nfs_nfsiod.c revision 15480
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 1.13 1996/01/13 23:27:58 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/filedesc.h> 46#include <sys/stat.h> 47#include <sys/vnode.h> 48#include <sys/mount.h> 49#include <sys/proc.h> 50#include <sys/uio.h> 51#include <sys/malloc.h> 52#include <sys/buf.h> 53#include <sys/mbuf.h> 54#include <sys/socket.h> 55#include <sys/socketvar.h> 56#include <sys/domain.h> 57#include <sys/protosw.h> 58#include <sys/namei.h> 59#include <sys/syslog.h> 60 61#include <netinet/in.h> 62#include <netinet/tcp.h> 63#ifdef ISO 64#include <netiso/iso.h> 65#endif 66#include <nfs/xdr_subs.h> 67#include <nfs/rpcv2.h> 68#include <nfs/nfsproto.h> 69#include <nfs/nfs.h> 70#include <nfs/nfsm_subs.h> 71#include <nfs/nfsrvcache.h> 72#include <nfs/nfsmount.h> 73#include <nfs/nfsnode.h> 74#include <nfs/nqnfs.h> 75#include <nfs/nfsrtt.h> 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 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON]; 83extern int nfs_numasync; 84extern time_t nqnfsstarttime; 85extern int nqsrv_writeslack; 86extern int nfsrtton; 87extern struct nfsstats nfsstats; 88extern int nfsrvw_procrastinate; 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]; 99 100 101#ifndef NFS_NOSERVER 102int nfsd_waiting = 0; 103static struct nfsdrt nfsdrt; 104static int nfs_numnfsd = 0; 105static int notstarted = 1; 106static int modify_flag = 0; 107static void nfsd_rt __P((int sotype, struct nfsrv_descript *nd, 108 int cacherep)); 109static int nfssvc_addsock __P((struct file *,struct mbuf *)); 110static int nfssvc_nfsd __P((struct nfsd_srvargs *,caddr_t,struct proc *)); 111/* 112 * NFS server system calls 113 * getfh() lives here too, but maybe should move to kern/vfs_syscalls.c 114 */ 115 116/* 117 * Get file handle system call 118 */ 119#ifndef _SYS_SYSPROTO_H_ 120struct getfh_args { 121 char *fname; 122 fhandle_t *fhp; 123}; 124#endif 125int 126getfh(p, uap, retval) 127 struct proc *p; 128 register struct getfh_args *uap; 129 int *retval; 130{ 131 register struct vnode *vp; 132 fhandle_t fh; 133 int error; 134 struct nameidata nd; 135 136 /* 137 * Must be super user 138 */ 139 error = suser(p->p_ucred, &p->p_acflag); 140 if(error) 141 return (error); 142 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, uap->fname, p); 143 error = namei(&nd); 144 if (error) 145 return (error); 146 vp = nd.ni_vp; 147 bzero((caddr_t)&fh, sizeof(fh)); 148 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; 149 error = VFS_VPTOFH(vp, &fh.fh_fid); 150 vput(vp); 151 if (error) 152 return (error); 153 error = copyout((caddr_t)&fh, (caddr_t)uap->fhp, sizeof (fh)); 154 return (error); 155} 156 157#endif /* NFS_NOSERVER */ 158/* 159 * Nfs server psuedo system call for the nfsd's 160 * Based on the flag value it either: 161 * - adds a socket to the selection list 162 * - remains in the kernel as an nfsd 163 * - remains in the kernel as an nfsiod 164 */ 165#ifndef _SYS_SYSPROTO_H_ 166struct nfssvc_args { 167 int flag; 168 caddr_t argp; 169}; 170#endif 171int 172nfssvc(p, uap, retval) 173 struct proc *p; 174 register struct nfssvc_args *uap; 175 int *retval; 176{ 177#ifndef NFS_NOSERVER 178 struct nameidata nd; 179 struct file *fp; 180 struct mbuf *nam; 181 struct nfsd_args nfsdarg; 182 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 183 struct nfsd_cargs ncd; 184 struct nfsd *nfsd; 185 struct nfssvc_sock *slp; 186 struct nfsuid *nuidp; 187 struct nfsmount *nmp; 188#endif /* NFS_NOSERVER */ 189 int error; 190 191 /* 192 * Must be super user 193 */ 194 error = suser(p->p_ucred, &p->p_acflag); 195 if(error) 196 return (error); 197 while (nfssvc_sockhead_flag & SLP_INIT) { 198 nfssvc_sockhead_flag |= SLP_WANTINIT; 199 (void) tsleep((caddr_t)&nfssvc_sockhead, PSOCK, "nfsd init", 0); 200 } 201 if (uap->flag & NFSSVC_BIOD) 202 error = nfssvc_iod(p); 203#ifdef NFS_NOSERVER 204 else 205 error = ENXIO; 206#else /* !NFS_NOSERVER */ 207 else if (uap->flag & NFSSVC_MNTD) { 208 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 209 if (error) 210 return (error); 211 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, 212 ncd.ncd_dirp, p); 213 error = namei(&nd); 214 if (error) 215 return (error); 216 if ((nd.ni_vp->v_flag & VROOT) == 0) 217 error = EINVAL; 218 nmp = VFSTONFS(nd.ni_vp->v_mount); 219 vput(nd.ni_vp); 220 if (error) 221 return (error); 222 if ((nmp->nm_flag & NFSMNT_MNTD) && 223 (uap->flag & NFSSVC_GOTAUTH) == 0) 224 return (0); 225 nmp->nm_flag |= NFSMNT_MNTD; 226 error = nqnfs_clientd(nmp, p->p_ucred, &ncd, uap->flag, 227 uap->argp, p); 228 } else if (uap->flag & NFSSVC_ADDSOCK) { 229 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 230 if (error) 231 return (error); 232 error = getsock(p->p_fd, nfsdarg.sock, &fp); 233 if (error) 234 return (error); 235 /* 236 * Get the client address for connected sockets. 237 */ 238 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 239 nam = (struct mbuf *)0; 240 else { 241 error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen, 242 MT_SONAME); 243 if (error) 244 return (error); 245 } 246 error = nfssvc_addsock(fp, nam); 247 } else { 248 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 249 if (error) 250 return (error); 251 if ((uap->flag & NFSSVC_AUTHIN) && ((nfsd = nsd->nsd_nfsd)) && 252 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 253 slp = nfsd->nfsd_slp; 254 255 /* 256 * First check to see if another nfsd has already 257 * added this credential. 258 */ 259 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 260 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 261 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 262 (!nfsd->nfsd_nd->nd_nam2 || 263 netaddr_match(NU_NETFAM(nuidp), 264 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 265 break; 266 } 267 if (nuidp) { 268 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 269 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 270 } else { 271 /* 272 * Nope, so we will. 273 */ 274 if (slp->ns_numuids < nuidhash_max) { 275 slp->ns_numuids++; 276 nuidp = (struct nfsuid *) 277 malloc(sizeof (struct nfsuid), M_NFSUID, 278 M_WAITOK); 279 } else 280 nuidp = (struct nfsuid *)0; 281 if ((slp->ns_flag & SLP_VALID) == 0) { 282 if (nuidp) 283 free((caddr_t)nuidp, M_NFSUID); 284 } else { 285 if (nuidp == (struct nfsuid *)0) { 286 nuidp = slp->ns_uidlruhead.tqh_first; 287 LIST_REMOVE(nuidp, nu_hash); 288 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 289 nu_lru); 290 if (nuidp->nu_flag & NU_NAM) 291 m_freem(nuidp->nu_nam); 292 } 293 nuidp->nu_flag = 0; 294 nuidp->nu_cr = nsd->nsd_cr; 295 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 296 nuidp->nu_cr.cr_ngroups = NGROUPS; 297 nuidp->nu_cr.cr_ref = 1; 298 nuidp->nu_timestamp = nsd->nsd_timestamp; 299 nuidp->nu_expire = time.tv_sec + nsd->nsd_ttl; 300 /* 301 * and save the session key in nu_key. 302 */ 303 bcopy(nsd->nsd_key, nuidp->nu_key, 304 sizeof (nsd->nsd_key)); 305 if (nfsd->nfsd_nd->nd_nam2) { 306 struct sockaddr_in *saddr; 307 308 saddr = mtod(nfsd->nfsd_nd->nd_nam2, 309 struct sockaddr_in *); 310 switch (saddr->sin_family) { 311 case AF_INET: 312 nuidp->nu_flag |= NU_INETADDR; 313 nuidp->nu_inetaddr = 314 saddr->sin_addr.s_addr; 315 break; 316 case AF_ISO: 317 default: 318 nuidp->nu_flag |= NU_NAM; 319 nuidp->nu_nam = m_copym( 320 nfsd->nfsd_nd->nd_nam2, 0, 321 M_COPYALL, M_WAIT); 322 break; 323 }; 324 } 325 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 326 nu_lru); 327 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 328 nuidp, nu_hash); 329 nfsrv_setcred(&nuidp->nu_cr, 330 &nfsd->nfsd_nd->nd_cr); 331 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 332 } 333 } 334 } 335 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 336 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 337 error = nfssvc_nfsd(nsd, uap->argp, p); 338 } 339#endif /* NFS_NOSERVER */ 340 if (error == EINTR || error == ERESTART) 341 error = 0; 342 return (error); 343} 344 345#ifndef NFS_NOSERVER 346/* 347 * Adds a socket to the list for servicing by nfsds. 348 */ 349static int 350nfssvc_addsock(fp, mynam) 351 struct file *fp; 352 struct mbuf *mynam; 353{ 354 register struct mbuf *m; 355 register int siz; 356 register struct nfssvc_sock *slp; 357 register struct socket *so; 358 struct nfssvc_sock *tslp; 359 int error, s; 360 361 so = (struct socket *)fp->f_data; 362 tslp = (struct nfssvc_sock *)0; 363 /* 364 * Add it to the list, as required. 365 */ 366 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 367 tslp = nfs_udpsock; 368 if (tslp->ns_flag & SLP_VALID) { 369 m_freem(mynam); 370 return (EPERM); 371 } 372#ifdef ISO 373 } else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) { 374 tslp = nfs_cltpsock; 375 if (tslp->ns_flag & SLP_VALID) { 376 m_freem(mynam); 377 return (EPERM); 378 } 379#endif /* ISO */ 380 } 381 if (so->so_type == SOCK_STREAM) 382 siz = NFS_MAXPACKET + sizeof (u_long); 383 else 384 siz = NFS_MAXPACKET; 385 error = soreserve(so, siz, siz); 386 if (error) { 387 m_freem(mynam); 388 return (error); 389 } 390 391 /* 392 * Set protocol specific options { for now TCP only } and 393 * reserve some space. For datagram sockets, this can get called 394 * repeatedly for the same socket, but that isn't harmful. 395 */ 396 if (so->so_type == SOCK_STREAM) { 397 MGET(m, M_WAIT, MT_SOOPTS); 398 *mtod(m, int *) = 1; 399 m->m_len = sizeof(int); 400 sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); 401 } 402 if (so->so_proto->pr_domain->dom_family == AF_INET && 403 so->so_proto->pr_protocol == IPPROTO_TCP) { 404 MGET(m, M_WAIT, MT_SOOPTS); 405 *mtod(m, int *) = 1; 406 m->m_len = sizeof(int); 407 sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); 408 } 409 so->so_rcv.sb_flags &= ~SB_NOINTR; 410 so->so_rcv.sb_timeo = 0; 411 so->so_snd.sb_flags &= ~SB_NOINTR; 412 so->so_snd.sb_timeo = 0; 413 if (tslp) 414 slp = tslp; 415 else { 416 slp = (struct nfssvc_sock *) 417 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 418 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 419 TAILQ_INIT(&slp->ns_uidlruhead); 420 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 421 } 422 slp->ns_so = so; 423 slp->ns_nam = mynam; 424 fp->f_count++; 425 slp->ns_fp = fp; 426 s = splnet(); 427 so->so_upcallarg = (caddr_t)slp; 428 so->so_upcall = nfsrv_rcv; 429 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 430 nfsrv_wakenfsd(slp); 431 splx(s); 432 return (0); 433} 434 435/* 436 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 437 * until it is killed by a signal. 438 */ 439static int 440nfssvc_nfsd(nsd, argp, p) 441 struct nfsd_srvargs *nsd; 442 caddr_t argp; 443 struct proc *p; 444{ 445 register struct mbuf *m; 446 register int siz; 447 register struct nfssvc_sock *slp; 448 register struct socket *so; 449 register int *solockp; 450 struct nfsd *nfsd = nsd->nsd_nfsd; 451 struct nfsrv_descript *nd = NULL; 452 struct mbuf *mreq; 453 int error = 0, cacherep, s, sotype, writes_todo; 454 u_quad_t cur_usec; 455 456#ifndef nolint 457 cacherep = RC_DOIT; 458 writes_todo = 0; 459#endif 460 s = splnet(); 461 if (nfsd == (struct nfsd *)0) { 462 nsd->nsd_nfsd = nfsd = (struct nfsd *) 463 malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK); 464 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 465 nfsd->nfsd_procp = p; 466 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 467 nfs_numnfsd++; 468 } 469 /* 470 * Loop getting rpc requests until SIGKILL. 471 */ 472 for (;;) { 473 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 474 while (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 475 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 476 nfsd->nfsd_flag |= NFSD_WAITING; 477 nfsd_waiting++; 478 error = tsleep((caddr_t)nfsd, PSOCK | PCATCH, 479 "nfsd", 0); 480 nfsd_waiting--; 481 if (error) 482 goto done; 483 } 484 if (nfsd->nfsd_slp == (struct nfssvc_sock *)0 && 485 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 486 for (slp = nfssvc_sockhead.tqh_first; slp != 0; 487 slp = slp->ns_chain.tqe_next) { 488 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 489 == (SLP_VALID | SLP_DOREC)) { 490 slp->ns_flag &= ~SLP_DOREC; 491 slp->ns_sref++; 492 nfsd->nfsd_slp = slp; 493 break; 494 } 495 } 496 if (slp == 0) 497 nfsd_head_flag &= ~NFSD_CHECKSLP; 498 } 499 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 500 continue; 501 if (slp->ns_flag & SLP_VALID) { 502 if (slp->ns_flag & SLP_DISCONN) 503 nfsrv_zapsock(slp); 504 else if (slp->ns_flag & SLP_NEEDQ) { 505 slp->ns_flag &= ~SLP_NEEDQ; 506 (void) nfs_sndlock(&slp->ns_solock, 507 (struct nfsreq *)0); 508 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 509 M_WAIT); 510 nfs_sndunlock(&slp->ns_solock); 511 } 512 error = nfsrv_dorec(slp, nfsd, &nd); 513 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 514 (u_quad_t)time.tv_usec; 515 if (error && slp->ns_tq.lh_first && 516 slp->ns_tq.lh_first->nd_time <= cur_usec) { 517 error = 0; 518 cacherep = RC_DOIT; 519 writes_todo = 1; 520 } else 521 writes_todo = 0; 522 nfsd->nfsd_flag |= NFSD_REQINPROG; 523 } 524 } else { 525 error = 0; 526 slp = nfsd->nfsd_slp; 527 } 528 if (error || (slp->ns_flag & SLP_VALID) == 0) { 529 if (nd) { 530 free((caddr_t)nd, M_NFSRVDESC); 531 nd = NULL; 532 } 533 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 534 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 535 nfsrv_slpderef(slp); 536 continue; 537 } 538 splx(s); 539 so = slp->ns_so; 540 sotype = so->so_type; 541 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 542 solockp = &slp->ns_solock; 543 else 544 solockp = (int *)0; 545 if (nd) { 546 nd->nd_starttime = time; 547 if (nd->nd_nam2) 548 nd->nd_nam = nd->nd_nam2; 549 else 550 nd->nd_nam = slp->ns_nam; 551 552 /* 553 * Check to see if authorization is needed. 554 */ 555 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 556 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 557 nsd->nsd_haddr = mtod(nd->nd_nam, 558 struct sockaddr_in *)->sin_addr.s_addr; 559 nsd->nsd_authlen = nfsd->nfsd_authlen; 560 nsd->nsd_verflen = nfsd->nfsd_verflen; 561 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 562 nfsd->nfsd_authlen) && 563 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 564 nfsd->nfsd_verflen) && 565 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 566 return (ENEEDAUTH); 567 cacherep = RC_DROPIT; 568 } else 569 cacherep = nfsrv_getcache(nd, slp, &mreq); 570 571 /* 572 * Check for just starting up for NQNFS and send 573 * fake "try again later" replies to the NQNFS clients. 574 */ 575 if (notstarted && nqnfsstarttime <= time.tv_sec) { 576 if (modify_flag) { 577 nqnfsstarttime = time.tv_sec + nqsrv_writeslack; 578 modify_flag = 0; 579 } else 580 notstarted = 0; 581 } 582 if (notstarted) { 583 if ((nd->nd_flag & ND_NQNFS) == 0) 584 cacherep = RC_DROPIT; 585 else if (nd->nd_procnum != NFSPROC_WRITE) { 586 nd->nd_procnum = NFSPROC_NOOP; 587 nd->nd_repstat = NQNFS_TRYLATER; 588 cacherep = RC_DOIT; 589 } else 590 modify_flag = 1; 591 } else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 592 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 593 nd->nd_procnum = NFSPROC_NOOP; 594 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 595 cacherep = RC_DOIT; 596 } 597 } 598 599 /* 600 * Loop to get all the write rpc relies that have been 601 * gathered together. 602 */ 603 do { 604 switch (cacherep) { 605 case RC_DOIT: 606 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 607 nfsrvw_procrastinate > 0 && !notstarted)) 608 error = nfsrv_writegather(&nd, slp, 609 nfsd->nfsd_procp, &mreq); 610 else 611 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 612 slp, nfsd->nfsd_procp, &mreq); 613 if (mreq == NULL) 614 break; 615 if (error) { 616 if (nd->nd_procnum != NQNFSPROC_VACATED) 617 nfsstats.srv_errs++; 618 nfsrv_updatecache(nd, FALSE, mreq); 619 if (nd->nd_nam2) 620 m_freem(nd->nd_nam2); 621 break; 622 } 623 nfsstats.srvrpccnt[nd->nd_procnum]++; 624 nfsrv_updatecache(nd, TRUE, mreq); 625 nd->nd_mrep = (struct mbuf *)0; 626 case RC_REPLY: 627 m = mreq; 628 siz = 0; 629 while (m) { 630 siz += m->m_len; 631 m = m->m_next; 632 } 633 if (siz <= 0 || siz > NFS_MAXPACKET) { 634 printf("mbuf siz=%d\n",siz); 635 panic("Bad nfs svc reply"); 636 } 637 m = mreq; 638 m->m_pkthdr.len = siz; 639 m->m_pkthdr.rcvif = (struct ifnet *)0; 640 /* 641 * For stream protocols, prepend a Sun RPC 642 * Record Mark. 643 */ 644 if (sotype == SOCK_STREAM) { 645 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 646 *mtod(m, u_long *) = htonl(0x80000000 | siz); 647 } 648 if (solockp) 649 (void) nfs_sndlock(solockp, (struct nfsreq *)0); 650 if (slp->ns_flag & SLP_VALID) 651 error = nfs_send(so, nd->nd_nam2, m, NULL); 652 else { 653 error = EPIPE; 654 m_freem(m); 655 } 656 if (nfsrtton) 657 nfsd_rt(sotype, nd, cacherep); 658 if (nd->nd_nam2) 659 MFREE(nd->nd_nam2, m); 660 if (nd->nd_mrep) 661 m_freem(nd->nd_mrep); 662 if (error == EPIPE) 663 nfsrv_zapsock(slp); 664 if (solockp) 665 nfs_sndunlock(solockp); 666 if (error == EINTR || error == ERESTART) { 667 free((caddr_t)nd, M_NFSRVDESC); 668 nfsrv_slpderef(slp); 669 s = splnet(); 670 goto done; 671 } 672 break; 673 case RC_DROPIT: 674 if (nfsrtton) 675 nfsd_rt(sotype, nd, cacherep); 676 m_freem(nd->nd_mrep); 677 m_freem(nd->nd_nam2); 678 break; 679 }; 680 if (nd) { 681 FREE((caddr_t)nd, M_NFSRVDESC); 682 nd = NULL; 683 } 684 685 /* 686 * Check to see if there are outstanding writes that 687 * need to be serviced. 688 */ 689 cur_usec = (u_quad_t)time.tv_sec * 1000000 + 690 (u_quad_t)time.tv_usec; 691 s = splsoftclock(); 692 if (slp->ns_tq.lh_first && 693 slp->ns_tq.lh_first->nd_time <= cur_usec) { 694 cacherep = RC_DOIT; 695 writes_todo = 1; 696 } else 697 writes_todo = 0; 698 splx(s); 699 } while (writes_todo); 700 s = splnet(); 701 if (nfsrv_dorec(slp, nfsd, &nd)) { 702 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 703 nfsd->nfsd_slp = NULL; 704 nfsrv_slpderef(slp); 705 } 706 } 707done: 708 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 709 splx(s); 710 free((caddr_t)nfsd, M_NFSD); 711 nsd->nsd_nfsd = (struct nfsd *)0; 712 if (--nfs_numnfsd == 0) 713 nfsrv_init(TRUE); /* Reinitialize everything */ 714 return (error); 715} 716#endif /* NFS_NOSERVER */ 717 718/* 719 * Asynchronous I/O daemons for client nfs. 720 * They do read-ahead and write-behind operations on the block I/O cache. 721 * Never returns unless it fails or gets killed. 722 */ 723static int 724nfssvc_iod(p) 725 struct proc *p; 726{ 727 register struct buf *bp, *nbp; 728 register int i, myiod; 729 struct vnode *vp; 730 int error = 0, s; 731 732 /* 733 * Assign my position or return error if too many already running 734 */ 735 myiod = -1; 736 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 737 if (nfs_asyncdaemon[i] == 0) { 738 nfs_asyncdaemon[i]++; 739 myiod = i; 740 break; 741 } 742 if (myiod == -1) 743 return (EBUSY); 744 nfs_numasync++; 745 /* 746 * Just loop around doin our stuff until SIGKILL 747 */ 748 for (;;) { 749 while (nfs_bufq.tqh_first == NULL && error == 0) { 750 nfs_iodwant[myiod] = p; 751 error = tsleep((caddr_t)&nfs_iodwant[myiod], 752 PWAIT | PCATCH, "nfsidl", 0); 753 } 754 while ((bp = nfs_bufq.tqh_first) != NULL) { 755 /* Take one off the front of the list */ 756 TAILQ_REMOVE(&nfs_bufq, bp, b_freelist); 757 if (bp->b_flags & B_READ) 758 (void) nfs_doio(bp, bp->b_rcred, (struct proc *)0); 759 else do { 760 /* 761 * Look for a delayed write for the same vnode, so I can do 762 * it now. We must grab it before calling nfs_doio() to 763 * avoid any risk of the vnode getting vclean()'d while 764 * we are doing the write rpc. 765 */ 766 vp = bp->b_vp; 767 s = splbio(); 768 for (nbp = vp->v_dirtyblkhd.lh_first; nbp; 769 nbp = nbp->b_vnbufs.le_next) { 770 if ((nbp->b_flags & 771 (B_BUSY|B_DELWRI|B_NEEDCOMMIT|B_NOCACHE))!=B_DELWRI) 772 continue; 773 bremfree(nbp); 774 vfs_busy_pages(nbp, 1); 775 nbp->b_flags |= (B_BUSY|B_ASYNC); 776 break; 777 } 778 splx(s); 779 /* 780 * For the delayed write, do the first part of nfs_bwrite() 781 * up to, but not including nfs_strategy(). 782 */ 783 if (nbp) { 784 nbp->b_flags &= ~(B_READ|B_DONE|B_ERROR|B_DELWRI); 785 reassignbuf(nbp, nbp->b_vp); 786 nbp->b_vp->v_numoutput++; 787 } 788 (void) nfs_doio(bp, bp->b_wcred, (struct proc *)0); 789 } while (bp = nbp); 790 } 791 if (error) { 792 nfs_asyncdaemon[myiod] = 0; 793 nfs_numasync--; 794 return (error); 795 } 796 } 797} 798 799/* 800 * Shut down a socket associated with an nfssvc_sock structure. 801 * Should be called with the send lock set, if required. 802 * The trick here is to increment the sref at the start, so that the nfsds 803 * will stop using it and clear ns_flag at the end so that it will not be 804 * reassigned during cleanup. 805 */ 806static void 807nfsrv_zapsock(slp) 808 register struct nfssvc_sock *slp; 809{ 810 register struct nfsuid *nuidp, *nnuidp; 811 register struct nfsrv_descript *nwp, *nnwp; 812 struct socket *so; 813 struct file *fp; 814 struct mbuf *m; 815 int s; 816 817 slp->ns_flag &= ~SLP_ALLFLAGS; 818 fp = slp->ns_fp; 819 if (fp) { 820 slp->ns_fp = (struct file *)0; 821 so = slp->ns_so; 822 so->so_upcall = NULL; 823 soshutdown(so, 2); 824 closef(fp, (struct proc *)0); 825 if (slp->ns_nam) 826 MFREE(slp->ns_nam, m); 827 m_freem(slp->ns_raw); 828 m_freem(slp->ns_rec); 829 for (nuidp = slp->ns_uidlruhead.tqh_first; nuidp != 0; 830 nuidp = nnuidp) { 831 nnuidp = nuidp->nu_lru.tqe_next; 832 LIST_REMOVE(nuidp, nu_hash); 833 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 834 if (nuidp->nu_flag & NU_NAM) 835 m_freem(nuidp->nu_nam); 836 free((caddr_t)nuidp, M_NFSUID); 837 } 838 s = splsoftclock(); 839 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 840 nnwp = nwp->nd_tq.le_next; 841 LIST_REMOVE(nwp, nd_tq); 842 free((caddr_t)nwp, M_NFSRVDESC); 843 } 844 LIST_INIT(&slp->ns_tq); 845 splx(s); 846 } 847} 848 849/* 850 * Get an authorization string for the uid by having the mount_nfs sitting 851 * on this mount point porpous out of the kernel and do it. 852 */ 853int 854nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key) 855 register struct nfsmount *nmp; 856 struct nfsreq *rep; 857 struct ucred *cred; 858 char **auth_str; 859 int *auth_len; 860 char *verf_str; 861 int *verf_len; 862 NFSKERBKEY_T key; /* return session key */ 863{ 864 int error = 0; 865 866 while ((nmp->nm_flag & NFSMNT_WAITAUTH) == 0) { 867 nmp->nm_flag |= NFSMNT_WANTAUTH; 868 (void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK, 869 "nfsauth1", 2 * hz); 870 error = nfs_sigintr(nmp, rep, rep->r_procp); 871 if (error) { 872 nmp->nm_flag &= ~NFSMNT_WANTAUTH; 873 return (error); 874 } 875 } 876 nmp->nm_flag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH); 877 nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 878 nmp->nm_authlen = RPCAUTH_MAXSIZ; 879 nmp->nm_verfstr = verf_str; 880 nmp->nm_verflen = *verf_len; 881 nmp->nm_authuid = cred->cr_uid; 882 wakeup((caddr_t)&nmp->nm_authstr); 883 884 /* 885 * And wait for mount_nfs to do its stuff. 886 */ 887 while ((nmp->nm_flag & NFSMNT_HASAUTH) == 0 && error == 0) { 888 (void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK, 889 "nfsauth2", 2 * hz); 890 error = nfs_sigintr(nmp, rep, rep->r_procp); 891 } 892 if (nmp->nm_flag & NFSMNT_AUTHERR) { 893 nmp->nm_flag &= ~NFSMNT_AUTHERR; 894 error = EAUTH; 895 } 896 if (error) 897 free((caddr_t)*auth_str, M_TEMP); 898 else { 899 *auth_len = nmp->nm_authlen; 900 *verf_len = nmp->nm_verflen; 901 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 902 } 903 nmp->nm_flag &= ~NFSMNT_HASAUTH; 904 nmp->nm_flag |= NFSMNT_WAITAUTH; 905 if (nmp->nm_flag & NFSMNT_WANTAUTH) { 906 nmp->nm_flag &= ~NFSMNT_WANTAUTH; 907 wakeup((caddr_t)&nmp->nm_authtype); 908 } 909 return (error); 910} 911 912/* 913 * Get a nickname authenticator and verifier. 914 */ 915int 916nfs_getnickauth(nmp, cred, auth_str, auth_len, verf_str, verf_len) 917 struct nfsmount *nmp; 918 struct ucred *cred; 919 char **auth_str; 920 int *auth_len; 921 char *verf_str; 922 int verf_len; 923{ 924 register struct nfsuid *nuidp; 925 register u_long *nickp, *verfp; 926 struct timeval ktvin, ktvout; 927 928#ifdef DIAGNOSTIC 929 if (verf_len < (4 * NFSX_UNSIGNED)) 930 panic("nfs_getnickauth verf too small"); 931#endif 932 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 933 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 934 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 935 break; 936 } 937 if (!nuidp || nuidp->nu_expire < time.tv_sec) 938 return (EACCES); 939 940 /* 941 * Move to the end of the lru list (end of lru == most recently used). 942 */ 943 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 944 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 945 946 nickp = (u_long *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 947 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 948 *nickp = txdr_unsigned(nuidp->nu_nickname); 949 *auth_str = (char *)nickp; 950 *auth_len = 2 * NFSX_UNSIGNED; 951 952 /* 953 * Now we must encrypt the verifier and package it up. 954 */ 955 verfp = (u_long *)verf_str; 956 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 957 if (time.tv_sec > nuidp->nu_timestamp.tv_sec || 958 (time.tv_sec == nuidp->nu_timestamp.tv_sec && 959 time.tv_usec > nuidp->nu_timestamp.tv_usec)) 960 nuidp->nu_timestamp = time; 961 else 962 nuidp->nu_timestamp.tv_usec++; 963 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 964 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 965 966 /* 967 * Now encrypt the timestamp verifier in ecb mode using the session 968 * key. 969 */ 970#ifdef NFSKERB 971 XXX 972#endif 973 974 *verfp++ = ktvout.tv_sec; 975 *verfp++ = ktvout.tv_usec; 976 *verfp = 0; 977 return (0); 978} 979 980/* 981 * Save the current nickname in a hash list entry on the mount point. 982 */ 983int 984nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep) 985 register struct nfsmount *nmp; 986 struct ucred *cred; 987 int len; 988 NFSKERBKEY_T key; 989 struct mbuf **mdp; 990 char **dposp; 991 struct mbuf *mrep; 992{ 993 register struct nfsuid *nuidp; 994 register u_long *tl; 995 register long t1; 996 struct mbuf *md = *mdp; 997 struct timeval ktvin, ktvout; 998 u_long nick; 999 char *dpos = *dposp, *cp2; 1000 int deltasec, error = 0; 1001 1002 if (len == (3 * NFSX_UNSIGNED)) { 1003 nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); 1004 ktvin.tv_sec = *tl++; 1005 ktvin.tv_usec = *tl++; 1006 nick = fxdr_unsigned(u_long, *tl); 1007 1008 /* 1009 * Decrypt the timestamp in ecb mode. 1010 */ 1011#ifdef NFSKERB 1012 XXX 1013#endif 1014 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1015 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1016 deltasec = time.tv_sec - ktvout.tv_sec; 1017 if (deltasec < 0) 1018 deltasec = -deltasec; 1019 /* 1020 * If ok, add it to the hash list for the mount point. 1021 */ 1022 if (deltasec <= NFS_KERBCLOCKSKEW) { 1023 if (nmp->nm_numuids < nuidhash_max) { 1024 nmp->nm_numuids++; 1025 nuidp = (struct nfsuid *) 1026 malloc(sizeof (struct nfsuid), M_NFSUID, 1027 M_WAITOK); 1028 } else { 1029 nuidp = nmp->nm_uidlruhead.tqh_first; 1030 LIST_REMOVE(nuidp, nu_hash); 1031 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1032 nu_lru); 1033 } 1034 nuidp->nu_flag = 0; 1035 nuidp->nu_cr.cr_uid = cred->cr_uid; 1036 nuidp->nu_expire = time.tv_sec + NFS_KERBTTL; 1037 nuidp->nu_timestamp = ktvout; 1038 nuidp->nu_nickname = nick; 1039 bcopy(key, nuidp->nu_key, sizeof (key)); 1040 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1041 nu_lru); 1042 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1043 nuidp, nu_hash); 1044 } 1045 } else 1046 nfsm_adv(nfsm_rndup(len)); 1047nfsmout: 1048 *mdp = md; 1049 *dposp = dpos; 1050 return (error); 1051} 1052 1053#ifndef NFS_NOSERVER 1054/* 1055 * Derefence a server socket structure. If it has no more references and 1056 * is no longer valid, you can throw it away. 1057 */ 1058void 1059nfsrv_slpderef(slp) 1060 register struct nfssvc_sock *slp; 1061{ 1062 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 1063 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1064 free((caddr_t)slp, M_NFSSVC); 1065 } 1066} 1067 1068/* 1069 * Initialize the data structures for the server. 1070 * Handshake with any new nfsds starting up to avoid any chance of 1071 * corruption. 1072 */ 1073void 1074nfsrv_init(terminating) 1075 int terminating; 1076{ 1077 register struct nfssvc_sock *slp, *nslp; 1078 1079 if (nfssvc_sockhead_flag & SLP_INIT) 1080 panic("nfsd init"); 1081 nfssvc_sockhead_flag |= SLP_INIT; 1082 if (terminating) { 1083 for (slp = nfssvc_sockhead.tqh_first; slp != 0; slp = nslp) { 1084 nslp = slp->ns_chain.tqe_next; 1085 if (slp->ns_flag & SLP_VALID) 1086 nfsrv_zapsock(slp); 1087 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 1088 free((caddr_t)slp, M_NFSSVC); 1089 } 1090 nfsrv_cleancache(); /* And clear out server cache */ 1091 } 1092 1093 TAILQ_INIT(&nfssvc_sockhead); 1094 nfssvc_sockhead_flag &= ~SLP_INIT; 1095 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 1096 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 1097 wakeup((caddr_t)&nfssvc_sockhead); 1098 } 1099 1100 TAILQ_INIT(&nfsd_head); 1101 nfsd_head_flag &= ~NFSD_CHECKSLP; 1102 1103 nfs_udpsock = (struct nfssvc_sock *) 1104 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1105 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 1106 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 1107 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 1108 1109 nfs_cltpsock = (struct nfssvc_sock *) 1110 malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 1111 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 1112 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 1113 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 1114} 1115 1116/* 1117 * Add entries to the server monitor log. 1118 */ 1119static void 1120nfsd_rt(sotype, nd, cacherep) 1121 int sotype; 1122 register struct nfsrv_descript *nd; 1123 int cacherep; 1124{ 1125 register struct drt *rt; 1126 1127 rt = &nfsdrt.drt[nfsdrt.pos]; 1128 if (cacherep == RC_DOIT) 1129 rt->flag = 0; 1130 else if (cacherep == RC_REPLY) 1131 rt->flag = DRT_CACHEREPLY; 1132 else 1133 rt->flag = DRT_CACHEDROP; 1134 if (sotype == SOCK_STREAM) 1135 rt->flag |= DRT_TCP; 1136 if (nd->nd_flag & ND_NQNFS) 1137 rt->flag |= DRT_NQNFS; 1138 else if (nd->nd_flag & ND_NFSV3) 1139 rt->flag |= DRT_NFSV3; 1140 rt->proc = nd->nd_procnum; 1141 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 1142 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 1143 else 1144 rt->ipadr = INADDR_ANY; 1145 rt->resptime = ((time.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 1146 (time.tv_usec - nd->nd_starttime.tv_usec); 1147 rt->tstamp = time; 1148 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1149} 1150#endif /* NFS_NOSERVER */ 1151