]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/nfs/nfs_gss.c
xnu-3248.50.21.tar.gz
[apple/xnu.git] / bsd / nfs / nfs_gss.c
index f1dd81b5f8447635fc0fabbd40a474ad5bd59031..0283fa918f7d540f266f004e2ccd44006bb0e4a2 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2007-2013 Apple Inc. All rights reserved.
+ * Copyright (c) 2007-2015 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
@@ -87,6 +87,7 @@
 #include <sys/ucred.h>
 
 #include <kern/host.h>
+#include <kern/task.h>
 #include <libkern/libkern.h>
 
 #include <mach/task.h>
 #include <nfs/xdr_subs.h>
 #include <nfs/nfsm_subs.h>
 #include <nfs/nfs_gss.h>
-
 #include "nfs_gss_crypto.h"
+#include <mach_assert.h>
+#include <kern/assert.h>
+
+#define ASSERT(EX) assert(EX)
 
 #define NFS_GSS_MACH_MAX_RETRIES 3
 
@@ -186,13 +190,16 @@ static u_char iv0[] = { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 }; // DES
 
 #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 *);
+void           nfs_gss_clnt_ctx_neg_cache_reap(struct nfsmount *);
+static void    nfs_gss_clnt_ctx_clean(struct nfs_gss_clnt_ctx *);
+static int     nfs_gss_clnt_ctx_copy(struct nfs_gss_clnt_ctx *, struct nfs_gss_clnt_ctx **, gss_key_info *);
+static void    nfs_gss_clnt_ctx_destroy(struct nfs_gss_clnt_ctx *);
+static void    nfs_gss_clnt_log_error(struct nfsreq *, struct nfs_gss_clnt_ctx *, uint32_t, uint32_t);
 #endif /* NFSCLIENT */
 
 #if NFSSERVER
@@ -257,25 +264,6 @@ nfs_gss_init(void)
 
 #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.
  *
@@ -292,6 +280,81 @@ nfs_gss_sysok(struct nfsreq *req)
 #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)
 
+#define SAFE_CAST_INTTYPE( type, intval ) \
+       ( (type)(intval)/(sizeof(type) < sizeof(intval) ? 0 : 1) )
+
+uid_t
+nfs_cred_getasid2uid(kauth_cred_t cred)
+{
+       uid_t result = SAFE_CAST_INTTYPE(uid_t, kauth_cred_getasid(cred));
+       return (result);
+}
+
+/*
+ * Debugging
+ */
+static void
+nfs_gss_clnt_ctx_dump(struct nfsmount *nmp)
+{
+       struct nfs_gss_clnt_ctx *cp;
+
+       lck_mtx_lock(&nmp->nm_lock);
+       NFS_GSS_DBG("Enter\n");
+       TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
+               lck_mtx_lock(cp->gss_clnt_mtx);
+               printf("context %d/%d: refcnt = %d, flags = %x\n",
+                      kauth_cred_getasid(cp->gss_clnt_cred),
+                      kauth_cred_getauid(cp->gss_clnt_cred),
+                      cp->gss_clnt_refcnt, cp->gss_clnt_flags);
+               lck_mtx_unlock(cp->gss_clnt_mtx);
+       }
+       NFS_GSS_DBG("Exit\n");
+       lck_mtx_unlock(&nmp->nm_lock);
+}
+
+static char *
+nfs_gss_clnt_ctx_name(struct nfsmount *nmp, struct nfs_gss_clnt_ctx *cp, char *buf, int len)
+{
+       char *np;
+       int nlen;
+       const char *server = "";
+
+       if (nmp && nmp->nm_mountp)
+               server = vfs_statfs(nmp->nm_mountp)->f_mntfromname;
+
+       if (cp == NULL) {
+               snprintf(buf, len, "[%s] NULL context", server);
+               return (buf);
+       }
+
+       if (cp->gss_clnt_principal && !cp->gss_clnt_display) {
+               np = (char *)cp->gss_clnt_principal;
+               nlen = cp->gss_clnt_prinlen;
+       } else {
+               np = cp->gss_clnt_display;
+               nlen = np ? strlen(cp->gss_clnt_display) : 0;
+       }
+       if (nlen)
+               snprintf(buf, len, "[%s] %.*s %d/%d %s", server, nlen, np,
+                        kauth_cred_getasid(cp->gss_clnt_cred),
+                        kauth_cred_getuid(cp->gss_clnt_cred),
+                        cp->gss_clnt_principal ? "" : "[from default cred] ");
+       else
+               snprintf(buf, len, "[%s] using default %d/%d ", server,
+                        kauth_cred_getasid(cp->gss_clnt_cred),
+                        kauth_cred_getuid(cp->gss_clnt_cred));
+       return (buf);
+}
+
+#define NFS_CTXBUFSZ 80
+#define NFS_GSS_CTX(req, cp) nfs_gss_clnt_ctx_name((req)->r_nmp, cp ? cp : (req)->r_gss_ctx, CTXBUF, sizeof(CTXBUF))
+
+#define NFS_GSS_CLNT_CTX_DUMP(nmp)             \
+       do {                  \
+               if (NFS_GSS_ISDBG && (NFS_DEBUG_FLAGS & 0x2))   \
+                       nfs_gss_clnt_ctx_dump((nmp));   \
+       } while (0)
+
 static int
 nfs_gss_clnt_ctx_cred_match(kauth_cred_t cred1, kauth_cred_t cred2)
 {
@@ -300,26 +363,162 @@ nfs_gss_clnt_ctx_cred_match(kauth_cred_t cred1, kauth_cred_t cred2)
        return (0);
 }
 
-                       
+/*
+ * Busy the mount for each principal set on the mount
+ * so that the automounter will not unmount the file
+ * system underneath us. With out this, if an unmount
+ * occurs the principal that is set for an audit session
+ * will be lost and we may end up with a different identity.
+ *
+ * Note setting principals on the mount is a bad idea. This
+ * really should be handle by KIM (Kerberos Identity Management)
+ * so that defaults can be set by service identities.
+ */
+
+static void
+nfs_gss_clnt_mnt_ref(struct nfsmount *nmp)
+{
+       int error;
+       vnode_t rvp;
+
+       if (nmp == NULL ||
+           !(vfs_flags(nmp->nm_mountp) & MNT_AUTOMOUNTED))
+               return;
+
+       error = VFS_ROOT(nmp->nm_mountp, &rvp, NULL);
+       if (!error) {
+               vnode_ref(rvp);
+               vnode_put(rvp);
+       }
+}
+
+/*
+ * Unbusy the mout. See above comment,
+ */
+
+static void
+nfs_gss_clnt_mnt_rele(struct nfsmount *nmp)
+{
+       int error;
+       vnode_t rvp;
+
+       if (nmp == NULL ||
+           !(vfs_flags(nmp->nm_mountp) & MNT_AUTOMOUNTED))
+               return;
+
+       error = VFS_ROOT(nmp->nm_mountp, &rvp, NULL);
+       if (!error) {
+               vnode_rele(rvp);
+               vnode_put(rvp);
+       }
+}
+
+int nfs_root_steals_ctx = 1;
+
 static int
-nfs_gss_clnt_ctx_find(struct nfsreq *req)
+nfs_gss_clnt_ctx_find_principal(struct nfsreq *req, uint8_t *principal, uint32_t plen, uint32_t nt)
 {
        struct nfsmount *nmp = req->r_nmp;
        struct nfs_gss_clnt_ctx *cp;
+       struct nfsreq treq;
        int error = 0;
+       struct timeval now;
+       gss_key_info *ki;
+       char CTXBUF[NFS_CTXBUFSZ];
 
+       bzero(&treq, sizeof (struct nfsreq));
+       treq.r_nmp = nmp;
+
+       microuptime(&now);
        lck_mtx_lock(&nmp->nm_lock);
        TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
+               lck_mtx_lock(cp->gss_clnt_mtx);
+               if (cp->gss_clnt_flags & GSS_CTX_DESTROY) {
+                       NFS_GSS_DBG("Found destroyed context %s refcnt = %d continuing\n",
+                                   NFS_GSS_CTX(req, cp),
+                                   cp->gss_clnt_refcnt);
+                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                       continue;
+               }
                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);
+                       if (nmp->nm_gsscl.tqh_first != cp) {
+                               TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);
+                               TAILQ_INSERT_HEAD(&nmp->nm_gsscl, cp, gss_clnt_entries);
+                       }
+                       if (principal) {
+                               /*
+                                * If we have a principal, but it does not match the current cred
+                                * mark it for removal
+                                */
+                               if (cp->gss_clnt_prinlen != plen || cp->gss_clnt_prinnt != nt ||
+                                   bcmp(cp->gss_clnt_principal, principal, plen) != 0) {
+                                       cp->gss_clnt_flags |= (GSS_CTX_INVAL | GSS_CTX_DESTROY);
+                                       cp->gss_clnt_refcnt++;
+                                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                                       NFS_GSS_DBG("Marking %s for deletion because %s does not match\n",
+                                                   NFS_GSS_CTX(req, cp), principal);
+                                       NFS_GSS_DBG("len = (%d,%d), nt = (%d,%d)\n", cp->gss_clnt_prinlen, plen,
+                                                   cp->gss_clnt_prinnt, nt);
+                                       treq.r_gss_ctx  = cp;
+                                       cp = NULL;
+                                       break;
+                               }
+                       }
+                       if (cp->gss_clnt_flags & GSS_CTX_INVAL) {
+                               /*
+                                * If we're still being used and we're not expired
+                                * just return and don't bother gssd again. Note if
+                                * gss_clnt_nctime is zero it is about to be set to now.
+                                */
+                               if (cp->gss_clnt_nctime + GSS_NEG_CACHE_TO >= now.tv_sec || cp->gss_clnt_nctime == 0) {
+                                       NFS_GSS_DBG("Context %s (refcnt = %d) not expired returning EAUTH nctime = %ld now = %ld\n",
+                                                   NFS_GSS_CTX(req, cp), cp->gss_clnt_refcnt, cp->gss_clnt_nctime, now.tv_sec);
+                                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                                       lck_mtx_unlock(&nmp->nm_lock);
+                                       return (NFSERR_EAUTH);
+                               }
+                               if (cp->gss_clnt_refcnt) {
+                                       struct nfs_gss_clnt_ctx *ncp;
+                                       /*
+                                        * If this context has references, we can't use it so we mark if for
+                                        * destruction and create a new context based on this one in the
+                                        * same manner as renewing one.
+                                        */
+                                       cp->gss_clnt_flags |= GSS_CTX_DESTROY;
+                                       NFS_GSS_DBG("Context %s has expired but we still have %d references\n",
+                                                   NFS_GSS_CTX(req, cp), cp->gss_clnt_refcnt);
+                                       error = nfs_gss_clnt_ctx_copy(cp, &ncp, NULL);
+                                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                                       if (error) {
+                                               lck_mtx_unlock(&nmp->nm_lock);
+                                               return (error);
+                                       }
+                                       cp = ncp;
+                                       break;
+                               } else {
+                                       /* cp->gss_clnt_kinfo should be NULL here */
+                                       if (cp->gss_clnt_kinfo) {
+                                               FREE(cp->gss_clnt_kinfo, M_TEMP);
+                                               cp->gss_clnt_kinfo = NULL;
+                                       }
+                                       if (cp->gss_clnt_nctime)
+                                               nmp->nm_ncentries--;
+                                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                                       TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);
+                                       break;
+                               }
+                       }
+                       /* Found a valid context to return */
+                       cp->gss_clnt_refcnt++;
+                       req->r_gss_ctx = cp;
+                       lck_mtx_unlock(cp->gss_clnt_mtx);
                        lck_mtx_unlock(&nmp->nm_lock);
                        return (0);
                }
+               lck_mtx_unlock(cp->gss_clnt_mtx);
        }
 
-       if (kauth_cred_getuid(req->r_cred) == 0) {
+       if (!cp && nfs_root_steals_ctx && principal == NULL && kauth_cred_getuid(req->r_cred) == 0) {
                /*
                 * If superuser is trying to get access, then co-opt
                 * the first valid context in the list.
@@ -328,101 +527,90 @@ nfs_gss_clnt_ctx_find(struct nfsreq *req)
                 * 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)) {
+                       if (!(cp->gss_clnt_flags & (GSS_CTX_INVAL|GSS_CTX_DESTROY))) {
                                nfs_gss_clnt_ctx_ref(req, cp);
                                lck_mtx_unlock(&nmp->nm_lock);
+                               NFS_GSS_DBG("Root stole context %s\n", NFS_GSS_CTX(req, NULL));
                                return (0);
                        }
                }
        }
 
-       /*
-        * Not found - create a new context
-        */
+       MALLOC(ki, gss_key_info *, sizeof (gss_key_info), M_TEMP, M_WAITOK|M_ZERO);
+       if (ki == NULL) {
+               lck_mtx_unlock(&nmp->nm_lock);
+               return (ENOMEM);
+       }
+
+       NFS_GSS_DBG("Context %s%sfound in Neg Cache @  %ld\n",
+                   NFS_GSS_CTX(req, cp),
+                   cp == NULL ? " not " : "",
+                   cp == NULL ? 0L : cp->gss_clnt_nctime);
 
        /*
-        * 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.
+        * Not found - create a new context
         */
-       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);
+               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_kinfo = ki;
+               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_ptime = now.tv_sec - GSS_PRINT_DELAY;
+               if (principal) {
+                       MALLOC(cp->gss_clnt_principal, uint8_t *, plen+1, M_TEMP, M_WAITOK|M_ZERO);
+                       memcpy(cp->gss_clnt_principal, principal, plen);
+                       cp->gss_clnt_prinlen = plen;
+                       cp->gss_clnt_prinnt = nt;
+                       cp->gss_clnt_flags |= GSS_CTX_STICKY;
+                       nfs_gss_clnt_mnt_ref(nmp);
+               }
+       } else {
+               cp->gss_clnt_kinfo = ki;
+               nfs_gss_clnt_ctx_clean(cp);
+               if (principal) {
+                       /*
+                        * If we have a principal and we found a matching audit
+                        * session, then to get here, the principal had to match.
+                        * In walking the context list if it has a principal
+                        * or the principal is not set then we mark the context
+                        * for destruction and set cp to NULL and we fall to the
+                        * if clause above. If the context still has references
+                        * again we copy the context which will preserve the principal
+                        * and we end up here with the correct principal set.
+                        * If we don't have references the the principal must have
+                        * match and we will fall through here.
+                        */
+                       cp->gss_clnt_flags |= GSS_CTX_STICKY;
+               }
        }
 
-       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);
+       TAILQ_INSERT_HEAD(&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)
+       if (error) {
+               NFS_GSS_DBG("nfs_gss_clnt_ctx_init_retry returned %d for %s\n", error, NFS_GSS_CTX(req, cp));
                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);
        }
 
+       /* Remove any old matching contex that had a different principal */
+       nfs_gss_clnt_ctx_unref(&treq);
+
        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)
+nfs_gss_clnt_ctx_find(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);
+       return (nfs_gss_clnt_ctx_find_principal(req, NULL, 0, 0));
 }
 
 /*
@@ -444,14 +632,14 @@ nfs_gss_clnt_cred_put(struct nfsreq *req, struct nfsm_chain *nmc, mbuf_t args)
        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) {
                /*
@@ -465,30 +653,6 @@ retry:
        }
        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
@@ -507,7 +671,7 @@ retry:
        }
        lck_mtx_unlock(cp->gss_clnt_mtx);
 
-       ki = &cp->gss_clnt_kinfo;
+       ki = cp->gss_clnt_kinfo;
        if (cp->gss_clnt_flags & GSS_CTX_COMPLETE) {
                /*
                 * Get a sequence number for this request.
@@ -702,7 +866,7 @@ nfs_gss_clnt_verf_get(
        struct gss_seq *gsp;
        uint32_t reslen, start, cksumlen, toklen;
        int error = 0;
-       gss_key_info *ki = &cp->gss_clnt_kinfo;
+       gss_key_info *ki = cp->gss_clnt_kinfo;
 
        reslen = cksumlen = 0;
        *accepted_statusp = 0;
@@ -719,8 +883,7 @@ nfs_gss_clnt_verf_get(
        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)
+               if (cp->gss_clnt_flags & GSS_CTX_COMPLETE)
                        return (NFSERR_EAUTH);
                if (verflen > 0)
                        nfsm_chain_adv(error, nmc, nfsm_rndup(verflen));
@@ -921,7 +1084,7 @@ nfsmout:
  * 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)
 {
@@ -929,7 +1092,7 @@ nfs_gss_clnt_args_restore(struct nfsreq *req)
        struct nfsm_chain mchain, *nmc = &mchain;
        int len, error = 0;
 
-       if (cp == NULL) 
+       if (cp == NULL)
                return (NFSERR_EAUTH);
 
        if ((cp->gss_clnt_flags & GSS_CTX_COMPLETE) == 0)
@@ -968,7 +1131,7 @@ nfs_gss_clnt_args_restore(struct nfsreq *req)
                 */
                len = req->r_gss_arglen;
                len += len % 8 > 0 ? 4 : 8;                     // add DES padding length
-               nfs_gss_encrypt_chain(&cp->gss_clnt_kinfo, nmc,
+               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)
@@ -999,15 +1162,16 @@ nfs_gss_clnt_ctx_init(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
        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;
+       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_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;
@@ -1063,11 +1227,7 @@ retry:
                        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;
        }
 
@@ -1078,9 +1238,6 @@ retry:
        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
@@ -1118,12 +1275,12 @@ retry:
        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) 
+       if (error == ENEEDAUTH)
                return (error);
 
        /*
@@ -1166,7 +1323,7 @@ nfs_gss_clnt_ctx_init_retry(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
        int retries = 0;
        int timeo = NFS_TRYLATERDEL;
 
-       if (nmp == NULL) {
+       if (nfs_mount_gone(nmp)) {
                error = ENXIO;
                goto bad;
        }
@@ -1188,7 +1345,7 @@ nfs_gss_clnt_ctx_init_retry(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
 
                retries++;
                /* If it's a soft mount just give up after a while */
-               if (NMFLAG(nmp, SOFT) && (retries > nmp->nm_retry)) {
+               if ((NMFLAG(nmp, SOFT) || (req->r_flags & R_SOFT)) && (retries > nmp->nm_retry)) {
                        error = ETIMEDOUT;
                        goto bad;
                }
@@ -1230,9 +1387,10 @@ nfs_gss_clnt_ctx_callserver(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
 {
        struct nfsm_chain nmreq, nmrep;
        int error = 0, status;
+       uint32_t major = cp->gss_clnt_major, minor = cp->gss_clnt_minor;
        int sz;
 
-       if (!req->r_nmp)
+       if (nfs_mount_gone(req->r_nmp))
                return (ENXIO);
        nfsm_chain_null(&nmreq);
        nfsm_chain_null(&nmrep);
@@ -1292,17 +1450,10 @@ nfs_gss_clnt_ctx_callserver(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
         */
        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);
+               printf("nfs_gss_clnt_ctx_callserver: gss_clnt_major = %d\n", cp->gss_clnt_major);
+               nfs_gss_clnt_log_error(req, cp, major, minor);
+               
        }
 
 nfsmout:
@@ -1330,7 +1481,7 @@ nfs_gss_clnt_svcname(struct nfsmount *nmp, gssd_nametype *nt, uint32_t *len)
        char *svcname, *d, *server;
        int lindx, sindx;
 
-       if (!nmp)
+       if (nfs_mount_gone(nmp))
                return (NULL);
 
        if (nmp->nm_sprinc) {
@@ -1400,10 +1551,79 @@ nfs_gss_clnt_get_upcall_port(kauth_cred_t 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);
+       host_release_special_port(gssd_host_port);
 
        return (uc_port);
 }
 
+
+static void
+nfs_gss_clnt_log_error(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp, uint32_t major, uint32_t minor)
+{
+#define GETMAJERROR(x) (((x) >> GSS_C_ROUTINE_ERROR_OFFSET) & GSS_C_ROUTINE_ERROR_MASK)        
+       struct nfsmount *nmp = req->r_nmp;
+       char who[] = "client";
+       uint32_t gss_error = GETMAJERROR(cp->gss_clnt_major);
+       const char *procn = "unkown";
+       proc_t proc;
+       pid_t pid = -1;
+       struct timeval now;
+
+       if (req->r_thread) {
+               proc = (proc_t)get_bsdthreadtask_info(req->r_thread);
+               if (proc != NULL && (proc->p_fd == NULL || (proc->p_lflag & P_LVFORK)))
+                       proc = NULL;
+               if (proc) {
+                       if (*proc->p_comm)
+                               procn = proc->p_comm;
+                       pid = proc->p_pid;
+               }
+       } else {
+               procn = "kernproc";
+               pid = 0;
+       }
+               
+       microuptime(&now);
+       if ((cp->gss_clnt_major != major || cp->gss_clnt_minor != minor ||
+            cp->gss_clnt_ptime + GSS_PRINT_DELAY < now.tv_sec) &&
+           (nmp->nm_state & NFSSTA_MOUNTED)) {
+               /* 
+                * Will let gssd do some logging in hopes that it can translate
+                * the minor code.
+                */
+               if (cp->gss_clnt_minor && cp->gss_clnt_minor != minor) {
+                       (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;
+
+               /*
+                *%%% It would be really nice to get the terminal from the proc or auditinfo_addr struct and print that here.
+                */
+               printf("NFS: gssd auth failure by %s on audit session %d uid %d proc %s/%d for mount %s. Error: major = %d minor = %d\n",
+                      cp->gss_clnt_display ? cp->gss_clnt_display : who, kauth_cred_getasid(req->r_cred), kauth_cred_getuid(req->r_cred),
+                      procn, pid, vfs_statfs(nmp->nm_mountp)->f_mntfromname, gss_error, (int32_t)cp->gss_clnt_minor);
+               cp->gss_clnt_ptime = now.tv_sec;
+               switch (gss_error) {
+               case 7: printf("NFS: gssd does not have credentials for session %d/%d, (kinit)?\n",
+                              kauth_cred_getasid(req->r_cred), kauth_cred_getauid(req->r_cred));
+                       break;
+               case 11: printf("NFS: gssd has expired credentals for session %d/%d, (kinit)?\n",
+                              kauth_cred_getasid(req->r_cred), kauth_cred_getauid(req->r_cred));
+                       break;
+               }
+       } else {
+               NFS_GSS_DBG("NFS: gssd auth failure by %s on audit session %d uid %d proc %s/%d for mount %s. Error: major = %d minor = %d\n",
+                           cp->gss_clnt_display ? cp->gss_clnt_display : who, kauth_cred_getasid(req->r_cred), kauth_cred_getuid(req->r_cred),
+                           procn, pid, vfs_statfs(nmp->nm_mountp)->f_mntfromname, gss_error, (int32_t)cp->gss_clnt_minor);
+       }
+}
+
 /*
  * Make an upcall to the gssd using Mach RPC
  * The upcall is made using a host special port.
@@ -1431,6 +1651,7 @@ nfs_gss_clnt_gssd_upcall(struct nfsreq *req, struct nfs_gss_clnt_ctx *cp)
        uint32_t ret_flags;
        uint32_t nfs_1des = (cp->gss_clnt_gssd_flags & GSSD_NFS_1DES);
        struct nfsmount *nmp;
+       uint32_t major = cp->gss_clnt_major, minor = cp->gss_clnt_minor;
        
        /*
         * NFS currently only supports default principals or
@@ -1541,21 +1762,7 @@ skip:
         */
        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);
+               nfs_gss_clnt_log_error(req, cp, major, minor);
        }
 
        if (skeylen > 0) {
@@ -1566,13 +1773,13 @@ skip:
                        goto out;
                }
                error = nfs_gss_mach_vmcopyout((vm_map_copy_t) okey, skeylen, 
-                               cp->gss_clnt_kinfo.skey);
+                               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);
+               error = gss_key_init(cp->gss_clnt_kinfo, skeylen);
                if (error)
                        goto out;
        }
@@ -1690,6 +1897,11 @@ nfs_gss_clnt_ctx_unref(struct nfsreq *req)
 {
        struct nfsmount *nmp = req->r_nmp;
        struct nfs_gss_clnt_ctx *cp = req->r_gss_ctx;
+       int on_neg_cache = 0;
+       int neg_cache = 0;
+       int destroy = 0;
+       struct timeval now;
+       char CTXBUF[NFS_CTXBUFSZ];
 
        if (cp == NULL)
                return;
@@ -1697,51 +1909,253 @@ nfs_gss_clnt_ctx_unref(struct nfsreq *req)
        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)
+       if (--cp->gss_clnt_refcnt < 0)
+               panic("Over release of gss context!\n");
+
+       if (cp->gss_clnt_refcnt == 0) {
+               if ((cp->gss_clnt_flags & GSS_CTX_INVAL) &&
+                   cp->gss_clnt_kinfo) {
+                       FREE(cp->gss_clnt_kinfo, M_TEMP);
+                       cp->gss_clnt_kinfo = NULL;
+               }
+               if (cp->gss_clnt_flags & GSS_CTX_DESTROY) {
+                       destroy = 1;
+                       if (cp->gss_clnt_flags & GSS_CTX_STICKY)
+                               nfs_gss_clnt_mnt_rele(nmp);
+                       if (cp->gss_clnt_nctime)
+                               on_neg_cache = 1;
+               }
+       }
+       if (!destroy && cp->gss_clnt_nctime == 0 &&
+           (cp->gss_clnt_flags & GSS_CTX_INVAL)) {
+               microuptime(&now);
+               cp->gss_clnt_nctime = now.tv_sec;
+               neg_cache = 1;
+       }
+       lck_mtx_unlock(cp->gss_clnt_mtx);
+       if (destroy) {
+               NFS_GSS_DBG("Destroying context %s\n", NFS_GSS_CTX(req, cp));
+               if (nmp) {
                        lck_mtx_lock(&nmp->nm_lock);
-               nfs_gss_clnt_ctx_remove(nmp, cp);
-               if (nmp)
+                       if (cp->gss_clnt_entries.tqe_next != NFSNOLIST) {
+                               TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);
+                       }
+                       if (on_neg_cache) {
+                               nmp->nm_ncentries--;
+                       }
+                       lck_mtx_unlock(&nmp->nm_lock);
+               }
+               nfs_gss_clnt_ctx_destroy(cp);
+       } else if (neg_cache) {
+               NFS_GSS_DBG("Entering context %s into negative cache\n", NFS_GSS_CTX(req, cp));
+               if (nmp) {
+                       lck_mtx_lock(&nmp->nm_lock);
+                       nmp->nm_ncentries++;
+                       nfs_gss_clnt_ctx_neg_cache_reap(nmp);
                        lck_mtx_unlock(&nmp->nm_lock);
+               }
+       }
+       NFS_GSS_CLNT_CTX_DUMP(nmp);
+}
 
-               return;
+/*
+ * Try and reap any old negative cache entries.
+ * cache queue.
+ */
+void
+nfs_gss_clnt_ctx_neg_cache_reap(struct nfsmount *nmp)
+{
+       struct nfs_gss_clnt_ctx *cp, *tcp;
+       struct timeval now;
+       int reaped = 0;
+
+       NFS_GSS_DBG("Reaping contexts ncentries = %d\n", nmp->nm_ncentries);
+       /* Try and reap old, unreferenced, expired contexts */
+
+       TAILQ_FOREACH_SAFE(cp, &nmp->nm_gsscl, gss_clnt_entries, tcp) {
+               int destroy = 0;
+
+               /* Don't reap STICKY contexts */
+               if ((cp->gss_clnt_flags & GSS_CTX_STICKY) ||
+                   !(cp->gss_clnt_flags & GSS_CTX_INVAL))
+                       continue;
+               /* Keep up to GSS_MAX_NEG_CACHE_ENTRIES */
+               if (nmp->nm_ncentries <= GSS_MAX_NEG_CACHE_ENTRIES)
+                       break;
+               /* Contexts too young */
+               if (cp->gss_clnt_nctime + GSS_NEG_CACHE_TO >= now.tv_sec)
+                       continue;
+               /* Not referenced, remove it. */
+               lck_mtx_lock(cp->gss_clnt_mtx);
+               if (cp->gss_clnt_refcnt == 0) {
+                       cp->gss_clnt_flags |= GSS_CTX_DESTROY;
+                       destroy = 1;
+               }
+               lck_mtx_unlock(cp->gss_clnt_mtx);
+               if (destroy) {
+                       TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);
+                       nmp->nm_ncentries++;
+                       reaped++;
+                       nfs_gss_clnt_ctx_destroy(cp);
+               }
        }
-       lck_mtx_unlock(cp->gss_clnt_mtx);
+       NFS_GSS_DBG("Reaped %d contexts ncentries = %d\n", reaped, nmp->nm_ncentries);
 }
 
 /*
- * Remove a context
+ * Clean a context to be cached
  */
 static void
-nfs_gss_clnt_ctx_remove(struct nfsmount *nmp, struct nfs_gss_clnt_ctx *cp)
+nfs_gss_clnt_ctx_clean(struct nfs_gss_clnt_ctx *cp)
 {
+       /* Preserve gss_clnt_mtx */
+       assert(cp->gss_clnt_thread == NULL);  /* Will be set to this thread */
+       /* gss_clnt_entries  we should not be on any list at this point */
+       cp->gss_clnt_flags = 0;
+       /* gss_clnt_refcnt should be zero */
+       assert(cp->gss_clnt_refcnt == 0);
        /*
-        * If dequeueing, assume nmp->nm_lock is held
+        * We are who we are preserve:
+        * gss_clnt_cred
+        * gss_clnt_principal
+        * gss_clnt_prinlen
+        * gss_clnt_prinnt
+        * gss_clnt_desplay
         */
-       if (nmp != NULL)
-               TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);
+       /* gss_clnt_proc will be set in nfs_gss_clnt_ctx_init */
+       cp->gss_clnt_seqnum = 0;
+       /* Preserve gss_clnt_service, we're not changing flavors */
+       if (cp->gss_clnt_handle) {
+               FREE(cp->gss_clnt_handle, M_TEMP);
+               cp->gss_clnt_handle = NULL;
+       }
+       cp->gss_clnt_handle_len = 0;
+       cp->gss_clnt_nctime = 0;
+       cp->gss_clnt_seqwin = 0;
+       if (cp->gss_clnt_seqbits) {
+               FREE(cp->gss_clnt_seqbits, M_TEMP);
+               cp->gss_clnt_seqbits = NULL;
+       }
+       /* Preserve gss_clnt_mport. Still talking to the same gssd */
+       if (cp->gss_clnt_verf) {
+               FREE(cp->gss_clnt_verf, M_TEMP);
+               cp->gss_clnt_verf = NULL;
+       }
+       /* Service name might change on failover, so reset it */
+       if (cp->gss_clnt_svcname) {
+               FREE(cp->gss_clnt_svcname, M_TEMP);
+               cp->gss_clnt_svcname = NULL;
+               cp->gss_clnt_svcnt = 0;
+       }
+       cp->gss_clnt_svcnamlen = 0;
+       cp->gss_clnt_cred_handle = 0;
+       cp->gss_clnt_context = 0;
+       if (cp->gss_clnt_token) {
+               FREE(cp->gss_clnt_token, M_TEMP);
+               cp->gss_clnt_token = NULL;
+       }
+       cp->gss_clnt_tokenlen = 0;
+       if (cp->gss_clnt_kinfo)
+               bzero(cp->gss_clnt_kinfo, sizeof(gss_key_info));
+       /*
+        * Preserve:
+        * gss_clnt_gssd_flags
+        * gss_clnt_major
+        * gss_clnt_minor
+        * gss_clnt_ptime
+        */
+}
+
+/*
+ * Copy a source context to a new context. This is used to create a new context
+ * with the identity of the old context for renewal. The old context is invalid
+ * at this point but may have reference still to it, so it is not safe to use that
+ * context.
+ */
+static int
+nfs_gss_clnt_ctx_copy(struct nfs_gss_clnt_ctx *scp, struct nfs_gss_clnt_ctx **dcpp, gss_key_info *ki)
+{
+       struct nfs_gss_clnt_ctx *dcp;
+
+       *dcpp = (struct nfs_gss_clnt_ctx *)NULL;
+       MALLOC(dcp, struct nfs_gss_clnt_ctx *, sizeof (struct nfs_gss_clnt_ctx), M_TEMP, M_WAITOK);
+       if (dcp == NULL)
+               return (ENOMEM);
+       bzero(dcp, sizeof (struct nfs_gss_clnt_ctx));
+       if (ki == NULL) {
+               MALLOC(dcp->gss_clnt_kinfo, gss_key_info *, sizeof (gss_key_info), M_TEMP, M_WAITOK);
+               if (dcp->gss_clnt_kinfo == NULL) {
+                       FREE(dcp, M_TEMP);
+                       return (ENOMEM);
+               }
+       } else {
+               dcp->gss_clnt_kinfo = ki;
+       }
+       bzero(dcp->gss_clnt_kinfo, sizeof (gss_key_info));
+       dcp->gss_clnt_mtx = lck_mtx_alloc_init(nfs_gss_clnt_grp, LCK_ATTR_NULL);
+       dcp->gss_clnt_cred = scp->gss_clnt_cred;
+       kauth_cred_ref(dcp->gss_clnt_cred);
+       dcp->gss_clnt_prinlen = scp->gss_clnt_prinlen;
+       dcp->gss_clnt_prinnt = scp->gss_clnt_prinnt;
+       if (scp->gss_clnt_principal) {
+               MALLOC(dcp->gss_clnt_principal, uint8_t *, dcp->gss_clnt_prinlen, M_TEMP, M_WAITOK | M_ZERO);
+               if (dcp->gss_clnt_principal == NULL) {
+                       FREE(dcp->gss_clnt_kinfo, M_TEMP);
+                       FREE(dcp, M_TEMP);
+                       return (ENOMEM);
+               }
+               bcopy(scp->gss_clnt_principal, dcp->gss_clnt_principal, dcp->gss_clnt_prinlen);
+       }
+       /* Note we don't preserve the display name, that will be set by a successful up call */
+       dcp->gss_clnt_service = scp->gss_clnt_service;
+       dcp->gss_clnt_mport = host_copy_special_port(scp->gss_clnt_mport);
+       /*  gss_clnt_kinfo allocated above */
+       dcp->gss_clnt_gssd_flags = scp->gss_clnt_gssd_flags;
+       dcp->gss_clnt_major = scp->gss_clnt_major;
+       dcp->gss_clnt_minor = scp->gss_clnt_minor;
+       dcp->gss_clnt_ptime = scp->gss_clnt_ptime;
+
+       *dcpp = dcp;
+
+       return (0);
+}
+
+/*
+ * Remove a context
+ */
+static void
+nfs_gss_clnt_ctx_destroy(struct nfs_gss_clnt_ctx *cp)
+{
+       NFS_GSS_DBG("Destroying context %d/%d\n",
+                   kauth_cred_getasid(cp->gss_clnt_cred),
+                   kauth_cred_getauid(cp->gss_clnt_cred));
 
        host_release_special_port(cp->gss_clnt_mport);
+       cp->gss_clnt_mport = IPC_PORT_NULL;
 
-       if (cp->gss_clnt_mtx)
+       if (cp->gss_clnt_mtx) {
                lck_mtx_destroy(cp->gss_clnt_mtx, nfs_gss_clnt_grp);
+               cp->gss_clnt_mtx = (lck_mtx_t *)NULL;
+       }
        if (IS_VALID_CRED(cp->gss_clnt_cred))
                kauth_cred_unref(&cp->gss_clnt_cred);
-       if (cp->gss_clnt_principal)
+       cp->gss_clnt_entries.tqe_next = NFSNOLIST;
+       cp->gss_clnt_entries.tqe_prev = NFSNOLIST;
+       if (cp->gss_clnt_principal) {
                FREE(cp->gss_clnt_principal, M_TEMP);
-       if (cp->gss_clnt_display)
+               cp->gss_clnt_principal = NULL;
+       }
+       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);
+               cp->gss_clnt_display = NULL;
+       }
+       if (cp->gss_clnt_kinfo) {
+               FREE(cp->gss_clnt_kinfo, M_TEMP);
+               cp->gss_clnt_kinfo = NULL;
+       }
+
+       nfs_gss_clnt_ctx_clean(cp);
+
        FREE(cp, M_TEMP);
 }
 
@@ -1754,71 +2168,57 @@ 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;
+       struct nfsmount *nmp;
        int error = 0;
-       kauth_cred_t saved_cred;
-       mach_port_t saved_mport;
+       char CTXBUF[NFS_CTXBUFSZ];
 
        if (cp == NULL)
                return (0);
 
+       if (req->r_nmp == NULL)
+               return (ENXIO);
+       nmp = req->r_nmp;
+
        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;
+       cp->gss_clnt_flags |= (GSS_CTX_INVAL | GSS_CTX_DESTROY);
 
-       /*
-        * 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);
 
+       error =  nfs_gss_clnt_ctx_copy(cp, &ncp, NULL);
+       NFS_GSS_DBG("Renewing context %s\n", NFS_GSS_CTX(req, ncp));
+       nfs_gss_clnt_ctx_unref(req);
+       if (error)
+               return (error);
+
+       lck_mtx_lock(&nmp->nm_lock);
        /*
-        * Create a new context
+        * Note we don't bother taking the new context mutex as we're
+        * not findable at the moment.
         */
-       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);
+       TAILQ_INSERT_HEAD(&nmp->nm_gsscl, ncp, gss_clnt_entries);
+       lck_mtx_unlock(&nmp->nm_lock);
 
        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.
@@ -1830,28 +2230,32 @@ nfs_gss_clnt_ctx_unmount(struct nfsmount *nmp)
        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++;
+       if (!nmp)
+               return;
+
+
+       lck_mtx_lock(&nmp->nm_lock);
+       while((cp = TAILQ_FIRST(&nmp->nm_gsscl))) {
+               TAILQ_REMOVE(&nmp->nm_gsscl, cp, gss_clnt_entries);
+               cp->gss_clnt_entries.tqe_next = NFSNOLIST;
+               lck_mtx_lock(cp->gss_clnt_mtx);
+               if (cp->gss_clnt_flags & GSS_CTX_DESTROY) {
                        lck_mtx_unlock(cp->gss_clnt_mtx);
-                       req.r_gss_ctx = cp;
+                       continue;
                }
+               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.
+                * But don't bother if it's a forced unmount.
                 */
-               if (!(nmp->nm_state & NFSSTA_FORCE) && (cp->gss_clnt_service != RPCSEC_GSS_SVC_SYS)) {
+               if (!nfs_mount_gone(nmp) &&
+                   (cp->gss_clnt_flags & (GSS_CTX_INVAL | GSS_CTX_DESTROY | GSS_CTX_COMPLETE)) == GSS_CTX_COMPLETE) {
                        cp->gss_clnt_proc = RPCSEC_GSS_DESTROY;
 
                        error = 0;
@@ -1872,51 +2276,169 @@ nfs_gss_clnt_ctx_unmount(struct nfsmount *nmp)
                 * refcount is zero.
                 */
                lck_mtx_lock(cp->gss_clnt_mtx);
-               cp->gss_clnt_flags |= GSS_CTX_INVAL;
+               cp->gss_clnt_flags |= (GSS_CTX_INVAL | GSS_CTX_DESTROY);
                lck_mtx_unlock(cp->gss_clnt_mtx);
                nfs_gss_clnt_ctx_unref(&req);
+               lck_mtx_lock(&nmp->nm_lock);
        }
+       lck_mtx_unlock(&nmp->nm_lock);
+       assert(TAILQ_EMPTY(&nmp->nm_gsscl));
 }
 
+
 /*
- * Destroy a mounts context for a credential
+ * Removes a mounts context for a credential
  */
 int
-nfs_gss_clnt_ctx_destroy(struct nfsmount *nmp, kauth_cred_t cred)
+nfs_gss_clnt_ctx_remove(struct nfsmount *nmp, kauth_cred_t cred)
 {
        struct nfs_gss_clnt_ctx *cp;
        struct nfsreq req;
 
        req.r_nmp = nmp;
 
+       NFS_GSS_DBG("Enter\n");
+       NFS_GSS_CLNT_CTX_DUMP(nmp);
        lck_mtx_lock(&nmp->nm_lock);
        TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
+               lck_mtx_lock(cp->gss_clnt_mtx);
                if (nfs_gss_clnt_ctx_cred_match(cp->gss_clnt_cred, cred)) {
-                       if (cp->gss_clnt_flags & GSS_CTX_INVAL)
+                       if (cp->gss_clnt_flags & GSS_CTX_DESTROY) {
+                               NFS_GSS_DBG("Found destroyed context %d/%d. refcnt = %d continuing\n",
+                                           kauth_cred_getasid(cp->gss_clnt_cred),
+                                           kauth_cred_getauid(cp->gss_clnt_cred),
+                                           cp->gss_clnt_refcnt);
+                               lck_mtx_unlock(cp->gss_clnt_mtx);
                                continue;
-                       lck_mtx_lock(cp->gss_clnt_mtx);
+                       }
                        cp->gss_clnt_refcnt++;
-                       cp->gss_clnt_flags |= GSS_CTX_INVAL;
+                       cp->gss_clnt_flags |= (GSS_CTX_INVAL | GSS_CTX_DESTROY);
                        lck_mtx_unlock(cp->gss_clnt_mtx);
                        req.r_gss_ctx = cp;
-                       break;
+                       lck_mtx_unlock(&nmp->nm_lock);
+                       /*
+                        * Drop the reference to remove it if its
+                        * refcount is zero.
+                        */
+                       NFS_GSS_DBG("Removed context %d/%d refcnt = %d\n",
+                                   kauth_cred_getasid(cp->gss_clnt_cred),
+                                   kauth_cred_getuid(cp->gss_clnt_cred),
+                                   cp->gss_clnt_refcnt);
+                       nfs_gss_clnt_ctx_unref(&req);
+                       return (0);
                }
+               lck_mtx_unlock(cp->gss_clnt_mtx);
        }
+
        lck_mtx_unlock(&nmp->nm_lock);
+       
+       NFS_GSS_DBG("Returning ENOENT\n");
+       return (ENOENT);
+}
 
-       if (cp == NULL)
-               return (ENOENT);
+/*
+ * Sets a mounts principal for a session associated with cred.
+ */
+int
+nfs_gss_clnt_ctx_set_principal(struct nfsmount *nmp, vfs_context_t ctx,
+                              uint8_t *principal, uint32_t princlen, uint32_t nametype)
+
+{
+       struct nfsreq req;
+       int error;
+
+       NFS_GSS_DBG("Enter:\n");
 
+       bzero(&req, sizeof(struct nfsreq));
+       req.r_nmp = nmp;
+       req.r_gss_ctx = NULL;
+       req.r_auth = nmp->nm_auth;
+       req.r_thread = vfs_context_thread(ctx);
+       req.r_cred = vfs_context_ucred(ctx);
+
+       error = nfs_gss_clnt_ctx_find_principal(&req, principal, princlen, nametype);
+       NFS_GSS_DBG("nfs_gss_clnt_ctx_find_principal returned %d\n", error);
        /*
-        * Drop the reference to remove it if its
-        * refcount is zero.
+        * We don't care about auth errors. Those would indicate that the context is in the
+        * neagative cache and if and when the user has credentials for the principal
+        * we should be good to go in that we will select those credentials for this principal.
         */
+       if (error == EACCES || error == EAUTH || error == ENEEDAUTH)
+               error = 0;
+
+       /* We're done with this request */
        nfs_gss_clnt_ctx_unref(&req);
 
-       return (0);
+       return (error);
 }
 
+/*
+ * Gets a mounts principal from a session associated with cred
+ */
+int
+nfs_gss_clnt_ctx_get_principal(struct nfsmount *nmp, vfs_context_t ctx,
+                              struct user_nfs_gss_principal *p)
+{
+       struct nfsreq req;
+       int error = 0;
+       struct nfs_gss_clnt_ctx *cp;
+       kauth_cred_t cred = vfs_context_ucred(ctx);
+       const char *princ;
+       char CTXBUF[NFS_CTXBUFSZ];
+
+       req.r_nmp = nmp;
+       lck_mtx_lock(&nmp->nm_lock);
+       TAILQ_FOREACH(cp, &nmp->nm_gsscl, gss_clnt_entries) {
+               lck_mtx_lock(cp->gss_clnt_mtx);
+               if (cp->gss_clnt_flags & GSS_CTX_DESTROY) {
+                       NFS_GSS_DBG("Found destroyed context %s refcnt = %d continuing\n",
+                                   NFS_GSS_CTX(&req, cp),
+                                   cp->gss_clnt_refcnt);
+                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                       continue;
+               }
+               if (nfs_gss_clnt_ctx_cred_match(cp->gss_clnt_cred, cred)) {
+                       cp->gss_clnt_refcnt++;
+                       lck_mtx_unlock(cp->gss_clnt_mtx);
+                       goto out;
+               }
+               lck_mtx_unlock(cp->gss_clnt_mtx);
+       }
+
+out:
+       if (cp == NULL) {
+               lck_mtx_unlock(&nmp->nm_lock);
+               p->princlen = 0;
+               p->principal = USER_ADDR_NULL;
+               p->nametype = GSSD_STRING_NAME;
+               p->flags |= NFS_IOC_NO_CRED_FLAG;
+               NFS_GSS_DBG("No context found for session %d by uid %d\n",
+                           kauth_cred_getasid(cred), kauth_cred_getuid(cred));
+               return (0);
+       }
+
+       princ = cp->gss_clnt_principal ? (char *)cp->gss_clnt_principal : cp->gss_clnt_display;
+       p->princlen = cp->gss_clnt_principal ? cp->gss_clnt_prinlen :
+               (cp->gss_clnt_display ? strlen(cp->gss_clnt_display) : 0);
+       p->nametype = cp->gss_clnt_prinnt;
+       if (princ) {
+               char *pp;
+
+               MALLOC(pp, char *, p->princlen, M_TEMP, M_WAITOK);
+               if (pp) {
+                       bcopy(princ, pp, p->princlen);
+                       p->principal = CAST_USER_ADDR_T(pp);
+               }
+               else
+                       error = ENOMEM;
+       }
+       lck_mtx_unlock(&nmp->nm_lock);
 
+       req.r_gss_ctx = cp;
+       NFS_GSS_DBG("Found context %s\n", NFS_GSS_CTX(&req, NULL));
+       nfs_gss_clnt_ctx_unref(&req);
+       return (error);
+}
 #endif /* NFSCLIENT */
 
 /*************
@@ -2035,6 +2557,8 @@ nfs_gss_svc_ctx_timer(__unused void *param1, __unused void *param2)
        lck_mtx_lock(nfs_gss_svc_ctx_mutex);
        clock_get_uptime(&timenow);
 
+       NFS_GSS_DBG("is running\n");
+
        /*
         * Scan all the hash chains
         */
@@ -2052,6 +2576,7 @@ nfs_gss_svc_ctx_timer(__unused void *param1, __unused void *param2)
                                 * A stale context - remove it
                                 */
                                LIST_REMOVE(cp, gss_svc_entries);
+                               NFS_GSS_DBG("Removing contex for %d\n", cp->gss_svc_uid);
                                if (cp->gss_svc_seqbits)
                                        FREE(cp->gss_svc_seqbits, M_TEMP);
                                lck_mtx_destroy(cp->gss_svc_mtx, nfs_gss_svc_grp);
@@ -2996,7 +3521,7 @@ nfs_gss_mach_alloc_buffer(u_char *buf, uint32_t buflen, vm_map_copy_t *addr)
 
        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);
+       kr = vm_allocate(ipc_kernel_map, &kmem_buf, tbuflen, VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_KERN_MEMORY_FILE));
        if (kr != 0) {
                printf("nfs_gss_mach_alloc_buffer: vm_allocate failed\n");
                return;
@@ -3007,7 +3532,7 @@ nfs_gss_mach_alloc_buffer(u_char *buf, uint32_t buflen, vm_map_copy_t *addr)
                                           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);
+               VM_PROT_READ|VM_PROT_WRITE|VM_PROT_MEMORY_TAG_MAKE(VM_KERN_MEMORY_FILE), FALSE);
        if (kr != 0) {
                printf("nfs_gss_mach_alloc_buffer: vm_map_wire failed\n");
                return;