xref: /csrg-svn/sys/nfs/nfs_nqlease.c (revision 56660)
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	7.15 (Berkeley) 11/01/92
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 			VOP_UNLOCK(vp);
232 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
233 			nqsrv_waitfor_expiry(lp);
234 			VOP_LOCK(vp);
235 			nqsrv_unlocklease(lp);
236 		}
237 doreply:
238 		/*
239 		 * Update the lease and return
240 		 */
241 		if ((flags & NQL_CHECK) == 0)
242 			nqsrv_instimeq(lp, *duration);
243 		if (lp->lc_flag & LC_NONCACHABLE)
244 			*cachablep = 0;
245 		else {
246 			*cachablep = 1;
247 			if (flags & NQL_WRITE)
248 				lp->lc_flag |= LC_WRITTEN;
249 		}
250 		splx(s);
251 		return (0);
252 	}
253 	splx(s);
254 	if (flags & NQL_CHECK)
255 		return (0);
256 
257 	/*
258 	 * Allocate new lease
259 	 * The value of nqsrv_maxnumlease should be set generously, so that
260 	 * the following "printf" happens infrequently.
261 	 */
262 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
263 		printf("Nqnfs server, too many leases\n");
264 		do {
265 			(void) tsleep((caddr_t)&lbolt, PSOCK,
266 					"nqsrvnuml", 0);
267 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
268 	}
269 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
270 	bzero((caddr_t)lp, sizeof (struct nqlease));
271 	if (flags & NQL_WRITE)
272 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
273 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nam);
274 	lp->lc_vp = vp;
275 	lp->lc_fsid = fh.fh_fsid;
276 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
277 	if (lq = *lpp)
278 		lq->lc_fhprev = &lp->lc_fhnext;
279 	lp->lc_fhnext = lq;
280 	lp->lc_fhprev = lpp;
281 	*lpp = lp;
282 	vp->v_lease = lp;
283 	s = splsoftclock();
284 	nqsrv_instimeq(lp, *duration);
285 	splx(s);
286 	*cachablep = 1;
287 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
288 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
289 	return (0);
290 }
291 
292 /*
293  * Local lease check for server syscalls.
294  * Just set up args and let nqsrv_getlease() do the rest.
295  */
296 void
297 lease_check(vp, p, cred, flag)
298 	struct vnode *vp;
299 	struct proc *p;
300 	struct ucred *cred;
301 	int flag;
302 {
303 	int duration = 0, cache;
304 	struct nfsd nfsd;
305 	u_quad_t frev;
306 
307 	nfsd.nd_slp = NQLOCALSLP;
308 	nfsd.nd_procp = p;
309 	(void) nqsrv_getlease(vp, &duration, NQL_CHECK | flag, &nfsd,
310 		(struct mbuf *)0, &cache, &frev, cred);
311 }
312 
313 /*
314  * Add a host to an nqhost structure for a lease.
315  */
316 void
317 nqsrv_addhost(lph, slp, nam)
318 	register struct nqhost *lph;
319 	struct nfssvc_sock *slp;
320 	struct mbuf *nam;
321 {
322 	register struct sockaddr_in *saddr;
323 
324 	if (slp == NQLOCALSLP)
325 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
326 	else if (slp == nfs_udpsock) {
327 		saddr = mtod(nam, struct sockaddr_in *);
328 		lph->lph_flag |= (LC_VALID | LC_UDP);
329 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
330 		lph->lph_port = saddr->sin_port;
331 	} else if (slp == nfs_cltpsock) {
332 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
333 		lph->lph_flag |= (LC_VALID | LC_CLTP);
334 	} else {
335 		lph->lph_flag |= (LC_VALID | LC_SREF);
336 		lph->lph_slp = slp;
337 		slp->ns_sref++;
338 	}
339 }
340 
341 /*
342  * Update the lease expiry time and position it in the timer queue correctly.
343  */
344 void
345 nqsrv_instimeq(lp, duration)
346 	register struct nqlease *lp;
347 	u_long duration;
348 {
349 	register struct nqlease *tlp;
350 	time_t newexpiry;
351 
352 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
353 	if (lp->lc_expiry == newexpiry)
354 		return;
355 	if (lp->lc_chain1[0])
356 		remque(lp);
357 	lp->lc_expiry = newexpiry;
358 
359 	/*
360 	 * Find where in the queue it should be.
361 	 */
362 	tlp = nqthead.th_chain[1];
363 	while (tlp->lc_expiry > newexpiry && tlp != (struct nqlease *)&nqthead)
364 		tlp = tlp->lc_chain1[1];
365 	if (tlp == nqthead.th_chain[1])
366 		NQSTORENOVRAM(newexpiry);
367 	insque(lp, tlp);
368 }
369 
370 /*
371  * Compare the requesting host address with the lph entry in the lease.
372  * Return true iff it is the same.
373  * This is somewhat messy due to the union in the nqhost structure.
374  * The local host is indicated by the special value of NQLOCALSLP for slp.
375  */
376 nqsrv_cmpnam(slp, nam, lph)
377 	register struct nfssvc_sock *slp;
378 	struct mbuf *nam;
379 	register struct nqhost *lph;
380 {
381 	register struct sockaddr_in *saddr;
382 	struct mbuf *addr;
383 	union nethostaddr lhaddr;
384 	int ret;
385 
386 	if (slp == NQLOCALSLP) {
387 		if (lph->lph_flag & LC_LOCAL)
388 			return (1);
389 		else
390 			return (0);
391 	}
392 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
393 		addr = nam;
394 	else
395 		addr = slp->ns_nam;
396 	if (lph->lph_flag & LC_UDP)
397 		ret = netaddr_match(AF_INET, &lph->lph_haddr, addr);
398 	else if (lph->lph_flag & LC_CLTP)
399 		ret = netaddr_match(AF_ISO, &lph->lph_claddr, addr);
400 	else {
401 		if ((lph->lph_slp->ns_flag & SLP_VALID) == 0)
402 			return (0);
403 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
404 		if (saddr->sin_family == AF_INET)
405 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
406 		else
407 			lhaddr.had_nam = lph->lph_slp->ns_nam;
408 		ret = netaddr_match(saddr->sin_family, &lhaddr, addr);
409 	}
410 	return (ret);
411 }
412 
413 /*
414  * Send out eviction notice messages to all other hosts for the lease.
415  */
416 void
417 nqsrv_send_eviction(vp, lp, slp, nam, cred)
418 	struct vnode *vp;
419 	register struct nqlease *lp;
420 	struct nfssvc_sock *slp;
421 	struct mbuf *nam;
422 	struct ucred *cred;
423 {
424 	register struct nqhost *lph = &lp->lc_host;
425 	register struct mbuf *m;
426 	register int siz;
427 	struct nqm *lphnext = lp->lc_morehosts;
428 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
429 	struct socket *so;
430 	struct sockaddr_in *saddr;
431 	fhandle_t *fhp;
432 	caddr_t bpos, cp;
433 	u_long xid;
434 	int len = 1, ok = 1, i = 0;
435 	int sotype, *solockp;
436 
437 	while (ok && (lph->lph_flag & LC_VALID)) {
438 		if (nqsrv_cmpnam(slp, nam, lph))
439 			lph->lph_flag |= LC_VACATED;
440 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
441 			if (lph->lph_flag & LC_UDP) {
442 				MGET(nam2, M_WAIT, MT_SONAME);
443 				saddr = mtod(nam2, struct sockaddr_in *);
444 				nam2->m_len = saddr->sin_len =
445 					sizeof (struct sockaddr_in);
446 				saddr->sin_family = AF_INET;
447 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
448 				saddr->sin_port = lph->lph_port;
449 				so = nfs_udpsock->ns_so;
450 			} else if (lph->lph_flag & LC_CLTP) {
451 				nam2 = lph->lph_nam;
452 				so = nfs_cltpsock->ns_so;
453 			} else if (lph->lph_slp->ns_flag & SLP_VALID) {
454 				nam2 = (struct mbuf *)0;
455 				so = lph->lph_slp->ns_so;
456 			} else
457 				goto nextone;
458 			sotype = so->so_type;
459 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
460 				solockp = &lph->lph_slp->ns_solock;
461 			else
462 				solockp = (int *)0;
463 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
464 				NFSX_FH);
465 			nfsm_build(cp, caddr_t, NFSX_FH);
466 			bzero(cp, NFSX_FH);
467 			fhp = (fhandle_t *)cp;
468 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
469 			VFS_VPTOFH(vp, &fhp->fh_fid);
470 			m = mreq;
471 			siz = 0;
472 			while (m) {
473 				siz += m->m_len;
474 				m = m->m_next;
475 			}
476 			if (siz <= 0 || siz > NFS_MAXPACKET) {
477 				printf("mbuf siz=%d\n",siz);
478 				panic("Bad nfs svc reply");
479 			}
480 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
481 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
482 				mreq, siz, &mheadend, &xid);
483 			/*
484 			 * For stream protocols, prepend a Sun RPC
485 			 * Record Mark.
486 			 */
487 			if (sotype == SOCK_STREAM) {
488 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
489 				*mtod(m, u_long *) = htonl(0x80000000 |
490 					(m->m_pkthdr.len - NFSX_UNSIGNED));
491 			}
492 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
493 			    (lph->lph_slp->ns_flag & SLP_VALID) == 0) ||
494 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
495 				m_freem(m);
496 			else {
497 				if (solockp)
498 					*solockp |= NFSMNT_SNDLOCK;
499 				(void) nfs_send(so, nam2, m,
500 						(struct nfsreq *)0);
501 				if (solockp)
502 					nfs_sndunlock(solockp);
503 			}
504 			if (lph->lph_flag & LC_UDP)
505 				MFREE(nam2, m);
506 		}
507 nextone:
508 		if (++i == len) {
509 			if (lphnext) {
510 				i = 0;
511 				len = LC_MOREHOSTSIZ;
512 				lph = lphnext->lpm_hosts;
513 				lphnext = lphnext->lpm_next;
514 			} else
515 				ok = 0;
516 		} else
517 			lph++;
518 	}
519 }
520 
521 /*
522  * Wait for the lease to expire.
523  * This will occur when all clients have sent "vacated" messages to
524  * this server OR when it expires do to timeout.
525  */
526 void
527 nqsrv_waitfor_expiry(lp)
528 	register struct nqlease *lp;
529 {
530 	register struct nqhost *lph;
531 	register int i;
532 	struct nqm *lphnext;
533 	int len, ok;
534 
535 tryagain:
536 	if (time.tv_sec > lp->lc_expiry)
537 		return;
538 	lph = &lp->lc_host;
539 	lphnext = lp->lc_morehosts;
540 	len = 1;
541 	i = 0;
542 	ok = 1;
543 	while (ok && (lph->lph_flag & LC_VALID)) {
544 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
545 			lp->lc_flag |= LC_EXPIREDWANTED;
546 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
547 					"nqexp", 0);
548 			goto tryagain;
549 		}
550 		if (++i == len) {
551 			if (lphnext) {
552 				i = 0;
553 				len = LC_MOREHOSTSIZ;
554 				lph = lphnext->lpm_hosts;
555 				lphnext = lphnext->lpm_next;
556 			} else
557 				ok = 0;
558 		} else
559 			lph++;
560 	}
561 }
562 
563 /*
564  * Nqnfs server timer that maintains the server lease queue.
565  * Scan the lease queue for expired entries:
566  * - when one is found, wakeup anyone waiting for it
567  *   else dequeue and free
568  */
569 void
570 nqnfs_serverd()
571 {
572 	register struct nqlease *lp, *lq;
573 	register struct nqhost *lph;
574 	struct nqlease *nextlp;
575 	struct nqm *lphnext, *olphnext;
576 	struct mbuf *n;
577 	union nqsrvthead *lhp;
578 	int i, len, ok;
579 
580 	lp = nqthead.th_chain[0];
581 	while (lp != (struct nqlease *)&nqthead) {
582 		if (lp->lc_expiry >= time.tv_sec)
583 			break;
584 		nextlp = lp->lc_chain1[0];
585 		if (lp->lc_flag & LC_EXPIREDWANTED) {
586 			lp->lc_flag &= ~LC_EXPIREDWANTED;
587 			wakeup((caddr_t)&lp->lc_flag);
588 		} else if ((lp->lc_flag & (LC_LOCKED | LC_WANTED)) == 0) {
589 		    /*
590 		     * Make a best effort at keeping a write caching lease long
591 		     * enough by not deleting it until it has been explicitly
592 		     * vacated or there have been no writes in the previous
593 		     * write_slack seconds since expiry and the nfsds are not
594 		     * all busy. The assumption is that if the nfsds are not
595 		     * all busy now (no queue of nfs requests), then the client
596 		     * would have been able to do at least one write to the
597 		     * file during the last write_slack seconds if it was still
598 		     * trying to push writes to the server.
599 		     */
600 		    if ((lp->lc_flag & (LC_WRITE | LC_VACATED)) == LC_WRITE &&
601 			((lp->lc_flag & LC_WRITTEN) || nfsd_waiting == 0)) {
602 			lp->lc_flag &= ~LC_WRITTEN;
603 			nqsrv_instimeq(lp, nqsrv_writeslack);
604 		    } else {
605 			remque(lp);
606 			if (lq = lp->lc_fhnext)
607 				lq->lc_fhprev = lp->lc_fhprev;
608 			*lp->lc_fhprev = lq;
609 			/*
610 			 * This soft reference may no longer be valid, but
611 			 * no harm done. The worst case is if the vnode was
612 			 * recycled and has another valid lease reference,
613 			 * which is dereferenced prematurely.
614 			 */
615 			lp->lc_vp->v_lease = (struct nqlease *)0;
616 			lph = &lp->lc_host;
617 			lphnext = lp->lc_morehosts;
618 			olphnext = (struct nqm *)0;
619 			len = 1;
620 			i = 0;
621 			ok = 1;
622 			while (ok && (lph->lph_flag & LC_VALID)) {
623 				if (lph->lph_flag & LC_CLTP)
624 					MFREE(lph->lph_nam, n);
625 				if (lph->lph_flag & LC_SREF)
626 					nfsrv_slpderef(lph->lph_slp);
627 				if (++i == len) {
628 					if (olphnext) {
629 						free((caddr_t)olphnext, M_NQMHOST);
630 						olphnext = (struct nqm *)0;
631 					}
632 					if (lphnext) {
633 						olphnext = lphnext;
634 						i = 0;
635 						len = LC_MOREHOSTSIZ;
636 						lph = lphnext->lpm_hosts;
637 						lphnext = lphnext->lpm_next;
638 					} else
639 						ok = 0;
640 				} else
641 					lph++;
642 			}
643 			FREE((caddr_t)lp, M_NQLEASE);
644 			if (olphnext)
645 				free((caddr_t)olphnext, M_NQMHOST);
646 			nfsstats.srvnqnfs_leases--;
647 		    }
648 		}
649 		lp = nextlp;
650 	}
651 }
652 
653 /*
654  * Called from nfssvc_nfsd() for a getlease rpc request.
655  * Do the from/to xdr translation and call nqsrv_getlease() to
656  * do the real work.
657  */
658 nqnfsrv_getlease(nfsd, mrep, md, dpos, cred, nam, mrq)
659 	struct nfsd *nfsd;
660 	struct mbuf *mrep, *md;
661 	caddr_t dpos;
662 	struct ucred *cred;
663 	struct mbuf *nam, **mrq;
664 {
665 	register struct nfsv2_fattr *fp;
666 	struct vattr va;
667 	register struct vattr *vap = &va;
668 	struct vnode *vp;
669 	nfsv2fh_t nfh;
670 	fhandle_t *fhp;
671 	register u_long *tl;
672 	register long t1;
673 	u_quad_t frev;
674 	caddr_t bpos;
675 	int error = 0;
676 	char *cp2;
677 	struct mbuf *mb, *mb2, *mreq;
678 	int flags, rdonly, cache;
679 
680 	fhp = &nfh.fh_generic;
681 	nfsm_srvmtofh(fhp);
682 	nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
683 	flags = fxdr_unsigned(int, *tl++);
684 	nfsd->nd_duration = fxdr_unsigned(int, *tl);
685 	if (error = nfsrv_fhtovp(fhp, TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
686 		nfsm_reply(0);
687 	if (rdonly && flags == NQL_WRITE) {
688 		error = EROFS;
689 		nfsm_reply(0);
690 	}
691 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
692 		nam, &cache, &frev, cred);
693 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
694 	vput(vp);
695 	nfsm_reply(NFSX_NQFATTR + 4*NFSX_UNSIGNED);
696 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
697 	*tl++ = txdr_unsigned(cache);
698 	*tl++ = txdr_unsigned(nfsd->nd_duration);
699 	txdr_hyper(&frev, tl);
700 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_NQFATTR);
701 	nfsm_srvfillattr;
702 	nfsm_srvdone;
703 }
704 
705 /*
706  * Called from nfssvc_nfsd() when a "vacated" message is received from a
707  * client. Find the entry and expire it.
708  */
709 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
710 	struct nfsd *nfsd;
711 	struct mbuf *mrep, *md;
712 	caddr_t dpos;
713 	struct ucred *cred;
714 	struct mbuf *nam, **mrq;
715 {
716 	register struct nqlease *lp;
717 	register struct nqhost *lph;
718 	struct nqlease *tlp = (struct nqlease *)0;
719 	struct vnode *vp;
720 	nfsv2fh_t nfh;
721 	fhandle_t *fhp;
722 	register u_long *tl;
723 	register long t1;
724 	struct nqm *lphnext;
725 	union nqsrvthead *lhp;
726 	u_quad_t frev;
727 	int error = 0, i, len, ok, rdonly, gotit = 0;
728 	char *cp2;
729 
730 	fhp = &nfh.fh_generic;
731 	nfsm_srvmtofh(fhp);
732 	if (error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
733 		return (error);
734 	m_freem(mrep);
735 	tlp = vp->v_lease;
736 	if (tlp == (struct nqlease *)0) {
737 		/*
738 		 * Find the lease by searching the hash list.
739 		 */
740 		for (lp = nqfhead[NQFHHASH(fhp->fh_fid.fid_data)]; lp;
741 		     lp = lp->lc_fhnext)
742 			if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
743 			    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
744 			    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
745 				  MAXFIDSZ)) {
746 				/* Found it */
747 				lp->lc_vp = vp;
748 				vp->v_lease = lp;
749 				tlp = lp;
750 				break;
751 			}
752 	}
753 	vrele(vp);
754 	if (tlp) {
755 		lp = tlp;
756 		len = 1;
757 		i = 0;
758 		lph = &lp->lc_host;
759 		lphnext = lp->lc_morehosts;
760 		ok = 1;
761 		while (ok && (lph->lph_flag & LC_VALID)) {
762 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
763 				lph->lph_flag |= LC_VACATED;
764 				gotit++;
765 				break;
766 			}
767 			if (++i == len) {
768 				if (lphnext) {
769 					len = LC_MOREHOSTSIZ;
770 					i = 0;
771 					lph = lphnext->lpm_hosts;
772 					lphnext = lphnext->lpm_next;
773 				} else
774 					ok = 0;
775 			} else
776 				lph++;
777 		}
778 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
779 			lp->lc_flag &= ~LC_EXPIREDWANTED;
780 			wakeup((caddr_t)&lp->lc_flag);
781 		}
782 nfsmout:
783 		return (EPERM);
784 	}
785 	return (EPERM);
786 }
787 
788 /*
789  * Client get lease rpc function.
790  */
791 nqnfs_getlease(vp, rwflag, cred, p)
792 	register struct vnode *vp;
793 	int rwflag;
794 	struct ucred *cred;
795 	struct proc *p;
796 {
797 	register u_long *tl;
798 	register caddr_t cp;
799 	register long t1;
800 	register struct nfsnode *np, *tp;
801 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
802 	caddr_t bpos, dpos, cp2;
803 	time_t reqtime;
804 	int error = 0;
805 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
806 	int cachable;
807 	u_quad_t frev;
808 
809 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
810 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
811 		 &bpos);
812 	nfsm_fhtom(vp);
813 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
814 	*tl++ = txdr_unsigned(rwflag);
815 	*tl = txdr_unsigned(nmp->nm_leaseterm);
816 	reqtime = time.tv_sec;
817 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
818 	np = VTONFS(vp);
819 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
820 	cachable = fxdr_unsigned(int, *tl++);
821 	reqtime += fxdr_unsigned(int, *tl++);
822 	if (reqtime > time.tv_sec) {
823 		fxdr_hyper(tl, &frev);
824 		nqnfs_clientlease(nmp, np, rwflag, cachable, reqtime, frev);
825 		nfsm_loadattr(vp, (struct vattr *)0);
826 	} else
827 		error = NQNFS_EXPIRED;
828 	nfsm_reqdone;
829 	return (error);
830 }
831 
832 /*
833  * Client vacated message function.
834  */
835 nqnfs_vacated(vp, cred)
836 	register struct vnode *vp;
837 	struct ucred *cred;
838 {
839 	register caddr_t cp;
840 	register struct mbuf *m;
841 	register int i;
842 	caddr_t bpos;
843 	u_long xid;
844 	int error = 0;
845 	struct mbuf *mreq, *mb, *mb2, *mheadend;
846 	struct nfsmount *nmp;
847 	struct nfsreq myrep;
848 
849 	nmp = VFSTONFS(vp->v_mount);
850 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
851 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
852 	nfsm_fhtom(vp);
853 	m = mreq;
854 	i = 0;
855 	while (m) {
856 		i += m->m_len;
857 		m = m->m_next;
858 	}
859 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
860 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
861 		mreq, i, &mheadend, &xid);
862 	if (nmp->nm_sotype == SOCK_STREAM) {
863 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
864 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
865 			NFSX_UNSIGNED));
866 	}
867 	myrep.r_flags = 0;
868 	myrep.r_nmp = nmp;
869 	if (nmp->nm_soflags & PR_CONNREQUIRED)
870 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
871 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
872 	if (nmp->nm_soflags & PR_CONNREQUIRED)
873 		nfs_sndunlock(&nmp->nm_flag);
874 	return (error);
875 }
876 
877 /*
878  * Called for client side callbacks
879  */
880 nqnfs_callback(nmp, mrep, md, dpos)
881 	struct nfsmount *nmp;
882 	struct mbuf *mrep, *md;
883 	caddr_t dpos;
884 {
885 	register struct vnode *vp;
886 	register u_long *tl;
887 	register long t1;
888 	nfsv2fh_t nfh;
889 	fhandle_t *fhp;
890 	struct nfsnode *np;
891 	struct nfsd nd;
892 	int error;
893 	char *cp2;
894 
895 	nd.nd_mrep = mrep;
896 	nd.nd_md = md;
897 	nd.nd_dpos = dpos;
898 	if (error = nfs_getreq(&nd, FALSE))
899 		return (error);
900 	md = nd.nd_md;
901 	dpos = nd.nd_dpos;
902 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
903 		m_freem(mrep);
904 		return (EPERM);
905 	}
906 	fhp = &nfh.fh_generic;
907 	nfsm_srvmtofh(fhp);
908 	m_freem(mrep);
909 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
910 		return (error);
911 	vp = NFSTOV(np);
912 	if (np->n_tnext) {
913 		np->n_expiry = 0;
914 		np->n_flag |= NQNFSEVICTED;
915 		if (np->n_tprev != (struct nfsnode *)nmp) {
916 			if (np->n_tnext == (struct nfsnode *)nmp)
917 				nmp->nm_tprev = np->n_tprev;
918 			else
919 				np->n_tnext->n_tprev = np->n_tprev;
920 			np->n_tprev->n_tnext = np->n_tnext;
921 			np->n_tnext = nmp->nm_tnext;
922 			nmp->nm_tnext = np;
923 			np->n_tprev = (struct nfsnode *)nmp;
924 			if (np->n_tnext == (struct nfsnode *)nmp)
925 				nmp->nm_tprev = np;
926 			else
927 				np->n_tnext->n_tprev = np;
928 		}
929 	}
930 	vrele(vp);
931 	nfsm_srvdone;
932 }
933 
934 /*
935  * Nqnfs client helper daemon. Runs once a second to expire leases.
936  * It also get authorization strings for "kerb" mounts.
937  * It must start at the beginning of the list again after any potential
938  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
939  * the list asynchronously.
940  */
941 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
942 	register struct nfsmount *nmp;
943 	struct ucred *cred;
944 	struct nfsd_cargs *ncd;
945 	int flag;
946 	caddr_t argp;
947 	struct proc *p;
948 {
949 	register struct nfsnode *np;
950 	struct vnode *vp;
951 	int error, vpid;
952 
953 	/*
954 	 * First initialize some variables
955 	 */
956 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
957 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
958 
959 	/*
960 	 * If an authorization string is being passed in, get it.
961 	 */
962 	if ((flag & NFSSVC_GOTAUTH) &&
963 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
964 		if (nmp->nm_flag & NFSMNT_HASAUTH)
965 			panic("cld kerb");
966 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
967 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
968 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
969 				ncd->ncd_authlen) == 0) {
970 				nmp->nm_authtype = ncd->ncd_authtype;
971 				nmp->nm_authlen = ncd->ncd_authlen;
972 			} else
973 				nmp->nm_flag |= NFSMNT_AUTHERR;
974 		} else
975 			nmp->nm_flag |= NFSMNT_AUTHERR;
976 		nmp->nm_flag |= NFSMNT_HASAUTH;
977 		wakeup((caddr_t)&nmp->nm_authlen);
978 	} else
979 		nmp->nm_flag |= NFSMNT_WAITAUTH;
980 
981 	/*
982 	 * Loop every second updating queue until there is a termination sig.
983 	 */
984 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
985 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
986 		np = nmp->nm_tnext;
987 		while (np != (struct nfsnode *)nmp &&
988 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
989 			vp = NFSTOV(np);
990 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
991 			vpid = vp->v_id;
992 			if (np->n_expiry < time.tv_sec) {
993 			   if (vget(vp) == 0) {
994 			     nmp->nm_inprog = vp;
995 			     if (vpid == vp->v_id) {
996 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
997 				if (np->n_tnext == (struct nfsnode *)nmp)
998 					nmp->nm_tprev = np->n_tprev;
999 				else
1000 					np->n_tnext->n_tprev = np->n_tprev;
1001 				if (np->n_tprev == (struct nfsnode *)nmp)
1002 					nmp->nm_tnext = np->n_tnext;
1003 				else
1004 					np->n_tprev->n_tnext = np->n_tnext;
1005 				np->n_tnext = (struct nfsnode *)0;
1006 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
1007 				    && vp->v_type == VREG) {
1008 					if (np->n_flag & NQNFSEVICTED) {
1009 						NFS_VINVBUF(np, vp,
1010 						            TRUE, cred, p);
1011 						np->n_flag &= ~NQNFSEVICTED;
1012 						(void) nqnfs_vacated(vp, cred);
1013 					} else {
1014 						np->n_flag &= ~NMODIFIED;
1015 						(void) VOP_FSYNC(vp, cred,
1016 						    MNT_WAIT, p);
1017 					}
1018 				}
1019 			      }
1020 			      vrele(vp);
1021 			      nmp->nm_inprog = NULLVP;
1022 			    }
1023 			    if (np != nmp->nm_tnext)
1024 				np = nmp->nm_tnext;
1025 			    else
1026 				break;
1027 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
1028 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1029 				 == NQNFSWRITE && vp->v_dirtyblkhd.le_next &&
1030 				 vget(vp) == 0) {
1031 				 nmp->nm_inprog = vp;
1032 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1033 				 if (vpid == vp->v_id &&
1034 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1035 					np->n_brev = np->n_lrev;
1036 				 vrele(vp);
1037 				 nmp->nm_inprog = NULLVP;
1038 			    }
1039 			    if (np != nmp->nm_tnext)
1040 				np = nmp->nm_tnext;
1041 			    else
1042 				break;
1043 			} else
1044 				break;
1045 		}
1046 	    }
1047 
1048 	    /*
1049 	     * Get an authorization string, if required.
1050 	     */
1051 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1052 		ncd->ncd_authuid = nmp->nm_authuid;
1053 		if (copyout((caddr_t)ncd, argp, sizeof (*ncd)))
1054 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1055 		else
1056 			return (ENEEDAUTH);
1057 	    }
1058 
1059 	    /*
1060 	     * Wait a bit (no pun) and do it again.
1061 	     */
1062 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1063 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1064 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1065 			"nqnfstimr", hz / 3);
1066 		    if (error == EINTR || error == ERESTART)
1067 			(void) dounmount(nmp->nm_mountp, MNT_NOFORCE);
1068 	    }
1069 	}
1070 	free((caddr_t)nmp, M_NFSMNT);
1071 	if (error == EWOULDBLOCK)
1072 		error = 0;
1073 	return (error);
1074 }
1075 
1076 /*
1077  * Adjust all timer queue expiry times when the time of day clock is changed.
1078  * Called from the settimeofday() syscall.
1079  */
1080 void
1081 lease_updatetime(deltat)
1082 	register int deltat;
1083 {
1084 	register struct nqlease *lp;
1085 	register struct nfsnode *np;
1086 	struct mount *mp;
1087 	struct nfsmount *nmp;
1088 	int s;
1089 
1090 	if (nqnfsstarttime != 0)
1091 		nqnfsstarttime += deltat;
1092 	s = splsoftclock();
1093 	lp = nqthead.th_chain[0];
1094 	while (lp != (struct nqlease *)&nqthead) {
1095 		lp->lc_expiry += deltat;
1096 		lp = lp->lc_chain1[0];
1097 	}
1098 	splx(s);
1099 
1100 	/*
1101 	 * Search the mount list for all nqnfs mounts and do their timer
1102 	 * queues.
1103 	 */
1104 	mp = rootfs;
1105 	do {
1106 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1107 			nmp = VFSTONFS(mp);
1108 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1109 				np = nmp->nm_tnext;
1110 				while (np != (struct nfsnode *)nmp) {
1111 					np->n_expiry += deltat;
1112 					np = np->n_tnext;
1113 				}
1114 			}
1115 		}
1116 		mp = mp->mnt_next;
1117 	} while (mp != rootfs);
1118 }
1119 
1120 /*
1121  * Lock a server lease.
1122  */
1123 void
1124 nqsrv_locklease(lp)
1125 	struct nqlease *lp;
1126 {
1127 
1128 	while (lp->lc_flag & LC_LOCKED) {
1129 		lp->lc_flag |= LC_WANTED;
1130 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1131 	}
1132 	lp->lc_flag |= LC_LOCKED;
1133 	lp->lc_flag &= ~LC_WANTED;
1134 }
1135 
1136 /*
1137  * Unlock a server lease.
1138  */
1139 void
1140 nqsrv_unlocklease(lp)
1141 	struct nqlease *lp;
1142 {
1143 
1144 	lp->lc_flag &= ~LC_LOCKED;
1145 	if (lp->lc_flag & LC_WANTED)
1146 		wakeup((caddr_t)lp);
1147 }
1148 
1149 /*
1150  * Update a client lease.
1151  */
1152 void
1153 nqnfs_clientlease(nmp, np, rwflag, cachable, expiry, frev)
1154 	register struct nfsmount *nmp;
1155 	register struct nfsnode *np;
1156 	int rwflag, cachable;
1157 	time_t expiry;
1158 	u_quad_t frev;
1159 {
1160 	register struct nfsnode *tp;
1161 
1162 	if (np->n_tnext) {
1163 		if (np->n_tnext == (struct nfsnode *)nmp)
1164 			nmp->nm_tprev = np->n_tprev;
1165 		else
1166 			np->n_tnext->n_tprev = np->n_tprev;
1167 		if (np->n_tprev == (struct nfsnode *)nmp)
1168 			nmp->nm_tnext = np->n_tnext;
1169 		else
1170 			np->n_tprev->n_tnext = np->n_tnext;
1171 		if (rwflag == NQL_WRITE)
1172 			np->n_flag |= NQNFSWRITE;
1173 	} else if (rwflag == NQL_READ)
1174 		np->n_flag &= ~NQNFSWRITE;
1175 	else
1176 		np->n_flag |= NQNFSWRITE;
1177 	if (cachable)
1178 		np->n_flag &= ~NQNFSNONCACHE;
1179 	else
1180 		np->n_flag |= NQNFSNONCACHE;
1181 	np->n_expiry = expiry;
1182 	np->n_lrev = frev;
1183 	tp = nmp->nm_tprev;
1184 	while (tp != (struct nfsnode *)nmp && tp->n_expiry > np->n_expiry)
1185 		tp = tp->n_tprev;
1186 	if (tp == (struct nfsnode *)nmp) {
1187 		np->n_tnext = nmp->nm_tnext;
1188 		nmp->nm_tnext = np;
1189 	} else {
1190 		np->n_tnext = tp->n_tnext;
1191 		tp->n_tnext = np;
1192 	}
1193 	np->n_tprev = tp;
1194 	if (np->n_tnext == (struct nfsnode *)nmp)
1195 		nmp->nm_tprev = np;
1196 	else
1197 		np->n_tnext->n_tprev = np;
1198 }
1199