1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 37 * $FreeBSD: src/sys/nfs/nfs_syscalls.c,v 1.58.2.1 2000/11/26 02:30:06 dillon Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_syscalls.c,v 1.17 2004/06/06 19:16:11 dillon Exp $ 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/file.h> 47 #include <sys/filedesc.h> 48 #include <sys/vnode.h> 49 #include <sys/malloc.h> 50 #include <sys/mount.h> 51 #include <sys/proc.h> 52 #include <sys/buf.h> 53 #include <sys/mbuf.h> 54 #include <sys/resourcevar.h> 55 #include <sys/socket.h> 56 #include <sys/socketvar.h> 57 #include <sys/domain.h> 58 #include <sys/protosw.h> 59 #include <sys/namei.h> 60 #include <vm/vm_zone.h> 61 62 #include <netinet/in.h> 63 #include <netinet/tcp.h> 64 #include "xdr_subs.h" 65 #include "rpcv2.h" 66 #include "nfsproto.h" 67 #include "nfs.h" 68 #include "nfsm_subs.h" 69 #include "nfsrvcache.h" 70 #include "nfsmount.h" 71 #include "nfsnode.h" 72 #include "nqnfs.h" 73 #include "nfsrtt.h" 74 75 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 76 77 /* Global defs. */ 78 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd, 79 struct nfssvc_sock *slp, 80 struct thread *td, 81 struct mbuf **mreqp); 82 extern int nfs_numasync; 83 extern time_t nqnfsstarttime; 84 extern int nqsrv_writeslack; 85 extern int nfsrtton; 86 extern struct nfsstats nfsstats; 87 extern int nfsrvw_procrastinate; 88 extern int nfsrvw_procrastinate_v3; 89 static int nuidhash_max = NFS_MAXUIDHASH; 90 91 #ifndef NFS_NOSERVER 92 static void nfsrv_zapsock (struct nfssvc_sock *slp); 93 #endif 94 static int nfssvc_iod (struct thread *); 95 96 #define TRUE 1 97 #define FALSE 0 98 99 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 100 101 SYSCTL_DECL(_vfs_nfs); 102 103 #ifndef NFS_NOSERVER 104 int nfsd_waiting = 0; 105 static struct nfsdrt nfsdrt; 106 static int nfs_numnfsd = 0; 107 static int notstarted = 1; 108 static int modify_flag = 0; 109 static void nfsd_rt (int sotype, struct nfsrv_descript *nd, 110 int cacherep); 111 static int nfssvc_addsock (struct file *, struct sockaddr *, 112 struct thread *); 113 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *); 114 115 static int nfs_privport = 0; 116 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, ""); 117 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, ""); 118 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, ""); 119 static int nfs_soreserve = 65535; 120 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, ""); 121 122 /* 123 * NFS server system calls 124 */ 125 126 #endif /* NFS_NOSERVER */ 127 /* 128 * nfssvc_args(int flag, caddr_t argp) 129 * 130 * Nfs server psuedo system call for the nfsd's 131 * Based on the flag value it either: 132 * - adds a socket to the selection list 133 * - remains in the kernel as an nfsd 134 * - remains in the kernel as an nfsiod 135 */ 136 int 137 nfssvc(struct nfssvc_args *uap) 138 { 139 #ifndef NFS_NOSERVER 140 struct nameidata nd; 141 struct file *fp; 142 struct sockaddr *nam; 143 struct nfsd_args nfsdarg; 144 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 145 struct nfsd_cargs ncd; 146 struct nfsd *nfsd; 147 struct nfssvc_sock *slp; 148 struct nfsuid *nuidp; 149 struct nfsmount *nmp; 150 #endif /* NFS_NOSERVER */ 151 int error; 152 struct thread *td = curthread; 153 154 /* 155 * Must be super user 156 */ 157 error = suser(td); 158 if(error) 159 return (error); 160 KKASSERT(td->td_proc); /* for ucred and p_fd */ 161 while (nfssvc_sockhead_flag & SLP_INIT) { 162 nfssvc_sockhead_flag |= SLP_WANTINIT; 163 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 164 } 165 if (uap->flag & NFSSVC_BIOD) 166 error = nfssvc_iod(td); 167 #ifdef NFS_NOSERVER 168 else 169 error = ENXIO; 170 #else /* !NFS_NOSERVER */ 171 else if (uap->flag & NFSSVC_MNTD) { 172 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 173 if (error) 174 return (error); 175 NDINIT(&nd, NAMEI_LOOKUP, CNP_FOLLOW | CNP_LOCKLEAF, 176 UIO_USERSPACE, ncd.ncd_dirp, td); 177 error = namei(&nd); 178 if (error) 179 return (error); 180 NDFREE(&nd, NDF_ONLY_PNBUF); 181 if ((nd.ni_vp->v_flag & VROOT) == 0) 182 error = EINVAL; 183 nmp = VFSTONFS(nd.ni_vp->v_mount); 184 vput(nd.ni_vp); 185 if (error) 186 return (error); 187 if ((nmp->nm_state & NFSSTA_MNTD) && 188 (uap->flag & NFSSVC_GOTAUTH) == 0) 189 return (0); 190 nmp->nm_state |= NFSSTA_MNTD; 191 error = nqnfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag, 192 uap->argp, td); 193 } else if (uap->flag & NFSSVC_ADDSOCK) { 194 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 195 if (error) 196 return (error); 197 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 198 if (error) 199 return (error); 200 /* 201 * Get the client address for connected sockets. 202 */ 203 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 204 nam = (struct sockaddr *)0; 205 else { 206 error = getsockaddr(&nam, nfsdarg.name, 207 nfsdarg.namelen); 208 if (error) { 209 fdrop(fp, td); 210 return (error); 211 } 212 } 213 error = nfssvc_addsock(fp, nam, td); 214 fdrop(fp, td); 215 } else { 216 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 217 if (error) 218 return (error); 219 if ((uap->flag & NFSSVC_AUTHIN) && 220 ((nfsd = nsd->nsd_nfsd)) != NULL && 221 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 222 slp = nfsd->nfsd_slp; 223 224 /* 225 * First check to see if another nfsd has already 226 * added this credential. 227 */ 228 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 229 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 230 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 231 (!nfsd->nfsd_nd->nd_nam2 || 232 netaddr_match(NU_NETFAM(nuidp), 233 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 234 break; 235 } 236 if (nuidp) { 237 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 238 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 239 } else { 240 /* 241 * Nope, so we will. 242 */ 243 if (slp->ns_numuids < nuidhash_max) { 244 slp->ns_numuids++; 245 nuidp = (struct nfsuid *) 246 malloc(sizeof (struct nfsuid), M_NFSUID, 247 M_WAITOK); 248 } else 249 nuidp = (struct nfsuid *)0; 250 if ((slp->ns_flag & SLP_VALID) == 0) { 251 if (nuidp) 252 free((caddr_t)nuidp, M_NFSUID); 253 } else { 254 if (nuidp == (struct nfsuid *)0) { 255 nuidp = slp->ns_uidlruhead.tqh_first; 256 LIST_REMOVE(nuidp, nu_hash); 257 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 258 nu_lru); 259 if (nuidp->nu_flag & NU_NAM) 260 FREE(nuidp->nu_nam, M_SONAME); 261 } 262 nuidp->nu_flag = 0; 263 nuidp->nu_cr = nsd->nsd_cr; 264 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 265 nuidp->nu_cr.cr_ngroups = NGROUPS; 266 nuidp->nu_cr.cr_ref = 1; 267 nuidp->nu_timestamp = nsd->nsd_timestamp; 268 nuidp->nu_expire = time_second + nsd->nsd_ttl; 269 /* 270 * and save the session key in nu_key. 271 */ 272 bcopy(nsd->nsd_key, nuidp->nu_key, 273 sizeof (nsd->nsd_key)); 274 if (nfsd->nfsd_nd->nd_nam2) { 275 struct sockaddr_in *saddr; 276 277 saddr = (struct sockaddr_in *) 278 nfsd->nfsd_nd->nd_nam2; 279 switch (saddr->sin_family) { 280 case AF_INET: 281 nuidp->nu_flag |= NU_INETADDR; 282 nuidp->nu_inetaddr = 283 saddr->sin_addr.s_addr; 284 break; 285 case AF_ISO: 286 default: 287 nuidp->nu_flag |= NU_NAM; 288 nuidp->nu_nam = 289 dup_sockaddr(nfsd->nfsd_nd->nd_nam2); 290 break; 291 }; 292 } 293 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 294 nu_lru); 295 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 296 nuidp, nu_hash); 297 nfsrv_setcred(&nuidp->nu_cr, 298 &nfsd->nfsd_nd->nd_cr); 299 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 300 } 301 } 302 } 303 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 304 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 305 error = nfssvc_nfsd(nsd, uap->argp, td); 306 } 307 #endif /* NFS_NOSERVER */ 308 if (error == EINTR || error == ERESTART) 309 error = 0; 310 return (error); 311 } 312 313 #ifndef NFS_NOSERVER 314 /* 315 * Adds a socket to the list for servicing by nfsds. 316 */ 317 static int 318 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td) 319 { 320 int siz; 321 struct nfssvc_sock *slp; 322 struct socket *so; 323 int error, s; 324 325 so = (struct socket *)fp->f_data; 326 #if 0 327 tslp = (struct nfssvc_sock *)0; 328 /* 329 * Add it to the list, as required. 330 */ 331 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 332 tslp = nfs_udpsock; 333 if (tslp->ns_flag & SLP_VALID) { 334 if (mynam != NULL) 335 FREE(mynam, M_SONAME); 336 return (EPERM); 337 } 338 } 339 #endif 340 /* 341 * Reserve buffer space in the socket. Note that due to bugs in 342 * Linux's delayed-ack code, serious performance degredation may 343 * occur with linux hosts if the minimum is used. 344 */ 345 if (so->so_type == SOCK_STREAM) 346 siz = NFS_MAXPACKET + sizeof (u_long); 347 else 348 siz = NFS_MAXPACKET; 349 if (siz < nfs_soreserve) 350 siz = nfs_soreserve; 351 if (siz > sb_max_adj) { 352 printf("Warning: vfs.nfs.soreserve (%d) " 353 "limited to adjusted sb_max (%ld)\n", 354 nfs_soreserve, sb_max_adj); 355 siz = sb_max_adj; 356 } 357 358 error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]); 359 if (error) { 360 if (mynam != NULL) 361 FREE(mynam, M_SONAME); 362 return (error); 363 } 364 365 /* 366 * Set protocol specific options { for now TCP only } and 367 * reserve some space. For datagram sockets, this can get called 368 * repeatedly for the same socket, but that isn't harmful. 369 */ 370 if (so->so_type == SOCK_STREAM) { 371 struct sockopt sopt; 372 int val; 373 374 bzero(&sopt, sizeof sopt); 375 sopt.sopt_level = SOL_SOCKET; 376 sopt.sopt_name = SO_KEEPALIVE; 377 sopt.sopt_val = &val; 378 sopt.sopt_valsize = sizeof val; 379 val = 1; 380 sosetopt(so, &sopt); 381 } 382 if (so->so_proto->pr_domain->dom_family == AF_INET && 383 so->so_proto->pr_protocol == IPPROTO_TCP) { 384 struct sockopt sopt; 385 int val; 386 387 bzero(&sopt, sizeof sopt); 388 sopt.sopt_level = IPPROTO_TCP; 389 sopt.sopt_name = TCP_NODELAY; 390 sopt.sopt_val = &val; 391 sopt.sopt_valsize = sizeof val; 392 val = 1; 393 sosetopt(so, &sopt); 394 } 395 so->so_rcv.sb_flags &= ~SB_NOINTR; 396 so->so_rcv.sb_timeo = 0; 397 so->so_snd.sb_flags &= ~SB_NOINTR; 398 so->so_snd.sb_timeo = 0; 399 400 slp = (struct nfssvc_sock *) 401 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 402 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 403 STAILQ_INIT(&slp->ns_rec); 404 TAILQ_INIT(&slp->ns_uidlruhead); 405 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 406 407 slp->ns_so = so; 408 slp->ns_nam = mynam; 409 fp->f_count++; 410 slp->ns_fp = fp; 411 s = splnet(); 412 so->so_upcallarg = (caddr_t)slp; 413 so->so_upcall = nfsrv_rcv; 414 so->so_rcv.sb_flags |= SB_UPCALL; 415 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 416 nfsrv_wakenfsd(slp); 417 splx(s); 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 static int 426 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td) 427 { 428 int siz; 429 struct nfssvc_sock *slp; 430 struct nfsd *nfsd = nsd->nsd_nfsd; 431 struct nfsrv_descript *nd = NULL; 432 struct mbuf *m, *mreq; 433 int error = 0, cacherep, s, sotype, writes_todo; 434 int procrastinate; 435 u_quad_t cur_usec; 436 437 #ifndef nolint 438 cacherep = RC_DOIT; 439 writes_todo = 0; 440 #endif 441 if (nfsd == (struct nfsd *)0) { 442 nsd->nsd_nfsd = nfsd = (struct nfsd *) 443 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 444 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 445 s = splnet(); 446 nfsd->nfsd_td = td; 447 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 448 nfs_numnfsd++; 449 } else 450 s = splnet(); 451 452 /* 453 * Loop getting rpc requests until SIGKILL. 454 */ 455 for (;;) { 456 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 457 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 458 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 459 nfsd->nfsd_flag |= NFSD_WAITING; 460 nfsd_waiting++; 461 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0); 462 nfsd_waiting--; 463 if (error) 464 goto done; 465 } 466 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 467 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 468 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 469 slp = slp->ns_chain.tqe_next) { 470 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 471 == (SLP_VALID | SLP_DOREC)) { 472 slp->ns_flag &= ~SLP_DOREC; 473 slp->ns_sref++; 474 nfsd->nfsd_slp = slp; 475 break; 476 } 477 } 478 if (slp == 0) 479 nfsd_head_flag &= ~NFSD_CHECKSLP; 480 } 481 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 482 continue; 483 if (slp->ns_flag & SLP_VALID) { 484 if (slp->ns_flag & SLP_DISCONN) 485 nfsrv_zapsock(slp); 486 else if (slp->ns_flag & SLP_NEEDQ) { 487 slp->ns_flag &= ~SLP_NEEDQ; 488 (void) nfs_slplock(slp, 1); 489 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 490 MB_WAIT); 491 nfs_slpunlock(slp); 492 } 493 error = nfsrv_dorec(slp, nfsd, &nd); 494 cur_usec = nfs_curusec(); 495 if (error && slp->ns_tq.lh_first && 496 slp->ns_tq.lh_first->nd_time <= cur_usec) { 497 error = 0; 498 cacherep = RC_DOIT; 499 writes_todo = 1; 500 } else 501 writes_todo = 0; 502 nfsd->nfsd_flag |= NFSD_REQINPROG; 503 } 504 } else { 505 error = 0; 506 slp = nfsd->nfsd_slp; 507 } 508 if (error || (slp->ns_flag & SLP_VALID) == 0) { 509 if (nd) { 510 free((caddr_t)nd, M_NFSRVDESC); 511 nd = NULL; 512 } 513 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 514 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 515 nfsrv_slpderef(slp); 516 continue; 517 } 518 splx(s); 519 sotype = slp->ns_so->so_type; 520 if (nd) { 521 getmicrotime(&nd->nd_starttime); 522 if (nd->nd_nam2) 523 nd->nd_nam = nd->nd_nam2; 524 else 525 nd->nd_nam = slp->ns_nam; 526 527 /* 528 * Check to see if authorization is needed. 529 */ 530 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 531 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 532 nsd->nsd_haddr = 533 ((struct sockaddr_in *) 534 nd->nd_nam)->sin_addr.s_addr; 535 nsd->nsd_authlen = nfsd->nfsd_authlen; 536 nsd->nsd_verflen = nfsd->nfsd_verflen; 537 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 538 nfsd->nfsd_authlen) && 539 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 540 nfsd->nfsd_verflen) && 541 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 542 return (ENEEDAUTH); 543 cacherep = RC_DROPIT; 544 } else 545 cacherep = nfsrv_getcache(nd, slp, &mreq); 546 547 /* 548 * Check for just starting up for NQNFS and send 549 * fake "try again later" replies to the NQNFS clients. 550 */ 551 if (notstarted && nqnfsstarttime <= time_second) { 552 if (modify_flag) { 553 nqnfsstarttime = time_second + nqsrv_writeslack; 554 modify_flag = 0; 555 } else 556 notstarted = 0; 557 } 558 if (notstarted) { 559 if ((nd->nd_flag & ND_NQNFS) == 0) 560 cacherep = RC_DROPIT; 561 else if (nd->nd_procnum != NFSPROC_WRITE) { 562 nd->nd_procnum = NFSPROC_NOOP; 563 nd->nd_repstat = NQNFS_TRYLATER; 564 cacherep = RC_DOIT; 565 } else 566 modify_flag = 1; 567 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 568 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 569 nd->nd_procnum = NFSPROC_NOOP; 570 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 571 cacherep = RC_DOIT; 572 } else if (nfs_privport) { 573 /* Check if source port is privileged */ 574 u_short port; 575 struct sockaddr *nam = nd->nd_nam; 576 struct sockaddr_in *sin; 577 578 sin = (struct sockaddr_in *)nam; 579 port = ntohs(sin->sin_port); 580 if (port >= IPPORT_RESERVED && 581 nd->nd_procnum != NFSPROC_NULL) { 582 nd->nd_procnum = NFSPROC_NOOP; 583 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 584 cacherep = RC_DOIT; 585 printf("NFS request from unprivileged port (%s:%d)\n", 586 inet_ntoa(sin->sin_addr), port); 587 } 588 } 589 590 } 591 592 /* 593 * Loop to get all the write rpc relies that have been 594 * gathered together. 595 */ 596 do { 597 switch (cacherep) { 598 case RC_DOIT: 599 if (nd && (nd->nd_flag & ND_NFSV3)) 600 procrastinate = nfsrvw_procrastinate_v3; 601 else 602 procrastinate = nfsrvw_procrastinate; 603 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 604 procrastinate > 0 && !notstarted)) 605 error = nfsrv_writegather(&nd, slp, 606 nfsd->nfsd_td, &mreq); 607 else 608 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 609 slp, nfsd->nfsd_td, &mreq); 610 if (mreq == NULL) 611 break; 612 if (error != 0 && error != NFSERR_RETVOID) { 613 if (nd->nd_procnum != NQNFSPROC_VACATED) 614 nfsstats.srv_errs++; 615 nfsrv_updatecache(nd, FALSE, mreq); 616 if (nd->nd_nam2) 617 FREE(nd->nd_nam2, M_SONAME); 618 break; 619 } 620 nfsstats.srvrpccnt[nd->nd_procnum]++; 621 nfsrv_updatecache(nd, TRUE, mreq); 622 nd->nd_mrep = (struct mbuf *)0; 623 case RC_REPLY: 624 m = mreq; 625 siz = 0; 626 while (m) { 627 siz += m->m_len; 628 m = m->m_next; 629 } 630 if (siz <= 0 || siz > NFS_MAXPACKET) { 631 printf("mbuf siz=%d\n",siz); 632 panic("Bad nfs svc reply"); 633 } 634 m = mreq; 635 m->m_pkthdr.len = siz; 636 m->m_pkthdr.rcvif = (struct ifnet *)0; 637 /* 638 * For stream protocols, prepend a Sun RPC 639 * Record Mark. 640 */ 641 if (sotype == SOCK_STREAM) { 642 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT); 643 if (m == NULL) 644 return (ENOBUFS); 645 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 646 } 647 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 648 (void) nfs_slplock(slp, 1); 649 if (slp->ns_flag & SLP_VALID) 650 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 651 else { 652 error = EPIPE; 653 m_freem(m); 654 } 655 if (nfsrtton) 656 nfsd_rt(sotype, nd, cacherep); 657 if (nd->nd_nam2) 658 FREE(nd->nd_nam2, M_SONAME); 659 if (nd->nd_mrep) 660 m_freem(nd->nd_mrep); 661 if (error == EPIPE) 662 nfsrv_zapsock(slp); 663 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 664 nfs_slpunlock(slp); 665 if (error == EINTR || error == ERESTART) { 666 free((caddr_t)nd, M_NFSRVDESC); 667 nfsrv_slpderef(slp); 668 s = splnet(); 669 goto done; 670 } 671 break; 672 case RC_DROPIT: 673 if (nfsrtton) 674 nfsd_rt(sotype, nd, cacherep); 675 m_freem(nd->nd_mrep); 676 if (nd->nd_nam2) 677 FREE(nd->nd_nam2, M_SONAME); 678 break; 679 }; 680 if (nd) { 681 FREE((caddr_t)nd, M_NFSRVDESC); 682 nd = NULL; 683 } 684 685 /* 686 * Check to see if there are outstanding writes that 687 * need to be serviced. 688 */ 689 cur_usec = nfs_curusec(); 690 s = splsoftclock(); 691 if (slp->ns_tq.lh_first && 692 slp->ns_tq.lh_first->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 s = splnet(); 700 if (nfsrv_dorec(slp, nfsd, &nd)) { 701 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 702 nfsd->nfsd_slp = NULL; 703 nfsrv_slpderef(slp); 704 } 705 } 706 done: 707 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 708 splx(s); 709 free((caddr_t)nfsd, M_NFSD); 710 nsd->nsd_nfsd = (struct nfsd *)0; 711 if (--nfs_numnfsd == 0) 712 nfsrv_init(TRUE); /* Reinitialize everything */ 713 return (error); 714 } 715 716 /* 717 * Shut down a socket associated with an nfssvc_sock structure. 718 * Should be called with the send lock set, if required. 719 * The trick here is to increment the sref at the start, so that the nfsds 720 * will stop using it and clear ns_flag at the end so that it will not be 721 * reassigned during cleanup. 722 */ 723 static void 724 nfsrv_zapsock(struct nfssvc_sock *slp) 725 { 726 struct nfsuid *nuidp, *nnuidp; 727 struct nfsrv_descript *nwp, *nnwp; 728 struct socket *so; 729 struct file *fp; 730 struct nfsrv_rec *rec; 731 int s; 732 733 slp->ns_flag &= ~SLP_ALLFLAGS; 734 fp = slp->ns_fp; 735 if (fp) { 736 slp->ns_fp = (struct file *)0; 737 so = slp->ns_so; 738 so->so_rcv.sb_flags &= ~SB_UPCALL; 739 so->so_upcall = NULL; 740 so->so_upcallarg = NULL; 741 soshutdown(so, 2); 742 closef(fp, NULL); 743 if (slp->ns_nam) 744 FREE(slp->ns_nam, M_SONAME); 745 m_freem(slp->ns_raw); 746 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 747 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 748 if (rec->nr_address) 749 FREE(rec->nr_address, M_SONAME); 750 m_freem(rec->nr_packet); 751 free(rec, M_NFSRVDESC); 752 } 753 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 754 nuidp = nnuidp) { 755 nnuidp = nuidp->nu_lru.tqe_next; 756 LIST_REMOVE(nuidp, nu_hash); 757 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 758 if (nuidp->nu_flag & NU_NAM) 759 FREE(nuidp->nu_nam, M_SONAME); 760 free((caddr_t)nuidp, M_NFSUID); 761 } 762 s = splsoftclock(); 763 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 764 nnwp = nwp->nd_tq.le_next; 765 LIST_REMOVE(nwp, nd_tq); 766 free((caddr_t)nwp, M_NFSRVDESC); 767 } 768 LIST_INIT(&slp->ns_tq); 769 splx(s); 770 } 771 } 772 773 /* 774 * Derefence a server socket structure. If it has no more references and 775 * is no longer valid, you can throw it away. 776 */ 777 void 778 nfsrv_slpderef(struct nfssvc_sock *slp) 779 { 780 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 781 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 782 free((caddr_t)slp, M_NFSSVC); 783 } 784 } 785 786 /* 787 * Lock a socket against others. 788 */ 789 int 790 nfs_slplock(struct nfssvc_sock *slp, int wait) 791 { 792 int *statep = &slp->ns_solock; 793 794 if (!wait && (*statep & NFSSTA_SNDLOCK)) 795 return(0); /* already locked, fail */ 796 while (*statep & NFSSTA_SNDLOCK) { 797 *statep |= NFSSTA_WANTSND; 798 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0); 799 } 800 *statep |= NFSSTA_SNDLOCK; 801 return (1); 802 } 803 804 /* 805 * Unlock the stream socket for others. 806 */ 807 void 808 nfs_slpunlock(struct nfssvc_sock *slp) 809 { 810 int *statep = &slp->ns_solock; 811 812 if ((*statep & NFSSTA_SNDLOCK) == 0) 813 panic("nfs slpunlock"); 814 *statep &= ~NFSSTA_SNDLOCK; 815 if (*statep & NFSSTA_WANTSND) { 816 *statep &= ~NFSSTA_WANTSND; 817 wakeup((caddr_t)statep); 818 } 819 } 820 821 /* 822 * Initialize the data structures for the server. 823 * Handshake with any new nfsds starting up to avoid any chance of 824 * corruption. 825 */ 826 void 827 nfsrv_init(int terminating) 828 { 829 struct nfssvc_sock *slp, *nslp; 830 831 if (nfssvc_sockhead_flag & SLP_INIT) 832 panic("nfsd init"); 833 nfssvc_sockhead_flag |= SLP_INIT; 834 if (terminating) { 835 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 836 nslp = slp->ns_chain.tqe_next; 837 if (slp->ns_flag & SLP_VALID) 838 nfsrv_zapsock(slp); 839 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 840 free((caddr_t)slp, M_NFSSVC); 841 } 842 nfsrv_cleancache(); /* And clear out server cache */ 843 } else 844 nfs_pub.np_valid = 0; 845 846 TAILQ_INIT(&nfssvc_sockhead); 847 nfssvc_sockhead_flag &= ~SLP_INIT; 848 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 849 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 850 wakeup((caddr_t)&nfssvc_sockhead); 851 } 852 853 TAILQ_INIT(&nfsd_head); 854 nfsd_head_flag &= ~NFSD_CHECKSLP; 855 856 #if 0 857 nfs_udpsock = (struct nfssvc_sock *) 858 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 859 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 860 STAILQ_INIT(&nfs_udpsock->ns_rec); 861 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 862 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 863 864 nfs_cltpsock = (struct nfssvc_sock *) 865 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 866 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 867 STAILQ_INIT(&nfs_cltpsock->ns_rec); 868 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 869 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 870 #endif 871 } 872 873 /* 874 * Add entries to the server monitor log. 875 */ 876 static void 877 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 878 { 879 struct drt *rt; 880 881 rt = &nfsdrt.drt[nfsdrt.pos]; 882 if (cacherep == RC_DOIT) 883 rt->flag = 0; 884 else if (cacherep == RC_REPLY) 885 rt->flag = DRT_CACHEREPLY; 886 else 887 rt->flag = DRT_CACHEDROP; 888 if (sotype == SOCK_STREAM) 889 rt->flag |= DRT_TCP; 890 if (nd->nd_flag & ND_NQNFS) 891 rt->flag |= DRT_NQNFS; 892 else if (nd->nd_flag & ND_NFSV3) 893 rt->flag |= DRT_NFSV3; 894 rt->proc = nd->nd_procnum; 895 if (nd->nd_nam->sa_family == AF_INET) 896 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 897 else 898 rt->ipadr = INADDR_ANY; 899 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 900 getmicrotime(&rt->tstamp); 901 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 902 } 903 #endif /* NFS_NOSERVER */ 904 905 static int nfs_defect = 0; 906 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 907 908 /* 909 * Asynchronous I/O daemons for client nfs. 910 * They do read-ahead and write-behind operations on the block I/O cache. 911 * Never returns unless it fails or gets killed. 912 */ 913 static int 914 nfssvc_iod(struct thread *td) 915 { 916 struct buf *bp; 917 int i, myiod; 918 struct nfsmount *nmp; 919 int error = 0; 920 921 /* 922 * Assign my position or return error if too many already running 923 */ 924 myiod = -1; 925 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 926 if (nfs_asyncdaemon[i] == 0) { 927 nfs_asyncdaemon[i]++; 928 myiod = i; 929 break; 930 } 931 if (myiod == -1) 932 return (EBUSY); 933 nfs_numasync++; 934 /* 935 * Just loop around doin our stuff until SIGKILL 936 */ 937 for (;;) { 938 while (((nmp = nfs_iodmount[myiod]) == NULL 939 || nmp->nm_bufq.tqh_first == NULL) 940 && error == 0) { 941 if (nmp) 942 nmp->nm_bufqiods--; 943 nfs_iodwant[myiod] = td; 944 nfs_iodmount[myiod] = NULL; 945 error = tsleep((caddr_t)&nfs_iodwant[myiod], 946 PCATCH, "nfsidl", 0); 947 } 948 if (error) { 949 nfs_asyncdaemon[myiod] = 0; 950 if (nmp) 951 nmp->nm_bufqiods--; 952 nfs_iodwant[myiod] = NULL; 953 nfs_iodmount[myiod] = NULL; 954 nfs_numasync--; 955 return (error); 956 } 957 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 958 /* Take one off the front of the list */ 959 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 960 nmp->nm_bufqlen--; 961 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 962 nmp->nm_bufqwant = FALSE; 963 wakeup(&nmp->nm_bufq); 964 } 965 (void) nfs_doio(bp, NULL); 966 /* 967 * If there are more than one iod on this mount, then defect 968 * so that the iods can be shared out fairly between the mounts 969 */ 970 if (nfs_defect && nmp->nm_bufqiods > 1) { 971 NFS_DPF(ASYNCIO, 972 ("nfssvc_iod: iod %d defecting from mount %p\n", 973 myiod, nmp)); 974 nfs_iodmount[myiod] = NULL; 975 nmp->nm_bufqiods--; 976 break; 977 } 978 } 979 } 980 } 981 982 983 /* 984 * Get an authorization string for the uid by having the mount_nfs sitting 985 * on this mount point porpous out of the kernel and do it. 986 */ 987 int 988 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, 989 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str, 990 int *verf_len, NFSKERBKEY_T key /* return session key */) 991 { 992 int error = 0; 993 994 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 995 nmp->nm_state |= NFSSTA_WANTAUTH; 996 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 997 "nfsauth1", 2 * hz); 998 error = nfs_sigintr(nmp, rep, rep->r_td); 999 if (error) { 1000 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1001 return (error); 1002 } 1003 } 1004 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1005 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1006 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1007 nmp->nm_verfstr = verf_str; 1008 nmp->nm_verflen = *verf_len; 1009 nmp->nm_authuid = cred->cr_uid; 1010 wakeup((caddr_t)&nmp->nm_authstr); 1011 1012 /* 1013 * And wait for mount_nfs to do its stuff. 1014 */ 1015 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1016 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1017 "nfsauth2", 2 * hz); 1018 error = nfs_sigintr(nmp, rep, rep->r_td); 1019 } 1020 if (nmp->nm_state & NFSSTA_AUTHERR) { 1021 nmp->nm_state &= ~NFSSTA_AUTHERR; 1022 error = EAUTH; 1023 } 1024 if (error) 1025 free((caddr_t)*auth_str, M_TEMP); 1026 else { 1027 *auth_len = nmp->nm_authlen; 1028 *verf_len = nmp->nm_verflen; 1029 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1030 } 1031 nmp->nm_state &= ~NFSSTA_HASAUTH; 1032 nmp->nm_state |= NFSSTA_WAITAUTH; 1033 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1034 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1035 wakeup((caddr_t)&nmp->nm_authtype); 1036 } 1037 return (error); 1038 } 1039 1040 /* 1041 * Get a nickname authenticator and verifier. 1042 */ 1043 int 1044 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str, 1045 int *auth_len, char *verf_str, int verf_len) 1046 { 1047 struct nfsuid *nuidp; 1048 u_int32_t *nickp, *verfp; 1049 struct timeval ktvin, ktvout; 1050 1051 #ifdef DIAGNOSTIC 1052 if (verf_len < (4 * NFSX_UNSIGNED)) 1053 panic("nfs_getnickauth verf too small"); 1054 #endif 1055 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1056 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1057 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1058 break; 1059 } 1060 if (!nuidp || nuidp->nu_expire < time_second) 1061 return (EACCES); 1062 1063 /* 1064 * Move to the end of the lru list (end of lru == most recently used). 1065 */ 1066 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1067 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1068 1069 nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1070 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1071 *nickp = txdr_unsigned(nuidp->nu_nickname); 1072 *auth_str = (char *)nickp; 1073 *auth_len = 2 * NFSX_UNSIGNED; 1074 1075 /* 1076 * Now we must encrypt the verifier and package it up. 1077 */ 1078 verfp = (u_int32_t *)verf_str; 1079 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1080 if (time_second > nuidp->nu_timestamp.tv_sec || 1081 (time_second == nuidp->nu_timestamp.tv_sec && 1082 time_second > nuidp->nu_timestamp.tv_usec)) 1083 getmicrotime(&nuidp->nu_timestamp); 1084 else 1085 nuidp->nu_timestamp.tv_usec++; 1086 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1087 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1088 1089 /* 1090 * Now encrypt the timestamp verifier in ecb mode using the session 1091 * key. 1092 */ 1093 #ifdef NFSKERB 1094 XXX 1095 #endif 1096 1097 *verfp++ = ktvout.tv_sec; 1098 *verfp++ = ktvout.tv_usec; 1099 *verfp = 0; 1100 return (0); 1101 } 1102 1103 /* 1104 * Save the current nickname in a hash list entry on the mount point. 1105 */ 1106 int 1107 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len, 1108 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, 1109 struct mbuf *mrep) 1110 { 1111 struct nfsuid *nuidp; 1112 u_int32_t *tl; 1113 int32_t t1; 1114 struct mbuf *md = *mdp; 1115 struct timeval ktvin, ktvout; 1116 u_int32_t nick; 1117 char *dpos = *dposp, *cp2; 1118 int deltasec, error = 0; 1119 1120 if (len == (3 * NFSX_UNSIGNED)) { 1121 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1122 ktvin.tv_sec = *tl++; 1123 ktvin.tv_usec = *tl++; 1124 nick = fxdr_unsigned(u_int32_t, *tl); 1125 1126 /* 1127 * Decrypt the timestamp in ecb mode. 1128 */ 1129 #ifdef NFSKERB 1130 XXX 1131 #endif 1132 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1133 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1134 deltasec = time_second - ktvout.tv_sec; 1135 if (deltasec < 0) 1136 deltasec = -deltasec; 1137 /* 1138 * If ok, add it to the hash list for the mount point. 1139 */ 1140 if (deltasec <= NFS_KERBCLOCKSKEW) { 1141 if (nmp->nm_numuids < nuidhash_max) { 1142 nmp->nm_numuids++; 1143 nuidp = (struct nfsuid *) 1144 malloc(sizeof (struct nfsuid), M_NFSUID, 1145 M_WAITOK); 1146 } else { 1147 nuidp = nmp->nm_uidlruhead.tqh_first; 1148 LIST_REMOVE(nuidp, nu_hash); 1149 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1150 nu_lru); 1151 } 1152 nuidp->nu_flag = 0; 1153 nuidp->nu_cr.cr_uid = cred->cr_uid; 1154 nuidp->nu_expire = time_second + NFS_KERBTTL; 1155 nuidp->nu_timestamp = ktvout; 1156 nuidp->nu_nickname = nick; 1157 bcopy(key, nuidp->nu_key, sizeof (key)); 1158 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1159 nu_lru); 1160 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1161 nuidp, nu_hash); 1162 } 1163 } else 1164 nfsm_adv(nfsm_rndup(len)); 1165 nfsmout: 1166 *mdp = md; 1167 *dposp = dpos; 1168 return (error); 1169 } 1170