sys/vfs/nfs/nfs_srvcache.c

   1 /*
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Rick Macklem at The University of Guelph.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      @(#)nfs_srvcache.c      8.3 (Berkeley) 3/30/95
  37  * $FreeBSD: src/sys/nfs/nfs_srvcache.c,v 1.21 2000/02/13 03:32:06 peter Exp $
  38  * $DragonFly: src/sys/vfs/nfs/nfs_srvcache.c,v 1.13 2008/01/05 14:02:41 swildner Exp $
  39  */
  40
  41 /*
  42  * Reference: Chet Juszczak, "Improving the Performance and Correctness
  43  *              of an NFS Server", in Proc. Winter 1989 USENIX Conference,
  44  *              pages 53-63. San Diego, February 1989.
  45  */
  46 #include <sys/param.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mount.h>
  49 #include <sys/systm.h>
  50 #include <sys/mbuf.h>
  51 #include <sys/socket.h>
  52 #include <sys/socketvar.h>      /* for dup_sockaddr */
  53
  54 #include <netinet/in.h>
  55 #include "rpcv2.h"
  56 #include "nfsproto.h"
  57 #include "nfs.h"
  58 #include "nfsrvcache.h"
  59
  60 #ifndef NFS_NOSERVER
  61 extern struct nfsstats nfsstats;
  62 extern int nfsv2_procid[NFS_NPROCS];
  63 static long numnfsrvcache;
  64 static long desirednfsrvcache = NFSRVCACHESIZ;
  65
  66 #define NFSRCHASH(xid) \
  67         (&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
  68 static LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
  69 static TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
  70 static u_long nfsrvhash;
  71
  72 #define TRUE    1
  73 #define FALSE   0
  74
  75 #define NETFAMILY(rp) \
  76                 (((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)
  77
  78 /*
  79  * Static array that defines which nfs rpc's are nonidempotent
  80  */
  81 static int nonidempotent[NFS_NPROCS] = {
  82         FALSE,
  83         FALSE,
  84         TRUE,
  85         FALSE,
  86         FALSE,
  87         FALSE,
  88         FALSE,
  89         TRUE,
  90         TRUE,
  91         TRUE,
  92         TRUE,
  93         TRUE,
  94         TRUE,
  95         TRUE,
  96         TRUE,
  97         TRUE,
  98         FALSE,
  99         FALSE,
 100         FALSE,
 101         FALSE,
 102         FALSE,
 103         FALSE,
 104         FALSE,
 105         FALSE,
 106         FALSE,
 107         FALSE,
 108 };
 109
 110 /* True iff the rpc reply is an nfs status ONLY! */
 111 static int nfsv2_repstat[NFS_NPROCS] = {
 112         FALSE,
 113         FALSE,
 114         FALSE,
 115         FALSE,
 116         FALSE,
 117         FALSE,
 118         FALSE,
 119         FALSE,
 120         FALSE,
 121         FALSE,
 122         TRUE,
 123         TRUE,
 124         TRUE,
 125         TRUE,
 126         FALSE,
 127         TRUE,
 128         FALSE,
 129         FALSE,
 130 };
 131
 132 /*
 133  * Initialize the server request cache list
 134  */
 135 void
 136 nfsrv_initcache(void)
 137 {
 138
 139         nfsrvhashtbl = hashinit(desirednfsrvcache, M_NFSD, &nfsrvhash);
 140         TAILQ_INIT(&nfsrvlruhead);
 141 }
 142
 143 /*
 144  * Look for the request in the cache
 145  * If found then
 146  *    return action and optionally reply
 147  * else
 148  *    insert it in the cache
 149  *
 150  * The rules are as follows:
 151  * - if in progress, return DROP request
 152  * - if completed within DELAY of the current time, return DROP it
 153  * - if completed a longer time ago return REPLY if the reply was cached or
 154  *   return DOIT
 155  * Update/add new request at end of lru list
 156  */
 157 int
 158 nfsrv_getcache(struct nfsrv_descript *nd, struct nfssvc_sock *slp,
 159                struct mbuf **repp)
 160 {
 161         struct nfsrvcache *rp;
 162         struct mbuf *mb;
 163         struct sockaddr_in *saddr;
 164         caddr_t bpos;
 165         int ret;
 166
 167         /*
 168          * Don't cache recent requests for reliable transport protocols.
 169          * (Maybe we should for the case of a reconnect, but..)
 170          */
 171         if (!nd->nd_nam2)
 172                 return (RC_DOIT);
 173 loop:
 174         for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
 175             rp = rp->rc_hash.le_next) {
 176             if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
 177                 netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
 178                         NFS_DPF(RC, ("H%03x", rp->rc_xid & 0xfff));
 179                         if ((rp->rc_flag & RC_LOCKED) != 0) {
 180                                 rp->rc_flag |= RC_WANTED;
 181                                 (void) tsleep((caddr_t)rp, 0, "nfsrc", 0);
 182                                 goto loop;
 183                         }
 184                         rp->rc_flag |= RC_LOCKED;
 185                         /* If not at end of LRU chain, move it there */
 186                         if (TAILQ_NEXT(rp, rc_lru) != NULL) {
 187                                 TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
 188                                 TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
 189                         }
 190                         if (rp->rc_state == RC_UNUSED)
 191                                 panic("nfsrv cache");
 192                         if (rp->rc_state == RC_INPROG) {
 193                                 nfsstats.srvcache_inproghits++;
 194                                 ret = RC_DROPIT;
 195                         } else if (rp->rc_flag & RC_REPSTATUS) {
 196                                 nfsstats.srvcache_nonidemdonehits++;
 197                                 nfs_rephead(0, nd, slp, rp->rc_status,
 198                                             repp, &mb, &bpos);
 199                                 ret = RC_REPLY;
 200                         } else if (rp->rc_flag & RC_REPMBUF) {
 201                                 nfsstats.srvcache_nonidemdonehits++;
 202                                 *repp = m_copym(rp->rc_reply, 0, M_COPYALL,
 203                                                 MB_WAIT);
 204                                 ret = RC_REPLY;
 205                         } else {
 206                                 nfsstats.srvcache_idemdonehits++;
 207                                 rp->rc_state = RC_INPROG;
 208                                 ret = RC_DOIT;
 209                         }
 210                         rp->rc_flag &= ~RC_LOCKED;
 211                         if (rp->rc_flag & RC_WANTED) {
 212                                 rp->rc_flag &= ~RC_WANTED;
 213                                 wakeup((caddr_t)rp);
 214                         }
 215                         return (ret);
 216                 }
 217         }
 218         nfsstats.srvcache_misses++;
 219         NFS_DPF(RC, ("M%03x", nd->nd_retxid & 0xfff));
 220         if (numnfsrvcache < desirednfsrvcache) {
 221                 rp = (struct nfsrvcache *)kmalloc((u_long)sizeof *rp,
 222                     M_NFSD, M_WAITOK | M_ZERO);
 223                 numnfsrvcache++;
 224                 rp->rc_flag = RC_LOCKED;
 225         } else {
 226                 rp = TAILQ_FIRST(&nfsrvlruhead);
 227                 while ((rp->rc_flag & RC_LOCKED) != 0) {
 228                         rp->rc_flag |= RC_WANTED;
 229                         (void) tsleep((caddr_t)rp, 0, "nfsrc", 0);
 230                         rp = TAILQ_FIRST(&nfsrvlruhead);
 231                 }
 232                 rp->rc_flag |= RC_LOCKED;
 233                 LIST_REMOVE(rp, rc_hash);
 234                 TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
 235                 if (rp->rc_flag & RC_REPMBUF)
 236                         m_freem(rp->rc_reply);
 237                 if (rp->rc_flag & RC_NAM)
 238                         FREE(rp->rc_nam, M_SONAME);
 239                 rp->rc_flag &= (RC_LOCKED | RC_WANTED);
 240         }
 241         TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
 242         rp->rc_state = RC_INPROG;
 243         rp->rc_xid = nd->nd_retxid;
 244         saddr = (struct sockaddr_in *)nd->nd_nam;
 245         switch (saddr->sin_family) {
 246         case AF_INET:
 247                 rp->rc_flag |= RC_INETADDR;
 248                 rp->rc_inetaddr = saddr->sin_addr.s_addr;
 249                 break;
 250         case AF_ISO:
 251         default:
 252                 rp->rc_flag |= RC_NAM;
 253                 rp->rc_nam = dup_sockaddr(nd->nd_nam);
 254                 break;
 255         };
 256         rp->rc_proc = nd->nd_procnum;
 257         LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
 258         rp->rc_flag &= ~RC_LOCKED;
 259         if (rp->rc_flag & RC_WANTED) {
 260                 rp->rc_flag &= ~RC_WANTED;
 261                 wakeup((caddr_t)rp);
 262         }
 263         return (RC_DOIT);
 264 }
 265
 266 /*
 267  * Update a request cache entry after the rpc has been done
 268  */
 269 void
 270 nfsrv_updatecache(struct nfsrv_descript *nd, int repvalid, struct mbuf *repmbuf)
 271 {
 272         struct nfsrvcache *rp;
 273
 274         if (!nd->nd_nam2)
 275                 return;
 276 loop:
 277         for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
 278             rp = rp->rc_hash.le_next) {
 279             if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
 280                 netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
 281                         NFS_DPF(RC, ("U%03x", rp->rc_xid & 0xfff));
 282                         if ((rp->rc_flag & RC_LOCKED) != 0) {
 283                                 rp->rc_flag |= RC_WANTED;
 284                                 (void) tsleep((caddr_t)rp, 0, "nfsrc", 0);
 285                                 goto loop;
 286                         }
 287                         rp->rc_flag |= RC_LOCKED;
 288                         if (rp->rc_state == RC_DONE) {
 289                                 /*
 290                                  * This can occur if the cache is too small.
 291                                  * Retransmits of the same request aren't
 292                                  * dropped so we may see the operation
 293                                  * complete more then once.
 294                                  */
 295                                 if (rp->rc_flag & RC_REPMBUF) {
 296                                         m_freem(rp->rc_reply);
 297                                         rp->rc_flag &= ~RC_REPMBUF;
 298                                 }
 299                         }
 300                         rp->rc_state = RC_DONE;
 301                         /*
 302                          * If we have a valid reply update status and save
 303                          * the reply for non-idempotent rpc's.
 304                          */
 305                         if (repvalid && nonidempotent[nd->nd_procnum]) {
 306                                 if ((nd->nd_flag & ND_NFSV3) == 0 &&
 307                                   nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
 308                                         rp->rc_status = nd->nd_repstat;
 309                                         rp->rc_flag |= RC_REPSTATUS;
 310                                 } else {
 311                                         rp->rc_reply = m_copym(repmbuf,
 312                                                 0, M_COPYALL, MB_WAIT);
 313                                         rp->rc_flag |= RC_REPMBUF;
 314                                 }
 315                         }
 316                         rp->rc_flag &= ~RC_LOCKED;
 317                         if (rp->rc_flag & RC_WANTED) {
 318                                 rp->rc_flag &= ~RC_WANTED;
 319                                 wakeup((caddr_t)rp);
 320                         }
 321                         return;
 322                 }
 323         }
 324         NFS_DPF(RC, ("L%03x", nd->nd_retxid & 0xfff));
 325 }
 326
 327 /*
 328  * Clean out the cache. Called when the last nfsd terminates.
 329  */
 330 void
 331 nfsrv_cleancache(void)
 332 {
 333         struct nfsrvcache *rp, *nextrp;
 334
 335         TAILQ_FOREACH_MUTABLE(rp, &nfsrvlruhead, rc_lru, nextrp) {
 336                 LIST_REMOVE(rp, rc_hash);
 337                 TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
 338                 if (rp->rc_flag & RC_REPMBUF)
 339                         m_freem(rp->rc_reply);
 340                 if (rp->rc_flag & RC_NAM)
 341                         kfree(rp->rc_nam, M_SONAME);
 342                 kfree(rp, M_NFSD);
 343         }
 344         numnfsrvcache = 0;
 345 }
 346
 347 #endif /* NFS_NOSERVER */