nfs_clntsocket.c revision 1.1.40.2
1 /* $NetBSD: nfs_clntsocket.c,v 1.1.40.2 2016/07/10 09:42:34 martin Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1991, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 /* 38 * Socket operations for use by nfs 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: nfs_clntsocket.c,v 1.1.40.2 2016/07/10 09:42:34 martin Exp $"); 43 44 #ifdef _KERNEL_OPT 45 #include "opt_nfs.h" 46 #include "opt_mbuftrace.h" 47 #endif 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/evcnt.h> 52 #include <sys/callout.h> 53 #include <sys/proc.h> 54 #include <sys/mount.h> 55 #include <sys/kernel.h> 56 #include <sys/kmem.h> 57 #include <sys/mbuf.h> 58 #include <sys/vnode.h> 59 #include <sys/domain.h> 60 #include <sys/protosw.h> 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/syslog.h> 64 #include <sys/tprintf.h> 65 #include <sys/namei.h> 66 #include <sys/signal.h> 67 #include <sys/signalvar.h> 68 #include <sys/kauth.h> 69 70 #include <netinet/in.h> 71 #include <netinet/tcp.h> 72 73 #include <nfs/rpcv2.h> 74 #include <nfs/nfsproto.h> 75 #include <nfs/nfs.h> 76 #include <nfs/xdr_subs.h> 77 #include <nfs/nfsm_subs.h> 78 #include <nfs/nfsmount.h> 79 #include <nfs/nfsnode.h> 80 #include <nfs/nfsrtt.h> 81 #include <nfs/nfs_var.h> 82 83 static int nfs_sndlock(struct nfsmount *, struct nfsreq *); 84 static void nfs_sndunlock(struct nfsmount *); 85 86 /* 87 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all 88 * done by soreceive(), but for SOCK_STREAM we must deal with the Record 89 * Mark and consolidate the data into a new mbuf list. 90 * nb: Sometimes TCP passes the data up to soreceive() in long lists of 91 * small mbufs. 92 * For SOCK_STREAM we must be very careful to read an entire record once 93 * we have read any of it, even if the system call has been interrupted. 94 */ 95 static int 96 nfs_receive(struct nfsreq *rep, struct mbuf **aname, struct mbuf **mp, 97 struct lwp *l) 98 { 99 struct socket *so; 100 struct uio auio; 101 struct iovec aio; 102 struct mbuf *m; 103 struct mbuf *control; 104 u_int32_t len; 105 struct mbuf **getnam; 106 int error, sotype, rcvflg; 107 108 /* 109 * Set up arguments for soreceive() 110 */ 111 *mp = NULL; 112 *aname = NULL; 113 sotype = rep->r_nmp->nm_sotype; 114 115 /* 116 * For reliable protocols, lock against other senders/receivers 117 * in case a reconnect is necessary. 118 * For SOCK_STREAM, first get the Record Mark to find out how much 119 * more there is to get. 120 * We must lock the socket against other receivers 121 * until we have an entire rpc request/reply. 122 */ 123 if (sotype != SOCK_DGRAM) { 124 error = nfs_sndlock(rep->r_nmp, rep); 125 if (error) 126 return (error); 127 tryagain: 128 /* 129 * Check for fatal errors and resending request. 130 */ 131 /* 132 * Ugh: If a reconnect attempt just happened, nm_so 133 * would have changed. NULL indicates a failed 134 * attempt that has essentially shut down this 135 * mount point. 136 */ 137 if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { 138 nfs_sndunlock(rep->r_nmp); 139 return (EINTR); 140 } 141 so = rep->r_nmp->nm_so; 142 if (!so) { 143 error = nfs_reconnect(rep); 144 if (error) { 145 nfs_sndunlock(rep->r_nmp); 146 return (error); 147 } 148 goto tryagain; 149 } 150 while (rep->r_flags & R_MUSTRESEND) { 151 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 152 nfsstats.rpcretries++; 153 rep->r_rtt = 0; 154 rep->r_flags &= ~R_TIMING; 155 error = nfs_send(so, rep->r_nmp->nm_nam, m, rep, l); 156 if (error) { 157 if (error == EINTR || error == ERESTART || 158 (error = nfs_reconnect(rep)) != 0) { 159 nfs_sndunlock(rep->r_nmp); 160 return (error); 161 } 162 goto tryagain; 163 } 164 } 165 nfs_sndunlock(rep->r_nmp); 166 if (sotype == SOCK_STREAM) { 167 aio.iov_base = (void *) &len; 168 aio.iov_len = sizeof(u_int32_t); 169 auio.uio_iov = &aio; 170 auio.uio_iovcnt = 1; 171 auio.uio_rw = UIO_READ; 172 auio.uio_offset = 0; 173 auio.uio_resid = sizeof(u_int32_t); 174 UIO_SETUP_SYSSPACE(&auio); 175 do { 176 rcvflg = MSG_WAITALL; 177 error = (*so->so_receive)(so, NULL, &auio, 178 NULL, NULL, &rcvflg); 179 if (error == EWOULDBLOCK && rep) { 180 if (rep->r_flags & R_SOFTTERM) 181 return (EINTR); 182 /* 183 * if it seems that the server died after it 184 * received our request, set EPIPE so that 185 * we'll reconnect and retransmit requests. 186 */ 187 if (rep->r_rexmit >= rep->r_nmp->nm_retry) { 188 nfsstats.rpctimeouts++; 189 error = EPIPE; 190 } 191 } 192 } while (error == EWOULDBLOCK); 193 if (!error && auio.uio_resid > 0) { 194 /* 195 * Don't log a 0 byte receive; it means 196 * that the socket has been closed, and 197 * can happen during normal operation 198 * (forcible unmount or Solaris server). 199 */ 200 if (auio.uio_resid != sizeof (u_int32_t)) 201 log(LOG_INFO, 202 "short receive (%lu/%lu) from nfs server %s\n", 203 (u_long)sizeof(u_int32_t) - auio.uio_resid, 204 (u_long)sizeof(u_int32_t), 205 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 206 error = EPIPE; 207 } 208 if (error) 209 goto errout; 210 len = ntohl(len) & ~0x80000000; 211 /* 212 * This is SERIOUS! We are out of sync with the sender 213 * and forcing a disconnect/reconnect is all I can do. 214 */ 215 if (len > NFS_MAXPACKET) { 216 log(LOG_ERR, "%s (%d) from nfs server %s\n", 217 "impossible packet length", 218 len, 219 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 220 error = EFBIG; 221 goto errout; 222 } 223 auio.uio_resid = len; 224 do { 225 rcvflg = MSG_WAITALL; 226 error = (*so->so_receive)(so, NULL, 227 &auio, mp, NULL, &rcvflg); 228 } while (error == EWOULDBLOCK || error == EINTR || 229 error == ERESTART); 230 if (!error && auio.uio_resid > 0) { 231 if (len != auio.uio_resid) 232 log(LOG_INFO, 233 "short receive (%lu/%d) from nfs server %s\n", 234 (u_long)len - auio.uio_resid, len, 235 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 236 error = EPIPE; 237 } 238 } else { 239 /* 240 * NB: Since uio_resid is big, MSG_WAITALL is ignored 241 * and soreceive() will return when it has either a 242 * control msg or a data msg. 243 * We have no use for control msg., but must grab them 244 * and then throw them away so we know what is going 245 * on. 246 */ 247 auio.uio_resid = len = 100000000; /* Anything Big */ 248 /* not need to setup uio_vmspace */ 249 do { 250 rcvflg = 0; 251 error = (*so->so_receive)(so, NULL, 252 &auio, mp, &control, &rcvflg); 253 if (control) 254 m_freem(control); 255 if (error == EWOULDBLOCK && rep) { 256 if (rep->r_flags & R_SOFTTERM) 257 return (EINTR); 258 } 259 } while (error == EWOULDBLOCK || 260 (!error && *mp == NULL && control)); 261 if ((rcvflg & MSG_EOR) == 0) 262 printf("Egad!!\n"); 263 if (!error && *mp == NULL) 264 error = EPIPE; 265 len -= auio.uio_resid; 266 } 267 errout: 268 if (error && error != EINTR && error != ERESTART) { 269 m_freem(*mp); 270 *mp = NULL; 271 if (error != EPIPE) 272 log(LOG_INFO, 273 "receive error %d from nfs server %s\n", 274 error, 275 rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); 276 error = nfs_sndlock(rep->r_nmp, rep); 277 if (!error) 278 error = nfs_reconnect(rep); 279 if (!error) 280 goto tryagain; 281 else 282 nfs_sndunlock(rep->r_nmp); 283 } 284 } else { 285 if ((so = rep->r_nmp->nm_so) == NULL) 286 return (EACCES); 287 if (so->so_state & SS_ISCONNECTED) 288 getnam = NULL; 289 else 290 getnam = aname; 291 auio.uio_resid = len = 1000000; 292 /* not need to setup uio_vmspace */ 293 do { 294 rcvflg = 0; 295 error = (*so->so_receive)(so, getnam, &auio, mp, 296 NULL, &rcvflg); 297 if (error == EWOULDBLOCK && 298 (rep->r_flags & R_SOFTTERM)) 299 return (EINTR); 300 } while (error == EWOULDBLOCK); 301 len -= auio.uio_resid; 302 if (!error && *mp == NULL) 303 error = EPIPE; 304 } 305 if (error) { 306 m_freem(*mp); 307 *mp = NULL; 308 } 309 return (error); 310 } 311 312 /* 313 * Implement receipt of reply on a socket. 314 * We must search through the list of received datagrams matching them 315 * with outstanding requests using the xid, until ours is found. 316 */ 317 /* ARGSUSED */ 318 static int 319 nfs_reply(struct nfsreq *myrep, struct lwp *lwp) 320 { 321 struct nfsreq *rep; 322 struct nfsmount *nmp = myrep->r_nmp; 323 int32_t t1; 324 struct mbuf *mrep, *nam, *md; 325 u_int32_t rxid, *tl; 326 char *dpos, *cp2; 327 int error, s; 328 329 /* 330 * Loop around until we get our own reply 331 */ 332 for (;;) { 333 /* 334 * Lock against other receivers so that I don't get stuck in 335 * sbwait() after someone else has received my reply for me. 336 * Also necessary for connection based protocols to avoid 337 * race conditions during a reconnect. 338 */ 339 error = nfs_rcvlock(nmp, myrep); 340 if (error == EALREADY) 341 return (0); 342 if (error) 343 return (error); 344 /* 345 * Get the next Rpc reply off the socket 346 */ 347 348 mutex_enter(&nmp->nm_lock); 349 nmp->nm_waiters++; 350 mutex_exit(&nmp->nm_lock); 351 352 error = nfs_receive(myrep, &nam, &mrep, lwp); 353 354 mutex_enter(&nmp->nm_lock); 355 nmp->nm_waiters--; 356 cv_signal(&nmp->nm_disconcv); 357 mutex_exit(&nmp->nm_lock); 358 359 if (error) { 360 nfs_rcvunlock(nmp); 361 362 if (nmp->nm_iflag & NFSMNT_DISMNT) { 363 /* 364 * Oops, we're going away now.. 365 */ 366 return error; 367 } 368 /* 369 * Ignore routing errors on connectionless protocols? ? 370 */ 371 if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { 372 nmp->nm_so->so_error = 0; 373 #ifdef DEBUG 374 if (ratecheck(&nfs_reply_last_err_time, 375 &nfs_err_interval)) 376 printf("%s: ignoring error %d\n", 377 __func__, error); 378 #endif 379 continue; 380 } 381 return (error); 382 } 383 if (nam) 384 m_freem(nam); 385 386 /* 387 * Get the xid and check that it is an rpc reply 388 */ 389 md = mrep; 390 dpos = mtod(md, void *); 391 nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); 392 rxid = *tl++; 393 if (*tl != rpc_reply) { 394 nfsstats.rpcinvalid++; 395 m_freem(mrep); 396 nfsmout: 397 nfs_rcvunlock(nmp); 398 continue; 399 } 400 401 /* 402 * Loop through the request list to match up the reply 403 * Iff no match, just drop the datagram 404 */ 405 s = splsoftnet(); 406 TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { 407 if (rep->r_mrep == NULL && rxid == rep->r_xid) { 408 /* Found it.. */ 409 rep->r_mrep = mrep; 410 rep->r_md = md; 411 rep->r_dpos = dpos; 412 if (nfsrtton) { 413 struct rttl *rt; 414 415 rt = &nfsrtt.rttl[nfsrtt.pos]; 416 rt->proc = rep->r_procnum; 417 rt->rto = NFS_RTO(nmp, nfs_proct[rep->r_procnum]); 418 rt->sent = nmp->nm_sent; 419 rt->cwnd = nmp->nm_cwnd; 420 rt->srtt = nmp->nm_srtt[nfs_proct[rep->r_procnum] - 1]; 421 rt->sdrtt = nmp->nm_sdrtt[nfs_proct[rep->r_procnum] - 1]; 422 rt->fsid = nmp->nm_mountp->mnt_stat.f_fsidx; 423 getmicrotime(&rt->tstamp); 424 if (rep->r_flags & R_TIMING) 425 rt->rtt = rep->r_rtt; 426 else 427 rt->rtt = 1000000; 428 nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; 429 } 430 /* 431 * Update congestion window. 432 * Do the additive increase of 433 * one rpc/rtt. 434 */ 435 if (nmp->nm_cwnd <= nmp->nm_sent) { 436 nmp->nm_cwnd += 437 (NFS_CWNDSCALE * NFS_CWNDSCALE + 438 (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; 439 if (nmp->nm_cwnd > NFS_MAXCWND) 440 nmp->nm_cwnd = NFS_MAXCWND; 441 } 442 rep->r_flags &= ~R_SENT; 443 nmp->nm_sent -= NFS_CWNDSCALE; 444 /* 445 * Update rtt using a gain of 0.125 on the mean 446 * and a gain of 0.25 on the deviation. 447 */ 448 if (rep->r_flags & R_TIMING) { 449 /* 450 * Since the timer resolution of 451 * NFS_HZ is so course, it can often 452 * result in r_rtt == 0. Since 453 * r_rtt == N means that the actual 454 * rtt is between N+dt and N+2-dt ticks, 455 * add 1. 456 */ 457 t1 = rep->r_rtt + 1; 458 t1 -= (NFS_SRTT(rep) >> 3); 459 NFS_SRTT(rep) += t1; 460 if (t1 < 0) 461 t1 = -t1; 462 t1 -= (NFS_SDRTT(rep) >> 2); 463 NFS_SDRTT(rep) += t1; 464 } 465 nmp->nm_timeouts = 0; 466 break; 467 } 468 } 469 splx(s); 470 nfs_rcvunlock(nmp); 471 /* 472 * If not matched to a request, drop it. 473 * If it's mine, get out. 474 */ 475 if (rep == 0) { 476 nfsstats.rpcunexpected++; 477 m_freem(mrep); 478 } else if (rep == myrep) { 479 if (rep->r_mrep == NULL) 480 panic("nfsreply nil"); 481 return (0); 482 } 483 } 484 } 485 486 /* 487 * nfs_request - goes something like this 488 * - fill in request struct 489 * - links it into list 490 * - calls nfs_send() for first transmit 491 * - calls nfs_receive() to get reply 492 * - break down rpc header and return with nfs reply pointed to 493 * by mrep or error 494 * nb: always frees up mreq mbuf list 495 */ 496 int 497 nfs_request(struct nfsnode *np, struct mbuf *mrest, int procnum, struct lwp *lwp, kauth_cred_t cred, struct mbuf **mrp, struct mbuf **mdp, char **dposp, int *rexmitp) 498 { 499 struct mbuf *m, *mrep; 500 struct nfsreq *rep; 501 u_int32_t *tl; 502 int i; 503 struct nfsmount *nmp = VFSTONFS(np->n_vnode->v_mount); 504 struct mbuf *md, *mheadend; 505 char nickv[RPCX_NICKVERF]; 506 time_t waituntil; 507 char *dpos, *cp2; 508 int t1, s, error = 0, mrest_len, auth_len, auth_type; 509 int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; 510 int verf_len, verf_type; 511 u_int32_t xid; 512 char *auth_str, *verf_str; 513 NFSKERBKEY_T key; /* save session key */ 514 kauth_cred_t acred; 515 struct mbuf *mrest_backup = NULL; 516 kauth_cred_t origcred = NULL; /* XXX: gcc */ 517 bool retry_cred = true; 518 bool use_opencred = (np->n_flag & NUSEOPENCRED) != 0; 519 520 if (rexmitp != NULL) 521 *rexmitp = 0; 522 523 acred = kauth_cred_alloc(); 524 525 tryagain_cred: 526 KASSERT(cred != NULL); 527 rep = kmem_alloc(sizeof(*rep), KM_SLEEP); 528 rep->r_nmp = nmp; 529 KASSERT(lwp == NULL || lwp == curlwp); 530 rep->r_lwp = lwp; 531 rep->r_procnum = procnum; 532 i = 0; 533 m = mrest; 534 while (m) { 535 i += m->m_len; 536 m = m->m_next; 537 } 538 mrest_len = i; 539 540 /* 541 * Get the RPC header with authorization. 542 */ 543 kerbauth: 544 verf_str = auth_str = NULL; 545 if (nmp->nm_flag & NFSMNT_KERB) { 546 verf_str = nickv; 547 verf_len = sizeof (nickv); 548 auth_type = RPCAUTH_KERB4; 549 memset((void *)key, 0, sizeof (key)); 550 if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, 551 &auth_len, verf_str, verf_len)) { 552 error = nfs_getauth(nmp, rep, cred, &auth_str, 553 &auth_len, verf_str, &verf_len, key); 554 if (error) { 555 kmem_free(rep, sizeof(*rep)); 556 m_freem(mrest); 557 KASSERT(kauth_cred_getrefcnt(acred) == 1); 558 kauth_cred_free(acred); 559 return (error); 560 } 561 } 562 retry_cred = false; 563 } else { 564 /* AUTH_UNIX */ 565 uid_t uid; 566 gid_t gid; 567 568 /* 569 * on the most unix filesystems, permission checks are 570 * done when the file is open(2)'ed. 571 * ie. once a file is successfully open'ed, 572 * following i/o operations never fail with EACCES. 573 * we try to follow the semantics as far as possible. 574 * 575 * note that we expect that the nfs server always grant 576 * accesses by the file's owner. 577 */ 578 origcred = cred; 579 switch (procnum) { 580 case NFSPROC_READ: 581 case NFSPROC_WRITE: 582 case NFSPROC_COMMIT: 583 uid = np->n_vattr->va_uid; 584 gid = np->n_vattr->va_gid; 585 if (kauth_cred_geteuid(cred) == uid && 586 kauth_cred_getegid(cred) == gid) { 587 retry_cred = false; 588 break; 589 } 590 if (use_opencred) 591 break; 592 kauth_cred_setuid(acred, uid); 593 kauth_cred_seteuid(acred, uid); 594 kauth_cred_setsvuid(acred, uid); 595 kauth_cred_setgid(acred, gid); 596 kauth_cred_setegid(acred, gid); 597 kauth_cred_setsvgid(acred, gid); 598 cred = acred; 599 break; 600 default: 601 retry_cred = false; 602 break; 603 } 604 /* 605 * backup mbuf chain if we can need it later to retry. 606 * 607 * XXX maybe we can keep a direct reference to 608 * mrest without doing m_copym, but it's ...ugly. 609 */ 610 if (retry_cred) 611 mrest_backup = m_copym(mrest, 0, M_COPYALL, M_WAIT); 612 auth_type = RPCAUTH_UNIX; 613 /* XXX elad - ngroups */ 614 auth_len = (((kauth_cred_ngroups(cred) > nmp->nm_numgrps) ? 615 nmp->nm_numgrps : kauth_cred_ngroups(cred)) << 2) + 616 5 * NFSX_UNSIGNED; 617 } 618 m = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, 619 auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid); 620 if (auth_str) 621 free(auth_str, M_TEMP); 622 623 /* 624 * For stream protocols, insert a Sun RPC Record Mark. 625 */ 626 if (nmp->nm_sotype == SOCK_STREAM) { 627 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 628 *mtod(m, u_int32_t *) = htonl(0x80000000 | 629 (m->m_pkthdr.len - NFSX_UNSIGNED)); 630 } 631 rep->r_mreq = m; 632 rep->r_xid = xid; 633 tryagain: 634 if (nmp->nm_flag & NFSMNT_SOFT) 635 rep->r_retry = nmp->nm_retry; 636 else 637 rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ 638 rep->r_rtt = rep->r_rexmit = 0; 639 if (nfs_proct[procnum] > 0) 640 rep->r_flags = R_TIMING; 641 else 642 rep->r_flags = 0; 643 rep->r_mrep = NULL; 644 645 /* 646 * Do the client side RPC. 647 */ 648 nfsstats.rpcrequests++; 649 /* 650 * Chain request into list of outstanding requests. Be sure 651 * to put it LAST so timer finds oldest requests first. 652 */ 653 s = splsoftnet(); 654 TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); 655 nfs_timer_start(); 656 657 /* 658 * If backing off another request or avoiding congestion, don't 659 * send this one now but let timer do it. If not timing a request, 660 * do it now. 661 */ 662 if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || 663 (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) { 664 splx(s); 665 if (nmp->nm_soflags & PR_CONNREQUIRED) 666 error = nfs_sndlock(nmp, rep); 667 if (!error) { 668 m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); 669 error = nfs_send(nmp->nm_so, nmp->nm_nam, m, rep, lwp); 670 if (nmp->nm_soflags & PR_CONNREQUIRED) 671 nfs_sndunlock(nmp); 672 } 673 s = splsoftnet(); 674 if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { 675 if ((rep->r_flags & R_SENT) == 0) { 676 nmp->nm_sent += NFS_CWNDSCALE; 677 rep->r_flags |= R_SENT; 678 } 679 } 680 splx(s); 681 } else { 682 splx(s); 683 rep->r_rtt = -1; 684 } 685 686 /* 687 * Wait for the reply from our send or the timer's. 688 */ 689 if (!error || error == EPIPE || error == EWOULDBLOCK) 690 error = nfs_reply(rep, lwp); 691 692 /* 693 * RPC done, unlink the request. 694 */ 695 s = splsoftnet(); 696 TAILQ_REMOVE(&nfs_reqq, rep, r_chain); 697 698 /* 699 * Decrement the outstanding request count. 700 */ 701 if (rep->r_flags & R_SENT) { 702 rep->r_flags &= ~R_SENT; /* paranoia */ 703 nmp->nm_sent -= NFS_CWNDSCALE; 704 } 705 splx(s); 706 707 if (rexmitp != NULL) { 708 int rexmit; 709 710 if (nmp->nm_sotype != SOCK_DGRAM) 711 rexmit = (rep->r_flags & R_REXMITTED) != 0; 712 else 713 rexmit = rep->r_rexmit; 714 *rexmitp = rexmit; 715 } 716 717 /* 718 * If there was a successful reply and a tprintf msg. 719 * tprintf a response. 720 */ 721 if (!error && (rep->r_flags & R_TPRINTFMSG)) 722 nfs_msg(rep->r_lwp, nmp->nm_mountp->mnt_stat.f_mntfromname, 723 "is alive again"); 724 mrep = rep->r_mrep; 725 md = rep->r_md; 726 dpos = rep->r_dpos; 727 if (error) 728 goto nfsmout; 729 730 /* 731 * break down the rpc header and check if ok 732 */ 733 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 734 if (*tl++ == rpc_msgdenied) { 735 if (*tl == rpc_mismatch) 736 error = EOPNOTSUPP; 737 else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { 738 if (!failed_auth) { 739 failed_auth++; 740 mheadend->m_next = NULL; 741 m_freem(mrep); 742 m_freem(rep->r_mreq); 743 goto kerbauth; 744 } else 745 error = EAUTH; 746 } else 747 error = EACCES; 748 m_freem(mrep); 749 goto nfsmout; 750 } 751 752 /* 753 * Grab any Kerberos verifier, otherwise just throw it away. 754 */ 755 verf_type = fxdr_unsigned(int, *tl++); 756 i = fxdr_unsigned(int32_t, *tl); 757 if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { 758 error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); 759 if (error) 760 goto nfsmout; 761 } else if (i > 0) 762 nfsm_adv(nfsm_rndup(i)); 763 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 764 /* 0 == ok */ 765 if (*tl == 0) { 766 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); 767 if (*tl != 0) { 768 error = fxdr_unsigned(int, *tl); 769 switch (error) { 770 case NFSERR_PERM: 771 error = EPERM; 772 break; 773 774 case NFSERR_NOENT: 775 error = ENOENT; 776 break; 777 778 case NFSERR_IO: 779 error = EIO; 780 break; 781 782 case NFSERR_NXIO: 783 error = ENXIO; 784 break; 785 786 case NFSERR_ACCES: 787 error = EACCES; 788 if (!retry_cred) 789 break; 790 m_freem(mrep); 791 m_freem(rep->r_mreq); 792 kmem_free(rep, sizeof(*rep)); 793 use_opencred = !use_opencred; 794 if (mrest_backup == NULL) { 795 /* m_copym failure */ 796 KASSERT( 797 kauth_cred_getrefcnt(acred) == 1); 798 kauth_cred_free(acred); 799 return ENOMEM; 800 } 801 mrest = mrest_backup; 802 mrest_backup = NULL; 803 cred = origcred; 804 error = 0; 805 retry_cred = false; 806 goto tryagain_cred; 807 808 case NFSERR_EXIST: 809 error = EEXIST; 810 break; 811 812 case NFSERR_XDEV: 813 error = EXDEV; 814 break; 815 816 case NFSERR_NODEV: 817 error = ENODEV; 818 break; 819 820 case NFSERR_NOTDIR: 821 error = ENOTDIR; 822 break; 823 824 case NFSERR_ISDIR: 825 error = EISDIR; 826 break; 827 828 case NFSERR_INVAL: 829 error = EINVAL; 830 break; 831 832 case NFSERR_FBIG: 833 error = EFBIG; 834 break; 835 836 case NFSERR_NOSPC: 837 error = ENOSPC; 838 break; 839 840 case NFSERR_ROFS: 841 error = EROFS; 842 break; 843 844 case NFSERR_MLINK: 845 error = EMLINK; 846 break; 847 848 case NFSERR_TIMEDOUT: 849 error = ETIMEDOUT; 850 break; 851 852 case NFSERR_NAMETOL: 853 error = ENAMETOOLONG; 854 break; 855 856 case NFSERR_NOTEMPTY: 857 error = ENOTEMPTY; 858 break; 859 860 case NFSERR_DQUOT: 861 error = EDQUOT; 862 break; 863 864 case NFSERR_STALE: 865 /* 866 * If the File Handle was stale, invalidate the 867 * lookup cache, just in case. 868 */ 869 error = ESTALE; 870 cache_purge(NFSTOV(np)); 871 break; 872 873 case NFSERR_REMOTE: 874 error = EREMOTE; 875 break; 876 877 case NFSERR_WFLUSH: 878 case NFSERR_BADHANDLE: 879 case NFSERR_NOT_SYNC: 880 case NFSERR_BAD_COOKIE: 881 error = EINVAL; 882 break; 883 884 case NFSERR_NOTSUPP: 885 error = ENOTSUP; 886 break; 887 888 case NFSERR_TOOSMALL: 889 case NFSERR_SERVERFAULT: 890 case NFSERR_BADTYPE: 891 error = EINVAL; 892 break; 893 894 case NFSERR_TRYLATER: 895 if ((nmp->nm_flag & NFSMNT_NFSV3) == 0) 896 break; 897 m_freem(mrep); 898 error = 0; 899 waituntil = time_second + trylater_delay; 900 while (time_second < waituntil) { 901 kpause("nfstrylater", false, hz, NULL); 902 } 903 trylater_delay *= NFS_TRYLATERDELMUL; 904 if (trylater_delay > NFS_TRYLATERDELMAX) 905 trylater_delay = NFS_TRYLATERDELMAX; 906 /* 907 * RFC1813: 908 * The client should wait and then try 909 * the request with a new RPC transaction ID. 910 */ 911 nfs_renewxid(rep); 912 goto tryagain; 913 914 default: 915 #ifdef DIAGNOSTIC 916 printf("Invalid rpc error code %d\n", error); 917 #endif 918 error = EINVAL; 919 break; 920 } 921 922 if (nmp->nm_flag & NFSMNT_NFSV3) { 923 *mrp = mrep; 924 *mdp = md; 925 *dposp = dpos; 926 error |= NFSERR_RETERR; 927 } else 928 m_freem(mrep); 929 goto nfsmout; 930 } 931 932 /* 933 * note which credential worked to minimize number of retries. 934 */ 935 if (use_opencred) 936 np->n_flag |= NUSEOPENCRED; 937 else 938 np->n_flag &= ~NUSEOPENCRED; 939 940 *mrp = mrep; 941 *mdp = md; 942 *dposp = dpos; 943 944 KASSERT(error == 0); 945 goto nfsmout; 946 } 947 m_freem(mrep); 948 error = EPROTONOSUPPORT; 949 nfsmout: 950 KASSERT(kauth_cred_getrefcnt(acred) == 1); 951 kauth_cred_free(acred); 952 m_freem(rep->r_mreq); 953 kmem_free(rep, sizeof(*rep)); 954 m_freem(mrest_backup); 955 return (error); 956 } 957 958 /* 959 * Lock a socket against others. 960 * Necessary for STREAM sockets to ensure you get an entire rpc request/reply 961 * and also to avoid race conditions between the processes with nfs requests 962 * in progress when a reconnect is necessary. 963 */ 964 static int 965 nfs_sndlock(struct nfsmount *nmp, struct nfsreq *rep) 966 { 967 struct lwp *l; 968 int timeo = 0; 969 bool catch = false; 970 int error = 0; 971 972 if (nmp->nm_flag & NFSMNT_SOFT) 973 timeo = nmp->nm_retry * nmp->nm_timeo; 974 975 if (nmp->nm_iflag & NFSMNT_DISMNTFORCE) 976 timeo = hz; 977 978 if (rep) { 979 l = rep->r_lwp; 980 if (rep->r_nmp->nm_flag & NFSMNT_INT) 981 catch = true; 982 } else 983 l = NULL; 984 mutex_enter(&nmp->nm_lock); 985 while ((nmp->nm_iflag & NFSMNT_SNDLOCK) != 0) { 986 if (rep && nfs_sigintr(rep->r_nmp, rep, l)) { 987 error = EINTR; 988 goto quit; 989 } 990 if (catch) { 991 error = cv_timedwait_sig(&nmp->nm_sndcv, 992 &nmp->nm_lock, timeo); 993 } else { 994 error = cv_timedwait(&nmp->nm_sndcv, 995 &nmp->nm_lock, timeo); 996 } 997 998 if (error) { 999 if ((error == EWOULDBLOCK) && 1000 (nmp->nm_flag & NFSMNT_SOFT)) { 1001 error = EIO; 1002 goto quit; 1003 } 1004 error = 0; 1005 } 1006 if (catch) { 1007 catch = false; 1008 timeo = 2 * hz; 1009 } 1010 } 1011 nmp->nm_iflag |= NFSMNT_SNDLOCK; 1012 quit: 1013 mutex_exit(&nmp->nm_lock); 1014 return error; 1015 } 1016 1017 /* 1018 * Unlock the stream socket for others. 1019 */ 1020 static void 1021 nfs_sndunlock(struct nfsmount *nmp) 1022 { 1023 1024 mutex_enter(&nmp->nm_lock); 1025 if ((nmp->nm_iflag & NFSMNT_SNDLOCK) == 0) 1026 panic("nfs sndunlock"); 1027 nmp->nm_iflag &= ~NFSMNT_SNDLOCK; 1028 cv_signal(&nmp->nm_sndcv); 1029 mutex_exit(&nmp->nm_lock); 1030 } 1031
Indexes created Mon Sep 29 03:10:08 GMT 2025