xref: /macosx-10.9.5/xnu-2422.115.4/bsd/nfs/nfs_gss.c

/*
 * Copyright (c) 2007-2013 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */

/*************
 * These functions implement RPCSEC_GSS security for the NFS client and server.
 * The code is specific to the use of Kerberos v5 and the use of DES MAC MD5
 * protection as described in Internet RFC 2203 and 2623.
 *
 * In contrast to the original AUTH_SYS authentication, RPCSEC_GSS is stateful.
 * It requires the client and server negotiate a secure connection as part of a
 * security context. The context state is maintained in client and server structures.
 * On the client side, each user of an NFS mount is assigned their own context,
 * identified by UID, on their first use of the mount, and it persists until the
 * unmount or until the context is renewed.  Each user context has a corresponding
 * server context which the server maintains until the client destroys it, or
 * until the context expires.
 *
 * The client and server contexts are set up dynamically.  When a user attempts
 * to send an NFS request, if there is no context for the user, then one is
 * set up via an exchange of NFS null procedure calls as described in RFC 2203.
 * During this exchange, the client and server pass a security token that is
 * forwarded via Mach upcall to the gssd, which invokes the GSS-API to authenticate
 * the user to the server (and vice-versa). The client and server also receive
 * a unique session key that can be used to digitally sign the credentials and
 * verifier or optionally to provide data integrity and/or privacy.
 *
 * Once the context is complete, the client and server enter a normal data
 * exchange phase - beginning with the NFS request that prompted the context
 * creation. During this phase, the client's RPC header contains an RPCSEC_GSS
 * credential and verifier, and the server returns a verifier as well.
 * For simple authentication, the verifier contains a signed checksum of the
 * RPC header, including the credential.  The server's verifier has a signed
 * checksum of the current sequence number.
 *
 * Each client call contains a sequence number that nominally increases by one
 * on each request.  The sequence number is intended to prevent replay attacks.
 * Since the protocol can be used over UDP, there is some allowance for
 * out-of-sequence requests, so the server checks whether the sequence numbers
 * are within a sequence "window". If a sequence number is outside the lower
 * bound of the window, the server silently drops the request. This has some
 * implications for retransmission. If a request needs to be retransmitted, the
 * client must bump the sequence number even if the request XID is unchanged.
 *
 * When the NFS mount is unmounted, the client sends a "destroy" credential
 * to delete the server's context for each user of the mount. Since it's
 * possible for the client to crash or disconnect without sending the destroy
 * message, the server has a thread that reaps contexts that have been idle
 * too long.
 */

#include <stdint.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/kauth.h>
#include <sys/kernel.h>
#include <sys/mount_internal.h>
#include <sys/vnode.h>
#include <sys/ubc.h>
#include <sys/malloc.h>
#include <sys/kpi_mbuf.h>
#include <sys/ucred.h>

#include <kern/host.h>
#include <libkern/libkern.h>

#include <mach/task.h>
#include <mach/host_special_ports.h>
#include <mach/host_priv.h>
#include <mach/thread_act.h>
#include <mach/mig_errors.h>
#include <mach/vm_map.h>
#include <vm/vm_map.h>
#include <vm/vm_kern.h>
#include <gssd/gssd_mach.h>

#include <nfs/rpcv2.h>
#include <nfs/nfsproto.h>
#include <nfs/nfs.h>
#include <nfs/nfsnode.h>
#include <nfs/nfs_gss.h>
#include <nfs/nfsmount.h>
#include <nfs/xdr_subs.h>
#include <nfs/nfsm_subs.h>
#include <nfs/nfs_gss.h>

#include "nfs_gss_crypto.h"

#define NFS_GSS_MACH_MAX_RETRIES 3

#define NFS_GSS_DBG(...) NFS_DBG(NFS_FAC_GSS, 7, ## __VA_ARGS__)
#define NFS_GSS_ISDBG  (NFS_DEBUG_FACILITY &  NFS_FAC_GSS)

typedef struct {
	int type;
	union {
		MD5_DESCBC_CTX m_ctx;
		HMAC_SHA1_DES3KD_CTX h_ctx;
	};
} GSS_DIGEST_CTX;

#define MAX_DIGEST SHA_DIGEST_LENGTH
#ifdef NFS_KERNEL_DEBUG
#define HASHLEN(ki)  (((ki)->hash_len > MAX_DIGEST) ? \
		(panic("nfs_gss.c:%d ki->hash_len is invalid = %d\n", __LINE__, (ki)->hash_len), MAX_DIGEST) : (ki)->hash_len)
#else
#define HASHLEN(ki)  (((ki)->hash_len > MAX_DIGEST) ? \
		(printf("nfs_gss.c:%d ki->hash_len is invalid = %d\n", __LINE__, (ki)->hash_len), MAX_DIGEST) : (ki)->hash_len)
#endif

#if NFSSERVER
u_long nfs_gss_svc_ctx_hash;
struct nfs_gss_svc_ctx_hashhead *nfs_gss_svc_ctx_hashtbl;
lck_mtx_t *nfs_gss_svc_ctx_mutex;
lck_grp_t *nfs_gss_svc_grp;
uint32_t nfsrv_gss_context_ttl = GSS_CTX_EXPIRE;
#define GSS_SVC_CTX_TTL ((uint64_t)max(2*GSS_CTX_PEND, nfsrv_gss_context_ttl) * NSEC_PER_SEC)
#endif /* NFSSERVER */

#if NFSCLIENT
lck_grp_t *nfs_gss_clnt_grp;
int nfs_single_des;
#endif /* NFSCLIENT */

/*
 * These octet strings are used to encode/decode ASN.1 tokens
 * in the RPCSEC_GSS verifiers.
 */
static u_char krb5_tokhead[] __attribute__((unused)) = { 0x60, 0x23 };
       u_char krb5_mech[11] = { 0x06, 0x09, 0x2a, 0x86, 0x48, 0x86, 0xf7, 0x12, 0x01, 0x02, 0x02 };
static u_char krb5_mic[]  = { 0x01, 0x01, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff };
static u_char krb5_mic3[]  = { 0x01, 0x01, 0x04, 0x00, 0xff, 0xff, 0xff, 0xff };
static u_char krb5_wrap[] = { 0x02, 0x01, 0x00, 0x00, 0x00, 0x00, 0xff, 0xff };
static u_char krb5_wrap3[] = { 0x02, 0x01, 0x04, 0x00, 0x02, 0x00, 0xff, 0xff };
static u_char iv0[] = { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 }; // DES MAC Initialization Vector

#define ALG_MIC(ki) (((ki)->type == NFS_GSS_1DES) ? krb5_mic : krb5_mic3)
#define ALG_WRAP(ki) (((ki)->type == NFS_GSS_1DES) ? krb5_wrap : krb5_wrap3)

/*
 * The size of the Kerberos v5 ASN.1 token
 * in the verifier.
 *
 * Note that the second octet of the krb5_tokhead (0x23) is a
 * DER-encoded size field that has variable length.  If the size
 * is 128 bytes or greater, then it uses two bytes, three bytes
 * if 65536 or greater, and so on.  Since the MIC tokens are
 * separate from the data, the size is always the same: 35 bytes (0x23).
 * However, the wrap token is different. Its size field includes the
 * size of the token + the encrypted data that follows. So the size
 * field may be two, three or four bytes.
 */
#define KRB5_SZ_TOKHEAD sizeof(krb5_tokhead)
#define KRB5_SZ_MECH	sizeof(krb5_mech)
#define KRB5_SZ_ALG	sizeof(krb5_mic) // 8 - same as krb5_wrap
#define KRB5_SZ_SEQ	8
#define KRB5_SZ_EXTRA	3  // a wrap token may be longer by up to this many octets
#define KRB5_SZ_TOKEN_NOSUM	(KRB5_SZ_TOKHEAD + KRB5_SZ_MECH + KRB5_SZ_ALG + KRB5_SZ_SEQ)
#define KRB5_SZ_TOKEN(cksumlen)		((cksumlen) + KRB5_SZ_TOKEN_NOSUM)
#define KRB5_SZ_TOKMAX(cksumlen)	(KRB5_SZ_TOKEN(cksumlen) + KRB5_SZ_EXTRA)

#if NFSCLIENT
static int	nfs_gss_clnt_ctx_find(struct nfsreq *);
static int	nfs_gss_clnt_ctx_failover(struct nfsreq *);
static int	nfs_gss_clnt_ctx_init(struct nfsreq *, struct nfs_gss_clnt_ctx *);
static int	nfs_gss_clnt_ctx_init_retry(struct nfsreq *, struct nfs_gss_clnt_ctx *);
static int	nfs_gss_clnt_ctx_callserver(struct nfsreq *, struct nfs_gss_clnt_ctx *);
static uint8_t	*nfs_gss_clnt_svcname(struct nfsmount *, gssd_nametype *, uint32_t *);
static int	nfs_gss_clnt_gssd_upcall(struct nfsreq *, struct nfs_gss_clnt_ctx *);
static void	nfs_gss_clnt_ctx_remove(struct nfsmount *, struct nfs_gss_clnt_ctx *);
#endif /* NFSCLIENT */

#if NFSSERVER
static struct nfs_gss_svc_ctx *nfs_gss_svc_ctx_find(uint32_t);
static void	nfs_gss_svc_ctx_insert(struct nfs_gss_svc_ctx *);
static void	nfs_gss_svc_ctx_timer(void *, void *);
static int	nfs_gss_svc_gssd_upcall(struct nfs_gss_svc_ctx *);
static int	nfs_gss_svc_seqnum_valid(struct nfs_gss_svc_ctx *, uint32_t);
#endif /* NFSSERVER */

static void	host_release_special_port(mach_port_t);
static mach_port_t host_copy_special_port(mach_port_t);
static void	nfs_gss_mach_alloc_buffer(u_char *, uint32_t, vm_map_copy_t *);
static int	nfs_gss_mach_vmcopyout(vm_map_copy_t, uint32_t, u_char *);
static int	nfs_gss_token_get(gss_key_info *ki, u_char *, u_char *, int, uint32_t *, u_char *);
static int	nfs_gss_token_put(gss_key_info *ki, u_char *, u_char *, int, int, u_char *);
static int	nfs_gss_der_length_size(int);
static void	nfs_gss_der_length_put(u_char **, int);
static int	nfs_gss_der_length_get(u_char **);
static int	nfs_gss_mchain_length(mbuf_t);
static int	nfs_gss_append_chain(struct nfsm_chain *, mbuf_t);
static void	nfs_gss_nfsm_chain(struct nfsm_chain *, mbuf_t);
static void	nfs_gss_cksum_mchain(gss_key_info *, mbuf_t, u_char *, int, int, u_char *);
static void	nfs_gss_cksum_chain(gss_key_info *, struct nfsm_chain *, u_char *, int, int, u_char *);
static void	nfs_gss_cksum_rep(gss_key_info *, uint32_t, u_char *);
static void	nfs_gss_encrypt_mchain(gss_key_info *, mbuf_t, int, int, int);
static void	nfs_gss_encrypt_chain(gss_key_info *, struct nfsm_chain *, int, int, int);

static void	gss_digest_Init(GSS_DIGEST_CTX *, gss_key_info *);
static void	gss_digest_Update(GSS_DIGEST_CTX *, void *, size_t);
static void	gss_digest_Final(GSS_DIGEST_CTX *, void *);
static void	gss_des_crypt(gss_key_info *, des_cblock *, des_cblock *,
				int32_t, des_cblock *, des_cblock *, int, int);
static int	gss_key_init(gss_key_info *, uint32_t);

#if NFSSERVER
thread_call_t nfs_gss_svc_ctx_timer_call;
int nfs_gss_timer_on = 0;
uint32_t nfs_gss_ctx_count = 0;
const uint32_t nfs_gss_ctx_max = GSS_SVC_MAXCONTEXTS;
#endif /* NFSSERVER */

/*
 * Initialization when NFS starts
 */
void
nfs_gss_init(void)
{
#if NFSCLIENT
	nfs_gss_clnt_grp = lck_grp_alloc_init("rpcsec_gss_clnt", LCK_GRP_ATTR_NULL);
#endif /* NFSCLIENT */

#if NFSSERVER
	nfs_gss_svc_grp  = lck_grp_alloc_init("rpcsec_gss_svc",  LCK_GRP_ATTR_NULL);

	nfs_gss_svc_ctx_hashtbl = hashinit(SVC_CTX_HASHSZ, M_TEMP, &nfs_gss_svc_ctx_hash);
	nfs_gss_svc_ctx_mutex = lck_mtx_alloc_init(nfs_gss_svc_grp, LCK_ATTR_NULL);

	nfs_gss_svc_ctx_timer_call = thread_call_allocate(nfs_gss_svc_ctx_timer, NULL);
#endif /* NFSSERVER */
}

#if NFSCLIENT

/*
 * Is it OK to fall back to using AUTH_SYS?
 */
static int
nfs_gss_sysok(struct nfsreq *req)
{
	struct nfsmount *nmp = req->r_nmp;
	int i;

	if (req->r_wrongsec) /* Not OK if we're trying to handle a wrongsec error */
		return (0);
	if (!nmp->nm_sec.count) /* assume it's OK if we don't have a set of flavors */
		return (1);
	for (i=0; i < nmp->nm_sec.count; i++)
		if (nmp->nm_sec.flavors[i] == RPCAUTH_SYS)
			return (1);
	return (0);
}

/*
 * Find the context for a particular user.
 *
 * If the context doesn't already exist
 * then create a new context for this user.
 *
 * Note that the code allows superuser (uid == 0)
 * to adopt the context of another user.
 *
 * We'll match on the audit session ids, since those
 * processes will have acccess to the same credential cache.
 */

#define kauth_cred_getasid(cred) ((cred)->cr_audit.as_aia_p->ai_asid)
#define kauth_cred_getauid(cred) ((cred)->cr_audit.as_aia_p->ai_auid)

static int
nfs_gss_clnt_ctx_cred_match(kauth_cred_t cred1, kauth_cred_t cred2)
{
	if (kauth_cred_getasid(cred1) == kauth_cred_getasid(cred2))
		return (1);
	return (0);
}


static int
nfs_gss_clnt_ctx_find(struct nfsreq *req)
{
	struct nfsmount *nmp = req->r_nmp;
	struct nfs_gss_clnt_ctx *cp;
	int error = 0;

	lck_mtx_lock(&nmp->nm_lock);
	TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
		if (nfs_gss_clnt_ctx_cred_match(cp->gss_clnt_cred, req->r_cred)) {
			if (cp->gss_clnt_flags & GSS_CTX_INVAL)
				continue;
			nfs_gss_clnt_ctx_ref(req, cp);
			lck_mtx_unlock(&nmp->nm_lock);
			return (0);
		}
	}

	if (kauth_cred_getuid(req->r_cred) == 0) {
		/*
		 * If superuser is trying to get access, then co-opt
		 * the first valid context in the list.
		 * XXX Ultimately, we need to allow superuser to
		 * go ahead and attempt to set up its own context
		 * in case one is set up for it.
		 */
		TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
			if (!(cp->gss_clnt_flags & GSS_CTX_INVAL)) {
				nfs_gss_clnt_ctx_ref(req, cp);
				lck_mtx_unlock(&nmp->nm_lock);
				return (0);
			}
		}
	}

	/*
	 * Not found - create a new context
	 */

	/*
	 * If the thread is async, then it cannot get
	 * kerberos creds and set up a proper context.
	 * If no sec= mount option is given, attempt
	 * to failover to sec=sys.
	 */
	if (req->r_thread == NULL) {
		if (nfs_gss_sysok(req)) {
			error = nfs_gss_clnt_ctx_failover(req);
		} else {
			printf("nfs_gss_clnt_ctx_find: no context for async\n");
			error = NFSERR_EAUTH;
		}

		lck_mtx_unlock(&nmp->nm_lock);
		return (error);
	}

	MALLOC(cp, struct nfs_gss_clnt_ctx *, sizeof(*cp), M_TEMP, M_WAITOK|M_ZERO);
	if (cp == NULL) {
		lck_mtx_unlock(&nmp->nm_lock);
		return (ENOMEM);
	}

	cp->gss_clnt_cred = req->r_cred;
	kauth_cred_ref(cp->gss_clnt_cred);
	cp->gss_clnt_mtx = lck_mtx_alloc_init(nfs_gss_clnt_grp, LCK_ATTR_NULL);
	cp->gss_clnt_thread = current_thread();
	nfs_gss_clnt_ctx_ref(req, cp);
	TAILQ_INSERT_TAIL(&nmp->nm_gsscl, cp, gss_clnt_entries);
	lck_mtx_unlock(&nmp->nm_lock);

	error = nfs_gss_clnt_ctx_init_retry(req, cp); // Initialize new context
	if (error)
		nfs_gss_clnt_ctx_unref(req);

	/*
	 * If we failed to set up a Kerberos context for this
	 * user and no sec= mount option was given, but the
	 * server indicated that it could support AUTH_SYS, then set
	 * up a dummy context that allows this user to attempt
	 * sec=sys calls.
	 */
	if (error && nfs_gss_sysok(req) &&
	    (error != ENXIO) && (error != ETIMEDOUT)) {
		lck_mtx_lock(&nmp->nm_lock);
		error = nfs_gss_clnt_ctx_failover(req);
		lck_mtx_unlock(&nmp->nm_lock);
	}

	return (error);
}

/*
 * Set up a dummy context to allow the use of sec=sys
 * for this user, if the server allows sec=sys.
 * The context is valid for GSS_CLNT_SYS_VALID seconds,
 * so that the user will periodically attempt to fail back
 * and get a real credential.
 *
 * Assumes context list (nm_lock) is locked
 */
static int
nfs_gss_clnt_ctx_failover(struct nfsreq *req)
{
	struct nfsmount *nmp = req->r_nmp;
	struct nfs_gss_clnt_ctx *cp;
	struct timeval now;

	MALLOC(cp, struct nfs_gss_clnt_ctx *, sizeof(*cp), M_TEMP, M_WAITOK|M_ZERO);
	if (cp == NULL)
		return (ENOMEM);

	cp->gss_clnt_service = RPCSEC_GSS_SVC_SYS;
	cp->gss_clnt_cred = req->r_cred;
	kauth_cred_ref(cp->gss_clnt_cred);
	cp->gss_clnt_mtx = lck_mtx_alloc_init(nfs_gss_clnt_grp, LCK_ATTR_NULL);
	microuptime(&now);
	cp->gss_clnt_ctime = now.tv_sec;	// time stamp
	nfs_gss_clnt_ctx_ref(req, cp);
	TAILQ_INSERT_TAIL(&nmp->nm_gsscl, cp, gss_clnt_entries);

	return (0);
}

/*
 * Inserts an RPCSEC_GSS credential into an RPC header.
 * After the credential is inserted, the code continues
 * to build the verifier which contains a signed checksum
 * of the RPC header.
 */
int
nfs_gss_clnt_cred_put(struct nfsreq *req, struct nfsm_chain *nmc, mbuf_t args)
{
	struct nfs_gss_clnt_ctx *cp;
	uint32_t seqnum = 0;
	int error = 0;
	int slpflag, recordmark = 0;
	int start, len, offset = 0;
	int pad, toklen;
	struct nfsm_chain nmc_tmp;
	struct gss_seq *gsp;
	u_char tokbuf[KRB5_SZ_TOKMAX(MAX_DIGEST)];
	u_char cksum[MAX_DIGEST];
	struct timeval now;
	gss_key_info *ki;

	slpflag = (PZERO-1);
	if (req->r_nmp) {
		slpflag |= (NMFLAG(req->r_nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR)) ? PCATCH : 0;
		recordmark = (req->r_nmp->nm_sotype == SOCK_STREAM);
	}
retry:
	if (req->r_gss_ctx == NULL) {
		/*
		 * Find the context for this user.
		 * If no context is found, one will
		 * be created.
		 */
		error = nfs_gss_clnt_ctx_find(req);
		if (error)
			return (error);
	}
	cp = req->r_gss_ctx;

	/*
	 * If it's a dummy context for a user that's using
	 * a fallback to sec=sys, then just return an error
	 * so rpchead can encode an RPCAUTH_UNIX cred.
	 */
	if (cp->gss_clnt_service == RPCSEC_GSS_SVC_SYS) {
		/*
		 * The dummy context is valid for just
		 * GSS_CLNT_SYS_VALID seconds.  If the context
		 * is older than this, mark it invalid and try
		 * again to get a real one.
		 */
		lck_mtx_lock(cp->gss_clnt_mtx);
		microuptime(&now);
		if (now.tv_sec > cp->gss_clnt_ctime + GSS_CLNT_SYS_VALID) {
			cp->gss_clnt_flags |= GSS_CTX_INVAL;
			lck_mtx_unlock(cp->gss_clnt_mtx);
			nfs_gss_clnt_ctx_unref(req);
			goto retry;
		}
		lck_mtx_unlock(cp->gss_clnt_mtx);
		return (ENEEDAUTH);
	}

	/*
	 * If the context thread isn't null, then the context isn't
	 * yet complete and is for the exclusive use of the thread
	 * doing the context setup. Wait until the context thread
	 * is null.
	 */
	lck_mtx_lock(cp->gss_clnt_mtx);
	if (cp->gss_clnt_thread && cp->gss_clnt_thread != current_thread()) {
		cp->gss_clnt_flags |= GSS_NEEDCTX;
		msleep(cp, cp->gss_clnt_mtx, slpflag | PDROP, "ctxwait", NULL);
		slpflag &= ~PCATCH;
		if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0)))
			return (error);
		nfs_gss_clnt_ctx_unref(req);
		goto retry;
	}
	lck_mtx_unlock(cp->gss_clnt_mtx);

	ki = &cp->gss_clnt_kinfo;
	if (cp->gss_clnt_flags & GSS_CTX_COMPLETE) {
		/*
		 * Get a sequence number for this request.
		 * Check whether the oldest request in the window is complete.
		 * If it's still pending, then wait until it's done before
		 * we allocate a new sequence number and allow this request
		 * to proceed.
		 */
		lck_mtx_lock(cp->gss_clnt_mtx);
		while (win_getbit(cp->gss_clnt_seqbits,
			((cp->gss_clnt_seqnum - cp->gss_clnt_seqwin) + 1) % cp->gss_clnt_seqwin)) {
			cp->gss_clnt_flags |= GSS_NEEDSEQ;
			msleep(cp, cp->gss_clnt_mtx, slpflag | PDROP, "seqwin", NULL);
			slpflag &= ~PCATCH;
			if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0))) {
				return (error);
			}
			lck_mtx_lock(cp->gss_clnt_mtx);
			if (cp->gss_clnt_flags & GSS_CTX_INVAL) {
				/* Renewed while while we were waiting */
				lck_mtx_unlock(cp->gss_clnt_mtx);
				nfs_gss_clnt_ctx_unref(req);
				goto retry;
			}
		}
		seqnum = ++cp->gss_clnt_seqnum;
		win_setbit(cp->gss_clnt_seqbits, seqnum % cp->gss_clnt_seqwin);
		lck_mtx_unlock(cp->gss_clnt_mtx);

		MALLOC(gsp, struct gss_seq *, sizeof(*gsp), M_TEMP, M_WAITOK|M_ZERO);
		if (gsp == NULL)
			return (ENOMEM);
		gsp->gss_seqnum = seqnum;
		SLIST_INSERT_HEAD(&req->r_gss_seqlist, gsp, gss_seqnext);
	}

	/* Insert the credential */
	nfsm_chain_add_32(error, nmc, RPCSEC_GSS);
	nfsm_chain_add_32(error, nmc, 5 * NFSX_UNSIGNED + cp->gss_clnt_handle_len);
	nfsm_chain_add_32(error, nmc, RPCSEC_GSS_VERS_1);
	nfsm_chain_add_32(error, nmc, cp->gss_clnt_proc);
	nfsm_chain_add_32(error, nmc, seqnum);
	nfsm_chain_add_32(error, nmc, cp->gss_clnt_service);
	nfsm_chain_add_32(error, nmc, cp->gss_clnt_handle_len);
	if (cp->gss_clnt_handle_len > 0) {
	   	if (cp->gss_clnt_handle == NULL)
		  	return (EBADRPC);
		nfsm_chain_add_opaque(error, nmc, cp->gss_clnt_handle, cp->gss_clnt_handle_len);
	}
	if (error)
	    return(error);
	/*
	 * Now add the verifier
	 */
	if (cp->gss_clnt_proc == RPCSEC_GSS_INIT ||
		cp->gss_clnt_proc == RPCSEC_GSS_CONTINUE_INIT) {
		/*
		 * If the context is still being created
		 * then use a null verifier.
		 */
		nfsm_chain_add_32(error, nmc, RPCAUTH_NULL);	// flavor
		nfsm_chain_add_32(error, nmc, 0);		// length
		nfsm_chain_build_done(error, nmc);
		if (!error)
			nfs_gss_append_chain(nmc, args);
		return (error);
	}

	offset = recordmark ? NFSX_UNSIGNED : 0; // record mark
	nfsm_chain_build_done(error, nmc);
	nfs_gss_cksum_chain(ki, nmc, ALG_MIC(ki), offset, 0, cksum);

	toklen = nfs_gss_token_put(ki, ALG_MIC(ki), tokbuf, 1, 0, cksum);
	nfsm_chain_add_32(error, nmc, RPCSEC_GSS);	// flavor
	nfsm_chain_add_32(error, nmc, toklen);		// length
	nfsm_chain_add_opaque(error, nmc, tokbuf, toklen);
	nfsm_chain_build_done(error, nmc);
	if (error)
		return (error);

	/*
	 * Now we may have to compute integrity or encrypt the call args
	 * per RFC 2203 Section 5.3.2
	 */
	switch (cp->gss_clnt_service) {
	case RPCSEC_GSS_SVC_NONE:
		nfs_gss_append_chain(nmc, args);
		break;
	case RPCSEC_GSS_SVC_INTEGRITY:
		len = nfs_gss_mchain_length(args);	// Find args length
		req->r_gss_arglen = len;		// Stash the args len
		len += NFSX_UNSIGNED;			// Add seqnum length
		nfsm_chain_add_32(error, nmc, len);	// and insert it
		start = nfsm_chain_offset(nmc);
		nfsm_chain_add_32(error, nmc, seqnum);	// Insert seqnum
		req->r_gss_argoff = nfsm_chain_offset(nmc); // Offset to args
		nfsm_chain_build_done(error, nmc);
		if (error)
			return (error);
		nfs_gss_append_chain(nmc, args);	// Append the args mbufs

		/* Now compute a checksum over the seqnum + args */
		nfs_gss_cksum_chain(ki, nmc, ALG_MIC(ki), start, len, cksum);

		/* Insert it into a token and append to the request */
		toklen = nfs_gss_token_put(ki, ALG_MIC(ki), tokbuf, 1, 0, cksum);
		nfsm_chain_finish_mbuf(error, nmc);	// force checksum into new mbuf
		nfsm_chain_add_32(error, nmc, toklen);
		nfsm_chain_add_opaque(error, nmc, tokbuf, toklen);
		nfsm_chain_build_done(error, nmc);
		break;
	case RPCSEC_GSS_SVC_PRIVACY:
		/* Prepend a new mbuf with the confounder & sequence number */
		nfsm_chain_build_alloc_init(error, &nmc_tmp, 3 * NFSX_UNSIGNED);
		nfsm_chain_add_32(error, &nmc_tmp, random());	// confounder bytes 1-4
		nfsm_chain_add_32(error, &nmc_tmp, random());	// confounder bytes 4-8
		nfsm_chain_add_32(error, &nmc_tmp, seqnum);
		nfsm_chain_build_done(error, &nmc_tmp);
		if (error)
			return (error);
		nfs_gss_append_chain(&nmc_tmp, args);		// Append the args mbufs

		len = nfs_gss_mchain_length(args);		// Find args length
		len += 3 * NFSX_UNSIGNED;			// add confounder & seqnum
		req->r_gss_arglen = len;			// Stash length

		/*
		 * Append a pad trailer - per RFC 1964 section 1.2.2.3
		 * Since XDR data is always 32-bit aligned, it
		 * needs to be padded either by 4 bytes or 8 bytes.
		 */
		nfsm_chain_finish_mbuf(error, &nmc_tmp);	// force padding into new mbuf
		if (len % 8 > 0) {
			nfsm_chain_add_32(error, &nmc_tmp, 0x04040404);
			len += NFSX_UNSIGNED;
		} else {
			nfsm_chain_add_32(error, &nmc_tmp, 0x08080808);
			nfsm_chain_add_32(error, &nmc_tmp, 0x08080808);
			len +=  2 * NFSX_UNSIGNED;
		}
		nfsm_chain_build_done(error, &nmc_tmp);

		/* Now compute a checksum over the confounder + seqnum + args */
		nfs_gss_cksum_chain(ki, &nmc_tmp, ALG_WRAP(ki), 0, len, cksum);

		/* Insert it into a token */
		toklen = nfs_gss_token_put(ki, ALG_WRAP(ki), tokbuf, 1, len, cksum);
		nfsm_chain_add_32(error, nmc, toklen + len);	// token + args length
		nfsm_chain_add_opaque_nopad(error, nmc, tokbuf, toklen);
		req->r_gss_argoff = nfsm_chain_offset(nmc);	// Stash offset
		nfsm_chain_build_done(error, nmc);
		if (error)
			return (error);
		nfs_gss_append_chain(nmc, nmc_tmp.nmc_mhead);	// Append the args mbufs

		/* Finally, encrypt the args */
		nfs_gss_encrypt_chain(ki, &nmc_tmp, 0, len, DES_ENCRYPT);

		/* Add null XDR pad if the ASN.1 token misaligned the data */
		pad = nfsm_pad(toklen + len);
		if (pad > 0) {
			nfsm_chain_add_opaque_nopad(error, nmc, iv0, pad);
			nfsm_chain_build_done(error, nmc);
		}
		break;
	}

	return (error);
}

/*
 * When receiving a reply, the client checks the verifier
 * returned by the server. Check that the verifier is the
 * correct type, then extract the sequence number checksum
 * from the token in the credential and compare it with a
 * computed checksum of the sequence number in the request
 * that was sent.
 */
int
nfs_gss_clnt_verf_get(
	struct nfsreq *req,
	struct nfsm_chain *nmc,
	uint32_t verftype,
	uint32_t verflen,
	uint32_t *accepted_statusp)
{
	u_char tokbuf[KRB5_SZ_TOKMAX(MAX_DIGEST)];
	u_char cksum1[MAX_DIGEST], cksum2[MAX_DIGEST];
	uint32_t seqnum = 0;
	struct nfs_gss_clnt_ctx *cp = req->r_gss_ctx;
	struct nfsm_chain nmc_tmp;
	struct gss_seq *gsp;
	uint32_t reslen, start, cksumlen, toklen;
	int error = 0;
	gss_key_info *ki = &cp->gss_clnt_kinfo;

	reslen = cksumlen = 0;
	*accepted_statusp = 0;

	if (cp == NULL)
		return (NFSERR_EAUTH);
	/*
	 * If it's not an RPCSEC_GSS verifier, then it has to
	 * be a null verifier that resulted from either
	 * a CONTINUE_NEEDED reply during context setup or
	 * from the reply to an AUTH_UNIX call from a dummy
	 * context that resulted from a fallback to sec=sys.
	 */
	if (verftype != RPCSEC_GSS) {
		if (verftype != RPCAUTH_NULL)
			return (NFSERR_EAUTH);
		if (cp->gss_clnt_flags & GSS_CTX_COMPLETE &&
			cp->gss_clnt_service != RPCSEC_GSS_SVC_SYS)
			return (NFSERR_EAUTH);
		if (verflen > 0)
			nfsm_chain_adv(error, nmc, nfsm_rndup(verflen));
		nfsm_chain_get_32(error, nmc, *accepted_statusp);
		return (error);
	}

	/*
	 * If we received an RPCSEC_GSS verifier but the
	 * context isn't yet complete, then it must be
	 * the context complete message from the server.
	 * The verifier will contain an encrypted checksum
	 * of the window but we don't have the session key
	 * yet so we can't decrypt it. Stash the verifier
	 * and check it later in nfs_gss_clnt_ctx_init() when
	 * the context is complete.
	 */
	if (!(cp->gss_clnt_flags & GSS_CTX_COMPLETE)) {
		MALLOC(cp->gss_clnt_verf, u_char *, verflen, M_TEMP, M_WAITOK|M_ZERO);
		if (cp->gss_clnt_verf == NULL)
			return (ENOMEM);
		nfsm_chain_get_opaque(error, nmc, verflen, cp->gss_clnt_verf);
		nfsm_chain_get_32(error, nmc, *accepted_statusp);
		return (error);
	}

	if (verflen != KRB5_SZ_TOKEN(ki->hash_len))
		return (NFSERR_EAUTH);

	/*
	 * Get the 8 octet sequence number
	 * checksum out of the verifier token.
	 */
	nfsm_chain_get_opaque(error, nmc, verflen, tokbuf);
	if (error)
		goto nfsmout;
	error = nfs_gss_token_get(ki, ALG_MIC(ki), tokbuf, 0, NULL, cksum1);
	if (error)
		goto nfsmout;

	/*
	 * Search the request sequence numbers for this reply, starting
	 * with the most recent, looking for a checksum that matches
	 * the one in the verifier returned by the server.
	 */
	SLIST_FOREACH(gsp, &req->r_gss_seqlist, gss_seqnext) {
		nfs_gss_cksum_rep(ki, gsp->gss_seqnum, cksum2);
		if (bcmp(cksum1, cksum2, HASHLEN(ki)) == 0)
			break;
	}
	if (gsp == NULL)
		return (NFSERR_EAUTH);

	/*
	 * Get the RPC accepted status
	 */
	nfsm_chain_get_32(error, nmc, *accepted_statusp);
	if (*accepted_statusp != RPC_SUCCESS)
		return (0);

	/*
	 * Now we may have to check integrity or decrypt the results
	 * per RFC 2203 Section 5.3.2
	 */
	switch (cp->gss_clnt_service) {
	case RPCSEC_GSS_SVC_NONE:
		/* nothing to do */
		break;
	case RPCSEC_GSS_SVC_INTEGRITY:
		/*
		 * Here's what we expect in the integrity results:
		 *
		 * - length of seq num + results (4 bytes)
		 * - sequence number (4 bytes)
		 * - results (variable bytes)
		 * - length of checksum token (37)
		 * - checksum of seqnum + results (37 bytes)
		 */
		nfsm_chain_get_32(error, nmc, reslen);		// length of results
		if (reslen > NFS_MAXPACKET) {
			error = EBADRPC;
			goto nfsmout;
		}

		/* Compute a checksum over the sequence number + results */
		start = nfsm_chain_offset(nmc);
		nfs_gss_cksum_chain(ki, nmc, ALG_MIC(ki), start, reslen, cksum1);

		/*
		 * Get the sequence number prepended to the results
		 * and compare it against the list in the request.
		 */
		nfsm_chain_get_32(error, nmc, seqnum);
		SLIST_FOREACH(gsp, &req->r_gss_seqlist, gss_seqnext) {
			if (seqnum == gsp->gss_seqnum)
				break;
		}
		if (gsp == NULL) {
			error = EBADRPC;
			goto nfsmout;
		}

		/*
		 * Advance to the end of the results and
		 * fetch the checksum computed by the server.
		 */
		nmc_tmp = *nmc;
		reslen -= NFSX_UNSIGNED;			// already skipped seqnum
		nfsm_chain_adv(error, &nmc_tmp, reslen);	// skip over the results
		nfsm_chain_get_32(error, &nmc_tmp, cksumlen);	// length of checksum
		if (cksumlen != KRB5_SZ_TOKEN(ki->hash_len)) {
			error = EBADRPC;
			goto nfsmout;
		}
		nfsm_chain_get_opaque(error, &nmc_tmp, cksumlen, tokbuf);
		if (error)
			goto nfsmout;
		error = nfs_gss_token_get(ki, ALG_MIC(ki), tokbuf, 0, NULL, cksum2);
		if (error)
			goto nfsmout;

		/* Verify that the checksums are the same */
		if (bcmp(cksum1, cksum2, HASHLEN(ki)) != 0) {
			error = EBADRPC;
			goto nfsmout;
		}
		break;
	case RPCSEC_GSS_SVC_PRIVACY:
		/*
		 * Here's what we expect in the privacy results:
		 *
		 * - length of confounder + seq num + token + results
		 * - wrap token (37-40 bytes)
		 * - confounder (8 bytes)
		 * - sequence number (4 bytes)
		 * - results (encrypted)
		 */
		nfsm_chain_get_32(error, nmc, reslen);		// length of results
		if (reslen > NFS_MAXPACKET) {
			error = EBADRPC;
			goto nfsmout;
		}

		/* Get the token that prepends the encrypted results */
		nfsm_chain_get_opaque(error, nmc, KRB5_SZ_TOKMAX(ki->hash_len), tokbuf);
		if (error)
			goto nfsmout;
		error = nfs_gss_token_get(ki, ALG_WRAP(ki), tokbuf, 0,
			&toklen, cksum1);
		if (error)
			goto nfsmout;
		nfsm_chain_reverse(nmc, nfsm_pad(toklen));
		reslen -= toklen;				// size of confounder + seqnum + results

		/* decrypt the confounder + sequence number + results */
		start = nfsm_chain_offset(nmc);
		nfs_gss_encrypt_chain(ki, nmc, start, reslen, DES_DECRYPT);

		/* Compute a checksum over the confounder + sequence number + results */
		nfs_gss_cksum_chain(ki, nmc, ALG_WRAP(ki), start, reslen, cksum2);

		/* Verify that the checksums are the same */
		if (bcmp(cksum1, cksum2, HASHLEN(ki)) != 0) {
			error = EBADRPC;
			goto nfsmout;
		}

		nfsm_chain_adv(error, nmc, 8);	// skip over the confounder

		/*
		 * Get the sequence number prepended to the results
		 * and compare it against the list in the request.
		 */
		nfsm_chain_get_32(error, nmc, seqnum);
		SLIST_FOREACH(gsp, &req->r_gss_seqlist, gss_seqnext) {
			if (seqnum == gsp->gss_seqnum)
				break;
		}
		if (gsp == NULL) {
			error = EBADRPC;
			goto nfsmout;
		}

		break;
	}
nfsmout:
	return (error);
}

/*
 * An RPCSEC_GSS request with no integrity or privacy consists
 * of just the header mbufs followed by the arg mbufs.
 *
 * However, integrity or privacy both trailer mbufs to the args,
 * which means we have to do some work to restore the arg mbuf
 * chain to its previous state in case we need to retransmit.
 *
 * The location and length of the args is marked by two fields
 * in the request structure: r_gss_argoff and r_gss_arglen,
 * which are stashed when the NFS request is built.
 */
int
nfs_gss_clnt_args_restore(struct nfsreq *req)
{
	struct nfs_gss_clnt_ctx *cp = req->r_gss_ctx;
	struct nfsm_chain mchain, *nmc = &mchain;
	int len, error = 0;

	if (cp == NULL)
		return (NFSERR_EAUTH);

	if ((cp->gss_clnt_flags & GSS_CTX_COMPLETE) == 0)
		return (ENEEDAUTH);

	nfsm_chain_dissect_init(error, nmc, req->r_mhead);	// start at RPC header
	nfsm_chain_adv(error, nmc, req->r_gss_argoff);		// advance to args
	if (error)
		return (error);

	switch (cp->gss_clnt_service) {
	case RPCSEC_GSS_SVC_NONE:
		/* nothing to do */
		break;
	case RPCSEC_GSS_SVC_INTEGRITY:
		/*
		 * All we have to do here is remove the appended checksum mbufs.
		 * We know that the checksum starts in a new mbuf beyond the end
		 * of the args.
		 */
		nfsm_chain_adv(error, nmc, req->r_gss_arglen);	// adv to last args mbuf
		if (error)
			return (error);

		mbuf_freem(mbuf_next(nmc->nmc_mcur));		// free the cksum mbuf
		error = mbuf_setnext(nmc->nmc_mcur, NULL);
		break;
	case RPCSEC_GSS_SVC_PRIVACY:
		/*
		 * The args are encrypted along with prepended confounders and seqnum.
		 * First we decrypt, the confounder, seqnum and args then skip to the
		 * final mbuf of the args.
		 * The arglen includes 8 bytes of confounder and 4 bytes of seqnum.
		 * Finally, we remove between 4 and 8 bytes of encryption padding
		 * as well as any alignment padding in the trailing mbuf.
		 */
		len = req->r_gss_arglen;
		len += len % 8 > 0 ? 4 : 8;			// add DES padding length
		nfs_gss_encrypt_chain(&cp->gss_clnt_kinfo, nmc,
					req->r_gss_argoff, len, DES_DECRYPT);
		nfsm_chain_adv(error, nmc, req->r_gss_arglen);
		if (error)
			return (error);
		mbuf_freem(mbuf_next(nmc->nmc_mcur));		// free the pad mbuf
		error = mbuf_setnext(nmc->nmc_mcur, NULL);
		break;
	}

	return (error);
}

/*
 * This function sets up  a new context on the client.
 * Context setup alternates upcalls to the gssd with NFS nullproc calls
 * to the server.  Each of these calls exchanges an opaque token, obtained
 * via the gssd's calls into the GSS-API on either the client or the server.
 * This cycle of calls ends when the client's upcall to the gssd and the
 * server's response both return GSS_S_COMPLETE.  At this point, the client
 * should have its session key and a handle that it can use to refer to its
 * new context on the server.
 */
static int
nfs_gss_clnt_ctx_init(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
{
	struct nfsmount *nmp = req->r_nmp;
	int client_complete = 0;
	int server_complete = 0;
	u_char cksum1[MAX_DIGEST], cksum2[MAX_DIGEST];
	int error = 0;
	struct timeval now;
	gss_key_info *ki = &cp->gss_clnt_kinfo;

	/* Initialize a new client context */

	cp->gss_clnt_svcname = nfs_gss_clnt_svcname(nmp, &cp->gss_clnt_svcnt, &cp->gss_clnt_svcnamlen);
	if (cp->gss_clnt_svcname == NULL) {
		error = NFSERR_EAUTH;
		goto nfsmout;
	}

	cp->gss_clnt_proc = RPCSEC_GSS_INIT;

	cp->gss_clnt_service =
		req->r_auth == RPCAUTH_KRB5  ? RPCSEC_GSS_SVC_NONE :
		req->r_auth == RPCAUTH_KRB5I ? RPCSEC_GSS_SVC_INTEGRITY :
		req->r_auth == RPCAUTH_KRB5P ? RPCSEC_GSS_SVC_PRIVACY : 0;

	cp->gss_clnt_gssd_flags = (nfs_single_des ? GSSD_NFS_1DES : 0);
	/*
	 * Now loop around alternating gss_init_sec_context and
	 * gss_accept_sec_context upcalls to the gssd on the client
	 * and server side until the context is complete - or fails.
	 */
	for (;;) {

retry:
		/* Upcall to the gss_init_sec_context in the gssd */
		error = nfs_gss_clnt_gssd_upcall(req, cp);
		if (error)
			goto nfsmout;

		if (cp->gss_clnt_major == GSS_S_COMPLETE) {
			client_complete = 1;
			if (server_complete)
				break;
		} else if (cp->gss_clnt_major != GSS_S_CONTINUE_NEEDED) {
			error = NFSERR_EAUTH;
			goto nfsmout;
		}

		/*
		 * Pass the token to the server.
		 */
		error = nfs_gss_clnt_ctx_callserver(req, cp);
		if (error) {
			if (error == ENEEDAUTH && cp->gss_clnt_proc == RPCSEC_GSS_INIT &&
				(cp->gss_clnt_gssd_flags & (GSSD_RESTART | GSSD_NFS_1DES)) == 0) {
				NFS_GSS_DBG("Retrying with single DES for req %p\n", req);
				cp->gss_clnt_gssd_flags = (GSSD_RESTART | GSSD_NFS_1DES);
				if (cp->gss_clnt_token)
					FREE(cp->gss_clnt_token, M_TEMP);
				cp->gss_clnt_token = NULL;
				cp->gss_clnt_tokenlen = 0;
				goto retry;
			}
			// Reset flags, if error = ENEEDAUTH we will try 3des again
			cp->gss_clnt_gssd_flags = 0;
			goto nfsmout;
		}
		if (cp->gss_clnt_major == GSS_S_COMPLETE) {
			server_complete = 1;
			if (client_complete)
				break;
		} else if (cp->gss_clnt_major != GSS_S_CONTINUE_NEEDED) {
			error = NFSERR_EAUTH;
			goto nfsmout;
		}

		cp->gss_clnt_proc = RPCSEC_GSS_CONTINUE_INIT;
	}

	/*
	 * The context is apparently established successfully
	 */
	lck_mtx_lock(cp->gss_clnt_mtx);
	cp->gss_clnt_flags |= GSS_CTX_COMPLETE;
	lck_mtx_unlock(cp->gss_clnt_mtx);
	cp->gss_clnt_proc = RPCSEC_GSS_DATA;
	microuptime(&now);
	cp->gss_clnt_ctime = now.tv_sec;	// time stamp


	/*
	 * Compute checksum of the server's window
	 */
	nfs_gss_cksum_rep(ki, cp->gss_clnt_seqwin, cksum1);

	/*
	 * and see if it matches the one in the
	 * verifier the server returned.
	 */
	error = nfs_gss_token_get(ki, ALG_MIC(ki), cp->gss_clnt_verf, 0,
		NULL, cksum2);
	FREE(cp->gss_clnt_verf, M_TEMP);
	cp->gss_clnt_verf = NULL;

	if (error || bcmp(cksum1, cksum2, HASHLEN(ki)) != 0) {
		error = NFSERR_EAUTH;
		goto nfsmout;
	}

	/*
	 * Set an initial sequence number somewhat randomized.
	 * Start small so we don't overflow GSS_MAXSEQ too quickly.
	 * Add the size of the sequence window so seqbits arithmetic
	 * doesn't go negative.
	 */
	cp->gss_clnt_seqnum = (random() & 0xffff) + cp->gss_clnt_seqwin;

	/*
	 * Allocate a bitmap to keep track of which requests
	 * are pending within the sequence number window.
	 */
	MALLOC(cp->gss_clnt_seqbits, uint32_t *,
		nfsm_rndup((cp->gss_clnt_seqwin + 7) / 8), M_TEMP, M_WAITOK|M_ZERO);
	if (cp->gss_clnt_seqbits == NULL)
		error = NFSERR_EAUTH;
nfsmout:
 	/*
	 * If the error is ENEEDAUTH we're not done, so no need
	 * to wake up other threads again. This thread will retry in
	 * the find or renew routines.
	 */
	if (error == ENEEDAUTH)
		return (error);

	/*
	 * If there's an error, just mark it as invalid.
	 * It will be removed when the reference count
	 * drops to zero.
	 */
	lck_mtx_lock(cp->gss_clnt_mtx);
	if (error)
		cp->gss_clnt_flags |= GSS_CTX_INVAL;

	/*
	 * Wake any threads waiting to use the context
	 */
	cp->gss_clnt_thread = NULL;
	if (cp->gss_clnt_flags & GSS_NEEDCTX) {
		cp->gss_clnt_flags &= ~GSS_NEEDCTX;
		wakeup(cp);
	}
	lck_mtx_unlock(cp->gss_clnt_mtx);

	return (error);
}

/*
 * This function calls nfs_gss_clnt_ctx_init() to set up a new context.
 * But if there's a failure in trying to establish the context it keeps
 * retrying at progressively longer intervals in case the failure is
 * due to some transient condition.  For instance, the server might be
 * failing the context setup because directory services is not coming
 * up in a timely fashion.
 */
static int
nfs_gss_clnt_ctx_init_retry(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
{
	struct nfsmount *nmp = req->r_nmp;
	struct timeval now;
	time_t waituntil;
	int error, slpflag;
	int retries = 0;
	int timeo = NFS_TRYLATERDEL;

	if (nmp == NULL) {
		error = ENXIO;
		goto bad;
	}

	/* For an "intr" mount allow a signal to interrupt the retries */
	slpflag = (NMFLAG(nmp, INTR) && !(req->r_flags & R_NOINTR)) ? PCATCH : 0;

	while ((error = nfs_gss_clnt_ctx_init(req, cp)) == ENEEDAUTH) {
		microuptime(&now);
		waituntil = now.tv_sec + timeo;
		while (now.tv_sec < waituntil) {
			tsleep(NULL, PSOCK | slpflag, "nfs_gss_clnt_ctx_init_retry", hz);
			slpflag = 0;
			error = nfs_sigintr(req->r_nmp, req, current_thread(), 0);
			if (error)
				goto bad;
			microuptime(&now);
		}

		retries++;
		/* If it's a soft mount just give up after a while */
		if (NMFLAG(nmp, SOFT) && (retries > nmp->nm_retry)) {
			error = ETIMEDOUT;
			goto bad;
		}
		timeo *= 2;
		if (timeo > 60)
			timeo = 60;
	}

	if (error == 0)
		return 0;	// success
bad:
	/*
	 * Give up on this context
	 */
	lck_mtx_lock(cp->gss_clnt_mtx);
	cp->gss_clnt_flags |= GSS_CTX_INVAL;

	/*
	 * Wake any threads waiting to use the context
	 */
	cp->gss_clnt_thread = NULL;
	if (cp->gss_clnt_flags & GSS_NEEDCTX) {
		cp->gss_clnt_flags &= ~GSS_NEEDCTX;
		wakeup(cp);
	}
	lck_mtx_unlock(cp->gss_clnt_mtx);

	return error;
}

/*
 * Call the NFS server using a null procedure for context setup.
 * Even though it's a null procedure and nominally has no arguments
 * RFC 2203 requires that the GSS-API token be passed as an argument
 * and received as a reply.
 */
static int
nfs_gss_clnt_ctx_callserver(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
{
	struct nfsm_chain nmreq, nmrep;
	int error = 0, status;
	int sz;

	if (!req->r_nmp)
		return (ENXIO);
	nfsm_chain_null(&nmreq);
	nfsm_chain_null(&nmrep);
	sz = NFSX_UNSIGNED + nfsm_rndup(cp->gss_clnt_tokenlen);
	nfsm_chain_build_alloc_init(error, &nmreq, sz);
	nfsm_chain_add_32(error, &nmreq, cp->gss_clnt_tokenlen);
	if (cp->gss_clnt_tokenlen > 0)
		nfsm_chain_add_opaque(error, &nmreq, cp->gss_clnt_token, cp->gss_clnt_tokenlen);
	nfsm_chain_build_done(error, &nmreq);
	if (error)
		goto nfsmout;

	/* Call the server */
	error = nfs_request_gss(req->r_nmp->nm_mountp, &nmreq, req->r_thread, req->r_cred,
				(req->r_flags & R_OPTMASK), cp, &nmrep, &status);
	if (cp->gss_clnt_token != NULL) {
		FREE(cp->gss_clnt_token, M_TEMP);
		cp->gss_clnt_token = NULL;
	}
	if (!error)
		error = status;
	if (error)
		goto nfsmout;

	/* Get the server's reply */

	nfsm_chain_get_32(error, &nmrep, cp->gss_clnt_handle_len);
	if (cp->gss_clnt_handle != NULL) {
		FREE(cp->gss_clnt_handle, M_TEMP);
		cp->gss_clnt_handle = NULL;
	}
	if (cp->gss_clnt_handle_len > 0) {
		MALLOC(cp->gss_clnt_handle, u_char *, cp->gss_clnt_handle_len, M_TEMP, M_WAITOK);
		if (cp->gss_clnt_handle == NULL) {
			error = ENOMEM;
			goto nfsmout;
		}
		nfsm_chain_get_opaque(error, &nmrep, cp->gss_clnt_handle_len, cp->gss_clnt_handle);
	}
	nfsm_chain_get_32(error, &nmrep, cp->gss_clnt_major);
	nfsm_chain_get_32(error, &nmrep, cp->gss_clnt_minor);
	nfsm_chain_get_32(error, &nmrep, cp->gss_clnt_seqwin);
	nfsm_chain_get_32(error, &nmrep, cp->gss_clnt_tokenlen);
	if (error)
		goto nfsmout;
	if (cp->gss_clnt_tokenlen > 0) {
		MALLOC(cp->gss_clnt_token, u_char *, cp->gss_clnt_tokenlen, M_TEMP, M_WAITOK);
		if (cp->gss_clnt_token == NULL) {
			error = ENOMEM;
			goto nfsmout;
		}
		nfsm_chain_get_opaque(error, &nmrep, cp->gss_clnt_tokenlen, cp->gss_clnt_token);
	}

	/*
	 * Make sure any unusual errors are expanded and logged by gssd
	 */
	if (cp->gss_clnt_major != GSS_S_COMPLETE &&
	    cp->gss_clnt_major != GSS_S_CONTINUE_NEEDED) {
		char who[] = "server";
		char unknown[] = "<unknown>";

		(void) mach_gss_log_error(
			cp->gss_clnt_mport,
			!req->r_nmp ? unknown :
			vfs_statfs(req->r_nmp->nm_mountp)->f_mntfromname,
			kauth_cred_getuid(cp->gss_clnt_cred),
			who,
			cp->gss_clnt_major,
			cp->gss_clnt_minor);
	}

nfsmout:
	nfsm_chain_cleanup(&nmreq);
	nfsm_chain_cleanup(&nmrep);

	return (error);
}

/*
 * We construct the service principal as a gss hostbased service principal of
 * the form nfs@<server>, unless the servers principal was passed down in the
 * mount arguments. If the arguments don't specify the service principal, the
 * server name is extracted the location passed in the mount argument if
 * available.  Otherwise assume a format of <server>:<path> in the
 * mntfromname. We don't currently support url's or other bizarre formats like
 * path@server. Mount_url will convert the nfs url into <server>:<path> when
 * calling mount, so this works out well in practice.
 *
 */

static uint8_t *
nfs_gss_clnt_svcname(struct nfsmount *nmp, gssd_nametype *nt, uint32_t *len)
{
	char *svcname, *d, *server;
	int lindx, sindx;

	if (!nmp)
		return (NULL);

	if (nmp->nm_sprinc) {
		*len = strlen(nmp->nm_sprinc) + 1;
		MALLOC(svcname, char *, *len, M_TEMP, M_WAITOK);
		*nt = GSSD_HOSTBASED;
		if (svcname == NULL)
			return (NULL);
		strlcpy(svcname, nmp->nm_sprinc, *len);

		return ((uint8_t *)svcname);
	}

	*nt = GSSD_HOSTBASED;
	if (nmp->nm_locations.nl_numlocs && !(NFS_GSS_ISDBG && (NFS_DEBUG_FLAGS & 0x1))) {
		lindx = nmp->nm_locations.nl_current.nli_loc;
		sindx = nmp->nm_locations.nl_current.nli_serv;
		server = nmp->nm_locations.nl_locations[lindx]->nl_servers[sindx]->ns_name;
		*len = (uint32_t)strlen(server);
	} else {
		/* Older binaries using older mount args end up here */
		server = vfs_statfs(nmp->nm_mountp)->f_mntfromname;
		NFS_GSS_DBG("nfs getting gss svcname from %s\n", server);
		d = strchr(server, ':');
		*len = (uint32_t)(d ? (d - server) : strlen(server));
	}

	*len +=  5; /* "nfs@" plus null */
	MALLOC(svcname, char *, *len, M_TEMP, M_WAITOK);
	strlcpy(svcname, "nfs", *len);
	strlcat(svcname, "@", *len);
	strlcat(svcname, server, *len);
	NFS_GSS_DBG("nfs svcname = %s\n", svcname);

	return ((uint8_t *)svcname);
}

/*
 * Get a mach port to talk to gssd.
 * gssd lives in the root bootstrap, so we call gssd's lookup routine
 * to get a send right to talk to a new gssd instance that launchd has launched
 * based on the cred's uid and audit session id.
 */

static mach_port_t
nfs_gss_clnt_get_upcall_port(kauth_cred_t credp)
{
	mach_port_t gssd_host_port, uc_port = IPC_PORT_NULL;
	kern_return_t kr;
	au_asid_t asid;
	uid_t uid;

	kr = host_get_gssd_port(host_priv_self(), &gssd_host_port);
	if (kr != KERN_SUCCESS) {
		printf("nfs_gss_get_upcall_port: can't get gssd port, status %x (%d)\n", kr, kr);
		return (IPC_PORT_NULL);
	}
	if (!IPC_PORT_VALID(gssd_host_port)) {
		printf("nfs_gss_get_upcall_port: gssd port not valid\n");
		return (IPC_PORT_NULL);
	}

	asid = kauth_cred_getasid(credp);
	uid = kauth_cred_getauid(credp);
	if (uid == AU_DEFAUDITID)
		uid = kauth_cred_getuid(credp);
	kr = mach_gss_lookup(gssd_host_port, uid, asid, &uc_port);
	if (kr != KERN_SUCCESS)
		printf("nfs_gss_clnt_get_upcall_port: mach_gssd_lookup failed: status %x (%d)\n", kr, kr);

	return (uc_port);
}

/*
 * Make an upcall to the gssd using Mach RPC
 * The upcall is made using a host special port.
 * This allows launchd to fire up the gssd in the
 * user's session.  This is important, since gssd
 * must have access to the user's credential cache.
 */
static int
nfs_gss_clnt_gssd_upcall(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
{
	kern_return_t kr;
	gssd_byte_buffer okey = NULL;
	uint32_t skeylen = 0;
	int retry_cnt = 0;
	vm_map_copy_t itoken = NULL;
	gssd_byte_buffer otoken = NULL;
	mach_msg_type_number_t otokenlen;
	int error = 0;
	uint8_t *principal = NULL;
	uint32_t plen = 0;
	int32_t nt = GSSD_STRING_NAME;
	vm_map_copy_t pname = NULL;
	vm_map_copy_t svcname = NULL;
	char display_name[MAX_DISPLAY_STR] = "";
	uint32_t ret_flags;
	uint32_t nfs_1des = (cp->gss_clnt_gssd_flags & GSSD_NFS_1DES);
	struct nfsmount *nmp;

	/*
	 * NFS currently only supports default principals or
	 * principals based on the uid of the caller, unless
	 * the principal to use for the mounting cred was specified
	 * in the mount argmuments. If the realm to use was specified
	 * then will send that up as the principal since the realm is
	 * preceed by an "@" gssd that will try and select the default
	 * principal for that realm.
	 */

	nmp = req->r_nmp;
	if (nmp == NULL || vfs_isforce(nmp->nm_mountp) || (nmp->nm_state & (NFSSTA_FORCE | NFSSTA_DEAD)))
		return (ENXIO);

	if (cp->gss_clnt_principal && cp->gss_clnt_prinlen) {
		principal = cp->gss_clnt_principal;
		plen = cp->gss_clnt_prinlen;
		nt = cp->gss_clnt_prinnt;
	} else if (nmp->nm_principal && IS_VALID_CRED(nmp->nm_mcred) && req->r_cred == nmp->nm_mcred) {
		plen = (uint32_t)strlen(nmp->nm_principal);
		MALLOC(principal, uint8_t *, plen, M_TEMP, M_WAITOK | M_ZERO);
		if (principal == NULL)
			return (ENOMEM);
		bcopy(nmp->nm_principal, principal, plen);
		cp->gss_clnt_prinnt = nt = GSSD_USER;
	}
	else if (nmp->nm_realm) {
		plen = (uint32_t)strlen(nmp->nm_realm);
		principal = (uint8_t *)nmp->nm_realm;
		nt = GSSD_USER;
	}

	if (!IPC_PORT_VALID(cp->gss_clnt_mport)) {
		cp->gss_clnt_mport = nfs_gss_clnt_get_upcall_port(req->r_cred);
		if (cp->gss_clnt_mport == IPC_PORT_NULL)
			goto out;
	}

	if (plen)
		nfs_gss_mach_alloc_buffer(principal, plen, &pname);
	if (cp->gss_clnt_svcnamlen)
		nfs_gss_mach_alloc_buffer(cp->gss_clnt_svcname, cp->gss_clnt_svcnamlen, &svcname);
	if (cp->gss_clnt_tokenlen)
		nfs_gss_mach_alloc_buffer(cp->gss_clnt_token, cp->gss_clnt_tokenlen, &itoken);

retry:
	kr = mach_gss_init_sec_context_v2(
		cp->gss_clnt_mport,
		GSSD_KRB5_MECH,
		(gssd_byte_buffer) itoken, (mach_msg_type_number_t) cp->gss_clnt_tokenlen,
		kauth_cred_getuid(cp->gss_clnt_cred),
		nt,
		(gssd_byte_buffer)pname, (mach_msg_type_number_t) plen,
		cp->gss_clnt_svcnt,
		(gssd_byte_buffer)svcname, (mach_msg_type_number_t) cp->gss_clnt_svcnamlen,
		GSSD_MUTUAL_FLAG,
		&cp->gss_clnt_gssd_flags,
		&cp->gss_clnt_context,
		&cp->gss_clnt_cred_handle,
		&ret_flags,
		&okey,  (mach_msg_type_number_t *) &skeylen,
		&otoken, &otokenlen,
		cp->gss_clnt_display ? NULL : display_name,
		&cp->gss_clnt_major,
		&cp->gss_clnt_minor);

	/* Should be cleared and set in gssd ? */
	cp->gss_clnt_gssd_flags &= ~GSSD_RESTART;
	cp->gss_clnt_gssd_flags |= nfs_1des;

	if (kr != KERN_SUCCESS) {
		printf("nfs_gss_clnt_gssd_upcall: mach_gss_init_sec_context failed: %x (%d)\n", kr, kr);
		if (kr == MIG_SERVER_DIED && cp->gss_clnt_cred_handle == 0 &&
			retry_cnt++ < NFS_GSS_MACH_MAX_RETRIES &&
			!vfs_isforce(nmp->nm_mountp) && (nmp->nm_state & (NFSSTA_FORCE | NFSSTA_DEAD)) == 0) {
			if (plen)
				nfs_gss_mach_alloc_buffer(principal, plen, &pname);
			if (cp->gss_clnt_svcnamlen)
				nfs_gss_mach_alloc_buffer(cp->gss_clnt_svcname, cp->gss_clnt_svcnamlen, &svcname);
			if (cp->gss_clnt_tokenlen > 0)
				nfs_gss_mach_alloc_buffer(cp->gss_clnt_token, cp->gss_clnt_tokenlen, &itoken);
			goto retry;
		}

		host_release_special_port(cp->gss_clnt_mport);
		cp->gss_clnt_mport = IPC_PORT_NULL;
		goto out;
	}

	if (cp->gss_clnt_display == NULL && *display_name != '\0') {
		int dlen = strnlen(display_name, MAX_DISPLAY_STR) + 1;  /* Add extra byte to include '\0' */

		if (dlen < MAX_DISPLAY_STR) {
			MALLOC(cp->gss_clnt_display, char *, dlen, M_TEMP, M_WAITOK);
			if (cp->gss_clnt_display == NULL)
				goto skip;
			bcopy(display_name, cp->gss_clnt_display, dlen);
		} else {
			goto skip;
		}
	}
skip:
	/*
	 * Make sure any unusual errors are expanded and logged by gssd
	 *
	 * XXXX, we need to rethink this and just have gssd return a string for the major and minor codes.
	 */
	if (cp->gss_clnt_major != GSS_S_COMPLETE &&
	    cp->gss_clnt_major != GSS_S_CONTINUE_NEEDED) {
#define GETMAJERROR(x) (((x) >> GSS_C_ROUTINE_ERROR_OFFSET) & GSS_C_ROUTINE_ERROR_MASK)
		char who[] = "client";
		uint32_t gss_error = GETMAJERROR(cp->gss_clnt_major);

		(void) mach_gss_log_error(
			cp->gss_clnt_mport,
			vfs_statfs(nmp->nm_mountp)->f_mntfromname,
			kauth_cred_getuid(cp->gss_clnt_cred),
			who,
			cp->gss_clnt_major,
			cp->gss_clnt_minor);
		gss_error = gss_error ? gss_error : cp->gss_clnt_major;
		printf("NFS gssd auth failure mount %s for %s major = %d minor = %d\n",
		       vfs_statfs(nmp->nm_mountp)->f_mntfromname, cp->gss_clnt_display ? cp->gss_clnt_display : who,
		       gss_error, (int32_t)cp->gss_clnt_minor);
	}

	if (skeylen > 0) {
		if (skeylen != SKEYLEN && skeylen != SKEYLEN3) {
			printf("nfs_gss_clnt_gssd_upcall: bad key length (%d)\n", skeylen);
			vm_map_copy_discard((vm_map_copy_t) okey);
			vm_map_copy_discard((vm_map_copy_t) otoken);
			goto out;
		}
		error = nfs_gss_mach_vmcopyout((vm_map_copy_t) okey, skeylen,
				cp->gss_clnt_kinfo.skey);
		if (error) {
			vm_map_copy_discard((vm_map_copy_t) otoken);
			goto out;
		}

		error = gss_key_init(&cp->gss_clnt_kinfo, skeylen);
		if (error)
			goto out;
	}

	/* Free context token used as input */
	if (cp->gss_clnt_token)
		FREE(cp->gss_clnt_token, M_TEMP);
	cp->gss_clnt_token = NULL;
	cp->gss_clnt_tokenlen = 0;

	if (otokenlen > 0) {
		/* Set context token to gss output token */
		MALLOC(cp->gss_clnt_token, u_char *, otokenlen, M_TEMP, M_WAITOK);
		if (cp->gss_clnt_token == NULL) {
			printf("nfs_gss_clnt_gssd_upcall: could not allocate %d bytes\n", otokenlen);
			vm_map_copy_discard((vm_map_copy_t) otoken);
			return (ENOMEM);
		}
		error = nfs_gss_mach_vmcopyout((vm_map_copy_t) otoken, otokenlen, cp->gss_clnt_token);
		if (error) {
			FREE(cp->gss_clnt_token, M_TEMP);
			cp->gss_clnt_token = NULL;
			return (NFSERR_EAUTH);
		}
		cp->gss_clnt_tokenlen = otokenlen;
	}

	return (0);

out:
	if (cp->gss_clnt_token)
		FREE(cp->gss_clnt_token, M_TEMP);
	cp->gss_clnt_token = NULL;
	cp->gss_clnt_tokenlen = 0;

	return (NFSERR_EAUTH);
}

/*
 * Invoked at the completion of an RPC call that uses an RPCSEC_GSS
 * credential. The sequence number window that the server returns
 * at context setup indicates the maximum number of client calls that
 * can be outstanding on a context. The client maintains a bitmap that
 * represents the server's window.  Each pending request has a bit set
 * in the window bitmap.  When a reply comes in or times out, we reset
 * the bit in the bitmap and if there are any other threads waiting for
 * a context slot we notify the waiting thread(s).
 *
 * Note that if a request is retransmitted, it will have a single XID
 * but it may be associated with multiple sequence numbers.  So we
 * may have to reset multiple sequence number bits in the window bitmap.
 */
void
nfs_gss_clnt_rpcdone(struct nfsreq *req)
{
	struct nfs_gss_clnt_ctx *cp = req->r_gss_ctx;
	struct gss_seq *gsp, *ngsp;
	int i = 0;

	if (cp == NULL || !(cp->gss_clnt_flags & GSS_CTX_COMPLETE))
		return;	// no context - don't bother
	/*
	 * Reset the bit for this request in the
	 * sequence number window to indicate it's done.
	 * We do this even if the request timed out.
	 */
	lck_mtx_lock(cp->gss_clnt_mtx);
	gsp = SLIST_FIRST(&req->r_gss_seqlist);
	if (gsp && gsp->gss_seqnum > (cp->gss_clnt_seqnum - cp->gss_clnt_seqwin))
		win_resetbit(cp->gss_clnt_seqbits,
			gsp->gss_seqnum % cp->gss_clnt_seqwin);

	/*
	 * Limit the seqnum list to GSS_CLNT_SEQLISTMAX entries
	 */
	SLIST_FOREACH_SAFE(gsp, &req->r_gss_seqlist, gss_seqnext, ngsp) {
		if (++i > GSS_CLNT_SEQLISTMAX) {
			SLIST_REMOVE(&req->r_gss_seqlist, gsp, gss_seq, gss_seqnext);
			FREE(gsp, M_TEMP);
		}
	}

	/*
	 * If there's a thread waiting for
	 * the window to advance, wake it up.
	 */
	if (cp->gss_clnt_flags & GSS_NEEDSEQ) {
		cp->gss_clnt_flags &= ~GSS_NEEDSEQ;
		wakeup(cp);
	}
	lck_mtx_unlock(cp->gss_clnt_mtx);
}

/*
 * Create a reference to a context from a request
 * and bump the reference count
 */
void
nfs_gss_clnt_ctx_ref(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
{
	req->r_gss_ctx = cp;

	lck_mtx_lock(cp->gss_clnt_mtx);
	cp->gss_clnt_refcnt++;
	lck_mtx_unlock(cp->gss_clnt_mtx);
}

/*
 * Remove a context reference from a request
 * If the reference count drops to zero, and the
 * context is invalid, destroy the context
 */
void
nfs_gss_clnt_ctx_unref(struct nfsreq *req)
{
	struct nfsmount *nmp = req->r_nmp;
	struct nfs_gss_clnt_ctx *cp = req->r_gss_ctx;

	if (cp == NULL)
		return;

	req->r_gss_ctx = NULL;

	lck_mtx_lock(cp->gss_clnt_mtx);
	if (--cp->gss_clnt_refcnt == 0
		&& cp->gss_clnt_flags & GSS_CTX_INVAL) {
		lck_mtx_unlock(cp->gss_clnt_mtx);

		if (nmp)
			lck_mtx_lock(&nmp->nm_lock);
		nfs_gss_clnt_ctx_remove(nmp, cp);
		if (nmp)
			lck_mtx_unlock(&nmp->nm_lock);

		return;
	}
	lck_mtx_unlock(cp->gss_clnt_mtx);
}

/*
 * Remove a context
 */
static void
nfs_gss_clnt_ctx_remove(struct nfsmount *nmp, struct nfs_gss_clnt_ctx *cp)
{
	/*
	 * If dequeueing, assume nmp->nm_lock is held
	 */
	if (nmp != NULL)
		TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);

	host_release_special_port(cp->gss_clnt_mport);

	if (cp->gss_clnt_mtx)
		lck_mtx_destroy(cp->gss_clnt_mtx, nfs_gss_clnt_grp);
	if (IS_VALID_CRED(cp->gss_clnt_cred))
		kauth_cred_unref(&cp->gss_clnt_cred);
	if (cp->gss_clnt_principal)
		FREE(cp->gss_clnt_principal, M_TEMP);
	if (cp->gss_clnt_display)
		FREE(cp->gss_clnt_display, M_TEMP);
	if (cp->gss_clnt_handle)
		FREE(cp->gss_clnt_handle, M_TEMP);
	if (cp->gss_clnt_seqbits)
		FREE(cp->gss_clnt_seqbits, M_TEMP);
	if (cp->gss_clnt_token)
		FREE(cp->gss_clnt_token, M_TEMP);
	if (cp->gss_clnt_svcname)
		FREE(cp->gss_clnt_svcname, M_TEMP);
	FREE(cp, M_TEMP);
}

/*
 * The context for a user is invalid.
 * Mark the context as invalid, then
 * create a new context.
 */
int
nfs_gss_clnt_ctx_renew(struct nfsreq *req)
{
	struct nfs_gss_clnt_ctx *cp = req->r_gss_ctx;
	struct nfsmount *nmp = req->r_nmp;
	struct nfs_gss_clnt_ctx *ncp;
	int error = 0;
	kauth_cred_t saved_cred;
	mach_port_t saved_mport;

	if (cp == NULL)
		return (0);

	lck_mtx_lock(cp->gss_clnt_mtx);
	if (cp->gss_clnt_flags & GSS_CTX_INVAL) {
		lck_mtx_unlock(cp->gss_clnt_mtx);
		nfs_gss_clnt_ctx_unref(req);
		return (0);	// already being renewed
	}
	saved_cred = cp->gss_clnt_cred;
	kauth_cred_ref(saved_cred);
	saved_mport = host_copy_special_port(cp->gss_clnt_mport);

	/* Remove the old context */
	cp->gss_clnt_flags |= GSS_CTX_INVAL;

	/*
	 * If there's a thread waiting
	 * in the old context, wake it up.
	 */
	if (cp->gss_clnt_flags & (GSS_NEEDCTX | GSS_NEEDSEQ)) {
		cp->gss_clnt_flags &= ~GSS_NEEDSEQ;
		wakeup(cp);
	}
	lck_mtx_unlock(cp->gss_clnt_mtx);

	/*
	 * Create a new context
	 */
	MALLOC(ncp, struct nfs_gss_clnt_ctx *, sizeof(*ncp),
		M_TEMP, M_WAITOK|M_ZERO);
	if (ncp == NULL) {
		error = ENOMEM;
		goto out;
	}

	ncp->gss_clnt_cred = saved_cred;
	kauth_cred_ref(ncp->gss_clnt_cred);
	ncp->gss_clnt_mport = host_copy_special_port(saved_mport); // re-use the gssd port
	ncp->gss_clnt_mtx = lck_mtx_alloc_init(nfs_gss_clnt_grp, LCK_ATTR_NULL);
	ncp->gss_clnt_thread = current_thread();
	lck_mtx_lock(&nmp->nm_lock);
	TAILQ_INSERT_TAIL(&nmp->nm_gsscl, ncp, gss_clnt_entries);
	lck_mtx_unlock(&nmp->nm_lock);

	/* Adjust reference counts to new and old context */
	nfs_gss_clnt_ctx_unref(req);
	nfs_gss_clnt_ctx_ref(req, ncp);

	error = nfs_gss_clnt_ctx_init_retry(req, ncp); // Initialize new context
out:
	host_release_special_port(saved_mport);
	kauth_cred_unref(&saved_cred);
	if (error)
		nfs_gss_clnt_ctx_unref(req);

	return (error);
}

/*
 * Destroy all the contexts associated with a mount.
 * The contexts are also destroyed by the server.
 */
void
nfs_gss_clnt_ctx_unmount(struct nfsmount *nmp)
{
	struct nfs_gss_clnt_ctx *cp;
	struct nfsm_chain nmreq, nmrep;
	int error, status;
	struct nfsreq req;

	req.r_nmp = nmp;

	for (;;) {
		lck_mtx_lock(&nmp->nm_lock);
		cp = TAILQ_FIRST(&nmp->nm_gsscl);
		if (cp) {
			lck_mtx_lock(cp->gss_clnt_mtx);
			cp->gss_clnt_refcnt++;
			lck_mtx_unlock(cp->gss_clnt_mtx);
			req.r_gss_ctx = cp;
		}
		lck_mtx_unlock(&nmp->nm_lock);
		if (cp == NULL)
			break;

		/*
		 * Tell the server to destroy its context.
		 * But don't bother if it's a forced unmount
		 * or if it's a dummy sec=sys context.
		 */
		if (!(nmp->nm_state & NFSSTA_FORCE) && (cp->gss_clnt_service != RPCSEC_GSS_SVC_SYS)) {
			cp->gss_clnt_proc = RPCSEC_GSS_DESTROY;

			error = 0;
			nfsm_chain_null(&nmreq);
			nfsm_chain_null(&nmrep);
			nfsm_chain_build_alloc_init(error, &nmreq, 0);
			nfsm_chain_build_done(error, &nmreq);
			if (!error)
				nfs_request_gss(nmp->nm_mountp, &nmreq,
					current_thread(), cp->gss_clnt_cred, 0, cp, &nmrep, &status);
			nfsm_chain_cleanup(&nmreq);
			nfsm_chain_cleanup(&nmrep);
		}

		/*
		 * Mark the context invalid then drop
		 * the reference to remove it if its
		 * refcount is zero.
		 */
		lck_mtx_lock(cp->gss_clnt_mtx);
		cp->gss_clnt_flags |= GSS_CTX_INVAL;
		lck_mtx_unlock(cp->gss_clnt_mtx);
		nfs_gss_clnt_ctx_unref(&req);
	}
}

/*
 * Destroy a mounts context for a credential
 */
int
nfs_gss_clnt_ctx_destroy(struct nfsmount *nmp, kauth_cred_t cred)
{
	struct nfs_gss_clnt_ctx *cp;
	struct nfsreq req;

	req.r_nmp = nmp;

	lck_mtx_lock(&nmp->nm_lock);
	TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
		if (nfs_gss_clnt_ctx_cred_match(cp->gss_clnt_cred, cred)) {
			if (cp->gss_clnt_flags & GSS_CTX_INVAL)
				continue;
			lck_mtx_lock(cp->gss_clnt_mtx);
			cp->gss_clnt_refcnt++;
			cp->gss_clnt_flags |= GSS_CTX_INVAL;
			lck_mtx_unlock(cp->gss_clnt_mtx);
			req.r_gss_ctx = cp;
			break;
		}
	}
	lck_mtx_unlock(&nmp->nm_lock);

	if (cp == NULL)
		return (ENOENT);

	/*
	 * Drop the reference to remove it if its
	 * refcount is zero.
	 */
	nfs_gss_clnt_ctx_unref(&req);

	return (0);
}


#endif /* NFSCLIENT */

/*************
 *
 * Server functions
 */

#if NFSSERVER

/*
 * Find a server context based on a handle value received
 * in an RPCSEC_GSS credential.
 */
static struct nfs_gss_svc_ctx *
nfs_gss_svc_ctx_find(uint32_t handle)
{
	struct nfs_gss_svc_ctx_hashhead *head;
	struct nfs_gss_svc_ctx *cp;
	uint64_t timenow;

	if (handle == 0)
		return (NULL);

	head = &nfs_gss_svc_ctx_hashtbl[SVC_CTX_HASH(handle)];
	/*
	 * Don't return a context that is going to expire in GSS_CTX_PEND seconds
	 */
	clock_interval_to_deadline(GSS_CTX_PEND, NSEC_PER_SEC, &timenow);

	lck_mtx_lock(nfs_gss_svc_ctx_mutex);

	LIST_FOREACH(cp, head, gss_svc_entries) {
		if (cp->gss_svc_handle == handle) {
			if (timenow > cp->gss_svc_incarnation + GSS_SVC_CTX_TTL) {
				/*
				 * Context has or is about to expire. Don't use.
				 * We'll return null and the client will have to create
				 * a new context.
				 */
				cp->gss_svc_handle = 0;
				/*
				 * Make sure though that we stay around for GSS_CTX_PEND seconds
				 * for other threads that might be using the context.
				 */
				cp->gss_svc_incarnation = timenow;

				cp = NULL;
				break;
			}
			lck_mtx_lock(cp->gss_svc_mtx);
			cp->gss_svc_refcnt++;
			lck_mtx_unlock(cp->gss_svc_mtx);
			break;
		}
	}

	lck_mtx_unlock(nfs_gss_svc_ctx_mutex);

	return (cp);
}

/*
 * Insert a new server context into the hash table
 * and start the context reap thread if necessary.
 */
static void
nfs_gss_svc_ctx_insert(struct nfs_gss_svc_ctx *cp)
{
	struct nfs_gss_svc_ctx_hashhead *head;
	struct nfs_gss_svc_ctx *p;

	lck_mtx_lock(nfs_gss_svc_ctx_mutex);

	/*
	 * Give the client a random handle so that if we reboot
	 * it's unlikely the client will get a bad context match.
	 * Make sure it's not zero or already assigned.
	 */
retry:
	cp->gss_svc_handle = random();
	if (cp->gss_svc_handle == 0)
		goto retry;
	head = &nfs_gss_svc_ctx_hashtbl[SVC_CTX_HASH(cp->gss_svc_handle)];
	LIST_FOREACH(p, head, gss_svc_entries)
		if (p->gss_svc_handle == cp->gss_svc_handle)
			goto retry;

	clock_interval_to_deadline(GSS_CTX_PEND, NSEC_PER_SEC,
		&cp->gss_svc_incarnation);
	LIST_INSERT_HEAD(head, cp, gss_svc_entries);
	nfs_gss_ctx_count++;

	if (!nfs_gss_timer_on) {
		nfs_gss_timer_on = 1;

		nfs_interval_timer_start(nfs_gss_svc_ctx_timer_call,
			min(GSS_TIMER_PERIOD, max(GSS_CTX_TTL_MIN, nfsrv_gss_context_ttl)) * MSECS_PER_SEC);
	}

	lck_mtx_unlock(nfs_gss_svc_ctx_mutex);
}

/*
 * This function is called via the kernel's callout
 * mechanism.  It runs only when there are
 * cached RPCSEC_GSS contexts.
 */
void
nfs_gss_svc_ctx_timer(__unused void *param1, __unused void *param2)
{
	struct nfs_gss_svc_ctx *cp, *next;
	uint64_t timenow;
	int contexts = 0;
	int i;

	lck_mtx_lock(nfs_gss_svc_ctx_mutex);
	clock_get_uptime(&timenow);

	/*
	 * Scan all the hash chains
	 */
	for (i = 0; i < SVC_CTX_HASHSZ; i++) {
		/*
		 * For each hash chain, look for entries
		 * that haven't been used in a while.
		 */
		LIST_FOREACH_SAFE(cp, &nfs_gss_svc_ctx_hashtbl[i], gss_svc_entries, next) {
			contexts++;
			if (timenow > cp->gss_svc_incarnation +
				(cp->gss_svc_handle ? GSS_SVC_CTX_TTL : 0)
				&& cp->gss_svc_refcnt == 0) {
				/*
				 * A stale context - remove it
				 */
				LIST_REMOVE(cp, gss_svc_entries);
				if (cp->gss_svc_seqbits)
					FREE(cp->gss_svc_seqbits, M_TEMP);
				lck_mtx_destroy(cp->gss_svc_mtx, nfs_gss_svc_grp);
				FREE(cp, M_TEMP);
				contexts--;
			}
		}
	}

	nfs_gss_ctx_count = contexts;

	/*
	 * If there are still some cached contexts left,
	 * set up another callout to check on them later.
	 */
	nfs_gss_timer_on = nfs_gss_ctx_count > 0;
	if (nfs_gss_timer_on)
		nfs_interval_timer_start(nfs_gss_svc_ctx_timer_call,
			min(GSS_TIMER_PERIOD, max(GSS_CTX_TTL_MIN, nfsrv_gss_context_ttl)) * MSECS_PER_SEC);

	lck_mtx_unlock(nfs_gss_svc_ctx_mutex);
}

/*
 * Here the server receives an RPCSEC_GSS credential in an
 * RPC call header.  First there's some checking to make sure
 * the credential is appropriate - whether the context is still
 * being set up, or is complete.  Then we use the handle to find
 * the server's context and validate the verifier, which contains
 * a signed checksum of the RPC header. If the verifier checks
 * out, we extract the user's UID and groups from the context
 * and use it to set up a UNIX credential for the user's request.
 */
int
nfs_gss_svc_cred_get(struct nfsrv_descript *nd, struct nfsm_chain *nmc)
{
	uint32_t vers, proc, seqnum, service;
	uint32_t handle, handle_len;
	struct nfs_gss_svc_ctx *cp = NULL;
	uint32_t flavor = 0, verflen = 0;
	int error = 0;
	uint32_t arglen, start, toklen, cksumlen;
	u_char tokbuf[KRB5_SZ_TOKMAX(MAX_DIGEST)];
	u_char cksum1[MAX_DIGEST], cksum2[MAX_DIGEST];
	struct nfsm_chain nmc_tmp;
	gss_key_info *ki;

	vers = proc = seqnum = service = handle_len = 0;
	arglen = cksumlen = 0;

	nfsm_chain_get_32(error, nmc, vers);
	if (vers != RPCSEC_GSS_VERS_1) {
		error = NFSERR_AUTHERR | AUTH_REJECTCRED;
		goto nfsmout;
	}

	nfsm_chain_get_32(error, nmc, proc);
	nfsm_chain_get_32(error, nmc, seqnum);
	nfsm_chain_get_32(error, nmc, service);
	nfsm_chain_get_32(error, nmc, handle_len);
	if (error)
		goto nfsmout;

	/*
	 * Make sure context setup/destroy is being done with a nullproc
	 */
	if (proc != RPCSEC_GSS_DATA && nd->nd_procnum != NFSPROC_NULL) {
		error = NFSERR_AUTHERR | RPCSEC_GSS_CREDPROBLEM;
		goto nfsmout;
	}

	/*
	 * If the sequence number is greater than the max
	 * allowable, reject and have the client init a
	 * new context.
	 */
	if (seqnum > GSS_MAXSEQ) {
		error = NFSERR_AUTHERR | RPCSEC_GSS_CTXPROBLEM;
		goto nfsmout;
	}

	nd->nd_sec =
		service == RPCSEC_GSS_SVC_NONE ?      RPCAUTH_KRB5 :
		service == RPCSEC_GSS_SVC_INTEGRITY ? RPCAUTH_KRB5I :
		service == RPCSEC_GSS_SVC_PRIVACY ?   RPCAUTH_KRB5P : 0;

	if (proc == RPCSEC_GSS_INIT) {
		/*
		 * Limit the total number of contexts
		 */
		if (nfs_gss_ctx_count > nfs_gss_ctx_max) {
			error = NFSERR_AUTHERR | RPCSEC_GSS_CTXPROBLEM;
			goto nfsmout;
		}

		/*
		 * Set up a new context
		 */
		MALLOC(cp, struct nfs_gss_svc_ctx *, sizeof(*cp), M_TEMP, M_WAITOK|M_ZERO);
		if (cp == NULL) {
			error = ENOMEM;
			goto nfsmout;
		}
		cp->gss_svc_mtx = lck_mtx_alloc_init(nfs_gss_svc_grp, LCK_ATTR_NULL);
		cp->gss_svc_refcnt = 1;
	} else {

		/*
		 * Use the handle to find the context
		 */
		if (handle_len != sizeof(handle)) {
			error = NFSERR_AUTHERR | RPCSEC_GSS_CREDPROBLEM;
			goto nfsmout;
		}
		nfsm_chain_get_32(error, nmc, handle);
		if (error)
			goto nfsmout;
		cp = nfs_gss_svc_ctx_find(handle);
		if (cp == NULL) {
			error = NFSERR_AUTHERR | RPCSEC_GSS_CTXPROBLEM;
			goto nfsmout;
		}
	}

	cp->gss_svc_proc = proc;
	ki = &cp->gss_svc_kinfo;

	if (proc == RPCSEC_GSS_DATA || proc == RPCSEC_GSS_DESTROY) {
		struct posix_cred temp_pcred;

		if (cp->gss_svc_seqwin == 0) {
			/*
			 * Context isn't complete
			 */
			error = NFSERR_AUTHERR | RPCSEC_GSS_CTXPROBLEM;
			goto nfsmout;
		}

		if (!nfs_gss_svc_seqnum_valid(cp, seqnum)) {
			/*
			 * Sequence number is bad
			 */
			error = EINVAL;	// drop the request
			goto nfsmout;
		}

		/* Now compute the client's call header checksum */
		nfs_gss_cksum_chain(ki, nmc, ALG_MIC(ki), 0, 0, cksum1);

		/*
		 * Validate the verifier.
		 * The verifier contains an encrypted checksum
		 * of the call header from the XID up to and
		 * including the credential.  We compute the
		 * checksum and compare it with what came in
		 * the verifier.
		 */
		nfsm_chain_get_32(error, nmc, flavor);
		nfsm_chain_get_32(error, nmc, verflen);
		if (error)
			goto nfsmout;
		if (flavor != RPCSEC_GSS || verflen != KRB5_SZ_TOKEN(ki->hash_len))
			error = NFSERR_AUTHERR | AUTH_BADVERF;
		nfsm_chain_get_opaque(error, nmc, verflen, tokbuf);
		if (error)
			goto nfsmout;

		/* Get the checksum from the token inside the verifier */
		error = nfs_gss_token_get(ki, ALG_MIC(ki), tokbuf, 1,
			NULL, cksum2);
		if (error)
			goto nfsmout;

		if (bcmp(cksum1, cksum2, HASHLEN(ki)) != 0) {
			error = NFSERR_AUTHERR | RPCSEC_GSS_CTXPROBLEM;
			goto nfsmout;
		}

		nd->nd_gss_seqnum = seqnum;

		/*
		 * Set up the user's cred
		 */
		bzero(&temp_pcred, sizeof(temp_pcred));
		temp_pcred.cr_uid = cp->gss_svc_uid;
		bcopy(cp->gss_svc_gids, temp_pcred.cr_groups,
				sizeof(gid_t) * cp->gss_svc_ngroups);
		temp_pcred.cr_ngroups = cp->gss_svc_ngroups;

		nd->nd_cr = posix_cred_create(&temp_pcred);
		if (nd->nd_cr == NULL) {
			error = ENOMEM;
			goto nfsmout;
		}
		clock_get_uptime(&cp->gss_svc_incarnation);

		/*
		 * If the call arguments are integrity or privacy protected
		 * then we need to check them here.
		 */
		switch (service) {
		case RPCSEC_GSS_SVC_NONE:
			/* nothing to do */
			break;
		case RPCSEC_GSS_SVC_INTEGRITY:
			/*
			 * Here's what we expect in the integrity call args:
			 *
			 * - length of seq num + call args (4 bytes)
			 * - sequence number (4 bytes)
			 * - call args (variable bytes)
			 * - length of checksum token (37)
			 * - checksum of seqnum + call args (37 bytes)
			 */
			nfsm_chain_get_32(error, nmc, arglen);		// length of args
			if (arglen > NFS_MAXPACKET) {
				error = EBADRPC;
				goto nfsmout;
			}

			/* Compute the checksum over the call args */
			start = nfsm_chain_offset(nmc);
			nfs_gss_cksum_chain(ki, nmc, ALG_MIC(ki), start, arglen, cksum1);

			/*
			 * Get the sequence number prepended to the args
			 * and compare it against the one sent in the
			 * call credential.
			 */
			nfsm_chain_get_32(error, nmc, seqnum);
			if (seqnum != nd->nd_gss_seqnum) {
				error = EBADRPC;			// returns as GARBAGEARGS
				goto nfsmout;
			}

			/*
			 * Advance to the end of the args and
			 * fetch the checksum computed by the client.
			 */
			nmc_tmp = *nmc;
			arglen -= NFSX_UNSIGNED;			// skipped seqnum
			nfsm_chain_adv(error, &nmc_tmp, arglen);	// skip args
			nfsm_chain_get_32(error, &nmc_tmp, cksumlen);	// length of checksum
			if (cksumlen != KRB5_SZ_TOKEN(ki->hash_len)) {
				error = EBADRPC;
				goto nfsmout;
			}
			nfsm_chain_get_opaque(error, &nmc_tmp, cksumlen, tokbuf);
			if (error)
				goto nfsmout;
			error = nfs_gss_token_get(ki, ALG_MIC(ki), tokbuf, 1,
				NULL, cksum2);

			/* Verify that the checksums are the same */
			if (error || bcmp(cksum1, cksum2, HASHLEN(ki)) != 0) {
				error = EBADRPC;
				goto nfsmout;
			}
			break;
		case RPCSEC_GSS_SVC_PRIVACY:
			/*
			 * Here's what we expect in the privacy call args:
			 *
			 * - length of confounder + seq num + token + call args
			 * - wrap token (37-40 bytes)
			 * - confounder (8 bytes)
			 * - sequence number (4 bytes)
			 * - call args (encrypted)
			 */
			nfsm_chain_get_32(error, nmc, arglen);		// length of args
			if (arglen > NFS_MAXPACKET) {
				error = EBADRPC;
				goto nfsmout;
			}

			/* Get the token that prepends the encrypted args */
			nfsm_chain_get_opaque(error, nmc, KRB5_SZ_TOKMAX(ki->hash_len), tokbuf);
			if (error)
				goto nfsmout;
			error = nfs_gss_token_get(ki, ALG_WRAP(ki), tokbuf, 1,
							&toklen, cksum1);
			if (error)
				goto nfsmout;
			nfsm_chain_reverse(nmc, nfsm_pad(toklen));

			/* decrypt the 8 byte confounder + seqnum + args */
			start = nfsm_chain_offset(nmc);
			arglen -= toklen;
			nfs_gss_encrypt_chain(ki, nmc, start, arglen, DES_DECRYPT);

			/* Compute a checksum over the sequence number + results */
			nfs_gss_cksum_chain(ki, nmc, ALG_WRAP(ki), start, arglen, cksum2);

			/* Verify that the checksums are the same */
			if (bcmp(cksum1, cksum2, HASHLEN(ki)) != 0) {
				error = EBADRPC;
				goto nfsmout;
			}

			/*
			 * Get the sequence number prepended to the args
			 * and compare it against the one sent in the
			 * call credential.
			 */
			nfsm_chain_adv(error, nmc, 8);			// skip over the confounder
			nfsm_chain_get_32(error, nmc, seqnum);
			if (seqnum != nd->nd_gss_seqnum) {
				error = EBADRPC;			// returns as GARBAGEARGS
				goto nfsmout;
			}
			break;
		}
	} else {
		/*
		 * If the proc is RPCSEC_GSS_INIT or RPCSEC_GSS_CONTINUE_INIT
		 * then we expect a null verifier.
		 */
		nfsm_chain_get_32(error, nmc, flavor);
		nfsm_chain_get_32(error, nmc, verflen);
		if (error || flavor != RPCAUTH_NULL || verflen > 0)
			error = NFSERR_AUTHERR | RPCSEC_GSS_CREDPROBLEM;
		if (error) {
			if (proc == RPCSEC_GSS_INIT) {
				lck_mtx_destroy(cp->gss_svc_mtx, nfs_gss_svc_grp);
				FREE(cp, M_TEMP);
				cp = NULL;
			}
			goto nfsmout;
		}
	}

	nd->nd_gss_context = cp;
	return 0;
nfsmout:
	if (cp)
		nfs_gss_svc_ctx_deref(cp);
	return (error);
}

/*
 * Insert the server's verifier into the RPC reply header.
 * It contains a signed checksum of the sequence number that
 * was received in the RPC call.
 * Then go on to add integrity or privacy if necessary.
 */
int
nfs_gss_svc_verf_put(struct nfsrv_descript *nd, struct nfsm_chain *nmc)
{
	struct nfs_gss_svc_ctx *cp;
	int error = 0;
	u_char tokbuf[KRB5_SZ_TOKEN(MAX_DIGEST)];
	int toklen;
	u_char cksum[MAX_DIGEST];
	gss_key_info *ki;

	cp = nd->nd_gss_context;
	ki = &cp->gss_svc_kinfo;

	if (cp->gss_svc_major != GSS_S_COMPLETE) {
		/*
		 * If the context isn't yet complete
		 * then return a null verifier.
		 */
		nfsm_chain_add_32(error, nmc, RPCAUTH_NULL);
		nfsm_chain_add_32(error, nmc, 0);
		return (error);
	}

	/*
	 * Compute checksum of the request seq number
	 * If it's the final reply of context setup
	 * then return the checksum of the context
	 * window size.
	 */
	if (cp->gss_svc_proc == RPCSEC_GSS_INIT ||
	    cp->gss_svc_proc == RPCSEC_GSS_CONTINUE_INIT)
		nfs_gss_cksum_rep(ki, cp->gss_svc_seqwin, cksum);
	else
		nfs_gss_cksum_rep(ki, nd->nd_gss_seqnum, cksum);
	/*
	 * Now wrap it in a token and add
	 * the verifier to the reply.
	 */
	toklen = nfs_gss_token_put(ki, ALG_MIC(ki), tokbuf, 0, 0, cksum);
	nfsm_chain_add_32(error, nmc, RPCSEC_GSS);
	nfsm_chain_add_32(error, nmc, toklen);
	nfsm_chain_add_opaque(error, nmc, tokbuf, toklen);

	return (error);
}

/*
 * The results aren't available yet, but if they need to be
 * checksummed for integrity protection or encrypted, then
 * we can record the start offset here, insert a place-holder
 * for the results length, as well as the sequence number.
 * The rest of the work is done later by nfs_gss_svc_protect_reply()
 * when the results are available.
 */
int
nfs_gss_svc_prepare_reply(struct nfsrv_descript *nd, struct nfsm_chain *nmc)
{
	struct nfs_gss_svc_ctx *cp = nd->nd_gss_context;
	int error = 0;

	if (cp->gss_svc_proc == RPCSEC_GSS_INIT ||
	    cp->gss_svc_proc == RPCSEC_GSS_CONTINUE_INIT)
		return (0);

	switch (nd->nd_sec) {
	case RPCAUTH_KRB5:
		/* Nothing to do */
		break;
	case RPCAUTH_KRB5I:
		nd->nd_gss_mb = nmc->nmc_mcur;			// record current mbuf
		nfsm_chain_finish_mbuf(error, nmc);		// split the chain here
		nfsm_chain_add_32(error, nmc, nd->nd_gss_seqnum); // req sequence number
		break;
	case RPCAUTH_KRB5P:
		nd->nd_gss_mb = nmc->nmc_mcur;			// record current mbuf
		nfsm_chain_finish_mbuf(error, nmc);		// split the chain here
		nfsm_chain_add_32(error, nmc, random());	// confounder bytes 1-4
		nfsm_chain_add_32(error, nmc, random());	// confounder bytes 5-8
		nfsm_chain_add_32(error, nmc, nd->nd_gss_seqnum); // req sequence number
		break;
	}

	return (error);
}

/*
 * The results are checksummed or encrypted for return to the client
 */
int
nfs_gss_svc_protect_reply(struct nfsrv_descript *nd, mbuf_t mrep)
{
	struct nfs_gss_svc_ctx *cp = nd->nd_gss_context;
	struct nfsm_chain nmrep_res, *nmc_res = &nmrep_res;
	struct nfsm_chain nmrep_pre, *nmc_pre = &nmrep_pre;
	mbuf_t mb, results;
	uint32_t reslen;
	u_char tokbuf[KRB5_SZ_TOKMAX(MAX_DIGEST)];
	int pad, toklen;
	u_char cksum[MAX_DIGEST];
	int error = 0;
	gss_key_info *ki = &cp->gss_svc_kinfo;

	/*
	 * Using a reference to the mbuf where we previously split the reply
	 * mbuf chain, we split the mbuf chain argument into two mbuf chains,
	 * one that allows us to prepend a length field or token, (nmc_pre)
	 * and the second which holds just the results that we're going to
	 * checksum and/or encrypt.  When we're done, we join the chains back
	 * together.
	 */
	nfs_gss_nfsm_chain(nmc_res, mrep);		// set up the results chain
	mb = nd->nd_gss_mb;				// the mbuf where we split
	results = mbuf_next(mb);			// first mbuf in the results
	reslen = nfs_gss_mchain_length(results);	// length of results
	error = mbuf_setnext(mb, NULL);			// disconnect the chains
	if (error)
		return (error);
	nfs_gss_nfsm_chain(nmc_pre, mb);		// set up the prepend chain

	if (nd->nd_sec == RPCAUTH_KRB5I) {
		nfsm_chain_add_32(error, nmc_pre, reslen);
		nfsm_chain_build_done(error, nmc_pre);
		if (error)
			return (error);
		nfs_gss_append_chain(nmc_pre, results);	// Append the results mbufs

		/* Now compute the checksum over the results data */
		nfs_gss_cksum_mchain(ki, results, ALG_MIC(ki), 0, reslen, cksum);

		/* Put it into a token and append to the request */
		toklen = nfs_gss_token_put(ki, ALG_MIC(ki), tokbuf, 0, 0, cksum);
		nfsm_chain_add_32(error, nmc_res, toklen);
		nfsm_chain_add_opaque(error, nmc_res, tokbuf, toklen);
		nfsm_chain_build_done(error, nmc_res);
	} else {
		/* RPCAUTH_KRB5P */
		/*
		 * Append a pad trailer - per RFC 1964 section 1.2.2.3
		 * Since XDR data is always 32-bit aligned, it
		 * needs to be padded either by 4 bytes or 8 bytes.
		 */
		if (reslen % 8 > 0) {
			nfsm_chain_add_32(error, nmc_res, 0x04040404);
			reslen += NFSX_UNSIGNED;
		} else {
			nfsm_chain_add_32(error, nmc_res, 0x08080808);
			nfsm_chain_add_32(error, nmc_res, 0x08080808);
			reslen +=  2 * NFSX_UNSIGNED;
		}
		nfsm_chain_build_done(error, nmc_res);

		/* Now compute the checksum over the results data */
		nfs_gss_cksum_mchain(ki, results, ALG_WRAP(ki), 0, reslen, cksum);

		/* Put it into a token and insert in the reply */
		toklen = nfs_gss_token_put(ki, ALG_WRAP(ki), tokbuf, 0, reslen, cksum);
		nfsm_chain_add_32(error, nmc_pre, toklen + reslen);
		nfsm_chain_add_opaque_nopad(error, nmc_pre, tokbuf, toklen);
		nfsm_chain_build_done(error, nmc_pre);
		if (error)
			return (error);
		nfs_gss_append_chain(nmc_pre, results);	// Append the results mbufs

		/* Encrypt the confounder + seqnum + results */
		nfs_gss_encrypt_mchain(ki, results, 0, reslen, DES_ENCRYPT);

		/* Add null XDR pad if the ASN.1 token misaligned the data */
		pad = nfsm_pad(toklen + reslen);
		if (pad > 0) {
			nfsm_chain_add_opaque_nopad(error, nmc_pre, iv0, pad);
			nfsm_chain_build_done(error, nmc_pre);
		}
	}

	return (error);
}

/*
 * This function handles the context setup calls from the client.
 * Essentially, it implements the NFS null procedure calls when
 * an RPCSEC_GSS credential is used.
 * This is the context maintenance function.  It creates and
 * destroys server contexts at the whim of the client.
 * During context creation, it receives GSS-API tokens from the
 * client, passes them up to gssd, and returns a received token
 * back to the client in the null procedure reply.
 */
int
nfs_gss_svc_ctx_init(struct nfsrv_descript *nd, struct nfsrv_sock *slp, mbuf_t *mrepp)
{
	struct nfs_gss_svc_ctx *cp = NULL;
	int error = 0;
	int autherr = 0;
	struct nfsm_chain *nmreq, nmrep;
	int sz;

	nmreq = &nd->nd_nmreq;
	nfsm_chain_null(&nmrep);
	*mrepp = NULL;
	cp = nd->nd_gss_context;
	nd->nd_repstat = 0;

	switch (cp->gss_svc_proc) {
	case RPCSEC_GSS_INIT:
		nfs_gss_svc_ctx_insert(cp);
		/* FALLTHRU */

	case RPCSEC_GSS_CONTINUE_INIT:
		/* Get the token from the request */
		nfsm_chain_get_32(error, nmreq, cp->gss_svc_tokenlen);
		if (cp->gss_svc_tokenlen == 0) {
			autherr = RPCSEC_GSS_CREDPROBLEM;
			break;
		}
		MALLOC(cp->gss_svc_token, u_char *, cp->gss_svc_tokenlen, M_TEMP, M_WAITOK);
		if (cp->gss_svc_token == NULL) {
			autherr = RPCSEC_GSS_CREDPROBLEM;
			break;
		}
		nfsm_chain_get_opaque(error, nmreq, cp->gss_svc_tokenlen, cp->gss_svc_token);

		/* Use the token in a gss_accept_sec_context upcall */
		error = nfs_gss_svc_gssd_upcall(cp);
		if (error) {
			autherr = RPCSEC_GSS_CREDPROBLEM;
			if (error == NFSERR_EAUTH)
				error = 0;
			break;
		}

		/*
		 * If the context isn't complete, pass the new token
		 * back to the client for another round.
		 */
		if (cp->gss_svc_major != GSS_S_COMPLETE)
			break;

		/*
		 * Now the server context is complete.
		 * Finish setup.
		 */
		clock_get_uptime(&cp->gss_svc_incarnation);

		cp->gss_svc_seqwin = GSS_SVC_SEQWINDOW;
		MALLOC(cp->gss_svc_seqbits, uint32_t *,
			nfsm_rndup((cp->gss_svc_seqwin + 7) / 8), M_TEMP, M_WAITOK|M_ZERO);
		if (cp->gss_svc_seqbits == NULL) {
			autherr = RPCSEC_GSS_CREDPROBLEM;
			break;
		}
		break;

	case RPCSEC_GSS_DATA:
		/* Just a nullproc ping - do nothing */
		break;

	case RPCSEC_GSS_DESTROY:
		/*
		 * Don't destroy the context immediately because
		 * other active requests might still be using it.
		 * Instead, schedule it for destruction after
		 * GSS_CTX_PEND time has elapsed.
		 */
		cp = nfs_gss_svc_ctx_find(cp->gss_svc_handle);
		if (cp != NULL) {
			cp->gss_svc_handle = 0;	// so it can't be found
			lck_mtx_lock(cp->gss_svc_mtx);
			clock_interval_to_deadline(GSS_CTX_PEND, NSEC_PER_SEC,
				&cp->gss_svc_incarnation);
			lck_mtx_unlock(cp->gss_svc_mtx);
		}
		break;
	default:
		autherr = RPCSEC_GSS_CREDPROBLEM;
		break;
	}

	/* Now build the reply  */

	if (nd->nd_repstat == 0)
		nd->nd_repstat = autherr ? (NFSERR_AUTHERR | autherr) : NFSERR_RETVOID;
	sz = 7 * NFSX_UNSIGNED + nfsm_rndup(cp->gss_svc_tokenlen); // size of results
	error = nfsrv_rephead(nd, slp, &nmrep, sz);
	*mrepp = nmrep.nmc_mhead;
	if (error || autherr)
		goto nfsmout;

	if (cp->gss_svc_proc == RPCSEC_GSS_INIT ||
	    cp->gss_svc_proc == RPCSEC_GSS_CONTINUE_INIT) {
		nfsm_chain_add_32(error, &nmrep, sizeof(cp->gss_svc_handle));
		nfsm_chain_add_32(error, &nmrep, cp->gss_svc_handle);

		nfsm_chain_add_32(error, &nmrep, cp->gss_svc_major);
		nfsm_chain_add_32(error, &nmrep, cp->gss_svc_minor);
		nfsm_chain_add_32(error, &nmrep, cp->gss_svc_seqwin);

		nfsm_chain_add_32(error, &nmrep, cp->gss_svc_tokenlen);
		if (cp->gss_svc_token != NULL) {
			nfsm_chain_add_opaque(error, &nmrep, cp->gss_svc_token, cp->gss_svc_tokenlen);
			FREE(cp->gss_svc_token, M_TEMP);
			cp->gss_svc_token = NULL;
		}
	}

nfsmout:
	if (autherr != 0) {
		nd->nd_gss_context = NULL;
		LIST_REMOVE(cp, gss_svc_entries);
		if (cp->gss_svc_seqbits != NULL)
			FREE(cp->gss_svc_seqbits, M_TEMP);
		if (cp->gss_svc_token != NULL)
			FREE(cp->gss_svc_token, M_TEMP);
		lck_mtx_destroy(cp->gss_svc_mtx, nfs_gss_svc_grp);
		FREE(cp, M_TEMP);
	}

	nfsm_chain_build_done(error, &nmrep);
	if (error) {
		nfsm_chain_cleanup(&nmrep);
		*mrepp = NULL;
	}
	return (error);
}

/*
 * This is almost a mirror-image of the client side upcall.
 * It passes and receives a token, but invokes gss_accept_sec_context.
 * If it's the final call of the context setup, then gssd also returns
 * the session key and the user's UID.
 */
static int
nfs_gss_svc_gssd_upcall(struct nfs_gss_svc_ctx *cp)
{
	kern_return_t kr;
	mach_port_t mp;
	int retry_cnt = 0;
	gssd_byte_buffer okey = NULL;
	uint32_t skeylen = 0;
	uint32_t ret_flags;
	vm_map_copy_t itoken = NULL;
	gssd_byte_buffer otoken = NULL;
	mach_msg_type_number_t otokenlen;
	int error = 0;
	char svcname[] = "nfs";

	kr = host_get_gssd_port(host_priv_self(), &mp);
	if (kr != KERN_SUCCESS) {
		printf("nfs_gss_svc_gssd_upcall: can't get gssd port, status %x (%d)\n", kr, kr);
		goto out;
	}
	if (!IPC_PORT_VALID(mp)) {
		printf("nfs_gss_svc_gssd_upcall: gssd port not valid\n");
		goto out;
	}

	if (cp->gss_svc_tokenlen > 0)
		nfs_gss_mach_alloc_buffer(cp->gss_svc_token, cp->gss_svc_tokenlen, &itoken);

retry:
	kr = mach_gss_accept_sec_context(
		mp,
		(gssd_byte_buffer) itoken, (mach_msg_type_number_t) cp->gss_svc_tokenlen,
		svcname,
		0,
		&cp->gss_svc_context,
		&cp->gss_svc_cred_handle,
		&ret_flags,
		&cp->gss_svc_uid,
		cp->gss_svc_gids,
		&cp->gss_svc_ngroups,
		&okey, (mach_msg_type_number_t *) &skeylen,
		&otoken, &otokenlen,
		&cp->gss_svc_major,
		&cp->gss_svc_minor);

	if (kr != KERN_SUCCESS) {
		printf("nfs_gss_svc_gssd_upcall failed: %x (%d)\n", kr, kr);
		if (kr == MIG_SERVER_DIED && cp->gss_svc_context == 0 &&
			retry_cnt++ < NFS_GSS_MACH_MAX_RETRIES) {
			if (cp->gss_svc_tokenlen > 0)
				nfs_gss_mach_alloc_buffer(cp->gss_svc_token, cp->gss_svc_tokenlen, &itoken);
			goto retry;
		}
		host_release_special_port(mp);
		goto out;
	}

	host_release_special_port(mp);

	if (skeylen > 0) {
		if (skeylen != SKEYLEN && skeylen != SKEYLEN3) {
			printf("nfs_gss_svc_gssd_upcall: bad key length (%d)\n", skeylen);
			vm_map_copy_discard((vm_map_copy_t) okey);
			vm_map_copy_discard((vm_map_copy_t) otoken);
			goto out;
		}
		error = nfs_gss_mach_vmcopyout((vm_map_copy_t) okey, skeylen, cp->gss_svc_kinfo.skey);
		if (error) {
			vm_map_copy_discard((vm_map_copy_t) otoken);
			goto out;
		}
		error = gss_key_init(&cp->gss_svc_kinfo, skeylen);
		if (error)
			goto out;

	}

	/* Free context token used as input */
	if (cp->gss_svc_token)
		FREE(cp->gss_svc_token, M_TEMP);
	cp->gss_svc_token = NULL;
	cp->gss_svc_tokenlen = 0;

	if (otokenlen > 0) {
		/* Set context token to gss output token */
		MALLOC(cp->gss_svc_token, u_char *, otokenlen, M_TEMP, M_WAITOK);
		if (cp->gss_svc_token == NULL) {
			printf("nfs_gss_svc_gssd_upcall: could not allocate %d bytes\n", otokenlen);
			vm_map_copy_discard((vm_map_copy_t) otoken);
			return (ENOMEM);
		}
		error = nfs_gss_mach_vmcopyout((vm_map_copy_t) otoken, otokenlen, cp->gss_svc_token);
		if (error) {
			FREE(cp->gss_svc_token, M_TEMP);
			cp->gss_svc_token = NULL;
			return (NFSERR_EAUTH);
		}
		cp->gss_svc_tokenlen = otokenlen;
	}

	return (0);

out:
	FREE(cp->gss_svc_token, M_TEMP);
	cp->gss_svc_tokenlen = 0;
	cp->gss_svc_token = NULL;

	return (NFSERR_EAUTH);
}

/*
 * Validate the sequence number in the credential as described
 * in RFC 2203 Section 5.3.3.1
 *
 * Here the window of valid sequence numbers is represented by
 * a bitmap.  As each sequence number is received, its bit is
 * set in the bitmap.  An invalid sequence number lies below
 * the lower bound of the window, or is within the window but
 * has its bit already set.
 */
static int
nfs_gss_svc_seqnum_valid(struct nfs_gss_svc_ctx *cp, uint32_t seq)
{
	uint32_t *bits = cp->gss_svc_seqbits;
	uint32_t win = cp->gss_svc_seqwin;
	uint32_t i;

	lck_mtx_lock(cp->gss_svc_mtx);

	/*
	 * If greater than the window upper bound,
	 * move the window up, and set the bit.
	 */
	if (seq > cp->gss_svc_seqmax) {
		if (seq - cp->gss_svc_seqmax > win)
			bzero(bits, nfsm_rndup((win + 7) / 8));
		else
			for (i = cp->gss_svc_seqmax + 1; i < seq; i++)
				win_resetbit(bits, i % win);
		win_setbit(bits, seq % win);
		cp->gss_svc_seqmax = seq;
		lck_mtx_unlock(cp->gss_svc_mtx);
		return (1);
	}

	/*
	 * Invalid if below the lower bound of the window
	 */
	if (seq <= cp->gss_svc_seqmax - win) {
		lck_mtx_unlock(cp->gss_svc_mtx);
		return (0);
	}

	/*
	 * In the window, invalid if the bit is already set
	 */
	if (win_getbit(bits, seq % win)) {
		lck_mtx_unlock(cp->gss_svc_mtx);
		return (0);
	}
	win_setbit(bits, seq % win);
	lck_mtx_unlock(cp->gss_svc_mtx);
	return (1);
}

/*
 * Drop a reference to a context
 *
 * Note that it's OK for the context to exist
 * with a refcount of zero.  The refcount isn't
 * checked until we're about to reap an expired one.
 */
void
nfs_gss_svc_ctx_deref(struct nfs_gss_svc_ctx *cp)
{
	lck_mtx_lock(cp->gss_svc_mtx);
	if (cp->gss_svc_refcnt > 0)
		cp->gss_svc_refcnt--;
	else
		printf("nfs_gss_ctx_deref: zero refcount\n");
	lck_mtx_unlock(cp->gss_svc_mtx);
}

/*
 * Called at NFS server shutdown - destroy all contexts
 */
void
nfs_gss_svc_cleanup(void)
{
	struct nfs_gss_svc_ctx_hashhead *head;
	struct nfs_gss_svc_ctx *cp, *ncp;
	int i;

	lck_mtx_lock(nfs_gss_svc_ctx_mutex);

	/*
	 * Run through all the buckets
	 */
	for (i = 0; i < SVC_CTX_HASHSZ; i++) {
		/*
		 * Remove and free all entries in the bucket
		 */
		head = &nfs_gss_svc_ctx_hashtbl[i];
		LIST_FOREACH_SAFE(cp, head, gss_svc_entries, ncp) {
			LIST_REMOVE(cp, gss_svc_entries);
			if (cp->gss_svc_seqbits)
				FREE(cp->gss_svc_seqbits, M_TEMP);
			lck_mtx_destroy(cp->gss_svc_mtx, nfs_gss_svc_grp);
			FREE(cp, M_TEMP);
		}
	}

	lck_mtx_unlock(nfs_gss_svc_ctx_mutex);
}

#endif /* NFSSERVER */


/*************
 * The following functions are used by both client and server.
 */

/*
 * Release a host special port that was obtained by host_get_special_port
 * or one of its macros (host_get_gssd_port in this case).
 * This really should be in a public kpi.
 */

/* This should be in a public header if this routine is not */
extern void ipc_port_release_send(ipc_port_t);
extern ipc_port_t ipc_port_copy_send(ipc_port_t);

static void
host_release_special_port(mach_port_t mp)
{
	if (IPC_PORT_VALID(mp))
		ipc_port_release_send(mp);
}

static mach_port_t
host_copy_special_port(mach_port_t mp)
{
	return (ipc_port_copy_send(mp));
}

/*
 * The token that is sent and received in the gssd upcall
 * has unbounded variable length.  Mach RPC does not pass
 * the token in-line.  Instead it uses page mapping to handle
 * these parameters.  This function allocates a VM buffer
 * to hold the token for an upcall and copies the token
 * (received from the client) into it.  The VM buffer is
 * marked with a src_destroy flag so that the upcall will
 * automatically de-allocate the buffer when the upcall is
 * complete.
 */
static void
nfs_gss_mach_alloc_buffer(u_char *buf, uint32_t buflen, vm_map_copy_t *addr)
{
	kern_return_t kr;
	vm_offset_t kmem_buf;
	vm_size_t tbuflen;

	*addr = NULL;
	if (buf == NULL || buflen == 0)
		return;

	tbuflen = vm_map_round_page(buflen,
				    vm_map_page_mask(ipc_kernel_map));
	kr = vm_allocate(ipc_kernel_map, &kmem_buf, tbuflen, VM_FLAGS_ANYWHERE);
	if (kr != 0) {
		printf("nfs_gss_mach_alloc_buffer: vm_allocate failed\n");
		return;
	}

	kr = vm_map_wire(ipc_kernel_map,
			 vm_map_trunc_page(kmem_buf,
					   vm_map_page_mask(ipc_kernel_map)),
			 vm_map_round_page(kmem_buf + tbuflen,
					   vm_map_page_mask(ipc_kernel_map)),
		VM_PROT_READ|VM_PROT_WRITE, FALSE);
	if (kr != 0) {
		printf("nfs_gss_mach_alloc_buffer: vm_map_wire failed\n");
		return;
	}

	bcopy(buf, (void *) kmem_buf, buflen);
	// Shouldn't need to bzero below since vm_allocate returns zeroed pages
	// bzero(kmem_buf + buflen, tbuflen - buflen);

	kr = vm_map_unwire(ipc_kernel_map,
			   vm_map_trunc_page(kmem_buf,
					     vm_map_page_mask(ipc_kernel_map)),
			   vm_map_round_page(kmem_buf + tbuflen,
					     vm_map_page_mask(ipc_kernel_map)),
			   FALSE);
	if (kr != 0) {
		printf("nfs_gss_mach_alloc_buffer: vm_map_unwire failed\n");
		return;
	}

	kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t) kmem_buf,
		(vm_map_size_t) buflen, TRUE, addr);
	if (kr != 0) {
		printf("nfs_gss_mach_alloc_buffer: vm_map_copyin failed\n");
		return;
	}
}

/*
 * Here we handle a token received from the gssd via an upcall.
 * The received token resides in an allocate VM buffer.
 * We copy the token out of this buffer to a chunk of malloc'ed
 * memory of the right size, then de-allocate the VM buffer.
 */
static int
nfs_gss_mach_vmcopyout(vm_map_copy_t in, uint32_t len, u_char *out)
{
	vm_map_offset_t map_data;
	vm_offset_t data;
	int error;

	error = vm_map_copyout(ipc_kernel_map, &map_data, in);
	if (error)
		return (error);

	data = CAST_DOWN(vm_offset_t, map_data);
	bcopy((void *) data, out, len);
	vm_deallocate(ipc_kernel_map, data, len);

	return (0);
}

/*
 * Encode an ASN.1 token to be wrapped in an RPCSEC_GSS verifier.
 * Returns the size of the token, since it contains a variable
 * length DER encoded size field.
 */
static int
nfs_gss_token_put(
	gss_key_info *ki,
	u_char *alg,
	u_char *p,
	int initiator,
	int datalen,
	u_char *cksum)
{
	static uint32_t seqnum = 0;
	u_char *psave = p;
	u_char plain[8];
	int toklen, i;

	/*
	 * Fill in the token header: 2 octets.
	 * This is 0x06 - an ASN.1 tag for APPLICATION, 0, SEQUENCE
	 * followed by the length of the token: 35 + 0 octets for a
	 * MIC token, or 35 + encrypted octets for a wrap token;
	 */
	*p++ = 0x060;
	toklen = KRB5_SZ_MECH + KRB5_SZ_ALG + KRB5_SZ_SEQ + HASHLEN(ki);
	nfs_gss_der_length_put(&p, toklen + datalen);

	/*
	 * Fill in the DER encoded mech OID for Kerberos v5.
	 * This represents the Kerberos OID 1.2.840.113554.1.2.2
	 * described in RFC 2623, section 4.2
	 */
	bcopy(krb5_mech, p, sizeof(krb5_mech));
	p += sizeof(krb5_mech);

	/*
	 * Now at the token described in RFC 1964, section 1.2.1
	 * Fill in the token ID, integrity algorithm indicator,
	 * for DES MAC MD5, and four filler octets.
	 * The alg string encodes the bytes to represent either
	 * a MIC token or a WRAP token for Kerberos.
	 */
	bcopy(alg, p, KRB5_SZ_ALG);
	p += KRB5_SZ_ALG;

	/*
	 * Now encode the sequence number according to
	 * RFC 1964, section 1.2.1.2 which dictates 4 octets
	 * of sequence number followed by 4 bytes of direction
	 * indicator: 0x00 for initiator or 0xff for acceptor.
	 * We DES CBC encrypt the sequence number using the first
	 * 8 octets of the checksum field as an initialization
	 * vector.
	 * Note that this sequence number is not at all related
	 * to the RPCSEC_GSS protocol sequence number.  This
	 * number is private to the ASN.1 token.  The only
	 * requirement is that it not be repeated in case the
	 * server has replay detection on, which normally should
	 * not be the case, since RFC 2203 section 5.2.3 says that
	 * replay detection and sequence checking must be turned off.
	 */
	seqnum++;
	for (i = 0; i < 4; i++)
		plain[i] = (u_char) ((seqnum >> (i * 8)) & 0xff);
	for (i = 4; i < 8; i++)
		plain[i] = initiator ? 0x00 : 0xff;
	gss_des_crypt(ki, (des_cblock *) plain, (des_cblock *) p, 8,
			(des_cblock *) cksum, NULL, DES_ENCRYPT, KG_USAGE_SEQ);
	p += 8;

	/*
	 * Finally, append the octets of the
	 * checksum of the alg + plaintext data.
	 * The plaintext could be an RPC call header,
	 * the window value, or a sequence number.
	 */
	bcopy(cksum, p, HASHLEN(ki));
	p += HASHLEN(ki);

	return (p - psave);
}

/*
 * Determine size of ASN.1 DER length
 */
static int
nfs_gss_der_length_size(int len)
{
	return
		len < (1 <<  7) ? 1 :
		len < (1 <<  8) ? 2 :
		len < (1 << 16) ? 3 :
		len < (1 << 24) ? 4 : 5;
}

/*
 * Encode an ASN.1 DER length field
 */
static void
nfs_gss_der_length_put(u_char **pp, int len)
{
	int sz = nfs_gss_der_length_size(len);
	u_char *p = *pp;

	if (sz == 1) {
		*p++ = (u_char) len;
	} else {
		*p++ = (u_char) ((sz-1) | 0x80);
		sz -= 1;
		while (sz--)
			*p++ = (u_char) ((len >> (sz * 8)) & 0xff);
	}

	*pp = p;
}

/*
 * Decode an ASN.1 DER length field
 */
static int
nfs_gss_der_length_get(u_char **pp)
{
	u_char *p = *pp;
	uint32_t flen, len = 0;

	flen = *p & 0x7f;

	if ((*p++ & 0x80) == 0)
		len = flen;
	else {
		if (flen > sizeof(uint32_t))
			return (-1);
		while (flen--)
			len = (len << 8) + *p++;
	}
	*pp = p;
	return (len);
}

/*
 * Decode an ASN.1 token from an RPCSEC_GSS verifier.
 */
static int
nfs_gss_token_get(
	gss_key_info *ki,
	u_char *alg,
	u_char *p,
	int initiator,
	uint32_t *len,
	u_char *cksum)
{
	u_char d, plain[8];
	u_char *psave = p;
	int seqnum, i;

	/*
	 * Check that we have a valid token header
	 */
	if (*p++ != 0x60)
		return (AUTH_BADCRED);
	(void) nfs_gss_der_length_get(&p);	// ignore the size

	/*
	 * Check that we have the DER encoded Kerberos v5 mech OID
	 */
	if (bcmp(p, krb5_mech, sizeof(krb5_mech) != 0))
		return (AUTH_BADCRED);
	p += sizeof(krb5_mech);

	/*
	 * Now check the token ID, DES MAC MD5 algorithm
	 * indicator, and filler octets.
	 */
	if (bcmp(p, alg, KRB5_SZ_ALG) != 0)
		return (AUTH_BADCRED);
	p += KRB5_SZ_ALG;

	/*
	 * Now decrypt the sequence number.
	 * Note that the gss decryption uses the first 8 octets
	 * of the checksum field as an initialization vector (p + 8).
	 * Per RFC 2203 section 5.2.2 we don't check the sequence number
	 * in the ASN.1 token because the RPCSEC_GSS protocol has its
	 * own sequence number described in section 5.3.3.1
	 */
	seqnum = 0;
	gss_des_crypt(ki, (des_cblock *)p, (des_cblock *) plain, 8,
			(des_cblock *) (p + 8), NULL, DES_DECRYPT, KG_USAGE_SEQ);
	p += 8;
	for (i = 0; i < 4; i++)
		seqnum |= plain[i] << (i * 8);

	/*
	 * Make sure the direction
	 * indicator octets are correct.
	 */
	d = initiator ? 0x00 : 0xff;
	for (i = 4; i < 8; i++)
		if (plain[i] != d)
			return (AUTH_BADCRED);

	/*
	 * Finally, get the checksum
	 */
	bcopy(p, cksum, HASHLEN(ki));
	p += HASHLEN(ki);

	if (len != NULL)
		*len = p - psave;

	return (0);
}

/*
 * Return the number of bytes in an mbuf chain.
 */
static int
nfs_gss_mchain_length(mbuf_t mhead)
{
	mbuf_t mb;
	int len = 0;

	for (mb = mhead; mb; mb = mbuf_next(mb))
		len += mbuf_len(mb);

	return (len);
}

/*
 * Append an args or results mbuf chain to the header chain
 */
static int
nfs_gss_append_chain(struct nfsm_chain *nmc, mbuf_t mc)
{
	int error = 0;
	mbuf_t mb, tail;

	/* Connect the mbuf chains */
	error = mbuf_setnext(nmc->nmc_mcur, mc);
	if (error)
		return (error);

	/* Find the last mbuf in the chain */
	tail = NULL;
	for (mb = mc; mb; mb = mbuf_next(mb))
		tail = mb;

	nmc->nmc_mcur = tail;
	nmc->nmc_ptr = (caddr_t) mbuf_data(tail) + mbuf_len(tail);
	nmc->nmc_left = mbuf_trailingspace(tail);

	return (0);
}

/*
 * Convert an mbuf chain to an NFS mbuf chain
 */
static void
nfs_gss_nfsm_chain(struct nfsm_chain *nmc, mbuf_t mc)
{
	mbuf_t mb, tail;

	/* Find the last mbuf in the chain */
	tail = NULL;
	for (mb = mc; mb; mb = mbuf_next(mb))
		tail = mb;

	nmc->nmc_mhead = mc;
	nmc->nmc_mcur = tail;
	nmc->nmc_ptr = (caddr_t) mbuf_data(tail) + mbuf_len(tail);
	nmc->nmc_left = mbuf_trailingspace(tail);
	nmc->nmc_flags = 0;
}


/*
 * Compute a checksum over an mbuf chain.
 * Start building an MD5 digest at the given offset and keep
 * going until the end of data in the current mbuf is reached.
 * Then convert the 16 byte MD5 digest to an 8 byte DES CBC
 * checksum.
 */
static void
nfs_gss_cksum_mchain(
	gss_key_info *ki,
	mbuf_t mhead,
	u_char *alg,
	int offset,
	int len,
	u_char *digest)
{
	mbuf_t mb;
	u_char *ptr;
	int left, bytes;
	GSS_DIGEST_CTX context;

	gss_digest_Init(&context, ki);

	/*
	 * Logically prepend the first 8 bytes of the algorithm
	 * field as required by RFC 1964, section 1.2.1.1
	 */
	gss_digest_Update(&context, alg, KRB5_SZ_ALG);

	/*
	 * Move down the mbuf chain until we reach the given
	 * byte offset, then start MD5 on the mbuf data until
	 * we've done len bytes.
	 */

	for (mb = mhead; mb && len > 0; mb = mbuf_next(mb)) {
		ptr  = mbuf_data(mb);
		left = mbuf_len(mb);
		if (offset >= left) {
			/* Offset not yet reached */
			offset -= left;
			continue;
		}
		/* At or beyond offset - checksum data */
		ptr += offset;
		left -= offset;
		offset = 0;

		bytes = left < len ? left : len;
		if (bytes > 0)
			gss_digest_Update(&context, ptr, bytes);
		len -= bytes;
	}

	gss_digest_Final(&context, digest);
}

/*
 * Compute a checksum over an NFS mbuf chain.
 * Start building an MD5 digest at the given offset and keep
 * going until the end of data in the current mbuf is reached.
 * Then convert the 16 byte MD5 digest to an 8 byte DES CBC
 * checksum.
 */
static void
nfs_gss_cksum_chain(
	gss_key_info *ki,
	struct nfsm_chain *nmc,
	u_char *alg,
	int offset,
	int len,
	u_char *cksum)
{
	/*
	 * If the length parameter is zero, then we need
	 * to use the length from the offset to the current
	 * encode/decode offset.
	 */
	if (len == 0)
		len = nfsm_chain_offset(nmc) - offset;

	return (nfs_gss_cksum_mchain(ki, nmc->nmc_mhead, alg, offset, len, cksum));
}

/*
 * Compute a checksum of the sequence number (or sequence window)
 * of an RPCSEC_GSS reply.
 */
static void
nfs_gss_cksum_rep(gss_key_info *ki, uint32_t seqnum, u_char *cksum)
{
	GSS_DIGEST_CTX context;
	uint32_t val = htonl(seqnum);

	gss_digest_Init(&context, ki);

	/*
	 * Logically prepend the first 8 bytes of the MIC
	 * token as required by RFC 1964, section 1.2.1.1
	 */
	gss_digest_Update(&context, ALG_MIC(ki), KRB5_SZ_ALG);

	/*
	 * Compute the digest of the seqnum in network order
	 */
	gss_digest_Update(&context, &val, 4);
	gss_digest_Final(&context, cksum);
}

/*
 * Encrypt or decrypt data in an mbuf chain with des-cbc.
 */
static void
nfs_gss_encrypt_mchain(
	gss_key_info *ki,
	mbuf_t mhead,
	int offset,
	int len,
	int encrypt)
{
	mbuf_t mb, mbn;
	u_char *ptr, *nptr;
	u_char tmp[8], ivec[8];
	int left, left8, remain;


	bzero(ivec, 8);

	/*
	 * Move down the mbuf chain until we reach the given
	 * byte offset, then start encrypting the mbuf data until
	 * we've done len bytes.
	 */

	for (mb = mhead; mb && len > 0; mb = mbn) {
		mbn  = mbuf_next(mb);
		ptr  = mbuf_data(mb);
		left = mbuf_len(mb);
		if (offset >= left) {
			/* Offset not yet reached */
			offset -= left;
			continue;
		}
		/* At or beyond offset - encrypt data */
		ptr += offset;
		left -= offset;
		offset = 0;

		/*
		 * DES or DES3 CBC has to encrypt 8 bytes at a time.
		 * If the number of bytes to be encrypted in this
		 * mbuf isn't some multiple of 8 bytes, encrypt all
		 * the 8 byte blocks, then combine the remaining
		 * bytes with enough from the next mbuf to make up
		 * an 8 byte block and encrypt that block separately,
		 * i.e. that block is split across two mbufs.
		 */
		remain = left % 8;
		left8 = left - remain;
		left = left8 < len ? left8 : len;
		if (left > 0) {
			gss_des_crypt(ki, (des_cblock *) ptr, (des_cblock *) ptr,
					left, &ivec, &ivec, encrypt, KG_USAGE_SEAL);
			len -= left;
		}

		if (mbn && remain > 0) {
			nptr = mbuf_data(mbn);
			offset = 8 - remain;
			bcopy(ptr + left, tmp, remain);		// grab from this mbuf
			bcopy(nptr, tmp + remain, offset);	// grab from next mbuf
			gss_des_crypt(ki, (des_cblock *) tmp, (des_cblock *) tmp, 8,
					&ivec, &ivec, encrypt, KG_USAGE_SEAL);
			bcopy(tmp, ptr + left, remain);		// return to this mbuf
			bcopy(tmp + remain, nptr, offset);	// return to next mbuf
			len -= 8;
		}
	}
}

/*
 * Encrypt or decrypt data in an NFS mbuf chain with des-cbc.
 */
static void
nfs_gss_encrypt_chain(
	gss_key_info *ki,
	struct nfsm_chain *nmc,
	int offset,
	int len,
	int encrypt)
{
	/*
	 * If the length parameter is zero, then we need
	 * to use the length from the offset to the current
	 * encode/decode offset.
	 */
	if (len == 0)
		len = nfsm_chain_offset(nmc) - offset;

	return (nfs_gss_encrypt_mchain(ki, nmc->nmc_mhead, offset, len, encrypt));
}

/*
 * The routines that follow provide abstractions for doing digests and crypto.
 */

static void
gss_digest_Init(GSS_DIGEST_CTX *ctx, gss_key_info *ki)
{
	ctx->type = ki->type;
	switch (ki->type) {
	case NFS_GSS_1DES:	MD5_DESCBC_Init(&ctx->m_ctx, &ki->ks_u.des.gss_sched);
				break;
	case NFS_GSS_3DES:	HMAC_SHA1_DES3KD_Init(&ctx->h_ctx, ki->ks_u.des3.ckey, 0);
				break;
	default:
			printf("gss_digest_Init: Unknown key info type %d\n", ki->type);
	}
}

static void
gss_digest_Update(GSS_DIGEST_CTX *ctx, void *data, size_t len)
{
	switch (ctx->type) {
	case NFS_GSS_1DES:	MD5_DESCBC_Update(&ctx->m_ctx, data, len);
				break;
	case NFS_GSS_3DES:	HMAC_SHA1_DES3KD_Update(&ctx->h_ctx, data, len);
				break;
	}
}

static void
gss_digest_Final(GSS_DIGEST_CTX *ctx, void *digest)
{
	switch (ctx->type) {
	case NFS_GSS_1DES:	MD5_DESCBC_Final(digest, &ctx->m_ctx);
				break;
	case NFS_GSS_3DES:	HMAC_SHA1_DES3KD_Final(digest, &ctx->h_ctx);
				break;
	}
}

static void
gss_des_crypt(gss_key_info *ki, des_cblock *in, des_cblock *out,
		int32_t len, des_cblock *iv, des_cblock *retiv, int encrypt, int usage)
{
	switch (ki->type) {
	case NFS_GSS_1DES:
			{
				des_cbc_key_schedule *sched = ((usage == KG_USAGE_SEAL) ?
							&ki->ks_u.des.gss_sched_Ke :
							&ki->ks_u.des.gss_sched);
				des_cbc_encrypt(in, out, len, sched, iv, retiv, encrypt);
			}
			break;
	case NFS_GSS_3DES:

			des3_cbc_encrypt(in, out, len, &ki->ks_u.des3.gss_sched, iv, retiv, encrypt);
			break;
	}
}

static int
gss_key_init(gss_key_info *ki, uint32_t skeylen)
{
	size_t i;
	int rc;
	des_cblock k[3];

	ki->keybytes = skeylen;
	switch (skeylen) {
	case sizeof(des_cblock):
				ki->type = NFS_GSS_1DES;
				ki->hash_len = MD5_DESCBC_DIGEST_LENGTH;
				ki->ks_u.des.key = (des_cblock *)ki->skey;
				rc = des_cbc_key_sched(ki->ks_u.des.key, &ki->ks_u.des.gss_sched);
				if (rc)
					return (rc);
				for (i = 0; i < ki->keybytes; i++)
					k[0][i] = 0xf0 ^ (*ki->ks_u.des.key)[i];
				rc = des_cbc_key_sched(&k[0], &ki->ks_u.des.gss_sched_Ke);
				break;
	case 3*sizeof(des_cblock):
				ki->type = NFS_GSS_3DES;
				ki->hash_len = SHA_DIGEST_LENGTH;
				ki->ks_u.des3.key = (des_cblock (*)[3])ki->skey;
				des3_derive_key(*ki->ks_u.des3.key, ki->ks_u.des3.ckey,
						KEY_USAGE_DES3_SIGN, KEY_USAGE_LEN);
				rc = des3_cbc_key_sched(*ki->ks_u.des3.key, &ki->ks_u.des3.gss_sched);
				if (rc)
					return (rc);
				break;
	default:
				printf("gss_key_init: Invalid key length %d\n", skeylen);
				rc = EINVAL;
				break;
	}

	return (rc);
}

#if 0
#define DISPLAYLEN 16
#define MAXDISPLAYLEN 256

static void
hexdump(const char *msg, void *data, size_t len)
{
	size_t i, j;
	u_char *d = data;
	char *p, disbuf[3*DISPLAYLEN+1];

	printf("NFS DEBUG %s len=%d:\n", msg, (uint32_t)len);
	if (len > MAXDISPLAYLEN)
		len = MAXDISPLAYLEN;

	for (i = 0; i < len; i += DISPLAYLEN) {
		for (p = disbuf, j = 0; (j + i) < len && j < DISPLAYLEN; j++, p += 3)
			snprintf(p, 4, "%02x ", d[i + j]);
		printf("\t%s\n", disbuf);
	}
}
#endif