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 1.19 2004/11/12 00:09:37 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/nlookup.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 nlookupdata 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 struct vnode *vp; 151 #endif /* NFS_NOSERVER */ 152 int error; 153 struct thread *td = curthread; 154 155 /* 156 * Must be super user 157 */ 158 error = suser(td); 159 if(error) 160 return (error); 161 KKASSERT(td->td_proc); /* for ucred and p_fd */ 162 while (nfssvc_sockhead_flag & SLP_INIT) { 163 nfssvc_sockhead_flag |= SLP_WANTINIT; 164 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 165 } 166 if (uap->flag & NFSSVC_BIOD) 167 error = nfssvc_iod(td); 168 #ifdef NFS_NOSERVER 169 else 170 error = ENXIO; 171 #else /* !NFS_NOSERVER */ 172 else if (uap->flag & NFSSVC_MNTD) { 173 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 174 if (error) 175 return (error); 176 vp = NULL; 177 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 178 NLC_FOLLOW); 179 if (error == 0) 180 error = nlookup(&nd); 181 if (error == 0) 182 error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp); 183 nlookup_done(&nd); 184 if (error) 185 return (error); 186 187 if ((vp->v_flag & VROOT) == 0) 188 error = EINVAL; 189 nmp = VFSTONFS(vp->v_mount); 190 vput(vp); 191 if (error) 192 return (error); 193 if ((nmp->nm_state & NFSSTA_MNTD) && 194 (uap->flag & NFSSVC_GOTAUTH) == 0) 195 return (0); 196 nmp->nm_state |= NFSSTA_MNTD; 197 error = nqnfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag, 198 uap->argp, td); 199 } else if (uap->flag & NFSSVC_ADDSOCK) { 200 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 201 if (error) 202 return (error); 203 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 204 if (error) 205 return (error); 206 /* 207 * Get the client address for connected sockets. 208 */ 209 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 210 nam = (struct sockaddr *)0; 211 else { 212 error = getsockaddr(&nam, nfsdarg.name, 213 nfsdarg.namelen); 214 if (error) { 215 fdrop(fp, td); 216 return (error); 217 } 218 } 219 error = nfssvc_addsock(fp, nam, td); 220 fdrop(fp, td); 221 } else { 222 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 223 if (error) 224 return (error); 225 if ((uap->flag & NFSSVC_AUTHIN) && 226 ((nfsd = nsd->nsd_nfsd)) != NULL && 227 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 228 slp = nfsd->nfsd_slp; 229 230 /* 231 * First check to see if another nfsd has already 232 * added this credential. 233 */ 234 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 235 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 236 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 237 (!nfsd->nfsd_nd->nd_nam2 || 238 netaddr_match(NU_NETFAM(nuidp), 239 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 240 break; 241 } 242 if (nuidp) { 243 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 244 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 245 } else { 246 /* 247 * Nope, so we will. 248 */ 249 if (slp->ns_numuids < nuidhash_max) { 250 slp->ns_numuids++; 251 nuidp = (struct nfsuid *) 252 malloc(sizeof (struct nfsuid), M_NFSUID, 253 M_WAITOK); 254 } else 255 nuidp = (struct nfsuid *)0; 256 if ((slp->ns_flag & SLP_VALID) == 0) { 257 if (nuidp) 258 free((caddr_t)nuidp, M_NFSUID); 259 } else { 260 if (nuidp == (struct nfsuid *)0) { 261 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 262 LIST_REMOVE(nuidp, nu_hash); 263 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 264 nu_lru); 265 if (nuidp->nu_flag & NU_NAM) 266 FREE(nuidp->nu_nam, M_SONAME); 267 } 268 nuidp->nu_flag = 0; 269 nuidp->nu_cr = nsd->nsd_cr; 270 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 271 nuidp->nu_cr.cr_ngroups = NGROUPS; 272 nuidp->nu_cr.cr_ref = 1; 273 nuidp->nu_timestamp = nsd->nsd_timestamp; 274 nuidp->nu_expire = time_second + nsd->nsd_ttl; 275 /* 276 * and save the session key in nu_key. 277 */ 278 bcopy(nsd->nsd_key, nuidp->nu_key, 279 sizeof (nsd->nsd_key)); 280 if (nfsd->nfsd_nd->nd_nam2) { 281 struct sockaddr_in *saddr; 282 283 saddr = (struct sockaddr_in *) 284 nfsd->nfsd_nd->nd_nam2; 285 switch (saddr->sin_family) { 286 case AF_INET: 287 nuidp->nu_flag |= NU_INETADDR; 288 nuidp->nu_inetaddr = 289 saddr->sin_addr.s_addr; 290 break; 291 case AF_ISO: 292 default: 293 nuidp->nu_flag |= NU_NAM; 294 nuidp->nu_nam = 295 dup_sockaddr(nfsd->nfsd_nd->nd_nam2); 296 break; 297 }; 298 } 299 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 300 nu_lru); 301 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 302 nuidp, nu_hash); 303 nfsrv_setcred(&nuidp->nu_cr, 304 &nfsd->nfsd_nd->nd_cr); 305 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 306 } 307 } 308 } 309 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 310 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 311 error = nfssvc_nfsd(nsd, uap->argp, td); 312 } 313 #endif /* NFS_NOSERVER */ 314 if (error == EINTR || error == ERESTART) 315 error = 0; 316 return (error); 317 } 318 319 #ifndef NFS_NOSERVER 320 /* 321 * Adds a socket to the list for servicing by nfsds. 322 */ 323 static int 324 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td) 325 { 326 int siz; 327 struct nfssvc_sock *slp; 328 struct socket *so; 329 int error, s; 330 331 so = (struct socket *)fp->f_data; 332 #if 0 333 tslp = (struct nfssvc_sock *)0; 334 /* 335 * Add it to the list, as required. 336 */ 337 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 338 tslp = nfs_udpsock; 339 if (tslp->ns_flag & SLP_VALID) { 340 if (mynam != NULL) 341 FREE(mynam, M_SONAME); 342 return (EPERM); 343 } 344 } 345 #endif 346 /* 347 * Reserve buffer space in the socket. Note that due to bugs in 348 * Linux's delayed-ack code, serious performance degredation may 349 * occur with linux hosts if the minimum is used. 350 */ 351 if (so->so_type == SOCK_STREAM) 352 siz = NFS_MAXPACKET + sizeof (u_long); 353 else 354 siz = NFS_MAXPACKET; 355 if (siz < nfs_soreserve) 356 siz = nfs_soreserve; 357 if (siz > sb_max_adj) { 358 printf("Warning: vfs.nfs.soreserve (%d) " 359 "limited to adjusted sb_max (%ld)\n", 360 nfs_soreserve, sb_max_adj); 361 siz = sb_max_adj; 362 } 363 364 error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]); 365 if (error) { 366 if (mynam != NULL) 367 FREE(mynam, M_SONAME); 368 return (error); 369 } 370 371 /* 372 * Set protocol specific options { for now TCP only } and 373 * reserve some space. For datagram sockets, this can get called 374 * repeatedly for the same socket, but that isn't harmful. 375 */ 376 if (so->so_type == SOCK_STREAM) { 377 struct sockopt sopt; 378 int val; 379 380 bzero(&sopt, sizeof sopt); 381 sopt.sopt_level = SOL_SOCKET; 382 sopt.sopt_name = SO_KEEPALIVE; 383 sopt.sopt_val = &val; 384 sopt.sopt_valsize = sizeof val; 385 val = 1; 386 sosetopt(so, &sopt); 387 } 388 if (so->so_proto->pr_domain->dom_family == AF_INET && 389 so->so_proto->pr_protocol == IPPROTO_TCP) { 390 struct sockopt sopt; 391 int val; 392 393 bzero(&sopt, sizeof sopt); 394 sopt.sopt_level = IPPROTO_TCP; 395 sopt.sopt_name = TCP_NODELAY; 396 sopt.sopt_val = &val; 397 sopt.sopt_valsize = sizeof val; 398 val = 1; 399 sosetopt(so, &sopt); 400 } 401 so->so_rcv.sb_flags &= ~SB_NOINTR; 402 so->so_rcv.sb_timeo = 0; 403 so->so_snd.sb_flags &= ~SB_NOINTR; 404 so->so_snd.sb_timeo = 0; 405 406 slp = (struct nfssvc_sock *) 407 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 408 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 409 STAILQ_INIT(&slp->ns_rec); 410 TAILQ_INIT(&slp->ns_uidlruhead); 411 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 412 413 slp->ns_so = so; 414 slp->ns_nam = mynam; 415 fp->f_count++; 416 slp->ns_fp = fp; 417 s = splnet(); 418 so->so_upcallarg = (caddr_t)slp; 419 so->so_upcall = nfsrv_rcv; 420 so->so_rcv.sb_flags |= SB_UPCALL; 421 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 422 nfsrv_wakenfsd(slp); 423 splx(s); 424 return (0); 425 } 426 427 /* 428 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 429 * until it is killed by a signal. 430 */ 431 static int 432 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td) 433 { 434 int siz; 435 struct nfssvc_sock *slp; 436 struct nfsd *nfsd = nsd->nsd_nfsd; 437 struct nfsrv_descript *nd = NULL; 438 struct mbuf *m, *mreq; 439 int error = 0, cacherep, s, sotype, writes_todo; 440 int procrastinate; 441 u_quad_t cur_usec; 442 443 #ifndef nolint 444 cacherep = RC_DOIT; 445 writes_todo = 0; 446 #endif 447 if (nfsd == (struct nfsd *)0) { 448 nsd->nsd_nfsd = nfsd = (struct nfsd *) 449 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 450 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 451 s = splnet(); 452 nfsd->nfsd_td = td; 453 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 454 nfs_numnfsd++; 455 } else 456 s = splnet(); 457 458 /* 459 * Loop getting rpc requests until SIGKILL. 460 */ 461 for (;;) { 462 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 463 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 464 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 465 nfsd->nfsd_flag |= NFSD_WAITING; 466 nfsd_waiting++; 467 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0); 468 nfsd_waiting--; 469 if (error) 470 goto done; 471 } 472 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 473 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 474 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 475 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 476 == (SLP_VALID | SLP_DOREC)) { 477 slp->ns_flag &= ~SLP_DOREC; 478 slp->ns_sref++; 479 nfsd->nfsd_slp = slp; 480 break; 481 } 482 } 483 if (slp == 0) 484 nfsd_head_flag &= ~NFSD_CHECKSLP; 485 } 486 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 487 continue; 488 if (slp->ns_flag & SLP_VALID) { 489 if (slp->ns_flag & SLP_DISCONN) 490 nfsrv_zapsock(slp); 491 else if (slp->ns_flag & SLP_NEEDQ) { 492 slp->ns_flag &= ~SLP_NEEDQ; 493 (void) nfs_slplock(slp, 1); 494 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 495 MB_WAIT); 496 nfs_slpunlock(slp); 497 } 498 error = nfsrv_dorec(slp, nfsd, &nd); 499 cur_usec = nfs_curusec(); 500 if (error && slp->ns_tq.lh_first && 501 slp->ns_tq.lh_first->nd_time <= cur_usec) { 502 error = 0; 503 cacherep = RC_DOIT; 504 writes_todo = 1; 505 } else 506 writes_todo = 0; 507 nfsd->nfsd_flag |= NFSD_REQINPROG; 508 } 509 } else { 510 error = 0; 511 slp = nfsd->nfsd_slp; 512 } 513 if (error || (slp->ns_flag & SLP_VALID) == 0) { 514 if (nd) { 515 free((caddr_t)nd, M_NFSRVDESC); 516 nd = NULL; 517 } 518 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 519 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 520 nfsrv_slpderef(slp); 521 continue; 522 } 523 splx(s); 524 sotype = slp->ns_so->so_type; 525 if (nd) { 526 getmicrotime(&nd->nd_starttime); 527 if (nd->nd_nam2) 528 nd->nd_nam = nd->nd_nam2; 529 else 530 nd->nd_nam = slp->ns_nam; 531 532 /* 533 * Check to see if authorization is needed. 534 */ 535 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 536 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 537 nsd->nsd_haddr = 538 ((struct sockaddr_in *) 539 nd->nd_nam)->sin_addr.s_addr; 540 nsd->nsd_authlen = nfsd->nfsd_authlen; 541 nsd->nsd_verflen = nfsd->nfsd_verflen; 542 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 543 nfsd->nfsd_authlen) && 544 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 545 nfsd->nfsd_verflen) && 546 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 547 return (ENEEDAUTH); 548 cacherep = RC_DROPIT; 549 } else 550 cacherep = nfsrv_getcache(nd, slp, &mreq); 551 552 /* 553 * Check for just starting up for NQNFS and send 554 * fake "try again later" replies to the NQNFS clients. 555 */ 556 if (notstarted && nqnfsstarttime <= time_second) { 557 if (modify_flag) { 558 nqnfsstarttime = time_second + nqsrv_writeslack; 559 modify_flag = 0; 560 } else 561 notstarted = 0; 562 } 563 if (notstarted) { 564 if ((nd->nd_flag & ND_NQNFS) == 0) 565 cacherep = RC_DROPIT; 566 else if (nd->nd_procnum != NFSPROC_WRITE) { 567 nd->nd_procnum = NFSPROC_NOOP; 568 nd->nd_repstat = NQNFS_TRYLATER; 569 cacherep = RC_DOIT; 570 } else 571 modify_flag = 1; 572 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 573 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 574 nd->nd_procnum = NFSPROC_NOOP; 575 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 576 cacherep = RC_DOIT; 577 } else if (nfs_privport) { 578 /* Check if source port is privileged */ 579 u_short port; 580 struct sockaddr *nam = nd->nd_nam; 581 struct sockaddr_in *sin; 582 583 sin = (struct sockaddr_in *)nam; 584 port = ntohs(sin->sin_port); 585 if (port >= IPPORT_RESERVED && 586 nd->nd_procnum != NFSPROC_NULL) { 587 nd->nd_procnum = NFSPROC_NOOP; 588 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 589 cacherep = RC_DOIT; 590 printf("NFS request from unprivileged port (%s:%d)\n", 591 inet_ntoa(sin->sin_addr), port); 592 } 593 } 594 595 } 596 597 /* 598 * Loop to get all the write rpc relies that have been 599 * gathered together. 600 */ 601 do { 602 switch (cacherep) { 603 case RC_DOIT: 604 if (nd && (nd->nd_flag & ND_NFSV3)) 605 procrastinate = nfsrvw_procrastinate_v3; 606 else 607 procrastinate = nfsrvw_procrastinate; 608 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 609 procrastinate > 0 && !notstarted)) 610 error = nfsrv_writegather(&nd, slp, 611 nfsd->nfsd_td, &mreq); 612 else 613 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 614 slp, nfsd->nfsd_td, &mreq); 615 if (mreq == NULL) 616 break; 617 if (error != 0 && error != NFSERR_RETVOID) { 618 if (nd->nd_procnum != NQNFSPROC_VACATED) 619 nfsstats.srv_errs++; 620 nfsrv_updatecache(nd, FALSE, mreq); 621 if (nd->nd_nam2) 622 FREE(nd->nd_nam2, M_SONAME); 623 break; 624 } 625 nfsstats.srvrpccnt[nd->nd_procnum]++; 626 nfsrv_updatecache(nd, TRUE, mreq); 627 nd->nd_mrep = (struct mbuf *)0; 628 case RC_REPLY: 629 m = mreq; 630 siz = 0; 631 while (m) { 632 siz += m->m_len; 633 m = m->m_next; 634 } 635 if (siz <= 0 || siz > NFS_MAXPACKET) { 636 printf("mbuf siz=%d\n",siz); 637 panic("Bad nfs svc reply"); 638 } 639 m = mreq; 640 m->m_pkthdr.len = siz; 641 m->m_pkthdr.rcvif = (struct ifnet *)0; 642 /* 643 * For stream protocols, prepend a Sun RPC 644 * Record Mark. 645 */ 646 if (sotype == SOCK_STREAM) { 647 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT); 648 if (m == NULL) 649 return (ENOBUFS); 650 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 651 } 652 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 653 (void) nfs_slplock(slp, 1); 654 if (slp->ns_flag & SLP_VALID) 655 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 656 else { 657 error = EPIPE; 658 m_freem(m); 659 } 660 if (nfsrtton) 661 nfsd_rt(sotype, nd, cacherep); 662 if (nd->nd_nam2) 663 FREE(nd->nd_nam2, M_SONAME); 664 if (nd->nd_mrep) 665 m_freem(nd->nd_mrep); 666 if (error == EPIPE) 667 nfsrv_zapsock(slp); 668 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 669 nfs_slpunlock(slp); 670 if (error == EINTR || error == ERESTART) { 671 free((caddr_t)nd, M_NFSRVDESC); 672 nfsrv_slpderef(slp); 673 s = splnet(); 674 goto done; 675 } 676 break; 677 case RC_DROPIT: 678 if (nfsrtton) 679 nfsd_rt(sotype, nd, cacherep); 680 m_freem(nd->nd_mrep); 681 if (nd->nd_nam2) 682 FREE(nd->nd_nam2, M_SONAME); 683 break; 684 }; 685 if (nd) { 686 FREE((caddr_t)nd, M_NFSRVDESC); 687 nd = NULL; 688 } 689 690 /* 691 * Check to see if there are outstanding writes that 692 * need to be serviced. 693 */ 694 cur_usec = nfs_curusec(); 695 s = splsoftclock(); 696 if (slp->ns_tq.lh_first && 697 slp->ns_tq.lh_first->nd_time <= cur_usec) { 698 cacherep = RC_DOIT; 699 writes_todo = 1; 700 } else 701 writes_todo = 0; 702 splx(s); 703 } while (writes_todo); 704 s = splnet(); 705 if (nfsrv_dorec(slp, nfsd, &nd)) { 706 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 707 nfsd->nfsd_slp = NULL; 708 nfsrv_slpderef(slp); 709 } 710 } 711 done: 712 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 713 splx(s); 714 free((caddr_t)nfsd, M_NFSD); 715 nsd->nsd_nfsd = (struct nfsd *)0; 716 if (--nfs_numnfsd == 0) 717 nfsrv_init(TRUE); /* Reinitialize everything */ 718 return (error); 719 } 720 721 /* 722 * Shut down a socket associated with an nfssvc_sock structure. 723 * Should be called with the send lock set, if required. 724 * The trick here is to increment the sref at the start, so that the nfsds 725 * will stop using it and clear ns_flag at the end so that it will not be 726 * reassigned during cleanup. 727 */ 728 static void 729 nfsrv_zapsock(struct nfssvc_sock *slp) 730 { 731 struct nfsuid *nuidp, *nnuidp; 732 struct nfsrv_descript *nwp, *nnwp; 733 struct socket *so; 734 struct file *fp; 735 struct nfsrv_rec *rec; 736 int s; 737 738 slp->ns_flag &= ~SLP_ALLFLAGS; 739 fp = slp->ns_fp; 740 if (fp) { 741 slp->ns_fp = (struct file *)0; 742 so = slp->ns_so; 743 so->so_rcv.sb_flags &= ~SB_UPCALL; 744 so->so_upcall = NULL; 745 so->so_upcallarg = NULL; 746 soshutdown(so, 2); 747 closef(fp, NULL); 748 if (slp->ns_nam) 749 FREE(slp->ns_nam, M_SONAME); 750 m_freem(slp->ns_raw); 751 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 752 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 753 if (rec->nr_address) 754 FREE(rec->nr_address, M_SONAME); 755 m_freem(rec->nr_packet); 756 free(rec, M_NFSRVDESC); 757 } 758 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru, 759 nnuidp) { 760 LIST_REMOVE(nuidp, nu_hash); 761 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 762 if (nuidp->nu_flag & NU_NAM) 763 FREE(nuidp->nu_nam, M_SONAME); 764 free((caddr_t)nuidp, M_NFSUID); 765 } 766 s = splsoftclock(); 767 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 768 nnwp = nwp->nd_tq.le_next; 769 LIST_REMOVE(nwp, nd_tq); 770 free((caddr_t)nwp, M_NFSRVDESC); 771 } 772 LIST_INIT(&slp->ns_tq); 773 splx(s); 774 } 775 } 776 777 /* 778 * Derefence a server socket structure. If it has no more references and 779 * is no longer valid, you can throw it away. 780 */ 781 void 782 nfsrv_slpderef(struct nfssvc_sock *slp) 783 { 784 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 785 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 786 free((caddr_t)slp, M_NFSSVC); 787 } 788 } 789 790 /* 791 * Lock a socket against others. 792 */ 793 int 794 nfs_slplock(struct nfssvc_sock *slp, int wait) 795 { 796 int *statep = &slp->ns_solock; 797 798 if (!wait && (*statep & NFSSTA_SNDLOCK)) 799 return(0); /* already locked, fail */ 800 while (*statep & NFSSTA_SNDLOCK) { 801 *statep |= NFSSTA_WANTSND; 802 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0); 803 } 804 *statep |= NFSSTA_SNDLOCK; 805 return (1); 806 } 807 808 /* 809 * Unlock the stream socket for others. 810 */ 811 void 812 nfs_slpunlock(struct nfssvc_sock *slp) 813 { 814 int *statep = &slp->ns_solock; 815 816 if ((*statep & NFSSTA_SNDLOCK) == 0) 817 panic("nfs slpunlock"); 818 *statep &= ~NFSSTA_SNDLOCK; 819 if (*statep & NFSSTA_WANTSND) { 820 *statep &= ~NFSSTA_WANTSND; 821 wakeup((caddr_t)statep); 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(int terminating) 832 { 833 struct nfssvc_sock *slp, *nslp; 834 835 if (nfssvc_sockhead_flag & SLP_INIT) 836 panic("nfsd init"); 837 nfssvc_sockhead_flag |= SLP_INIT; 838 if (terminating) { 839 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) { 840 if (slp->ns_flag & SLP_VALID) 841 nfsrv_zapsock(slp); 842 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 843 free((caddr_t)slp, M_NFSSVC); 844 } 845 nfsrv_cleancache(); /* And clear out server cache */ 846 } else 847 nfs_pub.np_valid = 0; 848 849 TAILQ_INIT(&nfssvc_sockhead); 850 nfssvc_sockhead_flag &= ~SLP_INIT; 851 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 852 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 853 wakeup((caddr_t)&nfssvc_sockhead); 854 } 855 856 TAILQ_INIT(&nfsd_head); 857 nfsd_head_flag &= ~NFSD_CHECKSLP; 858 859 #if 0 860 nfs_udpsock = (struct nfssvc_sock *) 861 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 862 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 863 STAILQ_INIT(&nfs_udpsock->ns_rec); 864 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 865 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 866 867 nfs_cltpsock = (struct nfssvc_sock *) 868 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 869 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 870 STAILQ_INIT(&nfs_cltpsock->ns_rec); 871 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 872 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 873 #endif 874 } 875 876 /* 877 * Add entries to the server monitor log. 878 */ 879 static void 880 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 881 { 882 struct drt *rt; 883 884 rt = &nfsdrt.drt[nfsdrt.pos]; 885 if (cacherep == RC_DOIT) 886 rt->flag = 0; 887 else if (cacherep == RC_REPLY) 888 rt->flag = DRT_CACHEREPLY; 889 else 890 rt->flag = DRT_CACHEDROP; 891 if (sotype == SOCK_STREAM) 892 rt->flag |= DRT_TCP; 893 if (nd->nd_flag & ND_NQNFS) 894 rt->flag |= DRT_NQNFS; 895 else if (nd->nd_flag & ND_NFSV3) 896 rt->flag |= DRT_NFSV3; 897 rt->proc = nd->nd_procnum; 898 if (nd->nd_nam->sa_family == AF_INET) 899 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 900 else 901 rt->ipadr = INADDR_ANY; 902 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 903 getmicrotime(&rt->tstamp); 904 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 905 } 906 #endif /* NFS_NOSERVER */ 907 908 static int nfs_defect = 0; 909 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 910 911 /* 912 * Asynchronous I/O daemons for client nfs. 913 * They do read-ahead and write-behind operations on the block I/O cache. 914 * Never returns unless it fails or gets killed. 915 */ 916 static int 917 nfssvc_iod(struct thread *td) 918 { 919 struct buf *bp; 920 int i, myiod; 921 struct nfsmount *nmp; 922 int error = 0; 923 924 /* 925 * Assign my position or return error if too many already running 926 */ 927 myiod = -1; 928 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 929 if (nfs_asyncdaemon[i] == 0) { 930 nfs_asyncdaemon[i]++; 931 myiod = i; 932 break; 933 } 934 if (myiod == -1) 935 return (EBUSY); 936 nfs_numasync++; 937 /* 938 * Just loop around doin our stuff until SIGKILL 939 */ 940 for (;;) { 941 while (((nmp = nfs_iodmount[myiod]) == NULL 942 || TAILQ_EMPTY(&nmp->nm_bufq)) 943 && error == 0) { 944 if (nmp) 945 nmp->nm_bufqiods--; 946 nfs_iodwant[myiod] = td; 947 nfs_iodmount[myiod] = NULL; 948 error = tsleep((caddr_t)&nfs_iodwant[myiod], 949 PCATCH, "nfsidl", 0); 950 } 951 if (error) { 952 nfs_asyncdaemon[myiod] = 0; 953 if (nmp) 954 nmp->nm_bufqiods--; 955 nfs_iodwant[myiod] = NULL; 956 nfs_iodmount[myiod] = NULL; 957 nfs_numasync--; 958 return (error); 959 } 960 while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) { 961 /* Take one off the front of the list */ 962 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 963 nmp->nm_bufqlen--; 964 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 965 nmp->nm_bufqwant = FALSE; 966 wakeup(&nmp->nm_bufq); 967 } 968 (void) nfs_doio(bp, NULL); 969 /* 970 * If there are more than one iod on this mount, then defect 971 * so that the iods can be shared out fairly between the mounts 972 */ 973 if (nfs_defect && nmp->nm_bufqiods > 1) { 974 NFS_DPF(ASYNCIO, 975 ("nfssvc_iod: iod %d defecting from mount %p\n", 976 myiod, nmp)); 977 nfs_iodmount[myiod] = NULL; 978 nmp->nm_bufqiods--; 979 break; 980 } 981 } 982 } 983 } 984 985 986 /* 987 * Get an authorization string for the uid by having the mount_nfs sitting 988 * on this mount point porpous out of the kernel and do it. 989 */ 990 int 991 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, 992 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str, 993 int *verf_len, NFSKERBKEY_T key /* return session key */) 994 { 995 int error = 0; 996 997 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 998 nmp->nm_state |= NFSSTA_WANTAUTH; 999 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 1000 "nfsauth1", 2 * hz); 1001 error = nfs_sigintr(nmp, rep, rep->r_td); 1002 if (error) { 1003 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1004 return (error); 1005 } 1006 } 1007 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1008 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1009 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1010 nmp->nm_verfstr = verf_str; 1011 nmp->nm_verflen = *verf_len; 1012 nmp->nm_authuid = cred->cr_uid; 1013 wakeup((caddr_t)&nmp->nm_authstr); 1014 1015 /* 1016 * And wait for mount_nfs to do its stuff. 1017 */ 1018 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1019 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1020 "nfsauth2", 2 * hz); 1021 error = nfs_sigintr(nmp, rep, rep->r_td); 1022 } 1023 if (nmp->nm_state & NFSSTA_AUTHERR) { 1024 nmp->nm_state &= ~NFSSTA_AUTHERR; 1025 error = EAUTH; 1026 } 1027 if (error) 1028 free((caddr_t)*auth_str, M_TEMP); 1029 else { 1030 *auth_len = nmp->nm_authlen; 1031 *verf_len = nmp->nm_verflen; 1032 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1033 } 1034 nmp->nm_state &= ~NFSSTA_HASAUTH; 1035 nmp->nm_state |= NFSSTA_WAITAUTH; 1036 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1037 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1038 wakeup((caddr_t)&nmp->nm_authtype); 1039 } 1040 return (error); 1041 } 1042 1043 /* 1044 * Get a nickname authenticator and verifier. 1045 */ 1046 int 1047 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str, 1048 int *auth_len, char *verf_str, int verf_len) 1049 { 1050 struct nfsuid *nuidp; 1051 u_int32_t *nickp, *verfp; 1052 struct timeval ktvin, ktvout; 1053 1054 #ifdef DIAGNOSTIC 1055 if (verf_len < (4 * NFSX_UNSIGNED)) 1056 panic("nfs_getnickauth verf too small"); 1057 #endif 1058 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1059 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1060 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1061 break; 1062 } 1063 if (!nuidp || nuidp->nu_expire < time_second) 1064 return (EACCES); 1065 1066 /* 1067 * Move to the end of the lru list (end of lru == most recently used). 1068 */ 1069 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1070 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1071 1072 nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1073 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1074 *nickp = txdr_unsigned(nuidp->nu_nickname); 1075 *auth_str = (char *)nickp; 1076 *auth_len = 2 * NFSX_UNSIGNED; 1077 1078 /* 1079 * Now we must encrypt the verifier and package it up. 1080 */ 1081 verfp = (u_int32_t *)verf_str; 1082 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1083 if (time_second > nuidp->nu_timestamp.tv_sec || 1084 (time_second == nuidp->nu_timestamp.tv_sec && 1085 time_second > nuidp->nu_timestamp.tv_usec)) 1086 getmicrotime(&nuidp->nu_timestamp); 1087 else 1088 nuidp->nu_timestamp.tv_usec++; 1089 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1090 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1091 1092 /* 1093 * Now encrypt the timestamp verifier in ecb mode using the session 1094 * key. 1095 */ 1096 #ifdef NFSKERB 1097 XXX 1098 #endif 1099 1100 *verfp++ = ktvout.tv_sec; 1101 *verfp++ = ktvout.tv_usec; 1102 *verfp = 0; 1103 return (0); 1104 } 1105 1106 /* 1107 * Save the current nickname in a hash list entry on the mount point. 1108 */ 1109 int 1110 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len, 1111 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, 1112 struct mbuf *mrep) 1113 { 1114 struct nfsuid *nuidp; 1115 u_int32_t *tl; 1116 int32_t t1; 1117 struct mbuf *md = *mdp; 1118 struct timeval ktvin, ktvout; 1119 u_int32_t nick; 1120 char *dpos = *dposp, *cp2; 1121 int deltasec, error = 0; 1122 1123 if (len == (3 * NFSX_UNSIGNED)) { 1124 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1125 ktvin.tv_sec = *tl++; 1126 ktvin.tv_usec = *tl++; 1127 nick = fxdr_unsigned(u_int32_t, *tl); 1128 1129 /* 1130 * Decrypt the timestamp in ecb mode. 1131 */ 1132 #ifdef NFSKERB 1133 XXX 1134 #endif 1135 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1136 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1137 deltasec = time_second - ktvout.tv_sec; 1138 if (deltasec < 0) 1139 deltasec = -deltasec; 1140 /* 1141 * If ok, add it to the hash list for the mount point. 1142 */ 1143 if (deltasec <= NFS_KERBCLOCKSKEW) { 1144 if (nmp->nm_numuids < nuidhash_max) { 1145 nmp->nm_numuids++; 1146 nuidp = (struct nfsuid *) 1147 malloc(sizeof (struct nfsuid), M_NFSUID, 1148 M_WAITOK); 1149 } else { 1150 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead); 1151 LIST_REMOVE(nuidp, nu_hash); 1152 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1153 nu_lru); 1154 } 1155 nuidp->nu_flag = 0; 1156 nuidp->nu_cr.cr_uid = cred->cr_uid; 1157 nuidp->nu_expire = time_second + NFS_KERBTTL; 1158 nuidp->nu_timestamp = ktvout; 1159 nuidp->nu_nickname = nick; 1160 bcopy(key, nuidp->nu_key, sizeof (key)); 1161 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1162 nu_lru); 1163 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1164 nuidp, nu_hash); 1165 } 1166 } else 1167 nfsm_adv(nfsm_rndup(len)); 1168 nfsmout: 1169 *mdp = md; 1170 *dposp = dpos; 1171 return (error); 1172 } 1173