nfs_nfsiod.c revision 59794
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: head/sys/nfsclient/nfs_nfsiod.c 59794 2000-04-30 18:52:11Z phk $ 38 */ 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/sysproto.h> 43#include <sys/kernel.h> 44#include <sys/sysctl.h> 45#include <sys/file.h> 46#include <sys/filedesc.h> 47#include <sys/vnode.h> 48#include <sys/malloc.h> 49#include <sys/mount.h> 50#include <sys/proc.h> 51#include <sys/buf.h> 52#include <sys/mbuf.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55#include <sys/domain.h> 56#include <sys/protosw.h> 57#include <sys/namei.h> 58 59#include <netinet/in.h> 60#include <netinet/tcp.h> 61#include <nfs/xdr_subs.h> 62#include <nfs/rpcv2.h> 63#include <nfs/nfsproto.h> 64#include <nfs/nfs.h> 65#include <nfs/nfsm_subs.h> 66#include <nfs/nfsrvcache.h> 67#include <nfs/nfsmount.h> 68#include <nfs/nfsnode.h> 69#include <nfs/nqnfs.h> 70#include <nfs/nfsrtt.h> 71 72static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 73 74/* Global defs. */ 75extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *nd, 76 struct nfssvc_sock *slp, 77 struct proc *procp, 78 struct mbuf **mreqp)); 79extern int nfs_numasync; 80extern time_t nqnfsstarttime; 81extern int nqsrv_writeslack; 82extern int nfsrtton; 83extern struct nfsstats nfsstats; 84extern int nfsrvw_procrastinate; 85extern int nfsrvw_procrastinate_v3; 86static int nuidhash_max = NFS_MAXUIDHASH; 87 88#ifndef NFS_NOSERVER 89static void nfsrv_zapsock __P((struct nfssvc_sock *slp)); 90#endif 91static int nfssvc_iod __P((struct proc *)); 92 93#define TRUE 1 94#define FALSE 0 95 96static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 97 98SYSCTL_DECL(_vfs_nfs); 99 100#ifndef NFS_NOSERVER 101int nfsd_waiting = 0; 102static struct nfsdrt nfsdrt; 103static int nfs_numnfsd = 0; 104static int notstarted = 1; 105static int modify_flag = 0; 106static void nfsd_rt __P((int sotype, struct nfsrv_descript *nd, 107 int cacherep)); 108static int nfssvc_addsock __P((struct file *, struct sockaddr *, 109 struct proc *)); 110static int nfssvc_nfsd __P((struct nfsd_srvargs *,caddr_t,struct proc *)); 111 112static int nfs_privport = 0; 113SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, ""); 114SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, ""); 115SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, ""); 116 117/* 118 * NFS server system calls 119 */ 120 121#endif /* NFS_NOSERVER */ 122/* 123 * Nfs server psuedo system call for the nfsd's 124 * Based on the flag value it either: 125 * - adds a socket to the selection list 126 * - remains in the kernel as an nfsd 127 * - remains in the kernel as an nfsiod 128 */ 129#ifndef _SYS_SYSPROTO_H_ 130struct nfssvc_args { 131 int flag; 132 caddr_t argp; 133}; 134#endif 135int 136nfssvc(p, uap) 137 struct proc *p; 138 register struct nfssvc_args *uap; 139{ 140#ifndef NFS_NOSERVER 141 struct nameidata nd; 142 struct file *fp; 143 struct sockaddr *nam; 144 struct nfsd_args nfsdarg; 145 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 146 struct nfsd_cargs ncd; 147 struct nfsd *nfsd; 148 struct nfssvc_sock *slp; 149 struct nfsuid *nuidp; 150 struct nfsmount *nmp; 151#endif /* NFS_NOSERVER */ 152 int error; 153 154 /* 155 * Must be super user 156 */ 157 error = suser(p); 158 if(error) 159 return (error); 160 while (nfssvc_sockhead_flag & SLP_INIT) { 161 nfssvc_sockhead_flag |= SLP_WANTINIT; 162 (void) tsleep((caddr_t)&nfssvc_sockhead, PSOCK, "nfsd init", 0); 163 } 164 if (uap->flag & NFSSVC_BIOD) 165 error = nfssvc_iod(p); 166#ifdef NFS_NOSERVER 167 else 168 error = ENXIO; 169#else /* !NFS_NOSERVER */ 170 else if (uap->flag & NFSSVC_MNTD) { 171 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 172 if (error) 173 return (error); 174 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 175 ncd.ncd_dirp, p); 176 error = namei(&nd); 177 if (error) 178 return (error); 179 NDFREE(&nd, NDF_ONLY_PNBUF); 180 if ((nd.ni_vp->v_flag & VROOT) == 0) 181 error = EINVAL; 182 nmp = VFSTONFS(nd.ni_vp->v_mount); 183 vput(nd.ni_vp); 184 if (error) 185 return (error); 186 if ((nmp->nm_state & NFSSTA_MNTD) && 187 (uap->flag & NFSSVC_GOTAUTH) == 0) 188 return (0); 189 nmp->nm_state |= NFSSTA_MNTD; 190 error = nqnfs_clientd(nmp, p->p_ucred, &ncd, uap->flag, 191 uap->argp, p); 192 } else if (uap->flag & NFSSVC_ADDSOCK) { 193 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 194 if (error) 195 return (error); 196 error = getsock(p->p_fd, nfsdarg.sock, &fp); 197 if (error) 198 return (error); 199 /* 200 * Get the client address for connected sockets. 201 */ 202 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 203 nam = (struct sockaddr *)0; 204 else { 205 error = getsockaddr(&nam, nfsdarg.name, 206 nfsdarg.namelen); 207 if (error) 208 return (error); 209 } 210 error = nfssvc_addsock(fp, nam, p); 211 } else { 212 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 213 if (error) 214 return (error); 215 if ((uap->flag & NFSSVC_AUTHIN) && 216 ((nfsd = nsd->nsd_nfsd)) != NULL && 217 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 218 slp = nfsd->nfsd_slp; 219 220 /* 221 * First check to see if another nfsd has already 222 * added this credential. 223 */ 224 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 225 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 226 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 227 (!nfsd->nfsd_nd->nd_nam2 || 228 netaddr_match(NU_NETFAM(nuidp), 229 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 230 break; 231 } 232 if (nuidp) { 233 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 234 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 235 } else { 236 /* 237 * Nope, so we will. 238 */ 239 if (slp->ns_numuids < nuidhash_max) { 240 slp->ns_numuids++; 241 nuidp = (struct nfsuid *) 242 malloc(sizeof (struct nfsuid), M_NFSUID, 243 M_WAITOK); 244 } else 245 nuidp = (struct nfsuid *)0; 246 if ((slp->ns_flag & SLP_VALID) == 0) { 247 if (nuidp) 248 free((caddr_t)nuidp, M_NFSUID); 249 } else { 250 if (nuidp == (struct nfsuid *)0) { 251 nuidp = slp->ns_uidlruhead.tqh_first; 252 LIST_REMOVE(nuidp, nu_hash); 253 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 254 nu_lru); 255 if (nuidp->nu_flag & NU_NAM) 256 FREE(nuidp->nu_nam, M_SONAME); 257 } 258 nuidp->nu_flag = 0; 259 nuidp->nu_cr = nsd->nsd_cr; 260 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 261 nuidp->nu_cr.cr_ngroups = NGROUPS; 262 nuidp->nu_cr.cr_ref = 1; 263 nuidp->nu_timestamp = nsd->nsd_timestamp; 264 nuidp->nu_expire = time_second + nsd->nsd_ttl; 265 /* 266 * and save the session key in nu_key. 267 */ 268 bcopy(nsd->nsd_key, nuidp->nu_key, 269 sizeof (nsd->nsd_key)); 270 if (nfsd->nfsd_nd->nd_nam2) { 271 struct sockaddr_in *saddr; 272 273 saddr = (struct sockaddr_in *) 274 nfsd->nfsd_nd->nd_nam2; 275 switch (saddr->sin_family) { 276 case AF_INET: 277 nuidp->nu_flag |= NU_INETADDR; 278 nuidp->nu_inetaddr = 279 saddr->sin_addr.s_addr; 280 break; 281 case AF_ISO: 282 default: 283 nuidp->nu_flag |= NU_NAM; 284 nuidp->nu_nam = 285 dup_sockaddr(nfsd->nfsd_nd-> 286 nd_nam2, 1); 287 break; 288 }; 289 } 290 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 291 nu_lru); 292 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 293 nuidp, nu_hash); 294 nfsrv_setcred(&nuidp->nu_cr, 295 &nfsd->nfsd_nd->nd_cr); 296 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 297 } 298 } 299 } 300 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 301 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 302 error = nfssvc_nfsd(nsd, uap->argp, p); 303 } 304#endif /* NFS_NOSERVER */ 305 if (error == EINTR || error == ERESTART) 306 error = 0; 307 return (error); 308} 309 310#ifndef NFS_NOSERVER 311/* 312 * Adds a socket to the list for servicing by nfsds. 313 */ 314static int 315nfssvc_addsock(fp, mynam, p) 316 struct file *fp; 317 struct sockaddr *mynam; 318 struct proc *p; 319{ 320 register int siz; 321 register struct nfssvc_sock *slp; 322 register struct socket *so; 323 int error, s; 324 325 so = (struct socket *)fp->f_data; 326#if 0 327 tslp = (struct nfssvc_sock *)0; 328 /* 329 * Add it to the list, as required. 330 */ 331 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 332 tslp = nfs_udpsock; 333 if (tslp->ns_flag & SLP_VALID) { 334 if (mynam != NULL) 335 FREE(mynam, M_SONAME); 336 return (EPERM); 337 } 338 } 339#endif 340 if (so->so_type == SOCK_STREAM) 341 siz = NFS_MAXPACKET + sizeof (u_long); 342 else 343 siz = NFS_MAXPACKET; 344 error = soreserve(so, siz, siz); 345 if (error) { 346 if (mynam != NULL) 347 FREE(mynam, M_SONAME); 348 return (error); 349 } 350 351 /* 352 * Set protocol specific options { for now TCP only } and 353 * reserve some space. For datagram sockets, this can get called 354 * repeatedly for the same socket, but that isn't harmful. 355 */ 356 if (so->so_type == SOCK_STREAM) { 357 struct sockopt sopt; 358 int val; 359 360 bzero(&sopt, sizeof sopt); 361 sopt.sopt_level = SOL_SOCKET; 362 sopt.sopt_name = SO_KEEPALIVE; 363 sopt.sopt_val = &val; 364 sopt.sopt_valsize = sizeof val; 365 val = 1; 366 sosetopt(so, &sopt); 367 } 368 if (so->so_proto->pr_domain->dom_family == AF_INET && 369 so->so_proto->pr_protocol == IPPROTO_TCP) { 370 struct sockopt sopt; 371 int val; 372 373 bzero(&sopt, sizeof sopt); 374 sopt.sopt_level = IPPROTO_TCP; 375 sopt.sopt_name = TCP_NODELAY; 376 sopt.sopt_val = &val; 377 sopt.sopt_valsize = sizeof val; 378 val = 1; 379 sosetopt(so, &sopt); 380 } 381 so->so_rcv.sb_flags &= ~SB_NOINTR; 382 so->so_rcv.sb_timeo = 0; 383 so->so_snd.sb_flags &= ~SB_NOINTR; 384 so->so_snd.sb_timeo = 0; 385 386 slp = (struct nfssvc_sock *) 387 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 388 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 389 STAILQ_INIT(&slp->ns_rec); 390 TAILQ_INIT(&slp->ns_uidlruhead); 391 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 392 393 slp->ns_so = so; 394 slp->ns_nam = mynam; 395 fp->f_count++; 396 slp->ns_fp = fp; 397 s = splnet(); 398 so->so_upcallarg = (caddr_t)slp; 399 so->so_upcall = nfsrv_rcv; 400 so->so_rcv.sb_flags |= SB_UPCALL; 401 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 402 nfsrv_wakenfsd(slp); 403 splx(s); 404 return (0); 405} 406 407/* 408 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 409 * until it is killed by a signal. 410 */ 411static int 412nfssvc_nfsd(nsd, argp, p) 413 struct nfsd_srvargs *nsd; 414 caddr_t argp; 415 struct proc *p; 416{ 417 register int siz; 418 register struct nfssvc_sock *slp; 419 struct nfsd *nfsd = nsd->nsd_nfsd; 420 struct nfsrv_descript *nd = NULL; 421 struct mbuf *m, *mreq; 422 int error = 0, cacherep, s, sotype, writes_todo; 423 int procrastinate; 424 u_quad_t cur_usec; 425 426#ifndef nolint 427 cacherep = RC_DOIT; 428 writes_todo = 0; 429#endif 430 if (nfsd == (struct nfsd *)0) { 431 nsd->nsd_nfsd = nfsd = (struct nfsd *) 432 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 433 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 434 s = splnet(); 435 nfsd->nfsd_procp = p; 436 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 437 nfs_numnfsd++; 438 } else 439 s = splnet(); 440 441 /* 442 * Loop getting rpc requests until SIGKILL. 443 */ 444 for (;;) { 445 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 446 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 447 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 448 nfsd->nfsd_flag |= NFSD_WAITING; 449 nfsd_waiting++; 450 error = tsleep((caddr_t)nfsd, PSOCK | PCATCH, 451 "nfsd", 0); 452 nfsd_waiting--; 453 if (error) 454 goto done; 455 } 456 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 457 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 458 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 459 slp = slp->ns_chain.tqe_next) { 460 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 461 == (SLP_VALID | SLP_DOREC)) { 462 slp->ns_flag &= ~SLP_DOREC; 463 slp->ns_sref++; 464 nfsd->nfsd_slp = slp; 465 break; 466 } 467 } 468 if (slp == 0) 469 nfsd_head_flag &= ~NFSD_CHECKSLP; 470 } 471 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 472 continue; 473 if (slp->ns_flag & SLP_VALID) { 474 if (slp->ns_flag & SLP_DISCONN) 475 nfsrv_zapsock(slp); 476 else if (slp->ns_flag & SLP_NEEDQ) { 477 slp->ns_flag &= ~SLP_NEEDQ; 478 (void) nfs_slplock(slp, 1); 479 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 480 M_WAIT); 481 nfs_slpunlock(slp); 482 } 483 error = nfsrv_dorec(slp, nfsd, &nd); 484 cur_usec = nfs_curusec(); 485 if (error && slp->ns_tq.lh_first && 486 slp->ns_tq.lh_first->nd_time <= cur_usec) { 487 error = 0; 488 cacherep = RC_DOIT; 489 writes_todo = 1; 490 } else 491 writes_todo = 0; 492 nfsd->nfsd_flag |= NFSD_REQINPROG; 493 } 494 } else { 495 error = 0; 496 slp = nfsd->nfsd_slp; 497 } 498 if (error || (slp->ns_flag & SLP_VALID) == 0) { 499 if (nd) { 500 free((caddr_t)nd, M_NFSRVDESC); 501 nd = NULL; 502 } 503 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 504 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 505 nfsrv_slpderef(slp); 506 continue; 507 } 508 splx(s); 509 sotype = slp->ns_so->so_type; 510 if (nd) { 511 getmicrotime(&nd->nd_starttime); 512 if (nd->nd_nam2) 513 nd->nd_nam = nd->nd_nam2; 514 else 515 nd->nd_nam = slp->ns_nam; 516 517 /* 518 * Check to see if authorization is needed. 519 */ 520 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 521 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 522 nsd->nsd_haddr = 523 ((struct sockaddr_in *) 524 nd->nd_nam)->sin_addr.s_addr; 525 nsd->nsd_authlen = nfsd->nfsd_authlen; 526 nsd->nsd_verflen = nfsd->nfsd_verflen; 527 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 528 nfsd->nfsd_authlen) && 529 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 530 nfsd->nfsd_verflen) && 531 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 532 return (ENEEDAUTH); 533 cacherep = RC_DROPIT; 534 } else 535 cacherep = nfsrv_getcache(nd, slp, &mreq); 536 537 /* 538 * Check for just starting up for NQNFS and send 539 * fake "try again later" replies to the NQNFS clients. 540 */ 541 if (notstarted && nqnfsstarttime <= time_second) { 542 if (modify_flag) { 543 nqnfsstarttime = time_second + nqsrv_writeslack; 544 modify_flag = 0; 545 } else 546 notstarted = 0; 547 } 548 if (notstarted) { 549 if ((nd->nd_flag & ND_NQNFS) == 0) 550 cacherep = RC_DROPIT; 551 else if (nd->nd_procnum != NFSPROC_WRITE) { 552 nd->nd_procnum = NFSPROC_NOOP; 553 nd->nd_repstat = NQNFS_TRYLATER; 554 cacherep = RC_DOIT; 555 } else 556 modify_flag = 1; 557 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 558 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 559 nd->nd_procnum = NFSPROC_NOOP; 560 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 561 cacherep = RC_DOIT; 562 } else if (nfs_privport) { 563 /* Check if source port is privileged */ 564 u_short port; 565 struct sockaddr *nam = nd->nd_nam; 566 struct sockaddr_in *sin; 567 568 sin = (struct sockaddr_in *)nam; 569 port = ntohs(sin->sin_port); 570 if (port >= IPPORT_RESERVED && 571 nd->nd_procnum != NFSPROC_NULL) { 572 nd->nd_procnum = NFSPROC_NOOP; 573 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 574 cacherep = RC_DOIT; 575 printf("NFS request from unprivileged port (%s:%d)\n", 576 inet_ntoa(sin->sin_addr), port); 577 } 578 } 579 580 } 581 582 /* 583 * Loop to get all the write rpc relies that have been 584 * gathered together. 585 */ 586 do { 587 switch (cacherep) { 588 case RC_DOIT: 589 if (nd && (nd->nd_flag & ND_NFSV3)) 590 procrastinate = nfsrvw_procrastinate_v3; 591 else 592 procrastinate = nfsrvw_procrastinate; 593 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 594 procrastinate > 0 && !notstarted)) 595 error = nfsrv_writegather(&nd, slp, 596 nfsd->nfsd_procp, &mreq); 597 else 598 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 599 slp, nfsd->nfsd_procp, &mreq); 600 if (mreq == NULL) 601 break; 602 if (error != 0 && error != NFSERR_RETVOID) { 603 if (nd->nd_procnum != NQNFSPROC_VACATED) 604 nfsstats.srv_errs++; 605 nfsrv_updatecache(nd, FALSE, mreq); 606 if (nd->nd_nam2) 607 FREE(nd->nd_nam2, M_SONAME); 608 break; 609 } 610 nfsstats.srvrpccnt[nd->nd_procnum]++; 611 nfsrv_updatecache(nd, TRUE, mreq); 612 nd->nd_mrep = (struct mbuf *)0; 613 case RC_REPLY: 614 m = mreq; 615 siz = 0; 616 while (m) { 617 siz += m->m_len; 618 m = m->m_next; 619 } 620 if (siz <= 0 || siz > NFS_MAXPACKET) { 621 printf("mbuf siz=%d\n",siz); 622 panic("Bad nfs svc reply"); 623 } 624 m = mreq; 625 m->m_pkthdr.len = siz; 626 m->m_pkthdr.rcvif = (struct ifnet *)0; 627 /* 628 * For stream protocols, prepend a Sun RPC 629 * Record Mark. 630 */ 631 if (sotype == SOCK_STREAM) { 632 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 633 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 634 } 635 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 636 (void) nfs_slplock(slp, 1); 637 if (slp->ns_flag & SLP_VALID) 638 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 639 else { 640 error = EPIPE; 641 m_freem(m); 642 } 643 if (nfsrtton) 644 nfsd_rt(sotype, nd, cacherep); 645 if (nd->nd_nam2) 646 FREE(nd->nd_nam2, M_SONAME); 647 if (nd->nd_mrep) 648 m_freem(nd->nd_mrep); 649 if (error == EPIPE) 650 nfsrv_zapsock(slp); 651 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 652 nfs_slpunlock(slp); 653 if (error == EINTR || error == ERESTART) { 654 free((caddr_t)nd, M_NFSRVDESC); 655 nfsrv_slpderef(slp); 656 s = splnet(); 657 goto done; 658 } 659 break; 660 case RC_DROPIT: 661 if (nfsrtton) 662 nfsd_rt(sotype, nd, cacherep); 663 m_freem(nd->nd_mrep); 664 if (nd->nd_nam2) 665 FREE(nd->nd_nam2, M_SONAME); 666 break; 667 }; 668 if (nd) { 669 FREE((caddr_t)nd, M_NFSRVDESC); 670 nd = NULL; 671 } 672 673 /* 674 * Check to see if there are outstanding writes that 675 * need to be serviced. 676 */ 677 cur_usec = nfs_curusec(); 678 s = splsoftclock(); 679 if (slp->ns_tq.lh_first && 680 slp->ns_tq.lh_first->nd_time <= cur_usec) { 681 cacherep = RC_DOIT; 682 writes_todo = 1; 683 } else 684 writes_todo = 0; 685 splx(s); 686 } while (writes_todo); 687 s = splnet(); 688 if (nfsrv_dorec(slp, nfsd, &nd)) { 689 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 690 nfsd->nfsd_slp = NULL; 691 nfsrv_slpderef(slp); 692 } 693 } 694done: 695 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 696 splx(s); 697 free((caddr_t)nfsd, M_NFSD); 698 nsd->nsd_nfsd = (struct nfsd *)0; 699 if (--nfs_numnfsd == 0) 700 nfsrv_init(TRUE); /* Reinitialize everything */ 701 return (error); 702} 703 704/* 705 * Shut down a socket associated with an nfssvc_sock structure. 706 * Should be called with the send lock set, if required. 707 * The trick here is to increment the sref at the start, so that the nfsds 708 * will stop using it and clear ns_flag at the end so that it will not be 709 * reassigned during cleanup. 710 */ 711static void 712nfsrv_zapsock(slp) 713 register struct nfssvc_sock *slp; 714{ 715 register struct nfsuid *nuidp, *nnuidp; 716 register struct nfsrv_descript *nwp, *nnwp; 717 struct socket *so; 718 struct file *fp; 719 struct nfsrv_rec *rec; 720 int s; 721 722 slp->ns_flag &= ~SLP_ALLFLAGS; 723 fp = slp->ns_fp; 724 if (fp) { 725 slp->ns_fp = (struct file *)0; 726 so = slp->ns_so; 727 so->so_rcv.sb_flags &= ~SB_UPCALL; 728 so->so_upcall = NULL; 729 so->so_upcallarg = NULL; 730 soshutdown(so, 2); 731 closef(fp, (struct proc *)0); 732 if (slp->ns_nam) 733 FREE(slp->ns_nam, M_SONAME); 734 m_freem(slp->ns_raw); 735 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 736 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 737 if (rec->nr_address) 738 FREE(rec->nr_address, M_SONAME); 739 m_freem(rec->nr_packet); 740 free(rec, M_NFSRVDESC); 741 } 742 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 743 nuidp = nnuidp) { 744 nnuidp = nuidp->nu_lru.tqe_next; 745 LIST_REMOVE(nuidp, nu_hash); 746 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 747 if (nuidp->nu_flag & NU_NAM) 748 FREE(nuidp->nu_nam, M_SONAME); 749 free((caddr_t)nuidp, M_NFSUID); 750 } 751 s = splsoftclock(); 752 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 753 nnwp = nwp->nd_tq.le_next; 754 LIST_REMOVE(nwp, nd_tq); 755 free((caddr_t)nwp, M_NFSRVDESC); 756 } 757 LIST_INIT(&slp->ns_tq); 758 splx(s); 759 } 760} 761 762/* 763 * Derefence a server socket structure. If it has no more references and 764 * is no longer valid, you can throw it away. 765 */ 766void 767nfsrv_slpderef(slp) 768 register struct nfssvc_sock *slp; 769{ 770 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 771 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 772 free((caddr_t)slp, M_NFSSVC); 773 } 774} 775 776/* 777 * Lock a socket against others. 778 */ 779int 780nfs_slplock(slp, wait) 781 register struct nfssvc_sock *slp; 782 int wait; 783{ 784 int *statep = &slp->ns_solock; 785 786 if (!wait && (*statep & NFSSTA_SNDLOCK)) 787 return(0); /* already locked, fail */ 788 while (*statep & NFSSTA_SNDLOCK) { 789 *statep |= NFSSTA_WANTSND; 790 (void) tsleep((caddr_t)statep, PZERO - 1, "nfsslplck", 0); 791 } 792 *statep |= NFSSTA_SNDLOCK; 793 return (1); 794} 795 796/* 797 * Unlock the stream socket for others. 798 */ 799void 800nfs_slpunlock(slp) 801 register struct nfssvc_sock *slp; 802{ 803 int *statep = &slp->ns_solock; 804 805 if ((*statep & NFSSTA_SNDLOCK) == 0) 806 panic("nfs slpunlock"); 807 *statep &= ~NFSSTA_SNDLOCK; 808 if (*statep & NFSSTA_WANTSND) { 809 *statep &= ~NFSSTA_WANTSND; 810 wakeup((caddr_t)statep); 811 } 812} 813 814/* 815 * Initialize the data structures for the server. 816 * Handshake with any new nfsds starting up to avoid any chance of 817 * corruption. 818 */ 819void 820nfsrv_init(terminating) 821 int terminating; 822{ 823 register struct nfssvc_sock *slp, *nslp; 824 825 if (nfssvc_sockhead_flag & SLP_INIT) 826 panic("nfsd init"); 827 nfssvc_sockhead_flag |= SLP_INIT; 828 if (terminating) { 829 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 830 nslp = slp->ns_chain.tqe_next; 831 if (slp->ns_flag & SLP_VALID) 832 nfsrv_zapsock(slp); 833 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 834 free((caddr_t)slp, M_NFSSVC); 835 } 836 nfsrv_cleancache(); /* And clear out server cache */ 837 } else 838 nfs_pub.np_valid = 0; 839 840 TAILQ_INIT(&nfssvc_sockhead); 841 nfssvc_sockhead_flag &= ~SLP_INIT; 842 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 843 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 844 wakeup((caddr_t)&nfssvc_sockhead); 845 } 846 847 TAILQ_INIT(&nfsd_head); 848 nfsd_head_flag &= ~NFSD_CHECKSLP; 849 850#if 0 851 nfs_udpsock = (struct nfssvc_sock *) 852 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 853 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 854 STAILQ_INIT(&nfs_udpsock->ns_rec); 855 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 856 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 857 858 nfs_cltpsock = (struct nfssvc_sock *) 859 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 860 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 861 STAILQ_INIT(&nfs_cltpsock->ns_rec); 862 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 863 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 864#endif 865} 866 867/* 868 * Add entries to the server monitor log. 869 */ 870static void 871nfsd_rt(sotype, nd, cacherep) 872 int sotype; 873 register struct nfsrv_descript *nd; 874 int cacherep; 875{ 876 register struct drt *rt; 877 878 rt = &nfsdrt.drt[nfsdrt.pos]; 879 if (cacherep == RC_DOIT) 880 rt->flag = 0; 881 else if (cacherep == RC_REPLY) 882 rt->flag = DRT_CACHEREPLY; 883 else 884 rt->flag = DRT_CACHEDROP; 885 if (sotype == SOCK_STREAM) 886 rt->flag |= DRT_TCP; 887 if (nd->nd_flag & ND_NQNFS) 888 rt->flag |= DRT_NQNFS; 889 else if (nd->nd_flag & ND_NFSV3) 890 rt->flag |= DRT_NFSV3; 891 rt->proc = nd->nd_procnum; 892 if (nd->nd_nam->sa_family == AF_INET) 893 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 894 else 895 rt->ipadr = INADDR_ANY; 896 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 897 getmicrotime(&rt->tstamp); 898 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 899} 900#endif /* NFS_NOSERVER */ 901 902static int nfs_defect = 0; 903SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 904 905/* 906 * Asynchronous I/O daemons for client nfs. 907 * They do read-ahead and write-behind operations on the block I/O cache. 908 * Never returns unless it fails or gets killed. 909 */ 910static int 911nfssvc_iod(p) 912 struct proc *p; 913{ 914 register struct buf *bp; 915 register int i, myiod; 916 struct nfsmount *nmp; 917 int error = 0; 918 919 /* 920 * Assign my position or return error if too many already running 921 */ 922 myiod = -1; 923 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 924 if (nfs_asyncdaemon[i] == 0) { 925 nfs_asyncdaemon[i]++; 926 myiod = i; 927 break; 928 } 929 if (myiod == -1) 930 return (EBUSY); 931 nfs_numasync++; 932 /* 933 * Just loop around doin our stuff until SIGKILL 934 */ 935 for (;;) { 936 while (((nmp = nfs_iodmount[myiod]) == NULL 937 || nmp->nm_bufq.tqh_first == NULL) 938 && error == 0) { 939 if (nmp) 940 nmp->nm_bufqiods--; 941 nfs_iodwant[myiod] = p; 942 nfs_iodmount[myiod] = NULL; 943 error = tsleep((caddr_t)&nfs_iodwant[myiod], 944 PWAIT | PCATCH, "nfsidl", 0); 945 } 946 if (error) { 947 nfs_asyncdaemon[myiod] = 0; 948 if (nmp) 949 nmp->nm_bufqiods--; 950 nfs_iodwant[myiod] = NULL; 951 nfs_iodmount[myiod] = NULL; 952 nfs_numasync--; 953 return (error); 954 } 955 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 956 /* Take one off the front of the list */ 957 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 958 nmp->nm_bufqlen--; 959 if (nmp->nm_bufqwant && nmp->nm_bufqlen <= nfs_numasync) { 960 nmp->nm_bufqwant = FALSE; 961 wakeup(&nmp->nm_bufq); 962 } 963 if (bp->b_iocmd == BIO_READ) 964 (void) nfs_doio(bp, bp->b_rcred, (struct proc *)0); 965 else 966 (void) nfs_doio(bp, bp->b_wcred, (struct proc *)0); 967 /* 968 * If there are more than one iod on this mount, then defect 969 * so that the iods can be shared out fairly between the mounts 970 */ 971 if (nfs_defect && nmp->nm_bufqiods > 1) { 972 NFS_DPF(ASYNCIO, 973 ("nfssvc_iod: iod %d defecting from mount %p\n", 974 myiod, nmp)); 975 nfs_iodmount[myiod] = NULL; 976 nmp->nm_bufqiods--; 977 break; 978 } 979 } 980 } 981} 982 983 984/* 985 * Get an authorization string for the uid by having the mount_nfs sitting 986 * on this mount point porpous out of the kernel and do it. 987 */ 988int 989nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 990 register struct nfsmount *nmp; 991 struct nfsreq *rep; 992 struct ucred *cred; 993 char **auth_str; 994 int *auth_len; 995 char *verf_str; 996 int *verf_len; 997 NFSKERBKEY_T key; /* return session key */ 998{ 999 int error = 0; 1000 1001 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 1002 nmp->nm_state |= NFSSTA_WANTAUTH; 1003 (void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK, 1004 "nfsauth1", 2 * hz); 1005 error = nfs_sigintr(nmp, rep, rep->r_procp); 1006 if (error) { 1007 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1008 return (error); 1009 } 1010 } 1011 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 1012 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 1013 nmp->nm_authlen = RPCAUTH_MAXSIZ; 1014 nmp->nm_verfstr = verf_str; 1015 nmp->nm_verflen = *verf_len; 1016 nmp->nm_authuid = cred->cr_uid; 1017 wakeup((caddr_t)&nmp->nm_authstr); 1018 1019 /* 1020 * And wait for mount_nfs to do its stuff. 1021 */ 1022 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1023 (void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK, 1024 "nfsauth2", 2 * hz); 1025 error = nfs_sigintr(nmp, rep, rep->r_procp); 1026 } 1027 if (nmp->nm_state & NFSSTA_AUTHERR) { 1028 nmp->nm_state &= ~NFSSTA_AUTHERR; 1029 error = EAUTH; 1030 } 1031 if (error) 1032 free((caddr_t)*auth_str, M_TEMP); 1033 else { 1034 *auth_len = nmp->nm_authlen; 1035 *verf_len = nmp->nm_verflen; 1036 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1037 } 1038 nmp->nm_state &= ~NFSSTA_HASAUTH; 1039 nmp->nm_state |= NFSSTA_WAITAUTH; 1040 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1041 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1042 wakeup((caddr_t)&nmp->nm_authtype); 1043 } 1044 return (error); 1045} 1046 1047/* 1048 * Get a nickname authenticator and verifier. 1049 */ 1050int 1051nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 1052 struct nfsmount *nmp; 1053 struct ucred *cred; 1054 char **auth_str; 1055 int *auth_len; 1056 char *verf_str; 1057 int verf_len; 1058{ 1059 register struct nfsuid *nuidp; 1060 register u_int32_t *nickp, *verfp; 1061 struct timeval ktvin, ktvout; 1062 1063#ifdef DIAGNOSTIC 1064 if (verf_len < (4 * NFSX_UNSIGNED)) 1065 panic("nfs_getnickauth verf too small"); 1066#endif 1067 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1068 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1069 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1070 break; 1071 } 1072 if (!nuidp || nuidp->nu_expire < time_second) 1073 return (EACCES); 1074 1075 /* 1076 * Move to the end of the lru list (end of lru == most recently used). 1077 */ 1078 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1079 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1080 1081 nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1082 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1083 *nickp = txdr_unsigned(nuidp->nu_nickname); 1084 *auth_str = (char *)nickp; 1085 *auth_len = 2 * NFSX_UNSIGNED; 1086 1087 /* 1088 * Now we must encrypt the verifier and package it up. 1089 */ 1090 verfp = (u_int32_t *)verf_str; 1091 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1092 if (time_second > nuidp->nu_timestamp.tv_sec || 1093 (time_second == nuidp->nu_timestamp.tv_sec && 1094 time_second > nuidp->nu_timestamp.tv_usec)) 1095 getmicrotime(&nuidp->nu_timestamp); 1096 else 1097 nuidp->nu_timestamp.tv_usec++; 1098 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1099 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1100 1101 /* 1102 * Now encrypt the timestamp verifier in ecb mode using the session 1103 * key. 1104 */ 1105#ifdef NFSKERB 1106 XXX 1107#endif 1108 1109 *verfp++ = ktvout.tv_sec; 1110 *verfp++ = ktvout.tv_usec; 1111 *verfp = 0; 1112 return (0); 1113} 1114 1115/* 1116 * Save the current nickname in a hash list entry on the mount point. 1117 */ 1118int 1119nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 1120 register struct nfsmount *nmp; 1121 struct ucred *cred; 1122 int len; 1123 NFSKERBKEY_T key; 1124 struct mbuf **mdp; 1125 char **dposp; 1126 struct mbuf *mrep; 1127{ 1128 register struct nfsuid *nuidp; 1129 register u_int32_t *tl; 1130 register int32_t t1; 1131 struct mbuf *md = *mdp; 1132 struct timeval ktvin, ktvout; 1133 u_int32_t nick; 1134 char *dpos = *dposp, *cp2; 1135 int deltasec, error = 0; 1136 1137 if (len == (3 * NFSX_UNSIGNED)) { 1138 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1139 ktvin.tv_sec = *tl++; 1140 ktvin.tv_usec = *tl++; 1141 nick = fxdr_unsigned(u_int32_t, *tl); 1142 1143 /* 1144 * Decrypt the timestamp in ecb mode. 1145 */ 1146#ifdef NFSKERB 1147 XXX 1148#endif 1149 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1150 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1151 deltasec = time_second - ktvout.tv_sec; 1152 if (deltasec < 0) 1153 deltasec = -deltasec; 1154 /* 1155 * If ok, add it to the hash list for the mount point. 1156 */ 1157 if (deltasec <= NFS_KERBCLOCKSKEW) { 1158 if (nmp->nm_numuids < nuidhash_max) { 1159 nmp->nm_numuids++; 1160 nuidp = (struct nfsuid *) 1161 malloc(sizeof (struct nfsuid), M_NFSUID, 1162 M_WAITOK); 1163 } else { 1164 nuidp = nmp->nm_uidlruhead.tqh_first; 1165 LIST_REMOVE(nuidp, nu_hash); 1166 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1167 nu_lru); 1168 } 1169 nuidp->nu_flag = 0; 1170 nuidp->nu_cr.cr_uid = cred->cr_uid; 1171 nuidp->nu_expire = time_second + NFS_KERBTTL; 1172 nuidp->nu_timestamp = ktvout; 1173 nuidp->nu_nickname = nick; 1174 bcopy(key, nuidp->nu_key, sizeof (key)); 1175 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1176 nu_lru); 1177 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1178 nuidp, nu_hash); 1179 } 1180 } else 1181 nfsm_adv(nfsm_rndup(len)); 1182nfsmout: 1183 *mdp = md; 1184 *dposp = dpos; 1185 return (error); 1186} 1187