xref: /openbsd-src/sys/nfs/nfs_srvcache.c (revision b725ae7711052a2233e31a66fefb8a752c388d7a)
1 /*	$OpenBSD: nfs_srvcache.c,v 1.10 2003/06/02 23:28:19 millert Exp $	*/
2 /*	$NetBSD: nfs_srvcache.c,v 1.12 1996/02/18 11:53:49 fvdl Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)nfs_srvcache.c	8.3 (Berkeley) 3/30/95
36  */
37 
38 /*
39  * Reference: Chet Juszczak, "Improving the Performance and Correctness
40  *		of an NFS Server", in Proc. Winter 1989 USENIX Conference,
41  *		pages 53-63. San Diego, February 1989.
42  */
43 #include <sys/param.h>
44 #include <sys/vnode.h>
45 #include <sys/mount.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/mbuf.h>
50 #include <sys/malloc.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 
54 #include <netinet/in.h>
55 #ifdef ISO
56 #include <netiso/iso.h>
57 #endif
58 #include <nfs/nfsm_subs.h>
59 #include <nfs/rpcv2.h>
60 #include <nfs/nfsproto.h>
61 #include <nfs/nfs.h>
62 #include <nfs/nfsrvcache.h>
63 #include <nfs/nfs_var.h>
64 
65 extern struct nfsstats nfsstats;
66 extern int nfsv2_procid[NFS_NPROCS];
67 long numnfsrvcache, desirednfsrvcache = NFSRVCACHESIZ;
68 
69 #define	NFSRCHASH(xid) \
70 	(&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
71 LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
72 TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
73 u_long nfsrvhash;
74 
75 #define TRUE	1
76 #define	FALSE	0
77 
78 #define	NETFAMILY(rp) \
79 		(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
80 
81 /*
82  * Static array that defines which nfs rpc's are nonidempotent
83  */
84 int nonidempotent[NFS_NPROCS] = {
85 	FALSE,
86 	FALSE,
87 	TRUE,
88 	FALSE,
89 	FALSE,
90 	FALSE,
91 	FALSE,
92 	TRUE,
93 	TRUE,
94 	TRUE,
95 	TRUE,
96 	TRUE,
97 	TRUE,
98 	TRUE,
99 	TRUE,
100 	TRUE,
101 	FALSE,
102 	FALSE,
103 	FALSE,
104 	FALSE,
105 	FALSE,
106 	FALSE,
107 	FALSE,
108 	FALSE,
109 	FALSE,
110 	FALSE,
111 };
112 
113 /* True iff the rpc reply is an nfs status ONLY! */
114 static int nfsv2_repstat[NFS_NPROCS] = {
115 	FALSE,
116 	FALSE,
117 	FALSE,
118 	FALSE,
119 	FALSE,
120 	FALSE,
121 	FALSE,
122 	FALSE,
123 	FALSE,
124 	FALSE,
125 	TRUE,
126 	TRUE,
127 	TRUE,
128 	TRUE,
129 	FALSE,
130 	TRUE,
131 	FALSE,
132 	FALSE,
133 };
134 
135 /*
136  * Initialize the server request cache list
137  */
138 void
139 nfsrv_initcache()
140 {
141 
142 	nfsrvhashtbl = hashinit(desirednfsrvcache, M_NFSD, M_WAITOK, &nfsrvhash);
143 	TAILQ_INIT(&nfsrvlruhead);
144 }
145 
146 /*
147  * Look for the request in the cache
148  * If found then
149  *    return action and optionally reply
150  * else
151  *    insert it in the cache
152  *
153  * The rules are as follows:
154  * - if in progress, return DROP request
155  * - if completed within DELAY of the current time, return DROP it
156  * - if completed a longer time ago return REPLY if the reply was cached or
157  *   return DOIT
158  * Update/add new request at end of lru list
159  */
160 int
161 nfsrv_getcache(nd, slp, repp)
162 	struct nfsrv_descript *nd;
163 	struct nfssvc_sock *slp;
164 	struct mbuf **repp;
165 {
166 	struct nfsrvcache *rp;
167 	struct mbuf *mb;
168 	struct sockaddr_in *saddr;
169 	caddr_t bpos;
170 	int ret;
171 
172 	/*
173 	 * Don't cache recent requests for reliable transport protocols.
174 	 * (Maybe we should for the case of a reconnect, but..)
175 	 */
176 	if (!nd->nd_nam2)
177 		return (RC_DOIT);
178 loop:
179 	for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != NULL;
180 	    rp = LIST_NEXT(rp, rc_hash)) {
181 	    if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
182 		netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
183 			if ((rp->rc_flag & RC_LOCKED) != 0) {
184 				rp->rc_flag |= RC_WANTED;
185 				(void) tsleep((caddr_t)rp, PZERO-1, "nfsrc", 0);
186 				goto loop;
187 			}
188 			rp->rc_flag |= RC_LOCKED;
189 			/* If not at end of LRU chain, move it there */
190 			if (TAILQ_NEXT(rp, rc_lru)) {
191 				TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
192 				TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
193 			}
194 			if (rp->rc_state == RC_UNUSED)
195 				panic("nfsrv cache");
196 			if (rp->rc_state == RC_INPROG) {
197 				nfsstats.srvcache_inproghits++;
198 				ret = RC_DROPIT;
199 			} else if (rp->rc_flag & RC_REPSTATUS) {
200 				nfsstats.srvcache_nonidemdonehits++;
201 				nfs_rephead(0, nd, slp, rp->rc_status,
202 				   (u_quad_t *)0, repp, &mb, &bpos);
203 				ret = RC_REPLY;
204 			} else if (rp->rc_flag & RC_REPMBUF) {
205 				nfsstats.srvcache_nonidemdonehits++;
206 				*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
207 						M_WAIT);
208 				ret = RC_REPLY;
209 			} else {
210 				nfsstats.srvcache_idemdonehits++;
211 				rp->rc_state = RC_INPROG;
212 				ret = RC_DOIT;
213 			}
214 			rp->rc_flag &= ~RC_LOCKED;
215 			if (rp->rc_flag & RC_WANTED) {
216 				rp->rc_flag &= ~RC_WANTED;
217 				wakeup((caddr_t)rp);
218 			}
219 			return (ret);
220 		}
221 	}
222 	nfsstats.srvcache_misses++;
223 	if (numnfsrvcache < desirednfsrvcache) {
224 		rp = (struct nfsrvcache *)malloc((u_long)sizeof *rp,
225 		    M_NFSD, M_WAITOK);
226 		bzero((char *)rp, sizeof *rp);
227 		numnfsrvcache++;
228 		rp->rc_flag = RC_LOCKED;
229 	} else {
230 		rp = TAILQ_FIRST(&nfsrvlruhead);
231 		while ((rp->rc_flag & RC_LOCKED) != 0) {
232 			rp->rc_flag |= RC_WANTED;
233 			(void) tsleep((caddr_t)rp, PZERO-1, "nfsrc", 0);
234 			rp = TAILQ_FIRST(&nfsrvlruhead);
235 		}
236 		rp->rc_flag |= RC_LOCKED;
237 		LIST_REMOVE(rp, rc_hash);
238 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
239 		if (rp->rc_flag & RC_REPMBUF)
240 			m_freem(rp->rc_reply);
241 		if (rp->rc_flag & RC_NAM)
242 			MFREE(rp->rc_nam, mb);
243 		rp->rc_flag &= (RC_LOCKED | RC_WANTED);
244 	}
245 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
246 	rp->rc_state = RC_INPROG;
247 	rp->rc_xid = nd->nd_retxid;
248 	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
249 	switch (saddr->sin_family) {
250 	case AF_INET:
251 		rp->rc_flag |= RC_INETADDR;
252 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
253 		break;
254 	case AF_ISO:
255 	default:
256 		rp->rc_flag |= RC_NAM;
257 		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
258 		break;
259 	};
260 	rp->rc_proc = nd->nd_procnum;
261 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
262 	rp->rc_flag &= ~RC_LOCKED;
263 	if (rp->rc_flag & RC_WANTED) {
264 		rp->rc_flag &= ~RC_WANTED;
265 		wakeup((caddr_t)rp);
266 	}
267 	return (RC_DOIT);
268 }
269 
270 /*
271  * Update a request cache entry after the rpc has been done
272  */
273 void
274 nfsrv_updatecache(nd, repvalid, repmbuf)
275 	struct nfsrv_descript *nd;
276 	int repvalid;
277 	struct mbuf *repmbuf;
278 {
279 	struct nfsrvcache *rp;
280 
281 	if (!nd->nd_nam2)
282 		return;
283 loop:
284 	for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != NULL;
285 	    rp = LIST_NEXT(rp, rc_hash)) {
286 	    if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
287 		netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
288 			if ((rp->rc_flag & RC_LOCKED) != 0) {
289 				rp->rc_flag |= RC_WANTED;
290 				(void) tsleep((caddr_t)rp, PZERO-1, "nfsrc", 0);
291 				goto loop;
292 			}
293 			rp->rc_flag |= RC_LOCKED;
294 			rp->rc_state = RC_DONE;
295 			/*
296 			 * If we have a valid reply update status and save
297 			 * the reply for non-idempotent rpc's.
298 			 */
299 			if (repvalid && nonidempotent[nd->nd_procnum]) {
300 				if ((nd->nd_flag & ND_NFSV3) == 0 &&
301 				  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
302 					rp->rc_status = nd->nd_repstat;
303 					rp->rc_flag |= RC_REPSTATUS;
304 				} else {
305 					rp->rc_reply = m_copym(repmbuf,
306 						0, M_COPYALL, M_WAIT);
307 					rp->rc_flag |= RC_REPMBUF;
308 				}
309 			}
310 			rp->rc_flag &= ~RC_LOCKED;
311 			if (rp->rc_flag & RC_WANTED) {
312 				rp->rc_flag &= ~RC_WANTED;
313 				wakeup((caddr_t)rp);
314 			}
315 			return;
316 		}
317 	}
318 }
319 
320 /*
321  * Clean out the cache. Called when the last nfsd terminates.
322  */
323 void
324 nfsrv_cleancache()
325 {
326 	struct nfsrvcache *rp, *nextrp;
327 
328 	for (rp = TAILQ_FIRST(&nfsrvlruhead); rp != NULL; rp = nextrp) {
329 		nextrp = TAILQ_NEXT(rp, rc_lru);
330 		LIST_REMOVE(rp, rc_hash);
331 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
332 		free(rp, M_NFSD);
333 	}
334 	numnfsrvcache = 0;
335 }
336