nfs_nfsiod.c revision 36176
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 1.37 1998/03/30 09:54:17 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_state & NFSSTA_MNTD) && 227 (uap->flag & NFSSVC_GOTAUTH) == 0) 228 return (0); 229 nmp->nm_state |= NFSSTA_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_second + 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 &slp->ns_solock, 518 (struct nfsreq *)0); 519 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 520 M_WAIT); 521 nfs_sndunlock(&slp->ns_solock, 522 &slp->ns_solock); 523 } 524 error = nfsrv_dorec(slp, nfsd, &nd); 525 cur_usec = nfs_curusec(); 526 if (error && slp->ns_tq.lh_first && 527 slp->ns_tq.lh_first->nd_time <= cur_usec) { 528 error = 0; 529 cacherep = RC_DOIT; 530 writes_todo = 1; 531 } else 532 writes_todo = 0; 533 nfsd->nfsd_flag |= NFSD_REQINPROG; 534 } 535 } else { 536 error = 0; 537 slp = nfsd->nfsd_slp; 538 } 539 if (error || (slp->ns_flag & SLP_VALID) == 0) { 540 if (nd) { 541 free((caddr_t)nd, M_NFSRVDESC); 542 nd = NULL; 543 } 544 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 545 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 546 nfsrv_slpderef(slp); 547 continue; 548 } 549 splx(s); 550 so = slp->ns_so; 551 sotype = so->so_type; 552 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 553 solockp = &slp->ns_solock; 554 else 555 solockp = (int *)0; 556 if (nd) { 557 getmicrotime(&nd->nd_starttime); 558 if (nd->nd_nam2) 559 nd->nd_nam = nd->nd_nam2; 560 else 561 nd->nd_nam = slp->ns_nam; 562 563 /* 564 * Check to see if authorization is needed. 565 */ 566 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 567 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 568 nsd->nsd_haddr = 569 ((struct sockaddr_in *) 570 nd->nd_nam)->sin_addr.s_addr; 571 nsd->nsd_authlen = nfsd->nfsd_authlen; 572 nsd->nsd_verflen = nfsd->nfsd_verflen; 573 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 574 nfsd->nfsd_authlen) && 575 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 576 nfsd->nfsd_verflen) && 577 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 578 return (ENEEDAUTH); 579 cacherep = RC_DROPIT; 580 } else 581 cacherep = nfsrv_getcache(nd, slp, &mreq); 582 583 /* 584 * Check for just starting up for NQNFS and send 585 * fake "try again later" replies to the NQNFS clients. 586 */ 587 if (notstarted && nqnfsstarttime <= time_second) { 588 if (modify_flag) { 589 nqnfsstarttime = time_second + nqsrv_writeslack; 590 modify_flag = 0; 591 } else 592 notstarted = 0; 593 } 594 if (notstarted) { 595 if ((nd->nd_flag & ND_NQNFS) == 0) 596 cacherep = RC_DROPIT; 597 else if (nd->nd_procnum != NFSPROC_WRITE) { 598 nd->nd_procnum = NFSPROC_NOOP; 599 nd->nd_repstat = NQNFS_TRYLATER; 600 cacherep = RC_DOIT; 601 } else 602 modify_flag = 1; 603 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 604 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 605 nd->nd_procnum = NFSPROC_NOOP; 606 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 607 cacherep = RC_DOIT; 608 } else if (nfs_privport) { 609 /* Check if source port is privileged */ 610 u_short port; 611 struct sockaddr *nam = nd->nd_nam; 612 struct sockaddr_in *sin; 613 614 sin = (struct sockaddr_in *)nam; 615 port = ntohs(sin->sin_port); 616 if (port >= IPPORT_RESERVED && 617 nd->nd_procnum != NFSPROC_NULL) { 618 nd->nd_procnum = NFSPROC_NOOP; 619 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 620 cacherep = RC_DOIT; 621 printf("NFS request from unprivileged port (%s:%d)\n", 622 inet_ntoa(sin->sin_addr), port); 623 } 624 } 625 626 } 627 628 /* 629 * Loop to get all the write rpc relies that have been 630 * gathered together. 631 */ 632 do { 633 switch (cacherep) { 634 case RC_DOIT: 635 if (nd && (nd->nd_flag & ND_NFSV3)) 636 procrastinate = nfsrvw_procrastinate_v3; 637 else 638 procrastinate = nfsrvw_procrastinate; 639 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 640 procrastinate > 0 && !notstarted)) 641 error = nfsrv_writegather(&nd, slp, 642 nfsd->nfsd_procp, &mreq); 643 else 644 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 645 slp, nfsd->nfsd_procp, &mreq); 646 if (mreq == NULL) 647 break; 648 if (error) { 649 if (nd->nd_procnum != NQNFSPROC_VACATED) 650 nfsstats.srv_errs++; 651 nfsrv_updatecache(nd, FALSE, mreq); 652 if (nd->nd_nam2) 653 FREE(nd->nd_nam2, M_SONAME); 654 break; 655 } 656 nfsstats.srvrpccnt[nd->nd_procnum]++; 657 nfsrv_updatecache(nd, TRUE, mreq); 658 nd->nd_mrep = (struct mbuf *)0; 659 case RC_REPLY: 660 m = mreq; 661 siz = 0; 662 while (m) { 663 siz += m->m_len; 664 m = m->m_next; 665 } 666 if (siz <= 0 || siz > NFS_MAXPACKET) { 667 printf("mbuf siz=%d\n",siz); 668 panic("Bad nfs svc reply"); 669 } 670 m = mreq; 671 m->m_pkthdr.len = siz; 672 m->m_pkthdr.rcvif = (struct ifnet *)0; 673 /* 674 * For stream protocols, prepend a Sun RPC 675 * Record Mark. 676 */ 677 if (sotype == SOCK_STREAM) { 678 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 679 *mtod(m, u_long *) = htonl(0x80000000 | siz); 680 } 681 if (solockp) 682 (void) nfs_sndlock(solockp, solockp, 683 (struct nfsreq *)0); 684 if (slp->ns_flag & SLP_VALID) 685 error = nfs_send(so, nd->nd_nam2, m, NULL); 686 else { 687 error = EPIPE; 688 m_freem(m); 689 } 690 if (nfsrtton) 691 nfsd_rt(sotype, nd, cacherep); 692 if (nd->nd_nam2) 693 FREE(nd->nd_nam2, M_SONAME); 694 if (nd->nd_mrep) 695 m_freem(nd->nd_mrep); 696 if (error == EPIPE) 697 nfsrv_zapsock(slp); 698 if (solockp) 699 nfs_sndunlock(solockp, solockp); 700 if (error == EINTR || error == ERESTART) { 701 free((caddr_t)nd, M_NFSRVDESC); 702 nfsrv_slpderef(slp); 703 s = splnet(); 704 goto done; 705 } 706 break; 707 case RC_DROPIT: 708 if (nfsrtton) 709 nfsd_rt(sotype, nd, cacherep); 710 m_freem(nd->nd_mrep); 711 FREE(nd->nd_nam2, M_SONAME); 712 break; 713 }; 714 if (nd) { 715 FREE((caddr_t)nd, M_NFSRVDESC); 716 nd = NULL; 717 } 718 719 /* 720 * Check to see if there are outstanding writes that 721 * need to be serviced. 722 */ 723 cur_usec = nfs_curusec(); 724 s = splsoftclock(); 725 if (slp->ns_tq.lh_first && 726 slp->ns_tq.lh_first->nd_time <= cur_usec) { 727 cacherep = RC_DOIT; 728 writes_todo = 1; 729 } else 730 writes_todo = 0; 731 splx(s); 732 } while (writes_todo); 733 s = splnet(); 734 if (nfsrv_dorec(slp, nfsd, &nd)) { 735 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 736 nfsd->nfsd_slp = NULL; 737 nfsrv_slpderef(slp); 738 } 739 } 740done: 741 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 742 splx(s); 743 free((caddr_t)nfsd, M_NFSD); 744 nsd->nsd_nfsd = (struct nfsd *)0; 745 if (--nfs_numnfsd == 0) 746 nfsrv_init(TRUE); /* Reinitialize everything */ 747 return (error); 748} 749#endif /* NFS_NOSERVER */ 750 751static int nfs_defect = 0; 752SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 753 754/* 755 * Asynchronous I/O daemons for client nfs. 756 * They do read-ahead and write-behind operations on the block I/O cache. 757 * Never returns unless it fails or gets killed. 758 */ 759static int 760nfssvc_iod(p) 761 struct proc *p; 762{ 763 register struct buf *bp; 764 register int i, myiod; 765 struct nfsmount *nmp; 766 int error = 0; 767 768 /* 769 * Assign my position or return error if too many already running 770 */ 771 myiod = -1; 772 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 773 if (nfs_asyncdaemon[i] == 0) { 774 nfs_asyncdaemon[i]++; 775 myiod = i; 776 break; 777 } 778 if (myiod == -1) 779 return (EBUSY); 780 nfs_numasync++; 781 /* 782 * Just loop around doin our stuff until SIGKILL 783 */ 784 for (;;) { 785 while (((nmp = nfs_iodmount[myiod]) == NULL 786 || nmp->nm_bufq.tqh_first == NULL) 787 && error == 0) { 788 if (nmp) 789 nmp->nm_bufqiods--; 790 nfs_iodwant[myiod] = p; 791 nfs_iodmount[myiod] = NULL; 792 error = tsleep((caddr_t)&nfs_iodwant[myiod], 793 PWAIT | PCATCH, "nfsidl", 0); 794 } 795 if (error) { 796 nfs_asyncdaemon[myiod] = 0; 797 if (nmp) nmp->nm_bufqiods--; 798 nfs_iodwant[myiod] = NULL; 799 nfs_iodmount[myiod] = NULL; 800 nfs_numasync--; 801 return (error); 802 } 803 while ((bp = nmp->nm_bufq.tqh_first) != NULL) { 804 /* Take one off the front of the list */ 805 TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist); 806 nmp->nm_bufqlen--; 807 if (nmp->nm_bufqwant && nmp->nm_bufqlen < 2 * nfs_numasync) { 808 nmp->nm_bufqwant = FALSE; 809 wakeup(&nmp->nm_bufq); 810 } 811 if (bp->b_flags & B_READ) 812 (void) nfs_doio(bp, bp->b_rcred, (struct proc *)0); 813 else 814 (void) nfs_doio(bp, bp->b_wcred, (struct proc *)0); 815 816 /* 817 * If there are more than one iod on this mount, then defect 818 * so that the iods can be shared out fairly between the mounts 819 */ 820 if (nfs_defect && nmp->nm_bufqiods > 1) { 821 NFS_DPF(ASYNCIO, 822 ("nfssvc_iod: iod %d defecting from mount %p\n", 823 myiod, nmp)); 824 nfs_iodmount[myiod] = NULL; 825 nmp->nm_bufqiods--; 826 break; 827 } 828 } 829 } 830} 831 832/* 833 * Shut down a socket associated with an nfssvc_sock structure. 834 * Should be called with the send lock set, if required. 835 * The trick here is to increment the sref at the start, so that the nfsds 836 * will stop using it and clear ns_flag at the end so that it will not be 837 * reassigned during cleanup. 838 */ 839static void 840nfsrv_zapsock(slp) 841 register struct nfssvc_sock *slp; 842{ 843 register struct nfsuid *nuidp, *nnuidp; 844 register struct nfsrv_descript *nwp, *nnwp; 845 struct socket *so; 846 struct file *fp; 847 struct nfsrv_rec *rec; 848 int s; 849 850 slp->ns_flag &= ~SLP_ALLFLAGS; 851 fp = slp->ns_fp; 852 if (fp) { 853 slp->ns_fp = (struct file *)0; 854 so = slp->ns_so; 855 so->so_upcall = NULL; 856 soshutdown(so, 2); 857 closef(fp, (struct proc *)0); 858 if (slp->ns_nam) 859 FREE(slp->ns_nam, M_SONAME); 860 m_freem(slp->ns_raw); 861 while (rec = STAILQ_FIRST(&slp->ns_rec)) { 862 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 863 if (rec->nr_address) 864 FREE(rec->nr_address, M_SONAME); 865 m_freem(rec->nr_packet); 866 free(rec, M_NFSRVDESC); 867 } 868 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 869 nuidp = nnuidp) { 870 nnuidp = nuidp->nu_lru.tqe_next; 871 LIST_REMOVE(nuidp, nu_hash); 872 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 873 if (nuidp->nu_flag & NU_NAM) 874 FREE(nuidp->nu_nam, M_SONAME); 875 free((caddr_t)nuidp, M_NFSUID); 876 } 877 s = splsoftclock(); 878 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 879 nnwp = nwp->nd_tq.le_next; 880 LIST_REMOVE(nwp, nd_tq); 881 free((caddr_t)nwp, M_NFSRVDESC); 882 } 883 LIST_INIT(&slp->ns_tq); 884 splx(s); 885 } 886} 887 888/* 889 * Get an authorization string for the uid by having the mount_nfs sitting 890 * on this mount point porpous out of the kernel and do it. 891 */ 892int 893nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 894 register struct nfsmount *nmp; 895 struct nfsreq *rep; 896 struct ucred *cred; 897 char **auth_str; 898 int *auth_len; 899 char *verf_str; 900 int *verf_len; 901 NFSKERBKEY_T key; /* return session key */ 902{ 903 int error = 0; 904 905 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 906 nmp->nm_state |= NFSSTA_WANTAUTH; 907 (void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK, 908 "nfsauth1", 2 * hz); 909 error = nfs_sigintr(nmp, rep, rep->r_procp); 910 if (error) { 911 nmp->nm_state &= ~NFSSTA_WANTAUTH; 912 return (error); 913 } 914 } 915 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 916 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 917 nmp->nm_authlen = RPCAUTH_MAXSIZ; 918 nmp->nm_verfstr = verf_str; 919 nmp->nm_verflen = *verf_len; 920 nmp->nm_authuid = cred->cr_uid; 921 wakeup((caddr_t)&nmp->nm_authstr); 922 923 /* 924 * And wait for mount_nfs to do its stuff. 925 */ 926 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 927 (void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK, 928 "nfsauth2", 2 * hz); 929 error = nfs_sigintr(nmp, rep, rep->r_procp); 930 } 931 if (nmp->nm_state & NFSSTA_AUTHERR) { 932 nmp->nm_state &= ~NFSSTA_AUTHERR; 933 error = EAUTH; 934 } 935 if (error) 936 free((caddr_t)*auth_str, M_TEMP); 937 else { 938 *auth_len = nmp->nm_authlen; 939 *verf_len = nmp->nm_verflen; 940 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 941 } 942 nmp->nm_state &= ~NFSSTA_HASAUTH; 943 nmp->nm_state |= NFSSTA_WAITAUTH; 944 if (nmp->nm_state & NFSSTA_WANTAUTH) { 945 nmp->nm_state &= ~NFSSTA_WANTAUTH; 946 wakeup((caddr_t)&nmp->nm_authtype); 947 } 948 return (error); 949} 950 951/* 952 * Get a nickname authenticator and verifier. 953 */ 954int 955nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 956 struct nfsmount *nmp; 957 struct ucred *cred; 958 char **auth_str; 959 int *auth_len; 960 char *verf_str; 961 int verf_len; 962{ 963 register struct nfsuid *nuidp; 964 register u_long *nickp, *verfp; 965 struct timeval ktvin, ktvout; 966 967#ifdef DIAGNOSTIC 968 if (verf_len < (4 * NFSX_UNSIGNED)) 969 panic("nfs_getnickauth verf too small"); 970#endif 971 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 972 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 973 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 974 break; 975 } 976 if (!nuidp || nuidp->nu_expire < time_second) 977 return (EACCES); 978 979 /* 980 * Move to the end of the lru list (end of lru == most recently used). 981 */ 982 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 983 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 984 985 nickp = (u_long *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 986 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 987 *nickp = txdr_unsigned(nuidp->nu_nickname); 988 *auth_str = (char *)nickp; 989 *auth_len = 2 * NFSX_UNSIGNED; 990 991 /* 992 * Now we must encrypt the verifier and package it up. 993 */ 994 verfp = (u_long *)verf_str; 995 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 996 if (time_second > nuidp->nu_timestamp.tv_sec || 997 (time_second == nuidp->nu_timestamp.tv_sec && 998 time_second > nuidp->nu_timestamp.tv_usec)) 999 getmicrotime(&nuidp->nu_timestamp); 1000 else 1001 nuidp->nu_timestamp.tv_usec++; 1002 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1003 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1004 1005 /* 1006 * Now encrypt the timestamp verifier in ecb mode using the session 1007 * key. 1008 */ 1009#ifdef NFSKERB 1010 XXX 1011#endif 1012 1013 *verfp++ = ktvout.tv_sec; 1014 *verfp++ = ktvout.tv_usec; 1015 *verfp = 0; 1016 return (0); 1017} 1018 1019/* 1020 * Save the current nickname in a hash list entry on the mount point. 1021 */ 1022int 1023nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 1024 register struct nfsmount *nmp; 1025 struct ucred *cred; 1026 int len; 1027 NFSKERBKEY_T key; 1028 struct mbuf **mdp; 1029 char **dposp; 1030 struct mbuf *mrep; 1031{ 1032 register struct nfsuid *nuidp; 1033 register u_long *tl; 1034 register long t1; 1035 struct mbuf *md = *mdp; 1036 struct timeval ktvin, ktvout; 1037 u_long nick; 1038 char *dpos = *dposp, *cp2; 1039 int deltasec, error = 0; 1040 1041 if (len == (3 * NFSX_UNSIGNED)) { 1042 nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); 1043 ktvin.tv_sec = *tl++; 1044 ktvin.tv_usec = *tl++; 1045 nick = fxdr_unsigned(u_long, *tl); 1046 1047 /* 1048 * Decrypt the timestamp in ecb mode. 1049 */ 1050#ifdef NFSKERB 1051 XXX 1052#endif 1053 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1054 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1055 deltasec = time_second - ktvout.tv_sec; 1056 if (deltasec < 0) 1057 deltasec = -deltasec; 1058 /* 1059 * If ok, add it to the hash list for the mount point. 1060 */ 1061 if (deltasec <= NFS_KERBCLOCKSKEW) { 1062 if (nmp->nm_numuids < nuidhash_max) { 1063 nmp->nm_numuids++; 1064 nuidp = (struct nfsuid *) 1065 malloc(sizeof (struct nfsuid), M_NFSUID, 1066 M_WAITOK); 1067 } else { 1068 nuidp = nmp->nm_uidlruhead.tqh_first; 1069 LIST_REMOVE(nuidp, nu_hash); 1070 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1071 nu_lru); 1072 } 1073 nuidp->nu_flag = 0; 1074 nuidp->nu_cr.cr_uid = cred->cr_uid; 1075 nuidp->nu_expire = time_second + NFS_KERBTTL; 1076 nuidp->nu_timestamp = ktvout; 1077 nuidp->nu_nickname = nick; 1078 bcopy(key, nuidp->nu_key, sizeof (key)); 1079 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1080 nu_lru); 1081 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1082 nuidp, nu_hash); 1083 } 1084 } else 1085 nfsm_adv(nfsm_rndup(len)); 1086nfsmout: 1087 *mdp = md; 1088 *dposp = dpos; 1089 return (error); 1090} 1091 1092#ifndef NFS_NOSERVER 1093 1094/* 1095 * Derefence a server socket structure. If it has no more references and 1096 * is no longer valid, you can throw it away. 1097 */ 1098void 1099nfsrv_slpderef(slp) 1100 register struct nfssvc_sock *slp; 1101{ 1102 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 1103 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1104 free((caddr_t)slp, M_NFSSVC); 1105 } 1106} 1107 1108/* 1109 * Initialize the data structures for the server. 1110 * Handshake with any new nfsds starting up to avoid any chance of 1111 * corruption. 1112 */ 1113void 1114nfsrv_init(terminating) 1115 int terminating; 1116{ 1117 register struct nfssvc_sock *slp, *nslp; 1118 1119 if (nfssvc_sockhead_flag & SLP_INIT) 1120 panic("nfsd init"); 1121 nfssvc_sockhead_flag |= SLP_INIT; 1122 if (terminating) { 1123 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 1124 nslp = slp->ns_chain.tqe_next; 1125 if (slp->ns_flag & SLP_VALID) 1126 nfsrv_zapsock(slp); 1127 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1128 free((caddr_t)slp, M_NFSSVC); 1129 } 1130 nfsrv_cleancache(); /* And clear out server cache */ 1131 } else 1132 nfs_pub.np_valid = 0; 1133 1134 TAILQ_INIT(&nfssvc_sockhead); 1135 nfssvc_sockhead_flag &= ~SLP_INIT; 1136 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 1137 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 1138 wakeup((caddr_t)&nfssvc_sockhead); 1139 } 1140 1141 TAILQ_INIT(&nfsd_head); 1142 nfsd_head_flag &= ~NFSD_CHECKSLP; 1143 1144 nfs_udpsock = (struct nfssvc_sock *) 1145 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1146 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 1147 STAILQ_INIT(&nfs_udpsock->ns_rec); 1148 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 1149 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 1150 1151 nfs_cltpsock = (struct nfssvc_sock *) 1152 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1153 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 1154 STAILQ_INIT(&nfs_cltpsock->ns_rec); 1155 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 1156 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 1157} 1158 1159/* 1160 * Add entries to the server monitor log. 1161 */ 1162static void 1163nfsd_rt(sotype, nd, cacherep) 1164 int sotype; 1165 register struct nfsrv_descript *nd; 1166 int cacherep; 1167{ 1168 register struct drt *rt; 1169 1170 rt = &nfsdrt.drt[nfsdrt.pos]; 1171 if (cacherep == RC_DOIT) 1172 rt->flag = 0; 1173 else if (cacherep == RC_REPLY) 1174 rt->flag = DRT_CACHEREPLY; 1175 else 1176 rt->flag = DRT_CACHEDROP; 1177 if (sotype == SOCK_STREAM) 1178 rt->flag |= DRT_TCP; 1179 if (nd->nd_flag & ND_NQNFS) 1180 rt->flag |= DRT_NQNFS; 1181 else if (nd->nd_flag & ND_NFSV3) 1182 rt->flag |= DRT_NFSV3; 1183 rt->proc = nd->nd_procnum; 1184 if (nd->nd_nam->sa_family == AF_INET) 1185 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 1186 else 1187 rt->ipadr = INADDR_ANY; 1188 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 1189 getmicrotime(&rt->tstamp); 1190 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1191} 1192#endif /* NFS_NOSERVER */ 1193