1 /* $NetBSD: nfs_syscalls.c,v 1.164 2024/07/05 04:31:54 rin 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 1.164 2024/07/05 04:31:54 rin 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 | NFSSVC_REPLACEEXPORTSLIST)) { 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 flag & (NFSSVC_SETEXPORTSLIST | NFSSVC_REPLACEEXPORTSLIST)); 362 363 free(args, M_TEMP); 364 } else { 365 error = ops->nsd_in(nsd, argp); 366 if (error) 367 return (error); 368 if ((uintptr_t)nsd->nsd_nfsd != 0 && 369 (nfsd = nfsd_get(nsd->nsd_nfsd)) == NULL) 370 return (EINVAL); 371 if ((flag & NFSSVC_AUTHIN) && 372 nfsd != NULL && 373 (nfsd->nfsd_slp->ns_flags & SLP_VALID)) { 374 slp = nfsd->nfsd_slp; 375 376 /* 377 * First check to see if another nfsd has already 378 * added this credential. 379 */ 380 LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid), 381 nu_hash) { 382 if (kauth_cred_geteuid(nuidp->nu_cr) == 383 nsd->nsd_cr.cr_uid && 384 (!nfsd->nfsd_nd->nd_nam2 || 385 netaddr_match(NU_NETFAM(nuidp), 386 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2))) 387 break; 388 } 389 if (nuidp) { 390 kauth_cred_hold(nuidp->nu_cr); 391 nfsd->nfsd_nd->nd_cr = nuidp->nu_cr; 392 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 393 } else { 394 /* 395 * Nope, so we will. 396 */ 397 if (slp->ns_numuids < nuidhash_max) { 398 slp->ns_numuids++; 399 nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP); 400 } else 401 nuidp = (struct nfsuid *)0; 402 if ((slp->ns_flags & SLP_VALID) == 0) { 403 if (nuidp) 404 kmem_free(nuidp, sizeof(*nuidp)); 405 } else { 406 if (nuidp == (struct nfsuid *)0) { 407 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); 408 LIST_REMOVE(nuidp, nu_hash); 409 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, 410 nu_lru); 411 if (nuidp->nu_flag & NU_NAM) 412 m_freem(nuidp->nu_nam); 413 } 414 nuidp->nu_flag = 0; 415 kauth_uucred_to_cred(nuidp->nu_cr, 416 &nsd->nsd_cr); 417 nuidp->nu_timestamp = nsd->nsd_timestamp; 418 nuidp->nu_expire = time_second + nsd->nsd_ttl; 419 /* 420 * and save the session key in nu_key. 421 */ 422 memcpy(nuidp->nu_key, nsd->nsd_key, 423 sizeof(nsd->nsd_key)); 424 if (nfsd->nfsd_nd->nd_nam2) { 425 struct sockaddr_in *saddr; 426 427 saddr = mtod(nfsd->nfsd_nd->nd_nam2, 428 struct sockaddr_in *); 429 switch (saddr->sin_family) { 430 case AF_INET: 431 nuidp->nu_flag |= NU_INETADDR; 432 nuidp->nu_inetaddr = 433 saddr->sin_addr.s_addr; 434 break; 435 case AF_INET6: 436 nuidp->nu_flag |= NU_NAM; 437 nuidp->nu_nam = m_copym( 438 nfsd->nfsd_nd->nd_nam2, 0, 439 M_COPYALL, M_WAIT); 440 break; 441 default: 442 kmem_free(nuidp, sizeof(*nuidp)); 443 return EAFNOSUPPORT; 444 }; 445 } 446 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp, 447 nu_lru); 448 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid), 449 nuidp, nu_hash); 450 kauth_cred_hold(nuidp->nu_cr); 451 nfsd->nfsd_nd->nd_cr = nuidp->nu_cr; 452 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL; 453 } 454 } 455 } 456 if ((flag & NFSSVC_AUTHINFAIL) && 457 nfsd != NULL) 458 nfsd->nfsd_flag |= NFSD_AUTHFAIL; 459 error = nfssvc_nfsd(ops, nsd, argp, l); 460 } 461 if (error == EINTR || error == ERESTART) 462 error = 0; 463 return (error); 464 } 465 466 static struct nfssvc_sock * 467 nfsrv_sockalloc(void) 468 { 469 struct nfssvc_sock *slp; 470 471 slp = kmem_alloc(sizeof(*slp), KM_SLEEP); 472 memset(slp, 0, sizeof (struct nfssvc_sock)); 473 mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET); 474 mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET); 475 cv_init(&slp->ns_cv, "nfsdsock"); 476 TAILQ_INIT(&slp->ns_uidlruhead); 477 LIST_INIT(&slp->ns_tq); 478 SIMPLEQ_INIT(&slp->ns_sendq); 479 mutex_enter(&nfsd_lock); 480 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain); 481 mutex_exit(&nfsd_lock); 482 483 return slp; 484 } 485 486 static void 487 nfsrv_sockfree(struct nfssvc_sock *slp) 488 { 489 490 KASSERT(slp->ns_so == NULL); 491 KASSERT(slp->ns_fp == NULL); 492 KASSERT((slp->ns_flags & SLP_VALID) == 0); 493 mutex_destroy(&slp->ns_lock); 494 mutex_destroy(&slp->ns_alock); 495 cv_destroy(&slp->ns_cv); 496 kmem_free(slp, sizeof(*slp)); 497 } 498 499 /* 500 * Adds a socket to the list for servicing by nfsds. 501 */ 502 int 503 nfssvc_addsock(file_t *fp, struct mbuf *mynam) 504 { 505 int siz; 506 struct nfssvc_sock *slp; 507 struct socket *so; 508 struct nfssvc_sock *tslp; 509 int error; 510 int val; 511 512 so = fp->f_socket; 513 tslp = (struct nfssvc_sock *)0; 514 /* 515 * Add it to the list, as required. 516 */ 517 if (so->so_proto->pr_protocol == IPPROTO_UDP) { 518 if (so->so_proto->pr_domain->dom_family == AF_INET6) 519 tslp = nfs_udp6sock; 520 else { 521 tslp = nfs_udpsock; 522 if (tslp->ns_flags & SLP_VALID) { 523 m_freem(mynam); 524 return (EPERM); 525 } 526 } 527 } 528 if (so->so_type == SOCK_STREAM) 529 siz = NFS_MAXPACKET + sizeof (u_long); 530 else 531 siz = NFS_MAXPACKET; 532 solock(so); 533 error = soreserve(so, siz, siz); 534 sounlock(so); 535 if (error) { 536 m_freem(mynam); 537 return (error); 538 } 539 540 /* 541 * Set protocol specific options { for now TCP only } and 542 * reserve some space. For datagram sockets, this can get called 543 * repeatedly for the same socket, but that isn't harmful. 544 */ 545 if (so->so_type == SOCK_STREAM) { 546 val = 1; 547 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 548 sizeof(val)); 549 } 550 if ((so->so_proto->pr_domain->dom_family == AF_INET || 551 so->so_proto->pr_domain->dom_family == AF_INET6) && 552 so->so_proto->pr_protocol == IPPROTO_TCP) { 553 val = 1; 554 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 555 sizeof(val)); 556 } 557 solock(so); 558 so->so_rcv.sb_flags &= ~SB_NOINTR; 559 so->so_rcv.sb_timeo = 0; 560 so->so_snd.sb_flags &= ~SB_NOINTR; 561 so->so_snd.sb_timeo = 0; 562 sounlock(so); 563 if (tslp) { 564 slp = tslp; 565 } else { 566 slp = nfsrv_sockalloc(); 567 } 568 slp->ns_so = so; 569 slp->ns_nam = mynam; 570 mutex_enter(&fp->f_lock); 571 fp->f_count++; 572 mutex_exit(&fp->f_lock); 573 slp->ns_fp = fp; 574 slp->ns_flags = SLP_VALID; 575 slp->ns_aflags = SLP_A_NEEDQ; 576 slp->ns_gflags = 0; 577 slp->ns_sflags = 0; 578 solock(so); 579 so->so_upcallarg = (void *)slp; 580 so->so_upcall = nfsrv_soupcall; 581 so->so_rcv.sb_flags |= SB_UPCALL; 582 sounlock(so); 583 nfsrv_wakenfsd(slp); 584 return (0); 585 } 586 587 /* 588 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests 589 * until it is killed by a signal. 590 */ 591 static int 592 nfssvc_nfsd(struct nfssvc_copy_ops *ops, struct nfsd_srvargs *nsd, 593 void *argp, struct lwp *l) 594 { 595 struct timeval tv; 596 struct mbuf *m; 597 struct nfssvc_sock *slp; 598 struct nfsd *nfsd; 599 struct nfsrv_descript *nd = NULL; 600 struct mbuf *mreq; 601 u_quad_t cur_usec; 602 int error = 0, cacherep, siz, sotype, writes_todo; 603 struct proc *p = l->l_proc; 604 bool doreinit; 605 606 #ifndef nolint 607 cacherep = RC_DOIT; 608 writes_todo = 0; 609 #endif 610 /* 611 * If userland didn't provide an nfsd cookie, bake a fresh one; 612 * if they did provide one, look it up. 613 */ 614 if ((uintptr_t)nsd->nsd_nfsd == 0) { 615 nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP); 616 memset(nfsd, 0, sizeof (struct nfsd)); 617 cv_init(&nfsd->nfsd_cv, "nfsd"); 618 nfsd->nfsd_procp = p; 619 mutex_enter(&nfsd_lock); 620 while ((nfssvc_sockhead_flag & SLP_INIT) != 0) { 621 KASSERT(nfs_numnfsd == 0); 622 cv_wait(&nfsd_initcv, &nfsd_lock); 623 } 624 nsd->nsd_nfsd = nfsd_bake_cookie(nfsd); 625 nfs_numnfsd++; 626 mutex_exit(&nfsd_lock); 627 } else if ((nfsd = nfsd_get(nsd->nsd_nfsd)) == NULL) { 628 return (EINVAL); 629 } 630 KASSERT(nfsd != NULL); 631 KASSERT(nsd->nsd_nfsd != (struct nfsd *)(uintptr_t)0); 632 633 /* 634 * Loop getting rpc requests until SIGKILL. 635 */ 636 for (;;) { 637 bool dummy; 638 639 preempt_point(); 640 641 if (nfsd->nfsd_slp == NULL) { 642 mutex_enter(&nfsd_lock); 643 while (nfsd->nfsd_slp == NULL && 644 (nfsd_head_flag & NFSD_CHECKSLP) == 0) { 645 SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd, 646 nfsd_idle); 647 error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock); 648 if (error) { 649 slp = nfsd->nfsd_slp; 650 nfsd->nfsd_slp = NULL; 651 if (!slp) 652 SLIST_REMOVE(&nfsd_idle_head, 653 nfsd, nfsd, nfsd_idle); 654 mutex_exit(&nfsd_lock); 655 if (slp) { 656 nfsrv_wakenfsd(slp); 657 nfsrv_slpderef(slp); 658 } 659 goto done; 660 } 661 } 662 if (nfsd->nfsd_slp == NULL && 663 (nfsd_head_flag & NFSD_CHECKSLP) != 0) { 664 slp = TAILQ_FIRST(&nfssvc_sockpending); 665 if (slp) { 666 KASSERT((slp->ns_gflags & SLP_G_DOREC) 667 != 0); 668 TAILQ_REMOVE(&nfssvc_sockpending, slp, 669 ns_pending); 670 slp->ns_gflags &= ~SLP_G_DOREC; 671 slp->ns_sref++; 672 nfsd->nfsd_slp = slp; 673 } else 674 nfsd_head_flag &= ~NFSD_CHECKSLP; 675 } 676 KASSERT(nfsd->nfsd_slp == NULL || 677 nfsd->nfsd_slp->ns_sref > 0); 678 mutex_exit(&nfsd_lock); 679 if ((slp = nfsd->nfsd_slp) == NULL) 680 continue; 681 if (slp->ns_flags & SLP_VALID) { 682 bool more; 683 684 if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) { 685 nfsrv_rcv(slp); 686 } 687 if (nfsdsock_testbits(slp, SLP_A_DISCONN)) { 688 nfsrv_zapsock(slp); 689 } 690 error = nfsrv_dorec(slp, nfsd, &nd, &more); 691 getmicrotime(&tv); 692 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 693 (u_quad_t)tv.tv_usec; 694 writes_todo = 0; 695 if (error) { 696 struct nfsrv_descript *nd2; 697 698 mutex_enter(&nfsd_lock); 699 nd2 = LIST_FIRST(&slp->ns_tq); 700 if (nd2 != NULL && 701 nd2->nd_time <= cur_usec) { 702 error = 0; 703 cacherep = RC_DOIT; 704 writes_todo = 1; 705 } 706 mutex_exit(&nfsd_lock); 707 } 708 if (error == 0 && more) { 709 nfsrv_wakenfsd(slp); 710 } 711 } 712 } else { 713 error = 0; 714 slp = nfsd->nfsd_slp; 715 } 716 KASSERT(slp != NULL); 717 KASSERT(nfsd->nfsd_slp == slp); 718 if (error || (slp->ns_flags & SLP_VALID) == 0) { 719 if (nd) { 720 nfsdreq_free(nd); 721 nd = NULL; 722 } 723 nfsd->nfsd_slp = NULL; 724 nfsrv_slpderef(slp); 725 continue; 726 } 727 sotype = slp->ns_so->so_type; 728 if (nd) { 729 getmicrotime(&nd->nd_starttime); 730 if (nd->nd_nam2) 731 nd->nd_nam = nd->nd_nam2; 732 else 733 nd->nd_nam = slp->ns_nam; 734 735 /* 736 * Check to see if authorization is needed. 737 */ 738 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) { 739 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH; 740 nsd->nsd_haddr = mtod(nd->nd_nam, 741 struct sockaddr_in *)->sin_addr.s_addr; 742 nsd->nsd_authlen = nfsd->nfsd_authlen; 743 nsd->nsd_verflen = nfsd->nfsd_verflen; 744 if (!copyout(nfsd->nfsd_authstr, 745 nsd->nsd_authstr, nfsd->nfsd_authlen) && 746 !copyout(nfsd->nfsd_verfstr, 747 nsd->nsd_verfstr, nfsd->nfsd_verflen) && 748 !ops->nsd_out(argp, nsd)) { 749 return (ENEEDAUTH); 750 } 751 cacherep = RC_DROPIT; 752 } else 753 cacherep = nfsrv_getcache(nd, slp, &mreq); 754 755 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) { 756 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL; 757 nd->nd_procnum = NFSPROC_NOOP; 758 nd->nd_repstat = 759 (NFSERR_AUTHERR | AUTH_TOOWEAK); 760 cacherep = RC_DOIT; 761 } 762 } 763 764 /* 765 * Loop to get all the write rpc relies that have been 766 * gathered together. 767 */ 768 do { 769 switch (cacherep) { 770 case RC_DOIT: 771 mreq = NULL; 772 netexport_rdlock(); 773 if (writes_todo || nd == NULL || 774 (!(nd->nd_flag & ND_NFSV3) && 775 nd->nd_procnum == NFSPROC_WRITE && 776 nfsrvw_procrastinate > 0)) 777 error = nfsrv_writegather(&nd, slp, 778 l, &mreq); 779 else 780 error = 781 (*(nfsrv3_procs[nd->nd_procnum])) 782 (nd, slp, l, &mreq); 783 netexport_rdunlock(); 784 if (mreq == NULL) { 785 if (nd != NULL) { 786 if (nd->nd_nam2) 787 m_free(nd->nd_nam2); 788 } 789 break; 790 } 791 if (error) { 792 nfsstats.srv_errs++; 793 if (nd) { 794 nfsrv_updatecache(nd, false, 795 mreq); 796 m_freem(nd->nd_nam2); 797 } 798 break; 799 } 800 if (nd) { 801 nfsstats.srvrpccnt[nd->nd_procnum]++; 802 nfsrv_updatecache(nd, true, mreq); 803 nd->nd_mrep = NULL; 804 } 805 /* FALLTHROUGH */ 806 case RC_REPLY: 807 m = mreq; 808 siz = 0; 809 while (m) { 810 siz += m->m_len; 811 m = m->m_next; 812 } 813 if (siz <= 0 || siz > NFS_MAXPACKET) { 814 printf("mbuf siz=%d\n",siz); 815 panic("Bad nfs svc reply"); 816 } 817 m = mreq; 818 m->m_pkthdr.len = siz; 819 m_reset_rcvif(m); 820 /* 821 * For stream protocols, prepend a Sun RPC 822 * Record Mark. 823 */ 824 if (sotype == SOCK_STREAM) { 825 M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); 826 *mtod(m, u_int32_t *) = 827 htonl(0x80000000 | siz); 828 } 829 if (nd) { 830 nd->nd_mreq = m; 831 if (nfsrtton) { 832 nfsd_rt(slp->ns_so->so_type, nd, 833 cacherep); 834 } 835 error = nfsdsock_sendreply(slp, nd); 836 nd = NULL; 837 } 838 if (error == EPIPE) 839 nfsrv_zapsock(slp); 840 if (error == EINTR || error == ERESTART) { 841 nfsd->nfsd_slp = NULL; 842 nfsrv_slpderef(slp); 843 goto done; 844 } 845 break; 846 case RC_DROPIT: 847 if (nd) { 848 if (nfsrtton) 849 nfsd_rt(sotype, nd, cacherep); 850 m_freem(nd->nd_mrep); 851 m_freem(nd->nd_nam2); 852 } 853 break; 854 } 855 if (nd) { 856 nfsdreq_free(nd); 857 nd = NULL; 858 } 859 860 /* 861 * Check to see if there are outstanding writes that 862 * need to be serviced. 863 */ 864 getmicrotime(&tv); 865 cur_usec = (u_quad_t)tv.tv_sec * 1000000 + 866 (u_quad_t)tv.tv_usec; 867 mutex_enter(&nfsd_lock); 868 if (LIST_FIRST(&slp->ns_tq) && 869 LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) { 870 cacherep = RC_DOIT; 871 writes_todo = 1; 872 } else 873 writes_todo = 0; 874 mutex_exit(&nfsd_lock); 875 } while (writes_todo); 876 if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) { 877 nfsd->nfsd_slp = NULL; 878 nfsrv_slpderef(slp); 879 } 880 } 881 done: 882 mutex_enter(&nfsd_lock); 883 nfsd_toss_cookie(nfsd); 884 doreinit = --nfs_numnfsd == 0; 885 if (doreinit) 886 nfssvc_sockhead_flag |= SLP_INIT; 887 mutex_exit(&nfsd_lock); 888 cv_destroy(&nfsd->nfsd_cv); 889 kmem_free(nfsd, sizeof(*nfsd)); 890 KASSERT(nsd->nsd_nfsd != (struct nfsd *)(uintptr_t)0); 891 nsd->nsd_nfsd = (struct nfsd *)(uintptr_t)0; 892 if (doreinit) 893 nfsrv_init(true); /* Reinitialize everything */ 894 return (error); 895 } 896 897 /* 898 * Shut down a socket associated with an nfssvc_sock structure. 899 * Should be called with the send lock set, if required. 900 * The trick here is to increment the sref at the start, so that the nfsds 901 * will stop using it and clear ns_flag at the end so that it will not be 902 * reassigned during cleanup. 903 * 904 * called at splsoftnet. 905 */ 906 void 907 nfsrv_zapsock(struct nfssvc_sock *slp) 908 { 909 struct nfsuid *nuidp, *nnuidp; 910 struct nfsrv_descript *nwp; 911 struct socket *so; 912 struct mbuf *m; 913 914 if (nfsdsock_drain(slp)) { 915 return; 916 } 917 mutex_enter(&nfsd_lock); 918 if (slp->ns_gflags & SLP_G_DOREC) { 919 TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending); 920 slp->ns_gflags &= ~SLP_G_DOREC; 921 } 922 mutex_exit(&nfsd_lock); 923 924 so = slp->ns_so; 925 KASSERT(so != NULL); 926 solock(so); 927 so->so_upcall = NULL; 928 so->so_upcallarg = NULL; 929 so->so_rcv.sb_flags &= ~SB_UPCALL; 930 soshutdown(so, SHUT_RDWR); 931 sounlock(so); 932 933 m_freem(slp->ns_raw); 934 m = slp->ns_rec; 935 while (m != NULL) { 936 struct mbuf *n; 937 938 n = m->m_nextpkt; 939 m_freem(m); 940 m = n; 941 } 942 /* XXX what about freeing ns_frag ? */ 943 for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0; 944 nuidp = nnuidp) { 945 nnuidp = TAILQ_NEXT(nuidp, nu_lru); 946 LIST_REMOVE(nuidp, nu_hash); 947 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru); 948 if (nuidp->nu_flag & NU_NAM) 949 m_freem(nuidp->nu_nam); 950 kmem_free(nuidp, sizeof(*nuidp)); 951 } 952 mutex_enter(&nfsd_lock); 953 while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) { 954 LIST_REMOVE(nwp, nd_tq); 955 mutex_exit(&nfsd_lock); 956 nfsdreq_free(nwp); 957 mutex_enter(&nfsd_lock); 958 } 959 mutex_exit(&nfsd_lock); 960 } 961 962 /* 963 * Derefence a server socket structure. If it has no more references and 964 * is no longer valid, you can throw it away. 965 */ 966 void 967 nfsrv_slpderef(struct nfssvc_sock *slp) 968 { 969 uint32_t ref; 970 971 mutex_enter(&nfsd_lock); 972 KASSERT(slp->ns_sref > 0); 973 ref = --slp->ns_sref; 974 if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) { 975 file_t *fp; 976 977 KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0); 978 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain); 979 mutex_exit(&nfsd_lock); 980 981 fp = slp->ns_fp; 982 if (fp != NULL) { 983 slp->ns_fp = NULL; 984 KASSERT(fp != NULL); 985 KASSERT(fp->f_socket == slp->ns_so); 986 KASSERT(fp->f_count > 0); 987 closef(fp); 988 slp->ns_so = NULL; 989 } 990 991 if (slp->ns_nam) 992 m_free(slp->ns_nam); 993 nfsrv_sockfree(slp); 994 } else 995 mutex_exit(&nfsd_lock); 996 } 997 998 /* 999 * Initialize the data structures for the server. 1000 * Handshake with any new nfsds starting up to avoid any chance of 1001 * corruption. 1002 */ 1003 void 1004 nfsrv_init(int terminating) 1005 { 1006 struct nfssvc_sock *slp; 1007 1008 if (!terminating) { 1009 mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET); 1010 cv_init(&nfsd_initcv, "nfsdinit"); 1011 } 1012 1013 mutex_enter(&nfsd_lock); 1014 if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0) 1015 panic("nfsd init"); 1016 nfssvc_sockhead_flag |= SLP_INIT; 1017 1018 if (terminating) { 1019 KASSERT(SLIST_EMPTY(&nfsd_idle_head)); 1020 KASSERT(RB_TREE_MIN(&nfsd_tree) == NULL); 1021 while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) { 1022 mutex_exit(&nfsd_lock); 1023 KASSERT(slp->ns_sref == 0); 1024 slp->ns_sref++; 1025 nfsrv_zapsock(slp); 1026 nfsrv_slpderef(slp); 1027 mutex_enter(&nfsd_lock); 1028 } 1029 KASSERT(TAILQ_EMPTY(&nfssvc_sockpending)); 1030 mutex_exit(&nfsd_lock); 1031 nfsrv_cleancache(); /* And clear out server cache */ 1032 } else { 1033 mutex_exit(&nfsd_lock); 1034 nfs_pub.np_valid = 0; 1035 } 1036 1037 TAILQ_INIT(&nfssvc_sockhead); 1038 TAILQ_INIT(&nfssvc_sockpending); 1039 1040 rb_tree_init(&nfsd_tree, &nfsd_tree_ops); 1041 SLIST_INIT(&nfsd_idle_head); 1042 nfsd_head_flag &= ~NFSD_CHECKSLP; 1043 1044 nfs_udpsock = nfsrv_sockalloc(); 1045 nfs_udp6sock = nfsrv_sockalloc(); 1046 1047 mutex_enter(&nfsd_lock); 1048 nfssvc_sockhead_flag &= ~SLP_INIT; 1049 cv_broadcast(&nfsd_initcv); 1050 mutex_exit(&nfsd_lock); 1051 } 1052 1053 void 1054 nfsrv_fini(void) 1055 { 1056 1057 nfsrv_init(true); 1058 cv_destroy(&nfsd_initcv); 1059 mutex_destroy(&nfsd_lock); 1060 } 1061 1062 /* 1063 * Add entries to the server monitor log. 1064 */ 1065 static void 1066 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep) 1067 { 1068 struct timeval tv; 1069 struct drt *rt; 1070 1071 rt = &nfsdrt.drt[nfsdrt.pos]; 1072 if (cacherep == RC_DOIT) 1073 rt->flag = 0; 1074 else if (cacherep == RC_REPLY) 1075 rt->flag = DRT_CACHEREPLY; 1076 else 1077 rt->flag = DRT_CACHEDROP; 1078 if (sotype == SOCK_STREAM) 1079 rt->flag |= DRT_TCP; 1080 if (nd->nd_flag & ND_NFSV3) 1081 rt->flag |= DRT_NFSV3; 1082 rt->proc = nd->nd_procnum; 1083 if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET) 1084 rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr; 1085 else 1086 rt->ipadr = INADDR_ANY; 1087 getmicrotime(&tv); 1088 rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) + 1089 (tv.tv_usec - nd->nd_starttime.tv_usec); 1090 rt->tstamp = tv; 1091 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ; 1092 } 1093