Home | History | Annotate | Line # | Download | only in nfs
nfs_syscalls.c revision 1.126
      1 /*	$NetBSD: nfs_syscalls.c,v 1.126 2007/11/22 13:30:39 yamt 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 1.126 2007/11/22 13:30:39 yamt 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 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
    120 
    121 #ifdef NFS
    122 /*
    123  * locking order:
    124  *	nfs_iodlist_lock -> nid_lock -> nm_lock
    125  */
    126 kmutex_t nfs_iodlist_lock;
    127 struct nfs_iodlist nfs_iodlist_idle;
    128 struct nfs_iodlist nfs_iodlist_all;
    129 int nfs_niothreads = -1; /* == "0, and has never been set" */
    130 #endif
    131 
    132 #ifdef NFSSERVER
    133 static struct nfssvc_sock *nfsrv_sockalloc __P((void));
    134 static void nfsrv_sockfree __P((struct nfssvc_sock *));
    135 static void nfsd_rt __P((int, struct nfsrv_descript *, int));
    136 #endif
    137 
    138 /*
    139  * NFS server system calls
    140  */
    141 
    142 
    143 /*
    144  * Nfs server pseudo system call for the nfsd's
    145  * Based on the flag value it either:
    146  * - adds a socket to the selection list
    147  * - remains in the kernel as an nfsd
    148  * - remains in the kernel as an nfsiod
    149  */
    150 int
    151 sys_nfssvc(struct lwp *l, void *v, register_t *retval)
    152 {
    153 	struct sys_nfssvc_args /* {
    154 		syscallarg(int) flag;
    155 		syscallarg(void *) argp;
    156 	} */ *uap = v;
    157 	int error;
    158 #ifdef NFSSERVER
    159 	struct file *fp;
    160 	struct mbuf *nam;
    161 	struct nfsd_args nfsdarg;
    162 	struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
    163 	struct nfsd *nfsd;
    164 	struct nfssvc_sock *slp;
    165 	struct nfsuid *nuidp;
    166 #endif
    167 
    168 	/*
    169 	 * Must be super user
    170 	 */
    171 	error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
    172 	    NULL);
    173 	if (error)
    174 		return (error);
    175 
    176 	/* Initialize NFS server / client shared data. */
    177 	nfs_init();
    178 
    179 #ifdef NFSSERVER
    180 	mutex_enter(&nfsd_lock);
    181 	while (nfssvc_sockhead_flag & SLP_INIT) {
    182 		cv_wait(&nfsd_initcv, &nfsd_lock);
    183 	}
    184 	mutex_exit(&nfsd_lock);
    185 #endif
    186 	if (SCARG(uap, flag) & NFSSVC_BIOD) {
    187 #if defined(NFS) && defined(COMPAT_14)
    188 		error = kpause("nfsbiod", true, 0, NULL); /* dummy impl */
    189 #else
    190 		error = ENOSYS;
    191 #endif
    192 	} else if (SCARG(uap, flag) & NFSSVC_MNTD) {
    193 		error = ENOSYS;
    194 	} else if (SCARG(uap, flag) & NFSSVC_ADDSOCK) {
    195 #ifndef NFSSERVER
    196 		error = ENOSYS;
    197 #else
    198 		error = copyin(SCARG(uap, argp), (void *)&nfsdarg,
    199 		    sizeof(nfsdarg));
    200 		if (error)
    201 			return (error);
    202 		/* getsock() will use the descriptor for us */
    203 		error = getsock(l->l_proc->p_fd, nfsdarg.sock, &fp);
    204 		if (error)
    205 			return (error);
    206 		/*
    207 		 * Get the client address for connected sockets.
    208 		 */
    209 		if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
    210 			nam = (struct mbuf *)0;
    211 		else {
    212 			error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen,
    213 				MT_SONAME);
    214 			if (error) {
    215 				FILE_UNUSE(fp, NULL);
    216 				return (error);
    217 			}
    218 		}
    219 		error = nfssvc_addsock(fp, nam);
    220 		FILE_UNUSE(fp, NULL);
    221 #endif /* !NFSSERVER */
    222 	} else if (SCARG(uap, flag) & NFSSVC_SETEXPORTSLIST) {
    223 #ifndef NFSSERVER
    224 		error = ENOSYS;
    225 #else
    226 		struct export_args *args;
    227 		struct mountd_exports_list mel;
    228 
    229 		error = copyin(SCARG(uap, argp), &mel, sizeof(mel));
    230 		if (error != 0)
    231 			return error;
    232 
    233 		args = (struct export_args *)malloc(mel.mel_nexports *
    234 		    sizeof(struct export_args), M_TEMP, M_WAITOK);
    235 		error = copyin(mel.mel_exports, args, mel.mel_nexports *
    236 		    sizeof(struct export_args));
    237 		if (error != 0) {
    238 			free(args, M_TEMP);
    239 			return error;
    240 		}
    241 		mel.mel_exports = args;
    242 
    243 		error = mountd_set_exports_list(&mel, l);
    244 
    245 		free(args, M_TEMP);
    246 #endif /* !NFSSERVER */
    247 	} else {
    248 #ifndef NFSSERVER
    249 		error = ENOSYS;
    250 #else
    251 		error = copyin(SCARG(uap, argp), (void *)nsd, sizeof (*nsd));
    252 		if (error)
    253 			return (error);
    254 		if ((SCARG(uap, flag) & NFSSVC_AUTHIN) &&
    255 		    ((nfsd = nsd->nsd_nfsd)) != NULL &&
    256 		    (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
    257 			slp = nfsd->nfsd_slp;
    258 
    259 			/*
    260 			 * First check to see if another nfsd has already
    261 			 * added this credential.
    262 			 */
    263 			LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid),
    264 			    nu_hash) {
    265 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
    266 				    nsd->nsd_cr.cr_uid &&
    267 				    (!nfsd->nfsd_nd->nd_nam2 ||
    268 				     netaddr_match(NU_NETFAM(nuidp),
    269 				     &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
    270 					break;
    271 			}
    272 			if (nuidp) {
    273 			    kauth_cred_hold(nuidp->nu_cr);
    274 			    nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    275 			    nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    276 			} else {
    277 			    /*
    278 			     * Nope, so we will.
    279 			     */
    280 			    if (slp->ns_numuids < nuidhash_max) {
    281 				slp->ns_numuids++;
    282 				nuidp = (struct nfsuid *)
    283 				   malloc(sizeof (struct nfsuid), M_NFSUID,
    284 					M_WAITOK);
    285 			    } else
    286 				nuidp = (struct nfsuid *)0;
    287 			    if ((slp->ns_flag & SLP_VALID) == 0) {
    288 				if (nuidp)
    289 				    free((void *)nuidp, M_NFSUID);
    290 			    } else {
    291 				if (nuidp == (struct nfsuid *)0) {
    292 				    nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
    293 				    LIST_REMOVE(nuidp, nu_hash);
    294 				    TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
    295 					nu_lru);
    296 				    if (nuidp->nu_flag & NU_NAM)
    297 					m_freem(nuidp->nu_nam);
    298 			        }
    299 				nuidp->nu_flag = 0;
    300 				kauth_uucred_to_cred(nuidp->nu_cr,
    301 				    &nsd->nsd_cr);
    302 				nuidp->nu_timestamp = nsd->nsd_timestamp;
    303 				nuidp->nu_expire = time_second + nsd->nsd_ttl;
    304 				/*
    305 				 * and save the session key in nu_key.
    306 				 */
    307 				memcpy(nuidp->nu_key, nsd->nsd_key,
    308 				    sizeof(nsd->nsd_key));
    309 				if (nfsd->nfsd_nd->nd_nam2) {
    310 				    struct sockaddr_in *saddr;
    311 
    312 				    saddr = mtod(nfsd->nfsd_nd->nd_nam2,
    313 					 struct sockaddr_in *);
    314 				    switch (saddr->sin_family) {
    315 				    case AF_INET:
    316 					nuidp->nu_flag |= NU_INETADDR;
    317 					nuidp->nu_inetaddr =
    318 					     saddr->sin_addr.s_addr;
    319 					break;
    320 				    case AF_ISO:
    321 				    default:
    322 					nuidp->nu_flag |= NU_NAM;
    323 					nuidp->nu_nam = m_copym(
    324 					    nfsd->nfsd_nd->nd_nam2, 0,
    325 					     M_COPYALL, M_WAIT);
    326 					break;
    327 				    };
    328 				}
    329 				TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
    330 					nu_lru);
    331 				LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
    332 					nuidp, nu_hash);
    333 				kauth_cred_hold(nuidp->nu_cr);
    334 				nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    335 				nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    336 			    }
    337 			}
    338 		}
    339 		if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) &&
    340 		    (nfsd = nsd->nsd_nfsd))
    341 			nfsd->nfsd_flag |= NFSD_AUTHFAIL;
    342 		error = nfssvc_nfsd(nsd, SCARG(uap, argp), l);
    343 #endif /* !NFSSERVER */
    344 	}
    345 	if (error == EINTR || error == ERESTART)
    346 		error = 0;
    347 	return (error);
    348 }
    349 
    350 #ifdef NFSSERVER
    351 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
    352 MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
    353 
    354 static struct nfssvc_sock *
    355 nfsrv_sockalloc()
    356 {
    357 	struct nfssvc_sock *slp;
    358 
    359 	slp = (struct nfssvc_sock *)
    360 	    malloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK);
    361 	memset(slp, 0, sizeof (struct nfssvc_sock));
    362 	mutex_init(&slp->ns_lock, 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_flag & SLP_VALID) == 0);
    381 	mutex_destroy(&slp->ns_lock);
    382 	cv_destroy(&slp->ns_cv);
    383 	free(slp, M_NFSSVC);
    384 }
    385 
    386 /*
    387  * Adds a socket to the list for servicing by nfsds.
    388  */
    389 int
    390 nfssvc_addsock(fp, mynam)
    391 	struct file *fp;
    392 	struct mbuf *mynam;
    393 {
    394 	struct mbuf *m;
    395 	int siz;
    396 	struct nfssvc_sock *slp;
    397 	struct socket *so;
    398 	struct nfssvc_sock *tslp;
    399 	int error, s;
    400 
    401 	so = (struct socket *)fp->f_data;
    402 	tslp = (struct nfssvc_sock *)0;
    403 	/*
    404 	 * Add it to the list, as required.
    405 	 */
    406 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
    407 #ifdef INET6
    408 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
    409 			tslp = nfs_udp6sock;
    410 		else
    411 #endif
    412 		tslp = nfs_udpsock;
    413 		if (tslp->ns_flag & SLP_VALID) {
    414 			m_freem(mynam);
    415 			return (EPERM);
    416 		}
    417 #ifdef ISO
    418 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
    419 		tslp = nfs_cltpsock;
    420 		if (tslp->ns_flag & SLP_VALID) {
    421 			m_freem(mynam);
    422 			return (EPERM);
    423 		}
    424 #endif /* ISO */
    425 	}
    426 	if (so->so_type == SOCK_STREAM)
    427 		siz = NFS_MAXPACKET + sizeof (u_long);
    428 	else
    429 		siz = NFS_MAXPACKET;
    430 	error = soreserve(so, siz, siz);
    431 	if (error) {
    432 		m_freem(mynam);
    433 		return (error);
    434 	}
    435 
    436 	/*
    437 	 * Set protocol specific options { for now TCP only } and
    438 	 * reserve some space. For datagram sockets, this can get called
    439 	 * repeatedly for the same socket, but that isn't harmful.
    440 	 */
    441 	if (so->so_type == SOCK_STREAM) {
    442 		m = m_get(M_WAIT, MT_SOOPTS);
    443 		MCLAIM(m, &nfs_mowner);
    444 		*mtod(m, int32_t *) = 1;
    445 		m->m_len = sizeof(int32_t);
    446 		sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
    447 	}
    448 	if ((so->so_proto->pr_domain->dom_family == AF_INET
    449 #ifdef INET6
    450 	    || so->so_proto->pr_domain->dom_family == AF_INET6
    451 #endif
    452 	    ) &&
    453 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
    454 		m = m_get(M_WAIT, MT_SOOPTS);
    455 		MCLAIM(m, &nfs_mowner);
    456 		*mtod(m, int32_t *) = 1;
    457 		m->m_len = sizeof(int32_t);
    458 		sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
    459 	}
    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 	if (tslp) {
    465 		slp = tslp;
    466 	} else {
    467 		slp = nfsrv_sockalloc();
    468 	}
    469 	slp->ns_so = so;
    470 	slp->ns_nam = mynam;
    471 	fp->f_count++;
    472 	slp->ns_fp = fp;
    473 	s = splsoftnet();
    474 	so->so_upcallarg = (void *)slp;
    475 	so->so_upcall = nfsrv_rcv;
    476 	so->so_rcv.sb_flags |= SB_UPCALL;
    477 	slp->ns_flag = SLP_VALID | SLP_NEEDQ;
    478 	nfsrv_wakenfsd(slp);
    479 	splx(s);
    480 	return (0);
    481 }
    482 
    483 /*
    484  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    485  * until it is killed by a signal.
    486  */
    487 int
    488 nfssvc_nfsd(nsd, argp, l)
    489 	struct nfsd_srvargs *nsd;
    490 	void *argp;
    491 	struct lwp *l;
    492 {
    493 	struct timeval tv;
    494 	struct mbuf *m;
    495 	struct nfssvc_sock *slp;
    496 	struct nfsd *nfsd = nsd->nsd_nfsd;
    497 	struct nfsrv_descript *nd = NULL;
    498 	struct mbuf *mreq;
    499 	u_quad_t cur_usec;
    500 	int error = 0, cacherep, siz, sotype, writes_todo;
    501 	struct proc *p = l->l_proc;
    502 	int s;
    503 
    504 #ifndef nolint
    505 	cacherep = RC_DOIT;
    506 	writes_todo = 0;
    507 #endif
    508 	uvm_lwp_hold(l);
    509 	if (nfsd == NULL) {
    510 		nsd->nsd_nfsd = nfsd =
    511 			malloc(sizeof (struct nfsd), M_NFSD, M_WAITOK);
    512 		memset(nfsd, 0, sizeof (struct nfsd));
    513 		cv_init(&nfsd->nfsd_cv, "nfsd");
    514 		nfsd->nfsd_procp = p;
    515 		mutex_enter(&nfsd_lock);
    516 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
    517 		nfs_numnfsd++;
    518 		mutex_exit(&nfsd_lock);
    519 	}
    520 	/*
    521 	 * Loop getting rpc requests until SIGKILL.
    522 	 */
    523 	for (;;) {
    524 		if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    525 		    != 0) {
    526 			preempt();
    527 		}
    528 		if (nfsd->nfsd_slp == NULL) {
    529 			mutex_enter(&nfsd_lock);
    530 			while (nfsd->nfsd_slp == NULL &&
    531 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
    532 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
    533 				    nfsd_idle);
    534 				error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock);
    535 				if (error) {
    536 					slp = nfsd->nfsd_slp;
    537 					nfsd->nfsd_slp = NULL;
    538 					if (!slp)
    539 						SLIST_REMOVE(&nfsd_idle_head,
    540 						    nfsd, nfsd, nfsd_idle);
    541 					mutex_exit(&nfsd_lock);
    542 					if (slp) {
    543 						nfsrv_wakenfsd(slp);
    544 						nfsrv_slpderef(slp);
    545 					}
    546 					goto done;
    547 				}
    548 			}
    549 			if (nfsd->nfsd_slp == NULL &&
    550 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
    551 				slp = TAILQ_FIRST(&nfssvc_sockpending);
    552 				if (slp) {
    553 					KASSERT((slp->ns_flag & SLP_DOREC)
    554 					    != 0);
    555 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
    556 					    ns_pending);
    557 					slp->ns_flag &= ~SLP_DOREC;
    558 					slp->ns_sref++;
    559 					nfsd->nfsd_slp = slp;
    560 				} else
    561 					nfsd_head_flag &= ~NFSD_CHECKSLP;
    562 			}
    563 			KASSERT(nfsd->nfsd_slp == NULL ||
    564 			    nfsd->nfsd_slp->ns_sref > 0);
    565 			mutex_exit(&nfsd_lock);
    566 			if ((slp = nfsd->nfsd_slp) == NULL)
    567 				continue;
    568 			KASSERT(slp->ns_sref > 0);
    569 			if (slp->ns_flag & SLP_VALID) {
    570 				if ((slp->ns_flag & SLP_NEEDQ) != 0) {
    571 					nfsrv_rcv(slp->ns_so, (void *)slp,
    572 					    M_WAIT);
    573 				}
    574 				if ((slp->ns_flag & SLP_DISCONN) != 0) {
    575 					nfsrv_zapsock(slp);
    576 				}
    577 				error = nfsrv_dorec(slp, nfsd, &nd);
    578 				getmicrotime(&tv);
    579 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    580 					(u_quad_t)tv.tv_usec;
    581 				if (error && LIST_FIRST(&slp->ns_tq) &&
    582 				    LIST_FIRST(&slp->ns_tq)->nd_time <=
    583 				    cur_usec) {
    584 					error = 0;
    585 					cacherep = RC_DOIT;
    586 					writes_todo = 1;
    587 				} else
    588 					writes_todo = 0;
    589 				if (error == 0 && slp->ns_rec != NULL) {
    590 					nfsrv_wakenfsd(slp);
    591 				}
    592 			}
    593 		} else {
    594 			error = 0;
    595 			slp = nfsd->nfsd_slp;
    596 		}
    597 		KASSERT(slp != NULL);
    598 		KASSERT(nfsd->nfsd_slp == slp);
    599 		if (error || (slp->ns_flag & SLP_VALID) == 0) {
    600 			if (nd) {
    601 				nfsdreq_free(nd);
    602 				nd = NULL;
    603 			}
    604 			nfsd->nfsd_slp = NULL;
    605 			nfsrv_slpderef(slp);
    606 			continue;
    607 		}
    608 		sotype = slp->ns_so->so_type;
    609 		if (nd) {
    610 			getmicrotime(&nd->nd_starttime);
    611 			if (nd->nd_nam2)
    612 				nd->nd_nam = nd->nd_nam2;
    613 			else
    614 				nd->nd_nam = slp->ns_nam;
    615 
    616 			/*
    617 			 * Check to see if authorization is needed.
    618 			 */
    619 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
    620 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
    621 				nsd->nsd_haddr = mtod(nd->nd_nam,
    622 				    struct sockaddr_in *)->sin_addr.s_addr;
    623 				nsd->nsd_authlen = nfsd->nfsd_authlen;
    624 				nsd->nsd_verflen = nfsd->nfsd_verflen;
    625 				if (!copyout(nfsd->nfsd_authstr,
    626 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
    627 				    !copyout(nfsd->nfsd_verfstr,
    628 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
    629 				    !copyout(nsd, argp, sizeof (*nsd))) {
    630 					uvm_lwp_rele(l);
    631 					return (ENEEDAUTH);
    632 				}
    633 				cacherep = RC_DROPIT;
    634 			} else
    635 				cacherep = nfsrv_getcache(nd, slp, &mreq);
    636 
    637 			if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
    638 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
    639 				nd->nd_procnum = NFSPROC_NOOP;
    640 				nd->nd_repstat =
    641 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
    642 				cacherep = RC_DOIT;
    643 			}
    644 		}
    645 
    646 		/*
    647 		 * Loop to get all the write rpc relies that have been
    648 		 * gathered together.
    649 		 */
    650 		do {
    651 #ifdef DIAGNOSTIC
    652 			int lockcount;
    653 #endif
    654 			switch (cacherep) {
    655 			case RC_DOIT:
    656 #ifdef DIAGNOSTIC
    657 				/*
    658 				 * NFS server procs should neither release
    659 				 * locks already held, nor leave things
    660 				 * locked.  Catch this sooner, rather than
    661 				 * later (when we try to relock something we
    662 				 * already have locked).  Careful inspection
    663 				 * of the failing routine usually turns up the
    664 				 * lock leak.. once we know what it is..
    665 				 */
    666 				lockcount = l->l_locks;
    667 #endif
    668 				mreq = NULL;
    669 				netexport_rdlock();
    670 				if (writes_todo || nd == NULL ||
    671 				     (!(nd->nd_flag & ND_NFSV3) &&
    672 				     nd->nd_procnum == NFSPROC_WRITE &&
    673 				     nfsrvw_procrastinate > 0))
    674 					error = nfsrv_writegather(&nd, slp,
    675 					    l, &mreq);
    676 				else
    677 					error =
    678 					    (*(nfsrv3_procs[nd->nd_procnum]))
    679 					    (nd, slp, l, &mreq);
    680 				netexport_rdunlock();
    681 #ifdef DIAGNOSTIC
    682 				if (l->l_locks != lockcount) {
    683 					/*
    684 					 * If you see this panic, audit
    685 					 * nfsrv3_procs[nd->nd_procnum] for
    686 					 * vnode locking errors (usually, it's
    687 					 * due to forgetting to vput()
    688 					 * something).
    689 					 */
    690 #ifdef DEBUG
    691 					extern void printlockedvnodes(void);
    692 					printlockedvnodes();
    693 #endif
    694 					printf("nfsd: locking botch in op %d"
    695 					    " (before %d, after %d)\n",
    696 					    nd ? nd->nd_procnum : -1,
    697 					    lockcount, l->l_locks);
    698 				}
    699 #endif
    700 				if (mreq == NULL) {
    701 					if (nd != NULL) {
    702 						if (nd->nd_nam2)
    703 							m_free(nd->nd_nam2);
    704 						if (nd->nd_mrep)
    705 							m_freem(nd->nd_mrep);
    706 					}
    707 					break;
    708 				}
    709 				if (error) {
    710 					nfsstats.srv_errs++;
    711 					nfsrv_updatecache(nd, false, mreq);
    712 					if (nd->nd_nam2)
    713 						m_freem(nd->nd_nam2);
    714 					break;
    715 				}
    716 				nfsstats.srvrpccnt[nd->nd_procnum]++;
    717 				nfsrv_updatecache(nd, true, mreq);
    718 				nd->nd_mrep = (struct mbuf *)0;
    719 			case RC_REPLY:
    720 				m = mreq;
    721 				siz = 0;
    722 				while (m) {
    723 					siz += m->m_len;
    724 					m = m->m_next;
    725 				}
    726 				if (siz <= 0 || siz > NFS_MAXPACKET) {
    727 					printf("mbuf siz=%d\n",siz);
    728 					panic("Bad nfs svc reply");
    729 				}
    730 				m = mreq;
    731 				m->m_pkthdr.len = siz;
    732 				m->m_pkthdr.rcvif = (struct ifnet *)0;
    733 				/*
    734 				 * For stream protocols, prepend a Sun RPC
    735 				 * Record Mark.
    736 				 */
    737 				if (sotype == SOCK_STREAM) {
    738 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
    739 					*mtod(m, u_int32_t *) =
    740 					    htonl(0x80000000 | siz);
    741 				}
    742 				nd->nd_mreq = m;
    743 				if (nfsrtton) {
    744 					nfsd_rt(slp->ns_so->so_type, nd,
    745 					    cacherep);
    746 				}
    747 				error = nfsdsock_sendreply(slp, nd);
    748 				nd = NULL;
    749 				if (error == EPIPE)
    750 					nfsrv_zapsock(slp);
    751 				if (error == EINTR || error == ERESTART) {
    752 					nfsd->nfsd_slp = NULL;
    753 					nfsrv_slpderef(slp);
    754 					goto done;
    755 				}
    756 				break;
    757 			case RC_DROPIT:
    758 				if (nfsrtton)
    759 					nfsd_rt(sotype, nd, cacherep);
    760 				m_freem(nd->nd_mrep);
    761 				m_freem(nd->nd_nam2);
    762 				break;
    763 			}
    764 			if (nd) {
    765 				nfsdreq_free(nd);
    766 				nd = NULL;
    767 			}
    768 
    769 			/*
    770 			 * Check to see if there are outstanding writes that
    771 			 * need to be serviced.
    772 			 */
    773 			getmicrotime(&tv);
    774 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    775 			    (u_quad_t)tv.tv_usec;
    776 			s = splsoftclock();
    777 			if (LIST_FIRST(&slp->ns_tq) &&
    778 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
    779 				cacherep = RC_DOIT;
    780 				writes_todo = 1;
    781 			} else
    782 				writes_todo = 0;
    783 			splx(s);
    784 		} while (writes_todo);
    785 		if (nfsrv_dorec(slp, nfsd, &nd)) {
    786 			nfsd->nfsd_slp = NULL;
    787 			nfsrv_slpderef(slp);
    788 		}
    789 	}
    790 done:
    791 	mutex_enter(&nfsd_lock);
    792 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
    793 	mutex_exit(&nfsd_lock);
    794 	cv_destroy(&nfsd->nfsd_cv);
    795 	free(nfsd, M_NFSD);
    796 	nsd->nsd_nfsd = NULL;
    797 	if (--nfs_numnfsd == 0)
    798 		nfsrv_init(true);	/* Reinitialize everything */
    799 	uvm_lwp_rele(l);
    800 	return (error);
    801 }
    802 
    803 /*
    804  * Shut down a socket associated with an nfssvc_sock structure.
    805  * Should be called with the send lock set, if required.
    806  * The trick here is to increment the sref at the start, so that the nfsds
    807  * will stop using it and clear ns_flag at the end so that it will not be
    808  * reassigned during cleanup.
    809  *
    810  * called at splsoftnet.
    811  */
    812 void
    813 nfsrv_zapsock(slp)
    814 	struct nfssvc_sock *slp;
    815 {
    816 	struct nfsuid *nuidp, *nnuidp;
    817 	struct nfsrv_descript *nwp, *nnwp;
    818 	struct socket *so;
    819 	struct mbuf *m;
    820 	int s;
    821 
    822 	if (nfsdsock_drain(slp)) {
    823 		return;
    824 	}
    825 	mutex_enter(&nfsd_lock);
    826 	if (slp->ns_flag & SLP_DOREC) {
    827 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
    828 		slp->ns_flag &= ~SLP_DOREC;
    829 	}
    830 	mutex_exit(&nfsd_lock);
    831 
    832 	so = slp->ns_so;
    833 	KASSERT(so != NULL);
    834 	so->so_upcall = NULL;
    835 	so->so_upcallarg = NULL;
    836 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    837 	soshutdown(so, SHUT_RDWR);
    838 
    839 	if (slp->ns_nam)
    840 		m_free(slp->ns_nam);
    841 	m_freem(slp->ns_raw);
    842 	m = slp->ns_rec;
    843 	while (m != NULL) {
    844 		struct mbuf *n;
    845 
    846 		n = m->m_nextpkt;
    847 		m_freem(m);
    848 		m = n;
    849 	}
    850 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
    851 	    nuidp = nnuidp) {
    852 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
    853 		LIST_REMOVE(nuidp, nu_hash);
    854 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
    855 		if (nuidp->nu_flag & NU_NAM)
    856 			m_freem(nuidp->nu_nam);
    857 		free((void *)nuidp, M_NFSUID);
    858 	}
    859 	s = splsoftclock();
    860 	for (nwp = LIST_FIRST(&slp->ns_tq); nwp; nwp = nnwp) {
    861 		nnwp = LIST_NEXT(nwp, nd_tq);
    862 		LIST_REMOVE(nwp, nd_tq);
    863 		nfsdreq_free(nwp);
    864 	}
    865 	splx(s);
    866 }
    867 
    868 /*
    869  * Derefence a server socket structure. If it has no more references and
    870  * is no longer valid, you can throw it away.
    871  */
    872 void
    873 nfsrv_slpderef(slp)
    874 	struct nfssvc_sock *slp;
    875 {
    876 	uint32_t ref;
    877 
    878 	mutex_enter(&nfsd_lock);
    879 	KASSERT(slp->ns_sref > 0);
    880 	ref = --slp->ns_sref;
    881 	mutex_exit(&nfsd_lock);
    882 	if (ref == 0 && (slp->ns_flag & SLP_VALID) == 0) {
    883 		struct file *fp;
    884 
    885 		mutex_enter(&nfsd_lock);
    886 		KASSERT((slp->ns_flag & SLP_DOREC) == 0);
    887 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
    888 		mutex_exit(&nfsd_lock);
    889 
    890 		fp = slp->ns_fp;
    891 		if (fp != NULL) {
    892 			slp->ns_fp = NULL;
    893 			KASSERT(fp != NULL);
    894 			KASSERT(fp->f_data == slp->ns_so);
    895 			mutex_enter(&fp->f_lock);
    896 			FILE_USE(fp);
    897 			closef(fp, (struct lwp *)0);
    898 			slp->ns_so = NULL;
    899 		}
    900 
    901 		nfsrv_sockfree(slp);
    902 	}
    903 }
    904 
    905 /*
    906  * Initialize the data structures for the server.
    907  * Handshake with any new nfsds starting up to avoid any chance of
    908  * corruption.
    909  */
    910 void
    911 nfsrv_init(terminating)
    912 	int terminating;
    913 {
    914 	struct nfssvc_sock *slp;
    915 
    916 	if (!terminating) {
    917 		mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET);
    918 		cv_init(&nfsd_initcv, "nfsdinit");
    919 	}
    920 
    921 	mutex_enter(&nfsd_lock);
    922 	if (nfssvc_sockhead_flag & SLP_INIT)
    923 		panic("nfsd init");
    924 	nfssvc_sockhead_flag |= SLP_INIT;
    925 
    926 	if (terminating) {
    927 		KASSERT(SLIST_EMPTY(&nfsd_idle_head));
    928 		KASSERT(TAILQ_EMPTY(&nfsd_head));
    929 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
    930 			mutex_exit(&nfsd_lock);
    931 			KASSERT(slp->ns_sref == 0);
    932 			slp->ns_sref++;
    933 			nfsrv_zapsock(slp);
    934 			nfsrv_slpderef(slp);
    935 			mutex_enter(&nfsd_lock);
    936 		}
    937 		KASSERT(TAILQ_EMPTY(&nfssvc_sockpending));
    938 		mutex_exit(&nfsd_lock);
    939 		nfsrv_cleancache();	/* And clear out server cache */
    940 	} else {
    941 		mutex_exit(&nfsd_lock);
    942 		nfs_pub.np_valid = 0;
    943 	}
    944 
    945 	TAILQ_INIT(&nfssvc_sockhead);
    946 	TAILQ_INIT(&nfssvc_sockpending);
    947 
    948 	TAILQ_INIT(&nfsd_head);
    949 	SLIST_INIT(&nfsd_idle_head);
    950 	nfsd_head_flag &= ~NFSD_CHECKSLP;
    951 
    952 	nfs_udpsock = nfsrv_sockalloc();
    953 
    954 #ifdef INET6
    955 	nfs_udp6sock = nfsrv_sockalloc();
    956 #endif
    957 
    958 #ifdef ISO
    959 	nfs_cltpsock = nfsrv_sockalloc();
    960 #endif
    961 
    962 	mutex_enter(&nfsd_lock);
    963 	nfssvc_sockhead_flag &= ~SLP_INIT;
    964 	cv_broadcast(&nfsd_initcv);
    965 	mutex_exit(&nfsd_lock);
    966 }
    967 
    968 /*
    969  * Add entries to the server monitor log.
    970  */
    971 static void
    972 nfsd_rt(sotype, nd, cacherep)
    973 	int sotype;
    974 	struct nfsrv_descript *nd;
    975 	int cacherep;
    976 {
    977 	struct timeval tv;
    978 	struct drt *rt;
    979 
    980 	rt = &nfsdrt.drt[nfsdrt.pos];
    981 	if (cacherep == RC_DOIT)
    982 		rt->flag = 0;
    983 	else if (cacherep == RC_REPLY)
    984 		rt->flag = DRT_CACHEREPLY;
    985 	else
    986 		rt->flag = DRT_CACHEDROP;
    987 	if (sotype == SOCK_STREAM)
    988 		rt->flag |= DRT_TCP;
    989 	if (nd->nd_flag & ND_NFSV3)
    990 		rt->flag |= DRT_NFSV3;
    991 	rt->proc = nd->nd_procnum;
    992 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
    993 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
    994 	else
    995 	    rt->ipadr = INADDR_ANY;
    996 	getmicrotime(&tv);
    997 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
    998 		(tv.tv_usec - nd->nd_starttime.tv_usec);
    999 	rt->tstamp = tv;
   1000 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
   1001 }
   1002 #endif /* NFSSERVER */
   1003 
   1004 #ifdef NFS
   1005 
   1006 int nfs_defect = 0;
   1007 /*
   1008  * Asynchronous I/O threads for client nfs.
   1009  * They do read-ahead and write-behind operations on the block I/O cache.
   1010  * Never returns unless it fails or gets killed.
   1011  */
   1012 
   1013 static void
   1014 nfssvc_iod(void *arg)
   1015 {
   1016 	struct buf *bp;
   1017 	struct nfs_iod *myiod;
   1018 	struct nfsmount *nmp;
   1019 
   1020 	KERNEL_LOCK(1, curlwp);
   1021 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
   1022 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
   1023 	KERNEL_UNLOCK_LAST(curlwp);
   1024 	cv_init(&myiod->nid_cv, "nfsiod");
   1025 	myiod->nid_exiting = false;
   1026 	myiod->nid_mount = NULL;
   1027 	mutex_enter(&nfs_iodlist_lock);
   1028 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
   1029 	mutex_exit(&nfs_iodlist_lock);
   1030 
   1031 	for (;;) {
   1032 		mutex_enter(&nfs_iodlist_lock);
   1033 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
   1034 		mutex_exit(&nfs_iodlist_lock);
   1035 
   1036 		mutex_enter(&myiod->nid_lock);
   1037 		while (/*CONSTCOND*/ true) {
   1038 			nmp = myiod->nid_mount;
   1039 			if (nmp) {
   1040 				myiod->nid_mount = NULL;
   1041 				break;
   1042 			}
   1043 			if (__predict_false(myiod->nid_exiting)) {
   1044 				/*
   1045 				 * drop nid_lock to preserve locking order.
   1046 				 */
   1047 				mutex_exit(&myiod->nid_lock);
   1048 				mutex_enter(&nfs_iodlist_lock);
   1049 				mutex_enter(&myiod->nid_lock);
   1050 				/*
   1051 				 * recheck nid_mount because nfs_asyncio can
   1052 				 * pick us in the meantime as we are still on
   1053 				 * nfs_iodlist_lock.
   1054 				 */
   1055 				if (myiod->nid_mount != NULL) {
   1056 					mutex_exit(&nfs_iodlist_lock);
   1057 					continue;
   1058 				}
   1059 				LIST_REMOVE(myiod, nid_idle);
   1060 				mutex_exit(&nfs_iodlist_lock);
   1061 				goto quit;
   1062 			}
   1063 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
   1064 		}
   1065 		mutex_exit(&myiod->nid_lock);
   1066 
   1067 		mutex_enter(&nmp->nm_lock);
   1068 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
   1069 			/* Take one off the front of the list */
   1070 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
   1071 			nmp->nm_bufqlen--;
   1072 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
   1073 				cv_broadcast(&nmp->nm_aiocv);
   1074 			}
   1075 			mutex_exit(&nmp->nm_lock);
   1076 			KERNEL_LOCK(1, curlwp);
   1077 			(void)nfs_doio(bp);
   1078 			KERNEL_UNLOCK_LAST(curlwp);
   1079 			mutex_enter(&nmp->nm_lock);
   1080 			/*
   1081 			 * If there are more than one iod on this mount,
   1082 			 * then defect so that the iods can be shared out
   1083 			 * fairly between the mounts
   1084 			 */
   1085 			if (nfs_defect && nmp->nm_bufqiods > 1) {
   1086 				break;
   1087 			}
   1088 		}
   1089 		KASSERT(nmp->nm_bufqiods > 0);
   1090 		nmp->nm_bufqiods--;
   1091 		mutex_exit(&nmp->nm_lock);
   1092 	}
   1093 quit:
   1094 	KASSERT(myiod->nid_mount == NULL);
   1095 	mutex_exit(&myiod->nid_lock);
   1096 
   1097 	cv_destroy(&myiod->nid_cv);
   1098 	KERNEL_LOCK(1, curlwp);
   1099 	mutex_destroy(&myiod->nid_lock);
   1100 	kmem_free(myiod, sizeof(*myiod));
   1101 	KERNEL_UNLOCK_LAST(curlwp);
   1102 
   1103 	kthread_exit(0);
   1104 }
   1105 
   1106 void
   1107 nfs_iodinit()
   1108 {
   1109 
   1110 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
   1111 	LIST_INIT(&nfs_iodlist_all);
   1112 	LIST_INIT(&nfs_iodlist_idle);
   1113 }
   1114 
   1115 int
   1116 nfs_set_niothreads(int newval)
   1117 {
   1118 	struct nfs_iod *nid;
   1119 	int error = 0;
   1120 
   1121 #if defined(MULTIPROCESSOR)
   1122         int hold_count;
   1123 #endif /* defined(MULTIPROCESSOR) */
   1124 
   1125 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
   1126 
   1127 	mutex_enter(&nfs_iodlist_lock);
   1128 	/* clamp to sane range */
   1129 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
   1130 
   1131 	while (nfs_numasync != nfs_niothreads && error == 0) {
   1132 		while (nfs_numasync < nfs_niothreads) {
   1133 
   1134 			/*
   1135 			 * kthread_create can wait for pagedaemon and
   1136 			 * pagedaemon can wait for nfsiod which needs to acquire
   1137 			 * nfs_iodlist_lock.
   1138 			 */
   1139 
   1140 			mutex_exit(&nfs_iodlist_lock);
   1141 			KERNEL_LOCK(1, curlwp);
   1142 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
   1143 			    nfssvc_iod, NULL, NULL, "nfsio");
   1144 			KERNEL_UNLOCK_LAST(curlwp);
   1145 			mutex_enter(&nfs_iodlist_lock);
   1146 			if (error) {
   1147 				/* give up */
   1148 				nfs_niothreads = nfs_numasync;
   1149 				break;
   1150 			}
   1151 			nfs_numasync++;
   1152 		}
   1153 		while (nfs_numasync > nfs_niothreads) {
   1154 			nid = LIST_FIRST(&nfs_iodlist_all);
   1155 			if (nid == NULL) {
   1156 				/* iod has not started yet. */
   1157 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
   1158 				continue;
   1159 			}
   1160 			LIST_REMOVE(nid, nid_all);
   1161 			mutex_enter(&nid->nid_lock);
   1162 			KASSERT(!nid->nid_exiting);
   1163 			nid->nid_exiting = true;
   1164 			cv_signal(&nid->nid_cv);
   1165 			mutex_exit(&nid->nid_lock);
   1166 			nfs_numasync--;
   1167 		}
   1168 	}
   1169 	mutex_exit(&nfs_iodlist_lock);
   1170 
   1171 	KERNEL_LOCK(hold_count, curlwp);
   1172 	return error;
   1173 }
   1174 
   1175 /*
   1176  * Get an authorization string for the uid by having the mount_nfs sitting
   1177  * on this mount point porpous out of the kernel and do it.
   1178  */
   1179 int
   1180 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
   1181 	struct nfsmount *nmp;
   1182 	struct nfsreq *rep;
   1183 	kauth_cred_t cred;
   1184 	char **auth_str;
   1185 	int *auth_len;
   1186 	char *verf_str;
   1187 	int *verf_len;
   1188 	NFSKERBKEY_T key;		/* return session key */
   1189 {
   1190 	int error = 0;
   1191 
   1192 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
   1193 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
   1194 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
   1195 			"nfsauth1", 2 * hz);
   1196 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1197 		if (error) {
   1198 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1199 			return (error);
   1200 		}
   1201 	}
   1202 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
   1203 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
   1204 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
   1205 	nmp->nm_verfstr = verf_str;
   1206 	nmp->nm_verflen = *verf_len;
   1207 	nmp->nm_authuid = kauth_cred_geteuid(cred);
   1208 	wakeup((void *)&nmp->nm_authstr);
   1209 
   1210 	/*
   1211 	 * And wait for mount_nfs to do its stuff.
   1212 	 */
   1213 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
   1214 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
   1215 			"nfsauth2", 2 * hz);
   1216 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1217 	}
   1218 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
   1219 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
   1220 		error = EAUTH;
   1221 	}
   1222 	if (error)
   1223 		free((void *)*auth_str, M_TEMP);
   1224 	else {
   1225 		*auth_len = nmp->nm_authlen;
   1226 		*verf_len = nmp->nm_verflen;
   1227 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
   1228 	}
   1229 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
   1230 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
   1231 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
   1232 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1233 		wakeup((void *)&nmp->nm_authtype);
   1234 	}
   1235 	return (error);
   1236 }
   1237 
   1238 /*
   1239  * Get a nickname authenticator and verifier.
   1240  */
   1241 int
   1242 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
   1243     int *auth_len, char *verf_str, int verf_len)
   1244 {
   1245 	struct timeval ktvin, ktvout, tv;
   1246 	struct nfsuid *nuidp;
   1247 	u_int32_t *nickp, *verfp;
   1248 
   1249 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
   1250 
   1251 #ifdef DIAGNOSTIC
   1252 	if (verf_len < (4 * NFSX_UNSIGNED))
   1253 		panic("nfs_getnickauth verf too small");
   1254 #endif
   1255 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
   1256 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
   1257 			break;
   1258 	}
   1259 	if (!nuidp || nuidp->nu_expire < time_second)
   1260 		return (EACCES);
   1261 
   1262 	/*
   1263 	 * Move to the end of the lru list (end of lru == most recently used).
   1264 	 */
   1265 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1266 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1267 
   1268 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
   1269 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1270 	*nickp = txdr_unsigned(nuidp->nu_nickname);
   1271 	*auth_str = (char *)nickp;
   1272 	*auth_len = 2 * NFSX_UNSIGNED;
   1273 
   1274 	/*
   1275 	 * Now we must encrypt the verifier and package it up.
   1276 	 */
   1277 	verfp = (u_int32_t *)verf_str;
   1278 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1279 	getmicrotime(&tv);
   1280 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
   1281 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
   1282 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
   1283 		nuidp->nu_timestamp = tv;
   1284 	else
   1285 		nuidp->nu_timestamp.tv_usec++;
   1286 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
   1287 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
   1288 
   1289 	/*
   1290 	 * Now encrypt the timestamp verifier in ecb mode using the session
   1291 	 * key.
   1292 	 */
   1293 #ifdef NFSKERB
   1294 	XXX
   1295 #endif
   1296 
   1297 	*verfp++ = ktvout.tv_sec;
   1298 	*verfp++ = ktvout.tv_usec;
   1299 	*verfp = 0;
   1300 	return (0);
   1301 }
   1302 
   1303 /*
   1304  * Save the current nickname in a hash list entry on the mount point.
   1305  */
   1306 int
   1307 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
   1308 	struct nfsmount *nmp;
   1309 	kauth_cred_t cred;
   1310 	int len;
   1311 	NFSKERBKEY_T key;
   1312 	struct mbuf **mdp;
   1313 	char **dposp;
   1314 	struct mbuf *mrep;
   1315 {
   1316 	struct nfsuid *nuidp;
   1317 	u_int32_t *tl;
   1318 	int32_t t1;
   1319 	struct mbuf *md = *mdp;
   1320 	struct timeval ktvin, ktvout;
   1321 	u_int32_t nick;
   1322 	char *dpos = *dposp, *cp2;
   1323 	int deltasec, error = 0;
   1324 
   1325 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
   1326 
   1327 	if (len == (3 * NFSX_UNSIGNED)) {
   1328 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
   1329 		ktvin.tv_sec = *tl++;
   1330 		ktvin.tv_usec = *tl++;
   1331 		nick = fxdr_unsigned(u_int32_t, *tl);
   1332 
   1333 		/*
   1334 		 * Decrypt the timestamp in ecb mode.
   1335 		 */
   1336 #ifdef NFSKERB
   1337 		XXX
   1338 #endif
   1339 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
   1340 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
   1341 		deltasec = time_second - ktvout.tv_sec;
   1342 		if (deltasec < 0)
   1343 			deltasec = -deltasec;
   1344 		/*
   1345 		 * If ok, add it to the hash list for the mount point.
   1346 		 */
   1347 		if (deltasec <= NFS_KERBCLOCKSKEW) {
   1348 			if (nmp->nm_numuids < nuidhash_max) {
   1349 				nmp->nm_numuids++;
   1350 				nuidp = (struct nfsuid *)
   1351 				   malloc(sizeof (struct nfsuid), M_NFSUID,
   1352 					M_WAITOK);
   1353 			} else {
   1354 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
   1355 				LIST_REMOVE(nuidp, nu_hash);
   1356 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
   1357 					nu_lru);
   1358 			}
   1359 			nuidp->nu_flag = 0;
   1360 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
   1361 			nuidp->nu_expire = time_second + NFS_KERBTTL;
   1362 			nuidp->nu_timestamp = ktvout;
   1363 			nuidp->nu_nickname = nick;
   1364 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
   1365 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
   1366 				nu_lru);
   1367 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
   1368 				nuidp, nu_hash);
   1369 		}
   1370 	} else
   1371 		nfsm_adv(nfsm_rndup(len));
   1372 nfsmout:
   1373 	*mdp = md;
   1374 	*dposp = dpos;
   1375 	return (error);
   1376 }
   1377 #endif /* NFS */
   1378