xref: /original-bsd/sys/nfs/nfs_nqlease.c (revision d45ebeed)
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	7.20 (Berkeley) 04/27/93
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 extern struct nfsreq nfsreqh;
107 
108 #define TRUE	1
109 #define	FALSE	0
110 
111 /*
112  * Get or check for a lease for "vp", based on NQL_CHECK flag.
113  * The rules are as follows:
114  * - if a current non-caching lease, reply non-caching
115  * - if a current lease for same host only, extend lease
116  * - if a read cachable lease and a read lease request
117  *	add host to list any reply cachable
118  * - else { set non-cachable for read-write sharing }
119  *	send eviction notice messages to all other hosts that have lease
120  *	wait for lease termination { either by receiving vacated messages
121  *					from all the other hosts or expiry
122  *					via. timeout }
123  *	modify lease to non-cachable
124  * - else if no current lease, issue new one
125  * - reply
126  * - return boolean TRUE iff nam should be m_freem()'d
127  * NB: Since nqnfs_serverd() is called from a timer, any potential tsleep()
128  *     in here must be framed by nqsrv_locklease() and nqsrv_unlocklease().
129  *     nqsrv_locklease() is coded such that at least one of LC_LOCKED and
130  *     LC_WANTED is set whenever a process is tsleeping in it. The exception
131  *     is when a new lease is being allocated, since it is not in the timer
132  *     queue yet. (Ditto for the splsoftclock() and splx(s) calls)
133  */
134 nqsrv_getlease(vp, duration, flags, nd, nam, cachablep, frev, cred)
135 	struct vnode *vp;
136 	u_long *duration;
137 	int flags;
138 	struct nfsd *nd;
139 	struct mbuf *nam;
140 	int *cachablep;
141 	u_quad_t *frev;
142 	struct ucred *cred;
143 {
144 	register struct nqlease *lp, *lq, **lpp;
145 	register struct nqhost *lph;
146 	struct nqlease *tlp;
147 	struct nqm **lphp;
148 	struct vattr vattr;
149 	fhandle_t fh;
150 	int i, ok, error, s;
151 
152 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
153 		return (0);
154 	if (*duration > nqsrv_maxlease)
155 		*duration = nqsrv_maxlease;
156 	if (error = VOP_GETATTR(vp, &vattr, cred, nd->nd_procp))
157 		return (error);
158 	*frev = vattr.va_filerev;
159 	s = splsoftclock();
160 	tlp = vp->v_lease;
161 	if ((flags & NQL_CHECK) == 0)
162 		nfsstats.srvnqnfs_getleases++;
163 	if (tlp == (struct nqlease *)0) {
164 
165 		/*
166 		 * Find the lease by searching the hash list.
167 		 */
168 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
169 		if (error = VFS_VPTOFH(vp, &fh.fh_fid)) {
170 			splx(s);
171 			return (error);
172 		}
173 		lpp = &nqfhead[NQFHHASH(fh.fh_fid.fid_data)];
174 		for (lp = *lpp; lp; lp = lp->lc_fhnext)
175 			if (fh.fh_fsid.val[0] == lp->lc_fsid.val[0] &&
176 			    fh.fh_fsid.val[1] == lp->lc_fsid.val[1] &&
177 			    !bcmp(fh.fh_fid.fid_data, lp->lc_fiddata,
178 				  fh.fh_fid.fid_len - sizeof (long))) {
179 				/* Found it */
180 				lp->lc_vp = vp;
181 				vp->v_lease = lp;
182 				tlp = lp;
183 				break;
184 			}
185 	}
186 	lp = tlp;
187 	if (lp) {
188 		if ((lp->lc_flag & LC_NONCACHABLE) ||
189 		    (lp->lc_morehosts == (struct nqm *)0 &&
190 		     nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host)))
191 			goto doreply;
192 		if ((flags & NQL_READ) && (lp->lc_flag & LC_WRITE)==0) {
193 			if (flags & NQL_CHECK)
194 				goto doreply;
195 			if (nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host))
196 				goto doreply;
197 			i = 0;
198 			if (lp->lc_morehosts) {
199 				lph = lp->lc_morehosts->lpm_hosts;
200 				lphp = &lp->lc_morehosts->lpm_next;
201 				ok = 1;
202 			} else {
203 				lphp = &lp->lc_morehosts;
204 				ok = 0;
205 			}
206 			while (ok && (lph->lph_flag & LC_VALID)) {
207 				if (nqsrv_cmpnam(nd->nd_slp, nam, lph))
208 					goto doreply;
209 				if (++i == LC_MOREHOSTSIZ) {
210 					i = 0;
211 					if (*lphp) {
212 						lph = (*lphp)->lpm_hosts;
213 						lphp = &((*lphp)->lpm_next);
214 					} else
215 						ok = 0;
216 				} else
217 					lph++;
218 			}
219 			nqsrv_locklease(lp);
220 			if (!ok) {
221 				*lphp = (struct nqm *)
222 					malloc(sizeof (struct nqm),
223 						M_NQMHOST, M_WAITOK);
224 				bzero((caddr_t)*lphp, sizeof (struct nqm));
225 				lph = (*lphp)->lpm_hosts;
226 			}
227 			nqsrv_addhost(lph, nd->nd_slp, nam);
228 			nqsrv_unlocklease(lp);
229 		} else {
230 			lp->lc_flag |= LC_NONCACHABLE;
231 			nqsrv_locklease(lp);
232 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
233 			nqsrv_waitfor_expiry(lp);
234 			nqsrv_unlocklease(lp);
235 		}
236 doreply:
237 		/*
238 		 * Update the lease and return
239 		 */
240 		if ((flags & NQL_CHECK) == 0)
241 			nqsrv_instimeq(lp, *duration);
242 		if (lp->lc_flag & LC_NONCACHABLE)
243 			*cachablep = 0;
244 		else {
245 			*cachablep = 1;
246 			if (flags & NQL_WRITE)
247 				lp->lc_flag |= LC_WRITTEN;
248 		}
249 		splx(s);
250 		return (0);
251 	}
252 	splx(s);
253 	if (flags & NQL_CHECK)
254 		return (0);
255 
256 	/*
257 	 * Allocate new lease
258 	 * The value of nqsrv_maxnumlease should be set generously, so that
259 	 * the following "printf" happens infrequently.
260 	 */
261 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
262 		printf("Nqnfs server, too many leases\n");
263 		do {
264 			(void) tsleep((caddr_t)&lbolt, PSOCK,
265 					"nqsrvnuml", 0);
266 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
267 	}
268 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
269 	bzero((caddr_t)lp, sizeof (struct nqlease));
270 	if (flags & NQL_WRITE)
271 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
272 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nam);
273 	lp->lc_vp = vp;
274 	lp->lc_fsid = fh.fh_fsid;
275 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
276 	if (lq = *lpp)
277 		lq->lc_fhprev = &lp->lc_fhnext;
278 	lp->lc_fhnext = lq;
279 	lp->lc_fhprev = lpp;
280 	*lpp = lp;
281 	vp->v_lease = lp;
282 	s = splsoftclock();
283 	nqsrv_instimeq(lp, *duration);
284 	splx(s);
285 	*cachablep = 1;
286 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
287 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
288 	return (0);
289 }
290 
291 /*
292  * Local lease check for server syscalls.
293  * Just set up args and let nqsrv_getlease() do the rest.
294  */
295 void
296 lease_check(vp, p, cred, flag)
297 	struct vnode *vp;
298 	struct proc *p;
299 	struct ucred *cred;
300 	int flag;
301 {
302 	int duration = 0, cache;
303 	struct nfsd nfsd;
304 	u_quad_t frev;
305 
306 	nfsd.nd_slp = NQLOCALSLP;
307 	nfsd.nd_procp = p;
308 	(void) nqsrv_getlease(vp, &duration, NQL_CHECK | flag, &nfsd,
309 		(struct mbuf *)0, &cache, &frev, cred);
310 }
311 
312 /*
313  * Add a host to an nqhost structure for a lease.
314  */
315 void
316 nqsrv_addhost(lph, slp, nam)
317 	register struct nqhost *lph;
318 	struct nfssvc_sock *slp;
319 	struct mbuf *nam;
320 {
321 	register struct sockaddr_in *saddr;
322 
323 	if (slp == NQLOCALSLP)
324 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
325 	else if (slp == nfs_udpsock) {
326 		saddr = mtod(nam, struct sockaddr_in *);
327 		lph->lph_flag |= (LC_VALID | LC_UDP);
328 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
329 		lph->lph_port = saddr->sin_port;
330 	} else if (slp == nfs_cltpsock) {
331 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
332 		lph->lph_flag |= (LC_VALID | LC_CLTP);
333 	} else {
334 		lph->lph_flag |= (LC_VALID | LC_SREF);
335 		lph->lph_slp = slp;
336 		slp->ns_sref++;
337 	}
338 }
339 
340 /*
341  * Update the lease expiry time and position it in the timer queue correctly.
342  */
343 void
344 nqsrv_instimeq(lp, duration)
345 	register struct nqlease *lp;
346 	u_long duration;
347 {
348 	register struct nqlease *tlp;
349 	time_t newexpiry;
350 
351 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
352 	if (lp->lc_expiry == newexpiry)
353 		return;
354 	if (lp->lc_chain1[0])
355 		remque(lp);
356 	lp->lc_expiry = newexpiry;
357 
358 	/*
359 	 * Find where in the queue it should be.
360 	 */
361 	tlp = nqthead.th_chain[1];
362 	while (tlp->lc_expiry > newexpiry && tlp != (struct nqlease *)&nqthead)
363 		tlp = tlp->lc_chain1[1];
364 	if (tlp == nqthead.th_chain[1])
365 		NQSTORENOVRAM(newexpiry);
366 	insque(lp, tlp);
367 }
368 
369 /*
370  * Compare the requesting host address with the lph entry in the lease.
371  * Return true iff it is the same.
372  * This is somewhat messy due to the union in the nqhost structure.
373  * The local host is indicated by the special value of NQLOCALSLP for slp.
374  */
375 nqsrv_cmpnam(slp, nam, lph)
376 	register struct nfssvc_sock *slp;
377 	struct mbuf *nam;
378 	register struct nqhost *lph;
379 {
380 	register struct sockaddr_in *saddr;
381 	struct mbuf *addr;
382 	union nethostaddr lhaddr;
383 	int ret;
384 
385 	if (slp == NQLOCALSLP) {
386 		if (lph->lph_flag & LC_LOCAL)
387 			return (1);
388 		else
389 			return (0);
390 	}
391 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
392 		addr = nam;
393 	else
394 		addr = slp->ns_nam;
395 	if (lph->lph_flag & LC_UDP)
396 		ret = netaddr_match(AF_INET, &lph->lph_haddr, addr);
397 	else if (lph->lph_flag & LC_CLTP)
398 		ret = netaddr_match(AF_ISO, &lph->lph_claddr, addr);
399 	else {
400 		if ((lph->lph_slp->ns_flag & SLP_VALID) == 0)
401 			return (0);
402 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
403 		if (saddr->sin_family == AF_INET)
404 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
405 		else
406 			lhaddr.had_nam = lph->lph_slp->ns_nam;
407 		ret = netaddr_match(saddr->sin_family, &lhaddr, addr);
408 	}
409 	return (ret);
410 }
411 
412 /*
413  * Send out eviction notice messages to all other hosts for the lease.
414  */
415 void
416 nqsrv_send_eviction(vp, lp, slp, nam, cred)
417 	struct vnode *vp;
418 	register struct nqlease *lp;
419 	struct nfssvc_sock *slp;
420 	struct mbuf *nam;
421 	struct ucred *cred;
422 {
423 	register struct nqhost *lph = &lp->lc_host;
424 	register struct mbuf *m;
425 	register int siz;
426 	struct nqm *lphnext = lp->lc_morehosts;
427 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
428 	struct socket *so;
429 	struct sockaddr_in *saddr;
430 	fhandle_t *fhp;
431 	caddr_t bpos, cp;
432 	u_long xid;
433 	int len = 1, ok = 1, i = 0;
434 	int sotype, *solockp;
435 
436 	while (ok && (lph->lph_flag & LC_VALID)) {
437 		if (nqsrv_cmpnam(slp, nam, lph))
438 			lph->lph_flag |= LC_VACATED;
439 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
440 			if (lph->lph_flag & LC_UDP) {
441 				MGET(nam2, M_WAIT, MT_SONAME);
442 				saddr = mtod(nam2, struct sockaddr_in *);
443 				nam2->m_len = saddr->sin_len =
444 					sizeof (struct sockaddr_in);
445 				saddr->sin_family = AF_INET;
446 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
447 				saddr->sin_port = lph->lph_port;
448 				so = nfs_udpsock->ns_so;
449 			} else if (lph->lph_flag & LC_CLTP) {
450 				nam2 = lph->lph_nam;
451 				so = nfs_cltpsock->ns_so;
452 			} else if (lph->lph_slp->ns_flag & SLP_VALID) {
453 				nam2 = (struct mbuf *)0;
454 				so = lph->lph_slp->ns_so;
455 			} else
456 				goto nextone;
457 			sotype = so->so_type;
458 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
459 				solockp = &lph->lph_slp->ns_solock;
460 			else
461 				solockp = (int *)0;
462 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
463 				NFSX_FH);
464 			nfsm_build(cp, caddr_t, NFSX_FH);
465 			bzero(cp, NFSX_FH);
466 			fhp = (fhandle_t *)cp;
467 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
468 			VFS_VPTOFH(vp, &fhp->fh_fid);
469 			m = mreq;
470 			siz = 0;
471 			while (m) {
472 				siz += m->m_len;
473 				m = m->m_next;
474 			}
475 			if (siz <= 0 || siz > NFS_MAXPACKET) {
476 				printf("mbuf siz=%d\n",siz);
477 				panic("Bad nfs svc reply");
478 			}
479 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
480 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
481 				mreq, siz, &mheadend, &xid);
482 			/*
483 			 * For stream protocols, prepend a Sun RPC
484 			 * Record Mark.
485 			 */
486 			if (sotype == SOCK_STREAM) {
487 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
488 				*mtod(m, u_long *) = htonl(0x80000000 |
489 					(m->m_pkthdr.len - NFSX_UNSIGNED));
490 			}
491 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
492 			    (lph->lph_slp->ns_flag & SLP_VALID) == 0) ||
493 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
494 				m_freem(m);
495 			else {
496 				if (solockp)
497 					*solockp |= NFSMNT_SNDLOCK;
498 				(void) nfs_send(so, nam2, m,
499 						(struct nfsreq *)0);
500 				if (solockp)
501 					nfs_sndunlock(solockp);
502 			}
503 			if (lph->lph_flag & LC_UDP)
504 				MFREE(nam2, m);
505 		}
506 nextone:
507 		if (++i == len) {
508 			if (lphnext) {
509 				i = 0;
510 				len = LC_MOREHOSTSIZ;
511 				lph = lphnext->lpm_hosts;
512 				lphnext = lphnext->lpm_next;
513 			} else
514 				ok = 0;
515 		} else
516 			lph++;
517 	}
518 }
519 
520 /*
521  * Wait for the lease to expire.
522  * This will occur when all clients have sent "vacated" messages to
523  * this server OR when it expires do to timeout.
524  */
525 void
526 nqsrv_waitfor_expiry(lp)
527 	register struct nqlease *lp;
528 {
529 	register struct nqhost *lph;
530 	register int i;
531 	struct nqm *lphnext;
532 	int len, ok;
533 
534 tryagain:
535 	if (time.tv_sec > lp->lc_expiry)
536 		return;
537 	lph = &lp->lc_host;
538 	lphnext = lp->lc_morehosts;
539 	len = 1;
540 	i = 0;
541 	ok = 1;
542 	while (ok && (lph->lph_flag & LC_VALID)) {
543 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
544 			lp->lc_flag |= LC_EXPIREDWANTED;
545 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
546 					"nqexp", 0);
547 			goto tryagain;
548 		}
549 		if (++i == len) {
550 			if (lphnext) {
551 				i = 0;
552 				len = LC_MOREHOSTSIZ;
553 				lph = lphnext->lpm_hosts;
554 				lphnext = lphnext->lpm_next;
555 			} else
556 				ok = 0;
557 		} else
558 			lph++;
559 	}
560 }
561 
562 /*
563  * Nqnfs server timer that maintains the server lease queue.
564  * Scan the lease queue for expired entries:
565  * - when one is found, wakeup anyone waiting for it
566  *   else dequeue and free
567  */
568 void
569 nqnfs_serverd()
570 {
571 	register struct nqlease *lp, *lq;
572 	register struct nqhost *lph;
573 	struct nqlease *nextlp;
574 	struct nqm *lphnext, *olphnext;
575 	struct mbuf *n;
576 	union nqsrvthead *lhp;
577 	int i, len, ok;
578 
579 	lp = nqthead.th_chain[0];
580 	while (lp != (struct nqlease *)&nqthead) {
581 		if (lp->lc_expiry >= time.tv_sec)
582 			break;
583 		nextlp = lp->lc_chain1[0];
584 		if (lp->lc_flag & LC_EXPIREDWANTED) {
585 			lp->lc_flag &= ~LC_EXPIREDWANTED;
586 			wakeup((caddr_t)&lp->lc_flag);
587 		} else if ((lp->lc_flag & (LC_LOCKED | LC_WANTED)) == 0) {
588 		    /*
589 		     * Make a best effort at keeping a write caching lease long
590 		     * enough by not deleting it until it has been explicitly
591 		     * vacated or there have been no writes in the previous
592 		     * write_slack seconds since expiry and the nfsds are not
593 		     * all busy. The assumption is that if the nfsds are not
594 		     * all busy now (no queue of nfs requests), then the client
595 		     * would have been able to do at least one write to the
596 		     * file during the last write_slack seconds if it was still
597 		     * trying to push writes to the server.
598 		     */
599 		    if ((lp->lc_flag & (LC_WRITE | LC_VACATED)) == LC_WRITE &&
600 			((lp->lc_flag & LC_WRITTEN) || nfsd_waiting == 0)) {
601 			lp->lc_flag &= ~LC_WRITTEN;
602 			nqsrv_instimeq(lp, nqsrv_writeslack);
603 		    } else {
604 			remque(lp);
605 			if (lq = lp->lc_fhnext)
606 				lq->lc_fhprev = lp->lc_fhprev;
607 			*lp->lc_fhprev = lq;
608 			/*
609 			 * This soft reference may no longer be valid, but
610 			 * no harm done. The worst case is if the vnode was
611 			 * recycled and has another valid lease reference,
612 			 * which is dereferenced prematurely.
613 			 */
614 			lp->lc_vp->v_lease = (struct nqlease *)0;
615 			lph = &lp->lc_host;
616 			lphnext = lp->lc_morehosts;
617 			olphnext = (struct nqm *)0;
618 			len = 1;
619 			i = 0;
620 			ok = 1;
621 			while (ok && (lph->lph_flag & LC_VALID)) {
622 				if (lph->lph_flag & LC_CLTP)
623 					MFREE(lph->lph_nam, n);
624 				if (lph->lph_flag & LC_SREF)
625 					nfsrv_slpderef(lph->lph_slp);
626 				if (++i == len) {
627 					if (olphnext) {
628 						free((caddr_t)olphnext, M_NQMHOST);
629 						olphnext = (struct nqm *)0;
630 					}
631 					if (lphnext) {
632 						olphnext = lphnext;
633 						i = 0;
634 						len = LC_MOREHOSTSIZ;
635 						lph = lphnext->lpm_hosts;
636 						lphnext = lphnext->lpm_next;
637 					} else
638 						ok = 0;
639 				} else
640 					lph++;
641 			}
642 			FREE((caddr_t)lp, M_NQLEASE);
643 			if (olphnext)
644 				free((caddr_t)olphnext, M_NQMHOST);
645 			nfsstats.srvnqnfs_leases--;
646 		    }
647 		}
648 		lp = nextlp;
649 	}
650 }
651 
652 /*
653  * Called from nfssvc_nfsd() for a getlease rpc request.
654  * Do the from/to xdr translation and call nqsrv_getlease() to
655  * do the real work.
656  */
657 nqnfsrv_getlease(nfsd, mrep, md, dpos, cred, nam, mrq)
658 	struct nfsd *nfsd;
659 	struct mbuf *mrep, *md;
660 	caddr_t dpos;
661 	struct ucred *cred;
662 	struct mbuf *nam, **mrq;
663 {
664 	register struct nfsv2_fattr *fp;
665 	struct vattr va;
666 	register struct vattr *vap = &va;
667 	struct vnode *vp;
668 	nfsv2fh_t nfh;
669 	fhandle_t *fhp;
670 	register u_long *tl;
671 	register long t1;
672 	u_quad_t frev;
673 	caddr_t bpos;
674 	int error = 0;
675 	char *cp2;
676 	struct mbuf *mb, *mb2, *mreq;
677 	int flags, rdonly, cache;
678 
679 	fhp = &nfh.fh_generic;
680 	nfsm_srvmtofh(fhp);
681 	nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
682 	flags = fxdr_unsigned(int, *tl++);
683 	nfsd->nd_duration = fxdr_unsigned(int, *tl);
684 	if (error = nfsrv_fhtovp(fhp, TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
685 		nfsm_reply(0);
686 	if (rdonly && flags == NQL_WRITE) {
687 		error = EROFS;
688 		nfsm_reply(0);
689 	}
690 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
691 		nam, &cache, &frev, cred);
692 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
693 	vput(vp);
694 	nfsm_reply(NFSX_NQFATTR + 4*NFSX_UNSIGNED);
695 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
696 	*tl++ = txdr_unsigned(cache);
697 	*tl++ = txdr_unsigned(nfsd->nd_duration);
698 	txdr_hyper(&frev, tl);
699 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_NQFATTR);
700 	nfsm_srvfillattr;
701 	nfsm_srvdone;
702 }
703 
704 /*
705  * Called from nfssvc_nfsd() when a "vacated" message is received from a
706  * client. Find the entry and expire it.
707  */
708 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
709 	struct nfsd *nfsd;
710 	struct mbuf *mrep, *md;
711 	caddr_t dpos;
712 	struct ucred *cred;
713 	struct mbuf *nam, **mrq;
714 {
715 	register struct nqlease *lp;
716 	register struct nqhost *lph;
717 	struct nqlease *tlp = (struct nqlease *)0;
718 	struct vnode *vp;
719 	nfsv2fh_t nfh;
720 	fhandle_t *fhp;
721 	register u_long *tl;
722 	register long t1;
723 	struct nqm *lphnext;
724 	union nqsrvthead *lhp;
725 	u_quad_t frev;
726 	int error = 0, i, len, ok, rdonly, gotit = 0;
727 	char *cp2;
728 
729 	fhp = &nfh.fh_generic;
730 	nfsm_srvmtofh(fhp);
731 	m_freem(mrep);
732 	/*
733 	 * Find the lease by searching the hash list.
734 	 */
735 	for (lp = nqfhead[NQFHHASH(fhp->fh_fid.fid_data)]; lp;
736 	     lp = lp->lc_fhnext)
737 		if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
738 		    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
739 		    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
740 			  MAXFIDSZ)) {
741 			/* Found it */
742 			tlp = lp;
743 			break;
744 		}
745 	if (tlp) {
746 		lp = tlp;
747 		len = 1;
748 		i = 0;
749 		lph = &lp->lc_host;
750 		lphnext = lp->lc_morehosts;
751 		ok = 1;
752 		while (ok && (lph->lph_flag & LC_VALID)) {
753 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
754 				lph->lph_flag |= LC_VACATED;
755 				gotit++;
756 				break;
757 			}
758 			if (++i == len) {
759 				if (lphnext) {
760 					len = LC_MOREHOSTSIZ;
761 					i = 0;
762 					lph = lphnext->lpm_hosts;
763 					lphnext = lphnext->lpm_next;
764 				} else
765 					ok = 0;
766 			} else
767 				lph++;
768 		}
769 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
770 			lp->lc_flag &= ~LC_EXPIREDWANTED;
771 			wakeup((caddr_t)&lp->lc_flag);
772 		}
773 nfsmout:
774 		return (EPERM);
775 	}
776 	return (EPERM);
777 }
778 
779 /*
780  * Client get lease rpc function.
781  */
782 nqnfs_getlease(vp, rwflag, cred, p)
783 	register struct vnode *vp;
784 	int rwflag;
785 	struct ucred *cred;
786 	struct proc *p;
787 {
788 	register u_long *tl;
789 	register caddr_t cp;
790 	register long t1;
791 	register struct nfsnode *np, *tp;
792 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
793 	caddr_t bpos, dpos, cp2;
794 	time_t reqtime;
795 	int error = 0;
796 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
797 	int cachable;
798 	u_quad_t frev;
799 
800 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
801 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
802 		 &bpos);
803 	nfsm_fhtom(vp);
804 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
805 	*tl++ = txdr_unsigned(rwflag);
806 	*tl = txdr_unsigned(nmp->nm_leaseterm);
807 	reqtime = time.tv_sec;
808 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
809 	np = VTONFS(vp);
810 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
811 	cachable = fxdr_unsigned(int, *tl++);
812 	reqtime += fxdr_unsigned(int, *tl++);
813 	if (reqtime > time.tv_sec) {
814 		fxdr_hyper(tl, &frev);
815 		nqnfs_clientlease(nmp, np, rwflag, cachable, reqtime, frev);
816 		nfsm_loadattr(vp, (struct vattr *)0);
817 	} else
818 		error = NQNFS_EXPIRED;
819 	nfsm_reqdone;
820 	return (error);
821 }
822 
823 /*
824  * Client vacated message function.
825  */
826 nqnfs_vacated(vp, cred)
827 	register struct vnode *vp;
828 	struct ucred *cred;
829 {
830 	register caddr_t cp;
831 	register struct mbuf *m;
832 	register int i;
833 	caddr_t bpos;
834 	u_long xid;
835 	int error = 0;
836 	struct mbuf *mreq, *mb, *mb2, *mheadend;
837 	struct nfsmount *nmp;
838 	struct nfsreq myrep;
839 
840 	nmp = VFSTONFS(vp->v_mount);
841 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
842 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
843 	nfsm_fhtom(vp);
844 	m = mreq;
845 	i = 0;
846 	while (m) {
847 		i += m->m_len;
848 		m = m->m_next;
849 	}
850 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
851 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
852 		mreq, i, &mheadend, &xid);
853 	if (nmp->nm_sotype == SOCK_STREAM) {
854 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
855 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
856 			NFSX_UNSIGNED));
857 	}
858 	myrep.r_flags = 0;
859 	myrep.r_nmp = nmp;
860 	if (nmp->nm_soflags & PR_CONNREQUIRED)
861 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
862 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
863 	if (nmp->nm_soflags & PR_CONNREQUIRED)
864 		nfs_sndunlock(&nmp->nm_flag);
865 	return (error);
866 }
867 
868 /*
869  * Called for client side callbacks
870  */
871 nqnfs_callback(nmp, mrep, md, dpos)
872 	struct nfsmount *nmp;
873 	struct mbuf *mrep, *md;
874 	caddr_t dpos;
875 {
876 	register struct vnode *vp;
877 	register u_long *tl;
878 	register long t1;
879 	nfsv2fh_t nfh;
880 	fhandle_t *fhp;
881 	struct nfsnode *np;
882 	struct nfsd nd;
883 	int error;
884 	char *cp2;
885 
886 	nd.nd_mrep = mrep;
887 	nd.nd_md = md;
888 	nd.nd_dpos = dpos;
889 	if (error = nfs_getreq(&nd, FALSE))
890 		return (error);
891 	md = nd.nd_md;
892 	dpos = nd.nd_dpos;
893 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
894 		m_freem(mrep);
895 		return (EPERM);
896 	}
897 	fhp = &nfh.fh_generic;
898 	nfsm_srvmtofh(fhp);
899 	m_freem(mrep);
900 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
901 		return (error);
902 	vp = NFSTOV(np);
903 	if (np->n_tnext) {
904 		np->n_expiry = 0;
905 		np->n_flag |= NQNFSEVICTED;
906 		if (np->n_tprev != (struct nfsnode *)nmp) {
907 			if (np->n_tnext == (struct nfsnode *)nmp)
908 				nmp->nm_tprev = np->n_tprev;
909 			else
910 				np->n_tnext->n_tprev = np->n_tprev;
911 			np->n_tprev->n_tnext = np->n_tnext;
912 			np->n_tnext = nmp->nm_tnext;
913 			nmp->nm_tnext = np;
914 			np->n_tprev = (struct nfsnode *)nmp;
915 			if (np->n_tnext == (struct nfsnode *)nmp)
916 				nmp->nm_tprev = np;
917 			else
918 				np->n_tnext->n_tprev = np;
919 		}
920 	}
921 	vrele(vp);
922 	nfsm_srvdone;
923 }
924 
925 /*
926  * Nqnfs client helper daemon. Runs once a second to expire leases.
927  * It also get authorization strings for "kerb" mounts.
928  * It must start at the beginning of the list again after any potential
929  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
930  * the list asynchronously.
931  */
932 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
933 	register struct nfsmount *nmp;
934 	struct ucred *cred;
935 	struct nfsd_cargs *ncd;
936 	int flag;
937 	caddr_t argp;
938 	struct proc *p;
939 {
940 	register struct nfsnode *np;
941 	struct vnode *vp;
942 	struct nfsreq myrep;
943 	int error, vpid;
944 
945 	/*
946 	 * First initialize some variables
947 	 */
948 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
949 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
950 
951 	/*
952 	 * If an authorization string is being passed in, get it.
953 	 */
954 	if ((flag & NFSSVC_GOTAUTH) &&
955 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
956 		if (nmp->nm_flag & NFSMNT_HASAUTH)
957 			panic("cld kerb");
958 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
959 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
960 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
961 				ncd->ncd_authlen) == 0) {
962 				nmp->nm_authtype = ncd->ncd_authtype;
963 				nmp->nm_authlen = ncd->ncd_authlen;
964 			} else
965 				nmp->nm_flag |= NFSMNT_AUTHERR;
966 		} else
967 			nmp->nm_flag |= NFSMNT_AUTHERR;
968 		nmp->nm_flag |= NFSMNT_HASAUTH;
969 		wakeup((caddr_t)&nmp->nm_authlen);
970 	} else
971 		nmp->nm_flag |= NFSMNT_WAITAUTH;
972 
973 	/*
974 	 * Loop every second updating queue until there is a termination sig.
975 	 */
976 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
977 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
978 		/*
979 		 * If there are no outstanding requests (and therefore no
980 		 * processes in nfs_reply) and there is data in the receive
981 		 * queue, poke for callbacks.
982 		 */
983 		if (nfsreqh.r_next == &nfsreqh && nmp->nm_so &&
984 		    nmp->nm_so->so_rcv.sb_cc > 0) {
985 		    myrep.r_flags = R_GETONEREP;
986 		    myrep.r_nmp = nmp;
987 		    myrep.r_mrep = (struct mbuf *)0;
988 		    myrep.r_procp = (struct proc *)0;
989 		    (void) nfs_reply(&myrep);
990 		}
991 
992 		/*
993 		 * Loop through the leases, updating as required.
994 		 */
995 		np = nmp->nm_tnext;
996 		while (np != (struct nfsnode *)nmp &&
997 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
998 			vp = NFSTOV(np);
999 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
1000 			vpid = vp->v_id;
1001 			if (np->n_expiry < time.tv_sec) {
1002 			   if (vget(vp) == 0) {
1003 			     nmp->nm_inprog = vp;
1004 			     if (vpid == vp->v_id) {
1005 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
1006 				if (np->n_tnext == (struct nfsnode *)nmp)
1007 					nmp->nm_tprev = np->n_tprev;
1008 				else
1009 					np->n_tnext->n_tprev = np->n_tprev;
1010 				if (np->n_tprev == (struct nfsnode *)nmp)
1011 					nmp->nm_tnext = np->n_tnext;
1012 				else
1013 					np->n_tprev->n_tnext = np->n_tnext;
1014 				np->n_tnext = (struct nfsnode *)0;
1015 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
1016 				    && vp->v_type == VREG) {
1017 					if (np->n_flag & NQNFSEVICTED) {
1018 						(void) nfs_vinvalbuf(vp,
1019 						       V_SAVE, cred, p, 0);
1020 						np->n_flag &= ~NQNFSEVICTED;
1021 						(void) nqnfs_vacated(vp, cred);
1022 					} else {
1023 						(void) VOP_FSYNC(vp, cred,
1024 						    MNT_WAIT, p);
1025 						np->n_flag &= ~NMODIFIED;
1026 					}
1027 				}
1028 			      }
1029 			      vrele(vp);
1030 			      nmp->nm_inprog = NULLVP;
1031 			    }
1032 			    if (np != nmp->nm_tnext)
1033 				np = nmp->nm_tnext;
1034 			    else
1035 				break;
1036 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
1037 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1038 				 == NQNFSWRITE && vp->v_dirtyblkhd.le_next &&
1039 				 vget(vp) == 0) {
1040 				 nmp->nm_inprog = vp;
1041 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1042 				 if (vpid == vp->v_id &&
1043 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1044 					np->n_brev = np->n_lrev;
1045 				 vrele(vp);
1046 				 nmp->nm_inprog = NULLVP;
1047 			    }
1048 			    if (np != nmp->nm_tnext)
1049 				np = nmp->nm_tnext;
1050 			    else
1051 				break;
1052 			} else
1053 				break;
1054 		}
1055 	    }
1056 
1057 	    /*
1058 	     * Get an authorization string, if required.
1059 	     */
1060 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1061 		ncd->ncd_authuid = nmp->nm_authuid;
1062 		if (copyout((caddr_t)ncd, argp, sizeof (struct nfsd_cargs)))
1063 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1064 		else
1065 			return (ENEEDAUTH);
1066 	    }
1067 
1068 	    /*
1069 	     * Wait a bit (no pun) and do it again.
1070 	     */
1071 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1072 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1073 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1074 			"nqnfstimr", hz / 3);
1075 		    if (error == EINTR || error == ERESTART)
1076 			(void) dounmount(nmp->nm_mountp, 0, p);
1077 	    }
1078 	}
1079 	free((caddr_t)nmp, M_NFSMNT);
1080 	if (error == EWOULDBLOCK)
1081 		error = 0;
1082 	return (error);
1083 }
1084 
1085 /*
1086  * Adjust all timer queue expiry times when the time of day clock is changed.
1087  * Called from the settimeofday() syscall.
1088  */
1089 void
1090 lease_updatetime(deltat)
1091 	register int deltat;
1092 {
1093 	register struct nqlease *lp;
1094 	register struct nfsnode *np;
1095 	struct mount *mp;
1096 	struct nfsmount *nmp;
1097 	int s;
1098 
1099 	if (nqnfsstarttime != 0)
1100 		nqnfsstarttime += deltat;
1101 	s = splsoftclock();
1102 	lp = nqthead.th_chain[0];
1103 	while (lp != (struct nqlease *)&nqthead) {
1104 		lp->lc_expiry += deltat;
1105 		lp = lp->lc_chain1[0];
1106 	}
1107 	splx(s);
1108 
1109 	/*
1110 	 * Search the mount list for all nqnfs mounts and do their timer
1111 	 * queues.
1112 	 */
1113 	mp = rootfs;
1114 	do {
1115 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1116 			nmp = VFSTONFS(mp);
1117 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1118 				np = nmp->nm_tnext;
1119 				while (np != (struct nfsnode *)nmp) {
1120 					np->n_expiry += deltat;
1121 					np = np->n_tnext;
1122 				}
1123 			}
1124 		}
1125 		mp = mp->mnt_next;
1126 	} while (mp != rootfs);
1127 }
1128 
1129 /*
1130  * Lock a server lease.
1131  */
1132 void
1133 nqsrv_locklease(lp)
1134 	struct nqlease *lp;
1135 {
1136 
1137 	while (lp->lc_flag & LC_LOCKED) {
1138 		lp->lc_flag |= LC_WANTED;
1139 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1140 	}
1141 	lp->lc_flag |= LC_LOCKED;
1142 	lp->lc_flag &= ~LC_WANTED;
1143 }
1144 
1145 /*
1146  * Unlock a server lease.
1147  */
1148 void
1149 nqsrv_unlocklease(lp)
1150 	struct nqlease *lp;
1151 {
1152 
1153 	lp->lc_flag &= ~LC_LOCKED;
1154 	if (lp->lc_flag & LC_WANTED)
1155 		wakeup((caddr_t)lp);
1156 }
1157 
1158 /*
1159  * Update a client lease.
1160  */
1161 void
1162 nqnfs_clientlease(nmp, np, rwflag, cachable, expiry, frev)
1163 	register struct nfsmount *nmp;
1164 	register struct nfsnode *np;
1165 	int rwflag, cachable;
1166 	time_t expiry;
1167 	u_quad_t frev;
1168 {
1169 	register struct nfsnode *tp;
1170 
1171 	if (np->n_tnext) {
1172 		if (np->n_tnext == (struct nfsnode *)nmp)
1173 			nmp->nm_tprev = np->n_tprev;
1174 		else
1175 			np->n_tnext->n_tprev = np->n_tprev;
1176 		if (np->n_tprev == (struct nfsnode *)nmp)
1177 			nmp->nm_tnext = np->n_tnext;
1178 		else
1179 			np->n_tprev->n_tnext = np->n_tnext;
1180 		if (rwflag == NQL_WRITE)
1181 			np->n_flag |= NQNFSWRITE;
1182 	} else if (rwflag == NQL_READ)
1183 		np->n_flag &= ~NQNFSWRITE;
1184 	else
1185 		np->n_flag |= NQNFSWRITE;
1186 	if (cachable)
1187 		np->n_flag &= ~NQNFSNONCACHE;
1188 	else
1189 		np->n_flag |= NQNFSNONCACHE;
1190 	np->n_expiry = expiry;
1191 	np->n_lrev = frev;
1192 	tp = nmp->nm_tprev;
1193 	while (tp != (struct nfsnode *)nmp && tp->n_expiry > np->n_expiry)
1194 		tp = tp->n_tprev;
1195 	if (tp == (struct nfsnode *)nmp) {
1196 		np->n_tnext = nmp->nm_tnext;
1197 		nmp->nm_tnext = np;
1198 	} else {
1199 		np->n_tnext = tp->n_tnext;
1200 		tp->n_tnext = np;
1201 	}
1202 	np->n_tprev = tp;
1203 	if (np->n_tnext == (struct nfsnode *)nmp)
1204 		nmp->nm_tprev = np;
1205 	else
1206 		np->n_tnext->n_tprev = np;
1207 }
1208