterm);
806 	reqtime = time.tv_sec;
807 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
808 	np = VTONFS(vp);
809 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
810 	cachable = fxdr_unsigned(int, *tl++);
811 	reqtime += fxdr_unsigned(int, *tl++);
812 	if (reqtime > time.tv_sec) {
813 		fxdr_hyper(tl, &frev);
814 		nqnfs_clientlease(nmp, np, rwflag, cachable, reqtime, frev);
815 		nfsm_loadattr(vp, (struct vattr *)0);
816 	} else
817 		error = NQNFS_EXPIRED;
818 	nfsm_reqdone;
819 	return (error);
820 }
821 
822 /*
823  * Client vacated message function.
824  */
825 nqnfs_vacated(vp, cred)
826 	register struct vnode *vp;
827 	struct ucred *cred;
828 {
829 	register caddr_t cp;
830 	register struct mbuf *m;
831 	register int i;
832 	caddr_t bpos;
833 	u_long xid;
834 	int error = 0;
835 	struct mbuf *mreq, *mb, *mb2, *mheadend;
836 	struct nfsmount *nmp;
837 	struct nfsreq myrep;
838 
839 	nmp = VFSTONFS(vp->v_mount);
840 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
841 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
842 	nfsm_fhtom(vp);
843 	m = mreq;
844 	i = 0;
845 	while (m) {
846 		i += m->m_len;
847 		m = m->m_next;
848 	}
849 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
850 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
851 		mreq, i, &mheadend, &xid);
852 	if (nmp->nm_sotype == SOCK_STREAM) {
853 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
854 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
855 			NFSX_UNSIGNED));
856 	}
857 	myrep.r_flags = 0;
858 	myrep.r_nmp = nmp;
859 	if (nmp->nm_soflags & PR_CONNREQUIRED)
860 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
861 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
862 	if (nmp->nm_soflags & PR_CONNREQUIRED)
863 		nfs_sndunlock(&nmp->nm_flag);
864 	return (error);
865 }
866 
867 /*
868  * Called for client side callbacks
869  */
870 nqnfs_callback(nmp, mrep, md, dpos)
871 	struct nfsmount *nmp;
872 	struct mbuf *mrep, *md;
873 	caddr_t dpos;
874 {
875 	register struct vnode *vp;
876 	register u_long *tl;
877 	register long t1;
878 	nfsv2fh_t nfh;
879 	fhandle_t *fhp;
880 	struct nfsnode *np;
881 	struct nfsd nd;
882 	int error;
883 	char *cp2;
884 
885 	nd.nd_mrep = mrep;
886 	nd.nd_md = md;
887 	nd.nd_dpos = dpos;
888 	if (error = nfs_getreq(&nd, FALSE))
889 		return (error);
890 	md = nd.nd_md;
891 	dpos = nd.nd_dpos;
892 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
893 		m_freem(mrep);
894 		return (EPERM);
895 	}
896 	fhp = &nfh.fh_generic;
897 	nfsm_srvmtofh(fhp);
898 	m_freem(mrep);
899 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
900 		return (error);
901 	vp = NFSTOV(np);
902 	if (np->n_tnext) {
903 		np->n_expiry = 0;
904 		np->n_flag |= NQNFSEVICTED;
905 		if (np->n_tprev != (struct nfsnode *)nmp) {
906 			if (np->n_tnext == (struct nfsnode *)nmp)
907 				nmp->nm_tprev = np->n_tprev;
908 			else
909 				np->n_tnext->n_tprev = np->n_tprev;
910 			np->n_tprev->n_tnext = np->n_tnext;
911 			np->n_tnext = nmp->nm_tnext;
912 			nmp->nm_tnext = np;
913 			np->n_tprev = (struct nfsnode *)nmp;
914 			if (np->n_tnext == (struct nfsnode *)nmp)
915 				nmp->nm_tprev = np;
916 			else
917 				np->n_tnext->n_tprev = np;
918 		}
919 	}
920 	vrele(vp);
921 	nfsm_srvdone;
922 }
923 
924 /*
925  * Nqnfs client helper daemon. Runs once a second to expire leases.
926  * It also get authorization strings for "kerb" mounts.
927  * It must start at the beginning of the list again after any potential
928  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
929  * the list asynchronously.
930  */
931 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
932 	register struct nfsmount *nmp;
933 	struct ucred *cred;
934 	struct nfsd_cargs *ncd;
935 	int flag;
936 	caddr_t argp;
937 	struct proc *p;
938 {
939 	register struct nfsnode *np;
940 	struct vnode *vp;
941 	int error, vpid;
942 
943 	/*
944 	 * First initialize some variables
945 	 */
946 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
947 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
948 
949 	/*
950 	 * If an authorization string is being passed in, get it.
951 	 */
952 	if ((flag & NFSSVC_GOTAUTH) &&
953 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
954 		if (nmp->nm_flag & NFSMNT_HASAUTH)
955 			panic("cld kerb");
956 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
957 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
958 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
959 				ncd->ncd_authlen) == 0) {
960 				nmp->nm_authtype = ncd->ncd_authtype;
961 				nmp->nm_authlen = ncd->ncd_authlen;
962 			} else
963 				nmp->nm_flag |= NFSMNT_AUTHERR;
964 		} else
965 			nmp->nm_flag |= NFSMNT_AUTHERR;
966 		nmp->nm_flag |= NFSMNT_HASAUTH;
967 		wakeup((caddr_t)&nmp->nm_authlen);
968 	} else
969 		nmp->nm_flag |= NFSMNT_WAITAUTH;
970 
971 	/*
972 	 * Loop every second updating queue until there is a termination sig.
973 	 */
974 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
975 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
976 		np = nmp->nm_tnext;
977 		while (np != (struct nfsnode *)nmp &&
978 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
979 			vp = NFSTOV(np);
980 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
981 			vpid = vp->v_id;
982 			if (np->n_expiry < time.tv_sec) {
983 			   if (vget(vp) == 0) {
984 			     nmp->nm_inprog = vp;
985 			     if (vpid == vp->v_id) {
986 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
987 				if (np->n_tnext == (struct nfsnode *)nmp)
988 					nmp->nm_tprev = np->n_tprev;
989 				else
990 					np->n_tnext->n_tprev = np->n_tprev;
991 				if (np->n_tprev == (struct nfsnode *)nmp)
992 					nmp->nm_tnext = np->n_tnext;
993 				else
994 					np->n_tprev->n_tnext = np->n_tnext;
995 				np->n_tnext = (struct nfsnode *)0;
996 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
997 				    && vp->v_type == VREG) {
998 					if (np->n_flag & NQNFSEVICTED) {
999 						NFS_VINVBUF(np, vp,
1000 						            TRUE, cred, p);
1001 						np->n_flag &= ~NQNFSEVICTED;
1002 						(void) nqnfs_vacated(vp, cred);
1003 					} else {
1004 						np->n_flag &= ~NMODIFIED;
1005 						(void) VOP_FSYNC(vp, cred,
1006 						    MNT_WAIT, p);
1007 					}
1008 				}
1009 			      }
1010 			      vrele(vp);
1011 			      nmp->nm_inprog = NULLVP;
1012 			    }
1013 			    if (np != nmp->nm_tnext)
1014 				np = nmp->nm_tnext;
1015 			    else
1016 				break;
1017 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
1018 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1019 				 == NQNFSWRITE && vp->v_dirtyblkhd.le_next &&
1020 				 vget(vp) == 0) {
1021 				 nmp->nm_inprog = vp;
1022 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1023 				 if (vpid == vp->v_id &&
1024 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1025 					np->n_brev = np->n_lrev;
1026 				 vrele(vp);
1027 				 nmp->nm_inprog = NULLVP;
1028 			    }
1029 			    if (np != nmp->nm_tnext)
1030 				np = nmp->nm_tnext;
1031 			    else
1032 				break;
1033 			} else
1034 				break;
1035 		}
1036 	    }
1037 
1038 	    /*
1039 	     * Get an authorization string, if required.
1040 	     */
1041 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1042 		ncd->ncd_authuid = nmp->nm_authuid;
1043 		if (copyout((caddr_t)ncd, argp, sizeof (*ncd)))
1044 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1045 		else
1046 			return (ENEEDAUTH);
1047 	    }
1048 
1049 	    /*
1050 	     * Wait a bit (no pun) and do it again.
1051 	     */
1052 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1053 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1054 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1055 			"nqnfstimr", hz / 3);
1056 		    if (error == EINTR || error == ERESTART)
1057 			(void) dounmount(nmp->nm_mountp, 0);
1058 	    }
1059 	}
1060 	free((caddr_t)nmp, M_NFSMNT);
1061 	if (error == EWOULDBLOCK)
1062 		error = 0;
1063 	return (error);
1064 }
1065 
1066 /*
1067  * Adjust all timer queue expiry times when the time of day clock is changed.
1068  * Called from the settimeofday() syscall.
1069  */
1070 void
1071 lease_updatetime(deltat)
1072 	register int deltat;
1073 {
1074 	register struct nqlease *lp;
1075 	register struct nfsnode *np;
1076 	struct mount *mp;
1077 	struct nfsmount *nmp;
1078 	int s;
1079 
1080 	if (nqnfsstarttime != 0)
1081 		nqnfsstarttime += deltat;
1082 	s = splsoftclock();
1083 	lp = nqthead.th_chain[0];
1084 	while (lp != (struct nqlease *)&nqthead) {
1085 		lp->lc_expiry += deltat;
1086 		lp = lp->lc_chain1[0];
1087 	}
1088 	splx(s);
1089 
1090 	/*
1091 	 * Search the mount list for all nqnfs mounts and do their timer
1092 	 * queues.
1093 	 */
1094 	mp = rootfs;
1095 	do {
1096 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1097 			nmp = VFSTONFS(mp);
1098 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1099 				np = nmp->nm_tnext;
1100 				while (np != (struct nfsnode *)nmp) {
1101 					np->n_expiry += deltat;
1102 					np = np->n_tnext;
1103 				}
1104 			}
1105 		}
1106 		mp = mp->mnt_next;
1107 	} while (mp != rootfs);
1108 }
1109 
1110 /*
1111  * Lock a server lease.
1112  */
1113 void
1114 nqsrv_locklease(lp)
1115 	struct nqlease *lp;
1116 {
1117 
1118 	while (lp->lc_flag & LC_LOCKED) {
1119 		lp->lc_flag |= LC_WANTED;
1120 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1121 	}
1122 	lp->lc_flag |= LC_LOCKED;
1123 	lp->lc_flag &= ~LC_WANTED;
1124 }
1125 
1126 /*
1127  * Unlock a server lease.
1128  */
1129 void
1130 nqsrv_unlocklease(lp)
1131 	struct nqlease *lp;
1132 {
1133 
1134 	lp->lc_flag &= ~LC_LOCKED;
1135 	if (lp->lc_flag & LC_WANTED)
1136 		wakeup((caddr_t)lp);
1137 }
1138 
1139 /*
1140  * Update a client lease.
1141  */
1142 void
1143 nqnfs_clientlease(nmp, np, rwflag, cachable, expiry, frev)
1144 	register struct nfsmount *nmp;
1145 	register struct nfsnode *np;
1146 	int rwflag, cachable;
1147 	time_t expiry;
1148 	u_quad_t frev;
1149 {
1150 	register struct nfsnode *tp;
1151 
1152 	if (np->n_tnext) {
1153 		if (np->n_tnext == (struct nfsnode *)nmp)
1154 			nmp->nm_tprev = np->n_tprev;
1155 		else
1156 			np->n_tnext->n_tprev = np->n_tprev;
1157 		if (np->n_tprev == (struct nfsnode *)nmp)
1158 			nmp->nm_tnext = np->n_tnext;
1159 		else
1160 			np->n_tprev->n_tnext = np->n_tnext;
1161 		if (rwflag == NQL_WRITE)
1162 			np->n_flag |= NQNFSWRITE;
1163 	} else if (rwflag == NQL_READ)
1164 		np->n_flag &= ~NQNFSWRITE;
1165 	else
1166 		np->n_flag |= NQNFSWRITE;
1167 	if (cachable)
1168 		np->n_flag &= ~NQNFSNONCACHE;
1169 	else
1170 		np->n_flag |= NQNFSNONCACHE;
1171 	np->n_expiry = expiry;
1172 	np->n_lrev = frev;
1173 	tp = nmp->nm_tprev;
1174 	while (tp != (struct nfsnode *)nmp && tp->n_expiry > np->n_expiry)
1175 		tp = tp->n_tprev;
1176 	if (tp == (struct nfsnode *)nmp) {
1177 		np->n_tnext = nmp->nm_tnext;
1178 		nmp->nm_tnext = np;
1179 	} else {
1180 		np->n_tnext = tp->n_tnext;
1181 		tp->n_tnext = np;
1182 	}
1183 	np->n_tprev = tp;
1184 	if (np->n_tnext == (struct nfsnode *)nmp)
1185 		nmp->nm_tprev = np;
1186 	else
1187 		np->n_tnext->n_tprev = np;
1188 }
1189