Home | History | Annotate | Line # | Download | only in nfs
nfs_syscalls.c revision 1.131
      1 /*	$NetBSD: nfs_syscalls.c,v 1.131 2008/02/28 17:07:49 elad 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 1.131 2008/02/28 17:07:49 elad 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 	struct file *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(l->l_proc->p_fd, 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 				FILE_UNUSE(fp, NULL);
    214 				return (error);
    215 			}
    216 		}
    217 		error = nfssvc_addsock(fp, nam);
    218 		FILE_UNUSE(fp, NULL);
    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 	/* XXX could be IPL_SOFTNET */
    357 	mutex_init(&slp->ns_lock, MUTEX_DRIVER, IPL_VM);
    358 	mutex_init(&slp->ns_alock, MUTEX_DRIVER, IPL_VM);
    359 	cv_init(&slp->ns_cv, "nfsdsock");
    360 	TAILQ_INIT(&slp->ns_uidlruhead);
    361 	LIST_INIT(&slp->ns_tq);
    362 	SIMPLEQ_INIT(&slp->ns_sendq);
    363 	mutex_enter(&nfsd_lock);
    364 	TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
    365 	mutex_exit(&nfsd_lock);
    366 
    367 	return slp;
    368 }
    369 
    370 static void
    371 nfsrv_sockfree(struct nfssvc_sock *slp)
    372 {
    373 
    374 	KASSERT(slp->ns_so == NULL);
    375 	KASSERT(slp->ns_fp == NULL);
    376 	KASSERT((slp->ns_flags & SLP_VALID) == 0);
    377 	mutex_destroy(&slp->ns_lock);
    378 	mutex_destroy(&slp->ns_alock);
    379 	cv_destroy(&slp->ns_cv);
    380 	kmem_free(slp, sizeof(*slp));
    381 }
    382 
    383 /*
    384  * Adds a socket to the list for servicing by nfsds.
    385  */
    386 int
    387 nfssvc_addsock(fp, mynam)
    388 	struct file *fp;
    389 	struct mbuf *mynam;
    390 {
    391 	struct mbuf *m;
    392 	int siz;
    393 	struct nfssvc_sock *slp;
    394 	struct socket *so;
    395 	struct nfssvc_sock *tslp;
    396 	int error, s;
    397 
    398 	so = (struct socket *)fp->f_data;
    399 	tslp = (struct nfssvc_sock *)0;
    400 	/*
    401 	 * Add it to the list, as required.
    402 	 */
    403 	if (so->so_proto->pr_protocol == IPPROTO_UDP) {
    404 #ifdef INET6
    405 		if (so->so_proto->pr_domain->dom_family == AF_INET6)
    406 			tslp = nfs_udp6sock;
    407 		else
    408 #endif
    409 		tslp = nfs_udpsock;
    410 		if (tslp->ns_flags & SLP_VALID) {
    411 			m_freem(mynam);
    412 			return (EPERM);
    413 		}
    414 #ifdef ISO
    415 	} else if (so->so_proto->pr_protocol == ISOPROTO_CLTP) {
    416 		tslp = nfs_cltpsock;
    417 		if (tslp->ns_flags & SLP_VALID) {
    418 			m_freem(mynam);
    419 			return (EPERM);
    420 		}
    421 #endif /* ISO */
    422 	}
    423 	if (so->so_type == SOCK_STREAM)
    424 		siz = NFS_MAXPACKET + sizeof (u_long);
    425 	else
    426 		siz = NFS_MAXPACKET;
    427 	error = soreserve(so, siz, siz);
    428 	if (error) {
    429 		m_freem(mynam);
    430 		return (error);
    431 	}
    432 
    433 	/*
    434 	 * Set protocol specific options { for now TCP only } and
    435 	 * reserve some space. For datagram sockets, this can get called
    436 	 * repeatedly for the same socket, but that isn't harmful.
    437 	 */
    438 	if (so->so_type == SOCK_STREAM) {
    439 		m = m_get(M_WAIT, MT_SOOPTS);
    440 		MCLAIM(m, &nfs_mowner);
    441 		*mtod(m, int32_t *) = 1;
    442 		m->m_len = sizeof(int32_t);
    443 		sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
    444 	}
    445 	if ((so->so_proto->pr_domain->dom_family == AF_INET
    446 #ifdef INET6
    447 	    || so->so_proto->pr_domain->dom_family == AF_INET6
    448 #endif
    449 	    ) &&
    450 	    so->so_proto->pr_protocol == IPPROTO_TCP) {
    451 		m = m_get(M_WAIT, MT_SOOPTS);
    452 		MCLAIM(m, &nfs_mowner);
    453 		*mtod(m, int32_t *) = 1;
    454 		m->m_len = sizeof(int32_t);
    455 		sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
    456 	}
    457 	so->so_rcv.sb_flags &= ~SB_NOINTR;
    458 	so->so_rcv.sb_timeo = 0;
    459 	so->so_snd.sb_flags &= ~SB_NOINTR;
    460 	so->so_snd.sb_timeo = 0;
    461 	if (tslp) {
    462 		slp = tslp;
    463 	} else {
    464 		slp = nfsrv_sockalloc();
    465 	}
    466 	slp->ns_so = so;
    467 	slp->ns_nam = mynam;
    468 	mutex_enter(&fp->f_lock);
    469 	fp->f_count++;
    470 	mutex_exit(&fp->f_lock);
    471 	slp->ns_fp = fp;
    472 	slp->ns_flags = SLP_VALID;
    473 	slp->ns_aflags = SLP_A_NEEDQ;
    474 	slp->ns_gflags = 0;
    475 	slp->ns_sflags = 0;
    476 	KERNEL_LOCK(1, curlwp);
    477 	s = splsoftnet();
    478 	so->so_upcallarg = (void *)slp;
    479 	so->so_upcall = nfsrv_soupcall;
    480 	so->so_rcv.sb_flags |= SB_UPCALL;
    481 	splx(s);
    482 	KERNEL_UNLOCK_ONE(curlwp);
    483 	nfsrv_wakenfsd(slp);
    484 	return (0);
    485 }
    486 
    487 /*
    488  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
    489  * until it is killed by a signal.
    490  */
    491 int
    492 nfssvc_nfsd(nsd, argp, l)
    493 	struct nfsd_srvargs *nsd;
    494 	void *argp;
    495 	struct lwp *l;
    496 {
    497 	struct timeval tv;
    498 	struct mbuf *m;
    499 	struct nfssvc_sock *slp;
    500 	struct nfsd *nfsd = nsd->nsd_nfsd;
    501 	struct nfsrv_descript *nd = NULL;
    502 	struct mbuf *mreq;
    503 	u_quad_t cur_usec;
    504 	int error = 0, cacherep, siz, sotype, writes_todo;
    505 	struct proc *p = l->l_proc;
    506 	int s;
    507 	bool doreinit;
    508 
    509 #ifndef nolint
    510 	cacherep = RC_DOIT;
    511 	writes_todo = 0;
    512 #endif
    513 	uvm_lwp_hold(l);
    514 	if (nfsd == NULL) {
    515 		nsd->nsd_nfsd = nfsd = kmem_alloc(sizeof(*nfsd), KM_SLEEP);
    516 		memset(nfsd, 0, sizeof (struct nfsd));
    517 		cv_init(&nfsd->nfsd_cv, "nfsd");
    518 		nfsd->nfsd_procp = p;
    519 		mutex_enter(&nfsd_lock);
    520 		while ((nfssvc_sockhead_flag & SLP_INIT) != 0) {
    521 			KASSERT(nfs_numnfsd == 0);
    522 			cv_wait(&nfsd_initcv, &nfsd_lock);
    523 		}
    524 		TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
    525 		nfs_numnfsd++;
    526 		mutex_exit(&nfsd_lock);
    527 	}
    528 	/*
    529 	 * Loop getting rpc requests until SIGKILL.
    530 	 */
    531 	for (;;) {
    532 		bool dummy;
    533 
    534 		if ((curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    535 		    != 0) {
    536 			preempt();
    537 		}
    538 		if (nfsd->nfsd_slp == NULL) {
    539 			mutex_enter(&nfsd_lock);
    540 			while (nfsd->nfsd_slp == NULL &&
    541 			    (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
    542 				SLIST_INSERT_HEAD(&nfsd_idle_head, nfsd,
    543 				    nfsd_idle);
    544 				error = cv_wait_sig(&nfsd->nfsd_cv, &nfsd_lock);
    545 				if (error) {
    546 					slp = nfsd->nfsd_slp;
    547 					nfsd->nfsd_slp = NULL;
    548 					if (!slp)
    549 						SLIST_REMOVE(&nfsd_idle_head,
    550 						    nfsd, nfsd, nfsd_idle);
    551 					mutex_exit(&nfsd_lock);
    552 					if (slp) {
    553 						nfsrv_wakenfsd(slp);
    554 						nfsrv_slpderef(slp);
    555 					}
    556 					goto done;
    557 				}
    558 			}
    559 			if (nfsd->nfsd_slp == NULL &&
    560 			    (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
    561 				slp = TAILQ_FIRST(&nfssvc_sockpending);
    562 				if (slp) {
    563 					KASSERT((slp->ns_gflags & SLP_G_DOREC)
    564 					    != 0);
    565 					TAILQ_REMOVE(&nfssvc_sockpending, slp,
    566 					    ns_pending);
    567 					slp->ns_gflags &= ~SLP_G_DOREC;
    568 					slp->ns_sref++;
    569 					nfsd->nfsd_slp = slp;
    570 				} else
    571 					nfsd_head_flag &= ~NFSD_CHECKSLP;
    572 			}
    573 			KASSERT(nfsd->nfsd_slp == NULL ||
    574 			    nfsd->nfsd_slp->ns_sref > 0);
    575 			mutex_exit(&nfsd_lock);
    576 			if ((slp = nfsd->nfsd_slp) == NULL)
    577 				continue;
    578 			if (slp->ns_flags & SLP_VALID) {
    579 				bool more;
    580 
    581 				if (nfsdsock_testbits(slp, SLP_A_NEEDQ)) {
    582 					nfsrv_rcv(slp);
    583 				}
    584 				if (nfsdsock_testbits(slp, SLP_A_DISCONN)) {
    585 					nfsrv_zapsock(slp);
    586 				}
    587 				error = nfsrv_dorec(slp, nfsd, &nd, &more);
    588 				getmicrotime(&tv);
    589 				cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    590 					(u_quad_t)tv.tv_usec;
    591 				writes_todo = 0;
    592 				if (error) {
    593 					struct nfsrv_descript *nd2;
    594 
    595 					mutex_enter(&nfsd_lock);
    596 					nd2 = LIST_FIRST(&slp->ns_tq);
    597 					if (nd2 != NULL &&
    598 					    nd2->nd_time <= cur_usec) {
    599 						error = 0;
    600 						cacherep = RC_DOIT;
    601 						writes_todo = 1;
    602 					}
    603 					mutex_exit(&nfsd_lock);
    604 				}
    605 				if (error == 0 && more) {
    606 					nfsrv_wakenfsd(slp);
    607 				}
    608 			}
    609 		} else {
    610 			error = 0;
    611 			slp = nfsd->nfsd_slp;
    612 		}
    613 		KASSERT(slp != NULL);
    614 		KASSERT(nfsd->nfsd_slp == slp);
    615 		if (error || (slp->ns_flags & SLP_VALID) == 0) {
    616 			if (nd) {
    617 				nfsdreq_free(nd);
    618 				nd = NULL;
    619 			}
    620 			nfsd->nfsd_slp = NULL;
    621 			nfsrv_slpderef(slp);
    622 			continue;
    623 		}
    624 		sotype = slp->ns_so->so_type;
    625 		if (nd) {
    626 			getmicrotime(&nd->nd_starttime);
    627 			if (nd->nd_nam2)
    628 				nd->nd_nam = nd->nd_nam2;
    629 			else
    630 				nd->nd_nam = slp->ns_nam;
    631 
    632 			/*
    633 			 * Check to see if authorization is needed.
    634 			 */
    635 			if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
    636 				nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
    637 				nsd->nsd_haddr = mtod(nd->nd_nam,
    638 				    struct sockaddr_in *)->sin_addr.s_addr;
    639 				nsd->nsd_authlen = nfsd->nfsd_authlen;
    640 				nsd->nsd_verflen = nfsd->nfsd_verflen;
    641 				if (!copyout(nfsd->nfsd_authstr,
    642 				    nsd->nsd_authstr, nfsd->nfsd_authlen) &&
    643 				    !copyout(nfsd->nfsd_verfstr,
    644 				    nsd->nsd_verfstr, nfsd->nfsd_verflen) &&
    645 				    !copyout(nsd, argp, sizeof (*nsd))) {
    646 					uvm_lwp_rele(l);
    647 					return (ENEEDAUTH);
    648 				}
    649 				cacherep = RC_DROPIT;
    650 			} else
    651 				cacherep = nfsrv_getcache(nd, slp, &mreq);
    652 
    653 			if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
    654 				nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
    655 				nd->nd_procnum = NFSPROC_NOOP;
    656 				nd->nd_repstat =
    657 				    (NFSERR_AUTHERR | AUTH_TOOWEAK);
    658 				cacherep = RC_DOIT;
    659 			}
    660 		}
    661 
    662 		/*
    663 		 * Loop to get all the write rpc relies that have been
    664 		 * gathered together.
    665 		 */
    666 		do {
    667 #ifdef DIAGNOSTIC
    668 			int lockcount;
    669 #endif
    670 			switch (cacherep) {
    671 			case RC_DOIT:
    672 #ifdef DIAGNOSTIC
    673 				/*
    674 				 * NFS server procs should neither release
    675 				 * locks already held, nor leave things
    676 				 * locked.  Catch this sooner, rather than
    677 				 * later (when we try to relock something we
    678 				 * already have locked).  Careful inspection
    679 				 * of the failing routine usually turns up the
    680 				 * lock leak.. once we know what it is..
    681 				 */
    682 				lockcount = l->l_locks;
    683 #endif
    684 				mreq = NULL;
    685 				netexport_rdlock();
    686 				if (writes_todo || nd == NULL ||
    687 				     (!(nd->nd_flag & ND_NFSV3) &&
    688 				     nd->nd_procnum == NFSPROC_WRITE &&
    689 				     nfsrvw_procrastinate > 0))
    690 					error = nfsrv_writegather(&nd, slp,
    691 					    l, &mreq);
    692 				else
    693 					error =
    694 					    (*(nfsrv3_procs[nd->nd_procnum]))
    695 					    (nd, slp, l, &mreq);
    696 				netexport_rdunlock();
    697 #ifdef DIAGNOSTIC
    698 				if (l->l_locks != lockcount) {
    699 					/*
    700 					 * If you see this panic, audit
    701 					 * nfsrv3_procs[nd->nd_procnum] for
    702 					 * vnode locking errors (usually, it's
    703 					 * due to forgetting to vput()
    704 					 * something).
    705 					 */
    706 #ifdef DEBUG
    707 					extern void printlockedvnodes(void);
    708 					printlockedvnodes();
    709 #endif
    710 					printf("nfsd: locking botch in op %d"
    711 					    " (before %d, after %d)\n",
    712 					    nd ? nd->nd_procnum : -1,
    713 					    lockcount, l->l_locks);
    714 				}
    715 #endif
    716 				if (mreq == NULL) {
    717 					if (nd != NULL) {
    718 						if (nd->nd_nam2)
    719 							m_free(nd->nd_nam2);
    720 						if (nd->nd_mrep)
    721 							m_freem(nd->nd_mrep);
    722 					}
    723 					break;
    724 				}
    725 				if (error) {
    726 					nfsstats.srv_errs++;
    727 					nfsrv_updatecache(nd, false, mreq);
    728 					if (nd->nd_nam2)
    729 						m_freem(nd->nd_nam2);
    730 					break;
    731 				}
    732 				nfsstats.srvrpccnt[nd->nd_procnum]++;
    733 				nfsrv_updatecache(nd, true, mreq);
    734 				nd->nd_mrep = (struct mbuf *)0;
    735 			case RC_REPLY:
    736 				m = mreq;
    737 				siz = 0;
    738 				while (m) {
    739 					siz += m->m_len;
    740 					m = m->m_next;
    741 				}
    742 				if (siz <= 0 || siz > NFS_MAXPACKET) {
    743 					printf("mbuf siz=%d\n",siz);
    744 					panic("Bad nfs svc reply");
    745 				}
    746 				m = mreq;
    747 				m->m_pkthdr.len = siz;
    748 				m->m_pkthdr.rcvif = (struct ifnet *)0;
    749 				/*
    750 				 * For stream protocols, prepend a Sun RPC
    751 				 * Record Mark.
    752 				 */
    753 				if (sotype == SOCK_STREAM) {
    754 					M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
    755 					*mtod(m, u_int32_t *) =
    756 					    htonl(0x80000000 | siz);
    757 				}
    758 				nd->nd_mreq = m;
    759 				if (nfsrtton) {
    760 					nfsd_rt(slp->ns_so->so_type, nd,
    761 					    cacherep);
    762 				}
    763 				error = nfsdsock_sendreply(slp, nd);
    764 				nd = NULL;
    765 				if (error == EPIPE)
    766 					nfsrv_zapsock(slp);
    767 				if (error == EINTR || error == ERESTART) {
    768 					nfsd->nfsd_slp = NULL;
    769 					nfsrv_slpderef(slp);
    770 					goto done;
    771 				}
    772 				break;
    773 			case RC_DROPIT:
    774 				if (nfsrtton)
    775 					nfsd_rt(sotype, nd, cacherep);
    776 				m_freem(nd->nd_mrep);
    777 				m_freem(nd->nd_nam2);
    778 				break;
    779 			}
    780 			if (nd) {
    781 				nfsdreq_free(nd);
    782 				nd = NULL;
    783 			}
    784 
    785 			/*
    786 			 * Check to see if there are outstanding writes that
    787 			 * need to be serviced.
    788 			 */
    789 			getmicrotime(&tv);
    790 			cur_usec = (u_quad_t)tv.tv_sec * 1000000 +
    791 			    (u_quad_t)tv.tv_usec;
    792 			s = splsoftclock();
    793 			if (LIST_FIRST(&slp->ns_tq) &&
    794 			    LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) {
    795 				cacherep = RC_DOIT;
    796 				writes_todo = 1;
    797 			} else
    798 				writes_todo = 0;
    799 			splx(s);
    800 		} while (writes_todo);
    801 		if (nfsrv_dorec(slp, nfsd, &nd, &dummy)) {
    802 			nfsd->nfsd_slp = NULL;
    803 			nfsrv_slpderef(slp);
    804 		}
    805 	}
    806 done:
    807 	mutex_enter(&nfsd_lock);
    808 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
    809 	doreinit = --nfs_numnfsd == 0;
    810 	if (doreinit)
    811 		nfssvc_sockhead_flag |= SLP_INIT;
    812 	mutex_exit(&nfsd_lock);
    813 	cv_destroy(&nfsd->nfsd_cv);
    814 	kmem_free(nfsd, sizeof(*nfsd));
    815 	nsd->nsd_nfsd = NULL;
    816 	if (doreinit)
    817 		nfsrv_init(true);	/* Reinitialize everything */
    818 	uvm_lwp_rele(l);
    819 	return (error);
    820 }
    821 
    822 /*
    823  * Shut down a socket associated with an nfssvc_sock structure.
    824  * Should be called with the send lock set, if required.
    825  * The trick here is to increment the sref at the start, so that the nfsds
    826  * will stop using it and clear ns_flag at the end so that it will not be
    827  * reassigned during cleanup.
    828  *
    829  * called at splsoftnet.
    830  */
    831 void
    832 nfsrv_zapsock(slp)
    833 	struct nfssvc_sock *slp;
    834 {
    835 	struct nfsuid *nuidp, *nnuidp;
    836 	struct nfsrv_descript *nwp;
    837 	struct socket *so;
    838 	struct mbuf *m;
    839 	int s;
    840 
    841 	if (nfsdsock_drain(slp)) {
    842 		return;
    843 	}
    844 	mutex_enter(&nfsd_lock);
    845 	if (slp->ns_gflags & SLP_G_DOREC) {
    846 		TAILQ_REMOVE(&nfssvc_sockpending, slp, ns_pending);
    847 		slp->ns_gflags &= ~SLP_G_DOREC;
    848 	}
    849 	mutex_exit(&nfsd_lock);
    850 
    851 	so = slp->ns_so;
    852 	KASSERT(so != NULL);
    853 	KERNEL_LOCK(1, curlwp);
    854 	s = splsoftnet();
    855 	so->so_upcall = NULL;
    856 	so->so_upcallarg = NULL;
    857 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    858 	splx(s);
    859 	soshutdown(so, SHUT_RDWR);
    860 	KERNEL_UNLOCK_ONE(curlwp);
    861 
    862 	if (slp->ns_nam)
    863 		m_free(slp->ns_nam);
    864 	m_freem(slp->ns_raw);
    865 	m = slp->ns_rec;
    866 	while (m != NULL) {
    867 		struct mbuf *n;
    868 
    869 		n = m->m_nextpkt;
    870 		m_freem(m);
    871 		m = n;
    872 	}
    873 	for (nuidp = TAILQ_FIRST(&slp->ns_uidlruhead); nuidp != 0;
    874 	    nuidp = nnuidp) {
    875 		nnuidp = TAILQ_NEXT(nuidp, nu_lru);
    876 		LIST_REMOVE(nuidp, nu_hash);
    877 		TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
    878 		if (nuidp->nu_flag & NU_NAM)
    879 			m_freem(nuidp->nu_nam);
    880 		kmem_free(nuidp, sizeof(*nuidp));
    881 	}
    882 	mutex_enter(&nfsd_lock);
    883 	while ((nwp = LIST_FIRST(&slp->ns_tq)) != NULL) {
    884 		LIST_REMOVE(nwp, nd_tq);
    885 		mutex_exit(&nfsd_lock);
    886 		nfsdreq_free(nwp);
    887 		mutex_enter(&nfsd_lock);
    888 	}
    889 	mutex_exit(&nfsd_lock);
    890 }
    891 
    892 /*
    893  * Derefence a server socket structure. If it has no more references and
    894  * is no longer valid, you can throw it away.
    895  */
    896 void
    897 nfsrv_slpderef(slp)
    898 	struct nfssvc_sock *slp;
    899 {
    900 	uint32_t ref;
    901 
    902 	mutex_enter(&nfsd_lock);
    903 	KASSERT(slp->ns_sref > 0);
    904 	ref = --slp->ns_sref;
    905 	mutex_exit(&nfsd_lock);
    906 	if (ref == 0 && (slp->ns_flags & SLP_VALID) == 0) {
    907 		struct file *fp;
    908 
    909 		mutex_enter(&nfsd_lock);
    910 		KASSERT((slp->ns_gflags & SLP_G_DOREC) == 0);
    911 		TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
    912 		mutex_exit(&nfsd_lock);
    913 
    914 		fp = slp->ns_fp;
    915 		if (fp != NULL) {
    916 			slp->ns_fp = NULL;
    917 			KASSERT(fp != NULL);
    918 			KASSERT(fp->f_data == slp->ns_so);
    919 			mutex_enter(&fp->f_lock);
    920 			FILE_USE(fp);
    921 			closef(fp, (struct lwp *)0);
    922 			slp->ns_so = NULL;
    923 		}
    924 
    925 		nfsrv_sockfree(slp);
    926 	}
    927 }
    928 
    929 /*
    930  * Initialize the data structures for the server.
    931  * Handshake with any new nfsds starting up to avoid any chance of
    932  * corruption.
    933  */
    934 void
    935 nfsrv_init(terminating)
    936 	int terminating;
    937 {
    938 	struct nfssvc_sock *slp;
    939 
    940 	if (!terminating) {
    941 		/* XXX could be IPL_SOFTNET */
    942 		mutex_init(&nfsd_lock, MUTEX_DRIVER, IPL_VM);
    943 		cv_init(&nfsd_initcv, "nfsdinit");
    944 	}
    945 
    946 	mutex_enter(&nfsd_lock);
    947 	if (!terminating && (nfssvc_sockhead_flag & SLP_INIT) != 0)
    948 		panic("nfsd init");
    949 	nfssvc_sockhead_flag |= SLP_INIT;
    950 
    951 	if (terminating) {
    952 		KASSERT(SLIST_EMPTY(&nfsd_idle_head));
    953 		KASSERT(TAILQ_EMPTY(&nfsd_head));
    954 		while ((slp = TAILQ_FIRST(&nfssvc_sockhead)) != NULL) {
    955 			mutex_exit(&nfsd_lock);
    956 			KASSERT(slp->ns_sref == 0);
    957 			slp->ns_sref++;
    958 			nfsrv_zapsock(slp);
    959 			nfsrv_slpderef(slp);
    960 			mutex_enter(&nfsd_lock);
    961 		}
    962 		KASSERT(TAILQ_EMPTY(&nfssvc_sockpending));
    963 		mutex_exit(&nfsd_lock);
    964 		nfsrv_cleancache();	/* And clear out server cache */
    965 	} else {
    966 		mutex_exit(&nfsd_lock);
    967 		nfs_pub.np_valid = 0;
    968 	}
    969 
    970 	TAILQ_INIT(&nfssvc_sockhead);
    971 	TAILQ_INIT(&nfssvc_sockpending);
    972 
    973 	TAILQ_INIT(&nfsd_head);
    974 	SLIST_INIT(&nfsd_idle_head);
    975 	nfsd_head_flag &= ~NFSD_CHECKSLP;
    976 
    977 	nfs_udpsock = nfsrv_sockalloc();
    978 
    979 #ifdef INET6
    980 	nfs_udp6sock = nfsrv_sockalloc();
    981 #endif
    982 
    983 #ifdef ISO
    984 	nfs_cltpsock = nfsrv_sockalloc();
    985 #endif
    986 
    987 	mutex_enter(&nfsd_lock);
    988 	nfssvc_sockhead_flag &= ~SLP_INIT;
    989 	cv_broadcast(&nfsd_initcv);
    990 	mutex_exit(&nfsd_lock);
    991 }
    992 
    993 /*
    994  * Add entries to the server monitor log.
    995  */
    996 static void
    997 nfsd_rt(sotype, nd, cacherep)
    998 	int sotype;
    999 	struct nfsrv_descript *nd;
   1000 	int cacherep;
   1001 {
   1002 	struct timeval tv;
   1003 	struct drt *rt;
   1004 
   1005 	rt = &nfsdrt.drt[nfsdrt.pos];
   1006 	if (cacherep == RC_DOIT)
   1007 		rt->flag = 0;
   1008 	else if (cacherep == RC_REPLY)
   1009 		rt->flag = DRT_CACHEREPLY;
   1010 	else
   1011 		rt->flag = DRT_CACHEDROP;
   1012 	if (sotype == SOCK_STREAM)
   1013 		rt->flag |= DRT_TCP;
   1014 	if (nd->nd_flag & ND_NFSV3)
   1015 		rt->flag |= DRT_NFSV3;
   1016 	rt->proc = nd->nd_procnum;
   1017 	if (mtod(nd->nd_nam, struct sockaddr *)->sa_family == AF_INET)
   1018 	    rt->ipadr = mtod(nd->nd_nam, struct sockaddr_in *)->sin_addr.s_addr;
   1019 	else
   1020 	    rt->ipadr = INADDR_ANY;
   1021 	getmicrotime(&tv);
   1022 	rt->resptime = ((tv.tv_sec - nd->nd_starttime.tv_sec) * 1000000) +
   1023 		(tv.tv_usec - nd->nd_starttime.tv_usec);
   1024 	rt->tstamp = tv;
   1025 	nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
   1026 }
   1027 #endif /* NFSSERVER */
   1028 
   1029 #ifdef NFS
   1030 
   1031 int nfs_defect = 0;
   1032 /*
   1033  * Asynchronous I/O threads for client nfs.
   1034  * They do read-ahead and write-behind operations on the block I/O cache.
   1035  * Never returns unless it fails or gets killed.
   1036  */
   1037 
   1038 static void
   1039 nfssvc_iod(void *arg)
   1040 {
   1041 	struct buf *bp;
   1042 	struct nfs_iod *myiod;
   1043 	struct nfsmount *nmp;
   1044 
   1045 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
   1046 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
   1047 	cv_init(&myiod->nid_cv, "nfsiod");
   1048 	myiod->nid_exiting = false;
   1049 	myiod->nid_mount = NULL;
   1050 	mutex_enter(&nfs_iodlist_lock);
   1051 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
   1052 	mutex_exit(&nfs_iodlist_lock);
   1053 
   1054 	for (;;) {
   1055 		mutex_enter(&nfs_iodlist_lock);
   1056 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
   1057 		mutex_exit(&nfs_iodlist_lock);
   1058 
   1059 		mutex_enter(&myiod->nid_lock);
   1060 		while (/*CONSTCOND*/ true) {
   1061 			nmp = myiod->nid_mount;
   1062 			if (nmp) {
   1063 				myiod->nid_mount = NULL;
   1064 				break;
   1065 			}
   1066 			if (__predict_false(myiod->nid_exiting)) {
   1067 				/*
   1068 				 * drop nid_lock to preserve locking order.
   1069 				 */
   1070 				mutex_exit(&myiod->nid_lock);
   1071 				mutex_enter(&nfs_iodlist_lock);
   1072 				mutex_enter(&myiod->nid_lock);
   1073 				/*
   1074 				 * recheck nid_mount because nfs_asyncio can
   1075 				 * pick us in the meantime as we are still on
   1076 				 * nfs_iodlist_lock.
   1077 				 */
   1078 				if (myiod->nid_mount != NULL) {
   1079 					mutex_exit(&nfs_iodlist_lock);
   1080 					continue;
   1081 				}
   1082 				LIST_REMOVE(myiod, nid_idle);
   1083 				mutex_exit(&nfs_iodlist_lock);
   1084 				goto quit;
   1085 			}
   1086 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
   1087 		}
   1088 		mutex_exit(&myiod->nid_lock);
   1089 
   1090 		mutex_enter(&nmp->nm_lock);
   1091 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
   1092 			/* Take one off the front of the list */
   1093 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
   1094 			nmp->nm_bufqlen--;
   1095 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
   1096 				cv_broadcast(&nmp->nm_aiocv);
   1097 			}
   1098 			mutex_exit(&nmp->nm_lock);
   1099 			KERNEL_LOCK(1, curlwp);
   1100 			(void)nfs_doio(bp);
   1101 			KERNEL_UNLOCK_LAST(curlwp);
   1102 			mutex_enter(&nmp->nm_lock);
   1103 			/*
   1104 			 * If there are more than one iod on this mount,
   1105 			 * then defect so that the iods can be shared out
   1106 			 * fairly between the mounts
   1107 			 */
   1108 			if (nfs_defect && nmp->nm_bufqiods > 1) {
   1109 				break;
   1110 			}
   1111 		}
   1112 		KASSERT(nmp->nm_bufqiods > 0);
   1113 		nmp->nm_bufqiods--;
   1114 		mutex_exit(&nmp->nm_lock);
   1115 	}
   1116 quit:
   1117 	KASSERT(myiod->nid_mount == NULL);
   1118 	mutex_exit(&myiod->nid_lock);
   1119 
   1120 	cv_destroy(&myiod->nid_cv);
   1121 	mutex_destroy(&myiod->nid_lock);
   1122 	kmem_free(myiod, sizeof(*myiod));
   1123 
   1124 	kthread_exit(0);
   1125 }
   1126 
   1127 void
   1128 nfs_iodinit()
   1129 {
   1130 
   1131 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
   1132 	LIST_INIT(&nfs_iodlist_all);
   1133 	LIST_INIT(&nfs_iodlist_idle);
   1134 }
   1135 
   1136 int
   1137 nfs_set_niothreads(int newval)
   1138 {
   1139 	struct nfs_iod *nid;
   1140 	int error = 0;
   1141 
   1142 #if defined(MULTIPROCESSOR)
   1143         int hold_count;
   1144 #endif /* defined(MULTIPROCESSOR) */
   1145 
   1146 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
   1147 
   1148 	mutex_enter(&nfs_iodlist_lock);
   1149 	/* clamp to sane range */
   1150 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
   1151 
   1152 	while (nfs_numasync != nfs_niothreads && error == 0) {
   1153 		while (nfs_numasync < nfs_niothreads) {
   1154 
   1155 			/*
   1156 			 * kthread_create can wait for pagedaemon and
   1157 			 * pagedaemon can wait for nfsiod which needs to acquire
   1158 			 * nfs_iodlist_lock.
   1159 			 */
   1160 
   1161 			mutex_exit(&nfs_iodlist_lock);
   1162 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
   1163 			    nfssvc_iod, NULL, NULL, "nfsio");
   1164 			mutex_enter(&nfs_iodlist_lock);
   1165 			if (error) {
   1166 				/* give up */
   1167 				nfs_niothreads = nfs_numasync;
   1168 				break;
   1169 			}
   1170 			nfs_numasync++;
   1171 		}
   1172 		while (nfs_numasync > nfs_niothreads) {
   1173 			nid = LIST_FIRST(&nfs_iodlist_all);
   1174 			if (nid == NULL) {
   1175 				/* iod has not started yet. */
   1176 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
   1177 				continue;
   1178 			}
   1179 			LIST_REMOVE(nid, nid_all);
   1180 			mutex_enter(&nid->nid_lock);
   1181 			KASSERT(!nid->nid_exiting);
   1182 			nid->nid_exiting = true;
   1183 			cv_signal(&nid->nid_cv);
   1184 			mutex_exit(&nid->nid_lock);
   1185 			nfs_numasync--;
   1186 		}
   1187 	}
   1188 	mutex_exit(&nfs_iodlist_lock);
   1189 
   1190 	KERNEL_LOCK(hold_count, curlwp);
   1191 	return error;
   1192 }
   1193 
   1194 /*
   1195  * Get an authorization string for the uid by having the mount_nfs sitting
   1196  * on this mount point porpous out of the kernel and do it.
   1197  */
   1198 int
   1199 nfs_getauth(nmp, rep, cred, auth_str, auth_len, verf_str, verf_len, key)
   1200 	struct nfsmount *nmp;
   1201 	struct nfsreq *rep;
   1202 	kauth_cred_t cred;
   1203 	char **auth_str;
   1204 	int *auth_len;
   1205 	char *verf_str;
   1206 	int *verf_len;
   1207 	NFSKERBKEY_T key;		/* return session key */
   1208 {
   1209 	int error = 0;
   1210 
   1211 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
   1212 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
   1213 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
   1214 			"nfsauth1", 2 * hz);
   1215 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1216 		if (error) {
   1217 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1218 			return (error);
   1219 		}
   1220 	}
   1221 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
   1222 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
   1223 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
   1224 	nmp->nm_verfstr = verf_str;
   1225 	nmp->nm_verflen = *verf_len;
   1226 	nmp->nm_authuid = kauth_cred_geteuid(cred);
   1227 	wakeup((void *)&nmp->nm_authstr);
   1228 
   1229 	/*
   1230 	 * And wait for mount_nfs to do its stuff.
   1231 	 */
   1232 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
   1233 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
   1234 			"nfsauth2", 2 * hz);
   1235 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
   1236 	}
   1237 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
   1238 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
   1239 		error = EAUTH;
   1240 	}
   1241 	if (error)
   1242 		free((void *)*auth_str, M_TEMP);
   1243 	else {
   1244 		*auth_len = nmp->nm_authlen;
   1245 		*verf_len = nmp->nm_verflen;
   1246 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
   1247 	}
   1248 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
   1249 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
   1250 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
   1251 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
   1252 		wakeup((void *)&nmp->nm_authtype);
   1253 	}
   1254 	return (error);
   1255 }
   1256 
   1257 /*
   1258  * Get a nickname authenticator and verifier.
   1259  */
   1260 int
   1261 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
   1262     int *auth_len, char *verf_str, int verf_len)
   1263 {
   1264 	struct timeval ktvin, ktvout, tv;
   1265 	struct nfsuid *nuidp;
   1266 	u_int32_t *nickp, *verfp;
   1267 
   1268 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
   1269 
   1270 #ifdef DIAGNOSTIC
   1271 	if (verf_len < (4 * NFSX_UNSIGNED))
   1272 		panic("nfs_getnickauth verf too small");
   1273 #endif
   1274 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
   1275 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
   1276 			break;
   1277 	}
   1278 	if (!nuidp || nuidp->nu_expire < time_second)
   1279 		return (EACCES);
   1280 
   1281 	/*
   1282 	 * Move to the end of the lru list (end of lru == most recently used).
   1283 	 */
   1284 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1285 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
   1286 
   1287 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
   1288 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1289 	*nickp = txdr_unsigned(nuidp->nu_nickname);
   1290 	*auth_str = (char *)nickp;
   1291 	*auth_len = 2 * NFSX_UNSIGNED;
   1292 
   1293 	/*
   1294 	 * Now we must encrypt the verifier and package it up.
   1295 	 */
   1296 	verfp = (u_int32_t *)verf_str;
   1297 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
   1298 	getmicrotime(&tv);
   1299 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
   1300 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
   1301 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
   1302 		nuidp->nu_timestamp = tv;
   1303 	else
   1304 		nuidp->nu_timestamp.tv_usec++;
   1305 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
   1306 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
   1307 
   1308 	/*
   1309 	 * Now encrypt the timestamp verifier in ecb mode using the session
   1310 	 * key.
   1311 	 */
   1312 #ifdef NFSKERB
   1313 	XXX
   1314 #endif
   1315 
   1316 	*verfp++ = ktvout.tv_sec;
   1317 	*verfp++ = ktvout.tv_usec;
   1318 	*verfp = 0;
   1319 	return (0);
   1320 }
   1321 
   1322 /*
   1323  * Save the current nickname in a hash list entry on the mount point.
   1324  */
   1325 int
   1326 nfs_savenickauth(nmp, cred, len, key, mdp, dposp, mrep)
   1327 	struct nfsmount *nmp;
   1328 	kauth_cred_t cred;
   1329 	int len;
   1330 	NFSKERBKEY_T key;
   1331 	struct mbuf **mdp;
   1332 	char **dposp;
   1333 	struct mbuf *mrep;
   1334 {
   1335 	struct nfsuid *nuidp;
   1336 	u_int32_t *tl;
   1337 	int32_t t1;
   1338 	struct mbuf *md = *mdp;
   1339 	struct timeval ktvin, ktvout;
   1340 	u_int32_t nick;
   1341 	char *dpos = *dposp, *cp2;
   1342 	int deltasec, error = 0;
   1343 
   1344 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
   1345 
   1346 	if (len == (3 * NFSX_UNSIGNED)) {
   1347 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
   1348 		ktvin.tv_sec = *tl++;
   1349 		ktvin.tv_usec = *tl++;
   1350 		nick = fxdr_unsigned(u_int32_t, *tl);
   1351 
   1352 		/*
   1353 		 * Decrypt the timestamp in ecb mode.
   1354 		 */
   1355 #ifdef NFSKERB
   1356 		XXX
   1357 #endif
   1358 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
   1359 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
   1360 		deltasec = time_second - ktvout.tv_sec;
   1361 		if (deltasec < 0)
   1362 			deltasec = -deltasec;
   1363 		/*
   1364 		 * If ok, add it to the hash list for the mount point.
   1365 		 */
   1366 		if (deltasec <= NFS_KERBCLOCKSKEW) {
   1367 			if (nmp->nm_numuids < nuidhash_max) {
   1368 				nmp->nm_numuids++;
   1369 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
   1370 			} else {
   1371 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
   1372 				LIST_REMOVE(nuidp, nu_hash);
   1373 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
   1374 					nu_lru);
   1375 			}
   1376 			nuidp->nu_flag = 0;
   1377 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
   1378 			nuidp->nu_expire = time_second + NFS_KERBTTL;
   1379 			nuidp->nu_timestamp = ktvout;
   1380 			nuidp->nu_nickname = nick;
   1381 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
   1382 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
   1383 				nu_lru);
   1384 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
   1385 				nuidp, nu_hash);
   1386 		}
   1387 	} else
   1388 		nfsm_adv(nfsm_rndup(len));
   1389 nfsmout:
   1390 	*mdp = md;
   1391 	*dposp = dpos;
   1392 	return (error);
   1393 }
   1394 #endif /* NFS */
   1395