Home | History | Annotate | Line # | Download | only in nfs
nfs_syscalls.c revision 1.136
      1 /*	$NetBSD: nfs_syscalls.c,v 1.136 2008/05/20 00:18:31 ad 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 1.136 2008/05/20 00:18:31 ad 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 		error = getsock(nfsdarg.sock, &fp);
    202 		if (error)
    203 			return (error);
    204 		/*
    205 		 * Get the client address for connected sockets.
    206 		 */
    207 		if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
    208 			nam = (struct mbuf *)0;
    209 		else {
    210 			error = sockargs(&nam, nfsdarg.name, nfsdarg.namelen,
    211 				MT_SONAME);
    212 			if (error) {
    213 				fd_putfile(nfsdarg.sock);
    214 				return (error);
    215 			}
    216 		}
    217 		error = nfssvc_addsock(fp, nam);
    218 		fd_putfile(nfsdarg.sock);
    219 #endif /* !NFSSERVER */
    220 	} else if (SCARG(uap, flag) & NFSSVC_SETEXPORTSLIST) {
    221 #ifndef NFSSERVER
    222 		error = ENOSYS;
    223 #else
    224 		struct export_args *args;
    225 		struct mountd_exports_list mel;
    226 
    227 		error = copyin(SCARG(uap, argp), &mel, sizeof(mel));
    228 		if (error != 0)
    229 			return error;
    230 
    231 		args = (struct export_args *)malloc(mel.mel_nexports *
    232 		    sizeof(struct export_args), M_TEMP, M_WAITOK);
    233 		error = copyin(mel.mel_exports, args, mel.mel_nexports *
    234 		    sizeof(struct export_args));
    235 		if (error != 0) {
    236 			free(args, M_TEMP);
    237 			return error;
    238 		}
    239 		mel.mel_exports = args;
    240 
    241 		error = mountd_set_exports_list(&mel, l);
    242 
    243 		free(args, M_TEMP);
    244 #endif /* !NFSSERVER */
    245 	} else {
    246 #ifndef NFSSERVER
    247 		error = ENOSYS;
    248 #else
    249 		error = copyin(SCARG(uap, argp), (void *)nsd, sizeof (*nsd));
    250 		if (error)
    251 			return (error);
    252 		if ((SCARG(uap, flag) & NFSSVC_AUTHIN) &&
    253 		    ((nfsd = nsd->nsd_nfsd)) != NULL &&
    254 		    (nfsd->nfsd_slp->ns_flags & SLP_VALID)) {
    255 			slp = nfsd->nfsd_slp;
    256 
    257 			/*
    258 			 * First check to see if another nfsd has already
    259 			 * added this credential.
    260 			 */
    261 			LIST_FOREACH(nuidp, NUIDHASH(slp, nsd->nsd_cr.cr_uid),
    262 			    nu_hash) {
    263 				if (kauth_cred_geteuid(nuidp->nu_cr) ==
    264 				    nsd->nsd_cr.cr_uid &&
    265 				    (!nfsd->nfsd_nd->nd_nam2 ||
    266 				     netaddr_match(NU_NETFAM(nuidp),
    267 				     &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
    268 					break;
    269 			}
    270 			if (nuidp) {
    271 			    kauth_cred_hold(nuidp->nu_cr);
    272 			    nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    273 			    nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    274 			} else {
    275 			    /*
    276 			     * Nope, so we will.
    277 			     */
    278 			    if (slp->ns_numuids < nuidhash_max) {
    279 				slp->ns_numuids++;
    280 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
    281 			    } else
    282 				nuidp = (struct nfsuid *)0;
    283 			    if ((slp->ns_flags & SLP_VALID) == 0) {
    284 				if (nuidp)
    285 				    kmem_free(nuidp, sizeof(*nuidp));
    286 			    } else {
    287 				if (nuidp == (struct nfsuid *)0) {
    288 				    nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
    289 				    LIST_REMOVE(nuidp, nu_hash);
    290 				    TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
    291 					nu_lru);
    292 				    if (nuidp->nu_flag & NU_NAM)
    293 					m_freem(nuidp->nu_nam);
    294 			        }
    295 				nuidp->nu_flag = 0;
    296 				kauth_uucred_to_cred(nuidp->nu_cr,
    297 				    &nsd->nsd_cr);
    298 				nuidp->nu_timestamp = nsd->nsd_timestamp;
    299 				nuidp->nu_expire = time_second + nsd->nsd_ttl;
    300 				/*
    301 				 * and save the session key in nu_key.
    302 				 */
    303 				memcpy(nuidp->nu_key, nsd->nsd_key,
    304 				    sizeof(nsd->nsd_key));
    305 				if (nfsd->nfsd_nd->nd_nam2) {
    306 				    struct sockaddr_in *saddr;
    307 
    308 				    saddr = mtod(nfsd->nfsd_nd->nd_nam2,
    309 					 struct sockaddr_in *);
    310 				    switch (saddr->sin_family) {
    311 				    case AF_INET:
    312 					nuidp->nu_flag |= NU_INETADDR;
    313 					nuidp->nu_inetaddr =
    314 					     saddr->sin_addr.s_addr;
    315 					break;
    316 				    case AF_ISO:
    317 				    default:
    318 					nuidp->nu_flag |= NU_NAM;
    319 					nuidp->nu_nam = m_copym(
    320 					    nfsd->nfsd_nd->nd_nam2, 0,
    321 					     M_COPYALL, M_WAIT);
    322 					break;
    323 				    };
    324 				}
    325 				TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
    326 					nu_lru);
    327 				LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
    328 					nuidp, nu_hash);
    329 				kauth_cred_hold(nuidp->nu_cr);
    330 				nfsd->nfsd_nd->nd_cr = nuidp->nu_cr;
    331 				nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
    332 			    }
    333 			}
    334 		}
    335 		if ((SCARG(uap, flag) & NFSSVC_AUTHINFAIL) &&
    336 		    (nfsd = nsd->nsd_nfsd))
    337 			nfsd->nfsd_flag |= NFSD_AUTHFAIL;
    338 		error = nfssvc_nfsd(nsd, SCARG(uap, argp), l);
    339 #endif /* !NFSSERVER */
    340 	}
    341 	if (error == EINTR || error == ERESTART)
    342 		error = 0;
    343 	return (error);
    344 }
    345 
    346 #ifdef NFSSERVER
    347 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
    348 
    349 static struct nfssvc_sock *
    350 nfsrv_sockalloc()
    351 {
    352 	struct nfssvc_sock *slp;
    353 
    354 	slp = kmem_alloc(sizeof(*slp), KM_SLEEP);
    355 	memset(slp, 0, sizeof (struct nfssvc_sock));
    356 	mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_SOFTNET);
    357 	mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_SOFTNET);
    358 	cv_init(&slp->ns_cv, "nfsdsock");
    359 	TAILQ_INIT(&slp->ns_uidlruhead);
    360 	LIST_INIT(&slp->ns_tq);
    361 	SIMPLEQ_INIT(&slp->ns_sendq);
    362 	mutex_enter(&nfsd_lock);
    363 	TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
    364 	mutex_exit(&nfsd_lock);
    365 
    366 	return slp;
    367 }
    368 
    369 static void
    370 nfsrv_sockfree(struct nfssvc_sock *slp)
    371 {
    372 
    373 	KASSERT(slp->ns_so == NULL);
    374 	KASSERT(slp->ns_fp == NULL);
    375 	KASSERT((slp->ns_flags & SLP_VALID) == 0);
    376 	mutex_destroy(&slp->ns_lock);
    377 	mutex_destroy(&slp->ns_alock);
    378 	cv_destroy(&slp->ns_cv);
    379 	kmem_free(slp, sizeof(*slp));
    380 }
    381 
    382 /*
    383  * Adds a socket to the list for servicing by nfsds.
    384  */
    385 int
    386 nfssvc_addsock(fp, mynam)
    387 	file_t *fp;
    388 	struct mbuf *mynam;
    389 {
    390 	struct mbuf *m;
    391 	int siz;
    392 	struct nfssvc_sock *slp;
    393 	struct socket *so;
    394 	struct nfssvc_sock *tslp;
    395 	int error;
    396 
    397 	so = (struct socket *)fp->f_data;
    398 	tslp = (struct nfssvc_sock *)0;
    399 	/*
    400 	 * Add it to the list, as required.
    401 	 */
    402 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
    403 #ifdef INET6
    404 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
    405 			tslp = nfs_udp6sock;
    406 		else
    407 #endif
    408 		tslp = nfs_udpsock;
    409 		if (tslp->ns_flags & SLP_VALID) {
    410 			m_freem(mynam);
    411 			return (EPERM);
    412 		}
    413 #ifdef ISO
    414 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
    415 		tslp = nfs_cltpsock;
    416 		if (tslp->ns_flags & SLP_VALID) {
    417 			m_freem(mynam);
    418 			return (EPERM);
    419 		}
    420 #endif /* ISO */
    421 	}
    422 	if (so->so_type == SOCK_STREAM)
    423 		siz = NFS_MAXPACKET + sizeof (u_long);
    424 	else
    425 		siz = NFS_MAXPACKET;
    426 	solock(so);
    427 	error = soreserve(so, siz, siz);
    428 	sounlock(so);
    429 	if (error) {
    430 		m_freem(mynam);
    431 		return (error);
    432 	}
    433 
    434 	/*
    435 	 * Set protocol specific options { for now TCP only } and
    436 	 * reserve some space. For datagram sockets, this can get called
    437 	 * repeatedly for the same socket, but that isn't harmful.
    438 	 */
    439 	if (so->so_type == SOCK_STREAM) {
    440 		m = m_get(M_WAIT, MT_SOOPTS);
    441 		MCLAIM(m, &nfs_mowner);
    442 		*mtod(m, int32_t *) = 1;
    443 		m->m_len = sizeof(int32_t);
    444 		sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
    445 	}
    446 	if ((so->so_proto->pr_domain->dom_family == AF_INET
    447 #ifdef INET6
    448 	    || so->so_proto->pr_domain->dom_family == AF_INET6
    449 #endif
    450 	    ) &&
    451 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
    452 		m = m_get(M_WAIT, MT_SOOPTS);
    453 		MCLAIM(m, &nfs_mowner);
    454 		*mtod(m, int32_t *) = 1;
    455 		m->m_len = sizeof(int32_t);
    456 		sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
    457 	}
    458 	solock(so);
    459 	so->so_rcv.sb_flags &= ~SB_NOINTR;
    460 	so->so_rcv.sb_timeo = 0;
    461 	so->so_snd.sb_flags &= ~SB_NOINTR;
    462 	so->so_snd.sb_timeo = 0;
    463 	sounlock(so);
    464 	if (tslp) {
    465 		slp = tslp;
    466 	} else {
    467 		slp = nfsrv_sockalloc();
    468 	}
    469 	slp->ns_so = so;
    470 	slp->ns_nam = mynam;
    471 	mutex_enter(&fp->f_lock);
    472 	fp->f_count++;
    473 	mutex_exit(&fp->f_lock);
    474 	slp->ns_fp = fp;
    475 	slp->ns_flags = SLP_VALID;
    476 	slp->ns_aflags = SLP_A_NEEDQ;
    477 	slp->ns_gflags = 0;
    478 	slp->ns_sflags = 0;
    479 	solock(so);
    480 	so->so_upcallarg = (void *)slp;
    481 	so->so_upcall = nfsrv_soupcall;
    482 	so->so_rcv.sb_flags |= SB_UPCALL;
    483 	sounlock(so);
    484 	nfsrv_wakenfsd(slp);
    485 	return (0);
    486 }
    487 
    488 /*
    489  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    490  * until it is killed by a signal.
    491  */
    492 int
    493 nfssvc_nfsd(nsd, argp, l)
    494 	struct nfsd_srvargs *nsd;
    495 	void *argp;
    496 	struct lwp *l;
    497 {
    498 	struct timeval tv;
    499 	struct mbuf *m;
    500 	struct nfssvc_sock *slp;
    501 	struct nfsd *nfsd = nsd->nsd_nfsd;
    502 	struct nfsrv_descript *nd = NULL;
    503 	struct mbuf *mreq;
    504 	u_quad_t cur_usec;
    505 	int error = 0, cacherep, siz, sotype, writes_todo;
    506 	struct proc *p = l->l_proc;
    507 	int s;
    508 	bool doreinit;
    509 
    510 #ifndef nolint
    511 	cacherep = RC_DOIT;
    512 	writes_todo = 0;
    513 #endif
    514 	uvm_lwp_hold(l);
    515 	if (nfsd == NULL) {
    516 		nsd->nsd_nfsd = nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP);
    517 		memset(nfsd, 0, sizeof (struct nfsd));
    518 		cv_init(&nfsd->nfsd_cv, "nfsd");
    519 		nfsd->nfsd_procp = p;
    520 		mutex_enter(&nfsd_lock);
    521 		while ((nfssvc_sockhead_flag & SLP_INIT) != 0) {
    522 			KASSERT(nfs_numnfsd == 0);
    523 			cv_wait(&nfsd_initcv, &nfsd_lock);
    524 		}
    525 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
    526 		nfs_numnfsd++;
    527 		mutex_exit(&nfsd_lock);
    528 	}
    529 	/*
    530 	 * Loop getting rpc requests until SIGKILL.
    531 	 */
    532 	for (;;) {
    533 		bool dummy;
    534 
    535 		if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    536 		    != 0) {
    537 			preempt();
    538 		}
    539 		if (nfsd->nfsd_slp == NULL) {
    540 			mutex_enter(&nfsd_lock);
    541 			while (nfsd->nfsd_slp == NULL &&
    542 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
    543 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
    544 				    nfsd_idle);
    545 				error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock);
    546 				if (error) {
    547 					slp = nfsd->nfsd_slp;
    548 					nfsd->nfsd_slp = NULL;
    549 					if (!slp)
    550 						SLIST_REMOVE(&nfsd_idle_head,
    551 						    nfsd, nfsd, nfsd_idle);
    552 					mutex_exit(&nfsd_lock);
    553 					if (slp) {
    554 						nfsrv_wakenfsd(slp);
    555 						nfsrv_slpderef(slp);
    556 					}
    557 					goto done;
    558 				}
    559 			}
    560 			if (nfsd->nfsd_slp == NULL &&
    561 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
    562 				slp = TAILQ_FIRST(&nfssvc_sockpending);
    563 				if (slp) {
    564 					KASSERT((slp->ns_gflags & SLP_G_DOREC)
    565 					    != 0);
    566 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
    567 					    ns_pending);
    568 					slp->ns_gflags &= ~SLP_G_DOREC;
    569 					slp->ns_sref++;
    570 					nfsd->nfsd_slp = slp;
    571 				} else
    572 					nfsd_head_flag &= ~NFSD_CHECKSLP;
    573 			}
    574 			KASSERT(nfsd->nfsd_slp == NULL ||
    575 			    nfsd->nfsd_slp->ns_sref > 0);
    576 			mutex_exit(&nfsd_lock);
    577 			if ((slp = nfsd->nfsd_slp) == NULL)
    578 				continue;
    579 			if (slp->ns_flags & SLP_VALID) {
    580 				bool more;
    581 
    582 				if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) {
    583 					nfsrv_rcv(slp);
    584 				}
    585 				if (nfsdsock_testbits(slp, SLP_A_DISCONN)) {
    586 					nfsrv_zapsock(slp);
    587 				}
    588 				error = nfsrv_dorec(slp, nfsd, &nd, &more);
    589 				getmicrotime(&tv);
    590 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    591 					(u_quad_t)tv.tv_usec;
    592 				writes_todo = 0;
    593 				if (error) {
    594 					struct nfsrv_descript *nd2;
    595 
    596 					mutex_enter(&nfsd_lock);
    597 					nd2 = LIST_FIRST(&slp->ns_tq);
    598 					if (nd2 != NULL &&
    599 					    nd2->nd_time <= cur_usec) {
    600 						error = 0;
    601 						cacherep = RC_DOIT;
    602 						writes_todo = 1;
    603 					}
    604 					mutex_exit(&nfsd_lock);
    605 				}
    606 				if (error == 0 && more) {
    607 					nfsrv_wakenfsd(slp);
    608 				}
    609 			}
    610 		} else {
    611 			error = 0;
    612 			slp = nfsd->nfsd_slp;
    613 		}
    614 		KASSERT(slp != NULL);
    615 		KASSERT(nfsd->nfsd_slp == slp);
    616 		if (error || (slp->ns_flags & SLP_VALID) == 0) {
    617 			if (nd) {
    618 				nfsdreq_free(nd);
    619 				nd = NULL;
    620 			}
    621 			nfsd->nfsd_slp = NULL;
    622 			nfsrv_slpderef(slp);
    623 			continue;
    624 		}
    625 		sotype = slp->ns_so->so_type;
    626 		if (nd) {
    627 			getmicrotime(&nd->nd_starttime);
    628 			if (nd->nd_nam2)
    629 				nd->nd_nam = nd->nd_nam2;
    630 			else
    631 				nd->nd_nam = slp->ns_nam;
    632 
    633 			/*
    634 			 * Check to see if authorization is needed.
    635 			 */
    636 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
    637 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
    638 				nsd->nsd_haddr = mtod(nd->nd_nam,
    639 				    struct sockaddr_in *)->sin_addr.s_addr;
    640 				nsd->nsd_authlen = nfsd->nfsd_authlen;
    641 				nsd->nsd_verflen = nfsd->nfsd_verflen;
    642 				if (!copyout(nfsd->nfsd_authstr,
    643 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
    644 				    !copyout(nfsd->nfsd_verfstr,
    645 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
    646 				    !copyout(nsd, argp, sizeof (*nsd))) {
    647 					uvm_lwp_rele(l);
    648 					return (ENEEDAUTH);
    649 				}
    650 				cacherep = RC_DROPIT;
    651 			} else
    652 				cacherep = nfsrv_getcache(nd, slp, &mreq);
    653 
    654 			if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
    655 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
    656 				nd->nd_procnum = NFSPROC_NOOP;
    657 				nd->nd_repstat =
    658 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
    659 				cacherep = RC_DOIT;
    660 			}
    661 		}
    662 
    663 		/*
    664 		 * Loop to get all the write rpc relies that have been
    665 		 * gathered together.
    666 		 */
    667 		do {
    668 			switch (cacherep) {
    669 			case RC_DOIT:
    670 				mreq = NULL;
    671 				netexport_rdlock();
    672 				if (writes_todo || nd == NULL ||
    673 				     (!(nd->nd_flag & ND_NFSV3) &&
    674 				     nd->nd_procnum == NFSPROC_WRITE &&
    675 				     nfsrvw_procrastinate > 0))
    676 					error = nfsrv_writegather(&nd, slp,
    677 					    l, &mreq);
    678 				else
    679 					error =
    680 					    (*(nfsrv3_procs[nd->nd_procnum]))
    681 					    (nd, slp, l, &mreq);
    682 				netexport_rdunlock();
    683 				if (mreq == NULL) {
    684 					if (nd != NULL) {
    685 						if (nd->nd_nam2)
    686 							m_free(nd->nd_nam2);
    687 						if (nd->nd_mrep)
    688 							m_freem(nd->nd_mrep);
    689 					}
    690 					break;
    691 				}
    692 				if (error) {
    693 					nfsstats.srv_errs++;
    694 					nfsrv_updatecache(nd, false, mreq);
    695 					if (nd->nd_nam2)
    696 						m_freem(nd->nd_nam2);
    697 					break;
    698 				}
    699 				nfsstats.srvrpccnt[nd->nd_procnum]++;
    700 				nfsrv_updatecache(nd, true, mreq);
    701 				nd->nd_mrep = (struct mbuf *)0;
    702 			case RC_REPLY:
    703 				m = mreq;
    704 				siz = 0;
    705 				while (m) {
    706 					siz += m->m_len;
    707 					m = m->m_next;
    708 				}
    709 				if (siz <= 0 || siz > NFS_MAXPACKET) {
    710 					printf("mbuf siz=%d\n",siz);
    711 					panic("Bad nfs svc reply");
    712 				}
    713 				m = mreq;
    714 				m->m_pkthdr.len = siz;
    715 				m->m_pkthdr.rcvif = (struct ifnet *)0;
    716 				/*
    717 				 * For stream protocols, prepend a Sun RPC
    718 				 * Record Mark.
    719 				 */
    720 				if (sotype == SOCK_STREAM) {
    721 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
    722 					*mtod(m, u_int32_t *) =
    723 					    htonl(0x80000000 | siz);
    724 				}
    725 				nd->nd_mreq = m;
    726 				if (nfsrtton) {
    727 					nfsd_rt(slp->ns_so->so_type, nd,
    728 					    cacherep);
    729 				}
    730 				error = nfsdsock_sendreply(slp, nd);
    731 				nd = NULL;
    732 				if (error == EPIPE)
    733 					nfsrv_zapsock(slp);
    734 				if (error == EINTR || error == ERESTART) {
    735 					nfsd->nfsd_slp = NULL;
    736 					nfsrv_slpderef(slp);
    737 					goto done;
    738 				}
    739 				break;
    740 			case RC_DROPIT:
    741 				if (nfsrtton)
    742 					nfsd_rt(sotype, nd, cacherep);
    743 				m_freem(nd->nd_mrep);
    744 				m_freem(nd->nd_nam2);
    745 				break;
    746 			}
    747 			if (nd) {
    748 				nfsdreq_free(nd);
    749 				nd = NULL;
    750 			}
    751 
    752 			/*
    753 			 * Check to see if there are outstanding writes that
    754 			 * need to be serviced.
    755 			 */
    756 			getmicrotime(&tv);
    757 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    758 			    (u_quad_t)tv.tv_usec;
    759 			s = splsoftclock();
    760 			if (LIST_FIRST(&slp->ns_tq) &&
    761 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
    762 				cacherep = RC_DOIT;
    763 				writes_todo = 1;
    764 			} else
    765 				writes_todo = 0;
    766 			splx(s);
    767 		} while (writes_todo);
    768 		if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) {
    769 			nfsd->nfsd_slp = NULL;
    770 			nfsrv_slpderef(slp);
    771 		}
    772 	}
    773 done:
    774 	mutex_enter(&nfsd_lock);
    775 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
    776 	doreinit = --nfs_numnfsd == 0;
    777 	if (doreinit)
    778 		nfssvc_sockhead_flag |= SLP_INIT;
    779 	mutex_exit(&nfsd_lock);
    780 	cv_destroy(&nfsd->nfsd_cv);
    781 	kmem_free(nfsd, sizeof(*nfsd));
    782 	nsd->nsd_nfsd = NULL;
    783 	if (doreinit)
    784 		nfsrv_init(true);	/* Reinitialize everything */
    785 	uvm_lwp_rele(l);
    786 	return (error);
    787 }
    788 
    789 /*
    790  * Shut down a socket associated with an nfssvc_sock structure.
    791  * Should be called with the send lock set, if required.
    792  * The trick here is to increment the sref at the start, so that the nfsds
    793  * will stop using it and clear ns_flag at the end so that it will not be
    794  * reassigned during cleanup.
    795  *
    796  * called at splsoftnet.
    797  */
    798 void
    799 nfsrv_zapsock(slp)
    800 	struct nfssvc_sock *slp;
    801 {
    802 	struct nfsuid *nuidp, *nnuidp;
    803 	struct nfsrv_descript *nwp;
    804 	struct socket *so;
    805 	struct mbuf *m;
    806 
    807 	if (nfsdsock_drain(slp)) {
    808 		return;
    809 	}
    810 	mutex_enter(&nfsd_lock);
    811 	if (slp->ns_gflags & SLP_G_DOREC) {
    812 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
    813 		slp->ns_gflags &= ~SLP_G_DOREC;
    814 	}
    815 	mutex_exit(&nfsd_lock);
    816 
    817 	so = slp->ns_so;
    818 	KASSERT(so != NULL);
    819 	solock(so);
    820 	so->so_upcall = NULL;
    821 	so->so_upcallarg = NULL;
    822 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    823 	soshutdown(so, SHUT_RDWR);
    824 	sounlock(so);
    825 
    826 	if (slp->ns_nam)
    827 		m_free(slp->ns_nam);
    828 	m_freem(slp->ns_raw);
    829 	m = slp->ns_rec;
    830 	while (m != NULL) {
    831 		struct mbuf *n;
    832 
    833 		n = m->m_nextpkt;
    834 		m_freem(m);
    835 		m = n;
    836 	}
    837 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
    838 	    nuidp = nnuidp) {
    839 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
    840 		LIST_REMOVE(nuidp, nu_hash);
    841 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
    842 		if (nuidp->nu_flag & NU_NAM)
    843 			m_freem(nuidp->nu_nam);
    844 		kmem_free(nuidp, sizeof(*nuidp));
    845 	}
    846 	mutex_enter(&nfsd_lock);
    847 	while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) {
    848 		LIST_REMOVE(nwp, nd_tq);
    849 		mutex_exit(&nfsd_lock);
    850 		nfsdreq_free(nwp);
    851 		mutex_enter(&nfsd_lock);
    852 	}
    853 	mutex_exit(&nfsd_lock);
    854 }
    855 
    856 /*
    857  * Derefence a server socket structure. If it has no more references and
    858  * is no longer valid, you can throw it away.
    859  */
    860 void
    861 nfsrv_slpderef(slp)
    862 	struct nfssvc_sock *slp;
    863 {
    864 	uint32_t ref;
    865 
    866 	mutex_enter(&nfsd_lock);
    867 	KASSERT(slp->ns_sref > 0);
    868 	ref = --slp->ns_sref;
    869 	mutex_exit(&nfsd_lock);
    870 	if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) {
    871 		file_t *fp;
    872 
    873 		mutex_enter(&nfsd_lock);
    874 		KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0);
    875 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
    876 		mutex_exit(&nfsd_lock);
    877 
    878 		fp = slp->ns_fp;
    879 		if (fp != NULL) {
    880 			slp->ns_fp = NULL;
    881 			KASSERT(fp != NULL);
    882 			KASSERT(fp->f_data == slp->ns_so);
    883 			KASSERT(fp->f_count > 0);
    884 			closef(fp);
    885 			slp->ns_so = NULL;
    886 		}
    887 
    888 		nfsrv_sockfree(slp);
    889 	}
    890 }
    891 
    892 /*
    893  * Initialize the data structures for the server.
    894  * Handshake with any new nfsds starting up to avoid any chance of
    895  * corruption.
    896  */
    897 void
    898 nfsrv_init(terminating)
    899 	int terminating;
    900 {
    901 	struct nfssvc_sock *slp;
    902 
    903 	if (!terminating) {
    904 		mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_SOFTNET);
    905 		cv_init(&nfsd_initcv, "nfsdinit");
    906 	}
    907 
    908 	mutex_enter(&nfsd_lock);
    909 	if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0)
    910 		panic("nfsd init");
    911 	nfssvc_sockhead_flag |= SLP_INIT;
    912 
    913 	if (terminating) {
    914 		KASSERT(SLIST_EMPTY(&nfsd_idle_head));
    915 		KASSERT(TAILQ_EMPTY(&nfsd_head));
    916 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
    917 			mutex_exit(&nfsd_lock);
    918 			KASSERT(slp->ns_sref == 0);
    919 			slp->ns_sref++;
    920 			nfsrv_zapsock(slp);
    921 			nfsrv_slpderef(slp);
    922 			mutex_enter(&nfsd_lock);
    923 		}
    924 		KASSERT(TAILQ_EMPTY(&nfssvc_sockpending));
    925 		mutex_exit(&nfsd_lock);
    926 		nfsrv_cleancache();	/* And clear out server cache */
    927 	} else {
    928 		mutex_exit(&nfsd_lock);
    929 		nfs_pub.np_valid = 0;
    930 	}
    931 
    932 	TAILQ_INIT(&nfssvc_sockhead);
    933 	TAILQ_INIT(&nfssvc_sockpending);
    934 
    935 	TAILQ_INIT(&nfsd_head);
    936 	SLIST_INIT(&nfsd_idle_head);
    937 	nfsd_head_flag &= ~NFSD_CHECKSLP;
    938 
    939 	nfs_udpsock = nfsrv_sockalloc();
    940 
    941 #ifdef INET6
    942 	nfs_udp6sock = nfsrv_sockalloc();
    943 #endif
    944 
    945 #ifdef ISO
    946 	nfs_cltpsock = nfsrv_sockalloc();
    947 #endif
    948 
    949 	mutex_enter(&nfsd_lock);
    950 	nfssvc_sockhead_flag &= ~SLP_INIT;
    951 	cv_broadcast(&nfsd_initcv);
    952 	mutex_exit(&nfsd_lock);
    953 }
    954 
    955 /*
    956  * Add entries to the server monitor log.
    957  */
    958 static void
    959 nfsd_rt(sotype, nd, cacherep)
    960 	int sotype;
    961 	struct nfsrv_descript *nd;
    962 	int cacherep;
    963 {
    964 	struct timeval tv;
    965 	struct drt *rt;
    966 
    967 	rt = &nfsdrt.drt[nfsdrt.pos];
    968 	if (cacherep == RC_DOIT)
    969 		rt->flag = 0;
    970 	else if (cacherep == RC_REPLY)
    971 		rt->flag = DRT_CACHEREPLY;
    972 	else
    973 		rt->flag = DRT_CACHEDROP;
    974 	if (sotype == SOCK_STREAM)
    975 		rt->flag |= DRT_TCP;
    976 	if (nd->nd_flag & ND_NFSV3)
    977 		rt->flag |= DRT_NFSV3;
    978 	rt->proc = nd->nd_procnum;
    979 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
    980 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
    981 	else
    982 	    rt->ipadr = INADDR_ANY;
    983 	getmicrotime(&tv);
    984 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
    985 		(tv.tv_usec - nd->nd_starttime.tv_usec);
    986 	rt->tstamp = tv;
    987 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
    988 }
    989 #endif /* NFSSERVER */
    990 
    991 #ifdef NFS
    992 
    993 int nfs_defect = 0;
    994 /*
    995  * Asynchronous I/O threads for client nfs.
    996  * They do read-ahead and write-behind operations on the block I/O cache.
    997  * Never returns unless it fails or gets killed.
    998  */
    999 
   1000 static void
   1001 nfssvc_iod(void *arg)
   1002 {
   1003 	struct buf *bp;
   1004 	struct nfs_iod *myiod;
   1005 	struct nfsmount *nmp;
   1006 
   1007 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
   1008 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
   1009 	cv_init(&myiod->nid_cv, "nfsiod");
   1010 	myiod->nid_exiting = false;
   1011 	myiod->nid_mount = NULL;
   1012 	mutex_enter(&nfs_iodlist_lock);
   1013 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
   1014 	mutex_exit(&nfs_iodlist_lock);
   1015 
   1016 	for (;;) {
   1017 		mutex_enter(&nfs_iodlist_lock);
   1018 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
   1019 		mutex_exit(&nfs_iodlist_lock);
   1020 
   1021 		mutex_enter(&myiod->nid_lock);
   1022 		while (/*CONSTCOND*/ true) {
   1023 			nmp = myiod->nid_mount;
   1024 			if (nmp) {
   1025 				myiod->nid_mount = NULL;
   1026 				break;
   1027 			}
   1028 			if (__predict_false(myiod->nid_exiting)) {
   1029 				/*
   1030 				 * drop nid_lock to preserve locking order.
   1031 				 */
   1032 				mutex_exit(&myiod->nid_lock);
   1033 				mutex_enter(&nfs_iodlist_lock);
   1034 				mutex_enter(&myiod->nid_lock);
   1035 				/*
   1036 				 * recheck nid_mount because nfs_asyncio can
   1037 				 * pick us in the meantime as we are still on
   1038 				 * nfs_iodlist_lock.
   1039 				 */
   1040 				if (myiod->nid_mount != NULL) {
   1041 					mutex_exit(&nfs_iodlist_lock);
   1042 					continue;
   1043 				}
   1044 				LIST_REMOVE(myiod, nid_idle);
   1045 				mutex_exit(&nfs_iodlist_lock);
   1046 				goto quit;
   1047 			}
   1048 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
   1049 		}
   1050 		mutex_exit(&myiod->nid_lock);
   1051 
   1052 		mutex_enter(&nmp->nm_lock);
   1053 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
   1054 			/* Take one off the front of the list */
   1055 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
   1056 			nmp->nm_bufqlen--;
   1057 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
   1058 				cv_broadcast(&nmp->nm_aiocv);
   1059 			}
   1060 			mutex_exit(&nmp->nm_lock);
   1061 			KERNEL_LOCK(1, curlwp);
   1062 			(void)nfs_doio(bp);
   1063 			KERNEL_UNLOCK_LAST(curlwp);
   1064 			mutex_enter(&nmp->nm_lock);
   1065 			/*
   1066 			 * If there are more than one iod on this mount,
   1067 			 * then defect so that the iods can be shared out
   1068 			 * fairly between the mounts
   1069 			 */
   1070 			if (nfs_defect && nmp->nm_bufqiods > 1) {
   1071 				break;
   1072 			}
   1073 		}
   1074 		KASSERT(nmp->nm_bufqiods > 0);
   1075 		nmp->nm_bufqiods--;
   1076 		mutex_exit(&nmp->nm_lock);
   1077 	}
   1078 quit:
   1079 	KASSERT(myiod->nid_mount == NULL);
   1080 	mutex_exit(&myiod->nid_lock);
   1081 
   1082 	cv_destroy(&myiod->nid_cv);
   1083 	mutex_destroy(&myiod->nid_lock);
   1084 	kmem_free(myiod, sizeof(*myiod));
   1085 
   1086 	kthread_exit(0);
   1087 }
   1088 
   1089 void
   1090 nfs_iodinit()
   1091 {
   1092 
   1093 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
   1094 	LIST_INIT(&nfs_iodlist_all);
   1095 	LIST_INIT(&nfs_iodlist_idle);
   1096 }
   1097 
   1098 int
   1099 nfs_set_niothreads(int newval)
   1100 {
   1101 	struct nfs_iod *nid;
   1102 	int error = 0;
   1103         int hold_count;
   1104 
   1105 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
   1106 
   1107 	mutex_enter(&nfs_iodlist_lock);
   1108 	/* clamp to sane range */
   1109 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
   1110 
   1111 	while (nfs_numasync != nfs_niothreads && error == 0) {
   1112 		while (nfs_numasync < nfs_niothreads) {
   1113 
   1114 			/*
   1115 			 * kthread_create can wait for pagedaemon and
   1116 			 * pagedaemon can wait for nfsiod which needs to acquire
   1117 			 * nfs_iodlist_lock.
   1118 			 */
   1119 
   1120 			mutex_exit(&nfs_iodlist_lock);
   1121 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
   1122 			    nfssvc_iod, NULL, NULL, "nfsio");
   1123 			mutex_enter(&nfs_iodlist_lock);
   1124 			if (error) {
   1125 				/* give up */
   1126 				nfs_niothreads = nfs_numasync;
   1127 				break;
   1128 			}
   1129 			nfs_numasync++;
   1130 		}
   1131 		while (nfs_numasync > nfs_niothreads) {
   1132 			nid = LIST_FIRST(&nfs_iodlist_all);
   1133 			if (nid == NULL) {
   1134 				/* iod has not started yet. */
   1135 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
   1136 				continue;
   1137 			}
   1138 			LIST_REMOVE(nid, nid_all);
   1139 			mutex_enter(&nid->nid_lock);
   1140 			KASSERT(!nid->nid_exiting);
   1141 			nid->nid_exiting = true;
   1142 			cv_signal(&nid->nid_cv);
   1143 			mutex_exit(&nid->nid_lock);
   1144 			nfs_numasync--;
   1145 		}
   1146 	}
   1147 	mutex_exit(&nfs_iodlist_lock);
   1148 
   1149 	KERNEL_LOCK(hold_count, curlwp);
   1150 	return error;
   1151 }
   1152 
   1153 /*
   1154  * Get an authorization string for the uid by having the mount_nfs sitting
   1155  * on this mount point porpous out of the kernel and do it.
   1156  */
   1157 int
   1158 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
   1159 	struct nfsmount *nmp;
   1160 	struct nfsreq *rep;
   1161 	kauth_cred_t cred;
   1162 	char **auth_str;
   1163 	int *auth_len;
   1164 	char *verf_str;
   1165 	int *verf_len;
   1166 	NFSKERBKEY_T key;		/* return session key */
   1167 {
   1168 	int error = 0;
   1169 
   1170 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
   1171 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
   1172 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
   1173 			"nfsauth1", 2 * hz);
   1174 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1175 		if (error) {
   1176 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1177 			return (error);
   1178 		}
   1179 	}
   1180 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
   1181 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
   1182 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
   1183 	nmp->nm_verfstr = verf_str;
   1184 	nmp->nm_verflen = *verf_len;
   1185 	nmp->nm_authuid = kauth_cred_geteuid(cred);
   1186 	wakeup((void *)&nmp->nm_authstr);
   1187 
   1188 	/*
   1189 	 * And wait for mount_nfs to do its stuff.
   1190 	 */
   1191 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
   1192 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
   1193 			"nfsauth2", 2 * hz);
   1194 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1195 	}
   1196 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
   1197 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
   1198 		error = EAUTH;
   1199 	}
   1200 	if (error)
   1201 		free((void *)*auth_str, M_TEMP);
   1202 	else {
   1203 		*auth_len = nmp->nm_authlen;
   1204 		*verf_len = nmp->nm_verflen;
   1205 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
   1206 	}
   1207 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
   1208 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
   1209 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
   1210 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1211 		wakeup((void *)&nmp->nm_authtype);
   1212 	}
   1213 	return (error);
   1214 }
   1215 
   1216 /*
   1217  * Get a nickname authenticator and verifier.
   1218  */
   1219 int
   1220 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
   1221     int *auth_len, char *verf_str, int verf_len)
   1222 {
   1223 	struct timeval ktvin, ktvout, tv;
   1224 	struct nfsuid *nuidp;
   1225 	u_int32_t *nickp, *verfp;
   1226 
   1227 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
   1228 
   1229 #ifdef DIAGNOSTIC
   1230 	if (verf_len < (4 * NFSX_UNSIGNED))
   1231 		panic("nfs_getnickauth verf too small");
   1232 #endif
   1233 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
   1234 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
   1235 			break;
   1236 	}
   1237 	if (!nuidp || nuidp->nu_expire < time_second)
   1238 		return (EACCES);
   1239 
   1240 	/*
   1241 	 * Move to the end of the lru list (end of lru == most recently used).
   1242 	 */
   1243 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1244 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1245 
   1246 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
   1247 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1248 	*nickp = txdr_unsigned(nuidp->nu_nickname);
   1249 	*auth_str = (char *)nickp;
   1250 	*auth_len = 2 * NFSX_UNSIGNED;
   1251 
   1252 	/*
   1253 	 * Now we must encrypt the verifier and package it up.
   1254 	 */
   1255 	verfp = (u_int32_t *)verf_str;
   1256 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1257 	getmicrotime(&tv);
   1258 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
   1259 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
   1260 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
   1261 		nuidp->nu_timestamp = tv;
   1262 	else
   1263 		nuidp->nu_timestamp.tv_usec++;
   1264 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
   1265 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
   1266 
   1267 	/*
   1268 	 * Now encrypt the timestamp verifier in ecb mode using the session
   1269 	 * key.
   1270 	 */
   1271 #ifdef NFSKERB
   1272 	XXX
   1273 #endif
   1274 
   1275 	*verfp++ = ktvout.tv_sec;
   1276 	*verfp++ = ktvout.tv_usec;
   1277 	*verfp = 0;
   1278 	return (0);
   1279 }
   1280 
   1281 /*
   1282  * Save the current nickname in a hash list entry on the mount point.
   1283  */
   1284 int
   1285 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
   1286 	struct nfsmount *nmp;
   1287 	kauth_cred_t cred;
   1288 	int len;
   1289 	NFSKERBKEY_T key;
   1290 	struct mbuf **mdp;
   1291 	char **dposp;
   1292 	struct mbuf *mrep;
   1293 {
   1294 	struct nfsuid *nuidp;
   1295 	u_int32_t *tl;
   1296 	int32_t t1;
   1297 	struct mbuf *md = *mdp;
   1298 	struct timeval ktvin, ktvout;
   1299 	u_int32_t nick;
   1300 	char *dpos = *dposp, *cp2;
   1301 	int deltasec, error = 0;
   1302 
   1303 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
   1304 
   1305 	if (len == (3 * NFSX_UNSIGNED)) {
   1306 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
   1307 		ktvin.tv_sec = *tl++;
   1308 		ktvin.tv_usec = *tl++;
   1309 		nick = fxdr_unsigned(u_int32_t, *tl);
   1310 
   1311 		/*
   1312 		 * Decrypt the timestamp in ecb mode.
   1313 		 */
   1314 #ifdef NFSKERB
   1315 		XXX
   1316 #endif
   1317 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
   1318 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
   1319 		deltasec = time_second - ktvout.tv_sec;
   1320 		if (deltasec < 0)
   1321 			deltasec = -deltasec;
   1322 		/*
   1323 		 * If ok, add it to the hash list for the mount point.
   1324 		 */
   1325 		if (deltasec <= NFS_KERBCLOCKSKEW) {
   1326 			if (nmp->nm_numuids < nuidhash_max) {
   1327 				nmp->nm_numuids++;
   1328 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
   1329 			} else {
   1330 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
   1331 				LIST_REMOVE(nuidp, nu_hash);
   1332 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
   1333 					nu_lru);
   1334 			}
   1335 			nuidp->nu_flag = 0;
   1336 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
   1337 			nuidp->nu_expire = time_second + NFS_KERBTTL;
   1338 			nuidp->nu_timestamp = ktvout;
   1339 			nuidp->nu_nickname = nick;
   1340 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
   1341 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
   1342 				nu_lru);
   1343 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
   1344 				nuidp, nu_hash);
   1345 		}
   1346 	} else
   1347 		nfsm_adv(nfsm_rndup(len));
   1348 nfsmout:
   1349 	*mdp = md;
   1350 	*dposp = dpos;
   1351 	return (error);
   1352 }
   1353 #endif /* NFS */
   1354