2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)nfs_srvcache.c 8.3 (Berkeley) 3/30/95
33 * $FreeBSD: src/sys/nfs/nfs_srvcache.c,v 1.21 2000/02/13 03:32:06 peter Exp $
37 * Reference: Chet Juszczak, "Improving the Performance and Correctness
38 * of an NFS Server", in Proc. Winter 1989 USENIX Conference,
39 * pages 53-63. San Diego, February 1989.
41 #include <sys/param.h>
42 #include <sys/malloc.h>
43 #include <sys/mount.h>
44 #include <sys/systm.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h> /* for dup_sockaddr */
49 #include <netinet/in.h>
53 #include "nfsrvcache.h"
56 static long numnfsrvcache;
57 static long desirednfsrvcache;
59 #define NFSRCHASH(xid) \
60 (&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
61 static LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
62 static TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
63 static u_long nfsrvhash;
68 #define NETFAMILY(rp) \
69 (((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
71 struct lwkt_token srvcache_token = LWKT_TOKEN_INITIALIZER(srvcache_token);
74 * Static array that defines which nfs rpc's are nonidempotent
76 static int nonidempotent[NFS_NPROCS] = {
105 /* True iff the rpc reply is an nfs status ONLY! */
106 static int nfsv2_repstat[NFS_NPROCS] = {
128 * Size the NFS server's duplicate request cache at 1/2 the nmbclusters,
129 * within a (64, 2048) range. This is to prevent all mbuf clusters being
130 * tied up in the NFS dupreq cache for small values of nmbclusters.
133 nfsrvcache_size_change(void)
135 desirednfsrvcache = nmbclusters / 2;
136 desirednfsrvcache = MIN(desirednfsrvcache, NFSRVCACHE_MAX_SIZE);
137 desirednfsrvcache = MAX(desirednfsrvcache, NFSRVCACHE_MIN_SIZE);
141 * Initialize the server request cache list
144 nfsrv_initcache(void)
146 nfsrvcache_size_change();
147 nfsrvhashtbl = hashinit(desirednfsrvcache, M_NFSD, &nfsrvhash);
148 TAILQ_INIT(&nfsrvlruhead);
152 * Look for the request in the cache
154 * return action and optionally reply
156 * insert it in the cache
158 * The rules are as follows:
159 * - if in progress, return DROP request
160 * - if completed within DELAY of the current time, return DROP it
161 * - if completed a longer time ago return REPLY if the reply was cached or
163 * Update/add new request at end of lru list
166 nfsrv_getcache(struct nfsrv_descript *nd, struct nfssvc_sock *slp,
169 struct nfsrvcache *rp;
171 struct sockaddr_in *saddr;
176 * Don't cache recent requests for reliable transport protocols.
177 * (Maybe we should for the case of a reconnect, but..)
182 lwkt_gettoken(&srvcache_token);
184 for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != NULL;
185 rp = rp->rc_hash.le_next) {
186 if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
187 netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
188 NFS_DPF(RC, ("H%03x", rp->rc_xid & 0xfff));
189 if ((rp->rc_flag & RC_LOCKED) != 0) {
190 rp->rc_flag |= RC_WANTED;
191 tsleep((caddr_t)rp, 0, "nfsrc", 0);
194 rp->rc_flag |= RC_LOCKED;
195 /* If not at end of LRU chain, move it there */
196 if (TAILQ_NEXT(rp, rc_lru) != NULL) {
197 TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
198 TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
200 if (rp->rc_state == RC_UNUSED)
201 panic("nfsrv cache");
202 if (rp->rc_state == RC_INPROG) {
203 nfsstats.srvcache_inproghits++;
205 } else if (rp->rc_flag & RC_REPSTATUS) {
206 nfsstats.srvcache_nonidemdonehits++;
207 nfs_rephead(0, nd, slp, rp->rc_status,
210 } else if (rp->rc_flag & RC_REPMBUF) {
211 nfsstats.srvcache_nonidemdonehits++;
212 *repp = m_copym(rp->rc_reply, 0, M_COPYALL,
216 nfsstats.srvcache_idemdonehits++;
217 rp->rc_state = RC_INPROG;
220 rp->rc_flag &= ~RC_LOCKED;
221 if (rp->rc_flag & RC_WANTED) {
222 rp->rc_flag &= ~RC_WANTED;
225 lwkt_reltoken(&srvcache_token);
230 nfsstats.srvcache_misses++;
231 NFS_DPF(RC, ("M%03x", nd->nd_retxid & 0xfff));
232 if (numnfsrvcache < desirednfsrvcache) {
233 rp = kmalloc((u_long)sizeof *rp, M_NFSD, M_WAITOK | M_ZERO);
235 rp->rc_flag = RC_LOCKED;
237 rp = TAILQ_FIRST(&nfsrvlruhead);
238 while ((rp->rc_flag & RC_LOCKED) != 0) {
239 rp->rc_flag |= RC_WANTED;
240 tsleep((caddr_t)rp, 0, "nfsrc", 0);
241 rp = TAILQ_FIRST(&nfsrvlruhead);
243 rp->rc_flag |= RC_LOCKED;
244 LIST_REMOVE(rp, rc_hash);
245 TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
246 if (rp->rc_flag & RC_REPMBUF) {
247 m_freem(rp->rc_reply);
249 rp->rc_flag &= ~RC_REPMBUF;
251 if (rp->rc_flag & RC_NAM) {
252 kfree(rp->rc_nam, M_SONAME);
254 rp->rc_flag &= ~RC_NAM;
257 TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
259 rp->rc_state = RC_INPROG;
260 rp->rc_xid = nd->nd_retxid;
261 saddr = (struct sockaddr_in *)nd->nd_nam;
262 switch (saddr->sin_family) {
264 rp->rc_flag |= RC_INETADDR;
265 rp->rc_inetaddr = saddr->sin_addr.s_addr;
269 rp->rc_flag |= RC_NAM;
270 rp->rc_nam = dup_sockaddr(nd->nd_nam);
273 rp->rc_proc = nd->nd_procnum;
274 LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
275 rp->rc_flag &= ~RC_LOCKED;
276 if (rp->rc_flag & RC_WANTED) {
277 rp->rc_flag &= ~RC_WANTED;
280 lwkt_reltoken(&srvcache_token);
286 * Update a request cache entry after the rpc has been done
289 nfsrv_updatecache(struct nfsrv_descript *nd, int repvalid, struct mbuf *repmbuf)
291 struct nfsrvcache *rp;
296 lwkt_gettoken(&srvcache_token);
298 for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != NULL;
299 rp = rp->rc_hash.le_next) {
300 if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
301 netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
302 NFS_DPF(RC, ("U%03x", rp->rc_xid & 0xfff));
303 if ((rp->rc_flag & RC_LOCKED) != 0) {
304 rp->rc_flag |= RC_WANTED;
305 tsleep((caddr_t)rp, 0, "nfsrc", 0);
308 rp->rc_flag |= RC_LOCKED;
309 if (rp->rc_state == RC_DONE) {
311 * This can occur if the cache is too small.
312 * Retransmits of the same request aren't
313 * dropped so we may see the operation
314 * complete more then once.
316 if (rp->rc_flag & RC_REPMBUF) {
317 m_freem(rp->rc_reply);
319 rp->rc_flag &= ~RC_REPMBUF;
322 rp->rc_state = RC_DONE;
325 * If we have a valid reply update status and save
326 * the reply for non-idempotent rpc's.
328 if (repvalid && nonidempotent[nd->nd_procnum]) {
329 if ((nd->nd_flag & ND_NFSV3) == 0 &&
330 nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
331 rp->rc_status = nd->nd_repstat;
332 rp->rc_flag |= RC_REPSTATUS;
334 if (rp->rc_flag & RC_REPMBUF) {
335 m_freem(rp->rc_reply);
337 rp->rc_flag &= ~RC_REPMBUF;
339 rp->rc_reply = m_copym(repmbuf, 0,
341 rp->rc_flag |= RC_REPMBUF;
344 rp->rc_flag &= ~RC_LOCKED;
345 if (rp->rc_flag & RC_WANTED) {
346 rp->rc_flag &= ~RC_WANTED;
352 lwkt_reltoken(&srvcache_token);
353 NFS_DPF(RC, ("L%03x", nd->nd_retxid & 0xfff));
357 * Clean out the cache. Called when the last nfsd terminates.
360 nfsrv_cleancache(void)
362 struct nfsrvcache *rp;
364 lwkt_gettoken(&srvcache_token);
365 while ((rp = TAILQ_FIRST(&nfsrvlruhead)) != NULL) {
366 if (rp->rc_flag & RC_LOCKED) {
367 rp->rc_flag |= RC_WANTED;
368 tsleep((caddr_t)rp, 0, "nfsrc", 0);
371 LIST_REMOVE(rp, rc_hash);
372 TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
373 if (rp->rc_flag & RC_REPMBUF) {
374 m_freem(rp->rc_reply);
376 rp->rc_flag &= ~RC_REPMBUF;
378 if (rp->rc_flag & RC_NAM) {
379 kfree(rp->rc_nam, M_SONAME);
381 rp->rc_flag &= ~RC_NAM;
386 lwkt_reltoken(&srvcache_token);
389 #endif /* NFS_NOSERVER */