Home | History | Annotate | Line # | Download | only in nfs
nfs_srvcache.c revision 1.34
      1 /*	$NetBSD: nfs_srvcache.c,v 1.34 2007/01/17 12:40:08 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 1.34 2007/01/17 12:40:08 yamt 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 #define	NETFAMILY(rp) \
     85 		(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
     86 
     87 static struct nfsrvcache *nfsrv_lookupcache(struct nfsrv_descript *nd);
     88 static void nfsrv_unlockcache(struct nfsrvcache *rp);
     89 
     90 /*
     91  * Static array that defines which nfs rpc's are nonidempotent
     92  */
     93 const int nonidempotent[NFS_NPROCS] = {
     94 	FALSE,	/* NULL */
     95 	FALSE,	/* GETATTR */
     96 	TRUE,	/* SETATTR */
     97 	FALSE,	/* LOOKUP */
     98 	FALSE,	/* ACCESS */
     99 	FALSE,	/* READLINK */
    100 	FALSE,	/* READ */
    101 	TRUE,	/* WRITE */
    102 	TRUE,	/* CREATE */
    103 	TRUE,	/* MKDIR */
    104 	TRUE,	/* SYMLINK */
    105 	TRUE,	/* MKNOD */
    106 	TRUE,	/* REMOVE */
    107 	TRUE,	/* RMDIR */
    108 	TRUE,	/* RENAME */
    109 	TRUE,	/* LINK */
    110 	FALSE,	/* READDIR */
    111 	FALSE,	/* READDIRPLUS */
    112 	FALSE,	/* FSSTAT */
    113 	FALSE,	/* FSINFO */
    114 	FALSE,	/* PATHCONF */
    115 	FALSE,	/* COMMIT */
    116 	FALSE,	/* NOOP */
    117 };
    118 
    119 /* True iff the rpc reply is an nfs status ONLY! */
    120 static const int nfsv2_repstat[NFS_NPROCS] = {
    121 	FALSE,	/* NULL */
    122 	FALSE,	/* GETATTR */
    123 	FALSE,	/* SETATTR */
    124 	FALSE,	/* NOOP */
    125 	FALSE,	/* LOOKUP */
    126 	FALSE,	/* READLINK */
    127 	FALSE,	/* READ */
    128 	FALSE,	/* Obsolete WRITECACHE */
    129 	FALSE,	/* WRITE */
    130 	FALSE,	/* CREATE */
    131 	TRUE,	/* REMOVE */
    132 	TRUE,	/* RENAME */
    133 	TRUE,	/* LINK */
    134 	TRUE,	/* SYMLINK */
    135 	FALSE,	/* MKDIR */
    136 	TRUE,	/* RMDIR */
    137 	FALSE,	/* READDIR */
    138 	FALSE,	/* STATFS */
    139 };
    140 
    141 static void
    142 cleanentry(struct nfsrvcache *rp)
    143 {
    144 
    145 	if ((rp->rc_flag & RC_REPMBUF) != 0) {
    146 		m_freem(rp->rc_reply);
    147 	}
    148 	if ((rp->rc_flag & RC_NAM) != 0) {
    149 		m_free(rp->rc_nam);
    150 	}
    151 	rp->rc_flag &= ~(RC_REPSTATUS|RC_REPMBUF);
    152 }
    153 
    154 /*
    155  * Initialize the server request cache list
    156  */
    157 void
    158 nfsrv_initcache()
    159 {
    160 
    161 	nfsrvhashtbl = hashinit(desirednfsrvcache, HASH_LIST, M_NFSD,
    162 	    M_WAITOK, &nfsrvhash);
    163 	TAILQ_INIT(&nfsrvlruhead);
    164 	pool_init(&nfs_reqcache_pool, sizeof(struct nfsrvcache), 0, 0, 0,
    165 	    "nfsreqcachepl", &pool_allocator_nointr);
    166 }
    167 
    168 /*
    169  * Lookup a cache and lock it
    170  */
    171 static struct nfsrvcache *
    172 nfsrv_lookupcache(nd)
    173 	struct nfsrv_descript *nd;
    174 {
    175 	struct nfsrvcache *rp;
    176 
    177 	LOCK_ASSERT(simple_lock_held(&nfsrv_reqcache_lock));
    178 
    179 loop:
    180 	LIST_FOREACH(rp, NFSRCHASH(nd->nd_retxid), rc_hash) {
    181 		if (nd->nd_retxid == rp->rc_xid &&
    182 		    nd->nd_procnum == rp->rc_proc &&
    183 		    netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
    184 			if ((rp->rc_flag & RC_LOCKED) != 0) {
    185 				rp->rc_flag |= RC_WANTED;
    186 				(void) ltsleep(rp, PZERO - 1, "nfsrc", 0,
    187 				    &nfsrv_reqcache_lock);
    188 				goto loop;
    189 			}
    190 			rp->rc_flag |= RC_LOCKED;
    191 			break;
    192 		}
    193 	}
    194 
    195 	return rp;
    196 }
    197 
    198 /*
    199  * Unlock a cache
    200  */
    201 static void
    202 nfsrv_unlockcache(rp)
    203 	struct nfsrvcache *rp;
    204 {
    205 
    206 	LOCK_ASSERT(simple_lock_held(&nfsrv_reqcache_lock));
    207 
    208 	rp->rc_flag &= ~RC_LOCKED;
    209 	if (rp->rc_flag & RC_WANTED) {
    210 		rp->rc_flag &= ~RC_WANTED;
    211 		wakeup(rp);
    212 	}
    213 }
    214 
    215 /*
    216  * Look for the request in the cache
    217  * If found then
    218  *    return action and optionally reply
    219  * else
    220  *    insert it in the cache
    221  *
    222  * The rules are as follows:
    223  * - if in progress, return DROP request
    224  * - if completed within DELAY of the current time, return DROP it
    225  * - if completed a longer time ago return REPLY if the reply was cached or
    226  *   return DOIT
    227  * Update/add new request at end of lru list
    228  */
    229 int
    230 nfsrv_getcache(nd, slp, repp)
    231 	struct nfsrv_descript *nd;
    232 	struct nfssvc_sock *slp;
    233 	struct mbuf **repp;
    234 {
    235 	struct nfsrvcache *rp, *rpdup;
    236 	struct mbuf *mb;
    237 	struct sockaddr_in *saddr;
    238 	caddr_t bpos;
    239 	int ret;
    240 
    241 	simple_lock(&nfsrv_reqcache_lock);
    242 	rp = nfsrv_lookupcache(nd);
    243 	if (rp) {
    244 		simple_unlock(&nfsrv_reqcache_lock);
    245 found:
    246 		/* If not at end of LRU chain, move it there */
    247 		if (TAILQ_NEXT(rp, rc_lru)) { /* racy but ok */
    248 			simple_lock(&nfsrv_reqcache_lock);
    249 			TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    250 			TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    251 			simple_unlock(&nfsrv_reqcache_lock);
    252 		}
    253 		if (rp->rc_state == RC_UNUSED)
    254 			panic("nfsrv cache");
    255 		if (rp->rc_state == RC_INPROG) {
    256 			nfsstats.srvcache_inproghits++;
    257 			ret = RC_DROPIT;
    258 		} else if (rp->rc_flag & RC_REPSTATUS) {
    259 			nfsstats.srvcache_nonidemdonehits++;
    260 			nfs_rephead(0, nd, slp, rp->rc_status,
    261 			   0, (u_quad_t *)0, repp, &mb, &bpos);
    262 			ret = RC_REPLY;
    263 		} else if (rp->rc_flag & RC_REPMBUF) {
    264 			nfsstats.srvcache_nonidemdonehits++;
    265 			*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
    266 					M_WAIT);
    267 			ret = RC_REPLY;
    268 		} else {
    269 			nfsstats.srvcache_idemdonehits++;
    270 			rp->rc_state = RC_INPROG;
    271 			ret = RC_DOIT;
    272 		}
    273 		simple_lock(&nfsrv_reqcache_lock);
    274 		nfsrv_unlockcache(rp);
    275 		simple_unlock(&nfsrv_reqcache_lock);
    276 		return ret;
    277 	}
    278 	nfsstats.srvcache_misses++;
    279 	if (numnfsrvcache < desirednfsrvcache) {
    280 		numnfsrvcache++;
    281 		simple_unlock(&nfsrv_reqcache_lock);
    282 		rp = pool_get(&nfs_reqcache_pool, PR_WAITOK);
    283 		memset(rp, 0, sizeof *rp);
    284 		rp->rc_flag = RC_LOCKED;
    285 	} else {
    286 		rp = TAILQ_FIRST(&nfsrvlruhead);
    287 		while ((rp->rc_flag & RC_LOCKED) != 0) {
    288 			rp->rc_flag |= RC_WANTED;
    289 			(void) ltsleep(rp, PZERO-1, "nfsrc", 0,
    290 			    &nfsrv_reqcache_lock);
    291 			rp = TAILQ_FIRST(&nfsrvlruhead);
    292 		}
    293 		rp->rc_flag |= RC_LOCKED;
    294 		LIST_REMOVE(rp, rc_hash);
    295 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    296 		simple_unlock(&nfsrv_reqcache_lock);
    297 		cleanentry(rp);
    298 		rp->rc_flag &= (RC_LOCKED | RC_WANTED);
    299 	}
    300 	rp->rc_state = RC_INPROG;
    301 	rp->rc_xid = nd->nd_retxid;
    302 	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
    303 	switch (saddr->sin_family) {
    304 	case AF_INET:
    305 		rp->rc_flag |= RC_INETADDR;
    306 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
    307 		break;
    308 	case AF_ISO:
    309 	default:
    310 		rp->rc_flag |= RC_NAM;
    311 		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
    312 		break;
    313 	};
    314 	rp->rc_proc = nd->nd_procnum;
    315 	simple_lock(&nfsrv_reqcache_lock);
    316 	rpdup = nfsrv_lookupcache(nd);
    317 	if (rpdup != NULL) {
    318 		/*
    319 		 * other thread made duplicate cache entry.
    320 		 */
    321 		simple_unlock(&nfsrv_reqcache_lock);
    322 		pool_put(&nfs_reqcache_pool, rp);
    323 		rp = rpdup;
    324 		goto found;
    325 	}
    326 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    327 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
    328 	nfsrv_unlockcache(rp);
    329 	simple_unlock(&nfsrv_reqcache_lock);
    330 	return RC_DOIT;
    331 }
    332 
    333 /*
    334  * Update a request cache entry after the rpc has been done
    335  */
    336 void
    337 nfsrv_updatecache(nd, repvalid, repmbuf)
    338 	struct nfsrv_descript *nd;
    339 	int repvalid;
    340 	struct mbuf *repmbuf;
    341 {
    342 	struct nfsrvcache *rp;
    343 
    344 	if (!nd->nd_nam2)
    345 		return;
    346 	simple_lock(&nfsrv_reqcache_lock);
    347 	rp = nfsrv_lookupcache(nd);
    348 	simple_unlock(&nfsrv_reqcache_lock);
    349 	if (rp) {
    350 		cleanentry(rp);
    351 		rp->rc_state = RC_DONE;
    352 		/*
    353 		 * If we have a valid reply update status and save
    354 		 * the reply for non-idempotent rpc's.
    355 		 */
    356 		if (repvalid && nonidempotent[nd->nd_procnum]) {
    357 			if ((nd->nd_flag & ND_NFSV3) == 0 &&
    358 			  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
    359 				rp->rc_status = nd->nd_repstat;
    360 				rp->rc_flag |= RC_REPSTATUS;
    361 			} else {
    362 				rp->rc_reply = m_copym(repmbuf,
    363 					0, M_COPYALL, M_WAIT);
    364 				rp->rc_flag |= RC_REPMBUF;
    365 			}
    366 		}
    367 		simple_lock(&nfsrv_reqcache_lock);
    368 		nfsrv_unlockcache(rp);
    369 		simple_unlock(&nfsrv_reqcache_lock);
    370 	}
    371 }
    372 
    373 /*
    374  * Clean out the cache. Called when the last nfsd terminates.
    375  */
    376 void
    377 nfsrv_cleancache()
    378 {
    379 	struct nfsrvcache *rp, *nextrp;
    380 
    381 	simple_lock(&nfsrv_reqcache_lock);
    382 	for (rp = TAILQ_FIRST(&nfsrvlruhead); rp != 0; rp = nextrp) {
    383 		nextrp = TAILQ_NEXT(rp, rc_lru);
    384 		LIST_REMOVE(rp, rc_hash);
    385 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    386 		KASSERT((rp->rc_flag & (RC_LOCKED|RC_WANTED)) == 0);
    387 		cleanentry(rp);
    388 		pool_put(&nfs_reqcache_pool, rp);
    389 	}
    390 	numnfsrvcache = 0;
    391 	simple_unlock(&nfsrv_reqcache_lock);
    392 }
    393