xref: /netbsd-src/sys/nfs/nfs_srvcache.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: nfs_srvcache.c,v 1.40 2007/06/01 11:56:03 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_srvcache.c	8.3 (Berkeley) 3/30/95
35  */
36 
37 /*
38  * Reference: Chet Juszczak, "Improving the Performance and Correctness
39  *		of an NFS Server", in Proc. Winter 1989 USENIX Conference,
40  *		pages 53-63. San Diego, February 1989.
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: nfs_srvcache.c,v 1.40 2007/06/01 11:56:03 yamt Exp $");
45 
46 #include "opt_iso.h"
47 
48 #include <sys/param.h>
49 #include <sys/vnode.h>
50 #include <sys/condvar.h>
51 #include <sys/mount.h>
52 #include <sys/kernel.h>
53 #include <sys/systm.h>
54 #include <sys/lock.h>
55 #include <sys/proc.h>
56 #include <sys/pool.h>
57 #include <sys/mbuf.h>
58 #include <sys/malloc.h>
59 #include <sys/mutex.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 
63 #include <netinet/in.h>
64 #ifdef ISO
65 #include <netiso/iso.h>
66 #endif
67 #include <nfs/nfsm_subs.h>
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfs/nfs.h>
71 #include <nfs/nfsrvcache.h>
72 #include <nfs/nfs_var.h>
73 
74 extern struct nfsstats nfsstats;
75 extern const int nfsv2_procid[NFS_NPROCS];
76 long numnfsrvcache, desirednfsrvcache = NFSRVCACHESIZ;
77 struct pool nfs_reqcache_pool;
78 
79 #define	NFSRCHASH(xid) \
80 	(&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
81 LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
82 TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
83 kmutex_t nfsrv_reqcache_lock;
84 u_long nfsrvhash;
85 
86 #if defined(MBUFTRACE)
87 static struct mowner nfsd_cache_mowner = MOWNER_INIT("nfsd", "cache");
88 #endif /* defined(MBUFTRACE) */
89 
90 #define	NETFAMILY(rp) \
91 		(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
92 
93 static struct nfsrvcache *nfsrv_lookupcache(struct nfsrv_descript *nd);
94 static void nfsrv_unlockcache(struct nfsrvcache *rp);
95 
96 /*
97  * Static array that defines which nfs rpc's are nonidempotent
98  */
99 const int nonidempotent[NFS_NPROCS] = {
100 	false,	/* NULL */
101 	false,	/* GETATTR */
102 	true,	/* SETATTR */
103 	false,	/* LOOKUP */
104 	false,	/* ACCESS */
105 	false,	/* READLINK */
106 	false,	/* READ */
107 	true,	/* WRITE */
108 	true,	/* CREATE */
109 	true,	/* MKDIR */
110 	true,	/* SYMLINK */
111 	true,	/* MKNOD */
112 	true,	/* REMOVE */
113 	true,	/* RMDIR */
114 	true,	/* RENAME */
115 	true,	/* LINK */
116 	false,	/* READDIR */
117 	false,	/* READDIRPLUS */
118 	false,	/* FSSTAT */
119 	false,	/* FSINFO */
120 	false,	/* PATHCONF */
121 	false,	/* COMMIT */
122 	false,	/* NOOP */
123 };
124 
125 /* True iff the rpc reply is an nfs status ONLY! */
126 static const int nfsv2_repstat[NFS_NPROCS] = {
127 	false,	/* NULL */
128 	false,	/* GETATTR */
129 	false,	/* SETATTR */
130 	false,	/* NOOP */
131 	false,	/* LOOKUP */
132 	false,	/* READLINK */
133 	false,	/* READ */
134 	false,	/* Obsolete WRITECACHE */
135 	false,	/* WRITE */
136 	false,	/* CREATE */
137 	true,	/* REMOVE */
138 	true,	/* RENAME */
139 	true,	/* LINK */
140 	true,	/* SYMLINK */
141 	false,	/* MKDIR */
142 	true,	/* RMDIR */
143 	false,	/* READDIR */
144 	false,	/* STATFS */
145 };
146 
147 static void
148 cleanentry(struct nfsrvcache *rp)
149 {
150 
151 	if ((rp->rc_flag & RC_REPMBUF) != 0) {
152 		m_freem(rp->rc_reply);
153 	}
154 	if ((rp->rc_flag & RC_NAM) != 0) {
155 		m_free(rp->rc_nam);
156 	}
157 	rp->rc_flag &= ~(RC_REPSTATUS|RC_REPMBUF);
158 }
159 
160 /*
161  * Initialize the server request cache list
162  */
163 void
164 nfsrv_initcache()
165 {
166 
167 	mutex_init(&nfsrv_reqcache_lock, MUTEX_DRIVER, IPL_NONE);
168 	nfsrvhashtbl = hashinit(desirednfsrvcache, HASH_LIST, M_NFSD,
169 	    M_WAITOK, &nfsrvhash);
170 	TAILQ_INIT(&nfsrvlruhead);
171 	pool_init(&nfs_reqcache_pool, sizeof(struct nfsrvcache), 0, 0, 0,
172 	    "nfsreqcachepl", &pool_allocator_nointr, IPL_NONE);
173 	MOWNER_ATTACH(&nfsd_cache_mowner);
174 }
175 
176 /*
177  * Lookup a cache and lock it
178  */
179 static struct nfsrvcache *
180 nfsrv_lookupcache(nd)
181 	struct nfsrv_descript *nd;
182 {
183 	struct nfsrvcache *rp;
184 
185 	KASSERT(mutex_owned(&nfsrv_reqcache_lock));
186 
187 loop:
188 	LIST_FOREACH(rp, NFSRCHASH(nd->nd_retxid), rc_hash) {
189 		if (nd->nd_retxid == rp->rc_xid &&
190 		    nd->nd_procnum == rp->rc_proc &&
191 		    netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
192 			if ((rp->rc_flag & RC_LOCKED) != 0) {
193 				cv_wait(&rp->rc_cv, &nfsrv_reqcache_lock);
194 				goto loop;
195 			}
196 			rp->rc_flag |= RC_LOCKED;
197 			break;
198 		}
199 	}
200 
201 	return rp;
202 }
203 
204 /*
205  * Unlock a cache
206  */
207 static void
208 nfsrv_unlockcache(rp)
209 	struct nfsrvcache *rp;
210 {
211 
212 	KASSERT(mutex_owned(&nfsrv_reqcache_lock));
213 
214 	rp->rc_flag &= ~RC_LOCKED;
215 	cv_broadcast(&rp->rc_cv);
216 }
217 
218 /*
219  * Look for the request in the cache
220  * If found then
221  *    return action and optionally reply
222  * else
223  *    insert it in the cache
224  *
225  * The rules are as follows:
226  * - if in progress, return DROP request
227  * - if completed within DELAY of the current time, return DROP it
228  * - if completed a longer time ago return REPLY if the reply was cached or
229  *   return DOIT
230  * Update/add new request at end of lru list
231  */
232 int
233 nfsrv_getcache(nd, slp, repp)
234 	struct nfsrv_descript *nd;
235 	struct nfssvc_sock *slp;
236 	struct mbuf **repp;
237 {
238 	struct nfsrvcache *rp, *rpdup;
239 	struct mbuf *mb;
240 	struct sockaddr_in *saddr;
241 	char *bpos;
242 	int ret;
243 
244 	mutex_enter(&nfsrv_reqcache_lock);
245 	rp = nfsrv_lookupcache(nd);
246 	if (rp) {
247 		mutex_exit(&nfsrv_reqcache_lock);
248 found:
249 		/* If not at end of LRU chain, move it there */
250 		if (TAILQ_NEXT(rp, rc_lru)) { /* racy but ok */
251 			mutex_enter(&nfsrv_reqcache_lock);
252 			TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
253 			TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
254 			mutex_exit(&nfsrv_reqcache_lock);
255 		}
256 		if (rp->rc_state == RC_UNUSED)
257 			panic("nfsrv cache");
258 		if (rp->rc_state == RC_INPROG) {
259 			nfsstats.srvcache_inproghits++;
260 			ret = RC_DROPIT;
261 		} else if (rp->rc_flag & RC_REPSTATUS) {
262 			nfsstats.srvcache_nonidemdonehits++;
263 			nfs_rephead(0, nd, slp, rp->rc_status,
264 			   0, (u_quad_t *)0, repp, &mb, &bpos);
265 			ret = RC_REPLY;
266 		} else if (rp->rc_flag & RC_REPMBUF) {
267 			nfsstats.srvcache_nonidemdonehits++;
268 			*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
269 					M_WAIT);
270 			ret = RC_REPLY;
271 		} else {
272 			nfsstats.srvcache_idemdonehits++;
273 			rp->rc_state = RC_INPROG;
274 			ret = RC_DOIT;
275 		}
276 		mutex_enter(&nfsrv_reqcache_lock);
277 		nfsrv_unlockcache(rp);
278 		mutex_exit(&nfsrv_reqcache_lock);
279 		return ret;
280 	}
281 	nfsstats.srvcache_misses++;
282 	if (numnfsrvcache < desirednfsrvcache) {
283 		numnfsrvcache++;
284 		mutex_exit(&nfsrv_reqcache_lock);
285 		rp = pool_get(&nfs_reqcache_pool, PR_WAITOK);
286 		memset(rp, 0, sizeof *rp);
287 		cv_init(&rp->rc_cv, "nfsdrc");
288 		rp->rc_flag = RC_LOCKED;
289 	} else {
290 		rp = TAILQ_FIRST(&nfsrvlruhead);
291 		while ((rp->rc_flag & RC_LOCKED) != 0) {
292 			rp->rc_flag |= RC_WANTED;
293 			cv_wait(&rp->rc_cv, &nfsrv_reqcache_lock);
294 			rp = TAILQ_FIRST(&nfsrvlruhead);
295 		}
296 		rp->rc_flag |= RC_LOCKED;
297 		LIST_REMOVE(rp, rc_hash);
298 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
299 		mutex_exit(&nfsrv_reqcache_lock);
300 		cleanentry(rp);
301 		rp->rc_flag &= (RC_LOCKED | RC_WANTED);
302 	}
303 	rp->rc_state = RC_INPROG;
304 	rp->rc_xid = nd->nd_retxid;
305 	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
306 	switch (saddr->sin_family) {
307 	case AF_INET:
308 		rp->rc_flag |= RC_INETADDR;
309 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
310 		break;
311 	case AF_ISO:
312 	default:
313 		rp->rc_flag |= RC_NAM;
314 		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
315 		m_claimm(rp->rc_nam, &nfsd_cache_mowner);
316 		break;
317 	};
318 	rp->rc_proc = nd->nd_procnum;
319 	mutex_enter(&nfsrv_reqcache_lock);
320 	rpdup = nfsrv_lookupcache(nd);
321 	if (rpdup != NULL) {
322 		/*
323 		 * other thread made duplicate cache entry.
324 		 */
325 		mutex_exit(&nfsrv_reqcache_lock);
326 		cv_destroy(&rp->rc_cv);
327 		pool_put(&nfs_reqcache_pool, rp);
328 		rp = rpdup;
329 		goto found;
330 	}
331 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
332 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
333 	nfsrv_unlockcache(rp);
334 	mutex_exit(&nfsrv_reqcache_lock);
335 	return RC_DOIT;
336 }
337 
338 /*
339  * Update a request cache entry after the rpc has been done
340  */
341 void
342 nfsrv_updatecache(nd, repvalid, repmbuf)
343 	struct nfsrv_descript *nd;
344 	int repvalid;
345 	struct mbuf *repmbuf;
346 {
347 	struct nfsrvcache *rp;
348 
349 	mutex_enter(&nfsrv_reqcache_lock);
350 	rp = nfsrv_lookupcache(nd);
351 	mutex_exit(&nfsrv_reqcache_lock);
352 	if (rp) {
353 		cleanentry(rp);
354 		rp->rc_state = RC_DONE;
355 		/*
356 		 * If we have a valid reply update status and save
357 		 * the reply for non-idempotent rpc's.
358 		 */
359 		if (repvalid && nonidempotent[nd->nd_procnum]) {
360 			if ((nd->nd_flag & ND_NFSV3) == 0 &&
361 			  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
362 				rp->rc_status = nd->nd_repstat;
363 				rp->rc_flag |= RC_REPSTATUS;
364 			} else {
365 				rp->rc_reply = m_copym(repmbuf,
366 					0, M_COPYALL, M_WAIT);
367 				m_claimm(rp->rc_reply, &nfsd_cache_mowner);
368 				rp->rc_flag |= RC_REPMBUF;
369 			}
370 		}
371 		mutex_enter(&nfsrv_reqcache_lock);
372 		nfsrv_unlockcache(rp);
373 		mutex_exit(&nfsrv_reqcache_lock);
374 	}
375 }
376 
377 /*
378  * Clean out the cache. Called when the last nfsd terminates.
379  */
380 void
381 nfsrv_cleancache()
382 {
383 	struct nfsrvcache *rp, *nextrp;
384 
385 	mutex_enter(&nfsrv_reqcache_lock);
386 	for (rp = TAILQ_FIRST(&nfsrvlruhead); rp != 0; rp = nextrp) {
387 		nextrp = TAILQ_NEXT(rp, rc_lru);
388 		LIST_REMOVE(rp, rc_hash);
389 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
390 		KASSERT((rp->rc_flag & (RC_LOCKED|RC_WANTED)) == 0);
391 		cleanentry(rp);
392 		cv_destroy(&rp->rc_cv);
393 		pool_put(&nfs_reqcache_pool, rp);
394 	}
395 	numnfsrvcache = 0;
396 	mutex_exit(&nfsrv_reqcache_lock);
397 }
398