xref: /original-bsd/sys/nfs/nfs_nqlease.c (revision 68d9582f)
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	7.6 (Berkeley) 05/14/92
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 "param.h"
27 #include "vnode.h"
28 #include "mount.h"
29 #include "kernel.h"
30 #include "proc.h"
31 #include "systm.h"
32 #include "mbuf.h"
33 #include "socket.h"
34 #include "socketvar.h"
35 #include "file.h"
36 #include "buf.h"
37 #include "stat.h"
38 #include "protosw.h"
39 #include "machine/endian.h"
40 #include "netinet/in.h"
41 #include "rpcv2.h"
42 #include "nfsv2.h"
43 #include "nfs.h"
44 #include "nfsm_subs.h"
45 #include "xdr_subs.h"
46 #include "nqnfs.h"
47 #include "nfsnode.h"
48 #include "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 #if	((NQLCHSZ&(NQLCHSZ-1)) == 0)
55 #define	NQFHHASH(f)	((*((u_long *)(f)))&(NQLCHSZ-1))
56 #else
57 #define	NQFHHASH(f)	((*((u_long *)(f)))%NQLCHSZ)
58 #endif
59 
60 union nqsrvthead nqthead;
61 union nqsrvthead nqfhead[NQLCHSZ];
62 time_t nqnfsstarttime = (time_t)0;
63 u_long nqnfs_prog, nqnfs_vers;
64 int nqsrv_clockskew = NQ_CLOCKSKEW;
65 int nqsrv_writeslack = NQ_WRITESLACK;
66 int nqsrv_maxlease = NQ_MAXLEASE;
67 int nqsrv_maxnumlease = NQ_MAXNUMLEASE;
68 void nqsrv_instimeq(), nqsrv_send_eviction(), nfs_sndunlock();
69 void nqsrv_unlocklease(), nqsrv_waitfor_expiry(), nfsrv_slpderef();
70 void nqsrv_addhost(), nqsrv_locklease(), nqnfs_serverd();
71 struct mbuf *nfsm_rpchead();
72 
73 /*
74  * Signifies which rpcs can have piggybacked lease requests
75  */
76 int nqnfs_piggy[NFS_NPROCS] = {
77 	0,
78 	NQL_READ,
79 	NQL_WRITE,
80 	0,
81 	NQL_READ,
82 	NQL_READ,
83 	NQL_READ,
84 	0,
85 	NQL_WRITE,
86 	0,
87 	0,
88 	0,
89 	0,
90 	0,
91 	0,
92 	0,
93 	NQL_READ,
94 	0,
95 	NQL_READ,
96 	0,
97 	0,
98 	0,
99 };
100 
101 int nnnnnn = sizeof (struct nqlease);
102 int oooooo = sizeof (struct nfsnode);
103 extern nfstype nfs_type[9];
104 extern struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock;
105 extern struct nfsd nfsd_head;
106 extern int nfsd_waiting;
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 	USES_VOP_GETATTR;
145 	USES_VOP_LOCK;
146 	USES_VOP_UNLOCK;
147 	register struct nqlease *lp;
148 	register struct nqhost *lph;
149 	struct nqlease *tlp = (struct nqlease *)0;
150 	struct nqm **lphp;
151 	struct vattr vattr;
152 	union nqsrvthead *lhp;
153 	fhandle_t fh;
154 	int i, ok, error, s;
155 
156 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
157 		return (0);
158 	if (*duration > nqsrv_maxlease)
159 		*duration = nqsrv_maxlease;
160 	if (error = VOP_GETATTR(vp, &vattr, cred, nd->nd_procp))
161 		return (error);
162 	*frev = vattr.va_filerev;
163 	s = splsoftclock();
164 	tlp = vp->v_lease;
165 	if ((flags & NQL_CHECK) == 0)
166 		nfsstats.srvnqnfs_getleases++;
167 	if (tlp == (struct nqlease *)0) {
168 
169 		/*
170 		 * Find the lease by searching the hash list.
171 		 */
172 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
173 		if (error = VFS_VPTOFH(vp, &fh.fh_fid)) {
174 			splx(s);
175 			return (error);
176 		}
177 		lhp = &nqfhead[NQFHHASH(fh.fh_fid.fid_data)];
178 		for (lp = lhp->th_chain[0]; lp != (struct nqlease *)lhp;
179 			lp = lp->lc_fhnext)
180 			if (fh.fh_fsid.val[0] == lp->lc_fsid.val[0] &&
181 			    fh.fh_fsid.val[1] == lp->lc_fsid.val[1] &&
182 			    !bcmp(fh.fh_fid.fid_data, lp->lc_fiddata,
183 				  fh.fh_fid.fid_len - sizeof (long))) {
184 				/* Found it */
185 				lp->lc_vp = vp;
186 				vp->v_lease = lp;
187 				tlp = lp;
188 				break;
189 			}
190 	}
191 	lp = tlp;
192 	if (lp) {
193 		if ((lp->lc_flag & LC_NONCACHABLE) ||
194 		    (lp->lc_morehosts == (struct nqm *)0 &&
195 		     nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host)))
196 			goto doreply;
197 		if ((flags & NQL_READ) && (lp->lc_flag & LC_WRITE)==0) {
198 			if (flags & NQL_CHECK)
199 				goto doreply;
200 			if (nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host))
201 				goto doreply;
202 			i = 0;
203 			if (lp->lc_morehosts) {
204 				lph = lp->lc_morehosts->lpm_hosts;
205 				lphp = &lp->lc_morehosts->lpm_next;
206 				ok = 1;
207 			} else {
208 				lphp = &lp->lc_morehosts;
209 				ok = 0;
210 			}
211 			while (ok && (lph->lph_flag & LC_VALID)) {
212 				if (nqsrv_cmpnam(nd->nd_slp, nam, lph))
213 					goto doreply;
214 				if (++i == LC_MOREHOSTSIZ) {
215 					i = 0;
216 					if (*lphp) {
217 						lph = (*lphp)->lpm_hosts;
218 						lphp = &((*lphp)->lpm_next);
219 					} else
220 						ok = 0;
221 				} else
222 					lph++;
223 			}
224 			nqsrv_locklease(lp);
225 			if (!ok) {
226 				*lphp = (struct nqm *)
227 					malloc(sizeof (struct nqm),
228 						M_NQMHOST, M_WAITOK);
229 				bzero((caddr_t)*lphp, sizeof (struct nqm));
230 				lph = (*lphp)->lpm_hosts;
231 			}
232 			nqsrv_addhost(lph, nd->nd_slp, nam);
233 			nqsrv_unlocklease(lp);
234 		} else {
235 			lp->lc_flag |= LC_NONCACHABLE;
236 			nqsrv_locklease(lp);
237 			VOP_UNLOCK(vp);
238 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
239 			nqsrv_waitfor_expiry(lp);
240 			VOP_LOCK(vp);
241 			nqsrv_unlocklease(lp);
242 		}
243 doreply:
244 		/*
245 		 * Update the lease and return
246 		 */
247 		if ((flags & NQL_CHECK) == 0)
248 			nqsrv_instimeq(lp, *duration);
249 		if (lp->lc_flag & LC_NONCACHABLE)
250 			*cachablep = 0;
251 		else {
252 			*cachablep = 1;
253 			if (flags & NQL_WRITE)
254 				lp->lc_flag |= LC_WRITTEN;
255 		}
256 		splx(s);
257 		return (0);
258 	}
259 	splx(s);
260 	if (flags & NQL_CHECK)
261 		return (0);
262 
263 	/*
264 	 * Allocate new lease
265 	 * The value of nqsrv_maxnumlease should be set generously, so that
266 	 * the following "printf" happens infrequently.
267 	 */
268 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
269 		printf("Nqnfs server, too many leases\n");
270 		do {
271 			(void) tsleep((caddr_t)&lbolt, PSOCK,
272 					"nqsrvnuml", 0);
273 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
274 	}
275 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
276 	bzero((caddr_t)lp, sizeof (struct nqlease));
277 	if (flags & NQL_WRITE)
278 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
279 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nam);
280 	lp->lc_vp = vp;
281 	lp->lc_fsid = fh.fh_fsid;
282 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
283 	lp->lc_fhnext = lhp->th_chain[0];
284 	if (lhp->th_head[0] == lhp)
285 		lhp->th_chain[1] = lp;
286 	else
287 		lhp->th_chain[0]->lc_fhprev = lp;
288 	lp->lc_fhprev = (struct nqlease *)lhp;
289 	lhp->th_chain[0] = lp;
290 	vp->v_lease = lp;
291 	s = splsoftclock();
292 	nqsrv_instimeq(lp, *duration);
293 	splx(s);
294 	*cachablep = 1;
295 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
296 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
297 	return (0);
298 }
299 
300 /*
301  * Local lease check for server syscalls.
302  * Just set up args and let nqsrv_getlease() do the rest.
303  */
304 void
305 lease_check(vp, p, cred, flag)
306 	struct vnode *vp;
307 	struct proc *p;
308 	struct ucred *cred;
309 	int flag;
310 {
311 	int duration, cache;
312 	struct nfsd nfsd;
313 	u_quad_t frev;
314 
315 	nfsd.nd_slp = NQLOCALSLP;
316 	nfsd.nd_procp = p;
317 	(void) nqsrv_getlease(vp, &duration, NQL_CHECK | flag, &nfsd,
318 		(struct mbuf *)0, &cache, &frev, cred);
319 }
320 
321 /*
322  * Add a host to an nqhost structure for a lease.
323  */
324 void
325 nqsrv_addhost(lph, slp, nam)
326 	register struct nqhost *lph;
327 	struct nfssvc_sock *slp;
328 	struct mbuf *nam;
329 {
330 	register struct sockaddr_in *saddr;
331 
332 	if (slp == NQLOCALSLP)
333 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
334 	else if (slp == nfs_udpsock) {
335 		saddr = mtod(nam, struct sockaddr_in *);
336 		lph->lph_flag |= (LC_VALID | LC_UDP);
337 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
338 		lph->lph_port = saddr->sin_port;
339 	} else if (slp == nfs_cltpsock) {
340 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
341 		lph->lph_flag |= (LC_VALID | LC_CLTP);
342 	} else {
343 		lph->lph_flag |= (LC_VALID | LC_SREF);
344 		lph->lph_slp = slp;
345 		slp->ns_sref++;
346 	}
347 }
348 
349 /*
350  * Update the lease expiry time and position it in the timer queue correctly.
351  */
352 void
353 nqsrv_instimeq(lp, duration)
354 	register struct nqlease *lp;
355 	u_long duration;
356 {
357 	register struct nqlease *tlp;
358 	time_t newexpiry;
359 
360 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
361 	if (lp->lc_expiry == newexpiry)
362 		return;
363 	if (lp->lc_chain1[0])
364 		remque(lp);
365 	lp->lc_expiry = newexpiry;
366 
367 	/*
368 	 * Find where in the queue it should be.
369 	 */
370 	tlp = nqthead.th_chain[1];
371 	while (tlp->lc_expiry > newexpiry && tlp != (struct nqlease *)&nqthead)
372 		tlp = tlp->lc_chain1[1];
373 	if (tlp == nqthead.th_chain[1])
374 		NQSTORENOVRAM(newexpiry);
375 	insque(lp, tlp);
376 }
377 
378 /*
379  * Compare the requesting host address with the lph entry in the lease.
380  * Return true iff it is the same.
381  * This is somewhat messy due to the union in the nqhost structure.
382  * The local host is indicated by the special value of NQLOCALSLP for slp.
383  */
384 nqsrv_cmpnam(slp, nam, lph)
385 	register struct nfssvc_sock *slp;
386 	struct mbuf *nam;
387 	register struct nqhost *lph;
388 {
389 	register struct sockaddr_in *saddr;
390 	struct mbuf *addr;
391 	union nethostaddr lhaddr;
392 	int ret;
393 
394 	if (slp == NQLOCALSLP) {
395 		if (lph->lph_flag & LC_LOCAL)
396 			return (1);
397 		else
398 			return (0);
399 	}
400 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
401 		addr = nam;
402 	else
403 		addr = slp->ns_nam;
404 	if (lph->lph_flag & LC_UDP)
405 		ret = nfs_netaddr_match(AF_INET, &lph->lph_haddr,
406 			(union nethostaddr *)0, addr);
407 	else if (lph->lph_flag & LC_CLTP)
408 		ret = nfs_netaddr_match(AF_ISO, &lph->lph_claddr,
409 			(union nethostaddr *)0, addr);
410 	else {
411 		if ((lph->lph_slp->ns_flag & SLP_VALID) == 0)
412 			return (0);
413 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
414 		if (saddr->sin_family == AF_INET)
415 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
416 		else
417 			lhaddr.had_nam = lph->lph_slp->ns_nam;
418 		ret = nfs_netaddr_match(saddr->sin_family, &lhaddr,
419 			(union nethostaddr *)0, addr);
420 	}
421 	return (ret);
422 }
423 
424 /*
425  * Send out eviction notice messages to all other hosts for the lease.
426  */
427 void
428 nqsrv_send_eviction(vp, lp, slp, nam, cred)
429 	struct vnode *vp;
430 	register struct nqlease *lp;
431 	struct nfssvc_sock *slp;
432 	struct mbuf *nam;
433 	struct ucred *cred;
434 {
435 	register struct nqhost *lph = &lp->lc_host;
436 	register struct mbuf *m;
437 	register int siz;
438 	struct nqm *lphnext = lp->lc_morehosts;
439 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
440 	struct socket *so;
441 	struct sockaddr_in *saddr;
442 	fhandle_t *fhp;
443 	caddr_t bpos, cp;
444 	u_long xid;
445 	int len = 1, ok = 1, i = 0;
446 	int sotype, *solockp;
447 
448 	while (ok && (lph->lph_flag & LC_VALID)) {
449 		if (nqsrv_cmpnam(slp, nam, lph))
450 			lph->lph_flag |= LC_VACATED;
451 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
452 			if (lph->lph_flag & LC_UDP) {
453 				MGET(nam2, M_WAIT, MT_SONAME);
454 				saddr = mtod(nam2, struct sockaddr_in *);
455 				nam2->m_len = saddr->sin_len =
456 					sizeof (struct sockaddr_in);
457 				saddr->sin_family = AF_INET;
458 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
459 				saddr->sin_port = lph->lph_port;
460 				so = nfs_udpsock->ns_so;
461 			} else if (lph->lph_flag & LC_CLTP) {
462 				nam2 = lph->lph_nam;
463 				so = nfs_cltpsock->ns_so;
464 			} else if (lph->lph_slp->ns_flag & SLP_VALID) {
465 				nam2 = (struct mbuf *)0;
466 				so = lph->lph_slp->ns_so;
467 			} else
468 				goto nextone;
469 			sotype = so->so_type;
470 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
471 				solockp = &lph->lph_slp->ns_solock;
472 			else
473 				solockp = (int *)0;
474 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
475 				NFSX_FH);
476 			nfsm_build(cp, caddr_t, NFSX_FH);
477 			bzero(cp, NFSX_FH);
478 			fhp = (fhandle_t *)cp;
479 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
480 			VFS_VPTOFH(vp, &fhp->fh_fid);
481 			m = mreq;
482 			siz = 0;
483 			while (m) {
484 				siz += m->m_len;
485 				m = m->m_next;
486 			}
487 			if (siz <= 0 || siz > NFS_MAXPACKET) {
488 				printf("mbuf siz=%d\n",siz);
489 				panic("Bad nfs svc reply");
490 			}
491 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
492 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
493 				mreq, siz, &mheadend, &xid);
494 			/*
495 			 * For stream protocols, prepend a Sun RPC
496 			 * Record Mark.
497 			 */
498 			if (sotype == SOCK_STREAM) {
499 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
500 				*mtod(m, u_long *) = htonl(0x80000000 |
501 					(m->m_pkthdr.len - NFSX_UNSIGNED));
502 			}
503 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
504 			    (lph->lph_slp->ns_flag & SLP_VALID) == 0) ||
505 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
506 				m_freem(m);
507 			else {
508 				if (solockp)
509 					*solockp |= NFSMNT_SNDLOCK;
510 				(void) nfs_send(so, nam2, m,
511 						(struct nfsreq *)0);
512 				if (solockp)
513 					nfs_sndunlock(solockp);
514 			}
515 			if (lph->lph_flag & LC_UDP)
516 				MFREE(nam2, m);
517 		}
518 nextone:
519 		if (++i == len) {
520 			if (lphnext) {
521 				i = 0;
522 				len = LC_MOREHOSTSIZ;
523 				lph = lphnext->lpm_hosts;
524 				lphnext = lphnext->lpm_next;
525 			} else
526 				ok = 0;
527 		} else
528 			lph++;
529 	}
530 }
531 
532 /*
533  * Wait for the lease to expire.
534  * This will occur when all clients have sent "vacated" messages to
535  * this server OR when it expires do to timeout.
536  */
537 void
538 nqsrv_waitfor_expiry(lp)
539 	register struct nqlease *lp;
540 {
541 	register struct nqhost *lph;
542 	register int i;
543 	struct nqm *lphnext;
544 	int len, ok;
545 
546 tryagain:
547 	if (time.tv_sec > lp->lc_expiry)
548 		return;
549 	lph = &lp->lc_host;
550 	lphnext = lp->lc_morehosts;
551 	len = 1;
552 	i = 0;
553 	ok = 1;
554 	while (ok && (lph->lph_flag & LC_VALID)) {
555 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
556 			lp->lc_flag |= LC_EXPIREDWANTED;
557 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
558 					"nqexp", 0);
559 			goto tryagain;
560 		}
561 		if (++i == len) {
562 			if (lphnext) {
563 				i = 0;
564 				len = LC_MOREHOSTSIZ;
565 				lph = lphnext->lpm_hosts;
566 				lphnext = lphnext->lpm_next;
567 			} else
568 				ok = 0;
569 		} else
570 			lph++;
571 	}
572 }
573 
574 /*
575  * Nqnfs server timer that maintains the server lease queue.
576  * Scan the lease queue for expired entries:
577  * - when one is found, wakeup anyone waiting for it
578  *   else dequeue and free
579  */
580 void
581 nqnfs_serverd()
582 {
583 	register struct nqlease *lp;
584 	register struct nqhost *lph;
585 	struct nqlease *nextlp;
586 	struct nqm *lphnext, *olphnext;
587 	struct mbuf *n;
588 	union nqsrvthead *lhp;
589 	int i, len, ok;
590 
591 	lp = nqthead.th_chain[0];
592 	while (lp != (struct nqlease *)&nqthead) {
593 		if (lp->lc_expiry >= time.tv_sec)
594 			break;
595 		nextlp = lp->lc_chain1[0];
596 		if (lp->lc_flag & LC_EXPIREDWANTED) {
597 			lp->lc_flag &= ~LC_EXPIREDWANTED;
598 			wakeup((caddr_t)&lp->lc_flag);
599 		} else if ((lp->lc_flag & (LC_LOCKED | LC_WANTED)) == 0) {
600 		    /*
601 		     * Make a best effort at keeping a write caching lease long
602 		     * enough by not deleting it until it has been explicitly
603 		     * vacated or there have been no writes in the previous
604 		     * write_slack seconds since expiry and the nfsds are not
605 		     * all busy. The assumption is that if the nfsds are not
606 		     * all busy now (no queue of nfs requests), then the client
607 		     * would have been able to do at least one write to the
608 		     * file during the last write_slack seconds if it was still
609 		     * trying to push writes to the server.
610 		     */
611 		    if ((lp->lc_flag & (LC_WRITE | LC_VACATED)) == LC_WRITE &&
612 			((lp->lc_flag & LC_WRITTEN) || nfsd_waiting == 0)) {
613 			lp->lc_flag &= ~LC_WRITTEN;
614 			nqsrv_instimeq(lp, nqsrv_writeslack);
615 		    } else {
616 			remque(lp);
617 			lhp = &nqfhead[NQFHHASH(lp->lc_fiddata)];
618 			if (lp->lc_fhprev == (struct nqlease *)lhp)
619 				lhp->th_chain[0] = lp->lc_fhnext;
620 			else
621 				lp->lc_fhprev->lc_fhnext = lp->lc_fhnext;
622 			if (lp->lc_fhnext == (struct nqlease *)lhp)
623 				lhp->th_chain[1] = lp->lc_fhprev;
624 			else
625 				lp->lc_fhnext->lc_fhprev = lp->lc_fhprev;
626 			/*
627 			 * This soft reference may no longer be valid, but
628 			 * no harm done. The worst case is if the vnode was
629 			 * recycled and has another valid lease reference,
630 			 * which is dereferenced prematurely.
631 			 */
632 			lp->lc_vp->v_lease = (struct nqlease *)0;
633 			lph = &lp->lc_host;
634 			lphnext = lp->lc_morehosts;
635 			olphnext = (struct nqm *)0;
636 			len = 1;
637 			i = 0;
638 			ok = 1;
639 			while (ok && (lph->lph_flag & LC_VALID)) {
640 				if (lph->lph_flag & LC_CLTP)
641 					MFREE(lph->lph_nam, n);
642 				if (lph->lph_flag & LC_SREF)
643 					nfsrv_slpderef(lph->lph_slp);
644 				if (++i == len) {
645 					if (olphnext) {
646 						free((caddr_t)olphnext, M_NQMHOST);
647 						olphnext = (struct nqm *)0;
648 					}
649 					if (lphnext) {
650 						olphnext = lphnext;
651 						i = 0;
652 						len = LC_MOREHOSTSIZ;
653 						lph = lphnext->lpm_hosts;
654 						lphnext = lphnext->lpm_next;
655 					} else
656 						ok = 0;
657 				} else
658 					lph++;
659 			}
660 			FREE((caddr_t)lp, M_NQLEASE);
661 			if (olphnext)
662 				free((caddr_t)olphnext, M_NQMHOST);
663 			nfsstats.srvnqnfs_leases--;
664 		    }
665 		}
666 		lp = nextlp;
667 	}
668 }
669 
670 /*
671  * Called from nfssvc_nfsd() for a getlease rpc request.
672  * Do the from/to xdr translation and call nqsrv_getlease() to
673  * do the real work.
674  */
675 nqnfsrv_getlease(nfsd, mrep, md, dpos, cred, nam, mrq)
676 	struct nfsd *nfsd;
677 	struct mbuf *mrep, *md;
678 	caddr_t dpos;
679 	struct ucred *cred;
680 	struct mbuf *nam, **mrq;
681 {
682 	USES_VOP_GETATTR;
683 	register struct nfsv2_fattr *fp;
684 	struct vattr va;
685 	register struct vattr *vap = &va;
686 	struct vnode *vp;
687 	nfsv2fh_t nfh;
688 	fhandle_t *fhp;
689 	register u_long *tl;
690 	register long t1;
691 	u_quad_t frev;
692 	caddr_t bpos;
693 	int error = 0;
694 	char *cp2;
695 	struct mbuf *mb, *mb2, *mreq;
696 	int flags, rdonly, cache;
697 
698 	fhp = &nfh.fh_generic;
699 	nfsm_srvmtofh(fhp);
700 	nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
701 	flags = fxdr_unsigned(int, *tl++);
702 	nfsd->nd_duration = fxdr_unsigned(int, *tl);
703 	if (error = nfsrv_fhtovp(fhp, TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
704 		nfsm_reply(0);
705 	if (rdonly && flags == NQL_WRITE) {
706 		error = EROFS;
707 		nfsm_reply(0);
708 	}
709 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
710 		nam, &cache, &frev, cred);
711 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
712 	vput(vp);
713 	nfsm_reply(NFSX_FATTR + 4*NFSX_UNSIGNED);
714 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
715 	*tl++ = txdr_unsigned(cache);
716 	*tl++ = txdr_unsigned(nfsd->nd_duration);
717 	txdr_hyper(&frev, tl);
718 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_FATTR);
719 	nfsm_srvfillattr;
720 	nfsm_srvdone;
721 }
722 
723 /*
724  * Called from nfssvc_nfsd() when a "vacated" message is received from a
725  * client. Find the entry and expire it.
726  */
727 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
728 	struct nfsd *nfsd;
729 	struct mbuf *mrep, *md;
730 	caddr_t dpos;
731 	struct ucred *cred;
732 	struct mbuf *nam, **mrq;
733 {
734 	register struct nqlease *lp;
735 	register struct nqhost *lph;
736 	struct nqlease *tlp = (struct nqlease *)0;
737 	struct vnode *vp;
738 	nfsv2fh_t nfh;
739 	fhandle_t *fhp;
740 	register u_long *tl;
741 	register long t1;
742 	struct nqm *lphnext;
743 	union nqsrvthead *lhp;
744 	u_quad_t frev;
745 	int error = 0, i, len, ok, rdonly, gotit = 0;
746 	char *cp2;
747 
748 	fhp = &nfh.fh_generic;
749 	nfsm_srvmtofh(fhp);
750 	if (error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
751 		return (error);
752 	m_freem(mrep);
753 	tlp = vp->v_lease;
754 	if (tlp == (struct nqlease *)0) {
755 		/*
756 		 * Find the lease by searching the hash list.
757 		 */
758 		lhp = &nqfhead[NQFHHASH(fhp->fh_fid.fid_data)];
759 		for (lp = lhp->th_chain[0]; lp != (struct nqlease *)lhp;
760 			lp = lp->lc_fhnext)
761 			if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
762 			    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
763 			    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
764 				  MAXFIDSZ)) {
765 				/* Found it */
766 				lp->lc_vp = vp;
767 				vp->v_lease = lp;
768 				tlp = lp;
769 				break;
770 			}
771 	}
772 	vrele(vp);
773 	if (tlp) {
774 		lp = tlp;
775 		len = 1;
776 		i = 0;
777 		lph = &lp->lc_host;
778 		lphnext = lp->lc_morehosts;
779 		ok = 1;
780 		while (ok && (lph->lph_flag & LC_VALID)) {
781 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
782 				lph->lph_flag |= LC_VACATED;
783 				gotit++;
784 				break;
785 			}
786 			if (++i == len) {
787 				if (lphnext) {
788 					len = LC_MOREHOSTSIZ;
789 					i = 0;
790 					lph = lphnext->lpm_hosts;
791 					lphnext = lphnext->lpm_next;
792 				} else
793 					ok = 0;
794 			} else
795 				lph++;
796 		}
797 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
798 			lp->lc_flag &= ~LC_EXPIREDWANTED;
799 			wakeup((caddr_t)&lp->lc_flag);
800 		}
801 nfsmout:
802 		return (EPERM);
803 	}
804 	return (EPERM);
805 }
806 
807 /*
808  * Client get lease rpc function.
809  */
810 nqnfs_getlease(vp, rwflag, cred, p)
811 	register struct vnode *vp;
812 	int rwflag;
813 	struct ucred *cred;
814 	struct proc *p;
815 {
816 	register u_long *tl;
817 	register caddr_t cp;
818 	register long t1;
819 	register struct nfsnode *np, *tp;
820 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
821 	caddr_t bpos, dpos, cp2;
822 	time_t reqtime;
823 	int error = 0;
824 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
825 	int cachable;
826 
827 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
828 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
829 		 &bpos);
830 	nfsm_fhtom(vp);
831 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
832 	*tl++ = txdr_unsigned(rwflag);
833 	*tl = txdr_unsigned(nmp->nm_leaseterm);
834 	reqtime = time.tv_sec;
835 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
836 	np = VTONFS(vp);
837 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
838 	cachable = fxdr_unsigned(int, *tl++);
839 	reqtime += fxdr_unsigned(int, *tl++);
840 	if (reqtime > time.tv_sec) {
841 		if (np->n_tnext) {
842 			if (np->n_tnext == (struct nfsnode *)nmp)
843 				nmp->nm_tprev = np->n_tprev;
844 			else
845 				np->n_tnext->n_tprev = np->n_tprev;
846 			if (np->n_tprev == (struct nfsnode *)nmp)
847 				nmp->nm_tnext = np->n_tnext;
848 			else
849 				np->n_tprev->n_tnext = np->n_tnext;
850 			if (rwflag == NQL_WRITE)
851 				np->n_flag |= NQNFSWRITE;
852 		} else if (rwflag == NQL_READ)
853 			np->n_flag &= ~NQNFSWRITE;
854 		else
855 			np->n_flag |= NQNFSWRITE;
856 		if (cachable)
857 			np->n_flag &= ~NQNFSNONCACHE;
858 		else
859 			np->n_flag |= NQNFSNONCACHE;
860 		np->n_expiry = reqtime;
861 		fxdr_hyper(tl, &np->n_lrev);
862 		tp = nmp->nm_tprev;
863 		while (tp != (struct nfsnode *)nmp && tp->n_expiry > np->n_expiry)
864 			tp = tp->n_tprev;
865 		if (tp == (struct nfsnode *)nmp) {
866 			np->n_tnext = nmp->nm_tnext;
867 			nmp->nm_tnext = np;
868 		} else {
869 			np->n_tnext = tp->n_tnext;
870 			tp->n_tnext = np;
871 		}
872 		np->n_tprev = tp;
873 		if (np->n_tnext == (struct nfsnode *)nmp)
874 			nmp->nm_tprev = np;
875 		else
876 			np->n_tnext->n_tprev = np;
877 		nfsm_loadattr(vp, (struct vattr *)0);
878 	} else
879 		error = NQNFS_EXPIRED;
880 	nfsm_reqdone;
881 	return (error);
882 }
883 
884 /*
885  * Client vacated message function.
886  */
887 nqnfs_vacated(vp, cred)
888 	register struct vnode *vp;
889 	struct ucred *cred;
890 {
891 	register caddr_t cp;
892 	register struct mbuf *m;
893 	register int i;
894 	caddr_t bpos;
895 	u_long xid;
896 	int error = 0;
897 	struct mbuf *mreq, *mb, *mb2, *mheadend;
898 	struct nfsmount *nmp;
899 	struct nfsreq myrep;
900 
901 	nmp = VFSTONFS(vp->v_mount);
902 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
903 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
904 	nfsm_fhtom(vp);
905 	m = mreq;
906 	i = 0;
907 	while (m) {
908 		i += m->m_len;
909 		m = m->m_next;
910 	}
911 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
912 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
913 		mreq, i, &mheadend, &xid);
914 	if (nmp->nm_sotype == SOCK_STREAM) {
915 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
916 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
917 			NFSX_UNSIGNED));
918 	}
919 	myrep.r_flags = 0;
920 	myrep.r_nmp = nmp;
921 	if (nmp->nm_soflags & PR_CONNREQUIRED)
922 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
923 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
924 	if (nmp->nm_soflags & PR_CONNREQUIRED)
925 		nfs_sndunlock(&nmp->nm_flag);
926 	return (error);
927 }
928 
929 /*
930  * Called for client side callbacks
931  */
932 nqnfs_callback(nmp, mrep, md, dpos)
933 	struct nfsmount *nmp;
934 	struct mbuf *mrep, *md;
935 	caddr_t dpos;
936 {
937 	register struct vnode *vp;
938 	register u_long *tl;
939 	register long t1;
940 	nfsv2fh_t nfh;
941 	fhandle_t *fhp;
942 	struct nfsnode *np;
943 	struct nfsd nd;
944 	int error;
945 	char *cp2;
946 
947 	nd.nd_mrep = mrep;
948 	nd.nd_md = md;
949 	nd.nd_dpos = dpos;
950 	if (error = nfs_getreq(&nd, FALSE))
951 		return (error);
952 	md = nd.nd_md;
953 	dpos = nd.nd_dpos;
954 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
955 		m_freem(mrep);
956 		return (EPERM);
957 	}
958 	fhp = &nfh.fh_generic;
959 	nfsm_srvmtofh(fhp);
960 	m_freem(mrep);
961 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
962 		return (error);
963 	vp = NFSTOV(np);
964 	if (np->n_tnext) {
965 		np->n_expiry = 0;
966 		np->n_flag |= NQNFSEVICTED;
967 		if (np->n_tprev != (struct nfsnode *)nmp) {
968 			if (np->n_tnext == (struct nfsnode *)nmp)
969 				nmp->nm_tprev = np->n_tprev;
970 			else
971 				np->n_tnext->n_tprev = np->n_tprev;
972 			np->n_tprev->n_tnext = np->n_tnext;
973 			np->n_tnext = nmp->nm_tnext;
974 			nmp->nm_tnext = np;
975 			np->n_tprev = (struct nfsnode *)nmp;
976 			if (np->n_tnext == (struct nfsnode *)nmp)
977 				nmp->nm_tprev = np;
978 			else
979 				np->n_tnext->n_tprev = np;
980 		}
981 	}
982 	vrele(vp);
983 	nfsm_srvdone;
984 }
985 
986 /*
987  * Nqnfs client helper daemon. Runs once a second to expire leases.
988  * It also get authorization strings for "kerb" mounts.
989  * It must start at the beginning of the list again after any potential
990  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
991  * the list asynchronously.
992  */
993 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
994 	register struct nfsmount *nmp;
995 	struct ucred *cred;
996 	struct nfsd_cargs *ncd;
997 	int flag;
998 	caddr_t argp;
999 	struct proc *p;
1000 {
1001 	register struct nfsnode *np;
1002 	struct vnode *vp;
1003 	int error, vpid;
1004 
1005 	/*
1006 	 * First initialize some variables
1007 	 */
1008 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1009 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
1010 
1011 	/*
1012 	 * If an authorization string is being passed in, get it.
1013 	 */
1014 	if ((flag & NFSSVC_GOTAUTH) &&
1015 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
1016 		if (nmp->nm_flag & NFSMNT_HASAUTH)
1017 			panic("cld kerb");
1018 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
1019 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
1020 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
1021 				ncd->ncd_authlen) == 0) {
1022 				nmp->nm_authtype = ncd->ncd_authtype;
1023 				nmp->nm_authlen = ncd->ncd_authlen;
1024 			} else
1025 				nmp->nm_flag |= NFSMNT_AUTHERR;
1026 		} else
1027 			nmp->nm_flag |= NFSMNT_AUTHERR;
1028 		nmp->nm_flag |= NFSMNT_HASAUTH;
1029 		wakeup((caddr_t)&nmp->nm_authlen);
1030 	} else
1031 		nmp->nm_flag |= NFSMNT_WAITAUTH;
1032 
1033 	/*
1034 	 * Loop every second updating queue until there is a termination sig.
1035 	 */
1036 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
1037 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
1038 		np = nmp->nm_tnext;
1039 		while (np != (struct nfsnode *)nmp &&
1040 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
1041 			vp = NFSTOV(np);
1042 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
1043 			vpid = vp->v_id;
1044 			if (np->n_expiry < time.tv_sec) {
1045 			   if (vget(vp) == 0) {
1046 			     nmp->nm_inprog = vp;
1047 			     if (vpid == vp->v_id) {
1048 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
1049 				if (np->n_tnext == (struct nfsnode *)nmp)
1050 					nmp->nm_tprev = np->n_tprev;
1051 				else
1052 					np->n_tnext->n_tprev = np->n_tprev;
1053 				if (np->n_tprev == (struct nfsnode *)nmp)
1054 					nmp->nm_tnext = np->n_tnext;
1055 				else
1056 					np->n_tprev->n_tnext = np->n_tnext;
1057 				np->n_tnext = (struct nfsnode *)0;
1058 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
1059 				    && vp->v_type == VREG) {
1060 					np->n_flag &= ~NMODIFIED;
1061 					if (np->n_flag & NQNFSEVICTED) {
1062 						vinvalbuf(vp, TRUE);
1063 						np->n_flag &= ~NQNFSEVICTED;
1064 						(void) nqnfs_vacated(vp, cred);
1065 					} else
1066 						vflushbuf(vp, B_SYNC);
1067 				}
1068 			      }
1069 			      vrele(vp);
1070 			      nmp->nm_inprog = NULLVP;
1071 			    }
1072 			    if (np != nmp->nm_tnext)
1073 				np = nmp->nm_tnext;
1074 			    else
1075 				break;
1076 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
1077 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1078 				 == NQNFSWRITE && vp->v_dirtyblkhd &&
1079 				 vget(vp) == 0) {
1080 				 nmp->nm_inprog = vp;
1081 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1082 				 if (vpid == vp->v_id &&
1083 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1084 					np->n_brev = np->n_lrev;
1085 				 vrele(vp);
1086 				 nmp->nm_inprog = NULLVP;
1087 			    }
1088 			    if (np != nmp->nm_tnext)
1089 				np = nmp->nm_tnext;
1090 			    else
1091 				break;
1092 			} else
1093 				break;
1094 		}
1095 	    }
1096 
1097 	    /*
1098 	     * Get an authorization string, if required.
1099 	     */
1100 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1101 		ncd->ncd_authuid = nmp->nm_authuid;
1102 		if (copyout((caddr_t)ncd, argp, sizeof (*ncd)))
1103 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1104 		else
1105 			return (ENEEDAUTH);
1106 	    }
1107 
1108 	    /*
1109 	     * Wait a bit (no pun) and do it again.
1110 	     */
1111 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1112 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1113 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1114 			"nqnfstimr", hz / 3);
1115 		    if (error == EINTR || error == ERESTART)
1116 			(void) dounmount(nmp->nm_mountp, MNT_NOFORCE);
1117 	    }
1118 	}
1119 	free((caddr_t)nmp, M_NFSMNT);
1120 	if (error == EWOULDBLOCK)
1121 		error = 0;
1122 	return (error);
1123 }
1124 
1125 /*
1126  * Adjust all timer queue expiry times when the time of day clock is changed.
1127  * Called from the settimeofday() syscall.
1128  */
1129 void
1130 lease_updatetime(deltat)
1131 	register int deltat;
1132 {
1133 	register struct nqlease *lp;
1134 	register struct nfsnode *np;
1135 	struct mount *mp;
1136 	struct nfsmount *nmp;
1137 	int s;
1138 
1139 	if (nqnfsstarttime != 0)
1140 		nqnfsstarttime += deltat;
1141 	s = splsoftclock();
1142 	lp = nqthead.th_chain[0];
1143 	while (lp != (struct nqlease *)&nqthead) {
1144 		lp->lc_expiry += deltat;
1145 		lp = lp->lc_chain1[0];
1146 	}
1147 	splx(s);
1148 
1149 	/*
1150 	 * Search the mount list for all nqnfs mounts and do their timer
1151 	 * queues.
1152 	 */
1153 	mp = rootfs;
1154 	do {
1155 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1156 			nmp = VFSTONFS(mp);
1157 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1158 				np = nmp->nm_tnext;
1159 				while (np != (struct nfsnode *)nmp) {
1160 					np->n_expiry += deltat;
1161 					np = np->n_tnext;
1162 				}
1163 			}
1164 		}
1165 		mp = mp->mnt_next;
1166 	} while (mp != rootfs);
1167 }
1168 
1169 /*
1170  * Lock a server lease.
1171  */
1172 void
1173 nqsrv_locklease(lp)
1174 	struct nqlease *lp;
1175 {
1176 
1177 	while (lp->lc_flag & LC_LOCKED) {
1178 		lp->lc_flag |= LC_WANTED;
1179 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1180 	}
1181 	lp->lc_flag |= LC_LOCKED;
1182 	lp->lc_flag &= ~LC_WANTED;
1183 }
1184 
1185 /*
1186  * Unlock a server lease.
1187  */
1188 void
1189 nqsrv_unlocklease(lp)
1190 	struct nqlease *lp;
1191 {
1192 
1193 	lp->lc_flag &= ~LC_LOCKED;
1194 	if (lp->lc_flag & LC_WANTED)
1195 		wakeup((caddr_t)lp);
1196 }
1197