Home | History | Annotate | Line # | Download | only in nfs
nfs_srvcache.c revision 1.42
      1 /*	$NetBSD: nfs_srvcache.c,v 1.42 2008/05/05 17:11:17 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 1.42 2008/05/05 17:11:17 ad 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/mutex.h>
     59 #include <sys/socket.h>
     60 #include <sys/socketvar.h>
     61 
     62 #include <netinet/in.h>
     63 #ifdef ISO
     64 #include <netiso/iso.h>
     65 #endif
     66 #include <nfs/nfsm_subs.h>
     67 #include <nfs/rpcv2.h>
     68 #include <nfs/nfsproto.h>
     69 #include <nfs/nfs.h>
     70 #include <nfs/nfsrvcache.h>
     71 #include <nfs/nfs_var.h>
     72 
     73 extern struct nfsstats nfsstats;
     74 extern const int nfsv2_procid[NFS_NPROCS];
     75 long numnfsrvcache, desirednfsrvcache = NFSRVCACHESIZ;
     76 struct pool nfs_reqcache_pool;
     77 
     78 #define	NFSRCHASH(xid) \
     79 	(&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
     80 LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
     81 TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
     82 kmutex_t nfsrv_reqcache_lock;
     83 u_long nfsrvhash;
     84 
     85 #if defined(MBUFTRACE)
     86 static struct mowner nfsd_cache_mowner = MOWNER_INIT("nfsd", "cache");
     87 #endif /* defined(MBUFTRACE) */
     88 
     89 #define	NETFAMILY(rp) \
     90 		(((rp)->rc_flags & RC_INETADDR) ? AF_INET : AF_ISO)
     91 
     92 static struct nfsrvcache *nfsrv_lookupcache(struct nfsrv_descript *nd);
     93 static void nfsrv_unlockcache(struct nfsrvcache *rp);
     94 
     95 /*
     96  * Static array that defines which nfs rpc's are nonidempotent
     97  */
     98 const int nonidempotent[NFS_NPROCS] = {
     99 	false,	/* NULL */
    100 	false,	/* GETATTR */
    101 	true,	/* SETATTR */
    102 	false,	/* LOOKUP */
    103 	false,	/* ACCESS */
    104 	false,	/* READLINK */
    105 	false,	/* READ */
    106 	true,	/* WRITE */
    107 	true,	/* CREATE */
    108 	true,	/* MKDIR */
    109 	true,	/* SYMLINK */
    110 	true,	/* MKNOD */
    111 	true,	/* REMOVE */
    112 	true,	/* RMDIR */
    113 	true,	/* RENAME */
    114 	true,	/* LINK */
    115 	false,	/* READDIR */
    116 	false,	/* READDIRPLUS */
    117 	false,	/* FSSTAT */
    118 	false,	/* FSINFO */
    119 	false,	/* PATHCONF */
    120 	false,	/* COMMIT */
    121 	false,	/* NOOP */
    122 };
    123 
    124 /* True iff the rpc reply is an nfs status ONLY! */
    125 static const int nfsv2_repstat[NFS_NPROCS] = {
    126 	false,	/* NULL */
    127 	false,	/* GETATTR */
    128 	false,	/* SETATTR */
    129 	false,	/* NOOP */
    130 	false,	/* LOOKUP */
    131 	false,	/* READLINK */
    132 	false,	/* READ */
    133 	false,	/* Obsolete WRITECACHE */
    134 	false,	/* WRITE */
    135 	false,	/* CREATE */
    136 	true,	/* REMOVE */
    137 	true,	/* RENAME */
    138 	true,	/* LINK */
    139 	true,	/* SYMLINK */
    140 	false,	/* MKDIR */
    141 	true,	/* RMDIR */
    142 	false,	/* READDIR */
    143 	false,	/* STATFS */
    144 };
    145 
    146 static void
    147 cleanentry(struct nfsrvcache *rp)
    148 {
    149 
    150 	if ((rp->rc_flags & RC_REPMBUF) != 0) {
    151 		m_freem(rp->rc_reply);
    152 	}
    153 	if ((rp->rc_flags & RC_NAM) != 0) {
    154 		m_free(rp->rc_nam);
    155 	}
    156 	rp->rc_flags &= ~(RC_REPSTATUS|RC_REPMBUF);
    157 }
    158 
    159 /*
    160  * Initialize the server request cache list
    161  */
    162 void
    163 nfsrv_initcache()
    164 {
    165 
    166 	mutex_init(&nfsrv_reqcache_lock, MUTEX_DEFAULT, IPL_NONE);
    167 	nfsrvhashtbl = hashinit(desirednfsrvcache, HASH_LIST, true,
    168 	    &nfsrvhash);
    169 	TAILQ_INIT(&nfsrvlruhead);
    170 	pool_init(&nfs_reqcache_pool, sizeof(struct nfsrvcache), 0, 0, 0,
    171 	    "nfsreqcachepl", &pool_allocator_nointr, IPL_NONE);
    172 	MOWNER_ATTACH(&nfsd_cache_mowner);
    173 }
    174 
    175 /*
    176  * Lookup a cache and lock it
    177  */
    178 static struct nfsrvcache *
    179 nfsrv_lookupcache(nd)
    180 	struct nfsrv_descript *nd;
    181 {
    182 	struct nfsrvcache *rp;
    183 
    184 	KASSERT(mutex_owned(&nfsrv_reqcache_lock));
    185 
    186 loop:
    187 	LIST_FOREACH(rp, NFSRCHASH(nd->nd_retxid), rc_hash) {
    188 		if (nd->nd_retxid == rp->rc_xid &&
    189 		    nd->nd_procnum == rp->rc_proc &&
    190 		    netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
    191 			if ((rp->rc_gflags & RC_G_LOCKED) != 0) {
    192 				cv_wait(&rp->rc_cv, &nfsrv_reqcache_lock);
    193 				goto loop;
    194 			}
    195 			rp->rc_gflags |= RC_G_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 	KASSERT(mutex_owned(&nfsrv_reqcache_lock));
    212 
    213 	KASSERT((rp->rc_gflags & RC_G_LOCKED) != 0);
    214 	rp->rc_gflags &= ~RC_G_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_flags & 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_flags & 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_gflags = RC_G_LOCKED;
    289 	} else {
    290 		rp = TAILQ_FIRST(&nfsrvlruhead);
    291 		while ((rp->rc_gflags & RC_G_LOCKED) != 0) {
    292 			cv_wait(&rp->rc_cv, &nfsrv_reqcache_lock);
    293 			rp = TAILQ_FIRST(&nfsrvlruhead);
    294 		}
    295 		rp->rc_gflags |= RC_G_LOCKED;
    296 		LIST_REMOVE(rp, rc_hash);
    297 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    298 		mutex_exit(&nfsrv_reqcache_lock);
    299 		cleanentry(rp);
    300 		rp->rc_flags = 0;
    301 	}
    302 	rp->rc_state = RC_INPROG;
    303 	rp->rc_xid = nd->nd_retxid;
    304 	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
    305 	switch (saddr->sin_family) {
    306 	case AF_INET:
    307 		rp->rc_flags |= RC_INETADDR;
    308 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
    309 		break;
    310 	case AF_ISO:
    311 	default:
    312 		rp->rc_flags |= RC_NAM;
    313 		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
    314 		m_claimm(rp->rc_nam, &nfsd_cache_mowner);
    315 		break;
    316 	};
    317 	rp->rc_proc = nd->nd_procnum;
    318 	mutex_enter(&nfsrv_reqcache_lock);
    319 	rpdup = nfsrv_lookupcache(nd);
    320 	if (rpdup != NULL) {
    321 		/*
    322 		 * other thread made duplicate cache entry.
    323 		 */
    324 		KASSERT(numnfsrvcache > 0);
    325 		numnfsrvcache--;
    326 		mutex_exit(&nfsrv_reqcache_lock);
    327 		cleanentry(rp);
    328 		cv_destroy(&rp->rc_cv);
    329 		pool_put(&nfs_reqcache_pool, rp);
    330 		rp = rpdup;
    331 		goto found;
    332 	}
    333 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    334 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
    335 	nfsrv_unlockcache(rp);
    336 	mutex_exit(&nfsrv_reqcache_lock);
    337 	return RC_DOIT;
    338 }
    339 
    340 /*
    341  * Update a request cache entry after the rpc has been done
    342  */
    343 void
    344 nfsrv_updatecache(nd, repvalid, repmbuf)
    345 	struct nfsrv_descript *nd;
    346 	int repvalid;
    347 	struct mbuf *repmbuf;
    348 {
    349 	struct nfsrvcache *rp;
    350 
    351 	mutex_enter(&nfsrv_reqcache_lock);
    352 	rp = nfsrv_lookupcache(nd);
    353 	mutex_exit(&nfsrv_reqcache_lock);
    354 	if (rp) {
    355 		cleanentry(rp);
    356 		rp->rc_state = RC_DONE;
    357 		/*
    358 		 * If we have a valid reply update status and save
    359 		 * the reply for non-idempotent rpc's.
    360 		 */
    361 		if (repvalid && nonidempotent[nd->nd_procnum]) {
    362 			if ((nd->nd_flag & ND_NFSV3) == 0 &&
    363 			  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
    364 				rp->rc_status = nd->nd_repstat;
    365 				rp->rc_flags |= RC_REPSTATUS;
    366 			} else {
    367 				rp->rc_reply = m_copym(repmbuf,
    368 					0, M_COPYALL, M_WAIT);
    369 				m_claimm(rp->rc_reply, &nfsd_cache_mowner);
    370 				rp->rc_flags |= RC_REPMBUF;
    371 			}
    372 		}
    373 		mutex_enter(&nfsrv_reqcache_lock);
    374 		nfsrv_unlockcache(rp);
    375 		mutex_exit(&nfsrv_reqcache_lock);
    376 	}
    377 }
    378 
    379 /*
    380  * Clean out the cache. Called when the last nfsd terminates.
    381  */
    382 void
    383 nfsrv_cleancache()
    384 {
    385 	struct nfsrvcache *rp;
    386 
    387 	mutex_enter(&nfsrv_reqcache_lock);
    388 	while ((rp = TAILQ_FIRST(&nfsrvlruhead)) != NULL) {
    389 		KASSERT((rp->rc_gflags & RC_G_LOCKED) == 0);
    390 		LIST_REMOVE(rp, rc_hash);
    391 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    392 		KASSERT(numnfsrvcache > 0);
    393 		numnfsrvcache--;
    394 		mutex_exit(&nfsrv_reqcache_lock);
    395 		cleanentry(rp);
    396 		cv_destroy(&rp->rc_cv);
    397 		pool_put(&nfs_reqcache_pool, rp);
    398 		mutex_enter(&nfsrv_reqcache_lock);
    399 	}
    400 	KASSERT(numnfsrvcache == 0);
    401 	mutex_exit(&nfsrv_reqcache_lock);
    402 }
    403