Home | History | Annotate | Line # | Download | only in nfs
nfs_syscalls.c revision 1.139
      1 /*	$NetBSD: nfs_syscalls.c,v 1.139 2008/09/28 00:10:18 pooka 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 1.139 2008/09/28 00:10:18 pooka Exp $");
     39 
     40 #include "fs_nfs.h"
     41 #include "opt_nfs.h"
     42 #include "opt_nfsserver.h"
     43 #include "opt_iso.h"
     44 #include "opt_inet.h"
     45 #include "opt_compat_netbsd.h"
     46 
     47 #include <sys/param.h>
     48 #include <sys/systm.h>
     49 #include <sys/kernel.h>
     50 #include <sys/file.h>
     51 #include <sys/stat.h>
     52 #include <sys/vnode.h>
     53 #include <sys/mount.h>
     54 #include <sys/proc.h>
     55 #include <sys/uio.h>
     56 #include <sys/malloc.h>
     57 #include <sys/kmem.h>
     58 #include <sys/buf.h>
     59 #include <sys/mbuf.h>
     60 #include <sys/socket.h>
     61 #include <sys/socketvar.h>
     62 #include <sys/signalvar.h>
     63 #include <sys/domain.h>
     64 #include <sys/protosw.h>
     65 #include <sys/namei.h>
     66 #include <sys/syslog.h>
     67 #include <sys/filedesc.h>
     68 #include <sys/kthread.h>
     69 #include <sys/kauth.h>
     70 #include <sys/syscallargs.h>
     71 
     72 #include <netinet/in.h>
     73 #include <netinet/tcp.h>
     74 #ifdef ISO
     75 #include <netiso/iso.h>
     76 #endif
     77 #include <nfs/xdr_subs.h>
     78 #include <nfs/rpcv2.h>
     79 #include <nfs/nfsproto.h>
     80 #include <nfs/nfs.h>
     81 #include <nfs/nfsm_subs.h>
     82 #include <nfs/nfsrvcache.h>
     83 #include <nfs/nfsmount.h>
     84 #include <nfs/nfsnode.h>
     85 #include <nfs/nfsrtt.h>
     86 #include <nfs/nfs_var.h>
     87 
     88 /* Global defs. */
     89 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
     90 						struct nfssvc_sock *,
     91 						struct lwp *, struct mbuf **));
     92 extern int nfsrvw_procrastinate;
     93 
     94 struct nfssvc_sock *nfs_udpsock;
     95 #ifdef ISO
     96 struct nfssvc_sock *nfs_cltpsock;
     97 #endif
     98 #ifdef INET6
     99 struct nfssvc_sock *nfs_udp6sock;
    100 #endif
    101 int nuidhash_max = NFS_MAXUIDHASH;
    102 #ifdef NFSSERVER
    103 static int nfs_numnfsd = 0;
    104 static struct nfsdrt nfsdrt;
    105 #endif
    106 
    107 #ifdef NFSSERVER
    108 kmutex_t nfsd_lock;
    109 struct nfssvc_sockhead nfssvc_sockhead;
    110 kcondvar_t nfsd_initcv;
    111 struct nfssvc_sockhead nfssvc_sockpending;
    112 struct nfsdhead nfsd_head;
    113 struct nfsdidlehead nfsd_idle_head;
    114 
    115 int nfssvc_sockhead_flag;
    116 int nfsd_head_flag;
    117 #endif
    118 
    119 #ifdef NFS
    120 /*
    121  * locking order:
    122  *	nfs_iodlist_lock -> nid_lock -> nm_lock
    123  */
    124 kmutex_t nfs_iodlist_lock;
    125 struct nfs_iodlist nfs_iodlist_idle;
    126 struct nfs_iodlist nfs_iodlist_all;
    127 int nfs_niothreads = -1; /* == "0, and has never been set" */
    128 #endif
    129 
    130 #ifdef NFSSERVER
    131 static struct nfssvc_sock *nfsrv_sockalloc __P((void));
    132 static void nfsrv_sockfree __P((struct nfssvc_sock *));
    133 static void nfsd_rt __P((int, struct nfsrv_descript *, int));
    134 #endif
    135 
    136 /*
    137  * NFS server system calls
    138  */
    139 
    140 
    141 /*
    142  * Nfs server pseudo system call for the nfsd's
    143  * Based on the flag value it either:
    144  * - adds a socket to the selection list
    145  * - remains in the kernel as an nfsd
    146  * - remains in the kernel as an nfsiod
    147  */
    148 int
    149 sys_nfssvc(struct lwp *l, const struct sys_nfssvc_args *uap, register_t *retval)
    150 {
    151 	/* {
    152 		syscallarg(int) flag;
    153 		syscallarg(void *) argp;
    154 	} */
    155 	int error;
    156 #ifdef NFSSERVER
    157 	file_t *fp;
    158 	struct mbuf *nam;
    159 	struct nfsd_args nfsdarg;
    160 	struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
    161 	struct nfsd *nfsd;
    162 	struct nfssvc_sock *slp;
    163 	struct nfsuid *nuidp;
    164 #endif
    165 
    166 	/*
    167 	 * Must be super user
    168 	 */
    169 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_NFS,
    170 	    KAUTH_REQ_NETWORK_NFS_SVC, NULL, NULL, NULL);
    171 	if (error)
    172 		return (error);
    173 
    174 	/* Initialize NFS server / client shared data. */
    175 	nfs_init();
    176 
    177 #ifdef NFSSERVER
    178 	mutex_enter(&nfsd_lock);
    179 	while (nfssvc_sockhead_flag & SLP_INIT) {
    180 		cv_wait(&nfsd_initcv, &nfsd_lock);
    181 	}
    182 	mutex_exit(&nfsd_lock);
    183 #endif
    184 	if (SCARG(uap, flag) & NFSSVC_BIOD) {
    185 #if defined(NFS) && defined(COMPAT_14)
    186 		error = kpause("nfsbiod", true, 0, NULL); /* dummy impl */
    187 #else
    188 		error = ENOSYS;
    189 #endif
    190 	} else if (SCARG(uap, flag) & NFSSVC_MNTD) {
    191 		error = ENOSYS;
    192 	} else if (SCARG(uap, flag) & NFSSVC_ADDSOCK) {
    193 #ifndef NFSSERVER
    194 		error = ENOSYS;
    195 #else
    196 		error = copyin(SCARG(uap, argp), (void *)&nfsdarg,
    197 		    sizeof(nfsdarg));
    198 		if (error)
    199 			return (error);
    200 		/* getsock() will use the descriptor for us */
    201 		if ((fp = fd_getfile(nfsdarg.sock)) == NULL)
    202 			return (EBADF);
    203 		if (fp->f_type != DTYPE_SOCKET) {
    204 			fd_putfile(nfsdarg.sock);
    205 			return (ENOTSOCK);
    206 		}
    207 		if (error)
    208 			return (error);
    209 		/*
    210 		 * Get the client address for connected sockets.
    211 		 */
    212 		if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
    213 			nam = (struct mbuf *)0;
    214 		else {
    215 			error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen,
    216 				MT_SONAME);
    217 			if (error) {
    218 				fd_putfile(nfsdarg.sock);
    219 				return (error);
    220 			}
    221 		}
    222 		error = nfssvc_addsock(fp, nam);
    223 		fd_putfile(nfsdarg.sock);
    224 #endif /* !NFSSERVER */
    225 	} else if (SCARG(uap, flag) & NFSSVC_SETEXPORTSLIST) {
    226 #ifndef NFSSERVER
    227 		error = ENOSYS;
    228 #else
    229 		struct export_args *args;
    230 		struct mountd_exports_list mel;
    231 
    232 		error = copyin(SCARG(uap, argp), &mel, sizeof(mel));
    233 		if (error != 0)
    234 			return error;
    235 
    236 		args = (struct export_args *)malloc(mel.mel_nexports *
    237 		    sizeof(struct export_args), M_TEMP, M_WAITOK);
    238 		error = copyin(mel.mel_exports, args, mel.mel_nexports *
    239 		    sizeof(struct export_args));
    240 		if (error != 0) {
    241 			free(args, M_TEMP);
    242 			return error;
    243 		}
    244 		mel.mel_exports = args;
    245 
    246 		error = mountd_set_exports_list(&mel, l);
    247 
    248 		free(args, M_TEMP);
    249 #endif /* !NFSSERVER */
    250 	} else {
    251 #ifndef NFSSERVER
    252 		error = ENOSYS;
    253 #else
    254 		error = copyin(SCARG(uap, argp), (void *)nsd, sizeof (*nsd));
    255 		if (error)
    256 			return (error);
    257 		if ((SCARG(uap, flag) & NFSSVC_AUTHIN) &&
    258 		    ((nfsd = nsd->nsd_nfsd)) != NULL &&
    259 		    (nfsd->nfsd_slp->ns_flags & SLP_VALID)) {
    260 			slp = nfsd->nfsd_slp;
    261 
    262 			/*
    263 			 * First check to see if another nfsd has already
    264 			 * added this credential.
    265 			 */
    266 			LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid),
    267 			    nu_hash) {
    268 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
    269 				    nsd->nsd_cr.cr_uid &&
    270 				    (!nfsd->nfsd_nd->nd_nam2 ||
    271 				     netaddr_match(NU_NETFAM(nuidp),
    272 				     &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
    273 					break;
    274 			}
    275 			if (nuidp) {
    276 			    kauth_cred_hold(nuidp->nu_cr);
    277 			    nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    278 			    nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    279 			} else {
    280 			    /*
    281 			     * Nope, so we will.
    282 			     */
    283 			    if (slp->ns_numuids < nuidhash_max) {
    284 				slp->ns_numuids++;
    285 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
    286 			    } else
    287 				nuidp = (struct nfsuid *)0;
    288 			    if ((slp->ns_flags & SLP_VALID) == 0) {
    289 				if (nuidp)
    290 				    kmem_free(nuidp, sizeof(*nuidp));
    291 			    } else {
    292 				if (nuidp == (struct nfsuid *)0) {
    293 				    nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
    294 				    LIST_REMOVE(nuidp, nu_hash);
    295 				    TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
    296 					nu_lru);
    297 				    if (nuidp->nu_flag & NU_NAM)
    298 					m_freem(nuidp->nu_nam);
    299 			        }
    300 				nuidp->nu_flag = 0;
    301 				kauth_uucred_to_cred(nuidp->nu_cr,
    302 				    &nsd->nsd_cr);
    303 				nuidp->nu_timestamp = nsd->nsd_timestamp;
    304 				nuidp->nu_expire = time_second + nsd->nsd_ttl;
    305 				/*
    306 				 * and save the session key in nu_key.
    307 				 */
    308 				memcpy(nuidp->nu_key, nsd->nsd_key,
    309 				    sizeof(nsd->nsd_key));
    310 				if (nfsd->nfsd_nd->nd_nam2) {
    311 				    struct sockaddr_in *saddr;
    312 
    313 				    saddr = mtod(nfsd->nfsd_nd->nd_nam2,
    314 					 struct sockaddr_in *);
    315 				    switch (saddr->sin_family) {
    316 				    case AF_INET:
    317 					nuidp->nu_flag |= NU_INETADDR;
    318 					nuidp->nu_inetaddr =
    319 					     saddr->sin_addr.s_addr;
    320 					break;
    321 				    case AF_ISO:
    322 				    default:
    323 					nuidp->nu_flag |= NU_NAM;
    324 					nuidp->nu_nam = m_copym(
    325 					    nfsd->nfsd_nd->nd_nam2, 0,
    326 					     M_COPYALL, M_WAIT);
    327 					break;
    328 				    };
    329 				}
    330 				TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
    331 					nu_lru);
    332 				LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
    333 					nuidp, nu_hash);
    334 				kauth_cred_hold(nuidp->nu_cr);
    335 				nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    336 				nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    337 			    }
    338 			}
    339 		}
    340 		if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) &&
    341 		    (nfsd = nsd->nsd_nfsd))
    342 			nfsd->nfsd_flag |= NFSD_AUTHFAIL;
    343 		error = nfssvc_nfsd(nsd, SCARG(uap, argp), l);
    344 #endif /* !NFSSERVER */
    345 	}
    346 	if (error == EINTR || error == ERESTART)
    347 		error = 0;
    348 	return (error);
    349 }
    350 
    351 #ifdef NFSSERVER
    352 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
    353 
    354 static struct nfssvc_sock *
    355 nfsrv_sockalloc()
    356 {
    357 	struct nfssvc_sock *slp;
    358 
    359 	slp = kmem_alloc(sizeof(*slp), KM_SLEEP);
    360 	memset(slp, 0, sizeof (struct nfssvc_sock));
    361 	mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET);
    362 	mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET);
    363 	cv_init(&slp->ns_cv, "nfsdsock");
    364 	TAILQ_INIT(&slp->ns_uidlruhead);
    365 	LIST_INIT(&slp->ns_tq);
    366 	SIMPLEQ_INIT(&slp->ns_sendq);
    367 	mutex_enter(&nfsd_lock);
    368 	TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
    369 	mutex_exit(&nfsd_lock);
    370 
    371 	return slp;
    372 }
    373 
    374 static void
    375 nfsrv_sockfree(struct nfssvc_sock *slp)
    376 {
    377 
    378 	KASSERT(slp->ns_so == NULL);
    379 	KASSERT(slp->ns_fp == NULL);
    380 	KASSERT((slp->ns_flags & SLP_VALID) == 0);
    381 	mutex_destroy(&slp->ns_lock);
    382 	mutex_destroy(&slp->ns_alock);
    383 	cv_destroy(&slp->ns_cv);
    384 	kmem_free(slp, sizeof(*slp));
    385 }
    386 
    387 /*
    388  * Adds a socket to the list for servicing by nfsds.
    389  */
    390 int
    391 nfssvc_addsock(fp, mynam)
    392 	file_t *fp;
    393 	struct mbuf *mynam;
    394 {
    395 	int siz;
    396 	struct nfssvc_sock *slp;
    397 	struct socket *so;
    398 	struct nfssvc_sock *tslp;
    399 	int error;
    400 	int val;
    401 
    402 	so = (struct socket *)fp->f_data;
    403 	tslp = (struct nfssvc_sock *)0;
    404 	/*
    405 	 * Add it to the list, as required.
    406 	 */
    407 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
    408 #ifdef INET6
    409 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
    410 			tslp = nfs_udp6sock;
    411 		else
    412 #endif
    413 		tslp = nfs_udpsock;
    414 		if (tslp->ns_flags & SLP_VALID) {
    415 			m_freem(mynam);
    416 			return (EPERM);
    417 		}
    418 #ifdef ISO
    419 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
    420 		tslp = nfs_cltpsock;
    421 		if (tslp->ns_flags & SLP_VALID) {
    422 			m_freem(mynam);
    423 			return (EPERM);
    424 		}
    425 #endif /* ISO */
    426 	}
    427 	if (so->so_type == SOCK_STREAM)
    428 		siz = NFS_MAXPACKET + sizeof (u_long);
    429 	else
    430 		siz = NFS_MAXPACKET;
    431 	solock(so);
    432 	error = soreserve(so, siz, siz);
    433 	sounlock(so);
    434 	if (error) {
    435 		m_freem(mynam);
    436 		return (error);
    437 	}
    438 
    439 	/*
    440 	 * Set protocol specific options { for now TCP only } and
    441 	 * reserve some space. For datagram sockets, this can get called
    442 	 * repeatedly for the same socket, but that isn't harmful.
    443 	 */
    444 	if (so->so_type == SOCK_STREAM) {
    445 		val = 1;
    446 		so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val,
    447 		    sizeof(val));
    448 	}
    449 	if ((so->so_proto->pr_domain->dom_family == AF_INET
    450 #ifdef INET6
    451 	    || so->so_proto->pr_domain->dom_family == AF_INET6
    452 #endif
    453 	    ) &&
    454 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
    455 		val = 1;
    456 		so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val,
    457 		    sizeof(val));
    458 	}
    459 	solock(so);
    460 	so->so_rcv.sb_flags &= ~SB_NOINTR;
    461 	so->so_rcv.sb_timeo = 0;
    462 	so->so_snd.sb_flags &= ~SB_NOINTR;
    463 	so->so_snd.sb_timeo = 0;
    464 	sounlock(so);
    465 	if (tslp) {
    466 		slp = tslp;
    467 	} else {
    468 		slp = nfsrv_sockalloc();
    469 	}
    470 	slp->ns_so = so;
    471 	slp->ns_nam = mynam;
    472 	mutex_enter(&fp->f_lock);
    473 	fp->f_count++;
    474 	mutex_exit(&fp->f_lock);
    475 	slp->ns_fp = fp;
    476 	slp->ns_flags = SLP_VALID;
    477 	slp->ns_aflags = SLP_A_NEEDQ;
    478 	slp->ns_gflags = 0;
    479 	slp->ns_sflags = 0;
    480 	solock(so);
    481 	so->so_upcallarg = (void *)slp;
    482 	so->so_upcall = nfsrv_soupcall;
    483 	so->so_rcv.sb_flags |= SB_UPCALL;
    484 	sounlock(so);
    485 	nfsrv_wakenfsd(slp);
    486 	return (0);
    487 }
    488 
    489 /*
    490  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    491  * until it is killed by a signal.
    492  */
    493 int
    494 nfssvc_nfsd(nsd, argp, l)
    495 	struct nfsd_srvargs *nsd;
    496 	void *argp;
    497 	struct lwp *l;
    498 {
    499 	struct timeval tv;
    500 	struct mbuf *m;
    501 	struct nfssvc_sock *slp;
    502 	struct nfsd *nfsd = nsd->nsd_nfsd;
    503 	struct nfsrv_descript *nd = NULL;
    504 	struct mbuf *mreq;
    505 	u_quad_t cur_usec;
    506 	int error = 0, cacherep, siz, sotype, writes_todo;
    507 	struct proc *p = l->l_proc;
    508 	int s;
    509 	bool doreinit;
    510 
    511 #ifndef nolint
    512 	cacherep = RC_DOIT;
    513 	writes_todo = 0;
    514 #endif
    515 	uvm_lwp_hold(l);
    516 	if (nfsd == NULL) {
    517 		nsd->nsd_nfsd = nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP);
    518 		memset(nfsd, 0, sizeof (struct nfsd));
    519 		cv_init(&nfsd->nfsd_cv, "nfsd");
    520 		nfsd->nfsd_procp = p;
    521 		mutex_enter(&nfsd_lock);
    522 		while ((nfssvc_sockhead_flag & SLP_INIT) != 0) {
    523 			KASSERT(nfs_numnfsd == 0);
    524 			cv_wait(&nfsd_initcv, &nfsd_lock);
    525 		}
    526 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
    527 		nfs_numnfsd++;
    528 		mutex_exit(&nfsd_lock);
    529 	}
    530 	/*
    531 	 * Loop getting rpc requests until SIGKILL.
    532 	 */
    533 	for (;;) {
    534 		bool dummy;
    535 
    536 		if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    537 		    != 0) {
    538 			preempt();
    539 		}
    540 		if (nfsd->nfsd_slp == NULL) {
    541 			mutex_enter(&nfsd_lock);
    542 			while (nfsd->nfsd_slp == NULL &&
    543 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
    544 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
    545 				    nfsd_idle);
    546 				error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock);
    547 				if (error) {
    548 					slp = nfsd->nfsd_slp;
    549 					nfsd->nfsd_slp = NULL;
    550 					if (!slp)
    551 						SLIST_REMOVE(&nfsd_idle_head,
    552 						    nfsd, nfsd, nfsd_idle);
    553 					mutex_exit(&nfsd_lock);
    554 					if (slp) {
    555 						nfsrv_wakenfsd(slp);
    556 						nfsrv_slpderef(slp);
    557 					}
    558 					goto done;
    559 				}
    560 			}
    561 			if (nfsd->nfsd_slp == NULL &&
    562 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
    563 				slp = TAILQ_FIRST(&nfssvc_sockpending);
    564 				if (slp) {
    565 					KASSERT((slp->ns_gflags & SLP_G_DOREC)
    566 					    != 0);
    567 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
    568 					    ns_pending);
    569 					slp->ns_gflags &= ~SLP_G_DOREC;
    570 					slp->ns_sref++;
    571 					nfsd->nfsd_slp = slp;
    572 				} else
    573 					nfsd_head_flag &= ~NFSD_CHECKSLP;
    574 			}
    575 			KASSERT(nfsd->nfsd_slp == NULL ||
    576 			    nfsd->nfsd_slp->ns_sref > 0);
    577 			mutex_exit(&nfsd_lock);
    578 			if ((slp = nfsd->nfsd_slp) == NULL)
    579 				continue;
    580 			if (slp->ns_flags & SLP_VALID) {
    581 				bool more;
    582 
    583 				if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) {
    584 					nfsrv_rcv(slp);
    585 				}
    586 				if (nfsdsock_testbits(slp, SLP_A_DISCONN)) {
    587 					nfsrv_zapsock(slp);
    588 				}
    589 				error = nfsrv_dorec(slp, nfsd, &nd, &more);
    590 				getmicrotime(&tv);
    591 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    592 					(u_quad_t)tv.tv_usec;
    593 				writes_todo = 0;
    594 				if (error) {
    595 					struct nfsrv_descript *nd2;
    596 
    597 					mutex_enter(&nfsd_lock);
    598 					nd2 = LIST_FIRST(&slp->ns_tq);
    599 					if (nd2 != NULL &&
    600 					    nd2->nd_time <= cur_usec) {
    601 						error = 0;
    602 						cacherep = RC_DOIT;
    603 						writes_todo = 1;
    604 					}
    605 					mutex_exit(&nfsd_lock);
    606 				}
    607 				if (error == 0 && more) {
    608 					nfsrv_wakenfsd(slp);
    609 				}
    610 			}
    611 		} else {
    612 			error = 0;
    613 			slp = nfsd->nfsd_slp;
    614 		}
    615 		KASSERT(slp != NULL);
    616 		KASSERT(nfsd->nfsd_slp == slp);
    617 		if (error || (slp->ns_flags & SLP_VALID) == 0) {
    618 			if (nd) {
    619 				nfsdreq_free(nd);
    620 				nd = NULL;
    621 			}
    622 			nfsd->nfsd_slp = NULL;
    623 			nfsrv_slpderef(slp);
    624 			continue;
    625 		}
    626 		sotype = slp->ns_so->so_type;
    627 		if (nd) {
    628 			getmicrotime(&nd->nd_starttime);
    629 			if (nd->nd_nam2)
    630 				nd->nd_nam = nd->nd_nam2;
    631 			else
    632 				nd->nd_nam = slp->ns_nam;
    633 
    634 			/*
    635 			 * Check to see if authorization is needed.
    636 			 */
    637 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
    638 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
    639 				nsd->nsd_haddr = mtod(nd->nd_nam,
    640 				    struct sockaddr_in *)->sin_addr.s_addr;
    641 				nsd->nsd_authlen = nfsd->nfsd_authlen;
    642 				nsd->nsd_verflen = nfsd->nfsd_verflen;
    643 				if (!copyout(nfsd->nfsd_authstr,
    644 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
    645 				    !copyout(nfsd->nfsd_verfstr,
    646 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
    647 				    !copyout(nsd, argp, sizeof (*nsd))) {
    648 					uvm_lwp_rele(l);
    649 					return (ENEEDAUTH);
    650 				}
    651 				cacherep = RC_DROPIT;
    652 			} else
    653 				cacherep = nfsrv_getcache(nd, slp, &mreq);
    654 
    655 			if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
    656 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
    657 				nd->nd_procnum = NFSPROC_NOOP;
    658 				nd->nd_repstat =
    659 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
    660 				cacherep = RC_DOIT;
    661 			}
    662 		}
    663 
    664 		/*
    665 		 * Loop to get all the write rpc relies that have been
    666 		 * gathered together.
    667 		 */
    668 		do {
    669 			switch (cacherep) {
    670 			case RC_DOIT:
    671 				mreq = NULL;
    672 				netexport_rdlock();
    673 				if (writes_todo || nd == NULL ||
    674 				     (!(nd->nd_flag & ND_NFSV3) &&
    675 				     nd->nd_procnum == NFSPROC_WRITE &&
    676 				     nfsrvw_procrastinate > 0))
    677 					error = nfsrv_writegather(&nd, slp,
    678 					    l, &mreq);
    679 				else
    680 					error =
    681 					    (*(nfsrv3_procs[nd->nd_procnum]))
    682 					    (nd, slp, l, &mreq);
    683 				netexport_rdunlock();
    684 				if (mreq == NULL) {
    685 					if (nd != NULL) {
    686 						if (nd->nd_nam2)
    687 							m_free(nd->nd_nam2);
    688 					}
    689 					break;
    690 				}
    691 				if (error) {
    692 					nfsstats.srv_errs++;
    693 					nfsrv_updatecache(nd, false, mreq);
    694 					if (nd->nd_nam2)
    695 						m_freem(nd->nd_nam2);
    696 					break;
    697 				}
    698 				nfsstats.srvrpccnt[nd->nd_procnum]++;
    699 				nfsrv_updatecache(nd, true, mreq);
    700 				nd->nd_mrep = (struct mbuf *)0;
    701 			case RC_REPLY:
    702 				m = mreq;
    703 				siz = 0;
    704 				while (m) {
    705 					siz += m->m_len;
    706 					m = m->m_next;
    707 				}
    708 				if (siz <= 0 || siz > NFS_MAXPACKET) {
    709 					printf("mbuf siz=%d\n",siz);
    710 					panic("Bad nfs svc reply");
    711 				}
    712 				m = mreq;
    713 				m->m_pkthdr.len = siz;
    714 				m->m_pkthdr.rcvif = (struct ifnet *)0;
    715 				/*
    716 				 * For stream protocols, prepend a Sun RPC
    717 				 * Record Mark.
    718 				 */
    719 				if (sotype == SOCK_STREAM) {
    720 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
    721 					*mtod(m, u_int32_t *) =
    722 					    htonl(0x80000000 | siz);
    723 				}
    724 				nd->nd_mreq = m;
    725 				if (nfsrtton) {
    726 					nfsd_rt(slp->ns_so->so_type, nd,
    727 					    cacherep);
    728 				}
    729 				error = nfsdsock_sendreply(slp, nd);
    730 				nd = NULL;
    731 				if (error == EPIPE)
    732 					nfsrv_zapsock(slp);
    733 				if (error == EINTR || error == ERESTART) {
    734 					nfsd->nfsd_slp = NULL;
    735 					nfsrv_slpderef(slp);
    736 					goto done;
    737 				}
    738 				break;
    739 			case RC_DROPIT:
    740 				if (nfsrtton)
    741 					nfsd_rt(sotype, nd, cacherep);
    742 				m_freem(nd->nd_mrep);
    743 				m_freem(nd->nd_nam2);
    744 				break;
    745 			}
    746 			if (nd) {
    747 				nfsdreq_free(nd);
    748 				nd = NULL;
    749 			}
    750 
    751 			/*
    752 			 * Check to see if there are outstanding writes that
    753 			 * need to be serviced.
    754 			 */
    755 			getmicrotime(&tv);
    756 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    757 			    (u_quad_t)tv.tv_usec;
    758 			s = splsoftclock();
    759 			if (LIST_FIRST(&slp->ns_tq) &&
    760 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
    761 				cacherep = RC_DOIT;
    762 				writes_todo = 1;
    763 			} else
    764 				writes_todo = 0;
    765 			splx(s);
    766 		} while (writes_todo);
    767 		if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) {
    768 			nfsd->nfsd_slp = NULL;
    769 			nfsrv_slpderef(slp);
    770 		}
    771 	}
    772 done:
    773 	mutex_enter(&nfsd_lock);
    774 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
    775 	doreinit = --nfs_numnfsd == 0;
    776 	if (doreinit)
    777 		nfssvc_sockhead_flag |= SLP_INIT;
    778 	mutex_exit(&nfsd_lock);
    779 	cv_destroy(&nfsd->nfsd_cv);
    780 	kmem_free(nfsd, sizeof(*nfsd));
    781 	nsd->nsd_nfsd = NULL;
    782 	if (doreinit)
    783 		nfsrv_init(true);	/* Reinitialize everything */
    784 	uvm_lwp_rele(l);
    785 	return (error);
    786 }
    787 
    788 /*
    789  * Shut down a socket associated with an nfssvc_sock structure.
    790  * Should be called with the send lock set, if required.
    791  * The trick here is to increment the sref at the start, so that the nfsds
    792  * will stop using it and clear ns_flag at the end so that it will not be
    793  * reassigned during cleanup.
    794  *
    795  * called at splsoftnet.
    796  */
    797 void
    798 nfsrv_zapsock(slp)
    799 	struct nfssvc_sock *slp;
    800 {
    801 	struct nfsuid *nuidp, *nnuidp;
    802 	struct nfsrv_descript *nwp;
    803 	struct socket *so;
    804 	struct mbuf *m;
    805 
    806 	if (nfsdsock_drain(slp)) {
    807 		return;
    808 	}
    809 	mutex_enter(&nfsd_lock);
    810 	if (slp->ns_gflags & SLP_G_DOREC) {
    811 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
    812 		slp->ns_gflags &= ~SLP_G_DOREC;
    813 	}
    814 	mutex_exit(&nfsd_lock);
    815 
    816 	so = slp->ns_so;
    817 	KASSERT(so != NULL);
    818 	solock(so);
    819 	so->so_upcall = NULL;
    820 	so->so_upcallarg = NULL;
    821 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    822 	soshutdown(so, SHUT_RDWR);
    823 	sounlock(so);
    824 
    825 	if (slp->ns_nam)
    826 		m_free(slp->ns_nam);
    827 	m_freem(slp->ns_raw);
    828 	m = slp->ns_rec;
    829 	while (m != NULL) {
    830 		struct mbuf *n;
    831 
    832 		n = m->m_nextpkt;
    833 		m_freem(m);
    834 		m = n;
    835 	}
    836 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
    837 	    nuidp = nnuidp) {
    838 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
    839 		LIST_REMOVE(nuidp, nu_hash);
    840 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
    841 		if (nuidp->nu_flag & NU_NAM)
    842 			m_freem(nuidp->nu_nam);
    843 		kmem_free(nuidp, sizeof(*nuidp));
    844 	}
    845 	mutex_enter(&nfsd_lock);
    846 	while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) {
    847 		LIST_REMOVE(nwp, nd_tq);
    848 		mutex_exit(&nfsd_lock);
    849 		nfsdreq_free(nwp);
    850 		mutex_enter(&nfsd_lock);
    851 	}
    852 	mutex_exit(&nfsd_lock);
    853 }
    854 
    855 /*
    856  * Derefence a server socket structure. If it has no more references and
    857  * is no longer valid, you can throw it away.
    858  */
    859 void
    860 nfsrv_slpderef(slp)
    861 	struct nfssvc_sock *slp;
    862 {
    863 	uint32_t ref;
    864 
    865 	mutex_enter(&nfsd_lock);
    866 	KASSERT(slp->ns_sref > 0);
    867 	ref = --slp->ns_sref;
    868 	mutex_exit(&nfsd_lock);
    869 	if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) {
    870 		file_t *fp;
    871 
    872 		mutex_enter(&nfsd_lock);
    873 		KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0);
    874 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
    875 		mutex_exit(&nfsd_lock);
    876 
    877 		fp = slp->ns_fp;
    878 		if (fp != NULL) {
    879 			slp->ns_fp = NULL;
    880 			KASSERT(fp != NULL);
    881 			KASSERT(fp->f_data == slp->ns_so);
    882 			KASSERT(fp->f_count > 0);
    883 			closef(fp);
    884 			slp->ns_so = NULL;
    885 		}
    886 
    887 		nfsrv_sockfree(slp);
    888 	}
    889 }
    890 
    891 /*
    892  * Initialize the data structures for the server.
    893  * Handshake with any new nfsds starting up to avoid any chance of
    894  * corruption.
    895  */
    896 void
    897 nfsrv_init(terminating)
    898 	int terminating;
    899 {
    900 	struct nfssvc_sock *slp;
    901 
    902 	if (!terminating) {
    903 		mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET);
    904 		cv_init(&nfsd_initcv, "nfsdinit");
    905 	}
    906 
    907 	mutex_enter(&nfsd_lock);
    908 	if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0)
    909 		panic("nfsd init");
    910 	nfssvc_sockhead_flag |= SLP_INIT;
    911 
    912 	if (terminating) {
    913 		KASSERT(SLIST_EMPTY(&nfsd_idle_head));
    914 		KASSERT(TAILQ_EMPTY(&nfsd_head));
    915 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
    916 			mutex_exit(&nfsd_lock);
    917 			KASSERT(slp->ns_sref == 0);
    918 			slp->ns_sref++;
    919 			nfsrv_zapsock(slp);
    920 			nfsrv_slpderef(slp);
    921 			mutex_enter(&nfsd_lock);
    922 		}
    923 		KASSERT(TAILQ_EMPTY(&nfssvc_sockpending));
    924 		mutex_exit(&nfsd_lock);
    925 		nfsrv_cleancache();	/* And clear out server cache */
    926 	} else {
    927 		mutex_exit(&nfsd_lock);
    928 		nfs_pub.np_valid = 0;
    929 	}
    930 
    931 	TAILQ_INIT(&nfssvc_sockhead);
    932 	TAILQ_INIT(&nfssvc_sockpending);
    933 
    934 	TAILQ_INIT(&nfsd_head);
    935 	SLIST_INIT(&nfsd_idle_head);
    936 	nfsd_head_flag &= ~NFSD_CHECKSLP;
    937 
    938 	nfs_udpsock = nfsrv_sockalloc();
    939 
    940 #ifdef INET6
    941 	nfs_udp6sock = nfsrv_sockalloc();
    942 #endif
    943 
    944 #ifdef ISO
    945 	nfs_cltpsock = nfsrv_sockalloc();
    946 #endif
    947 
    948 	mutex_enter(&nfsd_lock);
    949 	nfssvc_sockhead_flag &= ~SLP_INIT;
    950 	cv_broadcast(&nfsd_initcv);
    951 	mutex_exit(&nfsd_lock);
    952 }
    953 
    954 /*
    955  * Add entries to the server monitor log.
    956  */
    957 static void
    958 nfsd_rt(sotype, nd, cacherep)
    959 	int sotype;
    960 	struct nfsrv_descript *nd;
    961 	int cacherep;
    962 {
    963 	struct timeval tv;
    964 	struct drt *rt;
    965 
    966 	rt = &nfsdrt.drt[nfsdrt.pos];
    967 	if (cacherep == RC_DOIT)
    968 		rt->flag = 0;
    969 	else if (cacherep == RC_REPLY)
    970 		rt->flag = DRT_CACHEREPLY;
    971 	else
    972 		rt->flag = DRT_CACHEDROP;
    973 	if (sotype == SOCK_STREAM)
    974 		rt->flag |= DRT_TCP;
    975 	if (nd->nd_flag & ND_NFSV3)
    976 		rt->flag |= DRT_NFSV3;
    977 	rt->proc = nd->nd_procnum;
    978 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
    979 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
    980 	else
    981 	    rt->ipadr = INADDR_ANY;
    982 	getmicrotime(&tv);
    983 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
    984 		(tv.tv_usec - nd->nd_starttime.tv_usec);
    985 	rt->tstamp = tv;
    986 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
    987 }
    988 #endif /* NFSSERVER */
    989 
    990 #ifdef NFS
    991 
    992 int nfs_defect = 0;
    993 /*
    994  * Asynchronous I/O threads for client nfs.
    995  * They do read-ahead and write-behind operations on the block I/O cache.
    996  * Never returns unless it fails or gets killed.
    997  */
    998 
    999 static void
   1000 nfssvc_iod(void *arg)
   1001 {
   1002 	struct buf *bp;
   1003 	struct nfs_iod *myiod;
   1004 	struct nfsmount *nmp;
   1005 
   1006 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
   1007 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
   1008 	cv_init(&myiod->nid_cv, "nfsiod");
   1009 	myiod->nid_exiting = false;
   1010 	myiod->nid_mount = NULL;
   1011 	mutex_enter(&nfs_iodlist_lock);
   1012 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
   1013 	mutex_exit(&nfs_iodlist_lock);
   1014 
   1015 	for (;;) {
   1016 		mutex_enter(&nfs_iodlist_lock);
   1017 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
   1018 		mutex_exit(&nfs_iodlist_lock);
   1019 
   1020 		mutex_enter(&myiod->nid_lock);
   1021 		while (/*CONSTCOND*/ true) {
   1022 			nmp = myiod->nid_mount;
   1023 			if (nmp) {
   1024 				myiod->nid_mount = NULL;
   1025 				break;
   1026 			}
   1027 			if (__predict_false(myiod->nid_exiting)) {
   1028 				/*
   1029 				 * drop nid_lock to preserve locking order.
   1030 				 */
   1031 				mutex_exit(&myiod->nid_lock);
   1032 				mutex_enter(&nfs_iodlist_lock);
   1033 				mutex_enter(&myiod->nid_lock);
   1034 				/*
   1035 				 * recheck nid_mount because nfs_asyncio can
   1036 				 * pick us in the meantime as we are still on
   1037 				 * nfs_iodlist_lock.
   1038 				 */
   1039 				if (myiod->nid_mount != NULL) {
   1040 					mutex_exit(&nfs_iodlist_lock);
   1041 					continue;
   1042 				}
   1043 				LIST_REMOVE(myiod, nid_idle);
   1044 				mutex_exit(&nfs_iodlist_lock);
   1045 				goto quit;
   1046 			}
   1047 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
   1048 		}
   1049 		mutex_exit(&myiod->nid_lock);
   1050 
   1051 		mutex_enter(&nmp->nm_lock);
   1052 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
   1053 			/* Take one off the front of the list */
   1054 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
   1055 			nmp->nm_bufqlen--;
   1056 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
   1057 				cv_broadcast(&nmp->nm_aiocv);
   1058 			}
   1059 			mutex_exit(&nmp->nm_lock);
   1060 			KERNEL_LOCK(1, curlwp);
   1061 			(void)nfs_doio(bp);
   1062 			KERNEL_UNLOCK_LAST(curlwp);
   1063 			mutex_enter(&nmp->nm_lock);
   1064 			/*
   1065 			 * If there are more than one iod on this mount,
   1066 			 * then defect so that the iods can be shared out
   1067 			 * fairly between the mounts
   1068 			 */
   1069 			if (nfs_defect && nmp->nm_bufqiods > 1) {
   1070 				break;
   1071 			}
   1072 		}
   1073 		KASSERT(nmp->nm_bufqiods > 0);
   1074 		nmp->nm_bufqiods--;
   1075 		mutex_exit(&nmp->nm_lock);
   1076 	}
   1077 quit:
   1078 	KASSERT(myiod->nid_mount == NULL);
   1079 	mutex_exit(&myiod->nid_lock);
   1080 
   1081 	cv_destroy(&myiod->nid_cv);
   1082 	mutex_destroy(&myiod->nid_lock);
   1083 	kmem_free(myiod, sizeof(*myiod));
   1084 
   1085 	kthread_exit(0);
   1086 }
   1087 
   1088 void
   1089 nfs_iodinit()
   1090 {
   1091 
   1092 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
   1093 	LIST_INIT(&nfs_iodlist_all);
   1094 	LIST_INIT(&nfs_iodlist_idle);
   1095 }
   1096 
   1097 int
   1098 nfs_set_niothreads(int newval)
   1099 {
   1100 	struct nfs_iod *nid;
   1101 	int error = 0;
   1102         int hold_count;
   1103 
   1104 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
   1105 
   1106 	mutex_enter(&nfs_iodlist_lock);
   1107 	/* clamp to sane range */
   1108 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
   1109 
   1110 	while (nfs_numasync != nfs_niothreads && error == 0) {
   1111 		while (nfs_numasync < nfs_niothreads) {
   1112 
   1113 			/*
   1114 			 * kthread_create can wait for pagedaemon and
   1115 			 * pagedaemon can wait for nfsiod which needs to acquire
   1116 			 * nfs_iodlist_lock.
   1117 			 */
   1118 
   1119 			mutex_exit(&nfs_iodlist_lock);
   1120 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
   1121 			    nfssvc_iod, NULL, NULL, "nfsio");
   1122 			mutex_enter(&nfs_iodlist_lock);
   1123 			if (error) {
   1124 				/* give up */
   1125 				nfs_niothreads = nfs_numasync;
   1126 				break;
   1127 			}
   1128 			nfs_numasync++;
   1129 		}
   1130 		while (nfs_numasync > nfs_niothreads) {
   1131 			nid = LIST_FIRST(&nfs_iodlist_all);
   1132 			if (nid == NULL) {
   1133 				/* iod has not started yet. */
   1134 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
   1135 				continue;
   1136 			}
   1137 			LIST_REMOVE(nid, nid_all);
   1138 			mutex_enter(&nid->nid_lock);
   1139 			KASSERT(!nid->nid_exiting);
   1140 			nid->nid_exiting = true;
   1141 			cv_signal(&nid->nid_cv);
   1142 			mutex_exit(&nid->nid_lock);
   1143 			nfs_numasync--;
   1144 		}
   1145 	}
   1146 	mutex_exit(&nfs_iodlist_lock);
   1147 
   1148 	KERNEL_LOCK(hold_count, curlwp);
   1149 	return error;
   1150 }
   1151 
   1152 /*
   1153  * Get an authorization string for the uid by having the mount_nfs sitting
   1154  * on this mount point porpous out of the kernel and do it.
   1155  */
   1156 int
   1157 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
   1158 	struct nfsmount *nmp;
   1159 	struct nfsreq *rep;
   1160 	kauth_cred_t cred;
   1161 	char **auth_str;
   1162 	int *auth_len;
   1163 	char *verf_str;
   1164 	int *verf_len;
   1165 	NFSKERBKEY_T key;		/* return session key */
   1166 {
   1167 	int error = 0;
   1168 
   1169 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
   1170 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
   1171 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
   1172 			"nfsauth1", 2 * hz);
   1173 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1174 		if (error) {
   1175 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1176 			return (error);
   1177 		}
   1178 	}
   1179 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
   1180 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
   1181 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
   1182 	nmp->nm_verfstr = verf_str;
   1183 	nmp->nm_verflen = *verf_len;
   1184 	nmp->nm_authuid = kauth_cred_geteuid(cred);
   1185 	wakeup((void *)&nmp->nm_authstr);
   1186 
   1187 	/*
   1188 	 * And wait for mount_nfs to do its stuff.
   1189 	 */
   1190 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
   1191 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
   1192 			"nfsauth2", 2 * hz);
   1193 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1194 	}
   1195 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
   1196 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
   1197 		error = EAUTH;
   1198 	}
   1199 	if (error)
   1200 		free((void *)*auth_str, M_TEMP);
   1201 	else {
   1202 		*auth_len = nmp->nm_authlen;
   1203 		*verf_len = nmp->nm_verflen;
   1204 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
   1205 	}
   1206 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
   1207 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
   1208 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
   1209 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1210 		wakeup((void *)&nmp->nm_authtype);
   1211 	}
   1212 	return (error);
   1213 }
   1214 
   1215 /*
   1216  * Get a nickname authenticator and verifier.
   1217  */
   1218 int
   1219 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
   1220     int *auth_len, char *verf_str, int verf_len)
   1221 {
   1222 	struct timeval ktvin, ktvout, tv;
   1223 	struct nfsuid *nuidp;
   1224 	u_int32_t *nickp, *verfp;
   1225 
   1226 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
   1227 
   1228 #ifdef DIAGNOSTIC
   1229 	if (verf_len < (4 * NFSX_UNSIGNED))
   1230 		panic("nfs_getnickauth verf too small");
   1231 #endif
   1232 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
   1233 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
   1234 			break;
   1235 	}
   1236 	if (!nuidp || nuidp->nu_expire < time_second)
   1237 		return (EACCES);
   1238 
   1239 	/*
   1240 	 * Move to the end of the lru list (end of lru == most recently used).
   1241 	 */
   1242 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1243 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1244 
   1245 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
   1246 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1247 	*nickp = txdr_unsigned(nuidp->nu_nickname);
   1248 	*auth_str = (char *)nickp;
   1249 	*auth_len = 2 * NFSX_UNSIGNED;
   1250 
   1251 	/*
   1252 	 * Now we must encrypt the verifier and package it up.
   1253 	 */
   1254 	verfp = (u_int32_t *)verf_str;
   1255 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1256 	getmicrotime(&tv);
   1257 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
   1258 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
   1259 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
   1260 		nuidp->nu_timestamp = tv;
   1261 	else
   1262 		nuidp->nu_timestamp.tv_usec++;
   1263 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
   1264 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
   1265 
   1266 	/*
   1267 	 * Now encrypt the timestamp verifier in ecb mode using the session
   1268 	 * key.
   1269 	 */
   1270 #ifdef NFSKERB
   1271 	XXX
   1272 #endif
   1273 
   1274 	*verfp++ = ktvout.tv_sec;
   1275 	*verfp++ = ktvout.tv_usec;
   1276 	*verfp = 0;
   1277 	return (0);
   1278 }
   1279 
   1280 /*
   1281  * Save the current nickname in a hash list entry on the mount point.
   1282  */
   1283 int
   1284 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
   1285 	struct nfsmount *nmp;
   1286 	kauth_cred_t cred;
   1287 	int len;
   1288 	NFSKERBKEY_T key;
   1289 	struct mbuf **mdp;
   1290 	char **dposp;
   1291 	struct mbuf *mrep;
   1292 {
   1293 	struct nfsuid *nuidp;
   1294 	u_int32_t *tl;
   1295 	int32_t t1;
   1296 	struct mbuf *md = *mdp;
   1297 	struct timeval ktvin, ktvout;
   1298 	u_int32_t nick;
   1299 	char *dpos = *dposp, *cp2;
   1300 	int deltasec, error = 0;
   1301 
   1302 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
   1303 
   1304 	if (len == (3 * NFSX_UNSIGNED)) {
   1305 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
   1306 		ktvin.tv_sec = *tl++;
   1307 		ktvin.tv_usec = *tl++;
   1308 		nick = fxdr_unsigned(u_int32_t, *tl);
   1309 
   1310 		/*
   1311 		 * Decrypt the timestamp in ecb mode.
   1312 		 */
   1313 #ifdef NFSKERB
   1314 		XXX
   1315 #endif
   1316 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
   1317 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
   1318 		deltasec = time_second - ktvout.tv_sec;
   1319 		if (deltasec < 0)
   1320 			deltasec = -deltasec;
   1321 		/*
   1322 		 * If ok, add it to the hash list for the mount point.
   1323 		 */
   1324 		if (deltasec <= NFS_KERBCLOCKSKEW) {
   1325 			if (nmp->nm_numuids < nuidhash_max) {
   1326 				nmp->nm_numuids++;
   1327 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
   1328 			} else {
   1329 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
   1330 				LIST_REMOVE(nuidp, nu_hash);
   1331 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
   1332 					nu_lru);
   1333 			}
   1334 			nuidp->nu_flag = 0;
   1335 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
   1336 			nuidp->nu_expire = time_second + NFS_KERBTTL;
   1337 			nuidp->nu_timestamp = ktvout;
   1338 			nuidp->nu_nickname = nick;
   1339 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
   1340 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
   1341 				nu_lru);
   1342 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
   1343 				nuidp, nu_hash);
   1344 		}
   1345 	} else
   1346 		nfsm_adv(nfsm_rndup(len));
   1347 nfsmout:
   1348 	*mdp = md;
   1349 	*dposp = dpos;
   1350 	return (error);
   1351 }
   1352 #endif /* NFS */
   1353