Home | History | Annotate | Line # | Download | only in nfs
nfs_syscalls.c revision 1.99.2.1
      1 /*	$NetBSD: nfs_syscalls.c,v 1.99.2.1 2007/06/05 20:37:21 bouyer 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 1.99.2.1 2007/06/05 20:37:21 bouyer 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/buf.h>
     58 #include <sys/mbuf.h>
     59 #include <sys/socket.h>
     60 #include <sys/socketvar.h>
     61 #include <sys/signalvar.h>
     62 #include <sys/domain.h>
     63 #include <sys/protosw.h>
     64 #include <sys/namei.h>
     65 #include <sys/syslog.h>
     66 #include <sys/filedesc.h>
     67 #include <sys/kthread.h>
     68 #include <sys/kauth.h>
     69 
     70 #include <sys/sa.h>
     71 #include <sys/syscallargs.h>
     72 
     73 #include <netinet/in.h>
     74 #include <netinet/tcp.h>
     75 #ifdef ISO
     76 #include <netiso/iso.h>
     77 #endif
     78 #include <nfs/xdr_subs.h>
     79 #include <nfs/rpcv2.h>
     80 #include <nfs/nfsproto.h>
     81 #include <nfs/nfs.h>
     82 #include <nfs/nfsm_subs.h>
     83 #include <nfs/nfsrvcache.h>
     84 #include <nfs/nfsmount.h>
     85 #include <nfs/nfsnode.h>
     86 #include <nfs/nqnfs.h>
     87 #include <nfs/nfsrtt.h>
     88 #include <nfs/nfs_var.h>
     89 
     90 /* Global defs. */
     91 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) __P((struct nfsrv_descript *,
     92 						struct nfssvc_sock *,
     93 						struct lwp *, struct mbuf **));
     94 extern time_t nqnfsstarttime;
     95 extern int nfsrvw_procrastinate;
     96 
     97 struct nfssvc_sock *nfs_udpsock;
     98 #ifdef ISO
     99 struct nfssvc_sock *nfs_cltpsock;
    100 #endif
    101 #ifdef INET6
    102 struct nfssvc_sock *nfs_udp6sock;
    103 #endif
    104 int nuidhash_max = NFS_MAXUIDHASH;
    105 int nfsd_waiting = 0;
    106 #ifdef NFSSERVER
    107 static int nfs_numnfsd = 0;
    108 static int notstarted = 1;
    109 static int modify_flag = 0;
    110 static struct nfsdrt nfsdrt;
    111 #endif
    112 
    113 #ifdef NFSSERVER
    114 struct simplelock nfsd_slock = SIMPLELOCK_INITIALIZER;
    115 struct nfssvc_sockhead nfssvc_sockhead;
    116 struct nfssvc_sockhead nfssvc_sockpending;
    117 struct nfsdhead nfsd_head;
    118 struct nfsdidlehead nfsd_idle_head;
    119 
    120 int nfssvc_sockhead_flag;
    121 int nfsd_head_flag;
    122 #endif
    123 
    124 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
    125 
    126 #ifdef NFS
    127 struct nfs_iod nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
    128 int nfs_niothreads = -1; /* == "0, and has never been set" */
    129 #endif
    130 
    131 #ifdef NFSSERVER
    132 static struct nfssvc_sock *nfsrv_sockalloc __P((void));
    133 static void nfsrv_sockfree __P((struct nfssvc_sock *));
    134 static void nfsd_rt __P((int, struct nfsrv_descript *, int));
    135 #endif
    136 
    137 /*
    138  * NFS server system calls
    139  */
    140 
    141 
    142 /*
    143  * Nfs server pseudo system call for the nfsd's
    144  * Based on the flag value it either:
    145  * - adds a socket to the selection list
    146  * - remains in the kernel as an nfsd
    147  * - remains in the kernel as an nfsiod
    148  */
    149 int
    150 sys_nfssvc(struct lwp *l, void *v, register_t *retval)
    151 {
    152 	struct sys_nfssvc_args /* {
    153 		syscallarg(int) flag;
    154 		syscallarg(caddr_t) argp;
    155 	} */ *uap = v;
    156 	int error;
    157 #ifdef NFS
    158 	struct nameidata nd;
    159 	struct nfsmount *nmp;
    160 	struct nfsd_cargs ncd;
    161 #endif
    162 #ifdef NFSSERVER
    163 	int s;
    164 	struct file *fp;
    165 	struct mbuf *nam;
    166 	struct nfsd_args nfsdarg;
    167 	struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
    168 	struct nfsd *nfsd;
    169 	struct nfssvc_sock *slp;
    170 	struct nfsuid *nuidp;
    171 #endif
    172 
    173 	/*
    174 	 * Must be super user
    175 	 */
    176 	error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
    177 	    &l->l_acflag);
    178 	if (error)
    179 		return (error);
    180 
    181 	/* Initialize NFS server / client shared data. */
    182 	nfs_init();
    183 
    184 #ifdef NFSSERVER
    185 	s = splsoftnet();
    186 	simple_lock(&nfsd_slock);
    187 	while (nfssvc_sockhead_flag & SLP_INIT) {
    188 		nfssvc_sockhead_flag |= SLP_WANTINIT;
    189 		(void) ltsleep(&nfssvc_sockhead, PSOCK, "nfsd init", 0,
    190 		    &nfsd_slock);
    191 	}
    192 	simple_unlock(&nfsd_slock);
    193 	splx(s);
    194 #endif
    195 	if (SCARG(uap, flag) & NFSSVC_BIOD) {
    196 #if defined(NFS) && defined(COMPAT_14)
    197 		error = nfssvc_iod(l);
    198 #else
    199 		error = ENOSYS;
    200 #endif
    201 	} else if (SCARG(uap, flag) & NFSSVC_MNTD) {
    202 #ifndef NFS
    203 		error = ENOSYS;
    204 #else
    205 		error = copyin(SCARG(uap, argp), (caddr_t)&ncd, sizeof (ncd));
    206 		if (error)
    207 			return (error);
    208 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE,
    209 			ncd.ncd_dirp, l);
    210 		error = namei(&nd);
    211 		if (error)
    212 			return (error);
    213 		if ((nd.ni_vp->v_flag & VROOT) == 0)
    214 			error = EINVAL;
    215 		nmp = VFSTONFS(nd.ni_vp->v_mount);
    216 		vput(nd.ni_vp);
    217 		if (error)
    218 			return (error);
    219 		if ((nmp->nm_iflag & NFSMNT_MNTD) &&
    220 			(SCARG(uap, flag) & NFSSVC_GOTAUTH) == 0)
    221 			return (0);
    222 		nmp->nm_iflag |= NFSMNT_MNTD;
    223 		error = nqnfs_clientd(nmp, l->l_cred, &ncd, SCARG(uap, flag),
    224 			SCARG(uap, argp), l);
    225 #endif /* NFS */
    226 	} else if (SCARG(uap, flag) & NFSSVC_ADDSOCK) {
    227 #ifndef NFSSERVER
    228 		error = ENOSYS;
    229 #else
    230 		error = copyin(SCARG(uap, argp), (caddr_t)&nfsdarg,
    231 		    sizeof(nfsdarg));
    232 		if (error)
    233 			return (error);
    234 		/* getsock() will use the descriptor for us */
    235 		error = getsock(l->l_proc->p_fd, nfsdarg.sock, &fp);
    236 		if (error)
    237 			return (error);
    238 		/*
    239 		 * Get the client address for connected sockets.
    240 		 */
    241 		if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
    242 			nam = (struct mbuf *)0;
    243 		else {
    244 			error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen,
    245 				MT_SONAME);
    246 			if (error) {
    247 				FILE_UNUSE(fp, NULL);
    248 				return (error);
    249 			}
    250 		}
    251 		error = nfssvc_addsock(fp, nam);
    252 		FILE_UNUSE(fp, NULL);
    253 #endif /* !NFSSERVER */
    254 	} else if (SCARG(uap, flag) & NFSSVC_SETEXPORTSLIST) {
    255 #ifndef NFSSERVER
    256 		error = ENOSYS;
    257 #else
    258 		struct export_args *args;
    259 		struct mountd_exports_list mel;
    260 
    261 		error = copyin(SCARG(uap, argp), &mel, sizeof(mel));
    262 		if (error != 0)
    263 			return error;
    264 
    265 		args = (struct export_args *)malloc(mel.mel_nexports *
    266 		    sizeof(struct export_args), M_TEMP, M_WAITOK);
    267 		error = copyin(mel.mel_exports, args, mel.mel_nexports *
    268 		    sizeof(struct export_args));
    269 		if (error != 0) {
    270 			free(args, M_TEMP);
    271 			return error;
    272 		}
    273 		mel.mel_exports = args;
    274 
    275 		error = mountd_set_exports_list(&mel, l);
    276 
    277 		free(args, M_TEMP);
    278 #endif /* !NFSSERVER */
    279 	} else {
    280 #ifndef NFSSERVER
    281 		error = ENOSYS;
    282 #else
    283 		error = copyin(SCARG(uap, argp), (caddr_t)nsd, sizeof (*nsd));
    284 		if (error)
    285 			return (error);
    286 		if ((SCARG(uap, flag) & NFSSVC_AUTHIN) &&
    287 		    ((nfsd = nsd->nsd_nfsd)) != NULL &&
    288 		    (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
    289 			slp = nfsd->nfsd_slp;
    290 
    291 			/*
    292 			 * First check to see if another nfsd has already
    293 			 * added this credential.
    294 			 */
    295 			LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid),
    296 			    nu_hash) {
    297 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
    298 				    nsd->nsd_cr.cr_uid &&
    299 				    (!nfsd->nfsd_nd->nd_nam2 ||
    300 				     netaddr_match(NU_NETFAM(nuidp),
    301 				     &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
    302 					break;
    303 			}
    304 			if (nuidp) {
    305 			    kauth_cred_hold(nuidp->nu_cr);
    306 			    nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    307 			    nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    308 			} else {
    309 			    /*
    310 			     * Nope, so we will.
    311 			     */
    312 			    if (slp->ns_numuids < nuidhash_max) {
    313 				slp->ns_numuids++;
    314 				nuidp = (struct nfsuid *)
    315 				   malloc(sizeof (struct nfsuid), M_NFSUID,
    316 					M_WAITOK);
    317 			    } else
    318 				nuidp = (struct nfsuid *)0;
    319 			    if ((slp->ns_flag & SLP_VALID) == 0) {
    320 				if (nuidp)
    321 				    free((caddr_t)nuidp, M_NFSUID);
    322 			    } else {
    323 				if (nuidp == (struct nfsuid *)0) {
    324 				    nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
    325 				    LIST_REMOVE(nuidp, nu_hash);
    326 				    TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
    327 					nu_lru);
    328 				    if (nuidp->nu_flag & NU_NAM)
    329 					m_freem(nuidp->nu_nam);
    330 			        }
    331 				nuidp->nu_flag = 0;
    332 				kauth_uucred_to_cred(nuidp->nu_cr,
    333 				    &nsd->nsd_cr);
    334 				nuidp->nu_timestamp = nsd->nsd_timestamp;
    335 				nuidp->nu_expire = time_second + nsd->nsd_ttl;
    336 				/*
    337 				 * and save the session key in nu_key.
    338 				 */
    339 				memcpy(nuidp->nu_key, nsd->nsd_key,
    340 				    sizeof(nsd->nsd_key));
    341 				if (nfsd->nfsd_nd->nd_nam2) {
    342 				    struct sockaddr_in *saddr;
    343 
    344 				    saddr = mtod(nfsd->nfsd_nd->nd_nam2,
    345 					 struct sockaddr_in *);
    346 				    switch (saddr->sin_family) {
    347 				    case AF_INET:
    348 					nuidp->nu_flag |= NU_INETADDR;
    349 					nuidp->nu_inetaddr =
    350 					     saddr->sin_addr.s_addr;
    351 					break;
    352 				    case AF_ISO:
    353 				    default:
    354 					nuidp->nu_flag |= NU_NAM;
    355 					nuidp->nu_nam = m_copym(
    356 					    nfsd->nfsd_nd->nd_nam2, 0,
    357 					     M_COPYALL, M_WAIT);
    358 					break;
    359 				    };
    360 				}
    361 				TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
    362 					nu_lru);
    363 				LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
    364 					nuidp, nu_hash);
    365 				kauth_cred_hold(nuidp->nu_cr);
    366 				nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    367 				nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    368 			    }
    369 			}
    370 		}
    371 		if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) &&
    372 		    (nfsd = nsd->nsd_nfsd))
    373 			nfsd->nfsd_flag |= NFSD_AUTHFAIL;
    374 		error = nfssvc_nfsd(nsd, SCARG(uap, argp), l);
    375 #endif /* !NFSSERVER */
    376 	}
    377 	if (error == EINTR || error == ERESTART)
    378 		error = 0;
    379 	return (error);
    380 }
    381 
    382 #ifdef NFSSERVER
    383 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
    384 MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
    385 
    386 static struct nfssvc_sock *
    387 nfsrv_sockalloc()
    388 {
    389 	struct nfssvc_sock *slp;
    390 	int s;
    391 
    392 	slp = (struct nfssvc_sock *)
    393 	    malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK);
    394 	memset(slp, 0, sizeof (struct nfssvc_sock));
    395 	simple_lock_init(&slp->ns_lock);
    396 	TAILQ_INIT(&slp->ns_uidlruhead);
    397 	LIST_INIT(&slp->ns_tq);
    398 	SIMPLEQ_INIT(&slp->ns_sendq);
    399 	s = splsoftnet();
    400 	simple_lock(&nfsd_slock);
    401 	TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
    402 	simple_unlock(&nfsd_slock);
    403 	splx(s);
    404 
    405 	return slp;
    406 }
    407 
    408 static void
    409 nfsrv_sockfree(struct nfssvc_sock *slp)
    410 {
    411 
    412 	KASSERT(slp->ns_so == NULL);
    413 	KASSERT(slp->ns_fp == NULL);
    414 	KASSERT((slp->ns_flag & SLP_VALID) == 0);
    415 	free(slp, M_NFSSVC);
    416 }
    417 
    418 /*
    419  * Adds a socket to the list for servicing by nfsds.
    420  */
    421 int
    422 nfssvc_addsock(fp, mynam)
    423 	struct file *fp;
    424 	struct mbuf *mynam;
    425 {
    426 	struct mbuf *m;
    427 	int siz;
    428 	struct nfssvc_sock *slp;
    429 	struct socket *so;
    430 	struct nfssvc_sock *tslp;
    431 	int error, s;
    432 
    433 	so = (struct socket *)fp->f_data;
    434 	tslp = (struct nfssvc_sock *)0;
    435 	/*
    436 	 * Add it to the list, as required.
    437 	 */
    438 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
    439 #ifdef INET6
    440 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
    441 			tslp = nfs_udp6sock;
    442 		else
    443 #endif
    444 		tslp = nfs_udpsock;
    445 		if (tslp->ns_flag & SLP_VALID) {
    446 			m_freem(mynam);
    447 			return (EPERM);
    448 		}
    449 #ifdef ISO
    450 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
    451 		tslp = nfs_cltpsock;
    452 		if (tslp->ns_flag & SLP_VALID) {
    453 			m_freem(mynam);
    454 			return (EPERM);
    455 		}
    456 #endif /* ISO */
    457 	}
    458 	if (so->so_type == SOCK_STREAM)
    459 		siz = NFS_MAXPACKET + sizeof (u_long);
    460 	else
    461 		siz = NFS_MAXPACKET;
    462 	error = soreserve(so, siz, siz);
    463 	if (error) {
    464 		m_freem(mynam);
    465 		return (error);
    466 	}
    467 
    468 	/*
    469 	 * Set protocol specific options { for now TCP only } and
    470 	 * reserve some space. For datagram sockets, this can get called
    471 	 * repeatedly for the same socket, but that isn't harmful.
    472 	 */
    473 	if (so->so_type == SOCK_STREAM) {
    474 		m = m_get(M_WAIT, MT_SOOPTS);
    475 		MCLAIM(m, &nfs_mowner);
    476 		*mtod(m, int32_t *) = 1;
    477 		m->m_len = sizeof(int32_t);
    478 		sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
    479 	}
    480 	if ((so->so_proto->pr_domain->dom_family == AF_INET
    481 #ifdef INET6
    482 	    || so->so_proto->pr_domain->dom_family == AF_INET6
    483 #endif
    484 	    ) &&
    485 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
    486 		m = m_get(M_WAIT, MT_SOOPTS);
    487 		MCLAIM(m, &nfs_mowner);
    488 		*mtod(m, int32_t *) = 1;
    489 		m->m_len = sizeof(int32_t);
    490 		sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
    491 	}
    492 	so->so_rcv.sb_flags &= ~SB_NOINTR;
    493 	so->so_rcv.sb_timeo = 0;
    494 	so->so_snd.sb_flags &= ~SB_NOINTR;
    495 	so->so_snd.sb_timeo = 0;
    496 	if (tslp)
    497 		slp = tslp;
    498 	else {
    499 		slp = nfsrv_sockalloc();
    500 	}
    501 	slp->ns_so = so;
    502 	slp->ns_nam = mynam;
    503 	fp->f_count++;
    504 	slp->ns_fp = fp;
    505 	s = splsoftnet();
    506 	so->so_upcallarg = (caddr_t)slp;
    507 	so->so_upcall = nfsrv_rcv;
    508 	so->so_rcv.sb_flags |= SB_UPCALL;
    509 	slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
    510 	nfsrv_wakenfsd(slp);
    511 	splx(s);
    512 	return (0);
    513 }
    514 
    515 /*
    516  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    517  * until it is killed by a signal.
    518  */
    519 int
    520 nfssvc_nfsd(nsd, argp, l)
    521 	struct nfsd_srvargs *nsd;
    522 	caddr_t argp;
    523 	struct lwp *l;
    524 {
    525 	struct timeval tv;
    526 	struct mbuf *m;
    527 	struct nfssvc_sock *slp;
    528 	struct nfsd *nfsd = nsd->nsd_nfsd;
    529 	struct nfsrv_descript *nd = NULL;
    530 	struct mbuf *mreq;
    531 	u_quad_t cur_usec;
    532 	int error = 0, cacherep, s, siz, sotype, writes_todo;
    533 	struct proc *p = l->l_proc;
    534 
    535 #ifndef nolint
    536 	cacherep = RC_DOIT;
    537 	writes_todo = 0;
    538 #endif
    539 	s = splsoftnet();
    540 	if (nfsd == (struct nfsd *)0) {
    541 		nsd->nsd_nfsd = nfsd = (struct nfsd *)
    542 			malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK);
    543 		memset((caddr_t)nfsd, 0, sizeof (struct nfsd));
    544 		nfsd->nfsd_procp = p;
    545 		simple_lock(&nfsd_slock);
    546 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
    547 		nfs_numnfsd++;
    548 		simple_unlock(&nfsd_slock);
    549 	}
    550 	PHOLD(l);
    551 	/*
    552 	 * Loop getting rpc requests until SIGKILL.
    553 	 */
    554 	for (;;) {
    555 		if (nfsd->nfsd_slp == NULL) {
    556 			simple_lock(&nfsd_slock);
    557 			while (nfsd->nfsd_slp == NULL &&
    558 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
    559 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
    560 				    nfsd_idle);
    561 				nfsd_waiting++;
    562 				error = ltsleep(nfsd, PSOCK | PCATCH, "nfsd",
    563 				    0, &nfsd_slock);
    564 				nfsd_waiting--;
    565 				if (error) {
    566 					slp = nfsd->nfsd_slp;
    567 					nfsd->nfsd_slp = NULL;
    568 					if (!slp)
    569 						SLIST_REMOVE(&nfsd_idle_head,
    570 						    nfsd, nfsd, nfsd_idle);
    571 					simple_unlock(&nfsd_slock);
    572 					if (slp) {
    573 						nfsrv_wakenfsd(slp);
    574 						nfsrv_slpderef(slp);
    575 					}
    576 					goto done;
    577 				}
    578 			}
    579 			if (nfsd->nfsd_slp == NULL &&
    580 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
    581 				slp = TAILQ_FIRST(&nfssvc_sockpending);
    582 				if (slp) {
    583 					KASSERT((slp->ns_flag &
    584 					    (SLP_VALID | SLP_DOREC))
    585 					    == (SLP_VALID | SLP_DOREC));
    586 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
    587 					    ns_pending);
    588 					slp->ns_flag &= ~SLP_DOREC;
    589 					slp->ns_sref++;
    590 					nfsd->nfsd_slp = slp;
    591 				} else
    592 					nfsd_head_flag &= ~NFSD_CHECKSLP;
    593 			}
    594 			simple_unlock(&nfsd_slock);
    595 			if ((slp = nfsd->nfsd_slp) == NULL)
    596 				continue;
    597 			if (slp->ns_flag & SLP_VALID) {
    598 				if (slp->ns_flag & SLP_DISCONN)
    599 					nfsrv_zapsock(slp);
    600 				else if ((slp->ns_flag & SLP_NEEDQ) != 0) {
    601 					nfsrv_rcv(slp->ns_so, (void *)slp,
    602 					    M_WAIT);
    603 				}
    604 				error = nfsrv_dorec(slp, nfsd, &nd);
    605 				getmicrotime(&tv);
    606 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    607 					(u_quad_t)tv.tv_usec;
    608 				if (error && LIST_FIRST(&slp->ns_tq) &&
    609 				    LIST_FIRST(&slp->ns_tq)->nd_time <=
    610 				    cur_usec) {
    611 					error = 0;
    612 					cacherep = RC_DOIT;
    613 					writes_todo = 1;
    614 				} else
    615 					writes_todo = 0;
    616 				if (error == 0 && slp->ns_rec != NULL) {
    617 					nfsrv_wakenfsd(slp);
    618 				}
    619 			}
    620 		} else {
    621 			error = 0;
    622 			slp = nfsd->nfsd_slp;
    623 		}
    624 		if (error || (slp->ns_flag & SLP_VALID) == 0) {
    625 			if (nd) {
    626 				nfsdreq_free(nd);
    627 				nd = NULL;
    628 			}
    629 			nfsd->nfsd_slp = NULL;
    630 			nfsrv_slpderef(slp);
    631 			continue;
    632 		}
    633 		splx(s);
    634 		sotype = slp->ns_so->so_type;
    635 		if (nd) {
    636 			getmicrotime(&nd->nd_starttime);
    637 			if (nd->nd_nam2)
    638 				nd->nd_nam = nd->nd_nam2;
    639 			else
    640 				nd->nd_nam = slp->ns_nam;
    641 
    642 			/*
    643 			 * Check to see if authorization is needed.
    644 			 */
    645 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
    646 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
    647 				nsd->nsd_haddr = mtod(nd->nd_nam,
    648 				    struct sockaddr_in *)->sin_addr.s_addr;
    649 				nsd->nsd_authlen = nfsd->nfsd_authlen;
    650 				nsd->nsd_verflen = nfsd->nfsd_verflen;
    651 				if (!copyout(nfsd->nfsd_authstr,
    652 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
    653 				    !copyout(nfsd->nfsd_verfstr,
    654 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
    655 				    !copyout(nsd, argp, sizeof (*nsd))) {
    656 					PRELE(l);
    657 					return (ENEEDAUTH);
    658 				}
    659 				cacherep = RC_DROPIT;
    660 			} else
    661 				cacherep = nfsrv_getcache(nd, slp, &mreq);
    662 
    663 			/*
    664 			 * Check for just starting up for NQNFS and send
    665 			 * fake "try again later" replies to the NQNFS clients.
    666 			 */
    667 			if (notstarted && nqnfsstarttime <= time_second) {
    668 				if (modify_flag) {
    669 					nqnfsstarttime =
    670 					    time_second + nqsrv_writeslack;
    671 					modify_flag = 0;
    672 				} else
    673 					notstarted = 0;
    674 			}
    675 			if (notstarted) {
    676 				if ((nd->nd_flag & ND_NQNFS) == 0)
    677 					cacherep = RC_DROPIT;
    678 				else if (nd->nd_procnum != NFSPROC_WRITE) {
    679 					nd->nd_procnum = NFSPROC_NOOP;
    680 					nd->nd_repstat = NQNFS_TRYLATER;
    681 					cacherep = RC_DOIT;
    682 				} else
    683 					modify_flag = 1;
    684 			} else if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
    685 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
    686 				nd->nd_procnum = NFSPROC_NOOP;
    687 				nd->nd_repstat =
    688 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
    689 				cacherep = RC_DOIT;
    690 			}
    691 		}
    692 
    693 		/*
    694 		 * Loop to get all the write rpc relies that have been
    695 		 * gathered together.
    696 		 */
    697 		do {
    698 #ifdef DIAGNOSTIC
    699 			int lockcount;
    700 #endif
    701 			switch (cacherep) {
    702 			case RC_DOIT:
    703 #ifdef DIAGNOSTIC
    704 				/*
    705 				 * NFS server procs should neither release
    706 				 * locks already held, nor leave things
    707 				 * locked.  Catch this sooner, rather than
    708 				 * later (when we try to relock something we
    709 				 * already have locked).  Careful inspection
    710 				 * of the failing routine usually turns up the
    711 				 * lock leak.. once we know what it is..
    712 				 */
    713 				lockcount = l->l_locks;
    714 #endif
    715 				mreq = NULL;
    716 				netexport_rdlock();
    717 				if (writes_todo || nd == NULL ||
    718 				     (!(nd->nd_flag & ND_NFSV3) &&
    719 				     nd->nd_procnum == NFSPROC_WRITE &&
    720 				     nfsrvw_procrastinate > 0 && !notstarted))
    721 					error = nfsrv_writegather(&nd, slp,
    722 					    l, &mreq);
    723 				else
    724 					error =
    725 					    (*(nfsrv3_procs[nd->nd_procnum]))
    726 					    (nd, slp, l, &mreq);
    727 				netexport_rdunlock();
    728 #ifdef DIAGNOSTIC
    729 				if (l->l_locks != lockcount) {
    730 					/*
    731 					 * If you see this panic, audit
    732 					 * nfsrv3_procs[nd->nd_procnum] for
    733 					 * vnode locking errors (usually, it's
    734 					 * due to forgetting to vput()
    735 					 * something).
    736 					 */
    737 #ifdef DEBUG
    738 					extern void printlockedvnodes(void);
    739 					printlockedvnodes();
    740 #endif
    741 					printf("nfsd: locking botch in op %d"
    742 					    " (before %d, after %d)\n",
    743 					    nd ? nd->nd_procnum : -1,
    744 					    lockcount, l->l_locks);
    745 				}
    746 #endif
    747 				if (mreq == NULL) {
    748 					if (nd != NULL) {
    749 						if (nd->nd_nam2)
    750 							m_free(nd->nd_nam2);
    751 						if (nd->nd_mrep)
    752 							m_freem(nd->nd_mrep);
    753 					}
    754 					break;
    755 				}
    756 				if (error) {
    757 					if (nd->nd_procnum != NQNFSPROC_VACATED)
    758 						nfsstats.srv_errs++;
    759 					nfsrv_updatecache(nd, FALSE, mreq);
    760 					if (nd->nd_nam2)
    761 						m_freem(nd->nd_nam2);
    762 					break;
    763 				}
    764 				nfsstats.srvrpccnt[nd->nd_procnum]++;
    765 				nfsrv_updatecache(nd, TRUE, mreq);
    766 				nd->nd_mrep = (struct mbuf *)0;
    767 			case RC_REPLY:
    768 				m = mreq;
    769 				siz = 0;
    770 				while (m) {
    771 					siz += m->m_len;
    772 					m = m->m_next;
    773 				}
    774 				if (siz <= 0 || siz > NFS_MAXPACKET) {
    775 					printf("mbuf siz=%d\n",siz);
    776 					panic("Bad nfs svc reply");
    777 				}
    778 				m = mreq;
    779 				m->m_pkthdr.len = siz;
    780 				m->m_pkthdr.rcvif = (struct ifnet *)0;
    781 				/*
    782 				 * For stream protocols, prepend a Sun RPC
    783 				 * Record Mark.
    784 				 */
    785 				if (sotype == SOCK_STREAM) {
    786 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
    787 					*mtod(m, u_int32_t *) =
    788 					    htonl(0x80000000 | siz);
    789 				}
    790 				nd->nd_mreq = m;
    791 				if (nfsrtton) {
    792 					nfsd_rt(slp->ns_so->so_type, nd,
    793 					    cacherep);
    794 				}
    795 				s = splsoftnet();
    796 				error = nfsdsock_sendreply(slp, nd);
    797 				nd = NULL;
    798 				if (error == EPIPE)
    799 					nfsrv_zapsock(slp);
    800 				if (error == EINTR || error == ERESTART) {
    801 					nfsrv_slpderef(slp);
    802 					goto done;
    803 				}
    804 				splx(s);
    805 				break;
    806 			case RC_DROPIT:
    807 				if (nfsrtton)
    808 					nfsd_rt(sotype, nd, cacherep);
    809 				m_freem(nd->nd_mrep);
    810 				m_freem(nd->nd_nam2);
    811 				break;
    812 			}
    813 			if (nd) {
    814 				nfsdreq_free(nd);
    815 				nd = NULL;
    816 			}
    817 
    818 			/*
    819 			 * Check to see if there are outstanding writes that
    820 			 * need to be serviced.
    821 			 */
    822 			getmicrotime(&tv);
    823 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    824 			    (u_quad_t)tv.tv_usec;
    825 			s = splsoftclock();
    826 			if (LIST_FIRST(&slp->ns_tq) &&
    827 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
    828 				cacherep = RC_DOIT;
    829 				writes_todo = 1;
    830 			} else
    831 				writes_todo = 0;
    832 			splx(s);
    833 		} while (writes_todo);
    834 		s = splsoftnet();
    835 		if (nfsrv_dorec(slp, nfsd, &nd)) {
    836 			nfsd->nfsd_slp = NULL;
    837 			nfsrv_slpderef(slp);
    838 		}
    839 	}
    840 done:
    841 	PRELE(l);
    842 	simple_lock(&nfsd_slock);
    843 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
    844 	simple_unlock(&nfsd_slock);
    845 	splx(s);
    846 	free((caddr_t)nfsd, M_NFSD);
    847 	nsd->nsd_nfsd = (struct nfsd *)0;
    848 	if (--nfs_numnfsd == 0)
    849 		nfsrv_init(TRUE);	/* Reinitialize everything */
    850 	return (error);
    851 }
    852 
    853 /*
    854  * Shut down a socket associated with an nfssvc_sock structure.
    855  * Should be called with the send lock set, if required.
    856  * The trick here is to increment the sref at the start, so that the nfsds
    857  * will stop using it and clear ns_flag at the end so that it will not be
    858  * reassigned during cleanup.
    859  *
    860  * called at splsoftnet.
    861  */
    862 void
    863 nfsrv_zapsock(slp)
    864 	struct nfssvc_sock *slp;
    865 {
    866 	struct nfsuid *nuidp, *nnuidp;
    867 	struct nfsrv_descript *nwp, *nnwp;
    868 	struct socket *so;
    869 	struct mbuf *m;
    870 	int s;
    871 
    872 	if (nfsdsock_drain(slp)) {
    873 		return;
    874 	}
    875 	simple_lock(&nfsd_slock);
    876 	if (slp->ns_flag & SLP_DOREC) {
    877 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
    878 	}
    879 	simple_unlock(&nfsd_slock);
    880 
    881 	so = slp->ns_so;
    882 	KASSERT(so != NULL);
    883 	so->so_upcall = NULL;
    884 	so->so_upcallarg = NULL;
    885 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    886 	soshutdown(so, SHUT_RDWR);
    887 
    888 	if (slp->ns_nam)
    889 		m_free(slp->ns_nam);
    890 	m_freem(slp->ns_raw);
    891 	m = slp->ns_rec;
    892 	while (m != NULL) {
    893 		struct mbuf *n;
    894 
    895 		n = m->m_nextpkt;
    896 		m_freem(m);
    897 		m = n;
    898 	}
    899 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
    900 	    nuidp = nnuidp) {
    901 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
    902 		LIST_REMOVE(nuidp, nu_hash);
    903 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
    904 		if (nuidp->nu_flag & NU_NAM)
    905 			m_freem(nuidp->nu_nam);
    906 		free((caddr_t)nuidp, M_NFSUID);
    907 	}
    908 	s = splsoftclock();
    909 	for (nwp = LIST_FIRST(&slp->ns_tq); nwp; nwp = nnwp) {
    910 		nnwp = LIST_NEXT(nwp, nd_tq);
    911 		LIST_REMOVE(nwp, nd_tq);
    912 		nfsdreq_free(nwp);
    913 	}
    914 	splx(s);
    915 }
    916 
    917 /*
    918  * Derefence a server socket structure. If it has no more references and
    919  * is no longer valid, you can throw it away.
    920  */
    921 void
    922 nfsrv_slpderef(slp)
    923 	struct nfssvc_sock *slp;
    924 {
    925 	LOCK_ASSERT(!simple_lock_held(&nfsd_slock));
    926 
    927 	if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
    928 		struct file *fp;
    929 		int s = splsoftnet();
    930 		simple_lock(&nfsd_slock);
    931 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
    932 		simple_unlock(&nfsd_slock);
    933 		splx(s);
    934 
    935 		fp = slp->ns_fp;
    936 		if (fp != NULL) {
    937 			slp->ns_fp = NULL;
    938 			KASSERT(fp != NULL);
    939 			KASSERT(fp->f_data == slp->ns_so);
    940 			simple_lock(&fp->f_slock);
    941 			FILE_USE(fp);
    942 			closef(fp, (struct lwp *)0);
    943 			slp->ns_so = NULL;
    944 		}
    945 
    946 		nfsrv_sockfree(slp);
    947 	}
    948 }
    949 
    950 /*
    951  * Initialize the data structures for the server.
    952  * Handshake with any new nfsds starting up to avoid any chance of
    953  * corruption.
    954  */
    955 void
    956 nfsrv_init(terminating)
    957 	int terminating;
    958 {
    959 	struct nfssvc_sock *slp;
    960 	int s;
    961 
    962 	s = splsoftnet();
    963 	simple_lock(&nfsd_slock);
    964 	if (nfssvc_sockhead_flag & SLP_INIT)
    965 		panic("nfsd init");
    966 	nfssvc_sockhead_flag |= SLP_INIT;
    967 
    968 	if (terminating) {
    969 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
    970 			simple_unlock(&nfsd_slock);
    971 			KASSERT(slp->ns_sref == 0);
    972 			slp->ns_sref++;
    973 			nfsrv_zapsock(slp);
    974 			nfsrv_slpderef(slp);
    975 			simple_lock(&nfsd_slock);
    976 		}
    977 		simple_unlock(&nfsd_slock);
    978 		splx(s);
    979 		nfsrv_cleancache();	/* And clear out server cache */
    980 	} else {
    981 		simple_unlock(&nfsd_slock);
    982 		splx(s);
    983 		nfs_pub.np_valid = 0;
    984 	}
    985 
    986 	TAILQ_INIT(&nfssvc_sockhead);
    987 	TAILQ_INIT(&nfssvc_sockpending);
    988 	nfssvc_sockhead_flag &= ~SLP_INIT;
    989 
    990 	TAILQ_INIT(&nfsd_head);
    991 	SLIST_INIT(&nfsd_idle_head);
    992 	nfsd_head_flag &= ~NFSD_CHECKSLP;
    993 
    994 	nfs_udpsock = nfsrv_sockalloc();
    995 
    996 #ifdef INET6
    997 	nfs_udp6sock = nfsrv_sockalloc();
    998 #endif
    999 
   1000 #ifdef ISO
   1001 	nfs_cltpsock = nfsrv_sockalloc();
   1002 #endif
   1003 
   1004 	simple_lock(&nfsd_slock);
   1005 	if (nfssvc_sockhead_flag & SLP_WANTINIT) {
   1006 		nfssvc_sockhead_flag &= ~SLP_WANTINIT;
   1007 		wakeup(&nfssvc_sockhead);
   1008 	}
   1009 	simple_unlock(&nfsd_slock);
   1010 	splx(s);
   1011 }
   1012 
   1013 /*
   1014  * Add entries to the server monitor log.
   1015  */
   1016 static void
   1017 nfsd_rt(sotype, nd, cacherep)
   1018 	int sotype;
   1019 	struct nfsrv_descript *nd;
   1020 	int cacherep;
   1021 {
   1022 	struct timeval tv;
   1023 	struct drt *rt;
   1024 
   1025 	rt = &nfsdrt.drt[nfsdrt.pos];
   1026 	if (cacherep == RC_DOIT)
   1027 		rt->flag = 0;
   1028 	else if (cacherep == RC_REPLY)
   1029 		rt->flag = DRT_CACHEREPLY;
   1030 	else
   1031 		rt->flag = DRT_CACHEDROP;
   1032 	if (sotype == SOCK_STREAM)
   1033 		rt->flag |= DRT_TCP;
   1034 	if (nd->nd_flag & ND_NQNFS)
   1035 		rt->flag |= DRT_NQNFS;
   1036 	else if (nd->nd_flag & ND_NFSV3)
   1037 		rt->flag |= DRT_NFSV3;
   1038 	rt->proc = nd->nd_procnum;
   1039 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
   1040 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
   1041 	else
   1042 	    rt->ipadr = INADDR_ANY;
   1043 	getmicrotime(&tv);
   1044 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
   1045 		(tv.tv_usec - nd->nd_starttime.tv_usec);
   1046 	rt->tstamp = tv;
   1047 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
   1048 }
   1049 #endif /* NFSSERVER */
   1050 
   1051 #ifdef NFS
   1052 
   1053 int nfs_defect = 0;
   1054 /*
   1055  * Asynchronous I/O threads for client nfs.
   1056  * They do read-ahead and write-behind operations on the block I/O cache.
   1057  * Never returns unless it fails or gets killed.
   1058  */
   1059 
   1060 int
   1061 nfssvc_iod(l)
   1062 	struct lwp *l;
   1063 {
   1064 	struct buf *bp;
   1065 	int i;
   1066 	struct nfs_iod *myiod;
   1067 	struct nfsmount *nmp;
   1068 	int error = 0;
   1069 	struct proc *p = l->l_proc;
   1070 
   1071 	/*
   1072 	 * Assign my position or return error if too many already running
   1073 	 */
   1074 	myiod = NULL;
   1075 	for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
   1076 		if (nfs_asyncdaemon[i].nid_proc == NULL) {
   1077 			myiod = &nfs_asyncdaemon[i];
   1078 			break;
   1079 		}
   1080 	if (myiod == NULL)
   1081 		return (EBUSY);
   1082 	myiod->nid_proc = p;
   1083 	nfs_numasync++;
   1084 	PHOLD(l);
   1085 	/*
   1086 	 * Just loop around doing our stuff until SIGKILL
   1087 	 */
   1088 	for (;;) {
   1089 		while (/*CONSTCOND*/ TRUE) {
   1090 			simple_lock(&myiod->nid_slock);
   1091 			nmp = myiod->nid_mount;
   1092 			if (nmp) {
   1093 				simple_lock(&nmp->nm_slock);
   1094 				if (!TAILQ_EMPTY(&nmp->nm_bufq)) {
   1095 					simple_unlock(&myiod->nid_slock);
   1096 					break;
   1097 				}
   1098 				nmp->nm_bufqiods--;
   1099 				simple_unlock(&nmp->nm_slock);
   1100 			}
   1101 			myiod->nid_want = p;
   1102 			myiod->nid_mount = NULL;
   1103 			error = ltsleep(&myiod->nid_want,
   1104 			    PWAIT | PCATCH | PNORELOCK, "nfsidl", 0,
   1105 			    &myiod->nid_slock);
   1106 			if (error)
   1107 				goto quit;
   1108 		}
   1109 
   1110 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
   1111 			/* Take one off the front of the list */
   1112 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
   1113 			nmp->nm_bufqlen--;
   1114 			if (nmp->nm_bufqwant &&
   1115 			    nmp->nm_bufqlen < 2 * nfs_numasync) {
   1116 				nmp->nm_bufqwant = FALSE;
   1117 				wakeup(&nmp->nm_bufq);
   1118 			}
   1119 			simple_unlock(&nmp->nm_slock);
   1120 			(void)nfs_doio(bp);
   1121 			simple_lock(&nmp->nm_slock);
   1122 			/*
   1123 			 * If there are more than one iod on this mount,
   1124 			 * then defect so that the iods can be shared out
   1125 			 * fairly between the mounts
   1126 			 */
   1127 			if (nfs_defect && nmp->nm_bufqiods > 1) {
   1128 				myiod->nid_mount = NULL;
   1129 				nmp->nm_bufqiods--;
   1130 				break;
   1131 			}
   1132 		}
   1133 		simple_unlock(&nmp->nm_slock);
   1134 	}
   1135 quit:
   1136 	PRELE(l);
   1137 	if (nmp)
   1138 		nmp->nm_bufqiods--;
   1139 	myiod->nid_want = NULL;
   1140 	myiod->nid_mount = NULL;
   1141 	myiod->nid_proc = NULL;
   1142 	nfs_numasync--;
   1143 
   1144 	return error;
   1145 }
   1146 
   1147 void
   1148 nfs_iodinit()
   1149 {
   1150 	int i;
   1151 
   1152 	for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
   1153 		simple_lock_init(&nfs_asyncdaemon[i].nid_slock);
   1154 }
   1155 
   1156 void
   1157 start_nfsio(void *arg)
   1158 {
   1159 	nfssvc_iod(curlwp);
   1160 
   1161 	kthread_exit(0);
   1162 }
   1163 
   1164 void
   1165 nfs_getset_niothreads(set)
   1166 	int set;
   1167 {
   1168 	int i, have, start;
   1169 
   1170 	for (have = 0, i = 0; i < NFS_MAXASYNCDAEMON; i++)
   1171 		if (nfs_asyncdaemon[i].nid_proc != NULL)
   1172 			have++;
   1173 
   1174 	if (set) {
   1175 		/* clamp to sane range */
   1176 		nfs_niothreads = max(0, min(nfs_niothreads, NFS_MAXASYNCDAEMON));
   1177 
   1178 		start = nfs_niothreads - have;
   1179 
   1180 		while (start > 0) {
   1181 			kthread_create1(start_nfsio, NULL, NULL, "nfsio");
   1182 			start--;
   1183 		}
   1184 
   1185 		for (i = 0; (start < 0) && (i < NFS_MAXASYNCDAEMON); i++)
   1186 			if (nfs_asyncdaemon[i].nid_proc != NULL) {
   1187 				psignal(nfs_asyncdaemon[i].nid_proc, SIGKILL);
   1188 				start++;
   1189 			}
   1190 	} else {
   1191 		if (nfs_niothreads >= 0)
   1192 			nfs_niothreads = have;
   1193 	}
   1194 }
   1195 
   1196 /*
   1197  * Get an authorization string for the uid by having the mount_nfs sitting
   1198  * on this mount point porpous out of the kernel and do it.
   1199  */
   1200 int
   1201 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
   1202 	struct nfsmount *nmp;
   1203 	struct nfsreq *rep;
   1204 	kauth_cred_t cred;
   1205 	char **auth_str;
   1206 	int *auth_len;
   1207 	char *verf_str;
   1208 	int *verf_len;
   1209 	NFSKERBKEY_T key;		/* return session key */
   1210 {
   1211 	int error = 0;
   1212 
   1213 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
   1214 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
   1215 		(void) tsleep((caddr_t)&nmp->nm_authtype, PSOCK,
   1216 			"nfsauth1", 2 * hz);
   1217 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1218 		if (error) {
   1219 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1220 			return (error);
   1221 		}
   1222 	}
   1223 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
   1224 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
   1225 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
   1226 	nmp->nm_verfstr = verf_str;
   1227 	nmp->nm_verflen = *verf_len;
   1228 	nmp->nm_authuid = kauth_cred_geteuid(cred);
   1229 	wakeup((caddr_t)&nmp->nm_authstr);
   1230 
   1231 	/*
   1232 	 * And wait for mount_nfs to do its stuff.
   1233 	 */
   1234 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
   1235 		(void) tsleep((caddr_t)&nmp->nm_authlen, PSOCK,
   1236 			"nfsauth2", 2 * hz);
   1237 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1238 	}
   1239 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
   1240 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
   1241 		error = EAUTH;
   1242 	}
   1243 	if (error)
   1244 		free((caddr_t)*auth_str, M_TEMP);
   1245 	else {
   1246 		*auth_len = nmp->nm_authlen;
   1247 		*verf_len = nmp->nm_verflen;
   1248 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
   1249 	}
   1250 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
   1251 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
   1252 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
   1253 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1254 		wakeup((caddr_t)&nmp->nm_authtype);
   1255 	}
   1256 	return (error);
   1257 }
   1258 
   1259 /*
   1260  * Get a nickname authenticator and verifier.
   1261  */
   1262 int
   1263 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
   1264     int *auth_len, char *verf_str, int verf_len)
   1265 {
   1266 	struct timeval ktvin, ktvout, tv;
   1267 	struct nfsuid *nuidp;
   1268 	u_int32_t *nickp, *verfp;
   1269 
   1270 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
   1271 
   1272 #ifdef DIAGNOSTIC
   1273 	if (verf_len < (4 * NFSX_UNSIGNED))
   1274 		panic("nfs_getnickauth verf too small");
   1275 #endif
   1276 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
   1277 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
   1278 			break;
   1279 	}
   1280 	if (!nuidp || nuidp->nu_expire < time_second)
   1281 		return (EACCES);
   1282 
   1283 	/*
   1284 	 * Move to the end of the lru list (end of lru == most recently used).
   1285 	 */
   1286 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1287 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1288 
   1289 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
   1290 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1291 	*nickp = txdr_unsigned(nuidp->nu_nickname);
   1292 	*auth_str = (char *)nickp;
   1293 	*auth_len = 2 * NFSX_UNSIGNED;
   1294 
   1295 	/*
   1296 	 * Now we must encrypt the verifier and package it up.
   1297 	 */
   1298 	verfp = (u_int32_t *)verf_str;
   1299 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1300 	getmicrotime(&tv);
   1301 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
   1302 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
   1303 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
   1304 		nuidp->nu_timestamp = tv;
   1305 	else
   1306 		nuidp->nu_timestamp.tv_usec++;
   1307 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
   1308 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
   1309 
   1310 	/*
   1311 	 * Now encrypt the timestamp verifier in ecb mode using the session
   1312 	 * key.
   1313 	 */
   1314 #ifdef NFSKERB
   1315 	XXX
   1316 #endif
   1317 
   1318 	*verfp++ = ktvout.tv_sec;
   1319 	*verfp++ = ktvout.tv_usec;
   1320 	*verfp = 0;
   1321 	return (0);
   1322 }
   1323 
   1324 /*
   1325  * Save the current nickname in a hash list entry on the mount point.
   1326  */
   1327 int
   1328 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
   1329 	struct nfsmount *nmp;
   1330 	kauth_cred_t cred;
   1331 	int len;
   1332 	NFSKERBKEY_T key;
   1333 	struct mbuf **mdp;
   1334 	char **dposp;
   1335 	struct mbuf *mrep;
   1336 {
   1337 	struct nfsuid *nuidp;
   1338 	u_int32_t *tl;
   1339 	int32_t t1;
   1340 	struct mbuf *md = *mdp;
   1341 	struct timeval ktvin, ktvout;
   1342 	u_int32_t nick;
   1343 	char *dpos = *dposp, *cp2;
   1344 	int deltasec, error = 0;
   1345 
   1346 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
   1347 
   1348 	if (len == (3 * NFSX_UNSIGNED)) {
   1349 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
   1350 		ktvin.tv_sec = *tl++;
   1351 		ktvin.tv_usec = *tl++;
   1352 		nick = fxdr_unsigned(u_int32_t, *tl);
   1353 
   1354 		/*
   1355 		 * Decrypt the timestamp in ecb mode.
   1356 		 */
   1357 #ifdef NFSKERB
   1358 		XXX
   1359 #endif
   1360 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
   1361 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
   1362 		deltasec = time_second - ktvout.tv_sec;
   1363 		if (deltasec < 0)
   1364 			deltasec = -deltasec;
   1365 		/*
   1366 		 * If ok, add it to the hash list for the mount point.
   1367 		 */
   1368 		if (deltasec <= NFS_KERBCLOCKSKEW) {
   1369 			if (nmp->nm_numuids < nuidhash_max) {
   1370 				nmp->nm_numuids++;
   1371 				nuidp = (struct nfsuid *)
   1372 				   malloc(sizeof (struct nfsuid), M_NFSUID,
   1373 					M_WAITOK);
   1374 			} else {
   1375 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
   1376 				LIST_REMOVE(nuidp, nu_hash);
   1377 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
   1378 					nu_lru);
   1379 			}
   1380 			nuidp->nu_flag = 0;
   1381 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
   1382 			nuidp->nu_expire = time_second + NFS_KERBTTL;
   1383 			nuidp->nu_timestamp = ktvout;
   1384 			nuidp->nu_nickname = nick;
   1385 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
   1386 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
   1387 				nu_lru);
   1388 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
   1389 				nuidp, nu_hash);
   1390 		}
   1391 	} else
   1392 		nfsm_adv(nfsm_rndup(len));
   1393 nfsmout:
   1394 	*mdp = md;
   1395 	*dposp = dpos;
   1396 	return (error);
   1397 }
   1398 #endif /* NFS */
   1399