bsd/nfs/nfs_srvcache.c

   1 /*
   2  * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
  29 /*
  30  * Copyright (c) 1989, 1993
  31  *      The Regents of the University of California.  All rights reserved.
  32  *
  33  * This code is derived from software contributed to Berkeley by
  34  * Rick Macklem at The University of Guelph.
  35  *
  36  * Redistribution and use in source and binary forms, with or without
  37  * modification, are permitted provided that the following conditions
  38  * are met:
  39  * 1. Redistributions of source code must retain the above copyright
  40  *    notice, this list of conditions and the following disclaimer.
  41  * 2. Redistributions in binary form must reproduce the above copyright
  42  *    notice, this list of conditions and the following disclaimer in the
  43  *    documentation and/or other materials provided with the distribution.
  44  * 3. All advertising materials mentioning features or use of this software
  45  *    must display the following acknowledgement:
  46  *      This product includes software developed by the University of
  47  *      California, Berkeley and its contributors.
  48  * 4. Neither the name of the University nor the names of its contributors
  49  *    may be used to endorse or promote products derived from this software
  50  *    without specific prior written permission.
  51  *
  52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  62  * SUCH DAMAGE.
  63  *
  64  *      @(#)nfs_srvcache.c      8.3 (Berkeley) 3/30/95
  65  * FreeBSD-Id: nfs_srvcache.c,v 1.15 1997/10/12 20:25:46 phk Exp $
  66  */
  67
  68 #if NFSSERVER
  69 /*
  70  * Reference: Chet Juszczak, "Improving the Performance and Correctness
  71  *              of an NFS Server", in Proc. Winter 1989 USENIX Conference,
  72  *              pages 53-63. San Diego, February 1989.
  73  */
  74 #include <sys/param.h>
  75 #include <sys/vnode.h>
  76 #include <sys/mount_internal.h>
  77 #include <sys/kernel.h>
  78 #include <sys/systm.h>
  79 #include <sys/proc.h>
  80 #include <sys/kpi_mbuf.h>
  81 #include <sys/malloc.h>
  82 #include <sys/socket.h>
  83 #include <libkern/OSAtomic.h>
  84
  85 #include <netinet/in.h>
  86 #include <nfs/rpcv2.h>
  87 #include <nfs/nfsproto.h>
  88 #include <nfs/nfs.h>
  89 #include <nfs/nfsrvcache.h>
  90
  91 extern int nfsv2_procid[NFS_NPROCS];
  92 static int nfsrv_reqcache_count;
  93 int nfsrv_reqcache_size = NFSRVCACHESIZ;
  94
  95 #define NFSRCHASH(xid) \
  96         (&nfsrv_reqcache_hashtbl[((xid) + ((xid) >> 24)) & nfsrv_reqcache_hash])
  97 LIST_HEAD(nfsrv_reqcache_hash, nfsrvcache) *nfsrv_reqcache_hashtbl;
  98 TAILQ_HEAD(nfsrv_reqcache_lru, nfsrvcache) nfsrv_reqcache_lruhead;
  99 u_long nfsrv_reqcache_hash;
 100
 101 lck_grp_t *nfsrv_reqcache_lck_grp;
 102 lck_mtx_t *nfsrv_reqcache_mutex;
 103
 104 /*
 105  * Static array that defines which nfs rpc's are nonidempotent
 106  */
 107 static int nonidempotent[NFS_NPROCS] = {
 108         FALSE,
 109         FALSE,
 110         TRUE,
 111         FALSE,
 112         FALSE,
 113         FALSE,
 114         FALSE,
 115         TRUE,
 116         TRUE,
 117         TRUE,
 118         TRUE,
 119         TRUE,
 120         TRUE,
 121         TRUE,
 122         TRUE,
 123         TRUE,
 124         FALSE,
 125         FALSE,
 126         FALSE,
 127         FALSE,
 128         FALSE,
 129         FALSE,
 130         FALSE,
 131 };
 132
 133 /* True iff the rpc reply is an nfs status ONLY! */
 134 static int nfsv2_repstat[NFS_NPROCS] = {
 135         FALSE,
 136         FALSE,
 137         FALSE,
 138         FALSE,
 139         FALSE,
 140         FALSE,
 141         FALSE,
 142         FALSE,
 143         FALSE,
 144         FALSE,
 145         TRUE,
 146         TRUE,
 147         TRUE,
 148         TRUE,
 149         FALSE,
 150         TRUE,
 151         FALSE,
 152         FALSE,
 153 };
 154
 155 /*
 156  * Initialize the server request cache list
 157  */
 158 void
 159 nfsrv_initcache(void)
 160 {
 161         if (nfsrv_reqcache_size <= 0)
 162                 return;
 163
 164         lck_mtx_lock(nfsrv_reqcache_mutex);
 165         /* init nfs server request cache hash table */
 166         nfsrv_reqcache_hashtbl = hashinit(nfsrv_reqcache_size, M_NFSD, &nfsrv_reqcache_hash);
 167         TAILQ_INIT(&nfsrv_reqcache_lruhead);
 168         lck_mtx_unlock(nfsrv_reqcache_mutex);
 169 }
 170
 171 /*
 172  * This function compares two net addresses by family and returns TRUE
 173  * if they are the same host.
 174  * If there is any doubt, return FALSE.
 175  * The AF_INET family is handled as a special case so that address mbufs
 176  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
 177  */
 178 static int
 179 netaddr_match(
 180         int family,
 181         union nethostaddr *haddr,
 182         mbuf_t nam)
 183 {
 184         struct sockaddr_in *inetaddr;
 185
 186         switch (family) {
 187         case AF_INET:
 188                 inetaddr = mbuf_data(nam);
 189                 if (inetaddr->sin_family == AF_INET &&
 190                     inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
 191                         return (1);
 192                 break;
 193         default:
 194                 break;
 195         };
 196         return (0);
 197 }
 198
 199 /*
 200  * Look for the request in the cache
 201  * If found then
 202  *    return action and optionally reply
 203  * else
 204  *    insert it in the cache
 205  *
 206  * The rules are as follows:
 207  * - if in progress, return DROP request
 208  * - if completed within DELAY of the current time, return DROP it
 209  * - if completed a longer time ago return REPLY if the reply was cached or
 210  *   return DOIT
 211  * Update/add new request at end of lru list
 212  */
 213 int
 214 nfsrv_getcache(
 215         struct nfsrv_descript *nd,
 216         struct nfsrv_sock *slp,
 217         mbuf_t *mrepp)
 218 {
 219         struct nfsrvcache *rp;
 220         struct nfsm_chain nmrep;
 221         struct sockaddr_in *saddr;
 222         int ret, error;
 223
 224         /*
 225          * Don't cache recent requests for reliable transport protocols.
 226          * (Maybe we should for the case of a reconnect, but..)
 227          */
 228         if (!nd->nd_nam2)
 229                 return (RC_DOIT);
 230         lck_mtx_lock(nfsrv_reqcache_mutex);
 231 loop:
 232         for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
 233             rp = rp->rc_hash.le_next) {
 234             if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
 235                 netaddr_match(AF_INET, &rp->rc_haddr, nd->nd_nam)) {
 236                         if ((rp->rc_flag & RC_LOCKED) != 0) {
 237                                 rp->rc_flag |= RC_WANTED;
 238                                 msleep(rp, nfsrv_reqcache_mutex, PZERO-1, "nfsrc", NULL);
 239                                 goto loop;
 240                         }
 241                         rp->rc_flag |= RC_LOCKED;
 242                         /* If not at end of LRU chain, move it there */
 243                         if (rp->rc_lru.tqe_next) {
 244                                 TAILQ_REMOVE(&nfsrv_reqcache_lruhead, rp, rc_lru);
 245                                 TAILQ_INSERT_TAIL(&nfsrv_reqcache_lruhead, rp, rc_lru);
 246                         }
 247                         if (rp->rc_state == RC_UNUSED)
 248                                 panic("nfsrv cache");
 249                         if (rp->rc_state == RC_INPROG) {
 250                                 OSAddAtomic(1, (SInt32*)&nfsstats.srvcache_inproghits);
 251                                 ret = RC_DROPIT;
 252                         } else if (rp->rc_flag & RC_REPSTATUS) {
 253                                 OSAddAtomic(1, (SInt32*)&nfsstats.srvcache_nonidemdonehits);
 254                                 nd->nd_repstat = rp->rc_status;
 255                                 error = nfsrv_rephead(nd, slp, &nmrep, 0);
 256                                 if (error) {
 257                                         printf("nfsrv cache: reply alloc failed for nonidem request hit\n");
 258                                         ret = RC_DROPIT;
 259                                         *mrepp = NULL;
 260                                 } else {
 261                                         ret = RC_REPLY;
 262                                         *mrepp = nmrep.nmc_mhead;
 263                                 }
 264                         } else if (rp->rc_flag & RC_REPMBUF) {
 265                                 OSAddAtomic(1, (SInt32*)&nfsstats.srvcache_nonidemdonehits);
 266                                 error = mbuf_copym(rp->rc_reply, 0, MBUF_COPYALL, MBUF_WAITOK, mrepp);
 267                                 if (error) {
 268                                         printf("nfsrv cache: reply copym failed for nonidem request hit\n");
 269                                         ret = RC_DROPIT;
 270                                 } else {
 271                                         ret = RC_REPLY;
 272                                 }
 273                         } else {
 274                                 OSAddAtomic(1, (SInt32*)&nfsstats.srvcache_idemdonehits);
 275                                 rp->rc_state = RC_INPROG;
 276                                 ret = RC_DOIT;
 277                         }
 278                         rp->rc_flag &= ~RC_LOCKED;
 279                         if (rp->rc_flag & RC_WANTED) {
 280                                 rp->rc_flag &= ~RC_WANTED;
 281                                 wakeup(rp);
 282                         }
 283                         lck_mtx_unlock(nfsrv_reqcache_mutex);
 284                         return (ret);
 285                 }
 286         }
 287         OSAddAtomic(1, (SInt32*)&nfsstats.srvcache_misses);
 288         if (nfsrv_reqcache_count < nfsrv_reqcache_size) {
 289                 /* try to allocate a new entry */
 290                 MALLOC(rp, struct nfsrvcache *, sizeof *rp, M_NFSD, M_WAITOK);
 291                 if (rp) {
 292                         bzero((char *)rp, sizeof *rp);
 293                         nfsrv_reqcache_count++;
 294                         rp->rc_flag = RC_LOCKED;
 295                 }
 296         } else {
 297                 rp = NULL;
 298         }
 299         if (!rp) {
 300                 /* try to reuse the least recently used entry */
 301                 rp = nfsrv_reqcache_lruhead.tqh_first;
 302                 if (!rp) {
 303                         /* no entry to reuse? */
 304                         /* OK, we just won't be able to cache this request */
 305                         lck_mtx_unlock(nfsrv_reqcache_mutex);
 306                         return (RC_DOIT);
 307                 }
 308                 while ((rp->rc_flag & RC_LOCKED) != 0) {
 309                         rp->rc_flag |= RC_WANTED;
 310                         msleep(rp, nfsrv_reqcache_mutex, PZERO-1, "nfsrc", NULL);
 311                         rp = nfsrv_reqcache_lruhead.tqh_first;
 312                 }
 313                 rp->rc_flag |= RC_LOCKED;
 314                 LIST_REMOVE(rp, rc_hash);
 315                 TAILQ_REMOVE(&nfsrv_reqcache_lruhead, rp, rc_lru);
 316                 if (rp->rc_flag & RC_REPMBUF)
 317                         mbuf_freem(rp->rc_reply);
 318                 if (rp->rc_flag & RC_NAM)
 319                         mbuf_freem(rp->rc_nam);
 320                 rp->rc_flag &= (RC_LOCKED | RC_WANTED);
 321         }
 322         TAILQ_INSERT_TAIL(&nfsrv_reqcache_lruhead, rp, rc_lru);
 323         rp->rc_state = RC_INPROG;
 324         rp->rc_xid = nd->nd_retxid;
 325         saddr = mbuf_data(nd->nd_nam);
 326         switch (saddr->sin_family) {
 327         case AF_INET:
 328                 rp->rc_flag |= RC_INETADDR;
 329                 rp->rc_inetaddr = saddr->sin_addr.s_addr;
 330                 break;
 331         default:
 332                 error = mbuf_copym(nd->nd_nam, 0, MBUF_COPYALL, MBUF_WAITOK, &rp->rc_nam);
 333                 if (error)
 334                         printf("nfsrv cache: nam copym failed\n");
 335                 else
 336                         rp->rc_flag |= RC_NAM;
 337                 break;
 338         };
 339         rp->rc_proc = nd->nd_procnum;
 340         LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
 341         rp->rc_flag &= ~RC_LOCKED;
 342         if (rp->rc_flag & RC_WANTED) {
 343                 rp->rc_flag &= ~RC_WANTED;
 344                 wakeup(rp);
 345         }
 346         lck_mtx_unlock(nfsrv_reqcache_mutex);
 347         return (RC_DOIT);
 348 }
 349
 350 /*
 351  * Update a request cache entry after the rpc has been done
 352  */
 353 void
 354 nfsrv_updatecache(
 355         struct nfsrv_descript *nd,
 356         int repvalid,
 357         mbuf_t repmbuf)
 358 {
 359         struct nfsrvcache *rp;
 360         int error;
 361
 362         if (!nd->nd_nam2)
 363                 return;
 364         lck_mtx_lock(nfsrv_reqcache_mutex);
 365 loop:
 366         for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
 367             rp = rp->rc_hash.le_next) {
 368             if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
 369                 netaddr_match(AF_INET, &rp->rc_haddr, nd->nd_nam)) {
 370                         if ((rp->rc_flag & RC_LOCKED) != 0) {
 371                                 rp->rc_flag |= RC_WANTED;
 372                                 msleep(rp, nfsrv_reqcache_mutex, PZERO-1, "nfsrc", NULL);
 373                                 goto loop;
 374                         }
 375                         rp->rc_flag |= RC_LOCKED;
 376                         if (rp->rc_state == RC_DONE) {
 377                                 /*
 378                                  * This can occur if the cache is too small.
 379                                  * Retransmits of the same request aren't
 380                                  * dropped so we may see the operation
 381                                  * complete more then once.
 382                                  */
 383                                 if (rp->rc_flag & RC_REPMBUF) {
 384                                         mbuf_freem(rp->rc_reply);
 385                                         rp->rc_flag &= ~RC_REPMBUF;
 386                                 }
 387                         }
 388                         rp->rc_state = RC_DONE;
 389                         /*
 390                          * If we have a valid reply update status and save
 391                          * the reply for non-idempotent rpc's.
 392                          */
 393                         if (repvalid && nonidempotent[nd->nd_procnum]) {
 394                                 if ((nd->nd_vers == NFS_VER2) &&
 395                                   nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
 396                                         rp->rc_status = nd->nd_repstat;
 397                                         rp->rc_flag |= RC_REPSTATUS;
 398                                 } else {
 399                                         error = mbuf_copym(repmbuf, 0, MBUF_COPYALL, MBUF_WAITOK, &rp->rc_reply);
 400                                         if (!error)
 401                                                 rp->rc_flag |= RC_REPMBUF;
 402                                 }
 403                         }
 404                         rp->rc_flag &= ~RC_LOCKED;
 405                         if (rp->rc_flag & RC_WANTED) {
 406                                 rp->rc_flag &= ~RC_WANTED;
 407                                 wakeup(rp);
 408                         }
 409                         lck_mtx_unlock(nfsrv_reqcache_mutex);
 410                         return;
 411                 }
 412         }
 413         lck_mtx_unlock(nfsrv_reqcache_mutex);
 414 }
 415
 416 /*
 417  * Clean out the cache. Called when the last nfsd terminates.
 418  */
 419 void
 420 nfsrv_cleancache(void)
 421 {
 422         struct nfsrvcache *rp, *nextrp;
 423
 424         lck_mtx_lock(nfsrv_reqcache_mutex);
 425         for (rp = nfsrv_reqcache_lruhead.tqh_first; rp != 0; rp = nextrp) {
 426                 nextrp = rp->rc_lru.tqe_next;
 427                 LIST_REMOVE(rp, rc_hash);
 428                 TAILQ_REMOVE(&nfsrv_reqcache_lruhead, rp, rc_lru);
 429                 _FREE(rp, M_NFSD);
 430         }
 431         nfsrv_reqcache_count = 0;
 432         FREE(nfsrv_reqcache_hashtbl, M_TEMP);
 433         lck_mtx_unlock(nfsrv_reqcache_mutex);
 434 }
 435
 436 #endif /* NFSSERVER */