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 1.29 2007/04/22 01:13:17 dillon 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/buf.h> 53 #include <sys/mbuf.h> 54 #include <sys/resourcevar.h> 55 #include <sys/socket.h> 56 #include <sys/socketvar.h> 57 #include <sys/domain.h> 58 #include <sys/protosw.h> 59 #include <sys/nlookup.h> 60 #include <vm/vm_zone.h> 61 62 #include <netinet/in.h> 63 #include <netinet/tcp.h> 64 #include "xdr_subs.h" 65 #include "rpcv2.h" 66 #include "nfsproto.h" 67 #include "nfs.h" 68 #include "nfsm_subs.h" 69 #include "nfsrvcache.h" 70 #include "nfsmount.h" 71 #include "nfsnode.h" 72 #include "nfsrtt.h" 73 74 #include <sys/thread2.h> 75 76 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure"); 77 78 /* Global defs. */ 79 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd, 80 struct nfssvc_sock *slp, 81 struct thread *td, 82 struct mbuf **mreqp); 83 extern int nfs_numasync; 84 extern int nfsrtton; 85 extern struct nfsstats nfsstats; 86 extern int nfsrvw_procrastinate; 87 extern int nfsrvw_procrastinate_v3; 88 static int nuidhash_max = NFS_MAXUIDHASH; 89 90 #ifndef NFS_NOSERVER 91 static void nfsrv_zapsock (struct nfssvc_sock *slp); 92 #endif 93 static int nfssvc_iod (struct thread *); 94 95 #define TRUE 1 96 #define FALSE 0 97 98 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON]; 99 100 SYSCTL_DECL(_vfs_nfs); 101 102 #ifndef NFS_NOSERVER 103 int nfsd_waiting = 0; 104 static struct nfsdrt nfsdrt; 105 static int nfs_numnfsd = 0; 106 static void nfsd_rt (int sotype, struct nfsrv_descript *nd, 107 int cacherep); 108 static int nfssvc_addsock (struct file *, struct sockaddr *, 109 struct thread *); 110 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *); 111 112 static int nfs_privport = 0; 113 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, ""); 114 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, ""); 115 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, ""); 116 static int nfs_soreserve = 65535; 117 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, ""); 118 119 /* 120 * NFS server system calls 121 */ 122 123 #endif /* NFS_NOSERVER */ 124 /* 125 * nfssvc_args(int flag, caddr_t argp) 126 * 127 * Nfs server psuedo system call for the nfsd's 128 * Based on the flag value it either: 129 * - adds a socket to the selection list 130 * - remains in the kernel as an nfsd 131 * - remains in the kernel as an nfsiod 132 */ 133 int 134 sys_nfssvc(struct nfssvc_args *uap) 135 { 136 #ifndef NFS_NOSERVER 137 struct nlookupdata nd; 138 struct file *fp; 139 struct sockaddr *nam; 140 struct nfsd_args nfsdarg; 141 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 142 struct nfsd_cargs ncd; 143 struct nfsd *nfsd; 144 struct nfssvc_sock *slp; 145 struct nfsuid *nuidp; 146 struct nfsmount *nmp; 147 struct vnode *vp; 148 #endif /* NFS_NOSERVER */ 149 int error; 150 struct thread *td = curthread; 151 152 /* 153 * Must be super user 154 */ 155 error = suser(td); 156 if(error) 157 return (error); 158 KKASSERT(td->td_proc); /* for ucred and p_fd */ 159 while (nfssvc_sockhead_flag & SLP_INIT) { 160 nfssvc_sockhead_flag |= SLP_WANTINIT; 161 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0); 162 } 163 if (uap->flag & NFSSVC_BIOD) 164 error = nfssvc_iod(td); 165 #ifdef NFS_NOSERVER 166 else 167 error = ENXIO; 168 #else /* !NFS_NOSERVER */ 169 else if (uap->flag & NFSSVC_MNTD) { 170 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd)); 171 if (error) 172 return (error); 173 vp = NULL; 174 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 175 NLC_FOLLOW); 176 if (error == 0) 177 error = nlookup(&nd); 178 if (error == 0) 179 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 180 nlookup_done(&nd); 181 if (error) 182 return (error); 183 184 if ((vp->v_flag & VROOT) == 0) 185 error = EINVAL; 186 nmp = VFSTONFS(vp->v_mount); 187 vput(vp); 188 if (error) 189 return (error); 190 if ((nmp->nm_state & NFSSTA_MNTD) && 191 (uap->flag & NFSSVC_GOTAUTH) == 0) 192 return (0); 193 nmp->nm_state |= NFSSTA_MNTD; 194 error = nfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag, 195 uap->argp, td); 196 } else if (uap->flag & NFSSVC_ADDSOCK) { 197 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg)); 198 if (error) 199 return (error); 200 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp); 201 if (error) 202 return (error); 203 /* 204 * Get the client address for connected sockets. 205 */ 206 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 207 nam = (struct sockaddr *)0; 208 else { 209 error = getsockaddr(&nam, nfsdarg.name, 210 nfsdarg.namelen); 211 if (error) { 212 fdrop(fp); 213 return (error); 214 } 215 } 216 error = nfssvc_addsock(fp, nam, td); 217 fdrop(fp); 218 } else { 219 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd)); 220 if (error) 221 return (error); 222 if ((uap->flag & NFSSVC_AUTHIN) && 223 ((nfsd = nsd->nsd_nfsd)) != NULL && 224 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) { 225 slp = nfsd->nfsd_slp; 226 227 /* 228 * First check to see if another nfsd has already 229 * added this credential. 230 */ 231 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first; 232 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 233 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid && 234 (!nfsd->nfsd_nd->nd_nam2 || 235 netaddr_match(NU_NETFAM(nuidp), 236 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 237 break; 238 } 239 if (nuidp) { 240 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr); 241 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 242 } else { 243 /* 244 * Nope, so we will. 245 */ 246 if (slp->ns_numuids < nuidhash_max) { 247 slp->ns_numuids++; 248 nuidp = (struct nfsuid *) 249 kmalloc(sizeof (struct nfsuid), M_NFSUID, 250 M_WAITOK); 251 } else 252 nuidp = (struct nfsuid *)0; 253 if ((slp->ns_flag & SLP_VALID) == 0) { 254 if (nuidp) 255 kfree((caddr_t)nuidp, M_NFSUID); 256 } else { 257 if (nuidp == (struct nfsuid *)0) { 258 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 259 LIST_REMOVE(nuidp, nu_hash); 260 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 261 nu_lru); 262 if (nuidp->nu_flag & NU_NAM) 263 FREE(nuidp->nu_nam, M_SONAME); 264 } 265 nuidp->nu_flag = 0; 266 nuidp->nu_cr = nsd->nsd_cr; 267 if (nuidp->nu_cr.cr_ngroups > NGROUPS) 268 nuidp->nu_cr.cr_ngroups = NGROUPS; 269 nuidp->nu_cr.cr_ref = 1; 270 nuidp->nu_timestamp = nsd->nsd_timestamp; 271 nuidp->nu_expire = time_second + nsd->nsd_ttl; 272 /* 273 * and save the session key in nu_key. 274 */ 275 bcopy(nsd->nsd_key, nuidp->nu_key, 276 sizeof (nsd->nsd_key)); 277 if (nfsd->nfsd_nd->nd_nam2) { 278 struct sockaddr_in *saddr; 279 280 saddr = (struct sockaddr_in *) 281 nfsd->nfsd_nd->nd_nam2; 282 switch (saddr->sin_family) { 283 case AF_INET: 284 nuidp->nu_flag |= NU_INETADDR; 285 nuidp->nu_inetaddr = 286 saddr->sin_addr.s_addr; 287 break; 288 case AF_ISO: 289 default: 290 nuidp->nu_flag |= NU_NAM; 291 nuidp->nu_nam = 292 dup_sockaddr(nfsd->nfsd_nd->nd_nam2); 293 break; 294 }; 295 } 296 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 297 nu_lru); 298 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 299 nuidp, nu_hash); 300 nfsrv_setcred(&nuidp->nu_cr, 301 &nfsd->nfsd_nd->nd_cr); 302 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 303 } 304 } 305 } 306 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd)) 307 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 308 error = nfssvc_nfsd(nsd, uap->argp, td); 309 } 310 #endif /* NFS_NOSERVER */ 311 if (error == EINTR || error == ERESTART) 312 error = 0; 313 return (error); 314 } 315 316 #ifndef NFS_NOSERVER 317 /* 318 * Adds a socket to the list for servicing by nfsds. 319 */ 320 static int 321 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td) 322 { 323 int siz; 324 struct nfssvc_sock *slp; 325 struct socket *so; 326 int error; 327 328 so = (struct socket *)fp->f_data; 329 #if 0 330 tslp = (struct nfssvc_sock *)0; 331 /* 332 * Add it to the list, as required. 333 */ 334 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 335 tslp = nfs_udpsock; 336 if (tslp->ns_flag & SLP_VALID) { 337 if (mynam != NULL) 338 FREE(mynam, M_SONAME); 339 return (EPERM); 340 } 341 } 342 #endif 343 /* 344 * Reserve buffer space in the socket. Note that due to bugs in 345 * Linux's delayed-ack code, serious performance degredation may 346 * occur with linux hosts if the minimum is used. 347 */ 348 if (so->so_type == SOCK_STREAM) 349 siz = NFS_MAXPACKET + sizeof (u_long); 350 else 351 siz = NFS_MAXPACKET; 352 if (siz < nfs_soreserve) 353 siz = nfs_soreserve; 354 if (siz > sb_max_adj) { 355 kprintf("Warning: vfs.nfs.soreserve (%d) " 356 "limited to adjusted sb_max (%ld)\n", 357 nfs_soreserve, sb_max_adj); 358 siz = sb_max_adj; 359 } 360 361 error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]); 362 if (error) { 363 if (mynam != NULL) 364 FREE(mynam, M_SONAME); 365 return (error); 366 } 367 368 /* 369 * Set protocol specific options { for now TCP only } and 370 * reserve some space. For datagram sockets, this can get called 371 * repeatedly for the same socket, but that isn't harmful. 372 */ 373 if (so->so_type == SOCK_STREAM) { 374 struct sockopt sopt; 375 int val; 376 377 bzero(&sopt, sizeof sopt); 378 sopt.sopt_level = SOL_SOCKET; 379 sopt.sopt_name = SO_KEEPALIVE; 380 sopt.sopt_val = &val; 381 sopt.sopt_valsize = sizeof val; 382 val = 1; 383 sosetopt(so, &sopt); 384 } 385 if (so->so_proto->pr_domain->dom_family == AF_INET && 386 so->so_proto->pr_protocol == IPPROTO_TCP) { 387 struct sockopt sopt; 388 int val; 389 390 bzero(&sopt, sizeof sopt); 391 sopt.sopt_level = IPPROTO_TCP; 392 sopt.sopt_name = TCP_NODELAY; 393 sopt.sopt_val = &val; 394 sopt.sopt_valsize = sizeof val; 395 val = 1; 396 sosetopt(so, &sopt); 397 } 398 so->so_rcv.ssb_flags &= ~SSB_NOINTR; 399 so->so_rcv.ssb_timeo = 0; 400 so->so_snd.ssb_flags &= ~SSB_NOINTR; 401 so->so_snd.ssb_timeo = 0; 402 403 slp = (struct nfssvc_sock *) 404 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 405 bzero((caddr_t)slp, sizeof (struct nfssvc_sock)); 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); 447 bzero((caddr_t)nfsd, sizeof (struct nfsd)); 448 crit_enter(); 449 nfsd->nfsd_td = td; 450 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain); 451 nfs_numnfsd++; 452 } else 453 crit_enter(); 454 455 /* 456 * Loop getting rpc requests until SIGKILL. 457 */ 458 for (;;) { 459 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) { 460 while (nfsd->nfsd_slp == NULL && 461 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 462 nfsd->nfsd_flag |= NFSD_WAITING; 463 nfsd_waiting++; 464 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0); 465 nfsd_waiting--; 466 if (error) 467 goto done; 468 } 469 if (nfsd->nfsd_slp == NULL && 470 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 471 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { 472 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC)) 473 == (SLP_VALID | SLP_DOREC)) { 474 slp->ns_flag &= ~SLP_DOREC; 475 slp->ns_sref++; 476 nfsd->nfsd_slp = slp; 477 break; 478 } 479 } 480 if (slp == 0) 481 nfsd_head_flag &= ~NFSD_CHECKSLP; 482 } 483 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0) 484 continue; 485 if (slp->ns_flag & SLP_VALID) { 486 if (slp->ns_flag & SLP_DISCONN) 487 nfsrv_zapsock(slp); 488 else if (slp->ns_flag & SLP_NEEDQ) { 489 slp->ns_flag &= ~SLP_NEEDQ; 490 (void) nfs_slplock(slp, 1); 491 nfsrv_rcv(slp->ns_so, (caddr_t)slp, 492 MB_WAIT); 493 nfs_slpunlock(slp); 494 } 495 error = nfsrv_dorec(slp, nfsd, &nd); 496 cur_usec = nfs_curusec(); 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 kfree((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 crit_exit(); 521 sotype = slp->ns_so->so_type; 522 if (nd) { 523 getmicrotime(&nd->nd_starttime); 524 if (nd->nd_nam2) 525 nd->nd_nam = nd->nd_nam2; 526 else 527 nd->nd_nam = slp->ns_nam; 528 529 /* 530 * Check to see if authorization is needed. 531 */ 532 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 533 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 534 nsd->nsd_haddr = 535 ((struct sockaddr_in *) 536 nd->nd_nam)->sin_addr.s_addr; 537 nsd->nsd_authlen = nfsd->nfsd_authlen; 538 nsd->nsd_verflen = nfsd->nfsd_verflen; 539 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr, 540 nfsd->nfsd_authlen) && 541 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr, 542 nfsd->nfsd_verflen) && 543 !copyout((caddr_t)nsd, argp, sizeof (*nsd))) 544 return (ENEEDAUTH); 545 cacherep = RC_DROPIT; 546 } else { 547 cacherep = nfsrv_getcache(nd, slp, &mreq); 548 } 549 550 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 551 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 552 nd->nd_procnum = NFSPROC_NOOP; 553 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 554 cacherep = RC_DOIT; 555 } else if (nfs_privport) { 556 /* Check if source port is privileged */ 557 u_short port; 558 struct sockaddr *nam = nd->nd_nam; 559 struct sockaddr_in *sin; 560 561 sin = (struct sockaddr_in *)nam; 562 port = ntohs(sin->sin_port); 563 if (port >= IPPORT_RESERVED && 564 nd->nd_procnum != NFSPROC_NULL) { 565 nd->nd_procnum = NFSPROC_NOOP; 566 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 567 cacherep = RC_DOIT; 568 kprintf("NFS request from unprivileged port (%s:%d)\n", 569 inet_ntoa(sin->sin_addr), port); 570 } 571 } 572 573 } 574 575 /* 576 * Loop to get all the write rpc relies that have been 577 * gathered together. 578 */ 579 do { 580 switch (cacherep) { 581 case RC_DOIT: 582 if (nd && (nd->nd_flag & ND_NFSV3)) 583 procrastinate = nfsrvw_procrastinate_v3; 584 else 585 procrastinate = nfsrvw_procrastinate; 586 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE && 587 procrastinate > 0) 588 ) { 589 error = nfsrv_writegather(&nd, slp, 590 nfsd->nfsd_td, &mreq); 591 } else { 592 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, 593 slp, nfsd->nfsd_td, &mreq); 594 } 595 if (mreq == NULL) 596 break; 597 if (error != 0 && error != NFSERR_RETVOID) { 598 if (nd->nd_procnum != NQNFSPROC_VACATED) 599 nfsstats.srv_errs++; 600 nfsrv_updatecache(nd, FALSE, mreq); 601 if (nd->nd_nam2) 602 FREE(nd->nd_nam2, M_SONAME); 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 kprintf("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, MB_WAIT); 628 if (m == NULL) 629 return (ENOBUFS); 630 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz); 631 } 632 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 633 (void) nfs_slplock(slp, 1); 634 if (slp->ns_flag & SLP_VALID) 635 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL); 636 else { 637 error = EPIPE; 638 m_freem(m); 639 } 640 if (nfsrtton) 641 nfsd_rt(sotype, nd, cacherep); 642 if (nd->nd_nam2) 643 FREE(nd->nd_nam2, M_SONAME); 644 if (nd->nd_mrep) 645 m_freem(nd->nd_mrep); 646 if (error == EPIPE) 647 nfsrv_zapsock(slp); 648 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED) 649 nfs_slpunlock(slp); 650 if (error == EINTR || error == ERESTART) { 651 kfree((caddr_t)nd, M_NFSRVDESC); 652 nfsrv_slpderef(slp); 653 crit_enter(); 654 goto done; 655 } 656 break; 657 case RC_DROPIT: 658 if (nfsrtton) 659 nfsd_rt(sotype, nd, cacherep); 660 m_freem(nd->nd_mrep); 661 if (nd->nd_nam2) 662 FREE(nd->nd_nam2, M_SONAME); 663 break; 664 }; 665 if (nd) { 666 FREE((caddr_t)nd, M_NFSRVDESC); 667 nd = NULL; 668 } 669 670 /* 671 * Check to see if there are outstanding writes that 672 * need to be serviced. 673 */ 674 cur_usec = nfs_curusec(); 675 crit_enter(); 676 if (slp->ns_tq.lh_first && 677 slp->ns_tq.lh_first->nd_time <= cur_usec) { 678 cacherep = RC_DOIT; 679 writes_todo = 1; 680 } else 681 writes_todo = 0; 682 crit_exit(); 683 } while (writes_todo); 684 crit_enter(); 685 if (nfsrv_dorec(slp, nfsd, &nd)) { 686 nfsd->nfsd_flag &= ~NFSD_REQINPROG; 687 nfsd->nfsd_slp = NULL; 688 nfsrv_slpderef(slp); 689 } 690 } 691 done: 692 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain); 693 crit_exit(); 694 kfree((caddr_t)nfsd, M_NFSD); 695 nsd->nsd_nfsd = (struct nfsd *)0; 696 if (--nfs_numnfsd == 0) 697 nfsrv_init(TRUE); /* Reinitialize everything */ 698 return (error); 699 } 700 701 /* 702 * Shut down a socket associated with an nfssvc_sock structure. 703 * Should be called with the send lock set, if required. 704 * The trick here is to increment the sref at the start, so that the nfsds 705 * will stop using it and clear ns_flag at the end so that it will not be 706 * reassigned during cleanup. 707 */ 708 static void 709 nfsrv_zapsock(struct nfssvc_sock *slp) 710 { 711 struct nfsuid *nuidp, *nnuidp; 712 struct nfsrv_descript *nwp, *nnwp; 713 struct socket *so; 714 struct file *fp; 715 struct nfsrv_rec *rec; 716 717 slp->ns_flag &= ~SLP_ALLFLAGS; 718 fp = slp->ns_fp; 719 if (fp) { 720 slp->ns_fp = (struct file *)0; 721 so = slp->ns_so; 722 so->so_rcv.ssb_flags &= ~SSB_UPCALL; 723 so->so_upcall = NULL; 724 so->so_upcallarg = NULL; 725 soshutdown(so, 2); 726 closef(fp, NULL); 727 if (slp->ns_nam) 728 FREE(slp->ns_nam, M_SONAME); 729 m_freem(slp->ns_raw); 730 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) { 731 --slp->ns_numrec; 732 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link); 733 if (rec->nr_address) 734 FREE(rec->nr_address, M_SONAME); 735 m_freem(rec->nr_packet); 736 kfree(rec, M_NFSRVDESC); 737 } 738 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru, 739 nnuidp) { 740 LIST_REMOVE(nuidp, nu_hash); 741 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 742 if (nuidp->nu_flag & NU_NAM) 743 FREE(nuidp->nu_nam, M_SONAME); 744 kfree((caddr_t)nuidp, M_NFSUID); 745 } 746 crit_enter(); 747 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) { 748 nnwp = nwp->nd_tq.le_next; 749 LIST_REMOVE(nwp, nd_tq); 750 kfree((caddr_t)nwp, M_NFSRVDESC); 751 } 752 LIST_INIT(&slp->ns_tq); 753 crit_exit(); 754 } 755 } 756 757 /* 758 * Derefence a server socket structure. If it has no more references and 759 * is no longer valid, you can throw it away. 760 */ 761 void 762 nfsrv_slpderef(struct nfssvc_sock *slp) 763 { 764 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) { 765 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 766 kfree((caddr_t)slp, M_NFSSVC); 767 } 768 } 769 770 /* 771 * Lock a socket against others. 772 */ 773 int 774 nfs_slplock(struct nfssvc_sock *slp, int wait) 775 { 776 int *statep = &slp->ns_solock; 777 778 if (!wait && (*statep & NFSSTA_SNDLOCK)) 779 return(0); /* already locked, fail */ 780 while (*statep & NFSSTA_SNDLOCK) { 781 *statep |= NFSSTA_WANTSND; 782 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0); 783 } 784 *statep |= NFSSTA_SNDLOCK; 785 return (1); 786 } 787 788 /* 789 * Unlock the stream socket for others. 790 */ 791 void 792 nfs_slpunlock(struct nfssvc_sock *slp) 793 { 794 int *statep = &slp->ns_solock; 795 796 if ((*statep & NFSSTA_SNDLOCK) == 0) 797 panic("nfs slpunlock"); 798 *statep &= ~NFSSTA_SNDLOCK; 799 if (*statep & NFSSTA_WANTSND) { 800 *statep &= ~NFSSTA_WANTSND; 801 wakeup((caddr_t)statep); 802 } 803 } 804 805 /* 806 * Initialize the data structures for the server. 807 * Handshake with any new nfsds starting up to avoid any chance of 808 * corruption. 809 */ 810 void 811 nfsrv_init(int terminating) 812 { 813 struct nfssvc_sock *slp, *nslp; 814 815 if (nfssvc_sockhead_flag & SLP_INIT) 816 panic("nfsd init"); 817 nfssvc_sockhead_flag |= SLP_INIT; 818 if (terminating) { 819 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) { 820 if (slp->ns_flag & SLP_VALID) 821 nfsrv_zapsock(slp); 822 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 823 kfree((caddr_t)slp, M_NFSSVC); 824 } 825 nfsrv_cleancache(); /* And clear out server cache */ 826 } else 827 nfs_pub.np_valid = 0; 828 829 TAILQ_INIT(&nfssvc_sockhead); 830 nfssvc_sockhead_flag &= ~SLP_INIT; 831 if (nfssvc_sockhead_flag & SLP_WANTINIT) { 832 nfssvc_sockhead_flag &= ~SLP_WANTINIT; 833 wakeup((caddr_t)&nfssvc_sockhead); 834 } 835 836 TAILQ_INIT(&nfsd_head); 837 nfsd_head_flag &= ~NFSD_CHECKSLP; 838 839 #if 0 840 nfs_udpsock = (struct nfssvc_sock *) 841 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 842 bzero((caddr_t)nfs_udpsock, sizeof (struct nfssvc_sock)); 843 STAILQ_INIT(&nfs_udpsock->ns_rec); 844 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead); 845 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain); 846 847 nfs_cltpsock = (struct nfssvc_sock *) 848 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK); 849 bzero((caddr_t)nfs_cltpsock, sizeof (struct nfssvc_sock)); 850 STAILQ_INIT(&nfs_cltpsock->ns_rec); 851 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead); 852 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain); 853 #endif 854 } 855 856 /* 857 * Add entries to the server monitor log. 858 */ 859 static void 860 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 861 { 862 struct drt *rt; 863 864 rt = &nfsdrt.drt[nfsdrt.pos]; 865 if (cacherep == RC_DOIT) 866 rt->flag = 0; 867 else if (cacherep == RC_REPLY) 868 rt->flag = DRT_CACHEREPLY; 869 else 870 rt->flag = DRT_CACHEDROP; 871 if (sotype == SOCK_STREAM) 872 rt->flag |= DRT_TCP; 873 if (nd->nd_flag & ND_NFSV3) 874 rt->flag |= DRT_NFSV3; 875 rt->proc = nd->nd_procnum; 876 if (nd->nd_nam->sa_family == AF_INET) 877 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr; 878 else 879 rt->ipadr = INADDR_ANY; 880 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec); 881 getmicrotime(&rt->tstamp); 882 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 883 } 884 #endif /* NFS_NOSERVER */ 885 886 static int nfs_defect = 0; 887 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, ""); 888 889 /* 890 * Asynchronous I/O daemons for client nfs. 891 * They do read-ahead and write-behind operations on the block I/O cache. 892 * Never returns unless it fails or gets killed. 893 */ 894 static int 895 nfssvc_iod(struct thread *td) 896 { 897 struct bio *bio; 898 int i, myiod; 899 struct nfsmount *nmp; 900 int error = 0; 901 902 /* 903 * Assign my position or return error if too many already running 904 */ 905 myiod = -1; 906 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) 907 if (nfs_asyncdaemon[i] == 0) { 908 nfs_asyncdaemon[i]++; 909 myiod = i; 910 break; 911 } 912 if (myiod == -1) 913 return (EBUSY); 914 nfs_numasync++; 915 /* 916 * Just loop around doin our stuff until SIGKILL 917 */ 918 for (;;) { 919 while (((nmp = nfs_iodmount[myiod]) == NULL 920 || TAILQ_EMPTY(&nmp->nm_bioq)) 921 && error == 0) { 922 if (nmp) 923 nmp->nm_bioqiods--; 924 nfs_iodwant[myiod] = td; 925 nfs_iodmount[myiod] = NULL; 926 error = tsleep((caddr_t)&nfs_iodwant[myiod], 927 PCATCH, "nfsidl", 0); 928 } 929 if (error) { 930 nfs_asyncdaemon[myiod] = 0; 931 if (nmp) 932 nmp->nm_bioqiods--; 933 nfs_iodwant[myiod] = NULL; 934 nfs_iodmount[myiod] = NULL; 935 nfs_numasync--; 936 return (error); 937 } 938 while ((bio = TAILQ_FIRST(&nmp->nm_bioq)) != NULL) { 939 /* 940 * Take one off the front of the list. The BIO's 941 * block number is normalized for DEV_BSIZE. 942 */ 943 TAILQ_REMOVE(&nmp->nm_bioq, bio, bio_act); 944 nmp->nm_bioqlen--; 945 if (nmp->nm_bioqwant && nmp->nm_bioqlen <= nfs_numasync) { 946 nmp->nm_bioqwant = FALSE; 947 wakeup(&nmp->nm_bioq); 948 } 949 nfs_doio((struct vnode *)bio->bio_driver_info, bio, NULL); 950 /* 951 * If there are more than one iod on this mount, then defect 952 * so that the iods can be shared out fairly between the mounts 953 */ 954 if (nfs_defect && nmp->nm_bioqiods > 1) { 955 NFS_DPF(ASYNCIO, 956 ("nfssvc_iod: iod %d defecting from mount %p\n", 957 myiod, nmp)); 958 nfs_iodmount[myiod] = NULL; 959 nmp->nm_bioqiods--; 960 break; 961 } 962 } 963 } 964 } 965 966 967 /* 968 * Get an authorization string for the uid by having the mount_nfs sitting 969 * on this mount point porpous out of the kernel and do it. 970 */ 971 int 972 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, 973 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str, 974 int *verf_len, NFSKERBKEY_T key /* return session key */) 975 { 976 int error = 0; 977 978 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) { 979 nmp->nm_state |= NFSSTA_WANTAUTH; 980 (void) tsleep((caddr_t)&nmp->nm_authtype, 0, 981 "nfsauth1", 2 * hz); 982 error = nfs_sigintr(nmp, rep, rep->r_td); 983 if (error) { 984 nmp->nm_state &= ~NFSSTA_WANTAUTH; 985 return (error); 986 } 987 } 988 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH); 989 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK); 990 nmp->nm_authlen = RPCAUTH_MAXSIZ; 991 nmp->nm_verfstr = verf_str; 992 nmp->nm_verflen = *verf_len; 993 nmp->nm_authuid = cred->cr_uid; 994 wakeup((caddr_t)&nmp->nm_authstr); 995 996 /* 997 * And wait for mount_nfs to do its stuff. 998 */ 999 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) { 1000 (void) tsleep((caddr_t)&nmp->nm_authlen, 0, 1001 "nfsauth2", 2 * hz); 1002 error = nfs_sigintr(nmp, rep, rep->r_td); 1003 } 1004 if (nmp->nm_state & NFSSTA_AUTHERR) { 1005 nmp->nm_state &= ~NFSSTA_AUTHERR; 1006 error = EAUTH; 1007 } 1008 if (error) 1009 kfree((caddr_t)*auth_str, M_TEMP); 1010 else { 1011 *auth_len = nmp->nm_authlen; 1012 *verf_len = nmp->nm_verflen; 1013 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key)); 1014 } 1015 nmp->nm_state &= ~NFSSTA_HASAUTH; 1016 nmp->nm_state |= NFSSTA_WAITAUTH; 1017 if (nmp->nm_state & NFSSTA_WANTAUTH) { 1018 nmp->nm_state &= ~NFSSTA_WANTAUTH; 1019 wakeup((caddr_t)&nmp->nm_authtype); 1020 } 1021 return (error); 1022 } 1023 1024 /* 1025 * Get a nickname authenticator and verifier. 1026 */ 1027 int 1028 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str, 1029 int *auth_len, char *verf_str, int verf_len) 1030 { 1031 struct nfsuid *nuidp; 1032 u_int32_t *nickp, *verfp; 1033 struct timeval ktvin, ktvout; 1034 1035 #ifdef DIAGNOSTIC 1036 if (verf_len < (4 * NFSX_UNSIGNED)) 1037 panic("nfs_getnickauth verf too small"); 1038 #endif 1039 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first; 1040 nuidp != 0; nuidp = nuidp->nu_hash.le_next) { 1041 if (nuidp->nu_cr.cr_uid == cred->cr_uid) 1042 break; 1043 } 1044 if (!nuidp || nuidp->nu_expire < time_second) 1045 return (EACCES); 1046 1047 /* 1048 * Move to the end of the lru list (end of lru == most recently used). 1049 */ 1050 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru); 1051 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru); 1052 1053 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK); 1054 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME); 1055 *nickp = txdr_unsigned(nuidp->nu_nickname); 1056 *auth_str = (char *)nickp; 1057 *auth_len = 2 * NFSX_UNSIGNED; 1058 1059 /* 1060 * Now we must encrypt the verifier and package it up. 1061 */ 1062 verfp = (u_int32_t *)verf_str; 1063 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME); 1064 if (time_second > nuidp->nu_timestamp.tv_sec || 1065 (time_second == nuidp->nu_timestamp.tv_sec && 1066 time_second > nuidp->nu_timestamp.tv_usec)) 1067 getmicrotime(&nuidp->nu_timestamp); 1068 else 1069 nuidp->nu_timestamp.tv_usec++; 1070 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec); 1071 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); 1072 1073 /* 1074 * Now encrypt the timestamp verifier in ecb mode using the session 1075 * key. 1076 */ 1077 #ifdef NFSKERB 1078 XXX 1079 #endif 1080 1081 *verfp++ = ktvout.tv_sec; 1082 *verfp++ = ktvout.tv_usec; 1083 *verfp = 0; 1084 return (0); 1085 } 1086 1087 /* 1088 * Save the current nickname in a hash list entry on the mount point. 1089 */ 1090 int 1091 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len, 1092 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, 1093 struct mbuf *mrep) 1094 { 1095 struct nfsuid *nuidp; 1096 u_int32_t *tl; 1097 int32_t t1; 1098 struct mbuf *md = *mdp; 1099 struct timeval ktvin, ktvout; 1100 u_int32_t nick; 1101 char *dpos = *dposp, *cp2; 1102 int deltasec, error = 0; 1103 1104 if (len == (3 * NFSX_UNSIGNED)) { 1105 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1106 ktvin.tv_sec = *tl++; 1107 ktvin.tv_usec = *tl++; 1108 nick = fxdr_unsigned(u_int32_t, *tl); 1109 1110 /* 1111 * Decrypt the timestamp in ecb mode. 1112 */ 1113 #ifdef NFSKERB 1114 XXX 1115 #endif 1116 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec); 1117 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec); 1118 deltasec = time_second - ktvout.tv_sec; 1119 if (deltasec < 0) 1120 deltasec = -deltasec; 1121 /* 1122 * If ok, add it to the hash list for the mount point. 1123 */ 1124 if (deltasec <= NFS_KERBCLOCKSKEW) { 1125 if (nmp->nm_numuids < nuidhash_max) { 1126 nmp->nm_numuids++; 1127 nuidp = (struct nfsuid *) 1128 kmalloc(sizeof (struct nfsuid), M_NFSUID, 1129 M_WAITOK); 1130 } else { 1131 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead); 1132 LIST_REMOVE(nuidp, nu_hash); 1133 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, 1134 nu_lru); 1135 } 1136 nuidp->nu_flag = 0; 1137 nuidp->nu_cr.cr_uid = cred->cr_uid; 1138 nuidp->nu_expire = time_second + NFS_KERBTTL; 1139 nuidp->nu_timestamp = ktvout; 1140 nuidp->nu_nickname = nick; 1141 bcopy(key, nuidp->nu_key, sizeof (key)); 1142 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, 1143 nu_lru); 1144 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid), 1145 nuidp, nu_hash); 1146 } 1147 } else 1148 nfsm_adv(nfsm_rndup(len)); 1149 nfsmout: 1150 *mdp = md; 1151 *dposp = dpos; 1152 return (error); 1153 } 1154