xref: /csrg-svn/sys/nfs/nfs_nqlease.c (revision 52196)
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	7.2 (Berkeley) 01/14/92
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 "protosw.h"
38 #include "machine/endian.h"
39 #include "ufs/ufs/quota.h"
40 #include "ufs/ufs/inode.h"
41 #include "netinet/in.h"
42 #include "rpcv2.h"
43 #include "nfsv2.h"
44 #include "nfs.h"
45 #include "nfsm_subs.h"
46 #include "xdr_subs.h"
47 #include "nqnfs.h"
48 #include "nfsnode.h"
49 #include "nfsmount.h"
50 
51 /*
52  * List head for the lease queue and other global data.
53  * At any time a lease is linked into a list ordered by increasing expiry time.
54  */
55 #if	((NQLCHSZ&(NQLCHSZ-1)) == 0)
56 #define	NQFHHASH(f)	((*((u_long *)(f)))&(NQLCHSZ-1))
57 #else
58 #define	NQFHHASH(f)	((*((u_long *)(f)))%NQLCHSZ)
59 #endif
60 
61 union nqsrvthead nqthead;
62 union nqsrvthead nqfhead[NQLCHSZ];
63 time_t nqnfsstarttime = (time_t)0;
64 u_long nqnfs_prog, nqnfs_vers;
65 int nqsrv_clockskew = NQ_CLOCKSKEW;
66 int nqsrv_writeslack = NQ_WRITESLACK;
67 int nqsrv_maxlease = NQ_MAXLEASE;
68 int nqsrv_maxnumlease = NQ_MAXNUMLEASE;
69 void nqsrv_instimeq(), nqsrv_send_eviction(), nfs_sndunlock();
70 void nqsrv_unlocklease(), nqsrv_waitfor_expiry();
71 void nqsrv_addhost(), nqsrv_locklease(), nqnfs_serverd();
72 struct mbuf *nfsm_rpchead();
73 
74 /*
75  * Signifies which rpcs can have piggybacked lease requests
76  */
77 int nqnfs_piggy[NFS_NPROCS] = {
78 	0,
79 	NQL_READ,
80 	NQL_WRITE,
81 	0,
82 	NQL_READ,
83 	NQL_READ,
84 	NQL_READ,
85 	0,
86 	NQL_WRITE,
87 	0,
88 	0,
89 	0,
90 	0,
91 	0,
92 	0,
93 	0,
94 	NQL_READ,
95 	0,
96 	NQL_READ,
97 	0,
98 	0,
99 	0,
100 };
101 
102 int nnnnnn = sizeof (struct nqlease);
103 int oooooo = sizeof (struct nfsnode);
104 extern nfstype nfs_type[9];
105 extern struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock;
106 extern struct nfsd nfsd_head;
107 extern int nfsd_waiting;
108 
109 #define TRUE	1
110 #define	FALSE	0
111 
112 /*
113  * Get or check for a lease for "vp", based on NQL_CHECK flag.
114  * The rules are as follows:
115  * - if a current non-caching lease, reply non-caching
116  * - if a current lease for same host only, extend lease
117  * - if a read cachable lease and a read lease request
118  *	add host to list any reply cachable
119  * - else { set non-cachable for read-write sharing }
120  *	send eviction notice messages to all other hosts that have lease
121  *	wait for lease termination { either by receiving vacated messages
122  *					from all the other hosts or expiry
123  *					via. timeout }
124  *	modify lease to non-cachable
125  * - else if no current lease, issue new one
126  * - reply
127  * - return boolean TRUE iff nam should be m_freem()'d
128  * NB: Since nqnfs_serverd() is called from a timer, any potential tsleep()
129  *     in here must be framed by nqsrv_locklease() and nqsrv_unlocklease().
130  *     nqsrv_locklease() is coded such that at least one of LC_LOCKED and
131  *     LC_WANTED is set whenever a process is tsleeping in it. The exception
132  *     is when a new lease is being allocated, since it is not in the timer
133  *     queue yet. (Ditto for the splsoftclock() and splx(s) calls)
134  */
135 nqsrv_getlease(vp, duration, flags, nd, nam, cachablep, frev, cred)
136 	struct vnode *vp;
137 	u_long *duration;
138 	int flags;
139 	struct nfsd *nd;
140 	struct mbuf *nam;
141 	int *cachablep;
142 	u_quad_t *frev;
143 	struct ucred *cred;
144 {
145 	register struct nqlease *lp;
146 	register struct nqhost *lph;
147 	struct nqlease *tlp = (struct nqlease *)0;
148 	struct nqm **lphp;
149 	struct vattr vattr;
150 	union nqsrvthead *lhp;
151 	fhandle_t fh;
152 	int i, ok, error, s;
153 
154 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
155 		return (0);
156 	if (*duration > nqsrv_maxlease)
157 		*duration = nqsrv_maxlease;
158 	if (error = VOP_GETATTR(vp, &vattr, cred, nd->nd_procp))
159 		return (error);
160 	*frev = vattr.va_filerev;
161 	s = splsoftclock();
162 	tlp = vp->v_lease;
163 	if ((flags & NQL_CHECK) == 0)
164 		nfsstats.srvnqnfs_getleases++;
165 	if (tlp == (struct nqlease *)0) {
166 
167 		/*
168 		 * Find the lease by searching the hash list.
169 		 */
170 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
171 		if (error = VFS_VPTOFH(vp, &fh.fh_fid)) {
172 			splx(s);
173 			return (error);
174 		}
175 		lhp = &nqfhead[NQFHHASH(fh.fh_fid.fid_data)];
176 		for (lp = lhp->th_chain[0]; lp != (struct nqlease *)lhp;
177 			lp = lp->lc_fhnext)
178 			if (fh.fh_fsid.val[0] == lp->lc_fsid.val[0] &&
179 			    fh.fh_fsid.val[1] == lp->lc_fsid.val[1] &&
180 			    !bcmp(fh.fh_fid.fid_data, lp->lc_fiddata,
181 				  fh.fh_fid.fid_len - sizeof (long))) {
182 				/* Found it */
183 				lp->lc_vp = vp;
184 				vp->v_lease = lp;
185 				tlp = lp;
186 				break;
187 			}
188 	}
189 	lp = tlp;
190 	if (lp) {
191 		if ((lp->lc_flag & LC_NONCACHABLE) ||
192 		    (lp->lc_morehosts == (struct nqm *)0 &&
193 		     nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host)))
194 			goto doreply;
195 		if ((flags & NQL_READ) && (lp->lc_flag & LC_WRITE)==0) {
196 			if (flags & NQL_CHECK)
197 				goto doreply;
198 			if (nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host))
199 				goto doreply;
200 			i = 0;
201 			if (lp->lc_morehosts) {
202 				lph = lp->lc_morehosts->lpm_hosts;
203 				lphp = &lp->lc_morehosts->lpm_next;
204 				ok = 1;
205 			} else {
206 				lphp = &lp->lc_morehosts;
207 				ok = 0;
208 			}
209 			while (ok && (lph->lph_flag & LC_VALID)) {
210 				if (nqsrv_cmpnam(nd->nd_slp, nam, lph))
211 					goto doreply;
212 				if (++i == LC_MOREHOSTSIZ) {
213 					i = 0;
214 					if (*lphp) {
215 						lph = (*lphp)->lpm_hosts;
216 						lphp = &((*lphp)->lpm_next);
217 					} else
218 						ok = 0;
219 				} else
220 					lph++;
221 			}
222 			nqsrv_locklease(lp);
223 			if (!ok) {
224 				*lphp = (struct nqm *)
225 					malloc(sizeof (struct nqm),
226 						M_NQMHOST, M_WAITOK);
227 				bzero((caddr_t)*lphp, sizeof (struct nqm));
228 				lph = (*lphp)->lpm_hosts;
229 			}
230 			nqsrv_addhost(lph, nd->nd_slp, nd->nd_sref, nam);
231 			nqsrv_unlocklease(lp);
232 		} else {
233 			lp->lc_flag |= LC_NONCACHABLE;
234 			nqsrv_locklease(lp);
235 			VOP_UNLOCK(vp);
236 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
237 			nqsrv_waitfor_expiry(lp);
238 			VOP_LOCK(vp);
239 			nqsrv_unlocklease(lp);
240 		}
241 doreply:
242 		/*
243 		 * Update the lease and return
244 		 */
245 		if ((flags & NQL_CHECK) == 0)
246 			nqsrv_instimeq(lp, *duration);
247 		if (lp->lc_flag & LC_NONCACHABLE)
248 			*cachablep = 0;
249 		else {
250 			*cachablep = 1;
251 			if (flags & NQL_WRITE)
252 				lp->lc_flag |= LC_WRITTEN;
253 		}
254 		splx(s);
255 		return (0);
256 	}
257 	splx(s);
258 	if (flags & NQL_CHECK)
259 		return (0);
260 
261 	/*
262 	 * Allocate new lease
263 	 * The value of nqsrv_maxnumlease should be set generously, so that
264 	 * the following "printf" happens infrequently.
265 	 */
266 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
267 		printf("Nqnfs server, too many leases\n");
268 		do {
269 			(void) tsleep((caddr_t)&lbolt, PSOCK,
270 					"nqsrvnuml", 0);
271 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
272 	}
273 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
274 	bzero((caddr_t)lp, sizeof (struct nqlease));
275 	if (flags & NQL_WRITE)
276 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
277 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nd->nd_sref, nam);
278 	lp->lc_vp = vp;
279 	lp->lc_fsid = fh.fh_fsid;
280 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
281 	lp->lc_fhnext = lhp->th_chain[0];
282 	if (lhp->th_head[0] == lhp)
283 		lhp->th_chain[1] = lp;
284 	else
285 		lhp->th_chain[0]->lc_fhprev = lp;
286 	lp->lc_fhprev = (struct nqlease *)lhp;
287 	lhp->th_chain[0] = lp;
288 	vp->v_lease = lp;
289 	s = splsoftclock();
290 	nqsrv_instimeq(lp, *duration);
291 	splx(s);
292 	*cachablep = 1;
293 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
294 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
295 	return (0);
296 }
297 
298 /*
299  * Local lease check for server syscalls.
300  * Just set up args and let nqsrv_getlease() do the rest.
301  */
302 void
303 lease_check(vp, p, cred, flag)
304 	struct vnode *vp;
305 	struct proc *p;
306 	struct ucred *cred;
307 	int flag;
308 {
309 	int duration, cache;
310 	struct nfsd nfsd;
311 	u_quad_t frev;
312 
313 	nfsd.nd_slp = NQLOCALSLP;
314 	nfsd.nd_procp = p;
315 	(void) nqsrv_getlease(vp, &duration, NQL_CHECK | flag, &nfsd,
316 		(struct mbuf *)0, &cache, &frev, cred);
317 }
318 
319 /*
320  * Add a host to an nqhost structure for a lease.
321  */
322 void
323 nqsrv_addhost(lph, slp, sref, nam)
324 	register struct nqhost *lph;
325 	struct nfssvc_sock *slp;
326 	u_short sref;
327 	struct mbuf *nam;
328 {
329 	register struct sockaddr_in *saddr;
330 
331 	if (slp == NQLOCALSLP)
332 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
333 	else if (slp == nfs_udpsock) {
334 		saddr = mtod(nam, struct sockaddr_in *);
335 		lph->lph_flag |= (LC_VALID | LC_UDP);
336 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
337 		lph->lph_port = saddr->sin_port;
338 	} else if (slp == nfs_cltpsock) {
339 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
340 		lph->lph_flag |= (LC_VALID | LC_CLTP);
341 	} else {
342 		lph->lph_flag |= LC_VALID;
343 		lph->lph_slp = slp;
344 		lph->lph_sref = sref;
345 	}
346 }
347 
348 /*
349  * Update the lease expiry time and position it in the timer queue correctly.
350  */
351 void
352 nqsrv_instimeq(lp, duration)
353 	register struct nqlease *lp;
354 	u_long duration;
355 {
356 	register struct nqlease *tlp;
357 	time_t newexpiry;
358 
359 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
360 	if (lp->lc_expiry == newexpiry)
361 		return;
362 	if (lp->lc_chain1[0])
363 		remque(lp);
364 	lp->lc_expiry = newexpiry;
365 
366 	/*
367 	 * Find where in the queue it should be.
368 	 */
369 	tlp = nqthead.th_chain[1];
370 	while (tlp->lc_expiry > newexpiry && tlp != (struct nqlease *)&nqthead)
371 		tlp = tlp->lc_chain1[1];
372 	if (tlp == nqthead.th_chain[1])
373 		NQSTORENOVRAM(newexpiry);
374 	insque(lp, tlp);
375 }
376 
377 /*
378  * Compare the requesting host address with the lph entry in the lease.
379  * Return true iff it is the same.
380  * This is somewhat messy due to the union in the nqhost structure.
381  * The local host is indicated by the special value of NQLOCALSLP for slp.
382  */
383 nqsrv_cmpnam(slp, nam, lph)
384 	register struct nfssvc_sock *slp;
385 	struct mbuf *nam;
386 	register struct nqhost *lph;
387 {
388 	register struct sockaddr_in *saddr;
389 	struct mbuf *addr;
390 	union nethostaddr lhaddr;
391 	int ret;
392 
393 	if (slp == NQLOCALSLP) {
394 		if (lph->lph_flag & LC_LOCAL)
395 			return (1);
396 		else
397 			return (0);
398 	}
399 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
400 		addr = nam;
401 	else
402 		addr = slp->ns_nam;
403 	if (lph->lph_flag & LC_UDP)
404 		ret = nfs_netaddr_match(AF_INET, &lph->lph_haddr,
405 			(union nethostaddr *)0, addr);
406 	else if (lph->lph_flag & LC_CLTP)
407 		ret = nfs_netaddr_match(AF_ISO, &lph->lph_claddr,
408 			(union nethostaddr *)0, addr);
409 	else {
410 		if (lph->lph_sref != lph->lph_slp->ns_sref)
411 			return (0);
412 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
413 		if (saddr->sin_family == AF_INET)
414 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
415 		else
416 			lhaddr.had_nam = lph->lph_slp->ns_nam;
417 		ret = nfs_netaddr_match(saddr->sin_family, &lhaddr,
418 			(union nethostaddr *)0, addr);
419 	}
420 	return (ret);
421 }
422 
423 /*
424  * Send out eviction notice messages to all other hosts for the lease.
425  */
426 void
427 nqsrv_send_eviction(vp, lp, slp, nam, cred)
428 	struct vnode *vp;
429 	register struct nqlease *lp;
430 	struct nfssvc_sock *slp;
431 	struct mbuf *nam;
432 	struct ucred *cred;
433 {
434 	register struct nqhost *lph = &lp->lc_host;
435 	register struct mbuf *m;
436 	register int siz;
437 	struct nqm *lphnext = lp->lc_morehosts;
438 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
439 	struct socket *so;
440 	struct sockaddr_in *saddr;
441 	fhandle_t *fhp;
442 	caddr_t bpos, cp;
443 	u_long xid;
444 	int len = 1, ok = 1, i = 0;
445 	int sotype, *solockp;
446 
447 	while (ok && (lph->lph_flag & LC_VALID)) {
448 		if (nqsrv_cmpnam(slp, nam, lph))
449 			lph->lph_flag |= LC_VACATED;
450 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
451 			if (lph->lph_flag & LC_UDP) {
452 				MGET(nam2, M_WAIT, MT_SONAME);
453 				saddr = mtod(nam2, struct sockaddr_in *);
454 				nam2->m_len = saddr->sin_len =
455 					sizeof (struct sockaddr_in);
456 				saddr->sin_family = AF_INET;
457 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
458 				saddr->sin_port = lph->lph_port;
459 				so = nfs_udpsock->ns_so;
460 			} else if (lph->lph_flag & LC_CLTP) {
461 				nam2 = lph->lph_nam;
462 				so = nfs_cltpsock->ns_so;
463 			} else if (lph->lph_sref == lph->lph_slp->ns_sref) {
464 				nam2 = (struct mbuf *)0;
465 				so = lph->lph_slp->ns_so;
466 			} else
467 				goto nextone;
468 			sotype = so->so_type;
469 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
470 				solockp = &lph->lph_slp->ns_solock;
471 			else
472 				solockp = (int *)0;
473 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
474 				NFSX_FH);
475 			nfsm_build(cp, caddr_t, NFSX_FH);
476 			bzero(cp, NFSX_FH);
477 			fhp = (fhandle_t *)cp;
478 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
479 			VFS_VPTOFH(vp, &fhp->fh_fid);
480 			m = mreq;
481 			siz = 0;
482 			while (m) {
483 				siz += m->m_len;
484 				m = m->m_next;
485 			}
486 			if (siz <= 0 || siz > NFS_MAXPACKET) {
487 				printf("mbuf siz=%d\n",siz);
488 				panic("Bad nfs svc reply");
489 			}
490 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
491 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
492 				mreq, siz, &mheadend, &xid);
493 			/*
494 			 * For stream protocols, prepend a Sun RPC
495 			 * Record Mark.
496 			 */
497 			if (sotype == SOCK_STREAM) {
498 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
499 				*mtod(m, u_long *) = htonl(0x80000000 |
500 					(m->m_pkthdr.len - NFSX_UNSIGNED));
501 			}
502 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
503 			    lph->lph_sref != lph->lph_slp->ns_sref) ||
504 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
505 				m_freem(m);
506 			else {
507 				if (solockp)
508 					*solockp |= NFSMNT_SNDLOCK;
509 				(void) nfs_send(so, nam2, m,
510 						(struct nfsreq *)0);
511 				if (solockp)
512 					nfs_sndunlock(solockp);
513 			}
514 			if (lph->lph_flag & LC_UDP)
515 				MFREE(nam2, m);
516 		}
517 nextone:
518 		if (++i == len) {
519 			if (lphnext) {
520 				i = 0;
521 				len = LC_MOREHOSTSIZ;
522 				lph = lphnext->lpm_hosts;
523 				lphnext = lphnext->lpm_next;
524 			} else
525 				ok = 0;
526 		} else
527 			lph++;
528 	}
529 }
530 
531 /*
532  * Wait for the lease to expire.
533  * This will occur when all clients have sent "vacated" messages to
534  * this server OR when it expires do to timeout.
535  */
536 void
537 nqsrv_waitfor_expiry(lp)
538 	register struct nqlease *lp;
539 {
540 	register struct nqhost *lph;
541 	register int i;
542 	struct nqm *lphnext;
543 	int len, ok;
544 
545 tryagain:
546 	if (time.tv_sec > lp->lc_expiry)
547 		return;
548 	lph = &lp->lc_host;
549 	lphnext = lp->lc_morehosts;
550 	len = 1;
551 	i = 0;
552 	ok = 1;
553 	while (ok && (lph->lph_flag & LC_VALID)) {
554 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
555 			lp->lc_flag |= LC_EXPIREDWANTED;
556 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
557 					"nqexp", 0);
558 			goto tryagain;
559 		}
560 		if (++i == len) {
561 			if (lphnext) {
562 				i = 0;
563 				len = LC_MOREHOSTSIZ;
564 				lph = lphnext->lpm_hosts;
565 				lphnext = lphnext->lpm_next;
566 			} else
567 				ok = 0;
568 		} else
569 			lph++;
570 	}
571 }
572 
573 /*
574  * Nqnfs server timer that maintains the server lease queue.
575  * Scan the lease queue for expired entries:
576  * - when one is found, wakeup anyone waiting for it
577  *   else dequeue and free
578  */
579 void
580 nqnfs_serverd()
581 {
582 	register struct nqlease *lp;
583 	register struct nqhost *lph;
584 	struct nqlease *nextlp;
585 	struct nqm *lphnext, *olphnext;
586 	struct mbuf *n;
587 	union nqsrvthead *lhp;
588 	int i, len, ok;
589 
590 	lp = nqthead.th_chain[0];
591 	while (lp != (struct nqlease *)&nqthead) {
592 		if (lp->lc_expiry >= time.tv_sec)
593 			break;
594 		nextlp = lp->lc_chain1[0];
595 		if (lp->lc_flag & LC_EXPIREDWANTED) {
596 			lp->lc_flag &= ~LC_EXPIREDWANTED;
597 			wakeup((caddr_t)&lp->lc_flag);
598 		} else if ((lp->lc_flag & (LC_LOCKED | LC_WANTED)) == 0) {
599 		    /*
600 		     * Make a best effort at keeping a write caching lease long
601 		     * enough by not deleting it until it has been explicitly
602 		     * vacated or there have been no writes in the previous
603 		     * write_slack seconds since expiry and the nfsds are not
604 		     * all busy. The assumption is that if the nfsds are not
605 		     * all busy now (no queue of nfs requests), then the client
606 		     * would have been able to do at least one write to the
607 		     * file during the last write_slack seconds if it was still
608 		     * trying to push writes to the server.
609 		     */
610 		    if ((lp->lc_flag & (LC_WRITE | LC_VACATED)) == LC_WRITE &&
611 			((lp->lc_flag & LC_WRITTEN) || nfsd_waiting == 0)) {
612 			lp->lc_flag &= ~LC_WRITTEN;
613 			nqsrv_instimeq(lp, nqsrv_writeslack);
614 		    } else {
615 			remque(lp);
616 			lhp = &nqfhead[NQFHHASH(lp->lc_fiddata)];
617 			if (lp->lc_fhprev == (struct nqlease *)lhp)
618 				lhp->th_chain[0] = lp->lc_fhnext;
619 			else
620 				lp->lc_fhprev->lc_fhnext = lp->lc_fhnext;
621 			if (lp->lc_fhnext == (struct nqlease *)lhp)
622 				lhp->th_chain[1] = lp->lc_fhprev;
623 			else
624 				lp->lc_fhnext->lc_fhprev = lp->lc_fhprev;
625 			/*
626 			 * This soft reference may no longer be valid, but
627 			 * no harm done. The worst case is if the vnode was
628 			 * recycled and has another valid lease reference,
629 			 * which is dereferenced prematurely.
630 			 */
631 			lp->lc_vp->v_lease = (struct nqlease *)0;
632 			lph = &lp->lc_host;
633 			lphnext = lp->lc_morehosts;
634 			olphnext = (struct nqm *)0;
635 			len = 1;
636 			i = 0;
637 			ok = 1;
638 			while (ok && (lph->lph_flag & LC_VALID)) {
639 				if (lph->lph_flag & LC_CLTP)
640 					MFREE(lph->lph_nam, n);
641 				if (++i == len) {
642 					if (olphnext) {
643 						free((caddr_t)olphnext, M_NQMHOST);
644 						olphnext = (struct nqm *)0;
645 					}
646 					if (lphnext) {
647 						olphnext = lphnext;
648 						i = 0;
649 						len = LC_MOREHOSTSIZ;
650 						lph = lphnext->lpm_hosts;
651 						lphnext = lphnext->lpm_next;
652 					} else
653 						ok = 0;
654 				} else
655 					lph++;
656 			}
657 			FREE((caddr_t)lp, M_NQLEASE);
658 			if (olphnext)
659 				free((caddr_t)olphnext, M_NQMHOST);
660 			nfsstats.srvnqnfs_leases--;
661 		    }
662 		}
663 		lp = nextlp;
664 	}
665 }
666 
667 /*
668  * Called from nfssvc_nfsd() for a getlease rpc request.
669  * Do the from/to xdr translation and call nqsrv_getlease() to
670  * do the real work.
671  */
672 nqnfsrv_getlease(nfsd, mrep, md, dpos, cred, nam, mrq)
673 	struct nfsd *nfsd;
674 	struct mbuf *mrep, *md;
675 	caddr_t dpos;
676 	struct ucred *cred;
677 	struct mbuf *nam, **mrq;
678 {
679 	register struct nfsv2_fattr *fp;
680 	struct vattr va;
681 	register struct vattr *vap = &va;
682 	struct vnode *vp;
683 	nfsv2fh_t nfh;
684 	fhandle_t *fhp;
685 	register u_long *tl;
686 	register long t1;
687 	u_quad_t frev;
688 	caddr_t bpos;
689 	int error = 0;
690 	char *cp2;
691 	struct mbuf *mb, *mb2, *mreq;
692 	int flags, rdonly, cache;
693 
694 	fhp = &nfh.fh_generic;
695 	nfsm_srvmtofh(fhp);
696 	nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED);
697 	flags = fxdr_unsigned(int, *tl++);
698 	nfsd->nd_duration = fxdr_unsigned(int, *tl);
699 	if (error = nfsrv_fhtovp(fhp, TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
700 		nfsm_reply(0);
701 	if (rdonly && flags == NQL_WRITE) {
702 		error = EROFS;
703 		nfsm_reply(0);
704 	}
705 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
706 		nam, &cache, &frev, cred);
707 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
708 	vput(vp);
709 	nfsm_reply(NFSX_FATTR + 4*NFSX_UNSIGNED);
710 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
711 	*tl++ = txdr_unsigned(cache);
712 	*tl++ = txdr_unsigned(nfsd->nd_duration);
713 	txdr_hyper(&frev, tl);
714 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_FATTR);
715 	nfsm_srvfillattr;
716 	nfsm_srvdone;
717 }
718 
719 /*
720  * Called from nfssvc_nfsd() when a "vacated" message is received from a
721  * client. Find the entry and expire it.
722  */
723 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
724 	struct nfsd *nfsd;
725 	struct mbuf *mrep, *md;
726 	caddr_t dpos;
727 	struct ucred *cred;
728 	struct mbuf *nam, **mrq;
729 {
730 	register struct nqlease *lp;
731 	register struct nqhost *lph;
732 	struct nqlease *tlp = (struct nqlease *)0;
733 	struct vnode *vp;
734 	nfsv2fh_t nfh;
735 	fhandle_t *fhp;
736 	register u_long *tl;
737 	register long t1;
738 	struct nqm *lphnext;
739 	union nqsrvthead *lhp;
740 	u_quad_t frev;
741 	int error = 0, i, len, ok, rdonly, gotit = 0;
742 	char *cp2;
743 
744 	fhp = &nfh.fh_generic;
745 	nfsm_srvmtofh(fhp);
746 	if (error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
747 		return (error);
748 	m_freem(mrep);
749 	tlp = vp->v_lease;
750 	if (tlp == (struct nqlease *)0) {
751 		/*
752 		 * Find the lease by searching the hash list.
753 		 */
754 		lhp = &nqfhead[NQFHHASH(fhp->fh_fid.fid_data)];
755 		for (lp = lhp->th_chain[0]; lp != (struct nqlease *)lhp;
756 			lp = lp->lc_fhnext)
757 			if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
758 			    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
759 			    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
760 				  MAXFIDSZ)) {
761 				/* Found it */
762 				lp->lc_vp = vp;
763 				vp->v_lease = lp;
764 				tlp = lp;
765 				break;
766 			}
767 	}
768 	vrele(vp);
769 	if (tlp) {
770 		lp = tlp;
771 		len = 1;
772 		i = 0;
773 		lph = &lp->lc_host;
774 		lphnext = lp->lc_morehosts;
775 		ok = 1;
776 		while (ok && (lph->lph_flag & LC_VALID)) {
777 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
778 				lph->lph_flag |= LC_VACATED;
779 				gotit++;
780 				break;
781 			}
782 			if (++i == len) {
783 				if (lphnext) {
784 					len = LC_MOREHOSTSIZ;
785 					i = 0;
786 					lph = lphnext->lpm_hosts;
787 					lphnext = lphnext->lpm_next;
788 				} else
789 					ok = 0;
790 			} else
791 				lph++;
792 		}
793 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
794 			lp->lc_flag &= ~LC_EXPIREDWANTED;
795 			wakeup((caddr_t)&lp->lc_flag);
796 		}
797 nfsmout:
798 		return (EPERM);
799 	}
800 	return (EPERM);
801 }
802 
803 /*
804  * Client get lease rpc function.
805  */
806 nqnfs_getlease(vp, rwflag, cred, p)
807 	register struct vnode *vp;
808 	int rwflag;
809 	struct ucred *cred;
810 	struct proc *p;
811 {
812 	register u_long *tl;
813 	register caddr_t cp;
814 	register long t1;
815 	register struct nfsnode *np, *tp;
816 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
817 	caddr_t bpos, dpos, cp2;
818 	time_t reqtime;
819 	int error = 0;
820 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
821 	int cachable;
822 
823 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
824 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
825 		 &bpos);
826 	nfsm_fhtom(vp);
827 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
828 	*tl++ = txdr_unsigned(rwflag);
829 	*tl = txdr_unsigned(nmp->nm_leaseterm);
830 	reqtime = time.tv_sec;
831 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
832 	np = VTONFS(vp);
833 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
834 	cachable = fxdr_unsigned(int, *tl++);
835 	reqtime += fxdr_unsigned(int, *tl++);
836 	if (reqtime > time.tv_sec) {
837 		if (np->n_tnext) {
838 			if (np->n_tnext == (struct nfsnode *)nmp)
839 				nmp->nm_tprev = np->n_tprev;
840 			else
841 				np->n_tnext->n_tprev = np->n_tprev;
842 			if (np->n_tprev == (struct nfsnode *)nmp)
843 				nmp->nm_tnext = np->n_tnext;
844 			else
845 				np->n_tprev->n_tnext = np->n_tnext;
846 			if (rwflag == NQL_WRITE)
847 				np->n_flag |= NQNFSWRITE;
848 		} else if (rwflag == NQL_READ)
849 			np->n_flag &= ~NQNFSWRITE;
850 		else
851 			np->n_flag |= NQNFSWRITE;
852 		if (cachable)
853 			np->n_flag &= ~NQNFSNONCACHE;
854 		else
855 			np->n_flag |= NQNFSNONCACHE;
856 		np->n_expiry = reqtime;
857 		fxdr_hyper(tl, &np->n_lrev);
858 		tp = nmp->nm_tprev;
859 		while (tp != (struct nfsnode *)nmp && tp->n_expiry > np->n_expiry)
860 			tp = tp->n_tprev;
861 		if (tp == (struct nfsnode *)nmp) {
862 			np->n_tnext = nmp->nm_tnext;
863 			nmp->nm_tnext = np;
864 		} else {
865 			np->n_tnext = tp->n_tnext;
866 			tp->n_tnext = np;
867 		}
868 		np->n_tprev = tp;
869 		if (np->n_tnext == (struct nfsnode *)nmp)
870 			nmp->nm_tprev = np;
871 		else
872 			np->n_tnext->n_tprev = np;
873 		nfsm_loadattr(vp, (struct vattr *)0);
874 	} else
875 		error = NQNFS_EXPIRED;
876 	nfsm_reqdone;
877 	return (error);
878 }
879 
880 /*
881  * Client vacated message function.
882  */
883 nqnfs_vacated(vp, cred)
884 	register struct vnode *vp;
885 	struct ucred *cred;
886 {
887 	register caddr_t cp;
888 	register struct mbuf *m;
889 	register int i;
890 	caddr_t bpos;
891 	u_long xid;
892 	int error = 0;
893 	struct mbuf *mreq, *mb, *mb2, *mheadend;
894 	struct nfsmount *nmp;
895 	struct nfsreq myrep;
896 
897 	nmp = VFSTONFS(vp->v_mount);
898 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
899 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
900 	nfsm_fhtom(vp);
901 	m = mreq;
902 	i = 0;
903 	while (m) {
904 		i += m->m_len;
905 		m = m->m_next;
906 	}
907 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
908 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
909 		mreq, i, &mheadend, &xid);
910 	if (nmp->nm_sotype == SOCK_STREAM) {
911 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
912 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
913 			NFSX_UNSIGNED));
914 	}
915 	myrep.r_flags = 0;
916 	myrep.r_nmp = nmp;
917 	if (nmp->nm_soflags & PR_CONNREQUIRED)
918 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
919 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
920 	if (nmp->nm_soflags & PR_CONNREQUIRED)
921 		nfs_sndunlock(&nmp->nm_flag);
922 	return (error);
923 }
924 
925 /*
926  * Called for client side callbacks
927  */
928 nqnfs_callback(nmp, mrep, md, dpos)
929 	struct nfsmount *nmp;
930 	struct mbuf *mrep, *md;
931 	caddr_t dpos;
932 {
933 	register struct vnode *vp;
934 	register u_long *tl;
935 	register long t1;
936 	nfsv2fh_t nfh;
937 	fhandle_t *fhp;
938 	struct nfsnode *np;
939 	struct nfsd nd;
940 	int error;
941 	char *cp2;
942 
943 	nd.nd_mrep = mrep;
944 	nd.nd_md = md;
945 	nd.nd_dpos = dpos;
946 	if (error = nfs_getreq(&nd, FALSE))
947 		return (error);
948 	md = nd.nd_md;
949 	dpos = nd.nd_dpos;
950 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
951 		m_freem(mrep);
952 		return (EPERM);
953 	}
954 	fhp = &nfh.fh_generic;
955 	nfsm_srvmtofh(fhp);
956 	m_freem(mrep);
957 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
958 		return (error);
959 	vp = NFSTOV(np);
960 	if (np->n_tnext) {
961 		np->n_expiry = 0;
962 		np->n_flag |= NQNFSEVICTED;
963 		if (np->n_tprev != (struct nfsnode *)nmp) {
964 			if (np->n_tnext == (struct nfsnode *)nmp)
965 				nmp->nm_tprev = np->n_tprev;
966 			else
967 				np->n_tnext->n_tprev = np->n_tprev;
968 			np->n_tprev->n_tnext = np->n_tnext;
969 			np->n_tnext = nmp->nm_tnext;
970 			nmp->nm_tnext = np;
971 			np->n_tprev = (struct nfsnode *)nmp;
972 			if (np->n_tnext == (struct nfsnode *)nmp)
973 				nmp->nm_tprev = np;
974 			else
975 				np->n_tnext->n_tprev = np;
976 		}
977 	}
978 	vrele(vp);
979 	nfsm_srvdone;
980 }
981 
982 /*
983  * Nqnfs client helper daemon. Runs once a second to expire leases.
984  * It also get authorization strings for "kerb" mounts.
985  * It must start at the beginning of the list again after any potential
986  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
987  * the list asynchronously.
988  */
989 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
990 	register struct nfsmount *nmp;
991 	struct ucred *cred;
992 	struct nfsd_cargs *ncd;
993 	int flag;
994 	caddr_t argp;
995 	struct proc *p;
996 {
997 	register struct nfsnode *np;
998 	struct vnode *vp;
999 	int error, vpid;
1000 
1001 	/*
1002 	 * First initialize some variables
1003 	 */
1004 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1005 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
1006 
1007 	/*
1008 	 * If an authorization string is being passed in, get it.
1009 	 */
1010 	if ((flag & NFSSVC_GOTAUTH) &&
1011 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
1012 		if (nmp->nm_flag & NFSMNT_HASAUTH)
1013 			panic("cld kerb");
1014 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
1015 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
1016 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
1017 				ncd->ncd_authlen) == 0) {
1018 				nmp->nm_authtype = ncd->ncd_authtype;
1019 				nmp->nm_authlen = ncd->ncd_authlen;
1020 			} else
1021 				nmp->nm_flag |= NFSMNT_AUTHERR;
1022 		} else
1023 			nmp->nm_flag |= NFSMNT_AUTHERR;
1024 		nmp->nm_flag |= NFSMNT_HASAUTH;
1025 		wakeup((caddr_t)&nmp->nm_authlen);
1026 	} else
1027 		nmp->nm_flag |= NFSMNT_WAITAUTH;
1028 
1029 	/*
1030 	 * Loop every second updating queue until there is a termination sig.
1031 	 */
1032 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
1033 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
1034 		np = nmp->nm_tnext;
1035 		while (np != (struct nfsnode *)nmp &&
1036 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
1037 			vp = NFSTOV(np);
1038 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
1039 			vpid = vp->v_id;
1040 			if (np->n_expiry < time.tv_sec) {
1041 			   if (vget(vp) == 0) {
1042 			     nmp->nm_inprog = vp;
1043 			     if (vpid == vp->v_id) {
1044 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
1045 				if (np->n_tnext == (struct nfsnode *)nmp)
1046 					nmp->nm_tprev = np->n_tprev;
1047 				else
1048 					np->n_tnext->n_tprev = np->n_tprev;
1049 				if (np->n_tprev == (struct nfsnode *)nmp)
1050 					nmp->nm_tnext = np->n_tnext;
1051 				else
1052 					np->n_tprev->n_tnext = np->n_tnext;
1053 				np->n_tnext = (struct nfsnode *)0;
1054 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
1055 				    && vp->v_type == VREG) {
1056 					np->n_flag &= ~NMODIFIED;
1057 					if (np->n_flag & NQNFSEVICTED) {
1058 						vinvalbuf(vp, TRUE);
1059 						np->n_flag &= ~NQNFSEVICTED;
1060 					} else
1061 						vflushbuf(vp, B_SYNC);
1062 					(void) nqnfs_vacated(vp, cred);
1063 				}
1064 			      }
1065 			      vrele(vp);
1066 			      nmp->nm_inprog = NULLVP;
1067 			    }
1068 			    if (np != nmp->nm_tnext)
1069 				np = nmp->nm_tnext;
1070 			    else
1071 				break;
1072 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
1073 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1074 				 == NQNFSWRITE && vp->v_dirtyblkhd &&
1075 				 vget(vp) == 0) {
1076 				 nmp->nm_inprog = vp;
1077 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1078 				 if (vpid == vp->v_id &&
1079 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1080 					np->n_brev = np->n_lrev;
1081 				 vrele(vp);
1082 				 nmp->nm_inprog = NULLVP;
1083 			    }
1084 			    if (np != nmp->nm_tnext)
1085 				np = nmp->nm_tnext;
1086 			    else
1087 				break;
1088 			} else
1089 				break;
1090 		}
1091 	    }
1092 
1093 	    /*
1094 	     * Get an authorization string, if required.
1095 	     */
1096 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1097 		ncd->ncd_authuid = nmp->nm_authuid;
1098 		if (copyout((caddr_t)ncd, argp, sizeof (*ncd)))
1099 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1100 		else
1101 			return (ENEEDAUTH);
1102 	    }
1103 
1104 	    /*
1105 	     * Wait a bit (no pun) and do it again.
1106 	     */
1107 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1108 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1109 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1110 			"nqnfstimr", hz / 3);
1111 		    if (error == EINTR || error == ERESTART)
1112 			(void) dounmount(nmp->nm_mountp, MNT_NOFORCE);
1113 	    }
1114 	}
1115 	free((caddr_t)nmp, M_NFSMNT);
1116 	if (error == EWOULDBLOCK)
1117 		error = 0;
1118 	return (error);
1119 }
1120 
1121 /*
1122  * Adjust all timer queue expiry times when the time of day clock is changed.
1123  * Called from the settimeofday() syscall.
1124  */
1125 void
1126 lease_updatetime(deltat)
1127 	register int deltat;
1128 {
1129 	register struct nqlease *lp;
1130 	register struct nfsnode *np;
1131 	struct mount *mp;
1132 	struct nfsmount *nmp;
1133 	int s;
1134 
1135 	if (nqnfsstarttime != 0)
1136 		nqnfsstarttime += deltat;
1137 	s = splsoftclock();
1138 	lp = nqthead.th_chain[0];
1139 	while (lp != (struct nqlease *)&nqthead) {
1140 		lp->lc_expiry += deltat;
1141 		lp = lp->lc_chain1[0];
1142 	}
1143 	splx(s);
1144 
1145 	/*
1146 	 * Search the mount list for all nqnfs mounts and do their timer
1147 	 * queues.
1148 	 */
1149 	mp = rootfs;
1150 	do {
1151 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1152 			nmp = VFSTONFS(mp);
1153 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1154 				np = nmp->nm_tnext;
1155 				while (np != (struct nfsnode *)nmp) {
1156 					np->n_expiry += deltat;
1157 					np = np->n_tnext;
1158 				}
1159 			}
1160 		}
1161 		mp = mp->mnt_next;
1162 	} while (mp != rootfs);
1163 }
1164 
1165 /*
1166  * Lock a server lease.
1167  */
1168 void
1169 nqsrv_locklease(lp)
1170 	struct nqlease *lp;
1171 {
1172 
1173 	while (lp->lc_flag & LC_LOCKED) {
1174 		lp->lc_flag |= LC_WANTED;
1175 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1176 	}
1177 	lp->lc_flag |= LC_LOCKED;
1178 	lp->lc_flag &= ~LC_WANTED;
1179 }
1180 
1181 /*
1182  * Unlock a server lease.
1183  */
1184 void
1185 nqsrv_unlocklease(lp)
1186 	struct nqlease *lp;
1187 {
1188 
1189 	lp->lc_flag &= ~LC_LOCKED;
1190 	if (lp->lc_flag & LC_WANTED)
1191 		wakeup((caddr_t)lp);
1192 }
1193