Home | History | Annotate | Line # | Download | only in nfs
nfs_srvcache.c revision 1.39
      1 /*	$NetBSD: nfs_srvcache.c,v 1.39 2007/03/12 18:18:36 ad 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 1.39 2007/03/12 18:18:36 ad Exp $");
     45 
     46 #include "opt_iso.h"
     47 
     48 #include <sys/param.h>
     49 #include <sys/vnode.h>
     50 #include <sys/mount.h>
     51 #include <sys/kernel.h>
     52 #include <sys/systm.h>
     53 #include <sys/lock.h>
     54 #include <sys/proc.h>
     55 #include <sys/pool.h>
     56 #include <sys/mbuf.h>
     57 #include <sys/malloc.h>
     58 #include <sys/socket.h>
     59 #include <sys/socketvar.h>
     60 
     61 #include <netinet/in.h>
     62 #ifdef ISO
     63 #include <netiso/iso.h>
     64 #endif
     65 #include <nfs/nfsm_subs.h>
     66 #include <nfs/rpcv2.h>
     67 #include <nfs/nfsproto.h>
     68 #include <nfs/nfs.h>
     69 #include <nfs/nfsrvcache.h>
     70 #include <nfs/nfs_var.h>
     71 
     72 extern struct nfsstats nfsstats;
     73 extern const int nfsv2_procid[NFS_NPROCS];
     74 long numnfsrvcache, desirednfsrvcache = NFSRVCACHESIZ;
     75 struct pool nfs_reqcache_pool;
     76 
     77 #define	NFSRCHASH(xid) \
     78 	(&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
     79 LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
     80 TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
     81 struct simplelock nfsrv_reqcache_lock = SIMPLELOCK_INITIALIZER;
     82 u_long nfsrvhash;
     83 
     84 #if defined(MBUFTRACE)
     85 static struct mowner nfsd_cache_mowner = MOWNER_INIT("nfsd", "cache");
     86 #endif /* defined(MBUFTRACE) */
     87 
     88 #define	NETFAMILY(rp) \
     89 		(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
     90 
     91 static struct nfsrvcache *nfsrv_lookupcache(struct nfsrv_descript *nd);
     92 static void nfsrv_unlockcache(struct nfsrvcache *rp);
     93 
     94 /*
     95  * Static array that defines which nfs rpc's are nonidempotent
     96  */
     97 const int nonidempotent[NFS_NPROCS] = {
     98 	false,	/* NULL */
     99 	false,	/* GETATTR */
    100 	true,	/* SETATTR */
    101 	false,	/* LOOKUP */
    102 	false,	/* ACCESS */
    103 	false,	/* READLINK */
    104 	false,	/* READ */
    105 	true,	/* WRITE */
    106 	true,	/* CREATE */
    107 	true,	/* MKDIR */
    108 	true,	/* SYMLINK */
    109 	true,	/* MKNOD */
    110 	true,	/* REMOVE */
    111 	true,	/* RMDIR */
    112 	true,	/* RENAME */
    113 	true,	/* LINK */
    114 	false,	/* READDIR */
    115 	false,	/* READDIRPLUS */
    116 	false,	/* FSSTAT */
    117 	false,	/* FSINFO */
    118 	false,	/* PATHCONF */
    119 	false,	/* COMMIT */
    120 	false,	/* NOOP */
    121 };
    122 
    123 /* True iff the rpc reply is an nfs status ONLY! */
    124 static const int nfsv2_repstat[NFS_NPROCS] = {
    125 	false,	/* NULL */
    126 	false,	/* GETATTR */
    127 	false,	/* SETATTR */
    128 	false,	/* NOOP */
    129 	false,	/* LOOKUP */
    130 	false,	/* READLINK */
    131 	false,	/* READ */
    132 	false,	/* Obsolete WRITECACHE */
    133 	false,	/* WRITE */
    134 	false,	/* CREATE */
    135 	true,	/* REMOVE */
    136 	true,	/* RENAME */
    137 	true,	/* LINK */
    138 	true,	/* SYMLINK */
    139 	false,	/* MKDIR */
    140 	true,	/* RMDIR */
    141 	false,	/* READDIR */
    142 	false,	/* STATFS */
    143 };
    144 
    145 static void
    146 cleanentry(struct nfsrvcache *rp)
    147 {
    148 
    149 	if ((rp->rc_flag & RC_REPMBUF) != 0) {
    150 		m_freem(rp->rc_reply);
    151 	}
    152 	if ((rp->rc_flag & RC_NAM) != 0) {
    153 		m_free(rp->rc_nam);
    154 	}
    155 	rp->rc_flag &= ~(RC_REPSTATUS|RC_REPMBUF);
    156 }
    157 
    158 /*
    159  * Initialize the server request cache list
    160  */
    161 void
    162 nfsrv_initcache()
    163 {
    164 
    165 	nfsrvhashtbl = hashinit(desirednfsrvcache, HASH_LIST, M_NFSD,
    166 	    M_WAITOK, &nfsrvhash);
    167 	TAILQ_INIT(&nfsrvlruhead);
    168 	pool_init(&nfs_reqcache_pool, sizeof(struct nfsrvcache), 0, 0, 0,
    169 	    "nfsreqcachepl", &pool_allocator_nointr, IPL_NONE);
    170 	MOWNER_ATTACH(&nfsd_cache_mowner);
    171 }
    172 
    173 /*
    174  * Lookup a cache and lock it
    175  */
    176 static struct nfsrvcache *
    177 nfsrv_lookupcache(nd)
    178 	struct nfsrv_descript *nd;
    179 {
    180 	struct nfsrvcache *rp;
    181 
    182 	LOCK_ASSERT(simple_lock_held(&nfsrv_reqcache_lock));
    183 
    184 loop:
    185 	LIST_FOREACH(rp, NFSRCHASH(nd->nd_retxid), rc_hash) {
    186 		if (nd->nd_retxid == rp->rc_xid &&
    187 		    nd->nd_procnum == rp->rc_proc &&
    188 		    netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
    189 			if ((rp->rc_flag & RC_LOCKED) != 0) {
    190 				rp->rc_flag |= RC_WANTED;
    191 				(void) ltsleep(rp, PZERO - 1, "nfsrc", 0,
    192 				    &nfsrv_reqcache_lock);
    193 				goto loop;
    194 			}
    195 			rp->rc_flag |= RC_LOCKED;
    196 			break;
    197 		}
    198 	}
    199 
    200 	return rp;
    201 }
    202 
    203 /*
    204  * Unlock a cache
    205  */
    206 static void
    207 nfsrv_unlockcache(rp)
    208 	struct nfsrvcache *rp;
    209 {
    210 
    211 	LOCK_ASSERT(simple_lock_held(&nfsrv_reqcache_lock));
    212 
    213 	rp->rc_flag &= ~RC_LOCKED;
    214 	if (rp->rc_flag & RC_WANTED) {
    215 		rp->rc_flag &= ~RC_WANTED;
    216 		wakeup(rp);
    217 	}
    218 }
    219 
    220 /*
    221  * Look for the request in the cache
    222  * If found then
    223  *    return action and optionally reply
    224  * else
    225  *    insert it in the cache
    226  *
    227  * The rules are as follows:
    228  * - if in progress, return DROP request
    229  * - if completed within DELAY of the current time, return DROP it
    230  * - if completed a longer time ago return REPLY if the reply was cached or
    231  *   return DOIT
    232  * Update/add new request at end of lru list
    233  */
    234 int
    235 nfsrv_getcache(nd, slp, repp)
    236 	struct nfsrv_descript *nd;
    237 	struct nfssvc_sock *slp;
    238 	struct mbuf **repp;
    239 {
    240 	struct nfsrvcache *rp, *rpdup;
    241 	struct mbuf *mb;
    242 	struct sockaddr_in *saddr;
    243 	char *bpos;
    244 	int ret;
    245 
    246 	simple_lock(&nfsrv_reqcache_lock);
    247 	rp = nfsrv_lookupcache(nd);
    248 	if (rp) {
    249 		simple_unlock(&nfsrv_reqcache_lock);
    250 found:
    251 		/* If not at end of LRU chain, move it there */
    252 		if (TAILQ_NEXT(rp, rc_lru)) { /* racy but ok */
    253 			simple_lock(&nfsrv_reqcache_lock);
    254 			TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    255 			TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    256 			simple_unlock(&nfsrv_reqcache_lock);
    257 		}
    258 		if (rp->rc_state == RC_UNUSED)
    259 			panic("nfsrv cache");
    260 		if (rp->rc_state == RC_INPROG) {
    261 			nfsstats.srvcache_inproghits++;
    262 			ret = RC_DROPIT;
    263 		} else if (rp->rc_flag & RC_REPSTATUS) {
    264 			nfsstats.srvcache_nonidemdonehits++;
    265 			nfs_rephead(0, nd, slp, rp->rc_status,
    266 			   0, (u_quad_t *)0, repp, &mb, &bpos);
    267 			ret = RC_REPLY;
    268 		} else if (rp->rc_flag & RC_REPMBUF) {
    269 			nfsstats.srvcache_nonidemdonehits++;
    270 			*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
    271 					M_WAIT);
    272 			ret = RC_REPLY;
    273 		} else {
    274 			nfsstats.srvcache_idemdonehits++;
    275 			rp->rc_state = RC_INPROG;
    276 			ret = RC_DOIT;
    277 		}
    278 		simple_lock(&nfsrv_reqcache_lock);
    279 		nfsrv_unlockcache(rp);
    280 		simple_unlock(&nfsrv_reqcache_lock);
    281 		return ret;
    282 	}
    283 	nfsstats.srvcache_misses++;
    284 	if (numnfsrvcache < desirednfsrvcache) {
    285 		numnfsrvcache++;
    286 		simple_unlock(&nfsrv_reqcache_lock);
    287 		rp = pool_get(&nfs_reqcache_pool, PR_WAITOK);
    288 		memset(rp, 0, sizeof *rp);
    289 		rp->rc_flag = RC_LOCKED;
    290 	} else {
    291 		rp = TAILQ_FIRST(&nfsrvlruhead);
    292 		while ((rp->rc_flag & RC_LOCKED) != 0) {
    293 			rp->rc_flag |= RC_WANTED;
    294 			(void) ltsleep(rp, PZERO-1, "nfsrc", 0,
    295 			    &nfsrv_reqcache_lock);
    296 			rp = TAILQ_FIRST(&nfsrvlruhead);
    297 		}
    298 		rp->rc_flag |= RC_LOCKED;
    299 		LIST_REMOVE(rp, rc_hash);
    300 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    301 		simple_unlock(&nfsrv_reqcache_lock);
    302 		cleanentry(rp);
    303 		rp->rc_flag &= (RC_LOCKED | RC_WANTED);
    304 	}
    305 	rp->rc_state = RC_INPROG;
    306 	rp->rc_xid = nd->nd_retxid;
    307 	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
    308 	switch (saddr->sin_family) {
    309 	case AF_INET:
    310 		rp->rc_flag |= RC_INETADDR;
    311 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
    312 		break;
    313 	case AF_ISO:
    314 	default:
    315 		rp->rc_flag |= RC_NAM;
    316 		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
    317 		m_claimm(rp->rc_nam, &nfsd_cache_mowner);
    318 		break;
    319 	};
    320 	rp->rc_proc = nd->nd_procnum;
    321 	simple_lock(&nfsrv_reqcache_lock);
    322 	rpdup = nfsrv_lookupcache(nd);
    323 	if (rpdup != NULL) {
    324 		/*
    325 		 * other thread made duplicate cache entry.
    326 		 */
    327 		simple_unlock(&nfsrv_reqcache_lock);
    328 		pool_put(&nfs_reqcache_pool, rp);
    329 		rp = rpdup;
    330 		goto found;
    331 	}
    332 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    333 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
    334 	nfsrv_unlockcache(rp);
    335 	simple_unlock(&nfsrv_reqcache_lock);
    336 	return RC_DOIT;
    337 }
    338 
    339 /*
    340  * Update a request cache entry after the rpc has been done
    341  */
    342 void
    343 nfsrv_updatecache(nd, repvalid, repmbuf)
    344 	struct nfsrv_descript *nd;
    345 	int repvalid;
    346 	struct mbuf *repmbuf;
    347 {
    348 	struct nfsrvcache *rp;
    349 
    350 	simple_lock(&nfsrv_reqcache_lock);
    351 	rp = nfsrv_lookupcache(nd);
    352 	simple_unlock(&nfsrv_reqcache_lock);
    353 	if (rp) {
    354 		cleanentry(rp);
    355 		rp->rc_state = RC_DONE;
    356 		/*
    357 		 * If we have a valid reply update status and save
    358 		 * the reply for non-idempotent rpc's.
    359 		 */
    360 		if (repvalid && nonidempotent[nd->nd_procnum]) {
    361 			if ((nd->nd_flag & ND_NFSV3) == 0 &&
    362 			  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
    363 				rp->rc_status = nd->nd_repstat;
    364 				rp->rc_flag |= RC_REPSTATUS;
    365 			} else {
    366 				rp->rc_reply = m_copym(repmbuf,
    367 					0, M_COPYALL, M_WAIT);
    368 				m_claimm(rp->rc_reply, &nfsd_cache_mowner);
    369 				rp->rc_flag |= RC_REPMBUF;
    370 			}
    371 		}
    372 		simple_lock(&nfsrv_reqcache_lock);
    373 		nfsrv_unlockcache(rp);
    374 		simple_unlock(&nfsrv_reqcache_lock);
    375 	}
    376 }
    377 
    378 /*
    379  * Clean out the cache. Called when the last nfsd terminates.
    380  */
    381 void
    382 nfsrv_cleancache()
    383 {
    384 	struct nfsrvcache *rp, *nextrp;
    385 
    386 	simple_lock(&nfsrv_reqcache_lock);
    387 	for (rp = TAILQ_FIRST(&nfsrvlruhead); rp != 0; rp = nextrp) {
    388 		nextrp = TAILQ_NEXT(rp, rc_lru);
    389 		LIST_REMOVE(rp, rc_hash);
    390 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    391 		KASSERT((rp->rc_flag & (RC_LOCKED|RC_WANTED)) == 0);
    392 		cleanentry(rp);
    393 		pool_put(&nfs_reqcache_pool, rp);
    394 	}
    395 	numnfsrvcache = 0;
    396 	simple_unlock(&nfsrv_reqcache_lock);
    397 }
    398