Home | History | Annotate | Line # | Download | only in nfs
nfs_syscalls.c revision 1.140.4.1
      1 /*	$NetBSD: nfs_syscalls.c,v 1.140.4.1 2009/04/13 21:21:30 snj 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 1.140.4.1 2009/04/13 21:21:30 snj 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 #ifdef INET6
    322 				    case AF_INET6:
    323 #endif
    324 #ifdef ISO
    325 				    case AF_ISO:
    326 #endif
    327 					nuidp->nu_flag |= NU_NAM;
    328 					nuidp->nu_nam = m_copym(
    329 					    nfsd->nfsd_nd->nd_nam2, 0,
    330 					     M_COPYALL, M_WAIT);
    331 					break;
    332 				    default:
    333 					return EAFNOSUPPORT;
    334 				    };
    335 				}
    336 				TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
    337 					nu_lru);
    338 				LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
    339 					nuidp, nu_hash);
    340 				kauth_cred_hold(nuidp->nu_cr);
    341 				nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    342 				nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    343 			    }
    344 			}
    345 		}
    346 		if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) &&
    347 		    (nfsd = nsd->nsd_nfsd))
    348 			nfsd->nfsd_flag |= NFSD_AUTHFAIL;
    349 		error = nfssvc_nfsd(nsd, SCARG(uap, argp), l);
    350 #endif /* !NFSSERVER */
    351 	}
    352 	if (error == EINTR || error == ERESTART)
    353 		error = 0;
    354 	return (error);
    355 }
    356 
    357 #ifdef NFSSERVER
    358 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
    359 
    360 static struct nfssvc_sock *
    361 nfsrv_sockalloc()
    362 {
    363 	struct nfssvc_sock *slp;
    364 
    365 	slp = kmem_alloc(sizeof(*slp), KM_SLEEP);
    366 	memset(slp, 0, sizeof (struct nfssvc_sock));
    367 	mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET);
    368 	mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET);
    369 	cv_init(&slp->ns_cv, "nfsdsock");
    370 	TAILQ_INIT(&slp->ns_uidlruhead);
    371 	LIST_INIT(&slp->ns_tq);
    372 	SIMPLEQ_INIT(&slp->ns_sendq);
    373 	mutex_enter(&nfsd_lock);
    374 	TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
    375 	mutex_exit(&nfsd_lock);
    376 
    377 	return slp;
    378 }
    379 
    380 static void
    381 nfsrv_sockfree(struct nfssvc_sock *slp)
    382 {
    383 
    384 	KASSERT(slp->ns_so == NULL);
    385 	KASSERT(slp->ns_fp == NULL);
    386 	KASSERT((slp->ns_flags & SLP_VALID) == 0);
    387 	mutex_destroy(&slp->ns_lock);
    388 	mutex_destroy(&slp->ns_alock);
    389 	cv_destroy(&slp->ns_cv);
    390 	kmem_free(slp, sizeof(*slp));
    391 }
    392 
    393 /*
    394  * Adds a socket to the list for servicing by nfsds.
    395  */
    396 int
    397 nfssvc_addsock(fp, mynam)
    398 	file_t *fp;
    399 	struct mbuf *mynam;
    400 {
    401 	int siz;
    402 	struct nfssvc_sock *slp;
    403 	struct socket *so;
    404 	struct nfssvc_sock *tslp;
    405 	int error;
    406 	int val;
    407 
    408 	so = (struct socket *)fp->f_data;
    409 	tslp = (struct nfssvc_sock *)0;
    410 	/*
    411 	 * Add it to the list, as required.
    412 	 */
    413 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
    414 #ifdef INET6
    415 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
    416 			tslp = nfs_udp6sock;
    417 		else
    418 #endif
    419 		tslp = nfs_udpsock;
    420 		if (tslp->ns_flags & SLP_VALID) {
    421 			m_freem(mynam);
    422 			return (EPERM);
    423 		}
    424 #ifdef ISO
    425 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
    426 		tslp = nfs_cltpsock;
    427 		if (tslp->ns_flags & SLP_VALID) {
    428 			m_freem(mynam);
    429 			return (EPERM);
    430 		}
    431 #endif /* ISO */
    432 	}
    433 	if (so->so_type == SOCK_STREAM)
    434 		siz = NFS_MAXPACKET + sizeof (u_long);
    435 	else
    436 		siz = NFS_MAXPACKET;
    437 	solock(so);
    438 	error = soreserve(so, siz, siz);
    439 	sounlock(so);
    440 	if (error) {
    441 		m_freem(mynam);
    442 		return (error);
    443 	}
    444 
    445 	/*
    446 	 * Set protocol specific options { for now TCP only } and
    447 	 * reserve some space. For datagram sockets, this can get called
    448 	 * repeatedly for the same socket, but that isn't harmful.
    449 	 */
    450 	if (so->so_type == SOCK_STREAM) {
    451 		val = 1;
    452 		so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val,
    453 		    sizeof(val));
    454 	}
    455 	if ((so->so_proto->pr_domain->dom_family == AF_INET
    456 #ifdef INET6
    457 	    || so->so_proto->pr_domain->dom_family == AF_INET6
    458 #endif
    459 	    ) &&
    460 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
    461 		val = 1;
    462 		so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val,
    463 		    sizeof(val));
    464 	}
    465 	solock(so);
    466 	so->so_rcv.sb_flags &= ~SB_NOINTR;
    467 	so->so_rcv.sb_timeo = 0;
    468 	so->so_snd.sb_flags &= ~SB_NOINTR;
    469 	so->so_snd.sb_timeo = 0;
    470 	sounlock(so);
    471 	if (tslp) {
    472 		slp = tslp;
    473 	} else {
    474 		slp = nfsrv_sockalloc();
    475 	}
    476 	slp->ns_so = so;
    477 	slp->ns_nam = mynam;
    478 	mutex_enter(&fp->f_lock);
    479 	fp->f_count++;
    480 	mutex_exit(&fp->f_lock);
    481 	slp->ns_fp = fp;
    482 	slp->ns_flags = SLP_VALID;
    483 	slp->ns_aflags = SLP_A_NEEDQ;
    484 	slp->ns_gflags = 0;
    485 	slp->ns_sflags = 0;
    486 	solock(so);
    487 	so->so_upcallarg = (void *)slp;
    488 	so->so_upcall = nfsrv_soupcall;
    489 	so->so_rcv.sb_flags |= SB_UPCALL;
    490 	sounlock(so);
    491 	nfsrv_wakenfsd(slp);
    492 	return (0);
    493 }
    494 
    495 /*
    496  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    497  * until it is killed by a signal.
    498  */
    499 int
    500 nfssvc_nfsd(nsd, argp, l)
    501 	struct nfsd_srvargs *nsd;
    502 	void *argp;
    503 	struct lwp *l;
    504 {
    505 	struct timeval tv;
    506 	struct mbuf *m;
    507 	struct nfssvc_sock *slp;
    508 	struct nfsd *nfsd = nsd->nsd_nfsd;
    509 	struct nfsrv_descript *nd = NULL;
    510 	struct mbuf *mreq;
    511 	u_quad_t cur_usec;
    512 	int error = 0, cacherep, siz, sotype, writes_todo;
    513 	struct proc *p = l->l_proc;
    514 	bool doreinit;
    515 
    516 #ifndef nolint
    517 	cacherep = RC_DOIT;
    518 	writes_todo = 0;
    519 #endif
    520 	uvm_lwp_hold(l);
    521 	if (nfsd == NULL) {
    522 		nsd->nsd_nfsd = nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP);
    523 		memset(nfsd, 0, sizeof (struct nfsd));
    524 		cv_init(&nfsd->nfsd_cv, "nfsd");
    525 		nfsd->nfsd_procp = p;
    526 		mutex_enter(&nfsd_lock);
    527 		while ((nfssvc_sockhead_flag & SLP_INIT) != 0) {
    528 			KASSERT(nfs_numnfsd == 0);
    529 			cv_wait(&nfsd_initcv, &nfsd_lock);
    530 		}
    531 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
    532 		nfs_numnfsd++;
    533 		mutex_exit(&nfsd_lock);
    534 	}
    535 	/*
    536 	 * Loop getting rpc requests until SIGKILL.
    537 	 */
    538 	for (;;) {
    539 		bool dummy;
    540 
    541 		if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    542 		    != 0) {
    543 			preempt();
    544 		}
    545 		if (nfsd->nfsd_slp == NULL) {
    546 			mutex_enter(&nfsd_lock);
    547 			while (nfsd->nfsd_slp == NULL &&
    548 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
    549 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
    550 				    nfsd_idle);
    551 				error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock);
    552 				if (error) {
    553 					slp = nfsd->nfsd_slp;
    554 					nfsd->nfsd_slp = NULL;
    555 					if (!slp)
    556 						SLIST_REMOVE(&nfsd_idle_head,
    557 						    nfsd, nfsd, nfsd_idle);
    558 					mutex_exit(&nfsd_lock);
    559 					if (slp) {
    560 						nfsrv_wakenfsd(slp);
    561 						nfsrv_slpderef(slp);
    562 					}
    563 					goto done;
    564 				}
    565 			}
    566 			if (nfsd->nfsd_slp == NULL &&
    567 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
    568 				slp = TAILQ_FIRST(&nfssvc_sockpending);
    569 				if (slp) {
    570 					KASSERT((slp->ns_gflags & SLP_G_DOREC)
    571 					    != 0);
    572 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
    573 					    ns_pending);
    574 					slp->ns_gflags &= ~SLP_G_DOREC;
    575 					slp->ns_sref++;
    576 					nfsd->nfsd_slp = slp;
    577 				} else
    578 					nfsd_head_flag &= ~NFSD_CHECKSLP;
    579 			}
    580 			KASSERT(nfsd->nfsd_slp == NULL ||
    581 			    nfsd->nfsd_slp->ns_sref > 0);
    582 			mutex_exit(&nfsd_lock);
    583 			if ((slp = nfsd->nfsd_slp) == NULL)
    584 				continue;
    585 			if (slp->ns_flags & SLP_VALID) {
    586 				bool more;
    587 
    588 				if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) {
    589 					nfsrv_rcv(slp);
    590 				}
    591 				if (nfsdsock_testbits(slp, SLP_A_DISCONN)) {
    592 					nfsrv_zapsock(slp);
    593 				}
    594 				error = nfsrv_dorec(slp, nfsd, &nd, &more);
    595 				getmicrotime(&tv);
    596 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    597 					(u_quad_t)tv.tv_usec;
    598 				writes_todo = 0;
    599 				if (error) {
    600 					struct nfsrv_descript *nd2;
    601 
    602 					mutex_enter(&nfsd_lock);
    603 					nd2 = LIST_FIRST(&slp->ns_tq);
    604 					if (nd2 != NULL &&
    605 					    nd2->nd_time <= cur_usec) {
    606 						error = 0;
    607 						cacherep = RC_DOIT;
    608 						writes_todo = 1;
    609 					}
    610 					mutex_exit(&nfsd_lock);
    611 				}
    612 				if (error == 0 && more) {
    613 					nfsrv_wakenfsd(slp);
    614 				}
    615 			}
    616 		} else {
    617 			error = 0;
    618 			slp = nfsd->nfsd_slp;
    619 		}
    620 		KASSERT(slp != NULL);
    621 		KASSERT(nfsd->nfsd_slp == slp);
    622 		if (error || (slp->ns_flags & SLP_VALID) == 0) {
    623 			if (nd) {
    624 				nfsdreq_free(nd);
    625 				nd = NULL;
    626 			}
    627 			nfsd->nfsd_slp = NULL;
    628 			nfsrv_slpderef(slp);
    629 			continue;
    630 		}
    631 		sotype = slp->ns_so->so_type;
    632 		if (nd) {
    633 			getmicrotime(&nd->nd_starttime);
    634 			if (nd->nd_nam2)
    635 				nd->nd_nam = nd->nd_nam2;
    636 			else
    637 				nd->nd_nam = slp->ns_nam;
    638 
    639 			/*
    640 			 * Check to see if authorization is needed.
    641 			 */
    642 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
    643 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
    644 				nsd->nsd_haddr = mtod(nd->nd_nam,
    645 				    struct sockaddr_in *)->sin_addr.s_addr;
    646 				nsd->nsd_authlen = nfsd->nfsd_authlen;
    647 				nsd->nsd_verflen = nfsd->nfsd_verflen;
    648 				if (!copyout(nfsd->nfsd_authstr,
    649 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
    650 				    !copyout(nfsd->nfsd_verfstr,
    651 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
    652 				    !copyout(nsd, argp, sizeof (*nsd))) {
    653 					uvm_lwp_rele(l);
    654 					return (ENEEDAUTH);
    655 				}
    656 				cacherep = RC_DROPIT;
    657 			} else
    658 				cacherep = nfsrv_getcache(nd, slp, &mreq);
    659 
    660 			if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
    661 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
    662 				nd->nd_procnum = NFSPROC_NOOP;
    663 				nd->nd_repstat =
    664 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
    665 				cacherep = RC_DOIT;
    666 			}
    667 		}
    668 
    669 		/*
    670 		 * Loop to get all the write rpc relies that have been
    671 		 * gathered together.
    672 		 */
    673 		do {
    674 			switch (cacherep) {
    675 			case RC_DOIT:
    676 				mreq = NULL;
    677 				netexport_rdlock();
    678 				if (writes_todo || nd == NULL ||
    679 				     (!(nd->nd_flag & ND_NFSV3) &&
    680 				     nd->nd_procnum == NFSPROC_WRITE &&
    681 				     nfsrvw_procrastinate > 0))
    682 					error = nfsrv_writegather(&nd, slp,
    683 					    l, &mreq);
    684 				else
    685 					error =
    686 					    (*(nfsrv3_procs[nd->nd_procnum]))
    687 					    (nd, slp, l, &mreq);
    688 				netexport_rdunlock();
    689 				if (mreq == NULL) {
    690 					if (nd != NULL) {
    691 						if (nd->nd_nam2)
    692 							m_free(nd->nd_nam2);
    693 					}
    694 					break;
    695 				}
    696 				if (error) {
    697 					nfsstats.srv_errs++;
    698 					nfsrv_updatecache(nd, false, mreq);
    699 					if (nd->nd_nam2)
    700 						m_freem(nd->nd_nam2);
    701 					break;
    702 				}
    703 				nfsstats.srvrpccnt[nd->nd_procnum]++;
    704 				nfsrv_updatecache(nd, true, mreq);
    705 				nd->nd_mrep = (struct mbuf *)0;
    706 			case RC_REPLY:
    707 				m = mreq;
    708 				siz = 0;
    709 				while (m) {
    710 					siz += m->m_len;
    711 					m = m->m_next;
    712 				}
    713 				if (siz <= 0 || siz > NFS_MAXPACKET) {
    714 					printf("mbuf siz=%d\n",siz);
    715 					panic("Bad nfs svc reply");
    716 				}
    717 				m = mreq;
    718 				m->m_pkthdr.len = siz;
    719 				m->m_pkthdr.rcvif = (struct ifnet *)0;
    720 				/*
    721 				 * For stream protocols, prepend a Sun RPC
    722 				 * Record Mark.
    723 				 */
    724 				if (sotype == SOCK_STREAM) {
    725 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
    726 					*mtod(m, u_int32_t *) =
    727 					    htonl(0x80000000 | siz);
    728 				}
    729 				nd->nd_mreq = m;
    730 				if (nfsrtton) {
    731 					nfsd_rt(slp->ns_so->so_type, nd,
    732 					    cacherep);
    733 				}
    734 				error = nfsdsock_sendreply(slp, nd);
    735 				nd = NULL;
    736 				if (error == EPIPE)
    737 					nfsrv_zapsock(slp);
    738 				if (error == EINTR || error == ERESTART) {
    739 					nfsd->nfsd_slp = NULL;
    740 					nfsrv_slpderef(slp);
    741 					goto done;
    742 				}
    743 				break;
    744 			case RC_DROPIT:
    745 				if (nfsrtton)
    746 					nfsd_rt(sotype, nd, cacherep);
    747 				m_freem(nd->nd_mrep);
    748 				m_freem(nd->nd_nam2);
    749 				break;
    750 			}
    751 			if (nd) {
    752 				nfsdreq_free(nd);
    753 				nd = NULL;
    754 			}
    755 
    756 			/*
    757 			 * Check to see if there are outstanding writes that
    758 			 * need to be serviced.
    759 			 */
    760 			getmicrotime(&tv);
    761 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    762 			    (u_quad_t)tv.tv_usec;
    763 			mutex_enter(&nfsd_lock);
    764 			if (LIST_FIRST(&slp->ns_tq) &&
    765 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
    766 				cacherep = RC_DOIT;
    767 				writes_todo = 1;
    768 			} else
    769 				writes_todo = 0;
    770 			mutex_exit(&nfsd_lock);
    771 		} while (writes_todo);
    772 		if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) {
    773 			nfsd->nfsd_slp = NULL;
    774 			nfsrv_slpderef(slp);
    775 		}
    776 	}
    777 done:
    778 	mutex_enter(&nfsd_lock);
    779 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
    780 	doreinit = --nfs_numnfsd == 0;
    781 	if (doreinit)
    782 		nfssvc_sockhead_flag |= SLP_INIT;
    783 	mutex_exit(&nfsd_lock);
    784 	cv_destroy(&nfsd->nfsd_cv);
    785 	kmem_free(nfsd, sizeof(*nfsd));
    786 	nsd->nsd_nfsd = NULL;
    787 	if (doreinit)
    788 		nfsrv_init(true);	/* Reinitialize everything */
    789 	uvm_lwp_rele(l);
    790 	return (error);
    791 }
    792 
    793 /*
    794  * Shut down a socket associated with an nfssvc_sock structure.
    795  * Should be called with the send lock set, if required.
    796  * The trick here is to increment the sref at the start, so that the nfsds
    797  * will stop using it and clear ns_flag at the end so that it will not be
    798  * reassigned during cleanup.
    799  *
    800  * called at splsoftnet.
    801  */
    802 void
    803 nfsrv_zapsock(slp)
    804 	struct nfssvc_sock *slp;
    805 {
    806 	struct nfsuid *nuidp, *nnuidp;
    807 	struct nfsrv_descript *nwp;
    808 	struct socket *so;
    809 	struct mbuf *m;
    810 
    811 	if (nfsdsock_drain(slp)) {
    812 		return;
    813 	}
    814 	mutex_enter(&nfsd_lock);
    815 	if (slp->ns_gflags & SLP_G_DOREC) {
    816 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
    817 		slp->ns_gflags &= ~SLP_G_DOREC;
    818 	}
    819 	mutex_exit(&nfsd_lock);
    820 
    821 	so = slp->ns_so;
    822 	KASSERT(so != NULL);
    823 	solock(so);
    824 	so->so_upcall = NULL;
    825 	so->so_upcallarg = NULL;
    826 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    827 	soshutdown(so, SHUT_RDWR);
    828 	sounlock(so);
    829 
    830 	m_freem(slp->ns_raw);
    831 	m = slp->ns_rec;
    832 	while (m != NULL) {
    833 		struct mbuf *n;
    834 
    835 		n = m->m_nextpkt;
    836 		m_freem(m);
    837 		m = n;
    838 	}
    839 	/* XXX what about freeing ns_frag ? */
    840 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
    841 	    nuidp = nnuidp) {
    842 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
    843 		LIST_REMOVE(nuidp, nu_hash);
    844 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
    845 		if (nuidp->nu_flag & NU_NAM)
    846 			m_freem(nuidp->nu_nam);
    847 		kmem_free(nuidp, sizeof(*nuidp));
    848 	}
    849 	mutex_enter(&nfsd_lock);
    850 	while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) {
    851 		LIST_REMOVE(nwp, nd_tq);
    852 		mutex_exit(&nfsd_lock);
    853 		nfsdreq_free(nwp);
    854 		mutex_enter(&nfsd_lock);
    855 	}
    856 	mutex_exit(&nfsd_lock);
    857 }
    858 
    859 /*
    860  * Derefence a server socket structure. If it has no more references and
    861  * is no longer valid, you can throw it away.
    862  */
    863 void
    864 nfsrv_slpderef(slp)
    865 	struct nfssvc_sock *slp;
    866 {
    867 	uint32_t ref;
    868 
    869 	mutex_enter(&nfsd_lock);
    870 	KASSERT(slp->ns_sref > 0);
    871 	ref = --slp->ns_sref;
    872 	if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) {
    873 		file_t *fp;
    874 
    875 		KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0);
    876 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
    877 		mutex_exit(&nfsd_lock);
    878 
    879 		fp = slp->ns_fp;
    880 		if (fp != NULL) {
    881 			slp->ns_fp = NULL;
    882 			KASSERT(fp != NULL);
    883 			KASSERT(fp->f_data == slp->ns_so);
    884 			KASSERT(fp->f_count > 0);
    885 			closef(fp);
    886 			slp->ns_so = NULL;
    887 		}
    888 
    889 		if (slp->ns_nam)
    890 			m_free(slp->ns_nam);
    891 		nfsrv_sockfree(slp);
    892 	} else
    893 		mutex_exit(&nfsd_lock);
    894 }
    895 
    896 /*
    897  * Initialize the data structures for the server.
    898  * Handshake with any new nfsds starting up to avoid any chance of
    899  * corruption.
    900  */
    901 void
    902 nfsrv_init(terminating)
    903 	int terminating;
    904 {
    905 	struct nfssvc_sock *slp;
    906 
    907 	if (!terminating) {
    908 		mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET);
    909 		cv_init(&nfsd_initcv, "nfsdinit");
    910 	}
    911 
    912 	mutex_enter(&nfsd_lock);
    913 	if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0)
    914 		panic("nfsd init");
    915 	nfssvc_sockhead_flag |= SLP_INIT;
    916 
    917 	if (terminating) {
    918 		KASSERT(SLIST_EMPTY(&nfsd_idle_head));
    919 		KASSERT(TAILQ_EMPTY(&nfsd_head));
    920 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
    921 			mutex_exit(&nfsd_lock);
    922 			KASSERT(slp->ns_sref == 0);
    923 			slp->ns_sref++;
    924 			nfsrv_zapsock(slp);
    925 			nfsrv_slpderef(slp);
    926 			mutex_enter(&nfsd_lock);
    927 		}
    928 		KASSERT(TAILQ_EMPTY(&nfssvc_sockpending));
    929 		mutex_exit(&nfsd_lock);
    930 		nfsrv_cleancache();	/* And clear out server cache */
    931 	} else {
    932 		mutex_exit(&nfsd_lock);
    933 		nfs_pub.np_valid = 0;
    934 	}
    935 
    936 	TAILQ_INIT(&nfssvc_sockhead);
    937 	TAILQ_INIT(&nfssvc_sockpending);
    938 
    939 	TAILQ_INIT(&nfsd_head);
    940 	SLIST_INIT(&nfsd_idle_head);
    941 	nfsd_head_flag &= ~NFSD_CHECKSLP;
    942 
    943 	nfs_udpsock = nfsrv_sockalloc();
    944 
    945 #ifdef INET6
    946 	nfs_udp6sock = nfsrv_sockalloc();
    947 #endif
    948 
    949 #ifdef ISO
    950 	nfs_cltpsock = nfsrv_sockalloc();
    951 #endif
    952 
    953 	mutex_enter(&nfsd_lock);
    954 	nfssvc_sockhead_flag &= ~SLP_INIT;
    955 	cv_broadcast(&nfsd_initcv);
    956 	mutex_exit(&nfsd_lock);
    957 }
    958 
    959 /*
    960  * Add entries to the server monitor log.
    961  */
    962 static void
    963 nfsd_rt(sotype, nd, cacherep)
    964 	int sotype;
    965 	struct nfsrv_descript *nd;
    966 	int cacherep;
    967 {
    968 	struct timeval tv;
    969 	struct drt *rt;
    970 
    971 	rt = &nfsdrt.drt[nfsdrt.pos];
    972 	if (cacherep == RC_DOIT)
    973 		rt->flag = 0;
    974 	else if (cacherep == RC_REPLY)
    975 		rt->flag = DRT_CACHEREPLY;
    976 	else
    977 		rt->flag = DRT_CACHEDROP;
    978 	if (sotype == SOCK_STREAM)
    979 		rt->flag |= DRT_TCP;
    980 	if (nd->nd_flag & ND_NFSV3)
    981 		rt->flag |= DRT_NFSV3;
    982 	rt->proc = nd->nd_procnum;
    983 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
    984 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
    985 	else
    986 	    rt->ipadr = INADDR_ANY;
    987 	getmicrotime(&tv);
    988 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
    989 		(tv.tv_usec - nd->nd_starttime.tv_usec);
    990 	rt->tstamp = tv;
    991 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
    992 }
    993 #endif /* NFSSERVER */
    994 
    995 #ifdef NFS
    996 
    997 int nfs_defect = 0;
    998 /*
    999  * Asynchronous I/O threads for client nfs.
   1000  * They do read-ahead and write-behind operations on the block I/O cache.
   1001  * Never returns unless it fails or gets killed.
   1002  */
   1003 
   1004 static void
   1005 nfssvc_iod(void *arg)
   1006 {
   1007 	struct buf *bp;
   1008 	struct nfs_iod *myiod;
   1009 	struct nfsmount *nmp;
   1010 
   1011 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
   1012 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
   1013 	cv_init(&myiod->nid_cv, "nfsiod");
   1014 	myiod->nid_exiting = false;
   1015 	myiod->nid_mount = NULL;
   1016 	mutex_enter(&nfs_iodlist_lock);
   1017 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
   1018 	mutex_exit(&nfs_iodlist_lock);
   1019 
   1020 	for (;;) {
   1021 		mutex_enter(&nfs_iodlist_lock);
   1022 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
   1023 		mutex_exit(&nfs_iodlist_lock);
   1024 
   1025 		mutex_enter(&myiod->nid_lock);
   1026 		while (/*CONSTCOND*/ true) {
   1027 			nmp = myiod->nid_mount;
   1028 			if (nmp) {
   1029 				myiod->nid_mount = NULL;
   1030 				break;
   1031 			}
   1032 			if (__predict_false(myiod->nid_exiting)) {
   1033 				/*
   1034 				 * drop nid_lock to preserve locking order.
   1035 				 */
   1036 				mutex_exit(&myiod->nid_lock);
   1037 				mutex_enter(&nfs_iodlist_lock);
   1038 				mutex_enter(&myiod->nid_lock);
   1039 				/*
   1040 				 * recheck nid_mount because nfs_asyncio can
   1041 				 * pick us in the meantime as we are still on
   1042 				 * nfs_iodlist_lock.
   1043 				 */
   1044 				if (myiod->nid_mount != NULL) {
   1045 					mutex_exit(&nfs_iodlist_lock);
   1046 					continue;
   1047 				}
   1048 				LIST_REMOVE(myiod, nid_idle);
   1049 				mutex_exit(&nfs_iodlist_lock);
   1050 				goto quit;
   1051 			}
   1052 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
   1053 		}
   1054 		mutex_exit(&myiod->nid_lock);
   1055 
   1056 		mutex_enter(&nmp->nm_lock);
   1057 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
   1058 			/* Take one off the front of the list */
   1059 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
   1060 			nmp->nm_bufqlen--;
   1061 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
   1062 				cv_broadcast(&nmp->nm_aiocv);
   1063 			}
   1064 			mutex_exit(&nmp->nm_lock);
   1065 			KERNEL_LOCK(1, curlwp);
   1066 			(void)nfs_doio(bp);
   1067 			KERNEL_UNLOCK_LAST(curlwp);
   1068 			mutex_enter(&nmp->nm_lock);
   1069 			/*
   1070 			 * If there are more than one iod on this mount,
   1071 			 * then defect so that the iods can be shared out
   1072 			 * fairly between the mounts
   1073 			 */
   1074 			if (nfs_defect && nmp->nm_bufqiods > 1) {
   1075 				break;
   1076 			}
   1077 		}
   1078 		KASSERT(nmp->nm_bufqiods > 0);
   1079 		nmp->nm_bufqiods--;
   1080 		mutex_exit(&nmp->nm_lock);
   1081 	}
   1082 quit:
   1083 	KASSERT(myiod->nid_mount == NULL);
   1084 	mutex_exit(&myiod->nid_lock);
   1085 
   1086 	cv_destroy(&myiod->nid_cv);
   1087 	mutex_destroy(&myiod->nid_lock);
   1088 	kmem_free(myiod, sizeof(*myiod));
   1089 
   1090 	kthread_exit(0);
   1091 }
   1092 
   1093 void
   1094 nfs_iodinit()
   1095 {
   1096 
   1097 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
   1098 	LIST_INIT(&nfs_iodlist_all);
   1099 	LIST_INIT(&nfs_iodlist_idle);
   1100 }
   1101 
   1102 int
   1103 nfs_set_niothreads(int newval)
   1104 {
   1105 	struct nfs_iod *nid;
   1106 	int error = 0;
   1107         int hold_count;
   1108 
   1109 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
   1110 
   1111 	mutex_enter(&nfs_iodlist_lock);
   1112 	/* clamp to sane range */
   1113 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
   1114 
   1115 	while (nfs_numasync != nfs_niothreads && error == 0) {
   1116 		while (nfs_numasync < nfs_niothreads) {
   1117 
   1118 			/*
   1119 			 * kthread_create can wait for pagedaemon and
   1120 			 * pagedaemon can wait for nfsiod which needs to acquire
   1121 			 * nfs_iodlist_lock.
   1122 			 */
   1123 
   1124 			mutex_exit(&nfs_iodlist_lock);
   1125 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
   1126 			    nfssvc_iod, NULL, NULL, "nfsio");
   1127 			mutex_enter(&nfs_iodlist_lock);
   1128 			if (error) {
   1129 				/* give up */
   1130 				nfs_niothreads = nfs_numasync;
   1131 				break;
   1132 			}
   1133 			nfs_numasync++;
   1134 		}
   1135 		while (nfs_numasync > nfs_niothreads) {
   1136 			nid = LIST_FIRST(&nfs_iodlist_all);
   1137 			if (nid == NULL) {
   1138 				/* iod has not started yet. */
   1139 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
   1140 				continue;
   1141 			}
   1142 			LIST_REMOVE(nid, nid_all);
   1143 			mutex_enter(&nid->nid_lock);
   1144 			KASSERT(!nid->nid_exiting);
   1145 			nid->nid_exiting = true;
   1146 			cv_signal(&nid->nid_cv);
   1147 			mutex_exit(&nid->nid_lock);
   1148 			nfs_numasync--;
   1149 		}
   1150 	}
   1151 	mutex_exit(&nfs_iodlist_lock);
   1152 
   1153 	KERNEL_LOCK(hold_count, curlwp);
   1154 	return error;
   1155 }
   1156 
   1157 /*
   1158  * Get an authorization string for the uid by having the mount_nfs sitting
   1159  * on this mount point porpous out of the kernel and do it.
   1160  */
   1161 int
   1162 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
   1163 	struct nfsmount *nmp;
   1164 	struct nfsreq *rep;
   1165 	kauth_cred_t cred;
   1166 	char **auth_str;
   1167 	int *auth_len;
   1168 	char *verf_str;
   1169 	int *verf_len;
   1170 	NFSKERBKEY_T key;		/* return session key */
   1171 {
   1172 	int error = 0;
   1173 
   1174 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
   1175 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
   1176 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
   1177 			"nfsauth1", 2 * hz);
   1178 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1179 		if (error) {
   1180 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1181 			return (error);
   1182 		}
   1183 	}
   1184 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
   1185 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
   1186 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
   1187 	nmp->nm_verfstr = verf_str;
   1188 	nmp->nm_verflen = *verf_len;
   1189 	nmp->nm_authuid = kauth_cred_geteuid(cred);
   1190 	wakeup((void *)&nmp->nm_authstr);
   1191 
   1192 	/*
   1193 	 * And wait for mount_nfs to do its stuff.
   1194 	 */
   1195 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
   1196 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
   1197 			"nfsauth2", 2 * hz);
   1198 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1199 	}
   1200 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
   1201 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
   1202 		error = EAUTH;
   1203 	}
   1204 	if (error)
   1205 		free((void *)*auth_str, M_TEMP);
   1206 	else {
   1207 		*auth_len = nmp->nm_authlen;
   1208 		*verf_len = nmp->nm_verflen;
   1209 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
   1210 	}
   1211 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
   1212 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
   1213 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
   1214 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1215 		wakeup((void *)&nmp->nm_authtype);
   1216 	}
   1217 	return (error);
   1218 }
   1219 
   1220 /*
   1221  * Get a nickname authenticator and verifier.
   1222  */
   1223 int
   1224 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
   1225     int *auth_len, char *verf_str, int verf_len)
   1226 {
   1227 	struct timeval ktvin, ktvout, tv;
   1228 	struct nfsuid *nuidp;
   1229 	u_int32_t *nickp, *verfp;
   1230 
   1231 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
   1232 
   1233 #ifdef DIAGNOSTIC
   1234 	if (verf_len < (4 * NFSX_UNSIGNED))
   1235 		panic("nfs_getnickauth verf too small");
   1236 #endif
   1237 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
   1238 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
   1239 			break;
   1240 	}
   1241 	if (!nuidp || nuidp->nu_expire < time_second)
   1242 		return (EACCES);
   1243 
   1244 	/*
   1245 	 * Move to the end of the lru list (end of lru == most recently used).
   1246 	 */
   1247 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1248 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1249 
   1250 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
   1251 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1252 	*nickp = txdr_unsigned(nuidp->nu_nickname);
   1253 	*auth_str = (char *)nickp;
   1254 	*auth_len = 2 * NFSX_UNSIGNED;
   1255 
   1256 	/*
   1257 	 * Now we must encrypt the verifier and package it up.
   1258 	 */
   1259 	verfp = (u_int32_t *)verf_str;
   1260 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1261 	getmicrotime(&tv);
   1262 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
   1263 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
   1264 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
   1265 		nuidp->nu_timestamp = tv;
   1266 	else
   1267 		nuidp->nu_timestamp.tv_usec++;
   1268 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
   1269 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
   1270 
   1271 	/*
   1272 	 * Now encrypt the timestamp verifier in ecb mode using the session
   1273 	 * key.
   1274 	 */
   1275 #ifdef NFSKERB
   1276 	XXX
   1277 #endif
   1278 
   1279 	*verfp++ = ktvout.tv_sec;
   1280 	*verfp++ = ktvout.tv_usec;
   1281 	*verfp = 0;
   1282 	return (0);
   1283 }
   1284 
   1285 /*
   1286  * Save the current nickname in a hash list entry on the mount point.
   1287  */
   1288 int
   1289 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
   1290 	struct nfsmount *nmp;
   1291 	kauth_cred_t cred;
   1292 	int len;
   1293 	NFSKERBKEY_T key;
   1294 	struct mbuf **mdp;
   1295 	char **dposp;
   1296 	struct mbuf *mrep;
   1297 {
   1298 	struct nfsuid *nuidp;
   1299 	u_int32_t *tl;
   1300 	int32_t t1;
   1301 	struct mbuf *md = *mdp;
   1302 	struct timeval ktvin, ktvout;
   1303 	u_int32_t nick;
   1304 	char *dpos = *dposp, *cp2;
   1305 	int deltasec, error = 0;
   1306 
   1307 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
   1308 
   1309 	if (len == (3 * NFSX_UNSIGNED)) {
   1310 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
   1311 		ktvin.tv_sec = *tl++;
   1312 		ktvin.tv_usec = *tl++;
   1313 		nick = fxdr_unsigned(u_int32_t, *tl);
   1314 
   1315 		/*
   1316 		 * Decrypt the timestamp in ecb mode.
   1317 		 */
   1318 #ifdef NFSKERB
   1319 		XXX
   1320 #endif
   1321 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
   1322 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
   1323 		deltasec = time_second - ktvout.tv_sec;
   1324 		if (deltasec < 0)
   1325 			deltasec = -deltasec;
   1326 		/*
   1327 		 * If ok, add it to the hash list for the mount point.
   1328 		 */
   1329 		if (deltasec <= NFS_KERBCLOCKSKEW) {
   1330 			if (nmp->nm_numuids < nuidhash_max) {
   1331 				nmp->nm_numuids++;
   1332 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
   1333 			} else {
   1334 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
   1335 				LIST_REMOVE(nuidp, nu_hash);
   1336 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
   1337 					nu_lru);
   1338 			}
   1339 			nuidp->nu_flag = 0;
   1340 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
   1341 			nuidp->nu_expire = time_second + NFS_KERBTTL;
   1342 			nuidp->nu_timestamp = ktvout;
   1343 			nuidp->nu_nickname = nick;
   1344 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
   1345 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
   1346 				nu_lru);
   1347 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
   1348 				nuidp, nu_hash);
   1349 		}
   1350 	} else
   1351 		nfsm_adv(nfsm_rndup(len));
   1352 nfsmout:
   1353 	*mdp = md;
   1354 	*dposp = dpos;
   1355 	return (error);
   1356 }
   1357 #endif /* NFS */
   1358