xref: /original-bsd/sys/nfs/nfs_nqlease.c (revision c94cb6e0)
1 /*
2  * Copyright (c) 1992, 1993
3  *	The Regents of the University of California.  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	8.4 (Berkeley) 08/14/94
11  */
12 
13 /*
14  * References:
15  *	Cary G. Gray and David R. Cheriton, "Leases: An Efficient Fault-Tolerant
16  *		Mechanism for Distributed File Cache Consistency",
17  *		In Proc. of the Twelfth ACM Symposium on Operating Systems
18  *		Principals, pg. 202-210, Litchfield Park, AZ, Dec. 1989.
19  *	Michael N. Nelson, Brent B. Welch and John K. Ousterhout, "Caching
20  *		in the Sprite Network File System", ACM TOCS 6(1),
21  *		pages 134-154, February 1988.
22  *	V. Srinivasan and Jeffrey C. Mogul, "Spritely NFS: Implementation and
23  *		Performance of Cache-Consistency Protocols", Digital
24  *		Equipment Corporation WRL Research Report 89/5, May 1989.
25  */
26 #include <sys/param.h>
27 #include <sys/vnode.h>
28 #include <sys/mount.h>
29 #include <sys/kernel.h>
30 #include <sys/proc.h>
31 #include <sys/systm.h>
32 #include <sys/mbuf.h>
33 #include <sys/socket.h>
34 #include <sys/socketvar.h>
35 #include <sys/file.h>
36 #include <sys/buf.h>
37 #include <sys/stat.h>
38 #include <sys/protosw.h>
39 
40 #include <netinet/in.h>
41 #include <nfs/rpcv2.h>
42 #include <nfs/nfsv2.h>
43 #include <nfs/nfs.h>
44 #include <nfs/nfsm_subs.h>
45 #include <nfs/xdr_subs.h>
46 #include <nfs/nqnfs.h>
47 #include <nfs/nfsnode.h>
48 #include <nfs/nfsmount.h>
49 
50 /*
51  * List head for the lease queue and other global data.
52  * At any time a lease is linked into a list ordered by increasing expiry time.
53  */
54 #define	NQFHHASH(f)	((*((u_long *)(f)))&nqfheadhash)
55 
56 union nqsrvthead nqthead;
57 struct nqlease **nqfhead;
58 u_long nqfheadhash;
59 time_t nqnfsstarttime = (time_t)0;
60 u_long nqnfs_prog, nqnfs_vers;
61 int nqsrv_clockskew = NQ_CLOCKSKEW;
62 int nqsrv_writeslack = NQ_WRITESLACK;
63 int nqsrv_maxlease = NQ_MAXLEASE;
64 int nqsrv_maxnumlease = NQ_MAXNUMLEASE;
65 void nqsrv_instimeq(), nqsrv_send_eviction(), nfs_sndunlock();
66 void nqsrv_unlocklease(), nqsrv_waitfor_expiry(), nfsrv_slpderef();
67 void nqsrv_addhost(), nqsrv_locklease(), nqnfs_serverd();
68 void nqnfs_clientlease();
69 struct mbuf *nfsm_rpchead();
70 
71 /*
72  * Signifies which rpcs can have piggybacked lease requests
73  */
74 int nqnfs_piggy[NFS_NPROCS] = {
75 	0,
76 	NQL_READ,
77 	NQL_WRITE,
78 	0,
79 	NQL_READ,
80 	NQL_READ,
81 	NQL_READ,
82 	0,
83 	NQL_WRITE,
84 	0,
85 	0,
86 	0,
87 	0,
88 	0,
89 	0,
90 	0,
91 	NQL_READ,
92 	0,
93 	NQL_READ,
94 	0,
95 	0,
96 	0,
97 	0,
98 };
99 
100 int nnnnnn = sizeof (struct nqlease);
101 int oooooo = sizeof (struct nfsnode);
102 extern nfstype nfs_type[9];
103 extern struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock;
104 extern struct nfsd nfsd_head;
105 extern int nfsd_waiting;
106 extern struct nfsreq nfsreqh;
107 
108 #define TRUE	1
109 #define	FALSE	0
110 
111 /*
112  * Get or check for a lease for "vp", based on NQL_CHECK flag.
113  * The rules are as follows:
114  * - if a current non-caching lease, reply non-caching
115  * - if a current lease for same host only, extend lease
116  * - if a read cachable lease and a read lease request
117  *	add host to list any reply cachable
118  * - else { set non-cachable for read-write sharing }
119  *	send eviction notice messages to all other hosts that have lease
120  *	wait for lease termination { either by receiving vacated messages
121  *					from all the other hosts or expiry
122  *					via. timeout }
123  *	modify lease to non-cachable
124  * - else if no current lease, issue new one
125  * - reply
126  * - return boolean TRUE iff nam should be m_freem()'d
127  * NB: Since nqnfs_serverd() is called from a timer, any potential tsleep()
128  *     in here must be framed by nqsrv_locklease() and nqsrv_unlocklease().
129  *     nqsrv_locklease() is coded such that at least one of LC_LOCKED and
130  *     LC_WANTED is set whenever a process is tsleeping in it. The exception
131  *     is when a new lease is being allocated, since it is not in the timer
132  *     queue yet. (Ditto for the splsoftclock() and splx(s) calls)
133  */
134 nqsrv_getlease(vp, duration, flags, nd, nam, cachablep, frev, cred)
135 	struct vnode *vp;
136 	u_long *duration;
137 	int flags;
138 	struct nfsd *nd;
139 	struct mbuf *nam;
140 	int *cachablep;
141 	u_quad_t *frev;
142 	struct ucred *cred;
143 {
144 	register struct nqlease *lp, *lq, **lpp;
145 	register struct nqhost *lph;
146 	struct nqlease *tlp;
147 	struct nqm **lphp;
148 	struct vattr vattr;
149 	fhandle_t fh;
150 	int i, ok, error, s;
151 
152 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
153 		return (0);
154 	if (*duration > nqsrv_maxlease)
155 		*duration = nqsrv_maxlease;
156 	if (error = VOP_GETATTR(vp, &vattr, cred, nd->nd_procp))
157 		return (error);
158 	*frev = vattr.va_filerev;
159 	s = splsoftclock();
160 	tlp = vp->v_lease;
161 	if ((flags & NQL_CHECK) == 0)
162 		nfsstats.srvnqnfs_getleases++;
163 	if (tlp == (struct nqlease *)0) {
164 
165 		/*
166 		 * Find the lease by searching the hash list.
167 		 */
168 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
169 		if (error = VFS_VPTOFH(vp, &fh.fh_fid)) {
170 			splx(s);
171 			return (error);
172 		}
173 		lpp = &nqfhead[NQFHHASH(fh.fh_fid.fid_data)];
174 		for (lp = *lpp; lp; lp = lp->lc_fhnext)
175 			if (fh.fh_fsid.val[0] == lp->lc_fsid.val[0] &&
176 			    fh.fh_fsid.val[1] == lp->lc_fsid.val[1] &&
177 			    !bcmp(fh.fh_fid.fid_data, lp->lc_fiddata,
178 				  fh.fh_fid.fid_len - sizeof (long))) {
179 				/* Found it */
180 				lp->lc_vp = vp;
181 				vp->v_lease = lp;
182 				tlp = lp;
183 				break;
184 			}
185 	}
186 	lp = tlp;
187 	if (lp) {
188 		if ((lp->lc_flag & LC_NONCACHABLE) ||
189 		    (lp->lc_morehosts == (struct nqm *)0 &&
190 		     nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host)))
191 			goto doreply;
192 		if ((flags & NQL_READ) && (lp->lc_flag & LC_WRITE)==0) {
193 			if (flags & NQL_CHECK)
194 				goto doreply;
195 			if (nqsrv_cmpnam(nd->nd_slp, nam, &lp->lc_host))
196 				goto doreply;
197 			i = 0;
198 			if (lp->lc_morehosts) {
199 				lph = lp->lc_morehosts->lpm_hosts;
200 				lphp = &lp->lc_morehosts->lpm_next;
201 				ok = 1;
202 			} else {
203 				lphp = &lp->lc_morehosts;
204 				ok = 0;
205 			}
206 			while (ok && (lph->lph_flag & LC_VALID)) {
207 				if (nqsrv_cmpnam(nd->nd_slp, nam, lph))
208 					goto doreply;
209 				if (++i == LC_MOREHOSTSIZ) {
210 					i = 0;
211 					if (*lphp) {
212 						lph = (*lphp)->lpm_hosts;
213 						lphp = &((*lphp)->lpm_next);
214 					} else
215 						ok = 0;
216 				} else
217 					lph++;
218 			}
219 			nqsrv_locklease(lp);
220 			if (!ok) {
221 				*lphp = (struct nqm *)
222 					malloc(sizeof (struct nqm),
223 						M_NQMHOST, M_WAITOK);
224 				bzero((caddr_t)*lphp, sizeof (struct nqm));
225 				lph = (*lphp)->lpm_hosts;
226 			}
227 			nqsrv_addhost(lph, nd->nd_slp, nam);
228 			nqsrv_unlocklease(lp);
229 		} else {
230 			lp->lc_flag |= LC_NONCACHABLE;
231 			nqsrv_locklease(lp);
232 			nqsrv_send_eviction(vp, lp, nd->nd_slp, nam, cred);
233 			nqsrv_waitfor_expiry(lp);
234 			nqsrv_unlocklease(lp);
235 		}
236 doreply:
237 		/*
238 		 * Update the lease and return
239 		 */
240 		if ((flags & NQL_CHECK) == 0)
241 			nqsrv_instimeq(lp, *duration);
242 		if (lp->lc_flag & LC_NONCACHABLE)
243 			*cachablep = 0;
244 		else {
245 			*cachablep = 1;
246 			if (flags & NQL_WRITE)
247 				lp->lc_flag |= LC_WRITTEN;
248 		}
249 		splx(s);
250 		return (0);
251 	}
252 	splx(s);
253 	if (flags & NQL_CHECK)
254 		return (0);
255 
256 	/*
257 	 * Allocate new lease
258 	 * The value of nqsrv_maxnumlease should be set generously, so that
259 	 * the following "printf" happens infrequently.
260 	 */
261 	if (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease) {
262 		printf("Nqnfs server, too many leases\n");
263 		do {
264 			(void) tsleep((caddr_t)&lbolt, PSOCK,
265 					"nqsrvnuml", 0);
266 		} while (nfsstats.srvnqnfs_leases > nqsrv_maxnumlease);
267 	}
268 	MALLOC(lp, struct nqlease *, sizeof (struct nqlease), M_NQLEASE, M_WAITOK);
269 	bzero((caddr_t)lp, sizeof (struct nqlease));
270 	if (flags & NQL_WRITE)
271 		lp->lc_flag |= (LC_WRITE | LC_WRITTEN);
272 	nqsrv_addhost(&lp->lc_host, nd->nd_slp, nam);
273 	lp->lc_vp = vp;
274 	lp->lc_fsid = fh.fh_fsid;
275 	bcopy(fh.fh_fid.fid_data, lp->lc_fiddata, fh.fh_fid.fid_len - sizeof (long));
276 	if (lq = *lpp)
277 		lq->lc_fhprev = &lp->lc_fhnext;
278 	lp->lc_fhnext = lq;
279 	lp->lc_fhprev = lpp;
280 	*lpp = lp;
281 	vp->v_lease = lp;
282 	s = splsoftclock();
283 	nqsrv_instimeq(lp, *duration);
284 	splx(s);
285 	*cachablep = 1;
286 	if (++nfsstats.srvnqnfs_leases > nfsstats.srvnqnfs_maxleases)
287 		nfsstats.srvnqnfs_maxleases = nfsstats.srvnqnfs_leases;
288 	return (0);
289 }
290 
291 /*
292  * Local lease check for server syscalls.
293  * Just set up args and let nqsrv_getlease() do the rest.
294  */
295 lease_check(ap)
296 	struct vop_lease_args /* {
297 		struct vnode *a_vp;
298 		struct proc *a_p;
299 		struct ucred *a_cred;
300 		int a_flag;
301 	} */ *ap;
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 = ap->a_p;
309 	(void) nqsrv_getlease(ap->a_vp, &duration, NQL_CHECK | ap->a_flag,
310 		&nfsd, (struct mbuf *)0, &cache, &frev, ap->a_cred);
311 	return (0);
312 }
313 
314 /*
315  * Add a host to an nqhost structure for a lease.
316  */
317 void
318 nqsrv_addhost(lph, slp, nam)
319 	register struct nqhost *lph;
320 	struct nfssvc_sock *slp;
321 	struct mbuf *nam;
322 {
323 	register struct sockaddr_in *saddr;
324 
325 	if (slp == NQLOCALSLP)
326 		lph->lph_flag |= (LC_VALID | LC_LOCAL);
327 	else if (slp == nfs_udpsock) {
328 		saddr = mtod(nam, struct sockaddr_in *);
329 		lph->lph_flag |= (LC_VALID | LC_UDP);
330 		lph->lph_inetaddr = saddr->sin_addr.s_addr;
331 		lph->lph_port = saddr->sin_port;
332 	} else if (slp == nfs_cltpsock) {
333 		lph->lph_nam = m_copym(nam, 0, M_COPYALL, M_WAIT);
334 		lph->lph_flag |= (LC_VALID | LC_CLTP);
335 	} else {
336 		lph->lph_flag |= (LC_VALID | LC_SREF);
337 		lph->lph_slp = slp;
338 		slp->ns_sref++;
339 	}
340 }
341 
342 /*
343  * Update the lease expiry time and position it in the timer queue correctly.
344  */
345 void
346 nqsrv_instimeq(lp, duration)
347 	register struct nqlease *lp;
348 	u_long duration;
349 {
350 	register struct nqlease *tlp;
351 	time_t newexpiry;
352 
353 	newexpiry = time.tv_sec + duration + nqsrv_clockskew;
354 	if (lp->lc_expiry == newexpiry)
355 		return;
356 	if (lp->lc_chain1[0])
357 		remque(lp);
358 	lp->lc_expiry = newexpiry;
359 
360 	/*
361 	 * Find where in the queue it should be.
362 	 */
363 	tlp = nqthead.th_chain[1];
364 	while (tlp->lc_expiry > newexpiry && tlp != (struct nqlease *)&nqthead)
365 		tlp = tlp->lc_chain1[1];
366 	if (tlp == nqthead.th_chain[1])
367 		NQSTORENOVRAM(newexpiry);
368 	insque(lp, tlp);
369 }
370 
371 /*
372  * Compare the requesting host address with the lph entry in the lease.
373  * Return true iff it is the same.
374  * This is somewhat messy due to the union in the nqhost structure.
375  * The local host is indicated by the special value of NQLOCALSLP for slp.
376  */
377 nqsrv_cmpnam(slp, nam, lph)
378 	register struct nfssvc_sock *slp;
379 	struct mbuf *nam;
380 	register struct nqhost *lph;
381 {
382 	register struct sockaddr_in *saddr;
383 	struct mbuf *addr;
384 	union nethostaddr lhaddr;
385 	int ret;
386 
387 	if (slp == NQLOCALSLP) {
388 		if (lph->lph_flag & LC_LOCAL)
389 			return (1);
390 		else
391 			return (0);
392 	}
393 	if (slp == nfs_udpsock || slp == nfs_cltpsock)
394 		addr = nam;
395 	else
396 		addr = slp->ns_nam;
397 	if (lph->lph_flag & LC_UDP)
398 		ret = netaddr_match(AF_INET, &lph->lph_haddr, addr);
399 	else if (lph->lph_flag & LC_CLTP)
400 		ret = netaddr_match(AF_ISO, &lph->lph_claddr, addr);
401 	else {
402 		if ((lph->lph_slp->ns_flag & SLP_VALID) == 0)
403 			return (0);
404 		saddr = mtod(lph->lph_slp->ns_nam, struct sockaddr_in *);
405 		if (saddr->sin_family == AF_INET)
406 			lhaddr.had_inetaddr = saddr->sin_addr.s_addr;
407 		else
408 			lhaddr.had_nam = lph->lph_slp->ns_nam;
409 		ret = netaddr_match(saddr->sin_family, &lhaddr, addr);
410 	}
411 	return (ret);
412 }
413 
414 /*
415  * Send out eviction notice messages to all other hosts for the lease.
416  */
417 void
418 nqsrv_send_eviction(vp, lp, slp, nam, cred)
419 	struct vnode *vp;
420 	register struct nqlease *lp;
421 	struct nfssvc_sock *slp;
422 	struct mbuf *nam;
423 	struct ucred *cred;
424 {
425 	register struct nqhost *lph = &lp->lc_host;
426 	register struct mbuf *m;
427 	register int siz;
428 	struct nqm *lphnext = lp->lc_morehosts;
429 	struct mbuf *mreq, *mb, *mb2, *nam2, *mheadend;
430 	struct socket *so;
431 	struct sockaddr_in *saddr;
432 	fhandle_t *fhp;
433 	caddr_t bpos, cp;
434 	u_long xid;
435 	int len = 1, ok = 1, i = 0;
436 	int sotype, *solockp;
437 
438 	while (ok && (lph->lph_flag & LC_VALID)) {
439 		if (nqsrv_cmpnam(slp, nam, lph))
440 			lph->lph_flag |= LC_VACATED;
441 		else if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
442 			if (lph->lph_flag & LC_UDP) {
443 				MGET(nam2, M_WAIT, MT_SONAME);
444 				saddr = mtod(nam2, struct sockaddr_in *);
445 				nam2->m_len = saddr->sin_len =
446 					sizeof (struct sockaddr_in);
447 				saddr->sin_family = AF_INET;
448 				saddr->sin_addr.s_addr = lph->lph_inetaddr;
449 				saddr->sin_port = lph->lph_port;
450 				so = nfs_udpsock->ns_so;
451 			} else if (lph->lph_flag & LC_CLTP) {
452 				nam2 = lph->lph_nam;
453 				so = nfs_cltpsock->ns_so;
454 			} else if (lph->lph_slp->ns_flag & SLP_VALID) {
455 				nam2 = (struct mbuf *)0;
456 				so = lph->lph_slp->ns_so;
457 			} else
458 				goto nextone;
459 			sotype = so->so_type;
460 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
461 				solockp = &lph->lph_slp->ns_solock;
462 			else
463 				solockp = (int *)0;
464 			nfsm_reqhead((struct vnode *)0, NQNFSPROC_EVICTED,
465 				NFSX_FH);
466 			nfsm_build(cp, caddr_t, NFSX_FH);
467 			bzero(cp, NFSX_FH);
468 			fhp = (fhandle_t *)cp;
469 			fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
470 			VFS_VPTOFH(vp, &fhp->fh_fid);
471 			m = mreq;
472 			siz = 0;
473 			while (m) {
474 				siz += m->m_len;
475 				m = m->m_next;
476 			}
477 			if (siz <= 0 || siz > NFS_MAXPACKET) {
478 				printf("mbuf siz=%d\n",siz);
479 				panic("Bad nfs svc reply");
480 			}
481 			m = nfsm_rpchead(cred, TRUE, NQNFSPROC_EVICTED,
482 				RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
483 				mreq, siz, &mheadend, &xid);
484 			/*
485 			 * For stream protocols, prepend a Sun RPC
486 			 * Record Mark.
487 			 */
488 			if (sotype == SOCK_STREAM) {
489 				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
490 				*mtod(m, u_long *) = htonl(0x80000000 |
491 					(m->m_pkthdr.len - NFSX_UNSIGNED));
492 			}
493 			if (((lph->lph_flag & (LC_UDP | LC_CLTP)) == 0 &&
494 			    (lph->lph_slp->ns_flag & SLP_VALID) == 0) ||
495 			    (solockp && (*solockp & NFSMNT_SNDLOCK)))
496 				m_freem(m);
497 			else {
498 				if (solockp)
499 					*solockp |= NFSMNT_SNDLOCK;
500 				(void) nfs_send(so, nam2, m,
501 						(struct nfsreq *)0);
502 				if (solockp)
503 					nfs_sndunlock(solockp);
504 			}
505 			if (lph->lph_flag & LC_UDP)
506 				MFREE(nam2, m);
507 		}
508 nextone:
509 		if (++i == len) {
510 			if (lphnext) {
511 				i = 0;
512 				len = LC_MOREHOSTSIZ;
513 				lph = lphnext->lpm_hosts;
514 				lphnext = lphnext->lpm_next;
515 			} else
516 				ok = 0;
517 		} else
518 			lph++;
519 	}
520 }
521 
522 /*
523  * Wait for the lease to expire.
524  * This will occur when all clients have sent "vacated" messages to
525  * this server OR when it expires do to timeout.
526  */
527 void
528 nqsrv_waitfor_expiry(lp)
529 	register struct nqlease *lp;
530 {
531 	register struct nqhost *lph;
532 	register int i;
533 	struct nqm *lphnext;
534 	int len, ok;
535 
536 tryagain:
537 	if (time.tv_sec > lp->lc_expiry)
538 		return;
539 	lph = &lp->lc_host;
540 	lphnext = lp->lc_morehosts;
541 	len = 1;
542 	i = 0;
543 	ok = 1;
544 	while (ok && (lph->lph_flag & LC_VALID)) {
545 		if ((lph->lph_flag & (LC_LOCAL | LC_VACATED)) == 0) {
546 			lp->lc_flag |= LC_EXPIREDWANTED;
547 			(void) tsleep((caddr_t)&lp->lc_flag, PSOCK,
548 					"nqexp", 0);
549 			goto tryagain;
550 		}
551 		if (++i == len) {
552 			if (lphnext) {
553 				i = 0;
554 				len = LC_MOREHOSTSIZ;
555 				lph = lphnext->lpm_hosts;
556 				lphnext = lphnext->lpm_next;
557 			} else
558 				ok = 0;
559 		} else
560 			lph++;
561 	}
562 }
563 
564 /*
565  * Nqnfs server timer that maintains the server lease queue.
566  * Scan the lease queue for expired entries:
567  * - when one is found, wakeup anyone waiting for it
568  *   else dequeue and free
569  */
570 void
571 nqnfs_serverd()
572 {
573 	register struct nqlease *lp, *lq;
574 	register struct nqhost *lph;
575 	struct nqlease *nextlp;
576 	struct nqm *lphnext, *olphnext;
577 	struct mbuf *n;
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,
686 	    TRUE, &vp, cred, nfsd->nd_slp, nam, &rdonly))
687 		nfsm_reply(0);
688 	if (rdonly && flags == NQL_WRITE) {
689 		error = EROFS;
690 		nfsm_reply(0);
691 	}
692 	(void) nqsrv_getlease(vp, &nfsd->nd_duration, flags, nfsd,
693 		nam, &cache, &frev, cred);
694 	error = VOP_GETATTR(vp, vap, cred, nfsd->nd_procp);
695 	vput(vp);
696 	nfsm_reply(NFSX_NQFATTR + 4*NFSX_UNSIGNED);
697 	nfsm_build(tl, u_long *, 4*NFSX_UNSIGNED);
698 	*tl++ = txdr_unsigned(cache);
699 	*tl++ = txdr_unsigned(nfsd->nd_duration);
700 	txdr_hyper(&frev, tl);
701 	nfsm_build(fp, struct nfsv2_fattr *, NFSX_NQFATTR);
702 	nfsm_srvfillattr;
703 	nfsm_srvdone;
704 }
705 
706 /*
707  * Called from nfssvc_nfsd() when a "vacated" message is received from a
708  * client. Find the entry and expire it.
709  */
710 nqnfsrv_vacated(nfsd, mrep, md, dpos, cred, nam, mrq)
711 	struct nfsd *nfsd;
712 	struct mbuf *mrep, *md;
713 	caddr_t dpos;
714 	struct ucred *cred;
715 	struct mbuf *nam, **mrq;
716 {
717 	register struct nqlease *lp;
718 	register struct nqhost *lph;
719 	struct nqlease *tlp = (struct nqlease *)0;
720 	nfsv2fh_t nfh;
721 	fhandle_t *fhp;
722 	register u_long *tl;
723 	register long t1;
724 	struct nqm *lphnext;
725 	int error = 0, i, len, ok, gotit = 0;
726 	char *cp2;
727 
728 	fhp = &nfh.fh_generic;
729 	nfsm_srvmtofh(fhp);
730 	m_freem(mrep);
731 	/*
732 	 * Find the lease by searching the hash list.
733 	 */
734 	for (lp = nqfhead[NQFHHASH(fhp->fh_fid.fid_data)]; lp;
735 	     lp = lp->lc_fhnext)
736 		if (fhp->fh_fsid.val[0] == lp->lc_fsid.val[0] &&
737 		    fhp->fh_fsid.val[1] == lp->lc_fsid.val[1] &&
738 		    !bcmp(fhp->fh_fid.fid_data, lp->lc_fiddata,
739 			  MAXFIDSZ)) {
740 			/* Found it */
741 			tlp = lp;
742 			break;
743 		}
744 	if (tlp) {
745 		lp = tlp;
746 		len = 1;
747 		i = 0;
748 		lph = &lp->lc_host;
749 		lphnext = lp->lc_morehosts;
750 		ok = 1;
751 		while (ok && (lph->lph_flag & LC_VALID)) {
752 			if (nqsrv_cmpnam(nfsd->nd_slp, nam, lph)) {
753 				lph->lph_flag |= LC_VACATED;
754 				gotit++;
755 				break;
756 			}
757 			if (++i == len) {
758 				if (lphnext) {
759 					len = LC_MOREHOSTSIZ;
760 					i = 0;
761 					lph = lphnext->lpm_hosts;
762 					lphnext = lphnext->lpm_next;
763 				} else
764 					ok = 0;
765 			} else
766 				lph++;
767 		}
768 		if ((lp->lc_flag & LC_EXPIREDWANTED) && gotit) {
769 			lp->lc_flag &= ~LC_EXPIREDWANTED;
770 			wakeup((caddr_t)&lp->lc_flag);
771 		}
772 nfsmout:
773 		return (EPERM);
774 	}
775 	return (EPERM);
776 }
777 
778 /*
779  * Client get lease rpc function.
780  */
781 nqnfs_getlease(vp, rwflag, cred, p)
782 	register struct vnode *vp;
783 	int rwflag;
784 	struct ucred *cred;
785 	struct proc *p;
786 {
787 	register u_long *tl;
788 	register caddr_t cp;
789 	register long t1;
790 	register struct nfsnode *np;
791 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
792 	caddr_t bpos, dpos, cp2;
793 	time_t reqtime;
794 	int error = 0;
795 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
796 	int cachable;
797 	u_quad_t frev;
798 
799 	nfsstats.rpccnt[NQNFSPROC_GETLEASE]++;
800 	mb = mreq = nfsm_reqh(vp, NQNFSPROC_GETLEASE, NFSX_FH+2*NFSX_UNSIGNED,
801 		 &bpos);
802 	nfsm_fhtom(vp);
803 	nfsm_build(tl, u_long *, 2*NFSX_UNSIGNED);
804 	*tl++ = txdr_unsigned(rwflag);
805 	*tl = txdr_unsigned(nmp->nm_leaseterm);
806 	reqtime = time.tv_sec;
807 	nfsm_request(vp, NQNFSPROC_GETLEASE, p, cred);
808 	np = VTONFS(vp);
809 	nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED);
810 	cachable = fxdr_unsigned(int, *tl++);
811 	reqtime += fxdr_unsigned(int, *tl++);
812 	if (reqtime > time.tv_sec) {
813 		fxdr_hyper(tl, &frev);
814 		nqnfs_clientlease(nmp, np, rwflag, cachable, reqtime, frev);
815 		nfsm_loadattr(vp, (struct vattr *)0);
816 	} else
817 		error = NQNFS_EXPIRED;
818 	nfsm_reqdone;
819 	return (error);
820 }
821 
822 /*
823  * Client vacated message function.
824  */
825 nqnfs_vacated(vp, cred)
826 	register struct vnode *vp;
827 	struct ucred *cred;
828 {
829 	register caddr_t cp;
830 	register struct mbuf *m;
831 	register int i;
832 	caddr_t bpos;
833 	u_long xid;
834 	int error = 0;
835 	struct mbuf *mreq, *mb, *mb2, *mheadend;
836 	struct nfsmount *nmp;
837 	struct nfsreq myrep;
838 
839 	nmp = VFSTONFS(vp->v_mount);
840 	nfsstats.rpccnt[NQNFSPROC_VACATED]++;
841 	nfsm_reqhead(vp, NQNFSPROC_VACATED, NFSX_FH);
842 	nfsm_fhtom(vp);
843 	m = mreq;
844 	i = 0;
845 	while (m) {
846 		i += m->m_len;
847 		m = m->m_next;
848 	}
849 	m = nfsm_rpchead(cred, TRUE, NQNFSPROC_VACATED,
850 		RPCAUTH_UNIX, 5*NFSX_UNSIGNED, (char *)0,
851 		mreq, i, &mheadend, &xid);
852 	if (nmp->nm_sotype == SOCK_STREAM) {
853 		M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
854 		*mtod(m, u_long *) = htonl(0x80000000 | (m->m_pkthdr.len -
855 			NFSX_UNSIGNED));
856 	}
857 	myrep.r_flags = 0;
858 	myrep.r_nmp = nmp;
859 	if (nmp->nm_soflags & PR_CONNREQUIRED)
860 		(void) nfs_sndlock(&nmp->nm_flag, (struct nfsreq *)0);
861 	(void) nfs_send(nmp->nm_so, nmp->nm_nam, m, &myrep);
862 	if (nmp->nm_soflags & PR_CONNREQUIRED)
863 		nfs_sndunlock(&nmp->nm_flag);
864 	return (error);
865 }
866 
867 /*
868  * Called for client side callbacks
869  */
870 nqnfs_callback(nmp, mrep, md, dpos)
871 	struct nfsmount *nmp;
872 	struct mbuf *mrep, *md;
873 	caddr_t dpos;
874 {
875 	register struct vnode *vp;
876 	register u_long *tl;
877 	register long t1;
878 	nfsv2fh_t nfh;
879 	fhandle_t *fhp;
880 	struct nfsnode *np;
881 	struct nfsd nd;
882 	int error;
883 	char *cp2;
884 
885 	nd.nd_mrep = mrep;
886 	nd.nd_md = md;
887 	nd.nd_dpos = dpos;
888 	if (error = nfs_getreq(&nd, FALSE))
889 		return (error);
890 	md = nd.nd_md;
891 	dpos = nd.nd_dpos;
892 	if (nd.nd_procnum != NQNFSPROC_EVICTED) {
893 		m_freem(mrep);
894 		return (EPERM);
895 	}
896 	fhp = &nfh.fh_generic;
897 	nfsm_srvmtofh(fhp);
898 	m_freem(mrep);
899 	if (error = nfs_nget(nmp->nm_mountp, fhp, &np))
900 		return (error);
901 	vp = NFSTOV(np);
902 	if (np->n_tnext) {
903 		np->n_expiry = 0;
904 		np->n_flag |= NQNFSEVICTED;
905 		if (np->n_tprev != (struct nfsnode *)nmp) {
906 			if (np->n_tnext == (struct nfsnode *)nmp)
907 				nmp->nm_tprev = np->n_tprev;
908 			else
909 				np->n_tnext->n_tprev = np->n_tprev;
910 			np->n_tprev->n_tnext = np->n_tnext;
911 			np->n_tnext = nmp->nm_tnext;
912 			nmp->nm_tnext = np;
913 			np->n_tprev = (struct nfsnode *)nmp;
914 			if (np->n_tnext == (struct nfsnode *)nmp)
915 				nmp->nm_tprev = np;
916 			else
917 				np->n_tnext->n_tprev = np;
918 		}
919 	}
920 	vrele(vp);
921 	nfsm_srvdone;
922 }
923 
924 /*
925  * Nqnfs client helper daemon. Runs once a second to expire leases.
926  * It also get authorization strings for "kerb" mounts.
927  * It must start at the beginning of the list again after any potential
928  * "sleep" since nfs_reclaim() called from vclean() can pull a node off
929  * the list asynchronously.
930  */
931 nqnfs_clientd(nmp, cred, ncd, flag, argp, p)
932 	register struct nfsmount *nmp;
933 	struct ucred *cred;
934 	struct nfsd_cargs *ncd;
935 	int flag;
936 	caddr_t argp;
937 	struct proc *p;
938 {
939 	register struct nfsnode *np;
940 	struct vnode *vp;
941 	struct nfsreq myrep;
942 	int error, vpid;
943 
944 	/*
945 	 * First initialize some variables
946 	 */
947 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
948 	nqnfs_vers = txdr_unsigned(NQNFS_VER1);
949 
950 	/*
951 	 * If an authorization string is being passed in, get it.
952 	 */
953 	if ((flag & NFSSVC_GOTAUTH) &&
954 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT)) == 0) {
955 		if (nmp->nm_flag & NFSMNT_HASAUTH)
956 			panic("cld kerb");
957 		if ((flag & NFSSVC_AUTHINFAIL) == 0) {
958 			if (ncd->ncd_authlen <= RPCAUTH_MAXSIZ &&
959 				copyin(ncd->ncd_authstr, nmp->nm_authstr,
960 				ncd->ncd_authlen) == 0) {
961 				nmp->nm_authtype = ncd->ncd_authtype;
962 				nmp->nm_authlen = ncd->ncd_authlen;
963 			} else
964 				nmp->nm_flag |= NFSMNT_AUTHERR;
965 		} else
966 			nmp->nm_flag |= NFSMNT_AUTHERR;
967 		nmp->nm_flag |= NFSMNT_HASAUTH;
968 		wakeup((caddr_t)&nmp->nm_authlen);
969 	} else
970 		nmp->nm_flag |= NFSMNT_WAITAUTH;
971 
972 	/*
973 	 * Loop every second updating queue until there is a termination sig.
974 	 */
975 	while ((nmp->nm_flag & NFSMNT_DISMNT) == 0) {
976 	    if (nmp->nm_flag & NFSMNT_NQNFS) {
977 		/*
978 		 * If there are no outstanding requests (and therefore no
979 		 * processes in nfs_reply) and there is data in the receive
980 		 * queue, poke for callbacks.
981 		 */
982 		if (nfsreqh.r_next == &nfsreqh && nmp->nm_so &&
983 		    nmp->nm_so->so_rcv.sb_cc > 0) {
984 		    myrep.r_flags = R_GETONEREP;
985 		    myrep.r_nmp = nmp;
986 		    myrep.r_mrep = (struct mbuf *)0;
987 		    myrep.r_procp = (struct proc *)0;
988 		    (void) nfs_reply(&myrep);
989 		}
990 
991 		/*
992 		 * Loop through the leases, updating as required.
993 		 */
994 		np = nmp->nm_tnext;
995 		while (np != (struct nfsnode *)nmp &&
996 		       (nmp->nm_flag & NFSMNT_DISMINPROG) == 0) {
997 			vp = NFSTOV(np);
998 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash2");
999 			vpid = vp->v_id;
1000 			if (np->n_expiry < time.tv_sec) {
1001 			   if (vget(vp, 1) == 0) {
1002 			     nmp->nm_inprog = vp;
1003 			     if (vpid == vp->v_id) {
1004 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash3");
1005 				if (np->n_tnext == (struct nfsnode *)nmp)
1006 					nmp->nm_tprev = np->n_tprev;
1007 				else
1008 					np->n_tnext->n_tprev = np->n_tprev;
1009 				if (np->n_tprev == (struct nfsnode *)nmp)
1010 					nmp->nm_tnext = np->n_tnext;
1011 				else
1012 					np->n_tprev->n_tnext = np->n_tnext;
1013 				np->n_tnext = (struct nfsnode *)0;
1014 				if ((np->n_flag & (NMODIFIED | NQNFSEVICTED))
1015 				    && vp->v_type == VREG) {
1016 					if (np->n_flag & NQNFSEVICTED) {
1017 						(void) nfs_vinvalbuf(vp,
1018 						       V_SAVE, cred, p, 0);
1019 						np->n_flag &= ~NQNFSEVICTED;
1020 						(void) nqnfs_vacated(vp, cred);
1021 					} else {
1022 						(void) VOP_FSYNC(vp, cred,
1023 						    MNT_WAIT, p);
1024 						np->n_flag &= ~NMODIFIED;
1025 					}
1026 				}
1027 			      }
1028 			      vrele(vp);
1029 			      nmp->nm_inprog = NULLVP;
1030 			    }
1031 			    if (np != nmp->nm_tnext)
1032 				np = nmp->nm_tnext;
1033 			    else
1034 				break;
1035 			} else if ((np->n_expiry - NQ_RENEWAL) < time.tv_sec) {
1036 			    if ((np->n_flag & (NQNFSWRITE | NQNFSNONCACHE))
1037 				 == NQNFSWRITE && vp->v_dirtyblkhd.lh_first &&
1038 				 vget(vp, 1) == 0) {
1039 				 nmp->nm_inprog = vp;
1040 if (vp->v_mount->mnt_stat.f_fsid.val[1] != MOUNT_NFS) panic("trash4");
1041 				 if (vpid == vp->v_id &&
1042 				     nqnfs_getlease(vp, NQL_WRITE, cred, p)==0)
1043 					np->n_brev = np->n_lrev;
1044 				 vrele(vp);
1045 				 nmp->nm_inprog = NULLVP;
1046 			    }
1047 			    if (np != nmp->nm_tnext)
1048 				np = nmp->nm_tnext;
1049 			    else
1050 				break;
1051 			} else
1052 				break;
1053 		}
1054 	    }
1055 
1056 	    /*
1057 	     * Get an authorization string, if required.
1058 	     */
1059 	    if ((nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_DISMNT | NFSMNT_HASAUTH)) == 0) {
1060 		ncd->ncd_authuid = nmp->nm_authuid;
1061 		if (copyout((caddr_t)ncd, argp, sizeof (struct nfsd_cargs)))
1062 			nmp->nm_flag |= NFSMNT_WAITAUTH;
1063 		else
1064 			return (ENEEDAUTH);
1065 	    }
1066 
1067 	    /*
1068 	     * Wait a bit (no pun) and do it again.
1069 	     */
1070 	    if ((nmp->nm_flag & NFSMNT_DISMNT) == 0 &&
1071 		(nmp->nm_flag & (NFSMNT_WAITAUTH | NFSMNT_HASAUTH))) {
1072 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
1073 			"nqnfstimr", hz / 3);
1074 		    if (error == EINTR || error == ERESTART)
1075 			(void) dounmount(nmp->nm_mountp, 0, p);
1076 	    }
1077 	}
1078 	free((caddr_t)nmp, M_NFSMNT);
1079 	if (error == EWOULDBLOCK)
1080 		error = 0;
1081 	return (error);
1082 }
1083 
1084 /*
1085  * Adjust all timer queue expiry times when the time of day clock is changed.
1086  * Called from the settimeofday() syscall.
1087  */
1088 void
1089 lease_updatetime(deltat)
1090 	register int deltat;
1091 {
1092 	register struct nqlease *lp;
1093 	register struct nfsnode *np;
1094 	struct mount *mp;
1095 	struct nfsmount *nmp;
1096 	int s;
1097 
1098 	if (nqnfsstarttime != 0)
1099 		nqnfsstarttime += deltat;
1100 	s = splsoftclock();
1101 	lp = nqthead.th_chain[0];
1102 	while (lp != (struct nqlease *)&nqthead) {
1103 		lp->lc_expiry += deltat;
1104 		lp = lp->lc_chain1[0];
1105 	}
1106 	splx(s);
1107 
1108 	/*
1109 	 * Search the mount list for all nqnfs mounts and do their timer
1110 	 * queues.
1111 	 */
1112 	for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) {
1113 		if (mp->mnt_stat.f_fsid.val[1] == MOUNT_NFS) {
1114 			nmp = VFSTONFS(mp);
1115 			if (nmp->nm_flag & NFSMNT_NQNFS) {
1116 				np = nmp->nm_tnext;
1117 				while (np != (struct nfsnode *)nmp) {
1118 					np->n_expiry += deltat;
1119 					np = np->n_tnext;
1120 				}
1121 			}
1122 		}
1123 	}
1124 }
1125 
1126 /*
1127  * Lock a server lease.
1128  */
1129 void
1130 nqsrv_locklease(lp)
1131 	struct nqlease *lp;
1132 {
1133 
1134 	while (lp->lc_flag & LC_LOCKED) {
1135 		lp->lc_flag |= LC_WANTED;
1136 		(void) tsleep((caddr_t)lp, PSOCK, "nqlc", 0);
1137 	}
1138 	lp->lc_flag |= LC_LOCKED;
1139 	lp->lc_flag &= ~LC_WANTED;
1140 }
1141 
1142 /*
1143  * Unlock a server lease.
1144  */
1145 void
1146 nqsrv_unlocklease(lp)
1147 	struct nqlease *lp;
1148 {
1149 
1150 	lp->lc_flag &= ~LC_LOCKED;
1151 	if (lp->lc_flag & LC_WANTED)
1152 		wakeup((caddr_t)lp);
1153 }
1154 
1155 /*
1156  * Update a client lease.
1157  */
1158 void
1159 nqnfs_clientlease(nmp, np, rwflag, cachable, expiry, frev)
1160 	register struct nfsmount *nmp;
1161 	register struct nfsnode *np;
1162 	int rwflag, cachable;
1163 	time_t expiry;
1164 	u_quad_t frev;
1165 {
1166 	register struct nfsnode *tp;
1167 
1168 	if (np->n_tnext) {
1169 		if (np->n_tnext == (struct nfsnode *)nmp)
1170 			nmp->nm_tprev = np->n_tprev;
1171 		else
1172 			np->n_tnext->n_tprev = np->n_tprev;
1173 		if (np->n_tprev == (struct nfsnode *)nmp)
1174 			nmp->nm_tnext = np->n_tnext;
1175 		else
1176 			np->n_tprev->n_tnext = np->n_tnext;
1177 		if (rwflag == NQL_WRITE)
1178 			np->n_flag |= NQNFSWRITE;
1179 	} else if (rwflag == NQL_READ)
1180 		np->n_flag &= ~NQNFSWRITE;
1181 	else
1182 		np->n_flag |= NQNFSWRITE;
1183 	if (cachable)
1184 		np->n_flag &= ~NQNFSNONCACHE;
1185 	else
1186 		np->n_flag |= NQNFSNONCACHE;
1187 	np->n_expiry = expiry;
1188 	np->n_lrev = frev;
1189 	tp = nmp->nm_tprev;
1190 	while (tp != (struct nfsnode *)nmp && tp->n_expiry > np->n_expiry)
1191 		tp = tp->n_tprev;
1192 	if (tp == (struct nfsnode *)nmp) {
1193 		np->n_tnext = nmp->nm_tnext;
1194 		nmp->nm_tnext = np;
1195 	} else {
1196 		np->n_tnext = tp->n_tnext;
1197 		tp->n_tnext = np;
1198 	}
1199 	np->n_tprev = tp;
1200 	if (np->n_tnext == (struct nfsnode *)nmp)
1201 		nmp->nm_tprev = np;
1202 	else
1203 		np->n_tnext->n_tprev = np;
1204 }
1205