nfs_nfsdcache.c revision 1.2.10.3
1 1.2.10.2 tls /* $NetBSD: nfs_nfsdcache.c,v 1.2.10.3 2017/12/03 11:38:42 jdolecek Exp $ */ 2 1.2.10.2 tls /*- 3 1.2.10.2 tls * Copyright (c) 1989, 1993 4 1.2.10.2 tls * The Regents of the University of California. All rights reserved. 5 1.2.10.2 tls * 6 1.2.10.2 tls * This code is derived from software contributed to Berkeley by 7 1.2.10.2 tls * Rick Macklem at The University of Guelph. 8 1.2.10.2 tls * 9 1.2.10.2 tls * Redistribution and use in source and binary forms, with or without 10 1.2.10.2 tls * modification, are permitted provided that the following conditions 11 1.2.10.2 tls * are met: 12 1.2.10.2 tls * 1. Redistributions of source code must retain the above copyright 13 1.2.10.2 tls * notice, this list of conditions and the following disclaimer. 14 1.2.10.2 tls * 2. Redistributions in binary form must reproduce the above copyright 15 1.2.10.2 tls * notice, this list of conditions and the following disclaimer in the 16 1.2.10.2 tls * documentation and/or other materials provided with the distribution. 17 1.2.10.2 tls * 4. Neither the name of the University nor the names of its contributors 18 1.2.10.2 tls * may be used to endorse or promote products derived from this software 19 1.2.10.2 tls * without specific prior written permission. 20 1.2.10.2 tls * 21 1.2.10.2 tls * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 1.2.10.2 tls * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 1.2.10.2 tls * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 1.2.10.2 tls * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 1.2.10.2 tls * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 1.2.10.2 tls * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 1.2.10.2 tls * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 1.2.10.2 tls * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 1.2.10.2 tls * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 1.2.10.2 tls * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 1.2.10.2 tls * SUCH DAMAGE. 32 1.2.10.2 tls * 33 1.2.10.2 tls */ 34 1.2.10.2 tls 35 1.2.10.2 tls #include <sys/cdefs.h> 36 1.2.10.3 jdolecek /* __FBSDID("FreeBSD: head/sys/fs/nfsserver/nfs_nfsdcache.c 304026 2016-08-12 22:44:59Z rmacklem "); */ 37 1.2.10.2 tls __RCSID("$NetBSD: nfs_nfsdcache.c,v 1.2.10.3 2017/12/03 11:38:42 jdolecek Exp $"); 38 1.2.10.2 tls 39 1.2.10.2 tls /* 40 1.2.10.2 tls * Here is the basic algorithm: 41 1.2.10.2 tls * First, some design criteria I used: 42 1.2.10.2 tls * - I think a false hit is more serious than a false miss 43 1.2.10.2 tls * - A false hit for an RPC that has Op(s) that order via seqid# must be 44 1.2.10.2 tls * avoided at all cost 45 1.2.10.2 tls * - A valid hit will probably happen a long time after the original reply 46 1.2.10.2 tls * and the TCP socket that the original request was received on will no 47 1.2.10.2 tls * longer be active 48 1.2.10.2 tls * (The long time delay implies to me that LRU is not appropriate.) 49 1.2.10.2 tls * - The mechanism will satisfy the requirements of ordering Ops with seqid#s 50 1.2.10.2 tls * in them as well as minimizing the risk of redoing retried non-idempotent 51 1.2.10.2 tls * Ops. 52 1.2.10.2 tls * Because it is biased towards avoiding false hits, multiple entries with 53 1.2.10.2 tls * the same xid are to be expected, especially for the case of the entry 54 1.2.10.2 tls * in the cache being related to a seqid# sequenced Op. 55 1.2.10.2 tls * 56 1.2.10.2 tls * The basic algorithm I'm about to code up: 57 1.2.10.2 tls * - Null RPCs bypass the cache and are just done 58 1.2.10.2 tls * For TCP 59 1.2.10.2 tls * - key on <xid, NFS version> (as noted above, there can be several 60 1.2.10.2 tls * entries with the same key) 61 1.2.10.2 tls * When a request arrives: 62 1.2.10.2 tls * For all that match key 63 1.2.10.2 tls * - if RPC# != OR request_size != 64 1.2.10.2 tls * - not a match with this one 65 1.2.10.2 tls * - if NFSv4 and received on same TCP socket OR 66 1.2.10.2 tls * received on a TCP connection created before the 67 1.2.10.2 tls * entry was cached 68 1.2.10.2 tls * - not a match with this one 69 1.2.10.2 tls * (V2,3 clients might retry on same TCP socket) 70 1.2.10.2 tls * - calculate checksum on first N bytes of NFS XDR 71 1.2.10.2 tls * - if checksum != 72 1.2.10.2 tls * - not a match for this one 73 1.2.10.2 tls * If any of the remaining ones that match has a 74 1.2.10.2 tls * seqid_refcnt > 0 75 1.2.10.2 tls * - not a match (go do RPC, using new cache entry) 76 1.2.10.2 tls * If one match left 77 1.2.10.2 tls * - a hit (reply from cache) 78 1.2.10.2 tls * else 79 1.2.10.2 tls * - miss (go do RPC, using new cache entry) 80 1.2.10.2 tls * 81 1.2.10.2 tls * During processing of NFSv4 request: 82 1.2.10.2 tls * - set a flag when a non-idempotent Op is processed 83 1.2.10.2 tls * - when an Op that uses a seqid# (Open,...) is processed 84 1.2.10.2 tls * - if same seqid# as referenced entry in cache 85 1.2.10.2 tls * - free new cache entry 86 1.2.10.2 tls * - reply from referenced cache entry 87 1.2.10.2 tls * else if next seqid# in order 88 1.2.10.2 tls * - free referenced cache entry 89 1.2.10.2 tls * - increment seqid_refcnt on new cache entry 90 1.2.10.2 tls * - set pointer from Openowner/Lockowner to 91 1.2.10.2 tls * new cache entry (aka reference it) 92 1.2.10.2 tls * else if first seqid# in sequence 93 1.2.10.2 tls * - increment seqid_refcnt on new cache entry 94 1.2.10.2 tls * - set pointer from Openowner/Lockowner to 95 1.2.10.2 tls * new cache entry (aka reference it) 96 1.2.10.2 tls * 97 1.2.10.2 tls * At end of RPC processing: 98 1.2.10.2 tls * - if seqid_refcnt > 0 OR flagged non-idempotent on new 99 1.2.10.2 tls * cache entry 100 1.2.10.2 tls * - save reply in cache entry 101 1.2.10.2 tls * - calculate checksum on first N bytes of NFS XDR 102 1.2.10.2 tls * request 103 1.2.10.2 tls * - note op and length of XDR request (in bytes) 104 1.2.10.2 tls * - timestamp it 105 1.2.10.2 tls * else 106 1.2.10.2 tls * - free new cache entry 107 1.2.10.2 tls * - Send reply (noting info for socket activity check, below) 108 1.2.10.2 tls * 109 1.2.10.2 tls * For cache entries saved above: 110 1.2.10.2 tls * - if saved since seqid_refcnt was > 0 111 1.2.10.2 tls * - free when seqid_refcnt decrements to 0 112 1.2.10.2 tls * (when next one in sequence is processed above, or 113 1.2.10.2 tls * when Openowner/Lockowner is discarded) 114 1.2.10.2 tls * else { non-idempotent Op(s) } 115 1.2.10.2 tls * - free when 116 1.2.10.2 tls * - some further activity observed on same 117 1.2.10.2 tls * socket 118 1.2.10.2 tls * (I'm not yet sure how I'm going to do 119 1.2.10.2 tls * this. Maybe look at the TCP connection 120 1.2.10.2 tls * to see if the send_tcp_sequence# is well 121 1.2.10.2 tls * past sent reply OR K additional RPCs 122 1.2.10.2 tls * replied on same socket OR?) 123 1.2.10.2 tls * OR 124 1.2.10.2 tls * - when very old (hours, days, weeks?) 125 1.2.10.2 tls * 126 1.2.10.2 tls * For UDP (v2, 3 only), pretty much the old way: 127 1.2.10.2 tls * - key on <xid, NFS version, RPC#, Client host ip#> 128 1.2.10.2 tls * (at most one entry for each key) 129 1.2.10.2 tls * 130 1.2.10.2 tls * When a Request arrives: 131 1.2.10.2 tls * - if a match with entry via key 132 1.2.10.2 tls * - if RPC marked In_progress 133 1.2.10.2 tls * - discard request (don't send reply) 134 1.2.10.2 tls * else 135 1.2.10.2 tls * - reply from cache 136 1.2.10.2 tls * - timestamp cache entry 137 1.2.10.2 tls * else 138 1.2.10.2 tls * - add entry to cache, marked In_progress 139 1.2.10.2 tls * - do RPC 140 1.2.10.2 tls * - when RPC done 141 1.2.10.2 tls * - if RPC# non-idempotent 142 1.2.10.2 tls * - mark entry Done (not In_progress) 143 1.2.10.2 tls * - save reply 144 1.2.10.2 tls * - timestamp cache entry 145 1.2.10.2 tls * else 146 1.2.10.2 tls * - free cache entry 147 1.2.10.2 tls * - send reply 148 1.2.10.2 tls * 149 1.2.10.2 tls * Later, entries with saved replies are free'd a short time (few minutes) 150 1.2.10.2 tls * after reply sent (timestamp). 151 1.2.10.2 tls * Reference: Chet Juszczak, "Improving the Performance and Correctness 152 1.2.10.2 tls * of an NFS Server", in Proc. Winter 1989 USENIX Conference, 153 1.2.10.2 tls * pages 53-63. San Diego, February 1989. 154 1.2.10.2 tls * for the UDP case. 155 1.2.10.2 tls * nfsrc_floodlevel is set to the allowable upper limit for saved replies 156 1.2.10.2 tls * for TCP. For V3, a reply won't be saved when the flood level is 157 1.2.10.2 tls * hit. For V4, the non-idempotent Op will return NFSERR_RESOURCE in 158 1.2.10.2 tls * that case. This level should be set high enough that this almost 159 1.2.10.2 tls * never happens. 160 1.2.10.2 tls */ 161 1.2.10.2 tls #ifndef APPLEKEXT 162 1.2.10.3 jdolecek #include <fs/nfs/common/nfsport.h> 163 1.2.10.2 tls 164 1.2.10.3 jdolecek extern struct nfsstatsv1 nfsstatsv1; 165 1.2.10.2 tls extern struct mtx nfsrc_udpmtx; 166 1.2.10.2 tls extern struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; 167 1.2.10.3 jdolecek extern struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; 168 1.2.10.2 tls int nfsrc_floodlevel = NFSRVCACHE_FLOODLEVEL, nfsrc_tcpsavedreplies = 0; 169 1.2.10.2 tls #endif /* !APPLEKEXT */ 170 1.2.10.2 tls 171 1.2.10.2 tls SYSCTL_DECL(_vfs_nfsd); 172 1.2.10.2 tls 173 1.2.10.2 tls static u_int nfsrc_tcphighwater = 0; 174 1.2.10.2 tls static int 175 1.2.10.2 tls sysctl_tcphighwater(SYSCTL_HANDLER_ARGS) 176 1.2.10.2 tls { 177 1.2.10.2 tls int error, newhighwater; 178 1.2.10.2 tls 179 1.2.10.2 tls newhighwater = nfsrc_tcphighwater; 180 1.2.10.2 tls error = sysctl_handle_int(oidp, &newhighwater, 0, req); 181 1.2.10.2 tls if (error != 0 || req->newptr == NULL) 182 1.2.10.2 tls return (error); 183 1.2.10.2 tls if (newhighwater < 0) 184 1.2.10.2 tls return (EINVAL); 185 1.2.10.2 tls if (newhighwater >= nfsrc_floodlevel) 186 1.2.10.2 tls nfsrc_floodlevel = newhighwater + newhighwater / 5; 187 1.2.10.2 tls nfsrc_tcphighwater = newhighwater; 188 1.2.10.2 tls return (0); 189 1.2.10.2 tls } 190 1.2.10.2 tls SYSCTL_PROC(_vfs_nfsd, OID_AUTO, tcphighwater, CTLTYPE_UINT | CTLFLAG_RW, 0, 191 1.2.10.2 tls sizeof(nfsrc_tcphighwater), sysctl_tcphighwater, "IU", 192 1.2.10.2 tls "High water mark for TCP cache entries"); 193 1.2.10.2 tls 194 1.2.10.2 tls static u_int nfsrc_udphighwater = NFSRVCACHE_UDPHIGHWATER; 195 1.2.10.2 tls SYSCTL_UINT(_vfs_nfsd, OID_AUTO, udphighwater, CTLFLAG_RW, 196 1.2.10.2 tls &nfsrc_udphighwater, 0, 197 1.2.10.2 tls "High water mark for UDP cache entries"); 198 1.2.10.2 tls static u_int nfsrc_tcptimeout = NFSRVCACHE_TCPTIMEOUT; 199 1.2.10.2 tls SYSCTL_UINT(_vfs_nfsd, OID_AUTO, tcpcachetimeo, CTLFLAG_RW, 200 1.2.10.2 tls &nfsrc_tcptimeout, 0, 201 1.2.10.2 tls "Timeout for TCP entries in the DRC"); 202 1.2.10.2 tls static u_int nfsrc_tcpnonidempotent = 1; 203 1.2.10.2 tls SYSCTL_UINT(_vfs_nfsd, OID_AUTO, cachetcp, CTLFLAG_RW, 204 1.2.10.2 tls &nfsrc_tcpnonidempotent, 0, 205 1.2.10.2 tls "Enable the DRC for NFS over TCP"); 206 1.2.10.2 tls 207 1.2.10.2 tls static int nfsrc_udpcachesize = 0; 208 1.2.10.2 tls static TAILQ_HEAD(, nfsrvcache) nfsrvudplru; 209 1.2.10.2 tls static struct nfsrvhashhead nfsrvudphashtbl[NFSRVCACHE_HASHSIZE]; 210 1.2.10.2 tls 211 1.2.10.2 tls /* 212 1.2.10.2 tls * and the reverse mapping from generic to Version 2 procedure numbers 213 1.2.10.2 tls */ 214 1.2.10.2 tls static int newnfsv2_procid[NFS_V3NPROCS] = { 215 1.2.10.2 tls NFSV2PROC_NULL, 216 1.2.10.2 tls NFSV2PROC_GETATTR, 217 1.2.10.2 tls NFSV2PROC_SETATTR, 218 1.2.10.2 tls NFSV2PROC_LOOKUP, 219 1.2.10.2 tls NFSV2PROC_NOOP, 220 1.2.10.2 tls NFSV2PROC_READLINK, 221 1.2.10.2 tls NFSV2PROC_READ, 222 1.2.10.2 tls NFSV2PROC_WRITE, 223 1.2.10.2 tls NFSV2PROC_CREATE, 224 1.2.10.2 tls NFSV2PROC_MKDIR, 225 1.2.10.2 tls NFSV2PROC_SYMLINK, 226 1.2.10.2 tls NFSV2PROC_CREATE, 227 1.2.10.2 tls NFSV2PROC_REMOVE, 228 1.2.10.2 tls NFSV2PROC_RMDIR, 229 1.2.10.2 tls NFSV2PROC_RENAME, 230 1.2.10.2 tls NFSV2PROC_LINK, 231 1.2.10.2 tls NFSV2PROC_READDIR, 232 1.2.10.2 tls NFSV2PROC_NOOP, 233 1.2.10.2 tls NFSV2PROC_STATFS, 234 1.2.10.2 tls NFSV2PROC_NOOP, 235 1.2.10.2 tls NFSV2PROC_NOOP, 236 1.2.10.2 tls NFSV2PROC_NOOP, 237 1.2.10.2 tls }; 238 1.2.10.2 tls 239 1.2.10.2 tls #define nfsrc_hash(xid) (((xid) + ((xid) >> 24)) % NFSRVCACHE_HASHSIZE) 240 1.2.10.2 tls #define NFSRCUDPHASH(xid) \ 241 1.2.10.2 tls (&nfsrvudphashtbl[nfsrc_hash(xid)]) 242 1.2.10.2 tls #define NFSRCHASH(xid) \ 243 1.2.10.2 tls (&nfsrchash_table[nfsrc_hash(xid)].tbl) 244 1.2.10.3 jdolecek #define NFSRCAHASH(xid) (&nfsrcahash_table[nfsrc_hash(xid)]) 245 1.2.10.2 tls #define TRUE 1 246 1.2.10.2 tls #define FALSE 0 247 1.2.10.2 tls #define NFSRVCACHE_CHECKLEN 100 248 1.2.10.2 tls 249 1.2.10.2 tls /* True iff the rpc reply is an nfs status ONLY! */ 250 1.2.10.2 tls static int nfsv2_repstat[NFS_V3NPROCS] = { 251 1.2.10.2 tls FALSE, 252 1.2.10.2 tls FALSE, 253 1.2.10.2 tls FALSE, 254 1.2.10.2 tls FALSE, 255 1.2.10.2 tls FALSE, 256 1.2.10.2 tls FALSE, 257 1.2.10.2 tls FALSE, 258 1.2.10.2 tls FALSE, 259 1.2.10.2 tls FALSE, 260 1.2.10.2 tls FALSE, 261 1.2.10.2 tls TRUE, 262 1.2.10.2 tls TRUE, 263 1.2.10.2 tls TRUE, 264 1.2.10.2 tls TRUE, 265 1.2.10.2 tls FALSE, 266 1.2.10.2 tls TRUE, 267 1.2.10.2 tls FALSE, 268 1.2.10.2 tls FALSE, 269 1.2.10.2 tls FALSE, 270 1.2.10.2 tls FALSE, 271 1.2.10.2 tls FALSE, 272 1.2.10.2 tls FALSE, 273 1.2.10.2 tls }; 274 1.2.10.2 tls 275 1.2.10.2 tls /* 276 1.2.10.2 tls * Will NFS want to work over IPv6 someday? 277 1.2.10.2 tls */ 278 1.2.10.2 tls #define NETFAMILY(rp) \ 279 1.2.10.2 tls (((rp)->rc_flag & RC_INETIPV6) ? AF_INET6 : AF_INET) 280 1.2.10.2 tls 281 1.2.10.2 tls /* local functions */ 282 1.2.10.2 tls static int nfsrc_getudp(struct nfsrv_descript *nd, struct nfsrvcache *newrp); 283 1.2.10.2 tls static int nfsrc_gettcp(struct nfsrv_descript *nd, struct nfsrvcache *newrp); 284 1.2.10.2 tls static void nfsrc_lock(struct nfsrvcache *rp); 285 1.2.10.2 tls static void nfsrc_unlock(struct nfsrvcache *rp); 286 1.2.10.2 tls static void nfsrc_wanted(struct nfsrvcache *rp); 287 1.2.10.2 tls static void nfsrc_freecache(struct nfsrvcache *rp); 288 1.2.10.2 tls static int nfsrc_getlenandcksum(mbuf_t m1, u_int16_t *cksum); 289 1.2.10.2 tls static void nfsrc_marksametcpconn(u_int64_t); 290 1.2.10.2 tls 291 1.2.10.2 tls /* 292 1.2.10.2 tls * Return the correct mutex for this cache entry. 293 1.2.10.2 tls */ 294 1.2.10.2 tls static __inline struct mtx * 295 1.2.10.2 tls nfsrc_cachemutex(struct nfsrvcache *rp) 296 1.2.10.2 tls { 297 1.2.10.2 tls 298 1.2.10.2 tls if ((rp->rc_flag & RC_UDP) != 0) 299 1.2.10.2 tls return (&nfsrc_udpmtx); 300 1.2.10.2 tls return (&nfsrchash_table[nfsrc_hash(rp->rc_xid)].mtx); 301 1.2.10.2 tls } 302 1.2.10.2 tls 303 1.2.10.2 tls /* 304 1.2.10.2 tls * Initialize the server request cache list 305 1.2.10.2 tls */ 306 1.2.10.2 tls APPLESTATIC void 307 1.2.10.2 tls nfsrvd_initcache(void) 308 1.2.10.2 tls { 309 1.2.10.2 tls int i; 310 1.2.10.2 tls static int inited = 0; 311 1.2.10.2 tls 312 1.2.10.2 tls if (inited) 313 1.2.10.2 tls return; 314 1.2.10.2 tls inited = 1; 315 1.2.10.2 tls for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 316 1.2.10.2 tls LIST_INIT(&nfsrvudphashtbl[i]); 317 1.2.10.2 tls LIST_INIT(&nfsrchash_table[i].tbl); 318 1.2.10.3 jdolecek LIST_INIT(&nfsrcahash_table[i].tbl); 319 1.2.10.2 tls } 320 1.2.10.2 tls TAILQ_INIT(&nfsrvudplru); 321 1.2.10.2 tls nfsrc_tcpsavedreplies = 0; 322 1.2.10.2 tls nfsrc_udpcachesize = 0; 323 1.2.10.3 jdolecek nfsstatsv1.srvcache_tcppeak = 0; 324 1.2.10.3 jdolecek nfsstatsv1.srvcache_size = 0; 325 1.2.10.2 tls } 326 1.2.10.2 tls 327 1.2.10.2 tls /* 328 1.2.10.2 tls * Get a cache entry for this request. Basically just malloc a new one 329 1.2.10.2 tls * and then call nfsrc_getudp() or nfsrc_gettcp() to do the rest. 330 1.2.10.2 tls */ 331 1.2.10.2 tls APPLESTATIC int 332 1.2.10.3 jdolecek nfsrvd_getcache(struct nfsrv_descript *nd) 333 1.2.10.2 tls { 334 1.2.10.2 tls struct nfsrvcache *newrp; 335 1.2.10.2 tls int ret; 336 1.2.10.2 tls 337 1.2.10.2 tls if (nd->nd_procnum == NFSPROC_NULL) 338 1.2.10.2 tls panic("nfsd cache null"); 339 1.2.10.2 tls MALLOC(newrp, struct nfsrvcache *, sizeof (struct nfsrvcache), 340 1.2.10.2 tls M_NFSRVCACHE, M_WAITOK); 341 1.2.10.2 tls NFSBZERO((caddr_t)newrp, sizeof (struct nfsrvcache)); 342 1.2.10.2 tls if (nd->nd_flag & ND_NFSV4) 343 1.2.10.2 tls newrp->rc_flag = RC_NFSV4; 344 1.2.10.2 tls else if (nd->nd_flag & ND_NFSV3) 345 1.2.10.2 tls newrp->rc_flag = RC_NFSV3; 346 1.2.10.2 tls else 347 1.2.10.2 tls newrp->rc_flag = RC_NFSV2; 348 1.2.10.2 tls newrp->rc_xid = nd->nd_retxid; 349 1.2.10.2 tls newrp->rc_proc = nd->nd_procnum; 350 1.2.10.2 tls newrp->rc_sockref = nd->nd_sockref; 351 1.2.10.2 tls newrp->rc_cachetime = nd->nd_tcpconntime; 352 1.2.10.2 tls if (nd->nd_flag & ND_SAMETCPCONN) 353 1.2.10.2 tls newrp->rc_flag |= RC_SAMETCPCONN; 354 1.2.10.2 tls if (nd->nd_nam2 != NULL) { 355 1.2.10.2 tls newrp->rc_flag |= RC_UDP; 356 1.2.10.2 tls ret = nfsrc_getudp(nd, newrp); 357 1.2.10.2 tls } else { 358 1.2.10.2 tls ret = nfsrc_gettcp(nd, newrp); 359 1.2.10.2 tls } 360 1.2.10.2 tls NFSEXITCODE2(0, nd); 361 1.2.10.2 tls return (ret); 362 1.2.10.2 tls } 363 1.2.10.2 tls 364 1.2.10.2 tls /* 365 1.2.10.2 tls * For UDP (v2, v3): 366 1.2.10.2 tls * - key on <xid, NFS version, RPC#, Client host ip#> 367 1.2.10.2 tls * (at most one entry for each key) 368 1.2.10.2 tls */ 369 1.2.10.2 tls static int 370 1.2.10.2 tls nfsrc_getudp(struct nfsrv_descript *nd, struct nfsrvcache *newrp) 371 1.2.10.2 tls { 372 1.2.10.2 tls struct nfsrvcache *rp; 373 1.2.10.2 tls struct sockaddr_in *saddr; 374 1.2.10.2 tls struct sockaddr_in6 *saddr6; 375 1.2.10.2 tls struct nfsrvhashhead *hp; 376 1.2.10.2 tls int ret = 0; 377 1.2.10.2 tls struct mtx *mutex; 378 1.2.10.2 tls 379 1.2.10.2 tls mutex = nfsrc_cachemutex(newrp); 380 1.2.10.2 tls hp = NFSRCUDPHASH(newrp->rc_xid); 381 1.2.10.2 tls loop: 382 1.2.10.2 tls mtx_lock(mutex); 383 1.2.10.2 tls LIST_FOREACH(rp, hp, rc_hash) { 384 1.2.10.2 tls if (newrp->rc_xid == rp->rc_xid && 385 1.2.10.2 tls newrp->rc_proc == rp->rc_proc && 386 1.2.10.2 tls (newrp->rc_flag & rp->rc_flag & RC_NFSVERS) && 387 1.2.10.2 tls nfsaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) { 388 1.2.10.2 tls if ((rp->rc_flag & RC_LOCKED) != 0) { 389 1.2.10.2 tls rp->rc_flag |= RC_WANTED; 390 1.2.10.2 tls (void)mtx_sleep(rp, mutex, (PZERO - 1) | PDROP, 391 1.2.10.2 tls "nfsrc", 10 * hz); 392 1.2.10.2 tls goto loop; 393 1.2.10.2 tls } 394 1.2.10.2 tls if (rp->rc_flag == 0) 395 1.2.10.2 tls panic("nfs udp cache0"); 396 1.2.10.2 tls rp->rc_flag |= RC_LOCKED; 397 1.2.10.2 tls TAILQ_REMOVE(&nfsrvudplru, rp, rc_lru); 398 1.2.10.2 tls TAILQ_INSERT_TAIL(&nfsrvudplru, rp, rc_lru); 399 1.2.10.2 tls if (rp->rc_flag & RC_INPROG) { 400 1.2.10.3 jdolecek nfsstatsv1.srvcache_inproghits++; 401 1.2.10.2 tls mtx_unlock(mutex); 402 1.2.10.2 tls ret = RC_DROPIT; 403 1.2.10.2 tls } else if (rp->rc_flag & RC_REPSTATUS) { 404 1.2.10.2 tls /* 405 1.2.10.2 tls * V2 only. 406 1.2.10.2 tls */ 407 1.2.10.3 jdolecek nfsstatsv1.srvcache_nonidemdonehits++; 408 1.2.10.2 tls mtx_unlock(mutex); 409 1.2.10.2 tls nfsrvd_rephead(nd); 410 1.2.10.2 tls *(nd->nd_errp) = rp->rc_status; 411 1.2.10.2 tls ret = RC_REPLY; 412 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + 413 1.2.10.2 tls NFSRVCACHE_UDPTIMEOUT; 414 1.2.10.2 tls } else if (rp->rc_flag & RC_REPMBUF) { 415 1.2.10.3 jdolecek nfsstatsv1.srvcache_nonidemdonehits++; 416 1.2.10.2 tls mtx_unlock(mutex); 417 1.2.10.2 tls nd->nd_mreq = m_copym(rp->rc_reply, 0, 418 1.2.10.2 tls M_COPYALL, M_WAITOK); 419 1.2.10.2 tls ret = RC_REPLY; 420 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + 421 1.2.10.2 tls NFSRVCACHE_UDPTIMEOUT; 422 1.2.10.2 tls } else { 423 1.2.10.2 tls panic("nfs udp cache1"); 424 1.2.10.2 tls } 425 1.2.10.2 tls nfsrc_unlock(rp); 426 1.2.10.2 tls free((caddr_t)newrp, M_NFSRVCACHE); 427 1.2.10.2 tls goto out; 428 1.2.10.2 tls } 429 1.2.10.2 tls } 430 1.2.10.3 jdolecek nfsstatsv1.srvcache_misses++; 431 1.2.10.3 jdolecek atomic_add_int(&nfsstatsv1.srvcache_size, 1); 432 1.2.10.2 tls nfsrc_udpcachesize++; 433 1.2.10.2 tls 434 1.2.10.2 tls newrp->rc_flag |= RC_INPROG; 435 1.2.10.2 tls saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 436 1.2.10.2 tls if (saddr->sin_family == AF_INET) 437 1.2.10.2 tls newrp->rc_inet = saddr->sin_addr.s_addr; 438 1.2.10.2 tls else if (saddr->sin_family == AF_INET6) { 439 1.2.10.2 tls saddr6 = (struct sockaddr_in6 *)saddr; 440 1.2.10.2 tls NFSBCOPY((caddr_t)&saddr6->sin6_addr, (caddr_t)&newrp->rc_inet6, 441 1.2.10.2 tls sizeof (struct in6_addr)); 442 1.2.10.2 tls newrp->rc_flag |= RC_INETIPV6; 443 1.2.10.2 tls } 444 1.2.10.2 tls LIST_INSERT_HEAD(hp, newrp, rc_hash); 445 1.2.10.2 tls TAILQ_INSERT_TAIL(&nfsrvudplru, newrp, rc_lru); 446 1.2.10.2 tls mtx_unlock(mutex); 447 1.2.10.2 tls nd->nd_rp = newrp; 448 1.2.10.2 tls ret = RC_DOIT; 449 1.2.10.2 tls 450 1.2.10.2 tls out: 451 1.2.10.2 tls NFSEXITCODE2(0, nd); 452 1.2.10.2 tls return (ret); 453 1.2.10.2 tls } 454 1.2.10.2 tls 455 1.2.10.2 tls /* 456 1.2.10.2 tls * Update a request cache entry after the rpc has been done 457 1.2.10.2 tls */ 458 1.2.10.2 tls APPLESTATIC struct nfsrvcache * 459 1.2.10.3 jdolecek nfsrvd_updatecache(struct nfsrv_descript *nd) 460 1.2.10.2 tls { 461 1.2.10.2 tls struct nfsrvcache *rp; 462 1.2.10.2 tls struct nfsrvcache *retrp = NULL; 463 1.2.10.2 tls mbuf_t m; 464 1.2.10.2 tls struct mtx *mutex; 465 1.2.10.2 tls 466 1.2.10.2 tls rp = nd->nd_rp; 467 1.2.10.2 tls if (!rp) 468 1.2.10.2 tls panic("nfsrvd_updatecache null rp"); 469 1.2.10.2 tls nd->nd_rp = NULL; 470 1.2.10.2 tls mutex = nfsrc_cachemutex(rp); 471 1.2.10.2 tls mtx_lock(mutex); 472 1.2.10.2 tls nfsrc_lock(rp); 473 1.2.10.2 tls if (!(rp->rc_flag & RC_INPROG)) 474 1.2.10.2 tls panic("nfsrvd_updatecache not inprog"); 475 1.2.10.2 tls rp->rc_flag &= ~RC_INPROG; 476 1.2.10.2 tls if (rp->rc_flag & RC_UDP) { 477 1.2.10.2 tls TAILQ_REMOVE(&nfsrvudplru, rp, rc_lru); 478 1.2.10.2 tls TAILQ_INSERT_TAIL(&nfsrvudplru, rp, rc_lru); 479 1.2.10.2 tls } 480 1.2.10.2 tls 481 1.2.10.2 tls /* 482 1.2.10.2 tls * Reply from cache is a special case returned by nfsrv_checkseqid(). 483 1.2.10.2 tls */ 484 1.2.10.2 tls if (nd->nd_repstat == NFSERR_REPLYFROMCACHE) { 485 1.2.10.3 jdolecek nfsstatsv1.srvcache_nonidemdonehits++; 486 1.2.10.2 tls mtx_unlock(mutex); 487 1.2.10.2 tls nd->nd_repstat = 0; 488 1.2.10.2 tls if (nd->nd_mreq) 489 1.2.10.2 tls mbuf_freem(nd->nd_mreq); 490 1.2.10.2 tls if (!(rp->rc_flag & RC_REPMBUF)) 491 1.2.10.2 tls panic("reply from cache"); 492 1.2.10.2 tls nd->nd_mreq = m_copym(rp->rc_reply, 0, 493 1.2.10.2 tls M_COPYALL, M_WAITOK); 494 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; 495 1.2.10.2 tls nfsrc_unlock(rp); 496 1.2.10.2 tls goto out; 497 1.2.10.2 tls } 498 1.2.10.2 tls 499 1.2.10.2 tls /* 500 1.2.10.2 tls * If rc_refcnt > 0, save it 501 1.2.10.2 tls * For UDP, save it if ND_SAVEREPLY is set 502 1.2.10.2 tls * For TCP, save it if ND_SAVEREPLY and nfsrc_tcpnonidempotent is set 503 1.2.10.2 tls */ 504 1.2.10.2 tls if (nd->nd_repstat != NFSERR_DONTREPLY && 505 1.2.10.2 tls (rp->rc_refcnt > 0 || 506 1.2.10.2 tls ((nd->nd_flag & ND_SAVEREPLY) && (rp->rc_flag & RC_UDP)) || 507 1.2.10.2 tls ((nd->nd_flag & ND_SAVEREPLY) && !(rp->rc_flag & RC_UDP) && 508 1.2.10.2 tls nfsrc_tcpsavedreplies <= nfsrc_floodlevel && 509 1.2.10.2 tls nfsrc_tcpnonidempotent))) { 510 1.2.10.2 tls if (rp->rc_refcnt > 0) { 511 1.2.10.2 tls if (!(rp->rc_flag & RC_NFSV4)) 512 1.2.10.2 tls panic("update_cache refcnt"); 513 1.2.10.2 tls rp->rc_flag |= RC_REFCNT; 514 1.2.10.2 tls } 515 1.2.10.2 tls if ((nd->nd_flag & ND_NFSV2) && 516 1.2.10.2 tls nfsv2_repstat[newnfsv2_procid[nd->nd_procnum]]) { 517 1.2.10.2 tls rp->rc_status = nd->nd_repstat; 518 1.2.10.2 tls rp->rc_flag |= RC_REPSTATUS; 519 1.2.10.2 tls mtx_unlock(mutex); 520 1.2.10.2 tls } else { 521 1.2.10.2 tls if (!(rp->rc_flag & RC_UDP)) { 522 1.2.10.2 tls atomic_add_int(&nfsrc_tcpsavedreplies, 1); 523 1.2.10.2 tls if (nfsrc_tcpsavedreplies > 524 1.2.10.3 jdolecek nfsstatsv1.srvcache_tcppeak) 525 1.2.10.3 jdolecek nfsstatsv1.srvcache_tcppeak = 526 1.2.10.2 tls nfsrc_tcpsavedreplies; 527 1.2.10.2 tls } 528 1.2.10.2 tls mtx_unlock(mutex); 529 1.2.10.2 tls m = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK); 530 1.2.10.2 tls mtx_lock(mutex); 531 1.2.10.2 tls rp->rc_reply = m; 532 1.2.10.2 tls rp->rc_flag |= RC_REPMBUF; 533 1.2.10.2 tls mtx_unlock(mutex); 534 1.2.10.2 tls } 535 1.2.10.2 tls if (rp->rc_flag & RC_UDP) { 536 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + 537 1.2.10.2 tls NFSRVCACHE_UDPTIMEOUT; 538 1.2.10.2 tls nfsrc_unlock(rp); 539 1.2.10.2 tls } else { 540 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; 541 1.2.10.2 tls if (rp->rc_refcnt > 0) 542 1.2.10.2 tls nfsrc_unlock(rp); 543 1.2.10.2 tls else 544 1.2.10.2 tls retrp = rp; 545 1.2.10.2 tls } 546 1.2.10.2 tls } else { 547 1.2.10.2 tls nfsrc_freecache(rp); 548 1.2.10.2 tls mtx_unlock(mutex); 549 1.2.10.2 tls } 550 1.2.10.2 tls 551 1.2.10.2 tls out: 552 1.2.10.2 tls NFSEXITCODE2(0, nd); 553 1.2.10.2 tls return (retrp); 554 1.2.10.2 tls } 555 1.2.10.2 tls 556 1.2.10.2 tls /* 557 1.2.10.2 tls * Invalidate and, if possible, free an in prog cache entry. 558 1.2.10.2 tls * Must not sleep. 559 1.2.10.2 tls */ 560 1.2.10.2 tls APPLESTATIC void 561 1.2.10.2 tls nfsrvd_delcache(struct nfsrvcache *rp) 562 1.2.10.2 tls { 563 1.2.10.2 tls struct mtx *mutex; 564 1.2.10.2 tls 565 1.2.10.2 tls mutex = nfsrc_cachemutex(rp); 566 1.2.10.2 tls if (!(rp->rc_flag & RC_INPROG)) 567 1.2.10.2 tls panic("nfsrvd_delcache not in prog"); 568 1.2.10.2 tls mtx_lock(mutex); 569 1.2.10.2 tls rp->rc_flag &= ~RC_INPROG; 570 1.2.10.2 tls if (rp->rc_refcnt == 0 && !(rp->rc_flag & RC_LOCKED)) 571 1.2.10.2 tls nfsrc_freecache(rp); 572 1.2.10.2 tls mtx_unlock(mutex); 573 1.2.10.2 tls } 574 1.2.10.2 tls 575 1.2.10.2 tls /* 576 1.2.10.2 tls * Called after nfsrvd_updatecache() once the reply is sent, to update 577 1.2.10.3 jdolecek * the entry's sequence number and unlock it. The argument is 578 1.2.10.2 tls * the pointer returned by nfsrvd_updatecache(). 579 1.2.10.2 tls */ 580 1.2.10.2 tls APPLESTATIC void 581 1.2.10.3 jdolecek nfsrvd_sentcache(struct nfsrvcache *rp, int have_seq, uint32_t seq) 582 1.2.10.2 tls { 583 1.2.10.3 jdolecek struct nfsrchash_bucket *hbp; 584 1.2.10.2 tls 585 1.2.10.3 jdolecek KASSERT(rp->rc_flag & RC_LOCKED, ("nfsrvd_sentcache not locked")); 586 1.2.10.3 jdolecek if (have_seq) { 587 1.2.10.3 jdolecek hbp = NFSRCAHASH(rp->rc_sockref); 588 1.2.10.3 jdolecek mtx_lock(&hbp->mtx); 589 1.2.10.3 jdolecek rp->rc_tcpseq = seq; 590 1.2.10.3 jdolecek if (rp->rc_acked != RC_NO_ACK) 591 1.2.10.3 jdolecek LIST_INSERT_HEAD(&hbp->tbl, rp, rc_ahash); 592 1.2.10.3 jdolecek rp->rc_acked = RC_NO_ACK; 593 1.2.10.3 jdolecek mtx_unlock(&hbp->mtx); 594 1.2.10.2 tls } 595 1.2.10.2 tls nfsrc_unlock(rp); 596 1.2.10.2 tls } 597 1.2.10.2 tls 598 1.2.10.2 tls /* 599 1.2.10.2 tls * Get a cache entry for TCP 600 1.2.10.2 tls * - key on <xid, nfs version> 601 1.2.10.2 tls * (allow multiple entries for a given key) 602 1.2.10.2 tls */ 603 1.2.10.2 tls static int 604 1.2.10.2 tls nfsrc_gettcp(struct nfsrv_descript *nd, struct nfsrvcache *newrp) 605 1.2.10.2 tls { 606 1.2.10.2 tls struct nfsrvcache *rp, *nextrp; 607 1.2.10.2 tls int i; 608 1.2.10.2 tls struct nfsrvcache *hitrp; 609 1.2.10.2 tls struct nfsrvhashhead *hp, nfsrc_templist; 610 1.2.10.2 tls int hit, ret = 0; 611 1.2.10.2 tls struct mtx *mutex; 612 1.2.10.2 tls 613 1.2.10.2 tls mutex = nfsrc_cachemutex(newrp); 614 1.2.10.2 tls hp = NFSRCHASH(newrp->rc_xid); 615 1.2.10.2 tls newrp->rc_reqlen = nfsrc_getlenandcksum(nd->nd_mrep, &newrp->rc_cksum); 616 1.2.10.2 tls tryagain: 617 1.2.10.2 tls mtx_lock(mutex); 618 1.2.10.2 tls hit = 1; 619 1.2.10.2 tls LIST_INIT(&nfsrc_templist); 620 1.2.10.2 tls /* 621 1.2.10.2 tls * Get all the matches and put them on the temp list. 622 1.2.10.2 tls */ 623 1.2.10.2 tls rp = LIST_FIRST(hp); 624 1.2.10.2 tls while (rp != NULL) { 625 1.2.10.2 tls nextrp = LIST_NEXT(rp, rc_hash); 626 1.2.10.2 tls if (newrp->rc_xid == rp->rc_xid && 627 1.2.10.2 tls (!(rp->rc_flag & RC_INPROG) || 628 1.2.10.2 tls ((newrp->rc_flag & RC_SAMETCPCONN) && 629 1.2.10.2 tls newrp->rc_sockref == rp->rc_sockref)) && 630 1.2.10.2 tls (newrp->rc_flag & rp->rc_flag & RC_NFSVERS) && 631 1.2.10.2 tls newrp->rc_proc == rp->rc_proc && 632 1.2.10.2 tls ((newrp->rc_flag & RC_NFSV4) && 633 1.2.10.2 tls newrp->rc_sockref != rp->rc_sockref && 634 1.2.10.2 tls newrp->rc_cachetime >= rp->rc_cachetime) 635 1.2.10.2 tls && newrp->rc_reqlen == rp->rc_reqlen && 636 1.2.10.2 tls newrp->rc_cksum == rp->rc_cksum) { 637 1.2.10.2 tls LIST_REMOVE(rp, rc_hash); 638 1.2.10.2 tls LIST_INSERT_HEAD(&nfsrc_templist, rp, rc_hash); 639 1.2.10.2 tls } 640 1.2.10.2 tls rp = nextrp; 641 1.2.10.2 tls } 642 1.2.10.2 tls 643 1.2.10.2 tls /* 644 1.2.10.2 tls * Now, use nfsrc_templist to decide if there is a match. 645 1.2.10.2 tls */ 646 1.2.10.2 tls i = 0; 647 1.2.10.2 tls LIST_FOREACH(rp, &nfsrc_templist, rc_hash) { 648 1.2.10.2 tls i++; 649 1.2.10.2 tls if (rp->rc_refcnt > 0) { 650 1.2.10.2 tls hit = 0; 651 1.2.10.2 tls break; 652 1.2.10.2 tls } 653 1.2.10.2 tls } 654 1.2.10.2 tls /* 655 1.2.10.2 tls * Can be a hit only if one entry left. 656 1.2.10.2 tls * Note possible hit entry and put nfsrc_templist back on hash 657 1.2.10.2 tls * list. 658 1.2.10.2 tls */ 659 1.2.10.2 tls if (i != 1) 660 1.2.10.2 tls hit = 0; 661 1.2.10.2 tls hitrp = rp = LIST_FIRST(&nfsrc_templist); 662 1.2.10.2 tls while (rp != NULL) { 663 1.2.10.2 tls nextrp = LIST_NEXT(rp, rc_hash); 664 1.2.10.2 tls LIST_REMOVE(rp, rc_hash); 665 1.2.10.2 tls LIST_INSERT_HEAD(hp, rp, rc_hash); 666 1.2.10.2 tls rp = nextrp; 667 1.2.10.2 tls } 668 1.2.10.2 tls if (LIST_FIRST(&nfsrc_templist) != NULL) 669 1.2.10.2 tls panic("nfs gettcp cache templist"); 670 1.2.10.2 tls 671 1.2.10.2 tls if (hit) { 672 1.2.10.2 tls rp = hitrp; 673 1.2.10.2 tls if ((rp->rc_flag & RC_LOCKED) != 0) { 674 1.2.10.2 tls rp->rc_flag |= RC_WANTED; 675 1.2.10.2 tls (void)mtx_sleep(rp, mutex, (PZERO - 1) | PDROP, 676 1.2.10.2 tls "nfsrc", 10 * hz); 677 1.2.10.2 tls goto tryagain; 678 1.2.10.2 tls } 679 1.2.10.2 tls if (rp->rc_flag == 0) 680 1.2.10.2 tls panic("nfs tcp cache0"); 681 1.2.10.2 tls rp->rc_flag |= RC_LOCKED; 682 1.2.10.2 tls if (rp->rc_flag & RC_INPROG) { 683 1.2.10.3 jdolecek nfsstatsv1.srvcache_inproghits++; 684 1.2.10.2 tls mtx_unlock(mutex); 685 1.2.10.2 tls if (newrp->rc_sockref == rp->rc_sockref) 686 1.2.10.2 tls nfsrc_marksametcpconn(rp->rc_sockref); 687 1.2.10.2 tls ret = RC_DROPIT; 688 1.2.10.2 tls } else if (rp->rc_flag & RC_REPSTATUS) { 689 1.2.10.2 tls /* 690 1.2.10.2 tls * V2 only. 691 1.2.10.2 tls */ 692 1.2.10.3 jdolecek nfsstatsv1.srvcache_nonidemdonehits++; 693 1.2.10.2 tls mtx_unlock(mutex); 694 1.2.10.2 tls if (newrp->rc_sockref == rp->rc_sockref) 695 1.2.10.2 tls nfsrc_marksametcpconn(rp->rc_sockref); 696 1.2.10.2 tls ret = RC_REPLY; 697 1.2.10.2 tls nfsrvd_rephead(nd); 698 1.2.10.2 tls *(nd->nd_errp) = rp->rc_status; 699 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; 700 1.2.10.2 tls } else if (rp->rc_flag & RC_REPMBUF) { 701 1.2.10.3 jdolecek nfsstatsv1.srvcache_nonidemdonehits++; 702 1.2.10.2 tls mtx_unlock(mutex); 703 1.2.10.2 tls if (newrp->rc_sockref == rp->rc_sockref) 704 1.2.10.2 tls nfsrc_marksametcpconn(rp->rc_sockref); 705 1.2.10.2 tls ret = RC_REPLY; 706 1.2.10.2 tls nd->nd_mreq = m_copym(rp->rc_reply, 0, 707 1.2.10.2 tls M_COPYALL, M_WAITOK); 708 1.2.10.2 tls rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; 709 1.2.10.2 tls } else { 710 1.2.10.2 tls panic("nfs tcp cache1"); 711 1.2.10.2 tls } 712 1.2.10.2 tls nfsrc_unlock(rp); 713 1.2.10.2 tls free((caddr_t)newrp, M_NFSRVCACHE); 714 1.2.10.2 tls goto out; 715 1.2.10.2 tls } 716 1.2.10.3 jdolecek nfsstatsv1.srvcache_misses++; 717 1.2.10.3 jdolecek atomic_add_int(&nfsstatsv1.srvcache_size, 1); 718 1.2.10.2 tls 719 1.2.10.2 tls /* 720 1.2.10.2 tls * For TCP, multiple entries for a key are allowed, so don't 721 1.2.10.2 tls * chain it into the hash table until done. 722 1.2.10.2 tls */ 723 1.2.10.2 tls newrp->rc_cachetime = NFSD_MONOSEC; 724 1.2.10.2 tls newrp->rc_flag |= RC_INPROG; 725 1.2.10.2 tls LIST_INSERT_HEAD(hp, newrp, rc_hash); 726 1.2.10.2 tls mtx_unlock(mutex); 727 1.2.10.2 tls nd->nd_rp = newrp; 728 1.2.10.2 tls ret = RC_DOIT; 729 1.2.10.2 tls 730 1.2.10.2 tls out: 731 1.2.10.2 tls NFSEXITCODE2(0, nd); 732 1.2.10.2 tls return (ret); 733 1.2.10.2 tls } 734 1.2.10.2 tls 735 1.2.10.2 tls /* 736 1.2.10.2 tls * Lock a cache entry. 737 1.2.10.2 tls */ 738 1.2.10.2 tls static void 739 1.2.10.2 tls nfsrc_lock(struct nfsrvcache *rp) 740 1.2.10.2 tls { 741 1.2.10.2 tls struct mtx *mutex; 742 1.2.10.2 tls 743 1.2.10.2 tls mutex = nfsrc_cachemutex(rp); 744 1.2.10.2 tls mtx_assert(mutex, MA_OWNED); 745 1.2.10.2 tls while ((rp->rc_flag & RC_LOCKED) != 0) { 746 1.2.10.2 tls rp->rc_flag |= RC_WANTED; 747 1.2.10.2 tls (void)mtx_sleep(rp, mutex, PZERO - 1, "nfsrc", 0); 748 1.2.10.2 tls } 749 1.2.10.2 tls rp->rc_flag |= RC_LOCKED; 750 1.2.10.2 tls } 751 1.2.10.2 tls 752 1.2.10.2 tls /* 753 1.2.10.2 tls * Unlock a cache entry. 754 1.2.10.2 tls */ 755 1.2.10.2 tls static void 756 1.2.10.2 tls nfsrc_unlock(struct nfsrvcache *rp) 757 1.2.10.2 tls { 758 1.2.10.2 tls struct mtx *mutex; 759 1.2.10.2 tls 760 1.2.10.2 tls mutex = nfsrc_cachemutex(rp); 761 1.2.10.2 tls mtx_lock(mutex); 762 1.2.10.2 tls rp->rc_flag &= ~RC_LOCKED; 763 1.2.10.2 tls nfsrc_wanted(rp); 764 1.2.10.2 tls mtx_unlock(mutex); 765 1.2.10.2 tls } 766 1.2.10.2 tls 767 1.2.10.2 tls /* 768 1.2.10.2 tls * Wakeup anyone wanting entry. 769 1.2.10.2 tls */ 770 1.2.10.2 tls static void 771 1.2.10.2 tls nfsrc_wanted(struct nfsrvcache *rp) 772 1.2.10.2 tls { 773 1.2.10.2 tls if (rp->rc_flag & RC_WANTED) { 774 1.2.10.2 tls rp->rc_flag &= ~RC_WANTED; 775 1.2.10.2 tls wakeup((caddr_t)rp); 776 1.2.10.2 tls } 777 1.2.10.2 tls } 778 1.2.10.2 tls 779 1.2.10.2 tls /* 780 1.2.10.2 tls * Free up the entry. 781 1.2.10.2 tls * Must not sleep. 782 1.2.10.2 tls */ 783 1.2.10.2 tls static void 784 1.2.10.2 tls nfsrc_freecache(struct nfsrvcache *rp) 785 1.2.10.2 tls { 786 1.2.10.3 jdolecek struct nfsrchash_bucket *hbp; 787 1.2.10.2 tls 788 1.2.10.2 tls LIST_REMOVE(rp, rc_hash); 789 1.2.10.2 tls if (rp->rc_flag & RC_UDP) { 790 1.2.10.2 tls TAILQ_REMOVE(&nfsrvudplru, rp, rc_lru); 791 1.2.10.2 tls nfsrc_udpcachesize--; 792 1.2.10.3 jdolecek } else if (rp->rc_acked != RC_NO_SEQ) { 793 1.2.10.3 jdolecek hbp = NFSRCAHASH(rp->rc_sockref); 794 1.2.10.3 jdolecek mtx_lock(&hbp->mtx); 795 1.2.10.3 jdolecek if (rp->rc_acked == RC_NO_ACK) 796 1.2.10.3 jdolecek LIST_REMOVE(rp, rc_ahash); 797 1.2.10.3 jdolecek mtx_unlock(&hbp->mtx); 798 1.2.10.2 tls } 799 1.2.10.2 tls nfsrc_wanted(rp); 800 1.2.10.2 tls if (rp->rc_flag & RC_REPMBUF) { 801 1.2.10.2 tls mbuf_freem(rp->rc_reply); 802 1.2.10.2 tls if (!(rp->rc_flag & RC_UDP)) 803 1.2.10.2 tls atomic_add_int(&nfsrc_tcpsavedreplies, -1); 804 1.2.10.2 tls } 805 1.2.10.2 tls FREE((caddr_t)rp, M_NFSRVCACHE); 806 1.2.10.3 jdolecek atomic_add_int(&nfsstatsv1.srvcache_size, -1); 807 1.2.10.2 tls } 808 1.2.10.2 tls 809 1.2.10.2 tls /* 810 1.2.10.2 tls * Clean out the cache. Called when nfsserver module is unloaded. 811 1.2.10.2 tls */ 812 1.2.10.2 tls APPLESTATIC void 813 1.2.10.2 tls nfsrvd_cleancache(void) 814 1.2.10.2 tls { 815 1.2.10.2 tls struct nfsrvcache *rp, *nextrp; 816 1.2.10.2 tls int i; 817 1.2.10.2 tls 818 1.2.10.2 tls for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 819 1.2.10.2 tls mtx_lock(&nfsrchash_table[i].mtx); 820 1.2.10.2 tls LIST_FOREACH_SAFE(rp, &nfsrchash_table[i].tbl, rc_hash, nextrp) 821 1.2.10.2 tls nfsrc_freecache(rp); 822 1.2.10.2 tls mtx_unlock(&nfsrchash_table[i].mtx); 823 1.2.10.2 tls } 824 1.2.10.2 tls mtx_lock(&nfsrc_udpmtx); 825 1.2.10.2 tls for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 826 1.2.10.2 tls LIST_FOREACH_SAFE(rp, &nfsrvudphashtbl[i], rc_hash, nextrp) { 827 1.2.10.2 tls nfsrc_freecache(rp); 828 1.2.10.2 tls } 829 1.2.10.2 tls } 830 1.2.10.3 jdolecek nfsstatsv1.srvcache_size = 0; 831 1.2.10.2 tls mtx_unlock(&nfsrc_udpmtx); 832 1.2.10.2 tls nfsrc_tcpsavedreplies = 0; 833 1.2.10.2 tls } 834 1.2.10.2 tls 835 1.2.10.3 jdolecek #define HISTSIZE 16 836 1.2.10.2 tls /* 837 1.2.10.2 tls * The basic rule is to get rid of entries that are expired. 838 1.2.10.2 tls */ 839 1.2.10.3 jdolecek void 840 1.2.10.3 jdolecek nfsrc_trimcache(u_int64_t sockref, uint32_t snd_una, int final) 841 1.2.10.2 tls { 842 1.2.10.3 jdolecek struct nfsrchash_bucket *hbp; 843 1.2.10.2 tls struct nfsrvcache *rp, *nextrp; 844 1.2.10.3 jdolecek int force, lastslot, i, j, k, tto, time_histo[HISTSIZE]; 845 1.2.10.2 tls time_t thisstamp; 846 1.2.10.2 tls static time_t udp_lasttrim = 0, tcp_lasttrim = 0; 847 1.2.10.3 jdolecek static int onethread = 0, oneslot = 0; 848 1.2.10.3 jdolecek 849 1.2.10.3 jdolecek if (sockref != 0) { 850 1.2.10.3 jdolecek hbp = NFSRCAHASH(sockref); 851 1.2.10.3 jdolecek mtx_lock(&hbp->mtx); 852 1.2.10.3 jdolecek LIST_FOREACH_SAFE(rp, &hbp->tbl, rc_ahash, nextrp) { 853 1.2.10.3 jdolecek if (sockref == rp->rc_sockref) { 854 1.2.10.3 jdolecek if (SEQ_GEQ(snd_una, rp->rc_tcpseq)) { 855 1.2.10.3 jdolecek rp->rc_acked = RC_ACK; 856 1.2.10.3 jdolecek LIST_REMOVE(rp, rc_ahash); 857 1.2.10.3 jdolecek } else if (final) { 858 1.2.10.3 jdolecek rp->rc_acked = RC_NACK; 859 1.2.10.3 jdolecek LIST_REMOVE(rp, rc_ahash); 860 1.2.10.3 jdolecek } 861 1.2.10.3 jdolecek } 862 1.2.10.3 jdolecek } 863 1.2.10.3 jdolecek mtx_unlock(&hbp->mtx); 864 1.2.10.3 jdolecek } 865 1.2.10.2 tls 866 1.2.10.2 tls if (atomic_cmpset_acq_int(&onethread, 0, 1) == 0) 867 1.2.10.2 tls return; 868 1.2.10.2 tls if (NFSD_MONOSEC != udp_lasttrim || 869 1.2.10.2 tls nfsrc_udpcachesize >= (nfsrc_udphighwater + 870 1.2.10.2 tls nfsrc_udphighwater / 2)) { 871 1.2.10.2 tls mtx_lock(&nfsrc_udpmtx); 872 1.2.10.2 tls udp_lasttrim = NFSD_MONOSEC; 873 1.2.10.2 tls TAILQ_FOREACH_SAFE(rp, &nfsrvudplru, rc_lru, nextrp) { 874 1.2.10.2 tls if (!(rp->rc_flag & (RC_INPROG|RC_LOCKED|RC_WANTED)) 875 1.2.10.2 tls && rp->rc_refcnt == 0 876 1.2.10.2 tls && ((rp->rc_flag & RC_REFCNT) || 877 1.2.10.2 tls udp_lasttrim > rp->rc_timestamp || 878 1.2.10.2 tls nfsrc_udpcachesize > nfsrc_udphighwater)) 879 1.2.10.2 tls nfsrc_freecache(rp); 880 1.2.10.2 tls } 881 1.2.10.2 tls mtx_unlock(&nfsrc_udpmtx); 882 1.2.10.2 tls } 883 1.2.10.2 tls if (NFSD_MONOSEC != tcp_lasttrim || 884 1.2.10.2 tls nfsrc_tcpsavedreplies >= nfsrc_tcphighwater) { 885 1.2.10.3 jdolecek force = nfsrc_tcphighwater / 4; 886 1.2.10.3 jdolecek if (force > 0 && 887 1.2.10.3 jdolecek nfsrc_tcpsavedreplies + force >= nfsrc_tcphighwater) { 888 1.2.10.3 jdolecek for (i = 0; i < HISTSIZE; i++) 889 1.2.10.3 jdolecek time_histo[i] = 0; 890 1.2.10.3 jdolecek i = 0; 891 1.2.10.3 jdolecek lastslot = NFSRVCACHE_HASHSIZE - 1; 892 1.2.10.3 jdolecek } else { 893 1.2.10.3 jdolecek force = 0; 894 1.2.10.3 jdolecek if (NFSD_MONOSEC != tcp_lasttrim) { 895 1.2.10.3 jdolecek i = 0; 896 1.2.10.3 jdolecek lastslot = NFSRVCACHE_HASHSIZE - 1; 897 1.2.10.3 jdolecek } else { 898 1.2.10.3 jdolecek lastslot = i = oneslot; 899 1.2.10.3 jdolecek if (++oneslot >= NFSRVCACHE_HASHSIZE) 900 1.2.10.3 jdolecek oneslot = 0; 901 1.2.10.3 jdolecek } 902 1.2.10.3 jdolecek } 903 1.2.10.3 jdolecek tto = nfsrc_tcptimeout; 904 1.2.10.3 jdolecek tcp_lasttrim = NFSD_MONOSEC; 905 1.2.10.3 jdolecek for (; i <= lastslot; i++) { 906 1.2.10.2 tls mtx_lock(&nfsrchash_table[i].mtx); 907 1.2.10.2 tls LIST_FOREACH_SAFE(rp, &nfsrchash_table[i].tbl, rc_hash, 908 1.2.10.2 tls nextrp) { 909 1.2.10.2 tls if (!(rp->rc_flag & 910 1.2.10.2 tls (RC_INPROG|RC_LOCKED|RC_WANTED)) 911 1.2.10.2 tls && rp->rc_refcnt == 0) { 912 1.2.10.3 jdolecek if ((rp->rc_flag & RC_REFCNT) || 913 1.2.10.3 jdolecek tcp_lasttrim > rp->rc_timestamp || 914 1.2.10.3 jdolecek rp->rc_acked == RC_ACK) { 915 1.2.10.3 jdolecek nfsrc_freecache(rp); 916 1.2.10.3 jdolecek continue; 917 1.2.10.3 jdolecek } 918 1.2.10.3 jdolecek 919 1.2.10.3 jdolecek if (force == 0) 920 1.2.10.3 jdolecek continue; 921 1.2.10.2 tls /* 922 1.2.10.2 tls * The timestamps range from roughly the 923 1.2.10.2 tls * present (tcp_lasttrim) to the present 924 1.2.10.2 tls * + nfsrc_tcptimeout. Generate a simple 925 1.2.10.2 tls * histogram of where the timeouts fall. 926 1.2.10.2 tls */ 927 1.2.10.2 tls j = rp->rc_timestamp - tcp_lasttrim; 928 1.2.10.3 jdolecek if (j >= tto) 929 1.2.10.3 jdolecek j = HISTSIZE - 1; 930 1.2.10.3 jdolecek else if (j < 0) 931 1.2.10.2 tls j = 0; 932 1.2.10.3 jdolecek else 933 1.2.10.3 jdolecek j = j * HISTSIZE / tto; 934 1.2.10.2 tls time_histo[j]++; 935 1.2.10.2 tls } 936 1.2.10.2 tls } 937 1.2.10.2 tls mtx_unlock(&nfsrchash_table[i].mtx); 938 1.2.10.2 tls } 939 1.2.10.3 jdolecek if (force) { 940 1.2.10.2 tls /* 941 1.2.10.2 tls * Trim some more with a smaller timeout of as little 942 1.2.10.2 tls * as 20% of nfsrc_tcptimeout to try and get below 943 1.2.10.2 tls * 80% of the nfsrc_tcphighwater. 944 1.2.10.2 tls */ 945 1.2.10.2 tls k = 0; 946 1.2.10.3 jdolecek for (i = 0; i < (HISTSIZE - 2); i++) { 947 1.2.10.2 tls k += time_histo[i]; 948 1.2.10.3 jdolecek if (k > force) 949 1.2.10.2 tls break; 950 1.2.10.2 tls } 951 1.2.10.3 jdolecek k = tto * (i + 1) / HISTSIZE; 952 1.2.10.2 tls if (k < 1) 953 1.2.10.2 tls k = 1; 954 1.2.10.2 tls thisstamp = tcp_lasttrim + k; 955 1.2.10.2 tls for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 956 1.2.10.2 tls mtx_lock(&nfsrchash_table[i].mtx); 957 1.2.10.2 tls LIST_FOREACH_SAFE(rp, &nfsrchash_table[i].tbl, 958 1.2.10.2 tls rc_hash, nextrp) { 959 1.2.10.2 tls if (!(rp->rc_flag & 960 1.2.10.2 tls (RC_INPROG|RC_LOCKED|RC_WANTED)) 961 1.2.10.2 tls && rp->rc_refcnt == 0 962 1.2.10.2 tls && ((rp->rc_flag & RC_REFCNT) || 963 1.2.10.2 tls thisstamp > rp->rc_timestamp || 964 1.2.10.3 jdolecek rp->rc_acked == RC_ACK)) 965 1.2.10.2 tls nfsrc_freecache(rp); 966 1.2.10.2 tls } 967 1.2.10.2 tls mtx_unlock(&nfsrchash_table[i].mtx); 968 1.2.10.2 tls } 969 1.2.10.2 tls } 970 1.2.10.2 tls } 971 1.2.10.2 tls atomic_store_rel_int(&onethread, 0); 972 1.2.10.2 tls } 973 1.2.10.2 tls 974 1.2.10.2 tls /* 975 1.2.10.2 tls * Add a seqid# reference to the cache entry. 976 1.2.10.2 tls */ 977 1.2.10.2 tls APPLESTATIC void 978 1.2.10.2 tls nfsrvd_refcache(struct nfsrvcache *rp) 979 1.2.10.2 tls { 980 1.2.10.2 tls struct mtx *mutex; 981 1.2.10.2 tls 982 1.2.10.3 jdolecek if (rp == NULL) 983 1.2.10.3 jdolecek /* For NFSv4.1, there is no cache entry. */ 984 1.2.10.3 jdolecek return; 985 1.2.10.2 tls mutex = nfsrc_cachemutex(rp); 986 1.2.10.2 tls mtx_lock(mutex); 987 1.2.10.2 tls if (rp->rc_refcnt < 0) 988 1.2.10.2 tls panic("nfs cache refcnt"); 989 1.2.10.2 tls rp->rc_refcnt++; 990 1.2.10.2 tls mtx_unlock(mutex); 991 1.2.10.2 tls } 992 1.2.10.2 tls 993 1.2.10.2 tls /* 994 1.2.10.2 tls * Dereference a seqid# cache entry. 995 1.2.10.2 tls */ 996 1.2.10.2 tls APPLESTATIC void 997 1.2.10.2 tls nfsrvd_derefcache(struct nfsrvcache *rp) 998 1.2.10.2 tls { 999 1.2.10.2 tls struct mtx *mutex; 1000 1.2.10.2 tls 1001 1.2.10.2 tls mutex = nfsrc_cachemutex(rp); 1002 1.2.10.2 tls mtx_lock(mutex); 1003 1.2.10.2 tls if (rp->rc_refcnt <= 0) 1004 1.2.10.2 tls panic("nfs cache derefcnt"); 1005 1.2.10.2 tls rp->rc_refcnt--; 1006 1.2.10.2 tls if (rp->rc_refcnt == 0 && !(rp->rc_flag & (RC_LOCKED | RC_INPROG))) 1007 1.2.10.2 tls nfsrc_freecache(rp); 1008 1.2.10.2 tls mtx_unlock(mutex); 1009 1.2.10.2 tls } 1010 1.2.10.2 tls 1011 1.2.10.2 tls /* 1012 1.2.10.2 tls * Calculate the length of the mbuf list and a checksum on the first up to 1013 1.2.10.2 tls * NFSRVCACHE_CHECKLEN bytes. 1014 1.2.10.2 tls */ 1015 1.2.10.2 tls static int 1016 1.2.10.2 tls nfsrc_getlenandcksum(mbuf_t m1, u_int16_t *cksum) 1017 1.2.10.2 tls { 1018 1.2.10.2 tls int len = 0, cklen; 1019 1.2.10.2 tls mbuf_t m; 1020 1.2.10.2 tls 1021 1.2.10.2 tls m = m1; 1022 1.2.10.2 tls while (m) { 1023 1.2.10.2 tls len += mbuf_len(m); 1024 1.2.10.2 tls m = mbuf_next(m); 1025 1.2.10.2 tls } 1026 1.2.10.2 tls cklen = (len > NFSRVCACHE_CHECKLEN) ? NFSRVCACHE_CHECKLEN : len; 1027 1.2.10.2 tls *cksum = in_cksum(m1, cklen); 1028 1.2.10.2 tls return (len); 1029 1.2.10.2 tls } 1030 1.2.10.2 tls 1031 1.2.10.2 tls /* 1032 1.2.10.2 tls * Mark a TCP connection that is seeing retries. Should never happen for 1033 1.2.10.2 tls * NFSv4. 1034 1.2.10.2 tls */ 1035 1.2.10.2 tls static void 1036 1.2.10.2 tls nfsrc_marksametcpconn(u_int64_t sockref) 1037 1.2.10.2 tls { 1038 1.2.10.2 tls } 1039 1.2.10.2 tls 1040
Indexes created Wed Oct 01 19:09:53 GMT 2025