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 1.7 2003/07/26 21:48:49 rob 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/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/domain.h> 57 #include <sys/protosw.h> 58 #include <sys/namei.h> 59 #include <vm/vm_zone.h> 60 61 #include <netinet/in.h> 62 #include <netinet/tcp.h> 63 #include <nfs/xdr_subs.h> 64 #include <nfs/rpcv2.h> 65 #include <nfs/nfsproto.h> 66 #include <nfs/nfs.h> 67 #include <nfs/nfsm_subs.h> 68 #include <nfs/nfsrvcache.h> 69 #include <nfs/nfsmount.h> 70 #include <nfs/nfsnode.h> 71 #include <nfs/nqnfs.h> 72 #include <nfs/nfsrtt.h> 73 74 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 75 76 /* Global defs. */ 77 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *nd, 78 struct nfssvc_sock *slp, 79 struct thread *td, 80 struct mbuf **mreqp)); 81 extern int nfs_numasync; 82 extern time_t nqnfsstarttime; 83 extern int nqsrv_writeslack; 84 extern int nfsrtton; 85 extern struct nfsstats nfsstats; 86 extern int nfsrvw_procrastinate; 87 extern int nfsrvw_procrastinate_v3; 88 static int nuidhash_max = NFS_MAXUIDHASH; 89 90 #ifndef NFS_NOSERVER 91 static void nfsrv_zapsock __P((struct nfssvc_sock *slp)); 92 #endif 93 static int nfssvc_iod __P((struct thread *)); 94 95 #define TRUE 1 96 #define FALSE 0 97 98 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 99 100 SYSCTL_DECL(_vfs_nfs); 101 102 #ifndef NFS_NOSERVER 103 int nfsd_waiting = 0; 104 static struct nfsdrt nfsdrt; 105 static int nfs_numnfsd = 0; 106 static int notstarted = 1; 107 static int modify_flag = 0; 108 static void nfsd_rt __P((int sotype, struct nfsrv_descript *nd, 109 int cacherep)); 110 static int nfssvc_addsock __P((struct file *, struct sockaddr *, 111 struct thread *)); 112 static int nfssvc_nfsd __P((struct nfsd_srvargs *,caddr_t,struct thread *)); 113 114 static int nfs_privport = 0; 115 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, ""); 116 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, ""); 117 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, ""); 118 119 /* 120 * NFS server system calls 121 */ 122 123 #endif /* NFS_NOSERVER */ 124 /* 125 * nfssvc_args(int flag, caddr_t argp) 126 * 127 * Nfs server psuedo system call for the nfsd's 128 * Based on the flag value it either: 129 * - adds a socket to the selection list 130 * - remains in the kernel as an nfsd 131 * - remains in the kernel as an nfsiod 132 */ 133 int 134 nfssvc(struct nfssvc_args *uap) 135 { 136 #ifndef NFS_NOSERVER 137 struct nameidata nd; 138 struct file *fp; 139 struct sockaddr *nam; 140 struct nfsd_args nfsdarg; 141 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 142 struct nfsd_cargs ncd; 143 struct nfsd *nfsd; 144 struct nfssvc_sock *slp; 145 struct nfsuid *nuidp; 146 struct nfsmount *nmp; 147 #endif /* NFS_NOSERVER */ 148 int error; 149 struct thread *td = curthread; 150 151 /* 152 * Must be super user 153 */ 154 error = suser(td); 155 if(error) 156 return (error); 157 KKASSERT(td->td_proc); /* for ucred and p_fd */ 158 while (nfssvc_sockhead_flag & SLP_INIT) { 159 nfssvc_sockhead_flag |= SLP_WANTINIT; 160 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 161 } 162 if (uap->flag & NFSSVC_BIOD) 163 error = nfssvc_iod(td); 164 #ifdef NFS_NOSERVER 165 else 166 error = ENXIO; 167 #else /* !NFS_NOSERVER */ 168 else if (uap->flag & NFSSVC_MNTD) { 169 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 170 if (error) 171 return (error); 172 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 173 ncd.ncd_dirp, td); 174 error = namei(&nd); 175 if (error) 176 return (error); 177 NDFREE(&nd, NDF_ONLY_PNBUF); 178 if ((nd.ni_vp->v_flag & VROOT) == 0) 179 error = EINVAL; 180 nmp = VFSTONFS(nd.ni_vp->v_mount); 181 vput(nd.ni_vp); 182 if (error) 183 return (error); 184 if ((nmp->nm_state & NFSSTA_MNTD) && 185 (uap->flag & NFSSVC_GOTAUTH) == 0) 186 return (0); 187 nmp->nm_state |= NFSSTA_MNTD; 188 error = nqnfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag, 189 uap->argp, td); 190 } else if (uap->flag & NFSSVC_ADDSOCK) { 191 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 192 if (error) 193 return (error); 194 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 195 if (error) 196 return (error); 197 /* 198 * Get the client address for connected sockets. 199 */ 200 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 201 nam = (struct sockaddr *)0; 202 else { 203 error = getsockaddr(&nam, nfsdarg.name, 204 nfsdarg.namelen); 205 if (error) { 206 fdrop(fp, td); 207 return (error); 208 } 209 } 210 error = nfssvc_addsock(fp, nam, td); 211 fdrop(fp, td); 212 } else { 213 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 214 if (error) 215 return (error); 216 if ((uap->flag & NFSSVC_AUTHIN) && 217 ((nfsd = nsd->nsd_nfsd)) != NULL && 218 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 219 slp = nfsd->nfsd_slp; 220 221 /* 222 * First check to see if another nfsd has already 223 * added this credential. 224 */ 225 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 226 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 227 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 228 (!nfsd->nfsd_nd->nd_nam2 || 229 netaddr_match(NU_NETFAM(nuidp), 230 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 231 break; 232 } 233 if (nuidp) { 234 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 235 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 236 } else { 237 /* 238 * Nope, so we will. 239 */ 240 if (slp->ns_numuids < nuidhash_max) { 241 slp->ns_numuids++; 242 nuidp = (struct nfsuid *) 243 malloc(sizeof (struct nfsuid), M_NFSUID, 244 M_WAITOK); 245 } else 246 nuidp = (struct nfsuid *)0; 247 if ((slp->ns_flag & SLP_VALID) == 0) { 248 if (nuidp) 249 free((caddr_t)nuidp, M_NFSUID); 250 } else { 251 if (nuidp == (struct nfsuid *)0) { 252 nuidp = slp->ns_uidlruhead.tqh_first; 253 LIST_REMOVE(nuidp, nu_hash); 254 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 255 nu_lru); 256 if (nuidp->nu_flag & NU_NAM) 257 FREE(nuidp->nu_nam, M_SONAME); 258 } 259 nuidp->nu_flag = 0; 260 nuidp->nu_cr = nsd->nsd_cr; 261 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 262 nuidp->nu_cr.cr_ngroups = NGROUPS; 263 nuidp->nu_cr.cr_ref = 1; 264 nuidp->nu_timestamp = nsd->nsd_timestamp; 265 nuidp->nu_expire = time_second + nsd->nsd_ttl; 266 /* 267 * and save the session key in nu_key. 268 */ 269 bcopy(nsd->nsd_key, nuidp->nu_key, 270 sizeof (nsd->nsd_key)); 271 if (nfsd->nfsd_nd->nd_nam2) { 272 struct sockaddr_in *saddr; 273 274 saddr = (struct sockaddr_in *) 275 nfsd->nfsd_nd->nd_nam2; 276 switch (saddr->sin_family) { 277 case AF_INET: 278 nuidp->nu_flag |= NU_INETADDR; 279 nuidp->nu_inetaddr = 280 saddr->sin_addr.s_addr; 281 break; 282 case AF_ISO: 283 default: 284 nuidp->nu_flag |= NU_NAM; 285 nuidp->nu_nam = 286 dup_sockaddr(nfsd->nfsd_nd-> 287 nd_nam2, 1); 288 break; 289 }; 290 } 291 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 292 nu_lru); 293 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 294 nuidp, nu_hash); 295 nfsrv_setcred(&nuidp->nu_cr, 296 &nfsd->nfsd_nd->nd_cr); 297 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 298 } 299 } 300 } 301 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 302 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 303 error = nfssvc_nfsd(nsd, uap->argp, td); 304 } 305 #endif /* NFS_NOSERVER */ 306 if (error == EINTR || error == ERESTART) 307 error = 0; 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, s; 322 323 so = (struct socket *)fp->f_data; 324 #if 0 325 tslp = (struct nfssvc_sock *)0; 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 FREE(mynam, M_SONAME); 334 return (EPERM); 335 } 336 } 337 #endif 338 if (so->so_type == SOCK_STREAM) 339 siz = NFS_MAXPACKET + sizeof (u_long); 340 else 341 siz = NFS_MAXPACKET; 342 error = soreserve(so, siz, siz); 343 if (error) { 344 if (mynam != NULL) 345 FREE(mynam, M_SONAME); 346 return (error); 347 } 348 349 /* 350 * Set protocol specific options { for now TCP only } and 351 * reserve some space. For datagram sockets, this can get called 352 * repeatedly for the same socket, but that isn't harmful. 353 */ 354 if (so->so_type == SOCK_STREAM) { 355 struct sockopt sopt; 356 int val; 357 358 bzero(&sopt, sizeof sopt); 359 sopt.sopt_level = SOL_SOCKET; 360 sopt.sopt_name = SO_KEEPALIVE; 361 sopt.sopt_val = &val; 362 sopt.sopt_valsize = sizeof val; 363 val = 1; 364 sosetopt(so, &sopt); 365 } 366 if (so->so_proto->pr_domain->dom_family == AF_INET && 367 so->so_proto->pr_protocol == IPPROTO_TCP) { 368 struct sockopt sopt; 369 int val; 370 371 bzero(&sopt, sizeof sopt); 372 sopt.sopt_level = IPPROTO_TCP; 373 sopt.sopt_name = TCP_NODELAY; 374 sopt.sopt_val = &val; 375 sopt.sopt_valsize = sizeof val; 376 val = 1; 377 sosetopt(so, &sopt); 378 } 379 so->so_rcv.sb_flags &= ~SB_NOINTR; 380 so->so_rcv.sb_timeo = 0; 381 so->so_snd.sb_flags &= ~SB_NOINTR; 382 so->so_snd.sb_timeo = 0; 383 384 slp = (struct nfssvc_sock *) 385 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 386 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 387 STAILQ_INIT(&slp->ns_rec); 388 TAILQ_INIT(&slp->ns_uidlruhead); 389 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 390 391 slp->ns_so = so; 392 slp->ns_nam = mynam; 393 fp->f_count++; 394 slp->ns_fp = fp; 395 s = splnet(); 396 so->so_upcallarg = (caddr_t)slp; 397 so->so_upcall = nfsrv_rcv; 398 so->so_rcv.sb_flags |= SB_UPCALL; 399 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 400 nfsrv_wakenfsd(slp); 401 splx(s); 402 return (0); 403 } 404 405 /* 406 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 407 * until it is killed by a signal. 408 */ 409 static int 410 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td) 411 { 412 int siz; 413 struct nfssvc_sock *slp; 414 struct nfsd *nfsd = nsd->nsd_nfsd; 415 struct nfsrv_descript *nd = NULL; 416 struct mbuf *m, *mreq; 417 int error = 0, cacherep, s, sotype, writes_todo; 418 int procrastinate; 419 u_quad_t cur_usec; 420 421 #ifndef nolint 422 cacherep = RC_DOIT; 423 writes_todo = 0; 424 #endif 425 if (nfsd == (struct nfsd *)0) { 426 nsd->nsd_nfsd = nfsd = (struct nfsd *) 427 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 428 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 429 s = splnet(); 430 nfsd->nfsd_td = td; 431 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 432 nfs_numnfsd++; 433 } else 434 s = splnet(); 435 436 /* 437 * Loop getting rpc requests until SIGKILL. 438 */ 439 for (;;) { 440 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 441 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 442 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 443 nfsd->nfsd_flag |= NFSD_WAITING; 444 nfsd_waiting++; 445 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0); 446 nfsd_waiting--; 447 if (error) 448 goto done; 449 } 450 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 451 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 452 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 453 slp = slp->ns_chain.tqe_next) { 454 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 455 == (SLP_VALID | SLP_DOREC)) { 456 slp->ns_flag &= ~SLP_DOREC; 457 slp->ns_sref++; 458 nfsd->nfsd_slp = slp; 459 break; 460 } 461 } 462 if (slp == 0) 463 nfsd_head_flag &= ~NFSD_CHECKSLP; 464 } 465 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 466 continue; 467 if (slp->ns_flag & SLP_VALID) { 468 if (slp->ns_flag & SLP_DISCONN) 469 nfsrv_zapsock(slp); 470 else if (slp->ns_flag & SLP_NEEDQ) { 471 slp->ns_flag &= ~SLP_NEEDQ; 472 (void) nfs_slplock(slp, 1); 473 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 474 M_WAIT); 475 nfs_slpunlock(slp); 476 } 477 error = nfsrv_dorec(slp, nfsd, &nd); 478 cur_usec = nfs_curusec(); 479 if (error && slp->ns_tq.lh_first && 480 slp->ns_tq.lh_first->nd_time <= cur_usec) { 481 error = 0; 482 cacherep = RC_DOIT; 483 writes_todo = 1; 484 } else 485 writes_todo = 0; 486 nfsd->nfsd_flag |= NFSD_REQINPROG; 487 } 488 } else { 489 error = 0; 490 slp = nfsd->nfsd_slp; 491 } 492 if (error || (slp->ns_flag & SLP_VALID) == 0) { 493 if (nd) { 494 free((caddr_t)nd, M_NFSRVDESC); 495 nd = NULL; 496 } 497 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 498 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 499 nfsrv_slpderef(slp); 500 continue; 501 } 502 splx(s); 503 sotype = slp->ns_so->so_type; 504 if (nd) { 505 getmicrotime(&nd->nd_starttime); 506 if (nd->nd_nam2) 507 nd->nd_nam = nd->nd_nam2; 508 else 509 nd->nd_nam = slp->ns_nam; 510 511 /* 512 * Check to see if authorization is needed. 513 */ 514 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 515 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 516 nsd->nsd_haddr = 517 ((struct sockaddr_in *) 518 nd->nd_nam)->sin_addr.s_addr; 519 nsd->nsd_authlen = nfsd->nfsd_authlen; 520 nsd->nsd_verflen = nfsd->nfsd_verflen; 521 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 522 nfsd->nfsd_authlen) && 523 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 524 nfsd->nfsd_verflen) && 525 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 526 return (ENEEDAUTH); 527 cacherep = RC_DROPIT; 528 } else 529 cacherep = nfsrv_getcache(nd, slp, &mreq); 530 531 /* 532 * Check for just starting up for NQNFS and send 533 * fake "try again later" replies to the NQNFS clients. 534 */ 535 if (notstarted && nqnfsstarttime <= time_second) { 536 if (modify_flag) { 537 nqnfsstarttime = time_second + nqsrv_writeslack; 538 modify_flag = 0; 539 } else 540 notstarted = 0; 541 } 542 if (notstarted) { 543 if ((nd->nd_flag & ND_NQNFS) == 0) 544 cacherep = RC_DROPIT; 545 else if (nd->nd_procnum != NFSPROC_WRITE) { 546 nd->nd_procnum = NFSPROC_NOOP; 547 nd->nd_repstat = NQNFS_TRYLATER; 548 cacherep = RC_DOIT; 549 } else 550 modify_flag = 1; 551 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 552 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 553 nd->nd_procnum = NFSPROC_NOOP; 554 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 555 cacherep = RC_DOIT; 556 } else if (nfs_privport) { 557 /* Check if source port is privileged */ 558 u_short port; 559 struct sockaddr *nam = nd->nd_nam; 560 struct sockaddr_in *sin; 561 562 sin = (struct sockaddr_in *)nam; 563 port = ntohs(sin->sin_port); 564 if (port >= IPPORT_RESERVED && 565 nd->nd_procnum != NFSPROC_NULL) { 566 nd->nd_procnum = NFSPROC_NOOP; 567 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 568 cacherep = RC_DOIT; 569 printf("NFS request from unprivileged port (%s:%d)\n", 570 inet_ntoa(sin->sin_addr), port); 571 } 572 } 573 574 } 575 576 /* 577 * Loop to get all the write rpc relies that have been 578 * gathered together. 579 */ 580 do { 581 switch (cacherep) { 582 case RC_DOIT: 583 if (nd && (nd->nd_flag & ND_NFSV3)) 584 procrastinate = nfsrvw_procrastinate_v3; 585 else 586 procrastinate = nfsrvw_procrastinate; 587 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 588 procrastinate > 0 && !notstarted)) 589 error = nfsrv_writegather(&nd, slp, 590 nfsd->nfsd_td, &mreq); 591 else 592 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 593 slp, nfsd->nfsd_td, &mreq); 594 if (mreq == NULL) 595 break; 596 if (error != 0 && error != NFSERR_RETVOID) { 597 if (nd->nd_procnum != NQNFSPROC_VACATED) 598 nfsstats.srv_errs++; 599 nfsrv_updatecache(nd, FALSE, mreq); 600 if (nd->nd_nam2) 601 FREE(nd->nd_nam2, M_SONAME); 602 break; 603 } 604 nfsstats.srvrpccnt[nd->nd_procnum]++; 605 nfsrv_updatecache(nd, TRUE, mreq); 606 nd->nd_mrep = (struct mbuf *)0; 607 case RC_REPLY: 608 m = mreq; 609 siz = 0; 610 while (m) { 611 siz += m->m_len; 612 m = m->m_next; 613 } 614 if (siz <= 0 || siz > NFS_MAXPACKET) { 615 printf("mbuf siz=%d\n",siz); 616 panic("Bad nfs svc reply"); 617 } 618 m = mreq; 619 m->m_pkthdr.len = siz; 620 m->m_pkthdr.rcvif = (struct ifnet *)0; 621 /* 622 * For stream protocols, prepend a Sun RPC 623 * Record Mark. 624 */ 625 if (sotype == SOCK_STREAM) { 626 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 627 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 628 } 629 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 630 (void) nfs_slplock(slp, 1); 631 if (slp->ns_flag & SLP_VALID) 632 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 633 else { 634 error = EPIPE; 635 m_freem(m); 636 } 637 if (nfsrtton) 638 nfsd_rt(sotype, nd, cacherep); 639 if (nd->nd_nam2) 640 FREE(nd->nd_nam2, M_SONAME); 641 if (nd->nd_mrep) 642 m_freem(nd->nd_mrep); 643 if (error == EPIPE) 644 nfsrv_zapsock(slp); 645 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 646 nfs_slpunlock(slp); 647 if (error == EINTR || error == ERESTART) { 648 free((caddr_t)nd, M_NFSRVDESC); 649 nfsrv_slpderef(slp); 650 s = splnet(); 651 goto done; 652 } 653 break; 654 case RC_DROPIT: 655 if (nfsrtton) 656 nfsd_rt(sotype, nd, cacherep); 657 m_freem(nd->nd_mrep); 658 if (nd->nd_nam2) 659 FREE(nd->nd_nam2, M_SONAME); 660 break; 661 }; 662 if (nd) { 663 FREE((caddr_t)nd, M_NFSRVDESC); 664 nd = NULL; 665 } 666 667 /* 668 * Check to see if there are outstanding writes that 669 * need to be serviced. 670 */ 671 cur_usec = nfs_curusec(); 672 s = splsoftclock(); 673 if (slp->ns_tq.lh_first && 674 slp->ns_tq.lh_first->nd_time <= cur_usec) { 675 cacherep = RC_DOIT; 676 writes_todo = 1; 677 } else 678 writes_todo = 0; 679 splx(s); 680 } while (writes_todo); 681 s = splnet(); 682 if (nfsrv_dorec(slp, nfsd, &nd)) { 683 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 684 nfsd->nfsd_slp = NULL; 685 nfsrv_slpderef(slp); 686 } 687 } 688 done: 689 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 690 splx(s); 691 free((caddr_t)nfsd, M_NFSD); 692 nsd->nsd_nfsd = (struct nfsd *)0; 693 if (--nfs_numnfsd == 0) 694 nfsrv_init(TRUE); /* Reinitialize everything */ 695 return (error); 696 } 697 698 /* 699 * Shut down a socket associated with an nfssvc_sock structure. 700 * Should be called with the send lock set, if required. 701 * The trick here is to increment the sref at the start, so that the nfsds 702 * will stop using it and clear ns_flag at the end so that it will not be 703 * reassigned during cleanup. 704 */ 705 static void 706 nfsrv_zapsock(slp) 707 struct nfssvc_sock *slp; 708 { 709 struct nfsuid *nuidp, *nnuidp; 710 struct nfsrv_descript *nwp, *nnwp; 711 struct socket *so; 712 struct file *fp; 713 struct nfsrv_rec *rec; 714 int s; 715 716 slp->ns_flag &= ~SLP_ALLFLAGS; 717 fp = slp->ns_fp; 718 if (fp) { 719 slp->ns_fp = (struct file *)0; 720 so = slp->ns_so; 721 so->so_rcv.sb_flags &= ~SB_UPCALL; 722 so->so_upcall = NULL; 723 so->so_upcallarg = NULL; 724 soshutdown(so, 2); 725 closef(fp, NULL); 726 if (slp->ns_nam) 727 FREE(slp->ns_nam, M_SONAME); 728 m_freem(slp->ns_raw); 729 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 730 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 731 if (rec->nr_address) 732 FREE(rec->nr_address, M_SONAME); 733 m_freem(rec->nr_packet); 734 free(rec, M_NFSRVDESC); 735 } 736 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 737 nuidp = nnuidp) { 738 nnuidp = nuidp->nu_lru.tqe_next; 739 LIST_REMOVE(nuidp, nu_hash); 740 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 741 if (nuidp->nu_flag & NU_NAM) 742 FREE(nuidp->nu_nam, M_SONAME); 743 free((caddr_t)nuidp, M_NFSUID); 744 } 745 s = splsoftclock(); 746 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 747 nnwp = nwp->nd_tq.le_next; 748 LIST_REMOVE(nwp, nd_tq); 749 free((caddr_t)nwp, M_NFSRVDESC); 750 } 751 LIST_INIT(&slp->ns_tq); 752 splx(s); 753 } 754 } 755 756 /* 757 * Derefence a server socket structure. If it has no more references and 758 * is no longer valid, you can throw it away. 759 */ 760 void 761 nfsrv_slpderef(slp) 762 struct nfssvc_sock *slp; 763 { 764 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 765 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 766 free((caddr_t)slp, M_NFSSVC); 767 } 768 } 769 770 /* 771 * Lock a socket against others. 772 */ 773 int 774 nfs_slplock(slp, wait) 775 struct nfssvc_sock *slp; 776 int wait; 777 { 778 int *statep = &slp->ns_solock; 779 780 if (!wait && (*statep & NFSSTA_SNDLOCK)) 781 return(0); /* already locked, fail */ 782 while (*statep & NFSSTA_SNDLOCK) { 783 *statep |= NFSSTA_WANTSND; 784 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0); 785 } 786 *statep |= NFSSTA_SNDLOCK; 787 return (1); 788 } 789 790 /* 791 * Unlock the stream socket for others. 792 */ 793 void 794 nfs_slpunlock(slp) 795 struct nfssvc_sock *slp; 796 { 797 int *statep = &slp->ns_solock; 798 799 if ((*statep & NFSSTA_SNDLOCK) == 0) 800 panic("nfs slpunlock"); 801 *statep &= ~NFSSTA_SNDLOCK; 802 if (*statep & NFSSTA_WANTSND) { 803 *statep &= ~NFSSTA_WANTSND; 804 wakeup((caddr_t)statep); 805 } 806 } 807 808 /* 809 * Initialize the data structures for the server. 810 * Handshake with any new nfsds starting up to avoid any chance of 811 * corruption. 812 */ 813 void 814 nfsrv_init(terminating) 815 int terminating; 816 { 817 struct nfssvc_sock *slp, *nslp; 818 819 if (nfssvc_sockhead_flag & SLP_INIT) 820 panic("nfsd init"); 821 nfssvc_sockhead_flag |= SLP_INIT; 822 if (terminating) { 823 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 824 nslp = slp->ns_chain.tqe_next; 825 if (slp->ns_flag & SLP_VALID) 826 nfsrv_zapsock(slp); 827 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 828 free((caddr_t)slp, M_NFSSVC); 829 } 830 nfsrv_cleancache(); /* And clear out server cache */ 831 } else 832 nfs_pub.np_valid = 0; 833 834 TAILQ_INIT(&nfssvc_sockhead); 835 nfssvc_sockhead_flag &= ~SLP_INIT; 836 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 837 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 838 wakeup((caddr_t)&nfssvc_sockhead); 839 } 840 841 TAILQ_INIT(&nfsd_head); 842 nfsd_head_flag &= ~NFSD_CHECKSLP; 843 844 #if 0 845 nfs_udpsock = (struct nfssvc_sock *) 846 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 847 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 848 STAILQ_INIT(&nfs_udpsock->ns_rec); 849 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 850 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 851 852 nfs_cltpsock = (struct nfssvc_sock *) 853 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 854 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 855 STAILQ_INIT(&nfs_cltpsock->ns_rec); 856 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 857 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 858 #endif 859 } 860 861 /* 862 * Add entries to the server monitor log. 863 */ 864 static void 865 nfsd_rt(sotype, nd, cacherep) 866 int sotype; 867 struct nfsrv_descript *nd; 868 int cacherep; 869 { 870 struct drt *rt; 871 872 rt = &nfsdrt.drt[nfsdrt.pos]; 873 if (cacherep == RC_DOIT) 874 rt->flag = 0; 875 else if (cacherep == RC_REPLY) 876 rt->flag = DRT_CACHEREPLY; 877 else 878 rt->flag = DRT_CACHEDROP; 879 if (sotype == SOCK_STREAM) 880 rt->flag |= DRT_TCP; 881 if (nd->nd_flag & ND_NQNFS) 882 rt->flag |= DRT_NQNFS; 883 else if (nd->nd_flag & ND_NFSV3) 884 rt->flag |= DRT_NFSV3; 885 rt->proc = nd->nd_procnum; 886 if (nd->nd_nam->sa_family == AF_INET) 887 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 888 else 889 rt->ipadr = INADDR_ANY; 890 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 891 getmicrotime(&rt->tstamp); 892 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 893 } 894 #endif /* NFS_NOSERVER */ 895 896 static int nfs_defect = 0; 897 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 898 899 /* 900 * Asynchronous I/O daemons for client nfs. 901 * They do read-ahead and write-behind operations on the block I/O cache. 902 * Never returns unless it fails or gets killed. 903 */ 904 static int 905 nfssvc_iod(struct thread *td) 906 { 907 struct buf *bp; 908 int i, myiod; 909 struct nfsmount *nmp; 910 int error = 0; 911 912 /* 913 * Assign my position or return error if too many already running 914 */ 915 myiod = -1; 916 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 917 if (nfs_asyncdaemon[i] == 0) { 918 nfs_asyncdaemon[i]++; 919 myiod = i; 920 break; 921 } 922 if (myiod == -1) 923 return (EBUSY); 924 nfs_numasync++; 925 /* 926 * Just loop around doin our stuff until SIGKILL 927 */ 928 for (;;) { 929 while (((nmp = nfs_iodmount[myiod]) == NULL 930 || nmp->nm_bufq.tqh_first == NULL) 931 && error == 0) { 932 if (nmp) 933 nmp->nm_bufqiods--; 934 nfs_iodwant[myiod] = td; 935 nfs_iodmount[myiod] = NULL; 936 error = tsleep((caddr_t)&nfs_iodwant[myiod], 937 PCATCH, "nfsidl", 0); 938 } 939 if (error) { 940 nfs_asyncdaemon[myiod] = 0; 941 if (nmp) 942 nmp->nm_bufqiods--; 943 nfs_iodwant[myiod] = NULL; 944 nfs_iodmount[myiod] = NULL; 945 nfs_numasync--; 946 return (error); 947 } 948 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 949 /* Take one off the front of the list */ 950 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 951 nmp->nm_bufqlen--; 952 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 953 nmp->nm_bufqwant = FALSE; 954 wakeup(&nmp->nm_bufq); 955 } 956 (void) nfs_doio(bp, NULL); 957 /* 958 * If there are more than one iod on this mount, then defect 959 * so that the iods can be shared out fairly between the mounts 960 */ 961 if (nfs_defect && nmp->nm_bufqiods > 1) { 962 NFS_DPF(ASYNCIO, 963 ("nfssvc_iod: iod %d defecting from mount %p\n", 964 myiod, nmp)); 965 nfs_iodmount[myiod] = NULL; 966 nmp->nm_bufqiods--; 967 break; 968 } 969 } 970 } 971 } 972 973 974 /* 975 * Get an authorization string for the uid by having the mount_nfs sitting 976 * on this mount point porpous out of the kernel and do it. 977 */ 978 int 979 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 980 struct nfsmount *nmp; 981 struct nfsreq *rep; 982 struct ucred *cred; 983 char **auth_str; 984 int *auth_len; 985 char *verf_str; 986 int *verf_len; 987 NFSKERBKEY_T key; /* return session key */ 988 { 989 int error = 0; 990 991 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 992 nmp->nm_state |= NFSSTA_WANTAUTH; 993 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 994 "nfsauth1", 2 * hz); 995 error = nfs_sigintr(nmp, rep, rep->r_td); 996 if (error) { 997 nmp->nm_state &= ~NFSSTA_WANTAUTH; 998 return (error); 999 } 1000 } 1001 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1002 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1003 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1004 nmp->nm_verfstr = verf_str; 1005 nmp->nm_verflen = *verf_len; 1006 nmp->nm_authuid = cred->cr_uid; 1007 wakeup((caddr_t)&nmp->nm_authstr); 1008 1009 /* 1010 * And wait for mount_nfs to do its stuff. 1011 */ 1012 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1013 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1014 "nfsauth2", 2 * hz); 1015 error = nfs_sigintr(nmp, rep, rep->r_td); 1016 } 1017 if (nmp->nm_state & NFSSTA_AUTHERR) { 1018 nmp->nm_state &= ~NFSSTA_AUTHERR; 1019 error = EAUTH; 1020 } 1021 if (error) 1022 free((caddr_t)*auth_str, M_TEMP); 1023 else { 1024 *auth_len = nmp->nm_authlen; 1025 *verf_len = nmp->nm_verflen; 1026 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1027 } 1028 nmp->nm_state &= ~NFSSTA_HASAUTH; 1029 nmp->nm_state |= NFSSTA_WAITAUTH; 1030 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1031 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1032 wakeup((caddr_t)&nmp->nm_authtype); 1033 } 1034 return (error); 1035 } 1036 1037 /* 1038 * Get a nickname authenticator and verifier. 1039 */ 1040 int 1041 nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 1042 struct nfsmount *nmp; 1043 struct ucred *cred; 1044 char **auth_str; 1045 int *auth_len; 1046 char *verf_str; 1047 int verf_len; 1048 { 1049 struct nfsuid *nuidp; 1050 u_int32_t *nickp, *verfp; 1051 struct timeval ktvin, ktvout; 1052 1053 #ifdef DIAGNOSTIC 1054 if (verf_len < (4 * NFSX_UNSIGNED)) 1055 panic("nfs_getnickauth verf too small"); 1056 #endif 1057 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1058 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1059 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1060 break; 1061 } 1062 if (!nuidp || nuidp->nu_expire < time_second) 1063 return (EACCES); 1064 1065 /* 1066 * Move to the end of the lru list (end of lru == most recently used). 1067 */ 1068 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1069 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1070 1071 nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1072 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1073 *nickp = txdr_unsigned(nuidp->nu_nickname); 1074 *auth_str = (char *)nickp; 1075 *auth_len = 2 * NFSX_UNSIGNED; 1076 1077 /* 1078 * Now we must encrypt the verifier and package it up. 1079 */ 1080 verfp = (u_int32_t *)verf_str; 1081 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1082 if (time_second > nuidp->nu_timestamp.tv_sec || 1083 (time_second == nuidp->nu_timestamp.tv_sec && 1084 time_second > nuidp->nu_timestamp.tv_usec)) 1085 getmicrotime(&nuidp->nu_timestamp); 1086 else 1087 nuidp->nu_timestamp.tv_usec++; 1088 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1089 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1090 1091 /* 1092 * Now encrypt the timestamp verifier in ecb mode using the session 1093 * key. 1094 */ 1095 #ifdef NFSKERB 1096 XXX 1097 #endif 1098 1099 *verfp++ = ktvout.tv_sec; 1100 *verfp++ = ktvout.tv_usec; 1101 *verfp = 0; 1102 return (0); 1103 } 1104 1105 /* 1106 * Save the current nickname in a hash list entry on the mount point. 1107 */ 1108 int 1109 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 1110 struct nfsmount *nmp; 1111 struct ucred *cred; 1112 int len; 1113 NFSKERBKEY_T key; 1114 struct mbuf **mdp; 1115 char **dposp; 1116 struct mbuf *mrep; 1117 { 1118 struct nfsuid *nuidp; 1119 u_int32_t *tl; 1120 int32_t t1; 1121 struct mbuf *md = *mdp; 1122 struct timeval ktvin, ktvout; 1123 u_int32_t nick; 1124 char *dpos = *dposp, *cp2; 1125 int deltasec, error = 0; 1126 1127 if (len == (3 * NFSX_UNSIGNED)) { 1128 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1129 ktvin.tv_sec = *tl++; 1130 ktvin.tv_usec = *tl++; 1131 nick = fxdr_unsigned(u_int32_t, *tl); 1132 1133 /* 1134 * Decrypt the timestamp in ecb mode. 1135 */ 1136 #ifdef NFSKERB 1137 XXX 1138 #endif 1139 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1140 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1141 deltasec = time_second - ktvout.tv_sec; 1142 if (deltasec < 0) 1143 deltasec = -deltasec; 1144 /* 1145 * If ok, add it to the hash list for the mount point. 1146 */ 1147 if (deltasec <= NFS_KERBCLOCKSKEW) { 1148 if (nmp->nm_numuids < nuidhash_max) { 1149 nmp->nm_numuids++; 1150 nuidp = (struct nfsuid *) 1151 malloc(sizeof (struct nfsuid), M_NFSUID, 1152 M_WAITOK); 1153 } else { 1154 nuidp = nmp->nm_uidlruhead.tqh_first; 1155 LIST_REMOVE(nuidp, nu_hash); 1156 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1157 nu_lru); 1158 } 1159 nuidp->nu_flag = 0; 1160 nuidp->nu_cr.cr_uid = cred->cr_uid; 1161 nuidp->nu_expire = time_second + NFS_KERBTTL; 1162 nuidp->nu_timestamp = ktvout; 1163 nuidp->nu_nickname = nick; 1164 bcopy(key, nuidp->nu_key, sizeof (key)); 1165 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1166 nu_lru); 1167 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1168 nuidp, nu_hash); 1169 } 1170 } else 1171 nfsm_adv(nfsm_rndup(len)); 1172 nfsmout: 1173 *mdp = md; 1174 *dposp = dpos; 1175 return (error); 1176 } 1177