| 1 | /* |
| 2 | * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * The contents of this file constitute Original Code as defined in and |
| 7 | * are subject to the Apple Public Source License Version 1.1 (the |
| 8 | * "License"). You may not use this file except in compliance with the |
| 9 | * License. Please obtain a copy of the License at |
| 10 | * http://www.apple.com/publicsource and read it before using this file. |
| 11 | * |
| 12 | * This Original Code and all software distributed under the License are |
| 13 | * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 14 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 15 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 16 | * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the |
| 17 | * License for the specific language governing rights and limitations |
| 18 | * under the License. |
| 19 | * |
| 20 | * @APPLE_LICENSE_HEADER_END@ |
| 21 | */ |
| 22 | /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */ |
| 23 | /* |
| 24 | * Copyright (c) 1989, 1991, 1993, 1995 |
| 25 | * The Regents of the University of California. All rights reserved. |
| 26 | * |
| 27 | * This code is derived from software contributed to Berkeley by |
| 28 | * Rick Macklem at The University of Guelph. |
| 29 | * |
| 30 | * Redistribution and use in source and binary forms, with or without |
| 31 | * modification, are permitted provided that the following conditions |
| 32 | * are met: |
| 33 | * 1. Redistributions of source code must retain the above copyright |
| 34 | * notice, this list of conditions and the following disclaimer. |
| 35 | * 2. Redistributions in binary form must reproduce the above copyright |
| 36 | * notice, this list of conditions and the following disclaimer in the |
| 37 | * documentation and/or other materials provided with the distribution. |
| 38 | * 3. All advertising materials mentioning features or use of this software |
| 39 | * must display the following acknowledgement: |
| 40 | * This product includes software developed by the University of |
| 41 | * California, Berkeley and its contributors. |
| 42 | * 4. Neither the name of the University nor the names of its contributors |
| 43 | * may be used to endorse or promote products derived from this software |
| 44 | * without specific prior written permission. |
| 45 | * |
| 46 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 47 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 48 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 49 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 50 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 51 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 52 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 53 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 54 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 55 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 56 | * SUCH DAMAGE. |
| 57 | * |
| 58 | * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 |
| 59 | * FreeBSD-Id: nfs_socket.c,v 1.30 1997/10/28 15:59:07 bde Exp $ |
| 60 | */ |
| 61 | |
| 62 | /* |
| 63 | * Socket operations for use by nfs |
| 64 | */ |
| 65 | |
| 66 | #include <sys/param.h> |
| 67 | #include <sys/systm.h> |
| 68 | #include <sys/proc.h> |
| 69 | #include <sys/mount.h> |
| 70 | #include <sys/kernel.h> |
| 71 | #include <sys/mbuf.h> |
| 72 | #include <sys/malloc.h> |
| 73 | #include <sys/vnode.h> |
| 74 | #include <sys/domain.h> |
| 75 | #include <sys/protosw.h> |
| 76 | #include <sys/socket.h> |
| 77 | #include <sys/socketvar.h> |
| 78 | #include <sys/syslog.h> |
| 79 | #include <sys/tprintf.h> |
| 80 | #include <machine/spl.h> |
| 81 | |
| 82 | #include <sys/time.h> |
| 83 | #include <kern/clock.h> |
| 84 | #include <kern/task.h> |
| 85 | #include <kern/thread.h> |
| 86 | #include <sys/user.h> |
| 87 | |
| 88 | #include <netinet/in.h> |
| 89 | #include <netinet/tcp.h> |
| 90 | |
| 91 | #include <nfs/rpcv2.h> |
| 92 | #include <nfs/nfsproto.h> |
| 93 | #include <nfs/nfs.h> |
| 94 | #include <nfs/xdr_subs.h> |
| 95 | #include <nfs/nfsm_subs.h> |
| 96 | #include <nfs/nfsmount.h> |
| 97 | #include <nfs/nfsnode.h> |
| 98 | #include <nfs/nfsrtt.h> |
| 99 | #include <nfs/nqnfs.h> |
| 100 | |
| 101 | #include <sys/kdebug.h> |
| 102 | |
| 103 | #define FSDBG(A, B, C, D, E) \ |
| 104 | KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_NONE, \ |
| 105 | (int)(B), (int)(C), (int)(D), (int)(E), 0) |
| 106 | #define FSDBG_TOP(A, B, C, D, E) \ |
| 107 | KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_START, \ |
| 108 | (int)(B), (int)(C), (int)(D), (int)(E), 0) |
| 109 | #define FSDBG_BOT(A, B, C, D, E) \ |
| 110 | KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_END, \ |
| 111 | (int)(B), (int)(C), (int)(D), (int)(E), 0) |
| 112 | |
| 113 | #define TRUE 1 |
| 114 | #define FALSE 0 |
| 115 | |
| 116 | /* |
| 117 | * Estimate rto for an nfs rpc sent via. an unreliable datagram. |
| 118 | * Use the mean and mean deviation of rtt for the appropriate type of rpc |
| 119 | * for the frequent rpcs and a default for the others. |
| 120 | * The justification for doing "other" this way is that these rpcs |
| 121 | * happen so infrequently that timer est. would probably be stale. |
| 122 | * Also, since many of these rpcs are |
| 123 | * non-idempotent, a conservative timeout is desired. |
| 124 | * getattr, lookup - A+2D |
| 125 | * read, write - A+4D |
| 126 | * other - nm_timeo |
| 127 | */ |
| 128 | #define NFS_RTO(n, t) \ |
| 129 | ((t) == 0 ? (n)->nm_timeo : \ |
| 130 | ((t) < 3 ? \ |
| 131 | (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ |
| 132 | ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) |
| 133 | #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] |
| 134 | #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] |
| 135 | /* |
| 136 | * External data, mostly RPC constants in XDR form |
| 137 | */ |
| 138 | extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, |
| 139 | rpc_msgaccepted, rpc_call, rpc_autherr, |
| 140 | rpc_auth_kerb; |
| 141 | extern u_long nfs_prog, nqnfs_prog; |
| 142 | extern time_t nqnfsstarttime; |
| 143 | extern struct nfsstats nfsstats; |
| 144 | extern int nfsv3_procid[NFS_NPROCS]; |
| 145 | extern int nfs_ticks; |
| 146 | extern u_long nfs_xidwrap; |
| 147 | |
| 148 | /* |
| 149 | * Defines which timer to use for the procnum. |
| 150 | * 0 - default |
| 151 | * 1 - getattr |
| 152 | * 2 - lookup |
| 153 | * 3 - read |
| 154 | * 4 - write |
| 155 | */ |
| 156 | static int proct[NFS_NPROCS] = { |
| 157 | 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, |
| 158 | 0, 0, 0, |
| 159 | }; |
| 160 | |
| 161 | /* |
| 162 | * There is a congestion window for outstanding rpcs maintained per mount |
| 163 | * point. The cwnd size is adjusted in roughly the way that: |
| 164 | * Van Jacobson, Congestion avoidance and Control, In "Proceedings of |
| 165 | * SIGCOMM '88". ACM, August 1988. |
| 166 | * describes for TCP. The cwnd size is chopped in half on a retransmit timeout |
| 167 | * and incremented by 1/cwnd when each rpc reply is received and a full cwnd |
| 168 | * of rpcs is in progress. |
| 169 | * (The sent count and cwnd are scaled for integer arith.) |
| 170 | * Variants of "slow start" were tried and were found to be too much of a |
| 171 | * performance hit (ave. rtt 3 times larger), |
| 172 | * I suspect due to the large rtt that nfs rpcs have. |
| 173 | */ |
| 174 | #define NFS_CWNDSCALE 256 |
| 175 | #define NFS_MAXCWND (NFS_CWNDSCALE * 32) |
| 176 | static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; |
| 177 | int nfsrtton = 0; |
| 178 | struct nfsrtt nfsrtt; |
| 179 | |
| 180 | static int nfs_msg __P((struct proc *, const char *, const char *, int)); |
| 181 | static int nfs_rcvlock __P((struct nfsreq *)); |
| 182 | static void nfs_rcvunlock __P((struct nfsreq *)); |
| 183 | static int nfs_receive __P((struct nfsreq *rep, struct mbuf **aname, |
| 184 | struct mbuf **mp)); |
| 185 | static int nfs_reconnect __P((struct nfsreq *rep)); |
| 186 | static void nfs_repbusy(struct nfsreq *rep); |
| 187 | static struct nfsreq * nfs_repnext(struct nfsreq *rep); |
| 188 | static void nfs_repdequeue(struct nfsreq *rep); |
| 189 | |
| 190 | /* XXX */ |
| 191 | boolean_t current_thread_aborted(void); |
| 192 | kern_return_t thread_terminate(thread_act_t); |
| 193 | |
| 194 | #ifndef NFS_NOSERVER |
| 195 | static int nfsrv_getstream __P((struct nfssvc_sock *,int)); |
| 196 | |
| 197 | int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *nd, |
| 198 | struct nfssvc_sock *slp, |
| 199 | struct proc *procp, |
| 200 | struct mbuf **mreqp)) = { |
| 201 | nfsrv_null, |
| 202 | nfsrv_getattr, |
| 203 | nfsrv_setattr, |
| 204 | nfsrv_lookup, |
| 205 | nfsrv3_access, |
| 206 | nfsrv_readlink, |
| 207 | nfsrv_read, |
| 208 | nfsrv_write, |
| 209 | nfsrv_create, |
| 210 | nfsrv_mkdir, |
| 211 | nfsrv_symlink, |
| 212 | nfsrv_mknod, |
| 213 | nfsrv_remove, |
| 214 | nfsrv_rmdir, |
| 215 | nfsrv_rename, |
| 216 | nfsrv_link, |
| 217 | nfsrv_readdir, |
| 218 | nfsrv_readdirplus, |
| 219 | nfsrv_statfs, |
| 220 | nfsrv_fsinfo, |
| 221 | nfsrv_pathconf, |
| 222 | nfsrv_commit, |
| 223 | nqnfsrv_getlease, |
| 224 | nqnfsrv_vacated, |
| 225 | nfsrv_noop, |
| 226 | nfsrv_noop |
| 227 | }; |
| 228 | #endif /* NFS_NOSERVER */ |
| 229 | |
| 230 | /* |
| 231 | * NFSTRACE points were changed to FSDBG (KERNEL_DEBUG) |
| 232 | * But some of this code may prove useful someday... |
| 233 | */ |
| 234 | #undef NFSDIAG |
| 235 | #if NFSDIAG |
| 236 | int nfstraceindx = 0; |
| 237 | struct nfstracerec nfstracebuf[NFSTBUFSIZ] = {{0,0,0,0}}; |
| 238 | |
| 239 | #define NFSTRACESUSPENDERS |
| 240 | #ifdef NFSTRACESUSPENDERS |
| 241 | uint nfstracemask = 0xfff00200; |
| 242 | int nfstracexid = -1; |
| 243 | uint onfstracemask = 0; |
| 244 | int nfstracesuspend = -1; |
| 245 | #define NFSTRACE_SUSPEND \ |
| 246 | { \ |
| 247 | if (nfstracemask) { \ |
| 248 | onfstracemask = nfstracemask; \ |
| 249 | nfstracemask = 0; \ |
| 250 | } \ |
| 251 | } |
| 252 | #define NFSTRACE_RESUME \ |
| 253 | { \ |
| 254 | nfstracesuspend = -1; \ |
| 255 | if (!nfstracemask) \ |
| 256 | nfstracemask = onfstracemask; \ |
| 257 | } |
| 258 | #define NFSTRACE_STARTSUSPENDCOUNTDOWN \ |
| 259 | { \ |
| 260 | nfstracesuspend = (nfstraceindx+100) % NFSTBUFSIZ; \ |
| 261 | } |
| 262 | #define NFSTRACE_SUSPENDING (nfstracesuspend != -1) |
| 263 | #define NFSTRACE_SUSPENSEOVER \ |
| 264 | (nfstracesuspend > 100 ? \ |
| 265 | (nfstraceindx >= nfstracesuspend || \ |
| 266 | nfstraceindx < nfstracesuspend - 100) : \ |
| 267 | (nfstraceindx >= nfstracesuspend && \ |
| 268 | nfstraceindx < nfstracesuspend + 8192 - 100)) |
| 269 | #else |
| 270 | uint nfstracemask = 0; |
| 271 | #endif /* NFSTRACESUSPENDERS */ |
| 272 | |
| 273 | int nfsprnttimo = 1; |
| 274 | |
| 275 | int nfsodata[1024]; |
| 276 | int nfsoprocnum, nfsolen; |
| 277 | int nfsbt[32], nfsbtlen; |
| 278 | |
| 279 | #if defined(__ppc__) |
| 280 | int |
| 281 | backtrace(int *where, int size) |
| 282 | { |
| 283 | int register sp, *fp, numsaved; |
| 284 | |
| 285 | __asm__ volatile("mr %0,r1" : "=r" (sp)); |
| 286 | |
| 287 | fp = (int *)*((int *)sp); |
| 288 | size /= sizeof(int); |
| 289 | for (numsaved = 0; numsaved < size; numsaved++) { |
| 290 | *where++ = fp[2]; |
| 291 | if ((int)fp <= 0) |
| 292 | break; |
| 293 | fp = (int *)*fp; |
| 294 | } |
| 295 | return (numsaved); |
| 296 | } |
| 297 | #elif defined(__i386__) |
| 298 | int |
| 299 | backtrace() |
| 300 | { |
| 301 | return (0); /* Till someone implements a real routine */ |
| 302 | } |
| 303 | #else |
| 304 | #error architecture not implemented. |
| 305 | #endif |
| 306 | |
| 307 | void |
| 308 | nfsdup(struct nfsreq *rep) |
| 309 | { |
| 310 | int *ip, i, first = 1, end; |
| 311 | char *s, b[240]; |
| 312 | struct mbuf *mb; |
| 313 | |
| 314 | if ((nfs_debug & NFS_DEBUG_DUP) == 0) |
| 315 | return; |
| 316 | /* last mbuf in chain will be nfs content */ |
| 317 | for (mb = rep->r_mreq; mb->m_next; mb = mb->m_next) |
| 318 | ; |
| 319 | if (rep->r_procnum == nfsoprocnum && mb->m_len == nfsolen && |
| 320 | !bcmp((caddr_t)nfsodata, mb->m_data, nfsolen)) { |
| 321 | s = b + sprintf(b, "nfsdup x=%x p=%d h=", rep->r_xid, |
| 322 | rep->r_procnum); |
| 323 | end = (int)(VTONFS(rep->r_vp)->n_fhp); |
| 324 | ip = (int *)(end & ~3); |
| 325 | end += VTONFS(rep->r_vp)->n_fhsize; |
| 326 | while ((int)ip < end) { |
| 327 | i = *ip++; |
| 328 | if (first) { /* avoid leading zeroes */ |
| 329 | if (i == 0) |
| 330 | continue; |
| 331 | first = 0; |
| 332 | s += sprintf(s, "%x", i); |
| 333 | } else |
| 334 | s += sprintf(s, "%08x", i); |
| 335 | } |
| 336 | if (first) |
| 337 | sprintf(s, "%x", 0); |
| 338 | else /* eliminate trailing zeroes */ |
| 339 | while (*--s == '0') |
| 340 | *s = 0; |
| 341 | /* |
| 342 | * set a breakpoint here and you can view the |
| 343 | * current backtrace and the one saved in nfsbt |
| 344 | */ |
| 345 | kprintf("%s\n", b); |
| 346 | } |
| 347 | nfsoprocnum = rep->r_procnum; |
| 348 | nfsolen = mb->m_len; |
| 349 | bcopy(mb->m_data, (caddr_t)nfsodata, mb->m_len); |
| 350 | nfsbtlen = backtrace(&nfsbt, sizeof(nfsbt)); |
| 351 | } |
| 352 | #endif /* NFSDIAG */ |
| 353 | |
| 354 | |
| 355 | /* |
| 356 | * attempt to bind a socket to a reserved port |
| 357 | */ |
| 358 | static int |
| 359 | nfs_bind_resv(struct nfsmount *nmp) |
| 360 | { |
| 361 | struct socket *so = nmp->nm_so; |
| 362 | struct sockaddr_in sin; |
| 363 | int error; |
| 364 | u_short tport; |
| 365 | |
| 366 | if (!so) |
| 367 | return (EINVAL); |
| 368 | |
| 369 | sin.sin_len = sizeof (struct sockaddr_in); |
| 370 | sin.sin_family = AF_INET; |
| 371 | sin.sin_addr.s_addr = INADDR_ANY; |
| 372 | tport = IPPORT_RESERVED - 1; |
| 373 | sin.sin_port = htons(tport); |
| 374 | |
| 375 | while (((error = sobind(so, (struct sockaddr *) &sin)) == EADDRINUSE) && |
| 376 | (--tport > IPPORT_RESERVED / 2)) |
| 377 | sin.sin_port = htons(tport); |
| 378 | return (error); |
| 379 | } |
| 380 | |
| 381 | /* |
| 382 | * variables for managing the nfs_bind_resv_thread |
| 383 | */ |
| 384 | int nfs_resv_mounts = 0; |
| 385 | static int nfs_bind_resv_thread_state = 0; |
| 386 | #define NFS_BIND_RESV_THREAD_STATE_INITTED 1 |
| 387 | #define NFS_BIND_RESV_THREAD_STATE_RUNNING 2 |
| 388 | static struct slock nfs_bind_resv_slock; |
| 389 | struct nfs_bind_resv_request { |
| 390 | TAILQ_ENTRY(nfs_bind_resv_request) brr_chain; |
| 391 | struct nfsmount *brr_nmp; |
| 392 | int brr_error; |
| 393 | }; |
| 394 | static TAILQ_HEAD(, nfs_bind_resv_request) nfs_bind_resv_request_queue; |
| 395 | |
| 396 | /* |
| 397 | * thread to handle any reserved port bind requests |
| 398 | */ |
| 399 | static void |
| 400 | nfs_bind_resv_thread(void) |
| 401 | { |
| 402 | struct nfs_bind_resv_request *brreq; |
| 403 | boolean_t funnel_state; |
| 404 | |
| 405 | funnel_state = thread_funnel_set(network_flock, TRUE); |
| 406 | nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_RUNNING; |
| 407 | |
| 408 | while (nfs_resv_mounts > 0) { |
| 409 | simple_lock(&nfs_bind_resv_slock); |
| 410 | while ((brreq = TAILQ_FIRST(&nfs_bind_resv_request_queue))) { |
| 411 | TAILQ_REMOVE(&nfs_bind_resv_request_queue, brreq, brr_chain); |
| 412 | simple_unlock(&nfs_bind_resv_slock); |
| 413 | brreq->brr_error = nfs_bind_resv(brreq->brr_nmp); |
| 414 | wakeup(brreq); |
| 415 | simple_lock(&nfs_bind_resv_slock); |
| 416 | } |
| 417 | simple_unlock(&nfs_bind_resv_slock); |
| 418 | (void)tsleep((caddr_t)&nfs_bind_resv_request_queue, PSOCK, |
| 419 | "nfs_bind_resv_request_queue", 0); |
| 420 | } |
| 421 | |
| 422 | nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_INITTED; |
| 423 | (void) thread_funnel_set(network_flock, funnel_state); |
| 424 | (void) thread_terminate(current_act()); |
| 425 | } |
| 426 | |
| 427 | int |
| 428 | nfs_bind_resv_thread_wake(void) |
| 429 | { |
| 430 | if (nfs_bind_resv_thread_state < NFS_BIND_RESV_THREAD_STATE_RUNNING) |
| 431 | return (EIO); |
| 432 | wakeup(&nfs_bind_resv_request_queue); |
| 433 | return (0); |
| 434 | } |
| 435 | |
| 436 | /* |
| 437 | * underprivileged procs call this to request nfs_bind_resv_thread |
| 438 | * to perform the reserved port binding for them. |
| 439 | */ |
| 440 | static int |
| 441 | nfs_bind_resv_nopriv(struct nfsmount *nmp) |
| 442 | { |
| 443 | struct nfs_bind_resv_request brreq; |
| 444 | int error; |
| 445 | |
| 446 | if (nfs_bind_resv_thread_state < NFS_BIND_RESV_THREAD_STATE_RUNNING) { |
| 447 | if (nfs_bind_resv_thread_state < NFS_BIND_RESV_THREAD_STATE_INITTED) { |
| 448 | simple_lock_init(&nfs_bind_resv_slock); |
| 449 | TAILQ_INIT(&nfs_bind_resv_request_queue); |
| 450 | nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_INITTED; |
| 451 | } |
| 452 | kernel_thread(kernel_task, nfs_bind_resv_thread); |
| 453 | nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_RUNNING; |
| 454 | } |
| 455 | |
| 456 | brreq.brr_nmp = nmp; |
| 457 | brreq.brr_error = 0; |
| 458 | |
| 459 | simple_lock(&nfs_bind_resv_slock); |
| 460 | TAILQ_INSERT_TAIL(&nfs_bind_resv_request_queue, &brreq, brr_chain); |
| 461 | simple_unlock(&nfs_bind_resv_slock); |
| 462 | |
| 463 | error = nfs_bind_resv_thread_wake(); |
| 464 | if (error) { |
| 465 | TAILQ_REMOVE(&nfs_bind_resv_request_queue, &brreq, brr_chain); |
| 466 | /* Note: we might be able to simply restart the thread */ |
| 467 | return (error); |
| 468 | } |
| 469 | |
| 470 | (void) tsleep((caddr_t)&brreq, PSOCK, "nfsbindresv", 0); |
| 471 | |
| 472 | return (brreq.brr_error); |
| 473 | } |
| 474 | |
| 475 | /* |
| 476 | * Initialize sockets and congestion for a new NFS connection. |
| 477 | * We do not free the sockaddr if error. |
| 478 | */ |
| 479 | int |
| 480 | nfs_connect(nmp, rep) |
| 481 | struct nfsmount *nmp; |
| 482 | struct nfsreq *rep; |
| 483 | { |
| 484 | struct socket *so; |
| 485 | int s, error, rcvreserve, sndreserve; |
| 486 | struct sockaddr *saddr; |
| 487 | |
| 488 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 489 | nmp->nm_so = (struct socket *)0; |
| 490 | saddr = mtod(nmp->nm_nam, struct sockaddr *); |
| 491 | error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, |
| 492 | nmp->nm_soproto); |
| 493 | if (error) { |
| 494 | goto bad; |
| 495 | } |
| 496 | so = nmp->nm_so; |
| 497 | nmp->nm_soflags = so->so_proto->pr_flags; |
| 498 | |
| 499 | /* |
| 500 | * Some servers require that the client port be a reserved port number. |
| 501 | */ |
| 502 | if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { |
| 503 | struct proc *p; |
| 504 | /* |
| 505 | * sobind() requires current_proc() to have superuser privs. |
| 506 | * If this bind is part of a reconnect, and the current proc |
| 507 | * doesn't have superuser privs, we hand the sobind() off to |
| 508 | * a kernel thread to process. |
| 509 | */ |
| 510 | if ((nmp->nm_state & NFSSTA_MOUNTED) && |
| 511 | (p = current_proc()) && suser(p->p_ucred, &p->p_acflag)) { |
| 512 | /* request nfs_bind_resv_thread() to do bind */ |
| 513 | error = nfs_bind_resv_nopriv(nmp); |
| 514 | } else { |
| 515 | error = nfs_bind_resv(nmp); |
| 516 | } |
| 517 | if (error) |
| 518 | goto bad; |
| 519 | } |
| 520 | |
| 521 | /* |
| 522 | * Protocols that do not require connections may be optionally left |
| 523 | * unconnected for servers that reply from a port other than NFS_PORT. |
| 524 | */ |
| 525 | if (nmp->nm_flag & NFSMNT_NOCONN) { |
| 526 | if (nmp->nm_soflags & PR_CONNREQUIRED) { |
| 527 | error = ENOTCONN; |
| 528 | goto bad; |
| 529 | } |
| 530 | } else { |
| 531 | error = soconnect(so, mtod(nmp->nm_nam, struct sockaddr *)); |
| 532 | if (error) { |
| 533 | goto bad; |
| 534 | } |
| 535 | |
| 536 | /* |
| 537 | * Wait for the connection to complete. Cribbed from the |
| 538 | * connect system call but with the wait timing out so |
| 539 | * that interruptible mounts don't hang here for a long time. |
| 540 | */ |
| 541 | s = splnet(); |
| 542 | while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { |
| 543 | (void) tsleep((caddr_t)&so->so_timeo, PSOCK, |
| 544 | "nfscon", 2 * hz); |
| 545 | if ((so->so_state & SS_ISCONNECTING) && |
| 546 | so->so_error == 0 && rep && |
| 547 | (error = nfs_sigintr(nmp, rep, rep->r_procp))) { |
| 548 | so->so_state &= ~SS_ISCONNECTING; |
| 549 | splx(s); |
| 550 | goto bad; |
| 551 | } |
| 552 | } |
| 553 | if (so->so_error) { |
| 554 | error = so->so_error; |
| 555 | so->so_error = 0; |
| 556 | splx(s); |
| 557 | goto bad; |
| 558 | } |
| 559 | splx(s); |
| 560 | } |
| 561 | /* |
| 562 | * Always time out on recieve, this allows us to reconnect the |
| 563 | * socket to deal with network changes. |
| 564 | */ |
| 565 | so->so_rcv.sb_timeo = (2 * hz); |
| 566 | if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { |
| 567 | so->so_snd.sb_timeo = (5 * hz); |
| 568 | } else { |
| 569 | so->so_snd.sb_timeo = 0; |
| 570 | } |
| 571 | if (nmp->nm_sotype == SOCK_DGRAM) { |
| 572 | sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3; |
| 573 | rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * |
| 574 | (nmp->nm_readahead > 0 ? nmp->nm_readahead + 1 : 2); |
| 575 | } else if (nmp->nm_sotype == SOCK_SEQPACKET) { |
| 576 | sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3; |
| 577 | rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * |
| 578 | (nmp->nm_readahead > 0 ? nmp->nm_readahead + 1 : 2); |
| 579 | } else { |
| 580 | if (nmp->nm_sotype != SOCK_STREAM) |
| 581 | panic("nfscon sotype"); |
| 582 | |
| 583 | if (so->so_proto->pr_flags & PR_CONNREQUIRED) { |
| 584 | struct sockopt sopt; |
| 585 | int val; |
| 586 | |
| 587 | bzero(&sopt, sizeof sopt); |
| 588 | sopt.sopt_dir = SOPT_SET; |
| 589 | sopt.sopt_level = SOL_SOCKET; |
| 590 | sopt.sopt_name = SO_KEEPALIVE; |
| 591 | sopt.sopt_val = &val; |
| 592 | sopt.sopt_valsize = sizeof val; |
| 593 | val = 1; |
| 594 | sosetopt(so, &sopt); |
| 595 | } |
| 596 | if (so->so_proto->pr_protocol == IPPROTO_TCP) { |
| 597 | struct sockopt sopt; |
| 598 | int val; |
| 599 | |
| 600 | bzero(&sopt, sizeof sopt); |
| 601 | sopt.sopt_dir = SOPT_SET; |
| 602 | sopt.sopt_level = IPPROTO_TCP; |
| 603 | sopt.sopt_name = TCP_NODELAY; |
| 604 | sopt.sopt_val = &val; |
| 605 | sopt.sopt_valsize = sizeof val; |
| 606 | val = 1; |
| 607 | sosetopt(so, &sopt); |
| 608 | } |
| 609 | |
| 610 | sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long)) * 3; |
| 611 | rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long)) * |
| 612 | (nmp->nm_readahead > 0 ? nmp->nm_readahead + 1 : 2); |
| 613 | } |
| 614 | |
| 615 | if (sndreserve > NFS_MAXSOCKBUF) |
| 616 | sndreserve = NFS_MAXSOCKBUF; |
| 617 | if (rcvreserve > NFS_MAXSOCKBUF) |
| 618 | rcvreserve = NFS_MAXSOCKBUF; |
| 619 | error = soreserve(so, sndreserve, rcvreserve); |
| 620 | if (error) { |
| 621 | goto bad; |
| 622 | } |
| 623 | so->so_rcv.sb_flags |= SB_NOINTR; |
| 624 | so->so_snd.sb_flags |= SB_NOINTR; |
| 625 | |
| 626 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 627 | |
| 628 | /* Initialize other non-zero congestion variables */ |
| 629 | nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = |
| 630 | nmp->nm_srtt[3] = (NFS_TIMEO << 3); |
| 631 | nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = |
| 632 | nmp->nm_sdrtt[3] = 0; |
| 633 | nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ |
| 634 | nmp->nm_sent = 0; |
| 635 | FSDBG(529, nmp, nmp->nm_state, nmp->nm_soflags, nmp->nm_cwnd); |
| 636 | nmp->nm_timeouts = 0; |
| 637 | return (0); |
| 638 | |
| 639 | bad: |
| 640 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 641 | nfs_disconnect(nmp); |
| 642 | return (error); |
| 643 | } |
| 644 | |
| 645 | /* |
| 646 | * Reconnect routine: |
| 647 | * Called when a connection is broken on a reliable protocol. |
| 648 | * - clean up the old socket |
| 649 | * - nfs_connect() again |
| 650 | * - set R_MUSTRESEND for all outstanding requests on mount point |
| 651 | * If this fails the mount point is DEAD! |
| 652 | * nb: Must be called with the nfs_sndlock() set on the mount point. |
| 653 | */ |
| 654 | static int |
| 655 | nfs_reconnect(rep) |
| 656 | register struct nfsreq *rep; |
| 657 | { |
| 658 | register struct nfsreq *rp; |
| 659 | register struct nfsmount *nmp = rep->r_nmp; |
| 660 | int error; |
| 661 | |
| 662 | nfs_disconnect(nmp); |
| 663 | while ((error = nfs_connect(nmp, rep))) { |
| 664 | if (error == EINTR || error == ERESTART) |
| 665 | return (EINTR); |
| 666 | if (error == EIO) |
| 667 | return (EIO); |
| 668 | nfs_down(rep, rep->r_nmp, rep->r_procp, "can not connect", |
| 669 | error, NFSSTA_TIMEO); |
| 670 | if (!(nmp->nm_state & NFSSTA_MOUNTED)) { |
| 671 | /* we're not yet completely mounted and */ |
| 672 | /* we can't reconnect, so we fail */ |
| 673 | return (error); |
| 674 | } |
| 675 | if ((error = nfs_sigintr(rep->r_nmp, rep, rep->r_procp))) |
| 676 | return (error); |
| 677 | (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); |
| 678 | } |
| 679 | |
| 680 | NFS_DPF(DUP, ("nfs_reconnect RESEND\n")); |
| 681 | /* |
| 682 | * Loop through outstanding request list and fix up all requests |
| 683 | * on old socket. |
| 684 | */ |
| 685 | TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { |
| 686 | if (rp->r_nmp == nmp) |
| 687 | rp->r_flags |= R_MUSTRESEND; |
| 688 | } |
| 689 | return (0); |
| 690 | } |
| 691 | |
| 692 | /* |
| 693 | * NFS disconnect. Clean up and unlink. |
| 694 | */ |
| 695 | void |
| 696 | nfs_disconnect(nmp) |
| 697 | register struct nfsmount *nmp; |
| 698 | { |
| 699 | register struct socket *so; |
| 700 | |
| 701 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 702 | if (nmp->nm_so) { |
| 703 | so = nmp->nm_so; |
| 704 | nmp->nm_so = (struct socket *)0; |
| 705 | soshutdown(so, 2); |
| 706 | soclose(so); |
| 707 | } |
| 708 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 709 | } |
| 710 | |
| 711 | /* |
| 712 | * This is the nfs send routine. For connection based socket types, it |
| 713 | * must be called with an nfs_sndlock() on the socket. |
| 714 | * "rep == NULL" indicates that it has been called from a server. |
| 715 | * For the client side: |
| 716 | * - return EINTR if the RPC is terminated, 0 otherwise |
| 717 | * - set R_MUSTRESEND if the send fails for any reason |
| 718 | * - do any cleanup required by recoverable socket errors (???) |
| 719 | * For the server side: |
| 720 | * - return EINTR or ERESTART if interrupted by a signal |
| 721 | * - return EPIPE if a connection is lost for connection based sockets (TCP...) |
| 722 | * - do any cleanup required by recoverable socket errors (???) |
| 723 | */ |
| 724 | int |
| 725 | nfs_send(so, nam, top, rep) |
| 726 | register struct socket *so; |
| 727 | struct mbuf *nam; |
| 728 | register struct mbuf *top; |
| 729 | struct nfsreq *rep; |
| 730 | { |
| 731 | struct sockaddr *sendnam; |
| 732 | int error, error2, soflags, flags; |
| 733 | int xidqueued = 0; |
| 734 | struct nfsreq *rp; |
| 735 | char savenametolog[MNAMELEN]; |
| 736 | |
| 737 | if (rep) { |
| 738 | error = nfs_sigintr(rep->r_nmp, rep, rep->r_procp); |
| 739 | if (error) { |
| 740 | m_freem(top); |
| 741 | return (error); |
| 742 | } |
| 743 | if ((so = rep->r_nmp->nm_so) == NULL) { |
| 744 | rep->r_flags |= R_MUSTRESEND; |
| 745 | m_freem(top); |
| 746 | return (0); |
| 747 | } |
| 748 | rep->r_flags &= ~R_MUSTRESEND; |
| 749 | soflags = rep->r_nmp->nm_soflags; |
| 750 | TAILQ_FOREACH(rp, &nfs_reqq, r_chain) |
| 751 | if (rp == rep) |
| 752 | break; |
| 753 | if (rp) |
| 754 | xidqueued = rp->r_xid; |
| 755 | } else |
| 756 | soflags = so->so_proto->pr_flags; |
| 757 | if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED) || |
| 758 | (nam == 0)) |
| 759 | sendnam = (struct sockaddr *)0; |
| 760 | else |
| 761 | sendnam = mtod(nam, struct sockaddr *); |
| 762 | |
| 763 | if (so->so_type == SOCK_SEQPACKET) |
| 764 | flags = MSG_EOR; |
| 765 | else |
| 766 | flags = 0; |
| 767 | |
| 768 | #if NFSDIAG |
| 769 | if (rep) |
| 770 | nfsdup(rep); |
| 771 | #endif |
| 772 | /* |
| 773 | * Save the name here in case mount point goes away when we switch |
| 774 | * funnels. The name is using local stack and is large, but don't |
| 775 | * want to block if we malloc. |
| 776 | */ |
| 777 | if (rep) |
| 778 | strncpy(savenametolog, |
| 779 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname, |
| 780 | MNAMELEN); |
| 781 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 782 | error = sosend(so, sendnam, (struct uio *)0, top, |
| 783 | (struct mbuf *)0, flags); |
| 784 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 785 | |
| 786 | if (error) { |
| 787 | if (rep) { |
| 788 | if (xidqueued) { |
| 789 | TAILQ_FOREACH(rp, &nfs_reqq, r_chain) |
| 790 | if (rp == rep && rp->r_xid == xidqueued) |
| 791 | break; |
| 792 | if (!rp) |
| 793 | panic("nfs_send: error %d xid %x gone", |
| 794 | error, xidqueued); |
| 795 | } |
| 796 | log(LOG_INFO, "nfs send error %d for server %s\n", |
| 797 | error, savenametolog); |
| 798 | /* |
| 799 | * Deal with errors for the client side. |
| 800 | */ |
| 801 | error2 = nfs_sigintr(rep->r_nmp, rep, rep->r_procp); |
| 802 | if (error2) { |
| 803 | error = error2; |
| 804 | } else { |
| 805 | rep->r_flags |= R_MUSTRESEND; |
| 806 | NFS_DPF(DUP, |
| 807 | ("nfs_send RESEND error=%d\n", error)); |
| 808 | } |
| 809 | } else |
| 810 | log(LOG_INFO, "nfsd send error %d\n", error); |
| 811 | |
| 812 | /* |
| 813 | * Handle any recoverable (soft) socket errors here. (???) |
| 814 | */ |
| 815 | if (error != EINTR && error != ERESTART && error != EIO && |
| 816 | error != EWOULDBLOCK && error != EPIPE) { |
| 817 | error = 0; |
| 818 | } |
| 819 | } |
| 820 | return (error); |
| 821 | } |
| 822 | |
| 823 | /* |
| 824 | * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all |
| 825 | * done by soreceive(), but for SOCK_STREAM we must deal with the Record |
| 826 | * Mark and consolidate the data into a new mbuf list. |
| 827 | * nb: Sometimes TCP passes the data up to soreceive() in long lists of |
| 828 | * small mbufs. |
| 829 | * For SOCK_STREAM we must be very careful to read an entire record once |
| 830 | * we have read any of it, even if the system call has been interrupted. |
| 831 | */ |
| 832 | static int |
| 833 | nfs_receive(rep, aname, mp) |
| 834 | register struct nfsreq *rep; |
| 835 | struct mbuf **aname; |
| 836 | struct mbuf **mp; |
| 837 | { |
| 838 | register struct socket *so; |
| 839 | struct uio auio; |
| 840 | struct iovec aio; |
| 841 | register struct mbuf *m; |
| 842 | struct mbuf *control; |
| 843 | u_long len; |
| 844 | struct sockaddr **getnam; |
| 845 | struct sockaddr *tmp_nam; |
| 846 | struct mbuf *mhck; |
| 847 | struct sockaddr_in *sin; |
| 848 | int error, error2, sotype, rcvflg; |
| 849 | struct proc *p = current_proc(); /* XXX */ |
| 850 | |
| 851 | /* |
| 852 | * Set up arguments for soreceive() |
| 853 | */ |
| 854 | *mp = (struct mbuf *)0; |
| 855 | *aname = (struct mbuf *)0; |
| 856 | sotype = rep->r_nmp->nm_sotype; |
| 857 | |
| 858 | /* |
| 859 | * For reliable protocols, lock against other senders/receivers |
| 860 | * in case a reconnect is necessary. |
| 861 | * For SOCK_STREAM, first get the Record Mark to find out how much |
| 862 | * more there is to get. |
| 863 | * We must lock the socket against other receivers |
| 864 | * until we have an entire rpc request/reply. |
| 865 | */ |
| 866 | if (sotype != SOCK_DGRAM) { |
| 867 | error = nfs_sndlock(rep); |
| 868 | if (error) |
| 869 | return (error); |
| 870 | tryagain: |
| 871 | /* |
| 872 | * Check for fatal errors and resending request. |
| 873 | */ |
| 874 | /* |
| 875 | * Ugh: If a reconnect attempt just happened, nm_so |
| 876 | * would have changed. NULL indicates a failed |
| 877 | * attempt that has essentially shut down this |
| 878 | * mount point. |
| 879 | */ |
| 880 | if ((error = nfs_sigintr(rep->r_nmp, rep, p)) || rep->r_mrep) { |
| 881 | nfs_sndunlock(rep); |
| 882 | if (error) |
| 883 | return (error); |
| 884 | return (EINTR); |
| 885 | } |
| 886 | so = rep->r_nmp->nm_so; |
| 887 | if (!so) { |
| 888 | error = nfs_reconnect(rep); |
| 889 | if (error) { |
| 890 | nfs_sndunlock(rep); |
| 891 | return (error); |
| 892 | } |
| 893 | goto tryagain; |
| 894 | } |
| 895 | while (rep->r_flags & R_MUSTRESEND) { |
| 896 | m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); |
| 897 | nfsstats.rpcretries++; |
| 898 | NFS_DPF(DUP, |
| 899 | ("nfs_receive RESEND %s\n", |
| 900 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname)); |
| 901 | error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); |
| 902 | /* |
| 903 | * we also hold rcv lock so rep is still |
| 904 | * legit this point |
| 905 | */ |
| 906 | if (error) { |
| 907 | if (error == EINTR || error == ERESTART || |
| 908 | (error = nfs_reconnect(rep))) { |
| 909 | nfs_sndunlock(rep); |
| 910 | return (error); |
| 911 | } |
| 912 | goto tryagain; |
| 913 | } |
| 914 | } |
| 915 | nfs_sndunlock(rep); |
| 916 | if (sotype == SOCK_STREAM) { |
| 917 | aio.iov_base = (caddr_t) &len; |
| 918 | aio.iov_len = sizeof(u_long); |
| 919 | auio.uio_iov = &aio; |
| 920 | auio.uio_iovcnt = 1; |
| 921 | auio.uio_segflg = UIO_SYSSPACE; |
| 922 | auio.uio_rw = UIO_READ; |
| 923 | auio.uio_offset = 0; |
| 924 | auio.uio_resid = sizeof(u_long); |
| 925 | auio.uio_procp = p; |
| 926 | do { |
| 927 | rcvflg = MSG_WAITALL; |
| 928 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 929 | error = soreceive(so, (struct sockaddr **)0, &auio, |
| 930 | (struct mbuf **)0, (struct mbuf **)0, &rcvflg); |
| 931 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 932 | if (!rep->r_nmp) /* if unmounted then bailout */ |
| 933 | goto shutout; |
| 934 | if (error == EWOULDBLOCK && rep) { |
| 935 | error2 = nfs_sigintr(rep->r_nmp, rep, p); |
| 936 | if (error2) |
| 937 | error = error2; |
| 938 | } |
| 939 | } while (error == EWOULDBLOCK); |
| 940 | if (!error && auio.uio_resid > 0) { |
| 941 | log(LOG_INFO, |
| 942 | "short receive (%d/%d) from nfs server %s\n", |
| 943 | sizeof(u_long) - auio.uio_resid, |
| 944 | sizeof(u_long), |
| 945 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); |
| 946 | error = EPIPE; |
| 947 | } |
| 948 | if (error) |
| 949 | goto errout; |
| 950 | len = ntohl(len) & ~0x80000000; |
| 951 | /* |
| 952 | * This is SERIOUS! We are out of sync with the sender |
| 953 | * and forcing a disconnect/reconnect is all I can do. |
| 954 | */ |
| 955 | if (len > NFS_MAXPACKET) { |
| 956 | log(LOG_ERR, "%s (%d) from nfs server %s\n", |
| 957 | "impossible packet length", |
| 958 | len, |
| 959 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); |
| 960 | error = EFBIG; |
| 961 | goto errout; |
| 962 | } |
| 963 | auio.uio_resid = len; |
| 964 | |
| 965 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 966 | do { |
| 967 | rcvflg = MSG_WAITALL; |
| 968 | error = soreceive(so, (struct sockaddr **)0, |
| 969 | &auio, mp, (struct mbuf **)0, &rcvflg); |
| 970 | if (!rep->r_nmp) /* if unmounted then bailout */ { |
| 971 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 972 | goto shutout; |
| 973 | } |
| 974 | } while (error == EWOULDBLOCK || error == EINTR || |
| 975 | error == ERESTART); |
| 976 | |
| 977 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 978 | |
| 979 | if (!error && auio.uio_resid > 0) { |
| 980 | log(LOG_INFO, |
| 981 | "short receive (%d/%d) from nfs server %s\n", |
| 982 | len - auio.uio_resid, len, |
| 983 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); |
| 984 | error = EPIPE; |
| 985 | } |
| 986 | } else { |
| 987 | /* |
| 988 | * NB: Since uio_resid is big, MSG_WAITALL is ignored |
| 989 | * and soreceive() will return when it has either a |
| 990 | * control msg or a data msg. |
| 991 | * We have no use for control msg., but must grab them |
| 992 | * and then throw them away so we know what is going |
| 993 | * on. |
| 994 | */ |
| 995 | auio.uio_resid = len = 100000000; /* Anything Big */ |
| 996 | auio.uio_procp = p; |
| 997 | |
| 998 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 999 | do { |
| 1000 | control = NULL; |
| 1001 | rcvflg = 0; |
| 1002 | error = soreceive(so, (struct sockaddr **)0, |
| 1003 | &auio, mp, &control, &rcvflg); |
| 1004 | if (control) |
| 1005 | m_freem(control); |
| 1006 | if (!rep->r_nmp) /* if unmounted then bailout */ { |
| 1007 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 1008 | goto shutout; |
| 1009 | } |
| 1010 | if (error == EWOULDBLOCK && rep) { |
| 1011 | error2 = nfs_sigintr(rep->r_nmp, rep, p); |
| 1012 | if (error2) { |
| 1013 | thread_funnel_switch(NETWORK_FUNNEL, |
| 1014 | KERNEL_FUNNEL); |
| 1015 | return (error2); |
| 1016 | } |
| 1017 | } |
| 1018 | } while (error == EWOULDBLOCK || |
| 1019 | (!error && *mp == NULL && control)); |
| 1020 | |
| 1021 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 1022 | |
| 1023 | if ((rcvflg & MSG_EOR) == 0) |
| 1024 | printf("Egad!!\n"); |
| 1025 | if (!error && *mp == NULL) |
| 1026 | error = EPIPE; |
| 1027 | len -= auio.uio_resid; |
| 1028 | } |
| 1029 | errout: |
| 1030 | if (error && error != EINTR && error != ERESTART) { |
| 1031 | m_freem(*mp); |
| 1032 | *mp = (struct mbuf *)0; |
| 1033 | if (error != EPIPE) |
| 1034 | log(LOG_INFO, |
| 1035 | "receive error %d from nfs server %s\n", |
| 1036 | error, |
| 1037 | rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); |
| 1038 | error = nfs_sndlock(rep); |
| 1039 | if (!error) { |
| 1040 | error = nfs_reconnect(rep); |
| 1041 | if (!error) |
| 1042 | goto tryagain; |
| 1043 | nfs_sndunlock(rep); |
| 1044 | } |
| 1045 | } |
| 1046 | } else { |
| 1047 | /* |
| 1048 | * We could have failed while rebinding the datagram socket |
| 1049 | * so we need to attempt to rebind here. |
| 1050 | */ |
| 1051 | if ((so = rep->r_nmp->nm_so) == NULL) { |
| 1052 | error = nfs_sndlock(rep); |
| 1053 | if (!error) { |
| 1054 | error = nfs_reconnect(rep); |
| 1055 | nfs_sndunlock(rep); |
| 1056 | } |
| 1057 | if (error) |
| 1058 | return (error); |
| 1059 | if (!rep->r_nmp) /* if unmounted then bailout */ |
| 1060 | return (ENXIO); |
| 1061 | so = rep->r_nmp->nm_so; |
| 1062 | } |
| 1063 | if (so->so_state & SS_ISCONNECTED) |
| 1064 | getnam = (struct sockaddr **)0; |
| 1065 | else |
| 1066 | getnam = &tmp_nam;; |
| 1067 | auio.uio_resid = len = 1000000; |
| 1068 | auio.uio_procp = p; |
| 1069 | |
| 1070 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 1071 | do { |
| 1072 | rcvflg = 0; |
| 1073 | error = soreceive(so, getnam, &auio, mp, |
| 1074 | (struct mbuf **)0, &rcvflg); |
| 1075 | |
| 1076 | if ((getnam) && (*getnam)) { |
| 1077 | MGET(mhck, M_WAIT, MT_SONAME); |
| 1078 | mhck->m_len = (*getnam)->sa_len; |
| 1079 | sin = mtod(mhck, struct sockaddr_in *); |
| 1080 | bcopy(*getnam, sin, sizeof(struct sockaddr_in)); |
| 1081 | mhck->m_hdr.mh_len = sizeof(struct sockaddr_in); |
| 1082 | FREE(*getnam, M_SONAME); |
| 1083 | *aname = mhck; |
| 1084 | } |
| 1085 | if (!rep->r_nmp) /* if unmounted then bailout */ |
| 1086 | goto dgramout; |
| 1087 | if (error) { |
| 1088 | error2 = nfs_sigintr(rep->r_nmp, rep, p); |
| 1089 | if (error2) { |
| 1090 | error = error2; |
| 1091 | goto dgramout; |
| 1092 | } |
| 1093 | } |
| 1094 | /* Reconnect for all errors. We may be receiving |
| 1095 | * soft/hard/blocking errors because of a network |
| 1096 | * change. |
| 1097 | * XXX: we should rate limit or delay this |
| 1098 | * to once every N attempts or something. |
| 1099 | * although TCP doesn't seem to. |
| 1100 | */ |
| 1101 | if (error) { |
| 1102 | thread_funnel_switch(NETWORK_FUNNEL, |
| 1103 | KERNEL_FUNNEL); |
| 1104 | error2 = nfs_sndlock(rep); |
| 1105 | if (!error2) { |
| 1106 | error2 = nfs_reconnect(rep); |
| 1107 | if (error2) |
| 1108 | error = error2; |
| 1109 | else if (!rep->r_nmp) /* if unmounted then bailout */ |
| 1110 | error = ENXIO; |
| 1111 | else |
| 1112 | so = rep->r_nmp->nm_so; |
| 1113 | nfs_sndunlock(rep); |
| 1114 | } else { |
| 1115 | error = error2; |
| 1116 | } |
| 1117 | thread_funnel_switch(KERNEL_FUNNEL, |
| 1118 | NETWORK_FUNNEL); |
| 1119 | } |
| 1120 | } while (error == EWOULDBLOCK); |
| 1121 | |
| 1122 | dgramout: |
| 1123 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 1124 | len -= auio.uio_resid; |
| 1125 | } |
| 1126 | shutout: |
| 1127 | if (error) { |
| 1128 | m_freem(*mp); |
| 1129 | *mp = (struct mbuf *)0; |
| 1130 | } |
| 1131 | return (error); |
| 1132 | } |
| 1133 | |
| 1134 | /* |
| 1135 | * Implement receipt of reply on a socket. |
| 1136 | * We must search through the list of received datagrams matching them |
| 1137 | * with outstanding requests using the xid, until ours is found. |
| 1138 | */ |
| 1139 | /* ARGSUSED */ |
| 1140 | int |
| 1141 | nfs_reply(myrep) |
| 1142 | struct nfsreq *myrep; |
| 1143 | { |
| 1144 | register struct nfsreq *rep; |
| 1145 | register struct nfsmount *nmp = myrep->r_nmp; |
| 1146 | register long t1; |
| 1147 | struct mbuf *mrep, *md; |
| 1148 | struct mbuf *nam; |
| 1149 | u_long rxid, *tl; |
| 1150 | caddr_t dpos, cp2; |
| 1151 | int error; |
| 1152 | |
| 1153 | /* |
| 1154 | * Loop around until we get our own reply |
| 1155 | */ |
| 1156 | for (;;) { |
| 1157 | /* |
| 1158 | * Lock against other receivers so that I don't get stuck in |
| 1159 | * sbwait() after someone else has received my reply for me. |
| 1160 | * Also necessary for connection based protocols to avoid |
| 1161 | * race conditions during a reconnect. |
| 1162 | * If nfs_rcvlock() returns EALREADY, that means that |
| 1163 | * the reply has already been recieved by another |
| 1164 | * process and we can return immediately. In this |
| 1165 | * case, the lock is not taken to avoid races with |
| 1166 | * other processes. |
| 1167 | */ |
| 1168 | error = nfs_rcvlock(myrep); |
| 1169 | if (error == EALREADY) |
| 1170 | return (0); |
| 1171 | if (error) |
| 1172 | return (error); |
| 1173 | |
| 1174 | /* |
| 1175 | * If we slept after putting bits otw, then reply may have |
| 1176 | * arrived. In which case returning is required, or we |
| 1177 | * would hang trying to nfs_receive an already received reply. |
| 1178 | */ |
| 1179 | if (myrep->r_mrep != NULL) { |
| 1180 | nfs_rcvunlock(myrep); |
| 1181 | FSDBG(530, myrep->r_xid, myrep, myrep->r_nmp, -1); |
| 1182 | return (0); |
| 1183 | } |
| 1184 | /* |
| 1185 | * Get the next Rpc reply off the socket. Assume myrep->r_nmp |
| 1186 | * is still intact by checks done in nfs_rcvlock. |
| 1187 | */ |
| 1188 | /* XXX why do we ask for nam here? we don't use it! */ |
| 1189 | error = nfs_receive(myrep, &nam, &mrep); |
| 1190 | if (nam) |
| 1191 | m_freem(nam); |
| 1192 | /* |
| 1193 | * Bailout asap if nfsmount struct gone (unmounted). |
| 1194 | */ |
| 1195 | if (!myrep->r_nmp) { |
| 1196 | FSDBG(530, myrep->r_xid, myrep, nmp, -2); |
| 1197 | return (ENXIO); |
| 1198 | } |
| 1199 | if (error) { |
| 1200 | FSDBG(530, myrep->r_xid, myrep, nmp, error); |
| 1201 | nfs_rcvunlock(myrep); |
| 1202 | |
| 1203 | /* Bailout asap if nfsmount struct gone (unmounted). */ |
| 1204 | if (!myrep->r_nmp) |
| 1205 | return (ENXIO); |
| 1206 | |
| 1207 | /* |
| 1208 | * Ignore routing errors on connectionless protocols?? |
| 1209 | */ |
| 1210 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { |
| 1211 | if (nmp->nm_so) |
| 1212 | nmp->nm_so->so_error = 0; |
| 1213 | if (myrep->r_flags & R_GETONEREP) |
| 1214 | return (0); |
| 1215 | continue; |
| 1216 | } |
| 1217 | return (error); |
| 1218 | } |
| 1219 | |
| 1220 | /* |
| 1221 | * We assume all is fine, but if we did not have an error |
| 1222 | * and mrep is 0, better not dereference it. nfs_receieve |
| 1223 | * calls soreceive which carefully sets error=0 when it got |
| 1224 | * errors on sbwait (tsleep). In most cases, I assume that's |
| 1225 | * so we could go back again. In tcp case, EPIPE is returned. |
| 1226 | * In udp, case nfs_receive gets back here with no error and no |
| 1227 | * mrep. Is the right fix to have soreceive check for process |
| 1228 | * aborted after sbwait and return something non-zero? Should |
| 1229 | * nfs_receive give an EPIPE? Too risky to play with those |
| 1230 | * two this late in game for a shutdown problem. Instead, |
| 1231 | * just check here and get out. (ekn) |
| 1232 | */ |
| 1233 | if (!mrep) { |
| 1234 | nfs_rcvunlock(myrep); |
| 1235 | FSDBG(530, myrep->r_xid, myrep, nmp, -3); |
| 1236 | return (ENXIO); /* sounds good */ |
| 1237 | } |
| 1238 | |
| 1239 | /* |
| 1240 | * Get the xid and check that it is an rpc reply |
| 1241 | */ |
| 1242 | md = mrep; |
| 1243 | dpos = mtod(md, caddr_t); |
| 1244 | nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED); |
| 1245 | rxid = *tl++; |
| 1246 | if (*tl != rpc_reply) { |
| 1247 | #ifndef NFS_NOSERVER |
| 1248 | if (nmp->nm_flag & NFSMNT_NQNFS) { |
| 1249 | if (nqnfs_callback(nmp, mrep, md, dpos)) |
| 1250 | nfsstats.rpcinvalid++; |
| 1251 | } else { |
| 1252 | nfsstats.rpcinvalid++; |
| 1253 | m_freem(mrep); |
| 1254 | } |
| 1255 | #else |
| 1256 | nfsstats.rpcinvalid++; |
| 1257 | m_freem(mrep); |
| 1258 | #endif |
| 1259 | nfsmout: |
| 1260 | if (nmp->nm_state & NFSSTA_RCVLOCK) |
| 1261 | nfs_rcvunlock(myrep); |
| 1262 | if (myrep->r_flags & R_GETONEREP) |
| 1263 | return (0); /* this path used by NQNFS */ |
| 1264 | continue; |
| 1265 | } |
| 1266 | |
| 1267 | /* |
| 1268 | * Loop through the request list to match up the reply |
| 1269 | * Iff no match, just drop the datagram |
| 1270 | */ |
| 1271 | TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { |
| 1272 | if (rep->r_mrep == NULL && rxid == rep->r_xid) { |
| 1273 | /* Found it.. */ |
| 1274 | rep->r_mrep = mrep; |
| 1275 | rep->r_md = md; |
| 1276 | rep->r_dpos = dpos; |
| 1277 | /* |
| 1278 | * If we're tracking the round trip time |
| 1279 | * then we update the circular log here |
| 1280 | * with the stats from our current request. |
| 1281 | */ |
| 1282 | if (nfsrtton) { |
| 1283 | struct rttl *rt; |
| 1284 | |
| 1285 | rt = &nfsrtt.rttl[nfsrtt.pos]; |
| 1286 | rt->proc = rep->r_procnum; |
| 1287 | rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); |
| 1288 | rt->sent = nmp->nm_sent; |
| 1289 | rt->cwnd = nmp->nm_cwnd; |
| 1290 | if (proct[rep->r_procnum] == 0) |
| 1291 | panic("nfs_reply: proct[%d] is zero", rep->r_procnum); |
| 1292 | rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; |
| 1293 | rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; |
| 1294 | rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; |
| 1295 | microtime(&rt->tstamp); // XXX unused |
| 1296 | if (rep->r_flags & R_TIMING) |
| 1297 | rt->rtt = rep->r_rtt; |
| 1298 | else |
| 1299 | rt->rtt = 1000000; |
| 1300 | nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; |
| 1301 | } |
| 1302 | /* |
| 1303 | * Update congestion window. |
| 1304 | * Do the additive increase of |
| 1305 | * one rpc/rtt. |
| 1306 | */ |
| 1307 | FSDBG(530, rep->r_xid, rep, nmp->nm_sent, |
| 1308 | nmp->nm_cwnd); |
| 1309 | if (nmp->nm_cwnd <= nmp->nm_sent) { |
| 1310 | nmp->nm_cwnd += |
| 1311 | (NFS_CWNDSCALE * NFS_CWNDSCALE + |
| 1312 | (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; |
| 1313 | if (nmp->nm_cwnd > NFS_MAXCWND) |
| 1314 | nmp->nm_cwnd = NFS_MAXCWND; |
| 1315 | } |
| 1316 | if (rep->r_flags & R_SENT) { |
| 1317 | rep->r_flags &= ~R_SENT; |
| 1318 | nmp->nm_sent -= NFS_CWNDSCALE; |
| 1319 | } |
| 1320 | /* |
| 1321 | * Update rtt using a gain of 0.125 on the mean |
| 1322 | * and a gain of 0.25 on the deviation. |
| 1323 | */ |
| 1324 | if (rep->r_flags & R_TIMING) { |
| 1325 | /* |
| 1326 | * Since the timer resolution of |
| 1327 | * NFS_HZ is so course, it can often |
| 1328 | * result in r_rtt == 0. Since |
| 1329 | * r_rtt == N means that the actual |
| 1330 | * rtt is between N+dt and N+2-dt ticks, |
| 1331 | * add 1. |
| 1332 | */ |
| 1333 | if (proct[rep->r_procnum] == 0) |
| 1334 | panic("nfs_reply: proct[%d] is zero", rep->r_procnum); |
| 1335 | t1 = rep->r_rtt + 1; |
| 1336 | t1 -= (NFS_SRTT(rep) >> 3); |
| 1337 | NFS_SRTT(rep) += t1; |
| 1338 | if (t1 < 0) |
| 1339 | t1 = -t1; |
| 1340 | t1 -= (NFS_SDRTT(rep) >> 2); |
| 1341 | NFS_SDRTT(rep) += t1; |
| 1342 | } |
| 1343 | nmp->nm_timeouts = 0; |
| 1344 | break; |
| 1345 | } |
| 1346 | } |
| 1347 | nfs_rcvunlock(myrep); |
| 1348 | /* |
| 1349 | * If not matched to a request, drop it. |
| 1350 | * If it's mine, get out. |
| 1351 | */ |
| 1352 | if (rep == 0) { |
| 1353 | nfsstats.rpcunexpected++; |
| 1354 | m_freem(mrep); |
| 1355 | } else if (rep == myrep) { |
| 1356 | if (rep->r_mrep == NULL) |
| 1357 | panic("nfs_reply: nil r_mrep"); |
| 1358 | return (0); |
| 1359 | } |
| 1360 | FSDBG(530, myrep->r_xid, myrep, rep, |
| 1361 | rep ? rep->r_xid : myrep->r_flags); |
| 1362 | if (myrep->r_flags & R_GETONEREP) |
| 1363 | return (0); /* this path used by NQNFS */ |
| 1364 | } |
| 1365 | } |
| 1366 | |
| 1367 | /* |
| 1368 | * nfs_request - goes something like this |
| 1369 | * - fill in request struct |
| 1370 | * - links it into list |
| 1371 | * - calls nfs_send() for first transmit |
| 1372 | * - calls nfs_receive() to get reply |
| 1373 | * - break down rpc header and return with nfs reply pointed to |
| 1374 | * by mrep or error |
| 1375 | * nb: always frees up mreq mbuf list |
| 1376 | */ |
| 1377 | int |
| 1378 | nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp, xidp) |
| 1379 | struct vnode *vp; |
| 1380 | struct mbuf *mrest; |
| 1381 | int procnum; |
| 1382 | struct proc *procp; |
| 1383 | struct ucred *cred; |
| 1384 | struct mbuf **mrp; |
| 1385 | struct mbuf **mdp; |
| 1386 | caddr_t *dposp; |
| 1387 | u_int64_t *xidp; |
| 1388 | { |
| 1389 | register struct mbuf *m, *mrep, *m2; |
| 1390 | register struct nfsreq *rep, *rp; |
| 1391 | register u_long *tl; |
| 1392 | register int i; |
| 1393 | struct nfsmount *nmp; |
| 1394 | struct mbuf *md, *mheadend; |
| 1395 | struct nfsnode *np; |
| 1396 | char nickv[RPCX_NICKVERF]; |
| 1397 | time_t reqtime, waituntil; |
| 1398 | caddr_t dpos, cp2; |
| 1399 | int t1, nqlflag, cachable, s, error = 0, mrest_len, auth_len, auth_type; |
| 1400 | int trylater_delay = NQ_TRYLATERDEL, trylater_cnt = 0, failed_auth = 0; |
| 1401 | int verf_len, verf_type; |
| 1402 | u_long xid; |
| 1403 | u_quad_t frev; |
| 1404 | char *auth_str, *verf_str; |
| 1405 | NFSKERBKEY_T key; /* save session key */ |
| 1406 | int nmsotype; |
| 1407 | struct timeval now; |
| 1408 | |
| 1409 | if (mrp) |
| 1410 | *mrp = NULL; |
| 1411 | if (xidp) |
| 1412 | *xidp = 0; |
| 1413 | |
| 1414 | MALLOC_ZONE(rep, struct nfsreq *, |
| 1415 | sizeof(struct nfsreq), M_NFSREQ, M_WAITOK); |
| 1416 | |
| 1417 | nmp = VFSTONFS(vp->v_mount); |
| 1418 | if (nmp == NULL || |
| 1419 | (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_TIMEO)) == |
| 1420 | (NFSSTA_FORCE|NFSSTA_TIMEO)) { |
| 1421 | FREE_ZONE((caddr_t)rep, sizeof (struct nfsreq), M_NFSREQ); |
| 1422 | return (ENXIO); |
| 1423 | } |
| 1424 | nmsotype = nmp->nm_sotype; |
| 1425 | |
| 1426 | FSDBG_TOP(531, vp, procnum, nmp, rep); |
| 1427 | |
| 1428 | rep->r_nmp = nmp; |
| 1429 | rep->r_vp = vp; |
| 1430 | rep->r_procp = procp; |
| 1431 | rep->r_procnum = procnum; |
| 1432 | microuptime(&now); |
| 1433 | rep->r_lastmsg = now.tv_sec - |
| 1434 | ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay)); |
| 1435 | i = 0; |
| 1436 | m = mrest; |
| 1437 | while (m) { |
| 1438 | i += m->m_len; |
| 1439 | m = m->m_next; |
| 1440 | } |
| 1441 | mrest_len = i; |
| 1442 | |
| 1443 | /* |
| 1444 | * Get the RPC header with authorization. |
| 1445 | */ |
| 1446 | kerbauth: |
| 1447 | nmp = VFSTONFS(vp->v_mount); |
| 1448 | if (!nmp) { |
| 1449 | FSDBG_BOT(531, error, rep->r_xid, nmp, rep); |
| 1450 | FREE_ZONE((caddr_t)rep, sizeof (struct nfsreq), M_NFSREQ); |
| 1451 | return (ENXIO); |
| 1452 | } |
| 1453 | verf_str = auth_str = (char *)0; |
| 1454 | if (nmp->nm_flag & NFSMNT_KERB) { |
| 1455 | verf_str = nickv; |
| 1456 | verf_len = sizeof (nickv); |
| 1457 | auth_type = RPCAUTH_KERB4; |
| 1458 | bzero((caddr_t)key, sizeof (key)); |
| 1459 | if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, |
| 1460 | &auth_len, verf_str, verf_len)) { |
| 1461 | nmp = VFSTONFS(vp->v_mount); |
| 1462 | if (!nmp) { |
| 1463 | FSDBG_BOT(531, 2, vp, error, rep); |
| 1464 | FREE_ZONE((caddr_t)rep, |
| 1465 | sizeof (struct nfsreq), M_NFSREQ); |
| 1466 | m_freem(mrest); |
| 1467 | return (ENXIO); |
| 1468 | } |
| 1469 | error = nfs_getauth(nmp, rep, cred, &auth_str, |
| 1470 | &auth_len, verf_str, &verf_len, key); |
| 1471 | nmp = VFSTONFS(vp->v_mount); |
| 1472 | if (!error && !nmp) |
| 1473 | error = ENXIO; |
| 1474 | if (error) { |
| 1475 | FSDBG_BOT(531, 2, vp, error, rep); |
| 1476 | FREE_ZONE((caddr_t)rep, |
| 1477 | sizeof (struct nfsreq), M_NFSREQ); |
| 1478 | m_freem(mrest); |
| 1479 | return (error); |
| 1480 | } |
| 1481 | } |
| 1482 | } else { |
| 1483 | auth_type = RPCAUTH_UNIX; |
| 1484 | if (cred->cr_ngroups < 1) |
| 1485 | panic("nfsreq nogrps"); |
| 1486 | auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? |
| 1487 | nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + |
| 1488 | 5 * NFSX_UNSIGNED; |
| 1489 | } |
| 1490 | m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, |
| 1491 | auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); |
| 1492 | if (xidp) |
| 1493 | *xidp = ntohl(xid) + ((u_int64_t)nfs_xidwrap << 32); |
| 1494 | if (auth_str) |
| 1495 | _FREE(auth_str, M_TEMP); |
| 1496 | |
| 1497 | /* |
| 1498 | * For stream protocols, insert a Sun RPC Record Mark. |
| 1499 | */ |
| 1500 | if (nmsotype == SOCK_STREAM) { |
| 1501 | M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); |
| 1502 | *mtod(m, u_long *) = htonl(0x80000000 | |
| 1503 | (m->m_pkthdr.len - NFSX_UNSIGNED)); |
| 1504 | } |
| 1505 | rep->r_mreq = m; |
| 1506 | rep->r_xid = xid; |
| 1507 | tryagain: |
| 1508 | nmp = VFSTONFS(vp->v_mount); |
| 1509 | if (nmp && (nmp->nm_flag & NFSMNT_SOFT)) |
| 1510 | rep->r_retry = nmp->nm_retry; |
| 1511 | else |
| 1512 | rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ |
| 1513 | rep->r_rtt = rep->r_rexmit = 0; |
| 1514 | if (proct[procnum] > 0) |
| 1515 | rep->r_flags = R_TIMING; |
| 1516 | else |
| 1517 | rep->r_flags = 0; |
| 1518 | rep->r_mrep = NULL; |
| 1519 | |
| 1520 | /* |
| 1521 | * Do the client side RPC. |
| 1522 | */ |
| 1523 | nfsstats.rpcrequests++; |
| 1524 | /* |
| 1525 | * Chain request into list of outstanding requests. Be sure |
| 1526 | * to put it LAST so timer finds oldest requests first. |
| 1527 | */ |
| 1528 | s = splsoftclock(); |
| 1529 | TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); |
| 1530 | |
| 1531 | /* Get send time for nqnfs */ |
| 1532 | microtime(&now); |
| 1533 | reqtime = now.tv_sec; |
| 1534 | |
| 1535 | /* |
| 1536 | * If backing off another request or avoiding congestion, don't |
| 1537 | * send this one now but let timer do it. If not timing a request, |
| 1538 | * do it now. |
| 1539 | */ |
| 1540 | if (nmp && nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || |
| 1541 | (nmp->nm_flag & NFSMNT_DUMBTIMR) || |
| 1542 | nmp->nm_sent < nmp->nm_cwnd)) { |
| 1543 | int connrequired = (nmp->nm_soflags & PR_CONNREQUIRED); |
| 1544 | |
| 1545 | splx(s); |
| 1546 | if (connrequired) |
| 1547 | error = nfs_sndlock(rep); |
| 1548 | |
| 1549 | /* |
| 1550 | * Set the R_SENT before doing the send in case another thread |
| 1551 | * processes the reply before the nfs_send returns here |
| 1552 | */ |
| 1553 | if (!error) { |
| 1554 | if ((rep->r_flags & R_MUSTRESEND) == 0) { |
| 1555 | FSDBG(531, rep->r_xid, rep, nmp->nm_sent, |
| 1556 | nmp->nm_cwnd); |
| 1557 | nmp->nm_sent += NFS_CWNDSCALE; |
| 1558 | rep->r_flags |= R_SENT; |
| 1559 | } |
| 1560 | |
| 1561 | m2 = m_copym(m, 0, M_COPYALL, M_WAIT); |
| 1562 | error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep); |
| 1563 | if (connrequired) |
| 1564 | nfs_sndunlock(rep); |
| 1565 | } |
| 1566 | nmp = VFSTONFS(vp->v_mount); |
| 1567 | if (error) { |
| 1568 | if (nmp) |
| 1569 | nmp->nm_sent -= NFS_CWNDSCALE; |
| 1570 | rep->r_flags &= ~R_SENT; |
| 1571 | } |
| 1572 | } else { |
| 1573 | splx(s); |
| 1574 | rep->r_rtt = -1; |
| 1575 | } |
| 1576 | |
| 1577 | /* |
| 1578 | * Wait for the reply from our send or the timer's. |
| 1579 | */ |
| 1580 | if (!error || error == EPIPE) |
| 1581 | error = nfs_reply(rep); |
| 1582 | |
| 1583 | /* |
| 1584 | * RPC done, unlink the request. |
| 1585 | */ |
| 1586 | nfs_repdequeue(rep); |
| 1587 | |
| 1588 | nmp = VFSTONFS(vp->v_mount); |
| 1589 | |
| 1590 | /* |
| 1591 | * Decrement the outstanding request count. |
| 1592 | */ |
| 1593 | if (rep->r_flags & R_SENT) { |
| 1594 | rep->r_flags &= ~R_SENT; /* paranoia */ |
| 1595 | if (nmp) { |
| 1596 | FSDBG(531, rep->r_xid, rep, nmp->nm_sent, nmp->nm_cwnd); |
| 1597 | nmp->nm_sent -= NFS_CWNDSCALE; |
| 1598 | } |
| 1599 | } |
| 1600 | |
| 1601 | /* |
| 1602 | * If there was a successful reply and a tprintf msg. |
| 1603 | * tprintf a response. |
| 1604 | */ |
| 1605 | if (!error) |
| 1606 | nfs_up(rep, nmp, procp, "is alive again", NFSSTA_TIMEO); |
| 1607 | mrep = rep->r_mrep; |
| 1608 | md = rep->r_md; |
| 1609 | dpos = rep->r_dpos; |
| 1610 | if (!error && !nmp) |
| 1611 | error = ENXIO; |
| 1612 | if (error) { |
| 1613 | m_freem(rep->r_mreq); |
| 1614 | FSDBG_BOT(531, error, rep->r_xid, nmp, rep); |
| 1615 | FREE_ZONE((caddr_t)rep, sizeof (struct nfsreq), M_NFSREQ); |
| 1616 | return (error); |
| 1617 | } |
| 1618 | |
| 1619 | /* |
| 1620 | * break down the rpc header and check if ok |
| 1621 | */ |
| 1622 | nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); |
| 1623 | if (*tl++ == rpc_msgdenied) { |
| 1624 | if (*tl == rpc_mismatch) |
| 1625 | error = EOPNOTSUPP; |
| 1626 | else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { |
| 1627 | if (!failed_auth) { |
| 1628 | failed_auth++; |
| 1629 | mheadend->m_next = (struct mbuf *)0; |
| 1630 | m_freem(mrep); |
| 1631 | m_freem(rep->r_mreq); |
| 1632 | goto kerbauth; |
| 1633 | } else |
| 1634 | error = EAUTH; |
| 1635 | } else |
| 1636 | error = EACCES; |
| 1637 | m_freem(mrep); |
| 1638 | m_freem(rep->r_mreq); |
| 1639 | FSDBG_BOT(531, error, rep->r_xid, nmp, rep); |
| 1640 | FREE_ZONE((caddr_t)rep, sizeof (struct nfsreq), M_NFSREQ); |
| 1641 | return (error); |
| 1642 | } |
| 1643 | |
| 1644 | /* |
| 1645 | * Grab any Kerberos verifier, otherwise just throw it away. |
| 1646 | */ |
| 1647 | verf_type = fxdr_unsigned(int, *tl++); |
| 1648 | i = fxdr_unsigned(int, *tl); |
| 1649 | if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { |
| 1650 | error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); |
| 1651 | if (error) |
| 1652 | goto nfsmout; |
| 1653 | } else if (i > 0) |
| 1654 | nfsm_adv(nfsm_rndup(i)); |
| 1655 | nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); |
| 1656 | /* 0 == ok */ |
| 1657 | if (*tl == 0) { |
| 1658 | nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); |
| 1659 | if (*tl != 0) { |
| 1660 | error = fxdr_unsigned(int, *tl); |
| 1661 | if ((nmp->nm_flag & NFSMNT_NFSV3) && |
| 1662 | error == NFSERR_TRYLATER) { |
| 1663 | m_freem(mrep); |
| 1664 | error = 0; |
| 1665 | microuptime(&now); |
| 1666 | waituntil = now.tv_sec + trylater_delay; |
| 1667 | NFS_DPF(DUP, |
| 1668 | ("nfs_request %s flag=%x trylater_cnt=%x waituntil=%lx trylater_delay=%x\n", |
| 1669 | nmp->nm_mountp->mnt_stat.f_mntfromname, |
| 1670 | nmp->nm_flag, trylater_cnt, waituntil, |
| 1671 | trylater_delay)); |
| 1672 | while (now.tv_sec < waituntil) { |
| 1673 | (void)tsleep((caddr_t)&lbolt, |
| 1674 | PSOCK, "nqnfstry", 0); |
| 1675 | microuptime(&now); |
| 1676 | } |
| 1677 | trylater_delay *= 2; |
| 1678 | if (trylater_delay > 60) |
| 1679 | trylater_delay = 60; |
| 1680 | if (trylater_cnt < 7) |
| 1681 | trylater_cnt++; |
| 1682 | goto tryagain; |
| 1683 | } |
| 1684 | |
| 1685 | /* |
| 1686 | * If the File Handle was stale, invalidate the |
| 1687 | * lookup cache, just in case. |
| 1688 | */ |
| 1689 | if (error == ESTALE) |
| 1690 | cache_purge(vp); |
| 1691 | if (nmp->nm_flag & NFSMNT_NFSV3) { |
| 1692 | *mrp = mrep; |
| 1693 | *mdp = md; |
| 1694 | *dposp = dpos; |
| 1695 | error |= NFSERR_RETERR; |
| 1696 | } else { |
| 1697 | m_freem(mrep); |
| 1698 | error &= ~NFSERR_RETERR; |
| 1699 | } |
| 1700 | m_freem(rep->r_mreq); |
| 1701 | FSDBG_BOT(531, error, rep->r_xid, nmp, rep); |
| 1702 | FREE_ZONE((caddr_t)rep, |
| 1703 | sizeof (struct nfsreq), M_NFSREQ); |
| 1704 | return (error); |
| 1705 | } |
| 1706 | |
| 1707 | /* |
| 1708 | * For nqnfs, get any lease in reply |
| 1709 | */ |
| 1710 | if (nmp->nm_flag & NFSMNT_NQNFS) { |
| 1711 | nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); |
| 1712 | if (*tl) { |
| 1713 | np = VTONFS(vp); |
| 1714 | nqlflag = fxdr_unsigned(int, *tl); |
| 1715 | nfsm_dissect(tl, u_long *, 4*NFSX_UNSIGNED); |
| 1716 | cachable = fxdr_unsigned(int, *tl++); |
| 1717 | reqtime += fxdr_unsigned(int, *tl++); |
| 1718 | microtime(&now); |
| 1719 | if (reqtime > now.tv_sec) { |
| 1720 | fxdr_hyper(tl, &frev); |
| 1721 | nqnfs_clientlease(nmp, np, nqlflag, |
| 1722 | cachable, reqtime, frev); |
| 1723 | } |
| 1724 | } |
| 1725 | } |
| 1726 | *mrp = mrep; |
| 1727 | *mdp = md; |
| 1728 | *dposp = dpos; |
| 1729 | m_freem(rep->r_mreq); |
| 1730 | FSDBG_BOT(531, 0xf0f0f0f0, rep->r_xid, nmp, rep); |
| 1731 | FREE_ZONE((caddr_t)rep, sizeof (struct nfsreq), M_NFSREQ); |
| 1732 | return (0); |
| 1733 | } |
| 1734 | m_freem(mrep); |
| 1735 | error = EPROTONOSUPPORT; |
| 1736 | nfsmout: |
| 1737 | m_freem(rep->r_mreq); |
| 1738 | FSDBG_BOT(531, error, rep->r_xid, nmp, rep); |
| 1739 | FREE_ZONE((caddr_t)rep, sizeof (struct nfsreq), M_NFSREQ); |
| 1740 | return (error); |
| 1741 | } |
| 1742 | |
| 1743 | #ifndef NFS_NOSERVER |
| 1744 | /* |
| 1745 | * Generate the rpc reply header |
| 1746 | * siz arg. is used to decide if adding a cluster is worthwhile |
| 1747 | */ |
| 1748 | int |
| 1749 | nfs_rephead(siz, nd, slp, err, cache, frev, mrq, mbp, bposp) |
| 1750 | int siz; |
| 1751 | struct nfsrv_descript *nd; |
| 1752 | struct nfssvc_sock *slp; |
| 1753 | int err; |
| 1754 | int cache; |
| 1755 | u_quad_t *frev; |
| 1756 | struct mbuf **mrq; |
| 1757 | struct mbuf **mbp; |
| 1758 | caddr_t *bposp; |
| 1759 | { |
| 1760 | register u_long *tl; |
| 1761 | register struct mbuf *mreq; |
| 1762 | caddr_t bpos; |
| 1763 | struct mbuf *mb, *mb2; |
| 1764 | |
| 1765 | MGETHDR(mreq, M_WAIT, MT_DATA); |
| 1766 | mb = mreq; |
| 1767 | /* |
| 1768 | * If this is a big reply, use a cluster else |
| 1769 | * try and leave leading space for the lower level headers. |
| 1770 | */ |
| 1771 | siz += RPC_REPLYSIZ; |
| 1772 | if (siz >= MINCLSIZE) { |
| 1773 | MCLGET(mreq, M_WAIT); |
| 1774 | } else |
| 1775 | mreq->m_data += max_hdr; |
| 1776 | tl = mtod(mreq, u_long *); |
| 1777 | mreq->m_len = 6 * NFSX_UNSIGNED; |
| 1778 | bpos = ((caddr_t)tl) + mreq->m_len; |
| 1779 | *tl++ = txdr_unsigned(nd->nd_retxid); |
| 1780 | *tl++ = rpc_reply; |
| 1781 | if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { |
| 1782 | *tl++ = rpc_msgdenied; |
| 1783 | if (err & NFSERR_AUTHERR) { |
| 1784 | *tl++ = rpc_autherr; |
| 1785 | *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); |
| 1786 | mreq->m_len -= NFSX_UNSIGNED; |
| 1787 | bpos -= NFSX_UNSIGNED; |
| 1788 | } else { |
| 1789 | *tl++ = rpc_mismatch; |
| 1790 | *tl++ = txdr_unsigned(RPC_VER2); |
| 1791 | *tl = txdr_unsigned(RPC_VER2); |
| 1792 | } |
| 1793 | } else { |
| 1794 | *tl++ = rpc_msgaccepted; |
| 1795 | |
| 1796 | /* |
| 1797 | * For Kerberos authentication, we must send the nickname |
| 1798 | * verifier back, otherwise just RPCAUTH_NULL. |
| 1799 | */ |
| 1800 | if (nd->nd_flag & ND_KERBFULL) { |
| 1801 | register struct nfsuid *nuidp; |
| 1802 | struct timeval ktvin, ktvout; |
| 1803 | |
| 1804 | for (nuidp = NUIDHASH(slp, nd->nd_cr.cr_uid)->lh_first; |
| 1805 | nuidp != 0; nuidp = nuidp->nu_hash.le_next) { |
| 1806 | if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && |
| 1807 | (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), |
| 1808 | &nuidp->nu_haddr, nd->nd_nam2))) |
| 1809 | break; |
| 1810 | } |
| 1811 | if (nuidp) { |
| 1812 | ktvin.tv_sec = |
| 1813 | txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); |
| 1814 | ktvin.tv_usec = |
| 1815 | txdr_unsigned(nuidp->nu_timestamp.tv_usec); |
| 1816 | |
| 1817 | /* |
| 1818 | * Encrypt the timestamp in ecb mode using the |
| 1819 | * session key. |
| 1820 | */ |
| 1821 | #if NFSKERB |
| 1822 | XXX |
| 1823 | #endif |
| 1824 | |
| 1825 | *tl++ = rpc_auth_kerb; |
| 1826 | *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); |
| 1827 | *tl = ktvout.tv_sec; |
| 1828 | nfsm_build(tl, u_long *, 3 * NFSX_UNSIGNED); |
| 1829 | *tl++ = ktvout.tv_usec; |
| 1830 | *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); |
| 1831 | } else { |
| 1832 | *tl++ = 0; |
| 1833 | *tl++ = 0; |
| 1834 | } |
| 1835 | } else { |
| 1836 | *tl++ = 0; |
| 1837 | *tl++ = 0; |
| 1838 | } |
| 1839 | switch (err) { |
| 1840 | case EPROGUNAVAIL: |
| 1841 | *tl = txdr_unsigned(RPC_PROGUNAVAIL); |
| 1842 | break; |
| 1843 | case EPROGMISMATCH: |
| 1844 | *tl = txdr_unsigned(RPC_PROGMISMATCH); |
| 1845 | nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED); |
| 1846 | if (nd->nd_flag & ND_NQNFS) { |
| 1847 | *tl++ = txdr_unsigned(3); |
| 1848 | *tl = txdr_unsigned(3); |
| 1849 | } else { |
| 1850 | *tl++ = txdr_unsigned(2); |
| 1851 | *tl = txdr_unsigned(3); |
| 1852 | } |
| 1853 | break; |
| 1854 | case EPROCUNAVAIL: |
| 1855 | *tl = txdr_unsigned(RPC_PROCUNAVAIL); |
| 1856 | break; |
| 1857 | case EBADRPC: |
| 1858 | *tl = txdr_unsigned(RPC_GARBAGE); |
| 1859 | break; |
| 1860 | default: |
| 1861 | *tl = 0; |
| 1862 | if (err != NFSERR_RETVOID) { |
| 1863 | nfsm_build(tl, u_long *, NFSX_UNSIGNED); |
| 1864 | if (err) |
| 1865 | *tl = txdr_unsigned(nfsrv_errmap(nd, err)); |
| 1866 | else |
| 1867 | *tl = 0; |
| 1868 | } |
| 1869 | break; |
| 1870 | }; |
| 1871 | } |
| 1872 | |
| 1873 | /* |
| 1874 | * For nqnfs, piggyback lease as requested. |
| 1875 | */ |
| 1876 | if ((nd->nd_flag & ND_NQNFS) && err == 0) { |
| 1877 | if (nd->nd_flag & ND_LEASE) { |
| 1878 | nfsm_build(tl, u_long *, 5 * NFSX_UNSIGNED); |
| 1879 | *tl++ = txdr_unsigned(nd->nd_flag & ND_LEASE); |
| 1880 | *tl++ = txdr_unsigned(cache); |
| 1881 | *tl++ = txdr_unsigned(nd->nd_duration); |
| 1882 | txdr_hyper(frev, tl); |
| 1883 | } else { |
| 1884 | nfsm_build(tl, u_long *, NFSX_UNSIGNED); |
| 1885 | *tl = 0; |
| 1886 | } |
| 1887 | } |
| 1888 | if (mrq != NULL) |
| 1889 | *mrq = mreq; |
| 1890 | *mbp = mb; |
| 1891 | *bposp = bpos; |
| 1892 | if (err != 0 && err != NFSERR_RETVOID) |
| 1893 | nfsstats.srvrpc_errs++; |
| 1894 | return (0); |
| 1895 | } |
| 1896 | |
| 1897 | |
| 1898 | #endif /* NFS_NOSERVER */ |
| 1899 | |
| 1900 | |
| 1901 | /* |
| 1902 | * From FreeBSD 1.58, a Matt Dillon fix... |
| 1903 | * Flag a request as being about to terminate. |
| 1904 | * The nm_sent count is decremented now to avoid deadlocks when the process |
| 1905 | * in soreceive() hasn't yet managed to send its own request. |
| 1906 | */ |
| 1907 | static void |
| 1908 | nfs_softterm(struct nfsreq *rep) |
| 1909 | { |
| 1910 | |
| 1911 | rep->r_flags |= R_SOFTTERM; |
| 1912 | if (rep->r_flags & R_SENT) { |
| 1913 | FSDBG(532, rep->r_xid, rep, rep->r_nmp->nm_sent, |
| 1914 | rep->r_nmp->nm_cwnd); |
| 1915 | rep->r_nmp->nm_sent -= NFS_CWNDSCALE; |
| 1916 | rep->r_flags &= ~R_SENT; |
| 1917 | } |
| 1918 | } |
| 1919 | |
| 1920 | void |
| 1921 | nfs_timer_funnel(arg) |
| 1922 | void * arg; |
| 1923 | { |
| 1924 | (void) thread_funnel_set(kernel_flock, TRUE); |
| 1925 | nfs_timer(arg); |
| 1926 | (void) thread_funnel_set(kernel_flock, FALSE); |
| 1927 | |
| 1928 | } |
| 1929 | |
| 1930 | /* |
| 1931 | * Ensure rep isn't in use by the timer, then dequeue it. |
| 1932 | */ |
| 1933 | void |
| 1934 | nfs_repdequeue(struct nfsreq *rep) |
| 1935 | { |
| 1936 | int s; |
| 1937 | |
| 1938 | while ((rep->r_flags & R_BUSY)) { |
| 1939 | rep->r_flags |= R_WAITING; |
| 1940 | tsleep(rep, PSOCK, "repdeq", 0); |
| 1941 | } |
| 1942 | s = splsoftclock(); |
| 1943 | TAILQ_REMOVE(&nfs_reqq, rep, r_chain); |
| 1944 | splx(s); |
| 1945 | } |
| 1946 | |
| 1947 | /* |
| 1948 | * Busy (lock) a nfsreq, used by the nfs timer to make sure it's not |
| 1949 | * free()'d out from under it. |
| 1950 | */ |
| 1951 | void |
| 1952 | nfs_repbusy(struct nfsreq *rep) |
| 1953 | { |
| 1954 | |
| 1955 | if ((rep->r_flags & R_BUSY)) |
| 1956 | panic("rep locked"); |
| 1957 | rep->r_flags |= R_BUSY; |
| 1958 | } |
| 1959 | |
| 1960 | /* |
| 1961 | * Unbusy the nfsreq passed in, return the next nfsreq in the chain busied. |
| 1962 | */ |
| 1963 | struct nfsreq * |
| 1964 | nfs_repnext(struct nfsreq *rep) |
| 1965 | { |
| 1966 | struct nfsreq * nextrep; |
| 1967 | |
| 1968 | if (rep == NULL) |
| 1969 | return (NULL); |
| 1970 | /* |
| 1971 | * We need to get and busy the next req before signalling the |
| 1972 | * current one, otherwise wakeup() may block us and we'll race to |
| 1973 | * grab the next req. |
| 1974 | */ |
| 1975 | nextrep = TAILQ_NEXT(rep, r_chain); |
| 1976 | if (nextrep != NULL) |
| 1977 | nfs_repbusy(nextrep); |
| 1978 | /* unbusy and signal. */ |
| 1979 | rep->r_flags &= ~R_BUSY; |
| 1980 | if ((rep->r_flags & R_WAITING)) { |
| 1981 | rep->r_flags &= ~R_WAITING; |
| 1982 | wakeup(rep); |
| 1983 | } |
| 1984 | return (nextrep); |
| 1985 | } |
| 1986 | |
| 1987 | /* |
| 1988 | * Nfs timer routine |
| 1989 | * Scan the nfsreq list and retranmit any requests that have timed out |
| 1990 | * To avoid retransmission attempts on STREAM sockets (in the future) make |
| 1991 | * sure to set the r_retry field to 0 (implies nm_retry == 0). |
| 1992 | */ |
| 1993 | void |
| 1994 | nfs_timer(arg) |
| 1995 | void *arg; /* never used */ |
| 1996 | { |
| 1997 | register struct nfsreq *rep; |
| 1998 | register struct mbuf *m; |
| 1999 | register struct socket *so; |
| 2000 | register struct nfsmount *nmp; |
| 2001 | register int timeo; |
| 2002 | int s, error; |
| 2003 | #ifndef NFS_NOSERVER |
| 2004 | static long lasttime = 0; |
| 2005 | register struct nfssvc_sock *slp; |
| 2006 | u_quad_t cur_usec; |
| 2007 | #endif /* NFS_NOSERVER */ |
| 2008 | #if NFSDIAG |
| 2009 | int rttdiag; |
| 2010 | #endif |
| 2011 | int flags, rexmit, cwnd, sent; |
| 2012 | u_long xid; |
| 2013 | struct timeval now; |
| 2014 | |
| 2015 | s = splnet(); |
| 2016 | /* |
| 2017 | * XXX If preemptable threads are implemented the spls used for the |
| 2018 | * outstanding request queue must be replaced with mutexes. |
| 2019 | */ |
| 2020 | #ifdef NFSTRACESUSPENDERS |
| 2021 | if (NFSTRACE_SUSPENDING) { |
| 2022 | TAILQ_FOREACH(rep, &nfs_reqq, r_chain) |
| 2023 | if (rep->r_xid == nfstracexid) |
| 2024 | break; |
| 2025 | if (!rep) { |
| 2026 | NFSTRACE_RESUME; |
| 2027 | } else if (NFSTRACE_SUSPENSEOVER) { |
| 2028 | NFSTRACE_SUSPEND; |
| 2029 | } |
| 2030 | } |
| 2031 | #endif |
| 2032 | rep = TAILQ_FIRST(&nfs_reqq); |
| 2033 | if (rep != NULL) |
| 2034 | nfs_repbusy(rep); |
| 2035 | microuptime(&now); |
| 2036 | for ( ; rep != NULL ; rep = nfs_repnext(rep)) { |
| 2037 | #ifdef NFSTRACESUSPENDERS |
| 2038 | if (rep->r_mrep && !NFSTRACE_SUSPENDING) { |
| 2039 | nfstracexid = rep->r_xid; |
| 2040 | NFSTRACE_STARTSUSPENDCOUNTDOWN; |
| 2041 | } |
| 2042 | #endif |
| 2043 | nmp = rep->r_nmp; |
| 2044 | if (!nmp) /* unmounted */ |
| 2045 | continue; |
| 2046 | if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) |
| 2047 | continue; |
| 2048 | if (nfs_sigintr(nmp, rep, rep->r_procp)) |
| 2049 | continue; |
| 2050 | if (nmp->nm_tprintf_initial_delay != 0 && |
| 2051 | (rep->r_rexmit > 2 || (rep->r_flags & R_RESENDERR)) && |
| 2052 | rep->r_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) { |
| 2053 | rep->r_lastmsg = now.tv_sec; |
| 2054 | nfs_down(rep, rep->r_nmp, rep->r_procp, "not responding", |
| 2055 | 0, NFSSTA_TIMEO); |
| 2056 | if (!(nmp->nm_state & NFSSTA_MOUNTED)) { |
| 2057 | /* we're not yet completely mounted and */ |
| 2058 | /* we can't complete an RPC, so we fail */ |
| 2059 | nfsstats.rpctimeouts++; |
| 2060 | nfs_softterm(rep); |
| 2061 | continue; |
| 2062 | } |
| 2063 | } |
| 2064 | if (rep->r_rtt >= 0) { |
| 2065 | rep->r_rtt++; |
| 2066 | if (nmp->nm_flag & NFSMNT_DUMBTIMR) |
| 2067 | timeo = nmp->nm_timeo; |
| 2068 | else |
| 2069 | timeo = NFS_RTO(nmp, proct[rep->r_procnum]); |
| 2070 | /* ensure 62.5 ms floor */ |
| 2071 | while (16 * timeo < hz) |
| 2072 | timeo *= 2; |
| 2073 | if (nmp->nm_timeouts > 0) |
| 2074 | timeo *= nfs_backoff[nmp->nm_timeouts - 1]; |
| 2075 | if (rep->r_rtt <= timeo) |
| 2076 | continue; |
| 2077 | if (nmp->nm_timeouts < 8) |
| 2078 | nmp->nm_timeouts++; |
| 2079 | } |
| 2080 | /* |
| 2081 | * Check for too many retransmits. This is never true for |
| 2082 | * 'hard' mounts because we set r_retry to NFS_MAXREXMIT + 1 |
| 2083 | * and never allow r_rexmit to be more than NFS_MAXREXMIT. |
| 2084 | */ |
| 2085 | if (rep->r_rexmit >= rep->r_retry) { /* too many */ |
| 2086 | nfsstats.rpctimeouts++; |
| 2087 | nfs_softterm(rep); |
| 2088 | continue; |
| 2089 | } |
| 2090 | if (nmp->nm_sotype != SOCK_DGRAM) { |
| 2091 | if (++rep->r_rexmit > NFS_MAXREXMIT) |
| 2092 | rep->r_rexmit = NFS_MAXREXMIT; |
| 2093 | continue; |
| 2094 | } |
| 2095 | if ((so = nmp->nm_so) == NULL) |
| 2096 | continue; |
| 2097 | |
| 2098 | /* |
| 2099 | * If there is enough space and the window allows.. |
| 2100 | * Resend it |
| 2101 | * Set r_rtt to -1 in case we fail to send it now. |
| 2102 | */ |
| 2103 | #if NFSDIAG |
| 2104 | rttdiag = rep->r_rtt; |
| 2105 | #endif |
| 2106 | rep->r_rtt = -1; |
| 2107 | if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && |
| 2108 | ((nmp->nm_flag & NFSMNT_DUMBTIMR) || |
| 2109 | (rep->r_flags & R_SENT) || |
| 2110 | nmp->nm_sent < nmp->nm_cwnd) && |
| 2111 | (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ |
| 2112 | |
| 2113 | struct proc *p = current_proc(); |
| 2114 | |
| 2115 | #if NFSDIAG |
| 2116 | if (rep->r_flags & R_SENT && nfsprnttimo && |
| 2117 | nmp->nm_timeouts >= nfsprnttimo) { |
| 2118 | int t = proct[rep->r_procnum]; |
| 2119 | if (t) |
| 2120 | NFS_DPF(DUP, ("nfs_timer %s nmtm=%d tms=%d rtt=%d tm=%d p=%d A=%d D=%d\n", nmp->nm_mountp->mnt_stat.f_mntfromname, nmp->nm_timeo, nmp->nm_timeouts, rttdiag, timeo, rep->r_procnum, nmp->nm_srtt[t-1], nmp->nm_sdrtt[t-1])); |
| 2121 | else |
| 2122 | NFS_DPF(DUP, ("nfs_timer %s nmtm=%d tms=%d rtt=%d tm=%d p=%d\n", nmp->nm_mountp->mnt_stat.f_mntfromname, nmp->nm_timeo, nmp->nm_timeouts, rttdiag, timeo, rep->r_procnum)); |
| 2123 | } |
| 2124 | nfsdup(rep); |
| 2125 | #endif /* NFSDIAG */ |
| 2126 | /* |
| 2127 | * Iff first send, start timing |
| 2128 | * else turn timing off, backoff timer |
| 2129 | * and divide congestion window by 2. |
| 2130 | * We update these *before* the send to avoid |
| 2131 | * racing against receiving the reply. |
| 2132 | * We save them so we can restore them on send error. |
| 2133 | */ |
| 2134 | flags = rep->r_flags; |
| 2135 | rexmit = rep->r_rexmit; |
| 2136 | cwnd = nmp->nm_cwnd; |
| 2137 | sent = nmp->nm_sent; |
| 2138 | xid = rep->r_xid; |
| 2139 | if (rep->r_flags & R_SENT) { |
| 2140 | rep->r_flags &= ~R_TIMING; |
| 2141 | if (++rep->r_rexmit > NFS_MAXREXMIT) |
| 2142 | rep->r_rexmit = NFS_MAXREXMIT; |
| 2143 | nmp->nm_cwnd >>= 1; |
| 2144 | if (nmp->nm_cwnd < NFS_CWNDSCALE) |
| 2145 | nmp->nm_cwnd = NFS_CWNDSCALE; |
| 2146 | nfsstats.rpcretries++; |
| 2147 | } else { |
| 2148 | rep->r_flags |= R_SENT; |
| 2149 | nmp->nm_sent += NFS_CWNDSCALE; |
| 2150 | } |
| 2151 | FSDBG(535, xid, rep, nmp->nm_sent, nmp->nm_cwnd); |
| 2152 | |
| 2153 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 2154 | |
| 2155 | if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) |
| 2156 | error = (*so->so_proto->pr_usrreqs->pru_send) |
| 2157 | (so, 0, m, 0, 0, p); |
| 2158 | else |
| 2159 | error = (*so->so_proto->pr_usrreqs->pru_send) |
| 2160 | (so, 0, m, mtod(nmp->nm_nam, struct sockaddr *), 0, p); |
| 2161 | |
| 2162 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 2163 | |
| 2164 | FSDBG(535, xid, error, sent, cwnd); |
| 2165 | |
| 2166 | if (error) { |
| 2167 | if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) |
| 2168 | so->so_error = 0; |
| 2169 | rep->r_flags = flags | R_RESENDERR; |
| 2170 | rep->r_rexmit = rexmit; |
| 2171 | nmp->nm_cwnd = cwnd; |
| 2172 | nmp->nm_sent = sent; |
| 2173 | if (flags & R_SENT) |
| 2174 | nfsstats.rpcretries--; |
| 2175 | } else |
| 2176 | rep->r_rtt = 0; |
| 2177 | } |
| 2178 | } |
| 2179 | microuptime(&now); |
| 2180 | #ifndef NFS_NOSERVER |
| 2181 | /* |
| 2182 | * Call the nqnfs server timer once a second to handle leases. |
| 2183 | */ |
| 2184 | if (lasttime != now.tv_sec) { |
| 2185 | lasttime = now.tv_sec; |
| 2186 | nqnfs_serverd(); |
| 2187 | } |
| 2188 | |
| 2189 | /* |
| 2190 | * Scan the write gathering queues for writes that need to be |
| 2191 | * completed now. |
| 2192 | */ |
| 2193 | cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; |
| 2194 | TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { |
| 2195 | if (LIST_FIRST(&slp->ns_tq) && |
| 2196 | LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) |
| 2197 | nfsrv_wakenfsd(slp); |
| 2198 | } |
| 2199 | #endif /* NFS_NOSERVER */ |
| 2200 | splx(s); |
| 2201 | |
| 2202 | if (nfsbuffreeuptimestamp + 30 <= now.tv_sec) { |
| 2203 | /* |
| 2204 | * We haven't called nfs_buf_freeup() in a little while. |
| 2205 | * So, see if we can free up any stale/unused bufs now. |
| 2206 | */ |
| 2207 | nfs_buf_freeup(1); |
| 2208 | } |
| 2209 | |
| 2210 | timeout(nfs_timer_funnel, (void *)0, nfs_ticks); |
| 2211 | |
| 2212 | } |
| 2213 | |
| 2214 | |
| 2215 | /* |
| 2216 | * Test for a termination condition pending on the process. |
| 2217 | * This is used to determine if we need to bail on a mount. |
| 2218 | * EIO is returned if there has been a soft timeout. |
| 2219 | * EINTR is returned if there is a signal pending that is not being ignored |
| 2220 | * and the mount is interruptable, or if we are a thread that is in the process |
| 2221 | * of cancellation (also SIGKILL posted). |
| 2222 | */ |
| 2223 | int |
| 2224 | nfs_sigintr(nmp, rep, p) |
| 2225 | struct nfsmount *nmp; |
| 2226 | struct nfsreq *rep; |
| 2227 | struct proc *p; |
| 2228 | { |
| 2229 | struct uthread *curr_td; |
| 2230 | sigset_t pending_sigs; |
| 2231 | int context_good = 0; |
| 2232 | struct nfsmount *repnmp; |
| 2233 | |
| 2234 | if (nmp == NULL) |
| 2235 | return (ENXIO); |
| 2236 | if (rep != NULL) { |
| 2237 | repnmp = rep->r_nmp; |
| 2238 | /* we've had a forced unmount. */ |
| 2239 | if (repnmp == NULL) |
| 2240 | return (ENXIO); |
| 2241 | /* request has timed out on a 'soft' mount. */ |
| 2242 | if (rep->r_flags & R_SOFTTERM) |
| 2243 | return (EIO); |
| 2244 | /* |
| 2245 | * We're in the progress of a force unmount and there's |
| 2246 | * been a timeout we're dead and fail IO. |
| 2247 | */ |
| 2248 | if ((repnmp->nm_state & (NFSSTA_FORCE|NFSSTA_TIMEO)) == |
| 2249 | (NFSSTA_FORCE|NFSSTA_TIMEO)) |
| 2250 | return (EIO); |
| 2251 | /* Someone is unmounting us, go soft and mark it. */ |
| 2252 | if ((repnmp->nm_mountp->mnt_kern_flag & MNTK_FRCUNMOUNT)) { |
| 2253 | repnmp->nm_flag |= NFSMNT_SOFT; |
| 2254 | nmp->nm_state |= NFSSTA_FORCE; |
| 2255 | } |
| 2256 | /* |
| 2257 | * If the mount is hung and we've requested not to hang |
| 2258 | * on remote filesystems, then bail now. |
| 2259 | */ |
| 2260 | if (p != NULL && (p->p_flag & P_NOREMOTEHANG) != 0 && |
| 2261 | (repnmp->nm_state & NFSSTA_TIMEO) != 0) |
| 2262 | return (EIO); |
| 2263 | } |
| 2264 | /* XXX: is this valid? this probably should be an assertion. */ |
| 2265 | if (p == NULL) |
| 2266 | return (0); |
| 2267 | |
| 2268 | /* |
| 2269 | * XXX: Since nfs doesn't have a good shot at getting the current |
| 2270 | * thread we take a guess. (only struct proc * are passed to VOPs) |
| 2271 | * What we do is look at the current thread, if it belongs to the |
| 2272 | * passed in proc pointer then we have a "good/accurate" context |
| 2273 | * and can make an accurate guess as to what to do. |
| 2274 | * However if we have a bad context we have to make due with what |
| 2275 | * is in the proc struct which may not be as up to date as we'd |
| 2276 | * like. |
| 2277 | * This is ok because the process will call us with the correct |
| 2278 | * context after a short timeout while waiting for a response. |
| 2279 | */ |
| 2280 | curr_td = (struct uthread *)get_bsdthread_info(current_act()); |
| 2281 | if (curr_td->uu_proc == p) |
| 2282 | context_good = 1; |
| 2283 | if (context_good && current_thread_aborted()) |
| 2284 | return (EINTR); |
| 2285 | /* mask off thread and process blocked signals. */ |
| 2286 | if (context_good) |
| 2287 | pending_sigs = curr_td->uu_siglist & ~curr_td->uu_sigmask; |
| 2288 | else |
| 2289 | pending_sigs = p->p_siglist; |
| 2290 | /* mask off process level and NFS ignored signals. */ |
| 2291 | pending_sigs &= ~p->p_sigignore & NFSINT_SIGMASK; |
| 2292 | if (pending_sigs && (nmp->nm_flag & NFSMNT_INT) != 0) |
| 2293 | return (EINTR); |
| 2294 | return (0); |
| 2295 | } |
| 2296 | |
| 2297 | /* |
| 2298 | * Lock a socket against others. |
| 2299 | * Necessary for STREAM sockets to ensure you get an entire rpc request/reply |
| 2300 | * and also to avoid race conditions between the processes with nfs requests |
| 2301 | * in progress when a reconnect is necessary. |
| 2302 | */ |
| 2303 | int |
| 2304 | nfs_sndlock(rep) |
| 2305 | struct nfsreq *rep; |
| 2306 | { |
| 2307 | register int *statep; |
| 2308 | struct proc *p; |
| 2309 | int error, slpflag = 0, slptimeo = 0; |
| 2310 | |
| 2311 | if (rep->r_nmp == NULL) |
| 2312 | return (ENXIO); |
| 2313 | statep = &rep->r_nmp->nm_state; |
| 2314 | |
| 2315 | p = rep->r_procp; |
| 2316 | if (rep->r_nmp->nm_flag & NFSMNT_INT) |
| 2317 | slpflag = PCATCH; |
| 2318 | while (*statep & NFSSTA_SNDLOCK) { |
| 2319 | error = nfs_sigintr(rep->r_nmp, rep, p); |
| 2320 | if (error) |
| 2321 | return (error); |
| 2322 | *statep |= NFSSTA_WANTSND; |
| 2323 | if (p != NULL && (p->p_flag & P_NOREMOTEHANG) != 0) |
| 2324 | slptimeo = hz; |
| 2325 | (void) tsleep((caddr_t)statep, slpflag | (PZERO - 1), |
| 2326 | "nfsndlck", slptimeo); |
| 2327 | if (slpflag == PCATCH) { |
| 2328 | slpflag = 0; |
| 2329 | slptimeo = 2 * hz; |
| 2330 | } |
| 2331 | /* |
| 2332 | * Make sure while we slept that the mountpoint didn't go away. |
| 2333 | * nfs_sigintr and callers expect it in tact. |
| 2334 | */ |
| 2335 | if (!rep->r_nmp) |
| 2336 | return (ENXIO); /* don't have lock until out of loop */ |
| 2337 | } |
| 2338 | *statep |= NFSSTA_SNDLOCK; |
| 2339 | return (0); |
| 2340 | } |
| 2341 | |
| 2342 | /* |
| 2343 | * Unlock the stream socket for others. |
| 2344 | */ |
| 2345 | void |
| 2346 | nfs_sndunlock(rep) |
| 2347 | struct nfsreq *rep; |
| 2348 | { |
| 2349 | register int *statep; |
| 2350 | |
| 2351 | if (rep->r_nmp == NULL) |
| 2352 | return; |
| 2353 | statep = &rep->r_nmp->nm_state; |
| 2354 | if ((*statep & NFSSTA_SNDLOCK) == 0) |
| 2355 | panic("nfs sndunlock"); |
| 2356 | *statep &= ~NFSSTA_SNDLOCK; |
| 2357 | if (*statep & NFSSTA_WANTSND) { |
| 2358 | *statep &= ~NFSSTA_WANTSND; |
| 2359 | wakeup((caddr_t)statep); |
| 2360 | } |
| 2361 | } |
| 2362 | |
| 2363 | static int |
| 2364 | nfs_rcvlock(rep) |
| 2365 | register struct nfsreq *rep; |
| 2366 | { |
| 2367 | register int *statep; |
| 2368 | int error, slpflag, slptimeo = 0; |
| 2369 | |
| 2370 | /* make sure we still have our mountpoint */ |
| 2371 | if (!rep->r_nmp) { |
| 2372 | if (rep->r_mrep != NULL) |
| 2373 | return (EALREADY); |
| 2374 | return (ENXIO); |
| 2375 | } |
| 2376 | |
| 2377 | statep = &rep->r_nmp->nm_state; |
| 2378 | FSDBG_TOP(534, rep->r_xid, rep, rep->r_nmp, *statep); |
| 2379 | if (rep->r_nmp->nm_flag & NFSMNT_INT) |
| 2380 | slpflag = PCATCH; |
| 2381 | else |
| 2382 | slpflag = 0; |
| 2383 | while (*statep & NFSSTA_RCVLOCK) { |
| 2384 | if ((error = nfs_sigintr(rep->r_nmp, rep, rep->r_procp))) { |
| 2385 | FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, 0x100); |
| 2386 | return (error); |
| 2387 | } else if (rep->r_mrep != NULL) { |
| 2388 | /* |
| 2389 | * Don't bother sleeping if reply already arrived |
| 2390 | */ |
| 2391 | FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, 0x101); |
| 2392 | return (EALREADY); |
| 2393 | } |
| 2394 | FSDBG(534, rep->r_xid, rep, rep->r_nmp, 0x102); |
| 2395 | *statep |= NFSSTA_WANTRCV; |
| 2396 | /* |
| 2397 | * We need to poll if we're P_NOREMOTEHANG so that we |
| 2398 | * call nfs_sigintr periodically above. |
| 2399 | */ |
| 2400 | if (rep->r_procp != NULL && |
| 2401 | (rep->r_procp->p_flag & P_NOREMOTEHANG) != 0) |
| 2402 | slptimeo = hz; |
| 2403 | (void) tsleep((caddr_t)statep, slpflag | (PZERO - 1), |
| 2404 | "nfsrcvlk", slptimeo); |
| 2405 | if (slpflag == PCATCH) { |
| 2406 | slpflag = 0; |
| 2407 | slptimeo = 2 * hz; |
| 2408 | } |
| 2409 | /* |
| 2410 | * Make sure while we slept that the mountpoint didn't go away. |
| 2411 | * nfs_sigintr and caller nfs_reply expect it intact. |
| 2412 | */ |
| 2413 | if (!rep->r_nmp) { |
| 2414 | FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, 0x103); |
| 2415 | return (ENXIO); /* don't have lock until out of loop */ |
| 2416 | } |
| 2417 | } |
| 2418 | /* |
| 2419 | * nfs_reply will handle it if reply already arrived. |
| 2420 | * (We may have slept or been preempted while on network funnel). |
| 2421 | */ |
| 2422 | FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, *statep); |
| 2423 | *statep |= NFSSTA_RCVLOCK; |
| 2424 | return (0); |
| 2425 | } |
| 2426 | |
| 2427 | /* |
| 2428 | * Unlock the stream socket for others. |
| 2429 | */ |
| 2430 | static void |
| 2431 | nfs_rcvunlock(rep) |
| 2432 | register struct nfsreq *rep; |
| 2433 | { |
| 2434 | register int *statep; |
| 2435 | |
| 2436 | if (rep->r_nmp == NULL) |
| 2437 | return; |
| 2438 | statep = &rep->r_nmp->nm_state; |
| 2439 | |
| 2440 | FSDBG(533, statep, *statep, 0, 0); |
| 2441 | if ((*statep & NFSSTA_RCVLOCK) == 0) |
| 2442 | panic("nfs rcvunlock"); |
| 2443 | *statep &= ~NFSSTA_RCVLOCK; |
| 2444 | if (*statep & NFSSTA_WANTRCV) { |
| 2445 | *statep &= ~NFSSTA_WANTRCV; |
| 2446 | wakeup((caddr_t)statep); |
| 2447 | } |
| 2448 | } |
| 2449 | |
| 2450 | |
| 2451 | #ifndef NFS_NOSERVER |
| 2452 | /* |
| 2453 | * Socket upcall routine for the nfsd sockets. |
| 2454 | * The caddr_t arg is a pointer to the "struct nfssvc_sock". |
| 2455 | * Essentially do as much as possible non-blocking, else punt and it will |
| 2456 | * be called with M_WAIT from an nfsd. |
| 2457 | */ |
| 2458 | /* |
| 2459 | * Needs to run under network funnel |
| 2460 | */ |
| 2461 | void |
| 2462 | nfsrv_rcv(so, arg, waitflag) |
| 2463 | struct socket *so; |
| 2464 | caddr_t arg; |
| 2465 | int waitflag; |
| 2466 | { |
| 2467 | register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; |
| 2468 | register struct mbuf *m; |
| 2469 | struct mbuf *mp, *mhck; |
| 2470 | struct sockaddr *nam; |
| 2471 | struct uio auio; |
| 2472 | int flags, ns_nflag=0, error; |
| 2473 | struct sockaddr_in *sin; |
| 2474 | |
| 2475 | if ((slp->ns_flag & SLP_VALID) == 0) |
| 2476 | return; |
| 2477 | #ifdef notdef |
| 2478 | /* |
| 2479 | * Define this to test for nfsds handling this under heavy load. |
| 2480 | */ |
| 2481 | if (waitflag == M_DONTWAIT) { |
| 2482 | ns_nflag = SLPN_NEEDQ; |
| 2483 | goto dorecs; |
| 2484 | } |
| 2485 | #endif |
| 2486 | auio.uio_procp = NULL; |
| 2487 | if (so->so_type == SOCK_STREAM) { |
| 2488 | /* |
| 2489 | * If there are already records on the queue, defer soreceive() |
| 2490 | * to an nfsd so that there is feedback to the TCP layer that |
| 2491 | * the nfs servers are heavily loaded. |
| 2492 | */ |
| 2493 | if (slp->ns_rec && waitflag == M_DONTWAIT) { |
| 2494 | ns_nflag = SLPN_NEEDQ; |
| 2495 | goto dorecs; |
| 2496 | } |
| 2497 | |
| 2498 | /* |
| 2499 | * Do soreceive(). |
| 2500 | */ |
| 2501 | auio.uio_resid = 1000000000; |
| 2502 | flags = MSG_DONTWAIT; |
| 2503 | error = soreceive(so, (struct sockaddr **) 0, &auio, &mp, (struct mbuf **)0, &flags); |
| 2504 | if (error || mp == (struct mbuf *)0) { |
| 2505 | if (error == EWOULDBLOCK) |
| 2506 | ns_nflag = SLPN_NEEDQ; |
| 2507 | else |
| 2508 | ns_nflag = SLPN_DISCONN; |
| 2509 | goto dorecs; |
| 2510 | } |
| 2511 | m = mp; |
| 2512 | if (slp->ns_rawend) { |
| 2513 | slp->ns_rawend->m_next = m; |
| 2514 | slp->ns_cc += 1000000000 - auio.uio_resid; |
| 2515 | } else { |
| 2516 | slp->ns_raw = m; |
| 2517 | slp->ns_cc = 1000000000 - auio.uio_resid; |
| 2518 | } |
| 2519 | while (m->m_next) |
| 2520 | m = m->m_next; |
| 2521 | slp->ns_rawend = m; |
| 2522 | |
| 2523 | /* |
| 2524 | * Now try and parse record(s) out of the raw stream data. |
| 2525 | */ |
| 2526 | error = nfsrv_getstream(slp, waitflag); |
| 2527 | if (error) { |
| 2528 | if (error == EPERM) |
| 2529 | ns_nflag = SLPN_DISCONN; |
| 2530 | else |
| 2531 | ns_nflag = SLPN_NEEDQ; |
| 2532 | } |
| 2533 | } else { |
| 2534 | do { |
| 2535 | auio.uio_resid = 1000000000; |
| 2536 | flags = MSG_DONTWAIT | MSG_NEEDSA; |
| 2537 | nam = 0; |
| 2538 | mp = 0; |
| 2539 | error = soreceive(so, &nam, &auio, &mp, |
| 2540 | (struct mbuf **)0, &flags); |
| 2541 | |
| 2542 | if (mp) { |
| 2543 | if (nam) { |
| 2544 | MGET(mhck, M_WAIT, MT_SONAME); |
| 2545 | mhck->m_len = nam->sa_len; |
| 2546 | sin = mtod(mhck, struct sockaddr_in *); |
| 2547 | bcopy(nam, sin, sizeof(struct sockaddr_in)); |
| 2548 | mhck->m_hdr.mh_len = sizeof(struct sockaddr_in); |
| 2549 | |
| 2550 | m = mhck; |
| 2551 | m->m_next = mp; |
| 2552 | } else |
| 2553 | m = mp; |
| 2554 | if (slp->ns_recend) |
| 2555 | slp->ns_recend->m_nextpkt = m; |
| 2556 | else |
| 2557 | slp->ns_rec = m; |
| 2558 | slp->ns_recend = m; |
| 2559 | m->m_nextpkt = (struct mbuf *)0; |
| 2560 | } |
| 2561 | if (nam) { |
| 2562 | FREE(nam, M_SONAME); |
| 2563 | } |
| 2564 | if (error) { |
| 2565 | if ((so->so_proto->pr_flags & PR_CONNREQUIRED) |
| 2566 | && error != EWOULDBLOCK) { |
| 2567 | ns_nflag = SLPN_DISCONN; |
| 2568 | goto dorecs; |
| 2569 | } |
| 2570 | } |
| 2571 | } while (mp); |
| 2572 | } |
| 2573 | |
| 2574 | /* |
| 2575 | * Now try and process the request records, non-blocking. |
| 2576 | */ |
| 2577 | dorecs: |
| 2578 | if (ns_nflag) |
| 2579 | slp->ns_nflag |= ns_nflag; |
| 2580 | if (waitflag == M_DONTWAIT && |
| 2581 | (slp->ns_rec || (slp->ns_nflag & (SLPN_NEEDQ | SLPN_DISCONN)))) { |
| 2582 | thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL); |
| 2583 | nfsrv_wakenfsd(slp); |
| 2584 | thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL); |
| 2585 | } |
| 2586 | } |
| 2587 | |
| 2588 | /* |
| 2589 | * Try and extract an RPC request from the mbuf data list received on a |
| 2590 | * stream socket. The "waitflag" argument indicates whether or not it |
| 2591 | * can sleep. |
| 2592 | */ |
| 2593 | static int |
| 2594 | nfsrv_getstream(slp, waitflag) |
| 2595 | register struct nfssvc_sock *slp; |
| 2596 | int waitflag; |
| 2597 | { |
| 2598 | register struct mbuf *m, **mpp; |
| 2599 | register char *cp1, *cp2; |
| 2600 | register int len; |
| 2601 | struct mbuf *om, *m2, *recm; |
| 2602 | u_long recmark; |
| 2603 | |
| 2604 | if (slp->ns_nflag & SLPN_GETSTREAM) |
| 2605 | panic("nfs getstream"); |
| 2606 | slp->ns_nflag |= SLPN_GETSTREAM; |
| 2607 | for (;;) { |
| 2608 | if (slp->ns_reclen == 0) { |
| 2609 | if (slp->ns_cc < NFSX_UNSIGNED) { |
| 2610 | slp->ns_nflag &= ~SLPN_GETSTREAM; |
| 2611 | return (0); |
| 2612 | } |
| 2613 | m = slp->ns_raw; |
| 2614 | if (m->m_len >= NFSX_UNSIGNED) { |
| 2615 | bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); |
| 2616 | m->m_data += NFSX_UNSIGNED; |
| 2617 | m->m_len -= NFSX_UNSIGNED; |
| 2618 | } else { |
| 2619 | cp1 = (caddr_t)&recmark; |
| 2620 | cp2 = mtod(m, caddr_t); |
| 2621 | while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { |
| 2622 | while (m->m_len == 0) { |
| 2623 | m = m->m_next; |
| 2624 | cp2 = mtod(m, caddr_t); |
| 2625 | } |
| 2626 | *cp1++ = *cp2++; |
| 2627 | m->m_data++; |
| 2628 | m->m_len--; |
| 2629 | } |
| 2630 | } |
| 2631 | slp->ns_cc -= NFSX_UNSIGNED; |
| 2632 | recmark = ntohl(recmark); |
| 2633 | slp->ns_reclen = recmark & ~0x80000000; |
| 2634 | if (recmark & 0x80000000) |
| 2635 | slp->ns_nflag |= SLPN_LASTFRAG; |
| 2636 | else |
| 2637 | slp->ns_nflag &= ~SLPN_LASTFRAG; |
| 2638 | if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { |
| 2639 | slp->ns_nflag &= ~SLPN_GETSTREAM; |
| 2640 | return (EPERM); |
| 2641 | } |
| 2642 | } |
| 2643 | |
| 2644 | /* |
| 2645 | * Now get the record part. |
| 2646 | * |
| 2647 | * Note that slp->ns_reclen may be 0. Linux sometimes |
| 2648 | * generates 0-length RPCs |
| 2649 | */ |
| 2650 | recm = NULL; |
| 2651 | if (slp->ns_cc == slp->ns_reclen) { |
| 2652 | recm = slp->ns_raw; |
| 2653 | slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; |
| 2654 | slp->ns_cc = slp->ns_reclen = 0; |
| 2655 | } else if (slp->ns_cc > slp->ns_reclen) { |
| 2656 | len = 0; |
| 2657 | m = slp->ns_raw; |
| 2658 | om = (struct mbuf *)0; |
| 2659 | while (len < slp->ns_reclen) { |
| 2660 | if ((len + m->m_len) > slp->ns_reclen) { |
| 2661 | m2 = m_copym(m, 0, slp->ns_reclen - len, |
| 2662 | waitflag); |
| 2663 | if (m2) { |
| 2664 | if (om) { |
| 2665 | om->m_next = m2; |
| 2666 | recm = slp->ns_raw; |
| 2667 | } else |
| 2668 | recm = m2; |
| 2669 | m->m_data += slp->ns_reclen - len; |
| 2670 | m->m_len -= slp->ns_reclen - len; |
| 2671 | len = slp->ns_reclen; |
| 2672 | } else { |
| 2673 | slp->ns_nflag &= ~SLPN_GETSTREAM; |
| 2674 | return (EWOULDBLOCK); |
| 2675 | } |
| 2676 | } else if ((len + m->m_len) == slp->ns_reclen) { |
| 2677 | om = m; |
| 2678 | len += m->m_len; |
| 2679 | m = m->m_next; |
| 2680 | recm = slp->ns_raw; |
| 2681 | om->m_next = (struct mbuf *)0; |
| 2682 | } else { |
| 2683 | om = m; |
| 2684 | len += m->m_len; |
| 2685 | m = m->m_next; |
| 2686 | } |
| 2687 | } |
| 2688 | slp->ns_raw = m; |
| 2689 | slp->ns_cc -= len; |
| 2690 | slp->ns_reclen = 0; |
| 2691 | } else { |
| 2692 | slp->ns_nflag &= ~SLPN_GETSTREAM; |
| 2693 | return (0); |
| 2694 | } |
| 2695 | |
| 2696 | /* |
| 2697 | * Accumulate the fragments into a record. |
| 2698 | */ |
| 2699 | mpp = &slp->ns_frag; |
| 2700 | while (*mpp) |
| 2701 | mpp = &((*mpp)->m_next); |
| 2702 | *mpp = recm; |
| 2703 | if (slp->ns_nflag & SLPN_LASTFRAG) { |
| 2704 | if (slp->ns_recend) |
| 2705 | slp->ns_recend->m_nextpkt = slp->ns_frag; |
| 2706 | else |
| 2707 | slp->ns_rec = slp->ns_frag; |
| 2708 | slp->ns_recend = slp->ns_frag; |
| 2709 | slp->ns_frag = (struct mbuf *)0; |
| 2710 | } |
| 2711 | } |
| 2712 | } |
| 2713 | |
| 2714 | /* |
| 2715 | * Parse an RPC header. |
| 2716 | */ |
| 2717 | int |
| 2718 | nfsrv_dorec(slp, nfsd, ndp) |
| 2719 | register struct nfssvc_sock *slp; |
| 2720 | struct nfsd *nfsd; |
| 2721 | struct nfsrv_descript **ndp; |
| 2722 | { |
| 2723 | register struct mbuf *m; |
| 2724 | register struct mbuf *nam; |
| 2725 | register struct nfsrv_descript *nd; |
| 2726 | int error; |
| 2727 | |
| 2728 | *ndp = NULL; |
| 2729 | if ((slp->ns_flag & SLP_VALID) == 0 || |
| 2730 | (m = slp->ns_rec) == (struct mbuf *)0) |
| 2731 | return (ENOBUFS); |
| 2732 | slp->ns_rec = m->m_nextpkt; |
| 2733 | if (slp->ns_rec) |
| 2734 | m->m_nextpkt = (struct mbuf *)0; |
| 2735 | else |
| 2736 | slp->ns_recend = (struct mbuf *)0; |
| 2737 | if (m->m_type == MT_SONAME) { |
| 2738 | nam = m; |
| 2739 | m = m->m_next; |
| 2740 | nam->m_next = NULL; |
| 2741 | } else |
| 2742 | nam = NULL; |
| 2743 | MALLOC_ZONE(nd, struct nfsrv_descript *, |
| 2744 | sizeof (struct nfsrv_descript), M_NFSRVDESC, M_WAITOK); |
| 2745 | nd->nd_md = nd->nd_mrep = m; |
| 2746 | nd->nd_nam2 = nam; |
| 2747 | nd->nd_dpos = mtod(m, caddr_t); |
| 2748 | error = nfs_getreq(nd, nfsd, TRUE); |
| 2749 | if (error) { |
| 2750 | if (nam) |
| 2751 | m_freem(nam); |
| 2752 | FREE_ZONE((caddr_t)nd, sizeof *nd, M_NFSRVDESC); |
| 2753 | return (error); |
| 2754 | } |
| 2755 | *ndp = nd; |
| 2756 | nfsd->nfsd_nd = nd; |
| 2757 | return (0); |
| 2758 | } |
| 2759 | |
| 2760 | /* |
| 2761 | * Parse an RPC request |
| 2762 | * - verify it |
| 2763 | * - fill in the cred struct. |
| 2764 | */ |
| 2765 | int |
| 2766 | nfs_getreq(nd, nfsd, has_header) |
| 2767 | register struct nfsrv_descript *nd; |
| 2768 | struct nfsd *nfsd; |
| 2769 | int has_header; |
| 2770 | { |
| 2771 | register int len, i; |
| 2772 | register u_long *tl; |
| 2773 | register long t1; |
| 2774 | struct uio uio; |
| 2775 | struct iovec iov; |
| 2776 | caddr_t dpos, cp2, cp; |
| 2777 | u_long nfsvers, auth_type; |
| 2778 | uid_t nickuid; |
| 2779 | int error = 0, nqnfs = 0, ticklen; |
| 2780 | struct mbuf *mrep, *md; |
| 2781 | register struct nfsuid *nuidp; |
| 2782 | struct timeval tvin, tvout, now; |
| 2783 | #if 0 /* until encrypted keys are implemented */ |
| 2784 | NFSKERBKEYSCHED_T keys; /* stores key schedule */ |
| 2785 | #endif |
| 2786 | |
| 2787 | mrep = nd->nd_mrep; |
| 2788 | md = nd->nd_md; |
| 2789 | dpos = nd->nd_dpos; |
| 2790 | if (has_header) { |
| 2791 | nfsm_dissect(tl, u_long *, 10 * NFSX_UNSIGNED); |
| 2792 | nd->nd_retxid = fxdr_unsigned(u_long, *tl++); |
| 2793 | if (*tl++ != rpc_call) { |
| 2794 | m_freem(mrep); |
| 2795 | return (EBADRPC); |
| 2796 | } |
| 2797 | } else |
| 2798 | nfsm_dissect(tl, u_long *, 8 * NFSX_UNSIGNED); |
| 2799 | nd->nd_repstat = 0; |
| 2800 | nd->nd_flag = 0; |
| 2801 | if (*tl++ != rpc_vers) { |
| 2802 | nd->nd_repstat = ERPCMISMATCH; |
| 2803 | nd->nd_procnum = NFSPROC_NOOP; |
| 2804 | return (0); |
| 2805 | } |
| 2806 | if (*tl != nfs_prog) { |
| 2807 | if (*tl == nqnfs_prog) |
| 2808 | nqnfs++; |
| 2809 | else { |
| 2810 | nd->nd_repstat = EPROGUNAVAIL; |
| 2811 | nd->nd_procnum = NFSPROC_NOOP; |
| 2812 | return (0); |
| 2813 | } |
| 2814 | } |
| 2815 | tl++; |
| 2816 | nfsvers = fxdr_unsigned(u_long, *tl++); |
| 2817 | if (((nfsvers < NFS_VER2 || nfsvers > NFS_VER3) && !nqnfs) || |
| 2818 | (nfsvers != NQNFS_VER3 && nqnfs)) { |
| 2819 | nd->nd_repstat = EPROGMISMATCH; |
| 2820 | nd->nd_procnum = NFSPROC_NOOP; |
| 2821 | return (0); |
| 2822 | } |
| 2823 | if (nqnfs) |
| 2824 | nd->nd_flag = (ND_NFSV3 | ND_NQNFS); |
| 2825 | else if (nfsvers == NFS_VER3) |
| 2826 | nd->nd_flag = ND_NFSV3; |
| 2827 | nd->nd_procnum = fxdr_unsigned(u_long, *tl++); |
| 2828 | if (nd->nd_procnum == NFSPROC_NULL) |
| 2829 | return (0); |
| 2830 | if (nd->nd_procnum >= NFS_NPROCS || |
| 2831 | (!nqnfs && nd->nd_procnum >= NQNFSPROC_GETLEASE) || |
| 2832 | (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { |
| 2833 | nd->nd_repstat = EPROCUNAVAIL; |
| 2834 | nd->nd_procnum = NFSPROC_NOOP; |
| 2835 | return (0); |
| 2836 | } |
| 2837 | if ((nd->nd_flag & ND_NFSV3) == 0) |
| 2838 | nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; |
| 2839 | auth_type = *tl++; |
| 2840 | len = fxdr_unsigned(int, *tl++); |
| 2841 | if (len < 0 || len > RPCAUTH_MAXSIZ) { |
| 2842 | m_freem(mrep); |
| 2843 | return (EBADRPC); |
| 2844 | } |
| 2845 | |
| 2846 | nd->nd_flag &= ~ND_KERBAUTH; |
| 2847 | /* |
| 2848 | * Handle auth_unix or auth_kerb. |
| 2849 | */ |
| 2850 | if (auth_type == rpc_auth_unix) { |
| 2851 | len = fxdr_unsigned(int, *++tl); |
| 2852 | if (len < 0 || len > NFS_MAXNAMLEN) { |
| 2853 | m_freem(mrep); |
| 2854 | return (EBADRPC); |
| 2855 | } |
| 2856 | nfsm_adv(nfsm_rndup(len)); |
| 2857 | nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); |
| 2858 | bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); |
| 2859 | nd->nd_cr.cr_ref = 1; |
| 2860 | nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); |
| 2861 | nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); |
| 2862 | len = fxdr_unsigned(int, *tl); |
| 2863 | if (len < 0 || len > RPCAUTH_UNIXGIDS) { |
| 2864 | m_freem(mrep); |
| 2865 | return (EBADRPC); |
| 2866 | } |
| 2867 | nfsm_dissect(tl, u_long *, (len + 2) * NFSX_UNSIGNED); |
| 2868 | for (i = 1; i <= len; i++) |
| 2869 | if (i < NGROUPS) |
| 2870 | nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); |
| 2871 | else |
| 2872 | tl++; |
| 2873 | nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); |
| 2874 | if (nd->nd_cr.cr_ngroups > 1) |
| 2875 | nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); |
| 2876 | len = fxdr_unsigned(int, *++tl); |
| 2877 | if (len < 0 || len > RPCAUTH_MAXSIZ) { |
| 2878 | m_freem(mrep); |
| 2879 | return (EBADRPC); |
| 2880 | } |
| 2881 | if (len > 0) |
| 2882 | nfsm_adv(nfsm_rndup(len)); |
| 2883 | } else if (auth_type == rpc_auth_kerb) { |
| 2884 | switch (fxdr_unsigned(int, *tl++)) { |
| 2885 | case RPCAKN_FULLNAME: |
| 2886 | ticklen = fxdr_unsigned(int, *tl); |
| 2887 | *((u_long *)nfsd->nfsd_authstr) = *tl; |
| 2888 | uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; |
| 2889 | nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; |
| 2890 | if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { |
| 2891 | m_freem(mrep); |
| 2892 | return (EBADRPC); |
| 2893 | } |
| 2894 | uio.uio_offset = 0; |
| 2895 | uio.uio_iov = &iov; |
| 2896 | uio.uio_iovcnt = 1; |
| 2897 | uio.uio_segflg = UIO_SYSSPACE; |
| 2898 | iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; |
| 2899 | iov.iov_len = RPCAUTH_MAXSIZ - 4; |
| 2900 | nfsm_mtouio(&uio, uio.uio_resid); |
| 2901 | nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); |
| 2902 | if (*tl++ != rpc_auth_kerb || |
| 2903 | fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { |
| 2904 | printf("Bad kerb verifier\n"); |
| 2905 | nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); |
| 2906 | nd->nd_procnum = NFSPROC_NOOP; |
| 2907 | return (0); |
| 2908 | } |
| 2909 | nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); |
| 2910 | tl = (u_long *)cp; |
| 2911 | if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { |
| 2912 | printf("Not fullname kerb verifier\n"); |
| 2913 | nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); |
| 2914 | nd->nd_procnum = NFSPROC_NOOP; |
| 2915 | return (0); |
| 2916 | } |
| 2917 | cp += NFSX_UNSIGNED; |
| 2918 | bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); |
| 2919 | nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; |
| 2920 | nd->nd_flag |= ND_KERBFULL; |
| 2921 | nfsd->nfsd_flag |= NFSD_NEEDAUTH; |
| 2922 | break; |
| 2923 | case RPCAKN_NICKNAME: |
| 2924 | if (len != 2 * NFSX_UNSIGNED) { |
| 2925 | printf("Kerb nickname short\n"); |
| 2926 | nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); |
| 2927 | nd->nd_procnum = NFSPROC_NOOP; |
| 2928 | return (0); |
| 2929 | } |
| 2930 | nickuid = fxdr_unsigned(uid_t, *tl); |
| 2931 | nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); |
| 2932 | if (*tl++ != rpc_auth_kerb || |
| 2933 | fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { |
| 2934 | printf("Kerb nick verifier bad\n"); |
| 2935 | nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); |
| 2936 | nd->nd_procnum = NFSPROC_NOOP; |
| 2937 | return (0); |
| 2938 | } |
| 2939 | nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); |
| 2940 | tvin.tv_sec = *tl++; |
| 2941 | tvin.tv_usec = *tl; |
| 2942 | |
| 2943 | for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; |
| 2944 | nuidp != 0; nuidp = nuidp->nu_hash.le_next) { |
| 2945 | if (nuidp->nu_cr.cr_uid == nickuid && |
| 2946 | (!nd->nd_nam2 || |
| 2947 | netaddr_match(NU_NETFAM(nuidp), |
| 2948 | &nuidp->nu_haddr, nd->nd_nam2))) |
| 2949 | break; |
| 2950 | } |
| 2951 | if (!nuidp) { |
| 2952 | nd->nd_repstat = |
| 2953 | (NFSERR_AUTHERR|AUTH_REJECTCRED); |
| 2954 | nd->nd_procnum = NFSPROC_NOOP; |
| 2955 | return (0); |
| 2956 | } |
| 2957 | |
| 2958 | /* |
| 2959 | * Now, decrypt the timestamp using the session key |
| 2960 | * and validate it. |
| 2961 | */ |
| 2962 | #if NFSKERB |
| 2963 | XXX |
| 2964 | #endif |
| 2965 | |
| 2966 | tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); |
| 2967 | tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); |
| 2968 | microtime(&now); |
| 2969 | if (nuidp->nu_expire < now.tv_sec || |
| 2970 | nuidp->nu_timestamp.tv_sec > tvout.tv_sec || |
| 2971 | (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && |
| 2972 | nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { |
| 2973 | nuidp->nu_expire = 0; |
| 2974 | nd->nd_repstat = |
| 2975 | (NFSERR_AUTHERR|AUTH_REJECTVERF); |
| 2976 | nd->nd_procnum = NFSPROC_NOOP; |
| 2977 | return (0); |
| 2978 | } |
| 2979 | nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); |
| 2980 | nd->nd_flag |= ND_KERBNICK; |
| 2981 | }; |
| 2982 | } else { |
| 2983 | nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); |
| 2984 | nd->nd_procnum = NFSPROC_NOOP; |
| 2985 | return (0); |
| 2986 | } |
| 2987 | |
| 2988 | /* |
| 2989 | * For nqnfs, get piggybacked lease request. |
| 2990 | */ |
| 2991 | if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) { |
| 2992 | nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); |
| 2993 | nd->nd_flag |= fxdr_unsigned(int, *tl); |
| 2994 | if (nd->nd_flag & ND_LEASE) { |
| 2995 | nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); |
| 2996 | nd->nd_duration = fxdr_unsigned(int, *tl); |
| 2997 | } else |
| 2998 | nd->nd_duration = NQ_MINLEASE; |
| 2999 | } else |
| 3000 | nd->nd_duration = NQ_MINLEASE; |
| 3001 | nd->nd_md = md; |
| 3002 | nd->nd_dpos = dpos; |
| 3003 | return (0); |
| 3004 | nfsmout: |
| 3005 | return (error); |
| 3006 | } |
| 3007 | |
| 3008 | /* |
| 3009 | * Search for a sleeping nfsd and wake it up. |
| 3010 | * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the |
| 3011 | * running nfsds will go look for the work in the nfssvc_sock list. |
| 3012 | */ |
| 3013 | void |
| 3014 | nfsrv_wakenfsd(slp) |
| 3015 | struct nfssvc_sock *slp; |
| 3016 | { |
| 3017 | register struct nfsd *nd; |
| 3018 | |
| 3019 | if ((slp->ns_flag & SLP_VALID) == 0) |
| 3020 | return; |
| 3021 | TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) { |
| 3022 | if (nd->nfsd_flag & NFSD_WAITING) { |
| 3023 | nd->nfsd_flag &= ~NFSD_WAITING; |
| 3024 | if (nd->nfsd_slp) |
| 3025 | panic("nfsd wakeup"); |
| 3026 | slp->ns_sref++; |
| 3027 | nd->nfsd_slp = slp; |
| 3028 | wakeup((caddr_t)nd); |
| 3029 | return; |
| 3030 | } |
| 3031 | } |
| 3032 | slp->ns_flag |= SLP_DOREC; |
| 3033 | nfsd_head_flag |= NFSD_CHECKSLP; |
| 3034 | } |
| 3035 | #endif /* NFS_NOSERVER */ |
| 3036 | |
| 3037 | static int |
| 3038 | nfs_msg(p, server, msg, error) |
| 3039 | struct proc *p; |
| 3040 | const char *server, *msg; |
| 3041 | int error; |
| 3042 | { |
| 3043 | tpr_t tpr; |
| 3044 | |
| 3045 | if (p) |
| 3046 | tpr = tprintf_open(p); |
| 3047 | else |
| 3048 | tpr = NULL; |
| 3049 | if (error) |
| 3050 | tprintf(tpr, "nfs server %s: %s, error %d\n", server, msg, |
| 3051 | error); |
| 3052 | else |
| 3053 | tprintf(tpr, "nfs server %s: %s\n", server, msg); |
| 3054 | tprintf_close(tpr); |
| 3055 | return (0); |
| 3056 | } |
| 3057 | |
| 3058 | void |
| 3059 | nfs_down(rep, nmp, proc, msg, error, flags) |
| 3060 | struct nfsreq *rep; |
| 3061 | struct nfsmount *nmp; |
| 3062 | struct proc *proc; |
| 3063 | const char *msg; |
| 3064 | int error, flags; |
| 3065 | { |
| 3066 | if (nmp == NULL) |
| 3067 | return; |
| 3068 | if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) { |
| 3069 | vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, |
| 3070 | VQ_NOTRESP, 0); |
| 3071 | nmp->nm_state |= NFSSTA_TIMEO; |
| 3072 | } |
| 3073 | if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) { |
| 3074 | vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, |
| 3075 | VQ_NOTRESPLOCK, 0); |
| 3076 | nmp->nm_state |= NFSSTA_LOCKTIMEO; |
| 3077 | } |
| 3078 | if (rep) |
| 3079 | rep->r_flags |= R_TPRINTFMSG; |
| 3080 | nfs_msg(proc, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error); |
| 3081 | } |
| 3082 | |
| 3083 | void |
| 3084 | nfs_up(rep, nmp, proc, msg, flags) |
| 3085 | struct nfsreq *rep; |
| 3086 | struct nfsmount *nmp; |
| 3087 | struct proc *proc; |
| 3088 | const char *msg; |
| 3089 | int flags; |
| 3090 | { |
| 3091 | if (nmp == NULL) |
| 3092 | return; |
| 3093 | if ((rep == NULL) || (rep->r_flags & R_TPRINTFMSG) != 0) |
| 3094 | nfs_msg(proc, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0); |
| 3095 | if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) { |
| 3096 | nmp->nm_state &= ~NFSSTA_TIMEO; |
| 3097 | vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, |
| 3098 | VQ_NOTRESP, 1); |
| 3099 | } |
| 3100 | if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) { |
| 3101 | nmp->nm_state &= ~NFSSTA_LOCKTIMEO; |
| 3102 | vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, |
| 3103 | VQ_NOTRESPLOCK, 1); |
| 3104 | } |
| 3105 | } |
| 3106 | |