nfs_syscalls.c revision 1.142
1 /* $NetBSD: nfs_syscalls.c,v 1.142 2008/11/19 18:36:09 ad 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 1.142 2008/11/19 18:36:09 ad 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 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure"); 299 300 static struct nfssvc_sock * 301 nfsrv_sockalloc() 302 { 303 struct nfssvc_sock *slp; 304 305 slp = kmem_alloc(sizeof(*slp), KM_SLEEP); 306 memset(slp, 0, sizeof (struct nfssvc_sock)); 307 mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET); 308 mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET); 309 cv_init(&slp->ns_cv, "nfsdsock"); 310 TAILQ_INIT(&slp->ns_uidlruhead); 311 LIST_INIT(&slp->ns_tq); 312 SIMPLEQ_INIT(&slp->ns_sendq); 313 mutex_enter(&nfsd_lock); 314 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 315 mutex_exit(&nfsd_lock); 316 317 return slp; 318 } 319 320 static void 321 nfsrv_sockfree(struct nfssvc_sock *slp) 322 { 323 324 KASSERT(slp->ns_so == NULL); 325 KASSERT(slp->ns_fp == NULL); 326 KASSERT((slp->ns_flags & SLP_VALID) == 0); 327 mutex_destroy(&slp->ns_lock); 328 mutex_destroy(&slp->ns_alock); 329 cv_destroy(&slp->ns_cv); 330 kmem_free(slp, sizeof(*slp)); 331 } 332 333 /* 334 * Adds a socket to the list for servicing by nfsds. 335 */ 336 int 337 nfssvc_addsock(fp, mynam) 338 file_t *fp; 339 struct mbuf *mynam; 340 { 341 int siz; 342 struct nfssvc_sock *slp; 343 struct socket *so; 344 struct nfssvc_sock *tslp; 345 int error; 346 int val; 347 348 so = (struct socket *)fp->f_data; 349 tslp = (struct nfssvc_sock *)0; 350 /* 351 * Add it to the list, as required. 352 */ 353 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 354 if (so->so_proto->pr_domain->dom_family == AF_INET6) 355 tslp = nfs_udp6sock; 356 else { 357 tslp = nfs_udpsock; 358 if (tslp->ns_flags & SLP_VALID) { 359 m_freem(mynam); 360 return (EPERM); 361 } 362 } 363 } 364 if (so->so_type == SOCK_STREAM) 365 siz = NFS_MAXPACKET + sizeof (u_long); 366 else 367 siz = NFS_MAXPACKET; 368 solock(so); 369 error = soreserve(so, siz, siz); 370 sounlock(so); 371 if (error) { 372 m_freem(mynam); 373 return (error); 374 } 375 376 /* 377 * Set protocol specific options { for now TCP only } and 378 * reserve some space. For datagram sockets, this can get called 379 * repeatedly for the same socket, but that isn't harmful. 380 */ 381 if (so->so_type == SOCK_STREAM) { 382 val = 1; 383 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 384 sizeof(val)); 385 } 386 if ((so->so_proto->pr_domain->dom_family == AF_INET || 387 so->so_proto->pr_domain->dom_family == AF_INET6) && 388 so->so_proto->pr_protocol == IPPROTO_TCP) { 389 val = 1; 390 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 391 sizeof(val)); 392 } 393 solock(so); 394 so->so_rcv.sb_flags &= ~SB_NOINTR; 395 so->so_rcv.sb_timeo = 0; 396 so->so_snd.sb_flags &= ~SB_NOINTR; 397 so->so_snd.sb_timeo = 0; 398 sounlock(so); 399 if (tslp) { 400 slp = tslp; 401 } else { 402 slp = nfsrv_sockalloc(); 403 } 404 slp->ns_so = so; 405 slp->ns_nam = mynam; 406 mutex_enter(&fp->f_lock); 407 fp->f_count++; 408 mutex_exit(&fp->f_lock); 409 slp->ns_fp = fp; 410 slp->ns_flags = SLP_VALID; 411 slp->ns_aflags = SLP_A_NEEDQ; 412 slp->ns_gflags = 0; 413 slp->ns_sflags = 0; 414 solock(so); 415 so->so_upcallarg = (void *)slp; 416 so->so_upcall = nfsrv_soupcall; 417 so->so_rcv.sb_flags |= SB_UPCALL; 418 sounlock(so); 419 nfsrv_wakenfsd(slp); 420 return (0); 421 } 422 423 /* 424 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 425 * until it is killed by a signal. 426 */ 427 int 428 nfssvc_nfsd(nsd, argp, l) 429 struct nfsd_srvargs *nsd; 430 void *argp; 431 struct lwp *l; 432 { 433 struct timeval tv; 434 struct mbuf *m; 435 struct nfssvc_sock *slp; 436 struct nfsd *nfsd = nsd->nsd_nfsd; 437 struct nfsrv_descript *nd = NULL; 438 struct mbuf *mreq; 439 u_quad_t cur_usec; 440 int error = 0, cacherep, siz, sotype, writes_todo; 441 struct proc *p = l->l_proc; 442 int s; 443 bool doreinit; 444 445 #ifndef nolint 446 cacherep = RC_DOIT; 447 writes_todo = 0; 448 #endif 449 uvm_lwp_hold(l); 450 if (nfsd == NULL) { 451 nsd->nsd_nfsd = nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP); 452 memset(nfsd, 0, sizeof (struct nfsd)); 453 cv_init(&nfsd->nfsd_cv, "nfsd"); 454 nfsd->nfsd_procp = p; 455 mutex_enter(&nfsd_lock); 456 while ((nfssvc_sockhead_flag & SLP_INIT) != 0) { 457 KASSERT(nfs_numnfsd == 0); 458 cv_wait(&nfsd_initcv, &nfsd_lock); 459 } 460 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 461 nfs_numnfsd++; 462 mutex_exit(&nfsd_lock); 463 } 464 /* 465 * Loop getting rpc requests until SIGKILL. 466 */ 467 for (;;) { 468 bool dummy; 469 470 if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD) 471 != 0) { 472 preempt(); 473 } 474 if (nfsd->nfsd_slp == NULL) { 475 mutex_enter(&nfsd_lock); 476 while (nfsd->nfsd_slp == NULL && 477 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 478 SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd, 479 nfsd_idle); 480 error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock); 481 if (error) { 482 slp = nfsd->nfsd_slp; 483 nfsd->nfsd_slp = NULL; 484 if (!slp) 485 SLIST_REMOVE(&nfsd_idle_head, 486 nfsd, nfsd, nfsd_idle); 487 mutex_exit(&nfsd_lock); 488 if (slp) { 489 nfsrv_wakenfsd(slp); 490 nfsrv_slpderef(slp); 491 } 492 goto done; 493 } 494 } 495 if (nfsd->nfsd_slp == NULL && 496 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 497 slp = TAILQ_FIRST(&nfssvc_sockpending); 498 if (slp) { 499 KASSERT((slp->ns_gflags & SLP_G_DOREC) 500 != 0); 501 TAILQ_REMOVE(&nfssvc_sockpending, slp, 502 ns_pending); 503 slp->ns_gflags &= ~SLP_G_DOREC; 504 slp->ns_sref++; 505 nfsd->nfsd_slp = slp; 506 } else 507 nfsd_head_flag &= ~NFSD_CHECKSLP; 508 } 509 KASSERT(nfsd->nfsd_slp == NULL || 510 nfsd->nfsd_slp->ns_sref > 0); 511 mutex_exit(&nfsd_lock); 512 if ((slp = nfsd->nfsd_slp) == NULL) 513 continue; 514 if (slp->ns_flags & SLP_VALID) { 515 bool more; 516 517 if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) { 518 nfsrv_rcv(slp); 519 } 520 if (nfsdsock_testbits(slp, SLP_A_DISCONN)) { 521 nfsrv_zapsock(slp); 522 } 523 error = nfsrv_dorec(slp, nfsd, &nd, &more); 524 getmicrotime(&tv); 525 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 526 (u_quad_t)tv.tv_usec; 527 writes_todo = 0; 528 if (error) { 529 struct nfsrv_descript *nd2; 530 531 mutex_enter(&nfsd_lock); 532 nd2 = LIST_FIRST(&slp->ns_tq); 533 if (nd2 != NULL && 534 nd2->nd_time <= cur_usec) { 535 error = 0; 536 cacherep = RC_DOIT; 537 writes_todo = 1; 538 } 539 mutex_exit(&nfsd_lock); 540 } 541 if (error == 0 && more) { 542 nfsrv_wakenfsd(slp); 543 } 544 } 545 } else { 546 error = 0; 547 slp = nfsd->nfsd_slp; 548 } 549 KASSERT(slp != NULL); 550 KASSERT(nfsd->nfsd_slp == slp); 551 if (error || (slp->ns_flags & SLP_VALID) == 0) { 552 if (nd) { 553 nfsdreq_free(nd); 554 nd = NULL; 555 } 556 nfsd->nfsd_slp = NULL; 557 nfsrv_slpderef(slp); 558 continue; 559 } 560 sotype = slp->ns_so->so_type; 561 if (nd) { 562 getmicrotime(&nd->nd_starttime); 563 if (nd->nd_nam2) 564 nd->nd_nam = nd->nd_nam2; 565 else 566 nd->nd_nam = slp->ns_nam; 567 568 /* 569 * Check to see if authorization is needed. 570 */ 571 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 572 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 573 nsd->nsd_haddr = mtod(nd->nd_nam, 574 struct sockaddr_in *)->sin_addr.s_addr; 575 nsd->nsd_authlen = nfsd->nfsd_authlen; 576 nsd->nsd_verflen = nfsd->nfsd_verflen; 577 if (!copyout(nfsd->nfsd_authstr, 578 nsd->nsd_authstr, nfsd->nfsd_authlen) && 579 !copyout(nfsd->nfsd_verfstr, 580 nsd->nsd_verfstr, nfsd->nfsd_verflen) && 581 !copyout(nsd, argp, sizeof (*nsd))) { 582 uvm_lwp_rele(l); 583 return (ENEEDAUTH); 584 } 585 cacherep = RC_DROPIT; 586 } else 587 cacherep = nfsrv_getcache(nd, slp, &mreq); 588 589 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 590 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 591 nd->nd_procnum = NFSPROC_NOOP; 592 nd->nd_repstat = 593 (NFSERR_AUTHERR | AUTH_TOOWEAK); 594 cacherep = RC_DOIT; 595 } 596 } 597 598 /* 599 * Loop to get all the write rpc relies that have been 600 * gathered together. 601 */ 602 do { 603 switch (cacherep) { 604 case RC_DOIT: 605 mreq = NULL; 606 netexport_rdlock(); 607 if (writes_todo || nd == NULL || 608 (!(nd->nd_flag & ND_NFSV3) && 609 nd->nd_procnum == NFSPROC_WRITE && 610 nfsrvw_procrastinate > 0)) 611 error = nfsrv_writegather(&nd, slp, 612 l, &mreq); 613 else 614 error = 615 (*(nfsrv3_procs[nd->nd_procnum])) 616 (nd, slp, l, &mreq); 617 netexport_rdunlock(); 618 if (mreq == NULL) { 619 if (nd != NULL) { 620 if (nd->nd_nam2) 621 m_free(nd->nd_nam2); 622 } 623 break; 624 } 625 if (error) { 626 nfsstats.srv_errs++; 627 nfsrv_updatecache(nd, false, mreq); 628 if (nd->nd_nam2) 629 m_freem(nd->nd_nam2); 630 break; 631 } 632 nfsstats.srvrpccnt[nd->nd_procnum]++; 633 nfsrv_updatecache(nd, true, mreq); 634 nd->nd_mrep = (struct mbuf *)0; 635 case RC_REPLY: 636 m = mreq; 637 siz = 0; 638 while (m) { 639 siz += m->m_len; 640 m = m->m_next; 641 } 642 if (siz <= 0 || siz > NFS_MAXPACKET) { 643 printf("mbuf siz=%d\n",siz); 644 panic("Bad nfs svc reply"); 645 } 646 m = mreq; 647 m->m_pkthdr.len = siz; 648 m->m_pkthdr.rcvif = (struct ifnet *)0; 649 /* 650 * For stream protocols, prepend a Sun RPC 651 * Record Mark. 652 */ 653 if (sotype == SOCK_STREAM) { 654 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 655 *mtod(m, u_int32_t *) = 656 htonl(0x80000000 | siz); 657 } 658 nd->nd_mreq = m; 659 if (nfsrtton) { 660 nfsd_rt(slp->ns_so->so_type, nd, 661 cacherep); 662 } 663 error = nfsdsock_sendreply(slp, nd); 664 nd = NULL; 665 if (error == EPIPE) 666 nfsrv_zapsock(slp); 667 if (error == EINTR || error == ERESTART) { 668 nfsd->nfsd_slp = NULL; 669 nfsrv_slpderef(slp); 670 goto done; 671 } 672 break; 673 case RC_DROPIT: 674 if (nfsrtton) 675 nfsd_rt(sotype, nd, cacherep); 676 m_freem(nd->nd_mrep); 677 m_freem(nd->nd_nam2); 678 break; 679 } 680 if (nd) { 681 nfsdreq_free(nd); 682 nd = NULL; 683 } 684 685 /* 686 * Check to see if there are outstanding writes that 687 * need to be serviced. 688 */ 689 getmicrotime(&tv); 690 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 691 (u_quad_t)tv.tv_usec; 692 s = splsoftclock(); 693 if (LIST_FIRST(&slp->ns_tq) && 694 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) { 695 cacherep = RC_DOIT; 696 writes_todo = 1; 697 } else 698 writes_todo = 0; 699 splx(s); 700 } while (writes_todo); 701 if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) { 702 nfsd->nfsd_slp = NULL; 703 nfsrv_slpderef(slp); 704 } 705 } 706 done: 707 mutex_enter(&nfsd_lock); 708 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 709 doreinit = --nfs_numnfsd == 0; 710 if (doreinit) 711 nfssvc_sockhead_flag |= SLP_INIT; 712 mutex_exit(&nfsd_lock); 713 cv_destroy(&nfsd->nfsd_cv); 714 kmem_free(nfsd, sizeof(*nfsd)); 715 nsd->nsd_nfsd = NULL; 716 if (doreinit) 717 nfsrv_init(true); /* Reinitialize everything */ 718 uvm_lwp_rele(l); 719 return (error); 720 } 721 722 /* 723 * Shut down a socket associated with an nfssvc_sock structure. 724 * Should be called with the send lock set, if required. 725 * The trick here is to increment the sref at the start, so that the nfsds 726 * will stop using it and clear ns_flag at the end so that it will not be 727 * reassigned during cleanup. 728 * 729 * called at splsoftnet. 730 */ 731 void 732 nfsrv_zapsock(slp) 733 struct nfssvc_sock *slp; 734 { 735 struct nfsuid *nuidp, *nnuidp; 736 struct nfsrv_descript *nwp; 737 struct socket *so; 738 struct mbuf *m; 739 740 if (nfsdsock_drain(slp)) { 741 return; 742 } 743 mutex_enter(&nfsd_lock); 744 if (slp->ns_gflags & SLP_G_DOREC) { 745 TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending); 746 slp->ns_gflags &= ~SLP_G_DOREC; 747 } 748 mutex_exit(&nfsd_lock); 749 750 so = slp->ns_so; 751 KASSERT(so != NULL); 752 solock(so); 753 so->so_upcall = NULL; 754 so->so_upcallarg = NULL; 755 so->so_rcv.sb_flags &= ~SB_UPCALL; 756 soshutdown(so, SHUT_RDWR); 757 sounlock(so); 758 759 if (slp->ns_nam) 760 m_free(slp->ns_nam); 761 m_freem(slp->ns_raw); 762 m = slp->ns_rec; 763 while (m != NULL) { 764 struct mbuf *n; 765 766 n = m->m_nextpkt; 767 m_freem(m); 768 m = n; 769 } 770 for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0; 771 nuidp = nnuidp) { 772 nnuidp = TAILQ_NEXT(nuidp, nu_lru); 773 LIST_REMOVE(nuidp, nu_hash); 774 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 775 if (nuidp->nu_flag & NU_NAM) 776 m_freem(nuidp->nu_nam); 777 kmem_free(nuidp, sizeof(*nuidp)); 778 } 779 mutex_enter(&nfsd_lock); 780 while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) { 781 LIST_REMOVE(nwp, nd_tq); 782 mutex_exit(&nfsd_lock); 783 nfsdreq_free(nwp); 784 mutex_enter(&nfsd_lock); 785 } 786 mutex_exit(&nfsd_lock); 787 } 788 789 /* 790 * Derefence a server socket structure. If it has no more references and 791 * is no longer valid, you can throw it away. 792 */ 793 void 794 nfsrv_slpderef(slp) 795 struct nfssvc_sock *slp; 796 { 797 uint32_t ref; 798 799 mutex_enter(&nfsd_lock); 800 KASSERT(slp->ns_sref > 0); 801 ref = --slp->ns_sref; 802 mutex_exit(&nfsd_lock); 803 if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) { 804 file_t *fp; 805 806 mutex_enter(&nfsd_lock); 807 KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0); 808 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 809 mutex_exit(&nfsd_lock); 810 811 fp = slp->ns_fp; 812 if (fp != NULL) { 813 slp->ns_fp = NULL; 814 KASSERT(fp != NULL); 815 KASSERT(fp->f_data == slp->ns_so); 816 KASSERT(fp->f_count > 0); 817 closef(fp); 818 slp->ns_so = NULL; 819 } 820 821 nfsrv_sockfree(slp); 822 } 823 } 824 825 /* 826 * Initialize the data structures for the server. 827 * Handshake with any new nfsds starting up to avoid any chance of 828 * corruption. 829 */ 830 void 831 nfsrv_init(terminating) 832 int terminating; 833 { 834 struct nfssvc_sock *slp; 835 836 if (!terminating) { 837 mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET); 838 cv_init(&nfsd_initcv, "nfsdinit"); 839 } 840 841 mutex_enter(&nfsd_lock); 842 if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0) 843 panic("nfsd init"); 844 nfssvc_sockhead_flag |= SLP_INIT; 845 846 if (terminating) { 847 KASSERT(SLIST_EMPTY(&nfsd_idle_head)); 848 KASSERT(TAILQ_EMPTY(&nfsd_head)); 849 while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) { 850 mutex_exit(&nfsd_lock); 851 KASSERT(slp->ns_sref == 0); 852 slp->ns_sref++; 853 nfsrv_zapsock(slp); 854 nfsrv_slpderef(slp); 855 mutex_enter(&nfsd_lock); 856 } 857 KASSERT(TAILQ_EMPTY(&nfssvc_sockpending)); 858 mutex_exit(&nfsd_lock); 859 nfsrv_cleancache(); /* And clear out server cache */ 860 } else { 861 mutex_exit(&nfsd_lock); 862 nfs_pub.np_valid = 0; 863 } 864 865 TAILQ_INIT(&nfssvc_sockhead); 866 TAILQ_INIT(&nfssvc_sockpending); 867 868 TAILQ_INIT(&nfsd_head); 869 SLIST_INIT(&nfsd_idle_head); 870 nfsd_head_flag &= ~NFSD_CHECKSLP; 871 872 nfs_udpsock = nfsrv_sockalloc(); 873 nfs_udp6sock = nfsrv_sockalloc(); 874 875 mutex_enter(&nfsd_lock); 876 nfssvc_sockhead_flag &= ~SLP_INIT; 877 cv_broadcast(&nfsd_initcv); 878 mutex_exit(&nfsd_lock); 879 } 880 881 void 882 nfsrv_fini(void) 883 { 884 885 nfsrv_init(true); 886 cv_destroy(&nfsd_initcv); 887 mutex_destroy(&nfsd_lock); 888 } 889 890 /* 891 * Add entries to the server monitor log. 892 */ 893 static void 894 nfsd_rt(sotype, nd, cacherep) 895 int sotype; 896 struct nfsrv_descript *nd; 897 int cacherep; 898 { 899 struct timeval tv; 900 struct drt *rt; 901 902 rt = &nfsdrt.drt[nfsdrt.pos]; 903 if (cacherep == RC_DOIT) 904 rt->flag = 0; 905 else if (cacherep == RC_REPLY) 906 rt->flag = DRT_CACHEREPLY; 907 else 908 rt->flag = DRT_CACHEDROP; 909 if (sotype == SOCK_STREAM) 910 rt->flag |= DRT_TCP; 911 if (nd->nd_flag & ND_NFSV3) 912 rt->flag |= DRT_NFSV3; 913 rt->proc = nd->nd_procnum; 914 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 915 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 916 else 917 rt->ipadr = INADDR_ANY; 918 getmicrotime(&tv); 919 rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 920 (tv.tv_usec - nd->nd_starttime.tv_usec); 921 rt->tstamp = tv; 922 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 923 } 924
Indexes created Tue Oct 21 08:09:48 GMT 2025