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 1.31 2008/01/05 14:02:41 swildner 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/priv.h> 53 #include <sys/buf.h> 54 #include <sys/mbuf.h> 55 #include <sys/resourcevar.h> 56 #include <sys/socket.h> 57 #include <sys/socketvar.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/nlookup.h> 61 62 #include <sys/mutex2.h> 63 #include <sys/thread2.h> 64 65 #include <netinet/in.h> 66 #include <netinet/tcp.h> 67 #include "xdr_subs.h" 68 #include "rpcv2.h" 69 #include "nfsproto.h" 70 #include "nfs.h" 71 #include "nfsm_subs.h" 72 #include "nfsrvcache.h" 73 #include "nfsmount.h" 74 #include "nfsnode.h" 75 #include "nfsrtt.h" 76 77 #include <sys/thread2.h> 78 79 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 80 81 static int nuidhash_max = NFS_MAXUIDHASH; 82 83 #ifndef NFS_NOSERVER 84 static void nfsrv_zapsock (struct nfssvc_sock *slp); 85 #endif 86 87 #define TRUE 1 88 #define FALSE 0 89 90 SYSCTL_DECL(_vfs_nfs); 91 92 #ifndef NFS_NOSERVER 93 int nfsd_waiting = 0; 94 static struct nfsdrt nfsdrt; 95 static int nfs_numnfsd = 0; 96 static void nfsd_rt (int sotype, struct nfsrv_descript *nd, 97 int cacherep); 98 static int nfssvc_addsock (struct file *, struct sockaddr *, 99 struct thread *); 100 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *); 101 102 static int nfs_privport = 0; 103 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 104 0, "Enable privileged source port checks"); 105 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, 106 "Enable NFS request procrastination"); 107 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, 108 "Enable NFSv3 request procrastination"); 109 int nfs_soreserve = NFS_MAXPACKET * NFS_MAXASYNCBIO; 110 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, 111 "Minimum NFS socket buffer size reservation"); 112 113 /* 114 * NFS server system calls 115 */ 116 117 #endif /* NFS_NOSERVER */ 118 /* 119 * nfssvc_args(int flag, caddr_t argp) 120 * 121 * Nfs server psuedo system call for the nfsd's 122 * Based on the flag value it either: 123 * - adds a socket to the selection list 124 * - remains in the kernel as an nfsd 125 * - remains in the kernel as an nfsiod 126 * 127 * MPALMOSTSAFE 128 */ 129 int 130 sys_nfssvc(struct nfssvc_args *uap) 131 { 132 #ifndef NFS_NOSERVER 133 struct nlookupdata nd; 134 struct file *fp; 135 struct sockaddr *nam; 136 struct nfsd_args nfsdarg; 137 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 138 struct nfsd_cargs ncd; 139 struct nfsd *nfsd; 140 struct nfssvc_sock *slp; 141 struct nfsuid *nuidp; 142 struct nfsmount *nmp; 143 struct vnode *vp; 144 #endif /* NFS_NOSERVER */ 145 int error; 146 struct thread *td = curthread; 147 148 /* 149 * Must be super user 150 */ 151 error = priv_check(td, PRIV_ROOT); 152 if (error) 153 return (error); 154 155 lwkt_gettoken(&nfs_token); 156 157 while (nfssvc_sockhead_flag & SLP_INIT) { 158 nfssvc_sockhead_flag |= SLP_WANTINIT; 159 tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 160 } 161 if (uap->flag & NFSSVC_BIOD) 162 error = ENXIO; /* no longer need nfsiod's */ 163 #ifdef NFS_NOSERVER 164 else 165 error = ENXIO; 166 #else /* !NFS_NOSERVER */ 167 else if (uap->flag & NFSSVC_MNTD) { 168 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 169 if (error) 170 goto done; 171 vp = NULL; 172 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 173 NLC_FOLLOW); 174 if (error == 0) 175 error = nlookup(&nd); 176 if (error == 0) 177 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 178 nlookup_done(&nd); 179 if (error) 180 goto done; 181 182 if ((vp->v_flag & VROOT) == 0) 183 error = EINVAL; 184 nmp = VFSTONFS(vp->v_mount); 185 vput(vp); 186 if (error) 187 goto done; 188 if ((nmp->nm_state & NFSSTA_MNTD) && 189 (uap->flag & NFSSVC_GOTAUTH) == 0) { 190 error = 0; 191 goto done; 192 } 193 nmp->nm_state |= NFSSTA_MNTD; 194 error = nfs_clientd(nmp, td->td_ucred, &ncd, uap->flag, 195 uap->argp, td); 196 } else if (uap->flag & NFSSVC_ADDSOCK) { 197 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 198 if (error) 199 goto done; 200 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 201 if (error) 202 goto done; 203 /* 204 * Get the client address for connected sockets. 205 */ 206 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 207 nam = NULL; 208 else { 209 error = getsockaddr(&nam, nfsdarg.name, 210 nfsdarg.namelen); 211 if (error) { 212 fdrop(fp); 213 goto done; 214 } 215 } 216 error = nfssvc_addsock(fp, nam, td); 217 fdrop(fp); 218 } else { 219 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 220 if (error) 221 goto done; 222 if ((uap->flag & NFSSVC_AUTHIN) && 223 ((nfsd = nsd->nsd_nfsd)) != NULL && 224 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 225 slp = nfsd->nfsd_slp; 226 227 /* 228 * First check to see if another nfsd has already 229 * added this credential. 230 */ 231 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 232 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 233 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 234 (!nfsd->nfsd_nd->nd_nam2 || 235 netaddr_match(NU_NETFAM(nuidp), 236 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 237 break; 238 } 239 if (nuidp) { 240 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 241 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 242 } else { 243 /* 244 * Nope, so we will. 245 */ 246 if (slp->ns_numuids < nuidhash_max) { 247 slp->ns_numuids++; 248 nuidp = (struct nfsuid *) 249 kmalloc(sizeof (struct nfsuid), M_NFSUID, 250 M_WAITOK); 251 } else 252 nuidp = NULL; 253 if ((slp->ns_flag & SLP_VALID) == 0) { 254 if (nuidp) 255 kfree((caddr_t)nuidp, M_NFSUID); 256 } else { 257 if (nuidp == NULL) { 258 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 259 LIST_REMOVE(nuidp, nu_hash); 260 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 261 nu_lru); 262 if (nuidp->nu_flag & NU_NAM) 263 FREE(nuidp->nu_nam, M_SONAME); 264 } 265 nuidp->nu_flag = 0; 266 nuidp->nu_cr = nsd->nsd_cr; 267 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 268 nuidp->nu_cr.cr_ngroups = NGROUPS; 269 nuidp->nu_cr.cr_ref = 1; 270 nuidp->nu_timestamp = nsd->nsd_timestamp; 271 nuidp->nu_expire = time_second + nsd->nsd_ttl; 272 /* 273 * and save the session key in nu_key. 274 */ 275 bcopy(nsd->nsd_key, nuidp->nu_key, 276 sizeof (nsd->nsd_key)); 277 if (nfsd->nfsd_nd->nd_nam2) { 278 struct sockaddr_in *saddr; 279 280 saddr = (struct sockaddr_in *) 281 nfsd->nfsd_nd->nd_nam2; 282 switch (saddr->sin_family) { 283 case AF_INET: 284 nuidp->nu_flag |= NU_INETADDR; 285 nuidp->nu_inetaddr = 286 saddr->sin_addr.s_addr; 287 break; 288 case AF_ISO: 289 default: 290 nuidp->nu_flag |= NU_NAM; 291 nuidp->nu_nam = 292 dup_sockaddr(nfsd->nfsd_nd->nd_nam2); 293 break; 294 }; 295 } 296 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 297 nu_lru); 298 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 299 nuidp, nu_hash); 300 nfsrv_setcred(&nuidp->nu_cr, 301 &nfsd->nfsd_nd->nd_cr); 302 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 303 } 304 } 305 } 306 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 307 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 308 error = nfssvc_nfsd(nsd, uap->argp, td); 309 } 310 #endif /* NFS_NOSERVER */ 311 if (error == EINTR || error == ERESTART) 312 error = 0; 313 done: 314 lwkt_reltoken(&nfs_token); 315 return (error); 316 } 317 318 #ifndef NFS_NOSERVER 319 /* 320 * Adds a socket to the list for servicing by nfsds. 321 */ 322 static int 323 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td) 324 { 325 int siz; 326 struct nfssvc_sock *slp; 327 struct socket *so; 328 int error; 329 330 so = (struct socket *)fp->f_data; 331 #if 0 332 tslp = NULL; 333 /* 334 * Add it to the list, as required. 335 */ 336 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 337 tslp = nfs_udpsock; 338 if (tslp->ns_flag & SLP_VALID) { 339 if (mynam != NULL) 340 FREE(mynam, M_SONAME); 341 return (EPERM); 342 } 343 } 344 #endif 345 /* 346 * Reserve buffer space in the socket. Note that due to bugs in 347 * Linux's delayed-ack code, serious performance degredation may 348 * occur with linux hosts if the minimum is used. 349 * 350 * NFS sockets are not limited to the standard sb_max or by 351 * resource limits. 352 */ 353 if (so->so_type == SOCK_STREAM) 354 siz = NFS_MAXPACKET + sizeof (u_long); 355 else 356 siz = NFS_MAXPACKET; 357 if (siz < nfs_soreserve) 358 siz = nfs_soreserve; 359 360 error = soreserve(so, siz, siz, NULL); 361 if (error) { 362 if (mynam != NULL) 363 FREE(mynam, M_SONAME); 364 return (error); 365 } 366 367 /* 368 * Set protocol specific options { for now TCP only } and 369 * reserve some space. For datagram sockets, this can get called 370 * repeatedly for the same socket, but that isn't harmful. 371 */ 372 if (so->so_type == SOCK_STREAM) { 373 struct sockopt sopt; 374 int val; 375 376 bzero(&sopt, sizeof sopt); 377 sopt.sopt_level = SOL_SOCKET; 378 sopt.sopt_name = SO_KEEPALIVE; 379 sopt.sopt_val = &val; 380 sopt.sopt_valsize = sizeof val; 381 val = 1; 382 sosetopt(so, &sopt); 383 } 384 if (so->so_proto->pr_domain->dom_family == AF_INET && 385 so->so_proto->pr_protocol == IPPROTO_TCP) { 386 struct sockopt sopt; 387 int val; 388 389 bzero(&sopt, sizeof sopt); 390 sopt.sopt_level = IPPROTO_TCP; 391 sopt.sopt_name = TCP_NODELAY; 392 sopt.sopt_val = &val; 393 sopt.sopt_valsize = sizeof val; 394 val = 1; 395 sosetopt(so, &sopt); 396 397 bzero(&sopt, sizeof sopt); 398 sopt.sopt_level = IPPROTO_TCP; 399 sopt.sopt_name = TCP_FASTKEEP; 400 sopt.sopt_val = &val; 401 sopt.sopt_valsize = sizeof val; 402 val = 1; 403 sosetopt(so, &sopt); 404 } 405 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_NOINTR); 406 so->so_rcv.ssb_timeo = 0; 407 atomic_clear_int(&so->so_snd.ssb_flags, SSB_NOINTR); 408 so->so_snd.ssb_timeo = 0; 409 410 slp = kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 411 mtx_init(&slp->ns_solock); 412 STAILQ_INIT(&slp->ns_rec); 413 TAILQ_INIT(&slp->ns_uidlruhead); 414 lwkt_token_init(&slp->ns_token, "nfssrv_token"); 415 416 lwkt_gettoken(&nfs_token); 417 nfsrv_slpref(slp); 418 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 419 lwkt_gettoken(&slp->ns_token); 420 421 slp->ns_so = so; 422 slp->ns_nam = mynam; 423 fp->f_count++; 424 slp->ns_fp = fp; 425 426 so->so_upcallarg = (caddr_t)slp; 427 so->so_upcall = nfsrv_rcv_upcall; 428 atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 429 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 430 nfsrv_wakenfsd(slp, 1); 431 432 lwkt_reltoken(&slp->ns_token); 433 lwkt_reltoken(&nfs_token); 434 435 return (0); 436 } 437 438 /* 439 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 440 * until it is killed by a signal. 441 */ 442 static int 443 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td) 444 { 445 int siz; 446 struct nfssvc_sock *slp; 447 struct nfsd *nfsd = nsd->nsd_nfsd; 448 struct nfsrv_descript *nd = NULL; 449 struct mbuf *m, *mreq; 450 int error, cacherep, sotype, writes_todo; 451 int procrastinate; 452 u_quad_t cur_usec; 453 454 #ifndef nolint 455 cacherep = RC_DOIT; 456 writes_todo = 0; 457 #endif 458 lwkt_gettoken(&nfs_token); 459 460 if (nfsd == NULL) { 461 nsd->nsd_nfsd = nfsd = (struct nfsd *) 462 kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO); 463 nfsd->nfsd_td = td; 464 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 465 nfs_numnfsd++; 466 } 467 468 /* 469 * Loop getting rpc requests until SIGKILL. 470 */ 471 for (;;) { 472 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 473 while (nfsd->nfsd_slp == NULL && 474 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 475 nfsd->nfsd_flag |= NFSD_WAITING; 476 nfsd_waiting++; 477 error = tsleep(nfsd, PCATCH, "nfsd", 0); 478 nfsd_waiting--; 479 if (error && nfsd->nfsd_slp == NULL) 480 goto done; 481 } 482 if (nfsd->nfsd_slp == NULL && 483 (nfsd_head_flag & NFSD_CHECKSLP)) { 484 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 485 if ((slp->ns_flag & SLP_ACTION_MASK) || 486 slp->ns_needq_upcall) { 487 nfsrv_slpref(slp); 488 nfsd->nfsd_slp = slp; 489 break; 490 } 491 } 492 if (slp == NULL) 493 nfsd_head_flag &= ~NFSD_CHECKSLP; 494 } 495 if ((slp = nfsd->nfsd_slp) == NULL) 496 continue; 497 498 lwkt_reltoken(&nfs_token); 499 lwkt_gettoken(&slp->ns_token); 500 501 if (slp->ns_needq_upcall) { 502 slp->ns_needq_upcall = 0; 503 slp->ns_flag |= SLP_NEEDQ; 504 } 505 506 if (slp->ns_flag & SLP_VALID) { 507 /* 508 * We can both process additional received 509 * data into new records and process existing 510 * records. This keeps the pipeline hot by 511 * allowing the tcp socket to continue to 512 * drain while we are processing records. 513 */ 514 if (slp->ns_flag & SLP_DISCONN) { 515 nfsrv_zapsock(slp); 516 } else if (slp->ns_flag & SLP_NEEDQ) { 517 (void)nfs_slplock(slp, 1); 518 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 519 MB_WAIT); 520 nfs_slpunlock(slp); 521 } 522 error = nfsrv_dorec(slp, nfsd, &nd); 523 cur_usec = nfs_curusec(); 524 if (error && slp->ns_tq.lh_first && 525 slp->ns_tq.lh_first->nd_time <= cur_usec) { 526 error = 0; 527 cacherep = RC_DOIT; 528 writes_todo = 1; 529 } else { 530 writes_todo = 0; 531 } 532 nfsd->nfsd_flag |= NFSD_REQINPROG; 533 } else { 534 slp->ns_flag &= ~SLP_ACTION_MASK; 535 error = 0; 536 } 537 } else { 538 error = 0; 539 slp = nfsd->nfsd_slp; 540 541 lwkt_reltoken(&nfs_token); 542 lwkt_gettoken(&slp->ns_token); 543 544 if (slp->ns_needq_upcall) { 545 slp->ns_needq_upcall = 0; 546 slp->ns_flag |= SLP_NEEDQ; 547 } 548 if (NFSRV_RECLIMIT(slp) == 0 && 549 (slp->ns_flag & SLP_NEEDQ)) { 550 (void)nfs_slplock(slp, 1); 551 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 552 MB_WAIT); 553 nfs_slpunlock(slp); 554 } 555 } 556 557 /* 558 * nfs_token not held here. slp token is held. 559 */ 560 if (error || (slp->ns_flag & SLP_VALID) == 0) { 561 if (nd) { 562 kfree((caddr_t)nd, M_NFSRVDESC); 563 nd = NULL; 564 } 565 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 566 if (slp->ns_flag & SLP_ACTION_MASK) { 567 lwkt_reltoken(&slp->ns_token); 568 lwkt_gettoken(&nfs_token); 569 } else { 570 nfsd->nfsd_slp = NULL; 571 lwkt_reltoken(&slp->ns_token); 572 lwkt_gettoken(&nfs_token); 573 nfsrv_slpderef(slp); 574 } 575 continue; 576 } 577 578 /* 579 * Execute the NFS request - handle the server side cache 580 * 581 * nfs_token not held here. slp token is held. 582 */ 583 sotype = slp->ns_so->so_type; 584 if (nd) { 585 getmicrotime(&nd->nd_starttime); 586 if (nd->nd_nam2) 587 nd->nd_nam = nd->nd_nam2; 588 else 589 nd->nd_nam = slp->ns_nam; 590 591 /* 592 * Check to see if authorization is needed. 593 */ 594 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 595 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 596 nsd->nsd_haddr = 597 ((struct sockaddr_in *) 598 nd->nd_nam)->sin_addr.s_addr; 599 nsd->nsd_authlen = nfsd->nfsd_authlen; 600 nsd->nsd_verflen = nfsd->nfsd_verflen; 601 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 602 nfsd->nfsd_authlen) && 603 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 604 nfsd->nfsd_verflen) && 605 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 606 lwkt_reltoken(&slp->ns_token); 607 return (ENEEDAUTH); 608 cacherep = RC_DROPIT; 609 } else { 610 cacherep = nfsrv_getcache(nd, slp, &mreq); 611 } 612 613 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 614 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 615 nd->nd_procnum = NFSPROC_NOOP; 616 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 617 cacherep = RC_DOIT; 618 } else if (nfs_privport) { 619 /* Check if source port is privileged */ 620 u_short port; 621 struct sockaddr *nam = nd->nd_nam; 622 struct sockaddr_in *sin; 623 624 sin = (struct sockaddr_in *)nam; 625 port = ntohs(sin->sin_port); 626 if (port >= IPPORT_RESERVED && 627 nd->nd_procnum != NFSPROC_NULL) { 628 nd->nd_procnum = NFSPROC_NOOP; 629 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 630 cacherep = RC_DOIT; 631 kprintf("NFS request from unprivileged port (%s:%d)\n", 632 inet_ntoa(sin->sin_addr), port); 633 } 634 } 635 636 } 637 638 /* 639 * Execute the NFS request - direct execution 640 * 641 * Loop to get all the write rpc replies that have been 642 * gathered together. 643 * 644 * nfs_token not held here. slp token is held. 645 */ 646 do { 647 switch (cacherep) { 648 case RC_DOIT: 649 if (nd && (nd->nd_flag & ND_NFSV3)) 650 procrastinate = nfsrvw_procrastinate_v3; 651 else 652 procrastinate = nfsrvw_procrastinate; 653 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 654 procrastinate > 0) 655 ) { 656 error = nfsrv_writegather(&nd, slp, 657 nfsd->nfsd_td, &mreq); 658 } else { 659 /* NOT YET lwkt_reltoken(&slp->ns_token); */ 660 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 661 slp, nfsd->nfsd_td, &mreq); 662 /* NOT YET lwkt_gettoken(&slp->ns_token); */ 663 } 664 if (mreq == NULL) 665 break; 666 if (error != 0 && error != NFSERR_RETVOID) { 667 if (nd->nd_procnum != NQNFSPROC_VACATED) 668 nfsstats.srv_errs++; 669 nfsrv_updatecache(nd, FALSE, mreq); 670 if (nd->nd_nam2) 671 FREE(nd->nd_nam2, M_SONAME); 672 break; 673 } 674 nfsstats.srvrpccnt[nd->nd_procnum]++; 675 nfsrv_updatecache(nd, TRUE, mreq); 676 nd->nd_mrep = NULL; 677 /* FALL THROUGH */ 678 case RC_REPLY: 679 m = mreq; 680 siz = 0; 681 while (m) { 682 siz += m->m_len; 683 m = m->m_next; 684 } 685 if (siz <= 0 || siz > NFS_MAXPACKET) { 686 kprintf("mbuf siz=%d\n",siz); 687 panic("Bad nfs svc reply"); 688 } 689 m = mreq; 690 m->m_pkthdr.len = siz; 691 m->m_pkthdr.rcvif = NULL; 692 /* 693 * For stream protocols, prepend a Sun RPC 694 * Record Mark. 695 */ 696 if (sotype == SOCK_STREAM) { 697 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT); 698 if (m == NULL) { 699 error = ENOBUFS; 700 goto skip; 701 } 702 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 703 } 704 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 705 (void) nfs_slplock(slp, 1); 706 if (slp->ns_flag & SLP_VALID) 707 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 708 else { 709 error = EPIPE; 710 m_freem(m); 711 } 712 skip: 713 if (nfsrtton) 714 nfsd_rt(sotype, nd, cacherep); 715 if (nd->nd_nam2) 716 FREE(nd->nd_nam2, M_SONAME); 717 if (nd->nd_mrep) 718 m_freem(nd->nd_mrep); 719 if (error == EPIPE || error == ENOBUFS) 720 nfsrv_zapsock(slp); 721 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 722 nfs_slpunlock(slp); 723 if (error == EINTR || error == ERESTART) { 724 kfree((caddr_t)nd, M_NFSRVDESC); 725 lwkt_reltoken(&slp->ns_token); 726 lwkt_gettoken(&nfs_token); 727 nfsd->nfsd_slp = NULL; 728 nfsrv_slpderef(slp); 729 goto done; 730 } 731 break; 732 case RC_DROPIT: 733 if (nfsrtton) 734 nfsd_rt(sotype, nd, cacherep); 735 m_freem(nd->nd_mrep); 736 if (nd->nd_nam2) 737 FREE(nd->nd_nam2, M_SONAME); 738 break; 739 }; 740 if (nd) { 741 FREE((caddr_t)nd, M_NFSRVDESC); 742 nd = NULL; 743 } 744 745 /* 746 * Check to see if there are outstanding writes that 747 * need to be serviced. 748 */ 749 cur_usec = nfs_curusec(); 750 if (slp->ns_tq.lh_first && 751 slp->ns_tq.lh_first->nd_time <= cur_usec) { 752 cacherep = RC_DOIT; 753 writes_todo = 1; 754 } else { 755 writes_todo = 0; 756 } 757 } while (writes_todo); 758 759 /* 760 * nfs_token not held here. slp token is held. 761 */ 762 if (nfsrv_dorec(slp, nfsd, &nd)) { 763 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 764 if (slp->ns_flag & SLP_ACTION_MASK) { 765 lwkt_reltoken(&slp->ns_token); 766 lwkt_gettoken(&nfs_token); 767 } else { 768 nfsd->nfsd_slp = NULL; 769 lwkt_reltoken(&slp->ns_token); 770 lwkt_gettoken(&nfs_token); 771 nfsrv_slpderef(slp); 772 } 773 } else { 774 lwkt_reltoken(&slp->ns_token); 775 lwkt_gettoken(&nfs_token); 776 } 777 } 778 done: 779 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 780 kfree((caddr_t)nfsd, M_NFSD); 781 nsd->nsd_nfsd = NULL; 782 if (--nfs_numnfsd == 0) 783 nfsrv_init(TRUE); /* Reinitialize everything */ 784 785 lwkt_reltoken(&nfs_token); 786 return (error); 787 } 788 789 /* 790 * Shut down a socket associated with an nfssvc_sock structure. 791 * Should be called with the send lock set, if required. 792 * The trick here is to increment the sref at the start, so that the nfsds 793 * will stop using it and clear ns_flag at the end so that it will not be 794 * reassigned during cleanup. 795 */ 796 static void 797 nfsrv_zapsock(struct nfssvc_sock *slp) 798 { 799 struct nfsuid *nuidp, *nnuidp; 800 struct nfsrv_descript *nwp, *nnwp; 801 struct socket *so; 802 struct file *fp; 803 struct nfsrv_rec *rec; 804 805 slp->ns_flag &= ~SLP_ALLFLAGS; 806 fp = slp->ns_fp; 807 if (fp) { 808 slp->ns_fp = NULL; 809 so = slp->ns_so; 810 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 811 so->so_upcall = NULL; 812 so->so_upcallarg = NULL; 813 soshutdown(so, SHUT_RDWR); 814 closef(fp, NULL); 815 if (slp->ns_nam) 816 FREE(slp->ns_nam, M_SONAME); 817 m_freem(slp->ns_raw); 818 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 819 --slp->ns_numrec; 820 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 821 if (rec->nr_address) 822 FREE(rec->nr_address, M_SONAME); 823 m_freem(rec->nr_packet); 824 kfree(rec, M_NFSRVDESC); 825 } 826 KKASSERT(slp->ns_numrec == 0); 827 828 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru, 829 nnuidp) { 830 LIST_REMOVE(nuidp, nu_hash); 831 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 832 if (nuidp->nu_flag & NU_NAM) 833 FREE(nuidp->nu_nam, M_SONAME); 834 kfree((caddr_t)nuidp, M_NFSUID); 835 } 836 crit_enter(); 837 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 838 nnwp = nwp->nd_tq.le_next; 839 LIST_REMOVE(nwp, nd_tq); 840 kfree((caddr_t)nwp, M_NFSRVDESC); 841 } 842 LIST_INIT(&slp->ns_tq); 843 crit_exit(); 844 nfsrv_slpderef(slp); 845 } 846 } 847 848 /* 849 * Derefence a server socket structure. If it has no more references and 850 * is no longer valid, you can throw it away. 851 * 852 * Must be holding nfs_token! 853 */ 854 void 855 nfsrv_slpderef(struct nfssvc_sock *slp) 856 { 857 ASSERT_LWKT_TOKEN_HELD(&nfs_token); 858 if (slp->ns_sref == 1) { 859 KKASSERT((slp->ns_flag & SLP_VALID) == 0); 860 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 861 slp->ns_sref = 0; 862 kfree((caddr_t)slp, M_NFSSVC); 863 } else { 864 --slp->ns_sref; 865 } 866 } 867 868 void 869 nfsrv_slpref(struct nfssvc_sock *slp) 870 { 871 ASSERT_LWKT_TOKEN_HELD(&nfs_token); 872 ++slp->ns_sref; 873 } 874 875 /* 876 * Lock a socket against others. 877 * 878 * Returns 0 on failure, 1 on success. 879 */ 880 int 881 nfs_slplock(struct nfssvc_sock *slp, int wait) 882 { 883 mtx_t mtx = &slp->ns_solock; 884 885 if (wait) { 886 mtx_lock_ex(mtx, "nfsslplck", 0, 0); 887 return(1); 888 } else if (mtx_lock_ex_try(mtx) == 0) { 889 return(1); 890 } else { 891 return(0); 892 } 893 } 894 895 /* 896 * Unlock the stream socket for others. 897 */ 898 void 899 nfs_slpunlock(struct nfssvc_sock *slp) 900 { 901 mtx_t mtx = &slp->ns_solock; 902 903 mtx_unlock(mtx); 904 } 905 906 /* 907 * Initialize the data structures for the server. 908 * Handshake with any new nfsds starting up to avoid any chance of 909 * corruption. 910 */ 911 void 912 nfsrv_init(int terminating) 913 { 914 struct nfssvc_sock *slp, *nslp; 915 916 lwkt_gettoken(&nfs_token); 917 if (nfssvc_sockhead_flag & SLP_INIT) 918 panic("nfsd init"); 919 nfssvc_sockhead_flag |= SLP_INIT; 920 921 if (terminating) { 922 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) { 923 if (slp->ns_flag & SLP_VALID) 924 nfsrv_zapsock(slp); 925 /* 926 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 927 kfree((caddr_t)slp, M_NFSSVC); 928 */ 929 } 930 nfsrv_cleancache(); /* And clear out server cache */ 931 } else { 932 nfs_pub.np_valid = 0; 933 } 934 935 TAILQ_INIT(&nfssvc_sockhead); 936 nfssvc_sockhead_flag &= ~SLP_INIT; 937 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 938 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 939 wakeup((caddr_t)&nfssvc_sockhead); 940 } 941 942 TAILQ_INIT(&nfsd_head); 943 nfsd_head_flag &= ~NFSD_CHECKSLP; 944 945 lwkt_reltoken(&nfs_token); 946 947 #if 0 948 nfs_udpsock = (struct nfssvc_sock *) 949 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 950 mtx_init(&nfs_udpsock->ns_solock); 951 STAILQ_INIT(&nfs_udpsock->ns_rec); 952 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 953 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 954 955 nfs_cltpsock = (struct nfssvc_sock *) 956 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 957 mtx_init(&nfs_cltpsock->ns_solock); 958 STAILQ_INIT(&nfs_cltpsock->ns_rec); 959 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 960 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 961 #endif 962 } 963 964 /* 965 * Add entries to the server monitor log. 966 */ 967 static void 968 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 969 { 970 struct drt *rt; 971 972 rt = &nfsdrt.drt[nfsdrt.pos]; 973 if (cacherep == RC_DOIT) 974 rt->flag = 0; 975 else if (cacherep == RC_REPLY) 976 rt->flag = DRT_CACHEREPLY; 977 else 978 rt->flag = DRT_CACHEDROP; 979 if (sotype == SOCK_STREAM) 980 rt->flag |= DRT_TCP; 981 if (nd->nd_flag & ND_NFSV3) 982 rt->flag |= DRT_NFSV3; 983 rt->proc = nd->nd_procnum; 984 if (nd->nd_nam->sa_family == AF_INET) 985 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 986 else 987 rt->ipadr = INADDR_ANY; 988 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 989 getmicrotime(&rt->tstamp); 990 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 991 } 992 #endif /* NFS_NOSERVER */ 993 994 /* 995 * Get an authorization string for the uid by having the mount_nfs sitting 996 * on this mount point porpous out of the kernel and do it. 997 */ 998 int 999 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, 1000 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str, 1001 int *verf_len, NFSKERBKEY_T key /* return session key */) 1002 { 1003 int error = 0; 1004 1005 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 1006 nmp->nm_state |= NFSSTA_WANTAUTH; 1007 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 1008 "nfsauth1", 2 * hz); 1009 error = nfs_sigintr(nmp, rep, rep->r_td); 1010 if (error) { 1011 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1012 return (error); 1013 } 1014 } 1015 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1016 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1017 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1018 nmp->nm_verfstr = verf_str; 1019 nmp->nm_verflen = *verf_len; 1020 nmp->nm_authuid = cred->cr_uid; 1021 wakeup((caddr_t)&nmp->nm_authstr); 1022 1023 /* 1024 * And wait for mount_nfs to do its stuff. 1025 */ 1026 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1027 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1028 "nfsauth2", 2 * hz); 1029 error = nfs_sigintr(nmp, rep, rep->r_td); 1030 } 1031 if (nmp->nm_state & NFSSTA_AUTHERR) { 1032 nmp->nm_state &= ~NFSSTA_AUTHERR; 1033 error = EAUTH; 1034 } 1035 if (error) 1036 kfree((caddr_t)*auth_str, M_TEMP); 1037 else { 1038 *auth_len = nmp->nm_authlen; 1039 *verf_len = nmp->nm_verflen; 1040 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1041 } 1042 nmp->nm_state &= ~NFSSTA_HASAUTH; 1043 nmp->nm_state |= NFSSTA_WAITAUTH; 1044 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1045 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1046 wakeup((caddr_t)&nmp->nm_authtype); 1047 } 1048 return (error); 1049 } 1050 1051 /* 1052 * Get a nickname authenticator and verifier. 1053 */ 1054 int 1055 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str, 1056 int *auth_len, char *verf_str, int verf_len) 1057 { 1058 struct nfsuid *nuidp; 1059 u_int32_t *nickp, *verfp; 1060 struct timeval ktvin, ktvout; 1061 1062 #ifdef DIAGNOSTIC 1063 if (verf_len < (4 * NFSX_UNSIGNED)) 1064 panic("nfs_getnickauth verf too small"); 1065 #endif 1066 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1067 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1068 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1069 break; 1070 } 1071 if (!nuidp || nuidp->nu_expire < time_second) 1072 return (EACCES); 1073 1074 /* 1075 * Move to the end of the lru list (end of lru == most recently used). 1076 */ 1077 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1078 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1079 1080 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1081 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1082 *nickp = txdr_unsigned(nuidp->nu_nickname); 1083 *auth_str = (char *)nickp; 1084 *auth_len = 2 * NFSX_UNSIGNED; 1085 1086 /* 1087 * Now we must encrypt the verifier and package it up. 1088 */ 1089 verfp = (u_int32_t *)verf_str; 1090 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1091 if (time_second > nuidp->nu_timestamp.tv_sec || 1092 (time_second == nuidp->nu_timestamp.tv_sec && 1093 time_second > nuidp->nu_timestamp.tv_usec)) 1094 getmicrotime(&nuidp->nu_timestamp); 1095 else 1096 nuidp->nu_timestamp.tv_usec++; 1097 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1098 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1099 1100 /* 1101 * Now encrypt the timestamp verifier in ecb mode using the session 1102 * key. 1103 */ 1104 #ifdef NFSKERB 1105 XXX 1106 #else 1107 ktvout.tv_sec = 0; 1108 ktvout.tv_usec = 0; 1109 #endif 1110 1111 *verfp++ = ktvout.tv_sec; 1112 *verfp++ = ktvout.tv_usec; 1113 *verfp = 0; 1114 return (0); 1115 } 1116 1117 /* 1118 * Save the current nickname in a hash list entry on the mount point. 1119 */ 1120 int 1121 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len, 1122 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, 1123 struct mbuf *mrep) 1124 { 1125 struct nfsuid *nuidp; 1126 u_int32_t *tl; 1127 struct timeval ktvin, ktvout; 1128 u_int32_t nick; 1129 int deltasec, error = 0; 1130 struct nfsm_info info; 1131 1132 info.md = *mdp; 1133 info.dpos = *dposp; 1134 info.mrep = mrep; 1135 1136 if (len == (3 * NFSX_UNSIGNED)) { 1137 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 1138 ktvin.tv_sec = *tl++; 1139 ktvin.tv_usec = *tl++; 1140 nick = fxdr_unsigned(u_int32_t, *tl); 1141 1142 /* 1143 * Decrypt the timestamp in ecb mode. 1144 */ 1145 #ifdef NFSKERB 1146 XXX 1147 #else 1148 ktvout.tv_sec = 0; 1149 ktvout.tv_usec = 0; 1150 #endif 1151 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1152 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1153 deltasec = time_second - ktvout.tv_sec; 1154 if (deltasec < 0) 1155 deltasec = -deltasec; 1156 /* 1157 * If ok, add it to the hash list for the mount point. 1158 */ 1159 if (deltasec <= NFS_KERBCLOCKSKEW) { 1160 if (nmp->nm_numuids < nuidhash_max) { 1161 nmp->nm_numuids++; 1162 nuidp = (struct nfsuid *) 1163 kmalloc(sizeof (struct nfsuid), M_NFSUID, 1164 M_WAITOK); 1165 } else { 1166 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead); 1167 LIST_REMOVE(nuidp, nu_hash); 1168 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1169 nu_lru); 1170 } 1171 nuidp->nu_flag = 0; 1172 nuidp->nu_cr.cr_uid = cred->cr_uid; 1173 nuidp->nu_expire = time_second + NFS_KERBTTL; 1174 nuidp->nu_timestamp = ktvout; 1175 nuidp->nu_nickname = nick; 1176 bcopy(key, nuidp->nu_key, sizeof (key)); 1177 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1178 nu_lru); 1179 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1180 nuidp, nu_hash); 1181 } 1182 } else { 1183 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 1184 } 1185 nfsmout: 1186 *mdp = info.md; 1187 *dposp = info.dpos; 1188 return (error); 1189 } 1190