Home | History | Annotate | Line # | Download | only in nfs
nfs_srvcache.c revision 1.41
      1 /*	$NetBSD: nfs_srvcache.c,v 1.41 2007/12/04 17:42:31 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 1.41 2007/12/04 17:42:31 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_flags & 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_flags & RC_REPMBUF) != 0) {
    152 		m_freem(rp->rc_reply);
    153 	}
    154 	if ((rp->rc_flags & RC_NAM) != 0) {
    155 		m_free(rp->rc_nam);
    156 	}
    157 	rp->rc_flags &= ~(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_DEFAULT, 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_gflags & RC_G_LOCKED) != 0) {
    193 				cv_wait(&rp->rc_cv, &nfsrv_reqcache_lock);
    194 				goto loop;
    195 			}
    196 			rp->rc_gflags |= RC_G_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 	KASSERT((rp->rc_gflags & RC_G_LOCKED) != 0);
    215 	rp->rc_gflags &= ~RC_G_LOCKED;
    216 	cv_broadcast(&rp->rc_cv);
    217 }
    218 
    219 /*
    220  * Look for the request in the cache
    221  * If found then
    222  *    return action and optionally reply
    223  * else
    224  *    insert it in the cache
    225  *
    226  * The rules are as follows:
    227  * - if in progress, return DROP request
    228  * - if completed within DELAY of the current time, return DROP it
    229  * - if completed a longer time ago return REPLY if the reply was cached or
    230  *   return DOIT
    231  * Update/add new request at end of lru list
    232  */
    233 int
    234 nfsrv_getcache(nd, slp, repp)
    235 	struct nfsrv_descript *nd;
    236 	struct nfssvc_sock *slp;
    237 	struct mbuf **repp;
    238 {
    239 	struct nfsrvcache *rp, *rpdup;
    240 	struct mbuf *mb;
    241 	struct sockaddr_in *saddr;
    242 	char *bpos;
    243 	int ret;
    244 
    245 	mutex_enter(&nfsrv_reqcache_lock);
    246 	rp = nfsrv_lookupcache(nd);
    247 	if (rp) {
    248 		mutex_exit(&nfsrv_reqcache_lock);
    249 found:
    250 		/* If not at end of LRU chain, move it there */
    251 		if (TAILQ_NEXT(rp, rc_lru)) { /* racy but ok */
    252 			mutex_enter(&nfsrv_reqcache_lock);
    253 			TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    254 			TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    255 			mutex_exit(&nfsrv_reqcache_lock);
    256 		}
    257 		if (rp->rc_state == RC_UNUSED)
    258 			panic("nfsrv cache");
    259 		if (rp->rc_state == RC_INPROG) {
    260 			nfsstats.srvcache_inproghits++;
    261 			ret = RC_DROPIT;
    262 		} else if (rp->rc_flags & RC_REPSTATUS) {
    263 			nfsstats.srvcache_nonidemdonehits++;
    264 			nfs_rephead(0, nd, slp, rp->rc_status,
    265 			   0, (u_quad_t *)0, repp, &mb, &bpos);
    266 			ret = RC_REPLY;
    267 		} else if (rp->rc_flags & RC_REPMBUF) {
    268 			nfsstats.srvcache_nonidemdonehits++;
    269 			*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
    270 					M_WAIT);
    271 			ret = RC_REPLY;
    272 		} else {
    273 			nfsstats.srvcache_idemdonehits++;
    274 			rp->rc_state = RC_INPROG;
    275 			ret = RC_DOIT;
    276 		}
    277 		mutex_enter(&nfsrv_reqcache_lock);
    278 		nfsrv_unlockcache(rp);
    279 		mutex_exit(&nfsrv_reqcache_lock);
    280 		return ret;
    281 	}
    282 	nfsstats.srvcache_misses++;
    283 	if (numnfsrvcache < desirednfsrvcache) {
    284 		numnfsrvcache++;
    285 		mutex_exit(&nfsrv_reqcache_lock);
    286 		rp = pool_get(&nfs_reqcache_pool, PR_WAITOK);
    287 		memset(rp, 0, sizeof *rp);
    288 		cv_init(&rp->rc_cv, "nfsdrc");
    289 		rp->rc_gflags = RC_G_LOCKED;
    290 	} else {
    291 		rp = TAILQ_FIRST(&nfsrvlruhead);
    292 		while ((rp->rc_gflags & RC_G_LOCKED) != 0) {
    293 			cv_wait(&rp->rc_cv, &nfsrv_reqcache_lock);
    294 			rp = TAILQ_FIRST(&nfsrvlruhead);
    295 		}
    296 		rp->rc_gflags |= RC_G_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_flags = 0;
    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_flags |= RC_INETADDR;
    309 		rp->rc_inetaddr = saddr->sin_addr.s_addr;
    310 		break;
    311 	case AF_ISO:
    312 	default:
    313 		rp->rc_flags |= 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 		KASSERT(numnfsrvcache > 0);
    326 		numnfsrvcache--;
    327 		mutex_exit(&nfsrv_reqcache_lock);
    328 		cleanentry(rp);
    329 		cv_destroy(&rp->rc_cv);
    330 		pool_put(&nfs_reqcache_pool, rp);
    331 		rp = rpdup;
    332 		goto found;
    333 	}
    334 	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
    335 	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
    336 	nfsrv_unlockcache(rp);
    337 	mutex_exit(&nfsrv_reqcache_lock);
    338 	return RC_DOIT;
    339 }
    340 
    341 /*
    342  * Update a request cache entry after the rpc has been done
    343  */
    344 void
    345 nfsrv_updatecache(nd, repvalid, repmbuf)
    346 	struct nfsrv_descript *nd;
    347 	int repvalid;
    348 	struct mbuf *repmbuf;
    349 {
    350 	struct nfsrvcache *rp;
    351 
    352 	mutex_enter(&nfsrv_reqcache_lock);
    353 	rp = nfsrv_lookupcache(nd);
    354 	mutex_exit(&nfsrv_reqcache_lock);
    355 	if (rp) {
    356 		cleanentry(rp);
    357 		rp->rc_state = RC_DONE;
    358 		/*
    359 		 * If we have a valid reply update status and save
    360 		 * the reply for non-idempotent rpc's.
    361 		 */
    362 		if (repvalid && nonidempotent[nd->nd_procnum]) {
    363 			if ((nd->nd_flag & ND_NFSV3) == 0 &&
    364 			  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
    365 				rp->rc_status = nd->nd_repstat;
    366 				rp->rc_flags |= RC_REPSTATUS;
    367 			} else {
    368 				rp->rc_reply = m_copym(repmbuf,
    369 					0, M_COPYALL, M_WAIT);
    370 				m_claimm(rp->rc_reply, &nfsd_cache_mowner);
    371 				rp->rc_flags |= RC_REPMBUF;
    372 			}
    373 		}
    374 		mutex_enter(&nfsrv_reqcache_lock);
    375 		nfsrv_unlockcache(rp);
    376 		mutex_exit(&nfsrv_reqcache_lock);
    377 	}
    378 }
    379 
    380 /*
    381  * Clean out the cache. Called when the last nfsd terminates.
    382  */
    383 void
    384 nfsrv_cleancache()
    385 {
    386 	struct nfsrvcache *rp;
    387 
    388 	mutex_enter(&nfsrv_reqcache_lock);
    389 	while ((rp = TAILQ_FIRST(&nfsrvlruhead)) != NULL) {
    390 		KASSERT((rp->rc_gflags & RC_G_LOCKED) == 0);
    391 		LIST_REMOVE(rp, rc_hash);
    392 		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
    393 		KASSERT(numnfsrvcache > 0);
    394 		numnfsrvcache--;
    395 		mutex_exit(&nfsrv_reqcache_lock);
    396 		cleanentry(rp);
    397 		cv_destroy(&rp->rc_cv);
    398 		pool_put(&nfs_reqcache_pool, rp);
    399 		mutex_enter(&nfsrv_reqcache_lock);
    400 	}
    401 	KASSERT(numnfsrvcache == 0);
    402 	mutex_exit(&nfsrv_reqcache_lock);
    403 }
    404