1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 37 * $FreeBSD: src/sys/nfs/nfs_syscalls.c,v 1.58.2.1 2000/11/26 02:30:06 dillon Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_syscalls.c,v 1.31 2008/01/05 14:02:41 swildner Exp $ 39 */ 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/file.h> 47 #include <sys/filedesc.h> 48 #include <sys/vnode.h> 49 #include <sys/malloc.h> 50 #include <sys/mount.h> 51 #include <sys/proc.h> 52 #include <sys/priv.h> 53 #include <sys/buf.h> 54 #include <sys/mbuf.h> 55 #include <sys/resourcevar.h> 56 #include <sys/socket.h> 57 #include <sys/socketvar.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/nlookup.h> 61 #include <vm/vm_zone.h> 62 63 #include <netinet/in.h> 64 #include <netinet/tcp.h> 65 #include "xdr_subs.h" 66 #include "rpcv2.h" 67 #include "nfsproto.h" 68 #include "nfs.h" 69 #include "nfsm_subs.h" 70 #include "nfsrvcache.h" 71 #include "nfsmount.h" 72 #include "nfsnode.h" 73 #include "nfsrtt.h" 74 75 #include <sys/thread2.h> 76 77 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 78 79 /* Global defs. */ 80 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd, 81 struct nfssvc_sock *slp, 82 struct thread *td, 83 struct mbuf **mreqp); 84 extern int nfs_numasync; 85 extern int nfsrtton; 86 extern struct nfsstats nfsstats; 87 extern int nfsrvw_procrastinate; 88 extern int nfsrvw_procrastinate_v3; 89 static int nuidhash_max = NFS_MAXUIDHASH; 90 91 #ifndef NFS_NOSERVER 92 static void nfsrv_zapsock (struct nfssvc_sock *slp); 93 #endif 94 static int nfssvc_iod (struct thread *); 95 96 #define TRUE 1 97 #define FALSE 0 98 99 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 100 101 SYSCTL_DECL(_vfs_nfs); 102 103 #ifndef NFS_NOSERVER 104 int nfsd_waiting = 0; 105 static struct nfsdrt nfsdrt; 106 static int nfs_numnfsd = 0; 107 static void nfsd_rt (int sotype, struct nfsrv_descript *nd, 108 int cacherep); 109 static int nfssvc_addsock (struct file *, struct sockaddr *, 110 struct thread *); 111 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *); 112 113 static int nfs_privport = 0; 114 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, ""); 115 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, ""); 116 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, ""); 117 static int nfs_soreserve = 65535; 118 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, ""); 119 120 /* 121 * NFS server system calls 122 */ 123 124 #endif /* NFS_NOSERVER */ 125 /* 126 * nfssvc_args(int flag, caddr_t argp) 127 * 128 * Nfs server psuedo system call for the nfsd's 129 * Based on the flag value it either: 130 * - adds a socket to the selection list 131 * - remains in the kernel as an nfsd 132 * - remains in the kernel as an nfsiod 133 */ 134 int 135 sys_nfssvc(struct nfssvc_args *uap) 136 { 137 #ifndef NFS_NOSERVER 138 struct nlookupdata nd; 139 struct file *fp; 140 struct sockaddr *nam; 141 struct nfsd_args nfsdarg; 142 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 143 struct nfsd_cargs ncd; 144 struct nfsd *nfsd; 145 struct nfssvc_sock *slp; 146 struct nfsuid *nuidp; 147 struct nfsmount *nmp; 148 struct vnode *vp; 149 #endif /* NFS_NOSERVER */ 150 int error; 151 struct thread *td = curthread; 152 153 /* 154 * Must be super user 155 */ 156 error = priv_check(td, PRIV_ROOT); 157 if(error) 158 return (error); 159 KKASSERT(td->td_proc); /* for ucred and p_fd */ 160 while (nfssvc_sockhead_flag & SLP_INIT) { 161 nfssvc_sockhead_flag |= SLP_WANTINIT; 162 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 163 } 164 if (uap->flag & NFSSVC_BIOD) 165 error = nfssvc_iod(td); 166 #ifdef NFS_NOSERVER 167 else 168 error = ENXIO; 169 #else /* !NFS_NOSERVER */ 170 else if (uap->flag & NFSSVC_MNTD) { 171 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 172 if (error) 173 return (error); 174 vp = NULL; 175 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 176 NLC_FOLLOW); 177 if (error == 0) 178 error = nlookup(&nd); 179 if (error == 0) 180 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 181 nlookup_done(&nd); 182 if (error) 183 return (error); 184 185 if ((vp->v_flag & VROOT) == 0) 186 error = EINVAL; 187 nmp = VFSTONFS(vp->v_mount); 188 vput(vp); 189 if (error) 190 return (error); 191 if ((nmp->nm_state & NFSSTA_MNTD) && 192 (uap->flag & NFSSVC_GOTAUTH) == 0) 193 return (0); 194 nmp->nm_state |= NFSSTA_MNTD; 195 error = nfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag, 196 uap->argp, td); 197 } else if (uap->flag & NFSSVC_ADDSOCK) { 198 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 199 if (error) 200 return (error); 201 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 202 if (error) 203 return (error); 204 /* 205 * Get the client address for connected sockets. 206 */ 207 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 208 nam = (struct sockaddr *)0; 209 else { 210 error = getsockaddr(&nam, nfsdarg.name, 211 nfsdarg.namelen); 212 if (error) { 213 fdrop(fp); 214 return (error); 215 } 216 } 217 error = nfssvc_addsock(fp, nam, td); 218 fdrop(fp); 219 } else { 220 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 221 if (error) 222 return (error); 223 if ((uap->flag & NFSSVC_AUTHIN) && 224 ((nfsd = nsd->nsd_nfsd)) != NULL && 225 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 226 slp = nfsd->nfsd_slp; 227 228 /* 229 * First check to see if another nfsd has already 230 * added this credential. 231 */ 232 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 233 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 234 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 235 (!nfsd->nfsd_nd->nd_nam2 || 236 netaddr_match(NU_NETFAM(nuidp), 237 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 238 break; 239 } 240 if (nuidp) { 241 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 242 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 243 } else { 244 /* 245 * Nope, so we will. 246 */ 247 if (slp->ns_numuids < nuidhash_max) { 248 slp->ns_numuids++; 249 nuidp = (struct nfsuid *) 250 kmalloc(sizeof (struct nfsuid), M_NFSUID, 251 M_WAITOK); 252 } else 253 nuidp = (struct nfsuid *)0; 254 if ((slp->ns_flag & SLP_VALID) == 0) { 255 if (nuidp) 256 kfree((caddr_t)nuidp, M_NFSUID); 257 } else { 258 if (nuidp == (struct nfsuid *)0) { 259 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 260 LIST_REMOVE(nuidp, nu_hash); 261 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 262 nu_lru); 263 if (nuidp->nu_flag & NU_NAM) 264 FREE(nuidp->nu_nam, M_SONAME); 265 } 266 nuidp->nu_flag = 0; 267 nuidp->nu_cr = nsd->nsd_cr; 268 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 269 nuidp->nu_cr.cr_ngroups = NGROUPS; 270 nuidp->nu_cr.cr_ref = 1; 271 nuidp->nu_timestamp = nsd->nsd_timestamp; 272 nuidp->nu_expire = time_second + nsd->nsd_ttl; 273 /* 274 * and save the session key in nu_key. 275 */ 276 bcopy(nsd->nsd_key, nuidp->nu_key, 277 sizeof (nsd->nsd_key)); 278 if (nfsd->nfsd_nd->nd_nam2) { 279 struct sockaddr_in *saddr; 280 281 saddr = (struct sockaddr_in *) 282 nfsd->nfsd_nd->nd_nam2; 283 switch (saddr->sin_family) { 284 case AF_INET: 285 nuidp->nu_flag |= NU_INETADDR; 286 nuidp->nu_inetaddr = 287 saddr->sin_addr.s_addr; 288 break; 289 case AF_ISO: 290 default: 291 nuidp->nu_flag |= NU_NAM; 292 nuidp->nu_nam = 293 dup_sockaddr(nfsd->nfsd_nd->nd_nam2); 294 break; 295 }; 296 } 297 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 298 nu_lru); 299 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 300 nuidp, nu_hash); 301 nfsrv_setcred(&nuidp->nu_cr, 302 &nfsd->nfsd_nd->nd_cr); 303 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 304 } 305 } 306 } 307 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 308 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 309 error = nfssvc_nfsd(nsd, uap->argp, td); 310 } 311 #endif /* NFS_NOSERVER */ 312 if (error == EINTR || error == ERESTART) 313 error = 0; 314 return (error); 315 } 316 317 #ifndef NFS_NOSERVER 318 /* 319 * Adds a socket to the list for servicing by nfsds. 320 */ 321 static int 322 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td) 323 { 324 int siz; 325 struct nfssvc_sock *slp; 326 struct socket *so; 327 int error; 328 329 so = (struct socket *)fp->f_data; 330 #if 0 331 tslp = (struct nfssvc_sock *)0; 332 /* 333 * Add it to the list, as required. 334 */ 335 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 336 tslp = nfs_udpsock; 337 if (tslp->ns_flag & SLP_VALID) { 338 if (mynam != NULL) 339 FREE(mynam, M_SONAME); 340 return (EPERM); 341 } 342 } 343 #endif 344 /* 345 * Reserve buffer space in the socket. Note that due to bugs in 346 * Linux's delayed-ack code, serious performance degredation may 347 * occur with linux hosts if the minimum is used. 348 */ 349 if (so->so_type == SOCK_STREAM) 350 siz = NFS_MAXPACKET + sizeof (u_long); 351 else 352 siz = NFS_MAXPACKET; 353 if (siz < nfs_soreserve) 354 siz = nfs_soreserve; 355 if (siz > sb_max_adj) { 356 kprintf("Warning: vfs.nfs.soreserve (%d) " 357 "limited to adjusted sb_max (%ld)\n", 358 nfs_soreserve, sb_max_adj); 359 siz = sb_max_adj; 360 } 361 362 error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]); 363 if (error) { 364 if (mynam != NULL) 365 FREE(mynam, M_SONAME); 366 return (error); 367 } 368 369 /* 370 * Set protocol specific options { for now TCP only } and 371 * reserve some space. For datagram sockets, this can get called 372 * repeatedly for the same socket, but that isn't harmful. 373 */ 374 if (so->so_type == SOCK_STREAM) { 375 struct sockopt sopt; 376 int val; 377 378 bzero(&sopt, sizeof sopt); 379 sopt.sopt_level = SOL_SOCKET; 380 sopt.sopt_name = SO_KEEPALIVE; 381 sopt.sopt_val = &val; 382 sopt.sopt_valsize = sizeof val; 383 val = 1; 384 sosetopt(so, &sopt); 385 } 386 if (so->so_proto->pr_domain->dom_family == AF_INET && 387 so->so_proto->pr_protocol == IPPROTO_TCP) { 388 struct sockopt sopt; 389 int val; 390 391 bzero(&sopt, sizeof sopt); 392 sopt.sopt_level = IPPROTO_TCP; 393 sopt.sopt_name = TCP_NODELAY; 394 sopt.sopt_val = &val; 395 sopt.sopt_valsize = sizeof val; 396 val = 1; 397 sosetopt(so, &sopt); 398 } 399 so->so_rcv.ssb_flags &= ~SSB_NOINTR; 400 so->so_rcv.ssb_timeo = 0; 401 so->so_snd.ssb_flags &= ~SSB_NOINTR; 402 so->so_snd.ssb_timeo = 0; 403 404 slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock), 405 M_NFSSVC, M_WAITOK | M_ZERO); 406 STAILQ_INIT(&slp->ns_rec); 407 TAILQ_INIT(&slp->ns_uidlruhead); 408 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 409 410 slp->ns_so = so; 411 slp->ns_nam = mynam; 412 fp->f_count++; 413 slp->ns_fp = fp; 414 crit_enter(); 415 so->so_upcallarg = (caddr_t)slp; 416 so->so_upcall = nfsrv_rcv; 417 so->so_rcv.ssb_flags |= SSB_UPCALL; 418 slp->ns_flag = (SLP_VALID | SLP_NEEDQ); 419 nfsrv_wakenfsd(slp, 1); 420 crit_exit(); 421 return (0); 422 } 423 424 /* 425 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 426 * until it is killed by a signal. 427 */ 428 static int 429 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td) 430 { 431 int siz; 432 struct nfssvc_sock *slp; 433 struct nfsd *nfsd = nsd->nsd_nfsd; 434 struct nfsrv_descript *nd = NULL; 435 struct mbuf *m, *mreq; 436 int error = 0, cacherep, sotype, writes_todo; 437 int procrastinate; 438 u_quad_t cur_usec; 439 440 #ifndef nolint 441 cacherep = RC_DOIT; 442 writes_todo = 0; 443 #endif 444 if (nfsd == (struct nfsd *)0) { 445 nsd->nsd_nfsd = nfsd = (struct nfsd *) 446 kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO); 447 crit_enter(); 448 nfsd->nfsd_td = td; 449 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 450 nfs_numnfsd++; 451 } else 452 crit_enter(); 453 454 /* 455 * Loop getting rpc requests until SIGKILL. 456 */ 457 for (;;) { 458 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 459 while (nfsd->nfsd_slp == NULL && 460 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 461 nfsd->nfsd_flag |= NFSD_WAITING; 462 nfsd_waiting++; 463 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0); 464 nfsd_waiting--; 465 if (error) 466 goto done; 467 } 468 if (nfsd->nfsd_slp == NULL && 469 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 470 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 471 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 472 == (SLP_VALID | SLP_DOREC)) { 473 slp->ns_flag &= ~SLP_DOREC; 474 slp->ns_sref++; 475 nfsd->nfsd_slp = slp; 476 break; 477 } 478 } 479 if (slp == 0) 480 nfsd_head_flag &= ~NFSD_CHECKSLP; 481 } 482 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 483 continue; 484 if (slp->ns_flag & SLP_VALID) { 485 if (slp->ns_flag & SLP_DISCONN) 486 nfsrv_zapsock(slp); 487 else if (slp->ns_flag & SLP_NEEDQ) { 488 slp->ns_flag &= ~SLP_NEEDQ; 489 (void) nfs_slplock(slp, 1); 490 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 491 MB_WAIT); 492 nfs_slpunlock(slp); 493 } 494 error = nfsrv_dorec(slp, nfsd, &nd); 495 cur_usec = nfs_curusec(); 496 if (error && slp->ns_tq.lh_first && 497 slp->ns_tq.lh_first->nd_time <= cur_usec) { 498 error = 0; 499 cacherep = RC_DOIT; 500 writes_todo = 1; 501 } else 502 writes_todo = 0; 503 nfsd->nfsd_flag |= NFSD_REQINPROG; 504 } 505 } else { 506 error = 0; 507 slp = nfsd->nfsd_slp; 508 } 509 if (error || (slp->ns_flag & SLP_VALID) == 0) { 510 if (nd) { 511 kfree((caddr_t)nd, M_NFSRVDESC); 512 nd = NULL; 513 } 514 nfsd->nfsd_slp = (struct nfssvc_sock *)0; 515 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 516 nfsrv_slpderef(slp); 517 continue; 518 } 519 crit_exit(); 520 sotype = slp->ns_so->so_type; 521 if (nd) { 522 getmicrotime(&nd->nd_starttime); 523 if (nd->nd_nam2) 524 nd->nd_nam = nd->nd_nam2; 525 else 526 nd->nd_nam = slp->ns_nam; 527 528 /* 529 * Check to see if authorization is needed. 530 */ 531 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 532 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 533 nsd->nsd_haddr = 534 ((struct sockaddr_in *) 535 nd->nd_nam)->sin_addr.s_addr; 536 nsd->nsd_authlen = nfsd->nfsd_authlen; 537 nsd->nsd_verflen = nfsd->nfsd_verflen; 538 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 539 nfsd->nfsd_authlen) && 540 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 541 nfsd->nfsd_verflen) && 542 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 543 return (ENEEDAUTH); 544 cacherep = RC_DROPIT; 545 } else { 546 cacherep = nfsrv_getcache(nd, slp, &mreq); 547 } 548 549 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 550 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 551 nd->nd_procnum = NFSPROC_NOOP; 552 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 553 cacherep = RC_DOIT; 554 } else if (nfs_privport) { 555 /* Check if source port is privileged */ 556 u_short port; 557 struct sockaddr *nam = nd->nd_nam; 558 struct sockaddr_in *sin; 559 560 sin = (struct sockaddr_in *)nam; 561 port = ntohs(sin->sin_port); 562 if (port >= IPPORT_RESERVED && 563 nd->nd_procnum != NFSPROC_NULL) { 564 nd->nd_procnum = NFSPROC_NOOP; 565 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 566 cacherep = RC_DOIT; 567 kprintf("NFS request from unprivileged port (%s:%d)\n", 568 inet_ntoa(sin->sin_addr), port); 569 } 570 } 571 572 } 573 574 /* 575 * Loop to get all the write rpc relies that have been 576 * gathered together. 577 */ 578 do { 579 switch (cacherep) { 580 case RC_DOIT: 581 if (nd && (nd->nd_flag & ND_NFSV3)) 582 procrastinate = nfsrvw_procrastinate_v3; 583 else 584 procrastinate = nfsrvw_procrastinate; 585 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 586 procrastinate > 0) 587 ) { 588 error = nfsrv_writegather(&nd, slp, 589 nfsd->nfsd_td, &mreq); 590 } else { 591 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 592 slp, nfsd->nfsd_td, &mreq); 593 } 594 if (mreq == NULL) 595 break; 596 if (error != 0 && error != NFSERR_RETVOID) { 597 if (nd->nd_procnum != NQNFSPROC_VACATED) 598 nfsstats.srv_errs++; 599 nfsrv_updatecache(nd, FALSE, mreq); 600 if (nd->nd_nam2) 601 FREE(nd->nd_nam2, M_SONAME); 602 break; 603 } 604 nfsstats.srvrpccnt[nd->nd_procnum]++; 605 nfsrv_updatecache(nd, TRUE, mreq); 606 nd->nd_mrep = (struct mbuf *)0; 607 case RC_REPLY: 608 m = mreq; 609 siz = 0; 610 while (m) { 611 siz += m->m_len; 612 m = m->m_next; 613 } 614 if (siz <= 0 || siz > NFS_MAXPACKET) { 615 kprintf("mbuf siz=%d\n",siz); 616 panic("Bad nfs svc reply"); 617 } 618 m = mreq; 619 m->m_pkthdr.len = siz; 620 m->m_pkthdr.rcvif = (struct ifnet *)0; 621 /* 622 * For stream protocols, prepend a Sun RPC 623 * Record Mark. 624 */ 625 if (sotype == SOCK_STREAM) { 626 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT); 627 if (m == NULL) 628 return (ENOBUFS); 629 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 630 } 631 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 632 (void) nfs_slplock(slp, 1); 633 if (slp->ns_flag & SLP_VALID) 634 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 635 else { 636 error = EPIPE; 637 m_freem(m); 638 } 639 if (nfsrtton) 640 nfsd_rt(sotype, nd, cacherep); 641 if (nd->nd_nam2) 642 FREE(nd->nd_nam2, M_SONAME); 643 if (nd->nd_mrep) 644 m_freem(nd->nd_mrep); 645 if (error == EPIPE) 646 nfsrv_zapsock(slp); 647 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 648 nfs_slpunlock(slp); 649 if (error == EINTR || error == ERESTART) { 650 kfree((caddr_t)nd, M_NFSRVDESC); 651 nfsrv_slpderef(slp); 652 crit_enter(); 653 goto done; 654 } 655 break; 656 case RC_DROPIT: 657 if (nfsrtton) 658 nfsd_rt(sotype, nd, cacherep); 659 m_freem(nd->nd_mrep); 660 if (nd->nd_nam2) 661 FREE(nd->nd_nam2, M_SONAME); 662 break; 663 }; 664 if (nd) { 665 FREE((caddr_t)nd, M_NFSRVDESC); 666 nd = NULL; 667 } 668 669 /* 670 * Check to see if there are outstanding writes that 671 * need to be serviced. 672 */ 673 cur_usec = nfs_curusec(); 674 crit_enter(); 675 if (slp->ns_tq.lh_first && 676 slp->ns_tq.lh_first->nd_time <= cur_usec) { 677 cacherep = RC_DOIT; 678 writes_todo = 1; 679 } else 680 writes_todo = 0; 681 crit_exit(); 682 } while (writes_todo); 683 crit_enter(); 684 if (nfsrv_dorec(slp, nfsd, &nd)) { 685 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 686 nfsd->nfsd_slp = NULL; 687 nfsrv_slpderef(slp); 688 } 689 } 690 done: 691 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 692 crit_exit(); 693 kfree((caddr_t)nfsd, M_NFSD); 694 nsd->nsd_nfsd = (struct nfsd *)0; 695 if (--nfs_numnfsd == 0) 696 nfsrv_init(TRUE); /* Reinitialize everything */ 697 return (error); 698 } 699 700 /* 701 * Shut down a socket associated with an nfssvc_sock structure. 702 * Should be called with the send lock set, if required. 703 * The trick here is to increment the sref at the start, so that the nfsds 704 * will stop using it and clear ns_flag at the end so that it will not be 705 * reassigned during cleanup. 706 */ 707 static void 708 nfsrv_zapsock(struct nfssvc_sock *slp) 709 { 710 struct nfsuid *nuidp, *nnuidp; 711 struct nfsrv_descript *nwp, *nnwp; 712 struct socket *so; 713 struct file *fp; 714 struct nfsrv_rec *rec; 715 716 slp->ns_flag &= ~SLP_ALLFLAGS; 717 fp = slp->ns_fp; 718 if (fp) { 719 slp->ns_fp = (struct file *)0; 720 so = slp->ns_so; 721 so->so_rcv.ssb_flags &= ~SSB_UPCALL; 722 so->so_upcall = NULL; 723 so->so_upcallarg = NULL; 724 soshutdown(so, SHUT_RDWR); 725 closef(fp, NULL); 726 if (slp->ns_nam) 727 FREE(slp->ns_nam, M_SONAME); 728 m_freem(slp->ns_raw); 729 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 730 --slp->ns_numrec; 731 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 732 if (rec->nr_address) 733 FREE(rec->nr_address, M_SONAME); 734 m_freem(rec->nr_packet); 735 kfree(rec, M_NFSRVDESC); 736 } 737 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru, 738 nnuidp) { 739 LIST_REMOVE(nuidp, nu_hash); 740 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 741 if (nuidp->nu_flag & NU_NAM) 742 FREE(nuidp->nu_nam, M_SONAME); 743 kfree((caddr_t)nuidp, M_NFSUID); 744 } 745 crit_enter(); 746 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 747 nnwp = nwp->nd_tq.le_next; 748 LIST_REMOVE(nwp, nd_tq); 749 kfree((caddr_t)nwp, M_NFSRVDESC); 750 } 751 LIST_INIT(&slp->ns_tq); 752 crit_exit(); 753 } 754 } 755 756 /* 757 * Derefence a server socket structure. If it has no more references and 758 * is no longer valid, you can throw it away. 759 */ 760 void 761 nfsrv_slpderef(struct nfssvc_sock *slp) 762 { 763 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 764 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 765 kfree((caddr_t)slp, M_NFSSVC); 766 } 767 } 768 769 /* 770 * Lock a socket against others. 771 */ 772 int 773 nfs_slplock(struct nfssvc_sock *slp, int wait) 774 { 775 int *statep = &slp->ns_solock; 776 777 if (!wait && (*statep & NFSSTA_SNDLOCK)) 778 return(0); /* already locked, fail */ 779 while (*statep & NFSSTA_SNDLOCK) { 780 *statep |= NFSSTA_WANTSND; 781 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0); 782 } 783 *statep |= NFSSTA_SNDLOCK; 784 return (1); 785 } 786 787 /* 788 * Unlock the stream socket for others. 789 */ 790 void 791 nfs_slpunlock(struct nfssvc_sock *slp) 792 { 793 int *statep = &slp->ns_solock; 794 795 if ((*statep & NFSSTA_SNDLOCK) == 0) 796 panic("nfs slpunlock"); 797 *statep &= ~NFSSTA_SNDLOCK; 798 if (*statep & NFSSTA_WANTSND) { 799 *statep &= ~NFSSTA_WANTSND; 800 wakeup((caddr_t)statep); 801 } 802 } 803 804 /* 805 * Initialize the data structures for the server. 806 * Handshake with any new nfsds starting up to avoid any chance of 807 * corruption. 808 */ 809 void 810 nfsrv_init(int terminating) 811 { 812 struct nfssvc_sock *slp, *nslp; 813 814 if (nfssvc_sockhead_flag & SLP_INIT) 815 panic("nfsd init"); 816 nfssvc_sockhead_flag |= SLP_INIT; 817 if (terminating) { 818 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) { 819 if (slp->ns_flag & SLP_VALID) 820 nfsrv_zapsock(slp); 821 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 822 kfree((caddr_t)slp, M_NFSSVC); 823 } 824 nfsrv_cleancache(); /* And clear out server cache */ 825 } else 826 nfs_pub.np_valid = 0; 827 828 TAILQ_INIT(&nfssvc_sockhead); 829 nfssvc_sockhead_flag &= ~SLP_INIT; 830 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 831 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 832 wakeup((caddr_t)&nfssvc_sockhead); 833 } 834 835 TAILQ_INIT(&nfsd_head); 836 nfsd_head_flag &= ~NFSD_CHECKSLP; 837 838 #if 0 839 nfs_udpsock = (struct nfssvc_sock *) 840 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 841 STAILQ_INIT(&nfs_udpsock->ns_rec); 842 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 843 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 844 845 nfs_cltpsock = (struct nfssvc_sock *) 846 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO); 847 STAILQ_INIT(&nfs_cltpsock->ns_rec); 848 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 849 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 850 #endif 851 } 852 853 /* 854 * Add entries to the server monitor log. 855 */ 856 static void 857 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 858 { 859 struct drt *rt; 860 861 rt = &nfsdrt.drt[nfsdrt.pos]; 862 if (cacherep == RC_DOIT) 863 rt->flag = 0; 864 else if (cacherep == RC_REPLY) 865 rt->flag = DRT_CACHEREPLY; 866 else 867 rt->flag = DRT_CACHEDROP; 868 if (sotype == SOCK_STREAM) 869 rt->flag |= DRT_TCP; 870 if (nd->nd_flag & ND_NFSV3) 871 rt->flag |= DRT_NFSV3; 872 rt->proc = nd->nd_procnum; 873 if (nd->nd_nam->sa_family == AF_INET) 874 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 875 else 876 rt->ipadr = INADDR_ANY; 877 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 878 getmicrotime(&rt->tstamp); 879 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 880 } 881 #endif /* NFS_NOSERVER */ 882 883 static int nfs_defect = 0; 884 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 885 886 /* 887 * Asynchronous I/O daemons for client nfs. 888 * They do read-ahead and write-behind operations on the block I/O cache. 889 * Never returns unless it fails or gets killed. 890 */ 891 static int 892 nfssvc_iod(struct thread *td) 893 { 894 struct bio *bio; 895 int i, myiod; 896 struct nfsmount *nmp; 897 int error = 0; 898 899 /* 900 * Assign my position or return error if too many already running 901 */ 902 myiod = -1; 903 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 904 if (nfs_asyncdaemon[i] == 0) { 905 nfs_asyncdaemon[i]++; 906 myiod = i; 907 break; 908 } 909 if (myiod == -1) 910 return (EBUSY); 911 nfs_numasync++; 912 /* 913 * Just loop around doin our stuff until SIGKILL 914 */ 915 for (;;) { 916 while (((nmp = nfs_iodmount[myiod]) == NULL 917 || TAILQ_EMPTY(&nmp->nm_bioq)) 918 && error == 0) { 919 if (nmp) 920 nmp->nm_bioqiods--; 921 nfs_iodwant[myiod] = td; 922 nfs_iodmount[myiod] = NULL; 923 error = tsleep((caddr_t)&nfs_iodwant[myiod], 924 PCATCH, "nfsidl", 0); 925 } 926 if (error) { 927 nfs_asyncdaemon[myiod] = 0; 928 if (nmp) 929 nmp->nm_bioqiods--; 930 nfs_iodwant[myiod] = NULL; 931 nfs_iodmount[myiod] = NULL; 932 nfs_numasync--; 933 return (error); 934 } 935 while ((bio = TAILQ_FIRST(&nmp->nm_bioq)) != NULL) { 936 /* 937 * Take one off the front of the list. The BIO's 938 * block number is normalized for DEV_BSIZE. 939 */ 940 TAILQ_REMOVE(&nmp->nm_bioq, bio, bio_act); 941 nmp->nm_bioqlen--; 942 if (nmp->nm_bioqwant && nmp->nm_bioqlen <= nfs_numasync) { 943 nmp->nm_bioqwant = FALSE; 944 wakeup(&nmp->nm_bioq); 945 } 946 nfs_doio((struct vnode *)bio->bio_driver_info, bio, NULL); 947 /* 948 * If there are more than one iod on this mount, then defect 949 * so that the iods can be shared out fairly between the mounts 950 */ 951 if (nfs_defect && nmp->nm_bioqiods > 1) { 952 NFS_DPF(ASYNCIO, 953 ("nfssvc_iod: iod %d defecting from mount %p\n", 954 myiod, nmp)); 955 nfs_iodmount[myiod] = NULL; 956 nmp->nm_bioqiods--; 957 break; 958 } 959 } 960 } 961 } 962 963 964 /* 965 * Get an authorization string for the uid by having the mount_nfs sitting 966 * on this mount point porpous out of the kernel and do it. 967 */ 968 int 969 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, 970 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str, 971 int *verf_len, NFSKERBKEY_T key /* return session key */) 972 { 973 int error = 0; 974 975 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 976 nmp->nm_state |= NFSSTA_WANTAUTH; 977 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 978 "nfsauth1", 2 * hz); 979 error = nfs_sigintr(nmp, rep, rep->r_td); 980 if (error) { 981 nmp->nm_state &= ~NFSSTA_WANTAUTH; 982 return (error); 983 } 984 } 985 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 986 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 987 nmp->nm_authlen = RPCAUTH_MAXSIZ; 988 nmp->nm_verfstr = verf_str; 989 nmp->nm_verflen = *verf_len; 990 nmp->nm_authuid = cred->cr_uid; 991 wakeup((caddr_t)&nmp->nm_authstr); 992 993 /* 994 * And wait for mount_nfs to do its stuff. 995 */ 996 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 997 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 998 "nfsauth2", 2 * hz); 999 error = nfs_sigintr(nmp, rep, rep->r_td); 1000 } 1001 if (nmp->nm_state & NFSSTA_AUTHERR) { 1002 nmp->nm_state &= ~NFSSTA_AUTHERR; 1003 error = EAUTH; 1004 } 1005 if (error) 1006 kfree((caddr_t)*auth_str, M_TEMP); 1007 else { 1008 *auth_len = nmp->nm_authlen; 1009 *verf_len = nmp->nm_verflen; 1010 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1011 } 1012 nmp->nm_state &= ~NFSSTA_HASAUTH; 1013 nmp->nm_state |= NFSSTA_WAITAUTH; 1014 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1015 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1016 wakeup((caddr_t)&nmp->nm_authtype); 1017 } 1018 return (error); 1019 } 1020 1021 /* 1022 * Get a nickname authenticator and verifier. 1023 */ 1024 int 1025 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str, 1026 int *auth_len, char *verf_str, int verf_len) 1027 { 1028 struct nfsuid *nuidp; 1029 u_int32_t *nickp, *verfp; 1030 struct timeval ktvin, ktvout; 1031 1032 #ifdef DIAGNOSTIC 1033 if (verf_len < (4 * NFSX_UNSIGNED)) 1034 panic("nfs_getnickauth verf too small"); 1035 #endif 1036 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1037 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1038 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1039 break; 1040 } 1041 if (!nuidp || nuidp->nu_expire < time_second) 1042 return (EACCES); 1043 1044 /* 1045 * Move to the end of the lru list (end of lru == most recently used). 1046 */ 1047 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1048 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1049 1050 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1051 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1052 *nickp = txdr_unsigned(nuidp->nu_nickname); 1053 *auth_str = (char *)nickp; 1054 *auth_len = 2 * NFSX_UNSIGNED; 1055 1056 /* 1057 * Now we must encrypt the verifier and package it up. 1058 */ 1059 verfp = (u_int32_t *)verf_str; 1060 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1061 if (time_second > nuidp->nu_timestamp.tv_sec || 1062 (time_second == nuidp->nu_timestamp.tv_sec && 1063 time_second > nuidp->nu_timestamp.tv_usec)) 1064 getmicrotime(&nuidp->nu_timestamp); 1065 else 1066 nuidp->nu_timestamp.tv_usec++; 1067 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1068 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1069 1070 /* 1071 * Now encrypt the timestamp verifier in ecb mode using the session 1072 * key. 1073 */ 1074 #ifdef NFSKERB 1075 XXX 1076 #endif 1077 1078 *verfp++ = ktvout.tv_sec; 1079 *verfp++ = ktvout.tv_usec; 1080 *verfp = 0; 1081 return (0); 1082 } 1083 1084 /* 1085 * Save the current nickname in a hash list entry on the mount point. 1086 */ 1087 int 1088 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len, 1089 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, 1090 struct mbuf *mrep) 1091 { 1092 struct nfsuid *nuidp; 1093 u_int32_t *tl; 1094 int32_t t1; 1095 struct mbuf *md = *mdp; 1096 struct timeval ktvin, ktvout; 1097 u_int32_t nick; 1098 char *dpos = *dposp, *cp2; 1099 int deltasec, error = 0; 1100 1101 if (len == (3 * NFSX_UNSIGNED)) { 1102 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1103 ktvin.tv_sec = *tl++; 1104 ktvin.tv_usec = *tl++; 1105 nick = fxdr_unsigned(u_int32_t, *tl); 1106 1107 /* 1108 * Decrypt the timestamp in ecb mode. 1109 */ 1110 #ifdef NFSKERB 1111 XXX 1112 #endif 1113 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1114 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1115 deltasec = time_second - ktvout.tv_sec; 1116 if (deltasec < 0) 1117 deltasec = -deltasec; 1118 /* 1119 * If ok, add it to the hash list for the mount point. 1120 */ 1121 if (deltasec <= NFS_KERBCLOCKSKEW) { 1122 if (nmp->nm_numuids < nuidhash_max) { 1123 nmp->nm_numuids++; 1124 nuidp = (struct nfsuid *) 1125 kmalloc(sizeof (struct nfsuid), M_NFSUID, 1126 M_WAITOK); 1127 } else { 1128 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead); 1129 LIST_REMOVE(nuidp, nu_hash); 1130 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1131 nu_lru); 1132 } 1133 nuidp->nu_flag = 0; 1134 nuidp->nu_cr.cr_uid = cred->cr_uid; 1135 nuidp->nu_expire = time_second + NFS_KERBTTL; 1136 nuidp->nu_timestamp = ktvout; 1137 nuidp->nu_nickname = nick; 1138 bcopy(key, nuidp->nu_key, sizeof (key)); 1139 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1140 nu_lru); 1141 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1142 nuidp, nu_hash); 1143 } 1144 } else 1145 nfsm_adv(nfsm_rndup(len)); 1146 nfsmout: 1147 *mdp = md; 1148 *dposp = dpos; 1149 return (error); 1150 } 1151