nfs_syscalls.c revision 1.33.6.1.2.1
1 /* $NetBSD: nfs_syscalls.c,v 1.33.6.1.2.1 1999/06/21 01:28:56 thorpej Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 39 */ 40 41 #include "fs_nfs.h" 42 #include "opt_nfsserver.h" 43 #include "opt_iso.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/kernel.h> 48 #include <sys/file.h> 49 #include <sys/stat.h> 50 #include <sys/vnode.h> 51 #include <sys/mount.h> 52 #include <sys/proc.h> 53 #include <sys/uio.h> 54 #include <sys/malloc.h> 55 #include <sys/buf.h> 56 #include <sys/mbuf.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/domain.h> 60 #include <sys/protosw.h> 61 #include <sys/namei.h> 62 #include <sys/syslog.h> 63 #include <sys/filedesc.h> 64 65 #include <sys/syscallargs.h> 66 67 #include <netinet/in.h> 68 #include <netinet/tcp.h> 69 #ifdef ISO 70 #include <netiso/iso.h> 71 #endif 72 #include <nfs/xdr_subs.h> 73 #include <nfs/rpcv2.h> 74 #include <nfs/nfsproto.h> 75 #include <nfs/nfs.h> 76 #include <nfs/nfsm_subs.h> 77 #include <nfs/nfsrvcache.h> 78 #include <nfs/nfsmount.h> 79 #include <nfs/nfsnode.h> 80 #include <nfs/nqnfs.h> 81 #include <nfs/nfsrtt.h> 82 #include <nfs/nfs_var.h> 83 84 void nfsrv_zapsock __P((struct nfssvc_sock *)); 85 86 /* Global defs. */ 87 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *, 88 struct nfssvc_sock *, 89 struct proc *, struct mbuf **)); 90 extern int nfs_numasync; 91 extern time_t nqnfsstarttime; 92 extern int nqsrv_writeslack; 93 extern int nfsrtton; 94 extern struct nfsstats nfsstats; 95 extern int nfsrvw_procrastinate; 96 struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock; 97 int nuidhash_max = NFS_MAXUIDHASH; 98 int nfsd_waiting = 0; 99 #ifdef NFSSERVER 100 static int nfs_numnfsd = 0; 101 static int notstarted = 1; 102 static int modify_flag = 0; 103 static struct nfsdrt nfsdrt; 104 extern struct nfs_public nfs_pub; 105 #endif 106 107 #define TRUE 1 108 #define FALSE 0 109 110 #ifdef NFS 111 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 112 #endif 113 114 #ifdef NFSSERVER 115 static void nfsd_rt __P((int, struct nfsrv_descript *, int)); 116 #endif 117 118 /* 119 * NFS server system calls 120 * getfh() lives here too, but maybe should move to kern/vfs_syscalls.c 121 */ 122 123 /* 124 * Get file handle system call 125 */ 126 int 127 sys_getfh(p, v, retval) 128 struct proc *p; 129 register void *v; 130 register_t *retval; 131 { 132 register struct sys_getfh_args /* { 133 syscallarg(char *) fname; 134 syscallarg(fhandle_t *) fhp; 135 } */ *uap = v; 136 register struct vnode *vp; 137 fhandle_t fh; 138 int error; 139 struct nameidata nd; 140 141 /* 142 * Must be super user 143 */ 144 error = suser(p->p_ucred, &p->p_acflag); 145 if (error) 146 return (error); 147 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 148 SCARG(uap, fname), p); 149 error = namei(&nd); 150 if (error) 151 return (error); 152 vp = nd.ni_vp; 153 memset((caddr_t)&fh, 0, sizeof(fh)); 154 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 155 error = VFS_VPTOFH(vp, &fh.fh_fid); 156 vput(vp); 157 if (error) 158 return (error); 159 error = copyout((caddr_t)&fh, (caddr_t)SCARG(uap, fhp), sizeof (fh)); 160 return (error); 161 } 162 163 /* 164 * Nfs server pseudo system call for the nfsd's 165 * Based on the flag value it either: 166 * - adds a socket to the selection list 167 * - remains in the kernel as an nfsd 168 * - remains in the kernel as an nfsiod 169 */ 170 int 171 sys_nfssvc(p, v, retval) 172 struct proc *p; 173 void *v; 174 register_t *retval; 175 { 176 register struct sys_nfssvc_args /* { 177 syscallarg(int) flag; 178 syscallarg(caddr_t) argp; 179 } */ *uap = v; 180 int error; 181 #ifdef NFS 182 struct nameidata nd; 183 struct nfsmount *nmp; 184 struct nfsd_cargs ncd; 185 #endif 186 #ifdef NFSSERVER 187 struct file *fp; 188 struct mbuf *nam; 189 struct nfsd_args nfsdarg; 190 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 191 struct nfsd *nfsd; 192 struct nfssvc_sock *slp; 193 struct nfsuid *nuidp; 194 #endif 195 196 /* 197 * Must be super user 198 */ 199 error = suser(p->p_ucred, &p->p_acflag); 200 if(error) 201 return (error); 202 while (nfssvc_sockhead_flag & SLP_INIT) { 203 nfssvc_sockhead_flag |= SLP_WANTINIT; 204 (void) tsleep((caddr_t)&nfssvc_sockhead, PSOCK, "nfsd init", 0); 205 } 206 if (SCARG(uap, flag) & NFSSVC_BIOD) { 207 #ifdef NFS 208 error = nfssvc_iod(p); 209 #else 210 error = ENOSYS; 211 #endif 212 } else if (SCARG(uap, flag) & NFSSVC_MNTD) { 213 #ifndef NFS 214 error = ENOSYS; 215 #else 216 error = copyin(SCARG(uap, argp), (caddr_t)&ncd, sizeof (ncd)); 217 if (error) 218 return (error); 219 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 220 ncd.ncd_dirp, p); 221 error = namei(&nd); 222 if (error) 223 return (error); 224 if ((nd.ni_vp->v_flag & VROOT) == 0) 225 error = EINVAL; 226 nmp = VFSTONFS(nd.ni_vp->v_mount); 227 vput(nd.ni_vp); 228 if (error) 229 return (error); 230 if ((nmp->nm_iflag & NFSMNT_MNTD) && 231 (SCARG(uap, flag) & NFSSVC_GOTAUTH) == 0) 232 return (0); 233 nmp->nm_iflag |= NFSMNT_MNTD; 234 error = nqnfs_clientd(nmp, p->p_ucred, &ncd, SCARG(uap, flag), 235 SCARG(uap, argp), p); 236 #endif /* NFS */ 237 } else if (SCARG(uap, flag) & NFSSVC_ADDSOCK) { 238 #ifndef NFSSERVER 239 error = ENOSYS; 240 #else 241 error = copyin(SCARG(uap, argp), (caddr_t)&nfsdarg, 242 sizeof(nfsdarg)); 243 if (error) 244 return (error); 245 /* getsock() will use the descriptor for us */ 246 error = getsock(p->p_fd, nfsdarg.sock, &fp); 247 if (error) 248 return (error); 249 /* 250 * Get the client address for connected sockets. 251 */ 252 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 253 nam = (struct mbuf *)0; 254 else { 255 error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen, 256 MT_SONAME); 257 if (error) { 258 FILE_UNUSE(fp, NULL); 259 return (error); 260 } 261 } 262 error = nfssvc_addsock(fp, nam); 263 FILE_UNUSE(fp, NULL); 264 #endif /* !NFSSERVER */ 265 } else { 266 #ifndef NFSSERVER 267 error = ENOSYS; 268 #else 269 error = copyin(SCARG(uap, argp), (caddr_t)nsd, sizeof (*nsd)); 270 if (error) 271 return (error); 272 if ((SCARG(uap, flag) & NFSSVC_AUTHIN) && 273 ((nfsd = nsd->nsd_nfsd)) != NULL && 274 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 275 slp = nfsd->nfsd_slp; 276 277 /* 278 * First check to see if another nfsd has already 279 * added this credential. 280 */ 281 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 282 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 283 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 284 (!nfsd->nfsd_nd->nd_nam2 || 285 netaddr_match(NU_NETFAM(nuidp), 286 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 287 break; 288 } 289 if (nuidp) { 290 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 291 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 292 } else { 293 /* 294 * Nope, so we will. 295 */ 296 if (slp->ns_numuids < nuidhash_max) { 297 slp->ns_numuids++; 298 nuidp = (struct nfsuid *) 299 malloc(sizeof (struct nfsuid), M_NFSUID, 300 M_WAITOK); 301 } else 302 nuidp = (struct nfsuid *)0; 303 if ((slp->ns_flag & SLP_VALID) == 0) { 304 if (nuidp) 305 free((caddr_t)nuidp, M_NFSUID); 306 } else { 307 if (nuidp == (struct nfsuid *)0) { 308 nuidp = slp->ns_uidlruhead.tqh_first; 309 LIST_REMOVE(nuidp, nu_hash); 310 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 311 nu_lru); 312 if (nuidp->nu_flag & NU_NAM) 313 m_freem(nuidp->nu_nam); 314 } 315 nuidp->nu_flag = 0; 316 nuidp->nu_cr = nsd->nsd_cr; 317 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 318 nuidp->nu_cr.cr_ngroups = NGROUPS; 319 nuidp->nu_cr.cr_ref = 1; 320 nuidp->nu_timestamp = nsd->nsd_timestamp; 321 nuidp->nu_expire = time.tv_sec + nsd->nsd_ttl; 322 /* 323 * and save the session key in nu_key. 324 */ 325 memcpy(nuidp->nu_key, nsd->nsd_key, 326 sizeof(nsd->nsd_key)); 327 if (nfsd->nfsd_nd->nd_nam2) { 328 struct sockaddr_in *saddr; 329 330 saddr = mtod(nfsd->nfsd_nd->nd_nam2, 331 struct sockaddr_in *); 332 switch (saddr->sin_family) { 333 case AF_INET: 334 nuidp->nu_flag |= NU_INETADDR; 335 nuidp->nu_inetaddr = 336 saddr->sin_addr.s_addr; 337 break; 338 case AF_ISO: 339 default: 340 nuidp->nu_flag |= NU_NAM; 341 nuidp->nu_nam = m_copym( 342 nfsd->nfsd_nd->nd_nam2, 0, 343 M_COPYALL, M_WAIT); 344 break; 345 }; 346 } 347 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 348 nu_lru); 349 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 350 nuidp, nu_hash); 351 nfsrv_setcred(&nuidp->nu_cr, 352 &nfsd->nfsd_nd->nd_cr); 353 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 354 } 355 } 356 } 357 if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) && 358 (nfsd = nsd->nsd_nfsd)) 359 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 360 error = nfssvc_nfsd(nsd, SCARG(uap, argp), p); 361 #endif /* !NFSSERVER */ 362 } 363 if (error == EINTR || error == ERESTART) 364 error = 0; 365 return (error); 366 } 367 368 #ifdef NFSSERVER 369 /* 370 * Adds a socket to the list for servicing by nfsds. 371 */ 372 int 373 nfssvc_addsock(fp, mynam) 374 struct file *fp; 375 struct mbuf *mynam; 376 { 377 register struct mbuf *m; 378 register int siz; 379 register struct nfssvc_sock *slp; 380 register struct socket *so; 381 struct nfssvc_sock *tslp; 382 int error, s; 383 384 so = (struct socket *)fp->f_data; 385 tslp = (struct nfssvc_sock *)0; 386 /* 387 * Add it to the list, as required. 388 */ 389 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 390 tslp = nfs_udpsock; 391 if (tslp->ns_flag & SLP_VALID) { 392 m_freem(mynam); 393 return (EPERM); 394 } 395 #ifdef ISO 396 } else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) { 397 tslp = nfs_cltpsock; 398 if (tslp->ns_flag & SLP_VALID) { 399 m_freem(mynam); 400 return (EPERM); 401 } 402 #endif /* ISO */ 403 } 404 if (so->so_type == SOCK_STREAM) 405 siz = NFS_MAXPACKET + sizeof (u_long); 406 else 407 siz = NFS_MAXPACKET; 408 error = soreserve(so, siz, siz); 409 if (error) { 410 m_freem(mynam); 411 return (error); 412 } 413 414 /* 415 * Set protocol specific options { for now TCP only } and 416 * reserve some space. For datagram sockets, this can get called 417 * repeatedly for the same socket, but that isn't harmful. 418 */ 419 if (so->so_type == SOCK_STREAM) { 420 MGET(m, M_WAIT, MT_SOOPTS); 421 *mtod(m, int32_t *) = 1; 422 m->m_len = sizeof(int32_t); 423 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 424 } 425 if (so->so_proto->pr_domain->dom_family == AF_INET && 426 so->so_proto->pr_protocol == IPPROTO_TCP) { 427 MGET(m, M_WAIT, MT_SOOPTS); 428 *mtod(m, int32_t *) = 1; 429 m->m_len = sizeof(int32_t); 430 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 431 } 432 so->so_rcv.sb_flags &= ~SB_NOINTR; 433 so->so_rcv.sb_timeo = 0; 434 so->so_snd.sb_flags &= ~SB_NOINTR; 435 so->so_snd.sb_timeo = 0; 436 if (tslp) 437 slp = tslp; 438 else { 439 slp = (struct nfssvc_sock *) 440 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 441 memset((caddr_t)slp, 0, sizeof (struct nfssvc_sock)); 442 TAILQ_INIT(&slp->ns_uidlruhead); 443 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 444 } 445 slp->ns_so = so; 446 slp->ns_nam = mynam; 447 fp->f_count++; 448 slp->ns_fp = fp; 449 s = splsoftnet(); 450 so->so_upcallarg = (caddr_t)slp; 451 so->so_upcall = nfsrv_rcv; 452 so->so_rcv.sb_flags |= SB_UPCALL; 453 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 454 nfsrv_wakenfsd(slp); 455 splx(s); 456 return (0); 457 } 458 459 /* 460 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 461 * until it is killed by a signal. 462 */ 463 int 464 nfssvc_nfsd(nsd, argp, p) 465 struct nfsd_srvargs *nsd; 466 caddr_t argp; 467 struct proc *p; 468 { 469 register struct mbuf *m; 470 register int siz; 471 register struct nfssvc_sock *slp; 472 register struct socket *so; 473 register int *solockp; 474 struct nfsd *nfsd = nsd->nsd_nfsd; 475 struct nfsrv_descript *nd = NULL; 476 struct mbuf *mreq; 477 int error = 0, cacherep, s, sotype, writes_todo; 478 u_quad_t cur_usec; 479 480 #ifndef nolint 481 cacherep = RC_DOIT; 482 writes_todo = 0; 483 #endif 484 s = splsoftnet(); 485 if (nfsd == (struct nfsd *)0) { 486 nsd->nsd_nfsd = nfsd = (struct nfsd *) 487 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 488 memset((caddr_t)nfsd, 0, sizeof (struct nfsd)); 489 nfsd->nfsd_procp = p; 490 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 491 nfs_numnfsd++; 492 } 493 p->p_holdcnt++; 494 /* 495 * Loop getting rpc requests until SIGKILL. 496 */ 497 for (;;) { 498 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 499 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 500 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 501 nfsd->nfsd_flag |= NFSD_WAITING; 502 nfsd_waiting++; 503 error = tsleep((caddr_t)nfsd, PSOCK | PCATCH, 504 "nfsd", 0); 505 nfsd_waiting--; 506 if (error) 507 goto done; 508 } 509 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 510 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 511 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 512 slp = slp->ns_chain.tqe_next) { 513 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 514 == (SLP_VALID | SLP_DOREC)) { 515 slp->ns_flag &= ~SLP_DOREC; 516 slp->ns_sref++; 517 nfsd->nfsd_slp = slp; 518 break; 519 } 520 } 521 if (slp == 0) 522 nfsd_head_flag &= ~NFSD_CHECKSLP; 523 } 524 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 525 continue; 526 if (slp->ns_flag & SLP_VALID) { 527 if (slp->ns_flag & SLP_DISCONN) 528 nfsrv_zapsock(slp); 529 else if (slp->ns_flag & SLP_NEEDQ) { 530 slp->ns_flag &= ~SLP_NEEDQ; 531 (void) nfs_sndlock(&slp->ns_solock, 532 (struct nfsreq *)0); 533 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 534 M_WAIT); 535 nfs_sndunlock(&slp->ns_solock); 536 } 537 error = nfsrv_dorec(slp, nfsd, &nd); 538 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 539 (u_quad_t)time.tv_usec; 540 if (error && slp->ns_tq.lh_first && 541 slp->ns_tq.lh_first->nd_time <= cur_usec) { 542 error = 0; 543 cacherep = RC_DOIT; 544 writes_todo = 1; 545 } else 546 writes_todo = 0; 547 nfsd->nfsd_flag |= NFSD_REQINPROG; 548 } 549 } else { 550 error = 0; 551 slp = nfsd->nfsd_slp; 552 } 553 if (error || (slp->ns_flag & SLP_VALID) == 0) { 554 if (nd) { 555 free((caddr_t)nd, M_NFSRVDESC); 556 nd = NULL; 557 } 558 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 559 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 560 nfsrv_slpderef(slp); 561 continue; 562 } 563 splx(s); 564 so = slp->ns_so; 565 sotype = so->so_type; 566 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 567 solockp = &slp->ns_solock; 568 else 569 solockp = (int *)0; 570 if (nd) { 571 nd->nd_starttime = time; 572 if (nd->nd_nam2) 573 nd->nd_nam = nd->nd_nam2; 574 else 575 nd->nd_nam = slp->ns_nam; 576 577 /* 578 * Check to see if authorization is needed. 579 */ 580 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 581 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 582 nsd->nsd_haddr = mtod(nd->nd_nam, 583 struct sockaddr_in *)->sin_addr.s_addr; 584 nsd->nsd_authlen = nfsd->nfsd_authlen; 585 nsd->nsd_verflen = nfsd->nfsd_verflen; 586 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 587 nfsd->nfsd_authlen) && 588 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 589 nfsd->nfsd_verflen) && 590 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) { 591 p->p_holdcnt--; 592 return (ENEEDAUTH); 593 } 594 cacherep = RC_DROPIT; 595 } else 596 cacherep = nfsrv_getcache(nd, slp, &mreq); 597 598 /* 599 * Check for just starting up for NQNFS and send 600 * fake "try again later" replies to the NQNFS clients. 601 */ 602 if (notstarted && nqnfsstarttime <= time.tv_sec) { 603 if (modify_flag) { 604 nqnfsstarttime = time.tv_sec + nqsrv_writeslack; 605 modify_flag = 0; 606 } else 607 notstarted = 0; 608 } 609 if (notstarted) { 610 if ((nd->nd_flag & ND_NQNFS) == 0) 611 cacherep = RC_DROPIT; 612 else if (nd->nd_procnum != NFSPROC_WRITE) { 613 nd->nd_procnum = NFSPROC_NOOP; 614 nd->nd_repstat = NQNFS_TRYLATER; 615 cacherep = RC_DOIT; 616 } else 617 modify_flag = 1; 618 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 619 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 620 nd->nd_procnum = NFSPROC_NOOP; 621 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 622 cacherep = RC_DOIT; 623 } 624 } 625 626 /* 627 * Loop to get all the write rpc relies that have been 628 * gathered together. 629 */ 630 do { 631 #ifdef DIAGNOSTIC 632 int lockcount; 633 #endif 634 switch (cacherep) { 635 case RC_DOIT: 636 #ifdef DIAGNOSTIC 637 /* 638 * NFS server procs should neither release 639 * locks already held, nor leave things 640 * locked. Catch this sooner, rather than 641 * later (when we try to relock something we 642 * already have locked). Careful inspection 643 * of the failing routine usually turns up the 644 * lock leak.. once we know what it is.. 645 */ 646 lockcount = p->p_locks; 647 #endif 648 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 649 nfsrvw_procrastinate > 0 && !notstarted)) 650 error = nfsrv_writegather(&nd, slp, 651 nfsd->nfsd_procp, &mreq); 652 else 653 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 654 slp, nfsd->nfsd_procp, &mreq); 655 #ifdef DIAGNOSTIC 656 if (p->p_locks != lockcount) { 657 /* 658 * If you see this panic, audit 659 * nfsrv3_procs[nd->nd_procnum] for vnode 660 * locking errors (usually, it's due to 661 * forgetting to vput() something). 662 */ 663 panic("nfsd: locking botch in op %d", 664 nd ? nd->nd_procnum : -1); 665 } 666 #endif 667 if (mreq == NULL) 668 break; 669 if (error) { 670 if (nd->nd_procnum != NQNFSPROC_VACATED) 671 nfsstats.srv_errs++; 672 nfsrv_updatecache(nd, FALSE, mreq); 673 if (nd->nd_nam2) 674 m_freem(nd->nd_nam2); 675 break; 676 } 677 nfsstats.srvrpccnt[nd->nd_procnum]++; 678 nfsrv_updatecache(nd, TRUE, mreq); 679 nd->nd_mrep = (struct mbuf *)0; 680 case RC_REPLY: 681 m = mreq; 682 siz = 0; 683 while (m) { 684 siz += m->m_len; 685 m = m->m_next; 686 } 687 if (siz <= 0 || siz > NFS_MAXPACKET) { 688 printf("mbuf siz=%d\n",siz); 689 panic("Bad nfs svc reply"); 690 } 691 m = mreq; 692 m->m_pkthdr.len = siz; 693 m->m_pkthdr.rcvif = (struct ifnet *)0; 694 /* 695 * For stream protocols, prepend a Sun RPC 696 * Record Mark. 697 */ 698 if (sotype == SOCK_STREAM) { 699 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 700 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 701 } 702 if (solockp) 703 (void) nfs_sndlock(solockp, (struct nfsreq *)0); 704 if (slp->ns_flag & SLP_VALID) 705 error = nfs_send(so, nd->nd_nam2, m, NULL); 706 else { 707 error = EPIPE; 708 m_freem(m); 709 } 710 if (nfsrtton) 711 nfsd_rt(sotype, nd, cacherep); 712 if (nd->nd_nam2) 713 MFREE(nd->nd_nam2, m); 714 if (nd->nd_mrep) 715 m_freem(nd->nd_mrep); 716 if (error == EPIPE) 717 nfsrv_zapsock(slp); 718 if (solockp) 719 nfs_sndunlock(solockp); 720 if (error == EINTR || error == ERESTART) { 721 free((caddr_t)nd, M_NFSRVDESC); 722 nfsrv_slpderef(slp); 723 s = splsoftnet(); 724 goto done; 725 } 726 break; 727 case RC_DROPIT: 728 if (nfsrtton) 729 nfsd_rt(sotype, nd, cacherep); 730 m_freem(nd->nd_mrep); 731 m_freem(nd->nd_nam2); 732 break; 733 }; 734 if (nd) { 735 FREE((caddr_t)nd, M_NFSRVDESC); 736 nd = NULL; 737 } 738 739 /* 740 * Check to see if there are outstanding writes that 741 * need to be serviced. 742 */ 743 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 744 (u_quad_t)time.tv_usec; 745 s = splsoftclock(); 746 if (slp->ns_tq.lh_first && 747 slp->ns_tq.lh_first->nd_time <= cur_usec) { 748 cacherep = RC_DOIT; 749 writes_todo = 1; 750 } else 751 writes_todo = 0; 752 splx(s); 753 } while (writes_todo); 754 s = splsoftnet(); 755 if (nfsrv_dorec(slp, nfsd, &nd)) { 756 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 757 nfsd->nfsd_slp = NULL; 758 nfsrv_slpderef(slp); 759 } 760 } 761 done: 762 p->p_holdcnt--; 763 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 764 splx(s); 765 free((caddr_t)nfsd, M_NFSD); 766 nsd->nsd_nfsd = (struct nfsd *)0; 767 if (--nfs_numnfsd == 0) 768 nfsrv_init(TRUE); /* Reinitialize everything */ 769 return (error); 770 } 771 772 /* 773 * Shut down a socket associated with an nfssvc_sock structure. 774 * Should be called with the send lock set, if required. 775 * The trick here is to increment the sref at the start, so that the nfsds 776 * will stop using it and clear ns_flag at the end so that it will not be 777 * reassigned during cleanup. 778 */ 779 void 780 nfsrv_zapsock(slp) 781 register struct nfssvc_sock *slp; 782 { 783 register struct nfsuid *nuidp, *nnuidp; 784 register struct nfsrv_descript *nwp, *nnwp; 785 struct socket *so; 786 struct file *fp; 787 struct mbuf *m; 788 int s; 789 790 slp->ns_flag &= ~SLP_ALLFLAGS; 791 fp = slp->ns_fp; 792 if (fp) { 793 FILE_USE(fp); 794 slp->ns_fp = (struct file *)0; 795 so = slp->ns_so; 796 so->so_upcall = NULL; 797 so->so_upcallarg = NULL; 798 so->so_rcv.sb_flags &= ~SB_UPCALL; 799 soshutdown(so, 2); 800 closef(fp, (struct proc *)0); 801 if (slp->ns_nam) 802 MFREE(slp->ns_nam, m); 803 m_freem(slp->ns_raw); 804 m_freem(slp->ns_rec); 805 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 806 nuidp = nnuidp) { 807 nnuidp = nuidp->nu_lru.tqe_next; 808 LIST_REMOVE(nuidp, nu_hash); 809 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 810 if (nuidp->nu_flag & NU_NAM) 811 m_freem(nuidp->nu_nam); 812 free((caddr_t)nuidp, M_NFSUID); 813 } 814 s = splsoftclock(); 815 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 816 nnwp = nwp->nd_tq.le_next; 817 LIST_REMOVE(nwp, nd_tq); 818 free((caddr_t)nwp, M_NFSRVDESC); 819 } 820 LIST_INIT(&slp->ns_tq); 821 splx(s); 822 } 823 } 824 825 /* 826 * Derefence a server socket structure. If it has no more references and 827 * is no longer valid, you can throw it away. 828 */ 829 void 830 nfsrv_slpderef(slp) 831 register struct nfssvc_sock *slp; 832 { 833 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 834 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 835 free((caddr_t)slp, M_NFSSVC); 836 } 837 } 838 839 /* 840 * Initialize the data structures for the server. 841 * Handshake with any new nfsds starting up to avoid any chance of 842 * corruption. 843 */ 844 void 845 nfsrv_init(terminating) 846 int terminating; 847 { 848 register struct nfssvc_sock *slp, *nslp; 849 850 if (nfssvc_sockhead_flag & SLP_INIT) 851 panic("nfsd init"); 852 nfssvc_sockhead_flag |= SLP_INIT; 853 if (terminating) { 854 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 855 nslp = slp->ns_chain.tqe_next; 856 if (slp->ns_flag & SLP_VALID) 857 nfsrv_zapsock(slp); 858 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 859 free((caddr_t)slp, M_NFSSVC); 860 } 861 nfsrv_cleancache(); /* And clear out server cache */ 862 } else 863 nfs_pub.np_valid = 0; 864 865 TAILQ_INIT(&nfssvc_sockhead); 866 nfssvc_sockhead_flag &= ~SLP_INIT; 867 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 868 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 869 wakeup((caddr_t)&nfssvc_sockhead); 870 } 871 872 TAILQ_INIT(&nfsd_head); 873 nfsd_head_flag &= ~NFSD_CHECKSLP; 874 875 nfs_udpsock = (struct nfssvc_sock *) 876 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 877 memset((caddr_t)nfs_udpsock, 0, sizeof (struct nfssvc_sock)); 878 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 879 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 880 881 nfs_cltpsock = (struct nfssvc_sock *) 882 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 883 memset((caddr_t)nfs_cltpsock, 0, sizeof (struct nfssvc_sock)); 884 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 885 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 886 } 887 888 /* 889 * Add entries to the server monitor log. 890 */ 891 static void 892 nfsd_rt(sotype, nd, cacherep) 893 int sotype; 894 register struct nfsrv_descript *nd; 895 int cacherep; 896 { 897 register struct drt *rt; 898 899 rt = &nfsdrt.drt[nfsdrt.pos]; 900 if (cacherep == RC_DOIT) 901 rt->flag = 0; 902 else if (cacherep == RC_REPLY) 903 rt->flag = DRT_CACHEREPLY; 904 else 905 rt->flag = DRT_CACHEDROP; 906 if (sotype == SOCK_STREAM) 907 rt->flag |= DRT_TCP; 908 if (nd->nd_flag & ND_NQNFS) 909 rt->flag |= DRT_NQNFS; 910 else if (nd->nd_flag & ND_NFSV3) 911 rt->flag |= DRT_NFSV3; 912 rt->proc = nd->nd_procnum; 913 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 914 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 915 else 916 rt->ipadr = INADDR_ANY; 917 rt->resptime = ((time.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 918 (time.tv_usec - nd->nd_starttime.tv_usec); 919 rt->tstamp = time; 920 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 921 } 922 #endif /* NFSSERVER */ 923 924 #ifdef NFS 925 926 int nfs_defect = 0; 927 /* 928 * Asynchronous I/O daemons for client nfs. 929 * They do read-ahead and write-behind operations on the block I/O cache. 930 * Never returns unless it fails or gets killed. 931 */ 932 int 933 nfssvc_iod(p) 934 struct proc *p; 935 { 936 register struct buf *bp; 937 register int i, myiod; 938 struct nfsmount *nmp; 939 int error = 0; 940 941 /* 942 * Assign my position or return error if too many already running 943 */ 944 myiod = -1; 945 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 946 if (nfs_asyncdaemon[i] == 0) { 947 myiod = i; 948 break; 949 } 950 if (myiod == -1) 951 return (EBUSY); 952 nfs_asyncdaemon[myiod] = 1; 953 nfs_numasync++; 954 p->p_holdcnt++; 955 /* 956 * Just loop around doin our stuff until SIGKILL 957 */ 958 for (;;) { 959 while (((nmp = nfs_iodmount[myiod]) == NULL 960 || nmp->nm_bufq.tqh_first == NULL) 961 && error == 0) { 962 if (nmp) 963 nmp->nm_bufqiods--; 964 nfs_iodwant[myiod] = p; 965 nfs_iodmount[myiod] = NULL; 966 error = tsleep((caddr_t)&nfs_iodwant[myiod], 967 PWAIT | PCATCH, "nfsidl", 0); 968 } 969 if (error) { 970 if (nmp) 971 nmp->nm_bufqiods--; 972 nfs_iodmount[myiod] = NULL; 973 break; 974 } 975 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 976 /* Take one off the front of the list */ 977 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 978 nmp->nm_bufqlen--; 979 if (nmp->nm_bufqwant && nmp->nm_bufqlen < 2 * nfs_numasync) { 980 nmp->nm_bufqwant = FALSE; 981 wakeup(&nmp->nm_bufq); 982 } 983 if (bp->b_flags & B_READ) 984 (void) nfs_doio(bp, bp->b_rcred, (struct proc *)0); 985 else 986 (void) nfs_doio(bp, bp->b_wcred, (struct proc *)0); 987 /* 988 * If there are more than one iod on this mount, then defect 989 * so that the iods can be shared out fairly between the mounts 990 */ 991 if (nfs_defect && nmp->nm_bufqiods > 1) { 992 nfs_iodmount[myiod] = NULL; 993 nmp->nm_bufqiods--; 994 break; 995 } 996 } 997 } 998 p->p_holdcnt--; 999 nfs_asyncdaemon[myiod] = 0; 1000 nfs_numasync--; 1001 return (error); 1002 } 1003 1004 1005 /* 1006 * Get an authorization string for the uid by having the mount_nfs sitting 1007 * on this mount point porpous out of the kernel and do it. 1008 */ 1009 int 1010 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 1011 register struct nfsmount *nmp; 1012 struct nfsreq *rep; 1013 struct ucred *cred; 1014 char **auth_str; 1015 int *auth_len; 1016 char *verf_str; 1017 int *verf_len; 1018 NFSKERBKEY_T key; /* return session key */ 1019 { 1020 int error = 0; 1021 1022 while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) { 1023 nmp->nm_iflag |= NFSMNT_WANTAUTH; 1024 (void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK, 1025 "nfsauth1", 2 * hz); 1026 error = nfs_sigintr(nmp, rep, rep->r_procp); 1027 if (error) { 1028 nmp->nm_iflag &= ~NFSMNT_WANTAUTH; 1029 return (error); 1030 } 1031 } 1032 nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH); 1033 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1034 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1035 nmp->nm_verfstr = verf_str; 1036 nmp->nm_verflen = *verf_len; 1037 nmp->nm_authuid = cred->cr_uid; 1038 wakeup((caddr_t)&nmp->nm_authstr); 1039 1040 /* 1041 * And wait for mount_nfs to do its stuff. 1042 */ 1043 while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) { 1044 (void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK, 1045 "nfsauth2", 2 * hz); 1046 error = nfs_sigintr(nmp, rep, rep->r_procp); 1047 } 1048 if (nmp->nm_iflag & NFSMNT_AUTHERR) { 1049 nmp->nm_iflag &= ~NFSMNT_AUTHERR; 1050 error = EAUTH; 1051 } 1052 if (error) 1053 free((caddr_t)*auth_str, M_TEMP); 1054 else { 1055 *auth_len = nmp->nm_authlen; 1056 *verf_len = nmp->nm_verflen; 1057 memcpy((caddr_t)key, (caddr_t)nmp->nm_key, sizeof (key)); 1058 } 1059 nmp->nm_iflag &= ~NFSMNT_HASAUTH; 1060 nmp->nm_iflag |= NFSMNT_WAITAUTH; 1061 if (nmp->nm_iflag & NFSMNT_WANTAUTH) { 1062 nmp->nm_iflag &= ~NFSMNT_WANTAUTH; 1063 wakeup((caddr_t)&nmp->nm_authtype); 1064 } 1065 return (error); 1066 } 1067 1068 /* 1069 * Get a nickname authenticator and verifier. 1070 */ 1071 int 1072 nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 1073 struct nfsmount *nmp; 1074 struct ucred *cred; 1075 char **auth_str; 1076 int *auth_len; 1077 char *verf_str; 1078 int verf_len; 1079 { 1080 register struct nfsuid *nuidp; 1081 register u_int32_t *nickp, *verfp; 1082 struct timeval ktvin, ktvout; 1083 1084 #ifdef DIAGNOSTIC 1085 if (verf_len < (4 * NFSX_UNSIGNED)) 1086 panic("nfs_getnickauth verf too small"); 1087 #endif 1088 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1089 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1090 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1091 break; 1092 } 1093 if (!nuidp || nuidp->nu_expire < time.tv_sec) 1094 return (EACCES); 1095 1096 /* 1097 * Move to the end of the lru list (end of lru == most recently used). 1098 */ 1099 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1100 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1101 1102 nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1103 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1104 *nickp = txdr_unsigned(nuidp->nu_nickname); 1105 *auth_str = (char *)nickp; 1106 *auth_len = 2 * NFSX_UNSIGNED; 1107 1108 /* 1109 * Now we must encrypt the verifier and package it up. 1110 */ 1111 verfp = (u_int32_t *)verf_str; 1112 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1113 if (time.tv_sec > nuidp->nu_timestamp.tv_sec || 1114 (time.tv_sec == nuidp->nu_timestamp.tv_sec && 1115 time.tv_usec > nuidp->nu_timestamp.tv_usec)) 1116 nuidp->nu_timestamp = time; 1117 else 1118 nuidp->nu_timestamp.tv_usec++; 1119 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1120 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1121 1122 /* 1123 * Now encrypt the timestamp verifier in ecb mode using the session 1124 * key. 1125 */ 1126 #ifdef NFSKERB 1127 XXX 1128 #endif 1129 1130 *verfp++ = ktvout.tv_sec; 1131 *verfp++ = ktvout.tv_usec; 1132 *verfp = 0; 1133 return (0); 1134 } 1135 1136 /* 1137 * Save the current nickname in a hash list entry on the mount point. 1138 */ 1139 int 1140 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 1141 register struct nfsmount *nmp; 1142 struct ucred *cred; 1143 int len; 1144 NFSKERBKEY_T key; 1145 struct mbuf **mdp; 1146 char **dposp; 1147 struct mbuf *mrep; 1148 { 1149 register struct nfsuid *nuidp; 1150 register u_int32_t *tl; 1151 register int32_t t1; 1152 struct mbuf *md = *mdp; 1153 struct timeval ktvin, ktvout; 1154 u_int32_t nick; 1155 char *dpos = *dposp, *cp2; 1156 int deltasec, error = 0; 1157 1158 if (len == (3 * NFSX_UNSIGNED)) { 1159 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1160 ktvin.tv_sec = *tl++; 1161 ktvin.tv_usec = *tl++; 1162 nick = fxdr_unsigned(u_int32_t, *tl); 1163 1164 /* 1165 * Decrypt the timestamp in ecb mode. 1166 */ 1167 #ifdef NFSKERB 1168 XXX 1169 #endif 1170 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1171 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1172 deltasec = time.tv_sec - ktvout.tv_sec; 1173 if (deltasec < 0) 1174 deltasec = -deltasec; 1175 /* 1176 * If ok, add it to the hash list for the mount point. 1177 */ 1178 if (deltasec <= NFS_KERBCLOCKSKEW) { 1179 if (nmp->nm_numuids < nuidhash_max) { 1180 nmp->nm_numuids++; 1181 nuidp = (struct nfsuid *) 1182 malloc(sizeof (struct nfsuid), M_NFSUID, 1183 M_WAITOK); 1184 } else { 1185 nuidp = nmp->nm_uidlruhead.tqh_first; 1186 LIST_REMOVE(nuidp, nu_hash); 1187 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1188 nu_lru); 1189 } 1190 nuidp->nu_flag = 0; 1191 nuidp->nu_cr.cr_uid = cred->cr_uid; 1192 nuidp->nu_expire = time.tv_sec + NFS_KERBTTL; 1193 nuidp->nu_timestamp = ktvout; 1194 nuidp->nu_nickname = nick; 1195 memcpy(nuidp->nu_key, key, sizeof (key)); 1196 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1197 nu_lru); 1198 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1199 nuidp, nu_hash); 1200 } 1201 } else 1202 nfsm_adv(nfsm_rndup(len)); 1203 nfsmout: 1204 *mdp = md; 1205 *dposp = dpos; 1206 return (error); 1207 } 1208 #endif /* NFS */ 1209
Indexes created Tue Oct 21 08:09:48 GMT 2025