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 1.11 2003/09/23 05:03:53 dillon Exp $ 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/file.h> 47 #include <sys/filedesc.h> 48 #include <sys/vnode.h> 49 #include <sys/malloc.h> 50 #include <sys/mount.h> 51 #include <sys/proc.h> 52 #include <sys/buf.h> 53 #include <sys/mbuf.h> 54 #include <sys/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 "xdr_subs.h" 64 #include "rpcv2.h" 65 #include "nfsproto.h" 66 #include "nfs.h" 67 #include "nfsm_subs.h" 68 #include "nfsrvcache.h" 69 #include "nfsmount.h" 70 #include "nfsnode.h" 71 #include "nqnfs.h" 72 #include "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]) (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 (struct nfssvc_sock *slp); 92 #endif 93 static int nfssvc_iod (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 (int sotype, struct nfsrv_descript *nd, 109 int cacherep); 110 static int nfssvc_addsock (struct file *, struct sockaddr *, 111 struct thread *); 112 static int nfssvc_nfsd (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, NAMEI_LOOKUP, CNP_FOLLOW | CNP_LOCKLEAF, 173 UIO_USERSPACE, 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 if (m == NULL) 628 return (ENOBUFS); 629 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 630 } 631 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 632 (void) nfs_slplock(slp, 1); 633 if (slp->ns_flag & SLP_VALID) 634 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 635 else { 636 error = EPIPE; 637 m_freem(m); 638 } 639 if (nfsrtton) 640 nfsd_rt(sotype, nd, cacherep); 641 if (nd->nd_nam2) 642 FREE(nd->nd_nam2, M_SONAME); 643 if (nd->nd_mrep) 644 m_freem(nd->nd_mrep); 645 if (error == EPIPE) 646 nfsrv_zapsock(slp); 647 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 648 nfs_slpunlock(slp); 649 if (error == EINTR || error == ERESTART) { 650 free((caddr_t)nd, M_NFSRVDESC); 651 nfsrv_slpderef(slp); 652 s = splnet(); 653 goto done; 654 } 655 break; 656 case RC_DROPIT: 657 if (nfsrtton) 658 nfsd_rt(sotype, nd, cacherep); 659 m_freem(nd->nd_mrep); 660 if (nd->nd_nam2) 661 FREE(nd->nd_nam2, M_SONAME); 662 break; 663 }; 664 if (nd) { 665 FREE((caddr_t)nd, M_NFSRVDESC); 666 nd = NULL; 667 } 668 669 /* 670 * Check to see if there are outstanding writes that 671 * need to be serviced. 672 */ 673 cur_usec = nfs_curusec(); 674 s = splsoftclock(); 675 if (slp->ns_tq.lh_first && 676 slp->ns_tq.lh_first->nd_time <= cur_usec) { 677 cacherep = RC_DOIT; 678 writes_todo = 1; 679 } else 680 writes_todo = 0; 681 splx(s); 682 } while (writes_todo); 683 s = splnet(); 684 if (nfsrv_dorec(slp, nfsd, &nd)) { 685 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 686 nfsd->nfsd_slp = NULL; 687 nfsrv_slpderef(slp); 688 } 689 } 690 done: 691 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 692 splx(s); 693 free((caddr_t)nfsd, M_NFSD); 694 nsd->nsd_nfsd = (struct nfsd *)0; 695 if (--nfs_numnfsd == 0) 696 nfsrv_init(TRUE); /* Reinitialize everything */ 697 return (error); 698 } 699 700 /* 701 * Shut down a socket associated with an nfssvc_sock structure. 702 * Should be called with the send lock set, if required. 703 * The trick here is to increment the sref at the start, so that the nfsds 704 * will stop using it and clear ns_flag at the end so that it will not be 705 * reassigned during cleanup. 706 */ 707 static void 708 nfsrv_zapsock(slp) 709 struct nfssvc_sock *slp; 710 { 711 struct nfsuid *nuidp, *nnuidp; 712 struct nfsrv_descript *nwp, *nnwp; 713 struct socket *so; 714 struct file *fp; 715 struct nfsrv_rec *rec; 716 int s; 717 718 slp->ns_flag &= ~SLP_ALLFLAGS; 719 fp = slp->ns_fp; 720 if (fp) { 721 slp->ns_fp = (struct file *)0; 722 so = slp->ns_so; 723 so->so_rcv.sb_flags &= ~SB_UPCALL; 724 so->so_upcall = NULL; 725 so->so_upcallarg = NULL; 726 soshutdown(so, 2); 727 closef(fp, NULL); 728 if (slp->ns_nam) 729 FREE(slp->ns_nam, M_SONAME); 730 m_freem(slp->ns_raw); 731 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 732 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 733 if (rec->nr_address) 734 FREE(rec->nr_address, M_SONAME); 735 m_freem(rec->nr_packet); 736 free(rec, M_NFSRVDESC); 737 } 738 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 739 nuidp = nnuidp) { 740 nnuidp = nuidp->nu_lru.tqe_next; 741 LIST_REMOVE(nuidp, nu_hash); 742 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 743 if (nuidp->nu_flag & NU_NAM) 744 FREE(nuidp->nu_nam, M_SONAME); 745 free((caddr_t)nuidp, M_NFSUID); 746 } 747 s = splsoftclock(); 748 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 749 nnwp = nwp->nd_tq.le_next; 750 LIST_REMOVE(nwp, nd_tq); 751 free((caddr_t)nwp, M_NFSRVDESC); 752 } 753 LIST_INIT(&slp->ns_tq); 754 splx(s); 755 } 756 } 757 758 /* 759 * Derefence a server socket structure. If it has no more references and 760 * is no longer valid, you can throw it away. 761 */ 762 void 763 nfsrv_slpderef(slp) 764 struct nfssvc_sock *slp; 765 { 766 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 767 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 768 free((caddr_t)slp, M_NFSSVC); 769 } 770 } 771 772 /* 773 * Lock a socket against others. 774 */ 775 int 776 nfs_slplock(slp, wait) 777 struct nfssvc_sock *slp; 778 int wait; 779 { 780 int *statep = &slp->ns_solock; 781 782 if (!wait && (*statep & NFSSTA_SNDLOCK)) 783 return(0); /* already locked, fail */ 784 while (*statep & NFSSTA_SNDLOCK) { 785 *statep |= NFSSTA_WANTSND; 786 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0); 787 } 788 *statep |= NFSSTA_SNDLOCK; 789 return (1); 790 } 791 792 /* 793 * Unlock the stream socket for others. 794 */ 795 void 796 nfs_slpunlock(slp) 797 struct nfssvc_sock *slp; 798 { 799 int *statep = &slp->ns_solock; 800 801 if ((*statep & NFSSTA_SNDLOCK) == 0) 802 panic("nfs slpunlock"); 803 *statep &= ~NFSSTA_SNDLOCK; 804 if (*statep & NFSSTA_WANTSND) { 805 *statep &= ~NFSSTA_WANTSND; 806 wakeup((caddr_t)statep); 807 } 808 } 809 810 /* 811 * Initialize the data structures for the server. 812 * Handshake with any new nfsds starting up to avoid any chance of 813 * corruption. 814 */ 815 void 816 nfsrv_init(terminating) 817 int terminating; 818 { 819 struct nfssvc_sock *slp, *nslp; 820 821 if (nfssvc_sockhead_flag & SLP_INIT) 822 panic("nfsd init"); 823 nfssvc_sockhead_flag |= SLP_INIT; 824 if (terminating) { 825 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 826 nslp = slp->ns_chain.tqe_next; 827 if (slp->ns_flag & SLP_VALID) 828 nfsrv_zapsock(slp); 829 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 830 free((caddr_t)slp, M_NFSSVC); 831 } 832 nfsrv_cleancache(); /* And clear out server cache */ 833 } else 834 nfs_pub.np_valid = 0; 835 836 TAILQ_INIT(&nfssvc_sockhead); 837 nfssvc_sockhead_flag &= ~SLP_INIT; 838 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 839 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 840 wakeup((caddr_t)&nfssvc_sockhead); 841 } 842 843 TAILQ_INIT(&nfsd_head); 844 nfsd_head_flag &= ~NFSD_CHECKSLP; 845 846 #if 0 847 nfs_udpsock = (struct nfssvc_sock *) 848 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 849 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 850 STAILQ_INIT(&nfs_udpsock->ns_rec); 851 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 852 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 853 854 nfs_cltpsock = (struct nfssvc_sock *) 855 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 856 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 857 STAILQ_INIT(&nfs_cltpsock->ns_rec); 858 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 859 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 860 #endif 861 } 862 863 /* 864 * Add entries to the server monitor log. 865 */ 866 static void 867 nfsd_rt(sotype, nd, cacherep) 868 int sotype; 869 struct nfsrv_descript *nd; 870 int cacherep; 871 { 872 struct drt *rt; 873 874 rt = &nfsdrt.drt[nfsdrt.pos]; 875 if (cacherep == RC_DOIT) 876 rt->flag = 0; 877 else if (cacherep == RC_REPLY) 878 rt->flag = DRT_CACHEREPLY; 879 else 880 rt->flag = DRT_CACHEDROP; 881 if (sotype == SOCK_STREAM) 882 rt->flag |= DRT_TCP; 883 if (nd->nd_flag & ND_NQNFS) 884 rt->flag |= DRT_NQNFS; 885 else if (nd->nd_flag & ND_NFSV3) 886 rt->flag |= DRT_NFSV3; 887 rt->proc = nd->nd_procnum; 888 if (nd->nd_nam->sa_family == AF_INET) 889 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 890 else 891 rt->ipadr = INADDR_ANY; 892 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 893 getmicrotime(&rt->tstamp); 894 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 895 } 896 #endif /* NFS_NOSERVER */ 897 898 static int nfs_defect = 0; 899 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 900 901 /* 902 * Asynchronous I/O daemons for client nfs. 903 * They do read-ahead and write-behind operations on the block I/O cache. 904 * Never returns unless it fails or gets killed. 905 */ 906 static int 907 nfssvc_iod(struct thread *td) 908 { 909 struct buf *bp; 910 int i, myiod; 911 struct nfsmount *nmp; 912 int error = 0; 913 914 /* 915 * Assign my position or return error if too many already running 916 */ 917 myiod = -1; 918 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 919 if (nfs_asyncdaemon[i] == 0) { 920 nfs_asyncdaemon[i]++; 921 myiod = i; 922 break; 923 } 924 if (myiod == -1) 925 return (EBUSY); 926 nfs_numasync++; 927 /* 928 * Just loop around doin our stuff until SIGKILL 929 */ 930 for (;;) { 931 while (((nmp = nfs_iodmount[myiod]) == NULL 932 || nmp->nm_bufq.tqh_first == NULL) 933 && error == 0) { 934 if (nmp) 935 nmp->nm_bufqiods--; 936 nfs_iodwant[myiod] = td; 937 nfs_iodmount[myiod] = NULL; 938 error = tsleep((caddr_t)&nfs_iodwant[myiod], 939 PCATCH, "nfsidl", 0); 940 } 941 if (error) { 942 nfs_asyncdaemon[myiod] = 0; 943 if (nmp) 944 nmp->nm_bufqiods--; 945 nfs_iodwant[myiod] = NULL; 946 nfs_iodmount[myiod] = NULL; 947 nfs_numasync--; 948 return (error); 949 } 950 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 951 /* Take one off the front of the list */ 952 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 953 nmp->nm_bufqlen--; 954 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 955 nmp->nm_bufqwant = FALSE; 956 wakeup(&nmp->nm_bufq); 957 } 958 (void) nfs_doio(bp, NULL); 959 /* 960 * If there are more than one iod on this mount, then defect 961 * so that the iods can be shared out fairly between the mounts 962 */ 963 if (nfs_defect && nmp->nm_bufqiods > 1) { 964 NFS_DPF(ASYNCIO, 965 ("nfssvc_iod: iod %d defecting from mount %p\n", 966 myiod, nmp)); 967 nfs_iodmount[myiod] = NULL; 968 nmp->nm_bufqiods--; 969 break; 970 } 971 } 972 } 973 } 974 975 976 /* 977 * Get an authorization string for the uid by having the mount_nfs sitting 978 * on this mount point porpous out of the kernel and do it. 979 */ 980 int 981 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 982 struct nfsmount *nmp; 983 struct nfsreq *rep; 984 struct ucred *cred; 985 char **auth_str; 986 int *auth_len; 987 char *verf_str; 988 int *verf_len; 989 NFSKERBKEY_T key; /* return session key */ 990 { 991 int error = 0; 992 993 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 994 nmp->nm_state |= NFSSTA_WANTAUTH; 995 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 996 "nfsauth1", 2 * hz); 997 error = nfs_sigintr(nmp, rep, rep->r_td); 998 if (error) { 999 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1000 return (error); 1001 } 1002 } 1003 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1004 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1005 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1006 nmp->nm_verfstr = verf_str; 1007 nmp->nm_verflen = *verf_len; 1008 nmp->nm_authuid = cred->cr_uid; 1009 wakeup((caddr_t)&nmp->nm_authstr); 1010 1011 /* 1012 * And wait for mount_nfs to do its stuff. 1013 */ 1014 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1015 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1016 "nfsauth2", 2 * hz); 1017 error = nfs_sigintr(nmp, rep, rep->r_td); 1018 } 1019 if (nmp->nm_state & NFSSTA_AUTHERR) { 1020 nmp->nm_state &= ~NFSSTA_AUTHERR; 1021 error = EAUTH; 1022 } 1023 if (error) 1024 free((caddr_t)*auth_str, M_TEMP); 1025 else { 1026 *auth_len = nmp->nm_authlen; 1027 *verf_len = nmp->nm_verflen; 1028 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1029 } 1030 nmp->nm_state &= ~NFSSTA_HASAUTH; 1031 nmp->nm_state |= NFSSTA_WAITAUTH; 1032 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1033 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1034 wakeup((caddr_t)&nmp->nm_authtype); 1035 } 1036 return (error); 1037 } 1038 1039 /* 1040 * Get a nickname authenticator and verifier. 1041 */ 1042 int 1043 nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 1044 struct nfsmount *nmp; 1045 struct ucred *cred; 1046 char **auth_str; 1047 int *auth_len; 1048 char *verf_str; 1049 int verf_len; 1050 { 1051 struct nfsuid *nuidp; 1052 u_int32_t *nickp, *verfp; 1053 struct timeval ktvin, ktvout; 1054 1055 #ifdef DIAGNOSTIC 1056 if (verf_len < (4 * NFSX_UNSIGNED)) 1057 panic("nfs_getnickauth verf too small"); 1058 #endif 1059 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1060 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1061 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1062 break; 1063 } 1064 if (!nuidp || nuidp->nu_expire < time_second) 1065 return (EACCES); 1066 1067 /* 1068 * Move to the end of the lru list (end of lru == most recently used). 1069 */ 1070 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1071 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1072 1073 nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1074 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1075 *nickp = txdr_unsigned(nuidp->nu_nickname); 1076 *auth_str = (char *)nickp; 1077 *auth_len = 2 * NFSX_UNSIGNED; 1078 1079 /* 1080 * Now we must encrypt the verifier and package it up. 1081 */ 1082 verfp = (u_int32_t *)verf_str; 1083 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1084 if (time_second > nuidp->nu_timestamp.tv_sec || 1085 (time_second == nuidp->nu_timestamp.tv_sec && 1086 time_second > nuidp->nu_timestamp.tv_usec)) 1087 getmicrotime(&nuidp->nu_timestamp); 1088 else 1089 nuidp->nu_timestamp.tv_usec++; 1090 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1091 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1092 1093 /* 1094 * Now encrypt the timestamp verifier in ecb mode using the session 1095 * key. 1096 */ 1097 #ifdef NFSKERB 1098 XXX 1099 #endif 1100 1101 *verfp++ = ktvout.tv_sec; 1102 *verfp++ = ktvout.tv_usec; 1103 *verfp = 0; 1104 return (0); 1105 } 1106 1107 /* 1108 * Save the current nickname in a hash list entry on the mount point. 1109 */ 1110 int 1111 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 1112 struct nfsmount *nmp; 1113 struct ucred *cred; 1114 int len; 1115 NFSKERBKEY_T key; 1116 struct mbuf **mdp; 1117 char **dposp; 1118 struct mbuf *mrep; 1119 { 1120 struct nfsuid *nuidp; 1121 u_int32_t *tl; 1122 int32_t t1; 1123 struct mbuf *md = *mdp; 1124 struct timeval ktvin, ktvout; 1125 u_int32_t nick; 1126 char *dpos = *dposp, *cp2; 1127 int deltasec, error = 0; 1128 1129 if (len == (3 * NFSX_UNSIGNED)) { 1130 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1131 ktvin.tv_sec = *tl++; 1132 ktvin.tv_usec = *tl++; 1133 nick = fxdr_unsigned(u_int32_t, *tl); 1134 1135 /* 1136 * Decrypt the timestamp in ecb mode. 1137 */ 1138 #ifdef NFSKERB 1139 XXX 1140 #endif 1141 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1142 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1143 deltasec = time_second - ktvout.tv_sec; 1144 if (deltasec < 0) 1145 deltasec = -deltasec; 1146 /* 1147 * If ok, add it to the hash list for the mount point. 1148 */ 1149 if (deltasec <= NFS_KERBCLOCKSKEW) { 1150 if (nmp->nm_numuids < nuidhash_max) { 1151 nmp->nm_numuids++; 1152 nuidp = (struct nfsuid *) 1153 malloc(sizeof (struct nfsuid), M_NFSUID, 1154 M_WAITOK); 1155 } else { 1156 nuidp = nmp->nm_uidlruhead.tqh_first; 1157 LIST_REMOVE(nuidp, nu_hash); 1158 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1159 nu_lru); 1160 } 1161 nuidp->nu_flag = 0; 1162 nuidp->nu_cr.cr_uid = cred->cr_uid; 1163 nuidp->nu_expire = time_second + NFS_KERBTTL; 1164 nuidp->nu_timestamp = ktvout; 1165 nuidp->nu_nickname = nick; 1166 bcopy(key, nuidp->nu_key, sizeof (key)); 1167 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1168 nu_lru); 1169 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1170 nuidp, nu_hash); 1171 } 1172 } else 1173 nfsm_adv(nfsm_rndup(len)); 1174 nfsmout: 1175 *mdp = md; 1176 *dposp = dpos; 1177 return (error); 1178 } 1179