xref: /dragonfly/sys/vfs/nfs/nfs_srvcache.c (revision 25a2db75)
1 /*
2  * Copyright (c) 1989, 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  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_srvcache.c	8.3 (Berkeley) 3/30/95
37  * $FreeBSD: src/sys/nfs/nfs_srvcache.c,v 1.21 2000/02/13 03:32:06 peter Exp $
38  */
39 
40 /*
41  * Reference: Chet Juszczak, "Improving the Performance and Correctness
42  *		of an NFS Server", in Proc. Winter 1989 USENIX Conference,
43  *		pages 53-63. San Diego, February 1989.
44  */
45 #include <sys/param.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/systm.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>	/* for dup_sockaddr */
52 
53 #include <netinet/in.h>
54 #include "rpcv2.h"
55 #include "nfsproto.h"
56 #include "nfs.h"
57 #include "nfsrvcache.h"
58 
59 #ifndef NFS_NOSERVER
60 static long numnfsrvcache;
61 static long desirednfsrvcache;
62 
63 #define	NFSRCHASH(xid) \
64 	(&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
65 static LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
66 static TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
67 static u_long nfsrvhash;
68 
69 #define TRUE	1
70 #define	FALSE	0
71 
72 #define	NETFAMILY(rp) \
73 		(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
74 
75 struct lwkt_token srvcache_token = LWKT_TOKEN_INITIALIZER(srvcache_token);
76 
77 /*
78  * Static array that defines which nfs rpc's are nonidempotent
79  */
80 static int nonidempotent[NFS_NPROCS] = {
81 	FALSE,
82 	FALSE,
83 	TRUE,
84 	FALSE,
85 	FALSE,
86 	FALSE,
87 	FALSE,
88 	TRUE,
89 	TRUE,
90 	TRUE,
91 	TRUE,
92 	TRUE,
93 	TRUE,
94 	TRUE,
95 	TRUE,
96 	TRUE,
97 	FALSE,
98 	FALSE,
99 	FALSE,
100 	FALSE,
101 	FALSE,
102 	FALSE,
103 	FALSE,
104 	FALSE,
105 	FALSE,
106 	FALSE,
107 };
108 
109 /* True iff the rpc reply is an nfs status ONLY! */
110 static int nfsv2_repstat[NFS_NPROCS] = {
111 	FALSE,
112 	FALSE,
113 	FALSE,
114 	FALSE,
115 	FALSE,
116 	FALSE,
117 	FALSE,
118 	FALSE,
119 	FALSE,
120 	FALSE,
121 	TRUE,
122 	TRUE,
123 	TRUE,
124 	TRUE,
125 	FALSE,
126 	TRUE,
127 	FALSE,
128 	FALSE,
129 };
130 
131 /*
132  * Size the NFS server's duplicate request cache at 1/2 the nmbclusters,
133  * within a (64, 2048) range. This is to prevent all mbuf clusters being
134  * tied up in the NFS dupreq cache for small values of nmbclusters.
135  */
136 static void
137 nfsrvcache_size_change(void)
138 {
139 	desirednfsrvcache = nmbclusters / 2;
140 	desirednfsrvcache = MIN(desirednfsrvcache, NFSRVCACHE_MAX_SIZE);
141 	desirednfsrvcache = MAX(desirednfsrvcache, NFSRVCACHE_MIN_SIZE);
142 }
143 
144 /*
145  * Initialize the server request cache list
146  */
147 void
148 nfsrv_initcache(void)
149 {
150 	nfsrvcache_size_change();
151 	nfsrvhashtbl = hashinit(desirednfsrvcache, M_NFSD, &nfsrvhash);
152 	TAILQ_INIT(&nfsrvlruhead);
153 }
154 
155 /*
156  * Look for the request in the cache
157  * If found then
158  *    return action and optionally reply
159  * else
160  *    insert it in the cache
161  *
162  * The rules are as follows:
163  * - if in progress, return DROP request
164  * - if completed within DELAY of the current time, return DROP it
165  * - if completed a longer time ago return REPLY if the reply was cached or
166  *   return DOIT
167  * Update/add new request at end of lru list
168  */
169 int
170 nfsrv_getcache(struct nfsrv_descript *nd, struct nfssvc_sock *slp,
171 	       struct mbuf **repp)
172 {
173 	struct nfsrvcache *rp;
174 	struct mbuf *mb;
175 	struct sockaddr_in *saddr;
176 	caddr_t bpos;
177 	int ret;
178 
179 	/*
180 	 * Don't cache recent requests for reliable transport protocols.
181 	 * (Maybe we should for the case of a reconnect, but..)
182 	 */
183 	if (!nd->nd_nam2)
184 		return (RC_DOIT);
185 
186 	lwkt_gettoken(&srvcache_token);
187 loop:
188 	for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != NULL;
189 	    rp = rp->rc_hash.le_next) {
190 	    if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
191 		netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
192 		        NFS_DPF(RC, ("H%03x", rp->rc_xid & 0xfff));
193 			if ((rp->rc_flag & RC_LOCKED) != 0) {
194 				rp->rc_flag |= RC_WANTED;
195 				tsleep((caddr_t)rp, 0, "nfsrc", 0);
196 				goto loop;
197 			}
198 			rp->rc_flag |= RC_LOCKED;
199 			/* If not at end of LRU chain, move it there */
200 			if (TAILQ_NEXT(rp, rc_lru) != NULL) {
201 				TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
202 				TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
203 			}
204 			if (rp->rc_state == RC_UNUSED)
205 				panic("nfsrv cache");
206 			if (rp->rc_state == RC_INPROG) {
207 				nfsstats.srvcache_inproghits++;
208 				ret = RC_DROPIT;
209 			} else if (rp->rc_flag & RC_REPSTATUS) {
210 				nfsstats.srvcache_nonidemdonehits++;
211 				nfs_rephead(0, nd, slp, rp->rc_status,
212 					    repp, &mb, &bpos);
213 				ret = RC_REPLY;
214 			} else if (rp->rc_flag & RC_REPMBUF) {
215 				nfsstats.srvcache_nonidemdonehits++;
216 				*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
217 						MB_WAIT);
218 				ret = RC_REPLY;
219 			} else {
220 				nfsstats.srvcache_idemdonehits++;
221 				rp->rc_state = RC_INPROG;
222 				ret = RC_DOIT;
223 			}
224 			rp->rc_flag &= ~RC_LOCKED;
225 			if (rp->rc_flag & RC_WANTED) {
226 				rp->rc_flag &= ~RC_WANTED;
227 				wakeup((caddr_t)rp);
228 			}
229 			lwkt_reltoken(&srvcache_token);
230 			return (ret);
231 		}
232 	}
233 
234 	nfsstats.srvcache_misses++;
235 	NFS_DPF(RC, ("M%03x", nd->nd_retxid & 0xfff));
236 	if (numnfsrvcache < desirednfsrvcache) {
237 		rp = kmalloc((u_long)sizeof *rp, M_NFSD, M_WAITOK | M_ZERO);
238 		numnfsrvcache++;
239 		rp->rc_flag = RC_LOCKED;
240 	} else {
241 		rp = TAILQ_FIRST(&nfsrvlruhead);
242 		while ((rp->rc_flag & RC_LOCKED) != 0) {
243 			rp->rc_flag |= RC_WANTED;
244 			tsleep((caddr_t)rp, 0, "nfsrc", 0);
245 			rp = TAILQ_FIRST(&nfsrvlruhead);
246 		}
247 		rp->rc_flag |= RC_LOCKED;
248 		LIST_REMOVE(rp, rc_hash);
249 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
250 		if (rp->rc_flag & RC_REPMBUF) {
251 			m_freem(rp->rc_reply);
252 			rp->rc_reply = NULL;
253 			rp->rc_flag &= ~RC_REPMBUF;
254 		}
255 		if (rp->rc_flag & RC_NAM) {
256 			kfree(rp->rc_nam, M_SONAME);
257 			rp->rc_nam = NULL;
258 			rp->rc_flag &= ~RC_NAM;
259 		}
260 	}
261 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
262 
263 	rp->rc_state = RC_INPROG;
264 	rp->rc_xid = nd->nd_retxid;
265 	saddr = (struct sockaddr_in *)nd->nd_nam;
266 	switch (saddr->sin_family) {
267 	case AF_INET:
268 		rp->rc_flag |= RC_INETADDR;
269 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
270 		break;
271 	case AF_ISO:
272 	default:
273 		rp->rc_flag |= RC_NAM;
274 		rp->rc_nam = dup_sockaddr(nd->nd_nam);
275 		break;
276 	};
277 	rp->rc_proc = nd->nd_procnum;
278 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
279 	rp->rc_flag &= ~RC_LOCKED;
280 	if (rp->rc_flag & RC_WANTED) {
281 		rp->rc_flag &= ~RC_WANTED;
282 		wakeup((caddr_t)rp);
283 	}
284 	lwkt_reltoken(&srvcache_token);
285 
286 	return (RC_DOIT);
287 }
288 
289 /*
290  * Update a request cache entry after the rpc has been done
291  */
292 void
293 nfsrv_updatecache(struct nfsrv_descript *nd, int repvalid, struct mbuf *repmbuf)
294 {
295 	struct nfsrvcache *rp;
296 
297 	if (!nd->nd_nam2)
298 		return;
299 
300 	lwkt_gettoken(&srvcache_token);
301 loop:
302 	for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != NULL;
303 	    rp = rp->rc_hash.le_next) {
304 	    if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
305 		netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
306 			NFS_DPF(RC, ("U%03x", rp->rc_xid & 0xfff));
307 			if ((rp->rc_flag & RC_LOCKED) != 0) {
308 				rp->rc_flag |= RC_WANTED;
309 				tsleep((caddr_t)rp, 0, "nfsrc", 0);
310 				goto loop;
311 			}
312 			rp->rc_flag |= RC_LOCKED;
313 			if (rp->rc_state == RC_DONE) {
314 				/*
315 				 * This can occur if the cache is too small.
316 				 * Retransmits of the same request aren't
317 				 * dropped so we may see the operation
318 				 * complete more then once.
319 				 */
320 				if (rp->rc_flag & RC_REPMBUF) {
321 					m_freem(rp->rc_reply);
322 					rp->rc_reply = NULL;
323 					rp->rc_flag &= ~RC_REPMBUF;
324 				}
325 			}
326 			rp->rc_state = RC_DONE;
327 
328 			/*
329 			 * If we have a valid reply update status and save
330 			 * the reply for non-idempotent rpc's.
331 			 */
332 			if (repvalid && nonidempotent[nd->nd_procnum]) {
333 				if ((nd->nd_flag & ND_NFSV3) == 0 &&
334 				  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
335 					rp->rc_status = nd->nd_repstat;
336 					rp->rc_flag |= RC_REPSTATUS;
337 				} else {
338 					if (rp->rc_flag & RC_REPMBUF) {
339 						m_freem(rp->rc_reply);
340 						rp->rc_reply = NULL;
341 						rp->rc_flag &= ~RC_REPMBUF;
342 					}
343 					rp->rc_reply = m_copym(repmbuf, 0,
344 							M_COPYALL, MB_WAIT);
345 					rp->rc_flag |= RC_REPMBUF;
346 				}
347 			}
348 			rp->rc_flag &= ~RC_LOCKED;
349 			if (rp->rc_flag & RC_WANTED) {
350 				rp->rc_flag &= ~RC_WANTED;
351 				wakeup((caddr_t)rp);
352 			}
353 			break;
354 		}
355 	}
356 	lwkt_reltoken(&srvcache_token);
357 	NFS_DPF(RC, ("L%03x", nd->nd_retxid & 0xfff));
358 }
359 
360 /*
361  * Clean out the cache. Called when the last nfsd terminates.
362  */
363 void
364 nfsrv_cleancache(void)
365 {
366 	struct nfsrvcache *rp;
367 
368 	lwkt_gettoken(&srvcache_token);
369 	while ((rp = TAILQ_FIRST(&nfsrvlruhead)) != NULL) {
370 		if (rp->rc_flag & RC_LOCKED) {
371 			rp->rc_flag |= RC_WANTED;
372 			tsleep((caddr_t)rp, 0, "nfsrc", 0);
373 			continue;
374 		}
375 		LIST_REMOVE(rp, rc_hash);
376 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
377 		if (rp->rc_flag & RC_REPMBUF) {
378 			m_freem(rp->rc_reply);
379 			rp->rc_reply = NULL;
380 			rp->rc_flag &= ~RC_REPMBUF;
381 		}
382 		if (rp->rc_flag & RC_NAM) {
383 			kfree(rp->rc_nam, M_SONAME);
384 			rp->rc_nam = NULL;
385 			rp->rc_flag &= ~RC_NAM;
386 		}
387 		kfree(rp, M_NFSD);
388 	}
389 	numnfsrvcache = 0;
390 	lwkt_reltoken(&srvcache_token);
391 }
392 
393 #endif /* NFS_NOSERVER */
394