1 /* $NetBSD: nfs_syscalls.c,v 1.143 2008/11/28 06:47:08 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 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_syscalls.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 #include <sys/cdefs.h> 38 __KERNEL_RCSID(0, "$NetBSD: nfs_syscalls.c,v 1.143 2008/11/28 06:47:08 pooka Exp $"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/file.h> 44 #include <sys/stat.h> 45 #include <sys/vnode.h> 46 #include <sys/mount.h> 47 #include <sys/proc.h> 48 #include <sys/uio.h> 49 #include <sys/malloc.h> 50 #include <sys/kmem.h> 51 #include <sys/buf.h> 52 #include <sys/mbuf.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/signalvar.h> 56 #include <sys/domain.h> 57 #include <sys/protosw.h> 58 #include <sys/namei.h> 59 #include <sys/syslog.h> 60 #include <sys/filedesc.h> 61 #include <sys/kthread.h> 62 #include <sys/kauth.h> 63 #include <sys/syscallargs.h> 64 65 #include <netinet/in.h> 66 #include <netinet/tcp.h> 67 #include <nfs/xdr_subs.h> 68 #include <nfs/rpcv2.h> 69 #include <nfs/nfsproto.h> 70 #include <nfs/nfs.h> 71 #include <nfs/nfsm_subs.h> 72 #include <nfs/nfsrvcache.h> 73 #include <nfs/nfsmount.h> 74 #include <nfs/nfsnode.h> 75 #include <nfs/nfsrtt.h> 76 #include <nfs/nfs_var.h> 77 78 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 79 struct nfssvc_sock *, 80 struct lwp *, struct mbuf **)); 81 extern int nfsrvw_procrastinate; 82 extern int nuidhash_max; 83 84 static int nfs_numnfsd = 0; 85 static struct nfsdrt nfsdrt; 86 kmutex_t nfsd_lock; 87 struct nfssvc_sockhead nfssvc_sockhead; 88 kcondvar_t nfsd_initcv; 89 struct nfssvc_sockhead nfssvc_sockpending; 90 struct nfsdhead nfsd_head; 91 struct nfsdidlehead nfsd_idle_head; 92 93 int nfssvc_sockhead_flag; 94 int nfsd_head_flag; 95 96 struct nfssvc_sock *nfs_udpsock; 97 struct nfssvc_sock *nfs_udp6sock; 98 99 static struct nfssvc_sock *nfsrv_sockalloc __P((void)); 100 static void nfsrv_sockfree __P((struct nfssvc_sock *)); 101 static void nfsd_rt __P((int, struct nfsrv_descript *, int)); 102 103 /* 104 * NFS server system calls 105 */ 106 107 108 /* 109 * Nfs server pseudo system call for the nfsd's 110 * Based on the flag value it either: 111 * - adds a socket to the selection list 112 * - remains in the kernel as an nfsd 113 * - remains in the kernel as an nfsiod 114 */ 115 int 116 sys_nfssvc(struct lwp *l, const struct sys_nfssvc_args *uap, register_t *retval) 117 { 118 /* { 119 syscallarg(int) flag; 120 syscallarg(void *) argp; 121 } */ 122 int error; 123 file_t *fp; 124 struct mbuf *nam; 125 struct nfsd_args nfsdarg; 126 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 127 struct nfsd *nfsd; 128 struct nfssvc_sock *slp; 129 struct nfsuid *nuidp; 130 131 /* 132 * Must be super user 133 */ 134 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_NFS, 135 KAUTH_REQ_NETWORK_NFS_SVC, NULL, NULL, NULL); 136 if (error) 137 return (error); 138 139 mutex_enter(&nfsd_lock); 140 while (nfssvc_sockhead_flag & SLP_INIT) { 141 cv_wait(&nfsd_initcv, &nfsd_lock); 142 } 143 mutex_exit(&nfsd_lock); 144 145 if (SCARG(uap, flag) & NFSSVC_BIOD) { 146 /* Dummy implementation of nfsios for 1.4 and earlier. */ 147 error = kpause("nfsbiod", true, 0, NULL); 148 } else if (SCARG(uap, flag) & NFSSVC_MNTD) { 149 error = ENOSYS; 150 } else if (SCARG(uap, flag) & NFSSVC_ADDSOCK) { 151 error = copyin(SCARG(uap, argp), (void *)&nfsdarg, 152 sizeof(nfsdarg)); 153 if (error) 154 return (error); 155 /* getsock() will use the descriptor for us */ 156 if ((fp = fd_getfile(nfsdarg.sock)) == NULL) 157 return (EBADF); 158 if (fp->f_type != DTYPE_SOCKET) { 159 fd_putfile(nfsdarg.sock); 160 return (ENOTSOCK); 161 } 162 if (error) 163 return (error); 164 /* 165 * Get the client address for connected sockets. 166 */ 167 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 168 nam = (struct mbuf *)0; 169 else { 170 error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen, 171 MT_SONAME); 172 if (error) { 173 fd_putfile(nfsdarg.sock); 174 return (error); 175 } 176 } 177 error = nfssvc_addsock(fp, nam); 178 fd_putfile(nfsdarg.sock); 179 } else if (SCARG(uap, flag) & NFSSVC_SETEXPORTSLIST) { 180 struct export_args *args; 181 struct mountd_exports_list mel; 182 183 error = copyin(SCARG(uap, argp), &mel, sizeof(mel)); 184 if (error != 0) 185 return error; 186 187 args = (struct export_args *)malloc(mel.mel_nexports * 188 sizeof(struct export_args), M_TEMP, M_WAITOK); 189 error = copyin(mel.mel_exports, args, mel.mel_nexports * 190 sizeof(struct export_args)); 191 if (error != 0) { 192 free(args, M_TEMP); 193 return error; 194 } 195 mel.mel_exports = args; 196 197 error = mountd_set_exports_list(&mel, l); 198 199 free(args, M_TEMP); 200 } else { 201 error = copyin(SCARG(uap, argp), (void *)nsd, sizeof (*nsd)); 202 if (error) 203 return (error); 204 if ((SCARG(uap, flag) & NFSSVC_AUTHIN) && 205 ((nfsd = nsd->nsd_nfsd)) != NULL && 206 (nfsd->nfsd_slp->ns_flags & SLP_VALID)) { 207 slp = nfsd->nfsd_slp; 208 209 /* 210 * First check to see if another nfsd has already 211 * added this credential. 212 */ 213 LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid), 214 nu_hash) { 215 if (kauth_cred_geteuid(nuidp->nu_cr) == 216 nsd->nsd_cr.cr_uid && 217 (!nfsd->nfsd_nd->nd_nam2 || 218 netaddr_match(NU_NETFAM(nuidp), 219 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 220 break; 221 } 222 if (nuidp) { 223 kauth_cred_hold(nuidp->nu_cr); 224 nfsd->nfsd_nd->nd_cr = nuidp->nu_cr; 225 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 226 } else { 227 /* 228 * Nope, so we will. 229 */ 230 if (slp->ns_numuids < nuidhash_max) { 231 slp->ns_numuids++; 232 nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP); 233 } else 234 nuidp = (struct nfsuid *)0; 235 if ((slp->ns_flags & SLP_VALID) == 0) { 236 if (nuidp) 237 kmem_free(nuidp, sizeof(*nuidp)); 238 } else { 239 if (nuidp == (struct nfsuid *)0) { 240 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 241 LIST_REMOVE(nuidp, nu_hash); 242 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 243 nu_lru); 244 if (nuidp->nu_flag & NU_NAM) 245 m_freem(nuidp->nu_nam); 246 } 247 nuidp->nu_flag = 0; 248 kauth_uucred_to_cred(nuidp->nu_cr, 249 &nsd->nsd_cr); 250 nuidp->nu_timestamp = nsd->nsd_timestamp; 251 nuidp->nu_expire = time_second + nsd->nsd_ttl; 252 /* 253 * and save the session key in nu_key. 254 */ 255 memcpy(nuidp->nu_key, nsd->nsd_key, 256 sizeof(nsd->nsd_key)); 257 if (nfsd->nfsd_nd->nd_nam2) { 258 struct sockaddr_in *saddr; 259 260 saddr = mtod(nfsd->nfsd_nd->nd_nam2, 261 struct sockaddr_in *); 262 switch (saddr->sin_family) { 263 case AF_INET: 264 nuidp->nu_flag |= NU_INETADDR; 265 nuidp->nu_inetaddr = 266 saddr->sin_addr.s_addr; 267 break; 268 case AF_INET6: 269 nuidp->nu_flag |= NU_NAM; 270 nuidp->nu_nam = m_copym( 271 nfsd->nfsd_nd->nd_nam2, 0, 272 M_COPYALL, M_WAIT); 273 break; 274 default: 275 return EAFNOSUPPORT; 276 }; 277 } 278 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 279 nu_lru); 280 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 281 nuidp, nu_hash); 282 kauth_cred_hold(nuidp->nu_cr); 283 nfsd->nfsd_nd->nd_cr = nuidp->nu_cr; 284 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 285 } 286 } 287 } 288 if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) && 289 (nfsd = nsd->nsd_nfsd)) 290 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 291 error = nfssvc_nfsd(nsd, SCARG(uap, argp), l); 292 } 293 if (error == EINTR || error == ERESTART) 294 error = 0; 295 return (error); 296 } 297 298 static struct nfssvc_sock * 299 nfsrv_sockalloc() 300 { 301 struct nfssvc_sock *slp; 302 303 slp = kmem_alloc(sizeof(*slp), KM_SLEEP); 304 memset(slp, 0, sizeof (struct nfssvc_sock)); 305 mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET); 306 mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET); 307 cv_init(&slp->ns_cv, "nfsdsock"); 308 TAILQ_INIT(&slp->ns_uidlruhead); 309 LIST_INIT(&slp->ns_tq); 310 SIMPLEQ_INIT(&slp->ns_sendq); 311 mutex_enter(&nfsd_lock); 312 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 313 mutex_exit(&nfsd_lock); 314 315 return slp; 316 } 317 318 static void 319 nfsrv_sockfree(struct nfssvc_sock *slp) 320 { 321 322 KASSERT(slp->ns_so == NULL); 323 KASSERT(slp->ns_fp == NULL); 324 KASSERT((slp->ns_flags & SLP_VALID) == 0); 325 mutex_destroy(&slp->ns_lock); 326 mutex_destroy(&slp->ns_alock); 327 cv_destroy(&slp->ns_cv); 328 kmem_free(slp, sizeof(*slp)); 329 } 330 331 /* 332 * Adds a socket to the list for servicing by nfsds. 333 */ 334 int 335 nfssvc_addsock(fp, mynam) 336 file_t *fp; 337 struct mbuf *mynam; 338 { 339 int siz; 340 struct nfssvc_sock *slp; 341 struct socket *so; 342 struct nfssvc_sock *tslp; 343 int error; 344 int val; 345 346 so = (struct socket *)fp->f_data; 347 tslp = (struct nfssvc_sock *)0; 348 /* 349 * Add it to the list, as required. 350 */ 351 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 352 if (so->so_proto->pr_domain->dom_family == AF_INET6) 353 tslp = nfs_udp6sock; 354 else { 355 tslp = nfs_udpsock; 356 if (tslp->ns_flags & SLP_VALID) { 357 m_freem(mynam); 358 return (EPERM); 359 } 360 } 361 } 362 if (so->so_type == SOCK_STREAM) 363 siz = NFS_MAXPACKET + sizeof (u_long); 364 else 365 siz = NFS_MAXPACKET; 366 solock(so); 367 error = soreserve(so, siz, siz); 368 sounlock(so); 369 if (error) { 370 m_freem(mynam); 371 return (error); 372 } 373 374 /* 375 * Set protocol specific options { for now TCP only } and 376 * reserve some space. For datagram sockets, this can get called 377 * repeatedly for the same socket, but that isn't harmful. 378 */ 379 if (so->so_type == SOCK_STREAM) { 380 val = 1; 381 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 382 sizeof(val)); 383 } 384 if ((so->so_proto->pr_domain->dom_family == AF_INET || 385 so->so_proto->pr_domain->dom_family == AF_INET6) && 386 so->so_proto->pr_protocol == IPPROTO_TCP) { 387 val = 1; 388 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 389 sizeof(val)); 390 } 391 solock(so); 392 so->so_rcv.sb_flags &= ~SB_NOINTR; 393 so->so_rcv.sb_timeo = 0; 394 so->so_snd.sb_flags &= ~SB_NOINTR; 395 so->so_snd.sb_timeo = 0; 396 sounlock(so); 397 if (tslp) { 398 slp = tslp; 399 } else { 400 slp = nfsrv_sockalloc(); 401 } 402 slp->ns_so = so; 403 slp->ns_nam = mynam; 404 mutex_enter(&fp->f_lock); 405 fp->f_count++; 406 mutex_exit(&fp->f_lock); 407 slp->ns_fp = fp; 408 slp->ns_flags = SLP_VALID; 409 slp->ns_aflags = SLP_A_NEEDQ; 410 slp->ns_gflags = 0; 411 slp->ns_sflags = 0; 412 solock(so); 413 so->so_upcallarg = (void *)slp; 414 so->so_upcall = nfsrv_soupcall; 415 so->so_rcv.sb_flags |= SB_UPCALL; 416 sounlock(so); 417 nfsrv_wakenfsd(slp); 418 return (0); 419 } 420 421 /* 422 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 423 * until it is killed by a signal. 424 */ 425 int 426 nfssvc_nfsd(nsd, argp, l) 427 struct nfsd_srvargs *nsd; 428 void *argp; 429 struct lwp *l; 430 { 431 struct timeval tv; 432 struct mbuf *m; 433 struct nfssvc_sock *slp; 434 struct nfsd *nfsd = nsd->nsd_nfsd; 435 struct nfsrv_descript *nd = NULL; 436 struct mbuf *mreq; 437 u_quad_t cur_usec; 438 int error = 0, cacherep, siz, sotype, writes_todo; 439 struct proc *p = l->l_proc; 440 int s; 441 bool doreinit; 442 443 #ifndef nolint 444 cacherep = RC_DOIT; 445 writes_todo = 0; 446 #endif 447 uvm_lwp_hold(l); 448 if (nfsd == NULL) { 449 nsd->nsd_nfsd = nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP); 450 memset(nfsd, 0, sizeof (struct nfsd)); 451 cv_init(&nfsd->nfsd_cv, "nfsd"); 452 nfsd->nfsd_procp = p; 453 mutex_enter(&nfsd_lock); 454 while ((nfssvc_sockhead_flag & SLP_INIT) != 0) { 455 KASSERT(nfs_numnfsd == 0); 456 cv_wait(&nfsd_initcv, &nfsd_lock); 457 } 458 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 459 nfs_numnfsd++; 460 mutex_exit(&nfsd_lock); 461 } 462 /* 463 * Loop getting rpc requests until SIGKILL. 464 */ 465 for (;;) { 466 bool dummy; 467 468 if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD) 469 != 0) { 470 preempt(); 471 } 472 if (nfsd->nfsd_slp == NULL) { 473 mutex_enter(&nfsd_lock); 474 while (nfsd->nfsd_slp == NULL && 475 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 476 SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd, 477 nfsd_idle); 478 error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock); 479 if (error) { 480 slp = nfsd->nfsd_slp; 481 nfsd->nfsd_slp = NULL; 482 if (!slp) 483 SLIST_REMOVE(&nfsd_idle_head, 484 nfsd, nfsd, nfsd_idle); 485 mutex_exit(&nfsd_lock); 486 if (slp) { 487 nfsrv_wakenfsd(slp); 488 nfsrv_slpderef(slp); 489 } 490 goto done; 491 } 492 } 493 if (nfsd->nfsd_slp == NULL && 494 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 495 slp = TAILQ_FIRST(&nfssvc_sockpending); 496 if (slp) { 497 KASSERT((slp->ns_gflags & SLP_G_DOREC) 498 != 0); 499 TAILQ_REMOVE(&nfssvc_sockpending, slp, 500 ns_pending); 501 slp->ns_gflags &= ~SLP_G_DOREC; 502 slp->ns_sref++; 503 nfsd->nfsd_slp = slp; 504 } else 505 nfsd_head_flag &= ~NFSD_CHECKSLP; 506 } 507 KASSERT(nfsd->nfsd_slp == NULL || 508 nfsd->nfsd_slp->ns_sref > 0); 509 mutex_exit(&nfsd_lock); 510 if ((slp = nfsd->nfsd_slp) == NULL) 511 continue; 512 if (slp->ns_flags & SLP_VALID) { 513 bool more; 514 515 if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) { 516 nfsrv_rcv(slp); 517 } 518 if (nfsdsock_testbits(slp, SLP_A_DISCONN)) { 519 nfsrv_zapsock(slp); 520 } 521 error = nfsrv_dorec(slp, nfsd, &nd, &more); 522 getmicrotime(&tv); 523 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 524 (u_quad_t)tv.tv_usec; 525 writes_todo = 0; 526 if (error) { 527 struct nfsrv_descript *nd2; 528 529 mutex_enter(&nfsd_lock); 530 nd2 = LIST_FIRST(&slp->ns_tq); 531 if (nd2 != NULL && 532 nd2->nd_time <= cur_usec) { 533 error = 0; 534 cacherep = RC_DOIT; 535 writes_todo = 1; 536 } 537 mutex_exit(&nfsd_lock); 538 } 539 if (error == 0 && more) { 540 nfsrv_wakenfsd(slp); 541 } 542 } 543 } else { 544 error = 0; 545 slp = nfsd->nfsd_slp; 546 } 547 KASSERT(slp != NULL); 548 KASSERT(nfsd->nfsd_slp == slp); 549 if (error || (slp->ns_flags & SLP_VALID) == 0) { 550 if (nd) { 551 nfsdreq_free(nd); 552 nd = NULL; 553 } 554 nfsd->nfsd_slp = NULL; 555 nfsrv_slpderef(slp); 556 continue; 557 } 558 sotype = slp->ns_so->so_type; 559 if (nd) { 560 getmicrotime(&nd->nd_starttime); 561 if (nd->nd_nam2) 562 nd->nd_nam = nd->nd_nam2; 563 else 564 nd->nd_nam = slp->ns_nam; 565 566 /* 567 * Check to see if authorization is needed. 568 */ 569 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 570 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 571 nsd->nsd_haddr = mtod(nd->nd_nam, 572 struct sockaddr_in *)->sin_addr.s_addr; 573 nsd->nsd_authlen = nfsd->nfsd_authlen; 574 nsd->nsd_verflen = nfsd->nfsd_verflen; 575 if (!copyout(nfsd->nfsd_authstr, 576 nsd->nsd_authstr, nfsd->nfsd_authlen) && 577 !copyout(nfsd->nfsd_verfstr, 578 nsd->nsd_verfstr, nfsd->nfsd_verflen) && 579 !copyout(nsd, argp, sizeof (*nsd))) { 580 uvm_lwp_rele(l); 581 return (ENEEDAUTH); 582 } 583 cacherep = RC_DROPIT; 584 } else 585 cacherep = nfsrv_getcache(nd, slp, &mreq); 586 587 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 588 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 589 nd->nd_procnum = NFSPROC_NOOP; 590 nd->nd_repstat = 591 (NFSERR_AUTHERR | AUTH_TOOWEAK); 592 cacherep = RC_DOIT; 593 } 594 } 595 596 /* 597 * Loop to get all the write rpc relies that have been 598 * gathered together. 599 */ 600 do { 601 switch (cacherep) { 602 case RC_DOIT: 603 mreq = NULL; 604 netexport_rdlock(); 605 if (writes_todo || nd == NULL || 606 (!(nd->nd_flag & ND_NFSV3) && 607 nd->nd_procnum == NFSPROC_WRITE && 608 nfsrvw_procrastinate > 0)) 609 error = nfsrv_writegather(&nd, slp, 610 l, &mreq); 611 else 612 error = 613 (*(nfsrv3_procs[nd->nd_procnum])) 614 (nd, slp, l, &mreq); 615 netexport_rdunlock(); 616 if (mreq == NULL) { 617 if (nd != NULL) { 618 if (nd->nd_nam2) 619 m_free(nd->nd_nam2); 620 } 621 break; 622 } 623 if (error) { 624 nfsstats.srv_errs++; 625 nfsrv_updatecache(nd, false, mreq); 626 if (nd->nd_nam2) 627 m_freem(nd->nd_nam2); 628 break; 629 } 630 nfsstats.srvrpccnt[nd->nd_procnum]++; 631 nfsrv_updatecache(nd, true, mreq); 632 nd->nd_mrep = (struct mbuf *)0; 633 case RC_REPLY: 634 m = mreq; 635 siz = 0; 636 while (m) { 637 siz += m->m_len; 638 m = m->m_next; 639 } 640 if (siz <= 0 || siz > NFS_MAXPACKET) { 641 printf("mbuf siz=%d\n",siz); 642 panic("Bad nfs svc reply"); 643 } 644 m = mreq; 645 m->m_pkthdr.len = siz; 646 m->m_pkthdr.rcvif = (struct ifnet *)0; 647 /* 648 * For stream protocols, prepend a Sun RPC 649 * Record Mark. 650 */ 651 if (sotype == SOCK_STREAM) { 652 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 653 *mtod(m, u_int32_t *) = 654 htonl(0x80000000 | siz); 655 } 656 nd->nd_mreq = m; 657 if (nfsrtton) { 658 nfsd_rt(slp->ns_so->so_type, nd, 659 cacherep); 660 } 661 error = nfsdsock_sendreply(slp, nd); 662 nd = NULL; 663 if (error == EPIPE) 664 nfsrv_zapsock(slp); 665 if (error == EINTR || error == ERESTART) { 666 nfsd->nfsd_slp = NULL; 667 nfsrv_slpderef(slp); 668 goto done; 669 } 670 break; 671 case RC_DROPIT: 672 if (nfsrtton) 673 nfsd_rt(sotype, nd, cacherep); 674 m_freem(nd->nd_mrep); 675 m_freem(nd->nd_nam2); 676 break; 677 } 678 if (nd) { 679 nfsdreq_free(nd); 680 nd = NULL; 681 } 682 683 /* 684 * Check to see if there are outstanding writes that 685 * need to be serviced. 686 */ 687 getmicrotime(&tv); 688 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 689 (u_quad_t)tv.tv_usec; 690 s = splsoftclock(); 691 if (LIST_FIRST(&slp->ns_tq) && 692 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) { 693 cacherep = RC_DOIT; 694 writes_todo = 1; 695 } else 696 writes_todo = 0; 697 splx(s); 698 } while (writes_todo); 699 if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) { 700 nfsd->nfsd_slp = NULL; 701 nfsrv_slpderef(slp); 702 } 703 } 704 done: 705 mutex_enter(&nfsd_lock); 706 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 707 doreinit = --nfs_numnfsd == 0; 708 if (doreinit) 709 nfssvc_sockhead_flag |= SLP_INIT; 710 mutex_exit(&nfsd_lock); 711 cv_destroy(&nfsd->nfsd_cv); 712 kmem_free(nfsd, sizeof(*nfsd)); 713 nsd->nsd_nfsd = NULL; 714 if (doreinit) 715 nfsrv_init(true); /* Reinitialize everything */ 716 uvm_lwp_rele(l); 717 return (error); 718 } 719 720 /* 721 * Shut down a socket associated with an nfssvc_sock structure. 722 * Should be called with the send lock set, if required. 723 * The trick here is to increment the sref at the start, so that the nfsds 724 * will stop using it and clear ns_flag at the end so that it will not be 725 * reassigned during cleanup. 726 * 727 * called at splsoftnet. 728 */ 729 void 730 nfsrv_zapsock(slp) 731 struct nfssvc_sock *slp; 732 { 733 struct nfsuid *nuidp, *nnuidp; 734 struct nfsrv_descript *nwp; 735 struct socket *so; 736 struct mbuf *m; 737 738 if (nfsdsock_drain(slp)) { 739 return; 740 } 741 mutex_enter(&nfsd_lock); 742 if (slp->ns_gflags & SLP_G_DOREC) { 743 TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending); 744 slp->ns_gflags &= ~SLP_G_DOREC; 745 } 746 mutex_exit(&nfsd_lock); 747 748 so = slp->ns_so; 749 KASSERT(so != NULL); 750 solock(so); 751 so->so_upcall = NULL; 752 so->so_upcallarg = NULL; 753 so->so_rcv.sb_flags &= ~SB_UPCALL; 754 soshutdown(so, SHUT_RDWR); 755 sounlock(so); 756 757 if (slp->ns_nam) 758 m_free(slp->ns_nam); 759 m_freem(slp->ns_raw); 760 m = slp->ns_rec; 761 while (m != NULL) { 762 struct mbuf *n; 763 764 n = m->m_nextpkt; 765 m_freem(m); 766 m = n; 767 } 768 for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0; 769 nuidp = nnuidp) { 770 nnuidp = TAILQ_NEXT(nuidp, nu_lru); 771 LIST_REMOVE(nuidp, nu_hash); 772 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 773 if (nuidp->nu_flag & NU_NAM) 774 m_freem(nuidp->nu_nam); 775 kmem_free(nuidp, sizeof(*nuidp)); 776 } 777 mutex_enter(&nfsd_lock); 778 while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) { 779 LIST_REMOVE(nwp, nd_tq); 780 mutex_exit(&nfsd_lock); 781 nfsdreq_free(nwp); 782 mutex_enter(&nfsd_lock); 783 } 784 mutex_exit(&nfsd_lock); 785 } 786 787 /* 788 * Derefence a server socket structure. If it has no more references and 789 * is no longer valid, you can throw it away. 790 */ 791 void 792 nfsrv_slpderef(slp) 793 struct nfssvc_sock *slp; 794 { 795 uint32_t ref; 796 797 mutex_enter(&nfsd_lock); 798 KASSERT(slp->ns_sref > 0); 799 ref = --slp->ns_sref; 800 mutex_exit(&nfsd_lock); 801 if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) { 802 file_t *fp; 803 804 mutex_enter(&nfsd_lock); 805 KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0); 806 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 807 mutex_exit(&nfsd_lock); 808 809 fp = slp->ns_fp; 810 if (fp != NULL) { 811 slp->ns_fp = NULL; 812 KASSERT(fp != NULL); 813 KASSERT(fp->f_data == slp->ns_so); 814 KASSERT(fp->f_count > 0); 815 closef(fp); 816 slp->ns_so = NULL; 817 } 818 819 nfsrv_sockfree(slp); 820 } 821 } 822 823 /* 824 * Initialize the data structures for the server. 825 * Handshake with any new nfsds starting up to avoid any chance of 826 * corruption. 827 */ 828 void 829 nfsrv_init(terminating) 830 int terminating; 831 { 832 struct nfssvc_sock *slp; 833 834 if (!terminating) { 835 mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET); 836 cv_init(&nfsd_initcv, "nfsdinit"); 837 } 838 839 mutex_enter(&nfsd_lock); 840 if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0) 841 panic("nfsd init"); 842 nfssvc_sockhead_flag |= SLP_INIT; 843 844 if (terminating) { 845 KASSERT(SLIST_EMPTY(&nfsd_idle_head)); 846 KASSERT(TAILQ_EMPTY(&nfsd_head)); 847 while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) { 848 mutex_exit(&nfsd_lock); 849 KASSERT(slp->ns_sref == 0); 850 slp->ns_sref++; 851 nfsrv_zapsock(slp); 852 nfsrv_slpderef(slp); 853 mutex_enter(&nfsd_lock); 854 } 855 KASSERT(TAILQ_EMPTY(&nfssvc_sockpending)); 856 mutex_exit(&nfsd_lock); 857 nfsrv_cleancache(); /* And clear out server cache */ 858 } else { 859 mutex_exit(&nfsd_lock); 860 nfs_pub.np_valid = 0; 861 } 862 863 TAILQ_INIT(&nfssvc_sockhead); 864 TAILQ_INIT(&nfssvc_sockpending); 865 866 TAILQ_INIT(&nfsd_head); 867 SLIST_INIT(&nfsd_idle_head); 868 nfsd_head_flag &= ~NFSD_CHECKSLP; 869 870 nfs_udpsock = nfsrv_sockalloc(); 871 nfs_udp6sock = nfsrv_sockalloc(); 872 873 mutex_enter(&nfsd_lock); 874 nfssvc_sockhead_flag &= ~SLP_INIT; 875 cv_broadcast(&nfsd_initcv); 876 mutex_exit(&nfsd_lock); 877 } 878 879 void 880 nfsrv_fini(void) 881 { 882 883 nfsrv_init(true); 884 cv_destroy(&nfsd_initcv); 885 mutex_destroy(&nfsd_lock); 886 } 887 888 /* 889 * Add entries to the server monitor log. 890 */ 891 static void 892 nfsd_rt(sotype, nd, cacherep) 893 int sotype; 894 struct nfsrv_descript *nd; 895 int cacherep; 896 { 897 struct timeval tv; 898 struct drt *rt; 899 900 rt = &nfsdrt.drt[nfsdrt.pos]; 901 if (cacherep == RC_DOIT) 902 rt->flag = 0; 903 else if (cacherep == RC_REPLY) 904 rt->flag = DRT_CACHEREPLY; 905 else 906 rt->flag = DRT_CACHEDROP; 907 if (sotype == SOCK_STREAM) 908 rt->flag |= DRT_TCP; 909 if (nd->nd_flag & ND_NFSV3) 910 rt->flag |= DRT_NFSV3; 911 rt->proc = nd->nd_procnum; 912 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 913 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 914 else 915 rt->ipadr = INADDR_ANY; 916 getmicrotime(&tv); 917 rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 918 (tv.tv_usec - nd->nd_starttime.tv_usec); 919 rt->tstamp = tv; 920 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 921 } 922