nfs_nfsiod.c revision 12274
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 8.3 (Berkeley) 1/4/94 37 * $Id: nfs_syscalls.c,v 1.8 1995/10/29 15:33:11 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/file.h> 45#include <sys/stat.h> 46#include <sys/vnode.h> 47#include <sys/mount.h> 48#include <sys/proc.h> 49#include <sys/uio.h> 50#include <sys/malloc.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#include <sys/syslog.h> 59 60#include <netinet/in.h> 61#include <netinet/tcp.h> 62#ifdef ISO 63#include <netiso/iso.h> 64#endif 65#include <nfs/xdr_subs.h> 66#include <nfs/rpcv2.h> 67#include <nfs/nfsproto.h> 68#include <nfs/nfs.h> 69#include <nfs/nfsm_subs.h> 70#include <nfs/nfsrvcache.h> 71#include <nfs/nfsmount.h> 72#include <nfs/nfsnode.h> 73#include <nfs/nqnfs.h> 74#include <nfs/nfsrtt.h> 75 76void nfsrv_zapsock __P((struct nfssvc_sock *)); 77 78/* Global defs. */ 79extern int (*nfsrv3_procs[NFS_NPROCS])(); 80extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 81extern int nfs_numasync; 82extern time_t nqnfsstarttime; 83extern int nqsrv_writeslack; 84extern int nfsrtton; 85extern struct nfsstats nfsstats; 86extern int nfsrvw_procrastinate; 87struct nfssvc_sock *nfs_udpsock, *nfs_cltpsock; 88int nuidhash_max = NFS_MAXUIDHASH; 89static int nfs_numnfsd = 0; 90int nfsd_waiting = 0; 91static int notstarted = 1; 92static int modify_flag = 0; 93static struct nfsdrt nfsdrt; 94void nfsrv_cleancache(), nfsrv_rcv(), nfsrv_wakenfsd(), nfs_sndunlock(); 95static void nfsd_rt(); 96void nfsrv_slpderef(); 97 98#define TRUE 1 99#define FALSE 0 100 101static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 102/* 103 * NFS server system calls 104 * getfh() lives here too, but maybe should move to kern/vfs_syscalls.c 105 */ 106 107/* 108 * Get file handle system call 109 */ 110#ifndef _SYS_SYSPROTO_H_ 111struct getfh_args { 112 char *fname; 113 fhandle_t *fhp; 114}; 115#endif 116int 117getfh(p, uap, retval) 118 struct proc *p; 119 register struct getfh_args *uap; 120 int *retval; 121{ 122 register struct vnode *vp; 123 fhandle_t fh; 124 int error; 125 struct nameidata nd; 126 127 /* 128 * Must be super user 129 */ 130 error = suser(p->p_ucred, &p->p_acflag); 131 if(error) 132 return (error); 133 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, uap->fname, p); 134 error = namei(&nd); 135 if (error) 136 return (error); 137 vp = nd.ni_vp; 138 bzero((caddr_t)&fh, sizeof(fh)); 139 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 140 error = VFS_VPTOFH(vp, &fh.fh_fid); 141 vput(vp); 142 if (error) 143 return (error); 144 error = copyout((caddr_t)&fh, (caddr_t)uap->fhp, sizeof (fh)); 145 return (error); 146} 147 148/* 149 * Nfs server psuedo system call for the nfsd's 150 * Based on the flag value it either: 151 * - adds a socket to the selection list 152 * - remains in the kernel as an nfsd 153 * - remains in the kernel as an nfsiod 154 */ 155#ifndef _SYS_SYSPROTO_H_ 156struct nfssvc_args { 157 int flag; 158 caddr_t argp; 159}; 160#endif 161int 162nfssvc(p, uap, retval) 163 struct proc *p; 164 register struct nfssvc_args *uap; 165 int *retval; 166{ 167 struct nameidata nd; 168 struct file *fp; 169 struct mbuf *nam; 170 struct nfsd_args nfsdarg; 171 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 172 struct nfsd_cargs ncd; 173 struct nfsd *nfsd; 174 struct nfssvc_sock *slp; 175 struct nfsuid *nuidp; 176 struct nfsmount *nmp; 177 int error; 178 179 /* 180 * Must be super user 181 */ 182 error = suser(p->p_ucred, &p->p_acflag); 183 if(error) 184 return (error); 185 while (nfssvc_sockhead_flag & SLP_INIT) { 186 nfssvc_sockhead_flag |= SLP_WANTINIT; 187 (void) tsleep((caddr_t)&nfssvc_sockhead, PSOCK, "nfsd init", 0); 188 } 189 if (uap->flag & NFSSVC_BIOD) 190 error = nfssvc_iod(p); 191 else if (uap->flag & NFSSVC_MNTD) { 192 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 193 if (error) 194 return (error); 195 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 196 ncd.ncd_dirp, p); 197 error = namei(&nd); 198 if (error) 199 return (error); 200 if ((nd.ni_vp->v_flag & VROOT) == 0) 201 error = EINVAL; 202 nmp = VFSTONFS(nd.ni_vp->v_mount); 203 vput(nd.ni_vp); 204 if (error) 205 return (error); 206 if ((nmp->nm_flag & NFSMNT_MNTD) && 207 (uap->flag & NFSSVC_GOTAUTH) == 0) 208 return (0); 209 nmp->nm_flag |= NFSMNT_MNTD; 210 error = nqnfs_clientd(nmp, p->p_ucred, &ncd, uap->flag, 211 uap->argp, p); 212 } else if (uap->flag & NFSSVC_ADDSOCK) { 213 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 214 if (error) 215 return (error); 216 error = getsock(p->p_fd, nfsdarg.sock, &fp); 217 if (error) 218 return (error); 219 /* 220 * Get the client address for connected sockets. 221 */ 222 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 223 nam = (struct mbuf *)0; 224 else { 225 error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen, 226 MT_SONAME); 227 if (error) 228 return (error); 229 } 230 error = nfssvc_addsock(fp, nam); 231 } else { 232 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 233 if (error) 234 return (error); 235 if ((uap->flag & NFSSVC_AUTHIN) && ((nfsd = nsd->nsd_nfsd)) && 236 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 237 slp = nfsd->nfsd_slp; 238 239 /* 240 * First check to see if another nfsd has already 241 * added this credential. 242 */ 243 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 244 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 245 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 246 (!nfsd->nfsd_nd->nd_nam2 || 247 netaddr_match(NU_NETFAM(nuidp), 248 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 249 break; 250 } 251 if (nuidp) { 252 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 253 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 254 } else { 255 /* 256 * Nope, so we will. 257 */ 258 if (slp->ns_numuids < nuidhash_max) { 259 slp->ns_numuids++; 260 nuidp = (struct nfsuid *) 261 malloc(sizeof (struct nfsuid), M_NFSUID, 262 M_WAITOK); 263 } else 264 nuidp = (struct nfsuid *)0; 265 if ((slp->ns_flag & SLP_VALID) == 0) { 266 if (nuidp) 267 free((caddr_t)nuidp, M_NFSUID); 268 } else { 269 if (nuidp == (struct nfsuid *)0) { 270 nuidp = slp->ns_uidlruhead.tqh_first; 271 LIST_REMOVE(nuidp, nu_hash); 272 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 273 nu_lru); 274 if (nuidp->nu_flag & NU_NAM) 275 m_freem(nuidp->nu_nam); 276 } 277 nuidp->nu_flag = 0; 278 nuidp->nu_cr = nsd->nsd_cr; 279 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 280 nuidp->nu_cr.cr_ngroups = NGROUPS; 281 nuidp->nu_cr.cr_ref = 1; 282 nuidp->nu_timestamp = nsd->nsd_timestamp; 283 nuidp->nu_expire = time.tv_sec + nsd->nsd_ttl; 284 /* 285 * and save the session key in nu_key. 286 */ 287 bcopy(nsd->nsd_key, nuidp->nu_key, 288 sizeof (nsd->nsd_key)); 289 if (nfsd->nfsd_nd->nd_nam2) { 290 struct sockaddr_in *saddr; 291 292 saddr = mtod(nfsd->nfsd_nd->nd_nam2, 293 struct sockaddr_in *); 294 switch (saddr->sin_family) { 295 case AF_INET: 296 nuidp->nu_flag |= NU_INETADDR; 297 nuidp->nu_inetaddr = 298 saddr->sin_addr.s_addr; 299 break; 300 case AF_ISO: 301 default: 302 nuidp->nu_flag |= NU_NAM; 303 nuidp->nu_nam = m_copym( 304 nfsd->nfsd_nd->nd_nam2, 0, 305 M_COPYALL, M_WAIT); 306 break; 307 }; 308 } 309 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 310 nu_lru); 311 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 312 nuidp, nu_hash); 313 nfsrv_setcred(&nuidp->nu_cr, 314 &nfsd->nfsd_nd->nd_cr); 315 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 316 } 317 } 318 } 319 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 320 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 321 error = nfssvc_nfsd(nsd, uap->argp, p); 322 } 323 if (error == EINTR || error == ERESTART) 324 error = 0; 325 return (error); 326} 327 328/* 329 * Adds a socket to the list for servicing by nfsds. 330 */ 331int 332nfssvc_addsock(fp, mynam) 333 struct file *fp; 334 struct mbuf *mynam; 335{ 336 register struct mbuf *m; 337 register int siz; 338 register struct nfssvc_sock *slp; 339 register struct socket *so; 340 struct nfssvc_sock *tslp; 341 int error, s; 342 343 so = (struct socket *)fp->f_data; 344 tslp = (struct nfssvc_sock *)0; 345 /* 346 * Add it to the list, as required. 347 */ 348 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 349 tslp = nfs_udpsock; 350 if (tslp->ns_flag & SLP_VALID) { 351 m_freem(mynam); 352 return (EPERM); 353 } 354#ifdef ISO 355 } else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) { 356 tslp = nfs_cltpsock; 357 if (tslp->ns_flag & SLP_VALID) { 358 m_freem(mynam); 359 return (EPERM); 360 } 361#endif /* ISO */ 362 } 363 if (so->so_type == SOCK_STREAM) 364 siz = NFS_MAXPACKET + sizeof (u_long); 365 else 366 siz = NFS_MAXPACKET; 367 error = soreserve(so, siz, siz); 368 if (error) { 369 m_freem(mynam); 370 return (error); 371 } 372 373 /* 374 * Set protocol specific options { for now TCP only } and 375 * reserve some space. For datagram sockets, this can get called 376 * repeatedly for the same socket, but that isn't harmful. 377 */ 378 if (so->so_type == SOCK_STREAM) { 379 MGET(m, M_WAIT, MT_SOOPTS); 380 *mtod(m, int *) = 1; 381 m->m_len = sizeof(int); 382 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 383 } 384 if (so->so_proto->pr_domain->dom_family == AF_INET && 385 so->so_proto->pr_protocol == IPPROTO_TCP) { 386 MGET(m, M_WAIT, MT_SOOPTS); 387 *mtod(m, int *) = 1; 388 m->m_len = sizeof(int); 389 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 390 } 391 so->so_rcv.sb_flags &= ~SB_NOINTR; 392 so->so_rcv.sb_timeo = 0; 393 so->so_snd.sb_flags &= ~SB_NOINTR; 394 so->so_snd.sb_timeo = 0; 395 if (tslp) 396 slp = tslp; 397 else { 398 slp = (struct nfssvc_sock *) 399 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 400 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 401 TAILQ_INIT(&slp->ns_uidlruhead); 402 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 403 } 404 slp->ns_so = so; 405 slp->ns_nam = mynam; 406 fp->f_count++; 407 slp->ns_fp = fp; 408 s = splnet(); 409 so->so_upcallarg = (caddr_t)slp; 410 so->so_upcall = nfsrv_rcv; 411 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 412 nfsrv_wakenfsd(slp); 413 splx(s); 414 return (0); 415} 416 417/* 418 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 419 * until it is killed by a signal. 420 */ 421int 422nfssvc_nfsd(nsd, argp, p) 423 struct nfsd_srvargs *nsd; 424 caddr_t argp; 425 struct proc *p; 426{ 427 register struct mbuf *m; 428 register int siz; 429 register struct nfssvc_sock *slp; 430 register struct socket *so; 431 register int *solockp; 432 struct nfsd *nfsd = nsd->nsd_nfsd; 433 struct nfsrv_descript *nd = NULL; 434 struct mbuf *mreq; 435 int error = 0, cacherep, s, sotype, writes_todo; 436 u_quad_t cur_usec; 437 438#ifndef nolint 439 cacherep = RC_DOIT; 440 writes_todo = 0; 441#endif 442 s = splnet(); 443 if (nfsd == (struct nfsd *)0) { 444 nsd->nsd_nfsd = nfsd = (struct nfsd *) 445 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 446 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 447 nfsd->nfsd_procp = p; 448 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 449 nfs_numnfsd++; 450 } 451 /* 452 * Loop getting rpc requests until SIGKILL. 453 */ 454 for (;;) { 455 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 456 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 457 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 458 nfsd->nfsd_flag |= NFSD_WAITING; 459 nfsd_waiting++; 460 error = tsleep((caddr_t)nfsd, PSOCK | PCATCH, 461 "nfsd", 0); 462 nfsd_waiting--; 463 if (error) 464 goto done; 465 } 466 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 467 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 468 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 469 slp = slp->ns_chain.tqe_next) { 470 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 471 == (SLP_VALID | SLP_DOREC)) { 472 slp->ns_flag &= ~SLP_DOREC; 473 slp->ns_sref++; 474 nfsd->nfsd_slp = slp; 475 break; 476 } 477 } 478 if (slp == 0) 479 nfsd_head_flag &= ~NFSD_CHECKSLP; 480 } 481 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 482 continue; 483 if (slp->ns_flag & SLP_VALID) { 484 if (slp->ns_flag & SLP_DISCONN) 485 nfsrv_zapsock(slp); 486 else if (slp->ns_flag & SLP_NEEDQ) { 487 slp->ns_flag &= ~SLP_NEEDQ; 488 (void) nfs_sndlock(&slp->ns_solock, 489 (struct nfsreq *)0); 490 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 491 M_WAIT); 492 nfs_sndunlock(&slp->ns_solock); 493 } 494 error = nfsrv_dorec(slp, nfsd, &nd); 495 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 496 (u_quad_t)time.tv_usec; 497 if (error && slp->ns_tq.lh_first && 498 slp->ns_tq.lh_first->nd_time <= cur_usec) { 499 error = 0; 500 cacherep = RC_DOIT; 501 writes_todo = 1; 502 } else 503 writes_todo = 0; 504 nfsd->nfsd_flag |= NFSD_REQINPROG; 505 } 506 } else { 507 error = 0; 508 slp = nfsd->nfsd_slp; 509 } 510 if (error || (slp->ns_flag & SLP_VALID) == 0) { 511 if (nd) { 512 free((caddr_t)nd, M_NFSRVDESC); 513 nd = NULL; 514 } 515 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 516 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 517 nfsrv_slpderef(slp); 518 continue; 519 } 520 splx(s); 521 so = slp->ns_so; 522 sotype = so->so_type; 523 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 524 solockp = &slp->ns_solock; 525 else 526 solockp = (int *)0; 527 if (nd) { 528 nd->nd_starttime = time; 529 if (nd->nd_nam2) 530 nd->nd_nam = nd->nd_nam2; 531 else 532 nd->nd_nam = slp->ns_nam; 533 534 /* 535 * Check to see if authorization is needed. 536 */ 537 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 538 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 539 nsd->nsd_haddr = mtod(nd->nd_nam, 540 struct sockaddr_in *)->sin_addr.s_addr; 541 nsd->nsd_authlen = nfsd->nfsd_authlen; 542 nsd->nsd_verflen = nfsd->nfsd_verflen; 543 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 544 nfsd->nfsd_authlen) && 545 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 546 nfsd->nfsd_verflen) && 547 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 548 return (ENEEDAUTH); 549 cacherep = RC_DROPIT; 550 } else 551 cacherep = nfsrv_getcache(nd, slp, &mreq); 552 553 /* 554 * Check for just starting up for NQNFS and send 555 * fake "try again later" replies to the NQNFS clients. 556 */ 557 if (notstarted && nqnfsstarttime <= time.tv_sec) { 558 if (modify_flag) { 559 nqnfsstarttime = time.tv_sec + nqsrv_writeslack; 560 modify_flag = 0; 561 } else 562 notstarted = 0; 563 } 564 if (notstarted) { 565 if ((nd->nd_flag & ND_NQNFS) == 0) 566 cacherep = RC_DROPIT; 567 else if (nd->nd_procnum != NFSPROC_WRITE) { 568 nd->nd_procnum = NFSPROC_NOOP; 569 nd->nd_repstat = NQNFS_TRYLATER; 570 cacherep = RC_DOIT; 571 } else 572 modify_flag = 1; 573 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 574 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 575 nd->nd_procnum = NFSPROC_NOOP; 576 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 577 cacherep = RC_DOIT; 578 } 579 } 580 581 /* 582 * Loop to get all the write rpc relies that have been 583 * gathered together. 584 */ 585 do { 586 switch (cacherep) { 587 case RC_DOIT: 588 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 589 nfsrvw_procrastinate > 0 && !notstarted)) 590 error = nfsrv_writegather(&nd, slp, 591 nfsd->nfsd_procp, &mreq); 592 else 593 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 594 slp, nfsd->nfsd_procp, &mreq); 595 if (mreq == NULL) 596 break; 597 if (error) { 598 if (nd->nd_procnum != NQNFSPROC_VACATED) 599 nfsstats.srv_errs++; 600 nfsrv_updatecache(nd, FALSE, mreq); 601 if (nd->nd_nam2) 602 m_freem(nd->nd_nam2); 603 break; 604 } 605 nfsstats.srvrpccnt[nd->nd_procnum]++; 606 nfsrv_updatecache(nd, TRUE, mreq); 607 nd->nd_mrep = (struct mbuf *)0; 608 case RC_REPLY: 609 m = mreq; 610 siz = 0; 611 while (m) { 612 siz += m->m_len; 613 m = m->m_next; 614 } 615 if (siz <= 0 || siz > NFS_MAXPACKET) { 616 printf("mbuf siz=%d\n",siz); 617 panic("Bad nfs svc reply"); 618 } 619 m = mreq; 620 m->m_pkthdr.len = siz; 621 m->m_pkthdr.rcvif = (struct ifnet *)0; 622 /* 623 * For stream protocols, prepend a Sun RPC 624 * Record Mark. 625 */ 626 if (sotype == SOCK_STREAM) { 627 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 628 *mtod(m, u_long *) = htonl(0x80000000 | siz); 629 } 630 if (solockp) 631 (void) nfs_sndlock(solockp, (struct nfsreq *)0); 632 if (slp->ns_flag & SLP_VALID) 633 error = nfs_send(so, nd->nd_nam2, m, NULL); 634 else { 635 error = EPIPE; 636 m_freem(m); 637 } 638 if (nfsrtton) 639 nfsd_rt(sotype, nd, cacherep); 640 if (nd->nd_nam2) 641 MFREE(nd->nd_nam2, m); 642 if (nd->nd_mrep) 643 m_freem(nd->nd_mrep); 644 if (error == EPIPE) 645 nfsrv_zapsock(slp); 646 if (solockp) 647 nfs_sndunlock(solockp); 648 if (error == EINTR || error == ERESTART) { 649 free((caddr_t)nd, M_NFSRVDESC); 650 nfsrv_slpderef(slp); 651 s = splnet(); 652 goto done; 653 } 654 break; 655 case RC_DROPIT: 656 if (nfsrtton) 657 nfsd_rt(sotype, nd, cacherep); 658 m_freem(nd->nd_mrep); 659 m_freem(nd->nd_nam2); 660 break; 661 }; 662 if (nd) { 663 FREE((caddr_t)nd, M_NFSRVDESC); 664 nd = NULL; 665 } 666 667 /* 668 * Check to see if there are outstanding writes that 669 * need to be serviced. 670 */ 671 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 672 (u_quad_t)time.tv_usec; 673 s = splsoftclock(); 674 if (slp->ns_tq.lh_first && 675 slp->ns_tq.lh_first->nd_time <= cur_usec) { 676 cacherep = RC_DOIT; 677 writes_todo = 1; 678 } else 679 writes_todo = 0; 680 splx(s); 681 } while (writes_todo); 682 s = splnet(); 683 if (nfsrv_dorec(slp, nfsd, &nd)) { 684 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 685 nfsd->nfsd_slp = NULL; 686 nfsrv_slpderef(slp); 687 } 688 } 689done: 690 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 691 splx(s); 692 free((caddr_t)nfsd, M_NFSD); 693 nsd->nsd_nfsd = (struct nfsd *)0; 694 if (--nfs_numnfsd == 0) 695 nfsrv_init(TRUE); /* Reinitialize everything */ 696 return (error); 697} 698 699/* 700 * Asynchronous I/O daemons for client nfs. 701 * They do read-ahead and write-behind operations on the block I/O cache. 702 * Never returns unless it fails or gets killed. 703 */ 704int 705nfssvc_iod(p) 706 struct proc *p; 707{ 708 register struct buf *bp, *nbp; 709 register int i, myiod; 710 struct vnode *vp; 711 int error = 0, s; 712 713 /* 714 * Assign my position or return error if too many already running 715 */ 716 myiod = -1; 717 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 718 if (nfs_asyncdaemon[i] == 0) { 719 nfs_asyncdaemon[i]++; 720 myiod = i; 721 break; 722 } 723 if (myiod == -1) 724 return (EBUSY); 725 nfs_numasync++; 726 /* 727 * Just loop around doin our stuff until SIGKILL 728 */ 729 for (;;) { 730 while (nfs_bufq.tqh_first == NULL && error == 0) { 731 nfs_iodwant[myiod] = p; 732 error = tsleep((caddr_t)&nfs_iodwant[myiod], 733 PWAIT | PCATCH, "nfsidl", 0); 734 } 735 while ((bp = nfs_bufq.tqh_first) != NULL) { 736 /* Take one off the front of the list */ 737 TAILQ_REMOVE(&nfs_bufq, bp, b_freelist); 738 if (bp->b_flags & B_READ) 739 (void) nfs_doio(bp, bp->b_rcred, (struct proc *)0); 740 else do { 741 /* 742 * Look for a delayed write for the same vnode, so I can do 743 * it now. We must grab it before calling nfs_doio() to 744 * avoid any risk of the vnode getting vclean()'d while 745 * we are doing the write rpc. 746 */ 747 vp = bp->b_vp; 748 s = splbio(); 749 for (nbp = vp->v_dirtyblkhd.lh_first; nbp; 750 nbp = nbp->b_vnbufs.le_next) { 751 if ((nbp->b_flags & 752 (B_BUSY|B_DELWRI|B_NEEDCOMMIT|B_NOCACHE))!=B_DELWRI) 753 continue; 754 bremfree(nbp); 755 vfs_busy_pages(nbp, 1); 756 nbp->b_flags |= (B_BUSY|B_ASYNC); 757 break; 758 } 759 splx(s); 760 /* 761 * For the delayed write, do the first part of nfs_bwrite() 762 * up to, but not including nfs_strategy(). 763 */ 764 if (nbp) { 765 nbp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI); 766 reassignbuf(nbp, nbp->b_vp); 767 nbp->b_vp->v_numoutput++; 768 } 769 (void) nfs_doio(bp, bp->b_wcred, (struct proc *)0); 770 } while (bp = nbp); 771 } 772 if (error) { 773 nfs_asyncdaemon[myiod] = 0; 774 nfs_numasync--; 775 return (error); 776 } 777 } 778} 779 780/* 781 * Shut down a socket associated with an nfssvc_sock structure. 782 * Should be called with the send lock set, if required. 783 * The trick here is to increment the sref at the start, so that the nfsds 784 * will stop using it and clear ns_flag at the end so that it will not be 785 * reassigned during cleanup. 786 */ 787void 788nfsrv_zapsock(slp) 789 register struct nfssvc_sock *slp; 790{ 791 register struct nfsuid *nuidp, *nnuidp; 792 register struct nfsrv_descript *nwp, *nnwp; 793 struct socket *so; 794 struct file *fp; 795 struct mbuf *m; 796 int s; 797 798 slp->ns_flag &= ~SLP_ALLFLAGS; 799 fp = slp->ns_fp; 800 if (fp) { 801 slp->ns_fp = (struct file *)0; 802 so = slp->ns_so; 803 so->so_upcall = NULL; 804 soshutdown(so, 2); 805 closef(fp, (struct proc *)0); 806 if (slp->ns_nam) 807 MFREE(slp->ns_nam, m); 808 m_freem(slp->ns_raw); 809 m_freem(slp->ns_rec); 810 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 811 nuidp = nnuidp) { 812 nnuidp = nuidp->nu_lru.tqe_next; 813 LIST_REMOVE(nuidp, nu_hash); 814 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 815 if (nuidp->nu_flag & NU_NAM) 816 m_freem(nuidp->nu_nam); 817 free((caddr_t)nuidp, M_NFSUID); 818 } 819 s = splsoftclock(); 820 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 821 nnwp = nwp->nd_tq.le_next; 822 LIST_REMOVE(nwp, nd_tq); 823 free((caddr_t)nwp, M_NFSRVDESC); 824 } 825 LIST_INIT(&slp->ns_tq); 826 splx(s); 827 } 828} 829 830/* 831 * Get an authorization string for the uid by having the mount_nfs sitting 832 * on this mount point porpous out of the kernel and do it. 833 */ 834int 835nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 836 register struct nfsmount *nmp; 837 struct nfsreq *rep; 838 struct ucred *cred; 839 char **auth_str; 840 int *auth_len; 841 char *verf_str; 842 int *verf_len; 843 NFSKERBKEY_T key; /* return session key */ 844{ 845 int error = 0; 846 847 while ((nmp->nm_flag & NFSMNT_WAITAUTH) == 0) { 848 nmp->nm_flag |= NFSMNT_WANTAUTH; 849 (void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK, 850 "nfsauth1", 2 * hz); 851 error = nfs_sigintr(nmp, rep, rep->r_procp); 852 if (error) { 853 nmp->nm_flag &= ~NFSMNT_WANTAUTH; 854 return (error); 855 } 856 } 857 nmp->nm_flag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH); 858 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 859 nmp->nm_authlen = RPCAUTH_MAXSIZ; 860 nmp->nm_verfstr = verf_str; 861 nmp->nm_verflen = *verf_len; 862 nmp->nm_authuid = cred->cr_uid; 863 wakeup((caddr_t)&nmp->nm_authstr); 864 865 /* 866 * And wait for mount_nfs to do its stuff. 867 */ 868 while ((nmp->nm_flag & NFSMNT_HASAUTH) == 0 && error == 0) { 869 (void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK, 870 "nfsauth2", 2 * hz); 871 error = nfs_sigintr(nmp, rep, rep->r_procp); 872 } 873 if (nmp->nm_flag & NFSMNT_AUTHERR) { 874 nmp->nm_flag &= ~NFSMNT_AUTHERR; 875 error = EAUTH; 876 } 877 if (error) 878 free((caddr_t)*auth_str, M_TEMP); 879 else { 880 *auth_len = nmp->nm_authlen; 881 *verf_len = nmp->nm_verflen; 882 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 883 } 884 nmp->nm_flag &= ~NFSMNT_HASAUTH; 885 nmp->nm_flag |= NFSMNT_WAITAUTH; 886 if (nmp->nm_flag & NFSMNT_WANTAUTH) { 887 nmp->nm_flag &= ~NFSMNT_WANTAUTH; 888 wakeup((caddr_t)&nmp->nm_authtype); 889 } 890 return (error); 891} 892 893/* 894 * Get a nickname authenticator and verifier. 895 */ 896int 897nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 898 struct nfsmount *nmp; 899 struct ucred *cred; 900 char **auth_str; 901 int *auth_len; 902 char *verf_str; 903 int verf_len; 904{ 905 register struct nfsuid *nuidp; 906 register u_long *nickp, *verfp; 907 struct timeval ktvin, ktvout; 908 909#ifdef DIAGNOSTIC 910 if (verf_len < (4 * NFSX_UNSIGNED)) 911 panic("nfs_getnickauth verf too small"); 912#endif 913 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 914 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 915 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 916 break; 917 } 918 if (!nuidp || nuidp->nu_expire < time.tv_sec) 919 return (EACCES); 920 921 /* 922 * Move to the end of the lru list (end of lru == most recently used). 923 */ 924 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 925 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 926 927 nickp = (u_long *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 928 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 929 *nickp = txdr_unsigned(nuidp->nu_nickname); 930 *auth_str = (char *)nickp; 931 *auth_len = 2 * NFSX_UNSIGNED; 932 933 /* 934 * Now we must encrypt the verifier and package it up. 935 */ 936 verfp = (u_long *)verf_str; 937 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 938 if (time.tv_sec > nuidp->nu_timestamp.tv_sec || 939 (time.tv_sec == nuidp->nu_timestamp.tv_sec && 940 time.tv_usec > nuidp->nu_timestamp.tv_usec)) 941 nuidp->nu_timestamp = time; 942 else 943 nuidp->nu_timestamp.tv_usec++; 944 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 945 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 946 947 /* 948 * Now encrypt the timestamp verifier in ecb mode using the session 949 * key. 950 */ 951#ifdef NFSKERB 952 XXX 953#endif 954 955 *verfp++ = ktvout.tv_sec; 956 *verfp++ = ktvout.tv_usec; 957 *verfp = 0; 958 return (0); 959} 960 961/* 962 * Save the current nickname in a hash list entry on the mount point. 963 */ 964int 965nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 966 register struct nfsmount *nmp; 967 struct ucred *cred; 968 int len; 969 NFSKERBKEY_T key; 970 struct mbuf **mdp; 971 char **dposp; 972 struct mbuf *mrep; 973{ 974 register struct nfsuid *nuidp; 975 register u_long *tl; 976 register long t1; 977 struct mbuf *md = *mdp; 978 struct timeval ktvin, ktvout; 979 u_long nick; 980 char *dpos = *dposp, *cp2; 981 int deltasec, error = 0; 982 983 if (len == (3 * NFSX_UNSIGNED)) { 984 nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); 985 ktvin.tv_sec = *tl++; 986 ktvin.tv_usec = *tl++; 987 nick = fxdr_unsigned(u_long, *tl); 988 989 /* 990 * Decrypt the timestamp in ecb mode. 991 */ 992#ifdef NFSKERB 993 XXX 994#endif 995 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 996 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 997 deltasec = time.tv_sec - ktvout.tv_sec; 998 if (deltasec < 0) 999 deltasec = -deltasec; 1000 /* 1001 * If ok, add it to the hash list for the mount point. 1002 */ 1003 if (deltasec <= NFS_KERBCLOCKSKEW) { 1004 if (nmp->nm_numuids < nuidhash_max) { 1005 nmp->nm_numuids++; 1006 nuidp = (struct nfsuid *) 1007 malloc(sizeof (struct nfsuid), M_NFSUID, 1008 M_WAITOK); 1009 } else { 1010 nuidp = nmp->nm_uidlruhead.tqh_first; 1011 LIST_REMOVE(nuidp, nu_hash); 1012 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1013 nu_lru); 1014 } 1015 nuidp->nu_flag = 0; 1016 nuidp->nu_cr.cr_uid = cred->cr_uid; 1017 nuidp->nu_expire = time.tv_sec + NFS_KERBTTL; 1018 nuidp->nu_timestamp = ktvout; 1019 nuidp->nu_nickname = nick; 1020 bcopy(key, nuidp->nu_key, sizeof (key)); 1021 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1022 nu_lru); 1023 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1024 nuidp, nu_hash); 1025 } 1026 } else 1027 nfsm_adv(nfsm_rndup(len)); 1028nfsmout: 1029 *mdp = md; 1030 *dposp = dpos; 1031 return (error); 1032} 1033 1034/* 1035 * Derefence a server socket structure. If it has no more references and 1036 * is no longer valid, you can throw it away. 1037 */ 1038void 1039nfsrv_slpderef(slp) 1040 register struct nfssvc_sock *slp; 1041{ 1042 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 1043 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1044 free((caddr_t)slp, M_NFSSVC); 1045 } 1046} 1047 1048/* 1049 * Initialize the data structures for the server. 1050 * Handshake with any new nfsds starting up to avoid any chance of 1051 * corruption. 1052 */ 1053void 1054nfsrv_init(terminating) 1055 int terminating; 1056{ 1057 register struct nfssvc_sock *slp, *nslp; 1058 1059 if (nfssvc_sockhead_flag & SLP_INIT) 1060 panic("nfsd init"); 1061 nfssvc_sockhead_flag |= SLP_INIT; 1062 if (terminating) { 1063 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 1064 nslp = slp->ns_chain.tqe_next; 1065 if (slp->ns_flag & SLP_VALID) 1066 nfsrv_zapsock(slp); 1067 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1068 free((caddr_t)slp, M_NFSSVC); 1069 } 1070 nfsrv_cleancache(); /* And clear out server cache */ 1071 } 1072 1073 TAILQ_INIT(&nfssvc_sockhead); 1074 nfssvc_sockhead_flag &= ~SLP_INIT; 1075 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 1076 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 1077 wakeup((caddr_t)&nfssvc_sockhead); 1078 } 1079 1080 TAILQ_INIT(&nfsd_head); 1081 nfsd_head_flag &= ~NFSD_CHECKSLP; 1082 1083 nfs_udpsock = (struct nfssvc_sock *) 1084 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1085 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 1086 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 1087 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 1088 1089 nfs_cltpsock = (struct nfssvc_sock *) 1090 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1091 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 1092 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 1093 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 1094} 1095 1096/* 1097 * Add entries to the server monitor log. 1098 */ 1099static void 1100nfsd_rt(sotype, nd, cacherep) 1101 int sotype; 1102 register struct nfsrv_descript *nd; 1103 int cacherep; 1104{ 1105 register struct drt *rt; 1106 1107 rt = &nfsdrt.drt[nfsdrt.pos]; 1108 if (cacherep == RC_DOIT) 1109 rt->flag = 0; 1110 else if (cacherep == RC_REPLY) 1111 rt->flag = DRT_CACHEREPLY; 1112 else 1113 rt->flag = DRT_CACHEDROP; 1114 if (sotype == SOCK_STREAM) 1115 rt->flag |= DRT_TCP; 1116 if (nd->nd_flag & ND_NQNFS) 1117 rt->flag |= DRT_NQNFS; 1118 else if (nd->nd_flag & ND_NFSV3) 1119 rt->flag |= DRT_NFSV3; 1120 rt->proc = nd->nd_procnum; 1121 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 1122 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 1123 else 1124 rt->ipadr = INADDR_ANY; 1125 rt->resptime = ((time.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 1126 (time.tv_usec - nd->nd_starttime.tv_usec); 1127 rt->tstamp = time; 1128 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1129} 1130