xref: /openbsd/sys/nfs/nfs_srvcache.c (revision e5dd7070)
1 /*	$OpenBSD: nfs_srvcache.c,v 1.29 2019/12/05 10:41:57 mpi 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/mount.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/mbuf.h>
48 #include <sys/malloc.h>
49 #include <sys/socket.h>
50 #include <sys/queue.h>
51 
52 #include <crypto/siphash.h>
53 
54 #include <netinet/in.h>
55 #include <nfs/nfsproto.h>
56 #include <nfs/nfs.h>
57 #include <nfs/nfsrvcache.h>
58 #include <nfs/nfs_var.h>
59 
60 extern struct nfsstats nfsstats;
61 extern int nfsv2_procid[NFS_NPROCS];
62 long numnfsrvcache, desirednfsrvcache = NFSRVCACHESIZ;
63 
64 struct nfsrvcache	*nfsrv_lookupcache(struct nfsrv_descript *);
65 void			 nfsrv_cleanentry(struct nfsrvcache *);
66 
67 LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
68 SIPHASH_KEY nfsrvhashkey;
69 TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
70 u_long nfsrvhash;
71 #define	NFSRCHASH(xid) \
72     (&nfsrvhashtbl[SipHash24(&nfsrvhashkey, &(xid), sizeof(xid)) & nfsrvhash])
73 
74 #define	NETFAMILY(rp)	\
75 	(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_UNSPEC)
76 
77 /* Array that defines which nfs rpc's are nonidempotent */
78 int nonidempotent[NFS_NPROCS] = {
79 	0, 0, 1, 0, 0, 0, 0, 1,
80 	1, 1, 1, 1, 1, 1, 1, 1,
81 	0, 0, 0, 0, 0, 0, 0
82 };
83 
84 /* True iff the rpc reply is an nfs status ONLY! */
85 int nfsv2_repstat[NFS_NPROCS] = {
86 	0, 0, 0, 0, 0, 0, 0, 0,
87 	0, 0, 1, 1, 1, 1, 0, 1,
88 	0, 0
89 };
90 
91 void
92 nfsrv_cleanentry(struct nfsrvcache *rp)
93 {
94 	if ((rp->rc_flag & RC_REPMBUF) != 0)
95 		m_freem(rp->rc_reply);
96 
97 	if ((rp->rc_flag & RC_NAM) != 0)
98 		m_free(rp->rc_nam);
99 
100 	rp->rc_flag &= ~(RC_REPSTATUS|RC_REPMBUF);
101 }
102 
103 /* Initialize the server request cache list */
104 void
105 nfsrv_initcache(void)
106 {
107 
108 	nfsrvhashtbl = hashinit(desirednfsrvcache, M_NFSD, M_WAITOK, &nfsrvhash);
109 	arc4random_buf(&nfsrvhashkey, sizeof(nfsrvhashkey));
110 	TAILQ_INIT(&nfsrvlruhead);
111 }
112 
113 /*
114  * Look for the request in the cache
115  * If found then
116  *    return action and optionally reply
117  * else
118  *    insert it in the cache
119  *
120  * The rules are as follows:
121  * - if in progress, return DROP request
122  * - if completed within DELAY of the current time, return DROP it
123  * - if completed a longer time ago return REPLY if the reply was cached or
124  *   return DOIT
125  * Update/add new request at end of lru list
126  */
127 int
128 nfsrv_getcache(struct nfsrv_descript *nd, struct nfssvc_sock *slp,
129     struct mbuf **repp)
130 {
131 	struct nfsrvhash *hash;
132 	struct nfsrvcache *rp;
133 	struct mbuf *mb;
134 	struct sockaddr_in *saddr;
135 	int ret;
136 
137 	/*
138 	 * Don't cache recent requests for reliable transport protocols.
139 	 * (Maybe we should for the case of a reconnect, but..)
140 	 */
141 	if (!nd->nd_nam2)
142 		return (RC_DOIT);
143 
144 	rp = nfsrv_lookupcache(nd);
145 	if (rp) {
146 		/* If not at end of LRU chain, move it there */
147 		if (TAILQ_NEXT(rp, rc_lru)) {
148 			TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
149 			TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
150 		}
151 		if (rp->rc_state == RC_UNUSED)
152 			panic("nfsrv cache");
153 		if (rp->rc_state == RC_INPROG) {
154 			nfsstats.srvcache_inproghits++;
155 			ret = RC_DROPIT;
156 		} else if (rp->rc_flag & RC_REPSTATUS) {
157 			nfsstats.srvcache_nonidemdonehits++;
158 			nfs_rephead(0, nd, slp, rp->rc_status, repp, &mb);
159 			ret = RC_REPLY;
160 		} else if (rp->rc_flag & RC_REPMBUF) {
161 			nfsstats.srvcache_nonidemdonehits++;
162 			*repp = m_copym(rp->rc_reply, 0, M_COPYALL, M_WAIT);
163 			ret = RC_REPLY;
164 		} else {
165 			nfsstats.srvcache_idemdonehits++;
166 			rp->rc_state = RC_INPROG;
167 			ret = RC_DOIT;
168 		}
169 		rp->rc_flag &= ~RC_LOCKED;
170 		if (rp->rc_flag & RC_WANTED) {
171 			rp->rc_flag &= ~RC_WANTED;
172 			wakeup(rp);
173 		}
174 		return (ret);
175 	}
176 
177 	nfsstats.srvcache_misses++;
178 	if (numnfsrvcache < desirednfsrvcache) {
179 		rp = malloc(sizeof(*rp), M_NFSD, M_WAITOK|M_ZERO);
180 		numnfsrvcache++;
181 		rp->rc_flag = RC_LOCKED;
182 	} else {
183 		rp = TAILQ_FIRST(&nfsrvlruhead);
184 		while ((rp->rc_flag & RC_LOCKED) != 0) {
185 			rp->rc_flag |= RC_WANTED;
186 			tsleep_nsec(rp, PZERO-1, "nfsrc", INFSLP);
187 			rp = TAILQ_FIRST(&nfsrvlruhead);
188 		}
189 		rp->rc_flag |= RC_LOCKED;
190 		LIST_REMOVE(rp, rc_hash);
191 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
192 		nfsrv_cleanentry(rp);
193 		rp->rc_flag &= (RC_LOCKED | RC_WANTED);
194 	}
195 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
196 	rp->rc_state = RC_INPROG;
197 	rp->rc_xid = nd->nd_retxid;
198 	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
199 	switch (saddr->sin_family) {
200 	case AF_INET:
201 		rp->rc_flag |= RC_INETADDR;
202 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
203 		break;
204 	default:
205 		rp->rc_flag |= RC_NAM;
206 		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
207 		break;
208 	};
209 	rp->rc_proc = nd->nd_procnum;
210 	hash = NFSRCHASH(nd->nd_retxid);
211 	LIST_INSERT_HEAD(hash, rp, rc_hash);
212 	rp->rc_flag &= ~RC_LOCKED;
213 	if (rp->rc_flag & RC_WANTED) {
214 		rp->rc_flag &= ~RC_WANTED;
215 		wakeup(rp);
216 	}
217 	return (RC_DOIT);
218 }
219 
220 /* Update a request cache entry after the rpc has been done */
221 void
222 nfsrv_updatecache(struct nfsrv_descript *nd, int repvalid,
223     struct mbuf *repmbuf)
224 {
225 	struct nfsrvcache *rp;
226 
227 	if (!nd->nd_nam2)
228 		return;
229 
230 	rp = nfsrv_lookupcache(nd);
231 	if (rp) {
232 		nfsrv_cleanentry(rp);
233 		rp->rc_state = RC_DONE;
234 		/*
235 		 * If we have a valid reply update status and save
236 		 * the reply for non-idempotent rpc's.
237 		 */
238 		if (repvalid && nonidempotent[nd->nd_procnum]) {
239 			if ((nd->nd_flag & ND_NFSV3) == 0 &&
240 			  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
241 				rp->rc_status = nd->nd_repstat;
242 				rp->rc_flag |= RC_REPSTATUS;
243 			} else {
244 				rp->rc_reply = m_copym(repmbuf, 0, M_COPYALL,
245 				    M_WAIT);
246 				rp->rc_flag |= RC_REPMBUF;
247 			}
248 		}
249 		rp->rc_flag &= ~RC_LOCKED;
250 		if (rp->rc_flag & RC_WANTED) {
251 			rp->rc_flag &= ~RC_WANTED;
252 			wakeup(rp);
253 		}
254 		return;
255 	}
256 }
257 
258 /* Clean out the cache. Called when the last nfsd terminates. */
259 void
260 nfsrv_cleancache(void)
261 {
262 	struct nfsrvcache *rp, *nextrp;
263 
264 	for (rp = TAILQ_FIRST(&nfsrvlruhead); rp != NULL; rp = nextrp) {
265 		nextrp = TAILQ_NEXT(rp, rc_lru);
266 		LIST_REMOVE(rp, rc_hash);
267 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
268 		nfsrv_cleanentry(rp);
269 		free(rp, M_NFSD, sizeof(*rp));
270 	}
271 	numnfsrvcache = 0;
272 }
273 
274 struct nfsrvcache *
275 nfsrv_lookupcache(struct nfsrv_descript *nd)
276 {
277 	struct nfsrvhash	*hash;
278 	struct nfsrvcache	*rp;
279 
280 	hash = NFSRCHASH(nd->nd_retxid);
281 loop:
282 	LIST_FOREACH(rp, hash, rc_hash) {
283 		if (nd->nd_retxid == rp->rc_xid &&
284 		    nd->nd_procnum == rp->rc_proc &&
285 		    netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
286 			if ((rp->rc_flag & RC_LOCKED)) {
287 				rp->rc_flag |= RC_WANTED;
288 				tsleep_nsec(rp, PZERO - 1, "nfsrc", INFSLP);
289 				goto loop;
290 			}
291 			rp->rc_flag |= RC_LOCKED;
292 			return (rp);
293 		}
294 	}
295 
296 	return (NULL);
297 }
298