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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 33 * $FreeBSD: src/sys/nfs/nfs_syscalls.c,v 1.58.2.1 2000/11/26 02:30:06 dillon Exp $ 34 */ 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/sysproto.h> 39 #include <sys/kernel.h> 40 #include <sys/sysctl.h> 41 #include <sys/file.h> 42 #include <sys/filedesc.h> 43 #include <sys/vnode.h> 44 #include <sys/malloc.h> 45 #include <sys/mount.h> 46 #include <sys/proc.h> 47 #include <sys/priv.h> 48 #include <sys/buf.h> 49 #include <sys/mbuf.h> 50 #include <sys/resourcevar.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/domain.h> 54 #include <sys/protosw.h> 55 #include <sys/nlookup.h> 56 57 #include <sys/mutex2.h> 58 #include <sys/thread2.h> 59 60 #include <netinet/in.h> 61 #include <netinet/tcp.h> 62 #include "xdr_subs.h" 63 #include "rpcv2.h" 64 #include "nfsproto.h" 65 #include "nfs.h" 66 #include "nfsm_subs.h" 67 #include "nfsrvcache.h" 68 #include "nfsmount.h" 69 #include "nfsnode.h" 70 #include "nfsrtt.h" 71 72 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 73 74 static int nuidhash_max = NFS_MAXUIDHASH; 75 76 #ifndef NFS_NOSERVER 77 static void nfsrv_zapsock (struct nfssvc_sock *slp); 78 #endif 79 80 #define TRUE 1 81 #define FALSE 0 82 83 SYSCTL_DECL(_vfs_nfs); 84 85 #ifndef NFS_NOSERVER 86 int nfsd_waiting = 0; 87 static struct nfsdrt nfsdrt; 88 static int nfs_numnfsd = 0; 89 static void nfsd_rt (int sotype, struct nfsrv_descript *nd, 90 int cacherep); 91 static int nfssvc_addsock (struct file *, struct sockaddr *, 92 struct thread *); 93 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *); 94 95 static int nfs_privport = 0; 96 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 97 0, "Enable privileged source port checks"); 98 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, 99 "Enable NFS request procrastination"); 100 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, 101 "Enable NFSv3 request procrastination"); 102 int nfs_soreserve = NFS_MAXPACKET * NFS_MAXASYNCBIO; 103 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, 104 "Minimum NFS socket buffer size reservation"); 105 106 /* 107 * NFS server system calls 108 */ 109 110 #endif /* NFS_NOSERVER */ 111 /* 112 * nfssvc_args(int flag, caddr_t argp) 113 * 114 * Nfs server psuedo system call for the nfsd's 115 * Based on the flag value it either: 116 * - adds a socket to the selection list 117 * - remains in the kernel as an nfsd 118 * - remains in the kernel as an nfsiod 119 * 120 * MPALMOSTSAFE 121 */ 122 int 123 sys_nfssvc(struct nfssvc_args *uap) 124 { 125 #ifndef NFS_NOSERVER 126 struct nlookupdata nd; 127 struct file *fp; 128 struct sockaddr *nam; 129 struct nfsd_args nfsdarg; 130 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 131 struct nfsd_cargs ncd; 132 struct nfsd *nfsd; 133 struct nfssvc_sock *slp; 134 struct nfsuid *nuidp; 135 struct nfsmount *nmp; 136 struct vnode *vp; 137 #endif /* NFS_NOSERVER */ 138 int error; 139 struct thread *td = curthread; 140 141 /* 142 * Must be super user 143 */ 144 error = priv_check(td, PRIV_ROOT); 145 if (error) 146 return (error); 147 148 lwkt_gettoken(&nfs_token); 149 150 while (nfssvc_sockhead_flag & SLP_INIT) { 151 nfssvc_sockhead_flag |= SLP_WANTINIT; 152 tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 153 } 154 if (uap->flag & NFSSVC_BIOD) 155 error = ENXIO; /* no longer need nfsiod's */ 156 #ifdef NFS_NOSERVER 157 else 158 error = ENXIO; 159 #else /* !NFS_NOSERVER */ 160 else if (uap->flag & NFSSVC_MNTD) { 161 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 162 if (error) 163 goto done; 164 vp = NULL; 165 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 166 NLC_FOLLOW); 167 if (error == 0) 168 error = nlookup(&nd); 169 if (error == 0) 170 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 171 nlookup_done(&nd); 172 if (error) 173 goto done; 174 175 if ((vp->v_flag & VROOT) == 0) 176 error = EINVAL; 177 nmp = VFSTONFS(vp->v_mount); 178 vput(vp); 179 if (error) 180 goto done; 181 if ((nmp->nm_state & NFSSTA_MNTD) && 182 (uap->flag & NFSSVC_GOTAUTH) == 0) { 183 error = 0; 184 goto done; 185 } 186 nmp->nm_state |= NFSSTA_MNTD; 187 error = nfs_clientd(nmp, td->td_ucred, &ncd, uap->flag, 188 uap->argp, td); 189 } else if (uap->flag & NFSSVC_ADDSOCK) { 190 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 191 if (error) 192 goto done; 193 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 194 if (error) 195 goto done; 196 /* 197 * Get the client address for connected sockets. 198 */ 199 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 200 nam = NULL; 201 else { 202 error = getsockaddr(&nam, nfsdarg.name, 203 nfsdarg.namelen); 204 if (error) { 205 fdrop(fp); 206 goto done; 207 } 208 } 209 error = nfssvc_addsock(fp, nam, td); 210 fdrop(fp); 211 } else { 212 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 213 if (error) 214 goto done; 215 if ((uap->flag & NFSSVC_AUTHIN) && 216 ((nfsd = nsd->nsd_nfsd)) != NULL && 217 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 218 slp = nfsd->nfsd_slp; 219 220 /* 221 * First check to see if another nfsd has already 222 * added this credential. 223 */ 224 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 225 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 226 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 227 (!nfsd->nfsd_nd->nd_nam2 || 228 netaddr_match(NU_NETFAM(nuidp), 229 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 230 break; 231 } 232 if (nuidp) { 233 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 234 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 235 } else { 236 /* 237 * Nope, so we will. 238 */ 239 if (slp->ns_numuids < nuidhash_max) { 240 slp->ns_numuids++; 241 nuidp = (struct nfsuid *) 242 kmalloc(sizeof (struct nfsuid), M_NFSUID, 243 M_WAITOK); 244 } else 245 nuidp = NULL; 246 if ((slp->ns_flag & SLP_VALID) == 0) { 247 if (nuidp) 248 kfree((caddr_t)nuidp, M_NFSUID); 249 } else { 250 if (nuidp == NULL) { 251 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 252 LIST_REMOVE(nuidp, nu_hash); 253 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 254 nu_lru); 255 if (nuidp->nu_flag & NU_NAM) 256 kfree(nuidp->nu_nam, M_SONAME); 257 } 258 nuidp->nu_flag = 0; 259 nuidp->nu_cr = nsd->nsd_cr; 260 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 261 nuidp->nu_cr.cr_ngroups = NGROUPS; 262 nuidp->nu_cr.cr_ref = 1; 263 nuidp->nu_timestamp = nsd->nsd_timestamp; 264 nuidp->nu_expire = time_uptime + nsd->nsd_ttl; 265 /* 266 * and save the session key in nu_key. 267 */ 268 bcopy(nsd->nsd_key, nuidp->nu_key, 269 sizeof (nsd->nsd_key)); 270 if (nfsd->nfsd_nd->nd_nam2) { 271 struct sockaddr_in *saddr; 272 273 saddr = (struct sockaddr_in *) 274 nfsd->nfsd_nd->nd_nam2; 275 switch (saddr->sin_family) { 276 case AF_INET: 277 nuidp->nu_flag |= NU_INETADDR; 278 nuidp->nu_inetaddr = 279 saddr->sin_addr.s_addr; 280 break; 281 case AF_ISO: 282 default: 283 nuidp->nu_flag |= NU_NAM; 284 nuidp->nu_nam = 285 dup_sockaddr(nfsd->nfsd_nd->nd_nam2); 286 break; 287 }; 288 } 289 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 290 nu_lru); 291 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 292 nuidp, nu_hash); 293 nfsrv_setcred(&nuidp->nu_cr, 294 &nfsd->nfsd_nd->nd_cr); 295 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 296 } 297 } 298 } 299 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 300 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 301 error = nfssvc_nfsd(nsd, uap->argp, td); 302 } 303 #endif /* NFS_NOSERVER */ 304 if (error == EINTR || error == ERESTART) 305 error = 0; 306 done: 307 lwkt_reltoken(&nfs_token); 308 return (error); 309 } 310 311 #ifndef NFS_NOSERVER 312 /* 313 * Adds a socket to the list for servicing by nfsds. 314 */ 315 static int 316 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td) 317 { 318 int siz; 319 struct nfssvc_sock *slp; 320 struct socket *so; 321 int error; 322 323 so = (struct socket *)fp->f_data; 324 #if 0 325 tslp = NULL; 326 /* 327 * Add it to the list, as required. 328 */ 329 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 330 tslp = nfs_udpsock; 331 if (tslp->ns_flag & SLP_VALID) { 332 if (mynam != NULL) 333 kfree(mynam, M_SONAME); 334 return (EPERM); 335 } 336 } 337 #endif 338 /* 339 * Reserve buffer space in the socket. Note that due to bugs in 340 * Linux's delayed-ack code, serious performance degredation may 341 * occur with linux hosts if the minimum is used. 342 * 343 * NFS sockets are not limited to the standard sb_max or by 344 * resource limits. 345 */ 346 if (so->so_type == SOCK_STREAM) 347 siz = NFS_MAXPACKET + sizeof (u_long); 348 else 349 siz = NFS_MAXPACKET; 350 if (siz < nfs_soreserve) 351 siz = nfs_soreserve; 352 353 error = soreserve(so, siz, siz, NULL); 354 if (error) { 355 if (mynam != NULL) 356 kfree(mynam, M_SONAME); 357 return (error); 358 } 359 360 /* 361 * Set protocol specific options { for now TCP only } and 362 * reserve some space. For datagram sockets, this can get called 363 * repeatedly for the same socket, but that isn't harmful. 364 */ 365 if (so->so_type == SOCK_STREAM) { 366 struct sockopt sopt; 367 int val; 368 369 bzero(&sopt, sizeof sopt); 370 sopt.sopt_level = SOL_SOCKET; 371 sopt.sopt_name = SO_KEEPALIVE; 372 sopt.sopt_val = &val; 373 sopt.sopt_valsize = sizeof val; 374 val = 1; 375 sosetopt(so, &sopt); 376 } 377 if (so->so_proto->pr_domain->dom_family == AF_INET && 378 so->so_proto->pr_protocol == IPPROTO_TCP) { 379 struct sockopt sopt; 380 int val; 381 382 bzero(&sopt, sizeof sopt); 383 sopt.sopt_level = IPPROTO_TCP; 384 sopt.sopt_name = TCP_NODELAY; 385 sopt.sopt_val = &val; 386 sopt.sopt_valsize = sizeof val; 387 val = 1; 388 sosetopt(so, &sopt); 389 390 bzero(&sopt, sizeof sopt); 391 sopt.sopt_level = IPPROTO_TCP; 392 sopt.sopt_name = TCP_FASTKEEP; 393 sopt.sopt_val = &val; 394 sopt.sopt_valsize = sizeof val; 395 val = 1; 396 sosetopt(so, &sopt); 397 } 398 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_NOINTR); 399 so->so_rcv.ssb_timeo = 0; 400 atomic_clear_int(&so->so_snd.ssb_flags, SSB_NOINTR); 401 so->so_snd.ssb_timeo = 0; 402 403 slp = kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 404 mtx_init(&slp->ns_solock); 405 STAILQ_INIT(&slp->ns_rec); 406 TAILQ_INIT(&slp->ns_uidlruhead); 407 lwkt_token_init(&slp->ns_token, "nfssrv_token"); 408 409 lwkt_gettoken(&nfs_token); 410 nfsrv_slpref(slp); 411 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 412 lwkt_gettoken(&slp->ns_token); 413 414 slp->ns_so = so; 415 slp->ns_nam = mynam; 416 fp->f_count++; 417 slp->ns_fp = fp; 418 419 so->so_upcallarg = (caddr_t)slp; 420 so->so_upcall = nfsrv_rcv_upcall; 421 atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 422 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 423 nfsrv_wakenfsd(slp, 1); 424 425 lwkt_reltoken(&slp->ns_token); 426 lwkt_reltoken(&nfs_token); 427 428 return (0); 429 } 430 431 /* 432 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 433 * until it is killed by a signal. 434 */ 435 static int 436 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td) 437 { 438 int siz; 439 struct nfssvc_sock *slp; 440 struct nfsd *nfsd = nsd->nsd_nfsd; 441 struct nfsrv_descript *nd = NULL; 442 struct mbuf *m, *mreq; 443 int error, cacherep, sotype, writes_todo; 444 int procrastinate; 445 int slplocked; 446 u_quad_t cur_usec; 447 448 #ifndef nolint 449 cacherep = RC_DOIT; 450 writes_todo = 0; 451 #endif 452 lwkt_gettoken(&nfs_token); 453 454 if (nfsd == NULL) { 455 nsd->nsd_nfsd = nfsd = (struct nfsd *) 456 kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO); 457 nfsd->nfsd_td = td; 458 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 459 nfs_numnfsd++; 460 } 461 462 /* 463 * Loop getting rpc requests until SIGKILL. 464 */ 465 for (;;) { 466 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 467 while (nfsd->nfsd_slp == NULL && 468 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 469 nfsd->nfsd_flag |= NFSD_WAITING; 470 nfsd_waiting++; 471 error = tsleep(nfsd, PCATCH, "nfsd", 0); 472 nfsd_waiting--; 473 if (error && nfsd->nfsd_slp == NULL) 474 goto done; 475 } 476 if (nfsd->nfsd_slp == NULL && 477 (nfsd_head_flag & NFSD_CHECKSLP)) { 478 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 479 if ((slp->ns_flag & SLP_ACTION_MASK) || 480 slp->ns_needq_upcall) { 481 nfsrv_slpref(slp); 482 nfsd->nfsd_slp = slp; 483 break; 484 } 485 } 486 if (slp == NULL) 487 nfsd_head_flag &= ~NFSD_CHECKSLP; 488 } 489 if ((slp = nfsd->nfsd_slp) == NULL) 490 continue; 491 492 lwkt_reltoken(&nfs_token); 493 lwkt_gettoken(&slp->ns_token); 494 495 if (slp->ns_needq_upcall) { 496 slp->ns_needq_upcall = 0; 497 slp->ns_flag |= SLP_NEEDQ; 498 } 499 500 if (slp->ns_flag & SLP_VALID) { 501 /* 502 * We can both process additional received 503 * data into new records and process existing 504 * records. This keeps the pipeline hot by 505 * allowing the tcp socket to continue to 506 * drain while we are processing records. 507 */ 508 while (slp->ns_flag & (SLP_DISCONN|SLP_NEEDQ)) { 509 if (slp->ns_flag & SLP_DISCONN) { 510 nfsrv_zapsock(slp); 511 break; 512 } 513 if (slp->ns_flag & SLP_NEEDQ) { 514 nfs_slplock(slp, 1); 515 if ((slp->ns_flag & 516 SLP_NEEDQ) == 0) { 517 nfs_slpunlock(slp); 518 continue; 519 } 520 nfsrv_rcv(slp->ns_so, 521 (caddr_t)slp, 522 MB_WAIT); 523 nfs_slpunlock(slp); 524 break; 525 } 526 } 527 error = nfsrv_dorec(slp, nfsd, &nd); 528 cur_usec = nfs_curusec(); 529 if (error && slp->ns_tq.lh_first && 530 slp->ns_tq.lh_first->nd_time <= cur_usec) { 531 error = 0; 532 cacherep = RC_DOIT; 533 writes_todo = 1; 534 } else { 535 writes_todo = 0; 536 } 537 nfsd->nfsd_flag |= NFSD_REQINPROG; 538 } else { 539 slp->ns_flag &= ~SLP_ACTION_MASK; 540 error = 0; 541 } 542 } else { 543 error = 0; 544 slp = nfsd->nfsd_slp; 545 546 lwkt_reltoken(&nfs_token); 547 lwkt_gettoken(&slp->ns_token); 548 549 if (slp->ns_needq_upcall) { 550 slp->ns_needq_upcall = 0; 551 slp->ns_flag |= SLP_NEEDQ; 552 } 553 if (NFSRV_RECLIMIT(slp) == 0 && 554 (slp->ns_flag & SLP_NEEDQ)) { 555 nfs_slplock(slp, 1); 556 if (NFSRV_RECLIMIT(slp) == 0 && 557 (slp->ns_flag & SLP_NEEDQ)) { 558 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 559 MB_WAIT); 560 } 561 nfs_slpunlock(slp); 562 } 563 } 564 565 /* 566 * nfs_token not held here. slp token is held. 567 */ 568 if (error || (slp->ns_flag & SLP_VALID) == 0) { 569 if (nd) { 570 kfree((caddr_t)nd, M_NFSRVDESC); 571 nd = NULL; 572 } 573 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 574 if (slp->ns_flag & SLP_ACTION_MASK) { 575 lwkt_reltoken(&slp->ns_token); 576 lwkt_gettoken(&nfs_token); 577 } else { 578 nfsd->nfsd_slp = NULL; 579 lwkt_reltoken(&slp->ns_token); 580 lwkt_gettoken(&nfs_token); 581 nfsrv_slpderef(slp); 582 } 583 continue; 584 } 585 sotype = slp->ns_so->so_type; 586 587 /* 588 * Execute the NFS request - handle the server side cache 589 * 590 * nfs_token not held here. slp token is held. 591 */ 592 if (nd) { 593 getmicrotime(&nd->nd_starttime); 594 if (nd->nd_nam2) 595 nd->nd_nam = nd->nd_nam2; 596 else 597 nd->nd_nam = slp->ns_nam; 598 599 /* 600 * Check to see if authorization is needed. 601 */ 602 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 603 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 604 nsd->nsd_haddr = 605 ((struct sockaddr_in *) 606 nd->nd_nam)->sin_addr.s_addr; 607 nsd->nsd_authlen = nfsd->nfsd_authlen; 608 nsd->nsd_verflen = nfsd->nfsd_verflen; 609 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 610 nfsd->nfsd_authlen) && 611 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 612 nfsd->nfsd_verflen) && 613 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 614 { 615 lwkt_reltoken(&slp->ns_token); 616 return (ENEEDAUTH); 617 } 618 cacherep = RC_DROPIT; 619 } else { 620 cacherep = nfsrv_getcache(nd, slp, &mreq); 621 } 622 623 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 624 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 625 nd->nd_procnum = NFSPROC_NOOP; 626 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 627 cacherep = RC_DOIT; 628 } else if (nfs_privport) { 629 /* Check if source port is privileged */ 630 u_short port; 631 struct sockaddr *nam = nd->nd_nam; 632 struct sockaddr_in *sin; 633 634 sin = (struct sockaddr_in *)nam; 635 port = ntohs(sin->sin_port); 636 if (port >= IPPORT_RESERVED && 637 nd->nd_procnum != NFSPROC_NULL) { 638 nd->nd_procnum = NFSPROC_NOOP; 639 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 640 cacherep = RC_DOIT; 641 kprintf("NFS request from unprivileged port (%s:%d)\n", 642 inet_ntoa(sin->sin_addr), port); 643 } 644 } 645 } 646 647 /* 648 * Execute the NFS request - direct execution 649 * 650 * Loop to get all the write rpc replies that have been 651 * gathered together. 652 * 653 * nfs_token not held here. slp token is held. 654 */ 655 do { 656 switch (cacherep) { 657 case RC_DOIT: 658 if (nd && (nd->nd_flag & ND_NFSV3)) 659 procrastinate = nfsrvw_procrastinate_v3; 660 else 661 procrastinate = nfsrvw_procrastinate; 662 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 663 procrastinate > 0) 664 ) { 665 error = nfsrv_writegather(&nd, slp, 666 nfsd->nfsd_td, &mreq); 667 } else { 668 /* NOT YET lwkt_reltoken(&slp->ns_token); */ 669 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 670 slp, nfsd->nfsd_td, &mreq); 671 /* NOT YET lwkt_gettoken(&slp->ns_token); */ 672 lwpkthreaddeferred(); /* vnlru issues */ 673 } 674 if (mreq == NULL) 675 break; 676 if (error != 0 && error != NFSERR_RETVOID) { 677 if (nd->nd_procnum != NQNFSPROC_VACATED) 678 nfsstats.srv_errs++; 679 nfsrv_updatecache(nd, FALSE, mreq); 680 if (nd->nd_nam2) 681 kfree(nd->nd_nam2, M_SONAME); 682 break; 683 } 684 nfsstats.srvrpccnt[nd->nd_procnum]++; 685 nfsrv_updatecache(nd, TRUE, mreq); 686 nd->nd_mrep = NULL; 687 /* FALL THROUGH */ 688 case RC_REPLY: 689 m = mreq; 690 siz = 0; 691 while (m) { 692 siz += m->m_len; 693 m = m->m_next; 694 } 695 if (siz <= 0 || siz > NFS_MAXPACKET) { 696 kprintf("mbuf siz=%d\n",siz); 697 panic("Bad nfs svc reply"); 698 } 699 m = mreq; 700 m->m_pkthdr.len = siz; 701 m->m_pkthdr.rcvif = NULL; 702 /* 703 * For stream protocols, prepend a Sun RPC 704 * Record Mark. 705 */ 706 if (sotype == SOCK_STREAM) { 707 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT); 708 if (m == NULL) { 709 error = ENOBUFS; 710 slplocked = 0; 711 goto skip; 712 } 713 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 714 } 715 if ((slp->ns_flag & SLP_VALID) && 716 (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)){ 717 nfs_slplock(slp, 1); 718 slplocked = 1; 719 } else { 720 slplocked = 0; 721 } 722 if (slp->ns_flag & SLP_VALID) { 723 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 724 } else { 725 error = EPIPE; 726 m_freem(m); 727 } 728 skip: 729 if (nfsrtton) 730 nfsd_rt(sotype, nd, cacherep); 731 if (nd->nd_nam2) 732 kfree(nd->nd_nam2, M_SONAME); 733 if (nd->nd_mrep) 734 m_freem(nd->nd_mrep); 735 if (error == EPIPE || error == ENOBUFS) 736 nfsrv_zapsock(slp); 737 if (slplocked) 738 nfs_slpunlock(slp); 739 if (error == EINTR || error == ERESTART) { 740 kfree((caddr_t)nd, M_NFSRVDESC); 741 lwkt_reltoken(&slp->ns_token); 742 lwkt_gettoken(&nfs_token); 743 nfsd->nfsd_slp = NULL; 744 nfsrv_slpderef(slp); 745 goto done; 746 } 747 break; 748 case RC_DROPIT: 749 if (nfsrtton) 750 nfsd_rt(sotype, nd, cacherep); 751 m_freem(nd->nd_mrep); 752 if (nd->nd_nam2) 753 kfree(nd->nd_nam2, M_SONAME); 754 break; 755 } 756 if (nd) { 757 kfree((caddr_t)nd, M_NFSRVDESC); 758 nd = NULL; 759 } 760 761 /* 762 * Check to see if there are outstanding writes that 763 * need to be serviced. 764 */ 765 cur_usec = nfs_curusec(); 766 if (slp->ns_tq.lh_first && 767 slp->ns_tq.lh_first->nd_time <= cur_usec) { 768 cacherep = RC_DOIT; 769 writes_todo = 1; 770 } else { 771 writes_todo = 0; 772 } 773 } while (writes_todo); 774 775 /* 776 * nfs_token not held here. slp token is held. 777 */ 778 if (nfsrv_dorec(slp, nfsd, &nd)) { 779 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 780 if (slp->ns_flag & SLP_ACTION_MASK) { 781 lwkt_reltoken(&slp->ns_token); 782 lwkt_gettoken(&nfs_token); 783 } else { 784 nfsd->nfsd_slp = NULL; 785 lwkt_reltoken(&slp->ns_token); 786 lwkt_gettoken(&nfs_token); 787 nfsrv_slpderef(slp); 788 } 789 } else { 790 lwkt_reltoken(&slp->ns_token); 791 lwkt_gettoken(&nfs_token); 792 } 793 } 794 done: 795 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 796 kfree((caddr_t)nfsd, M_NFSD); 797 nsd->nsd_nfsd = NULL; 798 if (--nfs_numnfsd == 0) 799 nfsrv_init(TRUE); /* Reinitialize everything */ 800 801 lwkt_reltoken(&nfs_token); 802 return (error); 803 } 804 805 /* 806 * Shut down a socket associated with an nfssvc_sock structure. 807 * Should be called with the send lock set, if required. 808 * 809 * The trick here is to increment the sref at the start, so that the nfsds 810 * will stop using it and clear ns_flag at the end so that it will not be 811 * reassigned during cleanup. 812 * 813 * That said, while we shutdown() the socket here, we don't actually destroy 814 * it until the final deref as there might be other code in the middle of 815 * using it. 816 */ 817 static void 818 nfsrv_zapsock(struct nfssvc_sock *slp) 819 { 820 struct nfsuid *nuidp, *nnuidp; 821 struct nfsrv_descript *nwp, *nnwp; 822 struct socket *so; 823 struct nfsrv_rec *rec; 824 int wasvalid; 825 826 wasvalid = slp->ns_flag & SLP_VALID; 827 slp->ns_flag &= ~SLP_ALLFLAGS; 828 if (wasvalid && slp->ns_fp) { 829 so = slp->ns_so; 830 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 831 so->so_upcall = NULL; 832 so->so_upcallarg = NULL; 833 soshutdown(so, SHUT_RDWR); 834 if (slp->ns_nam) { 835 kfree(slp->ns_nam, M_SONAME); 836 slp->ns_nam = NULL; 837 } 838 839 m_freem(slp->ns_raw); 840 slp->ns_raw = NULL; 841 842 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 843 --slp->ns_numrec; 844 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 845 if (rec->nr_address) 846 kfree(rec->nr_address, M_SONAME); 847 m_freem(rec->nr_packet); 848 kfree(rec, M_NFSRVDESC); 849 } 850 KKASSERT(slp->ns_numrec == 0); 851 852 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru, 853 nnuidp) { 854 LIST_REMOVE(nuidp, nu_hash); 855 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 856 if (nuidp->nu_flag & NU_NAM) 857 kfree(nuidp->nu_nam, M_SONAME); 858 kfree((caddr_t)nuidp, M_NFSUID); 859 } 860 crit_enter(); /* XXX doesn't do anything any more */ 861 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 862 nnwp = nwp->nd_tq.le_next; 863 LIST_REMOVE(nwp, nd_tq); 864 kfree((caddr_t)nwp, M_NFSRVDESC); 865 } 866 LIST_INIT(&slp->ns_tq); 867 crit_exit(); 868 869 nfsrv_slpderef(slp); 870 } 871 } 872 873 /* 874 * Derefence a server socket structure. If it has no more references and 875 * is no longer valid, you can throw it away. 876 * 877 * Must be holding nfs_token! 878 */ 879 void 880 nfsrv_slpderef(struct nfssvc_sock *slp) 881 { 882 struct file *fp; 883 884 ASSERT_LWKT_TOKEN_HELD(&nfs_token); 885 if (slp->ns_sref == 1) { 886 KKASSERT((slp->ns_flag & SLP_VALID) == 0); 887 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 888 slp->ns_sref = 0; 889 fp = slp->ns_fp; 890 slp->ns_fp = NULL; 891 slp->ns_so = NULL; 892 if (fp) 893 closef(fp, NULL); 894 kfree((caddr_t)slp, M_NFSSVC); 895 } else { 896 --slp->ns_sref; 897 } 898 } 899 900 void 901 nfsrv_slpref(struct nfssvc_sock *slp) 902 { 903 ASSERT_LWKT_TOKEN_HELD(&nfs_token); 904 ++slp->ns_sref; 905 } 906 907 /* 908 * Lock a socket against others. 909 * 910 * Returns 0 on failure, 1 on success. 911 */ 912 int 913 nfs_slplock(struct nfssvc_sock *slp, int wait) 914 { 915 mtx_t mtx = &slp->ns_solock; 916 917 if (wait) { 918 mtx_lock_ex(mtx, "nfsslplck", 0, 0); 919 return(1); 920 } else if (mtx_lock_ex_try(mtx) == 0) { 921 return(1); 922 } else { 923 return(0); 924 } 925 } 926 927 /* 928 * Unlock the stream socket for others. 929 */ 930 void 931 nfs_slpunlock(struct nfssvc_sock *slp) 932 { 933 mtx_t mtx = &slp->ns_solock; 934 935 mtx_unlock(mtx); 936 } 937 938 /* 939 * Initialize the data structures for the server. 940 * Handshake with any new nfsds starting up to avoid any chance of 941 * corruption. 942 */ 943 void 944 nfsrv_init(int terminating) 945 { 946 struct nfssvc_sock *slp, *nslp; 947 948 lwkt_gettoken(&nfs_token); 949 if (nfssvc_sockhead_flag & SLP_INIT) 950 panic("nfsd init"); 951 nfssvc_sockhead_flag |= SLP_INIT; 952 953 if (terminating) { 954 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) { 955 nfsrv_slpref(slp); 956 lwkt_gettoken(&slp->ns_token); 957 if (slp->ns_flag & SLP_VALID) 958 nfsrv_zapsock(slp); 959 lwkt_reltoken(&slp->ns_token); 960 nfsrv_slpderef(slp); 961 } 962 nfsrv_cleancache(); /* And clear out server cache */ 963 } else { 964 nfs_pub.np_valid = 0; 965 } 966 967 TAILQ_INIT(&nfssvc_sockhead); 968 nfssvc_sockhead_flag &= ~SLP_INIT; 969 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 970 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 971 wakeup((caddr_t)&nfssvc_sockhead); 972 } 973 974 TAILQ_INIT(&nfsd_head); 975 nfsd_head_flag &= ~NFSD_CHECKSLP; 976 977 lwkt_reltoken(&nfs_token); 978 979 #if 0 980 nfs_udpsock = (struct nfssvc_sock *) 981 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 982 mtx_init(&nfs_udpsock->ns_solock); 983 STAILQ_INIT(&nfs_udpsock->ns_rec); 984 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 985 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 986 987 nfs_cltpsock = (struct nfssvc_sock *) 988 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 989 mtx_init(&nfs_cltpsock->ns_solock); 990 STAILQ_INIT(&nfs_cltpsock->ns_rec); 991 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 992 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 993 #endif 994 } 995 996 /* 997 * Add entries to the server monitor log. 998 */ 999 static void 1000 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 1001 { 1002 struct drt *rt; 1003 1004 rt = &nfsdrt.drt[nfsdrt.pos]; 1005 if (cacherep == RC_DOIT) 1006 rt->flag = 0; 1007 else if (cacherep == RC_REPLY) 1008 rt->flag = DRT_CACHEREPLY; 1009 else 1010 rt->flag = DRT_CACHEDROP; 1011 if (sotype == SOCK_STREAM) 1012 rt->flag |= DRT_TCP; 1013 if (nd->nd_flag & ND_NFSV3) 1014 rt->flag |= DRT_NFSV3; 1015 rt->proc = nd->nd_procnum; 1016 if (nd->nd_nam->sa_family == AF_INET) 1017 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 1018 else 1019 rt->ipadr = INADDR_ANY; 1020 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 1021 getmicrotime(&rt->tstamp); 1022 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1023 } 1024 #endif /* NFS_NOSERVER */ 1025 1026 /* 1027 * Get an authorization string for the uid by having the mount_nfs sitting 1028 * on this mount point porpous out of the kernel and do it. 1029 */ 1030 int 1031 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, 1032 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str, 1033 int *verf_len, NFSKERBKEY_T key /* return session key */) 1034 { 1035 int error = 0; 1036 1037 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 1038 nmp->nm_state |= NFSSTA_WANTAUTH; 1039 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 1040 "nfsauth1", 2 * hz); 1041 error = nfs_sigintr(nmp, rep, rep->r_td); 1042 if (error) { 1043 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1044 return (error); 1045 } 1046 } 1047 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1048 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1049 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1050 nmp->nm_verfstr = verf_str; 1051 nmp->nm_verflen = *verf_len; 1052 nmp->nm_authuid = cred->cr_uid; 1053 wakeup((caddr_t)&nmp->nm_authstr); 1054 1055 /* 1056 * And wait for mount_nfs to do its stuff. 1057 */ 1058 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1059 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1060 "nfsauth2", 2 * hz); 1061 error = nfs_sigintr(nmp, rep, rep->r_td); 1062 } 1063 if (nmp->nm_state & NFSSTA_AUTHERR) { 1064 nmp->nm_state &= ~NFSSTA_AUTHERR; 1065 error = EAUTH; 1066 } 1067 if (error) 1068 kfree((caddr_t)*auth_str, M_TEMP); 1069 else { 1070 *auth_len = nmp->nm_authlen; 1071 *verf_len = nmp->nm_verflen; 1072 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (NFSKERBKEY_T)); 1073 } 1074 nmp->nm_state &= ~NFSSTA_HASAUTH; 1075 nmp->nm_state |= NFSSTA_WAITAUTH; 1076 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1077 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1078 wakeup((caddr_t)&nmp->nm_authtype); 1079 } 1080 return (error); 1081 } 1082 1083 /* 1084 * Get a nickname authenticator and verifier. 1085 */ 1086 int 1087 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str, 1088 int *auth_len, char *verf_str, int verf_len) 1089 { 1090 struct nfsuid *nuidp; 1091 u_int32_t *nickp, *verfp; 1092 struct timeval ktvout; 1093 1094 #ifdef DIAGNOSTIC 1095 if (verf_len < (4 * NFSX_UNSIGNED)) 1096 panic("nfs_getnickauth verf too small"); 1097 #endif 1098 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1099 nuidp != NULL; nuidp = nuidp->nu_hash.le_next) { 1100 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1101 break; 1102 } 1103 if (!nuidp || nuidp->nu_expire < time_uptime) 1104 return (EACCES); 1105 1106 /* 1107 * Move to the end of the lru list (end of lru == most recently used). 1108 */ 1109 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1110 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1111 1112 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1113 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1114 *nickp = txdr_unsigned(nuidp->nu_nickname); 1115 *auth_str = (char *)nickp; 1116 *auth_len = 2 * NFSX_UNSIGNED; 1117 1118 /* 1119 * Now we must encrypt the verifier and package it up. 1120 */ 1121 verfp = (u_int32_t *)verf_str; 1122 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1123 if (time_second != nuidp->nu_timestamp.tv_sec || 1124 (time_second == nuidp->nu_timestamp.tv_sec && 1125 time_second > nuidp->nu_timestamp.tv_usec)) /* XXX */ 1126 getmicrotime(&nuidp->nu_timestamp); 1127 else 1128 nuidp->nu_timestamp.tv_usec++; 1129 1130 /* 1131 * Now encrypt the timestamp verifier in ecb mode using the session 1132 * key. 1133 */ 1134 #ifdef NFSKERB 1135 XXX 1136 #else 1137 ktvout.tv_sec = 0; 1138 ktvout.tv_usec = 0; 1139 #endif 1140 1141 *verfp++ = ktvout.tv_sec; 1142 *verfp++ = ktvout.tv_usec; 1143 *verfp = 0; 1144 return (0); 1145 } 1146 1147 /* 1148 * Save the current nickname in a hash list entry on the mount point. 1149 */ 1150 int 1151 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len, 1152 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, 1153 struct mbuf *mrep) 1154 { 1155 struct nfsuid *nuidp; 1156 u_int32_t *tl; 1157 struct timeval ktvin, ktvout; 1158 u_int32_t nick; 1159 int deltasec, error = 0; 1160 struct nfsm_info info; 1161 1162 info.md = *mdp; 1163 info.dpos = *dposp; 1164 info.mrep = mrep; 1165 1166 if (len == (3 * NFSX_UNSIGNED)) { 1167 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED)); 1168 ktvin.tv_sec = *tl++; 1169 ktvin.tv_usec = *tl++; 1170 nick = fxdr_unsigned(u_int32_t, *tl); 1171 1172 /* 1173 * Decrypt the timestamp in ecb mode. 1174 */ 1175 #ifdef NFSKERB 1176 XXX 1177 #else 1178 ktvout.tv_sec = 0; 1179 ktvout.tv_usec = 0; 1180 #endif 1181 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1182 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1183 deltasec = time_second - ktvout.tv_sec; 1184 if (deltasec < 0) 1185 deltasec = -deltasec; 1186 /* 1187 * If ok, add it to the hash list for the mount point. 1188 */ 1189 if (deltasec <= NFS_KERBCLOCKSKEW) { 1190 if (nmp->nm_numuids < nuidhash_max) { 1191 nmp->nm_numuids++; 1192 nuidp = (struct nfsuid *) 1193 kmalloc(sizeof (struct nfsuid), M_NFSUID, 1194 M_WAITOK); 1195 } else { 1196 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead); 1197 LIST_REMOVE(nuidp, nu_hash); 1198 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1199 nu_lru); 1200 } 1201 nuidp->nu_flag = 0; 1202 nuidp->nu_cr.cr_uid = cred->cr_uid; 1203 nuidp->nu_expire = time_uptime + NFS_KERBTTL; 1204 nuidp->nu_timestamp = ktvout; 1205 nuidp->nu_nickname = nick; 1206 bcopy(key, nuidp->nu_key, sizeof (NFSKERBKEY_T)); 1207 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1208 nu_lru); 1209 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1210 nuidp, nu_hash); 1211 } 1212 } else { 1213 ERROROUT(nfsm_adv(&info, nfsm_rndup(len))); 1214 } 1215 nfsmout: 1216 *mdp = info.md; 1217 *dposp = info.dpos; 1218 return (error); 1219 } 1220