1 /* $NetBSD: nfs_syscalls.c,v 1.160 2018/03/16 17:25:04 christos Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95 35 */ 36 37 #include <sys/cdefs.h> 38 __KERNEL_RCSID(0, "$NetBSD: nfs_syscalls.c,v 1.160 2018/03/16 17:25:04 christos Exp $"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/file.h> 44 #include <sys/stat.h> 45 #include <sys/vnode.h> 46 #include <sys/mount.h> 47 #include <sys/proc.h> 48 #include <sys/uio.h> 49 #include <sys/malloc.h> 50 #include <sys/kmem.h> 51 #include <sys/buf.h> 52 #include <sys/mbuf.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/signalvar.h> 56 #include <sys/domain.h> 57 #include <sys/protosw.h> 58 #include <sys/namei.h> 59 #include <sys/syslog.h> 60 #include <sys/filedesc.h> 61 #include <sys/kthread.h> 62 #include <sys/kauth.h> 63 #include <sys/syscallargs.h> 64 #include <sys/cprng.h> 65 #include <sys/rbtree.h> 66 67 #include <netinet/in.h> 68 #include <netinet/tcp.h> 69 #include <nfs/xdr_subs.h> 70 #include <nfs/rpcv2.h> 71 #include <nfs/nfsproto.h> 72 #include <nfs/nfs.h> 73 #include <nfs/nfsm_subs.h> 74 #include <nfs/nfsrvcache.h> 75 #include <nfs/nfsmount.h> 76 #include <nfs/nfsnode.h> 77 #include <nfs/nfsrtt.h> 78 #include <nfs/nfs_var.h> 79 80 extern int32_t (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *, 81 struct nfssvc_sock *, 82 struct lwp *, struct mbuf **); 83 extern int nfsrvw_procrastinate; 84 extern int nuidhash_max; 85 86 static int nfs_numnfsd = 0; 87 static struct nfsdrt nfsdrt; 88 kmutex_t nfsd_lock; 89 struct nfssvc_sockhead nfssvc_sockhead; 90 kcondvar_t nfsd_initcv; 91 struct nfssvc_sockhead nfssvc_sockpending; 92 struct nfsdidlehead nfsd_idle_head; 93 94 static rb_tree_t nfsd_tree; 95 static const rb_tree_ops_t nfsd_tree_ops; 96 97 int nfssvc_sockhead_flag; 98 int nfsd_head_flag; 99 100 struct nfssvc_sock *nfs_udpsock; 101 struct nfssvc_sock *nfs_udp6sock; 102 103 static struct nfssvc_sock *nfsrv_sockalloc(void); 104 static void nfsrv_sockfree(struct nfssvc_sock *); 105 static void nfsd_rt(int, struct nfsrv_descript *, int); 106 static int nfssvc_nfsd(struct nfssvc_copy_ops *, struct nfsd_srvargs *, void *, 107 struct lwp *); 108 109 static int nfsd_compare_nodes(void *, const void *, const void *); 110 static int nfsd_compare_key(void *, const void *, const void *); 111 112 static struct nfsd *nfsd_bake_cookie(struct nfsd *); 113 static void nfsd_toss_cookie(struct nfsd *); 114 static struct nfsd *nfsd_get(struct nfsd *); 115 116 static int nfssvc_addsock_in(struct nfsd_args *, const void *); 117 static int nfssvc_setexports_in(struct mountd_exports_list *, const void *); 118 static int nfssvc_nsd_in(struct nfsd_srvargs *, const void *); 119 static int nfssvc_nsd_out(void *, const struct nfsd_srvargs *); 120 static int nfssvc_exp_in(struct export_args *, const void *, size_t); 121 122 static const rb_tree_ops_t nfsd_tree_ops = { 123 .rbto_compare_nodes = nfsd_compare_nodes, 124 .rbto_compare_key = nfsd_compare_key, 125 .rbto_node_offset = offsetof(struct nfsd, nfsd_node), 126 }; 127 128 static int 129 nfsd_compare_nodes(void *cookie, const void *va, const void *vb) 130 { 131 const struct nfsd *na = va; 132 const struct nfsd *nb = vb; 133 134 if (na->nfsd_cookie < nb->nfsd_cookie) 135 return -1; 136 if (na->nfsd_cookie > nb->nfsd_cookie) 137 return +1; 138 return 0; 139 } 140 141 static int 142 nfsd_compare_key(void *cookie, const void *vn, const void *vk) 143 { 144 const struct nfsd *n = vn; 145 const uint32_t *k = vk; 146 147 if (n->nfsd_cookie < *k) 148 return -1; 149 if (n->nfsd_cookie > *k) 150 return +1; 151 return 0; 152 } 153 154 /* 155 * nfsd_bake_cookie(nfsd) 156 * 157 * Bake a cookie for nfsd, hang it on the tree of nfsds, and 158 * return a userland-safe pointer nfsdu neatly packed for 159 * transport in struct nfsd_srvargs::nsd_nfsd. 160 */ 161 static struct nfsd * 162 nfsd_bake_cookie(struct nfsd *nfsd) 163 { 164 165 KASSERT(mutex_owned(&nfsd_lock)); 166 167 do { 168 nfsd->nfsd_cookie = cprng_fast32(); 169 } while (nfsd->nfsd_cookie == 0 || 170 rb_tree_insert_node(&nfsd_tree, nfsd) != nfsd); 171 172 return (struct nfsd *)(uintptr_t)nfsd->nfsd_cookie; 173 } 174 175 /* 176 * nfsd_toss_cookie(nfsd) 177 * 178 * Toss nfsd's cookie. 179 */ 180 static void 181 nfsd_toss_cookie(struct nfsd *nfsd) 182 { 183 184 KASSERT(mutex_owned(&nfsd_lock)); 185 KASSERT(nfsd->nfsd_cookie != 0); 186 187 rb_tree_remove_node(&nfsd_tree, nfsd); 188 nfsd->nfsd_cookie = 0; /* paranoia */ 189 } 190 191 /* 192 * nfsd_get(nfsdu) 193 * 194 * Return the struct nfsd pointer for the userland nfsdu cookie, 195 * as stored in struct nfsd_srvargs::nsd_nfsd, or NULL if nfsdu is 196 * not a current valid nfsd cookie. 197 * 198 * Caller MUST NOT hold nfsd_lock. Caller MUST NOT pass (struct 199 * nfsd *)(uintptr_t)0, which is the sentinel value for no nfsd 200 * cookie, for which the caller should check first. 201 */ 202 static struct nfsd * 203 nfsd_get(struct nfsd *nfsdu) 204 { 205 uintptr_t cookie = (uintptr_t)nfsdu; 206 uint32_t key; 207 struct nfsd *nfsd; 208 209 KASSERT(cookie != 0); 210 if (cookie > UINT32_MAX) 211 return NULL; 212 key = cookie; 213 214 mutex_enter(&nfsd_lock); 215 nfsd = rb_tree_find_node(&nfsd_tree, &key); 216 mutex_exit(&nfsd_lock); 217 218 return nfsd; 219 } 220 221 static int 222 nfssvc_addsock_in(struct nfsd_args *nfsdarg, const void *argp) 223 { 224 225 return copyin(argp, nfsdarg, sizeof *nfsdarg); 226 } 227 228 static int 229 nfssvc_setexports_in(struct mountd_exports_list *mel, const void *argp) 230 { 231 232 return copyin(argp, mel, sizeof *mel); 233 } 234 235 static int 236 nfssvc_nsd_in(struct nfsd_srvargs *nsd, const void *argp) 237 { 238 239 return copyin(argp, nsd, sizeof *nsd); 240 } 241 242 static int 243 nfssvc_nsd_out(void *argp, const struct nfsd_srvargs *nsd) 244 { 245 246 return copyout(nsd, argp, sizeof *nsd); 247 } 248 249 static int 250 nfssvc_exp_in(struct export_args *exp, const void *argp, size_t nexports) 251 { 252 253 return copyin(argp, exp, sizeof(*exp) * nexports); 254 } 255 256 /* 257 * NFS server system calls 258 */ 259 260 static struct nfssvc_copy_ops native_ops = { 261 .addsock_in = nfssvc_addsock_in, 262 .setexports_in = nfssvc_setexports_in, 263 .nsd_in = nfssvc_nsd_in, 264 .nsd_out = nfssvc_nsd_out, 265 .exp_in = nfssvc_exp_in, 266 }; 267 268 /* 269 * Nfs server pseudo system call for the nfsd's 270 * Based on the flag value it either: 271 * - adds a socket to the selection list 272 * - remains in the kernel as an nfsd 273 * - remains in the kernel as an nfsiod 274 */ 275 276 int 277 sys_nfssvc(struct lwp *l, const struct sys_nfssvc_args *uap, register_t *retval) 278 { 279 /* { 280 syscallarg(int) flag; 281 syscallarg(void *) argp; 282 } */ 283 int flag = SCARG(uap, flag); 284 void *argp = SCARG(uap, argp); 285 286 return do_nfssvc(&native_ops, l, flag, argp, retval); 287 } 288 289 int 290 do_nfssvc(struct nfssvc_copy_ops *ops, struct lwp *l, int flag, void *argp, register_t *retval) 291 { 292 int error; 293 file_t *fp; 294 struct mbuf *nam; 295 struct nfsd_args nfsdarg; 296 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs; 297 struct nfsd *nfsd = NULL; 298 struct nfssvc_sock *slp; 299 struct nfsuid *nuidp; 300 301 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_NFS, 302 KAUTH_REQ_NETWORK_NFS_SVC, NULL, NULL, NULL); 303 if (error) 304 return (error); 305 306 mutex_enter(&nfsd_lock); 307 while (nfssvc_sockhead_flag & SLP_INIT) { 308 cv_wait(&nfsd_initcv, &nfsd_lock); 309 } 310 mutex_exit(&nfsd_lock); 311 312 if (flag & NFSSVC_BIOD) { 313 /* Dummy implementation of nfsios for 1.4 and earlier. */ 314 error = kpause("nfsbiod", true, 0, NULL); 315 } else if (flag & NFSSVC_MNTD) { 316 error = ENOSYS; 317 } else if (flag & NFSSVC_ADDSOCK) { 318 error = ops->addsock_in(&nfsdarg, argp); 319 if (error) 320 return (error); 321 /* getsock() will use the descriptor for us */ 322 if ((fp = fd_getfile(nfsdarg.sock)) == NULL) 323 return (EBADF); 324 if (fp->f_type != DTYPE_SOCKET) { 325 fd_putfile(nfsdarg.sock); 326 return (ENOTSOCK); 327 } 328 /* 329 * Get the client address for connected sockets. 330 */ 331 if (nfsdarg.name == NULL || nfsdarg.namelen == 0) 332 nam = (struct mbuf *)0; 333 else { 334 error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen, 335 UIO_USERSPACE, MT_SONAME); 336 if (error) { 337 fd_putfile(nfsdarg.sock); 338 return (error); 339 } 340 } 341 error = nfssvc_addsock(fp, nam); 342 fd_putfile(nfsdarg.sock); 343 } else if (flag & NFSSVC_SETEXPORTSLIST) { 344 struct export_args *args; 345 struct mountd_exports_list mel; 346 347 error = ops->setexports_in(&mel, argp); 348 if (error != 0) 349 return error; 350 351 args = (struct export_args *)malloc(mel.mel_nexports * 352 sizeof(struct export_args), M_TEMP, M_WAITOK); 353 error = ops->exp_in(args, mel.mel_exports, mel.mel_nexports); 354 if (error != 0) { 355 free(args, M_TEMP); 356 return error; 357 } 358 mel.mel_exports = args; 359 360 error = mountd_set_exports_list(&mel, l, NULL); 361 362 free(args, M_TEMP); 363 } else { 364 error = ops->nsd_in(nsd, argp); 365 if (error) 366 return (error); 367 if ((uintptr_t)nsd->nsd_nfsd != 0 && 368 (nfsd = nfsd_get(nsd->nsd_nfsd)) == NULL) 369 return (EINVAL); 370 if ((flag & NFSSVC_AUTHIN) && 371 nfsd != NULL && 372 (nfsd->nfsd_slp->ns_flags & SLP_VALID)) { 373 slp = nfsd->nfsd_slp; 374 375 /* 376 * First check to see if another nfsd has already 377 * added this credential. 378 */ 379 LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid), 380 nu_hash) { 381 if (kauth_cred_geteuid(nuidp->nu_cr) == 382 nsd->nsd_cr.cr_uid && 383 (!nfsd->nfsd_nd->nd_nam2 || 384 netaddr_match(NU_NETFAM(nuidp), 385 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 386 break; 387 } 388 if (nuidp) { 389 kauth_cred_hold(nuidp->nu_cr); 390 nfsd->nfsd_nd->nd_cr = nuidp->nu_cr; 391 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 392 } else { 393 /* 394 * Nope, so we will. 395 */ 396 if (slp->ns_numuids < nuidhash_max) { 397 slp->ns_numuids++; 398 nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP); 399 } else 400 nuidp = (struct nfsuid *)0; 401 if ((slp->ns_flags & SLP_VALID) == 0) { 402 if (nuidp) 403 kmem_free(nuidp, sizeof(*nuidp)); 404 } else { 405 if (nuidp == (struct nfsuid *)0) { 406 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 407 LIST_REMOVE(nuidp, nu_hash); 408 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 409 nu_lru); 410 if (nuidp->nu_flag & NU_NAM) 411 m_freem(nuidp->nu_nam); 412 } 413 nuidp->nu_flag = 0; 414 kauth_uucred_to_cred(nuidp->nu_cr, 415 &nsd->nsd_cr); 416 nuidp->nu_timestamp = nsd->nsd_timestamp; 417 nuidp->nu_expire = time_second + nsd->nsd_ttl; 418 /* 419 * and save the session key in nu_key. 420 */ 421 memcpy(nuidp->nu_key, nsd->nsd_key, 422 sizeof(nsd->nsd_key)); 423 if (nfsd->nfsd_nd->nd_nam2) { 424 struct sockaddr_in *saddr; 425 426 saddr = mtod(nfsd->nfsd_nd->nd_nam2, 427 struct sockaddr_in *); 428 switch (saddr->sin_family) { 429 case AF_INET: 430 nuidp->nu_flag |= NU_INETADDR; 431 nuidp->nu_inetaddr = 432 saddr->sin_addr.s_addr; 433 break; 434 case AF_INET6: 435 nuidp->nu_flag |= NU_NAM; 436 nuidp->nu_nam = m_copym( 437 nfsd->nfsd_nd->nd_nam2, 0, 438 M_COPYALL, M_WAIT); 439 break; 440 default: 441 kmem_free(nuidp, sizeof(*nuidp)); 442 return EAFNOSUPPORT; 443 }; 444 } 445 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 446 nu_lru); 447 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 448 nuidp, nu_hash); 449 kauth_cred_hold(nuidp->nu_cr); 450 nfsd->nfsd_nd->nd_cr = nuidp->nu_cr; 451 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 452 } 453 } 454 } 455 if ((flag & NFSSVC_AUTHINFAIL) && 456 nfsd != NULL) 457 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 458 error = nfssvc_nfsd(ops, nsd, argp, l); 459 } 460 if (error == EINTR || error == ERESTART) 461 error = 0; 462 return (error); 463 } 464 465 static struct nfssvc_sock * 466 nfsrv_sockalloc(void) 467 { 468 struct nfssvc_sock *slp; 469 470 slp = kmem_alloc(sizeof(*slp), KM_SLEEP); 471 memset(slp, 0, sizeof (struct nfssvc_sock)); 472 mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET); 473 mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET); 474 cv_init(&slp->ns_cv, "nfsdsock"); 475 TAILQ_INIT(&slp->ns_uidlruhead); 476 LIST_INIT(&slp->ns_tq); 477 SIMPLEQ_INIT(&slp->ns_sendq); 478 mutex_enter(&nfsd_lock); 479 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 480 mutex_exit(&nfsd_lock); 481 482 return slp; 483 } 484 485 static void 486 nfsrv_sockfree(struct nfssvc_sock *slp) 487 { 488 489 KASSERT(slp->ns_so == NULL); 490 KASSERT(slp->ns_fp == NULL); 491 KASSERT((slp->ns_flags & SLP_VALID) == 0); 492 mutex_destroy(&slp->ns_lock); 493 mutex_destroy(&slp->ns_alock); 494 cv_destroy(&slp->ns_cv); 495 kmem_free(slp, sizeof(*slp)); 496 } 497 498 /* 499 * Adds a socket to the list for servicing by nfsds. 500 */ 501 int 502 nfssvc_addsock(file_t *fp, struct mbuf *mynam) 503 { 504 int siz; 505 struct nfssvc_sock *slp; 506 struct socket *so; 507 struct nfssvc_sock *tslp; 508 int error; 509 int val; 510 511 so = fp->f_socket; 512 tslp = (struct nfssvc_sock *)0; 513 /* 514 * Add it to the list, as required. 515 */ 516 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 517 if (so->so_proto->pr_domain->dom_family == AF_INET6) 518 tslp = nfs_udp6sock; 519 else { 520 tslp = nfs_udpsock; 521 if (tslp->ns_flags & SLP_VALID) { 522 m_freem(mynam); 523 return (EPERM); 524 } 525 } 526 } 527 if (so->so_type == SOCK_STREAM) 528 siz = NFS_MAXPACKET + sizeof (u_long); 529 else 530 siz = NFS_MAXPACKET; 531 solock(so); 532 error = soreserve(so, siz, siz); 533 sounlock(so); 534 if (error) { 535 m_freem(mynam); 536 return (error); 537 } 538 539 /* 540 * Set protocol specific options { for now TCP only } and 541 * reserve some space. For datagram sockets, this can get called 542 * repeatedly for the same socket, but that isn't harmful. 543 */ 544 if (so->so_type == SOCK_STREAM) { 545 val = 1; 546 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 547 sizeof(val)); 548 } 549 if ((so->so_proto->pr_domain->dom_family == AF_INET || 550 so->so_proto->pr_domain->dom_family == AF_INET6) && 551 so->so_proto->pr_protocol == IPPROTO_TCP) { 552 val = 1; 553 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 554 sizeof(val)); 555 } 556 solock(so); 557 so->so_rcv.sb_flags &= ~SB_NOINTR; 558 so->so_rcv.sb_timeo = 0; 559 so->so_snd.sb_flags &= ~SB_NOINTR; 560 so->so_snd.sb_timeo = 0; 561 sounlock(so); 562 if (tslp) { 563 slp = tslp; 564 } else { 565 slp = nfsrv_sockalloc(); 566 } 567 slp->ns_so = so; 568 slp->ns_nam = mynam; 569 mutex_enter(&fp->f_lock); 570 fp->f_count++; 571 mutex_exit(&fp->f_lock); 572 slp->ns_fp = fp; 573 slp->ns_flags = SLP_VALID; 574 slp->ns_aflags = SLP_A_NEEDQ; 575 slp->ns_gflags = 0; 576 slp->ns_sflags = 0; 577 solock(so); 578 so->so_upcallarg = (void *)slp; 579 so->so_upcall = nfsrv_soupcall; 580 so->so_rcv.sb_flags |= SB_UPCALL; 581 sounlock(so); 582 nfsrv_wakenfsd(slp); 583 return (0); 584 } 585 586 /* 587 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 588 * until it is killed by a signal. 589 */ 590 static int 591 nfssvc_nfsd(struct nfssvc_copy_ops *ops, struct nfsd_srvargs *nsd, 592 void *argp, struct lwp *l) 593 { 594 struct timeval tv; 595 struct mbuf *m; 596 struct nfssvc_sock *slp; 597 struct nfsd *nfsd; 598 struct nfsrv_descript *nd = NULL; 599 struct mbuf *mreq; 600 u_quad_t cur_usec; 601 int error = 0, cacherep, siz, sotype, writes_todo; 602 struct proc *p = l->l_proc; 603 bool doreinit; 604 605 #ifndef nolint 606 cacherep = RC_DOIT; 607 writes_todo = 0; 608 #endif 609 /* 610 * If userland didn't provide an nfsd cookie, bake a fresh one; 611 * if they did provide one, look it up. 612 */ 613 if ((uintptr_t)nsd->nsd_nfsd == 0) { 614 nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP); 615 memset(nfsd, 0, sizeof (struct nfsd)); 616 cv_init(&nfsd->nfsd_cv, "nfsd"); 617 nfsd->nfsd_procp = p; 618 mutex_enter(&nfsd_lock); 619 while ((nfssvc_sockhead_flag & SLP_INIT) != 0) { 620 KASSERT(nfs_numnfsd == 0); 621 cv_wait(&nfsd_initcv, &nfsd_lock); 622 } 623 nsd->nsd_nfsd = nfsd_bake_cookie(nfsd); 624 nfs_numnfsd++; 625 mutex_exit(&nfsd_lock); 626 } else if ((nfsd = nfsd_get(nsd->nsd_nfsd)) == NULL) { 627 return (EINVAL); 628 } 629 KASSERT(nfsd != NULL); 630 KASSERT(nsd->nsd_nfsd != (struct nfsd *)(uintptr_t)0); 631 632 /* 633 * Loop getting rpc requests until SIGKILL. 634 */ 635 for (;;) { 636 bool dummy; 637 638 if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD) 639 != 0) { 640 preempt(); 641 } 642 if (nfsd->nfsd_slp == NULL) { 643 mutex_enter(&nfsd_lock); 644 while (nfsd->nfsd_slp == NULL && 645 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 646 SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd, 647 nfsd_idle); 648 error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock); 649 if (error) { 650 slp = nfsd->nfsd_slp; 651 nfsd->nfsd_slp = NULL; 652 if (!slp) 653 SLIST_REMOVE(&nfsd_idle_head, 654 nfsd, nfsd, nfsd_idle); 655 mutex_exit(&nfsd_lock); 656 if (slp) { 657 nfsrv_wakenfsd(slp); 658 nfsrv_slpderef(slp); 659 } 660 goto done; 661 } 662 } 663 if (nfsd->nfsd_slp == NULL && 664 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 665 slp = TAILQ_FIRST(&nfssvc_sockpending); 666 if (slp) { 667 KASSERT((slp->ns_gflags & SLP_G_DOREC) 668 != 0); 669 TAILQ_REMOVE(&nfssvc_sockpending, slp, 670 ns_pending); 671 slp->ns_gflags &= ~SLP_G_DOREC; 672 slp->ns_sref++; 673 nfsd->nfsd_slp = slp; 674 } else 675 nfsd_head_flag &= ~NFSD_CHECKSLP; 676 } 677 KASSERT(nfsd->nfsd_slp == NULL || 678 nfsd->nfsd_slp->ns_sref > 0); 679 mutex_exit(&nfsd_lock); 680 if ((slp = nfsd->nfsd_slp) == NULL) 681 continue; 682 if (slp->ns_flags & SLP_VALID) { 683 bool more; 684 685 if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) { 686 nfsrv_rcv(slp); 687 } 688 if (nfsdsock_testbits(slp, SLP_A_DISCONN)) { 689 nfsrv_zapsock(slp); 690 } 691 error = nfsrv_dorec(slp, nfsd, &nd, &more); 692 getmicrotime(&tv); 693 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 694 (u_quad_t)tv.tv_usec; 695 writes_todo = 0; 696 if (error) { 697 struct nfsrv_descript *nd2; 698 699 mutex_enter(&nfsd_lock); 700 nd2 = LIST_FIRST(&slp->ns_tq); 701 if (nd2 != NULL && 702 nd2->nd_time <= cur_usec) { 703 error = 0; 704 cacherep = RC_DOIT; 705 writes_todo = 1; 706 } 707 mutex_exit(&nfsd_lock); 708 } 709 if (error == 0 && more) { 710 nfsrv_wakenfsd(slp); 711 } 712 } 713 } else { 714 error = 0; 715 slp = nfsd->nfsd_slp; 716 } 717 KASSERT(slp != NULL); 718 KASSERT(nfsd->nfsd_slp == slp); 719 if (error || (slp->ns_flags & SLP_VALID) == 0) { 720 if (nd) { 721 nfsdreq_free(nd); 722 nd = NULL; 723 } 724 nfsd->nfsd_slp = NULL; 725 nfsrv_slpderef(slp); 726 continue; 727 } 728 sotype = slp->ns_so->so_type; 729 if (nd) { 730 getmicrotime(&nd->nd_starttime); 731 if (nd->nd_nam2) 732 nd->nd_nam = nd->nd_nam2; 733 else 734 nd->nd_nam = slp->ns_nam; 735 736 /* 737 * Check to see if authorization is needed. 738 */ 739 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 740 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 741 nsd->nsd_haddr = mtod(nd->nd_nam, 742 struct sockaddr_in *)->sin_addr.s_addr; 743 nsd->nsd_authlen = nfsd->nfsd_authlen; 744 nsd->nsd_verflen = nfsd->nfsd_verflen; 745 if (!copyout(nfsd->nfsd_authstr, 746 nsd->nsd_authstr, nfsd->nfsd_authlen) && 747 !copyout(nfsd->nfsd_verfstr, 748 nsd->nsd_verfstr, nfsd->nfsd_verflen) && 749 !ops->nsd_out(argp, nsd)) { 750 return (ENEEDAUTH); 751 } 752 cacherep = RC_DROPIT; 753 } else 754 cacherep = nfsrv_getcache(nd, slp, &mreq); 755 756 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 757 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 758 nd->nd_procnum = NFSPROC_NOOP; 759 nd->nd_repstat = 760 (NFSERR_AUTHERR | AUTH_TOOWEAK); 761 cacherep = RC_DOIT; 762 } 763 } 764 765 /* 766 * Loop to get all the write rpc relies that have been 767 * gathered together. 768 */ 769 do { 770 switch (cacherep) { 771 case RC_DOIT: 772 mreq = NULL; 773 netexport_rdlock(); 774 if (writes_todo || nd == NULL || 775 (!(nd->nd_flag & ND_NFSV3) && 776 nd->nd_procnum == NFSPROC_WRITE && 777 nfsrvw_procrastinate > 0)) 778 error = nfsrv_writegather(&nd, slp, 779 l, &mreq); 780 else 781 error = 782 (*(nfsrv3_procs[nd->nd_procnum])) 783 (nd, slp, l, &mreq); 784 netexport_rdunlock(); 785 if (mreq == NULL) { 786 if (nd != NULL) { 787 if (nd->nd_nam2) 788 m_free(nd->nd_nam2); 789 } 790 break; 791 } 792 if (error) { 793 nfsstats.srv_errs++; 794 if (nd) { 795 nfsrv_updatecache(nd, false, 796 mreq); 797 if (nd->nd_nam2) 798 m_freem(nd->nd_nam2); 799 } 800 break; 801 } 802 if (nd) { 803 nfsstats.srvrpccnt[nd->nd_procnum]++; 804 nfsrv_updatecache(nd, true, mreq); 805 nd->nd_mrep = NULL; 806 } 807 case RC_REPLY: 808 m = mreq; 809 siz = 0; 810 while (m) { 811 siz += m->m_len; 812 m = m->m_next; 813 } 814 if (siz <= 0 || siz > NFS_MAXPACKET) { 815 printf("mbuf siz=%d\n",siz); 816 panic("Bad nfs svc reply"); 817 } 818 m = mreq; 819 m->m_pkthdr.len = siz; 820 m_reset_rcvif(m); 821 /* 822 * For stream protocols, prepend a Sun RPC 823 * Record Mark. 824 */ 825 if (sotype == SOCK_STREAM) { 826 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 827 *mtod(m, u_int32_t *) = 828 htonl(0x80000000 | siz); 829 } 830 if (nd) { 831 nd->nd_mreq = m; 832 if (nfsrtton) { 833 nfsd_rt(slp->ns_so->so_type, nd, 834 cacherep); 835 } 836 error = nfsdsock_sendreply(slp, nd); 837 nd = NULL; 838 } 839 if (error == EPIPE) 840 nfsrv_zapsock(slp); 841 if (error == EINTR || error == ERESTART) { 842 nfsd->nfsd_slp = NULL; 843 nfsrv_slpderef(slp); 844 goto done; 845 } 846 break; 847 case RC_DROPIT: 848 if (nd) { 849 if (nfsrtton) 850 nfsd_rt(sotype, nd, cacherep); 851 m_freem(nd->nd_mrep); 852 m_freem(nd->nd_nam2); 853 } 854 break; 855 } 856 if (nd) { 857 nfsdreq_free(nd); 858 nd = NULL; 859 } 860 861 /* 862 * Check to see if there are outstanding writes that 863 * need to be serviced. 864 */ 865 getmicrotime(&tv); 866 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 867 (u_quad_t)tv.tv_usec; 868 mutex_enter(&nfsd_lock); 869 if (LIST_FIRST(&slp->ns_tq) && 870 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) { 871 cacherep = RC_DOIT; 872 writes_todo = 1; 873 } else 874 writes_todo = 0; 875 mutex_exit(&nfsd_lock); 876 } while (writes_todo); 877 if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) { 878 nfsd->nfsd_slp = NULL; 879 nfsrv_slpderef(slp); 880 } 881 } 882 done: 883 mutex_enter(&nfsd_lock); 884 nfsd_toss_cookie(nfsd); 885 doreinit = --nfs_numnfsd == 0; 886 if (doreinit) 887 nfssvc_sockhead_flag |= SLP_INIT; 888 mutex_exit(&nfsd_lock); 889 cv_destroy(&nfsd->nfsd_cv); 890 kmem_free(nfsd, sizeof(*nfsd)); 891 KASSERT(nsd->nsd_nfsd != (struct nfsd *)(uintptr_t)0); 892 nsd->nsd_nfsd = (struct nfsd *)(uintptr_t)0; 893 if (doreinit) 894 nfsrv_init(true); /* Reinitialize everything */ 895 return (error); 896 } 897 898 /* 899 * Shut down a socket associated with an nfssvc_sock structure. 900 * Should be called with the send lock set, if required. 901 * The trick here is to increment the sref at the start, so that the nfsds 902 * will stop using it and clear ns_flag at the end so that it will not be 903 * reassigned during cleanup. 904 * 905 * called at splsoftnet. 906 */ 907 void 908 nfsrv_zapsock(struct nfssvc_sock *slp) 909 { 910 struct nfsuid *nuidp, *nnuidp; 911 struct nfsrv_descript *nwp; 912 struct socket *so; 913 struct mbuf *m; 914 915 if (nfsdsock_drain(slp)) { 916 return; 917 } 918 mutex_enter(&nfsd_lock); 919 if (slp->ns_gflags & SLP_G_DOREC) { 920 TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending); 921 slp->ns_gflags &= ~SLP_G_DOREC; 922 } 923 mutex_exit(&nfsd_lock); 924 925 so = slp->ns_so; 926 KASSERT(so != NULL); 927 solock(so); 928 so->so_upcall = NULL; 929 so->so_upcallarg = NULL; 930 so->so_rcv.sb_flags &= ~SB_UPCALL; 931 soshutdown(so, SHUT_RDWR); 932 sounlock(so); 933 934 m_freem(slp->ns_raw); 935 m = slp->ns_rec; 936 while (m != NULL) { 937 struct mbuf *n; 938 939 n = m->m_nextpkt; 940 m_freem(m); 941 m = n; 942 } 943 /* XXX what about freeing ns_frag ? */ 944 for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0; 945 nuidp = nnuidp) { 946 nnuidp = TAILQ_NEXT(nuidp, nu_lru); 947 LIST_REMOVE(nuidp, nu_hash); 948 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 949 if (nuidp->nu_flag & NU_NAM) 950 m_freem(nuidp->nu_nam); 951 kmem_free(nuidp, sizeof(*nuidp)); 952 } 953 mutex_enter(&nfsd_lock); 954 while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) { 955 LIST_REMOVE(nwp, nd_tq); 956 mutex_exit(&nfsd_lock); 957 nfsdreq_free(nwp); 958 mutex_enter(&nfsd_lock); 959 } 960 mutex_exit(&nfsd_lock); 961 } 962 963 /* 964 * Derefence a server socket structure. If it has no more references and 965 * is no longer valid, you can throw it away. 966 */ 967 void 968 nfsrv_slpderef(struct nfssvc_sock *slp) 969 { 970 uint32_t ref; 971 972 mutex_enter(&nfsd_lock); 973 KASSERT(slp->ns_sref > 0); 974 ref = --slp->ns_sref; 975 if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) { 976 file_t *fp; 977 978 KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0); 979 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 980 mutex_exit(&nfsd_lock); 981 982 fp = slp->ns_fp; 983 if (fp != NULL) { 984 slp->ns_fp = NULL; 985 KASSERT(fp != NULL); 986 KASSERT(fp->f_socket == slp->ns_so); 987 KASSERT(fp->f_count > 0); 988 closef(fp); 989 slp->ns_so = NULL; 990 } 991 992 if (slp->ns_nam) 993 m_free(slp->ns_nam); 994 nfsrv_sockfree(slp); 995 } else 996 mutex_exit(&nfsd_lock); 997 } 998 999 /* 1000 * Initialize the data structures for the server. 1001 * Handshake with any new nfsds starting up to avoid any chance of 1002 * corruption. 1003 */ 1004 void 1005 nfsrv_init(int terminating) 1006 { 1007 struct nfssvc_sock *slp; 1008 1009 if (!terminating) { 1010 mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET); 1011 cv_init(&nfsd_initcv, "nfsdinit"); 1012 } 1013 1014 mutex_enter(&nfsd_lock); 1015 if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0) 1016 panic("nfsd init"); 1017 nfssvc_sockhead_flag |= SLP_INIT; 1018 1019 if (terminating) { 1020 KASSERT(SLIST_EMPTY(&nfsd_idle_head)); 1021 KASSERT(RB_TREE_MIN(&nfsd_tree) == NULL); 1022 while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) { 1023 mutex_exit(&nfsd_lock); 1024 KASSERT(slp->ns_sref == 0); 1025 slp->ns_sref++; 1026 nfsrv_zapsock(slp); 1027 nfsrv_slpderef(slp); 1028 mutex_enter(&nfsd_lock); 1029 } 1030 KASSERT(TAILQ_EMPTY(&nfssvc_sockpending)); 1031 mutex_exit(&nfsd_lock); 1032 nfsrv_cleancache(); /* And clear out server cache */ 1033 } else { 1034 mutex_exit(&nfsd_lock); 1035 nfs_pub.np_valid = 0; 1036 } 1037 1038 TAILQ_INIT(&nfssvc_sockhead); 1039 TAILQ_INIT(&nfssvc_sockpending); 1040 1041 rb_tree_init(&nfsd_tree, &nfsd_tree_ops); 1042 SLIST_INIT(&nfsd_idle_head); 1043 nfsd_head_flag &= ~NFSD_CHECKSLP; 1044 1045 nfs_udpsock = nfsrv_sockalloc(); 1046 nfs_udp6sock = nfsrv_sockalloc(); 1047 1048 mutex_enter(&nfsd_lock); 1049 nfssvc_sockhead_flag &= ~SLP_INIT; 1050 cv_broadcast(&nfsd_initcv); 1051 mutex_exit(&nfsd_lock); 1052 } 1053 1054 void 1055 nfsrv_fini(void) 1056 { 1057 1058 nfsrv_init(true); 1059 cv_destroy(&nfsd_initcv); 1060 mutex_destroy(&nfsd_lock); 1061 } 1062 1063 /* 1064 * Add entries to the server monitor log. 1065 */ 1066 static void 1067 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 1068 { 1069 struct timeval tv; 1070 struct drt *rt; 1071 1072 rt = &nfsdrt.drt[nfsdrt.pos]; 1073 if (cacherep == RC_DOIT) 1074 rt->flag = 0; 1075 else if (cacherep == RC_REPLY) 1076 rt->flag = DRT_CACHEREPLY; 1077 else 1078 rt->flag = DRT_CACHEDROP; 1079 if (sotype == SOCK_STREAM) 1080 rt->flag |= DRT_TCP; 1081 if (nd->nd_flag & ND_NFSV3) 1082 rt->flag |= DRT_NFSV3; 1083 rt->proc = nd->nd_procnum; 1084 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 1085 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 1086 else 1087 rt->ipadr = INADDR_ANY; 1088 getmicrotime(&tv); 1089 rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 1090 (tv.tv_usec - nd->nd_starttime.tv_usec); 1091 rt->tstamp = tv; 1092 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1093 } 1094