nfs_nfsiod.c revision 31391
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 * $Id: nfs_syscalls.c,v 1.32 1997/11/07 08:53:25 phk Exp $ 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#ifdef ISO 62#include <netiso/iso.h> 63#endif 64#include <nfs/xdr_subs.h> 65#include <nfs/rpcv2.h> 66#include <nfs/nfsproto.h> 67#include <nfs/nfs.h> 68#include <nfs/nfsm_subs.h> 69#include <nfs/nfsrvcache.h> 70#include <nfs/nfsmount.h> 71#include <nfs/nfsnode.h> 72#include <nfs/nqnfs.h> 73#include <nfs/nfsrtt.h> 74 75static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 76 77/* Global defs. */ 78extern int (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *nd, 79 struct nfssvc_sock *slp, 80 struct proc *procp, 81 struct mbuf **mreqp)); 82extern int nfs_numasync; 83extern time_t nqnfsstarttime; 84extern int nqsrv_writeslack; 85extern int nfsrtton; 86extern struct nfsstats nfsstats; 87extern int nfsrvw_procrastinate; 88extern int nfsrvw_procrastinate_v3; 89struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock; 90static int nuidhash_max = NFS_MAXUIDHASH; 91 92static void nfsrv_zapsock __P((struct nfssvc_sock *slp)); 93static int nfssvc_iod __P((struct proc *)); 94 95#define TRUE 1 96#define FALSE 0 97 98static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 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 * getfh() lives here too, but maybe should move to kern/vfs_syscalls.c 120 */ 121 122/* 123 * Get file handle system call 124 */ 125#ifndef _SYS_SYSPROTO_H_ 126struct getfh_args { 127 char *fname; 128 fhandle_t *fhp; 129}; 130#endif 131int 132getfh(p, uap) 133 struct proc *p; 134 register struct getfh_args *uap; 135{ 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, uap->fname, p); 148 error = namei(&nd); 149 if (error) 150 return (error); 151 vp = nd.ni_vp; 152 bzero((caddr_t)&fh, sizeof(fh)); 153 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 154 error = VFS_VPTOFH(vp, &fh.fh_fid); 155 vput(vp); 156 if (error) 157 return (error); 158 error = copyout((caddr_t)&fh, (caddr_t)uap->fhp, sizeof (fh)); 159 return (error); 160} 161 162#endif /* NFS_NOSERVER */ 163/* 164 * Nfs server psuedo 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#ifndef _SYS_SYSPROTO_H_ 171struct nfssvc_args { 172 int flag; 173 caddr_t argp; 174}; 175#endif 176int 177nfssvc(p, uap) 178 struct proc *p; 179 register struct nfssvc_args *uap; 180{ 181#ifndef NFS_NOSERVER 182 struct nameidata nd; 183 struct file *fp; 184 struct sockaddr *nam; 185 struct nfsd_args nfsdarg; 186 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 187 struct nfsd_cargs ncd; 188 struct nfsd *nfsd; 189 struct nfssvc_sock *slp; 190 struct nfsuid *nuidp; 191 struct nfsmount *nmp; 192#endif /* NFS_NOSERVER */ 193 int error; 194 195 /* 196 * Must be super user 197 */ 198 error = suser(p->p_ucred, &p->p_acflag); 199 if(error) 200 return (error); 201 while (nfssvc_sockhead_flag & SLP_INIT) { 202 nfssvc_sockhead_flag |= SLP_WANTINIT; 203 (void) tsleep((caddr_t)&nfssvc_sockhead, PSOCK, "nfsd init", 0); 204 } 205 if (uap->flag & NFSSVC_BIOD) 206 error = nfssvc_iod(p); 207#ifdef NFS_NOSERVER 208 else 209 error = ENXIO; 210#else /* !NFS_NOSERVER */ 211 else if (uap->flag & NFSSVC_MNTD) { 212 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 213 if (error) 214 return (error); 215 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 216 ncd.ncd_dirp, p); 217 error = namei(&nd); 218 if (error) 219 return (error); 220 if ((nd.ni_vp->v_flag & VROOT) == 0) 221 error = EINVAL; 222 nmp = VFSTONFS(nd.ni_vp->v_mount); 223 vput(nd.ni_vp); 224 if (error) 225 return (error); 226 if ((nmp->nm_flag & NFSMNT_MNTD) && 227 (uap->flag & NFSSVC_GOTAUTH) == 0) 228 return (0); 229 nmp->nm_flag |= NFSMNT_MNTD; 230 error = nqnfs_clientd(nmp, p->p_ucred, &ncd, uap->flag, 231 uap->argp, p); 232 } else if (uap->flag & NFSSVC_ADDSOCK) { 233 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 234 if (error) 235 return (error); 236 error = getsock(p->p_fd, nfsdarg.sock, &fp); 237 if (error) 238 return (error); 239 /* 240 * Get the client address for connected sockets. 241 */ 242 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 243 nam = (struct sockaddr *)0; 244 else { 245 error = getsockaddr(&nam, nfsdarg.name, 246 nfsdarg.namelen); 247 if (error) 248 return (error); 249 } 250 error = nfssvc_addsock(fp, nam, p); 251 } else { 252 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 253 if (error) 254 return (error); 255 if ((uap->flag & NFSSVC_AUTHIN) && ((nfsd = nsd->nsd_nfsd)) && 256 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 257 slp = nfsd->nfsd_slp; 258 259 /* 260 * First check to see if another nfsd has already 261 * added this credential. 262 */ 263 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 264 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 265 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 266 (!nfsd->nfsd_nd->nd_nam2 || 267 netaddr_match(NU_NETFAM(nuidp), 268 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 269 break; 270 } 271 if (nuidp) { 272 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 273 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 274 } else { 275 /* 276 * Nope, so we will. 277 */ 278 if (slp->ns_numuids < nuidhash_max) { 279 slp->ns_numuids++; 280 nuidp = (struct nfsuid *) 281 malloc(sizeof (struct nfsuid), M_NFSUID, 282 M_WAITOK); 283 } else 284 nuidp = (struct nfsuid *)0; 285 if ((slp->ns_flag & SLP_VALID) == 0) { 286 if (nuidp) 287 free((caddr_t)nuidp, M_NFSUID); 288 } else { 289 if (nuidp == (struct nfsuid *)0) { 290 nuidp = slp->ns_uidlruhead.tqh_first; 291 LIST_REMOVE(nuidp, nu_hash); 292 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 293 nu_lru); 294 if (nuidp->nu_flag & NU_NAM) 295 FREE(nuidp->nu_nam, M_SONAME); 296 } 297 nuidp->nu_flag = 0; 298 nuidp->nu_cr = nsd->nsd_cr; 299 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 300 nuidp->nu_cr.cr_ngroups = NGROUPS; 301 nuidp->nu_cr.cr_ref = 1; 302 nuidp->nu_timestamp = nsd->nsd_timestamp; 303 nuidp->nu_expire = time.tv_sec + nsd->nsd_ttl; 304 /* 305 * and save the session key in nu_key. 306 */ 307 bcopy(nsd->nsd_key, nuidp->nu_key, 308 sizeof (nsd->nsd_key)); 309 if (nfsd->nfsd_nd->nd_nam2) { 310 struct sockaddr_in *saddr; 311 312 saddr = (struct sockaddr_in *) 313 nfsd->nfsd_nd->nd_nam2; 314 switch (saddr->sin_family) { 315 case AF_INET: 316 nuidp->nu_flag |= NU_INETADDR; 317 nuidp->nu_inetaddr = 318 saddr->sin_addr.s_addr; 319 break; 320 case AF_ISO: 321 default: 322 nuidp->nu_flag |= NU_NAM; 323 nuidp->nu_nam = 324 dup_sockaddr(nfsd->nfsd_nd-> 325 nd_nam2, 1); 326 break; 327 }; 328 } 329 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 330 nu_lru); 331 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 332 nuidp, nu_hash); 333 nfsrv_setcred(&nuidp->nu_cr, 334 &nfsd->nfsd_nd->nd_cr); 335 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 336 } 337 } 338 } 339 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 340 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 341 error = nfssvc_nfsd(nsd, uap->argp, p); 342 } 343#endif /* NFS_NOSERVER */ 344 if (error == EINTR || error == ERESTART) 345 error = 0; 346 return (error); 347} 348 349#ifndef NFS_NOSERVER 350/* 351 * Adds a socket to the list for servicing by nfsds. 352 */ 353static int 354nfssvc_addsock(fp, mynam, p) 355 struct file *fp; 356 struct sockaddr *mynam; 357 struct proc *p; 358{ 359 register struct mbuf *m; 360 register int siz; 361 register struct nfssvc_sock *slp; 362 register struct socket *so; 363 struct nfssvc_sock *tslp; 364 int error, s; 365 366 so = (struct socket *)fp->f_data; 367 tslp = (struct nfssvc_sock *)0; 368 /* 369 * Add it to the list, as required. 370 */ 371 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 372 tslp = nfs_udpsock; 373 if (tslp->ns_flag & SLP_VALID) { 374 FREE(mynam, M_SONAME); 375 return (EPERM); 376 } 377#ifdef ISO 378 } else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) { 379 tslp = nfs_cltpsock; 380 if (tslp->ns_flag & SLP_VALID) { 381 FREE(mynam, M_SONAME); 382 return (EPERM); 383 } 384#endif /* ISO */ 385 } 386 if (so->so_type == SOCK_STREAM) 387 siz = NFS_MAXPACKET + sizeof (u_long); 388 else 389 siz = NFS_MAXPACKET; 390 error = soreserve(so, siz, siz); 391 if (error) { 392 FREE(mynam, M_SONAME); 393 return (error); 394 } 395 396 /* 397 * Set protocol specific options { for now TCP only } and 398 * reserve some space. For datagram sockets, this can get called 399 * repeatedly for the same socket, but that isn't harmful. 400 */ 401 if (so->so_type == SOCK_STREAM) { 402 MGET(m, M_WAIT, MT_SOOPTS); 403 *mtod(m, int *) = 1; 404 m->m_len = sizeof(int); 405 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m, p); 406 } 407 if (so->so_proto->pr_domain->dom_family == AF_INET && 408 so->so_proto->pr_protocol == IPPROTO_TCP) { 409 MGET(m, M_WAIT, MT_SOOPTS); 410 *mtod(m, int *) = 1; 411 m->m_len = sizeof(int); 412 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m, p); 413 } 414 so->so_rcv.sb_flags &= ~SB_NOINTR; 415 so->so_rcv.sb_timeo = 0; 416 so->so_snd.sb_flags &= ~SB_NOINTR; 417 so->so_snd.sb_timeo = 0; 418 if (tslp) 419 slp = tslp; 420 else { 421 slp = (struct nfssvc_sock *) 422 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 423 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 424 STAILQ_INIT(&slp->ns_rec); 425 TAILQ_INIT(&slp->ns_uidlruhead); 426 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 427 } 428 slp->ns_so = so; 429 slp->ns_nam = mynam; 430 fp->f_count++; 431 slp->ns_fp = fp; 432 s = splnet(); 433 so->so_upcallarg = (caddr_t)slp; 434 so->so_upcall = nfsrv_rcv; 435 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 436 nfsrv_wakenfsd(slp); 437 splx(s); 438 return (0); 439} 440 441/* 442 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 443 * until it is killed by a signal. 444 */ 445static int 446nfssvc_nfsd(nsd, argp, p) 447 struct nfsd_srvargs *nsd; 448 caddr_t argp; 449 struct proc *p; 450{ 451 register struct mbuf *m; 452 register int siz; 453 register struct nfssvc_sock *slp; 454 register struct socket *so; 455 register int *solockp; 456 struct nfsd *nfsd = nsd->nsd_nfsd; 457 struct nfsrv_descript *nd = NULL; 458 struct mbuf *mreq; 459 int error = 0, cacherep, s, sotype, writes_todo; 460 int procrastinate; 461 u_quad_t cur_usec; 462 463#ifndef nolint 464 cacherep = RC_DOIT; 465 writes_todo = 0; 466#endif 467 if (nfsd == (struct nfsd *)0) { 468 nfsd = (struct nfsd *) 469 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 470 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 471 s = splnet(); 472 nsd->nsd_nfsd = nfsd; 473 nfsd->nfsd_procp = p; 474 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 475 nfs_numnfsd++; 476 } else 477 s = splnet(); 478 479 /* 480 * Loop getting rpc requests until SIGKILL. 481 */ 482 for (;;) { 483 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 484 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 485 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 486 nfsd->nfsd_flag |= NFSD_WAITING; 487 nfsd_waiting++; 488 error = tsleep((caddr_t)nfsd, PSOCK | PCATCH, 489 "nfsd", 0); 490 nfsd_waiting--; 491 if (error) 492 goto done; 493 } 494 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 495 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 496 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 497 slp = slp->ns_chain.tqe_next) { 498 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 499 == (SLP_VALID | SLP_DOREC)) { 500 slp->ns_flag &= ~SLP_DOREC; 501 slp->ns_sref++; 502 nfsd->nfsd_slp = slp; 503 break; 504 } 505 } 506 if (slp == 0) 507 nfsd_head_flag &= ~NFSD_CHECKSLP; 508 } 509 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 510 continue; 511 if (slp->ns_flag & SLP_VALID) { 512 if (slp->ns_flag & SLP_DISCONN) 513 nfsrv_zapsock(slp); 514 else if (slp->ns_flag & SLP_NEEDQ) { 515 slp->ns_flag &= ~SLP_NEEDQ; 516 (void) nfs_sndlock(&slp->ns_solock, 517 (struct nfsreq *)0); 518 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 519 M_WAIT); 520 nfs_sndunlock(&slp->ns_solock); 521 } 522 error = nfsrv_dorec(slp, nfsd, &nd); 523 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 524 (u_quad_t)time.tv_usec; 525 if (error && slp->ns_tq.lh_first && 526 slp->ns_tq.lh_first->nd_time <= cur_usec) { 527 error = 0; 528 cacherep = RC_DOIT; 529 writes_todo = 1; 530 } else 531 writes_todo = 0; 532 nfsd->nfsd_flag |= NFSD_REQINPROG; 533 } 534 } else { 535 error = 0; 536 slp = nfsd->nfsd_slp; 537 } 538 if (error || (slp->ns_flag & SLP_VALID) == 0) { 539 if (nd) { 540 free((caddr_t)nd, M_NFSRVDESC); 541 nd = NULL; 542 } 543 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 544 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 545 nfsrv_slpderef(slp); 546 continue; 547 } 548 splx(s); 549 so = slp->ns_so; 550 sotype = so->so_type; 551 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 552 solockp = &slp->ns_solock; 553 else 554 solockp = (int *)0; 555 if (nd) { 556 gettime(&nd->nd_starttime); 557 if (nd->nd_nam2) 558 nd->nd_nam = nd->nd_nam2; 559 else 560 nd->nd_nam = slp->ns_nam; 561 562 /* 563 * Check to see if authorization is needed. 564 */ 565 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 566 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 567 nsd->nsd_haddr = 568 ((struct sockaddr_in *) 569 nd->nd_nam)->sin_addr.s_addr; 570 nsd->nsd_authlen = nfsd->nfsd_authlen; 571 nsd->nsd_verflen = nfsd->nfsd_verflen; 572 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 573 nfsd->nfsd_authlen) && 574 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 575 nfsd->nfsd_verflen) && 576 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 577 return (ENEEDAUTH); 578 cacherep = RC_DROPIT; 579 } else 580 cacherep = nfsrv_getcache(nd, slp, &mreq); 581 582 /* 583 * Check for just starting up for NQNFS and send 584 * fake "try again later" replies to the NQNFS clients. 585 */ 586 if (notstarted && nqnfsstarttime <= time.tv_sec) { 587 if (modify_flag) { 588 nqnfsstarttime = time.tv_sec + nqsrv_writeslack; 589 modify_flag = 0; 590 } else 591 notstarted = 0; 592 } 593 if (notstarted) { 594 if ((nd->nd_flag & ND_NQNFS) == 0) 595 cacherep = RC_DROPIT; 596 else if (nd->nd_procnum != NFSPROC_WRITE) { 597 nd->nd_procnum = NFSPROC_NOOP; 598 nd->nd_repstat = NQNFS_TRYLATER; 599 cacherep = RC_DOIT; 600 } else 601 modify_flag = 1; 602 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 603 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 604 nd->nd_procnum = NFSPROC_NOOP; 605 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 606 cacherep = RC_DOIT; 607 } else if (nfs_privport) { 608 /* Check if source port is privileged */ 609 u_short port; 610 struct sockaddr *nam = nd->nd_nam; 611 struct sockaddr_in *sin; 612 613 sin = (struct sockaddr_in *)nam; 614 port = ntohs(sin->sin_port); 615 if (port >= IPPORT_RESERVED && 616 nd->nd_procnum != NFSPROC_NULL) { 617 nd->nd_procnum = NFSPROC_NOOP; 618 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 619 cacherep = RC_DOIT; 620 printf("NFS request from unprivileged port (%s:%d)\n", 621 inet_ntoa(sin->sin_addr), port); 622 } 623 } 624 625 } 626 627 /* 628 * Loop to get all the write rpc relies that have been 629 * gathered together. 630 */ 631 do { 632 switch (cacherep) { 633 case RC_DOIT: 634 if (nd && (nd->nd_flag & ND_NFSV3)) 635 procrastinate = nfsrvw_procrastinate_v3; 636 else 637 procrastinate = nfsrvw_procrastinate; 638 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 639 procrastinate > 0 && !notstarted)) 640 error = nfsrv_writegather(&nd, slp, 641 nfsd->nfsd_procp, &mreq); 642 else 643 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 644 slp, nfsd->nfsd_procp, &mreq); 645 if (mreq == NULL) 646 break; 647 if (error) { 648 if (nd->nd_procnum != NQNFSPROC_VACATED) 649 nfsstats.srv_errs++; 650 nfsrv_updatecache(nd, FALSE, mreq); 651 if (nd->nd_nam2) 652 FREE(nd->nd_nam2, M_SONAME); 653 break; 654 } 655 nfsstats.srvrpccnt[nd->nd_procnum]++; 656 nfsrv_updatecache(nd, TRUE, mreq); 657 nd->nd_mrep = (struct mbuf *)0; 658 case RC_REPLY: 659 m = mreq; 660 siz = 0; 661 while (m) { 662 siz += m->m_len; 663 m = m->m_next; 664 } 665 if (siz <= 0 || siz > NFS_MAXPACKET) { 666 printf("mbuf siz=%d\n",siz); 667 panic("Bad nfs svc reply"); 668 } 669 m = mreq; 670 m->m_pkthdr.len = siz; 671 m->m_pkthdr.rcvif = (struct ifnet *)0; 672 /* 673 * For stream protocols, prepend a Sun RPC 674 * Record Mark. 675 */ 676 if (sotype == SOCK_STREAM) { 677 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 678 *mtod(m, u_long *) = htonl(0x80000000 | siz); 679 } 680 if (solockp) 681 (void) nfs_sndlock(solockp, (struct nfsreq *)0); 682 if (slp->ns_flag & SLP_VALID) 683 error = nfs_send(so, nd->nd_nam2, m, NULL); 684 else { 685 error = EPIPE; 686 m_freem(m); 687 } 688 if (nfsrtton) 689 nfsd_rt(sotype, nd, cacherep); 690 if (nd->nd_nam2) 691 FREE(nd->nd_nam2, M_SONAME); 692 if (nd->nd_mrep) 693 m_freem(nd->nd_mrep); 694 if (error == EPIPE) 695 nfsrv_zapsock(slp); 696 if (solockp) 697 nfs_sndunlock(solockp); 698 if (error == EINTR || error == ERESTART) { 699 free((caddr_t)nd, M_NFSRVDESC); 700 nfsrv_slpderef(slp); 701 s = splnet(); 702 goto done; 703 } 704 break; 705 case RC_DROPIT: 706 if (nfsrtton) 707 nfsd_rt(sotype, nd, cacherep); 708 m_freem(nd->nd_mrep); 709 FREE(nd->nd_nam2, M_SONAME); 710 break; 711 }; 712 if (nd) { 713 FREE((caddr_t)nd, M_NFSRVDESC); 714 nd = NULL; 715 } 716 717 /* 718 * Check to see if there are outstanding writes that 719 * need to be serviced. 720 */ 721 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 722 (u_quad_t)time.tv_usec; 723 s = splsoftclock(); 724 if (slp->ns_tq.lh_first && 725 slp->ns_tq.lh_first->nd_time <= cur_usec) { 726 cacherep = RC_DOIT; 727 writes_todo = 1; 728 } else 729 writes_todo = 0; 730 splx(s); 731 } while (writes_todo); 732 s = splnet(); 733 if (nfsrv_dorec(slp, nfsd, &nd)) { 734 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 735 nfsd->nfsd_slp = NULL; 736 nfsrv_slpderef(slp); 737 } 738 } 739done: 740 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 741 splx(s); 742 free((caddr_t)nfsd, M_NFSD); 743 nsd->nsd_nfsd = (struct nfsd *)0; 744 if (--nfs_numnfsd == 0) 745 nfsrv_init(TRUE); /* Reinitialize everything */ 746 return (error); 747} 748#endif /* NFS_NOSERVER */ 749 750int nfs_defect = 0; 751SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 752 753/* 754 * Asynchronous I/O daemons for client nfs. 755 * They do read-ahead and write-behind operations on the block I/O cache. 756 * Never returns unless it fails or gets killed. 757 */ 758static int 759nfssvc_iod(p) 760 struct proc *p; 761{ 762 register struct buf *bp; 763 register int i, myiod; 764 struct nfsmount *nmp; 765 int error = 0; 766 767 /* 768 * Assign my position or return error if too many already running 769 */ 770 myiod = -1; 771 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 772 if (nfs_asyncdaemon[i] == 0) { 773 nfs_asyncdaemon[i]++; 774 myiod = i; 775 break; 776 } 777 if (myiod == -1) 778 return (EBUSY); 779 nfs_numasync++; 780 /* 781 * Just loop around doin our stuff until SIGKILL 782 */ 783 for (;;) { 784 while (((nmp = nfs_iodmount[myiod]) == NULL 785 || nmp->nm_bufq.tqh_first == NULL) 786 && error == 0) { 787 if (nmp) 788 nmp->nm_bufqiods--; 789 nfs_iodwant[myiod] = p; 790 nfs_iodmount[myiod] = NULL; 791 error = tsleep((caddr_t)&nfs_iodwant[myiod], 792 PWAIT | PCATCH, "nfsidl", 0); 793 } 794 if (error) { 795 nfs_asyncdaemon[myiod] = 0; 796 if (nmp) nmp->nm_bufqiods--; 797 nfs_iodwant[myiod] = NULL; 798 nfs_iodmount[myiod] = NULL; 799 nfs_numasync--; 800 return (error); 801 } 802 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 803 /* Take one off the front of the list */ 804 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 805 nmp->nm_bufqlen--; 806 if (nmp->nm_bufqwant && nmp->nm_bufqlen < 2 * nfs_numasync) { 807 nmp->nm_bufqwant = FALSE; 808 wakeup(&nmp->nm_bufq); 809 } 810 if (bp->b_flags & B_READ) 811 (void) nfs_doio(bp, bp->b_rcred, (struct proc *)0); 812 else 813 (void) nfs_doio(bp, bp->b_wcred, (struct proc *)0); 814 815 /* 816 * If there are more than one iod on this mount, then defect 817 * so that the iods can be shared out fairly between the mounts 818 */ 819 if (nfs_defect && nmp->nm_bufqiods > 1) { 820 NFS_DPF(ASYNCIO, 821 ("nfssvc_iod: iod %d defecting from mount %p\n", 822 myiod, nmp)); 823 nfs_iodmount[myiod] = NULL; 824 nmp->nm_bufqiods--; 825 break; 826 } 827 } 828 } 829} 830 831/* 832 * Shut down a socket associated with an nfssvc_sock structure. 833 * Should be called with the send lock set, if required. 834 * The trick here is to increment the sref at the start, so that the nfsds 835 * will stop using it and clear ns_flag at the end so that it will not be 836 * reassigned during cleanup. 837 */ 838static void 839nfsrv_zapsock(slp) 840 register struct nfssvc_sock *slp; 841{ 842 register struct nfsuid *nuidp, *nnuidp; 843 register struct nfsrv_descript *nwp, *nnwp; 844 struct socket *so; 845 struct file *fp; 846 struct nfsrv_rec *rec; 847 int s; 848 849 slp->ns_flag &= ~SLP_ALLFLAGS; 850 fp = slp->ns_fp; 851 if (fp) { 852 slp->ns_fp = (struct file *)0; 853 so = slp->ns_so; 854 so->so_upcall = NULL; 855 soshutdown(so, 2); 856 closef(fp, (struct proc *)0); 857 if (slp->ns_nam) 858 FREE(slp->ns_nam, M_SONAME); 859 m_freem(slp->ns_raw); 860 while (rec = STAILQ_FIRST(&slp->ns_rec)) { 861 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 862 if (rec->nr_address) 863 FREE(rec->nr_address, M_SONAME); 864 m_freem(rec->nr_packet); 865 free(rec, M_NFSRVDESC); 866 } 867 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 868 nuidp = nnuidp) { 869 nnuidp = nuidp->nu_lru.tqe_next; 870 LIST_REMOVE(nuidp, nu_hash); 871 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 872 if (nuidp->nu_flag & NU_NAM) 873 FREE(nuidp->nu_nam, M_SONAME); 874 free((caddr_t)nuidp, M_NFSUID); 875 } 876 s = splsoftclock(); 877 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 878 nnwp = nwp->nd_tq.le_next; 879 LIST_REMOVE(nwp, nd_tq); 880 free((caddr_t)nwp, M_NFSRVDESC); 881 } 882 LIST_INIT(&slp->ns_tq); 883 splx(s); 884 } 885} 886 887/* 888 * Get an authorization string for the uid by having the mount_nfs sitting 889 * on this mount point porpous out of the kernel and do it. 890 */ 891int 892nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 893 register struct nfsmount *nmp; 894 struct nfsreq *rep; 895 struct ucred *cred; 896 char **auth_str; 897 int *auth_len; 898 char *verf_str; 899 int *verf_len; 900 NFSKERBKEY_T key; /* return session key */ 901{ 902 int error = 0; 903 904 while ((nmp->nm_flag & NFSMNT_WAITAUTH) == 0) { 905 nmp->nm_flag |= NFSMNT_WANTAUTH; 906 (void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK, 907 "nfsauth1", 2 * hz); 908 error = nfs_sigintr(nmp, rep, rep->r_procp); 909 if (error) { 910 nmp->nm_flag &= ~NFSMNT_WANTAUTH; 911 return (error); 912 } 913 } 914 nmp->nm_flag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH); 915 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 916 nmp->nm_authlen = RPCAUTH_MAXSIZ; 917 nmp->nm_verfstr = verf_str; 918 nmp->nm_verflen = *verf_len; 919 nmp->nm_authuid = cred->cr_uid; 920 wakeup((caddr_t)&nmp->nm_authstr); 921 922 /* 923 * And wait for mount_nfs to do its stuff. 924 */ 925 while ((nmp->nm_flag & NFSMNT_HASAUTH) == 0 && error == 0) { 926 (void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK, 927 "nfsauth2", 2 * hz); 928 error = nfs_sigintr(nmp, rep, rep->r_procp); 929 } 930 if (nmp->nm_flag & NFSMNT_AUTHERR) { 931 nmp->nm_flag &= ~NFSMNT_AUTHERR; 932 error = EAUTH; 933 } 934 if (error) 935 free((caddr_t)*auth_str, M_TEMP); 936 else { 937 *auth_len = nmp->nm_authlen; 938 *verf_len = nmp->nm_verflen; 939 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 940 } 941 nmp->nm_flag &= ~NFSMNT_HASAUTH; 942 nmp->nm_flag |= NFSMNT_WAITAUTH; 943 if (nmp->nm_flag & NFSMNT_WANTAUTH) { 944 nmp->nm_flag &= ~NFSMNT_WANTAUTH; 945 wakeup((caddr_t)&nmp->nm_authtype); 946 } 947 return (error); 948} 949 950/* 951 * Get a nickname authenticator and verifier. 952 */ 953int 954nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 955 struct nfsmount *nmp; 956 struct ucred *cred; 957 char **auth_str; 958 int *auth_len; 959 char *verf_str; 960 int verf_len; 961{ 962 register struct nfsuid *nuidp; 963 register u_long *nickp, *verfp; 964 struct timeval ktvin, ktvout; 965 966#ifdef DIAGNOSTIC 967 if (verf_len < (4 * NFSX_UNSIGNED)) 968 panic("nfs_getnickauth verf too small"); 969#endif 970 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 971 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 972 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 973 break; 974 } 975 if (!nuidp || nuidp->nu_expire < time.tv_sec) 976 return (EACCES); 977 978 /* 979 * Move to the end of the lru list (end of lru == most recently used). 980 */ 981 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 982 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 983 984 nickp = (u_long *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 985 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 986 *nickp = txdr_unsigned(nuidp->nu_nickname); 987 *auth_str = (char *)nickp; 988 *auth_len = 2 * NFSX_UNSIGNED; 989 990 /* 991 * Now we must encrypt the verifier and package it up. 992 */ 993 verfp = (u_long *)verf_str; 994 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 995 if (time.tv_sec > nuidp->nu_timestamp.tv_sec || 996 (time.tv_sec == nuidp->nu_timestamp.tv_sec && 997 time.tv_usec > nuidp->nu_timestamp.tv_usec)) 998 gettime(&nuidp->nu_timestamp); 999 else 1000 nuidp->nu_timestamp.tv_usec++; 1001 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1002 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1003 1004 /* 1005 * Now encrypt the timestamp verifier in ecb mode using the session 1006 * key. 1007 */ 1008#ifdef NFSKERB 1009 XXX 1010#endif 1011 1012 *verfp++ = ktvout.tv_sec; 1013 *verfp++ = ktvout.tv_usec; 1014 *verfp = 0; 1015 return (0); 1016} 1017 1018/* 1019 * Save the current nickname in a hash list entry on the mount point. 1020 */ 1021int 1022nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 1023 register struct nfsmount *nmp; 1024 struct ucred *cred; 1025 int len; 1026 NFSKERBKEY_T key; 1027 struct mbuf **mdp; 1028 char **dposp; 1029 struct mbuf *mrep; 1030{ 1031 register struct nfsuid *nuidp; 1032 register u_long *tl; 1033 register long t1; 1034 struct mbuf *md = *mdp; 1035 struct timeval ktvin, ktvout; 1036 u_long nick; 1037 char *dpos = *dposp, *cp2; 1038 int deltasec, error = 0; 1039 1040 if (len == (3 * NFSX_UNSIGNED)) { 1041 nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); 1042 ktvin.tv_sec = *tl++; 1043 ktvin.tv_usec = *tl++; 1044 nick = fxdr_unsigned(u_long, *tl); 1045 1046 /* 1047 * Decrypt the timestamp in ecb mode. 1048 */ 1049#ifdef NFSKERB 1050 XXX 1051#endif 1052 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1053 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1054 deltasec = time.tv_sec - ktvout.tv_sec; 1055 if (deltasec < 0) 1056 deltasec = -deltasec; 1057 /* 1058 * If ok, add it to the hash list for the mount point. 1059 */ 1060 if (deltasec <= NFS_KERBCLOCKSKEW) { 1061 if (nmp->nm_numuids < nuidhash_max) { 1062 nmp->nm_numuids++; 1063 nuidp = (struct nfsuid *) 1064 malloc(sizeof (struct nfsuid), M_NFSUID, 1065 M_WAITOK); 1066 } else { 1067 nuidp = nmp->nm_uidlruhead.tqh_first; 1068 LIST_REMOVE(nuidp, nu_hash); 1069 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1070 nu_lru); 1071 } 1072 nuidp->nu_flag = 0; 1073 nuidp->nu_cr.cr_uid = cred->cr_uid; 1074 nuidp->nu_expire = time.tv_sec + NFS_KERBTTL; 1075 nuidp->nu_timestamp = ktvout; 1076 nuidp->nu_nickname = nick; 1077 bcopy(key, nuidp->nu_key, sizeof (key)); 1078 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1079 nu_lru); 1080 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1081 nuidp, nu_hash); 1082 } 1083 } else 1084 nfsm_adv(nfsm_rndup(len)); 1085nfsmout: 1086 *mdp = md; 1087 *dposp = dpos; 1088 return (error); 1089} 1090 1091#ifndef NFS_NOSERVER 1092 1093/* 1094 * Derefence a server socket structure. If it has no more references and 1095 * is no longer valid, you can throw it away. 1096 */ 1097void 1098nfsrv_slpderef(slp) 1099 register struct nfssvc_sock *slp; 1100{ 1101 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 1102 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1103 free((caddr_t)slp, M_NFSSVC); 1104 } 1105} 1106 1107/* 1108 * Initialize the data structures for the server. 1109 * Handshake with any new nfsds starting up to avoid any chance of 1110 * corruption. 1111 */ 1112void 1113nfsrv_init(terminating) 1114 int terminating; 1115{ 1116 register struct nfssvc_sock *slp, *nslp; 1117 1118 if (nfssvc_sockhead_flag & SLP_INIT) 1119 panic("nfsd init"); 1120 nfssvc_sockhead_flag |= SLP_INIT; 1121 if (terminating) { 1122 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 1123 nslp = slp->ns_chain.tqe_next; 1124 if (slp->ns_flag & SLP_VALID) 1125 nfsrv_zapsock(slp); 1126 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1127 free((caddr_t)slp, M_NFSSVC); 1128 } 1129 nfsrv_cleancache(); /* And clear out server cache */ 1130 } else 1131 nfs_pub.np_valid = 0; 1132 1133 TAILQ_INIT(&nfssvc_sockhead); 1134 nfssvc_sockhead_flag &= ~SLP_INIT; 1135 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 1136 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 1137 wakeup((caddr_t)&nfssvc_sockhead); 1138 } 1139 1140 TAILQ_INIT(&nfsd_head); 1141 nfsd_head_flag &= ~NFSD_CHECKSLP; 1142 1143 nfs_udpsock = (struct nfssvc_sock *) 1144 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1145 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 1146 STAILQ_INIT(&nfs_udpsock->ns_rec); 1147 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 1148 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 1149 1150 nfs_cltpsock = (struct nfssvc_sock *) 1151 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1152 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 1153 STAILQ_INIT(&nfs_cltpsock->ns_rec); 1154 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 1155 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 1156} 1157 1158/* 1159 * Add entries to the server monitor log. 1160 */ 1161static void 1162nfsd_rt(sotype, nd, cacherep) 1163 int sotype; 1164 register struct nfsrv_descript *nd; 1165 int cacherep; 1166{ 1167 register struct drt *rt; 1168 1169 rt = &nfsdrt.drt[nfsdrt.pos]; 1170 if (cacherep == RC_DOIT) 1171 rt->flag = 0; 1172 else if (cacherep == RC_REPLY) 1173 rt->flag = DRT_CACHEREPLY; 1174 else 1175 rt->flag = DRT_CACHEDROP; 1176 if (sotype == SOCK_STREAM) 1177 rt->flag |= DRT_TCP; 1178 if (nd->nd_flag & ND_NQNFS) 1179 rt->flag |= DRT_NQNFS; 1180 else if (nd->nd_flag & ND_NFSV3) 1181 rt->flag |= DRT_NFSV3; 1182 rt->proc = nd->nd_procnum; 1183 if (nd->nd_nam->sa_family == AF_INET) 1184 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 1185 else 1186 rt->ipadr = INADDR_ANY; 1187 rt->resptime = ((time.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 1188 (time.tv_usec - nd->nd_starttime.tv_usec); 1189 gettime(&rt->tstamp); 1190 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1191} 1192#endif /* NFS_NOSERVER */ 1193